OpenCloudOS-Kernel/fs/dcache.c

3666 lines
94 KiB
C
Raw Normal View History

/*
* fs/dcache.c
*
* Complete reimplementation
* (C) 1997 Thomas Schoebel-Theuer,
* with heavy changes by Linus Torvalds
*/
/*
* Notes on the allocation strategy:
*
* The dcache is a master of the icache - whenever a dcache entry
* exists, the inode will always exist. "iput()" is done either when
* the dcache entry is deleted or garbage collected.
*/
#include <linux/syscalls.h>
#include <linux/string.h>
#include <linux/mm.h>
#include <linux/fs.h>
#include <linux/fsnotify.h>
#include <linux/slab.h>
#include <linux/init.h>
#include <linux/hash.h>
#include <linux/cache.h>
#include <linux/export.h>
#include <linux/mount.h>
#include <linux/file.h>
#include <asm/uaccess.h>
#include <linux/security.h>
#include <linux/seqlock.h>
#include <linux/swap.h>
#include <linux/bootmem.h>
#include <linux/fs_struct.h>
#include <linux/hardirq.h>
#include <linux/bit_spinlock.h>
#include <linux/rculist_bl.h>
#include <linux/prefetch.h>
#include <linux/ratelimit.h>
#include <linux/list_lru.h>
#include <linux/kasan.h>
#include "internal.h"
#include "mount.h"
/*
* Usage:
* dcache->d_inode->i_lock protects:
* - i_dentry, d_u.d_alias, d_inode of aliases
* dcache_hash_bucket lock protects:
* - the dcache hash table
* s_anon bl list spinlock protects:
* - the s_anon list (see __d_drop)
dentry: move to per-sb LRU locks With the dentry LRUs being per-sb structures, there is no real need for a global dentry_lru_lock. The locking can be made more fine-grained by moving to a per-sb LRU lock, isolating the LRU operations of different filesytsems completely from each other. The need for this is independent of any performance consideration that may arise: in the interest of abstracting the lru operations away, it is mandatory that each lru works around its own lock instead of a global lock for all of them. [glommer@openvz.org: updated changelog ] Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Glauber Costa <glommer@openvz.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Acked-by: Mel Gorman <mgorman@suse.de> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com> Cc: Arve Hjønnevåg <arve@android.com> Cc: Carlos Maiolino <cmaiolino@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Chuck Lever <chuck.lever@oracle.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: David Rientjes <rientjes@google.com> Cc: Gleb Natapov <gleb@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: J. Bruce Fields <bfields@redhat.com> Cc: Jan Kara <jack@suse.cz> Cc: Jerome Glisse <jglisse@redhat.com> Cc: John Stultz <john.stultz@linaro.org> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Kent Overstreet <koverstreet@google.com> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Steven Whitehouse <swhiteho@redhat.com> Cc: Thomas Hellstrom <thellstrom@vmware.com> Cc: Trond Myklebust <Trond.Myklebust@netapp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:17:55 +08:00
* dentry->d_sb->s_dentry_lru_lock protects:
* - the dcache lru lists and counters
* d_lock protects:
* - d_flags
* - d_name
* - d_lru
* - d_count
* - d_unhashed()
* - d_parent and d_subdirs
* - childrens' d_child and d_parent
* - d_u.d_alias, d_inode
*
* Ordering:
* dentry->d_inode->i_lock
* dentry->d_lock
dentry: move to per-sb LRU locks With the dentry LRUs being per-sb structures, there is no real need for a global dentry_lru_lock. The locking can be made more fine-grained by moving to a per-sb LRU lock, isolating the LRU operations of different filesytsems completely from each other. The need for this is independent of any performance consideration that may arise: in the interest of abstracting the lru operations away, it is mandatory that each lru works around its own lock instead of a global lock for all of them. [glommer@openvz.org: updated changelog ] Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Glauber Costa <glommer@openvz.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Acked-by: Mel Gorman <mgorman@suse.de> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com> Cc: Arve Hjønnevåg <arve@android.com> Cc: Carlos Maiolino <cmaiolino@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Chuck Lever <chuck.lever@oracle.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: David Rientjes <rientjes@google.com> Cc: Gleb Natapov <gleb@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: J. Bruce Fields <bfields@redhat.com> Cc: Jan Kara <jack@suse.cz> Cc: Jerome Glisse <jglisse@redhat.com> Cc: John Stultz <john.stultz@linaro.org> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Kent Overstreet <koverstreet@google.com> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Steven Whitehouse <swhiteho@redhat.com> Cc: Thomas Hellstrom <thellstrom@vmware.com> Cc: Trond Myklebust <Trond.Myklebust@netapp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:17:55 +08:00
* dentry->d_sb->s_dentry_lru_lock
* dcache_hash_bucket lock
* s_anon lock
*
* If there is an ancestor relationship:
* dentry->d_parent->...->d_parent->d_lock
* ...
* dentry->d_parent->d_lock
* dentry->d_lock
*
* If no ancestor relationship:
* if (dentry1 < dentry2)
* dentry1->d_lock
* dentry2->d_lock
*/
int sysctl_vfs_cache_pressure __read_mostly = 100;
EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
EXPORT_SYMBOL(rename_lock);
static struct kmem_cache *dentry_cache __read_mostly;
/*
* This is the single most critical data structure when it comes
* to the dcache: the hashtable for lookups. Somebody should try
* to make this good - I've just made it work.
*
* This hash-function tries to avoid losing too many bits of hash
* information, yet avoid using a prime hash-size or similar.
*/
static unsigned int d_hash_mask __read_mostly;
static unsigned int d_hash_shift __read_mostly;
static struct hlist_bl_head *dentry_hashtable __read_mostly;
static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
vfs: get rid of batshit-insane pointless dentry hash calculations For some odd historical reason, the final mixing round for the dentry cache hash table lookup had an insane "xor with big constant" logic. In two places. The big constant that is being xor'ed is GOLDEN_RATIO_PRIME, which is a fairly random-looking number that is designed to be *multiplied* with so that the bits get spread out over a whole long-word. But xor'ing with it is insane. It doesn't really even change the hash - it really only shifts the hash around in the hash table. To make matters worse, the insane big constant is different on 32-bit and 64-bit builds, even though the name hash bits we use are always 32-bit (and the bits from the pointer we mix in effectively are too). It's all total voodoo programming, in other words. Now, some testing and analysis of the hash chains shows that the rest of the hash function seems to be fairly good. It does pick the right bits of the parent dentry pointer, for example, and while it's generally a bad idea to use an xor to mix down the upper bits (because if there is a repeating pattern, the xor can cause "destructive interference"), it seems to not have been a disaster. For example, replacing the hash with the normal "hash_long()" code (that uses the GOLDEN_RATIO_PRIME constant correctly, btw) actually just makes the hash worse. The hand-picked hash knew which bits of the pointer had the highest entropy, and hash_long() ends up mixing bits less optimally at least in some trivial tests. So the hash function overall seems fine, it just has that really odd "shift result around by a constant xor". So get rid of the silly xor, and replace the down-mixing of the bits with an add instead of an xor that tends to not have the same kind of destructive interference issues. Some stats on the resulting hash chains shows that they look statistically identical before and after, but the code is simpler and no longer makes you go "WTF?". Also, the incoming hash really is just "unsigned int", not a long, and there's no real point to worry about the high 26 bits of the dentry pointer for the 64-bit case, because they are all going to be identical anyway. So also change the hashing to be done in the more natural 'unsigned int' that is the real size of the actual hashed data anyway. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-03-20 07:19:53 +08:00
unsigned int hash)
{
vfs: get rid of batshit-insane pointless dentry hash calculations For some odd historical reason, the final mixing round for the dentry cache hash table lookup had an insane "xor with big constant" logic. In two places. The big constant that is being xor'ed is GOLDEN_RATIO_PRIME, which is a fairly random-looking number that is designed to be *multiplied* with so that the bits get spread out over a whole long-word. But xor'ing with it is insane. It doesn't really even change the hash - it really only shifts the hash around in the hash table. To make matters worse, the insane big constant is different on 32-bit and 64-bit builds, even though the name hash bits we use are always 32-bit (and the bits from the pointer we mix in effectively are too). It's all total voodoo programming, in other words. Now, some testing and analysis of the hash chains shows that the rest of the hash function seems to be fairly good. It does pick the right bits of the parent dentry pointer, for example, and while it's generally a bad idea to use an xor to mix down the upper bits (because if there is a repeating pattern, the xor can cause "destructive interference"), it seems to not have been a disaster. For example, replacing the hash with the normal "hash_long()" code (that uses the GOLDEN_RATIO_PRIME constant correctly, btw) actually just makes the hash worse. The hand-picked hash knew which bits of the pointer had the highest entropy, and hash_long() ends up mixing bits less optimally at least in some trivial tests. So the hash function overall seems fine, it just has that really odd "shift result around by a constant xor". So get rid of the silly xor, and replace the down-mixing of the bits with an add instead of an xor that tends to not have the same kind of destructive interference issues. Some stats on the resulting hash chains shows that they look statistically identical before and after, but the code is simpler and no longer makes you go "WTF?". Also, the incoming hash really is just "unsigned int", not a long, and there's no real point to worry about the high 26 bits of the dentry pointer for the 64-bit case, because they are all going to be identical anyway. So also change the hashing to be done in the more natural 'unsigned int' that is the real size of the actual hashed data anyway. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-03-20 07:19:53 +08:00
hash += (unsigned long) parent / L1_CACHE_BYTES;
vfs: fix bad hashing of dentries Josef Bacik found a performance regression between 3.2 and 3.10 and narrowed it down to commit bfcfaa77bdf0 ("vfs: use 'unsigned long' accesses for dcache name comparison and hashing"). He reports: "The test case is essentially for (i = 0; i < 1000000; i++) mkdir("a$i"); On xfs on a fio card this goes at about 20k dir/sec with 3.2, and 12k dir/sec with 3.10. This is because we spend waaaaay more time in __d_lookup on 3.10 than in 3.2. The new hashing function for strings is suboptimal for < sizeof(unsigned long) string names (and hell even > sizeof(unsigned long) string names that I've tested). I broke out the old hashing function and the new one into a userspace helper to get real numbers and this is what I'm getting: Old hash table had 1000000 entries, 0 dupes, 0 max dupes New hash table had 12628 entries, 987372 dupes, 900 max dupes We had 11400 buckets with a p50 of 30 dupes, p90 of 240 dupes, p99 of 567 dupes for the new hash My test does the hash, and then does the d_hash into a integer pointer array the same size as the dentry hash table on my system, and then just increments the value at the address we got to see how many entries we overlap with. As you can see the old hash function ended up with all 1 million entries in their own bucket, whereas the new one they are only distributed among ~12.5k buckets, which is why we're using so much more CPU in __d_lookup". The reason for this hash regression is two-fold: - On 64-bit architectures the down-mixing of the original 64-bit word-at-a-time hash into the final 32-bit hash value is very simplistic and suboptimal, and just adds the two 32-bit parts together. In particular, because there is no bit shuffling and the mixing boundary is also a byte boundary, similar character patterns in the low and high word easily end up just canceling each other out. - the old byte-at-a-time hash mixed each byte into the final hash as it hashed the path component name, resulting in the low bits of the hash generally being a good source of hash data. That is not true for the word-at-a-time case, and the hash data is distributed among all the bits. The fix is the same in both cases: do a better job of mixing the bits up and using as much of the hash data as possible. We already have the "hash_32|64()" functions to do that. Reported-by: Josef Bacik <jbacik@fb.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Christoph Hellwig <hch@infradead.org> Cc: Chris Mason <clm@fb.com> Cc: linux-fsdevel@vger.kernel.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-09-14 02:30:10 +08:00
return dentry_hashtable + hash_32(hash, d_hash_shift);
}
parallel lookups machinery, part 3 We will need to be able to check if there is an in-lookup dentry with matching parent/name. Right now it's impossible, but as soon as start locking directories shared such beasts will appear. Add a secondary hash for locating those. Hash chains go through the same space where d_alias will be once it's not in-lookup anymore. Search is done under the same bitlock we use for modifications - with the primary hash we can rely on d_rehash() into the wrong chain being the worst that could happen, but here the pointers are buggered once it's removed from the chain. On the other hand, the chains are not going to be long and normally we'll end up adding to the chain anyway. That allows us to avoid bothering with ->d_lock when doing the comparisons - everything is stable until removed from chain. New helper: d_alloc_parallel(). Right now it allocates, verifies that no hashed and in-lookup matches exist and adds to in-lookup hash. Returns ERR_PTR() for error, hashed match (in the unlikely case it's been found) or new dentry. In-lookup matches trigger BUG() for now; that will change in the next commit when we introduce waiting for ongoing lookup to finish. Note that in-lookup matches won't be possible until we actually go for shared locking. lookup_slow() switched to use of d_alloc_parallel(). Again, these commits are separated only for making it easier to review. All this machinery will start doing something useful only when we go for shared locking; it's just that the combination is too large for my taste. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 14:42:04 +08:00
#define IN_LOOKUP_SHIFT 10
static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
unsigned int hash)
{
hash += (unsigned long) parent / L1_CACHE_BYTES;
return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
}
/* Statistics gathering. */
struct dentry_stat_t dentry_stat = {
.age_limit = 45,
};
fs: bump inode and dentry counters to long This series reworks our current object cache shrinking infrastructure in two main ways: * Noticing that a lot of users copy and paste their own version of LRU lists for objects, we put some effort in providing a generic version. It is modeled after the filesystem users: dentries, inodes, and xfs (for various tasks), but we expect that other users could benefit in the near future with little or no modification. Let us know if you have any issues. * The underlying list_lru being proposed automatically and transparently keeps the elements in per-node lists, and is able to manipulate the node lists individually. Given this infrastructure, we are able to modify the up-to-now hammer called shrink_slab to proceed with node-reclaim instead of always searching memory from all over like it has been doing. Per-node lru lists are also expected to lead to less contention in the lru locks on multi-node scans, since we are now no longer fighting for a global lock. The locks usually disappear from the profilers with this change. Although we have no official benchmarks for this version - be our guest to independently evaluate this - earlier versions of this series were performance tested (details at http://permalink.gmane.org/gmane.linux.kernel.mm/100537) yielding no visible performance regressions while yielding a better qualitative behavior in NUMA machines. With this infrastructure in place, we can use the list_lru entry point to provide memcg isolation and per-memcg targeted reclaim. Historically, those two pieces of work have been posted together. This version presents only the infrastructure work, deferring the memcg work for a later time, so we can focus on getting this part tested. You can see more about the history of such work at http://lwn.net/Articles/552769/ Dave Chinner (18): dcache: convert dentry_stat.nr_unused to per-cpu counters dentry: move to per-sb LRU locks dcache: remove dentries from LRU before putting on dispose list mm: new shrinker API shrinker: convert superblock shrinkers to new API list: add a new LRU list type inode: convert inode lru list to generic lru list code. dcache: convert to use new lru list infrastructure list_lru: per-node list infrastructure shrinker: add node awareness fs: convert inode and dentry shrinking to be node aware xfs: convert buftarg LRU to generic code xfs: rework buffer dispose list tracking xfs: convert dquot cache lru to list_lru fs: convert fs shrinkers to new scan/count API drivers: convert shrinkers to new count/scan API shrinker: convert remaining shrinkers to count/scan API shrinker: Kill old ->shrink API. Glauber Costa (7): fs: bump inode and dentry counters to long super: fix calculation of shrinkable objects for small numbers list_lru: per-node API vmscan: per-node deferred work i915: bail out earlier when shrinker cannot acquire mutex hugepage: convert huge zero page shrinker to new shrinker API list_lru: dynamically adjust node arrays This patch: There are situations in very large machines in which we can have a large quantity of dirty inodes, unused dentries, etc. This is particularly true when umounting a filesystem, where eventually since every live object will eventually be discarded. Dave Chinner reported a problem with this while experimenting with the shrinker revamp patchset. So we believe it is time for a change. This patch just moves int to longs. Machines where it matters should have a big long anyway. Signed-off-by: Glauber Costa <glommer@openvz.org> Cc: Dave Chinner <dchinner@redhat.com> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com> Cc: Arve Hjønnevåg <arve@android.com> Cc: Carlos Maiolino <cmaiolino@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Chuck Lever <chuck.lever@oracle.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: Dave Chinner <dchinner@redhat.com> Cc: David Rientjes <rientjes@google.com> Cc: Gleb Natapov <gleb@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: J. Bruce Fields <bfields@redhat.com> Cc: Jan Kara <jack@suse.cz> Cc: Jerome Glisse <jglisse@redhat.com> Cc: John Stultz <john.stultz@linaro.org> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Kent Overstreet <koverstreet@google.com> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Steven Whitehouse <swhiteho@redhat.com> Cc: Thomas Hellstrom <thellstrom@vmware.com> Cc: Trond Myklebust <Trond.Myklebust@netapp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:17:53 +08:00
static DEFINE_PER_CPU(long, nr_dentry);
dcache: convert dentry_stat.nr_unused to per-cpu counters Before we split up the dcache_lru_lock, the unused dentry counter needs to be made independent of the global dcache_lru_lock. Convert it to per-cpu counters to do this. Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Glauber Costa <glommer@openvz.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Acked-by: Mel Gorman <mgorman@suse.de> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com> Cc: Arve Hjønnevåg <arve@android.com> Cc: Carlos Maiolino <cmaiolino@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Chuck Lever <chuck.lever@oracle.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: David Rientjes <rientjes@google.com> Cc: Gleb Natapov <gleb@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: J. Bruce Fields <bfields@redhat.com> Cc: Jan Kara <jack@suse.cz> Cc: Jerome Glisse <jglisse@redhat.com> Cc: John Stultz <john.stultz@linaro.org> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Kent Overstreet <koverstreet@google.com> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Steven Whitehouse <swhiteho@redhat.com> Cc: Thomas Hellstrom <thellstrom@vmware.com> Cc: Trond Myklebust <Trond.Myklebust@netapp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:17:54 +08:00
static DEFINE_PER_CPU(long, nr_dentry_unused);
#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
dcache: convert dentry_stat.nr_unused to per-cpu counters Before we split up the dcache_lru_lock, the unused dentry counter needs to be made independent of the global dcache_lru_lock. Convert it to per-cpu counters to do this. Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Glauber Costa <glommer@openvz.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Acked-by: Mel Gorman <mgorman@suse.de> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com> Cc: Arve Hjønnevåg <arve@android.com> Cc: Carlos Maiolino <cmaiolino@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Chuck Lever <chuck.lever@oracle.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: David Rientjes <rientjes@google.com> Cc: Gleb Natapov <gleb@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: J. Bruce Fields <bfields@redhat.com> Cc: Jan Kara <jack@suse.cz> Cc: Jerome Glisse <jglisse@redhat.com> Cc: John Stultz <john.stultz@linaro.org> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Kent Overstreet <koverstreet@google.com> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Steven Whitehouse <swhiteho@redhat.com> Cc: Thomas Hellstrom <thellstrom@vmware.com> Cc: Trond Myklebust <Trond.Myklebust@netapp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:17:54 +08:00
/*
* Here we resort to our own counters instead of using generic per-cpu counters
* for consistency with what the vfs inode code does. We are expected to harvest
* better code and performance by having our own specialized counters.
*
* Please note that the loop is done over all possible CPUs, not over all online
* CPUs. The reason for this is that we don't want to play games with CPUs going
* on and off. If one of them goes off, we will just keep their counters.
*
* glommer: See cffbc8a for details, and if you ever intend to change this,
* please update all vfs counters to match.
*/
fs: bump inode and dentry counters to long This series reworks our current object cache shrinking infrastructure in two main ways: * Noticing that a lot of users copy and paste their own version of LRU lists for objects, we put some effort in providing a generic version. It is modeled after the filesystem users: dentries, inodes, and xfs (for various tasks), but we expect that other users could benefit in the near future with little or no modification. Let us know if you have any issues. * The underlying list_lru being proposed automatically and transparently keeps the elements in per-node lists, and is able to manipulate the node lists individually. Given this infrastructure, we are able to modify the up-to-now hammer called shrink_slab to proceed with node-reclaim instead of always searching memory from all over like it has been doing. Per-node lru lists are also expected to lead to less contention in the lru locks on multi-node scans, since we are now no longer fighting for a global lock. The locks usually disappear from the profilers with this change. Although we have no official benchmarks for this version - be our guest to independently evaluate this - earlier versions of this series were performance tested (details at http://permalink.gmane.org/gmane.linux.kernel.mm/100537) yielding no visible performance regressions while yielding a better qualitative behavior in NUMA machines. With this infrastructure in place, we can use the list_lru entry point to provide memcg isolation and per-memcg targeted reclaim. Historically, those two pieces of work have been posted together. This version presents only the infrastructure work, deferring the memcg work for a later time, so we can focus on getting this part tested. You can see more about the history of such work at http://lwn.net/Articles/552769/ Dave Chinner (18): dcache: convert dentry_stat.nr_unused to per-cpu counters dentry: move to per-sb LRU locks dcache: remove dentries from LRU before putting on dispose list mm: new shrinker API shrinker: convert superblock shrinkers to new API list: add a new LRU list type inode: convert inode lru list to generic lru list code. dcache: convert to use new lru list infrastructure list_lru: per-node list infrastructure shrinker: add node awareness fs: convert inode and dentry shrinking to be node aware xfs: convert buftarg LRU to generic code xfs: rework buffer dispose list tracking xfs: convert dquot cache lru to list_lru fs: convert fs shrinkers to new scan/count API drivers: convert shrinkers to new count/scan API shrinker: convert remaining shrinkers to count/scan API shrinker: Kill old ->shrink API. Glauber Costa (7): fs: bump inode and dentry counters to long super: fix calculation of shrinkable objects for small numbers list_lru: per-node API vmscan: per-node deferred work i915: bail out earlier when shrinker cannot acquire mutex hugepage: convert huge zero page shrinker to new shrinker API list_lru: dynamically adjust node arrays This patch: There are situations in very large machines in which we can have a large quantity of dirty inodes, unused dentries, etc. This is particularly true when umounting a filesystem, where eventually since every live object will eventually be discarded. Dave Chinner reported a problem with this while experimenting with the shrinker revamp patchset. So we believe it is time for a change. This patch just moves int to longs. Machines where it matters should have a big long anyway. Signed-off-by: Glauber Costa <glommer@openvz.org> Cc: Dave Chinner <dchinner@redhat.com> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com> Cc: Arve Hjønnevåg <arve@android.com> Cc: Carlos Maiolino <cmaiolino@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Chuck Lever <chuck.lever@oracle.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: Dave Chinner <dchinner@redhat.com> Cc: David Rientjes <rientjes@google.com> Cc: Gleb Natapov <gleb@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: J. Bruce Fields <bfields@redhat.com> Cc: Jan Kara <jack@suse.cz> Cc: Jerome Glisse <jglisse@redhat.com> Cc: John Stultz <john.stultz@linaro.org> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Kent Overstreet <koverstreet@google.com> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Steven Whitehouse <swhiteho@redhat.com> Cc: Thomas Hellstrom <thellstrom@vmware.com> Cc: Trond Myklebust <Trond.Myklebust@netapp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:17:53 +08:00
static long get_nr_dentry(void)
fs: use fast counters for vfs caches percpu_counter library generates quite nasty code, so unless you need to dynamically allocate counters or take fast approximate value, a simple per cpu set of counters is much better. The percpu_counter can never be made to work as well, because it has an indirection from pointer to percpu memory, and it can't use direct this_cpu_inc interfaces because it doesn't use static PER_CPU data, so code will always be worse. In the fastpath, it is the difference between this: incl %gs:nr_dentry # nr_dentry and this: movl percpu_counter_batch(%rip), %edx # percpu_counter_batch, movl $1, %esi #, movq $nr_dentry, %rdi #, call __percpu_counter_add # (plus I clobber registers) __percpu_counter_add: pushq %rbp # movq %rsp, %rbp #, subq $32, %rsp #, movq %rbx, -24(%rbp) #, movq %r12, -16(%rbp) #, movq %r13, -8(%rbp) #, movq %rdi, %rbx # fbc, fbc #APP # 216 "/home/npiggin/usr/src/linux-2.6/arch/x86/include/asm/thread_info.h" 1 movq %gs:kernel_stack,%rax #, pfo_ret__ # 0 "" 2 #NO_APP incl -8124(%rax) # <variable>.preempt_count movq 32(%rdi), %r12 # <variable>.counters, tcp_ptr__ #APP # 78 "lib/percpu_counter.c" 1 add %gs:this_cpu_off, %r12 # this_cpu_off, tcp_ptr__ # 0 "" 2 #NO_APP movslq (%r12),%r13 #* tcp_ptr__, tmp73 movslq %edx,%rax # batch, batch addq %rsi, %r13 # amount, count cmpq %rax, %r13 # batch, count jge .L27 #, negl %edx # tmp76 movslq %edx,%rdx # tmp76, tmp77 cmpq %rdx, %r13 # tmp77, count jg .L28 #, .L27: movq %rbx, %rdi # fbc, call _raw_spin_lock # addq %r13, 8(%rbx) # count, <variable>.count movq %rbx, %rdi # fbc, movl $0, (%r12) #,* tcp_ptr__ call _raw_spin_unlock # .L29: #APP # 216 "/home/npiggin/usr/src/linux-2.6/arch/x86/include/asm/thread_info.h" 1 movq %gs:kernel_stack,%rax #, pfo_ret__ # 0 "" 2 #NO_APP decl -8124(%rax) # <variable>.preempt_count movq -8136(%rax), %rax #, D.14625 testb $8, %al #, D.14625 jne .L32 #, .L31: movq -24(%rbp), %rbx #, movq -16(%rbp), %r12 #, movq -8(%rbp), %r13 #, leave ret .p2align 4,,10 .p2align 3 .L28: movl %r13d, (%r12) # count,* jmp .L29 # .L32: call preempt_schedule # .p2align 4,,6 jmp .L31 # .size __percpu_counter_add, .-__percpu_counter_add .p2align 4,,15 Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:19 +08:00
{
int i;
fs: bump inode and dentry counters to long This series reworks our current object cache shrinking infrastructure in two main ways: * Noticing that a lot of users copy and paste their own version of LRU lists for objects, we put some effort in providing a generic version. It is modeled after the filesystem users: dentries, inodes, and xfs (for various tasks), but we expect that other users could benefit in the near future with little or no modification. Let us know if you have any issues. * The underlying list_lru being proposed automatically and transparently keeps the elements in per-node lists, and is able to manipulate the node lists individually. Given this infrastructure, we are able to modify the up-to-now hammer called shrink_slab to proceed with node-reclaim instead of always searching memory from all over like it has been doing. Per-node lru lists are also expected to lead to less contention in the lru locks on multi-node scans, since we are now no longer fighting for a global lock. The locks usually disappear from the profilers with this change. Although we have no official benchmarks for this version - be our guest to independently evaluate this - earlier versions of this series were performance tested (details at http://permalink.gmane.org/gmane.linux.kernel.mm/100537) yielding no visible performance regressions while yielding a better qualitative behavior in NUMA machines. With this infrastructure in place, we can use the list_lru entry point to provide memcg isolation and per-memcg targeted reclaim. Historically, those two pieces of work have been posted together. This version presents only the infrastructure work, deferring the memcg work for a later time, so we can focus on getting this part tested. You can see more about the history of such work at http://lwn.net/Articles/552769/ Dave Chinner (18): dcache: convert dentry_stat.nr_unused to per-cpu counters dentry: move to per-sb LRU locks dcache: remove dentries from LRU before putting on dispose list mm: new shrinker API shrinker: convert superblock shrinkers to new API list: add a new LRU list type inode: convert inode lru list to generic lru list code. dcache: convert to use new lru list infrastructure list_lru: per-node list infrastructure shrinker: add node awareness fs: convert inode and dentry shrinking to be node aware xfs: convert buftarg LRU to generic code xfs: rework buffer dispose list tracking xfs: convert dquot cache lru to list_lru fs: convert fs shrinkers to new scan/count API drivers: convert shrinkers to new count/scan API shrinker: convert remaining shrinkers to count/scan API shrinker: Kill old ->shrink API. Glauber Costa (7): fs: bump inode and dentry counters to long super: fix calculation of shrinkable objects for small numbers list_lru: per-node API vmscan: per-node deferred work i915: bail out earlier when shrinker cannot acquire mutex hugepage: convert huge zero page shrinker to new shrinker API list_lru: dynamically adjust node arrays This patch: There are situations in very large machines in which we can have a large quantity of dirty inodes, unused dentries, etc. This is particularly true when umounting a filesystem, where eventually since every live object will eventually be discarded. Dave Chinner reported a problem with this while experimenting with the shrinker revamp patchset. So we believe it is time for a change. This patch just moves int to longs. Machines where it matters should have a big long anyway. Signed-off-by: Glauber Costa <glommer@openvz.org> Cc: Dave Chinner <dchinner@redhat.com> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com> Cc: Arve Hjønnevåg <arve@android.com> Cc: Carlos Maiolino <cmaiolino@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Chuck Lever <chuck.lever@oracle.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: Dave Chinner <dchinner@redhat.com> Cc: David Rientjes <rientjes@google.com> Cc: Gleb Natapov <gleb@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: J. Bruce Fields <bfields@redhat.com> Cc: Jan Kara <jack@suse.cz> Cc: Jerome Glisse <jglisse@redhat.com> Cc: John Stultz <john.stultz@linaro.org> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Kent Overstreet <koverstreet@google.com> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Steven Whitehouse <swhiteho@redhat.com> Cc: Thomas Hellstrom <thellstrom@vmware.com> Cc: Trond Myklebust <Trond.Myklebust@netapp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:17:53 +08:00
long sum = 0;
fs: use fast counters for vfs caches percpu_counter library generates quite nasty code, so unless you need to dynamically allocate counters or take fast approximate value, a simple per cpu set of counters is much better. The percpu_counter can never be made to work as well, because it has an indirection from pointer to percpu memory, and it can't use direct this_cpu_inc interfaces because it doesn't use static PER_CPU data, so code will always be worse. In the fastpath, it is the difference between this: incl %gs:nr_dentry # nr_dentry and this: movl percpu_counter_batch(%rip), %edx # percpu_counter_batch, movl $1, %esi #, movq $nr_dentry, %rdi #, call __percpu_counter_add # (plus I clobber registers) __percpu_counter_add: pushq %rbp # movq %rsp, %rbp #, subq $32, %rsp #, movq %rbx, -24(%rbp) #, movq %r12, -16(%rbp) #, movq %r13, -8(%rbp) #, movq %rdi, %rbx # fbc, fbc #APP # 216 "/home/npiggin/usr/src/linux-2.6/arch/x86/include/asm/thread_info.h" 1 movq %gs:kernel_stack,%rax #, pfo_ret__ # 0 "" 2 #NO_APP incl -8124(%rax) # <variable>.preempt_count movq 32(%rdi), %r12 # <variable>.counters, tcp_ptr__ #APP # 78 "lib/percpu_counter.c" 1 add %gs:this_cpu_off, %r12 # this_cpu_off, tcp_ptr__ # 0 "" 2 #NO_APP movslq (%r12),%r13 #* tcp_ptr__, tmp73 movslq %edx,%rax # batch, batch addq %rsi, %r13 # amount, count cmpq %rax, %r13 # batch, count jge .L27 #, negl %edx # tmp76 movslq %edx,%rdx # tmp76, tmp77 cmpq %rdx, %r13 # tmp77, count jg .L28 #, .L27: movq %rbx, %rdi # fbc, call _raw_spin_lock # addq %r13, 8(%rbx) # count, <variable>.count movq %rbx, %rdi # fbc, movl $0, (%r12) #,* tcp_ptr__ call _raw_spin_unlock # .L29: #APP # 216 "/home/npiggin/usr/src/linux-2.6/arch/x86/include/asm/thread_info.h" 1 movq %gs:kernel_stack,%rax #, pfo_ret__ # 0 "" 2 #NO_APP decl -8124(%rax) # <variable>.preempt_count movq -8136(%rax), %rax #, D.14625 testb $8, %al #, D.14625 jne .L32 #, .L31: movq -24(%rbp), %rbx #, movq -16(%rbp), %r12 #, movq -8(%rbp), %r13 #, leave ret .p2align 4,,10 .p2align 3 .L28: movl %r13d, (%r12) # count,* jmp .L29 # .L32: call preempt_schedule # .p2align 4,,6 jmp .L31 # .size __percpu_counter_add, .-__percpu_counter_add .p2align 4,,15 Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:19 +08:00
for_each_possible_cpu(i)
sum += per_cpu(nr_dentry, i);
return sum < 0 ? 0 : sum;
}
dcache: convert dentry_stat.nr_unused to per-cpu counters Before we split up the dcache_lru_lock, the unused dentry counter needs to be made independent of the global dcache_lru_lock. Convert it to per-cpu counters to do this. Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Glauber Costa <glommer@openvz.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Acked-by: Mel Gorman <mgorman@suse.de> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com> Cc: Arve Hjønnevåg <arve@android.com> Cc: Carlos Maiolino <cmaiolino@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Chuck Lever <chuck.lever@oracle.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: David Rientjes <rientjes@google.com> Cc: Gleb Natapov <gleb@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: J. Bruce Fields <bfields@redhat.com> Cc: Jan Kara <jack@suse.cz> Cc: Jerome Glisse <jglisse@redhat.com> Cc: John Stultz <john.stultz@linaro.org> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Kent Overstreet <koverstreet@google.com> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Steven Whitehouse <swhiteho@redhat.com> Cc: Thomas Hellstrom <thellstrom@vmware.com> Cc: Trond Myklebust <Trond.Myklebust@netapp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:17:54 +08:00
static long get_nr_dentry_unused(void)
{
int i;
long sum = 0;
for_each_possible_cpu(i)
sum += per_cpu(nr_dentry_unused, i);
return sum < 0 ? 0 : sum;
}
int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
size_t *lenp, loff_t *ppos)
{
fs: use fast counters for vfs caches percpu_counter library generates quite nasty code, so unless you need to dynamically allocate counters or take fast approximate value, a simple per cpu set of counters is much better. The percpu_counter can never be made to work as well, because it has an indirection from pointer to percpu memory, and it can't use direct this_cpu_inc interfaces because it doesn't use static PER_CPU data, so code will always be worse. In the fastpath, it is the difference between this: incl %gs:nr_dentry # nr_dentry and this: movl percpu_counter_batch(%rip), %edx # percpu_counter_batch, movl $1, %esi #, movq $nr_dentry, %rdi #, call __percpu_counter_add # (plus I clobber registers) __percpu_counter_add: pushq %rbp # movq %rsp, %rbp #, subq $32, %rsp #, movq %rbx, -24(%rbp) #, movq %r12, -16(%rbp) #, movq %r13, -8(%rbp) #, movq %rdi, %rbx # fbc, fbc #APP # 216 "/home/npiggin/usr/src/linux-2.6/arch/x86/include/asm/thread_info.h" 1 movq %gs:kernel_stack,%rax #, pfo_ret__ # 0 "" 2 #NO_APP incl -8124(%rax) # <variable>.preempt_count movq 32(%rdi), %r12 # <variable>.counters, tcp_ptr__ #APP # 78 "lib/percpu_counter.c" 1 add %gs:this_cpu_off, %r12 # this_cpu_off, tcp_ptr__ # 0 "" 2 #NO_APP movslq (%r12),%r13 #* tcp_ptr__, tmp73 movslq %edx,%rax # batch, batch addq %rsi, %r13 # amount, count cmpq %rax, %r13 # batch, count jge .L27 #, negl %edx # tmp76 movslq %edx,%rdx # tmp76, tmp77 cmpq %rdx, %r13 # tmp77, count jg .L28 #, .L27: movq %rbx, %rdi # fbc, call _raw_spin_lock # addq %r13, 8(%rbx) # count, <variable>.count movq %rbx, %rdi # fbc, movl $0, (%r12) #,* tcp_ptr__ call _raw_spin_unlock # .L29: #APP # 216 "/home/npiggin/usr/src/linux-2.6/arch/x86/include/asm/thread_info.h" 1 movq %gs:kernel_stack,%rax #, pfo_ret__ # 0 "" 2 #NO_APP decl -8124(%rax) # <variable>.preempt_count movq -8136(%rax), %rax #, D.14625 testb $8, %al #, D.14625 jne .L32 #, .L31: movq -24(%rbp), %rbx #, movq -16(%rbp), %r12 #, movq -8(%rbp), %r13 #, leave ret .p2align 4,,10 .p2align 3 .L28: movl %r13d, (%r12) # count,* jmp .L29 # .L32: call preempt_schedule # .p2align 4,,6 jmp .L31 # .size __percpu_counter_add, .-__percpu_counter_add .p2align 4,,15 Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:19 +08:00
dentry_stat.nr_dentry = get_nr_dentry();
dcache: convert dentry_stat.nr_unused to per-cpu counters Before we split up the dcache_lru_lock, the unused dentry counter needs to be made independent of the global dcache_lru_lock. Convert it to per-cpu counters to do this. Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Glauber Costa <glommer@openvz.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Acked-by: Mel Gorman <mgorman@suse.de> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com> Cc: Arve Hjønnevåg <arve@android.com> Cc: Carlos Maiolino <cmaiolino@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Chuck Lever <chuck.lever@oracle.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: David Rientjes <rientjes@google.com> Cc: Gleb Natapov <gleb@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: J. Bruce Fields <bfields@redhat.com> Cc: Jan Kara <jack@suse.cz> Cc: Jerome Glisse <jglisse@redhat.com> Cc: John Stultz <john.stultz@linaro.org> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Kent Overstreet <koverstreet@google.com> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Steven Whitehouse <swhiteho@redhat.com> Cc: Thomas Hellstrom <thellstrom@vmware.com> Cc: Trond Myklebust <Trond.Myklebust@netapp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:17:54 +08:00
dentry_stat.nr_unused = get_nr_dentry_unused();
fs: bump inode and dentry counters to long This series reworks our current object cache shrinking infrastructure in two main ways: * Noticing that a lot of users copy and paste their own version of LRU lists for objects, we put some effort in providing a generic version. It is modeled after the filesystem users: dentries, inodes, and xfs (for various tasks), but we expect that other users could benefit in the near future with little or no modification. Let us know if you have any issues. * The underlying list_lru being proposed automatically and transparently keeps the elements in per-node lists, and is able to manipulate the node lists individually. Given this infrastructure, we are able to modify the up-to-now hammer called shrink_slab to proceed with node-reclaim instead of always searching memory from all over like it has been doing. Per-node lru lists are also expected to lead to less contention in the lru locks on multi-node scans, since we are now no longer fighting for a global lock. The locks usually disappear from the profilers with this change. Although we have no official benchmarks for this version - be our guest to independently evaluate this - earlier versions of this series were performance tested (details at http://permalink.gmane.org/gmane.linux.kernel.mm/100537) yielding no visible performance regressions while yielding a better qualitative behavior in NUMA machines. With this infrastructure in place, we can use the list_lru entry point to provide memcg isolation and per-memcg targeted reclaim. Historically, those two pieces of work have been posted together. This version presents only the infrastructure work, deferring the memcg work for a later time, so we can focus on getting this part tested. You can see more about the history of such work at http://lwn.net/Articles/552769/ Dave Chinner (18): dcache: convert dentry_stat.nr_unused to per-cpu counters dentry: move to per-sb LRU locks dcache: remove dentries from LRU before putting on dispose list mm: new shrinker API shrinker: convert superblock shrinkers to new API list: add a new LRU list type inode: convert inode lru list to generic lru list code. dcache: convert to use new lru list infrastructure list_lru: per-node list infrastructure shrinker: add node awareness fs: convert inode and dentry shrinking to be node aware xfs: convert buftarg LRU to generic code xfs: rework buffer dispose list tracking xfs: convert dquot cache lru to list_lru fs: convert fs shrinkers to new scan/count API drivers: convert shrinkers to new count/scan API shrinker: convert remaining shrinkers to count/scan API shrinker: Kill old ->shrink API. Glauber Costa (7): fs: bump inode and dentry counters to long super: fix calculation of shrinkable objects for small numbers list_lru: per-node API vmscan: per-node deferred work i915: bail out earlier when shrinker cannot acquire mutex hugepage: convert huge zero page shrinker to new shrinker API list_lru: dynamically adjust node arrays This patch: There are situations in very large machines in which we can have a large quantity of dirty inodes, unused dentries, etc. This is particularly true when umounting a filesystem, where eventually since every live object will eventually be discarded. Dave Chinner reported a problem with this while experimenting with the shrinker revamp patchset. So we believe it is time for a change. This patch just moves int to longs. Machines where it matters should have a big long anyway. Signed-off-by: Glauber Costa <glommer@openvz.org> Cc: Dave Chinner <dchinner@redhat.com> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com> Cc: Arve Hjønnevåg <arve@android.com> Cc: Carlos Maiolino <cmaiolino@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Chuck Lever <chuck.lever@oracle.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: Dave Chinner <dchinner@redhat.com> Cc: David Rientjes <rientjes@google.com> Cc: Gleb Natapov <gleb@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: J. Bruce Fields <bfields@redhat.com> Cc: Jan Kara <jack@suse.cz> Cc: Jerome Glisse <jglisse@redhat.com> Cc: John Stultz <john.stultz@linaro.org> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Kent Overstreet <koverstreet@google.com> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Steven Whitehouse <swhiteho@redhat.com> Cc: Thomas Hellstrom <thellstrom@vmware.com> Cc: Trond Myklebust <Trond.Myklebust@netapp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:17:53 +08:00
return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
}
#endif
/*
* Compare 2 name strings, return 0 if they match, otherwise non-zero.
* The strings are both count bytes long, and count is non-zero.
*/
#ifdef CONFIG_DCACHE_WORD_ACCESS
#include <asm/word-at-a-time.h>
/*
* NOTE! 'cs' and 'scount' come from a dentry, so it has a
* aligned allocation for this particular component. We don't
* strictly need the load_unaligned_zeropad() safety, but it
* doesn't hurt either.
*
* In contrast, 'ct' and 'tcount' can be from a pathname, and do
* need the careful unaligned handling.
*/
static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
{
unsigned long a,b,mask;
for (;;) {
vfs: clean up __d_lookup_rcu() and dentry_cmp() interfaces The calling conventions for __d_lookup_rcu() and dentry_cmp() are annoying in different ways, and there is actually one single underlying reason for both of the annoyances. The fundamental reason is that we do the returned dentry sequence number check inside __d_lookup_rcu() instead of doing it in the caller. This results in two annoyances: - __d_lookup_rcu() now not only needs to return the dentry and the sequence number that goes along with the lookup, it also needs to return the inode pointer that was validated by that sequence number check. - and because we did the sequence number check early (to validate the name pointer and length) we also couldn't just pass the dentry itself to dentry_cmp(), we had to pass the counted string that contained the name. So that sequence number decision caused two separate ugly calling conventions. Both of these problems would be solved if we just did the sequence number check in the caller instead. There's only one caller, and that caller already has to do the sequence number check for the parent anyway, so just do that. That allows us to stop returning the dentry->d_inode in that in-out argument (pointer-to-pointer-to-inode), so we can make the inode argument just a regular input inode pointer. The caller can just load the inode from dentry->d_inode, and then do the sequence number check after that to make sure that it's synchronized with the name we looked up. And it allows us to just pass in the dentry to dentry_cmp(), which is what all the callers really wanted. Sure, dentry_cmp() has to be a bit careful about the dentry (which is not stable during RCU lookup), but that's actually very simple. And now that dentry_cmp() can clearly see that the first string argument is a dentry, we can use the direct word access for that, instead of the careful unaligned zero-padding. The dentry name is always properly aligned, since it is a single path component that is either embedded into the dentry itself, or was allocated with kmalloc() (see __d_alloc). Finally, this also uninlines the nasty slow-case for dentry comparisons: that one *does* need to do a sequence number check, since it will call in to the low-level filesystems, and we want to give those a stable inode pointer and path component length/start arguments. Doing an extra sequence check for that slow case is not a problem, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-05-05 05:59:14 +08:00
a = *(unsigned long *)cs;
b = load_unaligned_zeropad(ct);
if (tcount < sizeof(unsigned long))
break;
if (unlikely(a != b))
return 1;
cs += sizeof(unsigned long);
ct += sizeof(unsigned long);
tcount -= sizeof(unsigned long);
if (!tcount)
return 0;
}
mask = bytemask_from_count(tcount);
return unlikely(!!((a ^ b) & mask));
}
#else
static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
{
do {
if (*cs != *ct)
return 1;
cs++;
ct++;
tcount--;
} while (tcount);
return 0;
}
#endif
static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
{
2012-05-22 07:14:04 +08:00
const unsigned char *cs;
/*
* Be careful about RCU walk racing with rename:
* use ACCESS_ONCE to fetch the name pointer.
*
* NOTE! Even if a rename will mean that the length
* was not loaded atomically, we don't care. The
* RCU walk will check the sequence count eventually,
* and catch it. And we won't overrun the buffer,
* because we're reading the name pointer atomically,
* and a dentry name is guaranteed to be properly
* terminated with a NUL byte.
*
* End result: even if 'len' is wrong, we'll exit
* early because the data cannot match (there can
* be no NUL in the ct/tcount data)
*/
2012-05-22 07:14:04 +08:00
cs = ACCESS_ONCE(dentry->d_name.name);
smp_read_barrier_depends();
return dentry_string_cmp(cs, ct, tcount);
}
Allow sharing external names after __d_move() * external dentry names get a small structure prepended to them (struct external_name). * it contains an atomic refcount, matching the number of struct dentry instances that have ->d_name.name pointing to that external name. The first thing free_dentry() does is decrementing refcount of external name, so the instances that are between the call of free_dentry() and RCU-delayed actual freeing do not contribute. * __d_move(x, y, false) makes the name of x equal to the name of y, external or not. If y has an external name, extra reference is grabbed and put into x->d_name.name. If x used to have an external name, the reference to the old name is dropped and, should it reach zero, freeing is scheduled via kfree_rcu(). * free_dentry() in dentry with external name decrements the refcount of that name and, should it reach zero, does RCU-delayed call that will free both the dentry and external name. Otherwise it does what it used to do, except that __d_free() doesn't even look at ->d_name.name; it simply frees the dentry. All non-RCU accesses to dentry external name are safe wrt freeing since they all should happen before free_dentry() is called. RCU accesses might run into a dentry seen by free_dentry() or into an old name that got already dropped by __d_move(); however, in both cases dentry must have been alive and refer to that name at some point after we'd done rcu_read_lock(), which means that any freeing must be still pending. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2014-09-30 02:54:27 +08:00
struct external_name {
union {
atomic_t count;
struct rcu_head head;
} u;
unsigned char name[];
};
static inline struct external_name *external_name(struct dentry *dentry)
{
return container_of(dentry->d_name.name, struct external_name, name[0]);
}
static void __d_free(struct rcu_head *head)
{
struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
Allow sharing external names after __d_move() * external dentry names get a small structure prepended to them (struct external_name). * it contains an atomic refcount, matching the number of struct dentry instances that have ->d_name.name pointing to that external name. The first thing free_dentry() does is decrementing refcount of external name, so the instances that are between the call of free_dentry() and RCU-delayed actual freeing do not contribute. * __d_move(x, y, false) makes the name of x equal to the name of y, external or not. If y has an external name, extra reference is grabbed and put into x->d_name.name. If x used to have an external name, the reference to the old name is dropped and, should it reach zero, freeing is scheduled via kfree_rcu(). * free_dentry() in dentry with external name decrements the refcount of that name and, should it reach zero, does RCU-delayed call that will free both the dentry and external name. Otherwise it does what it used to do, except that __d_free() doesn't even look at ->d_name.name; it simply frees the dentry. All non-RCU accesses to dentry external name are safe wrt freeing since they all should happen before free_dentry() is called. RCU accesses might run into a dentry seen by free_dentry() or into an old name that got already dropped by __d_move(); however, in both cases dentry must have been alive and refer to that name at some point after we'd done rcu_read_lock(), which means that any freeing must be still pending. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2014-09-30 02:54:27 +08:00
kmem_cache_free(dentry_cache, dentry);
}
static void __d_free_external(struct rcu_head *head)
{
struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
kfree(external_name(dentry));
kmem_cache_free(dentry_cache, dentry);
}
static inline int dname_external(const struct dentry *dentry)
{
return dentry->d_name.name != dentry->d_iname;
}
static inline void __d_set_inode_and_type(struct dentry *dentry,
struct inode *inode,
unsigned type_flags)
{
unsigned flags;
dentry->d_inode = inode;
flags = READ_ONCE(dentry->d_flags);
flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
flags |= type_flags;
WRITE_ONCE(dentry->d_flags, flags);
}
static inline void __d_clear_type_and_inode(struct dentry *dentry)
{
unsigned flags = READ_ONCE(dentry->d_flags);
flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
WRITE_ONCE(dentry->d_flags, flags);
dentry->d_inode = NULL;
}
static void dentry_free(struct dentry *dentry)
{
WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
Allow sharing external names after __d_move() * external dentry names get a small structure prepended to them (struct external_name). * it contains an atomic refcount, matching the number of struct dentry instances that have ->d_name.name pointing to that external name. The first thing free_dentry() does is decrementing refcount of external name, so the instances that are between the call of free_dentry() and RCU-delayed actual freeing do not contribute. * __d_move(x, y, false) makes the name of x equal to the name of y, external or not. If y has an external name, extra reference is grabbed and put into x->d_name.name. If x used to have an external name, the reference to the old name is dropped and, should it reach zero, freeing is scheduled via kfree_rcu(). * free_dentry() in dentry with external name decrements the refcount of that name and, should it reach zero, does RCU-delayed call that will free both the dentry and external name. Otherwise it does what it used to do, except that __d_free() doesn't even look at ->d_name.name; it simply frees the dentry. All non-RCU accesses to dentry external name are safe wrt freeing since they all should happen before free_dentry() is called. RCU accesses might run into a dentry seen by free_dentry() or into an old name that got already dropped by __d_move(); however, in both cases dentry must have been alive and refer to that name at some point after we'd done rcu_read_lock(), which means that any freeing must be still pending. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2014-09-30 02:54:27 +08:00
if (unlikely(dname_external(dentry))) {
struct external_name *p = external_name(dentry);
if (likely(atomic_dec_and_test(&p->u.count))) {
call_rcu(&dentry->d_u.d_rcu, __d_free_external);
return;
}
}
/* if dentry was never visible to RCU, immediate free is OK */
if (!(dentry->d_flags & DCACHE_RCUACCESS))
__d_free(&dentry->d_u.d_rcu);
else
call_rcu(&dentry->d_u.d_rcu, __d_free);
}
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
/**
* dentry_rcuwalk_invalidate - invalidate in-progress rcu-walk lookups
* @dentry: the target dentry
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
* After this call, in-progress rcu-walk path lookup will fail. This
* should be called after unhashing, and after changing d_inode (if
* the dentry has not already been unhashed).
*/
static inline void dentry_rcuwalk_invalidate(struct dentry *dentry)
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
{
lockdep_assert_held(&dentry->d_lock);
/* Go through am invalidation barrier */
write_seqcount_invalidate(&dentry->d_seq);
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
}
/*
* Release the dentry's inode, using the filesystem
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
* d_iput() operation if defined. Dentry has no refcount
* and is unhashed.
*/
static void dentry_iput(struct dentry * dentry)
__releases(dentry->d_lock)
__releases(dentry->d_inode->i_lock)
{
struct inode *inode = dentry->d_inode;
if (inode) {
__d_clear_type_and_inode(dentry);
hlist_del_init(&dentry->d_u.d_alias);
spin_unlock(&dentry->d_lock);
spin_unlock(&inode->i_lock);
if (!inode->i_nlink)
fsnotify_inoderemove(inode);
if (dentry->d_op && dentry->d_op->d_iput)
dentry->d_op->d_iput(dentry, inode);
else
iput(inode);
} else {
spin_unlock(&dentry->d_lock);
}
}
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
/*
* Release the dentry's inode, using the filesystem
* d_iput() operation if defined. dentry remains in-use.
*/
static void dentry_unlink_inode(struct dentry * dentry)
__releases(dentry->d_lock)
__releases(dentry->d_inode->i_lock)
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
{
struct inode *inode = dentry->d_inode;
raw_write_seqcount_begin(&dentry->d_seq);
__d_clear_type_and_inode(dentry);
hlist_del_init(&dentry->d_u.d_alias);
raw_write_seqcount_end(&dentry->d_seq);
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
spin_unlock(&dentry->d_lock);
spin_unlock(&inode->i_lock);
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
if (!inode->i_nlink)
fsnotify_inoderemove(inode);
if (dentry->d_op && dentry->d_op->d_iput)
dentry->d_op->d_iput(dentry, inode);
else
iput(inode);
}
/*
* The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
* is in use - which includes both the "real" per-superblock
* LRU list _and_ the DCACHE_SHRINK_LIST use.
*
* The DCACHE_SHRINK_LIST bit is set whenever the dentry is
* on the shrink list (ie not on the superblock LRU list).
*
* The per-cpu "nr_dentry_unused" counters are updated with
* the DCACHE_LRU_LIST bit.
*
* These helper functions make sure we always follow the
* rules. d_lock must be held by the caller.
*/
#define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
static void d_lru_add(struct dentry *dentry)
{
D_FLAG_VERIFY(dentry, 0);
dentry->d_flags |= DCACHE_LRU_LIST;
this_cpu_inc(nr_dentry_unused);
WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
}
static void d_lru_del(struct dentry *dentry)
{
D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
dentry->d_flags &= ~DCACHE_LRU_LIST;
this_cpu_dec(nr_dentry_unused);
WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
}
static void d_shrink_del(struct dentry *dentry)
{
D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
list_del_init(&dentry->d_lru);
dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
this_cpu_dec(nr_dentry_unused);
}
static void d_shrink_add(struct dentry *dentry, struct list_head *list)
{
D_FLAG_VERIFY(dentry, 0);
list_add(&dentry->d_lru, list);
dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
this_cpu_inc(nr_dentry_unused);
}
/*
* These can only be called under the global LRU lock, ie during the
* callback for freeing the LRU list. "isolate" removes it from the
* LRU lists entirely, while shrink_move moves it to the indicated
* private list.
*/
list_lru: add helpers to isolate items Currently, the isolate callback passed to the list_lru_walk family of functions is supposed to just delete an item from the list upon returning LRU_REMOVED or LRU_REMOVED_RETRY, while nr_items counter is fixed by __list_lru_walk_one after the callback returns. Since the callback is allowed to drop the lock after removing an item (it has to return LRU_REMOVED_RETRY then), the nr_items can be less than the actual number of elements on the list even if we check them under the lock. This makes it difficult to move items from one list_lru_one to another, which is required for per-memcg list_lru reparenting - we can't just splice the lists, we have to move entries one by one. This patch therefore introduces helpers that must be used by callback functions to isolate items instead of raw list_del/list_move. These are list_lru_isolate and list_lru_isolate_move. They not only remove the entry from the list, but also fix the nr_items counter, making sure nr_items always reflects the actual number of elements on the list if checked under the appropriate lock. Signed-off-by: Vladimir Davydov <vdavydov@parallels.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@suse.cz> Cc: Tejun Heo <tj@kernel.org> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Dave Chinner <david@fromorbit.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-02-13 06:59:35 +08:00
static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
{
D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
dentry->d_flags &= ~DCACHE_LRU_LIST;
this_cpu_dec(nr_dentry_unused);
list_lru: add helpers to isolate items Currently, the isolate callback passed to the list_lru_walk family of functions is supposed to just delete an item from the list upon returning LRU_REMOVED or LRU_REMOVED_RETRY, while nr_items counter is fixed by __list_lru_walk_one after the callback returns. Since the callback is allowed to drop the lock after removing an item (it has to return LRU_REMOVED_RETRY then), the nr_items can be less than the actual number of elements on the list even if we check them under the lock. This makes it difficult to move items from one list_lru_one to another, which is required for per-memcg list_lru reparenting - we can't just splice the lists, we have to move entries one by one. This patch therefore introduces helpers that must be used by callback functions to isolate items instead of raw list_del/list_move. These are list_lru_isolate and list_lru_isolate_move. They not only remove the entry from the list, but also fix the nr_items counter, making sure nr_items always reflects the actual number of elements on the list if checked under the appropriate lock. Signed-off-by: Vladimir Davydov <vdavydov@parallels.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@suse.cz> Cc: Tejun Heo <tj@kernel.org> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Dave Chinner <david@fromorbit.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-02-13 06:59:35 +08:00
list_lru_isolate(lru, &dentry->d_lru);
}
list_lru: add helpers to isolate items Currently, the isolate callback passed to the list_lru_walk family of functions is supposed to just delete an item from the list upon returning LRU_REMOVED or LRU_REMOVED_RETRY, while nr_items counter is fixed by __list_lru_walk_one after the callback returns. Since the callback is allowed to drop the lock after removing an item (it has to return LRU_REMOVED_RETRY then), the nr_items can be less than the actual number of elements on the list even if we check them under the lock. This makes it difficult to move items from one list_lru_one to another, which is required for per-memcg list_lru reparenting - we can't just splice the lists, we have to move entries one by one. This patch therefore introduces helpers that must be used by callback functions to isolate items instead of raw list_del/list_move. These are list_lru_isolate and list_lru_isolate_move. They not only remove the entry from the list, but also fix the nr_items counter, making sure nr_items always reflects the actual number of elements on the list if checked under the appropriate lock. Signed-off-by: Vladimir Davydov <vdavydov@parallels.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@suse.cz> Cc: Tejun Heo <tj@kernel.org> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Dave Chinner <david@fromorbit.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-02-13 06:59:35 +08:00
static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
struct list_head *list)
{
D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
dentry->d_flags |= DCACHE_SHRINK_LIST;
list_lru: add helpers to isolate items Currently, the isolate callback passed to the list_lru_walk family of functions is supposed to just delete an item from the list upon returning LRU_REMOVED or LRU_REMOVED_RETRY, while nr_items counter is fixed by __list_lru_walk_one after the callback returns. Since the callback is allowed to drop the lock after removing an item (it has to return LRU_REMOVED_RETRY then), the nr_items can be less than the actual number of elements on the list even if we check them under the lock. This makes it difficult to move items from one list_lru_one to another, which is required for per-memcg list_lru reparenting - we can't just splice the lists, we have to move entries one by one. This patch therefore introduces helpers that must be used by callback functions to isolate items instead of raw list_del/list_move. These are list_lru_isolate and list_lru_isolate_move. They not only remove the entry from the list, but also fix the nr_items counter, making sure nr_items always reflects the actual number of elements on the list if checked under the appropriate lock. Signed-off-by: Vladimir Davydov <vdavydov@parallels.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@suse.cz> Cc: Tejun Heo <tj@kernel.org> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Dave Chinner <david@fromorbit.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-02-13 06:59:35 +08:00
list_lru_isolate_move(lru, &dentry->d_lru, list);
}
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries [Summary] Split LRU-list of unused dentries to one per superblock to avoid soft lock up during NFS mounts and remounting of any filesystem. Previously I posted here: http://lkml.org/lkml/2008/3/5/590 [Descriptions] - background dentry_unused is a list of dentries which are not referenced. dentry_unused grows up when references on directories or files are released. This list can be very long if there is huge free memory. - the problem When shrink_dcache_sb() is called, it scans all dentry_unused linearly under spin_lock(), and if dentry->d_sb is differnt from given superblock, scan next dentry. This scan costs very much if there are many entries, and very ineffective if there are many superblocks. IOW, When we need to shrink unused dentries on one dentry, but scans unused dentries on all superblocks in the system. For example, we scan 500 dentries to unmount a filesystem, but scans 1,000,000 or more unused dentries on other superblocks. In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink unused dentries on NFS, but scans 100,000,000 unused dentries on superblocks in the system such as local ext3 filesystems. I hear NFS mounting took 1 min on some system in use. * : NFS uses virtual filesystem in rpc layer, so NFS is affected by this problem. 100,000,000 is possible number on large systems. Per-superblock LRU of unused dentried can reduce the cost in reasonable manner. - How to fix I found this problem is solved by David Chinner's "Per-superblock unused dentry LRU lists V3"(1), so I rebase it and add some fix to reclaim with fairness, which is in Andrew Morton's comments(2). 1) http://lkml.org/lkml/2006/5/25/318 2) http://lkml.org/lkml/2006/5/25/320 Split LRU-list of unused dentries to each superblocks. Then, NFS mounting will check dentries under a superblock instead of all. But this spliting will break LRU of dentry-unused. So, I've attempted to make reclaim unused dentrins with fairness by calculate number of dentries to scan on this sb based on following way number of dentries to scan on this sb = count * (number of dentries on this sb / number of dentries in the machine) - ToDo - I have to measuring performance number and do stress tests. - When unmount occurs during prune_dcache(), scanning on same superblock, It is unable to reach next superblock because it is gone away. We restart scannig superblock from first one, it causes unfairness of reclaim unused dentries on first superblock. But I think this happens very rarely. - Test Results Result on 6GB boxes with excessive unused dentries. Without patch: $ cat /proc/sys/fs/dentry-state 10181835 10180203 45 0 0 0 # mount -t nfs 10.124.60.70:/work/kernel-src nfs real 0m1.830s user 0m0.001s sys 0m1.653s With this patch: $ cat /proc/sys/fs/dentry-state 10236610 10234751 45 0 0 0 # mount -t nfs 10.124.60.70:/work/kernel-src nfs real 0m0.106s user 0m0.002s sys 0m0.032s [akpm@linux-foundation.org: fix comments] Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com> Cc: Neil Brown <neilb@suse.de> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Cc: David Chinner <dgc@sgi.com> Cc: "J. Bruce Fields" <bfields@fieldses.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
/*
* dentry_lru_(add|del)_list) must be called with d_lock held.
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries [Summary] Split LRU-list of unused dentries to one per superblock to avoid soft lock up during NFS mounts and remounting of any filesystem. Previously I posted here: http://lkml.org/lkml/2008/3/5/590 [Descriptions] - background dentry_unused is a list of dentries which are not referenced. dentry_unused grows up when references on directories or files are released. This list can be very long if there is huge free memory. - the problem When shrink_dcache_sb() is called, it scans all dentry_unused linearly under spin_lock(), and if dentry->d_sb is differnt from given superblock, scan next dentry. This scan costs very much if there are many entries, and very ineffective if there are many superblocks. IOW, When we need to shrink unused dentries on one dentry, but scans unused dentries on all superblocks in the system. For example, we scan 500 dentries to unmount a filesystem, but scans 1,000,000 or more unused dentries on other superblocks. In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink unused dentries on NFS, but scans 100,000,000 unused dentries on superblocks in the system such as local ext3 filesystems. I hear NFS mounting took 1 min on some system in use. * : NFS uses virtual filesystem in rpc layer, so NFS is affected by this problem. 100,000,000 is possible number on large systems. Per-superblock LRU of unused dentried can reduce the cost in reasonable manner. - How to fix I found this problem is solved by David Chinner's "Per-superblock unused dentry LRU lists V3"(1), so I rebase it and add some fix to reclaim with fairness, which is in Andrew Morton's comments(2). 1) http://lkml.org/lkml/2006/5/25/318 2) http://lkml.org/lkml/2006/5/25/320 Split LRU-list of unused dentries to each superblocks. Then, NFS mounting will check dentries under a superblock instead of all. But this spliting will break LRU of dentry-unused. So, I've attempted to make reclaim unused dentrins with fairness by calculate number of dentries to scan on this sb based on following way number of dentries to scan on this sb = count * (number of dentries on this sb / number of dentries in the machine) - ToDo - I have to measuring performance number and do stress tests. - When unmount occurs during prune_dcache(), scanning on same superblock, It is unable to reach next superblock because it is gone away. We restart scannig superblock from first one, it causes unfairness of reclaim unused dentries on first superblock. But I think this happens very rarely. - Test Results Result on 6GB boxes with excessive unused dentries. Without patch: $ cat /proc/sys/fs/dentry-state 10181835 10180203 45 0 0 0 # mount -t nfs 10.124.60.70:/work/kernel-src nfs real 0m1.830s user 0m0.001s sys 0m1.653s With this patch: $ cat /proc/sys/fs/dentry-state 10236610 10234751 45 0 0 0 # mount -t nfs 10.124.60.70:/work/kernel-src nfs real 0m0.106s user 0m0.002s sys 0m0.032s [akpm@linux-foundation.org: fix comments] Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com> Cc: Neil Brown <neilb@suse.de> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Cc: David Chinner <dgc@sgi.com> Cc: "J. Bruce Fields" <bfields@fieldses.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
*/
static void dentry_lru_add(struct dentry *dentry)
{
if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
d_lru_add(dentry);
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries [Summary] Split LRU-list of unused dentries to one per superblock to avoid soft lock up during NFS mounts and remounting of any filesystem. Previously I posted here: http://lkml.org/lkml/2008/3/5/590 [Descriptions] - background dentry_unused is a list of dentries which are not referenced. dentry_unused grows up when references on directories or files are released. This list can be very long if there is huge free memory. - the problem When shrink_dcache_sb() is called, it scans all dentry_unused linearly under spin_lock(), and if dentry->d_sb is differnt from given superblock, scan next dentry. This scan costs very much if there are many entries, and very ineffective if there are many superblocks. IOW, When we need to shrink unused dentries on one dentry, but scans unused dentries on all superblocks in the system. For example, we scan 500 dentries to unmount a filesystem, but scans 1,000,000 or more unused dentries on other superblocks. In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink unused dentries on NFS, but scans 100,000,000 unused dentries on superblocks in the system such as local ext3 filesystems. I hear NFS mounting took 1 min on some system in use. * : NFS uses virtual filesystem in rpc layer, so NFS is affected by this problem. 100,000,000 is possible number on large systems. Per-superblock LRU of unused dentried can reduce the cost in reasonable manner. - How to fix I found this problem is solved by David Chinner's "Per-superblock unused dentry LRU lists V3"(1), so I rebase it and add some fix to reclaim with fairness, which is in Andrew Morton's comments(2). 1) http://lkml.org/lkml/2006/5/25/318 2) http://lkml.org/lkml/2006/5/25/320 Split LRU-list of unused dentries to each superblocks. Then, NFS mounting will check dentries under a superblock instead of all. But this spliting will break LRU of dentry-unused. So, I've attempted to make reclaim unused dentrins with fairness by calculate number of dentries to scan on this sb based on following way number of dentries to scan on this sb = count * (number of dentries on this sb / number of dentries in the machine) - ToDo - I have to measuring performance number and do stress tests. - When unmount occurs during prune_dcache(), scanning on same superblock, It is unable to reach next superblock because it is gone away. We restart scannig superblock from first one, it causes unfairness of reclaim unused dentries on first superblock. But I think this happens very rarely. - Test Results Result on 6GB boxes with excessive unused dentries. Without patch: $ cat /proc/sys/fs/dentry-state 10181835 10180203 45 0 0 0 # mount -t nfs 10.124.60.70:/work/kernel-src nfs real 0m1.830s user 0m0.001s sys 0m1.653s With this patch: $ cat /proc/sys/fs/dentry-state 10236610 10234751 45 0 0 0 # mount -t nfs 10.124.60.70:/work/kernel-src nfs real 0m0.106s user 0m0.002s sys 0m0.032s [akpm@linux-foundation.org: fix comments] Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com> Cc: Neil Brown <neilb@suse.de> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Cc: David Chinner <dgc@sgi.com> Cc: "J. Bruce Fields" <bfields@fieldses.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
}
/**
* d_drop - drop a dentry
* @dentry: dentry to drop
*
* d_drop() unhashes the entry from the parent dentry hashes, so that it won't
* be found through a VFS lookup any more. Note that this is different from
* deleting the dentry - d_delete will try to mark the dentry negative if
* possible, giving a successful _negative_ lookup, while d_drop will
* just make the cache lookup fail.
*
* d_drop() is used mainly for stuff that wants to invalidate a dentry for some
* reason (NFS timeouts or autofs deletes).
*
* __d_drop requires dentry->d_lock.
*/
void __d_drop(struct dentry *dentry)
{
vfs: get rid of insane dentry hashing rules The dentry hashing rules have been really quite complicated for a long while, in odd ways. That made functions like __d_drop() very fragile and non-obvious. In particular, whether a dentry was hashed or not was indicated with an explicit DCACHE_UNHASHED bit. That's despite the fact that the hash abstraction that the dentries use actually have a 'is this entry hashed or not' model (which is a simple test of the 'pprev' pointer). The reason that was done is because we used the normal 'is this entry unhashed' model to mark whether the dentry had _ever_ been hashed in the dentry hash tables, and that logic goes back many years (commit b3423415fbc2: "dcache: avoid RCU for never-hashed dentries"). That, in turn, meant that __d_drop had totally different unhashing logic for the dentry hash table case and for the anonymous dcache case, because in order to use the "is this dentry hashed" logic as a flag for whether it had ever been on the RCU hash table, we had to unhash such a dentry differently so that we'd never think that it wasn't 'unhashed' and wouldn't be free'd correctly. That's just insane. It made the logic really hard to follow, when there were two different kinds of "unhashed" states, and one of them (the one that used "list_bl_unhashed()") really had nothing at all to do with being unhashed per se, but with a very subtle lifetime rule instead. So turn all of it around, and make it logical. Instead of having a DENTRY_UNHASHED bit in d_flags to indicate whether the dentry is on the hash chains or not, use the hash chain unhashed logic for that. Suddenly "d_unhashed()" just uses "list_bl_unhashed()", and everything makes sense. And for the lifetime rule, just use an explicit DENTRY_RCUACCEES bit. If we ever insert the dentry into the dentry hash table so that it is visible to RCU lookup, we mark it DENTRY_RCUACCESS to show that it now needs the RCU lifetime rules. Now suddently that test at dentry free time makes sense too. And because unhashing now is sane and doesn't depend on where the dentry got unhashed from (because the dentry hash chain details doesn't have some subtle side effects), we can re-unify the __d_drop() logic and use common code for the unhashing. Also fix one more open-coded hash chain bit_spin_lock() that I missed in the previous chain locking cleanup commit. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-04-24 22:58:46 +08:00
if (!d_unhashed(dentry)) {
struct hlist_bl_head *b;
dcache: use IS_ROOT to decide where dentry is hashed Every hashed dentry is either hashed in the dentry_hashtable, or a superblock's s_anon list. __d_drop() assumes it can determine which is the case by checking DCACHE_DISCONNECTED; this is not true. It is true that when DCACHE_DISCONNECTED is cleared, the dentry is not only hashed on dentry_hashtable, but is fully connected to its parents back to the root. But the converse is *not* true: fs/exportfs/expfs.c:reconnect_path() attempts to connect a directory (found by filehandle lookup) back to root by ascending to parents and performing lookups one at a time. It does not clear DCACHE_DISCONNECTED until it's done, and that is not at all an atomic process. In particular, it is possible for DCACHE_DISCONNECTED to be set on a dentry which is hashed on the dentry_hashtable. Instead, use IS_ROOT() to check which hash chain a dentry is on. This *does* work: Dentries are hashed only by: - d_obtain_alias, which adds an IS_ROOT() dentry to sb_anon. - __d_rehash, called by _d_rehash: hashes to the dentry's parent, and all callers of _d_rehash appear to have d_parent set to a "real" parent. - __d_rehash, called by __d_move: rehashes the moved dentry to hash chain determined by target, and assigns target's d_parent to its d_parent, before dropping the dentry's d_lock. Therefore I believe it's safe for a holder of a dentry's d_lock to assume that it is hashed on sb_anon if and only if IS_ROOT(dentry) is true. I believe the incorrect assumption about DCACHE_DISCONNECTED was originally introduced by ceb5bdc2d246 "fs: dcache per-bucket dcache hash locking". Also add a comment while we're here. Cc: Nick Piggin <npiggin@kernel.dk> Acked-by: Christoph Hellwig <hch@infradead.org> Reviewed-by: NeilBrown <neilb@suse.de> Signed-off-by: J. Bruce Fields <bfields@redhat.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2012-06-29 00:10:55 +08:00
/*
* Hashed dentries are normally on the dentry hashtable,
* with the exception of those newly allocated by
* d_obtain_alias, which are always IS_ROOT:
*/
if (unlikely(IS_ROOT(dentry)))
b = &dentry->d_sb->s_anon;
else
b = d_hash(dentry->d_parent, dentry->d_name.hash);
hlist_bl_lock(b);
__hlist_bl_del(&dentry->d_hash);
dentry->d_hash.pprev = NULL;
hlist_bl_unlock(b);
dentry_rcuwalk_invalidate(dentry);
}
}
EXPORT_SYMBOL(__d_drop);
void d_drop(struct dentry *dentry)
{
spin_lock(&dentry->d_lock);
__d_drop(dentry);
spin_unlock(&dentry->d_lock);
}
EXPORT_SYMBOL(d_drop);
static void __dentry_kill(struct dentry *dentry)
{
struct dentry *parent = NULL;
bool can_free = true;
if (!IS_ROOT(dentry))
parent = dentry->d_parent;
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
vfs: use lockred "dead" flag to mark unrecoverably dead dentries This simplifies the RCU to refcounting code in particular. I was originally intending to leave this for later, but walking through all the dput() logic (see previous commit), I realized that the dput() "might_sleep()" check was misleadingly weak. And I removed it as misleading, both for performance profiling and for debugging. However, the might_sleep() debugging case is actually true: the final dput() can indeed sleep, if the inode of the dentry that you are releasing ends up sleeping at iput time (see dentry_iput()). So the problem with the might_sleep() in dput() wasn't that it wasn't true, it was that it wasn't actually testing and triggering on the interesting case. In particular, just about *any* dput() can indeed sleep, if you happen to race with another thread deleting the file in question, and you then lose the race to the be the last dput() for that file. But because it's a very rare race, the debugging code would never trigger it in practice. Why is this problematic? The new d_rcu_to_refcount() (see commit 15570086b590: "vfs: reimplement d_rcu_to_refcount() using lockref_get_or_lock()") does a dput() for the failure case, and it does it under the RCU lock. So potentially sleeping really is a bug. But there's no way I'm going to fix this with the previous complicated "lockref_get_or_lock()" interface. And rather than revert to the old and crufty nested dentry locking code (which did get this right by delaying the reference count updates until they were verified to be safe), let's make forward progress. Cc: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-09 04:46:52 +08:00
/*
* The dentry is now unrecoverably dead to the world.
*/
lockref_mark_dead(&dentry->d_lockref);
/*
* inform the fs via d_prune that this dentry is about to be
* unhashed and destroyed.
*/
if (dentry->d_flags & DCACHE_OP_PRUNE)
dentry->d_op->d_prune(dentry);
if (dentry->d_flags & DCACHE_LRU_LIST) {
if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
d_lru_del(dentry);
}
/* if it was on the hash then remove it */
__d_drop(dentry);
__list_del_entry(&dentry->d_child);
/*
* Inform d_walk() that we are no longer attached to the
* dentry tree
*/
dentry->d_flags |= DCACHE_DENTRY_KILLED;
if (parent)
spin_unlock(&parent->d_lock);
dentry_iput(dentry);
/*
* dentry_iput drops the locks, at which point nobody (except
* transient RCU lookups) can reach this dentry.
*/
BUG_ON(dentry->d_lockref.count > 0);
this_cpu_dec(nr_dentry);
if (dentry->d_op && dentry->d_op->d_release)
dentry->d_op->d_release(dentry);
spin_lock(&dentry->d_lock);
if (dentry->d_flags & DCACHE_SHRINK_LIST) {
dentry->d_flags |= DCACHE_MAY_FREE;
can_free = false;
}
spin_unlock(&dentry->d_lock);
if (likely(can_free))
dentry_free(dentry);
}
/*
* Finish off a dentry we've decided to kill.
* dentry->d_lock must be held, returns with it unlocked.
* If ref is non-zero, then decrement the refcount too.
* Returns dentry requiring refcount drop, or NULL if we're done.
*/
static struct dentry *dentry_kill(struct dentry *dentry)
__releases(dentry->d_lock)
{
struct inode *inode = dentry->d_inode;
struct dentry *parent = NULL;
if (inode && unlikely(!spin_trylock(&inode->i_lock)))
goto failed;
if (!IS_ROOT(dentry)) {
parent = dentry->d_parent;
if (unlikely(!spin_trylock(&parent->d_lock))) {
if (inode)
spin_unlock(&inode->i_lock);
goto failed;
}
}
__dentry_kill(dentry);
return parent;
failed:
spin_unlock(&dentry->d_lock);
cpu_relax();
return dentry; /* try again with same dentry */
}
shrink_dentry_list(): take parent's ->d_lock earlier The cause of livelocks there is that we are taking ->d_lock on dentry and its parent in the wrong order, forcing us to use trylock on the parent's one. d_walk() takes them in the right order, and unfortunately it's not hard to create a situation when shrink_dentry_list() can't make progress since trylock keeps failing, and shrink_dcache_parent() or check_submounts_and_drop() keeps calling d_walk() disrupting the very shrink_dentry_list() it's waiting for. Solution is straightforward - if that trylock fails, let's unlock the dentry itself and take locks in the right order. We need to stabilize ->d_parent without holding ->d_lock, but that's doable using RCU. And we'd better do that in the very beginning of the loop in shrink_dentry_list(), since the checks on refcount, etc. would need to be redone anyway. That deals with a half of the problem - killing dentries on the shrink list itself. Another one (dropping their parents) is in the next commit. locking parent is interesting - it would be easy to do rcu_read_lock(), lock whatever we think is a parent, lock dentry itself and check if the parent is still the right one. Except that we need to check that *before* locking the dentry, or we are risking taking ->d_lock out of order. Fortunately, once the D1 is locked, we can check if D2->d_parent is equal to D1 without the need to lock D2; D2->d_parent can start or stop pointing to D1 only under D1->d_lock, so taking D1->d_lock is enough. In other words, the right solution is rcu_read_lock/lock what looks like parent right now/check if it's still our parent/rcu_read_unlock/lock the child. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2014-05-29 20:54:52 +08:00
static inline struct dentry *lock_parent(struct dentry *dentry)
{
struct dentry *parent = dentry->d_parent;
if (IS_ROOT(dentry))
return NULL;
if (unlikely(dentry->d_lockref.count < 0))
return NULL;
shrink_dentry_list(): take parent's ->d_lock earlier The cause of livelocks there is that we are taking ->d_lock on dentry and its parent in the wrong order, forcing us to use trylock on the parent's one. d_walk() takes them in the right order, and unfortunately it's not hard to create a situation when shrink_dentry_list() can't make progress since trylock keeps failing, and shrink_dcache_parent() or check_submounts_and_drop() keeps calling d_walk() disrupting the very shrink_dentry_list() it's waiting for. Solution is straightforward - if that trylock fails, let's unlock the dentry itself and take locks in the right order. We need to stabilize ->d_parent without holding ->d_lock, but that's doable using RCU. And we'd better do that in the very beginning of the loop in shrink_dentry_list(), since the checks on refcount, etc. would need to be redone anyway. That deals with a half of the problem - killing dentries on the shrink list itself. Another one (dropping their parents) is in the next commit. locking parent is interesting - it would be easy to do rcu_read_lock(), lock whatever we think is a parent, lock dentry itself and check if the parent is still the right one. Except that we need to check that *before* locking the dentry, or we are risking taking ->d_lock out of order. Fortunately, once the D1 is locked, we can check if D2->d_parent is equal to D1 without the need to lock D2; D2->d_parent can start or stop pointing to D1 only under D1->d_lock, so taking D1->d_lock is enough. In other words, the right solution is rcu_read_lock/lock what looks like parent right now/check if it's still our parent/rcu_read_unlock/lock the child. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2014-05-29 20:54:52 +08:00
if (likely(spin_trylock(&parent->d_lock)))
return parent;
rcu_read_lock();
spin_unlock(&dentry->d_lock);
shrink_dentry_list(): take parent's ->d_lock earlier The cause of livelocks there is that we are taking ->d_lock on dentry and its parent in the wrong order, forcing us to use trylock on the parent's one. d_walk() takes them in the right order, and unfortunately it's not hard to create a situation when shrink_dentry_list() can't make progress since trylock keeps failing, and shrink_dcache_parent() or check_submounts_and_drop() keeps calling d_walk() disrupting the very shrink_dentry_list() it's waiting for. Solution is straightforward - if that trylock fails, let's unlock the dentry itself and take locks in the right order. We need to stabilize ->d_parent without holding ->d_lock, but that's doable using RCU. And we'd better do that in the very beginning of the loop in shrink_dentry_list(), since the checks on refcount, etc. would need to be redone anyway. That deals with a half of the problem - killing dentries on the shrink list itself. Another one (dropping their parents) is in the next commit. locking parent is interesting - it would be easy to do rcu_read_lock(), lock whatever we think is a parent, lock dentry itself and check if the parent is still the right one. Except that we need to check that *before* locking the dentry, or we are risking taking ->d_lock out of order. Fortunately, once the D1 is locked, we can check if D2->d_parent is equal to D1 without the need to lock D2; D2->d_parent can start or stop pointing to D1 only under D1->d_lock, so taking D1->d_lock is enough. In other words, the right solution is rcu_read_lock/lock what looks like parent right now/check if it's still our parent/rcu_read_unlock/lock the child. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2014-05-29 20:54:52 +08:00
again:
parent = ACCESS_ONCE(dentry->d_parent);
spin_lock(&parent->d_lock);
/*
* We can't blindly lock dentry until we are sure
* that we won't violate the locking order.
* Any changes of dentry->d_parent must have
* been done with parent->d_lock held, so
* spin_lock() above is enough of a barrier
* for checking if it's still our child.
*/
if (unlikely(parent != dentry->d_parent)) {
spin_unlock(&parent->d_lock);
goto again;
}
rcu_read_unlock();
if (parent != dentry)
spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
shrink_dentry_list(): take parent's ->d_lock earlier The cause of livelocks there is that we are taking ->d_lock on dentry and its parent in the wrong order, forcing us to use trylock on the parent's one. d_walk() takes them in the right order, and unfortunately it's not hard to create a situation when shrink_dentry_list() can't make progress since trylock keeps failing, and shrink_dcache_parent() or check_submounts_and_drop() keeps calling d_walk() disrupting the very shrink_dentry_list() it's waiting for. Solution is straightforward - if that trylock fails, let's unlock the dentry itself and take locks in the right order. We need to stabilize ->d_parent without holding ->d_lock, but that's doable using RCU. And we'd better do that in the very beginning of the loop in shrink_dentry_list(), since the checks on refcount, etc. would need to be redone anyway. That deals with a half of the problem - killing dentries on the shrink list itself. Another one (dropping their parents) is in the next commit. locking parent is interesting - it would be easy to do rcu_read_lock(), lock whatever we think is a parent, lock dentry itself and check if the parent is still the right one. Except that we need to check that *before* locking the dentry, or we are risking taking ->d_lock out of order. Fortunately, once the D1 is locked, we can check if D2->d_parent is equal to D1 without the need to lock D2; D2->d_parent can start or stop pointing to D1 only under D1->d_lock, so taking D1->d_lock is enough. In other words, the right solution is rcu_read_lock/lock what looks like parent right now/check if it's still our parent/rcu_read_unlock/lock the child. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2014-05-29 20:54:52 +08:00
else
parent = NULL;
return parent;
}
/*
* Try to do a lockless dput(), and return whether that was successful.
*
* If unsuccessful, we return false, having already taken the dentry lock.
*
* The caller needs to hold the RCU read lock, so that the dentry is
* guaranteed to stay around even if the refcount goes down to zero!
*/
static inline bool fast_dput(struct dentry *dentry)
{
int ret;
unsigned int d_flags;
/*
* If we have a d_op->d_delete() operation, we sould not
freeing unlinked file indefinitely delayed Normally opening a file, unlinking it and then closing will have the inode freed upon close() (provided that it's not otherwise busy and has no remaining links, of course). However, there's one case where that does *not* happen. Namely, if you open it by fhandle with cold dcache, then unlink() and close(). In normal case you get d_delete() in unlink(2) notice that dentry is busy and unhash it; on the final dput() it will be forcibly evicted from dcache, triggering iput() and inode removal. In this case, though, we end up with *two* dentries - disconnected (created by open-by-fhandle) and regular one (used by unlink()). The latter will have its reference to inode dropped just fine, but the former will not - it's considered hashed (it is on the ->s_anon list), so it will stay around until the memory pressure will finally do it in. As the result, we have the final iput() delayed indefinitely. It's trivial to reproduce - void flush_dcache(void) { system("mount -o remount,rw /"); } static char buf[20 * 1024 * 1024]; main() { int fd; union { struct file_handle f; char buf[MAX_HANDLE_SZ]; } x; int m; x.f.handle_bytes = sizeof(x); chdir("/root"); mkdir("foo", 0700); fd = open("foo/bar", O_CREAT | O_RDWR, 0600); close(fd); name_to_handle_at(AT_FDCWD, "foo/bar", &x.f, &m, 0); flush_dcache(); fd = open_by_handle_at(AT_FDCWD, &x.f, O_RDWR); unlink("foo/bar"); write(fd, buf, sizeof(buf)); system("df ."); /* 20Mb eaten */ close(fd); system("df ."); /* should've freed those 20Mb */ flush_dcache(); system("df ."); /* should be the same as #2 */ } will spit out something like Filesystem 1K-blocks Used Available Use% Mounted on /dev/root 322023 303843 1131 100% / Filesystem 1K-blocks Used Available Use% Mounted on /dev/root 322023 303843 1131 100% / Filesystem 1K-blocks Used Available Use% Mounted on /dev/root 322023 283282 21692 93% / - inode gets freed only when dentry is finally evicted (here we trigger than by remount; normally it would've happened in response to memory pressure hell knows when). Cc: stable@vger.kernel.org # v2.6.38+; earlier ones need s/kill_it/unhash_it/ Acked-by: J. Bruce Fields <bfields@fieldses.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2015-07-08 09:42:38 +08:00
* let the dentry count go to zero, so use "put_or_lock".
*/
if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
return lockref_put_or_lock(&dentry->d_lockref);
/*
* .. otherwise, we can try to just decrement the
* lockref optimistically.
*/
ret = lockref_put_return(&dentry->d_lockref);
/*
* If the lockref_put_return() failed due to the lock being held
* by somebody else, the fast path has failed. We will need to
* get the lock, and then check the count again.
*/
if (unlikely(ret < 0)) {
spin_lock(&dentry->d_lock);
if (dentry->d_lockref.count > 1) {
dentry->d_lockref.count--;
spin_unlock(&dentry->d_lock);
return 1;
}
return 0;
}
/*
* If we weren't the last ref, we're done.
*/
if (ret)
return 1;
/*
* Careful, careful. The reference count went down
* to zero, but we don't hold the dentry lock, so
* somebody else could get it again, and do another
* dput(), and we need to not race with that.
*
* However, there is a very special and common case
* where we don't care, because there is nothing to
* do: the dentry is still hashed, it does not have
* a 'delete' op, and it's referenced and already on
* the LRU list.
*
* NOTE! Since we aren't locked, these values are
* not "stable". However, it is sufficient that at
* some point after we dropped the reference the
* dentry was hashed and the flags had the proper
* value. Other dentry users may have re-gotten
* a reference to the dentry and change that, but
* our work is done - we can leave the dentry
* around with a zero refcount.
*/
smp_rmb();
d_flags = ACCESS_ONCE(dentry->d_flags);
freeing unlinked file indefinitely delayed Normally opening a file, unlinking it and then closing will have the inode freed upon close() (provided that it's not otherwise busy and has no remaining links, of course). However, there's one case where that does *not* happen. Namely, if you open it by fhandle with cold dcache, then unlink() and close(). In normal case you get d_delete() in unlink(2) notice that dentry is busy and unhash it; on the final dput() it will be forcibly evicted from dcache, triggering iput() and inode removal. In this case, though, we end up with *two* dentries - disconnected (created by open-by-fhandle) and regular one (used by unlink()). The latter will have its reference to inode dropped just fine, but the former will not - it's considered hashed (it is on the ->s_anon list), so it will stay around until the memory pressure will finally do it in. As the result, we have the final iput() delayed indefinitely. It's trivial to reproduce - void flush_dcache(void) { system("mount -o remount,rw /"); } static char buf[20 * 1024 * 1024]; main() { int fd; union { struct file_handle f; char buf[MAX_HANDLE_SZ]; } x; int m; x.f.handle_bytes = sizeof(x); chdir("/root"); mkdir("foo", 0700); fd = open("foo/bar", O_CREAT | O_RDWR, 0600); close(fd); name_to_handle_at(AT_FDCWD, "foo/bar", &x.f, &m, 0); flush_dcache(); fd = open_by_handle_at(AT_FDCWD, &x.f, O_RDWR); unlink("foo/bar"); write(fd, buf, sizeof(buf)); system("df ."); /* 20Mb eaten */ close(fd); system("df ."); /* should've freed those 20Mb */ flush_dcache(); system("df ."); /* should be the same as #2 */ } will spit out something like Filesystem 1K-blocks Used Available Use% Mounted on /dev/root 322023 303843 1131 100% / Filesystem 1K-blocks Used Available Use% Mounted on /dev/root 322023 303843 1131 100% / Filesystem 1K-blocks Used Available Use% Mounted on /dev/root 322023 283282 21692 93% / - inode gets freed only when dentry is finally evicted (here we trigger than by remount; normally it would've happened in response to memory pressure hell knows when). Cc: stable@vger.kernel.org # v2.6.38+; earlier ones need s/kill_it/unhash_it/ Acked-by: J. Bruce Fields <bfields@fieldses.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2015-07-08 09:42:38 +08:00
d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
/* Nothing to do? Dropping the reference was all we needed? */
if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
return 1;
/*
* Not the fast normal case? Get the lock. We've already decremented
* the refcount, but we'll need to re-check the situation after
* getting the lock.
*/
spin_lock(&dentry->d_lock);
/*
* Did somebody else grab a reference to it in the meantime, and
* we're no longer the last user after all? Alternatively, somebody
* else could have killed it and marked it dead. Either way, we
* don't need to do anything else.
*/
if (dentry->d_lockref.count) {
spin_unlock(&dentry->d_lock);
return 1;
}
/*
* Re-get the reference we optimistically dropped. We hold the
* lock, and we just tested that it was zero, so we can just
* set it to 1.
*/
dentry->d_lockref.count = 1;
return 0;
}
/*
* This is dput
*
* This is complicated by the fact that we do not want to put
* dentries that are no longer on any hash chain on the unused
* list: we'd much rather just get rid of them immediately.
*
* However, that implies that we have to traverse the dentry
* tree upwards to the parents which might _also_ now be
* scheduled for deletion (it may have been only waiting for
* its last child to go away).
*
* This tail recursion is done by hand as we don't want to depend
* on the compiler to always get this right (gcc generally doesn't).
* Real recursion would eat up our stack space.
*/
/*
* dput - release a dentry
* @dentry: dentry to release
*
* Release a dentry. This will drop the usage count and if appropriate
* call the dentry unlink method as well as removing it from the queues and
* releasing its resources. If the parent dentries were scheduled for release
* they too may now get deleted.
*/
void dput(struct dentry *dentry)
{
if (unlikely(!dentry))
return;
repeat:
rcu_read_lock();
if (likely(fast_dput(dentry))) {
rcu_read_unlock();
return;
}
/* Slow case: now with the dentry lock held */
rcu_read_unlock();
beginning of transition to parallel lookups - marking in-lookup dentries marked as such when (would be) parallel lookup is about to pass them to actual ->lookup(); unmarked when * __d_add() is about to make it hashed, positive or not. * __d_move() (from d_splice_alias(), directly or via __d_unalias()) puts a preexisting dentry in its place * in caller of ->lookup() if it has escaped all of the above. Bug (WARN_ON, actually) if it reaches the final dput() or d_instantiate() while still marked such. As the result, we are guaranteed that for as long as the flag is set, dentry will * remain negative unhashed with positive refcount * never have its ->d_alias looked at * never have its ->d_lru looked at * never have its ->d_parent and ->d_name changed Right now we have at most one such for any given parent directory. With parallel lookups that restriction will weaken to * only exist when parent is locked shared * at most one with given (parent,name) pair (comparison of names is according to ->d_compare()) * only exist when there's no hashed dentry with the same (parent,name) Transition will take the next several commits; unfortunately, we'll only be able to switch to rwsem at the end of this series. The reason for not making it a single patch is to simplify review. New primitives: d_in_lookup() (a predicate checking if dentry is in the in-lookup state) and d_lookup_done() (tells the system that we are done with lookup and if it's still marked as in-lookup, it should cease to be such). Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 07:52:13 +08:00
WARN_ON(d_in_lookup(dentry));
/* Unreachable? Get rid of it */
if (unlikely(d_unhashed(dentry)))
goto kill_it;
freeing unlinked file indefinitely delayed Normally opening a file, unlinking it and then closing will have the inode freed upon close() (provided that it's not otherwise busy and has no remaining links, of course). However, there's one case where that does *not* happen. Namely, if you open it by fhandle with cold dcache, then unlink() and close(). In normal case you get d_delete() in unlink(2) notice that dentry is busy and unhash it; on the final dput() it will be forcibly evicted from dcache, triggering iput() and inode removal. In this case, though, we end up with *two* dentries - disconnected (created by open-by-fhandle) and regular one (used by unlink()). The latter will have its reference to inode dropped just fine, but the former will not - it's considered hashed (it is on the ->s_anon list), so it will stay around until the memory pressure will finally do it in. As the result, we have the final iput() delayed indefinitely. It's trivial to reproduce - void flush_dcache(void) { system("mount -o remount,rw /"); } static char buf[20 * 1024 * 1024]; main() { int fd; union { struct file_handle f; char buf[MAX_HANDLE_SZ]; } x; int m; x.f.handle_bytes = sizeof(x); chdir("/root"); mkdir("foo", 0700); fd = open("foo/bar", O_CREAT | O_RDWR, 0600); close(fd); name_to_handle_at(AT_FDCWD, "foo/bar", &x.f, &m, 0); flush_dcache(); fd = open_by_handle_at(AT_FDCWD, &x.f, O_RDWR); unlink("foo/bar"); write(fd, buf, sizeof(buf)); system("df ."); /* 20Mb eaten */ close(fd); system("df ."); /* should've freed those 20Mb */ flush_dcache(); system("df ."); /* should be the same as #2 */ } will spit out something like Filesystem 1K-blocks Used Available Use% Mounted on /dev/root 322023 303843 1131 100% / Filesystem 1K-blocks Used Available Use% Mounted on /dev/root 322023 303843 1131 100% / Filesystem 1K-blocks Used Available Use% Mounted on /dev/root 322023 283282 21692 93% / - inode gets freed only when dentry is finally evicted (here we trigger than by remount; normally it would've happened in response to memory pressure hell knows when). Cc: stable@vger.kernel.org # v2.6.38+; earlier ones need s/kill_it/unhash_it/ Acked-by: J. Bruce Fields <bfields@fieldses.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2015-07-08 09:42:38 +08:00
if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
goto kill_it;
if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
if (dentry->d_op->d_delete(dentry))
goto kill_it;
}
if (!(dentry->d_flags & DCACHE_REFERENCED))
dentry->d_flags |= DCACHE_REFERENCED;
dentry_lru_add(dentry);
dentry->d_lockref.count--;
spin_unlock(&dentry->d_lock);
return;
fix quadratic behavior of shrink_dcache_parent() The time shrink_dcache_parent() takes, grows quadratically with the depth of the tree under 'parent'. This starts to get noticable at about 10,000. These kinds of depths don't occur normally, and filesystems which invoke shrink_dcache_parent() via d_invalidate() seem to have other depth dependent timings, so it's not even easy to expose this problem. However with FUSE it's easy to create a deep tree and d_invalidate() will also get called. This can make a syscall hang for a very long time. This is the original discovery of the problem by Russ Cox: http://article.gmane.org/gmane.comp.file-systems.fuse.devel/3826 The following patch fixes the quadratic behavior, by optionally allowing prune_dcache() to prune ancestors of a dentry in one go, instead of doing it one at a time. Common code in dput() and prune_one_dentry() is extracted into a new helper function d_kill(). shrink_dcache_parent() as well as shrink_dcache_sb() are converted to use the ancestry-pruner option. Only for shrink_dcache_memory() is this behavior not desirable, so it keeps using the old algorithm. Signed-off-by: Miklos Szeredi <mszeredi@suse.cz> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Maneesh Soni <maneesh@in.ibm.com> Acked-by: "Paul E. McKenney" <paulmck@us.ibm.com> Cc: Dipankar Sarma <dipankar@in.ibm.com> Cc: Neil Brown <neilb@suse.de> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-08 15:23:46 +08:00
kill_it:
dentry = dentry_kill(dentry);
fix quadratic behavior of shrink_dcache_parent() The time shrink_dcache_parent() takes, grows quadratically with the depth of the tree under 'parent'. This starts to get noticable at about 10,000. These kinds of depths don't occur normally, and filesystems which invoke shrink_dcache_parent() via d_invalidate() seem to have other depth dependent timings, so it's not even easy to expose this problem. However with FUSE it's easy to create a deep tree and d_invalidate() will also get called. This can make a syscall hang for a very long time. This is the original discovery of the problem by Russ Cox: http://article.gmane.org/gmane.comp.file-systems.fuse.devel/3826 The following patch fixes the quadratic behavior, by optionally allowing prune_dcache() to prune ancestors of a dentry in one go, instead of doing it one at a time. Common code in dput() and prune_one_dentry() is extracted into a new helper function d_kill(). shrink_dcache_parent() as well as shrink_dcache_sb() are converted to use the ancestry-pruner option. Only for shrink_dcache_memory() is this behavior not desirable, so it keeps using the old algorithm. Signed-off-by: Miklos Szeredi <mszeredi@suse.cz> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Maneesh Soni <maneesh@in.ibm.com> Acked-by: "Paul E. McKenney" <paulmck@us.ibm.com> Cc: Dipankar Sarma <dipankar@in.ibm.com> Cc: Neil Brown <neilb@suse.de> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-08 15:23:46 +08:00
if (dentry)
goto repeat;
}
EXPORT_SYMBOL(dput);
/* This must be called with d_lock held */
static inline void __dget_dlock(struct dentry *dentry)
{
dentry->d_lockref.count++;
}
static inline void __dget(struct dentry *dentry)
{
lockref_get(&dentry->d_lockref);
}
struct dentry *dget_parent(struct dentry *dentry)
{
int gotref;
struct dentry *ret;
/*
* Do optimistic parent lookup without any
* locking.
*/
rcu_read_lock();
ret = ACCESS_ONCE(dentry->d_parent);
gotref = lockref_get_not_zero(&ret->d_lockref);
rcu_read_unlock();
if (likely(gotref)) {
if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
return ret;
dput(ret);
}
repeat:
/*
* Don't need rcu_dereference because we re-check it was correct under
* the lock.
*/
rcu_read_lock();
ret = dentry->d_parent;
spin_lock(&ret->d_lock);
if (unlikely(ret != dentry->d_parent)) {
spin_unlock(&ret->d_lock);
rcu_read_unlock();
goto repeat;
}
rcu_read_unlock();
BUG_ON(!ret->d_lockref.count);
ret->d_lockref.count++;
spin_unlock(&ret->d_lock);
return ret;
}
EXPORT_SYMBOL(dget_parent);
/**
* d_find_alias - grab a hashed alias of inode
* @inode: inode in question
*
* If inode has a hashed alias, or is a directory and has any alias,
* acquire the reference to alias and return it. Otherwise return NULL.
* Notice that if inode is a directory there can be only one alias and
* it can be unhashed only if it has no children, or if it is the root
* of a filesystem, or if the directory was renamed and d_revalidate
* was the first vfs operation to notice.
*
[PATCH] knfsd: close a race-opportunity in d_splice_alias There is a possible race in d_splice_alias. Though __d_find_alias(inode, 1) will only return a dentry with DCACHE_DISCONNECTED set, it is possible for it to get cleared before the BUG_ON, and it is is not possible to lock against that. There are a couple of problems here. Firstly, the code doesn't match the comment. The comment describes a 'disconnected' dentry as being IS_ROOT as well as DCACHE_DISCONNECTED, however there is not testing of IS_ROOT anythere. A dentry is marked DCACHE_DISCONNECTED when allocated with d_alloc_anon, and remains DCACHE_DISCONNECTED while a path is built up towards the root. So a dentry can have a valid name and a valid parent and even grandparent, but will still be DCACHE_DISCONNECTED until a path to the root is created. Once the path to the root is complete, everything in the path gets DCACHE_DISCONNECTED cleared. So the fact that DCACHE_DISCONNECTED isn't enough to say that a dentry is free to be spliced in with a given name. This can only be allowed if the dentry does not yet have a name, so the IS_ROOT test is needed too. However even adding that test to __d_find_alias isn't enough. As d_splice_alias drops dcache_lock before calling d_move to perform the splice, it could race with another thread calling d_splice_alias to splice the inode in with a different name in a different part of the tree (in the case where a file has hard links). So that splicing code is only really safe for directories (as we know that directories only have one link). For directories, the caller of d_splice_alias will be holding i_mutex on the (unique) parent so there is no room for a race. A consequence of this is that a non-directory will never benefit from being spliced into a pre-exisiting dentry, but that isn't a problem. It is perfectly OK for a non-directory to have multiple dentries, some anonymous, some not. And the comment for d_splice_alias says that it only happens for directories anyway. Signed-off-by: Neil Brown <neilb@suse.de> Cc: Christoph Hellwig <hch@lst.de> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Dipankar Sarma <dipankar@in.ibm.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-10-04 17:16:16 +08:00
* If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
* any other hashed alias over that one.
*/
static struct dentry *__d_find_alias(struct inode *inode)
{
struct dentry *alias, *discon_alias;
again:
discon_alias = NULL;
hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
spin_lock(&alias->d_lock);
if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
[PATCH] knfsd: close a race-opportunity in d_splice_alias There is a possible race in d_splice_alias. Though __d_find_alias(inode, 1) will only return a dentry with DCACHE_DISCONNECTED set, it is possible for it to get cleared before the BUG_ON, and it is is not possible to lock against that. There are a couple of problems here. Firstly, the code doesn't match the comment. The comment describes a 'disconnected' dentry as being IS_ROOT as well as DCACHE_DISCONNECTED, however there is not testing of IS_ROOT anythere. A dentry is marked DCACHE_DISCONNECTED when allocated with d_alloc_anon, and remains DCACHE_DISCONNECTED while a path is built up towards the root. So a dentry can have a valid name and a valid parent and even grandparent, but will still be DCACHE_DISCONNECTED until a path to the root is created. Once the path to the root is complete, everything in the path gets DCACHE_DISCONNECTED cleared. So the fact that DCACHE_DISCONNECTED isn't enough to say that a dentry is free to be spliced in with a given name. This can only be allowed if the dentry does not yet have a name, so the IS_ROOT test is needed too. However even adding that test to __d_find_alias isn't enough. As d_splice_alias drops dcache_lock before calling d_move to perform the splice, it could race with another thread calling d_splice_alias to splice the inode in with a different name in a different part of the tree (in the case where a file has hard links). So that splicing code is only really safe for directories (as we know that directories only have one link). For directories, the caller of d_splice_alias will be holding i_mutex on the (unique) parent so there is no room for a race. A consequence of this is that a non-directory will never benefit from being spliced into a pre-exisiting dentry, but that isn't a problem. It is perfectly OK for a non-directory to have multiple dentries, some anonymous, some not. And the comment for d_splice_alias says that it only happens for directories anyway. Signed-off-by: Neil Brown <neilb@suse.de> Cc: Christoph Hellwig <hch@lst.de> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Dipankar Sarma <dipankar@in.ibm.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-10-04 17:16:16 +08:00
if (IS_ROOT(alias) &&
(alias->d_flags & DCACHE_DISCONNECTED)) {
discon_alias = alias;
} else {
__dget_dlock(alias);
spin_unlock(&alias->d_lock);
return alias;
}
}
spin_unlock(&alias->d_lock);
}
if (discon_alias) {
alias = discon_alias;
spin_lock(&alias->d_lock);
if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
__dget_dlock(alias);
spin_unlock(&alias->d_lock);
return alias;
}
spin_unlock(&alias->d_lock);
goto again;
}
return NULL;
}
struct dentry *d_find_alias(struct inode *inode)
{
struct dentry *de = NULL;
if (!hlist_empty(&inode->i_dentry)) {
spin_lock(&inode->i_lock);
de = __d_find_alias(inode);
spin_unlock(&inode->i_lock);
}
return de;
}
EXPORT_SYMBOL(d_find_alias);
/*
* Try to kill dentries associated with this inode.
* WARNING: you must own a reference to inode.
*/
void d_prune_aliases(struct inode *inode)
{
struct dentry *dentry;
restart:
spin_lock(&inode->i_lock);
hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
spin_lock(&dentry->d_lock);
if (!dentry->d_lockref.count) {
struct dentry *parent = lock_parent(dentry);
if (likely(!dentry->d_lockref.count)) {
__dentry_kill(dentry);
dput(parent);
goto restart;
}
if (parent)
spin_unlock(&parent->d_lock);
}
spin_unlock(&dentry->d_lock);
}
spin_unlock(&inode->i_lock);
}
EXPORT_SYMBOL(d_prune_aliases);
static void shrink_dentry_list(struct list_head *list)
{
struct dentry *dentry, *parent;
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries [Summary] Split LRU-list of unused dentries to one per superblock to avoid soft lock up during NFS mounts and remounting of any filesystem. Previously I posted here: http://lkml.org/lkml/2008/3/5/590 [Descriptions] - background dentry_unused is a list of dentries which are not referenced. dentry_unused grows up when references on directories or files are released. This list can be very long if there is huge free memory. - the problem When shrink_dcache_sb() is called, it scans all dentry_unused linearly under spin_lock(), and if dentry->d_sb is differnt from given superblock, scan next dentry. This scan costs very much if there are many entries, and very ineffective if there are many superblocks. IOW, When we need to shrink unused dentries on one dentry, but scans unused dentries on all superblocks in the system. For example, we scan 500 dentries to unmount a filesystem, but scans 1,000,000 or more unused dentries on other superblocks. In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink unused dentries on NFS, but scans 100,000,000 unused dentries on superblocks in the system such as local ext3 filesystems. I hear NFS mounting took 1 min on some system in use. * : NFS uses virtual filesystem in rpc layer, so NFS is affected by this problem. 100,000,000 is possible number on large systems. Per-superblock LRU of unused dentried can reduce the cost in reasonable manner. - How to fix I found this problem is solved by David Chinner's "Per-superblock unused dentry LRU lists V3"(1), so I rebase it and add some fix to reclaim with fairness, which is in Andrew Morton's comments(2). 1) http://lkml.org/lkml/2006/5/25/318 2) http://lkml.org/lkml/2006/5/25/320 Split LRU-list of unused dentries to each superblocks. Then, NFS mounting will check dentries under a superblock instead of all. But this spliting will break LRU of dentry-unused. So, I've attempted to make reclaim unused dentrins with fairness by calculate number of dentries to scan on this sb based on following way number of dentries to scan on this sb = count * (number of dentries on this sb / number of dentries in the machine) - ToDo - I have to measuring performance number and do stress tests. - When unmount occurs during prune_dcache(), scanning on same superblock, It is unable to reach next superblock because it is gone away. We restart scannig superblock from first one, it causes unfairness of reclaim unused dentries on first superblock. But I think this happens very rarely. - Test Results Result on 6GB boxes with excessive unused dentries. Without patch: $ cat /proc/sys/fs/dentry-state 10181835 10180203 45 0 0 0 # mount -t nfs 10.124.60.70:/work/kernel-src nfs real 0m1.830s user 0m0.001s sys 0m1.653s With this patch: $ cat /proc/sys/fs/dentry-state 10236610 10234751 45 0 0 0 # mount -t nfs 10.124.60.70:/work/kernel-src nfs real 0m0.106s user 0m0.002s sys 0m0.032s [akpm@linux-foundation.org: fix comments] Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com> Cc: Neil Brown <neilb@suse.de> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Cc: David Chinner <dgc@sgi.com> Cc: "J. Bruce Fields" <bfields@fieldses.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
while (!list_empty(list)) {
struct inode *inode;
dentry = list_entry(list->prev, struct dentry, d_lru);
spin_lock(&dentry->d_lock);
shrink_dentry_list(): take parent's ->d_lock earlier The cause of livelocks there is that we are taking ->d_lock on dentry and its parent in the wrong order, forcing us to use trylock on the parent's one. d_walk() takes them in the right order, and unfortunately it's not hard to create a situation when shrink_dentry_list() can't make progress since trylock keeps failing, and shrink_dcache_parent() or check_submounts_and_drop() keeps calling d_walk() disrupting the very shrink_dentry_list() it's waiting for. Solution is straightforward - if that trylock fails, let's unlock the dentry itself and take locks in the right order. We need to stabilize ->d_parent without holding ->d_lock, but that's doable using RCU. And we'd better do that in the very beginning of the loop in shrink_dentry_list(), since the checks on refcount, etc. would need to be redone anyway. That deals with a half of the problem - killing dentries on the shrink list itself. Another one (dropping their parents) is in the next commit. locking parent is interesting - it would be easy to do rcu_read_lock(), lock whatever we think is a parent, lock dentry itself and check if the parent is still the right one. Except that we need to check that *before* locking the dentry, or we are risking taking ->d_lock out of order. Fortunately, once the D1 is locked, we can check if D2->d_parent is equal to D1 without the need to lock D2; D2->d_parent can start or stop pointing to D1 only under D1->d_lock, so taking D1->d_lock is enough. In other words, the right solution is rcu_read_lock/lock what looks like parent right now/check if it's still our parent/rcu_read_unlock/lock the child. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2014-05-29 20:54:52 +08:00
parent = lock_parent(dentry);
dcache: remove dentries from LRU before putting on dispose list One of the big problems with modifying the way the dcache shrinker and LRU implementation works is that the LRU is abused in several ways. One of these is shrink_dentry_list(). Basically, we can move a dentry off the LRU onto a different list without doing any accounting changes, and then use dentry_lru_prune() to remove it from what-ever list it is now on to do the LRU accounting at that point. This makes it -really hard- to change the LRU implementation. The use of the per-sb LRU lock serialises movement of the dentries between the different lists and the removal of them, and this is the only reason that it works. If we want to break up the dentry LRU lock and lists into, say, per-node lists, we remove the only serialisation that allows this lru list/dispose list abuse to work. To make this work effectively, the dispose list has to be isolated from the LRU list - dentries have to be removed from the LRU *before* being placed on the dispose list. This means that the LRU accounting and isolation is completed before disposal is started, and that means we can change the LRU implementation freely in future. This means that dentries *must* be marked with DCACHE_SHRINK_LIST when they are placed on the dispose list so that we don't think that parent dentries found in try_prune_one_dentry() are on the LRU when the are actually on the dispose list. This would result in accounting the dentry to the LRU a second time. Hence dentry_lru_del() has to handle the DCACHE_SHRINK_LIST case Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Glauber Costa <glommer@openvz.org> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com> Cc: Arve Hjønnevåg <arve@android.com> Cc: Carlos Maiolino <cmaiolino@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Chuck Lever <chuck.lever@oracle.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: David Rientjes <rientjes@google.com> Cc: Gleb Natapov <gleb@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: J. Bruce Fields <bfields@redhat.com> Cc: Jan Kara <jack@suse.cz> Cc: Jerome Glisse <jglisse@redhat.com> Cc: John Stultz <john.stultz@linaro.org> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Kent Overstreet <koverstreet@google.com> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Steven Whitehouse <swhiteho@redhat.com> Cc: Thomas Hellstrom <thellstrom@vmware.com> Cc: Trond Myklebust <Trond.Myklebust@netapp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:17:55 +08:00
/*
* The dispose list is isolated and dentries are not accounted
* to the LRU here, so we can simply remove it from the list
* here regardless of whether it is referenced or not.
*/
d_shrink_del(dentry);
dcache: remove dentries from LRU before putting on dispose list One of the big problems with modifying the way the dcache shrinker and LRU implementation works is that the LRU is abused in several ways. One of these is shrink_dentry_list(). Basically, we can move a dentry off the LRU onto a different list without doing any accounting changes, and then use dentry_lru_prune() to remove it from what-ever list it is now on to do the LRU accounting at that point. This makes it -really hard- to change the LRU implementation. The use of the per-sb LRU lock serialises movement of the dentries between the different lists and the removal of them, and this is the only reason that it works. If we want to break up the dentry LRU lock and lists into, say, per-node lists, we remove the only serialisation that allows this lru list/dispose list abuse to work. To make this work effectively, the dispose list has to be isolated from the LRU list - dentries have to be removed from the LRU *before* being placed on the dispose list. This means that the LRU accounting and isolation is completed before disposal is started, and that means we can change the LRU implementation freely in future. This means that dentries *must* be marked with DCACHE_SHRINK_LIST when they are placed on the dispose list so that we don't think that parent dentries found in try_prune_one_dentry() are on the LRU when the are actually on the dispose list. This would result in accounting the dentry to the LRU a second time. Hence dentry_lru_del() has to handle the DCACHE_SHRINK_LIST case Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Glauber Costa <glommer@openvz.org> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com> Cc: Arve Hjønnevåg <arve@android.com> Cc: Carlos Maiolino <cmaiolino@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Chuck Lever <chuck.lever@oracle.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: David Rientjes <rientjes@google.com> Cc: Gleb Natapov <gleb@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: J. Bruce Fields <bfields@redhat.com> Cc: Jan Kara <jack@suse.cz> Cc: Jerome Glisse <jglisse@redhat.com> Cc: John Stultz <john.stultz@linaro.org> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Kent Overstreet <koverstreet@google.com> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Steven Whitehouse <swhiteho@redhat.com> Cc: Thomas Hellstrom <thellstrom@vmware.com> Cc: Trond Myklebust <Trond.Myklebust@netapp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:17:55 +08:00
/*
* We found an inuse dentry which was not removed from
dcache: remove dentries from LRU before putting on dispose list One of the big problems with modifying the way the dcache shrinker and LRU implementation works is that the LRU is abused in several ways. One of these is shrink_dentry_list(). Basically, we can move a dentry off the LRU onto a different list without doing any accounting changes, and then use dentry_lru_prune() to remove it from what-ever list it is now on to do the LRU accounting at that point. This makes it -really hard- to change the LRU implementation. The use of the per-sb LRU lock serialises movement of the dentries between the different lists and the removal of them, and this is the only reason that it works. If we want to break up the dentry LRU lock and lists into, say, per-node lists, we remove the only serialisation that allows this lru list/dispose list abuse to work. To make this work effectively, the dispose list has to be isolated from the LRU list - dentries have to be removed from the LRU *before* being placed on the dispose list. This means that the LRU accounting and isolation is completed before disposal is started, and that means we can change the LRU implementation freely in future. This means that dentries *must* be marked with DCACHE_SHRINK_LIST when they are placed on the dispose list so that we don't think that parent dentries found in try_prune_one_dentry() are on the LRU when the are actually on the dispose list. This would result in accounting the dentry to the LRU a second time. Hence dentry_lru_del() has to handle the DCACHE_SHRINK_LIST case Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Glauber Costa <glommer@openvz.org> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com> Cc: Arve Hjønnevåg <arve@android.com> Cc: Carlos Maiolino <cmaiolino@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Chuck Lever <chuck.lever@oracle.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: David Rientjes <rientjes@google.com> Cc: Gleb Natapov <gleb@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: J. Bruce Fields <bfields@redhat.com> Cc: Jan Kara <jack@suse.cz> Cc: Jerome Glisse <jglisse@redhat.com> Cc: John Stultz <john.stultz@linaro.org> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Kent Overstreet <koverstreet@google.com> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Steven Whitehouse <swhiteho@redhat.com> Cc: Thomas Hellstrom <thellstrom@vmware.com> Cc: Trond Myklebust <Trond.Myklebust@netapp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:17:55 +08:00
* the LRU because of laziness during lookup. Do not free it.
*/
if (dentry->d_lockref.count > 0) {
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries [Summary] Split LRU-list of unused dentries to one per superblock to avoid soft lock up during NFS mounts and remounting of any filesystem. Previously I posted here: http://lkml.org/lkml/2008/3/5/590 [Descriptions] - background dentry_unused is a list of dentries which are not referenced. dentry_unused grows up when references on directories or files are released. This list can be very long if there is huge free memory. - the problem When shrink_dcache_sb() is called, it scans all dentry_unused linearly under spin_lock(), and if dentry->d_sb is differnt from given superblock, scan next dentry. This scan costs very much if there are many entries, and very ineffective if there are many superblocks. IOW, When we need to shrink unused dentries on one dentry, but scans unused dentries on all superblocks in the system. For example, we scan 500 dentries to unmount a filesystem, but scans 1,000,000 or more unused dentries on other superblocks. In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink unused dentries on NFS, but scans 100,000,000 unused dentries on superblocks in the system such as local ext3 filesystems. I hear NFS mounting took 1 min on some system in use. * : NFS uses virtual filesystem in rpc layer, so NFS is affected by this problem. 100,000,000 is possible number on large systems. Per-superblock LRU of unused dentried can reduce the cost in reasonable manner. - How to fix I found this problem is solved by David Chinner's "Per-superblock unused dentry LRU lists V3"(1), so I rebase it and add some fix to reclaim with fairness, which is in Andrew Morton's comments(2). 1) http://lkml.org/lkml/2006/5/25/318 2) http://lkml.org/lkml/2006/5/25/320 Split LRU-list of unused dentries to each superblocks. Then, NFS mounting will check dentries under a superblock instead of all. But this spliting will break LRU of dentry-unused. So, I've attempted to make reclaim unused dentrins with fairness by calculate number of dentries to scan on this sb based on following way number of dentries to scan on this sb = count * (number of dentries on this sb / number of dentries in the machine) - ToDo - I have to measuring performance number and do stress tests. - When unmount occurs during prune_dcache(), scanning on same superblock, It is unable to reach next superblock because it is gone away. We restart scannig superblock from first one, it causes unfairness of reclaim unused dentries on first superblock. But I think this happens very rarely. - Test Results Result on 6GB boxes with excessive unused dentries. Without patch: $ cat /proc/sys/fs/dentry-state 10181835 10180203 45 0 0 0 # mount -t nfs 10.124.60.70:/work/kernel-src nfs real 0m1.830s user 0m0.001s sys 0m1.653s With this patch: $ cat /proc/sys/fs/dentry-state 10236610 10234751 45 0 0 0 # mount -t nfs 10.124.60.70:/work/kernel-src nfs real 0m0.106s user 0m0.002s sys 0m0.032s [akpm@linux-foundation.org: fix comments] Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com> Cc: Neil Brown <neilb@suse.de> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Cc: David Chinner <dgc@sgi.com> Cc: "J. Bruce Fields" <bfields@fieldses.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
spin_unlock(&dentry->d_lock);
shrink_dentry_list(): take parent's ->d_lock earlier The cause of livelocks there is that we are taking ->d_lock on dentry and its parent in the wrong order, forcing us to use trylock on the parent's one. d_walk() takes them in the right order, and unfortunately it's not hard to create a situation when shrink_dentry_list() can't make progress since trylock keeps failing, and shrink_dcache_parent() or check_submounts_and_drop() keeps calling d_walk() disrupting the very shrink_dentry_list() it's waiting for. Solution is straightforward - if that trylock fails, let's unlock the dentry itself and take locks in the right order. We need to stabilize ->d_parent without holding ->d_lock, but that's doable using RCU. And we'd better do that in the very beginning of the loop in shrink_dentry_list(), since the checks on refcount, etc. would need to be redone anyway. That deals with a half of the problem - killing dentries on the shrink list itself. Another one (dropping their parents) is in the next commit. locking parent is interesting - it would be easy to do rcu_read_lock(), lock whatever we think is a parent, lock dentry itself and check if the parent is still the right one. Except that we need to check that *before* locking the dentry, or we are risking taking ->d_lock out of order. Fortunately, once the D1 is locked, we can check if D2->d_parent is equal to D1 without the need to lock D2; D2->d_parent can start or stop pointing to D1 only under D1->d_lock, so taking D1->d_lock is enough. In other words, the right solution is rcu_read_lock/lock what looks like parent right now/check if it's still our parent/rcu_read_unlock/lock the child. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2014-05-29 20:54:52 +08:00
if (parent)
spin_unlock(&parent->d_lock);
continue;
}
if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
spin_unlock(&dentry->d_lock);
shrink_dentry_list(): take parent's ->d_lock earlier The cause of livelocks there is that we are taking ->d_lock on dentry and its parent in the wrong order, forcing us to use trylock on the parent's one. d_walk() takes them in the right order, and unfortunately it's not hard to create a situation when shrink_dentry_list() can't make progress since trylock keeps failing, and shrink_dcache_parent() or check_submounts_and_drop() keeps calling d_walk() disrupting the very shrink_dentry_list() it's waiting for. Solution is straightforward - if that trylock fails, let's unlock the dentry itself and take locks in the right order. We need to stabilize ->d_parent without holding ->d_lock, but that's doable using RCU. And we'd better do that in the very beginning of the loop in shrink_dentry_list(), since the checks on refcount, etc. would need to be redone anyway. That deals with a half of the problem - killing dentries on the shrink list itself. Another one (dropping their parents) is in the next commit. locking parent is interesting - it would be easy to do rcu_read_lock(), lock whatever we think is a parent, lock dentry itself and check if the parent is still the right one. Except that we need to check that *before* locking the dentry, or we are risking taking ->d_lock out of order. Fortunately, once the D1 is locked, we can check if D2->d_parent is equal to D1 without the need to lock D2; D2->d_parent can start or stop pointing to D1 only under D1->d_lock, so taking D1->d_lock is enough. In other words, the right solution is rcu_read_lock/lock what looks like parent right now/check if it's still our parent/rcu_read_unlock/lock the child. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2014-05-29 20:54:52 +08:00
if (parent)
spin_unlock(&parent->d_lock);
if (can_free)
dentry_free(dentry);
continue;
}
inode = dentry->d_inode;
if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
d_shrink_add(dentry, list);
dcache: remove dentries from LRU before putting on dispose list One of the big problems with modifying the way the dcache shrinker and LRU implementation works is that the LRU is abused in several ways. One of these is shrink_dentry_list(). Basically, we can move a dentry off the LRU onto a different list without doing any accounting changes, and then use dentry_lru_prune() to remove it from what-ever list it is now on to do the LRU accounting at that point. This makes it -really hard- to change the LRU implementation. The use of the per-sb LRU lock serialises movement of the dentries between the different lists and the removal of them, and this is the only reason that it works. If we want to break up the dentry LRU lock and lists into, say, per-node lists, we remove the only serialisation that allows this lru list/dispose list abuse to work. To make this work effectively, the dispose list has to be isolated from the LRU list - dentries have to be removed from the LRU *before* being placed on the dispose list. This means that the LRU accounting and isolation is completed before disposal is started, and that means we can change the LRU implementation freely in future. This means that dentries *must* be marked with DCACHE_SHRINK_LIST when they are placed on the dispose list so that we don't think that parent dentries found in try_prune_one_dentry() are on the LRU when the are actually on the dispose list. This would result in accounting the dentry to the LRU a second time. Hence dentry_lru_del() has to handle the DCACHE_SHRINK_LIST case Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Glauber Costa <glommer@openvz.org> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com> Cc: Arve Hjønnevåg <arve@android.com> Cc: Carlos Maiolino <cmaiolino@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Chuck Lever <chuck.lever@oracle.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: David Rientjes <rientjes@google.com> Cc: Gleb Natapov <gleb@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: J. Bruce Fields <bfields@redhat.com> Cc: Jan Kara <jack@suse.cz> Cc: Jerome Glisse <jglisse@redhat.com> Cc: John Stultz <john.stultz@linaro.org> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Kent Overstreet <koverstreet@google.com> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Steven Whitehouse <swhiteho@redhat.com> Cc: Thomas Hellstrom <thellstrom@vmware.com> Cc: Trond Myklebust <Trond.Myklebust@netapp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:17:55 +08:00
spin_unlock(&dentry->d_lock);
shrink_dentry_list(): take parent's ->d_lock earlier The cause of livelocks there is that we are taking ->d_lock on dentry and its parent in the wrong order, forcing us to use trylock on the parent's one. d_walk() takes them in the right order, and unfortunately it's not hard to create a situation when shrink_dentry_list() can't make progress since trylock keeps failing, and shrink_dcache_parent() or check_submounts_and_drop() keeps calling d_walk() disrupting the very shrink_dentry_list() it's waiting for. Solution is straightforward - if that trylock fails, let's unlock the dentry itself and take locks in the right order. We need to stabilize ->d_parent without holding ->d_lock, but that's doable using RCU. And we'd better do that in the very beginning of the loop in shrink_dentry_list(), since the checks on refcount, etc. would need to be redone anyway. That deals with a half of the problem - killing dentries on the shrink list itself. Another one (dropping their parents) is in the next commit. locking parent is interesting - it would be easy to do rcu_read_lock(), lock whatever we think is a parent, lock dentry itself and check if the parent is still the right one. Except that we need to check that *before* locking the dentry, or we are risking taking ->d_lock out of order. Fortunately, once the D1 is locked, we can check if D2->d_parent is equal to D1 without the need to lock D2; D2->d_parent can start or stop pointing to D1 only under D1->d_lock, so taking D1->d_lock is enough. In other words, the right solution is rcu_read_lock/lock what looks like parent right now/check if it's still our parent/rcu_read_unlock/lock the child. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2014-05-29 20:54:52 +08:00
if (parent)
spin_unlock(&parent->d_lock);
continue;
dcache: remove dentries from LRU before putting on dispose list One of the big problems with modifying the way the dcache shrinker and LRU implementation works is that the LRU is abused in several ways. One of these is shrink_dentry_list(). Basically, we can move a dentry off the LRU onto a different list without doing any accounting changes, and then use dentry_lru_prune() to remove it from what-ever list it is now on to do the LRU accounting at that point. This makes it -really hard- to change the LRU implementation. The use of the per-sb LRU lock serialises movement of the dentries between the different lists and the removal of them, and this is the only reason that it works. If we want to break up the dentry LRU lock and lists into, say, per-node lists, we remove the only serialisation that allows this lru list/dispose list abuse to work. To make this work effectively, the dispose list has to be isolated from the LRU list - dentries have to be removed from the LRU *before* being placed on the dispose list. This means that the LRU accounting and isolation is completed before disposal is started, and that means we can change the LRU implementation freely in future. This means that dentries *must* be marked with DCACHE_SHRINK_LIST when they are placed on the dispose list so that we don't think that parent dentries found in try_prune_one_dentry() are on the LRU when the are actually on the dispose list. This would result in accounting the dentry to the LRU a second time. Hence dentry_lru_del() has to handle the DCACHE_SHRINK_LIST case Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Glauber Costa <glommer@openvz.org> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com> Cc: Arve Hjønnevåg <arve@android.com> Cc: Carlos Maiolino <cmaiolino@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Chuck Lever <chuck.lever@oracle.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: David Rientjes <rientjes@google.com> Cc: Gleb Natapov <gleb@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: J. Bruce Fields <bfields@redhat.com> Cc: Jan Kara <jack@suse.cz> Cc: Jerome Glisse <jglisse@redhat.com> Cc: John Stultz <john.stultz@linaro.org> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Kent Overstreet <koverstreet@google.com> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Steven Whitehouse <swhiteho@redhat.com> Cc: Thomas Hellstrom <thellstrom@vmware.com> Cc: Trond Myklebust <Trond.Myklebust@netapp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:17:55 +08:00
}
__dentry_kill(dentry);
shrink_dentry_list(): take parent's ->d_lock earlier The cause of livelocks there is that we are taking ->d_lock on dentry and its parent in the wrong order, forcing us to use trylock on the parent's one. d_walk() takes them in the right order, and unfortunately it's not hard to create a situation when shrink_dentry_list() can't make progress since trylock keeps failing, and shrink_dcache_parent() or check_submounts_and_drop() keeps calling d_walk() disrupting the very shrink_dentry_list() it's waiting for. Solution is straightforward - if that trylock fails, let's unlock the dentry itself and take locks in the right order. We need to stabilize ->d_parent without holding ->d_lock, but that's doable using RCU. And we'd better do that in the very beginning of the loop in shrink_dentry_list(), since the checks on refcount, etc. would need to be redone anyway. That deals with a half of the problem - killing dentries on the shrink list itself. Another one (dropping their parents) is in the next commit. locking parent is interesting - it would be easy to do rcu_read_lock(), lock whatever we think is a parent, lock dentry itself and check if the parent is still the right one. Except that we need to check that *before* locking the dentry, or we are risking taking ->d_lock out of order. Fortunately, once the D1 is locked, we can check if D2->d_parent is equal to D1 without the need to lock D2; D2->d_parent can start or stop pointing to D1 only under D1->d_lock, so taking D1->d_lock is enough. In other words, the right solution is rcu_read_lock/lock what looks like parent right now/check if it's still our parent/rcu_read_unlock/lock the child. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2014-05-29 20:54:52 +08:00
/*
* We need to prune ancestors too. This is necessary to prevent
* quadratic behavior of shrink_dcache_parent(), but is also
* expected to be beneficial in reducing dentry cache
* fragmentation.
*/
dentry = parent;
while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
parent = lock_parent(dentry);
if (dentry->d_lockref.count != 1) {
dentry->d_lockref.count--;
spin_unlock(&dentry->d_lock);
if (parent)
spin_unlock(&parent->d_lock);
break;
}
inode = dentry->d_inode; /* can't be NULL */
if (unlikely(!spin_trylock(&inode->i_lock))) {
spin_unlock(&dentry->d_lock);
if (parent)
spin_unlock(&parent->d_lock);
cpu_relax();
continue;
}
__dentry_kill(dentry);
dentry = parent;
}
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries [Summary] Split LRU-list of unused dentries to one per superblock to avoid soft lock up during NFS mounts and remounting of any filesystem. Previously I posted here: http://lkml.org/lkml/2008/3/5/590 [Descriptions] - background dentry_unused is a list of dentries which are not referenced. dentry_unused grows up when references on directories or files are released. This list can be very long if there is huge free memory. - the problem When shrink_dcache_sb() is called, it scans all dentry_unused linearly under spin_lock(), and if dentry->d_sb is differnt from given superblock, scan next dentry. This scan costs very much if there are many entries, and very ineffective if there are many superblocks. IOW, When we need to shrink unused dentries on one dentry, but scans unused dentries on all superblocks in the system. For example, we scan 500 dentries to unmount a filesystem, but scans 1,000,000 or more unused dentries on other superblocks. In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink unused dentries on NFS, but scans 100,000,000 unused dentries on superblocks in the system such as local ext3 filesystems. I hear NFS mounting took 1 min on some system in use. * : NFS uses virtual filesystem in rpc layer, so NFS is affected by this problem. 100,000,000 is possible number on large systems. Per-superblock LRU of unused dentried can reduce the cost in reasonable manner. - How to fix I found this problem is solved by David Chinner's "Per-superblock unused dentry LRU lists V3"(1), so I rebase it and add some fix to reclaim with fairness, which is in Andrew Morton's comments(2). 1) http://lkml.org/lkml/2006/5/25/318 2) http://lkml.org/lkml/2006/5/25/320 Split LRU-list of unused dentries to each superblocks. Then, NFS mounting will check dentries under a superblock instead of all. But this spliting will break LRU of dentry-unused. So, I've attempted to make reclaim unused dentrins with fairness by calculate number of dentries to scan on this sb based on following way number of dentries to scan on this sb = count * (number of dentries on this sb / number of dentries in the machine) - ToDo - I have to measuring performance number and do stress tests. - When unmount occurs during prune_dcache(), scanning on same superblock, It is unable to reach next superblock because it is gone away. We restart scannig superblock from first one, it causes unfairness of reclaim unused dentries on first superblock. But I think this happens very rarely. - Test Results Result on 6GB boxes with excessive unused dentries. Without patch: $ cat /proc/sys/fs/dentry-state 10181835 10180203 45 0 0 0 # mount -t nfs 10.124.60.70:/work/kernel-src nfs real 0m1.830s user 0m0.001s sys 0m1.653s With this patch: $ cat /proc/sys/fs/dentry-state 10236610 10234751 45 0 0 0 # mount -t nfs 10.124.60.70:/work/kernel-src nfs real 0m0.106s user 0m0.002s sys 0m0.032s [akpm@linux-foundation.org: fix comments] Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com> Cc: Neil Brown <neilb@suse.de> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Cc: David Chinner <dgc@sgi.com> Cc: "J. Bruce Fields" <bfields@fieldses.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
}
}
list_lru: add helpers to isolate items Currently, the isolate callback passed to the list_lru_walk family of functions is supposed to just delete an item from the list upon returning LRU_REMOVED or LRU_REMOVED_RETRY, while nr_items counter is fixed by __list_lru_walk_one after the callback returns. Since the callback is allowed to drop the lock after removing an item (it has to return LRU_REMOVED_RETRY then), the nr_items can be less than the actual number of elements on the list even if we check them under the lock. This makes it difficult to move items from one list_lru_one to another, which is required for per-memcg list_lru reparenting - we can't just splice the lists, we have to move entries one by one. This patch therefore introduces helpers that must be used by callback functions to isolate items instead of raw list_del/list_move. These are list_lru_isolate and list_lru_isolate_move. They not only remove the entry from the list, but also fix the nr_items counter, making sure nr_items always reflects the actual number of elements on the list if checked under the appropriate lock. Signed-off-by: Vladimir Davydov <vdavydov@parallels.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@suse.cz> Cc: Tejun Heo <tj@kernel.org> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Dave Chinner <david@fromorbit.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-02-13 06:59:35 +08:00
static enum lru_status dentry_lru_isolate(struct list_head *item,
struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
{
struct list_head *freeable = arg;
struct dentry *dentry = container_of(item, struct dentry, d_lru);
/*
* we are inverting the lru lock/dentry->d_lock here,
* so use a trylock. If we fail to get the lock, just skip
* it
*/
if (!spin_trylock(&dentry->d_lock))
return LRU_SKIP;
/*
* Referenced dentries are still in use. If they have active
* counts, just remove them from the LRU. Otherwise give them
* another pass through the LRU.
*/
if (dentry->d_lockref.count) {
list_lru: add helpers to isolate items Currently, the isolate callback passed to the list_lru_walk family of functions is supposed to just delete an item from the list upon returning LRU_REMOVED or LRU_REMOVED_RETRY, while nr_items counter is fixed by __list_lru_walk_one after the callback returns. Since the callback is allowed to drop the lock after removing an item (it has to return LRU_REMOVED_RETRY then), the nr_items can be less than the actual number of elements on the list even if we check them under the lock. This makes it difficult to move items from one list_lru_one to another, which is required for per-memcg list_lru reparenting - we can't just splice the lists, we have to move entries one by one. This patch therefore introduces helpers that must be used by callback functions to isolate items instead of raw list_del/list_move. These are list_lru_isolate and list_lru_isolate_move. They not only remove the entry from the list, but also fix the nr_items counter, making sure nr_items always reflects the actual number of elements on the list if checked under the appropriate lock. Signed-off-by: Vladimir Davydov <vdavydov@parallels.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@suse.cz> Cc: Tejun Heo <tj@kernel.org> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Dave Chinner <david@fromorbit.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-02-13 06:59:35 +08:00
d_lru_isolate(lru, dentry);
spin_unlock(&dentry->d_lock);
return LRU_REMOVED;
}
if (dentry->d_flags & DCACHE_REFERENCED) {
dentry->d_flags &= ~DCACHE_REFERENCED;
spin_unlock(&dentry->d_lock);
/*
* The list move itself will be made by the common LRU code. At
* this point, we've dropped the dentry->d_lock but keep the
* lru lock. This is safe to do, since every list movement is
* protected by the lru lock even if both locks are held.
*
* This is guaranteed by the fact that all LRU management
* functions are intermediated by the LRU API calls like
* list_lru_add and list_lru_del. List movement in this file
* only ever occur through this functions or through callbacks
* like this one, that are called from the LRU API.
*
* The only exceptions to this are functions like
* shrink_dentry_list, and code that first checks for the
* DCACHE_SHRINK_LIST flag. Those are guaranteed to be
* operating only with stack provided lists after they are
* properly isolated from the main list. It is thus, always a
* local access.
*/
return LRU_ROTATE;
}
list_lru: add helpers to isolate items Currently, the isolate callback passed to the list_lru_walk family of functions is supposed to just delete an item from the list upon returning LRU_REMOVED or LRU_REMOVED_RETRY, while nr_items counter is fixed by __list_lru_walk_one after the callback returns. Since the callback is allowed to drop the lock after removing an item (it has to return LRU_REMOVED_RETRY then), the nr_items can be less than the actual number of elements on the list even if we check them under the lock. This makes it difficult to move items from one list_lru_one to another, which is required for per-memcg list_lru reparenting - we can't just splice the lists, we have to move entries one by one. This patch therefore introduces helpers that must be used by callback functions to isolate items instead of raw list_del/list_move. These are list_lru_isolate and list_lru_isolate_move. They not only remove the entry from the list, but also fix the nr_items counter, making sure nr_items always reflects the actual number of elements on the list if checked under the appropriate lock. Signed-off-by: Vladimir Davydov <vdavydov@parallels.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@suse.cz> Cc: Tejun Heo <tj@kernel.org> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Dave Chinner <david@fromorbit.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-02-13 06:59:35 +08:00
d_lru_shrink_move(lru, dentry, freeable);
spin_unlock(&dentry->d_lock);
return LRU_REMOVED;
}
/**
dcache: use a dispose list in select_parent select_parent currently abuses the dentry cache LRU to provide cleanup features for child dentries that need to be freed. It moves them to the tail of the LRU, then tells shrink_dcache_parent() to calls __shrink_dcache_sb to unconditionally move them to a dispose list (as DCACHE_REFERENCED is ignored). __shrink_dcache_sb() has to relock the dentries to move them off the LRU onto the dispose list, but otherwise does not touch the dentries that select_parent() moved to the tail of the LRU. It then passses the dispose list to shrink_dentry_list() which tries to free the dentries. IOWs, the use of __shrink_dcache_sb() is superfluous - we can build exactly the same list of dentries for disposal directly in select_parent() and call shrink_dentry_list() instead of calling __shrink_dcache_sb() to do that. This means that we avoid long holds on the lru lock walking the LRU moving dentries to the dispose list We also avoid the need to relock each dentry just to move it off the LRU, reducing the numebr of times we lock each dentry to dispose of them in shrink_dcache_parent() from 3 to 2 times. Further, we remove one of the two callers of __shrink_dcache_sb(). This also means that __shrink_dcache_sb can be moved into back into prune_dcache_sb() and we no longer have to handle referenced dentries conditionally, simplifying the code. Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-08-23 16:56:24 +08:00
* prune_dcache_sb - shrink the dcache
* @sb: superblock
list_lru: introduce list_lru_shrink_{count,walk} Kmem accounting of memcg is unusable now, because it lacks slab shrinker support. That means when we hit the limit we will get ENOMEM w/o any chance to recover. What we should do then is to call shrink_slab, which would reclaim old inode/dentry caches from this cgroup. This is what this patch set is intended to do. Basically, it does two things. First, it introduces the notion of per-memcg slab shrinker. A shrinker that wants to reclaim objects per cgroup should mark itself as SHRINKER_MEMCG_AWARE. Then it will be passed the memory cgroup to scan from in shrink_control->memcg. For such shrinkers shrink_slab iterates over the whole cgroup subtree under the target cgroup and calls the shrinker for each kmem-active memory cgroup. Secondly, this patch set makes the list_lru structure per-memcg. It's done transparently to list_lru users - everything they have to do is to tell list_lru_init that they want memcg-aware list_lru. Then the list_lru will automatically distribute objects among per-memcg lists basing on which cgroup the object is accounted to. This way to make FS shrinkers (icache, dcache) memcg-aware we only need to make them use memcg-aware list_lru, and this is what this patch set does. As before, this patch set only enables per-memcg kmem reclaim when the pressure goes from memory.limit, not from memory.kmem.limit. Handling memory.kmem.limit is going to be tricky due to GFP_NOFS allocations, and it is still unclear whether we will have this knob in the unified hierarchy. This patch (of 9): NUMA aware slab shrinkers use the list_lru structure to distribute objects coming from different NUMA nodes to different lists. Whenever such a shrinker needs to count or scan objects from a particular node, it issues commands like this: count = list_lru_count_node(lru, sc->nid); freed = list_lru_walk_node(lru, sc->nid, isolate_func, isolate_arg, &sc->nr_to_scan); where sc is an instance of the shrink_control structure passed to it from vmscan. To simplify this, let's add special list_lru functions to be used by shrinkers, list_lru_shrink_count() and list_lru_shrink_walk(), which consolidate the nid and nr_to_scan arguments in the shrink_control structure. This will also allow us to avoid patching shrinkers that use list_lru when we make shrink_slab() per-memcg - all we will have to do is extend the shrink_control structure to include the target memcg and make list_lru_shrink_{count,walk} handle this appropriately. Signed-off-by: Vladimir Davydov <vdavydov@parallels.com> Suggested-by: Dave Chinner <david@fromorbit.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@suse.cz> Cc: Greg Thelen <gthelen@google.com> Cc: Glauber Costa <glommer@gmail.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Tejun Heo <tj@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-02-13 06:58:47 +08:00
* @sc: shrink control, passed to list_lru_shrink_walk()
dcache: use a dispose list in select_parent select_parent currently abuses the dentry cache LRU to provide cleanup features for child dentries that need to be freed. It moves them to the tail of the LRU, then tells shrink_dcache_parent() to calls __shrink_dcache_sb to unconditionally move them to a dispose list (as DCACHE_REFERENCED is ignored). __shrink_dcache_sb() has to relock the dentries to move them off the LRU onto the dispose list, but otherwise does not touch the dentries that select_parent() moved to the tail of the LRU. It then passses the dispose list to shrink_dentry_list() which tries to free the dentries. IOWs, the use of __shrink_dcache_sb() is superfluous - we can build exactly the same list of dentries for disposal directly in select_parent() and call shrink_dentry_list() instead of calling __shrink_dcache_sb() to do that. This means that we avoid long holds on the lru lock walking the LRU moving dentries to the dispose list We also avoid the need to relock each dentry just to move it off the LRU, reducing the numebr of times we lock each dentry to dispose of them in shrink_dcache_parent() from 3 to 2 times. Further, we remove one of the two callers of __shrink_dcache_sb(). This also means that __shrink_dcache_sb can be moved into back into prune_dcache_sb() and we no longer have to handle referenced dentries conditionally, simplifying the code. Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-08-23 16:56:24 +08:00
*
list_lru: introduce list_lru_shrink_{count,walk} Kmem accounting of memcg is unusable now, because it lacks slab shrinker support. That means when we hit the limit we will get ENOMEM w/o any chance to recover. What we should do then is to call shrink_slab, which would reclaim old inode/dentry caches from this cgroup. This is what this patch set is intended to do. Basically, it does two things. First, it introduces the notion of per-memcg slab shrinker. A shrinker that wants to reclaim objects per cgroup should mark itself as SHRINKER_MEMCG_AWARE. Then it will be passed the memory cgroup to scan from in shrink_control->memcg. For such shrinkers shrink_slab iterates over the whole cgroup subtree under the target cgroup and calls the shrinker for each kmem-active memory cgroup. Secondly, this patch set makes the list_lru structure per-memcg. It's done transparently to list_lru users - everything they have to do is to tell list_lru_init that they want memcg-aware list_lru. Then the list_lru will automatically distribute objects among per-memcg lists basing on which cgroup the object is accounted to. This way to make FS shrinkers (icache, dcache) memcg-aware we only need to make them use memcg-aware list_lru, and this is what this patch set does. As before, this patch set only enables per-memcg kmem reclaim when the pressure goes from memory.limit, not from memory.kmem.limit. Handling memory.kmem.limit is going to be tricky due to GFP_NOFS allocations, and it is still unclear whether we will have this knob in the unified hierarchy. This patch (of 9): NUMA aware slab shrinkers use the list_lru structure to distribute objects coming from different NUMA nodes to different lists. Whenever such a shrinker needs to count or scan objects from a particular node, it issues commands like this: count = list_lru_count_node(lru, sc->nid); freed = list_lru_walk_node(lru, sc->nid, isolate_func, isolate_arg, &sc->nr_to_scan); where sc is an instance of the shrink_control structure passed to it from vmscan. To simplify this, let's add special list_lru functions to be used by shrinkers, list_lru_shrink_count() and list_lru_shrink_walk(), which consolidate the nid and nr_to_scan arguments in the shrink_control structure. This will also allow us to avoid patching shrinkers that use list_lru when we make shrink_slab() per-memcg - all we will have to do is extend the shrink_control structure to include the target memcg and make list_lru_shrink_{count,walk} handle this appropriately. Signed-off-by: Vladimir Davydov <vdavydov@parallels.com> Suggested-by: Dave Chinner <david@fromorbit.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@suse.cz> Cc: Greg Thelen <gthelen@google.com> Cc: Glauber Costa <glommer@gmail.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Tejun Heo <tj@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-02-13 06:58:47 +08:00
* Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
* is done when we need more memory and called from the superblock shrinker
dcache: use a dispose list in select_parent select_parent currently abuses the dentry cache LRU to provide cleanup features for child dentries that need to be freed. It moves them to the tail of the LRU, then tells shrink_dcache_parent() to calls __shrink_dcache_sb to unconditionally move them to a dispose list (as DCACHE_REFERENCED is ignored). __shrink_dcache_sb() has to relock the dentries to move them off the LRU onto the dispose list, but otherwise does not touch the dentries that select_parent() moved to the tail of the LRU. It then passses the dispose list to shrink_dentry_list() which tries to free the dentries. IOWs, the use of __shrink_dcache_sb() is superfluous - we can build exactly the same list of dentries for disposal directly in select_parent() and call shrink_dentry_list() instead of calling __shrink_dcache_sb() to do that. This means that we avoid long holds on the lru lock walking the LRU moving dentries to the dispose list We also avoid the need to relock each dentry just to move it off the LRU, reducing the numebr of times we lock each dentry to dispose of them in shrink_dcache_parent() from 3 to 2 times. Further, we remove one of the two callers of __shrink_dcache_sb(). This also means that __shrink_dcache_sb can be moved into back into prune_dcache_sb() and we no longer have to handle referenced dentries conditionally, simplifying the code. Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-08-23 16:56:24 +08:00
* function.
*
dcache: use a dispose list in select_parent select_parent currently abuses the dentry cache LRU to provide cleanup features for child dentries that need to be freed. It moves them to the tail of the LRU, then tells shrink_dcache_parent() to calls __shrink_dcache_sb to unconditionally move them to a dispose list (as DCACHE_REFERENCED is ignored). __shrink_dcache_sb() has to relock the dentries to move them off the LRU onto the dispose list, but otherwise does not touch the dentries that select_parent() moved to the tail of the LRU. It then passses the dispose list to shrink_dentry_list() which tries to free the dentries. IOWs, the use of __shrink_dcache_sb() is superfluous - we can build exactly the same list of dentries for disposal directly in select_parent() and call shrink_dentry_list() instead of calling __shrink_dcache_sb() to do that. This means that we avoid long holds on the lru lock walking the LRU moving dentries to the dispose list We also avoid the need to relock each dentry just to move it off the LRU, reducing the numebr of times we lock each dentry to dispose of them in shrink_dcache_parent() from 3 to 2 times. Further, we remove one of the two callers of __shrink_dcache_sb(). This also means that __shrink_dcache_sb can be moved into back into prune_dcache_sb() and we no longer have to handle referenced dentries conditionally, simplifying the code. Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-08-23 16:56:24 +08:00
* This function may fail to free any resources if all the dentries are in
* use.
*/
list_lru: introduce list_lru_shrink_{count,walk} Kmem accounting of memcg is unusable now, because it lacks slab shrinker support. That means when we hit the limit we will get ENOMEM w/o any chance to recover. What we should do then is to call shrink_slab, which would reclaim old inode/dentry caches from this cgroup. This is what this patch set is intended to do. Basically, it does two things. First, it introduces the notion of per-memcg slab shrinker. A shrinker that wants to reclaim objects per cgroup should mark itself as SHRINKER_MEMCG_AWARE. Then it will be passed the memory cgroup to scan from in shrink_control->memcg. For such shrinkers shrink_slab iterates over the whole cgroup subtree under the target cgroup and calls the shrinker for each kmem-active memory cgroup. Secondly, this patch set makes the list_lru structure per-memcg. It's done transparently to list_lru users - everything they have to do is to tell list_lru_init that they want memcg-aware list_lru. Then the list_lru will automatically distribute objects among per-memcg lists basing on which cgroup the object is accounted to. This way to make FS shrinkers (icache, dcache) memcg-aware we only need to make them use memcg-aware list_lru, and this is what this patch set does. As before, this patch set only enables per-memcg kmem reclaim when the pressure goes from memory.limit, not from memory.kmem.limit. Handling memory.kmem.limit is going to be tricky due to GFP_NOFS allocations, and it is still unclear whether we will have this knob in the unified hierarchy. This patch (of 9): NUMA aware slab shrinkers use the list_lru structure to distribute objects coming from different NUMA nodes to different lists. Whenever such a shrinker needs to count or scan objects from a particular node, it issues commands like this: count = list_lru_count_node(lru, sc->nid); freed = list_lru_walk_node(lru, sc->nid, isolate_func, isolate_arg, &sc->nr_to_scan); where sc is an instance of the shrink_control structure passed to it from vmscan. To simplify this, let's add special list_lru functions to be used by shrinkers, list_lru_shrink_count() and list_lru_shrink_walk(), which consolidate the nid and nr_to_scan arguments in the shrink_control structure. This will also allow us to avoid patching shrinkers that use list_lru when we make shrink_slab() per-memcg - all we will have to do is extend the shrink_control structure to include the target memcg and make list_lru_shrink_{count,walk} handle this appropriately. Signed-off-by: Vladimir Davydov <vdavydov@parallels.com> Suggested-by: Dave Chinner <david@fromorbit.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@suse.cz> Cc: Greg Thelen <gthelen@google.com> Cc: Glauber Costa <glommer@gmail.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Tejun Heo <tj@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-02-13 06:58:47 +08:00
long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
{
LIST_HEAD(dispose);
long freed;
list_lru: introduce list_lru_shrink_{count,walk} Kmem accounting of memcg is unusable now, because it lacks slab shrinker support. That means when we hit the limit we will get ENOMEM w/o any chance to recover. What we should do then is to call shrink_slab, which would reclaim old inode/dentry caches from this cgroup. This is what this patch set is intended to do. Basically, it does two things. First, it introduces the notion of per-memcg slab shrinker. A shrinker that wants to reclaim objects per cgroup should mark itself as SHRINKER_MEMCG_AWARE. Then it will be passed the memory cgroup to scan from in shrink_control->memcg. For such shrinkers shrink_slab iterates over the whole cgroup subtree under the target cgroup and calls the shrinker for each kmem-active memory cgroup. Secondly, this patch set makes the list_lru structure per-memcg. It's done transparently to list_lru users - everything they have to do is to tell list_lru_init that they want memcg-aware list_lru. Then the list_lru will automatically distribute objects among per-memcg lists basing on which cgroup the object is accounted to. This way to make FS shrinkers (icache, dcache) memcg-aware we only need to make them use memcg-aware list_lru, and this is what this patch set does. As before, this patch set only enables per-memcg kmem reclaim when the pressure goes from memory.limit, not from memory.kmem.limit. Handling memory.kmem.limit is going to be tricky due to GFP_NOFS allocations, and it is still unclear whether we will have this knob in the unified hierarchy. This patch (of 9): NUMA aware slab shrinkers use the list_lru structure to distribute objects coming from different NUMA nodes to different lists. Whenever such a shrinker needs to count or scan objects from a particular node, it issues commands like this: count = list_lru_count_node(lru, sc->nid); freed = list_lru_walk_node(lru, sc->nid, isolate_func, isolate_arg, &sc->nr_to_scan); where sc is an instance of the shrink_control structure passed to it from vmscan. To simplify this, let's add special list_lru functions to be used by shrinkers, list_lru_shrink_count() and list_lru_shrink_walk(), which consolidate the nid and nr_to_scan arguments in the shrink_control structure. This will also allow us to avoid patching shrinkers that use list_lru when we make shrink_slab() per-memcg - all we will have to do is extend the shrink_control structure to include the target memcg and make list_lru_shrink_{count,walk} handle this appropriately. Signed-off-by: Vladimir Davydov <vdavydov@parallels.com> Suggested-by: Dave Chinner <david@fromorbit.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@suse.cz> Cc: Greg Thelen <gthelen@google.com> Cc: Glauber Costa <glommer@gmail.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Tejun Heo <tj@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-02-13 06:58:47 +08:00
freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
dentry_lru_isolate, &dispose);
shrink_dentry_list(&dispose);
shrinker: convert superblock shrinkers to new API Convert superblock shrinker to use the new count/scan API, and propagate the API changes through to the filesystem callouts. The filesystem callouts already use a count/scan API, so it's just changing counters to longs to match the VM API. This requires the dentry and inode shrinker callouts to be converted to the count/scan API. This is mainly a mechanical change. [glommer@openvz.org: use mult_frac for fractional proportions, build fixes] Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Glauber Costa <glommer@openvz.org> Acked-by: Mel Gorman <mgorman@suse.de> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com> Cc: Arve Hjønnevåg <arve@android.com> Cc: Carlos Maiolino <cmaiolino@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Chuck Lever <chuck.lever@oracle.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: David Rientjes <rientjes@google.com> Cc: Gleb Natapov <gleb@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: J. Bruce Fields <bfields@redhat.com> Cc: Jan Kara <jack@suse.cz> Cc: Jerome Glisse <jglisse@redhat.com> Cc: John Stultz <john.stultz@linaro.org> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Kent Overstreet <koverstreet@google.com> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Steven Whitehouse <swhiteho@redhat.com> Cc: Thomas Hellstrom <thellstrom@vmware.com> Cc: Trond Myklebust <Trond.Myklebust@netapp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:17:57 +08:00
return freed;
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries [Summary] Split LRU-list of unused dentries to one per superblock to avoid soft lock up during NFS mounts and remounting of any filesystem. Previously I posted here: http://lkml.org/lkml/2008/3/5/590 [Descriptions] - background dentry_unused is a list of dentries which are not referenced. dentry_unused grows up when references on directories or files are released. This list can be very long if there is huge free memory. - the problem When shrink_dcache_sb() is called, it scans all dentry_unused linearly under spin_lock(), and if dentry->d_sb is differnt from given superblock, scan next dentry. This scan costs very much if there are many entries, and very ineffective if there are many superblocks. IOW, When we need to shrink unused dentries on one dentry, but scans unused dentries on all superblocks in the system. For example, we scan 500 dentries to unmount a filesystem, but scans 1,000,000 or more unused dentries on other superblocks. In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink unused dentries on NFS, but scans 100,000,000 unused dentries on superblocks in the system such as local ext3 filesystems. I hear NFS mounting took 1 min on some system in use. * : NFS uses virtual filesystem in rpc layer, so NFS is affected by this problem. 100,000,000 is possible number on large systems. Per-superblock LRU of unused dentried can reduce the cost in reasonable manner. - How to fix I found this problem is solved by David Chinner's "Per-superblock unused dentry LRU lists V3"(1), so I rebase it and add some fix to reclaim with fairness, which is in Andrew Morton's comments(2). 1) http://lkml.org/lkml/2006/5/25/318 2) http://lkml.org/lkml/2006/5/25/320 Split LRU-list of unused dentries to each superblocks. Then, NFS mounting will check dentries under a superblock instead of all. But this spliting will break LRU of dentry-unused. So, I've attempted to make reclaim unused dentrins with fairness by calculate number of dentries to scan on this sb based on following way number of dentries to scan on this sb = count * (number of dentries on this sb / number of dentries in the machine) - ToDo - I have to measuring performance number and do stress tests. - When unmount occurs during prune_dcache(), scanning on same superblock, It is unable to reach next superblock because it is gone away. We restart scannig superblock from first one, it causes unfairness of reclaim unused dentries on first superblock. But I think this happens very rarely. - Test Results Result on 6GB boxes with excessive unused dentries. Without patch: $ cat /proc/sys/fs/dentry-state 10181835 10180203 45 0 0 0 # mount -t nfs 10.124.60.70:/work/kernel-src nfs real 0m1.830s user 0m0.001s sys 0m1.653s With this patch: $ cat /proc/sys/fs/dentry-state 10236610 10234751 45 0 0 0 # mount -t nfs 10.124.60.70:/work/kernel-src nfs real 0m0.106s user 0m0.002s sys 0m0.032s [akpm@linux-foundation.org: fix comments] Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com> Cc: Neil Brown <neilb@suse.de> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Cc: David Chinner <dgc@sgi.com> Cc: "J. Bruce Fields" <bfields@fieldses.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
}
list_lru: remove special case function list_lru_dispose_all. The list_lru implementation has one function, list_lru_dispose_all, with only one user (the dentry code). At first, such function appears to make sense because we are really not interested in the result of isolating each dentry separately - all of them are going away anyway. However, it's implementation is buggy in the following way: When we call list_lru_dispose_all in fs/dcache.c, we scan all dentries marking them with DCACHE_SHRINK_LIST. However, this is done without the nlru->lock taken. The imediate result of that is that someone else may add or remove the dentry from the LRU at the same time. When list_lru_del happens in that scenario we will see an element that is not yet marked with DCACHE_SHRINK_LIST (even though it will be in the future) and obviously remove it from an lru where the element no longer is. Since list_lru_dispose_all will in effect count down nlru's nr_items and list_lru_del will do the same, this will lead to an imbalance. The solution for this would not be so simple: we can obviously just keep the lru_lock taken, but then we have no guarantees that we will be able to acquire the dentry lock (dentry->d_lock). To properly solve this, we need a communication mechanism between the lru and dentry code, so they can coordinate this with each other. Such mechanism already exists in the form of the list_lru_walk_cb callback. So it is possible to construct a dcache-side prune function that does the right thing only by calling list_lru_walk in a loop until no more dentries are available. With only one user, plus the fact that a sane solution for the problem would involve boucing between dcache and list_lru anyway, I see little justification to keep the special case list_lru_dispose_all in tree. Signed-off-by: Glauber Costa <glommer@openvz.org> Cc: Michal Hocko <mhocko@suse.cz> Acked-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:18:03 +08:00
static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
list_lru: add helpers to isolate items Currently, the isolate callback passed to the list_lru_walk family of functions is supposed to just delete an item from the list upon returning LRU_REMOVED or LRU_REMOVED_RETRY, while nr_items counter is fixed by __list_lru_walk_one after the callback returns. Since the callback is allowed to drop the lock after removing an item (it has to return LRU_REMOVED_RETRY then), the nr_items can be less than the actual number of elements on the list even if we check them under the lock. This makes it difficult to move items from one list_lru_one to another, which is required for per-memcg list_lru reparenting - we can't just splice the lists, we have to move entries one by one. This patch therefore introduces helpers that must be used by callback functions to isolate items instead of raw list_del/list_move. These are list_lru_isolate and list_lru_isolate_move. They not only remove the entry from the list, but also fix the nr_items counter, making sure nr_items always reflects the actual number of elements on the list if checked under the appropriate lock. Signed-off-by: Vladimir Davydov <vdavydov@parallels.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@suse.cz> Cc: Tejun Heo <tj@kernel.org> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Dave Chinner <david@fromorbit.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-02-13 06:59:35 +08:00
struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
dcache: remove dentries from LRU before putting on dispose list One of the big problems with modifying the way the dcache shrinker and LRU implementation works is that the LRU is abused in several ways. One of these is shrink_dentry_list(). Basically, we can move a dentry off the LRU onto a different list without doing any accounting changes, and then use dentry_lru_prune() to remove it from what-ever list it is now on to do the LRU accounting at that point. This makes it -really hard- to change the LRU implementation. The use of the per-sb LRU lock serialises movement of the dentries between the different lists and the removal of them, and this is the only reason that it works. If we want to break up the dentry LRU lock and lists into, say, per-node lists, we remove the only serialisation that allows this lru list/dispose list abuse to work. To make this work effectively, the dispose list has to be isolated from the LRU list - dentries have to be removed from the LRU *before* being placed on the dispose list. This means that the LRU accounting and isolation is completed before disposal is started, and that means we can change the LRU implementation freely in future. This means that dentries *must* be marked with DCACHE_SHRINK_LIST when they are placed on the dispose list so that we don't think that parent dentries found in try_prune_one_dentry() are on the LRU when the are actually on the dispose list. This would result in accounting the dentry to the LRU a second time. Hence dentry_lru_del() has to handle the DCACHE_SHRINK_LIST case Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Glauber Costa <glommer@openvz.org> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com> Cc: Arve Hjønnevåg <arve@android.com> Cc: Carlos Maiolino <cmaiolino@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Chuck Lever <chuck.lever@oracle.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: David Rientjes <rientjes@google.com> Cc: Gleb Natapov <gleb@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: J. Bruce Fields <bfields@redhat.com> Cc: Jan Kara <jack@suse.cz> Cc: Jerome Glisse <jglisse@redhat.com> Cc: John Stultz <john.stultz@linaro.org> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Kent Overstreet <koverstreet@google.com> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Steven Whitehouse <swhiteho@redhat.com> Cc: Thomas Hellstrom <thellstrom@vmware.com> Cc: Trond Myklebust <Trond.Myklebust@netapp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:17:55 +08:00
{
list_lru: remove special case function list_lru_dispose_all. The list_lru implementation has one function, list_lru_dispose_all, with only one user (the dentry code). At first, such function appears to make sense because we are really not interested in the result of isolating each dentry separately - all of them are going away anyway. However, it's implementation is buggy in the following way: When we call list_lru_dispose_all in fs/dcache.c, we scan all dentries marking them with DCACHE_SHRINK_LIST. However, this is done without the nlru->lock taken. The imediate result of that is that someone else may add or remove the dentry from the LRU at the same time. When list_lru_del happens in that scenario we will see an element that is not yet marked with DCACHE_SHRINK_LIST (even though it will be in the future) and obviously remove it from an lru where the element no longer is. Since list_lru_dispose_all will in effect count down nlru's nr_items and list_lru_del will do the same, this will lead to an imbalance. The solution for this would not be so simple: we can obviously just keep the lru_lock taken, but then we have no guarantees that we will be able to acquire the dentry lock (dentry->d_lock). To properly solve this, we need a communication mechanism between the lru and dentry code, so they can coordinate this with each other. Such mechanism already exists in the form of the list_lru_walk_cb callback. So it is possible to construct a dcache-side prune function that does the right thing only by calling list_lru_walk in a loop until no more dentries are available. With only one user, plus the fact that a sane solution for the problem would involve boucing between dcache and list_lru anyway, I see little justification to keep the special case list_lru_dispose_all in tree. Signed-off-by: Glauber Costa <glommer@openvz.org> Cc: Michal Hocko <mhocko@suse.cz> Acked-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:18:03 +08:00
struct list_head *freeable = arg;
struct dentry *dentry = container_of(item, struct dentry, d_lru);
dcache: remove dentries from LRU before putting on dispose list One of the big problems with modifying the way the dcache shrinker and LRU implementation works is that the LRU is abused in several ways. One of these is shrink_dentry_list(). Basically, we can move a dentry off the LRU onto a different list without doing any accounting changes, and then use dentry_lru_prune() to remove it from what-ever list it is now on to do the LRU accounting at that point. This makes it -really hard- to change the LRU implementation. The use of the per-sb LRU lock serialises movement of the dentries between the different lists and the removal of them, and this is the only reason that it works. If we want to break up the dentry LRU lock and lists into, say, per-node lists, we remove the only serialisation that allows this lru list/dispose list abuse to work. To make this work effectively, the dispose list has to be isolated from the LRU list - dentries have to be removed from the LRU *before* being placed on the dispose list. This means that the LRU accounting and isolation is completed before disposal is started, and that means we can change the LRU implementation freely in future. This means that dentries *must* be marked with DCACHE_SHRINK_LIST when they are placed on the dispose list so that we don't think that parent dentries found in try_prune_one_dentry() are on the LRU when the are actually on the dispose list. This would result in accounting the dentry to the LRU a second time. Hence dentry_lru_del() has to handle the DCACHE_SHRINK_LIST case Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Glauber Costa <glommer@openvz.org> Cc: "Theodore Ts'o" <tytso@mit.edu> Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: Artem Bityutskiy <artem.bityutskiy@linux.intel.com> Cc: Arve Hjønnevåg <arve@android.com> Cc: Carlos Maiolino <cmaiolino@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Chuck Lever <chuck.lever@oracle.com> Cc: Daniel Vetter <daniel.vetter@ffwll.ch> Cc: David Rientjes <rientjes@google.com> Cc: Gleb Natapov <gleb@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: J. Bruce Fields <bfields@redhat.com> Cc: Jan Kara <jack@suse.cz> Cc: Jerome Glisse <jglisse@redhat.com> Cc: John Stultz <john.stultz@linaro.org> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Kent Overstreet <koverstreet@google.com> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Steven Whitehouse <swhiteho@redhat.com> Cc: Thomas Hellstrom <thellstrom@vmware.com> Cc: Trond Myklebust <Trond.Myklebust@netapp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:17:55 +08:00
list_lru: remove special case function list_lru_dispose_all. The list_lru implementation has one function, list_lru_dispose_all, with only one user (the dentry code). At first, such function appears to make sense because we are really not interested in the result of isolating each dentry separately - all of them are going away anyway. However, it's implementation is buggy in the following way: When we call list_lru_dispose_all in fs/dcache.c, we scan all dentries marking them with DCACHE_SHRINK_LIST. However, this is done without the nlru->lock taken. The imediate result of that is that someone else may add or remove the dentry from the LRU at the same time. When list_lru_del happens in that scenario we will see an element that is not yet marked with DCACHE_SHRINK_LIST (even though it will be in the future) and obviously remove it from an lru where the element no longer is. Since list_lru_dispose_all will in effect count down nlru's nr_items and list_lru_del will do the same, this will lead to an imbalance. The solution for this would not be so simple: we can obviously just keep the lru_lock taken, but then we have no guarantees that we will be able to acquire the dentry lock (dentry->d_lock). To properly solve this, we need a communication mechanism between the lru and dentry code, so they can coordinate this with each other. Such mechanism already exists in the form of the list_lru_walk_cb callback. So it is possible to construct a dcache-side prune function that does the right thing only by calling list_lru_walk in a loop until no more dentries are available. With only one user, plus the fact that a sane solution for the problem would involve boucing between dcache and list_lru anyway, I see little justification to keep the special case list_lru_dispose_all in tree. Signed-off-by: Glauber Costa <glommer@openvz.org> Cc: Michal Hocko <mhocko@suse.cz> Acked-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:18:03 +08:00
/*
* we are inverting the lru lock/dentry->d_lock here,
* so use a trylock. If we fail to get the lock, just skip
* it
*/
if (!spin_trylock(&dentry->d_lock))
return LRU_SKIP;
list_lru: add helpers to isolate items Currently, the isolate callback passed to the list_lru_walk family of functions is supposed to just delete an item from the list upon returning LRU_REMOVED or LRU_REMOVED_RETRY, while nr_items counter is fixed by __list_lru_walk_one after the callback returns. Since the callback is allowed to drop the lock after removing an item (it has to return LRU_REMOVED_RETRY then), the nr_items can be less than the actual number of elements on the list even if we check them under the lock. This makes it difficult to move items from one list_lru_one to another, which is required for per-memcg list_lru reparenting - we can't just splice the lists, we have to move entries one by one. This patch therefore introduces helpers that must be used by callback functions to isolate items instead of raw list_del/list_move. These are list_lru_isolate and list_lru_isolate_move. They not only remove the entry from the list, but also fix the nr_items counter, making sure nr_items always reflects the actual number of elements on the list if checked under the appropriate lock. Signed-off-by: Vladimir Davydov <vdavydov@parallels.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@suse.cz> Cc: Tejun Heo <tj@kernel.org> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Dave Chinner <david@fromorbit.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-02-13 06:59:35 +08:00
d_lru_shrink_move(lru, dentry, freeable);
list_lru: remove special case function list_lru_dispose_all. The list_lru implementation has one function, list_lru_dispose_all, with only one user (the dentry code). At first, such function appears to make sense because we are really not interested in the result of isolating each dentry separately - all of them are going away anyway. However, it's implementation is buggy in the following way: When we call list_lru_dispose_all in fs/dcache.c, we scan all dentries marking them with DCACHE_SHRINK_LIST. However, this is done without the nlru->lock taken. The imediate result of that is that someone else may add or remove the dentry from the LRU at the same time. When list_lru_del happens in that scenario we will see an element that is not yet marked with DCACHE_SHRINK_LIST (even though it will be in the future) and obviously remove it from an lru where the element no longer is. Since list_lru_dispose_all will in effect count down nlru's nr_items and list_lru_del will do the same, this will lead to an imbalance. The solution for this would not be so simple: we can obviously just keep the lru_lock taken, but then we have no guarantees that we will be able to acquire the dentry lock (dentry->d_lock). To properly solve this, we need a communication mechanism between the lru and dentry code, so they can coordinate this with each other. Such mechanism already exists in the form of the list_lru_walk_cb callback. So it is possible to construct a dcache-side prune function that does the right thing only by calling list_lru_walk in a loop until no more dentries are available. With only one user, plus the fact that a sane solution for the problem would involve boucing between dcache and list_lru anyway, I see little justification to keep the special case list_lru_dispose_all in tree. Signed-off-by: Glauber Costa <glommer@openvz.org> Cc: Michal Hocko <mhocko@suse.cz> Acked-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:18:03 +08:00
spin_unlock(&dentry->d_lock);
list_lru: remove special case function list_lru_dispose_all. The list_lru implementation has one function, list_lru_dispose_all, with only one user (the dentry code). At first, such function appears to make sense because we are really not interested in the result of isolating each dentry separately - all of them are going away anyway. However, it's implementation is buggy in the following way: When we call list_lru_dispose_all in fs/dcache.c, we scan all dentries marking them with DCACHE_SHRINK_LIST. However, this is done without the nlru->lock taken. The imediate result of that is that someone else may add or remove the dentry from the LRU at the same time. When list_lru_del happens in that scenario we will see an element that is not yet marked with DCACHE_SHRINK_LIST (even though it will be in the future) and obviously remove it from an lru where the element no longer is. Since list_lru_dispose_all will in effect count down nlru's nr_items and list_lru_del will do the same, this will lead to an imbalance. The solution for this would not be so simple: we can obviously just keep the lru_lock taken, but then we have no guarantees that we will be able to acquire the dentry lock (dentry->d_lock). To properly solve this, we need a communication mechanism between the lru and dentry code, so they can coordinate this with each other. Such mechanism already exists in the form of the list_lru_walk_cb callback. So it is possible to construct a dcache-side prune function that does the right thing only by calling list_lru_walk in a loop until no more dentries are available. With only one user, plus the fact that a sane solution for the problem would involve boucing between dcache and list_lru anyway, I see little justification to keep the special case list_lru_dispose_all in tree. Signed-off-by: Glauber Costa <glommer@openvz.org> Cc: Michal Hocko <mhocko@suse.cz> Acked-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:18:03 +08:00
return LRU_REMOVED;
fix soft lock up at NFS mount via per-SB LRU-list of unused dentries [Summary] Split LRU-list of unused dentries to one per superblock to avoid soft lock up during NFS mounts and remounting of any filesystem. Previously I posted here: http://lkml.org/lkml/2008/3/5/590 [Descriptions] - background dentry_unused is a list of dentries which are not referenced. dentry_unused grows up when references on directories or files are released. This list can be very long if there is huge free memory. - the problem When shrink_dcache_sb() is called, it scans all dentry_unused linearly under spin_lock(), and if dentry->d_sb is differnt from given superblock, scan next dentry. This scan costs very much if there are many entries, and very ineffective if there are many superblocks. IOW, When we need to shrink unused dentries on one dentry, but scans unused dentries on all superblocks in the system. For example, we scan 500 dentries to unmount a filesystem, but scans 1,000,000 or more unused dentries on other superblocks. In our case , At mounting NFS*, shrink_dcache_sb() is called to shrink unused dentries on NFS, but scans 100,000,000 unused dentries on superblocks in the system such as local ext3 filesystems. I hear NFS mounting took 1 min on some system in use. * : NFS uses virtual filesystem in rpc layer, so NFS is affected by this problem. 100,000,000 is possible number on large systems. Per-superblock LRU of unused dentried can reduce the cost in reasonable manner. - How to fix I found this problem is solved by David Chinner's "Per-superblock unused dentry LRU lists V3"(1), so I rebase it and add some fix to reclaim with fairness, which is in Andrew Morton's comments(2). 1) http://lkml.org/lkml/2006/5/25/318 2) http://lkml.org/lkml/2006/5/25/320 Split LRU-list of unused dentries to each superblocks. Then, NFS mounting will check dentries under a superblock instead of all. But this spliting will break LRU of dentry-unused. So, I've attempted to make reclaim unused dentrins with fairness by calculate number of dentries to scan on this sb based on following way number of dentries to scan on this sb = count * (number of dentries on this sb / number of dentries in the machine) - ToDo - I have to measuring performance number and do stress tests. - When unmount occurs during prune_dcache(), scanning on same superblock, It is unable to reach next superblock because it is gone away. We restart scannig superblock from first one, it causes unfairness of reclaim unused dentries on first superblock. But I think this happens very rarely. - Test Results Result on 6GB boxes with excessive unused dentries. Without patch: $ cat /proc/sys/fs/dentry-state 10181835 10180203 45 0 0 0 # mount -t nfs 10.124.60.70:/work/kernel-src nfs real 0m1.830s user 0m0.001s sys 0m1.653s With this patch: $ cat /proc/sys/fs/dentry-state 10236610 10234751 45 0 0 0 # mount -t nfs 10.124.60.70:/work/kernel-src nfs real 0m0.106s user 0m0.002s sys 0m0.032s [akpm@linux-foundation.org: fix comments] Signed-off-by: Kentaro Makita <k-makita@np.css.fujitsu.com> Cc: Neil Brown <neilb@suse.de> Cc: Trond Myklebust <trond.myklebust@fys.uio.no> Cc: David Chinner <dgc@sgi.com> Cc: "J. Bruce Fields" <bfields@fieldses.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:27:13 +08:00
}
list_lru: remove special case function list_lru_dispose_all. The list_lru implementation has one function, list_lru_dispose_all, with only one user (the dentry code). At first, such function appears to make sense because we are really not interested in the result of isolating each dentry separately - all of them are going away anyway. However, it's implementation is buggy in the following way: When we call list_lru_dispose_all in fs/dcache.c, we scan all dentries marking them with DCACHE_SHRINK_LIST. However, this is done without the nlru->lock taken. The imediate result of that is that someone else may add or remove the dentry from the LRU at the same time. When list_lru_del happens in that scenario we will see an element that is not yet marked with DCACHE_SHRINK_LIST (even though it will be in the future) and obviously remove it from an lru where the element no longer is. Since list_lru_dispose_all will in effect count down nlru's nr_items and list_lru_del will do the same, this will lead to an imbalance. The solution for this would not be so simple: we can obviously just keep the lru_lock taken, but then we have no guarantees that we will be able to acquire the dentry lock (dentry->d_lock). To properly solve this, we need a communication mechanism between the lru and dentry code, so they can coordinate this with each other. Such mechanism already exists in the form of the list_lru_walk_cb callback. So it is possible to construct a dcache-side prune function that does the right thing only by calling list_lru_walk in a loop until no more dentries are available. With only one user, plus the fact that a sane solution for the problem would involve boucing between dcache and list_lru anyway, I see little justification to keep the special case list_lru_dispose_all in tree. Signed-off-by: Glauber Costa <glommer@openvz.org> Cc: Michal Hocko <mhocko@suse.cz> Acked-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:18:03 +08:00
/**
* shrink_dcache_sb - shrink dcache for a superblock
* @sb: superblock
*
* Shrink the dcache for the specified super block. This is used to free
* the dcache before unmounting a file system.
*/
void shrink_dcache_sb(struct super_block *sb)
{
list_lru: remove special case function list_lru_dispose_all. The list_lru implementation has one function, list_lru_dispose_all, with only one user (the dentry code). At first, such function appears to make sense because we are really not interested in the result of isolating each dentry separately - all of them are going away anyway. However, it's implementation is buggy in the following way: When we call list_lru_dispose_all in fs/dcache.c, we scan all dentries marking them with DCACHE_SHRINK_LIST. However, this is done without the nlru->lock taken. The imediate result of that is that someone else may add or remove the dentry from the LRU at the same time. When list_lru_del happens in that scenario we will see an element that is not yet marked with DCACHE_SHRINK_LIST (even though it will be in the future) and obviously remove it from an lru where the element no longer is. Since list_lru_dispose_all will in effect count down nlru's nr_items and list_lru_del will do the same, this will lead to an imbalance. The solution for this would not be so simple: we can obviously just keep the lru_lock taken, but then we have no guarantees that we will be able to acquire the dentry lock (dentry->d_lock). To properly solve this, we need a communication mechanism between the lru and dentry code, so they can coordinate this with each other. Such mechanism already exists in the form of the list_lru_walk_cb callback. So it is possible to construct a dcache-side prune function that does the right thing only by calling list_lru_walk in a loop until no more dentries are available. With only one user, plus the fact that a sane solution for the problem would involve boucing between dcache and list_lru anyway, I see little justification to keep the special case list_lru_dispose_all in tree. Signed-off-by: Glauber Costa <glommer@openvz.org> Cc: Michal Hocko <mhocko@suse.cz> Acked-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:18:03 +08:00
long freed;
do {
LIST_HEAD(dispose);
freed = list_lru_walk(&sb->s_dentry_lru,
dentry_lru_isolate_shrink, &dispose, UINT_MAX);
list_lru: remove special case function list_lru_dispose_all. The list_lru implementation has one function, list_lru_dispose_all, with only one user (the dentry code). At first, such function appears to make sense because we are really not interested in the result of isolating each dentry separately - all of them are going away anyway. However, it's implementation is buggy in the following way: When we call list_lru_dispose_all in fs/dcache.c, we scan all dentries marking them with DCACHE_SHRINK_LIST. However, this is done without the nlru->lock taken. The imediate result of that is that someone else may add or remove the dentry from the LRU at the same time. When list_lru_del happens in that scenario we will see an element that is not yet marked with DCACHE_SHRINK_LIST (even though it will be in the future) and obviously remove it from an lru where the element no longer is. Since list_lru_dispose_all will in effect count down nlru's nr_items and list_lru_del will do the same, this will lead to an imbalance. The solution for this would not be so simple: we can obviously just keep the lru_lock taken, but then we have no guarantees that we will be able to acquire the dentry lock (dentry->d_lock). To properly solve this, we need a communication mechanism between the lru and dentry code, so they can coordinate this with each other. Such mechanism already exists in the form of the list_lru_walk_cb callback. So it is possible to construct a dcache-side prune function that does the right thing only by calling list_lru_walk in a loop until no more dentries are available. With only one user, plus the fact that a sane solution for the problem would involve boucing between dcache and list_lru anyway, I see little justification to keep the special case list_lru_dispose_all in tree. Signed-off-by: Glauber Costa <glommer@openvz.org> Cc: Michal Hocko <mhocko@suse.cz> Acked-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-08-28 08:18:03 +08:00
this_cpu_sub(nr_dentry_unused, freed);
shrink_dentry_list(&dispose);
} while (freed > 0);
}
EXPORT_SYMBOL(shrink_dcache_sb);
/**
* enum d_walk_ret - action to talke during tree walk
* @D_WALK_CONTINUE: contrinue walk
* @D_WALK_QUIT: quit walk
* @D_WALK_NORETRY: quit when retry is needed
* @D_WALK_SKIP: skip this dentry and its children
*/
enum d_walk_ret {
D_WALK_CONTINUE,
D_WALK_QUIT,
D_WALK_NORETRY,
D_WALK_SKIP,
};
/**
* d_walk - walk the dentry tree
* @parent: start of walk
* @data: data passed to @enter() and @finish()
* @enter: callback when first entering the dentry
* @finish: callback when successfully finished the walk
*
* The @enter() and @finish() callbacks are called with d_lock held.
*/
static void d_walk(struct dentry *parent, void *data,
enum d_walk_ret (*enter)(void *, struct dentry *),
void (*finish)(void *))
{
struct dentry *this_parent;
struct list_head *next;
unsigned seq = 0;
enum d_walk_ret ret;
bool retry = true;
again:
read_seqbegin_or_lock(&rename_lock, &seq);
this_parent = parent;
spin_lock(&this_parent->d_lock);
ret = enter(data, this_parent);
switch (ret) {
case D_WALK_CONTINUE:
break;
case D_WALK_QUIT:
case D_WALK_SKIP:
goto out_unlock;
case D_WALK_NORETRY:
retry = false;
break;
}
repeat:
next = this_parent->d_subdirs.next;
resume:
while (next != &this_parent->d_subdirs) {
struct list_head *tmp = next;
struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
next = tmp->next;
spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
ret = enter(data, dentry);
switch (ret) {
case D_WALK_CONTINUE:
break;
case D_WALK_QUIT:
spin_unlock(&dentry->d_lock);
goto out_unlock;
case D_WALK_NORETRY:
retry = false;
break;
case D_WALK_SKIP:
spin_unlock(&dentry->d_lock);
continue;
}
if (!list_empty(&dentry->d_subdirs)) {
spin_unlock(&this_parent->d_lock);
spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
this_parent = dentry;
spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
goto repeat;
}
spin_unlock(&dentry->d_lock);
}
/*
* All done at this level ... ascend and resume the search.
*/
rcu_read_lock();
ascend:
if (this_parent != parent) {
struct dentry *child = this_parent;
this_parent = child->d_parent;
spin_unlock(&child->d_lock);
spin_lock(&this_parent->d_lock);
/* might go back up the wrong parent if we have had a rename. */
if (need_seqretry(&rename_lock, seq))
goto rename_retry;
/* go into the first sibling still alive */
do {
next = child->d_child.next;
if (next == &this_parent->d_subdirs)
goto ascend;
child = list_entry(next, struct dentry, d_child);
} while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
rcu_read_unlock();
goto resume;
}
if (need_seqretry(&rename_lock, seq))
goto rename_retry;
rcu_read_unlock();
if (finish)
finish(data);
out_unlock:
spin_unlock(&this_parent->d_lock);
done_seqretry(&rename_lock, seq);
return;
rename_retry:
spin_unlock(&this_parent->d_lock);
rcu_read_unlock();
BUG_ON(seq & 1);
if (!retry)
return;
seq = 1;
goto again;
}
/*
* Search for at least 1 mount point in the dentry's subdirs.
* We descend to the next level whenever the d_subdirs
* list is non-empty and continue searching.
*/
static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
{
int *ret = data;
if (d_mountpoint(dentry)) {
*ret = 1;
return D_WALK_QUIT;
}
return D_WALK_CONTINUE;
}
/**
* have_submounts - check for mounts over a dentry
* @parent: dentry to check.
*
* Return true if the parent or its subdirectories contain
* a mount point
*/
int have_submounts(struct dentry *parent)
{
int ret = 0;
d_walk(parent, &ret, check_mount, NULL);
return ret;
}
EXPORT_SYMBOL(have_submounts);
/*
* Called by mount code to set a mountpoint and check if the mountpoint is
* reachable (e.g. NFS can unhash a directory dentry and then the complete
* subtree can become unreachable).
*
* Only one of d_invalidate() and d_set_mounted() must succeed. For
* this reason take rename_lock and d_lock on dentry and ancestors.
*/
int d_set_mounted(struct dentry *dentry)
{
struct dentry *p;
int ret = -ENOENT;
write_seqlock(&rename_lock);
for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
/* Need exclusion wrt. d_invalidate() */
spin_lock(&p->d_lock);
if (unlikely(d_unhashed(p))) {
spin_unlock(&p->d_lock);
goto out;
}
spin_unlock(&p->d_lock);
}
spin_lock(&dentry->d_lock);
if (!d_unlinked(dentry)) {
dentry->d_flags |= DCACHE_MOUNTED;
ret = 0;
}
spin_unlock(&dentry->d_lock);
out:
write_sequnlock(&rename_lock);
return ret;
}
/*
* Search the dentry child list of the specified parent,
* and move any unused dentries to the end of the unused
* list for prune_dcache(). We descend to the next level
* whenever the d_subdirs list is non-empty and continue
* searching.
*
* It returns zero iff there are no unused children,
* otherwise it returns the number of children moved to
* the end of the unused list. This may not be the total
* number of unused children, because select_parent can
* drop the lock and return early due to latency
* constraints.
*/
struct select_data {
struct dentry *start;
struct list_head dispose;
int found;
};
static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
{
struct select_data *data = _data;
enum d_walk_ret ret = D_WALK_CONTINUE;
if (data->start == dentry)
goto out;
if (dentry->d_flags & DCACHE_SHRINK_LIST) {
data->found++;
} else {
if (dentry->d_flags & DCACHE_LRU_LIST)
d_lru_del(dentry);
if (!dentry->d_lockref.count) {
d_shrink_add(dentry, &data->dispose);
data->found++;
}
}
/*
* We can return to the caller if we have found some (this
* ensures forward progress). We'll be coming back to find
* the rest.
*/
if (!list_empty(&data->dispose))
ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
out:
return ret;
}
/**
* shrink_dcache_parent - prune dcache
* @parent: parent of entries to prune
*
* Prune the dcache to remove unused children of the parent dentry.
*/
void shrink_dcache_parent(struct dentry *parent)
{
for (;;) {
struct select_data data;
INIT_LIST_HEAD(&data.dispose);
data.start = parent;
data.found = 0;
d_walk(parent, &data, select_collect, NULL);
if (!data.found)
break;
shrink_dentry_list(&data.dispose);
fs/dcache.c: add cond_resched() to shrink_dcache_parent() Call cond_resched() in shrink_dcache_parent() to maintain interactivity. Before this patch: void shrink_dcache_parent(struct dentry * parent) { while ((found = select_parent(parent, &dispose)) != 0) shrink_dentry_list(&dispose); } select_parent() populates the dispose list with dentries which shrink_dentry_list() then deletes. select_parent() carefully uses need_resched() to avoid doing too much work at once. But neither shrink_dcache_parent() nor its called functions call cond_resched(). So once need_resched() is set select_parent() will return single dentry dispose list which is then deleted by shrink_dentry_list(). This is inefficient when there are a lot of dentry to process. This can cause softlockup and hurts interactivity on non preemptable kernels. This change adds cond_resched() in shrink_dcache_parent(). The benefit of this is that need_resched() is quickly cleared so that future calls to select_parent() are able to efficiently return a big batch of dentry. These additional cond_resched() do not seem to impact performance, at least for the workload below. Here is a program which can cause soft lockup if other system activity sets need_resched(). int main() { struct rlimit rlim; int i; int f[100000]; char buf[20]; struct timeval t1, t2; double diff; /* cleanup past run */ system("rm -rf x"); /* boost nfile rlimit */ rlim.rlim_cur = 200000; rlim.rlim_max = 200000; if (setrlimit(RLIMIT_NOFILE, &rlim)) err(1, "setrlimit"); /* make directory for files */ if (mkdir("x", 0700)) err(1, "mkdir"); if (gettimeofday(&t1, NULL)) err(1, "gettimeofday"); /* populate directory with open files */ for (i = 0; i < 100000; i++) { snprintf(buf, sizeof(buf), "x/%d", i); f[i] = open(buf, O_CREAT); if (f[i] == -1) err(1, "open"); } /* close some of the files */ for (i = 0; i < 85000; i++) close(f[i]); /* unlink all files, even open ones */ system("rm -rf x"); if (gettimeofday(&t2, NULL)) err(1, "gettimeofday"); diff = (((double)t2.tv_sec * 1000000 + t2.tv_usec) - ((double)t1.tv_sec * 1000000 + t1.tv_usec)); printf("done: %g elapsed\n", diff/1e6); return 0; } Signed-off-by: Greg Thelen <gthelen@google.com> Signed-off-by: Dave Chinner <david@fromorbit.com> Cc: <stable@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 06:26:48 +08:00
cond_resched();
}
}
EXPORT_SYMBOL(shrink_dcache_parent);
static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
{
/* it has busy descendents; complain about those instead */
if (!list_empty(&dentry->d_subdirs))
return D_WALK_CONTINUE;
/* root with refcount 1 is fine */
if (dentry == _data && dentry->d_lockref.count == 1)
return D_WALK_CONTINUE;
printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
" still in use (%d) [unmount of %s %s]\n",
dentry,
dentry->d_inode ?
dentry->d_inode->i_ino : 0UL,
dentry,
dentry->d_lockref.count,
dentry->d_sb->s_type->name,
dentry->d_sb->s_id);
WARN_ON(1);
return D_WALK_CONTINUE;
}
static void do_one_tree(struct dentry *dentry)
{
shrink_dcache_parent(dentry);
d_walk(dentry, dentry, umount_check, NULL);
d_drop(dentry);
dput(dentry);
}
/*
* destroy the dentries attached to a superblock on unmounting
*/
void shrink_dcache_for_umount(struct super_block *sb)
{
struct dentry *dentry;
WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
dentry = sb->s_root;
sb->s_root = NULL;
do_one_tree(dentry);
while (!hlist_bl_empty(&sb->s_anon)) {
dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
do_one_tree(dentry);
}
}
vfs: Lazily remove mounts on unlinked files and directories. With the introduction of mount namespaces and bind mounts it became possible to access files and directories that on some paths are mount points but are not mount points on other paths. It is very confusing when rm -rf somedir returns -EBUSY simply because somedir is mounted somewhere else. With the addition of user namespaces allowing unprivileged mounts this condition has gone from annoying to allowing a DOS attack on other users in the system. The possibility for mischief is removed by updating the vfs to support rename, unlink and rmdir on a dentry that is a mountpoint and by lazily unmounting mountpoints on deleted dentries. In particular this change allows rename, unlink and rmdir system calls on a dentry without a mountpoint in the current mount namespace to succeed, and it allows rename, unlink, and rmdir performed on a distributed filesystem to update the vfs cache even if when there is a mount in some namespace on the original dentry. There are two common patterns of maintaining mounts: Mounts on trusted paths with the parent directory of the mount point and all ancestory directories up to / owned by root and modifiable only by root (i.e. /media/xxx, /dev, /dev/pts, /proc, /sys, /sys/fs/cgroup/{cpu, cpuacct, ...}, /usr, /usr/local). Mounts on unprivileged directories maintained by fusermount. In the case of mounts in trusted directories owned by root and modifiable only by root the current parent directory permissions are sufficient to ensure a mount point on a trusted path is not removed or renamed by anyone other than root, even if there is a context where the there are no mount points to prevent this. In the case of mounts in directories owned by less privileged users races with users modifying the path of a mount point are already a danger. fusermount already uses a combination of chdir, /proc/<pid>/fd/NNN, and UMOUNT_NOFOLLOW to prevent these races. The removable of global rename, unlink, and rmdir protection really adds nothing new to consider only a widening of the attack window, and fusermount is already safe against unprivileged users modifying the directory simultaneously. In principle for perfect userspace programs returning -EBUSY for unlink, rmdir, and rename of dentires that have mounts in the local namespace is actually unnecessary. Unfortunately not all userspace programs are perfect so retaining -EBUSY for unlink, rmdir and rename of dentries that have mounts in the current mount namespace plays an important role of maintaining consistency with historical behavior and making imperfect userspace applications hard to exploit. v2: Remove spurious old_dentry. v3: Optimized shrink_submounts_and_drop Removed unsued afs label v4: Simplified the changes to check_submounts_and_drop Do not rename check_submounts_and_drop shrink_submounts_and_drop Document what why we need atomicity in check_submounts_and_drop Rely on the parent inode mutex to make d_revalidate and d_invalidate an atomic unit. v5: Refcount the mountpoint to detach in case of simultaneous renames. Reviewed-by: Miklos Szeredi <miklos@szeredi.hu> Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-10-02 09:33:48 +08:00
struct detach_data {
struct select_data select;
struct dentry *mountpoint;
};
static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
{
vfs: Lazily remove mounts on unlinked files and directories. With the introduction of mount namespaces and bind mounts it became possible to access files and directories that on some paths are mount points but are not mount points on other paths. It is very confusing when rm -rf somedir returns -EBUSY simply because somedir is mounted somewhere else. With the addition of user namespaces allowing unprivileged mounts this condition has gone from annoying to allowing a DOS attack on other users in the system. The possibility for mischief is removed by updating the vfs to support rename, unlink and rmdir on a dentry that is a mountpoint and by lazily unmounting mountpoints on deleted dentries. In particular this change allows rename, unlink and rmdir system calls on a dentry without a mountpoint in the current mount namespace to succeed, and it allows rename, unlink, and rmdir performed on a distributed filesystem to update the vfs cache even if when there is a mount in some namespace on the original dentry. There are two common patterns of maintaining mounts: Mounts on trusted paths with the parent directory of the mount point and all ancestory directories up to / owned by root and modifiable only by root (i.e. /media/xxx, /dev, /dev/pts, /proc, /sys, /sys/fs/cgroup/{cpu, cpuacct, ...}, /usr, /usr/local). Mounts on unprivileged directories maintained by fusermount. In the case of mounts in trusted directories owned by root and modifiable only by root the current parent directory permissions are sufficient to ensure a mount point on a trusted path is not removed or renamed by anyone other than root, even if there is a context where the there are no mount points to prevent this. In the case of mounts in directories owned by less privileged users races with users modifying the path of a mount point are already a danger. fusermount already uses a combination of chdir, /proc/<pid>/fd/NNN, and UMOUNT_NOFOLLOW to prevent these races. The removable of global rename, unlink, and rmdir protection really adds nothing new to consider only a widening of the attack window, and fusermount is already safe against unprivileged users modifying the directory simultaneously. In principle for perfect userspace programs returning -EBUSY for unlink, rmdir, and rename of dentires that have mounts in the local namespace is actually unnecessary. Unfortunately not all userspace programs are perfect so retaining -EBUSY for unlink, rmdir and rename of dentries that have mounts in the current mount namespace plays an important role of maintaining consistency with historical behavior and making imperfect userspace applications hard to exploit. v2: Remove spurious old_dentry. v3: Optimized shrink_submounts_and_drop Removed unsued afs label v4: Simplified the changes to check_submounts_and_drop Do not rename check_submounts_and_drop shrink_submounts_and_drop Document what why we need atomicity in check_submounts_and_drop Rely on the parent inode mutex to make d_revalidate and d_invalidate an atomic unit. v5: Refcount the mountpoint to detach in case of simultaneous renames. Reviewed-by: Miklos Szeredi <miklos@szeredi.hu> Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-10-02 09:33:48 +08:00
struct detach_data *data = _data;
if (d_mountpoint(dentry)) {
vfs: Lazily remove mounts on unlinked files and directories. With the introduction of mount namespaces and bind mounts it became possible to access files and directories that on some paths are mount points but are not mount points on other paths. It is very confusing when rm -rf somedir returns -EBUSY simply because somedir is mounted somewhere else. With the addition of user namespaces allowing unprivileged mounts this condition has gone from annoying to allowing a DOS attack on other users in the system. The possibility for mischief is removed by updating the vfs to support rename, unlink and rmdir on a dentry that is a mountpoint and by lazily unmounting mountpoints on deleted dentries. In particular this change allows rename, unlink and rmdir system calls on a dentry without a mountpoint in the current mount namespace to succeed, and it allows rename, unlink, and rmdir performed on a distributed filesystem to update the vfs cache even if when there is a mount in some namespace on the original dentry. There are two common patterns of maintaining mounts: Mounts on trusted paths with the parent directory of the mount point and all ancestory directories up to / owned by root and modifiable only by root (i.e. /media/xxx, /dev, /dev/pts, /proc, /sys, /sys/fs/cgroup/{cpu, cpuacct, ...}, /usr, /usr/local). Mounts on unprivileged directories maintained by fusermount. In the case of mounts in trusted directories owned by root and modifiable only by root the current parent directory permissions are sufficient to ensure a mount point on a trusted path is not removed or renamed by anyone other than root, even if there is a context where the there are no mount points to prevent this. In the case of mounts in directories owned by less privileged users races with users modifying the path of a mount point are already a danger. fusermount already uses a combination of chdir, /proc/<pid>/fd/NNN, and UMOUNT_NOFOLLOW to prevent these races. The removable of global rename, unlink, and rmdir protection really adds nothing new to consider only a widening of the attack window, and fusermount is already safe against unprivileged users modifying the directory simultaneously. In principle for perfect userspace programs returning -EBUSY for unlink, rmdir, and rename of dentires that have mounts in the local namespace is actually unnecessary. Unfortunately not all userspace programs are perfect so retaining -EBUSY for unlink, rmdir and rename of dentries that have mounts in the current mount namespace plays an important role of maintaining consistency with historical behavior and making imperfect userspace applications hard to exploit. v2: Remove spurious old_dentry. v3: Optimized shrink_submounts_and_drop Removed unsued afs label v4: Simplified the changes to check_submounts_and_drop Do not rename check_submounts_and_drop shrink_submounts_and_drop Document what why we need atomicity in check_submounts_and_drop Rely on the parent inode mutex to make d_revalidate and d_invalidate an atomic unit. v5: Refcount the mountpoint to detach in case of simultaneous renames. Reviewed-by: Miklos Szeredi <miklos@szeredi.hu> Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-10-02 09:33:48 +08:00
__dget_dlock(dentry);
data->mountpoint = dentry;
return D_WALK_QUIT;
}
vfs: Lazily remove mounts on unlinked files and directories. With the introduction of mount namespaces and bind mounts it became possible to access files and directories that on some paths are mount points but are not mount points on other paths. It is very confusing when rm -rf somedir returns -EBUSY simply because somedir is mounted somewhere else. With the addition of user namespaces allowing unprivileged mounts this condition has gone from annoying to allowing a DOS attack on other users in the system. The possibility for mischief is removed by updating the vfs to support rename, unlink and rmdir on a dentry that is a mountpoint and by lazily unmounting mountpoints on deleted dentries. In particular this change allows rename, unlink and rmdir system calls on a dentry without a mountpoint in the current mount namespace to succeed, and it allows rename, unlink, and rmdir performed on a distributed filesystem to update the vfs cache even if when there is a mount in some namespace on the original dentry. There are two common patterns of maintaining mounts: Mounts on trusted paths with the parent directory of the mount point and all ancestory directories up to / owned by root and modifiable only by root (i.e. /media/xxx, /dev, /dev/pts, /proc, /sys, /sys/fs/cgroup/{cpu, cpuacct, ...}, /usr, /usr/local). Mounts on unprivileged directories maintained by fusermount. In the case of mounts in trusted directories owned by root and modifiable only by root the current parent directory permissions are sufficient to ensure a mount point on a trusted path is not removed or renamed by anyone other than root, even if there is a context where the there are no mount points to prevent this. In the case of mounts in directories owned by less privileged users races with users modifying the path of a mount point are already a danger. fusermount already uses a combination of chdir, /proc/<pid>/fd/NNN, and UMOUNT_NOFOLLOW to prevent these races. The removable of global rename, unlink, and rmdir protection really adds nothing new to consider only a widening of the attack window, and fusermount is already safe against unprivileged users modifying the directory simultaneously. In principle for perfect userspace programs returning -EBUSY for unlink, rmdir, and rename of dentires that have mounts in the local namespace is actually unnecessary. Unfortunately not all userspace programs are perfect so retaining -EBUSY for unlink, rmdir and rename of dentries that have mounts in the current mount namespace plays an important role of maintaining consistency with historical behavior and making imperfect userspace applications hard to exploit. v2: Remove spurious old_dentry. v3: Optimized shrink_submounts_and_drop Removed unsued afs label v4: Simplified the changes to check_submounts_and_drop Do not rename check_submounts_and_drop shrink_submounts_and_drop Document what why we need atomicity in check_submounts_and_drop Rely on the parent inode mutex to make d_revalidate and d_invalidate an atomic unit. v5: Refcount the mountpoint to detach in case of simultaneous renames. Reviewed-by: Miklos Szeredi <miklos@szeredi.hu> Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-10-02 09:33:48 +08:00
return select_collect(&data->select, dentry);
}
static void check_and_drop(void *_data)
{
vfs: Lazily remove mounts on unlinked files and directories. With the introduction of mount namespaces and bind mounts it became possible to access files and directories that on some paths are mount points but are not mount points on other paths. It is very confusing when rm -rf somedir returns -EBUSY simply because somedir is mounted somewhere else. With the addition of user namespaces allowing unprivileged mounts this condition has gone from annoying to allowing a DOS attack on other users in the system. The possibility for mischief is removed by updating the vfs to support rename, unlink and rmdir on a dentry that is a mountpoint and by lazily unmounting mountpoints on deleted dentries. In particular this change allows rename, unlink and rmdir system calls on a dentry without a mountpoint in the current mount namespace to succeed, and it allows rename, unlink, and rmdir performed on a distributed filesystem to update the vfs cache even if when there is a mount in some namespace on the original dentry. There are two common patterns of maintaining mounts: Mounts on trusted paths with the parent directory of the mount point and all ancestory directories up to / owned by root and modifiable only by root (i.e. /media/xxx, /dev, /dev/pts, /proc, /sys, /sys/fs/cgroup/{cpu, cpuacct, ...}, /usr, /usr/local). Mounts on unprivileged directories maintained by fusermount. In the case of mounts in trusted directories owned by root and modifiable only by root the current parent directory permissions are sufficient to ensure a mount point on a trusted path is not removed or renamed by anyone other than root, even if there is a context where the there are no mount points to prevent this. In the case of mounts in directories owned by less privileged users races with users modifying the path of a mount point are already a danger. fusermount already uses a combination of chdir, /proc/<pid>/fd/NNN, and UMOUNT_NOFOLLOW to prevent these races. The removable of global rename, unlink, and rmdir protection really adds nothing new to consider only a widening of the attack window, and fusermount is already safe against unprivileged users modifying the directory simultaneously. In principle for perfect userspace programs returning -EBUSY for unlink, rmdir, and rename of dentires that have mounts in the local namespace is actually unnecessary. Unfortunately not all userspace programs are perfect so retaining -EBUSY for unlink, rmdir and rename of dentries that have mounts in the current mount namespace plays an important role of maintaining consistency with historical behavior and making imperfect userspace applications hard to exploit. v2: Remove spurious old_dentry. v3: Optimized shrink_submounts_and_drop Removed unsued afs label v4: Simplified the changes to check_submounts_and_drop Do not rename check_submounts_and_drop shrink_submounts_and_drop Document what why we need atomicity in check_submounts_and_drop Rely on the parent inode mutex to make d_revalidate and d_invalidate an atomic unit. v5: Refcount the mountpoint to detach in case of simultaneous renames. Reviewed-by: Miklos Szeredi <miklos@szeredi.hu> Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-10-02 09:33:48 +08:00
struct detach_data *data = _data;
vfs: Lazily remove mounts on unlinked files and directories. With the introduction of mount namespaces and bind mounts it became possible to access files and directories that on some paths are mount points but are not mount points on other paths. It is very confusing when rm -rf somedir returns -EBUSY simply because somedir is mounted somewhere else. With the addition of user namespaces allowing unprivileged mounts this condition has gone from annoying to allowing a DOS attack on other users in the system. The possibility for mischief is removed by updating the vfs to support rename, unlink and rmdir on a dentry that is a mountpoint and by lazily unmounting mountpoints on deleted dentries. In particular this change allows rename, unlink and rmdir system calls on a dentry without a mountpoint in the current mount namespace to succeed, and it allows rename, unlink, and rmdir performed on a distributed filesystem to update the vfs cache even if when there is a mount in some namespace on the original dentry. There are two common patterns of maintaining mounts: Mounts on trusted paths with the parent directory of the mount point and all ancestory directories up to / owned by root and modifiable only by root (i.e. /media/xxx, /dev, /dev/pts, /proc, /sys, /sys/fs/cgroup/{cpu, cpuacct, ...}, /usr, /usr/local). Mounts on unprivileged directories maintained by fusermount. In the case of mounts in trusted directories owned by root and modifiable only by root the current parent directory permissions are sufficient to ensure a mount point on a trusted path is not removed or renamed by anyone other than root, even if there is a context where the there are no mount points to prevent this. In the case of mounts in directories owned by less privileged users races with users modifying the path of a mount point are already a danger. fusermount already uses a combination of chdir, /proc/<pid>/fd/NNN, and UMOUNT_NOFOLLOW to prevent these races. The removable of global rename, unlink, and rmdir protection really adds nothing new to consider only a widening of the attack window, and fusermount is already safe against unprivileged users modifying the directory simultaneously. In principle for perfect userspace programs returning -EBUSY for unlink, rmdir, and rename of dentires that have mounts in the local namespace is actually unnecessary. Unfortunately not all userspace programs are perfect so retaining -EBUSY for unlink, rmdir and rename of dentries that have mounts in the current mount namespace plays an important role of maintaining consistency with historical behavior and making imperfect userspace applications hard to exploit. v2: Remove spurious old_dentry. v3: Optimized shrink_submounts_and_drop Removed unsued afs label v4: Simplified the changes to check_submounts_and_drop Do not rename check_submounts_and_drop shrink_submounts_and_drop Document what why we need atomicity in check_submounts_and_drop Rely on the parent inode mutex to make d_revalidate and d_invalidate an atomic unit. v5: Refcount the mountpoint to detach in case of simultaneous renames. Reviewed-by: Miklos Szeredi <miklos@szeredi.hu> Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-10-02 09:33:48 +08:00
if (!data->mountpoint && !data->select.found)
__d_drop(data->select.start);
}
/**
* d_invalidate - detach submounts, prune dcache, and drop
* @dentry: dentry to invalidate (aka detach, prune and drop)
*
* no dcache lock.
*
vfs: Lazily remove mounts on unlinked files and directories. With the introduction of mount namespaces and bind mounts it became possible to access files and directories that on some paths are mount points but are not mount points on other paths. It is very confusing when rm -rf somedir returns -EBUSY simply because somedir is mounted somewhere else. With the addition of user namespaces allowing unprivileged mounts this condition has gone from annoying to allowing a DOS attack on other users in the system. The possibility for mischief is removed by updating the vfs to support rename, unlink and rmdir on a dentry that is a mountpoint and by lazily unmounting mountpoints on deleted dentries. In particular this change allows rename, unlink and rmdir system calls on a dentry without a mountpoint in the current mount namespace to succeed, and it allows rename, unlink, and rmdir performed on a distributed filesystem to update the vfs cache even if when there is a mount in some namespace on the original dentry. There are two common patterns of maintaining mounts: Mounts on trusted paths with the parent directory of the mount point and all ancestory directories up to / owned by root and modifiable only by root (i.e. /media/xxx, /dev, /dev/pts, /proc, /sys, /sys/fs/cgroup/{cpu, cpuacct, ...}, /usr, /usr/local). Mounts on unprivileged directories maintained by fusermount. In the case of mounts in trusted directories owned by root and modifiable only by root the current parent directory permissions are sufficient to ensure a mount point on a trusted path is not removed or renamed by anyone other than root, even if there is a context where the there are no mount points to prevent this. In the case of mounts in directories owned by less privileged users races with users modifying the path of a mount point are already a danger. fusermount already uses a combination of chdir, /proc/<pid>/fd/NNN, and UMOUNT_NOFOLLOW to prevent these races. The removable of global rename, unlink, and rmdir protection really adds nothing new to consider only a widening of the attack window, and fusermount is already safe against unprivileged users modifying the directory simultaneously. In principle for perfect userspace programs returning -EBUSY for unlink, rmdir, and rename of dentires that have mounts in the local namespace is actually unnecessary. Unfortunately not all userspace programs are perfect so retaining -EBUSY for unlink, rmdir and rename of dentries that have mounts in the current mount namespace plays an important role of maintaining consistency with historical behavior and making imperfect userspace applications hard to exploit. v2: Remove spurious old_dentry. v3: Optimized shrink_submounts_and_drop Removed unsued afs label v4: Simplified the changes to check_submounts_and_drop Do not rename check_submounts_and_drop shrink_submounts_and_drop Document what why we need atomicity in check_submounts_and_drop Rely on the parent inode mutex to make d_revalidate and d_invalidate an atomic unit. v5: Refcount the mountpoint to detach in case of simultaneous renames. Reviewed-by: Miklos Szeredi <miklos@szeredi.hu> Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-10-02 09:33:48 +08:00
* The final d_drop is done as an atomic operation relative to
* rename_lock ensuring there are no races with d_set_mounted. This
* ensures there are no unhashed dentries on the path to a mountpoint.
*/
void d_invalidate(struct dentry *dentry)
{
/*
* If it's already been dropped, return OK.
*/
spin_lock(&dentry->d_lock);
if (d_unhashed(dentry)) {
spin_unlock(&dentry->d_lock);
return;
}
spin_unlock(&dentry->d_lock);
/* Negative dentries can be dropped without further checks */
if (!dentry->d_inode) {
d_drop(dentry);
return;
}
for (;;) {
vfs: Lazily remove mounts on unlinked files and directories. With the introduction of mount namespaces and bind mounts it became possible to access files and directories that on some paths are mount points but are not mount points on other paths. It is very confusing when rm -rf somedir returns -EBUSY simply because somedir is mounted somewhere else. With the addition of user namespaces allowing unprivileged mounts this condition has gone from annoying to allowing a DOS attack on other users in the system. The possibility for mischief is removed by updating the vfs to support rename, unlink and rmdir on a dentry that is a mountpoint and by lazily unmounting mountpoints on deleted dentries. In particular this change allows rename, unlink and rmdir system calls on a dentry without a mountpoint in the current mount namespace to succeed, and it allows rename, unlink, and rmdir performed on a distributed filesystem to update the vfs cache even if when there is a mount in some namespace on the original dentry. There are two common patterns of maintaining mounts: Mounts on trusted paths with the parent directory of the mount point and all ancestory directories up to / owned by root and modifiable only by root (i.e. /media/xxx, /dev, /dev/pts, /proc, /sys, /sys/fs/cgroup/{cpu, cpuacct, ...}, /usr, /usr/local). Mounts on unprivileged directories maintained by fusermount. In the case of mounts in trusted directories owned by root and modifiable only by root the current parent directory permissions are sufficient to ensure a mount point on a trusted path is not removed or renamed by anyone other than root, even if there is a context where the there are no mount points to prevent this. In the case of mounts in directories owned by less privileged users races with users modifying the path of a mount point are already a danger. fusermount already uses a combination of chdir, /proc/<pid>/fd/NNN, and UMOUNT_NOFOLLOW to prevent these races. The removable of global rename, unlink, and rmdir protection really adds nothing new to consider only a widening of the attack window, and fusermount is already safe against unprivileged users modifying the directory simultaneously. In principle for perfect userspace programs returning -EBUSY for unlink, rmdir, and rename of dentires that have mounts in the local namespace is actually unnecessary. Unfortunately not all userspace programs are perfect so retaining -EBUSY for unlink, rmdir and rename of dentries that have mounts in the current mount namespace plays an important role of maintaining consistency with historical behavior and making imperfect userspace applications hard to exploit. v2: Remove spurious old_dentry. v3: Optimized shrink_submounts_and_drop Removed unsued afs label v4: Simplified the changes to check_submounts_and_drop Do not rename check_submounts_and_drop shrink_submounts_and_drop Document what why we need atomicity in check_submounts_and_drop Rely on the parent inode mutex to make d_revalidate and d_invalidate an atomic unit. v5: Refcount the mountpoint to detach in case of simultaneous renames. Reviewed-by: Miklos Szeredi <miklos@szeredi.hu> Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-10-02 09:33:48 +08:00
struct detach_data data;
vfs: Lazily remove mounts on unlinked files and directories. With the introduction of mount namespaces and bind mounts it became possible to access files and directories that on some paths are mount points but are not mount points on other paths. It is very confusing when rm -rf somedir returns -EBUSY simply because somedir is mounted somewhere else. With the addition of user namespaces allowing unprivileged mounts this condition has gone from annoying to allowing a DOS attack on other users in the system. The possibility for mischief is removed by updating the vfs to support rename, unlink and rmdir on a dentry that is a mountpoint and by lazily unmounting mountpoints on deleted dentries. In particular this change allows rename, unlink and rmdir system calls on a dentry without a mountpoint in the current mount namespace to succeed, and it allows rename, unlink, and rmdir performed on a distributed filesystem to update the vfs cache even if when there is a mount in some namespace on the original dentry. There are two common patterns of maintaining mounts: Mounts on trusted paths with the parent directory of the mount point and all ancestory directories up to / owned by root and modifiable only by root (i.e. /media/xxx, /dev, /dev/pts, /proc, /sys, /sys/fs/cgroup/{cpu, cpuacct, ...}, /usr, /usr/local). Mounts on unprivileged directories maintained by fusermount. In the case of mounts in trusted directories owned by root and modifiable only by root the current parent directory permissions are sufficient to ensure a mount point on a trusted path is not removed or renamed by anyone other than root, even if there is a context where the there are no mount points to prevent this. In the case of mounts in directories owned by less privileged users races with users modifying the path of a mount point are already a danger. fusermount already uses a combination of chdir, /proc/<pid>/fd/NNN, and UMOUNT_NOFOLLOW to prevent these races. The removable of global rename, unlink, and rmdir protection really adds nothing new to consider only a widening of the attack window, and fusermount is already safe against unprivileged users modifying the directory simultaneously. In principle for perfect userspace programs returning -EBUSY for unlink, rmdir, and rename of dentires that have mounts in the local namespace is actually unnecessary. Unfortunately not all userspace programs are perfect so retaining -EBUSY for unlink, rmdir and rename of dentries that have mounts in the current mount namespace plays an important role of maintaining consistency with historical behavior and making imperfect userspace applications hard to exploit. v2: Remove spurious old_dentry. v3: Optimized shrink_submounts_and_drop Removed unsued afs label v4: Simplified the changes to check_submounts_and_drop Do not rename check_submounts_and_drop shrink_submounts_and_drop Document what why we need atomicity in check_submounts_and_drop Rely on the parent inode mutex to make d_revalidate and d_invalidate an atomic unit. v5: Refcount the mountpoint to detach in case of simultaneous renames. Reviewed-by: Miklos Szeredi <miklos@szeredi.hu> Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-10-02 09:33:48 +08:00
data.mountpoint = NULL;
INIT_LIST_HEAD(&data.select.dispose);
data.select.start = dentry;
data.select.found = 0;
d_walk(dentry, &data, detach_and_collect, check_and_drop);
vfs: Lazily remove mounts on unlinked files and directories. With the introduction of mount namespaces and bind mounts it became possible to access files and directories that on some paths are mount points but are not mount points on other paths. It is very confusing when rm -rf somedir returns -EBUSY simply because somedir is mounted somewhere else. With the addition of user namespaces allowing unprivileged mounts this condition has gone from annoying to allowing a DOS attack on other users in the system. The possibility for mischief is removed by updating the vfs to support rename, unlink and rmdir on a dentry that is a mountpoint and by lazily unmounting mountpoints on deleted dentries. In particular this change allows rename, unlink and rmdir system calls on a dentry without a mountpoint in the current mount namespace to succeed, and it allows rename, unlink, and rmdir performed on a distributed filesystem to update the vfs cache even if when there is a mount in some namespace on the original dentry. There are two common patterns of maintaining mounts: Mounts on trusted paths with the parent directory of the mount point and all ancestory directories up to / owned by root and modifiable only by root (i.e. /media/xxx, /dev, /dev/pts, /proc, /sys, /sys/fs/cgroup/{cpu, cpuacct, ...}, /usr, /usr/local). Mounts on unprivileged directories maintained by fusermount. In the case of mounts in trusted directories owned by root and modifiable only by root the current parent directory permissions are sufficient to ensure a mount point on a trusted path is not removed or renamed by anyone other than root, even if there is a context where the there are no mount points to prevent this. In the case of mounts in directories owned by less privileged users races with users modifying the path of a mount point are already a danger. fusermount already uses a combination of chdir, /proc/<pid>/fd/NNN, and UMOUNT_NOFOLLOW to prevent these races. The removable of global rename, unlink, and rmdir protection really adds nothing new to consider only a widening of the attack window, and fusermount is already safe against unprivileged users modifying the directory simultaneously. In principle for perfect userspace programs returning -EBUSY for unlink, rmdir, and rename of dentires that have mounts in the local namespace is actually unnecessary. Unfortunately not all userspace programs are perfect so retaining -EBUSY for unlink, rmdir and rename of dentries that have mounts in the current mount namespace plays an important role of maintaining consistency with historical behavior and making imperfect userspace applications hard to exploit. v2: Remove spurious old_dentry. v3: Optimized shrink_submounts_and_drop Removed unsued afs label v4: Simplified the changes to check_submounts_and_drop Do not rename check_submounts_and_drop shrink_submounts_and_drop Document what why we need atomicity in check_submounts_and_drop Rely on the parent inode mutex to make d_revalidate and d_invalidate an atomic unit. v5: Refcount the mountpoint to detach in case of simultaneous renames. Reviewed-by: Miklos Szeredi <miklos@szeredi.hu> Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-10-02 09:33:48 +08:00
if (data.select.found)
shrink_dentry_list(&data.select.dispose);
vfs: Lazily remove mounts on unlinked files and directories. With the introduction of mount namespaces and bind mounts it became possible to access files and directories that on some paths are mount points but are not mount points on other paths. It is very confusing when rm -rf somedir returns -EBUSY simply because somedir is mounted somewhere else. With the addition of user namespaces allowing unprivileged mounts this condition has gone from annoying to allowing a DOS attack on other users in the system. The possibility for mischief is removed by updating the vfs to support rename, unlink and rmdir on a dentry that is a mountpoint and by lazily unmounting mountpoints on deleted dentries. In particular this change allows rename, unlink and rmdir system calls on a dentry without a mountpoint in the current mount namespace to succeed, and it allows rename, unlink, and rmdir performed on a distributed filesystem to update the vfs cache even if when there is a mount in some namespace on the original dentry. There are two common patterns of maintaining mounts: Mounts on trusted paths with the parent directory of the mount point and all ancestory directories up to / owned by root and modifiable only by root (i.e. /media/xxx, /dev, /dev/pts, /proc, /sys, /sys/fs/cgroup/{cpu, cpuacct, ...}, /usr, /usr/local). Mounts on unprivileged directories maintained by fusermount. In the case of mounts in trusted directories owned by root and modifiable only by root the current parent directory permissions are sufficient to ensure a mount point on a trusted path is not removed or renamed by anyone other than root, even if there is a context where the there are no mount points to prevent this. In the case of mounts in directories owned by less privileged users races with users modifying the path of a mount point are already a danger. fusermount already uses a combination of chdir, /proc/<pid>/fd/NNN, and UMOUNT_NOFOLLOW to prevent these races. The removable of global rename, unlink, and rmdir protection really adds nothing new to consider only a widening of the attack window, and fusermount is already safe against unprivileged users modifying the directory simultaneously. In principle for perfect userspace programs returning -EBUSY for unlink, rmdir, and rename of dentires that have mounts in the local namespace is actually unnecessary. Unfortunately not all userspace programs are perfect so retaining -EBUSY for unlink, rmdir and rename of dentries that have mounts in the current mount namespace plays an important role of maintaining consistency with historical behavior and making imperfect userspace applications hard to exploit. v2: Remove spurious old_dentry. v3: Optimized shrink_submounts_and_drop Removed unsued afs label v4: Simplified the changes to check_submounts_and_drop Do not rename check_submounts_and_drop shrink_submounts_and_drop Document what why we need atomicity in check_submounts_and_drop Rely on the parent inode mutex to make d_revalidate and d_invalidate an atomic unit. v5: Refcount the mountpoint to detach in case of simultaneous renames. Reviewed-by: Miklos Szeredi <miklos@szeredi.hu> Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-10-02 09:33:48 +08:00
if (data.mountpoint) {
detach_mounts(data.mountpoint);
dput(data.mountpoint);
}
vfs: Lazily remove mounts on unlinked files and directories. With the introduction of mount namespaces and bind mounts it became possible to access files and directories that on some paths are mount points but are not mount points on other paths. It is very confusing when rm -rf somedir returns -EBUSY simply because somedir is mounted somewhere else. With the addition of user namespaces allowing unprivileged mounts this condition has gone from annoying to allowing a DOS attack on other users in the system. The possibility for mischief is removed by updating the vfs to support rename, unlink and rmdir on a dentry that is a mountpoint and by lazily unmounting mountpoints on deleted dentries. In particular this change allows rename, unlink and rmdir system calls on a dentry without a mountpoint in the current mount namespace to succeed, and it allows rename, unlink, and rmdir performed on a distributed filesystem to update the vfs cache even if when there is a mount in some namespace on the original dentry. There are two common patterns of maintaining mounts: Mounts on trusted paths with the parent directory of the mount point and all ancestory directories up to / owned by root and modifiable only by root (i.e. /media/xxx, /dev, /dev/pts, /proc, /sys, /sys/fs/cgroup/{cpu, cpuacct, ...}, /usr, /usr/local). Mounts on unprivileged directories maintained by fusermount. In the case of mounts in trusted directories owned by root and modifiable only by root the current parent directory permissions are sufficient to ensure a mount point on a trusted path is not removed or renamed by anyone other than root, even if there is a context where the there are no mount points to prevent this. In the case of mounts in directories owned by less privileged users races with users modifying the path of a mount point are already a danger. fusermount already uses a combination of chdir, /proc/<pid>/fd/NNN, and UMOUNT_NOFOLLOW to prevent these races. The removable of global rename, unlink, and rmdir protection really adds nothing new to consider only a widening of the attack window, and fusermount is already safe against unprivileged users modifying the directory simultaneously. In principle for perfect userspace programs returning -EBUSY for unlink, rmdir, and rename of dentires that have mounts in the local namespace is actually unnecessary. Unfortunately not all userspace programs are perfect so retaining -EBUSY for unlink, rmdir and rename of dentries that have mounts in the current mount namespace plays an important role of maintaining consistency with historical behavior and making imperfect userspace applications hard to exploit. v2: Remove spurious old_dentry. v3: Optimized shrink_submounts_and_drop Removed unsued afs label v4: Simplified the changes to check_submounts_and_drop Do not rename check_submounts_and_drop shrink_submounts_and_drop Document what why we need atomicity in check_submounts_and_drop Rely on the parent inode mutex to make d_revalidate and d_invalidate an atomic unit. v5: Refcount the mountpoint to detach in case of simultaneous renames. Reviewed-by: Miklos Szeredi <miklos@szeredi.hu> Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-10-02 09:33:48 +08:00
if (!data.mountpoint && !data.select.found)
break;
cond_resched();
}
}
EXPORT_SYMBOL(d_invalidate);
/**
* __d_alloc - allocate a dcache entry
* @sb: filesystem it will belong to
* @name: qstr of the name
*
* Allocates a dentry. It returns %NULL if there is insufficient memory
* available. On a success the dentry is returned. The name passed in is
* copied and the copy passed in may be reused after this call.
*/
struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
{
struct dentry *dentry;
char *dname;
dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
if (!dentry)
return NULL;
2012-05-22 07:14:04 +08:00
/*
* We guarantee that the inline name is always NUL-terminated.
* This way the memcpy() done by the name switching in rename
* will still always have a NUL at the end, even if we might
* be overwriting an internal NUL character
*/
dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
if (unlikely(!name)) {
static const struct qstr anon = QSTR_INIT("/", 1);
name = &anon;
dname = dentry->d_iname;
} else if (name->len > DNAME_INLINE_LEN-1) {
Allow sharing external names after __d_move() * external dentry names get a small structure prepended to them (struct external_name). * it contains an atomic refcount, matching the number of struct dentry instances that have ->d_name.name pointing to that external name. The first thing free_dentry() does is decrementing refcount of external name, so the instances that are between the call of free_dentry() and RCU-delayed actual freeing do not contribute. * __d_move(x, y, false) makes the name of x equal to the name of y, external or not. If y has an external name, extra reference is grabbed and put into x->d_name.name. If x used to have an external name, the reference to the old name is dropped and, should it reach zero, freeing is scheduled via kfree_rcu(). * free_dentry() in dentry with external name decrements the refcount of that name and, should it reach zero, does RCU-delayed call that will free both the dentry and external name. Otherwise it does what it used to do, except that __d_free() doesn't even look at ->d_name.name; it simply frees the dentry. All non-RCU accesses to dentry external name are safe wrt freeing since they all should happen before free_dentry() is called. RCU accesses might run into a dentry seen by free_dentry() or into an old name that got already dropped by __d_move(); however, in both cases dentry must have been alive and refer to that name at some point after we'd done rcu_read_lock(), which means that any freeing must be still pending. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2014-09-30 02:54:27 +08:00
size_t size = offsetof(struct external_name, name[1]);
2016-01-15 07:18:21 +08:00
struct external_name *p = kmalloc(size + name->len,
GFP_KERNEL_ACCOUNT);
Allow sharing external names after __d_move() * external dentry names get a small structure prepended to them (struct external_name). * it contains an atomic refcount, matching the number of struct dentry instances that have ->d_name.name pointing to that external name. The first thing free_dentry() does is decrementing refcount of external name, so the instances that are between the call of free_dentry() and RCU-delayed actual freeing do not contribute. * __d_move(x, y, false) makes the name of x equal to the name of y, external or not. If y has an external name, extra reference is grabbed and put into x->d_name.name. If x used to have an external name, the reference to the old name is dropped and, should it reach zero, freeing is scheduled via kfree_rcu(). * free_dentry() in dentry with external name decrements the refcount of that name and, should it reach zero, does RCU-delayed call that will free both the dentry and external name. Otherwise it does what it used to do, except that __d_free() doesn't even look at ->d_name.name; it simply frees the dentry. All non-RCU accesses to dentry external name are safe wrt freeing since they all should happen before free_dentry() is called. RCU accesses might run into a dentry seen by free_dentry() or into an old name that got already dropped by __d_move(); however, in both cases dentry must have been alive and refer to that name at some point after we'd done rcu_read_lock(), which means that any freeing must be still pending. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2014-09-30 02:54:27 +08:00
if (!p) {
kmem_cache_free(dentry_cache, dentry);
return NULL;
}
Allow sharing external names after __d_move() * external dentry names get a small structure prepended to them (struct external_name). * it contains an atomic refcount, matching the number of struct dentry instances that have ->d_name.name pointing to that external name. The first thing free_dentry() does is decrementing refcount of external name, so the instances that are between the call of free_dentry() and RCU-delayed actual freeing do not contribute. * __d_move(x, y, false) makes the name of x equal to the name of y, external or not. If y has an external name, extra reference is grabbed and put into x->d_name.name. If x used to have an external name, the reference to the old name is dropped and, should it reach zero, freeing is scheduled via kfree_rcu(). * free_dentry() in dentry with external name decrements the refcount of that name and, should it reach zero, does RCU-delayed call that will free both the dentry and external name. Otherwise it does what it used to do, except that __d_free() doesn't even look at ->d_name.name; it simply frees the dentry. All non-RCU accesses to dentry external name are safe wrt freeing since they all should happen before free_dentry() is called. RCU accesses might run into a dentry seen by free_dentry() or into an old name that got already dropped by __d_move(); however, in both cases dentry must have been alive and refer to that name at some point after we'd done rcu_read_lock(), which means that any freeing must be still pending. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2014-09-30 02:54:27 +08:00
atomic_set(&p->u.count, 1);
dname = p->name;
if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
kasan_unpoison_shadow(dname,
round_up(name->len + 1, sizeof(unsigned long)));
} else {
dname = dentry->d_iname;
}
dentry->d_name.len = name->len;
dentry->d_name.hash = name->hash;
memcpy(dname, name->name, name->len);
dname[name->len] = 0;
2012-05-22 07:14:04 +08:00
/* Make sure we always see the terminating NUL character */
smp_wmb();
dentry->d_name.name = dname;
dentry->d_lockref.count = 1;
vfs: get rid of insane dentry hashing rules The dentry hashing rules have been really quite complicated for a long while, in odd ways. That made functions like __d_drop() very fragile and non-obvious. In particular, whether a dentry was hashed or not was indicated with an explicit DCACHE_UNHASHED bit. That's despite the fact that the hash abstraction that the dentries use actually have a 'is this entry hashed or not' model (which is a simple test of the 'pprev' pointer). The reason that was done is because we used the normal 'is this entry unhashed' model to mark whether the dentry had _ever_ been hashed in the dentry hash tables, and that logic goes back many years (commit b3423415fbc2: "dcache: avoid RCU for never-hashed dentries"). That, in turn, meant that __d_drop had totally different unhashing logic for the dentry hash table case and for the anonymous dcache case, because in order to use the "is this dentry hashed" logic as a flag for whether it had ever been on the RCU hash table, we had to unhash such a dentry differently so that we'd never think that it wasn't 'unhashed' and wouldn't be free'd correctly. That's just insane. It made the logic really hard to follow, when there were two different kinds of "unhashed" states, and one of them (the one that used "list_bl_unhashed()") really had nothing at all to do with being unhashed per se, but with a very subtle lifetime rule instead. So turn all of it around, and make it logical. Instead of having a DENTRY_UNHASHED bit in d_flags to indicate whether the dentry is on the hash chains or not, use the hash chain unhashed logic for that. Suddenly "d_unhashed()" just uses "list_bl_unhashed()", and everything makes sense. And for the lifetime rule, just use an explicit DENTRY_RCUACCEES bit. If we ever insert the dentry into the dentry hash table so that it is visible to RCU lookup, we mark it DENTRY_RCUACCESS to show that it now needs the RCU lifetime rules. Now suddently that test at dentry free time makes sense too. And because unhashing now is sane and doesn't depend on where the dentry got unhashed from (because the dentry hash chain details doesn't have some subtle side effects), we can re-unify the __d_drop() logic and use common code for the unhashing. Also fix one more open-coded hash chain bit_spin_lock() that I missed in the previous chain locking cleanup commit. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-04-24 22:58:46 +08:00
dentry->d_flags = 0;
spin_lock_init(&dentry->d_lock);
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
seqcount_init(&dentry->d_seq);
dentry->d_inode = NULL;
dentry->d_parent = dentry;
dentry->d_sb = sb;
dentry->d_op = NULL;
dentry->d_fsdata = NULL;
INIT_HLIST_BL_NODE(&dentry->d_hash);
INIT_LIST_HEAD(&dentry->d_lru);
INIT_LIST_HEAD(&dentry->d_subdirs);
INIT_HLIST_NODE(&dentry->d_u.d_alias);
INIT_LIST_HEAD(&dentry->d_child);
d_set_d_op(dentry, dentry->d_sb->s_d_op);
fs: use fast counters for vfs caches percpu_counter library generates quite nasty code, so unless you need to dynamically allocate counters or take fast approximate value, a simple per cpu set of counters is much better. The percpu_counter can never be made to work as well, because it has an indirection from pointer to percpu memory, and it can't use direct this_cpu_inc interfaces because it doesn't use static PER_CPU data, so code will always be worse. In the fastpath, it is the difference between this: incl %gs:nr_dentry # nr_dentry and this: movl percpu_counter_batch(%rip), %edx # percpu_counter_batch, movl $1, %esi #, movq $nr_dentry, %rdi #, call __percpu_counter_add # (plus I clobber registers) __percpu_counter_add: pushq %rbp # movq %rsp, %rbp #, subq $32, %rsp #, movq %rbx, -24(%rbp) #, movq %r12, -16(%rbp) #, movq %r13, -8(%rbp) #, movq %rdi, %rbx # fbc, fbc #APP # 216 "/home/npiggin/usr/src/linux-2.6/arch/x86/include/asm/thread_info.h" 1 movq %gs:kernel_stack,%rax #, pfo_ret__ # 0 "" 2 #NO_APP incl -8124(%rax) # <variable>.preempt_count movq 32(%rdi), %r12 # <variable>.counters, tcp_ptr__ #APP # 78 "lib/percpu_counter.c" 1 add %gs:this_cpu_off, %r12 # this_cpu_off, tcp_ptr__ # 0 "" 2 #NO_APP movslq (%r12),%r13 #* tcp_ptr__, tmp73 movslq %edx,%rax # batch, batch addq %rsi, %r13 # amount, count cmpq %rax, %r13 # batch, count jge .L27 #, negl %edx # tmp76 movslq %edx,%rdx # tmp76, tmp77 cmpq %rdx, %r13 # tmp77, count jg .L28 #, .L27: movq %rbx, %rdi # fbc, call _raw_spin_lock # addq %r13, 8(%rbx) # count, <variable>.count movq %rbx, %rdi # fbc, movl $0, (%r12) #,* tcp_ptr__ call _raw_spin_unlock # .L29: #APP # 216 "/home/npiggin/usr/src/linux-2.6/arch/x86/include/asm/thread_info.h" 1 movq %gs:kernel_stack,%rax #, pfo_ret__ # 0 "" 2 #NO_APP decl -8124(%rax) # <variable>.preempt_count movq -8136(%rax), %rax #, D.14625 testb $8, %al #, D.14625 jne .L32 #, .L31: movq -24(%rbp), %rbx #, movq -16(%rbp), %r12 #, movq -8(%rbp), %r13 #, leave ret .p2align 4,,10 .p2align 3 .L28: movl %r13d, (%r12) # count,* jmp .L29 # .L32: call preempt_schedule # .p2align 4,,6 jmp .L31 # .size __percpu_counter_add, .-__percpu_counter_add .p2align 4,,15 Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:19 +08:00
this_cpu_inc(nr_dentry);
return dentry;
}
/**
* d_alloc - allocate a dcache entry
* @parent: parent of entry to allocate
* @name: qstr of the name
*
* Allocates a dentry. It returns %NULL if there is insufficient memory
* available. On a success the dentry is returned. The name passed in is
* copied and the copy passed in may be reused after this call.
*/
struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
{
struct dentry *dentry = __d_alloc(parent->d_sb, name);
if (!dentry)
return NULL;
spin_lock(&parent->d_lock);
/*
* don't need child lock because it is not subject
* to concurrency here
*/
__dget_dlock(parent);
dentry->d_parent = parent;
list_add(&dentry->d_child, &parent->d_subdirs);
spin_unlock(&parent->d_lock);
return dentry;
}
EXPORT_SYMBOL(d_alloc);
/**
* d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
* @sb: the superblock
* @name: qstr of the name
*
* For a filesystem that just pins its dentries in memory and never
* performs lookups at all, return an unhashed IS_ROOT dentry.
*/
struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
{
return __d_alloc(sb, name);
}
EXPORT_SYMBOL(d_alloc_pseudo);
struct dentry *d_alloc_name(struct dentry *parent, const char *name)
{
struct qstr q;
q.name = name;
q.hash_len = hashlen_string(name);
return d_alloc(parent, &q);
}
EXPORT_SYMBOL(d_alloc_name);
void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
{
WARN_ON_ONCE(dentry->d_op);
WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
DCACHE_OP_COMPARE |
DCACHE_OP_REVALIDATE |
vfs: kill FS_REVAL_DOT by adding a d_weak_revalidate dentry op The following set of operations on a NFS client and server will cause server# mkdir a client# cd a server# mv a a.bak client# sleep 30 # (or whatever the dir attrcache timeout is) client# stat . stat: cannot stat `.': Stale NFS file handle Obviously, we should not be getting an ESTALE error back there since the inode still exists on the server. The problem is that the lookup code will call d_revalidate on the dentry that "." refers to, because NFS has FS_REVAL_DOT set. nfs_lookup_revalidate will see that the parent directory has changed and will try to reverify the dentry by redoing a LOOKUP. That of course fails, so the lookup code returns ESTALE. The problem here is that d_revalidate is really a bad fit for this case. What we really want to know at this point is whether the inode is still good or not, but we don't really care what name it goes by or whether the dcache is still valid. Add a new d_op->d_weak_revalidate operation and have complete_walk call that instead of d_revalidate. The intent there is to allow for a "weaker" d_revalidate that just checks to see whether the inode is still good. This is also gives us an opportunity to kill off the FS_REVAL_DOT special casing. [AV: changed method name, added note in porting, fixed confusion re having it possibly called from RCU mode (it won't be)] Cc: NeilBrown <neilb@suse.de> Signed-off-by: Jeff Layton <jlayton@redhat.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-02-21 00:19:05 +08:00
DCACHE_OP_WEAK_REVALIDATE |
overlayfs: Make f_path always point to the overlay and f_inode to the underlay Make file->f_path always point to the overlay dentry so that the path in /proc/pid/fd is correct and to ensure that label-based LSMs have access to the overlay as well as the underlay (path-based LSMs probably don't need it). Using my union testsuite to set things up, before the patch I see: [root@andromeda union-testsuite]# bash 5</mnt/a/foo107 [root@andromeda union-testsuite]# ls -l /proc/$$/fd/ ... lr-x------. 1 root root 64 Jun 5 14:38 5 -> /a/foo107 [root@andromeda union-testsuite]# stat /mnt/a/foo107 ... Device: 23h/35d Inode: 13381 Links: 1 ... [root@andromeda union-testsuite]# stat -L /proc/$$/fd/5 ... Device: 23h/35d Inode: 13381 Links: 1 ... After the patch: [root@andromeda union-testsuite]# bash 5</mnt/a/foo107 [root@andromeda union-testsuite]# ls -l /proc/$$/fd/ ... lr-x------. 1 root root 64 Jun 5 14:22 5 -> /mnt/a/foo107 [root@andromeda union-testsuite]# stat /mnt/a/foo107 ... Device: 23h/35d Inode: 40346 Links: 1 ... [root@andromeda union-testsuite]# stat -L /proc/$$/fd/5 ... Device: 23h/35d Inode: 40346 Links: 1 ... Note the change in where /proc/$$/fd/5 points to in the ls command. It was pointing to /a/foo107 (which doesn't exist) and now points to /mnt/a/foo107 (which is correct). The inode accessed, however, is the lower layer. The union layer is on device 25h/37d and the upper layer on 24h/36d. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2015-06-18 21:32:31 +08:00
DCACHE_OP_DELETE |
DCACHE_OP_SELECT_INODE |
DCACHE_OP_REAL));
dentry->d_op = op;
if (!op)
return;
if (op->d_hash)
dentry->d_flags |= DCACHE_OP_HASH;
if (op->d_compare)
dentry->d_flags |= DCACHE_OP_COMPARE;
if (op->d_revalidate)
dentry->d_flags |= DCACHE_OP_REVALIDATE;
vfs: kill FS_REVAL_DOT by adding a d_weak_revalidate dentry op The following set of operations on a NFS client and server will cause server# mkdir a client# cd a server# mv a a.bak client# sleep 30 # (or whatever the dir attrcache timeout is) client# stat . stat: cannot stat `.': Stale NFS file handle Obviously, we should not be getting an ESTALE error back there since the inode still exists on the server. The problem is that the lookup code will call d_revalidate on the dentry that "." refers to, because NFS has FS_REVAL_DOT set. nfs_lookup_revalidate will see that the parent directory has changed and will try to reverify the dentry by redoing a LOOKUP. That of course fails, so the lookup code returns ESTALE. The problem here is that d_revalidate is really a bad fit for this case. What we really want to know at this point is whether the inode is still good or not, but we don't really care what name it goes by or whether the dcache is still valid. Add a new d_op->d_weak_revalidate operation and have complete_walk call that instead of d_revalidate. The intent there is to allow for a "weaker" d_revalidate that just checks to see whether the inode is still good. This is also gives us an opportunity to kill off the FS_REVAL_DOT special casing. [AV: changed method name, added note in porting, fixed confusion re having it possibly called from RCU mode (it won't be)] Cc: NeilBrown <neilb@suse.de> Signed-off-by: Jeff Layton <jlayton@redhat.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-02-21 00:19:05 +08:00
if (op->d_weak_revalidate)
dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
if (op->d_delete)
dentry->d_flags |= DCACHE_OP_DELETE;
if (op->d_prune)
dentry->d_flags |= DCACHE_OP_PRUNE;
overlayfs: Make f_path always point to the overlay and f_inode to the underlay Make file->f_path always point to the overlay dentry so that the path in /proc/pid/fd is correct and to ensure that label-based LSMs have access to the overlay as well as the underlay (path-based LSMs probably don't need it). Using my union testsuite to set things up, before the patch I see: [root@andromeda union-testsuite]# bash 5</mnt/a/foo107 [root@andromeda union-testsuite]# ls -l /proc/$$/fd/ ... lr-x------. 1 root root 64 Jun 5 14:38 5 -> /a/foo107 [root@andromeda union-testsuite]# stat /mnt/a/foo107 ... Device: 23h/35d Inode: 13381 Links: 1 ... [root@andromeda union-testsuite]# stat -L /proc/$$/fd/5 ... Device: 23h/35d Inode: 13381 Links: 1 ... After the patch: [root@andromeda union-testsuite]# bash 5</mnt/a/foo107 [root@andromeda union-testsuite]# ls -l /proc/$$/fd/ ... lr-x------. 1 root root 64 Jun 5 14:22 5 -> /mnt/a/foo107 [root@andromeda union-testsuite]# stat /mnt/a/foo107 ... Device: 23h/35d Inode: 40346 Links: 1 ... [root@andromeda union-testsuite]# stat -L /proc/$$/fd/5 ... Device: 23h/35d Inode: 40346 Links: 1 ... Note the change in where /proc/$$/fd/5 points to in the ls command. It was pointing to /a/foo107 (which doesn't exist) and now points to /mnt/a/foo107 (which is correct). The inode accessed, however, is the lower layer. The union layer is on device 25h/37d and the upper layer on 24h/36d. Signed-off-by: David Howells <dhowells@redhat.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2015-06-18 21:32:31 +08:00
if (op->d_select_inode)
dentry->d_flags |= DCACHE_OP_SELECT_INODE;
if (op->d_real)
dentry->d_flags |= DCACHE_OP_REAL;
}
EXPORT_SYMBOL(d_set_d_op);
/*
* d_set_fallthru - Mark a dentry as falling through to a lower layer
* @dentry - The dentry to mark
*
* Mark a dentry as falling through to the lower layer (as set with
* d_pin_lower()). This flag may be recorded on the medium.
*/
void d_set_fallthru(struct dentry *dentry)
{
spin_lock(&dentry->d_lock);
dentry->d_flags |= DCACHE_FALLTHRU;
spin_unlock(&dentry->d_lock);
}
EXPORT_SYMBOL(d_set_fallthru);
static unsigned d_flags_for_inode(struct inode *inode)
{
unsigned add_flags = DCACHE_REGULAR_TYPE;
if (!inode)
return DCACHE_MISS_TYPE;
if (S_ISDIR(inode->i_mode)) {
add_flags = DCACHE_DIRECTORY_TYPE;
if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
if (unlikely(!inode->i_op->lookup))
add_flags = DCACHE_AUTODIR_TYPE;
else
inode->i_opflags |= IOP_LOOKUP;
}
goto type_determined;
}
if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
if (unlikely(inode->i_op->get_link)) {
add_flags = DCACHE_SYMLINK_TYPE;
goto type_determined;
}
inode->i_opflags |= IOP_NOFOLLOW;
}
if (unlikely(!S_ISREG(inode->i_mode)))
add_flags = DCACHE_SPECIAL_TYPE;
type_determined:
if (unlikely(IS_AUTOMOUNT(inode)))
add_flags |= DCACHE_NEED_AUTOMOUNT;
return add_flags;
}
static void __d_instantiate(struct dentry *dentry, struct inode *inode)
{
unsigned add_flags = d_flags_for_inode(inode);
beginning of transition to parallel lookups - marking in-lookup dentries marked as such when (would be) parallel lookup is about to pass them to actual ->lookup(); unmarked when * __d_add() is about to make it hashed, positive or not. * __d_move() (from d_splice_alias(), directly or via __d_unalias()) puts a preexisting dentry in its place * in caller of ->lookup() if it has escaped all of the above. Bug (WARN_ON, actually) if it reaches the final dput() or d_instantiate() while still marked such. As the result, we are guaranteed that for as long as the flag is set, dentry will * remain negative unhashed with positive refcount * never have its ->d_alias looked at * never have its ->d_lru looked at * never have its ->d_parent and ->d_name changed Right now we have at most one such for any given parent directory. With parallel lookups that restriction will weaken to * only exist when parent is locked shared * at most one with given (parent,name) pair (comparison of names is according to ->d_compare()) * only exist when there's no hashed dentry with the same (parent,name) Transition will take the next several commits; unfortunately, we'll only be able to switch to rwsem at the end of this series. The reason for not making it a single patch is to simplify review. New primitives: d_in_lookup() (a predicate checking if dentry is in the in-lookup state) and d_lookup_done() (tells the system that we are done with lookup and if it's still marked as in-lookup, it should cease to be such). Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 07:52:13 +08:00
WARN_ON(d_in_lookup(dentry));
spin_lock(&dentry->d_lock);
hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
raw_write_seqcount_begin(&dentry->d_seq);
__d_set_inode_and_type(dentry, inode, add_flags);
raw_write_seqcount_end(&dentry->d_seq);
__fsnotify_d_instantiate(dentry);
spin_unlock(&dentry->d_lock);
}
/**
* d_instantiate - fill in inode information for a dentry
* @entry: dentry to complete
* @inode: inode to attach to this dentry
*
* Fill in inode information in the entry.
*
* This turns negative dentries into productive full members
* of society.
*
* NOTE! This assumes that the inode count has been incremented
* (or otherwise set) by the caller to indicate that it is now
* in use by the dcache.
*/
void d_instantiate(struct dentry *entry, struct inode * inode)
{
BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
if (inode) {
security_d_instantiate(entry, inode);
spin_lock(&inode->i_lock);
__d_instantiate(entry, inode);
spin_unlock(&inode->i_lock);
}
}
EXPORT_SYMBOL(d_instantiate);
/**
* d_instantiate_no_diralias - instantiate a non-aliased dentry
* @entry: dentry to complete
* @inode: inode to attach to this dentry
*
* Fill in inode information in the entry. If a directory alias is found, then
* return an error (and drop inode). Together with d_materialise_unique() this
* guarantees that a directory inode may never have more than one alias.
*/
int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
{
BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
security_d_instantiate(entry, inode);
spin_lock(&inode->i_lock);
if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
spin_unlock(&inode->i_lock);
iput(inode);
return -EBUSY;
}
__d_instantiate(entry, inode);
spin_unlock(&inode->i_lock);
return 0;
}
EXPORT_SYMBOL(d_instantiate_no_diralias);
struct dentry *d_make_root(struct inode *root_inode)
{
struct dentry *res = NULL;
if (root_inode) {
res = __d_alloc(root_inode->i_sb, NULL);
if (res)
d_instantiate(res, root_inode);
else
iput(root_inode);
}
return res;
}
EXPORT_SYMBOL(d_make_root);
fs/dcache: allow d_obtain_alias() to return unhashed dentries Without this patch, inodes are not promptly freed on last close of an unlinked file by an nfs client: client$ mount -tnfs4 server:/export/ /mnt/ client$ tail -f /mnt/FOO ... server$ df -i /export server$ rm /export/FOO (^C the tail -f) server$ df -i /export server$ echo 2 >/proc/sys/vm/drop_caches server$ df -i /export the df's will show that the inode is not freed on the filesystem until the last step, when it could have been freed after killing the client's tail -f. On-disk data won't be deallocated either, leading to possible spurious ENOSPC. This occurs because when the client does the close, it arrives in a compound with a putfh and a close, processed like: - putfh: look up the filehandle.  The only alias found for the inode will be DCACHE_UNHASHED alias referenced by the filp this, so it creates a new DCACHE_DISCONECTED dentry and returns that instead. - close: closes the existing filp, which is destroyed immediately by dput() since it's DCACHE_UNHASHED. - end of the compound: release the reference to the current filehandle, and dput() the new DCACHE_DISCONECTED dentry, which gets put on the unused list instead of being destroyed immediately. Nick Piggin suggested fixing this by allowing d_obtain_alias to return the unhashed dentry that is referenced by the filp, instead of making it create a new dentry. Leave __d_find_alias() alone to avoid changing behavior of other callers. Also nfsd doesn't need all the checks of __d_find_alias(); any dentry, hashed or unhashed, disconnected or not, should work. Signed-off-by: J. Bruce Fields <bfields@redhat.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-19 04:45:09 +08:00
static struct dentry * __d_find_any_alias(struct inode *inode)
{
struct dentry *alias;
if (hlist_empty(&inode->i_dentry))
fs/dcache: allow d_obtain_alias() to return unhashed dentries Without this patch, inodes are not promptly freed on last close of an unlinked file by an nfs client: client$ mount -tnfs4 server:/export/ /mnt/ client$ tail -f /mnt/FOO ... server$ df -i /export server$ rm /export/FOO (^C the tail -f) server$ df -i /export server$ echo 2 >/proc/sys/vm/drop_caches server$ df -i /export the df's will show that the inode is not freed on the filesystem until the last step, when it could have been freed after killing the client's tail -f. On-disk data won't be deallocated either, leading to possible spurious ENOSPC. This occurs because when the client does the close, it arrives in a compound with a putfh and a close, processed like: - putfh: look up the filehandle.  The only alias found for the inode will be DCACHE_UNHASHED alias referenced by the filp this, so it creates a new DCACHE_DISCONECTED dentry and returns that instead. - close: closes the existing filp, which is destroyed immediately by dput() since it's DCACHE_UNHASHED. - end of the compound: release the reference to the current filehandle, and dput() the new DCACHE_DISCONECTED dentry, which gets put on the unused list instead of being destroyed immediately. Nick Piggin suggested fixing this by allowing d_obtain_alias to return the unhashed dentry that is referenced by the filp, instead of making it create a new dentry. Leave __d_find_alias() alone to avoid changing behavior of other callers. Also nfsd doesn't need all the checks of __d_find_alias(); any dentry, hashed or unhashed, disconnected or not, should work. Signed-off-by: J. Bruce Fields <bfields@redhat.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-19 04:45:09 +08:00
return NULL;
alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
fs/dcache: allow d_obtain_alias() to return unhashed dentries Without this patch, inodes are not promptly freed on last close of an unlinked file by an nfs client: client$ mount -tnfs4 server:/export/ /mnt/ client$ tail -f /mnt/FOO ... server$ df -i /export server$ rm /export/FOO (^C the tail -f) server$ df -i /export server$ echo 2 >/proc/sys/vm/drop_caches server$ df -i /export the df's will show that the inode is not freed on the filesystem until the last step, when it could have been freed after killing the client's tail -f. On-disk data won't be deallocated either, leading to possible spurious ENOSPC. This occurs because when the client does the close, it arrives in a compound with a putfh and a close, processed like: - putfh: look up the filehandle.  The only alias found for the inode will be DCACHE_UNHASHED alias referenced by the filp this, so it creates a new DCACHE_DISCONECTED dentry and returns that instead. - close: closes the existing filp, which is destroyed immediately by dput() since it's DCACHE_UNHASHED. - end of the compound: release the reference to the current filehandle, and dput() the new DCACHE_DISCONECTED dentry, which gets put on the unused list instead of being destroyed immediately. Nick Piggin suggested fixing this by allowing d_obtain_alias to return the unhashed dentry that is referenced by the filp, instead of making it create a new dentry. Leave __d_find_alias() alone to avoid changing behavior of other callers. Also nfsd doesn't need all the checks of __d_find_alias(); any dentry, hashed or unhashed, disconnected or not, should work. Signed-off-by: J. Bruce Fields <bfields@redhat.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-19 04:45:09 +08:00
__dget(alias);
return alias;
}
/**
* d_find_any_alias - find any alias for a given inode
* @inode: inode to find an alias for
*
* If any aliases exist for the given inode, take and return a
* reference for one of them. If no aliases exist, return %NULL.
*/
struct dentry *d_find_any_alias(struct inode *inode)
fs/dcache: allow d_obtain_alias() to return unhashed dentries Without this patch, inodes are not promptly freed on last close of an unlinked file by an nfs client: client$ mount -tnfs4 server:/export/ /mnt/ client$ tail -f /mnt/FOO ... server$ df -i /export server$ rm /export/FOO (^C the tail -f) server$ df -i /export server$ echo 2 >/proc/sys/vm/drop_caches server$ df -i /export the df's will show that the inode is not freed on the filesystem until the last step, when it could have been freed after killing the client's tail -f. On-disk data won't be deallocated either, leading to possible spurious ENOSPC. This occurs because when the client does the close, it arrives in a compound with a putfh and a close, processed like: - putfh: look up the filehandle.  The only alias found for the inode will be DCACHE_UNHASHED alias referenced by the filp this, so it creates a new DCACHE_DISCONECTED dentry and returns that instead. - close: closes the existing filp, which is destroyed immediately by dput() since it's DCACHE_UNHASHED. - end of the compound: release the reference to the current filehandle, and dput() the new DCACHE_DISCONECTED dentry, which gets put on the unused list instead of being destroyed immediately. Nick Piggin suggested fixing this by allowing d_obtain_alias to return the unhashed dentry that is referenced by the filp, instead of making it create a new dentry. Leave __d_find_alias() alone to avoid changing behavior of other callers. Also nfsd doesn't need all the checks of __d_find_alias(); any dentry, hashed or unhashed, disconnected or not, should work. Signed-off-by: J. Bruce Fields <bfields@redhat.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-19 04:45:09 +08:00
{
struct dentry *de;
spin_lock(&inode->i_lock);
de = __d_find_any_alias(inode);
spin_unlock(&inode->i_lock);
return de;
}
EXPORT_SYMBOL(d_find_any_alias);
fs/dcache: allow d_obtain_alias() to return unhashed dentries Without this patch, inodes are not promptly freed on last close of an unlinked file by an nfs client: client$ mount -tnfs4 server:/export/ /mnt/ client$ tail -f /mnt/FOO ... server$ df -i /export server$ rm /export/FOO (^C the tail -f) server$ df -i /export server$ echo 2 >/proc/sys/vm/drop_caches server$ df -i /export the df's will show that the inode is not freed on the filesystem until the last step, when it could have been freed after killing the client's tail -f. On-disk data won't be deallocated either, leading to possible spurious ENOSPC. This occurs because when the client does the close, it arrives in a compound with a putfh and a close, processed like: - putfh: look up the filehandle.  The only alias found for the inode will be DCACHE_UNHASHED alias referenced by the filp this, so it creates a new DCACHE_DISCONECTED dentry and returns that instead. - close: closes the existing filp, which is destroyed immediately by dput() since it's DCACHE_UNHASHED. - end of the compound: release the reference to the current filehandle, and dput() the new DCACHE_DISCONECTED dentry, which gets put on the unused list instead of being destroyed immediately. Nick Piggin suggested fixing this by allowing d_obtain_alias to return the unhashed dentry that is referenced by the filp, instead of making it create a new dentry. Leave __d_find_alias() alone to avoid changing behavior of other callers. Also nfsd doesn't need all the checks of __d_find_alias(); any dentry, hashed or unhashed, disconnected or not, should work. Signed-off-by: J. Bruce Fields <bfields@redhat.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-19 04:45:09 +08:00
static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
{
struct dentry *tmp;
struct dentry *res;
unsigned add_flags;
if (!inode)
return ERR_PTR(-ESTALE);
if (IS_ERR(inode))
return ERR_CAST(inode);
fs/dcache: allow d_obtain_alias() to return unhashed dentries Without this patch, inodes are not promptly freed on last close of an unlinked file by an nfs client: client$ mount -tnfs4 server:/export/ /mnt/ client$ tail -f /mnt/FOO ... server$ df -i /export server$ rm /export/FOO (^C the tail -f) server$ df -i /export server$ echo 2 >/proc/sys/vm/drop_caches server$ df -i /export the df's will show that the inode is not freed on the filesystem until the last step, when it could have been freed after killing the client's tail -f. On-disk data won't be deallocated either, leading to possible spurious ENOSPC. This occurs because when the client does the close, it arrives in a compound with a putfh and a close, processed like: - putfh: look up the filehandle.  The only alias found for the inode will be DCACHE_UNHASHED alias referenced by the filp this, so it creates a new DCACHE_DISCONECTED dentry and returns that instead. - close: closes the existing filp, which is destroyed immediately by dput() since it's DCACHE_UNHASHED. - end of the compound: release the reference to the current filehandle, and dput() the new DCACHE_DISCONECTED dentry, which gets put on the unused list instead of being destroyed immediately. Nick Piggin suggested fixing this by allowing d_obtain_alias to return the unhashed dentry that is referenced by the filp, instead of making it create a new dentry. Leave __d_find_alias() alone to avoid changing behavior of other callers. Also nfsd doesn't need all the checks of __d_find_alias(); any dentry, hashed or unhashed, disconnected or not, should work. Signed-off-by: J. Bruce Fields <bfields@redhat.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-19 04:45:09 +08:00
res = d_find_any_alias(inode);
if (res)
goto out_iput;
tmp = __d_alloc(inode->i_sb, NULL);
if (!tmp) {
res = ERR_PTR(-ENOMEM);
goto out_iput;
}
security_d_instantiate(tmp, inode);
spin_lock(&inode->i_lock);
fs/dcache: allow d_obtain_alias() to return unhashed dentries Without this patch, inodes are not promptly freed on last close of an unlinked file by an nfs client: client$ mount -tnfs4 server:/export/ /mnt/ client$ tail -f /mnt/FOO ... server$ df -i /export server$ rm /export/FOO (^C the tail -f) server$ df -i /export server$ echo 2 >/proc/sys/vm/drop_caches server$ df -i /export the df's will show that the inode is not freed on the filesystem until the last step, when it could have been freed after killing the client's tail -f. On-disk data won't be deallocated either, leading to possible spurious ENOSPC. This occurs because when the client does the close, it arrives in a compound with a putfh and a close, processed like: - putfh: look up the filehandle.  The only alias found for the inode will be DCACHE_UNHASHED alias referenced by the filp this, so it creates a new DCACHE_DISCONECTED dentry and returns that instead. - close: closes the existing filp, which is destroyed immediately by dput() since it's DCACHE_UNHASHED. - end of the compound: release the reference to the current filehandle, and dput() the new DCACHE_DISCONECTED dentry, which gets put on the unused list instead of being destroyed immediately. Nick Piggin suggested fixing this by allowing d_obtain_alias to return the unhashed dentry that is referenced by the filp, instead of making it create a new dentry. Leave __d_find_alias() alone to avoid changing behavior of other callers. Also nfsd doesn't need all the checks of __d_find_alias(); any dentry, hashed or unhashed, disconnected or not, should work. Signed-off-by: J. Bruce Fields <bfields@redhat.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2011-01-19 04:45:09 +08:00
res = __d_find_any_alias(inode);
if (res) {
spin_unlock(&inode->i_lock);
dput(tmp);
goto out_iput;
}
/* attach a disconnected dentry */
add_flags = d_flags_for_inode(inode);
if (disconnected)
add_flags |= DCACHE_DISCONNECTED;
spin_lock(&tmp->d_lock);
__d_set_inode_and_type(tmp, inode, add_flags);
hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
hlist_bl_lock(&tmp->d_sb->s_anon);
hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
hlist_bl_unlock(&tmp->d_sb->s_anon);
spin_unlock(&tmp->d_lock);
spin_unlock(&inode->i_lock);
return tmp;
out_iput:
iput(inode);
return res;
}
/**
* d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
* @inode: inode to allocate the dentry for
*
* Obtain a dentry for an inode resulting from NFS filehandle conversion or
* similar open by handle operations. The returned dentry may be anonymous,
* or may have a full name (if the inode was already in the cache).
*
* When called on a directory inode, we must ensure that the inode only ever
* has one dentry. If a dentry is found, that is returned instead of
* allocating a new one.
*
* On successful return, the reference to the inode has been transferred
* to the dentry. In case of an error the reference on the inode is released.
* To make it easier to use in export operations a %NULL or IS_ERR inode may
* be passed in and the error will be propagated to the return value,
* with a %NULL @inode replaced by ERR_PTR(-ESTALE).
*/
struct dentry *d_obtain_alias(struct inode *inode)
{
return __d_obtain_alias(inode, 1);
}
EXPORT_SYMBOL(d_obtain_alias);
/**
* d_obtain_root - find or allocate a dentry for a given inode
* @inode: inode to allocate the dentry for
*
* Obtain an IS_ROOT dentry for the root of a filesystem.
*
* We must ensure that directory inodes only ever have one dentry. If a
* dentry is found, that is returned instead of allocating a new one.
*
* On successful return, the reference to the inode has been transferred
* to the dentry. In case of an error the reference on the inode is
* released. A %NULL or IS_ERR inode may be passed in and will be the
* error will be propagate to the return value, with a %NULL @inode
* replaced by ERR_PTR(-ESTALE).
*/
struct dentry *d_obtain_root(struct inode *inode)
{
return __d_obtain_alias(inode, 0);
}
EXPORT_SYMBOL(d_obtain_root);
/**
* d_add_ci - lookup or allocate new dentry with case-exact name
* @inode: the inode case-insensitive lookup has found
* @dentry: the negative dentry that was passed to the parent's lookup func
* @name: the case-exact name to be associated with the returned dentry
*
* This is to avoid filling the dcache with case-insensitive names to the
* same inode, only the actual correct case is stored in the dcache for
* case-insensitive filesystems.
*
* For a case-insensitive lookup match and if the the case-exact dentry
* already exists in in the dcache, use it and return it.
*
* If no entry exists with the exact case name, allocate new dentry with
* the exact case, and return the spliced entry.
*/
struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
struct qstr *name)
{
struct dentry *found, *res;
/*
* First check if a dentry matching the name already exists,
* if not go ahead and create it now.
*/
found = d_hash_and_lookup(dentry->d_parent, name);
if (found) {
iput(inode);
return found;
}
if (d_in_lookup(dentry)) {
found = d_alloc_parallel(dentry->d_parent, name,
dentry->d_wait);
if (IS_ERR(found) || !d_in_lookup(found)) {
iput(inode);
return found;
}
} else {
found = d_alloc(dentry->d_parent, name);
if (!found) {
iput(inode);
return ERR_PTR(-ENOMEM);
}
}
res = d_splice_alias(inode, found);
if (res) {
dput(found);
return res;
}
return found;
}
EXPORT_SYMBOL(d_add_ci);
vfs: clean up __d_lookup_rcu() and dentry_cmp() interfaces The calling conventions for __d_lookup_rcu() and dentry_cmp() are annoying in different ways, and there is actually one single underlying reason for both of the annoyances. The fundamental reason is that we do the returned dentry sequence number check inside __d_lookup_rcu() instead of doing it in the caller. This results in two annoyances: - __d_lookup_rcu() now not only needs to return the dentry and the sequence number that goes along with the lookup, it also needs to return the inode pointer that was validated by that sequence number check. - and because we did the sequence number check early (to validate the name pointer and length) we also couldn't just pass the dentry itself to dentry_cmp(), we had to pass the counted string that contained the name. So that sequence number decision caused two separate ugly calling conventions. Both of these problems would be solved if we just did the sequence number check in the caller instead. There's only one caller, and that caller already has to do the sequence number check for the parent anyway, so just do that. That allows us to stop returning the dentry->d_inode in that in-out argument (pointer-to-pointer-to-inode), so we can make the inode argument just a regular input inode pointer. The caller can just load the inode from dentry->d_inode, and then do the sequence number check after that to make sure that it's synchronized with the name we looked up. And it allows us to just pass in the dentry to dentry_cmp(), which is what all the callers really wanted. Sure, dentry_cmp() has to be a bit careful about the dentry (which is not stable during RCU lookup), but that's actually very simple. And now that dentry_cmp() can clearly see that the first string argument is a dentry, we can use the direct word access for that, instead of the careful unaligned zero-padding. The dentry name is always properly aligned, since it is a single path component that is either embedded into the dentry itself, or was allocated with kmalloc() (see __d_alloc). Finally, this also uninlines the nasty slow-case for dentry comparisons: that one *does* need to do a sequence number check, since it will call in to the low-level filesystems, and we want to give those a stable inode pointer and path component length/start arguments. Doing an extra sequence check for that slow case is not a problem, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-05-05 05:59:14 +08:00
/*
* Do the slow-case of the dentry name compare.
*
* Unlike the dentry_cmp() function, we need to atomically
* load the name and length information, so that the
vfs: clean up __d_lookup_rcu() and dentry_cmp() interfaces The calling conventions for __d_lookup_rcu() and dentry_cmp() are annoying in different ways, and there is actually one single underlying reason for both of the annoyances. The fundamental reason is that we do the returned dentry sequence number check inside __d_lookup_rcu() instead of doing it in the caller. This results in two annoyances: - __d_lookup_rcu() now not only needs to return the dentry and the sequence number that goes along with the lookup, it also needs to return the inode pointer that was validated by that sequence number check. - and because we did the sequence number check early (to validate the name pointer and length) we also couldn't just pass the dentry itself to dentry_cmp(), we had to pass the counted string that contained the name. So that sequence number decision caused two separate ugly calling conventions. Both of these problems would be solved if we just did the sequence number check in the caller instead. There's only one caller, and that caller already has to do the sequence number check for the parent anyway, so just do that. That allows us to stop returning the dentry->d_inode in that in-out argument (pointer-to-pointer-to-inode), so we can make the inode argument just a regular input inode pointer. The caller can just load the inode from dentry->d_inode, and then do the sequence number check after that to make sure that it's synchronized with the name we looked up. And it allows us to just pass in the dentry to dentry_cmp(), which is what all the callers really wanted. Sure, dentry_cmp() has to be a bit careful about the dentry (which is not stable during RCU lookup), but that's actually very simple. And now that dentry_cmp() can clearly see that the first string argument is a dentry, we can use the direct word access for that, instead of the careful unaligned zero-padding. The dentry name is always properly aligned, since it is a single path component that is either embedded into the dentry itself, or was allocated with kmalloc() (see __d_alloc). Finally, this also uninlines the nasty slow-case for dentry comparisons: that one *does* need to do a sequence number check, since it will call in to the low-level filesystems, and we want to give those a stable inode pointer and path component length/start arguments. Doing an extra sequence check for that slow case is not a problem, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-05-05 05:59:14 +08:00
* filesystem can rely on them, and can use the 'name' and
* 'len' information without worrying about walking off the
* end of memory etc.
*
* Thus the read_seqcount_retry() and the "duplicate" info
* in arguments (the low-level filesystem should not look
* at the dentry inode or name contents directly, since
* rename can change them while we're in RCU mode).
*/
enum slow_d_compare {
D_COMP_OK,
D_COMP_NOMATCH,
D_COMP_SEQRETRY,
};
static noinline enum slow_d_compare slow_dentry_cmp(
const struct dentry *parent,
struct dentry *dentry,
unsigned int seq,
const struct qstr *name)
{
int tlen = dentry->d_name.len;
const char *tname = dentry->d_name.name;
if (read_seqcount_retry(&dentry->d_seq, seq)) {
cpu_relax();
return D_COMP_SEQRETRY;
}
if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
vfs: clean up __d_lookup_rcu() and dentry_cmp() interfaces The calling conventions for __d_lookup_rcu() and dentry_cmp() are annoying in different ways, and there is actually one single underlying reason for both of the annoyances. The fundamental reason is that we do the returned dentry sequence number check inside __d_lookup_rcu() instead of doing it in the caller. This results in two annoyances: - __d_lookup_rcu() now not only needs to return the dentry and the sequence number that goes along with the lookup, it also needs to return the inode pointer that was validated by that sequence number check. - and because we did the sequence number check early (to validate the name pointer and length) we also couldn't just pass the dentry itself to dentry_cmp(), we had to pass the counted string that contained the name. So that sequence number decision caused two separate ugly calling conventions. Both of these problems would be solved if we just did the sequence number check in the caller instead. There's only one caller, and that caller already has to do the sequence number check for the parent anyway, so just do that. That allows us to stop returning the dentry->d_inode in that in-out argument (pointer-to-pointer-to-inode), so we can make the inode argument just a regular input inode pointer. The caller can just load the inode from dentry->d_inode, and then do the sequence number check after that to make sure that it's synchronized with the name we looked up. And it allows us to just pass in the dentry to dentry_cmp(), which is what all the callers really wanted. Sure, dentry_cmp() has to be a bit careful about the dentry (which is not stable during RCU lookup), but that's actually very simple. And now that dentry_cmp() can clearly see that the first string argument is a dentry, we can use the direct word access for that, instead of the careful unaligned zero-padding. The dentry name is always properly aligned, since it is a single path component that is either embedded into the dentry itself, or was allocated with kmalloc() (see __d_alloc). Finally, this also uninlines the nasty slow-case for dentry comparisons: that one *does* need to do a sequence number check, since it will call in to the low-level filesystems, and we want to give those a stable inode pointer and path component length/start arguments. Doing an extra sequence check for that slow case is not a problem, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-05-05 05:59:14 +08:00
return D_COMP_NOMATCH;
return D_COMP_OK;
}
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
/**
* __d_lookup_rcu - search for a dentry (racy, store-free)
* @parent: parent dentry
* @name: qstr of name we wish to find
* @seqp: returns d_seq value at the point where the dentry was found
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
* Returns: dentry, or NULL
*
* __d_lookup_rcu is the dcache lookup function for rcu-walk name
* resolution (store-free path walking) design described in
* Documentation/filesystems/path-lookup.txt.
*
* This is not to be used outside core vfs.
*
* __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
* held, and rcu_read_lock held. The returned dentry must not be stored into
* without taking d_lock and checking d_seq sequence count against @seq
* returned here.
*
* A refcount may be taken on the found dentry with the d_rcu_to_refcount
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
* function.
*
* Alternatively, __d_lookup_rcu may be called again to look up the child of
* the returned dentry, so long as its parent's seqlock is checked after the
* child is looked up. Thus, an interlocking stepping of sequence lock checks
* is formed, giving integrity down the path walk.
vfs: clean up __d_lookup_rcu() and dentry_cmp() interfaces The calling conventions for __d_lookup_rcu() and dentry_cmp() are annoying in different ways, and there is actually one single underlying reason for both of the annoyances. The fundamental reason is that we do the returned dentry sequence number check inside __d_lookup_rcu() instead of doing it in the caller. This results in two annoyances: - __d_lookup_rcu() now not only needs to return the dentry and the sequence number that goes along with the lookup, it also needs to return the inode pointer that was validated by that sequence number check. - and because we did the sequence number check early (to validate the name pointer and length) we also couldn't just pass the dentry itself to dentry_cmp(), we had to pass the counted string that contained the name. So that sequence number decision caused two separate ugly calling conventions. Both of these problems would be solved if we just did the sequence number check in the caller instead. There's only one caller, and that caller already has to do the sequence number check for the parent anyway, so just do that. That allows us to stop returning the dentry->d_inode in that in-out argument (pointer-to-pointer-to-inode), so we can make the inode argument just a regular input inode pointer. The caller can just load the inode from dentry->d_inode, and then do the sequence number check after that to make sure that it's synchronized with the name we looked up. And it allows us to just pass in the dentry to dentry_cmp(), which is what all the callers really wanted. Sure, dentry_cmp() has to be a bit careful about the dentry (which is not stable during RCU lookup), but that's actually very simple. And now that dentry_cmp() can clearly see that the first string argument is a dentry, we can use the direct word access for that, instead of the careful unaligned zero-padding. The dentry name is always properly aligned, since it is a single path component that is either embedded into the dentry itself, or was allocated with kmalloc() (see __d_alloc). Finally, this also uninlines the nasty slow-case for dentry comparisons: that one *does* need to do a sequence number check, since it will call in to the low-level filesystems, and we want to give those a stable inode pointer and path component length/start arguments. Doing an extra sequence check for that slow case is not a problem, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-05-05 05:59:14 +08:00
*
* NOTE! The caller *has* to check the resulting dentry against the sequence
* number we've returned before using any of the resulting dentry state!
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
*/
struct dentry *__d_lookup_rcu(const struct dentry *parent,
const struct qstr *name,
unsigned *seqp)
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
{
u64 hashlen = name->hash_len;
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
const unsigned char *str = name->name;
struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
struct hlist_bl_node *node;
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
struct dentry *dentry;
/*
* Note: There is significant duplication with __d_lookup_rcu which is
* required to prevent single threaded performance regressions
* especially on architectures where smp_rmb (in seqcounts) are costly.
* Keep the two functions in sync.
*/
/*
* The hash list is protected using RCU.
*
* Carefully use d_seq when comparing a candidate dentry, to avoid
* races with d_move().
*
* It is possible that concurrent renames can mess up our list
* walk here and result in missing our dentry, resulting in the
* false-negative result. d_lookup() protects against concurrent
* renames using rename_lock seqlock.
*
* See Documentation/filesystems/path-lookup.txt for more details.
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
*/
hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
unsigned seq;
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
seqretry:
vfs: clean up __d_lookup_rcu() and dentry_cmp() interfaces The calling conventions for __d_lookup_rcu() and dentry_cmp() are annoying in different ways, and there is actually one single underlying reason for both of the annoyances. The fundamental reason is that we do the returned dentry sequence number check inside __d_lookup_rcu() instead of doing it in the caller. This results in two annoyances: - __d_lookup_rcu() now not only needs to return the dentry and the sequence number that goes along with the lookup, it also needs to return the inode pointer that was validated by that sequence number check. - and because we did the sequence number check early (to validate the name pointer and length) we also couldn't just pass the dentry itself to dentry_cmp(), we had to pass the counted string that contained the name. So that sequence number decision caused two separate ugly calling conventions. Both of these problems would be solved if we just did the sequence number check in the caller instead. There's only one caller, and that caller already has to do the sequence number check for the parent anyway, so just do that. That allows us to stop returning the dentry->d_inode in that in-out argument (pointer-to-pointer-to-inode), so we can make the inode argument just a regular input inode pointer. The caller can just load the inode from dentry->d_inode, and then do the sequence number check after that to make sure that it's synchronized with the name we looked up. And it allows us to just pass in the dentry to dentry_cmp(), which is what all the callers really wanted. Sure, dentry_cmp() has to be a bit careful about the dentry (which is not stable during RCU lookup), but that's actually very simple. And now that dentry_cmp() can clearly see that the first string argument is a dentry, we can use the direct word access for that, instead of the careful unaligned zero-padding. The dentry name is always properly aligned, since it is a single path component that is either embedded into the dentry itself, or was allocated with kmalloc() (see __d_alloc). Finally, this also uninlines the nasty slow-case for dentry comparisons: that one *does* need to do a sequence number check, since it will call in to the low-level filesystems, and we want to give those a stable inode pointer and path component length/start arguments. Doing an extra sequence check for that slow case is not a problem, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-05-05 05:59:14 +08:00
/*
* The dentry sequence count protects us from concurrent
* renames, and thus protects parent and name fields.
vfs: clean up __d_lookup_rcu() and dentry_cmp() interfaces The calling conventions for __d_lookup_rcu() and dentry_cmp() are annoying in different ways, and there is actually one single underlying reason for both of the annoyances. The fundamental reason is that we do the returned dentry sequence number check inside __d_lookup_rcu() instead of doing it in the caller. This results in two annoyances: - __d_lookup_rcu() now not only needs to return the dentry and the sequence number that goes along with the lookup, it also needs to return the inode pointer that was validated by that sequence number check. - and because we did the sequence number check early (to validate the name pointer and length) we also couldn't just pass the dentry itself to dentry_cmp(), we had to pass the counted string that contained the name. So that sequence number decision caused two separate ugly calling conventions. Both of these problems would be solved if we just did the sequence number check in the caller instead. There's only one caller, and that caller already has to do the sequence number check for the parent anyway, so just do that. That allows us to stop returning the dentry->d_inode in that in-out argument (pointer-to-pointer-to-inode), so we can make the inode argument just a regular input inode pointer. The caller can just load the inode from dentry->d_inode, and then do the sequence number check after that to make sure that it's synchronized with the name we looked up. And it allows us to just pass in the dentry to dentry_cmp(), which is what all the callers really wanted. Sure, dentry_cmp() has to be a bit careful about the dentry (which is not stable during RCU lookup), but that's actually very simple. And now that dentry_cmp() can clearly see that the first string argument is a dentry, we can use the direct word access for that, instead of the careful unaligned zero-padding. The dentry name is always properly aligned, since it is a single path component that is either embedded into the dentry itself, or was allocated with kmalloc() (see __d_alloc). Finally, this also uninlines the nasty slow-case for dentry comparisons: that one *does* need to do a sequence number check, since it will call in to the low-level filesystems, and we want to give those a stable inode pointer and path component length/start arguments. Doing an extra sequence check for that slow case is not a problem, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-05-05 05:59:14 +08:00
*
* The caller must perform a seqcount check in order
* to do anything useful with the returned dentry.
vfs: clean up __d_lookup_rcu() and dentry_cmp() interfaces The calling conventions for __d_lookup_rcu() and dentry_cmp() are annoying in different ways, and there is actually one single underlying reason for both of the annoyances. The fundamental reason is that we do the returned dentry sequence number check inside __d_lookup_rcu() instead of doing it in the caller. This results in two annoyances: - __d_lookup_rcu() now not only needs to return the dentry and the sequence number that goes along with the lookup, it also needs to return the inode pointer that was validated by that sequence number check. - and because we did the sequence number check early (to validate the name pointer and length) we also couldn't just pass the dentry itself to dentry_cmp(), we had to pass the counted string that contained the name. So that sequence number decision caused two separate ugly calling conventions. Both of these problems would be solved if we just did the sequence number check in the caller instead. There's only one caller, and that caller already has to do the sequence number check for the parent anyway, so just do that. That allows us to stop returning the dentry->d_inode in that in-out argument (pointer-to-pointer-to-inode), so we can make the inode argument just a regular input inode pointer. The caller can just load the inode from dentry->d_inode, and then do the sequence number check after that to make sure that it's synchronized with the name we looked up. And it allows us to just pass in the dentry to dentry_cmp(), which is what all the callers really wanted. Sure, dentry_cmp() has to be a bit careful about the dentry (which is not stable during RCU lookup), but that's actually very simple. And now that dentry_cmp() can clearly see that the first string argument is a dentry, we can use the direct word access for that, instead of the careful unaligned zero-padding. The dentry name is always properly aligned, since it is a single path component that is either embedded into the dentry itself, or was allocated with kmalloc() (see __d_alloc). Finally, this also uninlines the nasty slow-case for dentry comparisons: that one *does* need to do a sequence number check, since it will call in to the low-level filesystems, and we want to give those a stable inode pointer and path component length/start arguments. Doing an extra sequence check for that slow case is not a problem, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-05-05 05:59:14 +08:00
*
* NOTE! We do a "raw" seqcount_begin here. That means that
* we don't wait for the sequence count to stabilize if it
* is in the middle of a sequence change. If we do the slow
* dentry compare, we will do seqretries until it is stable,
* and if we end up with a successful lookup, we actually
* want to exit RCU lookup anyway.
*/
seq = raw_seqcount_begin(&dentry->d_seq);
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
if (dentry->d_parent != parent)
continue;
if (d_unhashed(dentry))
continue;
vfs: clean up __d_lookup_rcu() and dentry_cmp() interfaces The calling conventions for __d_lookup_rcu() and dentry_cmp() are annoying in different ways, and there is actually one single underlying reason for both of the annoyances. The fundamental reason is that we do the returned dentry sequence number check inside __d_lookup_rcu() instead of doing it in the caller. This results in two annoyances: - __d_lookup_rcu() now not only needs to return the dentry and the sequence number that goes along with the lookup, it also needs to return the inode pointer that was validated by that sequence number check. - and because we did the sequence number check early (to validate the name pointer and length) we also couldn't just pass the dentry itself to dentry_cmp(), we had to pass the counted string that contained the name. So that sequence number decision caused two separate ugly calling conventions. Both of these problems would be solved if we just did the sequence number check in the caller instead. There's only one caller, and that caller already has to do the sequence number check for the parent anyway, so just do that. That allows us to stop returning the dentry->d_inode in that in-out argument (pointer-to-pointer-to-inode), so we can make the inode argument just a regular input inode pointer. The caller can just load the inode from dentry->d_inode, and then do the sequence number check after that to make sure that it's synchronized with the name we looked up. And it allows us to just pass in the dentry to dentry_cmp(), which is what all the callers really wanted. Sure, dentry_cmp() has to be a bit careful about the dentry (which is not stable during RCU lookup), but that's actually very simple. And now that dentry_cmp() can clearly see that the first string argument is a dentry, we can use the direct word access for that, instead of the careful unaligned zero-padding. The dentry name is always properly aligned, since it is a single path component that is either embedded into the dentry itself, or was allocated with kmalloc() (see __d_alloc). Finally, this also uninlines the nasty slow-case for dentry comparisons: that one *does* need to do a sequence number check, since it will call in to the low-level filesystems, and we want to give those a stable inode pointer and path component length/start arguments. Doing an extra sequence check for that slow case is not a problem, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-05-05 05:59:14 +08:00
if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
if (dentry->d_name.hash != hashlen_hash(hashlen))
continue;
*seqp = seq;
switch (slow_dentry_cmp(parent, dentry, seq, name)) {
vfs: clean up __d_lookup_rcu() and dentry_cmp() interfaces The calling conventions for __d_lookup_rcu() and dentry_cmp() are annoying in different ways, and there is actually one single underlying reason for both of the annoyances. The fundamental reason is that we do the returned dentry sequence number check inside __d_lookup_rcu() instead of doing it in the caller. This results in two annoyances: - __d_lookup_rcu() now not only needs to return the dentry and the sequence number that goes along with the lookup, it also needs to return the inode pointer that was validated by that sequence number check. - and because we did the sequence number check early (to validate the name pointer and length) we also couldn't just pass the dentry itself to dentry_cmp(), we had to pass the counted string that contained the name. So that sequence number decision caused two separate ugly calling conventions. Both of these problems would be solved if we just did the sequence number check in the caller instead. There's only one caller, and that caller already has to do the sequence number check for the parent anyway, so just do that. That allows us to stop returning the dentry->d_inode in that in-out argument (pointer-to-pointer-to-inode), so we can make the inode argument just a regular input inode pointer. The caller can just load the inode from dentry->d_inode, and then do the sequence number check after that to make sure that it's synchronized with the name we looked up. And it allows us to just pass in the dentry to dentry_cmp(), which is what all the callers really wanted. Sure, dentry_cmp() has to be a bit careful about the dentry (which is not stable during RCU lookup), but that's actually very simple. And now that dentry_cmp() can clearly see that the first string argument is a dentry, we can use the direct word access for that, instead of the careful unaligned zero-padding. The dentry name is always properly aligned, since it is a single path component that is either embedded into the dentry itself, or was allocated with kmalloc() (see __d_alloc). Finally, this also uninlines the nasty slow-case for dentry comparisons: that one *does* need to do a sequence number check, since it will call in to the low-level filesystems, and we want to give those a stable inode pointer and path component length/start arguments. Doing an extra sequence check for that slow case is not a problem, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-05-05 05:59:14 +08:00
case D_COMP_OK:
return dentry;
case D_COMP_NOMATCH:
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
continue;
vfs: clean up __d_lookup_rcu() and dentry_cmp() interfaces The calling conventions for __d_lookup_rcu() and dentry_cmp() are annoying in different ways, and there is actually one single underlying reason for both of the annoyances. The fundamental reason is that we do the returned dentry sequence number check inside __d_lookup_rcu() instead of doing it in the caller. This results in two annoyances: - __d_lookup_rcu() now not only needs to return the dentry and the sequence number that goes along with the lookup, it also needs to return the inode pointer that was validated by that sequence number check. - and because we did the sequence number check early (to validate the name pointer and length) we also couldn't just pass the dentry itself to dentry_cmp(), we had to pass the counted string that contained the name. So that sequence number decision caused two separate ugly calling conventions. Both of these problems would be solved if we just did the sequence number check in the caller instead. There's only one caller, and that caller already has to do the sequence number check for the parent anyway, so just do that. That allows us to stop returning the dentry->d_inode in that in-out argument (pointer-to-pointer-to-inode), so we can make the inode argument just a regular input inode pointer. The caller can just load the inode from dentry->d_inode, and then do the sequence number check after that to make sure that it's synchronized with the name we looked up. And it allows us to just pass in the dentry to dentry_cmp(), which is what all the callers really wanted. Sure, dentry_cmp() has to be a bit careful about the dentry (which is not stable during RCU lookup), but that's actually very simple. And now that dentry_cmp() can clearly see that the first string argument is a dentry, we can use the direct word access for that, instead of the careful unaligned zero-padding. The dentry name is always properly aligned, since it is a single path component that is either embedded into the dentry itself, or was allocated with kmalloc() (see __d_alloc). Finally, this also uninlines the nasty slow-case for dentry comparisons: that one *does* need to do a sequence number check, since it will call in to the low-level filesystems, and we want to give those a stable inode pointer and path component length/start arguments. Doing an extra sequence check for that slow case is not a problem, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-05-05 05:59:14 +08:00
default:
goto seqretry;
}
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
}
vfs: clean up __d_lookup_rcu() and dentry_cmp() interfaces The calling conventions for __d_lookup_rcu() and dentry_cmp() are annoying in different ways, and there is actually one single underlying reason for both of the annoyances. The fundamental reason is that we do the returned dentry sequence number check inside __d_lookup_rcu() instead of doing it in the caller. This results in two annoyances: - __d_lookup_rcu() now not only needs to return the dentry and the sequence number that goes along with the lookup, it also needs to return the inode pointer that was validated by that sequence number check. - and because we did the sequence number check early (to validate the name pointer and length) we also couldn't just pass the dentry itself to dentry_cmp(), we had to pass the counted string that contained the name. So that sequence number decision caused two separate ugly calling conventions. Both of these problems would be solved if we just did the sequence number check in the caller instead. There's only one caller, and that caller already has to do the sequence number check for the parent anyway, so just do that. That allows us to stop returning the dentry->d_inode in that in-out argument (pointer-to-pointer-to-inode), so we can make the inode argument just a regular input inode pointer. The caller can just load the inode from dentry->d_inode, and then do the sequence number check after that to make sure that it's synchronized with the name we looked up. And it allows us to just pass in the dentry to dentry_cmp(), which is what all the callers really wanted. Sure, dentry_cmp() has to be a bit careful about the dentry (which is not stable during RCU lookup), but that's actually very simple. And now that dentry_cmp() can clearly see that the first string argument is a dentry, we can use the direct word access for that, instead of the careful unaligned zero-padding. The dentry name is always properly aligned, since it is a single path component that is either embedded into the dentry itself, or was allocated with kmalloc() (see __d_alloc). Finally, this also uninlines the nasty slow-case for dentry comparisons: that one *does* need to do a sequence number check, since it will call in to the low-level filesystems, and we want to give those a stable inode pointer and path component length/start arguments. Doing an extra sequence check for that slow case is not a problem, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-05-05 05:59:14 +08:00
if (dentry->d_name.hash_len != hashlen)
continue;
*seqp = seq;
if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
vfs: clean up __d_lookup_rcu() and dentry_cmp() interfaces The calling conventions for __d_lookup_rcu() and dentry_cmp() are annoying in different ways, and there is actually one single underlying reason for both of the annoyances. The fundamental reason is that we do the returned dentry sequence number check inside __d_lookup_rcu() instead of doing it in the caller. This results in two annoyances: - __d_lookup_rcu() now not only needs to return the dentry and the sequence number that goes along with the lookup, it also needs to return the inode pointer that was validated by that sequence number check. - and because we did the sequence number check early (to validate the name pointer and length) we also couldn't just pass the dentry itself to dentry_cmp(), we had to pass the counted string that contained the name. So that sequence number decision caused two separate ugly calling conventions. Both of these problems would be solved if we just did the sequence number check in the caller instead. There's only one caller, and that caller already has to do the sequence number check for the parent anyway, so just do that. That allows us to stop returning the dentry->d_inode in that in-out argument (pointer-to-pointer-to-inode), so we can make the inode argument just a regular input inode pointer. The caller can just load the inode from dentry->d_inode, and then do the sequence number check after that to make sure that it's synchronized with the name we looked up. And it allows us to just pass in the dentry to dentry_cmp(), which is what all the callers really wanted. Sure, dentry_cmp() has to be a bit careful about the dentry (which is not stable during RCU lookup), but that's actually very simple. And now that dentry_cmp() can clearly see that the first string argument is a dentry, we can use the direct word access for that, instead of the careful unaligned zero-padding. The dentry name is always properly aligned, since it is a single path component that is either embedded into the dentry itself, or was allocated with kmalloc() (see __d_alloc). Finally, this also uninlines the nasty slow-case for dentry comparisons: that one *does* need to do a sequence number check, since it will call in to the low-level filesystems, and we want to give those a stable inode pointer and path component length/start arguments. Doing an extra sequence check for that slow case is not a problem, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-05-05 05:59:14 +08:00
return dentry;
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
}
return NULL;
}
/**
* d_lookup - search for a dentry
* @parent: parent dentry
* @name: qstr of name we wish to find
* Returns: dentry, or NULL
*
* d_lookup searches the children of the parent dentry for the name in
* question. If the dentry is found its reference count is incremented and the
* dentry is returned. The caller must use dput to free the entry when it has
* finished using it. %NULL is returned if the dentry does not exist.
*/
struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
{
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
struct dentry *dentry;
unsigned seq;
do {
seq = read_seqbegin(&rename_lock);
dentry = __d_lookup(parent, name);
if (dentry)
break;
} while (read_seqretry(&rename_lock, seq));
return dentry;
}
EXPORT_SYMBOL(d_lookup);
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
/**
* __d_lookup - search for a dentry (racy)
* @parent: parent dentry
* @name: qstr of name we wish to find
* Returns: dentry, or NULL
*
* __d_lookup is like d_lookup, however it may (rarely) return a
* false-negative result due to unrelated rename activity.
*
* __d_lookup is slightly faster by avoiding rename_lock read seqlock,
* however it must be used carefully, eg. with a following d_lookup in
* the case of failure.
*
* __d_lookup callers must be commented.
*/
struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
{
unsigned int len = name->len;
unsigned int hash = name->hash;
const unsigned char *str = name->name;
struct hlist_bl_head *b = d_hash(parent, hash);
struct hlist_bl_node *node;
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
struct dentry *found = NULL;
struct dentry *dentry;
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
/*
* Note: There is significant duplication with __d_lookup_rcu which is
* required to prevent single threaded performance regressions
* especially on architectures where smp_rmb (in seqcounts) are costly.
* Keep the two functions in sync.
*/
/*
* The hash list is protected using RCU.
*
* Take d_lock when comparing a candidate dentry, to avoid races
* with d_move().
*
* It is possible that concurrent renames can mess up our list
* walk here and result in missing our dentry, resulting in the
* false-negative result. d_lookup() protects against concurrent
* renames using rename_lock seqlock.
*
* See Documentation/filesystems/path-lookup.txt for more details.
*/
rcu_read_lock();
hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
if (dentry->d_name.hash != hash)
continue;
spin_lock(&dentry->d_lock);
if (dentry->d_parent != parent)
goto next;
Fix NULL pointer dereference in proc_sys_compare The VFS interface for the 'd_compare()' is a bit special (read: 'odd'), because it really just essentially replaces a memcmp(). The filesystem is supposed to just compare the two names with whatever case-independent or other function. And when I say 'is supposed to', I obviously mean that 'procfs does odd things, and actually looks at the dentry that we don't even pass down, rather than just the name'. Which results in problems, because we actually call d_compare before we have even verified that the dentry is still hashed at all. And that causes a problm since the inode that procfs looks at may have been free'd and the d_inode pointer is NULL. procfs just assumes that all dentries are positive, since procfs itself never generates a negative one. But memory pressure will still result in the dentry getting torn down, and as it is removed by RCU, it still remains visible on some lists - and to d_compare. If the filesystem just did a name comparison, we wouldn't care. And we could just fix procfs to know about negative dentries too. But rather than have the low-level filesystems know about internal VFS details, just move the check for a unhashed dentry up a bit, so that we will only call d_compare on dentries that are still active. The actual oops this caused didn't look like a NULL pointer dereference because procfs did a 'container_of(inode, struct proc_inode, vfs_inode)' to get at its internal proc_inode information from the inode pointer, and accessed a field below the inode. So the oops would look something like BUG: unable to handle kernel paging request at fffffffffffffff0 IP: [<ffffffff802bc6c6>] proc_sys_compare+0x36/0x50 and was seen on both x86-64 (Alexey Dobriyan and Hugh Dickins) and ppc64 (Hugh Dickins). Reported-by: Alexey Dobriyan <adobriyan@gmail.com> Acked-by: Hugh Dickins <hugh@veritas.com> Cc: Al Viro <viro@ZenIV.linux.org.uk> Reviewed-by: "Eric W. Biederman" <ebiederm@xmission.com> Signed-of-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-09-29 22:42:57 +08:00
if (d_unhashed(dentry))
goto next;
/*
* It is safe to compare names since d_move() cannot
* change the qstr (protected by d_lock).
*/
if (parent->d_flags & DCACHE_OP_COMPARE) {
vfs: clean up __d_lookup_rcu() and dentry_cmp() interfaces The calling conventions for __d_lookup_rcu() and dentry_cmp() are annoying in different ways, and there is actually one single underlying reason for both of the annoyances. The fundamental reason is that we do the returned dentry sequence number check inside __d_lookup_rcu() instead of doing it in the caller. This results in two annoyances: - __d_lookup_rcu() now not only needs to return the dentry and the sequence number that goes along with the lookup, it also needs to return the inode pointer that was validated by that sequence number check. - and because we did the sequence number check early (to validate the name pointer and length) we also couldn't just pass the dentry itself to dentry_cmp(), we had to pass the counted string that contained the name. So that sequence number decision caused two separate ugly calling conventions. Both of these problems would be solved if we just did the sequence number check in the caller instead. There's only one caller, and that caller already has to do the sequence number check for the parent anyway, so just do that. That allows us to stop returning the dentry->d_inode in that in-out argument (pointer-to-pointer-to-inode), so we can make the inode argument just a regular input inode pointer. The caller can just load the inode from dentry->d_inode, and then do the sequence number check after that to make sure that it's synchronized with the name we looked up. And it allows us to just pass in the dentry to dentry_cmp(), which is what all the callers really wanted. Sure, dentry_cmp() has to be a bit careful about the dentry (which is not stable during RCU lookup), but that's actually very simple. And now that dentry_cmp() can clearly see that the first string argument is a dentry, we can use the direct word access for that, instead of the careful unaligned zero-padding. The dentry name is always properly aligned, since it is a single path component that is either embedded into the dentry itself, or was allocated with kmalloc() (see __d_alloc). Finally, this also uninlines the nasty slow-case for dentry comparisons: that one *does* need to do a sequence number check, since it will call in to the low-level filesystems, and we want to give those a stable inode pointer and path component length/start arguments. Doing an extra sequence check for that slow case is not a problem, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-05-05 05:59:14 +08:00
int tlen = dentry->d_name.len;
const char *tname = dentry->d_name.name;
if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
goto next;
} else {
if (dentry->d_name.len != len)
goto next;
vfs: clean up __d_lookup_rcu() and dentry_cmp() interfaces The calling conventions for __d_lookup_rcu() and dentry_cmp() are annoying in different ways, and there is actually one single underlying reason for both of the annoyances. The fundamental reason is that we do the returned dentry sequence number check inside __d_lookup_rcu() instead of doing it in the caller. This results in two annoyances: - __d_lookup_rcu() now not only needs to return the dentry and the sequence number that goes along with the lookup, it also needs to return the inode pointer that was validated by that sequence number check. - and because we did the sequence number check early (to validate the name pointer and length) we also couldn't just pass the dentry itself to dentry_cmp(), we had to pass the counted string that contained the name. So that sequence number decision caused two separate ugly calling conventions. Both of these problems would be solved if we just did the sequence number check in the caller instead. There's only one caller, and that caller already has to do the sequence number check for the parent anyway, so just do that. That allows us to stop returning the dentry->d_inode in that in-out argument (pointer-to-pointer-to-inode), so we can make the inode argument just a regular input inode pointer. The caller can just load the inode from dentry->d_inode, and then do the sequence number check after that to make sure that it's synchronized with the name we looked up. And it allows us to just pass in the dentry to dentry_cmp(), which is what all the callers really wanted. Sure, dentry_cmp() has to be a bit careful about the dentry (which is not stable during RCU lookup), but that's actually very simple. And now that dentry_cmp() can clearly see that the first string argument is a dentry, we can use the direct word access for that, instead of the careful unaligned zero-padding. The dentry name is always properly aligned, since it is a single path component that is either embedded into the dentry itself, or was allocated with kmalloc() (see __d_alloc). Finally, this also uninlines the nasty slow-case for dentry comparisons: that one *does* need to do a sequence number check, since it will call in to the low-level filesystems, and we want to give those a stable inode pointer and path component length/start arguments. Doing an extra sequence check for that slow case is not a problem, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-05-05 05:59:14 +08:00
if (dentry_cmp(dentry, str, len))
goto next;
}
dentry->d_lockref.count++;
Fix NULL pointer dereference in proc_sys_compare The VFS interface for the 'd_compare()' is a bit special (read: 'odd'), because it really just essentially replaces a memcmp(). The filesystem is supposed to just compare the two names with whatever case-independent or other function. And when I say 'is supposed to', I obviously mean that 'procfs does odd things, and actually looks at the dentry that we don't even pass down, rather than just the name'. Which results in problems, because we actually call d_compare before we have even verified that the dentry is still hashed at all. And that causes a problm since the inode that procfs looks at may have been free'd and the d_inode pointer is NULL. procfs just assumes that all dentries are positive, since procfs itself never generates a negative one. But memory pressure will still result in the dentry getting torn down, and as it is removed by RCU, it still remains visible on some lists - and to d_compare. If the filesystem just did a name comparison, we wouldn't care. And we could just fix procfs to know about negative dentries too. But rather than have the low-level filesystems know about internal VFS details, just move the check for a unhashed dentry up a bit, so that we will only call d_compare on dentries that are still active. The actual oops this caused didn't look like a NULL pointer dereference because procfs did a 'container_of(inode, struct proc_inode, vfs_inode)' to get at its internal proc_inode information from the inode pointer, and accessed a field below the inode. So the oops would look something like BUG: unable to handle kernel paging request at fffffffffffffff0 IP: [<ffffffff802bc6c6>] proc_sys_compare+0x36/0x50 and was seen on both x86-64 (Alexey Dobriyan and Hugh Dickins) and ppc64 (Hugh Dickins). Reported-by: Alexey Dobriyan <adobriyan@gmail.com> Acked-by: Hugh Dickins <hugh@veritas.com> Cc: Al Viro <viro@ZenIV.linux.org.uk> Reviewed-by: "Eric W. Biederman" <ebiederm@xmission.com> Signed-of-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-09-29 22:42:57 +08:00
found = dentry;
spin_unlock(&dentry->d_lock);
break;
next:
spin_unlock(&dentry->d_lock);
}
rcu_read_unlock();
return found;
}
/**
* d_hash_and_lookup - hash the qstr then search for a dentry
* @dir: Directory to search in
* @name: qstr of name we wish to find
*
* On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
*/
struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
{
/*
* Check for a fs-specific hash function. Note that we must
* calculate the standard hash first, as the d_op->d_hash()
* routine may choose to leave the hash value unchanged.
*/
name->hash = full_name_hash(name->name, name->len);
if (dir->d_flags & DCACHE_OP_HASH) {
int err = dir->d_op->d_hash(dir, name);
if (unlikely(err < 0))
return ERR_PTR(err);
}
return d_lookup(dir, name);
}
EXPORT_SYMBOL(d_hash_and_lookup);
/*
* When a file is deleted, we have two options:
* - turn this dentry into a negative dentry
* - unhash this dentry and free it.
*
* Usually, we want to just turn this into
* a negative dentry, but if anybody else is
* currently using the dentry or the inode
* we can't do that and we fall back on removing
* it from the hash queues and waiting for
* it to be deleted later when it has no users
*/
/**
* d_delete - delete a dentry
* @dentry: The dentry to delete
*
* Turn the dentry into a negative dentry if possible, otherwise
* remove it from the hash queues so it can be deleted later
*/
void d_delete(struct dentry * dentry)
{
struct inode *inode;
int isdir = 0;
/*
* Are we the only user?
*/
again:
spin_lock(&dentry->d_lock);
inode = dentry->d_inode;
isdir = S_ISDIR(inode->i_mode);
if (dentry->d_lockref.count == 1) {
if (!spin_trylock(&inode->i_lock)) {
spin_unlock(&dentry->d_lock);
cpu_relax();
goto again;
}
dentry->d_flags &= ~DCACHE_CANT_MOUNT;
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
dentry_unlink_inode(dentry);
fsnotify_nameremove(dentry, isdir);
return;
}
if (!d_unhashed(dentry))
__d_drop(dentry);
spin_unlock(&dentry->d_lock);
fsnotify_nameremove(dentry, isdir);
}
EXPORT_SYMBOL(d_delete);
static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
{
BUG_ON(!d_unhashed(entry));
hlist_bl_lock(b);
vfs: get rid of insane dentry hashing rules The dentry hashing rules have been really quite complicated for a long while, in odd ways. That made functions like __d_drop() very fragile and non-obvious. In particular, whether a dentry was hashed or not was indicated with an explicit DCACHE_UNHASHED bit. That's despite the fact that the hash abstraction that the dentries use actually have a 'is this entry hashed or not' model (which is a simple test of the 'pprev' pointer). The reason that was done is because we used the normal 'is this entry unhashed' model to mark whether the dentry had _ever_ been hashed in the dentry hash tables, and that logic goes back many years (commit b3423415fbc2: "dcache: avoid RCU for never-hashed dentries"). That, in turn, meant that __d_drop had totally different unhashing logic for the dentry hash table case and for the anonymous dcache case, because in order to use the "is this dentry hashed" logic as a flag for whether it had ever been on the RCU hash table, we had to unhash such a dentry differently so that we'd never think that it wasn't 'unhashed' and wouldn't be free'd correctly. That's just insane. It made the logic really hard to follow, when there were two different kinds of "unhashed" states, and one of them (the one that used "list_bl_unhashed()") really had nothing at all to do with being unhashed per se, but with a very subtle lifetime rule instead. So turn all of it around, and make it logical. Instead of having a DENTRY_UNHASHED bit in d_flags to indicate whether the dentry is on the hash chains or not, use the hash chain unhashed logic for that. Suddenly "d_unhashed()" just uses "list_bl_unhashed()", and everything makes sense. And for the lifetime rule, just use an explicit DENTRY_RCUACCEES bit. If we ever insert the dentry into the dentry hash table so that it is visible to RCU lookup, we mark it DENTRY_RCUACCESS to show that it now needs the RCU lifetime rules. Now suddently that test at dentry free time makes sense too. And because unhashing now is sane and doesn't depend on where the dentry got unhashed from (because the dentry hash chain details doesn't have some subtle side effects), we can re-unify the __d_drop() logic and use common code for the unhashing. Also fix one more open-coded hash chain bit_spin_lock() that I missed in the previous chain locking cleanup commit. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-04-24 22:58:46 +08:00
entry->d_flags |= DCACHE_RCUACCESS;
hlist_bl_add_head_rcu(&entry->d_hash, b);
hlist_bl_unlock(b);
}
static void _d_rehash(struct dentry * entry)
{
__d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
}
/**
* d_rehash - add an entry back to the hash
* @entry: dentry to add to the hash
*
* Adds a dentry to the hash according to its name.
*/
void d_rehash(struct dentry * entry)
{
spin_lock(&entry->d_lock);
_d_rehash(entry);
spin_unlock(&entry->d_lock);
}
EXPORT_SYMBOL(d_rehash);
parallel lookups machinery, part 2 We'll need to verify that there's neither a hashed nor in-lookup dentry with desired parent/name before adding to in-lookup set. One possible solution would be to hold the parent's ->d_lock through both checks, but while the in-lookup set is relatively small at any time, dcache is not. And holding the parent's ->d_lock through something like __d_lookup_rcu() would suck too badly. So we leave the parent's ->d_lock alone, which means that we watch out for the following scenario: * we verify that there's no hashed match * existing in-lookup match gets hashed by another process * we verify that there's no in-lookup matches and decide that everything's fine. Solution: per-directory kinda-sorta seqlock, bumped around the times we hash something that used to be in-lookup or move (and hash) something in place of in-lookup. Then the above would turn into * read the counter * do dcache lookup * if no matches found, check for in-lookup matches * if there had been none of those either, check if the counter has changed; repeat if it has. The "kinda-sorta" part is due to the fact that we don't have much spare space in inode. There is a spare word (shared with i_bdev/i_cdev/i_pipe), so the counter part is not a problem, but spinlock is a different story. We could use the parent's ->d_lock, and it would be less painful in terms of contention, for __d_add() it would be rather inconvenient to grab; we could do that (using lock_parent()), but... Fortunately, we can get serialization on the counter itself, and it might be a good idea in general; we can use cmpxchg() in a loop to get from even to odd and smp_store_release() from odd to even. This commit adds the counter and updating logics; the readers will be added in the next commit. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 12:58:55 +08:00
static inline unsigned start_dir_add(struct inode *dir)
{
for (;;) {
unsigned n = dir->i_dir_seq;
if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
return n;
cpu_relax();
}
}
static inline void end_dir_add(struct inode *dir, unsigned n)
{
smp_store_release(&dir->i_dir_seq, n + 2);
}
static void d_wait_lookup(struct dentry *dentry)
{
if (d_in_lookup(dentry)) {
DECLARE_WAITQUEUE(wait, current);
add_wait_queue(dentry->d_wait, &wait);
do {
set_current_state(TASK_UNINTERRUPTIBLE);
spin_unlock(&dentry->d_lock);
schedule();
spin_lock(&dentry->d_lock);
} while (d_in_lookup(dentry));
}
}
parallel lookups machinery, part 3 We will need to be able to check if there is an in-lookup dentry with matching parent/name. Right now it's impossible, but as soon as start locking directories shared such beasts will appear. Add a secondary hash for locating those. Hash chains go through the same space where d_alias will be once it's not in-lookup anymore. Search is done under the same bitlock we use for modifications - with the primary hash we can rely on d_rehash() into the wrong chain being the worst that could happen, but here the pointers are buggered once it's removed from the chain. On the other hand, the chains are not going to be long and normally we'll end up adding to the chain anyway. That allows us to avoid bothering with ->d_lock when doing the comparisons - everything is stable until removed from chain. New helper: d_alloc_parallel(). Right now it allocates, verifies that no hashed and in-lookup matches exist and adds to in-lookup hash. Returns ERR_PTR() for error, hashed match (in the unlikely case it's been found) or new dentry. In-lookup matches trigger BUG() for now; that will change in the next commit when we introduce waiting for ongoing lookup to finish. Note that in-lookup matches won't be possible until we actually go for shared locking. lookup_slow() switched to use of d_alloc_parallel(). Again, these commits are separated only for making it easier to review. All this machinery will start doing something useful only when we go for shared locking; it's just that the combination is too large for my taste. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 14:42:04 +08:00
struct dentry *d_alloc_parallel(struct dentry *parent,
const struct qstr *name,
wait_queue_head_t *wq)
parallel lookups machinery, part 3 We will need to be able to check if there is an in-lookup dentry with matching parent/name. Right now it's impossible, but as soon as start locking directories shared such beasts will appear. Add a secondary hash for locating those. Hash chains go through the same space where d_alias will be once it's not in-lookup anymore. Search is done under the same bitlock we use for modifications - with the primary hash we can rely on d_rehash() into the wrong chain being the worst that could happen, but here the pointers are buggered once it's removed from the chain. On the other hand, the chains are not going to be long and normally we'll end up adding to the chain anyway. That allows us to avoid bothering with ->d_lock when doing the comparisons - everything is stable until removed from chain. New helper: d_alloc_parallel(). Right now it allocates, verifies that no hashed and in-lookup matches exist and adds to in-lookup hash. Returns ERR_PTR() for error, hashed match (in the unlikely case it's been found) or new dentry. In-lookup matches trigger BUG() for now; that will change in the next commit when we introduce waiting for ongoing lookup to finish. Note that in-lookup matches won't be possible until we actually go for shared locking. lookup_slow() switched to use of d_alloc_parallel(). Again, these commits are separated only for making it easier to review. All this machinery will start doing something useful only when we go for shared locking; it's just that the combination is too large for my taste. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 14:42:04 +08:00
{
unsigned int len = name->len;
unsigned int hash = name->hash;
const unsigned char *str = name->name;
struct hlist_bl_head *b = in_lookup_hash(parent, hash);
struct hlist_bl_node *node;
struct dentry *new = d_alloc(parent, name);
struct dentry *dentry;
unsigned seq, r_seq, d_seq;
if (unlikely(!new))
return ERR_PTR(-ENOMEM);
retry:
rcu_read_lock();
seq = smp_load_acquire(&parent->d_inode->i_dir_seq) & ~1;
r_seq = read_seqbegin(&rename_lock);
dentry = __d_lookup_rcu(parent, name, &d_seq);
if (unlikely(dentry)) {
if (!lockref_get_not_dead(&dentry->d_lockref)) {
rcu_read_unlock();
goto retry;
}
if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
rcu_read_unlock();
dput(dentry);
goto retry;
}
rcu_read_unlock();
dput(new);
return dentry;
}
if (unlikely(read_seqretry(&rename_lock, r_seq))) {
rcu_read_unlock();
goto retry;
}
hlist_bl_lock(b);
if (unlikely(parent->d_inode->i_dir_seq != seq)) {
hlist_bl_unlock(b);
rcu_read_unlock();
goto retry;
}
rcu_read_unlock();
/*
* No changes for the parent since the beginning of d_lookup().
* Since all removals from the chain happen with hlist_bl_lock(),
* any potential in-lookup matches are going to stay here until
* we unlock the chain. All fields are stable in everything
* we encounter.
*/
hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
if (dentry->d_name.hash != hash)
continue;
if (dentry->d_parent != parent)
continue;
if (d_unhashed(dentry))
continue;
if (parent->d_flags & DCACHE_OP_COMPARE) {
int tlen = dentry->d_name.len;
const char *tname = dentry->d_name.name;
if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
continue;
} else {
if (dentry->d_name.len != len)
continue;
if (dentry_cmp(dentry, str, len))
continue;
}
dget(dentry);
hlist_bl_unlock(b);
/* somebody is doing lookup for it right now; wait for it */
spin_lock(&dentry->d_lock);
d_wait_lookup(dentry);
/*
* it's not in-lookup anymore; in principle we should repeat
* everything from dcache lookup, but it's likely to be what
* d_lookup() would've found anyway. If it is, just return it;
* otherwise we really have to repeat the whole thing.
*/
if (unlikely(dentry->d_name.hash != hash))
goto mismatch;
if (unlikely(dentry->d_parent != parent))
goto mismatch;
if (unlikely(d_unhashed(dentry)))
goto mismatch;
if (parent->d_flags & DCACHE_OP_COMPARE) {
int tlen = dentry->d_name.len;
const char *tname = dentry->d_name.name;
if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
goto mismatch;
} else {
if (unlikely(dentry->d_name.len != len))
goto mismatch;
if (unlikely(dentry_cmp(dentry, str, len)))
goto mismatch;
}
/* OK, it *is* a hashed match; return it */
spin_unlock(&dentry->d_lock);
parallel lookups machinery, part 3 We will need to be able to check if there is an in-lookup dentry with matching parent/name. Right now it's impossible, but as soon as start locking directories shared such beasts will appear. Add a secondary hash for locating those. Hash chains go through the same space where d_alias will be once it's not in-lookup anymore. Search is done under the same bitlock we use for modifications - with the primary hash we can rely on d_rehash() into the wrong chain being the worst that could happen, but here the pointers are buggered once it's removed from the chain. On the other hand, the chains are not going to be long and normally we'll end up adding to the chain anyway. That allows us to avoid bothering with ->d_lock when doing the comparisons - everything is stable until removed from chain. New helper: d_alloc_parallel(). Right now it allocates, verifies that no hashed and in-lookup matches exist and adds to in-lookup hash. Returns ERR_PTR() for error, hashed match (in the unlikely case it's been found) or new dentry. In-lookup matches trigger BUG() for now; that will change in the next commit when we introduce waiting for ongoing lookup to finish. Note that in-lookup matches won't be possible until we actually go for shared locking. lookup_slow() switched to use of d_alloc_parallel(). Again, these commits are separated only for making it easier to review. All this machinery will start doing something useful only when we go for shared locking; it's just that the combination is too large for my taste. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 14:42:04 +08:00
dput(new);
return dentry;
}
/* we can't take ->d_lock here; it's OK, though. */
new->d_flags |= DCACHE_PAR_LOOKUP;
new->d_wait = wq;
parallel lookups machinery, part 3 We will need to be able to check if there is an in-lookup dentry with matching parent/name. Right now it's impossible, but as soon as start locking directories shared such beasts will appear. Add a secondary hash for locating those. Hash chains go through the same space where d_alias will be once it's not in-lookup anymore. Search is done under the same bitlock we use for modifications - with the primary hash we can rely on d_rehash() into the wrong chain being the worst that could happen, but here the pointers are buggered once it's removed from the chain. On the other hand, the chains are not going to be long and normally we'll end up adding to the chain anyway. That allows us to avoid bothering with ->d_lock when doing the comparisons - everything is stable until removed from chain. New helper: d_alloc_parallel(). Right now it allocates, verifies that no hashed and in-lookup matches exist and adds to in-lookup hash. Returns ERR_PTR() for error, hashed match (in the unlikely case it's been found) or new dentry. In-lookup matches trigger BUG() for now; that will change in the next commit when we introduce waiting for ongoing lookup to finish. Note that in-lookup matches won't be possible until we actually go for shared locking. lookup_slow() switched to use of d_alloc_parallel(). Again, these commits are separated only for making it easier to review. All this machinery will start doing something useful only when we go for shared locking; it's just that the combination is too large for my taste. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 14:42:04 +08:00
hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
hlist_bl_unlock(b);
return new;
mismatch:
spin_unlock(&dentry->d_lock);
dput(dentry);
goto retry;
parallel lookups machinery, part 3 We will need to be able to check if there is an in-lookup dentry with matching parent/name. Right now it's impossible, but as soon as start locking directories shared such beasts will appear. Add a secondary hash for locating those. Hash chains go through the same space where d_alias will be once it's not in-lookup anymore. Search is done under the same bitlock we use for modifications - with the primary hash we can rely on d_rehash() into the wrong chain being the worst that could happen, but here the pointers are buggered once it's removed from the chain. On the other hand, the chains are not going to be long and normally we'll end up adding to the chain anyway. That allows us to avoid bothering with ->d_lock when doing the comparisons - everything is stable until removed from chain. New helper: d_alloc_parallel(). Right now it allocates, verifies that no hashed and in-lookup matches exist and adds to in-lookup hash. Returns ERR_PTR() for error, hashed match (in the unlikely case it's been found) or new dentry. In-lookup matches trigger BUG() for now; that will change in the next commit when we introduce waiting for ongoing lookup to finish. Note that in-lookup matches won't be possible until we actually go for shared locking. lookup_slow() switched to use of d_alloc_parallel(). Again, these commits are separated only for making it easier to review. All this machinery will start doing something useful only when we go for shared locking; it's just that the combination is too large for my taste. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 14:42:04 +08:00
}
EXPORT_SYMBOL(d_alloc_parallel);
beginning of transition to parallel lookups - marking in-lookup dentries marked as such when (would be) parallel lookup is about to pass them to actual ->lookup(); unmarked when * __d_add() is about to make it hashed, positive or not. * __d_move() (from d_splice_alias(), directly or via __d_unalias()) puts a preexisting dentry in its place * in caller of ->lookup() if it has escaped all of the above. Bug (WARN_ON, actually) if it reaches the final dput() or d_instantiate() while still marked such. As the result, we are guaranteed that for as long as the flag is set, dentry will * remain negative unhashed with positive refcount * never have its ->d_alias looked at * never have its ->d_lru looked at * never have its ->d_parent and ->d_name changed Right now we have at most one such for any given parent directory. With parallel lookups that restriction will weaken to * only exist when parent is locked shared * at most one with given (parent,name) pair (comparison of names is according to ->d_compare()) * only exist when there's no hashed dentry with the same (parent,name) Transition will take the next several commits; unfortunately, we'll only be able to switch to rwsem at the end of this series. The reason for not making it a single patch is to simplify review. New primitives: d_in_lookup() (a predicate checking if dentry is in the in-lookup state) and d_lookup_done() (tells the system that we are done with lookup and if it's still marked as in-lookup, it should cease to be such). Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 07:52:13 +08:00
void __d_lookup_done(struct dentry *dentry)
{
parallel lookups machinery, part 3 We will need to be able to check if there is an in-lookup dentry with matching parent/name. Right now it's impossible, but as soon as start locking directories shared such beasts will appear. Add a secondary hash for locating those. Hash chains go through the same space where d_alias will be once it's not in-lookup anymore. Search is done under the same bitlock we use for modifications - with the primary hash we can rely on d_rehash() into the wrong chain being the worst that could happen, but here the pointers are buggered once it's removed from the chain. On the other hand, the chains are not going to be long and normally we'll end up adding to the chain anyway. That allows us to avoid bothering with ->d_lock when doing the comparisons - everything is stable until removed from chain. New helper: d_alloc_parallel(). Right now it allocates, verifies that no hashed and in-lookup matches exist and adds to in-lookup hash. Returns ERR_PTR() for error, hashed match (in the unlikely case it's been found) or new dentry. In-lookup matches trigger BUG() for now; that will change in the next commit when we introduce waiting for ongoing lookup to finish. Note that in-lookup matches won't be possible until we actually go for shared locking. lookup_slow() switched to use of d_alloc_parallel(). Again, these commits are separated only for making it easier to review. All this machinery will start doing something useful only when we go for shared locking; it's just that the combination is too large for my taste. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 14:42:04 +08:00
struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
dentry->d_name.hash);
hlist_bl_lock(b);
beginning of transition to parallel lookups - marking in-lookup dentries marked as such when (would be) parallel lookup is about to pass them to actual ->lookup(); unmarked when * __d_add() is about to make it hashed, positive or not. * __d_move() (from d_splice_alias(), directly or via __d_unalias()) puts a preexisting dentry in its place * in caller of ->lookup() if it has escaped all of the above. Bug (WARN_ON, actually) if it reaches the final dput() or d_instantiate() while still marked such. As the result, we are guaranteed that for as long as the flag is set, dentry will * remain negative unhashed with positive refcount * never have its ->d_alias looked at * never have its ->d_lru looked at * never have its ->d_parent and ->d_name changed Right now we have at most one such for any given parent directory. With parallel lookups that restriction will weaken to * only exist when parent is locked shared * at most one with given (parent,name) pair (comparison of names is according to ->d_compare()) * only exist when there's no hashed dentry with the same (parent,name) Transition will take the next several commits; unfortunately, we'll only be able to switch to rwsem at the end of this series. The reason for not making it a single patch is to simplify review. New primitives: d_in_lookup() (a predicate checking if dentry is in the in-lookup state) and d_lookup_done() (tells the system that we are done with lookup and if it's still marked as in-lookup, it should cease to be such). Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 07:52:13 +08:00
dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
parallel lookups machinery, part 3 We will need to be able to check if there is an in-lookup dentry with matching parent/name. Right now it's impossible, but as soon as start locking directories shared such beasts will appear. Add a secondary hash for locating those. Hash chains go through the same space where d_alias will be once it's not in-lookup anymore. Search is done under the same bitlock we use for modifications - with the primary hash we can rely on d_rehash() into the wrong chain being the worst that could happen, but here the pointers are buggered once it's removed from the chain. On the other hand, the chains are not going to be long and normally we'll end up adding to the chain anyway. That allows us to avoid bothering with ->d_lock when doing the comparisons - everything is stable until removed from chain. New helper: d_alloc_parallel(). Right now it allocates, verifies that no hashed and in-lookup matches exist and adds to in-lookup hash. Returns ERR_PTR() for error, hashed match (in the unlikely case it's been found) or new dentry. In-lookup matches trigger BUG() for now; that will change in the next commit when we introduce waiting for ongoing lookup to finish. Note that in-lookup matches won't be possible until we actually go for shared locking. lookup_slow() switched to use of d_alloc_parallel(). Again, these commits are separated only for making it easier to review. All this machinery will start doing something useful only when we go for shared locking; it's just that the combination is too large for my taste. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 14:42:04 +08:00
__hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
wake_up_all(dentry->d_wait);
dentry->d_wait = NULL;
parallel lookups machinery, part 3 We will need to be able to check if there is an in-lookup dentry with matching parent/name. Right now it's impossible, but as soon as start locking directories shared such beasts will appear. Add a secondary hash for locating those. Hash chains go through the same space where d_alias will be once it's not in-lookup anymore. Search is done under the same bitlock we use for modifications - with the primary hash we can rely on d_rehash() into the wrong chain being the worst that could happen, but here the pointers are buggered once it's removed from the chain. On the other hand, the chains are not going to be long and normally we'll end up adding to the chain anyway. That allows us to avoid bothering with ->d_lock when doing the comparisons - everything is stable until removed from chain. New helper: d_alloc_parallel(). Right now it allocates, verifies that no hashed and in-lookup matches exist and adds to in-lookup hash. Returns ERR_PTR() for error, hashed match (in the unlikely case it's been found) or new dentry. In-lookup matches trigger BUG() for now; that will change in the next commit when we introduce waiting for ongoing lookup to finish. Note that in-lookup matches won't be possible until we actually go for shared locking. lookup_slow() switched to use of d_alloc_parallel(). Again, these commits are separated only for making it easier to review. All this machinery will start doing something useful only when we go for shared locking; it's just that the combination is too large for my taste. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 14:42:04 +08:00
hlist_bl_unlock(b);
INIT_HLIST_NODE(&dentry->d_u.d_alias);
INIT_LIST_HEAD(&dentry->d_lru);
beginning of transition to parallel lookups - marking in-lookup dentries marked as such when (would be) parallel lookup is about to pass them to actual ->lookup(); unmarked when * __d_add() is about to make it hashed, positive or not. * __d_move() (from d_splice_alias(), directly or via __d_unalias()) puts a preexisting dentry in its place * in caller of ->lookup() if it has escaped all of the above. Bug (WARN_ON, actually) if it reaches the final dput() or d_instantiate() while still marked such. As the result, we are guaranteed that for as long as the flag is set, dentry will * remain negative unhashed with positive refcount * never have its ->d_alias looked at * never have its ->d_lru looked at * never have its ->d_parent and ->d_name changed Right now we have at most one such for any given parent directory. With parallel lookups that restriction will weaken to * only exist when parent is locked shared * at most one with given (parent,name) pair (comparison of names is according to ->d_compare()) * only exist when there's no hashed dentry with the same (parent,name) Transition will take the next several commits; unfortunately, we'll only be able to switch to rwsem at the end of this series. The reason for not making it a single patch is to simplify review. New primitives: d_in_lookup() (a predicate checking if dentry is in the in-lookup state) and d_lookup_done() (tells the system that we are done with lookup and if it's still marked as in-lookup, it should cease to be such). Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 07:52:13 +08:00
}
EXPORT_SYMBOL(__d_lookup_done);
/* inode->i_lock held if inode is non-NULL */
static inline void __d_add(struct dentry *dentry, struct inode *inode)
{
parallel lookups machinery, part 2 We'll need to verify that there's neither a hashed nor in-lookup dentry with desired parent/name before adding to in-lookup set. One possible solution would be to hold the parent's ->d_lock through both checks, but while the in-lookup set is relatively small at any time, dcache is not. And holding the parent's ->d_lock through something like __d_lookup_rcu() would suck too badly. So we leave the parent's ->d_lock alone, which means that we watch out for the following scenario: * we verify that there's no hashed match * existing in-lookup match gets hashed by another process * we verify that there's no in-lookup matches and decide that everything's fine. Solution: per-directory kinda-sorta seqlock, bumped around the times we hash something that used to be in-lookup or move (and hash) something in place of in-lookup. Then the above would turn into * read the counter * do dcache lookup * if no matches found, check for in-lookup matches * if there had been none of those either, check if the counter has changed; repeat if it has. The "kinda-sorta" part is due to the fact that we don't have much spare space in inode. There is a spare word (shared with i_bdev/i_cdev/i_pipe), so the counter part is not a problem, but spinlock is a different story. We could use the parent's ->d_lock, and it would be less painful in terms of contention, for __d_add() it would be rather inconvenient to grab; we could do that (using lock_parent()), but... Fortunately, we can get serialization on the counter itself, and it might be a good idea in general; we can use cmpxchg() in a loop to get from even to odd and smp_store_release() from odd to even. This commit adds the counter and updating logics; the readers will be added in the next commit. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 12:58:55 +08:00
struct inode *dir = NULL;
unsigned n;
spin_lock(&dentry->d_lock);
parallel lookups machinery, part 2 We'll need to verify that there's neither a hashed nor in-lookup dentry with desired parent/name before adding to in-lookup set. One possible solution would be to hold the parent's ->d_lock through both checks, but while the in-lookup set is relatively small at any time, dcache is not. And holding the parent's ->d_lock through something like __d_lookup_rcu() would suck too badly. So we leave the parent's ->d_lock alone, which means that we watch out for the following scenario: * we verify that there's no hashed match * existing in-lookup match gets hashed by another process * we verify that there's no in-lookup matches and decide that everything's fine. Solution: per-directory kinda-sorta seqlock, bumped around the times we hash something that used to be in-lookup or move (and hash) something in place of in-lookup. Then the above would turn into * read the counter * do dcache lookup * if no matches found, check for in-lookup matches * if there had been none of those either, check if the counter has changed; repeat if it has. The "kinda-sorta" part is due to the fact that we don't have much spare space in inode. There is a spare word (shared with i_bdev/i_cdev/i_pipe), so the counter part is not a problem, but spinlock is a different story. We could use the parent's ->d_lock, and it would be less painful in terms of contention, for __d_add() it would be rather inconvenient to grab; we could do that (using lock_parent()), but... Fortunately, we can get serialization on the counter itself, and it might be a good idea in general; we can use cmpxchg() in a loop to get from even to odd and smp_store_release() from odd to even. This commit adds the counter and updating logics; the readers will be added in the next commit. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 12:58:55 +08:00
if (unlikely(d_in_lookup(dentry))) {
dir = dentry->d_parent->d_inode;
n = start_dir_add(dir);
beginning of transition to parallel lookups - marking in-lookup dentries marked as such when (would be) parallel lookup is about to pass them to actual ->lookup(); unmarked when * __d_add() is about to make it hashed, positive or not. * __d_move() (from d_splice_alias(), directly or via __d_unalias()) puts a preexisting dentry in its place * in caller of ->lookup() if it has escaped all of the above. Bug (WARN_ON, actually) if it reaches the final dput() or d_instantiate() while still marked such. As the result, we are guaranteed that for as long as the flag is set, dentry will * remain negative unhashed with positive refcount * never have its ->d_alias looked at * never have its ->d_lru looked at * never have its ->d_parent and ->d_name changed Right now we have at most one such for any given parent directory. With parallel lookups that restriction will weaken to * only exist when parent is locked shared * at most one with given (parent,name) pair (comparison of names is according to ->d_compare()) * only exist when there's no hashed dentry with the same (parent,name) Transition will take the next several commits; unfortunately, we'll only be able to switch to rwsem at the end of this series. The reason for not making it a single patch is to simplify review. New primitives: d_in_lookup() (a predicate checking if dentry is in the in-lookup state) and d_lookup_done() (tells the system that we are done with lookup and if it's still marked as in-lookup, it should cease to be such). Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 07:52:13 +08:00
__d_lookup_done(dentry);
parallel lookups machinery, part 2 We'll need to verify that there's neither a hashed nor in-lookup dentry with desired parent/name before adding to in-lookup set. One possible solution would be to hold the parent's ->d_lock through both checks, but while the in-lookup set is relatively small at any time, dcache is not. And holding the parent's ->d_lock through something like __d_lookup_rcu() would suck too badly. So we leave the parent's ->d_lock alone, which means that we watch out for the following scenario: * we verify that there's no hashed match * existing in-lookup match gets hashed by another process * we verify that there's no in-lookup matches and decide that everything's fine. Solution: per-directory kinda-sorta seqlock, bumped around the times we hash something that used to be in-lookup or move (and hash) something in place of in-lookup. Then the above would turn into * read the counter * do dcache lookup * if no matches found, check for in-lookup matches * if there had been none of those either, check if the counter has changed; repeat if it has. The "kinda-sorta" part is due to the fact that we don't have much spare space in inode. There is a spare word (shared with i_bdev/i_cdev/i_pipe), so the counter part is not a problem, but spinlock is a different story. We could use the parent's ->d_lock, and it would be less painful in terms of contention, for __d_add() it would be rather inconvenient to grab; we could do that (using lock_parent()), but... Fortunately, we can get serialization on the counter itself, and it might be a good idea in general; we can use cmpxchg() in a loop to get from even to odd and smp_store_release() from odd to even. This commit adds the counter and updating logics; the readers will be added in the next commit. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 12:58:55 +08:00
}
if (inode) {
unsigned add_flags = d_flags_for_inode(inode);
hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
raw_write_seqcount_begin(&dentry->d_seq);
__d_set_inode_and_type(dentry, inode, add_flags);
raw_write_seqcount_end(&dentry->d_seq);
__fsnotify_d_instantiate(dentry);
}
_d_rehash(dentry);
parallel lookups machinery, part 2 We'll need to verify that there's neither a hashed nor in-lookup dentry with desired parent/name before adding to in-lookup set. One possible solution would be to hold the parent's ->d_lock through both checks, but while the in-lookup set is relatively small at any time, dcache is not. And holding the parent's ->d_lock through something like __d_lookup_rcu() would suck too badly. So we leave the parent's ->d_lock alone, which means that we watch out for the following scenario: * we verify that there's no hashed match * existing in-lookup match gets hashed by another process * we verify that there's no in-lookup matches and decide that everything's fine. Solution: per-directory kinda-sorta seqlock, bumped around the times we hash something that used to be in-lookup or move (and hash) something in place of in-lookup. Then the above would turn into * read the counter * do dcache lookup * if no matches found, check for in-lookup matches * if there had been none of those either, check if the counter has changed; repeat if it has. The "kinda-sorta" part is due to the fact that we don't have much spare space in inode. There is a spare word (shared with i_bdev/i_cdev/i_pipe), so the counter part is not a problem, but spinlock is a different story. We could use the parent's ->d_lock, and it would be less painful in terms of contention, for __d_add() it would be rather inconvenient to grab; we could do that (using lock_parent()), but... Fortunately, we can get serialization on the counter itself, and it might be a good idea in general; we can use cmpxchg() in a loop to get from even to odd and smp_store_release() from odd to even. This commit adds the counter and updating logics; the readers will be added in the next commit. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 12:58:55 +08:00
if (dir)
end_dir_add(dir, n);
spin_unlock(&dentry->d_lock);
if (inode)
spin_unlock(&inode->i_lock);
}
/**
* d_add - add dentry to hash queues
* @entry: dentry to add
* @inode: The inode to attach to this dentry
*
* This adds the entry to the hash queues and initializes @inode.
* The entry was actually filled in earlier during d_alloc().
*/
void d_add(struct dentry *entry, struct inode *inode)
{
if (inode) {
security_d_instantiate(entry, inode);
spin_lock(&inode->i_lock);
}
__d_add(entry, inode);
}
EXPORT_SYMBOL(d_add);
/**
* d_exact_alias - find and hash an exact unhashed alias
* @entry: dentry to add
* @inode: The inode to go with this dentry
*
* If an unhashed dentry with the same name/parent and desired
* inode already exists, hash and return it. Otherwise, return
* NULL.
*
* Parent directory should be locked.
*/
struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
{
struct dentry *alias;
int len = entry->d_name.len;
const char *name = entry->d_name.name;
unsigned int hash = entry->d_name.hash;
spin_lock(&inode->i_lock);
hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
/*
* Don't need alias->d_lock here, because aliases with
* d_parent == entry->d_parent are not subject to name or
* parent changes, because the parent inode i_mutex is held.
*/
if (alias->d_name.hash != hash)
continue;
if (alias->d_parent != entry->d_parent)
continue;
if (alias->d_name.len != len)
continue;
if (dentry_cmp(alias, name, len))
continue;
spin_lock(&alias->d_lock);
if (!d_unhashed(alias)) {
spin_unlock(&alias->d_lock);
alias = NULL;
} else {
__dget_dlock(alias);
_d_rehash(alias);
spin_unlock(&alias->d_lock);
}
spin_unlock(&inode->i_lock);
return alias;
}
spin_unlock(&inode->i_lock);
return NULL;
}
EXPORT_SYMBOL(d_exact_alias);
/**
* dentry_update_name_case - update case insensitive dentry with a new name
* @dentry: dentry to be updated
* @name: new name
*
* Update a case insensitive dentry with new case of name.
*
* dentry must have been returned by d_lookup with name @name. Old and new
* name lengths must match (ie. no d_compare which allows mismatched name
* lengths).
*
* Parent inode i_mutex must be held over d_lookup and into this call (to
* keep renames and concurrent inserts, and readdir(2) away).
*/
void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
{
BUG_ON(!inode_is_locked(dentry->d_parent->d_inode));
BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
spin_lock(&dentry->d_lock);
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
write_seqcount_begin(&dentry->d_seq);
memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
write_seqcount_end(&dentry->d_seq);
spin_unlock(&dentry->d_lock);
}
EXPORT_SYMBOL(dentry_update_name_case);
Allow sharing external names after __d_move() * external dentry names get a small structure prepended to them (struct external_name). * it contains an atomic refcount, matching the number of struct dentry instances that have ->d_name.name pointing to that external name. The first thing free_dentry() does is decrementing refcount of external name, so the instances that are between the call of free_dentry() and RCU-delayed actual freeing do not contribute. * __d_move(x, y, false) makes the name of x equal to the name of y, external or not. If y has an external name, extra reference is grabbed and put into x->d_name.name. If x used to have an external name, the reference to the old name is dropped and, should it reach zero, freeing is scheduled via kfree_rcu(). * free_dentry() in dentry with external name decrements the refcount of that name and, should it reach zero, does RCU-delayed call that will free both the dentry and external name. Otherwise it does what it used to do, except that __d_free() doesn't even look at ->d_name.name; it simply frees the dentry. All non-RCU accesses to dentry external name are safe wrt freeing since they all should happen before free_dentry() is called. RCU accesses might run into a dentry seen by free_dentry() or into an old name that got already dropped by __d_move(); however, in both cases dentry must have been alive and refer to that name at some point after we'd done rcu_read_lock(), which means that any freeing must be still pending. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2014-09-30 02:54:27 +08:00
static void swap_names(struct dentry *dentry, struct dentry *target)
{
Allow sharing external names after __d_move() * external dentry names get a small structure prepended to them (struct external_name). * it contains an atomic refcount, matching the number of struct dentry instances that have ->d_name.name pointing to that external name. The first thing free_dentry() does is decrementing refcount of external name, so the instances that are between the call of free_dentry() and RCU-delayed actual freeing do not contribute. * __d_move(x, y, false) makes the name of x equal to the name of y, external or not. If y has an external name, extra reference is grabbed and put into x->d_name.name. If x used to have an external name, the reference to the old name is dropped and, should it reach zero, freeing is scheduled via kfree_rcu(). * free_dentry() in dentry with external name decrements the refcount of that name and, should it reach zero, does RCU-delayed call that will free both the dentry and external name. Otherwise it does what it used to do, except that __d_free() doesn't even look at ->d_name.name; it simply frees the dentry. All non-RCU accesses to dentry external name are safe wrt freeing since they all should happen before free_dentry() is called. RCU accesses might run into a dentry seen by free_dentry() or into an old name that got already dropped by __d_move(); however, in both cases dentry must have been alive and refer to that name at some point after we'd done rcu_read_lock(), which means that any freeing must be still pending. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2014-09-30 02:54:27 +08:00
if (unlikely(dname_external(target))) {
if (unlikely(dname_external(dentry))) {
/*
* Both external: swap the pointers
*/
swap(target->d_name.name, dentry->d_name.name);
} else {
/*
* dentry:internal, target:external. Steal target's
* storage and make target internal.
*/
memcpy(target->d_iname, dentry->d_name.name,
dentry->d_name.len + 1);
dentry->d_name.name = target->d_name.name;
target->d_name.name = target->d_iname;
}
} else {
Allow sharing external names after __d_move() * external dentry names get a small structure prepended to them (struct external_name). * it contains an atomic refcount, matching the number of struct dentry instances that have ->d_name.name pointing to that external name. The first thing free_dentry() does is decrementing refcount of external name, so the instances that are between the call of free_dentry() and RCU-delayed actual freeing do not contribute. * __d_move(x, y, false) makes the name of x equal to the name of y, external or not. If y has an external name, extra reference is grabbed and put into x->d_name.name. If x used to have an external name, the reference to the old name is dropped and, should it reach zero, freeing is scheduled via kfree_rcu(). * free_dentry() in dentry with external name decrements the refcount of that name and, should it reach zero, does RCU-delayed call that will free both the dentry and external name. Otherwise it does what it used to do, except that __d_free() doesn't even look at ->d_name.name; it simply frees the dentry. All non-RCU accesses to dentry external name are safe wrt freeing since they all should happen before free_dentry() is called. RCU accesses might run into a dentry seen by free_dentry() or into an old name that got already dropped by __d_move(); however, in both cases dentry must have been alive and refer to that name at some point after we'd done rcu_read_lock(), which means that any freeing must be still pending. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2014-09-30 02:54:27 +08:00
if (unlikely(dname_external(dentry))) {
/*
* dentry:external, target:internal. Give dentry's
* storage to target and make dentry internal
*/
memcpy(dentry->d_iname, target->d_name.name,
target->d_name.len + 1);
target->d_name.name = dentry->d_name.name;
dentry->d_name.name = dentry->d_iname;
} else {
/*
* Both are internal.
*/
unsigned int i;
BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
swap(((long *) &dentry->d_iname)[i],
((long *) &target->d_iname)[i]);
}
}
}
swap(dentry->d_name.hash_len, target->d_name.hash_len);
}
Allow sharing external names after __d_move() * external dentry names get a small structure prepended to them (struct external_name). * it contains an atomic refcount, matching the number of struct dentry instances that have ->d_name.name pointing to that external name. The first thing free_dentry() does is decrementing refcount of external name, so the instances that are between the call of free_dentry() and RCU-delayed actual freeing do not contribute. * __d_move(x, y, false) makes the name of x equal to the name of y, external or not. If y has an external name, extra reference is grabbed and put into x->d_name.name. If x used to have an external name, the reference to the old name is dropped and, should it reach zero, freeing is scheduled via kfree_rcu(). * free_dentry() in dentry with external name decrements the refcount of that name and, should it reach zero, does RCU-delayed call that will free both the dentry and external name. Otherwise it does what it used to do, except that __d_free() doesn't even look at ->d_name.name; it simply frees the dentry. All non-RCU accesses to dentry external name are safe wrt freeing since they all should happen before free_dentry() is called. RCU accesses might run into a dentry seen by free_dentry() or into an old name that got already dropped by __d_move(); however, in both cases dentry must have been alive and refer to that name at some point after we'd done rcu_read_lock(), which means that any freeing must be still pending. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2014-09-30 02:54:27 +08:00
static void copy_name(struct dentry *dentry, struct dentry *target)
{
struct external_name *old_name = NULL;
if (unlikely(dname_external(dentry)))
old_name = external_name(dentry);
if (unlikely(dname_external(target))) {
atomic_inc(&external_name(target)->u.count);
dentry->d_name = target->d_name;
} else {
memcpy(dentry->d_iname, target->d_name.name,
target->d_name.len + 1);
dentry->d_name.name = dentry->d_iname;
dentry->d_name.hash_len = target->d_name.hash_len;
}
if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
kfree_rcu(old_name, u.head);
}
static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
{
/*
* XXXX: do we really need to take target->d_lock?
*/
if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
spin_lock(&target->d_parent->d_lock);
else {
if (d_ancestor(dentry->d_parent, target->d_parent)) {
spin_lock(&dentry->d_parent->d_lock);
spin_lock_nested(&target->d_parent->d_lock,
DENTRY_D_LOCK_NESTED);
} else {
spin_lock(&target->d_parent->d_lock);
spin_lock_nested(&dentry->d_parent->d_lock,
DENTRY_D_LOCK_NESTED);
}
}
if (target < dentry) {
spin_lock_nested(&target->d_lock, 2);
spin_lock_nested(&dentry->d_lock, 3);
} else {
spin_lock_nested(&dentry->d_lock, 2);
spin_lock_nested(&target->d_lock, 3);
}
}
static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
{
if (target->d_parent != dentry->d_parent)
spin_unlock(&dentry->d_parent->d_lock);
if (target->d_parent != target)
spin_unlock(&target->d_parent->d_lock);
spin_unlock(&target->d_lock);
spin_unlock(&dentry->d_lock);
}
/*
* When switching names, the actual string doesn't strictly have to
* be preserved in the target - because we're dropping the target
* anyway. As such, we can just do a simple memcpy() to copy over
vfs: Don't exchange "short" filenames unconditionally. Only exchange source and destination filenames if flags contain RENAME_EXCHANGE. In case if executable file was running and replaced by other file /proc/PID/exe should still show correct file name, not the old name of the file by which it was replaced. The scenario when this bug manifests itself was like this: * ALT Linux uses rpm and start-stop-daemon; * during a package upgrade rpm creates a temporary file for an executable to rename it upon successful unpacking; * start-stop-daemon is run subsequently and it obtains the (nonexistant) temporary filename via /proc/PID/exe thus failing to identify the running process. Note that "long" filenames (> DNAiME_INLINE_LEN) are still exchanged without RENAME_EXCHANGE and this behaviour exists long enough (should be fixed too apparently). So this patch is just an interim workaround that restores behavior for "short" names as it was before changes introduced by commit da1ce0670c14 ("vfs: add cross-rename"). See https://lkml.org/lkml/2014/9/7/6 for details. AV: the comments about being more careful with ->d_name.hash than with ->d_name.name are from back in 2.3.40s; they became obsolete by 2.3.60s, when we started to unhash the target instead of swapping hash chain positions followed by d_delete() as we used to do when dcache was first introduced. Acked-by: Miklos Szeredi <mszeredi@suse.cz> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: linux-fsdevel@vger.kernel.org Cc: stable@vger.kernel.org Fixes: da1ce0670c14 "vfs: add cross-rename" Signed-off-by: Mikhail Efremov <sem@altlinux.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2014-09-25 02:14:33 +08:00
* the new name before we switch, unless we are going to rehash
* it. Note that if we *do* unhash the target, we are not allowed
* to rehash it without giving it a new name/hash key - whether
* we swap or overwrite the names here, resulting name won't match
* the reality in filesystem; it's only there for d_path() purposes.
* Note that all of this is happening under rename_lock, so the
* any hash lookup seeing it in the middle of manipulations will
* be discarded anyway. So we do not care what happens to the hash
* key in that case.
*/
/*
* __d_move - move a dentry
* @dentry: entry to move
* @target: new dentry
* @exchange: exchange the two dentries
*
* Update the dcache to reflect the move of a file name. Negative
* dcache entries should not be moved in this way. Caller must hold
* rename_lock, the i_mutex of the source and target directories,
* and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
*/
static void __d_move(struct dentry *dentry, struct dentry *target,
bool exchange)
{
parallel lookups machinery, part 2 We'll need to verify that there's neither a hashed nor in-lookup dentry with desired parent/name before adding to in-lookup set. One possible solution would be to hold the parent's ->d_lock through both checks, but while the in-lookup set is relatively small at any time, dcache is not. And holding the parent's ->d_lock through something like __d_lookup_rcu() would suck too badly. So we leave the parent's ->d_lock alone, which means that we watch out for the following scenario: * we verify that there's no hashed match * existing in-lookup match gets hashed by another process * we verify that there's no in-lookup matches and decide that everything's fine. Solution: per-directory kinda-sorta seqlock, bumped around the times we hash something that used to be in-lookup or move (and hash) something in place of in-lookup. Then the above would turn into * read the counter * do dcache lookup * if no matches found, check for in-lookup matches * if there had been none of those either, check if the counter has changed; repeat if it has. The "kinda-sorta" part is due to the fact that we don't have much spare space in inode. There is a spare word (shared with i_bdev/i_cdev/i_pipe), so the counter part is not a problem, but spinlock is a different story. We could use the parent's ->d_lock, and it would be less painful in terms of contention, for __d_add() it would be rather inconvenient to grab; we could do that (using lock_parent()), but... Fortunately, we can get serialization on the counter itself, and it might be a good idea in general; we can use cmpxchg() in a loop to get from even to odd and smp_store_release() from odd to even. This commit adds the counter and updating logics; the readers will be added in the next commit. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 12:58:55 +08:00
struct inode *dir = NULL;
unsigned n;
if (!dentry->d_inode)
printk(KERN_WARNING "VFS: moving negative dcache entry\n");
BUG_ON(d_ancestor(dentry, target));
BUG_ON(d_ancestor(target, dentry));
dentry_lock_for_move(dentry, target);
parallel lookups machinery, part 2 We'll need to verify that there's neither a hashed nor in-lookup dentry with desired parent/name before adding to in-lookup set. One possible solution would be to hold the parent's ->d_lock through both checks, but while the in-lookup set is relatively small at any time, dcache is not. And holding the parent's ->d_lock through something like __d_lookup_rcu() would suck too badly. So we leave the parent's ->d_lock alone, which means that we watch out for the following scenario: * we verify that there's no hashed match * existing in-lookup match gets hashed by another process * we verify that there's no in-lookup matches and decide that everything's fine. Solution: per-directory kinda-sorta seqlock, bumped around the times we hash something that used to be in-lookup or move (and hash) something in place of in-lookup. Then the above would turn into * read the counter * do dcache lookup * if no matches found, check for in-lookup matches * if there had been none of those either, check if the counter has changed; repeat if it has. The "kinda-sorta" part is due to the fact that we don't have much spare space in inode. There is a spare word (shared with i_bdev/i_cdev/i_pipe), so the counter part is not a problem, but spinlock is a different story. We could use the parent's ->d_lock, and it would be less painful in terms of contention, for __d_add() it would be rather inconvenient to grab; we could do that (using lock_parent()), but... Fortunately, we can get serialization on the counter itself, and it might be a good idea in general; we can use cmpxchg() in a loop to get from even to odd and smp_store_release() from odd to even. This commit adds the counter and updating logics; the readers will be added in the next commit. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 12:58:55 +08:00
if (unlikely(d_in_lookup(target))) {
dir = target->d_parent->d_inode;
n = start_dir_add(dir);
beginning of transition to parallel lookups - marking in-lookup dentries marked as such when (would be) parallel lookup is about to pass them to actual ->lookup(); unmarked when * __d_add() is about to make it hashed, positive or not. * __d_move() (from d_splice_alias(), directly or via __d_unalias()) puts a preexisting dentry in its place * in caller of ->lookup() if it has escaped all of the above. Bug (WARN_ON, actually) if it reaches the final dput() or d_instantiate() while still marked such. As the result, we are guaranteed that for as long as the flag is set, dentry will * remain negative unhashed with positive refcount * never have its ->d_alias looked at * never have its ->d_lru looked at * never have its ->d_parent and ->d_name changed Right now we have at most one such for any given parent directory. With parallel lookups that restriction will weaken to * only exist when parent is locked shared * at most one with given (parent,name) pair (comparison of names is according to ->d_compare()) * only exist when there's no hashed dentry with the same (parent,name) Transition will take the next several commits; unfortunately, we'll only be able to switch to rwsem at the end of this series. The reason for not making it a single patch is to simplify review. New primitives: d_in_lookup() (a predicate checking if dentry is in the in-lookup state) and d_lookup_done() (tells the system that we are done with lookup and if it's still marked as in-lookup, it should cease to be such). Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 07:52:13 +08:00
__d_lookup_done(target);
parallel lookups machinery, part 2 We'll need to verify that there's neither a hashed nor in-lookup dentry with desired parent/name before adding to in-lookup set. One possible solution would be to hold the parent's ->d_lock through both checks, but while the in-lookup set is relatively small at any time, dcache is not. And holding the parent's ->d_lock through something like __d_lookup_rcu() would suck too badly. So we leave the parent's ->d_lock alone, which means that we watch out for the following scenario: * we verify that there's no hashed match * existing in-lookup match gets hashed by another process * we verify that there's no in-lookup matches and decide that everything's fine. Solution: per-directory kinda-sorta seqlock, bumped around the times we hash something that used to be in-lookup or move (and hash) something in place of in-lookup. Then the above would turn into * read the counter * do dcache lookup * if no matches found, check for in-lookup matches * if there had been none of those either, check if the counter has changed; repeat if it has. The "kinda-sorta" part is due to the fact that we don't have much spare space in inode. There is a spare word (shared with i_bdev/i_cdev/i_pipe), so the counter part is not a problem, but spinlock is a different story. We could use the parent's ->d_lock, and it would be less painful in terms of contention, for __d_add() it would be rather inconvenient to grab; we could do that (using lock_parent()), but... Fortunately, we can get serialization on the counter itself, and it might be a good idea in general; we can use cmpxchg() in a loop to get from even to odd and smp_store_release() from odd to even. This commit adds the counter and updating logics; the readers will be added in the next commit. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 12:58:55 +08:00
}
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
write_seqcount_begin(&dentry->d_seq);
write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
/* __d_drop does write_seqcount_barrier, but they're OK to nest. */
/*
* Move the dentry to the target hash queue. Don't bother checking
* for the same hash queue because of how unlikely it is.
*/
__d_drop(dentry);
__d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
/*
* Unhash the target (d_delete() is not usable here). If exchanging
* the two dentries, then rehash onto the other's hash queue.
*/
__d_drop(target);
if (exchange) {
__d_rehash(target,
d_hash(dentry->d_parent, dentry->d_name.hash));
}
/* Switch the names.. */
Allow sharing external names after __d_move() * external dentry names get a small structure prepended to them (struct external_name). * it contains an atomic refcount, matching the number of struct dentry instances that have ->d_name.name pointing to that external name. The first thing free_dentry() does is decrementing refcount of external name, so the instances that are between the call of free_dentry() and RCU-delayed actual freeing do not contribute. * __d_move(x, y, false) makes the name of x equal to the name of y, external or not. If y has an external name, extra reference is grabbed and put into x->d_name.name. If x used to have an external name, the reference to the old name is dropped and, should it reach zero, freeing is scheduled via kfree_rcu(). * free_dentry() in dentry with external name decrements the refcount of that name and, should it reach zero, does RCU-delayed call that will free both the dentry and external name. Otherwise it does what it used to do, except that __d_free() doesn't even look at ->d_name.name; it simply frees the dentry. All non-RCU accesses to dentry external name are safe wrt freeing since they all should happen before free_dentry() is called. RCU accesses might run into a dentry seen by free_dentry() or into an old name that got already dropped by __d_move(); however, in both cases dentry must have been alive and refer to that name at some point after we'd done rcu_read_lock(), which means that any freeing must be still pending. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2014-09-30 02:54:27 +08:00
if (exchange)
swap_names(dentry, target);
else
copy_name(dentry, target);
/* ... and switch them in the tree */
if (IS_ROOT(dentry)) {
/* splicing a tree */
dentry->d_parent = target->d_parent;
target->d_parent = target;
list_del_init(&target->d_child);
list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
} else {
/* swapping two dentries */
swap(dentry->d_parent, target->d_parent);
list_move(&target->d_child, &target->d_parent->d_subdirs);
list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
if (exchange)
fsnotify_d_move(target);
fsnotify_d_move(dentry);
}
fs: rcu-walk for path lookup Perform common cases of path lookups without any stores or locking in the ancestor dentry elements. This is called rcu-walk, as opposed to the current algorithm which is a refcount based walk, or ref-walk. This results in far fewer atomic operations on every path element, significantly improving path lookup performance. It also avoids cacheline bouncing on common dentries, significantly improving scalability. The overall design is like this: * LOOKUP_RCU is set in nd->flags, which distinguishes rcu-walk from ref-walk. * Take the RCU lock for the entire path walk, starting with the acquiring of the starting path (eg. root/cwd/fd-path). So now dentry refcounts are not required for dentry persistence. * synchronize_rcu is called when unregistering a filesystem, so we can access d_ops and i_ops during rcu-walk. * Similarly take the vfsmount lock for the entire path walk. So now mnt refcounts are not required for persistence. Also we are free to perform mount lookups, and to assume dentry mount points and mount roots are stable up and down the path. * Have a per-dentry seqlock to protect the dentry name, parent, and inode, so we can load this tuple atomically, and also check whether any of its members have changed. * Dentry lookups (based on parent, candidate string tuple) recheck the parent sequence after the child is found in case anything changed in the parent during the path walk. * inode is also RCU protected so we can load d_inode and use the inode for limited things. * i_mode, i_uid, i_gid can be tested for exec permissions during path walk. * i_op can be loaded. When we reach the destination dentry, we lock it, recheck lookup sequence, and increment its refcount and mountpoint refcount. RCU and vfsmount locks are dropped. This is termed "dropping rcu-walk". If the dentry refcount does not match, we can not drop rcu-walk gracefully at the current point in the lokup, so instead return -ECHILD (for want of a better errno). This signals the path walking code to re-do the entire lookup with a ref-walk. Aside from the final dentry, there are other situations that may be encounted where we cannot continue rcu-walk. In that case, we drop rcu-walk (ie. take a reference on the last good dentry) and continue with a ref-walk. Again, if we can drop rcu-walk gracefully, we return -ECHILD and do the whole lookup using ref-walk. But it is very important that we can continue with ref-walk for most cases, particularly to avoid the overhead of double lookups, and to gain the scalability advantages on common path elements (like cwd and root). The cases where rcu-walk cannot continue are: * NULL dentry (ie. any uncached path element) * parent with d_inode->i_op->permission or ACLs * dentries with d_revalidate * Following links In future patches, permission checks and d_revalidate become rcu-walk aware. It may be possible eventually to make following links rcu-walk aware. Uncached path elements will always require dropping to ref-walk mode, at the very least because i_mutex needs to be grabbed, and objects allocated. Signed-off-by: Nick Piggin <npiggin@kernel.dk>
2011-01-07 14:49:52 +08:00
write_seqcount_end(&target->d_seq);
write_seqcount_end(&dentry->d_seq);
parallel lookups machinery, part 2 We'll need to verify that there's neither a hashed nor in-lookup dentry with desired parent/name before adding to in-lookup set. One possible solution would be to hold the parent's ->d_lock through both checks, but while the in-lookup set is relatively small at any time, dcache is not. And holding the parent's ->d_lock through something like __d_lookup_rcu() would suck too badly. So we leave the parent's ->d_lock alone, which means that we watch out for the following scenario: * we verify that there's no hashed match * existing in-lookup match gets hashed by another process * we verify that there's no in-lookup matches and decide that everything's fine. Solution: per-directory kinda-sorta seqlock, bumped around the times we hash something that used to be in-lookup or move (and hash) something in place of in-lookup. Then the above would turn into * read the counter * do dcache lookup * if no matches found, check for in-lookup matches * if there had been none of those either, check if the counter has changed; repeat if it has. The "kinda-sorta" part is due to the fact that we don't have much spare space in inode. There is a spare word (shared with i_bdev/i_cdev/i_pipe), so the counter part is not a problem, but spinlock is a different story. We could use the parent's ->d_lock, and it would be less painful in terms of contention, for __d_add() it would be rather inconvenient to grab; we could do that (using lock_parent()), but... Fortunately, we can get serialization on the counter itself, and it might be a good idea in general; we can use cmpxchg() in a loop to get from even to odd and smp_store_release() from odd to even. This commit adds the counter and updating logics; the readers will be added in the next commit. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-04-15 12:58:55 +08:00
if (dir)
end_dir_add(dir, n);
dentry_unlock_for_move(dentry, target);
}
/*
* d_move - move a dentry
* @dentry: entry to move
* @target: new dentry
*
* Update the dcache to reflect the move of a file name. Negative
* dcache entries should not be moved in this way. See the locking
* requirements for __d_move.
*/
void d_move(struct dentry *dentry, struct dentry *target)
{
write_seqlock(&rename_lock);
__d_move(dentry, target, false);
write_sequnlock(&rename_lock);
}
EXPORT_SYMBOL(d_move);
/*
* d_exchange - exchange two dentries
* @dentry1: first dentry
* @dentry2: second dentry
*/
void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
{
write_seqlock(&rename_lock);
WARN_ON(!dentry1->d_inode);
WARN_ON(!dentry2->d_inode);
WARN_ON(IS_ROOT(dentry1));
WARN_ON(IS_ROOT(dentry2));
__d_move(dentry1, dentry2, true);
write_sequnlock(&rename_lock);
}
/**
* d_ancestor - search for an ancestor
* @p1: ancestor dentry
* @p2: child dentry
*
* Returns the ancestor dentry of p2 which is a child of p1, if p1 is
* an ancestor of p2, else NULL.
*/
struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
{
struct dentry *p;
for (p = p2; !IS_ROOT(p); p = p->d_parent) {
if (p->d_parent == p1)
return p;
}
return NULL;
}
/*
* This helper attempts to cope with remotely renamed directories
*
* It assumes that the caller is already holding
* dentry->d_parent->d_inode->i_mutex, and rename_lock
*
* Note: If ever the locking in lock_rename() changes, then please
* remember to update this too...
*/
static int __d_unalias(struct inode *inode,
struct dentry *dentry, struct dentry *alias)
{
struct mutex *m1 = NULL;
struct rw_semaphore *m2 = NULL;
dcache: return -ESTALE not -EBUSY on distributed fs race On a distributed filesystem it's possible for lookup to discover that a directory it just found is already cached elsewhere in the directory heirarchy. The dcache won't let us keep the directory in both places, so we have to move the dentry to the new location from the place we previously had it cached. If the parent has changed, then this requires all the same locks as we'd need to do a cross-directory rename. But we're already in lookup holding one parent's i_mutex, so it's too late to acquire those locks in the right order. The (unreliable) solution in __d_unalias is to trylock() the required locks and return -EBUSY if it fails. I see no particular reason for returning -EBUSY, and -ESTALE is already the result of some other lookup races on NFS. I think -ESTALE is the more helpful error return. It also allows us to take advantage of the logic Jeff Layton added in c6a9428401c0 "vfs: fix renameat to retry on ESTALE errors" and ancestors, which hopefully resolves some of these errors before they're returned to userspace. I can reproduce these cases using NFS with: ssh root@$client ' mount -olookupcache=pos '$server':'$export' /mnt/ mkdir /mnt/TO mkdir /mnt/DIR touch /mnt/DIR/test.txt while true; do strace -e open cat /mnt/DIR/test.txt 2>&1 | grep EBUSY done ' ssh root@$server ' while true; do mv $export/DIR $export/TO/DIR mv $export/TO/DIR $export/DIR done ' It also helps to add some other concurrent use of the directory on the client (e.g., "ls /mnt/TO"). And you can replace the server-side mv's by client-side mv's that are repeatedly killed. (If the client is interrupted while waiting for the RENAME response then it's left with a dentry that has to go under one parent or the other, but it doesn't yet know which.) Acked-by: Jeff Layton <jlayton@primarydata.com> Signed-off-by: J. Bruce Fields <bfields@redhat.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2015-02-10 23:55:53 +08:00
int ret = -ESTALE;
/* If alias and dentry share a parent, then no extra locks required */
if (alias->d_parent == dentry->d_parent)
goto out_unalias;
/* See lock_rename() */
if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
goto out_err;
m1 = &dentry->d_sb->s_vfs_rename_mutex;
if (!inode_trylock_shared(alias->d_parent->d_inode))
goto out_err;
m2 = &alias->d_parent->d_inode->i_rwsem;
out_unalias:
vfs: Lazily remove mounts on unlinked files and directories. With the introduction of mount namespaces and bind mounts it became possible to access files and directories that on some paths are mount points but are not mount points on other paths. It is very confusing when rm -rf somedir returns -EBUSY simply because somedir is mounted somewhere else. With the addition of user namespaces allowing unprivileged mounts this condition has gone from annoying to allowing a DOS attack on other users in the system. The possibility for mischief is removed by updating the vfs to support rename, unlink and rmdir on a dentry that is a mountpoint and by lazily unmounting mountpoints on deleted dentries. In particular this change allows rename, unlink and rmdir system calls on a dentry without a mountpoint in the current mount namespace to succeed, and it allows rename, unlink, and rmdir performed on a distributed filesystem to update the vfs cache even if when there is a mount in some namespace on the original dentry. There are two common patterns of maintaining mounts: Mounts on trusted paths with the parent directory of the mount point and all ancestory directories up to / owned by root and modifiable only by root (i.e. /media/xxx, /dev, /dev/pts, /proc, /sys, /sys/fs/cgroup/{cpu, cpuacct, ...}, /usr, /usr/local). Mounts on unprivileged directories maintained by fusermount. In the case of mounts in trusted directories owned by root and modifiable only by root the current parent directory permissions are sufficient to ensure a mount point on a trusted path is not removed or renamed by anyone other than root, even if there is a context where the there are no mount points to prevent this. In the case of mounts in directories owned by less privileged users races with users modifying the path of a mount point are already a danger. fusermount already uses a combination of chdir, /proc/<pid>/fd/NNN, and UMOUNT_NOFOLLOW to prevent these races. The removable of global rename, unlink, and rmdir protection really adds nothing new to consider only a widening of the attack window, and fusermount is already safe against unprivileged users modifying the directory simultaneously. In principle for perfect userspace programs returning -EBUSY for unlink, rmdir, and rename of dentires that have mounts in the local namespace is actually unnecessary. Unfortunately not all userspace programs are perfect so retaining -EBUSY for unlink, rmdir and rename of dentries that have mounts in the current mount namespace plays an important role of maintaining consistency with historical behavior and making imperfect userspace applications hard to exploit. v2: Remove spurious old_dentry. v3: Optimized shrink_submounts_and_drop Removed unsued afs label v4: Simplified the changes to check_submounts_and_drop Do not rename check_submounts_and_drop shrink_submounts_and_drop Document what why we need atomicity in check_submounts_and_drop Rely on the parent inode mutex to make d_revalidate and d_invalidate an atomic unit. v5: Refcount the mountpoint to detach in case of simultaneous renames. Reviewed-by: Miklos Szeredi <miklos@szeredi.hu> Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-10-02 09:33:48 +08:00
__d_move(alias, dentry, false);
ret = 0;
out_err:
if (m2)
up_read(m2);
if (m1)
mutex_unlock(m1);
return ret;
}
/**
* d_splice_alias - splice a disconnected dentry into the tree if one exists
* @inode: the inode which may have a disconnected dentry
* @dentry: a negative dentry which we want to point to the inode.
*
* If inode is a directory and has an IS_ROOT alias, then d_move that in
* place of the given dentry and return it, else simply d_add the inode
* to the dentry and return NULL.
*
* If a non-IS_ROOT directory is found, the filesystem is corrupt, and
* we should error out: directories can't have multiple aliases.
*
* This is needed in the lookup routine of any filesystem that is exportable
* (via knfsd) so that we can build dcache paths to directories effectively.
*
* If a dentry was found and moved, then it is returned. Otherwise NULL
* is returned. This matches the expected return value of ->lookup.
*
* Cluster filesystems may call this function with a negative, hashed dentry.
* In that case, we know that the inode will be a regular file, and also this
* will only occur during atomic_open. So we need to check for the dentry
* being already hashed only in the final case.
*/
struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
{
if (IS_ERR(inode))
return ERR_CAST(inode);
BUG_ON(!d_unhashed(dentry));
if (!inode)
goto out;
security_d_instantiate(dentry, inode);
spin_lock(&inode->i_lock);
if (S_ISDIR(inode->i_mode)) {
struct dentry *new = __d_find_any_alias(inode);
if (unlikely(new)) {
/* The reference to new ensures it remains an alias */
spin_unlock(&inode->i_lock);
write_seqlock(&rename_lock);
if (unlikely(d_ancestor(new, dentry))) {
write_sequnlock(&rename_lock);
dput(new);
new = ERR_PTR(-ELOOP);
pr_warn_ratelimited(
"VFS: Lookup of '%s' in %s %s"
" would have caused loop\n",
dentry->d_name.name,
inode->i_sb->s_type->name,
inode->i_sb->s_id);
} else if (!IS_ROOT(new)) {
int err = __d_unalias(inode, dentry, new);
write_sequnlock(&rename_lock);
if (err) {
dput(new);
new = ERR_PTR(err);
}
} else {
__d_move(new, dentry, false);
write_sequnlock(&rename_lock);
}
iput(inode);
return new;
}
}
out:
__d_add(dentry, inode);
return NULL;
}
EXPORT_SYMBOL(d_splice_alias);
static int prepend(char **buffer, int *buflen, const char *str, int namelen)
{
*buflen -= namelen;
if (*buflen < 0)
return -ENAMETOOLONG;
*buffer -= namelen;
memcpy(*buffer, str, namelen);
return 0;
}
dcache: Translating dentry into pathname without taking rename_lock When running the AIM7's short workload, Linus' lockref patch eliminated most of the spinlock contention. However, there were still some left: 8.46% reaim [kernel.kallsyms] [k] _raw_spin_lock |--42.21%-- d_path | proc_pid_readlink | SyS_readlinkat | SyS_readlink | system_call | __GI___readlink | |--40.97%-- sys_getcwd | system_call | __getcwd The big one here is the rename_lock (seqlock) contention in d_path() and the getcwd system call. This patch will eliminate the need to take the rename_lock while translating dentries into the full pathnames. The need to take the rename_lock is to make sure that no rename operation can be ongoing while the translation is in progress. However, only one thread can take the rename_lock thus blocking all the other threads that need it even though the translation process won't make any change to the dentries. This patch will replace the writer's write_seqlock/write_sequnlock sequence of the rename_lock of the callers of the prepend_path() and __dentry_path() functions with the reader's read_seqbegin/read_seqretry sequence within these 2 functions. As a result, the code will have to retry if one or more rename operations had been performed. In addition, RCU read lock will be taken during the translation process to make sure that no dentries will go away. To prevent live-lock from happening, the code will switch back to take the rename_lock if read_seqretry() fails for three times. To further reduce spinlock contention, this patch does not take the dentry's d_lock when copying the filename from the dentries. Instead, it treats the name pointer and length as unreliable and just copy the string byte-by-byte over until it hits a null byte or the end of string as specified by the length. This should avoid stepping into invalid memory address. The error cases are left to be handled by the sequence number check. The following code re-factoring are also made: 1. Move prepend('/') into prepend_name() to remove one conditional check. 2. Move the global root check in prepend_path() back to the top of the while loop. With this patch, the _raw_spin_lock will now account for only 1.2% of the total CPU cycles for the short workload. This patch also has the effect of reducing the effect of running perf on its profile since the perf command itself can be a heavy user of the d_path() function depending on the complexity of the workload. When taking the perf profile of the high-systime workload, the amount of spinlock contention contributed by running perf without this patch was about 16%. With this patch, the spinlock contention caused by the running of perf will go away and we will have a more accurate perf profile. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-10 00:18:13 +08:00
/**
* prepend_name - prepend a pathname in front of current buffer pointer
* @buffer: buffer pointer
* @buflen: allocated length of the buffer
* @name: name string and length qstr structure
dcache: Translating dentry into pathname without taking rename_lock When running the AIM7's short workload, Linus' lockref patch eliminated most of the spinlock contention. However, there were still some left: 8.46% reaim [kernel.kallsyms] [k] _raw_spin_lock |--42.21%-- d_path | proc_pid_readlink | SyS_readlinkat | SyS_readlink | system_call | __GI___readlink | |--40.97%-- sys_getcwd | system_call | __getcwd The big one here is the rename_lock (seqlock) contention in d_path() and the getcwd system call. This patch will eliminate the need to take the rename_lock while translating dentries into the full pathnames. The need to take the rename_lock is to make sure that no rename operation can be ongoing while the translation is in progress. However, only one thread can take the rename_lock thus blocking all the other threads that need it even though the translation process won't make any change to the dentries. This patch will replace the writer's write_seqlock/write_sequnlock sequence of the rename_lock of the callers of the prepend_path() and __dentry_path() functions with the reader's read_seqbegin/read_seqretry sequence within these 2 functions. As a result, the code will have to retry if one or more rename operations had been performed. In addition, RCU read lock will be taken during the translation process to make sure that no dentries will go away. To prevent live-lock from happening, the code will switch back to take the rename_lock if read_seqretry() fails for three times. To further reduce spinlock contention, this patch does not take the dentry's d_lock when copying the filename from the dentries. Instead, it treats the name pointer and length as unreliable and just copy the string byte-by-byte over until it hits a null byte or the end of string as specified by the length. This should avoid stepping into invalid memory address. The error cases are left to be handled by the sequence number check. The following code re-factoring are also made: 1. Move prepend('/') into prepend_name() to remove one conditional check. 2. Move the global root check in prepend_path() back to the top of the while loop. With this patch, the _raw_spin_lock will now account for only 1.2% of the total CPU cycles for the short workload. This patch also has the effect of reducing the effect of running perf on its profile since the perf command itself can be a heavy user of the d_path() function depending on the complexity of the workload. When taking the perf profile of the high-systime workload, the amount of spinlock contention contributed by running perf without this patch was about 16%. With this patch, the spinlock contention caused by the running of perf will go away and we will have a more accurate perf profile. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-10 00:18:13 +08:00
*
* With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
* make sure that either the old or the new name pointer and length are
* fetched. However, there may be mismatch between length and pointer.
* The length cannot be trusted, we need to copy it byte-by-byte until
* the length is reached or a null byte is found. It also prepends "/" at
* the beginning of the name. The sequence number check at the caller will
* retry it again when a d_move() does happen. So any garbage in the buffer
* due to mismatched pointer and length will be discarded.
*
* Data dependency barrier is needed to make sure that we see that terminating
* NUL. Alpha strikes again, film at 11...
dcache: Translating dentry into pathname without taking rename_lock When running the AIM7's short workload, Linus' lockref patch eliminated most of the spinlock contention. However, there were still some left: 8.46% reaim [kernel.kallsyms] [k] _raw_spin_lock |--42.21%-- d_path | proc_pid_readlink | SyS_readlinkat | SyS_readlink | system_call | __GI___readlink | |--40.97%-- sys_getcwd | system_call | __getcwd The big one here is the rename_lock (seqlock) contention in d_path() and the getcwd system call. This patch will eliminate the need to take the rename_lock while translating dentries into the full pathnames. The need to take the rename_lock is to make sure that no rename operation can be ongoing while the translation is in progress. However, only one thread can take the rename_lock thus blocking all the other threads that need it even though the translation process won't make any change to the dentries. This patch will replace the writer's write_seqlock/write_sequnlock sequence of the rename_lock of the callers of the prepend_path() and __dentry_path() functions with the reader's read_seqbegin/read_seqretry sequence within these 2 functions. As a result, the code will have to retry if one or more rename operations had been performed. In addition, RCU read lock will be taken during the translation process to make sure that no dentries will go away. To prevent live-lock from happening, the code will switch back to take the rename_lock if read_seqretry() fails for three times. To further reduce spinlock contention, this patch does not take the dentry's d_lock when copying the filename from the dentries. Instead, it treats the name pointer and length as unreliable and just copy the string byte-by-byte over until it hits a null byte or the end of string as specified by the length. This should avoid stepping into invalid memory address. The error cases are left to be handled by the sequence number check. The following code re-factoring are also made: 1. Move prepend('/') into prepend_name() to remove one conditional check. 2. Move the global root check in prepend_path() back to the top of the while loop. With this patch, the _raw_spin_lock will now account for only 1.2% of the total CPU cycles for the short workload. This patch also has the effect of reducing the effect of running perf on its profile since the perf command itself can be a heavy user of the d_path() function depending on the complexity of the workload. When taking the perf profile of the high-systime workload, the amount of spinlock contention contributed by running perf without this patch was about 16%. With this patch, the spinlock contention caused by the running of perf will go away and we will have a more accurate perf profile. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-10 00:18:13 +08:00
*/
static int prepend_name(char **buffer, int *buflen, struct qstr *name)
{
dcache: Translating dentry into pathname without taking rename_lock When running the AIM7's short workload, Linus' lockref patch eliminated most of the spinlock contention. However, there were still some left: 8.46% reaim [kernel.kallsyms] [k] _raw_spin_lock |--42.21%-- d_path | proc_pid_readlink | SyS_readlinkat | SyS_readlink | system_call | __GI___readlink | |--40.97%-- sys_getcwd | system_call | __getcwd The big one here is the rename_lock (seqlock) contention in d_path() and the getcwd system call. This patch will eliminate the need to take the rename_lock while translating dentries into the full pathnames. The need to take the rename_lock is to make sure that no rename operation can be ongoing while the translation is in progress. However, only one thread can take the rename_lock thus blocking all the other threads that need it even though the translation process won't make any change to the dentries. This patch will replace the writer's write_seqlock/write_sequnlock sequence of the rename_lock of the callers of the prepend_path() and __dentry_path() functions with the reader's read_seqbegin/read_seqretry sequence within these 2 functions. As a result, the code will have to retry if one or more rename operations had been performed. In addition, RCU read lock will be taken during the translation process to make sure that no dentries will go away. To prevent live-lock from happening, the code will switch back to take the rename_lock if read_seqretry() fails for three times. To further reduce spinlock contention, this patch does not take the dentry's d_lock when copying the filename from the dentries. Instead, it treats the name pointer and length as unreliable and just copy the string byte-by-byte over until it hits a null byte or the end of string as specified by the length. This should avoid stepping into invalid memory address. The error cases are left to be handled by the sequence number check. The following code re-factoring are also made: 1. Move prepend('/') into prepend_name() to remove one conditional check. 2. Move the global root check in prepend_path() back to the top of the while loop. With this patch, the _raw_spin_lock will now account for only 1.2% of the total CPU cycles for the short workload. This patch also has the effect of reducing the effect of running perf on its profile since the perf command itself can be a heavy user of the d_path() function depending on the complexity of the workload. When taking the perf profile of the high-systime workload, the amount of spinlock contention contributed by running perf without this patch was about 16%. With this patch, the spinlock contention caused by the running of perf will go away and we will have a more accurate perf profile. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-10 00:18:13 +08:00
const char *dname = ACCESS_ONCE(name->name);
u32 dlen = ACCESS_ONCE(name->len);
char *p;
smp_read_barrier_depends();
dcache: Translating dentry into pathname without taking rename_lock When running the AIM7's short workload, Linus' lockref patch eliminated most of the spinlock contention. However, there were still some left: 8.46% reaim [kernel.kallsyms] [k] _raw_spin_lock |--42.21%-- d_path | proc_pid_readlink | SyS_readlinkat | SyS_readlink | system_call | __GI___readlink | |--40.97%-- sys_getcwd | system_call | __getcwd The big one here is the rename_lock (seqlock) contention in d_path() and the getcwd system call. This patch will eliminate the need to take the rename_lock while translating dentries into the full pathnames. The need to take the rename_lock is to make sure that no rename operation can be ongoing while the translation is in progress. However, only one thread can take the rename_lock thus blocking all the other threads that need it even though the translation process won't make any change to the dentries. This patch will replace the writer's write_seqlock/write_sequnlock sequence of the rename_lock of the callers of the prepend_path() and __dentry_path() functions with the reader's read_seqbegin/read_seqretry sequence within these 2 functions. As a result, the code will have to retry if one or more rename operations had been performed. In addition, RCU read lock will be taken during the translation process to make sure that no dentries will go away. To prevent live-lock from happening, the code will switch back to take the rename_lock if read_seqretry() fails for three times. To further reduce spinlock contention, this patch does not take the dentry's d_lock when copying the filename from the dentries. Instead, it treats the name pointer and length as unreliable and just copy the string byte-by-byte over until it hits a null byte or the end of string as specified by the length. This should avoid stepping into invalid memory address. The error cases are left to be handled by the sequence number check. The following code re-factoring are also made: 1. Move prepend('/') into prepend_name() to remove one conditional check. 2. Move the global root check in prepend_path() back to the top of the while loop. With this patch, the _raw_spin_lock will now account for only 1.2% of the total CPU cycles for the short workload. This patch also has the effect of reducing the effect of running perf on its profile since the perf command itself can be a heavy user of the d_path() function depending on the complexity of the workload. When taking the perf profile of the high-systime workload, the amount of spinlock contention contributed by running perf without this patch was about 16%. With this patch, the spinlock contention caused by the running of perf will go away and we will have a more accurate perf profile. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-10 00:18:13 +08:00
*buflen -= dlen + 1;
if (*buflen < 0)
return -ENAMETOOLONG;
dcache: Translating dentry into pathname without taking rename_lock When running the AIM7's short workload, Linus' lockref patch eliminated most of the spinlock contention. However, there were still some left: 8.46% reaim [kernel.kallsyms] [k] _raw_spin_lock |--42.21%-- d_path | proc_pid_readlink | SyS_readlinkat | SyS_readlink | system_call | __GI___readlink | |--40.97%-- sys_getcwd | system_call | __getcwd The big one here is the rename_lock (seqlock) contention in d_path() and the getcwd system call. This patch will eliminate the need to take the rename_lock while translating dentries into the full pathnames. The need to take the rename_lock is to make sure that no rename operation can be ongoing while the translation is in progress. However, only one thread can take the rename_lock thus blocking all the other threads that need it even though the translation process won't make any change to the dentries. This patch will replace the writer's write_seqlock/write_sequnlock sequence of the rename_lock of the callers of the prepend_path() and __dentry_path() functions with the reader's read_seqbegin/read_seqretry sequence within these 2 functions. As a result, the code will have to retry if one or more rename operations had been performed. In addition, RCU read lock will be taken during the translation process to make sure that no dentries will go away. To prevent live-lock from happening, the code will switch back to take the rename_lock if read_seqretry() fails for three times. To further reduce spinlock contention, this patch does not take the dentry's d_lock when copying the filename from the dentries. Instead, it treats the name pointer and length as unreliable and just copy the string byte-by-byte over until it hits a null byte or the end of string as specified by the length. This should avoid stepping into invalid memory address. The error cases are left to be handled by the sequence number check. The following code re-factoring are also made: 1. Move prepend('/') into prepend_name() to remove one conditional check. 2. Move the global root check in prepend_path() back to the top of the while loop. With this patch, the _raw_spin_lock will now account for only 1.2% of the total CPU cycles for the short workload. This patch also has the effect of reducing the effect of running perf on its profile since the perf command itself can be a heavy user of the d_path() function depending on the complexity of the workload. When taking the perf profile of the high-systime workload, the amount of spinlock contention contributed by running perf without this patch was about 16%. With this patch, the spinlock contention caused by the running of perf will go away and we will have a more accurate perf profile. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-10 00:18:13 +08:00
p = *buffer -= dlen + 1;
*p++ = '/';
while (dlen--) {
char c = *dname++;
if (!c)
break;
*p++ = c;
}
return 0;
}
/**
* prepend_path - Prepend path string to a buffer
* @path: the dentry/vfsmount to report
fix apparmor dereferencing potentially freed dentry, sanitize __d_path() API __d_path() API is asking for trouble and in case of apparmor d_namespace_path() getting just that. The root cause is that when __d_path() misses the root it had been told to look for, it stores the location of the most remote ancestor in *root. Without grabbing references. Sure, at the moment of call it had been pinned down by what we have in *path. And if we raced with umount -l, we could have very well stopped at vfsmount/dentry that got freed as soon as prepend_path() dropped vfsmount_lock. It is safe to compare these pointers with pre-existing (and known to be still alive) vfsmount and dentry, as long as all we are asking is "is it the same address?". Dereferencing is not safe and apparmor ended up stepping into that. d_namespace_path() really wants to examine the place where we stopped, even if it's not connected to our namespace. As the result, it looked at ->d_sb->s_magic of a dentry that might've been already freed by that point. All other callers had been careful enough to avoid that, but it's really a bad interface - it invites that kind of trouble. The fix is fairly straightforward, even though it's bigger than I'd like: * prepend_path() root argument becomes const. * __d_path() is never called with NULL/NULL root. It was a kludge to start with. Instead, we have an explicit function - d_absolute_root(). Same as __d_path(), except that it doesn't get root passed and stops where it stops. apparmor and tomoyo are using it. * __d_path() returns NULL on path outside of root. The main caller is show_mountinfo() and that's precisely what we pass root for - to skip those outside chroot jail. Those who don't want that can (and do) use d_path(). * __d_path() root argument becomes const. Everyone agrees, I hope. * apparmor does *NOT* try to use __d_path() or any of its variants when it sees that path->mnt is an internal vfsmount. In that case it's definitely not mounted anywhere and dentry_path() is exactly what we want there. Handling of sysctl()-triggered weirdness is moved to that place. * if apparmor is asked to do pathname relative to chroot jail and __d_path() tells it we it's not in that jail, the sucker just calls d_absolute_path() instead. That's the other remaining caller of __d_path(), BTW. * seq_path_root() does _NOT_ return -ENAMETOOLONG (it's stupid anyway - the normal seq_file logics will take care of growing the buffer and redoing the call of ->show() just fine). However, if it gets path not reachable from root, it returns SEQ_SKIP. The only caller adjusted (i.e. stopped ignoring the return value as it used to do). Reviewed-by: John Johansen <john.johansen@canonical.com> ACKed-by: John Johansen <john.johansen@canonical.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Cc: stable@vger.kernel.org
2011-12-05 21:43:34 +08:00
* @root: root vfsmnt/dentry
* @buffer: pointer to the end of the buffer
* @buflen: pointer to buffer length
*
* The function will first try to write out the pathname without taking any
* lock other than the RCU read lock to make sure that dentries won't go away.
* It only checks the sequence number of the global rename_lock as any change
* in the dentry's d_seq will be preceded by changes in the rename_lock
* sequence number. If the sequence number had been changed, it will restart
* the whole pathname back-tracing sequence again by taking the rename_lock.
* In this case, there is no need to take the RCU read lock as the recursive
* parent pointer references will keep the dentry chain alive as long as no
* rename operation is performed.
*/
fix apparmor dereferencing potentially freed dentry, sanitize __d_path() API __d_path() API is asking for trouble and in case of apparmor d_namespace_path() getting just that. The root cause is that when __d_path() misses the root it had been told to look for, it stores the location of the most remote ancestor in *root. Without grabbing references. Sure, at the moment of call it had been pinned down by what we have in *path. And if we raced with umount -l, we could have very well stopped at vfsmount/dentry that got freed as soon as prepend_path() dropped vfsmount_lock. It is safe to compare these pointers with pre-existing (and known to be still alive) vfsmount and dentry, as long as all we are asking is "is it the same address?". Dereferencing is not safe and apparmor ended up stepping into that. d_namespace_path() really wants to examine the place where we stopped, even if it's not connected to our namespace. As the result, it looked at ->d_sb->s_magic of a dentry that might've been already freed by that point. All other callers had been careful enough to avoid that, but it's really a bad interface - it invites that kind of trouble. The fix is fairly straightforward, even though it's bigger than I'd like: * prepend_path() root argument becomes const. * __d_path() is never called with NULL/NULL root. It was a kludge to start with. Instead, we have an explicit function - d_absolute_root(). Same as __d_path(), except that it doesn't get root passed and stops where it stops. apparmor and tomoyo are using it. * __d_path() returns NULL on path outside of root. The main caller is show_mountinfo() and that's precisely what we pass root for - to skip those outside chroot jail. Those who don't want that can (and do) use d_path(). * __d_path() root argument becomes const. Everyone agrees, I hope. * apparmor does *NOT* try to use __d_path() or any of its variants when it sees that path->mnt is an internal vfsmount. In that case it's definitely not mounted anywhere and dentry_path() is exactly what we want there. Handling of sysctl()-triggered weirdness is moved to that place. * if apparmor is asked to do pathname relative to chroot jail and __d_path() tells it we it's not in that jail, the sucker just calls d_absolute_path() instead. That's the other remaining caller of __d_path(), BTW. * seq_path_root() does _NOT_ return -ENAMETOOLONG (it's stupid anyway - the normal seq_file logics will take care of growing the buffer and redoing the call of ->show() just fine). However, if it gets path not reachable from root, it returns SEQ_SKIP. The only caller adjusted (i.e. stopped ignoring the return value as it used to do). Reviewed-by: John Johansen <john.johansen@canonical.com> ACKed-by: John Johansen <john.johansen@canonical.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Cc: stable@vger.kernel.org
2011-12-05 21:43:34 +08:00
static int prepend_path(const struct path *path,
const struct path *root,
char **buffer, int *buflen)
{
struct dentry *dentry;
struct vfsmount *vfsmnt;
struct mount *mnt;
int error = 0;
RCU'd vfsmounts * RCU-delayed freeing of vfsmounts * vfsmount_lock replaced with a seqlock (mount_lock) * sequence number from mount_lock is stored in nameidata->m_seq and used when we exit RCU mode * new vfsmount flag - MNT_SYNC_UMOUNT. Set by umount_tree() when its caller knows that vfsmount will have no surviving references. * synchronize_rcu() done between unlocking namespace_sem in namespace_unlock() and doing pending mntput(). * new helper: legitimize_mnt(mnt, seq). Checks the mount_lock sequence number against seq, then grabs reference to mnt. Then it rechecks mount_lock again to close the race and either returns success or drops the reference it has acquired. The subtle point is that in case of MNT_SYNC_UMOUNT we can simply decrement the refcount and sod off - aforementioned synchronize_rcu() makes sure that final mntput() won't come until we leave RCU mode. We need that, since we don't want to end up with some lazy pathwalk racing with umount() and stealing the final mntput() from it - caller of umount() may expect it to return only once the fs is shut down and we don't want to break that. In other cases (i.e. with MNT_SYNC_UMOUNT absent) we have to do full-blown mntput() in case of mount_lock sequence number mismatch happening just as we'd grabbed the reference, but in those cases we won't be stealing the final mntput() from anything that would care. * mntput_no_expire() doesn't lock anything on the fast path now. Incidentally, SMP and UP cases are handled the same way - no ifdefs there. * normal pathname resolution does *not* do any writes to mount_lock. It does, of course, bump the refcounts of vfsmount and dentry in the very end, but that's it. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-30 10:06:07 +08:00
unsigned seq, m_seq = 0;
dcache: Translating dentry into pathname without taking rename_lock When running the AIM7's short workload, Linus' lockref patch eliminated most of the spinlock contention. However, there were still some left: 8.46% reaim [kernel.kallsyms] [k] _raw_spin_lock |--42.21%-- d_path | proc_pid_readlink | SyS_readlinkat | SyS_readlink | system_call | __GI___readlink | |--40.97%-- sys_getcwd | system_call | __getcwd The big one here is the rename_lock (seqlock) contention in d_path() and the getcwd system call. This patch will eliminate the need to take the rename_lock while translating dentries into the full pathnames. The need to take the rename_lock is to make sure that no rename operation can be ongoing while the translation is in progress. However, only one thread can take the rename_lock thus blocking all the other threads that need it even though the translation process won't make any change to the dentries. This patch will replace the writer's write_seqlock/write_sequnlock sequence of the rename_lock of the callers of the prepend_path() and __dentry_path() functions with the reader's read_seqbegin/read_seqretry sequence within these 2 functions. As a result, the code will have to retry if one or more rename operations had been performed. In addition, RCU read lock will be taken during the translation process to make sure that no dentries will go away. To prevent live-lock from happening, the code will switch back to take the rename_lock if read_seqretry() fails for three times. To further reduce spinlock contention, this patch does not take the dentry's d_lock when copying the filename from the dentries. Instead, it treats the name pointer and length as unreliable and just copy the string byte-by-byte over until it hits a null byte or the end of string as specified by the length. This should avoid stepping into invalid memory address. The error cases are left to be handled by the sequence number check. The following code re-factoring are also made: 1. Move prepend('/') into prepend_name() to remove one conditional check. 2. Move the global root check in prepend_path() back to the top of the while loop. With this patch, the _raw_spin_lock will now account for only 1.2% of the total CPU cycles for the short workload. This patch also has the effect of reducing the effect of running perf on its profile since the perf command itself can be a heavy user of the d_path() function depending on the complexity of the workload. When taking the perf profile of the high-systime workload, the amount of spinlock contention contributed by running perf without this patch was about 16%. With this patch, the spinlock contention caused by the running of perf will go away and we will have a more accurate perf profile. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-10 00:18:13 +08:00
char *bptr;
int blen;
rcu_read_lock();
RCU'd vfsmounts * RCU-delayed freeing of vfsmounts * vfsmount_lock replaced with a seqlock (mount_lock) * sequence number from mount_lock is stored in nameidata->m_seq and used when we exit RCU mode * new vfsmount flag - MNT_SYNC_UMOUNT. Set by umount_tree() when its caller knows that vfsmount will have no surviving references. * synchronize_rcu() done between unlocking namespace_sem in namespace_unlock() and doing pending mntput(). * new helper: legitimize_mnt(mnt, seq). Checks the mount_lock sequence number against seq, then grabs reference to mnt. Then it rechecks mount_lock again to close the race and either returns success or drops the reference it has acquired. The subtle point is that in case of MNT_SYNC_UMOUNT we can simply decrement the refcount and sod off - aforementioned synchronize_rcu() makes sure that final mntput() won't come until we leave RCU mode. We need that, since we don't want to end up with some lazy pathwalk racing with umount() and stealing the final mntput() from it - caller of umount() may expect it to return only once the fs is shut down and we don't want to break that. In other cases (i.e. with MNT_SYNC_UMOUNT absent) we have to do full-blown mntput() in case of mount_lock sequence number mismatch happening just as we'd grabbed the reference, but in those cases we won't be stealing the final mntput() from anything that would care. * mntput_no_expire() doesn't lock anything on the fast path now. Incidentally, SMP and UP cases are handled the same way - no ifdefs there. * normal pathname resolution does *not* do any writes to mount_lock. It does, of course, bump the refcounts of vfsmount and dentry in the very end, but that's it. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-30 10:06:07 +08:00
restart_mnt:
read_seqbegin_or_lock(&mount_lock, &m_seq);
seq = 0;
rcu_read_lock();
dcache: Translating dentry into pathname without taking rename_lock When running the AIM7's short workload, Linus' lockref patch eliminated most of the spinlock contention. However, there were still some left: 8.46% reaim [kernel.kallsyms] [k] _raw_spin_lock |--42.21%-- d_path | proc_pid_readlink | SyS_readlinkat | SyS_readlink | system_call | __GI___readlink | |--40.97%-- sys_getcwd | system_call | __getcwd The big one here is the rename_lock (seqlock) contention in d_path() and the getcwd system call. This patch will eliminate the need to take the rename_lock while translating dentries into the full pathnames. The need to take the rename_lock is to make sure that no rename operation can be ongoing while the translation is in progress. However, only one thread can take the rename_lock thus blocking all the other threads that need it even though the translation process won't make any change to the dentries. This patch will replace the writer's write_seqlock/write_sequnlock sequence of the rename_lock of the callers of the prepend_path() and __dentry_path() functions with the reader's read_seqbegin/read_seqretry sequence within these 2 functions. As a result, the code will have to retry if one or more rename operations had been performed. In addition, RCU read lock will be taken during the translation process to make sure that no dentries will go away. To prevent live-lock from happening, the code will switch back to take the rename_lock if read_seqretry() fails for three times. To further reduce spinlock contention, this patch does not take the dentry's d_lock when copying the filename from the dentries. Instead, it treats the name pointer and length as unreliable and just copy the string byte-by-byte over until it hits a null byte or the end of string as specified by the length. This should avoid stepping into invalid memory address. The error cases are left to be handled by the sequence number check. The following code re-factoring are also made: 1. Move prepend('/') into prepend_name() to remove one conditional check. 2. Move the global root check in prepend_path() back to the top of the while loop. With this patch, the _raw_spin_lock will now account for only 1.2% of the total CPU cycles for the short workload. This patch also has the effect of reducing the effect of running perf on its profile since the perf command itself can be a heavy user of the d_path() function depending on the complexity of the workload. When taking the perf profile of the high-systime workload, the amount of spinlock contention contributed by running perf without this patch was about 16%. With this patch, the spinlock contention caused by the running of perf will go away and we will have a more accurate perf profile. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-10 00:18:13 +08:00
restart:
bptr = *buffer;
blen = *buflen;
RCU'd vfsmounts * RCU-delayed freeing of vfsmounts * vfsmount_lock replaced with a seqlock (mount_lock) * sequence number from mount_lock is stored in nameidata->m_seq and used when we exit RCU mode * new vfsmount flag - MNT_SYNC_UMOUNT. Set by umount_tree() when its caller knows that vfsmount will have no surviving references. * synchronize_rcu() done between unlocking namespace_sem in namespace_unlock() and doing pending mntput(). * new helper: legitimize_mnt(mnt, seq). Checks the mount_lock sequence number against seq, then grabs reference to mnt. Then it rechecks mount_lock again to close the race and either returns success or drops the reference it has acquired. The subtle point is that in case of MNT_SYNC_UMOUNT we can simply decrement the refcount and sod off - aforementioned synchronize_rcu() makes sure that final mntput() won't come until we leave RCU mode. We need that, since we don't want to end up with some lazy pathwalk racing with umount() and stealing the final mntput() from it - caller of umount() may expect it to return only once the fs is shut down and we don't want to break that. In other cases (i.e. with MNT_SYNC_UMOUNT absent) we have to do full-blown mntput() in case of mount_lock sequence number mismatch happening just as we'd grabbed the reference, but in those cases we won't be stealing the final mntput() from anything that would care. * mntput_no_expire() doesn't lock anything on the fast path now. Incidentally, SMP and UP cases are handled the same way - no ifdefs there. * normal pathname resolution does *not* do any writes to mount_lock. It does, of course, bump the refcounts of vfsmount and dentry in the very end, but that's it. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-30 10:06:07 +08:00
error = 0;
dentry = path->dentry;
vfsmnt = path->mnt;
mnt = real_mount(vfsmnt);
dcache: Translating dentry into pathname without taking rename_lock When running the AIM7's short workload, Linus' lockref patch eliminated most of the spinlock contention. However, there were still some left: 8.46% reaim [kernel.kallsyms] [k] _raw_spin_lock |--42.21%-- d_path | proc_pid_readlink | SyS_readlinkat | SyS_readlink | system_call | __GI___readlink | |--40.97%-- sys_getcwd | system_call | __getcwd The big one here is the rename_lock (seqlock) contention in d_path() and the getcwd system call. This patch will eliminate the need to take the rename_lock while translating dentries into the full pathnames. The need to take the rename_lock is to make sure that no rename operation can be ongoing while the translation is in progress. However, only one thread can take the rename_lock thus blocking all the other threads that need it even though the translation process won't make any change to the dentries. This patch will replace the writer's write_seqlock/write_sequnlock sequence of the rename_lock of the callers of the prepend_path() and __dentry_path() functions with the reader's read_seqbegin/read_seqretry sequence within these 2 functions. As a result, the code will have to retry if one or more rename operations had been performed. In addition, RCU read lock will be taken during the translation process to make sure that no dentries will go away. To prevent live-lock from happening, the code will switch back to take the rename_lock if read_seqretry() fails for three times. To further reduce spinlock contention, this patch does not take the dentry's d_lock when copying the filename from the dentries. Instead, it treats the name pointer and length as unreliable and just copy the string byte-by-byte over until it hits a null byte or the end of string as specified by the length. This should avoid stepping into invalid memory address. The error cases are left to be handled by the sequence number check. The following code re-factoring are also made: 1. Move prepend('/') into prepend_name() to remove one conditional check. 2. Move the global root check in prepend_path() back to the top of the while loop. With this patch, the _raw_spin_lock will now account for only 1.2% of the total CPU cycles for the short workload. This patch also has the effect of reducing the effect of running perf on its profile since the perf command itself can be a heavy user of the d_path() function depending on the complexity of the workload. When taking the perf profile of the high-systime workload, the amount of spinlock contention contributed by running perf without this patch was about 16%. With this patch, the spinlock contention caused by the running of perf will go away and we will have a more accurate perf profile. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-10 00:18:13 +08:00
read_seqbegin_or_lock(&rename_lock, &seq);
while (dentry != root->dentry || vfsmnt != root->mnt) {
struct dentry * parent;
if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
RCU'd vfsmounts * RCU-delayed freeing of vfsmounts * vfsmount_lock replaced with a seqlock (mount_lock) * sequence number from mount_lock is stored in nameidata->m_seq and used when we exit RCU mode * new vfsmount flag - MNT_SYNC_UMOUNT. Set by umount_tree() when its caller knows that vfsmount will have no surviving references. * synchronize_rcu() done between unlocking namespace_sem in namespace_unlock() and doing pending mntput(). * new helper: legitimize_mnt(mnt, seq). Checks the mount_lock sequence number against seq, then grabs reference to mnt. Then it rechecks mount_lock again to close the race and either returns success or drops the reference it has acquired. The subtle point is that in case of MNT_SYNC_UMOUNT we can simply decrement the refcount and sod off - aforementioned synchronize_rcu() makes sure that final mntput() won't come until we leave RCU mode. We need that, since we don't want to end up with some lazy pathwalk racing with umount() and stealing the final mntput() from it - caller of umount() may expect it to return only once the fs is shut down and we don't want to break that. In other cases (i.e. with MNT_SYNC_UMOUNT absent) we have to do full-blown mntput() in case of mount_lock sequence number mismatch happening just as we'd grabbed the reference, but in those cases we won't be stealing the final mntput() from anything that would care. * mntput_no_expire() doesn't lock anything on the fast path now. Incidentally, SMP and UP cases are handled the same way - no ifdefs there. * normal pathname resolution does *not* do any writes to mount_lock. It does, of course, bump the refcounts of vfsmount and dentry in the very end, but that's it. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-30 10:06:07 +08:00
struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
/* Escaped? */
if (dentry != vfsmnt->mnt_root) {
bptr = *buffer;
blen = *buflen;
error = 3;
break;
}
/* Global root? */
RCU'd vfsmounts * RCU-delayed freeing of vfsmounts * vfsmount_lock replaced with a seqlock (mount_lock) * sequence number from mount_lock is stored in nameidata->m_seq and used when we exit RCU mode * new vfsmount flag - MNT_SYNC_UMOUNT. Set by umount_tree() when its caller knows that vfsmount will have no surviving references. * synchronize_rcu() done between unlocking namespace_sem in namespace_unlock() and doing pending mntput(). * new helper: legitimize_mnt(mnt, seq). Checks the mount_lock sequence number against seq, then grabs reference to mnt. Then it rechecks mount_lock again to close the race and either returns success or drops the reference it has acquired. The subtle point is that in case of MNT_SYNC_UMOUNT we can simply decrement the refcount and sod off - aforementioned synchronize_rcu() makes sure that final mntput() won't come until we leave RCU mode. We need that, since we don't want to end up with some lazy pathwalk racing with umount() and stealing the final mntput() from it - caller of umount() may expect it to return only once the fs is shut down and we don't want to break that. In other cases (i.e. with MNT_SYNC_UMOUNT absent) we have to do full-blown mntput() in case of mount_lock sequence number mismatch happening just as we'd grabbed the reference, but in those cases we won't be stealing the final mntput() from anything that would care. * mntput_no_expire() doesn't lock anything on the fast path now. Incidentally, SMP and UP cases are handled the same way - no ifdefs there. * normal pathname resolution does *not* do any writes to mount_lock. It does, of course, bump the refcounts of vfsmount and dentry in the very end, but that's it. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-30 10:06:07 +08:00
if (mnt != parent) {
dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
mnt = parent;
dcache: Translating dentry into pathname without taking rename_lock When running the AIM7's short workload, Linus' lockref patch eliminated most of the spinlock contention. However, there were still some left: 8.46% reaim [kernel.kallsyms] [k] _raw_spin_lock |--42.21%-- d_path | proc_pid_readlink | SyS_readlinkat | SyS_readlink | system_call | __GI___readlink | |--40.97%-- sys_getcwd | system_call | __getcwd The big one here is the rename_lock (seqlock) contention in d_path() and the getcwd system call. This patch will eliminate the need to take the rename_lock while translating dentries into the full pathnames. The need to take the rename_lock is to make sure that no rename operation can be ongoing while the translation is in progress. However, only one thread can take the rename_lock thus blocking all the other threads that need it even though the translation process won't make any change to the dentries. This patch will replace the writer's write_seqlock/write_sequnlock sequence of the rename_lock of the callers of the prepend_path() and __dentry_path() functions with the reader's read_seqbegin/read_seqretry sequence within these 2 functions. As a result, the code will have to retry if one or more rename operations had been performed. In addition, RCU read lock will be taken during the translation process to make sure that no dentries will go away. To prevent live-lock from happening, the code will switch back to take the rename_lock if read_seqretry() fails for three times. To further reduce spinlock contention, this patch does not take the dentry's d_lock when copying the filename from the dentries. Instead, it treats the name pointer and length as unreliable and just copy the string byte-by-byte over until it hits a null byte or the end of string as specified by the length. This should avoid stepping into invalid memory address. The error cases are left to be handled by the sequence number check. The following code re-factoring are also made: 1. Move prepend('/') into prepend_name() to remove one conditional check. 2. Move the global root check in prepend_path() back to the top of the while loop. With this patch, the _raw_spin_lock will now account for only 1.2% of the total CPU cycles for the short workload. This patch also has the effect of reducing the effect of running perf on its profile since the perf command itself can be a heavy user of the d_path() function depending on the complexity of the workload. When taking the perf profile of the high-systime workload, the amount of spinlock contention contributed by running perf without this patch was about 16%. With this patch, the spinlock contention caused by the running of perf will go away and we will have a more accurate perf profile. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-10 00:18:13 +08:00
vfsmnt = &mnt->mnt;
continue;
}
if (!error)
error = is_mounted(vfsmnt) ? 1 : 2;
break;
}
parent = dentry->d_parent;
prefetch(parent);
dcache: Translating dentry into pathname without taking rename_lock When running the AIM7's short workload, Linus' lockref patch eliminated most of the spinlock contention. However, there were still some left: 8.46% reaim [kernel.kallsyms] [k] _raw_spin_lock |--42.21%-- d_path | proc_pid_readlink | SyS_readlinkat | SyS_readlink | system_call | __GI___readlink | |--40.97%-- sys_getcwd | system_call | __getcwd The big one here is the rename_lock (seqlock) contention in d_path() and the getcwd system call. This patch will eliminate the need to take the rename_lock while translating dentries into the full pathnames. The need to take the rename_lock is to make sure that no rename operation can be ongoing while the translation is in progress. However, only one thread can take the rename_lock thus blocking all the other threads that need it even though the translation process won't make any change to the dentries. This patch will replace the writer's write_seqlock/write_sequnlock sequence of the rename_lock of the callers of the prepend_path() and __dentry_path() functions with the reader's read_seqbegin/read_seqretry sequence within these 2 functions. As a result, the code will have to retry if one or more rename operations had been performed. In addition, RCU read lock will be taken during the translation process to make sure that no dentries will go away. To prevent live-lock from happening, the code will switch back to take the rename_lock if read_seqretry() fails for three times. To further reduce spinlock contention, this patch does not take the dentry's d_lock when copying the filename from the dentries. Instead, it treats the name pointer and length as unreliable and just copy the string byte-by-byte over until it hits a null byte or the end of string as specified by the length. This should avoid stepping into invalid memory address. The error cases are left to be handled by the sequence number check. The following code re-factoring are also made: 1. Move prepend('/') into prepend_name() to remove one conditional check. 2. Move the global root check in prepend_path() back to the top of the while loop. With this patch, the _raw_spin_lock will now account for only 1.2% of the total CPU cycles for the short workload. This patch also has the effect of reducing the effect of running perf on its profile since the perf command itself can be a heavy user of the d_path() function depending on the complexity of the workload. When taking the perf profile of the high-systime workload, the amount of spinlock contention contributed by running perf without this patch was about 16%. With this patch, the spinlock contention caused by the running of perf will go away and we will have a more accurate perf profile. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-10 00:18:13 +08:00
error = prepend_name(&bptr, &blen, &dentry->d_name);
if (error)
break;
dentry = parent;
}
if (!(seq & 1))
rcu_read_unlock();
if (need_seqretry(&rename_lock, seq)) {
seq = 1;
dcache: Translating dentry into pathname without taking rename_lock When running the AIM7's short workload, Linus' lockref patch eliminated most of the spinlock contention. However, there were still some left: 8.46% reaim [kernel.kallsyms] [k] _raw_spin_lock |--42.21%-- d_path | proc_pid_readlink | SyS_readlinkat | SyS_readlink | system_call | __GI___readlink | |--40.97%-- sys_getcwd | system_call | __getcwd The big one here is the rename_lock (seqlock) contention in d_path() and the getcwd system call. This patch will eliminate the need to take the rename_lock while translating dentries into the full pathnames. The need to take the rename_lock is to make sure that no rename operation can be ongoing while the translation is in progress. However, only one thread can take the rename_lock thus blocking all the other threads that need it even though the translation process won't make any change to the dentries. This patch will replace the writer's write_seqlock/write_sequnlock sequence of the rename_lock of the callers of the prepend_path() and __dentry_path() functions with the reader's read_seqbegin/read_seqretry sequence within these 2 functions. As a result, the code will have to retry if one or more rename operations had been performed. In addition, RCU read lock will be taken during the translation process to make sure that no dentries will go away. To prevent live-lock from happening, the code will switch back to take the rename_lock if read_seqretry() fails for three times. To further reduce spinlock contention, this patch does not take the dentry's d_lock when copying the filename from the dentries. Instead, it treats the name pointer and length as unreliable and just copy the string byte-by-byte over until it hits a null byte or the end of string as specified by the length. This should avoid stepping into invalid memory address. The error cases are left to be handled by the sequence number check. The following code re-factoring are also made: 1. Move prepend('/') into prepend_name() to remove one conditional check. 2. Move the global root check in prepend_path() back to the top of the while loop. With this patch, the _raw_spin_lock will now account for only 1.2% of the total CPU cycles for the short workload. This patch also has the effect of reducing the effect of running perf on its profile since the perf command itself can be a heavy user of the d_path() function depending on the complexity of the workload. When taking the perf profile of the high-systime workload, the amount of spinlock contention contributed by running perf without this patch was about 16%. With this patch, the spinlock contention caused by the running of perf will go away and we will have a more accurate perf profile. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-10 00:18:13 +08:00
goto restart;
}
done_seqretry(&rename_lock, seq);
if (!(m_seq & 1))
rcu_read_unlock();
RCU'd vfsmounts * RCU-delayed freeing of vfsmounts * vfsmount_lock replaced with a seqlock (mount_lock) * sequence number from mount_lock is stored in nameidata->m_seq and used when we exit RCU mode * new vfsmount flag - MNT_SYNC_UMOUNT. Set by umount_tree() when its caller knows that vfsmount will have no surviving references. * synchronize_rcu() done between unlocking namespace_sem in namespace_unlock() and doing pending mntput(). * new helper: legitimize_mnt(mnt, seq). Checks the mount_lock sequence number against seq, then grabs reference to mnt. Then it rechecks mount_lock again to close the race and either returns success or drops the reference it has acquired. The subtle point is that in case of MNT_SYNC_UMOUNT we can simply decrement the refcount and sod off - aforementioned synchronize_rcu() makes sure that final mntput() won't come until we leave RCU mode. We need that, since we don't want to end up with some lazy pathwalk racing with umount() and stealing the final mntput() from it - caller of umount() may expect it to return only once the fs is shut down and we don't want to break that. In other cases (i.e. with MNT_SYNC_UMOUNT absent) we have to do full-blown mntput() in case of mount_lock sequence number mismatch happening just as we'd grabbed the reference, but in those cases we won't be stealing the final mntput() from anything that would care. * mntput_no_expire() doesn't lock anything on the fast path now. Incidentally, SMP and UP cases are handled the same way - no ifdefs there. * normal pathname resolution does *not* do any writes to mount_lock. It does, of course, bump the refcounts of vfsmount and dentry in the very end, but that's it. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-30 10:06:07 +08:00
if (need_seqretry(&mount_lock, m_seq)) {
m_seq = 1;
goto restart_mnt;
}
done_seqretry(&mount_lock, m_seq);
dcache: Translating dentry into pathname without taking rename_lock When running the AIM7's short workload, Linus' lockref patch eliminated most of the spinlock contention. However, there were still some left: 8.46% reaim [kernel.kallsyms] [k] _raw_spin_lock |--42.21%-- d_path | proc_pid_readlink | SyS_readlinkat | SyS_readlink | system_call | __GI___readlink | |--40.97%-- sys_getcwd | system_call | __getcwd The big one here is the rename_lock (seqlock) contention in d_path() and the getcwd system call. This patch will eliminate the need to take the rename_lock while translating dentries into the full pathnames. The need to take the rename_lock is to make sure that no rename operation can be ongoing while the translation is in progress. However, only one thread can take the rename_lock thus blocking all the other threads that need it even though the translation process won't make any change to the dentries. This patch will replace the writer's write_seqlock/write_sequnlock sequence of the rename_lock of the callers of the prepend_path() and __dentry_path() functions with the reader's read_seqbegin/read_seqretry sequence within these 2 functions. As a result, the code will have to retry if one or more rename operations had been performed. In addition, RCU read lock will be taken during the translation process to make sure that no dentries will go away. To prevent live-lock from happening, the code will switch back to take the rename_lock if read_seqretry() fails for three times. To further reduce spinlock contention, this patch does not take the dentry's d_lock when copying the filename from the dentries. Instead, it treats the name pointer and length as unreliable and just copy the string byte-by-byte over until it hits a null byte or the end of string as specified by the length. This should avoid stepping into invalid memory address. The error cases are left to be handled by the sequence number check. The following code re-factoring are also made: 1. Move prepend('/') into prepend_name() to remove one conditional check. 2. Move the global root check in prepend_path() back to the top of the while loop. With this patch, the _raw_spin_lock will now account for only 1.2% of the total CPU cycles for the short workload. This patch also has the effect of reducing the effect of running perf on its profile since the perf command itself can be a heavy user of the d_path() function depending on the complexity of the workload. When taking the perf profile of the high-systime workload, the amount of spinlock contention contributed by running perf without this patch was about 16%. With this patch, the spinlock contention caused by the running of perf will go away and we will have a more accurate perf profile. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-10 00:18:13 +08:00
if (error >= 0 && bptr == *buffer) {
if (--blen < 0)
error = -ENAMETOOLONG;
else
*--bptr = '/';
}
*buffer = bptr;
*buflen = blen;
return error;
}
/**
* __d_path - return the path of a dentry
* @path: the dentry/vfsmount to report
fix apparmor dereferencing potentially freed dentry, sanitize __d_path() API __d_path() API is asking for trouble and in case of apparmor d_namespace_path() getting just that. The root cause is that when __d_path() misses the root it had been told to look for, it stores the location of the most remote ancestor in *root. Without grabbing references. Sure, at the moment of call it had been pinned down by what we have in *path. And if we raced with umount -l, we could have very well stopped at vfsmount/dentry that got freed as soon as prepend_path() dropped vfsmount_lock. It is safe to compare these pointers with pre-existing (and known to be still alive) vfsmount and dentry, as long as all we are asking is "is it the same address?". Dereferencing is not safe and apparmor ended up stepping into that. d_namespace_path() really wants to examine the place where we stopped, even if it's not connected to our namespace. As the result, it looked at ->d_sb->s_magic of a dentry that might've been already freed by that point. All other callers had been careful enough to avoid that, but it's really a bad interface - it invites that kind of trouble. The fix is fairly straightforward, even though it's bigger than I'd like: * prepend_path() root argument becomes const. * __d_path() is never called with NULL/NULL root. It was a kludge to start with. Instead, we have an explicit function - d_absolute_root(). Same as __d_path(), except that it doesn't get root passed and stops where it stops. apparmor and tomoyo are using it. * __d_path() returns NULL on path outside of root. The main caller is show_mountinfo() and that's precisely what we pass root for - to skip those outside chroot jail. Those who don't want that can (and do) use d_path(). * __d_path() root argument becomes const. Everyone agrees, I hope. * apparmor does *NOT* try to use __d_path() or any of its variants when it sees that path->mnt is an internal vfsmount. In that case it's definitely not mounted anywhere and dentry_path() is exactly what we want there. Handling of sysctl()-triggered weirdness is moved to that place. * if apparmor is asked to do pathname relative to chroot jail and __d_path() tells it we it's not in that jail, the sucker just calls d_absolute_path() instead. That's the other remaining caller of __d_path(), BTW. * seq_path_root() does _NOT_ return -ENAMETOOLONG (it's stupid anyway - the normal seq_file logics will take care of growing the buffer and redoing the call of ->show() just fine). However, if it gets path not reachable from root, it returns SEQ_SKIP. The only caller adjusted (i.e. stopped ignoring the return value as it used to do). Reviewed-by: John Johansen <john.johansen@canonical.com> ACKed-by: John Johansen <john.johansen@canonical.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Cc: stable@vger.kernel.org
2011-12-05 21:43:34 +08:00
* @root: root vfsmnt/dentry
* @buf: buffer to return value in
* @buflen: buffer length
*
* Convert a dentry into an ASCII path name.
*
* Returns a pointer into the buffer or an error code if the
* path was too long.
*
* "buflen" should be positive.
*
fix apparmor dereferencing potentially freed dentry, sanitize __d_path() API __d_path() API is asking for trouble and in case of apparmor d_namespace_path() getting just that. The root cause is that when __d_path() misses the root it had been told to look for, it stores the location of the most remote ancestor in *root. Without grabbing references. Sure, at the moment of call it had been pinned down by what we have in *path. And if we raced with umount -l, we could have very well stopped at vfsmount/dentry that got freed as soon as prepend_path() dropped vfsmount_lock. It is safe to compare these pointers with pre-existing (and known to be still alive) vfsmount and dentry, as long as all we are asking is "is it the same address?". Dereferencing is not safe and apparmor ended up stepping into that. d_namespace_path() really wants to examine the place where we stopped, even if it's not connected to our namespace. As the result, it looked at ->d_sb->s_magic of a dentry that might've been already freed by that point. All other callers had been careful enough to avoid that, but it's really a bad interface - it invites that kind of trouble. The fix is fairly straightforward, even though it's bigger than I'd like: * prepend_path() root argument becomes const. * __d_path() is never called with NULL/NULL root. It was a kludge to start with. Instead, we have an explicit function - d_absolute_root(). Same as __d_path(), except that it doesn't get root passed and stops where it stops. apparmor and tomoyo are using it. * __d_path() returns NULL on path outside of root. The main caller is show_mountinfo() and that's precisely what we pass root for - to skip those outside chroot jail. Those who don't want that can (and do) use d_path(). * __d_path() root argument becomes const. Everyone agrees, I hope. * apparmor does *NOT* try to use __d_path() or any of its variants when it sees that path->mnt is an internal vfsmount. In that case it's definitely not mounted anywhere and dentry_path() is exactly what we want there. Handling of sysctl()-triggered weirdness is moved to that place. * if apparmor is asked to do pathname relative to chroot jail and __d_path() tells it we it's not in that jail, the sucker just calls d_absolute_path() instead. That's the other remaining caller of __d_path(), BTW. * seq_path_root() does _NOT_ return -ENAMETOOLONG (it's stupid anyway - the normal seq_file logics will take care of growing the buffer and redoing the call of ->show() just fine). However, if it gets path not reachable from root, it returns SEQ_SKIP. The only caller adjusted (i.e. stopped ignoring the return value as it used to do). Reviewed-by: John Johansen <john.johansen@canonical.com> ACKed-by: John Johansen <john.johansen@canonical.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Cc: stable@vger.kernel.org
2011-12-05 21:43:34 +08:00
* If the path is not reachable from the supplied root, return %NULL.
*/
fix apparmor dereferencing potentially freed dentry, sanitize __d_path() API __d_path() API is asking for trouble and in case of apparmor d_namespace_path() getting just that. The root cause is that when __d_path() misses the root it had been told to look for, it stores the location of the most remote ancestor in *root. Without grabbing references. Sure, at the moment of call it had been pinned down by what we have in *path. And if we raced with umount -l, we could have very well stopped at vfsmount/dentry that got freed as soon as prepend_path() dropped vfsmount_lock. It is safe to compare these pointers with pre-existing (and known to be still alive) vfsmount and dentry, as long as all we are asking is "is it the same address?". Dereferencing is not safe and apparmor ended up stepping into that. d_namespace_path() really wants to examine the place where we stopped, even if it's not connected to our namespace. As the result, it looked at ->d_sb->s_magic of a dentry that might've been already freed by that point. All other callers had been careful enough to avoid that, but it's really a bad interface - it invites that kind of trouble. The fix is fairly straightforward, even though it's bigger than I'd like: * prepend_path() root argument becomes const. * __d_path() is never called with NULL/NULL root. It was a kludge to start with. Instead, we have an explicit function - d_absolute_root(). Same as __d_path(), except that it doesn't get root passed and stops where it stops. apparmor and tomoyo are using it. * __d_path() returns NULL on path outside of root. The main caller is show_mountinfo() and that's precisely what we pass root for - to skip those outside chroot jail. Those who don't want that can (and do) use d_path(). * __d_path() root argument becomes const. Everyone agrees, I hope. * apparmor does *NOT* try to use __d_path() or any of its variants when it sees that path->mnt is an internal vfsmount. In that case it's definitely not mounted anywhere and dentry_path() is exactly what we want there. Handling of sysctl()-triggered weirdness is moved to that place. * if apparmor is asked to do pathname relative to chroot jail and __d_path() tells it we it's not in that jail, the sucker just calls d_absolute_path() instead. That's the other remaining caller of __d_path(), BTW. * seq_path_root() does _NOT_ return -ENAMETOOLONG (it's stupid anyway - the normal seq_file logics will take care of growing the buffer and redoing the call of ->show() just fine). However, if it gets path not reachable from root, it returns SEQ_SKIP. The only caller adjusted (i.e. stopped ignoring the return value as it used to do). Reviewed-by: John Johansen <john.johansen@canonical.com> ACKed-by: John Johansen <john.johansen@canonical.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Cc: stable@vger.kernel.org
2011-12-05 21:43:34 +08:00
char *__d_path(const struct path *path,
const struct path *root,
char *buf, int buflen)
{
char *res = buf + buflen;
int error;
prepend(&res, &buflen, "\0", 1);
error = prepend_path(path, root, &res, &buflen);
fix apparmor dereferencing potentially freed dentry, sanitize __d_path() API __d_path() API is asking for trouble and in case of apparmor d_namespace_path() getting just that. The root cause is that when __d_path() misses the root it had been told to look for, it stores the location of the most remote ancestor in *root. Without grabbing references. Sure, at the moment of call it had been pinned down by what we have in *path. And if we raced with umount -l, we could have very well stopped at vfsmount/dentry that got freed as soon as prepend_path() dropped vfsmount_lock. It is safe to compare these pointers with pre-existing (and known to be still alive) vfsmount and dentry, as long as all we are asking is "is it the same address?". Dereferencing is not safe and apparmor ended up stepping into that. d_namespace_path() really wants to examine the place where we stopped, even if it's not connected to our namespace. As the result, it looked at ->d_sb->s_magic of a dentry that might've been already freed by that point. All other callers had been careful enough to avoid that, but it's really a bad interface - it invites that kind of trouble. The fix is fairly straightforward, even though it's bigger than I'd like: * prepend_path() root argument becomes const. * __d_path() is never called with NULL/NULL root. It was a kludge to start with. Instead, we have an explicit function - d_absolute_root(). Same as __d_path(), except that it doesn't get root passed and stops where it stops. apparmor and tomoyo are using it. * __d_path() returns NULL on path outside of root. The main caller is show_mountinfo() and that's precisely what we pass root for - to skip those outside chroot jail. Those who don't want that can (and do) use d_path(). * __d_path() root argument becomes const. Everyone agrees, I hope. * apparmor does *NOT* try to use __d_path() or any of its variants when it sees that path->mnt is an internal vfsmount. In that case it's definitely not mounted anywhere and dentry_path() is exactly what we want there. Handling of sysctl()-triggered weirdness is moved to that place. * if apparmor is asked to do pathname relative to chroot jail and __d_path() tells it we it's not in that jail, the sucker just calls d_absolute_path() instead. That's the other remaining caller of __d_path(), BTW. * seq_path_root() does _NOT_ return -ENAMETOOLONG (it's stupid anyway - the normal seq_file logics will take care of growing the buffer and redoing the call of ->show() just fine). However, if it gets path not reachable from root, it returns SEQ_SKIP. The only caller adjusted (i.e. stopped ignoring the return value as it used to do). Reviewed-by: John Johansen <john.johansen@canonical.com> ACKed-by: John Johansen <john.johansen@canonical.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Cc: stable@vger.kernel.org
2011-12-05 21:43:34 +08:00
if (error < 0)
return ERR_PTR(error);
if (error > 0)
return NULL;
return res;
}
char *d_absolute_path(const struct path *path,
char *buf, int buflen)
{
struct path root = {};
char *res = buf + buflen;
int error;
prepend(&res, &buflen, "\0", 1);
error = prepend_path(path, &root, &res, &buflen);
if (error > 1)
error = -EINVAL;
if (error < 0)
return ERR_PTR(error);
return res;
}
/*
* same as __d_path but appends "(deleted)" for unlinked files.
*/
fix apparmor dereferencing potentially freed dentry, sanitize __d_path() API __d_path() API is asking for trouble and in case of apparmor d_namespace_path() getting just that. The root cause is that when __d_path() misses the root it had been told to look for, it stores the location of the most remote ancestor in *root. Without grabbing references. Sure, at the moment of call it had been pinned down by what we have in *path. And if we raced with umount -l, we could have very well stopped at vfsmount/dentry that got freed as soon as prepend_path() dropped vfsmount_lock. It is safe to compare these pointers with pre-existing (and known to be still alive) vfsmount and dentry, as long as all we are asking is "is it the same address?". Dereferencing is not safe and apparmor ended up stepping into that. d_namespace_path() really wants to examine the place where we stopped, even if it's not connected to our namespace. As the result, it looked at ->d_sb->s_magic of a dentry that might've been already freed by that point. All other callers had been careful enough to avoid that, but it's really a bad interface - it invites that kind of trouble. The fix is fairly straightforward, even though it's bigger than I'd like: * prepend_path() root argument becomes const. * __d_path() is never called with NULL/NULL root. It was a kludge to start with. Instead, we have an explicit function - d_absolute_root(). Same as __d_path(), except that it doesn't get root passed and stops where it stops. apparmor and tomoyo are using it. * __d_path() returns NULL on path outside of root. The main caller is show_mountinfo() and that's precisely what we pass root for - to skip those outside chroot jail. Those who don't want that can (and do) use d_path(). * __d_path() root argument becomes const. Everyone agrees, I hope. * apparmor does *NOT* try to use __d_path() or any of its variants when it sees that path->mnt is an internal vfsmount. In that case it's definitely not mounted anywhere and dentry_path() is exactly what we want there. Handling of sysctl()-triggered weirdness is moved to that place. * if apparmor is asked to do pathname relative to chroot jail and __d_path() tells it we it's not in that jail, the sucker just calls d_absolute_path() instead. That's the other remaining caller of __d_path(), BTW. * seq_path_root() does _NOT_ return -ENAMETOOLONG (it's stupid anyway - the normal seq_file logics will take care of growing the buffer and redoing the call of ->show() just fine). However, if it gets path not reachable from root, it returns SEQ_SKIP. The only caller adjusted (i.e. stopped ignoring the return value as it used to do). Reviewed-by: John Johansen <john.johansen@canonical.com> ACKed-by: John Johansen <john.johansen@canonical.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Cc: stable@vger.kernel.org
2011-12-05 21:43:34 +08:00
static int path_with_deleted(const struct path *path,
const struct path *root,
char **buf, int *buflen)
{
prepend(buf, buflen, "\0", 1);
if (d_unlinked(path->dentry)) {
int error = prepend(buf, buflen, " (deleted)", 10);
if (error)
return error;
}
return prepend_path(path, root, buf, buflen);
}
static int prepend_unreachable(char **buffer, int *buflen)
{
return prepend(buffer, buflen, "(unreachable)", 13);
}
static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
{
unsigned seq;
do {
seq = read_seqcount_begin(&fs->seq);
*root = fs->root;
} while (read_seqcount_retry(&fs->seq, seq));
}
/**
* d_path - return the path of a dentry
* @path: path to report
* @buf: buffer to return value in
* @buflen: buffer length
*
* Convert a dentry into an ASCII path name. If the entry has been deleted
* the string " (deleted)" is appended. Note that this is ambiguous.
*
* Returns a pointer into the buffer or an error code if the path was
* too long. Note: Callers should use the returned pointer, not the passed
* in buffer, to use the name! The implementation often starts at an offset
* into the buffer, and may leave 0 bytes at the start.
*
* "buflen" should be positive.
*/
char *d_path(const struct path *path, char *buf, int buflen)
{
char *res = buf + buflen;
struct path root;
int error;
/*
* We have various synthetic filesystems that never get mounted. On
* these filesystems dentries are never used for lookup purposes, and
* thus don't need to be hashed. They also don't need a name until a
* user wants to identify the object in /proc/pid/fd/. The little hack
* below allows us to generate a name for these objects on demand:
2013-11-09 08:31:29 +08:00
*
* Some pseudo inodes are mountable. When they are mounted
* path->dentry == path->mnt->mnt_root. In that case don't call d_dname
* and instead have d_path return the mounted path.
*/
2013-11-09 08:31:29 +08:00
if (path->dentry->d_op && path->dentry->d_op->d_dname &&
(!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
rcu_read_lock();
get_fs_root_rcu(current->fs, &root);
fix apparmor dereferencing potentially freed dentry, sanitize __d_path() API __d_path() API is asking for trouble and in case of apparmor d_namespace_path() getting just that. The root cause is that when __d_path() misses the root it had been told to look for, it stores the location of the most remote ancestor in *root. Without grabbing references. Sure, at the moment of call it had been pinned down by what we have in *path. And if we raced with umount -l, we could have very well stopped at vfsmount/dentry that got freed as soon as prepend_path() dropped vfsmount_lock. It is safe to compare these pointers with pre-existing (and known to be still alive) vfsmount and dentry, as long as all we are asking is "is it the same address?". Dereferencing is not safe and apparmor ended up stepping into that. d_namespace_path() really wants to examine the place where we stopped, even if it's not connected to our namespace. As the result, it looked at ->d_sb->s_magic of a dentry that might've been already freed by that point. All other callers had been careful enough to avoid that, but it's really a bad interface - it invites that kind of trouble. The fix is fairly straightforward, even though it's bigger than I'd like: * prepend_path() root argument becomes const. * __d_path() is never called with NULL/NULL root. It was a kludge to start with. Instead, we have an explicit function - d_absolute_root(). Same as __d_path(), except that it doesn't get root passed and stops where it stops. apparmor and tomoyo are using it. * __d_path() returns NULL on path outside of root. The main caller is show_mountinfo() and that's precisely what we pass root for - to skip those outside chroot jail. Those who don't want that can (and do) use d_path(). * __d_path() root argument becomes const. Everyone agrees, I hope. * apparmor does *NOT* try to use __d_path() or any of its variants when it sees that path->mnt is an internal vfsmount. In that case it's definitely not mounted anywhere and dentry_path() is exactly what we want there. Handling of sysctl()-triggered weirdness is moved to that place. * if apparmor is asked to do pathname relative to chroot jail and __d_path() tells it we it's not in that jail, the sucker just calls d_absolute_path() instead. That's the other remaining caller of __d_path(), BTW. * seq_path_root() does _NOT_ return -ENAMETOOLONG (it's stupid anyway - the normal seq_file logics will take care of growing the buffer and redoing the call of ->show() just fine). However, if it gets path not reachable from root, it returns SEQ_SKIP. The only caller adjusted (i.e. stopped ignoring the return value as it used to do). Reviewed-by: John Johansen <john.johansen@canonical.com> ACKed-by: John Johansen <john.johansen@canonical.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Cc: stable@vger.kernel.org
2011-12-05 21:43:34 +08:00
error = path_with_deleted(path, &root, &res, &buflen);
rcu_read_unlock();
fix apparmor dereferencing potentially freed dentry, sanitize __d_path() API __d_path() API is asking for trouble and in case of apparmor d_namespace_path() getting just that. The root cause is that when __d_path() misses the root it had been told to look for, it stores the location of the most remote ancestor in *root. Without grabbing references. Sure, at the moment of call it had been pinned down by what we have in *path. And if we raced with umount -l, we could have very well stopped at vfsmount/dentry that got freed as soon as prepend_path() dropped vfsmount_lock. It is safe to compare these pointers with pre-existing (and known to be still alive) vfsmount and dentry, as long as all we are asking is "is it the same address?". Dereferencing is not safe and apparmor ended up stepping into that. d_namespace_path() really wants to examine the place where we stopped, even if it's not connected to our namespace. As the result, it looked at ->d_sb->s_magic of a dentry that might've been already freed by that point. All other callers had been careful enough to avoid that, but it's really a bad interface - it invites that kind of trouble. The fix is fairly straightforward, even though it's bigger than I'd like: * prepend_path() root argument becomes const. * __d_path() is never called with NULL/NULL root. It was a kludge to start with. Instead, we have an explicit function - d_absolute_root(). Same as __d_path(), except that it doesn't get root passed and stops where it stops. apparmor and tomoyo are using it. * __d_path() returns NULL on path outside of root. The main caller is show_mountinfo() and that's precisely what we pass root for - to skip those outside chroot jail. Those who don't want that can (and do) use d_path(). * __d_path() root argument becomes const. Everyone agrees, I hope. * apparmor does *NOT* try to use __d_path() or any of its variants when it sees that path->mnt is an internal vfsmount. In that case it's definitely not mounted anywhere and dentry_path() is exactly what we want there. Handling of sysctl()-triggered weirdness is moved to that place. * if apparmor is asked to do pathname relative to chroot jail and __d_path() tells it we it's not in that jail, the sucker just calls d_absolute_path() instead. That's the other remaining caller of __d_path(), BTW. * seq_path_root() does _NOT_ return -ENAMETOOLONG (it's stupid anyway - the normal seq_file logics will take care of growing the buffer and redoing the call of ->show() just fine). However, if it gets path not reachable from root, it returns SEQ_SKIP. The only caller adjusted (i.e. stopped ignoring the return value as it used to do). Reviewed-by: John Johansen <john.johansen@canonical.com> ACKed-by: John Johansen <john.johansen@canonical.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Cc: stable@vger.kernel.org
2011-12-05 21:43:34 +08:00
if (error < 0)
res = ERR_PTR(error);
return res;
}
EXPORT_SYMBOL(d_path);
/*
* Helper function for dentry_operations.d_dname() members
*/
char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
const char *fmt, ...)
{
va_list args;
char temp[64];
int sz;
va_start(args, fmt);
sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
va_end(args);
if (sz > sizeof(temp) || sz > buflen)
return ERR_PTR(-ENAMETOOLONG);
buffer += buflen - sz;
return memcpy(buffer, temp, sz);
}
char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
{
char *end = buffer + buflen;
/* these dentries are never renamed, so d_lock is not needed */
if (prepend(&end, &buflen, " (deleted)", 11) ||
dcache: Translating dentry into pathname without taking rename_lock When running the AIM7's short workload, Linus' lockref patch eliminated most of the spinlock contention. However, there were still some left: 8.46% reaim [kernel.kallsyms] [k] _raw_spin_lock |--42.21%-- d_path | proc_pid_readlink | SyS_readlinkat | SyS_readlink | system_call | __GI___readlink | |--40.97%-- sys_getcwd | system_call | __getcwd The big one here is the rename_lock (seqlock) contention in d_path() and the getcwd system call. This patch will eliminate the need to take the rename_lock while translating dentries into the full pathnames. The need to take the rename_lock is to make sure that no rename operation can be ongoing while the translation is in progress. However, only one thread can take the rename_lock thus blocking all the other threads that need it even though the translation process won't make any change to the dentries. This patch will replace the writer's write_seqlock/write_sequnlock sequence of the rename_lock of the callers of the prepend_path() and __dentry_path() functions with the reader's read_seqbegin/read_seqretry sequence within these 2 functions. As a result, the code will have to retry if one or more rename operations had been performed. In addition, RCU read lock will be taken during the translation process to make sure that no dentries will go away. To prevent live-lock from happening, the code will switch back to take the rename_lock if read_seqretry() fails for three times. To further reduce spinlock contention, this patch does not take the dentry's d_lock when copying the filename from the dentries. Instead, it treats the name pointer and length as unreliable and just copy the string byte-by-byte over until it hits a null byte or the end of string as specified by the length. This should avoid stepping into invalid memory address. The error cases are left to be handled by the sequence number check. The following code re-factoring are also made: 1. Move prepend('/') into prepend_name() to remove one conditional check. 2. Move the global root check in prepend_path() back to the top of the while loop. With this patch, the _raw_spin_lock will now account for only 1.2% of the total CPU cycles for the short workload. This patch also has the effect of reducing the effect of running perf on its profile since the perf command itself can be a heavy user of the d_path() function depending on the complexity of the workload. When taking the perf profile of the high-systime workload, the amount of spinlock contention contributed by running perf without this patch was about 16%. With this patch, the spinlock contention caused by the running of perf will go away and we will have a more accurate perf profile. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-10 00:18:13 +08:00
prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
prepend(&end, &buflen, "/", 1))
end = ERR_PTR(-ENAMETOOLONG);
dcache: Translating dentry into pathname without taking rename_lock When running the AIM7's short workload, Linus' lockref patch eliminated most of the spinlock contention. However, there were still some left: 8.46% reaim [kernel.kallsyms] [k] _raw_spin_lock |--42.21%-- d_path | proc_pid_readlink | SyS_readlinkat | SyS_readlink | system_call | __GI___readlink | |--40.97%-- sys_getcwd | system_call | __getcwd The big one here is the rename_lock (seqlock) contention in d_path() and the getcwd system call. This patch will eliminate the need to take the rename_lock while translating dentries into the full pathnames. The need to take the rename_lock is to make sure that no rename operation can be ongoing while the translation is in progress. However, only one thread can take the rename_lock thus blocking all the other threads that need it even though the translation process won't make any change to the dentries. This patch will replace the writer's write_seqlock/write_sequnlock sequence of the rename_lock of the callers of the prepend_path() and __dentry_path() functions with the reader's read_seqbegin/read_seqretry sequence within these 2 functions. As a result, the code will have to retry if one or more rename operations had been performed. In addition, RCU read lock will be taken during the translation process to make sure that no dentries will go away. To prevent live-lock from happening, the code will switch back to take the rename_lock if read_seqretry() fails for three times. To further reduce spinlock contention, this patch does not take the dentry's d_lock when copying the filename from the dentries. Instead, it treats the name pointer and length as unreliable and just copy the string byte-by-byte over until it hits a null byte or the end of string as specified by the length. This should avoid stepping into invalid memory address. The error cases are left to be handled by the sequence number check. The following code re-factoring are also made: 1. Move prepend('/') into prepend_name() to remove one conditional check. 2. Move the global root check in prepend_path() back to the top of the while loop. With this patch, the _raw_spin_lock will now account for only 1.2% of the total CPU cycles for the short workload. This patch also has the effect of reducing the effect of running perf on its profile since the perf command itself can be a heavy user of the d_path() function depending on the complexity of the workload. When taking the perf profile of the high-systime workload, the amount of spinlock contention contributed by running perf without this patch was about 16%. With this patch, the spinlock contention caused by the running of perf will go away and we will have a more accurate perf profile. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-10 00:18:13 +08:00
return end;
}
drm: add pseudo filesystem for shared inodes Our current DRM design uses a single address_space for all users of the same DRM device. However, there is no way to create an anonymous address_space without an underlying inode. Therefore, we wait for the first ->open() callback on a registered char-dev and take-over the inode of the char-dev. This worked well so far, but has several drawbacks: - We screw with FS internals and rely on some non-obvious invariants like inode->i_mapping being the same as inode->i_data for char-devs. - We don't have any address_space prior to the first ->open() from user-space. This leads to ugly fallback code and we cannot allocate global objects early. As pointed out by Al-Viro, fs/anon_inode.c is *not* supposed to be used by drivers for anonymous inode-allocation. Therefore, this patch follows the proposed alternative solution and adds a pseudo filesystem mount-point to DRM. We can then allocate private inodes including a private address_space for each DRM device at initialization time. Note that we could use: sysfs_get_inode(sysfs_mnt->mnt_sb, drm_device->dev->kobj.sd); to get access to the underlying sysfs-inode of a "struct device" object. However, most of this information is currently hidden and it's not clear whether this address_space is suitable for driver access. Thus, unless linux allows anonymous address_space objects or driver-core provides a public inode per device, we're left with our own private internal mount point. Cc: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
2014-01-03 21:09:47 +08:00
EXPORT_SYMBOL(simple_dname);
/*
* Write full pathname from the root of the filesystem into the buffer.
*/
static char *__dentry_path(struct dentry *d, char *buf, int buflen)
{
struct dentry *dentry;
dcache: Translating dentry into pathname without taking rename_lock When running the AIM7's short workload, Linus' lockref patch eliminated most of the spinlock contention. However, there were still some left: 8.46% reaim [kernel.kallsyms] [k] _raw_spin_lock |--42.21%-- d_path | proc_pid_readlink | SyS_readlinkat | SyS_readlink | system_call | __GI___readlink | |--40.97%-- sys_getcwd | system_call | __getcwd The big one here is the rename_lock (seqlock) contention in d_path() and the getcwd system call. This patch will eliminate the need to take the rename_lock while translating dentries into the full pathnames. The need to take the rename_lock is to make sure that no rename operation can be ongoing while the translation is in progress. However, only one thread can take the rename_lock thus blocking all the other threads that need it even though the translation process won't make any change to the dentries. This patch will replace the writer's write_seqlock/write_sequnlock sequence of the rename_lock of the callers of the prepend_path() and __dentry_path() functions with the reader's read_seqbegin/read_seqretry sequence within these 2 functions. As a result, the code will have to retry if one or more rename operations had been performed. In addition, RCU read lock will be taken during the translation process to make sure that no dentries will go away. To prevent live-lock from happening, the code will switch back to take the rename_lock if read_seqretry() fails for three times. To further reduce spinlock contention, this patch does not take the dentry's d_lock when copying the filename from the dentries. Instead, it treats the name pointer and length as unreliable and just copy the string byte-by-byte over until it hits a null byte or the end of string as specified by the length. This should avoid stepping into invalid memory address. The error cases are left to be handled by the sequence number check. The following code re-factoring are also made: 1. Move prepend('/') into prepend_name() to remove one conditional check. 2. Move the global root check in prepend_path() back to the top of the while loop. With this patch, the _raw_spin_lock will now account for only 1.2% of the total CPU cycles for the short workload. This patch also has the effect of reducing the effect of running perf on its profile since the perf command itself can be a heavy user of the d_path() function depending on the complexity of the workload. When taking the perf profile of the high-systime workload, the amount of spinlock contention contributed by running perf without this patch was about 16%. With this patch, the spinlock contention caused by the running of perf will go away and we will have a more accurate perf profile. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-10 00:18:13 +08:00
char *end, *retval;
int len, seq = 0;
int error = 0;
if (buflen < 2)
goto Elong;
rcu_read_lock();
dcache: Translating dentry into pathname without taking rename_lock When running the AIM7's short workload, Linus' lockref patch eliminated most of the spinlock contention. However, there were still some left: 8.46% reaim [kernel.kallsyms] [k] _raw_spin_lock |--42.21%-- d_path | proc_pid_readlink | SyS_readlinkat | SyS_readlink | system_call | __GI___readlink | |--40.97%-- sys_getcwd | system_call | __getcwd The big one here is the rename_lock (seqlock) contention in d_path() and the getcwd system call. This patch will eliminate the need to take the rename_lock while translating dentries into the full pathnames. The need to take the rename_lock is to make sure that no rename operation can be ongoing while the translation is in progress. However, only one thread can take the rename_lock thus blocking all the other threads that need it even though the translation process won't make any change to the dentries. This patch will replace the writer's write_seqlock/write_sequnlock sequence of the rename_lock of the callers of the prepend_path() and __dentry_path() functions with the reader's read_seqbegin/read_seqretry sequence within these 2 functions. As a result, the code will have to retry if one or more rename operations had been performed. In addition, RCU read lock will be taken during the translation process to make sure that no dentries will go away. To prevent live-lock from happening, the code will switch back to take the rename_lock if read_seqretry() fails for three times. To further reduce spinlock contention, this patch does not take the dentry's d_lock when copying the filename from the dentries. Instead, it treats the name pointer and length as unreliable and just copy the string byte-by-byte over until it hits a null byte or the end of string as specified by the length. This should avoid stepping into invalid memory address. The error cases are left to be handled by the sequence number check. The following code re-factoring are also made: 1. Move prepend('/') into prepend_name() to remove one conditional check. 2. Move the global root check in prepend_path() back to the top of the while loop. With this patch, the _raw_spin_lock will now account for only 1.2% of the total CPU cycles for the short workload. This patch also has the effect of reducing the effect of running perf on its profile since the perf command itself can be a heavy user of the d_path() function depending on the complexity of the workload. When taking the perf profile of the high-systime workload, the amount of spinlock contention contributed by running perf without this patch was about 16%. With this patch, the spinlock contention caused by the running of perf will go away and we will have a more accurate perf profile. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-10 00:18:13 +08:00
restart:
dentry = d;
dcache: Translating dentry into pathname without taking rename_lock When running the AIM7's short workload, Linus' lockref patch eliminated most of the spinlock contention. However, there were still some left: 8.46% reaim [kernel.kallsyms] [k] _raw_spin_lock |--42.21%-- d_path | proc_pid_readlink | SyS_readlinkat | SyS_readlink | system_call | __GI___readlink | |--40.97%-- sys_getcwd | system_call | __getcwd The big one here is the rename_lock (seqlock) contention in d_path() and the getcwd system call. This patch will eliminate the need to take the rename_lock while translating dentries into the full pathnames. The need to take the rename_lock is to make sure that no rename operation can be ongoing while the translation is in progress. However, only one thread can take the rename_lock thus blocking all the other threads that need it even though the translation process won't make any change to the dentries. This patch will replace the writer's write_seqlock/write_sequnlock sequence of the rename_lock of the callers of the prepend_path() and __dentry_path() functions with the reader's read_seqbegin/read_seqretry sequence within these 2 functions. As a result, the code will have to retry if one or more rename operations had been performed. In addition, RCU read lock will be taken during the translation process to make sure that no dentries will go away. To prevent live-lock from happening, the code will switch back to take the rename_lock if read_seqretry() fails for three times. To further reduce spinlock contention, this patch does not take the dentry's d_lock when copying the filename from the dentries. Instead, it treats the name pointer and length as unreliable and just copy the string byte-by-byte over until it hits a null byte or the end of string as specified by the length. This should avoid stepping into invalid memory address. The error cases are left to be handled by the sequence number check. The following code re-factoring are also made: 1. Move prepend('/') into prepend_name() to remove one conditional check. 2. Move the global root check in prepend_path() back to the top of the while loop. With this patch, the _raw_spin_lock will now account for only 1.2% of the total CPU cycles for the short workload. This patch also has the effect of reducing the effect of running perf on its profile since the perf command itself can be a heavy user of the d_path() function depending on the complexity of the workload. When taking the perf profile of the high-systime workload, the amount of spinlock contention contributed by running perf without this patch was about 16%. With this patch, the spinlock contention caused by the running of perf will go away and we will have a more accurate perf profile. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-10 00:18:13 +08:00
end = buf + buflen;
len = buflen;
prepend(&end, &len, "\0", 1);
/* Get '/' right */
retval = end-1;
*retval = '/';
dcache: Translating dentry into pathname without taking rename_lock When running the AIM7's short workload, Linus' lockref patch eliminated most of the spinlock contention. However, there were still some left: 8.46% reaim [kernel.kallsyms] [k] _raw_spin_lock |--42.21%-- d_path | proc_pid_readlink | SyS_readlinkat | SyS_readlink | system_call | __GI___readlink | |--40.97%-- sys_getcwd | system_call | __getcwd The big one here is the rename_lock (seqlock) contention in d_path() and the getcwd system call. This patch will eliminate the need to take the rename_lock while translating dentries into the full pathnames. The need to take the rename_lock is to make sure that no rename operation can be ongoing while the translation is in progress. However, only one thread can take the rename_lock thus blocking all the other threads that need it even though the translation process won't make any change to the dentries. This patch will replace the writer's write_seqlock/write_sequnlock sequence of the rename_lock of the callers of the prepend_path() and __dentry_path() functions with the reader's read_seqbegin/read_seqretry sequence within these 2 functions. As a result, the code will have to retry if one or more rename operations had been performed. In addition, RCU read lock will be taken during the translation process to make sure that no dentries will go away. To prevent live-lock from happening, the code will switch back to take the rename_lock if read_seqretry() fails for three times. To further reduce spinlock contention, this patch does not take the dentry's d_lock when copying the filename from the dentries. Instead, it treats the name pointer and length as unreliable and just copy the string byte-by-byte over until it hits a null byte or the end of string as specified by the length. This should avoid stepping into invalid memory address. The error cases are left to be handled by the sequence number check. The following code re-factoring are also made: 1. Move prepend('/') into prepend_name() to remove one conditional check. 2. Move the global root check in prepend_path() back to the top of the while loop. With this patch, the _raw_spin_lock will now account for only 1.2% of the total CPU cycles for the short workload. This patch also has the effect of reducing the effect of running perf on its profile since the perf command itself can be a heavy user of the d_path() function depending on the complexity of the workload. When taking the perf profile of the high-systime workload, the amount of spinlock contention contributed by running perf without this patch was about 16%. With this patch, the spinlock contention caused by the running of perf will go away and we will have a more accurate perf profile. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-10 00:18:13 +08:00
read_seqbegin_or_lock(&rename_lock, &seq);
while (!IS_ROOT(dentry)) {
struct dentry *parent = dentry->d_parent;
prefetch(parent);
dcache: Translating dentry into pathname without taking rename_lock When running the AIM7's short workload, Linus' lockref patch eliminated most of the spinlock contention. However, there were still some left: 8.46% reaim [kernel.kallsyms] [k] _raw_spin_lock |--42.21%-- d_path | proc_pid_readlink | SyS_readlinkat | SyS_readlink | system_call | __GI___readlink | |--40.97%-- sys_getcwd | system_call | __getcwd The big one here is the rename_lock (seqlock) contention in d_path() and the getcwd system call. This patch will eliminate the need to take the rename_lock while translating dentries into the full pathnames. The need to take the rename_lock is to make sure that no rename operation can be ongoing while the translation is in progress. However, only one thread can take the rename_lock thus blocking all the other threads that need it even though the translation process won't make any change to the dentries. This patch will replace the writer's write_seqlock/write_sequnlock sequence of the rename_lock of the callers of the prepend_path() and __dentry_path() functions with the reader's read_seqbegin/read_seqretry sequence within these 2 functions. As a result, the code will have to retry if one or more rename operations had been performed. In addition, RCU read lock will be taken during the translation process to make sure that no dentries will go away. To prevent live-lock from happening, the code will switch back to take the rename_lock if read_seqretry() fails for three times. To further reduce spinlock contention, this patch does not take the dentry's d_lock when copying the filename from the dentries. Instead, it treats the name pointer and length as unreliable and just copy the string byte-by-byte over until it hits a null byte or the end of string as specified by the length. This should avoid stepping into invalid memory address. The error cases are left to be handled by the sequence number check. The following code re-factoring are also made: 1. Move prepend('/') into prepend_name() to remove one conditional check. 2. Move the global root check in prepend_path() back to the top of the while loop. With this patch, the _raw_spin_lock will now account for only 1.2% of the total CPU cycles for the short workload. This patch also has the effect of reducing the effect of running perf on its profile since the perf command itself can be a heavy user of the d_path() function depending on the complexity of the workload. When taking the perf profile of the high-systime workload, the amount of spinlock contention contributed by running perf without this patch was about 16%. With this patch, the spinlock contention caused by the running of perf will go away and we will have a more accurate perf profile. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-10 00:18:13 +08:00
error = prepend_name(&end, &len, &dentry->d_name);
if (error)
break;
retval = end;
dentry = parent;
}
if (!(seq & 1))
rcu_read_unlock();
if (need_seqretry(&rename_lock, seq)) {
seq = 1;
dcache: Translating dentry into pathname without taking rename_lock When running the AIM7's short workload, Linus' lockref patch eliminated most of the spinlock contention. However, there were still some left: 8.46% reaim [kernel.kallsyms] [k] _raw_spin_lock |--42.21%-- d_path | proc_pid_readlink | SyS_readlinkat | SyS_readlink | system_call | __GI___readlink | |--40.97%-- sys_getcwd | system_call | __getcwd The big one here is the rename_lock (seqlock) contention in d_path() and the getcwd system call. This patch will eliminate the need to take the rename_lock while translating dentries into the full pathnames. The need to take the rename_lock is to make sure that no rename operation can be ongoing while the translation is in progress. However, only one thread can take the rename_lock thus blocking all the other threads that need it even though the translation process won't make any change to the dentries. This patch will replace the writer's write_seqlock/write_sequnlock sequence of the rename_lock of the callers of the prepend_path() and __dentry_path() functions with the reader's read_seqbegin/read_seqretry sequence within these 2 functions. As a result, the code will have to retry if one or more rename operations had been performed. In addition, RCU read lock will be taken during the translation process to make sure that no dentries will go away. To prevent live-lock from happening, the code will switch back to take the rename_lock if read_seqretry() fails for three times. To further reduce spinlock contention, this patch does not take the dentry's d_lock when copying the filename from the dentries. Instead, it treats the name pointer and length as unreliable and just copy the string byte-by-byte over until it hits a null byte or the end of string as specified by the length. This should avoid stepping into invalid memory address. The error cases are left to be handled by the sequence number check. The following code re-factoring are also made: 1. Move prepend('/') into prepend_name() to remove one conditional check. 2. Move the global root check in prepend_path() back to the top of the while loop. With this patch, the _raw_spin_lock will now account for only 1.2% of the total CPU cycles for the short workload. This patch also has the effect of reducing the effect of running perf on its profile since the perf command itself can be a heavy user of the d_path() function depending on the complexity of the workload. When taking the perf profile of the high-systime workload, the amount of spinlock contention contributed by running perf without this patch was about 16%. With this patch, the spinlock contention caused by the running of perf will go away and we will have a more accurate perf profile. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-10 00:18:13 +08:00
goto restart;
}
done_seqretry(&rename_lock, seq);
dcache: Translating dentry into pathname without taking rename_lock When running the AIM7's short workload, Linus' lockref patch eliminated most of the spinlock contention. However, there were still some left: 8.46% reaim [kernel.kallsyms] [k] _raw_spin_lock |--42.21%-- d_path | proc_pid_readlink | SyS_readlinkat | SyS_readlink | system_call | __GI___readlink | |--40.97%-- sys_getcwd | system_call | __getcwd The big one here is the rename_lock (seqlock) contention in d_path() and the getcwd system call. This patch will eliminate the need to take the rename_lock while translating dentries into the full pathnames. The need to take the rename_lock is to make sure that no rename operation can be ongoing while the translation is in progress. However, only one thread can take the rename_lock thus blocking all the other threads that need it even though the translation process won't make any change to the dentries. This patch will replace the writer's write_seqlock/write_sequnlock sequence of the rename_lock of the callers of the prepend_path() and __dentry_path() functions with the reader's read_seqbegin/read_seqretry sequence within these 2 functions. As a result, the code will have to retry if one or more rename operations had been performed. In addition, RCU read lock will be taken during the translation process to make sure that no dentries will go away. To prevent live-lock from happening, the code will switch back to take the rename_lock if read_seqretry() fails for three times. To further reduce spinlock contention, this patch does not take the dentry's d_lock when copying the filename from the dentries. Instead, it treats the name pointer and length as unreliable and just copy the string byte-by-byte over until it hits a null byte or the end of string as specified by the length. This should avoid stepping into invalid memory address. The error cases are left to be handled by the sequence number check. The following code re-factoring are also made: 1. Move prepend('/') into prepend_name() to remove one conditional check. 2. Move the global root check in prepend_path() back to the top of the while loop. With this patch, the _raw_spin_lock will now account for only 1.2% of the total CPU cycles for the short workload. This patch also has the effect of reducing the effect of running perf on its profile since the perf command itself can be a heavy user of the d_path() function depending on the complexity of the workload. When taking the perf profile of the high-systime workload, the amount of spinlock contention contributed by running perf without this patch was about 16%. With this patch, the spinlock contention caused by the running of perf will go away and we will have a more accurate perf profile. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-10 00:18:13 +08:00
if (error)
goto Elong;
return retval;
Elong:
return ERR_PTR(-ENAMETOOLONG);
}
char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
{
dcache: Translating dentry into pathname without taking rename_lock When running the AIM7's short workload, Linus' lockref patch eliminated most of the spinlock contention. However, there were still some left: 8.46% reaim [kernel.kallsyms] [k] _raw_spin_lock |--42.21%-- d_path | proc_pid_readlink | SyS_readlinkat | SyS_readlink | system_call | __GI___readlink | |--40.97%-- sys_getcwd | system_call | __getcwd The big one here is the rename_lock (seqlock) contention in d_path() and the getcwd system call. This patch will eliminate the need to take the rename_lock while translating dentries into the full pathnames. The need to take the rename_lock is to make sure that no rename operation can be ongoing while the translation is in progress. However, only one thread can take the rename_lock thus blocking all the other threads that need it even though the translation process won't make any change to the dentries. This patch will replace the writer's write_seqlock/write_sequnlock sequence of the rename_lock of the callers of the prepend_path() and __dentry_path() functions with the reader's read_seqbegin/read_seqretry sequence within these 2 functions. As a result, the code will have to retry if one or more rename operations had been performed. In addition, RCU read lock will be taken during the translation process to make sure that no dentries will go away. To prevent live-lock from happening, the code will switch back to take the rename_lock if read_seqretry() fails for three times. To further reduce spinlock contention, this patch does not take the dentry's d_lock when copying the filename from the dentries. Instead, it treats the name pointer and length as unreliable and just copy the string byte-by-byte over until it hits a null byte or the end of string as specified by the length. This should avoid stepping into invalid memory address. The error cases are left to be handled by the sequence number check. The following code re-factoring are also made: 1. Move prepend('/') into prepend_name() to remove one conditional check. 2. Move the global root check in prepend_path() back to the top of the while loop. With this patch, the _raw_spin_lock will now account for only 1.2% of the total CPU cycles for the short workload. This patch also has the effect of reducing the effect of running perf on its profile since the perf command itself can be a heavy user of the d_path() function depending on the complexity of the workload. When taking the perf profile of the high-systime workload, the amount of spinlock contention contributed by running perf without this patch was about 16%. With this patch, the spinlock contention caused by the running of perf will go away and we will have a more accurate perf profile. Signed-off-by: Waiman Long <Waiman.Long@hp.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2013-09-10 00:18:13 +08:00
return __dentry_path(dentry, buf, buflen);
}
EXPORT_SYMBOL(dentry_path_raw);
char *dentry_path(struct dentry *dentry, char *buf, int buflen)
{
char *p = NULL;
char *retval;
if (d_unlinked(dentry)) {
p = buf + buflen;
if (prepend(&p, &buflen, "//deleted", 10) != 0)
goto Elong;
buflen++;
}
retval = __dentry_path(dentry, buf, buflen);
if (!IS_ERR(retval) && p)
*p = '/'; /* restore '/' overriden with '\0' */
return retval;
Elong:
return ERR_PTR(-ENAMETOOLONG);
}
static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
struct path *pwd)
{
unsigned seq;
do {
seq = read_seqcount_begin(&fs->seq);
*root = fs->root;
*pwd = fs->pwd;
} while (read_seqcount_retry(&fs->seq, seq));
}
/*
* NOTE! The user-level library version returns a
* character pointer. The kernel system call just
* returns the length of the buffer filled (which
* includes the ending '\0' character), or a negative
* error value. So libc would do something like
*
* char *getcwd(char * buf, size_t size)
* {
* int retval;
*
* retval = sys_getcwd(buf, size);
* if (retval >= 0)
* return buf;
* errno = -retval;
* return NULL;
* }
*/
SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
{
int error;
struct path pwd, root;
char *page = __getname();
if (!page)
return -ENOMEM;
rcu_read_lock();
get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
error = -ENOENT;
if (!d_unlinked(pwd.dentry)) {
unsigned long len;
char *cwd = page + PATH_MAX;
int buflen = PATH_MAX;
prepend(&cwd, &buflen, "\0", 1);
fix apparmor dereferencing potentially freed dentry, sanitize __d_path() API __d_path() API is asking for trouble and in case of apparmor d_namespace_path() getting just that. The root cause is that when __d_path() misses the root it had been told to look for, it stores the location of the most remote ancestor in *root. Without grabbing references. Sure, at the moment of call it had been pinned down by what we have in *path. And if we raced with umount -l, we could have very well stopped at vfsmount/dentry that got freed as soon as prepend_path() dropped vfsmount_lock. It is safe to compare these pointers with pre-existing (and known to be still alive) vfsmount and dentry, as long as all we are asking is "is it the same address?". Dereferencing is not safe and apparmor ended up stepping into that. d_namespace_path() really wants to examine the place where we stopped, even if it's not connected to our namespace. As the result, it looked at ->d_sb->s_magic of a dentry that might've been already freed by that point. All other callers had been careful enough to avoid that, but it's really a bad interface - it invites that kind of trouble. The fix is fairly straightforward, even though it's bigger than I'd like: * prepend_path() root argument becomes const. * __d_path() is never called with NULL/NULL root. It was a kludge to start with. Instead, we have an explicit function - d_absolute_root(). Same as __d_path(), except that it doesn't get root passed and stops where it stops. apparmor and tomoyo are using it. * __d_path() returns NULL on path outside of root. The main caller is show_mountinfo() and that's precisely what we pass root for - to skip those outside chroot jail. Those who don't want that can (and do) use d_path(). * __d_path() root argument becomes const. Everyone agrees, I hope. * apparmor does *NOT* try to use __d_path() or any of its variants when it sees that path->mnt is an internal vfsmount. In that case it's definitely not mounted anywhere and dentry_path() is exactly what we want there. Handling of sysctl()-triggered weirdness is moved to that place. * if apparmor is asked to do pathname relative to chroot jail and __d_path() tells it we it's not in that jail, the sucker just calls d_absolute_path() instead. That's the other remaining caller of __d_path(), BTW. * seq_path_root() does _NOT_ return -ENAMETOOLONG (it's stupid anyway - the normal seq_file logics will take care of growing the buffer and redoing the call of ->show() just fine). However, if it gets path not reachable from root, it returns SEQ_SKIP. The only caller adjusted (i.e. stopped ignoring the return value as it used to do). Reviewed-by: John Johansen <john.johansen@canonical.com> ACKed-by: John Johansen <john.johansen@canonical.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Cc: stable@vger.kernel.org
2011-12-05 21:43:34 +08:00
error = prepend_path(&pwd, &root, &cwd, &buflen);
rcu_read_unlock();
fix apparmor dereferencing potentially freed dentry, sanitize __d_path() API __d_path() API is asking for trouble and in case of apparmor d_namespace_path() getting just that. The root cause is that when __d_path() misses the root it had been told to look for, it stores the location of the most remote ancestor in *root. Without grabbing references. Sure, at the moment of call it had been pinned down by what we have in *path. And if we raced with umount -l, we could have very well stopped at vfsmount/dentry that got freed as soon as prepend_path() dropped vfsmount_lock. It is safe to compare these pointers with pre-existing (and known to be still alive) vfsmount and dentry, as long as all we are asking is "is it the same address?". Dereferencing is not safe and apparmor ended up stepping into that. d_namespace_path() really wants to examine the place where we stopped, even if it's not connected to our namespace. As the result, it looked at ->d_sb->s_magic of a dentry that might've been already freed by that point. All other callers had been careful enough to avoid that, but it's really a bad interface - it invites that kind of trouble. The fix is fairly straightforward, even though it's bigger than I'd like: * prepend_path() root argument becomes const. * __d_path() is never called with NULL/NULL root. It was a kludge to start with. Instead, we have an explicit function - d_absolute_root(). Same as __d_path(), except that it doesn't get root passed and stops where it stops. apparmor and tomoyo are using it. * __d_path() returns NULL on path outside of root. The main caller is show_mountinfo() and that's precisely what we pass root for - to skip those outside chroot jail. Those who don't want that can (and do) use d_path(). * __d_path() root argument becomes const. Everyone agrees, I hope. * apparmor does *NOT* try to use __d_path() or any of its variants when it sees that path->mnt is an internal vfsmount. In that case it's definitely not mounted anywhere and dentry_path() is exactly what we want there. Handling of sysctl()-triggered weirdness is moved to that place. * if apparmor is asked to do pathname relative to chroot jail and __d_path() tells it we it's not in that jail, the sucker just calls d_absolute_path() instead. That's the other remaining caller of __d_path(), BTW. * seq_path_root() does _NOT_ return -ENAMETOOLONG (it's stupid anyway - the normal seq_file logics will take care of growing the buffer and redoing the call of ->show() just fine). However, if it gets path not reachable from root, it returns SEQ_SKIP. The only caller adjusted (i.e. stopped ignoring the return value as it used to do). Reviewed-by: John Johansen <john.johansen@canonical.com> ACKed-by: John Johansen <john.johansen@canonical.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Cc: stable@vger.kernel.org
2011-12-05 21:43:34 +08:00
if (error < 0)
goto out;
/* Unreachable from current root */
fix apparmor dereferencing potentially freed dentry, sanitize __d_path() API __d_path() API is asking for trouble and in case of apparmor d_namespace_path() getting just that. The root cause is that when __d_path() misses the root it had been told to look for, it stores the location of the most remote ancestor in *root. Without grabbing references. Sure, at the moment of call it had been pinned down by what we have in *path. And if we raced with umount -l, we could have very well stopped at vfsmount/dentry that got freed as soon as prepend_path() dropped vfsmount_lock. It is safe to compare these pointers with pre-existing (and known to be still alive) vfsmount and dentry, as long as all we are asking is "is it the same address?". Dereferencing is not safe and apparmor ended up stepping into that. d_namespace_path() really wants to examine the place where we stopped, even if it's not connected to our namespace. As the result, it looked at ->d_sb->s_magic of a dentry that might've been already freed by that point. All other callers had been careful enough to avoid that, but it's really a bad interface - it invites that kind of trouble. The fix is fairly straightforward, even though it's bigger than I'd like: * prepend_path() root argument becomes const. * __d_path() is never called with NULL/NULL root. It was a kludge to start with. Instead, we have an explicit function - d_absolute_root(). Same as __d_path(), except that it doesn't get root passed and stops where it stops. apparmor and tomoyo are using it. * __d_path() returns NULL on path outside of root. The main caller is show_mountinfo() and that's precisely what we pass root for - to skip those outside chroot jail. Those who don't want that can (and do) use d_path(). * __d_path() root argument becomes const. Everyone agrees, I hope. * apparmor does *NOT* try to use __d_path() or any of its variants when it sees that path->mnt is an internal vfsmount. In that case it's definitely not mounted anywhere and dentry_path() is exactly what we want there. Handling of sysctl()-triggered weirdness is moved to that place. * if apparmor is asked to do pathname relative to chroot jail and __d_path() tells it we it's not in that jail, the sucker just calls d_absolute_path() instead. That's the other remaining caller of __d_path(), BTW. * seq_path_root() does _NOT_ return -ENAMETOOLONG (it's stupid anyway - the normal seq_file logics will take care of growing the buffer and redoing the call of ->show() just fine). However, if it gets path not reachable from root, it returns SEQ_SKIP. The only caller adjusted (i.e. stopped ignoring the return value as it used to do). Reviewed-by: John Johansen <john.johansen@canonical.com> ACKed-by: John Johansen <john.johansen@canonical.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Cc: stable@vger.kernel.org
2011-12-05 21:43:34 +08:00
if (error > 0) {
error = prepend_unreachable(&cwd, &buflen);
if (error)
goto out;
}
error = -ERANGE;
len = PATH_MAX + page - cwd;
if (len <= size) {
error = len;
if (copy_to_user(buf, cwd, len))
error = -EFAULT;
}
} else {
rcu_read_unlock();
}
out:
__putname(page);
return error;
}
/*
* Test whether new_dentry is a subdirectory of old_dentry.
*
* Trivially implemented using the dcache structure
*/
/**
* is_subdir - is new dentry a subdirectory of old_dentry
* @new_dentry: new dentry
* @old_dentry: old dentry
*
* Returns true if new_dentry is a subdirectory of the parent (at any depth).
* Returns false otherwise.
* Caller must ensure that "new_dentry" is pinned before calling is_subdir()
*/
bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
{
bool result;
unsigned seq;
if (new_dentry == old_dentry)
return true;
do {
/* for restarting inner loop in case of seq retry */
seq = read_seqbegin(&rename_lock);
/*
* Need rcu_readlock to protect against the d_parent trashing
* due to d_move
*/
rcu_read_lock();
if (d_ancestor(old_dentry, new_dentry))
result = true;
else
result = false;
rcu_read_unlock();
} while (read_seqretry(&rename_lock, seq));
return result;
}
static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
{
struct dentry *root = data;
if (dentry != root) {
if (d_unhashed(dentry) || !dentry->d_inode)
return D_WALK_SKIP;
if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
dentry->d_flags |= DCACHE_GENOCIDE;
dentry->d_lockref.count--;
}
}
return D_WALK_CONTINUE;
}
void d_genocide(struct dentry *parent)
{
d_walk(parent, parent, d_genocide_kill, NULL);
}
void d_tmpfile(struct dentry *dentry, struct inode *inode)
{
inode_dec_link_count(inode);
BUG_ON(dentry->d_name.name != dentry->d_iname ||
!hlist_unhashed(&dentry->d_u.d_alias) ||
!d_unlinked(dentry));
spin_lock(&dentry->d_parent->d_lock);
spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
(unsigned long long)inode->i_ino);
spin_unlock(&dentry->d_lock);
spin_unlock(&dentry->d_parent->d_lock);
d_instantiate(dentry, inode);
}
EXPORT_SYMBOL(d_tmpfile);
static __initdata unsigned long dhash_entries;
static int __init set_dhash_entries(char *str)
{
if (!str)
return 0;
dhash_entries = simple_strtoul(str, &str, 0);
return 1;
}
__setup("dhash_entries=", set_dhash_entries);
static void __init dcache_init_early(void)
{
unsigned int loop;
/* If hashes are distributed across NUMA nodes, defer
* hash allocation until vmalloc space is available.
*/
if (hashdist)
return;
dentry_hashtable =
alloc_large_system_hash("Dentry cache",
sizeof(struct hlist_bl_head),
dhash_entries,
13,
HASH_EARLY,
&d_hash_shift,
&d_hash_mask,
0,
0);
for (loop = 0; loop < (1U << d_hash_shift); loop++)
INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
}
static void __init dcache_init(void)
{
unsigned int loop;
/*
* A constructor could be added for stable state like the lists,
* but it is probably not worth it because of the cache nature
* of the dcache.
*/
dentry_cache = KMEM_CACHE(dentry,
2016-01-15 07:18:21 +08:00
SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT);
/* Hash may have been set up in dcache_init_early */
if (!hashdist)
return;
dentry_hashtable =
alloc_large_system_hash("Dentry cache",
sizeof(struct hlist_bl_head),
dhash_entries,
13,
0,
&d_hash_shift,
&d_hash_mask,
0,
0);
for (loop = 0; loop < (1U << d_hash_shift); loop++)
INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
}
/* SLAB cache for __getname() consumers */
struct kmem_cache *names_cachep __read_mostly;
EXPORT_SYMBOL(names_cachep);
EXPORT_SYMBOL(d_genocide);
void __init vfs_caches_init_early(void)
{
dcache_init_early();
inode_init_early();
}
void __init vfs_caches_init(void)
{
names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
dcache_init();
inode_init();
files_init();
files_maxfiles_init();
mnt_init();
bdev_cache_init();
chrdev_init();
}