Now that real_root is only used in ref-verify core gate it behind
CONFIG_BTRFS_FS_REF_VERIFY ifdef. This shrinks the size of pending
delayed refs by 8 bytes per ref, of which we can have many at any one
time depending on intensity of the workload. Also change the comment
about the member as it no longer deals with qgroups.
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Instead of checking whether qgroup processing for a dealyed ref has to
happen in the core of delayed ref, simply pull the check at init time of
respective delayed ref structures. This eliminates the final use of
real_root in delayed-ref core paving the way to making this member
optional.
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
In order to make 'real_root' used only in ref-verify it's required to
have the necessary context to perform the same checks that this member
is used for. So add 'mod_root' which will contain the root on behalf of
which a delayed ref was created and a 'skip_group' parameter which
will contain callsite-specific override of skip_qgroup.
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Both data and metadata delayed ref structures have fields named
root/ref_root respectively. Those are somewhat cryptic and don't really
convey the real meaning. In fact those roots are really the original
owners of the respective block (i.e in case of a snapshot a data delayed
ref will contain the original root that owns the given block). Rename
those fields accordingly and adjust comments.
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
I've been running a stress test that runs 20 workers in their own
subvolume, which are running an fsstress instance with 4 threads per
worker, which is 80 total fsstress threads. In addition to this I'm
running balance in the background as well as creating and deleting
snapshots. This test takes around 12 hours to run normally, going
slower and slower as the test goes on.
The reason for this is because fsstress is running fsync sometimes, and
because we're messing with block groups we often fall through to
btrfs_commit_transaction, so will often have 20-30 threads all calling
btrfs_commit_transaction at the same time.
These all get stuck contending on the extent tree while they try to run
delayed refs during the initial part of the commit.
This is suboptimal, really because the extent tree is a single point of
failure we only want one thread acting on that tree at once to reduce
lock contention.
Fix this by making the flushing mechanism a bit operation, to make it
easy to use test_and_set_bit() in order to make sure only one task does
this initial flush.
Once we're into the transaction commit we only have one thread doing
delayed ref running, it's just this initial pre-flush that is
problematic. With this patch my stress test takes around 90 minutes to
run, instead of 12 hours.
The memory barrier is not necessary for the flushing bit as it's
ordered, unlike plain int. The transaction state accessed in
btrfs_should_end_transaction could be affected by that too as it's not
always used under transaction lock. Upon Nikolay's analysis in [1]
it's not necessary:
In should_end_transaction it's read without holding any locks. (U)
It's modified in btrfs_cleanup_transaction without holding the
fs_info->trans_lock (U), but the STATE_ERROR flag is going to be set.
set in cleanup_transaction under fs_info->trans_lock (L)
set in btrfs_commit_trans to COMMIT_START under fs_info->trans_lock.(L)
set in btrfs_commit_trans to COMMIT_DOING under fs_info->trans_lock.(L)
set in btrfs_commit_trans to COMMIT_UNBLOCK under
fs_info->trans_lock.(L)
set in btrfs_commit_trans to COMMIT_COMPLETED without locks but at this
point the transaction is finished and fs_info->running_trans is NULL (U
but irrelevant).
So by the looks of it we can have a concurrent READ race with a WRITE,
due to reads not taking a lock. In this case what we want to ensure is
we either see new or old state. I consulted with Will Deacon and he said
that in such a case we'd want to annotate the accesses to ->state with
(READ|WRITE)_ONCE so as to avoid a theoretical tear, in this case I
don't think this could happen but I imagine at some point KCSAN would
flag such an access as racy (which it is).
[1] https://lore.kernel.org/linux-btrfs/e1fd5cc1-0f28-f670-69f4-e9958b4964e6@suse.com
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
[ add comments regarding memory barrier ]
Signed-off-by: David Sterba <dsterba@suse.com>
Currently we pass things around to figure out if we maybe freeing data
based on the state of the delayed refs head. This makes the accounting
sort of confusing and hard to follow, as it's distinctly separate from
the delayed ref heads stuff, but also depends on it entirely.
Fix this by explicitly adjusting the space_info->total_bytes_pinned in
the delayed refs code. We now have two places where we modify this
counter, once where we create the delayed and destroy the delayed refs,
and once when we pin and unpin the extents. This means there is a
slight overlap between delayed refs and the pin/unpin mechanisms, but
this is simply used by the ENOSPC infrastructure to determine if we need
to commit the transaction, so there's no adverse affect from this, we
might simply commit thinking it will give us enough space when it might
not.
CC: stable@vger.kernel.org # 5.10
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
These belong with the delayed refs related code, not in extent-tree.c.
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Just like btrfs_add_delayed_tree_ref(), use btrfs_ref to refactor
btrfs_add_delayed_data_ref().
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
btrfs_add_delayed_tree_ref() has a longer and longer parameter list, and
some callers like btrfs_inc_extent_ref() are using @owner as level for
delayed tree ref.
Instead of making the parameter list longer, use btrfs_ref to refactor
it, so each parameter assignment should be self-explaining without dirty
level/owner trick, and provides the basis for later refactoring.
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Current delayed ref interface has several problems:
- Longer and longer parameter lists
bytenr
num_bytes
parent
---------- so far so good
ref_root
owner
offset
---------- I don't feel good now
- Different interpretation of the same parameter
Above @owner for data ref is inode number (u64),
while for tree ref, it's level (int).
They are even in different size range.
For level we only need 0 ~ 8, while for ino it's
BTRFS_FIRST_FREE_OBJECTID ~ BTRFS_LAST_FREE_OBJECTID.
And @offset doesn't even make sense for tree ref.
Such parameter reuse may look clever as an hidden union, but it
destroys code readability.
To solve both problems, we introduce a new structure, btrfs_ref to solve
them:
- Structure instead of long parameter list
This makes later expansion easier, and is better documented.
- Use btrfs_ref::type to distinguish data and tree ref
- Use proper union to store data/tree ref specific structures.
- Use separate functions to fill data/tree ref data, with a common generic
function to fill common bytenr/num_bytes members.
All parameters will find its place in btrfs_ref, and an extra member,
@real_root, inspired by ref-verify code, is newly introduced for later
qgroup code, to record which tree is triggered by this extent modification.
This patch doesn't touch any code, but provides the basis for further
refactoring.
Signed-off-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
[BUG]
Btrfs/139 will fail with a high probability if the testing machine (VM)
has only 2G RAM.
Resulting the final write success while it should fail due to EDQUOT,
and the fs will have quota exceeding the limit by 16K.
The simplified reproducer will be: (needs a 2G ram VM)
$ mkfs.btrfs -f $dev
$ mount $dev $mnt
$ btrfs subv create $mnt/subv
$ btrfs quota enable $mnt
$ btrfs quota rescan -w $mnt
$ btrfs qgroup limit -e 1G $mnt/subv
$ for i in $(seq -w 1 8); do
xfs_io -f -c "pwrite 0 128M" $mnt/subv/file_$i > /dev/null
echo "file $i written" > /dev/kmsg
done
$ sync
$ btrfs qgroup show -pcre --raw $mnt
The last pwrite will not trigger EDQUOT and final 'qgroup show' will
show something like:
qgroupid rfer excl max_rfer max_excl parent child
-------- ---- ---- -------- -------- ------ -----
0/5 16384 16384 none none --- ---
0/256 1073758208 1073758208 none 1073741824 --- ---
And 1073758208 is larger than
> 1073741824.
[CAUSE]
It's a bug in btrfs qgroup data reserved space management.
For quota limit, we must ensure that:
reserved (data + metadata) + rfer/excl <= limit
Since rfer/excl is only updated at transaction commmit time, reserved
space needs to be taken special care.
One important part of reserved space is data, and for a new data extent
written to disk, we still need to take the reserved space until
rfer/excl numbers get updated.
Originally when an ordered extent finishes, we migrate the reserved
qgroup data space from extent_io tree to delayed ref head of the data
extent, expecting delayed ref will only be cleaned up at commit
transaction time.
However for small RAM machine, due to memory pressure dirty pages can be
flushed back to disk without committing a transaction.
The related events will be something like:
file 1 written
btrfs_finish_ordered_io: ino=258 ordered offset=0 len=54947840
btrfs_finish_ordered_io: ino=258 ordered offset=54947840 len=5636096
btrfs_finish_ordered_io: ino=258 ordered offset=61153280 len=57344
btrfs_finish_ordered_io: ino=258 ordered offset=61210624 len=8192
btrfs_finish_ordered_io: ino=258 ordered offset=60583936 len=569344
cleanup_ref_head: num_bytes=54947840
cleanup_ref_head: num_bytes=5636096
cleanup_ref_head: num_bytes=569344
cleanup_ref_head: num_bytes=57344
cleanup_ref_head: num_bytes=8192
^^^^^^^^^^^^^^^^ This will free qgroup data reserved space
file 2 written
...
file 8 written
cleanup_ref_head: num_bytes=8192
...
btrfs_commit_transaction <<< the only transaction committed during
the test
When file 2 is written, we have already freed 128M reserved qgroup data
space for ino 258. Thus later write won't trigger EDQUOT.
This allows us to write more data beyond qgroup limit.
In my 2G ram VM, it could reach about 1.2G before hitting EDQUOT.
[FIX]
By moving reserved qgroup data space from btrfs_delayed_ref_head to
btrfs_qgroup_extent_record, we can ensure that reserved qgroup data
space won't be freed half way before commit transaction, thus fix the
problem.
Fixes: f64d5ca868 ("btrfs: delayed_ref: Add new function to record reserved space into delayed ref")
Signed-off-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We do this dance in cleanup_ref_head and check_ref_cleanup, unify it
into a helper and cleanup the calling functions.
Reviewed-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Since trans is only used for referring to delayed_refs, there is no need
to pass it instead of delayed_refs to btrfs_delayed_ref_lock().
No functional change.
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Lu Fengqi <lufq.fnst@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Since trans is only used for referring to delayed_refs, there is no need
to pass it instead of delayed_refs to btrfs_select_ref_head(). No
functional change.
Signed-off-by: Lu Fengqi <lufq.fnst@cn.fujitsu.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
rb_first_cached() trades an extra pointer "leftmost" for doing the same
job as rb_first() but in O(1).
Functions manipulating href->ref_tree need to get the first entry, this
converts href->ref_tree to use rb_first_cached().
For more details about the optimization see patch "Btrfs: delayed-refs:
use rb_first_cached for href_root".
Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com>
Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
rb_first_cached() trades an extra pointer "leftmost" for doing the same
job as rb_first() but in O(1).
Functions manipulating href_root need to get the first entry, this
converts href_root to use rb_first_cached().
This patch is first in the sequenct of similar updates to other rbtrees
and this is analysis of the expected behaviour and improvements.
There's a common pattern:
while (node = rb_first) {
entry = rb_entry(node)
next = rb_next(node)
rb_erase(node)
cleanup(entry)
}
rb_first needs to traverse the tree up to logN depth, rb_erase can
completely reshuffle the tree. With the caching we'll skip the traversal
in rb_first. That's a cached memory access vs looped pointer
dereference trade-off that IMHO has a clear winner.
Measurements show there's not much difference in a sample tree with
10000 nodes: 4.5s / rb_first and 4.8s / rb_first_cached. Real effects of
caching and pointer chasing are unpredictable though.
Further optimzations can be done to avoid the expensive rb_erase step.
In some cases it's ok to process the nodes in any order, so the tree can
be traversed in post-order, not rebalancing the children nodes and just
calling free. Care must be taken regarding the next node.
Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com>
Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ update changelog from mail discussions ]
Signed-off-by: David Sterba <dsterba@suse.com>
This function is always called with a valid transaction handle from
where fs_info can be referenced. No functional changes.
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
This function is always called with a valid transaction handle from
where fs_info can be referenced. No functional changes.
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
It's used to print its pointer in a debug statement but doesn't really
bring any useful information to the error message.
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
When the delayed refs for a head are all run, eventually
cleanup_ref_head is called which (in case of deletion) obtains a
reference for the relevant btrfs_space_info struct by querying the bg
for the range. This is problematic because when the last extent of a
bg is deleted a race window emerges between removal of that bg and the
subsequent invocation of cleanup_ref_head. This can result in cache being null
and either a null pointer dereference or assertion failure.
task: ffff8d04d31ed080 task.stack: ffff9e5dc10cc000
RIP: 0010:assfail.constprop.78+0x18/0x1a [btrfs]
RSP: 0018:ffff9e5dc10cfbe8 EFLAGS: 00010292
RAX: 0000000000000044 RBX: 0000000000000000 RCX: 0000000000000000
RDX: ffff8d04ffc1f868 RSI: ffff8d04ffc178c8 RDI: ffff8d04ffc178c8
RBP: ffff8d04d29e5ea0 R08: 00000000000001f0 R09: 0000000000000001
R10: ffff9e5dc0507d58 R11: 0000000000000001 R12: ffff8d04d29e5ea0
R13: ffff8d04d29e5f08 R14: ffff8d04efe29b40 R15: ffff8d04efe203e0
FS: 00007fbf58ead500(0000) GS:ffff8d04ffc00000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00007fe6c6975648 CR3: 0000000013b2a000 CR4: 00000000000006f0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
__btrfs_run_delayed_refs+0x10e7/0x12c0 [btrfs]
btrfs_run_delayed_refs+0x68/0x250 [btrfs]
btrfs_should_end_transaction+0x42/0x60 [btrfs]
btrfs_truncate_inode_items+0xaac/0xfc0 [btrfs]
btrfs_evict_inode+0x4c6/0x5c0 [btrfs]
evict+0xc6/0x190
do_unlinkat+0x19c/0x300
do_syscall_64+0x74/0x140
entry_SYSCALL_64_after_hwframe+0x3d/0xa2
RIP: 0033:0x7fbf589c57a7
To fix this, introduce a new flag "is_system" to head_ref structs,
which is populated at insertion time. This allows to decouple the
querying for the spaceinfo from querying the possibly deleted bg.
Fixes: d7eae3403f ("Btrfs: rework delayed ref total_bytes_pinned accounting")
CC: stable@vger.kernel.org # 4.14+
Suggested-by: Omar Sandoval <osandov@osandov.com>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Remove GPL boilerplate text (long, short, one-line) and keep the rest,
ie. personal, company or original source copyright statements. Add the
SPDX header.
Unify the include protection macros to match the file names.
Signed-off-by: David Sterba <dsterba@suse.com>
The __cold functions are placed to a special section, as they're
expected to be called rarely. This could help i-cache prefetches or help
compiler to decide which branches are more/less likely to be taken
without any other annotations needed.
Though we can't add more __exit annotations, it's still possible to add
__cold (that's also added with __exit). That way the following function
categories are tagged:
- printf wrappers, error messages
- exit helpers
Signed-off-by: David Sterba <dsterba@suse.com>
Adding __init macro gives kernel a hint that this function is only used
during the initialization phase and its memory resources can be freed up
after.
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
If we get a significant amount of delayed refs for a single block (think
modifying multiple snapshots) we can end up spending an ungodly amount
of time looping through all of the entries trying to see if they can be
merged. This is because we only add them to a list, so we have O(2n)
for every ref head. This doesn't make any sense as we likely have refs
for different roots, and so they cannot be merged. Tracking in a tree
will allow us to break as soon as we hit an entry that doesn't match,
making our worst case O(n).
With this we can also merge entries more easily. Before we had to hope
that matching refs were on the ends of our list, but with the tree we
can search down to exact matches and merge them at insert time.
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
This is just excessive information in the ref_head, and makes the code
complicated. It is a relic from when we had the heads and the refs in
the same tree, which is no longer the case. With this removal I've
cleaned up a bunch of the cruft around this old assumption as well.
Signed-off-by: Josef Bacik <jbacik@fb.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We need this to decide when to account pinned bytes.
Signed-off-by: Omar Sandoval <osandov@fb.com>
Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com>
Signed-off-by: David Sterba <dsterba@suse.com>
refcount_t type and corresponding API should be
used instead of atomic_t when the variable is used as
a reference counter. This allows to avoid accidental
refcounter overflows that might lead to use-after-free
situations.
Signed-off-by: Elena Reshetova <elena.reshetova@intel.com>
Signed-off-by: Hans Liljestrand <ishkamiel@gmail.com>
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: David Windsor <dwindsor@gmail.com>
Signed-off-by: David Sterba <dsterba@suse.com>
All we need is @delayed_refs, all callers have get it ahead of calling
btrfs_find_delayed_ref_head since lock needs to be acquired firstly,
there is no reason to deference it again inside the function.
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
btrfs_add_delayed_data_ref is always called with a NULL extent_op,
so let's drop the argument.
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Patches queued up by Filipe:
The most important change is still the fix for the extent tree
corruption that happens due to balance when qgroups are enabled (a
regression introduced in 4.7 by a fix for a regression from the last
qgroups rework). This has been hitting SLE and openSUSE users and QA
very badly, where transactions keep getting aborted when running
delayed references leaving the root filesystem in RO mode and nearly
unusable. There are fixes here that allow us to run xfstests again
with the integrity checker enabled, which has been impossible since 4.8
(apparently I'm the only one running xfstests with the integrity
checker enabled, which is useful to validate dirtied leafs, like
checking if there are keys out of order, etc). The rest are just some
trivial fixes, most of them tagged for stable, and two cleanups.
Signed-off-by: Chris Mason <clm@fb.com>
This issue was found when I tried to delete a heavily reflinked file,
when deleting such files, other transaction operation will not have a
chance to make progress, for example, start_transaction() will blocked
in wait_current_trans(root) for long time, sometimes it even triggers
soft lockups, and the time taken to delete such heavily reflinked file
is also very large, often hundreds of seconds. Using perf top, it reports
that:
PerfTop: 7416 irqs/sec kernel:99.8% exact: 0.0% [4000Hz cpu-clock], (all, 4 CPUs)
---------------------------------------------------------------------------------------
84.37% [btrfs] [k] __btrfs_run_delayed_refs.constprop.80
11.02% [kernel] [k] delay_tsc
0.79% [kernel] [k] _raw_spin_unlock_irq
0.78% [kernel] [k] _raw_spin_unlock_irqrestore
0.45% [kernel] [k] do_raw_spin_lock
0.18% [kernel] [k] __slab_alloc
It seems __btrfs_run_delayed_refs() took most cpu time, after some debug
work, I found it's select_delayed_ref() causing this issue, for a delayed
head, in our case, it'll be full of BTRFS_DROP_DELAYED_REF nodes, but
select_delayed_ref() will firstly try to iterate node list to find
BTRFS_ADD_DELAYED_REF nodes, obviously it's a disaster in this case, and
waste much time.
To fix this issue, we introduce a new ref_add_list in struct btrfs_delayed_ref_head,
then in select_delayed_ref(), if this list is not empty, we can directly use
nodes in this list. With this patch, it just took about 10~15 seconds to
delte the same file. Now using perf top, it reports that:
PerfTop: 2734 irqs/sec kernel:99.5% exact: 0.0% [4000Hz cpu-clock], (all, 4 CPUs)
----------------------------------------------------------------------------------------
20.74% [kernel] [k] _raw_spin_unlock_irqrestore
16.33% [kernel] [k] __slab_alloc
5.41% [kernel] [k] lock_acquired
4.42% [kernel] [k] lock_acquire
4.05% [kernel] [k] lock_release
3.37% [kernel] [k] _raw_spin_unlock_irq
For normal files, this patch also gives help, at least we do not need to
iterate whole list to found BTRFS_ADD_DELAYED_REF nodes.
Signed-off-by: Wang Xiaoguang <wangxg.fnst@cn.fujitsu.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com>
Signed-off-by: David Sterba <dsterba@suse.com>
After the last big change in the delayed references code that was needed
for the last qgroups rework, the red black tree node field of struct
btrfs_delayed_ref_node is no longer used, so just remove it, this helps
us save some memory (since struct rb_node is 24 bytes on x86_64) for
these structures.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
No longer used as of commit 5846a3c268 ("btrfs: qgroup: Fix a race in
delayed_ref which leads to abort trans").
Signed-off-by: Filipe Manana <fdmanana@suse.com>
btrfs_delayed_extent_op can be packed in a better way, it's 40 bytes now
and has 8 unused bytes. Reducing the level type to u8 makes it possible
to squeeze it to the padding byte after key. The bitfields were switched
to bool as there's space to store the full byte without increasing the
whole structure, besides that the generated assembly is smaller.
struct btrfs_delayed_extent_op {
struct btrfs_disk_key key; /* 0 17 */
u8 level; /* 17 1 */
bool update_key; /* 18 1 */
bool update_flags; /* 19 1 */
bool is_data; /* 20 1 */
/* XXX 3 bytes hole, try to pack */
u64 flags_to_set; /* 24 8 */
/* size: 32, cachelines: 1, members: 6 */
/* sum members: 29, holes: 1, sum holes: 3 */
/* last cacheline: 32 bytes */
};
The final size is 32 bytes which gives +26 object per slab page.
text data bss dec hex filename
938811 43670 23144 1005625 f5839 fs/btrfs/btrfs.ko.before
938747 43670 23144 1005561 f57f9 fs/btrfs/btrfs.ko.after
Signed-off-by: David Sterba <dsterba@suse.com>
Between btrfs_allocerved_file_extent() and
btrfs_add_delayed_qgroup_reserve(), there is a window that delayed_refs
are run and delayed ref head maybe freed before
btrfs_add_delayed_qgroup_reserve().
This will cause btrfs_dad_delayed_qgroup_reserve() to return -ENOENT,
and cause transaction to be aborted.
This patch will record qgroup reserve space info into delayed_ref_head
at btrfs_add_delayed_ref(), to eliminate the race window.
Reported-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
In the kernel 4.2 merge window we had a big changes to the implementation
of delayed references and qgroups which made the no_quota field of delayed
references not used anymore. More specifically the no_quota field is not
used anymore as of:
commit 0ed4792af0 ("btrfs: qgroup: Switch to new extent-oriented qgroup mechanism.")
Leaving the no_quota field actually prevents delayed references from
getting merged, which in turn cause the following BUG_ON(), at
fs/btrfs/extent-tree.c, to be hit when qgroups are enabled:
static int run_delayed_tree_ref(...)
{
(...)
BUG_ON(node->ref_mod != 1);
(...)
}
This happens on a scenario like the following:
1) Ref1 bytenr X, action = BTRFS_ADD_DELAYED_REF, no_quota = 1, added.
2) Ref2 bytenr X, action = BTRFS_DROP_DELAYED_REF, no_quota = 0, added.
It's not merged with Ref1 because Ref1->no_quota != Ref2->no_quota.
3) Ref3 bytenr X, action = BTRFS_ADD_DELAYED_REF, no_quota = 1, added.
It's not merged with the reference at the tail of the list of refs
for bytenr X because the reference at the tail, Ref2 is incompatible
due to Ref2->no_quota != Ref3->no_quota.
4) Ref4 bytenr X, action = BTRFS_DROP_DELAYED_REF, no_quota = 0, added.
It's not merged with the reference at the tail of the list of refs
for bytenr X because the reference at the tail, Ref3 is incompatible
due to Ref3->no_quota != Ref4->no_quota.
5) We run delayed references, trigger merging of delayed references,
through __btrfs_run_delayed_refs() -> btrfs_merge_delayed_refs().
6) Ref1 and Ref3 are merged as Ref1->no_quota = Ref3->no_quota and
all other conditions are satisfied too. So Ref1 gets a ref_mod
value of 2.
7) Ref2 and Ref4 are merged as Ref2->no_quota = Ref4->no_quota and
all other conditions are satisfied too. So Ref2 gets a ref_mod
value of 2.
8) Ref1 and Ref2 aren't merged, because they have different values
for their no_quota field.
9) Delayed reference Ref1 is picked for running (select_delayed_ref()
always prefers references with an action == BTRFS_ADD_DELAYED_REF).
So run_delayed_tree_ref() is called for Ref1 which triggers the
BUG_ON because Ref1->red_mod != 1 (equals 2).
So fix this by removing the no_quota field, as it's not used anymore as
of commit 0ed4792af0 ("btrfs: qgroup: Switch to new extent-oriented
qgroup mechanism.").
The use of no_quota was also buggy in at least two places:
1) At delayed-refs.c:btrfs_add_delayed_tree_ref() - we were setting
no_quota to 0 instead of 1 when the following condition was true:
is_fstree(ref_root) || !fs_info->quota_enabled
2) At extent-tree.c:__btrfs_inc_extent_ref() - we were attempting to
reset a node's no_quota when the condition "!is_fstree(root_objectid)
|| !root->fs_info->quota_enabled" was true but we did it only in
an unused local stack variable, that is, we never reset the no_quota
value in the node itself.
This fixes the remainder of problems several people have been having when
running delayed references, mostly while a balance is running in parallel,
on a 4.2+ kernel.
Very special thanks to Stéphane Lesimple for helping debugging this issue
and testing this fix on his multi terabyte filesystem (which took more
than one day to balance alone, plus fsck, etc).
Also, this fixes deadlock issue when using the clone ioctl with qgroups
enabled, as reported by Elias Probst in the mailing list. The deadlock
happens because after calling btrfs_insert_empty_item we have our path
holding a write lock on a leaf of the fs/subvol tree and then before
releasing the path we called check_ref() which did backref walking, when
qgroups are enabled, and tried to read lock the same leaf. The trace for
this case is the following:
INFO: task systemd-nspawn:6095 blocked for more than 120 seconds.
(...)
Call Trace:
[<ffffffff86999201>] schedule+0x74/0x83
[<ffffffff863ef64c>] btrfs_tree_read_lock+0xc0/0xea
[<ffffffff86137ed7>] ? wait_woken+0x74/0x74
[<ffffffff8639f0a7>] btrfs_search_old_slot+0x51a/0x810
[<ffffffff863a129b>] btrfs_next_old_leaf+0xdf/0x3ce
[<ffffffff86413a00>] ? ulist_add_merge+0x1b/0x127
[<ffffffff86411688>] __resolve_indirect_refs+0x62a/0x667
[<ffffffff863ef546>] ? btrfs_clear_lock_blocking_rw+0x78/0xbe
[<ffffffff864122d3>] find_parent_nodes+0xaf3/0xfc6
[<ffffffff86412838>] __btrfs_find_all_roots+0x92/0xf0
[<ffffffff864128f2>] btrfs_find_all_roots+0x45/0x65
[<ffffffff8639a75b>] ? btrfs_get_tree_mod_seq+0x2b/0x88
[<ffffffff863e852e>] check_ref+0x64/0xc4
[<ffffffff863e9e01>] btrfs_clone+0x66e/0xb5d
[<ffffffff863ea77f>] btrfs_ioctl_clone+0x48f/0x5bb
[<ffffffff86048a68>] ? native_sched_clock+0x28/0x77
[<ffffffff863ed9b0>] btrfs_ioctl+0xabc/0x25cb
(...)
The problem goes away by eleminating check_ref(), which no longer is
needed as its purpose was to get a value for the no_quota field of
a delayed reference (this patch removes the no_quota field as mentioned
earlier).
Reported-by: Stéphane Lesimple <stephane_btrfs@lesimple.fr>
Tested-by: Stéphane Lesimple <stephane_btrfs@lesimple.fr>
Reported-by: Elias Probst <mail@eliasprobst.eu>
Reported-by: Peter Becker <floyd.net@gmail.com>
Reported-by: Malte Schröder <malte@tnxip.de>
Reported-by: Derek Dongray <derek@valedon.co.uk>
Reported-by: Erkki Seppala <flux-btrfs@inside.org>
Cc: stable@vger.kernel.org # 4.2+
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Add new function btrfs_add_delayed_qgroup_reserve() function to record
how much space is reserved for that extent.
As btrfs only accounts qgroup at run_delayed_refs() time, so newly
allocated extent should keep the reserved space until then.
So add needed function with related members to do it.
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
This is used by later qgroup fix patches for snapshot.
As current snapshot accounting is done by btrfs_qgroup_inherit(), but
new extent oriented quota mechanism will account extent from
btrfs_copy_root() and other snapshot things, causing wrong result.
So add this ability to handle snapshot accounting.
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
Add hook in add_delayed_ref_head() to record quota-related extent record
into delayed_ref_root->dirty_extent_record rb-tree for later qgroup
accounting.
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
This patch replace the rbtree used in ref_head to list.
This has the following advantage:
1) Easier merge logic.
With the new list implement, we only need to care merging the tail
ref_node with the new ref_node.
And this can be done quite easy at insert time, no need to do a
indicated merge at run_delayed_refs().
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: Chris Mason <clm@fb.com>
As we delete large extents, we end up doing huge amounts of COW in order
to delete the corresponding crcs. This adds accounting so that we keep
track of that space and flushing of delayed refs so that we don't build
up too much delayed crc work.
This helps limit the delayed work that must be done at commit time and
tries to avoid ENOSPC aborts because the crcs eat all the global
reserves.
Signed-off-by: Chris Mason <clm@fb.com>
Currently qgroups account for space by intercepting delayed ref updates to fs
trees. It does this by adding sequence numbers to delayed ref updates so that
it can figure out how the tree looked before the update so we can adjust the
counters properly. The problem with this is that it does not allow delayed refs
to be merged, so if you say are defragging an extent with 5k snapshots pointing
to it we will thrash the delayed ref lock because we need to go back and
manually merge these things together. Instead we want to process quota changes
when we know they are going to happen, like when we first allocate an extent, we
free a reference for an extent, we add new references etc. This patch
accomplishes this by only adding qgroup operations for real ref changes. We
only modify the sequence number when we need to lookup roots for bytenrs, this
reduces the amount of churn on the sequence number and allows us to merge
delayed refs as we add them most of the time. This patch encompasses a bunch of
architectural changes
1) qgroup ref operations: instead of tracking qgroup operations through the
delayed refs we simply add new ref operations whenever we notice that we need to
when we've modified the refs themselves.
2) tree mod seq: we no longer have this separation of major/minor counters.
this makes the sequence number stuff much more sane and we can remove some
locking that was needed to protect the counter.
3) delayed ref seq: we now read the tree mod seq number and use that as our
sequence. This means each new delayed ref doesn't have it's own unique sequence
number, rather whenever we go to lookup backrefs we inc the sequence number so
we can make sure to keep any new operations from screwing up our world view at
that given point. This allows us to merge delayed refs during runtime.
With all of these changes the delayed ref stuff is a little saner and the qgroup
accounting stuff no longer goes negative in some cases like it was before.
Thanks,
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
Currently we have two rb-trees, one for delayed ref heads and one for all of the
delayed refs, including the delayed ref heads. When we process the delayed refs
we have to hold onto the delayed ref lock for all of the selecting and merging
and such, which results in quite a bit of lock contention. This was solved by
having a waitqueue and only one flusher at a time, however this hurts if we get
a lot of delayed refs queued up.
So instead just have an rb tree for the delayed ref heads, and then attach the
delayed ref updates to an rb tree that is per delayed ref head. Then we only
need to take the delayed ref lock when adding new delayed refs and when
selecting a delayed ref head to process, all the rest of the time we deal with a
per delayed ref head lock which will be much less contentious.
The locking rules for this get a little more complicated since we have to lock
up to 3 things to properly process delayed refs, but I will address that problem
later. For now this passes all of xfstests and my overnight stress tests.
Thanks,
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>
The way how we process delayed refs is
1) get a bunch of head refs,
2) pick up one head ref,
3) go one node back for any delayed ref updates.
The head ref is also linked in the same rbtree as the delayed ref is,
so in 1) stage, we have to walk one by one including not only head refs, but
delayed refs.
When we have a great number of delayed refs pending to process,
this'll cost time a lot.
Here we introduce a head ref specific rbtree, it only has head refs, so troubles
go away.
Signed-off-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Chris Mason <clm@fb.com>
Chris hit a bug where we weren't finding extent records when running extent ops.
This is because we use the delayed_ref_head when running the extent op, which
means we can't use the ->type checks to see if we are metadata. We also lose
the level of the metadata we are working on. So to fix this we can just check
the ->is_data section of the extent_op, and we can store the level of the buffer
we were modifying in the extent_op. Thanks,
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Locking and unlocking delayed ref mutex are in the different functions,
and the name of lock functions is not uniform, so the readability is not
so good, this patch optimizes the lock logic and makes it more readable.
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
Nikolay Borisov