We got dentry pages from high_mem, and its address space directly goes into the
decryption path via f2fs_fname_disk_to_usr.
But, sg_init_one assumes the address is not from high_mem, so we can get this
panic since it doesn't call kmap_high but kunmap_high is triggered at the end.
kernel BUG at ../../../../../../kernel/mm/highmem.c:290!
Internal error: Oops - BUG: 0 [#1] PREEMPT SMP ARM
...
(kunmap_high+0xb0/0xb8) from [<c0114534>] (__kunmap_atomic+0xa0/0xa4)
(__kunmap_atomic+0xa0/0xa4) from [<c035f028>] (blkcipher_walk_done+0x128/0x1ec)
(blkcipher_walk_done+0x128/0x1ec) from [<c0366c24>] (crypto_cbc_decrypt+0xc0/0x170)
(crypto_cbc_decrypt+0xc0/0x170) from [<c0367148>] (crypto_cts_decrypt+0xc0/0x114)
(crypto_cts_decrypt+0xc0/0x114) from [<c035ea98>] (async_decrypt+0x40/0x48)
(async_decrypt+0x40/0x48) from [<c032ca34>] (f2fs_fname_disk_to_usr+0x124/0x304)
(f2fs_fname_disk_to_usr+0x124/0x304) from [<c03056fc>] (f2fs_fill_dentries+0xac/0x188)
(f2fs_fill_dentries+0xac/0x188) from [<c03059c8>] (f2fs_readdir+0x1f0/0x300)
(f2fs_readdir+0x1f0/0x300) from [<c0218054>] (vfs_readdir+0x90/0xb4)
(vfs_readdir+0x90/0xb4) from [<c0218418>] (SyS_getdents64+0x64/0xcc)
(SyS_getdents64+0x64/0xcc) from [<c0105ba0>] (ret_fast_syscall+0x0/0x30)
Cc: <stable@vger.kernel.org>
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In this patch, we try to reorganize f2fs_map_blocks to make block mapping
flow more clear by using following structure:
/* check status of mapping */
if (unmapped) {
/* blkaddr == NULL_ADDR || blkaddr == NEW_ADDR */
if (create) {
/* write path, handle dio write case here */
alloc_and_map;
} else {
/*
* handle read cases from all call paths:
* 1. generic read;
* 2. dio read;
* 3. fiemap;
* 4. bmap
*/
}
}
/* map buffer_header */
Besides, this patch handles the missing case correctly for dio write:
When we fail in __allocate_data_blocks, then in f2fs_map_blocks, we will
not allocate blocks correctly for preallocated blocks, but returning with
an unmapped buffer head, which will result in failure of dio write.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
We have potential overflow issue when calculating size of object, when
we left shift index with PAGE_CACHE_SHIFT bits, if type of index has only
32-bits space in 32-bit architecture, left shifting will incur overflow,
i.e:
pgoff_t index = 0xFFFFFFFF;
loff_t size = index << PAGE_CACHE_SHIFT;
size: 0xFFFFF000
So we should cast index with 64-bits type to avoid this issue.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
When shrinking extent cache, we have two steps in the flow:
1) shrink objects which are unreferenced by inodes;
2) shrink objects from LRU list of extent cache.
In step 1, if we haven't shrunk enough number of objects, we will try
step 2, but before that we didn't update the searching position which
may point to last inode index in global extent tree, result in failing
to shrink objects by traversing the all inodes' extent tree.
In this patch, we reset searching position to beginning of global extent
tree for fixing.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch changes to verify file type early in f2fs_fallocate for
cleanup, meanwhile this also fixes to add missing verification for
expand_inode_data.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
As comment says, we don't need to call f2fs_lock_op in write_inode to prevent
from producing dirty node pages all the time.
That happens only when there is not enough free sections and we can avoid that
by calling balance_fs in prior to that.
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This number is referenced by checkpoint under node_write lock.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This fixes the incorrect return statement at the end of the function
f2fs_ioc_release_volatile_write's body for returning zero as this is
incorrect due to the function call before this return statement to
the function punch_hole being able to fail and we should return this
function's return fail directly in order to signal to callers of the
function f2fs_ioc_release_volatile if a failure arises with this call
to punch_hole fails.
Signed-off-by: Nicholas Krause <xerofoify@gmail.com>
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Rename trace_f2fs_update_extent_tree to trace_f2fs_update_extent_tree_range,
then expand and enable it to trace in batches extent info updates.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Pull f2fs updates from Jaegeuk Kim:
"The major work includes fixing and enhancing the existing extent_cache
feature, which has been well settling down so far and now it becomes a
default mount option accordingly.
Also, this version newly registers a f2fs memory shrinker to reclaim
several objects consumed by a couple of data structures in order to
avoid memory pressures.
Another new feature is to add ioctl(F2FS_GARBAGE_COLLECT) which
triggers a cleaning job explicitly by users.
Most of the other patches are to fix bugs occurred in the corner cases
across the whole code area"
* tag 'for-f2fs-4.3' of git://git.kernel.org/pub/scm/linux/kernel/git/jaegeuk/f2fs: (85 commits)
f2fs: upset segment_info repair
f2fs: avoid accessing NULL pointer in f2fs_drop_largest_extent
f2fs: update extent tree in batches
f2fs: fix to release inode correctly
f2fs: handle f2fs_truncate error correctly
f2fs: avoid unneeded initializing when converting inline dentry
f2fs: atomically set inode->i_flags
f2fs: fix wrong pointer access during try_to_free_nids
f2fs: use __GFP_NOFAIL to avoid infinite loop
f2fs: lookup neighbor extent nodes for merging later
f2fs: split __insert_extent_tree_ret for readability
f2fs: kill dead code in __insert_extent_tree
f2fs: adjust showing of extent cache stat
f2fs: add largest/cached stat in extent cache
f2fs: fix incorrect mapping for bmap
f2fs: add annotation for space utilization of regular/inline dentry
f2fs: fix to update cached_en of extent tree properly
f2fs: fix typo
f2fs: check the node block address of newly allocated nid
f2fs: go out for insert_inode_locked failure
...
Pull core block updates from Jens Axboe:
"This first core part of the block IO changes contains:
- Cleanup of the bio IO error signaling from Christoph. We used to
rely on the uptodate bit and passing around of an error, now we
store the error in the bio itself.
- Improvement of the above from myself, by shrinking the bio size
down again to fit in two cachelines on x86-64.
- Revert of the max_hw_sectors cap removal from a revision again,
from Jeff Moyer. This caused performance regressions in various
tests. Reinstate the limit, bump it to a more reasonable size
instead.
- Make /sys/block/<dev>/queue/discard_max_bytes writeable, by me.
Most devices have huge trim limits, which can cause nasty latencies
when deleting files. Enable the admin to configure the size down.
We will look into having a more sane default instead of UINT_MAX
sectors.
- Improvement of the SGP gaps logic from Keith Busch.
- Enable the block core to handle arbitrarily sized bios, which
enables a nice simplification of bio_add_page() (which is an IO hot
path). From Kent.
- Improvements to the partition io stats accounting, making it
faster. From Ming Lei.
- Also from Ming Lei, a basic fixup for overflow of the sysfs pending
file in blk-mq, as well as a fix for a blk-mq timeout race
condition.
- Ming Lin has been carrying Kents above mentioned patches forward
for a while, and testing them. Ming also did a few fixes around
that.
- Sasha Levin found and fixed a use-after-free problem introduced by
the bio->bi_error changes from Christoph.
- Small blk cgroup cleanup from Viresh Kumar"
* 'for-4.3/core' of git://git.kernel.dk/linux-block: (26 commits)
blk: Fix bio_io_vec index when checking bvec gaps
block: Replace SG_GAPS with new queue limits mask
block: bump BLK_DEF_MAX_SECTORS to 2560
Revert "block: remove artifical max_hw_sectors cap"
blk-mq: fix race between timeout and freeing request
blk-mq: fix buffer overflow when reading sysfs file of 'pending'
Documentation: update notes in biovecs about arbitrarily sized bios
block: remove bio_get_nr_vecs()
fs: use helper bio_add_page() instead of open coding on bi_io_vec
block: kill merge_bvec_fn() completely
md/raid5: get rid of bio_fits_rdev()
md/raid5: split bio for chunk_aligned_read
block: remove split code in blkdev_issue_{discard,write_same}
btrfs: remove bio splitting and merge_bvec_fn() calls
bcache: remove driver private bio splitting code
block: simplify bio_add_page()
block: make generic_make_request handle arbitrarily sized bios
blk-cgroup: Drop unlikely before IS_ERR(_OR_NULL)
block: don't access bio->bi_error after bio_put()
block: shrink struct bio down to 2 cache lines again
...
This patch introduce a new helper f2fs_update_extent_tree_range which can
do extent mapping update at a specified range.
The main idea is:
1) punch all mapping info in extent node(s) which are at a specified range;
2) try to merge new extent mapping with adjacent node, or failing that,
insert the mapping into extent tree as a new node.
In order to see the benefit, I add a function for stating time stamping
count as below:
uint64_t rdtsc(void)
{
uint32_t lo, hi;
__asm__ __volatile__ ("rdtsc" : "=a" (lo), "=d" (hi));
return (uint64_t)hi << 32 | lo;
}
My test environment is: ubuntu, intel i7-3770, 16G memory, 256g micron ssd.
truncation path: update extent cache from truncate_data_blocks_range
non-truncataion path: update extent cache from other paths
total: all update paths
a) Removing 128MB file which has one extent node mapping whole range of
file:
1. dd if=/dev/zero of=/mnt/f2fs/128M bs=1M count=128
2. sync
3. rm /mnt/f2fs/128M
Before:
total count average
truncation: 7651022 32768 233.49
Patched:
total count average
truncation: 3321 33 100.64
b) fsstress:
fsstress -d /mnt/f2fs -l 5 -n 100 -p 20
Test times: 5 times.
Before:
total count average
truncation: 5812480.6 20911.6 277.95
non-truncation: 7783845.6 13440.8 579.12
total: 13596326.2 34352.4 395.79
Patched:
total count average
truncation: 1281283.0 3041.6 421.25
non-truncation: 7355844.4 13662.8 538.38
total: 8637127.4 16704.4 517.06
1) For the updates in truncation path:
- we can see updating in batches leads total tsc and update count reducing
explicitly;
- besides, for a single batched updating, punching multiple extent nodes
in a loop, result in executing more operations, so our average tsc
increase intensively.
2) For the updates in non-truncation path:
- there is a little improvement, that is because for the scenario that we
just need to update in the head or tail of extent node, new interface
optimize to update info in extent node directly, rather than removing
original extent node for updating and then inserting that updated one
into cache as new node.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In following call stack, if unfortunately we lose all chances to truncate
inode page in remove_inode_page, eventually we will add the nid allocated
previously into free nid cache, this nid is with NID_NEW status and with
NEW_ADDR in its blkaddr pointer:
- f2fs_create
- f2fs_add_link
- __f2fs_add_link
- init_inode_metadata
- new_inode_page
- new_node_page
- set_node_addr(, NEW_ADDR)
- f2fs_init_acl failed
- remove_inode_page failed
- handle_failed_inode
- remove_inode_page failed
- iput
- f2fs_evict_inode
- remove_inode_page failed
- alloc_nid_failed cache a nid with valid blkaddr: NEW_ADDR
This may not only cause resource leak of previous inode, but also may cause
incorrect use of the previous blkaddr which is located in NO.nid node entry
when this nid is reused by others.
This patch tries to add this inode to orphan list if we fail to truncate
inode, so that we can obtain a second chance to release it in orphan
recovery flow.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch fixes to return error number of f2fs_truncate, so that we
can handle the error correctly in callers.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
When converting inline dentry, we will zero out target dentry page before
duplicating data of inline dentry into target page, it become overhead
since inline dentry size is not small.
So this patch tries to remove unneeded initializing in the space of target
dentry page.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
According to commit 5f16f3225b ("ext4: atomically set inode->i_flags in
ext4_set_inode_flags()").
Signed-off-by: Zhang Zhen <zhenzhang.zhang@huawei.com>
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
__GFP_NOFAIL can avoid retrying the whole path of kmem_cache_alloc and
bio_alloc.
And, it also fixes the use cases of GFP_ATOMIC correctly.
Suggested-by: Chao Yu <chao2.yu@samsung.com>
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In __lookup_extent_tree_ret we will not try to find neighbor nodes if
we find the target node, in this condition, we will lost the chance to
merge the new mapping with exist extent node later.
So our extent cache of inode will be fragmented after overwrite exist
file, we can see the number of extent node increases intensively in
following test case:
dd if=/dev/zero of=/mnt/f2fs/4m bs=4K count=1024
Extent Cache:
- Hit Count: L1-1:0 L1-2:0 L2:0
- Hit Ratio: 0% (0 / 3072)
- Inner Struct Count: tree: 1, node: 1
dd if=/dev/zero of=/mnt/f2fs/4m bs=4K count=1024 conv=notrunc
Extent Cache:
- Hit Count: L1-1:2048 L1-2:0 L2:0
- Hit Ratio: 33% (2048 / 6144)
- Inner Struct Count: tree: 1, node: 961
This patch fixes to lookup neighbors of target node for further
merging.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch splits __insert_extent_tree_ret into __try_merge_extent_node &
__insert_extent_tree for code readability.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
After commit 0f825ee6e8 ("f2fs: add new interfaces for extent tree"),
f2fs_init_extent_tree becomes the only caller of __insert_extent_tree, and
in f2fs_init_extent_tree, we will only insert extent node in an empty tree,
so __try_{back,front}_merge in __insert_extent_tree will never be called.
This patch removes these dead codes, besides, rename __insert_extent_tree
to __init_extent_tree for readability.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch alters to replace total hit stat with rbtree hit stat,
and then adjust showing of extent cache stat:
Hit Count:
L1-1: for largest node hit count;
L1-2: for last cached node hit count;
L2: for extent node hit after lookuping in rbtree.
Hit Ratio:
ratio (hit count / total lookup count)
Inner Struct Count:
tree count, node count.
Before:
Extent Hit Ratio: 0 / 2
Extent Tree Count: 3
Extent Node Count: 2
Patched:
Exten Cacache:
- Hit Count: L1-1:4871 L1-2:2074 L2:208
- Hit Ratio: 1% (7153 / 550751)
- Inner Struct Count: tree: 26560, node: 11824
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch adds to stat the hit count of largest/cached node for showing
in debugfs.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
The test step is like below:
1. touch file
2. truncate -s $((1024*1024)) file
3. fallocate -o 0 -l $((1024*1024)) file
4. fibmap.f2fs file
Our result of fibmap.f2fs showed below is not correct:
file_pos start_blk end_blk blks
0 -937166132 -937166132 1
4096 -937166132 -937166132 1
8192 -937166132 -937166132 1
12288 -937166132 -937166132 1
16384 -937166132 -937166132 1
20480 -937166132 -937166132 1
...
1040384 -937166132 -937166132 1
1044480 -937166132 -937166132 1
This is because f2fs_map_blocks will return with no error when meeting
a hole or preallocated block, the caller __get_data_block will map the
uninitialized variable value to bh->b_blocknr.
Unfortunately generic_block_bmap will neither check the return value of
get_data() nor check mapping info of buffer_head, result in returning
the random block address.
After fixing the issue, our result shows correctly:
file_pos start_blk end_blk blks
0 0 0 256
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In f2fs_lookup_extent_tree, et->cached_en was read and updated with only
read lock held,
it could cause __lookup_extent_tree within return entirely wrong
extent_node, if other
thread update et->cached_en just before __lookup_extent_tree return.
However, there are two things about this patch that need to be noticed:
1. It does no good to arrange the order of concurrent read/write, the result
would still
be random in such case.
2. It's built on this assumption: the mix up of reads and writes on a single
pointer would
not make the pointer partially wrong at any time. Please let me know if I'm
wrong, thx.
Signed-off-by: Fan li <fanofcode.li@samsung.com>
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch adds a routine which checks the block address of newly allocated nid.
If an nid has already allocated by other thread due to subtle data races, it
will result in filesystem corruption.
So, it needs to check whether its block address was already allocated or not
in prior to nid allocation as the last chance.
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
We should not call unlock_new_inode when insert_inode_locked failed.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
If FG_GC failed to reclaim one section, let's retry with another section
from the start, since we can get anoterh good candidate.
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Previously, update_inode_page is not called under f2fs_lock_op.
Instead we should call with f2fs_write_inode.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
If we can reuse nids as many as possible, we can mitigate producing obsolete
node pages in the page cache.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
If node blocks were already moved, we don't need to move them again.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
As the below comment of bio_alloc_bioset, f2fs can allocate multiple bios at the
same time. So, we can't guarantee that bio is allocated all the time.
"
* When @bs is not NULL, if %__GFP_WAIT is set then bio_alloc will always be
* able to allocate a bio. This is due to the mempool guarantees. To make this
* work, callers must never allocate more than 1 bio at a time from this pool.
* Callers that need to allocate more than 1 bio must always submit the
* previously allocated bio for IO before attempting to allocate a new one.
* Failure to do so can cause deadlocks under memory pressure.
"
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch increases the number of maximum hard links for one file.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
We should avoid needless checkpoints when there is no dirty and prefree segment.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch introduces __count_free_nids/try_to_free_nids and registers
them in slab shrinker for shrinking under memory pressure.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In f2fs_delete_entry, if last dirent is remove from the dentry page,
we will try to punch that page since it has no valid date in it.
But truncate_hole which is used for punching could fail because of
no memory or IO error, if that happened, we'd better skip clearing
this valid dentry page.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
We should not write node pages when deleting orphan inodes.
In order to do that, we can eaisly set POR_DOING flag earlier before entering
orphan inode routine.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In recover_orphan_inode, whenever f2fs_iget fail, we will make kernel panic,
but it's not reasonable, because f2fs_iget can fail due to a lot of reasons
including out of memory.
So we change error handling method as below:
a) when finding no entry for the orphan inode, bug_on for catching bugs;
b) for other reasons, report it to caller.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
If there is not enough free segment, we should not assign a new segment
explicitly. Otherwise, we can run out of free segment.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
We can always fill up the bio now, no need to estimate the possible
size based on queue parameters.
Acked-by: Steven Whitehouse <swhiteho@redhat.com>
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
[hch: rebased and wrote a changelog]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lin <ming.l@ssi.samsung.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
Previously, we use radix tree to index all registered page entries for
atomic file, but now we only use radix tree to see whether current page
is indexed or not, since the other user of radix tree is gone in commit
042b7816aa ("f2fs: remove unnecessary call to invalidate inmemory pages").
So in this patch, we try to use one more efficient way:
Introducing a macro ATOMIC_WRITTEN_PAGE, and setting it as page private
value to indicate page indexing status. By using this way, we can save
memory and lookup time.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
We run ltp testcase with f2fs and obtain a TFAIL in diotest4, the result in
detail is as fallow:
dio04
<<<test_start>>>
tag=dio04 stime=1432278894
cmdline="diotest4"
contacts=""
analysis=exit
<<<test_output>>>
diotest4 1 TPASS : Negative Offset
diotest4 2 TPASS : removed
diotest4 3 TFAIL : diotest4.c:129: write allows odd count.returns 1: Success
diotest4 4 TFAIL : diotest4.c:183: Odd count of read and write
diotest4 5 TPASS : Read beyond the file size
......
the result of ext4 with same environment:
dio04
<<<test_start>>>
tag=dio04 stime=1432259643
cmdline="diotest4"
contacts=""
analysis=exit
<<<test_output>>>
diotest4 1 TPASS : Negative Offset
diotest4 2 TPASS : removed
diotest4 3 TPASS : Odd count of read and write
diotest4 4 TPASS : Read beyond the file size
......
The reason is that when triggering DIO in f2fs, we will return zero value
in ->direct_IO if writer's buffer offset, file offset and transfer size is
not alignment to block size of filesystem, resulting in falling back into
buffered write instead of returning -EINVAL.
This patch fixes that problem by returning correct error number for above
case, and removing the judgement condition in check_direct_IO to make sure
the verification will be enabled for direct reader too.
Besides, Jaegeuk Kim pointed out that there is expectional cases we should
always make direct-io falling back into buffered write, such as dio in
encrypted file.
Signed-off-by: Yunlei He <heyunlei@huawei.com>
[Chao Yu make small change and add detail description in commit message]
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
fill_zero can fail due to a lot of reason, but previously we do not handle
its return value, so its callers such as punch_hole/f2fs_zero_range may
report success, but actually can fail because of error occurs inside
fill_zero.
This patch fixes to report correct return value of fill_zero.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
When testing with generic/101 in xfstests, error message outputed as below:
--- tests/generic/101.out
+++ results//generic/101.out.bad
@@ -10,10 +10,14 @@
File foo content after log replay:
0000000 aa aa aa aa aa aa aa aa aa aa aa aa aa aa aa aa
*
-0200000 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
+0200000 bb bb bb bb bb bb bb bb bb bb bb bb bb bb bb bb
*
0372000
...
(Run 'diff -u tests/generic/101.out results/generic/101.out.bad' to see the entire diff)
The test flow is like below:
1. pwrite foo -S 0xaa 0 64K
2. pwrite foo -S 0xbb 64K 61K
3. sync
4. truncate foo 64K
5. truncate foo 125K
6. fsync foo
7. flakey drop writes
8. umount
After this test, we expect the data of recovered file will have the first
64k of data filling with value 0xaa and the next 61k of data filling with
value 0x00 because we have fsynced it before dropping writes in dm.
In f2fs, during recovering, we will only recover the valid block address
in direct node page if it is marked as a fsynced dnode, but block address
which means invalid/reserved (with value NULL_ADDR/NEW_ADDR) will not be
recovered. So, the file recovered shows its incorrect data 0xbb in range of
[61k, 125k].
In this patch, we fix to recover invalid/reserved block during recover flow.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In some cases, we only need the block address when we call
f2fs_reserve_block,
other fields of struct dnode_of_data aren't necessary.
We can try extent cache first for such cases in order to speed up the
process.
Signed-off-by: Fan li <fanofcode.li@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
To avoid meeting garbage data in next free node block at the end of warm
node chain when doing recovery, we will try to zero out that invalid block.
If the device is not support discard, our way for zeroing out block is:
grabbing a temporary zeroed page in meta inode, then, issue write request
with this page.
But, we forget to release that temporary page, so our memory usage will
increase without gaining any hit ratio benefit, so it's better to free it
for saving memory.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In following call path, we will pass a locked and referenced ipage
pointer to get_new_data_page:
- init_inode_metadata
- make_empty_dir
- get_new_data_page
There are two exit paths in get_new_data_page when error occurs:
1) grab_cache_page fails, ipage will not be released;
2) f2fs_reserve_block fails, ipage will be released in callee.
So, it's not consistent for error handling in get_new_data_page.
For f2fs_reserve_block, it's not very easy to change the rule
of error handling, since it's already complicated.
Here we deside to choose an easy way to fix this issue:
If any error occur in get_new_data_page, we will ensure releasing
ipage in this function.
The same issue is in f2fs_convert_inline_dir, fix that too.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Replace BUG_ON with f2fs_bug_on to deal with
block and segment validity check failed.
Signed-off-by: Xue Liu <liuxueliu.liu@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In get_meta_page, we guarantee no failure for the returned page,
but sometimes, IO error from device will incur returning an
non-updated page.
Then, we still use this page as updated one, exception could happen
when using this kind of page.
So in this condition, we'd better freeze fs by making fs readonly and
and stop doing checkpoint.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
some backing devices need pages to be stable during writeback. It doesn't
matter if
the page is completely overwritten or already uptodate, it needs to wait
before write.
Signed-off-by: Fan li <fanofcode.li@samsung.com>
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch adds to handle error cases in commit_inmem_pages.
If an error occurs, it stops to write the pages and return the error right
away.
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
When there is no enough free nids in free nid cache, we will try to
readahead FREE_NID_PAGES:4 nat pages into page cache of meta_inode,
then, reading nat entries in nat page for adding free nids to free nid
cache.
But when traversing all nat pages we readaheaded in a circulation,
our exit condition is not set right, one more nat page will be scanned
without readaheading, resulting worse read performance.
This patch fixes to read the correct number nat pages to avoid bad
performance.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
If we clear inline data/dentry flag in handle_failed_inode, we will fail
to decline the stat count of inline data/dentry in f2fs_evict_inode due
to no flag in inode. So remove the wrong clearing.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In f2fs_ioc_start_{atomic,volatile}_write, if we failed in converting
inline data, we will report error to user, but still remain atomic/volatile
flag in inode, it will impact further writes for this file. Fix it.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch fixes the incorrect range (0, LONG_MAX) which is used
in ranged fsync. If we use LONG_MAX as the parameter for indicating
the end of file we want to synchronize, in 32-bits architecture
machine, these datas after 4GB offset may not be persisted in
storage after ->fsync returned.
Here, we alter LONG_MAX to LLONG_MAX to fix this issue.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
When flushing comes from background, if there is no dirty page in the
mapping of inode, we'd better to skip seeking dirty page from mapping
for writebacking.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
The if statement "goto continue_unlock" is exactly the same when
each if condition is true that is depended on the value of both
"step" and "is_cold_data(page)" are 0 or 1. That means when the
value of "step" equals to "is_cold_data(page)", the if condition
is true and the if statement "goto continue_unlock" appears only
once, so it can be optimized to reduce the duplicated code.
Signed-off-by: Tiezhu Yang <kernelpatch@126.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In handle_failed_inode, there is a potential deadlock which can happen
in below call path:
- f2fs_create
- f2fs_lock_op down_read(cp_rwsem)
- f2fs_add_link
- __f2fs_add_link
- init_inode_metadata
- f2fs_init_security failed
- truncate_blocks failed
- handle_failed_inode
- f2fs_truncate
- truncate_blocks(..,true)
- write_checkpoint
- block_operations
- f2fs_lock_all down_write(cp_rwsem)
- f2fs_lock_op down_read(cp_rwsem)
So in this path, we pass parameter to f2fs_truncate to make sure
cp_rwsem in truncate_blocks will not be locked again.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In f2fs_do_collapse, region cp_rwsem covered is large, since it will be
held until all blocks are left shifted, so if we try to collapse small
area at the beginning of large file, checkpoint who want to grab writer's
lock of cp_rwsem will be delayed for long time.
In order to avoid this condition, altering to lock/unlock cp_rwsem each
shift operation.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Add a lookup and a insertion interface for extent tree.
The new lookup return the insert position and the prev/next
extents closest to the offset we lookup when find no match.
The new insertion uses above parameters to improve performance.
There are three possible insertions after the lookup in
f2fs_update_extent_tree, two of them insert parts of removed extent
back to tree, since no merge happens during this process, new insertion
skips the merge check in this scanario; the another insertion inserts a
new extent to tree, new insertion uses prev/next extent and insert
position to insert this extent directly, and save the time of searching
down the tree.
As long as tree remains unchanged between lookup and insertion, this
would work fine. And the new lookup would be useful when add
multi-blocks extent support for insertion interface.
Signed-off-by: Fan li <fanofcode.li@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch changes for a caller to handle the page after its bio gets an error.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Variables for recording extent cache ratio info were updated without
protection, this patch tries to alter them to atomic_t type for more
accurate stat.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch adds to stat the number of inline xattr inode for
showing in debugfs.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
That encrypted page is used temporarily, so we don't need to mark it accessed.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
If there are gced dirty pages and normal dirty pages in the mapping
of one inode, we might writeback them alternately with discontinuous
block address, resulting in low performance.
This patch introduces f2fs_write_cache_pages with codes copied from
write_cache_pages in mm/page-writeback.c.
In this function, we refactor flow with two steps:
1) writeback all cold type pages.
2) writeback all non-cold type pages.
By using this method, f2fs will writeback dirty pages with the same
temperature in bunch mode, it makes writeouted block being with
more continuous address, so they can be merged as much as possible
in f2fs bio cache, and also it will reduce the chance of submiting
small IO from block layer.
Test environment: 8g nokia sd card (very old sd card, but it shows
better effect when testing with this patch, and with a 32g kingston
sd card, I didn't see much more improvement).
Test step:
1. touch testfile;
2. truncate -s 512K testfile;
3. write all pages with odd index;
4. trigger gc by ioctl;
5. write all pages with even index;
6. time fsync testfile.
before:
real 0m0.402s
user 0m0.000s
sys 0m0.000s
after:
real 0m0.143s
user 0m0.004s
sys 0m0.004s
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch fixes to return correct error number of ->setxattr, which
is reported by xfstest tests/generic/026 as below:
generic/026 - output mismatch
--- tests/generic/026.out
+++ results/generic/026.out.bad
@@ -4,6 +4,6 @@
1 below acl max
acl max
1 above acl max
-chacl: cannot set access acl on "largeaclfile": Argument list too long
+chacl: cannot set access acl on "largeaclfile": Numerical result out of range
use 16 aces
use 17 aces
...
Ran: generic/026
Failures: generic/026
Failed 1 of 1 tests
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Previously, since 'commit 4531929e39 ("f2fs: move grabing orphan
pages out of protection region")' was committed, in write_orphan_inodes(),
we will grab all meta page in a batch before we use them under spinlock,
so that we can avoid large time delay of grabbing meta pages under
spinlock.
Now, 'commit d6c67a4fee ("f2fs: revmove spin_lock for
write_orphan_inodes")' remove the spinlock in write_orphan_inodes,
so there is no issue we describe above, we'd better recover to move
the grab operation to original place for readability.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
With cost-benifit method, background gc will consider old section with
fewer valid blocks as candidate victim, these old blocks in section will
be treated as cold data, and laterly will be moved into cold segment.
But if the gcing page is attached by user through buffered or mmaped
write, we should reset the page as non-cold one, because this page may
have more opportunity for further updating.
So fix to add clearing code for the missed 'mmap' case.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
When background gc is off, the only way to trigger gc is executing
a force gc in some operations who wants to grab space in disk.
The executing condition is limited: to execute force gc, we should
wait for the time when there is almost no more free section for LFS
allocation. This seems not reasonable for our user who wants to
control triggering gc by himself.
This patch introduces F2FS_IOC_GARBAGE_COLLECT interface for
triggering garbage collection by using ioctl. It provides our users
one more option to trigger gc.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch moves extent cache related code from data.c into extent_cache.c
since extent cache is independent feature, and its codes are not relate to
others in data.c, it's better for us to maintain them in separated place.
There is no functionality change, but several small coding style fixes
including:
* rename __drop_largest_extent to f2fs_drop_largest_extent for exporting;
* rename misspelled word 'untill' to 'until';
* remove unneeded 'return' in the end of f2fs_destroy_extent_tree().
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Since only parts of extents longer than F2FS_MIN_EXTENT_LEN will
be kept in extent cache after split, extents already shorter than
F2FS_MIN_EXTENT_LEN don't need to try split at all.
Signed-off-by: Fan Li <fanofcode.li@samsung.com>
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch fixes to update page flag (e.g. Uptodate/cold flag) in
->write_begin.
Otherwise, page will be non-uptodate when we try to write entire
page, and cold data flag in page will not be clean when gced page
is being rewritten.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
If an extent_tree entry has a zero reference count, we can drop it from the
cache in higher priority rather than currently referencing entries.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
In ->writepages, we use writepages mutex lock to serialize all block
address allocation and page submitting pairs from different inodes.
This method makes our delayed dirty pages of one inode being written
continously as many as possible.
But there is one problem that we did not submit current cached bio in
protection region of writepages mutex lock, so there is a small chance
that we submit the one of other thread's as below, resulting in
splitting more bios.
thread 1 thread 2
->writepages
lock(writepages)
->write_cache_pages
unlock(writepages)
lock(writepages)
->write_cache_pages
->f2fs_submit_merged_bio
->writepage
unlock(writepages)
fs_mark-6535 [002] .... 2242.270230: f2fs_submit_write_bio: dev = (1,0), WRITE_SYNC, DATA, sector = 5766152, size = 524288
fs_mark-6536 [000] .... 2242.270361: f2fs_submit_write_bio: dev = (1,0), WRITE_SYNC, DATA, sector = 5767176, size = 4096
fs_mark-6536 [000] .... 2242.270370: f2fs_submit_write_bio: dev = (1,0), WRITE_SYNC, NODE, sector = 8138112, size = 4096
fs_mark-6535 [002] .... 2242.270776: f2fs_submit_write_bio: dev = (1,0), WRITE_SYNC, DATA, sector = 5767184, size = 516096
This may really increase time of block layer works, and may cause
larger IO lantency.
This patch moves the submitting operation into region of writepages
mutex lock to avoid bio splits when concurrently writebacking is
intensive.
my test environment: virtual machine,
intel cpu i5 2500, 8GB size memory, 4GB size ramdisk
time fs_mark -t 16 -L 1 -s 524288 -S 1 -d /mnt/f2fs/
before:
real 0m4.244s
user 0m0.088s
sys 0m12.336s
after:
real 0m3.822s
user 0m0.072s
sys 0m10.760s
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
When testing with fs_mark, some blocks were written out as cold
data which were mixed with warm data, resulting in splitting more
bios.
This is because fs_mark will create file with random filename as
below:
559551ee~~~~~~~~15Z29OCC05JCKQP60JQ42MKV
559551ee~~~~~~~~NZAZ6X8OA8LHIIP6XD0L58RM
559551ef~~~~~~~~B15YDSWAK789HPSDZKYTW6WM
559551f1~~~~~~~~2DAE5DPS79785BUNTFWBEMP3
559551f1~~~~~~~~1MYDY0BKSQCJPI32Q8C514RM
559551f1~~~~~~~~YQOTMAOMN5CVRFOUNI026MP4
559551f3~~~~~~~~1WF42LPRTQJNPPGR3EINKMPE
559551f3~~~~~~~~8Y2NRK7CEPPAA02LY936PJPG
They are regarded as cold file since their filename are ended with
multimedia files' extension, but this should be wrong as we only
match the extension of filename, not the whole one.
In this patch, we try to fix the format of multimedia filename to:
"filename + '.' + extension", then we set cold file only its
filename matches the format.
So after this change, it will reduce the probability we set the
wrong cold file, also it helps a little for fs_mark's performance
on f2fs.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This makes the function check_dnode have a return type of bool
due to this particular function only ever returning either one
or zero as its return value and changes the name of the function
to is_alive in order to better explain this function's intended
work of checking if a dnode is still in use by the filesystem.
Signed-off-by: Nicholas Krause <xerofoify@gmail.com>
[Jaegeuk Kim: change the return value check for the renamed function]
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Because of the extent shrinker or other -ENOMEM scenarios, it cannot guarantee
that the largest extent would be cached in the tree all the time.
Instead of relying on extent_tree, we can simply check the cached one in extent
tree accordingly.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
We don't need to handle the duplicate extent information.
The integrated rule is:
- update on-disk extent with largest one tracked by in-memory extent_cache
- destroy extent_tree for the truncation case
- drop per-inode extent_cache by shrinker
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch introduces a shrinker targeting to reduce memory footprint consumed
by a number of in-memory f2fs data structures.
In addition, it newly adds:
- sbi->umount_mutex to avoid data races on shrinker and put_super
- sbi->shruinker_run_no to not revisit objects
Note that the basic implementation was copied from fs/ubifs/shrinker.c
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch relocates cached_en not only to be covered by spin_lock, but also
to set once after checking out completely.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Previously, f2fs_update_extent_cache() updates in-memory extent_cache all the
time, and then finally preserves its up-to-date extent into on-disk one during
f2fs_evict_inode.
But, in the following scenario:
1. mount
2. open & write an extent X
3. f2fs_evict_inode; on-disk extent is X
4. open & update the extent X with Y
5. sync; trigger checkpoint
6. power-cut
after power-on, f2fs should serve extent Y, but we have an on-disk extent X.
This causes a failure on xfstests/311.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
This patch fixes wrong calculation on block address field when an extent is
split.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
For newly added fallocate types, it should convert inline_data before handling
block swapping.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Before iput is called, the inode number used by a bad inode can be reassigned
to other new inode, resulting in any abnormal behaviors on the new inode.
This should not happen for the new inode.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
The write_checkpoint can update stat information, so we should destroy the stat
structure after it.
Reviewed-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Dirty page can be exist in mapping of newly created symlink, but previously
we did not maintain the counting of dirty page for symlink like we maintained
for regular/directory, so the counting we lookuped should be wrong.
This patch adds missed dirty page counting for symlink to fix this issue.
Signed-off-by: Chao Yu <chao2.yu@samsung.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
The key_put() function tests whether its argument is NULL and then
returns immediately. Thus the test around the call is not needed.
This issue was detected by using the Coccinelle software.
Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
Currently we have two different ways to signal an I/O error on a BIO:
(1) by clearing the BIO_UPTODATE flag
(2) by returning a Linux errno value to the bi_end_io callback
The first one has the drawback of only communicating a single possible
error (-EIO), and the second one has the drawback of not beeing persistent
when bios are queued up, and are not passed along from child to parent
bio in the ever more popular chaining scenario. Having both mechanisms
available has the additional drawback of utterly confusing driver authors
and introducing bugs where various I/O submitters only deal with one of
them, and the others have to add boilerplate code to deal with both kinds
of error returns.
So add a new bi_error field to store an errno value directly in struct
bio and remove the existing mechanisms to clean all this up.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: NeilBrown <neilb@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
Pull cgroup writeback support from Jens Axboe:
"This is the big pull request for adding cgroup writeback support.
This code has been in development for a long time, and it has been
simmering in for-next for a good chunk of this cycle too. This is one
of those problems that has been talked about for at least half a
decade, finally there's a solution and code to go with it.
Also see last weeks writeup on LWN:
http://lwn.net/Articles/648292/"
* 'for-4.2/writeback' of git://git.kernel.dk/linux-block: (85 commits)
writeback, blkio: add documentation for cgroup writeback support
vfs, writeback: replace FS_CGROUP_WRITEBACK with SB_I_CGROUPWB
writeback: do foreign inode detection iff cgroup writeback is enabled
v9fs: fix error handling in v9fs_session_init()
bdi: fix wrong error return value in cgwb_create()
buffer: remove unusued 'ret' variable
writeback: disassociate inodes from dying bdi_writebacks
writeback: implement foreign cgroup inode bdi_writeback switching
writeback: add lockdep annotation to inode_to_wb()
writeback: use unlocked_inode_to_wb transaction in inode_congested()
writeback: implement unlocked_inode_to_wb transaction and use it for stat updates
writeback: implement [locked_]inode_to_wb_and_lock_list()
writeback: implement foreign cgroup inode detection
writeback: make writeback_control track the inode being written back
writeback: relocate wb[_try]_get(), wb_put(), inode_{attach|detach}_wb()
mm: vmscan: disable memcg direct reclaim stalling if cgroup writeback support is in use
writeback: implement memcg writeback domain based throttling
writeback: reset wb_domain->dirty_limit[_tstmp] when memcg domain size changes
writeback: implement memcg wb_domain
writeback: update wb_over_bg_thresh() to use wb_domain aware operations
...
Pull f2fs updates from Jaegeuk Kim:
"New features:
- per-file encryption (e.g., ext4)
- FALLOC_FL_ZERO_RANGE
- FALLOC_FL_COLLAPSE_RANGE
- RENAME_WHITEOUT
Major enhancement/fixes:
- recovery broken superblocks
- enhance f2fs_trim_fs with a discard_map
- fix a race condition on dentry block allocation
- fix a deadlock during summary operation
- fix a missing fiemap result
.. and many minor bug fixes and clean-ups were done"
* tag 'for-f2fs-4.2' of git://git.kernel.org/pub/scm/linux/kernel/git/jaegeuk/f2fs: (83 commits)
f2fs: do not trim preallocated blocks when truncating after i_size
f2fs crypto: add alloc_bounce_page
f2fs crypto: fix to handle errors likewise ext4
f2fs: drop the volatile_write flag only
f2fs: skip committing valid superblock
f2fs: setting discard option in parse_options()
f2fs: fix to return exact trimmed size
f2fs: support FALLOC_FL_INSERT_RANGE
f2fs: hide common code in f2fs_replace_block
f2fs: disable the discard option when device doesn't support
f2fs crypto: remove alloc_page for bounce_page
f2fs: fix a deadlock for summary page lock vs. sentry_lock
f2fs crypto: clean up error handling in f2fs_fname_setup_filename
f2fs crypto: avoid f2fs_inherit_context for symlink
f2fs crypto: do not set encryption policy for non-directory by ioctl
f2fs crypto: allow setting encryption policy once
f2fs crypto: check context consistent for rename2
f2fs: avoid duplicated code by reusing f2fs_read_end_io
f2fs crypto: use per-inode tfm structure
f2fs: recovering broken superblock during mount
...