The patch increases the parallelism of mbcache by using the built-in
lock in the hlist_bl_node to protect the mb_cache's local block and
index hash chains. The global data mb_cache_lru_list and
mb_cache_list continue to be protected by the global
mb_cache_spinlock.
New block group spinlock, mb_cache_bg_lock is also added to serialize
accesses to mb_cache_entry's local data.
A new member e_refcnt is added to the mb_cache_entry structure to help
preventing an mb_cache_entry from being deallocated by a free while it
is being referenced by either mb_cache_entry_get() or
mb_cache_entry_find().
Signed-off-by: T. Makphaibulchoke <tmac@hp.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This patch changes each mb_cache's both block and index hash chains to
use a hlist_bl_node, which contains a built-in lock. This is the
first step in decoupling of locks serializing accesses to mb_cache
global data and each mb_cache_entry local data.
Signed-off-by: T. Makphaibulchoke <tmac@hp.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Introduce new FALLOC_FL_ZERO_RANGE flag for fallocate. This has the same
functionality as xfs ioctl XFS_IOC_ZERO_RANGE.
It can be used to convert a range of file to zeros preferably without
issuing data IO. Blocks should be preallocated for the regions that span
holes in the file, and the entire range is preferable converted to
unwritten extents
This can be also used to preallocate blocks past EOF in the same way as
with fallocate. Flag FALLOC_FL_KEEP_SIZE which should cause the inode
size to remain the same.
Also add appropriate tracepoints.
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Move block allocation out of the ext4_fallocate into separate function
called ext4_alloc_file_blocks(). This will allow us to use the same
allocation code for other allocation operations such as zero range which
is commit in the next patch.
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Currently in ext4_fallocate we would update inode size, c_time and sync
the file with every partial allocation which is entirely unnecessary. It
is true that if the crash happens in the middle of truncate we might end
up with unchanged i size, or c_time which I do not think is really a
problem - it does not mean file system corruption in any way. Note that
xfs is doing things the same way e.g. update all of the mentioned after
the allocation is done.
This commit moves all the updates after the allocation is done. In
addition we also need to change m_time as not only inode has been change
bot also data regions might have changed (unwritten extents). However
m_time will be only updated when i_size changed.
Also we do not need to be paranoid about changing the c_time only if the
actual allocation have happened, we can change it even if we try to
allocate only to find out that there are already block allocated. It's
not really a big deal and it will save us some additional complexity.
Also use ext4_debug, instead of ext4_warning in #ifdef EXT4FS_DEBUG
section.
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>-
--
v3: Do not remove the code to set EXT4_INODE_EOFBLOCKS flag
fs/ext4/extents.c | 96 ++++++++++++++++++++++++-------------------------------
1 file changed, 42 insertions(+), 54 deletions(-)
Commit 9cb00419fa, which enables hole punching for bigalloc file
systems, exposed a bug introduced by commit 6ae06ff51e in an earlier
release. When run on a bigalloc file system, xfstests generic/013, 068,
075, 083, 091, 100, 112, 127, 263, 269, and 270 fail with e2fsck errors
or cause kernel error messages indicating that previously freed blocks
are being freed again.
The latter commit optimizes the selection of the starting extent in
ext4_ext_rm_leaf() when hole punching by beginning with the extent
supplied in the path argument rather than with the last extent in the
leaf node (as is still done when truncating). However, the code in
rm_leaf that initially sets partial_cluster to track cluster sharing on
extent boundaries is only guaranteed to run if rm_leaf starts with the
last node in the leaf. Consequently, partial_cluster is not correctly
initialized when hole punching, and a cluster on the boundary of a
punched region that should be retained may instead be deallocated.
Signed-off-by: Eric Whitney <enwlinux@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
Code deallocating the extent path referenced by an argument to
ext4_ext_handle_uninitialized_extents was made redundant with identical
code in its one caller, ext4_ext_map_blocks, by commit 3779473246.
Allocating and deallocating the path in the same function also makes
the code clearer.
Signed-off-by: Eric Whitney <enwlinux@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Fix up error messages printed when the transaction pointers in a
journal head are inconsistent. This improves the error messages which
are printed when running xfstests generic/068 in data=journal mode.
See the bug report at: https://bugzilla.kernel.org/show_bug.cgi?id=60786
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
It's not needed until we start trying to modifying fields in the
journal_head which are protected by j_list_lock.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
It's not needed until we start trying to modifying fields in the
journal_head which are protected by j_list_lock.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
jh->b_transaction is adequately protected for reading by the
jbd_lock_bh_state(bh), so we don't need to take j_list_lock in
__journal_try_to_free_buffer().
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
We don't otherwise need j_list_lock during the rest of commit phase
#7, so add the transaction to the checkpoint list at the very end of
commit phase #6. This allows us to drop j_list_lock earlier, which is
a good thing since it is a super hot lock.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
The two hottest locks, and thus the biggest scalability bottlenecks,
in the jbd2 layer, are the j_list_lock and j_state_lock. This has
inspired some people to do some truly unnatural things[1].
[1] https://www.usenix.org/system/files/conference/fast14/fast14-paper_kang.pdf
We don't need to be holding both j_state_lock and j_list_lock while
calculating the journal statistics, so move those calculations to the
very end of jbd2_journal_commit_transaction.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
The j_state_lock is one of the hottest locks in the jbd2 layer and
thus one of its scalability bottlenecks.
We don't need to be holding the j_state_lock while we are calling
wake_up(&journal->j_wait_commit), so release the lock a little bit
earlier.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
During commit process, keep the block device plugged after we are done
writing the revoke records, until we are finished writing the rest of
the commit records in the journal. This will allow most of the
journal blocks to be written in a single I/O operation, instead of
separating the the revoke blocks from the rest of the journal blocks.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
When doing filesystem wide sync, there's no need to force transaction
commit (or synchronously write inode buffer) separately for each inode
because ext4_sync_fs() takes care of forcing commit at the end (VFS
takes care of flushing buffer cache, respectively). Most of the time
this slowness doesn't manifest because previous WB_SYNC_NONE writeback
doesn't leave much to write but when there are processes aggressively
creating new files and several filesystems to sync, the sync slowness
can be noticeable. In the following test script sync(1) takes around 6
minutes when there are two ext4 filesystems mounted on a standard SATA
drive. After this patch sync takes a couple of seconds so we have about
two orders of magnitude improvement.
function run_writers
{
for (( i = 0; i < 10; i++ )); do
mkdir $1/dir$i
for (( j = 0; j < 40000; j++ )); do
dd if=/dev/zero of=$1/dir$i/$j bs=4k count=4 &>/dev/null
done &
done
}
for dir in "$@"; do
run_writers $dir
done
sleep 40
time sync
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This patch implements fallocate's FALLOC_FL_COLLAPSE_RANGE for Ext4.
The semantics of this flag are following:
1) It collapses the range lying between offset and length by removing any data
blocks which are present in this range and than updates all the logical
offsets of extents beyond "offset + len" to nullify the hole created by
removing blocks. In short, it does not leave a hole.
2) It should be used exclusively. No other fallocate flag in combination.
3) Offset and length supplied to fallocate should be fs block size aligned
in case of xfs and ext4.
4) Collaspe range does not work beyond i_size.
Signed-off-by: Namjae Jeon <namjae.jeon@samsung.com>
Signed-off-by: Ashish Sangwan <a.sangwan@samsung.com>
Tested-by: Dongsu Park <dongsu.park@profitbricks.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
While handling punch-hole fallocate, it's useless to truncate page cache
before removing the range from extent tree (or block map in indirect case)
because page cache can be re-populated (by read-ahead or read(2) or mmap-ed
read) immediately after truncating page cache, but before updating extent
tree (or block map). In that case the user will see stale data even after
fallocate is completed.
Until the problem of data corruption resulting from pages backed by
already freed blocks is fully resolved, the simple thing we can do now
is to add another truncation of pagecache after punch hole is done.
Signed-off-by: Maxim Patlasov <mpatlasov@parallels.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
Adjust the conversion specifications in a few optionally compiled debug
messages to match the return type of ext4_es_status(). Also, make a
couple of minor grammatical message edits while we're at it.
Signed-off-by: Eric Whitney <enwlinux@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
When looking at a bug report with:
> kernel: EXT4-fs: 0 scanned, 0 found
I thought wow, 0 scanned, that's odd? But it's not odd; it's printing
a variable that is initialized to 0 and never touched again.
It's never been used since the original merge, so I don't really even
know what the original intent was, either.
If anyone knows how to hook it up, speak now via patch, otherwise just
yank it so it's not making a confusing situation more confusing in
kernel logs.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
The ext4_map_blocks() function returns the number of blocks which
satisfying the caller's request. This number of blocks requested by
the caller is specified by an unsigned integer, but the return value
of ext4_map_blocks() is a signed integer (to accomodate error codes
per the kernel's standard error signalling convention).
Historically, overflows could never happen since mballoc() will refuse
to allocate more than 2048 blocks at a time (which is something we
should fix), and if the blocks were already allocated, the fact that
there would be some number of intervening metadata blocks pretty much
guaranteed that there could never be a contiguous region of data
blocks that was greater than 2**31 blocks.
However, this is now possible if there is a file system which is a bit
bigger than 8TB, and is created using the new mke2fs hugeblock
feature, which can create a perfectly contiguous file. In that case,
if a userspace program attempted to call fallocate() on this already
fully allocated file, it's possible that ext4_map_blocks() could
return a number large enough that it would overflow a signed integer,
resulting in a ext4 thinking that the ext4_map_blocks() call had
failed with some strange error code.
Since ext4_map_blocks() is always free to return a smaller number of
blocks than what was requested by the caller, fix this by capping the
number of blocks that ext4_map_blocks() will ever try to map to 2**31
- 1. In practice this should never get hit, except by someone
deliberately trying to provke the above-described bug.
Thanks to the PaX team for asking whethre this could possibly happen
in some off-line discussions about using some static code checking
technology they are developing to find bugs in kernel code.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
The lowest levels of mballoc set all of the fields of struct
ext4_free_extent except for fe_logical, since they are just trying to
find the requested free set of blocks, and the logical block hasn't
been set yet. This makes some static code checkers sad. Set it to
various different debug values, which would be useful when
debugging mballoc if these values were to ever show up due to the
parts of mballoc triyng to use ac->ac_b_ex.fe_logical before it is
properly upper layers of mballoc failing to properly set, usually by
ext4_mb_use_best_found().
Addresses-Coverity-Id: #139697
Addresses-Coverity-Id: #139698
Addresses-Coverity-Id: #139699
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
The function ext4_expand_extra_isize_ea() doesn't need the size of all
of the extended attribute headers. So if we don't calculate it when
it is unneeded, it we can skip some undeeded memory references, and as
a bonus, we eliminate some kvetching by static code analysis tools.
Addresses-Coverity-Id: #741291
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Avoid false positives by static code analysis tools such as sparse and
coverity caused by the fact that we set the physical block, and then
the status in the extent_status structure. It is also more efficient
to set both of these values at once.
Addresses-Coverity-Id: #989077
Addresses-Coverity-Id: #989078
Addresses-Coverity-Id: #1080722
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Zheng Liu <wenqing.lz@taobao.com>
Commit 3779473246 breaks the return of error codes from
ext4_ext_handle_uninitialized_extents() in ext4_ext_map_blocks(). A
portion of the patch assigns that function's signed integer return
value to an unsigned int. Consequently, negatively valued error codes
are lost and can be treated as a bogus allocated block count.
Signed-off-by: Eric Whitney <enwlinux@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
When !defined(CONFIG_EXT4_DEBUG), mb_debug() should be defined as a
no_printk() statement instead of an empty statement in order to suppress
the following compiler warning:
fs/ext4/mballoc.c: In function ‘ext4_mb_cleanup_pa’:
fs/ext4/mballoc.c:2659:47: warning: suggest braces around empty body in an ‘if’ statement [-Wempty-body]
mb_debug(1, "mballoc: %u PAs left\n", count);
Signed-off-by: Patrick Palka <patrick@parcs.ath.cx>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Mark functions as static in jbd2/journal.c because they are not used
outside this file.
This eliminates the following warning in jbd2/journal.c:
fs/jbd2/journal.c:125:5: warning: no previous prototype for ‘jbd2_verify_csum_type’ [-Wmissing-prototypes]
fs/jbd2/journal.c:146:5: warning: no previous prototype for ‘jbd2_superblock_csum_verify’ [-Wmissing-prototypes]
fs/jbd2/journal.c:154:6: warning: no previous prototype for ‘jbd2_superblock_csum_set’ [-Wmissing-prototypes]
Signed-off-by: Rashika Kheria <rashika.kheria@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
"err" is zero here, there is no need to check again.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
If start_this_handle() fails then it leads to a use after free of
"handle".
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
If the i_crtime field is not present in the inode, don't leave the
field uninitialized.
Fixes: ef7f38359 ("ext4: Add nanosecond timestamps")
Reported-by: Vegard Nossum <vegard.nossum@oracle.com>
Tested-by: Vegard Nossum <vegard.nossum@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
The set_flexbg_block_bitmap() function assumed that the number of
blocks in a blockgroup was sb->blocksize * 8, which is normally true,
but not always! Use EXT4_BLOCKS_PER_GROUP(sb) instead, to fix block
bitmap corruption after:
mke2fs -t ext4 -g 3072 -i 4096 /dev/vdd 1G
mount -t ext4 /dev/vdd /vdd
resize2fs /dev/vdd 8G
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reported-by: Jon Bernard <jbernard@tuxion.com>
Cc: stable@vger.kernel.org
If a file system has a large number of inodes per block group, all of
the metadata blocks in a flex_bg may be larger than what can fit in a
single block group. Unfortunately, ext4_alloc_group_tables() in
resize.c was never tested to see if it would handle this case
correctly, and there were a large number of bugs which caused the
following sequence to result in a BUG_ON:
kernel bug at fs/ext4/resize.c:409!
...
call trace:
[<ffffffff81256768>] ext4_flex_group_add+0x1448/0x1830
[<ffffffff81257de2>] ext4_resize_fs+0x7b2/0xe80
[<ffffffff8123ac50>] ext4_ioctl+0xbf0/0xf00
[<ffffffff811c111d>] do_vfs_ioctl+0x2dd/0x4b0
[<ffffffff811b9df2>] ? final_putname+0x22/0x50
[<ffffffff811c1371>] sys_ioctl+0x81/0xa0
[<ffffffff81676aa9>] system_call_fastpath+0x16/0x1b
code: c8 4c 89 df e8 41 96 f8 ff 44 89 e8 49 01 c4 44 29 6d d4 0
rip [<ffffffff81254fa1>] set_flexbg_block_bitmap+0x171/0x180
This can be reproduced with the following command sequence:
mke2fs -t ext4 -i 4096 /dev/vdd 1G
mount -t ext4 /dev/vdd /vdd
resize2fs /dev/vdd 8G
To fix this, we need to make sure the right thing happens when a block
group's inode table straddles two block groups, which means the
following bugs had to be fixed:
1) Not clearing the BLOCK_UNINIT flag in the second block group in
ext4_alloc_group_tables --- the was proximate cause of the BUG_ON.
2) Incorrectly determining how many block groups contained contiguous
free blocks in ext4_alloc_group_tables().
3) Incorrectly setting the start of the next block range to be marked
in use after a discontinuity in setup_new_flex_group_blocks().
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
If an ext4 file system is created by some tool other than mke2fs
(perhaps by someone who has a pathalogical fear of the GPL) that
doesn't set one or the other of the EXT2_FLAGS_{UN}SIGNED_HASH flags,
and that file system is then mounted read-only, don't try to modify
the s_flags field. Otherwise, if dm_verity is in use, the superblock
will change, causing an dm_verity failure.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
In swap_inode_boot_loader() we forgot to release ->i_mutex and resume
unlocked dio for inode and inode_bl if there is an error starting the
journal handle. This commit fixes this issue.
Reported-by: Ahmed Tamrawi <ahmedtamrawi@gmail.com>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: Dr. Tilmann Bubeck <t.bubeck@reinform.de>
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org # v3.10+
Commit a115f749c1 (ext4: remove wait for unwritten extent conversion from
ext4_truncate) exposed a bug in ext4_ext_handle_uninitialized_extents().
It can be triggered by xfstest generic/299 when run on a test file
system created without a journal. This test continuously fallocates and
truncates files to which random dio/aio writes are simultaneously
performed by a separate process. The test completes successfully, but
if the test filesystem is mounted with the block_validity option, a
warning message stating that a logical block has been mapped to an
illegal physical block is posted in the kernel log.
The bug occurs when an extent is being converted to the written state
by ext4_end_io_dio() and ext4_ext_handle_uninitialized_extents()
discovers a mapping for an existing uninitialized extent. Although it
sets EXT4_MAP_MAPPED in map->m_flags, it fails to set map->m_pblk to
the discovered physical block number. Because map->m_pblk is not
otherwise initialized or set by this function or its callers, its
uninitialized value is returned to ext4_map_blocks(), where it is
stored as a bogus mapping in the extent status tree.
Since map->m_pblk can accidentally contain illegal values that are
larger than the physical size of the file system, calls to
check_block_validity() in ext4_map_blocks() that are enabled if the
block_validity mount option is used can fail, resulting in the logged
warning message.
Signed-off-by: Eric Whitney <enwlinux@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org # 3.11+
Pull SELinux fixes from James Morris.
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris/linux-security:
SELinux: Fix kernel BUG on empty security contexts.
selinux: add SOCK_DIAG_BY_FAMILY to the list of netlink message types
Pull vfs fixes from Al Viro:
"A couple of fixes, both -stable fodder. The O_SYNC bug is fairly
old..."
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
fix a kmap leak in virtio_console
fix O_SYNC|O_APPEND syncing the wrong range on write()
While we are at it, don't do kmap() under kmap_atomic(), *especially*
for a page we'd allocated with GFP_KERNEL. It's spelled "page_address",
and had that been more than that, we'd have a real trouble - kmap_high()
can block, and doing that while holding kmap_atomic() is a Bad Idea(tm).
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
It actually goes back to 2004 ([PATCH] Concurrent O_SYNC write support)
when sync_page_range() had been introduced; generic_file_write{,v}() correctly
synced
pos_after_write - written .. pos_after_write - 1
but generic_file_aio_write() synced
pos_before_write .. pos_before_write + written - 1
instead. Which is not the same thing with O_APPEND, obviously.
A couple of years later correct variant had been killed off when
everything switched to use of generic_file_aio_write().
All users of generic_file_aio_write() are affected, and the same bug
has been copied into other instances of ->aio_write().
The fix is trivial; the only subtle point is that generic_write_sync()
ought to be inlined to avoid calculations useless for the majority of
calls.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Pull btrfs fixes from Chris Mason:
"This is a small collection of fixes"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/linux-btrfs:
Btrfs: fix data corruption when reading/updating compressed extents
Btrfs: don't loop forever if we can't run because of the tree mod log
btrfs: reserve no transaction units in btrfs_ioctl_set_features
btrfs: commit transaction after setting label and features
Btrfs: fix assert screwup for the pending move stuff
Pull perf fixes from Ingo Molnar:
"Tooling fixes, mostly related to the KASLR fallout, but also other
fixes"
* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
perf buildid-cache: Check relocation when checking for existing kcore
perf tools: Adjust kallsyms for relocated kernel
perf tests: No need to set up ref_reloc_sym
perf symbols: Prevent the use of kcore if the kernel has moved
perf record: Get ref_reloc_sym from kernel map
perf machine: Set up ref_reloc_sym in machine__create_kernel_maps()
perf machine: Add machine__get_kallsyms_filename()
perf tools: Add kallsyms__get_function_start()
perf symbols: Fix symbol annotation for relocated kernel
perf tools: Fix include for non x86 architectures
perf tools: Fix AAAAARGH64 memory barriers
perf tools: Demangle kernel and kernel module symbols too
perf/doc: Remove mention of non-existent set_perf_event_pending() from design.txt
When using a mix of compressed file extents and prealloc extents, it
is possible to fill a page of a file with random, garbage data from
some unrelated previous use of the page, instead of a sequence of zeroes.
A simple sequence of steps to get into such case, taken from the test
case I made for xfstests, is:
_scratch_mkfs
_scratch_mount "-o compress-force=lzo"
$XFS_IO_PROG -f -c "pwrite -S 0x06 -b 18670 266978 18670" $SCRATCH_MNT/foobar
$XFS_IO_PROG -c "falloc 26450 665194" $SCRATCH_MNT/foobar
$XFS_IO_PROG -c "truncate 542872" $SCRATCH_MNT/foobar
$XFS_IO_PROG -c "fsync" $SCRATCH_MNT/foobar
This results in the following file items in the fs tree:
item 4 key (257 INODE_ITEM 0) itemoff 15879 itemsize 160
inode generation 6 transid 6 size 542872 block group 0 mode 100600
item 5 key (257 INODE_REF 256) itemoff 15863 itemsize 16
inode ref index 2 namelen 6 name: foobar
item 6 key (257 EXTENT_DATA 0) itemoff 15810 itemsize 53
extent data disk byte 0 nr 0 gen 6
extent data offset 0 nr 24576 ram 266240
extent compression 0
item 7 key (257 EXTENT_DATA 24576) itemoff 15757 itemsize 53
prealloc data disk byte 12849152 nr 241664 gen 6
prealloc data offset 0 nr 241664
item 8 key (257 EXTENT_DATA 266240) itemoff 15704 itemsize 53
extent data disk byte 12845056 nr 4096 gen 6
extent data offset 0 nr 20480 ram 20480
extent compression 2
item 9 key (257 EXTENT_DATA 286720) itemoff 15651 itemsize 53
prealloc data disk byte 13090816 nr 405504 gen 6
prealloc data offset 0 nr 258048
The on disk extent at offset 266240 (which corresponds to 1 single disk block),
contains 5 compressed chunks of file data. Each of the first 4 compress 4096
bytes of file data, while the last one only compresses 3024 bytes of file data.
Therefore a read into the file region [285648 ; 286720[ (length = 4096 - 3024 =
1072 bytes) should always return zeroes (our next extent is a prealloc one).
The solution here is the compression code path to zero the remaining (untouched)
bytes of the last page it uncompressed data into, as the information about how
much space the file data consumes in the last page is not known in the upper layer
fs/btrfs/extent_io.c:__do_readpage(). In __do_readpage we were correctly zeroing
the remainder of the page but only if it corresponds to the last page of the inode
and if the inode's size is not a multiple of the page size.
This would cause not only returning random data on reads, but also permanently
storing random data when updating parts of the region that should be zeroed.
For the example above, it means updating a single byte in the region [285648 ; 286720[
would store that byte correctly but also store random data on disk.
A test case for xfstests follows soon.
Signed-off-by: Filipe David Borba Manana <fdmanana@gmail.com>
Signed-off-by: Chris Mason <clm@fb.com>
