We have been seeing crashes on duplicate keys in
btrfs_set_item_key_safe():
BTRFS critical (device vdb): slot 4 key (450 108 8192) new key (450 108 8192)
------------[ cut here ]------------
kernel BUG at fs/btrfs/ctree.c:2620!
invalid opcode: 0000 [#1] PREEMPT SMP PTI
CPU: 0 PID: 3139 Comm: xfs_io Kdump: loaded Not tainted 6.9.0 #6
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.3-2.fc40 04/01/2014
RIP: 0010:btrfs_set_item_key_safe+0x11f/0x290 [btrfs]
With the following stack trace:
#0 btrfs_set_item_key_safe (fs/btrfs/ctree.c:2620:4)
#1 btrfs_drop_extents (fs/btrfs/file.c:411:4)
#2 log_one_extent (fs/btrfs/tree-log.c:4732:9)
#3 btrfs_log_changed_extents (fs/btrfs/tree-log.c:4955:9)
#4 btrfs_log_inode (fs/btrfs/tree-log.c:6626:9)
#5 btrfs_log_inode_parent (fs/btrfs/tree-log.c:7070:8)
#6 btrfs_log_dentry_safe (fs/btrfs/tree-log.c:7171:8)
#7 btrfs_sync_file (fs/btrfs/file.c:1933:8)
#8 vfs_fsync_range (fs/sync.c:188:9)
#9 vfs_fsync (fs/sync.c:202:9)
#10 do_fsync (fs/sync.c:212:9)
#11 __do_sys_fdatasync (fs/sync.c:225:9)
#12 __se_sys_fdatasync (fs/sync.c:223:1)
#13 __x64_sys_fdatasync (fs/sync.c:223:1)
#14 do_syscall_x64 (arch/x86/entry/common.c:52:14)
#15 do_syscall_64 (arch/x86/entry/common.c:83:7)
#16 entry_SYSCALL_64+0xaf/0x14c (arch/x86/entry/entry_64.S:121)
So we're logging a changed extent from fsync, which is splitting an
extent in the log tree. But this split part already exists in the tree,
triggering the BUG().
This is the state of the log tree at the time of the crash, dumped with
drgn (https://github.com/osandov/drgn/blob/main/contrib/btrfs_tree.py)
to get more details than btrfs_print_leaf() gives us:
>>> print_extent_buffer(prog.crashed_thread().stack_trace()[0]["eb"])
leaf 33439744 level 0 items 72 generation 9 owner 18446744073709551610
leaf 33439744 flags 0x100000000000000
fs uuid e5bd3946-400c-4223-8923-190ef1f18677
chunk uuid d58cb17e-6d02-494a-829a-18b7d8a399da
item 0 key (450 INODE_ITEM 0) itemoff 16123 itemsize 160
generation 7 transid 9 size 8192 nbytes 8473563889606862198
block group 0 mode 100600 links 1 uid 0 gid 0 rdev 0
sequence 204 flags 0x10(PREALLOC)
atime 1716417703.220000000 (2024-05-22 15:41:43)
ctime 1716417704.983333333 (2024-05-22 15:41:44)
mtime 1716417704.983333333 (2024-05-22 15:41:44)
otime 17592186044416.000000000 (559444-03-08 01:40:16)
item 1 key (450 INODE_REF 256) itemoff 16110 itemsize 13
index 195 namelen 3 name: 193
item 2 key (450 XATTR_ITEM 1640047104) itemoff 16073 itemsize 37
location key (0 UNKNOWN.0 0) type XATTR
transid 7 data_len 1 name_len 6
name: user.a
data a
item 3 key (450 EXTENT_DATA 0) itemoff 16020 itemsize 53
generation 9 type 1 (regular)
extent data disk byte 303144960 nr 12288
extent data offset 0 nr 4096 ram 12288
extent compression 0 (none)
item 4 key (450 EXTENT_DATA 4096) itemoff 15967 itemsize 53
generation 9 type 2 (prealloc)
prealloc data disk byte 303144960 nr 12288
prealloc data offset 4096 nr 8192
item 5 key (450 EXTENT_DATA 8192) itemoff 15914 itemsize 53
generation 9 type 2 (prealloc)
prealloc data disk byte 303144960 nr 12288
prealloc data offset 8192 nr 4096
...
So the real problem happened earlier: notice that items 4 (4k-12k) and 5
(8k-12k) overlap. Both are prealloc extents. Item 4 straddles i_size and
item 5 starts at i_size.
Here is the state of the filesystem tree at the time of the crash:
>>> root = prog.crashed_thread().stack_trace()[2]["inode"].root
>>> ret, nodes, slots = btrfs_search_slot(root, BtrfsKey(450, 0, 0))
>>> print_extent_buffer(nodes[0])
leaf 30425088 level 0 items 184 generation 9 owner 5
leaf 30425088 flags 0x100000000000000
fs uuid e5bd3946-400c-4223-8923-190ef1f18677
chunk uuid d58cb17e-6d02-494a-829a-18b7d8a399da
...
item 179 key (450 INODE_ITEM 0) itemoff 4907 itemsize 160
generation 7 transid 7 size 4096 nbytes 12288
block group 0 mode 100600 links 1 uid 0 gid 0 rdev 0
sequence 6 flags 0x10(PREALLOC)
atime 1716417703.220000000 (2024-05-22 15:41:43)
ctime 1716417703.220000000 (2024-05-22 15:41:43)
mtime 1716417703.220000000 (2024-05-22 15:41:43)
otime 1716417703.220000000 (2024-05-22 15:41:43)
item 180 key (450 INODE_REF 256) itemoff 4894 itemsize 13
index 195 namelen 3 name: 193
item 181 key (450 XATTR_ITEM 1640047104) itemoff 4857 itemsize 37
location key (0 UNKNOWN.0 0) type XATTR
transid 7 data_len 1 name_len 6
name: user.a
data a
item 182 key (450 EXTENT_DATA 0) itemoff 4804 itemsize 53
generation 9 type 1 (regular)
extent data disk byte 303144960 nr 12288
extent data offset 0 nr 8192 ram 12288
extent compression 0 (none)
item 183 key (450 EXTENT_DATA 8192) itemoff 4751 itemsize 53
generation 9 type 2 (prealloc)
prealloc data disk byte 303144960 nr 12288
prealloc data offset 8192 nr 4096
Item 5 in the log tree corresponds to item 183 in the filesystem tree,
but nothing matches item 4. Furthermore, item 183 is the last item in
the leaf.
btrfs_log_prealloc_extents() is responsible for logging prealloc extents
beyond i_size. It first truncates any previously logged prealloc extents
that start beyond i_size. Then, it walks the filesystem tree and copies
the prealloc extent items to the log tree.
If it hits the end of a leaf, then it calls btrfs_next_leaf(), which
unlocks the tree and does another search. However, while the filesystem
tree is unlocked, an ordered extent completion may modify the tree. In
particular, it may insert an extent item that overlaps with an extent
item that was already copied to the log tree.
This may manifest in several ways depending on the exact scenario,
including an EEXIST error that is silently translated to a full sync,
overlapping items in the log tree, or this crash. This particular crash
is triggered by the following sequence of events:
- Initially, the file has i_size=4k, a regular extent from 0-4k, and a
prealloc extent beyond i_size from 4k-12k. The prealloc extent item is
the last item in its B-tree leaf.
- The file is fsync'd, which copies its inode item and both extent items
to the log tree.
- An xattr is set on the file, which sets the
BTRFS_INODE_COPY_EVERYTHING flag.
- The range 4k-8k in the file is written using direct I/O. i_size is
extended to 8k, but the ordered extent is still in flight.
- The file is fsync'd. Since BTRFS_INODE_COPY_EVERYTHING is set, this
calls copy_inode_items_to_log(), which calls
btrfs_log_prealloc_extents().
- btrfs_log_prealloc_extents() finds the 4k-12k prealloc extent in the
filesystem tree. Since it starts before i_size, it skips it. Since it
is the last item in its B-tree leaf, it calls btrfs_next_leaf().
- btrfs_next_leaf() unlocks the path.
- The ordered extent completion runs, which converts the 4k-8k part of
the prealloc extent to written and inserts the remaining prealloc part
from 8k-12k.
- btrfs_next_leaf() does a search and finds the new prealloc extent
8k-12k.
- btrfs_log_prealloc_extents() copies the 8k-12k prealloc extent into
the log tree. Note that it overlaps with the 4k-12k prealloc extent
that was copied to the log tree by the first fsync.
- fsync calls btrfs_log_changed_extents(), which tries to log the 4k-8k
extent that was written.
- This tries to drop the range 4k-8k in the log tree, which requires
adjusting the start of the 4k-12k prealloc extent in the log tree to
8k.
- btrfs_set_item_key_safe() sees that there is already an extent
starting at 8k in the log tree and calls BUG().
Fix this by detecting when we're about to insert an overlapping file
extent item in the log tree and truncating the part that would overlap.
CC: stable@vger.kernel.org # 6.1+
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Extent maps are always associated to an inode's extent map tree, so
there's no need to pass the extent map tree explicitly to
clear_em_logging().
In order to facilitate an upcoming change that adds a shrinker for extent
maps, change clear_em_logging() to receive the inode instead of its extent
map tree.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
A comment from Filipe on one of my previous cleanups brought my
attention to a new helper we have for getting the root id of a root,
which makes it easier to read in the code.
The changes where made with the following Coccinelle semantic patch:
// <smpl>
@@
expression E,E1;
@@
(
E->root_key.objectid = E1
|
- E->root_key.objectid
+ btrfs_root_id(E)
)
// </smpl>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ minor style fixups ]
Signed-off-by: David Sterba <dsterba@suse.com>
We consistently use ->num_bytes everywhere through the delayed ref code,
except in btrfs_ref. Rename btrfs_ref to match all the other code.
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We have this in both btrfs_tree_ref and btrfs_data_ref, which is just
wasting space and making the code more complicated. Move this into
btrfs_ref proper and update all the call sites to do the assignment in
btrfs_ref.
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
btrfs_ref currently has ->owning_root, and ->ref_root is shared between
the tree ref and data ref, so in order to move that into btrfs_ref
proper I would need to add another root parameter to the initialization
function. This function has too many arguments, and adding another root
will make it easy to make mistakes about which root goes where.
Drop the generic ref init function and statically initialize the
btrfs_ref in every usage. This makes the code easier to read because we
can see what elements we're assigning, and will make the upcoming change
moving the ref_root into the btrfs_ref more clear and less error prone
than adding a new element to the initialization function.
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Before deciding if we can do a NOCOW write into a range, one of the things
we have to do is check if there are checksum items for that range. We do
that through the btrfs_lookup_csums_list() function, which searches for
checksums and adds them to a list supplied by the caller.
But all we need is to check if there is any checksum, we don't need to
look for all of them and collect them into a list, which requires more
search time in the checksums tree, allocating memory for checksums items
to add to the list, copy checksums from a leaf into those list items,
then free that memory, etc. This is all unnecessary overhead, wasting
mostly CPU time, and perhaps some occasional IO if we need to read from
disk any extent buffers.
So change btrfs_lookup_csums_list() to allow to return immediately in
case it finds any checksum, without the need to add it to a list and read
it from a leaf. This is accomplished by allowing a NULL list parameter and
making the function return 1 if it found any checksum, 0 if it didn't
found any, and a negative value in case of an error.
The following test with fio was used to measure performance:
$ cat test.sh
#!/bin/bash
DEV=/dev/nullb0
MNT=/mnt/nullb0
cat <<EOF > /tmp/fio-job.ini
[global]
name=fio-rand-write
filename=$MNT/fio-rand-write
rw=randwrite
bssplit=4k/20:8k/20:16k/20:32k/20:64k/20
direct=1
numjobs=16
fallocate=posix
time_based
runtime=300
[file1]
size=8G
ioengine=io_uring
iodepth=16
EOF
umount $MNT &> /dev/null
mkfs.btrfs -f $DEV
mount -o ssd $DEV $MNT
fio /tmp/fio-job.ini
umount $MNT
The test was run on a release kernel (Debian's default kernel config).
The results before this patch:
WRITE: bw=139MiB/s (146MB/s), 8204KiB/s-9504KiB/s (8401kB/s-9732kB/s), io=17.0GiB (18.3GB), run=125317-125344msec
The results after this patch:
WRITE: bw=153MiB/s (160MB/s), 9241KiB/s-10.0MiB/s (9463kB/s-10.5MB/s), io=17.0GiB (18.3GB), run=114054-114071msec
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
All the callers of btrfs_lookup_csums_list() pass a value of 0 as the
"search_commit" parameter. So remove it and make the function behave as
to always search from the regular root.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The mod_start and mod_len fields of struct extent_map were introduced by
commit 4e2f84e63d ("Btrfs: improve fsync by filtering extents that we
want") in order to avoid too low performance when fsyncing a file that
keeps getting extent maps merge, because it resulted in each fsync logging
again csum ranges that were already merged before.
We don't need this anymore as extent maps in the list of modified extents
are never merged with other extent maps and once we log an extent map we
remove it from the list of modified extent maps, so it's never logged
twice.
So remove the mod_start and mod_len fields from struct extent_map and use
instead the start and len fields when logging checksums in the fast fsync
path. This also makes EXTENT_FLAG_FILLING unused so remove it as well.
Running the reproducer from the commit mentioned before, with a larger
number of extents and against a null block device, so that IO is fast
and we can better see any impact from searching checksums items and
logging them, gave the following results from dd:
Before this change:
409600000 bytes (410 MB, 391 MiB) copied, 22.948 s, 17.8 MB/s
After this change:
409600000 bytes (410 MB, 391 MiB) copied, 22.9997 s, 17.8 MB/s
So no changes in throughput.
The test was done in a release kernel (non-debug, Debian's default kernel
config) and its steps are the following:
$ mkfs.btrfs -f /dev/nullb0
$ mount /dev/sdb /mnt
$ dd if=/dev/zero of=/mnt/foobar bs=4k count=100000 oflag=sync
$ umount /mnt
This also reduces the size of struct extent_map from 128 bytes down to 112
bytes, so now we can have 36 extents maps per 4K page instead of 32.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The helpers are doing an initialization or release work, none of which
is performance critical that it would require a static inline, so move
them to the .c file.
Signed-off-by: David Sterba <dsterba@suse.com>
When logging an inode and we require to copy items from subvolume leaves
to the log tree, we clone each subvolume leaf and than use that clone to
copy items to the log tree. This is required to avoid possible deadlocks
as stated in commit 796787c978 ("btrfs: do not modify log tree while
holding a leaf from fs tree locked").
The cloning requires allocating an extent buffer (struct extent_buffer)
and then allocating pages (folios) to attach to the extent buffer. This
may be slow in case we are under memory pressure, and since we are doing
the cloning while holding a read lock on a subvolume leaf, it means we
can be blocking other operations on that leaf for significant periods of
time, which can increase latency on operations like creating other files,
renaming files, etc. Similarly because we're under a log transaction, we
may also cause extra delay on other tasks doing an fsync, because syncing
the log requires waiting for tasks that joined a log transaction to exit
the transaction.
So to improve this, for any inode logging operation that needs to copy
items from a subvolume leaf ("full sync" or "copy everything" bit set
in the inode), preallocate a dummy extent buffer before locking any
extent buffer from the subvolume tree, and even before joining a log
transaction, add it to the log context and then use it when we need to
copy items from a subvolume leaf to the log tree. This avoids making
other operations get extra latency when waiting to lock a subvolume
leaf that is used during inode logging and we are under heavy memory
pressure.
The following test script with bonnie++ was used to test this:
$ cat test.sh
#!/bin/bash
DEV=/dev/sdh
MNT=/mnt/sdh
MOUNT_OPTIONS="-o ssd"
MEMTOTAL_BYTES=`free -b | grep Mem: | awk '{ print $2 }'`
NR_DIRECTORIES=20
NR_FILES=20480
DATASET_SIZE=$((MEMTOTAL_BYTES * 2 / 1048576))
DIRECTORY_SIZE=$((MEMTOTAL_BYTES * 2 / NR_FILES))
NR_FILES=$((NR_FILES / 1024))
echo "performance" | \
tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor
umount $DEV &> /dev/null
mkfs.btrfs -f $MKFS_OPTIONS $DEV
mount $MOUNT_OPTIONS $DEV $MNT
bonnie++ -u root -d $MNT \
-n $NR_FILES:$DIRECTORY_SIZE:$DIRECTORY_SIZE:$NR_DIRECTORIES \
-r 0 -s $DATASET_SIZE -b
umount $MNT
The results of this test on a 8G VM running a non-debug kernel (Debian's
default kernel config), were the following.
Before this change:
Version 2.00a ------Sequential Output------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Name:Size etc /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP
debian0 7501M 376k 99 1.4g 96 117m 14 1510k 99 2.5g 95 +++++ +++
Latency 35068us 24976us 2944ms 30725us 71770us 26152us
Version 2.00a ------Sequential Create------ --------Random Create--------
debian0 -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files:max:min /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP
20:384100:384100/20 20480 32 20480 58 20480 48 20480 39 20480 56 20480 61
Latency 411ms 11914us 119ms 617ms 10296us 110ms
After this change:
Version 2.00a ------Sequential Output------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Name:Size etc /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP
debian0 7501M 375k 99 1.4g 97 117m 14 1546k 99 2.3g 98 +++++ +++
Latency 35975us 20945us 2144ms 10297us 2217us 6004us
Version 2.00a ------Sequential Create------ --------Random Create--------
debian0 -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files:max:min /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP
20:384100:384100/20 20480 35 20480 58 20480 48 20480 40 20480 57 20480 59
Latency 320ms 11237us 77779us 518ms 6470us 86389us
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
With help of neovim, LSP and clangd we can identify header files that
are not actually needed to be included in the .c files. This is focused
only on removal (with minor fixups), further cleanups are possible but
will require doing the header files properly with forward declarations,
minimized includes and include-what-you-use care.
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Currently, in struct extent_map, we use an unsigned int (32 bits) to
identify the compression type of an extent and an unsigned long (64 bits
on a 64 bits platform, 32 bits otherwise) for flags. We are only using
6 different flags, so an unsigned long is excessive and we can use flags
to identify the compression type instead of using a dedicated 32 bits
field.
We can easily have tens or hundreds of thousands (or more) of extent maps
on busy and large filesystems, specially with compression enabled or many
or large files with tons of small extents. So it's convenient to have the
extent_map structure as small as possible in order to use less memory.
So remove the compression type field from struct extent_map, use flags
to identify the compression type and shorten the flags field from an
unsigned long to a u32. This saves 8 bytes (on 64 bits platforms) and
reduces the size of the structure from 136 bytes down to 128 bytes, using
now only two cache lines, and increases the number of extent maps we can
have per 4K page from 30 to 32. By using a u32 for the flags instead of
an unsigned long, we no longer use test_bit(), set_bit() and clear_bit(),
but that level of atomicity is not needed as most flags are never cleared
once set (before adding an extent map to the tree), and the ones that can
be cleared or set after an extent map is added to the tree, are always
performed while holding the write lock on the extent map tree, while the
reader holds a lock on the tree or tests for a flag that never changes
once the extent map is in the tree (such as compression flags).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Now that we're not clearing the dirty flag off of extent_buffers in zoned mode,
all that is left of btrfs_redirty_list_add() is a memzero() and some
ASSERT()ions.
As we're also memzero()ing the buffer on write-out btrfs_redirty_list_add()
has become obsolete and can be removed.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
-----BEGIN PGP SIGNATURE-----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=9TVx
-----END PGP SIGNATURE-----
Merge tag 'for-6.7-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux
Pull btrfs updates from David Sterba:
"New features:
- raid-stripe-tree
New tree for logical file extent mapping where the physical mapping
may not match on multiple devices. This is now used in zoned mode
to implement RAID0/RAID1* profiles, but can be used in non-zoned
mode as well. The support for RAID56 is in development and will
eventually fix the problems with the current implementation. This
is a backward incompatible feature and has to be enabled at mkfs
time.
- simple quota accounting (squota)
A simplified mode of qgroup that accounts all space on the initial
extent owners (a subvolume), the snapshots are then cheap to create
and delete. The deletion of snapshots in fully accounting qgroups
is a known CPU/IO performance bottleneck.
The squota is not suitable for the general use case but works well
for containers where the original subvolume exists for the whole
time. This is a backward incompatible feature as it needs extending
some structures, but can be enabled on an existing filesystem.
- temporary filesystem fsid (temp_fsid)
The fsid identifies a filesystem and is hard coded in the
structures, which disallows mounting the same fsid found on
different devices.
For a single device filesystem this is not strictly necessary, a
new temporary fsid can be generated on mount e.g. after a device is
cloned. This will be used by Steam Deck for root partition A/B
testing, or can be used for VM root images.
Other user visible changes:
- filesystems with partially finished metadata_uuid conversion cannot
be mounted anymore and the uuid fixup has to be done by btrfs-progs
(btrfstune).
Performance improvements:
- reduce reservations for checksum deletions (with enabled free space
tree by factor of 4), on a sample workload on file with many
extents the deletion time decreased by 12%
- make extent state merges more efficient during insertions, reduce
rb-tree iterations (run time of critical functions reduced by 5%)
Core changes:
- the integrity check functionality has been removed, this was a
debugging feature and removal does not affect other integrity
checks like checksums or tree-checker
- space reservation changes:
- more efficient delayed ref reservations, this avoids building up
too much work or overusing or exhausting the global block
reserve in some situations
- move delayed refs reservation to the transaction start time,
this prevents some ENOSPC corner cases related to exhaustion of
global reserve
- improvements in reducing excessive reservations for block group
items
- adjust overcommit logic in near full situations, account for one
more chunk to eventually allocate metadata chunk, this is mostly
relevant for small filesystems (<10GiB)
- single device filesystems are scanned but not registered (except
seed devices), this allows temp_fsid to work
- qgroup iterations do not need GFP_ATOMIC allocations anymore
- cleanups, refactoring, reduced data structure size, function
parameter simplifications, error handling fixes"
* tag 'for-6.7-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux: (156 commits)
btrfs: open code timespec64 in struct btrfs_inode
btrfs: remove redundant log root tree index assignment during log sync
btrfs: remove redundant initialization of variable dirty in btrfs_update_time()
btrfs: sysfs: show temp_fsid feature
btrfs: disable the device add feature for temp-fsid
btrfs: disable the seed feature for temp-fsid
btrfs: update comment for temp-fsid, fsid, and metadata_uuid
btrfs: remove pointless empty log context list check when syncing log
btrfs: update comment for struct btrfs_inode::lock
btrfs: remove pointless barrier from btrfs_sync_file()
btrfs: add and use helpers for reading and writing last_trans_committed
btrfs: add and use helpers for reading and writing fs_info->generation
btrfs: add and use helpers for reading and writing log_transid
btrfs: add and use helpers for reading and writing last_log_commit
btrfs: support cloned-device mount capability
btrfs: add helper function find_fsid_by_disk
btrfs: stop reserving excessive space for block group item insertions
btrfs: stop reserving excessive space for block group item updates
btrfs: reorder btrfs_inode to fill gaps
btrfs: open code btrfs_ordered_inode_tree in btrfs_inode
...
During log syncing, when we start updating the log root tree we compute
an index value, stored in variable 'index2', once we lock the log root
tree's mutex. This value depends on the log root's log_transid. And
shortly after we compute again the same value for 'index2' - the value
is exactly the same since we haven't released the mutex and therefore
the log_transid of the log root is the same as before.
This second 'index2' computation became pointless after commit
a93e01682e ("btrfs: remove no longer needed use of log_writers for the
log root tree"). So remove it.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
When syncing the log, if we get an error when updating the log root, we
check first if the log root tree context is in a log context list, and if
so it deletes from the log root tree context from the list. This check
however is pointless because at this moment the context is always in a
list, he have just added it to a context list. The check became pointless
after commit a93e01682e ("btrfs: remove no longer needed use of
log_writers for the log root tree"). So remove this now pointless empty
list check.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Currently the log_transid field of a root is always modified while holding
the root's log_mutex locked. Most readers of a root's log_transid are also
holding the root's log_mutex locked, however there is one exception which
is btrfs_set_inode_last_trans() where we don't take the lock to avoid
blocking several operations if log syncing is happening in parallel.
Any races here should be harmless, and in the worst case they may cause a
fsync to log an inode when it's not really needed, so nothing bad from a
functional perspective.
To avoid data race warnings from tools like KCSAN and other issues such
as load and store tearing (amongst others, see [1]), create helpers to
access the log_transid field of a root using READ_ONCE() and WRITE_ONCE(),
and use these helpers where needed.
[1] https://lwn.net/Articles/793253/
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Currently, the last_log_commit of a root can be accessed concurrently
without any lock protection. Readers can be calling btrfs_inode_in_log()
early in a fsync call, which reads a root's last_log_commit, while a
writer can change the last_log_commit while a log tree if being synced,
at btrfs_sync_log(). Any races here should be harmless, and in the worst
case they may cause a fsync to log an inode when it's not really needed,
so nothing bad from a functional perspective.
To avoid data race warnings from tools like KCSAN and other issues such
as load and store tearing (amongst others, see [1]), create helpers to
access the last_log_commit field of a root using READ_ONCE() and
WRITE_ONCE(), and use these helpers everywhere.
[1] https://lwn.net/Articles/793253/
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The structure btrfs_ordered_inode_tree is used only in one place, in
btrfs_inode. The structure itself has a 4 byte hole which is wasted
space.
Move the btrfs_ordered_inode_tree members to btrfs_inode with a common
prefix 'ordered_tree_' where the hole can be utilized and shrink inode
size.
Signed-off-by: David Sterba <dsterba@suse.com>
When freeing a log tree, during a transaction commit, we clear its dirty
log pages io tree by calling clear_extent_bits() using a range from 0 to
(u64)-1. This will iterate the io tree's rbtree and call rb_erase() on
each node before freeing it, which will often trigger rebalance operations
on the rbtree. A better alternative it to use extent_io_tree_release(),
which will not do deletions and trigger rebalances.
So use extent_io_tree_release() instead of clear_extent_bits().
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The root argument for fixup_inode_link_count() always matches the root of
the given inode, so remove the root argument and get it from the inode
argument. This also applies to the helpers count_inode_extrefs() and
count_inode_refs() used by fixup_inode_link_count() - they don't need the
root argument, as it always matches the root of the inode passed to them.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The root argument for btrfs_update_inode() always matches the root of the
given inode, so remove the root argument and get it from the inode
argument.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
While data extents require us to store additional inline refs to track
the original owner on free, this information is available implicitly for
metadata. It is found in the owner field of the header of the tree
block. Even if other trees refer to this block and the original ref goes
away, we will not rewrite that header field, so it will reliably give the
original owner.
In addition, there is a relocation case where a new data extent needs to
have an owning root separate from the referring root wired through
delayed refs.
To use it for recording simple quota deltas, we need to wire this root
id through from when we create the delayed ref until we fully process
it. Store it in the generic btrfs_ref struct of the delayed ref.
Signed-off-by: Boris Burkov <boris@bur.io>
Signed-off-by: David Sterba <dsterba@suse.com>
When marking an extent buffer as dirty, at btrfs_mark_buffer_dirty(),
we check if its generation matches the running transaction and if not we
just print a warning. Such mismatch is an indicator that something really
went wrong and only printing a warning message (and stack trace) is not
enough to prevent a corruption. Allowing a transaction to commit with such
an extent buffer will trigger an error if we ever try to read it from disk
due to a generation mismatch with its parent generation.
So abort the current transaction with -EUCLEAN if we notice a generation
mismatch. For this we need to pass a transaction handle to
btrfs_mark_buffer_dirty() which is always available except in test code,
in which case we can pass NULL since it operates on dummy extent buffers
and all test roots have a single node/leaf (root node at level 0).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Both callers of btrfs_pin_extent_for_log_replay expand the parameters to
extent buffer members. We can simply pass the extent buffer instead.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
There is only one caller of btrfs_pin_reserved_extent that expands the
parameters to extent buffer members. We can simply pass the extent
buffer instead.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Function name in the comment does not bring much value to code not
exposed as API and we don't stick to the kdoc format anymore. Update
formatting of parameter descriptions.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
-----BEGIN PGP SIGNATURE-----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=6WEW
-----END PGP SIGNATURE-----
Merge tag 'for-6.6-rc3-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux
Pull btrfs fixes from David Sterba:
- delayed refs fixes:
- fix race when refilling delayed refs block reserve
- prevent transaction block reserve underflow when starting
transaction
- error message and value adjustments
- fix build warnings with CONFIG_CC_OPTIMIZE_FOR_SIZE and
-Wmaybe-uninitialized
- fix for smatch report where uninitialized data from invalid extent
buffer range could be returned to the caller
- fix numeric overflow in statfs when calculating lower threshold
for a full filesystem
* tag 'for-6.6-rc3-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
btrfs: initialize start_slot in btrfs_log_prealloc_extents
btrfs: make sure to initialize start and len in find_free_dev_extent
btrfs: reset destination buffer when read_extent_buffer() gets invalid range
btrfs: properly report 0 avail for very full file systems
btrfs: log message if extent item not found when running delayed extent op
btrfs: remove redundant BUG_ON() from __btrfs_inc_extent_ref()
btrfs: return -EUCLEAN for delayed tree ref with a ref count not equals to 1
btrfs: prevent transaction block reserve underflow when starting transaction
btrfs: fix race when refilling delayed refs block reserve
Jens reported a compiler warning when using
CONFIG_CC_OPTIMIZE_FOR_SIZE=y that looks like this
fs/btrfs/tree-log.c: In function ‘btrfs_log_prealloc_extents’:
fs/btrfs/tree-log.c:4828:23: warning: ‘start_slot’ may be used
uninitialized [-Wmaybe-uninitialized]
4828 | ret = copy_items(trans, inode, dst_path, path,
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
4829 | start_slot, ins_nr, 1, 0);
| ~~~~~~~~~~~~~~~~~~~~~~~~~
fs/btrfs/tree-log.c:4725:13: note: ‘start_slot’ was declared here
4725 | int start_slot;
| ^~~~~~~~~~
The compiler is incorrect, as we only use this code when ins_len > 0,
and when ins_len > 0 we have start_slot properly initialized. However
we generally find the -Wmaybe-uninitialized warnings valuable, so
initialize start_slot to get rid of the warning.
Reported-by: Jens Axboe <axboe@kernel.dk>
Tested-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
-----BEGIN PGP SIGNATURE-----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=N7Yn
-----END PGP SIGNATURE-----
Merge tag 'for-6.6-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux
Pull btrfs updates from David Sterba:
"No new features, the bulk of the changes are fixes, refactoring and
cleanups. The notable fix is the scrub performance restoration after
rewrite in 6.4, though still only partial.
Fixes:
- scrub performance drop due to rewrite in 6.4 partially restored:
- do IO grouping by blg_plug/blk_unplug again
- avoid unnecessary tree searches when processing stripes, in
extent and checksum trees
- the drop is noticeable on fast PCIe devices, -66% and restored
to -33% of the original
- backports to 6.4 planned
- handle more corner cases of transaction commit during orphan
cleanup or delayed ref processing
- use correct fsid/metadata_uuid when validating super block
- copy directory permissions and time when creating a stub subvolume
Core:
- debugging feature integrity checker deprecated, to be removed in
6.7
- in zoned mode, zones are activated just before the write, making
error handling easier, now the overcommit mechanism can be enabled
again which improves performance by avoiding more frequent flushing
- v0 extent handling completely removed, deprecated long time ago
- error handling improvements
- tests:
- extent buffer bitmap tests
- pinned extent splitting tests
- cleanups and refactoring:
- compression writeback
- extent buffer bitmap
- space flushing, ENOSPC handling"
* tag 'for-6.6-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux: (110 commits)
btrfs: zoned: skip splitting and logical rewriting on pre-alloc write
btrfs: tests: test invalid splitting when skipping pinned drop extent_map
btrfs: tests: add a test for btrfs_add_extent_mapping
btrfs: tests: add extent_map tests for dropping with odd layouts
btrfs: scrub: move write back of repaired sectors to scrub_stripe_read_repair_worker()
btrfs: scrub: don't go ordered workqueue for dev-replace
btrfs: scrub: fix grouping of read IO
btrfs: scrub: avoid unnecessary csum tree search preparing stripes
btrfs: scrub: avoid unnecessary extent tree search preparing stripes
btrfs: copy dir permission and time when creating a stub subvolume
btrfs: remove pointless empty list check when reading delayed dir indexes
btrfs: drop redundant check to use fs_devices::metadata_uuid
btrfs: compare the correct fsid/metadata_uuid in btrfs_validate_super
btrfs: use the correct superblock to compare fsid in btrfs_validate_super
btrfs: simplify memcpy either of metadata_uuid or fsid
btrfs: add a helper to read the superblock metadata_uuid
btrfs: remove v0 extent handling
btrfs: output extra debug info if we failed to find an inline backref
btrfs: move the !zoned assert into run_delalloc_cow
btrfs: consolidate the error handling in run_delalloc_nocow
...
Use LIST_HEAD() to initialize the list_head instead of open-coding it.
Signed-off-by: Ruan Jinjie <ruanjinjie@huawei.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The variables leaf and slot are initialized when declared but the values
assigned to them are never read as they are being re-assigned later on.
The initializations are redundant and can be removed. Cleans up clang
scan build warnings:
fs/btrfs/tree-log.c:6797:25: warning: Value stored to 'leaf' during its
initialization is never read [deadcode.DeadStores]
fs/btrfs/tree-log.c:6798:7: warning: Value stored to 'slot' during its
initialization is never read [deadcode.DeadStores]
It's been there since b8aa330d2a ("Btrfs: improve performance on fsync
of files with multiple hardlinks") without any usage so it's safe to be
removed.
Signed-off-by: Colin Ian King <colin.i.king@gmail.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
In later patches, we're going to change how the inode's ctime field is
used. Switch to using accessor functions instead of raw accesses of
inode->i_ctime.
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: Jan Kara <jack@suse.cz>
Message-Id: <20230705190309.579783-27-jlayton@kernel.org>
Signed-off-by: Christian Brauner <brauner@kernel.org>
When dropping inode items from a log tree at drop_inode_items(), we this
BUG_ON() on the result of btrfs_search_slot() because we don't expect an
exact match since having a key with an offset of (u64)-1 is unexpected.
That is generally true, but for dir index keys for example, we can get a
key with such an offset value, even though it's very unlikely and it would
take ages to increase the sequence counter for a dir index up to (u64)-1.
We can deal with an exact match, we just have to delete the key at that
slot, so there is really no need to BUG_ON(), error out or trigger any
warning. So remove the BUG_ON().
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
btrfs_ordered_sum::bytendr stores a logical address. Make that clear by
renaming it to ->logical.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The for_rename argument of btrfs_record_unlink_dir() is defined as an
integer, but the argument is in fact used as a boolean. So change it to
a boolean to make its use more clear.
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
There's no point of having a label and goto at btrfs_record_unlink_dir()
because the function is trivial and can just return early if we are not
in a rename context. So remove the label and goto and instead return
early if we are not in a rename.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Update the comments at btrfs_record_unlink_dir() so that they mention
where new names are logged and where old names are removed. Also, while
at it make the width of the comments closer to 80 columns and capitalize
the sentences and finish them with punctuation.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
At btrfs_record_unlink_dir() we directly check the logged_trans field of
the given inodes to check if they were previously logged in the current
transaction, and if any of them were, then we can avoid setting the field
last_unlink_trans of the directory to the id of the current transaction if
we are in a rename path. Avoiding that can later prevent falling back to
a transaction commit if anyone attempts to log the directory.
However the logged_trans field, store in struct btrfs_inode, is transient,
not persisted in the inode item on its subvolume b+tree, so that means
that if an inode is evicted and then loaded again, its original value is
lost and it's reset to 0. So directly checking the logged_trans field can
lead to some false negative, and that only results in a performance impact
as mentioned before.
Instead of directly checking the logged_trans field of the inodes, use the
inode_logged() helper, which will check in the log tree if an inode was
logged before in case its logged_trans field has a value of 0. This way
we can avoid setting the directory inode's last_unlink_trans and cause
future logging attempts of it to fallback to transaction commits. The
following test script shows one example where this happens without this
patch:
$ cat test.sh
#!/bin/bash
DEV=/dev/nullb0
MNT=/mnt/nullb0
num_init_files=10000
num_new_files=10000
mkfs.btrfs -f $DEV
mount -o ssd $DEV $MNT
mkdir $MNT/testdir
for ((i = 1; i <= $num_init_files; i++)); do
echo -n > $MNT/testdir/file_$i
done
echo -n > $MNT/testdir/foo
sync
# Add some files so that there's more work in the transaction other
# than just renaming file foo.
for ((i = 1; i <= $num_new_files; i++)); do
echo -n > $MNT/testdir/new_file_$i
done
# Change the file, fsync it.
setfattr -n user.x1 -v 123 $MNT/testdir/foo
xfs_io -c "fsync" $MNT/testdir/foo
# Now triggger eviction of file foo but no eviction for our test
# directory, since it is being used by the process below. This will
# set logged_trans of the file's inode to 0 once it is loaded again.
(
cd $MNT/testdir
while true; do
:
done
) &
pid=$!
echo 2 > /proc/sys/vm/drop_caches
kill $pid
wait $pid
# Move foo out of our testdir. This will set last_unlink_trans
# of the directory inode to the current transaction, because
# logged_trans of both the directory and the file are set to 0.
mv $MNT/testdir/foo $MNT/foo
# Change file foo again and fsync it.
# This fsync will result in a transaction commit because the rename
# above has set last_unlink_trans of the parent directory to the id
# of the current transaction and because our inode for file foo has
# last_unlink_trans set to the current transaction, since it was
# evicted and reloaded and it was previously modified in the current
# transaction (the xattr addition).
xfs_io -c "pwrite 0 64K" $MNT/foo
start=$(date +%s%N)
xfs_io -c "fsync" $MNT/foo
end=$(date +%s%N)
dur=$(( (end - start) / 1000000 ))
echo "file fsync took: $dur milliseconds"
umount $MNT
Before this patch: fsync took 19 milliseconds
After this patch: fsync took 5 milliseconds
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
At need_log_inode() we directly check the ->logged_trans field of the
given inode to check if it was previously logged in the transaction, with
the goal of skipping logging the inode again when it's not necessary.
The ->logged_trans field in not persisted in the inode item or elsewhere,
it's only stored in memory (struct btrfs_inode), so it's transient and
lost once the inode is evicted and then loaded again. Once an inode is
loaded, we are conservative and set ->logged_trans to 0, which may mean
that either the inode was never logged in the current transaction or it
was logged but evicted before being loaded again.
Instead of checking the inode's ->logged_trans field directly, we can
use instead the helper inode_logged(), which will really check if the
inode was logged before in the current transaction in case we have a
->logged_trans field with a value of 0. This will prevent unnecessarily
logging an inode when it's not needed, and in some cases preventing a
transaction commit, in case the logging requires a fallback to a
transaction commit. The following test script shows a scenario where
due to eviction we fallback a transaction commit when trying to fsync
a file that was renamed:
$ cat test.sh
#!/bin/bash
DEV=/dev/nullb0
MNT=/mnt/nullb0
num_init_files=10000
num_new_files=10000
mkfs.btrfs -f $DEV
mount -o ssd $DEV $MNT
mkdir $MNT/testdir
for ((i = 1; i <= $num_init_files; i++)); do
echo -n > $MNT/testdir/file_$i
done
echo -n > $MNT/testdir/foo
sync
# Add some files so that there's more work in the transaction other
# than just renaming file foo.
for ((i = 1; i <= $num_new_files; i++)); do
echo -n > $MNT/testdir/new_file_$i
done
# Fsync the directory first.
xfs_io -c "fsync" $MNT/testdir
# Rename file foo.
mv $MNT/testdir/foo $MNT/testdir/bar
# Now trigger eviction of the test directory's inode.
# Once loaded again, it will have logged_trans set to 0 and
# last_unlink_trans set to the current transaction.
echo 2 > /proc/sys/vm/drop_caches
# Fsync file bar (ex-foo).
# Before the patch the fsync would result in a transaction commit
# because the inode for file bar has last_unlink_trans set to the
# current transaction, so it will attempt to log the parent directory
# as well, which will fallback to a full transaction commit because
# it also has its last_unlink_trans set to the current transaction,
# due to the inode eviction.
start=$(date +%s%N)
xfs_io -c "fsync" $MNT/testdir/bar
end=$(date +%s%N)
dur=$(( (end - start) / 1000000 ))
echo "file fsync took: $dur milliseconds"
umount $MNT
Before this patch: fsync took 22 milliseconds
After this patch: fsync took 8 milliseconds
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
This fixes the following warning reported by gcc 10.2.1 under x86_64:
../fs/btrfs/tree-log.c: In function ‘btrfs_log_inode’:
../fs/btrfs/tree-log.c:6211:9: error: ‘last_range_start’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
6211 | ret = insert_dir_log_key(trans, log, path, key.objectid,
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
6212 | first_dir_index, last_dir_index);
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
../fs/btrfs/tree-log.c:6161:6: note: ‘last_range_start’ was declared here
6161 | u64 last_range_start;
| ^~~~~~~~~~~~~~~~
This might be a false positive fixed in later compiler versions but we
want to have it fixed.
Reported-by: k2ci <kernel-bot@kylinos.cn>
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Shida Zhang <zhangshida@kylinos.cn>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
When logging dir dentries of a directory, we iterate over the subvolume
tree to find dir index keys on leaves modified in the current transaction.
This however is heavy on locking, since btrfs_search_forward() may often
keep locks on extent buffers for quite a while when walking the tree to
find a suitable leaf modified in the current transaction and with a key
not smaller than then the provided minimum key. That means it will block
other tasks trying to access the subvolume tree, which may be common fs
operations like creating, renaming, linking, unlinking, reflinking files,
etc.
A better solution is to iterate the log tree, since it's much smaller than
a subvolume tree and just use plain btrfs_search_slot() (or the wrapper
btrfs_for_each_slot()) and only contains dir index keys added in the
current transaction.
The following bonnie++ test on a non-debug kernel (with Debian's default
kernel config) on a 20G null block device, was used to measure the impact:
$ cat test.sh
#!/bin/bash
DEV=/dev/nullb0
MNT=/mnt/nullb0
NR_DIRECTORIES=20
NR_FILES=20480 # must be a multiple of 1024
DATASET_SIZE=$(( (8 * 1024 * 1024 * 1024) / 1048576 )) # 8 GiB as megabytes
DIRECTORY_SIZE=$(( DATASET_SIZE / NR_FILES ))
NR_FILES=$(( NR_FILES / 1024 ))
umount $DEV &> /dev/null
mkfs.btrfs -f $DEV
mount $DEV $MNT
bonnie++ -u root -d $MNT \
-n $NR_FILES:$DIRECTORY_SIZE:$DIRECTORY_SIZE:$NR_DIRECTORIES \
-r 0 -s $DATASET_SIZE -b
umount $MNT
Before patchset:
Version 2.00a ------Sequential Output------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Name:Size etc /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP
debian0 8G 376k 99 1.1g 98 939m 92 1527k 99 3.2g 99 9060 256
Latency 24920us 207us 680ms 5594us 171us 2891us
Version 2.00a ------Sequential Create------ --------Random Create--------
debian0 -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP
20/20 20480 96 +++++ +++ 20480 95 20480 99 +++++ +++ 20480 97
Latency 8708us 137us 5128us 6743us 60us 19712us
After patchset:
Version 2.00a ------Sequential Output------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Name:Size etc /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP
debian0 8G 384k 99 1.2g 99 971m 91 1533k 99 3.3g 99 9180 309
Latency 24930us 125us 661ms 5587us 46us 2020us
Version 2.00a ------Sequential Create------ --------Random Create--------
debian0 -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP
20/20 20480 90 +++++ +++ 20480 99 20480 99 +++++ +++ 20480 97
Latency 7030us 61us 1246us 4942us 56us 16855us
The patchset consists of this patch plus a previous one that has the
following subject:
"btrfs: avoid iterating over all indexes when logging directory"
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
When logging a directory, after copying all directory index items from the
subvolume tree to the log tree, we iterate over the subvolume tree to find
all dir index items that are located in leaves COWed (or created) in the
current transaction. If we keep logging a directory several times during
the same transaction, we end up iterating over the same dir index items
everytime we log the directory, wasting time and adding extra lock
contention on the subvolume tree.
So just keep track of the last logged dir index offset in order to start
the search for that index (+1) the next time the directory is logged, as
dir index values (key offsets) come from a monotonically increasing
counter.
The following test measures the difference before and after this change:
$ cat test.sh
#!/bin/bash
DEV=/dev/nullb0
MNT=/mnt/nullb0
umount $DEV &> /dev/null
mkfs.btrfs -f $DEV
mount -o ssd $DEV $MNT
# Time values in milliseconds.
declare -a fsync_times
# Total number of files added to the test directory.
num_files=1000000
# Fsync directory after every N files are added.
fsync_period=100
mkdir $MNT/testdir
fsync_total_time=0
for ((i = 1; i <= $num_files; i++)); do
echo -n > $MNT/testdir/file_$i
if [ $((i % fsync_period)) -eq 0 ]; then
start=$(date +%s%N)
xfs_io -c "fsync" $MNT/testdir
end=$(date +%s%N)
fsync_total_time=$((fsync_total_time + (end - start)))
fsync_times[i]=$(( (end - start) / 1000000 ))
echo -n -e "Progress $i / $num_files\r"
fi
done
echo -e "\nHistogram of directory fsync duration in ms:\n"
printf '%s\n' "${fsync_times[@]}" | \
perl -MStatistics::Histogram -e '@d = <>; print get_histogram(\@d);'
fsync_total_time=$((fsync_total_time / 1000000))
echo -e "\nTotal time spent in fsync: $fsync_total_time ms\n"
echo
umount $MNT
The test was run on a non-debug kernel (Debian's default kernel config)
against a 15G null block device.
Result before this change:
Histogram of directory fsync duration in ms:
Count: 10000
Range: 3.000 - 362.000; Mean: 34.556; Median: 31.000; Stddev: 25.751
Percentiles: 90th: 71.000; 95th: 77.000; 99th: 81.000
3.000 - 5.278: 1423 #################################
5.278 - 8.854: 1173 ###########################
8.854 - 14.467: 591 ##############
14.467 - 23.277: 1025 #######################
23.277 - 37.105: 1422 #################################
37.105 - 58.809: 2036 ###############################################
58.809 - 92.876: 2316 #####################################################
92.876 - 146.346: 6 |
146.346 - 230.271: 6 |
230.271 - 362.000: 2 |
Total time spent in fsync: 350527 ms
Result after this change:
Histogram of directory fsync duration in ms:
Count: 10000
Range: 3.000 - 1088.000; Mean: 8.704; Median: 8.000; Stddev: 12.576
Percentiles: 90th: 12.000; 95th: 14.000; 99th: 17.000
3.000 - 6.007: 3222 #################################
6.007 - 11.276: 5197 #####################################################
11.276 - 20.506: 1551 ################
20.506 - 36.674: 24 |
36.674 - 201.552: 1 |
201.552 - 353.841: 4 |
353.841 - 1088.000: 1 |
Total time spent in fsync: 92114 ms
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The tree-log code has three almost identical copies for the accounting on
an extent_buffer that doesn't need to be written any more. The only
difference is that walk_down_log_tree passed the bytenr used to find the
buffer instead of extent_buffer.start and calculates the length using the
nodesize, while the other two callers look at the extent_buffer.len
field that must always be equivalent to the nodesize.
Factor the code into a common helper.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
btrfs_bin_search() is a simple wrapper that searches for the whole slots
by calling btrfs_generic_bin_search() with the starting slot/first_slot
preset to 0.
This simple wrapper can be open coded as btrfs_bin_search().
Signed-off-by: Anand Jain <anand.jain@oracle.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
This is used in the tree-log code and is a holdover from previous
iterations of extent buffer writeback. We can simply use
wait_on_extent_buffer_writeback here, and remove
btrfs_wait_tree_block_writeback completely as it's equivalent (waiting
on page write writeback).
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
btrfs_clean_tree_block is a misnomer, it's just
clear_extent_buffer_dirty with some extra accounting around it. Rename
this to btrfs_clear_buffer_dirty to make it more clear it belongs with
it's setter, btrfs_mark_buffer_dirty.
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Now that we're passing in the trans into btrfs_clean_tree_block, we can
easily roll in the handling of the !trans case and replace all
occurrences of
if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags))
clear_extent_buffer_dirty(eb);
with
btrfs_tree_lock(eb);
btrfs_clean_tree_block(eb);
btrfs_tree_unlock(eb);
We need the lock because if we are actually dirty we need to make sure
we aren't racing with anything that's starting writeout currently. This
also makes sure that we're accounting fs_info->dirty_metadata_bytes
appropriately.
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Omar Sandoval