Commit 8d875f95da ("btrfs: disable strict file flushes for
renames and truncates") eliminated the notion of ordered operations and
instead BTRFS_INODE_ORDERED_DATA_CLOSE only remained as a flag
indicating that a file's content should be synced to disk in case a
file is truncated and any writes happen to it concurrently. In fact
this intendend behavior was broken until it was fixed in
f6dc45c7a9 ("Btrfs: fix filemap_flush call in btrfs_file_release").
All things considered let's give the flag a more descriptive name. Also
slightly reword comments.
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Since we now perform direct reads using i_rwsem, we can remove this
inode flag used to co-ordinate unlocked reads.
The truncate call takes i_rwsem. This means it is correctly synchronized
with concurrent direct reads.
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Johannes Thumshirn <jth@kernel.org>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
It's counterintuitive to have a function named btrfs_inode_xxx which
takes a generic inode. Also move the function to btrfs_inode.h so that
it has access to the definition of struct btrfs_inode.
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Currently regardless of a full or a fast fsync we always wait for ordered
extents to complete, and then start logging the inode after that. However
for fast fsyncs we can just wait for the writeback to complete, we don't
need to wait for the ordered extents to complete since we use the list of
modified extents maps to figure out which extents we must log and we can
get their checksums directly from the ordered extents that are still in
flight, otherwise look them up from the checksums tree.
Until commit b5e6c3e170 ("btrfs: always wait on ordered extents at
fsync time"), for fast fsyncs, we used to start logging without even
waiting for the writeback to complete first, we would wait for it to
complete after logging, while holding a transaction open, which lead to
performance issues when using cgroups and probably for other cases too,
as wait for IO while holding a transaction handle should be avoided as
much as possible. After that, for fast fsyncs, we started to wait for
ordered extents to complete before starting to log, which adds some
latency to fsyncs and we even got at least one report about a performance
drop which bisected to that particular change:
https://lore.kernel.org/linux-btrfs/20181109215148.GF23260@techsingularity.net/
This change makes fast fsyncs only wait for writeback to finish before
starting to log the inode, instead of waiting for both the writeback to
finish and for the ordered extents to complete. This brings back part of
the logic we had that extracts checksums from in flight ordered extents,
which are not yet in the checksums tree, and making sure transaction
commits wait for the completion of ordered extents previously logged
(by far most of the time they have already completed by the time a
transaction commit starts, resulting in no wait at all), to avoid any
data loss if an ordered extent completes after the transaction used to
log an inode is committed, followed by a power failure.
When there are no other tasks accessing the checksums and the subvolume
btrees, the ordered extent completion is pretty fast, typically taking
100 to 200 microseconds only in my observations. However when there are
other tasks accessing these btrees, ordered extent completion can take a
lot more time due to lock contention on nodes and leaves of these btrees.
I've seen cases over 2 milliseconds, which starts to be significant. In
particular when we do have concurrent fsyncs against different files there
is a lot of contention on the checksums btree, since we have many tasks
writing the checksums into the btree and other tasks that already started
the logging phase are doing lookups for checksums in the btree.
This change also turns all ranged fsyncs into full ranged fsyncs, which
is something we already did when not using the NO_HOLES features or when
doing a full fsync. This is to guarantee we never miss checksums due to
writeback having been triggered only for a part of an extent, and we end
up logging the full extent but only checksums for the written range, which
results in missing checksums after log replay. Allowing ranged fsyncs to
operate again only in the original range, when using the NO_HOLES feature
and doing a fast fsync is doable but requires some non trivial changes to
the writeback path, which can always be worked on later if needed, but I
don't think they are a very common use case.
Several tests were performed using fio for different numbers of concurrent
jobs, each writing and fsyncing its own file, for both sequential and
random file writes. The tests were run on bare metal, no virtualization,
on a box with 12 cores (Intel i7-8700), 64Gb of RAM and a NVMe device,
with a kernel configuration that is the default of typical distributions
(debian in this case), without debug options enabled (kasan, kmemleak,
slub debug, debug of page allocations, lock debugging, etc).
The following script that calls fio was used:
$ cat test-fsync.sh
#!/bin/bash
DEV=/dev/nvme0n1
MNT=/mnt/btrfs
MOUNT_OPTIONS="-o ssd -o space_cache=v2"
MKFS_OPTIONS="-d single -m single"
if [ $# -ne 5 ]; then
echo "Use $0 NUM_JOBS FILE_SIZE FSYNC_FREQ BLOCK_SIZE [write|randwrite]"
exit 1
fi
NUM_JOBS=$1
FILE_SIZE=$2
FSYNC_FREQ=$3
BLOCK_SIZE=$4
WRITE_MODE=$5
if [ "$WRITE_MODE" != "write" ] && [ "$WRITE_MODE" != "randwrite" ]; then
echo "Invalid WRITE_MODE, must be 'write' or 'randwrite'"
exit 1
fi
cat <<EOF > /tmp/fio-job.ini
[writers]
rw=$WRITE_MODE
fsync=$FSYNC_FREQ
fallocate=none
group_reporting=1
direct=0
bs=$BLOCK_SIZE
ioengine=sync
size=$FILE_SIZE
directory=$MNT
numjobs=$NUM_JOBS
EOF
echo "performance" | tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor
echo
echo "Using config:"
echo
cat /tmp/fio-job.ini
echo
umount $MNT &> /dev/null
mkfs.btrfs -f $MKFS_OPTIONS $DEV
mount $MOUNT_OPTIONS $DEV $MNT
fio /tmp/fio-job.ini
umount $MNT
The results were the following:
*************************
*** sequential writes ***
*************************
==== 1 job, 8GiB file, fsync frequency 1, block size 64KiB ====
Before patch:
WRITE: bw=36.6MiB/s (38.4MB/s), 36.6MiB/s-36.6MiB/s (38.4MB/s-38.4MB/s), io=8192MiB (8590MB), run=223689-223689msec
After patch:
WRITE: bw=40.2MiB/s (42.1MB/s), 40.2MiB/s-40.2MiB/s (42.1MB/s-42.1MB/s), io=8192MiB (8590MB), run=203980-203980msec
(+9.8%, -8.8% runtime)
==== 2 jobs, 4GiB files, fsync frequency 1, block size 64KiB ====
Before patch:
WRITE: bw=35.8MiB/s (37.5MB/s), 35.8MiB/s-35.8MiB/s (37.5MB/s-37.5MB/s), io=8192MiB (8590MB), run=228950-228950msec
After patch:
WRITE: bw=43.5MiB/s (45.6MB/s), 43.5MiB/s-43.5MiB/s (45.6MB/s-45.6MB/s), io=8192MiB (8590MB), run=188272-188272msec
(+21.5% throughput, -17.8% runtime)
==== 4 jobs, 2GiB files, fsync frequency 1, block size 64KiB ====
Before patch:
WRITE: bw=50.1MiB/s (52.6MB/s), 50.1MiB/s-50.1MiB/s (52.6MB/s-52.6MB/s), io=8192MiB (8590MB), run=163446-163446msec
After patch:
WRITE: bw=64.5MiB/s (67.6MB/s), 64.5MiB/s-64.5MiB/s (67.6MB/s-67.6MB/s), io=8192MiB (8590MB), run=126987-126987msec
(+28.7% throughput, -22.3% runtime)
==== 8 jobs, 1GiB files, fsync frequency 1, block size 64KiB ====
Before patch:
WRITE: bw=64.0MiB/s (68.1MB/s), 64.0MiB/s-64.0MiB/s (68.1MB/s-68.1MB/s), io=8192MiB (8590MB), run=126075-126075msec
After patch:
WRITE: bw=86.8MiB/s (91.0MB/s), 86.8MiB/s-86.8MiB/s (91.0MB/s-91.0MB/s), io=8192MiB (8590MB), run=94358-94358msec
(+35.6% throughput, -25.2% runtime)
==== 16 jobs, 512MiB files, fsync frequency 1, block size 64KiB ====
Before patch:
WRITE: bw=79.8MiB/s (83.6MB/s), 79.8MiB/s-79.8MiB/s (83.6MB/s-83.6MB/s), io=8192MiB (8590MB), run=102694-102694msec
After patch:
WRITE: bw=107MiB/s (112MB/s), 107MiB/s-107MiB/s (112MB/s-112MB/s), io=8192MiB (8590MB), run=76446-76446msec
(+34.1% throughput, -25.6% runtime)
==== 32 jobs, 512MiB files, fsync frequency 1, block size 64KiB ====
Before patch:
WRITE: bw=93.2MiB/s (97.7MB/s), 93.2MiB/s-93.2MiB/s (97.7MB/s-97.7MB/s), io=16.0GiB (17.2GB), run=175836-175836msec
After patch:
WRITE: bw=111MiB/s (117MB/s), 111MiB/s-111MiB/s (117MB/s-117MB/s), io=16.0GiB (17.2GB), run=147001-147001msec
(+19.1% throughput, -16.4% runtime)
==== 64 jobs, 512MiB files, fsync frequency 1, block size 64KiB ====
Before patch:
WRITE: bw=108MiB/s (114MB/s), 108MiB/s-108MiB/s (114MB/s-114MB/s), io=32.0GiB (34.4GB), run=302656-302656msec
After patch:
WRITE: bw=133MiB/s (140MB/s), 133MiB/s-133MiB/s (140MB/s-140MB/s), io=32.0GiB (34.4GB), run=246003-246003msec
(+23.1% throughput, -18.7% runtime)
************************
*** random writes ***
************************
==== 1 job, 8GiB file, fsync frequency 16, block size 4KiB ====
Before patch:
WRITE: bw=11.5MiB/s (12.0MB/s), 11.5MiB/s-11.5MiB/s (12.0MB/s-12.0MB/s), io=8192MiB (8590MB), run=714281-714281msec
After patch:
WRITE: bw=11.6MiB/s (12.2MB/s), 11.6MiB/s-11.6MiB/s (12.2MB/s-12.2MB/s), io=8192MiB (8590MB), run=705959-705959msec
(+0.9% throughput, -1.7% runtime)
==== 2 jobs, 4GiB files, fsync frequency 16, block size 4KiB ====
Before patch:
WRITE: bw=12.8MiB/s (13.5MB/s), 12.8MiB/s-12.8MiB/s (13.5MB/s-13.5MB/s), io=8192MiB (8590MB), run=638101-638101msec
After patch:
WRITE: bw=13.1MiB/s (13.7MB/s), 13.1MiB/s-13.1MiB/s (13.7MB/s-13.7MB/s), io=8192MiB (8590MB), run=625374-625374msec
(+2.3% throughput, -2.0% runtime)
==== 4 jobs, 2GiB files, fsync frequency 16, block size 4KiB ====
Before patch:
WRITE: bw=15.4MiB/s (16.2MB/s), 15.4MiB/s-15.4MiB/s (16.2MB/s-16.2MB/s), io=8192MiB (8590MB), run=531146-531146msec
After patch:
WRITE: bw=17.8MiB/s (18.7MB/s), 17.8MiB/s-17.8MiB/s (18.7MB/s-18.7MB/s), io=8192MiB (8590MB), run=460431-460431msec
(+15.6% throughput, -13.3% runtime)
==== 8 jobs, 1GiB files, fsync frequency 16, block size 4KiB ====
Before patch:
WRITE: bw=19.9MiB/s (20.8MB/s), 19.9MiB/s-19.9MiB/s (20.8MB/s-20.8MB/s), io=8192MiB (8590MB), run=412664-412664msec
After patch:
WRITE: bw=22.2MiB/s (23.3MB/s), 22.2MiB/s-22.2MiB/s (23.3MB/s-23.3MB/s), io=8192MiB (8590MB), run=368589-368589msec
(+11.6% throughput, -10.7% runtime)
==== 16 jobs, 512MiB files, fsync frequency 16, block size 4KiB ====
Before patch:
WRITE: bw=29.3MiB/s (30.7MB/s), 29.3MiB/s-29.3MiB/s (30.7MB/s-30.7MB/s), io=8192MiB (8590MB), run=279924-279924msec
After patch:
WRITE: bw=30.4MiB/s (31.9MB/s), 30.4MiB/s-30.4MiB/s (31.9MB/s-31.9MB/s), io=8192MiB (8590MB), run=269258-269258msec
(+3.8% throughput, -3.8% runtime)
==== 32 jobs, 512MiB files, fsync frequency 16, block size 4KiB ====
Before patch:
WRITE: bw=36.9MiB/s (38.7MB/s), 36.9MiB/s-36.9MiB/s (38.7MB/s-38.7MB/s), io=16.0GiB (17.2GB), run=443581-443581msec
After patch:
WRITE: bw=41.6MiB/s (43.6MB/s), 41.6MiB/s-41.6MiB/s (43.6MB/s-43.6MB/s), io=16.0GiB (17.2GB), run=394114-394114msec
(+12.7% throughput, -11.2% runtime)
==== 64 jobs, 512MiB files, fsync frequency 16, block size 4KiB ====
Before patch:
WRITE: bw=45.9MiB/s (48.1MB/s), 45.9MiB/s-45.9MiB/s (48.1MB/s-48.1MB/s), io=32.0GiB (34.4GB), run=714614-714614msec
After patch:
WRITE: bw=48.8MiB/s (51.1MB/s), 48.8MiB/s-48.8MiB/s (51.1MB/s-51.1MB/s), io=32.0GiB (34.4GB), run=672087-672087msec
(+6.3% throughput, -6.0% runtime)
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The possibility of extents being shared (through clone and deduplication
operations) requires special care when logging data checksums, to avoid
having a log tree with different checksum items that cover ranges which
overlap (which resulted in missing checksums after replaying a log tree).
Such problems were fixed in the past by the following commits:
commit 40e046acbd ("Btrfs: fix missing data checksums after replaying a
log tree")
commit e289f03ea7 ("btrfs: fix corrupt log due to concurrent fsync of
inodes with shared extents")
Test case generic/588 exercises the scenario solved by the first commit
(purely sequential and deterministic) while test case generic/457 often
triggered the case fixed by the second commit (not deterministic, requires
specific timings under concurrency).
The problems were addressed by deleting, from the log tree, any existing
checksums before logging the new ones. And also by doing the deletion and
logging of the cheksums while locking the checksum range in an extent io
tree (root->log_csum_range), to deal with the case where we have concurrent
fsyncs against files with shared extents.
That however causes more contention on the leaves of a log tree where we
store checksums (and all the nodes in the paths leading to them), even
when we do not have shared extents, or all the shared extents were created
by past transactions. It also adds a bit of contention on the spin lock of
the log_csums_range extent io tree of the log root.
This change adds a 'last_reflink_trans' field to the inode to keep track
of the last transaction where a new extent was shared between inodes
(through clone and deduplication operations). It is updated for both the
source and destination inodes of reflink operations whenever a new extent
(created in the current transaction) becomes shared by the inodes. This
field is kept in memory only, not persisted in the inode item, similar
to other existing fields (last_unlink_trans, logged_trans).
When logging checksums for an extent, if the value of 'last_reflink_trans'
is smaller then the current transaction's generation/id, we skip locking
the extent range and deletion of checksums from the log tree, since we
know we do not have new shared extents. This reduces contention on the
log tree's leaves where checksums are stored.
The following script, which uses fio, was used to measure the impact of
this change:
$ cat test-fsync.sh
#!/bin/bash
DEV=/dev/sdk
MNT=/mnt/sdk
MOUNT_OPTIONS="-o ssd"
MKFS_OPTIONS="-d single -m single"
if [ $# -ne 3 ]; then
echo "Use $0 NUM_JOBS FILE_SIZE FSYNC_FREQ"
exit 1
fi
NUM_JOBS=$1
FILE_SIZE=$2
FSYNC_FREQ=$3
cat <<EOF > /tmp/fio-job.ini
[writers]
rw=write
fsync=$FSYNC_FREQ
fallocate=none
group_reporting=1
direct=0
bs=64k
ioengine=sync
size=$FILE_SIZE
directory=$MNT
numjobs=$NUM_JOBS
EOF
echo "Using config:"
echo
cat /tmp/fio-job.ini
echo
mkfs.btrfs -f $MKFS_OPTIONS $DEV
mount $MOUNT_OPTIONS $DEV $MNT
fio /tmp/fio-job.ini
umount $MNT
The tests were performed for different numbers of jobs, file sizes and
fsync frequency. A qemu VM using kvm was used, with 8 cores (the host has
12 cores, with cpu governance set to performance mode on all cores), 16GiB
of ram (the host has 64GiB) and using a NVMe device directly (without an
intermediary filesystem in the host). While running the tests, the host
was not used for anything else, to avoid disturbing the tests.
The obtained results were the following (the last line of fio's output was
pasted). Starting with 16 jobs is where a significant difference is
observable in this particular setup and hardware (differences highlighted
below). The very small differences for tests with less than 16 jobs are
possibly just noise and random.
**** 1 job, file size 1G, fsync frequency 1 ****
before this change:
WRITE: bw=23.8MiB/s (24.9MB/s), 23.8MiB/s-23.8MiB/s (24.9MB/s-24.9MB/s), io=1024MiB (1074MB), run=43075-43075msec
after this change:
WRITE: bw=24.4MiB/s (25.6MB/s), 24.4MiB/s-24.4MiB/s (25.6MB/s-25.6MB/s), io=1024MiB (1074MB), run=41938-41938msec
**** 2 jobs, file size 1G, fsync frequency 1 ****
before this change:
WRITE: bw=37.7MiB/s (39.5MB/s), 37.7MiB/s-37.7MiB/s (39.5MB/s-39.5MB/s), io=2048MiB (2147MB), run=54351-54351msec
after this change:
WRITE: bw=37.7MiB/s (39.5MB/s), 37.6MiB/s-37.6MiB/s (39.5MB/s-39.5MB/s), io=2048MiB (2147MB), run=54428-54428msec
**** 4 jobs, file size 1G, fsync frequency 1 ****
before this change:
WRITE: bw=67.5MiB/s (70.8MB/s), 67.5MiB/s-67.5MiB/s (70.8MB/s-70.8MB/s), io=4096MiB (4295MB), run=60669-60669msec
after this change:
WRITE: bw=68.6MiB/s (71.0MB/s), 68.6MiB/s-68.6MiB/s (71.0MB/s-71.0MB/s), io=4096MiB (4295MB), run=59678-59678msec
**** 8 jobs, file size 1G, fsync frequency 1 ****
before this change:
WRITE: bw=128MiB/s (134MB/s), 128MiB/s-128MiB/s (134MB/s-134MB/s), io=8192MiB (8590MB), run=64048-64048msec
after this change:
WRITE: bw=129MiB/s (135MB/s), 129MiB/s-129MiB/s (135MB/s-135MB/s), io=8192MiB (8590MB), run=63405-63405msec
**** 16 jobs, file size 1G, fsync frequency 1 ****
before this change:
WRITE: bw=78.5MiB/s (82.3MB/s), 78.5MiB/s-78.5MiB/s (82.3MB/s-82.3MB/s), io=16.0GiB (17.2GB), run=208676-208676msec
after this change:
WRITE: bw=110MiB/s (115MB/s), 110MiB/s-110MiB/s (115MB/s-115MB/s), io=16.0GiB (17.2GB), run=149295-149295msec
(+40.1% throughput, -28.5% runtime)
**** 32 jobs, file size 1G, fsync frequency 1 ****
before this change:
WRITE: bw=58.8MiB/s (61.7MB/s), 58.8MiB/s-58.8MiB/s (61.7MB/s-61.7MB/s), io=32.0GiB (34.4GB), run=557134-557134msec
after this change:
WRITE: bw=76.1MiB/s (79.8MB/s), 76.1MiB/s-76.1MiB/s (79.8MB/s-79.8MB/s), io=32.0GiB (34.4GB), run=430550-430550msec
(+29.4% throughput, -22.7% runtime)
**** 64 jobs, file size 512M, fsync frequency 1 ****
before this change:
WRITE: bw=65.8MiB/s (68.0MB/s), 65.8MiB/s-65.8MiB/s (68.0MB/s-68.0MB/s), io=32.0GiB (34.4GB), run=498055-498055msec
after this change:
WRITE: bw=85.1MiB/s (89.2MB/s), 85.1MiB/s-85.1MiB/s (89.2MB/s-89.2MB/s), io=32.0GiB (34.4GB), run=385116-385116msec
(+29.3% throughput, -22.7% runtime)
**** 128 jobs, file size 256M, fsync frequency 1 ****
before this change:
WRITE: bw=54.7MiB/s (57.3MB/s), 54.7MiB/s-54.7MiB/s (57.3MB/s-57.3MB/s), io=32.0GiB (34.4GB), run=599373-599373msec
after this change:
WRITE: bw=121MiB/s (126MB/s), 121MiB/s-121MiB/s (126MB/s-126MB/s), io=32.0GiB (34.4GB), run=271907-271907msec
(+121.2% throughput, -54.6% runtime)
**** 256 jobs, file size 256M, fsync frequency 1 ****
before this change:
WRITE: bw=69.2MiB/s (72.5MB/s), 69.2MiB/s-69.2MiB/s (72.5MB/s-72.5MB/s), io=64.0GiB (68.7GB), run=947536-947536msec
after this change:
WRITE: bw=121MiB/s (127MB/s), 121MiB/s-121MiB/s (127MB/s-127MB/s), io=64.0GiB (68.7GB), run=541916-541916msec
(+74.9% throughput, -42.8% runtime)
**** 512 jobs, file size 128M, fsync frequency 1 ****
before this change:
WRITE: bw=85.4MiB/s (89.5MB/s), 85.4MiB/s-85.4MiB/s (89.5MB/s-89.5MB/s), io=64.0GiB (68.7GB), run=767734-767734msec
after this change:
WRITE: bw=141MiB/s (147MB/s), 141MiB/s-141MiB/s (147MB/s-147MB/s), io=64.0GiB (68.7GB), run=466022-466022msec
(+65.1% throughput, -39.3% runtime)
**** 1024 jobs, file size 128M, fsync frequency 1 ****
before this change:
WRITE: bw=115MiB/s (120MB/s), 115MiB/s-115MiB/s (120MB/s-120MB/s), io=128GiB (137GB), run=1143775-1143775msec
after this change:
WRITE: bw=171MiB/s (180MB/s), 171MiB/s-171MiB/s (180MB/s-180MB/s), io=128GiB (137GB), run=764843-764843msec
(+48.7% throughput, -33.1% runtime)
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
This reverts commit 5f008163a5.
The patch is a simplification after direct IO port to iomap
infrastructure, which gets reverted.
Signed-off-by: David Sterba <dsterba@suse.com>
Since we now perform direct reads using i_rwsem, we can remove this
inode flag used to co-ordinate unlocked reads.
The truncate call takes i_rwsem. This means it is correctly synchronized
with concurrent direct reads.
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Johannes Thumshirn <jth@kernel.org>
Signed-off-by: Goldwyn Rodrigues <rgoldwyn@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
In the worst case, there are _4_ layers of bios in the Btrfs direct I/O
path:
1. The bio created by the generic direct I/O code (dio_bio).
2. A clone of dio_bio we create in btrfs_submit_direct() to represent
the entire direct I/O range (orig_bio).
3. A partial clone of orig_bio limited to the size of a RAID stripe that
we create in btrfs_submit_direct_hook().
4. Clones of each of those split bios for each RAID stripe that we
create in btrfs_map_bio().
As of the previous commit, the second layer (orig_bio) is no longer
needed for anything: we can split dio_bio instead, and complete dio_bio
directly when all of the cloned bios complete. This lets us clean up a
bunch of cruft, including dip->subio_endio and dip->errors (we can use
dio_bio->bi_status instead). It also enables the next big cleanup of
direct I/O read repair.
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The next commit will get rid of btrfs_dio_private->orig_bio. The only
thing we really need it for is containing all of the checksums, but we
can easily put the checksum array in btrfs_dio_private and have the
submitted bios reference the array. We can also look the checksums up
while we're setting up instead of the current awkward logic that looks
them up for orig_bio when the first split bio is submitted.
(Interestingly, btrfs_dio_private did contain the
checksums before commit 23ea8e5a07 ("Btrfs: load checksum data once
when submitting a direct read io"), but it didn't look them up up
front.)
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
This is really a reference count now, so convert it to refcount_t and
rename it to refs.
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We haven't used this since commit 9be3395bcd ("Btrfs: use a btrfs
bioset instead of abusing bio internals").
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
In order to keep track of where we have file extents on disk, and thus
where it is safe to adjust the i_size to, we need to have a tree in
place to keep track of the contiguous areas we have file extents for.
Add helpers to use this tree, as it's not required for NO_HOLES file
systems. We will use this by setting DIRTY for areas we know we have
file extent item's set, and clearing it when we remove file extent items
for truncation.
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The inode delalloc mutex was added a long time ago by commit f248679e86
("Btrfs: add a delalloc mutex to inodes for delalloc reservations"), and
the reason for its introduction is not very clear from the change log. It
claims it solves bogus warnings from lockdep, however it lacks an example
report/warning from lockdep, or any explanation.
Since we have enough concurrentcy protection from the locks of the space
info and block reserve objects, and such lockdep warnings don't seem to
exist anymore (at least on a 5.3 kernel I couldn't get them with fstests,
ltp, fs_mark, etc), remove it, simplifying things a bit and decreasing
the size of the btrfs_inode structure. With some quick fio tests doing
direct IO and mmap writes I couldn't observe any significant performance
increase either (direct IO writes that don't increase the file's size
don't hold the inode's lock for their entire duration and mmap writes
don't hold the inode's lock at all), which are the only type of writes
that could see any performance gain due to less serialization.
Review feedback from Josef:
The problem was taking the i_mutex in mmap, which is how I was
protecting delalloc reservations originally. The delalloc mutex didn't
come with all of the other dependencies. That's what the lockdep
messages were about, removing the lock isn't going to make them appear
again.
We _had_ to lock around this because we used to do tricks to keep from
over-reserving, and if we didn't serialize delalloc reservations we'd
end up with ugly accounting problems when we tried to clean things up.
However with my recentish changes this isn't the case anymore. Every
operation is responsible for reserving its space, and then adding it to
the inode. Then cleaning up is straightforward and can't be mucked up
by other users. So we no longer need the delalloc mutex to safe us from
ourselves.
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
btrfs_print_data_csum_error() still assumed checksums to be 32 bit in
size. Make it size agnostic.
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Add a small helper for btrfs_print_data_csum_error() which formats the
checksum according to it's type for pretty printing.
Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: David Sterba <dsterba@suse.com>
[ shorten macro name ]
Signed-off-by: David Sterba <dsterba@suse.com>
Commit 41bd606769 ("Btrfs: fix fsync of files with multiple hard links
in new directories") introduced a path that makes fsync fallback to a full
transaction commit in order to avoid losing hard links and new ancestors
of the fsynced inode. That path is triggered only when the inode has more
than one hard link and either has a new hard link created in the current
transaction or the inode was evicted and reloaded in the current
transaction.
That path ends up getting triggered very often (hundreds of times) during
the course of pgbench benchmarks, resulting in performance drops of about
20%.
This change restores the performance by not triggering the full transaction
commit in those cases, and instead iterate the fs/subvolume tree in search
of all possible new ancestors, for all hard links, to log them.
Reported-by: Zhao Yuhu <zyuhu@suse.com>
Tested-by: James Wang <jnwang@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Deduplicate the btrfs file type conversion implementation - file systems
that use the same file types as defined by POSIX do not need to define
their own versions and can use the common helper functions decared in
fs_types.h and implemented in fs_types.c
Common implementation can be found via commit:
bbe7449e25 "fs: common implementation of file type"
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Phillip Potter <phil@philpotter.co.uk>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The log tree has a long standing problem that when a file is fsync'ed we
only check for new ancestors, created in the current transaction, by
following only the hard link for which the fsync was issued. We follow the
ancestors using the VFS' dget_parent() API. This means that if we create a
new link for a file in a directory that is new (or in an any other new
ancestor directory) and then fsync the file using an old hard link, we end
up not logging the new ancestor, and on log replay that new hard link and
ancestor do not exist. In some cases, involving renames, the file will not
exist at all.
Example:
mkfs.btrfs -f /dev/sdb
mount /dev/sdb /mnt
mkdir /mnt/A
touch /mnt/foo
ln /mnt/foo /mnt/A/bar
xfs_io -c fsync /mnt/foo
<power failure>
In this example after log replay only the hard link named 'foo' exists
and directory A does not exist, which is unexpected. In other major linux
filesystems, such as ext4, xfs and f2fs for example, both hard links exist
and so does directory A after mounting again the filesystem.
Checking if any new ancestors are new and need to be logged was added in
2009 by commit 12fcfd22fe ("Btrfs: tree logging unlink/rename fixes"),
however only for the ancestors of the hard link (dentry) for which the
fsync was issued, instead of checking for all ancestors for all of the
inode's hard links.
So fix this by tracking the id of the last transaction where a hard link
was created for an inode and then on fsync fallback to a full transaction
commit when an inode has more than one hard link and at least one new hard
link was created in the current transaction. This is the simplest solution
since this is not a common use case (adding frequently hard links for
which there's an ancestor created in the current transaction and then
fsync the file). In case it ever becomes a common use case, a solution
that consists of iterating the fs/subvol btree for each hard link and
check if any ancestor is new, could be implemented.
This solves many unexpected scenarios reported by Jayashree Mohan and
Vijay Chidambaram, and for which there is a new test case for fstests
under review.
Fixes: 12fcfd22fe ("Btrfs: tree logging unlink/rename fixes")
CC: stable@vger.kernel.org # 4.4+
Reported-by: Vijay Chidambaram <vvijay03@gmail.com>
Reported-by: Jayashree Mohan <jayashree2912@gmail.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The first auto-assigned value to enum is 0, we can use that and not
initialize all members where the auto-increment does the same. This is
used for values that are not part of on-disk format.
Reviewed-by: Omar Sandoval <osandov@fb.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: David Sterba <dsterba@suse.com>
Snapshot is expected to be fast. But if there are writers steadily
creating dirty pages in our subvolume, the snapshot may take a very long
time to complete. To fix the problem, we use tagged writepage for
snapshot flusher as we do in the generic write_cache_pages(), so we can
omit pages dirtied after the snapshot command.
This does not change the semantics regarding which data get to the
snapshot, if there are pages being dirtied during the snapshotting
operation. There's a sync called before snapshot is taken in old/new
case, any IO in flight just after that may be in the snapshot but this
depends on other system effects that might still sync the IO.
We do a simple snapshot speed test on a Intel D-1531 box:
fio --ioengine=libaio --iodepth=32 --bs=4k --rw=write --size=64G
--direct=0 --thread=1 --numjobs=1 --time_based --runtime=120
--filename=/mnt/sub/testfile --name=job1 --group_reporting & sleep 5;
time btrfs sub snap -r /mnt/sub /mnt/snap; killall fio
original: 1m58sec
patched: 6.54sec
This is the best case for this patch since for a sequential write case,
we omit nearly all pages dirtied after the snapshot command.
For a multi writers, random write test:
fio --ioengine=libaio --iodepth=32 --bs=4k --rw=randwrite --size=64G
--direct=0 --thread=1 --numjobs=4 --time_based --runtime=120
--filename=/mnt/sub/testfile --name=job1 --group_reporting & sleep 5;
time btrfs sub snap -r /mnt/sub /mnt/snap; killall fio
original: 15.83sec
patched: 10.35sec
The improvement is smaller compared to the sequential write case,
since we omit only half of the pages dirtied after snapshot command.
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Ethan Lien <ethanlien@synology.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
This will be used in future patches that remove the optional
extent_io_ops callbacks.
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
There are two members in struct btrfs_root which indicate root's
objectid: objectid and root_key.objectid.
They are both set to the same value in __setup_root():
static void __setup_root(struct btrfs_root *root,
struct btrfs_fs_info *fs_info,
u64 objectid)
{
...
root->objectid = objectid;
...
root->root_key.objectid = objecitd;
...
}
and not changed to other value after initialization.
grep in btrfs directory shows both are used in many places:
$ grep -rI "root->root_key.objectid" | wc -l
133
$ grep -rI "root->objectid" | wc -l
55
(4.17, inc. some noise)
It is confusing to have two similar variable names and it seems
that there is no rule about which should be used in a certain case.
Since ->root_key itself is needed for tree reloc tree, let's remove
'objecitd' member and unify code to use ->root_key.objectid in all places.
Signed-off-by: Misono Tomohiro <misono.tomohiro@jp.fujitsu.com>
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
While the regular inode timestamps all use timespec64 now, the i_otime
field is btrfs specific and still needs to be converted to correctly
represent times beyond 2038.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
We got rid of BTRFS_INODE_HAS_ORPHAN_ITEM and
BTRFS_INODE_ORPHAN_META_RESERVED, so we can renumber the flags to make
them consecutive again.
Signed-off-by: Omar Sandoval <osandov@fb.com>
[ switch them enums so we don't have to do that again ]
Signed-off-by: David Sterba <dsterba@suse.com>
Now that we don't keep long-standing reservations for orphan items,
root->orphan_block_rsv isn't used. We can git rid of it, along with:
- root->orphan_lock, which was used to protect root->orphan_block_rsv
- root->orphan_inodes, which was used as a refcount for root->orphan_block_rsv
- BTRFS_INODE_ORPHAN_META_RESERVED, which was used to track reservations
in root->orphan_block_rsv
- btrfs_orphan_commit_root(), which was the last user of any of these
and does nothing else
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Now that we don't add orphan items for truncate, there can't be races on
adding or deleting an orphan item, so this bit is unnecessary.
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Remove GPL boilerplate text (long, short, one-line) and keep the rest,
ie. personal, company or original source copyright statements. Add the
SPDX header.
Unify the include protection macros to match the file names.
Signed-off-by: David Sterba <dsterba@suse.com>
The current implementation of btrfs_page_exists_in_range() gives the
wrong answer if the workingset code has stored a shadow entry in the
page cache. The filemap_range_has_page() function does not have this
problem, and it's shared code, so use it instead.
eigned-off-by: Matthew Wilcox <mawilcox@microsoft.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
delayed_iput_count wa supposed to be used to implement, well, delayed
iput. The idea is that we keep accumulating the number of iputs we do
until eventually the inode is deleted. Turns out we never really
switched the delayed_iput_count from 0 to 1, hence all conditional
code relying on the value of that member being different than 0 was
never executed. This, as it turns out, didn't cause any problem due
to the simple fact that the generic inode's i_count member was always
used to count the number of iputs. So let's just remove the unused
member and all unused code. This patch essentially provides no
functional changes. While at it, also add proper documentation for
btrfs_add_delayed_iput
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ reformat comment ]
Signed-off-by: David Sterba <dsterba@suse.com>
The way we handle delalloc metadata reservations has gotten
progressively more complicated over the years. There is so much cruft
and weirdness around keeping the reserved count and outstanding counters
consistent and handling the error cases that it's impossible to
understand.
Fix this by making the delalloc block rsv per-inode. This way we can
calculate the actual size of the outstanding metadata reservations every
time we make a change, and then reserve the delta based on that amount.
This greatly simplifies the code everywhere, and makes the error
handling in btrfs_delalloc_reserve_metadata far less terrifying.
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
This is handy for tracing problems with modifying the outstanding
extents counters.
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Right now we do a lot of weird hoops around outstanding_extents in order
to keep the extent count consistent. This is because we logically
transfer the outstanding_extent count from the initial reservation
through the set_delalloc_bits. This makes it pretty difficult to get a
handle on how and when we need to mess with outstanding_extents.
Fix this by revamping the rules of how we deal with outstanding_extents.
Now instead everybody that is holding on to a delalloc extent is
required to increase the outstanding extents count for itself. This
means we'll have something like this
btrfs_delalloc_reserve_metadata - outstanding_extents = 1
btrfs_set_extent_delalloc - outstanding_extents = 2
btrfs_release_delalloc_extents - outstanding_extents = 1
for an initial file write. Now take the append write where we extend an
existing delalloc range but still under the maximum extent size
btrfs_delalloc_reserve_metadata - outstanding_extents = 2
btrfs_set_extent_delalloc
btrfs_set_bit_hook - outstanding_extents = 3
btrfs_merge_extent_hook - outstanding_extents = 2
btrfs_delalloc_release_extents - outstanding_extnets = 1
In order to make the ordered extent transition we of course must now
make ordered extents carry their own outstanding_extent reservation, so
for cow_file_range we end up with
btrfs_add_ordered_extent - outstanding_extents = 2
clear_extent_bit - outstanding_extents = 1
btrfs_remove_ordered_extent - outstanding_extents = 0
This makes all manipulations of outstanding_extents much more explicit.
Every successful call to btrfs_delalloc_reserve_metadata _must_ now be
combined with btrfs_release_delalloc_extents, even in the error case, as
that is the only function that actually modifies the
outstanding_extents counter.
The drawback to this is now we are much more likely to have transient
cases where outstanding_extents is much larger than it actually should
be. This could happen before as we manipulated the delalloc bits, but
now it happens basically at every write. This may put more pressure on
the ENOSPC flushing code, but I think making this code simpler is worth
the cost. I have another change coming to mitigate this side-effect
somewhat.
I also added trace points for the counter manipulation. These were used
by a bpf script I wrote to help track down leak issues.
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Add new value for compression to distinguish between defrag and
property. Previously, a single variable was used and this caused clashes
when the per-file 'compression' was set and a defrag -c was called.
The property-compression is loaded when the file is open, defrag will
overwrite the same variable and reset to 0 (ie. NONE) at when the file
defragmentaion is finished. That's considered a usability bug.
Now we won't touch the property value, use the defrag-compression. The
precedence of defrag is higher than for property (and whole-filesystem).
Signed-off-by: David Sterba <dsterba@suse.com>
This is preparatory for separating inode compression requested by defrag
and set via properties. This will fix a usability bug when defrag will
reset compression type to NONE. If the file has compression set via
property, it will not apply anymore (until next mount or reset through
command line).
We're going to fix that by adding another variable just for the defrag
call and won't touch the property. The defrag will have higher priority
when deciding whether to compress the data.
Signed-off-by: David Sterba <dsterba@suse.com>
Tracepoint arguments are all read-only. If we mark the arguments
as const, we're able to keep or convert those arguments to const
where appropriate.
Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Replace bi_error with a new bi_status to allow for a clear conversion.
Note that device mapper overloaded bi_error with a private value, which
we'll have to keep arround at least for now and thus propagate to a
proper blk_status_t value.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
Currently when there are buffered writes that were not yet flushed and
they fall within allocated ranges of the file (that is, not in holes or
beyond eof assuming there are no prealloc extents beyond eof), btrfs
simply reports an incorrect number of used blocks through the stat(2)
system call (or any of its variants), regardless of mount options or
inode flags (compress, compress-force, nodatacow). This is because the
number of blocks used that is reported is based on the current number
of bytes in the vfs inode plus the number of dealloc bytes in the btrfs
inode. The later covers bytes that both fall within allocated regions
of the file and holes.
Example scenarios where the number of reported blocks is wrong while the
buffered writes are not flushed:
$ mkfs.btrfs -f /dev/sdc
$ mount /dev/sdc /mnt/sdc
$ xfs_io -f -c "pwrite -S 0xaa 0 64K" /mnt/sdc/foo1
wrote 65536/65536 bytes at offset 0
64 KiB, 16 ops; 0.0000 sec (259.336 MiB/sec and 66390.0415 ops/sec)
$ sync
$ xfs_io -c "pwrite -S 0xbb 0 64K" /mnt/sdc/foo1
wrote 65536/65536 bytes at offset 0
64 KiB, 16 ops; 0.0000 sec (192.308 MiB/sec and 49230.7692 ops/sec)
# The following should have reported 64K...
$ du -h /mnt/sdc/foo1
128K /mnt/sdc/foo1
$ sync
# After flushing the buffered write, it now reports the correct value.
$ du -h /mnt/sdc/foo1
64K /mnt/sdc/foo1
$ xfs_io -f -c "falloc -k 0 128K" -c "pwrite -S 0xaa 0 64K" /mnt/sdc/foo2
wrote 65536/65536 bytes at offset 0
64 KiB, 16 ops; 0.0000 sec (520.833 MiB/sec and 133333.3333 ops/sec)
$ sync
$ xfs_io -c "pwrite -S 0xbb 64K 64K" /mnt/sdc/foo2
wrote 65536/65536 bytes at offset 65536
64 KiB, 16 ops; 0.0000 sec (260.417 MiB/sec and 66666.6667 ops/sec)
# The following should have reported 128K...
$ du -h /mnt/sdc/foo2
192K /mnt/sdc/foo2
$ sync
# After flushing the buffered write, it now reports the correct value.
$ du -h /mnt/sdc/foo2
128K /mnt/sdc/foo2
So the number of used file blocks is simply incorrect, unlike in other
filesystems such as ext4 and xfs for example, but only while the buffered
writes are not flushed.
Fix this by tracking the number of delalloc bytes that fall within holes
and beyond eof of a file, and use instead this new counter when reporting
the number of used blocks for an inode.
Another different problem that exists is that the delalloc bytes counter
is reset when writeback starts (by clearing the EXTENT_DEALLOC flag from
the respective range in the inode's iotree) and the vfs inode's bytes
counter is only incremented when writeback finishes (through
insert_reserved_file_extent()). Therefore while writeback is ongoing we
simply report a wrong number of blocks used by an inode if the write
operation covers a range previously unallocated. While this change does
not fix this problem, it does minimizes it a lot by shortening that time
window, as the new dealloc bytes counter (new_delalloc_bytes) is only
decremented when writeback finishes right before updating the vfs inode's
bytes counter. Fully fixing this second problem is not trivial and will
be addressed later by a different patch.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
The original csum error message only outputs inode number, offset, check
sum and expected check sum.
However no root objectid is outputted, which sometimes makes debugging
quite painful under multi-subvolume case (including relocation).
Also the checksum output is decimal, which seldom makes sense for
users/developers and is hard to read in most time.
This patch will add root objectid, which will be %lld for rootid larger
than LAST_FREE_OBJECTID, and hex csum output for better readability.
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Currently btrfs_ino takes a struct inode and this causes a lot of
internal btrfs functions which consume this ino to take a VFS inode,
rather than btrfs' own struct btrfs_inode. In order to fix this "leak"
of VFS structs into the internals of btrfs first it's necessary to
eliminate all uses of struct inode for the purpose of inode. This patch
does that by using BTRFS_I to convert an inode to btrfs_inode. With
this problem eliminated subsequent patches will start eliminating the
passing of struct inode altogether, eventually resulting in a lot cleaner
code.
Signed-off-by: Nikolay Borisov <n.borisov.lkml@gmail.com>
[ fix btrfs_get_extent tracepoint prototype ]
Signed-off-by: David Sterba <dsterba@suse.com>
We have a lot of random ints in btrfs_fs_info that can be put into flags. This
is mostly equivalent with the exception of how we deal with quota going on or
off, now instead we set a flag when we are turning it on or off and deal with
that appropriately, rather than just having a pending state that the current
quota_enabled gets set to. Thanks,
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Due to the optimization of lockless direct IO writes (the inode's i_mutex
is not held) introduced in commit 38851cc19a ("Btrfs: implement unlocked
dio write"), we started having races between such writes with concurrent
fsync operations that use the fast fsync path. These races were addressed
in the patches titled "Btrfs: fix race between fsync and lockless direct
IO writes" and "Btrfs: fix race between fsync and direct IO writes for
prealloc extents". The races happened because the direct IO path, like
every other write path, does create extent maps followed by the
corresponding ordered extents while the fast fsync path collected first
ordered extents and then it collected extent maps. This made it possible
to log file extent items (based on the collected extent maps) without
waiting for the corresponding ordered extents to complete (get their IO
done). The two fixes mentioned before added a solution that consists of
making the direct IO path create first the ordered extents and then the
extent maps, while the fsync path attempts to collect any new ordered
extents once it collects the extent maps. This was simple and did not
require adding any synchonization primitive to any data structure (struct
btrfs_inode for example) but it makes things more fragile for future
development endeavours and adds an exceptional approach compared to the
other write paths.
This change adds a read-write semaphore to the btrfs inode structure and
makes the direct IO path create the extent maps and the ordered extents
while holding read access on that semaphore, while the fast fsync path
collects extent maps and ordered extents while holding write access on
that semaphore. The logic for direct IO write path is encapsulated in a
new helper function that is used both for cow and nocow direct IO writes.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Josef Bacik <jbacik@fb.com>
Nikolay Borisov
Goldwyn Rodrigues
Filipe Manana
David Sterba
Omar Sandoval
Josef Bacik
Johannes Thumshirn
Phillip Potter
Ethan Lien
Misono Tomohiro
Arnd Bergmann
Matthew Wilcox
Jeff Mahoney
Christoph Hellwig
Qu Wenruo
Nikolay Borisov
Josef Bacik
Nicholas D Steeves