btrfs: change for_rename argument of btrfs_record_unlink_dir() to bool

The for_rename argument of btrfs_record_unlink_dir() is defined as an
integer, but the argument is in fact used as a boolean. S

btrfs: change for_rename argument of btrfs_record_unlink_dir() to bool

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>

show more ...

btrfs: use a negative value for BTRFS_LOG_FORCE_COMMIT

Currently we use the value 1 for BTRFS_LOG_FORCE_COMMIT, but that value
has a few inconveniences:

1) If it's ever used by btrfs_log_inode(), o

btrfs: use a negative value for BTRFS_LOG_FORCE_COMMIT

Currently we use the value 1 for BTRFS_LOG_FORCE_COMMIT, but that value
has a few inconveniences:

1) If it's ever used by btrfs_log_inode(), or any function down the call
chain, we have to remember to btrfs_set_log_full_commit(), which is
repetitive and has a chance to be forgotten in future use cases.
btrfs_log_inode_parent() only calls btrfs_set_log_full_commit() when
it gets a negative value from btrfs_log_inode();

2) Down the call chain of btrfs_log_inode(), we may have functions that
need to force a log commit, but can return either an error (negative
value), false (0) or true (1). So they are forced to return some
random negative to force a log commit - using BTRFS_LOG_FORCE_COMMIT
would make the intention more clear. Currently the only example is
flush_dir_items_batch().

So turn BTRFS_LOG_FORCE_COMMIT into a negative value. The chosen value
is -(MAX_ERRNO + 1), so that it does not overlap any errno value and makes
it easier to debug.

Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

show more ...

btrfs: simplify update of last_dir_index_offset when logging a directory

When logging a directory, we always set the inode's last_dir_index_offset
to the offset of the last dir index item we found.

btrfs: simplify update of last_dir_index_offset when logging a directory

When logging a directory, we always set the inode's last_dir_index_offset
to the offset of the last dir index item we found. This is using an extra
field in the log context structure, and it makes more sense to update it
only after we insert dir index items, and we could directly update the
inode's last_dir_index_offset field instead.

So make this simpler by updating the inode's last_dir_index_offset only
when we actually insert dir index keys in the log tree, and getting rid
of the last_dir_item_offset field in the log context structure.

Reported-by: David Arendt <admin@prnet.org>
Link: https://lore.kernel.org/linux-btrfs/ae169fc6-f504-28f0-a098-6fa6a4dfb612@leemhuis.info/
Reported-by: Maxim Mikityanskiy <maxtram95@gmail.com>
Link: https://lore.kernel.org/linux-btrfs/Y8voyTXdnPDz8xwY@mail.gmail.com/
Reported-by: Hunter Wardlaw <wardlawhunter@gmail.com>
Link: https://bugzilla.suse.com/show_bug.cgi?id=1207231
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=216851
CC: stable@vger.kernel.org # 6.1+
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

show more ...

btrfs: use struct fscrypt_str instead of struct qstr

While struct qstr is more natural without fscrypt, since it's provided
by dentries, struct fscrypt_str is provided by the fscrypt handlers
proces

btrfs: use struct fscrypt_str instead of struct qstr

While struct qstr is more natural without fscrypt, since it's provided
by dentries, struct fscrypt_str is provided by the fscrypt handlers
processing dentries, and is thus more natural in the fscrypt world.
Replace all of the struct qstr uses with struct fscrypt_str.

Signed-off-by: Sweet Tea Dorminy <sweettea-kernel@dorminy.me>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

show more ...

btrfs: use struct qstr instead of name and namelen pairs

Many functions throughout btrfs take name buffer and name length
arguments. Most of these functions at the highest level are usually
called w

btrfs: use struct qstr instead of name and namelen pairs

Many functions throughout btrfs take name buffer and name length
arguments. Most of these functions at the highest level are usually
called with these arguments extracted from a supplied dentry's name.
But the entire name can be passed instead, making each function a little
more elegant.

Each function whose arguments are currently the name and length
extracted from a dentry is herein converted to instead take a pointer to
the name in the dentry. The couple of calls to these calls without a
struct dentry are converted to create an appropriate qstr to pass in.
Additionally, every function which is only called with a name/len
extracted directly from a qstr is also converted.

This change has positive effect on stack consumption, frame of many
functions is reduced but this will be used in the future for fscrypt
related structures.

Signed-off-by: Sweet Tea Dorminy <sweettea-kernel@dorminy.me>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

show more ...

btrfs: move the printk helpers out of ctree.h

We have a bunch of printk helpers that are in ctree.h. These have
nothing to do with ctree.c, so move them into their own header.
Subsequent patches wi

btrfs: move the printk helpers out of ctree.h

We have a bunch of printk helpers that are in ctree.h. These have
nothing to do with ctree.c, so move them into their own header.
Subsequent patches will cleanup the printk helpers.

Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

show more ...

btrfs: use delayed items when logging a directory

When logging a directory we start by flushing all its delayed items.
That results in adding dir index items to the subvolume btree, for new
dentries

btrfs: use delayed items when logging a directory

When logging a directory we start by flushing all its delayed items.
That results in adding dir index items to the subvolume btree, for new
dentries, and removing dir index items from the subvolume btree for any
dentries that were deleted.

This makes it straightforward to log a directory simply by iterating over
all the modified subvolume btree leaves, especially when we used to log
both dir index keys and dir item keys (before commit 339d035424849c
("btrfs: only copy dir index keys when logging a directory") and when we
used to copy old dir index entries for leaves modified in the current
transaction (before commit 732d591a5d6c12 ("btrfs: stop copying old dir
items when logging a directory")).

From an efficiency point of view this has a couple of drawbacks:

1) Adds extra latency, due to copying delayed items to the subvolume btree
and deleting dir index items from the btree.

Further if there are other tasks accessing the btree, which is common
(syscalls like creat, mkdir, rename, link, unlink, truncate, reflinks,
etc, finishing an ordered extent, etc), lock contention can cause
further delays, both to the task logging a directory and to the other
tasks accessing the btree;

2) More time spent overall flushing delayed items, if after logging the
directory further changes are done to the directory in the same
transaction.

For example, if we add 10 dentries to a directory, fsync it, add more
10 dentries, fsync it again, then add more 10 dentries and fsync it
again, then we end up inserting 3 batches of 10 items to the subvolume
btree. With the changes from this patch, we flush all the delayed items
to the btree only once - a single batch of 30 items, and outside the
logging code (transaction commit or when delayed items are flushed
asynchronously).

This change simply skips the flushing of delayed items every time we log a
directory. Instead we copy the delayed insertion items directly to the log
tree and delete delayed deletion items directly from the log tree.
Therefore avoiding changing first the subvolume btree and then scanning it
for new items to copy from it to the log tree and detecting deletions
by observing gaps in consecutive dir index keys in subvolume btree leaves.

Running the following tests on a non-debug kernel (Debian's default kernel
config), on a box with a NVMe device, a 12 cores Intel CPU and 64G of ram,
produced the results below.

The results compare a branch without this patch and all the other patches
it depends on versus the same branch with the patchset applied.

The patchset is comprised of the following patches:

btrfs: don't drop dir index range items when logging a directory
btrfs: remove the root argument from log_new_dir_dentries()
btrfs: update stale comment for log_new_dir_dentries()
btrfs: free list element sooner at log_new_dir_dentries()
btrfs: avoid memory allocation at log_new_dir_dentries() for common case
btrfs: remove root argument from btrfs_delayed_item_reserve_metadata()
btrfs: store index number instead of key in struct btrfs_delayed_item
btrfs: remove unused logic when looking up delayed items
btrfs: shrink the size of struct btrfs_delayed_item
btrfs: search for last logged dir index if it's not cached in the inode
btrfs: move need_log_inode() to above log_conflicting_inodes()
btrfs: move log_new_dir_dentries() above btrfs_log_inode()
btrfs: log conflicting inodes without holding log mutex of the initial inode
btrfs: skip logging parent dir when conflicting inode is not a dir
btrfs: use delayed items when logging a directory

Custom test script for testing time spent at btrfs_log_inode():

#!/bin/bash

DEV=/dev/nvme0n1
MNT=/mnt/nvme0n1

# Total number of files to create in the test directory.
NUM_FILES=10000
# Fsync after creating or renaming N files.
FSYNC_AFTER=100

umount $DEV &> /dev/null
mkfs.btrfs -f $DEV
mount -o ssd $DEV $MNT

TEST_DIR=$MNT/testdir
mkdir $TEST_DIR

echo "Creating files..."
for ((i = 1; i <= $NUM_FILES; i++)); do
echo -n > $TEST_DIR/file_$i
if (( ($i % $FSYNC_AFTER) == 0 )); then
xfs_io -c "fsync" $TEST_DIR
fi
done

sync

echo "Renaming files..."
for ((i = 1; i <= $NUM_FILES; i++)); do
mv $TEST_DIR/file_$i $TEST_DIR/file_$i.renamed
if (( ($i % $FSYNC_AFTER) == 0 )); then
xfs_io -c "fsync" $TEST_DIR
fi
done

umount $MNT

And using the following bpftrace script to capture the total time that is
spent at btrfs_log_inode():

#!/usr/bin/bpftrace

k:btrfs_log_inode
{
@start_log_inode[tid] = nsecs;
}

kr:btrfs_log_inode
/@start_log_inode[tid]/
{
$dur = (nsecs - @start_log_inode[tid]) / 1000;
@btrfs_log_inode_total_time = sum($dur);
delete(@start_log_inode[tid]);
}

END
{
clear(@start_log_inode);
}

Result before applying patchset:

@btrfs_log_inode_total_time: 622642

Result after applying patchset:

@btrfs_log_inode_total_time: 354134 (-43.1% time spent)

The following dbench script was also used for testing:

#!/bin/bash

NUM_JOBS=$(nproc --all)

DEV=/dev/nvme0n1
MNT=/mnt/nvme0n1
MOUNT_OPTIONS="-o ssd"
MKFS_OPTIONS="-O no-holes -R free-space-tree"

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

dbench -D $MNT --skip-cleanup -t 120 -S $NUM_JOBS

umount $MNT

Before patchset:

Operation Count AvgLat MaxLat
----------------------------------------
NTCreateX 3322265 0.034 21.032
Close 2440562 0.002 0.994
Rename 140664 1.150 269.633
Unlink 670796 1.093 269.678
Deltree 96 5.481 15.510
Mkdir 48 0.004 0.052
Qpathinfo 3010924 0.014 8.127
Qfileinfo 528055 0.001 0.518
Qfsinfo 552113 0.003 0.372
Sfileinfo 270575 0.005 0.688
Find 1164176 0.052 13.931
WriteX 1658537 0.019 5.918
ReadX 5207412 0.003 1.034
LockX 10818 0.003 0.079
UnlockX 10818 0.002 0.313
Flush 232811 1.027 269.735

Throughput 869.867 MB/sec (sync dirs) 12 clients 12 procs max_latency=269.741 ms

After patchset:

Operation Count AvgLat MaxLat
----------------------------------------
NTCreateX 4152738 0.029 20.863
Close 3050770 0.002 1.119
Rename 175829 0.871 211.741
Unlink 838447 0.845 211.724
Deltree 120 4.798 14.162
Mkdir 60 0.003 0.005
Qpathinfo 3763807 0.011 4.673
Qfileinfo 660111 0.001 0.400
Qfsinfo 690141 0.003 0.429
Sfileinfo 338260 0.005 0.725
Find 1455273 0.046 6.787
WriteX 2073307 0.017 5.690
ReadX 6509193 0.003 1.171
LockX 13522 0.003 0.077
UnlockX 13522 0.002 0.125
Flush 291044 0.811 211.631

Throughput 1089.27 MB/sec (sync dirs) 12 clients 12 procs max_latency=211.750 ms

(+25.2% throughput, -21.5% max latency)

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

show more ...

btrfs: log conflicting inodes without holding log mutex of the initial inode

When logging an inode, if we detect the inode has a reference that
conflicts with some other inode that got renamed, we l

btrfs: log conflicting inodes without holding log mutex of the initial inode

When logging an inode, if we detect the inode has a reference that
conflicts with some other inode that got renamed, we log that other inode
while holding the log mutex of the current inode. We then find out if
there are other inodes that conflict with the first conflicting inode,
and log them while under the log mutex of the original inode. This is
fine because the recursion can only happen once.

For the upcoming work where we directly log delayed items without flushing
them first to the subvolume tree, this recursion adds a lot of complexity
and it's hard to keep lockdep happy about it.

So collect a list of conflicting inodes and then log the inodes after
unlocking the log mutex of the inode we started with.

Also limit the maximum number of conflict inodes we log to 10, to avoid
spending too much time logging (and maybe allocating too many list
elements too), as typically we don't have more than 1 or 2 conflicting
inodes - if we go over the limit, simply fallback to a transaction commit.

It is possible to have a very long list of conflicting inodes to be
intentionally created by a user if he/she creates a very long succession
of renames like this:

(...)
rename E to F
rename D to E
rename C to D
rename B to C
rename A to B
touch A (create a new file named A)
fsync A

If that happened for a sequence of hundreds or thousands of renames, it
could massively slow down the logging and cause other secondary effects
like for example blocking other fsync operations and transaction commits
for a very long time (assuming it wouldn't run into -ENOSPC or -ENOMEM
first). However such cases are very uncommon to happen in practice,
nevertheless it's better to be prepared for them and avoid chaos.
Such long sequence of conflicting inodes could be created before this
change.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

show more ...

btrfs: tree-log: make the return value for log syncing consistent

Currently we will return 1 or -EAGAIN if we decide we need to commit
the transaction rather than sync the log. In practice this doe

btrfs: tree-log: make the return value for log syncing consistent

Currently we will return 1 or -EAGAIN if we decide we need to commit
the transaction rather than sync the log. In practice this doesn't
really matter, we interpret any !0 and !BTRFS_NO_LOG_SYNC as needing to
commit the transaction. However this makes it hard to figure out what
the correct thing to do is.

Fix this up by defining BTRFS_LOG_FORCE_COMMIT and using this in all the
places where we want to force the transaction to be committed.

CC: stable@vger.kernel.org # 5.15+
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>

show more ...

btrfs: avoid inode logging during rename and link when possible

During a rename or link operation, we need to determine if an inode was
previously logged or not, and if it was, do some update to the

btrfs: avoid inode logging during rename and link when possible

During a rename or link operation, we need to determine if an inode was
previously logged or not, and if it was, do some update to the logged
inode. We used to rely exclusively on the logged_trans field of struct
btrfs_inode to determine that, but that was not reliable because the
value of that field is not persisted in the inode item, so it's lost
when an inode is evicted and loaded back again. That led to several
issues in the past, such as not persisting deletions (such as the case
fixed by commit 803f0f64d17769 ("Btrfs: fix fsync not persisting dentry
deletions due to inode evictions")), or resulting in losing a file
after an inode eviction followed by a rename (commit ecc64fab7d49c6
("btrfs: fix lost inode on log replay after mix of fsync, rename and
inode eviction")), besides other issues.

So the inode_logged() helper was introduced and used to determine if an
inode was possibly logged before in the current transaction, with the
caveat that it could return false positives, in the sense that even if an
inode was not logged before in the current transaction, it could still
return true, but never to return false in case the inode was logged.
>From a functional point of view that is fine, but from a performance
perspective it can introduce significant latencies to rename and link
operations, as they will end up doing inode logging even when it is not
necessary.

Recently on a 5.15 kernel, an openSUSE Tumbleweed user reported package
installations and upgrades, with the zypper tool, were often taking a
long time to complete. With strace it could be observed that zypper was
spending about 99% of its time on rename operations, and then with
further analysis we checked that directory logging was happening too
frequently. Taking into account that installation/upgrade of some of the
packages needed a few thousand file renames, the slowdown was very
noticeable for the user.

The issue was caused indirectly due to an excessive number of inode
evictions on a 5.15 kernel, about 100x more compared to a 5.13, 5.14 or
a 5.16-rc8 kernel. While triggering the inode evictions if something
outside btrfs' control, btrfs could still behave better by eliminating
the false positives from the inode_logged() helper.

So change inode_logged() to actually eliminate such false positives caused
by inode eviction and when an inode was never logged since the filesystem
was mounted, as both cases relate to when the logged_trans field of struct
btrfs_inode has a value of zero. When it can not determine if the inode
was logged based only on the logged_trans value, lookup for the existence
of the inode item in the log tree - if it's there then we known the inode
was logged, if it's not there then it can not have been logged in the
current transaction. Once we determine if the inode was logged, update
the logged_trans value to avoid future calls to have to search in the log
tree again.

Alternatively, we could start storing logged_trans in the on disk inode
item structure (struct btrfs_inode_item) in the unused space it still has,
but that would be a bit odd because:

1) We only care about logged_trans since the filesystem was mounted, we
don't care about its value from a previous mount. Having it persisted
in the inode item structure would not make the best use of the precious
unused space;

2) In order to get logged_trans persisted before inode eviction, we would
have to update the delayed inode when we finish logging the inode and
update its logged_trans in struct btrfs_inode, which makes it a bit
cumbersome since we need to check if the delayed inode exists, if not
create it and populate it and deal with any errors (-ENOMEM mostly).

This change is part of a patchset comprised of the following patches:

1/5 btrfs: add helper to delete a dir entry from a log tree
2/5 btrfs: pass the dentry to btrfs_log_new_name() instead of the inode
3/5 btrfs: avoid logging all directory changes during renames
4/5 btrfs: stop doing unnecessary log updates during a rename
5/5 btrfs: avoid inode logging during rename and link when possible

The following test script mimics part of what the zypper tool does during
package installations/upgrades. It does not triggers inode evictions, but
it's similar because it triggers false positives from the inode_logged()
helper, because the inodes have a logged_trans of 0, there's a log tree
due to a fsync of an unrelated file and the directory inode has its
last_trans field set to the current transaction:

$ cat test.sh

#!/bin/bash

DEV=/dev/nvme0n1
MNT=/mnt/nvme0n1

NUM_FILES=10000

mkfs.btrfs -f $DEV
mount $DEV $MNT

mkdir $MNT/testdir

for ((i = 1; i <= $NUM_FILES; i++)); do
echo -n > $MNT/testdir/file_$i
done

sync

# Now do some change to an unrelated file and fsync it.
# This is just to create a log tree to make sure that inode_logged()
# does not return false when called against "testdir".
xfs_io -f -c "pwrite 0 4K" -c "fsync" $MNT/foo

# Do some change to testdir. This is to make sure inode_logged()
# will return true when called against "testdir", because its
# logged_trans is 0, it was changed in the current transaction
# and there's a log tree.
echo -n > $MNT/testdir/file_$((NUM_FILES + 1))

echo "Renaming $NUM_FILES files..."
start=$(date +%s%N)
for ((i = 1; i <= $NUM_FILES; i++)); do
mv $MNT/testdir/file_$i $MNT/testdir/file_$i-RPMDELETE
done
end=$(date +%s%N)

dur=$(( (end - start) / 1000000 ))
echo "Renames took $dur milliseconds"

umount $MNT

Testing this change on a box using a non-debug kernel (Debian's default
kernel config) gave the following results:

NUM_FILES=10000, before patchset: 27837 ms
NUM_FILES=10000, after patches 1/5 to 4/5 applied: 9236 ms (-66.8%)
NUM_FILES=10000, after whole patchset applied: 8902 ms (-68.0%)

NUM_FILES=5000, before patchset: 9127 ms
NUM_FILES=5000, after patches 1/5 to 4/5 applied: 4640 ms (-49.2%)
NUM_FILES=5000, after whole patchset applied: 4441 ms (-51.3%)

NUM_FILES=2000, before patchset: 2528 ms
NUM_FILES=2000, after patches 1/5 to 4/5 applied: 1983 ms (-21.6%)
NUM_FILES=2000, after whole patchset applied: 1747 ms (-30.9%)

NUM_FILES=1000, before patchset: 1085 ms
NUM_FILES=1000, after patches 1/5 to 4/5 applied: 893 ms (-17.7%)
NUM_FILES=1000, after whole patchset applied: 867 ms (-20.1%)

Running dbench on the same physical machine with the following script:

$ cat run-dbench.sh
#!/bin/bash

NUM_JOBS=$(nproc --all)

DEV=/dev/nvme0n1
MNT=/mnt/nvme0n1
MOUNT_OPTIONS="-o ssd"
MKFS_OPTIONS="-O no-holes -R free-space-tree"

echo "performance" | \
tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor

mkfs.btrfs -f $MKFS_OPTIONS $DEV
mount $MOUNT_OPTIONS $DEV $MNT

dbench -D $MNT -t 120 $NUM_JOBS

umount $MNT

Before patchset:

Operation Count AvgLat MaxLat
----------------------------------------
NTCreateX 3761352 0.032 143.843
Close 2762770 0.002 2.273
Rename 159304 0.291 67.037
Unlink 759784 0.207 143.998
Deltree 72 4.028 15.977
Mkdir 36 0.003 0.006
Qpathinfo 3409780 0.013 9.678
Qfileinfo 596772 0.001 0.878
Qfsinfo 625189 0.003 1.245
Sfileinfo 306443 0.006 1.840
Find 1318106 0.063 19.798
WriteX 1871137 0.021 8.532
ReadX 5897325 0.003 3.567
LockX 12252 0.003 0.258
UnlockX 12252 0.002 0.100
Flush 263666 3.327 155.632

Throughput 980.047 MB/sec 12 clients 12 procs max_latency=155.636 ms

After whole patchset applied:

Operation Count AvgLat MaxLat
----------------------------------------
NTCreateX 4195584 0.033 107.742
Close 3081932 0.002 1.935
Rename 177641 0.218 14.905
Unlink 847333 0.166 107.822
Deltree 118 5.315 15.247
Mkdir 59 0.004 0.048
Qpathinfo 3802612 0.014 10.302
Qfileinfo 666748 0.001 1.034
Qfsinfo 697329 0.003 0.944
Sfileinfo 341712 0.006 2.099
Find 1470365 0.065 9.359
WriteX 2093921 0.021 8.087
ReadX 6576234 0.003 3.407
LockX 13660 0.003 0.308
UnlockX 13660 0.002 0.114
Flush 294090 2.906 115.539

Throughput 1093.11 MB/sec 12 clients 12 procs max_latency=115.544 ms

+11.5% throughput -25.8% max latency rename max latency -77.8%

Link: https://bugzilla.opensuse.org/show_bug.cgi?id=1193549
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

show more ...

btrfs: avoid logging all directory changes during renames

When doing a rename of a file, if the file or its old parent directory
were logged before, we log the new name of the file and then make sur

btrfs: avoid logging all directory changes during renames

When doing a rename of a file, if the file or its old parent directory
were logged before, we log the new name of the file and then make sure
we log the old parent directory, to ensure that after a log replay the
old name of the file is deleted and the new name added.

The logging of the old parent directory can take some time, because it
will scan all leaves modified in the current transaction, check which
directory entries were already logged, copy the ones that were not
logged before, etc. In this rename context all we need to do is make
sure that the old name of the file is deleted on log replay, so instead
of triggering a directory log operation, we can just delete the old
directory entry from the log if it's there, or in case it isn't there,
just log a range item to signal log replay that the old name must be
deleted. So change btrfs_log_new_name() to do that.

This scenario is actually not uncommon to trigger, and recently on a
5.15 kernel, an openSUSE Tumbleweed user reported package installations
and upgrades, with the zypper tool, were often taking a long time to
complete, much more than usual. With strace it could be observed that
zypper was spending over 99% of its time on rename operations, and then
with further analysis we checked that directory logging was happening
too frequently and causing high latencies for the rename operations.
Taking into account that installation/upgrade of some of these packages
needed about a few thousand file renames, the slowdown was very noticeable
for the user.

The issue was caused indirectly due to an excessive number of inode
evictions on a 5.15 kernel, about 100x more compared to a 5.13, 5.14
or a 5.16-rc8 kernel. After an inode eviction we can't tell for sure,
in an efficient way, if an inode was previously logged in the current
transaction, so we are pessimistic and assume it was, because in case
it was we need to update the logged inode. More details on that in one
of the patches in the same series (subject "btrfs: avoid inode logging
during rename and link when possible"). Either way, in case the parent
directory was logged before, we currently do more work then necessary
during a rename, and this change minimizes that amount of work.

The following script mimics part of what a package installation/upgrade
with zypper does, which is basically renaming a lot of files, in some
directory under /usr, to a name with a suffix of "-RPMDELETE":

$ cat test.sh
#!/bin/bash

DEV=/dev/nvme0n1
MNT=/mnt/nvme0n1

NUM_FILES=10000

mkfs.btrfs -f $DEV
mount $DEV $MNT

mkdir $MNT/testdir

for ((i = 1; i <= $NUM_FILES; i++)); do
echo -n > $MNT/testdir/file_$i
done

sync

# Do some change to testdir and fsync it.
echo -n > $MNT/testdir/file_$((NUM_FILES + 1))
xfs_io -c "fsync" $MNT/testdir

echo "Renaming $NUM_FILES files..."
start=$(date +%s%N)
for ((i = 1; i <= $NUM_FILES; i++)); do
mv $MNT/testdir/file_$i $MNT/testdir/file_$i-RPMDELETE
done
end=$(date +%s%N)

dur=$(( (end - start) / 1000000 ))
echo "Renames took $dur milliseconds"

umount $MNT

Testing this change on box using a non-debug kernel (Debian's default
kernel config) gave the following results:

NUM_FILES=10000, before this patch: 27399 ms
NUM_FILES=10000, after this patch: 9093 ms (-66.8%)

NUM_FILES=5000, before this patch: 9241 ms
NUM_FILES=5000, after this patch: 4642 ms (-49.8%)

NUM_FILES=2000, before this patch: 2550 ms
NUM_FILES=2000, after this patch: 1788 ms (-29.9%)

NUM_FILES=1000, before this patch: 1088 ms
NUM_FILES=1000, after this patch: 905 ms (-16.9%)

Link: https://bugzilla.opensuse.org/show_bug.cgi?id=1193549
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

show more ...

btrfs: pass the dentry to btrfs_log_new_name() instead of the inode

In the next patch in the series, there will be the need to access the old
name, and its length, of an inode when logging the inode

btrfs: pass the dentry to btrfs_log_new_name() instead of the inode

In the next patch in the series, there will be the need to access the old
name, and its length, of an inode when logging the inode during a rename.
So instead of passing the inode to btrfs_log_new_name() pass the dentry,
because from the dentry we can get the inode, the name and its length.

This will avoid passing 3 new parameters to btrfs_log_new_name() in the
next patch - the name, its length and an index number. This way we end
up passing only 1 new parameter, the index number.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

show more ...

btrfs: change error handling for btrfs_delete_*_in_log

Currently we will abort the transaction if we get a random error (like
-EIO) while trying to remove the directory entries from the root log
dur

btrfs: change error handling for btrfs_delete_*_in_log

Currently we will abort the transaction if we get a random error (like
-EIO) while trying to remove the directory entries from the root log
during rename.

However since these are simply log tree related errors, we can mark the
trans as needing a full commit. Then if the error was truly
catastrophic we'll hit it during the normal commit and abort as
appropriate.

Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>

show more ...

btrfs: keep track of the last logged keys when logging a directory

After the first time we log a directory in the current transaction, for
each directory item in a changed leaf of the subvolume tree

btrfs: keep track of the last logged keys when logging a directory

After the first time we log a directory in the current transaction, for
each directory item in a changed leaf of the subvolume tree, we have to
check if we previously logged the item, in order to overwrite it in case
its data changed or skip it in case its data hasn't changed.

Checking if we have logged each item before not only wastes times, but it
also adds lock contention on the log tree. So in order to minimize the
number of times we do such checks, keep track of the offset of the last
key we logged for a directory and, on the next time we log the directory,
skip the checks for any new keys that have an offset greater than the
offset we have previously saved. This is specially effective for index
keys, because the offset for these keys comes from a monotonically
increasing counter.

This patch is part of a patchset comprised of the following 5 patches:

btrfs: remove root argument from btrfs_log_inode() and its callees
btrfs: remove redundant log root assignment from log_dir_items()
btrfs: factor out the copying loop of dir items from log_dir_items()
btrfs: insert items in batches when logging a directory when possible
btrfs: keep track of the last logged keys when logging a directory

This is patch 5/5.

The following test was used on a non-debug kernel to measure the impact
it has on a directory fsync:

$ cat test-dir-fsync.sh
#!/bin/bash

DEV=/dev/nvme0n1
MNT=/mnt/nvme0n1

NUM_NEW_FILES=100000
NUM_FILE_DELETES=1000

mkfs.btrfs -f $DEV
mount -o ssd $DEV $MNT

mkdir $MNT/testdir

for ((i = 1; i <= $NUM_NEW_FILES; i++)); do
echo -n > $MNT/testdir/file_$i
done

# fsync the directory, this will log the new dir items and the inodes
# they point to, because these are new inodes.
start=$(date +%s%N)
xfs_io -c "fsync" $MNT/testdir
end=$(date +%s%N)

dur=$(( (end - start) / 1000000 ))
echo "dir fsync took $dur ms after adding $NUM_NEW_FILES files"

# sync to force transaction commit and wipeout the log.
sync

del_inc=$(( $NUM_NEW_FILES / $NUM_FILE_DELETES ))
for ((i = 1; i <= $NUM_NEW_FILES; i += $del_inc)); do
rm -f $MNT/testdir/file_$i
done

# fsync the directory, this will only log dir items, there are no
# dentries pointing to new inodes.
start=$(date +%s%N)
xfs_io -c "fsync" $MNT/testdir
end=$(date +%s%N)

dur=$(( (end - start) / 1000000 ))
echo "dir fsync took $dur ms after deleting $NUM_FILE_DELETES files"

umount $MNT

Test results with NUM_NEW_FILES set to 100 000 and 1 000 000:

**** before patchset, 100 000 files, 1000 deletes ****

dir fsync took 848 ms after adding 100000 files
dir fsync took 175 ms after deleting 1000 files

**** after patchset, 100 000 files, 1000 deletes ****

dir fsync took 758 ms after adding 100000 files (-11.2%)
dir fsync took 63 ms after deleting 1000 files (-94.1%)

**** before patchset, 1 000 000 files, 1000 deletes ****

dir fsync took 9945 ms after adding 1000000 files
dir fsync took 473 ms after deleting 1000 files

**** after patchset, 1 000 000 files, 1000 deletes ****

dir fsync took 8677 ms after adding 1000000 files (-13.6%)
dir fsync took 146 ms after deleting 1000 files (-105.6%)

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

show more ...

btrfs: pass the dentry to btrfs_log_new_name() instead of the inode

[ Upstream commit d5f5bd546552a94eefd68c42f40f778c40a89d2c ]

In the next patch in the series, there will be the need to access th

btrfs: pass the dentry to btrfs_log_new_name() instead of the inode

[ Upstream commit d5f5bd546552a94eefd68c42f40f778c40a89d2c ]

In the next patch in the series, there will be the need to access the old
name, and its length, of an inode when logging the inode during a rename.
So instead of passing the inode to btrfs_log_new_name() pass the dentry,
because from the dentry we can get the inode, the name and its length.

This will avoid passing 3 new parameters to btrfs_log_new_name() in the
next patch - the name, its length and an index number. This way we end
up passing only 1 new parameter, the index number.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>

show more ...

btrfs: pass the dentry to btrfs_log_new_name() instead of the inode

[ Upstream commit d5f5bd546552a94eefd68c42f40f778c40a89d2c ]

In the next patch in the series, there will be the need to access th

btrfs: pass the dentry to btrfs_log_new_name() instead of the inode

[ Upstream commit d5f5bd546552a94eefd68c42f40f778c40a89d2c ]

In the next patch in the series, there will be the need to access the old
name, and its length, of an inode when logging the inode during a rename.
So instead of passing the inode to btrfs_log_new_name() pass the dentry,
because from the dentry we can get the inode, the name and its length.

This will avoid passing 3 new parameters to btrfs_log_new_name() in the
next patch - the name, its length and an index number. This way we end
up passing only 1 new parameter, the index number.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>

show more ...

btrfs: pass the dentry to btrfs_log_new_name() instead of the inode

[ Upstream commit d5f5bd546552a94eefd68c42f40f778c40a89d2c ]

In the next patch in the series, there will be the need to access th

btrfs: pass the dentry to btrfs_log_new_name() instead of the inode

[ Upstream commit d5f5bd546552a94eefd68c42f40f778c40a89d2c ]

In the next patch in the series, there will be the need to access the old
name, and its length, of an inode when logging the inode during a rename.
So instead of passing the inode to btrfs_log_new_name() pass the dentry,
because from the dentry we can get the inode, the name and its length.

This will avoid passing 3 new parameters to btrfs_log_new_name() in the
next patch - the name, its length and an index number. This way we end
up passing only 1 new parameter, the index number.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>

show more ...

btrfs: pass the dentry to btrfs_log_new_name() instead of the inode

[ Upstream commit d5f5bd546552a94eefd68c42f40f778c40a89d2c ]

In the next patch in the series, there will be the need to access th

btrfs: pass the dentry to btrfs_log_new_name() instead of the inode

[ Upstream commit d5f5bd546552a94eefd68c42f40f778c40a89d2c ]

In the next patch in the series, there will be the need to access the old
name, and its length, of an inode when logging the inode during a rename.
So instead of passing the inode to btrfs_log_new_name() pass the dentry,
because from the dentry we can get the inode, the name and its length.

This will avoid passing 3 new parameters to btrfs_log_new_name() in the
next patch - the name, its length and an index number. This way we end
up passing only 1 new parameter, the index number.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>

show more ...

btrfs: pass the dentry to btrfs_log_new_name() instead of the inode

[ Upstream commit d5f5bd546552a94eefd68c42f40f778c40a89d2c ]

In the next patch in the series, there will be the need to access th

btrfs: pass the dentry to btrfs_log_new_name() instead of the inode

[ Upstream commit d5f5bd546552a94eefd68c42f40f778c40a89d2c ]

In the next patch in the series, there will be the need to access the old
name, and its length, of an inode when logging the inode during a rename.
So instead of passing the inode to btrfs_log_new_name() pass the dentry,
because from the dentry we can get the inode, the name and its length.

This will avoid passing 3 new parameters to btrfs_log_new_name() in the
next patch - the name, its length and an index number. This way we end
up passing only 1 new parameter, the index number.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>

show more ...

btrfs: pass the dentry to btrfs_log_new_name() instead of the inode

[ Upstream commit d5f5bd546552a94eefd68c42f40f778c40a89d2c ]

In the next patch in the series, there will be the need to access th

btrfs: pass the dentry to btrfs_log_new_name() instead of the inode

[ Upstream commit d5f5bd546552a94eefd68c42f40f778c40a89d2c ]

In the next patch in the series, there will be the need to access the old
name, and its length, of an inode when logging the inode during a rename.
So instead of passing the inode to btrfs_log_new_name() pass the dentry,
because from the dentry we can get the inode, the name and its length.

This will avoid passing 3 new parameters to btrfs_log_new_name() in the
next patch - the name, its length and an index number. This way we end
up passing only 1 new parameter, the index number.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>

show more ...

btrfs: pass the dentry to btrfs_log_new_name() instead of the inode

[ Upstream commit d5f5bd546552a94eefd68c42f40f778c40a89d2c ]

In the next patch in the series, there will be the need to access th

btrfs: pass the dentry to btrfs_log_new_name() instead of the inode

[ Upstream commit d5f5bd546552a94eefd68c42f40f778c40a89d2c ]

In the next patch in the series, there will be the need to access the old
name, and its length, of an inode when logging the inode during a rename.
So instead of passing the inode to btrfs_log_new_name() pass the dentry,
because from the dentry we can get the inode, the name and its length.

This will avoid passing 3 new parameters to btrfs_log_new_name() in the
next patch - the name, its length and an index number. This way we end
up passing only 1 new parameter, the index number.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>

show more ...

btrfs: pass the dentry to btrfs_log_new_name() instead of the inode

[ Upstream commit d5f5bd546552a94eefd68c42f40f778c40a89d2c ]

In the next patch in the series, there will be the need to access th

btrfs: pass the dentry to btrfs_log_new_name() instead of the inode

[ Upstream commit d5f5bd546552a94eefd68c42f40f778c40a89d2c ]

In the next patch in the series, there will be the need to access the old
name, and its length, of an inode when logging the inode during a rename.
So instead of passing the inode to btrfs_log_new_name() pass the dentry,
because from the dentry we can get the inode, the name and its length.

This will avoid passing 3 new parameters to btrfs_log_new_name() in the
next patch - the name, its length and an index number. This way we end
up passing only 1 new parameter, the index number.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>

show more ...

btrfs: pass the dentry to btrfs_log_new_name() instead of the inode

[ Upstream commit d5f5bd546552a94eefd68c42f40f778c40a89d2c ]

In the next patch in the series, there will be the need to access th

btrfs: pass the dentry to btrfs_log_new_name() instead of the inode

[ Upstream commit d5f5bd546552a94eefd68c42f40f778c40a89d2c ]

In the next patch in the series, there will be the need to access the old
name, and its length, of an inode when logging the inode during a rename.
So instead of passing the inode to btrfs_log_new_name() pass the dentry,
because from the dentry we can get the inode, the name and its length.

This will avoid passing 3 new parameters to btrfs_log_new_name() in the
next patch - the name, its length and an index number. This way we end
up passing only 1 new parameter, the index number.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>

show more ...

btrfs: make fast fsyncs wait only for writeback

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
fo

btrfs: make fast fsyncs wait only for writeback

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 b5e6c3e170b770 ("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>

show more ...

btrfs: do not commit logs and transactions during link and rename operations

Since commit d4682ba03ef618 ("Btrfs: sync log after logging new name") we
started to commit logs, and fallback to transac

btrfs: do not commit logs and transactions during link and rename operations

Since commit d4682ba03ef618 ("Btrfs: sync log after logging new name") we
started to commit logs, and fallback to transaction commits when we failed
to log the new names or commit the logs, after link and rename operations
when the target inodes (or their parents) were previously logged in the
current transaction. This was to avoid losing directories despite an
explicit fsync on them when they are ancestors of some inode that got a
new named logged, due to a link or rename operation. However that adds the
cost of starting IO and waiting for it to complete, which can cause higher
latencies for applications.

Instead of doing that, just make sure that when we log a new name for an
inode we don't mark any of its ancestors as logged, so that if any one
does an fsync against any of them, without doing any other change on them,
the fsync commits the log. This way we only pay the cost of a log commit
(or a transaction commit if something goes wrong or a new block group was
created) if the application explicitly asks to fsync any of the parent
directories.

Using dbench, which mixes several filesystems operations including renames,
revealed some significant latency gains. The following script that uses
dbench was used to test this:

#!/bin/bash

DEV=/dev/nvme0n1
MNT=/mnt/btrfs
MOUNT_OPTIONS="-o ssd -o space_cache=v2"
MKFS_OPTIONS="-m single -d single"
THREADS=16

echo "performance" | tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor
mkfs.btrfs -f $MKFS_OPTIONS $DEV
mount $MOUNT_OPTIONS $DEV $MNT

dbench -t 300 -D $MNT $THREADS

umount $MNT

The test was run on bare metal, no virtualization, on a box with 12 cores
(Intel i7-8700), 64Gb of RAM and using 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).

Results before this patch:

Operation Count AvgLat MaxLat
----------------------------------------
NTCreateX 10750455 0.011 155.088
Close 7896674 0.001 0.243
Rename 455222 2.158 1101.947
Unlink 2171189 0.067 121.638
Deltree 256 2.425 7.816
Mkdir 128 0.002 0.003
Qpathinfo 9744323 0.006 21.370
Qfileinfo 1707092 0.001 0.146
Qfsinfo 1786756 0.001 11.228
Sfileinfo 875612 0.003 21.263
Find 3767281 0.025 9.617
WriteX 5356924 0.011 211.390
ReadX 16852694 0.003 9.442
LockX 35008 0.002 0.119
UnlockX 35008 0.001 0.138
Flush 753458 4.252 1102.249

Throughput 1128.35 MB/sec 16 clients 16 procs max_latency=1102.255 ms

Results after this patch:

16 clients, after

Operation Count AvgLat MaxLat
----------------------------------------
NTCreateX 11471098 0.012 448.281
Close 8426396 0.001 0.925
Rename 485746 0.123 267.183
Unlink 2316477 0.080 63.433
Deltree 288 2.830 11.144
Mkdir 144 0.003 0.010
Qpathinfo 10397420 0.006 10.288
Qfileinfo 1822039 0.001 0.169
Qfsinfo 1906497 0.002 14.039
Sfileinfo 934433 0.004 2.438
Find 4019879 0.026 10.200
WriteX 5718932 0.011 200.985
ReadX 17981671 0.003 10.036
LockX 37352 0.002 0.076
UnlockX 37352 0.001 0.109
Flush 804018 5.015 778.033

Throughput 1201.98 MB/sec 16 clients 16 procs max_latency=778.036 ms
(+6.5% throughput, -29.4% max latency, -75.8% rename latency)

Test case generic/498 from fstests tests the scenario that the previously
mentioned commit fixed.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

show more ...