- 07 Oct, 2020 40 commits
-
-
Nikolay Borisov authored
It makes no sense to have sysfs-related routines be responsible for properly initialising the fs_info pointer of struct btrfs_fs_device. Instead this can be streamlined by making it the responsibility of btrfs_init_devices_late to initialize it. That function already initializes fs_info of every individual device in btrfs_fs_devices. As far as clearing it is concerned it makes sense to move it to close_fs_devices. That function is only called when struct btrfs_fs_devices is no longer in use - either for holding seeds or main devices for a mounted filesystem. Reviewed-by:
Josef Bacik <josef@toxicpanda.com> Reviewed-by:
Anand Jain <anand.jain@oracle.com> Signed-off-by:
Nikolay Borisov <nborisov@suse.com> Signed-off-by:
David Sterba <dsterba@suse.com>
-
Nikolay Borisov authored
The return value of this function conveys absolutely no information. All callers already check the state of fs_devices->opened to decide how to proceed. So convert the function to returning void. While at it make btrfs_close_devices also return void. Reviewed-by:
-
Nikolay Borisov authored
This prepares the code to switching seeds devices to a proper list. Reviewed-by:
-
Nikolay Borisov authored
This is in preparation for moving fs_devices to proper lists. Reviewed-by:
-
Nikolay Borisov authored
There's no practical reason too use 'err' as a variable to convey errors. In fact it's value is either set explicitly in the beginning of the function or it simply takes the value of 'ret'. Not conforming to the usual pattern of having ret be the only variable used to convey errors makes the code more error prone to bugs. In fact one such bug was introduced by 6bf9e4bd ("btrfs: inode: Verify inode mode toi avoid NULL pointer dereference") by assigning the error value to 'ret' and not 'err'. Let's fix that issue and make the function less tricky by leaving only ret to convey error values. Signed-off-by:
-
Josef Bacik authored
iomap dio will run generic_write_sync() for us if the iocb is DSYNC. This is problematic for us because of 2 reasons: 1. we hold the inode_lock() during this operation, and we take it in generic_write_sync() 2. we hold a read lock on the dio_sem but take the write lock in fsync Since we don't want to rip out this code right now, but reworking the locking is a bit much to do at this point, work around this problem with this masterpiece of a patch. First, we clear DSYNC on the iocb so that the iomap stuff doesn't know that it needs to handle the sync. We save this fact in current->journal_info, because we need to see do special things once we're in iomap_begin, and we have no way to pass private information into iomap_dio_rw(). Next we specify a separate iomap_dio_ops for sync, which implements an ->end_io() callback that gets called when the dio completes. This is important for AIO, because we really do need to run generic_write_sync() if we complete asynchronously. However if we're still in the submitting context when we enter ->end_io() we clear the flag so that the submitter knows they're the ones that needs to run generic_write_sync(). This is meant to be temporary. We need to work out how to eliminate the inode_lock() and the dio_sem in our fsync and use another mechanism to protect these operations. Tested-by:
Johannes Thumshirn <johannes.thumshirn@wdc.com> Signed-off-by:
-
Goldwyn Rodrigues authored
We're using direct io implementation based on buffer heads. This patch switches to the new iomap infrastructure. Switch from __blockdev_direct_IO() to iomap_dio_rw(). Rename btrfs_get_blocks_direct() to btrfs_dio_iomap_begin() and use it as iomap_begin() for iomap direct I/O functions. This function allocates and locks all the blocks required for the I/O. btrfs_submit_direct() is used as the submit_io() hook for direct I/O ops. Since we need direct I/O reads to go through iomap_dio_rw(), we change file_operations.read_iter() to a btrfs_file_read_iter() which calls btrfs_direct_IO() for direct reads and falls back to generic_file_buffered_read() for incomplete reads and buffered reads. We don't need address_space.direct_IO() anymore: set it to noop. Similarly, we don't need flags used in __blockdev_direct_IO(). iomap is capable of direct I/O reads from a hole, so we don't need to return -ENOENT. Btrfs direct I/O is now done under i_rwsem, shared in case of reads and exclusive in case of writes. This guards against simultaneous truncates. Use iomap->iomap_end() to check for failed or incomplete direct I/O: - for writes, call __endio_write_update_ordered() - for reads, unlock extents btrfs_dio_data is now hooked in iomap->private and not current->journal_info. It carries the reservation variable and the amount of data submitted, so we can calculate the amount of data to call __endio_write_update_ordered in case of an error. This patch removes last use of struct buffer_head from btrfs. Reviewed-by:
Goldwyn Rodrigues <rgoldwyn@suse.com> Reviewed-by:
-
Qu Wenruo authored
Commit 1c11b63e ("btrfs: replace pending/pinned chunks lists with io tree") introduced btrfs_device::alloc_state extent io tree, but it doesn't initialize the fs_info and owner member. This means the following features are not properly supported: - Fs owner report for insert_state() error Without fs_info initialized, although btrfs_err() won't panic, it won't output which fs is causing the error. - Wrong owner for trace events alloc_state will get the owner as pinned extents. Fix this by assiging proper fs_info and owner for btrfs_device::alloc_state. Fixes: 1c11b63e ("btrfs: replace pending/pinned chunks lists with io tree") Reviewed-by:
Qu Wenruo <wqu@suse.com> Reviewed-by:
-
Marcos Paulo de Souza authored
Since it's inclusion on 9afc6649 ("btrfs: block-group: refactor how we read one block group item") this function always returned 0, so there is no need to check for the returned value. Reviewed-by:
Marcos Paulo de Souza <mpdesouza@suse.com> Reviewed-by:
-
Leon Romanovsky authored
The compilation with W=1 generates the following warnings: fs/btrfs/sysfs.c:1630:6: warning: variable 'ret' set but not used [-Wunused-but-set-variable] 1630 | int ret; | ^~~ fs/btrfs/sysfs.c:1629:6: warning: variable 'features' set but not used [-Wunused-but-set-variable] 1629 | u64 features; | ^~~~~~~~ [ The unused variables are leftover from e410e34f ("Revert "btrfs: synchronize incompat feature bits with sysfs files""), which needs to be properly fixed by moving feature bit manipulation from the sysfs context. Silence the warning to save pepople time, we got several reports. ] Signed-off-by:Leon Romanovsky <leonro@nvidia.com> Reviewed-by:
-
Filipe Manana authored
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 b5e6c3e1 ("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:
-
Filipe Manana authored
Since commit d4682ba0 ("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 authored
During a rename we pin the log to make sure no one commits a log that reflects an ongoing rename operation, as it might result in a committed log where it recorded the unlink of the old name without having recorded the new name. However we are taking the subvolume's log_mutex before incrementing the log_writers counter, which is not necessary since that counter is atomic and we only remove the old name from the log and add the new name to the log after we have incremented log_writers, ensuring that no one can commit the log after we have removed the old name from the log and before we added the new name to the log. By taking the log_mutex lock we are just adding unnecessary contention on the lock, which can become visible for workloads that mix renames with fsyncs, writes for files opened with O_SYNC and unlink operations (if the inode or its parent were fsynced before in the current transaction). So just remove the lock and unlock of the subvolume's log_mutex at btrfs_pin_log_trans(). Using dbench, which mixes different types of operations that end up taking that mutex (fsyncs, renames, unlinks and writes into files opened with O_SYNC) revealed some small gains. The following script that calls dbench was used: #!/bin/bash DEV=/dev/nvme0n1 MNT=/mnt/btrfs MOUNT_OPTIONS="-o ssd -o space_cache=v2" MKFS_OPTIONS="-m single -d single" THREADS=32 echo "performance" | tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor mkfs.btrfs -f $MKFS_OPTIONS $DEV mount $MOUNT_OPTIONS $DEV $MNT dbench -s -t 600 -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 4410848 0.017 738.640 Close 3240222 0.001 0.834 Rename 186850 7.478 1272.476 Unlink 890875 0.128 785.018 Deltree 128 2.846 12.081 Mkdir 64 0.002 0.003 Qpathinfo 3997659 0.009 11.171 Qfileinfo 701307 0.001 0.478 Qfsinfo 733494 0.002 1.103 Sfileinfo 359362 0.004 3.266 Find 1546226 0.041 4.128 WriteX 2202803 7.905 1376.989 ReadX 6917775 0.003 3.887 LockX 14392 0.002 0.043 UnlockX 14392 0.001 0.085 Flush 309225 0.128 1033.936 Throughput 231.555 MB/sec (sync open) 32 clients 32 procs max_latency=1376.993 ms Results after this patch: Operation Count AvgLat MaxLat ---------------------------------------- NTCreateX 4603244 0.017 232.776 Close 3381299 0.001 1.041 Rename 194871 7.251 1073.165 Unlink 929730 0.133 119.233 Deltree 128 2.871 10.199 Mkdir 64 0.002 0.004 Qpathinfo 4171343 0.009 11.317 Qfileinfo 731227 0.001 1.635 Qfsinfo 765079 0.002 3.568 Sfileinfo 374881 0.004 1.220 Find 1612964 0.041 4.675 WriteX 2296720 7.569 1178.204 ReadX 7213633 0.003 3.075 LockX 14976 0.002 0.076 UnlockX 14976 0.001 0.061 Flush 322635 0.102 579.505 Throughput 241.4 MB/sec (sync open) 32 clients 32 procs max_latency=1178.207 ms (+4.3% throughput, -14.4% max latency) Reviewed-by:
-
David Sterba authored
There's a custom callback passed to btrfs_compare_trees which happens to be named exactly same as the existing function implementing it. This is confusing and the indirection is not necessary for our needs. Compiler is clever enough to call it directly so there's effectively no change. Reviewed-by:
-
David Sterba authored
There's already defined _rs within ctree.h:btrfs_printk_ratelimited, local variables should not use _ to avoid such name clashes with macro-local variables. Reviewed-by:
-
David Sterba authored
In btrfs_orphan_cleanup, there's another instance of fs_info, but it's the same as the one we already have. In btrfs_backref_finish_upper_links, rb_node is same type and used as temporary cursor to the tree. Reviewed-by:
-
David Sterba authored
The declarations of compression algorithm callbacks are defined in the .c file as they're used from there. Compiler warns that there are no declarations for public functions when compiling lzo.c/zlib.c/zstd.c. Fix that by moving the declarations to the header as it's the common place for all of them. Reviewed-by:
-
David Sterba authored
The function btrfs_feature_set_name returns a const char pointer, the second const is not necessary and reported as a warning: In file included from fs/btrfs/space-info.c:6: fs/btrfs/sysfs.h:16:1: warning: type qualifiers ignored on function return type [-Wignored-qualifiers] 16 | const char * const btrfs_feature_set_name(enum btrfs_feature_set set); | ^~~~~ Reviewed-by: -
Qu Wenruo authored
We're just doing rounding up to sectorsize to calculate the lockend. There is no need to do the unnecessary length calculation, just direct round_up() is enough. Reviewed-by:
-
Josef Bacik authored
Dave reported an issue where generic/102 would sometimes hang. This turned out to be because we'd get into this spot where we were no longer making progress on data reservations because our exit condition was not met. The log is basically while (!space_info->full && !list_empty(&space_info->tickets)) flush_space(space_info, flush_state); where flush state is our various flush states, but doesn't include ALLOC_CHUNK_FORCE. This is because we actually lead with allocating chunks, and so the assumption was that once you got to the actual flushing states you could no longer allocate chunks. This was a stupid assumption, because you could have deleted block groups that would be reclaimed by a transaction commit, thus unsetting space_info->full. This is essentially what happens with generic/102, and so sometimes you'd get stuck in the flushing loop because we weren't allocating chunks, but flushing space wasn't giving us what we needed to make progress. Fix this by adding ALLOC_CHUNK_FORCE to the end of our flushing states, that way we will eventually bail out because we did end up with space_info->full if we free'd a chunk previously. Otherwise, as is the case for this test, we'll allocate our chunk and continue on our happy merry way. Reported-by:
-
Josef Bacik authored
The data flushing steps are not obvious to people other than myself and Chris. Write a giant comment explaining the reasoning behind each flush step for data as well as why it is in that particular order. Reviewed-by:
-
Josef Bacik authored
Now that we have the data ticketing stuff in place, move normal data reservations to use an async reclaim helper to satisfy tickets. Before we could have multiple tasks race in and both allocate chunks, resulting in more data chunks than we would necessarily need. Serializing these allocations and making a single thread responsible for flushing will only allocate chunks as needed, as well as cut down on transaction commits and other flush related activities. Priority reservations will still work as they have before, simply trying to allocate a chunk until they can make their reservation. Reviewed-by:
-
Josef Bacik authored
We can end up with freed extents in the delayed refs, and thus may_commit_transaction() may not think we have enough pinned space to commit the transaction and we'll ENOSPC early. Handle this by running the delayed refs in order to make sure pinned is uptodate before we try to commit the transaction. Tested-by:
-
Josef Bacik authored
Before we were waiting on iputs after we committed the transaction, but this doesn't really make much sense. We want to reclaim any space we may have in order to be more likely to commit the transaction, due to pinned space being added by running the delayed iputs. Fix this by making delayed iputs run before committing the transaction. Reviewed-by:
-
Josef Bacik authored
We used to unconditionally commit the transaction at least 2 times and then on the 3rd try check against pinned space to make sure committing the transaction was worth the effort. This is overkill, we know nobody is going to steal our reservation, and if we can't make our reservation with the pinned amount simply bail out. This also cleans up the passing of bytes_needed to may_commit_transaction, as that was the thing we added into place in order to accomplish this behavior. We no longer need it so remove that mess. Reviewed-by:
-
Josef Bacik authored
This was an old wart left over from how we previously did data reservations. Before we could have people race in and take a reservation while we were flushing space, so we needed to make sure we looped a few times before giving up. Now that we're using the ticketing infrastructure we don't have to worry about this and can drop the logic altogether. Reviewed-by:
-
Josef Bacik authored
Now that data reservations follow the same pattern as metadata reservations we can simply rename __reserve_metadata_bytes to __reserve_bytes and use that helper for data reservations. Things to keep in mind, btrfs_can_overcommit() returns 0 for data, because we can never overcommit. We also will never pass in FLUSH_ALL for data, so we'll simply be added to the priority list and go straight into handle_reserve_ticket. Reviewed-by:
-
Josef Bacik authored
Nikolay reported a problem where generic/371 would fail sometimes with a slow drive. The gist of the test is that we fallocate a file in parallel with a pwrite of a different file. These two files combined are smaller than the file system, but sometimes the pwrite would ENOSPC. A fair bit of investigation uncovered the fact that the fallocate workload was racing in and grabbing the free space that the pwrite workload was trying to free up so it could make its own reservation. After a few loops of this eventually the pwrite workload would error out with an ENOSPC. We've had the same problem with metadata as well, and we serialized all metadata allocations to satisfy this problem. This wasn't usually a problem with data because data reservations are more straightforward, but obviously could still happen. Fix this by not allowing reservations to occur if there are any pending tickets waiting to be satisfied on the space info. Reviewed-by:
-
Josef Bacik authored
Now that we have all the infrastructure in place, use the ticketing infrastructure to make data allocations. This still maintains the exact same flushing behavior, but now we're using tickets to get our reservations satisfied. Reviewed-by:
-
Josef Bacik authored
Create a new function btrfs_reserve_data_bytes() in order to handle data reservations. This uses the new flush types and flush states to handle making data reservations. This patch specifically does not change any functionality, and is purposefully not cleaned up in order to make bisection easier for the future patches. The new helper is identical to the old helper in how it handles data reservations. We first try to force a chunk allocation, and then we run through the flush states all at once and in the same order that they were done with the old helper. Subsequent patches will clean this up and change the behavior of the flushing, and it is important to keep those changes separate so we can easily bisect down to the patch that caused the regression, rather than the patch that made us start using the new infrastructure. Reviewed-by:
-
Josef Bacik authored
Data space flushing currently unconditionally commits the transaction twice in a row, and the last time it checks if there's enough pinned extents to satisfy its reservation before deciding to commit the transaction for the 3rd and final time. Encode this logic into may_commit_transaction(). In the next patch we will pass in U64_MAX for bytes_needed the first two times, and the final time we will pass in the actual bytes we need so the normal logic will apply. This patch exists solely to make the logical changes I will make to the flushing state machine separate to make it easier to bisect any performance related regressions. Reviewed-by:
-
Josef Bacik authored
Currently the way we do data reservations is by seeing if we have enough space in our space_info. If we do not and we're a normal inode we'll 1) Attempt to force a chunk allocation until we can't anymore. 2) If that fails we'll flush delalloc, then commit the transaction, then run the delayed iputs. If we are a free space inode we're only allowed to force a chunk allocation. In order to use the normal flushing mechanism we need to encode this into a flush state array for normal inodes. Since both will start with allocating chunks until the space info is full there is no need to add this as a flush state, this will be handled specially. Reviewed-by:
-
Josef Bacik authored
Right now if the space is freed up after the ordered extents complete (which is likely since the reservations are held until they complete), we would do extra delalloc flushing before we'd notice that we didn't have any more tickets. Fix this by moving the tickets check after our wait_ordered_extents check. Tested-by:
-
Josef Bacik authored
The original iteration of flushing had us flushing delalloc and then checking to see if we could make our reservation, thus we were very careful about how many pages we would flush at once. But now that everything is async and we satisfy tickets as the space becomes available we don't have to keep track of any of this, simply try and flush the number of dirty inodes we may have in order to reclaim space to make our reservation. This cleans up our delalloc flushing significantly. The async_pages stuff is dropped because btrfs_start_delalloc_roots() handles the case that we generate async extents for us, so we no longer require this extra logic. Reviewed-by:
-
Josef Bacik authored
We are going to use the ticket infrastructure for data, so use the btrfs_space_info_free_bytes_may_use() helper in btrfs_free_reserved_data_space_noquota() so we get the btrfs_try_granting_tickets call when we free our reservation. Reviewed-by:
-
Josef Bacik authored
If we have compression on we could free up more space than we reserved, and thus be able to make a space reservation. Add the call for this scenario. Reviewed-by:
-
Josef Bacik authored
When unpinning we were only calling btrfs_try_granting_tickets() if global_rsv->space_info == space_info, which is problematic because we use ticketing for SYSTEM chunks, and want to use it for DATA as well. Fix this by moving this call outside of that if statement. Reviewed-by:
-
Josef Bacik authored
We were missing a call to btrfs_try_granting_tickets in btrfs_free_reserved_bytes, so add it to handle the case where we're able to satisfy an allocation because we've freed a pending reservation. Reviewed-by:
-
Josef Bacik authored
We have traditionally used flush_space() to flush metadata space, so we've been unconditionally using btrfs_metadata_alloc_profile() for our profile to allocate a chunk. However if we're going to use this for data we need to use btrfs_get_alloc_profile() on the space_info we pass in. Reviewed-by:
-
Josef Bacik authored
Currently shrink_delalloc just looks up the metadata space info, but this won't work if we're trying to reclaim space for data chunks. We get the right space_info we want passed into flush_space, so simply pass that along to shrink_delalloc. Reviewed-by:
-
