Compare commits

...

6 Commits

Author SHA1 Message Date
Auke Kok
cf635cd0da Ramping up data preallocation
Ramps up data preallocation based on the number of online
blocks. This results in a simple 2<<n block allocation pattern
until n=11 (2048) - the default value of data_prealloc_blocks.

Signed-off-by: Auke Kok <auke.kok@versity.com>
2026-04-21 18:15:42 -04:00
Zach Brown
e20765a9c7 Merge pull request #300 from versity/auke/more_false_positive_failures
Auke/more false positive failures: xfs lockdep miss, newline
2026-04-17 09:17:50 -07:00
Zach Brown
066da5c2a2 Merge pull request #297 from versity/auke/quota_mod_trans_hold
Hold transaction in scoutfs_quota_mod_rule to prevent alloc corruption.
2026-04-17 09:16:41 -07:00
Auke Kok
7eacc7139c Hold transaction in scoutfs_quota_mod_rule to prevent alloc corruption.
scoutfs_quota_mod_rule calls scoutfs_item_create/delete which use
the transaction allocator but it never held it. Without the hold,
a concurrent transaction commit can call scoutfs_alloc_init to
reinitialize the allocator while dirty_alloc_blocks is in the middle
of setting up the freed list block. This overwrites alloc->freed with
the server's fresh (empty) state, causing a blkno mismatch BUG_ON
in list_block_add.

Reproduced by stressing concurrent quota add/del operations across
mounts. Crashdump analysis confirms dirty_list_block COW'd a freed
block (fr_old=9842, new blkno=9852) but by the time list_block_add
ran, freed.ref.blkno was 0 with first_nr=0 and total_nr=0: the freed
list head had been zeroed by a concurrent alloc_init.

Fix by adding scoutfs_hold_trans/scoutfs_release_trans around the
item modification in scoutfs_quota_mod_rule, preventing transaction
commit from racing with the allocator use.

Rename the 'unlock' label to 'release' since 'out' now directly
does the unlock. The unlock safely handles a NULL lock.

Signed-off-by: Auke Kok <auke.kok@versity.com>
2026-04-16 16:20:47 -07:00
Auke Kok
9e3b01b3b4 Filter newlines out dmesg.new
Without overly broad filtering empty lines from dmesg, filter
them so dmesg.new doesn't trigger a test failure. I don't want
to overly process dmesg, so do this as late as possible.

The xfs lockdep patterns can forget a leading/trailing empty line,
causing a failure despite the explicit removal of the lockdep
false positive.

Signed-off-by: Auke Kok <auke.kok@versity.com>
2026-04-15 10:36:28 -07:00
Auke Kok
876c233f06 Ignore another xfs lockdep class
This already caught xfs_nondir_ilock_class, but recent CI runs
have been hitting xfs_dir_ilock_class, too.

Signed-off-by: Auke Kok <auke.kok@versity.com>
2026-04-15 10:36:28 -07:00
5 changed files with 39 additions and 14 deletions

View File

@@ -384,6 +384,12 @@ static inline u64 ext_last(struct scoutfs_extent *ext)
* This can waste a lot of space for small or sparse files but is
* reasonable when a file population is known to be large and dense but
* known to be written with non-streaming write patterns.
*
* In either strategy, preallocation ramps up proportionally with the
* file's online block count rather than jumping to the full prealloc
* size. This avoids overallocation for small files in mixed-size
* workloads while still allowing large files to benefit from full
* preallocation.
*/
static int alloc_block(struct super_block *sb, struct inode *inode,
struct scoutfs_extent *ext, u64 iblock,
@@ -400,6 +406,7 @@ static int alloc_block(struct super_block *sb, struct inode *inode,
struct scoutfs_extent found;
struct scoutfs_extent pre = {0,};
bool undo_pre = false;
bool have_onoff = false;
u64 blkno = 0;
u64 online;
u64 offline;
@@ -445,6 +452,7 @@ static int alloc_block(struct super_block *sb, struct inode *inode,
* blocks.
*/
scoutfs_inode_get_onoff(inode, &online, &offline);
have_onoff = true;
if (iblock > 1 && iblock == online) {
ret = scoutfs_ext_next(sb, &data_ext_ops, &args,
iblock, 1, &found);
@@ -491,6 +499,16 @@ static int alloc_block(struct super_block *sb, struct inode *inode,
/* overall prealloc limit */
count = min_t(u64, count, opts.data_prealloc_blocks);
/*
* Ramp preallocation up proportionally with the file's online
* block count rather than jumping to the full prealloc size.
*/
if (!ext->len) {
if (!have_onoff)
scoutfs_inode_get_onoff(inode, &online, &offline);
count = max_t(u64, 1, min(count, online));
}
ret = scoutfs_alloc_data(sb, datinf->alloc, datinf->wri,
&datinf->dalloc, count, &blkno, &count);
if (ret < 0)

View File

@@ -34,6 +34,7 @@
#include "totl.h"
#include "util.h"
#include "quota.h"
#include "trans.h"
#include "counters.h"
#include "scoutfs_trace.h"
@@ -1086,6 +1087,10 @@ int scoutfs_quota_mod_rule(struct super_block *sb, bool is_add,
if (ret < 0)
goto out;
ret = scoutfs_hold_trans(sb, true);
if (ret < 0)
goto out;
down_write(&qtinf->rwsem);
if (is_add) {
@@ -1095,28 +1100,30 @@ int scoutfs_quota_mod_rule(struct super_block *sb, bool is_add,
else if (ret == 0)
ret = -EEXIST;
if (ret < 0)
goto unlock;
goto release;
rule_to_rule_val(&rv, &rule);
ret = scoutfs_item_create(sb, &key, &rv, sizeof(rv), lock);
if (ret < 0)
goto unlock;
goto release;
} else {
ret = find_rule(sb, &rule, &key, lock) ?:
scoutfs_item_delete(sb, &key, lock);
if (ret < 0)
goto unlock;
goto release;
}
scoutfs_quota_invalidate(sb);
ret = 0;
unlock:
release:
up_write(&qtinf->rwsem);
scoutfs_unlock(sb, lock, SCOUTFS_LOCK_WRITE);
scoutfs_release_trans(sb);
out:
scoutfs_unlock(sb, lock, SCOUTFS_LOCK_WRITE);
if (is_add)
trace_scoutfs_quota_add_rule(sb, &rule, ret);
else

View File

@@ -87,7 +87,7 @@ awk '
( in_block == 2 ) {
buf = buf $0 "\n"
if ($0 ~ /<\/TASK>/) {
if (buf ~ /xfs_nondir_ilock_class/ && buf ~ /fs_reclaim/) {
if (buf ~ /xfs_(nondir_|dir_)?ilock_class/ && buf ~ /fs_reclaim/) {
# known xfs lockdep false positive, discard
} else {
printf "%s", buf

View File

@@ -8,10 +8,10 @@
/mnt/test/test/data-prealloc/file-1: extents: 32
/mnt/test/test/data-prealloc/file-2: extents: 32
== any writes to region prealloc get full extents
/mnt/test/test/data-prealloc/file-1: extents: 4
/mnt/test/test/data-prealloc/file-2: extents: 4
/mnt/test/test/data-prealloc/file-1: extents: 4
/mnt/test/test/data-prealloc/file-2: extents: 4
/mnt/test/test/data-prealloc/file-1: extents: 8
/mnt/test/test/data-prealloc/file-2: extents: 8
/mnt/test/test/data-prealloc/file-1: extents: 8
/mnt/test/test/data-prealloc/file-2: extents: 8
== streaming offline writes get full extents either way
/mnt/test/test/data-prealloc/file-1: extents: 4
/mnt/test/test/data-prealloc/file-2: extents: 4
@@ -20,8 +20,8 @@
== goofy preallocation amounts work
/mnt/test/test/data-prealloc/file-1: extents: 6
/mnt/test/test/data-prealloc/file-2: extents: 6
/mnt/test/test/data-prealloc/file-1: extents: 6
/mnt/test/test/data-prealloc/file-2: extents: 6
/mnt/test/test/data-prealloc/file-1: extents: 10
/mnt/test/test/data-prealloc/file-2: extents: 10
/mnt/test/test/data-prealloc/file-1: extents: 3
/mnt/test/test/data-prealloc/file-2: extents: 3
== block writes into region allocs hole

View File

@@ -694,8 +694,8 @@ for t in $tests; do
if [ "$sts" == "$T_PASS_STATUS" ]; then
dmesg | t_filter_dmesg > "$T_TMPDIR/dmesg.after"
diff --old-line-format="" --unchanged-line-format="" \
"$T_TMPDIR/dmesg.before" "$T_TMPDIR/dmesg.after" > \
"$T_TMPDIR/dmesg.new"
"$T_TMPDIR/dmesg.before" "$T_TMPDIR/dmesg.after" | \
grep -v '^$' > "$T_TMPDIR/dmesg.new"
if [ -s "$T_TMPDIR/dmesg.new" ]; then
message="unexpected messages in dmesg"