Archive-Team
/
zfs
2
0
Fork 0
Code Issues Pull Requests Packages Projects Releases 103 Wiki Activity

Do not persist user/group/project quota zap objects when unneeded 

In the zfs_id_over*quota functions, there is a short-circuit to skip
the zap_lookup when the quota zap does not exist. If quotas are never
used in a zpool, then the quota zap will never exist. But if
user/group/project quotas are ever used, the zap objects will be
created and will persist even if the quotas are deleted.

The quota zap_lookup in the write path can become a bottleneck for
write-heavy small I/O workloads. Before this commit, it was not
possible to remove this lookup without creating a new zpool.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Sam Atkinson <samatk@amazon.com>
Closes #14721
This commit is contained in:
Sam Atkinson 2023-10-20 17:22:04 -04:00 committed by GitHub
parent 57b4098562
commit 797f55ef12
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 16 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
module/zfs/zfs_quota.c
View File

@ -347,18 +347,32 @@ zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
if (*objp == 0) { if (*objp == 0) {
*objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA, *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
DMU_OT_NONE, 0, tx); DMU_OT_NONE, 0, tx);
VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ, VERIFY0(zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
zfs_userquota_prop_prefixes[type], 8, 1, objp, tx)); zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
} }
mutex_exit(&zfsvfs->z_lock);
if (quota == 0) { if (quota == 0) {
err = zap_remove(zfsvfs->z_os, *objp, buf, tx); err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
if (err == ENOENT) if (err == ENOENT)
err = 0; err = 0;
/*
* If the quota contains no more entries after the entry
* was removed, destroy the quota zap and remove the
* reference from zfsvfs. This will save us unnecessary
* zap_lookups for the quota during writes.
*/
uint64_t zap_nentries;
VERIFY0(zap_count(zfsvfs->z_os, *objp, &zap_nentries));
if (zap_nentries == 0) {
VERIFY0(zap_remove(zfsvfs->z_os, MASTER_NODE_OBJ,
zfs_userquota_prop_prefixes[type], tx));
VERIFY0(zap_destroy(zfsvfs->z_os, *objp, tx));
*objp = 0;
}
} else { } else {
err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, &quota, tx); err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, &quota, tx);
} }
mutex_exit(&zfsvfs->z_lock);
ASSERT(err == 0); ASSERT(err == 0);
if (fuid_dirtied) if (fuid_dirtied)
zfs_fuid_sync(zfsvfs, tx); zfs_fuid_sync(zfsvfs, tx);