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

Change zfs_snapshot_lock from mutex to rw lock 

By changing the zfs_snapshot_lock from a mutex to a rw lock the
zfsctl_lookup_objset() function can be allowed to run concurrently.
This should reduce the latency of fh_to_dentry lookups in ZFS
snapshots which are being accessed over NFS.

Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Chunwei Chen <david.chen@osnexus.com>
This commit is contained in:
Brian Behlendorf 2015-12-10 15:53:37 -08:00 committed by Ned Bass
parent 9b41d9c1b2
commit e16c04d643
1 changed files with 26 additions and 26 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

52
module/zfs/zfs_ctldir.c
View File

@ -102,7 +102,7 @@
*/ */
static avl_tree_t zfs_snapshots_by_name; static avl_tree_t zfs_snapshots_by_name;
static avl_tree_t zfs_snapshots_by_objsetid; static avl_tree_t zfs_snapshots_by_objsetid;
static kmutex_t zfs_snapshot_lock; static krwlock_t zfs_snapshot_lock;
/* /*
* Control Directory Tunables (.zfs) * Control Directory Tunables (.zfs)
@ -195,7 +195,7 @@ zfsctl_snapshot_rele(zfs_snapentry_t *se)
static void static void
zfsctl_snapshot_add(zfs_snapentry_t *se) zfsctl_snapshot_add(zfs_snapentry_t *se)
{ {
ASSERT(MUTEX_HELD(&zfs_snapshot_lock)); ASSERT(RW_WRITE_HELD(&zfs_snapshot_lock));
refcount_add(&se->se_refcount, NULL); refcount_add(&se->se_refcount, NULL);
avl_add(&zfs_snapshots_by_name, se); avl_add(&zfs_snapshots_by_name, se);
avl_add(&zfs_snapshots_by_objsetid, se); avl_add(&zfs_snapshots_by_objsetid, se);
@ -210,7 +210,7 @@ zfsctl_snapshot_add(zfs_snapentry_t *se)
static void static void
zfsctl_snapshot_remove(zfs_snapentry_t *se) zfsctl_snapshot_remove(zfs_snapentry_t *se)
{ {
ASSERT(MUTEX_HELD(&zfs_snapshot_lock)); ASSERT(RW_WRITE_HELD(&zfs_snapshot_lock));
avl_remove(&zfs_snapshots_by_name, se); avl_remove(&zfs_snapshots_by_name, se);
avl_remove(&zfs_snapshots_by_objsetid, se); avl_remove(&zfs_snapshots_by_objsetid, se);
zfsctl_snapshot_rele(se); zfsctl_snapshot_rele(se);
@ -268,7 +268,7 @@ zfsctl_snapshot_find_by_name(char *snapname)
{ {
zfs_snapentry_t *se, search; zfs_snapentry_t *se, search;
ASSERT(MUTEX_HELD(&zfs_snapshot_lock)); ASSERT(RW_LOCK_HELD(&zfs_snapshot_lock));
search.se_name = snapname; search.se_name = snapname;
se = avl_find(&zfs_snapshots_by_name, &search, NULL); se = avl_find(&zfs_snapshots_by_name, &search, NULL);
@ -288,7 +288,7 @@ zfsctl_snapshot_find_by_objsetid(spa_t *spa, uint64_t objsetid)
{ {
zfs_snapentry_t *se, search; zfs_snapentry_t *se, search;
ASSERT(MUTEX_HELD(&zfs_snapshot_lock)); ASSERT(RW_LOCK_HELD(&zfs_snapshot_lock));
search.se_spa = spa; search.se_spa = spa;
search.se_objsetid = objsetid; search.se_objsetid = objsetid;
@ -308,7 +308,7 @@ zfsctl_snapshot_rename(char *old_snapname, char *new_snapname)
{ {
zfs_snapentry_t *se; zfs_snapentry_t *se;
ASSERT(MUTEX_HELD(&zfs_snapshot_lock)); ASSERT(RW_WRITE_HELD(&zfs_snapshot_lock));
se = zfsctl_snapshot_find_by_name(old_snapname); se = zfsctl_snapshot_find_by_name(old_snapname);
if (se == NULL) if (se == NULL)
@ -346,12 +346,12 @@ snapentry_expire(void *data)
* Reschedule the unmount if the zfs_snapentry_t wasn't removed. * Reschedule the unmount if the zfs_snapentry_t wasn't removed.
* This can occur when the snapshot is busy. * This can occur when the snapshot is busy.
*/ */
mutex_enter(&zfs_snapshot_lock); rw_enter(&zfs_snapshot_lock, RW_READER);
if ((se = zfsctl_snapshot_find_by_objsetid(spa, objsetid)) != NULL) { if ((se = zfsctl_snapshot_find_by_objsetid(spa, objsetid)) != NULL) {
zfsctl_snapshot_unmount_delay_impl(se, zfs_expire_snapshot); zfsctl_snapshot_unmount_delay_impl(se, zfs_expire_snapshot);
zfsctl_snapshot_rele(se); zfsctl_snapshot_rele(se);
} }
mutex_exit(&zfs_snapshot_lock); rw_exit(&zfs_snapshot_lock);
} }
/* /*
@ -362,7 +362,7 @@ snapentry_expire(void *data)
static void static void
zfsctl_snapshot_unmount_cancel(zfs_snapentry_t *se) zfsctl_snapshot_unmount_cancel(zfs_snapentry_t *se)
{ {
ASSERT(MUTEX_HELD(&zfs_snapshot_lock)); ASSERT(RW_LOCK_HELD(&zfs_snapshot_lock));
if (taskq_cancel_id(zfs_expire_taskq, se->se_taskqid) == 0) { if (taskq_cancel_id(zfs_expire_taskq, se->se_taskqid) == 0) {
se->se_taskqid = -1; se->se_taskqid = -1;
@ -398,14 +398,14 @@ zfsctl_snapshot_unmount_delay(spa_t *spa, uint64_t objsetid, int delay)
zfs_snapentry_t *se; zfs_snapentry_t *se;
int error = ENOENT; int error = ENOENT;
mutex_enter(&zfs_snapshot_lock); rw_enter(&zfs_snapshot_lock, RW_READER);
if ((se = zfsctl_snapshot_find_by_objsetid(spa, objsetid)) != NULL) { if ((se = zfsctl_snapshot_find_by_objsetid(spa, objsetid)) != NULL) {
zfsctl_snapshot_unmount_cancel(se); zfsctl_snapshot_unmount_cancel(se);
zfsctl_snapshot_unmount_delay_impl(se, delay); zfsctl_snapshot_unmount_delay_impl(se, delay);
zfsctl_snapshot_rele(se); zfsctl_snapshot_rele(se);
error = 0; error = 0;
} }
mutex_exit(&zfs_snapshot_lock); rw_exit(&zfs_snapshot_lock);
return (error); return (error);
} }
@ -420,12 +420,12 @@ zfsctl_snapshot_ismounted(char *snapname)
zfs_snapentry_t *se; zfs_snapentry_t *se;
boolean_t ismounted = B_FALSE; boolean_t ismounted = B_FALSE;
mutex_enter(&zfs_snapshot_lock); rw_enter(&zfs_snapshot_lock, RW_READER);
if ((se = zfsctl_snapshot_find_by_name(snapname)) != NULL) { if ((se = zfsctl_snapshot_find_by_name(snapname)) != NULL) {
zfsctl_snapshot_rele(se); zfsctl_snapshot_rele(se);
ismounted = B_TRUE; ismounted = B_TRUE;
} }
mutex_exit(&zfs_snapshot_lock); rw_exit(&zfs_snapshot_lock);
return (ismounted); return (ismounted);
} }
@ -579,14 +579,14 @@ zfsctl_destroy(zfs_sb_t *zsb)
spa_t *spa = zsb->z_os->os_spa; spa_t *spa = zsb->z_os->os_spa;
uint64_t objsetid = dmu_objset_id(zsb->z_os); uint64_t objsetid = dmu_objset_id(zsb->z_os);
mutex_enter(&zfs_snapshot_lock); rw_enter(&zfs_snapshot_lock, RW_WRITER);
if ((se = zfsctl_snapshot_find_by_objsetid(spa, objsetid)) if ((se = zfsctl_snapshot_find_by_objsetid(spa, objsetid))
!= NULL) { != NULL) {
zfsctl_snapshot_unmount_cancel(se); zfsctl_snapshot_unmount_cancel(se);
zfsctl_snapshot_remove(se); zfsctl_snapshot_remove(se);
zfsctl_snapshot_rele(se); zfsctl_snapshot_rele(se);
} }
mutex_exit(&zfs_snapshot_lock); rw_exit(&zfs_snapshot_lock);
} else if (zsb->z_ctldir) { } else if (zsb->z_ctldir) {
iput(zsb->z_ctldir); iput(zsb->z_ctldir);
zsb->z_ctldir = NULL; zsb->z_ctldir = NULL;
@ -897,13 +897,13 @@ zfsctl_snapdir_rename(struct inode *sdip, char *snm,
goto out; goto out;
} }
mutex_enter(&zfs_snapshot_lock); rw_enter(&zfs_snapshot_lock, RW_WRITER);
error = dsl_dataset_rename_snapshot(fsname, snm, tnm, B_FALSE); error = dsl_dataset_rename_snapshot(fsname, snm, tnm, B_FALSE);
if (error == 0) if (error == 0)
(void) zfsctl_snapshot_rename(snm, tnm); (void) zfsctl_snapshot_rename(snm, tnm);
mutex_exit(&zfs_snapshot_lock); rw_exit(&zfs_snapshot_lock);
out: out:
kmem_free(from, MAXNAMELEN); kmem_free(from, MAXNAMELEN);
kmem_free(to, MAXNAMELEN); kmem_free(to, MAXNAMELEN);
@ -1024,12 +1024,12 @@ zfsctl_snapshot_unmount(char *snapname, int flags)
zfs_snapentry_t *se; zfs_snapentry_t *se;
int error; int error;
mutex_enter(&zfs_snapshot_lock); rw_enter(&zfs_snapshot_lock, RW_READER);
if ((se = zfsctl_snapshot_find_by_name(snapname)) == NULL) { if ((se = zfsctl_snapshot_find_by_name(snapname)) == NULL) {
mutex_exit(&zfs_snapshot_lock); rw_exit(&zfs_snapshot_lock);
return (ENOENT); return (ENOENT);
} }
mutex_exit(&zfs_snapshot_lock); rw_exit(&zfs_snapshot_lock);
argv[2] = kmem_asprintf(SET_UNMOUNT_CMD, argv[2] = kmem_asprintf(SET_UNMOUNT_CMD,
flags & MNT_FORCE ? "-f " : "", se->se_path); flags & MNT_FORCE ? "-f " : "", se->se_path);
@ -1145,13 +1145,13 @@ zfsctl_snapshot_mount(struct path *path, int flags)
dentry = spath.dentry; dentry = spath.dentry;
spath.mnt->mnt_flags |= MNT_SHRINKABLE; spath.mnt->mnt_flags |= MNT_SHRINKABLE;
mutex_enter(&zfs_snapshot_lock); rw_enter(&zfs_snapshot_lock, RW_WRITER);
se = zfsctl_snapshot_alloc(full_name, full_path, se = zfsctl_snapshot_alloc(full_name, full_path,
snap_zsb->z_os->os_spa, dmu_objset_id(snap_zsb->z_os), snap_zsb->z_os->os_spa, dmu_objset_id(snap_zsb->z_os),
dentry); dentry);
zfsctl_snapshot_add(se); zfsctl_snapshot_add(se);
zfsctl_snapshot_unmount_delay_impl(se, zfs_expire_snapshot); zfsctl_snapshot_unmount_delay_impl(se, zfs_expire_snapshot);
mutex_exit(&zfs_snapshot_lock); rw_exit(&zfs_snapshot_lock);
} }
path_put(&spath); path_put(&spath);
error: error:
@ -1180,7 +1180,7 @@ zfsctl_lookup_objset(struct super_block *sb, uint64_t objsetid, zfs_sb_t **zsbp)
* process. This race cannot occur to an expired mount point * process. This race cannot occur to an expired mount point
* because we hold the zfs_snapshot_lock to prevent the race. * because we hold the zfs_snapshot_lock to prevent the race.
*/ */
mutex_enter(&zfs_snapshot_lock); rw_enter(&zfs_snapshot_lock, RW_READER);
if ((se = zfsctl_snapshot_find_by_objsetid(spa, objsetid)) != NULL) { if ((se = zfsctl_snapshot_find_by_objsetid(spa, objsetid)) != NULL) {
zfs_sb_t *zsb; zfs_sb_t *zsb;
@ -1200,7 +1200,7 @@ zfsctl_lookup_objset(struct super_block *sb, uint64_t objsetid, zfs_sb_t **zsbp)
} else { } else {
error = SET_ERROR(ENOENT); error = SET_ERROR(ENOENT);
} }
mutex_exit(&zfs_snapshot_lock); rw_exit(&zfs_snapshot_lock);
/* /*
* Automount the snapshot given the objset id by constructing the * Automount the snapshot given the objset id by constructing the
@ -1274,7 +1274,7 @@ zfsctl_init(void)
avl_create(&zfs_snapshots_by_objsetid, snapentry_compare_by_objsetid, avl_create(&zfs_snapshots_by_objsetid, snapentry_compare_by_objsetid,
sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t,
se_node_objsetid)); se_node_objsetid));
mutex_init(&zfs_snapshot_lock, NULL, MUTEX_DEFAULT, NULL); rw_init(&zfs_snapshot_lock, NULL, RW_DEFAULT, NULL);
zfs_expire_taskq = taskq_create("z_unmount", 1, defclsyspri, zfs_expire_taskq = taskq_create("z_unmount", 1, defclsyspri,
1, 8, TASKQ_PREPOPULATE); 1, 8, TASKQ_PREPOPULATE);
@ -1291,7 +1291,7 @@ zfsctl_fini(void)
avl_destroy(&zfs_snapshots_by_name); avl_destroy(&zfs_snapshots_by_name);
avl_destroy(&zfs_snapshots_by_objsetid); avl_destroy(&zfs_snapshots_by_objsetid);
mutex_destroy(&zfs_snapshot_lock); rw_destroy(&zfs_snapshot_lock);
} }
module_param(zfs_admin_snapshot, int, 0644); module_param(zfs_admin_snapshot, int, 0644);