Use system_delay_taskq for long delay tasks

Use it for spa_deadman, zpl_posix_acl_free, snapentry_expire.
This free system_taskq from the above long delay tasks, and allow us to do
taskq_wait_outstanding on system_taskq without being blocked forever, making
system_taskq more generic and useful.

Signed-off-by: Chunwei Chen <[email protected]>

7 files changed, 18 insertions, 21 deletions

diff --git a/include/sys/zfs_context.h b/include/sys/zfs_context.h
index ed0a97c45..a1abe20da 100644
--- a/include/sys/zfs_context.h
+++ b/include/sys/zfs_context.h
@@ -498,6 +498,7 @@ typedef struct taskq {
 #define	TASKQID_INVALID		((taskqid_t)0)
 
 extern taskq_t *system_taskq;
+extern taskq_t *system_delay_taskq;
 
 extern taskq_t	*taskq_create(const char *, int, pri_t, int, int, uint_t);
 #define	taskq_create_proc(a, b, c, d, e, p, f) \
diff --git a/lib/libzpool/taskq.c b/lib/libzpool/taskq.c
index c1f87e173..791d50981 100644
--- a/lib/libzpool/taskq.c
+++ b/lib/libzpool/taskq.c
@@ -32,6 +32,7 @@
 
 int taskq_now;
 taskq_t *system_taskq;
+taskq_t *system_delay_taskq;
 
 #define	TASKQ_ACTIVE	0x00010000
 
@@ -353,6 +354,8 @@ system_taskq_init(void)
 {
 	system_taskq = taskq_create("system_taskq", 64, maxclsyspri, 4, 512,
 	    TASKQ_DYNAMIC | TASKQ_PREPOPULATE);
+	system_delay_taskq = taskq_create("delay_taskq", 4, maxclsyspri, 4,
+	    512, TASKQ_DYNAMIC | TASKQ_PREPOPULATE);
 }
 
 void
@@ -360,4 +363,6 @@ system_taskq_fini(void)
 {
 	taskq_destroy(system_taskq);
 	system_taskq = NULL; /* defensive */
+	taskq_destroy(system_delay_taskq);
+	system_delay_taskq = NULL;
 }
diff --git a/module/zfs/spa.c b/module/zfs/spa.c
index 05e15a2e6..9f652dc32 100644
--- a/module/zfs/spa.c
+++ b/module/zfs/spa.c
@@ -1208,7 +1208,7 @@ spa_deactivate(spa_t *spa)
 	list_destroy(&spa->spa_evicting_os_list);
 	list_destroy(&spa->spa_state_dirty_list);
 
-	taskq_cancel_id(system_taskq, spa->spa_deadman_tqid);
+	taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid);
 
 	for (t = 0; t < ZIO_TYPES; t++) {
 		for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
@@ -6517,8 +6517,8 @@ spa_sync(spa_t *spa, uint64_t txg)
 	tx = dmu_tx_create_assigned(dp, txg);
 
 	spa->spa_sync_starttime = gethrtime();
-	taskq_cancel_id(system_taskq, spa->spa_deadman_tqid);
-	spa->spa_deadman_tqid = taskq_dispatch_delay(system_taskq,
+	taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid);
+	spa->spa_deadman_tqid = taskq_dispatch_delay(system_delay_taskq,
 	    spa_deadman, spa, TQ_SLEEP, ddi_get_lbolt() +
 	    NSEC_TO_TICK(spa->spa_deadman_synctime));
 
@@ -6706,7 +6706,7 @@ spa_sync(spa_t *spa, uint64_t txg)
 	}
 	dmu_tx_commit(tx);
 
-	taskq_cancel_id(system_taskq, spa->spa_deadman_tqid);
+	taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid);
 	spa->spa_deadman_tqid = 0;
 
 	/*
diff --git a/module/zfs/spa_misc.c b/module/zfs/spa_misc.c
index 6ec05214e..42bd25980 100644
--- a/module/zfs/spa_misc.c
+++ b/module/zfs/spa_misc.c
@@ -530,7 +530,7 @@ spa_deadman(void *arg)
 	if (zfs_deadman_enabled)
 		vdev_deadman(spa->spa_root_vdev);
 
-	spa->spa_deadman_tqid = taskq_dispatch_delay(system_taskq,
+	spa->spa_deadman_tqid = taskq_dispatch_delay(system_delay_taskq,
 	    spa_deadman, spa, TQ_SLEEP, ddi_get_lbolt() +
 	    NSEC_TO_TICK(spa->spa_deadman_synctime));
 }
diff --git a/module/zfs/zfs_ctldir.c b/module/zfs/zfs_ctldir.c
index c7a93edfc..53674d975 100644
--- a/module/zfs/zfs_ctldir.c
+++ b/module/zfs/zfs_ctldir.c
@@ -111,11 +111,6 @@ static krwlock_t zfs_snapshot_lock;
 int zfs_expire_snapshot = ZFSCTL_EXPIRE_SNAPSHOT;
 int zfs_admin_snapshot = 1;
 
-/*
- * Dedicated task queue for unmounting snapshots.
- */
-static taskq_t *zfs_expire_taskq;
-
 typedef struct {
 	char		*se_name;	/* full snapshot name */
 	char		*se_path;	/* full mount path */
@@ -365,7 +360,7 @@ zfsctl_snapshot_unmount_cancel(zfs_snapentry_t *se)
 {
 	ASSERT(RW_LOCK_HELD(&zfs_snapshot_lock));
 
-	if (taskq_cancel_id(zfs_expire_taskq, se->se_taskqid) == 0) {
+	if (taskq_cancel_id(system_delay_taskq, se->se_taskqid) == 0) {
 		se->se_taskqid = TASKQID_INVALID;
 		zfsctl_snapshot_rele(se);
 	}
@@ -383,7 +378,7 @@ zfsctl_snapshot_unmount_delay_impl(zfs_snapentry_t *se, int delay)
 		return;
 
 	zfsctl_snapshot_hold(se);
-	se->se_taskqid = taskq_dispatch_delay(zfs_expire_taskq,
+	se->se_taskqid = taskq_dispatch_delay(system_delay_taskq,
 	    snapentry_expire, se, TQ_SLEEP, ddi_get_lbolt() + delay * HZ);
 }
 
@@ -1257,9 +1252,6 @@ zfsctl_init(void)
 	    sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t,
 	    se_node_objsetid));
 	rw_init(&zfs_snapshot_lock, NULL, RW_DEFAULT, NULL);
-
-	zfs_expire_taskq = taskq_create("z_unmount", 1, defclsyspri,
-	    1, 8, TASKQ_PREPOPULATE);
 }
 
 /*
@@ -1269,8 +1261,6 @@ zfsctl_init(void)
 void
 zfsctl_fini(void)
 {
-	taskq_destroy(zfs_expire_taskq);
-
 	avl_destroy(&zfs_snapshots_by_name);
 	avl_destroy(&zfs_snapshots_by_objsetid);
 	rw_destroy(&zfs_snapshot_lock);
diff --git a/module/zfs/zfs_vfsops.c b/module/zfs/zfs_vfsops.c
index 5417f2422..39e92ce21 100644
--- a/module/zfs/zfs_vfsops.c
+++ b/module/zfs/zfs_vfsops.c
@@ -1922,6 +1922,7 @@ zfs_fini(void)
 	/*
 	 * we don't use outstanding because zpl_posix_acl_free might add more.
 	 */
+	taskq_wait(system_delay_taskq);
 	taskq_wait(system_taskq);
 	unregister_filesystem(&zpl_fs_type);
 	zfs_znode_fini();
diff --git a/module/zfs/zpl_xattr.c b/module/zfs/zpl_xattr.c
index cec870824..9ab27f1c2 100644
--- a/module/zfs/zpl_xattr.c
+++ b/module/zfs/zpl_xattr.c
@@ -1511,8 +1511,8 @@ zpl_posix_acl_free(void *arg)
 	}
 
 	if (refire)
-		taskq_dispatch_delay(system_taskq, zpl_posix_acl_free, NULL,
-		    TQ_SLEEP, new_time);
+		taskq_dispatch_delay(system_delay_taskq, zpl_posix_acl_free,
+		    NULL, TQ_SLEEP, new_time);
 
 	while (freelist) {
 		a = freelist;
@@ -1537,7 +1537,7 @@ zpl_posix_acl_release_impl(struct posix_acl *acl)
 	*prev = a;
 	/* if it was empty before, schedule the free task */
 	if (prev == &acl_rel_head)
-		taskq_dispatch_delay(system_taskq, zpl_posix_acl_free, NULL,
-		    TQ_SLEEP, ddi_get_lbolt() + ACL_REL_SCHED);
+		taskq_dispatch_delay(system_delay_taskq, zpl_posix_acl_free,
+		    NULL, TQ_SLEEP, ddi_get_lbolt() + ACL_REL_SCHED);
 }
 #endif