4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
25 #include <sys/zfs_context.h>
26 #include <sys/spa_impl.h>
28 #include <sys/zio_checksum.h>
29 #include <sys/zio_compress.h>
31 #include <sys/dmu_tx.h>
34 #include <sys/vdev_impl.h>
35 #include <sys/metaslab.h>
36 #include <sys/uberblock_impl.h>
39 #include <sys/unique.h>
40 #include <sys/dsl_pool.h>
41 #include <sys/dsl_dir.h>
42 #include <sys/dsl_prop.h>
43 #include <sys/fm/util.h>
44 #include <sys/dsl_scan.h>
45 #include <sys/fs/zfs.h>
46 #include <sys/metaslab_impl.h>
54 * There are four basic locks for managing spa_t structures:
56 * spa_namespace_lock (global mutex)
58 * This lock must be acquired to do any of the following:
60 * - Lookup a spa_t by name
61 * - Add or remove a spa_t from the namespace
62 * - Increase spa_refcount from non-zero
63 * - Check if spa_refcount is zero
65 * - add/remove/attach/detach devices
66 * - Held for the duration of create/destroy/import/export
68 * It does not need to handle recursion. A create or destroy may
69 * reference objects (files or zvols) in other pools, but by
70 * definition they must have an existing reference, and will never need
71 * to lookup a spa_t by name.
73 * spa_refcount (per-spa refcount_t protected by mutex)
75 * This reference count keep track of any active users of the spa_t. The
76 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
77 * the refcount is never really 'zero' - opening a pool implicitly keeps
78 * some references in the DMU. Internally we check against spa_minref, but
79 * present the image of a zero/non-zero value to consumers.
81 * spa_config_lock[] (per-spa array of rwlocks)
83 * This protects the spa_t from config changes, and must be held in
84 * the following circumstances:
86 * - RW_READER to perform I/O to the spa
87 * - RW_WRITER to change the vdev config
89 * The locking order is fairly straightforward:
91 * spa_namespace_lock -> spa_refcount
93 * The namespace lock must be acquired to increase the refcount from 0
94 * or to check if it is zero.
96 * spa_refcount -> spa_config_lock[]
98 * There must be at least one valid reference on the spa_t to acquire
101 * spa_namespace_lock -> spa_config_lock[]
103 * The namespace lock must always be taken before the config lock.
106 * The spa_namespace_lock can be acquired directly and is globally visible.
108 * The namespace is manipulated using the following functions, all of which
109 * require the spa_namespace_lock to be held.
111 * spa_lookup() Lookup a spa_t by name.
113 * spa_add() Create a new spa_t in the namespace.
115 * spa_remove() Remove a spa_t from the namespace. This also
116 * frees up any memory associated with the spa_t.
118 * spa_next() Returns the next spa_t in the system, or the
119 * first if NULL is passed.
121 * spa_evict_all() Shutdown and remove all spa_t structures in
124 * spa_guid_exists() Determine whether a pool/device guid exists.
126 * The spa_refcount is manipulated using the following functions:
128 * spa_open_ref() Adds a reference to the given spa_t. Must be
129 * called with spa_namespace_lock held if the
130 * refcount is currently zero.
132 * spa_close() Remove a reference from the spa_t. This will
133 * not free the spa_t or remove it from the
134 * namespace. No locking is required.
136 * spa_refcount_zero() Returns true if the refcount is currently
137 * zero. Must be called with spa_namespace_lock
140 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
141 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
142 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
144 * To read the configuration, it suffices to hold one of these locks as reader.
145 * To modify the configuration, you must hold all locks as writer. To modify
146 * vdev state without altering the vdev tree's topology (e.g. online/offline),
147 * you must hold SCL_STATE and SCL_ZIO as writer.
149 * We use these distinct config locks to avoid recursive lock entry.
150 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
151 * block allocations (SCL_ALLOC), which may require reading space maps
152 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
154 * The spa config locks cannot be normal rwlocks because we need the
155 * ability to hand off ownership. For example, SCL_ZIO is acquired
156 * by the issuing thread and later released by an interrupt thread.
157 * They do, however, obey the usual write-wanted semantics to prevent
158 * writer (i.e. system administrator) starvation.
160 * The lock acquisition rules are as follows:
163 * Protects changes to the vdev tree topology, such as vdev
164 * add/remove/attach/detach. Protects the dirty config list
165 * (spa_config_dirty_list) and the set of spares and l2arc devices.
168 * Protects changes to pool state and vdev state, such as vdev
169 * online/offline/fault/degrade/clear. Protects the dirty state list
170 * (spa_state_dirty_list) and global pool state (spa_state).
173 * Protects changes to metaslab groups and classes.
174 * Held as reader by metaslab_alloc() and metaslab_claim().
177 * Held by bp-level zios (those which have no io_vd upon entry)
178 * to prevent changes to the vdev tree. The bp-level zio implicitly
179 * protects all of its vdev child zios, which do not hold SCL_ZIO.
182 * Protects changes to metaslab groups and classes.
183 * Held as reader by metaslab_free(). SCL_FREE is distinct from
184 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
185 * blocks in zio_done() while another i/o that holds either
186 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
189 * Held as reader to prevent changes to the vdev tree during trivial
190 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
191 * other locks, and lower than all of them, to ensure that it's safe
192 * to acquire regardless of caller context.
194 * In addition, the following rules apply:
196 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
197 * The lock ordering is SCL_CONFIG > spa_props_lock.
199 * (b) I/O operations on leaf vdevs. For any zio operation that takes
200 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
201 * or zio_write_phys() -- the caller must ensure that the config cannot
202 * cannot change in the interim, and that the vdev cannot be reopened.
203 * SCL_STATE as reader suffices for both.
205 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
207 * spa_vdev_enter() Acquire the namespace lock and the config lock
210 * spa_vdev_exit() Release the config lock, wait for all I/O
211 * to complete, sync the updated configs to the
212 * cache, and release the namespace lock.
214 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
215 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
216 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
218 * spa_rename() is also implemented within this file since is requires
219 * manipulation of the namespace.
222 static avl_tree_t spa_namespace_avl;
223 kmutex_t spa_namespace_lock;
224 static kcondvar_t spa_namespace_cv;
225 static int spa_active_count;
226 int spa_max_replication_override = SPA_DVAS_PER_BP;
228 static kmutex_t spa_spare_lock;
229 static avl_tree_t spa_spare_avl;
230 static kmutex_t spa_l2cache_lock;
231 static avl_tree_t spa_l2cache_avl;
233 kmem_cache_t *spa_buffer_pool;
237 /* Everything except dprintf is on by default in debug builds */
238 int zfs_flags = ~ZFS_DEBUG_DPRINTF;
244 * zfs_recover can be set to nonzero to attempt to recover from
245 * otherwise-fatal errors, typically caused by on-disk corruption. When
246 * set, calls to zfs_panic_recover() will turn into warning messages.
252 * ==========================================================================
254 * ==========================================================================
257 spa_config_lock_init(spa_t *spa)
261 for (i = 0; i < SCL_LOCKS; i++) {
262 spa_config_lock_t *scl = &spa->spa_config_lock[i];
263 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
264 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
265 refcount_create(&scl->scl_count);
266 scl->scl_writer = NULL;
267 scl->scl_write_wanted = 0;
272 spa_config_lock_destroy(spa_t *spa)
276 for (i = 0; i < SCL_LOCKS; i++) {
277 spa_config_lock_t *scl = &spa->spa_config_lock[i];
278 mutex_destroy(&scl->scl_lock);
279 cv_destroy(&scl->scl_cv);
280 refcount_destroy(&scl->scl_count);
281 ASSERT(scl->scl_writer == NULL);
282 ASSERT(scl->scl_write_wanted == 0);
287 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
291 for (i = 0; i < SCL_LOCKS; i++) {
292 spa_config_lock_t *scl = &spa->spa_config_lock[i];
293 if (!(locks & (1 << i)))
295 mutex_enter(&scl->scl_lock);
296 if (rw == RW_READER) {
297 if (scl->scl_writer || scl->scl_write_wanted) {
298 mutex_exit(&scl->scl_lock);
299 spa_config_exit(spa, locks ^ (1 << i), tag);
303 ASSERT(scl->scl_writer != curthread);
304 if (!refcount_is_zero(&scl->scl_count)) {
305 mutex_exit(&scl->scl_lock);
306 spa_config_exit(spa, locks ^ (1 << i), tag);
309 scl->scl_writer = curthread;
311 (void) refcount_add(&scl->scl_count, tag);
312 mutex_exit(&scl->scl_lock);
318 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
323 for (i = 0; i < SCL_LOCKS; i++) {
324 spa_config_lock_t *scl = &spa->spa_config_lock[i];
325 if (scl->scl_writer == curthread)
326 wlocks_held |= (1 << i);
327 if (!(locks & (1 << i)))
329 mutex_enter(&scl->scl_lock);
330 if (rw == RW_READER) {
331 while (scl->scl_writer || scl->scl_write_wanted) {
332 cv_wait(&scl->scl_cv, &scl->scl_lock);
335 ASSERT(scl->scl_writer != curthread);
336 while (!refcount_is_zero(&scl->scl_count)) {
337 scl->scl_write_wanted++;
338 cv_wait(&scl->scl_cv, &scl->scl_lock);
339 scl->scl_write_wanted--;
341 scl->scl_writer = curthread;
343 (void) refcount_add(&scl->scl_count, tag);
344 mutex_exit(&scl->scl_lock);
346 ASSERT(wlocks_held <= locks);
350 spa_config_exit(spa_t *spa, int locks, void *tag)
354 for (i = SCL_LOCKS - 1; i >= 0; i--) {
355 spa_config_lock_t *scl = &spa->spa_config_lock[i];
356 if (!(locks & (1 << i)))
358 mutex_enter(&scl->scl_lock);
359 ASSERT(!refcount_is_zero(&scl->scl_count));
360 if (refcount_remove(&scl->scl_count, tag) == 0) {
361 ASSERT(scl->scl_writer == NULL ||
362 scl->scl_writer == curthread);
363 scl->scl_writer = NULL; /* OK in either case */
364 cv_broadcast(&scl->scl_cv);
366 mutex_exit(&scl->scl_lock);
371 spa_config_held(spa_t *spa, int locks, krw_t rw)
373 int i, locks_held = 0;
375 for (i = 0; i < SCL_LOCKS; i++) {
376 spa_config_lock_t *scl = &spa->spa_config_lock[i];
377 if (!(locks & (1 << i)))
379 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
380 (rw == RW_WRITER && scl->scl_writer == curthread))
381 locks_held |= 1 << i;
388 * ==========================================================================
389 * SPA namespace functions
390 * ==========================================================================
394 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
395 * Returns NULL if no matching spa_t is found.
398 spa_lookup(const char *name)
400 static spa_t search; /* spa_t is large; don't allocate on stack */
406 ASSERT(MUTEX_HELD(&spa_namespace_lock));
409 * If it's a full dataset name, figure out the pool name and
412 cp = strpbrk(name, "/@");
418 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
419 spa = avl_find(&spa_namespace_avl, &search, &where);
428 * Create an uninitialized spa_t with the given name. Requires
429 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
430 * exist by calling spa_lookup() first.
433 spa_add(const char *name, nvlist_t *config, const char *altroot)
436 spa_config_dirent_t *dp;
439 ASSERT(MUTEX_HELD(&spa_namespace_lock));
441 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP | KM_NODEBUG);
443 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
444 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
445 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
446 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
447 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
448 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
449 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
450 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
451 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
453 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
454 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
455 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
456 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
458 for (t = 0; t < TXG_SIZE; t++)
459 bplist_create(&spa->spa_free_bplist[t]);
461 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
462 spa->spa_state = POOL_STATE_UNINITIALIZED;
463 spa->spa_freeze_txg = UINT64_MAX;
464 spa->spa_final_txg = UINT64_MAX;
465 spa->spa_load_max_txg = UINT64_MAX;
467 spa->spa_proc_state = SPA_PROC_NONE;
469 refcount_create(&spa->spa_refcount);
470 spa_config_lock_init(spa);
472 avl_add(&spa_namespace_avl, spa);
475 * Set the alternate root, if there is one.
478 spa->spa_root = spa_strdup(altroot);
483 * Every pool starts with the default cachefile
485 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
486 offsetof(spa_config_dirent_t, scd_link));
488 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
489 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
490 list_insert_head(&spa->spa_config_list, dp);
492 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
496 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
502 * Removes a spa_t from the namespace, freeing up any memory used. Requires
503 * spa_namespace_lock. This is called only after the spa_t has been closed and
507 spa_remove(spa_t *spa)
509 spa_config_dirent_t *dp;
512 ASSERT(MUTEX_HELD(&spa_namespace_lock));
513 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
515 nvlist_free(spa->spa_config_splitting);
517 avl_remove(&spa_namespace_avl, spa);
518 cv_broadcast(&spa_namespace_cv);
521 spa_strfree(spa->spa_root);
525 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
526 list_remove(&spa->spa_config_list, dp);
527 if (dp->scd_path != NULL)
528 spa_strfree(dp->scd_path);
529 kmem_free(dp, sizeof (spa_config_dirent_t));
532 list_destroy(&spa->spa_config_list);
534 nvlist_free(spa->spa_load_info);
535 spa_config_set(spa, NULL);
537 refcount_destroy(&spa->spa_refcount);
539 spa_config_lock_destroy(spa);
541 for (t = 0; t < TXG_SIZE; t++)
542 bplist_destroy(&spa->spa_free_bplist[t]);
544 cv_destroy(&spa->spa_async_cv);
545 cv_destroy(&spa->spa_proc_cv);
546 cv_destroy(&spa->spa_scrub_io_cv);
547 cv_destroy(&spa->spa_suspend_cv);
549 mutex_destroy(&spa->spa_async_lock);
550 mutex_destroy(&spa->spa_errlist_lock);
551 mutex_destroy(&spa->spa_errlog_lock);
552 mutex_destroy(&spa->spa_history_lock);
553 mutex_destroy(&spa->spa_proc_lock);
554 mutex_destroy(&spa->spa_props_lock);
555 mutex_destroy(&spa->spa_scrub_lock);
556 mutex_destroy(&spa->spa_suspend_lock);
557 mutex_destroy(&spa->spa_vdev_top_lock);
559 kmem_free(spa, sizeof (spa_t));
563 * Given a pool, return the next pool in the namespace, or NULL if there is
564 * none. If 'prev' is NULL, return the first pool.
567 spa_next(spa_t *prev)
569 ASSERT(MUTEX_HELD(&spa_namespace_lock));
572 return (AVL_NEXT(&spa_namespace_avl, prev));
574 return (avl_first(&spa_namespace_avl));
578 * ==========================================================================
579 * SPA refcount functions
580 * ==========================================================================
584 * Add a reference to the given spa_t. Must have at least one reference, or
585 * have the namespace lock held.
588 spa_open_ref(spa_t *spa, void *tag)
590 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
591 MUTEX_HELD(&spa_namespace_lock));
592 (void) refcount_add(&spa->spa_refcount, tag);
596 * Remove a reference to the given spa_t. Must have at least one reference, or
597 * have the namespace lock held.
600 spa_close(spa_t *spa, void *tag)
602 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
603 MUTEX_HELD(&spa_namespace_lock));
604 (void) refcount_remove(&spa->spa_refcount, tag);
608 * Check to see if the spa refcount is zero. Must be called with
609 * spa_namespace_lock held. We really compare against spa_minref, which is the
610 * number of references acquired when opening a pool
613 spa_refcount_zero(spa_t *spa)
615 ASSERT(MUTEX_HELD(&spa_namespace_lock));
617 return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
621 * ==========================================================================
622 * SPA spare and l2cache tracking
623 * ==========================================================================
627 * Hot spares and cache devices are tracked using the same code below,
628 * for 'auxiliary' devices.
631 typedef struct spa_aux {
639 spa_aux_compare(const void *a, const void *b)
641 const spa_aux_t *sa = a;
642 const spa_aux_t *sb = b;
644 if (sa->aux_guid < sb->aux_guid)
646 else if (sa->aux_guid > sb->aux_guid)
653 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
659 search.aux_guid = vd->vdev_guid;
660 if ((aux = avl_find(avl, &search, &where)) != NULL) {
663 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
664 aux->aux_guid = vd->vdev_guid;
666 avl_insert(avl, aux, where);
671 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
677 search.aux_guid = vd->vdev_guid;
678 aux = avl_find(avl, &search, &where);
682 if (--aux->aux_count == 0) {
683 avl_remove(avl, aux);
684 kmem_free(aux, sizeof (spa_aux_t));
685 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
686 aux->aux_pool = 0ULL;
691 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
693 spa_aux_t search, *found;
695 search.aux_guid = guid;
696 found = avl_find(avl, &search, NULL);
700 *pool = found->aux_pool;
707 *refcnt = found->aux_count;
712 return (found != NULL);
716 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
718 spa_aux_t search, *found;
721 search.aux_guid = vd->vdev_guid;
722 found = avl_find(avl, &search, &where);
723 ASSERT(found != NULL);
724 ASSERT(found->aux_pool == 0ULL);
726 found->aux_pool = spa_guid(vd->vdev_spa);
730 * Spares are tracked globally due to the following constraints:
732 * - A spare may be part of multiple pools.
733 * - A spare may be added to a pool even if it's actively in use within
735 * - A spare in use in any pool can only be the source of a replacement if
736 * the target is a spare in the same pool.
738 * We keep track of all spares on the system through the use of a reference
739 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
740 * spare, then we bump the reference count in the AVL tree. In addition, we set
741 * the 'vdev_isspare' member to indicate that the device is a spare (active or
742 * inactive). When a spare is made active (used to replace a device in the
743 * pool), we also keep track of which pool its been made a part of.
745 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
746 * called under the spa_namespace lock as part of vdev reconfiguration. The
747 * separate spare lock exists for the status query path, which does not need to
748 * be completely consistent with respect to other vdev configuration changes.
752 spa_spare_compare(const void *a, const void *b)
754 return (spa_aux_compare(a, b));
758 spa_spare_add(vdev_t *vd)
760 mutex_enter(&spa_spare_lock);
761 ASSERT(!vd->vdev_isspare);
762 spa_aux_add(vd, &spa_spare_avl);
763 vd->vdev_isspare = B_TRUE;
764 mutex_exit(&spa_spare_lock);
768 spa_spare_remove(vdev_t *vd)
770 mutex_enter(&spa_spare_lock);
771 ASSERT(vd->vdev_isspare);
772 spa_aux_remove(vd, &spa_spare_avl);
773 vd->vdev_isspare = B_FALSE;
774 mutex_exit(&spa_spare_lock);
778 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
782 mutex_enter(&spa_spare_lock);
783 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
784 mutex_exit(&spa_spare_lock);
790 spa_spare_activate(vdev_t *vd)
792 mutex_enter(&spa_spare_lock);
793 ASSERT(vd->vdev_isspare);
794 spa_aux_activate(vd, &spa_spare_avl);
795 mutex_exit(&spa_spare_lock);
799 * Level 2 ARC devices are tracked globally for the same reasons as spares.
800 * Cache devices currently only support one pool per cache device, and so
801 * for these devices the aux reference count is currently unused beyond 1.
805 spa_l2cache_compare(const void *a, const void *b)
807 return (spa_aux_compare(a, b));
811 spa_l2cache_add(vdev_t *vd)
813 mutex_enter(&spa_l2cache_lock);
814 ASSERT(!vd->vdev_isl2cache);
815 spa_aux_add(vd, &spa_l2cache_avl);
816 vd->vdev_isl2cache = B_TRUE;
817 mutex_exit(&spa_l2cache_lock);
821 spa_l2cache_remove(vdev_t *vd)
823 mutex_enter(&spa_l2cache_lock);
824 ASSERT(vd->vdev_isl2cache);
825 spa_aux_remove(vd, &spa_l2cache_avl);
826 vd->vdev_isl2cache = B_FALSE;
827 mutex_exit(&spa_l2cache_lock);
831 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
835 mutex_enter(&spa_l2cache_lock);
836 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
837 mutex_exit(&spa_l2cache_lock);
843 spa_l2cache_activate(vdev_t *vd)
845 mutex_enter(&spa_l2cache_lock);
846 ASSERT(vd->vdev_isl2cache);
847 spa_aux_activate(vd, &spa_l2cache_avl);
848 mutex_exit(&spa_l2cache_lock);
852 * ==========================================================================
854 * ==========================================================================
858 * Lock the given spa_t for the purpose of adding or removing a vdev.
859 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
860 * It returns the next transaction group for the spa_t.
863 spa_vdev_enter(spa_t *spa)
865 mutex_enter(&spa->spa_vdev_top_lock);
866 mutex_enter(&spa_namespace_lock);
867 return (spa_vdev_config_enter(spa));
871 * Internal implementation for spa_vdev_enter(). Used when a vdev
872 * operation requires multiple syncs (i.e. removing a device) while
873 * keeping the spa_namespace_lock held.
876 spa_vdev_config_enter(spa_t *spa)
878 ASSERT(MUTEX_HELD(&spa_namespace_lock));
880 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
882 return (spa_last_synced_txg(spa) + 1);
886 * Used in combination with spa_vdev_config_enter() to allow the syncing
887 * of multiple transactions without releasing the spa_namespace_lock.
890 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
892 int config_changed = B_FALSE;
894 ASSERT(MUTEX_HELD(&spa_namespace_lock));
895 ASSERT(txg > spa_last_synced_txg(spa));
897 spa->spa_pending_vdev = NULL;
902 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
904 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
905 config_changed = B_TRUE;
906 spa->spa_config_generation++;
910 * Verify the metaslab classes.
912 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
913 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
915 spa_config_exit(spa, SCL_ALL, spa);
918 * Panic the system if the specified tag requires it. This
919 * is useful for ensuring that configurations are updated
922 if (zio_injection_enabled)
923 zio_handle_panic_injection(spa, tag, 0);
926 * Note: this txg_wait_synced() is important because it ensures
927 * that there won't be more than one config change per txg.
928 * This allows us to use the txg as the generation number.
931 txg_wait_synced(spa->spa_dsl_pool, txg);
934 ASSERT(!vd->vdev_detached || vd->vdev_dtl_smo.smo_object == 0);
935 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
937 spa_config_exit(spa, SCL_ALL, spa);
941 * If the config changed, update the config cache.
944 spa_config_sync(spa, B_FALSE, B_TRUE);
948 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
949 * locking of spa_vdev_enter(), we also want make sure the transactions have
950 * synced to disk, and then update the global configuration cache with the new
954 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
956 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
957 mutex_exit(&spa_namespace_lock);
958 mutex_exit(&spa->spa_vdev_top_lock);
964 * Lock the given spa_t for the purpose of changing vdev state.
967 spa_vdev_state_enter(spa_t *spa, int oplocks)
969 int locks = SCL_STATE_ALL | oplocks;
972 * Root pools may need to read of the underlying devfs filesystem
973 * when opening up a vdev. Unfortunately if we're holding the
974 * SCL_ZIO lock it will result in a deadlock when we try to issue
975 * the read from the root filesystem. Instead we "prefetch"
976 * the associated vnodes that we need prior to opening the
977 * underlying devices and cache them so that we can prevent
978 * any I/O when we are doing the actual open.
980 if (spa_is_root(spa)) {
981 int low = locks & ~(SCL_ZIO - 1);
982 int high = locks & ~low;
984 spa_config_enter(spa, high, spa, RW_WRITER);
985 vdev_hold(spa->spa_root_vdev);
986 spa_config_enter(spa, low, spa, RW_WRITER);
988 spa_config_enter(spa, locks, spa, RW_WRITER);
990 spa->spa_vdev_locks = locks;
994 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
996 boolean_t config_changed = B_FALSE;
998 if (vd != NULL || error == 0)
999 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1003 vdev_state_dirty(vd->vdev_top);
1004 config_changed = B_TRUE;
1005 spa->spa_config_generation++;
1008 if (spa_is_root(spa))
1009 vdev_rele(spa->spa_root_vdev);
1011 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1012 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1015 * If anything changed, wait for it to sync. This ensures that,
1016 * from the system administrator's perspective, zpool(1M) commands
1017 * are synchronous. This is important for things like zpool offline:
1018 * when the command completes, you expect no further I/O from ZFS.
1021 txg_wait_synced(spa->spa_dsl_pool, 0);
1024 * If the config changed, update the config cache.
1026 if (config_changed) {
1027 mutex_enter(&spa_namespace_lock);
1028 spa_config_sync(spa, B_FALSE, B_TRUE);
1029 mutex_exit(&spa_namespace_lock);
1036 * ==========================================================================
1037 * Miscellaneous functions
1038 * ==========================================================================
1045 spa_rename(const char *name, const char *newname)
1051 * Lookup the spa_t and grab the config lock for writing. We need to
1052 * actually open the pool so that we can sync out the necessary labels.
1053 * It's OK to call spa_open() with the namespace lock held because we
1054 * allow recursive calls for other reasons.
1056 mutex_enter(&spa_namespace_lock);
1057 if ((err = spa_open(name, &spa, FTAG)) != 0) {
1058 mutex_exit(&spa_namespace_lock);
1062 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1064 avl_remove(&spa_namespace_avl, spa);
1065 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1066 avl_add(&spa_namespace_avl, spa);
1069 * Sync all labels to disk with the new names by marking the root vdev
1070 * dirty and waiting for it to sync. It will pick up the new pool name
1073 vdev_config_dirty(spa->spa_root_vdev);
1075 spa_config_exit(spa, SCL_ALL, FTAG);
1077 txg_wait_synced(spa->spa_dsl_pool, 0);
1080 * Sync the updated config cache.
1082 spa_config_sync(spa, B_FALSE, B_TRUE);
1084 spa_close(spa, FTAG);
1086 mutex_exit(&spa_namespace_lock);
1092 * Return the spa_t associated with given pool_guid, if it exists. If
1093 * device_guid is non-zero, determine whether the pool exists *and* contains
1094 * a device with the specified device_guid.
1097 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1100 avl_tree_t *t = &spa_namespace_avl;
1102 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1104 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1105 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1107 if (spa->spa_root_vdev == NULL)
1109 if (spa_guid(spa) == pool_guid) {
1110 if (device_guid == 0)
1113 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1114 device_guid) != NULL)
1118 * Check any devices we may be in the process of adding.
1120 if (spa->spa_pending_vdev) {
1121 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1122 device_guid) != NULL)
1132 * Determine whether a pool with the given pool_guid exists.
1135 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1137 return (spa_by_guid(pool_guid, device_guid) != NULL);
1141 spa_strdup(const char *s)
1147 new = kmem_alloc(len + 1, KM_SLEEP);
1155 spa_strfree(char *s)
1157 kmem_free(s, strlen(s) + 1);
1161 spa_get_random(uint64_t range)
1167 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1173 spa_generate_guid(spa_t *spa)
1175 uint64_t guid = spa_get_random(-1ULL);
1178 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1179 guid = spa_get_random(-1ULL);
1181 while (guid == 0 || spa_guid_exists(guid, 0))
1182 guid = spa_get_random(-1ULL);
1189 sprintf_blkptr(char *buf, const blkptr_t *bp)
1192 char *checksum = NULL;
1193 char *compress = NULL;
1196 type = dmu_ot[BP_GET_TYPE(bp)].ot_name;
1197 checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1198 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1201 SPRINTF_BLKPTR(snprintf, ' ', buf, bp, type, checksum, compress);
1205 spa_freeze(spa_t *spa)
1207 uint64_t freeze_txg = 0;
1209 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1210 if (spa->spa_freeze_txg == UINT64_MAX) {
1211 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1212 spa->spa_freeze_txg = freeze_txg;
1214 spa_config_exit(spa, SCL_ALL, FTAG);
1215 if (freeze_txg != 0)
1216 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1220 zfs_panic_recover(const char *fmt, ...)
1225 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1230 * This is a stripped-down version of strtoull, suitable only for converting
1231 * lowercase hexidecimal numbers that don't overflow.
1234 strtonum(const char *str, char **nptr)
1240 while ((c = *str) != '\0') {
1241 if (c >= '0' && c <= '9')
1243 else if (c >= 'a' && c <= 'f')
1244 digit = 10 + c - 'a';
1255 *nptr = (char *)str;
1261 * ==========================================================================
1262 * Accessor functions
1263 * ==========================================================================
1267 spa_shutting_down(spa_t *spa)
1269 return (spa->spa_async_suspended);
1273 spa_get_dsl(spa_t *spa)
1275 return (spa->spa_dsl_pool);
1279 spa_get_rootblkptr(spa_t *spa)
1281 return (&spa->spa_ubsync.ub_rootbp);
1285 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1287 spa->spa_uberblock.ub_rootbp = *bp;
1291 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1293 if (spa->spa_root == NULL)
1296 (void) strncpy(buf, spa->spa_root, buflen);
1300 spa_sync_pass(spa_t *spa)
1302 return (spa->spa_sync_pass);
1306 spa_name(spa_t *spa)
1308 return (spa->spa_name);
1312 spa_guid(spa_t *spa)
1315 * If we fail to parse the config during spa_load(), we can go through
1316 * the error path (which posts an ereport) and end up here with no root
1317 * vdev. We stash the original pool guid in 'spa_load_guid' to handle
1320 if (spa->spa_root_vdev != NULL)
1321 return (spa->spa_root_vdev->vdev_guid);
1323 return (spa->spa_load_guid);
1327 spa_last_synced_txg(spa_t *spa)
1329 return (spa->spa_ubsync.ub_txg);
1333 spa_first_txg(spa_t *spa)
1335 return (spa->spa_first_txg);
1339 spa_syncing_txg(spa_t *spa)
1341 return (spa->spa_syncing_txg);
1345 spa_state(spa_t *spa)
1347 return (spa->spa_state);
1351 spa_load_state(spa_t *spa)
1353 return (spa->spa_load_state);
1357 spa_freeze_txg(spa_t *spa)
1359 return (spa->spa_freeze_txg);
1364 spa_get_asize(spa_t *spa, uint64_t lsize)
1367 * The worst case is single-sector max-parity RAID-Z blocks, in which
1368 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
1369 * times the size; so just assume that. Add to this the fact that
1370 * we can have up to 3 DVAs per bp, and one more factor of 2 because
1371 * the block may be dittoed with up to 3 DVAs by ddt_sync().
1373 return (lsize * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2);
1377 spa_get_dspace(spa_t *spa)
1379 return (spa->spa_dspace);
1383 spa_update_dspace(spa_t *spa)
1385 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1386 ddt_get_dedup_dspace(spa);
1390 * Return the failure mode that has been set to this pool. The default
1391 * behavior will be to block all I/Os when a complete failure occurs.
1394 spa_get_failmode(spa_t *spa)
1396 return (spa->spa_failmode);
1400 spa_suspended(spa_t *spa)
1402 return (spa->spa_suspended);
1406 spa_version(spa_t *spa)
1408 return (spa->spa_ubsync.ub_version);
1412 spa_deflate(spa_t *spa)
1414 return (spa->spa_deflate);
1418 spa_normal_class(spa_t *spa)
1420 return (spa->spa_normal_class);
1424 spa_log_class(spa_t *spa)
1426 return (spa->spa_log_class);
1430 spa_max_replication(spa_t *spa)
1433 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1434 * handle BPs with more than one DVA allocated. Set our max
1435 * replication level accordingly.
1437 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1439 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1443 spa_prev_software_version(spa_t *spa)
1445 return (spa->spa_prev_software_version);
1449 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1451 uint64_t asize = DVA_GET_ASIZE(dva);
1452 uint64_t dsize = asize;
1454 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1456 if (asize != 0 && spa->spa_deflate) {
1457 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1458 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1465 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1470 for (d = 0; d < SPA_DVAS_PER_BP; d++)
1471 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1477 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1482 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1484 for (d = 0; d < SPA_DVAS_PER_BP; d++)
1485 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1487 spa_config_exit(spa, SCL_VDEV, FTAG);
1493 * ==========================================================================
1494 * Initialization and Termination
1495 * ==========================================================================
1499 spa_name_compare(const void *a1, const void *a2)
1501 const spa_t *s1 = a1;
1502 const spa_t *s2 = a2;
1505 s = strcmp(s1->spa_name, s2->spa_name);
1522 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1523 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1524 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1525 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1527 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1528 offsetof(spa_t, spa_avl));
1530 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1531 offsetof(spa_aux_t, aux_avl));
1533 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1534 offsetof(spa_aux_t, aux_avl));
1536 spa_mode_global = mode;
1544 vdev_cache_stat_init();
1558 vdev_cache_stat_fini();
1566 avl_destroy(&spa_namespace_avl);
1567 avl_destroy(&spa_spare_avl);
1568 avl_destroy(&spa_l2cache_avl);
1570 cv_destroy(&spa_namespace_cv);
1571 mutex_destroy(&spa_namespace_lock);
1572 mutex_destroy(&spa_spare_lock);
1573 mutex_destroy(&spa_l2cache_lock);
1577 * Return whether this pool has slogs. No locking needed.
1578 * It's not a problem if the wrong answer is returned as it's only for
1579 * performance and not correctness
1582 spa_has_slogs(spa_t *spa)
1584 return (spa->spa_log_class->mc_rotor != NULL);
1588 spa_get_log_state(spa_t *spa)
1590 return (spa->spa_log_state);
1594 spa_set_log_state(spa_t *spa, spa_log_state_t state)
1596 spa->spa_log_state = state;
1600 spa_is_root(spa_t *spa)
1602 return (spa->spa_is_root);
1606 spa_writeable(spa_t *spa)
1608 return (!!(spa->spa_mode & FWRITE));
1612 spa_mode(spa_t *spa)
1614 return (spa->spa_mode);
1618 spa_bootfs(spa_t *spa)
1620 return (spa->spa_bootfs);
1624 spa_delegation(spa_t *spa)
1626 return (spa->spa_delegation);
1630 spa_meta_objset(spa_t *spa)
1632 return (spa->spa_meta_objset);
1636 spa_dedup_checksum(spa_t *spa)
1638 return (spa->spa_dedup_checksum);
1642 * Reset pool scan stat per scan pass (or reboot).
1645 spa_scan_stat_init(spa_t *spa)
1647 /* data not stored on disk */
1648 spa->spa_scan_pass_start = gethrestime_sec();
1649 spa->spa_scan_pass_exam = 0;
1650 vdev_scan_stat_init(spa->spa_root_vdev);
1654 * Get scan stats for zpool status reports
1657 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
1659 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
1661 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
1663 bzero(ps, sizeof (pool_scan_stat_t));
1665 /* data stored on disk */
1666 ps->pss_func = scn->scn_phys.scn_func;
1667 ps->pss_start_time = scn->scn_phys.scn_start_time;
1668 ps->pss_end_time = scn->scn_phys.scn_end_time;
1669 ps->pss_to_examine = scn->scn_phys.scn_to_examine;
1670 ps->pss_examined = scn->scn_phys.scn_examined;
1671 ps->pss_to_process = scn->scn_phys.scn_to_process;
1672 ps->pss_processed = scn->scn_phys.scn_processed;
1673 ps->pss_errors = scn->scn_phys.scn_errors;
1674 ps->pss_state = scn->scn_phys.scn_state;
1676 /* data not stored on disk */
1677 ps->pss_pass_start = spa->spa_scan_pass_start;
1678 ps->pss_pass_exam = spa->spa_scan_pass_exam;
1683 #if defined(_KERNEL) && defined(HAVE_SPL)
1684 /* Namespace manipulation */
1685 EXPORT_SYMBOL(spa_lookup);
1686 EXPORT_SYMBOL(spa_add);
1687 EXPORT_SYMBOL(spa_remove);
1688 EXPORT_SYMBOL(spa_next);
1690 /* Refcount functions */
1691 EXPORT_SYMBOL(spa_open_ref);
1692 EXPORT_SYMBOL(spa_close);
1693 EXPORT_SYMBOL(spa_refcount_zero);
1695 /* Pool configuration lock */
1696 EXPORT_SYMBOL(spa_config_tryenter);
1697 EXPORT_SYMBOL(spa_config_enter);
1698 EXPORT_SYMBOL(spa_config_exit);
1699 EXPORT_SYMBOL(spa_config_held);
1701 /* Pool vdev add/remove lock */
1702 EXPORT_SYMBOL(spa_vdev_enter);
1703 EXPORT_SYMBOL(spa_vdev_exit);
1705 /* Pool vdev state change lock */
1706 EXPORT_SYMBOL(spa_vdev_state_enter);
1707 EXPORT_SYMBOL(spa_vdev_state_exit);
1709 /* Accessor functions */
1710 EXPORT_SYMBOL(spa_shutting_down);
1711 EXPORT_SYMBOL(spa_get_dsl);
1712 EXPORT_SYMBOL(spa_get_rootblkptr);
1713 EXPORT_SYMBOL(spa_set_rootblkptr);
1714 EXPORT_SYMBOL(spa_altroot);
1715 EXPORT_SYMBOL(spa_sync_pass);
1716 EXPORT_SYMBOL(spa_name);
1717 EXPORT_SYMBOL(spa_guid);
1718 EXPORT_SYMBOL(spa_last_synced_txg);
1719 EXPORT_SYMBOL(spa_first_txg);
1720 EXPORT_SYMBOL(spa_syncing_txg);
1721 EXPORT_SYMBOL(spa_version);
1722 EXPORT_SYMBOL(spa_state);
1723 EXPORT_SYMBOL(spa_load_state);
1724 EXPORT_SYMBOL(spa_freeze_txg);
1725 EXPORT_SYMBOL(spa_get_asize);
1726 EXPORT_SYMBOL(spa_get_dspace);
1727 EXPORT_SYMBOL(spa_update_dspace);
1728 EXPORT_SYMBOL(spa_deflate);
1729 EXPORT_SYMBOL(spa_normal_class);
1730 EXPORT_SYMBOL(spa_log_class);
1731 EXPORT_SYMBOL(spa_max_replication);
1732 EXPORT_SYMBOL(spa_prev_software_version);
1733 EXPORT_SYMBOL(spa_get_failmode);
1734 EXPORT_SYMBOL(spa_suspended);
1735 EXPORT_SYMBOL(spa_bootfs);
1736 EXPORT_SYMBOL(spa_delegation);
1737 EXPORT_SYMBOL(spa_meta_objset);
1739 /* Miscellaneous support routines */
1740 EXPORT_SYMBOL(spa_rename);
1741 EXPORT_SYMBOL(spa_guid_exists);
1742 EXPORT_SYMBOL(spa_strdup);
1743 EXPORT_SYMBOL(spa_strfree);
1744 EXPORT_SYMBOL(spa_get_random);
1745 EXPORT_SYMBOL(spa_generate_guid);
1746 EXPORT_SYMBOL(sprintf_blkptr);
1747 EXPORT_SYMBOL(spa_freeze);
1748 EXPORT_SYMBOL(spa_upgrade);
1749 EXPORT_SYMBOL(spa_evict_all);
1750 EXPORT_SYMBOL(spa_lookup_by_guid);
1751 EXPORT_SYMBOL(spa_has_spare);
1752 EXPORT_SYMBOL(dva_get_dsize_sync);
1753 EXPORT_SYMBOL(bp_get_dsize_sync);
1754 EXPORT_SYMBOL(bp_get_dsize);
1755 EXPORT_SYMBOL(spa_has_slogs);
1756 EXPORT_SYMBOL(spa_is_root);
1757 EXPORT_SYMBOL(spa_writeable);
1758 EXPORT_SYMBOL(spa_mode);
1760 EXPORT_SYMBOL(spa_namespace_lock);