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/dsl_scan.h>
44 #include <sys/fs/zfs.h>
45 #include <sys/metaslab_impl.h>
53 * There are four basic locks for managing spa_t structures:
55 * spa_namespace_lock (global mutex)
57 * This lock must be acquired to do any of the following:
59 * - Lookup a spa_t by name
60 * - Add or remove a spa_t from the namespace
61 * - Increase spa_refcount from non-zero
62 * - Check if spa_refcount is zero
64 * - add/remove/attach/detach devices
65 * - Held for the duration of create/destroy/import/export
67 * It does not need to handle recursion. A create or destroy may
68 * reference objects (files or zvols) in other pools, but by
69 * definition they must have an existing reference, and will never need
70 * to lookup a spa_t by name.
72 * spa_refcount (per-spa refcount_t protected by mutex)
74 * This reference count keep track of any active users of the spa_t. The
75 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
76 * the refcount is never really 'zero' - opening a pool implicitly keeps
77 * some references in the DMU. Internally we check against spa_minref, but
78 * present the image of a zero/non-zero value to consumers.
80 * spa_config_lock[] (per-spa array of rwlocks)
82 * This protects the spa_t from config changes, and must be held in
83 * the following circumstances:
85 * - RW_READER to perform I/O to the spa
86 * - RW_WRITER to change the vdev config
88 * The locking order is fairly straightforward:
90 * spa_namespace_lock -> spa_refcount
92 * The namespace lock must be acquired to increase the refcount from 0
93 * or to check if it is zero.
95 * spa_refcount -> spa_config_lock[]
97 * There must be at least one valid reference on the spa_t to acquire
100 * spa_namespace_lock -> spa_config_lock[]
102 * The namespace lock must always be taken before the config lock.
105 * The spa_namespace_lock can be acquired directly and is globally visible.
107 * The namespace is manipulated using the following functions, all of which
108 * require the spa_namespace_lock to be held.
110 * spa_lookup() Lookup a spa_t by name.
112 * spa_add() Create a new spa_t in the namespace.
114 * spa_remove() Remove a spa_t from the namespace. This also
115 * frees up any memory associated with the spa_t.
117 * spa_next() Returns the next spa_t in the system, or the
118 * first if NULL is passed.
120 * spa_evict_all() Shutdown and remove all spa_t structures in
123 * spa_guid_exists() Determine whether a pool/device guid exists.
125 * The spa_refcount is manipulated using the following functions:
127 * spa_open_ref() Adds a reference to the given spa_t. Must be
128 * called with spa_namespace_lock held if the
129 * refcount is currently zero.
131 * spa_close() Remove a reference from the spa_t. This will
132 * not free the spa_t or remove it from the
133 * namespace. No locking is required.
135 * spa_refcount_zero() Returns true if the refcount is currently
136 * zero. Must be called with spa_namespace_lock
139 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
140 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
141 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
143 * To read the configuration, it suffices to hold one of these locks as reader.
144 * To modify the configuration, you must hold all locks as writer. To modify
145 * vdev state without altering the vdev tree's topology (e.g. online/offline),
146 * you must hold SCL_STATE and SCL_ZIO as writer.
148 * We use these distinct config locks to avoid recursive lock entry.
149 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
150 * block allocations (SCL_ALLOC), which may require reading space maps
151 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
153 * The spa config locks cannot be normal rwlocks because we need the
154 * ability to hand off ownership. For example, SCL_ZIO is acquired
155 * by the issuing thread and later released by an interrupt thread.
156 * They do, however, obey the usual write-wanted semantics to prevent
157 * writer (i.e. system administrator) starvation.
159 * The lock acquisition rules are as follows:
162 * Protects changes to the vdev tree topology, such as vdev
163 * add/remove/attach/detach. Protects the dirty config list
164 * (spa_config_dirty_list) and the set of spares and l2arc devices.
167 * Protects changes to pool state and vdev state, such as vdev
168 * online/offline/fault/degrade/clear. Protects the dirty state list
169 * (spa_state_dirty_list) and global pool state (spa_state).
172 * Protects changes to metaslab groups and classes.
173 * Held as reader by metaslab_alloc() and metaslab_claim().
176 * Held by bp-level zios (those which have no io_vd upon entry)
177 * to prevent changes to the vdev tree. The bp-level zio implicitly
178 * protects all of its vdev child zios, which do not hold SCL_ZIO.
181 * Protects changes to metaslab groups and classes.
182 * Held as reader by metaslab_free(). SCL_FREE is distinct from
183 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
184 * blocks in zio_done() while another i/o that holds either
185 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
188 * Held as reader to prevent changes to the vdev tree during trivial
189 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
190 * other locks, and lower than all of them, to ensure that it's safe
191 * to acquire regardless of caller context.
193 * In addition, the following rules apply:
195 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
196 * The lock ordering is SCL_CONFIG > spa_props_lock.
198 * (b) I/O operations on leaf vdevs. For any zio operation that takes
199 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
200 * or zio_write_phys() -- the caller must ensure that the config cannot
201 * cannot change in the interim, and that the vdev cannot be reopened.
202 * SCL_STATE as reader suffices for both.
204 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
206 * spa_vdev_enter() Acquire the namespace lock and the config lock
209 * spa_vdev_exit() Release the config lock, wait for all I/O
210 * to complete, sync the updated configs to the
211 * cache, and release the namespace lock.
213 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
214 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
215 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
217 * spa_rename() is also implemented within this file since is requires
218 * manipulation of the namespace.
221 static avl_tree_t spa_namespace_avl;
222 kmutex_t spa_namespace_lock;
223 static kcondvar_t spa_namespace_cv;
224 static int spa_active_count;
225 int spa_max_replication_override = SPA_DVAS_PER_BP;
227 static kmutex_t spa_spare_lock;
228 static avl_tree_t spa_spare_avl;
229 static kmutex_t spa_l2cache_lock;
230 static avl_tree_t spa_l2cache_avl;
232 kmem_cache_t *spa_buffer_pool;
236 /* Everything except dprintf is on by default in debug builds */
237 int zfs_flags = ~ZFS_DEBUG_DPRINTF;
243 * zfs_recover can be set to nonzero to attempt to recover from
244 * otherwise-fatal errors, typically caused by on-disk corruption. When
245 * set, calls to zfs_panic_recover() will turn into warning messages.
251 * ==========================================================================
253 * ==========================================================================
256 spa_config_lock_init(spa_t *spa)
258 for (int i = 0; i < SCL_LOCKS; i++) {
259 spa_config_lock_t *scl = &spa->spa_config_lock[i];
260 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
261 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
262 refcount_create(&scl->scl_count);
263 scl->scl_writer = NULL;
264 scl->scl_write_wanted = 0;
269 spa_config_lock_destroy(spa_t *spa)
271 for (int i = 0; i < SCL_LOCKS; i++) {
272 spa_config_lock_t *scl = &spa->spa_config_lock[i];
273 mutex_destroy(&scl->scl_lock);
274 cv_destroy(&scl->scl_cv);
275 refcount_destroy(&scl->scl_count);
276 ASSERT(scl->scl_writer == NULL);
277 ASSERT(scl->scl_write_wanted == 0);
282 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
284 for (int i = 0; i < SCL_LOCKS; i++) {
285 spa_config_lock_t *scl = &spa->spa_config_lock[i];
286 if (!(locks & (1 << i)))
288 mutex_enter(&scl->scl_lock);
289 if (rw == RW_READER) {
290 if (scl->scl_writer || scl->scl_write_wanted) {
291 mutex_exit(&scl->scl_lock);
292 spa_config_exit(spa, locks ^ (1 << i), tag);
296 ASSERT(scl->scl_writer != curthread);
297 if (!refcount_is_zero(&scl->scl_count)) {
298 mutex_exit(&scl->scl_lock);
299 spa_config_exit(spa, locks ^ (1 << i), tag);
302 scl->scl_writer = curthread;
304 (void) refcount_add(&scl->scl_count, tag);
305 mutex_exit(&scl->scl_lock);
311 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
315 for (int i = 0; i < SCL_LOCKS; i++) {
316 spa_config_lock_t *scl = &spa->spa_config_lock[i];
317 if (scl->scl_writer == curthread)
318 wlocks_held |= (1 << i);
319 if (!(locks & (1 << i)))
321 mutex_enter(&scl->scl_lock);
322 if (rw == RW_READER) {
323 while (scl->scl_writer || scl->scl_write_wanted) {
324 cv_wait(&scl->scl_cv, &scl->scl_lock);
327 ASSERT(scl->scl_writer != curthread);
328 while (!refcount_is_zero(&scl->scl_count)) {
329 scl->scl_write_wanted++;
330 cv_wait(&scl->scl_cv, &scl->scl_lock);
331 scl->scl_write_wanted--;
333 scl->scl_writer = curthread;
335 (void) refcount_add(&scl->scl_count, tag);
336 mutex_exit(&scl->scl_lock);
338 ASSERT(wlocks_held <= locks);
342 spa_config_exit(spa_t *spa, int locks, void *tag)
344 for (int i = SCL_LOCKS - 1; i >= 0; i--) {
345 spa_config_lock_t *scl = &spa->spa_config_lock[i];
346 if (!(locks & (1 << i)))
348 mutex_enter(&scl->scl_lock);
349 ASSERT(!refcount_is_zero(&scl->scl_count));
350 if (refcount_remove(&scl->scl_count, tag) == 0) {
351 ASSERT(scl->scl_writer == NULL ||
352 scl->scl_writer == curthread);
353 scl->scl_writer = NULL; /* OK in either case */
354 cv_broadcast(&scl->scl_cv);
356 mutex_exit(&scl->scl_lock);
361 spa_config_held(spa_t *spa, int locks, krw_t rw)
365 for (int i = 0; i < SCL_LOCKS; i++) {
366 spa_config_lock_t *scl = &spa->spa_config_lock[i];
367 if (!(locks & (1 << i)))
369 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
370 (rw == RW_WRITER && scl->scl_writer == curthread))
371 locks_held |= 1 << i;
378 * ==========================================================================
379 * SPA namespace functions
380 * ==========================================================================
384 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
385 * Returns NULL if no matching spa_t is found.
388 spa_lookup(const char *name)
390 static spa_t search; /* spa_t is large; don't allocate on stack */
396 ASSERT(MUTEX_HELD(&spa_namespace_lock));
399 * If it's a full dataset name, figure out the pool name and
402 cp = strpbrk(name, "/@");
408 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
409 spa = avl_find(&spa_namespace_avl, &search, &where);
418 * Create an uninitialized spa_t with the given name. Requires
419 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
420 * exist by calling spa_lookup() first.
423 spa_add(const char *name, nvlist_t *config, const char *altroot)
426 spa_config_dirent_t *dp;
428 ASSERT(MUTEX_HELD(&spa_namespace_lock));
430 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
432 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
433 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
434 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
435 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
436 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
437 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
438 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
439 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
440 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
442 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
443 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
444 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
445 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
447 for (int t = 0; t < TXG_SIZE; t++)
448 bplist_create(&spa->spa_free_bplist[t]);
450 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
451 spa->spa_state = POOL_STATE_UNINITIALIZED;
452 spa->spa_freeze_txg = UINT64_MAX;
453 spa->spa_final_txg = UINT64_MAX;
454 spa->spa_load_max_txg = UINT64_MAX;
456 spa->spa_proc_state = SPA_PROC_NONE;
458 refcount_create(&spa->spa_refcount);
459 spa_config_lock_init(spa);
461 avl_add(&spa_namespace_avl, spa);
464 * Set the alternate root, if there is one.
467 spa->spa_root = spa_strdup(altroot);
472 * Every pool starts with the default cachefile
474 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
475 offsetof(spa_config_dirent_t, scd_link));
477 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
478 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
479 list_insert_head(&spa->spa_config_list, dp);
481 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
485 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
491 * Removes a spa_t from the namespace, freeing up any memory used. Requires
492 * spa_namespace_lock. This is called only after the spa_t has been closed and
496 spa_remove(spa_t *spa)
498 spa_config_dirent_t *dp;
500 ASSERT(MUTEX_HELD(&spa_namespace_lock));
501 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
503 nvlist_free(spa->spa_config_splitting);
505 avl_remove(&spa_namespace_avl, spa);
506 cv_broadcast(&spa_namespace_cv);
509 spa_strfree(spa->spa_root);
513 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
514 list_remove(&spa->spa_config_list, dp);
515 if (dp->scd_path != NULL)
516 spa_strfree(dp->scd_path);
517 kmem_free(dp, sizeof (spa_config_dirent_t));
520 list_destroy(&spa->spa_config_list);
522 nvlist_free(spa->spa_load_info);
523 spa_config_set(spa, NULL);
525 refcount_destroy(&spa->spa_refcount);
527 spa_config_lock_destroy(spa);
529 for (int t = 0; t < TXG_SIZE; t++)
530 bplist_destroy(&spa->spa_free_bplist[t]);
532 cv_destroy(&spa->spa_async_cv);
533 cv_destroy(&spa->spa_proc_cv);
534 cv_destroy(&spa->spa_scrub_io_cv);
535 cv_destroy(&spa->spa_suspend_cv);
537 mutex_destroy(&spa->spa_async_lock);
538 mutex_destroy(&spa->spa_errlist_lock);
539 mutex_destroy(&spa->spa_errlog_lock);
540 mutex_destroy(&spa->spa_history_lock);
541 mutex_destroy(&spa->spa_proc_lock);
542 mutex_destroy(&spa->spa_props_lock);
543 mutex_destroy(&spa->spa_scrub_lock);
544 mutex_destroy(&spa->spa_suspend_lock);
545 mutex_destroy(&spa->spa_vdev_top_lock);
547 kmem_free(spa, sizeof (spa_t));
551 * Given a pool, return the next pool in the namespace, or NULL if there is
552 * none. If 'prev' is NULL, return the first pool.
555 spa_next(spa_t *prev)
557 ASSERT(MUTEX_HELD(&spa_namespace_lock));
560 return (AVL_NEXT(&spa_namespace_avl, prev));
562 return (avl_first(&spa_namespace_avl));
566 * ==========================================================================
567 * SPA refcount functions
568 * ==========================================================================
572 * Add a reference to the given spa_t. Must have at least one reference, or
573 * have the namespace lock held.
576 spa_open_ref(spa_t *spa, void *tag)
578 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
579 MUTEX_HELD(&spa_namespace_lock));
580 (void) refcount_add(&spa->spa_refcount, tag);
584 * Remove a reference to the given spa_t. Must have at least one reference, or
585 * have the namespace lock held.
588 spa_close(spa_t *spa, void *tag)
590 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
591 MUTEX_HELD(&spa_namespace_lock));
592 (void) refcount_remove(&spa->spa_refcount, tag);
596 * Check to see if the spa refcount is zero. Must be called with
597 * spa_namespace_lock held. We really compare against spa_minref, which is the
598 * number of references acquired when opening a pool
601 spa_refcount_zero(spa_t *spa)
603 ASSERT(MUTEX_HELD(&spa_namespace_lock));
605 return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
609 * ==========================================================================
610 * SPA spare and l2cache tracking
611 * ==========================================================================
615 * Hot spares and cache devices are tracked using the same code below,
616 * for 'auxiliary' devices.
619 typedef struct spa_aux {
627 spa_aux_compare(const void *a, const void *b)
629 const spa_aux_t *sa = a;
630 const spa_aux_t *sb = b;
632 if (sa->aux_guid < sb->aux_guid)
634 else if (sa->aux_guid > sb->aux_guid)
641 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
647 search.aux_guid = vd->vdev_guid;
648 if ((aux = avl_find(avl, &search, &where)) != NULL) {
651 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
652 aux->aux_guid = vd->vdev_guid;
654 avl_insert(avl, aux, where);
659 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
665 search.aux_guid = vd->vdev_guid;
666 aux = avl_find(avl, &search, &where);
670 if (--aux->aux_count == 0) {
671 avl_remove(avl, aux);
672 kmem_free(aux, sizeof (spa_aux_t));
673 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
674 aux->aux_pool = 0ULL;
679 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
681 spa_aux_t search, *found;
683 search.aux_guid = guid;
684 found = avl_find(avl, &search, NULL);
688 *pool = found->aux_pool;
695 *refcnt = found->aux_count;
700 return (found != NULL);
704 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
706 spa_aux_t search, *found;
709 search.aux_guid = vd->vdev_guid;
710 found = avl_find(avl, &search, &where);
711 ASSERT(found != NULL);
712 ASSERT(found->aux_pool == 0ULL);
714 found->aux_pool = spa_guid(vd->vdev_spa);
718 * Spares are tracked globally due to the following constraints:
720 * - A spare may be part of multiple pools.
721 * - A spare may be added to a pool even if it's actively in use within
723 * - A spare in use in any pool can only be the source of a replacement if
724 * the target is a spare in the same pool.
726 * We keep track of all spares on the system through the use of a reference
727 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
728 * spare, then we bump the reference count in the AVL tree. In addition, we set
729 * the 'vdev_isspare' member to indicate that the device is a spare (active or
730 * inactive). When a spare is made active (used to replace a device in the
731 * pool), we also keep track of which pool its been made a part of.
733 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
734 * called under the spa_namespace lock as part of vdev reconfiguration. The
735 * separate spare lock exists for the status query path, which does not need to
736 * be completely consistent with respect to other vdev configuration changes.
740 spa_spare_compare(const void *a, const void *b)
742 return (spa_aux_compare(a, b));
746 spa_spare_add(vdev_t *vd)
748 mutex_enter(&spa_spare_lock);
749 ASSERT(!vd->vdev_isspare);
750 spa_aux_add(vd, &spa_spare_avl);
751 vd->vdev_isspare = B_TRUE;
752 mutex_exit(&spa_spare_lock);
756 spa_spare_remove(vdev_t *vd)
758 mutex_enter(&spa_spare_lock);
759 ASSERT(vd->vdev_isspare);
760 spa_aux_remove(vd, &spa_spare_avl);
761 vd->vdev_isspare = B_FALSE;
762 mutex_exit(&spa_spare_lock);
766 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
770 mutex_enter(&spa_spare_lock);
771 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
772 mutex_exit(&spa_spare_lock);
778 spa_spare_activate(vdev_t *vd)
780 mutex_enter(&spa_spare_lock);
781 ASSERT(vd->vdev_isspare);
782 spa_aux_activate(vd, &spa_spare_avl);
783 mutex_exit(&spa_spare_lock);
787 * Level 2 ARC devices are tracked globally for the same reasons as spares.
788 * Cache devices currently only support one pool per cache device, and so
789 * for these devices the aux reference count is currently unused beyond 1.
793 spa_l2cache_compare(const void *a, const void *b)
795 return (spa_aux_compare(a, b));
799 spa_l2cache_add(vdev_t *vd)
801 mutex_enter(&spa_l2cache_lock);
802 ASSERT(!vd->vdev_isl2cache);
803 spa_aux_add(vd, &spa_l2cache_avl);
804 vd->vdev_isl2cache = B_TRUE;
805 mutex_exit(&spa_l2cache_lock);
809 spa_l2cache_remove(vdev_t *vd)
811 mutex_enter(&spa_l2cache_lock);
812 ASSERT(vd->vdev_isl2cache);
813 spa_aux_remove(vd, &spa_l2cache_avl);
814 vd->vdev_isl2cache = B_FALSE;
815 mutex_exit(&spa_l2cache_lock);
819 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
823 mutex_enter(&spa_l2cache_lock);
824 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
825 mutex_exit(&spa_l2cache_lock);
831 spa_l2cache_activate(vdev_t *vd)
833 mutex_enter(&spa_l2cache_lock);
834 ASSERT(vd->vdev_isl2cache);
835 spa_aux_activate(vd, &spa_l2cache_avl);
836 mutex_exit(&spa_l2cache_lock);
840 * ==========================================================================
842 * ==========================================================================
846 * Lock the given spa_t for the purpose of adding or removing a vdev.
847 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
848 * It returns the next transaction group for the spa_t.
851 spa_vdev_enter(spa_t *spa)
853 mutex_enter(&spa->spa_vdev_top_lock);
854 mutex_enter(&spa_namespace_lock);
855 return (spa_vdev_config_enter(spa));
859 * Internal implementation for spa_vdev_enter(). Used when a vdev
860 * operation requires multiple syncs (i.e. removing a device) while
861 * keeping the spa_namespace_lock held.
864 spa_vdev_config_enter(spa_t *spa)
866 ASSERT(MUTEX_HELD(&spa_namespace_lock));
868 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
870 return (spa_last_synced_txg(spa) + 1);
874 * Used in combination with spa_vdev_config_enter() to allow the syncing
875 * of multiple transactions without releasing the spa_namespace_lock.
878 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
880 ASSERT(MUTEX_HELD(&spa_namespace_lock));
882 int config_changed = B_FALSE;
884 ASSERT(txg > spa_last_synced_txg(spa));
886 spa->spa_pending_vdev = NULL;
891 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
893 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
894 config_changed = B_TRUE;
895 spa->spa_config_generation++;
899 * Verify the metaslab classes.
901 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
902 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
904 spa_config_exit(spa, SCL_ALL, spa);
907 * Panic the system if the specified tag requires it. This
908 * is useful for ensuring that configurations are updated
911 if (zio_injection_enabled)
912 zio_handle_panic_injection(spa, tag, 0);
915 * Note: this txg_wait_synced() is important because it ensures
916 * that there won't be more than one config change per txg.
917 * This allows us to use the txg as the generation number.
920 txg_wait_synced(spa->spa_dsl_pool, txg);
923 ASSERT(!vd->vdev_detached || vd->vdev_dtl_smo.smo_object == 0);
924 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
926 spa_config_exit(spa, SCL_ALL, spa);
930 * If the config changed, update the config cache.
933 spa_config_sync(spa, B_FALSE, B_TRUE);
937 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
938 * locking of spa_vdev_enter(), we also want make sure the transactions have
939 * synced to disk, and then update the global configuration cache with the new
943 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
945 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
946 mutex_exit(&spa_namespace_lock);
947 mutex_exit(&spa->spa_vdev_top_lock);
953 * Lock the given spa_t for the purpose of changing vdev state.
956 spa_vdev_state_enter(spa_t *spa, int oplocks)
958 int locks = SCL_STATE_ALL | oplocks;
961 * Root pools may need to read of the underlying devfs filesystem
962 * when opening up a vdev. Unfortunately if we're holding the
963 * SCL_ZIO lock it will result in a deadlock when we try to issue
964 * the read from the root filesystem. Instead we "prefetch"
965 * the associated vnodes that we need prior to opening the
966 * underlying devices and cache them so that we can prevent
967 * any I/O when we are doing the actual open.
969 if (spa_is_root(spa)) {
970 int low = locks & ~(SCL_ZIO - 1);
971 int high = locks & ~low;
973 spa_config_enter(spa, high, spa, RW_WRITER);
974 vdev_hold(spa->spa_root_vdev);
975 spa_config_enter(spa, low, spa, RW_WRITER);
977 spa_config_enter(spa, locks, spa, RW_WRITER);
979 spa->spa_vdev_locks = locks;
983 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
985 boolean_t config_changed = B_FALSE;
987 if (vd != NULL || error == 0)
988 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
992 vdev_state_dirty(vd->vdev_top);
993 config_changed = B_TRUE;
994 spa->spa_config_generation++;
997 if (spa_is_root(spa))
998 vdev_rele(spa->spa_root_vdev);
1000 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1001 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1004 * If anything changed, wait for it to sync. This ensures that,
1005 * from the system administrator's perspective, zpool(1M) commands
1006 * are synchronous. This is important for things like zpool offline:
1007 * when the command completes, you expect no further I/O from ZFS.
1010 txg_wait_synced(spa->spa_dsl_pool, 0);
1013 * If the config changed, update the config cache.
1015 if (config_changed) {
1016 mutex_enter(&spa_namespace_lock);
1017 spa_config_sync(spa, B_FALSE, B_TRUE);
1018 mutex_exit(&spa_namespace_lock);
1025 * ==========================================================================
1026 * Miscellaneous functions
1027 * ==========================================================================
1034 spa_rename(const char *name, const char *newname)
1040 * Lookup the spa_t and grab the config lock for writing. We need to
1041 * actually open the pool so that we can sync out the necessary labels.
1042 * It's OK to call spa_open() with the namespace lock held because we
1043 * allow recursive calls for other reasons.
1045 mutex_enter(&spa_namespace_lock);
1046 if ((err = spa_open(name, &spa, FTAG)) != 0) {
1047 mutex_exit(&spa_namespace_lock);
1051 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1053 avl_remove(&spa_namespace_avl, spa);
1054 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1055 avl_add(&spa_namespace_avl, spa);
1058 * Sync all labels to disk with the new names by marking the root vdev
1059 * dirty and waiting for it to sync. It will pick up the new pool name
1062 vdev_config_dirty(spa->spa_root_vdev);
1064 spa_config_exit(spa, SCL_ALL, FTAG);
1066 txg_wait_synced(spa->spa_dsl_pool, 0);
1069 * Sync the updated config cache.
1071 spa_config_sync(spa, B_FALSE, B_TRUE);
1073 spa_close(spa, FTAG);
1075 mutex_exit(&spa_namespace_lock);
1081 * Return the spa_t associated with given pool_guid, if it exists. If
1082 * device_guid is non-zero, determine whether the pool exists *and* contains
1083 * a device with the specified device_guid.
1086 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1089 avl_tree_t *t = &spa_namespace_avl;
1091 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1093 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1094 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1096 if (spa->spa_root_vdev == NULL)
1098 if (spa_guid(spa) == pool_guid) {
1099 if (device_guid == 0)
1102 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1103 device_guid) != NULL)
1107 * Check any devices we may be in the process of adding.
1109 if (spa->spa_pending_vdev) {
1110 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1111 device_guid) != NULL)
1121 * Determine whether a pool with the given pool_guid exists.
1124 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1126 return (spa_by_guid(pool_guid, device_guid) != NULL);
1130 spa_strdup(const char *s)
1136 new = kmem_alloc(len + 1, KM_SLEEP);
1144 spa_strfree(char *s)
1146 kmem_free(s, strlen(s) + 1);
1150 spa_get_random(uint64_t range)
1156 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1162 spa_generate_guid(spa_t *spa)
1164 uint64_t guid = spa_get_random(-1ULL);
1167 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1168 guid = spa_get_random(-1ULL);
1170 while (guid == 0 || spa_guid_exists(guid, 0))
1171 guid = spa_get_random(-1ULL);
1178 sprintf_blkptr(char *buf, const blkptr_t *bp)
1181 char *checksum = NULL;
1182 char *compress = NULL;
1185 type = dmu_ot[BP_GET_TYPE(bp)].ot_name;
1186 checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1187 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1190 SPRINTF_BLKPTR(snprintf, ' ', buf, bp, type, checksum, compress);
1194 spa_freeze(spa_t *spa)
1196 uint64_t freeze_txg = 0;
1198 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1199 if (spa->spa_freeze_txg == UINT64_MAX) {
1200 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1201 spa->spa_freeze_txg = freeze_txg;
1203 spa_config_exit(spa, SCL_ALL, FTAG);
1204 if (freeze_txg != 0)
1205 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1209 zfs_panic_recover(const char *fmt, ...)
1214 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1219 * This is a stripped-down version of strtoull, suitable only for converting
1220 * lowercase hexidecimal numbers that don't overflow.
1223 strtonum(const char *str, char **nptr)
1229 while ((c = *str) != '\0') {
1230 if (c >= '0' && c <= '9')
1232 else if (c >= 'a' && c <= 'f')
1233 digit = 10 + c - 'a';
1244 *nptr = (char *)str;
1250 * ==========================================================================
1251 * Accessor functions
1252 * ==========================================================================
1256 spa_shutting_down(spa_t *spa)
1258 return (spa->spa_async_suspended);
1262 spa_get_dsl(spa_t *spa)
1264 return (spa->spa_dsl_pool);
1268 spa_get_rootblkptr(spa_t *spa)
1270 return (&spa->spa_ubsync.ub_rootbp);
1274 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1276 spa->spa_uberblock.ub_rootbp = *bp;
1280 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1282 if (spa->spa_root == NULL)
1285 (void) strncpy(buf, spa->spa_root, buflen);
1289 spa_sync_pass(spa_t *spa)
1291 return (spa->spa_sync_pass);
1295 spa_name(spa_t *spa)
1297 return (spa->spa_name);
1301 spa_guid(spa_t *spa)
1304 * If we fail to parse the config during spa_load(), we can go through
1305 * the error path (which posts an ereport) and end up here with no root
1306 * vdev. We stash the original pool guid in 'spa_load_guid' to handle
1309 if (spa->spa_root_vdev != NULL)
1310 return (spa->spa_root_vdev->vdev_guid);
1312 return (spa->spa_load_guid);
1316 spa_last_synced_txg(spa_t *spa)
1318 return (spa->spa_ubsync.ub_txg);
1322 spa_first_txg(spa_t *spa)
1324 return (spa->spa_first_txg);
1328 spa_syncing_txg(spa_t *spa)
1330 return (spa->spa_syncing_txg);
1334 spa_state(spa_t *spa)
1336 return (spa->spa_state);
1340 spa_load_state(spa_t *spa)
1342 return (spa->spa_load_state);
1346 spa_freeze_txg(spa_t *spa)
1348 return (spa->spa_freeze_txg);
1353 spa_get_asize(spa_t *spa, uint64_t lsize)
1356 * The worst case is single-sector max-parity RAID-Z blocks, in which
1357 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
1358 * times the size; so just assume that. Add to this the fact that
1359 * we can have up to 3 DVAs per bp, and one more factor of 2 because
1360 * the block may be dittoed with up to 3 DVAs by ddt_sync().
1362 return (lsize * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2);
1366 spa_get_dspace(spa_t *spa)
1368 return (spa->spa_dspace);
1372 spa_update_dspace(spa_t *spa)
1374 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1375 ddt_get_dedup_dspace(spa);
1379 * Return the failure mode that has been set to this pool. The default
1380 * behavior will be to block all I/Os when a complete failure occurs.
1383 spa_get_failmode(spa_t *spa)
1385 return (spa->spa_failmode);
1389 spa_suspended(spa_t *spa)
1391 return (spa->spa_suspended);
1395 spa_version(spa_t *spa)
1397 return (spa->spa_ubsync.ub_version);
1401 spa_deflate(spa_t *spa)
1403 return (spa->spa_deflate);
1407 spa_normal_class(spa_t *spa)
1409 return (spa->spa_normal_class);
1413 spa_log_class(spa_t *spa)
1415 return (spa->spa_log_class);
1419 spa_max_replication(spa_t *spa)
1422 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1423 * handle BPs with more than one DVA allocated. Set our max
1424 * replication level accordingly.
1426 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1428 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1432 spa_prev_software_version(spa_t *spa)
1434 return (spa->spa_prev_software_version);
1438 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1440 uint64_t asize = DVA_GET_ASIZE(dva);
1441 uint64_t dsize = asize;
1443 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1445 if (asize != 0 && spa->spa_deflate) {
1446 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1447 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1454 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1458 for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1459 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1465 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1469 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1471 for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1472 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1474 spa_config_exit(spa, SCL_VDEV, FTAG);
1480 * ==========================================================================
1481 * Initialization and Termination
1482 * ==========================================================================
1486 spa_name_compare(const void *a1, const void *a2)
1488 const spa_t *s1 = a1;
1489 const spa_t *s2 = a2;
1492 s = strcmp(s1->spa_name, s2->spa_name);
1503 return (spa_active_count);
1515 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1516 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1517 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1518 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1520 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1521 offsetof(spa_t, spa_avl));
1523 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1524 offsetof(spa_aux_t, aux_avl));
1526 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1527 offsetof(spa_aux_t, aux_avl));
1529 spa_mode_global = mode;
1536 vdev_cache_stat_init();
1550 vdev_cache_stat_fini();
1557 avl_destroy(&spa_namespace_avl);
1558 avl_destroy(&spa_spare_avl);
1559 avl_destroy(&spa_l2cache_avl);
1561 cv_destroy(&spa_namespace_cv);
1562 mutex_destroy(&spa_namespace_lock);
1563 mutex_destroy(&spa_spare_lock);
1564 mutex_destroy(&spa_l2cache_lock);
1568 * Return whether this pool has slogs. No locking needed.
1569 * It's not a problem if the wrong answer is returned as it's only for
1570 * performance and not correctness
1573 spa_has_slogs(spa_t *spa)
1575 return (spa->spa_log_class->mc_rotor != NULL);
1579 spa_get_log_state(spa_t *spa)
1581 return (spa->spa_log_state);
1585 spa_set_log_state(spa_t *spa, spa_log_state_t state)
1587 spa->spa_log_state = state;
1591 spa_is_root(spa_t *spa)
1593 return (spa->spa_is_root);
1597 spa_writeable(spa_t *spa)
1599 return (!!(spa->spa_mode & FWRITE));
1603 spa_mode(spa_t *spa)
1605 return (spa->spa_mode);
1609 spa_bootfs(spa_t *spa)
1611 return (spa->spa_bootfs);
1615 spa_delegation(spa_t *spa)
1617 return (spa->spa_delegation);
1621 spa_meta_objset(spa_t *spa)
1623 return (spa->spa_meta_objset);
1627 spa_dedup_checksum(spa_t *spa)
1629 return (spa->spa_dedup_checksum);
1633 * Reset pool scan stat per scan pass (or reboot).
1636 spa_scan_stat_init(spa_t *spa)
1638 /* data not stored on disk */
1639 spa->spa_scan_pass_start = gethrestime_sec();
1640 spa->spa_scan_pass_exam = 0;
1641 vdev_scan_stat_init(spa->spa_root_vdev);
1645 * Get scan stats for zpool status reports
1648 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
1650 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
1652 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
1654 bzero(ps, sizeof (pool_scan_stat_t));
1656 /* data stored on disk */
1657 ps->pss_func = scn->scn_phys.scn_func;
1658 ps->pss_start_time = scn->scn_phys.scn_start_time;
1659 ps->pss_end_time = scn->scn_phys.scn_end_time;
1660 ps->pss_to_examine = scn->scn_phys.scn_to_examine;
1661 ps->pss_examined = scn->scn_phys.scn_examined;
1662 ps->pss_to_process = scn->scn_phys.scn_to_process;
1663 ps->pss_processed = scn->scn_phys.scn_processed;
1664 ps->pss_errors = scn->scn_phys.scn_errors;
1665 ps->pss_state = scn->scn_phys.scn_state;
1667 /* data not stored on disk */
1668 ps->pss_pass_start = spa->spa_scan_pass_start;
1669 ps->pss_pass_exam = spa->spa_scan_pass_exam;