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)
260 for (i = 0; i < SCL_LOCKS; i++) {
261 spa_config_lock_t *scl = &spa->spa_config_lock[i];
262 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
263 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
264 refcount_create(&scl->scl_count);
265 scl->scl_writer = NULL;
266 scl->scl_write_wanted = 0;
271 spa_config_lock_destroy(spa_t *spa)
275 for (i = 0; i < SCL_LOCKS; i++) {
276 spa_config_lock_t *scl = &spa->spa_config_lock[i];
277 mutex_destroy(&scl->scl_lock);
278 cv_destroy(&scl->scl_cv);
279 refcount_destroy(&scl->scl_count);
280 ASSERT(scl->scl_writer == NULL);
281 ASSERT(scl->scl_write_wanted == 0);
286 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
290 for (i = 0; i < SCL_LOCKS; i++) {
291 spa_config_lock_t *scl = &spa->spa_config_lock[i];
292 if (!(locks & (1 << i)))
294 mutex_enter(&scl->scl_lock);
295 if (rw == RW_READER) {
296 if (scl->scl_writer || scl->scl_write_wanted) {
297 mutex_exit(&scl->scl_lock);
298 spa_config_exit(spa, locks ^ (1 << i), tag);
302 ASSERT(scl->scl_writer != curthread);
303 if (!refcount_is_zero(&scl->scl_count)) {
304 mutex_exit(&scl->scl_lock);
305 spa_config_exit(spa, locks ^ (1 << i), tag);
308 scl->scl_writer = curthread;
310 (void) refcount_add(&scl->scl_count, tag);
311 mutex_exit(&scl->scl_lock);
317 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
322 for (i = 0; i < SCL_LOCKS; i++) {
323 spa_config_lock_t *scl = &spa->spa_config_lock[i];
324 if (scl->scl_writer == curthread)
325 wlocks_held |= (1 << i);
326 if (!(locks & (1 << i)))
328 mutex_enter(&scl->scl_lock);
329 if (rw == RW_READER) {
330 while (scl->scl_writer || scl->scl_write_wanted) {
331 cv_wait(&scl->scl_cv, &scl->scl_lock);
334 ASSERT(scl->scl_writer != curthread);
335 while (!refcount_is_zero(&scl->scl_count)) {
336 scl->scl_write_wanted++;
337 cv_wait(&scl->scl_cv, &scl->scl_lock);
338 scl->scl_write_wanted--;
340 scl->scl_writer = curthread;
342 (void) refcount_add(&scl->scl_count, tag);
343 mutex_exit(&scl->scl_lock);
345 ASSERT(wlocks_held <= locks);
349 spa_config_exit(spa_t *spa, int locks, void *tag)
353 for (i = SCL_LOCKS - 1; i >= 0; i--) {
354 spa_config_lock_t *scl = &spa->spa_config_lock[i];
355 if (!(locks & (1 << i)))
357 mutex_enter(&scl->scl_lock);
358 ASSERT(!refcount_is_zero(&scl->scl_count));
359 if (refcount_remove(&scl->scl_count, tag) == 0) {
360 ASSERT(scl->scl_writer == NULL ||
361 scl->scl_writer == curthread);
362 scl->scl_writer = NULL; /* OK in either case */
363 cv_broadcast(&scl->scl_cv);
365 mutex_exit(&scl->scl_lock);
370 spa_config_held(spa_t *spa, int locks, krw_t rw)
372 int i, locks_held = 0;
374 for (i = 0; i < SCL_LOCKS; i++) {
375 spa_config_lock_t *scl = &spa->spa_config_lock[i];
376 if (!(locks & (1 << i)))
378 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
379 (rw == RW_WRITER && scl->scl_writer == curthread))
380 locks_held |= 1 << i;
387 * ==========================================================================
388 * SPA namespace functions
389 * ==========================================================================
393 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
394 * Returns NULL if no matching spa_t is found.
397 spa_lookup(const char *name)
399 static spa_t search; /* spa_t is large; don't allocate on stack */
405 ASSERT(MUTEX_HELD(&spa_namespace_lock));
408 * If it's a full dataset name, figure out the pool name and
411 cp = strpbrk(name, "/@");
417 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
418 spa = avl_find(&spa_namespace_avl, &search, &where);
427 * Create an uninitialized spa_t with the given name. Requires
428 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
429 * exist by calling spa_lookup() first.
432 spa_add(const char *name, nvlist_t *config, const char *altroot)
435 spa_config_dirent_t *dp;
438 ASSERT(MUTEX_HELD(&spa_namespace_lock));
440 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
442 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
443 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
444 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
445 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
446 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
447 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
448 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
449 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
450 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
452 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
453 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
454 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
455 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
457 for (t = 0; t < TXG_SIZE; t++)
458 bplist_create(&spa->spa_free_bplist[t]);
460 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
461 spa->spa_state = POOL_STATE_UNINITIALIZED;
462 spa->spa_freeze_txg = UINT64_MAX;
463 spa->spa_final_txg = UINT64_MAX;
464 spa->spa_load_max_txg = UINT64_MAX;
466 spa->spa_proc_state = SPA_PROC_NONE;
468 refcount_create(&spa->spa_refcount);
469 spa_config_lock_init(spa);
471 avl_add(&spa_namespace_avl, spa);
474 * Set the alternate root, if there is one.
477 spa->spa_root = spa_strdup(altroot);
482 * Every pool starts with the default cachefile
484 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
485 offsetof(spa_config_dirent_t, scd_link));
487 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
488 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
489 list_insert_head(&spa->spa_config_list, dp);
491 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
495 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
501 * Removes a spa_t from the namespace, freeing up any memory used. Requires
502 * spa_namespace_lock. This is called only after the spa_t has been closed and
506 spa_remove(spa_t *spa)
508 spa_config_dirent_t *dp;
511 ASSERT(MUTEX_HELD(&spa_namespace_lock));
512 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
514 nvlist_free(spa->spa_config_splitting);
516 avl_remove(&spa_namespace_avl, spa);
517 cv_broadcast(&spa_namespace_cv);
520 spa_strfree(spa->spa_root);
524 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
525 list_remove(&spa->spa_config_list, dp);
526 if (dp->scd_path != NULL)
527 spa_strfree(dp->scd_path);
528 kmem_free(dp, sizeof (spa_config_dirent_t));
531 list_destroy(&spa->spa_config_list);
533 nvlist_free(spa->spa_load_info);
534 spa_config_set(spa, NULL);
536 refcount_destroy(&spa->spa_refcount);
538 spa_config_lock_destroy(spa);
540 for (t = 0; t < TXG_SIZE; t++)
541 bplist_destroy(&spa->spa_free_bplist[t]);
543 cv_destroy(&spa->spa_async_cv);
544 cv_destroy(&spa->spa_proc_cv);
545 cv_destroy(&spa->spa_scrub_io_cv);
546 cv_destroy(&spa->spa_suspend_cv);
548 mutex_destroy(&spa->spa_async_lock);
549 mutex_destroy(&spa->spa_errlist_lock);
550 mutex_destroy(&spa->spa_errlog_lock);
551 mutex_destroy(&spa->spa_history_lock);
552 mutex_destroy(&spa->spa_proc_lock);
553 mutex_destroy(&spa->spa_props_lock);
554 mutex_destroy(&spa->spa_scrub_lock);
555 mutex_destroy(&spa->spa_suspend_lock);
556 mutex_destroy(&spa->spa_vdev_top_lock);
558 kmem_free(spa, sizeof (spa_t));
562 * Given a pool, return the next pool in the namespace, or NULL if there is
563 * none. If 'prev' is NULL, return the first pool.
566 spa_next(spa_t *prev)
568 ASSERT(MUTEX_HELD(&spa_namespace_lock));
571 return (AVL_NEXT(&spa_namespace_avl, prev));
573 return (avl_first(&spa_namespace_avl));
577 * ==========================================================================
578 * SPA refcount functions
579 * ==========================================================================
583 * Add a reference to the given spa_t. Must have at least one reference, or
584 * have the namespace lock held.
587 spa_open_ref(spa_t *spa, void *tag)
589 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
590 MUTEX_HELD(&spa_namespace_lock));
591 (void) refcount_add(&spa->spa_refcount, tag);
595 * Remove a reference to the given spa_t. Must have at least one reference, or
596 * have the namespace lock held.
599 spa_close(spa_t *spa, void *tag)
601 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
602 MUTEX_HELD(&spa_namespace_lock));
603 (void) refcount_remove(&spa->spa_refcount, tag);
607 * Check to see if the spa refcount is zero. Must be called with
608 * spa_namespace_lock held. We really compare against spa_minref, which is the
609 * number of references acquired when opening a pool
612 spa_refcount_zero(spa_t *spa)
614 ASSERT(MUTEX_HELD(&spa_namespace_lock));
616 return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
620 * ==========================================================================
621 * SPA spare and l2cache tracking
622 * ==========================================================================
626 * Hot spares and cache devices are tracked using the same code below,
627 * for 'auxiliary' devices.
630 typedef struct spa_aux {
638 spa_aux_compare(const void *a, const void *b)
640 const spa_aux_t *sa = a;
641 const spa_aux_t *sb = b;
643 if (sa->aux_guid < sb->aux_guid)
645 else if (sa->aux_guid > sb->aux_guid)
652 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
658 search.aux_guid = vd->vdev_guid;
659 if ((aux = avl_find(avl, &search, &where)) != NULL) {
662 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
663 aux->aux_guid = vd->vdev_guid;
665 avl_insert(avl, aux, where);
670 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
676 search.aux_guid = vd->vdev_guid;
677 aux = avl_find(avl, &search, &where);
681 if (--aux->aux_count == 0) {
682 avl_remove(avl, aux);
683 kmem_free(aux, sizeof (spa_aux_t));
684 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
685 aux->aux_pool = 0ULL;
690 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
692 spa_aux_t search, *found;
694 search.aux_guid = guid;
695 found = avl_find(avl, &search, NULL);
699 *pool = found->aux_pool;
706 *refcnt = found->aux_count;
711 return (found != NULL);
715 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
717 spa_aux_t search, *found;
720 search.aux_guid = vd->vdev_guid;
721 found = avl_find(avl, &search, &where);
722 ASSERT(found != NULL);
723 ASSERT(found->aux_pool == 0ULL);
725 found->aux_pool = spa_guid(vd->vdev_spa);
729 * Spares are tracked globally due to the following constraints:
731 * - A spare may be part of multiple pools.
732 * - A spare may be added to a pool even if it's actively in use within
734 * - A spare in use in any pool can only be the source of a replacement if
735 * the target is a spare in the same pool.
737 * We keep track of all spares on the system through the use of a reference
738 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
739 * spare, then we bump the reference count in the AVL tree. In addition, we set
740 * the 'vdev_isspare' member to indicate that the device is a spare (active or
741 * inactive). When a spare is made active (used to replace a device in the
742 * pool), we also keep track of which pool its been made a part of.
744 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
745 * called under the spa_namespace lock as part of vdev reconfiguration. The
746 * separate spare lock exists for the status query path, which does not need to
747 * be completely consistent with respect to other vdev configuration changes.
751 spa_spare_compare(const void *a, const void *b)
753 return (spa_aux_compare(a, b));
757 spa_spare_add(vdev_t *vd)
759 mutex_enter(&spa_spare_lock);
760 ASSERT(!vd->vdev_isspare);
761 spa_aux_add(vd, &spa_spare_avl);
762 vd->vdev_isspare = B_TRUE;
763 mutex_exit(&spa_spare_lock);
767 spa_spare_remove(vdev_t *vd)
769 mutex_enter(&spa_spare_lock);
770 ASSERT(vd->vdev_isspare);
771 spa_aux_remove(vd, &spa_spare_avl);
772 vd->vdev_isspare = B_FALSE;
773 mutex_exit(&spa_spare_lock);
777 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
781 mutex_enter(&spa_spare_lock);
782 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
783 mutex_exit(&spa_spare_lock);
789 spa_spare_activate(vdev_t *vd)
791 mutex_enter(&spa_spare_lock);
792 ASSERT(vd->vdev_isspare);
793 spa_aux_activate(vd, &spa_spare_avl);
794 mutex_exit(&spa_spare_lock);
798 * Level 2 ARC devices are tracked globally for the same reasons as spares.
799 * Cache devices currently only support one pool per cache device, and so
800 * for these devices the aux reference count is currently unused beyond 1.
804 spa_l2cache_compare(const void *a, const void *b)
806 return (spa_aux_compare(a, b));
810 spa_l2cache_add(vdev_t *vd)
812 mutex_enter(&spa_l2cache_lock);
813 ASSERT(!vd->vdev_isl2cache);
814 spa_aux_add(vd, &spa_l2cache_avl);
815 vd->vdev_isl2cache = B_TRUE;
816 mutex_exit(&spa_l2cache_lock);
820 spa_l2cache_remove(vdev_t *vd)
822 mutex_enter(&spa_l2cache_lock);
823 ASSERT(vd->vdev_isl2cache);
824 spa_aux_remove(vd, &spa_l2cache_avl);
825 vd->vdev_isl2cache = B_FALSE;
826 mutex_exit(&spa_l2cache_lock);
830 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
834 mutex_enter(&spa_l2cache_lock);
835 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
836 mutex_exit(&spa_l2cache_lock);
842 spa_l2cache_activate(vdev_t *vd)
844 mutex_enter(&spa_l2cache_lock);
845 ASSERT(vd->vdev_isl2cache);
846 spa_aux_activate(vd, &spa_l2cache_avl);
847 mutex_exit(&spa_l2cache_lock);
851 * ==========================================================================
853 * ==========================================================================
857 * Lock the given spa_t for the purpose of adding or removing a vdev.
858 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
859 * It returns the next transaction group for the spa_t.
862 spa_vdev_enter(spa_t *spa)
864 mutex_enter(&spa->spa_vdev_top_lock);
865 mutex_enter(&spa_namespace_lock);
866 return (spa_vdev_config_enter(spa));
870 * Internal implementation for spa_vdev_enter(). Used when a vdev
871 * operation requires multiple syncs (i.e. removing a device) while
872 * keeping the spa_namespace_lock held.
875 spa_vdev_config_enter(spa_t *spa)
877 ASSERT(MUTEX_HELD(&spa_namespace_lock));
879 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
881 return (spa_last_synced_txg(spa) + 1);
885 * Used in combination with spa_vdev_config_enter() to allow the syncing
886 * of multiple transactions without releasing the spa_namespace_lock.
889 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
891 int config_changed = B_FALSE;
893 ASSERT(MUTEX_HELD(&spa_namespace_lock));
894 ASSERT(txg > spa_last_synced_txg(spa));
896 spa->spa_pending_vdev = NULL;
901 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
903 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
904 config_changed = B_TRUE;
905 spa->spa_config_generation++;
909 * Verify the metaslab classes.
911 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
912 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
914 spa_config_exit(spa, SCL_ALL, spa);
917 * Panic the system if the specified tag requires it. This
918 * is useful for ensuring that configurations are updated
921 if (zio_injection_enabled)
922 zio_handle_panic_injection(spa, tag, 0);
925 * Note: this txg_wait_synced() is important because it ensures
926 * that there won't be more than one config change per txg.
927 * This allows us to use the txg as the generation number.
930 txg_wait_synced(spa->spa_dsl_pool, txg);
933 ASSERT(!vd->vdev_detached || vd->vdev_dtl_smo.smo_object == 0);
934 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
936 spa_config_exit(spa, SCL_ALL, spa);
940 * If the config changed, update the config cache.
943 spa_config_sync(spa, B_FALSE, B_TRUE);
947 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
948 * locking of spa_vdev_enter(), we also want make sure the transactions have
949 * synced to disk, and then update the global configuration cache with the new
953 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
955 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
956 mutex_exit(&spa_namespace_lock);
957 mutex_exit(&spa->spa_vdev_top_lock);
963 * Lock the given spa_t for the purpose of changing vdev state.
966 spa_vdev_state_enter(spa_t *spa, int oplocks)
968 int locks = SCL_STATE_ALL | oplocks;
971 * Root pools may need to read of the underlying devfs filesystem
972 * when opening up a vdev. Unfortunately if we're holding the
973 * SCL_ZIO lock it will result in a deadlock when we try to issue
974 * the read from the root filesystem. Instead we "prefetch"
975 * the associated vnodes that we need prior to opening the
976 * underlying devices and cache them so that we can prevent
977 * any I/O when we are doing the actual open.
979 if (spa_is_root(spa)) {
980 int low = locks & ~(SCL_ZIO - 1);
981 int high = locks & ~low;
983 spa_config_enter(spa, high, spa, RW_WRITER);
984 vdev_hold(spa->spa_root_vdev);
985 spa_config_enter(spa, low, spa, RW_WRITER);
987 spa_config_enter(spa, locks, spa, RW_WRITER);
989 spa->spa_vdev_locks = locks;
993 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
995 boolean_t config_changed = B_FALSE;
997 if (vd != NULL || error == 0)
998 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1002 vdev_state_dirty(vd->vdev_top);
1003 config_changed = B_TRUE;
1004 spa->spa_config_generation++;
1007 if (spa_is_root(spa))
1008 vdev_rele(spa->spa_root_vdev);
1010 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1011 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1014 * If anything changed, wait for it to sync. This ensures that,
1015 * from the system administrator's perspective, zpool(1M) commands
1016 * are synchronous. This is important for things like zpool offline:
1017 * when the command completes, you expect no further I/O from ZFS.
1020 txg_wait_synced(spa->spa_dsl_pool, 0);
1023 * If the config changed, update the config cache.
1025 if (config_changed) {
1026 mutex_enter(&spa_namespace_lock);
1027 spa_config_sync(spa, B_FALSE, B_TRUE);
1028 mutex_exit(&spa_namespace_lock);
1035 * ==========================================================================
1036 * Miscellaneous functions
1037 * ==========================================================================
1044 spa_rename(const char *name, const char *newname)
1050 * Lookup the spa_t and grab the config lock for writing. We need to
1051 * actually open the pool so that we can sync out the necessary labels.
1052 * It's OK to call spa_open() with the namespace lock held because we
1053 * allow recursive calls for other reasons.
1055 mutex_enter(&spa_namespace_lock);
1056 if ((err = spa_open(name, &spa, FTAG)) != 0) {
1057 mutex_exit(&spa_namespace_lock);
1061 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1063 avl_remove(&spa_namespace_avl, spa);
1064 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1065 avl_add(&spa_namespace_avl, spa);
1068 * Sync all labels to disk with the new names by marking the root vdev
1069 * dirty and waiting for it to sync. It will pick up the new pool name
1072 vdev_config_dirty(spa->spa_root_vdev);
1074 spa_config_exit(spa, SCL_ALL, FTAG);
1076 txg_wait_synced(spa->spa_dsl_pool, 0);
1079 * Sync the updated config cache.
1081 spa_config_sync(spa, B_FALSE, B_TRUE);
1083 spa_close(spa, FTAG);
1085 mutex_exit(&spa_namespace_lock);
1091 * Return the spa_t associated with given pool_guid, if it exists. If
1092 * device_guid is non-zero, determine whether the pool exists *and* contains
1093 * a device with the specified device_guid.
1096 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1099 avl_tree_t *t = &spa_namespace_avl;
1101 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1103 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1104 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1106 if (spa->spa_root_vdev == NULL)
1108 if (spa_guid(spa) == pool_guid) {
1109 if (device_guid == 0)
1112 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1113 device_guid) != NULL)
1117 * Check any devices we may be in the process of adding.
1119 if (spa->spa_pending_vdev) {
1120 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1121 device_guid) != NULL)
1131 * Determine whether a pool with the given pool_guid exists.
1134 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1136 return (spa_by_guid(pool_guid, device_guid) != NULL);
1140 spa_strdup(const char *s)
1146 new = kmem_alloc(len + 1, KM_SLEEP);
1154 spa_strfree(char *s)
1156 kmem_free(s, strlen(s) + 1);
1160 spa_get_random(uint64_t range)
1166 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1172 spa_generate_guid(spa_t *spa)
1174 uint64_t guid = spa_get_random(-1ULL);
1177 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1178 guid = spa_get_random(-1ULL);
1180 while (guid == 0 || spa_guid_exists(guid, 0))
1181 guid = spa_get_random(-1ULL);
1188 sprintf_blkptr(char *buf, const blkptr_t *bp)
1191 char *checksum = NULL;
1192 char *compress = NULL;
1195 type = dmu_ot[BP_GET_TYPE(bp)].ot_name;
1196 checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1197 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1200 SPRINTF_BLKPTR(snprintf, ' ', buf, bp, type, checksum, compress);
1204 spa_freeze(spa_t *spa)
1206 uint64_t freeze_txg = 0;
1208 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1209 if (spa->spa_freeze_txg == UINT64_MAX) {
1210 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1211 spa->spa_freeze_txg = freeze_txg;
1213 spa_config_exit(spa, SCL_ALL, FTAG);
1214 if (freeze_txg != 0)
1215 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1219 zfs_panic_recover(const char *fmt, ...)
1224 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1229 * This is a stripped-down version of strtoull, suitable only for converting
1230 * lowercase hexidecimal numbers that don't overflow.
1233 strtonum(const char *str, char **nptr)
1239 while ((c = *str) != '\0') {
1240 if (c >= '0' && c <= '9')
1242 else if (c >= 'a' && c <= 'f')
1243 digit = 10 + c - 'a';
1254 *nptr = (char *)str;
1260 * ==========================================================================
1261 * Accessor functions
1262 * ==========================================================================
1266 spa_shutting_down(spa_t *spa)
1268 return (spa->spa_async_suspended);
1272 spa_get_dsl(spa_t *spa)
1274 return (spa->spa_dsl_pool);
1278 spa_get_rootblkptr(spa_t *spa)
1280 return (&spa->spa_ubsync.ub_rootbp);
1284 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1286 spa->spa_uberblock.ub_rootbp = *bp;
1290 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1292 if (spa->spa_root == NULL)
1295 (void) strncpy(buf, spa->spa_root, buflen);
1299 spa_sync_pass(spa_t *spa)
1301 return (spa->spa_sync_pass);
1305 spa_name(spa_t *spa)
1307 return (spa->spa_name);
1311 spa_guid(spa_t *spa)
1314 * If we fail to parse the config during spa_load(), we can go through
1315 * the error path (which posts an ereport) and end up here with no root
1316 * vdev. We stash the original pool guid in 'spa_load_guid' to handle
1319 if (spa->spa_root_vdev != NULL)
1320 return (spa->spa_root_vdev->vdev_guid);
1322 return (spa->spa_load_guid);
1326 spa_last_synced_txg(spa_t *spa)
1328 return (spa->spa_ubsync.ub_txg);
1332 spa_first_txg(spa_t *spa)
1334 return (spa->spa_first_txg);
1338 spa_syncing_txg(spa_t *spa)
1340 return (spa->spa_syncing_txg);
1344 spa_state(spa_t *spa)
1346 return (spa->spa_state);
1350 spa_load_state(spa_t *spa)
1352 return (spa->spa_load_state);
1356 spa_freeze_txg(spa_t *spa)
1358 return (spa->spa_freeze_txg);
1363 spa_get_asize(spa_t *spa, uint64_t lsize)
1366 * The worst case is single-sector max-parity RAID-Z blocks, in which
1367 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
1368 * times the size; so just assume that. Add to this the fact that
1369 * we can have up to 3 DVAs per bp, and one more factor of 2 because
1370 * the block may be dittoed with up to 3 DVAs by ddt_sync().
1372 return (lsize * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2);
1376 spa_get_dspace(spa_t *spa)
1378 return (spa->spa_dspace);
1382 spa_update_dspace(spa_t *spa)
1384 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1385 ddt_get_dedup_dspace(spa);
1389 * Return the failure mode that has been set to this pool. The default
1390 * behavior will be to block all I/Os when a complete failure occurs.
1393 spa_get_failmode(spa_t *spa)
1395 return (spa->spa_failmode);
1399 spa_suspended(spa_t *spa)
1401 return (spa->spa_suspended);
1405 spa_version(spa_t *spa)
1407 return (spa->spa_ubsync.ub_version);
1411 spa_deflate(spa_t *spa)
1413 return (spa->spa_deflate);
1417 spa_normal_class(spa_t *spa)
1419 return (spa->spa_normal_class);
1423 spa_log_class(spa_t *spa)
1425 return (spa->spa_log_class);
1429 spa_max_replication(spa_t *spa)
1432 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1433 * handle BPs with more than one DVA allocated. Set our max
1434 * replication level accordingly.
1436 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1438 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1442 spa_prev_software_version(spa_t *spa)
1444 return (spa->spa_prev_software_version);
1448 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1450 uint64_t asize = DVA_GET_ASIZE(dva);
1451 uint64_t dsize = asize;
1453 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1455 if (asize != 0 && spa->spa_deflate) {
1456 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1457 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1464 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1469 for (d = 0; d < SPA_DVAS_PER_BP; d++)
1470 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1476 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1481 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1483 for (d = 0; d < SPA_DVAS_PER_BP; d++)
1484 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1486 spa_config_exit(spa, SCL_VDEV, FTAG);
1492 * ==========================================================================
1493 * Initialization and Termination
1494 * ==========================================================================
1498 spa_name_compare(const void *a1, const void *a2)
1500 const spa_t *s1 = a1;
1501 const spa_t *s2 = a2;
1504 s = strcmp(s1->spa_name, s2->spa_name);
1515 return (spa_active_count);
1527 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1528 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1529 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1530 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1532 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1533 offsetof(spa_t, spa_avl));
1535 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1536 offsetof(spa_aux_t, aux_avl));
1538 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1539 offsetof(spa_aux_t, aux_avl));
1541 spa_mode_global = mode;
1548 vdev_cache_stat_init();
1562 vdev_cache_stat_fini();
1569 avl_destroy(&spa_namespace_avl);
1570 avl_destroy(&spa_spare_avl);
1571 avl_destroy(&spa_l2cache_avl);
1573 cv_destroy(&spa_namespace_cv);
1574 mutex_destroy(&spa_namespace_lock);
1575 mutex_destroy(&spa_spare_lock);
1576 mutex_destroy(&spa_l2cache_lock);
1580 * Return whether this pool has slogs. No locking needed.
1581 * It's not a problem if the wrong answer is returned as it's only for
1582 * performance and not correctness
1585 spa_has_slogs(spa_t *spa)
1587 return (spa->spa_log_class->mc_rotor != NULL);
1591 spa_get_log_state(spa_t *spa)
1593 return (spa->spa_log_state);
1597 spa_set_log_state(spa_t *spa, spa_log_state_t state)
1599 spa->spa_log_state = state;
1603 spa_is_root(spa_t *spa)
1605 return (spa->spa_is_root);
1609 spa_writeable(spa_t *spa)
1611 return (!!(spa->spa_mode & FWRITE));
1615 spa_mode(spa_t *spa)
1617 return (spa->spa_mode);
1621 spa_bootfs(spa_t *spa)
1623 return (spa->spa_bootfs);
1627 spa_delegation(spa_t *spa)
1629 return (spa->spa_delegation);
1633 spa_meta_objset(spa_t *spa)
1635 return (spa->spa_meta_objset);
1639 spa_dedup_checksum(spa_t *spa)
1641 return (spa->spa_dedup_checksum);
1645 * Reset pool scan stat per scan pass (or reboot).
1648 spa_scan_stat_init(spa_t *spa)
1650 /* data not stored on disk */
1651 spa->spa_scan_pass_start = gethrestime_sec();
1652 spa->spa_scan_pass_exam = 0;
1653 vdev_scan_stat_init(spa->spa_root_vdev);
1657 * Get scan stats for zpool status reports
1660 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
1662 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
1664 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
1666 bzero(ps, sizeof (pool_scan_stat_t));
1668 /* data stored on disk */
1669 ps->pss_func = scn->scn_phys.scn_func;
1670 ps->pss_start_time = scn->scn_phys.scn_start_time;
1671 ps->pss_end_time = scn->scn_phys.scn_end_time;
1672 ps->pss_to_examine = scn->scn_phys.scn_to_examine;
1673 ps->pss_examined = scn->scn_phys.scn_examined;
1674 ps->pss_to_process = scn->scn_phys.scn_to_process;
1675 ps->pss_processed = scn->scn_phys.scn_processed;
1676 ps->pss_errors = scn->scn_phys.scn_errors;
1677 ps->pss_state = scn->scn_phys.scn_state;
1679 /* data not stored on disk */
1680 ps->pss_pass_start = spa->spa_scan_pass_start;
1681 ps->pss_pass_exam = spa->spa_scan_pass_exam;