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.
248 SYSCTL_DECL(_vfs_zfs);
249 TUNABLE_INT("vfs.zfs.recover", &zfs_recover);
250 SYSCTL_INT(_vfs_zfs, OID_AUTO, recover, CTLFLAG_RDTUN, &zfs_recover, 0,
251 "Try to recover from otherwise-fatal errors.");
255 * ==========================================================================
257 * ==========================================================================
260 spa_config_lock_init(spa_t *spa)
262 for (int i = 0; i < SCL_LOCKS; i++) {
263 spa_config_lock_t *scl = &spa->spa_config_lock[i];
264 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
265 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
266 refcount_create(&scl->scl_count);
267 scl->scl_writer = NULL;
268 scl->scl_write_wanted = 0;
273 spa_config_lock_destroy(spa_t *spa)
275 for (int 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)
288 for (int i = 0; i < SCL_LOCKS; i++) {
289 spa_config_lock_t *scl = &spa->spa_config_lock[i];
290 if (!(locks & (1 << i)))
292 mutex_enter(&scl->scl_lock);
293 if (rw == RW_READER) {
294 if (scl->scl_writer || scl->scl_write_wanted) {
295 mutex_exit(&scl->scl_lock);
296 spa_config_exit(spa, locks ^ (1 << i), tag);
300 ASSERT(scl->scl_writer != curthread);
301 if (!refcount_is_zero(&scl->scl_count)) {
302 mutex_exit(&scl->scl_lock);
303 spa_config_exit(spa, locks ^ (1 << i), tag);
306 scl->scl_writer = curthread;
308 (void) refcount_add(&scl->scl_count, tag);
309 mutex_exit(&scl->scl_lock);
315 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
319 for (int i = 0; i < SCL_LOCKS; i++) {
320 spa_config_lock_t *scl = &spa->spa_config_lock[i];
321 if (scl->scl_writer == curthread)
322 wlocks_held |= (1 << i);
323 if (!(locks & (1 << i)))
325 mutex_enter(&scl->scl_lock);
326 if (rw == RW_READER) {
327 while (scl->scl_writer || scl->scl_write_wanted) {
328 cv_wait(&scl->scl_cv, &scl->scl_lock);
331 ASSERT(scl->scl_writer != curthread);
332 while (!refcount_is_zero(&scl->scl_count)) {
333 scl->scl_write_wanted++;
334 cv_wait(&scl->scl_cv, &scl->scl_lock);
335 scl->scl_write_wanted--;
337 scl->scl_writer = curthread;
339 (void) refcount_add(&scl->scl_count, tag);
340 mutex_exit(&scl->scl_lock);
342 ASSERT(wlocks_held <= locks);
346 spa_config_exit(spa_t *spa, int locks, void *tag)
348 for (int i = SCL_LOCKS - 1; i >= 0; i--) {
349 spa_config_lock_t *scl = &spa->spa_config_lock[i];
350 if (!(locks & (1 << i)))
352 mutex_enter(&scl->scl_lock);
353 ASSERT(!refcount_is_zero(&scl->scl_count));
354 if (refcount_remove(&scl->scl_count, tag) == 0) {
355 ASSERT(scl->scl_writer == NULL ||
356 scl->scl_writer == curthread);
357 scl->scl_writer = NULL; /* OK in either case */
358 cv_broadcast(&scl->scl_cv);
360 mutex_exit(&scl->scl_lock);
365 spa_config_held(spa_t *spa, int locks, krw_t rw)
369 for (int i = 0; i < SCL_LOCKS; i++) {
370 spa_config_lock_t *scl = &spa->spa_config_lock[i];
371 if (!(locks & (1 << i)))
373 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
374 (rw == RW_WRITER && scl->scl_writer == curthread))
375 locks_held |= 1 << i;
382 * ==========================================================================
383 * SPA namespace functions
384 * ==========================================================================
388 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
389 * Returns NULL if no matching spa_t is found.
392 spa_lookup(const char *name)
394 static spa_t search; /* spa_t is large; don't allocate on stack */
400 ASSERT(MUTEX_HELD(&spa_namespace_lock));
403 * If it's a full dataset name, figure out the pool name and
406 cp = strpbrk(name, "/@");
412 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
413 spa = avl_find(&spa_namespace_avl, &search, &where);
422 * Create an uninitialized spa_t with the given name. Requires
423 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
424 * exist by calling spa_lookup() first.
427 spa_add(const char *name, nvlist_t *config, const char *altroot)
430 spa_config_dirent_t *dp;
432 ASSERT(MUTEX_HELD(&spa_namespace_lock));
434 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
436 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
437 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
438 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
439 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
440 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
441 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
442 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
443 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
444 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
446 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
447 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
448 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
449 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
451 for (int t = 0; t < TXG_SIZE; t++)
452 bplist_create(&spa->spa_free_bplist[t]);
454 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
455 spa->spa_state = POOL_STATE_UNINITIALIZED;
456 spa->spa_freeze_txg = UINT64_MAX;
457 spa->spa_final_txg = UINT64_MAX;
458 spa->spa_load_max_txg = UINT64_MAX;
460 spa->spa_proc_state = SPA_PROC_NONE;
462 refcount_create(&spa->spa_refcount);
463 spa_config_lock_init(spa);
465 avl_add(&spa_namespace_avl, spa);
468 * Set the alternate root, if there is one.
471 spa->spa_root = spa_strdup(altroot);
476 * Every pool starts with the default cachefile
478 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
479 offsetof(spa_config_dirent_t, scd_link));
481 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
482 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
483 list_insert_head(&spa->spa_config_list, dp);
485 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
489 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
495 * Removes a spa_t from the namespace, freeing up any memory used. Requires
496 * spa_namespace_lock. This is called only after the spa_t has been closed and
500 spa_remove(spa_t *spa)
502 spa_config_dirent_t *dp;
504 ASSERT(MUTEX_HELD(&spa_namespace_lock));
505 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
507 nvlist_free(spa->spa_config_splitting);
509 avl_remove(&spa_namespace_avl, spa);
510 cv_broadcast(&spa_namespace_cv);
513 spa_strfree(spa->spa_root);
517 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
518 list_remove(&spa->spa_config_list, dp);
519 if (dp->scd_path != NULL)
520 spa_strfree(dp->scd_path);
521 kmem_free(dp, sizeof (spa_config_dirent_t));
524 list_destroy(&spa->spa_config_list);
526 nvlist_free(spa->spa_load_info);
527 spa_config_set(spa, NULL);
529 refcount_destroy(&spa->spa_refcount);
531 spa_config_lock_destroy(spa);
533 for (int t = 0; t < TXG_SIZE; t++)
534 bplist_destroy(&spa->spa_free_bplist[t]);
536 cv_destroy(&spa->spa_async_cv);
537 cv_destroy(&spa->spa_proc_cv);
538 cv_destroy(&spa->spa_scrub_io_cv);
539 cv_destroy(&spa->spa_suspend_cv);
541 mutex_destroy(&spa->spa_async_lock);
542 mutex_destroy(&spa->spa_errlist_lock);
543 mutex_destroy(&spa->spa_errlog_lock);
544 mutex_destroy(&spa->spa_history_lock);
545 mutex_destroy(&spa->spa_proc_lock);
546 mutex_destroy(&spa->spa_props_lock);
547 mutex_destroy(&spa->spa_scrub_lock);
548 mutex_destroy(&spa->spa_suspend_lock);
549 mutex_destroy(&spa->spa_vdev_top_lock);
551 kmem_free(spa, sizeof (spa_t));
555 * Given a pool, return the next pool in the namespace, or NULL if there is
556 * none. If 'prev' is NULL, return the first pool.
559 spa_next(spa_t *prev)
561 ASSERT(MUTEX_HELD(&spa_namespace_lock));
564 return (AVL_NEXT(&spa_namespace_avl, prev));
566 return (avl_first(&spa_namespace_avl));
570 * ==========================================================================
571 * SPA refcount functions
572 * ==========================================================================
576 * Add a reference to the given spa_t. Must have at least one reference, or
577 * have the namespace lock held.
580 spa_open_ref(spa_t *spa, void *tag)
582 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
583 MUTEX_HELD(&spa_namespace_lock));
584 (void) refcount_add(&spa->spa_refcount, tag);
588 * Remove a reference to the given spa_t. Must have at least one reference, or
589 * have the namespace lock held.
592 spa_close(spa_t *spa, void *tag)
594 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
595 MUTEX_HELD(&spa_namespace_lock));
596 (void) refcount_remove(&spa->spa_refcount, tag);
600 * Check to see if the spa refcount is zero. Must be called with
601 * spa_namespace_lock held. We really compare against spa_minref, which is the
602 * number of references acquired when opening a pool
605 spa_refcount_zero(spa_t *spa)
607 ASSERT(MUTEX_HELD(&spa_namespace_lock));
609 return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
613 * ==========================================================================
614 * SPA spare and l2cache tracking
615 * ==========================================================================
619 * Hot spares and cache devices are tracked using the same code below,
620 * for 'auxiliary' devices.
623 typedef struct spa_aux {
631 spa_aux_compare(const void *a, const void *b)
633 const spa_aux_t *sa = a;
634 const spa_aux_t *sb = b;
636 if (sa->aux_guid < sb->aux_guid)
638 else if (sa->aux_guid > sb->aux_guid)
645 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
651 search.aux_guid = vd->vdev_guid;
652 if ((aux = avl_find(avl, &search, &where)) != NULL) {
655 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
656 aux->aux_guid = vd->vdev_guid;
658 avl_insert(avl, aux, where);
663 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
669 search.aux_guid = vd->vdev_guid;
670 aux = avl_find(avl, &search, &where);
674 if (--aux->aux_count == 0) {
675 avl_remove(avl, aux);
676 kmem_free(aux, sizeof (spa_aux_t));
677 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
678 aux->aux_pool = 0ULL;
683 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
685 spa_aux_t search, *found;
687 search.aux_guid = guid;
688 found = avl_find(avl, &search, NULL);
692 *pool = found->aux_pool;
699 *refcnt = found->aux_count;
704 return (found != NULL);
708 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
710 spa_aux_t search, *found;
713 search.aux_guid = vd->vdev_guid;
714 found = avl_find(avl, &search, &where);
715 ASSERT(found != NULL);
716 ASSERT(found->aux_pool == 0ULL);
718 found->aux_pool = spa_guid(vd->vdev_spa);
722 * Spares are tracked globally due to the following constraints:
724 * - A spare may be part of multiple pools.
725 * - A spare may be added to a pool even if it's actively in use within
727 * - A spare in use in any pool can only be the source of a replacement if
728 * the target is a spare in the same pool.
730 * We keep track of all spares on the system through the use of a reference
731 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
732 * spare, then we bump the reference count in the AVL tree. In addition, we set
733 * the 'vdev_isspare' member to indicate that the device is a spare (active or
734 * inactive). When a spare is made active (used to replace a device in the
735 * pool), we also keep track of which pool its been made a part of.
737 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
738 * called under the spa_namespace lock as part of vdev reconfiguration. The
739 * separate spare lock exists for the status query path, which does not need to
740 * be completely consistent with respect to other vdev configuration changes.
744 spa_spare_compare(const void *a, const void *b)
746 return (spa_aux_compare(a, b));
750 spa_spare_add(vdev_t *vd)
752 mutex_enter(&spa_spare_lock);
753 ASSERT(!vd->vdev_isspare);
754 spa_aux_add(vd, &spa_spare_avl);
755 vd->vdev_isspare = B_TRUE;
756 mutex_exit(&spa_spare_lock);
760 spa_spare_remove(vdev_t *vd)
762 mutex_enter(&spa_spare_lock);
763 ASSERT(vd->vdev_isspare);
764 spa_aux_remove(vd, &spa_spare_avl);
765 vd->vdev_isspare = B_FALSE;
766 mutex_exit(&spa_spare_lock);
770 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
774 mutex_enter(&spa_spare_lock);
775 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
776 mutex_exit(&spa_spare_lock);
782 spa_spare_activate(vdev_t *vd)
784 mutex_enter(&spa_spare_lock);
785 ASSERT(vd->vdev_isspare);
786 spa_aux_activate(vd, &spa_spare_avl);
787 mutex_exit(&spa_spare_lock);
791 * Level 2 ARC devices are tracked globally for the same reasons as spares.
792 * Cache devices currently only support one pool per cache device, and so
793 * for these devices the aux reference count is currently unused beyond 1.
797 spa_l2cache_compare(const void *a, const void *b)
799 return (spa_aux_compare(a, b));
803 spa_l2cache_add(vdev_t *vd)
805 mutex_enter(&spa_l2cache_lock);
806 ASSERT(!vd->vdev_isl2cache);
807 spa_aux_add(vd, &spa_l2cache_avl);
808 vd->vdev_isl2cache = B_TRUE;
809 mutex_exit(&spa_l2cache_lock);
813 spa_l2cache_remove(vdev_t *vd)
815 mutex_enter(&spa_l2cache_lock);
816 ASSERT(vd->vdev_isl2cache);
817 spa_aux_remove(vd, &spa_l2cache_avl);
818 vd->vdev_isl2cache = B_FALSE;
819 mutex_exit(&spa_l2cache_lock);
823 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
827 mutex_enter(&spa_l2cache_lock);
828 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
829 mutex_exit(&spa_l2cache_lock);
835 spa_l2cache_activate(vdev_t *vd)
837 mutex_enter(&spa_l2cache_lock);
838 ASSERT(vd->vdev_isl2cache);
839 spa_aux_activate(vd, &spa_l2cache_avl);
840 mutex_exit(&spa_l2cache_lock);
844 * ==========================================================================
846 * ==========================================================================
850 * Lock the given spa_t for the purpose of adding or removing a vdev.
851 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
852 * It returns the next transaction group for the spa_t.
855 spa_vdev_enter(spa_t *spa)
857 mutex_enter(&spa->spa_vdev_top_lock);
858 mutex_enter(&spa_namespace_lock);
859 return (spa_vdev_config_enter(spa));
863 * Internal implementation for spa_vdev_enter(). Used when a vdev
864 * operation requires multiple syncs (i.e. removing a device) while
865 * keeping the spa_namespace_lock held.
868 spa_vdev_config_enter(spa_t *spa)
870 ASSERT(MUTEX_HELD(&spa_namespace_lock));
872 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
874 return (spa_last_synced_txg(spa) + 1);
878 * Used in combination with spa_vdev_config_enter() to allow the syncing
879 * of multiple transactions without releasing the spa_namespace_lock.
882 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
884 ASSERT(MUTEX_HELD(&spa_namespace_lock));
886 int config_changed = B_FALSE;
888 ASSERT(txg > spa_last_synced_txg(spa));
890 spa->spa_pending_vdev = NULL;
895 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
897 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
898 config_changed = B_TRUE;
899 spa->spa_config_generation++;
903 * Verify the metaslab classes.
905 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
906 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
908 spa_config_exit(spa, SCL_ALL, spa);
911 * Panic the system if the specified tag requires it. This
912 * is useful for ensuring that configurations are updated
915 if (zio_injection_enabled)
916 zio_handle_panic_injection(spa, tag, 0);
919 * Note: this txg_wait_synced() is important because it ensures
920 * that there won't be more than one config change per txg.
921 * This allows us to use the txg as the generation number.
924 txg_wait_synced(spa->spa_dsl_pool, txg);
927 ASSERT(!vd->vdev_detached || vd->vdev_dtl_smo.smo_object == 0);
928 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
930 spa_config_exit(spa, SCL_ALL, spa);
934 * If the config changed, update the config cache.
937 spa_config_sync(spa, B_FALSE, B_TRUE);
941 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
942 * locking of spa_vdev_enter(), we also want make sure the transactions have
943 * synced to disk, and then update the global configuration cache with the new
947 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
949 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
950 mutex_exit(&spa_namespace_lock);
951 mutex_exit(&spa->spa_vdev_top_lock);
957 * Lock the given spa_t for the purpose of changing vdev state.
960 spa_vdev_state_enter(spa_t *spa, int oplocks)
962 int locks = SCL_STATE_ALL | oplocks;
965 * Root pools may need to read of the underlying devfs filesystem
966 * when opening up a vdev. Unfortunately if we're holding the
967 * SCL_ZIO lock it will result in a deadlock when we try to issue
968 * the read from the root filesystem. Instead we "prefetch"
969 * the associated vnodes that we need prior to opening the
970 * underlying devices and cache them so that we can prevent
971 * any I/O when we are doing the actual open.
973 if (spa_is_root(spa)) {
974 int low = locks & ~(SCL_ZIO - 1);
975 int high = locks & ~low;
977 spa_config_enter(spa, high, spa, RW_WRITER);
978 vdev_hold(spa->spa_root_vdev);
979 spa_config_enter(spa, low, spa, RW_WRITER);
981 spa_config_enter(spa, locks, spa, RW_WRITER);
983 spa->spa_vdev_locks = locks;
987 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
989 boolean_t config_changed = B_FALSE;
991 if (vd != NULL || error == 0)
992 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
996 vdev_state_dirty(vd->vdev_top);
997 config_changed = B_TRUE;
998 spa->spa_config_generation++;
1001 if (spa_is_root(spa))
1002 vdev_rele(spa->spa_root_vdev);
1004 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1005 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1008 * If anything changed, wait for it to sync. This ensures that,
1009 * from the system administrator's perspective, zpool(1M) commands
1010 * are synchronous. This is important for things like zpool offline:
1011 * when the command completes, you expect no further I/O from ZFS.
1014 txg_wait_synced(spa->spa_dsl_pool, 0);
1017 * If the config changed, update the config cache.
1019 if (config_changed) {
1020 mutex_enter(&spa_namespace_lock);
1021 spa_config_sync(spa, B_FALSE, B_TRUE);
1022 mutex_exit(&spa_namespace_lock);
1029 * ==========================================================================
1030 * Miscellaneous functions
1031 * ==========================================================================
1038 spa_rename(const char *name, const char *newname)
1044 * Lookup the spa_t and grab the config lock for writing. We need to
1045 * actually open the pool so that we can sync out the necessary labels.
1046 * It's OK to call spa_open() with the namespace lock held because we
1047 * allow recursive calls for other reasons.
1049 mutex_enter(&spa_namespace_lock);
1050 if ((err = spa_open(name, &spa, FTAG)) != 0) {
1051 mutex_exit(&spa_namespace_lock);
1055 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1057 avl_remove(&spa_namespace_avl, spa);
1058 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1059 avl_add(&spa_namespace_avl, spa);
1062 * Sync all labels to disk with the new names by marking the root vdev
1063 * dirty and waiting for it to sync. It will pick up the new pool name
1066 vdev_config_dirty(spa->spa_root_vdev);
1068 spa_config_exit(spa, SCL_ALL, FTAG);
1070 txg_wait_synced(spa->spa_dsl_pool, 0);
1073 * Sync the updated config cache.
1075 spa_config_sync(spa, B_FALSE, B_TRUE);
1077 spa_close(spa, FTAG);
1079 mutex_exit(&spa_namespace_lock);
1085 * Return the spa_t associated with given pool_guid, if it exists. If
1086 * device_guid is non-zero, determine whether the pool exists *and* contains
1087 * a device with the specified device_guid.
1090 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1093 avl_tree_t *t = &spa_namespace_avl;
1095 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1097 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1098 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1100 if (spa->spa_root_vdev == NULL)
1102 if (spa_guid(spa) == pool_guid) {
1103 if (device_guid == 0)
1106 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1107 device_guid) != NULL)
1111 * Check any devices we may be in the process of adding.
1113 if (spa->spa_pending_vdev) {
1114 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1115 device_guid) != NULL)
1125 * Determine whether a pool with the given pool_guid exists.
1128 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1130 return (spa_by_guid(pool_guid, device_guid) != NULL);
1134 spa_strdup(const char *s)
1140 new = kmem_alloc(len + 1, KM_SLEEP);
1148 spa_strfree(char *s)
1150 kmem_free(s, strlen(s) + 1);
1154 spa_get_random(uint64_t range)
1160 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1166 spa_generate_guid(spa_t *spa)
1168 uint64_t guid = spa_get_random(-1ULL);
1171 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1172 guid = spa_get_random(-1ULL);
1174 while (guid == 0 || spa_guid_exists(guid, 0))
1175 guid = spa_get_random(-1ULL);
1182 sprintf_blkptr(char *buf, const blkptr_t *bp)
1185 char *checksum = NULL;
1186 char *compress = NULL;
1189 type = dmu_ot[BP_GET_TYPE(bp)].ot_name;
1190 checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1191 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1194 SPRINTF_BLKPTR(snprintf, ' ', buf, bp, type, checksum, compress);
1198 spa_freeze(spa_t *spa)
1200 uint64_t freeze_txg = 0;
1202 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1203 if (spa->spa_freeze_txg == UINT64_MAX) {
1204 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1205 spa->spa_freeze_txg = freeze_txg;
1207 spa_config_exit(spa, SCL_ALL, FTAG);
1208 if (freeze_txg != 0)
1209 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1213 zfs_panic_recover(const char *fmt, ...)
1218 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1223 * This is a stripped-down version of strtoull, suitable only for converting
1224 * lowercase hexidecimal numbers that don't overflow.
1227 zfs_strtonum(const char *str, char **nptr)
1233 while ((c = *str) != '\0') {
1234 if (c >= '0' && c <= '9')
1236 else if (c >= 'a' && c <= 'f')
1237 digit = 10 + c - 'a';
1248 *nptr = (char *)str;
1254 * ==========================================================================
1255 * Accessor functions
1256 * ==========================================================================
1260 spa_shutting_down(spa_t *spa)
1262 return (spa->spa_async_suspended);
1266 spa_get_dsl(spa_t *spa)
1268 return (spa->spa_dsl_pool);
1272 spa_get_rootblkptr(spa_t *spa)
1274 return (&spa->spa_ubsync.ub_rootbp);
1278 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1280 spa->spa_uberblock.ub_rootbp = *bp;
1284 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1286 if (spa->spa_root == NULL)
1289 (void) strncpy(buf, spa->spa_root, buflen);
1293 spa_sync_pass(spa_t *spa)
1295 return (spa->spa_sync_pass);
1299 spa_name(spa_t *spa)
1301 return (spa->spa_name);
1305 spa_guid(spa_t *spa)
1308 * If we fail to parse the config during spa_load(), we can go through
1309 * the error path (which posts an ereport) and end up here with no root
1310 * vdev. We stash the original pool guid in 'spa_load_guid' to handle
1313 if (spa->spa_root_vdev != NULL)
1314 return (spa->spa_root_vdev->vdev_guid);
1316 return (spa->spa_load_guid);
1320 spa_last_synced_txg(spa_t *spa)
1322 return (spa->spa_ubsync.ub_txg);
1326 spa_first_txg(spa_t *spa)
1328 return (spa->spa_first_txg);
1332 spa_syncing_txg(spa_t *spa)
1334 return (spa->spa_syncing_txg);
1338 spa_state(spa_t *spa)
1340 return (spa->spa_state);
1344 spa_load_state(spa_t *spa)
1346 return (spa->spa_load_state);
1350 spa_freeze_txg(spa_t *spa)
1352 return (spa->spa_freeze_txg);
1357 spa_get_asize(spa_t *spa, uint64_t lsize)
1360 * The worst case is single-sector max-parity RAID-Z blocks, in which
1361 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
1362 * times the size; so just assume that. Add to this the fact that
1363 * we can have up to 3 DVAs per bp, and one more factor of 2 because
1364 * the block may be dittoed with up to 3 DVAs by ddt_sync().
1366 return (lsize * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2);
1370 spa_get_dspace(spa_t *spa)
1372 return (spa->spa_dspace);
1376 spa_update_dspace(spa_t *spa)
1378 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1379 ddt_get_dedup_dspace(spa);
1383 * Return the failure mode that has been set to this pool. The default
1384 * behavior will be to block all I/Os when a complete failure occurs.
1387 spa_get_failmode(spa_t *spa)
1389 return (spa->spa_failmode);
1393 spa_suspended(spa_t *spa)
1395 return (spa->spa_suspended);
1399 spa_version(spa_t *spa)
1401 return (spa->spa_ubsync.ub_version);
1405 spa_deflate(spa_t *spa)
1407 return (spa->spa_deflate);
1411 spa_normal_class(spa_t *spa)
1413 return (spa->spa_normal_class);
1417 spa_log_class(spa_t *spa)
1419 return (spa->spa_log_class);
1423 spa_max_replication(spa_t *spa)
1426 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1427 * handle BPs with more than one DVA allocated. Set our max
1428 * replication level accordingly.
1430 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1432 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1436 spa_prev_software_version(spa_t *spa)
1438 return (spa->spa_prev_software_version);
1442 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1444 uint64_t asize = DVA_GET_ASIZE(dva);
1445 uint64_t dsize = asize;
1447 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1449 if (asize != 0 && spa->spa_deflate) {
1450 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1451 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1458 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1462 for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1463 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1469 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1473 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1475 for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1476 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1478 spa_config_exit(spa, SCL_VDEV, FTAG);
1484 * ==========================================================================
1485 * Initialization and Termination
1486 * ==========================================================================
1490 spa_name_compare(const void *a1, const void *a2)
1492 const spa_t *s1 = a1;
1493 const spa_t *s2 = a2;
1496 s = strcmp(s1->spa_name, s2->spa_name);
1507 return (spa_active_count);
1519 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1520 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1521 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1522 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1524 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1525 offsetof(spa_t, spa_avl));
1527 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1528 offsetof(spa_aux_t, aux_avl));
1530 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1531 offsetof(spa_aux_t, aux_avl));
1533 spa_mode_global = mode;
1540 vdev_cache_stat_init();
1554 vdev_cache_stat_fini();
1561 avl_destroy(&spa_namespace_avl);
1562 avl_destroy(&spa_spare_avl);
1563 avl_destroy(&spa_l2cache_avl);
1565 cv_destroy(&spa_namespace_cv);
1566 mutex_destroy(&spa_namespace_lock);
1567 mutex_destroy(&spa_spare_lock);
1568 mutex_destroy(&spa_l2cache_lock);
1572 * Return whether this pool has slogs. No locking needed.
1573 * It's not a problem if the wrong answer is returned as it's only for
1574 * performance and not correctness
1577 spa_has_slogs(spa_t *spa)
1579 return (spa->spa_log_class->mc_rotor != NULL);
1583 spa_get_log_state(spa_t *spa)
1585 return (spa->spa_log_state);
1589 spa_set_log_state(spa_t *spa, spa_log_state_t state)
1591 spa->spa_log_state = state;
1595 spa_is_root(spa_t *spa)
1597 return (spa->spa_is_root);
1601 spa_writeable(spa_t *spa)
1603 return (!!(spa->spa_mode & FWRITE));
1607 spa_mode(spa_t *spa)
1609 return (spa->spa_mode);
1613 spa_bootfs(spa_t *spa)
1615 return (spa->spa_bootfs);
1619 spa_delegation(spa_t *spa)
1621 return (spa->spa_delegation);
1625 spa_meta_objset(spa_t *spa)
1627 return (spa->spa_meta_objset);
1631 spa_dedup_checksum(spa_t *spa)
1633 return (spa->spa_dedup_checksum);
1637 * Reset pool scan stat per scan pass (or reboot).
1640 spa_scan_stat_init(spa_t *spa)
1642 /* data not stored on disk */
1643 spa->spa_scan_pass_start = gethrestime_sec();
1644 spa->spa_scan_pass_exam = 0;
1645 vdev_scan_stat_init(spa->spa_root_vdev);
1649 * Get scan stats for zpool status reports
1652 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
1654 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
1656 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
1658 bzero(ps, sizeof (pool_scan_stat_t));
1660 /* data stored on disk */
1661 ps->pss_func = scn->scn_phys.scn_func;
1662 ps->pss_start_time = scn->scn_phys.scn_start_time;
1663 ps->pss_end_time = scn->scn_phys.scn_end_time;
1664 ps->pss_to_examine = scn->scn_phys.scn_to_examine;
1665 ps->pss_examined = scn->scn_phys.scn_examined;
1666 ps->pss_to_process = scn->scn_phys.scn_to_process;
1667 ps->pss_processed = scn->scn_phys.scn_processed;
1668 ps->pss_errors = scn->scn_phys.scn_errors;
1669 ps->pss_state = scn->scn_phys.scn_state;
1671 /* data not stored on disk */
1672 ps->pss_pass_start = spa->spa_scan_pass_start;
1673 ps->pss_pass_exam = spa->spa_scan_pass_exam;