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.
23 * Copyright (c) 2011 by Delphix. All rights reserved.
26 #include <sys/zfs_context.h>
27 #include <sys/spa_impl.h>
29 #include <sys/zio_checksum.h>
30 #include <sys/zio_compress.h>
32 #include <sys/dmu_tx.h>
35 #include <sys/vdev_impl.h>
36 #include <sys/metaslab.h>
37 #include <sys/uberblock_impl.h>
40 #include <sys/unique.h>
41 #include <sys/dsl_pool.h>
42 #include <sys/dsl_dir.h>
43 #include <sys/dsl_prop.h>
44 #include <sys/fm/util.h>
45 #include <sys/dsl_scan.h>
46 #include <sys/fs/zfs.h>
47 #include <sys/metaslab_impl.h>
55 * There are four basic locks for managing spa_t structures:
57 * spa_namespace_lock (global mutex)
59 * This lock must be acquired to do any of the following:
61 * - Lookup a spa_t by name
62 * - Add or remove a spa_t from the namespace
63 * - Increase spa_refcount from non-zero
64 * - Check if spa_refcount is zero
66 * - add/remove/attach/detach devices
67 * - Held for the duration of create/destroy/import/export
69 * It does not need to handle recursion. A create or destroy may
70 * reference objects (files or zvols) in other pools, but by
71 * definition they must have an existing reference, and will never need
72 * to lookup a spa_t by name.
74 * spa_refcount (per-spa refcount_t protected by mutex)
76 * This reference count keep track of any active users of the spa_t. The
77 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
78 * the refcount is never really 'zero' - opening a pool implicitly keeps
79 * some references in the DMU. Internally we check against spa_minref, but
80 * present the image of a zero/non-zero value to consumers.
82 * spa_config_lock[] (per-spa array of rwlocks)
84 * This protects the spa_t from config changes, and must be held in
85 * the following circumstances:
87 * - RW_READER to perform I/O to the spa
88 * - RW_WRITER to change the vdev config
90 * The locking order is fairly straightforward:
92 * spa_namespace_lock -> spa_refcount
94 * The namespace lock must be acquired to increase the refcount from 0
95 * or to check if it is zero.
97 * spa_refcount -> spa_config_lock[]
99 * There must be at least one valid reference on the spa_t to acquire
102 * spa_namespace_lock -> spa_config_lock[]
104 * The namespace lock must always be taken before the config lock.
107 * The spa_namespace_lock can be acquired directly and is globally visible.
109 * The namespace is manipulated using the following functions, all of which
110 * require the spa_namespace_lock to be held.
112 * spa_lookup() Lookup a spa_t by name.
114 * spa_add() Create a new spa_t in the namespace.
116 * spa_remove() Remove a spa_t from the namespace. This also
117 * frees up any memory associated with the spa_t.
119 * spa_next() Returns the next spa_t in the system, or the
120 * first if NULL is passed.
122 * spa_evict_all() Shutdown and remove all spa_t structures in
125 * spa_guid_exists() Determine whether a pool/device guid exists.
127 * The spa_refcount is manipulated using the following functions:
129 * spa_open_ref() Adds a reference to the given spa_t. Must be
130 * called with spa_namespace_lock held if the
131 * refcount is currently zero.
133 * spa_close() Remove a reference from the spa_t. This will
134 * not free the spa_t or remove it from the
135 * namespace. No locking is required.
137 * spa_refcount_zero() Returns true if the refcount is currently
138 * zero. Must be called with spa_namespace_lock
141 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
142 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
143 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
145 * To read the configuration, it suffices to hold one of these locks as reader.
146 * To modify the configuration, you must hold all locks as writer. To modify
147 * vdev state without altering the vdev tree's topology (e.g. online/offline),
148 * you must hold SCL_STATE and SCL_ZIO as writer.
150 * We use these distinct config locks to avoid recursive lock entry.
151 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
152 * block allocations (SCL_ALLOC), which may require reading space maps
153 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
155 * The spa config locks cannot be normal rwlocks because we need the
156 * ability to hand off ownership. For example, SCL_ZIO is acquired
157 * by the issuing thread and later released by an interrupt thread.
158 * They do, however, obey the usual write-wanted semantics to prevent
159 * writer (i.e. system administrator) starvation.
161 * The lock acquisition rules are as follows:
164 * Protects changes to the vdev tree topology, such as vdev
165 * add/remove/attach/detach. Protects the dirty config list
166 * (spa_config_dirty_list) and the set of spares and l2arc devices.
169 * Protects changes to pool state and vdev state, such as vdev
170 * online/offline/fault/degrade/clear. Protects the dirty state list
171 * (spa_state_dirty_list) and global pool state (spa_state).
174 * Protects changes to metaslab groups and classes.
175 * Held as reader by metaslab_alloc() and metaslab_claim().
178 * Held by bp-level zios (those which have no io_vd upon entry)
179 * to prevent changes to the vdev tree. The bp-level zio implicitly
180 * protects all of its vdev child zios, which do not hold SCL_ZIO.
183 * Protects changes to metaslab groups and classes.
184 * Held as reader by metaslab_free(). SCL_FREE is distinct from
185 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
186 * blocks in zio_done() while another i/o that holds either
187 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
190 * Held as reader to prevent changes to the vdev tree during trivial
191 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
192 * other locks, and lower than all of them, to ensure that it's safe
193 * to acquire regardless of caller context.
195 * In addition, the following rules apply:
197 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
198 * The lock ordering is SCL_CONFIG > spa_props_lock.
200 * (b) I/O operations on leaf vdevs. For any zio operation that takes
201 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
202 * or zio_write_phys() -- the caller must ensure that the config cannot
203 * cannot change in the interim, and that the vdev cannot be reopened.
204 * SCL_STATE as reader suffices for both.
206 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
208 * spa_vdev_enter() Acquire the namespace lock and the config lock
211 * spa_vdev_exit() Release the config lock, wait for all I/O
212 * to complete, sync the updated configs to the
213 * cache, and release the namespace lock.
215 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
216 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
217 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
219 * spa_rename() is also implemented within this file since is requires
220 * manipulation of the namespace.
223 static avl_tree_t spa_namespace_avl;
224 kmutex_t spa_namespace_lock;
225 static kcondvar_t spa_namespace_cv;
226 static int spa_active_count;
227 int spa_max_replication_override = SPA_DVAS_PER_BP;
229 static kmutex_t spa_spare_lock;
230 static avl_tree_t spa_spare_avl;
231 static kmutex_t spa_l2cache_lock;
232 static avl_tree_t spa_l2cache_avl;
234 kmem_cache_t *spa_buffer_pool;
238 * ==========================================================================
240 * ==========================================================================
243 spa_config_lock_init(spa_t *spa)
247 for (i = 0; i < SCL_LOCKS; i++) {
248 spa_config_lock_t *scl = &spa->spa_config_lock[i];
249 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
250 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
251 refcount_create(&scl->scl_count);
252 scl->scl_writer = NULL;
253 scl->scl_write_wanted = 0;
258 spa_config_lock_destroy(spa_t *spa)
262 for (i = 0; i < SCL_LOCKS; i++) {
263 spa_config_lock_t *scl = &spa->spa_config_lock[i];
264 mutex_destroy(&scl->scl_lock);
265 cv_destroy(&scl->scl_cv);
266 refcount_destroy(&scl->scl_count);
267 ASSERT(scl->scl_writer == NULL);
268 ASSERT(scl->scl_write_wanted == 0);
273 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
277 for (i = 0; i < SCL_LOCKS; i++) {
278 spa_config_lock_t *scl = &spa->spa_config_lock[i];
279 if (!(locks & (1 << i)))
281 mutex_enter(&scl->scl_lock);
282 if (rw == RW_READER) {
283 if (scl->scl_writer || scl->scl_write_wanted) {
284 mutex_exit(&scl->scl_lock);
285 spa_config_exit(spa, locks ^ (1 << i), tag);
289 ASSERT(scl->scl_writer != curthread);
290 if (!refcount_is_zero(&scl->scl_count)) {
291 mutex_exit(&scl->scl_lock);
292 spa_config_exit(spa, locks ^ (1 << i), tag);
295 scl->scl_writer = curthread;
297 (void) refcount_add(&scl->scl_count, tag);
298 mutex_exit(&scl->scl_lock);
304 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
309 for (i = 0; i < SCL_LOCKS; i++) {
310 spa_config_lock_t *scl = &spa->spa_config_lock[i];
311 if (scl->scl_writer == curthread)
312 wlocks_held |= (1 << i);
313 if (!(locks & (1 << i)))
315 mutex_enter(&scl->scl_lock);
316 if (rw == RW_READER) {
317 while (scl->scl_writer || scl->scl_write_wanted) {
318 cv_wait(&scl->scl_cv, &scl->scl_lock);
321 ASSERT(scl->scl_writer != curthread);
322 while (!refcount_is_zero(&scl->scl_count)) {
323 scl->scl_write_wanted++;
324 cv_wait(&scl->scl_cv, &scl->scl_lock);
325 scl->scl_write_wanted--;
327 scl->scl_writer = curthread;
329 (void) refcount_add(&scl->scl_count, tag);
330 mutex_exit(&scl->scl_lock);
332 ASSERT(wlocks_held <= locks);
336 spa_config_exit(spa_t *spa, int locks, void *tag)
340 for (i = SCL_LOCKS - 1; i >= 0; i--) {
341 spa_config_lock_t *scl = &spa->spa_config_lock[i];
342 if (!(locks & (1 << i)))
344 mutex_enter(&scl->scl_lock);
345 ASSERT(!refcount_is_zero(&scl->scl_count));
346 if (refcount_remove(&scl->scl_count, tag) == 0) {
347 ASSERT(scl->scl_writer == NULL ||
348 scl->scl_writer == curthread);
349 scl->scl_writer = NULL; /* OK in either case */
350 cv_broadcast(&scl->scl_cv);
352 mutex_exit(&scl->scl_lock);
357 spa_config_held(spa_t *spa, int locks, krw_t rw)
359 int i, locks_held = 0;
361 for (i = 0; i < SCL_LOCKS; i++) {
362 spa_config_lock_t *scl = &spa->spa_config_lock[i];
363 if (!(locks & (1 << i)))
365 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
366 (rw == RW_WRITER && scl->scl_writer == curthread))
367 locks_held |= 1 << i;
374 * ==========================================================================
375 * SPA namespace functions
376 * ==========================================================================
380 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
381 * Returns NULL if no matching spa_t is found.
384 spa_lookup(const char *name)
386 static spa_t search; /* spa_t is large; don't allocate on stack */
392 ASSERT(MUTEX_HELD(&spa_namespace_lock));
395 * If it's a full dataset name, figure out the pool name and
398 cp = strpbrk(name, "/@");
404 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
405 spa = avl_find(&spa_namespace_avl, &search, &where);
414 * Create an uninitialized spa_t with the given name. Requires
415 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
416 * exist by calling spa_lookup() first.
419 spa_add(const char *name, nvlist_t *config, const char *altroot)
422 spa_config_dirent_t *dp;
425 ASSERT(MUTEX_HELD(&spa_namespace_lock));
427 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP | KM_NODEBUG);
429 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
430 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
431 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
432 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
433 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
434 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
435 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
436 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
437 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
439 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
440 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
441 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
442 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
444 for (t = 0; t < TXG_SIZE; t++)
445 bplist_create(&spa->spa_free_bplist[t]);
447 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
448 spa->spa_state = POOL_STATE_UNINITIALIZED;
449 spa->spa_freeze_txg = UINT64_MAX;
450 spa->spa_final_txg = UINT64_MAX;
451 spa->spa_load_max_txg = UINT64_MAX;
453 spa->spa_proc_state = SPA_PROC_NONE;
455 refcount_create(&spa->spa_refcount);
456 spa_config_lock_init(spa);
458 avl_add(&spa_namespace_avl, spa);
461 * Set the alternate root, if there is one.
464 spa->spa_root = spa_strdup(altroot);
469 * Every pool starts with the default cachefile
471 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
472 offsetof(spa_config_dirent_t, scd_link));
474 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
475 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
476 list_insert_head(&spa->spa_config_list, dp);
478 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
482 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
488 * Removes a spa_t from the namespace, freeing up any memory used. Requires
489 * spa_namespace_lock. This is called only after the spa_t has been closed and
493 spa_remove(spa_t *spa)
495 spa_config_dirent_t *dp;
498 ASSERT(MUTEX_HELD(&spa_namespace_lock));
499 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
501 nvlist_free(spa->spa_config_splitting);
503 avl_remove(&spa_namespace_avl, spa);
504 cv_broadcast(&spa_namespace_cv);
507 spa_strfree(spa->spa_root);
511 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
512 list_remove(&spa->spa_config_list, dp);
513 if (dp->scd_path != NULL)
514 spa_strfree(dp->scd_path);
515 kmem_free(dp, sizeof (spa_config_dirent_t));
518 list_destroy(&spa->spa_config_list);
520 nvlist_free(spa->spa_load_info);
521 spa_config_set(spa, NULL);
523 refcount_destroy(&spa->spa_refcount);
525 spa_config_lock_destroy(spa);
527 for (t = 0; t < TXG_SIZE; t++)
528 bplist_destroy(&spa->spa_free_bplist[t]);
530 cv_destroy(&spa->spa_async_cv);
531 cv_destroy(&spa->spa_proc_cv);
532 cv_destroy(&spa->spa_scrub_io_cv);
533 cv_destroy(&spa->spa_suspend_cv);
535 mutex_destroy(&spa->spa_async_lock);
536 mutex_destroy(&spa->spa_errlist_lock);
537 mutex_destroy(&spa->spa_errlog_lock);
538 mutex_destroy(&spa->spa_history_lock);
539 mutex_destroy(&spa->spa_proc_lock);
540 mutex_destroy(&spa->spa_props_lock);
541 mutex_destroy(&spa->spa_scrub_lock);
542 mutex_destroy(&spa->spa_suspend_lock);
543 mutex_destroy(&spa->spa_vdev_top_lock);
545 kmem_free(spa, sizeof (spa_t));
549 * Given a pool, return the next pool in the namespace, or NULL if there is
550 * none. If 'prev' is NULL, return the first pool.
553 spa_next(spa_t *prev)
555 ASSERT(MUTEX_HELD(&spa_namespace_lock));
558 return (AVL_NEXT(&spa_namespace_avl, prev));
560 return (avl_first(&spa_namespace_avl));
564 * ==========================================================================
565 * SPA refcount functions
566 * ==========================================================================
570 * Add a reference to the given spa_t. Must have at least one reference, or
571 * have the namespace lock held.
574 spa_open_ref(spa_t *spa, void *tag)
576 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
577 MUTEX_HELD(&spa_namespace_lock));
578 (void) refcount_add(&spa->spa_refcount, tag);
582 * Remove a reference to the given spa_t. Must have at least one reference, or
583 * have the namespace lock held.
586 spa_close(spa_t *spa, void *tag)
588 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
589 MUTEX_HELD(&spa_namespace_lock));
590 (void) refcount_remove(&spa->spa_refcount, tag);
594 * Check to see if the spa refcount is zero. Must be called with
595 * spa_namespace_lock held. We really compare against spa_minref, which is the
596 * number of references acquired when opening a pool
599 spa_refcount_zero(spa_t *spa)
601 ASSERT(MUTEX_HELD(&spa_namespace_lock));
603 return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
607 * ==========================================================================
608 * SPA spare and l2cache tracking
609 * ==========================================================================
613 * Hot spares and cache devices are tracked using the same code below,
614 * for 'auxiliary' devices.
617 typedef struct spa_aux {
625 spa_aux_compare(const void *a, const void *b)
627 const spa_aux_t *sa = a;
628 const spa_aux_t *sb = b;
630 if (sa->aux_guid < sb->aux_guid)
632 else if (sa->aux_guid > sb->aux_guid)
639 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
645 search.aux_guid = vd->vdev_guid;
646 if ((aux = avl_find(avl, &search, &where)) != NULL) {
649 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
650 aux->aux_guid = vd->vdev_guid;
652 avl_insert(avl, aux, where);
657 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
663 search.aux_guid = vd->vdev_guid;
664 aux = avl_find(avl, &search, &where);
668 if (--aux->aux_count == 0) {
669 avl_remove(avl, aux);
670 kmem_free(aux, sizeof (spa_aux_t));
671 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
672 aux->aux_pool = 0ULL;
677 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
679 spa_aux_t search, *found;
681 search.aux_guid = guid;
682 found = avl_find(avl, &search, NULL);
686 *pool = found->aux_pool;
693 *refcnt = found->aux_count;
698 return (found != NULL);
702 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
704 spa_aux_t search, *found;
707 search.aux_guid = vd->vdev_guid;
708 found = avl_find(avl, &search, &where);
709 ASSERT(found != NULL);
710 ASSERT(found->aux_pool == 0ULL);
712 found->aux_pool = spa_guid(vd->vdev_spa);
716 * Spares are tracked globally due to the following constraints:
718 * - A spare may be part of multiple pools.
719 * - A spare may be added to a pool even if it's actively in use within
721 * - A spare in use in any pool can only be the source of a replacement if
722 * the target is a spare in the same pool.
724 * We keep track of all spares on the system through the use of a reference
725 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
726 * spare, then we bump the reference count in the AVL tree. In addition, we set
727 * the 'vdev_isspare' member to indicate that the device is a spare (active or
728 * inactive). When a spare is made active (used to replace a device in the
729 * pool), we also keep track of which pool its been made a part of.
731 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
732 * called under the spa_namespace lock as part of vdev reconfiguration. The
733 * separate spare lock exists for the status query path, which does not need to
734 * be completely consistent with respect to other vdev configuration changes.
738 spa_spare_compare(const void *a, const void *b)
740 return (spa_aux_compare(a, b));
744 spa_spare_add(vdev_t *vd)
746 mutex_enter(&spa_spare_lock);
747 ASSERT(!vd->vdev_isspare);
748 spa_aux_add(vd, &spa_spare_avl);
749 vd->vdev_isspare = B_TRUE;
750 mutex_exit(&spa_spare_lock);
754 spa_spare_remove(vdev_t *vd)
756 mutex_enter(&spa_spare_lock);
757 ASSERT(vd->vdev_isspare);
758 spa_aux_remove(vd, &spa_spare_avl);
759 vd->vdev_isspare = B_FALSE;
760 mutex_exit(&spa_spare_lock);
764 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
768 mutex_enter(&spa_spare_lock);
769 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
770 mutex_exit(&spa_spare_lock);
776 spa_spare_activate(vdev_t *vd)
778 mutex_enter(&spa_spare_lock);
779 ASSERT(vd->vdev_isspare);
780 spa_aux_activate(vd, &spa_spare_avl);
781 mutex_exit(&spa_spare_lock);
785 * Level 2 ARC devices are tracked globally for the same reasons as spares.
786 * Cache devices currently only support one pool per cache device, and so
787 * for these devices the aux reference count is currently unused beyond 1.
791 spa_l2cache_compare(const void *a, const void *b)
793 return (spa_aux_compare(a, b));
797 spa_l2cache_add(vdev_t *vd)
799 mutex_enter(&spa_l2cache_lock);
800 ASSERT(!vd->vdev_isl2cache);
801 spa_aux_add(vd, &spa_l2cache_avl);
802 vd->vdev_isl2cache = B_TRUE;
803 mutex_exit(&spa_l2cache_lock);
807 spa_l2cache_remove(vdev_t *vd)
809 mutex_enter(&spa_l2cache_lock);
810 ASSERT(vd->vdev_isl2cache);
811 spa_aux_remove(vd, &spa_l2cache_avl);
812 vd->vdev_isl2cache = B_FALSE;
813 mutex_exit(&spa_l2cache_lock);
817 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
821 mutex_enter(&spa_l2cache_lock);
822 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
823 mutex_exit(&spa_l2cache_lock);
829 spa_l2cache_activate(vdev_t *vd)
831 mutex_enter(&spa_l2cache_lock);
832 ASSERT(vd->vdev_isl2cache);
833 spa_aux_activate(vd, &spa_l2cache_avl);
834 mutex_exit(&spa_l2cache_lock);
838 * ==========================================================================
840 * ==========================================================================
844 * Lock the given spa_t for the purpose of adding or removing a vdev.
845 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
846 * It returns the next transaction group for the spa_t.
849 spa_vdev_enter(spa_t *spa)
851 mutex_enter(&spa->spa_vdev_top_lock);
852 mutex_enter(&spa_namespace_lock);
853 return (spa_vdev_config_enter(spa));
857 * Internal implementation for spa_vdev_enter(). Used when a vdev
858 * operation requires multiple syncs (i.e. removing a device) while
859 * keeping the spa_namespace_lock held.
862 spa_vdev_config_enter(spa_t *spa)
864 ASSERT(MUTEX_HELD(&spa_namespace_lock));
866 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
868 return (spa_last_synced_txg(spa) + 1);
872 * Used in combination with spa_vdev_config_enter() to allow the syncing
873 * of multiple transactions without releasing the spa_namespace_lock.
876 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
878 int config_changed = B_FALSE;
880 ASSERT(MUTEX_HELD(&spa_namespace_lock));
881 ASSERT(txg > spa_last_synced_txg(spa));
883 spa->spa_pending_vdev = NULL;
888 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
890 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
891 config_changed = B_TRUE;
892 spa->spa_config_generation++;
896 * Verify the metaslab classes.
898 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
899 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
901 spa_config_exit(spa, SCL_ALL, spa);
904 * Panic the system if the specified tag requires it. This
905 * is useful for ensuring that configurations are updated
908 if (zio_injection_enabled)
909 zio_handle_panic_injection(spa, tag, 0);
912 * Note: this txg_wait_synced() is important because it ensures
913 * that there won't be more than one config change per txg.
914 * This allows us to use the txg as the generation number.
917 txg_wait_synced(spa->spa_dsl_pool, txg);
920 ASSERT(!vd->vdev_detached || vd->vdev_dtl_smo.smo_object == 0);
921 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
923 spa_config_exit(spa, SCL_ALL, spa);
927 * If the config changed, update the config cache.
930 spa_config_sync(spa, B_FALSE, B_TRUE);
934 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
935 * locking of spa_vdev_enter(), we also want make sure the transactions have
936 * synced to disk, and then update the global configuration cache with the new
940 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
942 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
943 mutex_exit(&spa_namespace_lock);
944 mutex_exit(&spa->spa_vdev_top_lock);
950 * Lock the given spa_t for the purpose of changing vdev state.
953 spa_vdev_state_enter(spa_t *spa, int oplocks)
955 int locks = SCL_STATE_ALL | oplocks;
958 * Root pools may need to read of the underlying devfs filesystem
959 * when opening up a vdev. Unfortunately if we're holding the
960 * SCL_ZIO lock it will result in a deadlock when we try to issue
961 * the read from the root filesystem. Instead we "prefetch"
962 * the associated vnodes that we need prior to opening the
963 * underlying devices and cache them so that we can prevent
964 * any I/O when we are doing the actual open.
966 if (spa_is_root(spa)) {
967 int low = locks & ~(SCL_ZIO - 1);
968 int high = locks & ~low;
970 spa_config_enter(spa, high, spa, RW_WRITER);
971 vdev_hold(spa->spa_root_vdev);
972 spa_config_enter(spa, low, spa, RW_WRITER);
974 spa_config_enter(spa, locks, spa, RW_WRITER);
976 spa->spa_vdev_locks = locks;
980 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
982 boolean_t config_changed = B_FALSE;
984 if (vd != NULL || error == 0)
985 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
989 vdev_state_dirty(vd->vdev_top);
990 config_changed = B_TRUE;
991 spa->spa_config_generation++;
994 if (spa_is_root(spa))
995 vdev_rele(spa->spa_root_vdev);
997 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
998 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1001 * If anything changed, wait for it to sync. This ensures that,
1002 * from the system administrator's perspective, zpool(1M) commands
1003 * are synchronous. This is important for things like zpool offline:
1004 * when the command completes, you expect no further I/O from ZFS.
1007 txg_wait_synced(spa->spa_dsl_pool, 0);
1010 * If the config changed, update the config cache.
1012 if (config_changed) {
1013 mutex_enter(&spa_namespace_lock);
1014 spa_config_sync(spa, B_FALSE, B_TRUE);
1015 mutex_exit(&spa_namespace_lock);
1022 * ==========================================================================
1023 * Miscellaneous functions
1024 * ==========================================================================
1031 spa_rename(const char *name, const char *newname)
1037 * Lookup the spa_t and grab the config lock for writing. We need to
1038 * actually open the pool so that we can sync out the necessary labels.
1039 * It's OK to call spa_open() with the namespace lock held because we
1040 * allow recursive calls for other reasons.
1042 mutex_enter(&spa_namespace_lock);
1043 if ((err = spa_open(name, &spa, FTAG)) != 0) {
1044 mutex_exit(&spa_namespace_lock);
1048 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1050 avl_remove(&spa_namespace_avl, spa);
1051 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1052 avl_add(&spa_namespace_avl, spa);
1055 * Sync all labels to disk with the new names by marking the root vdev
1056 * dirty and waiting for it to sync. It will pick up the new pool name
1059 vdev_config_dirty(spa->spa_root_vdev);
1061 spa_config_exit(spa, SCL_ALL, FTAG);
1063 txg_wait_synced(spa->spa_dsl_pool, 0);
1066 * Sync the updated config cache.
1068 spa_config_sync(spa, B_FALSE, B_TRUE);
1070 spa_close(spa, FTAG);
1072 mutex_exit(&spa_namespace_lock);
1078 * Return the spa_t associated with given pool_guid, if it exists. If
1079 * device_guid is non-zero, determine whether the pool exists *and* contains
1080 * a device with the specified device_guid.
1083 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1086 avl_tree_t *t = &spa_namespace_avl;
1088 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1090 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1091 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1093 if (spa->spa_root_vdev == NULL)
1095 if (spa_guid(spa) == pool_guid) {
1096 if (device_guid == 0)
1099 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1100 device_guid) != NULL)
1104 * Check any devices we may be in the process of adding.
1106 if (spa->spa_pending_vdev) {
1107 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1108 device_guid) != NULL)
1118 * Determine whether a pool with the given pool_guid exists.
1121 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1123 return (spa_by_guid(pool_guid, device_guid) != NULL);
1127 spa_strdup(const char *s)
1133 new = kmem_alloc(len + 1, KM_SLEEP);
1141 spa_strfree(char *s)
1143 kmem_free(s, strlen(s) + 1);
1147 spa_get_random(uint64_t range)
1153 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1159 spa_generate_guid(spa_t *spa)
1161 uint64_t guid = spa_get_random(-1ULL);
1164 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1165 guid = spa_get_random(-1ULL);
1167 while (guid == 0 || spa_guid_exists(guid, 0))
1168 guid = spa_get_random(-1ULL);
1175 sprintf_blkptr(char *buf, const blkptr_t *bp)
1178 char *checksum = NULL;
1179 char *compress = NULL;
1182 type = dmu_ot[BP_GET_TYPE(bp)].ot_name;
1183 checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1184 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1187 SPRINTF_BLKPTR(snprintf, ' ', buf, bp, type, checksum, compress);
1191 spa_freeze(spa_t *spa)
1193 uint64_t freeze_txg = 0;
1195 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1196 if (spa->spa_freeze_txg == UINT64_MAX) {
1197 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1198 spa->spa_freeze_txg = freeze_txg;
1200 spa_config_exit(spa, SCL_ALL, FTAG);
1201 if (freeze_txg != 0)
1202 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1206 * This is a stripped-down version of strtoull, suitable only for converting
1207 * lowercase hexidecimal numbers that don't overflow.
1210 strtonum(const char *str, char **nptr)
1216 while ((c = *str) != '\0') {
1217 if (c >= '0' && c <= '9')
1219 else if (c >= 'a' && c <= 'f')
1220 digit = 10 + c - 'a';
1231 *nptr = (char *)str;
1237 * ==========================================================================
1238 * Accessor functions
1239 * ==========================================================================
1243 spa_shutting_down(spa_t *spa)
1245 return (spa->spa_async_suspended);
1249 spa_get_dsl(spa_t *spa)
1251 return (spa->spa_dsl_pool);
1255 spa_get_rootblkptr(spa_t *spa)
1257 return (&spa->spa_ubsync.ub_rootbp);
1261 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1263 spa->spa_uberblock.ub_rootbp = *bp;
1267 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1269 if (spa->spa_root == NULL)
1272 (void) strncpy(buf, spa->spa_root, buflen);
1276 spa_sync_pass(spa_t *spa)
1278 return (spa->spa_sync_pass);
1282 spa_name(spa_t *spa)
1284 return (spa->spa_name);
1288 spa_guid(spa_t *spa)
1291 * If we fail to parse the config during spa_load(), we can go through
1292 * the error path (which posts an ereport) and end up here with no root
1293 * vdev. We stash the original pool guid in 'spa_load_guid' to handle
1296 if (spa->spa_root_vdev != NULL)
1297 return (spa->spa_root_vdev->vdev_guid);
1299 return (spa->spa_load_guid);
1303 spa_last_synced_txg(spa_t *spa)
1305 return (spa->spa_ubsync.ub_txg);
1309 spa_first_txg(spa_t *spa)
1311 return (spa->spa_first_txg);
1315 spa_syncing_txg(spa_t *spa)
1317 return (spa->spa_syncing_txg);
1321 spa_state(spa_t *spa)
1323 return (spa->spa_state);
1327 spa_load_state(spa_t *spa)
1329 return (spa->spa_load_state);
1333 spa_freeze_txg(spa_t *spa)
1335 return (spa->spa_freeze_txg);
1340 spa_get_asize(spa_t *spa, uint64_t lsize)
1343 * The worst case is single-sector max-parity RAID-Z blocks, in which
1344 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
1345 * times the size; so just assume that. Add to this the fact that
1346 * we can have up to 3 DVAs per bp, and one more factor of 2 because
1347 * the block may be dittoed with up to 3 DVAs by ddt_sync().
1349 return (lsize * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2);
1353 spa_get_dspace(spa_t *spa)
1355 return (spa->spa_dspace);
1359 spa_update_dspace(spa_t *spa)
1361 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1362 ddt_get_dedup_dspace(spa);
1366 * Return the failure mode that has been set to this pool. The default
1367 * behavior will be to block all I/Os when a complete failure occurs.
1370 spa_get_failmode(spa_t *spa)
1372 return (spa->spa_failmode);
1376 spa_suspended(spa_t *spa)
1378 return (spa->spa_suspended);
1382 spa_version(spa_t *spa)
1384 return (spa->spa_ubsync.ub_version);
1388 spa_deflate(spa_t *spa)
1390 return (spa->spa_deflate);
1394 spa_normal_class(spa_t *spa)
1396 return (spa->spa_normal_class);
1400 spa_log_class(spa_t *spa)
1402 return (spa->spa_log_class);
1406 spa_max_replication(spa_t *spa)
1409 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1410 * handle BPs with more than one DVA allocated. Set our max
1411 * replication level accordingly.
1413 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1415 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1419 spa_prev_software_version(spa_t *spa)
1421 return (spa->spa_prev_software_version);
1425 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1427 uint64_t asize = DVA_GET_ASIZE(dva);
1428 uint64_t dsize = asize;
1430 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1432 if (asize != 0 && spa->spa_deflate) {
1433 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1434 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1441 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1446 for (d = 0; d < SPA_DVAS_PER_BP; d++)
1447 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1453 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1458 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1460 for (d = 0; d < SPA_DVAS_PER_BP; d++)
1461 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1463 spa_config_exit(spa, SCL_VDEV, FTAG);
1469 * ==========================================================================
1470 * Initialization and Termination
1471 * ==========================================================================
1475 spa_name_compare(const void *a1, const void *a2)
1477 const spa_t *s1 = a1;
1478 const spa_t *s2 = a2;
1481 s = strcmp(s1->spa_name, s2->spa_name);
1498 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1499 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1500 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1501 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1503 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1504 offsetof(spa_t, spa_avl));
1506 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1507 offsetof(spa_aux_t, aux_avl));
1509 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1510 offsetof(spa_aux_t, aux_avl));
1512 spa_mode_global = mode;
1520 vdev_cache_stat_init();
1534 vdev_cache_stat_fini();
1542 avl_destroy(&spa_namespace_avl);
1543 avl_destroy(&spa_spare_avl);
1544 avl_destroy(&spa_l2cache_avl);
1546 cv_destroy(&spa_namespace_cv);
1547 mutex_destroy(&spa_namespace_lock);
1548 mutex_destroy(&spa_spare_lock);
1549 mutex_destroy(&spa_l2cache_lock);
1553 * Return whether this pool has slogs. No locking needed.
1554 * It's not a problem if the wrong answer is returned as it's only for
1555 * performance and not correctness
1558 spa_has_slogs(spa_t *spa)
1560 return (spa->spa_log_class->mc_rotor != NULL);
1564 spa_get_log_state(spa_t *spa)
1566 return (spa->spa_log_state);
1570 spa_set_log_state(spa_t *spa, spa_log_state_t state)
1572 spa->spa_log_state = state;
1576 spa_is_root(spa_t *spa)
1578 return (spa->spa_is_root);
1582 spa_writeable(spa_t *spa)
1584 return (!!(spa->spa_mode & FWRITE));
1588 spa_mode(spa_t *spa)
1590 return (spa->spa_mode);
1594 spa_bootfs(spa_t *spa)
1596 return (spa->spa_bootfs);
1600 spa_delegation(spa_t *spa)
1602 return (spa->spa_delegation);
1606 spa_meta_objset(spa_t *spa)
1608 return (spa->spa_meta_objset);
1612 spa_dedup_checksum(spa_t *spa)
1614 return (spa->spa_dedup_checksum);
1618 * Reset pool scan stat per scan pass (or reboot).
1621 spa_scan_stat_init(spa_t *spa)
1623 /* data not stored on disk */
1624 spa->spa_scan_pass_start = gethrestime_sec();
1625 spa->spa_scan_pass_exam = 0;
1626 vdev_scan_stat_init(spa->spa_root_vdev);
1630 * Get scan stats for zpool status reports
1633 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
1635 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
1637 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
1639 bzero(ps, sizeof (pool_scan_stat_t));
1641 /* data stored on disk */
1642 ps->pss_func = scn->scn_phys.scn_func;
1643 ps->pss_start_time = scn->scn_phys.scn_start_time;
1644 ps->pss_end_time = scn->scn_phys.scn_end_time;
1645 ps->pss_to_examine = scn->scn_phys.scn_to_examine;
1646 ps->pss_examined = scn->scn_phys.scn_examined;
1647 ps->pss_to_process = scn->scn_phys.scn_to_process;
1648 ps->pss_processed = scn->scn_phys.scn_processed;
1649 ps->pss_errors = scn->scn_phys.scn_errors;
1650 ps->pss_state = scn->scn_phys.scn_state;
1652 /* data not stored on disk */
1653 ps->pss_pass_start = spa->spa_scan_pass_start;
1654 ps->pss_pass_exam = spa->spa_scan_pass_exam;
1660 spa_debug_enabled(spa_t *spa)
1662 return (spa->spa_debug);
1665 #if defined(_KERNEL) && defined(HAVE_SPL)
1666 /* Namespace manipulation */
1667 EXPORT_SYMBOL(spa_lookup);
1668 EXPORT_SYMBOL(spa_add);
1669 EXPORT_SYMBOL(spa_remove);
1670 EXPORT_SYMBOL(spa_next);
1672 /* Refcount functions */
1673 EXPORT_SYMBOL(spa_open_ref);
1674 EXPORT_SYMBOL(spa_close);
1675 EXPORT_SYMBOL(spa_refcount_zero);
1677 /* Pool configuration lock */
1678 EXPORT_SYMBOL(spa_config_tryenter);
1679 EXPORT_SYMBOL(spa_config_enter);
1680 EXPORT_SYMBOL(spa_config_exit);
1681 EXPORT_SYMBOL(spa_config_held);
1683 /* Pool vdev add/remove lock */
1684 EXPORT_SYMBOL(spa_vdev_enter);
1685 EXPORT_SYMBOL(spa_vdev_exit);
1687 /* Pool vdev state change lock */
1688 EXPORT_SYMBOL(spa_vdev_state_enter);
1689 EXPORT_SYMBOL(spa_vdev_state_exit);
1691 /* Accessor functions */
1692 EXPORT_SYMBOL(spa_shutting_down);
1693 EXPORT_SYMBOL(spa_get_dsl);
1694 EXPORT_SYMBOL(spa_get_rootblkptr);
1695 EXPORT_SYMBOL(spa_set_rootblkptr);
1696 EXPORT_SYMBOL(spa_altroot);
1697 EXPORT_SYMBOL(spa_sync_pass);
1698 EXPORT_SYMBOL(spa_name);
1699 EXPORT_SYMBOL(spa_guid);
1700 EXPORT_SYMBOL(spa_last_synced_txg);
1701 EXPORT_SYMBOL(spa_first_txg);
1702 EXPORT_SYMBOL(spa_syncing_txg);
1703 EXPORT_SYMBOL(spa_version);
1704 EXPORT_SYMBOL(spa_state);
1705 EXPORT_SYMBOL(spa_load_state);
1706 EXPORT_SYMBOL(spa_freeze_txg);
1707 EXPORT_SYMBOL(spa_get_asize);
1708 EXPORT_SYMBOL(spa_get_dspace);
1709 EXPORT_SYMBOL(spa_update_dspace);
1710 EXPORT_SYMBOL(spa_deflate);
1711 EXPORT_SYMBOL(spa_normal_class);
1712 EXPORT_SYMBOL(spa_log_class);
1713 EXPORT_SYMBOL(spa_max_replication);
1714 EXPORT_SYMBOL(spa_prev_software_version);
1715 EXPORT_SYMBOL(spa_get_failmode);
1716 EXPORT_SYMBOL(spa_suspended);
1717 EXPORT_SYMBOL(spa_bootfs);
1718 EXPORT_SYMBOL(spa_delegation);
1719 EXPORT_SYMBOL(spa_meta_objset);
1721 /* Miscellaneous support routines */
1722 EXPORT_SYMBOL(spa_rename);
1723 EXPORT_SYMBOL(spa_guid_exists);
1724 EXPORT_SYMBOL(spa_strdup);
1725 EXPORT_SYMBOL(spa_strfree);
1726 EXPORT_SYMBOL(spa_get_random);
1727 EXPORT_SYMBOL(spa_generate_guid);
1728 EXPORT_SYMBOL(sprintf_blkptr);
1729 EXPORT_SYMBOL(spa_freeze);
1730 EXPORT_SYMBOL(spa_upgrade);
1731 EXPORT_SYMBOL(spa_evict_all);
1732 EXPORT_SYMBOL(spa_lookup_by_guid);
1733 EXPORT_SYMBOL(spa_has_spare);
1734 EXPORT_SYMBOL(dva_get_dsize_sync);
1735 EXPORT_SYMBOL(bp_get_dsize_sync);
1736 EXPORT_SYMBOL(bp_get_dsize);
1737 EXPORT_SYMBOL(spa_has_slogs);
1738 EXPORT_SYMBOL(spa_is_root);
1739 EXPORT_SYMBOL(spa_writeable);
1740 EXPORT_SYMBOL(spa_mode);
1742 EXPORT_SYMBOL(spa_namespace_lock);