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/dsl_scan.h>
45 #include <sys/fs/zfs.h>
46 #include <sys/metaslab_impl.h>
54 * There are four basic locks for managing spa_t structures:
56 * spa_namespace_lock (global mutex)
58 * This lock must be acquired to do any of the following:
60 * - Lookup a spa_t by name
61 * - Add or remove a spa_t from the namespace
62 * - Increase spa_refcount from non-zero
63 * - Check if spa_refcount is zero
65 * - add/remove/attach/detach devices
66 * - Held for the duration of create/destroy/import/export
68 * It does not need to handle recursion. A create or destroy may
69 * reference objects (files or zvols) in other pools, but by
70 * definition they must have an existing reference, and will never need
71 * to lookup a spa_t by name.
73 * spa_refcount (per-spa refcount_t protected by mutex)
75 * This reference count keep track of any active users of the spa_t. The
76 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
77 * the refcount is never really 'zero' - opening a pool implicitly keeps
78 * some references in the DMU. Internally we check against spa_minref, but
79 * present the image of a zero/non-zero value to consumers.
81 * spa_config_lock[] (per-spa array of rwlocks)
83 * This protects the spa_t from config changes, and must be held in
84 * the following circumstances:
86 * - RW_READER to perform I/O to the spa
87 * - RW_WRITER to change the vdev config
89 * The locking order is fairly straightforward:
91 * spa_namespace_lock -> spa_refcount
93 * The namespace lock must be acquired to increase the refcount from 0
94 * or to check if it is zero.
96 * spa_refcount -> spa_config_lock[]
98 * There must be at least one valid reference on the spa_t to acquire
101 * spa_namespace_lock -> spa_config_lock[]
103 * The namespace lock must always be taken before the config lock.
106 * The spa_namespace_lock can be acquired directly and is globally visible.
108 * The namespace is manipulated using the following functions, all of which
109 * require the spa_namespace_lock to be held.
111 * spa_lookup() Lookup a spa_t by name.
113 * spa_add() Create a new spa_t in the namespace.
115 * spa_remove() Remove a spa_t from the namespace. This also
116 * frees up any memory associated with the spa_t.
118 * spa_next() Returns the next spa_t in the system, or the
119 * first if NULL is passed.
121 * spa_evict_all() Shutdown and remove all spa_t structures in
124 * spa_guid_exists() Determine whether a pool/device guid exists.
126 * The spa_refcount is manipulated using the following functions:
128 * spa_open_ref() Adds a reference to the given spa_t. Must be
129 * called with spa_namespace_lock held if the
130 * refcount is currently zero.
132 * spa_close() Remove a reference from the spa_t. This will
133 * not free the spa_t or remove it from the
134 * namespace. No locking is required.
136 * spa_refcount_zero() Returns true if the refcount is currently
137 * zero. Must be called with spa_namespace_lock
140 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
141 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
142 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
144 * To read the configuration, it suffices to hold one of these locks as reader.
145 * To modify the configuration, you must hold all locks as writer. To modify
146 * vdev state without altering the vdev tree's topology (e.g. online/offline),
147 * you must hold SCL_STATE and SCL_ZIO as writer.
149 * We use these distinct config locks to avoid recursive lock entry.
150 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
151 * block allocations (SCL_ALLOC), which may require reading space maps
152 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
154 * The spa config locks cannot be normal rwlocks because we need the
155 * ability to hand off ownership. For example, SCL_ZIO is acquired
156 * by the issuing thread and later released by an interrupt thread.
157 * They do, however, obey the usual write-wanted semantics to prevent
158 * writer (i.e. system administrator) starvation.
160 * The lock acquisition rules are as follows:
163 * Protects changes to the vdev tree topology, such as vdev
164 * add/remove/attach/detach. Protects the dirty config list
165 * (spa_config_dirty_list) and the set of spares and l2arc devices.
168 * Protects changes to pool state and vdev state, such as vdev
169 * online/offline/fault/degrade/clear. Protects the dirty state list
170 * (spa_state_dirty_list) and global pool state (spa_state).
173 * Protects changes to metaslab groups and classes.
174 * Held as reader by metaslab_alloc() and metaslab_claim().
177 * Held by bp-level zios (those which have no io_vd upon entry)
178 * to prevent changes to the vdev tree. The bp-level zio implicitly
179 * protects all of its vdev child zios, which do not hold SCL_ZIO.
182 * Protects changes to metaslab groups and classes.
183 * Held as reader by metaslab_free(). SCL_FREE is distinct from
184 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
185 * blocks in zio_done() while another i/o that holds either
186 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
189 * Held as reader to prevent changes to the vdev tree during trivial
190 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
191 * other locks, and lower than all of them, to ensure that it's safe
192 * to acquire regardless of caller context.
194 * In addition, the following rules apply:
196 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
197 * The lock ordering is SCL_CONFIG > spa_props_lock.
199 * (b) I/O operations on leaf vdevs. For any zio operation that takes
200 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
201 * or zio_write_phys() -- the caller must ensure that the config cannot
202 * cannot change in the interim, and that the vdev cannot be reopened.
203 * SCL_STATE as reader suffices for both.
205 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
207 * spa_vdev_enter() Acquire the namespace lock and the config lock
210 * spa_vdev_exit() Release the config lock, wait for all I/O
211 * to complete, sync the updated configs to the
212 * cache, and release the namespace lock.
214 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
215 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
216 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
218 * spa_rename() is also implemented within this file since is requires
219 * manipulation of the namespace.
222 static avl_tree_t spa_namespace_avl;
223 kmutex_t spa_namespace_lock;
224 static kcondvar_t spa_namespace_cv;
225 static int spa_active_count;
226 int spa_max_replication_override = SPA_DVAS_PER_BP;
228 static kmutex_t spa_spare_lock;
229 static avl_tree_t spa_spare_avl;
230 static kmutex_t spa_l2cache_lock;
231 static avl_tree_t spa_l2cache_avl;
233 kmem_cache_t *spa_buffer_pool;
237 /* Everything except dprintf is on by default in debug builds */
238 int zfs_flags = ~ZFS_DEBUG_DPRINTF;
244 * zfs_recover can be set to nonzero to attempt to recover from
245 * otherwise-fatal errors, typically caused by on-disk corruption. When
246 * set, calls to zfs_panic_recover() will turn into warning messages.
249 SYSCTL_DECL(_vfs_zfs);
250 TUNABLE_INT("vfs.zfs.recover", &zfs_recover);
251 SYSCTL_INT(_vfs_zfs, OID_AUTO, recover, CTLFLAG_RDTUN, &zfs_recover, 0,
252 "Try to recover from otherwise-fatal errors.");
256 * ==========================================================================
258 * ==========================================================================
261 spa_config_lock_init(spa_t *spa)
263 for (int i = 0; i < SCL_LOCKS; i++) {
264 spa_config_lock_t *scl = &spa->spa_config_lock[i];
265 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
266 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
267 refcount_create(&scl->scl_count);
268 scl->scl_writer = NULL;
269 scl->scl_write_wanted = 0;
274 spa_config_lock_destroy(spa_t *spa)
276 for (int i = 0; i < SCL_LOCKS; i++) {
277 spa_config_lock_t *scl = &spa->spa_config_lock[i];
278 mutex_destroy(&scl->scl_lock);
279 cv_destroy(&scl->scl_cv);
280 refcount_destroy(&scl->scl_count);
281 ASSERT(scl->scl_writer == NULL);
282 ASSERT(scl->scl_write_wanted == 0);
287 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
289 for (int i = 0; i < SCL_LOCKS; i++) {
290 spa_config_lock_t *scl = &spa->spa_config_lock[i];
291 if (!(locks & (1 << i)))
293 mutex_enter(&scl->scl_lock);
294 if (rw == RW_READER) {
295 if (scl->scl_writer || scl->scl_write_wanted) {
296 mutex_exit(&scl->scl_lock);
297 spa_config_exit(spa, locks ^ (1 << i), tag);
301 ASSERT(scl->scl_writer != curthread);
302 if (!refcount_is_zero(&scl->scl_count)) {
303 mutex_exit(&scl->scl_lock);
304 spa_config_exit(spa, locks ^ (1 << i), tag);
307 scl->scl_writer = curthread;
309 (void) refcount_add(&scl->scl_count, tag);
310 mutex_exit(&scl->scl_lock);
316 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
320 for (int i = 0; i < SCL_LOCKS; i++) {
321 spa_config_lock_t *scl = &spa->spa_config_lock[i];
322 if (scl->scl_writer == curthread)
323 wlocks_held |= (1 << i);
324 if (!(locks & (1 << i)))
326 mutex_enter(&scl->scl_lock);
327 if (rw == RW_READER) {
328 while (scl->scl_writer || scl->scl_write_wanted) {
329 cv_wait(&scl->scl_cv, &scl->scl_lock);
332 ASSERT(scl->scl_writer != curthread);
333 while (!refcount_is_zero(&scl->scl_count)) {
334 scl->scl_write_wanted++;
335 cv_wait(&scl->scl_cv, &scl->scl_lock);
336 scl->scl_write_wanted--;
338 scl->scl_writer = curthread;
340 (void) refcount_add(&scl->scl_count, tag);
341 mutex_exit(&scl->scl_lock);
343 ASSERT(wlocks_held <= locks);
347 spa_config_exit(spa_t *spa, int locks, void *tag)
349 for (int i = SCL_LOCKS - 1; i >= 0; i--) {
350 spa_config_lock_t *scl = &spa->spa_config_lock[i];
351 if (!(locks & (1 << i)))
353 mutex_enter(&scl->scl_lock);
354 ASSERT(!refcount_is_zero(&scl->scl_count));
355 if (refcount_remove(&scl->scl_count, tag) == 0) {
356 ASSERT(scl->scl_writer == NULL ||
357 scl->scl_writer == curthread);
358 scl->scl_writer = NULL; /* OK in either case */
359 cv_broadcast(&scl->scl_cv);
361 mutex_exit(&scl->scl_lock);
366 spa_config_held(spa_t *spa, int locks, krw_t rw)
370 for (int i = 0; i < SCL_LOCKS; i++) {
371 spa_config_lock_t *scl = &spa->spa_config_lock[i];
372 if (!(locks & (1 << i)))
374 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
375 (rw == RW_WRITER && scl->scl_writer == curthread))
376 locks_held |= 1 << i;
383 * ==========================================================================
384 * SPA namespace functions
385 * ==========================================================================
389 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
390 * Returns NULL if no matching spa_t is found.
393 spa_lookup(const char *name)
395 static spa_t search; /* spa_t is large; don't allocate on stack */
401 ASSERT(MUTEX_HELD(&spa_namespace_lock));
404 * If it's a full dataset name, figure out the pool name and
407 cp = strpbrk(name, "/@");
413 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
414 spa = avl_find(&spa_namespace_avl, &search, &where);
423 * Create an uninitialized spa_t with the given name. Requires
424 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
425 * exist by calling spa_lookup() first.
428 spa_add(const char *name, nvlist_t *config, const char *altroot)
431 spa_config_dirent_t *dp;
433 ASSERT(MUTEX_HELD(&spa_namespace_lock));
435 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
437 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
438 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
439 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
440 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
441 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
442 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
443 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
444 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
445 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
447 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
448 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
449 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
450 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
452 for (int t = 0; t < TXG_SIZE; t++)
453 bplist_create(&spa->spa_free_bplist[t]);
455 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
456 spa->spa_state = POOL_STATE_UNINITIALIZED;
457 spa->spa_freeze_txg = UINT64_MAX;
458 spa->spa_final_txg = UINT64_MAX;
459 spa->spa_load_max_txg = UINT64_MAX;
461 spa->spa_proc_state = SPA_PROC_NONE;
463 refcount_create(&spa->spa_refcount);
464 spa_config_lock_init(spa);
466 avl_add(&spa_namespace_avl, spa);
469 * Set the alternate root, if there is one.
472 spa->spa_root = spa_strdup(altroot);
477 * Every pool starts with the default cachefile
479 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
480 offsetof(spa_config_dirent_t, scd_link));
482 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
483 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
484 list_insert_head(&spa->spa_config_list, dp);
486 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
490 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
496 * Removes a spa_t from the namespace, freeing up any memory used. Requires
497 * spa_namespace_lock. This is called only after the spa_t has been closed and
501 spa_remove(spa_t *spa)
503 spa_config_dirent_t *dp;
505 ASSERT(MUTEX_HELD(&spa_namespace_lock));
506 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
508 nvlist_free(spa->spa_config_splitting);
510 avl_remove(&spa_namespace_avl, spa);
511 cv_broadcast(&spa_namespace_cv);
514 spa_strfree(spa->spa_root);
518 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
519 list_remove(&spa->spa_config_list, dp);
520 if (dp->scd_path != NULL)
521 spa_strfree(dp->scd_path);
522 kmem_free(dp, sizeof (spa_config_dirent_t));
525 list_destroy(&spa->spa_config_list);
527 nvlist_free(spa->spa_load_info);
528 spa_config_set(spa, NULL);
530 refcount_destroy(&spa->spa_refcount);
532 spa_config_lock_destroy(spa);
534 for (int t = 0; t < TXG_SIZE; t++)
535 bplist_destroy(&spa->spa_free_bplist[t]);
537 cv_destroy(&spa->spa_async_cv);
538 cv_destroy(&spa->spa_proc_cv);
539 cv_destroy(&spa->spa_scrub_io_cv);
540 cv_destroy(&spa->spa_suspend_cv);
542 mutex_destroy(&spa->spa_async_lock);
543 mutex_destroy(&spa->spa_errlist_lock);
544 mutex_destroy(&spa->spa_errlog_lock);
545 mutex_destroy(&spa->spa_history_lock);
546 mutex_destroy(&spa->spa_proc_lock);
547 mutex_destroy(&spa->spa_props_lock);
548 mutex_destroy(&spa->spa_scrub_lock);
549 mutex_destroy(&spa->spa_suspend_lock);
550 mutex_destroy(&spa->spa_vdev_top_lock);
552 kmem_free(spa, sizeof (spa_t));
556 * Given a pool, return the next pool in the namespace, or NULL if there is
557 * none. If 'prev' is NULL, return the first pool.
560 spa_next(spa_t *prev)
562 ASSERT(MUTEX_HELD(&spa_namespace_lock));
565 return (AVL_NEXT(&spa_namespace_avl, prev));
567 return (avl_first(&spa_namespace_avl));
571 * ==========================================================================
572 * SPA refcount functions
573 * ==========================================================================
577 * Add a reference to the given spa_t. Must have at least one reference, or
578 * have the namespace lock held.
581 spa_open_ref(spa_t *spa, void *tag)
583 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
584 MUTEX_HELD(&spa_namespace_lock));
585 (void) refcount_add(&spa->spa_refcount, tag);
589 * Remove a reference to the given spa_t. Must have at least one reference, or
590 * have the namespace lock held.
593 spa_close(spa_t *spa, void *tag)
595 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
596 MUTEX_HELD(&spa_namespace_lock));
597 (void) refcount_remove(&spa->spa_refcount, tag);
601 * Check to see if the spa refcount is zero. Must be called with
602 * spa_namespace_lock held. We really compare against spa_minref, which is the
603 * number of references acquired when opening a pool
606 spa_refcount_zero(spa_t *spa)
608 ASSERT(MUTEX_HELD(&spa_namespace_lock));
610 return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
614 * ==========================================================================
615 * SPA spare and l2cache tracking
616 * ==========================================================================
620 * Hot spares and cache devices are tracked using the same code below,
621 * for 'auxiliary' devices.
624 typedef struct spa_aux {
632 spa_aux_compare(const void *a, const void *b)
634 const spa_aux_t *sa = a;
635 const spa_aux_t *sb = b;
637 if (sa->aux_guid < sb->aux_guid)
639 else if (sa->aux_guid > sb->aux_guid)
646 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
652 search.aux_guid = vd->vdev_guid;
653 if ((aux = avl_find(avl, &search, &where)) != NULL) {
656 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
657 aux->aux_guid = vd->vdev_guid;
659 avl_insert(avl, aux, where);
664 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
670 search.aux_guid = vd->vdev_guid;
671 aux = avl_find(avl, &search, &where);
675 if (--aux->aux_count == 0) {
676 avl_remove(avl, aux);
677 kmem_free(aux, sizeof (spa_aux_t));
678 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
679 aux->aux_pool = 0ULL;
684 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
686 spa_aux_t search, *found;
688 search.aux_guid = guid;
689 found = avl_find(avl, &search, NULL);
693 *pool = found->aux_pool;
700 *refcnt = found->aux_count;
705 return (found != NULL);
709 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
711 spa_aux_t search, *found;
714 search.aux_guid = vd->vdev_guid;
715 found = avl_find(avl, &search, &where);
716 ASSERT(found != NULL);
717 ASSERT(found->aux_pool == 0ULL);
719 found->aux_pool = spa_guid(vd->vdev_spa);
723 * Spares are tracked globally due to the following constraints:
725 * - A spare may be part of multiple pools.
726 * - A spare may be added to a pool even if it's actively in use within
728 * - A spare in use in any pool can only be the source of a replacement if
729 * the target is a spare in the same pool.
731 * We keep track of all spares on the system through the use of a reference
732 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
733 * spare, then we bump the reference count in the AVL tree. In addition, we set
734 * the 'vdev_isspare' member to indicate that the device is a spare (active or
735 * inactive). When a spare is made active (used to replace a device in the
736 * pool), we also keep track of which pool its been made a part of.
738 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
739 * called under the spa_namespace lock as part of vdev reconfiguration. The
740 * separate spare lock exists for the status query path, which does not need to
741 * be completely consistent with respect to other vdev configuration changes.
745 spa_spare_compare(const void *a, const void *b)
747 return (spa_aux_compare(a, b));
751 spa_spare_add(vdev_t *vd)
753 mutex_enter(&spa_spare_lock);
754 ASSERT(!vd->vdev_isspare);
755 spa_aux_add(vd, &spa_spare_avl);
756 vd->vdev_isspare = B_TRUE;
757 mutex_exit(&spa_spare_lock);
761 spa_spare_remove(vdev_t *vd)
763 mutex_enter(&spa_spare_lock);
764 ASSERT(vd->vdev_isspare);
765 spa_aux_remove(vd, &spa_spare_avl);
766 vd->vdev_isspare = B_FALSE;
767 mutex_exit(&spa_spare_lock);
771 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
775 mutex_enter(&spa_spare_lock);
776 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
777 mutex_exit(&spa_spare_lock);
783 spa_spare_activate(vdev_t *vd)
785 mutex_enter(&spa_spare_lock);
786 ASSERT(vd->vdev_isspare);
787 spa_aux_activate(vd, &spa_spare_avl);
788 mutex_exit(&spa_spare_lock);
792 * Level 2 ARC devices are tracked globally for the same reasons as spares.
793 * Cache devices currently only support one pool per cache device, and so
794 * for these devices the aux reference count is currently unused beyond 1.
798 spa_l2cache_compare(const void *a, const void *b)
800 return (spa_aux_compare(a, b));
804 spa_l2cache_add(vdev_t *vd)
806 mutex_enter(&spa_l2cache_lock);
807 ASSERT(!vd->vdev_isl2cache);
808 spa_aux_add(vd, &spa_l2cache_avl);
809 vd->vdev_isl2cache = B_TRUE;
810 mutex_exit(&spa_l2cache_lock);
814 spa_l2cache_remove(vdev_t *vd)
816 mutex_enter(&spa_l2cache_lock);
817 ASSERT(vd->vdev_isl2cache);
818 spa_aux_remove(vd, &spa_l2cache_avl);
819 vd->vdev_isl2cache = B_FALSE;
820 mutex_exit(&spa_l2cache_lock);
824 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
828 mutex_enter(&spa_l2cache_lock);
829 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
830 mutex_exit(&spa_l2cache_lock);
836 spa_l2cache_activate(vdev_t *vd)
838 mutex_enter(&spa_l2cache_lock);
839 ASSERT(vd->vdev_isl2cache);
840 spa_aux_activate(vd, &spa_l2cache_avl);
841 mutex_exit(&spa_l2cache_lock);
845 * ==========================================================================
847 * ==========================================================================
851 * Lock the given spa_t for the purpose of adding or removing a vdev.
852 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
853 * It returns the next transaction group for the spa_t.
856 spa_vdev_enter(spa_t *spa)
858 mutex_enter(&spa->spa_vdev_top_lock);
859 mutex_enter(&spa_namespace_lock);
860 return (spa_vdev_config_enter(spa));
864 * Internal implementation for spa_vdev_enter(). Used when a vdev
865 * operation requires multiple syncs (i.e. removing a device) while
866 * keeping the spa_namespace_lock held.
869 spa_vdev_config_enter(spa_t *spa)
871 ASSERT(MUTEX_HELD(&spa_namespace_lock));
873 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
875 return (spa_last_synced_txg(spa) + 1);
879 * Used in combination with spa_vdev_config_enter() to allow the syncing
880 * of multiple transactions without releasing the spa_namespace_lock.
883 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
885 ASSERT(MUTEX_HELD(&spa_namespace_lock));
887 int config_changed = B_FALSE;
889 ASSERT(txg > spa_last_synced_txg(spa));
891 spa->spa_pending_vdev = NULL;
896 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
898 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
899 config_changed = B_TRUE;
900 spa->spa_config_generation++;
904 * Verify the metaslab classes.
906 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
907 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
909 spa_config_exit(spa, SCL_ALL, spa);
912 * Panic the system if the specified tag requires it. This
913 * is useful for ensuring that configurations are updated
916 if (zio_injection_enabled)
917 zio_handle_panic_injection(spa, tag, 0);
920 * Note: this txg_wait_synced() is important because it ensures
921 * that there won't be more than one config change per txg.
922 * This allows us to use the txg as the generation number.
925 txg_wait_synced(spa->spa_dsl_pool, txg);
928 ASSERT(!vd->vdev_detached || vd->vdev_dtl_smo.smo_object == 0);
929 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
931 spa_config_exit(spa, SCL_ALL, spa);
935 * If the config changed, update the config cache.
938 spa_config_sync(spa, B_FALSE, B_TRUE);
942 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
943 * locking of spa_vdev_enter(), we also want make sure the transactions have
944 * synced to disk, and then update the global configuration cache with the new
948 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
950 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
951 mutex_exit(&spa_namespace_lock);
952 mutex_exit(&spa->spa_vdev_top_lock);
958 * Lock the given spa_t for the purpose of changing vdev state.
961 spa_vdev_state_enter(spa_t *spa, int oplocks)
963 int locks = SCL_STATE_ALL | oplocks;
966 * Root pools may need to read of the underlying devfs filesystem
967 * when opening up a vdev. Unfortunately if we're holding the
968 * SCL_ZIO lock it will result in a deadlock when we try to issue
969 * the read from the root filesystem. Instead we "prefetch"
970 * the associated vnodes that we need prior to opening the
971 * underlying devices and cache them so that we can prevent
972 * any I/O when we are doing the actual open.
974 if (spa_is_root(spa)) {
975 int low = locks & ~(SCL_ZIO - 1);
976 int high = locks & ~low;
978 spa_config_enter(spa, high, spa, RW_WRITER);
979 vdev_hold(spa->spa_root_vdev);
980 spa_config_enter(spa, low, spa, RW_WRITER);
982 spa_config_enter(spa, locks, spa, RW_WRITER);
984 spa->spa_vdev_locks = locks;
988 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
990 boolean_t config_changed = B_FALSE;
992 if (vd != NULL || error == 0)
993 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
997 vdev_state_dirty(vd->vdev_top);
998 config_changed = B_TRUE;
999 spa->spa_config_generation++;
1002 if (spa_is_root(spa))
1003 vdev_rele(spa->spa_root_vdev);
1005 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1006 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1009 * If anything changed, wait for it to sync. This ensures that,
1010 * from the system administrator's perspective, zpool(1M) commands
1011 * are synchronous. This is important for things like zpool offline:
1012 * when the command completes, you expect no further I/O from ZFS.
1015 txg_wait_synced(spa->spa_dsl_pool, 0);
1018 * If the config changed, update the config cache.
1020 if (config_changed) {
1021 mutex_enter(&spa_namespace_lock);
1022 spa_config_sync(spa, B_FALSE, B_TRUE);
1023 mutex_exit(&spa_namespace_lock);
1030 * ==========================================================================
1031 * Miscellaneous functions
1032 * ==========================================================================
1039 spa_rename(const char *name, const char *newname)
1045 * Lookup the spa_t and grab the config lock for writing. We need to
1046 * actually open the pool so that we can sync out the necessary labels.
1047 * It's OK to call spa_open() with the namespace lock held because we
1048 * allow recursive calls for other reasons.
1050 mutex_enter(&spa_namespace_lock);
1051 if ((err = spa_open(name, &spa, FTAG)) != 0) {
1052 mutex_exit(&spa_namespace_lock);
1056 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1058 avl_remove(&spa_namespace_avl, spa);
1059 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1060 avl_add(&spa_namespace_avl, spa);
1063 * Sync all labels to disk with the new names by marking the root vdev
1064 * dirty and waiting for it to sync. It will pick up the new pool name
1067 vdev_config_dirty(spa->spa_root_vdev);
1069 spa_config_exit(spa, SCL_ALL, FTAG);
1071 txg_wait_synced(spa->spa_dsl_pool, 0);
1074 * Sync the updated config cache.
1076 spa_config_sync(spa, B_FALSE, B_TRUE);
1078 spa_close(spa, FTAG);
1080 mutex_exit(&spa_namespace_lock);
1086 * Return the spa_t associated with given pool_guid, if it exists. If
1087 * device_guid is non-zero, determine whether the pool exists *and* contains
1088 * a device with the specified device_guid.
1091 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1094 avl_tree_t *t = &spa_namespace_avl;
1096 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1098 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1099 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1101 if (spa->spa_root_vdev == NULL)
1103 if (spa_guid(spa) == pool_guid) {
1104 if (device_guid == 0)
1107 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1108 device_guid) != NULL)
1112 * Check any devices we may be in the process of adding.
1114 if (spa->spa_pending_vdev) {
1115 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1116 device_guid) != NULL)
1126 * Determine whether a pool with the given pool_guid exists.
1129 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1131 return (spa_by_guid(pool_guid, device_guid) != NULL);
1135 spa_strdup(const char *s)
1141 new = kmem_alloc(len + 1, KM_SLEEP);
1149 spa_strfree(char *s)
1151 kmem_free(s, strlen(s) + 1);
1155 spa_get_random(uint64_t range)
1161 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1167 spa_generate_guid(spa_t *spa)
1169 uint64_t guid = spa_get_random(-1ULL);
1172 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1173 guid = spa_get_random(-1ULL);
1175 while (guid == 0 || spa_guid_exists(guid, 0))
1176 guid = spa_get_random(-1ULL);
1183 sprintf_blkptr(char *buf, const blkptr_t *bp)
1186 char *checksum = NULL;
1187 char *compress = NULL;
1190 type = dmu_ot[BP_GET_TYPE(bp)].ot_name;
1191 checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1192 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1195 SPRINTF_BLKPTR(snprintf, ' ', buf, bp, type, checksum, compress);
1199 spa_freeze(spa_t *spa)
1201 uint64_t freeze_txg = 0;
1203 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1204 if (spa->spa_freeze_txg == UINT64_MAX) {
1205 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1206 spa->spa_freeze_txg = freeze_txg;
1208 spa_config_exit(spa, SCL_ALL, FTAG);
1209 if (freeze_txg != 0)
1210 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1214 zfs_panic_recover(const char *fmt, ...)
1219 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1224 * This is a stripped-down version of strtoull, suitable only for converting
1225 * lowercase hexidecimal numbers that don't overflow.
1228 zfs_strtonum(const char *str, char **nptr)
1234 while ((c = *str) != '\0') {
1235 if (c >= '0' && c <= '9')
1237 else if (c >= 'a' && c <= 'f')
1238 digit = 10 + c - 'a';
1249 *nptr = (char *)str;
1255 * ==========================================================================
1256 * Accessor functions
1257 * ==========================================================================
1261 spa_shutting_down(spa_t *spa)
1263 return (spa->spa_async_suspended);
1267 spa_get_dsl(spa_t *spa)
1269 return (spa->spa_dsl_pool);
1273 spa_get_rootblkptr(spa_t *spa)
1275 return (&spa->spa_ubsync.ub_rootbp);
1279 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1281 spa->spa_uberblock.ub_rootbp = *bp;
1285 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1287 if (spa->spa_root == NULL)
1290 (void) strncpy(buf, spa->spa_root, buflen);
1294 spa_sync_pass(spa_t *spa)
1296 return (spa->spa_sync_pass);
1300 spa_name(spa_t *spa)
1302 return (spa->spa_name);
1306 spa_guid(spa_t *spa)
1309 * If we fail to parse the config during spa_load(), we can go through
1310 * the error path (which posts an ereport) and end up here with no root
1311 * vdev. We stash the original pool guid in 'spa_load_guid' to handle
1314 if (spa->spa_root_vdev != NULL)
1315 return (spa->spa_root_vdev->vdev_guid);
1317 return (spa->spa_load_guid);
1321 spa_last_synced_txg(spa_t *spa)
1323 return (spa->spa_ubsync.ub_txg);
1327 spa_first_txg(spa_t *spa)
1329 return (spa->spa_first_txg);
1333 spa_syncing_txg(spa_t *spa)
1335 return (spa->spa_syncing_txg);
1339 spa_state(spa_t *spa)
1341 return (spa->spa_state);
1345 spa_load_state(spa_t *spa)
1347 return (spa->spa_load_state);
1351 spa_freeze_txg(spa_t *spa)
1353 return (spa->spa_freeze_txg);
1358 spa_get_asize(spa_t *spa, uint64_t lsize)
1361 * The worst case is single-sector max-parity RAID-Z blocks, in which
1362 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
1363 * times the size; so just assume that. Add to this the fact that
1364 * we can have up to 3 DVAs per bp, and one more factor of 2 because
1365 * the block may be dittoed with up to 3 DVAs by ddt_sync().
1367 return (lsize * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2);
1371 spa_get_dspace(spa_t *spa)
1373 return (spa->spa_dspace);
1377 spa_update_dspace(spa_t *spa)
1379 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1380 ddt_get_dedup_dspace(spa);
1384 * Return the failure mode that has been set to this pool. The default
1385 * behavior will be to block all I/Os when a complete failure occurs.
1388 spa_get_failmode(spa_t *spa)
1390 return (spa->spa_failmode);
1394 spa_suspended(spa_t *spa)
1396 return (spa->spa_suspended);
1400 spa_version(spa_t *spa)
1402 return (spa->spa_ubsync.ub_version);
1406 spa_deflate(spa_t *spa)
1408 return (spa->spa_deflate);
1412 spa_normal_class(spa_t *spa)
1414 return (spa->spa_normal_class);
1418 spa_log_class(spa_t *spa)
1420 return (spa->spa_log_class);
1424 spa_max_replication(spa_t *spa)
1427 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1428 * handle BPs with more than one DVA allocated. Set our max
1429 * replication level accordingly.
1431 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1433 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1437 spa_prev_software_version(spa_t *spa)
1439 return (spa->spa_prev_software_version);
1443 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1445 uint64_t asize = DVA_GET_ASIZE(dva);
1446 uint64_t dsize = asize;
1448 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1450 if (asize != 0 && spa->spa_deflate) {
1451 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1452 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1459 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1463 for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1464 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1470 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1474 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1476 for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1477 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1479 spa_config_exit(spa, SCL_VDEV, FTAG);
1485 * ==========================================================================
1486 * Initialization and Termination
1487 * ==========================================================================
1491 spa_name_compare(const void *a1, const void *a2)
1493 const spa_t *s1 = a1;
1494 const spa_t *s2 = a2;
1497 s = strcmp(s1->spa_name, s2->spa_name);
1508 return (spa_active_count);
1520 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1521 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1522 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1523 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1525 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1526 offsetof(spa_t, spa_avl));
1528 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1529 offsetof(spa_aux_t, aux_avl));
1531 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1532 offsetof(spa_aux_t, aux_avl));
1534 spa_mode_global = mode;
1541 vdev_cache_stat_init();
1555 vdev_cache_stat_fini();
1562 avl_destroy(&spa_namespace_avl);
1563 avl_destroy(&spa_spare_avl);
1564 avl_destroy(&spa_l2cache_avl);
1566 cv_destroy(&spa_namespace_cv);
1567 mutex_destroy(&spa_namespace_lock);
1568 mutex_destroy(&spa_spare_lock);
1569 mutex_destroy(&spa_l2cache_lock);
1573 * Return whether this pool has slogs. No locking needed.
1574 * It's not a problem if the wrong answer is returned as it's only for
1575 * performance and not correctness
1578 spa_has_slogs(spa_t *spa)
1580 return (spa->spa_log_class->mc_rotor != NULL);
1584 spa_get_log_state(spa_t *spa)
1586 return (spa->spa_log_state);
1590 spa_set_log_state(spa_t *spa, spa_log_state_t state)
1592 spa->spa_log_state = state;
1596 spa_is_root(spa_t *spa)
1598 return (spa->spa_is_root);
1602 spa_writeable(spa_t *spa)
1604 return (!!(spa->spa_mode & FWRITE));
1608 spa_mode(spa_t *spa)
1610 return (spa->spa_mode);
1614 spa_bootfs(spa_t *spa)
1616 return (spa->spa_bootfs);
1620 spa_delegation(spa_t *spa)
1622 return (spa->spa_delegation);
1626 spa_meta_objset(spa_t *spa)
1628 return (spa->spa_meta_objset);
1632 spa_dedup_checksum(spa_t *spa)
1634 return (spa->spa_dedup_checksum);
1638 * Reset pool scan stat per scan pass (or reboot).
1641 spa_scan_stat_init(spa_t *spa)
1643 /* data not stored on disk */
1644 spa->spa_scan_pass_start = gethrestime_sec();
1645 spa->spa_scan_pass_exam = 0;
1646 vdev_scan_stat_init(spa->spa_root_vdev);
1650 * Get scan stats for zpool status reports
1653 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
1655 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
1657 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
1659 bzero(ps, sizeof (pool_scan_stat_t));
1661 /* data stored on disk */
1662 ps->pss_func = scn->scn_phys.scn_func;
1663 ps->pss_start_time = scn->scn_phys.scn_start_time;
1664 ps->pss_end_time = scn->scn_phys.scn_end_time;
1665 ps->pss_to_examine = scn->scn_phys.scn_to_examine;
1666 ps->pss_examined = scn->scn_phys.scn_examined;
1667 ps->pss_to_process = scn->scn_phys.scn_to_process;
1668 ps->pss_processed = scn->scn_phys.scn_processed;
1669 ps->pss_errors = scn->scn_phys.scn_errors;
1670 ps->pss_state = scn->scn_phys.scn_state;
1672 /* data not stored on disk */
1673 ps->pss_pass_start = spa->spa_scan_pass_start;
1674 ps->pss_pass_exam = spa->spa_scan_pass_exam;
1680 spa_debug_enabled(spa_t *spa)
1682 return (spa->spa_debug);