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 2009 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
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/fs/zfs.h>
45 #include <sys/metaslab_impl.h>
46 #include <sys/sunddi.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_dasize(). 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)
317 for (int i = 0; i < SCL_LOCKS; i++) {
318 spa_config_lock_t *scl = &spa->spa_config_lock[i];
319 if (!(locks & (1 << i)))
321 mutex_enter(&scl->scl_lock);
322 if (rw == RW_READER) {
323 while (scl->scl_writer || scl->scl_write_wanted) {
324 cv_wait(&scl->scl_cv, &scl->scl_lock);
327 ASSERT(scl->scl_writer != curthread);
328 while (!refcount_is_zero(&scl->scl_count)) {
329 scl->scl_write_wanted++;
330 cv_wait(&scl->scl_cv, &scl->scl_lock);
331 scl->scl_write_wanted--;
333 scl->scl_writer = curthread;
335 (void) refcount_add(&scl->scl_count, tag);
336 mutex_exit(&scl->scl_lock);
341 spa_config_exit(spa_t *spa, int locks, void *tag)
343 for (int i = SCL_LOCKS - 1; i >= 0; i--) {
344 spa_config_lock_t *scl = &spa->spa_config_lock[i];
345 if (!(locks & (1 << i)))
347 mutex_enter(&scl->scl_lock);
348 ASSERT(!refcount_is_zero(&scl->scl_count));
349 if (refcount_remove(&scl->scl_count, tag) == 0) {
350 ASSERT(scl->scl_writer == NULL ||
351 scl->scl_writer == curthread);
352 scl->scl_writer = NULL; /* OK in either case */
353 cv_broadcast(&scl->scl_cv);
355 mutex_exit(&scl->scl_lock);
360 spa_config_held(spa_t *spa, int locks, krw_t rw)
364 for (int i = 0; i < SCL_LOCKS; i++) {
365 spa_config_lock_t *scl = &spa->spa_config_lock[i];
366 if (!(locks & (1 << i)))
368 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
369 (rw == RW_WRITER && scl->scl_writer == curthread))
370 locks_held |= 1 << i;
377 * ==========================================================================
378 * SPA namespace functions
379 * ==========================================================================
383 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
384 * Returns NULL if no matching spa_t is found.
387 spa_lookup(const char *name)
389 static spa_t search; /* spa_t is large; don't allocate on stack */
395 ASSERT(MUTEX_HELD(&spa_namespace_lock));
398 * If it's a full dataset name, figure out the pool name and
401 cp = strpbrk(name, "/@");
407 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
408 spa = avl_find(&spa_namespace_avl, &search, &where);
417 * Create an uninitialized spa_t with the given name. Requires
418 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
419 * exist by calling spa_lookup() first.
422 spa_add(const char *name, const char *altroot)
425 spa_config_dirent_t *dp;
427 ASSERT(MUTEX_HELD(&spa_namespace_lock));
429 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
431 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
432 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
433 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
434 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
435 mutex_init(&spa->spa_sync_bplist.bpl_lock, NULL, MUTEX_DEFAULT, NULL);
436 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
437 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
439 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
440 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
441 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
443 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
444 spa->spa_state = POOL_STATE_UNINITIALIZED;
445 spa->spa_freeze_txg = UINT64_MAX;
446 spa->spa_final_txg = UINT64_MAX;
448 refcount_create(&spa->spa_refcount);
449 spa_config_lock_init(spa);
451 avl_add(&spa_namespace_avl, spa);
453 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
456 * Set the alternate root, if there is one.
459 spa->spa_root = spa_strdup(altroot);
464 * Every pool starts with the default cachefile
466 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
467 offsetof(spa_config_dirent_t, scd_link));
469 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
470 dp->scd_path = spa_strdup(spa_config_path);
471 list_insert_head(&spa->spa_config_list, dp);
477 * Removes a spa_t from the namespace, freeing up any memory used. Requires
478 * spa_namespace_lock. This is called only after the spa_t has been closed and
482 spa_remove(spa_t *spa)
484 spa_config_dirent_t *dp;
486 ASSERT(MUTEX_HELD(&spa_namespace_lock));
487 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
489 avl_remove(&spa_namespace_avl, spa);
490 cv_broadcast(&spa_namespace_cv);
493 spa_strfree(spa->spa_root);
497 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
498 list_remove(&spa->spa_config_list, dp);
499 if (dp->scd_path != NULL)
500 spa_strfree(dp->scd_path);
501 kmem_free(dp, sizeof (spa_config_dirent_t));
504 list_destroy(&spa->spa_config_list);
506 spa_config_set(spa, NULL);
508 refcount_destroy(&spa->spa_refcount);
510 spa_config_lock_destroy(spa);
512 cv_destroy(&spa->spa_async_cv);
513 cv_destroy(&spa->spa_scrub_io_cv);
514 cv_destroy(&spa->spa_suspend_cv);
516 mutex_destroy(&spa->spa_async_lock);
517 mutex_destroy(&spa->spa_scrub_lock);
518 mutex_destroy(&spa->spa_errlog_lock);
519 mutex_destroy(&spa->spa_errlist_lock);
520 mutex_destroy(&spa->spa_sync_bplist.bpl_lock);
521 mutex_destroy(&spa->spa_history_lock);
522 mutex_destroy(&spa->spa_props_lock);
523 mutex_destroy(&spa->spa_suspend_lock);
525 kmem_free(spa, sizeof (spa_t));
529 * Given a pool, return the next pool in the namespace, or NULL if there is
530 * none. If 'prev' is NULL, return the first pool.
533 spa_next(spa_t *prev)
535 ASSERT(MUTEX_HELD(&spa_namespace_lock));
538 return (AVL_NEXT(&spa_namespace_avl, prev));
540 return (avl_first(&spa_namespace_avl));
544 * ==========================================================================
545 * SPA refcount functions
546 * ==========================================================================
550 * Add a reference to the given spa_t. Must have at least one reference, or
551 * have the namespace lock held.
554 spa_open_ref(spa_t *spa, void *tag)
556 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
557 MUTEX_HELD(&spa_namespace_lock));
558 (void) refcount_add(&spa->spa_refcount, tag);
562 * Remove a reference to the given spa_t. Must have at least one reference, or
563 * have the namespace lock held.
566 spa_close(spa_t *spa, void *tag)
568 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
569 MUTEX_HELD(&spa_namespace_lock));
570 (void) refcount_remove(&spa->spa_refcount, tag);
574 * Check to see if the spa refcount is zero. Must be called with
575 * spa_namespace_lock held. We really compare against spa_minref, which is the
576 * number of references acquired when opening a pool
579 spa_refcount_zero(spa_t *spa)
581 ASSERT(MUTEX_HELD(&spa_namespace_lock));
583 return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
587 * ==========================================================================
588 * SPA spare and l2cache tracking
589 * ==========================================================================
593 * Hot spares and cache devices are tracked using the same code below,
594 * for 'auxiliary' devices.
597 typedef struct spa_aux {
605 spa_aux_compare(const void *a, const void *b)
607 const spa_aux_t *sa = a;
608 const spa_aux_t *sb = b;
610 if (sa->aux_guid < sb->aux_guid)
612 else if (sa->aux_guid > sb->aux_guid)
619 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
625 search.aux_guid = vd->vdev_guid;
626 if ((aux = avl_find(avl, &search, &where)) != NULL) {
629 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
630 aux->aux_guid = vd->vdev_guid;
632 avl_insert(avl, aux, where);
637 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
643 search.aux_guid = vd->vdev_guid;
644 aux = avl_find(avl, &search, &where);
648 if (--aux->aux_count == 0) {
649 avl_remove(avl, aux);
650 kmem_free(aux, sizeof (spa_aux_t));
651 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
652 aux->aux_pool = 0ULL;
657 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
659 spa_aux_t search, *found;
661 search.aux_guid = guid;
662 found = avl_find(avl, &search, NULL);
666 *pool = found->aux_pool;
673 *refcnt = found->aux_count;
678 return (found != NULL);
682 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
684 spa_aux_t search, *found;
687 search.aux_guid = vd->vdev_guid;
688 found = avl_find(avl, &search, &where);
689 ASSERT(found != NULL);
690 ASSERT(found->aux_pool == 0ULL);
692 found->aux_pool = spa_guid(vd->vdev_spa);
696 * Spares are tracked globally due to the following constraints:
698 * - A spare may be part of multiple pools.
699 * - A spare may be added to a pool even if it's actively in use within
701 * - A spare in use in any pool can only be the source of a replacement if
702 * the target is a spare in the same pool.
704 * We keep track of all spares on the system through the use of a reference
705 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
706 * spare, then we bump the reference count in the AVL tree. In addition, we set
707 * the 'vdev_isspare' member to indicate that the device is a spare (active or
708 * inactive). When a spare is made active (used to replace a device in the
709 * pool), we also keep track of which pool its been made a part of.
711 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
712 * called under the spa_namespace lock as part of vdev reconfiguration. The
713 * separate spare lock exists for the status query path, which does not need to
714 * be completely consistent with respect to other vdev configuration changes.
718 spa_spare_compare(const void *a, const void *b)
720 return (spa_aux_compare(a, b));
724 spa_spare_add(vdev_t *vd)
726 mutex_enter(&spa_spare_lock);
727 ASSERT(!vd->vdev_isspare);
728 spa_aux_add(vd, &spa_spare_avl);
729 vd->vdev_isspare = B_TRUE;
730 mutex_exit(&spa_spare_lock);
734 spa_spare_remove(vdev_t *vd)
736 mutex_enter(&spa_spare_lock);
737 ASSERT(vd->vdev_isspare);
738 spa_aux_remove(vd, &spa_spare_avl);
739 vd->vdev_isspare = B_FALSE;
740 mutex_exit(&spa_spare_lock);
744 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
748 mutex_enter(&spa_spare_lock);
749 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
750 mutex_exit(&spa_spare_lock);
756 spa_spare_activate(vdev_t *vd)
758 mutex_enter(&spa_spare_lock);
759 ASSERT(vd->vdev_isspare);
760 spa_aux_activate(vd, &spa_spare_avl);
761 mutex_exit(&spa_spare_lock);
765 * Level 2 ARC devices are tracked globally for the same reasons as spares.
766 * Cache devices currently only support one pool per cache device, and so
767 * for these devices the aux reference count is currently unused beyond 1.
771 spa_l2cache_compare(const void *a, const void *b)
773 return (spa_aux_compare(a, b));
777 spa_l2cache_add(vdev_t *vd)
779 mutex_enter(&spa_l2cache_lock);
780 ASSERT(!vd->vdev_isl2cache);
781 spa_aux_add(vd, &spa_l2cache_avl);
782 vd->vdev_isl2cache = B_TRUE;
783 mutex_exit(&spa_l2cache_lock);
787 spa_l2cache_remove(vdev_t *vd)
789 mutex_enter(&spa_l2cache_lock);
790 ASSERT(vd->vdev_isl2cache);
791 spa_aux_remove(vd, &spa_l2cache_avl);
792 vd->vdev_isl2cache = B_FALSE;
793 mutex_exit(&spa_l2cache_lock);
797 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
801 mutex_enter(&spa_l2cache_lock);
802 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
803 mutex_exit(&spa_l2cache_lock);
809 spa_l2cache_activate(vdev_t *vd)
811 mutex_enter(&spa_l2cache_lock);
812 ASSERT(vd->vdev_isl2cache);
813 spa_aux_activate(vd, &spa_l2cache_avl);
814 mutex_exit(&spa_l2cache_lock);
818 spa_l2cache_space_update(vdev_t *vd, int64_t space, int64_t alloc)
820 vdev_space_update(vd, space, alloc, B_FALSE);
824 * ==========================================================================
826 * ==========================================================================
830 * Lock the given spa_t for the purpose of adding or removing a vdev.
831 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
832 * It returns the next transaction group for the spa_t.
835 spa_vdev_enter(spa_t *spa)
837 mutex_enter(&spa_namespace_lock);
839 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
841 return (spa_last_synced_txg(spa) + 1);
845 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
846 * locking of spa_vdev_enter(), we also want make sure the transactions have
847 * synced to disk, and then update the global configuration cache with the new
851 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
853 int config_changed = B_FALSE;
855 ASSERT(txg > spa_last_synced_txg(spa));
857 spa->spa_pending_vdev = NULL;
862 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
865 * If the config changed, notify the scrub thread that it must restart.
867 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
868 dsl_pool_scrub_restart(spa->spa_dsl_pool);
869 config_changed = B_TRUE;
872 spa_config_exit(spa, SCL_ALL, spa);
875 * Note: this txg_wait_synced() is important because it ensures
876 * that there won't be more than one config change per txg.
877 * This allows us to use the txg as the generation number.
880 txg_wait_synced(spa->spa_dsl_pool, txg);
883 ASSERT(!vd->vdev_detached || vd->vdev_dtl_smo.smo_object == 0);
884 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
886 spa_config_exit(spa, SCL_ALL, spa);
890 * If the config changed, update the config cache.
893 spa_config_sync(spa, B_FALSE, B_TRUE);
895 mutex_exit(&spa_namespace_lock);
901 * Lock the given spa_t for the purpose of changing vdev state.
904 spa_vdev_state_enter(spa_t *spa)
906 spa_config_enter(spa, SCL_STATE_ALL, spa, RW_WRITER);
910 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
913 vdev_state_dirty(vd->vdev_top);
915 spa_config_exit(spa, SCL_STATE_ALL, spa);
918 * If anything changed, wait for it to sync. This ensures that,
919 * from the system administrator's perspective, zpool(1M) commands
920 * are synchronous. This is important for things like zpool offline:
921 * when the command completes, you expect no further I/O from ZFS.
924 txg_wait_synced(spa->spa_dsl_pool, 0);
930 * ==========================================================================
931 * Miscellaneous functions
932 * ==========================================================================
939 spa_rename(const char *name, const char *newname)
945 * Lookup the spa_t and grab the config lock for writing. We need to
946 * actually open the pool so that we can sync out the necessary labels.
947 * It's OK to call spa_open() with the namespace lock held because we
948 * allow recursive calls for other reasons.
950 mutex_enter(&spa_namespace_lock);
951 if ((err = spa_open(name, &spa, FTAG)) != 0) {
952 mutex_exit(&spa_namespace_lock);
956 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
958 avl_remove(&spa_namespace_avl, spa);
959 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
960 avl_add(&spa_namespace_avl, spa);
963 * Sync all labels to disk with the new names by marking the root vdev
964 * dirty and waiting for it to sync. It will pick up the new pool name
967 vdev_config_dirty(spa->spa_root_vdev);
969 spa_config_exit(spa, SCL_ALL, FTAG);
971 txg_wait_synced(spa->spa_dsl_pool, 0);
974 * Sync the updated config cache.
976 spa_config_sync(spa, B_FALSE, B_TRUE);
978 spa_close(spa, FTAG);
980 mutex_exit(&spa_namespace_lock);
987 * Determine whether a pool with given pool_guid exists. If device_guid is
988 * non-zero, determine whether the pool exists *and* contains a device with the
989 * specified device_guid.
992 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
995 avl_tree_t *t = &spa_namespace_avl;
997 ASSERT(MUTEX_HELD(&spa_namespace_lock));
999 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1000 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1002 if (spa->spa_root_vdev == NULL)
1004 if (spa_guid(spa) == pool_guid) {
1005 if (device_guid == 0)
1008 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1009 device_guid) != NULL)
1013 * Check any devices we may be in the process of adding.
1015 if (spa->spa_pending_vdev) {
1016 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1017 device_guid) != NULL)
1023 return (spa != NULL);
1027 spa_strdup(const char *s)
1033 new = kmem_alloc(len + 1, KM_SLEEP);
1041 spa_strfree(char *s)
1043 kmem_free(s, strlen(s) + 1);
1047 spa_get_random(uint64_t range)
1053 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1059 sprintf_blkptr(char *buf, int len, const blkptr_t *bp)
1064 (void) snprintf(buf, len, "<NULL>");
1068 if (BP_IS_HOLE(bp)) {
1069 (void) snprintf(buf, len, "<hole>");
1073 (void) snprintf(buf, len, "[L%llu %s] %llxL/%llxP ",
1074 (u_longlong_t)BP_GET_LEVEL(bp),
1075 dmu_ot[BP_GET_TYPE(bp)].ot_name,
1076 (u_longlong_t)BP_GET_LSIZE(bp),
1077 (u_longlong_t)BP_GET_PSIZE(bp));
1079 for (d = 0; d < BP_GET_NDVAS(bp); d++) {
1080 const dva_t *dva = &bp->blk_dva[d];
1081 (void) snprintf(buf + strlen(buf), len - strlen(buf),
1082 "DVA[%d]=<%llu:%llx:%llx> ", d,
1083 (u_longlong_t)DVA_GET_VDEV(dva),
1084 (u_longlong_t)DVA_GET_OFFSET(dva),
1085 (u_longlong_t)DVA_GET_ASIZE(dva));
1088 (void) snprintf(buf + strlen(buf), len - strlen(buf),
1089 "%s %s %s %s birth=%llu fill=%llu cksum=%llx:%llx:%llx:%llx",
1090 zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name,
1091 zio_compress_table[BP_GET_COMPRESS(bp)].ci_name,
1092 BP_GET_BYTEORDER(bp) == 0 ? "BE" : "LE",
1093 BP_IS_GANG(bp) ? "gang" : "contiguous",
1094 (u_longlong_t)bp->blk_birth,
1095 (u_longlong_t)bp->blk_fill,
1096 (u_longlong_t)bp->blk_cksum.zc_word[0],
1097 (u_longlong_t)bp->blk_cksum.zc_word[1],
1098 (u_longlong_t)bp->blk_cksum.zc_word[2],
1099 (u_longlong_t)bp->blk_cksum.zc_word[3]);
1103 spa_freeze(spa_t *spa)
1105 uint64_t freeze_txg = 0;
1107 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1108 if (spa->spa_freeze_txg == UINT64_MAX) {
1109 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1110 spa->spa_freeze_txg = freeze_txg;
1112 spa_config_exit(spa, SCL_ALL, FTAG);
1113 if (freeze_txg != 0)
1114 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1118 zfs_panic_recover(const char *fmt, ...)
1123 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1128 * This is a stripped-down version of strtoull, suitable only for converting
1129 * lowercase hexidecimal numbers that don't overflow.
1132 zfs_strtonum(const char *str, char **nptr)
1138 while ((c = *str) != '\0') {
1139 if (c >= '0' && c <= '9')
1141 else if (c >= 'a' && c <= 'f')
1142 digit = 10 + c - 'a';
1153 *nptr = (char *)str;
1159 * ==========================================================================
1160 * Accessor functions
1161 * ==========================================================================
1165 spa_shutting_down(spa_t *spa)
1167 return (spa->spa_async_suspended);
1171 spa_get_dsl(spa_t *spa)
1173 return (spa->spa_dsl_pool);
1177 spa_get_rootblkptr(spa_t *spa)
1179 return (&spa->spa_ubsync.ub_rootbp);
1183 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1185 spa->spa_uberblock.ub_rootbp = *bp;
1189 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1191 if (spa->spa_root == NULL)
1194 (void) strncpy(buf, spa->spa_root, buflen);
1198 spa_sync_pass(spa_t *spa)
1200 return (spa->spa_sync_pass);
1204 spa_name(spa_t *spa)
1206 return (spa->spa_name);
1210 spa_guid(spa_t *spa)
1213 * If we fail to parse the config during spa_load(), we can go through
1214 * the error path (which posts an ereport) and end up here with no root
1215 * vdev. We stash the original pool guid in 'spa_load_guid' to handle
1218 if (spa->spa_root_vdev != NULL)
1219 return (spa->spa_root_vdev->vdev_guid);
1221 return (spa->spa_load_guid);
1225 spa_last_synced_txg(spa_t *spa)
1227 return (spa->spa_ubsync.ub_txg);
1231 spa_first_txg(spa_t *spa)
1233 return (spa->spa_first_txg);
1237 spa_state(spa_t *spa)
1239 return (spa->spa_state);
1243 spa_freeze_txg(spa_t *spa)
1245 return (spa->spa_freeze_txg);
1249 * Return how much space is allocated in the pool (ie. sum of all asize)
1252 spa_get_alloc(spa_t *spa)
1254 return (spa->spa_root_vdev->vdev_stat.vs_alloc);
1258 * Return how much (raid-z inflated) space there is in the pool.
1261 spa_get_space(spa_t *spa)
1263 return (spa->spa_root_vdev->vdev_stat.vs_space);
1267 * Return the amount of raid-z-deflated space in the pool.
1270 spa_get_dspace(spa_t *spa)
1272 if (spa->spa_deflate)
1273 return (spa->spa_root_vdev->vdev_stat.vs_dspace);
1275 return (spa->spa_root_vdev->vdev_stat.vs_space);
1280 spa_get_asize(spa_t *spa, uint64_t lsize)
1283 * For now, the worst case is 512-byte RAID-Z blocks, in which
1284 * case the space requirement is exactly 2x; so just assume that.
1285 * Add to this the fact that we can have up to 3 DVAs per bp, and
1286 * we have to multiply by a total of 6x.
1292 * Return the failure mode that has been set to this pool. The default
1293 * behavior will be to block all I/Os when a complete failure occurs.
1296 spa_get_failmode(spa_t *spa)
1298 return (spa->spa_failmode);
1302 spa_suspended(spa_t *spa)
1304 return (spa->spa_suspended);
1308 spa_version(spa_t *spa)
1310 return (spa->spa_ubsync.ub_version);
1314 spa_max_replication(spa_t *spa)
1317 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1318 * handle BPs with more than one DVA allocated. Set our max
1319 * replication level accordingly.
1321 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1323 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1327 bp_get_dasize(spa_t *spa, const blkptr_t *bp)
1331 if (!spa->spa_deflate)
1332 return (BP_GET_ASIZE(bp));
1334 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1335 for (i = 0; i < SPA_DVAS_PER_BP; i++) {
1337 vdev_lookup_top(spa, DVA_GET_VDEV(&bp->blk_dva[i]));
1339 sz += (DVA_GET_ASIZE(&bp->blk_dva[i]) >>
1340 SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1342 spa_config_exit(spa, SCL_VDEV, FTAG);
1347 * ==========================================================================
1348 * Initialization and Termination
1349 * ==========================================================================
1353 spa_name_compare(const void *a1, const void *a2)
1355 const spa_t *s1 = a1;
1356 const spa_t *s2 = a2;
1359 s = strcmp(s1->spa_name, s2->spa_name);
1370 return (spa_active_count);
1382 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1383 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1384 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1385 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1387 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1388 offsetof(spa_t, spa_avl));
1390 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1391 offsetof(spa_aux_t, aux_avl));
1393 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1394 offsetof(spa_aux_t, aux_avl));
1396 spa_mode_global = mode;
1403 vdev_cache_stat_init();
1417 vdev_cache_stat_fini();
1424 avl_destroy(&spa_namespace_avl);
1425 avl_destroy(&spa_spare_avl);
1426 avl_destroy(&spa_l2cache_avl);
1428 cv_destroy(&spa_namespace_cv);
1429 mutex_destroy(&spa_namespace_lock);
1430 mutex_destroy(&spa_spare_lock);
1431 mutex_destroy(&spa_l2cache_lock);
1435 * Return whether this pool has slogs. No locking needed.
1436 * It's not a problem if the wrong answer is returned as it's only for
1437 * performance and not correctness
1440 spa_has_slogs(spa_t *spa)
1442 return (spa->spa_log_class->mc_rotor != NULL);
1446 * Return whether this pool is the root pool.
1449 spa_is_root(spa_t *spa)
1451 return (spa->spa_is_root);
1455 spa_writeable(spa_t *spa)
1457 return (!!(spa->spa_mode & FWRITE));
1461 spa_mode(spa_t *spa)
1463 return (spa->spa_mode);