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.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
27 #include <sys/zfs_context.h>
28 #include <sys/spa_impl.h>
30 #include <sys/zio_checksum.h>
31 #include <sys/zio_compress.h>
33 #include <sys/dmu_tx.h>
36 #include <sys/vdev_impl.h>
37 #include <sys/metaslab.h>
38 #include <sys/uberblock_impl.h>
41 #include <sys/unique.h>
42 #include <sys/dsl_pool.h>
43 #include <sys/dsl_dir.h>
44 #include <sys/dsl_prop.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 /* Everything except dprintf is on by default in debug builds */
239 int zfs_flags = ~ZFS_DEBUG_DPRINTF;
245 * zfs_recover can be set to nonzero to attempt to recover from
246 * otherwise-fatal errors, typically caused by on-disk corruption. When
247 * set, calls to zfs_panic_recover() will turn into warning messages.
250 SYSCTL_DECL(_vfs_zfs);
251 TUNABLE_INT("vfs.zfs.recover", &zfs_recover);
252 SYSCTL_INT(_vfs_zfs, OID_AUTO, recover, CTLFLAG_RDTUN, &zfs_recover, 0,
253 "Try to recover from otherwise-fatal errors.");
257 * ==========================================================================
259 * ==========================================================================
262 spa_config_lock_init(spa_t *spa)
264 for (int i = 0; i < SCL_LOCKS; i++) {
265 spa_config_lock_t *scl = &spa->spa_config_lock[i];
266 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
267 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
268 refcount_create(&scl->scl_count);
269 scl->scl_writer = NULL;
270 scl->scl_write_wanted = 0;
275 spa_config_lock_destroy(spa_t *spa)
277 for (int i = 0; i < SCL_LOCKS; i++) {
278 spa_config_lock_t *scl = &spa->spa_config_lock[i];
279 mutex_destroy(&scl->scl_lock);
280 cv_destroy(&scl->scl_cv);
281 refcount_destroy(&scl->scl_count);
282 ASSERT(scl->scl_writer == NULL);
283 ASSERT(scl->scl_write_wanted == 0);
288 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
290 for (int i = 0; i < SCL_LOCKS; i++) {
291 spa_config_lock_t *scl = &spa->spa_config_lock[i];
292 if (!(locks & (1 << i)))
294 mutex_enter(&scl->scl_lock);
295 if (rw == RW_READER) {
296 if (scl->scl_writer || scl->scl_write_wanted) {
297 mutex_exit(&scl->scl_lock);
298 spa_config_exit(spa, locks ^ (1 << i), tag);
302 ASSERT(scl->scl_writer != curthread);
303 if (!refcount_is_zero(&scl->scl_count)) {
304 mutex_exit(&scl->scl_lock);
305 spa_config_exit(spa, locks ^ (1 << i), tag);
308 scl->scl_writer = curthread;
310 (void) refcount_add(&scl->scl_count, tag);
311 mutex_exit(&scl->scl_lock);
317 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
321 for (int i = 0; i < SCL_LOCKS; i++) {
322 spa_config_lock_t *scl = &spa->spa_config_lock[i];
323 if (scl->scl_writer == curthread)
324 wlocks_held |= (1 << i);
325 if (!(locks & (1 << i)))
327 mutex_enter(&scl->scl_lock);
328 if (rw == RW_READER) {
329 while (scl->scl_writer || scl->scl_write_wanted) {
330 cv_wait(&scl->scl_cv, &scl->scl_lock);
333 ASSERT(scl->scl_writer != curthread);
334 while (!refcount_is_zero(&scl->scl_count)) {
335 scl->scl_write_wanted++;
336 cv_wait(&scl->scl_cv, &scl->scl_lock);
337 scl->scl_write_wanted--;
339 scl->scl_writer = curthread;
341 (void) refcount_add(&scl->scl_count, tag);
342 mutex_exit(&scl->scl_lock);
344 ASSERT(wlocks_held <= locks);
348 spa_config_exit(spa_t *spa, int locks, void *tag)
350 for (int i = SCL_LOCKS - 1; i >= 0; i--) {
351 spa_config_lock_t *scl = &spa->spa_config_lock[i];
352 if (!(locks & (1 << i)))
354 mutex_enter(&scl->scl_lock);
355 ASSERT(!refcount_is_zero(&scl->scl_count));
356 if (refcount_remove(&scl->scl_count, tag) == 0) {
357 ASSERT(scl->scl_writer == NULL ||
358 scl->scl_writer == curthread);
359 scl->scl_writer = NULL; /* OK in either case */
360 cv_broadcast(&scl->scl_cv);
362 mutex_exit(&scl->scl_lock);
367 spa_config_held(spa_t *spa, int locks, krw_t rw)
371 for (int i = 0; i < SCL_LOCKS; i++) {
372 spa_config_lock_t *scl = &spa->spa_config_lock[i];
373 if (!(locks & (1 << i)))
375 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
376 (rw == RW_WRITER && scl->scl_writer == curthread))
377 locks_held |= 1 << i;
384 * ==========================================================================
385 * SPA namespace functions
386 * ==========================================================================
390 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
391 * Returns NULL if no matching spa_t is found.
394 spa_lookup(const char *name)
396 static spa_t search; /* spa_t is large; don't allocate on stack */
402 ASSERT(MUTEX_HELD(&spa_namespace_lock));
405 * If it's a full dataset name, figure out the pool name and
408 cp = strpbrk(name, "/@");
414 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
415 spa = avl_find(&spa_namespace_avl, &search, &where);
424 * Create an uninitialized spa_t with the given name. Requires
425 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
426 * exist by calling spa_lookup() first.
429 spa_add(const char *name, nvlist_t *config, const char *altroot)
432 spa_config_dirent_t *dp;
434 ASSERT(MUTEX_HELD(&spa_namespace_lock));
436 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
438 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
439 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
440 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
441 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
442 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
443 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
444 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
445 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
446 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
448 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
449 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
450 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
451 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
453 for (int t = 0; t < TXG_SIZE; t++)
454 bplist_create(&spa->spa_free_bplist[t]);
456 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
457 spa->spa_state = POOL_STATE_UNINITIALIZED;
458 spa->spa_freeze_txg = UINT64_MAX;
459 spa->spa_final_txg = UINT64_MAX;
460 spa->spa_load_max_txg = UINT64_MAX;
462 spa->spa_proc_state = SPA_PROC_NONE;
464 refcount_create(&spa->spa_refcount);
465 spa_config_lock_init(spa);
467 avl_add(&spa_namespace_avl, spa);
470 * Set the alternate root, if there is one.
473 spa->spa_root = spa_strdup(altroot);
478 * Every pool starts with the default cachefile
480 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
481 offsetof(spa_config_dirent_t, scd_link));
483 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
484 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
485 list_insert_head(&spa->spa_config_list, dp);
487 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
491 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
497 * Removes a spa_t from the namespace, freeing up any memory used. Requires
498 * spa_namespace_lock. This is called only after the spa_t has been closed and
502 spa_remove(spa_t *spa)
504 spa_config_dirent_t *dp;
506 ASSERT(MUTEX_HELD(&spa_namespace_lock));
507 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
509 nvlist_free(spa->spa_config_splitting);
511 avl_remove(&spa_namespace_avl, spa);
512 cv_broadcast(&spa_namespace_cv);
515 spa_strfree(spa->spa_root);
519 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
520 list_remove(&spa->spa_config_list, dp);
521 if (dp->scd_path != NULL)
522 spa_strfree(dp->scd_path);
523 kmem_free(dp, sizeof (spa_config_dirent_t));
526 list_destroy(&spa->spa_config_list);
528 nvlist_free(spa->spa_load_info);
529 spa_config_set(spa, NULL);
531 refcount_destroy(&spa->spa_refcount);
533 spa_config_lock_destroy(spa);
535 for (int t = 0; t < TXG_SIZE; t++)
536 bplist_destroy(&spa->spa_free_bplist[t]);
538 cv_destroy(&spa->spa_async_cv);
539 cv_destroy(&spa->spa_proc_cv);
540 cv_destroy(&spa->spa_scrub_io_cv);
541 cv_destroy(&spa->spa_suspend_cv);
543 mutex_destroy(&spa->spa_async_lock);
544 mutex_destroy(&spa->spa_errlist_lock);
545 mutex_destroy(&spa->spa_errlog_lock);
546 mutex_destroy(&spa->spa_history_lock);
547 mutex_destroy(&spa->spa_proc_lock);
548 mutex_destroy(&spa->spa_props_lock);
549 mutex_destroy(&spa->spa_scrub_lock);
550 mutex_destroy(&spa->spa_suspend_lock);
551 mutex_destroy(&spa->spa_vdev_top_lock);
553 kmem_free(spa, sizeof (spa_t));
557 * Given a pool, return the next pool in the namespace, or NULL if there is
558 * none. If 'prev' is NULL, return the first pool.
561 spa_next(spa_t *prev)
563 ASSERT(MUTEX_HELD(&spa_namespace_lock));
566 return (AVL_NEXT(&spa_namespace_avl, prev));
568 return (avl_first(&spa_namespace_avl));
572 * ==========================================================================
573 * SPA refcount functions
574 * ==========================================================================
578 * Add a reference to the given spa_t. Must have at least one reference, or
579 * have the namespace lock held.
582 spa_open_ref(spa_t *spa, void *tag)
584 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
585 MUTEX_HELD(&spa_namespace_lock));
586 (void) refcount_add(&spa->spa_refcount, tag);
590 * Remove a reference to the given spa_t. Must have at least one reference, or
591 * have the namespace lock held.
594 spa_close(spa_t *spa, void *tag)
596 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
597 MUTEX_HELD(&spa_namespace_lock));
598 (void) refcount_remove(&spa->spa_refcount, tag);
602 * Check to see if the spa refcount is zero. Must be called with
603 * spa_namespace_lock held. We really compare against spa_minref, which is the
604 * number of references acquired when opening a pool
607 spa_refcount_zero(spa_t *spa)
609 ASSERT(MUTEX_HELD(&spa_namespace_lock));
611 return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
615 * ==========================================================================
616 * SPA spare and l2cache tracking
617 * ==========================================================================
621 * Hot spares and cache devices are tracked using the same code below,
622 * for 'auxiliary' devices.
625 typedef struct spa_aux {
633 spa_aux_compare(const void *a, const void *b)
635 const spa_aux_t *sa = a;
636 const spa_aux_t *sb = b;
638 if (sa->aux_guid < sb->aux_guid)
640 else if (sa->aux_guid > sb->aux_guid)
647 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
653 search.aux_guid = vd->vdev_guid;
654 if ((aux = avl_find(avl, &search, &where)) != NULL) {
657 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
658 aux->aux_guid = vd->vdev_guid;
660 avl_insert(avl, aux, where);
665 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
671 search.aux_guid = vd->vdev_guid;
672 aux = avl_find(avl, &search, &where);
676 if (--aux->aux_count == 0) {
677 avl_remove(avl, aux);
678 kmem_free(aux, sizeof (spa_aux_t));
679 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
680 aux->aux_pool = 0ULL;
685 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
687 spa_aux_t search, *found;
689 search.aux_guid = guid;
690 found = avl_find(avl, &search, NULL);
694 *pool = found->aux_pool;
701 *refcnt = found->aux_count;
706 return (found != NULL);
710 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
712 spa_aux_t search, *found;
715 search.aux_guid = vd->vdev_guid;
716 found = avl_find(avl, &search, &where);
717 ASSERT(found != NULL);
718 ASSERT(found->aux_pool == 0ULL);
720 found->aux_pool = spa_guid(vd->vdev_spa);
724 * Spares are tracked globally due to the following constraints:
726 * - A spare may be part of multiple pools.
727 * - A spare may be added to a pool even if it's actively in use within
729 * - A spare in use in any pool can only be the source of a replacement if
730 * the target is a spare in the same pool.
732 * We keep track of all spares on the system through the use of a reference
733 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
734 * spare, then we bump the reference count in the AVL tree. In addition, we set
735 * the 'vdev_isspare' member to indicate that the device is a spare (active or
736 * inactive). When a spare is made active (used to replace a device in the
737 * pool), we also keep track of which pool its been made a part of.
739 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
740 * called under the spa_namespace lock as part of vdev reconfiguration. The
741 * separate spare lock exists for the status query path, which does not need to
742 * be completely consistent with respect to other vdev configuration changes.
746 spa_spare_compare(const void *a, const void *b)
748 return (spa_aux_compare(a, b));
752 spa_spare_add(vdev_t *vd)
754 mutex_enter(&spa_spare_lock);
755 ASSERT(!vd->vdev_isspare);
756 spa_aux_add(vd, &spa_spare_avl);
757 vd->vdev_isspare = B_TRUE;
758 mutex_exit(&spa_spare_lock);
762 spa_spare_remove(vdev_t *vd)
764 mutex_enter(&spa_spare_lock);
765 ASSERT(vd->vdev_isspare);
766 spa_aux_remove(vd, &spa_spare_avl);
767 vd->vdev_isspare = B_FALSE;
768 mutex_exit(&spa_spare_lock);
772 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
776 mutex_enter(&spa_spare_lock);
777 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
778 mutex_exit(&spa_spare_lock);
784 spa_spare_activate(vdev_t *vd)
786 mutex_enter(&spa_spare_lock);
787 ASSERT(vd->vdev_isspare);
788 spa_aux_activate(vd, &spa_spare_avl);
789 mutex_exit(&spa_spare_lock);
793 * Level 2 ARC devices are tracked globally for the same reasons as spares.
794 * Cache devices currently only support one pool per cache device, and so
795 * for these devices the aux reference count is currently unused beyond 1.
799 spa_l2cache_compare(const void *a, const void *b)
801 return (spa_aux_compare(a, b));
805 spa_l2cache_add(vdev_t *vd)
807 mutex_enter(&spa_l2cache_lock);
808 ASSERT(!vd->vdev_isl2cache);
809 spa_aux_add(vd, &spa_l2cache_avl);
810 vd->vdev_isl2cache = B_TRUE;
811 mutex_exit(&spa_l2cache_lock);
815 spa_l2cache_remove(vdev_t *vd)
817 mutex_enter(&spa_l2cache_lock);
818 ASSERT(vd->vdev_isl2cache);
819 spa_aux_remove(vd, &spa_l2cache_avl);
820 vd->vdev_isl2cache = B_FALSE;
821 mutex_exit(&spa_l2cache_lock);
825 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
829 mutex_enter(&spa_l2cache_lock);
830 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
831 mutex_exit(&spa_l2cache_lock);
837 spa_l2cache_activate(vdev_t *vd)
839 mutex_enter(&spa_l2cache_lock);
840 ASSERT(vd->vdev_isl2cache);
841 spa_aux_activate(vd, &spa_l2cache_avl);
842 mutex_exit(&spa_l2cache_lock);
846 * ==========================================================================
848 * ==========================================================================
852 * Lock the given spa_t for the purpose of adding or removing a vdev.
853 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
854 * It returns the next transaction group for the spa_t.
857 spa_vdev_enter(spa_t *spa)
859 mutex_enter(&spa->spa_vdev_top_lock);
860 mutex_enter(&spa_namespace_lock);
861 return (spa_vdev_config_enter(spa));
865 * Internal implementation for spa_vdev_enter(). Used when a vdev
866 * operation requires multiple syncs (i.e. removing a device) while
867 * keeping the spa_namespace_lock held.
870 spa_vdev_config_enter(spa_t *spa)
872 ASSERT(MUTEX_HELD(&spa_namespace_lock));
874 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
876 return (spa_last_synced_txg(spa) + 1);
880 * Used in combination with spa_vdev_config_enter() to allow the syncing
881 * of multiple transactions without releasing the spa_namespace_lock.
884 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
886 ASSERT(MUTEX_HELD(&spa_namespace_lock));
888 int config_changed = B_FALSE;
890 ASSERT(txg > spa_last_synced_txg(spa));
892 spa->spa_pending_vdev = NULL;
897 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
899 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
900 config_changed = B_TRUE;
901 spa->spa_config_generation++;
905 * Verify the metaslab classes.
907 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
908 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
910 spa_config_exit(spa, SCL_ALL, spa);
913 * Panic the system if the specified tag requires it. This
914 * is useful for ensuring that configurations are updated
917 if (zio_injection_enabled)
918 zio_handle_panic_injection(spa, tag, 0);
921 * Note: this txg_wait_synced() is important because it ensures
922 * that there won't be more than one config change per txg.
923 * This allows us to use the txg as the generation number.
926 txg_wait_synced(spa->spa_dsl_pool, txg);
929 ASSERT(!vd->vdev_detached || vd->vdev_dtl_smo.smo_object == 0);
930 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
932 spa_config_exit(spa, SCL_ALL, spa);
936 * If the config changed, update the config cache.
939 spa_config_sync(spa, B_FALSE, B_TRUE);
943 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
944 * locking of spa_vdev_enter(), we also want make sure the transactions have
945 * synced to disk, and then update the global configuration cache with the new
949 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
951 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
952 mutex_exit(&spa_namespace_lock);
953 mutex_exit(&spa->spa_vdev_top_lock);
959 * Lock the given spa_t for the purpose of changing vdev state.
962 spa_vdev_state_enter(spa_t *spa, int oplocks)
964 int locks = SCL_STATE_ALL | oplocks;
967 * Root pools may need to read of the underlying devfs filesystem
968 * when opening up a vdev. Unfortunately if we're holding the
969 * SCL_ZIO lock it will result in a deadlock when we try to issue
970 * the read from the root filesystem. Instead we "prefetch"
971 * the associated vnodes that we need prior to opening the
972 * underlying devices and cache them so that we can prevent
973 * any I/O when we are doing the actual open.
975 if (spa_is_root(spa)) {
976 int low = locks & ~(SCL_ZIO - 1);
977 int high = locks & ~low;
979 spa_config_enter(spa, high, spa, RW_WRITER);
980 vdev_hold(spa->spa_root_vdev);
981 spa_config_enter(spa, low, spa, RW_WRITER);
983 spa_config_enter(spa, locks, spa, RW_WRITER);
985 spa->spa_vdev_locks = locks;
989 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
991 boolean_t config_changed = B_FALSE;
993 if (vd != NULL || error == 0)
994 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
998 vdev_state_dirty(vd->vdev_top);
999 config_changed = B_TRUE;
1000 spa->spa_config_generation++;
1003 if (spa_is_root(spa))
1004 vdev_rele(spa->spa_root_vdev);
1006 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1007 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1010 * If anything changed, wait for it to sync. This ensures that,
1011 * from the system administrator's perspective, zpool(1M) commands
1012 * are synchronous. This is important for things like zpool offline:
1013 * when the command completes, you expect no further I/O from ZFS.
1016 txg_wait_synced(spa->spa_dsl_pool, 0);
1019 * If the config changed, update the config cache.
1021 if (config_changed) {
1022 mutex_enter(&spa_namespace_lock);
1023 spa_config_sync(spa, B_FALSE, B_TRUE);
1024 mutex_exit(&spa_namespace_lock);
1031 * ==========================================================================
1032 * Miscellaneous functions
1033 * ==========================================================================
1040 spa_rename(const char *name, const char *newname)
1046 * Lookup the spa_t and grab the config lock for writing. We need to
1047 * actually open the pool so that we can sync out the necessary labels.
1048 * It's OK to call spa_open() with the namespace lock held because we
1049 * allow recursive calls for other reasons.
1051 mutex_enter(&spa_namespace_lock);
1052 if ((err = spa_open(name, &spa, FTAG)) != 0) {
1053 mutex_exit(&spa_namespace_lock);
1057 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1059 avl_remove(&spa_namespace_avl, spa);
1060 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1061 avl_add(&spa_namespace_avl, spa);
1064 * Sync all labels to disk with the new names by marking the root vdev
1065 * dirty and waiting for it to sync. It will pick up the new pool name
1068 vdev_config_dirty(spa->spa_root_vdev);
1070 spa_config_exit(spa, SCL_ALL, FTAG);
1072 txg_wait_synced(spa->spa_dsl_pool, 0);
1075 * Sync the updated config cache.
1077 spa_config_sync(spa, B_FALSE, B_TRUE);
1079 spa_close(spa, FTAG);
1081 mutex_exit(&spa_namespace_lock);
1087 * Return the spa_t associated with given pool_guid, if it exists. If
1088 * device_guid is non-zero, determine whether the pool exists *and* contains
1089 * a device with the specified device_guid.
1092 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1095 avl_tree_t *t = &spa_namespace_avl;
1097 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1099 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1100 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1102 if (spa->spa_root_vdev == NULL)
1104 if (spa_guid(spa) == pool_guid) {
1105 if (device_guid == 0)
1108 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1109 device_guid) != NULL)
1113 * Check any devices we may be in the process of adding.
1115 if (spa->spa_pending_vdev) {
1116 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1117 device_guid) != NULL)
1127 * Determine whether a pool with the given pool_guid exists.
1130 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1132 return (spa_by_guid(pool_guid, device_guid) != NULL);
1136 spa_strdup(const char *s)
1142 new = kmem_alloc(len + 1, KM_SLEEP);
1150 spa_strfree(char *s)
1152 kmem_free(s, strlen(s) + 1);
1156 spa_get_random(uint64_t range)
1162 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1168 spa_generate_guid(spa_t *spa)
1170 uint64_t guid = spa_get_random(-1ULL);
1173 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1174 guid = spa_get_random(-1ULL);
1176 while (guid == 0 || spa_guid_exists(guid, 0))
1177 guid = spa_get_random(-1ULL);
1184 sprintf_blkptr(char *buf, const blkptr_t *bp)
1187 char *checksum = NULL;
1188 char *compress = NULL;
1191 type = dmu_ot[BP_GET_TYPE(bp)].ot_name;
1192 checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1193 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1196 SPRINTF_BLKPTR(snprintf, ' ', buf, bp, type, checksum, compress);
1200 spa_freeze(spa_t *spa)
1202 uint64_t freeze_txg = 0;
1204 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1205 if (spa->spa_freeze_txg == UINT64_MAX) {
1206 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1207 spa->spa_freeze_txg = freeze_txg;
1209 spa_config_exit(spa, SCL_ALL, FTAG);
1210 if (freeze_txg != 0)
1211 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1215 zfs_panic_recover(const char *fmt, ...)
1220 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1225 * This is a stripped-down version of strtoull, suitable only for converting
1226 * lowercase hexidecimal numbers that don't overflow.
1229 zfs_strtonum(const char *str, char **nptr)
1235 while ((c = *str) != '\0') {
1236 if (c >= '0' && c <= '9')
1238 else if (c >= 'a' && c <= 'f')
1239 digit = 10 + c - 'a';
1250 *nptr = (char *)str;
1256 * ==========================================================================
1257 * Accessor functions
1258 * ==========================================================================
1262 spa_shutting_down(spa_t *spa)
1264 return (spa->spa_async_suspended);
1268 spa_get_dsl(spa_t *spa)
1270 return (spa->spa_dsl_pool);
1274 spa_get_rootblkptr(spa_t *spa)
1276 return (&spa->spa_ubsync.ub_rootbp);
1280 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1282 spa->spa_uberblock.ub_rootbp = *bp;
1286 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1288 if (spa->spa_root == NULL)
1291 (void) strncpy(buf, spa->spa_root, buflen);
1295 spa_sync_pass(spa_t *spa)
1297 return (spa->spa_sync_pass);
1301 spa_name(spa_t *spa)
1303 return (spa->spa_name);
1307 spa_guid(spa_t *spa)
1310 * If we fail to parse the config during spa_load(), we can go through
1311 * the error path (which posts an ereport) and end up here with no root
1312 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1315 if (spa->spa_root_vdev != NULL)
1316 return (spa->spa_root_vdev->vdev_guid);
1318 return (spa->spa_config_guid);
1322 spa_load_guid(spa_t *spa)
1325 * This is a GUID that exists solely as a reference for the
1326 * purposes of the arc. It is generated at load time, and
1327 * is never written to persistent storage.
1329 return (spa->spa_load_guid);
1333 spa_last_synced_txg(spa_t *spa)
1335 return (spa->spa_ubsync.ub_txg);
1339 spa_first_txg(spa_t *spa)
1341 return (spa->spa_first_txg);
1345 spa_syncing_txg(spa_t *spa)
1347 return (spa->spa_syncing_txg);
1351 spa_state(spa_t *spa)
1353 return (spa->spa_state);
1357 spa_load_state(spa_t *spa)
1359 return (spa->spa_load_state);
1363 spa_freeze_txg(spa_t *spa)
1365 return (spa->spa_freeze_txg);
1370 spa_get_asize(spa_t *spa, uint64_t lsize)
1373 * The worst case is single-sector max-parity RAID-Z blocks, in which
1374 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
1375 * times the size; so just assume that. Add to this the fact that
1376 * we can have up to 3 DVAs per bp, and one more factor of 2 because
1377 * the block may be dittoed with up to 3 DVAs by ddt_sync().
1379 return (lsize * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2);
1383 spa_get_dspace(spa_t *spa)
1385 return (spa->spa_dspace);
1389 spa_update_dspace(spa_t *spa)
1391 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1392 ddt_get_dedup_dspace(spa);
1396 * Return the failure mode that has been set to this pool. The default
1397 * behavior will be to block all I/Os when a complete failure occurs.
1400 spa_get_failmode(spa_t *spa)
1402 return (spa->spa_failmode);
1406 spa_suspended(spa_t *spa)
1408 return (spa->spa_suspended);
1412 spa_version(spa_t *spa)
1414 return (spa->spa_ubsync.ub_version);
1418 spa_deflate(spa_t *spa)
1420 return (spa->spa_deflate);
1424 spa_normal_class(spa_t *spa)
1426 return (spa->spa_normal_class);
1430 spa_log_class(spa_t *spa)
1432 return (spa->spa_log_class);
1436 spa_max_replication(spa_t *spa)
1439 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1440 * handle BPs with more than one DVA allocated. Set our max
1441 * replication level accordingly.
1443 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1445 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1449 spa_prev_software_version(spa_t *spa)
1451 return (spa->spa_prev_software_version);
1455 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1457 uint64_t asize = DVA_GET_ASIZE(dva);
1458 uint64_t dsize = asize;
1460 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1462 if (asize != 0 && spa->spa_deflate) {
1463 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1464 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1471 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1475 for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1476 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1482 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1486 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1488 for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1489 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1491 spa_config_exit(spa, SCL_VDEV, FTAG);
1497 * ==========================================================================
1498 * Initialization and Termination
1499 * ==========================================================================
1503 spa_name_compare(const void *a1, const void *a2)
1505 const spa_t *s1 = a1;
1506 const spa_t *s2 = a2;
1509 s = strcmp(s1->spa_name, s2->spa_name);
1520 return (spa_active_count);
1532 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1533 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1534 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1535 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1537 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1538 offsetof(spa_t, spa_avl));
1540 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1541 offsetof(spa_aux_t, aux_avl));
1543 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1544 offsetof(spa_aux_t, aux_avl));
1546 spa_mode_global = mode;
1553 vdev_cache_stat_init();
1567 vdev_cache_stat_fini();
1574 avl_destroy(&spa_namespace_avl);
1575 avl_destroy(&spa_spare_avl);
1576 avl_destroy(&spa_l2cache_avl);
1578 cv_destroy(&spa_namespace_cv);
1579 mutex_destroy(&spa_namespace_lock);
1580 mutex_destroy(&spa_spare_lock);
1581 mutex_destroy(&spa_l2cache_lock);
1585 * Return whether this pool has slogs. No locking needed.
1586 * It's not a problem if the wrong answer is returned as it's only for
1587 * performance and not correctness
1590 spa_has_slogs(spa_t *spa)
1592 return (spa->spa_log_class->mc_rotor != NULL);
1596 spa_get_log_state(spa_t *spa)
1598 return (spa->spa_log_state);
1602 spa_set_log_state(spa_t *spa, spa_log_state_t state)
1604 spa->spa_log_state = state;
1608 spa_is_root(spa_t *spa)
1610 return (spa->spa_is_root);
1614 spa_writeable(spa_t *spa)
1616 return (!!(spa->spa_mode & FWRITE));
1620 spa_mode(spa_t *spa)
1622 return (spa->spa_mode);
1626 spa_bootfs(spa_t *spa)
1628 return (spa->spa_bootfs);
1632 spa_delegation(spa_t *spa)
1634 return (spa->spa_delegation);
1638 spa_meta_objset(spa_t *spa)
1640 return (spa->spa_meta_objset);
1644 spa_dedup_checksum(spa_t *spa)
1646 return (spa->spa_dedup_checksum);
1650 * Reset pool scan stat per scan pass (or reboot).
1653 spa_scan_stat_init(spa_t *spa)
1655 /* data not stored on disk */
1656 spa->spa_scan_pass_start = gethrestime_sec();
1657 spa->spa_scan_pass_exam = 0;
1658 vdev_scan_stat_init(spa->spa_root_vdev);
1662 * Get scan stats for zpool status reports
1665 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
1667 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
1669 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
1671 bzero(ps, sizeof (pool_scan_stat_t));
1673 /* data stored on disk */
1674 ps->pss_func = scn->scn_phys.scn_func;
1675 ps->pss_start_time = scn->scn_phys.scn_start_time;
1676 ps->pss_end_time = scn->scn_phys.scn_end_time;
1677 ps->pss_to_examine = scn->scn_phys.scn_to_examine;
1678 ps->pss_examined = scn->scn_phys.scn_examined;
1679 ps->pss_to_process = scn->scn_phys.scn_to_process;
1680 ps->pss_processed = scn->scn_phys.scn_processed;
1681 ps->pss_errors = scn->scn_phys.scn_errors;
1682 ps->pss_state = scn->scn_phys.scn_state;
1684 /* data not stored on disk */
1685 ps->pss_pass_start = spa->spa_scan_pass_start;
1686 ps->pss_pass_exam = spa->spa_scan_pass_exam;
1692 spa_debug_enabled(spa_t *spa)
1694 return (spa->spa_debug);