4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
25 #include <sys/zfs_context.h>
26 #include <sys/spa_impl.h>
28 #include <sys/zio_checksum.h>
29 #include <sys/zio_compress.h>
31 #include <sys/dmu_tx.h>
34 #include <sys/vdev_impl.h>
35 #include <sys/metaslab.h>
36 #include <sys/uberblock_impl.h>
39 #include <sys/unique.h>
40 #include <sys/dsl_pool.h>
41 #include <sys/dsl_dir.h>
42 #include <sys/dsl_prop.h>
43 #include <sys/dsl_scan.h>
44 #include <sys/fs/zfs.h>
45 #include <sys/metaslab_impl.h>
53 * There are four basic locks for managing spa_t structures:
55 * spa_namespace_lock (global mutex)
57 * This lock must be acquired to do any of the following:
59 * - Lookup a spa_t by name
60 * - Add or remove a spa_t from the namespace
61 * - Increase spa_refcount from non-zero
62 * - Check if spa_refcount is zero
64 * - add/remove/attach/detach devices
65 * - Held for the duration of create/destroy/import/export
67 * It does not need to handle recursion. A create or destroy may
68 * reference objects (files or zvols) in other pools, but by
69 * definition they must have an existing reference, and will never need
70 * to lookup a spa_t by name.
72 * spa_refcount (per-spa refcount_t protected by mutex)
74 * This reference count keep track of any active users of the spa_t. The
75 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
76 * the refcount is never really 'zero' - opening a pool implicitly keeps
77 * some references in the DMU. Internally we check against spa_minref, but
78 * present the image of a zero/non-zero value to consumers.
80 * spa_config_lock[] (per-spa array of rwlocks)
82 * This protects the spa_t from config changes, and must be held in
83 * the following circumstances:
85 * - RW_READER to perform I/O to the spa
86 * - RW_WRITER to change the vdev config
88 * The locking order is fairly straightforward:
90 * spa_namespace_lock -> spa_refcount
92 * The namespace lock must be acquired to increase the refcount from 0
93 * or to check if it is zero.
95 * spa_refcount -> spa_config_lock[]
97 * There must be at least one valid reference on the spa_t to acquire
100 * spa_namespace_lock -> spa_config_lock[]
102 * The namespace lock must always be taken before the config lock.
105 * The spa_namespace_lock can be acquired directly and is globally visible.
107 * The namespace is manipulated using the following functions, all of which
108 * require the spa_namespace_lock to be held.
110 * spa_lookup() Lookup a spa_t by name.
112 * spa_add() Create a new spa_t in the namespace.
114 * spa_remove() Remove a spa_t from the namespace. This also
115 * frees up any memory associated with the spa_t.
117 * spa_next() Returns the next spa_t in the system, or the
118 * first if NULL is passed.
120 * spa_evict_all() Shutdown and remove all spa_t structures in
123 * spa_guid_exists() Determine whether a pool/device guid exists.
125 * The spa_refcount is manipulated using the following functions:
127 * spa_open_ref() Adds a reference to the given spa_t. Must be
128 * called with spa_namespace_lock held if the
129 * refcount is currently zero.
131 * spa_close() Remove a reference from the spa_t. This will
132 * not free the spa_t or remove it from the
133 * namespace. No locking is required.
135 * spa_refcount_zero() Returns true if the refcount is currently
136 * zero. Must be called with spa_namespace_lock
139 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
140 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
141 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
143 * To read the configuration, it suffices to hold one of these locks as reader.
144 * To modify the configuration, you must hold all locks as writer. To modify
145 * vdev state without altering the vdev tree's topology (e.g. online/offline),
146 * you must hold SCL_STATE and SCL_ZIO as writer.
148 * We use these distinct config locks to avoid recursive lock entry.
149 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
150 * block allocations (SCL_ALLOC), which may require reading space maps
151 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
153 * The spa config locks cannot be normal rwlocks because we need the
154 * ability to hand off ownership. For example, SCL_ZIO is acquired
155 * by the issuing thread and later released by an interrupt thread.
156 * They do, however, obey the usual write-wanted semantics to prevent
157 * writer (i.e. system administrator) starvation.
159 * The lock acquisition rules are as follows:
162 * Protects changes to the vdev tree topology, such as vdev
163 * add/remove/attach/detach. Protects the dirty config list
164 * (spa_config_dirty_list) and the set of spares and l2arc devices.
167 * Protects changes to pool state and vdev state, such as vdev
168 * online/offline/fault/degrade/clear. Protects the dirty state list
169 * (spa_state_dirty_list) and global pool state (spa_state).
172 * Protects changes to metaslab groups and classes.
173 * Held as reader by metaslab_alloc() and metaslab_claim().
176 * Held by bp-level zios (those which have no io_vd upon entry)
177 * to prevent changes to the vdev tree. The bp-level zio implicitly
178 * protects all of its vdev child zios, which do not hold SCL_ZIO.
181 * Protects changes to metaslab groups and classes.
182 * Held as reader by metaslab_free(). SCL_FREE is distinct from
183 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
184 * blocks in zio_done() while another i/o that holds either
185 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
188 * Held as reader to prevent changes to the vdev tree during trivial
189 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
190 * other locks, and lower than all of them, to ensure that it's safe
191 * to acquire regardless of caller context.
193 * In addition, the following rules apply:
195 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
196 * The lock ordering is SCL_CONFIG > spa_props_lock.
198 * (b) I/O operations on leaf vdevs. For any zio operation that takes
199 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
200 * or zio_write_phys() -- the caller must ensure that the config cannot
201 * cannot change in the interim, and that the vdev cannot be reopened.
202 * SCL_STATE as reader suffices for both.
204 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
206 * spa_vdev_enter() Acquire the namespace lock and the config lock
209 * spa_vdev_exit() Release the config lock, wait for all I/O
210 * to complete, sync the updated configs to the
211 * cache, and release the namespace lock.
213 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
214 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
215 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
217 * spa_rename() is also implemented within this file since is requires
218 * manipulation of the namespace.
221 static avl_tree_t spa_namespace_avl
;
222 kmutex_t spa_namespace_lock
;
223 static kcondvar_t spa_namespace_cv
;
224 static int spa_active_count
;
225 int spa_max_replication_override
= SPA_DVAS_PER_BP
;
227 static kmutex_t spa_spare_lock
;
228 static avl_tree_t spa_spare_avl
;
229 static kmutex_t spa_l2cache_lock
;
230 static avl_tree_t spa_l2cache_avl
;
232 kmem_cache_t
*spa_buffer_pool
;
236 /* Everything except dprintf is on by default in debug builds */
237 int zfs_flags
= ~ZFS_DEBUG_DPRINTF
;
243 * zfs_recover can be set to nonzero to attempt to recover from
244 * otherwise-fatal errors, typically caused by on-disk corruption. When
245 * set, calls to zfs_panic_recover() will turn into warning messages.
251 * ==========================================================================
253 * ==========================================================================
256 spa_config_lock_init(spa_t
*spa
)
258 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
259 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
260 mutex_init(&scl
->scl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
261 cv_init(&scl
->scl_cv
, NULL
, CV_DEFAULT
, NULL
);
262 refcount_create(&scl
->scl_count
);
263 scl
->scl_writer
= NULL
;
264 scl
->scl_write_wanted
= 0;
269 spa_config_lock_destroy(spa_t
*spa
)
271 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
272 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
273 mutex_destroy(&scl
->scl_lock
);
274 cv_destroy(&scl
->scl_cv
);
275 refcount_destroy(&scl
->scl_count
);
276 ASSERT(scl
->scl_writer
== NULL
);
277 ASSERT(scl
->scl_write_wanted
== 0);
282 spa_config_tryenter(spa_t
*spa
, int locks
, void *tag
, krw_t rw
)
284 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
285 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
286 if (!(locks
& (1 << i
)))
288 mutex_enter(&scl
->scl_lock
);
289 if (rw
== RW_READER
) {
290 if (scl
->scl_writer
|| scl
->scl_write_wanted
) {
291 mutex_exit(&scl
->scl_lock
);
292 spa_config_exit(spa
, locks
^ (1 << i
), tag
);
296 ASSERT(scl
->scl_writer
!= curthread
);
297 if (!refcount_is_zero(&scl
->scl_count
)) {
298 mutex_exit(&scl
->scl_lock
);
299 spa_config_exit(spa
, locks
^ (1 << i
), tag
);
302 scl
->scl_writer
= curthread
;
304 (void) refcount_add(&scl
->scl_count
, tag
);
305 mutex_exit(&scl
->scl_lock
);
311 spa_config_enter(spa_t
*spa
, int locks
, void *tag
, krw_t rw
)
315 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
316 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
317 if (scl
->scl_writer
== curthread
)
318 wlocks_held
|= (1 << 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
);
338 ASSERT(wlocks_held
<= locks
);
342 spa_config_exit(spa_t
*spa
, int locks
, void *tag
)
344 for (int i
= SCL_LOCKS
- 1; i
>= 0; i
--) {
345 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
346 if (!(locks
& (1 << i
)))
348 mutex_enter(&scl
->scl_lock
);
349 ASSERT(!refcount_is_zero(&scl
->scl_count
));
350 if (refcount_remove(&scl
->scl_count
, tag
) == 0) {
351 ASSERT(scl
->scl_writer
== NULL
||
352 scl
->scl_writer
== curthread
);
353 scl
->scl_writer
= NULL
; /* OK in either case */
354 cv_broadcast(&scl
->scl_cv
);
356 mutex_exit(&scl
->scl_lock
);
361 spa_config_held(spa_t
*spa
, int locks
, krw_t rw
)
365 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
366 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
367 if (!(locks
& (1 << i
)))
369 if ((rw
== RW_READER
&& !refcount_is_zero(&scl
->scl_count
)) ||
370 (rw
== RW_WRITER
&& scl
->scl_writer
== curthread
))
371 locks_held
|= 1 << i
;
378 * ==========================================================================
379 * SPA namespace functions
380 * ==========================================================================
384 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
385 * Returns NULL if no matching spa_t is found.
388 spa_lookup(const char *name
)
390 static spa_t search
; /* spa_t is large; don't allocate on stack */
396 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
399 * If it's a full dataset name, figure out the pool name and
402 cp
= strpbrk(name
, "/@");
408 (void) strlcpy(search
.spa_name
, name
, sizeof (search
.spa_name
));
409 spa
= avl_find(&spa_namespace_avl
, &search
, &where
);
418 * Create an uninitialized spa_t with the given name. Requires
419 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
420 * exist by calling spa_lookup() first.
423 spa_add(const char *name
, nvlist_t
*config
, const char *altroot
)
426 spa_config_dirent_t
*dp
;
428 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
430 spa
= kmem_zalloc(sizeof (spa_t
), KM_SLEEP
);
432 mutex_init(&spa
->spa_async_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
433 mutex_init(&spa
->spa_errlist_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
434 mutex_init(&spa
->spa_errlog_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
435 mutex_init(&spa
->spa_history_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
436 mutex_init(&spa
->spa_proc_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
437 mutex_init(&spa
->spa_props_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
438 mutex_init(&spa
->spa_scrub_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
439 mutex_init(&spa
->spa_suspend_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
440 mutex_init(&spa
->spa_vdev_top_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
442 cv_init(&spa
->spa_async_cv
, NULL
, CV_DEFAULT
, NULL
);
443 cv_init(&spa
->spa_proc_cv
, NULL
, CV_DEFAULT
, NULL
);
444 cv_init(&spa
->spa_scrub_io_cv
, NULL
, CV_DEFAULT
, NULL
);
445 cv_init(&spa
->spa_suspend_cv
, NULL
, CV_DEFAULT
, NULL
);
447 for (int t
= 0; t
< TXG_SIZE
; t
++)
448 bplist_create(&spa
->spa_free_bplist
[t
]);
450 (void) strlcpy(spa
->spa_name
, name
, sizeof (spa
->spa_name
));
451 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
452 spa
->spa_freeze_txg
= UINT64_MAX
;
453 spa
->spa_final_txg
= UINT64_MAX
;
454 spa
->spa_load_max_txg
= UINT64_MAX
;
456 spa
->spa_proc_state
= SPA_PROC_NONE
;
458 refcount_create(&spa
->spa_refcount
);
459 spa_config_lock_init(spa
);
461 avl_add(&spa_namespace_avl
, spa
);
464 * Set the alternate root, if there is one.
467 spa
->spa_root
= spa_strdup(altroot
);
472 * Every pool starts with the default cachefile
474 list_create(&spa
->spa_config_list
, sizeof (spa_config_dirent_t
),
475 offsetof(spa_config_dirent_t
, scd_link
));
477 dp
= kmem_zalloc(sizeof (spa_config_dirent_t
), KM_SLEEP
);
478 dp
->scd_path
= altroot
? NULL
: spa_strdup(spa_config_path
);
479 list_insert_head(&spa
->spa_config_list
, dp
);
482 VERIFY(nvlist_dup(config
, &spa
->spa_config
, 0) == 0);
488 * Removes a spa_t from the namespace, freeing up any memory used. Requires
489 * spa_namespace_lock. This is called only after the spa_t has been closed and
493 spa_remove(spa_t
*spa
)
495 spa_config_dirent_t
*dp
;
497 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
498 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
500 nvlist_free(spa
->spa_config_splitting
);
502 avl_remove(&spa_namespace_avl
, spa
);
503 cv_broadcast(&spa_namespace_cv
);
506 spa_strfree(spa
->spa_root
);
510 while ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
511 list_remove(&spa
->spa_config_list
, dp
);
512 if (dp
->scd_path
!= NULL
)
513 spa_strfree(dp
->scd_path
);
514 kmem_free(dp
, sizeof (spa_config_dirent_t
));
517 list_destroy(&spa
->spa_config_list
);
519 spa_config_set(spa
, NULL
);
521 refcount_destroy(&spa
->spa_refcount
);
523 spa_config_lock_destroy(spa
);
525 for (int t
= 0; t
< TXG_SIZE
; t
++)
526 bplist_destroy(&spa
->spa_free_bplist
[t
]);
528 cv_destroy(&spa
->spa_async_cv
);
529 cv_destroy(&spa
->spa_proc_cv
);
530 cv_destroy(&spa
->spa_scrub_io_cv
);
531 cv_destroy(&spa
->spa_suspend_cv
);
533 mutex_destroy(&spa
->spa_async_lock
);
534 mutex_destroy(&spa
->spa_errlist_lock
);
535 mutex_destroy(&spa
->spa_errlog_lock
);
536 mutex_destroy(&spa
->spa_history_lock
);
537 mutex_destroy(&spa
->spa_proc_lock
);
538 mutex_destroy(&spa
->spa_props_lock
);
539 mutex_destroy(&spa
->spa_scrub_lock
);
540 mutex_destroy(&spa
->spa_suspend_lock
);
541 mutex_destroy(&spa
->spa_vdev_top_lock
);
543 kmem_free(spa
, sizeof (spa_t
));
547 * Given a pool, return the next pool in the namespace, or NULL if there is
548 * none. If 'prev' is NULL, return the first pool.
551 spa_next(spa_t
*prev
)
553 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
556 return (AVL_NEXT(&spa_namespace_avl
, prev
));
558 return (avl_first(&spa_namespace_avl
));
562 * ==========================================================================
563 * SPA refcount functions
564 * ==========================================================================
568 * Add a reference to the given spa_t. Must have at least one reference, or
569 * have the namespace lock held.
572 spa_open_ref(spa_t
*spa
, void *tag
)
574 ASSERT(refcount_count(&spa
->spa_refcount
) >= spa
->spa_minref
||
575 MUTEX_HELD(&spa_namespace_lock
));
576 (void) refcount_add(&spa
->spa_refcount
, tag
);
580 * Remove a reference to the given spa_t. Must have at least one reference, or
581 * have the namespace lock held.
584 spa_close(spa_t
*spa
, void *tag
)
586 ASSERT(refcount_count(&spa
->spa_refcount
) > spa
->spa_minref
||
587 MUTEX_HELD(&spa_namespace_lock
));
588 (void) refcount_remove(&spa
->spa_refcount
, tag
);
592 * Check to see if the spa refcount is zero. Must be called with
593 * spa_namespace_lock held. We really compare against spa_minref, which is the
594 * number of references acquired when opening a pool
597 spa_refcount_zero(spa_t
*spa
)
599 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
601 return (refcount_count(&spa
->spa_refcount
) == spa
->spa_minref
);
605 * ==========================================================================
606 * SPA spare and l2cache tracking
607 * ==========================================================================
611 * Hot spares and cache devices are tracked using the same code below,
612 * for 'auxiliary' devices.
615 typedef struct spa_aux
{
623 spa_aux_compare(const void *a
, const void *b
)
625 const spa_aux_t
*sa
= a
;
626 const spa_aux_t
*sb
= b
;
628 if (sa
->aux_guid
< sb
->aux_guid
)
630 else if (sa
->aux_guid
> sb
->aux_guid
)
637 spa_aux_add(vdev_t
*vd
, avl_tree_t
*avl
)
643 search
.aux_guid
= vd
->vdev_guid
;
644 if ((aux
= avl_find(avl
, &search
, &where
)) != NULL
) {
647 aux
= kmem_zalloc(sizeof (spa_aux_t
), KM_SLEEP
);
648 aux
->aux_guid
= vd
->vdev_guid
;
650 avl_insert(avl
, aux
, where
);
655 spa_aux_remove(vdev_t
*vd
, avl_tree_t
*avl
)
661 search
.aux_guid
= vd
->vdev_guid
;
662 aux
= avl_find(avl
, &search
, &where
);
666 if (--aux
->aux_count
== 0) {
667 avl_remove(avl
, aux
);
668 kmem_free(aux
, sizeof (spa_aux_t
));
669 } else if (aux
->aux_pool
== spa_guid(vd
->vdev_spa
)) {
670 aux
->aux_pool
= 0ULL;
675 spa_aux_exists(uint64_t guid
, uint64_t *pool
, int *refcnt
, avl_tree_t
*avl
)
677 spa_aux_t search
, *found
;
679 search
.aux_guid
= guid
;
680 found
= avl_find(avl
, &search
, NULL
);
684 *pool
= found
->aux_pool
;
691 *refcnt
= found
->aux_count
;
696 return (found
!= NULL
);
700 spa_aux_activate(vdev_t
*vd
, avl_tree_t
*avl
)
702 spa_aux_t search
, *found
;
705 search
.aux_guid
= vd
->vdev_guid
;
706 found
= avl_find(avl
, &search
, &where
);
707 ASSERT(found
!= NULL
);
708 ASSERT(found
->aux_pool
== 0ULL);
710 found
->aux_pool
= spa_guid(vd
->vdev_spa
);
714 * Spares are tracked globally due to the following constraints:
716 * - A spare may be part of multiple pools.
717 * - A spare may be added to a pool even if it's actively in use within
719 * - A spare in use in any pool can only be the source of a replacement if
720 * the target is a spare in the same pool.
722 * We keep track of all spares on the system through the use of a reference
723 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
724 * spare, then we bump the reference count in the AVL tree. In addition, we set
725 * the 'vdev_isspare' member to indicate that the device is a spare (active or
726 * inactive). When a spare is made active (used to replace a device in the
727 * pool), we also keep track of which pool its been made a part of.
729 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
730 * called under the spa_namespace lock as part of vdev reconfiguration. The
731 * separate spare lock exists for the status query path, which does not need to
732 * be completely consistent with respect to other vdev configuration changes.
736 spa_spare_compare(const void *a
, const void *b
)
738 return (spa_aux_compare(a
, b
));
742 spa_spare_add(vdev_t
*vd
)
744 mutex_enter(&spa_spare_lock
);
745 ASSERT(!vd
->vdev_isspare
);
746 spa_aux_add(vd
, &spa_spare_avl
);
747 vd
->vdev_isspare
= B_TRUE
;
748 mutex_exit(&spa_spare_lock
);
752 spa_spare_remove(vdev_t
*vd
)
754 mutex_enter(&spa_spare_lock
);
755 ASSERT(vd
->vdev_isspare
);
756 spa_aux_remove(vd
, &spa_spare_avl
);
757 vd
->vdev_isspare
= B_FALSE
;
758 mutex_exit(&spa_spare_lock
);
762 spa_spare_exists(uint64_t guid
, uint64_t *pool
, int *refcnt
)
766 mutex_enter(&spa_spare_lock
);
767 found
= spa_aux_exists(guid
, pool
, refcnt
, &spa_spare_avl
);
768 mutex_exit(&spa_spare_lock
);
774 spa_spare_activate(vdev_t
*vd
)
776 mutex_enter(&spa_spare_lock
);
777 ASSERT(vd
->vdev_isspare
);
778 spa_aux_activate(vd
, &spa_spare_avl
);
779 mutex_exit(&spa_spare_lock
);
783 * Level 2 ARC devices are tracked globally for the same reasons as spares.
784 * Cache devices currently only support one pool per cache device, and so
785 * for these devices the aux reference count is currently unused beyond 1.
789 spa_l2cache_compare(const void *a
, const void *b
)
791 return (spa_aux_compare(a
, b
));
795 spa_l2cache_add(vdev_t
*vd
)
797 mutex_enter(&spa_l2cache_lock
);
798 ASSERT(!vd
->vdev_isl2cache
);
799 spa_aux_add(vd
, &spa_l2cache_avl
);
800 vd
->vdev_isl2cache
= B_TRUE
;
801 mutex_exit(&spa_l2cache_lock
);
805 spa_l2cache_remove(vdev_t
*vd
)
807 mutex_enter(&spa_l2cache_lock
);
808 ASSERT(vd
->vdev_isl2cache
);
809 spa_aux_remove(vd
, &spa_l2cache_avl
);
810 vd
->vdev_isl2cache
= B_FALSE
;
811 mutex_exit(&spa_l2cache_lock
);
815 spa_l2cache_exists(uint64_t guid
, uint64_t *pool
)
819 mutex_enter(&spa_l2cache_lock
);
820 found
= spa_aux_exists(guid
, pool
, NULL
, &spa_l2cache_avl
);
821 mutex_exit(&spa_l2cache_lock
);
827 spa_l2cache_activate(vdev_t
*vd
)
829 mutex_enter(&spa_l2cache_lock
);
830 ASSERT(vd
->vdev_isl2cache
);
831 spa_aux_activate(vd
, &spa_l2cache_avl
);
832 mutex_exit(&spa_l2cache_lock
);
836 * ==========================================================================
838 * ==========================================================================
842 * Lock the given spa_t for the purpose of adding or removing a vdev.
843 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
844 * It returns the next transaction group for the spa_t.
847 spa_vdev_enter(spa_t
*spa
)
849 mutex_enter(&spa
->spa_vdev_top_lock
);
850 mutex_enter(&spa_namespace_lock
);
851 return (spa_vdev_config_enter(spa
));
855 * Internal implementation for spa_vdev_enter(). Used when a vdev
856 * operation requires multiple syncs (i.e. removing a device) while
857 * keeping the spa_namespace_lock held.
860 spa_vdev_config_enter(spa_t
*spa
)
862 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
864 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
866 return (spa_last_synced_txg(spa
) + 1);
870 * Used in combination with spa_vdev_config_enter() to allow the syncing
871 * of multiple transactions without releasing the spa_namespace_lock.
874 spa_vdev_config_exit(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int error
, char *tag
)
876 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
878 int config_changed
= B_FALSE
;
880 ASSERT(txg
> spa_last_synced_txg(spa
));
882 spa
->spa_pending_vdev
= NULL
;
887 vdev_dtl_reassess(spa
->spa_root_vdev
, 0, 0, B_FALSE
);
890 * If the config changed, notify the scrub that it must restart.
891 * This will initiate a resilver if needed.
893 if (error
== 0 && !list_is_empty(&spa
->spa_config_dirty_list
)) {
894 config_changed
= B_TRUE
;
895 spa
->spa_config_generation
++;
899 * Verify the metaslab classes.
901 ASSERT(metaslab_class_validate(spa_normal_class(spa
)) == 0);
902 ASSERT(metaslab_class_validate(spa_log_class(spa
)) == 0);
904 spa_config_exit(spa
, SCL_ALL
, spa
);
907 * Panic the system if the specified tag requires it. This
908 * is useful for ensuring that configurations are updated
911 if (zio_injection_enabled
)
912 zio_handle_panic_injection(spa
, tag
, 0);
915 * Note: this txg_wait_synced() is important because it ensures
916 * that there won't be more than one config change per txg.
917 * This allows us to use the txg as the generation number.
920 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
923 ASSERT(!vd
->vdev_detached
|| vd
->vdev_dtl_smo
.smo_object
== 0);
924 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
926 spa_config_exit(spa
, SCL_ALL
, spa
);
930 * If the config changed, update the config cache.
933 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
937 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
938 * locking of spa_vdev_enter(), we also want make sure the transactions have
939 * synced to disk, and then update the global configuration cache with the new
943 spa_vdev_exit(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int error
)
945 spa_vdev_config_exit(spa
, vd
, txg
, error
, FTAG
);
946 mutex_exit(&spa_namespace_lock
);
947 mutex_exit(&spa
->spa_vdev_top_lock
);
953 * Lock the given spa_t for the purpose of changing vdev state.
956 spa_vdev_state_enter(spa_t
*spa
, int oplocks
)
958 int locks
= SCL_STATE_ALL
| oplocks
;
961 * Root pools may need to read of the underlying devfs filesystem
962 * when opening up a vdev. Unfortunately if we're holding the
963 * SCL_ZIO lock it will result in a deadlock when we try to issue
964 * the read from the root filesystem. Instead we "prefetch"
965 * the associated vnodes that we need prior to opening the
966 * underlying devices and cache them so that we can prevent
967 * any I/O when we are doing the actual open.
969 if (spa_is_root(spa
)) {
970 int low
= locks
& ~(SCL_ZIO
- 1);
971 int high
= locks
& ~low
;
973 spa_config_enter(spa
, high
, spa
, RW_WRITER
);
974 vdev_hold(spa
->spa_root_vdev
);
975 spa_config_enter(spa
, low
, spa
, RW_WRITER
);
977 spa_config_enter(spa
, locks
, spa
, RW_WRITER
);
979 spa
->spa_vdev_locks
= locks
;
983 spa_vdev_state_exit(spa_t
*spa
, vdev_t
*vd
, int error
)
985 boolean_t config_changed
= B_FALSE
;
987 if (vd
!= NULL
|| error
== 0)
988 vdev_dtl_reassess(vd
? vd
->vdev_top
: spa
->spa_root_vdev
,
992 vdev_state_dirty(vd
->vdev_top
);
993 config_changed
= B_TRUE
;
994 spa
->spa_config_generation
++;
997 if (spa_is_root(spa
))
998 vdev_rele(spa
->spa_root_vdev
);
1000 ASSERT3U(spa
->spa_vdev_locks
, >=, SCL_STATE_ALL
);
1001 spa_config_exit(spa
, spa
->spa_vdev_locks
, spa
);
1004 * If anything changed, wait for it to sync. This ensures that,
1005 * from the system administrator's perspective, zpool(1M) commands
1006 * are synchronous. This is important for things like zpool offline:
1007 * when the command completes, you expect no further I/O from ZFS.
1010 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1013 * If the config changed, update the config cache.
1015 if (config_changed
) {
1016 mutex_enter(&spa_namespace_lock
);
1017 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
1018 mutex_exit(&spa_namespace_lock
);
1025 * ==========================================================================
1026 * Miscellaneous functions
1027 * ==========================================================================
1034 spa_rename(const char *name
, const char *newname
)
1040 * Lookup the spa_t and grab the config lock for writing. We need to
1041 * actually open the pool so that we can sync out the necessary labels.
1042 * It's OK to call spa_open() with the namespace lock held because we
1043 * allow recursive calls for other reasons.
1045 mutex_enter(&spa_namespace_lock
);
1046 if ((err
= spa_open(name
, &spa
, FTAG
)) != 0) {
1047 mutex_exit(&spa_namespace_lock
);
1051 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1053 avl_remove(&spa_namespace_avl
, spa
);
1054 (void) strlcpy(spa
->spa_name
, newname
, sizeof (spa
->spa_name
));
1055 avl_add(&spa_namespace_avl
, spa
);
1058 * Sync all labels to disk with the new names by marking the root vdev
1059 * dirty and waiting for it to sync. It will pick up the new pool name
1062 vdev_config_dirty(spa
->spa_root_vdev
);
1064 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1066 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1069 * Sync the updated config cache.
1071 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
1073 spa_close(spa
, FTAG
);
1075 mutex_exit(&spa_namespace_lock
);
1081 * Determine whether a pool with given pool_guid exists. If device_guid is
1082 * non-zero, determine whether the pool exists *and* contains a device with the
1083 * specified device_guid.
1086 spa_guid_exists(uint64_t pool_guid
, uint64_t device_guid
)
1089 avl_tree_t
*t
= &spa_namespace_avl
;
1091 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1093 for (spa
= avl_first(t
); spa
!= NULL
; spa
= AVL_NEXT(t
, spa
)) {
1094 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1096 if (spa
->spa_root_vdev
== NULL
)
1098 if (spa_guid(spa
) == pool_guid
) {
1099 if (device_guid
== 0)
1102 if (vdev_lookup_by_guid(spa
->spa_root_vdev
,
1103 device_guid
) != NULL
)
1107 * Check any devices we may be in the process of adding.
1109 if (spa
->spa_pending_vdev
) {
1110 if (vdev_lookup_by_guid(spa
->spa_pending_vdev
,
1111 device_guid
) != NULL
)
1117 return (spa
!= NULL
);
1121 spa_strdup(const char *s
)
1127 new = kmem_alloc(len
+ 1, KM_SLEEP
);
1135 spa_strfree(char *s
)
1137 kmem_free(s
, strlen(s
) + 1);
1141 spa_get_random(uint64_t range
)
1147 (void) random_get_pseudo_bytes((void *)&r
, sizeof (uint64_t));
1153 spa_generate_guid(spa_t
*spa
)
1155 uint64_t guid
= spa_get_random(-1ULL);
1158 while (guid
== 0 || spa_guid_exists(spa_guid(spa
), guid
))
1159 guid
= spa_get_random(-1ULL);
1161 while (guid
== 0 || spa_guid_exists(guid
, 0))
1162 guid
= spa_get_random(-1ULL);
1169 sprintf_blkptr(char *buf
, const blkptr_t
*bp
)
1172 char *checksum
= NULL
;
1173 char *compress
= NULL
;
1176 type
= dmu_ot
[BP_GET_TYPE(bp
)].ot_name
;
1177 checksum
= zio_checksum_table
[BP_GET_CHECKSUM(bp
)].ci_name
;
1178 compress
= zio_compress_table
[BP_GET_COMPRESS(bp
)].ci_name
;
1181 SPRINTF_BLKPTR(snprintf
, ' ', buf
, bp
, type
, checksum
, compress
);
1185 spa_freeze(spa_t
*spa
)
1187 uint64_t freeze_txg
= 0;
1189 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1190 if (spa
->spa_freeze_txg
== UINT64_MAX
) {
1191 freeze_txg
= spa_last_synced_txg(spa
) + TXG_SIZE
;
1192 spa
->spa_freeze_txg
= freeze_txg
;
1194 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1195 if (freeze_txg
!= 0)
1196 txg_wait_synced(spa_get_dsl(spa
), freeze_txg
);
1200 zfs_panic_recover(const char *fmt
, ...)
1205 vcmn_err(zfs_recover
? CE_WARN
: CE_PANIC
, fmt
, adx
);
1210 * This is a stripped-down version of strtoull, suitable only for converting
1211 * lowercase hexidecimal numbers that don't overflow.
1214 strtonum(const char *str
, char **nptr
)
1220 while ((c
= *str
) != '\0') {
1221 if (c
>= '0' && c
<= '9')
1223 else if (c
>= 'a' && c
<= 'f')
1224 digit
= 10 + c
- 'a';
1235 *nptr
= (char *)str
;
1241 * ==========================================================================
1242 * Accessor functions
1243 * ==========================================================================
1247 spa_shutting_down(spa_t
*spa
)
1249 return (spa
->spa_async_suspended
);
1253 spa_get_dsl(spa_t
*spa
)
1255 return (spa
->spa_dsl_pool
);
1259 spa_get_rootblkptr(spa_t
*spa
)
1261 return (&spa
->spa_ubsync
.ub_rootbp
);
1265 spa_set_rootblkptr(spa_t
*spa
, const blkptr_t
*bp
)
1267 spa
->spa_uberblock
.ub_rootbp
= *bp
;
1271 spa_altroot(spa_t
*spa
, char *buf
, size_t buflen
)
1273 if (spa
->spa_root
== NULL
)
1276 (void) strncpy(buf
, spa
->spa_root
, buflen
);
1280 spa_sync_pass(spa_t
*spa
)
1282 return (spa
->spa_sync_pass
);
1286 spa_name(spa_t
*spa
)
1288 return (spa
->spa_name
);
1292 spa_guid(spa_t
*spa
)
1295 * If we fail to parse the config during spa_load(), we can go through
1296 * the error path (which posts an ereport) and end up here with no root
1297 * vdev. We stash the original pool guid in 'spa_load_guid' to handle
1300 if (spa
->spa_root_vdev
!= NULL
)
1301 return (spa
->spa_root_vdev
->vdev_guid
);
1303 return (spa
->spa_load_guid
);
1307 spa_last_synced_txg(spa_t
*spa
)
1309 return (spa
->spa_ubsync
.ub_txg
);
1313 spa_first_txg(spa_t
*spa
)
1315 return (spa
->spa_first_txg
);
1319 spa_syncing_txg(spa_t
*spa
)
1321 return (spa
->spa_syncing_txg
);
1325 spa_state(spa_t
*spa
)
1327 return (spa
->spa_state
);
1331 spa_load_state(spa_t
*spa
)
1333 return (spa
->spa_load_state
);
1337 spa_freeze_txg(spa_t
*spa
)
1339 return (spa
->spa_freeze_txg
);
1344 spa_get_asize(spa_t
*spa
, uint64_t lsize
)
1347 * The worst case is single-sector max-parity RAID-Z blocks, in which
1348 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
1349 * times the size; so just assume that. Add to this the fact that
1350 * we can have up to 3 DVAs per bp, and one more factor of 2 because
1351 * the block may be dittoed with up to 3 DVAs by ddt_sync().
1353 return (lsize
* (VDEV_RAIDZ_MAXPARITY
+ 1) * SPA_DVAS_PER_BP
* 2);
1357 spa_get_dspace(spa_t
*spa
)
1359 return (spa
->spa_dspace
);
1363 spa_update_dspace(spa_t
*spa
)
1365 spa
->spa_dspace
= metaslab_class_get_dspace(spa_normal_class(spa
)) +
1366 ddt_get_dedup_dspace(spa
);
1370 * Return the failure mode that has been set to this pool. The default
1371 * behavior will be to block all I/Os when a complete failure occurs.
1374 spa_get_failmode(spa_t
*spa
)
1376 return (spa
->spa_failmode
);
1380 spa_suspended(spa_t
*spa
)
1382 return (spa
->spa_suspended
);
1386 spa_version(spa_t
*spa
)
1388 return (spa
->spa_ubsync
.ub_version
);
1392 spa_deflate(spa_t
*spa
)
1394 return (spa
->spa_deflate
);
1398 spa_normal_class(spa_t
*spa
)
1400 return (spa
->spa_normal_class
);
1404 spa_log_class(spa_t
*spa
)
1406 return (spa
->spa_log_class
);
1410 spa_max_replication(spa_t
*spa
)
1413 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1414 * handle BPs with more than one DVA allocated. Set our max
1415 * replication level accordingly.
1417 if (spa_version(spa
) < SPA_VERSION_DITTO_BLOCKS
)
1419 return (MIN(SPA_DVAS_PER_BP
, spa_max_replication_override
));
1423 spa_prev_software_version(spa_t
*spa
)
1425 return (spa
->spa_prev_software_version
);
1429 dva_get_dsize_sync(spa_t
*spa
, const dva_t
*dva
)
1431 uint64_t asize
= DVA_GET_ASIZE(dva
);
1432 uint64_t dsize
= asize
;
1434 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1436 if (asize
!= 0 && spa
->spa_deflate
) {
1437 vdev_t
*vd
= vdev_lookup_top(spa
, DVA_GET_VDEV(dva
));
1438 dsize
= (asize
>> SPA_MINBLOCKSHIFT
) * vd
->vdev_deflate_ratio
;
1445 bp_get_dsize_sync(spa_t
*spa
, const blkptr_t
*bp
)
1449 for (int d
= 0; d
< SPA_DVAS_PER_BP
; d
++)
1450 dsize
+= dva_get_dsize_sync(spa
, &bp
->blk_dva
[d
]);
1456 bp_get_dsize(spa_t
*spa
, const blkptr_t
*bp
)
1460 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
1462 for (int d
= 0; d
< SPA_DVAS_PER_BP
; d
++)
1463 dsize
+= dva_get_dsize_sync(spa
, &bp
->blk_dva
[d
]);
1465 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
1471 * ==========================================================================
1472 * Initialization and Termination
1473 * ==========================================================================
1477 spa_name_compare(const void *a1
, const void *a2
)
1479 const spa_t
*s1
= a1
;
1480 const spa_t
*s2
= a2
;
1483 s
= strcmp(s1
->spa_name
, s2
->spa_name
);
1494 return (spa_active_count
);
1506 mutex_init(&spa_namespace_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1507 mutex_init(&spa_spare_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1508 mutex_init(&spa_l2cache_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1509 cv_init(&spa_namespace_cv
, NULL
, CV_DEFAULT
, NULL
);
1511 avl_create(&spa_namespace_avl
, spa_name_compare
, sizeof (spa_t
),
1512 offsetof(spa_t
, spa_avl
));
1514 avl_create(&spa_spare_avl
, spa_spare_compare
, sizeof (spa_aux_t
),
1515 offsetof(spa_aux_t
, aux_avl
));
1517 avl_create(&spa_l2cache_avl
, spa_l2cache_compare
, sizeof (spa_aux_t
),
1518 offsetof(spa_aux_t
, aux_avl
));
1520 spa_mode_global
= mode
;
1527 vdev_cache_stat_init();
1541 vdev_cache_stat_fini();
1548 avl_destroy(&spa_namespace_avl
);
1549 avl_destroy(&spa_spare_avl
);
1550 avl_destroy(&spa_l2cache_avl
);
1552 cv_destroy(&spa_namespace_cv
);
1553 mutex_destroy(&spa_namespace_lock
);
1554 mutex_destroy(&spa_spare_lock
);
1555 mutex_destroy(&spa_l2cache_lock
);
1559 * Return whether this pool has slogs. No locking needed.
1560 * It's not a problem if the wrong answer is returned as it's only for
1561 * performance and not correctness
1564 spa_has_slogs(spa_t
*spa
)
1566 return (spa
->spa_log_class
->mc_rotor
!= NULL
);
1570 spa_get_log_state(spa_t
*spa
)
1572 return (spa
->spa_log_state
);
1576 spa_set_log_state(spa_t
*spa
, spa_log_state_t state
)
1578 spa
->spa_log_state
= state
;
1582 spa_is_root(spa_t
*spa
)
1584 return (spa
->spa_is_root
);
1588 spa_writeable(spa_t
*spa
)
1590 return (!!(spa
->spa_mode
& FWRITE
));
1594 spa_mode(spa_t
*spa
)
1596 return (spa
->spa_mode
);
1600 spa_bootfs(spa_t
*spa
)
1602 return (spa
->spa_bootfs
);
1606 spa_delegation(spa_t
*spa
)
1608 return (spa
->spa_delegation
);
1612 spa_meta_objset(spa_t
*spa
)
1614 return (spa
->spa_meta_objset
);
1618 spa_dedup_checksum(spa_t
*spa
)
1620 return (spa
->spa_dedup_checksum
);
1624 * Reset pool scan stat per scan pass (or reboot).
1627 spa_scan_stat_init(spa_t
*spa
)
1629 /* data not stored on disk */
1630 spa
->spa_scan_pass_start
= gethrestime_sec();
1631 spa
->spa_scan_pass_exam
= 0;
1632 vdev_scan_stat_init(spa
->spa_root_vdev
);
1636 * Get scan stats for zpool status reports
1639 spa_scan_get_stats(spa_t
*spa
, pool_scan_stat_t
*ps
)
1641 dsl_scan_t
*scn
= spa
->spa_dsl_pool
? spa
->spa_dsl_pool
->dp_scan
: NULL
;
1643 if (scn
== NULL
|| scn
->scn_phys
.scn_func
== POOL_SCAN_NONE
)
1645 bzero(ps
, sizeof (pool_scan_stat_t
));
1647 /* data stored on disk */
1648 ps
->pss_func
= scn
->scn_phys
.scn_func
;
1649 ps
->pss_start_time
= scn
->scn_phys
.scn_start_time
;
1650 ps
->pss_end_time
= scn
->scn_phys
.scn_end_time
;
1651 ps
->pss_to_examine
= scn
->scn_phys
.scn_to_examine
;
1652 ps
->pss_examined
= scn
->scn_phys
.scn_examined
;
1653 ps
->pss_to_process
= scn
->scn_phys
.scn_to_process
;
1654 ps
->pss_processed
= scn
->scn_phys
.scn_processed
;
1655 ps
->pss_errors
= scn
->scn_phys
.scn_errors
;
1656 ps
->pss_state
= scn
->scn_phys
.scn_state
;
1658 /* data not stored on disk */
1659 ps
->pss_pass_start
= spa
->spa_scan_pass_start
;
1660 ps
->pss_pass_exam
= spa
->spa_scan_pass_exam
;