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.
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
, 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_scrub_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
434 mutex_init(&spa
->spa_errlog_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
435 mutex_init(&spa
->spa_errlist_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
436 mutex_init(&spa
->spa_sync_bplist
.bpl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
437 mutex_init(&spa
->spa_history_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
438 mutex_init(&spa
->spa_props_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
440 cv_init(&spa
->spa_async_cv
, NULL
, CV_DEFAULT
, NULL
);
441 cv_init(&spa
->spa_scrub_io_cv
, NULL
, CV_DEFAULT
, NULL
);
442 cv_init(&spa
->spa_suspend_cv
, NULL
, CV_DEFAULT
, NULL
);
444 (void) strlcpy(spa
->spa_name
, name
, sizeof (spa
->spa_name
));
445 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
446 spa
->spa_freeze_txg
= UINT64_MAX
;
447 spa
->spa_final_txg
= UINT64_MAX
;
449 refcount_create(&spa
->spa_refcount
);
450 spa_config_lock_init(spa
);
452 avl_add(&spa_namespace_avl
, spa
);
454 mutex_init(&spa
->spa_suspend_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
457 * Set the alternate root, if there is one.
460 spa
->spa_root
= spa_strdup(altroot
);
465 * Every pool starts with the default cachefile
467 list_create(&spa
->spa_config_list
, sizeof (spa_config_dirent_t
),
468 offsetof(spa_config_dirent_t
, scd_link
));
470 dp
= kmem_zalloc(sizeof (spa_config_dirent_t
), KM_SLEEP
);
471 dp
->scd_path
= spa_strdup(spa_config_path
);
472 list_insert_head(&spa
->spa_config_list
, dp
);
478 * Removes a spa_t from the namespace, freeing up any memory used. Requires
479 * spa_namespace_lock. This is called only after the spa_t has been closed and
483 spa_remove(spa_t
*spa
)
485 spa_config_dirent_t
*dp
;
487 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
488 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
490 avl_remove(&spa_namespace_avl
, spa
);
491 cv_broadcast(&spa_namespace_cv
);
494 spa_strfree(spa
->spa_root
);
498 while ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
499 list_remove(&spa
->spa_config_list
, dp
);
500 if (dp
->scd_path
!= NULL
)
501 spa_strfree(dp
->scd_path
);
502 kmem_free(dp
, sizeof (spa_config_dirent_t
));
505 list_destroy(&spa
->spa_config_list
);
507 spa_config_set(spa
, NULL
);
509 refcount_destroy(&spa
->spa_refcount
);
511 spa_config_lock_destroy(spa
);
513 cv_destroy(&spa
->spa_async_cv
);
514 cv_destroy(&spa
->spa_scrub_io_cv
);
515 cv_destroy(&spa
->spa_suspend_cv
);
517 mutex_destroy(&spa
->spa_async_lock
);
518 mutex_destroy(&spa
->spa_scrub_lock
);
519 mutex_destroy(&spa
->spa_errlog_lock
);
520 mutex_destroy(&spa
->spa_errlist_lock
);
521 mutex_destroy(&spa
->spa_sync_bplist
.bpl_lock
);
522 mutex_destroy(&spa
->spa_history_lock
);
523 mutex_destroy(&spa
->spa_props_lock
);
524 mutex_destroy(&spa
->spa_suspend_lock
);
526 kmem_free(spa
, sizeof (spa_t
));
530 * Given a pool, return the next pool in the namespace, or NULL if there is
531 * none. If 'prev' is NULL, return the first pool.
534 spa_next(spa_t
*prev
)
536 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
539 return (AVL_NEXT(&spa_namespace_avl
, prev
));
541 return (avl_first(&spa_namespace_avl
));
545 * ==========================================================================
546 * SPA refcount functions
547 * ==========================================================================
551 * Add a reference to the given spa_t. Must have at least one reference, or
552 * have the namespace lock held.
555 spa_open_ref(spa_t
*spa
, void *tag
)
557 ASSERT(refcount_count(&spa
->spa_refcount
) >= spa
->spa_minref
||
558 MUTEX_HELD(&spa_namespace_lock
));
559 (void) refcount_add(&spa
->spa_refcount
, tag
);
563 * Remove a reference to the given spa_t. Must have at least one reference, or
564 * have the namespace lock held.
567 spa_close(spa_t
*spa
, void *tag
)
569 ASSERT(refcount_count(&spa
->spa_refcount
) > spa
->spa_minref
||
570 MUTEX_HELD(&spa_namespace_lock
));
571 (void) refcount_remove(&spa
->spa_refcount
, tag
);
575 * Check to see if the spa refcount is zero. Must be called with
576 * spa_namespace_lock held. We really compare against spa_minref, which is the
577 * number of references acquired when opening a pool
580 spa_refcount_zero(spa_t
*spa
)
582 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
584 return (refcount_count(&spa
->spa_refcount
) == spa
->spa_minref
);
588 * ==========================================================================
589 * SPA spare and l2cache tracking
590 * ==========================================================================
594 * Hot spares and cache devices are tracked using the same code below,
595 * for 'auxiliary' devices.
598 typedef struct spa_aux
{
606 spa_aux_compare(const void *a
, const void *b
)
608 const spa_aux_t
*sa
= a
;
609 const spa_aux_t
*sb
= b
;
611 if (sa
->aux_guid
< sb
->aux_guid
)
613 else if (sa
->aux_guid
> sb
->aux_guid
)
620 spa_aux_add(vdev_t
*vd
, avl_tree_t
*avl
)
626 search
.aux_guid
= vd
->vdev_guid
;
627 if ((aux
= avl_find(avl
, &search
, &where
)) != NULL
) {
630 aux
= kmem_zalloc(sizeof (spa_aux_t
), KM_SLEEP
);
631 aux
->aux_guid
= vd
->vdev_guid
;
633 avl_insert(avl
, aux
, where
);
638 spa_aux_remove(vdev_t
*vd
, avl_tree_t
*avl
)
644 search
.aux_guid
= vd
->vdev_guid
;
645 aux
= avl_find(avl
, &search
, &where
);
649 if (--aux
->aux_count
== 0) {
650 avl_remove(avl
, aux
);
651 kmem_free(aux
, sizeof (spa_aux_t
));
652 } else if (aux
->aux_pool
== spa_guid(vd
->vdev_spa
)) {
653 aux
->aux_pool
= 0ULL;
658 spa_aux_exists(uint64_t guid
, uint64_t *pool
, int *refcnt
, avl_tree_t
*avl
)
660 spa_aux_t search
, *found
;
662 search
.aux_guid
= guid
;
663 found
= avl_find(avl
, &search
, NULL
);
667 *pool
= found
->aux_pool
;
674 *refcnt
= found
->aux_count
;
679 return (found
!= NULL
);
683 spa_aux_activate(vdev_t
*vd
, avl_tree_t
*avl
)
685 spa_aux_t search
, *found
;
688 search
.aux_guid
= vd
->vdev_guid
;
689 found
= avl_find(avl
, &search
, &where
);
690 ASSERT(found
!= NULL
);
691 ASSERT(found
->aux_pool
== 0ULL);
693 found
->aux_pool
= spa_guid(vd
->vdev_spa
);
697 * Spares are tracked globally due to the following constraints:
699 * - A spare may be part of multiple pools.
700 * - A spare may be added to a pool even if it's actively in use within
702 * - A spare in use in any pool can only be the source of a replacement if
703 * the target is a spare in the same pool.
705 * We keep track of all spares on the system through the use of a reference
706 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
707 * spare, then we bump the reference count in the AVL tree. In addition, we set
708 * the 'vdev_isspare' member to indicate that the device is a spare (active or
709 * inactive). When a spare is made active (used to replace a device in the
710 * pool), we also keep track of which pool its been made a part of.
712 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
713 * called under the spa_namespace lock as part of vdev reconfiguration. The
714 * separate spare lock exists for the status query path, which does not need to
715 * be completely consistent with respect to other vdev configuration changes.
719 spa_spare_compare(const void *a
, const void *b
)
721 return (spa_aux_compare(a
, b
));
725 spa_spare_add(vdev_t
*vd
)
727 mutex_enter(&spa_spare_lock
);
728 ASSERT(!vd
->vdev_isspare
);
729 spa_aux_add(vd
, &spa_spare_avl
);
730 vd
->vdev_isspare
= B_TRUE
;
731 mutex_exit(&spa_spare_lock
);
735 spa_spare_remove(vdev_t
*vd
)
737 mutex_enter(&spa_spare_lock
);
738 ASSERT(vd
->vdev_isspare
);
739 spa_aux_remove(vd
, &spa_spare_avl
);
740 vd
->vdev_isspare
= B_FALSE
;
741 mutex_exit(&spa_spare_lock
);
745 spa_spare_exists(uint64_t guid
, uint64_t *pool
, int *refcnt
)
749 mutex_enter(&spa_spare_lock
);
750 found
= spa_aux_exists(guid
, pool
, refcnt
, &spa_spare_avl
);
751 mutex_exit(&spa_spare_lock
);
757 spa_spare_activate(vdev_t
*vd
)
759 mutex_enter(&spa_spare_lock
);
760 ASSERT(vd
->vdev_isspare
);
761 spa_aux_activate(vd
, &spa_spare_avl
);
762 mutex_exit(&spa_spare_lock
);
766 * Level 2 ARC devices are tracked globally for the same reasons as spares.
767 * Cache devices currently only support one pool per cache device, and so
768 * for these devices the aux reference count is currently unused beyond 1.
772 spa_l2cache_compare(const void *a
, const void *b
)
774 return (spa_aux_compare(a
, b
));
778 spa_l2cache_add(vdev_t
*vd
)
780 mutex_enter(&spa_l2cache_lock
);
781 ASSERT(!vd
->vdev_isl2cache
);
782 spa_aux_add(vd
, &spa_l2cache_avl
);
783 vd
->vdev_isl2cache
= B_TRUE
;
784 mutex_exit(&spa_l2cache_lock
);
788 spa_l2cache_remove(vdev_t
*vd
)
790 mutex_enter(&spa_l2cache_lock
);
791 ASSERT(vd
->vdev_isl2cache
);
792 spa_aux_remove(vd
, &spa_l2cache_avl
);
793 vd
->vdev_isl2cache
= B_FALSE
;
794 mutex_exit(&spa_l2cache_lock
);
798 spa_l2cache_exists(uint64_t guid
, uint64_t *pool
)
802 mutex_enter(&spa_l2cache_lock
);
803 found
= spa_aux_exists(guid
, pool
, NULL
, &spa_l2cache_avl
);
804 mutex_exit(&spa_l2cache_lock
);
810 spa_l2cache_activate(vdev_t
*vd
)
812 mutex_enter(&spa_l2cache_lock
);
813 ASSERT(vd
->vdev_isl2cache
);
814 spa_aux_activate(vd
, &spa_l2cache_avl
);
815 mutex_exit(&spa_l2cache_lock
);
819 spa_l2cache_space_update(vdev_t
*vd
, int64_t space
, int64_t alloc
)
821 vdev_space_update(vd
, space
, alloc
, B_FALSE
);
825 * ==========================================================================
827 * ==========================================================================
831 * Lock the given spa_t for the purpose of adding or removing a vdev.
832 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
833 * It returns the next transaction group for the spa_t.
836 spa_vdev_enter(spa_t
*spa
)
838 mutex_enter(&spa_namespace_lock
);
840 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
842 return (spa_last_synced_txg(spa
) + 1);
846 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
847 * locking of spa_vdev_enter(), we also want make sure the transactions have
848 * synced to disk, and then update the global configuration cache with the new
852 spa_vdev_exit(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int error
)
854 int config_changed
= B_FALSE
;
856 ASSERT(txg
> spa_last_synced_txg(spa
));
858 spa
->spa_pending_vdev
= NULL
;
863 vdev_dtl_reassess(spa
->spa_root_vdev
, 0, 0, B_FALSE
);
866 * If the config changed, notify the scrub thread that it must restart.
868 if (error
== 0 && !list_is_empty(&spa
->spa_config_dirty_list
)) {
869 dsl_pool_scrub_restart(spa
->spa_dsl_pool
);
870 config_changed
= B_TRUE
;
873 spa_config_exit(spa
, SCL_ALL
, spa
);
876 * Note: this txg_wait_synced() is important because it ensures
877 * that there won't be more than one config change per txg.
878 * This allows us to use the txg as the generation number.
881 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
884 ASSERT(!vd
->vdev_detached
|| vd
->vdev_dtl_smo
.smo_object
== 0);
885 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
887 spa_config_exit(spa
, SCL_ALL
, spa
);
891 * If the config changed, update the config cache.
894 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
896 mutex_exit(&spa_namespace_lock
);
902 * Lock the given spa_t for the purpose of changing vdev state.
905 spa_vdev_state_enter(spa_t
*spa
)
907 spa_config_enter(spa
, SCL_STATE_ALL
, spa
, RW_WRITER
);
911 spa_vdev_state_exit(spa_t
*spa
, vdev_t
*vd
, int error
)
914 vdev_state_dirty(vd
->vdev_top
);
916 spa_config_exit(spa
, SCL_STATE_ALL
, spa
);
919 * If anything changed, wait for it to sync. This ensures that,
920 * from the system administrator's perspective, zpool(1M) commands
921 * are synchronous. This is important for things like zpool offline:
922 * when the command completes, you expect no further I/O from ZFS.
925 txg_wait_synced(spa
->spa_dsl_pool
, 0);
931 * ==========================================================================
932 * Miscellaneous functions
933 * ==========================================================================
940 spa_rename(const char *name
, const char *newname
)
946 * Lookup the spa_t and grab the config lock for writing. We need to
947 * actually open the pool so that we can sync out the necessary labels.
948 * It's OK to call spa_open() with the namespace lock held because we
949 * allow recursive calls for other reasons.
951 mutex_enter(&spa_namespace_lock
);
952 if ((err
= spa_open(name
, &spa
, FTAG
)) != 0) {
953 mutex_exit(&spa_namespace_lock
);
957 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
959 avl_remove(&spa_namespace_avl
, spa
);
960 (void) strlcpy(spa
->spa_name
, newname
, sizeof (spa
->spa_name
));
961 avl_add(&spa_namespace_avl
, spa
);
964 * Sync all labels to disk with the new names by marking the root vdev
965 * dirty and waiting for it to sync. It will pick up the new pool name
968 vdev_config_dirty(spa
->spa_root_vdev
);
970 spa_config_exit(spa
, SCL_ALL
, FTAG
);
972 txg_wait_synced(spa
->spa_dsl_pool
, 0);
975 * Sync the updated config cache.
977 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
979 spa_close(spa
, FTAG
);
981 mutex_exit(&spa_namespace_lock
);
988 * Determine whether a pool with given pool_guid exists. If device_guid is
989 * non-zero, determine whether the pool exists *and* contains a device with the
990 * specified device_guid.
993 spa_guid_exists(uint64_t pool_guid
, uint64_t device_guid
)
996 avl_tree_t
*t
= &spa_namespace_avl
;
998 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1000 for (spa
= avl_first(t
); spa
!= NULL
; spa
= AVL_NEXT(t
, spa
)) {
1001 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1003 if (spa
->spa_root_vdev
== NULL
)
1005 if (spa_guid(spa
) == pool_guid
) {
1006 if (device_guid
== 0)
1009 if (vdev_lookup_by_guid(spa
->spa_root_vdev
,
1010 device_guid
) != NULL
)
1014 * Check any devices we may be in the process of adding.
1016 if (spa
->spa_pending_vdev
) {
1017 if (vdev_lookup_by_guid(spa
->spa_pending_vdev
,
1018 device_guid
) != NULL
)
1024 return (spa
!= NULL
);
1028 spa_strdup(const char *s
)
1034 new = kmem_alloc(len
+ 1, KM_SLEEP
);
1042 spa_strfree(char *s
)
1044 kmem_free(s
, strlen(s
) + 1);
1048 spa_get_random(uint64_t range
)
1054 (void) random_get_pseudo_bytes((void *)&r
, sizeof (uint64_t));
1060 sprintf_blkptr(char *buf
, int len
, const blkptr_t
*bp
)
1065 (void) snprintf(buf
, len
, "<NULL>");
1069 if (BP_IS_HOLE(bp
)) {
1070 (void) snprintf(buf
, len
, "<hole>");
1074 (void) snprintf(buf
, len
, "[L%llu %s] %llxL/%llxP ",
1075 (u_longlong_t
)BP_GET_LEVEL(bp
),
1076 dmu_ot
[BP_GET_TYPE(bp
)].ot_name
,
1077 (u_longlong_t
)BP_GET_LSIZE(bp
),
1078 (u_longlong_t
)BP_GET_PSIZE(bp
));
1080 for (d
= 0; d
< BP_GET_NDVAS(bp
); d
++) {
1081 const dva_t
*dva
= &bp
->blk_dva
[d
];
1082 (void) snprintf(buf
+ strlen(buf
), len
- strlen(buf
),
1083 "DVA[%d]=<%llu:%llx:%llx> ", d
,
1084 (u_longlong_t
)DVA_GET_VDEV(dva
),
1085 (u_longlong_t
)DVA_GET_OFFSET(dva
),
1086 (u_longlong_t
)DVA_GET_ASIZE(dva
));
1089 (void) snprintf(buf
+ strlen(buf
), len
- strlen(buf
),
1090 "%s %s %s %s birth=%llu fill=%llu cksum=%llx:%llx:%llx:%llx",
1091 zio_checksum_table
[BP_GET_CHECKSUM(bp
)].ci_name
,
1092 zio_compress_table
[BP_GET_COMPRESS(bp
)].ci_name
,
1093 BP_GET_BYTEORDER(bp
) == 0 ? "BE" : "LE",
1094 BP_IS_GANG(bp
) ? "gang" : "contiguous",
1095 (u_longlong_t
)bp
->blk_birth
,
1096 (u_longlong_t
)bp
->blk_fill
,
1097 (u_longlong_t
)bp
->blk_cksum
.zc_word
[0],
1098 (u_longlong_t
)bp
->blk_cksum
.zc_word
[1],
1099 (u_longlong_t
)bp
->blk_cksum
.zc_word
[2],
1100 (u_longlong_t
)bp
->blk_cksum
.zc_word
[3]);
1104 spa_freeze(spa_t
*spa
)
1106 uint64_t freeze_txg
= 0;
1108 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1109 if (spa
->spa_freeze_txg
== UINT64_MAX
) {
1110 freeze_txg
= spa_last_synced_txg(spa
) + TXG_SIZE
;
1111 spa
->spa_freeze_txg
= freeze_txg
;
1113 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1114 if (freeze_txg
!= 0)
1115 txg_wait_synced(spa_get_dsl(spa
), freeze_txg
);
1119 zfs_panic_recover(const char *fmt
, ...)
1124 vcmn_err(zfs_recover
? CE_WARN
: CE_PANIC
, fmt
, adx
);
1129 * ==========================================================================
1130 * Accessor functions
1131 * ==========================================================================
1135 spa_shutting_down(spa_t
*spa
)
1137 return (spa
->spa_async_suspended
);
1141 spa_get_dsl(spa_t
*spa
)
1143 return (spa
->spa_dsl_pool
);
1147 spa_get_rootblkptr(spa_t
*spa
)
1149 return (&spa
->spa_ubsync
.ub_rootbp
);
1153 spa_set_rootblkptr(spa_t
*spa
, const blkptr_t
*bp
)
1155 spa
->spa_uberblock
.ub_rootbp
= *bp
;
1159 spa_altroot(spa_t
*spa
, char *buf
, size_t buflen
)
1161 if (spa
->spa_root
== NULL
)
1164 (void) strncpy(buf
, spa
->spa_root
, buflen
);
1168 spa_sync_pass(spa_t
*spa
)
1170 return (spa
->spa_sync_pass
);
1174 spa_name(spa_t
*spa
)
1176 return (spa
->spa_name
);
1180 spa_guid(spa_t
*spa
)
1183 * If we fail to parse the config during spa_load(), we can go through
1184 * the error path (which posts an ereport) and end up here with no root
1185 * vdev. We stash the original pool guid in 'spa_load_guid' to handle
1188 if (spa
->spa_root_vdev
!= NULL
)
1189 return (spa
->spa_root_vdev
->vdev_guid
);
1191 return (spa
->spa_load_guid
);
1195 spa_last_synced_txg(spa_t
*spa
)
1197 return (spa
->spa_ubsync
.ub_txg
);
1201 spa_first_txg(spa_t
*spa
)
1203 return (spa
->spa_first_txg
);
1207 spa_state(spa_t
*spa
)
1209 return (spa
->spa_state
);
1213 spa_freeze_txg(spa_t
*spa
)
1215 return (spa
->spa_freeze_txg
);
1219 * Return how much space is allocated in the pool (ie. sum of all asize)
1222 spa_get_alloc(spa_t
*spa
)
1224 return (spa
->spa_root_vdev
->vdev_stat
.vs_alloc
);
1228 * Return how much (raid-z inflated) space there is in the pool.
1231 spa_get_space(spa_t
*spa
)
1233 return (spa
->spa_root_vdev
->vdev_stat
.vs_space
);
1237 * Return the amount of raid-z-deflated space in the pool.
1240 spa_get_dspace(spa_t
*spa
)
1242 if (spa
->spa_deflate
)
1243 return (spa
->spa_root_vdev
->vdev_stat
.vs_dspace
);
1245 return (spa
->spa_root_vdev
->vdev_stat
.vs_space
);
1250 spa_get_asize(spa_t
*spa
, uint64_t lsize
)
1253 * For now, the worst case is 512-byte RAID-Z blocks, in which
1254 * case the space requirement is exactly 2x; so just assume that.
1255 * Add to this the fact that we can have up to 3 DVAs per bp, and
1256 * we have to multiply by a total of 6x.
1262 * Return the failure mode that has been set to this pool. The default
1263 * behavior will be to block all I/Os when a complete failure occurs.
1266 spa_get_failmode(spa_t
*spa
)
1268 return (spa
->spa_failmode
);
1272 spa_suspended(spa_t
*spa
)
1274 return (spa
->spa_suspended
);
1278 spa_version(spa_t
*spa
)
1280 return (spa
->spa_ubsync
.ub_version
);
1284 spa_max_replication(spa_t
*spa
)
1287 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1288 * handle BPs with more than one DVA allocated. Set our max
1289 * replication level accordingly.
1291 if (spa_version(spa
) < SPA_VERSION_DITTO_BLOCKS
)
1293 return (MIN(SPA_DVAS_PER_BP
, spa_max_replication_override
));
1297 bp_get_dasize(spa_t
*spa
, const blkptr_t
*bp
)
1301 if (!spa
->spa_deflate
)
1302 return (BP_GET_ASIZE(bp
));
1304 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
1305 for (i
= 0; i
< SPA_DVAS_PER_BP
; i
++) {
1307 vdev_lookup_top(spa
, DVA_GET_VDEV(&bp
->blk_dva
[i
]));
1309 sz
+= (DVA_GET_ASIZE(&bp
->blk_dva
[i
]) >>
1310 SPA_MINBLOCKSHIFT
) * vd
->vdev_deflate_ratio
;
1312 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
1317 * ==========================================================================
1318 * Initialization and Termination
1319 * ==========================================================================
1323 spa_name_compare(const void *a1
, const void *a2
)
1325 const spa_t
*s1
= a1
;
1326 const spa_t
*s2
= a2
;
1329 s
= strcmp(s1
->spa_name
, s2
->spa_name
);
1340 return (spa_active_count
);
1352 mutex_init(&spa_namespace_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1353 mutex_init(&spa_spare_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1354 mutex_init(&spa_l2cache_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1355 cv_init(&spa_namespace_cv
, NULL
, CV_DEFAULT
, NULL
);
1357 avl_create(&spa_namespace_avl
, spa_name_compare
, sizeof (spa_t
),
1358 offsetof(spa_t
, spa_avl
));
1360 avl_create(&spa_spare_avl
, spa_spare_compare
, sizeof (spa_aux_t
),
1361 offsetof(spa_aux_t
, aux_avl
));
1363 avl_create(&spa_l2cache_avl
, spa_l2cache_compare
, sizeof (spa_aux_t
),
1364 offsetof(spa_aux_t
, aux_avl
));
1366 spa_mode_global
= mode
;
1373 vdev_cache_stat_init();
1387 vdev_cache_stat_fini();
1394 avl_destroy(&spa_namespace_avl
);
1395 avl_destroy(&spa_spare_avl
);
1396 avl_destroy(&spa_l2cache_avl
);
1398 cv_destroy(&spa_namespace_cv
);
1399 mutex_destroy(&spa_namespace_lock
);
1400 mutex_destroy(&spa_spare_lock
);
1401 mutex_destroy(&spa_l2cache_lock
);
1405 * Return whether this pool has slogs. No locking needed.
1406 * It's not a problem if the wrong answer is returned as it's only for
1407 * performance and not correctness
1410 spa_has_slogs(spa_t
*spa
)
1412 return (spa
->spa_log_class
->mc_rotor
!= NULL
);
1416 * Return whether this pool is the root pool.
1419 spa_is_root(spa_t
*spa
)
1421 return (spa
->spa_is_root
);
1425 spa_writeable(spa_t
*spa
)
1427 return (!!(spa
->spa_mode
& FWRITE
));
1431 spa_mode(spa_t
*spa
)
1433 return (spa
->spa_mode
);