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.
253 * ==========================================================================
255 * ==========================================================================
258 spa_config_lock_init(spa_t
*spa
)
260 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
261 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
262 mutex_init(&scl
->scl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
263 cv_init(&scl
->scl_cv
, NULL
, CV_DEFAULT
, NULL
);
264 refcount_create(&scl
->scl_count
);
265 scl
->scl_writer
= NULL
;
266 scl
->scl_write_wanted
= 0;
271 spa_config_lock_destroy(spa_t
*spa
)
273 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
274 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
275 mutex_destroy(&scl
->scl_lock
);
276 cv_destroy(&scl
->scl_cv
);
277 refcount_destroy(&scl
->scl_count
);
278 ASSERT(scl
->scl_writer
== NULL
);
279 ASSERT(scl
->scl_write_wanted
== 0);
284 spa_config_tryenter(spa_t
*spa
, int locks
, void *tag
, krw_t rw
)
286 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
287 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
288 if (!(locks
& (1 << i
)))
290 mutex_enter(&scl
->scl_lock
);
291 if (rw
== RW_READER
) {
292 if (scl
->scl_writer
|| scl
->scl_write_wanted
) {
293 mutex_exit(&scl
->scl_lock
);
294 spa_config_exit(spa
, locks
^ (1 << i
), tag
);
298 ASSERT(scl
->scl_writer
!= curthread
);
299 if (!refcount_is_zero(&scl
->scl_count
)) {
300 mutex_exit(&scl
->scl_lock
);
301 spa_config_exit(spa
, locks
^ (1 << i
), tag
);
304 scl
->scl_writer
= curthread
;
306 (void) refcount_add(&scl
->scl_count
, tag
);
307 mutex_exit(&scl
->scl_lock
);
313 spa_config_enter(spa_t
*spa
, int locks
, void *tag
, krw_t rw
)
317 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
318 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
319 if (scl
->scl_writer
== curthread
)
320 wlocks_held
|= (1 << i
);
321 if (!(locks
& (1 << i
)))
323 mutex_enter(&scl
->scl_lock
);
324 if (rw
== RW_READER
) {
325 while (scl
->scl_writer
|| scl
->scl_write_wanted
) {
326 cv_wait(&scl
->scl_cv
, &scl
->scl_lock
);
329 ASSERT(scl
->scl_writer
!= curthread
);
330 while (!refcount_is_zero(&scl
->scl_count
)) {
331 scl
->scl_write_wanted
++;
332 cv_wait(&scl
->scl_cv
, &scl
->scl_lock
);
333 scl
->scl_write_wanted
--;
335 scl
->scl_writer
= curthread
;
337 (void) refcount_add(&scl
->scl_count
, tag
);
338 mutex_exit(&scl
->scl_lock
);
340 ASSERT(wlocks_held
<= locks
);
344 spa_config_exit(spa_t
*spa
, int locks
, void *tag
)
346 for (int i
= SCL_LOCKS
- 1; i
>= 0; i
--) {
347 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
348 if (!(locks
& (1 << i
)))
350 mutex_enter(&scl
->scl_lock
);
351 ASSERT(!refcount_is_zero(&scl
->scl_count
));
352 if (refcount_remove(&scl
->scl_count
, tag
) == 0) {
353 ASSERT(scl
->scl_writer
== NULL
||
354 scl
->scl_writer
== curthread
);
355 scl
->scl_writer
= NULL
; /* OK in either case */
356 cv_broadcast(&scl
->scl_cv
);
358 mutex_exit(&scl
->scl_lock
);
363 spa_config_held(spa_t
*spa
, int locks
, krw_t rw
)
367 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
368 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
369 if (!(locks
& (1 << i
)))
371 if ((rw
== RW_READER
&& !refcount_is_zero(&scl
->scl_count
)) ||
372 (rw
== RW_WRITER
&& scl
->scl_writer
== curthread
))
373 locks_held
|= 1 << i
;
380 * ==========================================================================
381 * SPA namespace functions
382 * ==========================================================================
386 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
387 * Returns NULL if no matching spa_t is found.
390 spa_lookup(const char *name
)
392 static spa_t search
; /* spa_t is large; don't allocate on stack */
398 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
401 * If it's a full dataset name, figure out the pool name and
404 cp
= strpbrk(name
, "/@");
410 (void) strlcpy(search
.spa_name
, name
, sizeof (search
.spa_name
));
411 spa
= avl_find(&spa_namespace_avl
, &search
, &where
);
420 * Create an uninitialized spa_t with the given name. Requires
421 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
422 * exist by calling spa_lookup() first.
425 spa_add(const char *name
, nvlist_t
*config
, const char *altroot
)
428 spa_config_dirent_t
*dp
;
430 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
432 spa
= kmem_zalloc(sizeof (spa_t
), KM_SLEEP
);
434 mutex_init(&spa
->spa_async_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
435 mutex_init(&spa
->spa_errlist_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
436 mutex_init(&spa
->spa_errlog_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
437 mutex_init(&spa
->spa_history_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
438 mutex_init(&spa
->spa_proc_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
439 mutex_init(&spa
->spa_props_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
440 mutex_init(&spa
->spa_scrub_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
441 mutex_init(&spa
->spa_suspend_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
442 mutex_init(&spa
->spa_vdev_top_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
444 cv_init(&spa
->spa_async_cv
, NULL
, CV_DEFAULT
, NULL
);
445 cv_init(&spa
->spa_proc_cv
, NULL
, CV_DEFAULT
, NULL
);
446 cv_init(&spa
->spa_scrub_io_cv
, NULL
, CV_DEFAULT
, NULL
);
447 cv_init(&spa
->spa_suspend_cv
, NULL
, CV_DEFAULT
, NULL
);
449 for (int t
= 0; t
< TXG_SIZE
; t
++)
450 bplist_create(&spa
->spa_free_bplist
[t
]);
452 (void) strlcpy(spa
->spa_name
, name
, sizeof (spa
->spa_name
));
453 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
454 spa
->spa_freeze_txg
= UINT64_MAX
;
455 spa
->spa_final_txg
= UINT64_MAX
;
456 spa
->spa_load_max_txg
= UINT64_MAX
;
458 spa
->spa_proc_state
= SPA_PROC_NONE
;
460 refcount_create(&spa
->spa_refcount
);
461 spa_config_lock_init(spa
);
463 avl_add(&spa_namespace_avl
, spa
);
466 * Set the alternate root, if there is one.
469 spa
->spa_root
= spa_strdup(altroot
);
474 * Every pool starts with the default cachefile
476 list_create(&spa
->spa_config_list
, sizeof (spa_config_dirent_t
),
477 offsetof(spa_config_dirent_t
, scd_link
));
479 dp
= kmem_zalloc(sizeof (spa_config_dirent_t
), KM_SLEEP
);
480 dp
->scd_path
= altroot
? NULL
: spa_strdup(spa_config_path
);
481 list_insert_head(&spa
->spa_config_list
, dp
);
483 VERIFY(nvlist_alloc(&spa
->spa_load_info
, NV_UNIQUE_NAME
,
487 VERIFY(nvlist_dup(config
, &spa
->spa_config
, 0) == 0);
493 * Removes a spa_t from the namespace, freeing up any memory used. Requires
494 * spa_namespace_lock. This is called only after the spa_t has been closed and
498 spa_remove(spa_t
*spa
)
500 spa_config_dirent_t
*dp
;
502 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
503 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
505 nvlist_free(spa
->spa_config_splitting
);
507 avl_remove(&spa_namespace_avl
, spa
);
508 cv_broadcast(&spa_namespace_cv
);
511 spa_strfree(spa
->spa_root
);
515 while ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
516 list_remove(&spa
->spa_config_list
, dp
);
517 if (dp
->scd_path
!= NULL
)
518 spa_strfree(dp
->scd_path
);
519 kmem_free(dp
, sizeof (spa_config_dirent_t
));
522 list_destroy(&spa
->spa_config_list
);
524 nvlist_free(spa
->spa_load_info
);
525 spa_config_set(spa
, NULL
);
527 refcount_destroy(&spa
->spa_refcount
);
529 spa_config_lock_destroy(spa
);
531 for (int t
= 0; t
< TXG_SIZE
; t
++)
532 bplist_destroy(&spa
->spa_free_bplist
[t
]);
534 cv_destroy(&spa
->spa_async_cv
);
535 cv_destroy(&spa
->spa_proc_cv
);
536 cv_destroy(&spa
->spa_scrub_io_cv
);
537 cv_destroy(&spa
->spa_suspend_cv
);
539 mutex_destroy(&spa
->spa_async_lock
);
540 mutex_destroy(&spa
->spa_errlist_lock
);
541 mutex_destroy(&spa
->spa_errlog_lock
);
542 mutex_destroy(&spa
->spa_history_lock
);
543 mutex_destroy(&spa
->spa_proc_lock
);
544 mutex_destroy(&spa
->spa_props_lock
);
545 mutex_destroy(&spa
->spa_scrub_lock
);
546 mutex_destroy(&spa
->spa_suspend_lock
);
547 mutex_destroy(&spa
->spa_vdev_top_lock
);
549 kmem_free(spa
, sizeof (spa_t
));
553 * Given a pool, return the next pool in the namespace, or NULL if there is
554 * none. If 'prev' is NULL, return the first pool.
557 spa_next(spa_t
*prev
)
559 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
562 return (AVL_NEXT(&spa_namespace_avl
, prev
));
564 return (avl_first(&spa_namespace_avl
));
568 * ==========================================================================
569 * SPA refcount functions
570 * ==========================================================================
574 * Add a reference to the given spa_t. Must have at least one reference, or
575 * have the namespace lock held.
578 spa_open_ref(spa_t
*spa
, void *tag
)
580 ASSERT(refcount_count(&spa
->spa_refcount
) >= spa
->spa_minref
||
581 MUTEX_HELD(&spa_namespace_lock
));
582 (void) refcount_add(&spa
->spa_refcount
, tag
);
586 * Remove a reference to the given spa_t. Must have at least one reference, or
587 * have the namespace lock held.
590 spa_close(spa_t
*spa
, void *tag
)
592 ASSERT(refcount_count(&spa
->spa_refcount
) > spa
->spa_minref
||
593 MUTEX_HELD(&spa_namespace_lock
));
594 (void) refcount_remove(&spa
->spa_refcount
, tag
);
598 * Check to see if the spa refcount is zero. Must be called with
599 * spa_namespace_lock held. We really compare against spa_minref, which is the
600 * number of references acquired when opening a pool
603 spa_refcount_zero(spa_t
*spa
)
605 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
607 return (refcount_count(&spa
->spa_refcount
) == spa
->spa_minref
);
611 * ==========================================================================
612 * SPA spare and l2cache tracking
613 * ==========================================================================
617 * Hot spares and cache devices are tracked using the same code below,
618 * for 'auxiliary' devices.
621 typedef struct spa_aux
{
629 spa_aux_compare(const void *a
, const void *b
)
631 const spa_aux_t
*sa
= a
;
632 const spa_aux_t
*sb
= b
;
634 if (sa
->aux_guid
< sb
->aux_guid
)
636 else if (sa
->aux_guid
> sb
->aux_guid
)
643 spa_aux_add(vdev_t
*vd
, avl_tree_t
*avl
)
649 search
.aux_guid
= vd
->vdev_guid
;
650 if ((aux
= avl_find(avl
, &search
, &where
)) != NULL
) {
653 aux
= kmem_zalloc(sizeof (spa_aux_t
), KM_SLEEP
);
654 aux
->aux_guid
= vd
->vdev_guid
;
656 avl_insert(avl
, aux
, where
);
661 spa_aux_remove(vdev_t
*vd
, avl_tree_t
*avl
)
667 search
.aux_guid
= vd
->vdev_guid
;
668 aux
= avl_find(avl
, &search
, &where
);
672 if (--aux
->aux_count
== 0) {
673 avl_remove(avl
, aux
);
674 kmem_free(aux
, sizeof (spa_aux_t
));
675 } else if (aux
->aux_pool
== spa_guid(vd
->vdev_spa
)) {
676 aux
->aux_pool
= 0ULL;
681 spa_aux_exists(uint64_t guid
, uint64_t *pool
, int *refcnt
, avl_tree_t
*avl
)
683 spa_aux_t search
, *found
;
685 search
.aux_guid
= guid
;
686 found
= avl_find(avl
, &search
, NULL
);
690 *pool
= found
->aux_pool
;
697 *refcnt
= found
->aux_count
;
702 return (found
!= NULL
);
706 spa_aux_activate(vdev_t
*vd
, avl_tree_t
*avl
)
708 spa_aux_t search
, *found
;
711 search
.aux_guid
= vd
->vdev_guid
;
712 found
= avl_find(avl
, &search
, &where
);
713 ASSERT(found
!= NULL
);
714 ASSERT(found
->aux_pool
== 0ULL);
716 found
->aux_pool
= spa_guid(vd
->vdev_spa
);
720 * Spares are tracked globally due to the following constraints:
722 * - A spare may be part of multiple pools.
723 * - A spare may be added to a pool even if it's actively in use within
725 * - A spare in use in any pool can only be the source of a replacement if
726 * the target is a spare in the same pool.
728 * We keep track of all spares on the system through the use of a reference
729 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
730 * spare, then we bump the reference count in the AVL tree. In addition, we set
731 * the 'vdev_isspare' member to indicate that the device is a spare (active or
732 * inactive). When a spare is made active (used to replace a device in the
733 * pool), we also keep track of which pool its been made a part of.
735 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
736 * called under the spa_namespace lock as part of vdev reconfiguration. The
737 * separate spare lock exists for the status query path, which does not need to
738 * be completely consistent with respect to other vdev configuration changes.
742 spa_spare_compare(const void *a
, const void *b
)
744 return (spa_aux_compare(a
, b
));
748 spa_spare_add(vdev_t
*vd
)
750 mutex_enter(&spa_spare_lock
);
751 ASSERT(!vd
->vdev_isspare
);
752 spa_aux_add(vd
, &spa_spare_avl
);
753 vd
->vdev_isspare
= B_TRUE
;
754 mutex_exit(&spa_spare_lock
);
758 spa_spare_remove(vdev_t
*vd
)
760 mutex_enter(&spa_spare_lock
);
761 ASSERT(vd
->vdev_isspare
);
762 spa_aux_remove(vd
, &spa_spare_avl
);
763 vd
->vdev_isspare
= B_FALSE
;
764 mutex_exit(&spa_spare_lock
);
768 spa_spare_exists(uint64_t guid
, uint64_t *pool
, int *refcnt
)
772 mutex_enter(&spa_spare_lock
);
773 found
= spa_aux_exists(guid
, pool
, refcnt
, &spa_spare_avl
);
774 mutex_exit(&spa_spare_lock
);
780 spa_spare_activate(vdev_t
*vd
)
782 mutex_enter(&spa_spare_lock
);
783 ASSERT(vd
->vdev_isspare
);
784 spa_aux_activate(vd
, &spa_spare_avl
);
785 mutex_exit(&spa_spare_lock
);
789 * Level 2 ARC devices are tracked globally for the same reasons as spares.
790 * Cache devices currently only support one pool per cache device, and so
791 * for these devices the aux reference count is currently unused beyond 1.
795 spa_l2cache_compare(const void *a
, const void *b
)
797 return (spa_aux_compare(a
, b
));
801 spa_l2cache_add(vdev_t
*vd
)
803 mutex_enter(&spa_l2cache_lock
);
804 ASSERT(!vd
->vdev_isl2cache
);
805 spa_aux_add(vd
, &spa_l2cache_avl
);
806 vd
->vdev_isl2cache
= B_TRUE
;
807 mutex_exit(&spa_l2cache_lock
);
811 spa_l2cache_remove(vdev_t
*vd
)
813 mutex_enter(&spa_l2cache_lock
);
814 ASSERT(vd
->vdev_isl2cache
);
815 spa_aux_remove(vd
, &spa_l2cache_avl
);
816 vd
->vdev_isl2cache
= B_FALSE
;
817 mutex_exit(&spa_l2cache_lock
);
821 spa_l2cache_exists(uint64_t guid
, uint64_t *pool
)
825 mutex_enter(&spa_l2cache_lock
);
826 found
= spa_aux_exists(guid
, pool
, NULL
, &spa_l2cache_avl
);
827 mutex_exit(&spa_l2cache_lock
);
833 spa_l2cache_activate(vdev_t
*vd
)
835 mutex_enter(&spa_l2cache_lock
);
836 ASSERT(vd
->vdev_isl2cache
);
837 spa_aux_activate(vd
, &spa_l2cache_avl
);
838 mutex_exit(&spa_l2cache_lock
);
842 * ==========================================================================
844 * ==========================================================================
848 * Lock the given spa_t for the purpose of adding or removing a vdev.
849 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
850 * It returns the next transaction group for the spa_t.
853 spa_vdev_enter(spa_t
*spa
)
855 mutex_enter(&spa
->spa_vdev_top_lock
);
856 mutex_enter(&spa_namespace_lock
);
857 return (spa_vdev_config_enter(spa
));
861 * Internal implementation for spa_vdev_enter(). Used when a vdev
862 * operation requires multiple syncs (i.e. removing a device) while
863 * keeping the spa_namespace_lock held.
866 spa_vdev_config_enter(spa_t
*spa
)
868 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
870 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
872 return (spa_last_synced_txg(spa
) + 1);
876 * Used in combination with spa_vdev_config_enter() to allow the syncing
877 * of multiple transactions without releasing the spa_namespace_lock.
880 spa_vdev_config_exit(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int error
, char *tag
)
882 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
884 int config_changed
= B_FALSE
;
886 ASSERT(txg
> spa_last_synced_txg(spa
));
888 spa
->spa_pending_vdev
= NULL
;
893 vdev_dtl_reassess(spa
->spa_root_vdev
, 0, 0, B_FALSE
);
895 if (error
== 0 && !list_is_empty(&spa
->spa_config_dirty_list
)) {
896 config_changed
= B_TRUE
;
897 spa
->spa_config_generation
++;
901 * Verify the metaslab classes.
903 ASSERT(metaslab_class_validate(spa_normal_class(spa
)) == 0);
904 ASSERT(metaslab_class_validate(spa_log_class(spa
)) == 0);
906 spa_config_exit(spa
, SCL_ALL
, spa
);
909 * Panic the system if the specified tag requires it. This
910 * is useful for ensuring that configurations are updated
913 if (zio_injection_enabled
)
914 zio_handle_panic_injection(spa
, tag
, 0);
917 * Note: this txg_wait_synced() is important because it ensures
918 * that there won't be more than one config change per txg.
919 * This allows us to use the txg as the generation number.
922 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
925 ASSERT(!vd
->vdev_detached
|| vd
->vdev_dtl_smo
.smo_object
== 0);
926 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
928 spa_config_exit(spa
, SCL_ALL
, spa
);
932 * If the config changed, update the config cache.
935 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
939 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
940 * locking of spa_vdev_enter(), we also want make sure the transactions have
941 * synced to disk, and then update the global configuration cache with the new
945 spa_vdev_exit(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int error
)
947 spa_vdev_config_exit(spa
, vd
, txg
, error
, FTAG
);
948 mutex_exit(&spa_namespace_lock
);
949 mutex_exit(&spa
->spa_vdev_top_lock
);
955 * Lock the given spa_t for the purpose of changing vdev state.
958 spa_vdev_state_enter(spa_t
*spa
, int oplocks
)
960 int locks
= SCL_STATE_ALL
| oplocks
;
963 * Root pools may need to read of the underlying devfs filesystem
964 * when opening up a vdev. Unfortunately if we're holding the
965 * SCL_ZIO lock it will result in a deadlock when we try to issue
966 * the read from the root filesystem. Instead we "prefetch"
967 * the associated vnodes that we need prior to opening the
968 * underlying devices and cache them so that we can prevent
969 * any I/O when we are doing the actual open.
971 if (spa_is_root(spa
)) {
972 int low
= locks
& ~(SCL_ZIO
- 1);
973 int high
= locks
& ~low
;
975 spa_config_enter(spa
, high
, spa
, RW_WRITER
);
976 vdev_hold(spa
->spa_root_vdev
);
977 spa_config_enter(spa
, low
, spa
, RW_WRITER
);
979 spa_config_enter(spa
, locks
, spa
, RW_WRITER
);
981 spa
->spa_vdev_locks
= locks
;
985 spa_vdev_state_exit(spa_t
*spa
, vdev_t
*vd
, int error
)
987 boolean_t config_changed
= B_FALSE
;
989 if (vd
!= NULL
|| error
== 0)
990 vdev_dtl_reassess(vd
? vd
->vdev_top
: spa
->spa_root_vdev
,
994 vdev_state_dirty(vd
->vdev_top
);
995 config_changed
= B_TRUE
;
996 spa
->spa_config_generation
++;
999 if (spa_is_root(spa
))
1000 vdev_rele(spa
->spa_root_vdev
);
1002 ASSERT3U(spa
->spa_vdev_locks
, >=, SCL_STATE_ALL
);
1003 spa_config_exit(spa
, spa
->spa_vdev_locks
, spa
);
1006 * If anything changed, wait for it to sync. This ensures that,
1007 * from the system administrator's perspective, zpool(1M) commands
1008 * are synchronous. This is important for things like zpool offline:
1009 * when the command completes, you expect no further I/O from ZFS.
1012 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1015 * If the config changed, update the config cache.
1017 if (config_changed
) {
1018 mutex_enter(&spa_namespace_lock
);
1019 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
1020 mutex_exit(&spa_namespace_lock
);
1027 * ==========================================================================
1028 * Miscellaneous functions
1029 * ==========================================================================
1036 spa_rename(const char *name
, const char *newname
)
1042 * Lookup the spa_t and grab the config lock for writing. We need to
1043 * actually open the pool so that we can sync out the necessary labels.
1044 * It's OK to call spa_open() with the namespace lock held because we
1045 * allow recursive calls for other reasons.
1047 mutex_enter(&spa_namespace_lock
);
1048 if ((err
= spa_open(name
, &spa
, FTAG
)) != 0) {
1049 mutex_exit(&spa_namespace_lock
);
1053 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1055 avl_remove(&spa_namespace_avl
, spa
);
1056 (void) strlcpy(spa
->spa_name
, newname
, sizeof (spa
->spa_name
));
1057 avl_add(&spa_namespace_avl
, spa
);
1060 * Sync all labels to disk with the new names by marking the root vdev
1061 * dirty and waiting for it to sync. It will pick up the new pool name
1064 vdev_config_dirty(spa
->spa_root_vdev
);
1066 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1068 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1071 * Sync the updated config cache.
1073 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
1075 spa_close(spa
, FTAG
);
1077 mutex_exit(&spa_namespace_lock
);
1083 * Return the spa_t associated with given pool_guid, if it exists. If
1084 * device_guid is non-zero, determine whether the pool exists *and* contains
1085 * a device with the specified device_guid.
1088 spa_by_guid(uint64_t pool_guid
, uint64_t device_guid
)
1091 avl_tree_t
*t
= &spa_namespace_avl
;
1093 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1095 for (spa
= avl_first(t
); spa
!= NULL
; spa
= AVL_NEXT(t
, spa
)) {
1096 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1098 if (spa
->spa_root_vdev
== NULL
)
1100 if (spa_guid(spa
) == pool_guid
) {
1101 if (device_guid
== 0)
1104 if (vdev_lookup_by_guid(spa
->spa_root_vdev
,
1105 device_guid
) != NULL
)
1109 * Check any devices we may be in the process of adding.
1111 if (spa
->spa_pending_vdev
) {
1112 if (vdev_lookup_by_guid(spa
->spa_pending_vdev
,
1113 device_guid
) != NULL
)
1123 * Determine whether a pool with the given pool_guid exists.
1126 spa_guid_exists(uint64_t pool_guid
, uint64_t device_guid
)
1128 return (spa_by_guid(pool_guid
, device_guid
) != NULL
);
1132 spa_strdup(const char *s
)
1138 new = kmem_alloc(len
+ 1, KM_SLEEP
);
1146 spa_strfree(char *s
)
1148 kmem_free(s
, strlen(s
) + 1);
1152 spa_get_random(uint64_t range
)
1158 (void) random_get_pseudo_bytes((void *)&r
, sizeof (uint64_t));
1164 spa_generate_guid(spa_t
*spa
)
1166 uint64_t guid
= spa_get_random(-1ULL);
1169 while (guid
== 0 || spa_guid_exists(spa_guid(spa
), guid
))
1170 guid
= spa_get_random(-1ULL);
1172 while (guid
== 0 || spa_guid_exists(guid
, 0))
1173 guid
= spa_get_random(-1ULL);
1180 sprintf_blkptr(char *buf
, const blkptr_t
*bp
)
1183 char *checksum
= NULL
;
1184 char *compress
= NULL
;
1187 type
= dmu_ot
[BP_GET_TYPE(bp
)].ot_name
;
1188 checksum
= zio_checksum_table
[BP_GET_CHECKSUM(bp
)].ci_name
;
1189 compress
= zio_compress_table
[BP_GET_COMPRESS(bp
)].ci_name
;
1192 SPRINTF_BLKPTR(snprintf
, ' ', buf
, bp
, type
, checksum
, compress
);
1196 spa_freeze(spa_t
*spa
)
1198 uint64_t freeze_txg
= 0;
1200 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1201 if (spa
->spa_freeze_txg
== UINT64_MAX
) {
1202 freeze_txg
= spa_last_synced_txg(spa
) + TXG_SIZE
;
1203 spa
->spa_freeze_txg
= freeze_txg
;
1205 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1206 if (freeze_txg
!= 0)
1207 txg_wait_synced(spa_get_dsl(spa
), freeze_txg
);
1211 zfs_panic_recover(const char *fmt
, ...)
1216 vcmn_err(zfs_recover
? CE_WARN
: CE_PANIC
, fmt
, adx
);
1221 * This is a stripped-down version of strtoull, suitable only for converting
1222 * lowercase hexidecimal numbers that don't overflow.
1225 strtonum(const char *str
, char **nptr
)
1231 while ((c
= *str
) != '\0') {
1232 if (c
>= '0' && c
<= '9')
1234 else if (c
>= 'a' && c
<= 'f')
1235 digit
= 10 + c
- 'a';
1246 *nptr
= (char *)str
;
1252 * ==========================================================================
1253 * Accessor functions
1254 * ==========================================================================
1258 spa_shutting_down(spa_t
*spa
)
1260 return (spa
->spa_async_suspended
);
1264 spa_get_dsl(spa_t
*spa
)
1266 return (spa
->spa_dsl_pool
);
1270 spa_get_rootblkptr(spa_t
*spa
)
1272 return (&spa
->spa_ubsync
.ub_rootbp
);
1276 spa_set_rootblkptr(spa_t
*spa
, const blkptr_t
*bp
)
1278 spa
->spa_uberblock
.ub_rootbp
= *bp
;
1282 spa_altroot(spa_t
*spa
, char *buf
, size_t buflen
)
1284 if (spa
->spa_root
== NULL
)
1287 (void) strncpy(buf
, spa
->spa_root
, buflen
);
1291 spa_sync_pass(spa_t
*spa
)
1293 return (spa
->spa_sync_pass
);
1297 spa_name(spa_t
*spa
)
1299 return (spa
->spa_name
);
1303 spa_guid(spa_t
*spa
)
1306 * If we fail to parse the config during spa_load(), we can go through
1307 * the error path (which posts an ereport) and end up here with no root
1308 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1311 if (spa
->spa_root_vdev
!= NULL
)
1312 return (spa
->spa_root_vdev
->vdev_guid
);
1314 return (spa
->spa_config_guid
);
1318 spa_load_guid(spa_t
*spa
)
1321 * This is a GUID that exists solely as a reference for the
1322 * purposes of the arc. It is generated at load time, and
1323 * is never written to persistent storage.
1325 return (spa
->spa_load_guid
);
1329 spa_last_synced_txg(spa_t
*spa
)
1331 return (spa
->spa_ubsync
.ub_txg
);
1335 spa_first_txg(spa_t
*spa
)
1337 return (spa
->spa_first_txg
);
1341 spa_syncing_txg(spa_t
*spa
)
1343 return (spa
->spa_syncing_txg
);
1347 spa_state(spa_t
*spa
)
1349 return (spa
->spa_state
);
1353 spa_load_state(spa_t
*spa
)
1355 return (spa
->spa_load_state
);
1359 spa_freeze_txg(spa_t
*spa
)
1361 return (spa
->spa_freeze_txg
);
1366 spa_get_asize(spa_t
*spa
, uint64_t lsize
)
1369 * The worst case is single-sector max-parity RAID-Z blocks, in which
1370 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
1371 * times the size; so just assume that. Add to this the fact that
1372 * we can have up to 3 DVAs per bp, and one more factor of 2 because
1373 * the block may be dittoed with up to 3 DVAs by ddt_sync().
1375 return (lsize
* (VDEV_RAIDZ_MAXPARITY
+ 1) * SPA_DVAS_PER_BP
* 2);
1379 spa_get_dspace(spa_t
*spa
)
1381 return (spa
->spa_dspace
);
1385 spa_update_dspace(spa_t
*spa
)
1387 spa
->spa_dspace
= metaslab_class_get_dspace(spa_normal_class(spa
)) +
1388 ddt_get_dedup_dspace(spa
);
1392 * Return the failure mode that has been set to this pool. The default
1393 * behavior will be to block all I/Os when a complete failure occurs.
1396 spa_get_failmode(spa_t
*spa
)
1398 return (spa
->spa_failmode
);
1402 spa_suspended(spa_t
*spa
)
1404 return (spa
->spa_suspended
);
1408 spa_version(spa_t
*spa
)
1410 return (spa
->spa_ubsync
.ub_version
);
1414 spa_deflate(spa_t
*spa
)
1416 return (spa
->spa_deflate
);
1420 spa_normal_class(spa_t
*spa
)
1422 return (spa
->spa_normal_class
);
1426 spa_log_class(spa_t
*spa
)
1428 return (spa
->spa_log_class
);
1432 spa_max_replication(spa_t
*spa
)
1435 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1436 * handle BPs with more than one DVA allocated. Set our max
1437 * replication level accordingly.
1439 if (spa_version(spa
) < SPA_VERSION_DITTO_BLOCKS
)
1441 return (MIN(SPA_DVAS_PER_BP
, spa_max_replication_override
));
1445 spa_prev_software_version(spa_t
*spa
)
1447 return (spa
->spa_prev_software_version
);
1451 dva_get_dsize_sync(spa_t
*spa
, const dva_t
*dva
)
1453 uint64_t asize
= DVA_GET_ASIZE(dva
);
1454 uint64_t dsize
= asize
;
1456 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1458 if (asize
!= 0 && spa
->spa_deflate
) {
1459 vdev_t
*vd
= vdev_lookup_top(spa
, DVA_GET_VDEV(dva
));
1460 dsize
= (asize
>> SPA_MINBLOCKSHIFT
) * vd
->vdev_deflate_ratio
;
1467 bp_get_dsize_sync(spa_t
*spa
, const blkptr_t
*bp
)
1471 for (int d
= 0; d
< SPA_DVAS_PER_BP
; d
++)
1472 dsize
+= dva_get_dsize_sync(spa
, &bp
->blk_dva
[d
]);
1478 bp_get_dsize(spa_t
*spa
, const blkptr_t
*bp
)
1482 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
1484 for (int d
= 0; d
< SPA_DVAS_PER_BP
; d
++)
1485 dsize
+= dva_get_dsize_sync(spa
, &bp
->blk_dva
[d
]);
1487 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
1493 * ==========================================================================
1494 * Initialization and Termination
1495 * ==========================================================================
1499 spa_name_compare(const void *a1
, const void *a2
)
1501 const spa_t
*s1
= a1
;
1502 const spa_t
*s2
= a2
;
1505 s
= strcmp(s1
->spa_name
, s2
->spa_name
);
1516 return (spa_active_count
);
1528 mutex_init(&spa_namespace_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1529 mutex_init(&spa_spare_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1530 mutex_init(&spa_l2cache_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1531 cv_init(&spa_namespace_cv
, NULL
, CV_DEFAULT
, NULL
);
1533 avl_create(&spa_namespace_avl
, spa_name_compare
, sizeof (spa_t
),
1534 offsetof(spa_t
, spa_avl
));
1536 avl_create(&spa_spare_avl
, spa_spare_compare
, sizeof (spa_aux_t
),
1537 offsetof(spa_aux_t
, aux_avl
));
1539 avl_create(&spa_l2cache_avl
, spa_l2cache_compare
, sizeof (spa_aux_t
),
1540 offsetof(spa_aux_t
, aux_avl
));
1542 spa_mode_global
= mode
;
1549 vdev_cache_stat_init();
1563 vdev_cache_stat_fini();
1570 avl_destroy(&spa_namespace_avl
);
1571 avl_destroy(&spa_spare_avl
);
1572 avl_destroy(&spa_l2cache_avl
);
1574 cv_destroy(&spa_namespace_cv
);
1575 mutex_destroy(&spa_namespace_lock
);
1576 mutex_destroy(&spa_spare_lock
);
1577 mutex_destroy(&spa_l2cache_lock
);
1581 * Return whether this pool has slogs. No locking needed.
1582 * It's not a problem if the wrong answer is returned as it's only for
1583 * performance and not correctness
1586 spa_has_slogs(spa_t
*spa
)
1588 return (spa
->spa_log_class
->mc_rotor
!= NULL
);
1592 spa_get_log_state(spa_t
*spa
)
1594 return (spa
->spa_log_state
);
1598 spa_set_log_state(spa_t
*spa
, spa_log_state_t state
)
1600 spa
->spa_log_state
= state
;
1604 spa_is_root(spa_t
*spa
)
1606 return (spa
->spa_is_root
);
1610 spa_writeable(spa_t
*spa
)
1612 return (!!(spa
->spa_mode
& FWRITE
));
1616 spa_mode(spa_t
*spa
)
1618 return (spa
->spa_mode
);
1622 spa_bootfs(spa_t
*spa
)
1624 return (spa
->spa_bootfs
);
1628 spa_delegation(spa_t
*spa
)
1630 return (spa
->spa_delegation
);
1634 spa_meta_objset(spa_t
*spa
)
1636 return (spa
->spa_meta_objset
);
1640 spa_dedup_checksum(spa_t
*spa
)
1642 return (spa
->spa_dedup_checksum
);
1646 * Reset pool scan stat per scan pass (or reboot).
1649 spa_scan_stat_init(spa_t
*spa
)
1651 /* data not stored on disk */
1652 spa
->spa_scan_pass_start
= gethrestime_sec();
1653 spa
->spa_scan_pass_exam
= 0;
1654 vdev_scan_stat_init(spa
->spa_root_vdev
);
1658 * Get scan stats for zpool status reports
1661 spa_scan_get_stats(spa_t
*spa
, pool_scan_stat_t
*ps
)
1663 dsl_scan_t
*scn
= spa
->spa_dsl_pool
? spa
->spa_dsl_pool
->dp_scan
: NULL
;
1665 if (scn
== NULL
|| scn
->scn_phys
.scn_func
== POOL_SCAN_NONE
)
1667 bzero(ps
, sizeof (pool_scan_stat_t
));
1669 /* data stored on disk */
1670 ps
->pss_func
= scn
->scn_phys
.scn_func
;
1671 ps
->pss_start_time
= scn
->scn_phys
.scn_start_time
;
1672 ps
->pss_end_time
= scn
->scn_phys
.scn_end_time
;
1673 ps
->pss_to_examine
= scn
->scn_phys
.scn_to_examine
;
1674 ps
->pss_examined
= scn
->scn_phys
.scn_examined
;
1675 ps
->pss_to_process
= scn
->scn_phys
.scn_to_process
;
1676 ps
->pss_processed
= scn
->scn_phys
.scn_processed
;
1677 ps
->pss_errors
= scn
->scn_phys
.scn_errors
;
1678 ps
->pss_state
= scn
->scn_phys
.scn_state
;
1680 /* data not stored on disk */
1681 ps
->pss_pass_start
= spa
->spa_scan_pass_start
;
1682 ps
->pss_pass_exam
= spa
->spa_scan_pass_exam
;
1688 spa_debug_enabled(spa_t
*spa
)
1690 return (spa
->spa_debug
);