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) 2013 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/vdev_file.h>
38 #include <sys/metaslab.h>
39 #include <sys/uberblock_impl.h>
42 #include <sys/unique.h>
43 #include <sys/dsl_pool.h>
44 #include <sys/dsl_dir.h>
45 #include <sys/dsl_prop.h>
46 #include <sys/fm/util.h>
47 #include <sys/dsl_scan.h>
48 #include <sys/fs/zfs.h>
49 #include <sys/metaslab_impl.h>
52 #include <sys/kstat.h>
54 #include "zfeature_common.h"
59 * There are four basic locks for managing spa_t structures:
61 * spa_namespace_lock (global mutex)
63 * This lock must be acquired to do any of the following:
65 * - Lookup a spa_t by name
66 * - Add or remove a spa_t from the namespace
67 * - Increase spa_refcount from non-zero
68 * - Check if spa_refcount is zero
70 * - add/remove/attach/detach devices
71 * - Held for the duration of create/destroy/import/export
73 * It does not need to handle recursion. A create or destroy may
74 * reference objects (files or zvols) in other pools, but by
75 * definition they must have an existing reference, and will never need
76 * to lookup a spa_t by name.
78 * spa_refcount (per-spa refcount_t protected by mutex)
80 * This reference count keep track of any active users of the spa_t. The
81 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
82 * the refcount is never really 'zero' - opening a pool implicitly keeps
83 * some references in the DMU. Internally we check against spa_minref, but
84 * present the image of a zero/non-zero value to consumers.
86 * spa_config_lock[] (per-spa array of rwlocks)
88 * This protects the spa_t from config changes, and must be held in
89 * the following circumstances:
91 * - RW_READER to perform I/O to the spa
92 * - RW_WRITER to change the vdev config
94 * The locking order is fairly straightforward:
96 * spa_namespace_lock -> spa_refcount
98 * The namespace lock must be acquired to increase the refcount from 0
99 * or to check if it is zero.
101 * spa_refcount -> spa_config_lock[]
103 * There must be at least one valid reference on the spa_t to acquire
106 * spa_namespace_lock -> spa_config_lock[]
108 * The namespace lock must always be taken before the config lock.
111 * The spa_namespace_lock can be acquired directly and is globally visible.
113 * The namespace is manipulated using the following functions, all of which
114 * require the spa_namespace_lock to be held.
116 * spa_lookup() Lookup a spa_t by name.
118 * spa_add() Create a new spa_t in the namespace.
120 * spa_remove() Remove a spa_t from the namespace. This also
121 * frees up any memory associated with the spa_t.
123 * spa_next() Returns the next spa_t in the system, or the
124 * first if NULL is passed.
126 * spa_evict_all() Shutdown and remove all spa_t structures in
129 * spa_guid_exists() Determine whether a pool/device guid exists.
131 * The spa_refcount is manipulated using the following functions:
133 * spa_open_ref() Adds a reference to the given spa_t. Must be
134 * called with spa_namespace_lock held if the
135 * refcount is currently zero.
137 * spa_close() Remove a reference from the spa_t. This will
138 * not free the spa_t or remove it from the
139 * namespace. No locking is required.
141 * spa_refcount_zero() Returns true if the refcount is currently
142 * zero. Must be called with spa_namespace_lock
145 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
146 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
147 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
149 * To read the configuration, it suffices to hold one of these locks as reader.
150 * To modify the configuration, you must hold all locks as writer. To modify
151 * vdev state without altering the vdev tree's topology (e.g. online/offline),
152 * you must hold SCL_STATE and SCL_ZIO as writer.
154 * We use these distinct config locks to avoid recursive lock entry.
155 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
156 * block allocations (SCL_ALLOC), which may require reading space maps
157 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
159 * The spa config locks cannot be normal rwlocks because we need the
160 * ability to hand off ownership. For example, SCL_ZIO is acquired
161 * by the issuing thread and later released by an interrupt thread.
162 * They do, however, obey the usual write-wanted semantics to prevent
163 * writer (i.e. system administrator) starvation.
165 * The lock acquisition rules are as follows:
168 * Protects changes to the vdev tree topology, such as vdev
169 * add/remove/attach/detach. Protects the dirty config list
170 * (spa_config_dirty_list) and the set of spares and l2arc devices.
173 * Protects changes to pool state and vdev state, such as vdev
174 * online/offline/fault/degrade/clear. Protects the dirty state list
175 * (spa_state_dirty_list) and global pool state (spa_state).
178 * Protects changes to metaslab groups and classes.
179 * Held as reader by metaslab_alloc() and metaslab_claim().
182 * Held by bp-level zios (those which have no io_vd upon entry)
183 * to prevent changes to the vdev tree. The bp-level zio implicitly
184 * protects all of its vdev child zios, which do not hold SCL_ZIO.
187 * Protects changes to metaslab groups and classes.
188 * Held as reader by metaslab_free(). SCL_FREE is distinct from
189 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
190 * blocks in zio_done() while another i/o that holds either
191 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
194 * Held as reader to prevent changes to the vdev tree during trivial
195 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
196 * other locks, and lower than all of them, to ensure that it's safe
197 * to acquire regardless of caller context.
199 * In addition, the following rules apply:
201 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
202 * The lock ordering is SCL_CONFIG > spa_props_lock.
204 * (b) I/O operations on leaf vdevs. For any zio operation that takes
205 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
206 * or zio_write_phys() -- the caller must ensure that the config cannot
207 * cannot change in the interim, and that the vdev cannot be reopened.
208 * SCL_STATE as reader suffices for both.
210 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
212 * spa_vdev_enter() Acquire the namespace lock and the config lock
215 * spa_vdev_exit() Release the config lock, wait for all I/O
216 * to complete, sync the updated configs to the
217 * cache, and release the namespace lock.
219 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
220 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
221 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
223 * spa_rename() is also implemented within this file since it requires
224 * manipulation of the namespace.
227 static avl_tree_t spa_namespace_avl
;
228 kmutex_t spa_namespace_lock
;
229 static kcondvar_t spa_namespace_cv
;
230 static int spa_active_count
;
231 int spa_max_replication_override
= SPA_DVAS_PER_BP
;
233 static kmutex_t spa_spare_lock
;
234 static avl_tree_t spa_spare_avl
;
235 static kmutex_t spa_l2cache_lock
;
236 static avl_tree_t spa_l2cache_avl
;
238 kmem_cache_t
*spa_buffer_pool
;
242 * Expiration time in milliseconds. This value has two meanings. First it is
243 * used to determine when the spa_deadman() logic should fire. By default the
244 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
245 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
246 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
249 unsigned long zfs_deadman_synctime_ms
= 1000000ULL;
252 * By default the deadman is enabled.
254 int zfs_deadman_enabled
= 1;
257 * The worst case is single-sector max-parity RAID-Z blocks, in which
258 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
259 * times the size; so just assume that. Add to this the fact that
260 * we can have up to 3 DVAs per bp, and one more factor of 2 because
261 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together,
263 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
265 int spa_asize_inflation
= 24;
268 * ==========================================================================
270 * ==========================================================================
273 spa_config_lock_init(spa_t
*spa
)
277 for (i
= 0; i
< SCL_LOCKS
; i
++) {
278 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
279 mutex_init(&scl
->scl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
280 cv_init(&scl
->scl_cv
, NULL
, CV_DEFAULT
, NULL
);
281 refcount_create_untracked(&scl
->scl_count
);
282 scl
->scl_writer
= NULL
;
283 scl
->scl_write_wanted
= 0;
288 spa_config_lock_destroy(spa_t
*spa
)
292 for (i
= 0; i
< SCL_LOCKS
; i
++) {
293 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
294 mutex_destroy(&scl
->scl_lock
);
295 cv_destroy(&scl
->scl_cv
);
296 refcount_destroy(&scl
->scl_count
);
297 ASSERT(scl
->scl_writer
== NULL
);
298 ASSERT(scl
->scl_write_wanted
== 0);
303 spa_config_tryenter(spa_t
*spa
, int locks
, void *tag
, krw_t rw
)
307 for (i
= 0; i
< SCL_LOCKS
; i
++) {
308 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
309 if (!(locks
& (1 << i
)))
311 mutex_enter(&scl
->scl_lock
);
312 if (rw
== RW_READER
) {
313 if (scl
->scl_writer
|| scl
->scl_write_wanted
) {
314 mutex_exit(&scl
->scl_lock
);
315 spa_config_exit(spa
, locks
^ (1 << i
), tag
);
319 ASSERT(scl
->scl_writer
!= curthread
);
320 if (!refcount_is_zero(&scl
->scl_count
)) {
321 mutex_exit(&scl
->scl_lock
);
322 spa_config_exit(spa
, locks
^ (1 << i
), tag
);
325 scl
->scl_writer
= curthread
;
327 (void) refcount_add(&scl
->scl_count
, tag
);
328 mutex_exit(&scl
->scl_lock
);
334 spa_config_enter(spa_t
*spa
, int locks
, void *tag
, krw_t rw
)
339 ASSERT3U(SCL_LOCKS
, <, sizeof (wlocks_held
) * NBBY
);
341 for (i
= 0; i
< SCL_LOCKS
; i
++) {
342 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
343 if (scl
->scl_writer
== curthread
)
344 wlocks_held
|= (1 << i
);
345 if (!(locks
& (1 << i
)))
347 mutex_enter(&scl
->scl_lock
);
348 if (rw
== RW_READER
) {
349 while (scl
->scl_writer
|| scl
->scl_write_wanted
) {
350 cv_wait(&scl
->scl_cv
, &scl
->scl_lock
);
353 ASSERT(scl
->scl_writer
!= curthread
);
354 while (!refcount_is_zero(&scl
->scl_count
)) {
355 scl
->scl_write_wanted
++;
356 cv_wait(&scl
->scl_cv
, &scl
->scl_lock
);
357 scl
->scl_write_wanted
--;
359 scl
->scl_writer
= curthread
;
361 (void) refcount_add(&scl
->scl_count
, tag
);
362 mutex_exit(&scl
->scl_lock
);
364 ASSERT(wlocks_held
<= locks
);
368 spa_config_exit(spa_t
*spa
, int locks
, void *tag
)
372 for (i
= SCL_LOCKS
- 1; i
>= 0; i
--) {
373 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
374 if (!(locks
& (1 << i
)))
376 mutex_enter(&scl
->scl_lock
);
377 ASSERT(!refcount_is_zero(&scl
->scl_count
));
378 if (refcount_remove(&scl
->scl_count
, tag
) == 0) {
379 ASSERT(scl
->scl_writer
== NULL
||
380 scl
->scl_writer
== curthread
);
381 scl
->scl_writer
= NULL
; /* OK in either case */
382 cv_broadcast(&scl
->scl_cv
);
384 mutex_exit(&scl
->scl_lock
);
389 spa_config_held(spa_t
*spa
, int locks
, krw_t rw
)
391 int i
, locks_held
= 0;
393 for (i
= 0; i
< SCL_LOCKS
; i
++) {
394 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
395 if (!(locks
& (1 << i
)))
397 if ((rw
== RW_READER
&& !refcount_is_zero(&scl
->scl_count
)) ||
398 (rw
== RW_WRITER
&& scl
->scl_writer
== curthread
))
399 locks_held
|= 1 << i
;
406 * ==========================================================================
407 * SPA namespace functions
408 * ==========================================================================
412 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
413 * Returns NULL if no matching spa_t is found.
416 spa_lookup(const char *name
)
418 static spa_t search
; /* spa_t is large; don't allocate on stack */
423 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
425 (void) strlcpy(search
.spa_name
, name
, sizeof (search
.spa_name
));
428 * If it's a full dataset name, figure out the pool name and
431 cp
= strpbrk(search
.spa_name
, "/@");
435 spa
= avl_find(&spa_namespace_avl
, &search
, &where
);
441 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
442 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
443 * looking for potentially hung I/Os.
446 spa_deadman(void *arg
)
450 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
451 (gethrtime() - spa
->spa_sync_starttime
) / NANOSEC
,
452 ++spa
->spa_deadman_calls
);
453 if (zfs_deadman_enabled
)
454 vdev_deadman(spa
->spa_root_vdev
);
456 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_taskq
,
457 spa_deadman
, spa
, TQ_PUSHPAGE
, ddi_get_lbolt() +
458 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
462 * Create an uninitialized spa_t with the given name. Requires
463 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
464 * exist by calling spa_lookup() first.
467 spa_add(const char *name
, nvlist_t
*config
, const char *altroot
)
470 spa_config_dirent_t
*dp
;
473 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
475 spa
= kmem_zalloc(sizeof (spa_t
), KM_PUSHPAGE
| KM_NODEBUG
);
477 mutex_init(&spa
->spa_async_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
478 mutex_init(&spa
->spa_errlist_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
479 mutex_init(&spa
->spa_errlog_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
480 mutex_init(&spa
->spa_history_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
481 mutex_init(&spa
->spa_proc_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
482 mutex_init(&spa
->spa_props_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
483 mutex_init(&spa
->spa_scrub_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
484 mutex_init(&spa
->spa_suspend_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
485 mutex_init(&spa
->spa_vdev_top_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
487 cv_init(&spa
->spa_async_cv
, NULL
, CV_DEFAULT
, NULL
);
488 cv_init(&spa
->spa_proc_cv
, NULL
, CV_DEFAULT
, NULL
);
489 cv_init(&spa
->spa_scrub_io_cv
, NULL
, CV_DEFAULT
, NULL
);
490 cv_init(&spa
->spa_suspend_cv
, NULL
, CV_DEFAULT
, NULL
);
492 for (t
= 0; t
< TXG_SIZE
; t
++)
493 bplist_create(&spa
->spa_free_bplist
[t
]);
495 (void) strlcpy(spa
->spa_name
, name
, sizeof (spa
->spa_name
));
496 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
497 spa
->spa_freeze_txg
= UINT64_MAX
;
498 spa
->spa_final_txg
= UINT64_MAX
;
499 spa
->spa_load_max_txg
= UINT64_MAX
;
501 spa
->spa_proc_state
= SPA_PROC_NONE
;
503 spa
->spa_deadman_synctime
= MSEC2NSEC(zfs_deadman_synctime_ms
);
505 refcount_create(&spa
->spa_refcount
);
506 spa_config_lock_init(spa
);
509 avl_add(&spa_namespace_avl
, spa
);
512 * Set the alternate root, if there is one.
515 spa
->spa_root
= spa_strdup(altroot
);
520 * Every pool starts with the default cachefile
522 list_create(&spa
->spa_config_list
, sizeof (spa_config_dirent_t
),
523 offsetof(spa_config_dirent_t
, scd_link
));
525 dp
= kmem_zalloc(sizeof (spa_config_dirent_t
), KM_PUSHPAGE
);
526 dp
->scd_path
= altroot
? NULL
: spa_strdup(spa_config_path
);
527 list_insert_head(&spa
->spa_config_list
, dp
);
529 VERIFY(nvlist_alloc(&spa
->spa_load_info
, NV_UNIQUE_NAME
,
532 if (config
!= NULL
) {
535 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_FEATURES_FOR_READ
,
537 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
,
541 VERIFY(nvlist_dup(config
, &spa
->spa_config
, 0) == 0);
544 if (spa
->spa_label_features
== NULL
) {
545 VERIFY(nvlist_alloc(&spa
->spa_label_features
, NV_UNIQUE_NAME
,
549 spa
->spa_debug
= ((zfs_flags
& ZFS_DEBUG_SPA
) != 0);
555 * Removes a spa_t from the namespace, freeing up any memory used. Requires
556 * spa_namespace_lock. This is called only after the spa_t has been closed and
560 spa_remove(spa_t
*spa
)
562 spa_config_dirent_t
*dp
;
565 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
566 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
568 nvlist_free(spa
->spa_config_splitting
);
570 avl_remove(&spa_namespace_avl
, spa
);
571 cv_broadcast(&spa_namespace_cv
);
574 spa_strfree(spa
->spa_root
);
578 while ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
579 list_remove(&spa
->spa_config_list
, dp
);
580 if (dp
->scd_path
!= NULL
)
581 spa_strfree(dp
->scd_path
);
582 kmem_free(dp
, sizeof (spa_config_dirent_t
));
585 list_destroy(&spa
->spa_config_list
);
587 nvlist_free(spa
->spa_label_features
);
588 nvlist_free(spa
->spa_load_info
);
589 spa_config_set(spa
, NULL
);
591 refcount_destroy(&spa
->spa_refcount
);
593 spa_stats_destroy(spa
);
594 spa_config_lock_destroy(spa
);
596 for (t
= 0; t
< TXG_SIZE
; t
++)
597 bplist_destroy(&spa
->spa_free_bplist
[t
]);
599 cv_destroy(&spa
->spa_async_cv
);
600 cv_destroy(&spa
->spa_proc_cv
);
601 cv_destroy(&spa
->spa_scrub_io_cv
);
602 cv_destroy(&spa
->spa_suspend_cv
);
604 mutex_destroy(&spa
->spa_async_lock
);
605 mutex_destroy(&spa
->spa_errlist_lock
);
606 mutex_destroy(&spa
->spa_errlog_lock
);
607 mutex_destroy(&spa
->spa_history_lock
);
608 mutex_destroy(&spa
->spa_proc_lock
);
609 mutex_destroy(&spa
->spa_props_lock
);
610 mutex_destroy(&spa
->spa_scrub_lock
);
611 mutex_destroy(&spa
->spa_suspend_lock
);
612 mutex_destroy(&spa
->spa_vdev_top_lock
);
614 kmem_free(spa
, sizeof (spa_t
));
618 * Given a pool, return the next pool in the namespace, or NULL if there is
619 * none. If 'prev' is NULL, return the first pool.
622 spa_next(spa_t
*prev
)
624 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
627 return (AVL_NEXT(&spa_namespace_avl
, prev
));
629 return (avl_first(&spa_namespace_avl
));
633 * ==========================================================================
634 * SPA refcount functions
635 * ==========================================================================
639 * Add a reference to the given spa_t. Must have at least one reference, or
640 * have the namespace lock held.
643 spa_open_ref(spa_t
*spa
, void *tag
)
645 ASSERT(refcount_count(&spa
->spa_refcount
) >= spa
->spa_minref
||
646 MUTEX_HELD(&spa_namespace_lock
));
647 (void) refcount_add(&spa
->spa_refcount
, tag
);
651 * Remove a reference to the given spa_t. Must have at least one reference, or
652 * have the namespace lock held.
655 spa_close(spa_t
*spa
, void *tag
)
657 ASSERT(refcount_count(&spa
->spa_refcount
) > spa
->spa_minref
||
658 MUTEX_HELD(&spa_namespace_lock
));
659 (void) refcount_remove(&spa
->spa_refcount
, tag
);
663 * Check to see if the spa refcount is zero. Must be called with
664 * spa_namespace_lock held. We really compare against spa_minref, which is the
665 * number of references acquired when opening a pool
668 spa_refcount_zero(spa_t
*spa
)
670 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
672 return (refcount_count(&spa
->spa_refcount
) == spa
->spa_minref
);
676 * ==========================================================================
677 * SPA spare and l2cache tracking
678 * ==========================================================================
682 * Hot spares and cache devices are tracked using the same code below,
683 * for 'auxiliary' devices.
686 typedef struct spa_aux
{
694 spa_aux_compare(const void *a
, const void *b
)
696 const spa_aux_t
*sa
= a
;
697 const spa_aux_t
*sb
= b
;
699 if (sa
->aux_guid
< sb
->aux_guid
)
701 else if (sa
->aux_guid
> sb
->aux_guid
)
708 spa_aux_add(vdev_t
*vd
, avl_tree_t
*avl
)
714 search
.aux_guid
= vd
->vdev_guid
;
715 if ((aux
= avl_find(avl
, &search
, &where
)) != NULL
) {
718 aux
= kmem_zalloc(sizeof (spa_aux_t
), KM_PUSHPAGE
);
719 aux
->aux_guid
= vd
->vdev_guid
;
721 avl_insert(avl
, aux
, where
);
726 spa_aux_remove(vdev_t
*vd
, avl_tree_t
*avl
)
732 search
.aux_guid
= vd
->vdev_guid
;
733 aux
= avl_find(avl
, &search
, &where
);
737 if (--aux
->aux_count
== 0) {
738 avl_remove(avl
, aux
);
739 kmem_free(aux
, sizeof (spa_aux_t
));
740 } else if (aux
->aux_pool
== spa_guid(vd
->vdev_spa
)) {
741 aux
->aux_pool
= 0ULL;
746 spa_aux_exists(uint64_t guid
, uint64_t *pool
, int *refcnt
, avl_tree_t
*avl
)
748 spa_aux_t search
, *found
;
750 search
.aux_guid
= guid
;
751 found
= avl_find(avl
, &search
, NULL
);
755 *pool
= found
->aux_pool
;
762 *refcnt
= found
->aux_count
;
767 return (found
!= NULL
);
771 spa_aux_activate(vdev_t
*vd
, avl_tree_t
*avl
)
773 spa_aux_t search
, *found
;
776 search
.aux_guid
= vd
->vdev_guid
;
777 found
= avl_find(avl
, &search
, &where
);
778 ASSERT(found
!= NULL
);
779 ASSERT(found
->aux_pool
== 0ULL);
781 found
->aux_pool
= spa_guid(vd
->vdev_spa
);
785 * Spares are tracked globally due to the following constraints:
787 * - A spare may be part of multiple pools.
788 * - A spare may be added to a pool even if it's actively in use within
790 * - A spare in use in any pool can only be the source of a replacement if
791 * the target is a spare in the same pool.
793 * We keep track of all spares on the system through the use of a reference
794 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
795 * spare, then we bump the reference count in the AVL tree. In addition, we set
796 * the 'vdev_isspare' member to indicate that the device is a spare (active or
797 * inactive). When a spare is made active (used to replace a device in the
798 * pool), we also keep track of which pool its been made a part of.
800 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
801 * called under the spa_namespace lock as part of vdev reconfiguration. The
802 * separate spare lock exists for the status query path, which does not need to
803 * be completely consistent with respect to other vdev configuration changes.
807 spa_spare_compare(const void *a
, const void *b
)
809 return (spa_aux_compare(a
, b
));
813 spa_spare_add(vdev_t
*vd
)
815 mutex_enter(&spa_spare_lock
);
816 ASSERT(!vd
->vdev_isspare
);
817 spa_aux_add(vd
, &spa_spare_avl
);
818 vd
->vdev_isspare
= B_TRUE
;
819 mutex_exit(&spa_spare_lock
);
823 spa_spare_remove(vdev_t
*vd
)
825 mutex_enter(&spa_spare_lock
);
826 ASSERT(vd
->vdev_isspare
);
827 spa_aux_remove(vd
, &spa_spare_avl
);
828 vd
->vdev_isspare
= B_FALSE
;
829 mutex_exit(&spa_spare_lock
);
833 spa_spare_exists(uint64_t guid
, uint64_t *pool
, int *refcnt
)
837 mutex_enter(&spa_spare_lock
);
838 found
= spa_aux_exists(guid
, pool
, refcnt
, &spa_spare_avl
);
839 mutex_exit(&spa_spare_lock
);
845 spa_spare_activate(vdev_t
*vd
)
847 mutex_enter(&spa_spare_lock
);
848 ASSERT(vd
->vdev_isspare
);
849 spa_aux_activate(vd
, &spa_spare_avl
);
850 mutex_exit(&spa_spare_lock
);
854 * Level 2 ARC devices are tracked globally for the same reasons as spares.
855 * Cache devices currently only support one pool per cache device, and so
856 * for these devices the aux reference count is currently unused beyond 1.
860 spa_l2cache_compare(const void *a
, const void *b
)
862 return (spa_aux_compare(a
, b
));
866 spa_l2cache_add(vdev_t
*vd
)
868 mutex_enter(&spa_l2cache_lock
);
869 ASSERT(!vd
->vdev_isl2cache
);
870 spa_aux_add(vd
, &spa_l2cache_avl
);
871 vd
->vdev_isl2cache
= B_TRUE
;
872 mutex_exit(&spa_l2cache_lock
);
876 spa_l2cache_remove(vdev_t
*vd
)
878 mutex_enter(&spa_l2cache_lock
);
879 ASSERT(vd
->vdev_isl2cache
);
880 spa_aux_remove(vd
, &spa_l2cache_avl
);
881 vd
->vdev_isl2cache
= B_FALSE
;
882 mutex_exit(&spa_l2cache_lock
);
886 spa_l2cache_exists(uint64_t guid
, uint64_t *pool
)
890 mutex_enter(&spa_l2cache_lock
);
891 found
= spa_aux_exists(guid
, pool
, NULL
, &spa_l2cache_avl
);
892 mutex_exit(&spa_l2cache_lock
);
898 spa_l2cache_activate(vdev_t
*vd
)
900 mutex_enter(&spa_l2cache_lock
);
901 ASSERT(vd
->vdev_isl2cache
);
902 spa_aux_activate(vd
, &spa_l2cache_avl
);
903 mutex_exit(&spa_l2cache_lock
);
907 * ==========================================================================
909 * ==========================================================================
913 * Lock the given spa_t for the purpose of adding or removing a vdev.
914 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
915 * It returns the next transaction group for the spa_t.
918 spa_vdev_enter(spa_t
*spa
)
920 mutex_enter(&spa
->spa_vdev_top_lock
);
921 mutex_enter(&spa_namespace_lock
);
922 return (spa_vdev_config_enter(spa
));
926 * Internal implementation for spa_vdev_enter(). Used when a vdev
927 * operation requires multiple syncs (i.e. removing a device) while
928 * keeping the spa_namespace_lock held.
931 spa_vdev_config_enter(spa_t
*spa
)
933 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
935 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
937 return (spa_last_synced_txg(spa
) + 1);
941 * Used in combination with spa_vdev_config_enter() to allow the syncing
942 * of multiple transactions without releasing the spa_namespace_lock.
945 spa_vdev_config_exit(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int error
, char *tag
)
947 int config_changed
= B_FALSE
;
949 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
950 ASSERT(txg
> spa_last_synced_txg(spa
));
952 spa
->spa_pending_vdev
= NULL
;
957 vdev_dtl_reassess(spa
->spa_root_vdev
, 0, 0, B_FALSE
);
959 if (error
== 0 && !list_is_empty(&spa
->spa_config_dirty_list
)) {
960 config_changed
= B_TRUE
;
961 spa
->spa_config_generation
++;
965 * Verify the metaslab classes.
967 ASSERT(metaslab_class_validate(spa_normal_class(spa
)) == 0);
968 ASSERT(metaslab_class_validate(spa_log_class(spa
)) == 0);
970 spa_config_exit(spa
, SCL_ALL
, spa
);
973 * Panic the system if the specified tag requires it. This
974 * is useful for ensuring that configurations are updated
977 if (zio_injection_enabled
)
978 zio_handle_panic_injection(spa
, tag
, 0);
981 * Note: this txg_wait_synced() is important because it ensures
982 * that there won't be more than one config change per txg.
983 * This allows us to use the txg as the generation number.
986 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
989 ASSERT(!vd
->vdev_detached
|| vd
->vdev_dtl_sm
== NULL
);
990 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
992 spa_config_exit(spa
, SCL_ALL
, spa
);
996 * If the config changed, update the config cache.
999 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
1003 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1004 * locking of spa_vdev_enter(), we also want make sure the transactions have
1005 * synced to disk, and then update the global configuration cache with the new
1009 spa_vdev_exit(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int error
)
1011 spa_vdev_config_exit(spa
, vd
, txg
, error
, FTAG
);
1012 mutex_exit(&spa_namespace_lock
);
1013 mutex_exit(&spa
->spa_vdev_top_lock
);
1019 * Lock the given spa_t for the purpose of changing vdev state.
1022 spa_vdev_state_enter(spa_t
*spa
, int oplocks
)
1024 int locks
= SCL_STATE_ALL
| oplocks
;
1027 * Root pools may need to read of the underlying devfs filesystem
1028 * when opening up a vdev. Unfortunately if we're holding the
1029 * SCL_ZIO lock it will result in a deadlock when we try to issue
1030 * the read from the root filesystem. Instead we "prefetch"
1031 * the associated vnodes that we need prior to opening the
1032 * underlying devices and cache them so that we can prevent
1033 * any I/O when we are doing the actual open.
1035 if (spa_is_root(spa
)) {
1036 int low
= locks
& ~(SCL_ZIO
- 1);
1037 int high
= locks
& ~low
;
1039 spa_config_enter(spa
, high
, spa
, RW_WRITER
);
1040 vdev_hold(spa
->spa_root_vdev
);
1041 spa_config_enter(spa
, low
, spa
, RW_WRITER
);
1043 spa_config_enter(spa
, locks
, spa
, RW_WRITER
);
1045 spa
->spa_vdev_locks
= locks
;
1049 spa_vdev_state_exit(spa_t
*spa
, vdev_t
*vd
, int error
)
1051 boolean_t config_changed
= B_FALSE
;
1053 if (vd
!= NULL
|| error
== 0)
1054 vdev_dtl_reassess(vd
? vd
->vdev_top
: spa
->spa_root_vdev
,
1058 vdev_state_dirty(vd
->vdev_top
);
1059 config_changed
= B_TRUE
;
1060 spa
->spa_config_generation
++;
1063 if (spa_is_root(spa
))
1064 vdev_rele(spa
->spa_root_vdev
);
1066 ASSERT3U(spa
->spa_vdev_locks
, >=, SCL_STATE_ALL
);
1067 spa_config_exit(spa
, spa
->spa_vdev_locks
, spa
);
1070 * If anything changed, wait for it to sync. This ensures that,
1071 * from the system administrator's perspective, zpool(1M) commands
1072 * are synchronous. This is important for things like zpool offline:
1073 * when the command completes, you expect no further I/O from ZFS.
1076 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1079 * If the config changed, update the config cache.
1081 if (config_changed
) {
1082 mutex_enter(&spa_namespace_lock
);
1083 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
1084 mutex_exit(&spa_namespace_lock
);
1091 * ==========================================================================
1092 * Miscellaneous functions
1093 * ==========================================================================
1097 spa_activate_mos_feature(spa_t
*spa
, const char *feature
)
1099 (void) nvlist_add_boolean(spa
->spa_label_features
, feature
);
1100 vdev_config_dirty(spa
->spa_root_vdev
);
1104 spa_deactivate_mos_feature(spa_t
*spa
, const char *feature
)
1106 (void) nvlist_remove_all(spa
->spa_label_features
, feature
);
1107 vdev_config_dirty(spa
->spa_root_vdev
);
1114 spa_rename(const char *name
, const char *newname
)
1120 * Lookup the spa_t and grab the config lock for writing. We need to
1121 * actually open the pool so that we can sync out the necessary labels.
1122 * It's OK to call spa_open() with the namespace lock held because we
1123 * allow recursive calls for other reasons.
1125 mutex_enter(&spa_namespace_lock
);
1126 if ((err
= spa_open(name
, &spa
, FTAG
)) != 0) {
1127 mutex_exit(&spa_namespace_lock
);
1131 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1133 avl_remove(&spa_namespace_avl
, spa
);
1134 (void) strlcpy(spa
->spa_name
, newname
, sizeof (spa
->spa_name
));
1135 avl_add(&spa_namespace_avl
, spa
);
1138 * Sync all labels to disk with the new names by marking the root vdev
1139 * dirty and waiting for it to sync. It will pick up the new pool name
1142 vdev_config_dirty(spa
->spa_root_vdev
);
1144 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1146 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1149 * Sync the updated config cache.
1151 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
1153 spa_close(spa
, FTAG
);
1155 mutex_exit(&spa_namespace_lock
);
1161 * Return the spa_t associated with given pool_guid, if it exists. If
1162 * device_guid is non-zero, determine whether the pool exists *and* contains
1163 * a device with the specified device_guid.
1166 spa_by_guid(uint64_t pool_guid
, uint64_t device_guid
)
1169 avl_tree_t
*t
= &spa_namespace_avl
;
1171 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1173 for (spa
= avl_first(t
); spa
!= NULL
; spa
= AVL_NEXT(t
, spa
)) {
1174 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1176 if (spa
->spa_root_vdev
== NULL
)
1178 if (spa_guid(spa
) == pool_guid
) {
1179 if (device_guid
== 0)
1182 if (vdev_lookup_by_guid(spa
->spa_root_vdev
,
1183 device_guid
) != NULL
)
1187 * Check any devices we may be in the process of adding.
1189 if (spa
->spa_pending_vdev
) {
1190 if (vdev_lookup_by_guid(spa
->spa_pending_vdev
,
1191 device_guid
) != NULL
)
1201 * Determine whether a pool with the given pool_guid exists.
1204 spa_guid_exists(uint64_t pool_guid
, uint64_t device_guid
)
1206 return (spa_by_guid(pool_guid
, device_guid
) != NULL
);
1210 spa_strdup(const char *s
)
1216 new = kmem_alloc(len
+ 1, KM_PUSHPAGE
);
1224 spa_strfree(char *s
)
1226 kmem_free(s
, strlen(s
) + 1);
1230 spa_get_random(uint64_t range
)
1236 (void) random_get_pseudo_bytes((void *)&r
, sizeof (uint64_t));
1242 spa_generate_guid(spa_t
*spa
)
1244 uint64_t guid
= spa_get_random(-1ULL);
1247 while (guid
== 0 || spa_guid_exists(spa_guid(spa
), guid
))
1248 guid
= spa_get_random(-1ULL);
1250 while (guid
== 0 || spa_guid_exists(guid
, 0))
1251 guid
= spa_get_random(-1ULL);
1258 sprintf_blkptr(char *buf
, const blkptr_t
*bp
)
1261 char *checksum
= NULL
;
1262 char *compress
= NULL
;
1265 if (BP_GET_TYPE(bp
) & DMU_OT_NEWTYPE
) {
1266 dmu_object_byteswap_t bswap
=
1267 DMU_OT_BYTESWAP(BP_GET_TYPE(bp
));
1268 (void) snprintf(type
, sizeof (type
), "bswap %s %s",
1269 DMU_OT_IS_METADATA(BP_GET_TYPE(bp
)) ?
1270 "metadata" : "data",
1271 dmu_ot_byteswap
[bswap
].ob_name
);
1273 (void) strlcpy(type
, dmu_ot
[BP_GET_TYPE(bp
)].ot_name
,
1276 checksum
= zio_checksum_table
[BP_GET_CHECKSUM(bp
)].ci_name
;
1277 compress
= zio_compress_table
[BP_GET_COMPRESS(bp
)].ci_name
;
1280 SPRINTF_BLKPTR(snprintf
, ' ', buf
, bp
, type
, checksum
, compress
);
1284 spa_freeze(spa_t
*spa
)
1286 uint64_t freeze_txg
= 0;
1288 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1289 if (spa
->spa_freeze_txg
== UINT64_MAX
) {
1290 freeze_txg
= spa_last_synced_txg(spa
) + TXG_SIZE
;
1291 spa
->spa_freeze_txg
= freeze_txg
;
1293 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1294 if (freeze_txg
!= 0)
1295 txg_wait_synced(spa_get_dsl(spa
), freeze_txg
);
1299 * This is a stripped-down version of strtoull, suitable only for converting
1300 * lowercase hexadecimal numbers that don't overflow.
1303 strtonum(const char *str
, char **nptr
)
1309 while ((c
= *str
) != '\0') {
1310 if (c
>= '0' && c
<= '9')
1312 else if (c
>= 'a' && c
<= 'f')
1313 digit
= 10 + c
- 'a';
1324 *nptr
= (char *)str
;
1330 * ==========================================================================
1331 * Accessor functions
1332 * ==========================================================================
1336 spa_shutting_down(spa_t
*spa
)
1338 return (spa
->spa_async_suspended
);
1342 spa_get_dsl(spa_t
*spa
)
1344 return (spa
->spa_dsl_pool
);
1348 spa_is_initializing(spa_t
*spa
)
1350 return (spa
->spa_is_initializing
);
1354 spa_get_rootblkptr(spa_t
*spa
)
1356 return (&spa
->spa_ubsync
.ub_rootbp
);
1360 spa_set_rootblkptr(spa_t
*spa
, const blkptr_t
*bp
)
1362 spa
->spa_uberblock
.ub_rootbp
= *bp
;
1366 spa_altroot(spa_t
*spa
, char *buf
, size_t buflen
)
1368 if (spa
->spa_root
== NULL
)
1371 (void) strncpy(buf
, spa
->spa_root
, buflen
);
1375 spa_sync_pass(spa_t
*spa
)
1377 return (spa
->spa_sync_pass
);
1381 spa_name(spa_t
*spa
)
1383 return (spa
->spa_name
);
1387 spa_guid(spa_t
*spa
)
1389 dsl_pool_t
*dp
= spa_get_dsl(spa
);
1393 * If we fail to parse the config during spa_load(), we can go through
1394 * the error path (which posts an ereport) and end up here with no root
1395 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1398 if (spa
->spa_root_vdev
== NULL
)
1399 return (spa
->spa_config_guid
);
1401 guid
= spa
->spa_last_synced_guid
!= 0 ?
1402 spa
->spa_last_synced_guid
: spa
->spa_root_vdev
->vdev_guid
;
1405 * Return the most recently synced out guid unless we're
1406 * in syncing context.
1408 if (dp
&& dsl_pool_sync_context(dp
))
1409 return (spa
->spa_root_vdev
->vdev_guid
);
1415 spa_load_guid(spa_t
*spa
)
1418 * This is a GUID that exists solely as a reference for the
1419 * purposes of the arc. It is generated at load time, and
1420 * is never written to persistent storage.
1422 return (spa
->spa_load_guid
);
1426 spa_last_synced_txg(spa_t
*spa
)
1428 return (spa
->spa_ubsync
.ub_txg
);
1432 spa_first_txg(spa_t
*spa
)
1434 return (spa
->spa_first_txg
);
1438 spa_syncing_txg(spa_t
*spa
)
1440 return (spa
->spa_syncing_txg
);
1444 spa_state(spa_t
*spa
)
1446 return (spa
->spa_state
);
1450 spa_load_state(spa_t
*spa
)
1452 return (spa
->spa_load_state
);
1456 spa_freeze_txg(spa_t
*spa
)
1458 return (spa
->spa_freeze_txg
);
1463 spa_get_asize(spa_t
*spa
, uint64_t lsize
)
1465 return (lsize
* spa_asize_inflation
);
1469 spa_get_dspace(spa_t
*spa
)
1471 return (spa
->spa_dspace
);
1475 spa_update_dspace(spa_t
*spa
)
1477 spa
->spa_dspace
= metaslab_class_get_dspace(spa_normal_class(spa
)) +
1478 ddt_get_dedup_dspace(spa
);
1482 * Return the failure mode that has been set to this pool. The default
1483 * behavior will be to block all I/Os when a complete failure occurs.
1486 spa_get_failmode(spa_t
*spa
)
1488 return (spa
->spa_failmode
);
1492 spa_suspended(spa_t
*spa
)
1494 return (spa
->spa_suspended
);
1498 spa_version(spa_t
*spa
)
1500 return (spa
->spa_ubsync
.ub_version
);
1504 spa_deflate(spa_t
*spa
)
1506 return (spa
->spa_deflate
);
1510 spa_normal_class(spa_t
*spa
)
1512 return (spa
->spa_normal_class
);
1516 spa_log_class(spa_t
*spa
)
1518 return (spa
->spa_log_class
);
1522 spa_max_replication(spa_t
*spa
)
1525 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1526 * handle BPs with more than one DVA allocated. Set our max
1527 * replication level accordingly.
1529 if (spa_version(spa
) < SPA_VERSION_DITTO_BLOCKS
)
1531 return (MIN(SPA_DVAS_PER_BP
, spa_max_replication_override
));
1535 spa_prev_software_version(spa_t
*spa
)
1537 return (spa
->spa_prev_software_version
);
1541 spa_deadman_synctime(spa_t
*spa
)
1543 return (spa
->spa_deadman_synctime
);
1547 dva_get_dsize_sync(spa_t
*spa
, const dva_t
*dva
)
1549 uint64_t asize
= DVA_GET_ASIZE(dva
);
1550 uint64_t dsize
= asize
;
1552 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1554 if (asize
!= 0 && spa
->spa_deflate
) {
1555 vdev_t
*vd
= vdev_lookup_top(spa
, DVA_GET_VDEV(dva
));
1557 dsize
= (asize
>> SPA_MINBLOCKSHIFT
) *
1558 vd
->vdev_deflate_ratio
;
1565 bp_get_dsize_sync(spa_t
*spa
, const blkptr_t
*bp
)
1570 for (d
= 0; d
< SPA_DVAS_PER_BP
; d
++)
1571 dsize
+= dva_get_dsize_sync(spa
, &bp
->blk_dva
[d
]);
1577 bp_get_dsize(spa_t
*spa
, const blkptr_t
*bp
)
1582 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
1584 for (d
= 0; d
< SPA_DVAS_PER_BP
; d
++)
1585 dsize
+= dva_get_dsize_sync(spa
, &bp
->blk_dva
[d
]);
1587 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
1593 * ==========================================================================
1594 * Initialization and Termination
1595 * ==========================================================================
1599 spa_name_compare(const void *a1
, const void *a2
)
1601 const spa_t
*s1
= a1
;
1602 const spa_t
*s2
= a2
;
1605 s
= strcmp(s1
->spa_name
, s2
->spa_name
);
1622 mutex_init(&spa_namespace_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1623 mutex_init(&spa_spare_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1624 mutex_init(&spa_l2cache_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1625 cv_init(&spa_namespace_cv
, NULL
, CV_DEFAULT
, NULL
);
1627 avl_create(&spa_namespace_avl
, spa_name_compare
, sizeof (spa_t
),
1628 offsetof(spa_t
, spa_avl
));
1630 avl_create(&spa_spare_avl
, spa_spare_compare
, sizeof (spa_aux_t
),
1631 offsetof(spa_aux_t
, aux_avl
));
1633 avl_create(&spa_l2cache_avl
, spa_l2cache_compare
, sizeof (spa_aux_t
),
1634 offsetof(spa_aux_t
, aux_avl
));
1636 spa_mode_global
= mode
;
1639 if (spa_mode_global
!= FREAD
&& dprintf_find_string("watch")) {
1640 struct sigaction sa
;
1642 sa
.sa_flags
= SA_SIGINFO
;
1643 sigemptyset(&sa
.sa_mask
);
1644 sa
.sa_sigaction
= arc_buf_sigsegv
;
1646 if (sigaction(SIGSEGV
, &sa
, NULL
) == -1) {
1647 perror("could not enable watchpoints: "
1648 "sigaction(SIGSEGV, ...) = ");
1663 vdev_cache_stat_init();
1667 zpool_feature_init();
1680 vdev_cache_stat_fini();
1690 avl_destroy(&spa_namespace_avl
);
1691 avl_destroy(&spa_spare_avl
);
1692 avl_destroy(&spa_l2cache_avl
);
1694 cv_destroy(&spa_namespace_cv
);
1695 mutex_destroy(&spa_namespace_lock
);
1696 mutex_destroy(&spa_spare_lock
);
1697 mutex_destroy(&spa_l2cache_lock
);
1701 * Return whether this pool has slogs. No locking needed.
1702 * It's not a problem if the wrong answer is returned as it's only for
1703 * performance and not correctness
1706 spa_has_slogs(spa_t
*spa
)
1708 return (spa
->spa_log_class
->mc_rotor
!= NULL
);
1712 spa_get_log_state(spa_t
*spa
)
1714 return (spa
->spa_log_state
);
1718 spa_set_log_state(spa_t
*spa
, spa_log_state_t state
)
1720 spa
->spa_log_state
= state
;
1724 spa_is_root(spa_t
*spa
)
1726 return (spa
->spa_is_root
);
1730 spa_writeable(spa_t
*spa
)
1732 return (!!(spa
->spa_mode
& FWRITE
));
1736 spa_mode(spa_t
*spa
)
1738 return (spa
->spa_mode
);
1742 spa_bootfs(spa_t
*spa
)
1744 return (spa
->spa_bootfs
);
1748 spa_delegation(spa_t
*spa
)
1750 return (spa
->spa_delegation
);
1754 spa_meta_objset(spa_t
*spa
)
1756 return (spa
->spa_meta_objset
);
1760 spa_dedup_checksum(spa_t
*spa
)
1762 return (spa
->spa_dedup_checksum
);
1766 * Reset pool scan stat per scan pass (or reboot).
1769 spa_scan_stat_init(spa_t
*spa
)
1771 /* data not stored on disk */
1772 spa
->spa_scan_pass_start
= gethrestime_sec();
1773 spa
->spa_scan_pass_exam
= 0;
1774 vdev_scan_stat_init(spa
->spa_root_vdev
);
1778 * Get scan stats for zpool status reports
1781 spa_scan_get_stats(spa_t
*spa
, pool_scan_stat_t
*ps
)
1783 dsl_scan_t
*scn
= spa
->spa_dsl_pool
? spa
->spa_dsl_pool
->dp_scan
: NULL
;
1785 if (scn
== NULL
|| scn
->scn_phys
.scn_func
== POOL_SCAN_NONE
)
1786 return (SET_ERROR(ENOENT
));
1787 bzero(ps
, sizeof (pool_scan_stat_t
));
1789 /* data stored on disk */
1790 ps
->pss_func
= scn
->scn_phys
.scn_func
;
1791 ps
->pss_start_time
= scn
->scn_phys
.scn_start_time
;
1792 ps
->pss_end_time
= scn
->scn_phys
.scn_end_time
;
1793 ps
->pss_to_examine
= scn
->scn_phys
.scn_to_examine
;
1794 ps
->pss_examined
= scn
->scn_phys
.scn_examined
;
1795 ps
->pss_to_process
= scn
->scn_phys
.scn_to_process
;
1796 ps
->pss_processed
= scn
->scn_phys
.scn_processed
;
1797 ps
->pss_errors
= scn
->scn_phys
.scn_errors
;
1798 ps
->pss_state
= scn
->scn_phys
.scn_state
;
1800 /* data not stored on disk */
1801 ps
->pss_pass_start
= spa
->spa_scan_pass_start
;
1802 ps
->pss_pass_exam
= spa
->spa_scan_pass_exam
;
1808 spa_debug_enabled(spa_t
*spa
)
1810 return (spa
->spa_debug
);
1813 #if defined(_KERNEL) && defined(HAVE_SPL)
1814 /* Namespace manipulation */
1815 EXPORT_SYMBOL(spa_lookup
);
1816 EXPORT_SYMBOL(spa_add
);
1817 EXPORT_SYMBOL(spa_remove
);
1818 EXPORT_SYMBOL(spa_next
);
1820 /* Refcount functions */
1821 EXPORT_SYMBOL(spa_open_ref
);
1822 EXPORT_SYMBOL(spa_close
);
1823 EXPORT_SYMBOL(spa_refcount_zero
);
1825 /* Pool configuration lock */
1826 EXPORT_SYMBOL(spa_config_tryenter
);
1827 EXPORT_SYMBOL(spa_config_enter
);
1828 EXPORT_SYMBOL(spa_config_exit
);
1829 EXPORT_SYMBOL(spa_config_held
);
1831 /* Pool vdev add/remove lock */
1832 EXPORT_SYMBOL(spa_vdev_enter
);
1833 EXPORT_SYMBOL(spa_vdev_exit
);
1835 /* Pool vdev state change lock */
1836 EXPORT_SYMBOL(spa_vdev_state_enter
);
1837 EXPORT_SYMBOL(spa_vdev_state_exit
);
1839 /* Accessor functions */
1840 EXPORT_SYMBOL(spa_shutting_down
);
1841 EXPORT_SYMBOL(spa_get_dsl
);
1842 EXPORT_SYMBOL(spa_get_rootblkptr
);
1843 EXPORT_SYMBOL(spa_set_rootblkptr
);
1844 EXPORT_SYMBOL(spa_altroot
);
1845 EXPORT_SYMBOL(spa_sync_pass
);
1846 EXPORT_SYMBOL(spa_name
);
1847 EXPORT_SYMBOL(spa_guid
);
1848 EXPORT_SYMBOL(spa_last_synced_txg
);
1849 EXPORT_SYMBOL(spa_first_txg
);
1850 EXPORT_SYMBOL(spa_syncing_txg
);
1851 EXPORT_SYMBOL(spa_version
);
1852 EXPORT_SYMBOL(spa_state
);
1853 EXPORT_SYMBOL(spa_load_state
);
1854 EXPORT_SYMBOL(spa_freeze_txg
);
1855 EXPORT_SYMBOL(spa_get_asize
);
1856 EXPORT_SYMBOL(spa_get_dspace
);
1857 EXPORT_SYMBOL(spa_update_dspace
);
1858 EXPORT_SYMBOL(spa_deflate
);
1859 EXPORT_SYMBOL(spa_normal_class
);
1860 EXPORT_SYMBOL(spa_log_class
);
1861 EXPORT_SYMBOL(spa_max_replication
);
1862 EXPORT_SYMBOL(spa_prev_software_version
);
1863 EXPORT_SYMBOL(spa_get_failmode
);
1864 EXPORT_SYMBOL(spa_suspended
);
1865 EXPORT_SYMBOL(spa_bootfs
);
1866 EXPORT_SYMBOL(spa_delegation
);
1867 EXPORT_SYMBOL(spa_meta_objset
);
1869 /* Miscellaneous support routines */
1870 EXPORT_SYMBOL(spa_rename
);
1871 EXPORT_SYMBOL(spa_guid_exists
);
1872 EXPORT_SYMBOL(spa_strdup
);
1873 EXPORT_SYMBOL(spa_strfree
);
1874 EXPORT_SYMBOL(spa_get_random
);
1875 EXPORT_SYMBOL(spa_generate_guid
);
1876 EXPORT_SYMBOL(sprintf_blkptr
);
1877 EXPORT_SYMBOL(spa_freeze
);
1878 EXPORT_SYMBOL(spa_upgrade
);
1879 EXPORT_SYMBOL(spa_evict_all
);
1880 EXPORT_SYMBOL(spa_lookup_by_guid
);
1881 EXPORT_SYMBOL(spa_has_spare
);
1882 EXPORT_SYMBOL(dva_get_dsize_sync
);
1883 EXPORT_SYMBOL(bp_get_dsize_sync
);
1884 EXPORT_SYMBOL(bp_get_dsize
);
1885 EXPORT_SYMBOL(spa_has_slogs
);
1886 EXPORT_SYMBOL(spa_is_root
);
1887 EXPORT_SYMBOL(spa_writeable
);
1888 EXPORT_SYMBOL(spa_mode
);
1890 EXPORT_SYMBOL(spa_namespace_lock
);
1892 module_param(zfs_deadman_synctime_ms
, ulong
, 0644);
1893 MODULE_PARM_DESC(zfs_deadman_synctime_ms
, "Expiration time in milliseconds");
1895 module_param(zfs_deadman_enabled
, int, 0644);
1896 MODULE_PARM_DESC(zfs_deadman_enabled
, "Enable deadman timer");
1898 module_param(spa_asize_inflation
, int, 0644);
1899 MODULE_PARM_DESC(spa_asize_inflation
,
1900 "SPA size estimate multiplication factor");