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>
29 #include <sys/spa_boot.h>
31 #include <sys/zio_checksum.h>
32 #include <sys/zio_compress.h>
34 #include <sys/dmu_tx.h>
37 #include <sys/vdev_impl.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/dsl_scan.h>
47 #include <sys/fs/zfs.h>
48 #include <sys/metaslab_impl.h>
52 #include "zfeature_common.h"
57 * There are four basic locks for managing spa_t structures:
59 * spa_namespace_lock (global mutex)
61 * This lock must be acquired to do any of the following:
63 * - Lookup a spa_t by name
64 * - Add or remove a spa_t from the namespace
65 * - Increase spa_refcount from non-zero
66 * - Check if spa_refcount is zero
68 * - add/remove/attach/detach devices
69 * - Held for the duration of create/destroy/import/export
71 * It does not need to handle recursion. A create or destroy may
72 * reference objects (files or zvols) in other pools, but by
73 * definition they must have an existing reference, and will never need
74 * to lookup a spa_t by name.
76 * spa_refcount (per-spa refcount_t protected by mutex)
78 * This reference count keep track of any active users of the spa_t. The
79 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
80 * the refcount is never really 'zero' - opening a pool implicitly keeps
81 * some references in the DMU. Internally we check against spa_minref, but
82 * present the image of a zero/non-zero value to consumers.
84 * spa_config_lock[] (per-spa array of rwlocks)
86 * This protects the spa_t from config changes, and must be held in
87 * the following circumstances:
89 * - RW_READER to perform I/O to the spa
90 * - RW_WRITER to change the vdev config
92 * The locking order is fairly straightforward:
94 * spa_namespace_lock -> spa_refcount
96 * The namespace lock must be acquired to increase the refcount from 0
97 * or to check if it is zero.
99 * spa_refcount -> spa_config_lock[]
101 * There must be at least one valid reference on the spa_t to acquire
104 * spa_namespace_lock -> spa_config_lock[]
106 * The namespace lock must always be taken before the config lock.
109 * The spa_namespace_lock can be acquired directly and is globally visible.
111 * The namespace is manipulated using the following functions, all of which
112 * require the spa_namespace_lock to be held.
114 * spa_lookup() Lookup a spa_t by name.
116 * spa_add() Create a new spa_t in the namespace.
118 * spa_remove() Remove a spa_t from the namespace. This also
119 * frees up any memory associated with the spa_t.
121 * spa_next() Returns the next spa_t in the system, or the
122 * first if NULL is passed.
124 * spa_evict_all() Shutdown and remove all spa_t structures in
127 * spa_guid_exists() Determine whether a pool/device guid exists.
129 * The spa_refcount is manipulated using the following functions:
131 * spa_open_ref() Adds a reference to the given spa_t. Must be
132 * called with spa_namespace_lock held if the
133 * refcount is currently zero.
135 * spa_close() Remove a reference from the spa_t. This will
136 * not free the spa_t or remove it from the
137 * namespace. No locking is required.
139 * spa_refcount_zero() Returns true if the refcount is currently
140 * zero. Must be called with spa_namespace_lock
143 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
144 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
145 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
147 * To read the configuration, it suffices to hold one of these locks as reader.
148 * To modify the configuration, you must hold all locks as writer. To modify
149 * vdev state without altering the vdev tree's topology (e.g. online/offline),
150 * you must hold SCL_STATE and SCL_ZIO as writer.
152 * We use these distinct config locks to avoid recursive lock entry.
153 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
154 * block allocations (SCL_ALLOC), which may require reading space maps
155 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
157 * The spa config locks cannot be normal rwlocks because we need the
158 * ability to hand off ownership. For example, SCL_ZIO is acquired
159 * by the issuing thread and later released by an interrupt thread.
160 * They do, however, obey the usual write-wanted semantics to prevent
161 * writer (i.e. system administrator) starvation.
163 * The lock acquisition rules are as follows:
166 * Protects changes to the vdev tree topology, such as vdev
167 * add/remove/attach/detach. Protects the dirty config list
168 * (spa_config_dirty_list) and the set of spares and l2arc devices.
171 * Protects changes to pool state and vdev state, such as vdev
172 * online/offline/fault/degrade/clear. Protects the dirty state list
173 * (spa_state_dirty_list) and global pool state (spa_state).
176 * Protects changes to metaslab groups and classes.
177 * Held as reader by metaslab_alloc() and metaslab_claim().
180 * Held by bp-level zios (those which have no io_vd upon entry)
181 * to prevent changes to the vdev tree. The bp-level zio implicitly
182 * protects all of its vdev child zios, which do not hold SCL_ZIO.
185 * Protects changes to metaslab groups and classes.
186 * Held as reader by metaslab_free(). SCL_FREE is distinct from
187 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
188 * blocks in zio_done() while another i/o that holds either
189 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
192 * Held as reader to prevent changes to the vdev tree during trivial
193 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
194 * other locks, and lower than all of them, to ensure that it's safe
195 * to acquire regardless of caller context.
197 * In addition, the following rules apply:
199 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
200 * The lock ordering is SCL_CONFIG > spa_props_lock.
202 * (b) I/O operations on leaf vdevs. For any zio operation that takes
203 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
204 * or zio_write_phys() -- the caller must ensure that the config cannot
205 * cannot change in the interim, and that the vdev cannot be reopened.
206 * SCL_STATE as reader suffices for both.
208 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
210 * spa_vdev_enter() Acquire the namespace lock and the config lock
213 * spa_vdev_exit() Release the config lock, wait for all I/O
214 * to complete, sync the updated configs to the
215 * cache, and release the namespace lock.
217 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
218 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
219 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
221 * spa_rename() is also implemented within this file since it requires
222 * manipulation of the namespace.
225 static avl_tree_t spa_namespace_avl
;
226 kmutex_t spa_namespace_lock
;
227 static kcondvar_t spa_namespace_cv
;
228 static int spa_active_count
;
229 int spa_max_replication_override
= SPA_DVAS_PER_BP
;
231 static kmutex_t spa_spare_lock
;
232 static avl_tree_t spa_spare_avl
;
233 static kmutex_t spa_l2cache_lock
;
234 static avl_tree_t spa_l2cache_avl
;
236 kmem_cache_t
*spa_buffer_pool
;
240 /* Everything except dprintf and spa is on by default in debug builds */
241 int zfs_flags
= ~(ZFS_DEBUG_DPRINTF
| ZFS_DEBUG_SPA
);
247 * zfs_recover can be set to nonzero to attempt to recover from
248 * otherwise-fatal errors, typically caused by on-disk corruption. When
249 * set, calls to zfs_panic_recover() will turn into warning messages.
250 * This should only be used as a last resort, as it typically results
251 * in leaked space, or worse.
256 * Expiration time in milliseconds. This value has two meanings. First it is
257 * used to determine when the spa_deadman() logic should fire. By default the
258 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
259 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
260 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
263 uint64_t zfs_deadman_synctime_ms
= 1000000ULL;
266 * Check time in milliseconds. This defines the frequency at which we check
269 uint64_t zfs_deadman_checktime_ms
= 5000ULL;
272 * Override the zfs deadman behavior via /etc/system. By default the
273 * deadman is enabled except on VMware and sparc deployments.
275 int zfs_deadman_enabled
= -1;
278 * The worst case is single-sector max-parity RAID-Z blocks, in which
279 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
280 * times the size; so just assume that. Add to this the fact that
281 * we can have up to 3 DVAs per bp, and one more factor of 2 because
282 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together,
284 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
286 int spa_asize_inflation
= 24;
289 * ==========================================================================
291 * ==========================================================================
294 spa_config_lock_init(spa_t
*spa
)
296 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
297 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
298 mutex_init(&scl
->scl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
299 cv_init(&scl
->scl_cv
, NULL
, CV_DEFAULT
, NULL
);
300 refcount_create_untracked(&scl
->scl_count
);
301 scl
->scl_writer
= NULL
;
302 scl
->scl_write_wanted
= 0;
307 spa_config_lock_destroy(spa_t
*spa
)
309 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
310 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
311 mutex_destroy(&scl
->scl_lock
);
312 cv_destroy(&scl
->scl_cv
);
313 refcount_destroy(&scl
->scl_count
);
314 ASSERT(scl
->scl_writer
== NULL
);
315 ASSERT(scl
->scl_write_wanted
== 0);
320 spa_config_tryenter(spa_t
*spa
, int locks
, void *tag
, krw_t rw
)
322 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
323 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
324 if (!(locks
& (1 << i
)))
326 mutex_enter(&scl
->scl_lock
);
327 if (rw
== RW_READER
) {
328 if (scl
->scl_writer
|| scl
->scl_write_wanted
) {
329 mutex_exit(&scl
->scl_lock
);
330 spa_config_exit(spa
, locks
^ (1 << i
), tag
);
334 ASSERT(scl
->scl_writer
!= curthread
);
335 if (!refcount_is_zero(&scl
->scl_count
)) {
336 mutex_exit(&scl
->scl_lock
);
337 spa_config_exit(spa
, locks
^ (1 << i
), tag
);
340 scl
->scl_writer
= curthread
;
342 (void) refcount_add(&scl
->scl_count
, tag
);
343 mutex_exit(&scl
->scl_lock
);
349 spa_config_enter(spa_t
*spa
, int locks
, void *tag
, krw_t rw
)
353 ASSERT3U(SCL_LOCKS
, <, sizeof (wlocks_held
) * NBBY
);
355 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
356 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
357 if (scl
->scl_writer
== curthread
)
358 wlocks_held
|= (1 << i
);
359 if (!(locks
& (1 << i
)))
361 mutex_enter(&scl
->scl_lock
);
362 if (rw
== RW_READER
) {
363 while (scl
->scl_writer
|| scl
->scl_write_wanted
) {
364 cv_wait(&scl
->scl_cv
, &scl
->scl_lock
);
367 ASSERT(scl
->scl_writer
!= curthread
);
368 while (!refcount_is_zero(&scl
->scl_count
)) {
369 scl
->scl_write_wanted
++;
370 cv_wait(&scl
->scl_cv
, &scl
->scl_lock
);
371 scl
->scl_write_wanted
--;
373 scl
->scl_writer
= curthread
;
375 (void) refcount_add(&scl
->scl_count
, tag
);
376 mutex_exit(&scl
->scl_lock
);
378 ASSERT(wlocks_held
<= locks
);
382 spa_config_exit(spa_t
*spa
, int locks
, void *tag
)
384 for (int i
= SCL_LOCKS
- 1; i
>= 0; i
--) {
385 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
386 if (!(locks
& (1 << i
)))
388 mutex_enter(&scl
->scl_lock
);
389 ASSERT(!refcount_is_zero(&scl
->scl_count
));
390 if (refcount_remove(&scl
->scl_count
, tag
) == 0) {
391 ASSERT(scl
->scl_writer
== NULL
||
392 scl
->scl_writer
== curthread
);
393 scl
->scl_writer
= NULL
; /* OK in either case */
394 cv_broadcast(&scl
->scl_cv
);
396 mutex_exit(&scl
->scl_lock
);
401 spa_config_held(spa_t
*spa
, int locks
, krw_t rw
)
405 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
406 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
407 if (!(locks
& (1 << i
)))
409 if ((rw
== RW_READER
&& !refcount_is_zero(&scl
->scl_count
)) ||
410 (rw
== RW_WRITER
&& scl
->scl_writer
== curthread
))
411 locks_held
|= 1 << i
;
418 * ==========================================================================
419 * SPA namespace functions
420 * ==========================================================================
424 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
425 * Returns NULL if no matching spa_t is found.
428 spa_lookup(const char *name
)
430 static spa_t search
; /* spa_t is large; don't allocate on stack */
435 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
437 (void) strlcpy(search
.spa_name
, name
, sizeof (search
.spa_name
));
440 * If it's a full dataset name, figure out the pool name and
443 cp
= strpbrk(search
.spa_name
, "/@#");
447 spa
= avl_find(&spa_namespace_avl
, &search
, &where
);
453 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
454 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
455 * looking for potentially hung I/Os.
458 spa_deadman(void *arg
)
463 * Disable the deadman timer if the pool is suspended.
465 if (spa_suspended(spa
)) {
466 VERIFY(cyclic_reprogram(spa
->spa_deadman_cycid
, CY_INFINITY
));
470 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
471 (gethrtime() - spa
->spa_sync_starttime
) / NANOSEC
,
472 ++spa
->spa_deadman_calls
);
473 if (zfs_deadman_enabled
)
474 vdev_deadman(spa
->spa_root_vdev
);
478 * Create an uninitialized spa_t with the given name. Requires
479 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
480 * exist by calling spa_lookup() first.
483 spa_add(const char *name
, nvlist_t
*config
, const char *altroot
)
486 spa_config_dirent_t
*dp
;
490 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
492 spa
= kmem_zalloc(sizeof (spa_t
), KM_SLEEP
);
494 mutex_init(&spa
->spa_async_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
495 mutex_init(&spa
->spa_errlist_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
496 mutex_init(&spa
->spa_errlog_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
497 mutex_init(&spa
->spa_history_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
498 mutex_init(&spa
->spa_proc_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
499 mutex_init(&spa
->spa_props_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
500 mutex_init(&spa
->spa_scrub_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
501 mutex_init(&spa
->spa_suspend_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
502 mutex_init(&spa
->spa_vdev_top_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
503 mutex_init(&spa
->spa_iokstat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
505 cv_init(&spa
->spa_async_cv
, NULL
, CV_DEFAULT
, NULL
);
506 cv_init(&spa
->spa_proc_cv
, NULL
, CV_DEFAULT
, NULL
);
507 cv_init(&spa
->spa_scrub_io_cv
, NULL
, CV_DEFAULT
, NULL
);
508 cv_init(&spa
->spa_suspend_cv
, NULL
, CV_DEFAULT
, NULL
);
510 for (int t
= 0; t
< TXG_SIZE
; t
++)
511 bplist_create(&spa
->spa_free_bplist
[t
]);
513 (void) strlcpy(spa
->spa_name
, name
, sizeof (spa
->spa_name
));
514 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
515 spa
->spa_freeze_txg
= UINT64_MAX
;
516 spa
->spa_final_txg
= UINT64_MAX
;
517 spa
->spa_load_max_txg
= UINT64_MAX
;
519 spa
->spa_proc_state
= SPA_PROC_NONE
;
521 hdlr
.cyh_func
= spa_deadman
;
523 hdlr
.cyh_level
= CY_LOW_LEVEL
;
525 spa
->spa_deadman_synctime
= MSEC2NSEC(zfs_deadman_synctime_ms
);
528 * This determines how often we need to check for hung I/Os after
529 * the cyclic has already fired. Since checking for hung I/Os is
530 * an expensive operation we don't want to check too frequently.
531 * Instead wait for 5 seconds before checking again.
533 when
.cyt_interval
= MSEC2NSEC(zfs_deadman_checktime_ms
);
534 when
.cyt_when
= CY_INFINITY
;
535 mutex_enter(&cpu_lock
);
536 spa
->spa_deadman_cycid
= cyclic_add(&hdlr
, &when
);
537 mutex_exit(&cpu_lock
);
539 refcount_create(&spa
->spa_refcount
);
540 spa_config_lock_init(spa
);
542 avl_add(&spa_namespace_avl
, spa
);
545 * Set the alternate root, if there is one.
548 spa
->spa_root
= spa_strdup(altroot
);
553 * Every pool starts with the default cachefile
555 list_create(&spa
->spa_config_list
, sizeof (spa_config_dirent_t
),
556 offsetof(spa_config_dirent_t
, scd_link
));
558 dp
= kmem_zalloc(sizeof (spa_config_dirent_t
), KM_SLEEP
);
559 dp
->scd_path
= altroot
? NULL
: spa_strdup(spa_config_path
);
560 list_insert_head(&spa
->spa_config_list
, dp
);
562 VERIFY(nvlist_alloc(&spa
->spa_load_info
, NV_UNIQUE_NAME
,
565 if (config
!= NULL
) {
568 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_FEATURES_FOR_READ
,
570 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
,
574 VERIFY(nvlist_dup(config
, &spa
->spa_config
, 0) == 0);
577 if (spa
->spa_label_features
== NULL
) {
578 VERIFY(nvlist_alloc(&spa
->spa_label_features
, NV_UNIQUE_NAME
,
582 spa
->spa_iokstat
= kstat_create("zfs", 0, name
,
583 "disk", KSTAT_TYPE_IO
, 1, 0);
584 if (spa
->spa_iokstat
) {
585 spa
->spa_iokstat
->ks_lock
= &spa
->spa_iokstat_lock
;
586 kstat_install(spa
->spa_iokstat
);
589 spa
->spa_debug
= ((zfs_flags
& ZFS_DEBUG_SPA
) != 0);
592 * As a pool is being created, treat all features as disabled by
593 * setting SPA_FEATURE_DISABLED for all entries in the feature
596 for (int i
= 0; i
< SPA_FEATURES
; i
++) {
597 spa
->spa_feat_refcount_cache
[i
] = SPA_FEATURE_DISABLED
;
604 * Removes a spa_t from the namespace, freeing up any memory used. Requires
605 * spa_namespace_lock. This is called only after the spa_t has been closed and
609 spa_remove(spa_t
*spa
)
611 spa_config_dirent_t
*dp
;
613 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
614 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
616 nvlist_free(spa
->spa_config_splitting
);
618 avl_remove(&spa_namespace_avl
, spa
);
619 cv_broadcast(&spa_namespace_cv
);
622 spa_strfree(spa
->spa_root
);
626 while ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
627 list_remove(&spa
->spa_config_list
, dp
);
628 if (dp
->scd_path
!= NULL
)
629 spa_strfree(dp
->scd_path
);
630 kmem_free(dp
, sizeof (spa_config_dirent_t
));
633 list_destroy(&spa
->spa_config_list
);
635 nvlist_free(spa
->spa_label_features
);
636 nvlist_free(spa
->spa_load_info
);
637 spa_config_set(spa
, NULL
);
639 mutex_enter(&cpu_lock
);
640 if (spa
->spa_deadman_cycid
!= CYCLIC_NONE
)
641 cyclic_remove(spa
->spa_deadman_cycid
);
642 mutex_exit(&cpu_lock
);
643 spa
->spa_deadman_cycid
= CYCLIC_NONE
;
645 refcount_destroy(&spa
->spa_refcount
);
647 spa_config_lock_destroy(spa
);
649 kstat_delete(spa
->spa_iokstat
);
650 spa
->spa_iokstat
= NULL
;
652 for (int t
= 0; t
< TXG_SIZE
; t
++)
653 bplist_destroy(&spa
->spa_free_bplist
[t
]);
655 cv_destroy(&spa
->spa_async_cv
);
656 cv_destroy(&spa
->spa_proc_cv
);
657 cv_destroy(&spa
->spa_scrub_io_cv
);
658 cv_destroy(&spa
->spa_suspend_cv
);
660 mutex_destroy(&spa
->spa_async_lock
);
661 mutex_destroy(&spa
->spa_errlist_lock
);
662 mutex_destroy(&spa
->spa_errlog_lock
);
663 mutex_destroy(&spa
->spa_history_lock
);
664 mutex_destroy(&spa
->spa_proc_lock
);
665 mutex_destroy(&spa
->spa_props_lock
);
666 mutex_destroy(&spa
->spa_scrub_lock
);
667 mutex_destroy(&spa
->spa_suspend_lock
);
668 mutex_destroy(&spa
->spa_vdev_top_lock
);
669 mutex_destroy(&spa
->spa_iokstat_lock
);
671 kmem_free(spa
, sizeof (spa_t
));
675 * Given a pool, return the next pool in the namespace, or NULL if there is
676 * none. If 'prev' is NULL, return the first pool.
679 spa_next(spa_t
*prev
)
681 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
684 return (AVL_NEXT(&spa_namespace_avl
, prev
));
686 return (avl_first(&spa_namespace_avl
));
690 * ==========================================================================
691 * SPA refcount functions
692 * ==========================================================================
696 * Add a reference to the given spa_t. Must have at least one reference, or
697 * have the namespace lock held.
700 spa_open_ref(spa_t
*spa
, void *tag
)
702 ASSERT(refcount_count(&spa
->spa_refcount
) >= spa
->spa_minref
||
703 MUTEX_HELD(&spa_namespace_lock
));
704 (void) refcount_add(&spa
->spa_refcount
, tag
);
708 * Remove a reference to the given spa_t. Must have at least one reference, or
709 * have the namespace lock held.
712 spa_close(spa_t
*spa
, void *tag
)
714 ASSERT(refcount_count(&spa
->spa_refcount
) > spa
->spa_minref
||
715 MUTEX_HELD(&spa_namespace_lock
));
716 (void) refcount_remove(&spa
->spa_refcount
, tag
);
720 * Check to see if the spa refcount is zero. Must be called with
721 * spa_namespace_lock held. We really compare against spa_minref, which is the
722 * number of references acquired when opening a pool
725 spa_refcount_zero(spa_t
*spa
)
727 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
729 return (refcount_count(&spa
->spa_refcount
) == spa
->spa_minref
);
733 * ==========================================================================
734 * SPA spare and l2cache tracking
735 * ==========================================================================
739 * Hot spares and cache devices are tracked using the same code below,
740 * for 'auxiliary' devices.
743 typedef struct spa_aux
{
751 spa_aux_compare(const void *a
, const void *b
)
753 const spa_aux_t
*sa
= a
;
754 const spa_aux_t
*sb
= b
;
756 if (sa
->aux_guid
< sb
->aux_guid
)
758 else if (sa
->aux_guid
> sb
->aux_guid
)
765 spa_aux_add(vdev_t
*vd
, avl_tree_t
*avl
)
771 search
.aux_guid
= vd
->vdev_guid
;
772 if ((aux
= avl_find(avl
, &search
, &where
)) != NULL
) {
775 aux
= kmem_zalloc(sizeof (spa_aux_t
), KM_SLEEP
);
776 aux
->aux_guid
= vd
->vdev_guid
;
778 avl_insert(avl
, aux
, where
);
783 spa_aux_remove(vdev_t
*vd
, avl_tree_t
*avl
)
789 search
.aux_guid
= vd
->vdev_guid
;
790 aux
= avl_find(avl
, &search
, &where
);
794 if (--aux
->aux_count
== 0) {
795 avl_remove(avl
, aux
);
796 kmem_free(aux
, sizeof (spa_aux_t
));
797 } else if (aux
->aux_pool
== spa_guid(vd
->vdev_spa
)) {
798 aux
->aux_pool
= 0ULL;
803 spa_aux_exists(uint64_t guid
, uint64_t *pool
, int *refcnt
, avl_tree_t
*avl
)
805 spa_aux_t search
, *found
;
807 search
.aux_guid
= guid
;
808 found
= avl_find(avl
, &search
, NULL
);
812 *pool
= found
->aux_pool
;
819 *refcnt
= found
->aux_count
;
824 return (found
!= NULL
);
828 spa_aux_activate(vdev_t
*vd
, avl_tree_t
*avl
)
830 spa_aux_t search
, *found
;
833 search
.aux_guid
= vd
->vdev_guid
;
834 found
= avl_find(avl
, &search
, &where
);
835 ASSERT(found
!= NULL
);
836 ASSERT(found
->aux_pool
== 0ULL);
838 found
->aux_pool
= spa_guid(vd
->vdev_spa
);
842 * Spares are tracked globally due to the following constraints:
844 * - A spare may be part of multiple pools.
845 * - A spare may be added to a pool even if it's actively in use within
847 * - A spare in use in any pool can only be the source of a replacement if
848 * the target is a spare in the same pool.
850 * We keep track of all spares on the system through the use of a reference
851 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
852 * spare, then we bump the reference count in the AVL tree. In addition, we set
853 * the 'vdev_isspare' member to indicate that the device is a spare (active or
854 * inactive). When a spare is made active (used to replace a device in the
855 * pool), we also keep track of which pool its been made a part of.
857 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
858 * called under the spa_namespace lock as part of vdev reconfiguration. The
859 * separate spare lock exists for the status query path, which does not need to
860 * be completely consistent with respect to other vdev configuration changes.
864 spa_spare_compare(const void *a
, const void *b
)
866 return (spa_aux_compare(a
, b
));
870 spa_spare_add(vdev_t
*vd
)
872 mutex_enter(&spa_spare_lock
);
873 ASSERT(!vd
->vdev_isspare
);
874 spa_aux_add(vd
, &spa_spare_avl
);
875 vd
->vdev_isspare
= B_TRUE
;
876 mutex_exit(&spa_spare_lock
);
880 spa_spare_remove(vdev_t
*vd
)
882 mutex_enter(&spa_spare_lock
);
883 ASSERT(vd
->vdev_isspare
);
884 spa_aux_remove(vd
, &spa_spare_avl
);
885 vd
->vdev_isspare
= B_FALSE
;
886 mutex_exit(&spa_spare_lock
);
890 spa_spare_exists(uint64_t guid
, uint64_t *pool
, int *refcnt
)
894 mutex_enter(&spa_spare_lock
);
895 found
= spa_aux_exists(guid
, pool
, refcnt
, &spa_spare_avl
);
896 mutex_exit(&spa_spare_lock
);
902 spa_spare_activate(vdev_t
*vd
)
904 mutex_enter(&spa_spare_lock
);
905 ASSERT(vd
->vdev_isspare
);
906 spa_aux_activate(vd
, &spa_spare_avl
);
907 mutex_exit(&spa_spare_lock
);
911 * Level 2 ARC devices are tracked globally for the same reasons as spares.
912 * Cache devices currently only support one pool per cache device, and so
913 * for these devices the aux reference count is currently unused beyond 1.
917 spa_l2cache_compare(const void *a
, const void *b
)
919 return (spa_aux_compare(a
, b
));
923 spa_l2cache_add(vdev_t
*vd
)
925 mutex_enter(&spa_l2cache_lock
);
926 ASSERT(!vd
->vdev_isl2cache
);
927 spa_aux_add(vd
, &spa_l2cache_avl
);
928 vd
->vdev_isl2cache
= B_TRUE
;
929 mutex_exit(&spa_l2cache_lock
);
933 spa_l2cache_remove(vdev_t
*vd
)
935 mutex_enter(&spa_l2cache_lock
);
936 ASSERT(vd
->vdev_isl2cache
);
937 spa_aux_remove(vd
, &spa_l2cache_avl
);
938 vd
->vdev_isl2cache
= B_FALSE
;
939 mutex_exit(&spa_l2cache_lock
);
943 spa_l2cache_exists(uint64_t guid
, uint64_t *pool
)
947 mutex_enter(&spa_l2cache_lock
);
948 found
= spa_aux_exists(guid
, pool
, NULL
, &spa_l2cache_avl
);
949 mutex_exit(&spa_l2cache_lock
);
955 spa_l2cache_activate(vdev_t
*vd
)
957 mutex_enter(&spa_l2cache_lock
);
958 ASSERT(vd
->vdev_isl2cache
);
959 spa_aux_activate(vd
, &spa_l2cache_avl
);
960 mutex_exit(&spa_l2cache_lock
);
964 * ==========================================================================
966 * ==========================================================================
970 * Lock the given spa_t for the purpose of adding or removing a vdev.
971 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
972 * It returns the next transaction group for the spa_t.
975 spa_vdev_enter(spa_t
*spa
)
977 mutex_enter(&spa
->spa_vdev_top_lock
);
978 mutex_enter(&spa_namespace_lock
);
979 return (spa_vdev_config_enter(spa
));
983 * Internal implementation for spa_vdev_enter(). Used when a vdev
984 * operation requires multiple syncs (i.e. removing a device) while
985 * keeping the spa_namespace_lock held.
988 spa_vdev_config_enter(spa_t
*spa
)
990 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
992 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
994 return (spa_last_synced_txg(spa
) + 1);
998 * Used in combination with spa_vdev_config_enter() to allow the syncing
999 * of multiple transactions without releasing the spa_namespace_lock.
1002 spa_vdev_config_exit(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int error
, char *tag
)
1004 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1006 int config_changed
= B_FALSE
;
1008 ASSERT(txg
> spa_last_synced_txg(spa
));
1010 spa
->spa_pending_vdev
= NULL
;
1013 * Reassess the DTLs.
1015 vdev_dtl_reassess(spa
->spa_root_vdev
, 0, 0, B_FALSE
);
1017 if (error
== 0 && !list_is_empty(&spa
->spa_config_dirty_list
)) {
1018 config_changed
= B_TRUE
;
1019 spa
->spa_config_generation
++;
1023 * Verify the metaslab classes.
1025 ASSERT(metaslab_class_validate(spa_normal_class(spa
)) == 0);
1026 ASSERT(metaslab_class_validate(spa_log_class(spa
)) == 0);
1028 spa_config_exit(spa
, SCL_ALL
, spa
);
1031 * Panic the system if the specified tag requires it. This
1032 * is useful for ensuring that configurations are updated
1035 if (zio_injection_enabled
)
1036 zio_handle_panic_injection(spa
, tag
, 0);
1039 * Note: this txg_wait_synced() is important because it ensures
1040 * that there won't be more than one config change per txg.
1041 * This allows us to use the txg as the generation number.
1044 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
1047 ASSERT(!vd
->vdev_detached
|| vd
->vdev_dtl_sm
== NULL
);
1048 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
1050 spa_config_exit(spa
, SCL_ALL
, spa
);
1054 * If the config changed, update the config cache.
1057 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
1061 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1062 * locking of spa_vdev_enter(), we also want make sure the transactions have
1063 * synced to disk, and then update the global configuration cache with the new
1067 spa_vdev_exit(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int error
)
1069 spa_vdev_config_exit(spa
, vd
, txg
, error
, FTAG
);
1070 mutex_exit(&spa_namespace_lock
);
1071 mutex_exit(&spa
->spa_vdev_top_lock
);
1077 * Lock the given spa_t for the purpose of changing vdev state.
1080 spa_vdev_state_enter(spa_t
*spa
, int oplocks
)
1082 int locks
= SCL_STATE_ALL
| oplocks
;
1085 * Root pools may need to read of the underlying devfs filesystem
1086 * when opening up a vdev. Unfortunately if we're holding the
1087 * SCL_ZIO lock it will result in a deadlock when we try to issue
1088 * the read from the root filesystem. Instead we "prefetch"
1089 * the associated vnodes that we need prior to opening the
1090 * underlying devices and cache them so that we can prevent
1091 * any I/O when we are doing the actual open.
1093 if (spa_is_root(spa
)) {
1094 int low
= locks
& ~(SCL_ZIO
- 1);
1095 int high
= locks
& ~low
;
1097 spa_config_enter(spa
, high
, spa
, RW_WRITER
);
1098 vdev_hold(spa
->spa_root_vdev
);
1099 spa_config_enter(spa
, low
, spa
, RW_WRITER
);
1101 spa_config_enter(spa
, locks
, spa
, RW_WRITER
);
1103 spa
->spa_vdev_locks
= locks
;
1107 spa_vdev_state_exit(spa_t
*spa
, vdev_t
*vd
, int error
)
1109 boolean_t config_changed
= B_FALSE
;
1111 if (vd
!= NULL
|| error
== 0)
1112 vdev_dtl_reassess(vd
? vd
->vdev_top
: spa
->spa_root_vdev
,
1116 vdev_state_dirty(vd
->vdev_top
);
1117 config_changed
= B_TRUE
;
1118 spa
->spa_config_generation
++;
1121 if (spa_is_root(spa
))
1122 vdev_rele(spa
->spa_root_vdev
);
1124 ASSERT3U(spa
->spa_vdev_locks
, >=, SCL_STATE_ALL
);
1125 spa_config_exit(spa
, spa
->spa_vdev_locks
, spa
);
1128 * If anything changed, wait for it to sync. This ensures that,
1129 * from the system administrator's perspective, zpool(1M) commands
1130 * are synchronous. This is important for things like zpool offline:
1131 * when the command completes, you expect no further I/O from ZFS.
1134 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1137 * If the config changed, update the config cache.
1139 if (config_changed
) {
1140 mutex_enter(&spa_namespace_lock
);
1141 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
1142 mutex_exit(&spa_namespace_lock
);
1149 * ==========================================================================
1150 * Miscellaneous functions
1151 * ==========================================================================
1155 spa_activate_mos_feature(spa_t
*spa
, const char *feature
, dmu_tx_t
*tx
)
1157 if (!nvlist_exists(spa
->spa_label_features
, feature
)) {
1158 fnvlist_add_boolean(spa
->spa_label_features
, feature
);
1160 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1161 * dirty the vdev config because lock SCL_CONFIG is not held.
1162 * Thankfully, in this case we don't need to dirty the config
1163 * because it will be written out anyway when we finish
1164 * creating the pool.
1166 if (tx
->tx_txg
!= TXG_INITIAL
)
1167 vdev_config_dirty(spa
->spa_root_vdev
);
1172 spa_deactivate_mos_feature(spa_t
*spa
, const char *feature
)
1174 if (nvlist_remove_all(spa
->spa_label_features
, feature
) == 0)
1175 vdev_config_dirty(spa
->spa_root_vdev
);
1182 spa_rename(const char *name
, const char *newname
)
1188 * Lookup the spa_t and grab the config lock for writing. We need to
1189 * actually open the pool so that we can sync out the necessary labels.
1190 * It's OK to call spa_open() with the namespace lock held because we
1191 * allow recursive calls for other reasons.
1193 mutex_enter(&spa_namespace_lock
);
1194 if ((err
= spa_open(name
, &spa
, FTAG
)) != 0) {
1195 mutex_exit(&spa_namespace_lock
);
1199 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1201 avl_remove(&spa_namespace_avl
, spa
);
1202 (void) strlcpy(spa
->spa_name
, newname
, sizeof (spa
->spa_name
));
1203 avl_add(&spa_namespace_avl
, spa
);
1206 * Sync all labels to disk with the new names by marking the root vdev
1207 * dirty and waiting for it to sync. It will pick up the new pool name
1210 vdev_config_dirty(spa
->spa_root_vdev
);
1212 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1214 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1217 * Sync the updated config cache.
1219 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
1221 spa_close(spa
, FTAG
);
1223 mutex_exit(&spa_namespace_lock
);
1229 * Return the spa_t associated with given pool_guid, if it exists. If
1230 * device_guid is non-zero, determine whether the pool exists *and* contains
1231 * a device with the specified device_guid.
1234 spa_by_guid(uint64_t pool_guid
, uint64_t device_guid
)
1237 avl_tree_t
*t
= &spa_namespace_avl
;
1239 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1241 for (spa
= avl_first(t
); spa
!= NULL
; spa
= AVL_NEXT(t
, spa
)) {
1242 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1244 if (spa
->spa_root_vdev
== NULL
)
1246 if (spa_guid(spa
) == pool_guid
) {
1247 if (device_guid
== 0)
1250 if (vdev_lookup_by_guid(spa
->spa_root_vdev
,
1251 device_guid
) != NULL
)
1255 * Check any devices we may be in the process of adding.
1257 if (spa
->spa_pending_vdev
) {
1258 if (vdev_lookup_by_guid(spa
->spa_pending_vdev
,
1259 device_guid
) != NULL
)
1269 * Determine whether a pool with the given pool_guid exists.
1272 spa_guid_exists(uint64_t pool_guid
, uint64_t device_guid
)
1274 return (spa_by_guid(pool_guid
, device_guid
) != NULL
);
1278 spa_strdup(const char *s
)
1284 new = kmem_alloc(len
+ 1, KM_SLEEP
);
1292 spa_strfree(char *s
)
1294 kmem_free(s
, strlen(s
) + 1);
1298 spa_get_random(uint64_t range
)
1304 (void) random_get_pseudo_bytes((void *)&r
, sizeof (uint64_t));
1310 spa_generate_guid(spa_t
*spa
)
1312 uint64_t guid
= spa_get_random(-1ULL);
1315 while (guid
== 0 || spa_guid_exists(spa_guid(spa
), guid
))
1316 guid
= spa_get_random(-1ULL);
1318 while (guid
== 0 || spa_guid_exists(guid
, 0))
1319 guid
= spa_get_random(-1ULL);
1326 snprintf_blkptr(char *buf
, size_t buflen
, const blkptr_t
*bp
)
1329 char *checksum
= NULL
;
1330 char *compress
= NULL
;
1333 if (BP_GET_TYPE(bp
) & DMU_OT_NEWTYPE
) {
1334 dmu_object_byteswap_t bswap
=
1335 DMU_OT_BYTESWAP(BP_GET_TYPE(bp
));
1336 (void) snprintf(type
, sizeof (type
), "bswap %s %s",
1337 DMU_OT_IS_METADATA(BP_GET_TYPE(bp
)) ?
1338 "metadata" : "data",
1339 dmu_ot_byteswap
[bswap
].ob_name
);
1341 (void) strlcpy(type
, dmu_ot
[BP_GET_TYPE(bp
)].ot_name
,
1344 if (!BP_IS_EMBEDDED(bp
)) {
1346 zio_checksum_table
[BP_GET_CHECKSUM(bp
)].ci_name
;
1348 compress
= zio_compress_table
[BP_GET_COMPRESS(bp
)].ci_name
;
1351 SNPRINTF_BLKPTR(snprintf
, ' ', buf
, buflen
, bp
, type
, checksum
,
1356 spa_freeze(spa_t
*spa
)
1358 uint64_t freeze_txg
= 0;
1360 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1361 if (spa
->spa_freeze_txg
== UINT64_MAX
) {
1362 freeze_txg
= spa_last_synced_txg(spa
) + TXG_SIZE
;
1363 spa
->spa_freeze_txg
= freeze_txg
;
1365 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1366 if (freeze_txg
!= 0)
1367 txg_wait_synced(spa_get_dsl(spa
), freeze_txg
);
1371 zfs_panic_recover(const char *fmt
, ...)
1376 vcmn_err(zfs_recover
? CE_WARN
: CE_PANIC
, fmt
, adx
);
1381 * This is a stripped-down version of strtoull, suitable only for converting
1382 * lowercase hexadecimal numbers that don't overflow.
1385 strtonum(const char *str
, char **nptr
)
1391 while ((c
= *str
) != '\0') {
1392 if (c
>= '0' && c
<= '9')
1394 else if (c
>= 'a' && c
<= 'f')
1395 digit
= 10 + c
- 'a';
1406 *nptr
= (char *)str
;
1412 * ==========================================================================
1413 * Accessor functions
1414 * ==========================================================================
1418 spa_shutting_down(spa_t
*spa
)
1420 return (spa
->spa_async_suspended
);
1424 spa_get_dsl(spa_t
*spa
)
1426 return (spa
->spa_dsl_pool
);
1430 spa_is_initializing(spa_t
*spa
)
1432 return (spa
->spa_is_initializing
);
1436 spa_get_rootblkptr(spa_t
*spa
)
1438 return (&spa
->spa_ubsync
.ub_rootbp
);
1442 spa_set_rootblkptr(spa_t
*spa
, const blkptr_t
*bp
)
1444 spa
->spa_uberblock
.ub_rootbp
= *bp
;
1448 spa_altroot(spa_t
*spa
, char *buf
, size_t buflen
)
1450 if (spa
->spa_root
== NULL
)
1453 (void) strncpy(buf
, spa
->spa_root
, buflen
);
1457 spa_sync_pass(spa_t
*spa
)
1459 return (spa
->spa_sync_pass
);
1463 spa_name(spa_t
*spa
)
1465 return (spa
->spa_name
);
1469 spa_guid(spa_t
*spa
)
1471 dsl_pool_t
*dp
= spa_get_dsl(spa
);
1475 * If we fail to parse the config during spa_load(), we can go through
1476 * the error path (which posts an ereport) and end up here with no root
1477 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1480 if (spa
->spa_root_vdev
== NULL
)
1481 return (spa
->spa_config_guid
);
1483 guid
= spa
->spa_last_synced_guid
!= 0 ?
1484 spa
->spa_last_synced_guid
: spa
->spa_root_vdev
->vdev_guid
;
1487 * Return the most recently synced out guid unless we're
1488 * in syncing context.
1490 if (dp
&& dsl_pool_sync_context(dp
))
1491 return (spa
->spa_root_vdev
->vdev_guid
);
1497 spa_load_guid(spa_t
*spa
)
1500 * This is a GUID that exists solely as a reference for the
1501 * purposes of the arc. It is generated at load time, and
1502 * is never written to persistent storage.
1504 return (spa
->spa_load_guid
);
1508 spa_last_synced_txg(spa_t
*spa
)
1510 return (spa
->spa_ubsync
.ub_txg
);
1514 spa_first_txg(spa_t
*spa
)
1516 return (spa
->spa_first_txg
);
1520 spa_syncing_txg(spa_t
*spa
)
1522 return (spa
->spa_syncing_txg
);
1526 spa_state(spa_t
*spa
)
1528 return (spa
->spa_state
);
1532 spa_load_state(spa_t
*spa
)
1534 return (spa
->spa_load_state
);
1538 spa_freeze_txg(spa_t
*spa
)
1540 return (spa
->spa_freeze_txg
);
1545 spa_get_asize(spa_t
*spa
, uint64_t lsize
)
1547 return (lsize
* spa_asize_inflation
);
1551 spa_get_dspace(spa_t
*spa
)
1553 return (spa
->spa_dspace
);
1557 spa_update_dspace(spa_t
*spa
)
1559 spa
->spa_dspace
= metaslab_class_get_dspace(spa_normal_class(spa
)) +
1560 ddt_get_dedup_dspace(spa
);
1564 * Return the failure mode that has been set to this pool. The default
1565 * behavior will be to block all I/Os when a complete failure occurs.
1568 spa_get_failmode(spa_t
*spa
)
1570 return (spa
->spa_failmode
);
1574 spa_suspended(spa_t
*spa
)
1576 return (spa
->spa_suspended
);
1580 spa_version(spa_t
*spa
)
1582 return (spa
->spa_ubsync
.ub_version
);
1586 spa_deflate(spa_t
*spa
)
1588 return (spa
->spa_deflate
);
1592 spa_normal_class(spa_t
*spa
)
1594 return (spa
->spa_normal_class
);
1598 spa_log_class(spa_t
*spa
)
1600 return (spa
->spa_log_class
);
1604 spa_max_replication(spa_t
*spa
)
1607 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1608 * handle BPs with more than one DVA allocated. Set our max
1609 * replication level accordingly.
1611 if (spa_version(spa
) < SPA_VERSION_DITTO_BLOCKS
)
1613 return (MIN(SPA_DVAS_PER_BP
, spa_max_replication_override
));
1617 spa_prev_software_version(spa_t
*spa
)
1619 return (spa
->spa_prev_software_version
);
1623 spa_deadman_synctime(spa_t
*spa
)
1625 return (spa
->spa_deadman_synctime
);
1629 dva_get_dsize_sync(spa_t
*spa
, const dva_t
*dva
)
1631 uint64_t asize
= DVA_GET_ASIZE(dva
);
1632 uint64_t dsize
= asize
;
1634 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1636 if (asize
!= 0 && spa
->spa_deflate
) {
1637 vdev_t
*vd
= vdev_lookup_top(spa
, DVA_GET_VDEV(dva
));
1638 dsize
= (asize
>> SPA_MINBLOCKSHIFT
) * vd
->vdev_deflate_ratio
;
1645 bp_get_dsize_sync(spa_t
*spa
, const blkptr_t
*bp
)
1649 for (int d
= 0; d
< BP_GET_NDVAS(bp
); d
++)
1650 dsize
+= dva_get_dsize_sync(spa
, &bp
->blk_dva
[d
]);
1656 bp_get_dsize(spa_t
*spa
, const blkptr_t
*bp
)
1660 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
1662 for (int d
= 0; d
< BP_GET_NDVAS(bp
); d
++)
1663 dsize
+= dva_get_dsize_sync(spa
, &bp
->blk_dva
[d
]);
1665 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
1671 * ==========================================================================
1672 * Initialization and Termination
1673 * ==========================================================================
1677 spa_name_compare(const void *a1
, const void *a2
)
1679 const spa_t
*s1
= a1
;
1680 const spa_t
*s2
= a2
;
1683 s
= strcmp(s1
->spa_name
, s2
->spa_name
);
1694 return (spa_active_count
);
1706 mutex_init(&spa_namespace_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1707 mutex_init(&spa_spare_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1708 mutex_init(&spa_l2cache_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1709 cv_init(&spa_namespace_cv
, NULL
, CV_DEFAULT
, NULL
);
1711 avl_create(&spa_namespace_avl
, spa_name_compare
, sizeof (spa_t
),
1712 offsetof(spa_t
, spa_avl
));
1714 avl_create(&spa_spare_avl
, spa_spare_compare
, sizeof (spa_aux_t
),
1715 offsetof(spa_aux_t
, aux_avl
));
1717 avl_create(&spa_l2cache_avl
, spa_l2cache_compare
, sizeof (spa_aux_t
),
1718 offsetof(spa_aux_t
, aux_avl
));
1720 spa_mode_global
= mode
;
1725 if (spa_mode_global
!= FREAD
&& dprintf_find_string("watch")) {
1726 arc_procfd
= open("/proc/self/ctl", O_WRONLY
);
1727 if (arc_procfd
== -1) {
1728 perror("could not enable watchpoints: "
1729 "opening /proc/self/ctl failed: ");
1742 vdev_cache_stat_init();
1745 zpool_feature_init();
1757 vdev_cache_stat_fini();
1765 avl_destroy(&spa_namespace_avl
);
1766 avl_destroy(&spa_spare_avl
);
1767 avl_destroy(&spa_l2cache_avl
);
1769 cv_destroy(&spa_namespace_cv
);
1770 mutex_destroy(&spa_namespace_lock
);
1771 mutex_destroy(&spa_spare_lock
);
1772 mutex_destroy(&spa_l2cache_lock
);
1776 * Return whether this pool has slogs. No locking needed.
1777 * It's not a problem if the wrong answer is returned as it's only for
1778 * performance and not correctness
1781 spa_has_slogs(spa_t
*spa
)
1783 return (spa
->spa_log_class
->mc_rotor
!= NULL
);
1787 spa_get_log_state(spa_t
*spa
)
1789 return (spa
->spa_log_state
);
1793 spa_set_log_state(spa_t
*spa
, spa_log_state_t state
)
1795 spa
->spa_log_state
= state
;
1799 spa_is_root(spa_t
*spa
)
1801 return (spa
->spa_is_root
);
1805 spa_writeable(spa_t
*spa
)
1807 return (!!(spa
->spa_mode
& FWRITE
));
1811 spa_mode(spa_t
*spa
)
1813 return (spa
->spa_mode
);
1817 spa_bootfs(spa_t
*spa
)
1819 return (spa
->spa_bootfs
);
1823 spa_delegation(spa_t
*spa
)
1825 return (spa
->spa_delegation
);
1829 spa_meta_objset(spa_t
*spa
)
1831 return (spa
->spa_meta_objset
);
1835 spa_dedup_checksum(spa_t
*spa
)
1837 return (spa
->spa_dedup_checksum
);
1841 * Reset pool scan stat per scan pass (or reboot).
1844 spa_scan_stat_init(spa_t
*spa
)
1846 /* data not stored on disk */
1847 spa
->spa_scan_pass_start
= gethrestime_sec();
1848 spa
->spa_scan_pass_exam
= 0;
1849 vdev_scan_stat_init(spa
->spa_root_vdev
);
1853 * Get scan stats for zpool status reports
1856 spa_scan_get_stats(spa_t
*spa
, pool_scan_stat_t
*ps
)
1858 dsl_scan_t
*scn
= spa
->spa_dsl_pool
? spa
->spa_dsl_pool
->dp_scan
: NULL
;
1860 if (scn
== NULL
|| scn
->scn_phys
.scn_func
== POOL_SCAN_NONE
)
1861 return (SET_ERROR(ENOENT
));
1862 bzero(ps
, sizeof (pool_scan_stat_t
));
1864 /* data stored on disk */
1865 ps
->pss_func
= scn
->scn_phys
.scn_func
;
1866 ps
->pss_start_time
= scn
->scn_phys
.scn_start_time
;
1867 ps
->pss_end_time
= scn
->scn_phys
.scn_end_time
;
1868 ps
->pss_to_examine
= scn
->scn_phys
.scn_to_examine
;
1869 ps
->pss_examined
= scn
->scn_phys
.scn_examined
;
1870 ps
->pss_to_process
= scn
->scn_phys
.scn_to_process
;
1871 ps
->pss_processed
= scn
->scn_phys
.scn_processed
;
1872 ps
->pss_errors
= scn
->scn_phys
.scn_errors
;
1873 ps
->pss_state
= scn
->scn_phys
.scn_state
;
1875 /* data not stored on disk */
1876 ps
->pss_pass_start
= spa
->spa_scan_pass_start
;
1877 ps
->pss_pass_exam
= spa
->spa_scan_pass_exam
;
1883 spa_debug_enabled(spa_t
*spa
)
1885 return (spa
->spa_debug
);