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, 2015 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 #include <sys/zfs_context.h>
29 #include <sys/spa_impl.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/vdev_file.h>
39 #include <sys/metaslab.h>
40 #include <sys/uberblock_impl.h>
43 #include <sys/unique.h>
44 #include <sys/dsl_pool.h>
45 #include <sys/dsl_dir.h>
46 #include <sys/dsl_prop.h>
47 #include <sys/fm/util.h>
48 #include <sys/dsl_scan.h>
49 #include <sys/fs/zfs.h>
50 #include <sys/metaslab_impl.h>
53 #include <sys/kstat.h>
55 #include "zfeature_common.h"
60 * There are four basic locks for managing spa_t structures:
62 * spa_namespace_lock (global mutex)
64 * This lock must be acquired to do any of the following:
66 * - Lookup a spa_t by name
67 * - Add or remove a spa_t from the namespace
68 * - Increase spa_refcount from non-zero
69 * - Check if spa_refcount is zero
71 * - add/remove/attach/detach devices
72 * - Held for the duration of create/destroy/import/export
74 * It does not need to handle recursion. A create or destroy may
75 * reference objects (files or zvols) in other pools, but by
76 * definition they must have an existing reference, and will never need
77 * to lookup a spa_t by name.
79 * spa_refcount (per-spa refcount_t protected by mutex)
81 * This reference count keep track of any active users of the spa_t. The
82 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
83 * the refcount is never really 'zero' - opening a pool implicitly keeps
84 * some references in the DMU. Internally we check against spa_minref, but
85 * present the image of a zero/non-zero value to consumers.
87 * spa_config_lock[] (per-spa array of rwlocks)
89 * This protects the spa_t from config changes, and must be held in
90 * the following circumstances:
92 * - RW_READER to perform I/O to the spa
93 * - RW_WRITER to change the vdev config
95 * The locking order is fairly straightforward:
97 * spa_namespace_lock -> spa_refcount
99 * The namespace lock must be acquired to increase the refcount from 0
100 * or to check if it is zero.
102 * spa_refcount -> spa_config_lock[]
104 * There must be at least one valid reference on the spa_t to acquire
107 * spa_namespace_lock -> spa_config_lock[]
109 * The namespace lock must always be taken before the config lock.
112 * The spa_namespace_lock can be acquired directly and is globally visible.
114 * The namespace is manipulated using the following functions, all of which
115 * require the spa_namespace_lock to be held.
117 * spa_lookup() Lookup a spa_t by name.
119 * spa_add() Create a new spa_t in the namespace.
121 * spa_remove() Remove a spa_t from the namespace. This also
122 * frees up any memory associated with the spa_t.
124 * spa_next() Returns the next spa_t in the system, or the
125 * first if NULL is passed.
127 * spa_evict_all() Shutdown and remove all spa_t structures in
130 * spa_guid_exists() Determine whether a pool/device guid exists.
132 * The spa_refcount is manipulated using the following functions:
134 * spa_open_ref() Adds a reference to the given spa_t. Must be
135 * called with spa_namespace_lock held if the
136 * refcount is currently zero.
138 * spa_close() Remove a reference from the spa_t. This will
139 * not free the spa_t or remove it from the
140 * namespace. No locking is required.
142 * spa_refcount_zero() Returns true if the refcount is currently
143 * zero. Must be called with spa_namespace_lock
146 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
147 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
148 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
150 * To read the configuration, it suffices to hold one of these locks as reader.
151 * To modify the configuration, you must hold all locks as writer. To modify
152 * vdev state without altering the vdev tree's topology (e.g. online/offline),
153 * you must hold SCL_STATE and SCL_ZIO as writer.
155 * We use these distinct config locks to avoid recursive lock entry.
156 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
157 * block allocations (SCL_ALLOC), which may require reading space maps
158 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
160 * The spa config locks cannot be normal rwlocks because we need the
161 * ability to hand off ownership. For example, SCL_ZIO is acquired
162 * by the issuing thread and later released by an interrupt thread.
163 * They do, however, obey the usual write-wanted semantics to prevent
164 * writer (i.e. system administrator) starvation.
166 * The lock acquisition rules are as follows:
169 * Protects changes to the vdev tree topology, such as vdev
170 * add/remove/attach/detach. Protects the dirty config list
171 * (spa_config_dirty_list) and the set of spares and l2arc devices.
174 * Protects changes to pool state and vdev state, such as vdev
175 * online/offline/fault/degrade/clear. Protects the dirty state list
176 * (spa_state_dirty_list) and global pool state (spa_state).
179 * Protects changes to metaslab groups and classes.
180 * Held as reader by metaslab_alloc() and metaslab_claim().
183 * Held by bp-level zios (those which have no io_vd upon entry)
184 * to prevent changes to the vdev tree. The bp-level zio implicitly
185 * protects all of its vdev child zios, which do not hold SCL_ZIO.
188 * Protects changes to metaslab groups and classes.
189 * Held as reader by metaslab_free(). SCL_FREE is distinct from
190 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
191 * blocks in zio_done() while another i/o that holds either
192 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
195 * Held as reader to prevent changes to the vdev tree during trivial
196 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
197 * other locks, and lower than all of them, to ensure that it's safe
198 * to acquire regardless of caller context.
200 * In addition, the following rules apply:
202 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
203 * The lock ordering is SCL_CONFIG > spa_props_lock.
205 * (b) I/O operations on leaf vdevs. For any zio operation that takes
206 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
207 * or zio_write_phys() -- the caller must ensure that the config cannot
208 * cannot change in the interim, and that the vdev cannot be reopened.
209 * SCL_STATE as reader suffices for both.
211 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
213 * spa_vdev_enter() Acquire the namespace lock and the config lock
216 * spa_vdev_exit() Release the config lock, wait for all I/O
217 * to complete, sync the updated configs to the
218 * cache, and release the namespace lock.
220 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
221 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
222 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
224 * spa_rename() is also implemented within this file since it requires
225 * manipulation of the namespace.
228 static avl_tree_t spa_namespace_avl
;
229 kmutex_t spa_namespace_lock
;
230 static kcondvar_t spa_namespace_cv
;
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 /* Everything except dprintf and spa is on by default in debug builds */
243 int zfs_flags
= ~(ZFS_DEBUG_DPRINTF
| ZFS_DEBUG_SPA
);
249 * zfs_recover can be set to nonzero to attempt to recover from
250 * otherwise-fatal errors, typically caused by on-disk corruption. When
251 * set, calls to zfs_panic_recover() will turn into warning messages.
252 * This should only be used as a last resort, as it typically results
253 * in leaked space, or worse.
255 int zfs_recover
= B_FALSE
;
258 * If destroy encounters an EIO while reading metadata (e.g. indirect
259 * blocks), space referenced by the missing metadata can not be freed.
260 * Normally this causes the background destroy to become "stalled", as
261 * it is unable to make forward progress. While in this stalled state,
262 * all remaining space to free from the error-encountering filesystem is
263 * "temporarily leaked". Set this flag to cause it to ignore the EIO,
264 * permanently leak the space from indirect blocks that can not be read,
265 * and continue to free everything else that it can.
267 * The default, "stalling" behavior is useful if the storage partially
268 * fails (i.e. some but not all i/os fail), and then later recovers. In
269 * this case, we will be able to continue pool operations while it is
270 * partially failed, and when it recovers, we can continue to free the
271 * space, with no leaks. However, note that this case is actually
274 * Typically pools either (a) fail completely (but perhaps temporarily,
275 * e.g. a top-level vdev going offline), or (b) have localized,
276 * permanent errors (e.g. disk returns the wrong data due to bit flip or
277 * firmware bug). In case (a), this setting does not matter because the
278 * pool will be suspended and the sync thread will not be able to make
279 * forward progress regardless. In case (b), because the error is
280 * permanent, the best we can do is leak the minimum amount of space,
281 * which is what setting this flag will do. Therefore, it is reasonable
282 * for this flag to normally be set, but we chose the more conservative
283 * approach of not setting it, so that there is no possibility of
284 * leaking space in the "partial temporary" failure case.
286 int zfs_free_leak_on_eio
= B_FALSE
;
289 * Expiration time in milliseconds. This value has two meanings. First it is
290 * used to determine when the spa_deadman() logic should fire. By default the
291 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
292 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
293 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
296 unsigned long zfs_deadman_synctime_ms
= 1000000ULL;
299 * By default the deadman is enabled.
301 int zfs_deadman_enabled
= 1;
304 * The worst case is single-sector max-parity RAID-Z blocks, in which
305 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
306 * times the size; so just assume that. Add to this the fact that
307 * we can have up to 3 DVAs per bp, and one more factor of 2 because
308 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together,
310 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
312 int spa_asize_inflation
= 24;
315 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
316 * the pool to be consumed. This ensures that we don't run the pool
317 * completely out of space, due to unaccounted changes (e.g. to the MOS).
318 * It also limits the worst-case time to allocate space. If we have
319 * less than this amount of free space, most ZPL operations (e.g. write,
320 * create) will return ENOSPC.
322 * Certain operations (e.g. file removal, most administrative actions) can
323 * use half the slop space. They will only return ENOSPC if less than half
324 * the slop space is free. Typically, once the pool has less than the slop
325 * space free, the user will use these operations to free up space in the pool.
326 * These are the operations that call dsl_pool_adjustedsize() with the netfree
327 * argument set to TRUE.
329 * A very restricted set of operations are always permitted, regardless of
330 * the amount of free space. These are the operations that call
331 * dsl_sync_task(ZFS_SPACE_CHECK_NONE), e.g. "zfs destroy". If these
332 * operations result in a net increase in the amount of space used,
333 * it is possible to run the pool completely out of space, causing it to
334 * be permanently read-only.
336 * See also the comments in zfs_space_check_t.
338 int spa_slop_shift
= 5;
341 * ==========================================================================
343 * ==========================================================================
346 spa_config_lock_init(spa_t
*spa
)
350 for (i
= 0; i
< SCL_LOCKS
; i
++) {
351 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
352 mutex_init(&scl
->scl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
353 cv_init(&scl
->scl_cv
, NULL
, CV_DEFAULT
, NULL
);
354 refcount_create_untracked(&scl
->scl_count
);
355 scl
->scl_writer
= NULL
;
356 scl
->scl_write_wanted
= 0;
361 spa_config_lock_destroy(spa_t
*spa
)
365 for (i
= 0; i
< SCL_LOCKS
; i
++) {
366 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
367 mutex_destroy(&scl
->scl_lock
);
368 cv_destroy(&scl
->scl_cv
);
369 refcount_destroy(&scl
->scl_count
);
370 ASSERT(scl
->scl_writer
== NULL
);
371 ASSERT(scl
->scl_write_wanted
== 0);
376 spa_config_tryenter(spa_t
*spa
, int locks
, void *tag
, krw_t rw
)
380 for (i
= 0; i
< SCL_LOCKS
; i
++) {
381 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
382 if (!(locks
& (1 << i
)))
384 mutex_enter(&scl
->scl_lock
);
385 if (rw
== RW_READER
) {
386 if (scl
->scl_writer
|| scl
->scl_write_wanted
) {
387 mutex_exit(&scl
->scl_lock
);
388 spa_config_exit(spa
, locks
^ (1 << i
), tag
);
392 ASSERT(scl
->scl_writer
!= curthread
);
393 if (!refcount_is_zero(&scl
->scl_count
)) {
394 mutex_exit(&scl
->scl_lock
);
395 spa_config_exit(spa
, locks
^ (1 << i
), tag
);
398 scl
->scl_writer
= curthread
;
400 (void) refcount_add(&scl
->scl_count
, tag
);
401 mutex_exit(&scl
->scl_lock
);
407 spa_config_enter(spa_t
*spa
, int locks
, void *tag
, krw_t rw
)
412 ASSERT3U(SCL_LOCKS
, <, sizeof (wlocks_held
) * NBBY
);
414 for (i
= 0; i
< SCL_LOCKS
; i
++) {
415 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
416 if (scl
->scl_writer
== curthread
)
417 wlocks_held
|= (1 << i
);
418 if (!(locks
& (1 << i
)))
420 mutex_enter(&scl
->scl_lock
);
421 if (rw
== RW_READER
) {
422 while (scl
->scl_writer
|| scl
->scl_write_wanted
) {
423 cv_wait(&scl
->scl_cv
, &scl
->scl_lock
);
426 ASSERT(scl
->scl_writer
!= curthread
);
427 while (!refcount_is_zero(&scl
->scl_count
)) {
428 scl
->scl_write_wanted
++;
429 cv_wait(&scl
->scl_cv
, &scl
->scl_lock
);
430 scl
->scl_write_wanted
--;
432 scl
->scl_writer
= curthread
;
434 (void) refcount_add(&scl
->scl_count
, tag
);
435 mutex_exit(&scl
->scl_lock
);
437 ASSERT(wlocks_held
<= locks
);
441 spa_config_exit(spa_t
*spa
, int locks
, void *tag
)
445 for (i
= SCL_LOCKS
- 1; i
>= 0; i
--) {
446 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
447 if (!(locks
& (1 << i
)))
449 mutex_enter(&scl
->scl_lock
);
450 ASSERT(!refcount_is_zero(&scl
->scl_count
));
451 if (refcount_remove(&scl
->scl_count
, tag
) == 0) {
452 ASSERT(scl
->scl_writer
== NULL
||
453 scl
->scl_writer
== curthread
);
454 scl
->scl_writer
= NULL
; /* OK in either case */
455 cv_broadcast(&scl
->scl_cv
);
457 mutex_exit(&scl
->scl_lock
);
462 spa_config_held(spa_t
*spa
, int locks
, krw_t rw
)
464 int i
, locks_held
= 0;
466 for (i
= 0; i
< SCL_LOCKS
; i
++) {
467 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
468 if (!(locks
& (1 << i
)))
470 if ((rw
== RW_READER
&& !refcount_is_zero(&scl
->scl_count
)) ||
471 (rw
== RW_WRITER
&& scl
->scl_writer
== curthread
))
472 locks_held
|= 1 << i
;
479 * ==========================================================================
480 * SPA namespace functions
481 * ==========================================================================
485 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
486 * Returns NULL if no matching spa_t is found.
489 spa_lookup(const char *name
)
491 static spa_t search
; /* spa_t is large; don't allocate on stack */
496 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
498 (void) strlcpy(search
.spa_name
, name
, sizeof (search
.spa_name
));
501 * If it's a full dataset name, figure out the pool name and
504 cp
= strpbrk(search
.spa_name
, "/@#");
508 spa
= avl_find(&spa_namespace_avl
, &search
, &where
);
514 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
515 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
516 * looking for potentially hung I/Os.
519 spa_deadman(void *arg
)
523 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
524 (gethrtime() - spa
->spa_sync_starttime
) / NANOSEC
,
525 ++spa
->spa_deadman_calls
);
526 if (zfs_deadman_enabled
)
527 vdev_deadman(spa
->spa_root_vdev
);
529 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_taskq
,
530 spa_deadman
, spa
, KM_SLEEP
, ddi_get_lbolt() +
531 NSEC_TO_TICK(spa
->spa_deadman_synctime
));
535 * Create an uninitialized spa_t with the given name. Requires
536 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
537 * exist by calling spa_lookup() first.
540 spa_add(const char *name
, nvlist_t
*config
, const char *altroot
)
543 spa_config_dirent_t
*dp
;
547 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
549 spa
= kmem_zalloc(sizeof (spa_t
), KM_SLEEP
);
551 mutex_init(&spa
->spa_async_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
552 mutex_init(&spa
->spa_errlist_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
553 mutex_init(&spa
->spa_errlog_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
554 mutex_init(&spa
->spa_evicting_os_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
555 mutex_init(&spa
->spa_history_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
556 mutex_init(&spa
->spa_proc_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
557 mutex_init(&spa
->spa_props_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
558 mutex_init(&spa
->spa_scrub_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
559 mutex_init(&spa
->spa_suspend_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
560 mutex_init(&spa
->spa_vdev_top_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
561 mutex_init(&spa
->spa_feat_stats_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
563 cv_init(&spa
->spa_async_cv
, NULL
, CV_DEFAULT
, NULL
);
564 cv_init(&spa
->spa_evicting_os_cv
, NULL
, CV_DEFAULT
, NULL
);
565 cv_init(&spa
->spa_proc_cv
, NULL
, CV_DEFAULT
, NULL
);
566 cv_init(&spa
->spa_scrub_io_cv
, NULL
, CV_DEFAULT
, NULL
);
567 cv_init(&spa
->spa_suspend_cv
, NULL
, CV_DEFAULT
, NULL
);
569 for (t
= 0; t
< TXG_SIZE
; t
++)
570 bplist_create(&spa
->spa_free_bplist
[t
]);
572 (void) strlcpy(spa
->spa_name
, name
, sizeof (spa
->spa_name
));
573 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
574 spa
->spa_freeze_txg
= UINT64_MAX
;
575 spa
->spa_final_txg
= UINT64_MAX
;
576 spa
->spa_load_max_txg
= UINT64_MAX
;
578 spa
->spa_proc_state
= SPA_PROC_NONE
;
580 spa
->spa_deadman_synctime
= MSEC2NSEC(zfs_deadman_synctime_ms
);
582 refcount_create(&spa
->spa_refcount
);
583 spa_config_lock_init(spa
);
586 avl_add(&spa_namespace_avl
, spa
);
589 * Set the alternate root, if there is one.
592 spa
->spa_root
= spa_strdup(altroot
);
595 * Every pool starts with the default cachefile
597 list_create(&spa
->spa_config_list
, sizeof (spa_config_dirent_t
),
598 offsetof(spa_config_dirent_t
, scd_link
));
600 dp
= kmem_zalloc(sizeof (spa_config_dirent_t
), KM_SLEEP
);
601 dp
->scd_path
= altroot
? NULL
: spa_strdup(spa_config_path
);
602 list_insert_head(&spa
->spa_config_list
, dp
);
604 VERIFY(nvlist_alloc(&spa
->spa_load_info
, NV_UNIQUE_NAME
,
607 if (config
!= NULL
) {
610 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_FEATURES_FOR_READ
,
612 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
,
616 VERIFY(nvlist_dup(config
, &spa
->spa_config
, 0) == 0);
619 if (spa
->spa_label_features
== NULL
) {
620 VERIFY(nvlist_alloc(&spa
->spa_label_features
, NV_UNIQUE_NAME
,
624 spa
->spa_debug
= ((zfs_flags
& ZFS_DEBUG_SPA
) != 0);
626 spa
->spa_min_ashift
= INT_MAX
;
627 spa
->spa_max_ashift
= 0;
630 * As a pool is being created, treat all features as disabled by
631 * setting SPA_FEATURE_DISABLED for all entries in the feature
634 for (i
= 0; i
< SPA_FEATURES
; i
++) {
635 spa
->spa_feat_refcount_cache
[i
] = SPA_FEATURE_DISABLED
;
642 * Removes a spa_t from the namespace, freeing up any memory used. Requires
643 * spa_namespace_lock. This is called only after the spa_t has been closed and
647 spa_remove(spa_t
*spa
)
649 spa_config_dirent_t
*dp
;
652 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
653 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
654 ASSERT3U(refcount_count(&spa
->spa_refcount
), ==, 0);
656 nvlist_free(spa
->spa_config_splitting
);
658 avl_remove(&spa_namespace_avl
, spa
);
659 cv_broadcast(&spa_namespace_cv
);
662 spa_strfree(spa
->spa_root
);
664 while ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
665 list_remove(&spa
->spa_config_list
, dp
);
666 if (dp
->scd_path
!= NULL
)
667 spa_strfree(dp
->scd_path
);
668 kmem_free(dp
, sizeof (spa_config_dirent_t
));
671 list_destroy(&spa
->spa_config_list
);
673 nvlist_free(spa
->spa_label_features
);
674 nvlist_free(spa
->spa_load_info
);
675 nvlist_free(spa
->spa_feat_stats
);
676 spa_config_set(spa
, NULL
);
678 refcount_destroy(&spa
->spa_refcount
);
680 spa_stats_destroy(spa
);
681 spa_config_lock_destroy(spa
);
683 for (t
= 0; t
< TXG_SIZE
; t
++)
684 bplist_destroy(&spa
->spa_free_bplist
[t
]);
686 cv_destroy(&spa
->spa_async_cv
);
687 cv_destroy(&spa
->spa_evicting_os_cv
);
688 cv_destroy(&spa
->spa_proc_cv
);
689 cv_destroy(&spa
->spa_scrub_io_cv
);
690 cv_destroy(&spa
->spa_suspend_cv
);
692 mutex_destroy(&spa
->spa_async_lock
);
693 mutex_destroy(&spa
->spa_errlist_lock
);
694 mutex_destroy(&spa
->spa_errlog_lock
);
695 mutex_destroy(&spa
->spa_evicting_os_lock
);
696 mutex_destroy(&spa
->spa_history_lock
);
697 mutex_destroy(&spa
->spa_proc_lock
);
698 mutex_destroy(&spa
->spa_props_lock
);
699 mutex_destroy(&spa
->spa_scrub_lock
);
700 mutex_destroy(&spa
->spa_suspend_lock
);
701 mutex_destroy(&spa
->spa_vdev_top_lock
);
702 mutex_destroy(&spa
->spa_feat_stats_lock
);
704 kmem_free(spa
, sizeof (spa_t
));
708 * Given a pool, return the next pool in the namespace, or NULL if there is
709 * none. If 'prev' is NULL, return the first pool.
712 spa_next(spa_t
*prev
)
714 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
717 return (AVL_NEXT(&spa_namespace_avl
, prev
));
719 return (avl_first(&spa_namespace_avl
));
723 * ==========================================================================
724 * SPA refcount functions
725 * ==========================================================================
729 * Add a reference to the given spa_t. Must have at least one reference, or
730 * have the namespace lock held.
733 spa_open_ref(spa_t
*spa
, void *tag
)
735 ASSERT(refcount_count(&spa
->spa_refcount
) >= spa
->spa_minref
||
736 MUTEX_HELD(&spa_namespace_lock
));
737 (void) refcount_add(&spa
->spa_refcount
, tag
);
741 * Remove a reference to the given spa_t. Must have at least one reference, or
742 * have the namespace lock held.
745 spa_close(spa_t
*spa
, void *tag
)
747 ASSERT(refcount_count(&spa
->spa_refcount
) > spa
->spa_minref
||
748 MUTEX_HELD(&spa_namespace_lock
));
749 (void) refcount_remove(&spa
->spa_refcount
, tag
);
753 * Remove a reference to the given spa_t held by a dsl dir that is
754 * being asynchronously released. Async releases occur from a taskq
755 * performing eviction of dsl datasets and dirs. The namespace lock
756 * isn't held and the hold by the object being evicted may contribute to
757 * spa_minref (e.g. dataset or directory released during pool export),
758 * so the asserts in spa_close() do not apply.
761 spa_async_close(spa_t
*spa
, void *tag
)
763 (void) refcount_remove(&spa
->spa_refcount
, tag
);
767 * Check to see if the spa refcount is zero. Must be called with
768 * spa_namespace_lock held. We really compare against spa_minref, which is the
769 * number of references acquired when opening a pool
772 spa_refcount_zero(spa_t
*spa
)
774 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
776 return (refcount_count(&spa
->spa_refcount
) == spa
->spa_minref
);
780 * ==========================================================================
781 * SPA spare and l2cache tracking
782 * ==========================================================================
786 * Hot spares and cache devices are tracked using the same code below,
787 * for 'auxiliary' devices.
790 typedef struct spa_aux
{
798 spa_aux_compare(const void *a
, const void *b
)
800 const spa_aux_t
*sa
= a
;
801 const spa_aux_t
*sb
= b
;
803 if (sa
->aux_guid
< sb
->aux_guid
)
805 else if (sa
->aux_guid
> sb
->aux_guid
)
812 spa_aux_add(vdev_t
*vd
, avl_tree_t
*avl
)
818 search
.aux_guid
= vd
->vdev_guid
;
819 if ((aux
= avl_find(avl
, &search
, &where
)) != NULL
) {
822 aux
= kmem_zalloc(sizeof (spa_aux_t
), KM_SLEEP
);
823 aux
->aux_guid
= vd
->vdev_guid
;
825 avl_insert(avl
, aux
, where
);
830 spa_aux_remove(vdev_t
*vd
, avl_tree_t
*avl
)
836 search
.aux_guid
= vd
->vdev_guid
;
837 aux
= avl_find(avl
, &search
, &where
);
841 if (--aux
->aux_count
== 0) {
842 avl_remove(avl
, aux
);
843 kmem_free(aux
, sizeof (spa_aux_t
));
844 } else if (aux
->aux_pool
== spa_guid(vd
->vdev_spa
)) {
845 aux
->aux_pool
= 0ULL;
850 spa_aux_exists(uint64_t guid
, uint64_t *pool
, int *refcnt
, avl_tree_t
*avl
)
852 spa_aux_t search
, *found
;
854 search
.aux_guid
= guid
;
855 found
= avl_find(avl
, &search
, NULL
);
859 *pool
= found
->aux_pool
;
866 *refcnt
= found
->aux_count
;
871 return (found
!= NULL
);
875 spa_aux_activate(vdev_t
*vd
, avl_tree_t
*avl
)
877 spa_aux_t search
, *found
;
880 search
.aux_guid
= vd
->vdev_guid
;
881 found
= avl_find(avl
, &search
, &where
);
882 ASSERT(found
!= NULL
);
883 ASSERT(found
->aux_pool
== 0ULL);
885 found
->aux_pool
= spa_guid(vd
->vdev_spa
);
889 * Spares are tracked globally due to the following constraints:
891 * - A spare may be part of multiple pools.
892 * - A spare may be added to a pool even if it's actively in use within
894 * - A spare in use in any pool can only be the source of a replacement if
895 * the target is a spare in the same pool.
897 * We keep track of all spares on the system through the use of a reference
898 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
899 * spare, then we bump the reference count in the AVL tree. In addition, we set
900 * the 'vdev_isspare' member to indicate that the device is a spare (active or
901 * inactive). When a spare is made active (used to replace a device in the
902 * pool), we also keep track of which pool its been made a part of.
904 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
905 * called under the spa_namespace lock as part of vdev reconfiguration. The
906 * separate spare lock exists for the status query path, which does not need to
907 * be completely consistent with respect to other vdev configuration changes.
911 spa_spare_compare(const void *a
, const void *b
)
913 return (spa_aux_compare(a
, b
));
917 spa_spare_add(vdev_t
*vd
)
919 mutex_enter(&spa_spare_lock
);
920 ASSERT(!vd
->vdev_isspare
);
921 spa_aux_add(vd
, &spa_spare_avl
);
922 vd
->vdev_isspare
= B_TRUE
;
923 mutex_exit(&spa_spare_lock
);
927 spa_spare_remove(vdev_t
*vd
)
929 mutex_enter(&spa_spare_lock
);
930 ASSERT(vd
->vdev_isspare
);
931 spa_aux_remove(vd
, &spa_spare_avl
);
932 vd
->vdev_isspare
= B_FALSE
;
933 mutex_exit(&spa_spare_lock
);
937 spa_spare_exists(uint64_t guid
, uint64_t *pool
, int *refcnt
)
941 mutex_enter(&spa_spare_lock
);
942 found
= spa_aux_exists(guid
, pool
, refcnt
, &spa_spare_avl
);
943 mutex_exit(&spa_spare_lock
);
949 spa_spare_activate(vdev_t
*vd
)
951 mutex_enter(&spa_spare_lock
);
952 ASSERT(vd
->vdev_isspare
);
953 spa_aux_activate(vd
, &spa_spare_avl
);
954 mutex_exit(&spa_spare_lock
);
958 * Level 2 ARC devices are tracked globally for the same reasons as spares.
959 * Cache devices currently only support one pool per cache device, and so
960 * for these devices the aux reference count is currently unused beyond 1.
964 spa_l2cache_compare(const void *a
, const void *b
)
966 return (spa_aux_compare(a
, b
));
970 spa_l2cache_add(vdev_t
*vd
)
972 mutex_enter(&spa_l2cache_lock
);
973 ASSERT(!vd
->vdev_isl2cache
);
974 spa_aux_add(vd
, &spa_l2cache_avl
);
975 vd
->vdev_isl2cache
= B_TRUE
;
976 mutex_exit(&spa_l2cache_lock
);
980 spa_l2cache_remove(vdev_t
*vd
)
982 mutex_enter(&spa_l2cache_lock
);
983 ASSERT(vd
->vdev_isl2cache
);
984 spa_aux_remove(vd
, &spa_l2cache_avl
);
985 vd
->vdev_isl2cache
= B_FALSE
;
986 mutex_exit(&spa_l2cache_lock
);
990 spa_l2cache_exists(uint64_t guid
, uint64_t *pool
)
994 mutex_enter(&spa_l2cache_lock
);
995 found
= spa_aux_exists(guid
, pool
, NULL
, &spa_l2cache_avl
);
996 mutex_exit(&spa_l2cache_lock
);
1002 spa_l2cache_activate(vdev_t
*vd
)
1004 mutex_enter(&spa_l2cache_lock
);
1005 ASSERT(vd
->vdev_isl2cache
);
1006 spa_aux_activate(vd
, &spa_l2cache_avl
);
1007 mutex_exit(&spa_l2cache_lock
);
1011 * ==========================================================================
1013 * ==========================================================================
1017 * Lock the given spa_t for the purpose of adding or removing a vdev.
1018 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1019 * It returns the next transaction group for the spa_t.
1022 spa_vdev_enter(spa_t
*spa
)
1024 mutex_enter(&spa
->spa_vdev_top_lock
);
1025 mutex_enter(&spa_namespace_lock
);
1026 return (spa_vdev_config_enter(spa
));
1030 * Internal implementation for spa_vdev_enter(). Used when a vdev
1031 * operation requires multiple syncs (i.e. removing a device) while
1032 * keeping the spa_namespace_lock held.
1035 spa_vdev_config_enter(spa_t
*spa
)
1037 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1039 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
1041 return (spa_last_synced_txg(spa
) + 1);
1045 * Used in combination with spa_vdev_config_enter() to allow the syncing
1046 * of multiple transactions without releasing the spa_namespace_lock.
1049 spa_vdev_config_exit(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int error
, char *tag
)
1051 int config_changed
= B_FALSE
;
1053 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1054 ASSERT(txg
> spa_last_synced_txg(spa
));
1056 spa
->spa_pending_vdev
= NULL
;
1059 * Reassess the DTLs.
1061 vdev_dtl_reassess(spa
->spa_root_vdev
, 0, 0, B_FALSE
);
1063 if (error
== 0 && !list_is_empty(&spa
->spa_config_dirty_list
)) {
1064 config_changed
= B_TRUE
;
1065 spa
->spa_config_generation
++;
1069 * Verify the metaslab classes.
1071 ASSERT(metaslab_class_validate(spa_normal_class(spa
)) == 0);
1072 ASSERT(metaslab_class_validate(spa_log_class(spa
)) == 0);
1074 spa_config_exit(spa
, SCL_ALL
, spa
);
1077 * Panic the system if the specified tag requires it. This
1078 * is useful for ensuring that configurations are updated
1081 if (zio_injection_enabled
)
1082 zio_handle_panic_injection(spa
, tag
, 0);
1085 * Note: this txg_wait_synced() is important because it ensures
1086 * that there won't be more than one config change per txg.
1087 * This allows us to use the txg as the generation number.
1090 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
1093 ASSERT(!vd
->vdev_detached
|| vd
->vdev_dtl_sm
== NULL
);
1094 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
1096 spa_config_exit(spa
, SCL_ALL
, spa
);
1100 * If the config changed, update the config cache.
1103 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
1107 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1108 * locking of spa_vdev_enter(), we also want make sure the transactions have
1109 * synced to disk, and then update the global configuration cache with the new
1113 spa_vdev_exit(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int error
)
1115 spa_vdev_config_exit(spa
, vd
, txg
, error
, FTAG
);
1116 mutex_exit(&spa_namespace_lock
);
1117 mutex_exit(&spa
->spa_vdev_top_lock
);
1123 * Lock the given spa_t for the purpose of changing vdev state.
1126 spa_vdev_state_enter(spa_t
*spa
, int oplocks
)
1128 int locks
= SCL_STATE_ALL
| oplocks
;
1131 * Root pools may need to read of the underlying devfs filesystem
1132 * when opening up a vdev. Unfortunately if we're holding the
1133 * SCL_ZIO lock it will result in a deadlock when we try to issue
1134 * the read from the root filesystem. Instead we "prefetch"
1135 * the associated vnodes that we need prior to opening the
1136 * underlying devices and cache them so that we can prevent
1137 * any I/O when we are doing the actual open.
1139 if (spa_is_root(spa
)) {
1140 int low
= locks
& ~(SCL_ZIO
- 1);
1141 int high
= locks
& ~low
;
1143 spa_config_enter(spa
, high
, spa
, RW_WRITER
);
1144 vdev_hold(spa
->spa_root_vdev
);
1145 spa_config_enter(spa
, low
, spa
, RW_WRITER
);
1147 spa_config_enter(spa
, locks
, spa
, RW_WRITER
);
1149 spa
->spa_vdev_locks
= locks
;
1153 spa_vdev_state_exit(spa_t
*spa
, vdev_t
*vd
, int error
)
1155 boolean_t config_changed
= B_FALSE
;
1157 if (vd
!= NULL
|| error
== 0)
1158 vdev_dtl_reassess(vd
? vd
->vdev_top
: spa
->spa_root_vdev
,
1162 vdev_state_dirty(vd
->vdev_top
);
1163 config_changed
= B_TRUE
;
1164 spa
->spa_config_generation
++;
1167 if (spa_is_root(spa
))
1168 vdev_rele(spa
->spa_root_vdev
);
1170 ASSERT3U(spa
->spa_vdev_locks
, >=, SCL_STATE_ALL
);
1171 spa_config_exit(spa
, spa
->spa_vdev_locks
, spa
);
1174 * If anything changed, wait for it to sync. This ensures that,
1175 * from the system administrator's perspective, zpool(1M) commands
1176 * are synchronous. This is important for things like zpool offline:
1177 * when the command completes, you expect no further I/O from ZFS.
1180 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1183 * If the config changed, update the config cache.
1185 if (config_changed
) {
1186 mutex_enter(&spa_namespace_lock
);
1187 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
1188 mutex_exit(&spa_namespace_lock
);
1195 * ==========================================================================
1196 * Miscellaneous functions
1197 * ==========================================================================
1201 spa_activate_mos_feature(spa_t
*spa
, const char *feature
, dmu_tx_t
*tx
)
1203 if (!nvlist_exists(spa
->spa_label_features
, feature
)) {
1204 fnvlist_add_boolean(spa
->spa_label_features
, feature
);
1206 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1207 * dirty the vdev config because lock SCL_CONFIG is not held.
1208 * Thankfully, in this case we don't need to dirty the config
1209 * because it will be written out anyway when we finish
1210 * creating the pool.
1212 if (tx
->tx_txg
!= TXG_INITIAL
)
1213 vdev_config_dirty(spa
->spa_root_vdev
);
1218 spa_deactivate_mos_feature(spa_t
*spa
, const char *feature
)
1220 if (nvlist_remove_all(spa
->spa_label_features
, feature
) == 0)
1221 vdev_config_dirty(spa
->spa_root_vdev
);
1228 spa_rename(const char *name
, const char *newname
)
1234 * Lookup the spa_t and grab the config lock for writing. We need to
1235 * actually open the pool so that we can sync out the necessary labels.
1236 * It's OK to call spa_open() with the namespace lock held because we
1237 * allow recursive calls for other reasons.
1239 mutex_enter(&spa_namespace_lock
);
1240 if ((err
= spa_open(name
, &spa
, FTAG
)) != 0) {
1241 mutex_exit(&spa_namespace_lock
);
1245 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1247 avl_remove(&spa_namespace_avl
, spa
);
1248 (void) strlcpy(spa
->spa_name
, newname
, sizeof (spa
->spa_name
));
1249 avl_add(&spa_namespace_avl
, spa
);
1252 * Sync all labels to disk with the new names by marking the root vdev
1253 * dirty and waiting for it to sync. It will pick up the new pool name
1256 vdev_config_dirty(spa
->spa_root_vdev
);
1258 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1260 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1263 * Sync the updated config cache.
1265 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
1267 spa_close(spa
, FTAG
);
1269 mutex_exit(&spa_namespace_lock
);
1275 * Return the spa_t associated with given pool_guid, if it exists. If
1276 * device_guid is non-zero, determine whether the pool exists *and* contains
1277 * a device with the specified device_guid.
1280 spa_by_guid(uint64_t pool_guid
, uint64_t device_guid
)
1283 avl_tree_t
*t
= &spa_namespace_avl
;
1285 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1287 for (spa
= avl_first(t
); spa
!= NULL
; spa
= AVL_NEXT(t
, spa
)) {
1288 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1290 if (spa
->spa_root_vdev
== NULL
)
1292 if (spa_guid(spa
) == pool_guid
) {
1293 if (device_guid
== 0)
1296 if (vdev_lookup_by_guid(spa
->spa_root_vdev
,
1297 device_guid
) != NULL
)
1301 * Check any devices we may be in the process of adding.
1303 if (spa
->spa_pending_vdev
) {
1304 if (vdev_lookup_by_guid(spa
->spa_pending_vdev
,
1305 device_guid
) != NULL
)
1315 * Determine whether a pool with the given pool_guid exists.
1318 spa_guid_exists(uint64_t pool_guid
, uint64_t device_guid
)
1320 return (spa_by_guid(pool_guid
, device_guid
) != NULL
);
1324 spa_strdup(const char *s
)
1330 new = kmem_alloc(len
+ 1, KM_SLEEP
);
1338 spa_strfree(char *s
)
1340 kmem_free(s
, strlen(s
) + 1);
1344 spa_get_random(uint64_t range
)
1350 (void) random_get_pseudo_bytes((void *)&r
, sizeof (uint64_t));
1356 spa_generate_guid(spa_t
*spa
)
1358 uint64_t guid
= spa_get_random(-1ULL);
1361 while (guid
== 0 || spa_guid_exists(spa_guid(spa
), guid
))
1362 guid
= spa_get_random(-1ULL);
1364 while (guid
== 0 || spa_guid_exists(guid
, 0))
1365 guid
= spa_get_random(-1ULL);
1372 snprintf_blkptr(char *buf
, size_t buflen
, const blkptr_t
*bp
)
1375 char *checksum
= NULL
;
1376 char *compress
= NULL
;
1379 if (BP_GET_TYPE(bp
) & DMU_OT_NEWTYPE
) {
1380 dmu_object_byteswap_t bswap
=
1381 DMU_OT_BYTESWAP(BP_GET_TYPE(bp
));
1382 (void) snprintf(type
, sizeof (type
), "bswap %s %s",
1383 DMU_OT_IS_METADATA(BP_GET_TYPE(bp
)) ?
1384 "metadata" : "data",
1385 dmu_ot_byteswap
[bswap
].ob_name
);
1387 (void) strlcpy(type
, dmu_ot
[BP_GET_TYPE(bp
)].ot_name
,
1390 if (!BP_IS_EMBEDDED(bp
)) {
1392 zio_checksum_table
[BP_GET_CHECKSUM(bp
)].ci_name
;
1394 compress
= zio_compress_table
[BP_GET_COMPRESS(bp
)].ci_name
;
1397 SNPRINTF_BLKPTR(snprintf
, ' ', buf
, buflen
, bp
, type
, checksum
,
1402 spa_freeze(spa_t
*spa
)
1404 uint64_t freeze_txg
= 0;
1406 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1407 if (spa
->spa_freeze_txg
== UINT64_MAX
) {
1408 freeze_txg
= spa_last_synced_txg(spa
) + TXG_SIZE
;
1409 spa
->spa_freeze_txg
= freeze_txg
;
1411 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1412 if (freeze_txg
!= 0)
1413 txg_wait_synced(spa_get_dsl(spa
), freeze_txg
);
1417 zfs_panic_recover(const char *fmt
, ...)
1422 vcmn_err(zfs_recover
? CE_WARN
: CE_PANIC
, fmt
, adx
);
1427 * This is a stripped-down version of strtoull, suitable only for converting
1428 * lowercase hexadecimal numbers that don't overflow.
1431 strtonum(const char *str
, char **nptr
)
1437 while ((c
= *str
) != '\0') {
1438 if (c
>= '0' && c
<= '9')
1440 else if (c
>= 'a' && c
<= 'f')
1441 digit
= 10 + c
- 'a';
1452 *nptr
= (char *)str
;
1458 * ==========================================================================
1459 * Accessor functions
1460 * ==========================================================================
1464 spa_shutting_down(spa_t
*spa
)
1466 return (spa
->spa_async_suspended
);
1470 spa_get_dsl(spa_t
*spa
)
1472 return (spa
->spa_dsl_pool
);
1476 spa_is_initializing(spa_t
*spa
)
1478 return (spa
->spa_is_initializing
);
1482 spa_get_rootblkptr(spa_t
*spa
)
1484 return (&spa
->spa_ubsync
.ub_rootbp
);
1488 spa_set_rootblkptr(spa_t
*spa
, const blkptr_t
*bp
)
1490 spa
->spa_uberblock
.ub_rootbp
= *bp
;
1494 spa_altroot(spa_t
*spa
, char *buf
, size_t buflen
)
1496 if (spa
->spa_root
== NULL
)
1499 (void) strncpy(buf
, spa
->spa_root
, buflen
);
1503 spa_sync_pass(spa_t
*spa
)
1505 return (spa
->spa_sync_pass
);
1509 spa_name(spa_t
*spa
)
1511 return (spa
->spa_name
);
1515 spa_guid(spa_t
*spa
)
1517 dsl_pool_t
*dp
= spa_get_dsl(spa
);
1521 * If we fail to parse the config during spa_load(), we can go through
1522 * the error path (which posts an ereport) and end up here with no root
1523 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1526 if (spa
->spa_root_vdev
== NULL
)
1527 return (spa
->spa_config_guid
);
1529 guid
= spa
->spa_last_synced_guid
!= 0 ?
1530 spa
->spa_last_synced_guid
: spa
->spa_root_vdev
->vdev_guid
;
1533 * Return the most recently synced out guid unless we're
1534 * in syncing context.
1536 if (dp
&& dsl_pool_sync_context(dp
))
1537 return (spa
->spa_root_vdev
->vdev_guid
);
1543 spa_load_guid(spa_t
*spa
)
1546 * This is a GUID that exists solely as a reference for the
1547 * purposes of the arc. It is generated at load time, and
1548 * is never written to persistent storage.
1550 return (spa
->spa_load_guid
);
1554 spa_last_synced_txg(spa_t
*spa
)
1556 return (spa
->spa_ubsync
.ub_txg
);
1560 spa_first_txg(spa_t
*spa
)
1562 return (spa
->spa_first_txg
);
1566 spa_syncing_txg(spa_t
*spa
)
1568 return (spa
->spa_syncing_txg
);
1572 spa_state(spa_t
*spa
)
1574 return (spa
->spa_state
);
1578 spa_load_state(spa_t
*spa
)
1580 return (spa
->spa_load_state
);
1584 spa_freeze_txg(spa_t
*spa
)
1586 return (spa
->spa_freeze_txg
);
1591 spa_get_asize(spa_t
*spa
, uint64_t lsize
)
1593 return (lsize
* spa_asize_inflation
);
1597 * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%),
1600 * See the comment above spa_slop_shift for details.
1603 spa_get_slop_space(spa_t
*spa
) {
1604 uint64_t space
= spa_get_dspace(spa
);
1605 return (MAX(space
>> spa_slop_shift
, SPA_MINDEVSIZE
>> 1));
1609 spa_get_dspace(spa_t
*spa
)
1611 return (spa
->spa_dspace
);
1615 spa_update_dspace(spa_t
*spa
)
1617 spa
->spa_dspace
= metaslab_class_get_dspace(spa_normal_class(spa
)) +
1618 ddt_get_dedup_dspace(spa
);
1622 * Return the failure mode that has been set to this pool. The default
1623 * behavior will be to block all I/Os when a complete failure occurs.
1626 spa_get_failmode(spa_t
*spa
)
1628 return (spa
->spa_failmode
);
1632 spa_suspended(spa_t
*spa
)
1634 return (spa
->spa_suspended
);
1638 spa_version(spa_t
*spa
)
1640 return (spa
->spa_ubsync
.ub_version
);
1644 spa_deflate(spa_t
*spa
)
1646 return (spa
->spa_deflate
);
1650 spa_normal_class(spa_t
*spa
)
1652 return (spa
->spa_normal_class
);
1656 spa_log_class(spa_t
*spa
)
1658 return (spa
->spa_log_class
);
1662 spa_evicting_os_register(spa_t
*spa
, objset_t
*os
)
1664 mutex_enter(&spa
->spa_evicting_os_lock
);
1665 list_insert_head(&spa
->spa_evicting_os_list
, os
);
1666 mutex_exit(&spa
->spa_evicting_os_lock
);
1670 spa_evicting_os_deregister(spa_t
*spa
, objset_t
*os
)
1672 mutex_enter(&spa
->spa_evicting_os_lock
);
1673 list_remove(&spa
->spa_evicting_os_list
, os
);
1674 cv_broadcast(&spa
->spa_evicting_os_cv
);
1675 mutex_exit(&spa
->spa_evicting_os_lock
);
1679 spa_evicting_os_wait(spa_t
*spa
)
1681 mutex_enter(&spa
->spa_evicting_os_lock
);
1682 while (!list_is_empty(&spa
->spa_evicting_os_list
))
1683 cv_wait(&spa
->spa_evicting_os_cv
, &spa
->spa_evicting_os_lock
);
1684 mutex_exit(&spa
->spa_evicting_os_lock
);
1686 dmu_buf_user_evict_wait();
1690 spa_max_replication(spa_t
*spa
)
1693 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1694 * handle BPs with more than one DVA allocated. Set our max
1695 * replication level accordingly.
1697 if (spa_version(spa
) < SPA_VERSION_DITTO_BLOCKS
)
1699 return (MIN(SPA_DVAS_PER_BP
, spa_max_replication_override
));
1703 spa_prev_software_version(spa_t
*spa
)
1705 return (spa
->spa_prev_software_version
);
1709 spa_deadman_synctime(spa_t
*spa
)
1711 return (spa
->spa_deadman_synctime
);
1715 dva_get_dsize_sync(spa_t
*spa
, const dva_t
*dva
)
1717 uint64_t asize
= DVA_GET_ASIZE(dva
);
1718 uint64_t dsize
= asize
;
1720 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1722 if (asize
!= 0 && spa
->spa_deflate
) {
1723 vdev_t
*vd
= vdev_lookup_top(spa
, DVA_GET_VDEV(dva
));
1725 dsize
= (asize
>> SPA_MINBLOCKSHIFT
) *
1726 vd
->vdev_deflate_ratio
;
1733 bp_get_dsize_sync(spa_t
*spa
, const blkptr_t
*bp
)
1738 for (d
= 0; d
< BP_GET_NDVAS(bp
); d
++)
1739 dsize
+= dva_get_dsize_sync(spa
, &bp
->blk_dva
[d
]);
1745 bp_get_dsize(spa_t
*spa
, const blkptr_t
*bp
)
1750 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
1752 for (d
= 0; d
< BP_GET_NDVAS(bp
); d
++)
1753 dsize
+= dva_get_dsize_sync(spa
, &bp
->blk_dva
[d
]);
1755 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
1761 * ==========================================================================
1762 * Initialization and Termination
1763 * ==========================================================================
1767 spa_name_compare(const void *a1
, const void *a2
)
1769 const spa_t
*s1
= a1
;
1770 const spa_t
*s2
= a2
;
1773 s
= strcmp(s1
->spa_name
, s2
->spa_name
);
1790 mutex_init(&spa_namespace_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1791 mutex_init(&spa_spare_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1792 mutex_init(&spa_l2cache_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1793 cv_init(&spa_namespace_cv
, NULL
, CV_DEFAULT
, NULL
);
1795 avl_create(&spa_namespace_avl
, spa_name_compare
, sizeof (spa_t
),
1796 offsetof(spa_t
, spa_avl
));
1798 avl_create(&spa_spare_avl
, spa_spare_compare
, sizeof (spa_aux_t
),
1799 offsetof(spa_aux_t
, aux_avl
));
1801 avl_create(&spa_l2cache_avl
, spa_l2cache_compare
, sizeof (spa_aux_t
),
1802 offsetof(spa_aux_t
, aux_avl
));
1804 spa_mode_global
= mode
;
1807 if (spa_mode_global
!= FREAD
&& dprintf_find_string("watch")) {
1808 struct sigaction sa
;
1810 sa
.sa_flags
= SA_SIGINFO
;
1811 sigemptyset(&sa
.sa_mask
);
1812 sa
.sa_sigaction
= arc_buf_sigsegv
;
1814 if (sigaction(SIGSEGV
, &sa
, NULL
) == -1) {
1815 perror("could not enable watchpoints: "
1816 "sigaction(SIGSEGV, ...) = ");
1831 vdev_cache_stat_init();
1834 zpool_feature_init();
1846 vdev_cache_stat_fini();
1856 avl_destroy(&spa_namespace_avl
);
1857 avl_destroy(&spa_spare_avl
);
1858 avl_destroy(&spa_l2cache_avl
);
1860 cv_destroy(&spa_namespace_cv
);
1861 mutex_destroy(&spa_namespace_lock
);
1862 mutex_destroy(&spa_spare_lock
);
1863 mutex_destroy(&spa_l2cache_lock
);
1867 * Return whether this pool has slogs. No locking needed.
1868 * It's not a problem if the wrong answer is returned as it's only for
1869 * performance and not correctness
1872 spa_has_slogs(spa_t
*spa
)
1874 return (spa
->spa_log_class
->mc_rotor
!= NULL
);
1878 spa_get_log_state(spa_t
*spa
)
1880 return (spa
->spa_log_state
);
1884 spa_set_log_state(spa_t
*spa
, spa_log_state_t state
)
1886 spa
->spa_log_state
= state
;
1890 spa_is_root(spa_t
*spa
)
1892 return (spa
->spa_is_root
);
1896 spa_writeable(spa_t
*spa
)
1898 return (!!(spa
->spa_mode
& FWRITE
));
1902 * Returns true if there is a pending sync task in any of the current
1903 * syncing txg, the current quiescing txg, or the current open txg.
1906 spa_has_pending_synctask(spa_t
*spa
)
1908 return (!txg_all_lists_empty(&spa
->spa_dsl_pool
->dp_sync_tasks
));
1912 spa_mode(spa_t
*spa
)
1914 return (spa
->spa_mode
);
1918 spa_bootfs(spa_t
*spa
)
1920 return (spa
->spa_bootfs
);
1924 spa_delegation(spa_t
*spa
)
1926 return (spa
->spa_delegation
);
1930 spa_meta_objset(spa_t
*spa
)
1932 return (spa
->spa_meta_objset
);
1936 spa_dedup_checksum(spa_t
*spa
)
1938 return (spa
->spa_dedup_checksum
);
1942 * Reset pool scan stat per scan pass (or reboot).
1945 spa_scan_stat_init(spa_t
*spa
)
1947 /* data not stored on disk */
1948 spa
->spa_scan_pass_start
= gethrestime_sec();
1949 spa
->spa_scan_pass_exam
= 0;
1950 vdev_scan_stat_init(spa
->spa_root_vdev
);
1954 * Get scan stats for zpool status reports
1957 spa_scan_get_stats(spa_t
*spa
, pool_scan_stat_t
*ps
)
1959 dsl_scan_t
*scn
= spa
->spa_dsl_pool
? spa
->spa_dsl_pool
->dp_scan
: NULL
;
1961 if (scn
== NULL
|| scn
->scn_phys
.scn_func
== POOL_SCAN_NONE
)
1962 return (SET_ERROR(ENOENT
));
1963 bzero(ps
, sizeof (pool_scan_stat_t
));
1965 /* data stored on disk */
1966 ps
->pss_func
= scn
->scn_phys
.scn_func
;
1967 ps
->pss_start_time
= scn
->scn_phys
.scn_start_time
;
1968 ps
->pss_end_time
= scn
->scn_phys
.scn_end_time
;
1969 ps
->pss_to_examine
= scn
->scn_phys
.scn_to_examine
;
1970 ps
->pss_examined
= scn
->scn_phys
.scn_examined
;
1971 ps
->pss_to_process
= scn
->scn_phys
.scn_to_process
;
1972 ps
->pss_processed
= scn
->scn_phys
.scn_processed
;
1973 ps
->pss_errors
= scn
->scn_phys
.scn_errors
;
1974 ps
->pss_state
= scn
->scn_phys
.scn_state
;
1976 /* data not stored on disk */
1977 ps
->pss_pass_start
= spa
->spa_scan_pass_start
;
1978 ps
->pss_pass_exam
= spa
->spa_scan_pass_exam
;
1984 spa_debug_enabled(spa_t
*spa
)
1986 return (spa
->spa_debug
);
1990 spa_maxblocksize(spa_t
*spa
)
1992 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_BLOCKS
))
1993 return (SPA_MAXBLOCKSIZE
);
1995 return (SPA_OLD_MAXBLOCKSIZE
);
1998 #if defined(_KERNEL) && defined(HAVE_SPL)
1999 /* Namespace manipulation */
2000 EXPORT_SYMBOL(spa_lookup
);
2001 EXPORT_SYMBOL(spa_add
);
2002 EXPORT_SYMBOL(spa_remove
);
2003 EXPORT_SYMBOL(spa_next
);
2005 /* Refcount functions */
2006 EXPORT_SYMBOL(spa_open_ref
);
2007 EXPORT_SYMBOL(spa_close
);
2008 EXPORT_SYMBOL(spa_refcount_zero
);
2010 /* Pool configuration lock */
2011 EXPORT_SYMBOL(spa_config_tryenter
);
2012 EXPORT_SYMBOL(spa_config_enter
);
2013 EXPORT_SYMBOL(spa_config_exit
);
2014 EXPORT_SYMBOL(spa_config_held
);
2016 /* Pool vdev add/remove lock */
2017 EXPORT_SYMBOL(spa_vdev_enter
);
2018 EXPORT_SYMBOL(spa_vdev_exit
);
2020 /* Pool vdev state change lock */
2021 EXPORT_SYMBOL(spa_vdev_state_enter
);
2022 EXPORT_SYMBOL(spa_vdev_state_exit
);
2024 /* Accessor functions */
2025 EXPORT_SYMBOL(spa_shutting_down
);
2026 EXPORT_SYMBOL(spa_get_dsl
);
2027 EXPORT_SYMBOL(spa_get_rootblkptr
);
2028 EXPORT_SYMBOL(spa_set_rootblkptr
);
2029 EXPORT_SYMBOL(spa_altroot
);
2030 EXPORT_SYMBOL(spa_sync_pass
);
2031 EXPORT_SYMBOL(spa_name
);
2032 EXPORT_SYMBOL(spa_guid
);
2033 EXPORT_SYMBOL(spa_last_synced_txg
);
2034 EXPORT_SYMBOL(spa_first_txg
);
2035 EXPORT_SYMBOL(spa_syncing_txg
);
2036 EXPORT_SYMBOL(spa_version
);
2037 EXPORT_SYMBOL(spa_state
);
2038 EXPORT_SYMBOL(spa_load_state
);
2039 EXPORT_SYMBOL(spa_freeze_txg
);
2040 EXPORT_SYMBOL(spa_get_asize
);
2041 EXPORT_SYMBOL(spa_get_dspace
);
2042 EXPORT_SYMBOL(spa_update_dspace
);
2043 EXPORT_SYMBOL(spa_deflate
);
2044 EXPORT_SYMBOL(spa_normal_class
);
2045 EXPORT_SYMBOL(spa_log_class
);
2046 EXPORT_SYMBOL(spa_max_replication
);
2047 EXPORT_SYMBOL(spa_prev_software_version
);
2048 EXPORT_SYMBOL(spa_get_failmode
);
2049 EXPORT_SYMBOL(spa_suspended
);
2050 EXPORT_SYMBOL(spa_bootfs
);
2051 EXPORT_SYMBOL(spa_delegation
);
2052 EXPORT_SYMBOL(spa_meta_objset
);
2053 EXPORT_SYMBOL(spa_maxblocksize
);
2055 /* Miscellaneous support routines */
2056 EXPORT_SYMBOL(spa_rename
);
2057 EXPORT_SYMBOL(spa_guid_exists
);
2058 EXPORT_SYMBOL(spa_strdup
);
2059 EXPORT_SYMBOL(spa_strfree
);
2060 EXPORT_SYMBOL(spa_get_random
);
2061 EXPORT_SYMBOL(spa_generate_guid
);
2062 EXPORT_SYMBOL(snprintf_blkptr
);
2063 EXPORT_SYMBOL(spa_freeze
);
2064 EXPORT_SYMBOL(spa_upgrade
);
2065 EXPORT_SYMBOL(spa_evict_all
);
2066 EXPORT_SYMBOL(spa_lookup_by_guid
);
2067 EXPORT_SYMBOL(spa_has_spare
);
2068 EXPORT_SYMBOL(dva_get_dsize_sync
);
2069 EXPORT_SYMBOL(bp_get_dsize_sync
);
2070 EXPORT_SYMBOL(bp_get_dsize
);
2071 EXPORT_SYMBOL(spa_has_slogs
);
2072 EXPORT_SYMBOL(spa_is_root
);
2073 EXPORT_SYMBOL(spa_writeable
);
2074 EXPORT_SYMBOL(spa_mode
);
2076 EXPORT_SYMBOL(spa_namespace_lock
);
2078 module_param(zfs_flags
, uint
, 0644);
2079 MODULE_PARM_DESC(zfs_flags
, "Set additional debugging flags");
2081 module_param(zfs_recover
, int, 0644);
2082 MODULE_PARM_DESC(zfs_recover
, "Set to attempt to recover from fatal errors");
2084 module_param(zfs_free_leak_on_eio
, int, 0644);
2085 MODULE_PARM_DESC(zfs_free_leak_on_eio
,
2086 "Set to ignore IO errors during free and permanently leak the space");
2088 module_param(zfs_deadman_synctime_ms
, ulong
, 0644);
2089 MODULE_PARM_DESC(zfs_deadman_synctime_ms
, "Expiration time in milliseconds");
2091 module_param(zfs_deadman_enabled
, int, 0644);
2092 MODULE_PARM_DESC(zfs_deadman_enabled
, "Enable deadman timer");
2094 module_param(spa_asize_inflation
, int, 0644);
2095 MODULE_PARM_DESC(spa_asize_inflation
,
2096 "SPA size estimate multiplication factor");
2098 module_param(spa_slop_shift
, int, 0644);
2099 MODULE_PARM_DESC(spa_slop_shift
, "Reserved free space in pool");