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 2015 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
26 * Copyright 2013 Saso Kiselkov. All rights reserved.
29 #include <sys/zfs_context.h>
30 #include <sys/spa_impl.h>
32 #include <sys/zio_checksum.h>
33 #include <sys/zio_compress.h>
35 #include <sys/dmu_tx.h>
38 #include <sys/vdev_impl.h>
39 #include <sys/vdev_file.h>
40 #include <sys/vdev_raidz.h>
41 #include <sys/metaslab.h>
42 #include <sys/uberblock_impl.h>
45 #include <sys/unique.h>
46 #include <sys/dsl_pool.h>
47 #include <sys/dsl_dir.h>
48 #include <sys/dsl_prop.h>
49 #include <sys/fm/util.h>
50 #include <sys/dsl_scan.h>
51 #include <sys/fs/zfs.h>
52 #include <sys/metaslab_impl.h>
55 #include <sys/kstat.h>
57 #include <sys/zfeature.h>
62 * There are four basic locks for managing spa_t structures:
64 * spa_namespace_lock (global mutex)
66 * This lock must be acquired to do any of the following:
68 * - Lookup a spa_t by name
69 * - Add or remove a spa_t from the namespace
70 * - Increase spa_refcount from non-zero
71 * - Check if spa_refcount is zero
73 * - add/remove/attach/detach devices
74 * - Held for the duration of create/destroy/import/export
76 * It does not need to handle recursion. A create or destroy may
77 * reference objects (files or zvols) in other pools, but by
78 * definition they must have an existing reference, and will never need
79 * to lookup a spa_t by name.
81 * spa_refcount (per-spa refcount_t protected by mutex)
83 * This reference count keep track of any active users of the spa_t. The
84 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
85 * the refcount is never really 'zero' - opening a pool implicitly keeps
86 * some references in the DMU. Internally we check against spa_minref, but
87 * present the image of a zero/non-zero value to consumers.
89 * spa_config_lock[] (per-spa array of rwlocks)
91 * This protects the spa_t from config changes, and must be held in
92 * the following circumstances:
94 * - RW_READER to perform I/O to the spa
95 * - RW_WRITER to change the vdev config
97 * The locking order is fairly straightforward:
99 * spa_namespace_lock -> spa_refcount
101 * The namespace lock must be acquired to increase the refcount from 0
102 * or to check if it is zero.
104 * spa_refcount -> spa_config_lock[]
106 * There must be at least one valid reference on the spa_t to acquire
109 * spa_namespace_lock -> spa_config_lock[]
111 * The namespace lock must always be taken before the config lock.
114 * The spa_namespace_lock can be acquired directly and is globally visible.
116 * The namespace is manipulated using the following functions, all of which
117 * require the spa_namespace_lock to be held.
119 * spa_lookup() Lookup a spa_t by name.
121 * spa_add() Create a new spa_t in the namespace.
123 * spa_remove() Remove a spa_t from the namespace. This also
124 * frees up any memory associated with the spa_t.
126 * spa_next() Returns the next spa_t in the system, or the
127 * first if NULL is passed.
129 * spa_evict_all() Shutdown and remove all spa_t structures in
132 * spa_guid_exists() Determine whether a pool/device guid exists.
134 * The spa_refcount is manipulated using the following functions:
136 * spa_open_ref() Adds a reference to the given spa_t. Must be
137 * called with spa_namespace_lock held if the
138 * refcount is currently zero.
140 * spa_close() Remove a reference from the spa_t. This will
141 * not free the spa_t or remove it from the
142 * namespace. No locking is required.
144 * spa_refcount_zero() Returns true if the refcount is currently
145 * zero. Must be called with spa_namespace_lock
148 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
149 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
150 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
152 * To read the configuration, it suffices to hold one of these locks as reader.
153 * To modify the configuration, you must hold all locks as writer. To modify
154 * vdev state without altering the vdev tree's topology (e.g. online/offline),
155 * you must hold SCL_STATE and SCL_ZIO as writer.
157 * We use these distinct config locks to avoid recursive lock entry.
158 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
159 * block allocations (SCL_ALLOC), which may require reading space maps
160 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
162 * The spa config locks cannot be normal rwlocks because we need the
163 * ability to hand off ownership. For example, SCL_ZIO is acquired
164 * by the issuing thread and later released by an interrupt thread.
165 * They do, however, obey the usual write-wanted semantics to prevent
166 * writer (i.e. system administrator) starvation.
168 * The lock acquisition rules are as follows:
171 * Protects changes to the vdev tree topology, such as vdev
172 * add/remove/attach/detach. Protects the dirty config list
173 * (spa_config_dirty_list) and the set of spares and l2arc devices.
176 * Protects changes to pool state and vdev state, such as vdev
177 * online/offline/fault/degrade/clear. Protects the dirty state list
178 * (spa_state_dirty_list) and global pool state (spa_state).
181 * Protects changes to metaslab groups and classes.
182 * Held as reader by metaslab_alloc() and metaslab_claim().
185 * Held by bp-level zios (those which have no io_vd upon entry)
186 * to prevent changes to the vdev tree. The bp-level zio implicitly
187 * protects all of its vdev child zios, which do not hold SCL_ZIO.
190 * Protects changes to metaslab groups and classes.
191 * Held as reader by metaslab_free(). SCL_FREE is distinct from
192 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
193 * blocks in zio_done() while another i/o that holds either
194 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
197 * Held as reader to prevent changes to the vdev tree during trivial
198 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
199 * other locks, and lower than all of them, to ensure that it's safe
200 * to acquire regardless of caller context.
202 * In addition, the following rules apply:
204 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
205 * The lock ordering is SCL_CONFIG > spa_props_lock.
207 * (b) I/O operations on leaf vdevs. For any zio operation that takes
208 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
209 * or zio_write_phys() -- the caller must ensure that the config cannot
210 * cannot change in the interim, and that the vdev cannot be reopened.
211 * SCL_STATE as reader suffices for both.
213 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
215 * spa_vdev_enter() Acquire the namespace lock and the config lock
218 * spa_vdev_exit() Release the config lock, wait for all I/O
219 * to complete, sync the updated configs to the
220 * cache, and release the namespace lock.
222 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
223 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
224 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
226 * spa_rename() is also implemented within this file since it requires
227 * manipulation of the namespace.
230 static avl_tree_t spa_namespace_avl
;
231 kmutex_t spa_namespace_lock
;
232 static kcondvar_t spa_namespace_cv
;
233 int spa_max_replication_override
= SPA_DVAS_PER_BP
;
235 static kmutex_t spa_spare_lock
;
236 static avl_tree_t spa_spare_avl
;
237 static kmutex_t spa_l2cache_lock
;
238 static avl_tree_t spa_l2cache_avl
;
240 kmem_cache_t
*spa_buffer_pool
;
244 /* Everything except dprintf and spa is on by default in debug builds */
245 int zfs_flags
= ~(ZFS_DEBUG_DPRINTF
| ZFS_DEBUG_SPA
);
251 * zfs_recover can be set to nonzero to attempt to recover from
252 * otherwise-fatal errors, typically caused by on-disk corruption. When
253 * set, calls to zfs_panic_recover() will turn into warning messages.
254 * This should only be used as a last resort, as it typically results
255 * in leaked space, or worse.
257 int zfs_recover
= B_FALSE
;
260 * If destroy encounters an EIO while reading metadata (e.g. indirect
261 * blocks), space referenced by the missing metadata can not be freed.
262 * Normally this causes the background destroy to become "stalled", as
263 * it is unable to make forward progress. While in this stalled state,
264 * all remaining space to free from the error-encountering filesystem is
265 * "temporarily leaked". Set this flag to cause it to ignore the EIO,
266 * permanently leak the space from indirect blocks that can not be read,
267 * and continue to free everything else that it can.
269 * The default, "stalling" behavior is useful if the storage partially
270 * fails (i.e. some but not all i/os fail), and then later recovers. In
271 * this case, we will be able to continue pool operations while it is
272 * partially failed, and when it recovers, we can continue to free the
273 * space, with no leaks. However, note that this case is actually
276 * Typically pools either (a) fail completely (but perhaps temporarily,
277 * e.g. a top-level vdev going offline), or (b) have localized,
278 * permanent errors (e.g. disk returns the wrong data due to bit flip or
279 * firmware bug). In case (a), this setting does not matter because the
280 * pool will be suspended and the sync thread will not be able to make
281 * forward progress regardless. In case (b), because the error is
282 * permanent, the best we can do is leak the minimum amount of space,
283 * which is what setting this flag will do. Therefore, it is reasonable
284 * for this flag to normally be set, but we chose the more conservative
285 * approach of not setting it, so that there is no possibility of
286 * leaking space in the "partial temporary" failure case.
288 int zfs_free_leak_on_eio
= B_FALSE
;
291 * Expiration time in milliseconds. This value has two meanings. First it is
292 * used to determine when the spa_deadman() logic should fire. By default the
293 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
294 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
295 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
298 unsigned long zfs_deadman_synctime_ms
= 1000000ULL;
301 * Check time in milliseconds. This defines the frequency at which we check
304 unsigned long zfs_deadman_checktime_ms
= 5000ULL;
307 * By default the deadman is enabled.
309 int zfs_deadman_enabled
= 1;
312 * The worst case is single-sector max-parity RAID-Z blocks, in which
313 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
314 * times the size; so just assume that. Add to this the fact that
315 * we can have up to 3 DVAs per bp, and one more factor of 2 because
316 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together,
318 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
320 int spa_asize_inflation
= 24;
323 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
324 * the pool to be consumed. This ensures that we don't run the pool
325 * completely out of space, due to unaccounted changes (e.g. to the MOS).
326 * It also limits the worst-case time to allocate space. If we have
327 * less than this amount of free space, most ZPL operations (e.g. write,
328 * create) will return ENOSPC.
330 * Certain operations (e.g. file removal, most administrative actions) can
331 * use half the slop space. They will only return ENOSPC if less than half
332 * the slop space is free. Typically, once the pool has less than the slop
333 * space free, the user will use these operations to free up space in the pool.
334 * These are the operations that call dsl_pool_adjustedsize() with the netfree
335 * argument set to TRUE.
337 * A very restricted set of operations are always permitted, regardless of
338 * the amount of free space. These are the operations that call
339 * dsl_sync_task(ZFS_SPACE_CHECK_NONE), e.g. "zfs destroy". If these
340 * operations result in a net increase in the amount of space used,
341 * it is possible to run the pool completely out of space, causing it to
342 * be permanently read-only.
344 * Note that on very small pools, the slop space will be larger than
345 * 3.2%, in an effort to have it be at least spa_min_slop (128MB),
346 * but we never allow it to be more than half the pool size.
348 * See also the comments in zfs_space_check_t.
350 int spa_slop_shift
= 5;
351 uint64_t spa_min_slop
= 128 * 1024 * 1024;
354 * ==========================================================================
356 * ==========================================================================
359 spa_config_lock_init(spa_t
*spa
)
363 for (i
= 0; i
< SCL_LOCKS
; i
++) {
364 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
365 mutex_init(&scl
->scl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
366 cv_init(&scl
->scl_cv
, NULL
, CV_DEFAULT
, NULL
);
367 refcount_create_untracked(&scl
->scl_count
);
368 scl
->scl_writer
= NULL
;
369 scl
->scl_write_wanted
= 0;
374 spa_config_lock_destroy(spa_t
*spa
)
378 for (i
= 0; i
< SCL_LOCKS
; i
++) {
379 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
380 mutex_destroy(&scl
->scl_lock
);
381 cv_destroy(&scl
->scl_cv
);
382 refcount_destroy(&scl
->scl_count
);
383 ASSERT(scl
->scl_writer
== NULL
);
384 ASSERT(scl
->scl_write_wanted
== 0);
389 spa_config_tryenter(spa_t
*spa
, int locks
, void *tag
, krw_t rw
)
393 for (i
= 0; i
< SCL_LOCKS
; i
++) {
394 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
395 if (!(locks
& (1 << i
)))
397 mutex_enter(&scl
->scl_lock
);
398 if (rw
== RW_READER
) {
399 if (scl
->scl_writer
|| scl
->scl_write_wanted
) {
400 mutex_exit(&scl
->scl_lock
);
401 spa_config_exit(spa
, locks
& ((1 << i
) - 1),
406 ASSERT(scl
->scl_writer
!= curthread
);
407 if (!refcount_is_zero(&scl
->scl_count
)) {
408 mutex_exit(&scl
->scl_lock
);
409 spa_config_exit(spa
, locks
& ((1 << i
) - 1),
413 scl
->scl_writer
= curthread
;
415 (void) refcount_add(&scl
->scl_count
, tag
);
416 mutex_exit(&scl
->scl_lock
);
422 spa_config_enter(spa_t
*spa
, int locks
, void *tag
, krw_t rw
)
427 ASSERT3U(SCL_LOCKS
, <, sizeof (wlocks_held
) * NBBY
);
429 for (i
= 0; i
< SCL_LOCKS
; i
++) {
430 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
431 if (scl
->scl_writer
== curthread
)
432 wlocks_held
|= (1 << i
);
433 if (!(locks
& (1 << i
)))
435 mutex_enter(&scl
->scl_lock
);
436 if (rw
== RW_READER
) {
437 while (scl
->scl_writer
|| scl
->scl_write_wanted
) {
438 cv_wait(&scl
->scl_cv
, &scl
->scl_lock
);
441 ASSERT(scl
->scl_writer
!= curthread
);
442 while (!refcount_is_zero(&scl
->scl_count
)) {
443 scl
->scl_write_wanted
++;
444 cv_wait(&scl
->scl_cv
, &scl
->scl_lock
);
445 scl
->scl_write_wanted
--;
447 scl
->scl_writer
= curthread
;
449 (void) refcount_add(&scl
->scl_count
, tag
);
450 mutex_exit(&scl
->scl_lock
);
452 ASSERT(wlocks_held
<= locks
);
456 spa_config_exit(spa_t
*spa
, int locks
, void *tag
)
460 for (i
= SCL_LOCKS
- 1; i
>= 0; i
--) {
461 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
462 if (!(locks
& (1 << i
)))
464 mutex_enter(&scl
->scl_lock
);
465 ASSERT(!refcount_is_zero(&scl
->scl_count
));
466 if (refcount_remove(&scl
->scl_count
, tag
) == 0) {
467 ASSERT(scl
->scl_writer
== NULL
||
468 scl
->scl_writer
== curthread
);
469 scl
->scl_writer
= NULL
; /* OK in either case */
470 cv_broadcast(&scl
->scl_cv
);
472 mutex_exit(&scl
->scl_lock
);
477 spa_config_held(spa_t
*spa
, int locks
, krw_t rw
)
479 int i
, locks_held
= 0;
481 for (i
= 0; i
< SCL_LOCKS
; i
++) {
482 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
483 if (!(locks
& (1 << i
)))
485 if ((rw
== RW_READER
&& !refcount_is_zero(&scl
->scl_count
)) ||
486 (rw
== RW_WRITER
&& scl
->scl_writer
== curthread
))
487 locks_held
|= 1 << i
;
494 * ==========================================================================
495 * SPA namespace functions
496 * ==========================================================================
500 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
501 * Returns NULL if no matching spa_t is found.
504 spa_lookup(const char *name
)
506 static spa_t search
; /* spa_t is large; don't allocate on stack */
511 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
513 (void) strlcpy(search
.spa_name
, name
, sizeof (search
.spa_name
));
516 * If it's a full dataset name, figure out the pool name and
519 cp
= strpbrk(search
.spa_name
, "/@#");
523 spa
= avl_find(&spa_namespace_avl
, &search
, &where
);
529 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
530 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
531 * looking for potentially hung I/Os.
534 spa_deadman(void *arg
)
538 /* Disable the deadman if the pool is suspended. */
539 if (spa_suspended(spa
))
542 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
543 (gethrtime() - spa
->spa_sync_starttime
) / NANOSEC
,
544 ++spa
->spa_deadman_calls
);
545 if (zfs_deadman_enabled
)
546 vdev_deadman(spa
->spa_root_vdev
);
548 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_delay_taskq
,
549 spa_deadman
, spa
, TQ_SLEEP
, ddi_get_lbolt() +
550 MSEC_TO_TICK(zfs_deadman_checktime_ms
));
554 * Create an uninitialized spa_t with the given name. Requires
555 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
556 * exist by calling spa_lookup() first.
559 spa_add(const char *name
, nvlist_t
*config
, const char *altroot
)
562 spa_config_dirent_t
*dp
;
566 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
568 spa
= kmem_zalloc(sizeof (spa_t
), KM_SLEEP
);
570 mutex_init(&spa
->spa_async_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
571 mutex_init(&spa
->spa_errlist_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
572 mutex_init(&spa
->spa_errlog_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
573 mutex_init(&spa
->spa_evicting_os_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
574 mutex_init(&spa
->spa_history_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
575 mutex_init(&spa
->spa_proc_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
576 mutex_init(&spa
->spa_props_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
577 mutex_init(&spa
->spa_cksum_tmpls_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
578 mutex_init(&spa
->spa_scrub_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
579 mutex_init(&spa
->spa_suspend_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
580 mutex_init(&spa
->spa_vdev_top_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
581 mutex_init(&spa
->spa_feat_stats_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
582 mutex_init(&spa
->spa_alloc_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
584 cv_init(&spa
->spa_async_cv
, NULL
, CV_DEFAULT
, NULL
);
585 cv_init(&spa
->spa_evicting_os_cv
, NULL
, CV_DEFAULT
, NULL
);
586 cv_init(&spa
->spa_proc_cv
, NULL
, CV_DEFAULT
, NULL
);
587 cv_init(&spa
->spa_scrub_io_cv
, NULL
, CV_DEFAULT
, NULL
);
588 cv_init(&spa
->spa_suspend_cv
, NULL
, CV_DEFAULT
, NULL
);
590 for (t
= 0; t
< TXG_SIZE
; t
++)
591 bplist_create(&spa
->spa_free_bplist
[t
]);
593 (void) strlcpy(spa
->spa_name
, name
, sizeof (spa
->spa_name
));
594 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
595 spa
->spa_freeze_txg
= UINT64_MAX
;
596 spa
->spa_final_txg
= UINT64_MAX
;
597 spa
->spa_load_max_txg
= UINT64_MAX
;
599 spa
->spa_proc_state
= SPA_PROC_NONE
;
601 spa
->spa_deadman_synctime
= MSEC2NSEC(zfs_deadman_synctime_ms
);
603 refcount_create(&spa
->spa_refcount
);
604 spa_config_lock_init(spa
);
607 avl_add(&spa_namespace_avl
, spa
);
610 * Set the alternate root, if there is one.
613 spa
->spa_root
= spa_strdup(altroot
);
615 avl_create(&spa
->spa_alloc_tree
, zio_timestamp_compare
,
616 sizeof (zio_t
), offsetof(zio_t
, io_alloc_node
));
619 * Every pool starts with the default cachefile
621 list_create(&spa
->spa_config_list
, sizeof (spa_config_dirent_t
),
622 offsetof(spa_config_dirent_t
, scd_link
));
624 dp
= kmem_zalloc(sizeof (spa_config_dirent_t
), KM_SLEEP
);
625 dp
->scd_path
= altroot
? NULL
: spa_strdup(spa_config_path
);
626 list_insert_head(&spa
->spa_config_list
, dp
);
628 VERIFY(nvlist_alloc(&spa
->spa_load_info
, NV_UNIQUE_NAME
,
631 if (config
!= NULL
) {
634 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_FEATURES_FOR_READ
,
636 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
,
640 VERIFY(nvlist_dup(config
, &spa
->spa_config
, 0) == 0);
643 if (spa
->spa_label_features
== NULL
) {
644 VERIFY(nvlist_alloc(&spa
->spa_label_features
, NV_UNIQUE_NAME
,
648 spa
->spa_debug
= ((zfs_flags
& ZFS_DEBUG_SPA
) != 0);
650 spa
->spa_min_ashift
= INT_MAX
;
651 spa
->spa_max_ashift
= 0;
653 /* Reset cached value */
654 spa
->spa_dedup_dspace
= ~0ULL;
657 * As a pool is being created, treat all features as disabled by
658 * setting SPA_FEATURE_DISABLED for all entries in the feature
661 for (i
= 0; i
< SPA_FEATURES
; i
++) {
662 spa
->spa_feat_refcount_cache
[i
] = SPA_FEATURE_DISABLED
;
669 * Removes a spa_t from the namespace, freeing up any memory used. Requires
670 * spa_namespace_lock. This is called only after the spa_t has been closed and
674 spa_remove(spa_t
*spa
)
676 spa_config_dirent_t
*dp
;
679 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
680 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
681 ASSERT3U(refcount_count(&spa
->spa_refcount
), ==, 0);
683 nvlist_free(spa
->spa_config_splitting
);
685 avl_remove(&spa_namespace_avl
, spa
);
686 cv_broadcast(&spa_namespace_cv
);
689 spa_strfree(spa
->spa_root
);
691 while ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
692 list_remove(&spa
->spa_config_list
, dp
);
693 if (dp
->scd_path
!= NULL
)
694 spa_strfree(dp
->scd_path
);
695 kmem_free(dp
, sizeof (spa_config_dirent_t
));
698 avl_destroy(&spa
->spa_alloc_tree
);
699 list_destroy(&spa
->spa_config_list
);
701 nvlist_free(spa
->spa_label_features
);
702 nvlist_free(spa
->spa_load_info
);
703 nvlist_free(spa
->spa_feat_stats
);
704 spa_config_set(spa
, NULL
);
706 refcount_destroy(&spa
->spa_refcount
);
708 spa_stats_destroy(spa
);
709 spa_config_lock_destroy(spa
);
711 for (t
= 0; t
< TXG_SIZE
; t
++)
712 bplist_destroy(&spa
->spa_free_bplist
[t
]);
714 zio_checksum_templates_free(spa
);
716 cv_destroy(&spa
->spa_async_cv
);
717 cv_destroy(&spa
->spa_evicting_os_cv
);
718 cv_destroy(&spa
->spa_proc_cv
);
719 cv_destroy(&spa
->spa_scrub_io_cv
);
720 cv_destroy(&spa
->spa_suspend_cv
);
722 mutex_destroy(&spa
->spa_alloc_lock
);
723 mutex_destroy(&spa
->spa_async_lock
);
724 mutex_destroy(&spa
->spa_errlist_lock
);
725 mutex_destroy(&spa
->spa_errlog_lock
);
726 mutex_destroy(&spa
->spa_evicting_os_lock
);
727 mutex_destroy(&spa
->spa_history_lock
);
728 mutex_destroy(&spa
->spa_proc_lock
);
729 mutex_destroy(&spa
->spa_props_lock
);
730 mutex_destroy(&spa
->spa_cksum_tmpls_lock
);
731 mutex_destroy(&spa
->spa_scrub_lock
);
732 mutex_destroy(&spa
->spa_suspend_lock
);
733 mutex_destroy(&spa
->spa_vdev_top_lock
);
734 mutex_destroy(&spa
->spa_feat_stats_lock
);
736 kmem_free(spa
, sizeof (spa_t
));
740 * Given a pool, return the next pool in the namespace, or NULL if there is
741 * none. If 'prev' is NULL, return the first pool.
744 spa_next(spa_t
*prev
)
746 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
749 return (AVL_NEXT(&spa_namespace_avl
, prev
));
751 return (avl_first(&spa_namespace_avl
));
755 * ==========================================================================
756 * SPA refcount functions
757 * ==========================================================================
761 * Add a reference to the given spa_t. Must have at least one reference, or
762 * have the namespace lock held.
765 spa_open_ref(spa_t
*spa
, void *tag
)
767 ASSERT(refcount_count(&spa
->spa_refcount
) >= spa
->spa_minref
||
768 MUTEX_HELD(&spa_namespace_lock
));
769 (void) refcount_add(&spa
->spa_refcount
, tag
);
773 * Remove a reference to the given spa_t. Must have at least one reference, or
774 * have the namespace lock held.
777 spa_close(spa_t
*spa
, void *tag
)
779 ASSERT(refcount_count(&spa
->spa_refcount
) > spa
->spa_minref
||
780 MUTEX_HELD(&spa_namespace_lock
));
781 (void) refcount_remove(&spa
->spa_refcount
, tag
);
785 * Remove a reference to the given spa_t held by a dsl dir that is
786 * being asynchronously released. Async releases occur from a taskq
787 * performing eviction of dsl datasets and dirs. The namespace lock
788 * isn't held and the hold by the object being evicted may contribute to
789 * spa_minref (e.g. dataset or directory released during pool export),
790 * so the asserts in spa_close() do not apply.
793 spa_async_close(spa_t
*spa
, void *tag
)
795 (void) refcount_remove(&spa
->spa_refcount
, tag
);
799 * Check to see if the spa refcount is zero. Must be called with
800 * spa_namespace_lock held. We really compare against spa_minref, which is the
801 * number of references acquired when opening a pool
804 spa_refcount_zero(spa_t
*spa
)
806 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
808 return (refcount_count(&spa
->spa_refcount
) == spa
->spa_minref
);
812 * ==========================================================================
813 * SPA spare and l2cache tracking
814 * ==========================================================================
818 * Hot spares and cache devices are tracked using the same code below,
819 * for 'auxiliary' devices.
822 typedef struct spa_aux
{
830 spa_aux_compare(const void *a
, const void *b
)
832 const spa_aux_t
*sa
= (const spa_aux_t
*)a
;
833 const spa_aux_t
*sb
= (const spa_aux_t
*)b
;
835 return (AVL_CMP(sa
->aux_guid
, sb
->aux_guid
));
839 spa_aux_add(vdev_t
*vd
, avl_tree_t
*avl
)
845 search
.aux_guid
= vd
->vdev_guid
;
846 if ((aux
= avl_find(avl
, &search
, &where
)) != NULL
) {
849 aux
= kmem_zalloc(sizeof (spa_aux_t
), KM_SLEEP
);
850 aux
->aux_guid
= vd
->vdev_guid
;
852 avl_insert(avl
, aux
, where
);
857 spa_aux_remove(vdev_t
*vd
, avl_tree_t
*avl
)
863 search
.aux_guid
= vd
->vdev_guid
;
864 aux
= avl_find(avl
, &search
, &where
);
868 if (--aux
->aux_count
== 0) {
869 avl_remove(avl
, aux
);
870 kmem_free(aux
, sizeof (spa_aux_t
));
871 } else if (aux
->aux_pool
== spa_guid(vd
->vdev_spa
)) {
872 aux
->aux_pool
= 0ULL;
877 spa_aux_exists(uint64_t guid
, uint64_t *pool
, int *refcnt
, avl_tree_t
*avl
)
879 spa_aux_t search
, *found
;
881 search
.aux_guid
= guid
;
882 found
= avl_find(avl
, &search
, NULL
);
886 *pool
= found
->aux_pool
;
893 *refcnt
= found
->aux_count
;
898 return (found
!= NULL
);
902 spa_aux_activate(vdev_t
*vd
, avl_tree_t
*avl
)
904 spa_aux_t search
, *found
;
907 search
.aux_guid
= vd
->vdev_guid
;
908 found
= avl_find(avl
, &search
, &where
);
909 ASSERT(found
!= NULL
);
910 ASSERT(found
->aux_pool
== 0ULL);
912 found
->aux_pool
= spa_guid(vd
->vdev_spa
);
916 * Spares are tracked globally due to the following constraints:
918 * - A spare may be part of multiple pools.
919 * - A spare may be added to a pool even if it's actively in use within
921 * - A spare in use in any pool can only be the source of a replacement if
922 * the target is a spare in the same pool.
924 * We keep track of all spares on the system through the use of a reference
925 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
926 * spare, then we bump the reference count in the AVL tree. In addition, we set
927 * the 'vdev_isspare' member to indicate that the device is a spare (active or
928 * inactive). When a spare is made active (used to replace a device in the
929 * pool), we also keep track of which pool its been made a part of.
931 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
932 * called under the spa_namespace lock as part of vdev reconfiguration. The
933 * separate spare lock exists for the status query path, which does not need to
934 * be completely consistent with respect to other vdev configuration changes.
938 spa_spare_compare(const void *a
, const void *b
)
940 return (spa_aux_compare(a
, b
));
944 spa_spare_add(vdev_t
*vd
)
946 mutex_enter(&spa_spare_lock
);
947 ASSERT(!vd
->vdev_isspare
);
948 spa_aux_add(vd
, &spa_spare_avl
);
949 vd
->vdev_isspare
= B_TRUE
;
950 mutex_exit(&spa_spare_lock
);
954 spa_spare_remove(vdev_t
*vd
)
956 mutex_enter(&spa_spare_lock
);
957 ASSERT(vd
->vdev_isspare
);
958 spa_aux_remove(vd
, &spa_spare_avl
);
959 vd
->vdev_isspare
= B_FALSE
;
960 mutex_exit(&spa_spare_lock
);
964 spa_spare_exists(uint64_t guid
, uint64_t *pool
, int *refcnt
)
968 mutex_enter(&spa_spare_lock
);
969 found
= spa_aux_exists(guid
, pool
, refcnt
, &spa_spare_avl
);
970 mutex_exit(&spa_spare_lock
);
976 spa_spare_activate(vdev_t
*vd
)
978 mutex_enter(&spa_spare_lock
);
979 ASSERT(vd
->vdev_isspare
);
980 spa_aux_activate(vd
, &spa_spare_avl
);
981 mutex_exit(&spa_spare_lock
);
985 * Level 2 ARC devices are tracked globally for the same reasons as spares.
986 * Cache devices currently only support one pool per cache device, and so
987 * for these devices the aux reference count is currently unused beyond 1.
991 spa_l2cache_compare(const void *a
, const void *b
)
993 return (spa_aux_compare(a
, b
));
997 spa_l2cache_add(vdev_t
*vd
)
999 mutex_enter(&spa_l2cache_lock
);
1000 ASSERT(!vd
->vdev_isl2cache
);
1001 spa_aux_add(vd
, &spa_l2cache_avl
);
1002 vd
->vdev_isl2cache
= B_TRUE
;
1003 mutex_exit(&spa_l2cache_lock
);
1007 spa_l2cache_remove(vdev_t
*vd
)
1009 mutex_enter(&spa_l2cache_lock
);
1010 ASSERT(vd
->vdev_isl2cache
);
1011 spa_aux_remove(vd
, &spa_l2cache_avl
);
1012 vd
->vdev_isl2cache
= B_FALSE
;
1013 mutex_exit(&spa_l2cache_lock
);
1017 spa_l2cache_exists(uint64_t guid
, uint64_t *pool
)
1021 mutex_enter(&spa_l2cache_lock
);
1022 found
= spa_aux_exists(guid
, pool
, NULL
, &spa_l2cache_avl
);
1023 mutex_exit(&spa_l2cache_lock
);
1029 spa_l2cache_activate(vdev_t
*vd
)
1031 mutex_enter(&spa_l2cache_lock
);
1032 ASSERT(vd
->vdev_isl2cache
);
1033 spa_aux_activate(vd
, &spa_l2cache_avl
);
1034 mutex_exit(&spa_l2cache_lock
);
1038 * ==========================================================================
1040 * ==========================================================================
1044 * Lock the given spa_t for the purpose of adding or removing a vdev.
1045 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1046 * It returns the next transaction group for the spa_t.
1049 spa_vdev_enter(spa_t
*spa
)
1051 mutex_enter(&spa
->spa_vdev_top_lock
);
1052 mutex_enter(&spa_namespace_lock
);
1053 return (spa_vdev_config_enter(spa
));
1057 * Internal implementation for spa_vdev_enter(). Used when a vdev
1058 * operation requires multiple syncs (i.e. removing a device) while
1059 * keeping the spa_namespace_lock held.
1062 spa_vdev_config_enter(spa_t
*spa
)
1064 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1066 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
1068 return (spa_last_synced_txg(spa
) + 1);
1072 * Used in combination with spa_vdev_config_enter() to allow the syncing
1073 * of multiple transactions without releasing the spa_namespace_lock.
1076 spa_vdev_config_exit(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int error
, char *tag
)
1078 int config_changed
= B_FALSE
;
1080 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1081 ASSERT(txg
> spa_last_synced_txg(spa
));
1083 spa
->spa_pending_vdev
= NULL
;
1086 * Reassess the DTLs.
1088 vdev_dtl_reassess(spa
->spa_root_vdev
, 0, 0, B_FALSE
);
1090 if (error
== 0 && !list_is_empty(&spa
->spa_config_dirty_list
)) {
1091 config_changed
= B_TRUE
;
1092 spa
->spa_config_generation
++;
1096 * Verify the metaslab classes.
1098 ASSERT(metaslab_class_validate(spa_normal_class(spa
)) == 0);
1099 ASSERT(metaslab_class_validate(spa_log_class(spa
)) == 0);
1101 spa_config_exit(spa
, SCL_ALL
, spa
);
1104 * Panic the system if the specified tag requires it. This
1105 * is useful for ensuring that configurations are updated
1108 if (zio_injection_enabled
)
1109 zio_handle_panic_injection(spa
, tag
, 0);
1112 * Note: this txg_wait_synced() is important because it ensures
1113 * that there won't be more than one config change per txg.
1114 * This allows us to use the txg as the generation number.
1117 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
1120 ASSERT(!vd
->vdev_detached
|| vd
->vdev_dtl_sm
== NULL
);
1121 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
1123 spa_config_exit(spa
, SCL_ALL
, spa
);
1127 * If the config changed, update the config cache.
1130 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
1134 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1135 * locking of spa_vdev_enter(), we also want make sure the transactions have
1136 * synced to disk, and then update the global configuration cache with the new
1140 spa_vdev_exit(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int error
)
1142 spa_vdev_config_exit(spa
, vd
, txg
, error
, FTAG
);
1143 mutex_exit(&spa_namespace_lock
);
1144 mutex_exit(&spa
->spa_vdev_top_lock
);
1150 * Lock the given spa_t for the purpose of changing vdev state.
1153 spa_vdev_state_enter(spa_t
*spa
, int oplocks
)
1155 int locks
= SCL_STATE_ALL
| oplocks
;
1158 * Root pools may need to read of the underlying devfs filesystem
1159 * when opening up a vdev. Unfortunately if we're holding the
1160 * SCL_ZIO lock it will result in a deadlock when we try to issue
1161 * the read from the root filesystem. Instead we "prefetch"
1162 * the associated vnodes that we need prior to opening the
1163 * underlying devices and cache them so that we can prevent
1164 * any I/O when we are doing the actual open.
1166 if (spa_is_root(spa
)) {
1167 int low
= locks
& ~(SCL_ZIO
- 1);
1168 int high
= locks
& ~low
;
1170 spa_config_enter(spa
, high
, spa
, RW_WRITER
);
1171 vdev_hold(spa
->spa_root_vdev
);
1172 spa_config_enter(spa
, low
, spa
, RW_WRITER
);
1174 spa_config_enter(spa
, locks
, spa
, RW_WRITER
);
1176 spa
->spa_vdev_locks
= locks
;
1180 spa_vdev_state_exit(spa_t
*spa
, vdev_t
*vd
, int error
)
1182 boolean_t config_changed
= B_FALSE
;
1184 if (vd
!= NULL
|| error
== 0)
1185 vdev_dtl_reassess(vd
? vd
->vdev_top
: spa
->spa_root_vdev
,
1189 vdev_state_dirty(vd
->vdev_top
);
1190 config_changed
= B_TRUE
;
1191 spa
->spa_config_generation
++;
1194 if (spa_is_root(spa
))
1195 vdev_rele(spa
->spa_root_vdev
);
1197 ASSERT3U(spa
->spa_vdev_locks
, >=, SCL_STATE_ALL
);
1198 spa_config_exit(spa
, spa
->spa_vdev_locks
, spa
);
1201 * If anything changed, wait for it to sync. This ensures that,
1202 * from the system administrator's perspective, zpool(1M) commands
1203 * are synchronous. This is important for things like zpool offline:
1204 * when the command completes, you expect no further I/O from ZFS.
1207 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1210 * If the config changed, update the config cache.
1212 if (config_changed
) {
1213 mutex_enter(&spa_namespace_lock
);
1214 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
1215 mutex_exit(&spa_namespace_lock
);
1222 * ==========================================================================
1223 * Miscellaneous functions
1224 * ==========================================================================
1228 spa_activate_mos_feature(spa_t
*spa
, const char *feature
, dmu_tx_t
*tx
)
1230 if (!nvlist_exists(spa
->spa_label_features
, feature
)) {
1231 fnvlist_add_boolean(spa
->spa_label_features
, feature
);
1233 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1234 * dirty the vdev config because lock SCL_CONFIG is not held.
1235 * Thankfully, in this case we don't need to dirty the config
1236 * because it will be written out anyway when we finish
1237 * creating the pool.
1239 if (tx
->tx_txg
!= TXG_INITIAL
)
1240 vdev_config_dirty(spa
->spa_root_vdev
);
1245 spa_deactivate_mos_feature(spa_t
*spa
, const char *feature
)
1247 if (nvlist_remove_all(spa
->spa_label_features
, feature
) == 0)
1248 vdev_config_dirty(spa
->spa_root_vdev
);
1255 spa_rename(const char *name
, const char *newname
)
1261 * Lookup the spa_t and grab the config lock for writing. We need to
1262 * actually open the pool so that we can sync out the necessary labels.
1263 * It's OK to call spa_open() with the namespace lock held because we
1264 * allow recursive calls for other reasons.
1266 mutex_enter(&spa_namespace_lock
);
1267 if ((err
= spa_open(name
, &spa
, FTAG
)) != 0) {
1268 mutex_exit(&spa_namespace_lock
);
1272 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1274 avl_remove(&spa_namespace_avl
, spa
);
1275 (void) strlcpy(spa
->spa_name
, newname
, sizeof (spa
->spa_name
));
1276 avl_add(&spa_namespace_avl
, spa
);
1279 * Sync all labels to disk with the new names by marking the root vdev
1280 * dirty and waiting for it to sync. It will pick up the new pool name
1283 vdev_config_dirty(spa
->spa_root_vdev
);
1285 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1287 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1290 * Sync the updated config cache.
1292 spa_config_sync(spa
, B_FALSE
, B_TRUE
);
1294 spa_close(spa
, FTAG
);
1296 mutex_exit(&spa_namespace_lock
);
1302 * Return the spa_t associated with given pool_guid, if it exists. If
1303 * device_guid is non-zero, determine whether the pool exists *and* contains
1304 * a device with the specified device_guid.
1307 spa_by_guid(uint64_t pool_guid
, uint64_t device_guid
)
1310 avl_tree_t
*t
= &spa_namespace_avl
;
1312 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1314 for (spa
= avl_first(t
); spa
!= NULL
; spa
= AVL_NEXT(t
, spa
)) {
1315 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1317 if (spa
->spa_root_vdev
== NULL
)
1319 if (spa_guid(spa
) == pool_guid
) {
1320 if (device_guid
== 0)
1323 if (vdev_lookup_by_guid(spa
->spa_root_vdev
,
1324 device_guid
) != NULL
)
1328 * Check any devices we may be in the process of adding.
1330 if (spa
->spa_pending_vdev
) {
1331 if (vdev_lookup_by_guid(spa
->spa_pending_vdev
,
1332 device_guid
) != NULL
)
1342 * Determine whether a pool with the given pool_guid exists.
1345 spa_guid_exists(uint64_t pool_guid
, uint64_t device_guid
)
1347 return (spa_by_guid(pool_guid
, device_guid
) != NULL
);
1351 spa_strdup(const char *s
)
1357 new = kmem_alloc(len
+ 1, KM_SLEEP
);
1365 spa_strfree(char *s
)
1367 kmem_free(s
, strlen(s
) + 1);
1371 spa_get_random(uint64_t range
)
1377 (void) random_get_pseudo_bytes((void *)&r
, sizeof (uint64_t));
1383 spa_generate_guid(spa_t
*spa
)
1385 uint64_t guid
= spa_get_random(-1ULL);
1388 while (guid
== 0 || spa_guid_exists(spa_guid(spa
), guid
))
1389 guid
= spa_get_random(-1ULL);
1391 while (guid
== 0 || spa_guid_exists(guid
, 0))
1392 guid
= spa_get_random(-1ULL);
1399 snprintf_blkptr(char *buf
, size_t buflen
, const blkptr_t
*bp
)
1402 char *checksum
= NULL
;
1403 char *compress
= NULL
;
1406 if (BP_GET_TYPE(bp
) & DMU_OT_NEWTYPE
) {
1407 dmu_object_byteswap_t bswap
=
1408 DMU_OT_BYTESWAP(BP_GET_TYPE(bp
));
1409 (void) snprintf(type
, sizeof (type
), "bswap %s %s",
1410 DMU_OT_IS_METADATA(BP_GET_TYPE(bp
)) ?
1411 "metadata" : "data",
1412 dmu_ot_byteswap
[bswap
].ob_name
);
1414 (void) strlcpy(type
, dmu_ot
[BP_GET_TYPE(bp
)].ot_name
,
1417 if (!BP_IS_EMBEDDED(bp
)) {
1419 zio_checksum_table
[BP_GET_CHECKSUM(bp
)].ci_name
;
1421 compress
= zio_compress_table
[BP_GET_COMPRESS(bp
)].ci_name
;
1424 SNPRINTF_BLKPTR(snprintf
, ' ', buf
, buflen
, bp
, type
, checksum
,
1429 spa_freeze(spa_t
*spa
)
1431 uint64_t freeze_txg
= 0;
1433 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1434 if (spa
->spa_freeze_txg
== UINT64_MAX
) {
1435 freeze_txg
= spa_last_synced_txg(spa
) + TXG_SIZE
;
1436 spa
->spa_freeze_txg
= freeze_txg
;
1438 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1439 if (freeze_txg
!= 0)
1440 txg_wait_synced(spa_get_dsl(spa
), freeze_txg
);
1444 zfs_panic_recover(const char *fmt
, ...)
1449 vcmn_err(zfs_recover
? CE_WARN
: CE_PANIC
, fmt
, adx
);
1454 * This is a stripped-down version of strtoull, suitable only for converting
1455 * lowercase hexadecimal numbers that don't overflow.
1458 strtonum(const char *str
, char **nptr
)
1464 while ((c
= *str
) != '\0') {
1465 if (c
>= '0' && c
<= '9')
1467 else if (c
>= 'a' && c
<= 'f')
1468 digit
= 10 + c
- 'a';
1479 *nptr
= (char *)str
;
1485 * ==========================================================================
1486 * Accessor functions
1487 * ==========================================================================
1491 spa_shutting_down(spa_t
*spa
)
1493 return (spa
->spa_async_suspended
);
1497 spa_get_dsl(spa_t
*spa
)
1499 return (spa
->spa_dsl_pool
);
1503 spa_is_initializing(spa_t
*spa
)
1505 return (spa
->spa_is_initializing
);
1509 spa_get_rootblkptr(spa_t
*spa
)
1511 return (&spa
->spa_ubsync
.ub_rootbp
);
1515 spa_set_rootblkptr(spa_t
*spa
, const blkptr_t
*bp
)
1517 spa
->spa_uberblock
.ub_rootbp
= *bp
;
1521 spa_altroot(spa_t
*spa
, char *buf
, size_t buflen
)
1523 if (spa
->spa_root
== NULL
)
1526 (void) strncpy(buf
, spa
->spa_root
, buflen
);
1530 spa_sync_pass(spa_t
*spa
)
1532 return (spa
->spa_sync_pass
);
1536 spa_name(spa_t
*spa
)
1538 return (spa
->spa_name
);
1542 spa_guid(spa_t
*spa
)
1544 dsl_pool_t
*dp
= spa_get_dsl(spa
);
1548 * If we fail to parse the config during spa_load(), we can go through
1549 * the error path (which posts an ereport) and end up here with no root
1550 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1553 if (spa
->spa_root_vdev
== NULL
)
1554 return (spa
->spa_config_guid
);
1556 guid
= spa
->spa_last_synced_guid
!= 0 ?
1557 spa
->spa_last_synced_guid
: spa
->spa_root_vdev
->vdev_guid
;
1560 * Return the most recently synced out guid unless we're
1561 * in syncing context.
1563 if (dp
&& dsl_pool_sync_context(dp
))
1564 return (spa
->spa_root_vdev
->vdev_guid
);
1570 spa_load_guid(spa_t
*spa
)
1573 * This is a GUID that exists solely as a reference for the
1574 * purposes of the arc. It is generated at load time, and
1575 * is never written to persistent storage.
1577 return (spa
->spa_load_guid
);
1581 spa_last_synced_txg(spa_t
*spa
)
1583 return (spa
->spa_ubsync
.ub_txg
);
1587 spa_first_txg(spa_t
*spa
)
1589 return (spa
->spa_first_txg
);
1593 spa_syncing_txg(spa_t
*spa
)
1595 return (spa
->spa_syncing_txg
);
1599 spa_state(spa_t
*spa
)
1601 return (spa
->spa_state
);
1605 spa_load_state(spa_t
*spa
)
1607 return (spa
->spa_load_state
);
1611 spa_freeze_txg(spa_t
*spa
)
1613 return (spa
->spa_freeze_txg
);
1618 spa_get_asize(spa_t
*spa
, uint64_t lsize
)
1620 return (lsize
* spa_asize_inflation
);
1624 * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%),
1625 * or at least 128MB, unless that would cause it to be more than half the
1628 * See the comment above spa_slop_shift for details.
1631 spa_get_slop_space(spa_t
*spa
)
1633 uint64_t space
= spa_get_dspace(spa
);
1634 return (MAX(space
>> spa_slop_shift
, MIN(space
>> 1, spa_min_slop
)));
1638 spa_get_dspace(spa_t
*spa
)
1640 return (spa
->spa_dspace
);
1644 spa_update_dspace(spa_t
*spa
)
1646 spa
->spa_dspace
= metaslab_class_get_dspace(spa_normal_class(spa
)) +
1647 ddt_get_dedup_dspace(spa
);
1651 * Return the failure mode that has been set to this pool. The default
1652 * behavior will be to block all I/Os when a complete failure occurs.
1655 spa_get_failmode(spa_t
*spa
)
1657 return (spa
->spa_failmode
);
1661 spa_suspended(spa_t
*spa
)
1663 return (spa
->spa_suspended
);
1667 spa_version(spa_t
*spa
)
1669 return (spa
->spa_ubsync
.ub_version
);
1673 spa_deflate(spa_t
*spa
)
1675 return (spa
->spa_deflate
);
1679 spa_normal_class(spa_t
*spa
)
1681 return (spa
->spa_normal_class
);
1685 spa_log_class(spa_t
*spa
)
1687 return (spa
->spa_log_class
);
1691 spa_evicting_os_register(spa_t
*spa
, objset_t
*os
)
1693 mutex_enter(&spa
->spa_evicting_os_lock
);
1694 list_insert_head(&spa
->spa_evicting_os_list
, os
);
1695 mutex_exit(&spa
->spa_evicting_os_lock
);
1699 spa_evicting_os_deregister(spa_t
*spa
, objset_t
*os
)
1701 mutex_enter(&spa
->spa_evicting_os_lock
);
1702 list_remove(&spa
->spa_evicting_os_list
, os
);
1703 cv_broadcast(&spa
->spa_evicting_os_cv
);
1704 mutex_exit(&spa
->spa_evicting_os_lock
);
1708 spa_evicting_os_wait(spa_t
*spa
)
1710 mutex_enter(&spa
->spa_evicting_os_lock
);
1711 while (!list_is_empty(&spa
->spa_evicting_os_list
))
1712 cv_wait(&spa
->spa_evicting_os_cv
, &spa
->spa_evicting_os_lock
);
1713 mutex_exit(&spa
->spa_evicting_os_lock
);
1715 dmu_buf_user_evict_wait();
1719 spa_max_replication(spa_t
*spa
)
1722 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1723 * handle BPs with more than one DVA allocated. Set our max
1724 * replication level accordingly.
1726 if (spa_version(spa
) < SPA_VERSION_DITTO_BLOCKS
)
1728 return (MIN(SPA_DVAS_PER_BP
, spa_max_replication_override
));
1732 spa_prev_software_version(spa_t
*spa
)
1734 return (spa
->spa_prev_software_version
);
1738 spa_deadman_synctime(spa_t
*spa
)
1740 return (spa
->spa_deadman_synctime
);
1744 dva_get_dsize_sync(spa_t
*spa
, const dva_t
*dva
)
1746 uint64_t asize
= DVA_GET_ASIZE(dva
);
1747 uint64_t dsize
= asize
;
1749 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1751 if (asize
!= 0 && spa
->spa_deflate
) {
1752 vdev_t
*vd
= vdev_lookup_top(spa
, DVA_GET_VDEV(dva
));
1754 dsize
= (asize
>> SPA_MINBLOCKSHIFT
) *
1755 vd
->vdev_deflate_ratio
;
1762 bp_get_dsize_sync(spa_t
*spa
, const blkptr_t
*bp
)
1767 for (d
= 0; d
< BP_GET_NDVAS(bp
); d
++)
1768 dsize
+= dva_get_dsize_sync(spa
, &bp
->blk_dva
[d
]);
1774 bp_get_dsize(spa_t
*spa
, const blkptr_t
*bp
)
1779 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
1781 for (d
= 0; d
< BP_GET_NDVAS(bp
); d
++)
1782 dsize
+= dva_get_dsize_sync(spa
, &bp
->blk_dva
[d
]);
1784 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
1790 * ==========================================================================
1791 * Initialization and Termination
1792 * ==========================================================================
1796 spa_name_compare(const void *a1
, const void *a2
)
1798 const spa_t
*s1
= a1
;
1799 const spa_t
*s2
= a2
;
1802 s
= strcmp(s1
->spa_name
, s2
->spa_name
);
1804 return (AVL_ISIGN(s
));
1816 mutex_init(&spa_namespace_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1817 mutex_init(&spa_spare_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1818 mutex_init(&spa_l2cache_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1819 cv_init(&spa_namespace_cv
, NULL
, CV_DEFAULT
, NULL
);
1821 avl_create(&spa_namespace_avl
, spa_name_compare
, sizeof (spa_t
),
1822 offsetof(spa_t
, spa_avl
));
1824 avl_create(&spa_spare_avl
, spa_spare_compare
, sizeof (spa_aux_t
),
1825 offsetof(spa_aux_t
, aux_avl
));
1827 avl_create(&spa_l2cache_avl
, spa_l2cache_compare
, sizeof (spa_aux_t
),
1828 offsetof(spa_aux_t
, aux_avl
));
1830 spa_mode_global
= mode
;
1833 if (spa_mode_global
!= FREAD
&& dprintf_find_string("watch")) {
1834 struct sigaction sa
;
1836 sa
.sa_flags
= SA_SIGINFO
;
1837 sigemptyset(&sa
.sa_mask
);
1838 sa
.sa_sigaction
= arc_buf_sigsegv
;
1840 if (sigaction(SIGSEGV
, &sa
, NULL
) == -1) {
1841 perror("could not enable watchpoints: "
1842 "sigaction(SIGSEGV, ...) = ");
1853 metaslab_alloc_trace_init();
1858 vdev_cache_stat_init();
1859 vdev_raidz_math_init();
1863 zpool_feature_init();
1876 vdev_cache_stat_fini();
1877 vdev_raidz_math_fini();
1882 metaslab_alloc_trace_fini();
1888 avl_destroy(&spa_namespace_avl
);
1889 avl_destroy(&spa_spare_avl
);
1890 avl_destroy(&spa_l2cache_avl
);
1892 cv_destroy(&spa_namespace_cv
);
1893 mutex_destroy(&spa_namespace_lock
);
1894 mutex_destroy(&spa_spare_lock
);
1895 mutex_destroy(&spa_l2cache_lock
);
1899 * Return whether this pool has slogs. No locking needed.
1900 * It's not a problem if the wrong answer is returned as it's only for
1901 * performance and not correctness
1904 spa_has_slogs(spa_t
*spa
)
1906 return (spa
->spa_log_class
->mc_rotor
!= NULL
);
1910 spa_get_log_state(spa_t
*spa
)
1912 return (spa
->spa_log_state
);
1916 spa_set_log_state(spa_t
*spa
, spa_log_state_t state
)
1918 spa
->spa_log_state
= state
;
1922 spa_is_root(spa_t
*spa
)
1924 return (spa
->spa_is_root
);
1928 spa_writeable(spa_t
*spa
)
1930 return (!!(spa
->spa_mode
& FWRITE
));
1934 * Returns true if there is a pending sync task in any of the current
1935 * syncing txg, the current quiescing txg, or the current open txg.
1938 spa_has_pending_synctask(spa_t
*spa
)
1940 return (!txg_all_lists_empty(&spa
->spa_dsl_pool
->dp_sync_tasks
));
1944 spa_mode(spa_t
*spa
)
1946 return (spa
->spa_mode
);
1950 spa_bootfs(spa_t
*spa
)
1952 return (spa
->spa_bootfs
);
1956 spa_delegation(spa_t
*spa
)
1958 return (spa
->spa_delegation
);
1962 spa_meta_objset(spa_t
*spa
)
1964 return (spa
->spa_meta_objset
);
1968 spa_dedup_checksum(spa_t
*spa
)
1970 return (spa
->spa_dedup_checksum
);
1974 * Reset pool scan stat per scan pass (or reboot).
1977 spa_scan_stat_init(spa_t
*spa
)
1979 /* data not stored on disk */
1980 spa
->spa_scan_pass_start
= gethrestime_sec();
1981 spa
->spa_scan_pass_exam
= 0;
1982 vdev_scan_stat_init(spa
->spa_root_vdev
);
1986 * Get scan stats for zpool status reports
1989 spa_scan_get_stats(spa_t
*spa
, pool_scan_stat_t
*ps
)
1991 dsl_scan_t
*scn
= spa
->spa_dsl_pool
? spa
->spa_dsl_pool
->dp_scan
: NULL
;
1993 if (scn
== NULL
|| scn
->scn_phys
.scn_func
== POOL_SCAN_NONE
)
1994 return (SET_ERROR(ENOENT
));
1995 bzero(ps
, sizeof (pool_scan_stat_t
));
1997 /* data stored on disk */
1998 ps
->pss_func
= scn
->scn_phys
.scn_func
;
1999 ps
->pss_start_time
= scn
->scn_phys
.scn_start_time
;
2000 ps
->pss_end_time
= scn
->scn_phys
.scn_end_time
;
2001 ps
->pss_to_examine
= scn
->scn_phys
.scn_to_examine
;
2002 ps
->pss_examined
= scn
->scn_phys
.scn_examined
;
2003 ps
->pss_to_process
= scn
->scn_phys
.scn_to_process
;
2004 ps
->pss_processed
= scn
->scn_phys
.scn_processed
;
2005 ps
->pss_errors
= scn
->scn_phys
.scn_errors
;
2006 ps
->pss_state
= scn
->scn_phys
.scn_state
;
2008 /* data not stored on disk */
2009 ps
->pss_pass_start
= spa
->spa_scan_pass_start
;
2010 ps
->pss_pass_exam
= spa
->spa_scan_pass_exam
;
2016 spa_debug_enabled(spa_t
*spa
)
2018 return (spa
->spa_debug
);
2022 spa_maxblocksize(spa_t
*spa
)
2024 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_BLOCKS
))
2025 return (SPA_MAXBLOCKSIZE
);
2027 return (SPA_OLD_MAXBLOCKSIZE
);
2031 spa_maxdnodesize(spa_t
*spa
)
2033 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_DNODE
))
2034 return (DNODE_MAX_SIZE
);
2036 return (DNODE_MIN_SIZE
);
2039 #if defined(_KERNEL) && defined(HAVE_SPL)
2040 /* Namespace manipulation */
2041 EXPORT_SYMBOL(spa_lookup
);
2042 EXPORT_SYMBOL(spa_add
);
2043 EXPORT_SYMBOL(spa_remove
);
2044 EXPORT_SYMBOL(spa_next
);
2046 /* Refcount functions */
2047 EXPORT_SYMBOL(spa_open_ref
);
2048 EXPORT_SYMBOL(spa_close
);
2049 EXPORT_SYMBOL(spa_refcount_zero
);
2051 /* Pool configuration lock */
2052 EXPORT_SYMBOL(spa_config_tryenter
);
2053 EXPORT_SYMBOL(spa_config_enter
);
2054 EXPORT_SYMBOL(spa_config_exit
);
2055 EXPORT_SYMBOL(spa_config_held
);
2057 /* Pool vdev add/remove lock */
2058 EXPORT_SYMBOL(spa_vdev_enter
);
2059 EXPORT_SYMBOL(spa_vdev_exit
);
2061 /* Pool vdev state change lock */
2062 EXPORT_SYMBOL(spa_vdev_state_enter
);
2063 EXPORT_SYMBOL(spa_vdev_state_exit
);
2065 /* Accessor functions */
2066 EXPORT_SYMBOL(spa_shutting_down
);
2067 EXPORT_SYMBOL(spa_get_dsl
);
2068 EXPORT_SYMBOL(spa_get_rootblkptr
);
2069 EXPORT_SYMBOL(spa_set_rootblkptr
);
2070 EXPORT_SYMBOL(spa_altroot
);
2071 EXPORT_SYMBOL(spa_sync_pass
);
2072 EXPORT_SYMBOL(spa_name
);
2073 EXPORT_SYMBOL(spa_guid
);
2074 EXPORT_SYMBOL(spa_last_synced_txg
);
2075 EXPORT_SYMBOL(spa_first_txg
);
2076 EXPORT_SYMBOL(spa_syncing_txg
);
2077 EXPORT_SYMBOL(spa_version
);
2078 EXPORT_SYMBOL(spa_state
);
2079 EXPORT_SYMBOL(spa_load_state
);
2080 EXPORT_SYMBOL(spa_freeze_txg
);
2081 EXPORT_SYMBOL(spa_get_asize
);
2082 EXPORT_SYMBOL(spa_get_dspace
);
2083 EXPORT_SYMBOL(spa_update_dspace
);
2084 EXPORT_SYMBOL(spa_deflate
);
2085 EXPORT_SYMBOL(spa_normal_class
);
2086 EXPORT_SYMBOL(spa_log_class
);
2087 EXPORT_SYMBOL(spa_max_replication
);
2088 EXPORT_SYMBOL(spa_prev_software_version
);
2089 EXPORT_SYMBOL(spa_get_failmode
);
2090 EXPORT_SYMBOL(spa_suspended
);
2091 EXPORT_SYMBOL(spa_bootfs
);
2092 EXPORT_SYMBOL(spa_delegation
);
2093 EXPORT_SYMBOL(spa_meta_objset
);
2094 EXPORT_SYMBOL(spa_maxblocksize
);
2095 EXPORT_SYMBOL(spa_maxdnodesize
);
2097 /* Miscellaneous support routines */
2098 EXPORT_SYMBOL(spa_rename
);
2099 EXPORT_SYMBOL(spa_guid_exists
);
2100 EXPORT_SYMBOL(spa_strdup
);
2101 EXPORT_SYMBOL(spa_strfree
);
2102 EXPORT_SYMBOL(spa_get_random
);
2103 EXPORT_SYMBOL(spa_generate_guid
);
2104 EXPORT_SYMBOL(snprintf_blkptr
);
2105 EXPORT_SYMBOL(spa_freeze
);
2106 EXPORT_SYMBOL(spa_upgrade
);
2107 EXPORT_SYMBOL(spa_evict_all
);
2108 EXPORT_SYMBOL(spa_lookup_by_guid
);
2109 EXPORT_SYMBOL(spa_has_spare
);
2110 EXPORT_SYMBOL(dva_get_dsize_sync
);
2111 EXPORT_SYMBOL(bp_get_dsize_sync
);
2112 EXPORT_SYMBOL(bp_get_dsize
);
2113 EXPORT_SYMBOL(spa_has_slogs
);
2114 EXPORT_SYMBOL(spa_is_root
);
2115 EXPORT_SYMBOL(spa_writeable
);
2116 EXPORT_SYMBOL(spa_mode
);
2117 EXPORT_SYMBOL(spa_namespace_lock
);
2120 module_param(zfs_flags
, uint
, 0644);
2121 MODULE_PARM_DESC(zfs_flags
, "Set additional debugging flags");
2123 module_param(zfs_recover
, int, 0644);
2124 MODULE_PARM_DESC(zfs_recover
, "Set to attempt to recover from fatal errors");
2126 module_param(zfs_free_leak_on_eio
, int, 0644);
2127 MODULE_PARM_DESC(zfs_free_leak_on_eio
,
2128 "Set to ignore IO errors during free and permanently leak the space");
2130 module_param(zfs_deadman_synctime_ms
, ulong
, 0644);
2131 MODULE_PARM_DESC(zfs_deadman_synctime_ms
, "Expiration time in milliseconds");
2133 module_param(zfs_deadman_checktime_ms
, ulong
, 0644);
2134 MODULE_PARM_DESC(zfs_deadman_checktime_ms
,
2135 "Dead I/O check interval in milliseconds");
2137 module_param(zfs_deadman_enabled
, int, 0644);
2138 MODULE_PARM_DESC(zfs_deadman_enabled
, "Enable deadman timer");
2140 module_param(spa_asize_inflation
, int, 0644);
2141 MODULE_PARM_DESC(spa_asize_inflation
,
2142 "SPA size estimate multiplication factor");
2144 module_param(spa_slop_shift
, int, 0644);
2145 MODULE_PARM_DESC(spa_slop_shift
, "Reserved free space in pool");