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, 2019 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.
27 * Copyright (c) 2017 Datto Inc.
28 * Copyright (c) 2017, Intel Corporation.
29 * Copyright (c) 2019, loli10K <ezomori.nozomu@gmail.com>. All rights reserved.
32 #include <sys/zfs_context.h>
33 #include <sys/spa_impl.h>
35 #include <sys/zio_checksum.h>
36 #include <sys/zio_compress.h>
38 #include <sys/dmu_tx.h>
41 #include <sys/vdev_impl.h>
42 #include <sys/vdev_initialize.h>
43 #include <sys/vdev_trim.h>
44 #include <sys/vdev_file.h>
45 #include <sys/vdev_raidz.h>
46 #include <sys/metaslab.h>
47 #include <sys/uberblock_impl.h>
50 #include <sys/unique.h>
51 #include <sys/dsl_pool.h>
52 #include <sys/dsl_dir.h>
53 #include <sys/dsl_prop.h>
54 #include <sys/fm/util.h>
55 #include <sys/dsl_scan.h>
56 #include <sys/fs/zfs.h>
57 #include <sys/metaslab_impl.h>
60 #include <sys/kstat.h>
62 #include <sys/btree.h>
63 #include <sys/zfeature.h>
65 #include <sys/zstd/zstd.h>
70 * There are three basic locks for managing spa_t structures:
72 * spa_namespace_lock (global mutex)
74 * This lock must be acquired to do any of the following:
76 * - Lookup a spa_t by name
77 * - Add or remove a spa_t from the namespace
78 * - Increase spa_refcount from non-zero
79 * - Check if spa_refcount is zero
81 * - add/remove/attach/detach devices
82 * - Held for the duration of create/destroy/import/export
84 * It does not need to handle recursion. A create or destroy may
85 * reference objects (files or zvols) in other pools, but by
86 * definition they must have an existing reference, and will never need
87 * to lookup a spa_t by name.
89 * spa_refcount (per-spa zfs_refcount_t protected by mutex)
91 * This reference count keep track of any active users of the spa_t. The
92 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
93 * the refcount is never really 'zero' - opening a pool implicitly keeps
94 * some references in the DMU. Internally we check against spa_minref, but
95 * present the image of a zero/non-zero value to consumers.
97 * spa_config_lock[] (per-spa array of rwlocks)
99 * This protects the spa_t from config changes, and must be held in
100 * the following circumstances:
102 * - RW_READER to perform I/O to the spa
103 * - RW_WRITER to change the vdev config
105 * The locking order is fairly straightforward:
107 * spa_namespace_lock -> spa_refcount
109 * The namespace lock must be acquired to increase the refcount from 0
110 * or to check if it is zero.
112 * spa_refcount -> spa_config_lock[]
114 * There must be at least one valid reference on the spa_t to acquire
117 * spa_namespace_lock -> spa_config_lock[]
119 * The namespace lock must always be taken before the config lock.
122 * The spa_namespace_lock can be acquired directly and is globally visible.
124 * The namespace is manipulated using the following functions, all of which
125 * require the spa_namespace_lock to be held.
127 * spa_lookup() Lookup a spa_t by name.
129 * spa_add() Create a new spa_t in the namespace.
131 * spa_remove() Remove a spa_t from the namespace. This also
132 * frees up any memory associated with the spa_t.
134 * spa_next() Returns the next spa_t in the system, or the
135 * first if NULL is passed.
137 * spa_evict_all() Shutdown and remove all spa_t structures in
140 * spa_guid_exists() Determine whether a pool/device guid exists.
142 * The spa_refcount is manipulated using the following functions:
144 * spa_open_ref() Adds a reference to the given spa_t. Must be
145 * called with spa_namespace_lock held if the
146 * refcount is currently zero.
148 * spa_close() Remove a reference from the spa_t. This will
149 * not free the spa_t or remove it from the
150 * namespace. No locking is required.
152 * spa_refcount_zero() Returns true if the refcount is currently
153 * zero. Must be called with spa_namespace_lock
156 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
157 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
158 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
160 * To read the configuration, it suffices to hold one of these locks as reader.
161 * To modify the configuration, you must hold all locks as writer. To modify
162 * vdev state without altering the vdev tree's topology (e.g. online/offline),
163 * you must hold SCL_STATE and SCL_ZIO as writer.
165 * We use these distinct config locks to avoid recursive lock entry.
166 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
167 * block allocations (SCL_ALLOC), which may require reading space maps
168 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
170 * The spa config locks cannot be normal rwlocks because we need the
171 * ability to hand off ownership. For example, SCL_ZIO is acquired
172 * by the issuing thread and later released by an interrupt thread.
173 * They do, however, obey the usual write-wanted semantics to prevent
174 * writer (i.e. system administrator) starvation.
176 * The lock acquisition rules are as follows:
179 * Protects changes to the vdev tree topology, such as vdev
180 * add/remove/attach/detach. Protects the dirty config list
181 * (spa_config_dirty_list) and the set of spares and l2arc devices.
184 * Protects changes to pool state and vdev state, such as vdev
185 * online/offline/fault/degrade/clear. Protects the dirty state list
186 * (spa_state_dirty_list) and global pool state (spa_state).
189 * Protects changes to metaslab groups and classes.
190 * Held as reader by metaslab_alloc() and metaslab_claim().
193 * Held by bp-level zios (those which have no io_vd upon entry)
194 * to prevent changes to the vdev tree. The bp-level zio implicitly
195 * protects all of its vdev child zios, which do not hold SCL_ZIO.
198 * Protects changes to metaslab groups and classes.
199 * Held as reader by metaslab_free(). SCL_FREE is distinct from
200 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
201 * blocks in zio_done() while another i/o that holds either
202 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
205 * Held as reader to prevent changes to the vdev tree during trivial
206 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
207 * other locks, and lower than all of them, to ensure that it's safe
208 * to acquire regardless of caller context.
210 * In addition, the following rules apply:
212 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
213 * The lock ordering is SCL_CONFIG > spa_props_lock.
215 * (b) I/O operations on leaf vdevs. For any zio operation that takes
216 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
217 * or zio_write_phys() -- the caller must ensure that the config cannot
218 * cannot change in the interim, and that the vdev cannot be reopened.
219 * SCL_STATE as reader suffices for both.
221 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
223 * spa_vdev_enter() Acquire the namespace lock and the config lock
226 * spa_vdev_exit() Release the config lock, wait for all I/O
227 * to complete, sync the updated configs to the
228 * cache, and release the namespace lock.
230 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
231 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
232 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
235 static avl_tree_t spa_namespace_avl
;
236 kmutex_t spa_namespace_lock
;
237 static kcondvar_t spa_namespace_cv
;
238 int spa_max_replication_override
= SPA_DVAS_PER_BP
;
240 static kmutex_t spa_spare_lock
;
241 static avl_tree_t spa_spare_avl
;
242 static kmutex_t spa_l2cache_lock
;
243 static avl_tree_t spa_l2cache_avl
;
245 kmem_cache_t
*spa_buffer_pool
;
246 spa_mode_t spa_mode_global
= SPA_MODE_UNINIT
;
250 * Everything except dprintf, set_error, spa, and indirect_remap is on
251 * by default in debug builds.
253 int zfs_flags
= ~(ZFS_DEBUG_DPRINTF
| ZFS_DEBUG_SET_ERROR
|
254 ZFS_DEBUG_INDIRECT_REMAP
);
260 * zfs_recover can be set to nonzero to attempt to recover from
261 * otherwise-fatal errors, typically caused by on-disk corruption. When
262 * set, calls to zfs_panic_recover() will turn into warning messages.
263 * This should only be used as a last resort, as it typically results
264 * in leaked space, or worse.
266 int zfs_recover
= B_FALSE
;
269 * If destroy encounters an EIO while reading metadata (e.g. indirect
270 * blocks), space referenced by the missing metadata can not be freed.
271 * Normally this causes the background destroy to become "stalled", as
272 * it is unable to make forward progress. While in this stalled state,
273 * all remaining space to free from the error-encountering filesystem is
274 * "temporarily leaked". Set this flag to cause it to ignore the EIO,
275 * permanently leak the space from indirect blocks that can not be read,
276 * and continue to free everything else that it can.
278 * The default, "stalling" behavior is useful if the storage partially
279 * fails (i.e. some but not all i/os fail), and then later recovers. In
280 * this case, we will be able to continue pool operations while it is
281 * partially failed, and when it recovers, we can continue to free the
282 * space, with no leaks. However, note that this case is actually
285 * Typically pools either (a) fail completely (but perhaps temporarily,
286 * e.g. a top-level vdev going offline), or (b) have localized,
287 * permanent errors (e.g. disk returns the wrong data due to bit flip or
288 * firmware bug). In case (a), this setting does not matter because the
289 * pool will be suspended and the sync thread will not be able to make
290 * forward progress regardless. In case (b), because the error is
291 * permanent, the best we can do is leak the minimum amount of space,
292 * which is what setting this flag will do. Therefore, it is reasonable
293 * for this flag to normally be set, but we chose the more conservative
294 * approach of not setting it, so that there is no possibility of
295 * leaking space in the "partial temporary" failure case.
297 int zfs_free_leak_on_eio
= B_FALSE
;
300 * Expiration time in milliseconds. This value has two meanings. First it is
301 * used to determine when the spa_deadman() logic should fire. By default the
302 * spa_deadman() will fire if spa_sync() has not completed in 600 seconds.
303 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
304 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
305 * in one of three behaviors controlled by zfs_deadman_failmode.
307 unsigned long zfs_deadman_synctime_ms
= 600000UL;
310 * This value controls the maximum amount of time zio_wait() will block for an
311 * outstanding IO. By default this is 300 seconds at which point the "hung"
312 * behavior will be applied as described for zfs_deadman_synctime_ms.
314 unsigned long zfs_deadman_ziotime_ms
= 300000UL;
317 * Check time in milliseconds. This defines the frequency at which we check
320 unsigned long zfs_deadman_checktime_ms
= 60000UL;
323 * By default the deadman is enabled.
325 int zfs_deadman_enabled
= 1;
328 * Controls the behavior of the deadman when it detects a "hung" I/O.
329 * Valid values are zfs_deadman_failmode=<wait|continue|panic>.
331 * wait - Wait for the "hung" I/O (default)
332 * continue - Attempt to recover from a "hung" I/O
333 * panic - Panic the system
335 char *zfs_deadman_failmode
= "wait";
338 * The worst case is single-sector max-parity RAID-Z blocks, in which
339 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
340 * times the size; so just assume that. Add to this the fact that
341 * we can have up to 3 DVAs per bp, and one more factor of 2 because
342 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together,
344 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
346 int spa_asize_inflation
= 24;
349 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
350 * the pool to be consumed. This ensures that we don't run the pool
351 * completely out of space, due to unaccounted changes (e.g. to the MOS).
352 * It also limits the worst-case time to allocate space. If we have
353 * less than this amount of free space, most ZPL operations (e.g. write,
354 * create) will return ENOSPC.
356 * Certain operations (e.g. file removal, most administrative actions) can
357 * use half the slop space. They will only return ENOSPC if less than half
358 * the slop space is free. Typically, once the pool has less than the slop
359 * space free, the user will use these operations to free up space in the pool.
360 * These are the operations that call dsl_pool_adjustedsize() with the netfree
361 * argument set to TRUE.
363 * Operations that are almost guaranteed to free up space in the absence of
364 * a pool checkpoint can use up to three quarters of the slop space
367 * A very restricted set of operations are always permitted, regardless of
368 * the amount of free space. These are the operations that call
369 * dsl_sync_task(ZFS_SPACE_CHECK_NONE). If these operations result in a net
370 * increase in the amount of space used, it is possible to run the pool
371 * completely out of space, causing it to be permanently read-only.
373 * Note that on very small pools, the slop space will be larger than
374 * 3.2%, in an effort to have it be at least spa_min_slop (128MB),
375 * but we never allow it to be more than half the pool size.
377 * See also the comments in zfs_space_check_t.
379 int spa_slop_shift
= 5;
380 uint64_t spa_min_slop
= 128 * 1024 * 1024;
381 int spa_allocators
= 4;
386 spa_load_failed(spa_t
*spa
, const char *fmt
, ...)
392 (void) vsnprintf(buf
, sizeof (buf
), fmt
, adx
);
395 zfs_dbgmsg("spa_load(%s, config %s): FAILED: %s", spa
->spa_name
,
396 spa
->spa_trust_config
? "trusted" : "untrusted", buf
);
401 spa_load_note(spa_t
*spa
, const char *fmt
, ...)
407 (void) vsnprintf(buf
, sizeof (buf
), fmt
, adx
);
410 zfs_dbgmsg("spa_load(%s, config %s): %s", spa
->spa_name
,
411 spa
->spa_trust_config
? "trusted" : "untrusted", buf
);
415 * By default dedup and user data indirects land in the special class
417 int zfs_ddt_data_is_special
= B_TRUE
;
418 int zfs_user_indirect_is_special
= B_TRUE
;
421 * The percentage of special class final space reserved for metadata only.
422 * Once we allocate 100 - zfs_special_class_metadata_reserve_pct we only
423 * let metadata into the class.
425 int zfs_special_class_metadata_reserve_pct
= 25;
428 * ==========================================================================
430 * ==========================================================================
433 spa_config_lock_init(spa_t
*spa
)
435 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
436 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
437 mutex_init(&scl
->scl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
438 cv_init(&scl
->scl_cv
, NULL
, CV_DEFAULT
, NULL
);
439 zfs_refcount_create_untracked(&scl
->scl_count
);
440 scl
->scl_writer
= NULL
;
441 scl
->scl_write_wanted
= 0;
446 spa_config_lock_destroy(spa_t
*spa
)
448 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
449 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
450 mutex_destroy(&scl
->scl_lock
);
451 cv_destroy(&scl
->scl_cv
);
452 zfs_refcount_destroy(&scl
->scl_count
);
453 ASSERT(scl
->scl_writer
== NULL
);
454 ASSERT(scl
->scl_write_wanted
== 0);
459 spa_config_tryenter(spa_t
*spa
, int locks
, void *tag
, krw_t rw
)
461 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
462 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
463 if (!(locks
& (1 << i
)))
465 mutex_enter(&scl
->scl_lock
);
466 if (rw
== RW_READER
) {
467 if (scl
->scl_writer
|| scl
->scl_write_wanted
) {
468 mutex_exit(&scl
->scl_lock
);
469 spa_config_exit(spa
, locks
& ((1 << i
) - 1),
474 ASSERT(scl
->scl_writer
!= curthread
);
475 if (!zfs_refcount_is_zero(&scl
->scl_count
)) {
476 mutex_exit(&scl
->scl_lock
);
477 spa_config_exit(spa
, locks
& ((1 << i
) - 1),
481 scl
->scl_writer
= curthread
;
483 (void) zfs_refcount_add(&scl
->scl_count
, tag
);
484 mutex_exit(&scl
->scl_lock
);
490 spa_config_enter(spa_t
*spa
, int locks
, const void *tag
, krw_t rw
)
494 ASSERT3U(SCL_LOCKS
, <, sizeof (wlocks_held
) * NBBY
);
496 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
497 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
498 if (scl
->scl_writer
== curthread
)
499 wlocks_held
|= (1 << i
);
500 if (!(locks
& (1 << i
)))
502 mutex_enter(&scl
->scl_lock
);
503 if (rw
== RW_READER
) {
504 while (scl
->scl_writer
|| scl
->scl_write_wanted
) {
505 cv_wait(&scl
->scl_cv
, &scl
->scl_lock
);
508 ASSERT(scl
->scl_writer
!= curthread
);
509 while (!zfs_refcount_is_zero(&scl
->scl_count
)) {
510 scl
->scl_write_wanted
++;
511 cv_wait(&scl
->scl_cv
, &scl
->scl_lock
);
512 scl
->scl_write_wanted
--;
514 scl
->scl_writer
= curthread
;
516 (void) zfs_refcount_add(&scl
->scl_count
, tag
);
517 mutex_exit(&scl
->scl_lock
);
519 ASSERT3U(wlocks_held
, <=, locks
);
523 spa_config_exit(spa_t
*spa
, int locks
, const void *tag
)
525 for (int i
= SCL_LOCKS
- 1; i
>= 0; i
--) {
526 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
527 if (!(locks
& (1 << i
)))
529 mutex_enter(&scl
->scl_lock
);
530 ASSERT(!zfs_refcount_is_zero(&scl
->scl_count
));
531 if (zfs_refcount_remove(&scl
->scl_count
, tag
) == 0) {
532 ASSERT(scl
->scl_writer
== NULL
||
533 scl
->scl_writer
== curthread
);
534 scl
->scl_writer
= NULL
; /* OK in either case */
535 cv_broadcast(&scl
->scl_cv
);
537 mutex_exit(&scl
->scl_lock
);
542 spa_config_held(spa_t
*spa
, int locks
, krw_t rw
)
546 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
547 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
548 if (!(locks
& (1 << i
)))
550 if ((rw
== RW_READER
&&
551 !zfs_refcount_is_zero(&scl
->scl_count
)) ||
552 (rw
== RW_WRITER
&& scl
->scl_writer
== curthread
))
553 locks_held
|= 1 << i
;
560 * ==========================================================================
561 * SPA namespace functions
562 * ==========================================================================
566 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
567 * Returns NULL if no matching spa_t is found.
570 spa_lookup(const char *name
)
572 static spa_t search
; /* spa_t is large; don't allocate on stack */
577 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
579 (void) strlcpy(search
.spa_name
, name
, sizeof (search
.spa_name
));
582 * If it's a full dataset name, figure out the pool name and
585 cp
= strpbrk(search
.spa_name
, "/@#");
589 spa
= avl_find(&spa_namespace_avl
, &search
, &where
);
595 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
596 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
597 * looking for potentially hung I/Os.
600 spa_deadman(void *arg
)
604 /* Disable the deadman if the pool is suspended. */
605 if (spa_suspended(spa
))
608 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
609 (gethrtime() - spa
->spa_sync_starttime
) / NANOSEC
,
610 ++spa
->spa_deadman_calls
);
611 if (zfs_deadman_enabled
)
612 vdev_deadman(spa
->spa_root_vdev
, FTAG
);
614 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_delay_taskq
,
615 spa_deadman
, spa
, TQ_SLEEP
, ddi_get_lbolt() +
616 MSEC_TO_TICK(zfs_deadman_checktime_ms
));
620 spa_log_sm_sort_by_txg(const void *va
, const void *vb
)
622 const spa_log_sm_t
*a
= va
;
623 const spa_log_sm_t
*b
= vb
;
625 return (TREE_CMP(a
->sls_txg
, b
->sls_txg
));
629 * Create an uninitialized spa_t with the given name. Requires
630 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
631 * exist by calling spa_lookup() first.
634 spa_add(const char *name
, nvlist_t
*config
, const char *altroot
)
637 spa_config_dirent_t
*dp
;
639 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
641 spa
= kmem_zalloc(sizeof (spa_t
), KM_SLEEP
);
643 mutex_init(&spa
->spa_async_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
644 mutex_init(&spa
->spa_errlist_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
645 mutex_init(&spa
->spa_errlog_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
646 mutex_init(&spa
->spa_evicting_os_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
647 mutex_init(&spa
->spa_history_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
648 mutex_init(&spa
->spa_proc_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
649 mutex_init(&spa
->spa_props_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
650 mutex_init(&spa
->spa_cksum_tmpls_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
651 mutex_init(&spa
->spa_scrub_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
652 mutex_init(&spa
->spa_suspend_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
653 mutex_init(&spa
->spa_vdev_top_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
654 mutex_init(&spa
->spa_feat_stats_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
655 mutex_init(&spa
->spa_flushed_ms_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
656 mutex_init(&spa
->spa_activities_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
658 cv_init(&spa
->spa_async_cv
, NULL
, CV_DEFAULT
, NULL
);
659 cv_init(&spa
->spa_evicting_os_cv
, NULL
, CV_DEFAULT
, NULL
);
660 cv_init(&spa
->spa_proc_cv
, NULL
, CV_DEFAULT
, NULL
);
661 cv_init(&spa
->spa_scrub_io_cv
, NULL
, CV_DEFAULT
, NULL
);
662 cv_init(&spa
->spa_suspend_cv
, NULL
, CV_DEFAULT
, NULL
);
663 cv_init(&spa
->spa_activities_cv
, NULL
, CV_DEFAULT
, NULL
);
664 cv_init(&spa
->spa_waiters_cv
, NULL
, CV_DEFAULT
, NULL
);
666 for (int t
= 0; t
< TXG_SIZE
; t
++)
667 bplist_create(&spa
->spa_free_bplist
[t
]);
669 (void) strlcpy(spa
->spa_name
, name
, sizeof (spa
->spa_name
));
670 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
671 spa
->spa_freeze_txg
= UINT64_MAX
;
672 spa
->spa_final_txg
= UINT64_MAX
;
673 spa
->spa_load_max_txg
= UINT64_MAX
;
675 spa
->spa_proc_state
= SPA_PROC_NONE
;
676 spa
->spa_trust_config
= B_TRUE
;
677 spa
->spa_hostid
= zone_get_hostid(NULL
);
679 spa
->spa_deadman_synctime
= MSEC2NSEC(zfs_deadman_synctime_ms
);
680 spa
->spa_deadman_ziotime
= MSEC2NSEC(zfs_deadman_ziotime_ms
);
681 spa_set_deadman_failmode(spa
, zfs_deadman_failmode
);
683 zfs_refcount_create(&spa
->spa_refcount
);
684 spa_config_lock_init(spa
);
687 avl_add(&spa_namespace_avl
, spa
);
690 * Set the alternate root, if there is one.
693 spa
->spa_root
= spa_strdup(altroot
);
695 spa
->spa_alloc_count
= spa_allocators
;
696 spa
->spa_alloc_locks
= kmem_zalloc(spa
->spa_alloc_count
*
697 sizeof (kmutex_t
), KM_SLEEP
);
698 spa
->spa_alloc_trees
= kmem_zalloc(spa
->spa_alloc_count
*
699 sizeof (avl_tree_t
), KM_SLEEP
);
700 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++) {
701 mutex_init(&spa
->spa_alloc_locks
[i
], NULL
, MUTEX_DEFAULT
, NULL
);
702 avl_create(&spa
->spa_alloc_trees
[i
], zio_bookmark_compare
,
703 sizeof (zio_t
), offsetof(zio_t
, io_alloc_node
));
705 avl_create(&spa
->spa_metaslabs_by_flushed
, metaslab_sort_by_flushed
,
706 sizeof (metaslab_t
), offsetof(metaslab_t
, ms_spa_txg_node
));
707 avl_create(&spa
->spa_sm_logs_by_txg
, spa_log_sm_sort_by_txg
,
708 sizeof (spa_log_sm_t
), offsetof(spa_log_sm_t
, sls_node
));
709 list_create(&spa
->spa_log_summary
, sizeof (log_summary_entry_t
),
710 offsetof(log_summary_entry_t
, lse_node
));
713 * Every pool starts with the default cachefile
715 list_create(&spa
->spa_config_list
, sizeof (spa_config_dirent_t
),
716 offsetof(spa_config_dirent_t
, scd_link
));
718 dp
= kmem_zalloc(sizeof (spa_config_dirent_t
), KM_SLEEP
);
719 dp
->scd_path
= altroot
? NULL
: spa_strdup(spa_config_path
);
720 list_insert_head(&spa
->spa_config_list
, dp
);
722 VERIFY(nvlist_alloc(&spa
->spa_load_info
, NV_UNIQUE_NAME
,
725 if (config
!= NULL
) {
728 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_FEATURES_FOR_READ
,
730 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
,
734 VERIFY(nvlist_dup(config
, &spa
->spa_config
, 0) == 0);
737 if (spa
->spa_label_features
== NULL
) {
738 VERIFY(nvlist_alloc(&spa
->spa_label_features
, NV_UNIQUE_NAME
,
742 spa
->spa_min_ashift
= INT_MAX
;
743 spa
->spa_max_ashift
= 0;
744 spa
->spa_min_alloc
= INT_MAX
;
746 /* Reset cached value */
747 spa
->spa_dedup_dspace
= ~0ULL;
750 * As a pool is being created, treat all features as disabled by
751 * setting SPA_FEATURE_DISABLED for all entries in the feature
754 for (int i
= 0; i
< SPA_FEATURES
; i
++) {
755 spa
->spa_feat_refcount_cache
[i
] = SPA_FEATURE_DISABLED
;
758 list_create(&spa
->spa_leaf_list
, sizeof (vdev_t
),
759 offsetof(vdev_t
, vdev_leaf_node
));
765 * Removes a spa_t from the namespace, freeing up any memory used. Requires
766 * spa_namespace_lock. This is called only after the spa_t has been closed and
770 spa_remove(spa_t
*spa
)
772 spa_config_dirent_t
*dp
;
774 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
775 ASSERT(spa_state(spa
) == POOL_STATE_UNINITIALIZED
);
776 ASSERT3U(zfs_refcount_count(&spa
->spa_refcount
), ==, 0);
777 ASSERT0(spa
->spa_waiters
);
779 nvlist_free(spa
->spa_config_splitting
);
781 avl_remove(&spa_namespace_avl
, spa
);
782 cv_broadcast(&spa_namespace_cv
);
785 spa_strfree(spa
->spa_root
);
787 while ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
788 list_remove(&spa
->spa_config_list
, dp
);
789 if (dp
->scd_path
!= NULL
)
790 spa_strfree(dp
->scd_path
);
791 kmem_free(dp
, sizeof (spa_config_dirent_t
));
794 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++) {
795 avl_destroy(&spa
->spa_alloc_trees
[i
]);
796 mutex_destroy(&spa
->spa_alloc_locks
[i
]);
798 kmem_free(spa
->spa_alloc_locks
, spa
->spa_alloc_count
*
800 kmem_free(spa
->spa_alloc_trees
, spa
->spa_alloc_count
*
801 sizeof (avl_tree_t
));
803 avl_destroy(&spa
->spa_metaslabs_by_flushed
);
804 avl_destroy(&spa
->spa_sm_logs_by_txg
);
805 list_destroy(&spa
->spa_log_summary
);
806 list_destroy(&spa
->spa_config_list
);
807 list_destroy(&spa
->spa_leaf_list
);
809 nvlist_free(spa
->spa_label_features
);
810 nvlist_free(spa
->spa_load_info
);
811 nvlist_free(spa
->spa_feat_stats
);
812 spa_config_set(spa
, NULL
);
814 zfs_refcount_destroy(&spa
->spa_refcount
);
816 spa_stats_destroy(spa
);
817 spa_config_lock_destroy(spa
);
819 for (int t
= 0; t
< TXG_SIZE
; t
++)
820 bplist_destroy(&spa
->spa_free_bplist
[t
]);
822 zio_checksum_templates_free(spa
);
824 cv_destroy(&spa
->spa_async_cv
);
825 cv_destroy(&spa
->spa_evicting_os_cv
);
826 cv_destroy(&spa
->spa_proc_cv
);
827 cv_destroy(&spa
->spa_scrub_io_cv
);
828 cv_destroy(&spa
->spa_suspend_cv
);
829 cv_destroy(&spa
->spa_activities_cv
);
830 cv_destroy(&spa
->spa_waiters_cv
);
832 mutex_destroy(&spa
->spa_flushed_ms_lock
);
833 mutex_destroy(&spa
->spa_async_lock
);
834 mutex_destroy(&spa
->spa_errlist_lock
);
835 mutex_destroy(&spa
->spa_errlog_lock
);
836 mutex_destroy(&spa
->spa_evicting_os_lock
);
837 mutex_destroy(&spa
->spa_history_lock
);
838 mutex_destroy(&spa
->spa_proc_lock
);
839 mutex_destroy(&spa
->spa_props_lock
);
840 mutex_destroy(&spa
->spa_cksum_tmpls_lock
);
841 mutex_destroy(&spa
->spa_scrub_lock
);
842 mutex_destroy(&spa
->spa_suspend_lock
);
843 mutex_destroy(&spa
->spa_vdev_top_lock
);
844 mutex_destroy(&spa
->spa_feat_stats_lock
);
845 mutex_destroy(&spa
->spa_activities_lock
);
847 kmem_free(spa
, sizeof (spa_t
));
851 * Given a pool, return the next pool in the namespace, or NULL if there is
852 * none. If 'prev' is NULL, return the first pool.
855 spa_next(spa_t
*prev
)
857 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
860 return (AVL_NEXT(&spa_namespace_avl
, prev
));
862 return (avl_first(&spa_namespace_avl
));
866 * ==========================================================================
867 * SPA refcount functions
868 * ==========================================================================
872 * Add a reference to the given spa_t. Must have at least one reference, or
873 * have the namespace lock held.
876 spa_open_ref(spa_t
*spa
, void *tag
)
878 ASSERT(zfs_refcount_count(&spa
->spa_refcount
) >= spa
->spa_minref
||
879 MUTEX_HELD(&spa_namespace_lock
));
880 (void) zfs_refcount_add(&spa
->spa_refcount
, tag
);
884 * Remove a reference to the given spa_t. Must have at least one reference, or
885 * have the namespace lock held.
888 spa_close(spa_t
*spa
, void *tag
)
890 ASSERT(zfs_refcount_count(&spa
->spa_refcount
) > spa
->spa_minref
||
891 MUTEX_HELD(&spa_namespace_lock
));
892 (void) zfs_refcount_remove(&spa
->spa_refcount
, tag
);
896 * Remove a reference to the given spa_t held by a dsl dir that is
897 * being asynchronously released. Async releases occur from a taskq
898 * performing eviction of dsl datasets and dirs. The namespace lock
899 * isn't held and the hold by the object being evicted may contribute to
900 * spa_minref (e.g. dataset or directory released during pool export),
901 * so the asserts in spa_close() do not apply.
904 spa_async_close(spa_t
*spa
, void *tag
)
906 (void) zfs_refcount_remove(&spa
->spa_refcount
, tag
);
910 * Check to see if the spa refcount is zero. Must be called with
911 * spa_namespace_lock held. We really compare against spa_minref, which is the
912 * number of references acquired when opening a pool
915 spa_refcount_zero(spa_t
*spa
)
917 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
919 return (zfs_refcount_count(&spa
->spa_refcount
) == spa
->spa_minref
);
923 * ==========================================================================
924 * SPA spare and l2cache tracking
925 * ==========================================================================
929 * Hot spares and cache devices are tracked using the same code below,
930 * for 'auxiliary' devices.
933 typedef struct spa_aux
{
941 spa_aux_compare(const void *a
, const void *b
)
943 const spa_aux_t
*sa
= (const spa_aux_t
*)a
;
944 const spa_aux_t
*sb
= (const spa_aux_t
*)b
;
946 return (TREE_CMP(sa
->aux_guid
, sb
->aux_guid
));
950 spa_aux_add(vdev_t
*vd
, avl_tree_t
*avl
)
956 search
.aux_guid
= vd
->vdev_guid
;
957 if ((aux
= avl_find(avl
, &search
, &where
)) != NULL
) {
960 aux
= kmem_zalloc(sizeof (spa_aux_t
), KM_SLEEP
);
961 aux
->aux_guid
= vd
->vdev_guid
;
963 avl_insert(avl
, aux
, where
);
968 spa_aux_remove(vdev_t
*vd
, avl_tree_t
*avl
)
974 search
.aux_guid
= vd
->vdev_guid
;
975 aux
= avl_find(avl
, &search
, &where
);
979 if (--aux
->aux_count
== 0) {
980 avl_remove(avl
, aux
);
981 kmem_free(aux
, sizeof (spa_aux_t
));
982 } else if (aux
->aux_pool
== spa_guid(vd
->vdev_spa
)) {
983 aux
->aux_pool
= 0ULL;
988 spa_aux_exists(uint64_t guid
, uint64_t *pool
, int *refcnt
, avl_tree_t
*avl
)
990 spa_aux_t search
, *found
;
992 search
.aux_guid
= guid
;
993 found
= avl_find(avl
, &search
, NULL
);
997 *pool
= found
->aux_pool
;
1004 *refcnt
= found
->aux_count
;
1009 return (found
!= NULL
);
1013 spa_aux_activate(vdev_t
*vd
, avl_tree_t
*avl
)
1015 spa_aux_t search
, *found
;
1018 search
.aux_guid
= vd
->vdev_guid
;
1019 found
= avl_find(avl
, &search
, &where
);
1020 ASSERT(found
!= NULL
);
1021 ASSERT(found
->aux_pool
== 0ULL);
1023 found
->aux_pool
= spa_guid(vd
->vdev_spa
);
1027 * Spares are tracked globally due to the following constraints:
1029 * - A spare may be part of multiple pools.
1030 * - A spare may be added to a pool even if it's actively in use within
1032 * - A spare in use in any pool can only be the source of a replacement if
1033 * the target is a spare in the same pool.
1035 * We keep track of all spares on the system through the use of a reference
1036 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
1037 * spare, then we bump the reference count in the AVL tree. In addition, we set
1038 * the 'vdev_isspare' member to indicate that the device is a spare (active or
1039 * inactive). When a spare is made active (used to replace a device in the
1040 * pool), we also keep track of which pool its been made a part of.
1042 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
1043 * called under the spa_namespace lock as part of vdev reconfiguration. The
1044 * separate spare lock exists for the status query path, which does not need to
1045 * be completely consistent with respect to other vdev configuration changes.
1049 spa_spare_compare(const void *a
, const void *b
)
1051 return (spa_aux_compare(a
, b
));
1055 spa_spare_add(vdev_t
*vd
)
1057 mutex_enter(&spa_spare_lock
);
1058 ASSERT(!vd
->vdev_isspare
);
1059 spa_aux_add(vd
, &spa_spare_avl
);
1060 vd
->vdev_isspare
= B_TRUE
;
1061 mutex_exit(&spa_spare_lock
);
1065 spa_spare_remove(vdev_t
*vd
)
1067 mutex_enter(&spa_spare_lock
);
1068 ASSERT(vd
->vdev_isspare
);
1069 spa_aux_remove(vd
, &spa_spare_avl
);
1070 vd
->vdev_isspare
= B_FALSE
;
1071 mutex_exit(&spa_spare_lock
);
1075 spa_spare_exists(uint64_t guid
, uint64_t *pool
, int *refcnt
)
1079 mutex_enter(&spa_spare_lock
);
1080 found
= spa_aux_exists(guid
, pool
, refcnt
, &spa_spare_avl
);
1081 mutex_exit(&spa_spare_lock
);
1087 spa_spare_activate(vdev_t
*vd
)
1089 mutex_enter(&spa_spare_lock
);
1090 ASSERT(vd
->vdev_isspare
);
1091 spa_aux_activate(vd
, &spa_spare_avl
);
1092 mutex_exit(&spa_spare_lock
);
1096 * Level 2 ARC devices are tracked globally for the same reasons as spares.
1097 * Cache devices currently only support one pool per cache device, and so
1098 * for these devices the aux reference count is currently unused beyond 1.
1102 spa_l2cache_compare(const void *a
, const void *b
)
1104 return (spa_aux_compare(a
, b
));
1108 spa_l2cache_add(vdev_t
*vd
)
1110 mutex_enter(&spa_l2cache_lock
);
1111 ASSERT(!vd
->vdev_isl2cache
);
1112 spa_aux_add(vd
, &spa_l2cache_avl
);
1113 vd
->vdev_isl2cache
= B_TRUE
;
1114 mutex_exit(&spa_l2cache_lock
);
1118 spa_l2cache_remove(vdev_t
*vd
)
1120 mutex_enter(&spa_l2cache_lock
);
1121 ASSERT(vd
->vdev_isl2cache
);
1122 spa_aux_remove(vd
, &spa_l2cache_avl
);
1123 vd
->vdev_isl2cache
= B_FALSE
;
1124 mutex_exit(&spa_l2cache_lock
);
1128 spa_l2cache_exists(uint64_t guid
, uint64_t *pool
)
1132 mutex_enter(&spa_l2cache_lock
);
1133 found
= spa_aux_exists(guid
, pool
, NULL
, &spa_l2cache_avl
);
1134 mutex_exit(&spa_l2cache_lock
);
1140 spa_l2cache_activate(vdev_t
*vd
)
1142 mutex_enter(&spa_l2cache_lock
);
1143 ASSERT(vd
->vdev_isl2cache
);
1144 spa_aux_activate(vd
, &spa_l2cache_avl
);
1145 mutex_exit(&spa_l2cache_lock
);
1149 * ==========================================================================
1151 * ==========================================================================
1155 * Lock the given spa_t for the purpose of adding or removing a vdev.
1156 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1157 * It returns the next transaction group for the spa_t.
1160 spa_vdev_enter(spa_t
*spa
)
1162 mutex_enter(&spa
->spa_vdev_top_lock
);
1163 mutex_enter(&spa_namespace_lock
);
1165 vdev_autotrim_stop_all(spa
);
1167 return (spa_vdev_config_enter(spa
));
1171 * The same as spa_vdev_enter() above but additionally takes the guid of
1172 * the vdev being detached. When there is a rebuild in process it will be
1173 * suspended while the vdev tree is modified then resumed by spa_vdev_exit().
1174 * The rebuild is canceled if only a single child remains after the detach.
1177 spa_vdev_detach_enter(spa_t
*spa
, uint64_t guid
)
1179 mutex_enter(&spa
->spa_vdev_top_lock
);
1180 mutex_enter(&spa_namespace_lock
);
1182 vdev_autotrim_stop_all(spa
);
1185 vdev_t
*vd
= spa_lookup_by_guid(spa
, guid
, B_FALSE
);
1187 vdev_rebuild_stop_wait(vd
->vdev_top
);
1191 return (spa_vdev_config_enter(spa
));
1195 * Internal implementation for spa_vdev_enter(). Used when a vdev
1196 * operation requires multiple syncs (i.e. removing a device) while
1197 * keeping the spa_namespace_lock held.
1200 spa_vdev_config_enter(spa_t
*spa
)
1202 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1204 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
1206 return (spa_last_synced_txg(spa
) + 1);
1210 * Used in combination with spa_vdev_config_enter() to allow the syncing
1211 * of multiple transactions without releasing the spa_namespace_lock.
1214 spa_vdev_config_exit(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int error
, char *tag
)
1216 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1218 int config_changed
= B_FALSE
;
1220 ASSERT(txg
> spa_last_synced_txg(spa
));
1222 spa
->spa_pending_vdev
= NULL
;
1225 * Reassess the DTLs.
1227 vdev_dtl_reassess(spa
->spa_root_vdev
, 0, 0, B_FALSE
, B_FALSE
);
1229 if (error
== 0 && !list_is_empty(&spa
->spa_config_dirty_list
)) {
1230 config_changed
= B_TRUE
;
1231 spa
->spa_config_generation
++;
1235 * Verify the metaslab classes.
1237 ASSERT(metaslab_class_validate(spa_normal_class(spa
)) == 0);
1238 ASSERT(metaslab_class_validate(spa_log_class(spa
)) == 0);
1239 ASSERT(metaslab_class_validate(spa_special_class(spa
)) == 0);
1240 ASSERT(metaslab_class_validate(spa_dedup_class(spa
)) == 0);
1242 spa_config_exit(spa
, SCL_ALL
, spa
);
1245 * Panic the system if the specified tag requires it. This
1246 * is useful for ensuring that configurations are updated
1249 if (zio_injection_enabled
)
1250 zio_handle_panic_injection(spa
, tag
, 0);
1253 * Note: this txg_wait_synced() is important because it ensures
1254 * that there won't be more than one config change per txg.
1255 * This allows us to use the txg as the generation number.
1258 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
1261 ASSERT(!vd
->vdev_detached
|| vd
->vdev_dtl_sm
== NULL
);
1262 if (vd
->vdev_ops
->vdev_op_leaf
) {
1263 mutex_enter(&vd
->vdev_initialize_lock
);
1264 vdev_initialize_stop(vd
, VDEV_INITIALIZE_CANCELED
,
1266 mutex_exit(&vd
->vdev_initialize_lock
);
1268 mutex_enter(&vd
->vdev_trim_lock
);
1269 vdev_trim_stop(vd
, VDEV_TRIM_CANCELED
, NULL
);
1270 mutex_exit(&vd
->vdev_trim_lock
);
1274 * The vdev may be both a leaf and top-level device.
1276 vdev_autotrim_stop_wait(vd
);
1278 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
1280 spa_config_exit(spa
, SCL_ALL
, spa
);
1284 * If the config changed, update the config cache.
1287 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
1291 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1292 * locking of spa_vdev_enter(), we also want make sure the transactions have
1293 * synced to disk, and then update the global configuration cache with the new
1297 spa_vdev_exit(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int error
)
1299 vdev_autotrim_restart(spa
);
1300 vdev_rebuild_restart(spa
);
1302 spa_vdev_config_exit(spa
, vd
, txg
, error
, FTAG
);
1303 mutex_exit(&spa_namespace_lock
);
1304 mutex_exit(&spa
->spa_vdev_top_lock
);
1310 * Lock the given spa_t for the purpose of changing vdev state.
1313 spa_vdev_state_enter(spa_t
*spa
, int oplocks
)
1315 int locks
= SCL_STATE_ALL
| oplocks
;
1318 * Root pools may need to read of the underlying devfs filesystem
1319 * when opening up a vdev. Unfortunately if we're holding the
1320 * SCL_ZIO lock it will result in a deadlock when we try to issue
1321 * the read from the root filesystem. Instead we "prefetch"
1322 * the associated vnodes that we need prior to opening the
1323 * underlying devices and cache them so that we can prevent
1324 * any I/O when we are doing the actual open.
1326 if (spa_is_root(spa
)) {
1327 int low
= locks
& ~(SCL_ZIO
- 1);
1328 int high
= locks
& ~low
;
1330 spa_config_enter(spa
, high
, spa
, RW_WRITER
);
1331 vdev_hold(spa
->spa_root_vdev
);
1332 spa_config_enter(spa
, low
, spa
, RW_WRITER
);
1334 spa_config_enter(spa
, locks
, spa
, RW_WRITER
);
1336 spa
->spa_vdev_locks
= locks
;
1340 spa_vdev_state_exit(spa_t
*spa
, vdev_t
*vd
, int error
)
1342 boolean_t config_changed
= B_FALSE
;
1345 if (vd
== NULL
|| vd
== spa
->spa_root_vdev
) {
1346 vdev_top
= spa
->spa_root_vdev
;
1348 vdev_top
= vd
->vdev_top
;
1351 if (vd
!= NULL
|| error
== 0)
1352 vdev_dtl_reassess(vdev_top
, 0, 0, B_FALSE
, B_FALSE
);
1355 if (vd
!= spa
->spa_root_vdev
)
1356 vdev_state_dirty(vdev_top
);
1358 config_changed
= B_TRUE
;
1359 spa
->spa_config_generation
++;
1362 if (spa_is_root(spa
))
1363 vdev_rele(spa
->spa_root_vdev
);
1365 ASSERT3U(spa
->spa_vdev_locks
, >=, SCL_STATE_ALL
);
1366 spa_config_exit(spa
, spa
->spa_vdev_locks
, spa
);
1369 * If anything changed, wait for it to sync. This ensures that,
1370 * from the system administrator's perspective, zpool(8) commands
1371 * are synchronous. This is important for things like zpool offline:
1372 * when the command completes, you expect no further I/O from ZFS.
1375 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1378 * If the config changed, update the config cache.
1380 if (config_changed
) {
1381 mutex_enter(&spa_namespace_lock
);
1382 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
1383 mutex_exit(&spa_namespace_lock
);
1390 * ==========================================================================
1391 * Miscellaneous functions
1392 * ==========================================================================
1396 spa_activate_mos_feature(spa_t
*spa
, const char *feature
, dmu_tx_t
*tx
)
1398 if (!nvlist_exists(spa
->spa_label_features
, feature
)) {
1399 fnvlist_add_boolean(spa
->spa_label_features
, feature
);
1401 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1402 * dirty the vdev config because lock SCL_CONFIG is not held.
1403 * Thankfully, in this case we don't need to dirty the config
1404 * because it will be written out anyway when we finish
1405 * creating the pool.
1407 if (tx
->tx_txg
!= TXG_INITIAL
)
1408 vdev_config_dirty(spa
->spa_root_vdev
);
1413 spa_deactivate_mos_feature(spa_t
*spa
, const char *feature
)
1415 if (nvlist_remove_all(spa
->spa_label_features
, feature
) == 0)
1416 vdev_config_dirty(spa
->spa_root_vdev
);
1420 * Return the spa_t associated with given pool_guid, if it exists. If
1421 * device_guid is non-zero, determine whether the pool exists *and* contains
1422 * a device with the specified device_guid.
1425 spa_by_guid(uint64_t pool_guid
, uint64_t device_guid
)
1428 avl_tree_t
*t
= &spa_namespace_avl
;
1430 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1432 for (spa
= avl_first(t
); spa
!= NULL
; spa
= AVL_NEXT(t
, spa
)) {
1433 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1435 if (spa
->spa_root_vdev
== NULL
)
1437 if (spa_guid(spa
) == pool_guid
) {
1438 if (device_guid
== 0)
1441 if (vdev_lookup_by_guid(spa
->spa_root_vdev
,
1442 device_guid
) != NULL
)
1446 * Check any devices we may be in the process of adding.
1448 if (spa
->spa_pending_vdev
) {
1449 if (vdev_lookup_by_guid(spa
->spa_pending_vdev
,
1450 device_guid
) != NULL
)
1460 * Determine whether a pool with the given pool_guid exists.
1463 spa_guid_exists(uint64_t pool_guid
, uint64_t device_guid
)
1465 return (spa_by_guid(pool_guid
, device_guid
) != NULL
);
1469 spa_strdup(const char *s
)
1475 new = kmem_alloc(len
+ 1, KM_SLEEP
);
1483 spa_strfree(char *s
)
1485 kmem_free(s
, strlen(s
) + 1);
1489 spa_get_random(uint64_t range
)
1498 (void) random_get_pseudo_bytes((void *)&r
, sizeof (uint64_t));
1504 spa_generate_guid(spa_t
*spa
)
1506 uint64_t guid
= spa_get_random(-1ULL);
1509 while (guid
== 0 || spa_guid_exists(spa_guid(spa
), guid
))
1510 guid
= spa_get_random(-1ULL);
1512 while (guid
== 0 || spa_guid_exists(guid
, 0))
1513 guid
= spa_get_random(-1ULL);
1520 snprintf_blkptr(char *buf
, size_t buflen
, const blkptr_t
*bp
)
1523 char *checksum
= NULL
;
1524 char *compress
= NULL
;
1527 if (BP_GET_TYPE(bp
) & DMU_OT_NEWTYPE
) {
1528 dmu_object_byteswap_t bswap
=
1529 DMU_OT_BYTESWAP(BP_GET_TYPE(bp
));
1530 (void) snprintf(type
, sizeof (type
), "bswap %s %s",
1531 DMU_OT_IS_METADATA(BP_GET_TYPE(bp
)) ?
1532 "metadata" : "data",
1533 dmu_ot_byteswap
[bswap
].ob_name
);
1535 (void) strlcpy(type
, dmu_ot
[BP_GET_TYPE(bp
)].ot_name
,
1538 if (!BP_IS_EMBEDDED(bp
)) {
1540 zio_checksum_table
[BP_GET_CHECKSUM(bp
)].ci_name
;
1542 compress
= zio_compress_table
[BP_GET_COMPRESS(bp
)].ci_name
;
1545 SNPRINTF_BLKPTR(snprintf
, ' ', buf
, buflen
, bp
, type
, checksum
,
1550 spa_freeze(spa_t
*spa
)
1552 uint64_t freeze_txg
= 0;
1554 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1555 if (spa
->spa_freeze_txg
== UINT64_MAX
) {
1556 freeze_txg
= spa_last_synced_txg(spa
) + TXG_SIZE
;
1557 spa
->spa_freeze_txg
= freeze_txg
;
1559 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1560 if (freeze_txg
!= 0)
1561 txg_wait_synced(spa_get_dsl(spa
), freeze_txg
);
1565 zfs_panic_recover(const char *fmt
, ...)
1570 vcmn_err(zfs_recover
? CE_WARN
: CE_PANIC
, fmt
, adx
);
1575 * This is a stripped-down version of strtoull, suitable only for converting
1576 * lowercase hexadecimal numbers that don't overflow.
1579 zfs_strtonum(const char *str
, char **nptr
)
1585 while ((c
= *str
) != '\0') {
1586 if (c
>= '0' && c
<= '9')
1588 else if (c
>= 'a' && c
<= 'f')
1589 digit
= 10 + c
- 'a';
1600 *nptr
= (char *)str
;
1606 spa_activate_allocation_classes(spa_t
*spa
, dmu_tx_t
*tx
)
1609 * We bump the feature refcount for each special vdev added to the pool
1611 ASSERT(spa_feature_is_enabled(spa
, SPA_FEATURE_ALLOCATION_CLASSES
));
1612 spa_feature_incr(spa
, SPA_FEATURE_ALLOCATION_CLASSES
, tx
);
1616 * ==========================================================================
1617 * Accessor functions
1618 * ==========================================================================
1622 spa_shutting_down(spa_t
*spa
)
1624 return (spa
->spa_async_suspended
);
1628 spa_get_dsl(spa_t
*spa
)
1630 return (spa
->spa_dsl_pool
);
1634 spa_is_initializing(spa_t
*spa
)
1636 return (spa
->spa_is_initializing
);
1640 spa_indirect_vdevs_loaded(spa_t
*spa
)
1642 return (spa
->spa_indirect_vdevs_loaded
);
1646 spa_get_rootblkptr(spa_t
*spa
)
1648 return (&spa
->spa_ubsync
.ub_rootbp
);
1652 spa_set_rootblkptr(spa_t
*spa
, const blkptr_t
*bp
)
1654 spa
->spa_uberblock
.ub_rootbp
= *bp
;
1658 spa_altroot(spa_t
*spa
, char *buf
, size_t buflen
)
1660 if (spa
->spa_root
== NULL
)
1663 (void) strncpy(buf
, spa
->spa_root
, buflen
);
1667 spa_sync_pass(spa_t
*spa
)
1669 return (spa
->spa_sync_pass
);
1673 spa_name(spa_t
*spa
)
1675 return (spa
->spa_name
);
1679 spa_guid(spa_t
*spa
)
1681 dsl_pool_t
*dp
= spa_get_dsl(spa
);
1685 * If we fail to parse the config during spa_load(), we can go through
1686 * the error path (which posts an ereport) and end up here with no root
1687 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1690 if (spa
->spa_root_vdev
== NULL
)
1691 return (spa
->spa_config_guid
);
1693 guid
= spa
->spa_last_synced_guid
!= 0 ?
1694 spa
->spa_last_synced_guid
: spa
->spa_root_vdev
->vdev_guid
;
1697 * Return the most recently synced out guid unless we're
1698 * in syncing context.
1700 if (dp
&& dsl_pool_sync_context(dp
))
1701 return (spa
->spa_root_vdev
->vdev_guid
);
1707 spa_load_guid(spa_t
*spa
)
1710 * This is a GUID that exists solely as a reference for the
1711 * purposes of the arc. It is generated at load time, and
1712 * is never written to persistent storage.
1714 return (spa
->spa_load_guid
);
1718 spa_last_synced_txg(spa_t
*spa
)
1720 return (spa
->spa_ubsync
.ub_txg
);
1724 spa_first_txg(spa_t
*spa
)
1726 return (spa
->spa_first_txg
);
1730 spa_syncing_txg(spa_t
*spa
)
1732 return (spa
->spa_syncing_txg
);
1736 * Return the last txg where data can be dirtied. The final txgs
1737 * will be used to just clear out any deferred frees that remain.
1740 spa_final_dirty_txg(spa_t
*spa
)
1742 return (spa
->spa_final_txg
- TXG_DEFER_SIZE
);
1746 spa_state(spa_t
*spa
)
1748 return (spa
->spa_state
);
1752 spa_load_state(spa_t
*spa
)
1754 return (spa
->spa_load_state
);
1758 spa_freeze_txg(spa_t
*spa
)
1760 return (spa
->spa_freeze_txg
);
1764 * Return the inflated asize for a logical write in bytes. This is used by the
1765 * DMU to calculate the space a logical write will require on disk.
1766 * If lsize is smaller than the largest physical block size allocatable on this
1767 * pool we use its value instead, since the write will end up using the whole
1771 spa_get_worst_case_asize(spa_t
*spa
, uint64_t lsize
)
1774 return (0); /* No inflation needed */
1775 return (MAX(lsize
, 1 << spa
->spa_max_ashift
) * spa_asize_inflation
);
1779 * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%),
1780 * or at least 128MB, unless that would cause it to be more than half the
1783 * See the comment above spa_slop_shift for details.
1786 spa_get_slop_space(spa_t
*spa
)
1788 uint64_t space
= spa_get_dspace(spa
);
1789 return (MAX(space
>> spa_slop_shift
, MIN(space
>> 1, spa_min_slop
)));
1793 spa_get_dspace(spa_t
*spa
)
1795 return (spa
->spa_dspace
);
1799 spa_get_checkpoint_space(spa_t
*spa
)
1801 return (spa
->spa_checkpoint_info
.sci_dspace
);
1805 spa_update_dspace(spa_t
*spa
)
1807 spa
->spa_dspace
= metaslab_class_get_dspace(spa_normal_class(spa
)) +
1808 ddt_get_dedup_dspace(spa
);
1809 if (spa
->spa_vdev_removal
!= NULL
) {
1811 * We can't allocate from the removing device, so
1812 * subtract its size. This prevents the DMU/DSL from
1813 * filling up the (now smaller) pool while we are in the
1814 * middle of removing the device.
1816 * Note that the DMU/DSL doesn't actually know or care
1817 * how much space is allocated (it does its own tracking
1818 * of how much space has been logically used). So it
1819 * doesn't matter that the data we are moving may be
1820 * allocated twice (on the old device and the new
1823 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
1825 vdev_lookup_top(spa
, spa
->spa_vdev_removal
->svr_vdev_id
);
1826 spa
->spa_dspace
-= spa_deflate(spa
) ?
1827 vd
->vdev_stat
.vs_dspace
: vd
->vdev_stat
.vs_space
;
1828 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
1833 * Return the failure mode that has been set to this pool. The default
1834 * behavior will be to block all I/Os when a complete failure occurs.
1837 spa_get_failmode(spa_t
*spa
)
1839 return (spa
->spa_failmode
);
1843 spa_suspended(spa_t
*spa
)
1845 return (spa
->spa_suspended
!= ZIO_SUSPEND_NONE
);
1849 spa_version(spa_t
*spa
)
1851 return (spa
->spa_ubsync
.ub_version
);
1855 spa_deflate(spa_t
*spa
)
1857 return (spa
->spa_deflate
);
1861 spa_normal_class(spa_t
*spa
)
1863 return (spa
->spa_normal_class
);
1867 spa_log_class(spa_t
*spa
)
1869 return (spa
->spa_log_class
);
1873 spa_special_class(spa_t
*spa
)
1875 return (spa
->spa_special_class
);
1879 spa_dedup_class(spa_t
*spa
)
1881 return (spa
->spa_dedup_class
);
1885 * Locate an appropriate allocation class
1888 spa_preferred_class(spa_t
*spa
, uint64_t size
, dmu_object_type_t objtype
,
1889 uint_t level
, uint_t special_smallblk
)
1891 if (DMU_OT_IS_ZIL(objtype
)) {
1892 if (spa
->spa_log_class
->mc_groups
!= 0)
1893 return (spa_log_class(spa
));
1895 return (spa_normal_class(spa
));
1898 boolean_t has_special_class
= spa
->spa_special_class
->mc_groups
!= 0;
1900 if (DMU_OT_IS_DDT(objtype
)) {
1901 if (spa
->spa_dedup_class
->mc_groups
!= 0)
1902 return (spa_dedup_class(spa
));
1903 else if (has_special_class
&& zfs_ddt_data_is_special
)
1904 return (spa_special_class(spa
));
1906 return (spa_normal_class(spa
));
1909 /* Indirect blocks for user data can land in special if allowed */
1910 if (level
> 0 && (DMU_OT_IS_FILE(objtype
) || objtype
== DMU_OT_ZVOL
)) {
1911 if (has_special_class
&& zfs_user_indirect_is_special
)
1912 return (spa_special_class(spa
));
1914 return (spa_normal_class(spa
));
1917 if (DMU_OT_IS_METADATA(objtype
) || level
> 0) {
1918 if (has_special_class
)
1919 return (spa_special_class(spa
));
1921 return (spa_normal_class(spa
));
1925 * Allow small file blocks in special class in some cases (like
1926 * for the dRAID vdev feature). But always leave a reserve of
1927 * zfs_special_class_metadata_reserve_pct exclusively for metadata.
1929 if (DMU_OT_IS_FILE(objtype
) &&
1930 has_special_class
&& size
<= special_smallblk
) {
1931 metaslab_class_t
*special
= spa_special_class(spa
);
1932 uint64_t alloc
= metaslab_class_get_alloc(special
);
1933 uint64_t space
= metaslab_class_get_space(special
);
1935 (space
* (100 - zfs_special_class_metadata_reserve_pct
))
1942 return (spa_normal_class(spa
));
1946 spa_evicting_os_register(spa_t
*spa
, objset_t
*os
)
1948 mutex_enter(&spa
->spa_evicting_os_lock
);
1949 list_insert_head(&spa
->spa_evicting_os_list
, os
);
1950 mutex_exit(&spa
->spa_evicting_os_lock
);
1954 spa_evicting_os_deregister(spa_t
*spa
, objset_t
*os
)
1956 mutex_enter(&spa
->spa_evicting_os_lock
);
1957 list_remove(&spa
->spa_evicting_os_list
, os
);
1958 cv_broadcast(&spa
->spa_evicting_os_cv
);
1959 mutex_exit(&spa
->spa_evicting_os_lock
);
1963 spa_evicting_os_wait(spa_t
*spa
)
1965 mutex_enter(&spa
->spa_evicting_os_lock
);
1966 while (!list_is_empty(&spa
->spa_evicting_os_list
))
1967 cv_wait(&spa
->spa_evicting_os_cv
, &spa
->spa_evicting_os_lock
);
1968 mutex_exit(&spa
->spa_evicting_os_lock
);
1970 dmu_buf_user_evict_wait();
1974 spa_max_replication(spa_t
*spa
)
1977 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1978 * handle BPs with more than one DVA allocated. Set our max
1979 * replication level accordingly.
1981 if (spa_version(spa
) < SPA_VERSION_DITTO_BLOCKS
)
1983 return (MIN(SPA_DVAS_PER_BP
, spa_max_replication_override
));
1987 spa_prev_software_version(spa_t
*spa
)
1989 return (spa
->spa_prev_software_version
);
1993 spa_deadman_synctime(spa_t
*spa
)
1995 return (spa
->spa_deadman_synctime
);
1999 spa_get_autotrim(spa_t
*spa
)
2001 return (spa
->spa_autotrim
);
2005 spa_deadman_ziotime(spa_t
*spa
)
2007 return (spa
->spa_deadman_ziotime
);
2011 spa_get_deadman_failmode(spa_t
*spa
)
2013 return (spa
->spa_deadman_failmode
);
2017 spa_set_deadman_failmode(spa_t
*spa
, const char *failmode
)
2019 if (strcmp(failmode
, "wait") == 0)
2020 spa
->spa_deadman_failmode
= ZIO_FAILURE_MODE_WAIT
;
2021 else if (strcmp(failmode
, "continue") == 0)
2022 spa
->spa_deadman_failmode
= ZIO_FAILURE_MODE_CONTINUE
;
2023 else if (strcmp(failmode
, "panic") == 0)
2024 spa
->spa_deadman_failmode
= ZIO_FAILURE_MODE_PANIC
;
2026 spa
->spa_deadman_failmode
= ZIO_FAILURE_MODE_WAIT
;
2030 spa_set_deadman_ziotime(hrtime_t ns
)
2034 if (spa_mode_global
!= SPA_MODE_UNINIT
) {
2035 mutex_enter(&spa_namespace_lock
);
2036 while ((spa
= spa_next(spa
)) != NULL
)
2037 spa
->spa_deadman_ziotime
= ns
;
2038 mutex_exit(&spa_namespace_lock
);
2043 spa_set_deadman_synctime(hrtime_t ns
)
2047 if (spa_mode_global
!= SPA_MODE_UNINIT
) {
2048 mutex_enter(&spa_namespace_lock
);
2049 while ((spa
= spa_next(spa
)) != NULL
)
2050 spa
->spa_deadman_synctime
= ns
;
2051 mutex_exit(&spa_namespace_lock
);
2056 dva_get_dsize_sync(spa_t
*spa
, const dva_t
*dva
)
2058 uint64_t asize
= DVA_GET_ASIZE(dva
);
2059 uint64_t dsize
= asize
;
2061 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
2063 if (asize
!= 0 && spa
->spa_deflate
) {
2064 vdev_t
*vd
= vdev_lookup_top(spa
, DVA_GET_VDEV(dva
));
2066 dsize
= (asize
>> SPA_MINBLOCKSHIFT
) *
2067 vd
->vdev_deflate_ratio
;
2074 bp_get_dsize_sync(spa_t
*spa
, const blkptr_t
*bp
)
2078 for (int d
= 0; d
< BP_GET_NDVAS(bp
); d
++)
2079 dsize
+= dva_get_dsize_sync(spa
, &bp
->blk_dva
[d
]);
2085 bp_get_dsize(spa_t
*spa
, const blkptr_t
*bp
)
2089 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2091 for (int d
= 0; d
< BP_GET_NDVAS(bp
); d
++)
2092 dsize
+= dva_get_dsize_sync(spa
, &bp
->blk_dva
[d
]);
2094 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2100 spa_dirty_data(spa_t
*spa
)
2102 return (spa
->spa_dsl_pool
->dp_dirty_total
);
2106 * ==========================================================================
2107 * SPA Import Progress Routines
2108 * ==========================================================================
2111 typedef struct spa_import_progress
{
2112 uint64_t pool_guid
; /* unique id for updates */
2114 spa_load_state_t spa_load_state
;
2115 uint64_t mmp_sec_remaining
; /* MMP activity check */
2116 uint64_t spa_load_max_txg
; /* rewind txg */
2117 procfs_list_node_t smh_node
;
2118 } spa_import_progress_t
;
2120 spa_history_list_t
*spa_import_progress_list
= NULL
;
2123 spa_import_progress_show_header(struct seq_file
*f
)
2125 seq_printf(f
, "%-20s %-14s %-14s %-12s %s\n", "pool_guid",
2126 "load_state", "multihost_secs", "max_txg",
2132 spa_import_progress_show(struct seq_file
*f
, void *data
)
2134 spa_import_progress_t
*sip
= (spa_import_progress_t
*)data
;
2136 seq_printf(f
, "%-20llu %-14llu %-14llu %-12llu %s\n",
2137 (u_longlong_t
)sip
->pool_guid
, (u_longlong_t
)sip
->spa_load_state
,
2138 (u_longlong_t
)sip
->mmp_sec_remaining
,
2139 (u_longlong_t
)sip
->spa_load_max_txg
,
2140 (sip
->pool_name
? sip
->pool_name
: "-"));
2145 /* Remove oldest elements from list until there are no more than 'size' left */
2147 spa_import_progress_truncate(spa_history_list_t
*shl
, unsigned int size
)
2149 spa_import_progress_t
*sip
;
2150 while (shl
->size
> size
) {
2151 sip
= list_remove_head(&shl
->procfs_list
.pl_list
);
2153 spa_strfree(sip
->pool_name
);
2154 kmem_free(sip
, sizeof (spa_import_progress_t
));
2158 IMPLY(size
== 0, list_is_empty(&shl
->procfs_list
.pl_list
));
2162 spa_import_progress_init(void)
2164 spa_import_progress_list
= kmem_zalloc(sizeof (spa_history_list_t
),
2167 spa_import_progress_list
->size
= 0;
2169 spa_import_progress_list
->procfs_list
.pl_private
=
2170 spa_import_progress_list
;
2172 procfs_list_install("zfs",
2176 &spa_import_progress_list
->procfs_list
,
2177 spa_import_progress_show
,
2178 spa_import_progress_show_header
,
2180 offsetof(spa_import_progress_t
, smh_node
));
2184 spa_import_progress_destroy(void)
2186 spa_history_list_t
*shl
= spa_import_progress_list
;
2187 procfs_list_uninstall(&shl
->procfs_list
);
2188 spa_import_progress_truncate(shl
, 0);
2189 procfs_list_destroy(&shl
->procfs_list
);
2190 kmem_free(shl
, sizeof (spa_history_list_t
));
2194 spa_import_progress_set_state(uint64_t pool_guid
,
2195 spa_load_state_t load_state
)
2197 spa_history_list_t
*shl
= spa_import_progress_list
;
2198 spa_import_progress_t
*sip
;
2204 mutex_enter(&shl
->procfs_list
.pl_lock
);
2205 for (sip
= list_tail(&shl
->procfs_list
.pl_list
); sip
!= NULL
;
2206 sip
= list_prev(&shl
->procfs_list
.pl_list
, sip
)) {
2207 if (sip
->pool_guid
== pool_guid
) {
2208 sip
->spa_load_state
= load_state
;
2213 mutex_exit(&shl
->procfs_list
.pl_lock
);
2219 spa_import_progress_set_max_txg(uint64_t pool_guid
, uint64_t load_max_txg
)
2221 spa_history_list_t
*shl
= spa_import_progress_list
;
2222 spa_import_progress_t
*sip
;
2228 mutex_enter(&shl
->procfs_list
.pl_lock
);
2229 for (sip
= list_tail(&shl
->procfs_list
.pl_list
); sip
!= NULL
;
2230 sip
= list_prev(&shl
->procfs_list
.pl_list
, sip
)) {
2231 if (sip
->pool_guid
== pool_guid
) {
2232 sip
->spa_load_max_txg
= load_max_txg
;
2237 mutex_exit(&shl
->procfs_list
.pl_lock
);
2243 spa_import_progress_set_mmp_check(uint64_t pool_guid
,
2244 uint64_t mmp_sec_remaining
)
2246 spa_history_list_t
*shl
= spa_import_progress_list
;
2247 spa_import_progress_t
*sip
;
2253 mutex_enter(&shl
->procfs_list
.pl_lock
);
2254 for (sip
= list_tail(&shl
->procfs_list
.pl_list
); sip
!= NULL
;
2255 sip
= list_prev(&shl
->procfs_list
.pl_list
, sip
)) {
2256 if (sip
->pool_guid
== pool_guid
) {
2257 sip
->mmp_sec_remaining
= mmp_sec_remaining
;
2262 mutex_exit(&shl
->procfs_list
.pl_lock
);
2268 * A new import is in progress, add an entry.
2271 spa_import_progress_add(spa_t
*spa
)
2273 spa_history_list_t
*shl
= spa_import_progress_list
;
2274 spa_import_progress_t
*sip
;
2275 char *poolname
= NULL
;
2277 sip
= kmem_zalloc(sizeof (spa_import_progress_t
), KM_SLEEP
);
2278 sip
->pool_guid
= spa_guid(spa
);
2280 (void) nvlist_lookup_string(spa
->spa_config
, ZPOOL_CONFIG_POOL_NAME
,
2282 if (poolname
== NULL
)
2283 poolname
= spa_name(spa
);
2284 sip
->pool_name
= spa_strdup(poolname
);
2285 sip
->spa_load_state
= spa_load_state(spa
);
2287 mutex_enter(&shl
->procfs_list
.pl_lock
);
2288 procfs_list_add(&shl
->procfs_list
, sip
);
2290 mutex_exit(&shl
->procfs_list
.pl_lock
);
2294 spa_import_progress_remove(uint64_t pool_guid
)
2296 spa_history_list_t
*shl
= spa_import_progress_list
;
2297 spa_import_progress_t
*sip
;
2299 mutex_enter(&shl
->procfs_list
.pl_lock
);
2300 for (sip
= list_tail(&shl
->procfs_list
.pl_list
); sip
!= NULL
;
2301 sip
= list_prev(&shl
->procfs_list
.pl_list
, sip
)) {
2302 if (sip
->pool_guid
== pool_guid
) {
2304 spa_strfree(sip
->pool_name
);
2305 list_remove(&shl
->procfs_list
.pl_list
, sip
);
2307 kmem_free(sip
, sizeof (spa_import_progress_t
));
2311 mutex_exit(&shl
->procfs_list
.pl_lock
);
2315 * ==========================================================================
2316 * Initialization and Termination
2317 * ==========================================================================
2321 spa_name_compare(const void *a1
, const void *a2
)
2323 const spa_t
*s1
= a1
;
2324 const spa_t
*s2
= a2
;
2327 s
= strcmp(s1
->spa_name
, s2
->spa_name
);
2329 return (TREE_ISIGN(s
));
2339 spa_init(spa_mode_t mode
)
2341 mutex_init(&spa_namespace_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
2342 mutex_init(&spa_spare_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
2343 mutex_init(&spa_l2cache_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
2344 cv_init(&spa_namespace_cv
, NULL
, CV_DEFAULT
, NULL
);
2346 avl_create(&spa_namespace_avl
, spa_name_compare
, sizeof (spa_t
),
2347 offsetof(spa_t
, spa_avl
));
2349 avl_create(&spa_spare_avl
, spa_spare_compare
, sizeof (spa_aux_t
),
2350 offsetof(spa_aux_t
, aux_avl
));
2352 avl_create(&spa_l2cache_avl
, spa_l2cache_compare
, sizeof (spa_aux_t
),
2353 offsetof(spa_aux_t
, aux_avl
));
2355 spa_mode_global
= mode
;
2358 if (spa_mode_global
!= SPA_MODE_READ
&& dprintf_find_string("watch")) {
2359 struct sigaction sa
;
2361 sa
.sa_flags
= SA_SIGINFO
;
2362 sigemptyset(&sa
.sa_mask
);
2363 sa
.sa_sigaction
= arc_buf_sigsegv
;
2365 if (sigaction(SIGSEGV
, &sa
, NULL
) == -1) {
2366 perror("could not enable watchpoints: "
2367 "sigaction(SIGSEGV, ...) = ");
2375 zfs_refcount_init();
2378 metaslab_stat_init();
2383 vdev_cache_stat_init();
2384 vdev_mirror_stat_init();
2385 vdev_raidz_math_init();
2389 zpool_feature_init();
2394 spa_import_progress_init();
2405 vdev_cache_stat_fini();
2406 vdev_mirror_stat_fini();
2407 vdev_raidz_math_fini();
2412 metaslab_stat_fini();
2415 zfs_refcount_fini();
2419 spa_import_progress_destroy();
2421 avl_destroy(&spa_namespace_avl
);
2422 avl_destroy(&spa_spare_avl
);
2423 avl_destroy(&spa_l2cache_avl
);
2425 cv_destroy(&spa_namespace_cv
);
2426 mutex_destroy(&spa_namespace_lock
);
2427 mutex_destroy(&spa_spare_lock
);
2428 mutex_destroy(&spa_l2cache_lock
);
2432 * Return whether this pool has slogs. No locking needed.
2433 * It's not a problem if the wrong answer is returned as it's only for
2434 * performance and not correctness
2437 spa_has_slogs(spa_t
*spa
)
2439 return (spa
->spa_log_class
->mc_rotor
!= NULL
);
2443 spa_get_log_state(spa_t
*spa
)
2445 return (spa
->spa_log_state
);
2449 spa_set_log_state(spa_t
*spa
, spa_log_state_t state
)
2451 spa
->spa_log_state
= state
;
2455 spa_is_root(spa_t
*spa
)
2457 return (spa
->spa_is_root
);
2461 spa_writeable(spa_t
*spa
)
2463 return (!!(spa
->spa_mode
& SPA_MODE_WRITE
) && spa
->spa_trust_config
);
2467 * Returns true if there is a pending sync task in any of the current
2468 * syncing txg, the current quiescing txg, or the current open txg.
2471 spa_has_pending_synctask(spa_t
*spa
)
2473 return (!txg_all_lists_empty(&spa
->spa_dsl_pool
->dp_sync_tasks
) ||
2474 !txg_all_lists_empty(&spa
->spa_dsl_pool
->dp_early_sync_tasks
));
2478 spa_mode(spa_t
*spa
)
2480 return (spa
->spa_mode
);
2484 spa_bootfs(spa_t
*spa
)
2486 return (spa
->spa_bootfs
);
2490 spa_delegation(spa_t
*spa
)
2492 return (spa
->spa_delegation
);
2496 spa_meta_objset(spa_t
*spa
)
2498 return (spa
->spa_meta_objset
);
2502 spa_dedup_checksum(spa_t
*spa
)
2504 return (spa
->spa_dedup_checksum
);
2508 * Reset pool scan stat per scan pass (or reboot).
2511 spa_scan_stat_init(spa_t
*spa
)
2513 /* data not stored on disk */
2514 spa
->spa_scan_pass_start
= gethrestime_sec();
2515 if (dsl_scan_is_paused_scrub(spa
->spa_dsl_pool
->dp_scan
))
2516 spa
->spa_scan_pass_scrub_pause
= spa
->spa_scan_pass_start
;
2518 spa
->spa_scan_pass_scrub_pause
= 0;
2519 spa
->spa_scan_pass_scrub_spent_paused
= 0;
2520 spa
->spa_scan_pass_exam
= 0;
2521 spa
->spa_scan_pass_issued
= 0;
2522 vdev_scan_stat_init(spa
->spa_root_vdev
);
2526 * Get scan stats for zpool status reports
2529 spa_scan_get_stats(spa_t
*spa
, pool_scan_stat_t
*ps
)
2531 dsl_scan_t
*scn
= spa
->spa_dsl_pool
? spa
->spa_dsl_pool
->dp_scan
: NULL
;
2533 if (scn
== NULL
|| scn
->scn_phys
.scn_func
== POOL_SCAN_NONE
)
2534 return (SET_ERROR(ENOENT
));
2535 bzero(ps
, sizeof (pool_scan_stat_t
));
2537 /* data stored on disk */
2538 ps
->pss_func
= scn
->scn_phys
.scn_func
;
2539 ps
->pss_state
= scn
->scn_phys
.scn_state
;
2540 ps
->pss_start_time
= scn
->scn_phys
.scn_start_time
;
2541 ps
->pss_end_time
= scn
->scn_phys
.scn_end_time
;
2542 ps
->pss_to_examine
= scn
->scn_phys
.scn_to_examine
;
2543 ps
->pss_examined
= scn
->scn_phys
.scn_examined
;
2544 ps
->pss_to_process
= scn
->scn_phys
.scn_to_process
;
2545 ps
->pss_processed
= scn
->scn_phys
.scn_processed
;
2546 ps
->pss_errors
= scn
->scn_phys
.scn_errors
;
2548 /* data not stored on disk */
2549 ps
->pss_pass_exam
= spa
->spa_scan_pass_exam
;
2550 ps
->pss_pass_start
= spa
->spa_scan_pass_start
;
2551 ps
->pss_pass_scrub_pause
= spa
->spa_scan_pass_scrub_pause
;
2552 ps
->pss_pass_scrub_spent_paused
= spa
->spa_scan_pass_scrub_spent_paused
;
2553 ps
->pss_pass_issued
= spa
->spa_scan_pass_issued
;
2555 scn
->scn_issued_before_pass
+ spa
->spa_scan_pass_issued
;
2561 spa_maxblocksize(spa_t
*spa
)
2563 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_BLOCKS
))
2564 return (SPA_MAXBLOCKSIZE
);
2566 return (SPA_OLD_MAXBLOCKSIZE
);
2571 * Returns the txg that the last device removal completed. No indirect mappings
2572 * have been added since this txg.
2575 spa_get_last_removal_txg(spa_t
*spa
)
2578 uint64_t ret
= -1ULL;
2580 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2582 * sr_prev_indirect_vdev is only modified while holding all the
2583 * config locks, so it is sufficient to hold SCL_VDEV as reader when
2586 vdevid
= spa
->spa_removing_phys
.sr_prev_indirect_vdev
;
2588 while (vdevid
!= -1ULL) {
2589 vdev_t
*vd
= vdev_lookup_top(spa
, vdevid
);
2590 vdev_indirect_births_t
*vib
= vd
->vdev_indirect_births
;
2592 ASSERT3P(vd
->vdev_ops
, ==, &vdev_indirect_ops
);
2595 * If the removal did not remap any data, we don't care.
2597 if (vdev_indirect_births_count(vib
) != 0) {
2598 ret
= vdev_indirect_births_last_entry_txg(vib
);
2602 vdevid
= vd
->vdev_indirect_config
.vic_prev_indirect_vdev
;
2604 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2607 spa_feature_is_active(spa
, SPA_FEATURE_DEVICE_REMOVAL
));
2613 spa_maxdnodesize(spa_t
*spa
)
2615 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_DNODE
))
2616 return (DNODE_MAX_SIZE
);
2618 return (DNODE_MIN_SIZE
);
2622 spa_multihost(spa_t
*spa
)
2624 return (spa
->spa_multihost
? B_TRUE
: B_FALSE
);
2628 spa_get_hostid(spa_t
*spa
)
2630 return (spa
->spa_hostid
);
2634 spa_trust_config(spa_t
*spa
)
2636 return (spa
->spa_trust_config
);
2640 spa_missing_tvds_allowed(spa_t
*spa
)
2642 return (spa
->spa_missing_tvds_allowed
);
2646 spa_syncing_log_sm(spa_t
*spa
)
2648 return (spa
->spa_syncing_log_sm
);
2652 spa_set_missing_tvds(spa_t
*spa
, uint64_t missing
)
2654 spa
->spa_missing_tvds
= missing
;
2658 * Return the pool state string ("ONLINE", "DEGRADED", "SUSPENDED", etc).
2661 spa_state_to_name(spa_t
*spa
)
2663 ASSERT3P(spa
, !=, NULL
);
2666 * it is possible for the spa to exist, without root vdev
2667 * as the spa transitions during import/export
2669 vdev_t
*rvd
= spa
->spa_root_vdev
;
2671 return ("TRANSITIONING");
2673 vdev_state_t state
= rvd
->vdev_state
;
2674 vdev_aux_t aux
= rvd
->vdev_stat
.vs_aux
;
2676 if (spa_suspended(spa
) &&
2677 (spa_get_failmode(spa
) != ZIO_FAILURE_MODE_CONTINUE
))
2678 return ("SUSPENDED");
2681 case VDEV_STATE_CLOSED
:
2682 case VDEV_STATE_OFFLINE
:
2684 case VDEV_STATE_REMOVED
:
2686 case VDEV_STATE_CANT_OPEN
:
2687 if (aux
== VDEV_AUX_CORRUPT_DATA
|| aux
== VDEV_AUX_BAD_LOG
)
2689 else if (aux
== VDEV_AUX_SPLIT_POOL
)
2693 case VDEV_STATE_FAULTED
:
2695 case VDEV_STATE_DEGRADED
:
2696 return ("DEGRADED");
2697 case VDEV_STATE_HEALTHY
:
2707 spa_top_vdevs_spacemap_addressable(spa_t
*spa
)
2709 vdev_t
*rvd
= spa
->spa_root_vdev
;
2710 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
2711 if (!vdev_is_spacemap_addressable(rvd
->vdev_child
[c
]))
2718 spa_has_checkpoint(spa_t
*spa
)
2720 return (spa
->spa_checkpoint_txg
!= 0);
2724 spa_importing_readonly_checkpoint(spa_t
*spa
)
2726 return ((spa
->spa_import_flags
& ZFS_IMPORT_CHECKPOINT
) &&
2727 spa
->spa_mode
== SPA_MODE_READ
);
2731 spa_min_claim_txg(spa_t
*spa
)
2733 uint64_t checkpoint_txg
= spa
->spa_uberblock
.ub_checkpoint_txg
;
2735 if (checkpoint_txg
!= 0)
2736 return (checkpoint_txg
+ 1);
2738 return (spa
->spa_first_txg
);
2742 * If there is a checkpoint, async destroys may consume more space from
2743 * the pool instead of freeing it. In an attempt to save the pool from
2744 * getting suspended when it is about to run out of space, we stop
2745 * processing async destroys.
2748 spa_suspend_async_destroy(spa_t
*spa
)
2750 dsl_pool_t
*dp
= spa_get_dsl(spa
);
2752 uint64_t unreserved
= dsl_pool_unreserved_space(dp
,
2753 ZFS_SPACE_CHECK_EXTRA_RESERVED
);
2754 uint64_t used
= dsl_dir_phys(dp
->dp_root_dir
)->dd_used_bytes
;
2755 uint64_t avail
= (unreserved
> used
) ? (unreserved
- used
) : 0;
2757 if (spa_has_checkpoint(spa
) && avail
== 0)
2763 #if defined(_KERNEL)
2766 param_set_deadman_failmode_common(const char *val
)
2772 return (SET_ERROR(EINVAL
));
2774 if ((p
= strchr(val
, '\n')) != NULL
)
2777 if (strcmp(val
, "wait") != 0 && strcmp(val
, "continue") != 0 &&
2778 strcmp(val
, "panic"))
2779 return (SET_ERROR(EINVAL
));
2781 if (spa_mode_global
!= SPA_MODE_UNINIT
) {
2782 mutex_enter(&spa_namespace_lock
);
2783 while ((spa
= spa_next(spa
)) != NULL
)
2784 spa_set_deadman_failmode(spa
, val
);
2785 mutex_exit(&spa_namespace_lock
);
2792 /* Namespace manipulation */
2793 EXPORT_SYMBOL(spa_lookup
);
2794 EXPORT_SYMBOL(spa_add
);
2795 EXPORT_SYMBOL(spa_remove
);
2796 EXPORT_SYMBOL(spa_next
);
2798 /* Refcount functions */
2799 EXPORT_SYMBOL(spa_open_ref
);
2800 EXPORT_SYMBOL(spa_close
);
2801 EXPORT_SYMBOL(spa_refcount_zero
);
2803 /* Pool configuration lock */
2804 EXPORT_SYMBOL(spa_config_tryenter
);
2805 EXPORT_SYMBOL(spa_config_enter
);
2806 EXPORT_SYMBOL(spa_config_exit
);
2807 EXPORT_SYMBOL(spa_config_held
);
2809 /* Pool vdev add/remove lock */
2810 EXPORT_SYMBOL(spa_vdev_enter
);
2811 EXPORT_SYMBOL(spa_vdev_exit
);
2813 /* Pool vdev state change lock */
2814 EXPORT_SYMBOL(spa_vdev_state_enter
);
2815 EXPORT_SYMBOL(spa_vdev_state_exit
);
2817 /* Accessor functions */
2818 EXPORT_SYMBOL(spa_shutting_down
);
2819 EXPORT_SYMBOL(spa_get_dsl
);
2820 EXPORT_SYMBOL(spa_get_rootblkptr
);
2821 EXPORT_SYMBOL(spa_set_rootblkptr
);
2822 EXPORT_SYMBOL(spa_altroot
);
2823 EXPORT_SYMBOL(spa_sync_pass
);
2824 EXPORT_SYMBOL(spa_name
);
2825 EXPORT_SYMBOL(spa_guid
);
2826 EXPORT_SYMBOL(spa_last_synced_txg
);
2827 EXPORT_SYMBOL(spa_first_txg
);
2828 EXPORT_SYMBOL(spa_syncing_txg
);
2829 EXPORT_SYMBOL(spa_version
);
2830 EXPORT_SYMBOL(spa_state
);
2831 EXPORT_SYMBOL(spa_load_state
);
2832 EXPORT_SYMBOL(spa_freeze_txg
);
2833 EXPORT_SYMBOL(spa_get_dspace
);
2834 EXPORT_SYMBOL(spa_update_dspace
);
2835 EXPORT_SYMBOL(spa_deflate
);
2836 EXPORT_SYMBOL(spa_normal_class
);
2837 EXPORT_SYMBOL(spa_log_class
);
2838 EXPORT_SYMBOL(spa_special_class
);
2839 EXPORT_SYMBOL(spa_preferred_class
);
2840 EXPORT_SYMBOL(spa_max_replication
);
2841 EXPORT_SYMBOL(spa_prev_software_version
);
2842 EXPORT_SYMBOL(spa_get_failmode
);
2843 EXPORT_SYMBOL(spa_suspended
);
2844 EXPORT_SYMBOL(spa_bootfs
);
2845 EXPORT_SYMBOL(spa_delegation
);
2846 EXPORT_SYMBOL(spa_meta_objset
);
2847 EXPORT_SYMBOL(spa_maxblocksize
);
2848 EXPORT_SYMBOL(spa_maxdnodesize
);
2850 /* Miscellaneous support routines */
2851 EXPORT_SYMBOL(spa_guid_exists
);
2852 EXPORT_SYMBOL(spa_strdup
);
2853 EXPORT_SYMBOL(spa_strfree
);
2854 EXPORT_SYMBOL(spa_get_random
);
2855 EXPORT_SYMBOL(spa_generate_guid
);
2856 EXPORT_SYMBOL(snprintf_blkptr
);
2857 EXPORT_SYMBOL(spa_freeze
);
2858 EXPORT_SYMBOL(spa_upgrade
);
2859 EXPORT_SYMBOL(spa_evict_all
);
2860 EXPORT_SYMBOL(spa_lookup_by_guid
);
2861 EXPORT_SYMBOL(spa_has_spare
);
2862 EXPORT_SYMBOL(dva_get_dsize_sync
);
2863 EXPORT_SYMBOL(bp_get_dsize_sync
);
2864 EXPORT_SYMBOL(bp_get_dsize
);
2865 EXPORT_SYMBOL(spa_has_slogs
);
2866 EXPORT_SYMBOL(spa_is_root
);
2867 EXPORT_SYMBOL(spa_writeable
);
2868 EXPORT_SYMBOL(spa_mode
);
2869 EXPORT_SYMBOL(spa_namespace_lock
);
2870 EXPORT_SYMBOL(spa_trust_config
);
2871 EXPORT_SYMBOL(spa_missing_tvds_allowed
);
2872 EXPORT_SYMBOL(spa_set_missing_tvds
);
2873 EXPORT_SYMBOL(spa_state_to_name
);
2874 EXPORT_SYMBOL(spa_importing_readonly_checkpoint
);
2875 EXPORT_SYMBOL(spa_min_claim_txg
);
2876 EXPORT_SYMBOL(spa_suspend_async_destroy
);
2877 EXPORT_SYMBOL(spa_has_checkpoint
);
2878 EXPORT_SYMBOL(spa_top_vdevs_spacemap_addressable
);
2880 ZFS_MODULE_PARAM(zfs
, zfs_
, flags
, UINT
, ZMOD_RW
,
2881 "Set additional debugging flags");
2883 ZFS_MODULE_PARAM(zfs
, zfs_
, recover
, INT
, ZMOD_RW
,
2884 "Set to attempt to recover from fatal errors");
2886 ZFS_MODULE_PARAM(zfs
, zfs_
, free_leak_on_eio
, INT
, ZMOD_RW
,
2887 "Set to ignore IO errors during free and permanently leak the space");
2889 ZFS_MODULE_PARAM(zfs
, zfs_
, deadman_checktime_ms
, ULONG
, ZMOD_RW
,
2890 "Dead I/O check interval in milliseconds");
2892 ZFS_MODULE_PARAM(zfs
, zfs_
, deadman_enabled
, INT
, ZMOD_RW
,
2893 "Enable deadman timer");
2895 ZFS_MODULE_PARAM(zfs_spa
, spa_
, asize_inflation
, INT
, ZMOD_RW
,
2896 "SPA size estimate multiplication factor");
2898 ZFS_MODULE_PARAM(zfs
, zfs_
, ddt_data_is_special
, INT
, ZMOD_RW
,
2899 "Place DDT data into the special class");
2901 ZFS_MODULE_PARAM(zfs
, zfs_
, user_indirect_is_special
, INT
, ZMOD_RW
,
2902 "Place user data indirect blocks into the special class");
2905 ZFS_MODULE_PARAM_CALL(zfs_deadman
, zfs_deadman_
, failmode
,
2906 param_set_deadman_failmode
, param_get_charp
, ZMOD_RW
,
2907 "Failmode for deadman timer");
2909 ZFS_MODULE_PARAM_CALL(zfs_deadman
, zfs_deadman_
, synctime_ms
,
2910 param_set_deadman_synctime
, param_get_ulong
, ZMOD_RW
,
2911 "Pool sync expiration time in milliseconds");
2913 ZFS_MODULE_PARAM_CALL(zfs_deadman
, zfs_deadman_
, ziotime_ms
,
2914 param_set_deadman_ziotime
, param_get_ulong
, ZMOD_RW
,
2915 "IO expiration time in milliseconds");
2917 ZFS_MODULE_PARAM(zfs
, zfs_
, special_class_metadata_reserve_pct
, INT
, ZMOD_RW
,
2918 "Small file blocks in special vdevs depends on this much "
2919 "free space available");
2922 ZFS_MODULE_PARAM_CALL(zfs_spa
, spa_
, slop_shift
, param_set_slop_shift
,
2923 param_get_int
, ZMOD_RW
, "Reserved free space in pool");