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, 2018 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) 2014 Integros [integros.com]
28 * Copyright (c) 2017 Datto Inc.
31 #include <sys/zfs_context.h>
32 #include <sys/spa_impl.h>
33 #include <sys/spa_boot.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/metaslab.h>
44 #include <sys/uberblock_impl.h>
47 #include <sys/unique.h>
48 #include <sys/dsl_pool.h>
49 #include <sys/dsl_dir.h>
50 #include <sys/dsl_prop.h>
51 #include <sys/dsl_scan.h>
52 #include <sys/fs/zfs.h>
53 #include <sys/metaslab_impl.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 static int spa_active_count
;
234 int spa_max_replication_override
= SPA_DVAS_PER_BP
;
236 static kmutex_t spa_spare_lock
;
237 static avl_tree_t spa_spare_avl
;
238 static kmutex_t spa_l2cache_lock
;
239 static avl_tree_t spa_l2cache_avl
;
241 kmem_cache_t
*spa_buffer_pool
;
246 * Everything except dprintf, spa, and indirect_remap is on by default
249 int zfs_flags
= ~(ZFS_DEBUG_DPRINTF
| ZFS_DEBUG_INDIRECT_REMAP
);
255 * zfs_recover can be set to nonzero to attempt to recover from
256 * otherwise-fatal errors, typically caused by on-disk corruption. When
257 * set, calls to zfs_panic_recover() will turn into warning messages.
258 * This should only be used as a last resort, as it typically results
259 * in leaked space, or worse.
261 boolean_t zfs_recover
= B_FALSE
;
264 * If destroy encounters an EIO while reading metadata (e.g. indirect
265 * blocks), space referenced by the missing metadata can not be freed.
266 * Normally this causes the background destroy to become "stalled", as
267 * it is unable to make forward progress. While in this stalled state,
268 * all remaining space to free from the error-encountering filesystem is
269 * "temporarily leaked". Set this flag to cause it to ignore the EIO,
270 * permanently leak the space from indirect blocks that can not be read,
271 * and continue to free everything else that it can.
273 * The default, "stalling" behavior is useful if the storage partially
274 * fails (i.e. some but not all i/os fail), and then later recovers. In
275 * this case, we will be able to continue pool operations while it is
276 * partially failed, and when it recovers, we can continue to free the
277 * space, with no leaks. However, note that this case is actually
280 * Typically pools either (a) fail completely (but perhaps temporarily,
281 * e.g. a top-level vdev going offline), or (b) have localized,
282 * permanent errors (e.g. disk returns the wrong data due to bit flip or
283 * firmware bug). In case (a), this setting does not matter because the
284 * pool will be suspended and the sync thread will not be able to make
285 * forward progress regardless. In case (b), because the error is
286 * permanent, the best we can do is leak the minimum amount of space,
287 * which is what setting this flag will do. Therefore, it is reasonable
288 * for this flag to normally be set, but we chose the more conservative
289 * approach of not setting it, so that there is no possibility of
290 * leaking space in the "partial temporary" failure case.
292 boolean_t zfs_free_leak_on_eio
= B_FALSE
;
295 * Expiration time in milliseconds. This value has two meanings. First it is
296 * used to determine when the spa_deadman() logic should fire. By default the
297 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
298 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
299 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
302 uint64_t zfs_deadman_synctime_ms
= 1000000ULL;
305 * Check time in milliseconds. This defines the frequency at which we check
308 uint64_t zfs_deadman_checktime_ms
= 5000ULL;
311 * Override the zfs deadman behavior via /etc/system. By default the
312 * deadman is enabled except on VMware and sparc deployments.
314 int zfs_deadman_enabled
= -1;
317 * The worst case is single-sector max-parity RAID-Z blocks, in which
318 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
319 * times the size; so just assume that. Add to this the fact that
320 * we can have up to 3 DVAs per bp, and one more factor of 2 because
321 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together,
323 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
325 int spa_asize_inflation
= 24;
328 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
329 * the pool to be consumed. This ensures that we don't run the pool
330 * completely out of space, due to unaccounted changes (e.g. to the MOS).
331 * It also limits the worst-case time to allocate space. If we have
332 * less than this amount of free space, most ZPL operations (e.g. write,
333 * create) will return ENOSPC.
335 * Certain operations (e.g. file removal, most administrative actions) can
336 * use half the slop space. They will only return ENOSPC if less than half
337 * the slop space is free. Typically, once the pool has less than the slop
338 * space free, the user will use these operations to free up space in the pool.
339 * These are the operations that call dsl_pool_adjustedsize() with the netfree
340 * argument set to TRUE.
342 * Operations that are almost guaranteed to free up space in the absence of
343 * a pool checkpoint can use up to three quarters of the slop space
346 * A very restricted set of operations are always permitted, regardless of
347 * the amount of free space. These are the operations that call
348 * dsl_sync_task(ZFS_SPACE_CHECK_NONE). If these operations result in a net
349 * increase in the amount of space used, it is possible to run the pool
350 * completely out of space, causing it to be permanently read-only.
352 * Note that on very small pools, the slop space will be larger than
353 * 3.2%, in an effort to have it be at least spa_min_slop (128MB),
354 * but we never allow it to be more than half the pool size.
356 * See also the comments in zfs_space_check_t.
358 int spa_slop_shift
= 5;
359 uint64_t spa_min_slop
= 128 * 1024 * 1024;
361 int spa_allocators
= 4;
365 spa_load_failed(spa_t
*spa
, const char *fmt
, ...)
371 (void) vsnprintf(buf
, sizeof (buf
), fmt
, adx
);
374 zfs_dbgmsg("spa_load(%s, config %s): FAILED: %s", spa
->spa_name
,
375 spa
->spa_trust_config
? "trusted" : "untrusted", buf
);
380 spa_load_note(spa_t
*spa
, const char *fmt
, ...)
386 (void) vsnprintf(buf
, sizeof (buf
), fmt
, adx
);
389 zfs_dbgmsg("spa_load(%s, config %s): %s", spa
->spa_name
,
390 spa
->spa_trust_config
? "trusted" : "untrusted", buf
);
394 * ==========================================================================
396 * ==========================================================================
399 spa_config_lock_init(spa_t
*spa
)
401 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
402 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
403 mutex_init(&scl
->scl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
404 cv_init(&scl
->scl_cv
, NULL
, CV_DEFAULT
, NULL
);
405 refcount_create_untracked(&scl
->scl_count
);
406 scl
->scl_writer
= NULL
;
407 scl
->scl_write_wanted
= 0;
412 spa_config_lock_destroy(spa_t
*spa
)
414 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
415 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
416 mutex_destroy(&scl
->scl_lock
);
417 cv_destroy(&scl
->scl_cv
);
418 refcount_destroy(&scl
->scl_count
);
419 ASSERT(scl
->scl_writer
== NULL
);
420 ASSERT(scl
->scl_write_wanted
== 0);
425 spa_config_tryenter(spa_t
*spa
, int locks
, void *tag
, krw_t rw
)
427 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
428 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
429 if (!(locks
& (1 << i
)))
431 mutex_enter(&scl
->scl_lock
);
432 if (rw
== RW_READER
) {
433 if (scl
->scl_writer
|| scl
->scl_write_wanted
) {
434 mutex_exit(&scl
->scl_lock
);
435 spa_config_exit(spa
, locks
& ((1 << i
) - 1),
440 ASSERT(scl
->scl_writer
!= curthread
);
441 if (!refcount_is_zero(&scl
->scl_count
)) {
442 mutex_exit(&scl
->scl_lock
);
443 spa_config_exit(spa
, locks
& ((1 << i
) - 1),
447 scl
->scl_writer
= curthread
;
449 (void) refcount_add(&scl
->scl_count
, tag
);
450 mutex_exit(&scl
->scl_lock
);
456 spa_config_enter(spa_t
*spa
, int locks
, void *tag
, krw_t rw
)
460 ASSERT3U(SCL_LOCKS
, <, sizeof (wlocks_held
) * NBBY
);
462 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
463 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
464 if (scl
->scl_writer
== curthread
)
465 wlocks_held
|= (1 << i
);
466 if (!(locks
& (1 << i
)))
468 mutex_enter(&scl
->scl_lock
);
469 if (rw
== RW_READER
) {
470 while (scl
->scl_writer
|| scl
->scl_write_wanted
) {
471 cv_wait(&scl
->scl_cv
, &scl
->scl_lock
);
474 ASSERT(scl
->scl_writer
!= curthread
);
475 while (!refcount_is_zero(&scl
->scl_count
)) {
476 scl
->scl_write_wanted
++;
477 cv_wait(&scl
->scl_cv
, &scl
->scl_lock
);
478 scl
->scl_write_wanted
--;
480 scl
->scl_writer
= curthread
;
482 (void) refcount_add(&scl
->scl_count
, tag
);
483 mutex_exit(&scl
->scl_lock
);
485 ASSERT3U(wlocks_held
, <=, locks
);
489 spa_config_exit(spa_t
*spa
, int locks
, void *tag
)
491 for (int i
= SCL_LOCKS
- 1; i
>= 0; i
--) {
492 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
493 if (!(locks
& (1 << i
)))
495 mutex_enter(&scl
->scl_lock
);
496 ASSERT(!refcount_is_zero(&scl
->scl_count
));
497 if (refcount_remove(&scl
->scl_count
, tag
) == 0) {
498 ASSERT(scl
->scl_writer
== NULL
||
499 scl
->scl_writer
== curthread
);
500 scl
->scl_writer
= NULL
; /* OK in either case */
501 cv_broadcast(&scl
->scl_cv
);
503 mutex_exit(&scl
->scl_lock
);
508 spa_config_held(spa_t
*spa
, int locks
, krw_t rw
)
512 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
513 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
514 if (!(locks
& (1 << i
)))
516 if ((rw
== RW_READER
&& !refcount_is_zero(&scl
->scl_count
)) ||
517 (rw
== RW_WRITER
&& scl
->scl_writer
== curthread
))
518 locks_held
|= 1 << i
;
525 * ==========================================================================
526 * SPA namespace functions
527 * ==========================================================================
531 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
532 * Returns NULL if no matching spa_t is found.
535 spa_lookup(const char *name
)
537 static spa_t search
; /* spa_t is large; don't allocate on stack */
542 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
544 (void) strlcpy(search
.spa_name
, name
, sizeof (search
.spa_name
));
547 * If it's a full dataset name, figure out the pool name and
550 cp
= strpbrk(search
.spa_name
, "/@#");
554 spa
= avl_find(&spa_namespace_avl
, &search
, &where
);
560 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
561 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
562 * looking for potentially hung I/Os.
565 spa_deadman(void *arg
)
570 * Disable the deadman timer if the pool is suspended.
572 if (spa_suspended(spa
)) {
573 VERIFY(cyclic_reprogram(spa
->spa_deadman_cycid
, CY_INFINITY
));
577 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
578 (gethrtime() - spa
->spa_sync_starttime
) / NANOSEC
,
579 ++spa
->spa_deadman_calls
);
580 if (zfs_deadman_enabled
)
581 vdev_deadman(spa
->spa_root_vdev
);
585 * Create an uninitialized spa_t with the given name. Requires
586 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
587 * exist by calling spa_lookup() first.
590 spa_add(const char *name
, nvlist_t
*config
, const char *altroot
)
593 spa_config_dirent_t
*dp
;
597 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
599 spa
= kmem_zalloc(sizeof (spa_t
), KM_SLEEP
);
601 mutex_init(&spa
->spa_async_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
602 mutex_init(&spa
->spa_errlist_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
603 mutex_init(&spa
->spa_errlog_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
604 mutex_init(&spa
->spa_evicting_os_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
605 mutex_init(&spa
->spa_history_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
606 mutex_init(&spa
->spa_proc_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
607 mutex_init(&spa
->spa_props_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
608 mutex_init(&spa
->spa_cksum_tmpls_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
609 mutex_init(&spa
->spa_scrub_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
610 mutex_init(&spa
->spa_suspend_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
611 mutex_init(&spa
->spa_vdev_top_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
612 mutex_init(&spa
->spa_iokstat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
614 cv_init(&spa
->spa_async_cv
, NULL
, CV_DEFAULT
, NULL
);
615 cv_init(&spa
->spa_evicting_os_cv
, NULL
, CV_DEFAULT
, NULL
);
616 cv_init(&spa
->spa_proc_cv
, NULL
, CV_DEFAULT
, NULL
);
617 cv_init(&spa
->spa_scrub_io_cv
, NULL
, CV_DEFAULT
, NULL
);
618 cv_init(&spa
->spa_suspend_cv
, NULL
, CV_DEFAULT
, NULL
);
620 for (int t
= 0; t
< TXG_SIZE
; t
++)
621 bplist_create(&spa
->spa_free_bplist
[t
]);
623 (void) strlcpy(spa
->spa_name
, name
, sizeof (spa
->spa_name
));
624 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
625 spa
->spa_freeze_txg
= UINT64_MAX
;
626 spa
->spa_final_txg
= UINT64_MAX
;
627 spa
->spa_load_max_txg
= UINT64_MAX
;
629 spa
->spa_proc_state
= SPA_PROC_NONE
;
630 spa
->spa_trust_config
= B_TRUE
;
632 hdlr
.cyh_func
= spa_deadman
;
634 hdlr
.cyh_level
= CY_LOW_LEVEL
;
636 spa
->spa_deadman_synctime
= MSEC2NSEC(zfs_deadman_synctime_ms
);
639 * This determines how often we need to check for hung I/Os after
640 * the cyclic has already fired. Since checking for hung I/Os is
641 * an expensive operation we don't want to check too frequently.
642 * Instead wait for 5 seconds before checking again.
644 when
.cyt_interval
= MSEC2NSEC(zfs_deadman_checktime_ms
);
645 when
.cyt_when
= CY_INFINITY
;
646 mutex_enter(&cpu_lock
);
647 spa
->spa_deadman_cycid
= cyclic_add(&hdlr
, &when
);
648 mutex_exit(&cpu_lock
);
650 refcount_create(&spa
->spa_refcount
);
651 spa_config_lock_init(spa
);
653 avl_add(&spa_namespace_avl
, spa
);
656 * Set the alternate root, if there is one.
659 spa
->spa_root
= spa_strdup(altroot
);
663 spa
->spa_alloc_count
= spa_allocators
;
664 spa
->spa_alloc_locks
= kmem_zalloc(spa
->spa_alloc_count
*
665 sizeof (kmutex_t
), KM_SLEEP
);
666 spa
->spa_alloc_trees
= kmem_zalloc(spa
->spa_alloc_count
*
667 sizeof (avl_tree_t
), KM_SLEEP
);
668 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++) {
669 mutex_init(&spa
->spa_alloc_locks
[i
], NULL
, MUTEX_DEFAULT
, NULL
);
670 avl_create(&spa
->spa_alloc_trees
[i
], zio_bookmark_compare
,
671 sizeof (zio_t
), offsetof(zio_t
, io_alloc_node
));
675 * Every pool starts with the default cachefile
677 list_create(&spa
->spa_config_list
, sizeof (spa_config_dirent_t
),
678 offsetof(spa_config_dirent_t
, scd_link
));
680 dp
= kmem_zalloc(sizeof (spa_config_dirent_t
), KM_SLEEP
);
681 dp
->scd_path
= altroot
? NULL
: spa_strdup(spa_config_path
);
682 list_insert_head(&spa
->spa_config_list
, dp
);
684 VERIFY(nvlist_alloc(&spa
->spa_load_info
, NV_UNIQUE_NAME
,
687 if (config
!= NULL
) {
690 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_FEATURES_FOR_READ
,
692 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
,
696 VERIFY(nvlist_dup(config
, &spa
->spa_config
, 0) == 0);
699 if (spa
->spa_label_features
== NULL
) {
700 VERIFY(nvlist_alloc(&spa
->spa_label_features
, NV_UNIQUE_NAME
,
704 spa
->spa_iokstat
= kstat_create("zfs", 0, name
,
705 "disk", KSTAT_TYPE_IO
, 1, 0);
706 if (spa
->spa_iokstat
) {
707 spa
->spa_iokstat
->ks_lock
= &spa
->spa_iokstat_lock
;
708 kstat_install(spa
->spa_iokstat
);
711 spa
->spa_min_ashift
= INT_MAX
;
712 spa
->spa_max_ashift
= 0;
715 * As a pool is being created, treat all features as disabled by
716 * setting SPA_FEATURE_DISABLED for all entries in the feature
719 for (int i
= 0; i
< SPA_FEATURES
; i
++) {
720 spa
->spa_feat_refcount_cache
[i
] = SPA_FEATURE_DISABLED
;
727 * Removes a spa_t from the namespace, freeing up any memory used. Requires
728 * spa_namespace_lock. This is called only after the spa_t has been closed and
732 spa_remove(spa_t
*spa
)
734 spa_config_dirent_t
*dp
;
736 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
737 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
738 ASSERT3U(refcount_count(&spa
->spa_refcount
), ==, 0);
740 nvlist_free(spa
->spa_config_splitting
);
742 avl_remove(&spa_namespace_avl
, spa
);
743 cv_broadcast(&spa_namespace_cv
);
746 spa_strfree(spa
->spa_root
);
750 while ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
751 list_remove(&spa
->spa_config_list
, dp
);
752 if (dp
->scd_path
!= NULL
)
753 spa_strfree(dp
->scd_path
);
754 kmem_free(dp
, sizeof (spa_config_dirent_t
));
757 for (int i
= 0; i
< spa
->spa_alloc_count
; i
++) {
758 avl_destroy(&spa
->spa_alloc_trees
[i
]);
759 mutex_destroy(&spa
->spa_alloc_locks
[i
]);
761 kmem_free(spa
->spa_alloc_locks
, spa
->spa_alloc_count
*
763 kmem_free(spa
->spa_alloc_trees
, spa
->spa_alloc_count
*
764 sizeof (avl_tree_t
));
766 list_destroy(&spa
->spa_config_list
);
768 nvlist_free(spa
->spa_label_features
);
769 nvlist_free(spa
->spa_load_info
);
770 spa_config_set(spa
, NULL
);
772 mutex_enter(&cpu_lock
);
773 if (spa
->spa_deadman_cycid
!= CYCLIC_NONE
)
774 cyclic_remove(spa
->spa_deadman_cycid
);
775 mutex_exit(&cpu_lock
);
776 spa
->spa_deadman_cycid
= CYCLIC_NONE
;
778 refcount_destroy(&spa
->spa_refcount
);
780 spa_config_lock_destroy(spa
);
782 kstat_delete(spa
->spa_iokstat
);
783 spa
->spa_iokstat
= NULL
;
785 for (int t
= 0; t
< TXG_SIZE
; t
++)
786 bplist_destroy(&spa
->spa_free_bplist
[t
]);
788 zio_checksum_templates_free(spa
);
790 cv_destroy(&spa
->spa_async_cv
);
791 cv_destroy(&spa
->spa_evicting_os_cv
);
792 cv_destroy(&spa
->spa_proc_cv
);
793 cv_destroy(&spa
->spa_scrub_io_cv
);
794 cv_destroy(&spa
->spa_suspend_cv
);
796 mutex_destroy(&spa
->spa_async_lock
);
797 mutex_destroy(&spa
->spa_errlist_lock
);
798 mutex_destroy(&spa
->spa_errlog_lock
);
799 mutex_destroy(&spa
->spa_evicting_os_lock
);
800 mutex_destroy(&spa
->spa_history_lock
);
801 mutex_destroy(&spa
->spa_proc_lock
);
802 mutex_destroy(&spa
->spa_props_lock
);
803 mutex_destroy(&spa
->spa_cksum_tmpls_lock
);
804 mutex_destroy(&spa
->spa_scrub_lock
);
805 mutex_destroy(&spa
->spa_suspend_lock
);
806 mutex_destroy(&spa
->spa_vdev_top_lock
);
807 mutex_destroy(&spa
->spa_iokstat_lock
);
809 kmem_free(spa
, sizeof (spa_t
));
813 * Given a pool, return the next pool in the namespace, or NULL if there is
814 * none. If 'prev' is NULL, return the first pool.
817 spa_next(spa_t
*prev
)
819 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
822 return (AVL_NEXT(&spa_namespace_avl
, prev
));
824 return (avl_first(&spa_namespace_avl
));
828 * ==========================================================================
829 * SPA refcount functions
830 * ==========================================================================
834 * Add a reference to the given spa_t. Must have at least one reference, or
835 * have the namespace lock held.
838 spa_open_ref(spa_t
*spa
, void *tag
)
840 ASSERT(refcount_count(&spa
->spa_refcount
) >= spa
->spa_minref
||
841 MUTEX_HELD(&spa_namespace_lock
));
842 (void) refcount_add(&spa
->spa_refcount
, tag
);
846 * Remove a reference to the given spa_t. Must have at least one reference, or
847 * have the namespace lock held.
850 spa_close(spa_t
*spa
, void *tag
)
852 ASSERT(refcount_count(&spa
->spa_refcount
) > spa
->spa_minref
||
853 MUTEX_HELD(&spa_namespace_lock
));
854 (void) refcount_remove(&spa
->spa_refcount
, tag
);
858 * Remove a reference to the given spa_t held by a dsl dir that is
859 * being asynchronously released. Async releases occur from a taskq
860 * performing eviction of dsl datasets and dirs. The namespace lock
861 * isn't held and the hold by the object being evicted may contribute to
862 * spa_minref (e.g. dataset or directory released during pool export),
863 * so the asserts in spa_close() do not apply.
866 spa_async_close(spa_t
*spa
, void *tag
)
868 (void) refcount_remove(&spa
->spa_refcount
, tag
);
872 * Check to see if the spa refcount is zero. Must be called with
873 * spa_namespace_lock held. We really compare against spa_minref, which is the
874 * number of references acquired when opening a pool
877 spa_refcount_zero(spa_t
*spa
)
879 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
881 return (refcount_count(&spa
->spa_refcount
) == spa
->spa_minref
);
885 * ==========================================================================
886 * SPA spare and l2cache tracking
887 * ==========================================================================
891 * Hot spares and cache devices are tracked using the same code below,
892 * for 'auxiliary' devices.
895 typedef struct spa_aux
{
903 spa_aux_compare(const void *a
, const void *b
)
905 const spa_aux_t
*sa
= a
;
906 const spa_aux_t
*sb
= b
;
908 if (sa
->aux_guid
< sb
->aux_guid
)
910 else if (sa
->aux_guid
> sb
->aux_guid
)
917 spa_aux_add(vdev_t
*vd
, avl_tree_t
*avl
)
923 search
.aux_guid
= vd
->vdev_guid
;
924 if ((aux
= avl_find(avl
, &search
, &where
)) != NULL
) {
927 aux
= kmem_zalloc(sizeof (spa_aux_t
), KM_SLEEP
);
928 aux
->aux_guid
= vd
->vdev_guid
;
930 avl_insert(avl
, aux
, where
);
935 spa_aux_remove(vdev_t
*vd
, avl_tree_t
*avl
)
941 search
.aux_guid
= vd
->vdev_guid
;
942 aux
= avl_find(avl
, &search
, &where
);
946 if (--aux
->aux_count
== 0) {
947 avl_remove(avl
, aux
);
948 kmem_free(aux
, sizeof (spa_aux_t
));
949 } else if (aux
->aux_pool
== spa_guid(vd
->vdev_spa
)) {
950 aux
->aux_pool
= 0ULL;
955 spa_aux_exists(uint64_t guid
, uint64_t *pool
, int *refcnt
, avl_tree_t
*avl
)
957 spa_aux_t search
, *found
;
959 search
.aux_guid
= guid
;
960 found
= avl_find(avl
, &search
, NULL
);
964 *pool
= found
->aux_pool
;
971 *refcnt
= found
->aux_count
;
976 return (found
!= NULL
);
980 spa_aux_activate(vdev_t
*vd
, avl_tree_t
*avl
)
982 spa_aux_t search
, *found
;
985 search
.aux_guid
= vd
->vdev_guid
;
986 found
= avl_find(avl
, &search
, &where
);
987 ASSERT(found
!= NULL
);
988 ASSERT(found
->aux_pool
== 0ULL);
990 found
->aux_pool
= spa_guid(vd
->vdev_spa
);
994 * Spares are tracked globally due to the following constraints:
996 * - A spare may be part of multiple pools.
997 * - A spare may be added to a pool even if it's actively in use within
999 * - A spare in use in any pool can only be the source of a replacement if
1000 * the target is a spare in the same pool.
1002 * We keep track of all spares on the system through the use of a reference
1003 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
1004 * spare, then we bump the reference count in the AVL tree. In addition, we set
1005 * the 'vdev_isspare' member to indicate that the device is a spare (active or
1006 * inactive). When a spare is made active (used to replace a device in the
1007 * pool), we also keep track of which pool its been made a part of.
1009 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
1010 * called under the spa_namespace lock as part of vdev reconfiguration. The
1011 * separate spare lock exists for the status query path, which does not need to
1012 * be completely consistent with respect to other vdev configuration changes.
1016 spa_spare_compare(const void *a
, const void *b
)
1018 return (spa_aux_compare(a
, b
));
1022 spa_spare_add(vdev_t
*vd
)
1024 mutex_enter(&spa_spare_lock
);
1025 ASSERT(!vd
->vdev_isspare
);
1026 spa_aux_add(vd
, &spa_spare_avl
);
1027 vd
->vdev_isspare
= B_TRUE
;
1028 mutex_exit(&spa_spare_lock
);
1032 spa_spare_remove(vdev_t
*vd
)
1034 mutex_enter(&spa_spare_lock
);
1035 ASSERT(vd
->vdev_isspare
);
1036 spa_aux_remove(vd
, &spa_spare_avl
);
1037 vd
->vdev_isspare
= B_FALSE
;
1038 mutex_exit(&spa_spare_lock
);
1042 spa_spare_exists(uint64_t guid
, uint64_t *pool
, int *refcnt
)
1046 mutex_enter(&spa_spare_lock
);
1047 found
= spa_aux_exists(guid
, pool
, refcnt
, &spa_spare_avl
);
1048 mutex_exit(&spa_spare_lock
);
1054 spa_spare_activate(vdev_t
*vd
)
1056 mutex_enter(&spa_spare_lock
);
1057 ASSERT(vd
->vdev_isspare
);
1058 spa_aux_activate(vd
, &spa_spare_avl
);
1059 mutex_exit(&spa_spare_lock
);
1063 * Level 2 ARC devices are tracked globally for the same reasons as spares.
1064 * Cache devices currently only support one pool per cache device, and so
1065 * for these devices the aux reference count is currently unused beyond 1.
1069 spa_l2cache_compare(const void *a
, const void *b
)
1071 return (spa_aux_compare(a
, b
));
1075 spa_l2cache_add(vdev_t
*vd
)
1077 mutex_enter(&spa_l2cache_lock
);
1078 ASSERT(!vd
->vdev_isl2cache
);
1079 spa_aux_add(vd
, &spa_l2cache_avl
);
1080 vd
->vdev_isl2cache
= B_TRUE
;
1081 mutex_exit(&spa_l2cache_lock
);
1085 spa_l2cache_remove(vdev_t
*vd
)
1087 mutex_enter(&spa_l2cache_lock
);
1088 ASSERT(vd
->vdev_isl2cache
);
1089 spa_aux_remove(vd
, &spa_l2cache_avl
);
1090 vd
->vdev_isl2cache
= B_FALSE
;
1091 mutex_exit(&spa_l2cache_lock
);
1095 spa_l2cache_exists(uint64_t guid
, uint64_t *pool
)
1099 mutex_enter(&spa_l2cache_lock
);
1100 found
= spa_aux_exists(guid
, pool
, NULL
, &spa_l2cache_avl
);
1101 mutex_exit(&spa_l2cache_lock
);
1107 spa_l2cache_activate(vdev_t
*vd
)
1109 mutex_enter(&spa_l2cache_lock
);
1110 ASSERT(vd
->vdev_isl2cache
);
1111 spa_aux_activate(vd
, &spa_l2cache_avl
);
1112 mutex_exit(&spa_l2cache_lock
);
1116 * ==========================================================================
1118 * ==========================================================================
1122 * Lock the given spa_t for the purpose of adding or removing a vdev.
1123 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1124 * It returns the next transaction group for the spa_t.
1127 spa_vdev_enter(spa_t
*spa
)
1129 mutex_enter(&spa
->spa_vdev_top_lock
);
1130 mutex_enter(&spa_namespace_lock
);
1131 return (spa_vdev_config_enter(spa
));
1135 * Internal implementation for spa_vdev_enter(). Used when a vdev
1136 * operation requires multiple syncs (i.e. removing a device) while
1137 * keeping the spa_namespace_lock held.
1140 spa_vdev_config_enter(spa_t
*spa
)
1142 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1144 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
1146 return (spa_last_synced_txg(spa
) + 1);
1150 * Used in combination with spa_vdev_config_enter() to allow the syncing
1151 * of multiple transactions without releasing the spa_namespace_lock.
1154 spa_vdev_config_exit(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int error
, char *tag
)
1156 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1158 int config_changed
= B_FALSE
;
1160 ASSERT(txg
> spa_last_synced_txg(spa
));
1162 spa
->spa_pending_vdev
= NULL
;
1165 * Reassess the DTLs.
1167 vdev_dtl_reassess(spa
->spa_root_vdev
, 0, 0, B_FALSE
);
1169 if (error
== 0 && !list_is_empty(&spa
->spa_config_dirty_list
)) {
1170 config_changed
= B_TRUE
;
1171 spa
->spa_config_generation
++;
1175 * Verify the metaslab classes.
1177 ASSERT(metaslab_class_validate(spa_normal_class(spa
)) == 0);
1178 ASSERT(metaslab_class_validate(spa_log_class(spa
)) == 0);
1180 spa_config_exit(spa
, SCL_ALL
, spa
);
1183 * Panic the system if the specified tag requires it. This
1184 * is useful for ensuring that configurations are updated
1187 if (zio_injection_enabled
)
1188 zio_handle_panic_injection(spa
, tag
, 0);
1191 * Note: this txg_wait_synced() is important because it ensures
1192 * that there won't be more than one config change per txg.
1193 * This allows us to use the txg as the generation number.
1196 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
1199 ASSERT(!vd
->vdev_detached
|| vd
->vdev_dtl_sm
== NULL
);
1200 if (vd
->vdev_ops
->vdev_op_leaf
) {
1201 mutex_enter(&vd
->vdev_initialize_lock
);
1202 vdev_initialize_stop(vd
, VDEV_INITIALIZE_CANCELED
);
1203 mutex_exit(&vd
->vdev_initialize_lock
);
1206 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
1208 spa_config_exit(spa
, SCL_ALL
, spa
);
1212 * If the config changed, update the config cache.
1215 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
1219 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1220 * locking of spa_vdev_enter(), we also want make sure the transactions have
1221 * synced to disk, and then update the global configuration cache with the new
1225 spa_vdev_exit(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int error
)
1227 spa_vdev_config_exit(spa
, vd
, txg
, error
, FTAG
);
1228 mutex_exit(&spa_namespace_lock
);
1229 mutex_exit(&spa
->spa_vdev_top_lock
);
1235 * Lock the given spa_t for the purpose of changing vdev state.
1238 spa_vdev_state_enter(spa_t
*spa
, int oplocks
)
1240 int locks
= SCL_STATE_ALL
| oplocks
;
1243 * Root pools may need to read of the underlying devfs filesystem
1244 * when opening up a vdev. Unfortunately if we're holding the
1245 * SCL_ZIO lock it will result in a deadlock when we try to issue
1246 * the read from the root filesystem. Instead we "prefetch"
1247 * the associated vnodes that we need prior to opening the
1248 * underlying devices and cache them so that we can prevent
1249 * any I/O when we are doing the actual open.
1251 if (spa_is_root(spa
)) {
1252 int low
= locks
& ~(SCL_ZIO
- 1);
1253 int high
= locks
& ~low
;
1255 spa_config_enter(spa
, high
, spa
, RW_WRITER
);
1256 vdev_hold(spa
->spa_root_vdev
);
1257 spa_config_enter(spa
, low
, spa
, RW_WRITER
);
1259 spa_config_enter(spa
, locks
, spa
, RW_WRITER
);
1261 spa
->spa_vdev_locks
= locks
;
1265 spa_vdev_state_exit(spa_t
*spa
, vdev_t
*vd
, int error
)
1267 boolean_t config_changed
= B_FALSE
;
1269 if (vd
!= NULL
|| error
== 0)
1270 vdev_dtl_reassess(vd
? vd
->vdev_top
: spa
->spa_root_vdev
,
1274 vdev_state_dirty(vd
->vdev_top
);
1275 config_changed
= B_TRUE
;
1276 spa
->spa_config_generation
++;
1279 if (spa_is_root(spa
))
1280 vdev_rele(spa
->spa_root_vdev
);
1282 ASSERT3U(spa
->spa_vdev_locks
, >=, SCL_STATE_ALL
);
1283 spa_config_exit(spa
, spa
->spa_vdev_locks
, spa
);
1286 * If anything changed, wait for it to sync. This ensures that,
1287 * from the system administrator's perspective, zpool(8) commands
1288 * are synchronous. This is important for things like zpool offline:
1289 * when the command completes, you expect no further I/O from ZFS.
1292 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1295 * If the config changed, update the config cache.
1297 if (config_changed
) {
1298 mutex_enter(&spa_namespace_lock
);
1299 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
1300 mutex_exit(&spa_namespace_lock
);
1307 * ==========================================================================
1308 * Miscellaneous functions
1309 * ==========================================================================
1313 spa_activate_mos_feature(spa_t
*spa
, const char *feature
, dmu_tx_t
*tx
)
1315 if (!nvlist_exists(spa
->spa_label_features
, feature
)) {
1316 fnvlist_add_boolean(spa
->spa_label_features
, feature
);
1318 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1319 * dirty the vdev config because lock SCL_CONFIG is not held.
1320 * Thankfully, in this case we don't need to dirty the config
1321 * because it will be written out anyway when we finish
1322 * creating the pool.
1324 if (tx
->tx_txg
!= TXG_INITIAL
)
1325 vdev_config_dirty(spa
->spa_root_vdev
);
1330 spa_deactivate_mos_feature(spa_t
*spa
, const char *feature
)
1332 if (nvlist_remove_all(spa
->spa_label_features
, feature
) == 0)
1333 vdev_config_dirty(spa
->spa_root_vdev
);
1340 spa_rename(const char *name
, const char *newname
)
1346 * Lookup the spa_t and grab the config lock for writing. We need to
1347 * actually open the pool so that we can sync out the necessary labels.
1348 * It's OK to call spa_open() with the namespace lock held because we
1349 * allow recursive calls for other reasons.
1351 mutex_enter(&spa_namespace_lock
);
1352 if ((err
= spa_open(name
, &spa
, FTAG
)) != 0) {
1353 mutex_exit(&spa_namespace_lock
);
1357 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1359 avl_remove(&spa_namespace_avl
, spa
);
1360 (void) strlcpy(spa
->spa_name
, newname
, sizeof (spa
->spa_name
));
1361 avl_add(&spa_namespace_avl
, spa
);
1364 * Sync all labels to disk with the new names by marking the root vdev
1365 * dirty and waiting for it to sync. It will pick up the new pool name
1368 vdev_config_dirty(spa
->spa_root_vdev
);
1370 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1372 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1375 * Sync the updated config cache.
1377 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
1379 spa_close(spa
, FTAG
);
1381 mutex_exit(&spa_namespace_lock
);
1387 * Return the spa_t associated with given pool_guid, if it exists. If
1388 * device_guid is non-zero, determine whether the pool exists *and* contains
1389 * a device with the specified device_guid.
1392 spa_by_guid(uint64_t pool_guid
, uint64_t device_guid
)
1395 avl_tree_t
*t
= &spa_namespace_avl
;
1397 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1399 for (spa
= avl_first(t
); spa
!= NULL
; spa
= AVL_NEXT(t
, spa
)) {
1400 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1402 if (spa
->spa_root_vdev
== NULL
)
1404 if (spa_guid(spa
) == pool_guid
) {
1405 if (device_guid
== 0)
1408 if (vdev_lookup_by_guid(spa
->spa_root_vdev
,
1409 device_guid
) != NULL
)
1413 * Check any devices we may be in the process of adding.
1415 if (spa
->spa_pending_vdev
) {
1416 if (vdev_lookup_by_guid(spa
->spa_pending_vdev
,
1417 device_guid
) != NULL
)
1427 * Determine whether a pool with the given pool_guid exists.
1430 spa_guid_exists(uint64_t pool_guid
, uint64_t device_guid
)
1432 return (spa_by_guid(pool_guid
, device_guid
) != NULL
);
1436 spa_strdup(const char *s
)
1442 new = kmem_alloc(len
+ 1, KM_SLEEP
);
1450 spa_strfree(char *s
)
1452 kmem_free(s
, strlen(s
) + 1);
1456 spa_get_random(uint64_t range
)
1462 (void) random_get_pseudo_bytes((void *)&r
, sizeof (uint64_t));
1468 spa_generate_guid(spa_t
*spa
)
1470 uint64_t guid
= spa_get_random(-1ULL);
1473 while (guid
== 0 || spa_guid_exists(spa_guid(spa
), guid
))
1474 guid
= spa_get_random(-1ULL);
1476 while (guid
== 0 || spa_guid_exists(guid
, 0))
1477 guid
= spa_get_random(-1ULL);
1484 snprintf_blkptr(char *buf
, size_t buflen
, const blkptr_t
*bp
)
1487 char *checksum
= NULL
;
1488 char *compress
= NULL
;
1491 if (BP_GET_TYPE(bp
) & DMU_OT_NEWTYPE
) {
1492 dmu_object_byteswap_t bswap
=
1493 DMU_OT_BYTESWAP(BP_GET_TYPE(bp
));
1494 (void) snprintf(type
, sizeof (type
), "bswap %s %s",
1495 DMU_OT_IS_METADATA(BP_GET_TYPE(bp
)) ?
1496 "metadata" : "data",
1497 dmu_ot_byteswap
[bswap
].ob_name
);
1499 (void) strlcpy(type
, dmu_ot
[BP_GET_TYPE(bp
)].ot_name
,
1502 if (!BP_IS_EMBEDDED(bp
)) {
1504 zio_checksum_table
[BP_GET_CHECKSUM(bp
)].ci_name
;
1506 compress
= zio_compress_table
[BP_GET_COMPRESS(bp
)].ci_name
;
1509 SNPRINTF_BLKPTR(snprintf
, ' ', buf
, buflen
, bp
, type
, checksum
,
1514 spa_freeze(spa_t
*spa
)
1516 uint64_t freeze_txg
= 0;
1518 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1519 if (spa
->spa_freeze_txg
== UINT64_MAX
) {
1520 freeze_txg
= spa_last_synced_txg(spa
) + TXG_SIZE
;
1521 spa
->spa_freeze_txg
= freeze_txg
;
1523 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1524 if (freeze_txg
!= 0)
1525 txg_wait_synced(spa_get_dsl(spa
), freeze_txg
);
1529 zfs_panic_recover(const char *fmt
, ...)
1534 vcmn_err(zfs_recover
? CE_WARN
: CE_PANIC
, fmt
, adx
);
1539 * This is a stripped-down version of strtoull, suitable only for converting
1540 * lowercase hexadecimal numbers that don't overflow.
1543 zfs_strtonum(const char *str
, char **nptr
)
1549 while ((c
= *str
) != '\0') {
1550 if (c
>= '0' && c
<= '9')
1552 else if (c
>= 'a' && c
<= 'f')
1553 digit
= 10 + c
- 'a';
1564 *nptr
= (char *)str
;
1570 * ==========================================================================
1571 * Accessor functions
1572 * ==========================================================================
1576 spa_shutting_down(spa_t
*spa
)
1578 return (spa
->spa_async_suspended
);
1582 spa_get_dsl(spa_t
*spa
)
1584 return (spa
->spa_dsl_pool
);
1588 spa_is_initializing(spa_t
*spa
)
1590 return (spa
->spa_is_initializing
);
1594 spa_indirect_vdevs_loaded(spa_t
*spa
)
1596 return (spa
->spa_indirect_vdevs_loaded
);
1600 spa_get_rootblkptr(spa_t
*spa
)
1602 return (&spa
->spa_ubsync
.ub_rootbp
);
1606 spa_set_rootblkptr(spa_t
*spa
, const blkptr_t
*bp
)
1608 spa
->spa_uberblock
.ub_rootbp
= *bp
;
1612 spa_altroot(spa_t
*spa
, char *buf
, size_t buflen
)
1614 if (spa
->spa_root
== NULL
)
1617 (void) strncpy(buf
, spa
->spa_root
, buflen
);
1621 spa_sync_pass(spa_t
*spa
)
1623 return (spa
->spa_sync_pass
);
1627 spa_name(spa_t
*spa
)
1629 return (spa
->spa_name
);
1633 spa_guid(spa_t
*spa
)
1635 dsl_pool_t
*dp
= spa_get_dsl(spa
);
1639 * If we fail to parse the config during spa_load(), we can go through
1640 * the error path (which posts an ereport) and end up here with no root
1641 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1644 if (spa
->spa_root_vdev
== NULL
)
1645 return (spa
->spa_config_guid
);
1647 guid
= spa
->spa_last_synced_guid
!= 0 ?
1648 spa
->spa_last_synced_guid
: spa
->spa_root_vdev
->vdev_guid
;
1651 * Return the most recently synced out guid unless we're
1652 * in syncing context.
1654 if (dp
&& dsl_pool_sync_context(dp
))
1655 return (spa
->spa_root_vdev
->vdev_guid
);
1661 spa_load_guid(spa_t
*spa
)
1664 * This is a GUID that exists solely as a reference for the
1665 * purposes of the arc. It is generated at load time, and
1666 * is never written to persistent storage.
1668 return (spa
->spa_load_guid
);
1672 spa_last_synced_txg(spa_t
*spa
)
1674 return (spa
->spa_ubsync
.ub_txg
);
1678 spa_first_txg(spa_t
*spa
)
1680 return (spa
->spa_first_txg
);
1684 spa_syncing_txg(spa_t
*spa
)
1686 return (spa
->spa_syncing_txg
);
1690 * Return the last txg where data can be dirtied. The final txgs
1691 * will be used to just clear out any deferred frees that remain.
1694 spa_final_dirty_txg(spa_t
*spa
)
1696 return (spa
->spa_final_txg
- TXG_DEFER_SIZE
);
1700 spa_state(spa_t
*spa
)
1702 return (spa
->spa_state
);
1706 spa_load_state(spa_t
*spa
)
1708 return (spa
->spa_load_state
);
1712 spa_freeze_txg(spa_t
*spa
)
1714 return (spa
->spa_freeze_txg
);
1719 spa_get_worst_case_asize(spa_t
*spa
, uint64_t lsize
)
1721 return (lsize
* spa_asize_inflation
);
1725 * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%),
1726 * or at least 128MB, unless that would cause it to be more than half the
1729 * See the comment above spa_slop_shift for details.
1732 spa_get_slop_space(spa_t
*spa
)
1734 uint64_t space
= spa_get_dspace(spa
);
1735 return (MAX(space
>> spa_slop_shift
, MIN(space
>> 1, spa_min_slop
)));
1739 spa_get_dspace(spa_t
*spa
)
1741 return (spa
->spa_dspace
);
1745 spa_get_checkpoint_space(spa_t
*spa
)
1747 return (spa
->spa_checkpoint_info
.sci_dspace
);
1751 spa_update_dspace(spa_t
*spa
)
1753 spa
->spa_dspace
= metaslab_class_get_dspace(spa_normal_class(spa
)) +
1754 ddt_get_dedup_dspace(spa
);
1755 if (spa
->spa_vdev_removal
!= NULL
) {
1757 * We can't allocate from the removing device, so
1758 * subtract its size. This prevents the DMU/DSL from
1759 * filling up the (now smaller) pool while we are in the
1760 * middle of removing the device.
1762 * Note that the DMU/DSL doesn't actually know or care
1763 * how much space is allocated (it does its own tracking
1764 * of how much space has been logically used). So it
1765 * doesn't matter that the data we are moving may be
1766 * allocated twice (on the old device and the new
1769 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
1771 vdev_lookup_top(spa
, spa
->spa_vdev_removal
->svr_vdev_id
);
1772 spa
->spa_dspace
-= spa_deflate(spa
) ?
1773 vd
->vdev_stat
.vs_dspace
: vd
->vdev_stat
.vs_space
;
1774 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
1779 * Return the failure mode that has been set to this pool. The default
1780 * behavior will be to block all I/Os when a complete failure occurs.
1783 spa_get_failmode(spa_t
*spa
)
1785 return (spa
->spa_failmode
);
1789 spa_suspended(spa_t
*spa
)
1791 return (spa
->spa_suspended
);
1795 spa_version(spa_t
*spa
)
1797 return (spa
->spa_ubsync
.ub_version
);
1801 spa_deflate(spa_t
*spa
)
1803 return (spa
->spa_deflate
);
1807 spa_normal_class(spa_t
*spa
)
1809 return (spa
->spa_normal_class
);
1813 spa_log_class(spa_t
*spa
)
1815 return (spa
->spa_log_class
);
1819 spa_evicting_os_register(spa_t
*spa
, objset_t
*os
)
1821 mutex_enter(&spa
->spa_evicting_os_lock
);
1822 list_insert_head(&spa
->spa_evicting_os_list
, os
);
1823 mutex_exit(&spa
->spa_evicting_os_lock
);
1827 spa_evicting_os_deregister(spa_t
*spa
, objset_t
*os
)
1829 mutex_enter(&spa
->spa_evicting_os_lock
);
1830 list_remove(&spa
->spa_evicting_os_list
, os
);
1831 cv_broadcast(&spa
->spa_evicting_os_cv
);
1832 mutex_exit(&spa
->spa_evicting_os_lock
);
1836 spa_evicting_os_wait(spa_t
*spa
)
1838 mutex_enter(&spa
->spa_evicting_os_lock
);
1839 while (!list_is_empty(&spa
->spa_evicting_os_list
))
1840 cv_wait(&spa
->spa_evicting_os_cv
, &spa
->spa_evicting_os_lock
);
1841 mutex_exit(&spa
->spa_evicting_os_lock
);
1843 dmu_buf_user_evict_wait();
1847 spa_max_replication(spa_t
*spa
)
1850 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1851 * handle BPs with more than one DVA allocated. Set our max
1852 * replication level accordingly.
1854 if (spa_version(spa
) < SPA_VERSION_DITTO_BLOCKS
)
1856 return (MIN(SPA_DVAS_PER_BP
, spa_max_replication_override
));
1860 spa_prev_software_version(spa_t
*spa
)
1862 return (spa
->spa_prev_software_version
);
1866 spa_deadman_synctime(spa_t
*spa
)
1868 return (spa
->spa_deadman_synctime
);
1872 dva_get_dsize_sync(spa_t
*spa
, const dva_t
*dva
)
1874 uint64_t asize
= DVA_GET_ASIZE(dva
);
1875 uint64_t dsize
= asize
;
1877 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1879 if (asize
!= 0 && spa
->spa_deflate
) {
1880 vdev_t
*vd
= vdev_lookup_top(spa
, DVA_GET_VDEV(dva
));
1881 dsize
= (asize
>> SPA_MINBLOCKSHIFT
) * vd
->vdev_deflate_ratio
;
1888 bp_get_dsize_sync(spa_t
*spa
, const blkptr_t
*bp
)
1892 for (int d
= 0; d
< BP_GET_NDVAS(bp
); d
++)
1893 dsize
+= dva_get_dsize_sync(spa
, &bp
->blk_dva
[d
]);
1899 bp_get_dsize(spa_t
*spa
, const blkptr_t
*bp
)
1903 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
1905 for (int d
= 0; d
< BP_GET_NDVAS(bp
); d
++)
1906 dsize
+= dva_get_dsize_sync(spa
, &bp
->blk_dva
[d
]);
1908 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
1914 spa_dirty_data(spa_t
*spa
)
1916 return (spa
->spa_dsl_pool
->dp_dirty_total
);
1920 * ==========================================================================
1921 * Initialization and Termination
1922 * ==========================================================================
1926 spa_name_compare(const void *a1
, const void *a2
)
1928 const spa_t
*s1
= a1
;
1929 const spa_t
*s2
= a2
;
1932 s
= strcmp(s1
->spa_name
, s2
->spa_name
);
1943 return (spa_active_count
);
1955 mutex_init(&spa_namespace_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1956 mutex_init(&spa_spare_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1957 mutex_init(&spa_l2cache_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1958 cv_init(&spa_namespace_cv
, NULL
, CV_DEFAULT
, NULL
);
1960 avl_create(&spa_namespace_avl
, spa_name_compare
, sizeof (spa_t
),
1961 offsetof(spa_t
, spa_avl
));
1963 avl_create(&spa_spare_avl
, spa_spare_compare
, sizeof (spa_aux_t
),
1964 offsetof(spa_aux_t
, aux_avl
));
1966 avl_create(&spa_l2cache_avl
, spa_l2cache_compare
, sizeof (spa_aux_t
),
1967 offsetof(spa_aux_t
, aux_avl
));
1969 spa_mode_global
= mode
;
1974 if (spa_mode_global
!= FREAD
&& dprintf_find_string("watch")) {
1975 arc_procfd
= open("/proc/self/ctl", O_WRONLY
);
1976 if (arc_procfd
== -1) {
1977 perror("could not enable watchpoints: "
1978 "opening /proc/self/ctl failed: ");
1988 metaslab_alloc_trace_init();
1992 vdev_cache_stat_init();
1995 zpool_feature_init();
2007 vdev_cache_stat_fini();
2011 metaslab_alloc_trace_fini();
2016 avl_destroy(&spa_namespace_avl
);
2017 avl_destroy(&spa_spare_avl
);
2018 avl_destroy(&spa_l2cache_avl
);
2020 cv_destroy(&spa_namespace_cv
);
2021 mutex_destroy(&spa_namespace_lock
);
2022 mutex_destroy(&spa_spare_lock
);
2023 mutex_destroy(&spa_l2cache_lock
);
2027 * Return whether this pool has slogs. No locking needed.
2028 * It's not a problem if the wrong answer is returned as it's only for
2029 * performance and not correctness
2032 spa_has_slogs(spa_t
*spa
)
2034 return (spa
->spa_log_class
->mc_rotor
!= NULL
);
2038 spa_get_log_state(spa_t
*spa
)
2040 return (spa
->spa_log_state
);
2044 spa_set_log_state(spa_t
*spa
, spa_log_state_t state
)
2046 spa
->spa_log_state
= state
;
2050 spa_is_root(spa_t
*spa
)
2052 return (spa
->spa_is_root
);
2056 spa_writeable(spa_t
*spa
)
2058 return (!!(spa
->spa_mode
& FWRITE
) && spa
->spa_trust_config
);
2062 * Returns true if there is a pending sync task in any of the current
2063 * syncing txg, the current quiescing txg, or the current open txg.
2066 spa_has_pending_synctask(spa_t
*spa
)
2068 return (!txg_all_lists_empty(&spa
->spa_dsl_pool
->dp_sync_tasks
) ||
2069 !txg_all_lists_empty(&spa
->spa_dsl_pool
->dp_early_sync_tasks
));
2073 spa_mode(spa_t
*spa
)
2075 return (spa
->spa_mode
);
2079 spa_bootfs(spa_t
*spa
)
2081 return (spa
->spa_bootfs
);
2085 spa_delegation(spa_t
*spa
)
2087 return (spa
->spa_delegation
);
2091 spa_meta_objset(spa_t
*spa
)
2093 return (spa
->spa_meta_objset
);
2097 spa_dedup_checksum(spa_t
*spa
)
2099 return (spa
->spa_dedup_checksum
);
2103 * Reset pool scan stat per scan pass (or reboot).
2106 spa_scan_stat_init(spa_t
*spa
)
2108 /* data not stored on disk */
2109 spa
->spa_scan_pass_start
= gethrestime_sec();
2110 if (dsl_scan_is_paused_scrub(spa
->spa_dsl_pool
->dp_scan
))
2111 spa
->spa_scan_pass_scrub_pause
= spa
->spa_scan_pass_start
;
2113 spa
->spa_scan_pass_scrub_pause
= 0;
2114 spa
->spa_scan_pass_scrub_spent_paused
= 0;
2115 spa
->spa_scan_pass_exam
= 0;
2116 vdev_scan_stat_init(spa
->spa_root_vdev
);
2120 * Get scan stats for zpool status reports
2123 spa_scan_get_stats(spa_t
*spa
, pool_scan_stat_t
*ps
)
2125 dsl_scan_t
*scn
= spa
->spa_dsl_pool
? spa
->spa_dsl_pool
->dp_scan
: NULL
;
2127 if (scn
== NULL
|| scn
->scn_phys
.scn_func
== POOL_SCAN_NONE
)
2128 return (SET_ERROR(ENOENT
));
2129 bzero(ps
, sizeof (pool_scan_stat_t
));
2131 /* data stored on disk */
2132 ps
->pss_func
= scn
->scn_phys
.scn_func
;
2133 ps
->pss_start_time
= scn
->scn_phys
.scn_start_time
;
2134 ps
->pss_end_time
= scn
->scn_phys
.scn_end_time
;
2135 ps
->pss_to_examine
= scn
->scn_phys
.scn_to_examine
;
2136 ps
->pss_examined
= scn
->scn_phys
.scn_examined
;
2137 ps
->pss_to_process
= scn
->scn_phys
.scn_to_process
;
2138 ps
->pss_processed
= scn
->scn_phys
.scn_processed
;
2139 ps
->pss_errors
= scn
->scn_phys
.scn_errors
;
2140 ps
->pss_state
= scn
->scn_phys
.scn_state
;
2142 /* data not stored on disk */
2143 ps
->pss_pass_start
= spa
->spa_scan_pass_start
;
2144 ps
->pss_pass_exam
= spa
->spa_scan_pass_exam
;
2145 ps
->pss_pass_scrub_pause
= spa
->spa_scan_pass_scrub_pause
;
2146 ps
->pss_pass_scrub_spent_paused
= spa
->spa_scan_pass_scrub_spent_paused
;
2152 spa_maxblocksize(spa_t
*spa
)
2154 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_BLOCKS
))
2155 return (SPA_MAXBLOCKSIZE
);
2157 return (SPA_OLD_MAXBLOCKSIZE
);
2161 * Returns the txg that the last device removal completed. No indirect mappings
2162 * have been added since this txg.
2165 spa_get_last_removal_txg(spa_t
*spa
)
2168 uint64_t ret
= -1ULL;
2170 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2172 * sr_prev_indirect_vdev is only modified while holding all the
2173 * config locks, so it is sufficient to hold SCL_VDEV as reader when
2176 vdevid
= spa
->spa_removing_phys
.sr_prev_indirect_vdev
;
2178 while (vdevid
!= -1ULL) {
2179 vdev_t
*vd
= vdev_lookup_top(spa
, vdevid
);
2180 vdev_indirect_births_t
*vib
= vd
->vdev_indirect_births
;
2182 ASSERT3P(vd
->vdev_ops
, ==, &vdev_indirect_ops
);
2185 * If the removal did not remap any data, we don't care.
2187 if (vdev_indirect_births_count(vib
) != 0) {
2188 ret
= vdev_indirect_births_last_entry_txg(vib
);
2192 vdevid
= vd
->vdev_indirect_config
.vic_prev_indirect_vdev
;
2194 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2197 spa_feature_is_active(spa
, SPA_FEATURE_DEVICE_REMOVAL
));
2203 spa_trust_config(spa_t
*spa
)
2205 return (spa
->spa_trust_config
);
2209 spa_missing_tvds_allowed(spa_t
*spa
)
2211 return (spa
->spa_missing_tvds_allowed
);
2215 spa_set_missing_tvds(spa_t
*spa
, uint64_t missing
)
2217 spa
->spa_missing_tvds
= missing
;
2221 spa_top_vdevs_spacemap_addressable(spa_t
*spa
)
2223 vdev_t
*rvd
= spa
->spa_root_vdev
;
2224 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
2225 if (!vdev_is_spacemap_addressable(rvd
->vdev_child
[c
]))
2232 spa_has_checkpoint(spa_t
*spa
)
2234 return (spa
->spa_checkpoint_txg
!= 0);
2238 spa_importing_readonly_checkpoint(spa_t
*spa
)
2240 return ((spa
->spa_import_flags
& ZFS_IMPORT_CHECKPOINT
) &&
2241 spa
->spa_mode
== FREAD
);
2245 spa_min_claim_txg(spa_t
*spa
)
2247 uint64_t checkpoint_txg
= spa
->spa_uberblock
.ub_checkpoint_txg
;
2249 if (checkpoint_txg
!= 0)
2250 return (checkpoint_txg
+ 1);
2252 return (spa
->spa_first_txg
);
2256 * If there is a checkpoint, async destroys may consume more space from
2257 * the pool instead of freeing it. In an attempt to save the pool from
2258 * getting suspended when it is about to run out of space, we stop
2259 * processing async destroys.
2262 spa_suspend_async_destroy(spa_t
*spa
)
2264 dsl_pool_t
*dp
= spa_get_dsl(spa
);
2266 uint64_t unreserved
= dsl_pool_unreserved_space(dp
,
2267 ZFS_SPACE_CHECK_EXTRA_RESERVED
);
2268 uint64_t used
= dsl_dir_phys(dp
->dp_root_dir
)->dd_used_bytes
;
2269 uint64_t avail
= (unreserved
> used
) ? (unreserved
- used
) : 0;
2271 if (spa_has_checkpoint(spa
) && avail
== 0)