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, 2017 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.
30 #include <sys/zfs_context.h>
31 #include <sys/spa_impl.h>
33 #include <sys/zio_checksum.h>
34 #include <sys/zio_compress.h>
36 #include <sys/dmu_tx.h>
39 #include <sys/vdev_impl.h>
40 #include <sys/vdev_file.h>
41 #include <sys/vdev_raidz.h>
42 #include <sys/metaslab.h>
43 #include <sys/uberblock_impl.h>
46 #include <sys/unique.h>
47 #include <sys/dsl_pool.h>
48 #include <sys/dsl_dir.h>
49 #include <sys/dsl_prop.h>
50 #include <sys/fm/util.h>
51 #include <sys/dsl_scan.h>
52 #include <sys/fs/zfs.h>
53 #include <sys/metaslab_impl.h>
56 #include <sys/kstat.h>
58 #include <sys/zfeature.h>
64 * There are four basic locks for managing spa_t structures:
66 * spa_namespace_lock (global mutex)
68 * This lock must be acquired to do any of the following:
70 * - Lookup a spa_t by name
71 * - Add or remove a spa_t from the namespace
72 * - Increase spa_refcount from non-zero
73 * - Check if spa_refcount is zero
75 * - add/remove/attach/detach devices
76 * - Held for the duration of create/destroy/import/export
78 * It does not need to handle recursion. A create or destroy may
79 * reference objects (files or zvols) in other pools, but by
80 * definition they must have an existing reference, and will never need
81 * to lookup a spa_t by name.
83 * spa_refcount (per-spa refcount_t protected by mutex)
85 * This reference count keep track of any active users of the spa_t. The
86 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
87 * the refcount is never really 'zero' - opening a pool implicitly keeps
88 * some references in the DMU. Internally we check against spa_minref, but
89 * present the image of a zero/non-zero value to consumers.
91 * spa_config_lock[] (per-spa array of rwlocks)
93 * This protects the spa_t from config changes, and must be held in
94 * the following circumstances:
96 * - RW_READER to perform I/O to the spa
97 * - RW_WRITER to change the vdev config
99 * The locking order is fairly straightforward:
101 * spa_namespace_lock -> spa_refcount
103 * The namespace lock must be acquired to increase the refcount from 0
104 * or to check if it is zero.
106 * spa_refcount -> spa_config_lock[]
108 * There must be at least one valid reference on the spa_t to acquire
111 * spa_namespace_lock -> spa_config_lock[]
113 * The namespace lock must always be taken before the config lock.
116 * The spa_namespace_lock can be acquired directly and is globally visible.
118 * The namespace is manipulated using the following functions, all of which
119 * require the spa_namespace_lock to be held.
121 * spa_lookup() Lookup a spa_t by name.
123 * spa_add() Create a new spa_t in the namespace.
125 * spa_remove() Remove a spa_t from the namespace. This also
126 * frees up any memory associated with the spa_t.
128 * spa_next() Returns the next spa_t in the system, or the
129 * first if NULL is passed.
131 * spa_evict_all() Shutdown and remove all spa_t structures in
134 * spa_guid_exists() Determine whether a pool/device guid exists.
136 * The spa_refcount is manipulated using the following functions:
138 * spa_open_ref() Adds a reference to the given spa_t. Must be
139 * called with spa_namespace_lock held if the
140 * refcount is currently zero.
142 * spa_close() Remove a reference from the spa_t. This will
143 * not free the spa_t or remove it from the
144 * namespace. No locking is required.
146 * spa_refcount_zero() Returns true if the refcount is currently
147 * zero. Must be called with spa_namespace_lock
150 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
151 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
152 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
154 * To read the configuration, it suffices to hold one of these locks as reader.
155 * To modify the configuration, you must hold all locks as writer. To modify
156 * vdev state without altering the vdev tree's topology (e.g. online/offline),
157 * you must hold SCL_STATE and SCL_ZIO as writer.
159 * We use these distinct config locks to avoid recursive lock entry.
160 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
161 * block allocations (SCL_ALLOC), which may require reading space maps
162 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
164 * The spa config locks cannot be normal rwlocks because we need the
165 * ability to hand off ownership. For example, SCL_ZIO is acquired
166 * by the issuing thread and later released by an interrupt thread.
167 * They do, however, obey the usual write-wanted semantics to prevent
168 * writer (i.e. system administrator) starvation.
170 * The lock acquisition rules are as follows:
173 * Protects changes to the vdev tree topology, such as vdev
174 * add/remove/attach/detach. Protects the dirty config list
175 * (spa_config_dirty_list) and the set of spares and l2arc devices.
178 * Protects changes to pool state and vdev state, such as vdev
179 * online/offline/fault/degrade/clear. Protects the dirty state list
180 * (spa_state_dirty_list) and global pool state (spa_state).
183 * Protects changes to metaslab groups and classes.
184 * Held as reader by metaslab_alloc() and metaslab_claim().
187 * Held by bp-level zios (those which have no io_vd upon entry)
188 * to prevent changes to the vdev tree. The bp-level zio implicitly
189 * protects all of its vdev child zios, which do not hold SCL_ZIO.
192 * Protects changes to metaslab groups and classes.
193 * Held as reader by metaslab_free(). SCL_FREE is distinct from
194 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
195 * blocks in zio_done() while another i/o that holds either
196 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
199 * Held as reader to prevent changes to the vdev tree during trivial
200 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
201 * other locks, and lower than all of them, to ensure that it's safe
202 * to acquire regardless of caller context.
204 * In addition, the following rules apply:
206 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
207 * The lock ordering is SCL_CONFIG > spa_props_lock.
209 * (b) I/O operations on leaf vdevs. For any zio operation that takes
210 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
211 * or zio_write_phys() -- the caller must ensure that the config cannot
212 * cannot change in the interim, and that the vdev cannot be reopened.
213 * SCL_STATE as reader suffices for both.
215 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
217 * spa_vdev_enter() Acquire the namespace lock and the config lock
220 * spa_vdev_exit() Release the config lock, wait for all I/O
221 * to complete, sync the updated configs to the
222 * cache, and release the namespace lock.
224 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
225 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
226 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
228 * spa_rename() is also implemented within this file since it requires
229 * manipulation of the namespace.
232 static avl_tree_t spa_namespace_avl
;
233 kmutex_t spa_namespace_lock
;
234 static kcondvar_t spa_namespace_cv
;
235 int spa_max_replication_override
= SPA_DVAS_PER_BP
;
237 static kmutex_t spa_spare_lock
;
238 static avl_tree_t spa_spare_avl
;
239 static kmutex_t spa_l2cache_lock
;
240 static avl_tree_t spa_l2cache_avl
;
242 kmem_cache_t
*spa_buffer_pool
;
247 * Everything except dprintf, set_error, spa, and indirect_remap is on
248 * by default in debug builds.
250 int zfs_flags
= ~(ZFS_DEBUG_DPRINTF
| ZFS_DEBUG_SET_ERROR
|
251 ZFS_DEBUG_INDIRECT_REMAP
);
257 * zfs_recover can be set to nonzero to attempt to recover from
258 * otherwise-fatal errors, typically caused by on-disk corruption. When
259 * set, calls to zfs_panic_recover() will turn into warning messages.
260 * This should only be used as a last resort, as it typically results
261 * in leaked space, or worse.
263 int zfs_recover
= B_FALSE
;
266 * If destroy encounters an EIO while reading metadata (e.g. indirect
267 * blocks), space referenced by the missing metadata can not be freed.
268 * Normally this causes the background destroy to become "stalled", as
269 * it is unable to make forward progress. While in this stalled state,
270 * all remaining space to free from the error-encountering filesystem is
271 * "temporarily leaked". Set this flag to cause it to ignore the EIO,
272 * permanently leak the space from indirect blocks that can not be read,
273 * and continue to free everything else that it can.
275 * The default, "stalling" behavior is useful if the storage partially
276 * fails (i.e. some but not all i/os fail), and then later recovers. In
277 * this case, we will be able to continue pool operations while it is
278 * partially failed, and when it recovers, we can continue to free the
279 * space, with no leaks. However, note that this case is actually
282 * Typically pools either (a) fail completely (but perhaps temporarily,
283 * e.g. a top-level vdev going offline), or (b) have localized,
284 * permanent errors (e.g. disk returns the wrong data due to bit flip or
285 * firmware bug). In case (a), this setting does not matter because the
286 * pool will be suspended and the sync thread will not be able to make
287 * forward progress regardless. In case (b), because the error is
288 * permanent, the best we can do is leak the minimum amount of space,
289 * which is what setting this flag will do. Therefore, it is reasonable
290 * for this flag to normally be set, but we chose the more conservative
291 * approach of not setting it, so that there is no possibility of
292 * leaking space in the "partial temporary" failure case.
294 int zfs_free_leak_on_eio
= B_FALSE
;
297 * Expiration time in milliseconds. This value has two meanings. First it is
298 * used to determine when the spa_deadman() logic should fire. By default the
299 * spa_deadman() will fire if spa_sync() has not completed in 600 seconds.
300 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
301 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
302 * in one of three behaviors controlled by zfs_deadman_failmode.
304 unsigned long zfs_deadman_synctime_ms
= 600000ULL;
307 * This value controls the maximum amount of time zio_wait() will block for an
308 * outstanding IO. By default this is 300 seconds at which point the "hung"
309 * behavior will be applied as described for zfs_deadman_synctime_ms.
311 unsigned long zfs_deadman_ziotime_ms
= 300000ULL;
314 * Check time in milliseconds. This defines the frequency at which we check
317 unsigned long zfs_deadman_checktime_ms
= 60000ULL;
320 * By default the deadman is enabled.
322 int zfs_deadman_enabled
= 1;
325 * Controls the behavior of the deadman when it detects a "hung" I/O.
326 * Valid values are zfs_deadman_failmode=<wait|continue|panic>.
328 * wait - Wait for the "hung" I/O (default)
329 * continue - Attempt to recover from a "hung" I/O
330 * panic - Panic the system
332 char *zfs_deadman_failmode
= "wait";
335 * The worst case is single-sector max-parity RAID-Z blocks, in which
336 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
337 * times the size; so just assume that. Add to this the fact that
338 * we can have up to 3 DVAs per bp, and one more factor of 2 because
339 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together,
341 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
343 int spa_asize_inflation
= 24;
346 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
347 * the pool to be consumed. This ensures that we don't run the pool
348 * completely out of space, due to unaccounted changes (e.g. to the MOS).
349 * It also limits the worst-case time to allocate space. If we have
350 * less than this amount of free space, most ZPL operations (e.g. write,
351 * create) will return ENOSPC.
353 * Certain operations (e.g. file removal, most administrative actions) can
354 * use half the slop space. They will only return ENOSPC if less than half
355 * the slop space is free. Typically, once the pool has less than the slop
356 * space free, the user will use these operations to free up space in the pool.
357 * These are the operations that call dsl_pool_adjustedsize() with the netfree
358 * argument set to TRUE.
360 * Operations that are almost guaranteed to free up space in the absence of
361 * a pool checkpoint can use up to three quarters of the slop space
364 * A very restricted set of operations are always permitted, regardless of
365 * the amount of free space. These are the operations that call
366 * dsl_sync_task(ZFS_SPACE_CHECK_NONE). If these operations result in a net
367 * increase in the amount of space used, it is possible to run the pool
368 * completely out of space, causing it to be permanently read-only.
370 * Note that on very small pools, the slop space will be larger than
371 * 3.2%, in an effort to have it be at least spa_min_slop (128MB),
372 * but we never allow it to be more than half the pool size.
374 * See also the comments in zfs_space_check_t.
376 int spa_slop_shift
= 5;
377 uint64_t spa_min_slop
= 128 * 1024 * 1024;
381 spa_load_failed(spa_t
*spa
, const char *fmt
, ...)
387 (void) vsnprintf(buf
, sizeof (buf
), fmt
, adx
);
390 zfs_dbgmsg("spa_load(%s, config %s): FAILED: %s", spa
->spa_name
,
391 spa
->spa_trust_config
? "trusted" : "untrusted", buf
);
396 spa_load_note(spa_t
*spa
, const char *fmt
, ...)
402 (void) vsnprintf(buf
, sizeof (buf
), fmt
, adx
);
405 zfs_dbgmsg("spa_load(%s, config %s): %s", spa
->spa_name
,
406 spa
->spa_trust_config
? "trusted" : "untrusted", buf
);
410 * ==========================================================================
412 * ==========================================================================
415 spa_config_lock_init(spa_t
*spa
)
417 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
418 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
419 mutex_init(&scl
->scl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
420 cv_init(&scl
->scl_cv
, NULL
, CV_DEFAULT
, NULL
);
421 refcount_create_untracked(&scl
->scl_count
);
422 scl
->scl_writer
= NULL
;
423 scl
->scl_write_wanted
= 0;
428 spa_config_lock_destroy(spa_t
*spa
)
430 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
431 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
432 mutex_destroy(&scl
->scl_lock
);
433 cv_destroy(&scl
->scl_cv
);
434 refcount_destroy(&scl
->scl_count
);
435 ASSERT(scl
->scl_writer
== NULL
);
436 ASSERT(scl
->scl_write_wanted
== 0);
441 spa_config_tryenter(spa_t
*spa
, int locks
, void *tag
, krw_t rw
)
443 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
444 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
445 if (!(locks
& (1 << i
)))
447 mutex_enter(&scl
->scl_lock
);
448 if (rw
== RW_READER
) {
449 if (scl
->scl_writer
|| scl
->scl_write_wanted
) {
450 mutex_exit(&scl
->scl_lock
);
451 spa_config_exit(spa
, locks
& ((1 << i
) - 1),
456 ASSERT(scl
->scl_writer
!= curthread
);
457 if (!refcount_is_zero(&scl
->scl_count
)) {
458 mutex_exit(&scl
->scl_lock
);
459 spa_config_exit(spa
, locks
& ((1 << i
) - 1),
463 scl
->scl_writer
= curthread
;
465 (void) refcount_add(&scl
->scl_count
, tag
);
466 mutex_exit(&scl
->scl_lock
);
472 spa_config_enter(spa_t
*spa
, int locks
, void *tag
, krw_t rw
)
476 ASSERT3U(SCL_LOCKS
, <, sizeof (wlocks_held
) * NBBY
);
478 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
479 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
480 if (scl
->scl_writer
== curthread
)
481 wlocks_held
|= (1 << i
);
482 if (!(locks
& (1 << i
)))
484 mutex_enter(&scl
->scl_lock
);
485 if (rw
== RW_READER
) {
486 while (scl
->scl_writer
|| scl
->scl_write_wanted
) {
487 cv_wait(&scl
->scl_cv
, &scl
->scl_lock
);
490 ASSERT(scl
->scl_writer
!= curthread
);
491 while (!refcount_is_zero(&scl
->scl_count
)) {
492 scl
->scl_write_wanted
++;
493 cv_wait(&scl
->scl_cv
, &scl
->scl_lock
);
494 scl
->scl_write_wanted
--;
496 scl
->scl_writer
= curthread
;
498 (void) refcount_add(&scl
->scl_count
, tag
);
499 mutex_exit(&scl
->scl_lock
);
501 ASSERT3U(wlocks_held
, <=, locks
);
505 spa_config_exit(spa_t
*spa
, int locks
, void *tag
)
507 for (int i
= SCL_LOCKS
- 1; i
>= 0; i
--) {
508 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
509 if (!(locks
& (1 << i
)))
511 mutex_enter(&scl
->scl_lock
);
512 ASSERT(!refcount_is_zero(&scl
->scl_count
));
513 if (refcount_remove(&scl
->scl_count
, tag
) == 0) {
514 ASSERT(scl
->scl_writer
== NULL
||
515 scl
->scl_writer
== curthread
);
516 scl
->scl_writer
= NULL
; /* OK in either case */
517 cv_broadcast(&scl
->scl_cv
);
519 mutex_exit(&scl
->scl_lock
);
524 spa_config_held(spa_t
*spa
, int locks
, krw_t rw
)
528 for (int i
= 0; i
< SCL_LOCKS
; i
++) {
529 spa_config_lock_t
*scl
= &spa
->spa_config_lock
[i
];
530 if (!(locks
& (1 << i
)))
532 if ((rw
== RW_READER
&& !refcount_is_zero(&scl
->scl_count
)) ||
533 (rw
== RW_WRITER
&& scl
->scl_writer
== curthread
))
534 locks_held
|= 1 << i
;
541 * ==========================================================================
542 * SPA namespace functions
543 * ==========================================================================
547 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
548 * Returns NULL if no matching spa_t is found.
551 spa_lookup(const char *name
)
553 static spa_t search
; /* spa_t is large; don't allocate on stack */
558 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
560 (void) strlcpy(search
.spa_name
, name
, sizeof (search
.spa_name
));
563 * If it's a full dataset name, figure out the pool name and
566 cp
= strpbrk(search
.spa_name
, "/@#");
570 spa
= avl_find(&spa_namespace_avl
, &search
, &where
);
576 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
577 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
578 * looking for potentially hung I/Os.
581 spa_deadman(void *arg
)
585 /* Disable the deadman if the pool is suspended. */
586 if (spa_suspended(spa
))
589 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
590 (gethrtime() - spa
->spa_sync_starttime
) / NANOSEC
,
591 ++spa
->spa_deadman_calls
);
592 if (zfs_deadman_enabled
)
593 vdev_deadman(spa
->spa_root_vdev
, FTAG
);
595 spa
->spa_deadman_tqid
= taskq_dispatch_delay(system_delay_taskq
,
596 spa_deadman
, spa
, TQ_SLEEP
, ddi_get_lbolt() +
597 MSEC_TO_TICK(zfs_deadman_checktime_ms
));
601 * Create an uninitialized spa_t with the given name. Requires
602 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
603 * exist by calling spa_lookup() first.
606 spa_add(const char *name
, nvlist_t
*config
, const char *altroot
)
609 spa_config_dirent_t
*dp
;
611 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
613 spa
= kmem_zalloc(sizeof (spa_t
), KM_SLEEP
);
615 mutex_init(&spa
->spa_async_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
616 mutex_init(&spa
->spa_errlist_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
617 mutex_init(&spa
->spa_errlog_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
618 mutex_init(&spa
->spa_evicting_os_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
619 mutex_init(&spa
->spa_history_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
620 mutex_init(&spa
->spa_proc_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
621 mutex_init(&spa
->spa_props_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
622 mutex_init(&spa
->spa_cksum_tmpls_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
623 mutex_init(&spa
->spa_scrub_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
624 mutex_init(&spa
->spa_suspend_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
625 mutex_init(&spa
->spa_vdev_top_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
626 mutex_init(&spa
->spa_feat_stats_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
627 mutex_init(&spa
->spa_alloc_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
629 cv_init(&spa
->spa_async_cv
, NULL
, CV_DEFAULT
, NULL
);
630 cv_init(&spa
->spa_evicting_os_cv
, NULL
, CV_DEFAULT
, NULL
);
631 cv_init(&spa
->spa_proc_cv
, NULL
, CV_DEFAULT
, NULL
);
632 cv_init(&spa
->spa_scrub_io_cv
, NULL
, CV_DEFAULT
, NULL
);
633 cv_init(&spa
->spa_suspend_cv
, NULL
, CV_DEFAULT
, NULL
);
635 for (int t
= 0; t
< TXG_SIZE
; t
++)
636 bplist_create(&spa
->spa_free_bplist
[t
]);
638 (void) strlcpy(spa
->spa_name
, name
, sizeof (spa
->spa_name
));
639 spa
->spa_state
= POOL_STATE_UNINITIALIZED
;
640 spa
->spa_freeze_txg
= UINT64_MAX
;
641 spa
->spa_final_txg
= UINT64_MAX
;
642 spa
->spa_load_max_txg
= UINT64_MAX
;
644 spa
->spa_proc_state
= SPA_PROC_NONE
;
645 spa
->spa_trust_config
= B_TRUE
;
647 spa
->spa_deadman_synctime
= MSEC2NSEC(zfs_deadman_synctime_ms
);
648 spa
->spa_deadman_ziotime
= MSEC2NSEC(zfs_deadman_ziotime_ms
);
649 spa_set_deadman_failmode(spa
, zfs_deadman_failmode
);
651 refcount_create(&spa
->spa_refcount
);
652 spa_config_lock_init(spa
);
655 avl_add(&spa_namespace_avl
, spa
);
658 * Set the alternate root, if there is one.
661 spa
->spa_root
= spa_strdup(altroot
);
663 avl_create(&spa
->spa_alloc_tree
, zio_bookmark_compare
,
664 sizeof (zio_t
), offsetof(zio_t
, io_alloc_node
));
667 * Every pool starts with the default cachefile
669 list_create(&spa
->spa_config_list
, sizeof (spa_config_dirent_t
),
670 offsetof(spa_config_dirent_t
, scd_link
));
672 dp
= kmem_zalloc(sizeof (spa_config_dirent_t
), KM_SLEEP
);
673 dp
->scd_path
= altroot
? NULL
: spa_strdup(spa_config_path
);
674 list_insert_head(&spa
->spa_config_list
, dp
);
676 VERIFY(nvlist_alloc(&spa
->spa_load_info
, NV_UNIQUE_NAME
,
679 if (config
!= NULL
) {
682 if (nvlist_lookup_nvlist(config
, ZPOOL_CONFIG_FEATURES_FOR_READ
,
684 VERIFY(nvlist_dup(features
, &spa
->spa_label_features
,
688 VERIFY(nvlist_dup(config
, &spa
->spa_config
, 0) == 0);
691 if (spa
->spa_label_features
== NULL
) {
692 VERIFY(nvlist_alloc(&spa
->spa_label_features
, NV_UNIQUE_NAME
,
696 spa
->spa_min_ashift
= INT_MAX
;
697 spa
->spa_max_ashift
= 0;
699 /* Reset cached value */
700 spa
->spa_dedup_dspace
= ~0ULL;
703 * As a pool is being created, treat all features as disabled by
704 * setting SPA_FEATURE_DISABLED for all entries in the feature
707 for (int i
= 0; i
< SPA_FEATURES
; i
++) {
708 spa
->spa_feat_refcount_cache
[i
] = SPA_FEATURE_DISABLED
;
715 * Removes a spa_t from the namespace, freeing up any memory used. Requires
716 * spa_namespace_lock. This is called only after the spa_t has been closed and
720 spa_remove(spa_t
*spa
)
722 spa_config_dirent_t
*dp
;
724 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
725 ASSERT(spa
->spa_state
== POOL_STATE_UNINITIALIZED
);
726 ASSERT3U(refcount_count(&spa
->spa_refcount
), ==, 0);
728 nvlist_free(spa
->spa_config_splitting
);
730 avl_remove(&spa_namespace_avl
, spa
);
731 cv_broadcast(&spa_namespace_cv
);
734 spa_strfree(spa
->spa_root
);
736 while ((dp
= list_head(&spa
->spa_config_list
)) != NULL
) {
737 list_remove(&spa
->spa_config_list
, dp
);
738 if (dp
->scd_path
!= NULL
)
739 spa_strfree(dp
->scd_path
);
740 kmem_free(dp
, sizeof (spa_config_dirent_t
));
743 avl_destroy(&spa
->spa_alloc_tree
);
744 list_destroy(&spa
->spa_config_list
);
746 nvlist_free(spa
->spa_label_features
);
747 nvlist_free(spa
->spa_load_info
);
748 nvlist_free(spa
->spa_feat_stats
);
749 spa_config_set(spa
, NULL
);
751 refcount_destroy(&spa
->spa_refcount
);
753 spa_stats_destroy(spa
);
754 spa_config_lock_destroy(spa
);
756 for (int t
= 0; t
< TXG_SIZE
; t
++)
757 bplist_destroy(&spa
->spa_free_bplist
[t
]);
759 zio_checksum_templates_free(spa
);
761 cv_destroy(&spa
->spa_async_cv
);
762 cv_destroy(&spa
->spa_evicting_os_cv
);
763 cv_destroy(&spa
->spa_proc_cv
);
764 cv_destroy(&spa
->spa_scrub_io_cv
);
765 cv_destroy(&spa
->spa_suspend_cv
);
767 mutex_destroy(&spa
->spa_alloc_lock
);
768 mutex_destroy(&spa
->spa_async_lock
);
769 mutex_destroy(&spa
->spa_errlist_lock
);
770 mutex_destroy(&spa
->spa_errlog_lock
);
771 mutex_destroy(&spa
->spa_evicting_os_lock
);
772 mutex_destroy(&spa
->spa_history_lock
);
773 mutex_destroy(&spa
->spa_proc_lock
);
774 mutex_destroy(&spa
->spa_props_lock
);
775 mutex_destroy(&spa
->spa_cksum_tmpls_lock
);
776 mutex_destroy(&spa
->spa_scrub_lock
);
777 mutex_destroy(&spa
->spa_suspend_lock
);
778 mutex_destroy(&spa
->spa_vdev_top_lock
);
779 mutex_destroy(&spa
->spa_feat_stats_lock
);
781 kmem_free(spa
, sizeof (spa_t
));
785 * Given a pool, return the next pool in the namespace, or NULL if there is
786 * none. If 'prev' is NULL, return the first pool.
789 spa_next(spa_t
*prev
)
791 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
794 return (AVL_NEXT(&spa_namespace_avl
, prev
));
796 return (avl_first(&spa_namespace_avl
));
800 * ==========================================================================
801 * SPA refcount functions
802 * ==========================================================================
806 * Add a reference to the given spa_t. Must have at least one reference, or
807 * have the namespace lock held.
810 spa_open_ref(spa_t
*spa
, void *tag
)
812 ASSERT(refcount_count(&spa
->spa_refcount
) >= spa
->spa_minref
||
813 MUTEX_HELD(&spa_namespace_lock
));
814 (void) refcount_add(&spa
->spa_refcount
, tag
);
818 * Remove a reference to the given spa_t. Must have at least one reference, or
819 * have the namespace lock held.
822 spa_close(spa_t
*spa
, void *tag
)
824 ASSERT(refcount_count(&spa
->spa_refcount
) > spa
->spa_minref
||
825 MUTEX_HELD(&spa_namespace_lock
));
826 (void) refcount_remove(&spa
->spa_refcount
, tag
);
830 * Remove a reference to the given spa_t held by a dsl dir that is
831 * being asynchronously released. Async releases occur from a taskq
832 * performing eviction of dsl datasets and dirs. The namespace lock
833 * isn't held and the hold by the object being evicted may contribute to
834 * spa_minref (e.g. dataset or directory released during pool export),
835 * so the asserts in spa_close() do not apply.
838 spa_async_close(spa_t
*spa
, void *tag
)
840 (void) refcount_remove(&spa
->spa_refcount
, tag
);
844 * Check to see if the spa refcount is zero. Must be called with
845 * spa_namespace_lock held. We really compare against spa_minref, which is the
846 * number of references acquired when opening a pool
849 spa_refcount_zero(spa_t
*spa
)
851 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
853 return (refcount_count(&spa
->spa_refcount
) == spa
->spa_minref
);
857 * ==========================================================================
858 * SPA spare and l2cache tracking
859 * ==========================================================================
863 * Hot spares and cache devices are tracked using the same code below,
864 * for 'auxiliary' devices.
867 typedef struct spa_aux
{
875 spa_aux_compare(const void *a
, const void *b
)
877 const spa_aux_t
*sa
= (const spa_aux_t
*)a
;
878 const spa_aux_t
*sb
= (const spa_aux_t
*)b
;
880 return (AVL_CMP(sa
->aux_guid
, sb
->aux_guid
));
884 spa_aux_add(vdev_t
*vd
, avl_tree_t
*avl
)
890 search
.aux_guid
= vd
->vdev_guid
;
891 if ((aux
= avl_find(avl
, &search
, &where
)) != NULL
) {
894 aux
= kmem_zalloc(sizeof (spa_aux_t
), KM_SLEEP
);
895 aux
->aux_guid
= vd
->vdev_guid
;
897 avl_insert(avl
, aux
, where
);
902 spa_aux_remove(vdev_t
*vd
, avl_tree_t
*avl
)
908 search
.aux_guid
= vd
->vdev_guid
;
909 aux
= avl_find(avl
, &search
, &where
);
913 if (--aux
->aux_count
== 0) {
914 avl_remove(avl
, aux
);
915 kmem_free(aux
, sizeof (spa_aux_t
));
916 } else if (aux
->aux_pool
== spa_guid(vd
->vdev_spa
)) {
917 aux
->aux_pool
= 0ULL;
922 spa_aux_exists(uint64_t guid
, uint64_t *pool
, int *refcnt
, avl_tree_t
*avl
)
924 spa_aux_t search
, *found
;
926 search
.aux_guid
= guid
;
927 found
= avl_find(avl
, &search
, NULL
);
931 *pool
= found
->aux_pool
;
938 *refcnt
= found
->aux_count
;
943 return (found
!= NULL
);
947 spa_aux_activate(vdev_t
*vd
, avl_tree_t
*avl
)
949 spa_aux_t search
, *found
;
952 search
.aux_guid
= vd
->vdev_guid
;
953 found
= avl_find(avl
, &search
, &where
);
954 ASSERT(found
!= NULL
);
955 ASSERT(found
->aux_pool
== 0ULL);
957 found
->aux_pool
= spa_guid(vd
->vdev_spa
);
961 * Spares are tracked globally due to the following constraints:
963 * - A spare may be part of multiple pools.
964 * - A spare may be added to a pool even if it's actively in use within
966 * - A spare in use in any pool can only be the source of a replacement if
967 * the target is a spare in the same pool.
969 * We keep track of all spares on the system through the use of a reference
970 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
971 * spare, then we bump the reference count in the AVL tree. In addition, we set
972 * the 'vdev_isspare' member to indicate that the device is a spare (active or
973 * inactive). When a spare is made active (used to replace a device in the
974 * pool), we also keep track of which pool its been made a part of.
976 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
977 * called under the spa_namespace lock as part of vdev reconfiguration. The
978 * separate spare lock exists for the status query path, which does not need to
979 * be completely consistent with respect to other vdev configuration changes.
983 spa_spare_compare(const void *a
, const void *b
)
985 return (spa_aux_compare(a
, b
));
989 spa_spare_add(vdev_t
*vd
)
991 mutex_enter(&spa_spare_lock
);
992 ASSERT(!vd
->vdev_isspare
);
993 spa_aux_add(vd
, &spa_spare_avl
);
994 vd
->vdev_isspare
= B_TRUE
;
995 mutex_exit(&spa_spare_lock
);
999 spa_spare_remove(vdev_t
*vd
)
1001 mutex_enter(&spa_spare_lock
);
1002 ASSERT(vd
->vdev_isspare
);
1003 spa_aux_remove(vd
, &spa_spare_avl
);
1004 vd
->vdev_isspare
= B_FALSE
;
1005 mutex_exit(&spa_spare_lock
);
1009 spa_spare_exists(uint64_t guid
, uint64_t *pool
, int *refcnt
)
1013 mutex_enter(&spa_spare_lock
);
1014 found
= spa_aux_exists(guid
, pool
, refcnt
, &spa_spare_avl
);
1015 mutex_exit(&spa_spare_lock
);
1021 spa_spare_activate(vdev_t
*vd
)
1023 mutex_enter(&spa_spare_lock
);
1024 ASSERT(vd
->vdev_isspare
);
1025 spa_aux_activate(vd
, &spa_spare_avl
);
1026 mutex_exit(&spa_spare_lock
);
1030 * Level 2 ARC devices are tracked globally for the same reasons as spares.
1031 * Cache devices currently only support one pool per cache device, and so
1032 * for these devices the aux reference count is currently unused beyond 1.
1036 spa_l2cache_compare(const void *a
, const void *b
)
1038 return (spa_aux_compare(a
, b
));
1042 spa_l2cache_add(vdev_t
*vd
)
1044 mutex_enter(&spa_l2cache_lock
);
1045 ASSERT(!vd
->vdev_isl2cache
);
1046 spa_aux_add(vd
, &spa_l2cache_avl
);
1047 vd
->vdev_isl2cache
= B_TRUE
;
1048 mutex_exit(&spa_l2cache_lock
);
1052 spa_l2cache_remove(vdev_t
*vd
)
1054 mutex_enter(&spa_l2cache_lock
);
1055 ASSERT(vd
->vdev_isl2cache
);
1056 spa_aux_remove(vd
, &spa_l2cache_avl
);
1057 vd
->vdev_isl2cache
= B_FALSE
;
1058 mutex_exit(&spa_l2cache_lock
);
1062 spa_l2cache_exists(uint64_t guid
, uint64_t *pool
)
1066 mutex_enter(&spa_l2cache_lock
);
1067 found
= spa_aux_exists(guid
, pool
, NULL
, &spa_l2cache_avl
);
1068 mutex_exit(&spa_l2cache_lock
);
1074 spa_l2cache_activate(vdev_t
*vd
)
1076 mutex_enter(&spa_l2cache_lock
);
1077 ASSERT(vd
->vdev_isl2cache
);
1078 spa_aux_activate(vd
, &spa_l2cache_avl
);
1079 mutex_exit(&spa_l2cache_lock
);
1083 * ==========================================================================
1085 * ==========================================================================
1089 * Lock the given spa_t for the purpose of adding or removing a vdev.
1090 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1091 * It returns the next transaction group for the spa_t.
1094 spa_vdev_enter(spa_t
*spa
)
1096 mutex_enter(&spa
->spa_vdev_top_lock
);
1097 mutex_enter(&spa_namespace_lock
);
1098 return (spa_vdev_config_enter(spa
));
1102 * Internal implementation for spa_vdev_enter(). Used when a vdev
1103 * operation requires multiple syncs (i.e. removing a device) while
1104 * keeping the spa_namespace_lock held.
1107 spa_vdev_config_enter(spa_t
*spa
)
1109 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1111 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
1113 return (spa_last_synced_txg(spa
) + 1);
1117 * Used in combination with spa_vdev_config_enter() to allow the syncing
1118 * of multiple transactions without releasing the spa_namespace_lock.
1121 spa_vdev_config_exit(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int error
, char *tag
)
1123 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1125 int config_changed
= B_FALSE
;
1127 ASSERT(txg
> spa_last_synced_txg(spa
));
1129 spa
->spa_pending_vdev
= NULL
;
1132 * Reassess the DTLs.
1134 vdev_dtl_reassess(spa
->spa_root_vdev
, 0, 0, B_FALSE
);
1136 if (error
== 0 && !list_is_empty(&spa
->spa_config_dirty_list
)) {
1137 config_changed
= B_TRUE
;
1138 spa
->spa_config_generation
++;
1142 * Verify the metaslab classes.
1144 ASSERT(metaslab_class_validate(spa_normal_class(spa
)) == 0);
1145 ASSERT(metaslab_class_validate(spa_log_class(spa
)) == 0);
1147 spa_config_exit(spa
, SCL_ALL
, spa
);
1150 * Panic the system if the specified tag requires it. This
1151 * is useful for ensuring that configurations are updated
1154 if (zio_injection_enabled
)
1155 zio_handle_panic_injection(spa
, tag
, 0);
1158 * Note: this txg_wait_synced() is important because it ensures
1159 * that there won't be more than one config change per txg.
1160 * This allows us to use the txg as the generation number.
1163 txg_wait_synced(spa
->spa_dsl_pool
, txg
);
1166 ASSERT(!vd
->vdev_detached
|| vd
->vdev_dtl_sm
== NULL
);
1167 spa_config_enter(spa
, SCL_ALL
, spa
, RW_WRITER
);
1169 spa_config_exit(spa
, SCL_ALL
, spa
);
1173 * If the config changed, update the config cache.
1176 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
1180 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1181 * locking of spa_vdev_enter(), we also want make sure the transactions have
1182 * synced to disk, and then update the global configuration cache with the new
1186 spa_vdev_exit(spa_t
*spa
, vdev_t
*vd
, uint64_t txg
, int error
)
1188 spa_vdev_config_exit(spa
, vd
, txg
, error
, FTAG
);
1189 mutex_exit(&spa_namespace_lock
);
1190 mutex_exit(&spa
->spa_vdev_top_lock
);
1196 * Lock the given spa_t for the purpose of changing vdev state.
1199 spa_vdev_state_enter(spa_t
*spa
, int oplocks
)
1201 int locks
= SCL_STATE_ALL
| oplocks
;
1204 * Root pools may need to read of the underlying devfs filesystem
1205 * when opening up a vdev. Unfortunately if we're holding the
1206 * SCL_ZIO lock it will result in a deadlock when we try to issue
1207 * the read from the root filesystem. Instead we "prefetch"
1208 * the associated vnodes that we need prior to opening the
1209 * underlying devices and cache them so that we can prevent
1210 * any I/O when we are doing the actual open.
1212 if (spa_is_root(spa
)) {
1213 int low
= locks
& ~(SCL_ZIO
- 1);
1214 int high
= locks
& ~low
;
1216 spa_config_enter(spa
, high
, spa
, RW_WRITER
);
1217 vdev_hold(spa
->spa_root_vdev
);
1218 spa_config_enter(spa
, low
, spa
, RW_WRITER
);
1220 spa_config_enter(spa
, locks
, spa
, RW_WRITER
);
1222 spa
->spa_vdev_locks
= locks
;
1226 spa_vdev_state_exit(spa_t
*spa
, vdev_t
*vd
, int error
)
1228 boolean_t config_changed
= B_FALSE
;
1231 if (vd
== NULL
|| vd
== spa
->spa_root_vdev
) {
1232 vdev_top
= spa
->spa_root_vdev
;
1234 vdev_top
= vd
->vdev_top
;
1237 if (vd
!= NULL
|| error
== 0)
1238 vdev_dtl_reassess(vdev_top
, 0, 0, B_FALSE
);
1241 if (vd
!= spa
->spa_root_vdev
)
1242 vdev_state_dirty(vdev_top
);
1244 config_changed
= B_TRUE
;
1245 spa
->spa_config_generation
++;
1248 if (spa_is_root(spa
))
1249 vdev_rele(spa
->spa_root_vdev
);
1251 ASSERT3U(spa
->spa_vdev_locks
, >=, SCL_STATE_ALL
);
1252 spa_config_exit(spa
, spa
->spa_vdev_locks
, spa
);
1255 * If anything changed, wait for it to sync. This ensures that,
1256 * from the system administrator's perspective, zpool(1M) commands
1257 * are synchronous. This is important for things like zpool offline:
1258 * when the command completes, you expect no further I/O from ZFS.
1261 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1264 * If the config changed, update the config cache.
1266 if (config_changed
) {
1267 mutex_enter(&spa_namespace_lock
);
1268 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
1269 mutex_exit(&spa_namespace_lock
);
1276 * ==========================================================================
1277 * Miscellaneous functions
1278 * ==========================================================================
1282 spa_activate_mos_feature(spa_t
*spa
, const char *feature
, dmu_tx_t
*tx
)
1284 if (!nvlist_exists(spa
->spa_label_features
, feature
)) {
1285 fnvlist_add_boolean(spa
->spa_label_features
, feature
);
1287 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1288 * dirty the vdev config because lock SCL_CONFIG is not held.
1289 * Thankfully, in this case we don't need to dirty the config
1290 * because it will be written out anyway when we finish
1291 * creating the pool.
1293 if (tx
->tx_txg
!= TXG_INITIAL
)
1294 vdev_config_dirty(spa
->spa_root_vdev
);
1299 spa_deactivate_mos_feature(spa_t
*spa
, const char *feature
)
1301 if (nvlist_remove_all(spa
->spa_label_features
, feature
) == 0)
1302 vdev_config_dirty(spa
->spa_root_vdev
);
1309 spa_rename(const char *name
, const char *newname
)
1315 * Lookup the spa_t and grab the config lock for writing. We need to
1316 * actually open the pool so that we can sync out the necessary labels.
1317 * It's OK to call spa_open() with the namespace lock held because we
1318 * allow recursive calls for other reasons.
1320 mutex_enter(&spa_namespace_lock
);
1321 if ((err
= spa_open(name
, &spa
, FTAG
)) != 0) {
1322 mutex_exit(&spa_namespace_lock
);
1326 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1328 avl_remove(&spa_namespace_avl
, spa
);
1329 (void) strlcpy(spa
->spa_name
, newname
, sizeof (spa
->spa_name
));
1330 avl_add(&spa_namespace_avl
, spa
);
1333 * Sync all labels to disk with the new names by marking the root vdev
1334 * dirty and waiting for it to sync. It will pick up the new pool name
1337 vdev_config_dirty(spa
->spa_root_vdev
);
1339 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1341 txg_wait_synced(spa
->spa_dsl_pool
, 0);
1344 * Sync the updated config cache.
1346 spa_write_cachefile(spa
, B_FALSE
, B_TRUE
);
1348 spa_close(spa
, FTAG
);
1350 mutex_exit(&spa_namespace_lock
);
1356 * Return the spa_t associated with given pool_guid, if it exists. If
1357 * device_guid is non-zero, determine whether the pool exists *and* contains
1358 * a device with the specified device_guid.
1361 spa_by_guid(uint64_t pool_guid
, uint64_t device_guid
)
1364 avl_tree_t
*t
= &spa_namespace_avl
;
1366 ASSERT(MUTEX_HELD(&spa_namespace_lock
));
1368 for (spa
= avl_first(t
); spa
!= NULL
; spa
= AVL_NEXT(t
, spa
)) {
1369 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
1371 if (spa
->spa_root_vdev
== NULL
)
1373 if (spa_guid(spa
) == pool_guid
) {
1374 if (device_guid
== 0)
1377 if (vdev_lookup_by_guid(spa
->spa_root_vdev
,
1378 device_guid
) != NULL
)
1382 * Check any devices we may be in the process of adding.
1384 if (spa
->spa_pending_vdev
) {
1385 if (vdev_lookup_by_guid(spa
->spa_pending_vdev
,
1386 device_guid
) != NULL
)
1396 * Determine whether a pool with the given pool_guid exists.
1399 spa_guid_exists(uint64_t pool_guid
, uint64_t device_guid
)
1401 return (spa_by_guid(pool_guid
, device_guid
) != NULL
);
1405 spa_strdup(const char *s
)
1411 new = kmem_alloc(len
+ 1, KM_SLEEP
);
1419 spa_strfree(char *s
)
1421 kmem_free(s
, strlen(s
) + 1);
1425 spa_get_random(uint64_t range
)
1434 (void) random_get_pseudo_bytes((void *)&r
, sizeof (uint64_t));
1440 spa_generate_guid(spa_t
*spa
)
1442 uint64_t guid
= spa_get_random(-1ULL);
1445 while (guid
== 0 || spa_guid_exists(spa_guid(spa
), guid
))
1446 guid
= spa_get_random(-1ULL);
1448 while (guid
== 0 || spa_guid_exists(guid
, 0))
1449 guid
= spa_get_random(-1ULL);
1456 snprintf_blkptr(char *buf
, size_t buflen
, const blkptr_t
*bp
)
1459 char *checksum
= NULL
;
1460 char *compress
= NULL
;
1463 if (BP_GET_TYPE(bp
) & DMU_OT_NEWTYPE
) {
1464 dmu_object_byteswap_t bswap
=
1465 DMU_OT_BYTESWAP(BP_GET_TYPE(bp
));
1466 (void) snprintf(type
, sizeof (type
), "bswap %s %s",
1467 DMU_OT_IS_METADATA(BP_GET_TYPE(bp
)) ?
1468 "metadata" : "data",
1469 dmu_ot_byteswap
[bswap
].ob_name
);
1471 (void) strlcpy(type
, dmu_ot
[BP_GET_TYPE(bp
)].ot_name
,
1474 if (!BP_IS_EMBEDDED(bp
)) {
1476 zio_checksum_table
[BP_GET_CHECKSUM(bp
)].ci_name
;
1478 compress
= zio_compress_table
[BP_GET_COMPRESS(bp
)].ci_name
;
1481 SNPRINTF_BLKPTR(snprintf
, ' ', buf
, buflen
, bp
, type
, checksum
,
1486 spa_freeze(spa_t
*spa
)
1488 uint64_t freeze_txg
= 0;
1490 spa_config_enter(spa
, SCL_ALL
, FTAG
, RW_WRITER
);
1491 if (spa
->spa_freeze_txg
== UINT64_MAX
) {
1492 freeze_txg
= spa_last_synced_txg(spa
) + TXG_SIZE
;
1493 spa
->spa_freeze_txg
= freeze_txg
;
1495 spa_config_exit(spa
, SCL_ALL
, FTAG
);
1496 if (freeze_txg
!= 0)
1497 txg_wait_synced(spa_get_dsl(spa
), freeze_txg
);
1501 zfs_panic_recover(const char *fmt
, ...)
1506 vcmn_err(zfs_recover
? CE_WARN
: CE_PANIC
, fmt
, adx
);
1511 * This is a stripped-down version of strtoull, suitable only for converting
1512 * lowercase hexadecimal numbers that don't overflow.
1515 zfs_strtonum(const char *str
, char **nptr
)
1521 while ((c
= *str
) != '\0') {
1522 if (c
>= '0' && c
<= '9')
1524 else if (c
>= 'a' && c
<= 'f')
1525 digit
= 10 + c
- 'a';
1536 *nptr
= (char *)str
;
1542 * ==========================================================================
1543 * Accessor functions
1544 * ==========================================================================
1548 spa_shutting_down(spa_t
*spa
)
1550 return (spa
->spa_async_suspended
);
1554 spa_get_dsl(spa_t
*spa
)
1556 return (spa
->spa_dsl_pool
);
1560 spa_is_initializing(spa_t
*spa
)
1562 return (spa
->spa_is_initializing
);
1566 spa_indirect_vdevs_loaded(spa_t
*spa
)
1568 return (spa
->spa_indirect_vdevs_loaded
);
1572 spa_get_rootblkptr(spa_t
*spa
)
1574 return (&spa
->spa_ubsync
.ub_rootbp
);
1578 spa_set_rootblkptr(spa_t
*spa
, const blkptr_t
*bp
)
1580 spa
->spa_uberblock
.ub_rootbp
= *bp
;
1584 spa_altroot(spa_t
*spa
, char *buf
, size_t buflen
)
1586 if (spa
->spa_root
== NULL
)
1589 (void) strncpy(buf
, spa
->spa_root
, buflen
);
1593 spa_sync_pass(spa_t
*spa
)
1595 return (spa
->spa_sync_pass
);
1599 spa_name(spa_t
*spa
)
1601 return (spa
->spa_name
);
1605 spa_guid(spa_t
*spa
)
1607 dsl_pool_t
*dp
= spa_get_dsl(spa
);
1611 * If we fail to parse the config during spa_load(), we can go through
1612 * the error path (which posts an ereport) and end up here with no root
1613 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1616 if (spa
->spa_root_vdev
== NULL
)
1617 return (spa
->spa_config_guid
);
1619 guid
= spa
->spa_last_synced_guid
!= 0 ?
1620 spa
->spa_last_synced_guid
: spa
->spa_root_vdev
->vdev_guid
;
1623 * Return the most recently synced out guid unless we're
1624 * in syncing context.
1626 if (dp
&& dsl_pool_sync_context(dp
))
1627 return (spa
->spa_root_vdev
->vdev_guid
);
1633 spa_load_guid(spa_t
*spa
)
1636 * This is a GUID that exists solely as a reference for the
1637 * purposes of the arc. It is generated at load time, and
1638 * is never written to persistent storage.
1640 return (spa
->spa_load_guid
);
1644 spa_last_synced_txg(spa_t
*spa
)
1646 return (spa
->spa_ubsync
.ub_txg
);
1650 spa_first_txg(spa_t
*spa
)
1652 return (spa
->spa_first_txg
);
1656 spa_syncing_txg(spa_t
*spa
)
1658 return (spa
->spa_syncing_txg
);
1662 * Return the last txg where data can be dirtied. The final txgs
1663 * will be used to just clear out any deferred frees that remain.
1666 spa_final_dirty_txg(spa_t
*spa
)
1668 return (spa
->spa_final_txg
- TXG_DEFER_SIZE
);
1672 spa_state(spa_t
*spa
)
1674 return (spa
->spa_state
);
1678 spa_load_state(spa_t
*spa
)
1680 return (spa
->spa_load_state
);
1684 spa_freeze_txg(spa_t
*spa
)
1686 return (spa
->spa_freeze_txg
);
1690 * Return the inflated asize for a logical write in bytes. This is used by the
1691 * DMU to calculate the space a logical write will require on disk.
1692 * If lsize is smaller than the largest physical block size allocatable on this
1693 * pool we use its value instead, since the write will end up using the whole
1697 spa_get_worst_case_asize(spa_t
*spa
, uint64_t lsize
)
1700 return (0); /* No inflation needed */
1701 return (MAX(lsize
, 1 << spa
->spa_max_ashift
) * spa_asize_inflation
);
1705 * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%),
1706 * or at least 128MB, unless that would cause it to be more than half the
1709 * See the comment above spa_slop_shift for details.
1712 spa_get_slop_space(spa_t
*spa
)
1714 uint64_t space
= spa_get_dspace(spa
);
1715 return (MAX(space
>> spa_slop_shift
, MIN(space
>> 1, spa_min_slop
)));
1719 spa_get_dspace(spa_t
*spa
)
1721 return (spa
->spa_dspace
);
1725 spa_get_checkpoint_space(spa_t
*spa
)
1727 return (spa
->spa_checkpoint_info
.sci_dspace
);
1731 spa_update_dspace(spa_t
*spa
)
1733 spa
->spa_dspace
= metaslab_class_get_dspace(spa_normal_class(spa
)) +
1734 ddt_get_dedup_dspace(spa
);
1735 if (spa
->spa_vdev_removal
!= NULL
) {
1737 * We can't allocate from the removing device, so
1738 * subtract its size. This prevents the DMU/DSL from
1739 * filling up the (now smaller) pool while we are in the
1740 * middle of removing the device.
1742 * Note that the DMU/DSL doesn't actually know or care
1743 * how much space is allocated (it does its own tracking
1744 * of how much space has been logically used). So it
1745 * doesn't matter that the data we are moving may be
1746 * allocated twice (on the old device and the new
1749 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
1751 vdev_lookup_top(spa
, spa
->spa_vdev_removal
->svr_vdev_id
);
1752 spa
->spa_dspace
-= spa_deflate(spa
) ?
1753 vd
->vdev_stat
.vs_dspace
: vd
->vdev_stat
.vs_space
;
1754 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
1759 * Return the failure mode that has been set to this pool. The default
1760 * behavior will be to block all I/Os when a complete failure occurs.
1763 spa_get_failmode(spa_t
*spa
)
1765 return (spa
->spa_failmode
);
1769 spa_suspended(spa_t
*spa
)
1771 return (spa
->spa_suspended
!= ZIO_SUSPEND_NONE
);
1775 spa_version(spa_t
*spa
)
1777 return (spa
->spa_ubsync
.ub_version
);
1781 spa_deflate(spa_t
*spa
)
1783 return (spa
->spa_deflate
);
1787 spa_normal_class(spa_t
*spa
)
1789 return (spa
->spa_normal_class
);
1793 spa_log_class(spa_t
*spa
)
1795 return (spa
->spa_log_class
);
1799 spa_evicting_os_register(spa_t
*spa
, objset_t
*os
)
1801 mutex_enter(&spa
->spa_evicting_os_lock
);
1802 list_insert_head(&spa
->spa_evicting_os_list
, os
);
1803 mutex_exit(&spa
->spa_evicting_os_lock
);
1807 spa_evicting_os_deregister(spa_t
*spa
, objset_t
*os
)
1809 mutex_enter(&spa
->spa_evicting_os_lock
);
1810 list_remove(&spa
->spa_evicting_os_list
, os
);
1811 cv_broadcast(&spa
->spa_evicting_os_cv
);
1812 mutex_exit(&spa
->spa_evicting_os_lock
);
1816 spa_evicting_os_wait(spa_t
*spa
)
1818 mutex_enter(&spa
->spa_evicting_os_lock
);
1819 while (!list_is_empty(&spa
->spa_evicting_os_list
))
1820 cv_wait(&spa
->spa_evicting_os_cv
, &spa
->spa_evicting_os_lock
);
1821 mutex_exit(&spa
->spa_evicting_os_lock
);
1823 dmu_buf_user_evict_wait();
1827 spa_max_replication(spa_t
*spa
)
1830 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1831 * handle BPs with more than one DVA allocated. Set our max
1832 * replication level accordingly.
1834 if (spa_version(spa
) < SPA_VERSION_DITTO_BLOCKS
)
1836 return (MIN(SPA_DVAS_PER_BP
, spa_max_replication_override
));
1840 spa_prev_software_version(spa_t
*spa
)
1842 return (spa
->spa_prev_software_version
);
1846 spa_deadman_synctime(spa_t
*spa
)
1848 return (spa
->spa_deadman_synctime
);
1852 spa_deadman_ziotime(spa_t
*spa
)
1854 return (spa
->spa_deadman_ziotime
);
1858 spa_get_deadman_failmode(spa_t
*spa
)
1860 return (spa
->spa_deadman_failmode
);
1864 spa_set_deadman_failmode(spa_t
*spa
, const char *failmode
)
1866 if (strcmp(failmode
, "wait") == 0)
1867 spa
->spa_deadman_failmode
= ZIO_FAILURE_MODE_WAIT
;
1868 else if (strcmp(failmode
, "continue") == 0)
1869 spa
->spa_deadman_failmode
= ZIO_FAILURE_MODE_CONTINUE
;
1870 else if (strcmp(failmode
, "panic") == 0)
1871 spa
->spa_deadman_failmode
= ZIO_FAILURE_MODE_PANIC
;
1873 spa
->spa_deadman_failmode
= ZIO_FAILURE_MODE_WAIT
;
1877 dva_get_dsize_sync(spa_t
*spa
, const dva_t
*dva
)
1879 uint64_t asize
= DVA_GET_ASIZE(dva
);
1880 uint64_t dsize
= asize
;
1882 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
1884 if (asize
!= 0 && spa
->spa_deflate
) {
1885 vdev_t
*vd
= vdev_lookup_top(spa
, DVA_GET_VDEV(dva
));
1887 dsize
= (asize
>> SPA_MINBLOCKSHIFT
) *
1888 vd
->vdev_deflate_ratio
;
1895 bp_get_dsize_sync(spa_t
*spa
, const blkptr_t
*bp
)
1899 for (int d
= 0; d
< BP_GET_NDVAS(bp
); d
++)
1900 dsize
+= dva_get_dsize_sync(spa
, &bp
->blk_dva
[d
]);
1906 bp_get_dsize(spa_t
*spa
, const blkptr_t
*bp
)
1910 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
1912 for (int d
= 0; d
< BP_GET_NDVAS(bp
); d
++)
1913 dsize
+= dva_get_dsize_sync(spa
, &bp
->blk_dva
[d
]);
1915 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
1921 spa_dirty_data(spa_t
*spa
)
1923 return (spa
->spa_dsl_pool
->dp_dirty_total
);
1927 * ==========================================================================
1928 * Initialization and Termination
1929 * ==========================================================================
1933 spa_name_compare(const void *a1
, const void *a2
)
1935 const spa_t
*s1
= a1
;
1936 const spa_t
*s2
= a2
;
1939 s
= strcmp(s1
->spa_name
, s2
->spa_name
);
1941 return (AVL_ISIGN(s
));
1953 mutex_init(&spa_namespace_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1954 mutex_init(&spa_spare_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1955 mutex_init(&spa_l2cache_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1956 cv_init(&spa_namespace_cv
, NULL
, CV_DEFAULT
, NULL
);
1958 avl_create(&spa_namespace_avl
, spa_name_compare
, sizeof (spa_t
),
1959 offsetof(spa_t
, spa_avl
));
1961 avl_create(&spa_spare_avl
, spa_spare_compare
, sizeof (spa_aux_t
),
1962 offsetof(spa_aux_t
, aux_avl
));
1964 avl_create(&spa_l2cache_avl
, spa_l2cache_compare
, sizeof (spa_aux_t
),
1965 offsetof(spa_aux_t
, aux_avl
));
1967 spa_mode_global
= mode
;
1970 if (spa_mode_global
!= FREAD
&& dprintf_find_string("watch")) {
1971 struct sigaction sa
;
1973 sa
.sa_flags
= SA_SIGINFO
;
1974 sigemptyset(&sa
.sa_mask
);
1975 sa
.sa_sigaction
= arc_buf_sigsegv
;
1977 if (sigaction(SIGSEGV
, &sa
, NULL
) == -1) {
1978 perror("could not enable watchpoints: "
1979 "sigaction(SIGSEGV, ...) = ");
1990 metaslab_alloc_trace_init();
1995 vdev_cache_stat_init();
1996 vdev_mirror_stat_init();
1997 vdev_raidz_math_init();
2001 zpool_feature_init();
2016 vdev_cache_stat_fini();
2017 vdev_mirror_stat_fini();
2018 vdev_raidz_math_fini();
2023 metaslab_alloc_trace_fini();
2031 avl_destroy(&spa_namespace_avl
);
2032 avl_destroy(&spa_spare_avl
);
2033 avl_destroy(&spa_l2cache_avl
);
2035 cv_destroy(&spa_namespace_cv
);
2036 mutex_destroy(&spa_namespace_lock
);
2037 mutex_destroy(&spa_spare_lock
);
2038 mutex_destroy(&spa_l2cache_lock
);
2042 * Return whether this pool has slogs. No locking needed.
2043 * It's not a problem if the wrong answer is returned as it's only for
2044 * performance and not correctness
2047 spa_has_slogs(spa_t
*spa
)
2049 return (spa
->spa_log_class
->mc_rotor
!= NULL
);
2053 spa_get_log_state(spa_t
*spa
)
2055 return (spa
->spa_log_state
);
2059 spa_set_log_state(spa_t
*spa
, spa_log_state_t state
)
2061 spa
->spa_log_state
= state
;
2065 spa_is_root(spa_t
*spa
)
2067 return (spa
->spa_is_root
);
2071 spa_writeable(spa_t
*spa
)
2073 return (!!(spa
->spa_mode
& FWRITE
) && spa
->spa_trust_config
);
2077 * Returns true if there is a pending sync task in any of the current
2078 * syncing txg, the current quiescing txg, or the current open txg.
2081 spa_has_pending_synctask(spa_t
*spa
)
2083 return (!txg_all_lists_empty(&spa
->spa_dsl_pool
->dp_sync_tasks
) ||
2084 !txg_all_lists_empty(&spa
->spa_dsl_pool
->dp_early_sync_tasks
));
2088 spa_mode(spa_t
*spa
)
2090 return (spa
->spa_mode
);
2094 spa_bootfs(spa_t
*spa
)
2096 return (spa
->spa_bootfs
);
2100 spa_delegation(spa_t
*spa
)
2102 return (spa
->spa_delegation
);
2106 spa_meta_objset(spa_t
*spa
)
2108 return (spa
->spa_meta_objset
);
2112 spa_dedup_checksum(spa_t
*spa
)
2114 return (spa
->spa_dedup_checksum
);
2118 * Reset pool scan stat per scan pass (or reboot).
2121 spa_scan_stat_init(spa_t
*spa
)
2123 /* data not stored on disk */
2124 spa
->spa_scan_pass_start
= gethrestime_sec();
2125 if (dsl_scan_is_paused_scrub(spa
->spa_dsl_pool
->dp_scan
))
2126 spa
->spa_scan_pass_scrub_pause
= spa
->spa_scan_pass_start
;
2128 spa
->spa_scan_pass_scrub_pause
= 0;
2129 spa
->spa_scan_pass_scrub_spent_paused
= 0;
2130 spa
->spa_scan_pass_exam
= 0;
2131 spa
->spa_scan_pass_issued
= 0;
2132 vdev_scan_stat_init(spa
->spa_root_vdev
);
2136 * Get scan stats for zpool status reports
2139 spa_scan_get_stats(spa_t
*spa
, pool_scan_stat_t
*ps
)
2141 dsl_scan_t
*scn
= spa
->spa_dsl_pool
? spa
->spa_dsl_pool
->dp_scan
: NULL
;
2143 if (scn
== NULL
|| scn
->scn_phys
.scn_func
== POOL_SCAN_NONE
)
2144 return (SET_ERROR(ENOENT
));
2145 bzero(ps
, sizeof (pool_scan_stat_t
));
2147 /* data stored on disk */
2148 ps
->pss_func
= scn
->scn_phys
.scn_func
;
2149 ps
->pss_state
= scn
->scn_phys
.scn_state
;
2150 ps
->pss_start_time
= scn
->scn_phys
.scn_start_time
;
2151 ps
->pss_end_time
= scn
->scn_phys
.scn_end_time
;
2152 ps
->pss_to_examine
= scn
->scn_phys
.scn_to_examine
;
2153 ps
->pss_examined
= scn
->scn_phys
.scn_examined
;
2154 ps
->pss_to_process
= scn
->scn_phys
.scn_to_process
;
2155 ps
->pss_processed
= scn
->scn_phys
.scn_processed
;
2156 ps
->pss_errors
= scn
->scn_phys
.scn_errors
;
2158 /* data not stored on disk */
2159 ps
->pss_pass_exam
= spa
->spa_scan_pass_exam
;
2160 ps
->pss_pass_start
= spa
->spa_scan_pass_start
;
2161 ps
->pss_pass_scrub_pause
= spa
->spa_scan_pass_scrub_pause
;
2162 ps
->pss_pass_scrub_spent_paused
= spa
->spa_scan_pass_scrub_spent_paused
;
2163 ps
->pss_pass_issued
= spa
->spa_scan_pass_issued
;
2165 scn
->scn_issued_before_pass
+ spa
->spa_scan_pass_issued
;
2171 spa_maxblocksize(spa_t
*spa
)
2173 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_BLOCKS
))
2174 return (SPA_MAXBLOCKSIZE
);
2176 return (SPA_OLD_MAXBLOCKSIZE
);
2181 * Returns the txg that the last device removal completed. No indirect mappings
2182 * have been added since this txg.
2185 spa_get_last_removal_txg(spa_t
*spa
)
2188 uint64_t ret
= -1ULL;
2190 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
2192 * sr_prev_indirect_vdev is only modified while holding all the
2193 * config locks, so it is sufficient to hold SCL_VDEV as reader when
2196 vdevid
= spa
->spa_removing_phys
.sr_prev_indirect_vdev
;
2198 while (vdevid
!= -1ULL) {
2199 vdev_t
*vd
= vdev_lookup_top(spa
, vdevid
);
2200 vdev_indirect_births_t
*vib
= vd
->vdev_indirect_births
;
2202 ASSERT3P(vd
->vdev_ops
, ==, &vdev_indirect_ops
);
2205 * If the removal did not remap any data, we don't care.
2207 if (vdev_indirect_births_count(vib
) != 0) {
2208 ret
= vdev_indirect_births_last_entry_txg(vib
);
2212 vdevid
= vd
->vdev_indirect_config
.vic_prev_indirect_vdev
;
2214 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
2217 spa_feature_is_active(spa
, SPA_FEATURE_DEVICE_REMOVAL
));
2223 spa_maxdnodesize(spa_t
*spa
)
2225 if (spa_feature_is_enabled(spa
, SPA_FEATURE_LARGE_DNODE
))
2226 return (DNODE_MAX_SIZE
);
2228 return (DNODE_MIN_SIZE
);
2232 spa_multihost(spa_t
*spa
)
2234 return (spa
->spa_multihost
? B_TRUE
: B_FALSE
);
2238 spa_get_hostid(void)
2240 unsigned long myhostid
;
2243 myhostid
= zone_get_hostid(NULL
);
2246 * We're emulating the system's hostid in userland, so
2247 * we can't use zone_get_hostid().
2249 (void) ddi_strtoul(hw_serial
, NULL
, 10, &myhostid
);
2250 #endif /* _KERNEL */
2256 spa_trust_config(spa_t
*spa
)
2258 return (spa
->spa_trust_config
);
2262 spa_missing_tvds_allowed(spa_t
*spa
)
2264 return (spa
->spa_missing_tvds_allowed
);
2268 spa_set_missing_tvds(spa_t
*spa
, uint64_t missing
)
2270 spa
->spa_missing_tvds
= missing
;
2274 * Return the pool state string ("ONLINE", "DEGRADED", "SUSPENDED", etc).
2277 spa_state_to_name(spa_t
*spa
)
2279 vdev_state_t state
= spa
->spa_root_vdev
->vdev_state
;
2280 vdev_aux_t aux
= spa
->spa_root_vdev
->vdev_stat
.vs_aux
;
2282 if (spa_suspended(spa
) &&
2283 (spa_get_failmode(spa
) != ZIO_FAILURE_MODE_CONTINUE
))
2284 return ("SUSPENDED");
2287 case VDEV_STATE_CLOSED
:
2288 case VDEV_STATE_OFFLINE
:
2290 case VDEV_STATE_REMOVED
:
2292 case VDEV_STATE_CANT_OPEN
:
2293 if (aux
== VDEV_AUX_CORRUPT_DATA
|| aux
== VDEV_AUX_BAD_LOG
)
2295 else if (aux
== VDEV_AUX_SPLIT_POOL
)
2299 case VDEV_STATE_FAULTED
:
2301 case VDEV_STATE_DEGRADED
:
2302 return ("DEGRADED");
2303 case VDEV_STATE_HEALTHY
:
2313 spa_top_vdevs_spacemap_addressable(spa_t
*spa
)
2315 vdev_t
*rvd
= spa
->spa_root_vdev
;
2316 for (uint64_t c
= 0; c
< rvd
->vdev_children
; c
++) {
2317 if (!vdev_is_spacemap_addressable(rvd
->vdev_child
[c
]))
2324 spa_has_checkpoint(spa_t
*spa
)
2326 return (spa
->spa_checkpoint_txg
!= 0);
2330 spa_importing_readonly_checkpoint(spa_t
*spa
)
2332 return ((spa
->spa_import_flags
& ZFS_IMPORT_CHECKPOINT
) &&
2333 spa
->spa_mode
== FREAD
);
2337 spa_min_claim_txg(spa_t
*spa
)
2339 uint64_t checkpoint_txg
= spa
->spa_uberblock
.ub_checkpoint_txg
;
2341 if (checkpoint_txg
!= 0)
2342 return (checkpoint_txg
+ 1);
2344 return (spa
->spa_first_txg
);
2348 * If there is a checkpoint, async destroys may consume more space from
2349 * the pool instead of freeing it. In an attempt to save the pool from
2350 * getting suspended when it is about to run out of space, we stop
2351 * processing async destroys.
2354 spa_suspend_async_destroy(spa_t
*spa
)
2356 dsl_pool_t
*dp
= spa_get_dsl(spa
);
2358 uint64_t unreserved
= dsl_pool_unreserved_space(dp
,
2359 ZFS_SPACE_CHECK_EXTRA_RESERVED
);
2360 uint64_t used
= dsl_dir_phys(dp
->dp_root_dir
)->dd_used_bytes
;
2361 uint64_t avail
= (unreserved
> used
) ? (unreserved
- used
) : 0;
2363 if (spa_has_checkpoint(spa
) && avail
== 0)
2369 #if defined(_KERNEL)
2371 #include <linux/mod_compat.h>
2374 param_set_deadman_failmode(const char *val
, zfs_kernel_param_t
*kp
)
2380 return (SET_ERROR(-EINVAL
));
2382 if ((p
= strchr(val
, '\n')) != NULL
)
2385 if (strcmp(val
, "wait") != 0 && strcmp(val
, "continue") != 0 &&
2386 strcmp(val
, "panic"))
2387 return (SET_ERROR(-EINVAL
));
2389 if (spa_mode_global
!= 0) {
2390 mutex_enter(&spa_namespace_lock
);
2391 while ((spa
= spa_next(spa
)) != NULL
)
2392 spa_set_deadman_failmode(spa
, val
);
2393 mutex_exit(&spa_namespace_lock
);
2396 return (param_set_charp(val
, kp
));
2400 param_set_deadman_ziotime(const char *val
, zfs_kernel_param_t
*kp
)
2405 error
= param_set_ulong(val
, kp
);
2407 return (SET_ERROR(error
));
2409 if (spa_mode_global
!= 0) {
2410 mutex_enter(&spa_namespace_lock
);
2411 while ((spa
= spa_next(spa
)) != NULL
)
2412 spa
->spa_deadman_ziotime
=
2413 MSEC2NSEC(zfs_deadman_ziotime_ms
);
2414 mutex_exit(&spa_namespace_lock
);
2421 param_set_deadman_synctime(const char *val
, zfs_kernel_param_t
*kp
)
2426 error
= param_set_ulong(val
, kp
);
2428 return (SET_ERROR(error
));
2430 if (spa_mode_global
!= 0) {
2431 mutex_enter(&spa_namespace_lock
);
2432 while ((spa
= spa_next(spa
)) != NULL
)
2433 spa
->spa_deadman_synctime
=
2434 MSEC2NSEC(zfs_deadman_synctime_ms
);
2435 mutex_exit(&spa_namespace_lock
);
2441 /* Namespace manipulation */
2442 EXPORT_SYMBOL(spa_lookup
);
2443 EXPORT_SYMBOL(spa_add
);
2444 EXPORT_SYMBOL(spa_remove
);
2445 EXPORT_SYMBOL(spa_next
);
2447 /* Refcount functions */
2448 EXPORT_SYMBOL(spa_open_ref
);
2449 EXPORT_SYMBOL(spa_close
);
2450 EXPORT_SYMBOL(spa_refcount_zero
);
2452 /* Pool configuration lock */
2453 EXPORT_SYMBOL(spa_config_tryenter
);
2454 EXPORT_SYMBOL(spa_config_enter
);
2455 EXPORT_SYMBOL(spa_config_exit
);
2456 EXPORT_SYMBOL(spa_config_held
);
2458 /* Pool vdev add/remove lock */
2459 EXPORT_SYMBOL(spa_vdev_enter
);
2460 EXPORT_SYMBOL(spa_vdev_exit
);
2462 /* Pool vdev state change lock */
2463 EXPORT_SYMBOL(spa_vdev_state_enter
);
2464 EXPORT_SYMBOL(spa_vdev_state_exit
);
2466 /* Accessor functions */
2467 EXPORT_SYMBOL(spa_shutting_down
);
2468 EXPORT_SYMBOL(spa_get_dsl
);
2469 EXPORT_SYMBOL(spa_get_rootblkptr
);
2470 EXPORT_SYMBOL(spa_set_rootblkptr
);
2471 EXPORT_SYMBOL(spa_altroot
);
2472 EXPORT_SYMBOL(spa_sync_pass
);
2473 EXPORT_SYMBOL(spa_name
);
2474 EXPORT_SYMBOL(spa_guid
);
2475 EXPORT_SYMBOL(spa_last_synced_txg
);
2476 EXPORT_SYMBOL(spa_first_txg
);
2477 EXPORT_SYMBOL(spa_syncing_txg
);
2478 EXPORT_SYMBOL(spa_version
);
2479 EXPORT_SYMBOL(spa_state
);
2480 EXPORT_SYMBOL(spa_load_state
);
2481 EXPORT_SYMBOL(spa_freeze_txg
);
2482 EXPORT_SYMBOL(spa_get_dspace
);
2483 EXPORT_SYMBOL(spa_update_dspace
);
2484 EXPORT_SYMBOL(spa_deflate
);
2485 EXPORT_SYMBOL(spa_normal_class
);
2486 EXPORT_SYMBOL(spa_log_class
);
2487 EXPORT_SYMBOL(spa_max_replication
);
2488 EXPORT_SYMBOL(spa_prev_software_version
);
2489 EXPORT_SYMBOL(spa_get_failmode
);
2490 EXPORT_SYMBOL(spa_suspended
);
2491 EXPORT_SYMBOL(spa_bootfs
);
2492 EXPORT_SYMBOL(spa_delegation
);
2493 EXPORT_SYMBOL(spa_meta_objset
);
2494 EXPORT_SYMBOL(spa_maxblocksize
);
2495 EXPORT_SYMBOL(spa_maxdnodesize
);
2497 /* Miscellaneous support routines */
2498 EXPORT_SYMBOL(spa_rename
);
2499 EXPORT_SYMBOL(spa_guid_exists
);
2500 EXPORT_SYMBOL(spa_strdup
);
2501 EXPORT_SYMBOL(spa_strfree
);
2502 EXPORT_SYMBOL(spa_get_random
);
2503 EXPORT_SYMBOL(spa_generate_guid
);
2504 EXPORT_SYMBOL(snprintf_blkptr
);
2505 EXPORT_SYMBOL(spa_freeze
);
2506 EXPORT_SYMBOL(spa_upgrade
);
2507 EXPORT_SYMBOL(spa_evict_all
);
2508 EXPORT_SYMBOL(spa_lookup_by_guid
);
2509 EXPORT_SYMBOL(spa_has_spare
);
2510 EXPORT_SYMBOL(dva_get_dsize_sync
);
2511 EXPORT_SYMBOL(bp_get_dsize_sync
);
2512 EXPORT_SYMBOL(bp_get_dsize
);
2513 EXPORT_SYMBOL(spa_has_slogs
);
2514 EXPORT_SYMBOL(spa_is_root
);
2515 EXPORT_SYMBOL(spa_writeable
);
2516 EXPORT_SYMBOL(spa_mode
);
2517 EXPORT_SYMBOL(spa_namespace_lock
);
2518 EXPORT_SYMBOL(spa_trust_config
);
2519 EXPORT_SYMBOL(spa_missing_tvds_allowed
);
2520 EXPORT_SYMBOL(spa_set_missing_tvds
);
2521 EXPORT_SYMBOL(spa_state_to_name
);
2522 EXPORT_SYMBOL(spa_importing_readonly_checkpoint
);
2523 EXPORT_SYMBOL(spa_min_claim_txg
);
2524 EXPORT_SYMBOL(spa_suspend_async_destroy
);
2525 EXPORT_SYMBOL(spa_has_checkpoint
);
2526 EXPORT_SYMBOL(spa_top_vdevs_spacemap_addressable
);
2529 module_param(zfs_flags
, uint
, 0644);
2530 MODULE_PARM_DESC(zfs_flags
, "Set additional debugging flags");
2532 module_param(zfs_recover
, int, 0644);
2533 MODULE_PARM_DESC(zfs_recover
, "Set to attempt to recover from fatal errors");
2535 module_param(zfs_free_leak_on_eio
, int, 0644);
2536 MODULE_PARM_DESC(zfs_free_leak_on_eio
,
2537 "Set to ignore IO errors during free and permanently leak the space");
2539 module_param_call(zfs_deadman_synctime_ms
, param_set_deadman_synctime
,
2540 param_get_ulong
, &zfs_deadman_synctime_ms
, 0644);
2541 MODULE_PARM_DESC(zfs_deadman_synctime_ms
,
2542 "Pool sync expiration time in milliseconds");
2544 module_param_call(zfs_deadman_ziotime_ms
, param_set_deadman_ziotime
,
2545 param_get_ulong
, &zfs_deadman_ziotime_ms
, 0644);
2546 MODULE_PARM_DESC(zfs_deadman_ziotime_ms
,
2547 "IO expiration time in milliseconds");
2549 module_param(zfs_deadman_checktime_ms
, ulong
, 0644);
2550 MODULE_PARM_DESC(zfs_deadman_checktime_ms
,
2551 "Dead I/O check interval in milliseconds");
2553 module_param(zfs_deadman_enabled
, int, 0644);
2554 MODULE_PARM_DESC(zfs_deadman_enabled
, "Enable deadman timer");
2556 module_param_call(zfs_deadman_failmode
, param_set_deadman_failmode
,
2557 param_get_charp
, &zfs_deadman_failmode
, 0644);
2558 MODULE_PARM_DESC(zfs_deadman_failmode
, "Failmode for deadman timer");
2560 module_param(spa_asize_inflation
, int, 0644);
2561 MODULE_PARM_DESC(spa_asize_inflation
,
2562 "SPA size estimate multiplication factor");
2564 module_param(spa_slop_shift
, int, 0644);
2565 MODULE_PARM_DESC(spa_slop_shift
, "Reserved free space in pool");