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 https://opensource.org/licenses/CDDL-1.0.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2019 by Delphix. All rights reserved.
24 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
26 * Copyright 2013 Saso Kiselkov. All rights reserved.
27 * Copyright (c) 2017 Datto Inc.
28 * Copyright (c) 2017, Intel Corporation.
29 * Copyright (c) 2019, loli10K <ezomori.nozomu@gmail.com>. All rights reserved.
32 #include <sys/zfs_context.h>
33 #include <sys/zfs_chksum.h>
34 #include <sys/spa_impl.h>
36 #include <sys/zio_checksum.h>
37 #include <sys/zio_compress.h>
39 #include <sys/dmu_tx.h>
42 #include <sys/vdev_impl.h>
43 #include <sys/vdev_initialize.h>
44 #include <sys/vdev_trim.h>
45 #include <sys/vdev_file.h>
46 #include <sys/vdev_raidz.h>
47 #include <sys/metaslab.h>
48 #include <sys/uberblock_impl.h>
51 #include <sys/unique.h>
52 #include <sys/dsl_pool.h>
53 #include <sys/dsl_dir.h>
54 #include <sys/dsl_prop.h>
55 #include <sys/fm/util.h>
56 #include <sys/dsl_scan.h>
57 #include <sys/fs/zfs.h>
58 #include <sys/metaslab_impl.h>
61 #include <sys/kstat.h>
63 #include <sys/btree.h>
64 #include <sys/zfeature.h>
66 #include <sys/zstd/zstd.h>
71 * There are three basic locks for managing spa_t structures:
73 * spa_namespace_lock (global mutex)
75 * This lock must be acquired to do any of the following:
77 * - Lookup a spa_t by name
78 * - Add or remove a spa_t from the namespace
79 * - Increase spa_refcount from non-zero
80 * - Check if spa_refcount is zero
82 * - add/remove/attach/detach devices
83 * - Held for the duration of create/destroy/import/export
85 * It does not need to handle recursion. A create or destroy may
86 * reference objects (files or zvols) in other pools, but by
87 * definition they must have an existing reference, and will never need
88 * to lookup a spa_t by name.
90 * spa_refcount (per-spa zfs_refcount_t protected by mutex)
92 * This reference count keep track of any active users of the spa_t. The
93 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
94 * the refcount is never really 'zero' - opening a pool implicitly keeps
95 * some references in the DMU. Internally we check against spa_minref, but
96 * present the image of a zero/non-zero value to consumers.
98 * spa_config_lock[] (per-spa array of rwlocks)
100 * This protects the spa_t from config changes, and must be held in
101 * the following circumstances:
103 * - RW_READER to perform I/O to the spa
104 * - RW_WRITER to change the vdev config
106 * The locking order is fairly straightforward:
108 * spa_namespace_lock -> spa_refcount
110 * The namespace lock must be acquired to increase the refcount from 0
111 * or to check if it is zero.
113 * spa_refcount -> spa_config_lock[]
115 * There must be at least one valid reference on the spa_t to acquire
118 * spa_namespace_lock -> spa_config_lock[]
120 * The namespace lock must always be taken before the config lock.
123 * The spa_namespace_lock can be acquired directly and is globally visible.
125 * The namespace is manipulated using the following functions, all of which
126 * require the spa_namespace_lock to be held.
128 * spa_lookup() Lookup a spa_t by name.
130 * spa_add() Create a new spa_t in the namespace.
132 * spa_remove() Remove a spa_t from the namespace. This also
133 * frees up any memory associated with the spa_t.
135 * spa_next() Returns the next spa_t in the system, or the
136 * first if NULL is passed.
138 * spa_evict_all() Shutdown and remove all spa_t structures in
141 * spa_guid_exists() Determine whether a pool/device guid exists.
143 * The spa_refcount is manipulated using the following functions:
145 * spa_open_ref() Adds a reference to the given spa_t. Must be
146 * called with spa_namespace_lock held if the
147 * refcount is currently zero.
149 * spa_close() Remove a reference from the spa_t. This will
150 * not free the spa_t or remove it from the
151 * namespace. No locking is required.
153 * spa_refcount_zero() Returns true if the refcount is currently
154 * zero. Must be called with spa_namespace_lock
157 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
158 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
159 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
161 * To read the configuration, it suffices to hold one of these locks as reader.
162 * To modify the configuration, you must hold all locks as writer. To modify
163 * vdev state without altering the vdev tree's topology (e.g. online/offline),
164 * you must hold SCL_STATE and SCL_ZIO as writer.
166 * We use these distinct config locks to avoid recursive lock entry.
167 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
168 * block allocations (SCL_ALLOC), which may require reading space maps
169 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
171 * The spa config locks cannot be normal rwlocks because we need the
172 * ability to hand off ownership. For example, SCL_ZIO is acquired
173 * by the issuing thread and later released by an interrupt thread.
174 * They do, however, obey the usual write-wanted semantics to prevent
175 * writer (i.e. system administrator) starvation.
177 * The lock acquisition rules are as follows:
180 * Protects changes to the vdev tree topology, such as vdev
181 * add/remove/attach/detach. Protects the dirty config list
182 * (spa_config_dirty_list) and the set of spares and l2arc devices.
185 * Protects changes to pool state and vdev state, such as vdev
186 * online/offline/fault/degrade/clear. Protects the dirty state list
187 * (spa_state_dirty_list) and global pool state (spa_state).
190 * Protects changes to metaslab groups and classes.
191 * Held as reader by metaslab_alloc() and metaslab_claim().
194 * Held by bp-level zios (those which have no io_vd upon entry)
195 * to prevent changes to the vdev tree. The bp-level zio implicitly
196 * protects all of its vdev child zios, which do not hold SCL_ZIO.
199 * Protects changes to metaslab groups and classes.
200 * Held as reader by metaslab_free(). SCL_FREE is distinct from
201 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
202 * blocks in zio_done() while another i/o that holds either
203 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
206 * Held as reader to prevent changes to the vdev tree during trivial
207 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
208 * other locks, and lower than all of them, to ensure that it's safe
209 * to acquire regardless of caller context.
211 * In addition, the following rules apply:
213 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
214 * The lock ordering is SCL_CONFIG > spa_props_lock.
216 * (b) I/O operations on leaf vdevs. For any zio operation that takes
217 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
218 * or zio_write_phys() -- the caller must ensure that the config cannot
219 * cannot change in the interim, and that the vdev cannot be reopened.
220 * SCL_STATE as reader suffices for both.
222 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
224 * spa_vdev_enter() Acquire the namespace lock and the config lock
227 * spa_vdev_exit() Release the config lock, wait for all I/O
228 * to complete, sync the updated configs to the
229 * cache, and release the namespace lock.
231 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
232 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
233 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
236 static avl_tree_t spa_namespace_avl;
237 kmutex_t spa_namespace_lock;
238 static kcondvar_t spa_namespace_cv;
239 static const int spa_max_replication_override = SPA_DVAS_PER_BP;
241 static kmutex_t spa_spare_lock;
242 static avl_tree_t spa_spare_avl;
243 static kmutex_t spa_l2cache_lock;
244 static avl_tree_t spa_l2cache_avl;
246 spa_mode_t spa_mode_global = SPA_MODE_UNINIT;
250 * Everything except dprintf, set_error, spa, and indirect_remap is on
251 * by default in debug builds.
253 int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SET_ERROR |
254 ZFS_DEBUG_INDIRECT_REMAP);
260 * zfs_recover can be set to nonzero to attempt to recover from
261 * otherwise-fatal errors, typically caused by on-disk corruption. When
262 * set, calls to zfs_panic_recover() will turn into warning messages.
263 * This should only be used as a last resort, as it typically results
264 * in leaked space, or worse.
266 int zfs_recover = B_FALSE;
269 * If destroy encounters an EIO while reading metadata (e.g. indirect
270 * blocks), space referenced by the missing metadata can not be freed.
271 * Normally this causes the background destroy to become "stalled", as
272 * it is unable to make forward progress. While in this stalled state,
273 * all remaining space to free from the error-encountering filesystem is
274 * "temporarily leaked". Set this flag to cause it to ignore the EIO,
275 * permanently leak the space from indirect blocks that can not be read,
276 * and continue to free everything else that it can.
278 * The default, "stalling" behavior is useful if the storage partially
279 * fails (i.e. some but not all i/os fail), and then later recovers. In
280 * this case, we will be able to continue pool operations while it is
281 * partially failed, and when it recovers, we can continue to free the
282 * space, with no leaks. However, note that this case is actually
285 * Typically pools either (a) fail completely (but perhaps temporarily,
286 * e.g. a top-level vdev going offline), or (b) have localized,
287 * permanent errors (e.g. disk returns the wrong data due to bit flip or
288 * firmware bug). In case (a), this setting does not matter because the
289 * pool will be suspended and the sync thread will not be able to make
290 * forward progress regardless. In case (b), because the error is
291 * permanent, the best we can do is leak the minimum amount of space,
292 * which is what setting this flag will do. Therefore, it is reasonable
293 * for this flag to normally be set, but we chose the more conservative
294 * approach of not setting it, so that there is no possibility of
295 * leaking space in the "partial temporary" failure case.
297 int zfs_free_leak_on_eio = B_FALSE;
300 * Expiration time in milliseconds. This value has two meanings. First it is
301 * used to determine when the spa_deadman() logic should fire. By default the
302 * spa_deadman() will fire if spa_sync() has not completed in 600 seconds.
303 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
304 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
305 * in one of three behaviors controlled by zfs_deadman_failmode.
307 uint64_t zfs_deadman_synctime_ms = 600000UL; /* 10 min. */
310 * This value controls the maximum amount of time zio_wait() will block for an
311 * outstanding IO. By default this is 300 seconds at which point the "hung"
312 * behavior will be applied as described for zfs_deadman_synctime_ms.
314 uint64_t zfs_deadman_ziotime_ms = 300000UL; /* 5 min. */
317 * Check time in milliseconds. This defines the frequency at which we check
320 uint64_t zfs_deadman_checktime_ms = 60000UL; /* 1 min. */
323 * By default the deadman is enabled.
325 int zfs_deadman_enabled = B_TRUE;
328 * Controls the behavior of the deadman when it detects a "hung" I/O.
329 * Valid values are zfs_deadman_failmode=<wait|continue|panic>.
331 * wait - Wait for the "hung" I/O (default)
332 * continue - Attempt to recover from a "hung" I/O
333 * panic - Panic the system
335 const char *zfs_deadman_failmode = "wait";
338 * The worst case is single-sector max-parity RAID-Z blocks, in which
339 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
340 * times the size; so just assume that. Add to this the fact that
341 * we can have up to 3 DVAs per bp, and one more factor of 2 because
342 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together,
344 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
346 uint_t spa_asize_inflation = 24;
349 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
350 * the pool to be consumed (bounded by spa_max_slop). This ensures that we
351 * don't run the pool completely out of space, due to unaccounted changes (e.g.
352 * to the MOS). It also limits the worst-case time to allocate space. If we
353 * have less than this amount of free space, most ZPL operations (e.g. write,
354 * create) will return ENOSPC. The ZIL metaslabs (spa_embedded_log_class) are
355 * also part of this 3.2% of space which can't be consumed by normal writes;
356 * the slop space "proper" (spa_get_slop_space()) is decreased by the embedded
359 * Certain operations (e.g. file removal, most administrative actions) can
360 * use half the slop space. They will only return ENOSPC if less than half
361 * the slop space is free. Typically, once the pool has less than the slop
362 * space free, the user will use these operations to free up space in the pool.
363 * These are the operations that call dsl_pool_adjustedsize() with the netfree
364 * argument set to TRUE.
366 * Operations that are almost guaranteed to free up space in the absence of
367 * a pool checkpoint can use up to three quarters of the slop space
370 * A very restricted set of operations are always permitted, regardless of
371 * the amount of free space. These are the operations that call
372 * dsl_sync_task(ZFS_SPACE_CHECK_NONE). If these operations result in a net
373 * increase in the amount of space used, it is possible to run the pool
374 * completely out of space, causing it to be permanently read-only.
376 * Note that on very small pools, the slop space will be larger than
377 * 3.2%, in an effort to have it be at least spa_min_slop (128MB),
378 * but we never allow it to be more than half the pool size.
380 * Further, on very large pools, the slop space will be smaller than
381 * 3.2%, to avoid reserving much more space than we actually need; bounded
382 * by spa_max_slop (128GB).
384 * See also the comments in zfs_space_check_t.
386 uint_t spa_slop_shift = 5;
387 static const uint64_t spa_min_slop = 128ULL * 1024 * 1024;
388 static const uint64_t spa_max_slop = 128ULL * 1024 * 1024 * 1024;
389 static const int spa_allocators = 4;
393 spa_load_failed(spa_t *spa, const char *fmt, ...)
399 (void) vsnprintf(buf, sizeof (buf), fmt, adx);
402 zfs_dbgmsg("spa_load(%s, config %s): FAILED: %s", spa->spa_name,
403 spa->spa_trust_config ? "trusted" : "untrusted", buf);
407 spa_load_note(spa_t *spa, const char *fmt, ...)
413 (void) vsnprintf(buf, sizeof (buf), fmt, adx);
416 zfs_dbgmsg("spa_load(%s, config %s): %s", spa->spa_name,
417 spa->spa_trust_config ? "trusted" : "untrusted", buf);
421 * By default dedup and user data indirects land in the special class
423 static int zfs_ddt_data_is_special = B_TRUE;
424 static int zfs_user_indirect_is_special = B_TRUE;
427 * The percentage of special class final space reserved for metadata only.
428 * Once we allocate 100 - zfs_special_class_metadata_reserve_pct we only
429 * let metadata into the class.
431 static uint_t zfs_special_class_metadata_reserve_pct = 25;
434 * ==========================================================================
436 * ==========================================================================
439 spa_config_lock_init(spa_t *spa)
441 for (int i = 0; i < SCL_LOCKS; i++) {
442 spa_config_lock_t *scl = &spa->spa_config_lock[i];
443 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
444 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
445 scl->scl_writer = NULL;
446 scl->scl_write_wanted = 0;
452 spa_config_lock_destroy(spa_t *spa)
454 for (int i = 0; i < SCL_LOCKS; i++) {
455 spa_config_lock_t *scl = &spa->spa_config_lock[i];
456 mutex_destroy(&scl->scl_lock);
457 cv_destroy(&scl->scl_cv);
458 ASSERT(scl->scl_writer == NULL);
459 ASSERT(scl->scl_write_wanted == 0);
460 ASSERT(scl->scl_count == 0);
465 spa_config_tryenter(spa_t *spa, int locks, const void *tag, krw_t rw)
467 for (int i = 0; i < SCL_LOCKS; i++) {
468 spa_config_lock_t *scl = &spa->spa_config_lock[i];
469 if (!(locks & (1 << i)))
471 mutex_enter(&scl->scl_lock);
472 if (rw == RW_READER) {
473 if (scl->scl_writer || scl->scl_write_wanted) {
474 mutex_exit(&scl->scl_lock);
475 spa_config_exit(spa, locks & ((1 << i) - 1),
480 ASSERT(scl->scl_writer != curthread);
481 if (scl->scl_count != 0) {
482 mutex_exit(&scl->scl_lock);
483 spa_config_exit(spa, locks & ((1 << i) - 1),
487 scl->scl_writer = curthread;
490 mutex_exit(&scl->scl_lock);
496 spa_config_enter(spa_t *spa, int locks, const void *tag, krw_t rw)
501 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
503 for (int i = 0; i < SCL_LOCKS; i++) {
504 spa_config_lock_t *scl = &spa->spa_config_lock[i];
505 if (scl->scl_writer == curthread)
506 wlocks_held |= (1 << i);
507 if (!(locks & (1 << i)))
509 mutex_enter(&scl->scl_lock);
510 if (rw == RW_READER) {
511 while (scl->scl_writer || scl->scl_write_wanted) {
512 cv_wait(&scl->scl_cv, &scl->scl_lock);
515 ASSERT(scl->scl_writer != curthread);
516 while (scl->scl_count != 0) {
517 scl->scl_write_wanted++;
518 cv_wait(&scl->scl_cv, &scl->scl_lock);
519 scl->scl_write_wanted--;
521 scl->scl_writer = curthread;
524 mutex_exit(&scl->scl_lock);
526 ASSERT3U(wlocks_held, <=, locks);
530 spa_config_exit(spa_t *spa, int locks, const void *tag)
533 for (int i = SCL_LOCKS - 1; i >= 0; i--) {
534 spa_config_lock_t *scl = &spa->spa_config_lock[i];
535 if (!(locks & (1 << i)))
537 mutex_enter(&scl->scl_lock);
538 ASSERT(scl->scl_count > 0);
539 if (--scl->scl_count == 0) {
540 ASSERT(scl->scl_writer == NULL ||
541 scl->scl_writer == curthread);
542 scl->scl_writer = NULL; /* OK in either case */
543 cv_broadcast(&scl->scl_cv);
545 mutex_exit(&scl->scl_lock);
550 spa_config_held(spa_t *spa, int locks, krw_t rw)
554 for (int i = 0; i < SCL_LOCKS; i++) {
555 spa_config_lock_t *scl = &spa->spa_config_lock[i];
556 if (!(locks & (1 << i)))
558 if ((rw == RW_READER && scl->scl_count != 0) ||
559 (rw == RW_WRITER && scl->scl_writer == curthread))
560 locks_held |= 1 << i;
567 * ==========================================================================
568 * SPA namespace functions
569 * ==========================================================================
573 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
574 * Returns NULL if no matching spa_t is found.
577 spa_lookup(const char *name)
579 static spa_t search; /* spa_t is large; don't allocate on stack */
584 ASSERT(MUTEX_HELD(&spa_namespace_lock));
586 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
589 * If it's a full dataset name, figure out the pool name and
592 cp = strpbrk(search.spa_name, "/@#");
596 spa = avl_find(&spa_namespace_avl, &search, &where);
602 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
603 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
604 * looking for potentially hung I/Os.
607 spa_deadman(void *arg)
611 /* Disable the deadman if the pool is suspended. */
612 if (spa_suspended(spa))
615 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
616 (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
617 (u_longlong_t)++spa->spa_deadman_calls);
618 if (zfs_deadman_enabled)
619 vdev_deadman(spa->spa_root_vdev, FTAG);
621 spa->spa_deadman_tqid = taskq_dispatch_delay(system_delay_taskq,
622 spa_deadman, spa, TQ_SLEEP, ddi_get_lbolt() +
623 MSEC_TO_TICK(zfs_deadman_checktime_ms));
627 spa_log_sm_sort_by_txg(const void *va, const void *vb)
629 const spa_log_sm_t *a = va;
630 const spa_log_sm_t *b = vb;
632 return (TREE_CMP(a->sls_txg, b->sls_txg));
636 * Create an uninitialized spa_t with the given name. Requires
637 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
638 * exist by calling spa_lookup() first.
641 spa_add(const char *name, nvlist_t *config, const char *altroot)
644 spa_config_dirent_t *dp;
646 ASSERT(MUTEX_HELD(&spa_namespace_lock));
648 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
650 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
651 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
652 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
653 mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL);
654 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
655 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
656 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
657 mutex_init(&spa->spa_cksum_tmpls_lock, NULL, MUTEX_DEFAULT, NULL);
658 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
659 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
660 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
661 mutex_init(&spa->spa_feat_stats_lock, NULL, MUTEX_DEFAULT, NULL);
662 mutex_init(&spa->spa_flushed_ms_lock, NULL, MUTEX_DEFAULT, NULL);
663 mutex_init(&spa->spa_activities_lock, NULL, MUTEX_DEFAULT, NULL);
665 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
666 cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL);
667 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
668 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
669 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
670 cv_init(&spa->spa_activities_cv, NULL, CV_DEFAULT, NULL);
671 cv_init(&spa->spa_waiters_cv, NULL, CV_DEFAULT, NULL);
673 for (int t = 0; t < TXG_SIZE; t++)
674 bplist_create(&spa->spa_free_bplist[t]);
676 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
677 spa->spa_state = POOL_STATE_UNINITIALIZED;
678 spa->spa_freeze_txg = UINT64_MAX;
679 spa->spa_final_txg = UINT64_MAX;
680 spa->spa_load_max_txg = UINT64_MAX;
682 spa->spa_proc_state = SPA_PROC_NONE;
683 spa->spa_trust_config = B_TRUE;
684 spa->spa_hostid = zone_get_hostid(NULL);
686 spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
687 spa->spa_deadman_ziotime = MSEC2NSEC(zfs_deadman_ziotime_ms);
688 spa_set_deadman_failmode(spa, zfs_deadman_failmode);
690 zfs_refcount_create(&spa->spa_refcount);
691 spa_config_lock_init(spa);
694 avl_add(&spa_namespace_avl, spa);
697 * Set the alternate root, if there is one.
700 spa->spa_root = spa_strdup(altroot);
702 spa->spa_alloc_count = spa_allocators;
703 spa->spa_allocs = kmem_zalloc(spa->spa_alloc_count *
704 sizeof (spa_alloc_t), KM_SLEEP);
705 for (int i = 0; i < spa->spa_alloc_count; i++) {
706 mutex_init(&spa->spa_allocs[i].spaa_lock, NULL, MUTEX_DEFAULT,
708 avl_create(&spa->spa_allocs[i].spaa_tree, zio_bookmark_compare,
709 sizeof (zio_t), offsetof(zio_t, io_alloc_node));
711 avl_create(&spa->spa_metaslabs_by_flushed, metaslab_sort_by_flushed,
712 sizeof (metaslab_t), offsetof(metaslab_t, ms_spa_txg_node));
713 avl_create(&spa->spa_sm_logs_by_txg, spa_log_sm_sort_by_txg,
714 sizeof (spa_log_sm_t), offsetof(spa_log_sm_t, sls_node));
715 list_create(&spa->spa_log_summary, sizeof (log_summary_entry_t),
716 offsetof(log_summary_entry_t, lse_node));
719 * Every pool starts with the default cachefile
721 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
722 offsetof(spa_config_dirent_t, scd_link));
724 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
725 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
726 list_insert_head(&spa->spa_config_list, dp);
728 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
731 if (config != NULL) {
734 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
736 VERIFY(nvlist_dup(features, &spa->spa_label_features,
740 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
743 if (spa->spa_label_features == NULL) {
744 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
748 spa->spa_min_ashift = INT_MAX;
749 spa->spa_max_ashift = 0;
750 spa->spa_min_alloc = INT_MAX;
752 /* Reset cached value */
753 spa->spa_dedup_dspace = ~0ULL;
756 * As a pool is being created, treat all features as disabled by
757 * setting SPA_FEATURE_DISABLED for all entries in the feature
760 for (int i = 0; i < SPA_FEATURES; i++) {
761 spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED;
764 list_create(&spa->spa_leaf_list, sizeof (vdev_t),
765 offsetof(vdev_t, vdev_leaf_node));
771 * Removes a spa_t from the namespace, freeing up any memory used. Requires
772 * spa_namespace_lock. This is called only after the spa_t has been closed and
776 spa_remove(spa_t *spa)
778 spa_config_dirent_t *dp;
780 ASSERT(MUTEX_HELD(&spa_namespace_lock));
781 ASSERT(spa_state(spa) == POOL_STATE_UNINITIALIZED);
782 ASSERT3U(zfs_refcount_count(&spa->spa_refcount), ==, 0);
783 ASSERT0(spa->spa_waiters);
785 nvlist_free(spa->spa_config_splitting);
787 avl_remove(&spa_namespace_avl, spa);
788 cv_broadcast(&spa_namespace_cv);
791 spa_strfree(spa->spa_root);
793 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
794 list_remove(&spa->spa_config_list, dp);
795 if (dp->scd_path != NULL)
796 spa_strfree(dp->scd_path);
797 kmem_free(dp, sizeof (spa_config_dirent_t));
800 for (int i = 0; i < spa->spa_alloc_count; i++) {
801 avl_destroy(&spa->spa_allocs[i].spaa_tree);
802 mutex_destroy(&spa->spa_allocs[i].spaa_lock);
804 kmem_free(spa->spa_allocs, spa->spa_alloc_count *
805 sizeof (spa_alloc_t));
807 avl_destroy(&spa->spa_metaslabs_by_flushed);
808 avl_destroy(&spa->spa_sm_logs_by_txg);
809 list_destroy(&spa->spa_log_summary);
810 list_destroy(&spa->spa_config_list);
811 list_destroy(&spa->spa_leaf_list);
813 nvlist_free(spa->spa_label_features);
814 nvlist_free(spa->spa_load_info);
815 nvlist_free(spa->spa_feat_stats);
816 spa_config_set(spa, NULL);
818 zfs_refcount_destroy(&spa->spa_refcount);
820 spa_stats_destroy(spa);
821 spa_config_lock_destroy(spa);
823 for (int t = 0; t < TXG_SIZE; t++)
824 bplist_destroy(&spa->spa_free_bplist[t]);
826 zio_checksum_templates_free(spa);
828 cv_destroy(&spa->spa_async_cv);
829 cv_destroy(&spa->spa_evicting_os_cv);
830 cv_destroy(&spa->spa_proc_cv);
831 cv_destroy(&spa->spa_scrub_io_cv);
832 cv_destroy(&spa->spa_suspend_cv);
833 cv_destroy(&spa->spa_activities_cv);
834 cv_destroy(&spa->spa_waiters_cv);
836 mutex_destroy(&spa->spa_flushed_ms_lock);
837 mutex_destroy(&spa->spa_async_lock);
838 mutex_destroy(&spa->spa_errlist_lock);
839 mutex_destroy(&spa->spa_errlog_lock);
840 mutex_destroy(&spa->spa_evicting_os_lock);
841 mutex_destroy(&spa->spa_history_lock);
842 mutex_destroy(&spa->spa_proc_lock);
843 mutex_destroy(&spa->spa_props_lock);
844 mutex_destroy(&spa->spa_cksum_tmpls_lock);
845 mutex_destroy(&spa->spa_scrub_lock);
846 mutex_destroy(&spa->spa_suspend_lock);
847 mutex_destroy(&spa->spa_vdev_top_lock);
848 mutex_destroy(&spa->spa_feat_stats_lock);
849 mutex_destroy(&spa->spa_activities_lock);
851 kmem_free(spa, sizeof (spa_t));
855 * Given a pool, return the next pool in the namespace, or NULL if there is
856 * none. If 'prev' is NULL, return the first pool.
859 spa_next(spa_t *prev)
861 ASSERT(MUTEX_HELD(&spa_namespace_lock));
864 return (AVL_NEXT(&spa_namespace_avl, prev));
866 return (avl_first(&spa_namespace_avl));
870 * ==========================================================================
871 * SPA refcount functions
872 * ==========================================================================
876 * Add a reference to the given spa_t. Must have at least one reference, or
877 * have the namespace lock held.
880 spa_open_ref(spa_t *spa, const void *tag)
882 ASSERT(zfs_refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
883 MUTEX_HELD(&spa_namespace_lock));
884 (void) zfs_refcount_add(&spa->spa_refcount, tag);
888 * Remove a reference to the given spa_t. Must have at least one reference, or
889 * have the namespace lock held.
892 spa_close(spa_t *spa, const void *tag)
894 ASSERT(zfs_refcount_count(&spa->spa_refcount) > spa->spa_minref ||
895 MUTEX_HELD(&spa_namespace_lock));
896 (void) zfs_refcount_remove(&spa->spa_refcount, tag);
900 * Remove a reference to the given spa_t held by a dsl dir that is
901 * being asynchronously released. Async releases occur from a taskq
902 * performing eviction of dsl datasets and dirs. The namespace lock
903 * isn't held and the hold by the object being evicted may contribute to
904 * spa_minref (e.g. dataset or directory released during pool export),
905 * so the asserts in spa_close() do not apply.
908 spa_async_close(spa_t *spa, const void *tag)
910 (void) zfs_refcount_remove(&spa->spa_refcount, tag);
914 * Check to see if the spa refcount is zero. Must be called with
915 * spa_namespace_lock held. We really compare against spa_minref, which is the
916 * number of references acquired when opening a pool
919 spa_refcount_zero(spa_t *spa)
921 ASSERT(MUTEX_HELD(&spa_namespace_lock));
923 return (zfs_refcount_count(&spa->spa_refcount) == spa->spa_minref);
927 * ==========================================================================
928 * SPA spare and l2cache tracking
929 * ==========================================================================
933 * Hot spares and cache devices are tracked using the same code below,
934 * for 'auxiliary' devices.
937 typedef struct spa_aux {
945 spa_aux_compare(const void *a, const void *b)
947 const spa_aux_t *sa = (const spa_aux_t *)a;
948 const spa_aux_t *sb = (const spa_aux_t *)b;
950 return (TREE_CMP(sa->aux_guid, sb->aux_guid));
954 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
960 search.aux_guid = vd->vdev_guid;
961 if ((aux = avl_find(avl, &search, &where)) != NULL) {
964 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
965 aux->aux_guid = vd->vdev_guid;
967 avl_insert(avl, aux, where);
972 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
978 search.aux_guid = vd->vdev_guid;
979 aux = avl_find(avl, &search, &where);
983 if (--aux->aux_count == 0) {
984 avl_remove(avl, aux);
985 kmem_free(aux, sizeof (spa_aux_t));
986 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
987 aux->aux_pool = 0ULL;
992 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
994 spa_aux_t search, *found;
996 search.aux_guid = guid;
997 found = avl_find(avl, &search, NULL);
1001 *pool = found->aux_pool;
1008 *refcnt = found->aux_count;
1013 return (found != NULL);
1017 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
1019 spa_aux_t search, *found;
1022 search.aux_guid = vd->vdev_guid;
1023 found = avl_find(avl, &search, &where);
1024 ASSERT(found != NULL);
1025 ASSERT(found->aux_pool == 0ULL);
1027 found->aux_pool = spa_guid(vd->vdev_spa);
1031 * Spares are tracked globally due to the following constraints:
1033 * - A spare may be part of multiple pools.
1034 * - A spare may be added to a pool even if it's actively in use within
1036 * - A spare in use in any pool can only be the source of a replacement if
1037 * the target is a spare in the same pool.
1039 * We keep track of all spares on the system through the use of a reference
1040 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
1041 * spare, then we bump the reference count in the AVL tree. In addition, we set
1042 * the 'vdev_isspare' member to indicate that the device is a spare (active or
1043 * inactive). When a spare is made active (used to replace a device in the
1044 * pool), we also keep track of which pool its been made a part of.
1046 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
1047 * called under the spa_namespace lock as part of vdev reconfiguration. The
1048 * separate spare lock exists for the status query path, which does not need to
1049 * be completely consistent with respect to other vdev configuration changes.
1053 spa_spare_compare(const void *a, const void *b)
1055 return (spa_aux_compare(a, b));
1059 spa_spare_add(vdev_t *vd)
1061 mutex_enter(&spa_spare_lock);
1062 ASSERT(!vd->vdev_isspare);
1063 spa_aux_add(vd, &spa_spare_avl);
1064 vd->vdev_isspare = B_TRUE;
1065 mutex_exit(&spa_spare_lock);
1069 spa_spare_remove(vdev_t *vd)
1071 mutex_enter(&spa_spare_lock);
1072 ASSERT(vd->vdev_isspare);
1073 spa_aux_remove(vd, &spa_spare_avl);
1074 vd->vdev_isspare = B_FALSE;
1075 mutex_exit(&spa_spare_lock);
1079 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
1083 mutex_enter(&spa_spare_lock);
1084 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
1085 mutex_exit(&spa_spare_lock);
1091 spa_spare_activate(vdev_t *vd)
1093 mutex_enter(&spa_spare_lock);
1094 ASSERT(vd->vdev_isspare);
1095 spa_aux_activate(vd, &spa_spare_avl);
1096 mutex_exit(&spa_spare_lock);
1100 * Level 2 ARC devices are tracked globally for the same reasons as spares.
1101 * Cache devices currently only support one pool per cache device, and so
1102 * for these devices the aux reference count is currently unused beyond 1.
1106 spa_l2cache_compare(const void *a, const void *b)
1108 return (spa_aux_compare(a, b));
1112 spa_l2cache_add(vdev_t *vd)
1114 mutex_enter(&spa_l2cache_lock);
1115 ASSERT(!vd->vdev_isl2cache);
1116 spa_aux_add(vd, &spa_l2cache_avl);
1117 vd->vdev_isl2cache = B_TRUE;
1118 mutex_exit(&spa_l2cache_lock);
1122 spa_l2cache_remove(vdev_t *vd)
1124 mutex_enter(&spa_l2cache_lock);
1125 ASSERT(vd->vdev_isl2cache);
1126 spa_aux_remove(vd, &spa_l2cache_avl);
1127 vd->vdev_isl2cache = B_FALSE;
1128 mutex_exit(&spa_l2cache_lock);
1132 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
1136 mutex_enter(&spa_l2cache_lock);
1137 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
1138 mutex_exit(&spa_l2cache_lock);
1144 spa_l2cache_activate(vdev_t *vd)
1146 mutex_enter(&spa_l2cache_lock);
1147 ASSERT(vd->vdev_isl2cache);
1148 spa_aux_activate(vd, &spa_l2cache_avl);
1149 mutex_exit(&spa_l2cache_lock);
1153 * ==========================================================================
1155 * ==========================================================================
1159 * Lock the given spa_t for the purpose of adding or removing a vdev.
1160 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1161 * It returns the next transaction group for the spa_t.
1164 spa_vdev_enter(spa_t *spa)
1166 mutex_enter(&spa->spa_vdev_top_lock);
1167 mutex_enter(&spa_namespace_lock);
1169 vdev_autotrim_stop_all(spa);
1171 return (spa_vdev_config_enter(spa));
1175 * The same as spa_vdev_enter() above but additionally takes the guid of
1176 * the vdev being detached. When there is a rebuild in process it will be
1177 * suspended while the vdev tree is modified then resumed by spa_vdev_exit().
1178 * The rebuild is canceled if only a single child remains after the detach.
1181 spa_vdev_detach_enter(spa_t *spa, uint64_t guid)
1183 mutex_enter(&spa->spa_vdev_top_lock);
1184 mutex_enter(&spa_namespace_lock);
1186 vdev_autotrim_stop_all(spa);
1189 vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE);
1191 vdev_rebuild_stop_wait(vd->vdev_top);
1195 return (spa_vdev_config_enter(spa));
1199 * Internal implementation for spa_vdev_enter(). Used when a vdev
1200 * operation requires multiple syncs (i.e. removing a device) while
1201 * keeping the spa_namespace_lock held.
1204 spa_vdev_config_enter(spa_t *spa)
1206 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1208 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1210 return (spa_last_synced_txg(spa) + 1);
1214 * Used in combination with spa_vdev_config_enter() to allow the syncing
1215 * of multiple transactions without releasing the spa_namespace_lock.
1218 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error,
1221 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1223 int config_changed = B_FALSE;
1225 ASSERT(txg > spa_last_synced_txg(spa));
1227 spa->spa_pending_vdev = NULL;
1230 * Reassess the DTLs.
1232 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE, B_FALSE);
1234 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
1235 config_changed = B_TRUE;
1236 spa->spa_config_generation++;
1240 * Verify the metaslab classes.
1242 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
1243 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
1244 ASSERT(metaslab_class_validate(spa_embedded_log_class(spa)) == 0);
1245 ASSERT(metaslab_class_validate(spa_special_class(spa)) == 0);
1246 ASSERT(metaslab_class_validate(spa_dedup_class(spa)) == 0);
1248 spa_config_exit(spa, SCL_ALL, spa);
1251 * Panic the system if the specified tag requires it. This
1252 * is useful for ensuring that configurations are updated
1255 if (zio_injection_enabled)
1256 zio_handle_panic_injection(spa, tag, 0);
1259 * Note: this txg_wait_synced() is important because it ensures
1260 * that there won't be more than one config change per txg.
1261 * This allows us to use the txg as the generation number.
1264 txg_wait_synced(spa->spa_dsl_pool, txg);
1267 ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
1268 if (vd->vdev_ops->vdev_op_leaf) {
1269 mutex_enter(&vd->vdev_initialize_lock);
1270 vdev_initialize_stop(vd, VDEV_INITIALIZE_CANCELED,
1272 mutex_exit(&vd->vdev_initialize_lock);
1274 mutex_enter(&vd->vdev_trim_lock);
1275 vdev_trim_stop(vd, VDEV_TRIM_CANCELED, NULL);
1276 mutex_exit(&vd->vdev_trim_lock);
1280 * The vdev may be both a leaf and top-level device.
1282 vdev_autotrim_stop_wait(vd);
1284 spa_config_enter(spa, SCL_STATE_ALL, spa, RW_WRITER);
1286 spa_config_exit(spa, SCL_STATE_ALL, spa);
1290 * If the config changed, update the config cache.
1293 spa_write_cachefile(spa, B_FALSE, B_TRUE, B_FALSE);
1297 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1298 * locking of spa_vdev_enter(), we also want make sure the transactions have
1299 * synced to disk, and then update the global configuration cache with the new
1303 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1305 vdev_autotrim_restart(spa);
1306 vdev_rebuild_restart(spa);
1308 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1309 mutex_exit(&spa_namespace_lock);
1310 mutex_exit(&spa->spa_vdev_top_lock);
1316 * Lock the given spa_t for the purpose of changing vdev state.
1319 spa_vdev_state_enter(spa_t *spa, int oplocks)
1321 int locks = SCL_STATE_ALL | oplocks;
1324 * Root pools may need to read of the underlying devfs filesystem
1325 * when opening up a vdev. Unfortunately if we're holding the
1326 * SCL_ZIO lock it will result in a deadlock when we try to issue
1327 * the read from the root filesystem. Instead we "prefetch"
1328 * the associated vnodes that we need prior to opening the
1329 * underlying devices and cache them so that we can prevent
1330 * any I/O when we are doing the actual open.
1332 if (spa_is_root(spa)) {
1333 int low = locks & ~(SCL_ZIO - 1);
1334 int high = locks & ~low;
1336 spa_config_enter(spa, high, spa, RW_WRITER);
1337 vdev_hold(spa->spa_root_vdev);
1338 spa_config_enter(spa, low, spa, RW_WRITER);
1340 spa_config_enter(spa, locks, spa, RW_WRITER);
1342 spa->spa_vdev_locks = locks;
1346 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1348 boolean_t config_changed = B_FALSE;
1351 if (vd == NULL || vd == spa->spa_root_vdev) {
1352 vdev_top = spa->spa_root_vdev;
1354 vdev_top = vd->vdev_top;
1357 if (vd != NULL || error == 0)
1358 vdev_dtl_reassess(vdev_top, 0, 0, B_FALSE, B_FALSE);
1361 if (vd != spa->spa_root_vdev)
1362 vdev_state_dirty(vdev_top);
1364 config_changed = B_TRUE;
1365 spa->spa_config_generation++;
1368 if (spa_is_root(spa))
1369 vdev_rele(spa->spa_root_vdev);
1371 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1372 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1375 * If anything changed, wait for it to sync. This ensures that,
1376 * from the system administrator's perspective, zpool(8) commands
1377 * are synchronous. This is important for things like zpool offline:
1378 * when the command completes, you expect no further I/O from ZFS.
1381 txg_wait_synced(spa->spa_dsl_pool, 0);
1384 * If the config changed, update the config cache.
1386 if (config_changed) {
1387 mutex_enter(&spa_namespace_lock);
1388 spa_write_cachefile(spa, B_FALSE, B_TRUE, B_FALSE);
1389 mutex_exit(&spa_namespace_lock);
1396 * ==========================================================================
1397 * Miscellaneous functions
1398 * ==========================================================================
1402 spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx)
1404 if (!nvlist_exists(spa->spa_label_features, feature)) {
1405 fnvlist_add_boolean(spa->spa_label_features, feature);
1407 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1408 * dirty the vdev config because lock SCL_CONFIG is not held.
1409 * Thankfully, in this case we don't need to dirty the config
1410 * because it will be written out anyway when we finish
1411 * creating the pool.
1413 if (tx->tx_txg != TXG_INITIAL)
1414 vdev_config_dirty(spa->spa_root_vdev);
1419 spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1421 if (nvlist_remove_all(spa->spa_label_features, feature) == 0)
1422 vdev_config_dirty(spa->spa_root_vdev);
1426 * Return the spa_t associated with given pool_guid, if it exists. If
1427 * device_guid is non-zero, determine whether the pool exists *and* contains
1428 * a device with the specified device_guid.
1431 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1434 avl_tree_t *t = &spa_namespace_avl;
1436 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1438 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1439 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1441 if (spa->spa_root_vdev == NULL)
1443 if (spa_guid(spa) == pool_guid) {
1444 if (device_guid == 0)
1447 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1448 device_guid) != NULL)
1452 * Check any devices we may be in the process of adding.
1454 if (spa->spa_pending_vdev) {
1455 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1456 device_guid) != NULL)
1466 * Determine whether a pool with the given pool_guid exists.
1469 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1471 return (spa_by_guid(pool_guid, device_guid) != NULL);
1475 spa_strdup(const char *s)
1481 new = kmem_alloc(len + 1, KM_SLEEP);
1482 memcpy(new, s, len + 1);
1488 spa_strfree(char *s)
1490 kmem_free(s, strlen(s) + 1);
1494 spa_generate_guid(spa_t *spa)
1500 (void) random_get_pseudo_bytes((void *)&guid,
1502 } while (guid == 0 || spa_guid_exists(spa_guid(spa), guid));
1505 (void) random_get_pseudo_bytes((void *)&guid,
1507 } while (guid == 0 || spa_guid_exists(guid, 0));
1514 snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp)
1517 const char *checksum = NULL;
1518 const char *compress = NULL;
1521 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1522 dmu_object_byteswap_t bswap =
1523 DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1524 (void) snprintf(type, sizeof (type), "bswap %s %s",
1525 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1526 "metadata" : "data",
1527 dmu_ot_byteswap[bswap].ob_name);
1529 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1532 if (!BP_IS_EMBEDDED(bp)) {
1534 zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1536 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1539 SNPRINTF_BLKPTR(kmem_scnprintf, ' ', buf, buflen, bp, type, checksum,
1544 spa_freeze(spa_t *spa)
1546 uint64_t freeze_txg = 0;
1548 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1549 if (spa->spa_freeze_txg == UINT64_MAX) {
1550 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1551 spa->spa_freeze_txg = freeze_txg;
1553 spa_config_exit(spa, SCL_ALL, FTAG);
1554 if (freeze_txg != 0)
1555 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1559 zfs_panic_recover(const char *fmt, ...)
1564 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1569 * This is a stripped-down version of strtoull, suitable only for converting
1570 * lowercase hexadecimal numbers that don't overflow.
1573 zfs_strtonum(const char *str, char **nptr)
1579 while ((c = *str) != '\0') {
1580 if (c >= '0' && c <= '9')
1582 else if (c >= 'a' && c <= 'f')
1583 digit = 10 + c - 'a';
1594 *nptr = (char *)str;
1600 spa_activate_allocation_classes(spa_t *spa, dmu_tx_t *tx)
1603 * We bump the feature refcount for each special vdev added to the pool
1605 ASSERT(spa_feature_is_enabled(spa, SPA_FEATURE_ALLOCATION_CLASSES));
1606 spa_feature_incr(spa, SPA_FEATURE_ALLOCATION_CLASSES, tx);
1610 * ==========================================================================
1611 * Accessor functions
1612 * ==========================================================================
1616 spa_shutting_down(spa_t *spa)
1618 return (spa->spa_async_suspended);
1622 spa_get_dsl(spa_t *spa)
1624 return (spa->spa_dsl_pool);
1628 spa_is_initializing(spa_t *spa)
1630 return (spa->spa_is_initializing);
1634 spa_indirect_vdevs_loaded(spa_t *spa)
1636 return (spa->spa_indirect_vdevs_loaded);
1640 spa_get_rootblkptr(spa_t *spa)
1642 return (&spa->spa_ubsync.ub_rootbp);
1646 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1648 spa->spa_uberblock.ub_rootbp = *bp;
1652 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1654 if (spa->spa_root == NULL)
1657 (void) strlcpy(buf, spa->spa_root, buflen);
1661 spa_sync_pass(spa_t *spa)
1663 return (spa->spa_sync_pass);
1667 spa_name(spa_t *spa)
1669 return (spa->spa_name);
1673 spa_guid(spa_t *spa)
1675 dsl_pool_t *dp = spa_get_dsl(spa);
1679 * If we fail to parse the config during spa_load(), we can go through
1680 * the error path (which posts an ereport) and end up here with no root
1681 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1684 if (spa->spa_root_vdev == NULL)
1685 return (spa->spa_config_guid);
1687 guid = spa->spa_last_synced_guid != 0 ?
1688 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1691 * Return the most recently synced out guid unless we're
1692 * in syncing context.
1694 if (dp && dsl_pool_sync_context(dp))
1695 return (spa->spa_root_vdev->vdev_guid);
1701 spa_load_guid(spa_t *spa)
1704 * This is a GUID that exists solely as a reference for the
1705 * purposes of the arc. It is generated at load time, and
1706 * is never written to persistent storage.
1708 return (spa->spa_load_guid);
1712 spa_last_synced_txg(spa_t *spa)
1714 return (spa->spa_ubsync.ub_txg);
1718 spa_first_txg(spa_t *spa)
1720 return (spa->spa_first_txg);
1724 spa_syncing_txg(spa_t *spa)
1726 return (spa->spa_syncing_txg);
1730 * Return the last txg where data can be dirtied. The final txgs
1731 * will be used to just clear out any deferred frees that remain.
1734 spa_final_dirty_txg(spa_t *spa)
1736 return (spa->spa_final_txg - TXG_DEFER_SIZE);
1740 spa_state(spa_t *spa)
1742 return (spa->spa_state);
1746 spa_load_state(spa_t *spa)
1748 return (spa->spa_load_state);
1752 spa_freeze_txg(spa_t *spa)
1754 return (spa->spa_freeze_txg);
1758 * Return the inflated asize for a logical write in bytes. This is used by the
1759 * DMU to calculate the space a logical write will require on disk.
1760 * If lsize is smaller than the largest physical block size allocatable on this
1761 * pool we use its value instead, since the write will end up using the whole
1765 spa_get_worst_case_asize(spa_t *spa, uint64_t lsize)
1768 return (0); /* No inflation needed */
1769 return (MAX(lsize, 1 << spa->spa_max_ashift) * spa_asize_inflation);
1773 * Return the amount of slop space in bytes. It is typically 1/32 of the pool
1774 * (3.2%), minus the embedded log space. On very small pools, it may be
1775 * slightly larger than this. On very large pools, it will be capped to
1776 * the value of spa_max_slop. The embedded log space is not included in
1777 * spa_dspace. By subtracting it, the usable space (per "zfs list") is a
1778 * constant 97% of the total space, regardless of metaslab size (assuming the
1779 * default spa_slop_shift=5 and a non-tiny pool).
1781 * See the comment above spa_slop_shift for more details.
1784 spa_get_slop_space(spa_t *spa)
1790 * Make sure spa_dedup_dspace has been set.
1792 if (spa->spa_dedup_dspace == ~0ULL)
1793 spa_update_dspace(spa);
1796 * spa_get_dspace() includes the space only logically "used" by
1797 * deduplicated data, so since it's not useful to reserve more
1798 * space with more deduplicated data, we subtract that out here.
1800 space = spa_get_dspace(spa) - spa->spa_dedup_dspace;
1801 slop = MIN(space >> spa_slop_shift, spa_max_slop);
1804 * Subtract the embedded log space, but no more than half the (3.2%)
1805 * unusable space. Note, the "no more than half" is only relevant if
1806 * zfs_embedded_slog_min_ms >> spa_slop_shift < 2, which is not true by
1809 uint64_t embedded_log =
1810 metaslab_class_get_dspace(spa_embedded_log_class(spa));
1811 slop -= MIN(embedded_log, slop >> 1);
1814 * Slop space should be at least spa_min_slop, but no more than half
1817 slop = MAX(slop, MIN(space >> 1, spa_min_slop));
1822 spa_get_dspace(spa_t *spa)
1824 return (spa->spa_dspace);
1828 spa_get_checkpoint_space(spa_t *spa)
1830 return (spa->spa_checkpoint_info.sci_dspace);
1834 spa_update_dspace(spa_t *spa)
1836 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1837 ddt_get_dedup_dspace(spa);
1838 if (spa->spa_nonallocating_dspace > 0) {
1840 * Subtract the space provided by all non-allocating vdevs that
1841 * contribute to dspace. If a file is overwritten, its old
1842 * blocks are freed and new blocks are allocated. If there are
1843 * no snapshots of the file, the available space should remain
1844 * the same. The old blocks could be freed from the
1845 * non-allocating vdev, but the new blocks must be allocated on
1846 * other (allocating) vdevs. By reserving the entire size of
1847 * the non-allocating vdevs (including allocated space), we
1848 * ensure that there will be enough space on the allocating
1849 * vdevs for this file overwrite to succeed.
1851 * Note that the DMU/DSL doesn't actually know or care
1852 * how much space is allocated (it does its own tracking
1853 * of how much space has been logically used). So it
1854 * doesn't matter that the data we are moving may be
1855 * allocated twice (on the old device and the new device).
1857 ASSERT3U(spa->spa_dspace, >=, spa->spa_nonallocating_dspace);
1858 spa->spa_dspace -= spa->spa_nonallocating_dspace;
1863 * Return the failure mode that has been set to this pool. The default
1864 * behavior will be to block all I/Os when a complete failure occurs.
1867 spa_get_failmode(spa_t *spa)
1869 return (spa->spa_failmode);
1873 spa_suspended(spa_t *spa)
1875 return (spa->spa_suspended != ZIO_SUSPEND_NONE);
1879 spa_version(spa_t *spa)
1881 return (spa->spa_ubsync.ub_version);
1885 spa_deflate(spa_t *spa)
1887 return (spa->spa_deflate);
1891 spa_normal_class(spa_t *spa)
1893 return (spa->spa_normal_class);
1897 spa_log_class(spa_t *spa)
1899 return (spa->spa_log_class);
1903 spa_embedded_log_class(spa_t *spa)
1905 return (spa->spa_embedded_log_class);
1909 spa_special_class(spa_t *spa)
1911 return (spa->spa_special_class);
1915 spa_dedup_class(spa_t *spa)
1917 return (spa->spa_dedup_class);
1921 * Locate an appropriate allocation class
1924 spa_preferred_class(spa_t *spa, uint64_t size, dmu_object_type_t objtype,
1925 uint_t level, uint_t special_smallblk)
1928 * ZIL allocations determine their class in zio_alloc_zil().
1930 ASSERT(objtype != DMU_OT_INTENT_LOG);
1932 boolean_t has_special_class = spa->spa_special_class->mc_groups != 0;
1934 if (DMU_OT_IS_DDT(objtype)) {
1935 if (spa->spa_dedup_class->mc_groups != 0)
1936 return (spa_dedup_class(spa));
1937 else if (has_special_class && zfs_ddt_data_is_special)
1938 return (spa_special_class(spa));
1940 return (spa_normal_class(spa));
1943 /* Indirect blocks for user data can land in special if allowed */
1944 if (level > 0 && (DMU_OT_IS_FILE(objtype) || objtype == DMU_OT_ZVOL)) {
1945 if (has_special_class && zfs_user_indirect_is_special)
1946 return (spa_special_class(spa));
1948 return (spa_normal_class(spa));
1951 if (DMU_OT_IS_METADATA(objtype) || level > 0) {
1952 if (has_special_class)
1953 return (spa_special_class(spa));
1955 return (spa_normal_class(spa));
1959 * Allow small file blocks in special class in some cases (like
1960 * for the dRAID vdev feature). But always leave a reserve of
1961 * zfs_special_class_metadata_reserve_pct exclusively for metadata.
1963 if (DMU_OT_IS_FILE(objtype) &&
1964 has_special_class && size <= special_smallblk) {
1965 metaslab_class_t *special = spa_special_class(spa);
1966 uint64_t alloc = metaslab_class_get_alloc(special);
1967 uint64_t space = metaslab_class_get_space(special);
1969 (space * (100 - zfs_special_class_metadata_reserve_pct))
1976 return (spa_normal_class(spa));
1980 spa_evicting_os_register(spa_t *spa, objset_t *os)
1982 mutex_enter(&spa->spa_evicting_os_lock);
1983 list_insert_head(&spa->spa_evicting_os_list, os);
1984 mutex_exit(&spa->spa_evicting_os_lock);
1988 spa_evicting_os_deregister(spa_t *spa, objset_t *os)
1990 mutex_enter(&spa->spa_evicting_os_lock);
1991 list_remove(&spa->spa_evicting_os_list, os);
1992 cv_broadcast(&spa->spa_evicting_os_cv);
1993 mutex_exit(&spa->spa_evicting_os_lock);
1997 spa_evicting_os_wait(spa_t *spa)
1999 mutex_enter(&spa->spa_evicting_os_lock);
2000 while (!list_is_empty(&spa->spa_evicting_os_list))
2001 cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock);
2002 mutex_exit(&spa->spa_evicting_os_lock);
2004 dmu_buf_user_evict_wait();
2008 spa_max_replication(spa_t *spa)
2011 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
2012 * handle BPs with more than one DVA allocated. Set our max
2013 * replication level accordingly.
2015 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
2017 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
2021 spa_prev_software_version(spa_t *spa)
2023 return (spa->spa_prev_software_version);
2027 spa_deadman_synctime(spa_t *spa)
2029 return (spa->spa_deadman_synctime);
2033 spa_get_autotrim(spa_t *spa)
2035 return (spa->spa_autotrim);
2039 spa_deadman_ziotime(spa_t *spa)
2041 return (spa->spa_deadman_ziotime);
2045 spa_get_deadman_failmode(spa_t *spa)
2047 return (spa->spa_deadman_failmode);
2051 spa_set_deadman_failmode(spa_t *spa, const char *failmode)
2053 if (strcmp(failmode, "wait") == 0)
2054 spa->spa_deadman_failmode = ZIO_FAILURE_MODE_WAIT;
2055 else if (strcmp(failmode, "continue") == 0)
2056 spa->spa_deadman_failmode = ZIO_FAILURE_MODE_CONTINUE;
2057 else if (strcmp(failmode, "panic") == 0)
2058 spa->spa_deadman_failmode = ZIO_FAILURE_MODE_PANIC;
2060 spa->spa_deadman_failmode = ZIO_FAILURE_MODE_WAIT;
2064 spa_set_deadman_ziotime(hrtime_t ns)
2068 if (spa_mode_global != SPA_MODE_UNINIT) {
2069 mutex_enter(&spa_namespace_lock);
2070 while ((spa = spa_next(spa)) != NULL)
2071 spa->spa_deadman_ziotime = ns;
2072 mutex_exit(&spa_namespace_lock);
2077 spa_set_deadman_synctime(hrtime_t ns)
2081 if (spa_mode_global != SPA_MODE_UNINIT) {
2082 mutex_enter(&spa_namespace_lock);
2083 while ((spa = spa_next(spa)) != NULL)
2084 spa->spa_deadman_synctime = ns;
2085 mutex_exit(&spa_namespace_lock);
2090 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
2092 uint64_t asize = DVA_GET_ASIZE(dva);
2093 uint64_t dsize = asize;
2095 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2097 if (asize != 0 && spa->spa_deflate) {
2098 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
2100 dsize = (asize >> SPA_MINBLOCKSHIFT) *
2101 vd->vdev_deflate_ratio;
2108 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
2112 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
2113 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
2119 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
2123 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2125 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
2126 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
2128 spa_config_exit(spa, SCL_VDEV, FTAG);
2134 spa_dirty_data(spa_t *spa)
2136 return (spa->spa_dsl_pool->dp_dirty_total);
2140 * ==========================================================================
2141 * SPA Import Progress Routines
2142 * ==========================================================================
2145 typedef struct spa_import_progress {
2146 uint64_t pool_guid; /* unique id for updates */
2148 spa_load_state_t spa_load_state;
2149 uint64_t mmp_sec_remaining; /* MMP activity check */
2150 uint64_t spa_load_max_txg; /* rewind txg */
2151 procfs_list_node_t smh_node;
2152 } spa_import_progress_t;
2154 spa_history_list_t *spa_import_progress_list = NULL;
2157 spa_import_progress_show_header(struct seq_file *f)
2159 seq_printf(f, "%-20s %-14s %-14s %-12s %s\n", "pool_guid",
2160 "load_state", "multihost_secs", "max_txg",
2166 spa_import_progress_show(struct seq_file *f, void *data)
2168 spa_import_progress_t *sip = (spa_import_progress_t *)data;
2170 seq_printf(f, "%-20llu %-14llu %-14llu %-12llu %s\n",
2171 (u_longlong_t)sip->pool_guid, (u_longlong_t)sip->spa_load_state,
2172 (u_longlong_t)sip->mmp_sec_remaining,
2173 (u_longlong_t)sip->spa_load_max_txg,
2174 (sip->pool_name ? sip->pool_name : "-"));
2179 /* Remove oldest elements from list until there are no more than 'size' left */
2181 spa_import_progress_truncate(spa_history_list_t *shl, unsigned int size)
2183 spa_import_progress_t *sip;
2184 while (shl->size > size) {
2185 sip = list_remove_head(&shl->procfs_list.pl_list);
2187 spa_strfree(sip->pool_name);
2188 kmem_free(sip, sizeof (spa_import_progress_t));
2192 IMPLY(size == 0, list_is_empty(&shl->procfs_list.pl_list));
2196 spa_import_progress_init(void)
2198 spa_import_progress_list = kmem_zalloc(sizeof (spa_history_list_t),
2201 spa_import_progress_list->size = 0;
2203 spa_import_progress_list->procfs_list.pl_private =
2204 spa_import_progress_list;
2206 procfs_list_install("zfs",
2210 &spa_import_progress_list->procfs_list,
2211 spa_import_progress_show,
2212 spa_import_progress_show_header,
2214 offsetof(spa_import_progress_t, smh_node));
2218 spa_import_progress_destroy(void)
2220 spa_history_list_t *shl = spa_import_progress_list;
2221 procfs_list_uninstall(&shl->procfs_list);
2222 spa_import_progress_truncate(shl, 0);
2223 procfs_list_destroy(&shl->procfs_list);
2224 kmem_free(shl, sizeof (spa_history_list_t));
2228 spa_import_progress_set_state(uint64_t pool_guid,
2229 spa_load_state_t load_state)
2231 spa_history_list_t *shl = spa_import_progress_list;
2232 spa_import_progress_t *sip;
2238 mutex_enter(&shl->procfs_list.pl_lock);
2239 for (sip = list_tail(&shl->procfs_list.pl_list); sip != NULL;
2240 sip = list_prev(&shl->procfs_list.pl_list, sip)) {
2241 if (sip->pool_guid == pool_guid) {
2242 sip->spa_load_state = load_state;
2247 mutex_exit(&shl->procfs_list.pl_lock);
2253 spa_import_progress_set_max_txg(uint64_t pool_guid, uint64_t load_max_txg)
2255 spa_history_list_t *shl = spa_import_progress_list;
2256 spa_import_progress_t *sip;
2262 mutex_enter(&shl->procfs_list.pl_lock);
2263 for (sip = list_tail(&shl->procfs_list.pl_list); sip != NULL;
2264 sip = list_prev(&shl->procfs_list.pl_list, sip)) {
2265 if (sip->pool_guid == pool_guid) {
2266 sip->spa_load_max_txg = load_max_txg;
2271 mutex_exit(&shl->procfs_list.pl_lock);
2277 spa_import_progress_set_mmp_check(uint64_t pool_guid,
2278 uint64_t mmp_sec_remaining)
2280 spa_history_list_t *shl = spa_import_progress_list;
2281 spa_import_progress_t *sip;
2287 mutex_enter(&shl->procfs_list.pl_lock);
2288 for (sip = list_tail(&shl->procfs_list.pl_list); sip != NULL;
2289 sip = list_prev(&shl->procfs_list.pl_list, sip)) {
2290 if (sip->pool_guid == pool_guid) {
2291 sip->mmp_sec_remaining = mmp_sec_remaining;
2296 mutex_exit(&shl->procfs_list.pl_lock);
2302 * A new import is in progress, add an entry.
2305 spa_import_progress_add(spa_t *spa)
2307 spa_history_list_t *shl = spa_import_progress_list;
2308 spa_import_progress_t *sip;
2309 char *poolname = NULL;
2311 sip = kmem_zalloc(sizeof (spa_import_progress_t), KM_SLEEP);
2312 sip->pool_guid = spa_guid(spa);
2314 (void) nvlist_lookup_string(spa->spa_config, ZPOOL_CONFIG_POOL_NAME,
2316 if (poolname == NULL)
2317 poolname = spa_name(spa);
2318 sip->pool_name = spa_strdup(poolname);
2319 sip->spa_load_state = spa_load_state(spa);
2321 mutex_enter(&shl->procfs_list.pl_lock);
2322 procfs_list_add(&shl->procfs_list, sip);
2324 mutex_exit(&shl->procfs_list.pl_lock);
2328 spa_import_progress_remove(uint64_t pool_guid)
2330 spa_history_list_t *shl = spa_import_progress_list;
2331 spa_import_progress_t *sip;
2333 mutex_enter(&shl->procfs_list.pl_lock);
2334 for (sip = list_tail(&shl->procfs_list.pl_list); sip != NULL;
2335 sip = list_prev(&shl->procfs_list.pl_list, sip)) {
2336 if (sip->pool_guid == pool_guid) {
2338 spa_strfree(sip->pool_name);
2339 list_remove(&shl->procfs_list.pl_list, sip);
2341 kmem_free(sip, sizeof (spa_import_progress_t));
2345 mutex_exit(&shl->procfs_list.pl_lock);
2349 * ==========================================================================
2350 * Initialization and Termination
2351 * ==========================================================================
2355 spa_name_compare(const void *a1, const void *a2)
2357 const spa_t *s1 = a1;
2358 const spa_t *s2 = a2;
2361 s = strcmp(s1->spa_name, s2->spa_name);
2363 return (TREE_ISIGN(s));
2373 spa_init(spa_mode_t mode)
2375 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
2376 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
2377 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
2378 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
2380 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
2381 offsetof(spa_t, spa_avl));
2383 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
2384 offsetof(spa_aux_t, aux_avl));
2386 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
2387 offsetof(spa_aux_t, aux_avl));
2389 spa_mode_global = mode;
2392 if (spa_mode_global != SPA_MODE_READ && dprintf_find_string("watch")) {
2393 struct sigaction sa;
2395 sa.sa_flags = SA_SIGINFO;
2396 sigemptyset(&sa.sa_mask);
2397 sa.sa_sigaction = arc_buf_sigsegv;
2399 if (sigaction(SIGSEGV, &sa, NULL) == -1) {
2400 perror("could not enable watchpoints: "
2401 "sigaction(SIGSEGV, ...) = ");
2409 zfs_refcount_init();
2412 metaslab_stat_init();
2417 vdev_cache_stat_init();
2418 vdev_mirror_stat_init();
2419 vdev_raidz_math_init();
2424 zpool_feature_init();
2430 spa_import_progress_init();
2441 vdev_cache_stat_fini();
2442 vdev_mirror_stat_fini();
2443 vdev_raidz_math_fini();
2449 metaslab_stat_fini();
2452 zfs_refcount_fini();
2456 spa_import_progress_destroy();
2458 avl_destroy(&spa_namespace_avl);
2459 avl_destroy(&spa_spare_avl);
2460 avl_destroy(&spa_l2cache_avl);
2462 cv_destroy(&spa_namespace_cv);
2463 mutex_destroy(&spa_namespace_lock);
2464 mutex_destroy(&spa_spare_lock);
2465 mutex_destroy(&spa_l2cache_lock);
2469 * Return whether this pool has a dedicated slog device. No locking needed.
2470 * It's not a problem if the wrong answer is returned as it's only for
2471 * performance and not correctness.
2474 spa_has_slogs(spa_t *spa)
2476 return (spa->spa_log_class->mc_groups != 0);
2480 spa_get_log_state(spa_t *spa)
2482 return (spa->spa_log_state);
2486 spa_set_log_state(spa_t *spa, spa_log_state_t state)
2488 spa->spa_log_state = state;
2492 spa_is_root(spa_t *spa)
2494 return (spa->spa_is_root);
2498 spa_writeable(spa_t *spa)
2500 return (!!(spa->spa_mode & SPA_MODE_WRITE) && spa->spa_trust_config);
2504 * Returns true if there is a pending sync task in any of the current
2505 * syncing txg, the current quiescing txg, or the current open txg.
2508 spa_has_pending_synctask(spa_t *spa)
2510 return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks) ||
2511 !txg_all_lists_empty(&spa->spa_dsl_pool->dp_early_sync_tasks));
2515 spa_mode(spa_t *spa)
2517 return (spa->spa_mode);
2521 spa_bootfs(spa_t *spa)
2523 return (spa->spa_bootfs);
2527 spa_delegation(spa_t *spa)
2529 return (spa->spa_delegation);
2533 spa_meta_objset(spa_t *spa)
2535 return (spa->spa_meta_objset);
2539 spa_dedup_checksum(spa_t *spa)
2541 return (spa->spa_dedup_checksum);
2545 * Reset pool scan stat per scan pass (or reboot).
2548 spa_scan_stat_init(spa_t *spa)
2550 /* data not stored on disk */
2551 spa->spa_scan_pass_start = gethrestime_sec();
2552 if (dsl_scan_is_paused_scrub(spa->spa_dsl_pool->dp_scan))
2553 spa->spa_scan_pass_scrub_pause = spa->spa_scan_pass_start;
2555 spa->spa_scan_pass_scrub_pause = 0;
2556 spa->spa_scan_pass_scrub_spent_paused = 0;
2557 spa->spa_scan_pass_exam = 0;
2558 spa->spa_scan_pass_issued = 0;
2559 vdev_scan_stat_init(spa->spa_root_vdev);
2563 * Get scan stats for zpool status reports
2566 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
2568 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
2570 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
2571 return (SET_ERROR(ENOENT));
2572 memset(ps, 0, sizeof (pool_scan_stat_t));
2574 /* data stored on disk */
2575 ps->pss_func = scn->scn_phys.scn_func;
2576 ps->pss_state = scn->scn_phys.scn_state;
2577 ps->pss_start_time = scn->scn_phys.scn_start_time;
2578 ps->pss_end_time = scn->scn_phys.scn_end_time;
2579 ps->pss_to_examine = scn->scn_phys.scn_to_examine;
2580 ps->pss_examined = scn->scn_phys.scn_examined;
2581 ps->pss_to_process = scn->scn_phys.scn_to_process;
2582 ps->pss_processed = scn->scn_phys.scn_processed;
2583 ps->pss_errors = scn->scn_phys.scn_errors;
2585 /* data not stored on disk */
2586 ps->pss_pass_exam = spa->spa_scan_pass_exam;
2587 ps->pss_pass_start = spa->spa_scan_pass_start;
2588 ps->pss_pass_scrub_pause = spa->spa_scan_pass_scrub_pause;
2589 ps->pss_pass_scrub_spent_paused = spa->spa_scan_pass_scrub_spent_paused;
2590 ps->pss_pass_issued = spa->spa_scan_pass_issued;
2592 scn->scn_issued_before_pass + spa->spa_scan_pass_issued;
2598 spa_maxblocksize(spa_t *spa)
2600 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS))
2601 return (SPA_MAXBLOCKSIZE);
2603 return (SPA_OLD_MAXBLOCKSIZE);
2608 * Returns the txg that the last device removal completed. No indirect mappings
2609 * have been added since this txg.
2612 spa_get_last_removal_txg(spa_t *spa)
2615 uint64_t ret = -1ULL;
2617 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2619 * sr_prev_indirect_vdev is only modified while holding all the
2620 * config locks, so it is sufficient to hold SCL_VDEV as reader when
2623 vdevid = spa->spa_removing_phys.sr_prev_indirect_vdev;
2625 while (vdevid != -1ULL) {
2626 vdev_t *vd = vdev_lookup_top(spa, vdevid);
2627 vdev_indirect_births_t *vib = vd->vdev_indirect_births;
2629 ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops);
2632 * If the removal did not remap any data, we don't care.
2634 if (vdev_indirect_births_count(vib) != 0) {
2635 ret = vdev_indirect_births_last_entry_txg(vib);
2639 vdevid = vd->vdev_indirect_config.vic_prev_indirect_vdev;
2641 spa_config_exit(spa, SCL_VDEV, FTAG);
2644 spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL));
2650 spa_maxdnodesize(spa_t *spa)
2652 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE))
2653 return (DNODE_MAX_SIZE);
2655 return (DNODE_MIN_SIZE);
2659 spa_multihost(spa_t *spa)
2661 return (spa->spa_multihost ? B_TRUE : B_FALSE);
2665 spa_get_hostid(spa_t *spa)
2667 return (spa->spa_hostid);
2671 spa_trust_config(spa_t *spa)
2673 return (spa->spa_trust_config);
2677 spa_missing_tvds_allowed(spa_t *spa)
2679 return (spa->spa_missing_tvds_allowed);
2683 spa_syncing_log_sm(spa_t *spa)
2685 return (spa->spa_syncing_log_sm);
2689 spa_set_missing_tvds(spa_t *spa, uint64_t missing)
2691 spa->spa_missing_tvds = missing;
2695 * Return the pool state string ("ONLINE", "DEGRADED", "SUSPENDED", etc).
2698 spa_state_to_name(spa_t *spa)
2700 ASSERT3P(spa, !=, NULL);
2703 * it is possible for the spa to exist, without root vdev
2704 * as the spa transitions during import/export
2706 vdev_t *rvd = spa->spa_root_vdev;
2708 return ("TRANSITIONING");
2710 vdev_state_t state = rvd->vdev_state;
2711 vdev_aux_t aux = rvd->vdev_stat.vs_aux;
2713 if (spa_suspended(spa) &&
2714 (spa_get_failmode(spa) != ZIO_FAILURE_MODE_CONTINUE))
2715 return ("SUSPENDED");
2718 case VDEV_STATE_CLOSED:
2719 case VDEV_STATE_OFFLINE:
2721 case VDEV_STATE_REMOVED:
2723 case VDEV_STATE_CANT_OPEN:
2724 if (aux == VDEV_AUX_CORRUPT_DATA || aux == VDEV_AUX_BAD_LOG)
2726 else if (aux == VDEV_AUX_SPLIT_POOL)
2730 case VDEV_STATE_FAULTED:
2732 case VDEV_STATE_DEGRADED:
2733 return ("DEGRADED");
2734 case VDEV_STATE_HEALTHY:
2744 spa_top_vdevs_spacemap_addressable(spa_t *spa)
2746 vdev_t *rvd = spa->spa_root_vdev;
2747 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
2748 if (!vdev_is_spacemap_addressable(rvd->vdev_child[c]))
2755 spa_has_checkpoint(spa_t *spa)
2757 return (spa->spa_checkpoint_txg != 0);
2761 spa_importing_readonly_checkpoint(spa_t *spa)
2763 return ((spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT) &&
2764 spa->spa_mode == SPA_MODE_READ);
2768 spa_min_claim_txg(spa_t *spa)
2770 uint64_t checkpoint_txg = spa->spa_uberblock.ub_checkpoint_txg;
2772 if (checkpoint_txg != 0)
2773 return (checkpoint_txg + 1);
2775 return (spa->spa_first_txg);
2779 * If there is a checkpoint, async destroys may consume more space from
2780 * the pool instead of freeing it. In an attempt to save the pool from
2781 * getting suspended when it is about to run out of space, we stop
2782 * processing async destroys.
2785 spa_suspend_async_destroy(spa_t *spa)
2787 dsl_pool_t *dp = spa_get_dsl(spa);
2789 uint64_t unreserved = dsl_pool_unreserved_space(dp,
2790 ZFS_SPACE_CHECK_EXTRA_RESERVED);
2791 uint64_t used = dsl_dir_phys(dp->dp_root_dir)->dd_used_bytes;
2792 uint64_t avail = (unreserved > used) ? (unreserved - used) : 0;
2794 if (spa_has_checkpoint(spa) && avail == 0)
2800 #if defined(_KERNEL)
2803 param_set_deadman_failmode_common(const char *val)
2809 return (SET_ERROR(EINVAL));
2811 if ((p = strchr(val, '\n')) != NULL)
2814 if (strcmp(val, "wait") != 0 && strcmp(val, "continue") != 0 &&
2815 strcmp(val, "panic"))
2816 return (SET_ERROR(EINVAL));
2818 if (spa_mode_global != SPA_MODE_UNINIT) {
2819 mutex_enter(&spa_namespace_lock);
2820 while ((spa = spa_next(spa)) != NULL)
2821 spa_set_deadman_failmode(spa, val);
2822 mutex_exit(&spa_namespace_lock);
2829 /* Namespace manipulation */
2830 EXPORT_SYMBOL(spa_lookup);
2831 EXPORT_SYMBOL(spa_add);
2832 EXPORT_SYMBOL(spa_remove);
2833 EXPORT_SYMBOL(spa_next);
2835 /* Refcount functions */
2836 EXPORT_SYMBOL(spa_open_ref);
2837 EXPORT_SYMBOL(spa_close);
2838 EXPORT_SYMBOL(spa_refcount_zero);
2840 /* Pool configuration lock */
2841 EXPORT_SYMBOL(spa_config_tryenter);
2842 EXPORT_SYMBOL(spa_config_enter);
2843 EXPORT_SYMBOL(spa_config_exit);
2844 EXPORT_SYMBOL(spa_config_held);
2846 /* Pool vdev add/remove lock */
2847 EXPORT_SYMBOL(spa_vdev_enter);
2848 EXPORT_SYMBOL(spa_vdev_exit);
2850 /* Pool vdev state change lock */
2851 EXPORT_SYMBOL(spa_vdev_state_enter);
2852 EXPORT_SYMBOL(spa_vdev_state_exit);
2854 /* Accessor functions */
2855 EXPORT_SYMBOL(spa_shutting_down);
2856 EXPORT_SYMBOL(spa_get_dsl);
2857 EXPORT_SYMBOL(spa_get_rootblkptr);
2858 EXPORT_SYMBOL(spa_set_rootblkptr);
2859 EXPORT_SYMBOL(spa_altroot);
2860 EXPORT_SYMBOL(spa_sync_pass);
2861 EXPORT_SYMBOL(spa_name);
2862 EXPORT_SYMBOL(spa_guid);
2863 EXPORT_SYMBOL(spa_last_synced_txg);
2864 EXPORT_SYMBOL(spa_first_txg);
2865 EXPORT_SYMBOL(spa_syncing_txg);
2866 EXPORT_SYMBOL(spa_version);
2867 EXPORT_SYMBOL(spa_state);
2868 EXPORT_SYMBOL(spa_load_state);
2869 EXPORT_SYMBOL(spa_freeze_txg);
2870 EXPORT_SYMBOL(spa_get_dspace);
2871 EXPORT_SYMBOL(spa_update_dspace);
2872 EXPORT_SYMBOL(spa_deflate);
2873 EXPORT_SYMBOL(spa_normal_class);
2874 EXPORT_SYMBOL(spa_log_class);
2875 EXPORT_SYMBOL(spa_special_class);
2876 EXPORT_SYMBOL(spa_preferred_class);
2877 EXPORT_SYMBOL(spa_max_replication);
2878 EXPORT_SYMBOL(spa_prev_software_version);
2879 EXPORT_SYMBOL(spa_get_failmode);
2880 EXPORT_SYMBOL(spa_suspended);
2881 EXPORT_SYMBOL(spa_bootfs);
2882 EXPORT_SYMBOL(spa_delegation);
2883 EXPORT_SYMBOL(spa_meta_objset);
2884 EXPORT_SYMBOL(spa_maxblocksize);
2885 EXPORT_SYMBOL(spa_maxdnodesize);
2887 /* Miscellaneous support routines */
2888 EXPORT_SYMBOL(spa_guid_exists);
2889 EXPORT_SYMBOL(spa_strdup);
2890 EXPORT_SYMBOL(spa_strfree);
2891 EXPORT_SYMBOL(spa_generate_guid);
2892 EXPORT_SYMBOL(snprintf_blkptr);
2893 EXPORT_SYMBOL(spa_freeze);
2894 EXPORT_SYMBOL(spa_upgrade);
2895 EXPORT_SYMBOL(spa_evict_all);
2896 EXPORT_SYMBOL(spa_lookup_by_guid);
2897 EXPORT_SYMBOL(spa_has_spare);
2898 EXPORT_SYMBOL(dva_get_dsize_sync);
2899 EXPORT_SYMBOL(bp_get_dsize_sync);
2900 EXPORT_SYMBOL(bp_get_dsize);
2901 EXPORT_SYMBOL(spa_has_slogs);
2902 EXPORT_SYMBOL(spa_is_root);
2903 EXPORT_SYMBOL(spa_writeable);
2904 EXPORT_SYMBOL(spa_mode);
2905 EXPORT_SYMBOL(spa_namespace_lock);
2906 EXPORT_SYMBOL(spa_trust_config);
2907 EXPORT_SYMBOL(spa_missing_tvds_allowed);
2908 EXPORT_SYMBOL(spa_set_missing_tvds);
2909 EXPORT_SYMBOL(spa_state_to_name);
2910 EXPORT_SYMBOL(spa_importing_readonly_checkpoint);
2911 EXPORT_SYMBOL(spa_min_claim_txg);
2912 EXPORT_SYMBOL(spa_suspend_async_destroy);
2913 EXPORT_SYMBOL(spa_has_checkpoint);
2914 EXPORT_SYMBOL(spa_top_vdevs_spacemap_addressable);
2916 ZFS_MODULE_PARAM(zfs, zfs_, flags, UINT, ZMOD_RW,
2917 "Set additional debugging flags");
2919 ZFS_MODULE_PARAM(zfs, zfs_, recover, INT, ZMOD_RW,
2920 "Set to attempt to recover from fatal errors");
2922 ZFS_MODULE_PARAM(zfs, zfs_, free_leak_on_eio, INT, ZMOD_RW,
2923 "Set to ignore IO errors during free and permanently leak the space");
2925 ZFS_MODULE_PARAM(zfs_deadman, zfs_deadman_, checktime_ms, U64, ZMOD_RW,
2926 "Dead I/O check interval in milliseconds");
2928 ZFS_MODULE_PARAM(zfs_deadman, zfs_deadman_, enabled, INT, ZMOD_RW,
2929 "Enable deadman timer");
2931 ZFS_MODULE_PARAM(zfs_spa, spa_, asize_inflation, UINT, ZMOD_RW,
2932 "SPA size estimate multiplication factor");
2934 ZFS_MODULE_PARAM(zfs, zfs_, ddt_data_is_special, INT, ZMOD_RW,
2935 "Place DDT data into the special class");
2937 ZFS_MODULE_PARAM(zfs, zfs_, user_indirect_is_special, INT, ZMOD_RW,
2938 "Place user data indirect blocks into the special class");
2941 ZFS_MODULE_PARAM_CALL(zfs_deadman, zfs_deadman_, failmode,
2942 param_set_deadman_failmode, param_get_charp, ZMOD_RW,
2943 "Failmode for deadman timer");
2945 ZFS_MODULE_PARAM_CALL(zfs_deadman, zfs_deadman_, synctime_ms,
2946 param_set_deadman_synctime, spl_param_get_u64, ZMOD_RW,
2947 "Pool sync expiration time in milliseconds");
2949 ZFS_MODULE_PARAM_CALL(zfs_deadman, zfs_deadman_, ziotime_ms,
2950 param_set_deadman_ziotime, spl_param_get_u64, ZMOD_RW,
2951 "IO expiration time in milliseconds");
2953 ZFS_MODULE_PARAM(zfs, zfs_, special_class_metadata_reserve_pct, UINT, ZMOD_RW,
2954 "Small file blocks in special vdevs depends on this much "
2955 "free space available");
2958 ZFS_MODULE_PARAM_CALL(zfs_spa, spa_, slop_shift, param_set_slop_shift,
2959 param_get_uint, ZMOD_RW, "Reserved free space in pool");