4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2019 by Delphix. All rights reserved.
24 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
26 * Copyright 2013 Saso Kiselkov. All rights reserved.
27 * Copyright (c) 2017 Datto Inc.
28 * Copyright (c) 2017, Intel Corporation.
29 * Copyright (c) 2019, loli10K <ezomori.nozomu@gmail.com>. All rights reserved.
32 #include <sys/zfs_context.h>
33 #include <sys/spa_impl.h>
35 #include <sys/zio_checksum.h>
36 #include <sys/zio_compress.h>
38 #include <sys/dmu_tx.h>
41 #include <sys/vdev_impl.h>
42 #include <sys/vdev_initialize.h>
43 #include <sys/vdev_trim.h>
44 #include <sys/vdev_file.h>
45 #include <sys/vdev_raidz.h>
46 #include <sys/metaslab.h>
47 #include <sys/uberblock_impl.h>
50 #include <sys/unique.h>
51 #include <sys/dsl_pool.h>
52 #include <sys/dsl_dir.h>
53 #include <sys/dsl_prop.h>
54 #include <sys/fm/util.h>
55 #include <sys/dsl_scan.h>
56 #include <sys/fs/zfs.h>
57 #include <sys/metaslab_impl.h>
60 #include <sys/kstat.h>
62 #include <sys/btree.h>
63 #include <sys/zfeature.h>
65 #include <sys/zstd/zstd.h>
70 * There are three basic locks for managing spa_t structures:
72 * spa_namespace_lock (global mutex)
74 * This lock must be acquired to do any of the following:
76 * - Lookup a spa_t by name
77 * - Add or remove a spa_t from the namespace
78 * - Increase spa_refcount from non-zero
79 * - Check if spa_refcount is zero
81 * - add/remove/attach/detach devices
82 * - Held for the duration of create/destroy/import/export
84 * It does not need to handle recursion. A create or destroy may
85 * reference objects (files or zvols) in other pools, but by
86 * definition they must have an existing reference, and will never need
87 * to lookup a spa_t by name.
89 * spa_refcount (per-spa zfs_refcount_t protected by mutex)
91 * This reference count keep track of any active users of the spa_t. The
92 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
93 * the refcount is never really 'zero' - opening a pool implicitly keeps
94 * some references in the DMU. Internally we check against spa_minref, but
95 * present the image of a zero/non-zero value to consumers.
97 * spa_config_lock[] (per-spa array of rwlocks)
99 * This protects the spa_t from config changes, and must be held in
100 * the following circumstances:
102 * - RW_READER to perform I/O to the spa
103 * - RW_WRITER to change the vdev config
105 * The locking order is fairly straightforward:
107 * spa_namespace_lock -> spa_refcount
109 * The namespace lock must be acquired to increase the refcount from 0
110 * or to check if it is zero.
112 * spa_refcount -> spa_config_lock[]
114 * There must be at least one valid reference on the spa_t to acquire
117 * spa_namespace_lock -> spa_config_lock[]
119 * The namespace lock must always be taken before the config lock.
122 * The spa_namespace_lock can be acquired directly and is globally visible.
124 * The namespace is manipulated using the following functions, all of which
125 * require the spa_namespace_lock to be held.
127 * spa_lookup() Lookup a spa_t by name.
129 * spa_add() Create a new spa_t in the namespace.
131 * spa_remove() Remove a spa_t from the namespace. This also
132 * frees up any memory associated with the spa_t.
134 * spa_next() Returns the next spa_t in the system, or the
135 * first if NULL is passed.
137 * spa_evict_all() Shutdown and remove all spa_t structures in
140 * spa_guid_exists() Determine whether a pool/device guid exists.
142 * The spa_refcount is manipulated using the following functions:
144 * spa_open_ref() Adds a reference to the given spa_t. Must be
145 * called with spa_namespace_lock held if the
146 * refcount is currently zero.
148 * spa_close() Remove a reference from the spa_t. This will
149 * not free the spa_t or remove it from the
150 * namespace. No locking is required.
152 * spa_refcount_zero() Returns true if the refcount is currently
153 * zero. Must be called with spa_namespace_lock
156 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
157 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
158 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
160 * To read the configuration, it suffices to hold one of these locks as reader.
161 * To modify the configuration, you must hold all locks as writer. To modify
162 * vdev state without altering the vdev tree's topology (e.g. online/offline),
163 * you must hold SCL_STATE and SCL_ZIO as writer.
165 * We use these distinct config locks to avoid recursive lock entry.
166 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
167 * block allocations (SCL_ALLOC), which may require reading space maps
168 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
170 * The spa config locks cannot be normal rwlocks because we need the
171 * ability to hand off ownership. For example, SCL_ZIO is acquired
172 * by the issuing thread and later released by an interrupt thread.
173 * They do, however, obey the usual write-wanted semantics to prevent
174 * writer (i.e. system administrator) starvation.
176 * The lock acquisition rules are as follows:
179 * Protects changes to the vdev tree topology, such as vdev
180 * add/remove/attach/detach. Protects the dirty config list
181 * (spa_config_dirty_list) and the set of spares and l2arc devices.
184 * Protects changes to pool state and vdev state, such as vdev
185 * online/offline/fault/degrade/clear. Protects the dirty state list
186 * (spa_state_dirty_list) and global pool state (spa_state).
189 * Protects changes to metaslab groups and classes.
190 * Held as reader by metaslab_alloc() and metaslab_claim().
193 * Held by bp-level zios (those which have no io_vd upon entry)
194 * to prevent changes to the vdev tree. The bp-level zio implicitly
195 * protects all of its vdev child zios, which do not hold SCL_ZIO.
198 * Protects changes to metaslab groups and classes.
199 * Held as reader by metaslab_free(). SCL_FREE is distinct from
200 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
201 * blocks in zio_done() while another i/o that holds either
202 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
205 * Held as reader to prevent changes to the vdev tree during trivial
206 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
207 * other locks, and lower than all of them, to ensure that it's safe
208 * to acquire regardless of caller context.
210 * In addition, the following rules apply:
212 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
213 * The lock ordering is SCL_CONFIG > spa_props_lock.
215 * (b) I/O operations on leaf vdevs. For any zio operation that takes
216 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
217 * or zio_write_phys() -- the caller must ensure that the config cannot
218 * cannot change in the interim, and that the vdev cannot be reopened.
219 * SCL_STATE as reader suffices for both.
221 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
223 * spa_vdev_enter() Acquire the namespace lock and the config lock
226 * spa_vdev_exit() Release the config lock, wait for all I/O
227 * to complete, sync the updated configs to the
228 * cache, and release the namespace lock.
230 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
231 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
232 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
235 static avl_tree_t spa_namespace_avl;
236 kmutex_t spa_namespace_lock;
237 static kcondvar_t spa_namespace_cv;
238 int spa_max_replication_override = SPA_DVAS_PER_BP;
240 static kmutex_t spa_spare_lock;
241 static avl_tree_t spa_spare_avl;
242 static kmutex_t spa_l2cache_lock;
243 static avl_tree_t spa_l2cache_avl;
245 kmem_cache_t *spa_buffer_pool;
246 spa_mode_t spa_mode_global = SPA_MODE_UNINIT;
250 * Everything except dprintf, set_error, spa, and indirect_remap is on
251 * by default in debug builds.
253 int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SET_ERROR |
254 ZFS_DEBUG_INDIRECT_REMAP);
260 * zfs_recover can be set to nonzero to attempt to recover from
261 * otherwise-fatal errors, typically caused by on-disk corruption. When
262 * set, calls to zfs_panic_recover() will turn into warning messages.
263 * This should only be used as a last resort, as it typically results
264 * in leaked space, or worse.
266 int zfs_recover = B_FALSE;
269 * If destroy encounters an EIO while reading metadata (e.g. indirect
270 * blocks), space referenced by the missing metadata can not be freed.
271 * Normally this causes the background destroy to become "stalled", as
272 * it is unable to make forward progress. While in this stalled state,
273 * all remaining space to free from the error-encountering filesystem is
274 * "temporarily leaked". Set this flag to cause it to ignore the EIO,
275 * permanently leak the space from indirect blocks that can not be read,
276 * and continue to free everything else that it can.
278 * The default, "stalling" behavior is useful if the storage partially
279 * fails (i.e. some but not all i/os fail), and then later recovers. In
280 * this case, we will be able to continue pool operations while it is
281 * partially failed, and when it recovers, we can continue to free the
282 * space, with no leaks. However, note that this case is actually
285 * Typically pools either (a) fail completely (but perhaps temporarily,
286 * e.g. a top-level vdev going offline), or (b) have localized,
287 * permanent errors (e.g. disk returns the wrong data due to bit flip or
288 * firmware bug). In case (a), this setting does not matter because the
289 * pool will be suspended and the sync thread will not be able to make
290 * forward progress regardless. In case (b), because the error is
291 * permanent, the best we can do is leak the minimum amount of space,
292 * which is what setting this flag will do. Therefore, it is reasonable
293 * for this flag to normally be set, but we chose the more conservative
294 * approach of not setting it, so that there is no possibility of
295 * leaking space in the "partial temporary" failure case.
297 int zfs_free_leak_on_eio = B_FALSE;
300 * Expiration time in milliseconds. This value has two meanings. First it is
301 * used to determine when the spa_deadman() logic should fire. By default the
302 * spa_deadman() will fire if spa_sync() has not completed in 600 seconds.
303 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
304 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
305 * in one of three behaviors controlled by zfs_deadman_failmode.
307 unsigned long zfs_deadman_synctime_ms = 600000UL;
310 * This value controls the maximum amount of time zio_wait() will block for an
311 * outstanding IO. By default this is 300 seconds at which point the "hung"
312 * behavior will be applied as described for zfs_deadman_synctime_ms.
314 unsigned long zfs_deadman_ziotime_ms = 300000UL;
317 * Check time in milliseconds. This defines the frequency at which we check
320 unsigned long zfs_deadman_checktime_ms = 60000UL;
323 * By default the deadman is enabled.
325 int zfs_deadman_enabled = 1;
328 * Controls the behavior of the deadman when it detects a "hung" I/O.
329 * Valid values are zfs_deadman_failmode=<wait|continue|panic>.
331 * wait - Wait for the "hung" I/O (default)
332 * continue - Attempt to recover from a "hung" I/O
333 * panic - Panic the system
335 char *zfs_deadman_failmode = "wait";
338 * The worst case is single-sector max-parity RAID-Z blocks, in which
339 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
340 * times the size; so just assume that. Add to this the fact that
341 * we can have up to 3 DVAs per bp, and one more factor of 2 because
342 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together,
344 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
346 int spa_asize_inflation = 24;
349 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
350 * the pool to be consumed (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 int spa_slop_shift = 5;
387 uint64_t spa_min_slop = 128ULL * 1024 * 1024;
388 uint64_t spa_max_slop = 128ULL * 1024 * 1024 * 1024;
389 int spa_allocators = 4;
394 spa_load_failed(spa_t *spa, const char *fmt, ...)
400 (void) vsnprintf(buf, sizeof (buf), fmt, adx);
403 zfs_dbgmsg("spa_load(%s, config %s): FAILED: %s", spa->spa_name,
404 spa->spa_trust_config ? "trusted" : "untrusted", buf);
409 spa_load_note(spa_t *spa, const char *fmt, ...)
415 (void) vsnprintf(buf, sizeof (buf), fmt, adx);
418 zfs_dbgmsg("spa_load(%s, config %s): %s", spa->spa_name,
419 spa->spa_trust_config ? "trusted" : "untrusted", buf);
423 * By default dedup and user data indirects land in the special class
425 int zfs_ddt_data_is_special = B_TRUE;
426 int zfs_user_indirect_is_special = B_TRUE;
429 * The percentage of special class final space reserved for metadata only.
430 * Once we allocate 100 - zfs_special_class_metadata_reserve_pct we only
431 * let metadata into the class.
433 int zfs_special_class_metadata_reserve_pct = 25;
436 * ==========================================================================
438 * ==========================================================================
441 spa_config_lock_init(spa_t *spa)
443 for (int i = 0; i < SCL_LOCKS; i++) {
444 spa_config_lock_t *scl = &spa->spa_config_lock[i];
445 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
446 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
447 zfs_refcount_create_untracked(&scl->scl_count);
448 scl->scl_writer = NULL;
449 scl->scl_write_wanted = 0;
454 spa_config_lock_destroy(spa_t *spa)
456 for (int i = 0; i < SCL_LOCKS; i++) {
457 spa_config_lock_t *scl = &spa->spa_config_lock[i];
458 mutex_destroy(&scl->scl_lock);
459 cv_destroy(&scl->scl_cv);
460 zfs_refcount_destroy(&scl->scl_count);
461 ASSERT(scl->scl_writer == NULL);
462 ASSERT(scl->scl_write_wanted == 0);
467 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
469 for (int i = 0; i < SCL_LOCKS; i++) {
470 spa_config_lock_t *scl = &spa->spa_config_lock[i];
471 if (!(locks & (1 << i)))
473 mutex_enter(&scl->scl_lock);
474 if (rw == RW_READER) {
475 if (scl->scl_writer || scl->scl_write_wanted) {
476 mutex_exit(&scl->scl_lock);
477 spa_config_exit(spa, locks & ((1 << i) - 1),
482 ASSERT(scl->scl_writer != curthread);
483 if (!zfs_refcount_is_zero(&scl->scl_count)) {
484 mutex_exit(&scl->scl_lock);
485 spa_config_exit(spa, locks & ((1 << i) - 1),
489 scl->scl_writer = curthread;
491 (void) zfs_refcount_add(&scl->scl_count, tag);
492 mutex_exit(&scl->scl_lock);
498 spa_config_enter(spa_t *spa, int locks, const void *tag, krw_t rw)
502 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
504 for (int i = 0; i < SCL_LOCKS; i++) {
505 spa_config_lock_t *scl = &spa->spa_config_lock[i];
506 if (scl->scl_writer == curthread)
507 wlocks_held |= (1 << i);
508 if (!(locks & (1 << i)))
510 mutex_enter(&scl->scl_lock);
511 if (rw == RW_READER) {
512 while (scl->scl_writer || scl->scl_write_wanted) {
513 cv_wait(&scl->scl_cv, &scl->scl_lock);
516 ASSERT(scl->scl_writer != curthread);
517 while (!zfs_refcount_is_zero(&scl->scl_count)) {
518 scl->scl_write_wanted++;
519 cv_wait(&scl->scl_cv, &scl->scl_lock);
520 scl->scl_write_wanted--;
522 scl->scl_writer = curthread;
524 (void) zfs_refcount_add(&scl->scl_count, tag);
525 mutex_exit(&scl->scl_lock);
527 ASSERT3U(wlocks_held, <=, locks);
531 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(!zfs_refcount_is_zero(&scl->scl_count));
539 if (zfs_refcount_remove(&scl->scl_count, tag) == 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 &&
559 !zfs_refcount_is_zero(&scl->scl_count)) ||
560 (rw == RW_WRITER && scl->scl_writer == curthread))
561 locks_held |= 1 << i;
568 * ==========================================================================
569 * SPA namespace functions
570 * ==========================================================================
574 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
575 * Returns NULL if no matching spa_t is found.
578 spa_lookup(const char *name)
580 static spa_t search; /* spa_t is large; don't allocate on stack */
585 ASSERT(MUTEX_HELD(&spa_namespace_lock));
587 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
590 * If it's a full dataset name, figure out the pool name and
593 cp = strpbrk(search.spa_name, "/@#");
597 spa = avl_find(&spa_namespace_avl, &search, &where);
603 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
604 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
605 * looking for potentially hung I/Os.
608 spa_deadman(void *arg)
612 /* Disable the deadman if the pool is suspended. */
613 if (spa_suspended(spa))
616 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
617 (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
618 ++spa->spa_deadman_calls);
619 if (zfs_deadman_enabled)
620 vdev_deadman(spa->spa_root_vdev, FTAG);
622 spa->spa_deadman_tqid = taskq_dispatch_delay(system_delay_taskq,
623 spa_deadman, spa, TQ_SLEEP, ddi_get_lbolt() +
624 MSEC_TO_TICK(zfs_deadman_checktime_ms));
628 spa_log_sm_sort_by_txg(const void *va, const void *vb)
630 const spa_log_sm_t *a = va;
631 const spa_log_sm_t *b = vb;
633 return (TREE_CMP(a->sls_txg, b->sls_txg));
637 * Create an uninitialized spa_t with the given name. Requires
638 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
639 * exist by calling spa_lookup() first.
642 spa_add(const char *name, nvlist_t *config, const char *altroot)
645 spa_config_dirent_t *dp;
647 ASSERT(MUTEX_HELD(&spa_namespace_lock));
649 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
651 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
652 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
653 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
654 mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL);
655 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
656 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
657 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
658 mutex_init(&spa->spa_cksum_tmpls_lock, NULL, MUTEX_DEFAULT, NULL);
659 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
660 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
661 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
662 mutex_init(&spa->spa_feat_stats_lock, NULL, MUTEX_DEFAULT, NULL);
663 mutex_init(&spa->spa_flushed_ms_lock, NULL, MUTEX_DEFAULT, NULL);
664 mutex_init(&spa->spa_activities_lock, NULL, MUTEX_DEFAULT, NULL);
666 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
667 cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL);
668 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
669 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
670 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
671 cv_init(&spa->spa_activities_cv, NULL, CV_DEFAULT, NULL);
672 cv_init(&spa->spa_waiters_cv, NULL, CV_DEFAULT, NULL);
674 for (int t = 0; t < TXG_SIZE; t++)
675 bplist_create(&spa->spa_free_bplist[t]);
677 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
678 spa->spa_state = POOL_STATE_UNINITIALIZED;
679 spa->spa_freeze_txg = UINT64_MAX;
680 spa->spa_final_txg = UINT64_MAX;
681 spa->spa_load_max_txg = UINT64_MAX;
683 spa->spa_proc_state = SPA_PROC_NONE;
684 spa->spa_trust_config = B_TRUE;
685 spa->spa_hostid = zone_get_hostid(NULL);
687 spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
688 spa->spa_deadman_ziotime = MSEC2NSEC(zfs_deadman_ziotime_ms);
689 spa_set_deadman_failmode(spa, zfs_deadman_failmode);
691 zfs_refcount_create(&spa->spa_refcount);
692 spa_config_lock_init(spa);
695 avl_add(&spa_namespace_avl, spa);
698 * Set the alternate root, if there is one.
701 spa->spa_root = spa_strdup(altroot);
703 spa->spa_alloc_count = spa_allocators;
704 spa->spa_alloc_locks = kmem_zalloc(spa->spa_alloc_count *
705 sizeof (kmutex_t), KM_SLEEP);
706 spa->spa_alloc_trees = kmem_zalloc(spa->spa_alloc_count *
707 sizeof (avl_tree_t), KM_SLEEP);
708 for (int i = 0; i < spa->spa_alloc_count; i++) {
709 mutex_init(&spa->spa_alloc_locks[i], NULL, MUTEX_DEFAULT, NULL);
710 avl_create(&spa->spa_alloc_trees[i], zio_bookmark_compare,
711 sizeof (zio_t), offsetof(zio_t, io_alloc_node));
713 avl_create(&spa->spa_metaslabs_by_flushed, metaslab_sort_by_flushed,
714 sizeof (metaslab_t), offsetof(metaslab_t, ms_spa_txg_node));
715 avl_create(&spa->spa_sm_logs_by_txg, spa_log_sm_sort_by_txg,
716 sizeof (spa_log_sm_t), offsetof(spa_log_sm_t, sls_node));
717 list_create(&spa->spa_log_summary, sizeof (log_summary_entry_t),
718 offsetof(log_summary_entry_t, lse_node));
721 * Every pool starts with the default cachefile
723 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
724 offsetof(spa_config_dirent_t, scd_link));
726 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
727 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
728 list_insert_head(&spa->spa_config_list, dp);
730 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
733 if (config != NULL) {
736 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
738 VERIFY(nvlist_dup(features, &spa->spa_label_features,
742 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
745 if (spa->spa_label_features == NULL) {
746 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
750 spa->spa_min_ashift = INT_MAX;
751 spa->spa_max_ashift = 0;
752 spa->spa_min_alloc = INT_MAX;
754 /* Reset cached value */
755 spa->spa_dedup_dspace = ~0ULL;
758 * As a pool is being created, treat all features as disabled by
759 * setting SPA_FEATURE_DISABLED for all entries in the feature
762 for (int i = 0; i < SPA_FEATURES; i++) {
763 spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED;
766 list_create(&spa->spa_leaf_list, sizeof (vdev_t),
767 offsetof(vdev_t, vdev_leaf_node));
773 * Removes a spa_t from the namespace, freeing up any memory used. Requires
774 * spa_namespace_lock. This is called only after the spa_t has been closed and
778 spa_remove(spa_t *spa)
780 spa_config_dirent_t *dp;
782 ASSERT(MUTEX_HELD(&spa_namespace_lock));
783 ASSERT(spa_state(spa) == POOL_STATE_UNINITIALIZED);
784 ASSERT3U(zfs_refcount_count(&spa->spa_refcount), ==, 0);
785 ASSERT0(spa->spa_waiters);
787 nvlist_free(spa->spa_config_splitting);
789 avl_remove(&spa_namespace_avl, spa);
790 cv_broadcast(&spa_namespace_cv);
793 spa_strfree(spa->spa_root);
795 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
796 list_remove(&spa->spa_config_list, dp);
797 if (dp->scd_path != NULL)
798 spa_strfree(dp->scd_path);
799 kmem_free(dp, sizeof (spa_config_dirent_t));
802 for (int i = 0; i < spa->spa_alloc_count; i++) {
803 avl_destroy(&spa->spa_alloc_trees[i]);
804 mutex_destroy(&spa->spa_alloc_locks[i]);
806 kmem_free(spa->spa_alloc_locks, spa->spa_alloc_count *
808 kmem_free(spa->spa_alloc_trees, spa->spa_alloc_count *
809 sizeof (avl_tree_t));
811 avl_destroy(&spa->spa_metaslabs_by_flushed);
812 avl_destroy(&spa->spa_sm_logs_by_txg);
813 list_destroy(&spa->spa_log_summary);
814 list_destroy(&spa->spa_config_list);
815 list_destroy(&spa->spa_leaf_list);
817 nvlist_free(spa->spa_label_features);
818 nvlist_free(spa->spa_load_info);
819 nvlist_free(spa->spa_feat_stats);
820 spa_config_set(spa, NULL);
822 zfs_refcount_destroy(&spa->spa_refcount);
824 spa_stats_destroy(spa);
825 spa_config_lock_destroy(spa);
827 for (int t = 0; t < TXG_SIZE; t++)
828 bplist_destroy(&spa->spa_free_bplist[t]);
830 zio_checksum_templates_free(spa);
832 cv_destroy(&spa->spa_async_cv);
833 cv_destroy(&spa->spa_evicting_os_cv);
834 cv_destroy(&spa->spa_proc_cv);
835 cv_destroy(&spa->spa_scrub_io_cv);
836 cv_destroy(&spa->spa_suspend_cv);
837 cv_destroy(&spa->spa_activities_cv);
838 cv_destroy(&spa->spa_waiters_cv);
840 mutex_destroy(&spa->spa_flushed_ms_lock);
841 mutex_destroy(&spa->spa_async_lock);
842 mutex_destroy(&spa->spa_errlist_lock);
843 mutex_destroy(&spa->spa_errlog_lock);
844 mutex_destroy(&spa->spa_evicting_os_lock);
845 mutex_destroy(&spa->spa_history_lock);
846 mutex_destroy(&spa->spa_proc_lock);
847 mutex_destroy(&spa->spa_props_lock);
848 mutex_destroy(&spa->spa_cksum_tmpls_lock);
849 mutex_destroy(&spa->spa_scrub_lock);
850 mutex_destroy(&spa->spa_suspend_lock);
851 mutex_destroy(&spa->spa_vdev_top_lock);
852 mutex_destroy(&spa->spa_feat_stats_lock);
853 mutex_destroy(&spa->spa_activities_lock);
855 kmem_free(spa, sizeof (spa_t));
859 * Given a pool, return the next pool in the namespace, or NULL if there is
860 * none. If 'prev' is NULL, return the first pool.
863 spa_next(spa_t *prev)
865 ASSERT(MUTEX_HELD(&spa_namespace_lock));
868 return (AVL_NEXT(&spa_namespace_avl, prev));
870 return (avl_first(&spa_namespace_avl));
874 * ==========================================================================
875 * SPA refcount functions
876 * ==========================================================================
880 * Add a reference to the given spa_t. Must have at least one reference, or
881 * have the namespace lock held.
884 spa_open_ref(spa_t *spa, void *tag)
886 ASSERT(zfs_refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
887 MUTEX_HELD(&spa_namespace_lock));
888 (void) zfs_refcount_add(&spa->spa_refcount, tag);
892 * Remove a reference to the given spa_t. Must have at least one reference, or
893 * have the namespace lock held.
896 spa_close(spa_t *spa, void *tag)
898 ASSERT(zfs_refcount_count(&spa->spa_refcount) > spa->spa_minref ||
899 MUTEX_HELD(&spa_namespace_lock));
900 (void) zfs_refcount_remove(&spa->spa_refcount, tag);
904 * Remove a reference to the given spa_t held by a dsl dir that is
905 * being asynchronously released. Async releases occur from a taskq
906 * performing eviction of dsl datasets and dirs. The namespace lock
907 * isn't held and the hold by the object being evicted may contribute to
908 * spa_minref (e.g. dataset or directory released during pool export),
909 * so the asserts in spa_close() do not apply.
912 spa_async_close(spa_t *spa, void *tag)
914 (void) zfs_refcount_remove(&spa->spa_refcount, tag);
918 * Check to see if the spa refcount is zero. Must be called with
919 * spa_namespace_lock held. We really compare against spa_minref, which is the
920 * number of references acquired when opening a pool
923 spa_refcount_zero(spa_t *spa)
925 ASSERT(MUTEX_HELD(&spa_namespace_lock));
927 return (zfs_refcount_count(&spa->spa_refcount) == spa->spa_minref);
931 * ==========================================================================
932 * SPA spare and l2cache tracking
933 * ==========================================================================
937 * Hot spares and cache devices are tracked using the same code below,
938 * for 'auxiliary' devices.
941 typedef struct spa_aux {
949 spa_aux_compare(const void *a, const void *b)
951 const spa_aux_t *sa = (const spa_aux_t *)a;
952 const spa_aux_t *sb = (const spa_aux_t *)b;
954 return (TREE_CMP(sa->aux_guid, sb->aux_guid));
958 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
964 search.aux_guid = vd->vdev_guid;
965 if ((aux = avl_find(avl, &search, &where)) != NULL) {
968 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
969 aux->aux_guid = vd->vdev_guid;
971 avl_insert(avl, aux, where);
976 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
982 search.aux_guid = vd->vdev_guid;
983 aux = avl_find(avl, &search, &where);
987 if (--aux->aux_count == 0) {
988 avl_remove(avl, aux);
989 kmem_free(aux, sizeof (spa_aux_t));
990 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
991 aux->aux_pool = 0ULL;
996 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
998 spa_aux_t search, *found;
1000 search.aux_guid = guid;
1001 found = avl_find(avl, &search, NULL);
1005 *pool = found->aux_pool;
1012 *refcnt = found->aux_count;
1017 return (found != NULL);
1021 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
1023 spa_aux_t search, *found;
1026 search.aux_guid = vd->vdev_guid;
1027 found = avl_find(avl, &search, &where);
1028 ASSERT(found != NULL);
1029 ASSERT(found->aux_pool == 0ULL);
1031 found->aux_pool = spa_guid(vd->vdev_spa);
1035 * Spares are tracked globally due to the following constraints:
1037 * - A spare may be part of multiple pools.
1038 * - A spare may be added to a pool even if it's actively in use within
1040 * - A spare in use in any pool can only be the source of a replacement if
1041 * the target is a spare in the same pool.
1043 * We keep track of all spares on the system through the use of a reference
1044 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
1045 * spare, then we bump the reference count in the AVL tree. In addition, we set
1046 * the 'vdev_isspare' member to indicate that the device is a spare (active or
1047 * inactive). When a spare is made active (used to replace a device in the
1048 * pool), we also keep track of which pool its been made a part of.
1050 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
1051 * called under the spa_namespace lock as part of vdev reconfiguration. The
1052 * separate spare lock exists for the status query path, which does not need to
1053 * be completely consistent with respect to other vdev configuration changes.
1057 spa_spare_compare(const void *a, const void *b)
1059 return (spa_aux_compare(a, b));
1063 spa_spare_add(vdev_t *vd)
1065 mutex_enter(&spa_spare_lock);
1066 ASSERT(!vd->vdev_isspare);
1067 spa_aux_add(vd, &spa_spare_avl);
1068 vd->vdev_isspare = B_TRUE;
1069 mutex_exit(&spa_spare_lock);
1073 spa_spare_remove(vdev_t *vd)
1075 mutex_enter(&spa_spare_lock);
1076 ASSERT(vd->vdev_isspare);
1077 spa_aux_remove(vd, &spa_spare_avl);
1078 vd->vdev_isspare = B_FALSE;
1079 mutex_exit(&spa_spare_lock);
1083 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
1087 mutex_enter(&spa_spare_lock);
1088 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
1089 mutex_exit(&spa_spare_lock);
1095 spa_spare_activate(vdev_t *vd)
1097 mutex_enter(&spa_spare_lock);
1098 ASSERT(vd->vdev_isspare);
1099 spa_aux_activate(vd, &spa_spare_avl);
1100 mutex_exit(&spa_spare_lock);
1104 * Level 2 ARC devices are tracked globally for the same reasons as spares.
1105 * Cache devices currently only support one pool per cache device, and so
1106 * for these devices the aux reference count is currently unused beyond 1.
1110 spa_l2cache_compare(const void *a, const void *b)
1112 return (spa_aux_compare(a, b));
1116 spa_l2cache_add(vdev_t *vd)
1118 mutex_enter(&spa_l2cache_lock);
1119 ASSERT(!vd->vdev_isl2cache);
1120 spa_aux_add(vd, &spa_l2cache_avl);
1121 vd->vdev_isl2cache = B_TRUE;
1122 mutex_exit(&spa_l2cache_lock);
1126 spa_l2cache_remove(vdev_t *vd)
1128 mutex_enter(&spa_l2cache_lock);
1129 ASSERT(vd->vdev_isl2cache);
1130 spa_aux_remove(vd, &spa_l2cache_avl);
1131 vd->vdev_isl2cache = B_FALSE;
1132 mutex_exit(&spa_l2cache_lock);
1136 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
1140 mutex_enter(&spa_l2cache_lock);
1141 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
1142 mutex_exit(&spa_l2cache_lock);
1148 spa_l2cache_activate(vdev_t *vd)
1150 mutex_enter(&spa_l2cache_lock);
1151 ASSERT(vd->vdev_isl2cache);
1152 spa_aux_activate(vd, &spa_l2cache_avl);
1153 mutex_exit(&spa_l2cache_lock);
1157 * ==========================================================================
1159 * ==========================================================================
1163 * Lock the given spa_t for the purpose of adding or removing a vdev.
1164 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1165 * It returns the next transaction group for the spa_t.
1168 spa_vdev_enter(spa_t *spa)
1170 mutex_enter(&spa->spa_vdev_top_lock);
1171 mutex_enter(&spa_namespace_lock);
1173 vdev_autotrim_stop_all(spa);
1175 return (spa_vdev_config_enter(spa));
1179 * The same as spa_vdev_enter() above but additionally takes the guid of
1180 * the vdev being detached. When there is a rebuild in process it will be
1181 * suspended while the vdev tree is modified then resumed by spa_vdev_exit().
1182 * The rebuild is canceled if only a single child remains after the detach.
1185 spa_vdev_detach_enter(spa_t *spa, uint64_t guid)
1187 mutex_enter(&spa->spa_vdev_top_lock);
1188 mutex_enter(&spa_namespace_lock);
1190 vdev_autotrim_stop_all(spa);
1193 vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE);
1195 vdev_rebuild_stop_wait(vd->vdev_top);
1199 return (spa_vdev_config_enter(spa));
1203 * Internal implementation for spa_vdev_enter(). Used when a vdev
1204 * operation requires multiple syncs (i.e. removing a device) while
1205 * keeping the spa_namespace_lock held.
1208 spa_vdev_config_enter(spa_t *spa)
1210 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1212 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1214 return (spa_last_synced_txg(spa) + 1);
1218 * Used in combination with spa_vdev_config_enter() to allow the syncing
1219 * of multiple transactions without releasing the spa_namespace_lock.
1222 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
1224 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1226 int config_changed = B_FALSE;
1228 ASSERT(txg > spa_last_synced_txg(spa));
1230 spa->spa_pending_vdev = NULL;
1233 * Reassess the DTLs.
1235 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE, B_FALSE);
1237 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
1238 config_changed = B_TRUE;
1239 spa->spa_config_generation++;
1243 * Verify the metaslab classes.
1245 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
1246 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
1247 ASSERT(metaslab_class_validate(spa_embedded_log_class(spa)) == 0);
1248 ASSERT(metaslab_class_validate(spa_special_class(spa)) == 0);
1249 ASSERT(metaslab_class_validate(spa_dedup_class(spa)) == 0);
1251 spa_config_exit(spa, SCL_ALL, spa);
1254 * Panic the system if the specified tag requires it. This
1255 * is useful for ensuring that configurations are updated
1258 if (zio_injection_enabled)
1259 zio_handle_panic_injection(spa, tag, 0);
1262 * Note: this txg_wait_synced() is important because it ensures
1263 * that there won't be more than one config change per txg.
1264 * This allows us to use the txg as the generation number.
1267 txg_wait_synced(spa->spa_dsl_pool, txg);
1270 ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
1271 if (vd->vdev_ops->vdev_op_leaf) {
1272 mutex_enter(&vd->vdev_initialize_lock);
1273 vdev_initialize_stop(vd, VDEV_INITIALIZE_CANCELED,
1275 mutex_exit(&vd->vdev_initialize_lock);
1277 mutex_enter(&vd->vdev_trim_lock);
1278 vdev_trim_stop(vd, VDEV_TRIM_CANCELED, NULL);
1279 mutex_exit(&vd->vdev_trim_lock);
1283 * The vdev may be both a leaf and top-level device.
1285 vdev_autotrim_stop_wait(vd);
1287 spa_config_enter(spa, SCL_STATE_ALL, spa, RW_WRITER);
1289 spa_config_exit(spa, SCL_STATE_ALL, spa);
1293 * If the config changed, update the config cache.
1296 spa_write_cachefile(spa, B_FALSE, B_TRUE);
1300 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1301 * locking of spa_vdev_enter(), we also want make sure the transactions have
1302 * synced to disk, and then update the global configuration cache with the new
1306 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1308 vdev_autotrim_restart(spa);
1309 vdev_rebuild_restart(spa);
1311 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1312 mutex_exit(&spa_namespace_lock);
1313 mutex_exit(&spa->spa_vdev_top_lock);
1319 * Lock the given spa_t for the purpose of changing vdev state.
1322 spa_vdev_state_enter(spa_t *spa, int oplocks)
1324 int locks = SCL_STATE_ALL | oplocks;
1327 * Root pools may need to read of the underlying devfs filesystem
1328 * when opening up a vdev. Unfortunately if we're holding the
1329 * SCL_ZIO lock it will result in a deadlock when we try to issue
1330 * the read from the root filesystem. Instead we "prefetch"
1331 * the associated vnodes that we need prior to opening the
1332 * underlying devices and cache them so that we can prevent
1333 * any I/O when we are doing the actual open.
1335 if (spa_is_root(spa)) {
1336 int low = locks & ~(SCL_ZIO - 1);
1337 int high = locks & ~low;
1339 spa_config_enter(spa, high, spa, RW_WRITER);
1340 vdev_hold(spa->spa_root_vdev);
1341 spa_config_enter(spa, low, spa, RW_WRITER);
1343 spa_config_enter(spa, locks, spa, RW_WRITER);
1345 spa->spa_vdev_locks = locks;
1349 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1351 boolean_t config_changed = B_FALSE;
1354 if (vd == NULL || vd == spa->spa_root_vdev) {
1355 vdev_top = spa->spa_root_vdev;
1357 vdev_top = vd->vdev_top;
1360 if (vd != NULL || error == 0)
1361 vdev_dtl_reassess(vdev_top, 0, 0, B_FALSE, B_FALSE);
1364 if (vd != spa->spa_root_vdev)
1365 vdev_state_dirty(vdev_top);
1367 config_changed = B_TRUE;
1368 spa->spa_config_generation++;
1371 if (spa_is_root(spa))
1372 vdev_rele(spa->spa_root_vdev);
1374 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1375 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1378 * If anything changed, wait for it to sync. This ensures that,
1379 * from the system administrator's perspective, zpool(8) commands
1380 * are synchronous. This is important for things like zpool offline:
1381 * when the command completes, you expect no further I/O from ZFS.
1384 txg_wait_synced(spa->spa_dsl_pool, 0);
1387 * If the config changed, update the config cache.
1389 if (config_changed) {
1390 mutex_enter(&spa_namespace_lock);
1391 spa_write_cachefile(spa, B_FALSE, B_TRUE);
1392 mutex_exit(&spa_namespace_lock);
1399 * ==========================================================================
1400 * Miscellaneous functions
1401 * ==========================================================================
1405 spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx)
1407 if (!nvlist_exists(spa->spa_label_features, feature)) {
1408 fnvlist_add_boolean(spa->spa_label_features, feature);
1410 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1411 * dirty the vdev config because lock SCL_CONFIG is not held.
1412 * Thankfully, in this case we don't need to dirty the config
1413 * because it will be written out anyway when we finish
1414 * creating the pool.
1416 if (tx->tx_txg != TXG_INITIAL)
1417 vdev_config_dirty(spa->spa_root_vdev);
1422 spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1424 if (nvlist_remove_all(spa->spa_label_features, feature) == 0)
1425 vdev_config_dirty(spa->spa_root_vdev);
1429 * Return the spa_t associated with given pool_guid, if it exists. If
1430 * device_guid is non-zero, determine whether the pool exists *and* contains
1431 * a device with the specified device_guid.
1434 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1437 avl_tree_t *t = &spa_namespace_avl;
1439 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1441 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1442 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1444 if (spa->spa_root_vdev == NULL)
1446 if (spa_guid(spa) == pool_guid) {
1447 if (device_guid == 0)
1450 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1451 device_guid) != NULL)
1455 * Check any devices we may be in the process of adding.
1457 if (spa->spa_pending_vdev) {
1458 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1459 device_guid) != NULL)
1469 * Determine whether a pool with the given pool_guid exists.
1472 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1474 return (spa_by_guid(pool_guid, device_guid) != NULL);
1478 spa_strdup(const char *s)
1484 new = kmem_alloc(len + 1, KM_SLEEP);
1492 spa_strfree(char *s)
1494 kmem_free(s, strlen(s) + 1);
1498 spa_get_random(uint64_t range)
1507 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1513 spa_generate_guid(spa_t *spa)
1515 uint64_t guid = spa_get_random(-1ULL);
1518 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1519 guid = spa_get_random(-1ULL);
1521 while (guid == 0 || spa_guid_exists(guid, 0))
1522 guid = spa_get_random(-1ULL);
1529 snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp)
1532 char *checksum = NULL;
1533 char *compress = NULL;
1536 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1537 dmu_object_byteswap_t bswap =
1538 DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1539 (void) snprintf(type, sizeof (type), "bswap %s %s",
1540 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1541 "metadata" : "data",
1542 dmu_ot_byteswap[bswap].ob_name);
1544 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1547 if (!BP_IS_EMBEDDED(bp)) {
1549 zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1551 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1554 SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum,
1559 spa_freeze(spa_t *spa)
1561 uint64_t freeze_txg = 0;
1563 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1564 if (spa->spa_freeze_txg == UINT64_MAX) {
1565 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1566 spa->spa_freeze_txg = freeze_txg;
1568 spa_config_exit(spa, SCL_ALL, FTAG);
1569 if (freeze_txg != 0)
1570 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1574 zfs_panic_recover(const char *fmt, ...)
1579 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1584 * This is a stripped-down version of strtoull, suitable only for converting
1585 * lowercase hexadecimal numbers that don't overflow.
1588 zfs_strtonum(const char *str, char **nptr)
1594 while ((c = *str) != '\0') {
1595 if (c >= '0' && c <= '9')
1597 else if (c >= 'a' && c <= 'f')
1598 digit = 10 + c - 'a';
1609 *nptr = (char *)str;
1615 spa_activate_allocation_classes(spa_t *spa, dmu_tx_t *tx)
1618 * We bump the feature refcount for each special vdev added to the pool
1620 ASSERT(spa_feature_is_enabled(spa, SPA_FEATURE_ALLOCATION_CLASSES));
1621 spa_feature_incr(spa, SPA_FEATURE_ALLOCATION_CLASSES, tx);
1625 * ==========================================================================
1626 * Accessor functions
1627 * ==========================================================================
1631 spa_shutting_down(spa_t *spa)
1633 return (spa->spa_async_suspended);
1637 spa_get_dsl(spa_t *spa)
1639 return (spa->spa_dsl_pool);
1643 spa_is_initializing(spa_t *spa)
1645 return (spa->spa_is_initializing);
1649 spa_indirect_vdevs_loaded(spa_t *spa)
1651 return (spa->spa_indirect_vdevs_loaded);
1655 spa_get_rootblkptr(spa_t *spa)
1657 return (&spa->spa_ubsync.ub_rootbp);
1661 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1663 spa->spa_uberblock.ub_rootbp = *bp;
1667 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1669 if (spa->spa_root == NULL)
1672 (void) strncpy(buf, spa->spa_root, buflen);
1676 spa_sync_pass(spa_t *spa)
1678 return (spa->spa_sync_pass);
1682 spa_name(spa_t *spa)
1684 return (spa->spa_name);
1688 spa_guid(spa_t *spa)
1690 dsl_pool_t *dp = spa_get_dsl(spa);
1694 * If we fail to parse the config during spa_load(), we can go through
1695 * the error path (which posts an ereport) and end up here with no root
1696 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1699 if (spa->spa_root_vdev == NULL)
1700 return (spa->spa_config_guid);
1702 guid = spa->spa_last_synced_guid != 0 ?
1703 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1706 * Return the most recently synced out guid unless we're
1707 * in syncing context.
1709 if (dp && dsl_pool_sync_context(dp))
1710 return (spa->spa_root_vdev->vdev_guid);
1716 spa_load_guid(spa_t *spa)
1719 * This is a GUID that exists solely as a reference for the
1720 * purposes of the arc. It is generated at load time, and
1721 * is never written to persistent storage.
1723 return (spa->spa_load_guid);
1727 spa_last_synced_txg(spa_t *spa)
1729 return (spa->spa_ubsync.ub_txg);
1733 spa_first_txg(spa_t *spa)
1735 return (spa->spa_first_txg);
1739 spa_syncing_txg(spa_t *spa)
1741 return (spa->spa_syncing_txg);
1745 * Return the last txg where data can be dirtied. The final txgs
1746 * will be used to just clear out any deferred frees that remain.
1749 spa_final_dirty_txg(spa_t *spa)
1751 return (spa->spa_final_txg - TXG_DEFER_SIZE);
1755 spa_state(spa_t *spa)
1757 return (spa->spa_state);
1761 spa_load_state(spa_t *spa)
1763 return (spa->spa_load_state);
1767 spa_freeze_txg(spa_t *spa)
1769 return (spa->spa_freeze_txg);
1773 * Return the inflated asize for a logical write in bytes. This is used by the
1774 * DMU to calculate the space a logical write will require on disk.
1775 * If lsize is smaller than the largest physical block size allocatable on this
1776 * pool we use its value instead, since the write will end up using the whole
1780 spa_get_worst_case_asize(spa_t *spa, uint64_t lsize)
1783 return (0); /* No inflation needed */
1784 return (MAX(lsize, 1 << spa->spa_max_ashift) * spa_asize_inflation);
1788 * Return the amount of slop space in bytes. It is typically 1/32 of the pool
1789 * (3.2%), minus the embedded log space. On very small pools, it may be
1790 * slightly larger than this. On very large pools, it will be capped to
1791 * the value of spa_max_slop. The embedded log space is not included in
1792 * spa_dspace. By subtracting it, the usable space (per "zfs list") is a
1793 * constant 97% of the total space, regardless of metaslab size (assuming the
1794 * default spa_slop_shift=5 and a non-tiny pool).
1796 * See the comment above spa_slop_shift for more details.
1799 spa_get_slop_space(spa_t *spa)
1801 uint64_t space = spa_get_dspace(spa);
1802 uint64_t slop = MIN(space >> spa_slop_shift, spa_max_slop);
1805 * Subtract the embedded log space, but no more than half the (3.2%)
1806 * unusable space. Note, the "no more than half" is only relevant if
1807 * zfs_embedded_slog_min_ms >> spa_slop_shift < 2, which is not true by
1810 uint64_t embedded_log =
1811 metaslab_class_get_dspace(spa_embedded_log_class(spa));
1812 slop -= MIN(embedded_log, slop >> 1);
1815 * Slop space should be at least spa_min_slop, but no more than half
1818 slop = MAX(slop, MIN(space >> 1, spa_min_slop));
1823 spa_get_dspace(spa_t *spa)
1825 return (spa->spa_dspace);
1829 spa_get_checkpoint_space(spa_t *spa)
1831 return (spa->spa_checkpoint_info.sci_dspace);
1835 spa_update_dspace(spa_t *spa)
1837 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1838 ddt_get_dedup_dspace(spa);
1839 if (spa->spa_vdev_removal != NULL) {
1841 * We can't allocate from the removing device, so subtract
1842 * its size if it was included in dspace (i.e. if this is a
1843 * normal-class vdev, not special/dedup). This prevents the
1844 * DMU/DSL from filling up the (now smaller) pool while we
1845 * are in the middle of removing the device.
1847 * Note that the DMU/DSL doesn't actually know or care
1848 * how much space is allocated (it does its own tracking
1849 * of how much space has been logically used). So it
1850 * doesn't matter that the data we are moving may be
1851 * allocated twice (on the old device and the new
1854 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1856 vdev_lookup_top(spa, spa->spa_vdev_removal->svr_vdev_id);
1857 if (vd->vdev_mg->mg_class == spa_normal_class(spa)) {
1858 spa->spa_dspace -= spa_deflate(spa) ?
1859 vd->vdev_stat.vs_dspace : vd->vdev_stat.vs_space;
1861 spa_config_exit(spa, SCL_VDEV, FTAG);
1866 * Return the failure mode that has been set to this pool. The default
1867 * behavior will be to block all I/Os when a complete failure occurs.
1870 spa_get_failmode(spa_t *spa)
1872 return (spa->spa_failmode);
1876 spa_suspended(spa_t *spa)
1878 return (spa->spa_suspended != ZIO_SUSPEND_NONE);
1882 spa_version(spa_t *spa)
1884 return (spa->spa_ubsync.ub_version);
1888 spa_deflate(spa_t *spa)
1890 return (spa->spa_deflate);
1894 spa_normal_class(spa_t *spa)
1896 return (spa->spa_normal_class);
1900 spa_log_class(spa_t *spa)
1902 return (spa->spa_log_class);
1906 spa_embedded_log_class(spa_t *spa)
1908 return (spa->spa_embedded_log_class);
1912 spa_special_class(spa_t *spa)
1914 return (spa->spa_special_class);
1918 spa_dedup_class(spa_t *spa)
1920 return (spa->spa_dedup_class);
1924 * Locate an appropriate allocation class
1927 spa_preferred_class(spa_t *spa, uint64_t size, dmu_object_type_t objtype,
1928 uint_t level, uint_t special_smallblk)
1931 * ZIL allocations determine their class in zio_alloc_zil().
1933 ASSERT(objtype != DMU_OT_INTENT_LOG);
1935 boolean_t has_special_class = spa->spa_special_class->mc_groups != 0;
1937 if (DMU_OT_IS_DDT(objtype)) {
1938 if (spa->spa_dedup_class->mc_groups != 0)
1939 return (spa_dedup_class(spa));
1940 else if (has_special_class && zfs_ddt_data_is_special)
1941 return (spa_special_class(spa));
1943 return (spa_normal_class(spa));
1946 /* Indirect blocks for user data can land in special if allowed */
1947 if (level > 0 && (DMU_OT_IS_FILE(objtype) || objtype == DMU_OT_ZVOL)) {
1948 if (has_special_class && zfs_user_indirect_is_special)
1949 return (spa_special_class(spa));
1951 return (spa_normal_class(spa));
1954 if (DMU_OT_IS_METADATA(objtype) || level > 0) {
1955 if (has_special_class)
1956 return (spa_special_class(spa));
1958 return (spa_normal_class(spa));
1962 * Allow small file blocks in special class in some cases (like
1963 * for the dRAID vdev feature). But always leave a reserve of
1964 * zfs_special_class_metadata_reserve_pct exclusively for metadata.
1966 if (DMU_OT_IS_FILE(objtype) &&
1967 has_special_class && size <= special_smallblk) {
1968 metaslab_class_t *special = spa_special_class(spa);
1969 uint64_t alloc = metaslab_class_get_alloc(special);
1970 uint64_t space = metaslab_class_get_space(special);
1972 (space * (100 - zfs_special_class_metadata_reserve_pct))
1979 return (spa_normal_class(spa));
1983 spa_evicting_os_register(spa_t *spa, objset_t *os)
1985 mutex_enter(&spa->spa_evicting_os_lock);
1986 list_insert_head(&spa->spa_evicting_os_list, os);
1987 mutex_exit(&spa->spa_evicting_os_lock);
1991 spa_evicting_os_deregister(spa_t *spa, objset_t *os)
1993 mutex_enter(&spa->spa_evicting_os_lock);
1994 list_remove(&spa->spa_evicting_os_list, os);
1995 cv_broadcast(&spa->spa_evicting_os_cv);
1996 mutex_exit(&spa->spa_evicting_os_lock);
2000 spa_evicting_os_wait(spa_t *spa)
2002 mutex_enter(&spa->spa_evicting_os_lock);
2003 while (!list_is_empty(&spa->spa_evicting_os_list))
2004 cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock);
2005 mutex_exit(&spa->spa_evicting_os_lock);
2007 dmu_buf_user_evict_wait();
2011 spa_max_replication(spa_t *spa)
2014 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
2015 * handle BPs with more than one DVA allocated. Set our max
2016 * replication level accordingly.
2018 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
2020 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
2024 spa_prev_software_version(spa_t *spa)
2026 return (spa->spa_prev_software_version);
2030 spa_deadman_synctime(spa_t *spa)
2032 return (spa->spa_deadman_synctime);
2036 spa_get_autotrim(spa_t *spa)
2038 return (spa->spa_autotrim);
2042 spa_deadman_ziotime(spa_t *spa)
2044 return (spa->spa_deadman_ziotime);
2048 spa_get_deadman_failmode(spa_t *spa)
2050 return (spa->spa_deadman_failmode);
2054 spa_set_deadman_failmode(spa_t *spa, const char *failmode)
2056 if (strcmp(failmode, "wait") == 0)
2057 spa->spa_deadman_failmode = ZIO_FAILURE_MODE_WAIT;
2058 else if (strcmp(failmode, "continue") == 0)
2059 spa->spa_deadman_failmode = ZIO_FAILURE_MODE_CONTINUE;
2060 else if (strcmp(failmode, "panic") == 0)
2061 spa->spa_deadman_failmode = ZIO_FAILURE_MODE_PANIC;
2063 spa->spa_deadman_failmode = ZIO_FAILURE_MODE_WAIT;
2067 spa_set_deadman_ziotime(hrtime_t ns)
2071 if (spa_mode_global != SPA_MODE_UNINIT) {
2072 mutex_enter(&spa_namespace_lock);
2073 while ((spa = spa_next(spa)) != NULL)
2074 spa->spa_deadman_ziotime = ns;
2075 mutex_exit(&spa_namespace_lock);
2080 spa_set_deadman_synctime(hrtime_t ns)
2084 if (spa_mode_global != SPA_MODE_UNINIT) {
2085 mutex_enter(&spa_namespace_lock);
2086 while ((spa = spa_next(spa)) != NULL)
2087 spa->spa_deadman_synctime = ns;
2088 mutex_exit(&spa_namespace_lock);
2093 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
2095 uint64_t asize = DVA_GET_ASIZE(dva);
2096 uint64_t dsize = asize;
2098 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
2100 if (asize != 0 && spa->spa_deflate) {
2101 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
2103 dsize = (asize >> SPA_MINBLOCKSHIFT) *
2104 vd->vdev_deflate_ratio;
2111 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
2115 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
2116 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
2122 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
2126 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2128 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
2129 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
2131 spa_config_exit(spa, SCL_VDEV, FTAG);
2137 spa_dirty_data(spa_t *spa)
2139 return (spa->spa_dsl_pool->dp_dirty_total);
2143 * ==========================================================================
2144 * SPA Import Progress Routines
2145 * ==========================================================================
2148 typedef struct spa_import_progress {
2149 uint64_t pool_guid; /* unique id for updates */
2151 spa_load_state_t spa_load_state;
2152 uint64_t mmp_sec_remaining; /* MMP activity check */
2153 uint64_t spa_load_max_txg; /* rewind txg */
2154 procfs_list_node_t smh_node;
2155 } spa_import_progress_t;
2157 spa_history_list_t *spa_import_progress_list = NULL;
2160 spa_import_progress_show_header(struct seq_file *f)
2162 seq_printf(f, "%-20s %-14s %-14s %-12s %s\n", "pool_guid",
2163 "load_state", "multihost_secs", "max_txg",
2169 spa_import_progress_show(struct seq_file *f, void *data)
2171 spa_import_progress_t *sip = (spa_import_progress_t *)data;
2173 seq_printf(f, "%-20llu %-14llu %-14llu %-12llu %s\n",
2174 (u_longlong_t)sip->pool_guid, (u_longlong_t)sip->spa_load_state,
2175 (u_longlong_t)sip->mmp_sec_remaining,
2176 (u_longlong_t)sip->spa_load_max_txg,
2177 (sip->pool_name ? sip->pool_name : "-"));
2182 /* Remove oldest elements from list until there are no more than 'size' left */
2184 spa_import_progress_truncate(spa_history_list_t *shl, unsigned int size)
2186 spa_import_progress_t *sip;
2187 while (shl->size > size) {
2188 sip = list_remove_head(&shl->procfs_list.pl_list);
2190 spa_strfree(sip->pool_name);
2191 kmem_free(sip, sizeof (spa_import_progress_t));
2195 IMPLY(size == 0, list_is_empty(&shl->procfs_list.pl_list));
2199 spa_import_progress_init(void)
2201 spa_import_progress_list = kmem_zalloc(sizeof (spa_history_list_t),
2204 spa_import_progress_list->size = 0;
2206 spa_import_progress_list->procfs_list.pl_private =
2207 spa_import_progress_list;
2209 procfs_list_install("zfs",
2213 &spa_import_progress_list->procfs_list,
2214 spa_import_progress_show,
2215 spa_import_progress_show_header,
2217 offsetof(spa_import_progress_t, smh_node));
2221 spa_import_progress_destroy(void)
2223 spa_history_list_t *shl = spa_import_progress_list;
2224 procfs_list_uninstall(&shl->procfs_list);
2225 spa_import_progress_truncate(shl, 0);
2226 procfs_list_destroy(&shl->procfs_list);
2227 kmem_free(shl, sizeof (spa_history_list_t));
2231 spa_import_progress_set_state(uint64_t pool_guid,
2232 spa_load_state_t load_state)
2234 spa_history_list_t *shl = spa_import_progress_list;
2235 spa_import_progress_t *sip;
2241 mutex_enter(&shl->procfs_list.pl_lock);
2242 for (sip = list_tail(&shl->procfs_list.pl_list); sip != NULL;
2243 sip = list_prev(&shl->procfs_list.pl_list, sip)) {
2244 if (sip->pool_guid == pool_guid) {
2245 sip->spa_load_state = load_state;
2250 mutex_exit(&shl->procfs_list.pl_lock);
2256 spa_import_progress_set_max_txg(uint64_t pool_guid, uint64_t load_max_txg)
2258 spa_history_list_t *shl = spa_import_progress_list;
2259 spa_import_progress_t *sip;
2265 mutex_enter(&shl->procfs_list.pl_lock);
2266 for (sip = list_tail(&shl->procfs_list.pl_list); sip != NULL;
2267 sip = list_prev(&shl->procfs_list.pl_list, sip)) {
2268 if (sip->pool_guid == pool_guid) {
2269 sip->spa_load_max_txg = load_max_txg;
2274 mutex_exit(&shl->procfs_list.pl_lock);
2280 spa_import_progress_set_mmp_check(uint64_t pool_guid,
2281 uint64_t mmp_sec_remaining)
2283 spa_history_list_t *shl = spa_import_progress_list;
2284 spa_import_progress_t *sip;
2290 mutex_enter(&shl->procfs_list.pl_lock);
2291 for (sip = list_tail(&shl->procfs_list.pl_list); sip != NULL;
2292 sip = list_prev(&shl->procfs_list.pl_list, sip)) {
2293 if (sip->pool_guid == pool_guid) {
2294 sip->mmp_sec_remaining = mmp_sec_remaining;
2299 mutex_exit(&shl->procfs_list.pl_lock);
2305 * A new import is in progress, add an entry.
2308 spa_import_progress_add(spa_t *spa)
2310 spa_history_list_t *shl = spa_import_progress_list;
2311 spa_import_progress_t *sip;
2312 char *poolname = NULL;
2314 sip = kmem_zalloc(sizeof (spa_import_progress_t), KM_SLEEP);
2315 sip->pool_guid = spa_guid(spa);
2317 (void) nvlist_lookup_string(spa->spa_config, ZPOOL_CONFIG_POOL_NAME,
2319 if (poolname == NULL)
2320 poolname = spa_name(spa);
2321 sip->pool_name = spa_strdup(poolname);
2322 sip->spa_load_state = spa_load_state(spa);
2324 mutex_enter(&shl->procfs_list.pl_lock);
2325 procfs_list_add(&shl->procfs_list, sip);
2327 mutex_exit(&shl->procfs_list.pl_lock);
2331 spa_import_progress_remove(uint64_t pool_guid)
2333 spa_history_list_t *shl = spa_import_progress_list;
2334 spa_import_progress_t *sip;
2336 mutex_enter(&shl->procfs_list.pl_lock);
2337 for (sip = list_tail(&shl->procfs_list.pl_list); sip != NULL;
2338 sip = list_prev(&shl->procfs_list.pl_list, sip)) {
2339 if (sip->pool_guid == pool_guid) {
2341 spa_strfree(sip->pool_name);
2342 list_remove(&shl->procfs_list.pl_list, sip);
2344 kmem_free(sip, sizeof (spa_import_progress_t));
2348 mutex_exit(&shl->procfs_list.pl_lock);
2352 * ==========================================================================
2353 * Initialization and Termination
2354 * ==========================================================================
2358 spa_name_compare(const void *a1, const void *a2)
2360 const spa_t *s1 = a1;
2361 const spa_t *s2 = a2;
2364 s = strcmp(s1->spa_name, s2->spa_name);
2366 return (TREE_ISIGN(s));
2376 spa_init(spa_mode_t mode)
2378 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
2379 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
2380 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
2381 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
2383 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
2384 offsetof(spa_t, spa_avl));
2386 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
2387 offsetof(spa_aux_t, aux_avl));
2389 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
2390 offsetof(spa_aux_t, aux_avl));
2392 spa_mode_global = mode;
2395 if (spa_mode_global != SPA_MODE_READ && dprintf_find_string("watch")) {
2396 struct sigaction sa;
2398 sa.sa_flags = SA_SIGINFO;
2399 sigemptyset(&sa.sa_mask);
2400 sa.sa_sigaction = arc_buf_sigsegv;
2402 if (sigaction(SIGSEGV, &sa, NULL) == -1) {
2403 perror("could not enable watchpoints: "
2404 "sigaction(SIGSEGV, ...) = ");
2412 zfs_refcount_init();
2415 metaslab_stat_init();
2420 vdev_cache_stat_init();
2421 vdev_mirror_stat_init();
2422 vdev_raidz_math_init();
2426 zpool_feature_init();
2431 spa_import_progress_init();
2442 vdev_cache_stat_fini();
2443 vdev_mirror_stat_fini();
2444 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 bzero(ps, 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_get_random);
2892 EXPORT_SYMBOL(spa_generate_guid);
2893 EXPORT_SYMBOL(snprintf_blkptr);
2894 EXPORT_SYMBOL(spa_freeze);
2895 EXPORT_SYMBOL(spa_upgrade);
2896 EXPORT_SYMBOL(spa_evict_all);
2897 EXPORT_SYMBOL(spa_lookup_by_guid);
2898 EXPORT_SYMBOL(spa_has_spare);
2899 EXPORT_SYMBOL(dva_get_dsize_sync);
2900 EXPORT_SYMBOL(bp_get_dsize_sync);
2901 EXPORT_SYMBOL(bp_get_dsize);
2902 EXPORT_SYMBOL(spa_has_slogs);
2903 EXPORT_SYMBOL(spa_is_root);
2904 EXPORT_SYMBOL(spa_writeable);
2905 EXPORT_SYMBOL(spa_mode);
2906 EXPORT_SYMBOL(spa_namespace_lock);
2907 EXPORT_SYMBOL(spa_trust_config);
2908 EXPORT_SYMBOL(spa_missing_tvds_allowed);
2909 EXPORT_SYMBOL(spa_set_missing_tvds);
2910 EXPORT_SYMBOL(spa_state_to_name);
2911 EXPORT_SYMBOL(spa_importing_readonly_checkpoint);
2912 EXPORT_SYMBOL(spa_min_claim_txg);
2913 EXPORT_SYMBOL(spa_suspend_async_destroy);
2914 EXPORT_SYMBOL(spa_has_checkpoint);
2915 EXPORT_SYMBOL(spa_top_vdevs_spacemap_addressable);
2917 ZFS_MODULE_PARAM(zfs, zfs_, flags, UINT, ZMOD_RW,
2918 "Set additional debugging flags");
2920 ZFS_MODULE_PARAM(zfs, zfs_, recover, INT, ZMOD_RW,
2921 "Set to attempt to recover from fatal errors");
2923 ZFS_MODULE_PARAM(zfs, zfs_, free_leak_on_eio, INT, ZMOD_RW,
2924 "Set to ignore IO errors during free and permanently leak the space");
2926 ZFS_MODULE_PARAM(zfs_deadman, zfs_deadman_, checktime_ms, ULONG, ZMOD_RW,
2927 "Dead I/O check interval in milliseconds");
2929 ZFS_MODULE_PARAM(zfs_deadman, zfs_deadman_, enabled, INT, ZMOD_RW,
2930 "Enable deadman timer");
2932 ZFS_MODULE_PARAM(zfs_spa, spa_, asize_inflation, INT, ZMOD_RW,
2933 "SPA size estimate multiplication factor");
2935 ZFS_MODULE_PARAM(zfs, zfs_, ddt_data_is_special, INT, ZMOD_RW,
2936 "Place DDT data into the special class");
2938 ZFS_MODULE_PARAM(zfs, zfs_, user_indirect_is_special, INT, ZMOD_RW,
2939 "Place user data indirect blocks into the special class");
2942 ZFS_MODULE_PARAM_CALL(zfs_deadman, zfs_deadman_, failmode,
2943 param_set_deadman_failmode, param_get_charp, ZMOD_RW,
2944 "Failmode for deadman timer");
2946 ZFS_MODULE_PARAM_CALL(zfs_deadman, zfs_deadman_, synctime_ms,
2947 param_set_deadman_synctime, param_get_ulong, ZMOD_RW,
2948 "Pool sync expiration time in milliseconds");
2950 ZFS_MODULE_PARAM_CALL(zfs_deadman, zfs_deadman_, ziotime_ms,
2951 param_set_deadman_ziotime, param_get_ulong, ZMOD_RW,
2952 "IO expiration time in milliseconds");
2954 ZFS_MODULE_PARAM(zfs, zfs_, special_class_metadata_reserve_pct, INT, ZMOD_RW,
2955 "Small file blocks in special vdevs depends on this much "
2956 "free space available");
2959 ZFS_MODULE_PARAM_CALL(zfs_spa, spa_, slop_shift, param_set_slop_shift,
2960 param_get_int, ZMOD_RW, "Reserved free space in pool");