4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
24 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
25 * Copyright 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 * Copyright 2013 Saso Kiselkov. All rights reserved.
28 * Copyright (c) 2014 Integros [integros.com]
29 * Copyright (c) 2017 Datto Inc.
32 #include <sys/zfs_context.h>
33 #include <sys/spa_impl.h>
34 #include <sys/spa_boot.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_file.h>
44 #include <sys/vdev_initialize.h>
45 #include <sys/metaslab.h>
46 #include <sys/uberblock_impl.h>
49 #include <sys/unique.h>
50 #include <sys/dsl_pool.h>
51 #include <sys/dsl_dir.h>
52 #include <sys/dsl_prop.h>
53 #include <sys/dsl_scan.h>
54 #include <sys/fs/zfs.h>
55 #include <sys/metaslab_impl.h>
59 #include <sys/zfeature.h>
61 #if defined(__FreeBSD__) && defined(_KERNEL)
62 #include <sys/types.h>
63 #include <sys/sysctl.h>
69 * There are four basic locks for managing spa_t structures:
71 * spa_namespace_lock (global mutex)
73 * This lock must be acquired to do any of the following:
75 * - Lookup a spa_t by name
76 * - Add or remove a spa_t from the namespace
77 * - Increase spa_refcount from non-zero
78 * - Check if spa_refcount is zero
80 * - add/remove/attach/detach devices
81 * - Held for the duration of create/destroy/import/export
83 * It does not need to handle recursion. A create or destroy may
84 * reference objects (files or zvols) in other pools, but by
85 * definition they must have an existing reference, and will never need
86 * to lookup a spa_t by name.
88 * spa_refcount (per-spa refcount_t protected by mutex)
90 * This reference count keep track of any active users of the spa_t. The
91 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
92 * the refcount is never really 'zero' - opening a pool implicitly keeps
93 * some references in the DMU. Internally we check against spa_minref, but
94 * present the image of a zero/non-zero value to consumers.
96 * spa_config_lock[] (per-spa array of rwlocks)
98 * This protects the spa_t from config changes, and must be held in
99 * the following circumstances:
101 * - RW_READER to perform I/O to the spa
102 * - RW_WRITER to change the vdev config
104 * The locking order is fairly straightforward:
106 * spa_namespace_lock -> spa_refcount
108 * The namespace lock must be acquired to increase the refcount from 0
109 * or to check if it is zero.
111 * spa_refcount -> spa_config_lock[]
113 * There must be at least one valid reference on the spa_t to acquire
116 * spa_namespace_lock -> spa_config_lock[]
118 * The namespace lock must always be taken before the config lock.
121 * The spa_namespace_lock can be acquired directly and is globally visible.
123 * The namespace is manipulated using the following functions, all of which
124 * require the spa_namespace_lock to be held.
126 * spa_lookup() Lookup a spa_t by name.
128 * spa_add() Create a new spa_t in the namespace.
130 * spa_remove() Remove a spa_t from the namespace. This also
131 * frees up any memory associated with the spa_t.
133 * spa_next() Returns the next spa_t in the system, or the
134 * first if NULL is passed.
136 * spa_evict_all() Shutdown and remove all spa_t structures in
139 * spa_guid_exists() Determine whether a pool/device guid exists.
141 * The spa_refcount is manipulated using the following functions:
143 * spa_open_ref() Adds a reference to the given spa_t. Must be
144 * called with spa_namespace_lock held if the
145 * refcount is currently zero.
147 * spa_close() Remove a reference from the spa_t. This will
148 * not free the spa_t or remove it from the
149 * namespace. No locking is required.
151 * spa_refcount_zero() Returns true if the refcount is currently
152 * zero. Must be called with spa_namespace_lock
155 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
156 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
157 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
159 * To read the configuration, it suffices to hold one of these locks as reader.
160 * To modify the configuration, you must hold all locks as writer. To modify
161 * vdev state without altering the vdev tree's topology (e.g. online/offline),
162 * you must hold SCL_STATE and SCL_ZIO as writer.
164 * We use these distinct config locks to avoid recursive lock entry.
165 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
166 * block allocations (SCL_ALLOC), which may require reading space maps
167 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
169 * The spa config locks cannot be normal rwlocks because we need the
170 * ability to hand off ownership. For example, SCL_ZIO is acquired
171 * by the issuing thread and later released by an interrupt thread.
172 * They do, however, obey the usual write-wanted semantics to prevent
173 * writer (i.e. system administrator) starvation.
175 * The lock acquisition rules are as follows:
178 * Protects changes to the vdev tree topology, such as vdev
179 * add/remove/attach/detach. Protects the dirty config list
180 * (spa_config_dirty_list) and the set of spares and l2arc devices.
183 * Protects changes to pool state and vdev state, such as vdev
184 * online/offline/fault/degrade/clear. Protects the dirty state list
185 * (spa_state_dirty_list) and global pool state (spa_state).
188 * Protects changes to metaslab groups and classes.
189 * Held as reader by metaslab_alloc() and metaslab_claim().
192 * Held by bp-level zios (those which have no io_vd upon entry)
193 * to prevent changes to the vdev tree. The bp-level zio implicitly
194 * protects all of its vdev child zios, which do not hold SCL_ZIO.
197 * Protects changes to metaslab groups and classes.
198 * Held as reader by metaslab_free(). SCL_FREE is distinct from
199 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
200 * blocks in zio_done() while another i/o that holds either
201 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
204 * Held as reader to prevent changes to the vdev tree during trivial
205 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
206 * other locks, and lower than all of them, to ensure that it's safe
207 * to acquire regardless of caller context.
209 * In addition, the following rules apply:
211 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
212 * The lock ordering is SCL_CONFIG > spa_props_lock.
214 * (b) I/O operations on leaf vdevs. For any zio operation that takes
215 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
216 * or zio_write_phys() -- the caller must ensure that the config cannot
217 * cannot change in the interim, and that the vdev cannot be reopened.
218 * SCL_STATE as reader suffices for both.
220 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
222 * spa_vdev_enter() Acquire the namespace lock and the config lock
225 * spa_vdev_exit() Release the config lock, wait for all I/O
226 * to complete, sync the updated configs to the
227 * cache, and release the namespace lock.
229 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
230 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
231 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
234 static avl_tree_t spa_namespace_avl;
235 kmutex_t spa_namespace_lock;
236 static kcondvar_t spa_namespace_cv;
237 static int spa_active_count;
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;
250 * Everything except dprintf, spa, and indirect_remap is on by default
253 int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_INDIRECT_REMAP);
259 * zfs_recover can be set to nonzero to attempt to recover from
260 * otherwise-fatal errors, typically caused by on-disk corruption. When
261 * set, calls to zfs_panic_recover() will turn into warning messages.
262 * This should only be used as a last resort, as it typically results
263 * in leaked space, or worse.
265 boolean_t zfs_recover = B_FALSE;
268 * If destroy encounters an EIO while reading metadata (e.g. indirect
269 * blocks), space referenced by the missing metadata can not be freed.
270 * Normally this causes the background destroy to become "stalled", as
271 * it is unable to make forward progress. While in this stalled state,
272 * all remaining space to free from the error-encountering filesystem is
273 * "temporarily leaked". Set this flag to cause it to ignore the EIO,
274 * permanently leak the space from indirect blocks that can not be read,
275 * and continue to free everything else that it can.
277 * The default, "stalling" behavior is useful if the storage partially
278 * fails (i.e. some but not all i/os fail), and then later recovers. In
279 * this case, we will be able to continue pool operations while it is
280 * partially failed, and when it recovers, we can continue to free the
281 * space, with no leaks. However, note that this case is actually
284 * Typically pools either (a) fail completely (but perhaps temporarily,
285 * e.g. a top-level vdev going offline), or (b) have localized,
286 * permanent errors (e.g. disk returns the wrong data due to bit flip or
287 * firmware bug). In case (a), this setting does not matter because the
288 * pool will be suspended and the sync thread will not be able to make
289 * forward progress regardless. In case (b), because the error is
290 * permanent, the best we can do is leak the minimum amount of space,
291 * which is what setting this flag will do. Therefore, it is reasonable
292 * for this flag to normally be set, but we chose the more conservative
293 * approach of not setting it, so that there is no possibility of
294 * leaking space in the "partial temporary" failure case.
296 boolean_t zfs_free_leak_on_eio = B_FALSE;
299 * Expiration time in milliseconds. This value has two meanings. First it is
300 * used to determine when the spa_deadman() logic should fire. By default the
301 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
302 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
303 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
306 uint64_t zfs_deadman_synctime_ms = 1000000ULL;
309 * Check time in milliseconds. This defines the frequency at which we check
312 uint64_t zfs_deadman_checktime_ms = 5000ULL;
315 * Default value of -1 for zfs_deadman_enabled is resolved in
318 int zfs_deadman_enabled = -1;
321 * The worst case is single-sector max-parity RAID-Z blocks, in which
322 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
323 * times the size; so just assume that. Add to this the fact that
324 * we can have up to 3 DVAs per bp, and one more factor of 2 because
325 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together,
327 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
329 int spa_asize_inflation = 24;
331 #if defined(__FreeBSD__) && defined(_KERNEL)
332 SYSCTL_DECL(_vfs_zfs);
333 SYSCTL_INT(_vfs_zfs, OID_AUTO, recover, CTLFLAG_RWTUN, &zfs_recover, 0,
334 "Try to recover from otherwise-fatal errors.");
337 sysctl_vfs_zfs_debug_flags(SYSCTL_HANDLER_ARGS)
342 err = sysctl_handle_int(oidp, &val, 0, req);
343 if (err != 0 || req->newptr == NULL)
347 * ZFS_DEBUG_MODIFY must be enabled prior to boot so all
348 * arc buffers in the system have the necessary additional
349 * checksum data. However, it is safe to disable at any
352 if (!(zfs_flags & ZFS_DEBUG_MODIFY))
353 val &= ~ZFS_DEBUG_MODIFY;
359 SYSCTL_PROC(_vfs_zfs, OID_AUTO, debugflags,
360 CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RWTUN, 0, sizeof(int),
361 sysctl_vfs_zfs_debug_flags, "IU", "Debug flags for ZFS testing.");
363 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_synctime_ms, CTLFLAG_RWTUN,
364 &zfs_deadman_synctime_ms, 0,
365 "Stalled ZFS I/O expiration time in milliseconds");
366 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_checktime_ms, CTLFLAG_RWTUN,
367 &zfs_deadman_checktime_ms, 0,
368 "Period of checks for stalled ZFS I/O in milliseconds");
369 SYSCTL_INT(_vfs_zfs, OID_AUTO, deadman_enabled, CTLFLAG_RWTUN,
370 &zfs_deadman_enabled, 0, "Kernel panic on stalled ZFS I/O");
371 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_asize_inflation, CTLFLAG_RWTUN,
372 &spa_asize_inflation, 0, "Worst case inflation factor for single sector writes");
381 * If we are not i386 or amd64 or in a virtual machine,
382 * disable ZFS deadman thread by default
384 if (zfs_deadman_enabled == -1) {
385 #if defined(__amd64__) || defined(__i386__)
386 zfs_deadman_enabled = (vm_guest == VM_GUEST_NO) ? 1 : 0;
388 zfs_deadman_enabled = 0;
393 #endif /* !illumos */
396 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
397 * the pool to be consumed. This ensures that we don't run the pool
398 * completely out of space, due to unaccounted changes (e.g. to the MOS).
399 * It also limits the worst-case time to allocate space. If we have
400 * less than this amount of free space, most ZPL operations (e.g. write,
401 * create) will return ENOSPC.
403 * Certain operations (e.g. file removal, most administrative actions) can
404 * use half the slop space. They will only return ENOSPC if less than half
405 * the slop space is free. Typically, once the pool has less than the slop
406 * space free, the user will use these operations to free up space in the pool.
407 * These are the operations that call dsl_pool_adjustedsize() with the netfree
408 * argument set to TRUE.
410 * Operations that are almost guaranteed to free up space in the absence of
411 * a pool checkpoint can use up to three quarters of the slop space
414 * A very restricted set of operations are always permitted, regardless of
415 * the amount of free space. These are the operations that call
416 * dsl_sync_task(ZFS_SPACE_CHECK_NONE). If these operations result in a net
417 * increase in the amount of space used, it is possible to run the pool
418 * completely out of space, causing it to be permanently read-only.
420 * Note that on very small pools, the slop space will be larger than
421 * 3.2%, in an effort to have it be at least spa_min_slop (128MB),
422 * but we never allow it to be more than half the pool size.
424 * See also the comments in zfs_space_check_t.
426 int spa_slop_shift = 5;
427 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_slop_shift, CTLFLAG_RWTUN,
429 "Shift value of reserved space (1/(2^spa_slop_shift)).");
430 uint64_t spa_min_slop = 128 * 1024 * 1024;
431 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, spa_min_slop, CTLFLAG_RWTUN,
433 "Minimal value of reserved space");
435 int spa_allocators = 4;
437 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_allocators, CTLFLAG_RWTUN,
439 "Number of allocators per metaslab group");
443 spa_load_failed(spa_t *spa, const char *fmt, ...)
449 (void) vsnprintf(buf, sizeof (buf), fmt, adx);
452 zfs_dbgmsg("spa_load(%s, config %s): FAILED: %s", spa->spa_name,
453 spa->spa_trust_config ? "trusted" : "untrusted", buf);
458 spa_load_note(spa_t *spa, const char *fmt, ...)
464 (void) vsnprintf(buf, sizeof (buf), fmt, adx);
467 zfs_dbgmsg("spa_load(%s, config %s): %s", spa->spa_name,
468 spa->spa_trust_config ? "trusted" : "untrusted", buf);
472 * ==========================================================================
474 * ==========================================================================
477 spa_config_lock_init(spa_t *spa)
479 for (int i = 0; i < SCL_LOCKS; i++) {
480 spa_config_lock_t *scl = &spa->spa_config_lock[i];
481 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
482 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
483 refcount_create_untracked(&scl->scl_count);
484 scl->scl_writer = NULL;
485 scl->scl_write_wanted = 0;
490 spa_config_lock_destroy(spa_t *spa)
492 for (int i = 0; i < SCL_LOCKS; i++) {
493 spa_config_lock_t *scl = &spa->spa_config_lock[i];
494 mutex_destroy(&scl->scl_lock);
495 cv_destroy(&scl->scl_cv);
496 refcount_destroy(&scl->scl_count);
497 ASSERT(scl->scl_writer == NULL);
498 ASSERT(scl->scl_write_wanted == 0);
503 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
505 for (int i = 0; i < SCL_LOCKS; i++) {
506 spa_config_lock_t *scl = &spa->spa_config_lock[i];
507 if (!(locks & (1 << i)))
509 mutex_enter(&scl->scl_lock);
510 if (rw == RW_READER) {
511 if (scl->scl_writer || scl->scl_write_wanted) {
512 mutex_exit(&scl->scl_lock);
513 spa_config_exit(spa, locks & ((1 << i) - 1),
518 ASSERT(scl->scl_writer != curthread);
519 if (!refcount_is_zero(&scl->scl_count)) {
520 mutex_exit(&scl->scl_lock);
521 spa_config_exit(spa, locks & ((1 << i) - 1),
525 scl->scl_writer = curthread;
527 (void) refcount_add(&scl->scl_count, tag);
528 mutex_exit(&scl->scl_lock);
534 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
538 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
540 for (int i = 0; i < SCL_LOCKS; i++) {
541 spa_config_lock_t *scl = &spa->spa_config_lock[i];
542 if (scl->scl_writer == curthread)
543 wlocks_held |= (1 << i);
544 if (!(locks & (1 << i)))
546 mutex_enter(&scl->scl_lock);
547 if (rw == RW_READER) {
548 while (scl->scl_writer || scl->scl_write_wanted) {
549 cv_wait(&scl->scl_cv, &scl->scl_lock);
552 ASSERT(scl->scl_writer != curthread);
553 while (!refcount_is_zero(&scl->scl_count)) {
554 scl->scl_write_wanted++;
555 cv_wait(&scl->scl_cv, &scl->scl_lock);
556 scl->scl_write_wanted--;
558 scl->scl_writer = curthread;
560 (void) refcount_add(&scl->scl_count, tag);
561 mutex_exit(&scl->scl_lock);
563 ASSERT3U(wlocks_held, <=, locks);
567 spa_config_exit(spa_t *spa, int locks, void *tag)
569 for (int i = SCL_LOCKS - 1; i >= 0; i--) {
570 spa_config_lock_t *scl = &spa->spa_config_lock[i];
571 if (!(locks & (1 << i)))
573 mutex_enter(&scl->scl_lock);
574 ASSERT(!refcount_is_zero(&scl->scl_count));
575 if (refcount_remove(&scl->scl_count, tag) == 0) {
576 ASSERT(scl->scl_writer == NULL ||
577 scl->scl_writer == curthread);
578 scl->scl_writer = NULL; /* OK in either case */
579 cv_broadcast(&scl->scl_cv);
581 mutex_exit(&scl->scl_lock);
586 spa_config_held(spa_t *spa, int locks, krw_t rw)
590 for (int i = 0; i < SCL_LOCKS; i++) {
591 spa_config_lock_t *scl = &spa->spa_config_lock[i];
592 if (!(locks & (1 << i)))
594 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
595 (rw == RW_WRITER && scl->scl_writer == curthread))
596 locks_held |= 1 << i;
603 * ==========================================================================
604 * SPA namespace functions
605 * ==========================================================================
609 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
610 * Returns NULL if no matching spa_t is found.
613 spa_lookup(const char *name)
615 static spa_t search; /* spa_t is large; don't allocate on stack */
620 ASSERT(MUTEX_HELD(&spa_namespace_lock));
622 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
625 * If it's a full dataset name, figure out the pool name and
628 cp = strpbrk(search.spa_name, "/@#");
632 spa = avl_find(&spa_namespace_avl, &search, &where);
638 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
639 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
640 * looking for potentially hung I/Os.
643 spa_deadman(void *arg, int pending)
648 * Disable the deadman timer if the pool is suspended.
650 if (spa_suspended(spa)) {
652 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
654 /* Nothing. just don't schedule any future callouts. */
659 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
660 (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
661 ++spa->spa_deadman_calls);
662 if (zfs_deadman_enabled)
663 vdev_deadman(spa->spa_root_vdev);
666 callout_schedule(&spa->spa_deadman_cycid,
667 hz * zfs_deadman_checktime_ms / MILLISEC);
672 #if defined(__FreeBSD__) && defined(_KERNEL)
674 spa_deadman_timeout(void *arg)
678 taskqueue_enqueue(taskqueue_thread, &spa->spa_deadman_task);
683 * Create an uninitialized spa_t with the given name. Requires
684 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
685 * exist by calling spa_lookup() first.
688 spa_add(const char *name, nvlist_t *config, const char *altroot)
691 spa_config_dirent_t *dp;
697 ASSERT(MUTEX_HELD(&spa_namespace_lock));
699 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
701 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
702 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
703 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
704 mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL);
705 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
706 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
707 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
708 mutex_init(&spa->spa_cksum_tmpls_lock, NULL, MUTEX_DEFAULT, NULL);
709 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
710 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
711 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
712 mutex_init(&spa->spa_feat_stats_lock, NULL, MUTEX_DEFAULT, NULL);
714 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
715 cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL);
716 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
717 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
718 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
720 for (int t = 0; t < TXG_SIZE; t++)
721 bplist_create(&spa->spa_free_bplist[t]);
723 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
724 spa->spa_state = POOL_STATE_UNINITIALIZED;
725 spa->spa_freeze_txg = UINT64_MAX;
726 spa->spa_final_txg = UINT64_MAX;
727 spa->spa_load_max_txg = UINT64_MAX;
729 spa->spa_proc_state = SPA_PROC_NONE;
730 spa->spa_trust_config = B_TRUE;
733 hdlr.cyh_func = spa_deadman;
735 hdlr.cyh_level = CY_LOW_LEVEL;
738 spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
742 * This determines how often we need to check for hung I/Os after
743 * the cyclic has already fired. Since checking for hung I/Os is
744 * an expensive operation we don't want to check too frequently.
745 * Instead wait for 5 seconds before checking again.
747 when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms);
748 when.cyt_when = CY_INFINITY;
749 mutex_enter(&cpu_lock);
750 spa->spa_deadman_cycid = cyclic_add(&hdlr, &when);
751 mutex_exit(&cpu_lock);
755 * callout(9) does not provide a way to initialize a callout with
756 * a function and an argument, so we use callout_reset() to schedule
757 * the callout in the very distant future. Even if that event ever
758 * fires, it should be okayas we won't have any active zio-s.
759 * But normally spa_sync() will reschedule the callout with a proper
761 * callout(9) does not allow the callback function to sleep but
762 * vdev_deadman() needs to acquire vq_lock and illumos mutexes are
763 * emulated using sx(9). For this reason spa_deadman_timeout()
764 * will schedule spa_deadman() as task on a taskqueue that allows
767 TASK_INIT(&spa->spa_deadman_task, 0, spa_deadman, spa);
768 callout_init(&spa->spa_deadman_cycid, 1);
769 callout_reset_sbt(&spa->spa_deadman_cycid, SBT_MAX, 0,
770 spa_deadman_timeout, spa, 0);
773 refcount_create(&spa->spa_refcount);
774 spa_config_lock_init(spa);
776 avl_add(&spa_namespace_avl, spa);
779 * Set the alternate root, if there is one.
782 spa->spa_root = spa_strdup(altroot);
786 spa->spa_alloc_count = spa_allocators;
787 spa->spa_alloc_locks = kmem_zalloc(spa->spa_alloc_count *
788 sizeof (kmutex_t), KM_SLEEP);
789 spa->spa_alloc_trees = kmem_zalloc(spa->spa_alloc_count *
790 sizeof (avl_tree_t), KM_SLEEP);
791 for (int i = 0; i < spa->spa_alloc_count; i++) {
792 mutex_init(&spa->spa_alloc_locks[i], NULL, MUTEX_DEFAULT, NULL);
793 avl_create(&spa->spa_alloc_trees[i], zio_bookmark_compare,
794 sizeof (zio_t), offsetof(zio_t, io_alloc_node));
798 * Every pool starts with the default cachefile
800 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
801 offsetof(spa_config_dirent_t, scd_link));
803 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
804 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
805 list_insert_head(&spa->spa_config_list, dp);
807 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
810 if (config != NULL) {
813 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
815 VERIFY(nvlist_dup(features, &spa->spa_label_features,
819 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
822 if (spa->spa_label_features == NULL) {
823 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
827 spa->spa_min_ashift = INT_MAX;
828 spa->spa_max_ashift = 0;
831 * As a pool is being created, treat all features as disabled by
832 * setting SPA_FEATURE_DISABLED for all entries in the feature
835 for (int i = 0; i < SPA_FEATURES; i++) {
836 spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED;
843 * Removes a spa_t from the namespace, freeing up any memory used. Requires
844 * spa_namespace_lock. This is called only after the spa_t has been closed and
848 spa_remove(spa_t *spa)
850 spa_config_dirent_t *dp;
852 ASSERT(MUTEX_HELD(&spa_namespace_lock));
853 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
854 ASSERT3U(refcount_count(&spa->spa_refcount), ==, 0);
856 nvlist_free(spa->spa_config_splitting);
858 avl_remove(&spa_namespace_avl, spa);
859 cv_broadcast(&spa_namespace_cv);
862 spa_strfree(spa->spa_root);
866 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
867 list_remove(&spa->spa_config_list, dp);
868 if (dp->scd_path != NULL)
869 spa_strfree(dp->scd_path);
870 kmem_free(dp, sizeof (spa_config_dirent_t));
873 for (int i = 0; i < spa->spa_alloc_count; i++) {
874 avl_destroy(&spa->spa_alloc_trees[i]);
875 mutex_destroy(&spa->spa_alloc_locks[i]);
877 kmem_free(spa->spa_alloc_locks, spa->spa_alloc_count *
879 kmem_free(spa->spa_alloc_trees, spa->spa_alloc_count *
880 sizeof (avl_tree_t));
882 list_destroy(&spa->spa_config_list);
884 nvlist_free(spa->spa_label_features);
885 nvlist_free(spa->spa_load_info);
886 nvlist_free(spa->spa_feat_stats);
887 spa_config_set(spa, NULL);
890 mutex_enter(&cpu_lock);
891 if (spa->spa_deadman_cycid != CYCLIC_NONE)
892 cyclic_remove(spa->spa_deadman_cycid);
893 mutex_exit(&cpu_lock);
894 spa->spa_deadman_cycid = CYCLIC_NONE;
897 callout_drain(&spa->spa_deadman_cycid);
898 taskqueue_drain(taskqueue_thread, &spa->spa_deadman_task);
902 refcount_destroy(&spa->spa_refcount);
904 spa_config_lock_destroy(spa);
906 for (int t = 0; t < TXG_SIZE; t++)
907 bplist_destroy(&spa->spa_free_bplist[t]);
909 zio_checksum_templates_free(spa);
911 cv_destroy(&spa->spa_async_cv);
912 cv_destroy(&spa->spa_evicting_os_cv);
913 cv_destroy(&spa->spa_proc_cv);
914 cv_destroy(&spa->spa_scrub_io_cv);
915 cv_destroy(&spa->spa_suspend_cv);
917 mutex_destroy(&spa->spa_async_lock);
918 mutex_destroy(&spa->spa_errlist_lock);
919 mutex_destroy(&spa->spa_errlog_lock);
920 mutex_destroy(&spa->spa_evicting_os_lock);
921 mutex_destroy(&spa->spa_history_lock);
922 mutex_destroy(&spa->spa_proc_lock);
923 mutex_destroy(&spa->spa_props_lock);
924 mutex_destroy(&spa->spa_cksum_tmpls_lock);
925 mutex_destroy(&spa->spa_scrub_lock);
926 mutex_destroy(&spa->spa_suspend_lock);
927 mutex_destroy(&spa->spa_vdev_top_lock);
928 mutex_destroy(&spa->spa_feat_stats_lock);
930 kmem_free(spa, sizeof (spa_t));
934 * Given a pool, return the next pool in the namespace, or NULL if there is
935 * none. If 'prev' is NULL, return the first pool.
938 spa_next(spa_t *prev)
940 ASSERT(MUTEX_HELD(&spa_namespace_lock));
943 return (AVL_NEXT(&spa_namespace_avl, prev));
945 return (avl_first(&spa_namespace_avl));
949 * ==========================================================================
950 * SPA refcount functions
951 * ==========================================================================
955 * Add a reference to the given spa_t. Must have at least one reference, or
956 * have the namespace lock held.
959 spa_open_ref(spa_t *spa, void *tag)
961 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
962 MUTEX_HELD(&spa_namespace_lock));
963 (void) refcount_add(&spa->spa_refcount, tag);
967 * Remove a reference to the given spa_t. Must have at least one reference, or
968 * have the namespace lock held.
971 spa_close(spa_t *spa, void *tag)
973 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
974 MUTEX_HELD(&spa_namespace_lock));
975 (void) refcount_remove(&spa->spa_refcount, tag);
979 * Remove a reference to the given spa_t held by a dsl dir that is
980 * being asynchronously released. Async releases occur from a taskq
981 * performing eviction of dsl datasets and dirs. The namespace lock
982 * isn't held and the hold by the object being evicted may contribute to
983 * spa_minref (e.g. dataset or directory released during pool export),
984 * so the asserts in spa_close() do not apply.
987 spa_async_close(spa_t *spa, void *tag)
989 (void) refcount_remove(&spa->spa_refcount, tag);
993 * Check to see if the spa refcount is zero. Must be called with
994 * spa_namespace_lock held. We really compare against spa_minref, which is the
995 * number of references acquired when opening a pool
998 spa_refcount_zero(spa_t *spa)
1000 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1002 return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
1006 * ==========================================================================
1007 * SPA spare and l2cache tracking
1008 * ==========================================================================
1012 * Hot spares and cache devices are tracked using the same code below,
1013 * for 'auxiliary' devices.
1016 typedef struct spa_aux {
1024 spa_aux_compare(const void *a, const void *b)
1026 const spa_aux_t *sa = (const spa_aux_t *)a;
1027 const spa_aux_t *sb = (const spa_aux_t *)b;
1029 return (AVL_CMP(sa->aux_guid, sb->aux_guid));
1033 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
1039 search.aux_guid = vd->vdev_guid;
1040 if ((aux = avl_find(avl, &search, &where)) != NULL) {
1043 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
1044 aux->aux_guid = vd->vdev_guid;
1046 avl_insert(avl, aux, where);
1051 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
1057 search.aux_guid = vd->vdev_guid;
1058 aux = avl_find(avl, &search, &where);
1060 ASSERT(aux != NULL);
1062 if (--aux->aux_count == 0) {
1063 avl_remove(avl, aux);
1064 kmem_free(aux, sizeof (spa_aux_t));
1065 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
1066 aux->aux_pool = 0ULL;
1071 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
1073 spa_aux_t search, *found;
1075 search.aux_guid = guid;
1076 found = avl_find(avl, &search, NULL);
1080 *pool = found->aux_pool;
1087 *refcnt = found->aux_count;
1092 return (found != NULL);
1096 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
1098 spa_aux_t search, *found;
1101 search.aux_guid = vd->vdev_guid;
1102 found = avl_find(avl, &search, &where);
1103 ASSERT(found != NULL);
1104 ASSERT(found->aux_pool == 0ULL);
1106 found->aux_pool = spa_guid(vd->vdev_spa);
1110 * Spares are tracked globally due to the following constraints:
1112 * - A spare may be part of multiple pools.
1113 * - A spare may be added to a pool even if it's actively in use within
1115 * - A spare in use in any pool can only be the source of a replacement if
1116 * the target is a spare in the same pool.
1118 * We keep track of all spares on the system through the use of a reference
1119 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
1120 * spare, then we bump the reference count in the AVL tree. In addition, we set
1121 * the 'vdev_isspare' member to indicate that the device is a spare (active or
1122 * inactive). When a spare is made active (used to replace a device in the
1123 * pool), we also keep track of which pool its been made a part of.
1125 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
1126 * called under the spa_namespace lock as part of vdev reconfiguration. The
1127 * separate spare lock exists for the status query path, which does not need to
1128 * be completely consistent with respect to other vdev configuration changes.
1132 spa_spare_compare(const void *a, const void *b)
1134 return (spa_aux_compare(a, b));
1138 spa_spare_add(vdev_t *vd)
1140 mutex_enter(&spa_spare_lock);
1141 ASSERT(!vd->vdev_isspare);
1142 spa_aux_add(vd, &spa_spare_avl);
1143 vd->vdev_isspare = B_TRUE;
1144 mutex_exit(&spa_spare_lock);
1148 spa_spare_remove(vdev_t *vd)
1150 mutex_enter(&spa_spare_lock);
1151 ASSERT(vd->vdev_isspare);
1152 spa_aux_remove(vd, &spa_spare_avl);
1153 vd->vdev_isspare = B_FALSE;
1154 mutex_exit(&spa_spare_lock);
1158 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
1162 mutex_enter(&spa_spare_lock);
1163 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
1164 mutex_exit(&spa_spare_lock);
1170 spa_spare_activate(vdev_t *vd)
1172 mutex_enter(&spa_spare_lock);
1173 ASSERT(vd->vdev_isspare);
1174 spa_aux_activate(vd, &spa_spare_avl);
1175 mutex_exit(&spa_spare_lock);
1179 * Level 2 ARC devices are tracked globally for the same reasons as spares.
1180 * Cache devices currently only support one pool per cache device, and so
1181 * for these devices the aux reference count is currently unused beyond 1.
1185 spa_l2cache_compare(const void *a, const void *b)
1187 return (spa_aux_compare(a, b));
1191 spa_l2cache_add(vdev_t *vd)
1193 mutex_enter(&spa_l2cache_lock);
1194 ASSERT(!vd->vdev_isl2cache);
1195 spa_aux_add(vd, &spa_l2cache_avl);
1196 vd->vdev_isl2cache = B_TRUE;
1197 mutex_exit(&spa_l2cache_lock);
1201 spa_l2cache_remove(vdev_t *vd)
1203 mutex_enter(&spa_l2cache_lock);
1204 ASSERT(vd->vdev_isl2cache);
1205 spa_aux_remove(vd, &spa_l2cache_avl);
1206 vd->vdev_isl2cache = B_FALSE;
1207 mutex_exit(&spa_l2cache_lock);
1211 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
1215 mutex_enter(&spa_l2cache_lock);
1216 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
1217 mutex_exit(&spa_l2cache_lock);
1223 spa_l2cache_activate(vdev_t *vd)
1225 mutex_enter(&spa_l2cache_lock);
1226 ASSERT(vd->vdev_isl2cache);
1227 spa_aux_activate(vd, &spa_l2cache_avl);
1228 mutex_exit(&spa_l2cache_lock);
1232 * ==========================================================================
1234 * ==========================================================================
1238 * Lock the given spa_t for the purpose of adding or removing a vdev.
1239 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1240 * It returns the next transaction group for the spa_t.
1243 spa_vdev_enter(spa_t *spa)
1245 mutex_enter(&spa->spa_vdev_top_lock);
1246 mutex_enter(&spa_namespace_lock);
1247 return (spa_vdev_config_enter(spa));
1251 * Internal implementation for spa_vdev_enter(). Used when a vdev
1252 * operation requires multiple syncs (i.e. removing a device) while
1253 * keeping the spa_namespace_lock held.
1256 spa_vdev_config_enter(spa_t *spa)
1258 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1260 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1262 return (spa_last_synced_txg(spa) + 1);
1266 * Used in combination with spa_vdev_config_enter() to allow the syncing
1267 * of multiple transactions without releasing the spa_namespace_lock.
1270 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
1272 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1274 int config_changed = B_FALSE;
1276 ASSERT(txg > spa_last_synced_txg(spa));
1278 spa->spa_pending_vdev = NULL;
1281 * Reassess the DTLs.
1283 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
1285 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
1286 config_changed = B_TRUE;
1287 spa->spa_config_generation++;
1291 * Verify the metaslab classes.
1293 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
1294 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
1296 spa_config_exit(spa, SCL_ALL, spa);
1299 * Panic the system if the specified tag requires it. This
1300 * is useful for ensuring that configurations are updated
1303 if (zio_injection_enabled)
1304 zio_handle_panic_injection(spa, tag, 0);
1307 * Note: this txg_wait_synced() is important because it ensures
1308 * that there won't be more than one config change per txg.
1309 * This allows us to use the txg as the generation number.
1312 txg_wait_synced(spa->spa_dsl_pool, txg);
1315 ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
1316 if (vd->vdev_ops->vdev_op_leaf) {
1317 mutex_enter(&vd->vdev_initialize_lock);
1318 vdev_initialize_stop(vd, VDEV_INITIALIZE_CANCELED);
1319 mutex_exit(&vd->vdev_initialize_lock);
1322 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1324 spa_config_exit(spa, SCL_ALL, spa);
1328 * If the config changed, update the config cache.
1331 spa_write_cachefile(spa, B_FALSE, B_TRUE);
1335 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1336 * locking of spa_vdev_enter(), we also want make sure the transactions have
1337 * synced to disk, and then update the global configuration cache with the new
1341 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1343 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1344 mutex_exit(&spa_namespace_lock);
1345 mutex_exit(&spa->spa_vdev_top_lock);
1351 * Lock the given spa_t for the purpose of changing vdev state.
1354 spa_vdev_state_enter(spa_t *spa, int oplocks)
1356 int locks = SCL_STATE_ALL | oplocks;
1359 * Root pools may need to read of the underlying devfs filesystem
1360 * when opening up a vdev. Unfortunately if we're holding the
1361 * SCL_ZIO lock it will result in a deadlock when we try to issue
1362 * the read from the root filesystem. Instead we "prefetch"
1363 * the associated vnodes that we need prior to opening the
1364 * underlying devices and cache them so that we can prevent
1365 * any I/O when we are doing the actual open.
1367 if (spa_is_root(spa)) {
1368 int low = locks & ~(SCL_ZIO - 1);
1369 int high = locks & ~low;
1371 spa_config_enter(spa, high, spa, RW_WRITER);
1372 vdev_hold(spa->spa_root_vdev);
1373 spa_config_enter(spa, low, spa, RW_WRITER);
1375 spa_config_enter(spa, locks, spa, RW_WRITER);
1377 spa->spa_vdev_locks = locks;
1381 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1383 boolean_t config_changed = B_FALSE;
1385 if (vd != NULL || error == 0)
1386 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1390 vdev_state_dirty(vd->vdev_top);
1391 config_changed = B_TRUE;
1392 spa->spa_config_generation++;
1395 if (spa_is_root(spa))
1396 vdev_rele(spa->spa_root_vdev);
1398 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1399 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1402 * If anything changed, wait for it to sync. This ensures that,
1403 * from the system administrator's perspective, zpool(1M) commands
1404 * are synchronous. This is important for things like zpool offline:
1405 * when the command completes, you expect no further I/O from ZFS.
1408 txg_wait_synced(spa->spa_dsl_pool, 0);
1411 * If the config changed, update the config cache.
1413 if (config_changed) {
1414 mutex_enter(&spa_namespace_lock);
1415 spa_write_cachefile(spa, B_FALSE, B_TRUE);
1416 mutex_exit(&spa_namespace_lock);
1423 * ==========================================================================
1424 * Miscellaneous functions
1425 * ==========================================================================
1429 spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx)
1431 if (!nvlist_exists(spa->spa_label_features, feature)) {
1432 fnvlist_add_boolean(spa->spa_label_features, feature);
1434 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1435 * dirty the vdev config because lock SCL_CONFIG is not held.
1436 * Thankfully, in this case we don't need to dirty the config
1437 * because it will be written out anyway when we finish
1438 * creating the pool.
1440 if (tx->tx_txg != TXG_INITIAL)
1441 vdev_config_dirty(spa->spa_root_vdev);
1446 spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1448 if (nvlist_remove_all(spa->spa_label_features, feature) == 0)
1449 vdev_config_dirty(spa->spa_root_vdev);
1453 * Return the spa_t associated with given pool_guid, if it exists. If
1454 * device_guid is non-zero, determine whether the pool exists *and* contains
1455 * a device with the specified device_guid.
1458 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1461 avl_tree_t *t = &spa_namespace_avl;
1463 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1465 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1466 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1468 if (spa->spa_root_vdev == NULL)
1470 if (spa_guid(spa) == pool_guid) {
1471 if (device_guid == 0)
1474 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1475 device_guid) != NULL)
1479 * Check any devices we may be in the process of adding.
1481 if (spa->spa_pending_vdev) {
1482 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1483 device_guid) != NULL)
1493 * Determine whether a pool with the given pool_guid exists.
1496 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1498 return (spa_by_guid(pool_guid, device_guid) != NULL);
1502 spa_strdup(const char *s)
1508 new = kmem_alloc(len + 1, KM_SLEEP);
1516 spa_strfree(char *s)
1518 kmem_free(s, strlen(s) + 1);
1522 spa_get_random(uint64_t range)
1528 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1534 spa_generate_guid(spa_t *spa)
1536 uint64_t guid = spa_get_random(-1ULL);
1539 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1540 guid = spa_get_random(-1ULL);
1542 while (guid == 0 || spa_guid_exists(guid, 0))
1543 guid = spa_get_random(-1ULL);
1550 snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp)
1553 char *checksum = NULL;
1554 char *compress = NULL;
1557 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1558 dmu_object_byteswap_t bswap =
1559 DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1560 (void) snprintf(type, sizeof (type), "bswap %s %s",
1561 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1562 "metadata" : "data",
1563 dmu_ot_byteswap[bswap].ob_name);
1565 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1568 if (!BP_IS_EMBEDDED(bp)) {
1570 zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1572 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1575 SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum,
1580 spa_freeze(spa_t *spa)
1582 uint64_t freeze_txg = 0;
1584 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1585 if (spa->spa_freeze_txg == UINT64_MAX) {
1586 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1587 spa->spa_freeze_txg = freeze_txg;
1589 spa_config_exit(spa, SCL_ALL, FTAG);
1590 if (freeze_txg != 0)
1591 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1595 zfs_panic_recover(const char *fmt, ...)
1600 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1605 * This is a stripped-down version of strtoull, suitable only for converting
1606 * lowercase hexadecimal numbers that don't overflow.
1609 zfs_strtonum(const char *str, char **nptr)
1615 while ((c = *str) != '\0') {
1616 if (c >= '0' && c <= '9')
1618 else if (c >= 'a' && c <= 'f')
1619 digit = 10 + c - 'a';
1630 *nptr = (char *)str;
1636 * ==========================================================================
1637 * Accessor functions
1638 * ==========================================================================
1642 spa_shutting_down(spa_t *spa)
1644 return (spa->spa_async_suspended);
1648 spa_get_dsl(spa_t *spa)
1650 return (spa->spa_dsl_pool);
1654 spa_is_initializing(spa_t *spa)
1656 return (spa->spa_is_initializing);
1660 spa_indirect_vdevs_loaded(spa_t *spa)
1662 return (spa->spa_indirect_vdevs_loaded);
1666 spa_get_rootblkptr(spa_t *spa)
1668 return (&spa->spa_ubsync.ub_rootbp);
1672 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1674 spa->spa_uberblock.ub_rootbp = *bp;
1678 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1680 if (spa->spa_root == NULL)
1683 (void) strncpy(buf, spa->spa_root, buflen);
1687 spa_sync_pass(spa_t *spa)
1689 return (spa->spa_sync_pass);
1693 spa_name(spa_t *spa)
1695 return (spa->spa_name);
1699 spa_guid(spa_t *spa)
1701 dsl_pool_t *dp = spa_get_dsl(spa);
1705 * If we fail to parse the config during spa_load(), we can go through
1706 * the error path (which posts an ereport) and end up here with no root
1707 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1710 if (spa->spa_root_vdev == NULL)
1711 return (spa->spa_config_guid);
1713 guid = spa->spa_last_synced_guid != 0 ?
1714 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1717 * Return the most recently synced out guid unless we're
1718 * in syncing context.
1720 if (dp && dsl_pool_sync_context(dp))
1721 return (spa->spa_root_vdev->vdev_guid);
1727 spa_load_guid(spa_t *spa)
1730 * This is a GUID that exists solely as a reference for the
1731 * purposes of the arc. It is generated at load time, and
1732 * is never written to persistent storage.
1734 return (spa->spa_load_guid);
1738 spa_last_synced_txg(spa_t *spa)
1740 return (spa->spa_ubsync.ub_txg);
1744 spa_first_txg(spa_t *spa)
1746 return (spa->spa_first_txg);
1750 spa_syncing_txg(spa_t *spa)
1752 return (spa->spa_syncing_txg);
1756 * Return the last txg where data can be dirtied. The final txgs
1757 * will be used to just clear out any deferred frees that remain.
1760 spa_final_dirty_txg(spa_t *spa)
1762 return (spa->spa_final_txg - TXG_DEFER_SIZE);
1766 spa_state(spa_t *spa)
1768 return (spa->spa_state);
1772 spa_load_state(spa_t *spa)
1774 return (spa->spa_load_state);
1778 spa_freeze_txg(spa_t *spa)
1780 return (spa->spa_freeze_txg);
1785 spa_get_worst_case_asize(spa_t *spa, uint64_t lsize)
1787 return (lsize * spa_asize_inflation);
1791 * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%),
1792 * or at least 128MB, unless that would cause it to be more than half the
1795 * See the comment above spa_slop_shift for details.
1798 spa_get_slop_space(spa_t *spa)
1800 uint64_t space = spa_get_dspace(spa);
1801 return (MAX(space >> spa_slop_shift, MIN(space >> 1, spa_min_slop)));
1805 spa_get_dspace(spa_t *spa)
1807 return (spa->spa_dspace);
1811 spa_get_checkpoint_space(spa_t *spa)
1813 return (spa->spa_checkpoint_info.sci_dspace);
1817 spa_update_dspace(spa_t *spa)
1819 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1820 ddt_get_dedup_dspace(spa);
1821 if (spa->spa_vdev_removal != NULL) {
1823 * We can't allocate from the removing device, so
1824 * subtract its size. This prevents the DMU/DSL from
1825 * filling up the (now smaller) pool while we are in the
1826 * middle of removing the device.
1828 * Note that the DMU/DSL doesn't actually know or care
1829 * how much space is allocated (it does its own tracking
1830 * of how much space has been logically used). So it
1831 * doesn't matter that the data we are moving may be
1832 * allocated twice (on the old device and the new
1835 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1837 vdev_lookup_top(spa, spa->spa_vdev_removal->svr_vdev_id);
1838 spa->spa_dspace -= spa_deflate(spa) ?
1839 vd->vdev_stat.vs_dspace : vd->vdev_stat.vs_space;
1840 spa_config_exit(spa, SCL_VDEV, FTAG);
1845 * Return the failure mode that has been set to this pool. The default
1846 * behavior will be to block all I/Os when a complete failure occurs.
1849 spa_get_failmode(spa_t *spa)
1851 return (spa->spa_failmode);
1855 spa_suspended(spa_t *spa)
1857 return (spa->spa_suspended);
1861 spa_version(spa_t *spa)
1863 return (spa->spa_ubsync.ub_version);
1867 spa_deflate(spa_t *spa)
1869 return (spa->spa_deflate);
1873 spa_normal_class(spa_t *spa)
1875 return (spa->spa_normal_class);
1879 spa_log_class(spa_t *spa)
1881 return (spa->spa_log_class);
1885 spa_evicting_os_register(spa_t *spa, objset_t *os)
1887 mutex_enter(&spa->spa_evicting_os_lock);
1888 list_insert_head(&spa->spa_evicting_os_list, os);
1889 mutex_exit(&spa->spa_evicting_os_lock);
1893 spa_evicting_os_deregister(spa_t *spa, objset_t *os)
1895 mutex_enter(&spa->spa_evicting_os_lock);
1896 list_remove(&spa->spa_evicting_os_list, os);
1897 cv_broadcast(&spa->spa_evicting_os_cv);
1898 mutex_exit(&spa->spa_evicting_os_lock);
1902 spa_evicting_os_wait(spa_t *spa)
1904 mutex_enter(&spa->spa_evicting_os_lock);
1905 while (!list_is_empty(&spa->spa_evicting_os_list))
1906 cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock);
1907 mutex_exit(&spa->spa_evicting_os_lock);
1909 dmu_buf_user_evict_wait();
1913 spa_max_replication(spa_t *spa)
1916 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1917 * handle BPs with more than one DVA allocated. Set our max
1918 * replication level accordingly.
1920 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1922 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1926 spa_prev_software_version(spa_t *spa)
1928 return (spa->spa_prev_software_version);
1932 spa_deadman_synctime(spa_t *spa)
1934 return (spa->spa_deadman_synctime);
1938 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1940 uint64_t asize = DVA_GET_ASIZE(dva);
1941 uint64_t dsize = asize;
1943 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1945 if (asize != 0 && spa->spa_deflate) {
1946 uint64_t vdev = DVA_GET_VDEV(dva);
1947 vdev_t *vd = vdev_lookup_top(spa, vdev);
1950 "dva_get_dsize_sync(): bad DVA %llu:%llu",
1951 (u_longlong_t)vdev, (u_longlong_t)asize);
1953 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1960 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1964 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1965 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1971 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1975 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1977 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1978 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1980 spa_config_exit(spa, SCL_VDEV, FTAG);
1986 spa_dirty_data(spa_t *spa)
1988 return (spa->spa_dsl_pool->dp_dirty_total);
1992 * ==========================================================================
1993 * Initialization and Termination
1994 * ==========================================================================
1998 spa_name_compare(const void *a1, const void *a2)
2000 const spa_t *s1 = a1;
2001 const spa_t *s2 = a2;
2004 s = strcmp(s1->spa_name, s2->spa_name);
2006 return (AVL_ISIGN(s));
2012 return (spa_active_count);
2022 EVENTHANDLER_DEFINE(mountroot, spa_boot_init, NULL, 0);
2028 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
2029 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
2030 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
2031 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
2033 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
2034 offsetof(spa_t, spa_avl));
2036 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
2037 offsetof(spa_aux_t, aux_avl));
2039 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
2040 offsetof(spa_aux_t, aux_avl));
2042 spa_mode_global = mode;
2048 if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
2049 arc_procfd = open("/proc/self/ctl", O_WRONLY);
2050 if (arc_procfd == -1) {
2051 perror("could not enable watchpoints: "
2052 "opening /proc/self/ctl failed: ");
2058 #endif /* illumos */
2062 metaslab_alloc_trace_init();
2067 vdev_cache_stat_init();
2071 zpool_feature_init();
2075 dsl_scan_global_init();
2080 #endif /* !illumos */
2091 vdev_cache_stat_fini();
2096 metaslab_alloc_trace_fini();
2102 avl_destroy(&spa_namespace_avl);
2103 avl_destroy(&spa_spare_avl);
2104 avl_destroy(&spa_l2cache_avl);
2106 cv_destroy(&spa_namespace_cv);
2107 mutex_destroy(&spa_namespace_lock);
2108 mutex_destroy(&spa_spare_lock);
2109 mutex_destroy(&spa_l2cache_lock);
2113 * Return whether this pool has slogs. No locking needed.
2114 * It's not a problem if the wrong answer is returned as it's only for
2115 * performance and not correctness
2118 spa_has_slogs(spa_t *spa)
2120 return (spa->spa_log_class->mc_rotor != NULL);
2124 spa_get_log_state(spa_t *spa)
2126 return (spa->spa_log_state);
2130 spa_set_log_state(spa_t *spa, spa_log_state_t state)
2132 spa->spa_log_state = state;
2136 spa_is_root(spa_t *spa)
2138 return (spa->spa_is_root);
2142 spa_writeable(spa_t *spa)
2144 return (!!(spa->spa_mode & FWRITE) && spa->spa_trust_config);
2148 * Returns true if there is a pending sync task in any of the current
2149 * syncing txg, the current quiescing txg, or the current open txg.
2152 spa_has_pending_synctask(spa_t *spa)
2154 return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks) ||
2155 !txg_all_lists_empty(&spa->spa_dsl_pool->dp_early_sync_tasks));
2159 spa_mode(spa_t *spa)
2161 return (spa->spa_mode);
2165 spa_bootfs(spa_t *spa)
2167 return (spa->spa_bootfs);
2171 spa_delegation(spa_t *spa)
2173 return (spa->spa_delegation);
2177 spa_meta_objset(spa_t *spa)
2179 return (spa->spa_meta_objset);
2183 spa_dedup_checksum(spa_t *spa)
2185 return (spa->spa_dedup_checksum);
2189 * Reset pool scan stat per scan pass (or reboot).
2192 spa_scan_stat_init(spa_t *spa)
2194 /* data not stored on disk */
2195 spa->spa_scan_pass_start = gethrestime_sec();
2196 if (dsl_scan_is_paused_scrub(spa->spa_dsl_pool->dp_scan))
2197 spa->spa_scan_pass_scrub_pause = spa->spa_scan_pass_start;
2199 spa->spa_scan_pass_scrub_pause = 0;
2200 spa->spa_scan_pass_scrub_spent_paused = 0;
2201 spa->spa_scan_pass_exam = 0;
2202 spa->spa_scan_pass_issued = 0;
2203 vdev_scan_stat_init(spa->spa_root_vdev);
2207 * Get scan stats for zpool status reports
2210 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
2212 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
2214 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
2215 return (SET_ERROR(ENOENT));
2216 bzero(ps, sizeof (pool_scan_stat_t));
2218 /* data stored on disk */
2219 ps->pss_func = scn->scn_phys.scn_func;
2220 ps->pss_state = scn->scn_phys.scn_state;
2221 ps->pss_start_time = scn->scn_phys.scn_start_time;
2222 ps->pss_end_time = scn->scn_phys.scn_end_time;
2223 ps->pss_to_examine = scn->scn_phys.scn_to_examine;
2224 ps->pss_to_process = scn->scn_phys.scn_to_process;
2225 ps->pss_processed = scn->scn_phys.scn_processed;
2226 ps->pss_errors = scn->scn_phys.scn_errors;
2227 ps->pss_examined = scn->scn_phys.scn_examined;
2229 scn->scn_issued_before_pass + spa->spa_scan_pass_issued;
2230 /* data not stored on disk */
2231 ps->pss_pass_start = spa->spa_scan_pass_start;
2232 ps->pss_pass_exam = spa->spa_scan_pass_exam;
2233 ps->pss_pass_issued = spa->spa_scan_pass_issued;
2234 ps->pss_pass_scrub_pause = spa->spa_scan_pass_scrub_pause;
2235 ps->pss_pass_scrub_spent_paused = spa->spa_scan_pass_scrub_spent_paused;
2241 spa_maxblocksize(spa_t *spa)
2243 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS))
2244 return (SPA_MAXBLOCKSIZE);
2246 return (SPA_OLD_MAXBLOCKSIZE);
2250 spa_maxdnodesize(spa_t *spa)
2252 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE))
2253 return (DNODE_MAX_SIZE);
2255 return (DNODE_MIN_SIZE);
2260 * Returns the txg that the last device removal completed. No indirect mappings
2261 * have been added since this txg.
2264 spa_get_last_removal_txg(spa_t *spa)
2267 uint64_t ret = -1ULL;
2269 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2271 * sr_prev_indirect_vdev is only modified while holding all the
2272 * config locks, so it is sufficient to hold SCL_VDEV as reader when
2275 vdevid = spa->spa_removing_phys.sr_prev_indirect_vdev;
2277 while (vdevid != -1ULL) {
2278 vdev_t *vd = vdev_lookup_top(spa, vdevid);
2279 vdev_indirect_births_t *vib = vd->vdev_indirect_births;
2281 ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops);
2284 * If the removal did not remap any data, we don't care.
2286 if (vdev_indirect_births_count(vib) != 0) {
2287 ret = vdev_indirect_births_last_entry_txg(vib);
2291 vdevid = vd->vdev_indirect_config.vic_prev_indirect_vdev;
2293 spa_config_exit(spa, SCL_VDEV, FTAG);
2296 spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL));
2302 spa_trust_config(spa_t *spa)
2304 return (spa->spa_trust_config);
2308 spa_missing_tvds_allowed(spa_t *spa)
2310 return (spa->spa_missing_tvds_allowed);
2314 spa_set_missing_tvds(spa_t *spa, uint64_t missing)
2316 spa->spa_missing_tvds = missing;
2320 spa_top_vdevs_spacemap_addressable(spa_t *spa)
2322 vdev_t *rvd = spa->spa_root_vdev;
2323 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
2324 if (!vdev_is_spacemap_addressable(rvd->vdev_child[c]))
2331 spa_has_checkpoint(spa_t *spa)
2333 return (spa->spa_checkpoint_txg != 0);
2337 spa_importing_readonly_checkpoint(spa_t *spa)
2339 return ((spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT) &&
2340 spa->spa_mode == FREAD);
2344 spa_min_claim_txg(spa_t *spa)
2346 uint64_t checkpoint_txg = spa->spa_uberblock.ub_checkpoint_txg;
2348 if (checkpoint_txg != 0)
2349 return (checkpoint_txg + 1);
2351 return (spa->spa_first_txg);
2355 * If there is a checkpoint, async destroys may consume more space from
2356 * the pool instead of freeing it. In an attempt to save the pool from
2357 * getting suspended when it is about to run out of space, we stop
2358 * processing async destroys.
2361 spa_suspend_async_destroy(spa_t *spa)
2363 dsl_pool_t *dp = spa_get_dsl(spa);
2365 uint64_t unreserved = dsl_pool_unreserved_space(dp,
2366 ZFS_SPACE_CHECK_EXTRA_RESERVED);
2367 uint64_t used = dsl_dir_phys(dp->dp_root_dir)->dd_used_bytes;
2368 uint64_t avail = (unreserved > used) ? (unreserved - used) : 0;
2370 if (spa_has_checkpoint(spa) && avail == 0)