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, 2015 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
28 #include <sys/zfs_context.h>
29 #include <sys/spa_impl.h>
30 #include <sys/spa_boot.h>
32 #include <sys/zio_checksum.h>
33 #include <sys/zio_compress.h>
35 #include <sys/dmu_tx.h>
38 #include <sys/vdev_impl.h>
39 #include <sys/metaslab.h>
40 #include <sys/uberblock_impl.h>
43 #include <sys/unique.h>
44 #include <sys/dsl_pool.h>
45 #include <sys/dsl_dir.h>
46 #include <sys/dsl_prop.h>
47 #include <sys/dsl_scan.h>
48 #include <sys/fs/zfs.h>
49 #include <sys/metaslab_impl.h>
53 #include "zfeature_common.h"
58 * There are four basic locks for managing spa_t structures:
60 * spa_namespace_lock (global mutex)
62 * This lock must be acquired to do any of the following:
64 * - Lookup a spa_t by name
65 * - Add or remove a spa_t from the namespace
66 * - Increase spa_refcount from non-zero
67 * - Check if spa_refcount is zero
69 * - add/remove/attach/detach devices
70 * - Held for the duration of create/destroy/import/export
72 * It does not need to handle recursion. A create or destroy may
73 * reference objects (files or zvols) in other pools, but by
74 * definition they must have an existing reference, and will never need
75 * to lookup a spa_t by name.
77 * spa_refcount (per-spa refcount_t protected by mutex)
79 * This reference count keep track of any active users of the spa_t. The
80 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
81 * the refcount is never really 'zero' - opening a pool implicitly keeps
82 * some references in the DMU. Internally we check against spa_minref, but
83 * present the image of a zero/non-zero value to consumers.
85 * spa_config_lock[] (per-spa array of rwlocks)
87 * This protects the spa_t from config changes, and must be held in
88 * the following circumstances:
90 * - RW_READER to perform I/O to the spa
91 * - RW_WRITER to change the vdev config
93 * The locking order is fairly straightforward:
95 * spa_namespace_lock -> spa_refcount
97 * The namespace lock must be acquired to increase the refcount from 0
98 * or to check if it is zero.
100 * spa_refcount -> spa_config_lock[]
102 * There must be at least one valid reference on the spa_t to acquire
105 * spa_namespace_lock -> spa_config_lock[]
107 * The namespace lock must always be taken before the config lock.
110 * The spa_namespace_lock can be acquired directly and is globally visible.
112 * The namespace is manipulated using the following functions, all of which
113 * require the spa_namespace_lock to be held.
115 * spa_lookup() Lookup a spa_t by name.
117 * spa_add() Create a new spa_t in the namespace.
119 * spa_remove() Remove a spa_t from the namespace. This also
120 * frees up any memory associated with the spa_t.
122 * spa_next() Returns the next spa_t in the system, or the
123 * first if NULL is passed.
125 * spa_evict_all() Shutdown and remove all spa_t structures in
128 * spa_guid_exists() Determine whether a pool/device guid exists.
130 * The spa_refcount is manipulated using the following functions:
132 * spa_open_ref() Adds a reference to the given spa_t. Must be
133 * called with spa_namespace_lock held if the
134 * refcount is currently zero.
136 * spa_close() Remove a reference from the spa_t. This will
137 * not free the spa_t or remove it from the
138 * namespace. No locking is required.
140 * spa_refcount_zero() Returns true if the refcount is currently
141 * zero. Must be called with spa_namespace_lock
144 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
145 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
146 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
148 * To read the configuration, it suffices to hold one of these locks as reader.
149 * To modify the configuration, you must hold all locks as writer. To modify
150 * vdev state without altering the vdev tree's topology (e.g. online/offline),
151 * you must hold SCL_STATE and SCL_ZIO as writer.
153 * We use these distinct config locks to avoid recursive lock entry.
154 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
155 * block allocations (SCL_ALLOC), which may require reading space maps
156 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
158 * The spa config locks cannot be normal rwlocks because we need the
159 * ability to hand off ownership. For example, SCL_ZIO is acquired
160 * by the issuing thread and later released by an interrupt thread.
161 * They do, however, obey the usual write-wanted semantics to prevent
162 * writer (i.e. system administrator) starvation.
164 * The lock acquisition rules are as follows:
167 * Protects changes to the vdev tree topology, such as vdev
168 * add/remove/attach/detach. Protects the dirty config list
169 * (spa_config_dirty_list) and the set of spares and l2arc devices.
172 * Protects changes to pool state and vdev state, such as vdev
173 * online/offline/fault/degrade/clear. Protects the dirty state list
174 * (spa_state_dirty_list) and global pool state (spa_state).
177 * Protects changes to metaslab groups and classes.
178 * Held as reader by metaslab_alloc() and metaslab_claim().
181 * Held by bp-level zios (those which have no io_vd upon entry)
182 * to prevent changes to the vdev tree. The bp-level zio implicitly
183 * protects all of its vdev child zios, which do not hold SCL_ZIO.
186 * Protects changes to metaslab groups and classes.
187 * Held as reader by metaslab_free(). SCL_FREE is distinct from
188 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
189 * blocks in zio_done() while another i/o that holds either
190 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
193 * Held as reader to prevent changes to the vdev tree during trivial
194 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
195 * other locks, and lower than all of them, to ensure that it's safe
196 * to acquire regardless of caller context.
198 * In addition, the following rules apply:
200 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
201 * The lock ordering is SCL_CONFIG > spa_props_lock.
203 * (b) I/O operations on leaf vdevs. For any zio operation that takes
204 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
205 * or zio_write_phys() -- the caller must ensure that the config cannot
206 * cannot change in the interim, and that the vdev cannot be reopened.
207 * SCL_STATE as reader suffices for both.
209 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
211 * spa_vdev_enter() Acquire the namespace lock and the config lock
214 * spa_vdev_exit() Release the config lock, wait for all I/O
215 * to complete, sync the updated configs to the
216 * cache, and release the namespace lock.
218 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
219 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
220 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
222 * spa_rename() is also implemented within this file since it requires
223 * manipulation of the namespace.
226 static avl_tree_t spa_namespace_avl;
227 kmutex_t spa_namespace_lock;
228 static kcondvar_t spa_namespace_cv;
229 static int spa_active_count;
230 int spa_max_replication_override = SPA_DVAS_PER_BP;
232 static kmutex_t spa_spare_lock;
233 static avl_tree_t spa_spare_avl;
234 static kmutex_t spa_l2cache_lock;
235 static avl_tree_t spa_l2cache_avl;
237 kmem_cache_t *spa_buffer_pool;
241 /* Everything except dprintf and spa is on by default in debug builds */
242 int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA);
247 TUNABLE_INT("debug.zfs_flags", &zfs_flags);
248 SYSCTL_INT(_debug, OID_AUTO, zfs_flags, CTLFLAG_RWTUN, &zfs_flags, 0,
252 * zfs_recover can be set to nonzero to attempt to recover from
253 * otherwise-fatal errors, typically caused by on-disk corruption. When
254 * set, calls to zfs_panic_recover() will turn into warning messages.
255 * This should only be used as a last resort, as it typically results
256 * in leaked space, or worse.
258 boolean_t zfs_recover = B_FALSE;
259 SYSCTL_DECL(_vfs_zfs);
260 TUNABLE_INT("vfs.zfs.recover", &zfs_recover);
261 SYSCTL_INT(_vfs_zfs, OID_AUTO, recover, CTLFLAG_RDTUN, &zfs_recover, 0,
262 "Try to recover from otherwise-fatal errors.");
265 * If destroy encounters an EIO while reading metadata (e.g. indirect
266 * blocks), space referenced by the missing metadata can not be freed.
267 * Normally this causes the background destroy to become "stalled", as
268 * it is unable to make forward progress. While in this stalled state,
269 * all remaining space to free from the error-encountering filesystem is
270 * "temporarily leaked". Set this flag to cause it to ignore the EIO,
271 * permanently leak the space from indirect blocks that can not be read,
272 * and continue to free everything else that it can.
274 * The default, "stalling" behavior is useful if the storage partially
275 * fails (i.e. some but not all i/os fail), and then later recovers. In
276 * this case, we will be able to continue pool operations while it is
277 * partially failed, and when it recovers, we can continue to free the
278 * space, with no leaks. However, note that this case is actually
281 * Typically pools either (a) fail completely (but perhaps temporarily,
282 * e.g. a top-level vdev going offline), or (b) have localized,
283 * permanent errors (e.g. disk returns the wrong data due to bit flip or
284 * firmware bug). In case (a), this setting does not matter because the
285 * pool will be suspended and the sync thread will not be able to make
286 * forward progress regardless. In case (b), because the error is
287 * permanent, the best we can do is leak the minimum amount of space,
288 * which is what setting this flag will do. Therefore, it is reasonable
289 * for this flag to normally be set, but we chose the more conservative
290 * approach of not setting it, so that there is no possibility of
291 * leaking space in the "partial temporary" failure case.
293 boolean_t zfs_free_leak_on_eio = B_FALSE;
296 * Expiration time in milliseconds. This value has two meanings. First it is
297 * used to determine when the spa_deadman() logic should fire. By default the
298 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
299 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
300 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
303 uint64_t zfs_deadman_synctime_ms = 1000000ULL;
304 TUNABLE_QUAD("vfs.zfs.deadman_synctime_ms", &zfs_deadman_synctime_ms);
305 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_synctime_ms, CTLFLAG_RDTUN,
306 &zfs_deadman_synctime_ms, 0,
307 "Stalled ZFS I/O expiration time in milliseconds");
310 * Check time in milliseconds. This defines the frequency at which we check
313 uint64_t zfs_deadman_checktime_ms = 5000ULL;
314 TUNABLE_QUAD("vfs.zfs.deadman_checktime_ms", &zfs_deadman_checktime_ms);
315 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_checktime_ms, CTLFLAG_RDTUN,
316 &zfs_deadman_checktime_ms, 0,
317 "Period of checks for stalled ZFS I/O in milliseconds");
320 * Default value of -1 for zfs_deadman_enabled is resolved in
323 int zfs_deadman_enabled = -1;
324 TUNABLE_INT("vfs.zfs.deadman_enabled", &zfs_deadman_enabled);
325 SYSCTL_INT(_vfs_zfs, OID_AUTO, deadman_enabled, CTLFLAG_RDTUN,
326 &zfs_deadman_enabled, 0, "Kernel panic on stalled ZFS I/O");
329 * The worst case is single-sector max-parity RAID-Z blocks, in which
330 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
331 * times the size; so just assume that. Add to this the fact that
332 * we can have up to 3 DVAs per bp, and one more factor of 2 because
333 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together,
335 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
337 int spa_asize_inflation = 24;
338 TUNABLE_INT("vfs.zfs.spa_asize_inflation", &spa_asize_inflation);
339 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_asize_inflation, CTLFLAG_RWTUN,
340 &spa_asize_inflation, 0, "Worst case inflation factor for single sector writes");
348 * If we are not i386 or amd64 or in a virtual machine,
349 * disable ZFS deadman thread by default
351 if (zfs_deadman_enabled == -1) {
352 #if defined(__amd64__) || defined(__i386__)
353 zfs_deadman_enabled = (vm_guest == VM_GUEST_NO) ? 1 : 0;
355 zfs_deadman_enabled = 0;
360 #endif /* !illumos */
363 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
364 * the pool to be consumed. This ensures that we don't run the pool
365 * completely out of space, due to unaccounted changes (e.g. to the MOS).
366 * It also limits the worst-case time to allocate space. If we have
367 * less than this amount of free space, most ZPL operations (e.g. write,
368 * create) will return ENOSPC.
370 * Certain operations (e.g. file removal, most administrative actions) can
371 * use half the slop space. They will only return ENOSPC if less than half
372 * the slop space is free. Typically, once the pool has less than the slop
373 * space free, the user will use these operations to free up space in the pool.
374 * These are the operations that call dsl_pool_adjustedsize() with the netfree
375 * argument set to TRUE.
377 * A very restricted set of operations are always permitted, regardless of
378 * the amount of free space. These are the operations that call
379 * dsl_sync_task(ZFS_SPACE_CHECK_NONE), e.g. "zfs destroy". If these
380 * operations result in a net increase in the amount of space used,
381 * it is possible to run the pool completely out of space, causing it to
382 * be permanently read-only.
384 * See also the comments in zfs_space_check_t.
386 int spa_slop_shift = 5;
387 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_slop_shift, CTLFLAG_RWTUN,
389 "Shift value of reserved space (1/(2^spa_slop_shift)).");
392 * ==========================================================================
394 * ==========================================================================
397 spa_config_lock_init(spa_t *spa)
399 for (int i = 0; i < SCL_LOCKS; i++) {
400 spa_config_lock_t *scl = &spa->spa_config_lock[i];
401 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
402 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
403 refcount_create_untracked(&scl->scl_count);
404 scl->scl_writer = NULL;
405 scl->scl_write_wanted = 0;
410 spa_config_lock_destroy(spa_t *spa)
412 for (int i = 0; i < SCL_LOCKS; i++) {
413 spa_config_lock_t *scl = &spa->spa_config_lock[i];
414 mutex_destroy(&scl->scl_lock);
415 cv_destroy(&scl->scl_cv);
416 refcount_destroy(&scl->scl_count);
417 ASSERT(scl->scl_writer == NULL);
418 ASSERT(scl->scl_write_wanted == 0);
423 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
425 for (int i = 0; i < SCL_LOCKS; i++) {
426 spa_config_lock_t *scl = &spa->spa_config_lock[i];
427 if (!(locks & (1 << i)))
429 mutex_enter(&scl->scl_lock);
430 if (rw == RW_READER) {
431 if (scl->scl_writer || scl->scl_write_wanted) {
432 mutex_exit(&scl->scl_lock);
433 spa_config_exit(spa, locks ^ (1 << i), tag);
437 ASSERT(scl->scl_writer != curthread);
438 if (!refcount_is_zero(&scl->scl_count)) {
439 mutex_exit(&scl->scl_lock);
440 spa_config_exit(spa, locks ^ (1 << i), tag);
443 scl->scl_writer = curthread;
445 (void) refcount_add(&scl->scl_count, tag);
446 mutex_exit(&scl->scl_lock);
452 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
456 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
458 for (int i = 0; i < SCL_LOCKS; i++) {
459 spa_config_lock_t *scl = &spa->spa_config_lock[i];
460 if (scl->scl_writer == curthread)
461 wlocks_held |= (1 << i);
462 if (!(locks & (1 << i)))
464 mutex_enter(&scl->scl_lock);
465 if (rw == RW_READER) {
466 while (scl->scl_writer || scl->scl_write_wanted) {
467 cv_wait(&scl->scl_cv, &scl->scl_lock);
470 ASSERT(scl->scl_writer != curthread);
471 while (!refcount_is_zero(&scl->scl_count)) {
472 scl->scl_write_wanted++;
473 cv_wait(&scl->scl_cv, &scl->scl_lock);
474 scl->scl_write_wanted--;
476 scl->scl_writer = curthread;
478 (void) refcount_add(&scl->scl_count, tag);
479 mutex_exit(&scl->scl_lock);
481 ASSERT(wlocks_held <= locks);
485 spa_config_exit(spa_t *spa, int locks, void *tag)
487 for (int i = SCL_LOCKS - 1; i >= 0; i--) {
488 spa_config_lock_t *scl = &spa->spa_config_lock[i];
489 if (!(locks & (1 << i)))
491 mutex_enter(&scl->scl_lock);
492 ASSERT(!refcount_is_zero(&scl->scl_count));
493 if (refcount_remove(&scl->scl_count, tag) == 0) {
494 ASSERT(scl->scl_writer == NULL ||
495 scl->scl_writer == curthread);
496 scl->scl_writer = NULL; /* OK in either case */
497 cv_broadcast(&scl->scl_cv);
499 mutex_exit(&scl->scl_lock);
504 spa_config_held(spa_t *spa, int locks, krw_t rw)
508 for (int i = 0; i < SCL_LOCKS; i++) {
509 spa_config_lock_t *scl = &spa->spa_config_lock[i];
510 if (!(locks & (1 << i)))
512 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
513 (rw == RW_WRITER && scl->scl_writer == curthread))
514 locks_held |= 1 << i;
521 * ==========================================================================
522 * SPA namespace functions
523 * ==========================================================================
527 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
528 * Returns NULL if no matching spa_t is found.
531 spa_lookup(const char *name)
533 static spa_t search; /* spa_t is large; don't allocate on stack */
538 ASSERT(MUTEX_HELD(&spa_namespace_lock));
540 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
543 * If it's a full dataset name, figure out the pool name and
546 cp = strpbrk(search.spa_name, "/@#");
550 spa = avl_find(&spa_namespace_avl, &search, &where);
556 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
557 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
558 * looking for potentially hung I/Os.
561 spa_deadman(void *arg)
566 * Disable the deadman timer if the pool is suspended.
568 if (spa_suspended(spa)) {
570 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
572 /* Nothing. just don't schedule any future callouts. */
577 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
578 (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
579 ++spa->spa_deadman_calls);
580 if (zfs_deadman_enabled)
581 vdev_deadman(spa->spa_root_vdev);
585 * Create an uninitialized spa_t with the given name. Requires
586 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
587 * exist by calling spa_lookup() first.
590 spa_add(const char *name, nvlist_t *config, const char *altroot)
593 spa_config_dirent_t *dp;
599 ASSERT(MUTEX_HELD(&spa_namespace_lock));
601 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
603 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
604 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
605 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
606 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
607 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
608 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
609 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
610 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
611 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
613 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
614 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
615 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
616 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
618 for (int t = 0; t < TXG_SIZE; t++)
619 bplist_create(&spa->spa_free_bplist[t]);
621 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
622 spa->spa_state = POOL_STATE_UNINITIALIZED;
623 spa->spa_freeze_txg = UINT64_MAX;
624 spa->spa_final_txg = UINT64_MAX;
625 spa->spa_load_max_txg = UINT64_MAX;
627 spa->spa_proc_state = SPA_PROC_NONE;
630 hdlr.cyh_func = spa_deadman;
632 hdlr.cyh_level = CY_LOW_LEVEL;
635 spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
639 * This determines how often we need to check for hung I/Os after
640 * the cyclic has already fired. Since checking for hung I/Os is
641 * an expensive operation we don't want to check too frequently.
642 * Instead wait for 5 seconds before checking again.
644 when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms);
645 when.cyt_when = CY_INFINITY;
646 mutex_enter(&cpu_lock);
647 spa->spa_deadman_cycid = cyclic_add(&hdlr, &when);
648 mutex_exit(&cpu_lock);
651 callout_init(&spa->spa_deadman_cycid, CALLOUT_MPSAFE);
654 refcount_create(&spa->spa_refcount);
655 spa_config_lock_init(spa);
657 avl_add(&spa_namespace_avl, spa);
660 * Set the alternate root, if there is one.
663 spa->spa_root = spa_strdup(altroot);
668 * Every pool starts with the default cachefile
670 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
671 offsetof(spa_config_dirent_t, scd_link));
673 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
674 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
675 list_insert_head(&spa->spa_config_list, dp);
677 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
680 if (config != NULL) {
683 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
685 VERIFY(nvlist_dup(features, &spa->spa_label_features,
689 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
692 if (spa->spa_label_features == NULL) {
693 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
697 spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0);
699 spa->spa_min_ashift = INT_MAX;
700 spa->spa_max_ashift = 0;
703 * As a pool is being created, treat all features as disabled by
704 * setting SPA_FEATURE_DISABLED for all entries in the feature
707 for (int i = 0; i < SPA_FEATURES; i++) {
708 spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED;
715 * Removes a spa_t from the namespace, freeing up any memory used. Requires
716 * spa_namespace_lock. This is called only after the spa_t has been closed and
720 spa_remove(spa_t *spa)
722 spa_config_dirent_t *dp;
724 ASSERT(MUTEX_HELD(&spa_namespace_lock));
725 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
727 nvlist_free(spa->spa_config_splitting);
729 avl_remove(&spa_namespace_avl, spa);
730 cv_broadcast(&spa_namespace_cv);
733 spa_strfree(spa->spa_root);
737 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
738 list_remove(&spa->spa_config_list, dp);
739 if (dp->scd_path != NULL)
740 spa_strfree(dp->scd_path);
741 kmem_free(dp, sizeof (spa_config_dirent_t));
744 list_destroy(&spa->spa_config_list);
746 nvlist_free(spa->spa_label_features);
747 nvlist_free(spa->spa_load_info);
748 spa_config_set(spa, NULL);
751 mutex_enter(&cpu_lock);
752 if (spa->spa_deadman_cycid != CYCLIC_NONE)
753 cyclic_remove(spa->spa_deadman_cycid);
754 mutex_exit(&cpu_lock);
755 spa->spa_deadman_cycid = CYCLIC_NONE;
758 callout_drain(&spa->spa_deadman_cycid);
762 refcount_destroy(&spa->spa_refcount);
764 spa_config_lock_destroy(spa);
766 for (int t = 0; t < TXG_SIZE; t++)
767 bplist_destroy(&spa->spa_free_bplist[t]);
769 cv_destroy(&spa->spa_async_cv);
770 cv_destroy(&spa->spa_proc_cv);
771 cv_destroy(&spa->spa_scrub_io_cv);
772 cv_destroy(&spa->spa_suspend_cv);
774 mutex_destroy(&spa->spa_async_lock);
775 mutex_destroy(&spa->spa_errlist_lock);
776 mutex_destroy(&spa->spa_errlog_lock);
777 mutex_destroy(&spa->spa_history_lock);
778 mutex_destroy(&spa->spa_proc_lock);
779 mutex_destroy(&spa->spa_props_lock);
780 mutex_destroy(&spa->spa_scrub_lock);
781 mutex_destroy(&spa->spa_suspend_lock);
782 mutex_destroy(&spa->spa_vdev_top_lock);
784 kmem_free(spa, sizeof (spa_t));
788 * Given a pool, return the next pool in the namespace, or NULL if there is
789 * none. If 'prev' is NULL, return the first pool.
792 spa_next(spa_t *prev)
794 ASSERT(MUTEX_HELD(&spa_namespace_lock));
797 return (AVL_NEXT(&spa_namespace_avl, prev));
799 return (avl_first(&spa_namespace_avl));
803 * ==========================================================================
804 * SPA refcount functions
805 * ==========================================================================
809 * Add a reference to the given spa_t. Must have at least one reference, or
810 * have the namespace lock held.
813 spa_open_ref(spa_t *spa, void *tag)
815 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
816 MUTEX_HELD(&spa_namespace_lock));
817 (void) refcount_add(&spa->spa_refcount, tag);
821 * Remove a reference to the given spa_t. Must have at least one reference, or
822 * have the namespace lock held.
825 spa_close(spa_t *spa, void *tag)
827 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
828 MUTEX_HELD(&spa_namespace_lock));
829 (void) refcount_remove(&spa->spa_refcount, tag);
833 * Check to see if the spa refcount is zero. Must be called with
834 * spa_namespace_lock held. We really compare against spa_minref, which is the
835 * number of references acquired when opening a pool
838 spa_refcount_zero(spa_t *spa)
840 ASSERT(MUTEX_HELD(&spa_namespace_lock));
842 return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
846 * ==========================================================================
847 * SPA spare and l2cache tracking
848 * ==========================================================================
852 * Hot spares and cache devices are tracked using the same code below,
853 * for 'auxiliary' devices.
856 typedef struct spa_aux {
864 spa_aux_compare(const void *a, const void *b)
866 const spa_aux_t *sa = a;
867 const spa_aux_t *sb = b;
869 if (sa->aux_guid < sb->aux_guid)
871 else if (sa->aux_guid > sb->aux_guid)
878 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
884 search.aux_guid = vd->vdev_guid;
885 if ((aux = avl_find(avl, &search, &where)) != NULL) {
888 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
889 aux->aux_guid = vd->vdev_guid;
891 avl_insert(avl, aux, where);
896 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
902 search.aux_guid = vd->vdev_guid;
903 aux = avl_find(avl, &search, &where);
907 if (--aux->aux_count == 0) {
908 avl_remove(avl, aux);
909 kmem_free(aux, sizeof (spa_aux_t));
910 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
911 aux->aux_pool = 0ULL;
916 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
918 spa_aux_t search, *found;
920 search.aux_guid = guid;
921 found = avl_find(avl, &search, NULL);
925 *pool = found->aux_pool;
932 *refcnt = found->aux_count;
937 return (found != NULL);
941 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
943 spa_aux_t search, *found;
946 search.aux_guid = vd->vdev_guid;
947 found = avl_find(avl, &search, &where);
948 ASSERT(found != NULL);
949 ASSERT(found->aux_pool == 0ULL);
951 found->aux_pool = spa_guid(vd->vdev_spa);
955 * Spares are tracked globally due to the following constraints:
957 * - A spare may be part of multiple pools.
958 * - A spare may be added to a pool even if it's actively in use within
960 * - A spare in use in any pool can only be the source of a replacement if
961 * the target is a spare in the same pool.
963 * We keep track of all spares on the system through the use of a reference
964 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
965 * spare, then we bump the reference count in the AVL tree. In addition, we set
966 * the 'vdev_isspare' member to indicate that the device is a spare (active or
967 * inactive). When a spare is made active (used to replace a device in the
968 * pool), we also keep track of which pool its been made a part of.
970 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
971 * called under the spa_namespace lock as part of vdev reconfiguration. The
972 * separate spare lock exists for the status query path, which does not need to
973 * be completely consistent with respect to other vdev configuration changes.
977 spa_spare_compare(const void *a, const void *b)
979 return (spa_aux_compare(a, b));
983 spa_spare_add(vdev_t *vd)
985 mutex_enter(&spa_spare_lock);
986 ASSERT(!vd->vdev_isspare);
987 spa_aux_add(vd, &spa_spare_avl);
988 vd->vdev_isspare = B_TRUE;
989 mutex_exit(&spa_spare_lock);
993 spa_spare_remove(vdev_t *vd)
995 mutex_enter(&spa_spare_lock);
996 ASSERT(vd->vdev_isspare);
997 spa_aux_remove(vd, &spa_spare_avl);
998 vd->vdev_isspare = B_FALSE;
999 mutex_exit(&spa_spare_lock);
1003 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
1007 mutex_enter(&spa_spare_lock);
1008 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
1009 mutex_exit(&spa_spare_lock);
1015 spa_spare_activate(vdev_t *vd)
1017 mutex_enter(&spa_spare_lock);
1018 ASSERT(vd->vdev_isspare);
1019 spa_aux_activate(vd, &spa_spare_avl);
1020 mutex_exit(&spa_spare_lock);
1024 * Level 2 ARC devices are tracked globally for the same reasons as spares.
1025 * Cache devices currently only support one pool per cache device, and so
1026 * for these devices the aux reference count is currently unused beyond 1.
1030 spa_l2cache_compare(const void *a, const void *b)
1032 return (spa_aux_compare(a, b));
1036 spa_l2cache_add(vdev_t *vd)
1038 mutex_enter(&spa_l2cache_lock);
1039 ASSERT(!vd->vdev_isl2cache);
1040 spa_aux_add(vd, &spa_l2cache_avl);
1041 vd->vdev_isl2cache = B_TRUE;
1042 mutex_exit(&spa_l2cache_lock);
1046 spa_l2cache_remove(vdev_t *vd)
1048 mutex_enter(&spa_l2cache_lock);
1049 ASSERT(vd->vdev_isl2cache);
1050 spa_aux_remove(vd, &spa_l2cache_avl);
1051 vd->vdev_isl2cache = B_FALSE;
1052 mutex_exit(&spa_l2cache_lock);
1056 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
1060 mutex_enter(&spa_l2cache_lock);
1061 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
1062 mutex_exit(&spa_l2cache_lock);
1068 spa_l2cache_activate(vdev_t *vd)
1070 mutex_enter(&spa_l2cache_lock);
1071 ASSERT(vd->vdev_isl2cache);
1072 spa_aux_activate(vd, &spa_l2cache_avl);
1073 mutex_exit(&spa_l2cache_lock);
1077 * ==========================================================================
1079 * ==========================================================================
1083 * Lock the given spa_t for the purpose of adding or removing a vdev.
1084 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1085 * It returns the next transaction group for the spa_t.
1088 spa_vdev_enter(spa_t *spa)
1090 mutex_enter(&spa->spa_vdev_top_lock);
1091 mutex_enter(&spa_namespace_lock);
1092 return (spa_vdev_config_enter(spa));
1096 * Internal implementation for spa_vdev_enter(). Used when a vdev
1097 * operation requires multiple syncs (i.e. removing a device) while
1098 * keeping the spa_namespace_lock held.
1101 spa_vdev_config_enter(spa_t *spa)
1103 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1105 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1107 return (spa_last_synced_txg(spa) + 1);
1111 * Used in combination with spa_vdev_config_enter() to allow the syncing
1112 * of multiple transactions without releasing the spa_namespace_lock.
1115 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
1117 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1119 int config_changed = B_FALSE;
1121 ASSERT(txg > spa_last_synced_txg(spa));
1123 spa->spa_pending_vdev = NULL;
1126 * Reassess the DTLs.
1128 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
1130 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
1131 config_changed = B_TRUE;
1132 spa->spa_config_generation++;
1136 * Verify the metaslab classes.
1138 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
1139 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
1141 spa_config_exit(spa, SCL_ALL, spa);
1144 * Panic the system if the specified tag requires it. This
1145 * is useful for ensuring that configurations are updated
1148 if (zio_injection_enabled)
1149 zio_handle_panic_injection(spa, tag, 0);
1152 * Note: this txg_wait_synced() is important because it ensures
1153 * that there won't be more than one config change per txg.
1154 * This allows us to use the txg as the generation number.
1157 txg_wait_synced(spa->spa_dsl_pool, txg);
1160 ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
1161 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1163 spa_config_exit(spa, SCL_ALL, spa);
1167 * If the config changed, update the config cache.
1170 spa_config_sync(spa, B_FALSE, B_TRUE);
1174 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1175 * locking of spa_vdev_enter(), we also want make sure the transactions have
1176 * synced to disk, and then update the global configuration cache with the new
1180 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1182 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1183 mutex_exit(&spa_namespace_lock);
1184 mutex_exit(&spa->spa_vdev_top_lock);
1190 * Lock the given spa_t for the purpose of changing vdev state.
1193 spa_vdev_state_enter(spa_t *spa, int oplocks)
1195 int locks = SCL_STATE_ALL | oplocks;
1198 * Root pools may need to read of the underlying devfs filesystem
1199 * when opening up a vdev. Unfortunately if we're holding the
1200 * SCL_ZIO lock it will result in a deadlock when we try to issue
1201 * the read from the root filesystem. Instead we "prefetch"
1202 * the associated vnodes that we need prior to opening the
1203 * underlying devices and cache them so that we can prevent
1204 * any I/O when we are doing the actual open.
1206 if (spa_is_root(spa)) {
1207 int low = locks & ~(SCL_ZIO - 1);
1208 int high = locks & ~low;
1210 spa_config_enter(spa, high, spa, RW_WRITER);
1211 vdev_hold(spa->spa_root_vdev);
1212 spa_config_enter(spa, low, spa, RW_WRITER);
1214 spa_config_enter(spa, locks, spa, RW_WRITER);
1216 spa->spa_vdev_locks = locks;
1220 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1222 boolean_t config_changed = B_FALSE;
1224 if (vd != NULL || error == 0)
1225 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1229 vdev_state_dirty(vd->vdev_top);
1230 config_changed = B_TRUE;
1231 spa->spa_config_generation++;
1234 if (spa_is_root(spa))
1235 vdev_rele(spa->spa_root_vdev);
1237 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1238 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1241 * If anything changed, wait for it to sync. This ensures that,
1242 * from the system administrator's perspective, zpool(1M) commands
1243 * are synchronous. This is important for things like zpool offline:
1244 * when the command completes, you expect no further I/O from ZFS.
1247 txg_wait_synced(spa->spa_dsl_pool, 0);
1250 * If the config changed, update the config cache.
1252 if (config_changed) {
1253 mutex_enter(&spa_namespace_lock);
1254 spa_config_sync(spa, B_FALSE, B_TRUE);
1255 mutex_exit(&spa_namespace_lock);
1262 * ==========================================================================
1263 * Miscellaneous functions
1264 * ==========================================================================
1268 spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx)
1270 if (!nvlist_exists(spa->spa_label_features, feature)) {
1271 fnvlist_add_boolean(spa->spa_label_features, feature);
1273 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1274 * dirty the vdev config because lock SCL_CONFIG is not held.
1275 * Thankfully, in this case we don't need to dirty the config
1276 * because it will be written out anyway when we finish
1277 * creating the pool.
1279 if (tx->tx_txg != TXG_INITIAL)
1280 vdev_config_dirty(spa->spa_root_vdev);
1285 spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1287 if (nvlist_remove_all(spa->spa_label_features, feature) == 0)
1288 vdev_config_dirty(spa->spa_root_vdev);
1295 spa_rename(const char *name, const char *newname)
1301 * Lookup the spa_t and grab the config lock for writing. We need to
1302 * actually open the pool so that we can sync out the necessary labels.
1303 * It's OK to call spa_open() with the namespace lock held because we
1304 * allow recursive calls for other reasons.
1306 mutex_enter(&spa_namespace_lock);
1307 if ((err = spa_open(name, &spa, FTAG)) != 0) {
1308 mutex_exit(&spa_namespace_lock);
1312 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1314 avl_remove(&spa_namespace_avl, spa);
1315 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1316 avl_add(&spa_namespace_avl, spa);
1319 * Sync all labels to disk with the new names by marking the root vdev
1320 * dirty and waiting for it to sync. It will pick up the new pool name
1323 vdev_config_dirty(spa->spa_root_vdev);
1325 spa_config_exit(spa, SCL_ALL, FTAG);
1327 txg_wait_synced(spa->spa_dsl_pool, 0);
1330 * Sync the updated config cache.
1332 spa_config_sync(spa, B_FALSE, B_TRUE);
1334 spa_close(spa, FTAG);
1336 mutex_exit(&spa_namespace_lock);
1342 * Return the spa_t associated with given pool_guid, if it exists. If
1343 * device_guid is non-zero, determine whether the pool exists *and* contains
1344 * a device with the specified device_guid.
1347 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1350 avl_tree_t *t = &spa_namespace_avl;
1352 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1354 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1355 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1357 if (spa->spa_root_vdev == NULL)
1359 if (spa_guid(spa) == pool_guid) {
1360 if (device_guid == 0)
1363 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1364 device_guid) != NULL)
1368 * Check any devices we may be in the process of adding.
1370 if (spa->spa_pending_vdev) {
1371 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1372 device_guid) != NULL)
1382 * Determine whether a pool with the given pool_guid exists.
1385 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1387 return (spa_by_guid(pool_guid, device_guid) != NULL);
1391 spa_strdup(const char *s)
1397 new = kmem_alloc(len + 1, KM_SLEEP);
1405 spa_strfree(char *s)
1407 kmem_free(s, strlen(s) + 1);
1411 spa_get_random(uint64_t range)
1417 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1423 spa_generate_guid(spa_t *spa)
1425 uint64_t guid = spa_get_random(-1ULL);
1428 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1429 guid = spa_get_random(-1ULL);
1431 while (guid == 0 || spa_guid_exists(guid, 0))
1432 guid = spa_get_random(-1ULL);
1439 snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp)
1442 char *checksum = NULL;
1443 char *compress = NULL;
1446 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1447 dmu_object_byteswap_t bswap =
1448 DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1449 (void) snprintf(type, sizeof (type), "bswap %s %s",
1450 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1451 "metadata" : "data",
1452 dmu_ot_byteswap[bswap].ob_name);
1454 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1457 if (!BP_IS_EMBEDDED(bp)) {
1459 zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1461 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1464 SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum,
1469 spa_freeze(spa_t *spa)
1471 uint64_t freeze_txg = 0;
1473 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1474 if (spa->spa_freeze_txg == UINT64_MAX) {
1475 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1476 spa->spa_freeze_txg = freeze_txg;
1478 spa_config_exit(spa, SCL_ALL, FTAG);
1479 if (freeze_txg != 0)
1480 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1484 zfs_panic_recover(const char *fmt, ...)
1489 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1494 * This is a stripped-down version of strtoull, suitable only for converting
1495 * lowercase hexadecimal numbers that don't overflow.
1498 zfs_strtonum(const char *str, char **nptr)
1504 while ((c = *str) != '\0') {
1505 if (c >= '0' && c <= '9')
1507 else if (c >= 'a' && c <= 'f')
1508 digit = 10 + c - 'a';
1519 *nptr = (char *)str;
1525 * ==========================================================================
1526 * Accessor functions
1527 * ==========================================================================
1531 spa_shutting_down(spa_t *spa)
1533 return (spa->spa_async_suspended);
1537 spa_get_dsl(spa_t *spa)
1539 return (spa->spa_dsl_pool);
1543 spa_is_initializing(spa_t *spa)
1545 return (spa->spa_is_initializing);
1549 spa_get_rootblkptr(spa_t *spa)
1551 return (&spa->spa_ubsync.ub_rootbp);
1555 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1557 spa->spa_uberblock.ub_rootbp = *bp;
1561 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1563 if (spa->spa_root == NULL)
1566 (void) strncpy(buf, spa->spa_root, buflen);
1570 spa_sync_pass(spa_t *spa)
1572 return (spa->spa_sync_pass);
1576 spa_name(spa_t *spa)
1578 return (spa->spa_name);
1582 spa_guid(spa_t *spa)
1584 dsl_pool_t *dp = spa_get_dsl(spa);
1588 * If we fail to parse the config during spa_load(), we can go through
1589 * the error path (which posts an ereport) and end up here with no root
1590 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1593 if (spa->spa_root_vdev == NULL)
1594 return (spa->spa_config_guid);
1596 guid = spa->spa_last_synced_guid != 0 ?
1597 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1600 * Return the most recently synced out guid unless we're
1601 * in syncing context.
1603 if (dp && dsl_pool_sync_context(dp))
1604 return (spa->spa_root_vdev->vdev_guid);
1610 spa_load_guid(spa_t *spa)
1613 * This is a GUID that exists solely as a reference for the
1614 * purposes of the arc. It is generated at load time, and
1615 * is never written to persistent storage.
1617 return (spa->spa_load_guid);
1621 spa_last_synced_txg(spa_t *spa)
1623 return (spa->spa_ubsync.ub_txg);
1627 spa_first_txg(spa_t *spa)
1629 return (spa->spa_first_txg);
1633 spa_syncing_txg(spa_t *spa)
1635 return (spa->spa_syncing_txg);
1639 spa_state(spa_t *spa)
1641 return (spa->spa_state);
1645 spa_load_state(spa_t *spa)
1647 return (spa->spa_load_state);
1651 spa_freeze_txg(spa_t *spa)
1653 return (spa->spa_freeze_txg);
1658 spa_get_asize(spa_t *spa, uint64_t lsize)
1660 return (lsize * spa_asize_inflation);
1664 * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%),
1667 * See the comment above spa_slop_shift for details.
1670 spa_get_slop_space(spa_t *spa) {
1671 uint64_t space = spa_get_dspace(spa);
1672 return (MAX(space >> spa_slop_shift, SPA_MINDEVSIZE >> 1));
1676 spa_get_dspace(spa_t *spa)
1678 return (spa->spa_dspace);
1682 spa_update_dspace(spa_t *spa)
1684 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1685 ddt_get_dedup_dspace(spa);
1689 * Return the failure mode that has been set to this pool. The default
1690 * behavior will be to block all I/Os when a complete failure occurs.
1693 spa_get_failmode(spa_t *spa)
1695 return (spa->spa_failmode);
1699 spa_suspended(spa_t *spa)
1701 return (spa->spa_suspended);
1705 spa_version(spa_t *spa)
1707 return (spa->spa_ubsync.ub_version);
1711 spa_deflate(spa_t *spa)
1713 return (spa->spa_deflate);
1717 spa_normal_class(spa_t *spa)
1719 return (spa->spa_normal_class);
1723 spa_log_class(spa_t *spa)
1725 return (spa->spa_log_class);
1729 spa_max_replication(spa_t *spa)
1732 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1733 * handle BPs with more than one DVA allocated. Set our max
1734 * replication level accordingly.
1736 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1738 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1742 spa_prev_software_version(spa_t *spa)
1744 return (spa->spa_prev_software_version);
1748 spa_deadman_synctime(spa_t *spa)
1750 return (spa->spa_deadman_synctime);
1754 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1756 uint64_t asize = DVA_GET_ASIZE(dva);
1757 uint64_t dsize = asize;
1759 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1761 if (asize != 0 && spa->spa_deflate) {
1762 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1763 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1770 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1774 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1775 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1781 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1785 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1787 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1788 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1790 spa_config_exit(spa, SCL_VDEV, FTAG);
1796 * ==========================================================================
1797 * Initialization and Termination
1798 * ==========================================================================
1802 spa_name_compare(const void *a1, const void *a2)
1804 const spa_t *s1 = a1;
1805 const spa_t *s2 = a2;
1808 s = strcmp(s1->spa_name, s2->spa_name);
1819 return (spa_active_count);
1829 EVENTHANDLER_DEFINE(mountroot, spa_boot_init, NULL, 0);
1835 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1836 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1837 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1838 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1840 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1841 offsetof(spa_t, spa_avl));
1843 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1844 offsetof(spa_aux_t, aux_avl));
1846 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1847 offsetof(spa_aux_t, aux_avl));
1849 spa_mode_global = mode;
1855 if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
1856 arc_procfd = open("/proc/self/ctl", O_WRONLY);
1857 if (arc_procfd == -1) {
1858 perror("could not enable watchpoints: "
1859 "opening /proc/self/ctl failed: ");
1865 #endif /* illumos */
1873 vdev_cache_stat_init();
1876 zpool_feature_init();
1883 #endif /* !illumos */
1893 vdev_cache_stat_fini();
1902 avl_destroy(&spa_namespace_avl);
1903 avl_destroy(&spa_spare_avl);
1904 avl_destroy(&spa_l2cache_avl);
1906 cv_destroy(&spa_namespace_cv);
1907 mutex_destroy(&spa_namespace_lock);
1908 mutex_destroy(&spa_spare_lock);
1909 mutex_destroy(&spa_l2cache_lock);
1913 * Return whether this pool has slogs. No locking needed.
1914 * It's not a problem if the wrong answer is returned as it's only for
1915 * performance and not correctness
1918 spa_has_slogs(spa_t *spa)
1920 return (spa->spa_log_class->mc_rotor != NULL);
1924 spa_get_log_state(spa_t *spa)
1926 return (spa->spa_log_state);
1930 spa_set_log_state(spa_t *spa, spa_log_state_t state)
1932 spa->spa_log_state = state;
1936 spa_is_root(spa_t *spa)
1938 return (spa->spa_is_root);
1942 spa_writeable(spa_t *spa)
1944 return (!!(spa->spa_mode & FWRITE));
1948 * Returns true if there is a pending sync task in any of the current
1949 * syncing txg, the current quiescing txg, or the current open txg.
1952 spa_has_pending_synctask(spa_t *spa)
1954 return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks));
1958 spa_mode(spa_t *spa)
1960 return (spa->spa_mode);
1964 spa_bootfs(spa_t *spa)
1966 return (spa->spa_bootfs);
1970 spa_delegation(spa_t *spa)
1972 return (spa->spa_delegation);
1976 spa_meta_objset(spa_t *spa)
1978 return (spa->spa_meta_objset);
1982 spa_dedup_checksum(spa_t *spa)
1984 return (spa->spa_dedup_checksum);
1988 * Reset pool scan stat per scan pass (or reboot).
1991 spa_scan_stat_init(spa_t *spa)
1993 /* data not stored on disk */
1994 spa->spa_scan_pass_start = gethrestime_sec();
1995 spa->spa_scan_pass_exam = 0;
1996 vdev_scan_stat_init(spa->spa_root_vdev);
2000 * Get scan stats for zpool status reports
2003 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
2005 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
2007 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
2008 return (SET_ERROR(ENOENT));
2009 bzero(ps, sizeof (pool_scan_stat_t));
2011 /* data stored on disk */
2012 ps->pss_func = scn->scn_phys.scn_func;
2013 ps->pss_start_time = scn->scn_phys.scn_start_time;
2014 ps->pss_end_time = scn->scn_phys.scn_end_time;
2015 ps->pss_to_examine = scn->scn_phys.scn_to_examine;
2016 ps->pss_examined = scn->scn_phys.scn_examined;
2017 ps->pss_to_process = scn->scn_phys.scn_to_process;
2018 ps->pss_processed = scn->scn_phys.scn_processed;
2019 ps->pss_errors = scn->scn_phys.scn_errors;
2020 ps->pss_state = scn->scn_phys.scn_state;
2022 /* data not stored on disk */
2023 ps->pss_pass_start = spa->spa_scan_pass_start;
2024 ps->pss_pass_exam = spa->spa_scan_pass_exam;
2030 spa_debug_enabled(spa_t *spa)
2032 return (spa->spa_debug);
2036 spa_maxblocksize(spa_t *spa)
2038 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS))
2039 return (SPA_MAXBLOCKSIZE);
2041 return (SPA_OLD_MAXBLOCKSIZE);