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.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
29 #include <sys/zfs_context.h>
30 #include <sys/spa_impl.h>
31 #include <sys/spa_boot.h>
33 #include <sys/zio_checksum.h>
34 #include <sys/zio_compress.h>
36 #include <sys/dmu_tx.h>
39 #include <sys/vdev_impl.h>
40 #include <sys/metaslab.h>
41 #include <sys/uberblock_impl.h>
44 #include <sys/unique.h>
45 #include <sys/dsl_pool.h>
46 #include <sys/dsl_dir.h>
47 #include <sys/dsl_prop.h>
48 #include <sys/dsl_scan.h>
49 #include <sys/fs/zfs.h>
50 #include <sys/metaslab_impl.h>
54 #include "zfeature_common.h"
59 * There are four basic locks for managing spa_t structures:
61 * spa_namespace_lock (global mutex)
63 * This lock must be acquired to do any of the following:
65 * - Lookup a spa_t by name
66 * - Add or remove a spa_t from the namespace
67 * - Increase spa_refcount from non-zero
68 * - Check if spa_refcount is zero
70 * - add/remove/attach/detach devices
71 * - Held for the duration of create/destroy/import/export
73 * It does not need to handle recursion. A create or destroy may
74 * reference objects (files or zvols) in other pools, but by
75 * definition they must have an existing reference, and will never need
76 * to lookup a spa_t by name.
78 * spa_refcount (per-spa refcount_t protected by mutex)
80 * This reference count keep track of any active users of the spa_t. The
81 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
82 * the refcount is never really 'zero' - opening a pool implicitly keeps
83 * some references in the DMU. Internally we check against spa_minref, but
84 * present the image of a zero/non-zero value to consumers.
86 * spa_config_lock[] (per-spa array of rwlocks)
88 * This protects the spa_t from config changes, and must be held in
89 * the following circumstances:
91 * - RW_READER to perform I/O to the spa
92 * - RW_WRITER to change the vdev config
94 * The locking order is fairly straightforward:
96 * spa_namespace_lock -> spa_refcount
98 * The namespace lock must be acquired to increase the refcount from 0
99 * or to check if it is zero.
101 * spa_refcount -> spa_config_lock[]
103 * There must be at least one valid reference on the spa_t to acquire
106 * spa_namespace_lock -> spa_config_lock[]
108 * The namespace lock must always be taken before the config lock.
111 * The spa_namespace_lock can be acquired directly and is globally visible.
113 * The namespace is manipulated using the following functions, all of which
114 * require the spa_namespace_lock to be held.
116 * spa_lookup() Lookup a spa_t by name.
118 * spa_add() Create a new spa_t in the namespace.
120 * spa_remove() Remove a spa_t from the namespace. This also
121 * frees up any memory associated with the spa_t.
123 * spa_next() Returns the next spa_t in the system, or the
124 * first if NULL is passed.
126 * spa_evict_all() Shutdown and remove all spa_t structures in
129 * spa_guid_exists() Determine whether a pool/device guid exists.
131 * The spa_refcount is manipulated using the following functions:
133 * spa_open_ref() Adds a reference to the given spa_t. Must be
134 * called with spa_namespace_lock held if the
135 * refcount is currently zero.
137 * spa_close() Remove a reference from the spa_t. This will
138 * not free the spa_t or remove it from the
139 * namespace. No locking is required.
141 * spa_refcount_zero() Returns true if the refcount is currently
142 * zero. Must be called with spa_namespace_lock
145 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
146 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
147 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
149 * To read the configuration, it suffices to hold one of these locks as reader.
150 * To modify the configuration, you must hold all locks as writer. To modify
151 * vdev state without altering the vdev tree's topology (e.g. online/offline),
152 * you must hold SCL_STATE and SCL_ZIO as writer.
154 * We use these distinct config locks to avoid recursive lock entry.
155 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
156 * block allocations (SCL_ALLOC), which may require reading space maps
157 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
159 * The spa config locks cannot be normal rwlocks because we need the
160 * ability to hand off ownership. For example, SCL_ZIO is acquired
161 * by the issuing thread and later released by an interrupt thread.
162 * They do, however, obey the usual write-wanted semantics to prevent
163 * writer (i.e. system administrator) starvation.
165 * The lock acquisition rules are as follows:
168 * Protects changes to the vdev tree topology, such as vdev
169 * add/remove/attach/detach. Protects the dirty config list
170 * (spa_config_dirty_list) and the set of spares and l2arc devices.
173 * Protects changes to pool state and vdev state, such as vdev
174 * online/offline/fault/degrade/clear. Protects the dirty state list
175 * (spa_state_dirty_list) and global pool state (spa_state).
178 * Protects changes to metaslab groups and classes.
179 * Held as reader by metaslab_alloc() and metaslab_claim().
182 * Held by bp-level zios (those which have no io_vd upon entry)
183 * to prevent changes to the vdev tree. The bp-level zio implicitly
184 * protects all of its vdev child zios, which do not hold SCL_ZIO.
187 * Protects changes to metaslab groups and classes.
188 * Held as reader by metaslab_free(). SCL_FREE is distinct from
189 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
190 * blocks in zio_done() while another i/o that holds either
191 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
194 * Held as reader to prevent changes to the vdev tree during trivial
195 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
196 * other locks, and lower than all of them, to ensure that it's safe
197 * to acquire regardless of caller context.
199 * In addition, the following rules apply:
201 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
202 * The lock ordering is SCL_CONFIG > spa_props_lock.
204 * (b) I/O operations on leaf vdevs. For any zio operation that takes
205 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
206 * or zio_write_phys() -- the caller must ensure that the config cannot
207 * cannot change in the interim, and that the vdev cannot be reopened.
208 * SCL_STATE as reader suffices for both.
210 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
212 * spa_vdev_enter() Acquire the namespace lock and the config lock
215 * spa_vdev_exit() Release the config lock, wait for all I/O
216 * to complete, sync the updated configs to the
217 * cache, and release the namespace lock.
219 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
220 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
221 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
223 * spa_rename() is also implemented within this file since it requires
224 * manipulation of the namespace.
227 static avl_tree_t spa_namespace_avl;
228 kmutex_t spa_namespace_lock;
229 static kcondvar_t spa_namespace_cv;
230 static int spa_active_count;
231 int spa_max_replication_override = SPA_DVAS_PER_BP;
233 static kmutex_t spa_spare_lock;
234 static avl_tree_t spa_spare_avl;
235 static kmutex_t spa_l2cache_lock;
236 static avl_tree_t spa_l2cache_avl;
238 kmem_cache_t *spa_buffer_pool;
242 /* Everything except dprintf and spa is on by default in debug builds */
243 int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA);
248 TUNABLE_INT("debug.zfs_flags", &zfs_flags);
249 SYSCTL_INT(_debug, OID_AUTO, zfs_flags, CTLFLAG_RWTUN, &zfs_flags, 0,
253 * zfs_recover can be set to nonzero to attempt to recover from
254 * otherwise-fatal errors, typically caused by on-disk corruption. When
255 * set, calls to zfs_panic_recover() will turn into warning messages.
256 * This should only be used as a last resort, as it typically results
257 * in leaked space, or worse.
259 boolean_t zfs_recover = B_FALSE;
260 SYSCTL_DECL(_vfs_zfs);
261 TUNABLE_INT("vfs.zfs.recover", &zfs_recover);
262 SYSCTL_INT(_vfs_zfs, OID_AUTO, recover, CTLFLAG_RWTUN, &zfs_recover, 0,
263 "Try to recover from otherwise-fatal errors.");
266 * If destroy encounters an EIO while reading metadata (e.g. indirect
267 * blocks), space referenced by the missing metadata can not be freed.
268 * Normally this causes the background destroy to become "stalled", as
269 * it is unable to make forward progress. While in this stalled state,
270 * all remaining space to free from the error-encountering filesystem is
271 * "temporarily leaked". Set this flag to cause it to ignore the EIO,
272 * permanently leak the space from indirect blocks that can not be read,
273 * and continue to free everything else that it can.
275 * The default, "stalling" behavior is useful if the storage partially
276 * fails (i.e. some but not all i/os fail), and then later recovers. In
277 * this case, we will be able to continue pool operations while it is
278 * partially failed, and when it recovers, we can continue to free the
279 * space, with no leaks. However, note that this case is actually
282 * Typically pools either (a) fail completely (but perhaps temporarily,
283 * e.g. a top-level vdev going offline), or (b) have localized,
284 * permanent errors (e.g. disk returns the wrong data due to bit flip or
285 * firmware bug). In case (a), this setting does not matter because the
286 * pool will be suspended and the sync thread will not be able to make
287 * forward progress regardless. In case (b), because the error is
288 * permanent, the best we can do is leak the minimum amount of space,
289 * which is what setting this flag will do. Therefore, it is reasonable
290 * for this flag to normally be set, but we chose the more conservative
291 * approach of not setting it, so that there is no possibility of
292 * leaking space in the "partial temporary" failure case.
294 boolean_t zfs_free_leak_on_eio = B_FALSE;
297 * Expiration time in milliseconds. This value has two meanings. First it is
298 * used to determine when the spa_deadman() logic should fire. By default the
299 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
300 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
301 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
304 uint64_t zfs_deadman_synctime_ms = 1000000ULL;
305 TUNABLE_QUAD("vfs.zfs.deadman_synctime_ms", &zfs_deadman_synctime_ms);
306 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_synctime_ms, CTLFLAG_RDTUN,
307 &zfs_deadman_synctime_ms, 0,
308 "Stalled ZFS I/O expiration time in milliseconds");
311 * Check time in milliseconds. This defines the frequency at which we check
314 uint64_t zfs_deadman_checktime_ms = 5000ULL;
315 TUNABLE_QUAD("vfs.zfs.deadman_checktime_ms", &zfs_deadman_checktime_ms);
316 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_checktime_ms, CTLFLAG_RDTUN,
317 &zfs_deadman_checktime_ms, 0,
318 "Period of checks for stalled ZFS I/O in milliseconds");
321 * Default value of -1 for zfs_deadman_enabled is resolved in
324 int zfs_deadman_enabled = -1;
325 TUNABLE_INT("vfs.zfs.deadman_enabled", &zfs_deadman_enabled);
326 SYSCTL_INT(_vfs_zfs, OID_AUTO, deadman_enabled, CTLFLAG_RDTUN,
327 &zfs_deadman_enabled, 0, "Kernel panic on stalled ZFS I/O");
330 * The worst case is single-sector max-parity RAID-Z blocks, in which
331 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
332 * times the size; so just assume that. Add to this the fact that
333 * we can have up to 3 DVAs per bp, and one more factor of 2 because
334 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together,
336 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
338 int spa_asize_inflation = 24;
339 TUNABLE_INT("vfs.zfs.spa_asize_inflation", &spa_asize_inflation);
340 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_asize_inflation, CTLFLAG_RWTUN,
341 &spa_asize_inflation, 0, "Worst case inflation factor for single sector writes");
349 * If we are not i386 or amd64 or in a virtual machine,
350 * disable ZFS deadman thread by default
352 if (zfs_deadman_enabled == -1) {
353 #if defined(__amd64__) || defined(__i386__)
354 zfs_deadman_enabled = (vm_guest == VM_GUEST_NO) ? 1 : 0;
356 zfs_deadman_enabled = 0;
361 #endif /* !illumos */
364 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
365 * the pool to be consumed. This ensures that we don't run the pool
366 * completely out of space, due to unaccounted changes (e.g. to the MOS).
367 * It also limits the worst-case time to allocate space. If we have
368 * less than this amount of free space, most ZPL operations (e.g. write,
369 * create) will return ENOSPC.
371 * Certain operations (e.g. file removal, most administrative actions) can
372 * use half the slop space. They will only return ENOSPC if less than half
373 * the slop space is free. Typically, once the pool has less than the slop
374 * space free, the user will use these operations to free up space in the pool.
375 * These are the operations that call dsl_pool_adjustedsize() with the netfree
376 * argument set to TRUE.
378 * A very restricted set of operations are always permitted, regardless of
379 * the amount of free space. These are the operations that call
380 * dsl_sync_task(ZFS_SPACE_CHECK_NONE), e.g. "zfs destroy". If these
381 * operations result in a net increase in the amount of space used,
382 * it is possible to run the pool completely out of space, causing it to
383 * be permanently read-only.
385 * See also the comments in zfs_space_check_t.
387 int spa_slop_shift = 5;
388 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_slop_shift, CTLFLAG_RWTUN,
390 "Shift value of reserved space (1/(2^spa_slop_shift)).");
393 * ==========================================================================
395 * ==========================================================================
398 spa_config_lock_init(spa_t *spa)
400 for (int i = 0; i < SCL_LOCKS; i++) {
401 spa_config_lock_t *scl = &spa->spa_config_lock[i];
402 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
403 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
404 refcount_create_untracked(&scl->scl_count);
405 scl->scl_writer = NULL;
406 scl->scl_write_wanted = 0;
411 spa_config_lock_destroy(spa_t *spa)
413 for (int i = 0; i < SCL_LOCKS; i++) {
414 spa_config_lock_t *scl = &spa->spa_config_lock[i];
415 mutex_destroy(&scl->scl_lock);
416 cv_destroy(&scl->scl_cv);
417 refcount_destroy(&scl->scl_count);
418 ASSERT(scl->scl_writer == NULL);
419 ASSERT(scl->scl_write_wanted == 0);
424 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
426 for (int i = 0; i < SCL_LOCKS; i++) {
427 spa_config_lock_t *scl = &spa->spa_config_lock[i];
428 if (!(locks & (1 << i)))
430 mutex_enter(&scl->scl_lock);
431 if (rw == RW_READER) {
432 if (scl->scl_writer || scl->scl_write_wanted) {
433 mutex_exit(&scl->scl_lock);
434 spa_config_exit(spa, locks ^ (1 << i), tag);
438 ASSERT(scl->scl_writer != curthread);
439 if (!refcount_is_zero(&scl->scl_count)) {
440 mutex_exit(&scl->scl_lock);
441 spa_config_exit(spa, locks ^ (1 << i), tag);
444 scl->scl_writer = curthread;
446 (void) refcount_add(&scl->scl_count, tag);
447 mutex_exit(&scl->scl_lock);
453 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
457 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
459 for (int i = 0; i < SCL_LOCKS; i++) {
460 spa_config_lock_t *scl = &spa->spa_config_lock[i];
461 if (scl->scl_writer == curthread)
462 wlocks_held |= (1 << i);
463 if (!(locks & (1 << i)))
465 mutex_enter(&scl->scl_lock);
466 if (rw == RW_READER) {
467 while (scl->scl_writer || scl->scl_write_wanted) {
468 cv_wait(&scl->scl_cv, &scl->scl_lock);
471 ASSERT(scl->scl_writer != curthread);
472 while (!refcount_is_zero(&scl->scl_count)) {
473 scl->scl_write_wanted++;
474 cv_wait(&scl->scl_cv, &scl->scl_lock);
475 scl->scl_write_wanted--;
477 scl->scl_writer = curthread;
479 (void) refcount_add(&scl->scl_count, tag);
480 mutex_exit(&scl->scl_lock);
482 ASSERT(wlocks_held <= locks);
486 spa_config_exit(spa_t *spa, int locks, void *tag)
488 for (int i = SCL_LOCKS - 1; i >= 0; i--) {
489 spa_config_lock_t *scl = &spa->spa_config_lock[i];
490 if (!(locks & (1 << i)))
492 mutex_enter(&scl->scl_lock);
493 ASSERT(!refcount_is_zero(&scl->scl_count));
494 if (refcount_remove(&scl->scl_count, tag) == 0) {
495 ASSERT(scl->scl_writer == NULL ||
496 scl->scl_writer == curthread);
497 scl->scl_writer = NULL; /* OK in either case */
498 cv_broadcast(&scl->scl_cv);
500 mutex_exit(&scl->scl_lock);
505 spa_config_held(spa_t *spa, int locks, krw_t rw)
509 for (int i = 0; i < SCL_LOCKS; i++) {
510 spa_config_lock_t *scl = &spa->spa_config_lock[i];
511 if (!(locks & (1 << i)))
513 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
514 (rw == RW_WRITER && scl->scl_writer == curthread))
515 locks_held |= 1 << i;
522 * ==========================================================================
523 * SPA namespace functions
524 * ==========================================================================
528 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
529 * Returns NULL if no matching spa_t is found.
532 spa_lookup(const char *name)
534 static spa_t search; /* spa_t is large; don't allocate on stack */
539 ASSERT(MUTEX_HELD(&spa_namespace_lock));
541 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
544 * If it's a full dataset name, figure out the pool name and
547 cp = strpbrk(search.spa_name, "/@#");
551 spa = avl_find(&spa_namespace_avl, &search, &where);
557 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
558 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
559 * looking for potentially hung I/Os.
562 spa_deadman(void *arg)
567 * Disable the deadman timer if the pool is suspended.
569 if (spa_suspended(spa)) {
571 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
573 /* Nothing. just don't schedule any future callouts. */
578 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
579 (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
580 ++spa->spa_deadman_calls);
581 if (zfs_deadman_enabled)
582 vdev_deadman(spa->spa_root_vdev);
586 * Create an uninitialized spa_t with the given name. Requires
587 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
588 * exist by calling spa_lookup() first.
591 spa_add(const char *name, nvlist_t *config, const char *altroot)
594 spa_config_dirent_t *dp;
600 ASSERT(MUTEX_HELD(&spa_namespace_lock));
602 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
604 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
605 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
606 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
607 mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL);
608 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
609 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
610 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
611 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
612 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
613 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
615 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
616 cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL);
617 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
618 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
619 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
621 for (int t = 0; t < TXG_SIZE; t++)
622 bplist_create(&spa->spa_free_bplist[t]);
624 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
625 spa->spa_state = POOL_STATE_UNINITIALIZED;
626 spa->spa_freeze_txg = UINT64_MAX;
627 spa->spa_final_txg = UINT64_MAX;
628 spa->spa_load_max_txg = UINT64_MAX;
630 spa->spa_proc_state = SPA_PROC_NONE;
633 hdlr.cyh_func = spa_deadman;
635 hdlr.cyh_level = CY_LOW_LEVEL;
638 spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
642 * This determines how often we need to check for hung I/Os after
643 * the cyclic has already fired. Since checking for hung I/Os is
644 * an expensive operation we don't want to check too frequently.
645 * Instead wait for 5 seconds before checking again.
647 when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms);
648 when.cyt_when = CY_INFINITY;
649 mutex_enter(&cpu_lock);
650 spa->spa_deadman_cycid = cyclic_add(&hdlr, &when);
651 mutex_exit(&cpu_lock);
654 callout_init(&spa->spa_deadman_cycid, CALLOUT_MPSAFE);
657 refcount_create(&spa->spa_refcount);
658 spa_config_lock_init(spa);
660 avl_add(&spa_namespace_avl, spa);
663 * Set the alternate root, if there is one.
666 spa->spa_root = spa_strdup(altroot);
671 * Every pool starts with the default cachefile
673 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
674 offsetof(spa_config_dirent_t, scd_link));
676 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
677 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
678 list_insert_head(&spa->spa_config_list, dp);
680 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
683 if (config != NULL) {
686 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
688 VERIFY(nvlist_dup(features, &spa->spa_label_features,
692 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
695 if (spa->spa_label_features == NULL) {
696 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
700 spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0);
702 spa->spa_min_ashift = INT_MAX;
703 spa->spa_max_ashift = 0;
706 * As a pool is being created, treat all features as disabled by
707 * setting SPA_FEATURE_DISABLED for all entries in the feature
710 for (int i = 0; i < SPA_FEATURES; i++) {
711 spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED;
718 * Removes a spa_t from the namespace, freeing up any memory used. Requires
719 * spa_namespace_lock. This is called only after the spa_t has been closed and
723 spa_remove(spa_t *spa)
725 spa_config_dirent_t *dp;
727 ASSERT(MUTEX_HELD(&spa_namespace_lock));
728 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
729 ASSERT3U(refcount_count(&spa->spa_refcount), ==, 0);
731 nvlist_free(spa->spa_config_splitting);
733 avl_remove(&spa_namespace_avl, spa);
734 cv_broadcast(&spa_namespace_cv);
737 spa_strfree(spa->spa_root);
741 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
742 list_remove(&spa->spa_config_list, dp);
743 if (dp->scd_path != NULL)
744 spa_strfree(dp->scd_path);
745 kmem_free(dp, sizeof (spa_config_dirent_t));
748 list_destroy(&spa->spa_config_list);
750 nvlist_free(spa->spa_label_features);
751 nvlist_free(spa->spa_load_info);
752 spa_config_set(spa, NULL);
755 mutex_enter(&cpu_lock);
756 if (spa->spa_deadman_cycid != CYCLIC_NONE)
757 cyclic_remove(spa->spa_deadman_cycid);
758 mutex_exit(&cpu_lock);
759 spa->spa_deadman_cycid = CYCLIC_NONE;
762 callout_drain(&spa->spa_deadman_cycid);
766 refcount_destroy(&spa->spa_refcount);
768 spa_config_lock_destroy(spa);
770 for (int t = 0; t < TXG_SIZE; t++)
771 bplist_destroy(&spa->spa_free_bplist[t]);
773 cv_destroy(&spa->spa_async_cv);
774 cv_destroy(&spa->spa_evicting_os_cv);
775 cv_destroy(&spa->spa_proc_cv);
776 cv_destroy(&spa->spa_scrub_io_cv);
777 cv_destroy(&spa->spa_suspend_cv);
779 mutex_destroy(&spa->spa_async_lock);
780 mutex_destroy(&spa->spa_errlist_lock);
781 mutex_destroy(&spa->spa_errlog_lock);
782 mutex_destroy(&spa->spa_evicting_os_lock);
783 mutex_destroy(&spa->spa_history_lock);
784 mutex_destroy(&spa->spa_proc_lock);
785 mutex_destroy(&spa->spa_props_lock);
786 mutex_destroy(&spa->spa_scrub_lock);
787 mutex_destroy(&spa->spa_suspend_lock);
788 mutex_destroy(&spa->spa_vdev_top_lock);
790 kmem_free(spa, sizeof (spa_t));
794 * Given a pool, return the next pool in the namespace, or NULL if there is
795 * none. If 'prev' is NULL, return the first pool.
798 spa_next(spa_t *prev)
800 ASSERT(MUTEX_HELD(&spa_namespace_lock));
803 return (AVL_NEXT(&spa_namespace_avl, prev));
805 return (avl_first(&spa_namespace_avl));
809 * ==========================================================================
810 * SPA refcount functions
811 * ==========================================================================
815 * Add a reference to the given spa_t. Must have at least one reference, or
816 * have the namespace lock held.
819 spa_open_ref(spa_t *spa, void *tag)
821 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
822 MUTEX_HELD(&spa_namespace_lock));
823 (void) refcount_add(&spa->spa_refcount, tag);
827 * Remove a reference to the given spa_t. Must have at least one reference, or
828 * have the namespace lock held.
831 spa_close(spa_t *spa, void *tag)
833 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
834 MUTEX_HELD(&spa_namespace_lock));
835 (void) refcount_remove(&spa->spa_refcount, tag);
839 * Remove a reference to the given spa_t held by a dsl dir that is
840 * being asynchronously released. Async releases occur from a taskq
841 * performing eviction of dsl datasets and dirs. The namespace lock
842 * isn't held and the hold by the object being evicted may contribute to
843 * spa_minref (e.g. dataset or directory released during pool export),
844 * so the asserts in spa_close() do not apply.
847 spa_async_close(spa_t *spa, void *tag)
849 (void) refcount_remove(&spa->spa_refcount, tag);
853 * Check to see if the spa refcount is zero. Must be called with
854 * spa_namespace_lock held. We really compare against spa_minref, which is the
855 * number of references acquired when opening a pool
858 spa_refcount_zero(spa_t *spa)
860 ASSERT(MUTEX_HELD(&spa_namespace_lock));
862 return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
866 * ==========================================================================
867 * SPA spare and l2cache tracking
868 * ==========================================================================
872 * Hot spares and cache devices are tracked using the same code below,
873 * for 'auxiliary' devices.
876 typedef struct spa_aux {
884 spa_aux_compare(const void *a, const void *b)
886 const spa_aux_t *sa = a;
887 const spa_aux_t *sb = b;
889 if (sa->aux_guid < sb->aux_guid)
891 else if (sa->aux_guid > sb->aux_guid)
898 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
904 search.aux_guid = vd->vdev_guid;
905 if ((aux = avl_find(avl, &search, &where)) != NULL) {
908 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
909 aux->aux_guid = vd->vdev_guid;
911 avl_insert(avl, aux, where);
916 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
922 search.aux_guid = vd->vdev_guid;
923 aux = avl_find(avl, &search, &where);
927 if (--aux->aux_count == 0) {
928 avl_remove(avl, aux);
929 kmem_free(aux, sizeof (spa_aux_t));
930 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
931 aux->aux_pool = 0ULL;
936 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
938 spa_aux_t search, *found;
940 search.aux_guid = guid;
941 found = avl_find(avl, &search, NULL);
945 *pool = found->aux_pool;
952 *refcnt = found->aux_count;
957 return (found != NULL);
961 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
963 spa_aux_t search, *found;
966 search.aux_guid = vd->vdev_guid;
967 found = avl_find(avl, &search, &where);
968 ASSERT(found != NULL);
969 ASSERT(found->aux_pool == 0ULL);
971 found->aux_pool = spa_guid(vd->vdev_spa);
975 * Spares are tracked globally due to the following constraints:
977 * - A spare may be part of multiple pools.
978 * - A spare may be added to a pool even if it's actively in use within
980 * - A spare in use in any pool can only be the source of a replacement if
981 * the target is a spare in the same pool.
983 * We keep track of all spares on the system through the use of a reference
984 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
985 * spare, then we bump the reference count in the AVL tree. In addition, we set
986 * the 'vdev_isspare' member to indicate that the device is a spare (active or
987 * inactive). When a spare is made active (used to replace a device in the
988 * pool), we also keep track of which pool its been made a part of.
990 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
991 * called under the spa_namespace lock as part of vdev reconfiguration. The
992 * separate spare lock exists for the status query path, which does not need to
993 * be completely consistent with respect to other vdev configuration changes.
997 spa_spare_compare(const void *a, const void *b)
999 return (spa_aux_compare(a, b));
1003 spa_spare_add(vdev_t *vd)
1005 mutex_enter(&spa_spare_lock);
1006 ASSERT(!vd->vdev_isspare);
1007 spa_aux_add(vd, &spa_spare_avl);
1008 vd->vdev_isspare = B_TRUE;
1009 mutex_exit(&spa_spare_lock);
1013 spa_spare_remove(vdev_t *vd)
1015 mutex_enter(&spa_spare_lock);
1016 ASSERT(vd->vdev_isspare);
1017 spa_aux_remove(vd, &spa_spare_avl);
1018 vd->vdev_isspare = B_FALSE;
1019 mutex_exit(&spa_spare_lock);
1023 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
1027 mutex_enter(&spa_spare_lock);
1028 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
1029 mutex_exit(&spa_spare_lock);
1035 spa_spare_activate(vdev_t *vd)
1037 mutex_enter(&spa_spare_lock);
1038 ASSERT(vd->vdev_isspare);
1039 spa_aux_activate(vd, &spa_spare_avl);
1040 mutex_exit(&spa_spare_lock);
1044 * Level 2 ARC devices are tracked globally for the same reasons as spares.
1045 * Cache devices currently only support one pool per cache device, and so
1046 * for these devices the aux reference count is currently unused beyond 1.
1050 spa_l2cache_compare(const void *a, const void *b)
1052 return (spa_aux_compare(a, b));
1056 spa_l2cache_add(vdev_t *vd)
1058 mutex_enter(&spa_l2cache_lock);
1059 ASSERT(!vd->vdev_isl2cache);
1060 spa_aux_add(vd, &spa_l2cache_avl);
1061 vd->vdev_isl2cache = B_TRUE;
1062 mutex_exit(&spa_l2cache_lock);
1066 spa_l2cache_remove(vdev_t *vd)
1068 mutex_enter(&spa_l2cache_lock);
1069 ASSERT(vd->vdev_isl2cache);
1070 spa_aux_remove(vd, &spa_l2cache_avl);
1071 vd->vdev_isl2cache = B_FALSE;
1072 mutex_exit(&spa_l2cache_lock);
1076 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
1080 mutex_enter(&spa_l2cache_lock);
1081 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
1082 mutex_exit(&spa_l2cache_lock);
1088 spa_l2cache_activate(vdev_t *vd)
1090 mutex_enter(&spa_l2cache_lock);
1091 ASSERT(vd->vdev_isl2cache);
1092 spa_aux_activate(vd, &spa_l2cache_avl);
1093 mutex_exit(&spa_l2cache_lock);
1097 * ==========================================================================
1099 * ==========================================================================
1103 * Lock the given spa_t for the purpose of adding or removing a vdev.
1104 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1105 * It returns the next transaction group for the spa_t.
1108 spa_vdev_enter(spa_t *spa)
1110 mutex_enter(&spa->spa_vdev_top_lock);
1111 mutex_enter(&spa_namespace_lock);
1112 return (spa_vdev_config_enter(spa));
1116 * Internal implementation for spa_vdev_enter(). Used when a vdev
1117 * operation requires multiple syncs (i.e. removing a device) while
1118 * keeping the spa_namespace_lock held.
1121 spa_vdev_config_enter(spa_t *spa)
1123 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1125 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1127 return (spa_last_synced_txg(spa) + 1);
1131 * Used in combination with spa_vdev_config_enter() to allow the syncing
1132 * of multiple transactions without releasing the spa_namespace_lock.
1135 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
1137 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1139 int config_changed = B_FALSE;
1141 ASSERT(txg > spa_last_synced_txg(spa));
1143 spa->spa_pending_vdev = NULL;
1146 * Reassess the DTLs.
1148 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
1150 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
1151 config_changed = B_TRUE;
1152 spa->spa_config_generation++;
1156 * Verify the metaslab classes.
1158 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
1159 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
1161 spa_config_exit(spa, SCL_ALL, spa);
1164 * Panic the system if the specified tag requires it. This
1165 * is useful for ensuring that configurations are updated
1168 if (zio_injection_enabled)
1169 zio_handle_panic_injection(spa, tag, 0);
1172 * Note: this txg_wait_synced() is important because it ensures
1173 * that there won't be more than one config change per txg.
1174 * This allows us to use the txg as the generation number.
1177 txg_wait_synced(spa->spa_dsl_pool, txg);
1180 ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
1181 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1183 spa_config_exit(spa, SCL_ALL, spa);
1187 * If the config changed, update the config cache.
1190 spa_config_sync(spa, B_FALSE, B_TRUE);
1194 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1195 * locking of spa_vdev_enter(), we also want make sure the transactions have
1196 * synced to disk, and then update the global configuration cache with the new
1200 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1202 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1203 mutex_exit(&spa_namespace_lock);
1204 mutex_exit(&spa->spa_vdev_top_lock);
1210 * Lock the given spa_t for the purpose of changing vdev state.
1213 spa_vdev_state_enter(spa_t *spa, int oplocks)
1215 int locks = SCL_STATE_ALL | oplocks;
1218 * Root pools may need to read of the underlying devfs filesystem
1219 * when opening up a vdev. Unfortunately if we're holding the
1220 * SCL_ZIO lock it will result in a deadlock when we try to issue
1221 * the read from the root filesystem. Instead we "prefetch"
1222 * the associated vnodes that we need prior to opening the
1223 * underlying devices and cache them so that we can prevent
1224 * any I/O when we are doing the actual open.
1226 if (spa_is_root(spa)) {
1227 int low = locks & ~(SCL_ZIO - 1);
1228 int high = locks & ~low;
1230 spa_config_enter(spa, high, spa, RW_WRITER);
1231 vdev_hold(spa->spa_root_vdev);
1232 spa_config_enter(spa, low, spa, RW_WRITER);
1234 spa_config_enter(spa, locks, spa, RW_WRITER);
1236 spa->spa_vdev_locks = locks;
1240 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1242 boolean_t config_changed = B_FALSE;
1244 if (vd != NULL || error == 0)
1245 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1249 vdev_state_dirty(vd->vdev_top);
1250 config_changed = B_TRUE;
1251 spa->spa_config_generation++;
1254 if (spa_is_root(spa))
1255 vdev_rele(spa->spa_root_vdev);
1257 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1258 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1261 * If anything changed, wait for it to sync. This ensures that,
1262 * from the system administrator's perspective, zpool(1M) commands
1263 * are synchronous. This is important for things like zpool offline:
1264 * when the command completes, you expect no further I/O from ZFS.
1267 txg_wait_synced(spa->spa_dsl_pool, 0);
1270 * If the config changed, update the config cache.
1272 if (config_changed) {
1273 mutex_enter(&spa_namespace_lock);
1274 spa_config_sync(spa, B_FALSE, B_TRUE);
1275 mutex_exit(&spa_namespace_lock);
1282 * ==========================================================================
1283 * Miscellaneous functions
1284 * ==========================================================================
1288 spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx)
1290 if (!nvlist_exists(spa->spa_label_features, feature)) {
1291 fnvlist_add_boolean(spa->spa_label_features, feature);
1293 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1294 * dirty the vdev config because lock SCL_CONFIG is not held.
1295 * Thankfully, in this case we don't need to dirty the config
1296 * because it will be written out anyway when we finish
1297 * creating the pool.
1299 if (tx->tx_txg != TXG_INITIAL)
1300 vdev_config_dirty(spa->spa_root_vdev);
1305 spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1307 if (nvlist_remove_all(spa->spa_label_features, feature) == 0)
1308 vdev_config_dirty(spa->spa_root_vdev);
1315 spa_rename(const char *name, const char *newname)
1321 * Lookup the spa_t and grab the config lock for writing. We need to
1322 * actually open the pool so that we can sync out the necessary labels.
1323 * It's OK to call spa_open() with the namespace lock held because we
1324 * allow recursive calls for other reasons.
1326 mutex_enter(&spa_namespace_lock);
1327 if ((err = spa_open(name, &spa, FTAG)) != 0) {
1328 mutex_exit(&spa_namespace_lock);
1332 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1334 avl_remove(&spa_namespace_avl, spa);
1335 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1336 avl_add(&spa_namespace_avl, spa);
1339 * Sync all labels to disk with the new names by marking the root vdev
1340 * dirty and waiting for it to sync. It will pick up the new pool name
1343 vdev_config_dirty(spa->spa_root_vdev);
1345 spa_config_exit(spa, SCL_ALL, FTAG);
1347 txg_wait_synced(spa->spa_dsl_pool, 0);
1350 * Sync the updated config cache.
1352 spa_config_sync(spa, B_FALSE, B_TRUE);
1354 spa_close(spa, FTAG);
1356 mutex_exit(&spa_namespace_lock);
1362 * Return the spa_t associated with given pool_guid, if it exists. If
1363 * device_guid is non-zero, determine whether the pool exists *and* contains
1364 * a device with the specified device_guid.
1367 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1370 avl_tree_t *t = &spa_namespace_avl;
1372 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1374 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1375 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1377 if (spa->spa_root_vdev == NULL)
1379 if (spa_guid(spa) == pool_guid) {
1380 if (device_guid == 0)
1383 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1384 device_guid) != NULL)
1388 * Check any devices we may be in the process of adding.
1390 if (spa->spa_pending_vdev) {
1391 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1392 device_guid) != NULL)
1402 * Determine whether a pool with the given pool_guid exists.
1405 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1407 return (spa_by_guid(pool_guid, device_guid) != NULL);
1411 spa_strdup(const char *s)
1417 new = kmem_alloc(len + 1, KM_SLEEP);
1425 spa_strfree(char *s)
1427 kmem_free(s, strlen(s) + 1);
1431 spa_get_random(uint64_t range)
1437 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1443 spa_generate_guid(spa_t *spa)
1445 uint64_t guid = spa_get_random(-1ULL);
1448 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1449 guid = spa_get_random(-1ULL);
1451 while (guid == 0 || spa_guid_exists(guid, 0))
1452 guid = spa_get_random(-1ULL);
1459 snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp)
1462 char *checksum = NULL;
1463 char *compress = NULL;
1466 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1467 dmu_object_byteswap_t bswap =
1468 DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1469 (void) snprintf(type, sizeof (type), "bswap %s %s",
1470 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1471 "metadata" : "data",
1472 dmu_ot_byteswap[bswap].ob_name);
1474 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1477 if (!BP_IS_EMBEDDED(bp)) {
1479 zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1481 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1484 SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum,
1489 spa_freeze(spa_t *spa)
1491 uint64_t freeze_txg = 0;
1493 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1494 if (spa->spa_freeze_txg == UINT64_MAX) {
1495 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1496 spa->spa_freeze_txg = freeze_txg;
1498 spa_config_exit(spa, SCL_ALL, FTAG);
1499 if (freeze_txg != 0)
1500 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1504 zfs_panic_recover(const char *fmt, ...)
1509 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1514 * This is a stripped-down version of strtoull, suitable only for converting
1515 * lowercase hexadecimal numbers that don't overflow.
1518 zfs_strtonum(const char *str, char **nptr)
1524 while ((c = *str) != '\0') {
1525 if (c >= '0' && c <= '9')
1527 else if (c >= 'a' && c <= 'f')
1528 digit = 10 + c - 'a';
1539 *nptr = (char *)str;
1545 * ==========================================================================
1546 * Accessor functions
1547 * ==========================================================================
1551 spa_shutting_down(spa_t *spa)
1553 return (spa->spa_async_suspended);
1557 spa_get_dsl(spa_t *spa)
1559 return (spa->spa_dsl_pool);
1563 spa_is_initializing(spa_t *spa)
1565 return (spa->spa_is_initializing);
1569 spa_get_rootblkptr(spa_t *spa)
1571 return (&spa->spa_ubsync.ub_rootbp);
1575 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1577 spa->spa_uberblock.ub_rootbp = *bp;
1581 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1583 if (spa->spa_root == NULL)
1586 (void) strncpy(buf, spa->spa_root, buflen);
1590 spa_sync_pass(spa_t *spa)
1592 return (spa->spa_sync_pass);
1596 spa_name(spa_t *spa)
1598 return (spa->spa_name);
1602 spa_guid(spa_t *spa)
1604 dsl_pool_t *dp = spa_get_dsl(spa);
1608 * If we fail to parse the config during spa_load(), we can go through
1609 * the error path (which posts an ereport) and end up here with no root
1610 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1613 if (spa->spa_root_vdev == NULL)
1614 return (spa->spa_config_guid);
1616 guid = spa->spa_last_synced_guid != 0 ?
1617 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1620 * Return the most recently synced out guid unless we're
1621 * in syncing context.
1623 if (dp && dsl_pool_sync_context(dp))
1624 return (spa->spa_root_vdev->vdev_guid);
1630 spa_load_guid(spa_t *spa)
1633 * This is a GUID that exists solely as a reference for the
1634 * purposes of the arc. It is generated at load time, and
1635 * is never written to persistent storage.
1637 return (spa->spa_load_guid);
1641 spa_last_synced_txg(spa_t *spa)
1643 return (spa->spa_ubsync.ub_txg);
1647 spa_first_txg(spa_t *spa)
1649 return (spa->spa_first_txg);
1653 spa_syncing_txg(spa_t *spa)
1655 return (spa->spa_syncing_txg);
1659 spa_state(spa_t *spa)
1661 return (spa->spa_state);
1665 spa_load_state(spa_t *spa)
1667 return (spa->spa_load_state);
1671 spa_freeze_txg(spa_t *spa)
1673 return (spa->spa_freeze_txg);
1678 spa_get_asize(spa_t *spa, uint64_t lsize)
1680 return (lsize * spa_asize_inflation);
1684 * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%),
1687 * See the comment above spa_slop_shift for details.
1690 spa_get_slop_space(spa_t *spa) {
1691 uint64_t space = spa_get_dspace(spa);
1692 return (MAX(space >> spa_slop_shift, SPA_MINDEVSIZE >> 1));
1696 spa_get_dspace(spa_t *spa)
1698 return (spa->spa_dspace);
1702 spa_update_dspace(spa_t *spa)
1704 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1705 ddt_get_dedup_dspace(spa);
1709 * Return the failure mode that has been set to this pool. The default
1710 * behavior will be to block all I/Os when a complete failure occurs.
1713 spa_get_failmode(spa_t *spa)
1715 return (spa->spa_failmode);
1719 spa_suspended(spa_t *spa)
1721 return (spa->spa_suspended);
1725 spa_version(spa_t *spa)
1727 return (spa->spa_ubsync.ub_version);
1731 spa_deflate(spa_t *spa)
1733 return (spa->spa_deflate);
1737 spa_normal_class(spa_t *spa)
1739 return (spa->spa_normal_class);
1743 spa_log_class(spa_t *spa)
1745 return (spa->spa_log_class);
1749 spa_evicting_os_register(spa_t *spa, objset_t *os)
1751 mutex_enter(&spa->spa_evicting_os_lock);
1752 list_insert_head(&spa->spa_evicting_os_list, os);
1753 mutex_exit(&spa->spa_evicting_os_lock);
1757 spa_evicting_os_deregister(spa_t *spa, objset_t *os)
1759 mutex_enter(&spa->spa_evicting_os_lock);
1760 list_remove(&spa->spa_evicting_os_list, os);
1761 cv_broadcast(&spa->spa_evicting_os_cv);
1762 mutex_exit(&spa->spa_evicting_os_lock);
1766 spa_evicting_os_wait(spa_t *spa)
1768 mutex_enter(&spa->spa_evicting_os_lock);
1769 while (!list_is_empty(&spa->spa_evicting_os_list))
1770 cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock);
1771 mutex_exit(&spa->spa_evicting_os_lock);
1773 dmu_buf_user_evict_wait();
1777 spa_max_replication(spa_t *spa)
1780 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1781 * handle BPs with more than one DVA allocated. Set our max
1782 * replication level accordingly.
1784 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1786 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1790 spa_prev_software_version(spa_t *spa)
1792 return (spa->spa_prev_software_version);
1796 spa_deadman_synctime(spa_t *spa)
1798 return (spa->spa_deadman_synctime);
1802 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1804 uint64_t asize = DVA_GET_ASIZE(dva);
1805 uint64_t dsize = asize;
1807 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1809 if (asize != 0 && spa->spa_deflate) {
1810 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1811 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1818 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1822 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1823 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1829 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1833 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1835 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1836 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1838 spa_config_exit(spa, SCL_VDEV, FTAG);
1844 * ==========================================================================
1845 * Initialization and Termination
1846 * ==========================================================================
1850 spa_name_compare(const void *a1, const void *a2)
1852 const spa_t *s1 = a1;
1853 const spa_t *s2 = a2;
1856 s = strcmp(s1->spa_name, s2->spa_name);
1867 return (spa_active_count);
1877 EVENTHANDLER_DEFINE(mountroot, spa_boot_init, NULL, 0);
1883 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1884 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1885 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1886 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1888 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1889 offsetof(spa_t, spa_avl));
1891 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1892 offsetof(spa_aux_t, aux_avl));
1894 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1895 offsetof(spa_aux_t, aux_avl));
1897 spa_mode_global = mode;
1903 if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
1904 arc_procfd = open("/proc/self/ctl", O_WRONLY);
1905 if (arc_procfd == -1) {
1906 perror("could not enable watchpoints: "
1907 "opening /proc/self/ctl failed: ");
1913 #endif /* illumos */
1921 vdev_cache_stat_init();
1924 zpool_feature_init();
1931 #endif /* !illumos */
1941 vdev_cache_stat_fini();
1950 avl_destroy(&spa_namespace_avl);
1951 avl_destroy(&spa_spare_avl);
1952 avl_destroy(&spa_l2cache_avl);
1954 cv_destroy(&spa_namespace_cv);
1955 mutex_destroy(&spa_namespace_lock);
1956 mutex_destroy(&spa_spare_lock);
1957 mutex_destroy(&spa_l2cache_lock);
1961 * Return whether this pool has slogs. No locking needed.
1962 * It's not a problem if the wrong answer is returned as it's only for
1963 * performance and not correctness
1966 spa_has_slogs(spa_t *spa)
1968 return (spa->spa_log_class->mc_rotor != NULL);
1972 spa_get_log_state(spa_t *spa)
1974 return (spa->spa_log_state);
1978 spa_set_log_state(spa_t *spa, spa_log_state_t state)
1980 spa->spa_log_state = state;
1984 spa_is_root(spa_t *spa)
1986 return (spa->spa_is_root);
1990 spa_writeable(spa_t *spa)
1992 return (!!(spa->spa_mode & FWRITE));
1996 * Returns true if there is a pending sync task in any of the current
1997 * syncing txg, the current quiescing txg, or the current open txg.
2000 spa_has_pending_synctask(spa_t *spa)
2002 return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks));
2006 spa_mode(spa_t *spa)
2008 return (spa->spa_mode);
2012 spa_bootfs(spa_t *spa)
2014 return (spa->spa_bootfs);
2018 spa_delegation(spa_t *spa)
2020 return (spa->spa_delegation);
2024 spa_meta_objset(spa_t *spa)
2026 return (spa->spa_meta_objset);
2030 spa_dedup_checksum(spa_t *spa)
2032 return (spa->spa_dedup_checksum);
2036 * Reset pool scan stat per scan pass (or reboot).
2039 spa_scan_stat_init(spa_t *spa)
2041 /* data not stored on disk */
2042 spa->spa_scan_pass_start = gethrestime_sec();
2043 spa->spa_scan_pass_exam = 0;
2044 vdev_scan_stat_init(spa->spa_root_vdev);
2048 * Get scan stats for zpool status reports
2051 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
2053 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
2055 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
2056 return (SET_ERROR(ENOENT));
2057 bzero(ps, sizeof (pool_scan_stat_t));
2059 /* data stored on disk */
2060 ps->pss_func = scn->scn_phys.scn_func;
2061 ps->pss_start_time = scn->scn_phys.scn_start_time;
2062 ps->pss_end_time = scn->scn_phys.scn_end_time;
2063 ps->pss_to_examine = scn->scn_phys.scn_to_examine;
2064 ps->pss_examined = scn->scn_phys.scn_examined;
2065 ps->pss_to_process = scn->scn_phys.scn_to_process;
2066 ps->pss_processed = scn->scn_phys.scn_processed;
2067 ps->pss_errors = scn->scn_phys.scn_errors;
2068 ps->pss_state = scn->scn_phys.scn_state;
2070 /* data not stored on disk */
2071 ps->pss_pass_start = spa->spa_scan_pass_start;
2072 ps->pss_pass_exam = spa->spa_scan_pass_exam;
2078 spa_debug_enabled(spa_t *spa)
2080 return (spa->spa_debug);
2084 spa_maxblocksize(spa_t *spa)
2086 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS))
2087 return (SPA_MAXBLOCKSIZE);
2089 return (SPA_OLD_MAXBLOCKSIZE);