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, 2017 by Delphix. All rights reserved.
24 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
26 * Copyright 2013 Saso Kiselkov. All rights reserved.
27 * Copyright (c) 2017 Datto Inc.
30 #include <sys/zfs_context.h>
31 #include <sys/spa_impl.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/vdev_file.h>
41 #include <sys/vdev_raidz.h>
42 #include <sys/metaslab.h>
43 #include <sys/uberblock_impl.h>
46 #include <sys/unique.h>
47 #include <sys/dsl_pool.h>
48 #include <sys/dsl_dir.h>
49 #include <sys/dsl_prop.h>
50 #include <sys/fm/util.h>
51 #include <sys/dsl_scan.h>
52 #include <sys/fs/zfs.h>
53 #include <sys/metaslab_impl.h>
56 #include <sys/kstat.h>
58 #include <sys/zfeature.h>
64 * There are four basic locks for managing spa_t structures:
66 * spa_namespace_lock (global mutex)
68 * This lock must be acquired to do any of the following:
70 * - Lookup a spa_t by name
71 * - Add or remove a spa_t from the namespace
72 * - Increase spa_refcount from non-zero
73 * - Check if spa_refcount is zero
75 * - add/remove/attach/detach devices
76 * - Held for the duration of create/destroy/import/export
78 * It does not need to handle recursion. A create or destroy may
79 * reference objects (files or zvols) in other pools, but by
80 * definition they must have an existing reference, and will never need
81 * to lookup a spa_t by name.
83 * spa_refcount (per-spa refcount_t protected by mutex)
85 * This reference count keep track of any active users of the spa_t. The
86 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
87 * the refcount is never really 'zero' - opening a pool implicitly keeps
88 * some references in the DMU. Internally we check against spa_minref, but
89 * present the image of a zero/non-zero value to consumers.
91 * spa_config_lock[] (per-spa array of rwlocks)
93 * This protects the spa_t from config changes, and must be held in
94 * the following circumstances:
96 * - RW_READER to perform I/O to the spa
97 * - RW_WRITER to change the vdev config
99 * The locking order is fairly straightforward:
101 * spa_namespace_lock -> spa_refcount
103 * The namespace lock must be acquired to increase the refcount from 0
104 * or to check if it is zero.
106 * spa_refcount -> spa_config_lock[]
108 * There must be at least one valid reference on the spa_t to acquire
111 * spa_namespace_lock -> spa_config_lock[]
113 * The namespace lock must always be taken before the config lock.
116 * The spa_namespace_lock can be acquired directly and is globally visible.
118 * The namespace is manipulated using the following functions, all of which
119 * require the spa_namespace_lock to be held.
121 * spa_lookup() Lookup a spa_t by name.
123 * spa_add() Create a new spa_t in the namespace.
125 * spa_remove() Remove a spa_t from the namespace. This also
126 * frees up any memory associated with the spa_t.
128 * spa_next() Returns the next spa_t in the system, or the
129 * first if NULL is passed.
131 * spa_evict_all() Shutdown and remove all spa_t structures in
134 * spa_guid_exists() Determine whether a pool/device guid exists.
136 * The spa_refcount is manipulated using the following functions:
138 * spa_open_ref() Adds a reference to the given spa_t. Must be
139 * called with spa_namespace_lock held if the
140 * refcount is currently zero.
142 * spa_close() Remove a reference from the spa_t. This will
143 * not free the spa_t or remove it from the
144 * namespace. No locking is required.
146 * spa_refcount_zero() Returns true if the refcount is currently
147 * zero. Must be called with spa_namespace_lock
150 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
151 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
152 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
154 * To read the configuration, it suffices to hold one of these locks as reader.
155 * To modify the configuration, you must hold all locks as writer. To modify
156 * vdev state without altering the vdev tree's topology (e.g. online/offline),
157 * you must hold SCL_STATE and SCL_ZIO as writer.
159 * We use these distinct config locks to avoid recursive lock entry.
160 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
161 * block allocations (SCL_ALLOC), which may require reading space maps
162 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
164 * The spa config locks cannot be normal rwlocks because we need the
165 * ability to hand off ownership. For example, SCL_ZIO is acquired
166 * by the issuing thread and later released by an interrupt thread.
167 * They do, however, obey the usual write-wanted semantics to prevent
168 * writer (i.e. system administrator) starvation.
170 * The lock acquisition rules are as follows:
173 * Protects changes to the vdev tree topology, such as vdev
174 * add/remove/attach/detach. Protects the dirty config list
175 * (spa_config_dirty_list) and the set of spares and l2arc devices.
178 * Protects changes to pool state and vdev state, such as vdev
179 * online/offline/fault/degrade/clear. Protects the dirty state list
180 * (spa_state_dirty_list) and global pool state (spa_state).
183 * Protects changes to metaslab groups and classes.
184 * Held as reader by metaslab_alloc() and metaslab_claim().
187 * Held by bp-level zios (those which have no io_vd upon entry)
188 * to prevent changes to the vdev tree. The bp-level zio implicitly
189 * protects all of its vdev child zios, which do not hold SCL_ZIO.
192 * Protects changes to metaslab groups and classes.
193 * Held as reader by metaslab_free(). SCL_FREE is distinct from
194 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
195 * blocks in zio_done() while another i/o that holds either
196 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
199 * Held as reader to prevent changes to the vdev tree during trivial
200 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
201 * other locks, and lower than all of them, to ensure that it's safe
202 * to acquire regardless of caller context.
204 * In addition, the following rules apply:
206 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
207 * The lock ordering is SCL_CONFIG > spa_props_lock.
209 * (b) I/O operations on leaf vdevs. For any zio operation that takes
210 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
211 * or zio_write_phys() -- the caller must ensure that the config cannot
212 * cannot change in the interim, and that the vdev cannot be reopened.
213 * SCL_STATE as reader suffices for both.
215 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
217 * spa_vdev_enter() Acquire the namespace lock and the config lock
220 * spa_vdev_exit() Release the config lock, wait for all I/O
221 * to complete, sync the updated configs to the
222 * cache, and release the namespace lock.
224 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
225 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
226 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
228 * spa_rename() is also implemented within this file since it requires
229 * manipulation of the namespace.
232 static avl_tree_t spa_namespace_avl;
233 kmutex_t spa_namespace_lock;
234 static kcondvar_t spa_namespace_cv;
235 int spa_max_replication_override = SPA_DVAS_PER_BP;
237 static kmutex_t spa_spare_lock;
238 static avl_tree_t spa_spare_avl;
239 static kmutex_t spa_l2cache_lock;
240 static avl_tree_t spa_l2cache_avl;
242 kmem_cache_t *spa_buffer_pool;
246 int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SET_ERROR | ZFS_DEBUG_SPA);
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 int zfs_recover = B_FALSE;
261 * If destroy encounters an EIO while reading metadata (e.g. indirect
262 * blocks), space referenced by the missing metadata can not be freed.
263 * Normally this causes the background destroy to become "stalled", as
264 * it is unable to make forward progress. While in this stalled state,
265 * all remaining space to free from the error-encountering filesystem is
266 * "temporarily leaked". Set this flag to cause it to ignore the EIO,
267 * permanently leak the space from indirect blocks that can not be read,
268 * and continue to free everything else that it can.
270 * The default, "stalling" behavior is useful if the storage partially
271 * fails (i.e. some but not all i/os fail), and then later recovers. In
272 * this case, we will be able to continue pool operations while it is
273 * partially failed, and when it recovers, we can continue to free the
274 * space, with no leaks. However, note that this case is actually
277 * Typically pools either (a) fail completely (but perhaps temporarily,
278 * e.g. a top-level vdev going offline), or (b) have localized,
279 * permanent errors (e.g. disk returns the wrong data due to bit flip or
280 * firmware bug). In case (a), this setting does not matter because the
281 * pool will be suspended and the sync thread will not be able to make
282 * forward progress regardless. In case (b), because the error is
283 * permanent, the best we can do is leak the minimum amount of space,
284 * which is what setting this flag will do. Therefore, it is reasonable
285 * for this flag to normally be set, but we chose the more conservative
286 * approach of not setting it, so that there is no possibility of
287 * leaking space in the "partial temporary" failure case.
289 int zfs_free_leak_on_eio = B_FALSE;
292 * Expiration time in milliseconds. This value has two meanings. First it is
293 * used to determine when the spa_deadman() logic should fire. By default the
294 * spa_deadman() will fire if spa_sync() has not completed in 600 seconds.
295 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
296 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
297 * in one of three behaviors controlled by zfs_deadman_failmode.
299 unsigned long zfs_deadman_synctime_ms = 600000ULL;
302 * This value controls the maximum amount of time zio_wait() will block for an
303 * outstanding IO. By default this is 300 seconds at which point the "hung"
304 * behavior will be applied as described for zfs_deadman_synctime_ms.
306 unsigned long zfs_deadman_ziotime_ms = 300000ULL;
309 * Check time in milliseconds. This defines the frequency at which we check
312 unsigned long zfs_deadman_checktime_ms = 60000ULL;
315 * By default the deadman is enabled.
317 int zfs_deadman_enabled = 1;
320 * Controls the behavior of the deadman when it detects a "hung" I/O.
321 * Valid values are zfs_deadman_failmode=<wait|continue|panic>.
323 * wait - Wait for the "hung" I/O (default)
324 * continue - Attempt to recover from a "hung" I/O
325 * panic - Panic the system
327 char *zfs_deadman_failmode = "wait";
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;
341 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
342 * the pool to be consumed. This ensures that we don't run the pool
343 * completely out of space, due to unaccounted changes (e.g. to the MOS).
344 * It also limits the worst-case time to allocate space. If we have
345 * less than this amount of free space, most ZPL operations (e.g. write,
346 * create) will return ENOSPC.
348 * Certain operations (e.g. file removal, most administrative actions) can
349 * use half the slop space. They will only return ENOSPC if less than half
350 * the slop space is free. Typically, once the pool has less than the slop
351 * space free, the user will use these operations to free up space in the pool.
352 * These are the operations that call dsl_pool_adjustedsize() with the netfree
353 * argument set to TRUE.
355 * A very restricted set of operations are always permitted, regardless of
356 * the amount of free space. These are the operations that call
357 * dsl_sync_task(ZFS_SPACE_CHECK_NONE), e.g. "zfs destroy". If these
358 * operations result in a net increase in the amount of space used,
359 * it is possible to run the pool completely out of space, causing it to
360 * be permanently read-only.
362 * Note that on very small pools, the slop space will be larger than
363 * 3.2%, in an effort to have it be at least spa_min_slop (128MB),
364 * but we never allow it to be more than half the pool size.
366 * See also the comments in zfs_space_check_t.
368 int spa_slop_shift = 5;
369 uint64_t spa_min_slop = 128 * 1024 * 1024;
372 * ==========================================================================
374 * ==========================================================================
377 spa_config_lock_init(spa_t *spa)
379 for (int i = 0; i < SCL_LOCKS; i++) {
380 spa_config_lock_t *scl = &spa->spa_config_lock[i];
381 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
382 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
383 refcount_create_untracked(&scl->scl_count);
384 scl->scl_writer = NULL;
385 scl->scl_write_wanted = 0;
390 spa_config_lock_destroy(spa_t *spa)
392 for (int i = 0; i < SCL_LOCKS; i++) {
393 spa_config_lock_t *scl = &spa->spa_config_lock[i];
394 mutex_destroy(&scl->scl_lock);
395 cv_destroy(&scl->scl_cv);
396 refcount_destroy(&scl->scl_count);
397 ASSERT(scl->scl_writer == NULL);
398 ASSERT(scl->scl_write_wanted == 0);
403 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
405 for (int i = 0; i < SCL_LOCKS; i++) {
406 spa_config_lock_t *scl = &spa->spa_config_lock[i];
407 if (!(locks & (1 << i)))
409 mutex_enter(&scl->scl_lock);
410 if (rw == RW_READER) {
411 if (scl->scl_writer || scl->scl_write_wanted) {
412 mutex_exit(&scl->scl_lock);
413 spa_config_exit(spa, locks & ((1 << i) - 1),
418 ASSERT(scl->scl_writer != curthread);
419 if (!refcount_is_zero(&scl->scl_count)) {
420 mutex_exit(&scl->scl_lock);
421 spa_config_exit(spa, locks & ((1 << i) - 1),
425 scl->scl_writer = curthread;
427 (void) refcount_add(&scl->scl_count, tag);
428 mutex_exit(&scl->scl_lock);
434 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
438 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
440 for (int i = 0; i < SCL_LOCKS; i++) {
441 spa_config_lock_t *scl = &spa->spa_config_lock[i];
442 if (scl->scl_writer == curthread)
443 wlocks_held |= (1 << i);
444 if (!(locks & (1 << i)))
446 mutex_enter(&scl->scl_lock);
447 if (rw == RW_READER) {
448 while (scl->scl_writer || scl->scl_write_wanted) {
449 cv_wait(&scl->scl_cv, &scl->scl_lock);
452 ASSERT(scl->scl_writer != curthread);
453 while (!refcount_is_zero(&scl->scl_count)) {
454 scl->scl_write_wanted++;
455 cv_wait(&scl->scl_cv, &scl->scl_lock);
456 scl->scl_write_wanted--;
458 scl->scl_writer = curthread;
460 (void) refcount_add(&scl->scl_count, tag);
461 mutex_exit(&scl->scl_lock);
463 ASSERT(wlocks_held <= locks);
467 spa_config_exit(spa_t *spa, int locks, void *tag)
469 for (int i = SCL_LOCKS - 1; i >= 0; i--) {
470 spa_config_lock_t *scl = &spa->spa_config_lock[i];
471 if (!(locks & (1 << i)))
473 mutex_enter(&scl->scl_lock);
474 ASSERT(!refcount_is_zero(&scl->scl_count));
475 if (refcount_remove(&scl->scl_count, tag) == 0) {
476 ASSERT(scl->scl_writer == NULL ||
477 scl->scl_writer == curthread);
478 scl->scl_writer = NULL; /* OK in either case */
479 cv_broadcast(&scl->scl_cv);
481 mutex_exit(&scl->scl_lock);
486 spa_config_held(spa_t *spa, int locks, krw_t rw)
490 for (int i = 0; i < SCL_LOCKS; i++) {
491 spa_config_lock_t *scl = &spa->spa_config_lock[i];
492 if (!(locks & (1 << i)))
494 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
495 (rw == RW_WRITER && scl->scl_writer == curthread))
496 locks_held |= 1 << i;
503 * ==========================================================================
504 * SPA namespace functions
505 * ==========================================================================
509 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
510 * Returns NULL if no matching spa_t is found.
513 spa_lookup(const char *name)
515 static spa_t search; /* spa_t is large; don't allocate on stack */
520 ASSERT(MUTEX_HELD(&spa_namespace_lock));
522 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
525 * If it's a full dataset name, figure out the pool name and
528 cp = strpbrk(search.spa_name, "/@#");
532 spa = avl_find(&spa_namespace_avl, &search, &where);
538 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
539 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
540 * looking for potentially hung I/Os.
543 spa_deadman(void *arg)
547 /* Disable the deadman if the pool is suspended. */
548 if (spa_suspended(spa))
551 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
552 (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
553 ++spa->spa_deadman_calls);
554 if (zfs_deadman_enabled)
555 vdev_deadman(spa->spa_root_vdev, FTAG);
557 spa->spa_deadman_tqid = taskq_dispatch_delay(system_delay_taskq,
558 spa_deadman, spa, TQ_SLEEP, ddi_get_lbolt() +
559 MSEC_TO_TICK(zfs_deadman_checktime_ms));
563 * Create an uninitialized spa_t with the given name. Requires
564 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
565 * exist by calling spa_lookup() first.
568 spa_add(const char *name, nvlist_t *config, const char *altroot)
571 spa_config_dirent_t *dp;
573 ASSERT(MUTEX_HELD(&spa_namespace_lock));
575 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
577 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
578 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
579 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
580 mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL);
581 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
582 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
583 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
584 mutex_init(&spa->spa_cksum_tmpls_lock, NULL, MUTEX_DEFAULT, NULL);
585 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
586 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
587 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
588 mutex_init(&spa->spa_feat_stats_lock, NULL, MUTEX_DEFAULT, NULL);
589 mutex_init(&spa->spa_alloc_lock, NULL, MUTEX_DEFAULT, NULL);
591 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
592 cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL);
593 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
594 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
595 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
597 for (int t = 0; t < TXG_SIZE; t++)
598 bplist_create(&spa->spa_free_bplist[t]);
600 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
601 spa->spa_state = POOL_STATE_UNINITIALIZED;
602 spa->spa_freeze_txg = UINT64_MAX;
603 spa->spa_final_txg = UINT64_MAX;
604 spa->spa_load_max_txg = UINT64_MAX;
606 spa->spa_proc_state = SPA_PROC_NONE;
608 spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
609 spa->spa_deadman_ziotime = MSEC2NSEC(zfs_deadman_ziotime_ms);
610 spa_set_deadman_failmode(spa, zfs_deadman_failmode);
612 refcount_create(&spa->spa_refcount);
613 spa_config_lock_init(spa);
616 avl_add(&spa_namespace_avl, spa);
619 * Set the alternate root, if there is one.
622 spa->spa_root = spa_strdup(altroot);
624 avl_create(&spa->spa_alloc_tree, zio_bookmark_compare,
625 sizeof (zio_t), offsetof(zio_t, io_alloc_node));
628 * Every pool starts with the default cachefile
630 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
631 offsetof(spa_config_dirent_t, scd_link));
633 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
634 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
635 list_insert_head(&spa->spa_config_list, dp);
637 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
640 if (config != NULL) {
643 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
645 VERIFY(nvlist_dup(features, &spa->spa_label_features,
649 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
652 if (spa->spa_label_features == NULL) {
653 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
657 spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0);
659 spa->spa_min_ashift = INT_MAX;
660 spa->spa_max_ashift = 0;
662 /* Reset cached value */
663 spa->spa_dedup_dspace = ~0ULL;
666 * As a pool is being created, treat all features as disabled by
667 * setting SPA_FEATURE_DISABLED for all entries in the feature
670 for (int i = 0; i < SPA_FEATURES; i++) {
671 spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED;
678 * Removes a spa_t from the namespace, freeing up any memory used. Requires
679 * spa_namespace_lock. This is called only after the spa_t has been closed and
683 spa_remove(spa_t *spa)
685 spa_config_dirent_t *dp;
687 ASSERT(MUTEX_HELD(&spa_namespace_lock));
688 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
689 ASSERT3U(refcount_count(&spa->spa_refcount), ==, 0);
691 nvlist_free(spa->spa_config_splitting);
693 avl_remove(&spa_namespace_avl, spa);
694 cv_broadcast(&spa_namespace_cv);
697 spa_strfree(spa->spa_root);
699 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
700 list_remove(&spa->spa_config_list, dp);
701 if (dp->scd_path != NULL)
702 spa_strfree(dp->scd_path);
703 kmem_free(dp, sizeof (spa_config_dirent_t));
706 avl_destroy(&spa->spa_alloc_tree);
707 list_destroy(&spa->spa_config_list);
709 nvlist_free(spa->spa_label_features);
710 nvlist_free(spa->spa_load_info);
711 nvlist_free(spa->spa_feat_stats);
712 spa_config_set(spa, NULL);
714 refcount_destroy(&spa->spa_refcount);
716 spa_stats_destroy(spa);
717 spa_config_lock_destroy(spa);
719 for (int t = 0; t < TXG_SIZE; t++)
720 bplist_destroy(&spa->spa_free_bplist[t]);
722 zio_checksum_templates_free(spa);
724 cv_destroy(&spa->spa_async_cv);
725 cv_destroy(&spa->spa_evicting_os_cv);
726 cv_destroy(&spa->spa_proc_cv);
727 cv_destroy(&spa->spa_scrub_io_cv);
728 cv_destroy(&spa->spa_suspend_cv);
730 mutex_destroy(&spa->spa_alloc_lock);
731 mutex_destroy(&spa->spa_async_lock);
732 mutex_destroy(&spa->spa_errlist_lock);
733 mutex_destroy(&spa->spa_errlog_lock);
734 mutex_destroy(&spa->spa_evicting_os_lock);
735 mutex_destroy(&spa->spa_history_lock);
736 mutex_destroy(&spa->spa_proc_lock);
737 mutex_destroy(&spa->spa_props_lock);
738 mutex_destroy(&spa->spa_cksum_tmpls_lock);
739 mutex_destroy(&spa->spa_scrub_lock);
740 mutex_destroy(&spa->spa_suspend_lock);
741 mutex_destroy(&spa->spa_vdev_top_lock);
742 mutex_destroy(&spa->spa_feat_stats_lock);
744 kmem_free(spa, sizeof (spa_t));
748 * Given a pool, return the next pool in the namespace, or NULL if there is
749 * none. If 'prev' is NULL, return the first pool.
752 spa_next(spa_t *prev)
754 ASSERT(MUTEX_HELD(&spa_namespace_lock));
757 return (AVL_NEXT(&spa_namespace_avl, prev));
759 return (avl_first(&spa_namespace_avl));
763 * ==========================================================================
764 * SPA refcount functions
765 * ==========================================================================
769 * Add a reference to the given spa_t. Must have at least one reference, or
770 * have the namespace lock held.
773 spa_open_ref(spa_t *spa, void *tag)
775 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
776 MUTEX_HELD(&spa_namespace_lock));
777 (void) refcount_add(&spa->spa_refcount, tag);
781 * Remove a reference to the given spa_t. Must have at least one reference, or
782 * have the namespace lock held.
785 spa_close(spa_t *spa, void *tag)
787 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
788 MUTEX_HELD(&spa_namespace_lock));
789 (void) refcount_remove(&spa->spa_refcount, tag);
793 * Remove a reference to the given spa_t held by a dsl dir that is
794 * being asynchronously released. Async releases occur from a taskq
795 * performing eviction of dsl datasets and dirs. The namespace lock
796 * isn't held and the hold by the object being evicted may contribute to
797 * spa_minref (e.g. dataset or directory released during pool export),
798 * so the asserts in spa_close() do not apply.
801 spa_async_close(spa_t *spa, void *tag)
803 (void) refcount_remove(&spa->spa_refcount, tag);
807 * Check to see if the spa refcount is zero. Must be called with
808 * spa_namespace_lock held. We really compare against spa_minref, which is the
809 * number of references acquired when opening a pool
812 spa_refcount_zero(spa_t *spa)
814 ASSERT(MUTEX_HELD(&spa_namespace_lock));
816 return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
820 * ==========================================================================
821 * SPA spare and l2cache tracking
822 * ==========================================================================
826 * Hot spares and cache devices are tracked using the same code below,
827 * for 'auxiliary' devices.
830 typedef struct spa_aux {
838 spa_aux_compare(const void *a, const void *b)
840 const spa_aux_t *sa = (const spa_aux_t *)a;
841 const spa_aux_t *sb = (const spa_aux_t *)b;
843 return (AVL_CMP(sa->aux_guid, sb->aux_guid));
847 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
853 search.aux_guid = vd->vdev_guid;
854 if ((aux = avl_find(avl, &search, &where)) != NULL) {
857 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
858 aux->aux_guid = vd->vdev_guid;
860 avl_insert(avl, aux, where);
865 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
871 search.aux_guid = vd->vdev_guid;
872 aux = avl_find(avl, &search, &where);
876 if (--aux->aux_count == 0) {
877 avl_remove(avl, aux);
878 kmem_free(aux, sizeof (spa_aux_t));
879 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
880 aux->aux_pool = 0ULL;
885 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
887 spa_aux_t search, *found;
889 search.aux_guid = guid;
890 found = avl_find(avl, &search, NULL);
894 *pool = found->aux_pool;
901 *refcnt = found->aux_count;
906 return (found != NULL);
910 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
912 spa_aux_t search, *found;
915 search.aux_guid = vd->vdev_guid;
916 found = avl_find(avl, &search, &where);
917 ASSERT(found != NULL);
918 ASSERT(found->aux_pool == 0ULL);
920 found->aux_pool = spa_guid(vd->vdev_spa);
924 * Spares are tracked globally due to the following constraints:
926 * - A spare may be part of multiple pools.
927 * - A spare may be added to a pool even if it's actively in use within
929 * - A spare in use in any pool can only be the source of a replacement if
930 * the target is a spare in the same pool.
932 * We keep track of all spares on the system through the use of a reference
933 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
934 * spare, then we bump the reference count in the AVL tree. In addition, we set
935 * the 'vdev_isspare' member to indicate that the device is a spare (active or
936 * inactive). When a spare is made active (used to replace a device in the
937 * pool), we also keep track of which pool its been made a part of.
939 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
940 * called under the spa_namespace lock as part of vdev reconfiguration. The
941 * separate spare lock exists for the status query path, which does not need to
942 * be completely consistent with respect to other vdev configuration changes.
946 spa_spare_compare(const void *a, const void *b)
948 return (spa_aux_compare(a, b));
952 spa_spare_add(vdev_t *vd)
954 mutex_enter(&spa_spare_lock);
955 ASSERT(!vd->vdev_isspare);
956 spa_aux_add(vd, &spa_spare_avl);
957 vd->vdev_isspare = B_TRUE;
958 mutex_exit(&spa_spare_lock);
962 spa_spare_remove(vdev_t *vd)
964 mutex_enter(&spa_spare_lock);
965 ASSERT(vd->vdev_isspare);
966 spa_aux_remove(vd, &spa_spare_avl);
967 vd->vdev_isspare = B_FALSE;
968 mutex_exit(&spa_spare_lock);
972 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
976 mutex_enter(&spa_spare_lock);
977 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
978 mutex_exit(&spa_spare_lock);
984 spa_spare_activate(vdev_t *vd)
986 mutex_enter(&spa_spare_lock);
987 ASSERT(vd->vdev_isspare);
988 spa_aux_activate(vd, &spa_spare_avl);
989 mutex_exit(&spa_spare_lock);
993 * Level 2 ARC devices are tracked globally for the same reasons as spares.
994 * Cache devices currently only support one pool per cache device, and so
995 * for these devices the aux reference count is currently unused beyond 1.
999 spa_l2cache_compare(const void *a, const void *b)
1001 return (spa_aux_compare(a, b));
1005 spa_l2cache_add(vdev_t *vd)
1007 mutex_enter(&spa_l2cache_lock);
1008 ASSERT(!vd->vdev_isl2cache);
1009 spa_aux_add(vd, &spa_l2cache_avl);
1010 vd->vdev_isl2cache = B_TRUE;
1011 mutex_exit(&spa_l2cache_lock);
1015 spa_l2cache_remove(vdev_t *vd)
1017 mutex_enter(&spa_l2cache_lock);
1018 ASSERT(vd->vdev_isl2cache);
1019 spa_aux_remove(vd, &spa_l2cache_avl);
1020 vd->vdev_isl2cache = B_FALSE;
1021 mutex_exit(&spa_l2cache_lock);
1025 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
1029 mutex_enter(&spa_l2cache_lock);
1030 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
1031 mutex_exit(&spa_l2cache_lock);
1037 spa_l2cache_activate(vdev_t *vd)
1039 mutex_enter(&spa_l2cache_lock);
1040 ASSERT(vd->vdev_isl2cache);
1041 spa_aux_activate(vd, &spa_l2cache_avl);
1042 mutex_exit(&spa_l2cache_lock);
1046 * ==========================================================================
1048 * ==========================================================================
1052 * Lock the given spa_t for the purpose of adding or removing a vdev.
1053 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1054 * It returns the next transaction group for the spa_t.
1057 spa_vdev_enter(spa_t *spa)
1059 mutex_enter(&spa->spa_vdev_top_lock);
1060 mutex_enter(&spa_namespace_lock);
1061 return (spa_vdev_config_enter(spa));
1065 * Internal implementation for spa_vdev_enter(). Used when a vdev
1066 * operation requires multiple syncs (i.e. removing a device) while
1067 * keeping the spa_namespace_lock held.
1070 spa_vdev_config_enter(spa_t *spa)
1072 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1074 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1076 return (spa_last_synced_txg(spa) + 1);
1080 * Used in combination with spa_vdev_config_enter() to allow the syncing
1081 * of multiple transactions without releasing the spa_namespace_lock.
1084 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
1086 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1088 int config_changed = B_FALSE;
1090 ASSERT(txg > spa_last_synced_txg(spa));
1092 spa->spa_pending_vdev = NULL;
1095 * Reassess the DTLs.
1097 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
1099 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
1100 config_changed = B_TRUE;
1101 spa->spa_config_generation++;
1105 * Verify the metaslab classes.
1107 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
1108 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
1110 spa_config_exit(spa, SCL_ALL, spa);
1113 * Panic the system if the specified tag requires it. This
1114 * is useful for ensuring that configurations are updated
1117 if (zio_injection_enabled)
1118 zio_handle_panic_injection(spa, tag, 0);
1121 * Note: this txg_wait_synced() is important because it ensures
1122 * that there won't be more than one config change per txg.
1123 * This allows us to use the txg as the generation number.
1126 txg_wait_synced(spa->spa_dsl_pool, txg);
1129 ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
1130 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1132 spa_config_exit(spa, SCL_ALL, spa);
1136 * If the config changed, update the config cache.
1139 spa_config_sync(spa, B_FALSE, B_TRUE);
1143 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1144 * locking of spa_vdev_enter(), we also want make sure the transactions have
1145 * synced to disk, and then update the global configuration cache with the new
1149 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1151 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1152 mutex_exit(&spa_namespace_lock);
1153 mutex_exit(&spa->spa_vdev_top_lock);
1159 * Lock the given spa_t for the purpose of changing vdev state.
1162 spa_vdev_state_enter(spa_t *spa, int oplocks)
1164 int locks = SCL_STATE_ALL | oplocks;
1167 * Root pools may need to read of the underlying devfs filesystem
1168 * when opening up a vdev. Unfortunately if we're holding the
1169 * SCL_ZIO lock it will result in a deadlock when we try to issue
1170 * the read from the root filesystem. Instead we "prefetch"
1171 * the associated vnodes that we need prior to opening the
1172 * underlying devices and cache them so that we can prevent
1173 * any I/O when we are doing the actual open.
1175 if (spa_is_root(spa)) {
1176 int low = locks & ~(SCL_ZIO - 1);
1177 int high = locks & ~low;
1179 spa_config_enter(spa, high, spa, RW_WRITER);
1180 vdev_hold(spa->spa_root_vdev);
1181 spa_config_enter(spa, low, spa, RW_WRITER);
1183 spa_config_enter(spa, locks, spa, RW_WRITER);
1185 spa->spa_vdev_locks = locks;
1189 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1191 boolean_t config_changed = B_FALSE;
1194 if (vd == NULL || vd == spa->spa_root_vdev) {
1195 vdev_top = spa->spa_root_vdev;
1197 vdev_top = vd->vdev_top;
1200 if (vd != NULL || error == 0)
1201 vdev_dtl_reassess(vdev_top, 0, 0, B_FALSE);
1204 if (vd != spa->spa_root_vdev)
1205 vdev_state_dirty(vdev_top);
1207 config_changed = B_TRUE;
1208 spa->spa_config_generation++;
1211 if (spa_is_root(spa))
1212 vdev_rele(spa->spa_root_vdev);
1214 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1215 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1218 * If anything changed, wait for it to sync. This ensures that,
1219 * from the system administrator's perspective, zpool(1M) commands
1220 * are synchronous. This is important for things like zpool offline:
1221 * when the command completes, you expect no further I/O from ZFS.
1224 txg_wait_synced(spa->spa_dsl_pool, 0);
1227 * If the config changed, update the config cache.
1229 if (config_changed) {
1230 mutex_enter(&spa_namespace_lock);
1231 spa_config_sync(spa, B_FALSE, B_TRUE);
1232 mutex_exit(&spa_namespace_lock);
1239 * ==========================================================================
1240 * Miscellaneous functions
1241 * ==========================================================================
1245 spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx)
1247 if (!nvlist_exists(spa->spa_label_features, feature)) {
1248 fnvlist_add_boolean(spa->spa_label_features, feature);
1250 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1251 * dirty the vdev config because lock SCL_CONFIG is not held.
1252 * Thankfully, in this case we don't need to dirty the config
1253 * because it will be written out anyway when we finish
1254 * creating the pool.
1256 if (tx->tx_txg != TXG_INITIAL)
1257 vdev_config_dirty(spa->spa_root_vdev);
1262 spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1264 if (nvlist_remove_all(spa->spa_label_features, feature) == 0)
1265 vdev_config_dirty(spa->spa_root_vdev);
1272 spa_rename(const char *name, const char *newname)
1278 * Lookup the spa_t and grab the config lock for writing. We need to
1279 * actually open the pool so that we can sync out the necessary labels.
1280 * It's OK to call spa_open() with the namespace lock held because we
1281 * allow recursive calls for other reasons.
1283 mutex_enter(&spa_namespace_lock);
1284 if ((err = spa_open(name, &spa, FTAG)) != 0) {
1285 mutex_exit(&spa_namespace_lock);
1289 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1291 avl_remove(&spa_namespace_avl, spa);
1292 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1293 avl_add(&spa_namespace_avl, spa);
1296 * Sync all labels to disk with the new names by marking the root vdev
1297 * dirty and waiting for it to sync. It will pick up the new pool name
1300 vdev_config_dirty(spa->spa_root_vdev);
1302 spa_config_exit(spa, SCL_ALL, FTAG);
1304 txg_wait_synced(spa->spa_dsl_pool, 0);
1307 * Sync the updated config cache.
1309 spa_config_sync(spa, B_FALSE, B_TRUE);
1311 spa_close(spa, FTAG);
1313 mutex_exit(&spa_namespace_lock);
1319 * Return the spa_t associated with given pool_guid, if it exists. If
1320 * device_guid is non-zero, determine whether the pool exists *and* contains
1321 * a device with the specified device_guid.
1324 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1327 avl_tree_t *t = &spa_namespace_avl;
1329 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1331 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1332 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1334 if (spa->spa_root_vdev == NULL)
1336 if (spa_guid(spa) == pool_guid) {
1337 if (device_guid == 0)
1340 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1341 device_guid) != NULL)
1345 * Check any devices we may be in the process of adding.
1347 if (spa->spa_pending_vdev) {
1348 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1349 device_guid) != NULL)
1359 * Determine whether a pool with the given pool_guid exists.
1362 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1364 return (spa_by_guid(pool_guid, device_guid) != NULL);
1368 spa_strdup(const char *s)
1374 new = kmem_alloc(len + 1, KM_SLEEP);
1382 spa_strfree(char *s)
1384 kmem_free(s, strlen(s) + 1);
1388 spa_get_random(uint64_t range)
1397 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1403 spa_generate_guid(spa_t *spa)
1405 uint64_t guid = spa_get_random(-1ULL);
1408 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1409 guid = spa_get_random(-1ULL);
1411 while (guid == 0 || spa_guid_exists(guid, 0))
1412 guid = spa_get_random(-1ULL);
1419 snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp)
1422 char *checksum = NULL;
1423 char *compress = NULL;
1426 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1427 dmu_object_byteswap_t bswap =
1428 DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1429 (void) snprintf(type, sizeof (type), "bswap %s %s",
1430 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1431 "metadata" : "data",
1432 dmu_ot_byteswap[bswap].ob_name);
1434 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1437 if (!BP_IS_EMBEDDED(bp)) {
1439 zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1441 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1444 SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum,
1449 spa_freeze(spa_t *spa)
1451 uint64_t freeze_txg = 0;
1453 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1454 if (spa->spa_freeze_txg == UINT64_MAX) {
1455 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1456 spa->spa_freeze_txg = freeze_txg;
1458 spa_config_exit(spa, SCL_ALL, FTAG);
1459 if (freeze_txg != 0)
1460 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1464 zfs_panic_recover(const char *fmt, ...)
1469 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1474 * This is a stripped-down version of strtoull, suitable only for converting
1475 * lowercase hexadecimal numbers that don't overflow.
1478 zfs_strtonum(const char *str, char **nptr)
1484 while ((c = *str) != '\0') {
1485 if (c >= '0' && c <= '9')
1487 else if (c >= 'a' && c <= 'f')
1488 digit = 10 + c - 'a';
1499 *nptr = (char *)str;
1505 * ==========================================================================
1506 * Accessor functions
1507 * ==========================================================================
1511 spa_shutting_down(spa_t *spa)
1513 return (spa->spa_async_suspended);
1517 spa_get_dsl(spa_t *spa)
1519 return (spa->spa_dsl_pool);
1523 spa_is_initializing(spa_t *spa)
1525 return (spa->spa_is_initializing);
1529 spa_get_rootblkptr(spa_t *spa)
1531 return (&spa->spa_ubsync.ub_rootbp);
1535 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1537 spa->spa_uberblock.ub_rootbp = *bp;
1541 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1543 if (spa->spa_root == NULL)
1546 (void) strncpy(buf, spa->spa_root, buflen);
1550 spa_sync_pass(spa_t *spa)
1552 return (spa->spa_sync_pass);
1556 spa_name(spa_t *spa)
1558 return (spa->spa_name);
1562 spa_guid(spa_t *spa)
1564 dsl_pool_t *dp = spa_get_dsl(spa);
1568 * If we fail to parse the config during spa_load(), we can go through
1569 * the error path (which posts an ereport) and end up here with no root
1570 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1573 if (spa->spa_root_vdev == NULL)
1574 return (spa->spa_config_guid);
1576 guid = spa->spa_last_synced_guid != 0 ?
1577 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1580 * Return the most recently synced out guid unless we're
1581 * in syncing context.
1583 if (dp && dsl_pool_sync_context(dp))
1584 return (spa->spa_root_vdev->vdev_guid);
1590 spa_load_guid(spa_t *spa)
1593 * This is a GUID that exists solely as a reference for the
1594 * purposes of the arc. It is generated at load time, and
1595 * is never written to persistent storage.
1597 return (spa->spa_load_guid);
1601 spa_last_synced_txg(spa_t *spa)
1603 return (spa->spa_ubsync.ub_txg);
1607 spa_first_txg(spa_t *spa)
1609 return (spa->spa_first_txg);
1613 spa_syncing_txg(spa_t *spa)
1615 return (spa->spa_syncing_txg);
1619 * Return the last txg where data can be dirtied. The final txgs
1620 * will be used to just clear out any deferred frees that remain.
1623 spa_final_dirty_txg(spa_t *spa)
1625 return (spa->spa_final_txg - TXG_DEFER_SIZE);
1629 spa_state(spa_t *spa)
1631 return (spa->spa_state);
1635 spa_load_state(spa_t *spa)
1637 return (spa->spa_load_state);
1641 spa_freeze_txg(spa_t *spa)
1643 return (spa->spa_freeze_txg);
1647 * Return the inflated asize for a logical write in bytes. This is used by the
1648 * DMU to calculate the space a logical write will require on disk.
1649 * If lsize is smaller than the largest physical block size allocatable on this
1650 * pool we use its value instead, since the write will end up using the whole
1654 spa_get_worst_case_asize(spa_t *spa, uint64_t lsize)
1657 return (0); /* No inflation needed */
1658 return (MAX(lsize, 1 << spa->spa_max_ashift) * spa_asize_inflation);
1662 * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%),
1663 * or at least 128MB, unless that would cause it to be more than half the
1666 * See the comment above spa_slop_shift for details.
1669 spa_get_slop_space(spa_t *spa)
1671 uint64_t space = spa_get_dspace(spa);
1672 return (MAX(space >> spa_slop_shift, MIN(space >> 1, spa_min_slop)));
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 != ZIO_SUSPEND_NONE);
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_evicting_os_register(spa_t *spa, objset_t *os)
1731 mutex_enter(&spa->spa_evicting_os_lock);
1732 list_insert_head(&spa->spa_evicting_os_list, os);
1733 mutex_exit(&spa->spa_evicting_os_lock);
1737 spa_evicting_os_deregister(spa_t *spa, objset_t *os)
1739 mutex_enter(&spa->spa_evicting_os_lock);
1740 list_remove(&spa->spa_evicting_os_list, os);
1741 cv_broadcast(&spa->spa_evicting_os_cv);
1742 mutex_exit(&spa->spa_evicting_os_lock);
1746 spa_evicting_os_wait(spa_t *spa)
1748 mutex_enter(&spa->spa_evicting_os_lock);
1749 while (!list_is_empty(&spa->spa_evicting_os_list))
1750 cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock);
1751 mutex_exit(&spa->spa_evicting_os_lock);
1753 dmu_buf_user_evict_wait();
1757 spa_max_replication(spa_t *spa)
1760 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1761 * handle BPs with more than one DVA allocated. Set our max
1762 * replication level accordingly.
1764 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1766 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1770 spa_prev_software_version(spa_t *spa)
1772 return (spa->spa_prev_software_version);
1776 spa_deadman_synctime(spa_t *spa)
1778 return (spa->spa_deadman_synctime);
1782 spa_deadman_ziotime(spa_t *spa)
1784 return (spa->spa_deadman_ziotime);
1788 spa_get_deadman_failmode(spa_t *spa)
1790 return (spa->spa_deadman_failmode);
1794 spa_set_deadman_failmode(spa_t *spa, const char *failmode)
1796 if (strcmp(failmode, "wait") == 0)
1797 spa->spa_deadman_failmode = ZIO_FAILURE_MODE_WAIT;
1798 else if (strcmp(failmode, "continue") == 0)
1799 spa->spa_deadman_failmode = ZIO_FAILURE_MODE_CONTINUE;
1800 else if (strcmp(failmode, "panic") == 0)
1801 spa->spa_deadman_failmode = ZIO_FAILURE_MODE_PANIC;
1803 spa->spa_deadman_failmode = ZIO_FAILURE_MODE_WAIT;
1807 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1809 uint64_t asize = DVA_GET_ASIZE(dva);
1810 uint64_t dsize = asize;
1812 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1814 if (asize != 0 && spa->spa_deflate) {
1815 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1817 dsize = (asize >> SPA_MINBLOCKSHIFT) *
1818 vd->vdev_deflate_ratio;
1825 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1829 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1830 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1836 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1840 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1842 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1843 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1845 spa_config_exit(spa, SCL_VDEV, FTAG);
1851 * ==========================================================================
1852 * Initialization and Termination
1853 * ==========================================================================
1857 spa_name_compare(const void *a1, const void *a2)
1859 const spa_t *s1 = a1;
1860 const spa_t *s2 = a2;
1863 s = strcmp(s1->spa_name, s2->spa_name);
1865 return (AVL_ISIGN(s));
1877 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1878 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1879 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1880 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1882 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1883 offsetof(spa_t, spa_avl));
1885 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1886 offsetof(spa_aux_t, aux_avl));
1888 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1889 offsetof(spa_aux_t, aux_avl));
1891 spa_mode_global = mode;
1894 if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
1895 struct sigaction sa;
1897 sa.sa_flags = SA_SIGINFO;
1898 sigemptyset(&sa.sa_mask);
1899 sa.sa_sigaction = arc_buf_sigsegv;
1901 if (sigaction(SIGSEGV, &sa, NULL) == -1) {
1902 perror("could not enable watchpoints: "
1903 "sigaction(SIGSEGV, ...) = ");
1914 metaslab_alloc_trace_init();
1919 vdev_cache_stat_init();
1920 vdev_mirror_stat_init();
1921 vdev_raidz_math_init();
1925 zpool_feature_init();
1940 vdev_cache_stat_fini();
1941 vdev_mirror_stat_fini();
1942 vdev_raidz_math_fini();
1947 metaslab_alloc_trace_fini();
1955 avl_destroy(&spa_namespace_avl);
1956 avl_destroy(&spa_spare_avl);
1957 avl_destroy(&spa_l2cache_avl);
1959 cv_destroy(&spa_namespace_cv);
1960 mutex_destroy(&spa_namespace_lock);
1961 mutex_destroy(&spa_spare_lock);
1962 mutex_destroy(&spa_l2cache_lock);
1966 * Return whether this pool has slogs. No locking needed.
1967 * It's not a problem if the wrong answer is returned as it's only for
1968 * performance and not correctness
1971 spa_has_slogs(spa_t *spa)
1973 return (spa->spa_log_class->mc_rotor != NULL);
1977 spa_get_log_state(spa_t *spa)
1979 return (spa->spa_log_state);
1983 spa_set_log_state(spa_t *spa, spa_log_state_t state)
1985 spa->spa_log_state = state;
1989 spa_is_root(spa_t *spa)
1991 return (spa->spa_is_root);
1995 spa_writeable(spa_t *spa)
1997 return (!!(spa->spa_mode & FWRITE));
2001 * Returns true if there is a pending sync task in any of the current
2002 * syncing txg, the current quiescing txg, or the current open txg.
2005 spa_has_pending_synctask(spa_t *spa)
2007 return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks));
2011 spa_mode(spa_t *spa)
2013 return (spa->spa_mode);
2017 spa_bootfs(spa_t *spa)
2019 return (spa->spa_bootfs);
2023 spa_delegation(spa_t *spa)
2025 return (spa->spa_delegation);
2029 spa_meta_objset(spa_t *spa)
2031 return (spa->spa_meta_objset);
2035 spa_dedup_checksum(spa_t *spa)
2037 return (spa->spa_dedup_checksum);
2041 * Reset pool scan stat per scan pass (or reboot).
2044 spa_scan_stat_init(spa_t *spa)
2046 /* data not stored on disk */
2047 spa->spa_scan_pass_start = gethrestime_sec();
2048 if (dsl_scan_is_paused_scrub(spa->spa_dsl_pool->dp_scan))
2049 spa->spa_scan_pass_scrub_pause = spa->spa_scan_pass_start;
2051 spa->spa_scan_pass_scrub_pause = 0;
2052 spa->spa_scan_pass_scrub_spent_paused = 0;
2053 spa->spa_scan_pass_exam = 0;
2054 spa->spa_scan_pass_issued = 0;
2055 vdev_scan_stat_init(spa->spa_root_vdev);
2059 * Get scan stats for zpool status reports
2062 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
2064 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
2066 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
2067 return (SET_ERROR(ENOENT));
2068 bzero(ps, sizeof (pool_scan_stat_t));
2070 /* data stored on disk */
2071 ps->pss_func = scn->scn_phys.scn_func;
2072 ps->pss_state = scn->scn_phys.scn_state;
2073 ps->pss_start_time = scn->scn_phys.scn_start_time;
2074 ps->pss_end_time = scn->scn_phys.scn_end_time;
2075 ps->pss_to_examine = scn->scn_phys.scn_to_examine;
2076 ps->pss_examined = scn->scn_phys.scn_examined;
2077 ps->pss_to_process = scn->scn_phys.scn_to_process;
2078 ps->pss_processed = scn->scn_phys.scn_processed;
2079 ps->pss_errors = scn->scn_phys.scn_errors;
2081 /* data not stored on disk */
2082 ps->pss_pass_exam = spa->spa_scan_pass_exam;
2083 ps->pss_pass_start = spa->spa_scan_pass_start;
2084 ps->pss_pass_scrub_pause = spa->spa_scan_pass_scrub_pause;
2085 ps->pss_pass_scrub_spent_paused = spa->spa_scan_pass_scrub_spent_paused;
2086 ps->pss_pass_issued = spa->spa_scan_pass_issued;
2088 scn->scn_issued_before_pass + spa->spa_scan_pass_issued;
2094 spa_debug_enabled(spa_t *spa)
2096 return (spa->spa_debug);
2100 spa_maxblocksize(spa_t *spa)
2102 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS))
2103 return (SPA_MAXBLOCKSIZE);
2105 return (SPA_OLD_MAXBLOCKSIZE);
2109 spa_maxdnodesize(spa_t *spa)
2111 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE))
2112 return (DNODE_MAX_SIZE);
2114 return (DNODE_MIN_SIZE);
2118 spa_multihost(spa_t *spa)
2120 return (spa->spa_multihost ? B_TRUE : B_FALSE);
2124 spa_get_hostid(void)
2126 unsigned long myhostid;
2129 myhostid = zone_get_hostid(NULL);
2132 * We're emulating the system's hostid in userland, so
2133 * we can't use zone_get_hostid().
2135 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2136 #endif /* _KERNEL */
2141 #if defined(_KERNEL) && defined(HAVE_SPL)
2143 #include <linux/mod_compat.h>
2146 param_set_deadman_failmode(const char *val, zfs_kernel_param_t *kp)
2152 return (SET_ERROR(-EINVAL));
2154 if ((p = strchr(val, '\n')) != NULL)
2157 if (strcmp(val, "wait") != 0 && strcmp(val, "continue") != 0 &&
2158 strcmp(val, "panic"))
2159 return (SET_ERROR(-EINVAL));
2161 mutex_enter(&spa_namespace_lock);
2162 while ((spa = spa_next(spa)) != NULL)
2163 spa_set_deadman_failmode(spa, val);
2164 mutex_exit(&spa_namespace_lock);
2166 return (param_set_charp(val, kp));
2169 /* Namespace manipulation */
2170 EXPORT_SYMBOL(spa_lookup);
2171 EXPORT_SYMBOL(spa_add);
2172 EXPORT_SYMBOL(spa_remove);
2173 EXPORT_SYMBOL(spa_next);
2175 /* Refcount functions */
2176 EXPORT_SYMBOL(spa_open_ref);
2177 EXPORT_SYMBOL(spa_close);
2178 EXPORT_SYMBOL(spa_refcount_zero);
2180 /* Pool configuration lock */
2181 EXPORT_SYMBOL(spa_config_tryenter);
2182 EXPORT_SYMBOL(spa_config_enter);
2183 EXPORT_SYMBOL(spa_config_exit);
2184 EXPORT_SYMBOL(spa_config_held);
2186 /* Pool vdev add/remove lock */
2187 EXPORT_SYMBOL(spa_vdev_enter);
2188 EXPORT_SYMBOL(spa_vdev_exit);
2190 /* Pool vdev state change lock */
2191 EXPORT_SYMBOL(spa_vdev_state_enter);
2192 EXPORT_SYMBOL(spa_vdev_state_exit);
2194 /* Accessor functions */
2195 EXPORT_SYMBOL(spa_shutting_down);
2196 EXPORT_SYMBOL(spa_get_dsl);
2197 EXPORT_SYMBOL(spa_get_rootblkptr);
2198 EXPORT_SYMBOL(spa_set_rootblkptr);
2199 EXPORT_SYMBOL(spa_altroot);
2200 EXPORT_SYMBOL(spa_sync_pass);
2201 EXPORT_SYMBOL(spa_name);
2202 EXPORT_SYMBOL(spa_guid);
2203 EXPORT_SYMBOL(spa_last_synced_txg);
2204 EXPORT_SYMBOL(spa_first_txg);
2205 EXPORT_SYMBOL(spa_syncing_txg);
2206 EXPORT_SYMBOL(spa_version);
2207 EXPORT_SYMBOL(spa_state);
2208 EXPORT_SYMBOL(spa_load_state);
2209 EXPORT_SYMBOL(spa_freeze_txg);
2210 EXPORT_SYMBOL(spa_get_dspace);
2211 EXPORT_SYMBOL(spa_update_dspace);
2212 EXPORT_SYMBOL(spa_deflate);
2213 EXPORT_SYMBOL(spa_normal_class);
2214 EXPORT_SYMBOL(spa_log_class);
2215 EXPORT_SYMBOL(spa_max_replication);
2216 EXPORT_SYMBOL(spa_prev_software_version);
2217 EXPORT_SYMBOL(spa_get_failmode);
2218 EXPORT_SYMBOL(spa_suspended);
2219 EXPORT_SYMBOL(spa_bootfs);
2220 EXPORT_SYMBOL(spa_delegation);
2221 EXPORT_SYMBOL(spa_meta_objset);
2222 EXPORT_SYMBOL(spa_maxblocksize);
2223 EXPORT_SYMBOL(spa_maxdnodesize);
2225 /* Miscellaneous support routines */
2226 EXPORT_SYMBOL(spa_rename);
2227 EXPORT_SYMBOL(spa_guid_exists);
2228 EXPORT_SYMBOL(spa_strdup);
2229 EXPORT_SYMBOL(spa_strfree);
2230 EXPORT_SYMBOL(spa_get_random);
2231 EXPORT_SYMBOL(spa_generate_guid);
2232 EXPORT_SYMBOL(snprintf_blkptr);
2233 EXPORT_SYMBOL(spa_freeze);
2234 EXPORT_SYMBOL(spa_upgrade);
2235 EXPORT_SYMBOL(spa_evict_all);
2236 EXPORT_SYMBOL(spa_lookup_by_guid);
2237 EXPORT_SYMBOL(spa_has_spare);
2238 EXPORT_SYMBOL(dva_get_dsize_sync);
2239 EXPORT_SYMBOL(bp_get_dsize_sync);
2240 EXPORT_SYMBOL(bp_get_dsize);
2241 EXPORT_SYMBOL(spa_has_slogs);
2242 EXPORT_SYMBOL(spa_is_root);
2243 EXPORT_SYMBOL(spa_writeable);
2244 EXPORT_SYMBOL(spa_mode);
2245 EXPORT_SYMBOL(spa_namespace_lock);
2248 module_param(zfs_flags, uint, 0644);
2249 MODULE_PARM_DESC(zfs_flags, "Set additional debugging flags");
2251 module_param(zfs_recover, int, 0644);
2252 MODULE_PARM_DESC(zfs_recover, "Set to attempt to recover from fatal errors");
2254 module_param(zfs_free_leak_on_eio, int, 0644);
2255 MODULE_PARM_DESC(zfs_free_leak_on_eio,
2256 "Set to ignore IO errors during free and permanently leak the space");
2258 module_param(zfs_deadman_synctime_ms, ulong, 0644);
2259 MODULE_PARM_DESC(zfs_deadman_synctime_ms,
2260 "Pool sync expiration time in milliseconds");
2262 module_param(zfs_deadman_ziotime_ms, ulong, 0644);
2263 MODULE_PARM_DESC(zfs_deadman_ziotime_ms,
2264 "IO expiration time in milliseconds");
2266 module_param(zfs_deadman_checktime_ms, ulong, 0644);
2267 MODULE_PARM_DESC(zfs_deadman_checktime_ms,
2268 "Dead I/O check interval in milliseconds");
2270 module_param(zfs_deadman_enabled, int, 0644);
2271 MODULE_PARM_DESC(zfs_deadman_enabled, "Enable deadman timer");
2273 module_param_call(zfs_deadman_failmode, param_set_deadman_failmode,
2274 param_get_charp, &zfs_deadman_failmode, 0644);
2275 MODULE_PARM_DESC(zfs_deadman_failmode, "Failmode for deadman timer");
2277 module_param(spa_asize_inflation, int, 0644);
2278 MODULE_PARM_DESC(spa_asize_inflation,
2279 "SPA size estimate multiplication factor");
2281 module_param(spa_slop_shift, int, 0644);
2282 MODULE_PARM_DESC(spa_slop_shift, "Reserved free space in pool");