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 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.
29 #include <sys/zfs_context.h>
30 #include <sys/spa_impl.h>
32 #include <sys/zio_checksum.h>
33 #include <sys/zio_compress.h>
35 #include <sys/dmu_tx.h>
38 #include <sys/vdev_impl.h>
39 #include <sys/vdev_file.h>
40 #include <sys/vdev_raidz.h>
41 #include <sys/metaslab.h>
42 #include <sys/uberblock_impl.h>
45 #include <sys/unique.h>
46 #include <sys/dsl_pool.h>
47 #include <sys/dsl_dir.h>
48 #include <sys/dsl_prop.h>
49 #include <sys/fm/util.h>
50 #include <sys/dsl_scan.h>
51 #include <sys/fs/zfs.h>
52 #include <sys/metaslab_impl.h>
55 #include <sys/kstat.h>
57 #include <sys/zfeature.h>
62 * There are four basic locks for managing spa_t structures:
64 * spa_namespace_lock (global mutex)
66 * This lock must be acquired to do any of the following:
68 * - Lookup a spa_t by name
69 * - Add or remove a spa_t from the namespace
70 * - Increase spa_refcount from non-zero
71 * - Check if spa_refcount is zero
73 * - add/remove/attach/detach devices
74 * - Held for the duration of create/destroy/import/export
76 * It does not need to handle recursion. A create or destroy may
77 * reference objects (files or zvols) in other pools, but by
78 * definition they must have an existing reference, and will never need
79 * to lookup a spa_t by name.
81 * spa_refcount (per-spa refcount_t protected by mutex)
83 * This reference count keep track of any active users of the spa_t. The
84 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
85 * the refcount is never really 'zero' - opening a pool implicitly keeps
86 * some references in the DMU. Internally we check against spa_minref, but
87 * present the image of a zero/non-zero value to consumers.
89 * spa_config_lock[] (per-spa array of rwlocks)
91 * This protects the spa_t from config changes, and must be held in
92 * the following circumstances:
94 * - RW_READER to perform I/O to the spa
95 * - RW_WRITER to change the vdev config
97 * The locking order is fairly straightforward:
99 * spa_namespace_lock -> spa_refcount
101 * The namespace lock must be acquired to increase the refcount from 0
102 * or to check if it is zero.
104 * spa_refcount -> spa_config_lock[]
106 * There must be at least one valid reference on the spa_t to acquire
109 * spa_namespace_lock -> spa_config_lock[]
111 * The namespace lock must always be taken before the config lock.
114 * The spa_namespace_lock can be acquired directly and is globally visible.
116 * The namespace is manipulated using the following functions, all of which
117 * require the spa_namespace_lock to be held.
119 * spa_lookup() Lookup a spa_t by name.
121 * spa_add() Create a new spa_t in the namespace.
123 * spa_remove() Remove a spa_t from the namespace. This also
124 * frees up any memory associated with the spa_t.
126 * spa_next() Returns the next spa_t in the system, or the
127 * first if NULL is passed.
129 * spa_evict_all() Shutdown and remove all spa_t structures in
132 * spa_guid_exists() Determine whether a pool/device guid exists.
134 * The spa_refcount is manipulated using the following functions:
136 * spa_open_ref() Adds a reference to the given spa_t. Must be
137 * called with spa_namespace_lock held if the
138 * refcount is currently zero.
140 * spa_close() Remove a reference from the spa_t. This will
141 * not free the spa_t or remove it from the
142 * namespace. No locking is required.
144 * spa_refcount_zero() Returns true if the refcount is currently
145 * zero. Must be called with spa_namespace_lock
148 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
149 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
150 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
152 * To read the configuration, it suffices to hold one of these locks as reader.
153 * To modify the configuration, you must hold all locks as writer. To modify
154 * vdev state without altering the vdev tree's topology (e.g. online/offline),
155 * you must hold SCL_STATE and SCL_ZIO as writer.
157 * We use these distinct config locks to avoid recursive lock entry.
158 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
159 * block allocations (SCL_ALLOC), which may require reading space maps
160 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
162 * The spa config locks cannot be normal rwlocks because we need the
163 * ability to hand off ownership. For example, SCL_ZIO is acquired
164 * by the issuing thread and later released by an interrupt thread.
165 * They do, however, obey the usual write-wanted semantics to prevent
166 * writer (i.e. system administrator) starvation.
168 * The lock acquisition rules are as follows:
171 * Protects changes to the vdev tree topology, such as vdev
172 * add/remove/attach/detach. Protects the dirty config list
173 * (spa_config_dirty_list) and the set of spares and l2arc devices.
176 * Protects changes to pool state and vdev state, such as vdev
177 * online/offline/fault/degrade/clear. Protects the dirty state list
178 * (spa_state_dirty_list) and global pool state (spa_state).
181 * Protects changes to metaslab groups and classes.
182 * Held as reader by metaslab_alloc() and metaslab_claim().
185 * Held by bp-level zios (those which have no io_vd upon entry)
186 * to prevent changes to the vdev tree. The bp-level zio implicitly
187 * protects all of its vdev child zios, which do not hold SCL_ZIO.
190 * Protects changes to metaslab groups and classes.
191 * Held as reader by metaslab_free(). SCL_FREE is distinct from
192 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
193 * blocks in zio_done() while another i/o that holds either
194 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
197 * Held as reader to prevent changes to the vdev tree during trivial
198 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
199 * other locks, and lower than all of them, to ensure that it's safe
200 * to acquire regardless of caller context.
202 * In addition, the following rules apply:
204 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
205 * The lock ordering is SCL_CONFIG > spa_props_lock.
207 * (b) I/O operations on leaf vdevs. For any zio operation that takes
208 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
209 * or zio_write_phys() -- the caller must ensure that the config cannot
210 * cannot change in the interim, and that the vdev cannot be reopened.
211 * SCL_STATE as reader suffices for both.
213 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
215 * spa_vdev_enter() Acquire the namespace lock and the config lock
218 * spa_vdev_exit() Release the config lock, wait for all I/O
219 * to complete, sync the updated configs to the
220 * cache, and release the namespace lock.
222 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
223 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
224 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
226 * spa_rename() is also implemented within this file since it requires
227 * manipulation of the namespace.
230 static avl_tree_t spa_namespace_avl;
231 kmutex_t spa_namespace_lock;
232 static kcondvar_t spa_namespace_cv;
233 int spa_max_replication_override = SPA_DVAS_PER_BP;
235 static kmutex_t spa_spare_lock;
236 static avl_tree_t spa_spare_avl;
237 static kmutex_t spa_l2cache_lock;
238 static avl_tree_t spa_l2cache_avl;
240 kmem_cache_t *spa_buffer_pool;
244 /* Everything except dprintf and spa is on by default in debug builds */
245 int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA);
251 * zfs_recover can be set to nonzero to attempt to recover from
252 * otherwise-fatal errors, typically caused by on-disk corruption. When
253 * set, calls to zfs_panic_recover() will turn into warning messages.
254 * This should only be used as a last resort, as it typically results
255 * in leaked space, or worse.
257 int zfs_recover = B_FALSE;
260 * If destroy encounters an EIO while reading metadata (e.g. indirect
261 * blocks), space referenced by the missing metadata can not be freed.
262 * Normally this causes the background destroy to become "stalled", as
263 * it is unable to make forward progress. While in this stalled state,
264 * all remaining space to free from the error-encountering filesystem is
265 * "temporarily leaked". Set this flag to cause it to ignore the EIO,
266 * permanently leak the space from indirect blocks that can not be read,
267 * and continue to free everything else that it can.
269 * The default, "stalling" behavior is useful if the storage partially
270 * fails (i.e. some but not all i/os fail), and then later recovers. In
271 * this case, we will be able to continue pool operations while it is
272 * partially failed, and when it recovers, we can continue to free the
273 * space, with no leaks. However, note that this case is actually
276 * Typically pools either (a) fail completely (but perhaps temporarily,
277 * e.g. a top-level vdev going offline), or (b) have localized,
278 * permanent errors (e.g. disk returns the wrong data due to bit flip or
279 * firmware bug). In case (a), this setting does not matter because the
280 * pool will be suspended and the sync thread will not be able to make
281 * forward progress regardless. In case (b), because the error is
282 * permanent, the best we can do is leak the minimum amount of space,
283 * which is what setting this flag will do. Therefore, it is reasonable
284 * for this flag to normally be set, but we chose the more conservative
285 * approach of not setting it, so that there is no possibility of
286 * leaking space in the "partial temporary" failure case.
288 int zfs_free_leak_on_eio = B_FALSE;
291 * Expiration time in milliseconds. This value has two meanings. First it is
292 * used to determine when the spa_deadman() logic should fire. By default the
293 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
294 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
295 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
298 unsigned long zfs_deadman_synctime_ms = 1000000ULL;
301 * By default the deadman is enabled.
303 int zfs_deadman_enabled = 1;
306 * The worst case is single-sector max-parity RAID-Z blocks, in which
307 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
308 * times the size; so just assume that. Add to this the fact that
309 * we can have up to 3 DVAs per bp, and one more factor of 2 because
310 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together,
312 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
314 int spa_asize_inflation = 24;
317 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
318 * the pool to be consumed. This ensures that we don't run the pool
319 * completely out of space, due to unaccounted changes (e.g. to the MOS).
320 * It also limits the worst-case time to allocate space. If we have
321 * less than this amount of free space, most ZPL operations (e.g. write,
322 * create) will return ENOSPC.
324 * Certain operations (e.g. file removal, most administrative actions) can
325 * use half the slop space. They will only return ENOSPC if less than half
326 * the slop space is free. Typically, once the pool has less than the slop
327 * space free, the user will use these operations to free up space in the pool.
328 * These are the operations that call dsl_pool_adjustedsize() with the netfree
329 * argument set to TRUE.
331 * A very restricted set of operations are always permitted, regardless of
332 * the amount of free space. These are the operations that call
333 * dsl_sync_task(ZFS_SPACE_CHECK_NONE), e.g. "zfs destroy". If these
334 * operations result in a net increase in the amount of space used,
335 * it is possible to run the pool completely out of space, causing it to
336 * be permanently read-only.
338 * See also the comments in zfs_space_check_t.
340 int spa_slop_shift = 5;
343 * ==========================================================================
345 * ==========================================================================
348 spa_config_lock_init(spa_t *spa)
352 for (i = 0; i < SCL_LOCKS; i++) {
353 spa_config_lock_t *scl = &spa->spa_config_lock[i];
354 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
355 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
356 refcount_create_untracked(&scl->scl_count);
357 scl->scl_writer = NULL;
358 scl->scl_write_wanted = 0;
363 spa_config_lock_destroy(spa_t *spa)
367 for (i = 0; i < SCL_LOCKS; i++) {
368 spa_config_lock_t *scl = &spa->spa_config_lock[i];
369 mutex_destroy(&scl->scl_lock);
370 cv_destroy(&scl->scl_cv);
371 refcount_destroy(&scl->scl_count);
372 ASSERT(scl->scl_writer == NULL);
373 ASSERT(scl->scl_write_wanted == 0);
378 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
382 for (i = 0; i < SCL_LOCKS; i++) {
383 spa_config_lock_t *scl = &spa->spa_config_lock[i];
384 if (!(locks & (1 << i)))
386 mutex_enter(&scl->scl_lock);
387 if (rw == RW_READER) {
388 if (scl->scl_writer || scl->scl_write_wanted) {
389 mutex_exit(&scl->scl_lock);
390 spa_config_exit(spa, locks & ((1 << i) - 1),
395 ASSERT(scl->scl_writer != curthread);
396 if (!refcount_is_zero(&scl->scl_count)) {
397 mutex_exit(&scl->scl_lock);
398 spa_config_exit(spa, locks & ((1 << i) - 1),
402 scl->scl_writer = curthread;
404 (void) refcount_add(&scl->scl_count, tag);
405 mutex_exit(&scl->scl_lock);
411 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
416 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
418 for (i = 0; i < SCL_LOCKS; i++) {
419 spa_config_lock_t *scl = &spa->spa_config_lock[i];
420 if (scl->scl_writer == curthread)
421 wlocks_held |= (1 << i);
422 if (!(locks & (1 << i)))
424 mutex_enter(&scl->scl_lock);
425 if (rw == RW_READER) {
426 while (scl->scl_writer || scl->scl_write_wanted) {
427 cv_wait(&scl->scl_cv, &scl->scl_lock);
430 ASSERT(scl->scl_writer != curthread);
431 while (!refcount_is_zero(&scl->scl_count)) {
432 scl->scl_write_wanted++;
433 cv_wait(&scl->scl_cv, &scl->scl_lock);
434 scl->scl_write_wanted--;
436 scl->scl_writer = curthread;
438 (void) refcount_add(&scl->scl_count, tag);
439 mutex_exit(&scl->scl_lock);
441 ASSERT(wlocks_held <= locks);
445 spa_config_exit(spa_t *spa, int locks, void *tag)
449 for (i = SCL_LOCKS - 1; i >= 0; i--) {
450 spa_config_lock_t *scl = &spa->spa_config_lock[i];
451 if (!(locks & (1 << i)))
453 mutex_enter(&scl->scl_lock);
454 ASSERT(!refcount_is_zero(&scl->scl_count));
455 if (refcount_remove(&scl->scl_count, tag) == 0) {
456 ASSERT(scl->scl_writer == NULL ||
457 scl->scl_writer == curthread);
458 scl->scl_writer = NULL; /* OK in either case */
459 cv_broadcast(&scl->scl_cv);
461 mutex_exit(&scl->scl_lock);
466 spa_config_held(spa_t *spa, int locks, krw_t rw)
468 int i, locks_held = 0;
470 for (i = 0; i < SCL_LOCKS; i++) {
471 spa_config_lock_t *scl = &spa->spa_config_lock[i];
472 if (!(locks & (1 << i)))
474 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
475 (rw == RW_WRITER && scl->scl_writer == curthread))
476 locks_held |= 1 << i;
483 * ==========================================================================
484 * SPA namespace functions
485 * ==========================================================================
489 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
490 * Returns NULL if no matching spa_t is found.
493 spa_lookup(const char *name)
495 static spa_t search; /* spa_t is large; don't allocate on stack */
500 ASSERT(MUTEX_HELD(&spa_namespace_lock));
502 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
505 * If it's a full dataset name, figure out the pool name and
508 cp = strpbrk(search.spa_name, "/@#");
512 spa = avl_find(&spa_namespace_avl, &search, &where);
518 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
519 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
520 * looking for potentially hung I/Os.
523 spa_deadman(void *arg)
527 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
528 (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
529 ++spa->spa_deadman_calls);
530 if (zfs_deadman_enabled)
531 vdev_deadman(spa->spa_root_vdev);
533 spa->spa_deadman_tqid = taskq_dispatch_delay(system_taskq,
534 spa_deadman, spa, TQ_SLEEP, ddi_get_lbolt() +
535 NSEC_TO_TICK(spa->spa_deadman_synctime));
539 * Create an uninitialized spa_t with the given name. Requires
540 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
541 * exist by calling spa_lookup() first.
544 spa_add(const char *name, nvlist_t *config, const char *altroot)
547 spa_config_dirent_t *dp;
551 ASSERT(MUTEX_HELD(&spa_namespace_lock));
553 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
555 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
556 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
557 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
558 mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL);
559 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
560 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
561 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
562 mutex_init(&spa->spa_cksum_tmpls_lock, NULL, MUTEX_DEFAULT, NULL);
563 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
564 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
565 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
566 mutex_init(&spa->spa_feat_stats_lock, NULL, MUTEX_DEFAULT, NULL);
567 mutex_init(&spa->spa_alloc_lock, NULL, MUTEX_DEFAULT, NULL);
569 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
570 cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL);
571 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
572 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
573 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
575 for (t = 0; t < TXG_SIZE; t++)
576 bplist_create(&spa->spa_free_bplist[t]);
578 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
579 spa->spa_state = POOL_STATE_UNINITIALIZED;
580 spa->spa_freeze_txg = UINT64_MAX;
581 spa->spa_final_txg = UINT64_MAX;
582 spa->spa_load_max_txg = UINT64_MAX;
584 spa->spa_proc_state = SPA_PROC_NONE;
586 spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
588 refcount_create(&spa->spa_refcount);
589 spa_config_lock_init(spa);
592 avl_add(&spa_namespace_avl, spa);
595 * Set the alternate root, if there is one.
598 spa->spa_root = spa_strdup(altroot);
600 avl_create(&spa->spa_alloc_tree, zio_timestamp_compare,
601 sizeof (zio_t), offsetof(zio_t, io_alloc_node));
604 * Every pool starts with the default cachefile
606 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
607 offsetof(spa_config_dirent_t, scd_link));
609 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
610 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
611 list_insert_head(&spa->spa_config_list, dp);
613 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
616 if (config != NULL) {
619 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
621 VERIFY(nvlist_dup(features, &spa->spa_label_features,
625 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
628 if (spa->spa_label_features == NULL) {
629 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
633 spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0);
635 spa->spa_min_ashift = INT_MAX;
636 spa->spa_max_ashift = 0;
639 * As a pool is being created, treat all features as disabled by
640 * setting SPA_FEATURE_DISABLED for all entries in the feature
643 for (i = 0; i < SPA_FEATURES; i++) {
644 spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED;
651 * Removes a spa_t from the namespace, freeing up any memory used. Requires
652 * spa_namespace_lock. This is called only after the spa_t has been closed and
656 spa_remove(spa_t *spa)
658 spa_config_dirent_t *dp;
661 ASSERT(MUTEX_HELD(&spa_namespace_lock));
662 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
663 ASSERT3U(refcount_count(&spa->spa_refcount), ==, 0);
665 nvlist_free(spa->spa_config_splitting);
667 avl_remove(&spa_namespace_avl, spa);
668 cv_broadcast(&spa_namespace_cv);
671 spa_strfree(spa->spa_root);
673 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
674 list_remove(&spa->spa_config_list, dp);
675 if (dp->scd_path != NULL)
676 spa_strfree(dp->scd_path);
677 kmem_free(dp, sizeof (spa_config_dirent_t));
680 avl_destroy(&spa->spa_alloc_tree);
681 list_destroy(&spa->spa_config_list);
683 nvlist_free(spa->spa_label_features);
684 nvlist_free(spa->spa_load_info);
685 nvlist_free(spa->spa_feat_stats);
686 spa_config_set(spa, NULL);
688 refcount_destroy(&spa->spa_refcount);
690 spa_stats_destroy(spa);
691 spa_config_lock_destroy(spa);
693 for (t = 0; t < TXG_SIZE; t++)
694 bplist_destroy(&spa->spa_free_bplist[t]);
696 zio_checksum_templates_free(spa);
698 cv_destroy(&spa->spa_async_cv);
699 cv_destroy(&spa->spa_evicting_os_cv);
700 cv_destroy(&spa->spa_proc_cv);
701 cv_destroy(&spa->spa_scrub_io_cv);
702 cv_destroy(&spa->spa_suspend_cv);
704 mutex_destroy(&spa->spa_alloc_lock);
705 mutex_destroy(&spa->spa_async_lock);
706 mutex_destroy(&spa->spa_errlist_lock);
707 mutex_destroy(&spa->spa_errlog_lock);
708 mutex_destroy(&spa->spa_evicting_os_lock);
709 mutex_destroy(&spa->spa_history_lock);
710 mutex_destroy(&spa->spa_proc_lock);
711 mutex_destroy(&spa->spa_props_lock);
712 mutex_destroy(&spa->spa_cksum_tmpls_lock);
713 mutex_destroy(&spa->spa_scrub_lock);
714 mutex_destroy(&spa->spa_suspend_lock);
715 mutex_destroy(&spa->spa_vdev_top_lock);
716 mutex_destroy(&spa->spa_feat_stats_lock);
718 kmem_free(spa, sizeof (spa_t));
722 * Given a pool, return the next pool in the namespace, or NULL if there is
723 * none. If 'prev' is NULL, return the first pool.
726 spa_next(spa_t *prev)
728 ASSERT(MUTEX_HELD(&spa_namespace_lock));
731 return (AVL_NEXT(&spa_namespace_avl, prev));
733 return (avl_first(&spa_namespace_avl));
737 * ==========================================================================
738 * SPA refcount functions
739 * ==========================================================================
743 * Add a reference to the given spa_t. Must have at least one reference, or
744 * have the namespace lock held.
747 spa_open_ref(spa_t *spa, void *tag)
749 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
750 MUTEX_HELD(&spa_namespace_lock));
751 (void) refcount_add(&spa->spa_refcount, tag);
755 * Remove a reference to the given spa_t. Must have at least one reference, or
756 * have the namespace lock held.
759 spa_close(spa_t *spa, void *tag)
761 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
762 MUTEX_HELD(&spa_namespace_lock));
763 (void) refcount_remove(&spa->spa_refcount, tag);
767 * Remove a reference to the given spa_t held by a dsl dir that is
768 * being asynchronously released. Async releases occur from a taskq
769 * performing eviction of dsl datasets and dirs. The namespace lock
770 * isn't held and the hold by the object being evicted may contribute to
771 * spa_minref (e.g. dataset or directory released during pool export),
772 * so the asserts in spa_close() do not apply.
775 spa_async_close(spa_t *spa, void *tag)
777 (void) refcount_remove(&spa->spa_refcount, tag);
781 * Check to see if the spa refcount is zero. Must be called with
782 * spa_namespace_lock held. We really compare against spa_minref, which is the
783 * number of references acquired when opening a pool
786 spa_refcount_zero(spa_t *spa)
788 ASSERT(MUTEX_HELD(&spa_namespace_lock));
790 return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
794 * ==========================================================================
795 * SPA spare and l2cache tracking
796 * ==========================================================================
800 * Hot spares and cache devices are tracked using the same code below,
801 * for 'auxiliary' devices.
804 typedef struct spa_aux {
812 spa_aux_compare(const void *a, const void *b)
814 const spa_aux_t *sa = (const spa_aux_t *)a;
815 const spa_aux_t *sb = (const spa_aux_t *)b;
817 return (AVL_CMP(sa->aux_guid, sb->aux_guid));
821 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
827 search.aux_guid = vd->vdev_guid;
828 if ((aux = avl_find(avl, &search, &where)) != NULL) {
831 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
832 aux->aux_guid = vd->vdev_guid;
834 avl_insert(avl, aux, where);
839 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
845 search.aux_guid = vd->vdev_guid;
846 aux = avl_find(avl, &search, &where);
850 if (--aux->aux_count == 0) {
851 avl_remove(avl, aux);
852 kmem_free(aux, sizeof (spa_aux_t));
853 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
854 aux->aux_pool = 0ULL;
859 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
861 spa_aux_t search, *found;
863 search.aux_guid = guid;
864 found = avl_find(avl, &search, NULL);
868 *pool = found->aux_pool;
875 *refcnt = found->aux_count;
880 return (found != NULL);
884 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
886 spa_aux_t search, *found;
889 search.aux_guid = vd->vdev_guid;
890 found = avl_find(avl, &search, &where);
891 ASSERT(found != NULL);
892 ASSERT(found->aux_pool == 0ULL);
894 found->aux_pool = spa_guid(vd->vdev_spa);
898 * Spares are tracked globally due to the following constraints:
900 * - A spare may be part of multiple pools.
901 * - A spare may be added to a pool even if it's actively in use within
903 * - A spare in use in any pool can only be the source of a replacement if
904 * the target is a spare in the same pool.
906 * We keep track of all spares on the system through the use of a reference
907 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
908 * spare, then we bump the reference count in the AVL tree. In addition, we set
909 * the 'vdev_isspare' member to indicate that the device is a spare (active or
910 * inactive). When a spare is made active (used to replace a device in the
911 * pool), we also keep track of which pool its been made a part of.
913 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
914 * called under the spa_namespace lock as part of vdev reconfiguration. The
915 * separate spare lock exists for the status query path, which does not need to
916 * be completely consistent with respect to other vdev configuration changes.
920 spa_spare_compare(const void *a, const void *b)
922 return (spa_aux_compare(a, b));
926 spa_spare_add(vdev_t *vd)
928 mutex_enter(&spa_spare_lock);
929 ASSERT(!vd->vdev_isspare);
930 spa_aux_add(vd, &spa_spare_avl);
931 vd->vdev_isspare = B_TRUE;
932 mutex_exit(&spa_spare_lock);
936 spa_spare_remove(vdev_t *vd)
938 mutex_enter(&spa_spare_lock);
939 ASSERT(vd->vdev_isspare);
940 spa_aux_remove(vd, &spa_spare_avl);
941 vd->vdev_isspare = B_FALSE;
942 mutex_exit(&spa_spare_lock);
946 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
950 mutex_enter(&spa_spare_lock);
951 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
952 mutex_exit(&spa_spare_lock);
958 spa_spare_activate(vdev_t *vd)
960 mutex_enter(&spa_spare_lock);
961 ASSERT(vd->vdev_isspare);
962 spa_aux_activate(vd, &spa_spare_avl);
963 mutex_exit(&spa_spare_lock);
967 * Level 2 ARC devices are tracked globally for the same reasons as spares.
968 * Cache devices currently only support one pool per cache device, and so
969 * for these devices the aux reference count is currently unused beyond 1.
973 spa_l2cache_compare(const void *a, const void *b)
975 return (spa_aux_compare(a, b));
979 spa_l2cache_add(vdev_t *vd)
981 mutex_enter(&spa_l2cache_lock);
982 ASSERT(!vd->vdev_isl2cache);
983 spa_aux_add(vd, &spa_l2cache_avl);
984 vd->vdev_isl2cache = B_TRUE;
985 mutex_exit(&spa_l2cache_lock);
989 spa_l2cache_remove(vdev_t *vd)
991 mutex_enter(&spa_l2cache_lock);
992 ASSERT(vd->vdev_isl2cache);
993 spa_aux_remove(vd, &spa_l2cache_avl);
994 vd->vdev_isl2cache = B_FALSE;
995 mutex_exit(&spa_l2cache_lock);
999 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
1003 mutex_enter(&spa_l2cache_lock);
1004 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
1005 mutex_exit(&spa_l2cache_lock);
1011 spa_l2cache_activate(vdev_t *vd)
1013 mutex_enter(&spa_l2cache_lock);
1014 ASSERT(vd->vdev_isl2cache);
1015 spa_aux_activate(vd, &spa_l2cache_avl);
1016 mutex_exit(&spa_l2cache_lock);
1020 * ==========================================================================
1022 * ==========================================================================
1026 * Lock the given spa_t for the purpose of adding or removing a vdev.
1027 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1028 * It returns the next transaction group for the spa_t.
1031 spa_vdev_enter(spa_t *spa)
1033 mutex_enter(&spa->spa_vdev_top_lock);
1034 mutex_enter(&spa_namespace_lock);
1035 return (spa_vdev_config_enter(spa));
1039 * Internal implementation for spa_vdev_enter(). Used when a vdev
1040 * operation requires multiple syncs (i.e. removing a device) while
1041 * keeping the spa_namespace_lock held.
1044 spa_vdev_config_enter(spa_t *spa)
1046 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1048 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1050 return (spa_last_synced_txg(spa) + 1);
1054 * Used in combination with spa_vdev_config_enter() to allow the syncing
1055 * of multiple transactions without releasing the spa_namespace_lock.
1058 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
1060 int config_changed = B_FALSE;
1062 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1063 ASSERT(txg > spa_last_synced_txg(spa));
1065 spa->spa_pending_vdev = NULL;
1068 * Reassess the DTLs.
1070 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
1072 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
1073 config_changed = B_TRUE;
1074 spa->spa_config_generation++;
1078 * Verify the metaslab classes.
1080 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
1081 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
1083 spa_config_exit(spa, SCL_ALL, spa);
1086 * Panic the system if the specified tag requires it. This
1087 * is useful for ensuring that configurations are updated
1090 if (zio_injection_enabled)
1091 zio_handle_panic_injection(spa, tag, 0);
1094 * Note: this txg_wait_synced() is important because it ensures
1095 * that there won't be more than one config change per txg.
1096 * This allows us to use the txg as the generation number.
1099 txg_wait_synced(spa->spa_dsl_pool, txg);
1102 ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
1103 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1105 spa_config_exit(spa, SCL_ALL, spa);
1109 * If the config changed, update the config cache.
1112 spa_config_sync(spa, B_FALSE, B_TRUE);
1116 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1117 * locking of spa_vdev_enter(), we also want make sure the transactions have
1118 * synced to disk, and then update the global configuration cache with the new
1122 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1124 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1125 mutex_exit(&spa_namespace_lock);
1126 mutex_exit(&spa->spa_vdev_top_lock);
1132 * Lock the given spa_t for the purpose of changing vdev state.
1135 spa_vdev_state_enter(spa_t *spa, int oplocks)
1137 int locks = SCL_STATE_ALL | oplocks;
1140 * Root pools may need to read of the underlying devfs filesystem
1141 * when opening up a vdev. Unfortunately if we're holding the
1142 * SCL_ZIO lock it will result in a deadlock when we try to issue
1143 * the read from the root filesystem. Instead we "prefetch"
1144 * the associated vnodes that we need prior to opening the
1145 * underlying devices and cache them so that we can prevent
1146 * any I/O when we are doing the actual open.
1148 if (spa_is_root(spa)) {
1149 int low = locks & ~(SCL_ZIO - 1);
1150 int high = locks & ~low;
1152 spa_config_enter(spa, high, spa, RW_WRITER);
1153 vdev_hold(spa->spa_root_vdev);
1154 spa_config_enter(spa, low, spa, RW_WRITER);
1156 spa_config_enter(spa, locks, spa, RW_WRITER);
1158 spa->spa_vdev_locks = locks;
1162 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1164 boolean_t config_changed = B_FALSE;
1166 if (vd != NULL || error == 0)
1167 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1171 vdev_state_dirty(vd->vdev_top);
1172 config_changed = B_TRUE;
1173 spa->spa_config_generation++;
1176 if (spa_is_root(spa))
1177 vdev_rele(spa->spa_root_vdev);
1179 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1180 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1183 * If anything changed, wait for it to sync. This ensures that,
1184 * from the system administrator's perspective, zpool(1M) commands
1185 * are synchronous. This is important for things like zpool offline:
1186 * when the command completes, you expect no further I/O from ZFS.
1189 txg_wait_synced(spa->spa_dsl_pool, 0);
1192 * If the config changed, update the config cache.
1194 if (config_changed) {
1195 mutex_enter(&spa_namespace_lock);
1196 spa_config_sync(spa, B_FALSE, B_TRUE);
1197 mutex_exit(&spa_namespace_lock);
1204 * ==========================================================================
1205 * Miscellaneous functions
1206 * ==========================================================================
1210 spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx)
1212 if (!nvlist_exists(spa->spa_label_features, feature)) {
1213 fnvlist_add_boolean(spa->spa_label_features, feature);
1215 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1216 * dirty the vdev config because lock SCL_CONFIG is not held.
1217 * Thankfully, in this case we don't need to dirty the config
1218 * because it will be written out anyway when we finish
1219 * creating the pool.
1221 if (tx->tx_txg != TXG_INITIAL)
1222 vdev_config_dirty(spa->spa_root_vdev);
1227 spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1229 if (nvlist_remove_all(spa->spa_label_features, feature) == 0)
1230 vdev_config_dirty(spa->spa_root_vdev);
1237 spa_rename(const char *name, const char *newname)
1243 * Lookup the spa_t and grab the config lock for writing. We need to
1244 * actually open the pool so that we can sync out the necessary labels.
1245 * It's OK to call spa_open() with the namespace lock held because we
1246 * allow recursive calls for other reasons.
1248 mutex_enter(&spa_namespace_lock);
1249 if ((err = spa_open(name, &spa, FTAG)) != 0) {
1250 mutex_exit(&spa_namespace_lock);
1254 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1256 avl_remove(&spa_namespace_avl, spa);
1257 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1258 avl_add(&spa_namespace_avl, spa);
1261 * Sync all labels to disk with the new names by marking the root vdev
1262 * dirty and waiting for it to sync. It will pick up the new pool name
1265 vdev_config_dirty(spa->spa_root_vdev);
1267 spa_config_exit(spa, SCL_ALL, FTAG);
1269 txg_wait_synced(spa->spa_dsl_pool, 0);
1272 * Sync the updated config cache.
1274 spa_config_sync(spa, B_FALSE, B_TRUE);
1276 spa_close(spa, FTAG);
1278 mutex_exit(&spa_namespace_lock);
1284 * Return the spa_t associated with given pool_guid, if it exists. If
1285 * device_guid is non-zero, determine whether the pool exists *and* contains
1286 * a device with the specified device_guid.
1289 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1292 avl_tree_t *t = &spa_namespace_avl;
1294 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1296 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1297 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1299 if (spa->spa_root_vdev == NULL)
1301 if (spa_guid(spa) == pool_guid) {
1302 if (device_guid == 0)
1305 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1306 device_guid) != NULL)
1310 * Check any devices we may be in the process of adding.
1312 if (spa->spa_pending_vdev) {
1313 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1314 device_guid) != NULL)
1324 * Determine whether a pool with the given pool_guid exists.
1327 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1329 return (spa_by_guid(pool_guid, device_guid) != NULL);
1333 spa_strdup(const char *s)
1339 new = kmem_alloc(len + 1, KM_SLEEP);
1347 spa_strfree(char *s)
1349 kmem_free(s, strlen(s) + 1);
1353 spa_get_random(uint64_t range)
1359 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1365 spa_generate_guid(spa_t *spa)
1367 uint64_t guid = spa_get_random(-1ULL);
1370 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1371 guid = spa_get_random(-1ULL);
1373 while (guid == 0 || spa_guid_exists(guid, 0))
1374 guid = spa_get_random(-1ULL);
1381 snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp)
1384 char *checksum = NULL;
1385 char *compress = NULL;
1388 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1389 dmu_object_byteswap_t bswap =
1390 DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1391 (void) snprintf(type, sizeof (type), "bswap %s %s",
1392 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1393 "metadata" : "data",
1394 dmu_ot_byteswap[bswap].ob_name);
1396 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1399 if (!BP_IS_EMBEDDED(bp)) {
1401 zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1403 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1406 SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum,
1411 spa_freeze(spa_t *spa)
1413 uint64_t freeze_txg = 0;
1415 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1416 if (spa->spa_freeze_txg == UINT64_MAX) {
1417 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1418 spa->spa_freeze_txg = freeze_txg;
1420 spa_config_exit(spa, SCL_ALL, FTAG);
1421 if (freeze_txg != 0)
1422 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1426 zfs_panic_recover(const char *fmt, ...)
1431 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1436 * This is a stripped-down version of strtoull, suitable only for converting
1437 * lowercase hexadecimal numbers that don't overflow.
1440 strtonum(const char *str, char **nptr)
1446 while ((c = *str) != '\0') {
1447 if (c >= '0' && c <= '9')
1449 else if (c >= 'a' && c <= 'f')
1450 digit = 10 + c - 'a';
1461 *nptr = (char *)str;
1467 * ==========================================================================
1468 * Accessor functions
1469 * ==========================================================================
1473 spa_shutting_down(spa_t *spa)
1475 return (spa->spa_async_suspended);
1479 spa_get_dsl(spa_t *spa)
1481 return (spa->spa_dsl_pool);
1485 spa_is_initializing(spa_t *spa)
1487 return (spa->spa_is_initializing);
1491 spa_get_rootblkptr(spa_t *spa)
1493 return (&spa->spa_ubsync.ub_rootbp);
1497 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1499 spa->spa_uberblock.ub_rootbp = *bp;
1503 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1505 if (spa->spa_root == NULL)
1508 (void) strncpy(buf, spa->spa_root, buflen);
1512 spa_sync_pass(spa_t *spa)
1514 return (spa->spa_sync_pass);
1518 spa_name(spa_t *spa)
1520 return (spa->spa_name);
1524 spa_guid(spa_t *spa)
1526 dsl_pool_t *dp = spa_get_dsl(spa);
1530 * If we fail to parse the config during spa_load(), we can go through
1531 * the error path (which posts an ereport) and end up here with no root
1532 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1535 if (spa->spa_root_vdev == NULL)
1536 return (spa->spa_config_guid);
1538 guid = spa->spa_last_synced_guid != 0 ?
1539 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1542 * Return the most recently synced out guid unless we're
1543 * in syncing context.
1545 if (dp && dsl_pool_sync_context(dp))
1546 return (spa->spa_root_vdev->vdev_guid);
1552 spa_load_guid(spa_t *spa)
1555 * This is a GUID that exists solely as a reference for the
1556 * purposes of the arc. It is generated at load time, and
1557 * is never written to persistent storage.
1559 return (spa->spa_load_guid);
1563 spa_last_synced_txg(spa_t *spa)
1565 return (spa->spa_ubsync.ub_txg);
1569 spa_first_txg(spa_t *spa)
1571 return (spa->spa_first_txg);
1575 spa_syncing_txg(spa_t *spa)
1577 return (spa->spa_syncing_txg);
1581 spa_state(spa_t *spa)
1583 return (spa->spa_state);
1587 spa_load_state(spa_t *spa)
1589 return (spa->spa_load_state);
1593 spa_freeze_txg(spa_t *spa)
1595 return (spa->spa_freeze_txg);
1600 spa_get_asize(spa_t *spa, uint64_t lsize)
1602 return (lsize * spa_asize_inflation);
1606 * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%),
1609 * See the comment above spa_slop_shift for details.
1612 spa_get_slop_space(spa_t *spa) {
1613 uint64_t space = spa_get_dspace(spa);
1614 return (MAX(space >> spa_slop_shift, SPA_MINDEVSIZE >> 1));
1618 spa_get_dspace(spa_t *spa)
1620 return (spa->spa_dspace);
1624 spa_update_dspace(spa_t *spa)
1626 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1627 ddt_get_dedup_dspace(spa);
1631 * Return the failure mode that has been set to this pool. The default
1632 * behavior will be to block all I/Os when a complete failure occurs.
1635 spa_get_failmode(spa_t *spa)
1637 return (spa->spa_failmode);
1641 spa_suspended(spa_t *spa)
1643 return (spa->spa_suspended);
1647 spa_version(spa_t *spa)
1649 return (spa->spa_ubsync.ub_version);
1653 spa_deflate(spa_t *spa)
1655 return (spa->spa_deflate);
1659 spa_normal_class(spa_t *spa)
1661 return (spa->spa_normal_class);
1665 spa_log_class(spa_t *spa)
1667 return (spa->spa_log_class);
1671 spa_evicting_os_register(spa_t *spa, objset_t *os)
1673 mutex_enter(&spa->spa_evicting_os_lock);
1674 list_insert_head(&spa->spa_evicting_os_list, os);
1675 mutex_exit(&spa->spa_evicting_os_lock);
1679 spa_evicting_os_deregister(spa_t *spa, objset_t *os)
1681 mutex_enter(&spa->spa_evicting_os_lock);
1682 list_remove(&spa->spa_evicting_os_list, os);
1683 cv_broadcast(&spa->spa_evicting_os_cv);
1684 mutex_exit(&spa->spa_evicting_os_lock);
1688 spa_evicting_os_wait(spa_t *spa)
1690 mutex_enter(&spa->spa_evicting_os_lock);
1691 while (!list_is_empty(&spa->spa_evicting_os_list))
1692 cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock);
1693 mutex_exit(&spa->spa_evicting_os_lock);
1695 dmu_buf_user_evict_wait();
1699 spa_max_replication(spa_t *spa)
1702 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1703 * handle BPs with more than one DVA allocated. Set our max
1704 * replication level accordingly.
1706 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1708 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1712 spa_prev_software_version(spa_t *spa)
1714 return (spa->spa_prev_software_version);
1718 spa_deadman_synctime(spa_t *spa)
1720 return (spa->spa_deadman_synctime);
1724 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1726 uint64_t asize = DVA_GET_ASIZE(dva);
1727 uint64_t dsize = asize;
1729 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1731 if (asize != 0 && spa->spa_deflate) {
1732 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1734 dsize = (asize >> SPA_MINBLOCKSHIFT) *
1735 vd->vdev_deflate_ratio;
1742 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1747 for (d = 0; d < BP_GET_NDVAS(bp); d++)
1748 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1754 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1759 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1761 for (d = 0; d < BP_GET_NDVAS(bp); d++)
1762 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1764 spa_config_exit(spa, SCL_VDEV, FTAG);
1770 * ==========================================================================
1771 * Initialization and Termination
1772 * ==========================================================================
1776 spa_name_compare(const void *a1, const void *a2)
1778 const spa_t *s1 = a1;
1779 const spa_t *s2 = a2;
1782 s = strcmp(s1->spa_name, s2->spa_name);
1784 return (AVL_ISIGN(s));
1796 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1797 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1798 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1799 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1801 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1802 offsetof(spa_t, spa_avl));
1804 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1805 offsetof(spa_aux_t, aux_avl));
1807 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1808 offsetof(spa_aux_t, aux_avl));
1810 spa_mode_global = mode;
1813 if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
1814 struct sigaction sa;
1816 sa.sa_flags = SA_SIGINFO;
1817 sigemptyset(&sa.sa_mask);
1818 sa.sa_sigaction = arc_buf_sigsegv;
1820 if (sigaction(SIGSEGV, &sa, NULL) == -1) {
1821 perror("could not enable watchpoints: "
1822 "sigaction(SIGSEGV, ...) = ");
1837 vdev_cache_stat_init();
1838 vdev_raidz_math_init();
1841 zpool_feature_init();
1853 vdev_cache_stat_fini();
1854 vdev_raidz_math_fini();
1864 avl_destroy(&spa_namespace_avl);
1865 avl_destroy(&spa_spare_avl);
1866 avl_destroy(&spa_l2cache_avl);
1868 cv_destroy(&spa_namespace_cv);
1869 mutex_destroy(&spa_namespace_lock);
1870 mutex_destroy(&spa_spare_lock);
1871 mutex_destroy(&spa_l2cache_lock);
1875 * Return whether this pool has slogs. No locking needed.
1876 * It's not a problem if the wrong answer is returned as it's only for
1877 * performance and not correctness
1880 spa_has_slogs(spa_t *spa)
1882 return (spa->spa_log_class->mc_rotor != NULL);
1886 spa_get_log_state(spa_t *spa)
1888 return (spa->spa_log_state);
1892 spa_set_log_state(spa_t *spa, spa_log_state_t state)
1894 spa->spa_log_state = state;
1898 spa_is_root(spa_t *spa)
1900 return (spa->spa_is_root);
1904 spa_writeable(spa_t *spa)
1906 return (!!(spa->spa_mode & FWRITE));
1910 * Returns true if there is a pending sync task in any of the current
1911 * syncing txg, the current quiescing txg, or the current open txg.
1914 spa_has_pending_synctask(spa_t *spa)
1916 return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks));
1920 spa_mode(spa_t *spa)
1922 return (spa->spa_mode);
1926 spa_bootfs(spa_t *spa)
1928 return (spa->spa_bootfs);
1932 spa_delegation(spa_t *spa)
1934 return (spa->spa_delegation);
1938 spa_meta_objset(spa_t *spa)
1940 return (spa->spa_meta_objset);
1944 spa_dedup_checksum(spa_t *spa)
1946 return (spa->spa_dedup_checksum);
1950 * Reset pool scan stat per scan pass (or reboot).
1953 spa_scan_stat_init(spa_t *spa)
1955 /* data not stored on disk */
1956 spa->spa_scan_pass_start = gethrestime_sec();
1957 spa->spa_scan_pass_exam = 0;
1958 vdev_scan_stat_init(spa->spa_root_vdev);
1962 * Get scan stats for zpool status reports
1965 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
1967 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
1969 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
1970 return (SET_ERROR(ENOENT));
1971 bzero(ps, sizeof (pool_scan_stat_t));
1973 /* data stored on disk */
1974 ps->pss_func = scn->scn_phys.scn_func;
1975 ps->pss_start_time = scn->scn_phys.scn_start_time;
1976 ps->pss_end_time = scn->scn_phys.scn_end_time;
1977 ps->pss_to_examine = scn->scn_phys.scn_to_examine;
1978 ps->pss_examined = scn->scn_phys.scn_examined;
1979 ps->pss_to_process = scn->scn_phys.scn_to_process;
1980 ps->pss_processed = scn->scn_phys.scn_processed;
1981 ps->pss_errors = scn->scn_phys.scn_errors;
1982 ps->pss_state = scn->scn_phys.scn_state;
1984 /* data not stored on disk */
1985 ps->pss_pass_start = spa->spa_scan_pass_start;
1986 ps->pss_pass_exam = spa->spa_scan_pass_exam;
1992 spa_debug_enabled(spa_t *spa)
1994 return (spa->spa_debug);
1998 spa_maxblocksize(spa_t *spa)
2000 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS))
2001 return (SPA_MAXBLOCKSIZE);
2003 return (SPA_OLD_MAXBLOCKSIZE);
2007 spa_maxdnodesize(spa_t *spa)
2009 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE))
2010 return (DNODE_MAX_SIZE);
2012 return (DNODE_MIN_SIZE);
2015 #if defined(_KERNEL) && defined(HAVE_SPL)
2016 /* Namespace manipulation */
2017 EXPORT_SYMBOL(spa_lookup);
2018 EXPORT_SYMBOL(spa_add);
2019 EXPORT_SYMBOL(spa_remove);
2020 EXPORT_SYMBOL(spa_next);
2022 /* Refcount functions */
2023 EXPORT_SYMBOL(spa_open_ref);
2024 EXPORT_SYMBOL(spa_close);
2025 EXPORT_SYMBOL(spa_refcount_zero);
2027 /* Pool configuration lock */
2028 EXPORT_SYMBOL(spa_config_tryenter);
2029 EXPORT_SYMBOL(spa_config_enter);
2030 EXPORT_SYMBOL(spa_config_exit);
2031 EXPORT_SYMBOL(spa_config_held);
2033 /* Pool vdev add/remove lock */
2034 EXPORT_SYMBOL(spa_vdev_enter);
2035 EXPORT_SYMBOL(spa_vdev_exit);
2037 /* Pool vdev state change lock */
2038 EXPORT_SYMBOL(spa_vdev_state_enter);
2039 EXPORT_SYMBOL(spa_vdev_state_exit);
2041 /* Accessor functions */
2042 EXPORT_SYMBOL(spa_shutting_down);
2043 EXPORT_SYMBOL(spa_get_dsl);
2044 EXPORT_SYMBOL(spa_get_rootblkptr);
2045 EXPORT_SYMBOL(spa_set_rootblkptr);
2046 EXPORT_SYMBOL(spa_altroot);
2047 EXPORT_SYMBOL(spa_sync_pass);
2048 EXPORT_SYMBOL(spa_name);
2049 EXPORT_SYMBOL(spa_guid);
2050 EXPORT_SYMBOL(spa_last_synced_txg);
2051 EXPORT_SYMBOL(spa_first_txg);
2052 EXPORT_SYMBOL(spa_syncing_txg);
2053 EXPORT_SYMBOL(spa_version);
2054 EXPORT_SYMBOL(spa_state);
2055 EXPORT_SYMBOL(spa_load_state);
2056 EXPORT_SYMBOL(spa_freeze_txg);
2057 EXPORT_SYMBOL(spa_get_asize);
2058 EXPORT_SYMBOL(spa_get_dspace);
2059 EXPORT_SYMBOL(spa_update_dspace);
2060 EXPORT_SYMBOL(spa_deflate);
2061 EXPORT_SYMBOL(spa_normal_class);
2062 EXPORT_SYMBOL(spa_log_class);
2063 EXPORT_SYMBOL(spa_max_replication);
2064 EXPORT_SYMBOL(spa_prev_software_version);
2065 EXPORT_SYMBOL(spa_get_failmode);
2066 EXPORT_SYMBOL(spa_suspended);
2067 EXPORT_SYMBOL(spa_bootfs);
2068 EXPORT_SYMBOL(spa_delegation);
2069 EXPORT_SYMBOL(spa_meta_objset);
2070 EXPORT_SYMBOL(spa_maxblocksize);
2071 EXPORT_SYMBOL(spa_maxdnodesize);
2073 /* Miscellaneous support routines */
2074 EXPORT_SYMBOL(spa_rename);
2075 EXPORT_SYMBOL(spa_guid_exists);
2076 EXPORT_SYMBOL(spa_strdup);
2077 EXPORT_SYMBOL(spa_strfree);
2078 EXPORT_SYMBOL(spa_get_random);
2079 EXPORT_SYMBOL(spa_generate_guid);
2080 EXPORT_SYMBOL(snprintf_blkptr);
2081 EXPORT_SYMBOL(spa_freeze);
2082 EXPORT_SYMBOL(spa_upgrade);
2083 EXPORT_SYMBOL(spa_evict_all);
2084 EXPORT_SYMBOL(spa_lookup_by_guid);
2085 EXPORT_SYMBOL(spa_has_spare);
2086 EXPORT_SYMBOL(dva_get_dsize_sync);
2087 EXPORT_SYMBOL(bp_get_dsize_sync);
2088 EXPORT_SYMBOL(bp_get_dsize);
2089 EXPORT_SYMBOL(spa_has_slogs);
2090 EXPORT_SYMBOL(spa_is_root);
2091 EXPORT_SYMBOL(spa_writeable);
2092 EXPORT_SYMBOL(spa_mode);
2094 EXPORT_SYMBOL(spa_namespace_lock);
2096 module_param(zfs_flags, uint, 0644);
2097 MODULE_PARM_DESC(zfs_flags, "Set additional debugging flags");
2099 module_param(zfs_recover, int, 0644);
2100 MODULE_PARM_DESC(zfs_recover, "Set to attempt to recover from fatal errors");
2102 module_param(zfs_free_leak_on_eio, int, 0644);
2103 MODULE_PARM_DESC(zfs_free_leak_on_eio,
2104 "Set to ignore IO errors during free and permanently leak the space");
2106 module_param(zfs_deadman_synctime_ms, ulong, 0644);
2107 MODULE_PARM_DESC(zfs_deadman_synctime_ms, "Expiration time in milliseconds");
2109 module_param(zfs_deadman_enabled, int, 0644);
2110 MODULE_PARM_DESC(zfs_deadman_enabled, "Enable deadman timer");
2112 module_param(spa_asize_inflation, int, 0644);
2113 MODULE_PARM_DESC(spa_asize_inflation,
2114 "SPA size estimate multiplication factor");
2116 module_param(spa_slop_shift, int, 0644);
2117 MODULE_PARM_DESC(spa_slop_shift, "Reserved free space in pool");