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, 2014 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
28 #include <sys/zfs_context.h>
29 #include <sys/spa_impl.h>
30 #include <sys/spa_boot.h>
32 #include <sys/zio_checksum.h>
33 #include <sys/zio_compress.h>
35 #include <sys/dmu_tx.h>
38 #include <sys/vdev_impl.h>
39 #include <sys/metaslab.h>
40 #include <sys/uberblock_impl.h>
43 #include <sys/unique.h>
44 #include <sys/dsl_pool.h>
45 #include <sys/dsl_dir.h>
46 #include <sys/dsl_prop.h>
47 #include <sys/dsl_scan.h>
48 #include <sys/fs/zfs.h>
49 #include <sys/metaslab_impl.h>
53 #include "zfeature_common.h"
58 * There are four basic locks for managing spa_t structures:
60 * spa_namespace_lock (global mutex)
62 * This lock must be acquired to do any of the following:
64 * - Lookup a spa_t by name
65 * - Add or remove a spa_t from the namespace
66 * - Increase spa_refcount from non-zero
67 * - Check if spa_refcount is zero
69 * - add/remove/attach/detach devices
70 * - Held for the duration of create/destroy/import/export
72 * It does not need to handle recursion. A create or destroy may
73 * reference objects (files or zvols) in other pools, but by
74 * definition they must have an existing reference, and will never need
75 * to lookup a spa_t by name.
77 * spa_refcount (per-spa refcount_t protected by mutex)
79 * This reference count keep track of any active users of the spa_t. The
80 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
81 * the refcount is never really 'zero' - opening a pool implicitly keeps
82 * some references in the DMU. Internally we check against spa_minref, but
83 * present the image of a zero/non-zero value to consumers.
85 * spa_config_lock[] (per-spa array of rwlocks)
87 * This protects the spa_t from config changes, and must be held in
88 * the following circumstances:
90 * - RW_READER to perform I/O to the spa
91 * - RW_WRITER to change the vdev config
93 * The locking order is fairly straightforward:
95 * spa_namespace_lock -> spa_refcount
97 * The namespace lock must be acquired to increase the refcount from 0
98 * or to check if it is zero.
100 * spa_refcount -> spa_config_lock[]
102 * There must be at least one valid reference on the spa_t to acquire
105 * spa_namespace_lock -> spa_config_lock[]
107 * The namespace lock must always be taken before the config lock.
110 * The spa_namespace_lock can be acquired directly and is globally visible.
112 * The namespace is manipulated using the following functions, all of which
113 * require the spa_namespace_lock to be held.
115 * spa_lookup() Lookup a spa_t by name.
117 * spa_add() Create a new spa_t in the namespace.
119 * spa_remove() Remove a spa_t from the namespace. This also
120 * frees up any memory associated with the spa_t.
122 * spa_next() Returns the next spa_t in the system, or the
123 * first if NULL is passed.
125 * spa_evict_all() Shutdown and remove all spa_t structures in
128 * spa_guid_exists() Determine whether a pool/device guid exists.
130 * The spa_refcount is manipulated using the following functions:
132 * spa_open_ref() Adds a reference to the given spa_t. Must be
133 * called with spa_namespace_lock held if the
134 * refcount is currently zero.
136 * spa_close() Remove a reference from the spa_t. This will
137 * not free the spa_t or remove it from the
138 * namespace. No locking is required.
140 * spa_refcount_zero() Returns true if the refcount is currently
141 * zero. Must be called with spa_namespace_lock
144 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
145 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
146 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
148 * To read the configuration, it suffices to hold one of these locks as reader.
149 * To modify the configuration, you must hold all locks as writer. To modify
150 * vdev state without altering the vdev tree's topology (e.g. online/offline),
151 * you must hold SCL_STATE and SCL_ZIO as writer.
153 * We use these distinct config locks to avoid recursive lock entry.
154 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
155 * block allocations (SCL_ALLOC), which may require reading space maps
156 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
158 * The spa config locks cannot be normal rwlocks because we need the
159 * ability to hand off ownership. For example, SCL_ZIO is acquired
160 * by the issuing thread and later released by an interrupt thread.
161 * They do, however, obey the usual write-wanted semantics to prevent
162 * writer (i.e. system administrator) starvation.
164 * The lock acquisition rules are as follows:
167 * Protects changes to the vdev tree topology, such as vdev
168 * add/remove/attach/detach. Protects the dirty config list
169 * (spa_config_dirty_list) and the set of spares and l2arc devices.
172 * Protects changes to pool state and vdev state, such as vdev
173 * online/offline/fault/degrade/clear. Protects the dirty state list
174 * (spa_state_dirty_list) and global pool state (spa_state).
177 * Protects changes to metaslab groups and classes.
178 * Held as reader by metaslab_alloc() and metaslab_claim().
181 * Held by bp-level zios (those which have no io_vd upon entry)
182 * to prevent changes to the vdev tree. The bp-level zio implicitly
183 * protects all of its vdev child zios, which do not hold SCL_ZIO.
186 * Protects changes to metaslab groups and classes.
187 * Held as reader by metaslab_free(). SCL_FREE is distinct from
188 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
189 * blocks in zio_done() while another i/o that holds either
190 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
193 * Held as reader to prevent changes to the vdev tree during trivial
194 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
195 * other locks, and lower than all of them, to ensure that it's safe
196 * to acquire regardless of caller context.
198 * In addition, the following rules apply:
200 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
201 * The lock ordering is SCL_CONFIG > spa_props_lock.
203 * (b) I/O operations on leaf vdevs. For any zio operation that takes
204 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
205 * or zio_write_phys() -- the caller must ensure that the config cannot
206 * cannot change in the interim, and that the vdev cannot be reopened.
207 * SCL_STATE as reader suffices for both.
209 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
211 * spa_vdev_enter() Acquire the namespace lock and the config lock
214 * spa_vdev_exit() Release the config lock, wait for all I/O
215 * to complete, sync the updated configs to the
216 * cache, and release the namespace lock.
218 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
219 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
220 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
222 * spa_rename() is also implemented within this file since it requires
223 * manipulation of the namespace.
226 static avl_tree_t spa_namespace_avl;
227 kmutex_t spa_namespace_lock;
228 static kcondvar_t spa_namespace_cv;
229 static int spa_active_count;
230 int spa_max_replication_override = SPA_DVAS_PER_BP;
232 static kmutex_t spa_spare_lock;
233 static avl_tree_t spa_spare_avl;
234 static kmutex_t spa_l2cache_lock;
235 static avl_tree_t spa_l2cache_avl;
237 kmem_cache_t *spa_buffer_pool;
241 /* Everything except dprintf and spa is on by default in debug builds */
242 int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA);
247 TUNABLE_INT("debug.zfs_flags", &zfs_flags);
248 SYSCTL_INT(_debug, OID_AUTO, zfs_flags, CTLFLAG_RWTUN, &zfs_flags, 0,
252 * zfs_recover can be set to nonzero to attempt to recover from
253 * otherwise-fatal errors, typically caused by on-disk corruption. When
254 * set, calls to zfs_panic_recover() will turn into warning messages.
255 * This should only be used as a last resort, as it typically results
256 * in leaked space, or worse.
258 boolean_t zfs_recover = B_FALSE;
259 SYSCTL_DECL(_vfs_zfs);
260 TUNABLE_INT("vfs.zfs.recover", &zfs_recover);
261 SYSCTL_INT(_vfs_zfs, OID_AUTO, recover, CTLFLAG_RDTUN, &zfs_recover, 0,
262 "Try to recover from otherwise-fatal errors.");
265 * If destroy encounters an EIO while reading metadata (e.g. indirect
266 * blocks), space referenced by the missing metadata can not be freed.
267 * Normally this causes the background destroy to become "stalled", as
268 * it is unable to make forward progress. While in this stalled state,
269 * all remaining space to free from the error-encountering filesystem is
270 * "temporarily leaked". Set this flag to cause it to ignore the EIO,
271 * permanently leak the space from indirect blocks that can not be read,
272 * and continue to free everything else that it can.
274 * The default, "stalling" behavior is useful if the storage partially
275 * fails (i.e. some but not all i/os fail), and then later recovers. In
276 * this case, we will be able to continue pool operations while it is
277 * partially failed, and when it recovers, we can continue to free the
278 * space, with no leaks. However, note that this case is actually
281 * Typically pools either (a) fail completely (but perhaps temporarily,
282 * e.g. a top-level vdev going offline), or (b) have localized,
283 * permanent errors (e.g. disk returns the wrong data due to bit flip or
284 * firmware bug). In case (a), this setting does not matter because the
285 * pool will be suspended and the sync thread will not be able to make
286 * forward progress regardless. In case (b), because the error is
287 * permanent, the best we can do is leak the minimum amount of space,
288 * which is what setting this flag will do. Therefore, it is reasonable
289 * for this flag to normally be set, but we chose the more conservative
290 * approach of not setting it, so that there is no possibility of
291 * leaking space in the "partial temporary" failure case.
293 boolean_t zfs_free_leak_on_eio = B_FALSE;
296 * Expiration time in milliseconds. This value has two meanings. First it is
297 * used to determine when the spa_deadman() logic should fire. By default the
298 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
299 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
300 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
303 uint64_t zfs_deadman_synctime_ms = 1000000ULL;
304 TUNABLE_QUAD("vfs.zfs.deadman_synctime_ms", &zfs_deadman_synctime_ms);
305 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_synctime_ms, CTLFLAG_RDTUN,
306 &zfs_deadman_synctime_ms, 0,
307 "Stalled ZFS I/O expiration time in milliseconds");
310 * Check time in milliseconds. This defines the frequency at which we check
313 uint64_t zfs_deadman_checktime_ms = 5000ULL;
314 TUNABLE_QUAD("vfs.zfs.deadman_checktime_ms", &zfs_deadman_checktime_ms);
315 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_checktime_ms, CTLFLAG_RDTUN,
316 &zfs_deadman_checktime_ms, 0,
317 "Period of checks for stalled ZFS I/O in milliseconds");
320 * Default value of -1 for zfs_deadman_enabled is resolved in
323 int zfs_deadman_enabled = -1;
324 TUNABLE_INT("vfs.zfs.deadman_enabled", &zfs_deadman_enabled);
325 SYSCTL_INT(_vfs_zfs, OID_AUTO, deadman_enabled, CTLFLAG_RDTUN,
326 &zfs_deadman_enabled, 0, "Kernel panic on stalled ZFS I/O");
329 * The worst case is single-sector max-parity RAID-Z blocks, in which
330 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
331 * times the size; so just assume that. Add to this the fact that
332 * we can have up to 3 DVAs per bp, and one more factor of 2 because
333 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together,
335 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
337 int spa_asize_inflation = 24;
338 TUNABLE_INT("vfs.zfs.spa_asize_inflation", &spa_asize_inflation);
339 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_asize_inflation, CTLFLAG_RWTUN,
340 &spa_asize_inflation, 0, "Worst case inflation factor for single sector writes");
348 * If we are not i386 or amd64 or in a virtual machine,
349 * disable ZFS deadman thread by default
351 if (zfs_deadman_enabled == -1) {
352 #if defined(__amd64__) || defined(__i386__)
353 zfs_deadman_enabled = (vm_guest == VM_GUEST_NO) ? 1 : 0;
355 zfs_deadman_enabled = 0;
360 #endif /* !illumos */
363 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
364 * the pool to be consumed. This ensures that we don't run the pool
365 * completely out of space, due to unaccounted changes (e.g. to the MOS).
366 * It also limits the worst-case time to allocate space. If we have
367 * less than this amount of free space, most ZPL operations (e.g. write,
368 * create) will return ENOSPC.
370 * Certain operations (e.g. file removal, most administrative actions) can
371 * use half the slop space. They will only return ENOSPC if less than half
372 * the slop space is free. Typically, once the pool has less than the slop
373 * space free, the user will use these operations to free up space in the pool.
374 * These are the operations that call dsl_pool_adjustedsize() with the netfree
375 * argument set to TRUE.
377 * A very restricted set of operations are always permitted, regardless of
378 * the amount of free space. These are the operations that call
379 * dsl_sync_task(ZFS_SPACE_CHECK_NONE), e.g. "zfs destroy". If these
380 * operations result in a net increase in the amount of space used,
381 * it is possible to run the pool completely out of space, causing it to
382 * be permanently read-only.
384 * See also the comments in zfs_space_check_t.
386 int spa_slop_shift = 5;
389 * ==========================================================================
391 * ==========================================================================
394 spa_config_lock_init(spa_t *spa)
396 for (int i = 0; i < SCL_LOCKS; i++) {
397 spa_config_lock_t *scl = &spa->spa_config_lock[i];
398 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
399 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
400 refcount_create_untracked(&scl->scl_count);
401 scl->scl_writer = NULL;
402 scl->scl_write_wanted = 0;
407 spa_config_lock_destroy(spa_t *spa)
409 for (int i = 0; i < SCL_LOCKS; i++) {
410 spa_config_lock_t *scl = &spa->spa_config_lock[i];
411 mutex_destroy(&scl->scl_lock);
412 cv_destroy(&scl->scl_cv);
413 refcount_destroy(&scl->scl_count);
414 ASSERT(scl->scl_writer == NULL);
415 ASSERT(scl->scl_write_wanted == 0);
420 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
422 for (int i = 0; i < SCL_LOCKS; i++) {
423 spa_config_lock_t *scl = &spa->spa_config_lock[i];
424 if (!(locks & (1 << i)))
426 mutex_enter(&scl->scl_lock);
427 if (rw == RW_READER) {
428 if (scl->scl_writer || scl->scl_write_wanted) {
429 mutex_exit(&scl->scl_lock);
430 spa_config_exit(spa, locks ^ (1 << i), tag);
434 ASSERT(scl->scl_writer != curthread);
435 if (!refcount_is_zero(&scl->scl_count)) {
436 mutex_exit(&scl->scl_lock);
437 spa_config_exit(spa, locks ^ (1 << i), tag);
440 scl->scl_writer = curthread;
442 (void) refcount_add(&scl->scl_count, tag);
443 mutex_exit(&scl->scl_lock);
449 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
453 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
455 for (int i = 0; i < SCL_LOCKS; i++) {
456 spa_config_lock_t *scl = &spa->spa_config_lock[i];
457 if (scl->scl_writer == curthread)
458 wlocks_held |= (1 << i);
459 if (!(locks & (1 << i)))
461 mutex_enter(&scl->scl_lock);
462 if (rw == RW_READER) {
463 while (scl->scl_writer || scl->scl_write_wanted) {
464 cv_wait(&scl->scl_cv, &scl->scl_lock);
467 ASSERT(scl->scl_writer != curthread);
468 while (!refcount_is_zero(&scl->scl_count)) {
469 scl->scl_write_wanted++;
470 cv_wait(&scl->scl_cv, &scl->scl_lock);
471 scl->scl_write_wanted--;
473 scl->scl_writer = curthread;
475 (void) refcount_add(&scl->scl_count, tag);
476 mutex_exit(&scl->scl_lock);
478 ASSERT(wlocks_held <= locks);
482 spa_config_exit(spa_t *spa, int locks, void *tag)
484 for (int i = SCL_LOCKS - 1; i >= 0; i--) {
485 spa_config_lock_t *scl = &spa->spa_config_lock[i];
486 if (!(locks & (1 << i)))
488 mutex_enter(&scl->scl_lock);
489 ASSERT(!refcount_is_zero(&scl->scl_count));
490 if (refcount_remove(&scl->scl_count, tag) == 0) {
491 ASSERT(scl->scl_writer == NULL ||
492 scl->scl_writer == curthread);
493 scl->scl_writer = NULL; /* OK in either case */
494 cv_broadcast(&scl->scl_cv);
496 mutex_exit(&scl->scl_lock);
501 spa_config_held(spa_t *spa, int locks, krw_t rw)
505 for (int i = 0; i < SCL_LOCKS; i++) {
506 spa_config_lock_t *scl = &spa->spa_config_lock[i];
507 if (!(locks & (1 << i)))
509 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
510 (rw == RW_WRITER && scl->scl_writer == curthread))
511 locks_held |= 1 << i;
518 * ==========================================================================
519 * SPA namespace functions
520 * ==========================================================================
524 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
525 * Returns NULL if no matching spa_t is found.
528 spa_lookup(const char *name)
530 static spa_t search; /* spa_t is large; don't allocate on stack */
535 ASSERT(MUTEX_HELD(&spa_namespace_lock));
537 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
540 * If it's a full dataset name, figure out the pool name and
543 cp = strpbrk(search.spa_name, "/@#");
547 spa = avl_find(&spa_namespace_avl, &search, &where);
553 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
554 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
555 * looking for potentially hung I/Os.
558 spa_deadman(void *arg)
563 * Disable the deadman timer if the pool is suspended.
565 if (spa_suspended(spa)) {
567 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
569 /* Nothing. just don't schedule any future callouts. */
574 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
575 (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
576 ++spa->spa_deadman_calls);
577 if (zfs_deadman_enabled)
578 vdev_deadman(spa->spa_root_vdev);
582 * Create an uninitialized spa_t with the given name. Requires
583 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
584 * exist by calling spa_lookup() first.
587 spa_add(const char *name, nvlist_t *config, const char *altroot)
590 spa_config_dirent_t *dp;
596 ASSERT(MUTEX_HELD(&spa_namespace_lock));
598 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
600 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
601 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
602 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
603 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
604 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
605 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
606 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
607 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
608 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
610 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
611 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
612 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
613 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
615 for (int t = 0; t < TXG_SIZE; t++)
616 bplist_create(&spa->spa_free_bplist[t]);
618 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
619 spa->spa_state = POOL_STATE_UNINITIALIZED;
620 spa->spa_freeze_txg = UINT64_MAX;
621 spa->spa_final_txg = UINT64_MAX;
622 spa->spa_load_max_txg = UINT64_MAX;
624 spa->spa_proc_state = SPA_PROC_NONE;
627 hdlr.cyh_func = spa_deadman;
629 hdlr.cyh_level = CY_LOW_LEVEL;
632 spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
636 * This determines how often we need to check for hung I/Os after
637 * the cyclic has already fired. Since checking for hung I/Os is
638 * an expensive operation we don't want to check too frequently.
639 * Instead wait for 5 seconds before checking again.
641 when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms);
642 when.cyt_when = CY_INFINITY;
643 mutex_enter(&cpu_lock);
644 spa->spa_deadman_cycid = cyclic_add(&hdlr, &when);
645 mutex_exit(&cpu_lock);
648 callout_init(&spa->spa_deadman_cycid, CALLOUT_MPSAFE);
651 refcount_create(&spa->spa_refcount);
652 spa_config_lock_init(spa);
654 avl_add(&spa_namespace_avl, spa);
657 * Set the alternate root, if there is one.
660 spa->spa_root = spa_strdup(altroot);
665 * Every pool starts with the default cachefile
667 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
668 offsetof(spa_config_dirent_t, scd_link));
670 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
671 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
672 list_insert_head(&spa->spa_config_list, dp);
674 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
677 if (config != NULL) {
680 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
682 VERIFY(nvlist_dup(features, &spa->spa_label_features,
686 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
689 if (spa->spa_label_features == NULL) {
690 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
694 spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0);
697 * As a pool is being created, treat all features as disabled by
698 * setting SPA_FEATURE_DISABLED for all entries in the feature
701 for (int i = 0; i < SPA_FEATURES; i++) {
702 spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED;
709 * Removes a spa_t from the namespace, freeing up any memory used. Requires
710 * spa_namespace_lock. This is called only after the spa_t has been closed and
714 spa_remove(spa_t *spa)
716 spa_config_dirent_t *dp;
718 ASSERT(MUTEX_HELD(&spa_namespace_lock));
719 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
721 nvlist_free(spa->spa_config_splitting);
723 avl_remove(&spa_namespace_avl, spa);
724 cv_broadcast(&spa_namespace_cv);
727 spa_strfree(spa->spa_root);
731 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
732 list_remove(&spa->spa_config_list, dp);
733 if (dp->scd_path != NULL)
734 spa_strfree(dp->scd_path);
735 kmem_free(dp, sizeof (spa_config_dirent_t));
738 list_destroy(&spa->spa_config_list);
740 nvlist_free(spa->spa_label_features);
741 nvlist_free(spa->spa_load_info);
742 spa_config_set(spa, NULL);
745 mutex_enter(&cpu_lock);
746 if (spa->spa_deadman_cycid != CYCLIC_NONE)
747 cyclic_remove(spa->spa_deadman_cycid);
748 mutex_exit(&cpu_lock);
749 spa->spa_deadman_cycid = CYCLIC_NONE;
752 callout_drain(&spa->spa_deadman_cycid);
756 refcount_destroy(&spa->spa_refcount);
758 spa_config_lock_destroy(spa);
760 for (int t = 0; t < TXG_SIZE; t++)
761 bplist_destroy(&spa->spa_free_bplist[t]);
763 cv_destroy(&spa->spa_async_cv);
764 cv_destroy(&spa->spa_proc_cv);
765 cv_destroy(&spa->spa_scrub_io_cv);
766 cv_destroy(&spa->spa_suspend_cv);
768 mutex_destroy(&spa->spa_async_lock);
769 mutex_destroy(&spa->spa_errlist_lock);
770 mutex_destroy(&spa->spa_errlog_lock);
771 mutex_destroy(&spa->spa_history_lock);
772 mutex_destroy(&spa->spa_proc_lock);
773 mutex_destroy(&spa->spa_props_lock);
774 mutex_destroy(&spa->spa_scrub_lock);
775 mutex_destroy(&spa->spa_suspend_lock);
776 mutex_destroy(&spa->spa_vdev_top_lock);
778 kmem_free(spa, sizeof (spa_t));
782 * Given a pool, return the next pool in the namespace, or NULL if there is
783 * none. If 'prev' is NULL, return the first pool.
786 spa_next(spa_t *prev)
788 ASSERT(MUTEX_HELD(&spa_namespace_lock));
791 return (AVL_NEXT(&spa_namespace_avl, prev));
793 return (avl_first(&spa_namespace_avl));
797 * ==========================================================================
798 * SPA refcount functions
799 * ==========================================================================
803 * Add a reference to the given spa_t. Must have at least one reference, or
804 * have the namespace lock held.
807 spa_open_ref(spa_t *spa, void *tag)
809 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
810 MUTEX_HELD(&spa_namespace_lock));
811 (void) refcount_add(&spa->spa_refcount, tag);
815 * Remove a reference to the given spa_t. Must have at least one reference, or
816 * have the namespace lock held.
819 spa_close(spa_t *spa, void *tag)
821 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
822 MUTEX_HELD(&spa_namespace_lock));
823 (void) refcount_remove(&spa->spa_refcount, tag);
827 * Check to see if the spa refcount is zero. Must be called with
828 * spa_namespace_lock held. We really compare against spa_minref, which is the
829 * number of references acquired when opening a pool
832 spa_refcount_zero(spa_t *spa)
834 ASSERT(MUTEX_HELD(&spa_namespace_lock));
836 return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
840 * ==========================================================================
841 * SPA spare and l2cache tracking
842 * ==========================================================================
846 * Hot spares and cache devices are tracked using the same code below,
847 * for 'auxiliary' devices.
850 typedef struct spa_aux {
858 spa_aux_compare(const void *a, const void *b)
860 const spa_aux_t *sa = a;
861 const spa_aux_t *sb = b;
863 if (sa->aux_guid < sb->aux_guid)
865 else if (sa->aux_guid > sb->aux_guid)
872 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
878 search.aux_guid = vd->vdev_guid;
879 if ((aux = avl_find(avl, &search, &where)) != NULL) {
882 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
883 aux->aux_guid = vd->vdev_guid;
885 avl_insert(avl, aux, where);
890 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
896 search.aux_guid = vd->vdev_guid;
897 aux = avl_find(avl, &search, &where);
901 if (--aux->aux_count == 0) {
902 avl_remove(avl, aux);
903 kmem_free(aux, sizeof (spa_aux_t));
904 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
905 aux->aux_pool = 0ULL;
910 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
912 spa_aux_t search, *found;
914 search.aux_guid = guid;
915 found = avl_find(avl, &search, NULL);
919 *pool = found->aux_pool;
926 *refcnt = found->aux_count;
931 return (found != NULL);
935 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
937 spa_aux_t search, *found;
940 search.aux_guid = vd->vdev_guid;
941 found = avl_find(avl, &search, &where);
942 ASSERT(found != NULL);
943 ASSERT(found->aux_pool == 0ULL);
945 found->aux_pool = spa_guid(vd->vdev_spa);
949 * Spares are tracked globally due to the following constraints:
951 * - A spare may be part of multiple pools.
952 * - A spare may be added to a pool even if it's actively in use within
954 * - A spare in use in any pool can only be the source of a replacement if
955 * the target is a spare in the same pool.
957 * We keep track of all spares on the system through the use of a reference
958 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
959 * spare, then we bump the reference count in the AVL tree. In addition, we set
960 * the 'vdev_isspare' member to indicate that the device is a spare (active or
961 * inactive). When a spare is made active (used to replace a device in the
962 * pool), we also keep track of which pool its been made a part of.
964 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
965 * called under the spa_namespace lock as part of vdev reconfiguration. The
966 * separate spare lock exists for the status query path, which does not need to
967 * be completely consistent with respect to other vdev configuration changes.
971 spa_spare_compare(const void *a, const void *b)
973 return (spa_aux_compare(a, b));
977 spa_spare_add(vdev_t *vd)
979 mutex_enter(&spa_spare_lock);
980 ASSERT(!vd->vdev_isspare);
981 spa_aux_add(vd, &spa_spare_avl);
982 vd->vdev_isspare = B_TRUE;
983 mutex_exit(&spa_spare_lock);
987 spa_spare_remove(vdev_t *vd)
989 mutex_enter(&spa_spare_lock);
990 ASSERT(vd->vdev_isspare);
991 spa_aux_remove(vd, &spa_spare_avl);
992 vd->vdev_isspare = B_FALSE;
993 mutex_exit(&spa_spare_lock);
997 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
1001 mutex_enter(&spa_spare_lock);
1002 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
1003 mutex_exit(&spa_spare_lock);
1009 spa_spare_activate(vdev_t *vd)
1011 mutex_enter(&spa_spare_lock);
1012 ASSERT(vd->vdev_isspare);
1013 spa_aux_activate(vd, &spa_spare_avl);
1014 mutex_exit(&spa_spare_lock);
1018 * Level 2 ARC devices are tracked globally for the same reasons as spares.
1019 * Cache devices currently only support one pool per cache device, and so
1020 * for these devices the aux reference count is currently unused beyond 1.
1024 spa_l2cache_compare(const void *a, const void *b)
1026 return (spa_aux_compare(a, b));
1030 spa_l2cache_add(vdev_t *vd)
1032 mutex_enter(&spa_l2cache_lock);
1033 ASSERT(!vd->vdev_isl2cache);
1034 spa_aux_add(vd, &spa_l2cache_avl);
1035 vd->vdev_isl2cache = B_TRUE;
1036 mutex_exit(&spa_l2cache_lock);
1040 spa_l2cache_remove(vdev_t *vd)
1042 mutex_enter(&spa_l2cache_lock);
1043 ASSERT(vd->vdev_isl2cache);
1044 spa_aux_remove(vd, &spa_l2cache_avl);
1045 vd->vdev_isl2cache = B_FALSE;
1046 mutex_exit(&spa_l2cache_lock);
1050 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
1054 mutex_enter(&spa_l2cache_lock);
1055 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
1056 mutex_exit(&spa_l2cache_lock);
1062 spa_l2cache_activate(vdev_t *vd)
1064 mutex_enter(&spa_l2cache_lock);
1065 ASSERT(vd->vdev_isl2cache);
1066 spa_aux_activate(vd, &spa_l2cache_avl);
1067 mutex_exit(&spa_l2cache_lock);
1071 * ==========================================================================
1073 * ==========================================================================
1077 * Lock the given spa_t for the purpose of adding or removing a vdev.
1078 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1079 * It returns the next transaction group for the spa_t.
1082 spa_vdev_enter(spa_t *spa)
1084 mutex_enter(&spa->spa_vdev_top_lock);
1085 mutex_enter(&spa_namespace_lock);
1086 return (spa_vdev_config_enter(spa));
1090 * Internal implementation for spa_vdev_enter(). Used when a vdev
1091 * operation requires multiple syncs (i.e. removing a device) while
1092 * keeping the spa_namespace_lock held.
1095 spa_vdev_config_enter(spa_t *spa)
1097 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1099 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1101 return (spa_last_synced_txg(spa) + 1);
1105 * Used in combination with spa_vdev_config_enter() to allow the syncing
1106 * of multiple transactions without releasing the spa_namespace_lock.
1109 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
1111 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1113 int config_changed = B_FALSE;
1115 ASSERT(txg > spa_last_synced_txg(spa));
1117 spa->spa_pending_vdev = NULL;
1120 * Reassess the DTLs.
1122 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
1124 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
1125 config_changed = B_TRUE;
1126 spa->spa_config_generation++;
1130 * Verify the metaslab classes.
1132 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
1133 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
1135 spa_config_exit(spa, SCL_ALL, spa);
1138 * Panic the system if the specified tag requires it. This
1139 * is useful for ensuring that configurations are updated
1142 if (zio_injection_enabled)
1143 zio_handle_panic_injection(spa, tag, 0);
1146 * Note: this txg_wait_synced() is important because it ensures
1147 * that there won't be more than one config change per txg.
1148 * This allows us to use the txg as the generation number.
1151 txg_wait_synced(spa->spa_dsl_pool, txg);
1154 ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
1155 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1157 spa_config_exit(spa, SCL_ALL, spa);
1161 * If the config changed, update the config cache.
1164 spa_config_sync(spa, B_FALSE, B_TRUE);
1168 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1169 * locking of spa_vdev_enter(), we also want make sure the transactions have
1170 * synced to disk, and then update the global configuration cache with the new
1174 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1176 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1177 mutex_exit(&spa_namespace_lock);
1178 mutex_exit(&spa->spa_vdev_top_lock);
1184 * Lock the given spa_t for the purpose of changing vdev state.
1187 spa_vdev_state_enter(spa_t *spa, int oplocks)
1189 int locks = SCL_STATE_ALL | oplocks;
1192 * Root pools may need to read of the underlying devfs filesystem
1193 * when opening up a vdev. Unfortunately if we're holding the
1194 * SCL_ZIO lock it will result in a deadlock when we try to issue
1195 * the read from the root filesystem. Instead we "prefetch"
1196 * the associated vnodes that we need prior to opening the
1197 * underlying devices and cache them so that we can prevent
1198 * any I/O when we are doing the actual open.
1200 if (spa_is_root(spa)) {
1201 int low = locks & ~(SCL_ZIO - 1);
1202 int high = locks & ~low;
1204 spa_config_enter(spa, high, spa, RW_WRITER);
1205 vdev_hold(spa->spa_root_vdev);
1206 spa_config_enter(spa, low, spa, RW_WRITER);
1208 spa_config_enter(spa, locks, spa, RW_WRITER);
1210 spa->spa_vdev_locks = locks;
1214 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1216 boolean_t config_changed = B_FALSE;
1218 if (vd != NULL || error == 0)
1219 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1223 vdev_state_dirty(vd->vdev_top);
1224 config_changed = B_TRUE;
1225 spa->spa_config_generation++;
1228 if (spa_is_root(spa))
1229 vdev_rele(spa->spa_root_vdev);
1231 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1232 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1235 * If anything changed, wait for it to sync. This ensures that,
1236 * from the system administrator's perspective, zpool(1M) commands
1237 * are synchronous. This is important for things like zpool offline:
1238 * when the command completes, you expect no further I/O from ZFS.
1241 txg_wait_synced(spa->spa_dsl_pool, 0);
1244 * If the config changed, update the config cache.
1246 if (config_changed) {
1247 mutex_enter(&spa_namespace_lock);
1248 spa_config_sync(spa, B_FALSE, B_TRUE);
1249 mutex_exit(&spa_namespace_lock);
1256 * ==========================================================================
1257 * Miscellaneous functions
1258 * ==========================================================================
1262 spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx)
1264 if (!nvlist_exists(spa->spa_label_features, feature)) {
1265 fnvlist_add_boolean(spa->spa_label_features, feature);
1267 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1268 * dirty the vdev config because lock SCL_CONFIG is not held.
1269 * Thankfully, in this case we don't need to dirty the config
1270 * because it will be written out anyway when we finish
1271 * creating the pool.
1273 if (tx->tx_txg != TXG_INITIAL)
1274 vdev_config_dirty(spa->spa_root_vdev);
1279 spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1281 if (nvlist_remove_all(spa->spa_label_features, feature) == 0)
1282 vdev_config_dirty(spa->spa_root_vdev);
1289 spa_rename(const char *name, const char *newname)
1295 * Lookup the spa_t and grab the config lock for writing. We need to
1296 * actually open the pool so that we can sync out the necessary labels.
1297 * It's OK to call spa_open() with the namespace lock held because we
1298 * allow recursive calls for other reasons.
1300 mutex_enter(&spa_namespace_lock);
1301 if ((err = spa_open(name, &spa, FTAG)) != 0) {
1302 mutex_exit(&spa_namespace_lock);
1306 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1308 avl_remove(&spa_namespace_avl, spa);
1309 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1310 avl_add(&spa_namespace_avl, spa);
1313 * Sync all labels to disk with the new names by marking the root vdev
1314 * dirty and waiting for it to sync. It will pick up the new pool name
1317 vdev_config_dirty(spa->spa_root_vdev);
1319 spa_config_exit(spa, SCL_ALL, FTAG);
1321 txg_wait_synced(spa->spa_dsl_pool, 0);
1324 * Sync the updated config cache.
1326 spa_config_sync(spa, B_FALSE, B_TRUE);
1328 spa_close(spa, FTAG);
1330 mutex_exit(&spa_namespace_lock);
1336 * Return the spa_t associated with given pool_guid, if it exists. If
1337 * device_guid is non-zero, determine whether the pool exists *and* contains
1338 * a device with the specified device_guid.
1341 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1344 avl_tree_t *t = &spa_namespace_avl;
1346 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1348 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1349 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1351 if (spa->spa_root_vdev == NULL)
1353 if (spa_guid(spa) == pool_guid) {
1354 if (device_guid == 0)
1357 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1358 device_guid) != NULL)
1362 * Check any devices we may be in the process of adding.
1364 if (spa->spa_pending_vdev) {
1365 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1366 device_guid) != NULL)
1376 * Determine whether a pool with the given pool_guid exists.
1379 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1381 return (spa_by_guid(pool_guid, device_guid) != NULL);
1385 spa_strdup(const char *s)
1391 new = kmem_alloc(len + 1, KM_SLEEP);
1399 spa_strfree(char *s)
1401 kmem_free(s, strlen(s) + 1);
1405 spa_get_random(uint64_t range)
1411 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1417 spa_generate_guid(spa_t *spa)
1419 uint64_t guid = spa_get_random(-1ULL);
1422 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1423 guid = spa_get_random(-1ULL);
1425 while (guid == 0 || spa_guid_exists(guid, 0))
1426 guid = spa_get_random(-1ULL);
1433 snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp)
1436 char *checksum = NULL;
1437 char *compress = NULL;
1440 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1441 dmu_object_byteswap_t bswap =
1442 DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1443 (void) snprintf(type, sizeof (type), "bswap %s %s",
1444 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1445 "metadata" : "data",
1446 dmu_ot_byteswap[bswap].ob_name);
1448 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1451 if (!BP_IS_EMBEDDED(bp)) {
1453 zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1455 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1458 SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum,
1463 spa_freeze(spa_t *spa)
1465 uint64_t freeze_txg = 0;
1467 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1468 if (spa->spa_freeze_txg == UINT64_MAX) {
1469 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1470 spa->spa_freeze_txg = freeze_txg;
1472 spa_config_exit(spa, SCL_ALL, FTAG);
1473 if (freeze_txg != 0)
1474 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1478 zfs_panic_recover(const char *fmt, ...)
1483 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1488 * This is a stripped-down version of strtoull, suitable only for converting
1489 * lowercase hexadecimal numbers that don't overflow.
1492 zfs_strtonum(const char *str, char **nptr)
1498 while ((c = *str) != '\0') {
1499 if (c >= '0' && c <= '9')
1501 else if (c >= 'a' && c <= 'f')
1502 digit = 10 + c - 'a';
1513 *nptr = (char *)str;
1519 * ==========================================================================
1520 * Accessor functions
1521 * ==========================================================================
1525 spa_shutting_down(spa_t *spa)
1527 return (spa->spa_async_suspended);
1531 spa_get_dsl(spa_t *spa)
1533 return (spa->spa_dsl_pool);
1537 spa_is_initializing(spa_t *spa)
1539 return (spa->spa_is_initializing);
1543 spa_get_rootblkptr(spa_t *spa)
1545 return (&spa->spa_ubsync.ub_rootbp);
1549 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1551 spa->spa_uberblock.ub_rootbp = *bp;
1555 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1557 if (spa->spa_root == NULL)
1560 (void) strncpy(buf, spa->spa_root, buflen);
1564 spa_sync_pass(spa_t *spa)
1566 return (spa->spa_sync_pass);
1570 spa_name(spa_t *spa)
1572 return (spa->spa_name);
1576 spa_guid(spa_t *spa)
1578 dsl_pool_t *dp = spa_get_dsl(spa);
1582 * If we fail to parse the config during spa_load(), we can go through
1583 * the error path (which posts an ereport) and end up here with no root
1584 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1587 if (spa->spa_root_vdev == NULL)
1588 return (spa->spa_config_guid);
1590 guid = spa->spa_last_synced_guid != 0 ?
1591 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1594 * Return the most recently synced out guid unless we're
1595 * in syncing context.
1597 if (dp && dsl_pool_sync_context(dp))
1598 return (spa->spa_root_vdev->vdev_guid);
1604 spa_load_guid(spa_t *spa)
1607 * This is a GUID that exists solely as a reference for the
1608 * purposes of the arc. It is generated at load time, and
1609 * is never written to persistent storage.
1611 return (spa->spa_load_guid);
1615 spa_last_synced_txg(spa_t *spa)
1617 return (spa->spa_ubsync.ub_txg);
1621 spa_first_txg(spa_t *spa)
1623 return (spa->spa_first_txg);
1627 spa_syncing_txg(spa_t *spa)
1629 return (spa->spa_syncing_txg);
1633 spa_state(spa_t *spa)
1635 return (spa->spa_state);
1639 spa_load_state(spa_t *spa)
1641 return (spa->spa_load_state);
1645 spa_freeze_txg(spa_t *spa)
1647 return (spa->spa_freeze_txg);
1652 spa_get_asize(spa_t *spa, uint64_t lsize)
1654 return (lsize * spa_asize_inflation);
1658 * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%),
1661 * See the comment above spa_slop_shift for details.
1664 spa_get_slop_space(spa_t *spa) {
1665 uint64_t space = spa_get_dspace(spa);
1666 return (MAX(space >> spa_slop_shift, SPA_MINDEVSIZE >> 1));
1670 spa_get_dspace(spa_t *spa)
1672 return (spa->spa_dspace);
1676 spa_update_dspace(spa_t *spa)
1678 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1679 ddt_get_dedup_dspace(spa);
1683 * Return the failure mode that has been set to this pool. The default
1684 * behavior will be to block all I/Os when a complete failure occurs.
1687 spa_get_failmode(spa_t *spa)
1689 return (spa->spa_failmode);
1693 spa_suspended(spa_t *spa)
1695 return (spa->spa_suspended);
1699 spa_version(spa_t *spa)
1701 return (spa->spa_ubsync.ub_version);
1705 spa_deflate(spa_t *spa)
1707 return (spa->spa_deflate);
1711 spa_normal_class(spa_t *spa)
1713 return (spa->spa_normal_class);
1717 spa_log_class(spa_t *spa)
1719 return (spa->spa_log_class);
1723 spa_max_replication(spa_t *spa)
1726 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1727 * handle BPs with more than one DVA allocated. Set our max
1728 * replication level accordingly.
1730 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1732 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1736 spa_prev_software_version(spa_t *spa)
1738 return (spa->spa_prev_software_version);
1742 spa_deadman_synctime(spa_t *spa)
1744 return (spa->spa_deadman_synctime);
1748 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1750 uint64_t asize = DVA_GET_ASIZE(dva);
1751 uint64_t dsize = asize;
1753 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1755 if (asize != 0 && spa->spa_deflate) {
1756 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1757 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1764 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1768 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1769 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1775 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1779 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1781 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1782 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1784 spa_config_exit(spa, SCL_VDEV, FTAG);
1790 * ==========================================================================
1791 * Initialization and Termination
1792 * ==========================================================================
1796 spa_name_compare(const void *a1, const void *a2)
1798 const spa_t *s1 = a1;
1799 const spa_t *s2 = a2;
1802 s = strcmp(s1->spa_name, s2->spa_name);
1813 return (spa_active_count);
1823 EVENTHANDLER_DEFINE(mountroot, spa_boot_init, NULL, 0);
1829 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1830 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1831 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1832 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1834 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1835 offsetof(spa_t, spa_avl));
1837 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1838 offsetof(spa_aux_t, aux_avl));
1840 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1841 offsetof(spa_aux_t, aux_avl));
1843 spa_mode_global = mode;
1849 if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
1850 arc_procfd = open("/proc/self/ctl", O_WRONLY);
1851 if (arc_procfd == -1) {
1852 perror("could not enable watchpoints: "
1853 "opening /proc/self/ctl failed: ");
1859 #endif /* illumos */
1867 vdev_cache_stat_init();
1870 zpool_feature_init();
1877 #endif /* !illumos */
1887 vdev_cache_stat_fini();
1896 avl_destroy(&spa_namespace_avl);
1897 avl_destroy(&spa_spare_avl);
1898 avl_destroy(&spa_l2cache_avl);
1900 cv_destroy(&spa_namespace_cv);
1901 mutex_destroy(&spa_namespace_lock);
1902 mutex_destroy(&spa_spare_lock);
1903 mutex_destroy(&spa_l2cache_lock);
1907 * Return whether this pool has slogs. No locking needed.
1908 * It's not a problem if the wrong answer is returned as it's only for
1909 * performance and not correctness
1912 spa_has_slogs(spa_t *spa)
1914 return (spa->spa_log_class->mc_rotor != NULL);
1918 spa_get_log_state(spa_t *spa)
1920 return (spa->spa_log_state);
1924 spa_set_log_state(spa_t *spa, spa_log_state_t state)
1926 spa->spa_log_state = state;
1930 spa_is_root(spa_t *spa)
1932 return (spa->spa_is_root);
1936 spa_writeable(spa_t *spa)
1938 return (!!(spa->spa_mode & FWRITE));
1942 spa_mode(spa_t *spa)
1944 return (spa->spa_mode);
1948 spa_bootfs(spa_t *spa)
1950 return (spa->spa_bootfs);
1954 spa_delegation(spa_t *spa)
1956 return (spa->spa_delegation);
1960 spa_meta_objset(spa_t *spa)
1962 return (spa->spa_meta_objset);
1966 spa_dedup_checksum(spa_t *spa)
1968 return (spa->spa_dedup_checksum);
1972 * Reset pool scan stat per scan pass (or reboot).
1975 spa_scan_stat_init(spa_t *spa)
1977 /* data not stored on disk */
1978 spa->spa_scan_pass_start = gethrestime_sec();
1979 spa->spa_scan_pass_exam = 0;
1980 vdev_scan_stat_init(spa->spa_root_vdev);
1984 * Get scan stats for zpool status reports
1987 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
1989 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
1991 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
1992 return (SET_ERROR(ENOENT));
1993 bzero(ps, sizeof (pool_scan_stat_t));
1995 /* data stored on disk */
1996 ps->pss_func = scn->scn_phys.scn_func;
1997 ps->pss_start_time = scn->scn_phys.scn_start_time;
1998 ps->pss_end_time = scn->scn_phys.scn_end_time;
1999 ps->pss_to_examine = scn->scn_phys.scn_to_examine;
2000 ps->pss_examined = scn->scn_phys.scn_examined;
2001 ps->pss_to_process = scn->scn_phys.scn_to_process;
2002 ps->pss_processed = scn->scn_phys.scn_processed;
2003 ps->pss_errors = scn->scn_phys.scn_errors;
2004 ps->pss_state = scn->scn_phys.scn_state;
2006 /* data not stored on disk */
2007 ps->pss_pass_start = spa->spa_scan_pass_start;
2008 ps->pss_pass_exam = spa->spa_scan_pass_exam;
2014 spa_debug_enabled(spa_t *spa)
2016 return (spa->spa_debug);