4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 * Copyright 2013 Saso Kiselkov. All rights reserved.
30 #include <sys/zfs_context.h>
31 #include <sys/spa_impl.h>
32 #include <sys/spa_boot.h>
34 #include <sys/zio_checksum.h>
35 #include <sys/zio_compress.h>
37 #include <sys/dmu_tx.h>
40 #include <sys/vdev_impl.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/dsl_scan.h>
50 #include <sys/fs/zfs.h>
51 #include <sys/metaslab_impl.h>
55 #include <sys/zfeature.h>
60 * There are four basic locks for managing spa_t structures:
62 * spa_namespace_lock (global mutex)
64 * This lock must be acquired to do any of the following:
66 * - Lookup a spa_t by name
67 * - Add or remove a spa_t from the namespace
68 * - Increase spa_refcount from non-zero
69 * - Check if spa_refcount is zero
71 * - add/remove/attach/detach devices
72 * - Held for the duration of create/destroy/import/export
74 * It does not need to handle recursion. A create or destroy may
75 * reference objects (files or zvols) in other pools, but by
76 * definition they must have an existing reference, and will never need
77 * to lookup a spa_t by name.
79 * spa_refcount (per-spa refcount_t protected by mutex)
81 * This reference count keep track of any active users of the spa_t. The
82 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
83 * the refcount is never really 'zero' - opening a pool implicitly keeps
84 * some references in the DMU. Internally we check against spa_minref, but
85 * present the image of a zero/non-zero value to consumers.
87 * spa_config_lock[] (per-spa array of rwlocks)
89 * This protects the spa_t from config changes, and must be held in
90 * the following circumstances:
92 * - RW_READER to perform I/O to the spa
93 * - RW_WRITER to change the vdev config
95 * The locking order is fairly straightforward:
97 * spa_namespace_lock -> spa_refcount
99 * The namespace lock must be acquired to increase the refcount from 0
100 * or to check if it is zero.
102 * spa_refcount -> spa_config_lock[]
104 * There must be at least one valid reference on the spa_t to acquire
107 * spa_namespace_lock -> spa_config_lock[]
109 * The namespace lock must always be taken before the config lock.
112 * The spa_namespace_lock can be acquired directly and is globally visible.
114 * The namespace is manipulated using the following functions, all of which
115 * require the spa_namespace_lock to be held.
117 * spa_lookup() Lookup a spa_t by name.
119 * spa_add() Create a new spa_t in the namespace.
121 * spa_remove() Remove a spa_t from the namespace. This also
122 * frees up any memory associated with the spa_t.
124 * spa_next() Returns the next spa_t in the system, or the
125 * first if NULL is passed.
127 * spa_evict_all() Shutdown and remove all spa_t structures in
130 * spa_guid_exists() Determine whether a pool/device guid exists.
132 * The spa_refcount is manipulated using the following functions:
134 * spa_open_ref() Adds a reference to the given spa_t. Must be
135 * called with spa_namespace_lock held if the
136 * refcount is currently zero.
138 * spa_close() Remove a reference from the spa_t. This will
139 * not free the spa_t or remove it from the
140 * namespace. No locking is required.
142 * spa_refcount_zero() Returns true if the refcount is currently
143 * zero. Must be called with spa_namespace_lock
146 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
147 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
148 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
150 * To read the configuration, it suffices to hold one of these locks as reader.
151 * To modify the configuration, you must hold all locks as writer. To modify
152 * vdev state without altering the vdev tree's topology (e.g. online/offline),
153 * you must hold SCL_STATE and SCL_ZIO as writer.
155 * We use these distinct config locks to avoid recursive lock entry.
156 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
157 * block allocations (SCL_ALLOC), which may require reading space maps
158 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
160 * The spa config locks cannot be normal rwlocks because we need the
161 * ability to hand off ownership. For example, SCL_ZIO is acquired
162 * by the issuing thread and later released by an interrupt thread.
163 * They do, however, obey the usual write-wanted semantics to prevent
164 * writer (i.e. system administrator) starvation.
166 * The lock acquisition rules are as follows:
169 * Protects changes to the vdev tree topology, such as vdev
170 * add/remove/attach/detach. Protects the dirty config list
171 * (spa_config_dirty_list) and the set of spares and l2arc devices.
174 * Protects changes to pool state and vdev state, such as vdev
175 * online/offline/fault/degrade/clear. Protects the dirty state list
176 * (spa_state_dirty_list) and global pool state (spa_state).
179 * Protects changes to metaslab groups and classes.
180 * Held as reader by metaslab_alloc() and metaslab_claim().
183 * Held by bp-level zios (those which have no io_vd upon entry)
184 * to prevent changes to the vdev tree. The bp-level zio implicitly
185 * protects all of its vdev child zios, which do not hold SCL_ZIO.
188 * Protects changes to metaslab groups and classes.
189 * Held as reader by metaslab_free(). SCL_FREE is distinct from
190 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
191 * blocks in zio_done() while another i/o that holds either
192 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
195 * Held as reader to prevent changes to the vdev tree during trivial
196 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
197 * other locks, and lower than all of them, to ensure that it's safe
198 * to acquire regardless of caller context.
200 * In addition, the following rules apply:
202 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
203 * The lock ordering is SCL_CONFIG > spa_props_lock.
205 * (b) I/O operations on leaf vdevs. For any zio operation that takes
206 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
207 * or zio_write_phys() -- the caller must ensure that the config cannot
208 * cannot change in the interim, and that the vdev cannot be reopened.
209 * SCL_STATE as reader suffices for both.
211 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
213 * spa_vdev_enter() Acquire the namespace lock and the config lock
216 * spa_vdev_exit() Release the config lock, wait for all I/O
217 * to complete, sync the updated configs to the
218 * cache, and release the namespace lock.
220 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
221 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
222 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
224 * spa_rename() is also implemented within this file since it requires
225 * manipulation of the namespace.
228 static avl_tree_t spa_namespace_avl;
229 kmutex_t spa_namespace_lock;
230 static kcondvar_t spa_namespace_cv;
231 static int spa_active_count;
232 int spa_max_replication_override = SPA_DVAS_PER_BP;
234 static kmutex_t spa_spare_lock;
235 static avl_tree_t spa_spare_avl;
236 static kmutex_t spa_l2cache_lock;
237 static avl_tree_t spa_l2cache_avl;
239 kmem_cache_t *spa_buffer_pool;
243 /* Everything except dprintf and spa is on by default in debug builds */
244 int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA);
250 * zfs_recover can be set to nonzero to attempt to recover from
251 * otherwise-fatal errors, typically caused by on-disk corruption. When
252 * set, calls to zfs_panic_recover() will turn into warning messages.
253 * This should only be used as a last resort, as it typically results
254 * in leaked space, or worse.
256 boolean_t zfs_recover = B_FALSE;
257 SYSCTL_DECL(_vfs_zfs);
258 SYSCTL_INT(_vfs_zfs, OID_AUTO, recover, CTLFLAG_RDTUN, &zfs_recover, 0,
259 "Try to recover from otherwise-fatal errors.");
262 sysctl_vfs_zfs_debug_flags(SYSCTL_HANDLER_ARGS)
267 err = sysctl_handle_int(oidp, &val, 0, req);
268 if (err != 0 || req->newptr == NULL)
272 * ZFS_DEBUG_MODIFY must be enabled prior to boot so all
273 * arc buffers in the system have the necessary additional
274 * checksum data. However, it is safe to disable at any
277 if (!(zfs_flags & ZFS_DEBUG_MODIFY))
278 val &= ~ZFS_DEBUG_MODIFY;
283 SYSCTL_PROC(_vfs_zfs, OID_AUTO, debug_flags,
284 CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RWTUN, 0, sizeof(int),
285 sysctl_vfs_zfs_debug_flags, "IU", "Debug flags for ZFS testing.");
288 * If destroy encounters an EIO while reading metadata (e.g. indirect
289 * blocks), space referenced by the missing metadata can not be freed.
290 * Normally this causes the background destroy to become "stalled", as
291 * it is unable to make forward progress. While in this stalled state,
292 * all remaining space to free from the error-encountering filesystem is
293 * "temporarily leaked". Set this flag to cause it to ignore the EIO,
294 * permanently leak the space from indirect blocks that can not be read,
295 * and continue to free everything else that it can.
297 * The default, "stalling" behavior is useful if the storage partially
298 * fails (i.e. some but not all i/os fail), and then later recovers. In
299 * this case, we will be able to continue pool operations while it is
300 * partially failed, and when it recovers, we can continue to free the
301 * space, with no leaks. However, note that this case is actually
304 * Typically pools either (a) fail completely (but perhaps temporarily,
305 * e.g. a top-level vdev going offline), or (b) have localized,
306 * permanent errors (e.g. disk returns the wrong data due to bit flip or
307 * firmware bug). In case (a), this setting does not matter because the
308 * pool will be suspended and the sync thread will not be able to make
309 * forward progress regardless. In case (b), because the error is
310 * permanent, the best we can do is leak the minimum amount of space,
311 * which is what setting this flag will do. Therefore, it is reasonable
312 * for this flag to normally be set, but we chose the more conservative
313 * approach of not setting it, so that there is no possibility of
314 * leaking space in the "partial temporary" failure case.
316 boolean_t zfs_free_leak_on_eio = B_FALSE;
319 * Expiration time in milliseconds. This value has two meanings. First it is
320 * used to determine when the spa_deadman() logic should fire. By default the
321 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
322 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
323 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
326 uint64_t zfs_deadman_synctime_ms = 1000000ULL;
327 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_synctime_ms, CTLFLAG_RDTUN,
328 &zfs_deadman_synctime_ms, 0,
329 "Stalled ZFS I/O expiration time in milliseconds");
332 * Check time in milliseconds. This defines the frequency at which we check
335 uint64_t zfs_deadman_checktime_ms = 5000ULL;
336 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_checktime_ms, CTLFLAG_RDTUN,
337 &zfs_deadman_checktime_ms, 0,
338 "Period of checks for stalled ZFS I/O in milliseconds");
341 * Default value of -1 for zfs_deadman_enabled is resolved in
344 int zfs_deadman_enabled = -1;
345 SYSCTL_INT(_vfs_zfs, OID_AUTO, deadman_enabled, CTLFLAG_RDTUN,
346 &zfs_deadman_enabled, 0, "Kernel panic on stalled ZFS I/O");
349 * The worst case is single-sector max-parity RAID-Z blocks, in which
350 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
351 * times the size; so just assume that. Add to this the fact that
352 * we can have up to 3 DVAs per bp, and one more factor of 2 because
353 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together,
355 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
357 int spa_asize_inflation = 24;
358 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_asize_inflation, CTLFLAG_RWTUN,
359 &spa_asize_inflation, 0, "Worst case inflation factor for single sector writes");
367 * If we are not i386 or amd64 or in a virtual machine,
368 * disable ZFS deadman thread by default
370 if (zfs_deadman_enabled == -1) {
371 #if defined(__amd64__) || defined(__i386__)
372 zfs_deadman_enabled = (vm_guest == VM_GUEST_NO) ? 1 : 0;
374 zfs_deadman_enabled = 0;
379 #endif /* !illumos */
382 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
383 * the pool to be consumed. This ensures that we don't run the pool
384 * completely out of space, due to unaccounted changes (e.g. to the MOS).
385 * It also limits the worst-case time to allocate space. If we have
386 * less than this amount of free space, most ZPL operations (e.g. write,
387 * create) will return ENOSPC.
389 * Certain operations (e.g. file removal, most administrative actions) can
390 * use half the slop space. They will only return ENOSPC if less than half
391 * the slop space is free. Typically, once the pool has less than the slop
392 * space free, the user will use these operations to free up space in the pool.
393 * These are the operations that call dsl_pool_adjustedsize() with the netfree
394 * argument set to TRUE.
396 * A very restricted set of operations are always permitted, regardless of
397 * the amount of free space. These are the operations that call
398 * dsl_sync_task(ZFS_SPACE_CHECK_NONE), e.g. "zfs destroy". If these
399 * operations result in a net increase in the amount of space used,
400 * it is possible to run the pool completely out of space, causing it to
401 * be permanently read-only.
403 * See also the comments in zfs_space_check_t.
405 int spa_slop_shift = 5;
406 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_slop_shift, CTLFLAG_RWTUN,
408 "Shift value of reserved space (1/(2^spa_slop_shift)).");
411 * ==========================================================================
413 * ==========================================================================
416 spa_config_lock_init(spa_t *spa)
418 for (int i = 0; i < SCL_LOCKS; i++) {
419 spa_config_lock_t *scl = &spa->spa_config_lock[i];
420 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
421 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
422 refcount_create_untracked(&scl->scl_count);
423 scl->scl_writer = NULL;
424 scl->scl_write_wanted = 0;
429 spa_config_lock_destroy(spa_t *spa)
431 for (int i = 0; i < SCL_LOCKS; i++) {
432 spa_config_lock_t *scl = &spa->spa_config_lock[i];
433 mutex_destroy(&scl->scl_lock);
434 cv_destroy(&scl->scl_cv);
435 refcount_destroy(&scl->scl_count);
436 ASSERT(scl->scl_writer == NULL);
437 ASSERT(scl->scl_write_wanted == 0);
442 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
444 for (int i = 0; i < SCL_LOCKS; i++) {
445 spa_config_lock_t *scl = &spa->spa_config_lock[i];
446 if (!(locks & (1 << i)))
448 mutex_enter(&scl->scl_lock);
449 if (rw == RW_READER) {
450 if (scl->scl_writer || scl->scl_write_wanted) {
451 mutex_exit(&scl->scl_lock);
452 spa_config_exit(spa, locks ^ (1 << i), tag);
456 ASSERT(scl->scl_writer != curthread);
457 if (!refcount_is_zero(&scl->scl_count)) {
458 mutex_exit(&scl->scl_lock);
459 spa_config_exit(spa, locks ^ (1 << i), tag);
462 scl->scl_writer = curthread;
464 (void) refcount_add(&scl->scl_count, tag);
465 mutex_exit(&scl->scl_lock);
471 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
475 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
477 for (int i = 0; i < SCL_LOCKS; i++) {
478 spa_config_lock_t *scl = &spa->spa_config_lock[i];
479 if (scl->scl_writer == curthread)
480 wlocks_held |= (1 << i);
481 if (!(locks & (1 << i)))
483 mutex_enter(&scl->scl_lock);
484 if (rw == RW_READER) {
485 while (scl->scl_writer || scl->scl_write_wanted) {
486 cv_wait(&scl->scl_cv, &scl->scl_lock);
489 ASSERT(scl->scl_writer != curthread);
490 while (!refcount_is_zero(&scl->scl_count)) {
491 scl->scl_write_wanted++;
492 cv_wait(&scl->scl_cv, &scl->scl_lock);
493 scl->scl_write_wanted--;
495 scl->scl_writer = curthread;
497 (void) refcount_add(&scl->scl_count, tag);
498 mutex_exit(&scl->scl_lock);
500 ASSERT(wlocks_held <= locks);
504 spa_config_exit(spa_t *spa, int locks, void *tag)
506 for (int i = SCL_LOCKS - 1; i >= 0; i--) {
507 spa_config_lock_t *scl = &spa->spa_config_lock[i];
508 if (!(locks & (1 << i)))
510 mutex_enter(&scl->scl_lock);
511 ASSERT(!refcount_is_zero(&scl->scl_count));
512 if (refcount_remove(&scl->scl_count, tag) == 0) {
513 ASSERT(scl->scl_writer == NULL ||
514 scl->scl_writer == curthread);
515 scl->scl_writer = NULL; /* OK in either case */
516 cv_broadcast(&scl->scl_cv);
518 mutex_exit(&scl->scl_lock);
523 spa_config_held(spa_t *spa, int locks, krw_t rw)
527 for (int i = 0; i < SCL_LOCKS; i++) {
528 spa_config_lock_t *scl = &spa->spa_config_lock[i];
529 if (!(locks & (1 << i)))
531 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
532 (rw == RW_WRITER && scl->scl_writer == curthread))
533 locks_held |= 1 << i;
540 * ==========================================================================
541 * SPA namespace functions
542 * ==========================================================================
546 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
547 * Returns NULL if no matching spa_t is found.
550 spa_lookup(const char *name)
552 static spa_t search; /* spa_t is large; don't allocate on stack */
557 ASSERT(MUTEX_HELD(&spa_namespace_lock));
559 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
562 * If it's a full dataset name, figure out the pool name and
565 cp = strpbrk(search.spa_name, "/@#");
569 spa = avl_find(&spa_namespace_avl, &search, &where);
575 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
576 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
577 * looking for potentially hung I/Os.
580 spa_deadman(void *arg)
585 * Disable the deadman timer if the pool is suspended.
587 if (spa_suspended(spa)) {
589 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
591 /* Nothing. just don't schedule any future callouts. */
596 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
597 (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
598 ++spa->spa_deadman_calls);
599 if (zfs_deadman_enabled)
600 vdev_deadman(spa->spa_root_vdev);
603 callout_schedule(&spa->spa_deadman_cycid,
604 hz * zfs_deadman_checktime_ms / MILLISEC);
610 * Create an uninitialized spa_t with the given name. Requires
611 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
612 * exist by calling spa_lookup() first.
615 spa_add(const char *name, nvlist_t *config, const char *altroot)
618 spa_config_dirent_t *dp;
624 ASSERT(MUTEX_HELD(&spa_namespace_lock));
626 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
628 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
629 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
630 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
631 mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL);
632 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
633 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
634 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
635 mutex_init(&spa->spa_cksum_tmpls_lock, NULL, MUTEX_DEFAULT, NULL);
636 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
637 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
638 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
640 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
641 cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL);
642 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
643 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
644 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
646 for (int t = 0; t < TXG_SIZE; t++)
647 bplist_create(&spa->spa_free_bplist[t]);
649 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
650 spa->spa_state = POOL_STATE_UNINITIALIZED;
651 spa->spa_freeze_txg = UINT64_MAX;
652 spa->spa_final_txg = UINT64_MAX;
653 spa->spa_load_max_txg = UINT64_MAX;
655 spa->spa_proc_state = SPA_PROC_NONE;
658 hdlr.cyh_func = spa_deadman;
660 hdlr.cyh_level = CY_LOW_LEVEL;
663 spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
667 * This determines how often we need to check for hung I/Os after
668 * the cyclic has already fired. Since checking for hung I/Os is
669 * an expensive operation we don't want to check too frequently.
670 * Instead wait for 5 seconds before checking again.
672 when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms);
673 when.cyt_when = CY_INFINITY;
674 mutex_enter(&cpu_lock);
675 spa->spa_deadman_cycid = cyclic_add(&hdlr, &when);
676 mutex_exit(&cpu_lock);
679 callout_init(&spa->spa_deadman_cycid, 1);
682 refcount_create(&spa->spa_refcount);
683 spa_config_lock_init(spa);
685 avl_add(&spa_namespace_avl, spa);
688 * Set the alternate root, if there is one.
691 spa->spa_root = spa_strdup(altroot);
696 * Every pool starts with the default cachefile
698 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
699 offsetof(spa_config_dirent_t, scd_link));
701 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
702 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
703 list_insert_head(&spa->spa_config_list, dp);
705 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
708 if (config != NULL) {
711 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
713 VERIFY(nvlist_dup(features, &spa->spa_label_features,
717 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
720 if (spa->spa_label_features == NULL) {
721 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
725 spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0);
727 spa->spa_min_ashift = INT_MAX;
728 spa->spa_max_ashift = 0;
731 * As a pool is being created, treat all features as disabled by
732 * setting SPA_FEATURE_DISABLED for all entries in the feature
735 for (int i = 0; i < SPA_FEATURES; i++) {
736 spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED;
743 * Removes a spa_t from the namespace, freeing up any memory used. Requires
744 * spa_namespace_lock. This is called only after the spa_t has been closed and
748 spa_remove(spa_t *spa)
750 spa_config_dirent_t *dp;
752 ASSERT(MUTEX_HELD(&spa_namespace_lock));
753 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
754 ASSERT3U(refcount_count(&spa->spa_refcount), ==, 0);
756 nvlist_free(spa->spa_config_splitting);
758 avl_remove(&spa_namespace_avl, spa);
759 cv_broadcast(&spa_namespace_cv);
762 spa_strfree(spa->spa_root);
766 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
767 list_remove(&spa->spa_config_list, dp);
768 if (dp->scd_path != NULL)
769 spa_strfree(dp->scd_path);
770 kmem_free(dp, sizeof (spa_config_dirent_t));
773 list_destroy(&spa->spa_config_list);
775 nvlist_free(spa->spa_label_features);
776 nvlist_free(spa->spa_load_info);
777 spa_config_set(spa, NULL);
780 mutex_enter(&cpu_lock);
781 if (spa->spa_deadman_cycid != CYCLIC_NONE)
782 cyclic_remove(spa->spa_deadman_cycid);
783 mutex_exit(&cpu_lock);
784 spa->spa_deadman_cycid = CYCLIC_NONE;
787 callout_drain(&spa->spa_deadman_cycid);
791 refcount_destroy(&spa->spa_refcount);
793 spa_config_lock_destroy(spa);
795 for (int t = 0; t < TXG_SIZE; t++)
796 bplist_destroy(&spa->spa_free_bplist[t]);
798 zio_checksum_templates_free(spa);
800 cv_destroy(&spa->spa_async_cv);
801 cv_destroy(&spa->spa_evicting_os_cv);
802 cv_destroy(&spa->spa_proc_cv);
803 cv_destroy(&spa->spa_scrub_io_cv);
804 cv_destroy(&spa->spa_suspend_cv);
806 mutex_destroy(&spa->spa_async_lock);
807 mutex_destroy(&spa->spa_errlist_lock);
808 mutex_destroy(&spa->spa_errlog_lock);
809 mutex_destroy(&spa->spa_evicting_os_lock);
810 mutex_destroy(&spa->spa_history_lock);
811 mutex_destroy(&spa->spa_proc_lock);
812 mutex_destroy(&spa->spa_props_lock);
813 mutex_destroy(&spa->spa_cksum_tmpls_lock);
814 mutex_destroy(&spa->spa_scrub_lock);
815 mutex_destroy(&spa->spa_suspend_lock);
816 mutex_destroy(&spa->spa_vdev_top_lock);
818 kmem_free(spa, sizeof (spa_t));
822 * Given a pool, return the next pool in the namespace, or NULL if there is
823 * none. If 'prev' is NULL, return the first pool.
826 spa_next(spa_t *prev)
828 ASSERT(MUTEX_HELD(&spa_namespace_lock));
831 return (AVL_NEXT(&spa_namespace_avl, prev));
833 return (avl_first(&spa_namespace_avl));
837 * ==========================================================================
838 * SPA refcount functions
839 * ==========================================================================
843 * Add a reference to the given spa_t. Must have at least one reference, or
844 * have the namespace lock held.
847 spa_open_ref(spa_t *spa, void *tag)
849 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
850 MUTEX_HELD(&spa_namespace_lock));
851 (void) refcount_add(&spa->spa_refcount, tag);
855 * Remove a reference to the given spa_t. Must have at least one reference, or
856 * have the namespace lock held.
859 spa_close(spa_t *spa, void *tag)
861 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
862 MUTEX_HELD(&spa_namespace_lock));
863 (void) refcount_remove(&spa->spa_refcount, tag);
867 * Remove a reference to the given spa_t held by a dsl dir that is
868 * being asynchronously released. Async releases occur from a taskq
869 * performing eviction of dsl datasets and dirs. The namespace lock
870 * isn't held and the hold by the object being evicted may contribute to
871 * spa_minref (e.g. dataset or directory released during pool export),
872 * so the asserts in spa_close() do not apply.
875 spa_async_close(spa_t *spa, void *tag)
877 (void) refcount_remove(&spa->spa_refcount, tag);
881 * Check to see if the spa refcount is zero. Must be called with
882 * spa_namespace_lock held. We really compare against spa_minref, which is the
883 * number of references acquired when opening a pool
886 spa_refcount_zero(spa_t *spa)
888 ASSERT(MUTEX_HELD(&spa_namespace_lock));
890 return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
894 * ==========================================================================
895 * SPA spare and l2cache tracking
896 * ==========================================================================
900 * Hot spares and cache devices are tracked using the same code below,
901 * for 'auxiliary' devices.
904 typedef struct spa_aux {
912 spa_aux_compare(const void *a, const void *b)
914 const spa_aux_t *sa = a;
915 const spa_aux_t *sb = b;
917 if (sa->aux_guid < sb->aux_guid)
919 else if (sa->aux_guid > sb->aux_guid)
926 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
932 search.aux_guid = vd->vdev_guid;
933 if ((aux = avl_find(avl, &search, &where)) != NULL) {
936 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
937 aux->aux_guid = vd->vdev_guid;
939 avl_insert(avl, aux, where);
944 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
950 search.aux_guid = vd->vdev_guid;
951 aux = avl_find(avl, &search, &where);
955 if (--aux->aux_count == 0) {
956 avl_remove(avl, aux);
957 kmem_free(aux, sizeof (spa_aux_t));
958 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
959 aux->aux_pool = 0ULL;
964 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
966 spa_aux_t search, *found;
968 search.aux_guid = guid;
969 found = avl_find(avl, &search, NULL);
973 *pool = found->aux_pool;
980 *refcnt = found->aux_count;
985 return (found != NULL);
989 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
991 spa_aux_t search, *found;
994 search.aux_guid = vd->vdev_guid;
995 found = avl_find(avl, &search, &where);
996 ASSERT(found != NULL);
997 ASSERT(found->aux_pool == 0ULL);
999 found->aux_pool = spa_guid(vd->vdev_spa);
1003 * Spares are tracked globally due to the following constraints:
1005 * - A spare may be part of multiple pools.
1006 * - A spare may be added to a pool even if it's actively in use within
1008 * - A spare in use in any pool can only be the source of a replacement if
1009 * the target is a spare in the same pool.
1011 * We keep track of all spares on the system through the use of a reference
1012 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
1013 * spare, then we bump the reference count in the AVL tree. In addition, we set
1014 * the 'vdev_isspare' member to indicate that the device is a spare (active or
1015 * inactive). When a spare is made active (used to replace a device in the
1016 * pool), we also keep track of which pool its been made a part of.
1018 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
1019 * called under the spa_namespace lock as part of vdev reconfiguration. The
1020 * separate spare lock exists for the status query path, which does not need to
1021 * be completely consistent with respect to other vdev configuration changes.
1025 spa_spare_compare(const void *a, const void *b)
1027 return (spa_aux_compare(a, b));
1031 spa_spare_add(vdev_t *vd)
1033 mutex_enter(&spa_spare_lock);
1034 ASSERT(!vd->vdev_isspare);
1035 spa_aux_add(vd, &spa_spare_avl);
1036 vd->vdev_isspare = B_TRUE;
1037 mutex_exit(&spa_spare_lock);
1041 spa_spare_remove(vdev_t *vd)
1043 mutex_enter(&spa_spare_lock);
1044 ASSERT(vd->vdev_isspare);
1045 spa_aux_remove(vd, &spa_spare_avl);
1046 vd->vdev_isspare = B_FALSE;
1047 mutex_exit(&spa_spare_lock);
1051 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
1055 mutex_enter(&spa_spare_lock);
1056 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
1057 mutex_exit(&spa_spare_lock);
1063 spa_spare_activate(vdev_t *vd)
1065 mutex_enter(&spa_spare_lock);
1066 ASSERT(vd->vdev_isspare);
1067 spa_aux_activate(vd, &spa_spare_avl);
1068 mutex_exit(&spa_spare_lock);
1072 * Level 2 ARC devices are tracked globally for the same reasons as spares.
1073 * Cache devices currently only support one pool per cache device, and so
1074 * for these devices the aux reference count is currently unused beyond 1.
1078 spa_l2cache_compare(const void *a, const void *b)
1080 return (spa_aux_compare(a, b));
1084 spa_l2cache_add(vdev_t *vd)
1086 mutex_enter(&spa_l2cache_lock);
1087 ASSERT(!vd->vdev_isl2cache);
1088 spa_aux_add(vd, &spa_l2cache_avl);
1089 vd->vdev_isl2cache = B_TRUE;
1090 mutex_exit(&spa_l2cache_lock);
1094 spa_l2cache_remove(vdev_t *vd)
1096 mutex_enter(&spa_l2cache_lock);
1097 ASSERT(vd->vdev_isl2cache);
1098 spa_aux_remove(vd, &spa_l2cache_avl);
1099 vd->vdev_isl2cache = B_FALSE;
1100 mutex_exit(&spa_l2cache_lock);
1104 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
1108 mutex_enter(&spa_l2cache_lock);
1109 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
1110 mutex_exit(&spa_l2cache_lock);
1116 spa_l2cache_activate(vdev_t *vd)
1118 mutex_enter(&spa_l2cache_lock);
1119 ASSERT(vd->vdev_isl2cache);
1120 spa_aux_activate(vd, &spa_l2cache_avl);
1121 mutex_exit(&spa_l2cache_lock);
1125 * ==========================================================================
1127 * ==========================================================================
1131 * Lock the given spa_t for the purpose of adding or removing a vdev.
1132 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1133 * It returns the next transaction group for the spa_t.
1136 spa_vdev_enter(spa_t *spa)
1138 mutex_enter(&spa->spa_vdev_top_lock);
1139 mutex_enter(&spa_namespace_lock);
1140 return (spa_vdev_config_enter(spa));
1144 * Internal implementation for spa_vdev_enter(). Used when a vdev
1145 * operation requires multiple syncs (i.e. removing a device) while
1146 * keeping the spa_namespace_lock held.
1149 spa_vdev_config_enter(spa_t *spa)
1151 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1153 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1155 return (spa_last_synced_txg(spa) + 1);
1159 * Used in combination with spa_vdev_config_enter() to allow the syncing
1160 * of multiple transactions without releasing the spa_namespace_lock.
1163 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
1165 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1167 int config_changed = B_FALSE;
1169 ASSERT(txg > spa_last_synced_txg(spa));
1171 spa->spa_pending_vdev = NULL;
1174 * Reassess the DTLs.
1176 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
1178 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
1179 config_changed = B_TRUE;
1180 spa->spa_config_generation++;
1184 * Verify the metaslab classes.
1186 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
1187 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
1189 spa_config_exit(spa, SCL_ALL, spa);
1192 * Panic the system if the specified tag requires it. This
1193 * is useful for ensuring that configurations are updated
1196 if (zio_injection_enabled)
1197 zio_handle_panic_injection(spa, tag, 0);
1200 * Note: this txg_wait_synced() is important because it ensures
1201 * that there won't be more than one config change per txg.
1202 * This allows us to use the txg as the generation number.
1205 txg_wait_synced(spa->spa_dsl_pool, txg);
1208 ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
1209 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1211 spa_config_exit(spa, SCL_ALL, spa);
1215 * If the config changed, update the config cache.
1218 spa_config_sync(spa, B_FALSE, B_TRUE);
1222 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1223 * locking of spa_vdev_enter(), we also want make sure the transactions have
1224 * synced to disk, and then update the global configuration cache with the new
1228 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1230 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1231 mutex_exit(&spa_namespace_lock);
1232 mutex_exit(&spa->spa_vdev_top_lock);
1238 * Lock the given spa_t for the purpose of changing vdev state.
1241 spa_vdev_state_enter(spa_t *spa, int oplocks)
1243 int locks = SCL_STATE_ALL | oplocks;
1246 * Root pools may need to read of the underlying devfs filesystem
1247 * when opening up a vdev. Unfortunately if we're holding the
1248 * SCL_ZIO lock it will result in a deadlock when we try to issue
1249 * the read from the root filesystem. Instead we "prefetch"
1250 * the associated vnodes that we need prior to opening the
1251 * underlying devices and cache them so that we can prevent
1252 * any I/O when we are doing the actual open.
1254 if (spa_is_root(spa)) {
1255 int low = locks & ~(SCL_ZIO - 1);
1256 int high = locks & ~low;
1258 spa_config_enter(spa, high, spa, RW_WRITER);
1259 vdev_hold(spa->spa_root_vdev);
1260 spa_config_enter(spa, low, spa, RW_WRITER);
1262 spa_config_enter(spa, locks, spa, RW_WRITER);
1264 spa->spa_vdev_locks = locks;
1268 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1270 boolean_t config_changed = B_FALSE;
1272 if (vd != NULL || error == 0)
1273 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1277 vdev_state_dirty(vd->vdev_top);
1278 config_changed = B_TRUE;
1279 spa->spa_config_generation++;
1282 if (spa_is_root(spa))
1283 vdev_rele(spa->spa_root_vdev);
1285 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1286 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1289 * If anything changed, wait for it to sync. This ensures that,
1290 * from the system administrator's perspective, zpool(1M) commands
1291 * are synchronous. This is important for things like zpool offline:
1292 * when the command completes, you expect no further I/O from ZFS.
1295 txg_wait_synced(spa->spa_dsl_pool, 0);
1298 * If the config changed, update the config cache.
1300 if (config_changed) {
1301 mutex_enter(&spa_namespace_lock);
1302 spa_config_sync(spa, B_FALSE, B_TRUE);
1303 mutex_exit(&spa_namespace_lock);
1310 * ==========================================================================
1311 * Miscellaneous functions
1312 * ==========================================================================
1316 spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx)
1318 if (!nvlist_exists(spa->spa_label_features, feature)) {
1319 fnvlist_add_boolean(spa->spa_label_features, feature);
1321 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1322 * dirty the vdev config because lock SCL_CONFIG is not held.
1323 * Thankfully, in this case we don't need to dirty the config
1324 * because it will be written out anyway when we finish
1325 * creating the pool.
1327 if (tx->tx_txg != TXG_INITIAL)
1328 vdev_config_dirty(spa->spa_root_vdev);
1333 spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1335 if (nvlist_remove_all(spa->spa_label_features, feature) == 0)
1336 vdev_config_dirty(spa->spa_root_vdev);
1343 spa_rename(const char *name, const char *newname)
1349 * Lookup the spa_t and grab the config lock for writing. We need to
1350 * actually open the pool so that we can sync out the necessary labels.
1351 * It's OK to call spa_open() with the namespace lock held because we
1352 * allow recursive calls for other reasons.
1354 mutex_enter(&spa_namespace_lock);
1355 if ((err = spa_open(name, &spa, FTAG)) != 0) {
1356 mutex_exit(&spa_namespace_lock);
1360 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1362 avl_remove(&spa_namespace_avl, spa);
1363 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1364 avl_add(&spa_namespace_avl, spa);
1367 * Sync all labels to disk with the new names by marking the root vdev
1368 * dirty and waiting for it to sync. It will pick up the new pool name
1371 vdev_config_dirty(spa->spa_root_vdev);
1373 spa_config_exit(spa, SCL_ALL, FTAG);
1375 txg_wait_synced(spa->spa_dsl_pool, 0);
1378 * Sync the updated config cache.
1380 spa_config_sync(spa, B_FALSE, B_TRUE);
1382 spa_close(spa, FTAG);
1384 mutex_exit(&spa_namespace_lock);
1390 * Return the spa_t associated with given pool_guid, if it exists. If
1391 * device_guid is non-zero, determine whether the pool exists *and* contains
1392 * a device with the specified device_guid.
1395 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1398 avl_tree_t *t = &spa_namespace_avl;
1400 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1402 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1403 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1405 if (spa->spa_root_vdev == NULL)
1407 if (spa_guid(spa) == pool_guid) {
1408 if (device_guid == 0)
1411 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1412 device_guid) != NULL)
1416 * Check any devices we may be in the process of adding.
1418 if (spa->spa_pending_vdev) {
1419 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1420 device_guid) != NULL)
1430 * Determine whether a pool with the given pool_guid exists.
1433 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1435 return (spa_by_guid(pool_guid, device_guid) != NULL);
1439 spa_strdup(const char *s)
1445 new = kmem_alloc(len + 1, KM_SLEEP);
1453 spa_strfree(char *s)
1455 kmem_free(s, strlen(s) + 1);
1459 spa_get_random(uint64_t range)
1465 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1471 spa_generate_guid(spa_t *spa)
1473 uint64_t guid = spa_get_random(-1ULL);
1476 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1477 guid = spa_get_random(-1ULL);
1479 while (guid == 0 || spa_guid_exists(guid, 0))
1480 guid = spa_get_random(-1ULL);
1487 snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp)
1490 char *checksum = NULL;
1491 char *compress = NULL;
1494 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1495 dmu_object_byteswap_t bswap =
1496 DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1497 (void) snprintf(type, sizeof (type), "bswap %s %s",
1498 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1499 "metadata" : "data",
1500 dmu_ot_byteswap[bswap].ob_name);
1502 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1505 if (!BP_IS_EMBEDDED(bp)) {
1507 zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1509 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1512 SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum,
1517 spa_freeze(spa_t *spa)
1519 uint64_t freeze_txg = 0;
1521 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1522 if (spa->spa_freeze_txg == UINT64_MAX) {
1523 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1524 spa->spa_freeze_txg = freeze_txg;
1526 spa_config_exit(spa, SCL_ALL, FTAG);
1527 if (freeze_txg != 0)
1528 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1532 zfs_panic_recover(const char *fmt, ...)
1537 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1542 * This is a stripped-down version of strtoull, suitable only for converting
1543 * lowercase hexadecimal numbers that don't overflow.
1546 zfs_strtonum(const char *str, char **nptr)
1552 while ((c = *str) != '\0') {
1553 if (c >= '0' && c <= '9')
1555 else if (c >= 'a' && c <= 'f')
1556 digit = 10 + c - 'a';
1567 *nptr = (char *)str;
1573 * ==========================================================================
1574 * Accessor functions
1575 * ==========================================================================
1579 spa_shutting_down(spa_t *spa)
1581 return (spa->spa_async_suspended);
1585 spa_get_dsl(spa_t *spa)
1587 return (spa->spa_dsl_pool);
1591 spa_is_initializing(spa_t *spa)
1593 return (spa->spa_is_initializing);
1597 spa_get_rootblkptr(spa_t *spa)
1599 return (&spa->spa_ubsync.ub_rootbp);
1603 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1605 spa->spa_uberblock.ub_rootbp = *bp;
1609 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1611 if (spa->spa_root == NULL)
1614 (void) strncpy(buf, spa->spa_root, buflen);
1618 spa_sync_pass(spa_t *spa)
1620 return (spa->spa_sync_pass);
1624 spa_name(spa_t *spa)
1626 return (spa->spa_name);
1630 spa_guid(spa_t *spa)
1632 dsl_pool_t *dp = spa_get_dsl(spa);
1636 * If we fail to parse the config during spa_load(), we can go through
1637 * the error path (which posts an ereport) and end up here with no root
1638 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1641 if (spa->spa_root_vdev == NULL)
1642 return (spa->spa_config_guid);
1644 guid = spa->spa_last_synced_guid != 0 ?
1645 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1648 * Return the most recently synced out guid unless we're
1649 * in syncing context.
1651 if (dp && dsl_pool_sync_context(dp))
1652 return (spa->spa_root_vdev->vdev_guid);
1658 spa_load_guid(spa_t *spa)
1661 * This is a GUID that exists solely as a reference for the
1662 * purposes of the arc. It is generated at load time, and
1663 * is never written to persistent storage.
1665 return (spa->spa_load_guid);
1669 spa_last_synced_txg(spa_t *spa)
1671 return (spa->spa_ubsync.ub_txg);
1675 spa_first_txg(spa_t *spa)
1677 return (spa->spa_first_txg);
1681 spa_syncing_txg(spa_t *spa)
1683 return (spa->spa_syncing_txg);
1687 spa_state(spa_t *spa)
1689 return (spa->spa_state);
1693 spa_load_state(spa_t *spa)
1695 return (spa->spa_load_state);
1699 spa_freeze_txg(spa_t *spa)
1701 return (spa->spa_freeze_txg);
1706 spa_get_asize(spa_t *spa, uint64_t lsize)
1708 return (lsize * spa_asize_inflation);
1712 * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%),
1715 * See the comment above spa_slop_shift for details.
1718 spa_get_slop_space(spa_t *spa) {
1719 uint64_t space = spa_get_dspace(spa);
1720 return (MAX(space >> spa_slop_shift, SPA_MINDEVSIZE >> 1));
1724 spa_get_dspace(spa_t *spa)
1726 return (spa->spa_dspace);
1730 spa_update_dspace(spa_t *spa)
1732 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1733 ddt_get_dedup_dspace(spa);
1737 * Return the failure mode that has been set to this pool. The default
1738 * behavior will be to block all I/Os when a complete failure occurs.
1741 spa_get_failmode(spa_t *spa)
1743 return (spa->spa_failmode);
1747 spa_suspended(spa_t *spa)
1749 return (spa->spa_suspended);
1753 spa_version(spa_t *spa)
1755 return (spa->spa_ubsync.ub_version);
1759 spa_deflate(spa_t *spa)
1761 return (spa->spa_deflate);
1765 spa_normal_class(spa_t *spa)
1767 return (spa->spa_normal_class);
1771 spa_log_class(spa_t *spa)
1773 return (spa->spa_log_class);
1777 spa_evicting_os_register(spa_t *spa, objset_t *os)
1779 mutex_enter(&spa->spa_evicting_os_lock);
1780 list_insert_head(&spa->spa_evicting_os_list, os);
1781 mutex_exit(&spa->spa_evicting_os_lock);
1785 spa_evicting_os_deregister(spa_t *spa, objset_t *os)
1787 mutex_enter(&spa->spa_evicting_os_lock);
1788 list_remove(&spa->spa_evicting_os_list, os);
1789 cv_broadcast(&spa->spa_evicting_os_cv);
1790 mutex_exit(&spa->spa_evicting_os_lock);
1794 spa_evicting_os_wait(spa_t *spa)
1796 mutex_enter(&spa->spa_evicting_os_lock);
1797 while (!list_is_empty(&spa->spa_evicting_os_list))
1798 cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock);
1799 mutex_exit(&spa->spa_evicting_os_lock);
1801 dmu_buf_user_evict_wait();
1805 spa_max_replication(spa_t *spa)
1808 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1809 * handle BPs with more than one DVA allocated. Set our max
1810 * replication level accordingly.
1812 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1814 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1818 spa_prev_software_version(spa_t *spa)
1820 return (spa->spa_prev_software_version);
1824 spa_deadman_synctime(spa_t *spa)
1826 return (spa->spa_deadman_synctime);
1830 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1832 uint64_t asize = DVA_GET_ASIZE(dva);
1833 uint64_t dsize = asize;
1835 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1837 if (asize != 0 && spa->spa_deflate) {
1838 vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1839 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1846 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1850 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1851 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1857 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1861 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1863 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1864 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1866 spa_config_exit(spa, SCL_VDEV, FTAG);
1872 * ==========================================================================
1873 * Initialization and Termination
1874 * ==========================================================================
1878 spa_name_compare(const void *a1, const void *a2)
1880 const spa_t *s1 = a1;
1881 const spa_t *s2 = a2;
1884 s = strcmp(s1->spa_name, s2->spa_name);
1895 return (spa_active_count);
1905 EVENTHANDLER_DEFINE(mountroot, spa_boot_init, NULL, 0);
1911 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1912 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1913 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1914 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1916 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1917 offsetof(spa_t, spa_avl));
1919 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1920 offsetof(spa_aux_t, aux_avl));
1922 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1923 offsetof(spa_aux_t, aux_avl));
1925 spa_mode_global = mode;
1931 if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
1932 arc_procfd = open("/proc/self/ctl", O_WRONLY);
1933 if (arc_procfd == -1) {
1934 perror("could not enable watchpoints: "
1935 "opening /proc/self/ctl failed: ");
1941 #endif /* illumos */
1949 vdev_cache_stat_init();
1952 zpool_feature_init();
1959 #endif /* !illumos */
1969 vdev_cache_stat_fini();
1978 avl_destroy(&spa_namespace_avl);
1979 avl_destroy(&spa_spare_avl);
1980 avl_destroy(&spa_l2cache_avl);
1982 cv_destroy(&spa_namespace_cv);
1983 mutex_destroy(&spa_namespace_lock);
1984 mutex_destroy(&spa_spare_lock);
1985 mutex_destroy(&spa_l2cache_lock);
1989 * Return whether this pool has slogs. No locking needed.
1990 * It's not a problem if the wrong answer is returned as it's only for
1991 * performance and not correctness
1994 spa_has_slogs(spa_t *spa)
1996 return (spa->spa_log_class->mc_rotor != NULL);
2000 spa_get_log_state(spa_t *spa)
2002 return (spa->spa_log_state);
2006 spa_set_log_state(spa_t *spa, spa_log_state_t state)
2008 spa->spa_log_state = state;
2012 spa_is_root(spa_t *spa)
2014 return (spa->spa_is_root);
2018 spa_writeable(spa_t *spa)
2020 return (!!(spa->spa_mode & FWRITE));
2024 * Returns true if there is a pending sync task in any of the current
2025 * syncing txg, the current quiescing txg, or the current open txg.
2028 spa_has_pending_synctask(spa_t *spa)
2030 return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks));
2034 spa_mode(spa_t *spa)
2036 return (spa->spa_mode);
2040 spa_bootfs(spa_t *spa)
2042 return (spa->spa_bootfs);
2046 spa_delegation(spa_t *spa)
2048 return (spa->spa_delegation);
2052 spa_meta_objset(spa_t *spa)
2054 return (spa->spa_meta_objset);
2058 spa_dedup_checksum(spa_t *spa)
2060 return (spa->spa_dedup_checksum);
2064 * Reset pool scan stat per scan pass (or reboot).
2067 spa_scan_stat_init(spa_t *spa)
2069 /* data not stored on disk */
2070 spa->spa_scan_pass_start = gethrestime_sec();
2071 spa->spa_scan_pass_exam = 0;
2072 vdev_scan_stat_init(spa->spa_root_vdev);
2076 * Get scan stats for zpool status reports
2079 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
2081 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
2083 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
2084 return (SET_ERROR(ENOENT));
2085 bzero(ps, sizeof (pool_scan_stat_t));
2087 /* data stored on disk */
2088 ps->pss_func = scn->scn_phys.scn_func;
2089 ps->pss_start_time = scn->scn_phys.scn_start_time;
2090 ps->pss_end_time = scn->scn_phys.scn_end_time;
2091 ps->pss_to_examine = scn->scn_phys.scn_to_examine;
2092 ps->pss_examined = scn->scn_phys.scn_examined;
2093 ps->pss_to_process = scn->scn_phys.scn_to_process;
2094 ps->pss_processed = scn->scn_phys.scn_processed;
2095 ps->pss_errors = scn->scn_phys.scn_errors;
2096 ps->pss_state = scn->scn_phys.scn_state;
2098 /* data not stored on disk */
2099 ps->pss_pass_start = spa->spa_scan_pass_start;
2100 ps->pss_pass_exam = spa->spa_scan_pass_exam;
2106 spa_debug_enabled(spa_t *spa)
2108 return (spa->spa_debug);
2112 spa_maxblocksize(spa_t *spa)
2114 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS))
2115 return (SPA_MAXBLOCKSIZE);
2117 return (SPA_OLD_MAXBLOCKSIZE);