4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
24 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
25 * Copyright 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.
28 * Copyright (c) 2014 Integros [integros.com]
31 #include <sys/zfs_context.h>
32 #include <sys/spa_impl.h>
33 #include <sys/spa_boot.h>
35 #include <sys/zio_checksum.h>
36 #include <sys/zio_compress.h>
38 #include <sys/dmu_tx.h>
41 #include <sys/vdev_impl.h>
42 #include <sys/vdev_file.h>
43 #include <sys/metaslab.h>
44 #include <sys/uberblock_impl.h>
47 #include <sys/unique.h>
48 #include <sys/dsl_pool.h>
49 #include <sys/dsl_dir.h>
50 #include <sys/dsl_prop.h>
51 #include <sys/dsl_scan.h>
52 #include <sys/fs/zfs.h>
53 #include <sys/metaslab_impl.h>
57 #include <sys/zfeature.h>
59 #if defined(__FreeBSD__) && defined(_KERNEL)
60 #include <sys/types.h>
61 #include <sys/sysctl.h>
67 * There are four basic locks for managing spa_t structures:
69 * spa_namespace_lock (global mutex)
71 * This lock must be acquired to do any of the following:
73 * - Lookup a spa_t by name
74 * - Add or remove a spa_t from the namespace
75 * - Increase spa_refcount from non-zero
76 * - Check if spa_refcount is zero
78 * - add/remove/attach/detach devices
79 * - Held for the duration of create/destroy/import/export
81 * It does not need to handle recursion. A create or destroy may
82 * reference objects (files or zvols) in other pools, but by
83 * definition they must have an existing reference, and will never need
84 * to lookup a spa_t by name.
86 * spa_refcount (per-spa refcount_t protected by mutex)
88 * This reference count keep track of any active users of the spa_t. The
89 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
90 * the refcount is never really 'zero' - opening a pool implicitly keeps
91 * some references in the DMU. Internally we check against spa_minref, but
92 * present the image of a zero/non-zero value to consumers.
94 * spa_config_lock[] (per-spa array of rwlocks)
96 * This protects the spa_t from config changes, and must be held in
97 * the following circumstances:
99 * - RW_READER to perform I/O to the spa
100 * - RW_WRITER to change the vdev config
102 * The locking order is fairly straightforward:
104 * spa_namespace_lock -> spa_refcount
106 * The namespace lock must be acquired to increase the refcount from 0
107 * or to check if it is zero.
109 * spa_refcount -> spa_config_lock[]
111 * There must be at least one valid reference on the spa_t to acquire
114 * spa_namespace_lock -> spa_config_lock[]
116 * The namespace lock must always be taken before the config lock.
119 * The spa_namespace_lock can be acquired directly and is globally visible.
121 * The namespace is manipulated using the following functions, all of which
122 * require the spa_namespace_lock to be held.
124 * spa_lookup() Lookup a spa_t by name.
126 * spa_add() Create a new spa_t in the namespace.
128 * spa_remove() Remove a spa_t from the namespace. This also
129 * frees up any memory associated with the spa_t.
131 * spa_next() Returns the next spa_t in the system, or the
132 * first if NULL is passed.
134 * spa_evict_all() Shutdown and remove all spa_t structures in
137 * spa_guid_exists() Determine whether a pool/device guid exists.
139 * The spa_refcount is manipulated using the following functions:
141 * spa_open_ref() Adds a reference to the given spa_t. Must be
142 * called with spa_namespace_lock held if the
143 * refcount is currently zero.
145 * spa_close() Remove a reference from the spa_t. This will
146 * not free the spa_t or remove it from the
147 * namespace. No locking is required.
149 * spa_refcount_zero() Returns true if the refcount is currently
150 * zero. Must be called with spa_namespace_lock
153 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
154 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
155 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
157 * To read the configuration, it suffices to hold one of these locks as reader.
158 * To modify the configuration, you must hold all locks as writer. To modify
159 * vdev state without altering the vdev tree's topology (e.g. online/offline),
160 * you must hold SCL_STATE and SCL_ZIO as writer.
162 * We use these distinct config locks to avoid recursive lock entry.
163 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
164 * block allocations (SCL_ALLOC), which may require reading space maps
165 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
167 * The spa config locks cannot be normal rwlocks because we need the
168 * ability to hand off ownership. For example, SCL_ZIO is acquired
169 * by the issuing thread and later released by an interrupt thread.
170 * They do, however, obey the usual write-wanted semantics to prevent
171 * writer (i.e. system administrator) starvation.
173 * The lock acquisition rules are as follows:
176 * Protects changes to the vdev tree topology, such as vdev
177 * add/remove/attach/detach. Protects the dirty config list
178 * (spa_config_dirty_list) and the set of spares and l2arc devices.
181 * Protects changes to pool state and vdev state, such as vdev
182 * online/offline/fault/degrade/clear. Protects the dirty state list
183 * (spa_state_dirty_list) and global pool state (spa_state).
186 * Protects changes to metaslab groups and classes.
187 * Held as reader by metaslab_alloc() and metaslab_claim().
190 * Held by bp-level zios (those which have no io_vd upon entry)
191 * to prevent changes to the vdev tree. The bp-level zio implicitly
192 * protects all of its vdev child zios, which do not hold SCL_ZIO.
195 * Protects changes to metaslab groups and classes.
196 * Held as reader by metaslab_free(). SCL_FREE is distinct from
197 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
198 * blocks in zio_done() while another i/o that holds either
199 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
202 * Held as reader to prevent changes to the vdev tree during trivial
203 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
204 * other locks, and lower than all of them, to ensure that it's safe
205 * to acquire regardless of caller context.
207 * In addition, the following rules apply:
209 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
210 * The lock ordering is SCL_CONFIG > spa_props_lock.
212 * (b) I/O operations on leaf vdevs. For any zio operation that takes
213 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
214 * or zio_write_phys() -- the caller must ensure that the config cannot
215 * cannot change in the interim, and that the vdev cannot be reopened.
216 * SCL_STATE as reader suffices for both.
218 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
220 * spa_vdev_enter() Acquire the namespace lock and the config lock
223 * spa_vdev_exit() Release the config lock, wait for all I/O
224 * to complete, sync the updated configs to the
225 * cache, and release the namespace lock.
227 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
228 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
229 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
231 * spa_rename() is also implemented within this file since it requires
232 * manipulation of the namespace.
235 static avl_tree_t spa_namespace_avl;
236 kmutex_t spa_namespace_lock;
237 static kcondvar_t spa_namespace_cv;
238 static int spa_active_count;
239 int spa_max_replication_override = SPA_DVAS_PER_BP;
241 static kmutex_t spa_spare_lock;
242 static avl_tree_t spa_spare_avl;
243 static kmutex_t spa_l2cache_lock;
244 static avl_tree_t spa_l2cache_avl;
246 kmem_cache_t *spa_buffer_pool;
250 /* Everything except dprintf and spa is on by default in debug builds */
251 int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA);
257 * zfs_recover can be set to nonzero to attempt to recover from
258 * otherwise-fatal errors, typically caused by on-disk corruption. When
259 * set, calls to zfs_panic_recover() will turn into warning messages.
260 * This should only be used as a last resort, as it typically results
261 * in leaked space, or worse.
263 boolean_t zfs_recover = B_FALSE;
266 * If destroy encounters an EIO while reading metadata (e.g. indirect
267 * blocks), space referenced by the missing metadata can not be freed.
268 * Normally this causes the background destroy to become "stalled", as
269 * it is unable to make forward progress. While in this stalled state,
270 * all remaining space to free from the error-encountering filesystem is
271 * "temporarily leaked". Set this flag to cause it to ignore the EIO,
272 * permanently leak the space from indirect blocks that can not be read,
273 * and continue to free everything else that it can.
275 * The default, "stalling" behavior is useful if the storage partially
276 * fails (i.e. some but not all i/os fail), and then later recovers. In
277 * this case, we will be able to continue pool operations while it is
278 * partially failed, and when it recovers, we can continue to free the
279 * space, with no leaks. However, note that this case is actually
282 * Typically pools either (a) fail completely (but perhaps temporarily,
283 * e.g. a top-level vdev going offline), or (b) have localized,
284 * permanent errors (e.g. disk returns the wrong data due to bit flip or
285 * firmware bug). In case (a), this setting does not matter because the
286 * pool will be suspended and the sync thread will not be able to make
287 * forward progress regardless. In case (b), because the error is
288 * permanent, the best we can do is leak the minimum amount of space,
289 * which is what setting this flag will do. Therefore, it is reasonable
290 * for this flag to normally be set, but we chose the more conservative
291 * approach of not setting it, so that there is no possibility of
292 * leaking space in the "partial temporary" failure case.
294 boolean_t zfs_free_leak_on_eio = B_FALSE;
297 * Expiration time in milliseconds. This value has two meanings. First it is
298 * used to determine when the spa_deadman() logic should fire. By default the
299 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
300 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
301 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
304 uint64_t zfs_deadman_synctime_ms = 1000000ULL;
307 * Check time in milliseconds. This defines the frequency at which we check
310 uint64_t zfs_deadman_checktime_ms = 5000ULL;
313 * Default value of -1 for zfs_deadman_enabled is resolved in
316 int zfs_deadman_enabled = -1;
319 * The worst case is single-sector max-parity RAID-Z blocks, in which
320 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
321 * times the size; so just assume that. Add to this the fact that
322 * we can have up to 3 DVAs per bp, and one more factor of 2 because
323 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together,
325 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
327 int spa_asize_inflation = 24;
329 #if defined(__FreeBSD__) && defined(_KERNEL)
330 SYSCTL_DECL(_vfs_zfs);
331 SYSCTL_INT(_vfs_zfs, OID_AUTO, recover, CTLFLAG_RWTUN, &zfs_recover, 0,
332 "Try to recover from otherwise-fatal errors.");
335 sysctl_vfs_zfs_debug_flags(SYSCTL_HANDLER_ARGS)
340 err = sysctl_handle_int(oidp, &val, 0, req);
341 if (err != 0 || req->newptr == NULL)
345 * ZFS_DEBUG_MODIFY must be enabled prior to boot so all
346 * arc buffers in the system have the necessary additional
347 * checksum data. However, it is safe to disable at any
350 if (!(zfs_flags & ZFS_DEBUG_MODIFY))
351 val &= ~ZFS_DEBUG_MODIFY;
357 SYSCTL_PROC(_vfs_zfs, OID_AUTO, debugflags,
358 CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RWTUN, 0, sizeof(int),
359 sysctl_vfs_zfs_debug_flags, "IU", "Debug flags for ZFS testing.");
361 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_synctime_ms, CTLFLAG_RDTUN,
362 &zfs_deadman_synctime_ms, 0,
363 "Stalled ZFS I/O expiration time in milliseconds");
364 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_checktime_ms, CTLFLAG_RDTUN,
365 &zfs_deadman_checktime_ms, 0,
366 "Period of checks for stalled ZFS I/O in milliseconds");
367 SYSCTL_INT(_vfs_zfs, OID_AUTO, deadman_enabled, CTLFLAG_RDTUN,
368 &zfs_deadman_enabled, 0, "Kernel panic on stalled ZFS I/O");
369 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_asize_inflation, CTLFLAG_RWTUN,
370 &spa_asize_inflation, 0, "Worst case inflation factor for single sector writes");
379 * If we are not i386 or amd64 or in a virtual machine,
380 * disable ZFS deadman thread by default
382 if (zfs_deadman_enabled == -1) {
383 #if defined(__amd64__) || defined(__i386__)
384 zfs_deadman_enabled = (vm_guest == VM_GUEST_NO) ? 1 : 0;
386 zfs_deadman_enabled = 0;
391 #endif /* !illumos */
394 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
395 * the pool to be consumed. This ensures that we don't run the pool
396 * completely out of space, due to unaccounted changes (e.g. to the MOS).
397 * It also limits the worst-case time to allocate space. If we have
398 * less than this amount of free space, most ZPL operations (e.g. write,
399 * create) will return ENOSPC.
401 * Certain operations (e.g. file removal, most administrative actions) can
402 * use half the slop space. They will only return ENOSPC if less than half
403 * the slop space is free. Typically, once the pool has less than the slop
404 * space free, the user will use these operations to free up space in the pool.
405 * These are the operations that call dsl_pool_adjustedsize() with the netfree
406 * argument set to TRUE.
408 * A very restricted set of operations are always permitted, regardless of
409 * the amount of free space. These are the operations that call
410 * dsl_sync_task(ZFS_SPACE_CHECK_NONE), e.g. "zfs destroy". If these
411 * operations result in a net increase in the amount of space used,
412 * it is possible to run the pool completely out of space, causing it to
413 * be permanently read-only.
415 * Note that on very small pools, the slop space will be larger than
416 * 3.2%, in an effort to have it be at least spa_min_slop (128MB),
417 * but we never allow it to be more than half the pool size.
419 * See also the comments in zfs_space_check_t.
421 int spa_slop_shift = 5;
422 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_slop_shift, CTLFLAG_RWTUN,
424 "Shift value of reserved space (1/(2^spa_slop_shift)).");
425 uint64_t spa_min_slop = 128 * 1024 * 1024;
426 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, spa_min_slop, CTLFLAG_RWTUN,
428 "Minimal value of reserved space");
431 * ==========================================================================
433 * ==========================================================================
436 spa_config_lock_init(spa_t *spa)
438 for (int i = 0; i < SCL_LOCKS; i++) {
439 spa_config_lock_t *scl = &spa->spa_config_lock[i];
440 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
441 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
442 refcount_create_untracked(&scl->scl_count);
443 scl->scl_writer = NULL;
444 scl->scl_write_wanted = 0;
449 spa_config_lock_destroy(spa_t *spa)
451 for (int i = 0; i < SCL_LOCKS; i++) {
452 spa_config_lock_t *scl = &spa->spa_config_lock[i];
453 mutex_destroy(&scl->scl_lock);
454 cv_destroy(&scl->scl_cv);
455 refcount_destroy(&scl->scl_count);
456 ASSERT(scl->scl_writer == NULL);
457 ASSERT(scl->scl_write_wanted == 0);
462 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
464 for (int i = 0; i < SCL_LOCKS; i++) {
465 spa_config_lock_t *scl = &spa->spa_config_lock[i];
466 if (!(locks & (1 << i)))
468 mutex_enter(&scl->scl_lock);
469 if (rw == RW_READER) {
470 if (scl->scl_writer || scl->scl_write_wanted) {
471 mutex_exit(&scl->scl_lock);
472 spa_config_exit(spa, locks & ((1 << i) - 1),
477 ASSERT(scl->scl_writer != curthread);
478 if (!refcount_is_zero(&scl->scl_count)) {
479 mutex_exit(&scl->scl_lock);
480 spa_config_exit(spa, locks & ((1 << i) - 1),
484 scl->scl_writer = curthread;
486 (void) refcount_add(&scl->scl_count, tag);
487 mutex_exit(&scl->scl_lock);
493 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
497 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
499 for (int i = 0; i < SCL_LOCKS; i++) {
500 spa_config_lock_t *scl = &spa->spa_config_lock[i];
501 if (scl->scl_writer == curthread)
502 wlocks_held |= (1 << i);
503 if (!(locks & (1 << i)))
505 mutex_enter(&scl->scl_lock);
506 if (rw == RW_READER) {
507 while (scl->scl_writer || scl->scl_write_wanted) {
508 cv_wait(&scl->scl_cv, &scl->scl_lock);
511 ASSERT(scl->scl_writer != curthread);
512 while (!refcount_is_zero(&scl->scl_count)) {
513 scl->scl_write_wanted++;
514 cv_wait(&scl->scl_cv, &scl->scl_lock);
515 scl->scl_write_wanted--;
517 scl->scl_writer = curthread;
519 (void) refcount_add(&scl->scl_count, tag);
520 mutex_exit(&scl->scl_lock);
522 ASSERT(wlocks_held <= locks);
526 spa_config_exit(spa_t *spa, int locks, void *tag)
528 for (int i = SCL_LOCKS - 1; i >= 0; i--) {
529 spa_config_lock_t *scl = &spa->spa_config_lock[i];
530 if (!(locks & (1 << i)))
532 mutex_enter(&scl->scl_lock);
533 ASSERT(!refcount_is_zero(&scl->scl_count));
534 if (refcount_remove(&scl->scl_count, tag) == 0) {
535 ASSERT(scl->scl_writer == NULL ||
536 scl->scl_writer == curthread);
537 scl->scl_writer = NULL; /* OK in either case */
538 cv_broadcast(&scl->scl_cv);
540 mutex_exit(&scl->scl_lock);
545 spa_config_held(spa_t *spa, int locks, krw_t rw)
549 for (int i = 0; i < SCL_LOCKS; i++) {
550 spa_config_lock_t *scl = &spa->spa_config_lock[i];
551 if (!(locks & (1 << i)))
553 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
554 (rw == RW_WRITER && scl->scl_writer == curthread))
555 locks_held |= 1 << i;
562 * ==========================================================================
563 * SPA namespace functions
564 * ==========================================================================
568 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
569 * Returns NULL if no matching spa_t is found.
572 spa_lookup(const char *name)
574 static spa_t search; /* spa_t is large; don't allocate on stack */
579 ASSERT(MUTEX_HELD(&spa_namespace_lock));
581 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
584 * If it's a full dataset name, figure out the pool name and
587 cp = strpbrk(search.spa_name, "/@#");
591 spa = avl_find(&spa_namespace_avl, &search, &where);
597 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
598 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
599 * looking for potentially hung I/Os.
602 spa_deadman(void *arg, int pending)
607 * Disable the deadman timer if the pool is suspended.
609 if (spa_suspended(spa)) {
611 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
613 /* Nothing. just don't schedule any future callouts. */
618 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
619 (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
620 ++spa->spa_deadman_calls);
621 if (zfs_deadman_enabled)
622 vdev_deadman(spa->spa_root_vdev);
625 callout_schedule(&spa->spa_deadman_cycid,
626 hz * zfs_deadman_checktime_ms / MILLISEC);
631 #if defined(__FreeBSD__) && defined(_KERNEL)
633 spa_deadman_timeout(void *arg)
637 taskqueue_enqueue(taskqueue_thread, &spa->spa_deadman_task);
642 * Create an uninitialized spa_t with the given name. Requires
643 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
644 * exist by calling spa_lookup() first.
647 spa_add(const char *name, nvlist_t *config, const char *altroot)
650 spa_config_dirent_t *dp;
656 ASSERT(MUTEX_HELD(&spa_namespace_lock));
658 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
660 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
661 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
662 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
663 mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL);
664 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
665 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
666 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
667 mutex_init(&spa->spa_cksum_tmpls_lock, NULL, MUTEX_DEFAULT, NULL);
668 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
669 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
670 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
671 mutex_init(&spa->spa_alloc_lock, NULL, MUTEX_DEFAULT, NULL);
673 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
674 cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL);
675 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
676 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
677 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
679 for (int t = 0; t < TXG_SIZE; t++)
680 bplist_create(&spa->spa_free_bplist[t]);
682 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
683 spa->spa_state = POOL_STATE_UNINITIALIZED;
684 spa->spa_freeze_txg = UINT64_MAX;
685 spa->spa_final_txg = UINT64_MAX;
686 spa->spa_load_max_txg = UINT64_MAX;
688 spa->spa_proc_state = SPA_PROC_NONE;
691 hdlr.cyh_func = spa_deadman;
693 hdlr.cyh_level = CY_LOW_LEVEL;
696 spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
700 * This determines how often we need to check for hung I/Os after
701 * the cyclic has already fired. Since checking for hung I/Os is
702 * an expensive operation we don't want to check too frequently.
703 * Instead wait for 5 seconds before checking again.
705 when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms);
706 when.cyt_when = CY_INFINITY;
707 mutex_enter(&cpu_lock);
708 spa->spa_deadman_cycid = cyclic_add(&hdlr, &when);
709 mutex_exit(&cpu_lock);
713 * callout(9) does not provide a way to initialize a callout with
714 * a function and an argument, so we use callout_reset() to schedule
715 * the callout in the very distant future. Even if that event ever
716 * fires, it should be okayas we won't have any active zio-s.
717 * But normally spa_sync() will reschedule the callout with a proper
719 * callout(9) does not allow the callback function to sleep but
720 * vdev_deadman() needs to acquire vq_lock and illumos mutexes are
721 * emulated using sx(9). For this reason spa_deadman_timeout()
722 * will schedule spa_deadman() as task on a taskqueue that allows
725 TASK_INIT(&spa->spa_deadman_task, 0, spa_deadman, spa);
726 callout_init(&spa->spa_deadman_cycid, 1);
727 callout_reset_sbt(&spa->spa_deadman_cycid, SBT_MAX, 0,
728 spa_deadman_timeout, spa, 0);
731 refcount_create(&spa->spa_refcount);
732 spa_config_lock_init(spa);
734 avl_add(&spa_namespace_avl, spa);
737 * Set the alternate root, if there is one.
740 spa->spa_root = spa_strdup(altroot);
744 avl_create(&spa->spa_alloc_tree, zio_bookmark_compare,
745 sizeof (zio_t), offsetof(zio_t, io_alloc_node));
748 * Every pool starts with the default cachefile
750 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
751 offsetof(spa_config_dirent_t, scd_link));
753 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
754 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
755 list_insert_head(&spa->spa_config_list, dp);
757 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
760 if (config != NULL) {
763 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
765 VERIFY(nvlist_dup(features, &spa->spa_label_features,
769 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
772 if (spa->spa_label_features == NULL) {
773 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
777 spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0);
779 spa->spa_min_ashift = INT_MAX;
780 spa->spa_max_ashift = 0;
783 * As a pool is being created, treat all features as disabled by
784 * setting SPA_FEATURE_DISABLED for all entries in the feature
787 for (int i = 0; i < SPA_FEATURES; i++) {
788 spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED;
795 * Removes a spa_t from the namespace, freeing up any memory used. Requires
796 * spa_namespace_lock. This is called only after the spa_t has been closed and
800 spa_remove(spa_t *spa)
802 spa_config_dirent_t *dp;
804 ASSERT(MUTEX_HELD(&spa_namespace_lock));
805 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
806 ASSERT3U(refcount_count(&spa->spa_refcount), ==, 0);
808 nvlist_free(spa->spa_config_splitting);
810 avl_remove(&spa_namespace_avl, spa);
811 cv_broadcast(&spa_namespace_cv);
814 spa_strfree(spa->spa_root);
818 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
819 list_remove(&spa->spa_config_list, dp);
820 if (dp->scd_path != NULL)
821 spa_strfree(dp->scd_path);
822 kmem_free(dp, sizeof (spa_config_dirent_t));
825 avl_destroy(&spa->spa_alloc_tree);
826 list_destroy(&spa->spa_config_list);
828 nvlist_free(spa->spa_label_features);
829 nvlist_free(spa->spa_load_info);
830 spa_config_set(spa, NULL);
833 mutex_enter(&cpu_lock);
834 if (spa->spa_deadman_cycid != CYCLIC_NONE)
835 cyclic_remove(spa->spa_deadman_cycid);
836 mutex_exit(&cpu_lock);
837 spa->spa_deadman_cycid = CYCLIC_NONE;
840 callout_drain(&spa->spa_deadman_cycid);
841 taskqueue_drain(taskqueue_thread, &spa->spa_deadman_task);
845 refcount_destroy(&spa->spa_refcount);
847 spa_config_lock_destroy(spa);
849 for (int t = 0; t < TXG_SIZE; t++)
850 bplist_destroy(&spa->spa_free_bplist[t]);
852 zio_checksum_templates_free(spa);
854 cv_destroy(&spa->spa_async_cv);
855 cv_destroy(&spa->spa_evicting_os_cv);
856 cv_destroy(&spa->spa_proc_cv);
857 cv_destroy(&spa->spa_scrub_io_cv);
858 cv_destroy(&spa->spa_suspend_cv);
860 mutex_destroy(&spa->spa_alloc_lock);
861 mutex_destroy(&spa->spa_async_lock);
862 mutex_destroy(&spa->spa_errlist_lock);
863 mutex_destroy(&spa->spa_errlog_lock);
864 mutex_destroy(&spa->spa_evicting_os_lock);
865 mutex_destroy(&spa->spa_history_lock);
866 mutex_destroy(&spa->spa_proc_lock);
867 mutex_destroy(&spa->spa_props_lock);
868 mutex_destroy(&spa->spa_cksum_tmpls_lock);
869 mutex_destroy(&spa->spa_scrub_lock);
870 mutex_destroy(&spa->spa_suspend_lock);
871 mutex_destroy(&spa->spa_vdev_top_lock);
873 kmem_free(spa, sizeof (spa_t));
877 * Given a pool, return the next pool in the namespace, or NULL if there is
878 * none. If 'prev' is NULL, return the first pool.
881 spa_next(spa_t *prev)
883 ASSERT(MUTEX_HELD(&spa_namespace_lock));
886 return (AVL_NEXT(&spa_namespace_avl, prev));
888 return (avl_first(&spa_namespace_avl));
892 * ==========================================================================
893 * SPA refcount functions
894 * ==========================================================================
898 * Add a reference to the given spa_t. Must have at least one reference, or
899 * have the namespace lock held.
902 spa_open_ref(spa_t *spa, void *tag)
904 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
905 MUTEX_HELD(&spa_namespace_lock));
906 (void) refcount_add(&spa->spa_refcount, tag);
910 * Remove a reference to the given spa_t. Must have at least one reference, or
911 * have the namespace lock held.
914 spa_close(spa_t *spa, void *tag)
916 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
917 MUTEX_HELD(&spa_namespace_lock));
918 (void) refcount_remove(&spa->spa_refcount, tag);
922 * Remove a reference to the given spa_t held by a dsl dir that is
923 * being asynchronously released. Async releases occur from a taskq
924 * performing eviction of dsl datasets and dirs. The namespace lock
925 * isn't held and the hold by the object being evicted may contribute to
926 * spa_minref (e.g. dataset or directory released during pool export),
927 * so the asserts in spa_close() do not apply.
930 spa_async_close(spa_t *spa, void *tag)
932 (void) refcount_remove(&spa->spa_refcount, tag);
936 * Check to see if the spa refcount is zero. Must be called with
937 * spa_namespace_lock held. We really compare against spa_minref, which is the
938 * number of references acquired when opening a pool
941 spa_refcount_zero(spa_t *spa)
943 ASSERT(MUTEX_HELD(&spa_namespace_lock));
945 return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
949 * ==========================================================================
950 * SPA spare and l2cache tracking
951 * ==========================================================================
955 * Hot spares and cache devices are tracked using the same code below,
956 * for 'auxiliary' devices.
959 typedef struct spa_aux {
967 spa_aux_compare(const void *a, const void *b)
969 const spa_aux_t *sa = a;
970 const spa_aux_t *sb = b;
972 if (sa->aux_guid < sb->aux_guid)
974 else if (sa->aux_guid > sb->aux_guid)
981 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
987 search.aux_guid = vd->vdev_guid;
988 if ((aux = avl_find(avl, &search, &where)) != NULL) {
991 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
992 aux->aux_guid = vd->vdev_guid;
994 avl_insert(avl, aux, where);
999 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
1005 search.aux_guid = vd->vdev_guid;
1006 aux = avl_find(avl, &search, &where);
1008 ASSERT(aux != NULL);
1010 if (--aux->aux_count == 0) {
1011 avl_remove(avl, aux);
1012 kmem_free(aux, sizeof (spa_aux_t));
1013 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
1014 aux->aux_pool = 0ULL;
1019 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
1021 spa_aux_t search, *found;
1023 search.aux_guid = guid;
1024 found = avl_find(avl, &search, NULL);
1028 *pool = found->aux_pool;
1035 *refcnt = found->aux_count;
1040 return (found != NULL);
1044 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
1046 spa_aux_t search, *found;
1049 search.aux_guid = vd->vdev_guid;
1050 found = avl_find(avl, &search, &where);
1051 ASSERT(found != NULL);
1052 ASSERT(found->aux_pool == 0ULL);
1054 found->aux_pool = spa_guid(vd->vdev_spa);
1058 * Spares are tracked globally due to the following constraints:
1060 * - A spare may be part of multiple pools.
1061 * - A spare may be added to a pool even if it's actively in use within
1063 * - A spare in use in any pool can only be the source of a replacement if
1064 * the target is a spare in the same pool.
1066 * We keep track of all spares on the system through the use of a reference
1067 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
1068 * spare, then we bump the reference count in the AVL tree. In addition, we set
1069 * the 'vdev_isspare' member to indicate that the device is a spare (active or
1070 * inactive). When a spare is made active (used to replace a device in the
1071 * pool), we also keep track of which pool its been made a part of.
1073 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
1074 * called under the spa_namespace lock as part of vdev reconfiguration. The
1075 * separate spare lock exists for the status query path, which does not need to
1076 * be completely consistent with respect to other vdev configuration changes.
1080 spa_spare_compare(const void *a, const void *b)
1082 return (spa_aux_compare(a, b));
1086 spa_spare_add(vdev_t *vd)
1088 mutex_enter(&spa_spare_lock);
1089 ASSERT(!vd->vdev_isspare);
1090 spa_aux_add(vd, &spa_spare_avl);
1091 vd->vdev_isspare = B_TRUE;
1092 mutex_exit(&spa_spare_lock);
1096 spa_spare_remove(vdev_t *vd)
1098 mutex_enter(&spa_spare_lock);
1099 ASSERT(vd->vdev_isspare);
1100 spa_aux_remove(vd, &spa_spare_avl);
1101 vd->vdev_isspare = B_FALSE;
1102 mutex_exit(&spa_spare_lock);
1106 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
1110 mutex_enter(&spa_spare_lock);
1111 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
1112 mutex_exit(&spa_spare_lock);
1118 spa_spare_activate(vdev_t *vd)
1120 mutex_enter(&spa_spare_lock);
1121 ASSERT(vd->vdev_isspare);
1122 spa_aux_activate(vd, &spa_spare_avl);
1123 mutex_exit(&spa_spare_lock);
1127 * Level 2 ARC devices are tracked globally for the same reasons as spares.
1128 * Cache devices currently only support one pool per cache device, and so
1129 * for these devices the aux reference count is currently unused beyond 1.
1133 spa_l2cache_compare(const void *a, const void *b)
1135 return (spa_aux_compare(a, b));
1139 spa_l2cache_add(vdev_t *vd)
1141 mutex_enter(&spa_l2cache_lock);
1142 ASSERT(!vd->vdev_isl2cache);
1143 spa_aux_add(vd, &spa_l2cache_avl);
1144 vd->vdev_isl2cache = B_TRUE;
1145 mutex_exit(&spa_l2cache_lock);
1149 spa_l2cache_remove(vdev_t *vd)
1151 mutex_enter(&spa_l2cache_lock);
1152 ASSERT(vd->vdev_isl2cache);
1153 spa_aux_remove(vd, &spa_l2cache_avl);
1154 vd->vdev_isl2cache = B_FALSE;
1155 mutex_exit(&spa_l2cache_lock);
1159 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
1163 mutex_enter(&spa_l2cache_lock);
1164 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
1165 mutex_exit(&spa_l2cache_lock);
1171 spa_l2cache_activate(vdev_t *vd)
1173 mutex_enter(&spa_l2cache_lock);
1174 ASSERT(vd->vdev_isl2cache);
1175 spa_aux_activate(vd, &spa_l2cache_avl);
1176 mutex_exit(&spa_l2cache_lock);
1180 * ==========================================================================
1182 * ==========================================================================
1186 * Lock the given spa_t for the purpose of adding or removing a vdev.
1187 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1188 * It returns the next transaction group for the spa_t.
1191 spa_vdev_enter(spa_t *spa)
1193 mutex_enter(&spa->spa_vdev_top_lock);
1194 mutex_enter(&spa_namespace_lock);
1195 return (spa_vdev_config_enter(spa));
1199 * Internal implementation for spa_vdev_enter(). Used when a vdev
1200 * operation requires multiple syncs (i.e. removing a device) while
1201 * keeping the spa_namespace_lock held.
1204 spa_vdev_config_enter(spa_t *spa)
1206 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1208 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1210 return (spa_last_synced_txg(spa) + 1);
1214 * Used in combination with spa_vdev_config_enter() to allow the syncing
1215 * of multiple transactions without releasing the spa_namespace_lock.
1218 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
1220 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1222 int config_changed = B_FALSE;
1224 ASSERT(txg > spa_last_synced_txg(spa));
1226 spa->spa_pending_vdev = NULL;
1229 * Reassess the DTLs.
1231 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
1233 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
1234 config_changed = B_TRUE;
1235 spa->spa_config_generation++;
1239 * Verify the metaslab classes.
1241 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
1242 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
1244 spa_config_exit(spa, SCL_ALL, spa);
1247 * Panic the system if the specified tag requires it. This
1248 * is useful for ensuring that configurations are updated
1251 if (zio_injection_enabled)
1252 zio_handle_panic_injection(spa, tag, 0);
1255 * Note: this txg_wait_synced() is important because it ensures
1256 * that there won't be more than one config change per txg.
1257 * This allows us to use the txg as the generation number.
1260 txg_wait_synced(spa->spa_dsl_pool, txg);
1263 ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
1264 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1266 spa_config_exit(spa, SCL_ALL, spa);
1270 * If the config changed, update the config cache.
1273 spa_config_sync(spa, B_FALSE, B_TRUE);
1277 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1278 * locking of spa_vdev_enter(), we also want make sure the transactions have
1279 * synced to disk, and then update the global configuration cache with the new
1283 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1285 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1286 mutex_exit(&spa_namespace_lock);
1287 mutex_exit(&spa->spa_vdev_top_lock);
1293 * Lock the given spa_t for the purpose of changing vdev state.
1296 spa_vdev_state_enter(spa_t *spa, int oplocks)
1298 int locks = SCL_STATE_ALL | oplocks;
1301 * Root pools may need to read of the underlying devfs filesystem
1302 * when opening up a vdev. Unfortunately if we're holding the
1303 * SCL_ZIO lock it will result in a deadlock when we try to issue
1304 * the read from the root filesystem. Instead we "prefetch"
1305 * the associated vnodes that we need prior to opening the
1306 * underlying devices and cache them so that we can prevent
1307 * any I/O when we are doing the actual open.
1309 if (spa_is_root(spa)) {
1310 int low = locks & ~(SCL_ZIO - 1);
1311 int high = locks & ~low;
1313 spa_config_enter(spa, high, spa, RW_WRITER);
1314 vdev_hold(spa->spa_root_vdev);
1315 spa_config_enter(spa, low, spa, RW_WRITER);
1317 spa_config_enter(spa, locks, spa, RW_WRITER);
1319 spa->spa_vdev_locks = locks;
1323 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1325 boolean_t config_changed = B_FALSE;
1327 if (vd != NULL || error == 0)
1328 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1332 vdev_state_dirty(vd->vdev_top);
1333 config_changed = B_TRUE;
1334 spa->spa_config_generation++;
1337 if (spa_is_root(spa))
1338 vdev_rele(spa->spa_root_vdev);
1340 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1341 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1344 * If anything changed, wait for it to sync. This ensures that,
1345 * from the system administrator's perspective, zpool(1M) commands
1346 * are synchronous. This is important for things like zpool offline:
1347 * when the command completes, you expect no further I/O from ZFS.
1350 txg_wait_synced(spa->spa_dsl_pool, 0);
1353 * If the config changed, update the config cache.
1355 if (config_changed) {
1356 mutex_enter(&spa_namespace_lock);
1357 spa_config_sync(spa, B_FALSE, B_TRUE);
1358 mutex_exit(&spa_namespace_lock);
1365 * ==========================================================================
1366 * Miscellaneous functions
1367 * ==========================================================================
1371 spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx)
1373 if (!nvlist_exists(spa->spa_label_features, feature)) {
1374 fnvlist_add_boolean(spa->spa_label_features, feature);
1376 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1377 * dirty the vdev config because lock SCL_CONFIG is not held.
1378 * Thankfully, in this case we don't need to dirty the config
1379 * because it will be written out anyway when we finish
1380 * creating the pool.
1382 if (tx->tx_txg != TXG_INITIAL)
1383 vdev_config_dirty(spa->spa_root_vdev);
1388 spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1390 if (nvlist_remove_all(spa->spa_label_features, feature) == 0)
1391 vdev_config_dirty(spa->spa_root_vdev);
1398 spa_rename(const char *name, const char *newname)
1404 * Lookup the spa_t and grab the config lock for writing. We need to
1405 * actually open the pool so that we can sync out the necessary labels.
1406 * It's OK to call spa_open() with the namespace lock held because we
1407 * allow recursive calls for other reasons.
1409 mutex_enter(&spa_namespace_lock);
1410 if ((err = spa_open(name, &spa, FTAG)) != 0) {
1411 mutex_exit(&spa_namespace_lock);
1415 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1417 avl_remove(&spa_namespace_avl, spa);
1418 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1419 avl_add(&spa_namespace_avl, spa);
1422 * Sync all labels to disk with the new names by marking the root vdev
1423 * dirty and waiting for it to sync. It will pick up the new pool name
1426 vdev_config_dirty(spa->spa_root_vdev);
1428 spa_config_exit(spa, SCL_ALL, FTAG);
1430 txg_wait_synced(spa->spa_dsl_pool, 0);
1433 * Sync the updated config cache.
1435 spa_config_sync(spa, B_FALSE, B_TRUE);
1437 spa_close(spa, FTAG);
1439 mutex_exit(&spa_namespace_lock);
1445 * Return the spa_t associated with given pool_guid, if it exists. If
1446 * device_guid is non-zero, determine whether the pool exists *and* contains
1447 * a device with the specified device_guid.
1450 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1453 avl_tree_t *t = &spa_namespace_avl;
1455 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1457 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1458 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1460 if (spa->spa_root_vdev == NULL)
1462 if (spa_guid(spa) == pool_guid) {
1463 if (device_guid == 0)
1466 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1467 device_guid) != NULL)
1471 * Check any devices we may be in the process of adding.
1473 if (spa->spa_pending_vdev) {
1474 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1475 device_guid) != NULL)
1485 * Determine whether a pool with the given pool_guid exists.
1488 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1490 return (spa_by_guid(pool_guid, device_guid) != NULL);
1494 spa_strdup(const char *s)
1500 new = kmem_alloc(len + 1, KM_SLEEP);
1508 spa_strfree(char *s)
1510 kmem_free(s, strlen(s) + 1);
1514 spa_get_random(uint64_t range)
1520 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1526 spa_generate_guid(spa_t *spa)
1528 uint64_t guid = spa_get_random(-1ULL);
1531 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1532 guid = spa_get_random(-1ULL);
1534 while (guid == 0 || spa_guid_exists(guid, 0))
1535 guid = spa_get_random(-1ULL);
1542 snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp)
1545 char *checksum = NULL;
1546 char *compress = NULL;
1549 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1550 dmu_object_byteswap_t bswap =
1551 DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1552 (void) snprintf(type, sizeof (type), "bswap %s %s",
1553 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1554 "metadata" : "data",
1555 dmu_ot_byteswap[bswap].ob_name);
1557 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1560 if (!BP_IS_EMBEDDED(bp)) {
1562 zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1564 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1567 SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum,
1572 spa_freeze(spa_t *spa)
1574 uint64_t freeze_txg = 0;
1576 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1577 if (spa->spa_freeze_txg == UINT64_MAX) {
1578 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1579 spa->spa_freeze_txg = freeze_txg;
1581 spa_config_exit(spa, SCL_ALL, FTAG);
1582 if (freeze_txg != 0)
1583 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1587 zfs_panic_recover(const char *fmt, ...)
1592 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1597 * This is a stripped-down version of strtoull, suitable only for converting
1598 * lowercase hexadecimal numbers that don't overflow.
1601 zfs_strtonum(const char *str, char **nptr)
1607 while ((c = *str) != '\0') {
1608 if (c >= '0' && c <= '9')
1610 else if (c >= 'a' && c <= 'f')
1611 digit = 10 + c - 'a';
1622 *nptr = (char *)str;
1628 * ==========================================================================
1629 * Accessor functions
1630 * ==========================================================================
1634 spa_shutting_down(spa_t *spa)
1636 return (spa->spa_async_suspended);
1640 spa_get_dsl(spa_t *spa)
1642 return (spa->spa_dsl_pool);
1646 spa_is_initializing(spa_t *spa)
1648 return (spa->spa_is_initializing);
1652 spa_get_rootblkptr(spa_t *spa)
1654 return (&spa->spa_ubsync.ub_rootbp);
1658 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1660 spa->spa_uberblock.ub_rootbp = *bp;
1664 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1666 if (spa->spa_root == NULL)
1669 (void) strncpy(buf, spa->spa_root, buflen);
1673 spa_sync_pass(spa_t *spa)
1675 return (spa->spa_sync_pass);
1679 spa_name(spa_t *spa)
1681 return (spa->spa_name);
1685 spa_guid(spa_t *spa)
1687 dsl_pool_t *dp = spa_get_dsl(spa);
1691 * If we fail to parse the config during spa_load(), we can go through
1692 * the error path (which posts an ereport) and end up here with no root
1693 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1696 if (spa->spa_root_vdev == NULL)
1697 return (spa->spa_config_guid);
1699 guid = spa->spa_last_synced_guid != 0 ?
1700 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1703 * Return the most recently synced out guid unless we're
1704 * in syncing context.
1706 if (dp && dsl_pool_sync_context(dp))
1707 return (spa->spa_root_vdev->vdev_guid);
1713 spa_load_guid(spa_t *spa)
1716 * This is a GUID that exists solely as a reference for the
1717 * purposes of the arc. It is generated at load time, and
1718 * is never written to persistent storage.
1720 return (spa->spa_load_guid);
1724 spa_last_synced_txg(spa_t *spa)
1726 return (spa->spa_ubsync.ub_txg);
1730 spa_first_txg(spa_t *spa)
1732 return (spa->spa_first_txg);
1736 spa_syncing_txg(spa_t *spa)
1738 return (spa->spa_syncing_txg);
1742 * Return the last txg where data can be dirtied. The final txgs
1743 * will be used to just clear out any deferred frees that remain.
1746 spa_final_dirty_txg(spa_t *spa)
1748 return (spa->spa_final_txg - TXG_DEFER_SIZE);
1752 spa_state(spa_t *spa)
1754 return (spa->spa_state);
1758 spa_load_state(spa_t *spa)
1760 return (spa->spa_load_state);
1764 spa_freeze_txg(spa_t *spa)
1766 return (spa->spa_freeze_txg);
1771 spa_get_worst_case_asize(spa_t *spa, uint64_t lsize)
1773 return (lsize * spa_asize_inflation);
1777 * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%),
1778 * or at least 128MB, unless that would cause it to be more than half the
1781 * See the comment above spa_slop_shift for details.
1784 spa_get_slop_space(spa_t *spa)
1786 uint64_t space = spa_get_dspace(spa);
1787 return (MAX(space >> spa_slop_shift, MIN(space >> 1, spa_min_slop)));
1791 spa_get_dspace(spa_t *spa)
1793 return (spa->spa_dspace);
1797 spa_update_dspace(spa_t *spa)
1799 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1800 ddt_get_dedup_dspace(spa);
1804 * Return the failure mode that has been set to this pool. The default
1805 * behavior will be to block all I/Os when a complete failure occurs.
1808 spa_get_failmode(spa_t *spa)
1810 return (spa->spa_failmode);
1814 spa_suspended(spa_t *spa)
1816 return (spa->spa_suspended);
1820 spa_version(spa_t *spa)
1822 return (spa->spa_ubsync.ub_version);
1826 spa_deflate(spa_t *spa)
1828 return (spa->spa_deflate);
1832 spa_normal_class(spa_t *spa)
1834 return (spa->spa_normal_class);
1838 spa_log_class(spa_t *spa)
1840 return (spa->spa_log_class);
1844 spa_evicting_os_register(spa_t *spa, objset_t *os)
1846 mutex_enter(&spa->spa_evicting_os_lock);
1847 list_insert_head(&spa->spa_evicting_os_list, os);
1848 mutex_exit(&spa->spa_evicting_os_lock);
1852 spa_evicting_os_deregister(spa_t *spa, objset_t *os)
1854 mutex_enter(&spa->spa_evicting_os_lock);
1855 list_remove(&spa->spa_evicting_os_list, os);
1856 cv_broadcast(&spa->spa_evicting_os_cv);
1857 mutex_exit(&spa->spa_evicting_os_lock);
1861 spa_evicting_os_wait(spa_t *spa)
1863 mutex_enter(&spa->spa_evicting_os_lock);
1864 while (!list_is_empty(&spa->spa_evicting_os_list))
1865 cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock);
1866 mutex_exit(&spa->spa_evicting_os_lock);
1868 dmu_buf_user_evict_wait();
1872 spa_max_replication(spa_t *spa)
1875 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1876 * handle BPs with more than one DVA allocated. Set our max
1877 * replication level accordingly.
1879 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1881 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1885 spa_prev_software_version(spa_t *spa)
1887 return (spa->spa_prev_software_version);
1891 spa_deadman_synctime(spa_t *spa)
1893 return (spa->spa_deadman_synctime);
1897 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1899 uint64_t asize = DVA_GET_ASIZE(dva);
1900 uint64_t dsize = asize;
1902 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1904 if (asize != 0 && spa->spa_deflate) {
1905 uint64_t vdev = DVA_GET_VDEV(dva);
1906 vdev_t *vd = vdev_lookup_top(spa, vdev);
1909 "dva_get_dsize_sync(): bad DVA %llu:%llu",
1910 (u_longlong_t)vdev, (u_longlong_t)asize);
1912 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1919 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1923 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1924 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1930 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1934 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1936 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1937 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1939 spa_config_exit(spa, SCL_VDEV, FTAG);
1945 * ==========================================================================
1946 * Initialization and Termination
1947 * ==========================================================================
1951 spa_name_compare(const void *a1, const void *a2)
1953 const spa_t *s1 = a1;
1954 const spa_t *s2 = a2;
1957 s = strcmp(s1->spa_name, s2->spa_name);
1968 return (spa_active_count);
1978 EVENTHANDLER_DEFINE(mountroot, spa_boot_init, NULL, 0);
1984 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1985 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1986 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1987 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1989 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1990 offsetof(spa_t, spa_avl));
1992 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1993 offsetof(spa_aux_t, aux_avl));
1995 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1996 offsetof(spa_aux_t, aux_avl));
1998 spa_mode_global = mode;
2004 if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
2005 arc_procfd = open("/proc/self/ctl", O_WRONLY);
2006 if (arc_procfd == -1) {
2007 perror("could not enable watchpoints: "
2008 "opening /proc/self/ctl failed: ");
2014 #endif /* illumos */
2018 metaslab_alloc_trace_init();
2023 vdev_cache_stat_init();
2027 zpool_feature_init();
2034 #endif /* !illumos */
2045 vdev_cache_stat_fini();
2050 metaslab_alloc_trace_fini();
2055 avl_destroy(&spa_namespace_avl);
2056 avl_destroy(&spa_spare_avl);
2057 avl_destroy(&spa_l2cache_avl);
2059 cv_destroy(&spa_namespace_cv);
2060 mutex_destroy(&spa_namespace_lock);
2061 mutex_destroy(&spa_spare_lock);
2062 mutex_destroy(&spa_l2cache_lock);
2066 * Return whether this pool has slogs. No locking needed.
2067 * It's not a problem if the wrong answer is returned as it's only for
2068 * performance and not correctness
2071 spa_has_slogs(spa_t *spa)
2073 return (spa->spa_log_class->mc_rotor != NULL);
2077 spa_get_log_state(spa_t *spa)
2079 return (spa->spa_log_state);
2083 spa_set_log_state(spa_t *spa, spa_log_state_t state)
2085 spa->spa_log_state = state;
2089 spa_is_root(spa_t *spa)
2091 return (spa->spa_is_root);
2095 spa_writeable(spa_t *spa)
2097 return (!!(spa->spa_mode & FWRITE));
2101 * Returns true if there is a pending sync task in any of the current
2102 * syncing txg, the current quiescing txg, or the current open txg.
2105 spa_has_pending_synctask(spa_t *spa)
2107 return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks));
2111 spa_mode(spa_t *spa)
2113 return (spa->spa_mode);
2117 spa_bootfs(spa_t *spa)
2119 return (spa->spa_bootfs);
2123 spa_delegation(spa_t *spa)
2125 return (spa->spa_delegation);
2129 spa_meta_objset(spa_t *spa)
2131 return (spa->spa_meta_objset);
2135 spa_dedup_checksum(spa_t *spa)
2137 return (spa->spa_dedup_checksum);
2141 * Reset pool scan stat per scan pass (or reboot).
2144 spa_scan_stat_init(spa_t *spa)
2146 /* data not stored on disk */
2147 spa->spa_scan_pass_start = gethrestime_sec();
2148 spa->spa_scan_pass_exam = 0;
2149 vdev_scan_stat_init(spa->spa_root_vdev);
2153 * Get scan stats for zpool status reports
2156 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
2158 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
2160 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
2161 return (SET_ERROR(ENOENT));
2162 bzero(ps, sizeof (pool_scan_stat_t));
2164 /* data stored on disk */
2165 ps->pss_func = scn->scn_phys.scn_func;
2166 ps->pss_start_time = scn->scn_phys.scn_start_time;
2167 ps->pss_end_time = scn->scn_phys.scn_end_time;
2168 ps->pss_to_examine = scn->scn_phys.scn_to_examine;
2169 ps->pss_examined = scn->scn_phys.scn_examined;
2170 ps->pss_to_process = scn->scn_phys.scn_to_process;
2171 ps->pss_processed = scn->scn_phys.scn_processed;
2172 ps->pss_errors = scn->scn_phys.scn_errors;
2173 ps->pss_state = scn->scn_phys.scn_state;
2175 /* data not stored on disk */
2176 ps->pss_pass_start = spa->spa_scan_pass_start;
2177 ps->pss_pass_exam = spa->spa_scan_pass_exam;
2183 spa_debug_enabled(spa_t *spa)
2185 return (spa->spa_debug);
2189 spa_maxblocksize(spa_t *spa)
2191 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS))
2192 return (SPA_MAXBLOCKSIZE);
2194 return (SPA_OLD_MAXBLOCKSIZE);