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]
29 * Copyright (c) 2017 Datto Inc.
32 #include <sys/zfs_context.h>
33 #include <sys/spa_impl.h>
34 #include <sys/spa_boot.h>
36 #include <sys/zio_checksum.h>
37 #include <sys/zio_compress.h>
39 #include <sys/dmu_tx.h>
42 #include <sys/vdev_impl.h>
43 #include <sys/vdev_file.h>
44 #include <sys/metaslab.h>
45 #include <sys/uberblock_impl.h>
48 #include <sys/unique.h>
49 #include <sys/dsl_pool.h>
50 #include <sys/dsl_dir.h>
51 #include <sys/dsl_prop.h>
52 #include <sys/dsl_scan.h>
53 #include <sys/fs/zfs.h>
54 #include <sys/metaslab_impl.h>
58 #include <sys/zfeature.h>
60 #if defined(__FreeBSD__) && defined(_KERNEL)
61 #include <sys/types.h>
62 #include <sys/sysctl.h>
68 * There are four basic locks for managing spa_t structures:
70 * spa_namespace_lock (global mutex)
72 * This lock must be acquired to do any of the following:
74 * - Lookup a spa_t by name
75 * - Add or remove a spa_t from the namespace
76 * - Increase spa_refcount from non-zero
77 * - Check if spa_refcount is zero
79 * - add/remove/attach/detach devices
80 * - Held for the duration of create/destroy/import/export
82 * It does not need to handle recursion. A create or destroy may
83 * reference objects (files or zvols) in other pools, but by
84 * definition they must have an existing reference, and will never need
85 * to lookup a spa_t by name.
87 * spa_refcount (per-spa refcount_t protected by mutex)
89 * This reference count keep track of any active users of the spa_t. The
90 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
91 * the refcount is never really 'zero' - opening a pool implicitly keeps
92 * some references in the DMU. Internally we check against spa_minref, but
93 * present the image of a zero/non-zero value to consumers.
95 * spa_config_lock[] (per-spa array of rwlocks)
97 * This protects the spa_t from config changes, and must be held in
98 * the following circumstances:
100 * - RW_READER to perform I/O to the spa
101 * - RW_WRITER to change the vdev config
103 * The locking order is fairly straightforward:
105 * spa_namespace_lock -> spa_refcount
107 * The namespace lock must be acquired to increase the refcount from 0
108 * or to check if it is zero.
110 * spa_refcount -> spa_config_lock[]
112 * There must be at least one valid reference on the spa_t to acquire
115 * spa_namespace_lock -> spa_config_lock[]
117 * The namespace lock must always be taken before the config lock.
120 * The spa_namespace_lock can be acquired directly and is globally visible.
122 * The namespace is manipulated using the following functions, all of which
123 * require the spa_namespace_lock to be held.
125 * spa_lookup() Lookup a spa_t by name.
127 * spa_add() Create a new spa_t in the namespace.
129 * spa_remove() Remove a spa_t from the namespace. This also
130 * frees up any memory associated with the spa_t.
132 * spa_next() Returns the next spa_t in the system, or the
133 * first if NULL is passed.
135 * spa_evict_all() Shutdown and remove all spa_t structures in
138 * spa_guid_exists() Determine whether a pool/device guid exists.
140 * The spa_refcount is manipulated using the following functions:
142 * spa_open_ref() Adds a reference to the given spa_t. Must be
143 * called with spa_namespace_lock held if the
144 * refcount is currently zero.
146 * spa_close() Remove a reference from the spa_t. This will
147 * not free the spa_t or remove it from the
148 * namespace. No locking is required.
150 * spa_refcount_zero() Returns true if the refcount is currently
151 * zero. Must be called with spa_namespace_lock
154 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
155 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
156 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
158 * To read the configuration, it suffices to hold one of these locks as reader.
159 * To modify the configuration, you must hold all locks as writer. To modify
160 * vdev state without altering the vdev tree's topology (e.g. online/offline),
161 * you must hold SCL_STATE and SCL_ZIO as writer.
163 * We use these distinct config locks to avoid recursive lock entry.
164 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
165 * block allocations (SCL_ALLOC), which may require reading space maps
166 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
168 * The spa config locks cannot be normal rwlocks because we need the
169 * ability to hand off ownership. For example, SCL_ZIO is acquired
170 * by the issuing thread and later released by an interrupt thread.
171 * They do, however, obey the usual write-wanted semantics to prevent
172 * writer (i.e. system administrator) starvation.
174 * The lock acquisition rules are as follows:
177 * Protects changes to the vdev tree topology, such as vdev
178 * add/remove/attach/detach. Protects the dirty config list
179 * (spa_config_dirty_list) and the set of spares and l2arc devices.
182 * Protects changes to pool state and vdev state, such as vdev
183 * online/offline/fault/degrade/clear. Protects the dirty state list
184 * (spa_state_dirty_list) and global pool state (spa_state).
187 * Protects changes to metaslab groups and classes.
188 * Held as reader by metaslab_alloc() and metaslab_claim().
191 * Held by bp-level zios (those which have no io_vd upon entry)
192 * to prevent changes to the vdev tree. The bp-level zio implicitly
193 * protects all of its vdev child zios, which do not hold SCL_ZIO.
196 * Protects changes to metaslab groups and classes.
197 * Held as reader by metaslab_free(). SCL_FREE is distinct from
198 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
199 * blocks in zio_done() while another i/o that holds either
200 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
203 * Held as reader to prevent changes to the vdev tree during trivial
204 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
205 * other locks, and lower than all of them, to ensure that it's safe
206 * to acquire regardless of caller context.
208 * In addition, the following rules apply:
210 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
211 * The lock ordering is SCL_CONFIG > spa_props_lock.
213 * (b) I/O operations on leaf vdevs. For any zio operation that takes
214 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
215 * or zio_write_phys() -- the caller must ensure that the config cannot
216 * cannot change in the interim, and that the vdev cannot be reopened.
217 * SCL_STATE as reader suffices for both.
219 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
221 * spa_vdev_enter() Acquire the namespace lock and the config lock
224 * spa_vdev_exit() Release the config lock, wait for all I/O
225 * to complete, sync the updated configs to the
226 * cache, and release the namespace lock.
228 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
229 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
230 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
232 * spa_rename() is also implemented within this file since it requires
233 * manipulation of the namespace.
236 static avl_tree_t spa_namespace_avl;
237 kmutex_t spa_namespace_lock;
238 static kcondvar_t spa_namespace_cv;
239 static int spa_active_count;
240 int spa_max_replication_override = SPA_DVAS_PER_BP;
242 static kmutex_t spa_spare_lock;
243 static avl_tree_t spa_spare_avl;
244 static kmutex_t spa_l2cache_lock;
245 static avl_tree_t spa_l2cache_avl;
247 kmem_cache_t *spa_buffer_pool;
251 /* Everything except dprintf and spa is on by default in debug builds */
252 int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA);
258 * zfs_recover can be set to nonzero to attempt to recover from
259 * otherwise-fatal errors, typically caused by on-disk corruption. When
260 * set, calls to zfs_panic_recover() will turn into warning messages.
261 * This should only be used as a last resort, as it typically results
262 * in leaked space, or worse.
264 boolean_t zfs_recover = B_FALSE;
267 * If destroy encounters an EIO while reading metadata (e.g. indirect
268 * blocks), space referenced by the missing metadata can not be freed.
269 * Normally this causes the background destroy to become "stalled", as
270 * it is unable to make forward progress. While in this stalled state,
271 * all remaining space to free from the error-encountering filesystem is
272 * "temporarily leaked". Set this flag to cause it to ignore the EIO,
273 * permanently leak the space from indirect blocks that can not be read,
274 * and continue to free everything else that it can.
276 * The default, "stalling" behavior is useful if the storage partially
277 * fails (i.e. some but not all i/os fail), and then later recovers. In
278 * this case, we will be able to continue pool operations while it is
279 * partially failed, and when it recovers, we can continue to free the
280 * space, with no leaks. However, note that this case is actually
283 * Typically pools either (a) fail completely (but perhaps temporarily,
284 * e.g. a top-level vdev going offline), or (b) have localized,
285 * permanent errors (e.g. disk returns the wrong data due to bit flip or
286 * firmware bug). In case (a), this setting does not matter because the
287 * pool will be suspended and the sync thread will not be able to make
288 * forward progress regardless. In case (b), because the error is
289 * permanent, the best we can do is leak the minimum amount of space,
290 * which is what setting this flag will do. Therefore, it is reasonable
291 * for this flag to normally be set, but we chose the more conservative
292 * approach of not setting it, so that there is no possibility of
293 * leaking space in the "partial temporary" failure case.
295 boolean_t zfs_free_leak_on_eio = B_FALSE;
298 * Expiration time in milliseconds. This value has two meanings. First it is
299 * used to determine when the spa_deadman() logic should fire. By default the
300 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
301 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
302 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
305 uint64_t zfs_deadman_synctime_ms = 1000000ULL;
308 * Check time in milliseconds. This defines the frequency at which we check
311 uint64_t zfs_deadman_checktime_ms = 5000ULL;
314 * Default value of -1 for zfs_deadman_enabled is resolved in
317 int zfs_deadman_enabled = -1;
320 * The worst case is single-sector max-parity RAID-Z blocks, in which
321 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
322 * times the size; so just assume that. Add to this the fact that
323 * we can have up to 3 DVAs per bp, and one more factor of 2 because
324 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together,
326 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
328 int spa_asize_inflation = 24;
330 #if defined(__FreeBSD__) && defined(_KERNEL)
331 SYSCTL_DECL(_vfs_zfs);
332 SYSCTL_INT(_vfs_zfs, OID_AUTO, recover, CTLFLAG_RWTUN, &zfs_recover, 0,
333 "Try to recover from otherwise-fatal errors.");
336 sysctl_vfs_zfs_debug_flags(SYSCTL_HANDLER_ARGS)
341 err = sysctl_handle_int(oidp, &val, 0, req);
342 if (err != 0 || req->newptr == NULL)
346 * ZFS_DEBUG_MODIFY must be enabled prior to boot so all
347 * arc buffers in the system have the necessary additional
348 * checksum data. However, it is safe to disable at any
351 if (!(zfs_flags & ZFS_DEBUG_MODIFY))
352 val &= ~ZFS_DEBUG_MODIFY;
358 SYSCTL_PROC(_vfs_zfs, OID_AUTO, debugflags,
359 CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RWTUN, 0, sizeof(int),
360 sysctl_vfs_zfs_debug_flags, "IU", "Debug flags for ZFS testing.");
362 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_synctime_ms, CTLFLAG_RDTUN,
363 &zfs_deadman_synctime_ms, 0,
364 "Stalled ZFS I/O expiration time in milliseconds");
365 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_checktime_ms, CTLFLAG_RDTUN,
366 &zfs_deadman_checktime_ms, 0,
367 "Period of checks for stalled ZFS I/O in milliseconds");
368 SYSCTL_INT(_vfs_zfs, OID_AUTO, deadman_enabled, CTLFLAG_RDTUN,
369 &zfs_deadman_enabled, 0, "Kernel panic on stalled ZFS I/O");
370 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_asize_inflation, CTLFLAG_RWTUN,
371 &spa_asize_inflation, 0, "Worst case inflation factor for single sector writes");
380 * If we are not i386 or amd64 or in a virtual machine,
381 * disable ZFS deadman thread by default
383 if (zfs_deadman_enabled == -1) {
384 #if defined(__amd64__) || defined(__i386__)
385 zfs_deadman_enabled = (vm_guest == VM_GUEST_NO) ? 1 : 0;
387 zfs_deadman_enabled = 0;
392 #endif /* !illumos */
395 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
396 * the pool to be consumed. This ensures that we don't run the pool
397 * completely out of space, due to unaccounted changes (e.g. to the MOS).
398 * It also limits the worst-case time to allocate space. If we have
399 * less than this amount of free space, most ZPL operations (e.g. write,
400 * create) will return ENOSPC.
402 * Certain operations (e.g. file removal, most administrative actions) can
403 * use half the slop space. They will only return ENOSPC if less than half
404 * the slop space is free. Typically, once the pool has less than the slop
405 * space free, the user will use these operations to free up space in the pool.
406 * These are the operations that call dsl_pool_adjustedsize() with the netfree
407 * argument set to TRUE.
409 * A very restricted set of operations are always permitted, regardless of
410 * the amount of free space. These are the operations that call
411 * dsl_sync_task(ZFS_SPACE_CHECK_NONE), e.g. "zfs destroy". If these
412 * operations result in a net increase in the amount of space used,
413 * it is possible to run the pool completely out of space, causing it to
414 * be permanently read-only.
416 * Note that on very small pools, the slop space will be larger than
417 * 3.2%, in an effort to have it be at least spa_min_slop (128MB),
418 * but we never allow it to be more than half the pool size.
420 * See also the comments in zfs_space_check_t.
422 int spa_slop_shift = 5;
423 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_slop_shift, CTLFLAG_RWTUN,
425 "Shift value of reserved space (1/(2^spa_slop_shift)).");
426 uint64_t spa_min_slop = 128 * 1024 * 1024;
427 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, spa_min_slop, CTLFLAG_RWTUN,
429 "Minimal value of reserved space");
432 * ==========================================================================
434 * ==========================================================================
437 spa_config_lock_init(spa_t *spa)
439 for (int i = 0; i < SCL_LOCKS; i++) {
440 spa_config_lock_t *scl = &spa->spa_config_lock[i];
441 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
442 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
443 refcount_create_untracked(&scl->scl_count);
444 scl->scl_writer = NULL;
445 scl->scl_write_wanted = 0;
450 spa_config_lock_destroy(spa_t *spa)
452 for (int i = 0; i < SCL_LOCKS; i++) {
453 spa_config_lock_t *scl = &spa->spa_config_lock[i];
454 mutex_destroy(&scl->scl_lock);
455 cv_destroy(&scl->scl_cv);
456 refcount_destroy(&scl->scl_count);
457 ASSERT(scl->scl_writer == NULL);
458 ASSERT(scl->scl_write_wanted == 0);
463 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
465 for (int i = 0; i < SCL_LOCKS; i++) {
466 spa_config_lock_t *scl = &spa->spa_config_lock[i];
467 if (!(locks & (1 << i)))
469 mutex_enter(&scl->scl_lock);
470 if (rw == RW_READER) {
471 if (scl->scl_writer || scl->scl_write_wanted) {
472 mutex_exit(&scl->scl_lock);
473 spa_config_exit(spa, locks & ((1 << i) - 1),
478 ASSERT(scl->scl_writer != curthread);
479 if (!refcount_is_zero(&scl->scl_count)) {
480 mutex_exit(&scl->scl_lock);
481 spa_config_exit(spa, locks & ((1 << i) - 1),
485 scl->scl_writer = curthread;
487 (void) refcount_add(&scl->scl_count, tag);
488 mutex_exit(&scl->scl_lock);
494 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
498 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
500 for (int i = 0; i < SCL_LOCKS; i++) {
501 spa_config_lock_t *scl = &spa->spa_config_lock[i];
502 if (scl->scl_writer == curthread)
503 wlocks_held |= (1 << i);
504 if (!(locks & (1 << i)))
506 mutex_enter(&scl->scl_lock);
507 if (rw == RW_READER) {
508 while (scl->scl_writer || scl->scl_write_wanted) {
509 cv_wait(&scl->scl_cv, &scl->scl_lock);
512 ASSERT(scl->scl_writer != curthread);
513 while (!refcount_is_zero(&scl->scl_count)) {
514 scl->scl_write_wanted++;
515 cv_wait(&scl->scl_cv, &scl->scl_lock);
516 scl->scl_write_wanted--;
518 scl->scl_writer = curthread;
520 (void) refcount_add(&scl->scl_count, tag);
521 mutex_exit(&scl->scl_lock);
523 ASSERT(wlocks_held <= locks);
527 spa_config_exit(spa_t *spa, int locks, void *tag)
529 for (int i = SCL_LOCKS - 1; i >= 0; i--) {
530 spa_config_lock_t *scl = &spa->spa_config_lock[i];
531 if (!(locks & (1 << i)))
533 mutex_enter(&scl->scl_lock);
534 ASSERT(!refcount_is_zero(&scl->scl_count));
535 if (refcount_remove(&scl->scl_count, tag) == 0) {
536 ASSERT(scl->scl_writer == NULL ||
537 scl->scl_writer == curthread);
538 scl->scl_writer = NULL; /* OK in either case */
539 cv_broadcast(&scl->scl_cv);
541 mutex_exit(&scl->scl_lock);
546 spa_config_held(spa_t *spa, int locks, krw_t rw)
550 for (int i = 0; i < SCL_LOCKS; i++) {
551 spa_config_lock_t *scl = &spa->spa_config_lock[i];
552 if (!(locks & (1 << i)))
554 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
555 (rw == RW_WRITER && scl->scl_writer == curthread))
556 locks_held |= 1 << i;
563 * ==========================================================================
564 * SPA namespace functions
565 * ==========================================================================
569 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
570 * Returns NULL if no matching spa_t is found.
573 spa_lookup(const char *name)
575 static spa_t search; /* spa_t is large; don't allocate on stack */
580 ASSERT(MUTEX_HELD(&spa_namespace_lock));
582 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
585 * If it's a full dataset name, figure out the pool name and
588 cp = strpbrk(search.spa_name, "/@#");
592 spa = avl_find(&spa_namespace_avl, &search, &where);
598 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
599 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
600 * looking for potentially hung I/Os.
603 spa_deadman(void *arg, int pending)
608 * Disable the deadman timer if the pool is suspended.
610 if (spa_suspended(spa)) {
612 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
614 /* Nothing. just don't schedule any future callouts. */
619 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
620 (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
621 ++spa->spa_deadman_calls);
622 if (zfs_deadman_enabled)
623 vdev_deadman(spa->spa_root_vdev);
626 callout_schedule(&spa->spa_deadman_cycid,
627 hz * zfs_deadman_checktime_ms / MILLISEC);
632 #if defined(__FreeBSD__) && defined(_KERNEL)
634 spa_deadman_timeout(void *arg)
638 taskqueue_enqueue(taskqueue_thread, &spa->spa_deadman_task);
643 * Create an uninitialized spa_t with the given name. Requires
644 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
645 * exist by calling spa_lookup() first.
648 spa_add(const char *name, nvlist_t *config, const char *altroot)
651 spa_config_dirent_t *dp;
657 ASSERT(MUTEX_HELD(&spa_namespace_lock));
659 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
661 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
662 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
663 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
664 mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL);
665 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
666 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
667 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
668 mutex_init(&spa->spa_cksum_tmpls_lock, NULL, MUTEX_DEFAULT, NULL);
669 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
670 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
671 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
672 mutex_init(&spa->spa_alloc_lock, NULL, MUTEX_DEFAULT, NULL);
674 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
675 cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL);
676 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
677 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
678 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
680 for (int t = 0; t < TXG_SIZE; t++)
681 bplist_create(&spa->spa_free_bplist[t]);
683 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
684 spa->spa_state = POOL_STATE_UNINITIALIZED;
685 spa->spa_freeze_txg = UINT64_MAX;
686 spa->spa_final_txg = UINT64_MAX;
687 spa->spa_load_max_txg = UINT64_MAX;
689 spa->spa_proc_state = SPA_PROC_NONE;
692 hdlr.cyh_func = spa_deadman;
694 hdlr.cyh_level = CY_LOW_LEVEL;
697 spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
701 * This determines how often we need to check for hung I/Os after
702 * the cyclic has already fired. Since checking for hung I/Os is
703 * an expensive operation we don't want to check too frequently.
704 * Instead wait for 5 seconds before checking again.
706 when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms);
707 when.cyt_when = CY_INFINITY;
708 mutex_enter(&cpu_lock);
709 spa->spa_deadman_cycid = cyclic_add(&hdlr, &when);
710 mutex_exit(&cpu_lock);
714 * callout(9) does not provide a way to initialize a callout with
715 * a function and an argument, so we use callout_reset() to schedule
716 * the callout in the very distant future. Even if that event ever
717 * fires, it should be okayas we won't have any active zio-s.
718 * But normally spa_sync() will reschedule the callout with a proper
720 * callout(9) does not allow the callback function to sleep but
721 * vdev_deadman() needs to acquire vq_lock and illumos mutexes are
722 * emulated using sx(9). For this reason spa_deadman_timeout()
723 * will schedule spa_deadman() as task on a taskqueue that allows
726 TASK_INIT(&spa->spa_deadman_task, 0, spa_deadman, spa);
727 callout_init(&spa->spa_deadman_cycid, 1);
728 callout_reset_sbt(&spa->spa_deadman_cycid, SBT_MAX, 0,
729 spa_deadman_timeout, spa, 0);
732 refcount_create(&spa->spa_refcount);
733 spa_config_lock_init(spa);
735 avl_add(&spa_namespace_avl, spa);
738 * Set the alternate root, if there is one.
741 spa->spa_root = spa_strdup(altroot);
745 avl_create(&spa->spa_alloc_tree, zio_bookmark_compare,
746 sizeof (zio_t), offsetof(zio_t, io_alloc_node));
749 * Every pool starts with the default cachefile
751 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
752 offsetof(spa_config_dirent_t, scd_link));
754 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
755 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
756 list_insert_head(&spa->spa_config_list, dp);
758 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
761 if (config != NULL) {
764 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
766 VERIFY(nvlist_dup(features, &spa->spa_label_features,
770 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
773 if (spa->spa_label_features == NULL) {
774 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
778 spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0);
780 spa->spa_min_ashift = INT_MAX;
781 spa->spa_max_ashift = 0;
784 * As a pool is being created, treat all features as disabled by
785 * setting SPA_FEATURE_DISABLED for all entries in the feature
788 for (int i = 0; i < SPA_FEATURES; i++) {
789 spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED;
796 * Removes a spa_t from the namespace, freeing up any memory used. Requires
797 * spa_namespace_lock. This is called only after the spa_t has been closed and
801 spa_remove(spa_t *spa)
803 spa_config_dirent_t *dp;
805 ASSERT(MUTEX_HELD(&spa_namespace_lock));
806 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
807 ASSERT3U(refcount_count(&spa->spa_refcount), ==, 0);
809 nvlist_free(spa->spa_config_splitting);
811 avl_remove(&spa_namespace_avl, spa);
812 cv_broadcast(&spa_namespace_cv);
815 spa_strfree(spa->spa_root);
819 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
820 list_remove(&spa->spa_config_list, dp);
821 if (dp->scd_path != NULL)
822 spa_strfree(dp->scd_path);
823 kmem_free(dp, sizeof (spa_config_dirent_t));
826 avl_destroy(&spa->spa_alloc_tree);
827 list_destroy(&spa->spa_config_list);
829 nvlist_free(spa->spa_label_features);
830 nvlist_free(spa->spa_load_info);
831 spa_config_set(spa, NULL);
834 mutex_enter(&cpu_lock);
835 if (spa->spa_deadman_cycid != CYCLIC_NONE)
836 cyclic_remove(spa->spa_deadman_cycid);
837 mutex_exit(&cpu_lock);
838 spa->spa_deadman_cycid = CYCLIC_NONE;
841 callout_drain(&spa->spa_deadman_cycid);
842 taskqueue_drain(taskqueue_thread, &spa->spa_deadman_task);
846 refcount_destroy(&spa->spa_refcount);
848 spa_config_lock_destroy(spa);
850 for (int t = 0; t < TXG_SIZE; t++)
851 bplist_destroy(&spa->spa_free_bplist[t]);
853 zio_checksum_templates_free(spa);
855 cv_destroy(&spa->spa_async_cv);
856 cv_destroy(&spa->spa_evicting_os_cv);
857 cv_destroy(&spa->spa_proc_cv);
858 cv_destroy(&spa->spa_scrub_io_cv);
859 cv_destroy(&spa->spa_suspend_cv);
861 mutex_destroy(&spa->spa_alloc_lock);
862 mutex_destroy(&spa->spa_async_lock);
863 mutex_destroy(&spa->spa_errlist_lock);
864 mutex_destroy(&spa->spa_errlog_lock);
865 mutex_destroy(&spa->spa_evicting_os_lock);
866 mutex_destroy(&spa->spa_history_lock);
867 mutex_destroy(&spa->spa_proc_lock);
868 mutex_destroy(&spa->spa_props_lock);
869 mutex_destroy(&spa->spa_cksum_tmpls_lock);
870 mutex_destroy(&spa->spa_scrub_lock);
871 mutex_destroy(&spa->spa_suspend_lock);
872 mutex_destroy(&spa->spa_vdev_top_lock);
874 kmem_free(spa, sizeof (spa_t));
878 * Given a pool, return the next pool in the namespace, or NULL if there is
879 * none. If 'prev' is NULL, return the first pool.
882 spa_next(spa_t *prev)
884 ASSERT(MUTEX_HELD(&spa_namespace_lock));
887 return (AVL_NEXT(&spa_namespace_avl, prev));
889 return (avl_first(&spa_namespace_avl));
893 * ==========================================================================
894 * SPA refcount functions
895 * ==========================================================================
899 * Add a reference to the given spa_t. Must have at least one reference, or
900 * have the namespace lock held.
903 spa_open_ref(spa_t *spa, void *tag)
905 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
906 MUTEX_HELD(&spa_namespace_lock));
907 (void) refcount_add(&spa->spa_refcount, tag);
911 * Remove a reference to the given spa_t. Must have at least one reference, or
912 * have the namespace lock held.
915 spa_close(spa_t *spa, void *tag)
917 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
918 MUTEX_HELD(&spa_namespace_lock));
919 (void) refcount_remove(&spa->spa_refcount, tag);
923 * Remove a reference to the given spa_t held by a dsl dir that is
924 * being asynchronously released. Async releases occur from a taskq
925 * performing eviction of dsl datasets and dirs. The namespace lock
926 * isn't held and the hold by the object being evicted may contribute to
927 * spa_minref (e.g. dataset or directory released during pool export),
928 * so the asserts in spa_close() do not apply.
931 spa_async_close(spa_t *spa, void *tag)
933 (void) refcount_remove(&spa->spa_refcount, tag);
937 * Check to see if the spa refcount is zero. Must be called with
938 * spa_namespace_lock held. We really compare against spa_minref, which is the
939 * number of references acquired when opening a pool
942 spa_refcount_zero(spa_t *spa)
944 ASSERT(MUTEX_HELD(&spa_namespace_lock));
946 return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
950 * ==========================================================================
951 * SPA spare and l2cache tracking
952 * ==========================================================================
956 * Hot spares and cache devices are tracked using the same code below,
957 * for 'auxiliary' devices.
960 typedef struct spa_aux {
968 spa_aux_compare(const void *a, const void *b)
970 const spa_aux_t *sa = a;
971 const spa_aux_t *sb = b;
973 if (sa->aux_guid < sb->aux_guid)
975 else if (sa->aux_guid > sb->aux_guid)
982 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
988 search.aux_guid = vd->vdev_guid;
989 if ((aux = avl_find(avl, &search, &where)) != NULL) {
992 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
993 aux->aux_guid = vd->vdev_guid;
995 avl_insert(avl, aux, where);
1000 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
1006 search.aux_guid = vd->vdev_guid;
1007 aux = avl_find(avl, &search, &where);
1009 ASSERT(aux != NULL);
1011 if (--aux->aux_count == 0) {
1012 avl_remove(avl, aux);
1013 kmem_free(aux, sizeof (spa_aux_t));
1014 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
1015 aux->aux_pool = 0ULL;
1020 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
1022 spa_aux_t search, *found;
1024 search.aux_guid = guid;
1025 found = avl_find(avl, &search, NULL);
1029 *pool = found->aux_pool;
1036 *refcnt = found->aux_count;
1041 return (found != NULL);
1045 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
1047 spa_aux_t search, *found;
1050 search.aux_guid = vd->vdev_guid;
1051 found = avl_find(avl, &search, &where);
1052 ASSERT(found != NULL);
1053 ASSERT(found->aux_pool == 0ULL);
1055 found->aux_pool = spa_guid(vd->vdev_spa);
1059 * Spares are tracked globally due to the following constraints:
1061 * - A spare may be part of multiple pools.
1062 * - A spare may be added to a pool even if it's actively in use within
1064 * - A spare in use in any pool can only be the source of a replacement if
1065 * the target is a spare in the same pool.
1067 * We keep track of all spares on the system through the use of a reference
1068 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
1069 * spare, then we bump the reference count in the AVL tree. In addition, we set
1070 * the 'vdev_isspare' member to indicate that the device is a spare (active or
1071 * inactive). When a spare is made active (used to replace a device in the
1072 * pool), we also keep track of which pool its been made a part of.
1074 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
1075 * called under the spa_namespace lock as part of vdev reconfiguration. The
1076 * separate spare lock exists for the status query path, which does not need to
1077 * be completely consistent with respect to other vdev configuration changes.
1081 spa_spare_compare(const void *a, const void *b)
1083 return (spa_aux_compare(a, b));
1087 spa_spare_add(vdev_t *vd)
1089 mutex_enter(&spa_spare_lock);
1090 ASSERT(!vd->vdev_isspare);
1091 spa_aux_add(vd, &spa_spare_avl);
1092 vd->vdev_isspare = B_TRUE;
1093 mutex_exit(&spa_spare_lock);
1097 spa_spare_remove(vdev_t *vd)
1099 mutex_enter(&spa_spare_lock);
1100 ASSERT(vd->vdev_isspare);
1101 spa_aux_remove(vd, &spa_spare_avl);
1102 vd->vdev_isspare = B_FALSE;
1103 mutex_exit(&spa_spare_lock);
1107 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
1111 mutex_enter(&spa_spare_lock);
1112 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
1113 mutex_exit(&spa_spare_lock);
1119 spa_spare_activate(vdev_t *vd)
1121 mutex_enter(&spa_spare_lock);
1122 ASSERT(vd->vdev_isspare);
1123 spa_aux_activate(vd, &spa_spare_avl);
1124 mutex_exit(&spa_spare_lock);
1128 * Level 2 ARC devices are tracked globally for the same reasons as spares.
1129 * Cache devices currently only support one pool per cache device, and so
1130 * for these devices the aux reference count is currently unused beyond 1.
1134 spa_l2cache_compare(const void *a, const void *b)
1136 return (spa_aux_compare(a, b));
1140 spa_l2cache_add(vdev_t *vd)
1142 mutex_enter(&spa_l2cache_lock);
1143 ASSERT(!vd->vdev_isl2cache);
1144 spa_aux_add(vd, &spa_l2cache_avl);
1145 vd->vdev_isl2cache = B_TRUE;
1146 mutex_exit(&spa_l2cache_lock);
1150 spa_l2cache_remove(vdev_t *vd)
1152 mutex_enter(&spa_l2cache_lock);
1153 ASSERT(vd->vdev_isl2cache);
1154 spa_aux_remove(vd, &spa_l2cache_avl);
1155 vd->vdev_isl2cache = B_FALSE;
1156 mutex_exit(&spa_l2cache_lock);
1160 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
1164 mutex_enter(&spa_l2cache_lock);
1165 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
1166 mutex_exit(&spa_l2cache_lock);
1172 spa_l2cache_activate(vdev_t *vd)
1174 mutex_enter(&spa_l2cache_lock);
1175 ASSERT(vd->vdev_isl2cache);
1176 spa_aux_activate(vd, &spa_l2cache_avl);
1177 mutex_exit(&spa_l2cache_lock);
1181 * ==========================================================================
1183 * ==========================================================================
1187 * Lock the given spa_t for the purpose of adding or removing a vdev.
1188 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1189 * It returns the next transaction group for the spa_t.
1192 spa_vdev_enter(spa_t *spa)
1194 mutex_enter(&spa->spa_vdev_top_lock);
1195 mutex_enter(&spa_namespace_lock);
1196 return (spa_vdev_config_enter(spa));
1200 * Internal implementation for spa_vdev_enter(). Used when a vdev
1201 * operation requires multiple syncs (i.e. removing a device) while
1202 * keeping the spa_namespace_lock held.
1205 spa_vdev_config_enter(spa_t *spa)
1207 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1209 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1211 return (spa_last_synced_txg(spa) + 1);
1215 * Used in combination with spa_vdev_config_enter() to allow the syncing
1216 * of multiple transactions without releasing the spa_namespace_lock.
1219 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
1221 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1223 int config_changed = B_FALSE;
1225 ASSERT(txg > spa_last_synced_txg(spa));
1227 spa->spa_pending_vdev = NULL;
1230 * Reassess the DTLs.
1232 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
1234 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
1235 config_changed = B_TRUE;
1236 spa->spa_config_generation++;
1240 * Verify the metaslab classes.
1242 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
1243 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
1245 spa_config_exit(spa, SCL_ALL, spa);
1248 * Panic the system if the specified tag requires it. This
1249 * is useful for ensuring that configurations are updated
1252 if (zio_injection_enabled)
1253 zio_handle_panic_injection(spa, tag, 0);
1256 * Note: this txg_wait_synced() is important because it ensures
1257 * that there won't be more than one config change per txg.
1258 * This allows us to use the txg as the generation number.
1261 txg_wait_synced(spa->spa_dsl_pool, txg);
1264 ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
1265 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1267 spa_config_exit(spa, SCL_ALL, spa);
1271 * If the config changed, update the config cache.
1274 spa_config_sync(spa, B_FALSE, B_TRUE);
1278 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1279 * locking of spa_vdev_enter(), we also want make sure the transactions have
1280 * synced to disk, and then update the global configuration cache with the new
1284 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1286 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1287 mutex_exit(&spa_namespace_lock);
1288 mutex_exit(&spa->spa_vdev_top_lock);
1294 * Lock the given spa_t for the purpose of changing vdev state.
1297 spa_vdev_state_enter(spa_t *spa, int oplocks)
1299 int locks = SCL_STATE_ALL | oplocks;
1302 * Root pools may need to read of the underlying devfs filesystem
1303 * when opening up a vdev. Unfortunately if we're holding the
1304 * SCL_ZIO lock it will result in a deadlock when we try to issue
1305 * the read from the root filesystem. Instead we "prefetch"
1306 * the associated vnodes that we need prior to opening the
1307 * underlying devices and cache them so that we can prevent
1308 * any I/O when we are doing the actual open.
1310 if (spa_is_root(spa)) {
1311 int low = locks & ~(SCL_ZIO - 1);
1312 int high = locks & ~low;
1314 spa_config_enter(spa, high, spa, RW_WRITER);
1315 vdev_hold(spa->spa_root_vdev);
1316 spa_config_enter(spa, low, spa, RW_WRITER);
1318 spa_config_enter(spa, locks, spa, RW_WRITER);
1320 spa->spa_vdev_locks = locks;
1324 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1326 boolean_t config_changed = B_FALSE;
1328 if (vd != NULL || error == 0)
1329 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1333 vdev_state_dirty(vd->vdev_top);
1334 config_changed = B_TRUE;
1335 spa->spa_config_generation++;
1338 if (spa_is_root(spa))
1339 vdev_rele(spa->spa_root_vdev);
1341 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1342 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1345 * If anything changed, wait for it to sync. This ensures that,
1346 * from the system administrator's perspective, zpool(1M) commands
1347 * are synchronous. This is important for things like zpool offline:
1348 * when the command completes, you expect no further I/O from ZFS.
1351 txg_wait_synced(spa->spa_dsl_pool, 0);
1354 * If the config changed, update the config cache.
1356 if (config_changed) {
1357 mutex_enter(&spa_namespace_lock);
1358 spa_config_sync(spa, B_FALSE, B_TRUE);
1359 mutex_exit(&spa_namespace_lock);
1366 * ==========================================================================
1367 * Miscellaneous functions
1368 * ==========================================================================
1372 spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx)
1374 if (!nvlist_exists(spa->spa_label_features, feature)) {
1375 fnvlist_add_boolean(spa->spa_label_features, feature);
1377 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1378 * dirty the vdev config because lock SCL_CONFIG is not held.
1379 * Thankfully, in this case we don't need to dirty the config
1380 * because it will be written out anyway when we finish
1381 * creating the pool.
1383 if (tx->tx_txg != TXG_INITIAL)
1384 vdev_config_dirty(spa->spa_root_vdev);
1389 spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1391 if (nvlist_remove_all(spa->spa_label_features, feature) == 0)
1392 vdev_config_dirty(spa->spa_root_vdev);
1399 spa_rename(const char *name, const char *newname)
1405 * Lookup the spa_t and grab the config lock for writing. We need to
1406 * actually open the pool so that we can sync out the necessary labels.
1407 * It's OK to call spa_open() with the namespace lock held because we
1408 * allow recursive calls for other reasons.
1410 mutex_enter(&spa_namespace_lock);
1411 if ((err = spa_open(name, &spa, FTAG)) != 0) {
1412 mutex_exit(&spa_namespace_lock);
1416 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1418 avl_remove(&spa_namespace_avl, spa);
1419 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1420 avl_add(&spa_namespace_avl, spa);
1423 * Sync all labels to disk with the new names by marking the root vdev
1424 * dirty and waiting for it to sync. It will pick up the new pool name
1427 vdev_config_dirty(spa->spa_root_vdev);
1429 spa_config_exit(spa, SCL_ALL, FTAG);
1431 txg_wait_synced(spa->spa_dsl_pool, 0);
1434 * Sync the updated config cache.
1436 spa_config_sync(spa, B_FALSE, B_TRUE);
1438 spa_close(spa, FTAG);
1440 mutex_exit(&spa_namespace_lock);
1446 * Return the spa_t associated with given pool_guid, if it exists. If
1447 * device_guid is non-zero, determine whether the pool exists *and* contains
1448 * a device with the specified device_guid.
1451 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1454 avl_tree_t *t = &spa_namespace_avl;
1456 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1458 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1459 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1461 if (spa->spa_root_vdev == NULL)
1463 if (spa_guid(spa) == pool_guid) {
1464 if (device_guid == 0)
1467 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1468 device_guid) != NULL)
1472 * Check any devices we may be in the process of adding.
1474 if (spa->spa_pending_vdev) {
1475 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1476 device_guid) != NULL)
1486 * Determine whether a pool with the given pool_guid exists.
1489 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1491 return (spa_by_guid(pool_guid, device_guid) != NULL);
1495 spa_strdup(const char *s)
1501 new = kmem_alloc(len + 1, KM_SLEEP);
1509 spa_strfree(char *s)
1511 kmem_free(s, strlen(s) + 1);
1515 spa_get_random(uint64_t range)
1521 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1527 spa_generate_guid(spa_t *spa)
1529 uint64_t guid = spa_get_random(-1ULL);
1532 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1533 guid = spa_get_random(-1ULL);
1535 while (guid == 0 || spa_guid_exists(guid, 0))
1536 guid = spa_get_random(-1ULL);
1543 snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp)
1546 char *checksum = NULL;
1547 char *compress = NULL;
1550 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1551 dmu_object_byteswap_t bswap =
1552 DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1553 (void) snprintf(type, sizeof (type), "bswap %s %s",
1554 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1555 "metadata" : "data",
1556 dmu_ot_byteswap[bswap].ob_name);
1558 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1561 if (!BP_IS_EMBEDDED(bp)) {
1563 zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1565 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1568 SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum,
1573 spa_freeze(spa_t *spa)
1575 uint64_t freeze_txg = 0;
1577 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1578 if (spa->spa_freeze_txg == UINT64_MAX) {
1579 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1580 spa->spa_freeze_txg = freeze_txg;
1582 spa_config_exit(spa, SCL_ALL, FTAG);
1583 if (freeze_txg != 0)
1584 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1588 zfs_panic_recover(const char *fmt, ...)
1593 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1598 * This is a stripped-down version of strtoull, suitable only for converting
1599 * lowercase hexadecimal numbers that don't overflow.
1602 zfs_strtonum(const char *str, char **nptr)
1608 while ((c = *str) != '\0') {
1609 if (c >= '0' && c <= '9')
1611 else if (c >= 'a' && c <= 'f')
1612 digit = 10 + c - 'a';
1623 *nptr = (char *)str;
1629 * ==========================================================================
1630 * Accessor functions
1631 * ==========================================================================
1635 spa_shutting_down(spa_t *spa)
1637 return (spa->spa_async_suspended);
1641 spa_get_dsl(spa_t *spa)
1643 return (spa->spa_dsl_pool);
1647 spa_is_initializing(spa_t *spa)
1649 return (spa->spa_is_initializing);
1653 spa_get_rootblkptr(spa_t *spa)
1655 return (&spa->spa_ubsync.ub_rootbp);
1659 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1661 spa->spa_uberblock.ub_rootbp = *bp;
1665 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1667 if (spa->spa_root == NULL)
1670 (void) strncpy(buf, spa->spa_root, buflen);
1674 spa_sync_pass(spa_t *spa)
1676 return (spa->spa_sync_pass);
1680 spa_name(spa_t *spa)
1682 return (spa->spa_name);
1686 spa_guid(spa_t *spa)
1688 dsl_pool_t *dp = spa_get_dsl(spa);
1692 * If we fail to parse the config during spa_load(), we can go through
1693 * the error path (which posts an ereport) and end up here with no root
1694 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1697 if (spa->spa_root_vdev == NULL)
1698 return (spa->spa_config_guid);
1700 guid = spa->spa_last_synced_guid != 0 ?
1701 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1704 * Return the most recently synced out guid unless we're
1705 * in syncing context.
1707 if (dp && dsl_pool_sync_context(dp))
1708 return (spa->spa_root_vdev->vdev_guid);
1714 spa_load_guid(spa_t *spa)
1717 * This is a GUID that exists solely as a reference for the
1718 * purposes of the arc. It is generated at load time, and
1719 * is never written to persistent storage.
1721 return (spa->spa_load_guid);
1725 spa_last_synced_txg(spa_t *spa)
1727 return (spa->spa_ubsync.ub_txg);
1731 spa_first_txg(spa_t *spa)
1733 return (spa->spa_first_txg);
1737 spa_syncing_txg(spa_t *spa)
1739 return (spa->spa_syncing_txg);
1743 * Return the last txg where data can be dirtied. The final txgs
1744 * will be used to just clear out any deferred frees that remain.
1747 spa_final_dirty_txg(spa_t *spa)
1749 return (spa->spa_final_txg - TXG_DEFER_SIZE);
1753 spa_state(spa_t *spa)
1755 return (spa->spa_state);
1759 spa_load_state(spa_t *spa)
1761 return (spa->spa_load_state);
1765 spa_freeze_txg(spa_t *spa)
1767 return (spa->spa_freeze_txg);
1772 spa_get_worst_case_asize(spa_t *spa, uint64_t lsize)
1774 return (lsize * spa_asize_inflation);
1778 * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%),
1779 * or at least 128MB, unless that would cause it to be more than half the
1782 * See the comment above spa_slop_shift for details.
1785 spa_get_slop_space(spa_t *spa)
1787 uint64_t space = spa_get_dspace(spa);
1788 return (MAX(space >> spa_slop_shift, MIN(space >> 1, spa_min_slop)));
1792 spa_get_dspace(spa_t *spa)
1794 return (spa->spa_dspace);
1798 spa_update_dspace(spa_t *spa)
1800 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1801 ddt_get_dedup_dspace(spa);
1805 * Return the failure mode that has been set to this pool. The default
1806 * behavior will be to block all I/Os when a complete failure occurs.
1809 spa_get_failmode(spa_t *spa)
1811 return (spa->spa_failmode);
1815 spa_suspended(spa_t *spa)
1817 return (spa->spa_suspended);
1821 spa_version(spa_t *spa)
1823 return (spa->spa_ubsync.ub_version);
1827 spa_deflate(spa_t *spa)
1829 return (spa->spa_deflate);
1833 spa_normal_class(spa_t *spa)
1835 return (spa->spa_normal_class);
1839 spa_log_class(spa_t *spa)
1841 return (spa->spa_log_class);
1845 spa_evicting_os_register(spa_t *spa, objset_t *os)
1847 mutex_enter(&spa->spa_evicting_os_lock);
1848 list_insert_head(&spa->spa_evicting_os_list, os);
1849 mutex_exit(&spa->spa_evicting_os_lock);
1853 spa_evicting_os_deregister(spa_t *spa, objset_t *os)
1855 mutex_enter(&spa->spa_evicting_os_lock);
1856 list_remove(&spa->spa_evicting_os_list, os);
1857 cv_broadcast(&spa->spa_evicting_os_cv);
1858 mutex_exit(&spa->spa_evicting_os_lock);
1862 spa_evicting_os_wait(spa_t *spa)
1864 mutex_enter(&spa->spa_evicting_os_lock);
1865 while (!list_is_empty(&spa->spa_evicting_os_list))
1866 cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock);
1867 mutex_exit(&spa->spa_evicting_os_lock);
1869 dmu_buf_user_evict_wait();
1873 spa_max_replication(spa_t *spa)
1876 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1877 * handle BPs with more than one DVA allocated. Set our max
1878 * replication level accordingly.
1880 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1882 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1886 spa_prev_software_version(spa_t *spa)
1888 return (spa->spa_prev_software_version);
1892 spa_deadman_synctime(spa_t *spa)
1894 return (spa->spa_deadman_synctime);
1898 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1900 uint64_t asize = DVA_GET_ASIZE(dva);
1901 uint64_t dsize = asize;
1903 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1905 if (asize != 0 && spa->spa_deflate) {
1906 uint64_t vdev = DVA_GET_VDEV(dva);
1907 vdev_t *vd = vdev_lookup_top(spa, vdev);
1910 "dva_get_dsize_sync(): bad DVA %llu:%llu",
1911 (u_longlong_t)vdev, (u_longlong_t)asize);
1913 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1920 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1924 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1925 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1931 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1935 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1937 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1938 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1940 spa_config_exit(spa, SCL_VDEV, FTAG);
1946 * ==========================================================================
1947 * Initialization and Termination
1948 * ==========================================================================
1952 spa_name_compare(const void *a1, const void *a2)
1954 const spa_t *s1 = a1;
1955 const spa_t *s2 = a2;
1958 s = strcmp(s1->spa_name, s2->spa_name);
1969 return (spa_active_count);
1979 EVENTHANDLER_DEFINE(mountroot, spa_boot_init, NULL, 0);
1985 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1986 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1987 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1988 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1990 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1991 offsetof(spa_t, spa_avl));
1993 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1994 offsetof(spa_aux_t, aux_avl));
1996 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1997 offsetof(spa_aux_t, aux_avl));
1999 spa_mode_global = mode;
2005 if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
2006 arc_procfd = open("/proc/self/ctl", O_WRONLY);
2007 if (arc_procfd == -1) {
2008 perror("could not enable watchpoints: "
2009 "opening /proc/self/ctl failed: ");
2015 #endif /* illumos */
2019 metaslab_alloc_trace_init();
2024 vdev_cache_stat_init();
2028 zpool_feature_init();
2035 #endif /* !illumos */
2046 vdev_cache_stat_fini();
2051 metaslab_alloc_trace_fini();
2056 avl_destroy(&spa_namespace_avl);
2057 avl_destroy(&spa_spare_avl);
2058 avl_destroy(&spa_l2cache_avl);
2060 cv_destroy(&spa_namespace_cv);
2061 mutex_destroy(&spa_namespace_lock);
2062 mutex_destroy(&spa_spare_lock);
2063 mutex_destroy(&spa_l2cache_lock);
2067 * Return whether this pool has slogs. No locking needed.
2068 * It's not a problem if the wrong answer is returned as it's only for
2069 * performance and not correctness
2072 spa_has_slogs(spa_t *spa)
2074 return (spa->spa_log_class->mc_rotor != NULL);
2078 spa_get_log_state(spa_t *spa)
2080 return (spa->spa_log_state);
2084 spa_set_log_state(spa_t *spa, spa_log_state_t state)
2086 spa->spa_log_state = state;
2090 spa_is_root(spa_t *spa)
2092 return (spa->spa_is_root);
2096 spa_writeable(spa_t *spa)
2098 return (!!(spa->spa_mode & FWRITE));
2102 * Returns true if there is a pending sync task in any of the current
2103 * syncing txg, the current quiescing txg, or the current open txg.
2106 spa_has_pending_synctask(spa_t *spa)
2108 return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks));
2112 spa_mode(spa_t *spa)
2114 return (spa->spa_mode);
2118 spa_bootfs(spa_t *spa)
2120 return (spa->spa_bootfs);
2124 spa_delegation(spa_t *spa)
2126 return (spa->spa_delegation);
2130 spa_meta_objset(spa_t *spa)
2132 return (spa->spa_meta_objset);
2136 spa_dedup_checksum(spa_t *spa)
2138 return (spa->spa_dedup_checksum);
2142 * Reset pool scan stat per scan pass (or reboot).
2145 spa_scan_stat_init(spa_t *spa)
2147 /* data not stored on disk */
2148 spa->spa_scan_pass_start = gethrestime_sec();
2149 if (dsl_scan_is_paused_scrub(spa->spa_dsl_pool->dp_scan))
2150 spa->spa_scan_pass_scrub_pause = spa->spa_scan_pass_start;
2152 spa->spa_scan_pass_scrub_pause = 0;
2153 spa->spa_scan_pass_scrub_spent_paused = 0;
2154 spa->spa_scan_pass_exam = 0;
2155 vdev_scan_stat_init(spa->spa_root_vdev);
2159 * Get scan stats for zpool status reports
2162 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
2164 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
2166 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
2167 return (SET_ERROR(ENOENT));
2168 bzero(ps, sizeof (pool_scan_stat_t));
2170 /* data stored on disk */
2171 ps->pss_func = scn->scn_phys.scn_func;
2172 ps->pss_start_time = scn->scn_phys.scn_start_time;
2173 ps->pss_end_time = scn->scn_phys.scn_end_time;
2174 ps->pss_to_examine = scn->scn_phys.scn_to_examine;
2175 ps->pss_examined = scn->scn_phys.scn_examined;
2176 ps->pss_to_process = scn->scn_phys.scn_to_process;
2177 ps->pss_processed = scn->scn_phys.scn_processed;
2178 ps->pss_errors = scn->scn_phys.scn_errors;
2179 ps->pss_state = scn->scn_phys.scn_state;
2181 /* data not stored on disk */
2182 ps->pss_pass_start = spa->spa_scan_pass_start;
2183 ps->pss_pass_exam = spa->spa_scan_pass_exam;
2184 ps->pss_pass_scrub_pause = spa->spa_scan_pass_scrub_pause;
2185 ps->pss_pass_scrub_spent_paused = spa->spa_scan_pass_scrub_spent_paused;
2191 spa_debug_enabled(spa_t *spa)
2193 return (spa->spa_debug);
2197 spa_maxblocksize(spa_t *spa)
2199 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS))
2200 return (SPA_MAXBLOCKSIZE);
2202 return (SPA_OLD_MAXBLOCKSIZE);