4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
24 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
25 * Copyright 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 * Copyright 2013 Saso Kiselkov. All rights reserved.
30 #include <sys/zfs_context.h>
31 #include <sys/spa_impl.h>
32 #include <sys/spa_boot.h>
34 #include <sys/zio_checksum.h>
35 #include <sys/zio_compress.h>
37 #include <sys/dmu_tx.h>
40 #include <sys/vdev_impl.h>
41 #include <sys/metaslab.h>
42 #include <sys/uberblock_impl.h>
45 #include <sys/unique.h>
46 #include <sys/dsl_pool.h>
47 #include <sys/dsl_dir.h>
48 #include <sys/dsl_prop.h>
49 #include <sys/dsl_scan.h>
50 #include <sys/fs/zfs.h>
51 #include <sys/metaslab_impl.h>
55 #include <sys/zfeature.h>
60 * There are four basic locks for managing spa_t structures:
62 * spa_namespace_lock (global mutex)
64 * This lock must be acquired to do any of the following:
66 * - Lookup a spa_t by name
67 * - Add or remove a spa_t from the namespace
68 * - Increase spa_refcount from non-zero
69 * - Check if spa_refcount is zero
71 * - add/remove/attach/detach devices
72 * - Held for the duration of create/destroy/import/export
74 * It does not need to handle recursion. A create or destroy may
75 * reference objects (files or zvols) in other pools, but by
76 * definition they must have an existing reference, and will never need
77 * to lookup a spa_t by name.
79 * spa_refcount (per-spa refcount_t protected by mutex)
81 * This reference count keep track of any active users of the spa_t. The
82 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
83 * the refcount is never really 'zero' - opening a pool implicitly keeps
84 * some references in the DMU. Internally we check against spa_minref, but
85 * present the image of a zero/non-zero value to consumers.
87 * spa_config_lock[] (per-spa array of rwlocks)
89 * This protects the spa_t from config changes, and must be held in
90 * the following circumstances:
92 * - RW_READER to perform I/O to the spa
93 * - RW_WRITER to change the vdev config
95 * The locking order is fairly straightforward:
97 * spa_namespace_lock -> spa_refcount
99 * The namespace lock must be acquired to increase the refcount from 0
100 * or to check if it is zero.
102 * spa_refcount -> spa_config_lock[]
104 * There must be at least one valid reference on the spa_t to acquire
107 * spa_namespace_lock -> spa_config_lock[]
109 * The namespace lock must always be taken before the config lock.
112 * The spa_namespace_lock can be acquired directly and is globally visible.
114 * The namespace is manipulated using the following functions, all of which
115 * require the spa_namespace_lock to be held.
117 * spa_lookup() Lookup a spa_t by name.
119 * spa_add() Create a new spa_t in the namespace.
121 * spa_remove() Remove a spa_t from the namespace. This also
122 * frees up any memory associated with the spa_t.
124 * spa_next() Returns the next spa_t in the system, or the
125 * first if NULL is passed.
127 * spa_evict_all() Shutdown and remove all spa_t structures in
130 * spa_guid_exists() Determine whether a pool/device guid exists.
132 * The spa_refcount is manipulated using the following functions:
134 * spa_open_ref() Adds a reference to the given spa_t. Must be
135 * called with spa_namespace_lock held if the
136 * refcount is currently zero.
138 * spa_close() Remove a reference from the spa_t. This will
139 * not free the spa_t or remove it from the
140 * namespace. No locking is required.
142 * spa_refcount_zero() Returns true if the refcount is currently
143 * zero. Must be called with spa_namespace_lock
146 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
147 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
148 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
150 * To read the configuration, it suffices to hold one of these locks as reader.
151 * To modify the configuration, you must hold all locks as writer. To modify
152 * vdev state without altering the vdev tree's topology (e.g. online/offline),
153 * you must hold SCL_STATE and SCL_ZIO as writer.
155 * We use these distinct config locks to avoid recursive lock entry.
156 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
157 * block allocations (SCL_ALLOC), which may require reading space maps
158 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
160 * The spa config locks cannot be normal rwlocks because we need the
161 * ability to hand off ownership. For example, SCL_ZIO is acquired
162 * by the issuing thread and later released by an interrupt thread.
163 * They do, however, obey the usual write-wanted semantics to prevent
164 * writer (i.e. system administrator) starvation.
166 * The lock acquisition rules are as follows:
169 * Protects changes to the vdev tree topology, such as vdev
170 * add/remove/attach/detach. Protects the dirty config list
171 * (spa_config_dirty_list) and the set of spares and l2arc devices.
174 * Protects changes to pool state and vdev state, such as vdev
175 * online/offline/fault/degrade/clear. Protects the dirty state list
176 * (spa_state_dirty_list) and global pool state (spa_state).
179 * Protects changes to metaslab groups and classes.
180 * Held as reader by metaslab_alloc() and metaslab_claim().
183 * Held by bp-level zios (those which have no io_vd upon entry)
184 * to prevent changes to the vdev tree. The bp-level zio implicitly
185 * protects all of its vdev child zios, which do not hold SCL_ZIO.
188 * Protects changes to metaslab groups and classes.
189 * Held as reader by metaslab_free(). SCL_FREE is distinct from
190 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
191 * blocks in zio_done() while another i/o that holds either
192 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
195 * Held as reader to prevent changes to the vdev tree during trivial
196 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
197 * other locks, and lower than all of them, to ensure that it's safe
198 * to acquire regardless of caller context.
200 * In addition, the following rules apply:
202 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
203 * The lock ordering is SCL_CONFIG > spa_props_lock.
205 * (b) I/O operations on leaf vdevs. For any zio operation that takes
206 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
207 * or zio_write_phys() -- the caller must ensure that the config cannot
208 * cannot change in the interim, and that the vdev cannot be reopened.
209 * SCL_STATE as reader suffices for both.
211 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
213 * spa_vdev_enter() Acquire the namespace lock and the config lock
216 * spa_vdev_exit() Release the config lock, wait for all I/O
217 * to complete, sync the updated configs to the
218 * cache, and release the namespace lock.
220 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
221 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
222 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
224 * spa_rename() is also implemented within this file since it requires
225 * manipulation of the namespace.
228 static avl_tree_t spa_namespace_avl;
229 kmutex_t spa_namespace_lock;
230 static kcondvar_t spa_namespace_cv;
231 static int spa_active_count;
232 int spa_max_replication_override = SPA_DVAS_PER_BP;
234 static kmutex_t spa_spare_lock;
235 static avl_tree_t spa_spare_avl;
236 static kmutex_t spa_l2cache_lock;
237 static avl_tree_t spa_l2cache_avl;
239 kmem_cache_t *spa_buffer_pool;
243 /* Everything except dprintf and spa is on by default in debug builds */
244 int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA);
249 TUNABLE_INT("debug.zfs_flags", &zfs_flags);
250 SYSCTL_INT(_debug, OID_AUTO, zfs_flags, CTLFLAG_RWTUN, &zfs_flags, 0,
254 * zfs_recover can be set to nonzero to attempt to recover from
255 * otherwise-fatal errors, typically caused by on-disk corruption. When
256 * set, calls to zfs_panic_recover() will turn into warning messages.
257 * This should only be used as a last resort, as it typically results
258 * in leaked space, or worse.
260 boolean_t zfs_recover = B_FALSE;
261 SYSCTL_DECL(_vfs_zfs);
262 TUNABLE_INT("vfs.zfs.recover", &zfs_recover);
263 SYSCTL_INT(_vfs_zfs, OID_AUTO, recover, CTLFLAG_RWTUN, &zfs_recover, 0,
264 "Try to recover from otherwise-fatal errors.");
267 sysctl_vfs_zfs_debug_flags(SYSCTL_HANDLER_ARGS)
272 err = sysctl_handle_int(oidp, &val, 0, req);
273 if (err != 0 || req->newptr == NULL)
277 * ZFS_DEBUG_MODIFY must be enabled prior to boot so all
278 * arc buffers in the system have the necessary additional
279 * checksum data. However, it is safe to disable at any
282 if (!(zfs_flags & ZFS_DEBUG_MODIFY))
283 val &= ~ZFS_DEBUG_MODIFY;
288 TUNABLE_INT("vfs.zfs.debug_flags", &zfs_flags);
289 SYSCTL_PROC(_vfs_zfs, OID_AUTO, debug_flags,
290 CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(int),
291 sysctl_vfs_zfs_debug_flags, "IU", "Debug flags for ZFS testing.");
294 * If destroy encounters an EIO while reading metadata (e.g. indirect
295 * blocks), space referenced by the missing metadata can not be freed.
296 * Normally this causes the background destroy to become "stalled", as
297 * it is unable to make forward progress. While in this stalled state,
298 * all remaining space to free from the error-encountering filesystem is
299 * "temporarily leaked". Set this flag to cause it to ignore the EIO,
300 * permanently leak the space from indirect blocks that can not be read,
301 * and continue to free everything else that it can.
303 * The default, "stalling" behavior is useful if the storage partially
304 * fails (i.e. some but not all i/os fail), and then later recovers. In
305 * this case, we will be able to continue pool operations while it is
306 * partially failed, and when it recovers, we can continue to free the
307 * space, with no leaks. However, note that this case is actually
310 * Typically pools either (a) fail completely (but perhaps temporarily,
311 * e.g. a top-level vdev going offline), or (b) have localized,
312 * permanent errors (e.g. disk returns the wrong data due to bit flip or
313 * firmware bug). In case (a), this setting does not matter because the
314 * pool will be suspended and the sync thread will not be able to make
315 * forward progress regardless. In case (b), because the error is
316 * permanent, the best we can do is leak the minimum amount of space,
317 * which is what setting this flag will do. Therefore, it is reasonable
318 * for this flag to normally be set, but we chose the more conservative
319 * approach of not setting it, so that there is no possibility of
320 * leaking space in the "partial temporary" failure case.
322 boolean_t zfs_free_leak_on_eio = B_FALSE;
325 * Expiration time in milliseconds. This value has two meanings. First it is
326 * used to determine when the spa_deadman() logic should fire. By default the
327 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
328 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
329 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
332 uint64_t zfs_deadman_synctime_ms = 1000000ULL;
333 TUNABLE_QUAD("vfs.zfs.deadman_synctime_ms", &zfs_deadman_synctime_ms);
334 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_synctime_ms, CTLFLAG_RDTUN,
335 &zfs_deadman_synctime_ms, 0,
336 "Stalled ZFS I/O expiration time in milliseconds");
339 * Check time in milliseconds. This defines the frequency at which we check
342 uint64_t zfs_deadman_checktime_ms = 5000ULL;
343 TUNABLE_QUAD("vfs.zfs.deadman_checktime_ms", &zfs_deadman_checktime_ms);
344 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_checktime_ms, CTLFLAG_RDTUN,
345 &zfs_deadman_checktime_ms, 0,
346 "Period of checks for stalled ZFS I/O in milliseconds");
349 * Default value of -1 for zfs_deadman_enabled is resolved in
352 int zfs_deadman_enabled = -1;
353 TUNABLE_INT("vfs.zfs.deadman_enabled", &zfs_deadman_enabled);
354 SYSCTL_INT(_vfs_zfs, OID_AUTO, deadman_enabled, CTLFLAG_RDTUN,
355 &zfs_deadman_enabled, 0, "Kernel panic on stalled ZFS I/O");
358 * The worst case is single-sector max-parity RAID-Z blocks, in which
359 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
360 * times the size; so just assume that. Add to this the fact that
361 * we can have up to 3 DVAs per bp, and one more factor of 2 because
362 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together,
364 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
366 int spa_asize_inflation = 24;
367 TUNABLE_INT("vfs.zfs.spa_asize_inflation", &spa_asize_inflation);
368 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_asize_inflation, CTLFLAG_RWTUN,
369 &spa_asize_inflation, 0, "Worst case inflation factor for single sector writes");
377 * If we are not i386 or amd64 or in a virtual machine,
378 * disable ZFS deadman thread by default
380 if (zfs_deadman_enabled == -1) {
381 #if defined(__amd64__) || defined(__i386__)
382 zfs_deadman_enabled = (vm_guest == VM_GUEST_NO) ? 1 : 0;
384 zfs_deadman_enabled = 0;
389 #endif /* !illumos */
392 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
393 * the pool to be consumed. This ensures that we don't run the pool
394 * completely out of space, due to unaccounted changes (e.g. to the MOS).
395 * It also limits the worst-case time to allocate space. If we have
396 * less than this amount of free space, most ZPL operations (e.g. write,
397 * create) will return ENOSPC.
399 * Certain operations (e.g. file removal, most administrative actions) can
400 * use half the slop space. They will only return ENOSPC if less than half
401 * the slop space is free. Typically, once the pool has less than the slop
402 * space free, the user will use these operations to free up space in the pool.
403 * These are the operations that call dsl_pool_adjustedsize() with the netfree
404 * argument set to TRUE.
406 * A very restricted set of operations are always permitted, regardless of
407 * the amount of free space. These are the operations that call
408 * dsl_sync_task(ZFS_SPACE_CHECK_NONE), e.g. "zfs destroy". If these
409 * operations result in a net increase in the amount of space used,
410 * it is possible to run the pool completely out of space, causing it to
411 * be permanently read-only.
413 * See also the comments in zfs_space_check_t.
415 int spa_slop_shift = 5;
416 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_slop_shift, CTLFLAG_RWTUN,
418 "Shift value of reserved space (1/(2^spa_slop_shift)).");
421 * ==========================================================================
423 * ==========================================================================
426 spa_config_lock_init(spa_t *spa)
428 for (int i = 0; i < SCL_LOCKS; i++) {
429 spa_config_lock_t *scl = &spa->spa_config_lock[i];
430 mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
431 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
432 refcount_create_untracked(&scl->scl_count);
433 scl->scl_writer = NULL;
434 scl->scl_write_wanted = 0;
439 spa_config_lock_destroy(spa_t *spa)
441 for (int i = 0; i < SCL_LOCKS; i++) {
442 spa_config_lock_t *scl = &spa->spa_config_lock[i];
443 mutex_destroy(&scl->scl_lock);
444 cv_destroy(&scl->scl_cv);
445 refcount_destroy(&scl->scl_count);
446 ASSERT(scl->scl_writer == NULL);
447 ASSERT(scl->scl_write_wanted == 0);
452 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
454 for (int i = 0; i < SCL_LOCKS; i++) {
455 spa_config_lock_t *scl = &spa->spa_config_lock[i];
456 if (!(locks & (1 << i)))
458 mutex_enter(&scl->scl_lock);
459 if (rw == RW_READER) {
460 if (scl->scl_writer || scl->scl_write_wanted) {
461 mutex_exit(&scl->scl_lock);
462 spa_config_exit(spa, locks & ((1 << i) - 1),
467 ASSERT(scl->scl_writer != curthread);
468 if (!refcount_is_zero(&scl->scl_count)) {
469 mutex_exit(&scl->scl_lock);
470 spa_config_exit(spa, locks & ((1 << i) - 1),
474 scl->scl_writer = curthread;
476 (void) refcount_add(&scl->scl_count, tag);
477 mutex_exit(&scl->scl_lock);
483 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
487 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
489 for (int i = 0; i < SCL_LOCKS; i++) {
490 spa_config_lock_t *scl = &spa->spa_config_lock[i];
491 if (scl->scl_writer == curthread)
492 wlocks_held |= (1 << i);
493 if (!(locks & (1 << i)))
495 mutex_enter(&scl->scl_lock);
496 if (rw == RW_READER) {
497 while (scl->scl_writer || scl->scl_write_wanted) {
498 cv_wait(&scl->scl_cv, &scl->scl_lock);
501 ASSERT(scl->scl_writer != curthread);
502 while (!refcount_is_zero(&scl->scl_count)) {
503 scl->scl_write_wanted++;
504 cv_wait(&scl->scl_cv, &scl->scl_lock);
505 scl->scl_write_wanted--;
507 scl->scl_writer = curthread;
509 (void) refcount_add(&scl->scl_count, tag);
510 mutex_exit(&scl->scl_lock);
512 ASSERT(wlocks_held <= locks);
516 spa_config_exit(spa_t *spa, int locks, void *tag)
518 for (int i = SCL_LOCKS - 1; i >= 0; i--) {
519 spa_config_lock_t *scl = &spa->spa_config_lock[i];
520 if (!(locks & (1 << i)))
522 mutex_enter(&scl->scl_lock);
523 ASSERT(!refcount_is_zero(&scl->scl_count));
524 if (refcount_remove(&scl->scl_count, tag) == 0) {
525 ASSERT(scl->scl_writer == NULL ||
526 scl->scl_writer == curthread);
527 scl->scl_writer = NULL; /* OK in either case */
528 cv_broadcast(&scl->scl_cv);
530 mutex_exit(&scl->scl_lock);
535 spa_config_held(spa_t *spa, int locks, krw_t rw)
539 for (int i = 0; i < SCL_LOCKS; i++) {
540 spa_config_lock_t *scl = &spa->spa_config_lock[i];
541 if (!(locks & (1 << i)))
543 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
544 (rw == RW_WRITER && scl->scl_writer == curthread))
545 locks_held |= 1 << i;
552 * ==========================================================================
553 * SPA namespace functions
554 * ==========================================================================
558 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
559 * Returns NULL if no matching spa_t is found.
562 spa_lookup(const char *name)
564 static spa_t search; /* spa_t is large; don't allocate on stack */
569 ASSERT(MUTEX_HELD(&spa_namespace_lock));
571 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
574 * If it's a full dataset name, figure out the pool name and
577 cp = strpbrk(search.spa_name, "/@#");
581 spa = avl_find(&spa_namespace_avl, &search, &where);
587 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
588 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
589 * looking for potentially hung I/Os.
592 spa_deadman(void *arg)
597 * Disable the deadman timer if the pool is suspended.
599 if (spa_suspended(spa)) {
601 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
603 /* Nothing. just don't schedule any future callouts. */
608 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
609 (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
610 ++spa->spa_deadman_calls);
611 if (zfs_deadman_enabled)
612 vdev_deadman(spa->spa_root_vdev);
615 callout_schedule(&spa->spa_deadman_cycid,
616 hz * zfs_deadman_checktime_ms / MILLISEC);
622 * Create an uninitialized spa_t with the given name. Requires
623 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
624 * exist by calling spa_lookup() first.
627 spa_add(const char *name, nvlist_t *config, const char *altroot)
630 spa_config_dirent_t *dp;
636 ASSERT(MUTEX_HELD(&spa_namespace_lock));
638 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
640 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
641 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
642 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
643 mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL);
644 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
645 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
646 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
647 mutex_init(&spa->spa_cksum_tmpls_lock, NULL, MUTEX_DEFAULT, NULL);
648 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
649 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
650 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
652 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
653 cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL);
654 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
655 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
656 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
658 for (int t = 0; t < TXG_SIZE; t++)
659 bplist_create(&spa->spa_free_bplist[t]);
661 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
662 spa->spa_state = POOL_STATE_UNINITIALIZED;
663 spa->spa_freeze_txg = UINT64_MAX;
664 spa->spa_final_txg = UINT64_MAX;
665 spa->spa_load_max_txg = UINT64_MAX;
667 spa->spa_proc_state = SPA_PROC_NONE;
670 hdlr.cyh_func = spa_deadman;
672 hdlr.cyh_level = CY_LOW_LEVEL;
675 spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
679 * This determines how often we need to check for hung I/Os after
680 * the cyclic has already fired. Since checking for hung I/Os is
681 * an expensive operation we don't want to check too frequently.
682 * Instead wait for 5 seconds before checking again.
684 when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms);
685 when.cyt_when = CY_INFINITY;
686 mutex_enter(&cpu_lock);
687 spa->spa_deadman_cycid = cyclic_add(&hdlr, &when);
688 mutex_exit(&cpu_lock);
691 callout_init(&spa->spa_deadman_cycid, CALLOUT_MPSAFE);
694 refcount_create(&spa->spa_refcount);
695 spa_config_lock_init(spa);
697 avl_add(&spa_namespace_avl, spa);
700 * Set the alternate root, if there is one.
703 spa->spa_root = spa_strdup(altroot);
708 * Every pool starts with the default cachefile
710 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
711 offsetof(spa_config_dirent_t, scd_link));
713 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
714 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
715 list_insert_head(&spa->spa_config_list, dp);
717 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
720 if (config != NULL) {
723 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
725 VERIFY(nvlist_dup(features, &spa->spa_label_features,
729 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
732 if (spa->spa_label_features == NULL) {
733 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
737 spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0);
739 spa->spa_min_ashift = INT_MAX;
740 spa->spa_max_ashift = 0;
743 * As a pool is being created, treat all features as disabled by
744 * setting SPA_FEATURE_DISABLED for all entries in the feature
747 for (int i = 0; i < SPA_FEATURES; i++) {
748 spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED;
755 * Removes a spa_t from the namespace, freeing up any memory used. Requires
756 * spa_namespace_lock. This is called only after the spa_t has been closed and
760 spa_remove(spa_t *spa)
762 spa_config_dirent_t *dp;
764 ASSERT(MUTEX_HELD(&spa_namespace_lock));
765 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
766 ASSERT3U(refcount_count(&spa->spa_refcount), ==, 0);
768 nvlist_free(spa->spa_config_splitting);
770 avl_remove(&spa_namespace_avl, spa);
771 cv_broadcast(&spa_namespace_cv);
774 spa_strfree(spa->spa_root);
778 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
779 list_remove(&spa->spa_config_list, dp);
780 if (dp->scd_path != NULL)
781 spa_strfree(dp->scd_path);
782 kmem_free(dp, sizeof (spa_config_dirent_t));
785 list_destroy(&spa->spa_config_list);
787 nvlist_free(spa->spa_label_features);
788 nvlist_free(spa->spa_load_info);
789 spa_config_set(spa, NULL);
792 mutex_enter(&cpu_lock);
793 if (spa->spa_deadman_cycid != CYCLIC_NONE)
794 cyclic_remove(spa->spa_deadman_cycid);
795 mutex_exit(&cpu_lock);
796 spa->spa_deadman_cycid = CYCLIC_NONE;
799 callout_drain(&spa->spa_deadman_cycid);
803 refcount_destroy(&spa->spa_refcount);
805 spa_config_lock_destroy(spa);
807 for (int t = 0; t < TXG_SIZE; t++)
808 bplist_destroy(&spa->spa_free_bplist[t]);
810 zio_checksum_templates_free(spa);
812 cv_destroy(&spa->spa_async_cv);
813 cv_destroy(&spa->spa_evicting_os_cv);
814 cv_destroy(&spa->spa_proc_cv);
815 cv_destroy(&spa->spa_scrub_io_cv);
816 cv_destroy(&spa->spa_suspend_cv);
818 mutex_destroy(&spa->spa_async_lock);
819 mutex_destroy(&spa->spa_errlist_lock);
820 mutex_destroy(&spa->spa_errlog_lock);
821 mutex_destroy(&spa->spa_evicting_os_lock);
822 mutex_destroy(&spa->spa_history_lock);
823 mutex_destroy(&spa->spa_proc_lock);
824 mutex_destroy(&spa->spa_props_lock);
825 mutex_destroy(&spa->spa_cksum_tmpls_lock);
826 mutex_destroy(&spa->spa_scrub_lock);
827 mutex_destroy(&spa->spa_suspend_lock);
828 mutex_destroy(&spa->spa_vdev_top_lock);
830 kmem_free(spa, sizeof (spa_t));
834 * Given a pool, return the next pool in the namespace, or NULL if there is
835 * none. If 'prev' is NULL, return the first pool.
838 spa_next(spa_t *prev)
840 ASSERT(MUTEX_HELD(&spa_namespace_lock));
843 return (AVL_NEXT(&spa_namespace_avl, prev));
845 return (avl_first(&spa_namespace_avl));
849 * ==========================================================================
850 * SPA refcount functions
851 * ==========================================================================
855 * Add a reference to the given spa_t. Must have at least one reference, or
856 * have the namespace lock held.
859 spa_open_ref(spa_t *spa, void *tag)
861 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
862 MUTEX_HELD(&spa_namespace_lock));
863 (void) refcount_add(&spa->spa_refcount, tag);
867 * Remove a reference to the given spa_t. Must have at least one reference, or
868 * have the namespace lock held.
871 spa_close(spa_t *spa, void *tag)
873 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
874 MUTEX_HELD(&spa_namespace_lock));
875 (void) refcount_remove(&spa->spa_refcount, tag);
879 * Remove a reference to the given spa_t held by a dsl dir that is
880 * being asynchronously released. Async releases occur from a taskq
881 * performing eviction of dsl datasets and dirs. The namespace lock
882 * isn't held and the hold by the object being evicted may contribute to
883 * spa_minref (e.g. dataset or directory released during pool export),
884 * so the asserts in spa_close() do not apply.
887 spa_async_close(spa_t *spa, void *tag)
889 (void) refcount_remove(&spa->spa_refcount, tag);
893 * Check to see if the spa refcount is zero. Must be called with
894 * spa_namespace_lock held. We really compare against spa_minref, which is the
895 * number of references acquired when opening a pool
898 spa_refcount_zero(spa_t *spa)
900 ASSERT(MUTEX_HELD(&spa_namespace_lock));
902 return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
906 * ==========================================================================
907 * SPA spare and l2cache tracking
908 * ==========================================================================
912 * Hot spares and cache devices are tracked using the same code below,
913 * for 'auxiliary' devices.
916 typedef struct spa_aux {
924 spa_aux_compare(const void *a, const void *b)
926 const spa_aux_t *sa = a;
927 const spa_aux_t *sb = b;
929 if (sa->aux_guid < sb->aux_guid)
931 else if (sa->aux_guid > sb->aux_guid)
938 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
944 search.aux_guid = vd->vdev_guid;
945 if ((aux = avl_find(avl, &search, &where)) != NULL) {
948 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
949 aux->aux_guid = vd->vdev_guid;
951 avl_insert(avl, aux, where);
956 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
962 search.aux_guid = vd->vdev_guid;
963 aux = avl_find(avl, &search, &where);
967 if (--aux->aux_count == 0) {
968 avl_remove(avl, aux);
969 kmem_free(aux, sizeof (spa_aux_t));
970 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
971 aux->aux_pool = 0ULL;
976 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
978 spa_aux_t search, *found;
980 search.aux_guid = guid;
981 found = avl_find(avl, &search, NULL);
985 *pool = found->aux_pool;
992 *refcnt = found->aux_count;
997 return (found != NULL);
1001 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
1003 spa_aux_t search, *found;
1006 search.aux_guid = vd->vdev_guid;
1007 found = avl_find(avl, &search, &where);
1008 ASSERT(found != NULL);
1009 ASSERT(found->aux_pool == 0ULL);
1011 found->aux_pool = spa_guid(vd->vdev_spa);
1015 * Spares are tracked globally due to the following constraints:
1017 * - A spare may be part of multiple pools.
1018 * - A spare may be added to a pool even if it's actively in use within
1020 * - A spare in use in any pool can only be the source of a replacement if
1021 * the target is a spare in the same pool.
1023 * We keep track of all spares on the system through the use of a reference
1024 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
1025 * spare, then we bump the reference count in the AVL tree. In addition, we set
1026 * the 'vdev_isspare' member to indicate that the device is a spare (active or
1027 * inactive). When a spare is made active (used to replace a device in the
1028 * pool), we also keep track of which pool its been made a part of.
1030 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
1031 * called under the spa_namespace lock as part of vdev reconfiguration. The
1032 * separate spare lock exists for the status query path, which does not need to
1033 * be completely consistent with respect to other vdev configuration changes.
1037 spa_spare_compare(const void *a, const void *b)
1039 return (spa_aux_compare(a, b));
1043 spa_spare_add(vdev_t *vd)
1045 mutex_enter(&spa_spare_lock);
1046 ASSERT(!vd->vdev_isspare);
1047 spa_aux_add(vd, &spa_spare_avl);
1048 vd->vdev_isspare = B_TRUE;
1049 mutex_exit(&spa_spare_lock);
1053 spa_spare_remove(vdev_t *vd)
1055 mutex_enter(&spa_spare_lock);
1056 ASSERT(vd->vdev_isspare);
1057 spa_aux_remove(vd, &spa_spare_avl);
1058 vd->vdev_isspare = B_FALSE;
1059 mutex_exit(&spa_spare_lock);
1063 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
1067 mutex_enter(&spa_spare_lock);
1068 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
1069 mutex_exit(&spa_spare_lock);
1075 spa_spare_activate(vdev_t *vd)
1077 mutex_enter(&spa_spare_lock);
1078 ASSERT(vd->vdev_isspare);
1079 spa_aux_activate(vd, &spa_spare_avl);
1080 mutex_exit(&spa_spare_lock);
1084 * Level 2 ARC devices are tracked globally for the same reasons as spares.
1085 * Cache devices currently only support one pool per cache device, and so
1086 * for these devices the aux reference count is currently unused beyond 1.
1090 spa_l2cache_compare(const void *a, const void *b)
1092 return (spa_aux_compare(a, b));
1096 spa_l2cache_add(vdev_t *vd)
1098 mutex_enter(&spa_l2cache_lock);
1099 ASSERT(!vd->vdev_isl2cache);
1100 spa_aux_add(vd, &spa_l2cache_avl);
1101 vd->vdev_isl2cache = B_TRUE;
1102 mutex_exit(&spa_l2cache_lock);
1106 spa_l2cache_remove(vdev_t *vd)
1108 mutex_enter(&spa_l2cache_lock);
1109 ASSERT(vd->vdev_isl2cache);
1110 spa_aux_remove(vd, &spa_l2cache_avl);
1111 vd->vdev_isl2cache = B_FALSE;
1112 mutex_exit(&spa_l2cache_lock);
1116 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
1120 mutex_enter(&spa_l2cache_lock);
1121 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
1122 mutex_exit(&spa_l2cache_lock);
1128 spa_l2cache_activate(vdev_t *vd)
1130 mutex_enter(&spa_l2cache_lock);
1131 ASSERT(vd->vdev_isl2cache);
1132 spa_aux_activate(vd, &spa_l2cache_avl);
1133 mutex_exit(&spa_l2cache_lock);
1137 * ==========================================================================
1139 * ==========================================================================
1143 * Lock the given spa_t for the purpose of adding or removing a vdev.
1144 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1145 * It returns the next transaction group for the spa_t.
1148 spa_vdev_enter(spa_t *spa)
1150 mutex_enter(&spa->spa_vdev_top_lock);
1151 mutex_enter(&spa_namespace_lock);
1152 return (spa_vdev_config_enter(spa));
1156 * Internal implementation for spa_vdev_enter(). Used when a vdev
1157 * operation requires multiple syncs (i.e. removing a device) while
1158 * keeping the spa_namespace_lock held.
1161 spa_vdev_config_enter(spa_t *spa)
1163 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1165 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1167 return (spa_last_synced_txg(spa) + 1);
1171 * Used in combination with spa_vdev_config_enter() to allow the syncing
1172 * of multiple transactions without releasing the spa_namespace_lock.
1175 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
1177 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1179 int config_changed = B_FALSE;
1181 ASSERT(txg > spa_last_synced_txg(spa));
1183 spa->spa_pending_vdev = NULL;
1186 * Reassess the DTLs.
1188 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
1190 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
1191 config_changed = B_TRUE;
1192 spa->spa_config_generation++;
1196 * Verify the metaslab classes.
1198 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
1199 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
1201 spa_config_exit(spa, SCL_ALL, spa);
1204 * Panic the system if the specified tag requires it. This
1205 * is useful for ensuring that configurations are updated
1208 if (zio_injection_enabled)
1209 zio_handle_panic_injection(spa, tag, 0);
1212 * Note: this txg_wait_synced() is important because it ensures
1213 * that there won't be more than one config change per txg.
1214 * This allows us to use the txg as the generation number.
1217 txg_wait_synced(spa->spa_dsl_pool, txg);
1220 ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
1221 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1223 spa_config_exit(spa, SCL_ALL, spa);
1227 * If the config changed, update the config cache.
1230 spa_config_sync(spa, B_FALSE, B_TRUE);
1234 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1235 * locking of spa_vdev_enter(), we also want make sure the transactions have
1236 * synced to disk, and then update the global configuration cache with the new
1240 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1242 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1243 mutex_exit(&spa_namespace_lock);
1244 mutex_exit(&spa->spa_vdev_top_lock);
1250 * Lock the given spa_t for the purpose of changing vdev state.
1253 spa_vdev_state_enter(spa_t *spa, int oplocks)
1255 int locks = SCL_STATE_ALL | oplocks;
1258 * Root pools may need to read of the underlying devfs filesystem
1259 * when opening up a vdev. Unfortunately if we're holding the
1260 * SCL_ZIO lock it will result in a deadlock when we try to issue
1261 * the read from the root filesystem. Instead we "prefetch"
1262 * the associated vnodes that we need prior to opening the
1263 * underlying devices and cache them so that we can prevent
1264 * any I/O when we are doing the actual open.
1266 if (spa_is_root(spa)) {
1267 int low = locks & ~(SCL_ZIO - 1);
1268 int high = locks & ~low;
1270 spa_config_enter(spa, high, spa, RW_WRITER);
1271 vdev_hold(spa->spa_root_vdev);
1272 spa_config_enter(spa, low, spa, RW_WRITER);
1274 spa_config_enter(spa, locks, spa, RW_WRITER);
1276 spa->spa_vdev_locks = locks;
1280 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1282 boolean_t config_changed = B_FALSE;
1284 if (vd != NULL || error == 0)
1285 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1289 vdev_state_dirty(vd->vdev_top);
1290 config_changed = B_TRUE;
1291 spa->spa_config_generation++;
1294 if (spa_is_root(spa))
1295 vdev_rele(spa->spa_root_vdev);
1297 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1298 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1301 * If anything changed, wait for it to sync. This ensures that,
1302 * from the system administrator's perspective, zpool(1M) commands
1303 * are synchronous. This is important for things like zpool offline:
1304 * when the command completes, you expect no further I/O from ZFS.
1307 txg_wait_synced(spa->spa_dsl_pool, 0);
1310 * If the config changed, update the config cache.
1312 if (config_changed) {
1313 mutex_enter(&spa_namespace_lock);
1314 spa_config_sync(spa, B_FALSE, B_TRUE);
1315 mutex_exit(&spa_namespace_lock);
1322 * ==========================================================================
1323 * Miscellaneous functions
1324 * ==========================================================================
1328 spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx)
1330 if (!nvlist_exists(spa->spa_label_features, feature)) {
1331 fnvlist_add_boolean(spa->spa_label_features, feature);
1333 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1334 * dirty the vdev config because lock SCL_CONFIG is not held.
1335 * Thankfully, in this case we don't need to dirty the config
1336 * because it will be written out anyway when we finish
1337 * creating the pool.
1339 if (tx->tx_txg != TXG_INITIAL)
1340 vdev_config_dirty(spa->spa_root_vdev);
1345 spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1347 if (nvlist_remove_all(spa->spa_label_features, feature) == 0)
1348 vdev_config_dirty(spa->spa_root_vdev);
1355 spa_rename(const char *name, const char *newname)
1361 * Lookup the spa_t and grab the config lock for writing. We need to
1362 * actually open the pool so that we can sync out the necessary labels.
1363 * It's OK to call spa_open() with the namespace lock held because we
1364 * allow recursive calls for other reasons.
1366 mutex_enter(&spa_namespace_lock);
1367 if ((err = spa_open(name, &spa, FTAG)) != 0) {
1368 mutex_exit(&spa_namespace_lock);
1372 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1374 avl_remove(&spa_namespace_avl, spa);
1375 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1376 avl_add(&spa_namespace_avl, spa);
1379 * Sync all labels to disk with the new names by marking the root vdev
1380 * dirty and waiting for it to sync. It will pick up the new pool name
1383 vdev_config_dirty(spa->spa_root_vdev);
1385 spa_config_exit(spa, SCL_ALL, FTAG);
1387 txg_wait_synced(spa->spa_dsl_pool, 0);
1390 * Sync the updated config cache.
1392 spa_config_sync(spa, B_FALSE, B_TRUE);
1394 spa_close(spa, FTAG);
1396 mutex_exit(&spa_namespace_lock);
1402 * Return the spa_t associated with given pool_guid, if it exists. If
1403 * device_guid is non-zero, determine whether the pool exists *and* contains
1404 * a device with the specified device_guid.
1407 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1410 avl_tree_t *t = &spa_namespace_avl;
1412 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1414 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1415 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1417 if (spa->spa_root_vdev == NULL)
1419 if (spa_guid(spa) == pool_guid) {
1420 if (device_guid == 0)
1423 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1424 device_guid) != NULL)
1428 * Check any devices we may be in the process of adding.
1430 if (spa->spa_pending_vdev) {
1431 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1432 device_guid) != NULL)
1442 * Determine whether a pool with the given pool_guid exists.
1445 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1447 return (spa_by_guid(pool_guid, device_guid) != NULL);
1451 spa_strdup(const char *s)
1457 new = kmem_alloc(len + 1, KM_SLEEP);
1465 spa_strfree(char *s)
1467 kmem_free(s, strlen(s) + 1);
1471 spa_get_random(uint64_t range)
1477 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1483 spa_generate_guid(spa_t *spa)
1485 uint64_t guid = spa_get_random(-1ULL);
1488 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1489 guid = spa_get_random(-1ULL);
1491 while (guid == 0 || spa_guid_exists(guid, 0))
1492 guid = spa_get_random(-1ULL);
1499 snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp)
1502 char *checksum = NULL;
1503 char *compress = NULL;
1506 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1507 dmu_object_byteswap_t bswap =
1508 DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1509 (void) snprintf(type, sizeof (type), "bswap %s %s",
1510 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1511 "metadata" : "data",
1512 dmu_ot_byteswap[bswap].ob_name);
1514 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1517 if (!BP_IS_EMBEDDED(bp)) {
1519 zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1521 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1524 SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum,
1529 spa_freeze(spa_t *spa)
1531 uint64_t freeze_txg = 0;
1533 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1534 if (spa->spa_freeze_txg == UINT64_MAX) {
1535 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1536 spa->spa_freeze_txg = freeze_txg;
1538 spa_config_exit(spa, SCL_ALL, FTAG);
1539 if (freeze_txg != 0)
1540 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1544 zfs_panic_recover(const char *fmt, ...)
1549 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1554 * This is a stripped-down version of strtoull, suitable only for converting
1555 * lowercase hexadecimal numbers that don't overflow.
1558 zfs_strtonum(const char *str, char **nptr)
1564 while ((c = *str) != '\0') {
1565 if (c >= '0' && c <= '9')
1567 else if (c >= 'a' && c <= 'f')
1568 digit = 10 + c - 'a';
1579 *nptr = (char *)str;
1585 * ==========================================================================
1586 * Accessor functions
1587 * ==========================================================================
1591 spa_shutting_down(spa_t *spa)
1593 return (spa->spa_async_suspended);
1597 spa_get_dsl(spa_t *spa)
1599 return (spa->spa_dsl_pool);
1603 spa_is_initializing(spa_t *spa)
1605 return (spa->spa_is_initializing);
1609 spa_get_rootblkptr(spa_t *spa)
1611 return (&spa->spa_ubsync.ub_rootbp);
1615 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1617 spa->spa_uberblock.ub_rootbp = *bp;
1621 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1623 if (spa->spa_root == NULL)
1626 (void) strncpy(buf, spa->spa_root, buflen);
1630 spa_sync_pass(spa_t *spa)
1632 return (spa->spa_sync_pass);
1636 spa_name(spa_t *spa)
1638 return (spa->spa_name);
1642 spa_guid(spa_t *spa)
1644 dsl_pool_t *dp = spa_get_dsl(spa);
1648 * If we fail to parse the config during spa_load(), we can go through
1649 * the error path (which posts an ereport) and end up here with no root
1650 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1653 if (spa->spa_root_vdev == NULL)
1654 return (spa->spa_config_guid);
1656 guid = spa->spa_last_synced_guid != 0 ?
1657 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1660 * Return the most recently synced out guid unless we're
1661 * in syncing context.
1663 if (dp && dsl_pool_sync_context(dp))
1664 return (spa->spa_root_vdev->vdev_guid);
1670 spa_load_guid(spa_t *spa)
1673 * This is a GUID that exists solely as a reference for the
1674 * purposes of the arc. It is generated at load time, and
1675 * is never written to persistent storage.
1677 return (spa->spa_load_guid);
1681 spa_last_synced_txg(spa_t *spa)
1683 return (spa->spa_ubsync.ub_txg);
1687 spa_first_txg(spa_t *spa)
1689 return (spa->spa_first_txg);
1693 spa_syncing_txg(spa_t *spa)
1695 return (spa->spa_syncing_txg);
1699 spa_state(spa_t *spa)
1701 return (spa->spa_state);
1705 spa_load_state(spa_t *spa)
1707 return (spa->spa_load_state);
1711 spa_freeze_txg(spa_t *spa)
1713 return (spa->spa_freeze_txg);
1718 spa_get_asize(spa_t *spa, uint64_t lsize)
1720 return (lsize * spa_asize_inflation);
1724 * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%),
1727 * See the comment above spa_slop_shift for details.
1730 spa_get_slop_space(spa_t *spa) {
1731 uint64_t space = spa_get_dspace(spa);
1732 return (MAX(space >> spa_slop_shift, SPA_MINDEVSIZE >> 1));
1736 spa_get_dspace(spa_t *spa)
1738 return (spa->spa_dspace);
1742 spa_update_dspace(spa_t *spa)
1744 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1745 ddt_get_dedup_dspace(spa);
1749 * Return the failure mode that has been set to this pool. The default
1750 * behavior will be to block all I/Os when a complete failure occurs.
1753 spa_get_failmode(spa_t *spa)
1755 return (spa->spa_failmode);
1759 spa_suspended(spa_t *spa)
1761 return (spa->spa_suspended);
1765 spa_version(spa_t *spa)
1767 return (spa->spa_ubsync.ub_version);
1771 spa_deflate(spa_t *spa)
1773 return (spa->spa_deflate);
1777 spa_normal_class(spa_t *spa)
1779 return (spa->spa_normal_class);
1783 spa_log_class(spa_t *spa)
1785 return (spa->spa_log_class);
1789 spa_evicting_os_register(spa_t *spa, objset_t *os)
1791 mutex_enter(&spa->spa_evicting_os_lock);
1792 list_insert_head(&spa->spa_evicting_os_list, os);
1793 mutex_exit(&spa->spa_evicting_os_lock);
1797 spa_evicting_os_deregister(spa_t *spa, objset_t *os)
1799 mutex_enter(&spa->spa_evicting_os_lock);
1800 list_remove(&spa->spa_evicting_os_list, os);
1801 cv_broadcast(&spa->spa_evicting_os_cv);
1802 mutex_exit(&spa->spa_evicting_os_lock);
1806 spa_evicting_os_wait(spa_t *spa)
1808 mutex_enter(&spa->spa_evicting_os_lock);
1809 while (!list_is_empty(&spa->spa_evicting_os_list))
1810 cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock);
1811 mutex_exit(&spa->spa_evicting_os_lock);
1813 dmu_buf_user_evict_wait();
1817 spa_max_replication(spa_t *spa)
1820 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1821 * handle BPs with more than one DVA allocated. Set our max
1822 * replication level accordingly.
1824 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1826 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1830 spa_prev_software_version(spa_t *spa)
1832 return (spa->spa_prev_software_version);
1836 spa_deadman_synctime(spa_t *spa)
1838 return (spa->spa_deadman_synctime);
1842 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1844 uint64_t asize = DVA_GET_ASIZE(dva);
1845 uint64_t dsize = asize;
1847 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1849 if (asize != 0 && spa->spa_deflate) {
1850 uint64_t vdev = DVA_GET_VDEV(dva);
1851 vdev_t *vd = vdev_lookup_top(spa, vdev);
1854 "dva_get_dsize_sync(): bad DVA %llu:%llu",
1855 (u_longlong_t)vdev, (u_longlong_t)asize);
1857 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1864 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1868 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1869 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1875 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1879 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1881 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1882 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1884 spa_config_exit(spa, SCL_VDEV, FTAG);
1890 * ==========================================================================
1891 * Initialization and Termination
1892 * ==========================================================================
1896 spa_name_compare(const void *a1, const void *a2)
1898 const spa_t *s1 = a1;
1899 const spa_t *s2 = a2;
1902 s = strcmp(s1->spa_name, s2->spa_name);
1913 return (spa_active_count);
1923 EVENTHANDLER_DEFINE(mountroot, spa_boot_init, NULL, 0);
1929 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1930 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1931 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1932 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1934 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1935 offsetof(spa_t, spa_avl));
1937 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1938 offsetof(spa_aux_t, aux_avl));
1940 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1941 offsetof(spa_aux_t, aux_avl));
1943 spa_mode_global = mode;
1949 if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
1950 arc_procfd = open("/proc/self/ctl", O_WRONLY);
1951 if (arc_procfd == -1) {
1952 perror("could not enable watchpoints: "
1953 "opening /proc/self/ctl failed: ");
1959 #endif /* illumos */
1967 vdev_cache_stat_init();
1970 zpool_feature_init();
1977 #endif /* !illumos */
1987 vdev_cache_stat_fini();
1996 avl_destroy(&spa_namespace_avl);
1997 avl_destroy(&spa_spare_avl);
1998 avl_destroy(&spa_l2cache_avl);
2000 cv_destroy(&spa_namespace_cv);
2001 mutex_destroy(&spa_namespace_lock);
2002 mutex_destroy(&spa_spare_lock);
2003 mutex_destroy(&spa_l2cache_lock);
2007 * Return whether this pool has slogs. No locking needed.
2008 * It's not a problem if the wrong answer is returned as it's only for
2009 * performance and not correctness
2012 spa_has_slogs(spa_t *spa)
2014 return (spa->spa_log_class->mc_rotor != NULL);
2018 spa_get_log_state(spa_t *spa)
2020 return (spa->spa_log_state);
2024 spa_set_log_state(spa_t *spa, spa_log_state_t state)
2026 spa->spa_log_state = state;
2030 spa_is_root(spa_t *spa)
2032 return (spa->spa_is_root);
2036 spa_writeable(spa_t *spa)
2038 return (!!(spa->spa_mode & FWRITE));
2042 * Returns true if there is a pending sync task in any of the current
2043 * syncing txg, the current quiescing txg, or the current open txg.
2046 spa_has_pending_synctask(spa_t *spa)
2048 return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks));
2052 spa_mode(spa_t *spa)
2054 return (spa->spa_mode);
2058 spa_bootfs(spa_t *spa)
2060 return (spa->spa_bootfs);
2064 spa_delegation(spa_t *spa)
2066 return (spa->spa_delegation);
2070 spa_meta_objset(spa_t *spa)
2072 return (spa->spa_meta_objset);
2076 spa_dedup_checksum(spa_t *spa)
2078 return (spa->spa_dedup_checksum);
2082 * Reset pool scan stat per scan pass (or reboot).
2085 spa_scan_stat_init(spa_t *spa)
2087 /* data not stored on disk */
2088 spa->spa_scan_pass_start = gethrestime_sec();
2089 spa->spa_scan_pass_exam = 0;
2090 vdev_scan_stat_init(spa->spa_root_vdev);
2094 * Get scan stats for zpool status reports
2097 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
2099 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
2101 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
2102 return (SET_ERROR(ENOENT));
2103 bzero(ps, sizeof (pool_scan_stat_t));
2105 /* data stored on disk */
2106 ps->pss_func = scn->scn_phys.scn_func;
2107 ps->pss_start_time = scn->scn_phys.scn_start_time;
2108 ps->pss_end_time = scn->scn_phys.scn_end_time;
2109 ps->pss_to_examine = scn->scn_phys.scn_to_examine;
2110 ps->pss_examined = scn->scn_phys.scn_examined;
2111 ps->pss_to_process = scn->scn_phys.scn_to_process;
2112 ps->pss_processed = scn->scn_phys.scn_processed;
2113 ps->pss_errors = scn->scn_phys.scn_errors;
2114 ps->pss_state = scn->scn_phys.scn_state;
2116 /* data not stored on disk */
2117 ps->pss_pass_start = spa->spa_scan_pass_start;
2118 ps->pss_pass_exam = spa->spa_scan_pass_exam;
2124 spa_debug_enabled(spa_t *spa)
2126 return (spa->spa_debug);
2130 spa_maxblocksize(spa_t *spa)
2132 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS))
2133 return (SPA_MAXBLOCKSIZE);
2135 return (SPA_OLD_MAXBLOCKSIZE);