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.
28 * Copyright (c) 2014 Integros [integros.com]
31 #include <sys/zfs_context.h>
32 #include <sys/spa_impl.h>
33 #include <sys/spa_boot.h>
35 #include <sys/zio_checksum.h>
36 #include <sys/zio_compress.h>
38 #include <sys/dmu_tx.h>
41 #include <sys/vdev_impl.h>
42 #include <sys/metaslab.h>
43 #include <sys/uberblock_impl.h>
46 #include <sys/unique.h>
47 #include <sys/dsl_pool.h>
48 #include <sys/dsl_dir.h>
49 #include <sys/dsl_prop.h>
50 #include <sys/dsl_scan.h>
51 #include <sys/fs/zfs.h>
52 #include <sys/metaslab_impl.h>
56 #include <sys/zfeature.h>
61 * There are four basic locks for managing spa_t structures:
63 * spa_namespace_lock (global mutex)
65 * This lock must be acquired to do any of the following:
67 * - Lookup a spa_t by name
68 * - Add or remove a spa_t from the namespace
69 * - Increase spa_refcount from non-zero
70 * - Check if spa_refcount is zero
72 * - add/remove/attach/detach devices
73 * - Held for the duration of create/destroy/import/export
75 * It does not need to handle recursion. A create or destroy may
76 * reference objects (files or zvols) in other pools, but by
77 * definition they must have an existing reference, and will never need
78 * to lookup a spa_t by name.
80 * spa_refcount (per-spa refcount_t protected by mutex)
82 * This reference count keep track of any active users of the spa_t. The
83 * spa_t cannot be destroyed or freed while this is non-zero. Internally,
84 * the refcount is never really 'zero' - opening a pool implicitly keeps
85 * some references in the DMU. Internally we check against spa_minref, but
86 * present the image of a zero/non-zero value to consumers.
88 * spa_config_lock[] (per-spa array of rwlocks)
90 * This protects the spa_t from config changes, and must be held in
91 * the following circumstances:
93 * - RW_READER to perform I/O to the spa
94 * - RW_WRITER to change the vdev config
96 * The locking order is fairly straightforward:
98 * spa_namespace_lock -> spa_refcount
100 * The namespace lock must be acquired to increase the refcount from 0
101 * or to check if it is zero.
103 * spa_refcount -> spa_config_lock[]
105 * There must be at least one valid reference on the spa_t to acquire
108 * spa_namespace_lock -> spa_config_lock[]
110 * The namespace lock must always be taken before the config lock.
113 * The spa_namespace_lock can be acquired directly and is globally visible.
115 * The namespace is manipulated using the following functions, all of which
116 * require the spa_namespace_lock to be held.
118 * spa_lookup() Lookup a spa_t by name.
120 * spa_add() Create a new spa_t in the namespace.
122 * spa_remove() Remove a spa_t from the namespace. This also
123 * frees up any memory associated with the spa_t.
125 * spa_next() Returns the next spa_t in the system, or the
126 * first if NULL is passed.
128 * spa_evict_all() Shutdown and remove all spa_t structures in
131 * spa_guid_exists() Determine whether a pool/device guid exists.
133 * The spa_refcount is manipulated using the following functions:
135 * spa_open_ref() Adds a reference to the given spa_t. Must be
136 * called with spa_namespace_lock held if the
137 * refcount is currently zero.
139 * spa_close() Remove a reference from the spa_t. This will
140 * not free the spa_t or remove it from the
141 * namespace. No locking is required.
143 * spa_refcount_zero() Returns true if the refcount is currently
144 * zero. Must be called with spa_namespace_lock
147 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
148 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
149 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
151 * To read the configuration, it suffices to hold one of these locks as reader.
152 * To modify the configuration, you must hold all locks as writer. To modify
153 * vdev state without altering the vdev tree's topology (e.g. online/offline),
154 * you must hold SCL_STATE and SCL_ZIO as writer.
156 * We use these distinct config locks to avoid recursive lock entry.
157 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
158 * block allocations (SCL_ALLOC), which may require reading space maps
159 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
161 * The spa config locks cannot be normal rwlocks because we need the
162 * ability to hand off ownership. For example, SCL_ZIO is acquired
163 * by the issuing thread and later released by an interrupt thread.
164 * They do, however, obey the usual write-wanted semantics to prevent
165 * writer (i.e. system administrator) starvation.
167 * The lock acquisition rules are as follows:
170 * Protects changes to the vdev tree topology, such as vdev
171 * add/remove/attach/detach. Protects the dirty config list
172 * (spa_config_dirty_list) and the set of spares and l2arc devices.
175 * Protects changes to pool state and vdev state, such as vdev
176 * online/offline/fault/degrade/clear. Protects the dirty state list
177 * (spa_state_dirty_list) and global pool state (spa_state).
180 * Protects changes to metaslab groups and classes.
181 * Held as reader by metaslab_alloc() and metaslab_claim().
184 * Held by bp-level zios (those which have no io_vd upon entry)
185 * to prevent changes to the vdev tree. The bp-level zio implicitly
186 * protects all of its vdev child zios, which do not hold SCL_ZIO.
189 * Protects changes to metaslab groups and classes.
190 * Held as reader by metaslab_free(). SCL_FREE is distinct from
191 * SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
192 * blocks in zio_done() while another i/o that holds either
193 * SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
196 * Held as reader to prevent changes to the vdev tree during trivial
197 * inquiries such as bp_get_dsize(). SCL_VDEV is distinct from the
198 * other locks, and lower than all of them, to ensure that it's safe
199 * to acquire regardless of caller context.
201 * In addition, the following rules apply:
203 * (a) spa_props_lock protects pool properties, spa_config and spa_config_list.
204 * The lock ordering is SCL_CONFIG > spa_props_lock.
206 * (b) I/O operations on leaf vdevs. For any zio operation that takes
207 * an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
208 * or zio_write_phys() -- the caller must ensure that the config cannot
209 * cannot change in the interim, and that the vdev cannot be reopened.
210 * SCL_STATE as reader suffices for both.
212 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
214 * spa_vdev_enter() Acquire the namespace lock and the config lock
217 * spa_vdev_exit() Release the config lock, wait for all I/O
218 * to complete, sync the updated configs to the
219 * cache, and release the namespace lock.
221 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
222 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
223 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
225 * spa_rename() is also implemented within this file since it requires
226 * manipulation of the namespace.
229 static avl_tree_t spa_namespace_avl;
230 kmutex_t spa_namespace_lock;
231 static kcondvar_t spa_namespace_cv;
232 static int spa_active_count;
233 int spa_max_replication_override = SPA_DVAS_PER_BP;
235 static kmutex_t spa_spare_lock;
236 static avl_tree_t spa_spare_avl;
237 static kmutex_t spa_l2cache_lock;
238 static avl_tree_t spa_l2cache_avl;
240 kmem_cache_t *spa_buffer_pool;
244 /* Everything except dprintf and spa is on by default in debug builds */
245 int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA);
250 TUNABLE_INT("debug.zfs_flags", &zfs_flags);
251 SYSCTL_INT(_debug, OID_AUTO, zfs_flags, CTLFLAG_RWTUN, &zfs_flags, 0,
255 * zfs_recover can be set to nonzero to attempt to recover from
256 * otherwise-fatal errors, typically caused by on-disk corruption. When
257 * set, calls to zfs_panic_recover() will turn into warning messages.
258 * This should only be used as a last resort, as it typically results
259 * in leaked space, or worse.
261 boolean_t zfs_recover = B_FALSE;
262 SYSCTL_DECL(_vfs_zfs);
263 TUNABLE_INT("vfs.zfs.recover", &zfs_recover);
264 SYSCTL_INT(_vfs_zfs, OID_AUTO, recover, CTLFLAG_RWTUN, &zfs_recover, 0,
265 "Try to recover from otherwise-fatal errors.");
268 sysctl_vfs_zfs_debug_flags(SYSCTL_HANDLER_ARGS)
273 err = sysctl_handle_int(oidp, &val, 0, req);
274 if (err != 0 || req->newptr == NULL)
278 * ZFS_DEBUG_MODIFY must be enabled prior to boot so all
279 * arc buffers in the system have the necessary additional
280 * checksum data. However, it is safe to disable at any
283 if (!(zfs_flags & ZFS_DEBUG_MODIFY))
284 val &= ~ZFS_DEBUG_MODIFY;
289 TUNABLE_INT("vfs.zfs.debugflags", &zfs_flags);
290 SYSCTL_PROC(_vfs_zfs, OID_AUTO, debugflags,
291 CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(int),
292 sysctl_vfs_zfs_debug_flags, "IU", "Debug flags for ZFS testing.");
293 SYSCTL_PROC(_vfs_zfs, OID_AUTO, debug_flags,
294 CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(int),
295 sysctl_vfs_zfs_debug_flags, "IU",
296 "Debug flags for ZFS testing (deprecated, see vfs.zfs.debugflags).");
299 * If destroy encounters an EIO while reading metadata (e.g. indirect
300 * blocks), space referenced by the missing metadata can not be freed.
301 * Normally this causes the background destroy to become "stalled", as
302 * it is unable to make forward progress. While in this stalled state,
303 * all remaining space to free from the error-encountering filesystem is
304 * "temporarily leaked". Set this flag to cause it to ignore the EIO,
305 * permanently leak the space from indirect blocks that can not be read,
306 * and continue to free everything else that it can.
308 * The default, "stalling" behavior is useful if the storage partially
309 * fails (i.e. some but not all i/os fail), and then later recovers. In
310 * this case, we will be able to continue pool operations while it is
311 * partially failed, and when it recovers, we can continue to free the
312 * space, with no leaks. However, note that this case is actually
315 * Typically pools either (a) fail completely (but perhaps temporarily,
316 * e.g. a top-level vdev going offline), or (b) have localized,
317 * permanent errors (e.g. disk returns the wrong data due to bit flip or
318 * firmware bug). In case (a), this setting does not matter because the
319 * pool will be suspended and the sync thread will not be able to make
320 * forward progress regardless. In case (b), because the error is
321 * permanent, the best we can do is leak the minimum amount of space,
322 * which is what setting this flag will do. Therefore, it is reasonable
323 * for this flag to normally be set, but we chose the more conservative
324 * approach of not setting it, so that there is no possibility of
325 * leaking space in the "partial temporary" failure case.
327 boolean_t zfs_free_leak_on_eio = B_FALSE;
330 * Expiration time in milliseconds. This value has two meanings. First it is
331 * used to determine when the spa_deadman() logic should fire. By default the
332 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
333 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
334 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
337 uint64_t zfs_deadman_synctime_ms = 1000000ULL;
338 TUNABLE_QUAD("vfs.zfs.deadman_synctime_ms", &zfs_deadman_synctime_ms);
339 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_synctime_ms, CTLFLAG_RDTUN,
340 &zfs_deadman_synctime_ms, 0,
341 "Stalled ZFS I/O expiration time in milliseconds");
344 * Check time in milliseconds. This defines the frequency at which we check
347 uint64_t zfs_deadman_checktime_ms = 5000ULL;
348 TUNABLE_QUAD("vfs.zfs.deadman_checktime_ms", &zfs_deadman_checktime_ms);
349 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, deadman_checktime_ms, CTLFLAG_RDTUN,
350 &zfs_deadman_checktime_ms, 0,
351 "Period of checks for stalled ZFS I/O in milliseconds");
354 * Default value of -1 for zfs_deadman_enabled is resolved in
357 int zfs_deadman_enabled = -1;
358 TUNABLE_INT("vfs.zfs.deadman_enabled", &zfs_deadman_enabled);
359 SYSCTL_INT(_vfs_zfs, OID_AUTO, deadman_enabled, CTLFLAG_RDTUN,
360 &zfs_deadman_enabled, 0, "Kernel panic on stalled ZFS I/O");
363 * The worst case is single-sector max-parity RAID-Z blocks, in which
364 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
365 * times the size; so just assume that. Add to this the fact that
366 * we can have up to 3 DVAs per bp, and one more factor of 2 because
367 * the block may be dittoed with up to 3 DVAs by ddt_sync(). All together,
369 * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
371 int spa_asize_inflation = 24;
372 TUNABLE_INT("vfs.zfs.spa_asize_inflation", &spa_asize_inflation);
373 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_asize_inflation, CTLFLAG_RWTUN,
374 &spa_asize_inflation, 0, "Worst case inflation factor for single sector writes");
382 * If we are not i386 or amd64 or in a virtual machine,
383 * disable ZFS deadman thread by default
385 if (zfs_deadman_enabled == -1) {
386 #if defined(__amd64__) || defined(__i386__)
387 zfs_deadman_enabled = (vm_guest == VM_GUEST_NO) ? 1 : 0;
389 zfs_deadman_enabled = 0;
394 #endif /* !illumos */
397 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
398 * the pool to be consumed. This ensures that we don't run the pool
399 * completely out of space, due to unaccounted changes (e.g. to the MOS).
400 * It also limits the worst-case time to allocate space. If we have
401 * less than this amount of free space, most ZPL operations (e.g. write,
402 * create) will return ENOSPC.
404 * Certain operations (e.g. file removal, most administrative actions) can
405 * use half the slop space. They will only return ENOSPC if less than half
406 * the slop space is free. Typically, once the pool has less than the slop
407 * space free, the user will use these operations to free up space in the pool.
408 * These are the operations that call dsl_pool_adjustedsize() with the netfree
409 * argument set to TRUE.
411 * A very restricted set of operations are always permitted, regardless of
412 * the amount of free space. These are the operations that call
413 * dsl_sync_task(ZFS_SPACE_CHECK_NONE), e.g. "zfs destroy". If these
414 * operations result in a net increase in the amount of space used,
415 * it is possible to run the pool completely out of space, causing it to
416 * be permanently read-only.
418 * Note that on very small pools, the slop space will be larger than
419 * 3.2%, in an effort to have it be at least spa_min_slop (128MB),
420 * but we never allow it to be more than half the pool size.
422 * See also the comments in zfs_space_check_t.
424 int spa_slop_shift = 5;
425 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_slop_shift, CTLFLAG_RWTUN,
427 "Shift value of reserved space (1/(2^spa_slop_shift)).");
428 uint64_t spa_min_slop = 128 * 1024 * 1024;
429 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, spa_min_slop, CTLFLAG_RWTUN,
431 "Minimal value of reserved space");
434 * ==========================================================================
436 * ==========================================================================
439 spa_config_lock_init(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_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
444 cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
445 refcount_create_untracked(&scl->scl_count);
446 scl->scl_writer = NULL;
447 scl->scl_write_wanted = 0;
452 spa_config_lock_destroy(spa_t *spa)
454 for (int i = 0; i < SCL_LOCKS; i++) {
455 spa_config_lock_t *scl = &spa->spa_config_lock[i];
456 mutex_destroy(&scl->scl_lock);
457 cv_destroy(&scl->scl_cv);
458 refcount_destroy(&scl->scl_count);
459 ASSERT(scl->scl_writer == NULL);
460 ASSERT(scl->scl_write_wanted == 0);
465 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
467 for (int i = 0; i < SCL_LOCKS; i++) {
468 spa_config_lock_t *scl = &spa->spa_config_lock[i];
469 if (!(locks & (1 << i)))
471 mutex_enter(&scl->scl_lock);
472 if (rw == RW_READER) {
473 if (scl->scl_writer || scl->scl_write_wanted) {
474 mutex_exit(&scl->scl_lock);
475 spa_config_exit(spa, locks & ((1 << i) - 1),
480 ASSERT(scl->scl_writer != curthread);
481 if (!refcount_is_zero(&scl->scl_count)) {
482 mutex_exit(&scl->scl_lock);
483 spa_config_exit(spa, locks & ((1 << i) - 1),
487 scl->scl_writer = curthread;
489 (void) refcount_add(&scl->scl_count, tag);
490 mutex_exit(&scl->scl_lock);
496 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
500 ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
502 for (int i = 0; i < SCL_LOCKS; i++) {
503 spa_config_lock_t *scl = &spa->spa_config_lock[i];
504 if (scl->scl_writer == curthread)
505 wlocks_held |= (1 << i);
506 if (!(locks & (1 << i)))
508 mutex_enter(&scl->scl_lock);
509 if (rw == RW_READER) {
510 while (scl->scl_writer || scl->scl_write_wanted) {
511 cv_wait(&scl->scl_cv, &scl->scl_lock);
514 ASSERT(scl->scl_writer != curthread);
515 while (!refcount_is_zero(&scl->scl_count)) {
516 scl->scl_write_wanted++;
517 cv_wait(&scl->scl_cv, &scl->scl_lock);
518 scl->scl_write_wanted--;
520 scl->scl_writer = curthread;
522 (void) refcount_add(&scl->scl_count, tag);
523 mutex_exit(&scl->scl_lock);
525 ASSERT(wlocks_held <= locks);
529 spa_config_exit(spa_t *spa, int locks, void *tag)
531 for (int i = SCL_LOCKS - 1; i >= 0; i--) {
532 spa_config_lock_t *scl = &spa->spa_config_lock[i];
533 if (!(locks & (1 << i)))
535 mutex_enter(&scl->scl_lock);
536 ASSERT(!refcount_is_zero(&scl->scl_count));
537 if (refcount_remove(&scl->scl_count, tag) == 0) {
538 ASSERT(scl->scl_writer == NULL ||
539 scl->scl_writer == curthread);
540 scl->scl_writer = NULL; /* OK in either case */
541 cv_broadcast(&scl->scl_cv);
543 mutex_exit(&scl->scl_lock);
548 spa_config_held(spa_t *spa, int locks, krw_t rw)
552 for (int i = 0; i < SCL_LOCKS; i++) {
553 spa_config_lock_t *scl = &spa->spa_config_lock[i];
554 if (!(locks & (1 << i)))
556 if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
557 (rw == RW_WRITER && scl->scl_writer == curthread))
558 locks_held |= 1 << i;
565 * ==========================================================================
566 * SPA namespace functions
567 * ==========================================================================
571 * Lookup the named spa_t in the AVL tree. The spa_namespace_lock must be held.
572 * Returns NULL if no matching spa_t is found.
575 spa_lookup(const char *name)
577 static spa_t search; /* spa_t is large; don't allocate on stack */
582 ASSERT(MUTEX_HELD(&spa_namespace_lock));
584 (void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
587 * If it's a full dataset name, figure out the pool name and
590 cp = strpbrk(search.spa_name, "/@#");
594 spa = avl_find(&spa_namespace_avl, &search, &where);
600 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
601 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
602 * looking for potentially hung I/Os.
605 spa_deadman(void *arg, int pending)
610 * Disable the deadman timer if the pool is suspended.
612 if (spa_suspended(spa)) {
614 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
616 /* Nothing. just don't schedule any future callouts. */
621 zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
622 (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
623 ++spa->spa_deadman_calls);
624 if (zfs_deadman_enabled)
625 vdev_deadman(spa->spa_root_vdev);
628 callout_schedule(&spa->spa_deadman_cycid,
629 hz * zfs_deadman_checktime_ms / MILLISEC);
634 #if defined(__FreeBSD__) && defined(_KERNEL)
636 spa_deadman_timeout(void *arg)
640 taskqueue_enqueue(taskqueue_thread, &spa->spa_deadman_task);
645 * Create an uninitialized spa_t with the given name. Requires
646 * spa_namespace_lock. The caller must ensure that the spa_t doesn't already
647 * exist by calling spa_lookup() first.
650 spa_add(const char *name, nvlist_t *config, const char *altroot)
653 spa_config_dirent_t *dp;
659 ASSERT(MUTEX_HELD(&spa_namespace_lock));
661 spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
663 mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
664 mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
665 mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
666 mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL);
667 mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
668 mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
669 mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
670 mutex_init(&spa->spa_cksum_tmpls_lock, NULL, MUTEX_DEFAULT, NULL);
671 mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
672 mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
673 mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
674 mutex_init(&spa->spa_alloc_lock, NULL, MUTEX_DEFAULT, NULL);
676 cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
677 cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL);
678 cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
679 cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
680 cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
682 for (int t = 0; t < TXG_SIZE; t++)
683 bplist_create(&spa->spa_free_bplist[t]);
685 (void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
686 spa->spa_state = POOL_STATE_UNINITIALIZED;
687 spa->spa_freeze_txg = UINT64_MAX;
688 spa->spa_final_txg = UINT64_MAX;
689 spa->spa_load_max_txg = UINT64_MAX;
691 spa->spa_proc_state = SPA_PROC_NONE;
694 hdlr.cyh_func = spa_deadman;
696 hdlr.cyh_level = CY_LOW_LEVEL;
699 spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
703 * This determines how often we need to check for hung I/Os after
704 * the cyclic has already fired. Since checking for hung I/Os is
705 * an expensive operation we don't want to check too frequently.
706 * Instead wait for 5 seconds before checking again.
708 when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms);
709 when.cyt_when = CY_INFINITY;
710 mutex_enter(&cpu_lock);
711 spa->spa_deadman_cycid = cyclic_add(&hdlr, &when);
712 mutex_exit(&cpu_lock);
716 * callout(9) does not provide a way to initialize a callout with
717 * a function and an argument, so we use callout_reset() to schedule
718 * the callout in the very distant future. Even if that event ever
719 * fires, it should be okayas we won't have any active zio-s.
720 * But normally spa_sync() will reschedule the callout with a proper
722 * callout(9) does not allow the callback function to sleep but
723 * vdev_deadman() needs to acquire vq_lock and illumos mutexes are
724 * emulated using sx(9). For this reason spa_deadman_timeout()
725 * will schedule spa_deadman() as task on a taskqueue that allows
728 TASK_INIT(&spa->spa_deadman_task, 0, spa_deadman, spa);
729 callout_init(&spa->spa_deadman_cycid, 1);
730 callout_reset_sbt(&spa->spa_deadman_cycid, SBT_MAX, 0,
731 spa_deadman_timeout, spa, 0);
734 refcount_create(&spa->spa_refcount);
735 spa_config_lock_init(spa);
737 avl_add(&spa_namespace_avl, spa);
740 * Set the alternate root, if there is one.
743 spa->spa_root = spa_strdup(altroot);
747 avl_create(&spa->spa_alloc_tree, zio_timestamp_compare,
748 sizeof (zio_t), offsetof(zio_t, io_alloc_node));
751 * Every pool starts with the default cachefile
753 list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
754 offsetof(spa_config_dirent_t, scd_link));
756 dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
757 dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
758 list_insert_head(&spa->spa_config_list, dp);
760 VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
763 if (config != NULL) {
766 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
768 VERIFY(nvlist_dup(features, &spa->spa_label_features,
772 VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
775 if (spa->spa_label_features == NULL) {
776 VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
780 spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0);
782 spa->spa_min_ashift = INT_MAX;
783 spa->spa_max_ashift = 0;
786 * As a pool is being created, treat all features as disabled by
787 * setting SPA_FEATURE_DISABLED for all entries in the feature
790 for (int i = 0; i < SPA_FEATURES; i++) {
791 spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED;
798 * Removes a spa_t from the namespace, freeing up any memory used. Requires
799 * spa_namespace_lock. This is called only after the spa_t has been closed and
803 spa_remove(spa_t *spa)
805 spa_config_dirent_t *dp;
807 ASSERT(MUTEX_HELD(&spa_namespace_lock));
808 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
809 ASSERT3U(refcount_count(&spa->spa_refcount), ==, 0);
811 nvlist_free(spa->spa_config_splitting);
813 avl_remove(&spa_namespace_avl, spa);
814 cv_broadcast(&spa_namespace_cv);
817 spa_strfree(spa->spa_root);
821 while ((dp = list_head(&spa->spa_config_list)) != NULL) {
822 list_remove(&spa->spa_config_list, dp);
823 if (dp->scd_path != NULL)
824 spa_strfree(dp->scd_path);
825 kmem_free(dp, sizeof (spa_config_dirent_t));
828 avl_destroy(&spa->spa_alloc_tree);
829 list_destroy(&spa->spa_config_list);
831 nvlist_free(spa->spa_label_features);
832 nvlist_free(spa->spa_load_info);
833 spa_config_set(spa, NULL);
836 mutex_enter(&cpu_lock);
837 if (spa->spa_deadman_cycid != CYCLIC_NONE)
838 cyclic_remove(spa->spa_deadman_cycid);
839 mutex_exit(&cpu_lock);
840 spa->spa_deadman_cycid = CYCLIC_NONE;
843 callout_drain(&spa->spa_deadman_cycid);
844 taskqueue_drain(taskqueue_thread, &spa->spa_deadman_task);
848 refcount_destroy(&spa->spa_refcount);
850 spa_config_lock_destroy(spa);
852 for (int t = 0; t < TXG_SIZE; t++)
853 bplist_destroy(&spa->spa_free_bplist[t]);
855 zio_checksum_templates_free(spa);
857 cv_destroy(&spa->spa_async_cv);
858 cv_destroy(&spa->spa_evicting_os_cv);
859 cv_destroy(&spa->spa_proc_cv);
860 cv_destroy(&spa->spa_scrub_io_cv);
861 cv_destroy(&spa->spa_suspend_cv);
863 mutex_destroy(&spa->spa_alloc_lock);
864 mutex_destroy(&spa->spa_async_lock);
865 mutex_destroy(&spa->spa_errlist_lock);
866 mutex_destroy(&spa->spa_errlog_lock);
867 mutex_destroy(&spa->spa_evicting_os_lock);
868 mutex_destroy(&spa->spa_history_lock);
869 mutex_destroy(&spa->spa_proc_lock);
870 mutex_destroy(&spa->spa_props_lock);
871 mutex_destroy(&spa->spa_cksum_tmpls_lock);
872 mutex_destroy(&spa->spa_scrub_lock);
873 mutex_destroy(&spa->spa_suspend_lock);
874 mutex_destroy(&spa->spa_vdev_top_lock);
876 kmem_free(spa, sizeof (spa_t));
880 * Given a pool, return the next pool in the namespace, or NULL if there is
881 * none. If 'prev' is NULL, return the first pool.
884 spa_next(spa_t *prev)
886 ASSERT(MUTEX_HELD(&spa_namespace_lock));
889 return (AVL_NEXT(&spa_namespace_avl, prev));
891 return (avl_first(&spa_namespace_avl));
895 * ==========================================================================
896 * SPA refcount functions
897 * ==========================================================================
901 * Add a reference to the given spa_t. Must have at least one reference, or
902 * have the namespace lock held.
905 spa_open_ref(spa_t *spa, void *tag)
907 ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
908 MUTEX_HELD(&spa_namespace_lock));
909 (void) refcount_add(&spa->spa_refcount, tag);
913 * Remove a reference to the given spa_t. Must have at least one reference, or
914 * have the namespace lock held.
917 spa_close(spa_t *spa, void *tag)
919 ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
920 MUTEX_HELD(&spa_namespace_lock));
921 (void) refcount_remove(&spa->spa_refcount, tag);
925 * Remove a reference to the given spa_t held by a dsl dir that is
926 * being asynchronously released. Async releases occur from a taskq
927 * performing eviction of dsl datasets and dirs. The namespace lock
928 * isn't held and the hold by the object being evicted may contribute to
929 * spa_minref (e.g. dataset or directory released during pool export),
930 * so the asserts in spa_close() do not apply.
933 spa_async_close(spa_t *spa, void *tag)
935 (void) refcount_remove(&spa->spa_refcount, tag);
939 * Check to see if the spa refcount is zero. Must be called with
940 * spa_namespace_lock held. We really compare against spa_minref, which is the
941 * number of references acquired when opening a pool
944 spa_refcount_zero(spa_t *spa)
946 ASSERT(MUTEX_HELD(&spa_namespace_lock));
948 return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
952 * ==========================================================================
953 * SPA spare and l2cache tracking
954 * ==========================================================================
958 * Hot spares and cache devices are tracked using the same code below,
959 * for 'auxiliary' devices.
962 typedef struct spa_aux {
970 spa_aux_compare(const void *a, const void *b)
972 const spa_aux_t *sa = a;
973 const spa_aux_t *sb = b;
975 if (sa->aux_guid < sb->aux_guid)
977 else if (sa->aux_guid > sb->aux_guid)
984 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
990 search.aux_guid = vd->vdev_guid;
991 if ((aux = avl_find(avl, &search, &where)) != NULL) {
994 aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
995 aux->aux_guid = vd->vdev_guid;
997 avl_insert(avl, aux, where);
1002 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
1008 search.aux_guid = vd->vdev_guid;
1009 aux = avl_find(avl, &search, &where);
1011 ASSERT(aux != NULL);
1013 if (--aux->aux_count == 0) {
1014 avl_remove(avl, aux);
1015 kmem_free(aux, sizeof (spa_aux_t));
1016 } else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
1017 aux->aux_pool = 0ULL;
1022 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
1024 spa_aux_t search, *found;
1026 search.aux_guid = guid;
1027 found = avl_find(avl, &search, NULL);
1031 *pool = found->aux_pool;
1038 *refcnt = found->aux_count;
1043 return (found != NULL);
1047 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
1049 spa_aux_t search, *found;
1052 search.aux_guid = vd->vdev_guid;
1053 found = avl_find(avl, &search, &where);
1054 ASSERT(found != NULL);
1055 ASSERT(found->aux_pool == 0ULL);
1057 found->aux_pool = spa_guid(vd->vdev_spa);
1061 * Spares are tracked globally due to the following constraints:
1063 * - A spare may be part of multiple pools.
1064 * - A spare may be added to a pool even if it's actively in use within
1066 * - A spare in use in any pool can only be the source of a replacement if
1067 * the target is a spare in the same pool.
1069 * We keep track of all spares on the system through the use of a reference
1070 * counted AVL tree. When a vdev is added as a spare, or used as a replacement
1071 * spare, then we bump the reference count in the AVL tree. In addition, we set
1072 * the 'vdev_isspare' member to indicate that the device is a spare (active or
1073 * inactive). When a spare is made active (used to replace a device in the
1074 * pool), we also keep track of which pool its been made a part of.
1076 * The 'spa_spare_lock' protects the AVL tree. These functions are normally
1077 * called under the spa_namespace lock as part of vdev reconfiguration. The
1078 * separate spare lock exists for the status query path, which does not need to
1079 * be completely consistent with respect to other vdev configuration changes.
1083 spa_spare_compare(const void *a, const void *b)
1085 return (spa_aux_compare(a, b));
1089 spa_spare_add(vdev_t *vd)
1091 mutex_enter(&spa_spare_lock);
1092 ASSERT(!vd->vdev_isspare);
1093 spa_aux_add(vd, &spa_spare_avl);
1094 vd->vdev_isspare = B_TRUE;
1095 mutex_exit(&spa_spare_lock);
1099 spa_spare_remove(vdev_t *vd)
1101 mutex_enter(&spa_spare_lock);
1102 ASSERT(vd->vdev_isspare);
1103 spa_aux_remove(vd, &spa_spare_avl);
1104 vd->vdev_isspare = B_FALSE;
1105 mutex_exit(&spa_spare_lock);
1109 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
1113 mutex_enter(&spa_spare_lock);
1114 found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
1115 mutex_exit(&spa_spare_lock);
1121 spa_spare_activate(vdev_t *vd)
1123 mutex_enter(&spa_spare_lock);
1124 ASSERT(vd->vdev_isspare);
1125 spa_aux_activate(vd, &spa_spare_avl);
1126 mutex_exit(&spa_spare_lock);
1130 * Level 2 ARC devices are tracked globally for the same reasons as spares.
1131 * Cache devices currently only support one pool per cache device, and so
1132 * for these devices the aux reference count is currently unused beyond 1.
1136 spa_l2cache_compare(const void *a, const void *b)
1138 return (spa_aux_compare(a, b));
1142 spa_l2cache_add(vdev_t *vd)
1144 mutex_enter(&spa_l2cache_lock);
1145 ASSERT(!vd->vdev_isl2cache);
1146 spa_aux_add(vd, &spa_l2cache_avl);
1147 vd->vdev_isl2cache = B_TRUE;
1148 mutex_exit(&spa_l2cache_lock);
1152 spa_l2cache_remove(vdev_t *vd)
1154 mutex_enter(&spa_l2cache_lock);
1155 ASSERT(vd->vdev_isl2cache);
1156 spa_aux_remove(vd, &spa_l2cache_avl);
1157 vd->vdev_isl2cache = B_FALSE;
1158 mutex_exit(&spa_l2cache_lock);
1162 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
1166 mutex_enter(&spa_l2cache_lock);
1167 found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
1168 mutex_exit(&spa_l2cache_lock);
1174 spa_l2cache_activate(vdev_t *vd)
1176 mutex_enter(&spa_l2cache_lock);
1177 ASSERT(vd->vdev_isl2cache);
1178 spa_aux_activate(vd, &spa_l2cache_avl);
1179 mutex_exit(&spa_l2cache_lock);
1183 * ==========================================================================
1185 * ==========================================================================
1189 * Lock the given spa_t for the purpose of adding or removing a vdev.
1190 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1191 * It returns the next transaction group for the spa_t.
1194 spa_vdev_enter(spa_t *spa)
1196 mutex_enter(&spa->spa_vdev_top_lock);
1197 mutex_enter(&spa_namespace_lock);
1198 return (spa_vdev_config_enter(spa));
1202 * Internal implementation for spa_vdev_enter(). Used when a vdev
1203 * operation requires multiple syncs (i.e. removing a device) while
1204 * keeping the spa_namespace_lock held.
1207 spa_vdev_config_enter(spa_t *spa)
1209 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1211 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1213 return (spa_last_synced_txg(spa) + 1);
1217 * Used in combination with spa_vdev_config_enter() to allow the syncing
1218 * of multiple transactions without releasing the spa_namespace_lock.
1221 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
1223 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1225 int config_changed = B_FALSE;
1227 ASSERT(txg > spa_last_synced_txg(spa));
1229 spa->spa_pending_vdev = NULL;
1232 * Reassess the DTLs.
1234 vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
1236 if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
1237 config_changed = B_TRUE;
1238 spa->spa_config_generation++;
1242 * Verify the metaslab classes.
1244 ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
1245 ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
1247 spa_config_exit(spa, SCL_ALL, spa);
1250 * Panic the system if the specified tag requires it. This
1251 * is useful for ensuring that configurations are updated
1254 if (zio_injection_enabled)
1255 zio_handle_panic_injection(spa, tag, 0);
1258 * Note: this txg_wait_synced() is important because it ensures
1259 * that there won't be more than one config change per txg.
1260 * This allows us to use the txg as the generation number.
1263 txg_wait_synced(spa->spa_dsl_pool, txg);
1266 ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
1267 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1269 spa_config_exit(spa, SCL_ALL, spa);
1273 * If the config changed, update the config cache.
1276 spa_config_sync(spa, B_FALSE, B_TRUE);
1280 * Unlock the spa_t after adding or removing a vdev. Besides undoing the
1281 * locking of spa_vdev_enter(), we also want make sure the transactions have
1282 * synced to disk, and then update the global configuration cache with the new
1286 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1288 spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1289 mutex_exit(&spa_namespace_lock);
1290 mutex_exit(&spa->spa_vdev_top_lock);
1296 * Lock the given spa_t for the purpose of changing vdev state.
1299 spa_vdev_state_enter(spa_t *spa, int oplocks)
1301 int locks = SCL_STATE_ALL | oplocks;
1304 * Root pools may need to read of the underlying devfs filesystem
1305 * when opening up a vdev. Unfortunately if we're holding the
1306 * SCL_ZIO lock it will result in a deadlock when we try to issue
1307 * the read from the root filesystem. Instead we "prefetch"
1308 * the associated vnodes that we need prior to opening the
1309 * underlying devices and cache them so that we can prevent
1310 * any I/O when we are doing the actual open.
1312 if (spa_is_root(spa)) {
1313 int low = locks & ~(SCL_ZIO - 1);
1314 int high = locks & ~low;
1316 spa_config_enter(spa, high, spa, RW_WRITER);
1317 vdev_hold(spa->spa_root_vdev);
1318 spa_config_enter(spa, low, spa, RW_WRITER);
1320 spa_config_enter(spa, locks, spa, RW_WRITER);
1322 spa->spa_vdev_locks = locks;
1326 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1328 boolean_t config_changed = B_FALSE;
1330 if (vd != NULL || error == 0)
1331 vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1335 vdev_state_dirty(vd->vdev_top);
1336 config_changed = B_TRUE;
1337 spa->spa_config_generation++;
1340 if (spa_is_root(spa))
1341 vdev_rele(spa->spa_root_vdev);
1343 ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1344 spa_config_exit(spa, spa->spa_vdev_locks, spa);
1347 * If anything changed, wait for it to sync. This ensures that,
1348 * from the system administrator's perspective, zpool(1M) commands
1349 * are synchronous. This is important for things like zpool offline:
1350 * when the command completes, you expect no further I/O from ZFS.
1353 txg_wait_synced(spa->spa_dsl_pool, 0);
1356 * If the config changed, update the config cache.
1358 if (config_changed) {
1359 mutex_enter(&spa_namespace_lock);
1360 spa_config_sync(spa, B_FALSE, B_TRUE);
1361 mutex_exit(&spa_namespace_lock);
1368 * ==========================================================================
1369 * Miscellaneous functions
1370 * ==========================================================================
1374 spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx)
1376 if (!nvlist_exists(spa->spa_label_features, feature)) {
1377 fnvlist_add_boolean(spa->spa_label_features, feature);
1379 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1380 * dirty the vdev config because lock SCL_CONFIG is not held.
1381 * Thankfully, in this case we don't need to dirty the config
1382 * because it will be written out anyway when we finish
1383 * creating the pool.
1385 if (tx->tx_txg != TXG_INITIAL)
1386 vdev_config_dirty(spa->spa_root_vdev);
1391 spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1393 if (nvlist_remove_all(spa->spa_label_features, feature) == 0)
1394 vdev_config_dirty(spa->spa_root_vdev);
1401 spa_rename(const char *name, const char *newname)
1407 * Lookup the spa_t and grab the config lock for writing. We need to
1408 * actually open the pool so that we can sync out the necessary labels.
1409 * It's OK to call spa_open() with the namespace lock held because we
1410 * allow recursive calls for other reasons.
1412 mutex_enter(&spa_namespace_lock);
1413 if ((err = spa_open(name, &spa, FTAG)) != 0) {
1414 mutex_exit(&spa_namespace_lock);
1418 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1420 avl_remove(&spa_namespace_avl, spa);
1421 (void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1422 avl_add(&spa_namespace_avl, spa);
1425 * Sync all labels to disk with the new names by marking the root vdev
1426 * dirty and waiting for it to sync. It will pick up the new pool name
1429 vdev_config_dirty(spa->spa_root_vdev);
1431 spa_config_exit(spa, SCL_ALL, FTAG);
1433 txg_wait_synced(spa->spa_dsl_pool, 0);
1436 * Sync the updated config cache.
1438 spa_config_sync(spa, B_FALSE, B_TRUE);
1440 spa_close(spa, FTAG);
1442 mutex_exit(&spa_namespace_lock);
1448 * Return the spa_t associated with given pool_guid, if it exists. If
1449 * device_guid is non-zero, determine whether the pool exists *and* contains
1450 * a device with the specified device_guid.
1453 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1456 avl_tree_t *t = &spa_namespace_avl;
1458 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1460 for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1461 if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1463 if (spa->spa_root_vdev == NULL)
1465 if (spa_guid(spa) == pool_guid) {
1466 if (device_guid == 0)
1469 if (vdev_lookup_by_guid(spa->spa_root_vdev,
1470 device_guid) != NULL)
1474 * Check any devices we may be in the process of adding.
1476 if (spa->spa_pending_vdev) {
1477 if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1478 device_guid) != NULL)
1488 * Determine whether a pool with the given pool_guid exists.
1491 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1493 return (spa_by_guid(pool_guid, device_guid) != NULL);
1497 spa_strdup(const char *s)
1503 new = kmem_alloc(len + 1, KM_SLEEP);
1511 spa_strfree(char *s)
1513 kmem_free(s, strlen(s) + 1);
1517 spa_get_random(uint64_t range)
1523 (void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1529 spa_generate_guid(spa_t *spa)
1531 uint64_t guid = spa_get_random(-1ULL);
1534 while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1535 guid = spa_get_random(-1ULL);
1537 while (guid == 0 || spa_guid_exists(guid, 0))
1538 guid = spa_get_random(-1ULL);
1545 snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp)
1548 char *checksum = NULL;
1549 char *compress = NULL;
1552 if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1553 dmu_object_byteswap_t bswap =
1554 DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1555 (void) snprintf(type, sizeof (type), "bswap %s %s",
1556 DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1557 "metadata" : "data",
1558 dmu_ot_byteswap[bswap].ob_name);
1560 (void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1563 if (!BP_IS_EMBEDDED(bp)) {
1565 zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1567 compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1570 SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum,
1575 spa_freeze(spa_t *spa)
1577 uint64_t freeze_txg = 0;
1579 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1580 if (spa->spa_freeze_txg == UINT64_MAX) {
1581 freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1582 spa->spa_freeze_txg = freeze_txg;
1584 spa_config_exit(spa, SCL_ALL, FTAG);
1585 if (freeze_txg != 0)
1586 txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1590 zfs_panic_recover(const char *fmt, ...)
1595 vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1600 * This is a stripped-down version of strtoull, suitable only for converting
1601 * lowercase hexadecimal numbers that don't overflow.
1604 zfs_strtonum(const char *str, char **nptr)
1610 while ((c = *str) != '\0') {
1611 if (c >= '0' && c <= '9')
1613 else if (c >= 'a' && c <= 'f')
1614 digit = 10 + c - 'a';
1625 *nptr = (char *)str;
1631 * ==========================================================================
1632 * Accessor functions
1633 * ==========================================================================
1637 spa_shutting_down(spa_t *spa)
1639 return (spa->spa_async_suspended);
1643 spa_get_dsl(spa_t *spa)
1645 return (spa->spa_dsl_pool);
1649 spa_is_initializing(spa_t *spa)
1651 return (spa->spa_is_initializing);
1655 spa_get_rootblkptr(spa_t *spa)
1657 return (&spa->spa_ubsync.ub_rootbp);
1661 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1663 spa->spa_uberblock.ub_rootbp = *bp;
1667 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1669 if (spa->spa_root == NULL)
1672 (void) strncpy(buf, spa->spa_root, buflen);
1676 spa_sync_pass(spa_t *spa)
1678 return (spa->spa_sync_pass);
1682 spa_name(spa_t *spa)
1684 return (spa->spa_name);
1688 spa_guid(spa_t *spa)
1690 dsl_pool_t *dp = spa_get_dsl(spa);
1694 * If we fail to parse the config during spa_load(), we can go through
1695 * the error path (which posts an ereport) and end up here with no root
1696 * vdev. We stash the original pool guid in 'spa_config_guid' to handle
1699 if (spa->spa_root_vdev == NULL)
1700 return (spa->spa_config_guid);
1702 guid = spa->spa_last_synced_guid != 0 ?
1703 spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1706 * Return the most recently synced out guid unless we're
1707 * in syncing context.
1709 if (dp && dsl_pool_sync_context(dp))
1710 return (spa->spa_root_vdev->vdev_guid);
1716 spa_load_guid(spa_t *spa)
1719 * This is a GUID that exists solely as a reference for the
1720 * purposes of the arc. It is generated at load time, and
1721 * is never written to persistent storage.
1723 return (spa->spa_load_guid);
1727 spa_last_synced_txg(spa_t *spa)
1729 return (spa->spa_ubsync.ub_txg);
1733 spa_first_txg(spa_t *spa)
1735 return (spa->spa_first_txg);
1739 spa_syncing_txg(spa_t *spa)
1741 return (spa->spa_syncing_txg);
1745 spa_state(spa_t *spa)
1747 return (spa->spa_state);
1751 spa_load_state(spa_t *spa)
1753 return (spa->spa_load_state);
1757 spa_freeze_txg(spa_t *spa)
1759 return (spa->spa_freeze_txg);
1764 spa_get_asize(spa_t *spa, uint64_t lsize)
1766 return (lsize * spa_asize_inflation);
1770 * Return the amount of slop space in bytes. It is 1/32 of the pool (3.2%),
1771 * or at least 128MB, unless that would cause it to be more than half the
1774 * See the comment above spa_slop_shift for details.
1777 spa_get_slop_space(spa_t *spa)
1779 uint64_t space = spa_get_dspace(spa);
1780 return (MAX(space >> spa_slop_shift, MIN(space >> 1, spa_min_slop)));
1784 spa_get_dspace(spa_t *spa)
1786 return (spa->spa_dspace);
1790 spa_update_dspace(spa_t *spa)
1792 spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1793 ddt_get_dedup_dspace(spa);
1797 * Return the failure mode that has been set to this pool. The default
1798 * behavior will be to block all I/Os when a complete failure occurs.
1801 spa_get_failmode(spa_t *spa)
1803 return (spa->spa_failmode);
1807 spa_suspended(spa_t *spa)
1809 return (spa->spa_suspended);
1813 spa_version(spa_t *spa)
1815 return (spa->spa_ubsync.ub_version);
1819 spa_deflate(spa_t *spa)
1821 return (spa->spa_deflate);
1825 spa_normal_class(spa_t *spa)
1827 return (spa->spa_normal_class);
1831 spa_log_class(spa_t *spa)
1833 return (spa->spa_log_class);
1837 spa_evicting_os_register(spa_t *spa, objset_t *os)
1839 mutex_enter(&spa->spa_evicting_os_lock);
1840 list_insert_head(&spa->spa_evicting_os_list, os);
1841 mutex_exit(&spa->spa_evicting_os_lock);
1845 spa_evicting_os_deregister(spa_t *spa, objset_t *os)
1847 mutex_enter(&spa->spa_evicting_os_lock);
1848 list_remove(&spa->spa_evicting_os_list, os);
1849 cv_broadcast(&spa->spa_evicting_os_cv);
1850 mutex_exit(&spa->spa_evicting_os_lock);
1854 spa_evicting_os_wait(spa_t *spa)
1856 mutex_enter(&spa->spa_evicting_os_lock);
1857 while (!list_is_empty(&spa->spa_evicting_os_list))
1858 cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock);
1859 mutex_exit(&spa->spa_evicting_os_lock);
1861 dmu_buf_user_evict_wait();
1865 spa_max_replication(spa_t *spa)
1868 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1869 * handle BPs with more than one DVA allocated. Set our max
1870 * replication level accordingly.
1872 if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1874 return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1878 spa_prev_software_version(spa_t *spa)
1880 return (spa->spa_prev_software_version);
1884 spa_deadman_synctime(spa_t *spa)
1886 return (spa->spa_deadman_synctime);
1890 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1892 uint64_t asize = DVA_GET_ASIZE(dva);
1893 uint64_t dsize = asize;
1895 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1897 if (asize != 0 && spa->spa_deflate) {
1898 uint64_t vdev = DVA_GET_VDEV(dva);
1899 vdev_t *vd = vdev_lookup_top(spa, vdev);
1902 "dva_get_dsize_sync(): bad DVA %llu:%llu",
1903 (u_longlong_t)vdev, (u_longlong_t)asize);
1905 dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1912 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1916 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1917 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1923 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1927 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1929 for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1930 dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1932 spa_config_exit(spa, SCL_VDEV, FTAG);
1938 * ==========================================================================
1939 * Initialization and Termination
1940 * ==========================================================================
1944 spa_name_compare(const void *a1, const void *a2)
1946 const spa_t *s1 = a1;
1947 const spa_t *s2 = a2;
1950 s = strcmp(s1->spa_name, s2->spa_name);
1961 return (spa_active_count);
1971 EVENTHANDLER_DEFINE(mountroot, spa_boot_init, NULL, 0);
1977 mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1978 mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1979 mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1980 cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1982 avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1983 offsetof(spa_t, spa_avl));
1985 avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1986 offsetof(spa_aux_t, aux_avl));
1988 avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1989 offsetof(spa_aux_t, aux_avl));
1991 spa_mode_global = mode;
1997 if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
1998 arc_procfd = open("/proc/self/ctl", O_WRONLY);
1999 if (arc_procfd == -1) {
2000 perror("could not enable watchpoints: "
2001 "opening /proc/self/ctl failed: ");
2007 #endif /* illumos */
2015 vdev_cache_stat_init();
2018 zpool_feature_init();
2025 #endif /* !illumos */
2035 vdev_cache_stat_fini();
2044 avl_destroy(&spa_namespace_avl);
2045 avl_destroy(&spa_spare_avl);
2046 avl_destroy(&spa_l2cache_avl);
2048 cv_destroy(&spa_namespace_cv);
2049 mutex_destroy(&spa_namespace_lock);
2050 mutex_destroy(&spa_spare_lock);
2051 mutex_destroy(&spa_l2cache_lock);
2055 * Return whether this pool has slogs. No locking needed.
2056 * It's not a problem if the wrong answer is returned as it's only for
2057 * performance and not correctness
2060 spa_has_slogs(spa_t *spa)
2062 return (spa->spa_log_class->mc_rotor != NULL);
2066 spa_get_log_state(spa_t *spa)
2068 return (spa->spa_log_state);
2072 spa_set_log_state(spa_t *spa, spa_log_state_t state)
2074 spa->spa_log_state = state;
2078 spa_is_root(spa_t *spa)
2080 return (spa->spa_is_root);
2084 spa_writeable(spa_t *spa)
2086 return (!!(spa->spa_mode & FWRITE));
2090 * Returns true if there is a pending sync task in any of the current
2091 * syncing txg, the current quiescing txg, or the current open txg.
2094 spa_has_pending_synctask(spa_t *spa)
2096 return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks));
2100 spa_mode(spa_t *spa)
2102 return (spa->spa_mode);
2106 spa_bootfs(spa_t *spa)
2108 return (spa->spa_bootfs);
2112 spa_delegation(spa_t *spa)
2114 return (spa->spa_delegation);
2118 spa_meta_objset(spa_t *spa)
2120 return (spa->spa_meta_objset);
2124 spa_dedup_checksum(spa_t *spa)
2126 return (spa->spa_dedup_checksum);
2130 * Reset pool scan stat per scan pass (or reboot).
2133 spa_scan_stat_init(spa_t *spa)
2135 /* data not stored on disk */
2136 spa->spa_scan_pass_start = gethrestime_sec();
2137 spa->spa_scan_pass_exam = 0;
2138 vdev_scan_stat_init(spa->spa_root_vdev);
2142 * Get scan stats for zpool status reports
2145 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
2147 dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
2149 if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
2150 return (SET_ERROR(ENOENT));
2151 bzero(ps, sizeof (pool_scan_stat_t));
2153 /* data stored on disk */
2154 ps->pss_func = scn->scn_phys.scn_func;
2155 ps->pss_start_time = scn->scn_phys.scn_start_time;
2156 ps->pss_end_time = scn->scn_phys.scn_end_time;
2157 ps->pss_to_examine = scn->scn_phys.scn_to_examine;
2158 ps->pss_examined = scn->scn_phys.scn_examined;
2159 ps->pss_to_process = scn->scn_phys.scn_to_process;
2160 ps->pss_processed = scn->scn_phys.scn_processed;
2161 ps->pss_errors = scn->scn_phys.scn_errors;
2162 ps->pss_state = scn->scn_phys.scn_state;
2164 /* data not stored on disk */
2165 ps->pss_pass_start = spa->spa_scan_pass_start;
2166 ps->pss_pass_exam = spa->spa_scan_pass_exam;
2172 spa_debug_enabled(spa_t *spa)
2174 return (spa->spa_debug);
2178 spa_maxblocksize(spa_t *spa)
2180 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS))
2181 return (SPA_MAXBLOCKSIZE);
2183 return (SPA_OLD_MAXBLOCKSIZE);