spa_misc.c revision e0f1c0afa46cc84d4b1e40124032a9a87310386e
1/*
2 * CDDL HEADER START
3 *
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.
7 *
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.
12 *
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]
18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
24 * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
25 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
26 * Copyright 2013 Saso Kiselkov. All rights reserved.
27 * Copyright (c) 2014 Integros [integros.com]
28 * Copyright (c) 2017 Datto Inc.
29 */
30
31#include <sys/zfs_context.h>
32#include <sys/spa_impl.h>
33#include <sys/spa_boot.h>
34#include <sys/zio.h>
35#include <sys/zio_checksum.h>
36#include <sys/zio_compress.h>
37#include <sys/dmu.h>
38#include <sys/dmu_tx.h>
39#include <sys/zap.h>
40#include <sys/zil.h>
41#include <sys/vdev_impl.h>
42#include <sys/vdev_initialize.h>
43#include <sys/metaslab.h>
44#include <sys/uberblock_impl.h>
45#include <sys/txg.h>
46#include <sys/avl.h>
47#include <sys/unique.h>
48#include <sys/dsl_pool.h>
49#include <sys/dsl_dir.h>
50#include <sys/dsl_prop.h>
51#include <sys/dsl_scan.h>
52#include <sys/fs/zfs.h>
53#include <sys/metaslab_impl.h>
54#include <sys/arc.h>
55#include <sys/ddt.h>
56#include "zfs_prop.h"
57#include <sys/zfeature.h>
58
59/*
60 * SPA locking
61 *
62 * There are four basic locks for managing spa_t structures:
63 *
64 * spa_namespace_lock (global mutex)
65 *
66 *	This lock must be acquired to do any of the following:
67 *
68 *		- Lookup a spa_t by name
69 *		- Add or remove a spa_t from the namespace
70 *		- Increase spa_refcount from non-zero
71 *		- Check if spa_refcount is zero
72 *		- Rename a spa_t
73 *		- add/remove/attach/detach devices
74 *		- Held for the duration of create/destroy/import/export
75 *
76 *	It does not need to handle recursion.  A create or destroy may
77 *	reference objects (files or zvols) in other pools, but by
78 *	definition they must have an existing reference, and will never need
79 *	to lookup a spa_t by name.
80 *
81 * spa_refcount (per-spa zfs_refcount_t protected by mutex)
82 *
83 *	This reference count keep track of any active users of the spa_t.  The
84 *	spa_t cannot be destroyed or freed while this is non-zero.  Internally,
85 *	the refcount is never really 'zero' - opening a pool implicitly keeps
86 *	some references in the DMU.  Internally we check against spa_minref, but
87 *	present the image of a zero/non-zero value to consumers.
88 *
89 * spa_config_lock[] (per-spa array of rwlocks)
90 *
91 *	This protects the spa_t from config changes, and must be held in
92 *	the following circumstances:
93 *
94 *		- RW_READER to perform I/O to the spa
95 *		- RW_WRITER to change the vdev config
96 *
97 * The locking order is fairly straightforward:
98 *
99 *		spa_namespace_lock	->	spa_refcount
100 *
101 *	The namespace lock must be acquired to increase the refcount from 0
102 *	or to check if it is zero.
103 *
104 *		spa_refcount		->	spa_config_lock[]
105 *
106 *	There must be at least one valid reference on the spa_t to acquire
107 *	the config lock.
108 *
109 *		spa_namespace_lock	->	spa_config_lock[]
110 *
111 *	The namespace lock must always be taken before the config lock.
112 *
113 *
114 * The spa_namespace_lock can be acquired directly and is globally visible.
115 *
116 * The namespace is manipulated using the following functions, all of which
117 * require the spa_namespace_lock to be held.
118 *
119 *	spa_lookup()		Lookup a spa_t by name.
120 *
121 *	spa_add()		Create a new spa_t in the namespace.
122 *
123 *	spa_remove()		Remove a spa_t from the namespace.  This also
124 *				frees up any memory associated with the spa_t.
125 *
126 *	spa_next()		Returns the next spa_t in the system, or the
127 *				first if NULL is passed.
128 *
129 *	spa_evict_all()		Shutdown and remove all spa_t structures in
130 *				the system.
131 *
132 *	spa_guid_exists()	Determine whether a pool/device guid exists.
133 *
134 * The spa_refcount is manipulated using the following functions:
135 *
136 *	spa_open_ref()		Adds a reference to the given spa_t.  Must be
137 *				called with spa_namespace_lock held if the
138 *				refcount is currently zero.
139 *
140 *	spa_close()		Remove a reference from the spa_t.  This will
141 *				not free the spa_t or remove it from the
142 *				namespace.  No locking is required.
143 *
144 *	spa_refcount_zero()	Returns true if the refcount is currently
145 *				zero.  Must be called with spa_namespace_lock
146 *				held.
147 *
148 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
149 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
150 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
151 *
152 * To read the configuration, it suffices to hold one of these locks as reader.
153 * To modify the configuration, you must hold all locks as writer.  To modify
154 * vdev state without altering the vdev tree's topology (e.g. online/offline),
155 * you must hold SCL_STATE and SCL_ZIO as writer.
156 *
157 * We use these distinct config locks to avoid recursive lock entry.
158 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
159 * block allocations (SCL_ALLOC), which may require reading space maps
160 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
161 *
162 * The spa config locks cannot be normal rwlocks because we need the
163 * ability to hand off ownership.  For example, SCL_ZIO is acquired
164 * by the issuing thread and later released by an interrupt thread.
165 * They do, however, obey the usual write-wanted semantics to prevent
166 * writer (i.e. system administrator) starvation.
167 *
168 * The lock acquisition rules are as follows:
169 *
170 * SCL_CONFIG
171 *	Protects changes to the vdev tree topology, such as vdev
172 *	add/remove/attach/detach.  Protects the dirty config list
173 *	(spa_config_dirty_list) and the set of spares and l2arc devices.
174 *
175 * SCL_STATE
176 *	Protects changes to pool state and vdev state, such as vdev
177 *	online/offline/fault/degrade/clear.  Protects the dirty state list
178 *	(spa_state_dirty_list) and global pool state (spa_state).
179 *
180 * SCL_ALLOC
181 *	Protects changes to metaslab groups and classes.
182 *	Held as reader by metaslab_alloc() and metaslab_claim().
183 *
184 * SCL_ZIO
185 *	Held by bp-level zios (those which have no io_vd upon entry)
186 *	to prevent changes to the vdev tree.  The bp-level zio implicitly
187 *	protects all of its vdev child zios, which do not hold SCL_ZIO.
188 *
189 * SCL_FREE
190 *	Protects changes to metaslab groups and classes.
191 *	Held as reader by metaslab_free().  SCL_FREE is distinct from
192 *	SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
193 *	blocks in zio_done() while another i/o that holds either
194 *	SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
195 *
196 * SCL_VDEV
197 *	Held as reader to prevent changes to the vdev tree during trivial
198 *	inquiries such as bp_get_dsize().  SCL_VDEV is distinct from the
199 *	other locks, and lower than all of them, to ensure that it's safe
200 *	to acquire regardless of caller context.
201 *
202 * In addition, the following rules apply:
203 *
204 * (a)	spa_props_lock protects pool properties, spa_config and spa_config_list.
205 *	The lock ordering is SCL_CONFIG > spa_props_lock.
206 *
207 * (b)	I/O operations on leaf vdevs.  For any zio operation that takes
208 *	an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
209 *	or zio_write_phys() -- the caller must ensure that the config cannot
210 *	cannot change in the interim, and that the vdev cannot be reopened.
211 *	SCL_STATE as reader suffices for both.
212 *
213 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
214 *
215 *	spa_vdev_enter()	Acquire the namespace lock and the config lock
216 *				for writing.
217 *
218 *	spa_vdev_exit()		Release the config lock, wait for all I/O
219 *				to complete, sync the updated configs to the
220 *				cache, and release the namespace lock.
221 *
222 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
223 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
224 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
225 */
226
227static avl_tree_t spa_namespace_avl;
228kmutex_t spa_namespace_lock;
229static kcondvar_t spa_namespace_cv;
230static int spa_active_count;
231int spa_max_replication_override = SPA_DVAS_PER_BP;
232
233static kmutex_t spa_spare_lock;
234static avl_tree_t spa_spare_avl;
235static kmutex_t spa_l2cache_lock;
236static avl_tree_t spa_l2cache_avl;
237
238kmem_cache_t *spa_buffer_pool;
239int spa_mode_global;
240
241#ifdef ZFS_DEBUG
242/*
243 * Everything except dprintf, spa, and indirect_remap is on by default
244 * in debug builds.
245 */
246int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_INDIRECT_REMAP);
247#else
248int zfs_flags = 0;
249#endif
250
251/*
252 * zfs_recover can be set to nonzero to attempt to recover from
253 * otherwise-fatal errors, typically caused by on-disk corruption.  When
254 * set, calls to zfs_panic_recover() will turn into warning messages.
255 * This should only be used as a last resort, as it typically results
256 * in leaked space, or worse.
257 */
258boolean_t zfs_recover = B_FALSE;
259
260/*
261 * If destroy encounters an EIO while reading metadata (e.g. indirect
262 * blocks), space referenced by the missing metadata can not be freed.
263 * Normally this causes the background destroy to become "stalled", as
264 * it is unable to make forward progress.  While in this stalled state,
265 * all remaining space to free from the error-encountering filesystem is
266 * "temporarily leaked".  Set this flag to cause it to ignore the EIO,
267 * permanently leak the space from indirect blocks that can not be read,
268 * and continue to free everything else that it can.
269 *
270 * The default, "stalling" behavior is useful if the storage partially
271 * fails (i.e. some but not all i/os fail), and then later recovers.  In
272 * this case, we will be able to continue pool operations while it is
273 * partially failed, and when it recovers, we can continue to free the
274 * space, with no leaks.  However, note that this case is actually
275 * fairly rare.
276 *
277 * Typically pools either (a) fail completely (but perhaps temporarily,
278 * e.g. a top-level vdev going offline), or (b) have localized,
279 * permanent errors (e.g. disk returns the wrong data due to bit flip or
280 * firmware bug).  In case (a), this setting does not matter because the
281 * pool will be suspended and the sync thread will not be able to make
282 * forward progress regardless.  In case (b), because the error is
283 * permanent, the best we can do is leak the minimum amount of space,
284 * which is what setting this flag will do.  Therefore, it is reasonable
285 * for this flag to normally be set, but we chose the more conservative
286 * approach of not setting it, so that there is no possibility of
287 * leaking space in the "partial temporary" failure case.
288 */
289boolean_t zfs_free_leak_on_eio = B_FALSE;
290
291/*
292 * Expiration time in milliseconds. This value has two meanings. First it is
293 * used to determine when the spa_deadman() logic should fire. By default the
294 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
295 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
296 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
297 * in a system panic.
298 */
299uint64_t zfs_deadman_synctime_ms = 1000000ULL;
300
301/*
302 * Check time in milliseconds. This defines the frequency at which we check
303 * for hung I/O.
304 */
305uint64_t zfs_deadman_checktime_ms = 5000ULL;
306
307/*
308 * Override the zfs deadman behavior via /etc/system. By default the
309 * deadman is enabled except on VMware and sparc deployments.
310 */
311int zfs_deadman_enabled = -1;
312
313/*
314 * The worst case is single-sector max-parity RAID-Z blocks, in which
315 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
316 * times the size; so just assume that.  Add to this the fact that
317 * we can have up to 3 DVAs per bp, and one more factor of 2 because
318 * the block may be dittoed with up to 3 DVAs by ddt_sync().  All together,
319 * the worst case is:
320 *     (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
321 */
322int spa_asize_inflation = 24;
323
324/*
325 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
326 * the pool to be consumed.  This ensures that we don't run the pool
327 * completely out of space, due to unaccounted changes (e.g. to the MOS).
328 * It also limits the worst-case time to allocate space.  If we have
329 * less than this amount of free space, most ZPL operations (e.g. write,
330 * create) will return ENOSPC.
331 *
332 * Certain operations (e.g. file removal, most administrative actions) can
333 * use half the slop space.  They will only return ENOSPC if less than half
334 * the slop space is free.  Typically, once the pool has less than the slop
335 * space free, the user will use these operations to free up space in the pool.
336 * These are the operations that call dsl_pool_adjustedsize() with the netfree
337 * argument set to TRUE.
338 *
339 * Operations that are almost guaranteed to free up space in the absence of
340 * a pool checkpoint can use up to three quarters of the slop space
341 * (e.g zfs destroy).
342 *
343 * A very restricted set of operations are always permitted, regardless of
344 * the amount of free space.  These are the operations that call
345 * dsl_sync_task(ZFS_SPACE_CHECK_NONE). If these operations result in a net
346 * increase in the amount of space used, it is possible to run the pool
347 * completely out of space, causing it to be permanently read-only.
348 *
349 * Note that on very small pools, the slop space will be larger than
350 * 3.2%, in an effort to have it be at least spa_min_slop (128MB),
351 * but we never allow it to be more than half the pool size.
352 *
353 * See also the comments in zfs_space_check_t.
354 */
355int spa_slop_shift = 5;
356uint64_t spa_min_slop = 128 * 1024 * 1024;
357
358int spa_allocators = 4;
359
360/*PRINTFLIKE2*/
361void
362spa_load_failed(spa_t *spa, const char *fmt, ...)
363{
364	va_list adx;
365	char buf[256];
366
367	va_start(adx, fmt);
368	(void) vsnprintf(buf, sizeof (buf), fmt, adx);
369	va_end(adx);
370
371	zfs_dbgmsg("spa_load(%s, config %s): FAILED: %s", spa->spa_name,
372	    spa->spa_trust_config ? "trusted" : "untrusted", buf);
373}
374
375/*PRINTFLIKE2*/
376void
377spa_load_note(spa_t *spa, const char *fmt, ...)
378{
379	va_list adx;
380	char buf[256];
381
382	va_start(adx, fmt);
383	(void) vsnprintf(buf, sizeof (buf), fmt, adx);
384	va_end(adx);
385
386	zfs_dbgmsg("spa_load(%s, config %s): %s", spa->spa_name,
387	    spa->spa_trust_config ? "trusted" : "untrusted", buf);
388}
389
390/*
391 * ==========================================================================
392 * SPA config locking
393 * ==========================================================================
394 */
395static void
396spa_config_lock_init(spa_t *spa)
397{
398	for (int i = 0; i < SCL_LOCKS; i++) {
399		spa_config_lock_t *scl = &spa->spa_config_lock[i];
400		mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
401		cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
402		zfs_refcount_create_untracked(&scl->scl_count);
403		scl->scl_writer = NULL;
404		scl->scl_write_wanted = 0;
405	}
406}
407
408static void
409spa_config_lock_destroy(spa_t *spa)
410{
411	for (int i = 0; i < SCL_LOCKS; i++) {
412		spa_config_lock_t *scl = &spa->spa_config_lock[i];
413		mutex_destroy(&scl->scl_lock);
414		cv_destroy(&scl->scl_cv);
415		zfs_refcount_destroy(&scl->scl_count);
416		ASSERT(scl->scl_writer == NULL);
417		ASSERT(scl->scl_write_wanted == 0);
418	}
419}
420
421int
422spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
423{
424	for (int i = 0; i < SCL_LOCKS; i++) {
425		spa_config_lock_t *scl = &spa->spa_config_lock[i];
426		if (!(locks & (1 << i)))
427			continue;
428		mutex_enter(&scl->scl_lock);
429		if (rw == RW_READER) {
430			if (scl->scl_writer || scl->scl_write_wanted) {
431				mutex_exit(&scl->scl_lock);
432				spa_config_exit(spa, locks & ((1 << i) - 1),
433				    tag);
434				return (0);
435			}
436		} else {
437			ASSERT(scl->scl_writer != curthread);
438			if (!zfs_refcount_is_zero(&scl->scl_count)) {
439				mutex_exit(&scl->scl_lock);
440				spa_config_exit(spa, locks & ((1 << i) - 1),
441				    tag);
442				return (0);
443			}
444			scl->scl_writer = curthread;
445		}
446		(void) zfs_refcount_add(&scl->scl_count, tag);
447		mutex_exit(&scl->scl_lock);
448	}
449	return (1);
450}
451
452void
453spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
454{
455	int wlocks_held = 0;
456
457	ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
458
459	for (int i = 0; i < SCL_LOCKS; i++) {
460		spa_config_lock_t *scl = &spa->spa_config_lock[i];
461		if (scl->scl_writer == curthread)
462			wlocks_held |= (1 << i);
463		if (!(locks & (1 << i)))
464			continue;
465		mutex_enter(&scl->scl_lock);
466		if (rw == RW_READER) {
467			while (scl->scl_writer || scl->scl_write_wanted) {
468				cv_wait(&scl->scl_cv, &scl->scl_lock);
469			}
470		} else {
471			ASSERT(scl->scl_writer != curthread);
472			while (!zfs_refcount_is_zero(&scl->scl_count)) {
473				scl->scl_write_wanted++;
474				cv_wait(&scl->scl_cv, &scl->scl_lock);
475				scl->scl_write_wanted--;
476			}
477			scl->scl_writer = curthread;
478		}
479		(void) zfs_refcount_add(&scl->scl_count, tag);
480		mutex_exit(&scl->scl_lock);
481	}
482	ASSERT3U(wlocks_held, <=, locks);
483}
484
485void
486spa_config_exit(spa_t *spa, int locks, void *tag)
487{
488	for (int i = SCL_LOCKS - 1; i >= 0; i--) {
489		spa_config_lock_t *scl = &spa->spa_config_lock[i];
490		if (!(locks & (1 << i)))
491			continue;
492		mutex_enter(&scl->scl_lock);
493		ASSERT(!zfs_refcount_is_zero(&scl->scl_count));
494		if (zfs_refcount_remove(&scl->scl_count, tag) == 0) {
495			ASSERT(scl->scl_writer == NULL ||
496			    scl->scl_writer == curthread);
497			scl->scl_writer = NULL;	/* OK in either case */
498			cv_broadcast(&scl->scl_cv);
499		}
500		mutex_exit(&scl->scl_lock);
501	}
502}
503
504int
505spa_config_held(spa_t *spa, int locks, krw_t rw)
506{
507	int locks_held = 0;
508
509	for (int i = 0; i < SCL_LOCKS; i++) {
510		spa_config_lock_t *scl = &spa->spa_config_lock[i];
511		if (!(locks & (1 << i)))
512			continue;
513		if ((rw == RW_READER &&
514		    !zfs_refcount_is_zero(&scl->scl_count)) ||
515		    (rw == RW_WRITER && scl->scl_writer == curthread))
516			locks_held |= 1 << i;
517	}
518
519	return (locks_held);
520}
521
522/*
523 * ==========================================================================
524 * SPA namespace functions
525 * ==========================================================================
526 */
527
528/*
529 * Lookup the named spa_t in the AVL tree.  The spa_namespace_lock must be held.
530 * Returns NULL if no matching spa_t is found.
531 */
532spa_t *
533spa_lookup(const char *name)
534{
535	static spa_t search;	/* spa_t is large; don't allocate on stack */
536	spa_t *spa;
537	avl_index_t where;
538	char *cp;
539
540	ASSERT(MUTEX_HELD(&spa_namespace_lock));
541
542	(void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
543
544	/*
545	 * If it's a full dataset name, figure out the pool name and
546	 * just use that.
547	 */
548	cp = strpbrk(search.spa_name, "/@#");
549	if (cp != NULL)
550		*cp = '\0';
551
552	spa = avl_find(&spa_namespace_avl, &search, &where);
553
554	return (spa);
555}
556
557/*
558 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
559 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
560 * looking for potentially hung I/Os.
561 */
562void
563spa_deadman(void *arg)
564{
565	spa_t *spa = arg;
566
567	/*
568	 * Disable the deadman timer if the pool is suspended.
569	 */
570	if (spa_suspended(spa)) {
571		VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
572		return;
573	}
574
575	zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
576	    (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
577	    ++spa->spa_deadman_calls);
578	if (zfs_deadman_enabled)
579		vdev_deadman(spa->spa_root_vdev);
580}
581
582/*
583 * Create an uninitialized spa_t with the given name.  Requires
584 * spa_namespace_lock.  The caller must ensure that the spa_t doesn't already
585 * exist by calling spa_lookup() first.
586 */
587spa_t *
588spa_add(const char *name, nvlist_t *config, const char *altroot)
589{
590	spa_t *spa;
591	spa_config_dirent_t *dp;
592	cyc_handler_t hdlr;
593	cyc_time_t when;
594
595	ASSERT(MUTEX_HELD(&spa_namespace_lock));
596
597	spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
598
599	mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
600	mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
601	mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
602	mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL);
603	mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
604	mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
605	mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
606	mutex_init(&spa->spa_cksum_tmpls_lock, NULL, MUTEX_DEFAULT, NULL);
607	mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
608	mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
609	mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
610	mutex_init(&spa->spa_iokstat_lock, NULL, MUTEX_DEFAULT, NULL);
611
612	cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
613	cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL);
614	cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
615	cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
616	cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
617
618	for (int t = 0; t < TXG_SIZE; t++)
619		bplist_create(&spa->spa_free_bplist[t]);
620
621	(void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
622	spa->spa_state = POOL_STATE_UNINITIALIZED;
623	spa->spa_freeze_txg = UINT64_MAX;
624	spa->spa_final_txg = UINT64_MAX;
625	spa->spa_load_max_txg = UINT64_MAX;
626	spa->spa_proc = &p0;
627	spa->spa_proc_state = SPA_PROC_NONE;
628	spa->spa_trust_config = B_TRUE;
629
630	hdlr.cyh_func = spa_deadman;
631	hdlr.cyh_arg = spa;
632	hdlr.cyh_level = CY_LOW_LEVEL;
633
634	spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
635
636	/*
637	 * This determines how often we need to check for hung I/Os after
638	 * the cyclic has already fired. Since checking for hung I/Os is
639	 * an expensive operation we don't want to check too frequently.
640	 * Instead wait for 5 seconds before checking again.
641	 */
642	when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms);
643	when.cyt_when = CY_INFINITY;
644	mutex_enter(&cpu_lock);
645	spa->spa_deadman_cycid = cyclic_add(&hdlr, &when);
646	mutex_exit(&cpu_lock);
647
648	zfs_refcount_create(&spa->spa_refcount);
649	spa_config_lock_init(spa);
650
651	avl_add(&spa_namespace_avl, spa);
652
653	/*
654	 * Set the alternate root, if there is one.
655	 */
656	if (altroot) {
657		spa->spa_root = spa_strdup(altroot);
658		spa_active_count++;
659	}
660
661	spa->spa_alloc_count = spa_allocators;
662	spa->spa_alloc_locks = kmem_zalloc(spa->spa_alloc_count *
663	    sizeof (kmutex_t), KM_SLEEP);
664	spa->spa_alloc_trees = kmem_zalloc(spa->spa_alloc_count *
665	    sizeof (avl_tree_t), KM_SLEEP);
666	for (int i = 0; i < spa->spa_alloc_count; i++) {
667		mutex_init(&spa->spa_alloc_locks[i], NULL, MUTEX_DEFAULT, NULL);
668		avl_create(&spa->spa_alloc_trees[i], zio_bookmark_compare,
669		    sizeof (zio_t), offsetof(zio_t, io_alloc_node));
670	}
671
672	/*
673	 * Every pool starts with the default cachefile
674	 */
675	list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
676	    offsetof(spa_config_dirent_t, scd_link));
677
678	dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
679	dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
680	list_insert_head(&spa->spa_config_list, dp);
681
682	VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
683	    KM_SLEEP) == 0);
684
685	if (config != NULL) {
686		nvlist_t *features;
687
688		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
689		    &features) == 0) {
690			VERIFY(nvlist_dup(features, &spa->spa_label_features,
691			    0) == 0);
692		}
693
694		VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
695	}
696
697	if (spa->spa_label_features == NULL) {
698		VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
699		    KM_SLEEP) == 0);
700	}
701
702	spa->spa_iokstat = kstat_create("zfs", 0, name,
703	    "disk", KSTAT_TYPE_IO, 1, 0);
704	if (spa->spa_iokstat) {
705		spa->spa_iokstat->ks_lock = &spa->spa_iokstat_lock;
706		kstat_install(spa->spa_iokstat);
707	}
708
709	spa->spa_min_ashift = INT_MAX;
710	spa->spa_max_ashift = 0;
711
712	/*
713	 * As a pool is being created, treat all features as disabled by
714	 * setting SPA_FEATURE_DISABLED for all entries in the feature
715	 * refcount cache.
716	 */
717	for (int i = 0; i < SPA_FEATURES; i++) {
718		spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED;
719	}
720
721	list_create(&spa->spa_leaf_list, sizeof (vdev_t),
722	    offsetof(vdev_t, vdev_leaf_node));
723
724	return (spa);
725}
726
727/*
728 * Removes a spa_t from the namespace, freeing up any memory used.  Requires
729 * spa_namespace_lock.  This is called only after the spa_t has been closed and
730 * deactivated.
731 */
732void
733spa_remove(spa_t *spa)
734{
735	spa_config_dirent_t *dp;
736
737	ASSERT(MUTEX_HELD(&spa_namespace_lock));
738	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
739	ASSERT3U(zfs_refcount_count(&spa->spa_refcount), ==, 0);
740
741	nvlist_free(spa->spa_config_splitting);
742
743	avl_remove(&spa_namespace_avl, spa);
744	cv_broadcast(&spa_namespace_cv);
745
746	if (spa->spa_root) {
747		spa_strfree(spa->spa_root);
748		spa_active_count--;
749	}
750
751	while ((dp = list_head(&spa->spa_config_list)) != NULL) {
752		list_remove(&spa->spa_config_list, dp);
753		if (dp->scd_path != NULL)
754			spa_strfree(dp->scd_path);
755		kmem_free(dp, sizeof (spa_config_dirent_t));
756	}
757
758	for (int i = 0; i < spa->spa_alloc_count; i++) {
759		avl_destroy(&spa->spa_alloc_trees[i]);
760		mutex_destroy(&spa->spa_alloc_locks[i]);
761	}
762	kmem_free(spa->spa_alloc_locks, spa->spa_alloc_count *
763	    sizeof (kmutex_t));
764	kmem_free(spa->spa_alloc_trees, spa->spa_alloc_count *
765	    sizeof (avl_tree_t));
766
767	list_destroy(&spa->spa_config_list);
768	list_destroy(&spa->spa_leaf_list);
769
770	nvlist_free(spa->spa_label_features);
771	nvlist_free(spa->spa_load_info);
772	spa_config_set(spa, NULL);
773
774	mutex_enter(&cpu_lock);
775	if (spa->spa_deadman_cycid != CYCLIC_NONE)
776		cyclic_remove(spa->spa_deadman_cycid);
777	mutex_exit(&cpu_lock);
778	spa->spa_deadman_cycid = CYCLIC_NONE;
779
780	zfs_refcount_destroy(&spa->spa_refcount);
781
782	spa_config_lock_destroy(spa);
783
784	kstat_delete(spa->spa_iokstat);
785	spa->spa_iokstat = NULL;
786
787	for (int t = 0; t < TXG_SIZE; t++)
788		bplist_destroy(&spa->spa_free_bplist[t]);
789
790	zio_checksum_templates_free(spa);
791
792	cv_destroy(&spa->spa_async_cv);
793	cv_destroy(&spa->spa_evicting_os_cv);
794	cv_destroy(&spa->spa_proc_cv);
795	cv_destroy(&spa->spa_scrub_io_cv);
796	cv_destroy(&spa->spa_suspend_cv);
797
798	mutex_destroy(&spa->spa_async_lock);
799	mutex_destroy(&spa->spa_errlist_lock);
800	mutex_destroy(&spa->spa_errlog_lock);
801	mutex_destroy(&spa->spa_evicting_os_lock);
802	mutex_destroy(&spa->spa_history_lock);
803	mutex_destroy(&spa->spa_proc_lock);
804	mutex_destroy(&spa->spa_props_lock);
805	mutex_destroy(&spa->spa_cksum_tmpls_lock);
806	mutex_destroy(&spa->spa_scrub_lock);
807	mutex_destroy(&spa->spa_suspend_lock);
808	mutex_destroy(&spa->spa_vdev_top_lock);
809	mutex_destroy(&spa->spa_iokstat_lock);
810
811	kmem_free(spa, sizeof (spa_t));
812}
813
814/*
815 * Given a pool, return the next pool in the namespace, or NULL if there is
816 * none.  If 'prev' is NULL, return the first pool.
817 */
818spa_t *
819spa_next(spa_t *prev)
820{
821	ASSERT(MUTEX_HELD(&spa_namespace_lock));
822
823	if (prev)
824		return (AVL_NEXT(&spa_namespace_avl, prev));
825	else
826		return (avl_first(&spa_namespace_avl));
827}
828
829/*
830 * ==========================================================================
831 * SPA refcount functions
832 * ==========================================================================
833 */
834
835/*
836 * Add a reference to the given spa_t.  Must have at least one reference, or
837 * have the namespace lock held.
838 */
839void
840spa_open_ref(spa_t *spa, void *tag)
841{
842	ASSERT(zfs_refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
843	    MUTEX_HELD(&spa_namespace_lock));
844	(void) zfs_refcount_add(&spa->spa_refcount, tag);
845}
846
847/*
848 * Remove a reference to the given spa_t.  Must have at least one reference, or
849 * have the namespace lock held.
850 */
851void
852spa_close(spa_t *spa, void *tag)
853{
854	ASSERT(zfs_refcount_count(&spa->spa_refcount) > spa->spa_minref ||
855	    MUTEX_HELD(&spa_namespace_lock));
856	(void) zfs_refcount_remove(&spa->spa_refcount, tag);
857}
858
859/*
860 * Remove a reference to the given spa_t held by a dsl dir that is
861 * being asynchronously released.  Async releases occur from a taskq
862 * performing eviction of dsl datasets and dirs.  The namespace lock
863 * isn't held and the hold by the object being evicted may contribute to
864 * spa_minref (e.g. dataset or directory released during pool export),
865 * so the asserts in spa_close() do not apply.
866 */
867void
868spa_async_close(spa_t *spa, void *tag)
869{
870	(void) zfs_refcount_remove(&spa->spa_refcount, tag);
871}
872
873/*
874 * Check to see if the spa refcount is zero.  Must be called with
875 * spa_namespace_lock held.  We really compare against spa_minref, which is the
876 * number of references acquired when opening a pool
877 */
878boolean_t
879spa_refcount_zero(spa_t *spa)
880{
881	ASSERT(MUTEX_HELD(&spa_namespace_lock));
882
883	return (zfs_refcount_count(&spa->spa_refcount) == spa->spa_minref);
884}
885
886/*
887 * ==========================================================================
888 * SPA spare and l2cache tracking
889 * ==========================================================================
890 */
891
892/*
893 * Hot spares and cache devices are tracked using the same code below,
894 * for 'auxiliary' devices.
895 */
896
897typedef struct spa_aux {
898	uint64_t	aux_guid;
899	uint64_t	aux_pool;
900	avl_node_t	aux_avl;
901	int		aux_count;
902} spa_aux_t;
903
904static int
905spa_aux_compare(const void *a, const void *b)
906{
907	const spa_aux_t *sa = a;
908	const spa_aux_t *sb = b;
909
910	if (sa->aux_guid < sb->aux_guid)
911		return (-1);
912	else if (sa->aux_guid > sb->aux_guid)
913		return (1);
914	else
915		return (0);
916}
917
918void
919spa_aux_add(vdev_t *vd, avl_tree_t *avl)
920{
921	avl_index_t where;
922	spa_aux_t search;
923	spa_aux_t *aux;
924
925	search.aux_guid = vd->vdev_guid;
926	if ((aux = avl_find(avl, &search, &where)) != NULL) {
927		aux->aux_count++;
928	} else {
929		aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
930		aux->aux_guid = vd->vdev_guid;
931		aux->aux_count = 1;
932		avl_insert(avl, aux, where);
933	}
934}
935
936void
937spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
938{
939	spa_aux_t search;
940	spa_aux_t *aux;
941	avl_index_t where;
942
943	search.aux_guid = vd->vdev_guid;
944	aux = avl_find(avl, &search, &where);
945
946	ASSERT(aux != NULL);
947
948	if (--aux->aux_count == 0) {
949		avl_remove(avl, aux);
950		kmem_free(aux, sizeof (spa_aux_t));
951	} else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
952		aux->aux_pool = 0ULL;
953	}
954}
955
956boolean_t
957spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
958{
959	spa_aux_t search, *found;
960
961	search.aux_guid = guid;
962	found = avl_find(avl, &search, NULL);
963
964	if (pool) {
965		if (found)
966			*pool = found->aux_pool;
967		else
968			*pool = 0ULL;
969	}
970
971	if (refcnt) {
972		if (found)
973			*refcnt = found->aux_count;
974		else
975			*refcnt = 0;
976	}
977
978	return (found != NULL);
979}
980
981void
982spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
983{
984	spa_aux_t search, *found;
985	avl_index_t where;
986
987	search.aux_guid = vd->vdev_guid;
988	found = avl_find(avl, &search, &where);
989	ASSERT(found != NULL);
990	ASSERT(found->aux_pool == 0ULL);
991
992	found->aux_pool = spa_guid(vd->vdev_spa);
993}
994
995/*
996 * Spares are tracked globally due to the following constraints:
997 *
998 *	- A spare may be part of multiple pools.
999 *	- A spare may be added to a pool even if it's actively in use within
1000 *	  another pool.
1001 *	- A spare in use in any pool can only be the source of a replacement if
1002 *	  the target is a spare in the same pool.
1003 *
1004 * We keep track of all spares on the system through the use of a reference
1005 * counted AVL tree.  When a vdev is added as a spare, or used as a replacement
1006 * spare, then we bump the reference count in the AVL tree.  In addition, we set
1007 * the 'vdev_isspare' member to indicate that the device is a spare (active or
1008 * inactive).  When a spare is made active (used to replace a device in the
1009 * pool), we also keep track of which pool its been made a part of.
1010 *
1011 * The 'spa_spare_lock' protects the AVL tree.  These functions are normally
1012 * called under the spa_namespace lock as part of vdev reconfiguration.  The
1013 * separate spare lock exists for the status query path, which does not need to
1014 * be completely consistent with respect to other vdev configuration changes.
1015 */
1016
1017static int
1018spa_spare_compare(const void *a, const void *b)
1019{
1020	return (spa_aux_compare(a, b));
1021}
1022
1023void
1024spa_spare_add(vdev_t *vd)
1025{
1026	mutex_enter(&spa_spare_lock);
1027	ASSERT(!vd->vdev_isspare);
1028	spa_aux_add(vd, &spa_spare_avl);
1029	vd->vdev_isspare = B_TRUE;
1030	mutex_exit(&spa_spare_lock);
1031}
1032
1033void
1034spa_spare_remove(vdev_t *vd)
1035{
1036	mutex_enter(&spa_spare_lock);
1037	ASSERT(vd->vdev_isspare);
1038	spa_aux_remove(vd, &spa_spare_avl);
1039	vd->vdev_isspare = B_FALSE;
1040	mutex_exit(&spa_spare_lock);
1041}
1042
1043boolean_t
1044spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
1045{
1046	boolean_t found;
1047
1048	mutex_enter(&spa_spare_lock);
1049	found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
1050	mutex_exit(&spa_spare_lock);
1051
1052	return (found);
1053}
1054
1055void
1056spa_spare_activate(vdev_t *vd)
1057{
1058	mutex_enter(&spa_spare_lock);
1059	ASSERT(vd->vdev_isspare);
1060	spa_aux_activate(vd, &spa_spare_avl);
1061	mutex_exit(&spa_spare_lock);
1062}
1063
1064/*
1065 * Level 2 ARC devices are tracked globally for the same reasons as spares.
1066 * Cache devices currently only support one pool per cache device, and so
1067 * for these devices the aux reference count is currently unused beyond 1.
1068 */
1069
1070static int
1071spa_l2cache_compare(const void *a, const void *b)
1072{
1073	return (spa_aux_compare(a, b));
1074}
1075
1076void
1077spa_l2cache_add(vdev_t *vd)
1078{
1079	mutex_enter(&spa_l2cache_lock);
1080	ASSERT(!vd->vdev_isl2cache);
1081	spa_aux_add(vd, &spa_l2cache_avl);
1082	vd->vdev_isl2cache = B_TRUE;
1083	mutex_exit(&spa_l2cache_lock);
1084}
1085
1086void
1087spa_l2cache_remove(vdev_t *vd)
1088{
1089	mutex_enter(&spa_l2cache_lock);
1090	ASSERT(vd->vdev_isl2cache);
1091	spa_aux_remove(vd, &spa_l2cache_avl);
1092	vd->vdev_isl2cache = B_FALSE;
1093	mutex_exit(&spa_l2cache_lock);
1094}
1095
1096boolean_t
1097spa_l2cache_exists(uint64_t guid, uint64_t *pool)
1098{
1099	boolean_t found;
1100
1101	mutex_enter(&spa_l2cache_lock);
1102	found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
1103	mutex_exit(&spa_l2cache_lock);
1104
1105	return (found);
1106}
1107
1108void
1109spa_l2cache_activate(vdev_t *vd)
1110{
1111	mutex_enter(&spa_l2cache_lock);
1112	ASSERT(vd->vdev_isl2cache);
1113	spa_aux_activate(vd, &spa_l2cache_avl);
1114	mutex_exit(&spa_l2cache_lock);
1115}
1116
1117/*
1118 * ==========================================================================
1119 * SPA vdev locking
1120 * ==========================================================================
1121 */
1122
1123/*
1124 * Lock the given spa_t for the purpose of adding or removing a vdev.
1125 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1126 * It returns the next transaction group for the spa_t.
1127 */
1128uint64_t
1129spa_vdev_enter(spa_t *spa)
1130{
1131	mutex_enter(&spa->spa_vdev_top_lock);
1132	mutex_enter(&spa_namespace_lock);
1133	return (spa_vdev_config_enter(spa));
1134}
1135
1136/*
1137 * Internal implementation for spa_vdev_enter().  Used when a vdev
1138 * operation requires multiple syncs (i.e. removing a device) while
1139 * keeping the spa_namespace_lock held.
1140 */
1141uint64_t
1142spa_vdev_config_enter(spa_t *spa)
1143{
1144	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1145
1146	spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1147
1148	return (spa_last_synced_txg(spa) + 1);
1149}
1150
1151/*
1152 * Used in combination with spa_vdev_config_enter() to allow the syncing
1153 * of multiple transactions without releasing the spa_namespace_lock.
1154 */
1155void
1156spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
1157{
1158	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1159
1160	int config_changed = B_FALSE;
1161
1162	ASSERT(txg > spa_last_synced_txg(spa));
1163
1164	spa->spa_pending_vdev = NULL;
1165
1166	/*
1167	 * Reassess the DTLs.
1168	 */
1169	vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
1170
1171	if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
1172		config_changed = B_TRUE;
1173		spa->spa_config_generation++;
1174	}
1175
1176	/*
1177	 * Verify the metaslab classes.
1178	 */
1179	ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
1180	ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
1181
1182	spa_config_exit(spa, SCL_ALL, spa);
1183
1184	/*
1185	 * Panic the system if the specified tag requires it.  This
1186	 * is useful for ensuring that configurations are updated
1187	 * transactionally.
1188	 */
1189	if (zio_injection_enabled)
1190		zio_handle_panic_injection(spa, tag, 0);
1191
1192	/*
1193	 * Note: this txg_wait_synced() is important because it ensures
1194	 * that there won't be more than one config change per txg.
1195	 * This allows us to use the txg as the generation number.
1196	 */
1197	if (error == 0)
1198		txg_wait_synced(spa->spa_dsl_pool, txg);
1199
1200	if (vd != NULL) {
1201		ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
1202		if (vd->vdev_ops->vdev_op_leaf) {
1203			mutex_enter(&vd->vdev_initialize_lock);
1204			vdev_initialize_stop(vd, VDEV_INITIALIZE_CANCELED);
1205			mutex_exit(&vd->vdev_initialize_lock);
1206		}
1207
1208		spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1209		vdev_free(vd);
1210		spa_config_exit(spa, SCL_ALL, spa);
1211	}
1212
1213	/*
1214	 * If the config changed, update the config cache.
1215	 */
1216	if (config_changed)
1217		spa_write_cachefile(spa, B_FALSE, B_TRUE);
1218}
1219
1220/*
1221 * Unlock the spa_t after adding or removing a vdev.  Besides undoing the
1222 * locking of spa_vdev_enter(), we also want make sure the transactions have
1223 * synced to disk, and then update the global configuration cache with the new
1224 * information.
1225 */
1226int
1227spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1228{
1229	spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1230	mutex_exit(&spa_namespace_lock);
1231	mutex_exit(&spa->spa_vdev_top_lock);
1232
1233	return (error);
1234}
1235
1236/*
1237 * Lock the given spa_t for the purpose of changing vdev state.
1238 */
1239void
1240spa_vdev_state_enter(spa_t *spa, int oplocks)
1241{
1242	int locks = SCL_STATE_ALL | oplocks;
1243
1244	/*
1245	 * Root pools may need to read of the underlying devfs filesystem
1246	 * when opening up a vdev.  Unfortunately if we're holding the
1247	 * SCL_ZIO lock it will result in a deadlock when we try to issue
1248	 * the read from the root filesystem.  Instead we "prefetch"
1249	 * the associated vnodes that we need prior to opening the
1250	 * underlying devices and cache them so that we can prevent
1251	 * any I/O when we are doing the actual open.
1252	 */
1253	if (spa_is_root(spa)) {
1254		int low = locks & ~(SCL_ZIO - 1);
1255		int high = locks & ~low;
1256
1257		spa_config_enter(spa, high, spa, RW_WRITER);
1258		vdev_hold(spa->spa_root_vdev);
1259		spa_config_enter(spa, low, spa, RW_WRITER);
1260	} else {
1261		spa_config_enter(spa, locks, spa, RW_WRITER);
1262	}
1263	spa->spa_vdev_locks = locks;
1264}
1265
1266int
1267spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1268{
1269	boolean_t config_changed = B_FALSE;
1270
1271	if (vd != NULL || error == 0)
1272		vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1273		    0, 0, B_FALSE);
1274
1275	if (vd != NULL) {
1276		vdev_state_dirty(vd->vdev_top);
1277		config_changed = B_TRUE;
1278		spa->spa_config_generation++;
1279	}
1280
1281	if (spa_is_root(spa))
1282		vdev_rele(spa->spa_root_vdev);
1283
1284	ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1285	spa_config_exit(spa, spa->spa_vdev_locks, spa);
1286
1287	/*
1288	 * If anything changed, wait for it to sync.  This ensures that,
1289	 * from the system administrator's perspective, zpool(1M) commands
1290	 * are synchronous.  This is important for things like zpool offline:
1291	 * when the command completes, you expect no further I/O from ZFS.
1292	 */
1293	if (vd != NULL)
1294		txg_wait_synced(spa->spa_dsl_pool, 0);
1295
1296	/*
1297	 * If the config changed, update the config cache.
1298	 */
1299	if (config_changed) {
1300		mutex_enter(&spa_namespace_lock);
1301		spa_write_cachefile(spa, B_FALSE, B_TRUE);
1302		mutex_exit(&spa_namespace_lock);
1303	}
1304
1305	return (error);
1306}
1307
1308/*
1309 * ==========================================================================
1310 * Miscellaneous functions
1311 * ==========================================================================
1312 */
1313
1314void
1315spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx)
1316{
1317	if (!nvlist_exists(spa->spa_label_features, feature)) {
1318		fnvlist_add_boolean(spa->spa_label_features, feature);
1319		/*
1320		 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1321		 * dirty the vdev config because lock SCL_CONFIG is not held.
1322		 * Thankfully, in this case we don't need to dirty the config
1323		 * because it will be written out anyway when we finish
1324		 * creating the pool.
1325		 */
1326		if (tx->tx_txg != TXG_INITIAL)
1327			vdev_config_dirty(spa->spa_root_vdev);
1328	}
1329}
1330
1331void
1332spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1333{
1334	if (nvlist_remove_all(spa->spa_label_features, feature) == 0)
1335		vdev_config_dirty(spa->spa_root_vdev);
1336}
1337
1338/*
1339 * Return the spa_t associated with given pool_guid, if it exists.  If
1340 * device_guid is non-zero, determine whether the pool exists *and* contains
1341 * a device with the specified device_guid.
1342 */
1343spa_t *
1344spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1345{
1346	spa_t *spa;
1347	avl_tree_t *t = &spa_namespace_avl;
1348
1349	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1350
1351	for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1352		if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1353			continue;
1354		if (spa->spa_root_vdev == NULL)
1355			continue;
1356		if (spa_guid(spa) == pool_guid) {
1357			if (device_guid == 0)
1358				break;
1359
1360			if (vdev_lookup_by_guid(spa->spa_root_vdev,
1361			    device_guid) != NULL)
1362				break;
1363
1364			/*
1365			 * Check any devices we may be in the process of adding.
1366			 */
1367			if (spa->spa_pending_vdev) {
1368				if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1369				    device_guid) != NULL)
1370					break;
1371			}
1372		}
1373	}
1374
1375	return (spa);
1376}
1377
1378/*
1379 * Determine whether a pool with the given pool_guid exists.
1380 */
1381boolean_t
1382spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1383{
1384	return (spa_by_guid(pool_guid, device_guid) != NULL);
1385}
1386
1387char *
1388spa_strdup(const char *s)
1389{
1390	size_t len;
1391	char *new;
1392
1393	len = strlen(s);
1394	new = kmem_alloc(len + 1, KM_SLEEP);
1395	bcopy(s, new, len);
1396	new[len] = '\0';
1397
1398	return (new);
1399}
1400
1401void
1402spa_strfree(char *s)
1403{
1404	kmem_free(s, strlen(s) + 1);
1405}
1406
1407uint64_t
1408spa_get_random(uint64_t range)
1409{
1410	uint64_t r;
1411
1412	ASSERT(range != 0);
1413
1414	if (range == 1)
1415		return (0);
1416
1417	(void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1418
1419	return (r % range);
1420}
1421
1422uint64_t
1423spa_generate_guid(spa_t *spa)
1424{
1425	uint64_t guid = spa_get_random(-1ULL);
1426
1427	if (spa != NULL) {
1428		while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1429			guid = spa_get_random(-1ULL);
1430	} else {
1431		while (guid == 0 || spa_guid_exists(guid, 0))
1432			guid = spa_get_random(-1ULL);
1433	}
1434
1435	return (guid);
1436}
1437
1438void
1439snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp)
1440{
1441	char type[256];
1442	char *checksum = NULL;
1443	char *compress = NULL;
1444
1445	if (bp != NULL) {
1446		if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1447			dmu_object_byteswap_t bswap =
1448			    DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1449			(void) snprintf(type, sizeof (type), "bswap %s %s",
1450			    DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1451			    "metadata" : "data",
1452			    dmu_ot_byteswap[bswap].ob_name);
1453		} else {
1454			(void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1455			    sizeof (type));
1456		}
1457		if (!BP_IS_EMBEDDED(bp)) {
1458			checksum =
1459			    zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1460		}
1461		compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1462	}
1463
1464	SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum,
1465	    compress);
1466}
1467
1468void
1469spa_freeze(spa_t *spa)
1470{
1471	uint64_t freeze_txg = 0;
1472
1473	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1474	if (spa->spa_freeze_txg == UINT64_MAX) {
1475		freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1476		spa->spa_freeze_txg = freeze_txg;
1477	}
1478	spa_config_exit(spa, SCL_ALL, FTAG);
1479	if (freeze_txg != 0)
1480		txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1481}
1482
1483void
1484zfs_panic_recover(const char *fmt, ...)
1485{
1486	va_list adx;
1487
1488	va_start(adx, fmt);
1489	vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1490	va_end(adx);
1491}
1492
1493/*
1494 * This is a stripped-down version of strtoull, suitable only for converting
1495 * lowercase hexadecimal numbers that don't overflow.
1496 */
1497uint64_t
1498zfs_strtonum(const char *str, char **nptr)
1499{
1500	uint64_t val = 0;
1501	char c;
1502	int digit;
1503
1504	while ((c = *str) != '\0') {
1505		if (c >= '0' && c <= '9')
1506			digit = c - '0';
1507		else if (c >= 'a' && c <= 'f')
1508			digit = 10 + c - 'a';
1509		else
1510			break;
1511
1512		val *= 16;
1513		val += digit;
1514
1515		str++;
1516	}
1517
1518	if (nptr)
1519		*nptr = (char *)str;
1520
1521	return (val);
1522}
1523
1524/*
1525 * ==========================================================================
1526 * Accessor functions
1527 * ==========================================================================
1528 */
1529
1530boolean_t
1531spa_shutting_down(spa_t *spa)
1532{
1533	return (spa->spa_async_suspended);
1534}
1535
1536dsl_pool_t *
1537spa_get_dsl(spa_t *spa)
1538{
1539	return (spa->spa_dsl_pool);
1540}
1541
1542boolean_t
1543spa_is_initializing(spa_t *spa)
1544{
1545	return (spa->spa_is_initializing);
1546}
1547
1548boolean_t
1549spa_indirect_vdevs_loaded(spa_t *spa)
1550{
1551	return (spa->spa_indirect_vdevs_loaded);
1552}
1553
1554blkptr_t *
1555spa_get_rootblkptr(spa_t *spa)
1556{
1557	return (&spa->spa_ubsync.ub_rootbp);
1558}
1559
1560void
1561spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1562{
1563	spa->spa_uberblock.ub_rootbp = *bp;
1564}
1565
1566void
1567spa_altroot(spa_t *spa, char *buf, size_t buflen)
1568{
1569	if (spa->spa_root == NULL)
1570		buf[0] = '\0';
1571	else
1572		(void) strncpy(buf, spa->spa_root, buflen);
1573}
1574
1575int
1576spa_sync_pass(spa_t *spa)
1577{
1578	return (spa->spa_sync_pass);
1579}
1580
1581char *
1582spa_name(spa_t *spa)
1583{
1584	return (spa->spa_name);
1585}
1586
1587uint64_t
1588spa_guid(spa_t *spa)
1589{
1590	dsl_pool_t *dp = spa_get_dsl(spa);
1591	uint64_t guid;
1592
1593	/*
1594	 * If we fail to parse the config during spa_load(), we can go through
1595	 * the error path (which posts an ereport) and end up here with no root
1596	 * vdev.  We stash the original pool guid in 'spa_config_guid' to handle
1597	 * this case.
1598	 */
1599	if (spa->spa_root_vdev == NULL)
1600		return (spa->spa_config_guid);
1601
1602	guid = spa->spa_last_synced_guid != 0 ?
1603	    spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1604
1605	/*
1606	 * Return the most recently synced out guid unless we're
1607	 * in syncing context.
1608	 */
1609	if (dp && dsl_pool_sync_context(dp))
1610		return (spa->spa_root_vdev->vdev_guid);
1611	else
1612		return (guid);
1613}
1614
1615uint64_t
1616spa_load_guid(spa_t *spa)
1617{
1618	/*
1619	 * This is a GUID that exists solely as a reference for the
1620	 * purposes of the arc.  It is generated at load time, and
1621	 * is never written to persistent storage.
1622	 */
1623	return (spa->spa_load_guid);
1624}
1625
1626uint64_t
1627spa_last_synced_txg(spa_t *spa)
1628{
1629	return (spa->spa_ubsync.ub_txg);
1630}
1631
1632uint64_t
1633spa_first_txg(spa_t *spa)
1634{
1635	return (spa->spa_first_txg);
1636}
1637
1638uint64_t
1639spa_syncing_txg(spa_t *spa)
1640{
1641	return (spa->spa_syncing_txg);
1642}
1643
1644/*
1645 * Return the last txg where data can be dirtied. The final txgs
1646 * will be used to just clear out any deferred frees that remain.
1647 */
1648uint64_t
1649spa_final_dirty_txg(spa_t *spa)
1650{
1651	return (spa->spa_final_txg - TXG_DEFER_SIZE);
1652}
1653
1654pool_state_t
1655spa_state(spa_t *spa)
1656{
1657	return (spa->spa_state);
1658}
1659
1660spa_load_state_t
1661spa_load_state(spa_t *spa)
1662{
1663	return (spa->spa_load_state);
1664}
1665
1666uint64_t
1667spa_freeze_txg(spa_t *spa)
1668{
1669	return (spa->spa_freeze_txg);
1670}
1671
1672/* ARGSUSED */
1673uint64_t
1674spa_get_worst_case_asize(spa_t *spa, uint64_t lsize)
1675{
1676	return (lsize * spa_asize_inflation);
1677}
1678
1679/*
1680 * Return the amount of slop space in bytes.  It is 1/32 of the pool (3.2%),
1681 * or at least 128MB, unless that would cause it to be more than half the
1682 * pool size.
1683 *
1684 * See the comment above spa_slop_shift for details.
1685 */
1686uint64_t
1687spa_get_slop_space(spa_t *spa)
1688{
1689	uint64_t space = spa_get_dspace(spa);
1690	return (MAX(space >> spa_slop_shift, MIN(space >> 1, spa_min_slop)));
1691}
1692
1693uint64_t
1694spa_get_dspace(spa_t *spa)
1695{
1696	return (spa->spa_dspace);
1697}
1698
1699uint64_t
1700spa_get_checkpoint_space(spa_t *spa)
1701{
1702	return (spa->spa_checkpoint_info.sci_dspace);
1703}
1704
1705void
1706spa_update_dspace(spa_t *spa)
1707{
1708	spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1709	    ddt_get_dedup_dspace(spa);
1710	if (spa->spa_vdev_removal != NULL) {
1711		/*
1712		 * We can't allocate from the removing device, so
1713		 * subtract its size.  This prevents the DMU/DSL from
1714		 * filling up the (now smaller) pool while we are in the
1715		 * middle of removing the device.
1716		 *
1717		 * Note that the DMU/DSL doesn't actually know or care
1718		 * how much space is allocated (it does its own tracking
1719		 * of how much space has been logically used).  So it
1720		 * doesn't matter that the data we are moving may be
1721		 * allocated twice (on the old device and the new
1722		 * device).
1723		 */
1724		spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1725		vdev_t *vd =
1726		    vdev_lookup_top(spa, spa->spa_vdev_removal->svr_vdev_id);
1727		spa->spa_dspace -= spa_deflate(spa) ?
1728		    vd->vdev_stat.vs_dspace : vd->vdev_stat.vs_space;
1729		spa_config_exit(spa, SCL_VDEV, FTAG);
1730	}
1731}
1732
1733/*
1734 * Return the failure mode that has been set to this pool. The default
1735 * behavior will be to block all I/Os when a complete failure occurs.
1736 */
1737uint8_t
1738spa_get_failmode(spa_t *spa)
1739{
1740	return (spa->spa_failmode);
1741}
1742
1743boolean_t
1744spa_suspended(spa_t *spa)
1745{
1746	return (spa->spa_suspended != ZIO_SUSPEND_NONE);
1747}
1748
1749uint64_t
1750spa_version(spa_t *spa)
1751{
1752	return (spa->spa_ubsync.ub_version);
1753}
1754
1755boolean_t
1756spa_deflate(spa_t *spa)
1757{
1758	return (spa->spa_deflate);
1759}
1760
1761metaslab_class_t *
1762spa_normal_class(spa_t *spa)
1763{
1764	return (spa->spa_normal_class);
1765}
1766
1767metaslab_class_t *
1768spa_log_class(spa_t *spa)
1769{
1770	return (spa->spa_log_class);
1771}
1772
1773void
1774spa_evicting_os_register(spa_t *spa, objset_t *os)
1775{
1776	mutex_enter(&spa->spa_evicting_os_lock);
1777	list_insert_head(&spa->spa_evicting_os_list, os);
1778	mutex_exit(&spa->spa_evicting_os_lock);
1779}
1780
1781void
1782spa_evicting_os_deregister(spa_t *spa, objset_t *os)
1783{
1784	mutex_enter(&spa->spa_evicting_os_lock);
1785	list_remove(&spa->spa_evicting_os_list, os);
1786	cv_broadcast(&spa->spa_evicting_os_cv);
1787	mutex_exit(&spa->spa_evicting_os_lock);
1788}
1789
1790void
1791spa_evicting_os_wait(spa_t *spa)
1792{
1793	mutex_enter(&spa->spa_evicting_os_lock);
1794	while (!list_is_empty(&spa->spa_evicting_os_list))
1795		cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock);
1796	mutex_exit(&spa->spa_evicting_os_lock);
1797
1798	dmu_buf_user_evict_wait();
1799}
1800
1801int
1802spa_max_replication(spa_t *spa)
1803{
1804	/*
1805	 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1806	 * handle BPs with more than one DVA allocated.  Set our max
1807	 * replication level accordingly.
1808	 */
1809	if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1810		return (1);
1811	return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1812}
1813
1814int
1815spa_prev_software_version(spa_t *spa)
1816{
1817	return (spa->spa_prev_software_version);
1818}
1819
1820uint64_t
1821spa_deadman_synctime(spa_t *spa)
1822{
1823	return (spa->spa_deadman_synctime);
1824}
1825
1826uint64_t
1827dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1828{
1829	uint64_t asize = DVA_GET_ASIZE(dva);
1830	uint64_t dsize = asize;
1831
1832	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1833
1834	if (asize != 0 && spa->spa_deflate) {
1835		vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1836		dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1837	}
1838
1839	return (dsize);
1840}
1841
1842uint64_t
1843bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1844{
1845	uint64_t dsize = 0;
1846
1847	for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1848		dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1849
1850	return (dsize);
1851}
1852
1853uint64_t
1854bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1855{
1856	uint64_t dsize = 0;
1857
1858	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1859
1860	for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1861		dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1862
1863	spa_config_exit(spa, SCL_VDEV, FTAG);
1864
1865	return (dsize);
1866}
1867
1868uint64_t
1869spa_dirty_data(spa_t *spa)
1870{
1871	return (spa->spa_dsl_pool->dp_dirty_total);
1872}
1873
1874/*
1875 * ==========================================================================
1876 * Initialization and Termination
1877 * ==========================================================================
1878 */
1879
1880static int
1881spa_name_compare(const void *a1, const void *a2)
1882{
1883	const spa_t *s1 = a1;
1884	const spa_t *s2 = a2;
1885	int s;
1886
1887	s = strcmp(s1->spa_name, s2->spa_name);
1888	if (s > 0)
1889		return (1);
1890	if (s < 0)
1891		return (-1);
1892	return (0);
1893}
1894
1895int
1896spa_busy(void)
1897{
1898	return (spa_active_count);
1899}
1900
1901void
1902spa_boot_init()
1903{
1904	spa_config_load();
1905}
1906
1907void
1908spa_init(int mode)
1909{
1910	mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1911	mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1912	mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1913	cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1914
1915	avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1916	    offsetof(spa_t, spa_avl));
1917
1918	avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1919	    offsetof(spa_aux_t, aux_avl));
1920
1921	avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1922	    offsetof(spa_aux_t, aux_avl));
1923
1924	spa_mode_global = mode;
1925
1926#ifdef _KERNEL
1927	spa_arch_init();
1928#else
1929	if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
1930		arc_procfd = open("/proc/self/ctl", O_WRONLY);
1931		if (arc_procfd == -1) {
1932			perror("could not enable watchpoints: "
1933			    "opening /proc/self/ctl failed: ");
1934		} else {
1935			arc_watch = B_TRUE;
1936		}
1937	}
1938#endif
1939
1940	zfs_refcount_init();
1941	unique_init();
1942	range_tree_init();
1943	metaslab_alloc_trace_init();
1944	zio_init();
1945	dmu_init();
1946	zil_init();
1947	vdev_cache_stat_init();
1948	zfs_prop_init();
1949	zpool_prop_init();
1950	zpool_feature_init();
1951	spa_config_load();
1952	l2arc_start();
1953}
1954
1955void
1956spa_fini(void)
1957{
1958	l2arc_stop();
1959
1960	spa_evict_all();
1961
1962	vdev_cache_stat_fini();
1963	zil_fini();
1964	dmu_fini();
1965	zio_fini();
1966	metaslab_alloc_trace_fini();
1967	range_tree_fini();
1968	unique_fini();
1969	zfs_refcount_fini();
1970
1971	avl_destroy(&spa_namespace_avl);
1972	avl_destroy(&spa_spare_avl);
1973	avl_destroy(&spa_l2cache_avl);
1974
1975	cv_destroy(&spa_namespace_cv);
1976	mutex_destroy(&spa_namespace_lock);
1977	mutex_destroy(&spa_spare_lock);
1978	mutex_destroy(&spa_l2cache_lock);
1979}
1980
1981/*
1982 * Return whether this pool has slogs. No locking needed.
1983 * It's not a problem if the wrong answer is returned as it's only for
1984 * performance and not correctness
1985 */
1986boolean_t
1987spa_has_slogs(spa_t *spa)
1988{
1989	return (spa->spa_log_class->mc_rotor != NULL);
1990}
1991
1992spa_log_state_t
1993spa_get_log_state(spa_t *spa)
1994{
1995	return (spa->spa_log_state);
1996}
1997
1998void
1999spa_set_log_state(spa_t *spa, spa_log_state_t state)
2000{
2001	spa->spa_log_state = state;
2002}
2003
2004boolean_t
2005spa_is_root(spa_t *spa)
2006{
2007	return (spa->spa_is_root);
2008}
2009
2010boolean_t
2011spa_writeable(spa_t *spa)
2012{
2013	return (!!(spa->spa_mode & FWRITE) && spa->spa_trust_config);
2014}
2015
2016/*
2017 * Returns true if there is a pending sync task in any of the current
2018 * syncing txg, the current quiescing txg, or the current open txg.
2019 */
2020boolean_t
2021spa_has_pending_synctask(spa_t *spa)
2022{
2023	return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks) ||
2024	    !txg_all_lists_empty(&spa->spa_dsl_pool->dp_early_sync_tasks));
2025}
2026
2027int
2028spa_mode(spa_t *spa)
2029{
2030	return (spa->spa_mode);
2031}
2032
2033uint64_t
2034spa_bootfs(spa_t *spa)
2035{
2036	return (spa->spa_bootfs);
2037}
2038
2039uint64_t
2040spa_delegation(spa_t *spa)
2041{
2042	return (spa->spa_delegation);
2043}
2044
2045objset_t *
2046spa_meta_objset(spa_t *spa)
2047{
2048	return (spa->spa_meta_objset);
2049}
2050
2051enum zio_checksum
2052spa_dedup_checksum(spa_t *spa)
2053{
2054	return (spa->spa_dedup_checksum);
2055}
2056
2057/*
2058 * Reset pool scan stat per scan pass (or reboot).
2059 */
2060void
2061spa_scan_stat_init(spa_t *spa)
2062{
2063	/* data not stored on disk */
2064	spa->spa_scan_pass_start = gethrestime_sec();
2065	if (dsl_scan_is_paused_scrub(spa->spa_dsl_pool->dp_scan))
2066		spa->spa_scan_pass_scrub_pause = spa->spa_scan_pass_start;
2067	else
2068		spa->spa_scan_pass_scrub_pause = 0;
2069	spa->spa_scan_pass_scrub_spent_paused = 0;
2070	spa->spa_scan_pass_exam = 0;
2071	vdev_scan_stat_init(spa->spa_root_vdev);
2072}
2073
2074/*
2075 * Get scan stats for zpool status reports
2076 */
2077int
2078spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
2079{
2080	dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
2081
2082	if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
2083		return (SET_ERROR(ENOENT));
2084	bzero(ps, sizeof (pool_scan_stat_t));
2085
2086	/* data stored on disk */
2087	ps->pss_func = scn->scn_phys.scn_func;
2088	ps->pss_start_time = scn->scn_phys.scn_start_time;
2089	ps->pss_end_time = scn->scn_phys.scn_end_time;
2090	ps->pss_to_examine = scn->scn_phys.scn_to_examine;
2091	ps->pss_examined = scn->scn_phys.scn_examined;
2092	ps->pss_to_process = scn->scn_phys.scn_to_process;
2093	ps->pss_processed = scn->scn_phys.scn_processed;
2094	ps->pss_errors = scn->scn_phys.scn_errors;
2095	ps->pss_state = scn->scn_phys.scn_state;
2096
2097	/* data not stored on disk */
2098	ps->pss_pass_start = spa->spa_scan_pass_start;
2099	ps->pss_pass_exam = spa->spa_scan_pass_exam;
2100	ps->pss_pass_scrub_pause = spa->spa_scan_pass_scrub_pause;
2101	ps->pss_pass_scrub_spent_paused = spa->spa_scan_pass_scrub_spent_paused;
2102
2103	return (0);
2104}
2105
2106int
2107spa_maxblocksize(spa_t *spa)
2108{
2109	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS))
2110		return (SPA_MAXBLOCKSIZE);
2111	else
2112		return (SPA_OLD_MAXBLOCKSIZE);
2113}
2114
2115int
2116spa_maxdnodesize(spa_t *spa)
2117{
2118	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE))
2119		return (DNODE_MAX_SIZE);
2120	else
2121		return (DNODE_MIN_SIZE);
2122}
2123
2124boolean_t
2125spa_multihost(spa_t *spa)
2126{
2127	return (spa->spa_multihost ? B_TRUE : B_FALSE);
2128}
2129
2130unsigned long
2131spa_get_hostid(void)
2132{
2133	unsigned long myhostid;
2134
2135#ifdef	_KERNEL
2136	myhostid = zone_get_hostid(NULL);
2137#else	/* _KERNEL */
2138	/*
2139	 * We're emulating the system's hostid in userland, so
2140	 * we can't use zone_get_hostid().
2141	 */
2142	(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2143#endif	/* _KERNEL */
2144
2145	return (myhostid);
2146}
2147
2148/*
2149 * Returns the txg that the last device removal completed. No indirect mappings
2150 * have been added since this txg.
2151 */
2152uint64_t
2153spa_get_last_removal_txg(spa_t *spa)
2154{
2155	uint64_t vdevid;
2156	uint64_t ret = -1ULL;
2157
2158	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2159	/*
2160	 * sr_prev_indirect_vdev is only modified while holding all the
2161	 * config locks, so it is sufficient to hold SCL_VDEV as reader when
2162	 * examining it.
2163	 */
2164	vdevid = spa->spa_removing_phys.sr_prev_indirect_vdev;
2165
2166	while (vdevid != -1ULL) {
2167		vdev_t *vd = vdev_lookup_top(spa, vdevid);
2168		vdev_indirect_births_t *vib = vd->vdev_indirect_births;
2169
2170		ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops);
2171
2172		/*
2173		 * If the removal did not remap any data, we don't care.
2174		 */
2175		if (vdev_indirect_births_count(vib) != 0) {
2176			ret = vdev_indirect_births_last_entry_txg(vib);
2177			break;
2178		}
2179
2180		vdevid = vd->vdev_indirect_config.vic_prev_indirect_vdev;
2181	}
2182	spa_config_exit(spa, SCL_VDEV, FTAG);
2183
2184	IMPLY(ret != -1ULL,
2185	    spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL));
2186
2187	return (ret);
2188}
2189
2190boolean_t
2191spa_trust_config(spa_t *spa)
2192{
2193	return (spa->spa_trust_config);
2194}
2195
2196uint64_t
2197spa_missing_tvds_allowed(spa_t *spa)
2198{
2199	return (spa->spa_missing_tvds_allowed);
2200}
2201
2202void
2203spa_set_missing_tvds(spa_t *spa, uint64_t missing)
2204{
2205	spa->spa_missing_tvds = missing;
2206}
2207
2208boolean_t
2209spa_top_vdevs_spacemap_addressable(spa_t *spa)
2210{
2211	vdev_t *rvd = spa->spa_root_vdev;
2212	for (uint64_t c = 0; c < rvd->vdev_children; c++) {
2213		if (!vdev_is_spacemap_addressable(rvd->vdev_child[c]))
2214			return (B_FALSE);
2215	}
2216	return (B_TRUE);
2217}
2218
2219boolean_t
2220spa_has_checkpoint(spa_t *spa)
2221{
2222	return (spa->spa_checkpoint_txg != 0);
2223}
2224
2225boolean_t
2226spa_importing_readonly_checkpoint(spa_t *spa)
2227{
2228	return ((spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT) &&
2229	    spa->spa_mode == FREAD);
2230}
2231
2232uint64_t
2233spa_min_claim_txg(spa_t *spa)
2234{
2235	uint64_t checkpoint_txg = spa->spa_uberblock.ub_checkpoint_txg;
2236
2237	if (checkpoint_txg != 0)
2238		return (checkpoint_txg + 1);
2239
2240	return (spa->spa_first_txg);
2241}
2242
2243/*
2244 * If there is a checkpoint, async destroys may consume more space from
2245 * the pool instead of freeing it. In an attempt to save the pool from
2246 * getting suspended when it is about to run out of space, we stop
2247 * processing async destroys.
2248 */
2249boolean_t
2250spa_suspend_async_destroy(spa_t *spa)
2251{
2252	dsl_pool_t *dp = spa_get_dsl(spa);
2253
2254	uint64_t unreserved = dsl_pool_unreserved_space(dp,
2255	    ZFS_SPACE_CHECK_EXTRA_RESERVED);
2256	uint64_t used = dsl_dir_phys(dp->dp_root_dir)->dd_used_bytes;
2257	uint64_t avail = (unreserved > used) ? (unreserved - used) : 0;
2258
2259	if (spa_has_checkpoint(spa) && avail == 0)
2260		return (B_TRUE);
2261
2262	return (B_FALSE);
2263}
2264