spa_misc.c revision a15215608b8bd90f714f6db21ee623b584607cb6
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 2009 Sun Microsystems, Inc.  All rights reserved.
23 * Use is subject to license terms.
24 */
25
26#include <sys/zfs_context.h>
27#include <sys/spa_impl.h>
28#include <sys/zio.h>
29#include <sys/zio_checksum.h>
30#include <sys/zio_compress.h>
31#include <sys/dmu.h>
32#include <sys/dmu_tx.h>
33#include <sys/zap.h>
34#include <sys/zil.h>
35#include <sys/vdev_impl.h>
36#include <sys/metaslab.h>
37#include <sys/uberblock_impl.h>
38#include <sys/txg.h>
39#include <sys/avl.h>
40#include <sys/unique.h>
41#include <sys/dsl_pool.h>
42#include <sys/dsl_dir.h>
43#include <sys/dsl_prop.h>
44#include <sys/fs/zfs.h>
45#include <sys/metaslab_impl.h>
46#include <sys/sunddi.h>
47#include <sys/arc.h>
48#include <sys/ddt.h>
49#include "zfs_prop.h"
50
51/*
52 * SPA locking
53 *
54 * There are four basic locks for managing spa_t structures:
55 *
56 * spa_namespace_lock (global mutex)
57 *
58 *	This lock must be acquired to do any of the following:
59 *
60 *		- Lookup a spa_t by name
61 *		- Add or remove a spa_t from the namespace
62 *		- Increase spa_refcount from non-zero
63 *		- Check if spa_refcount is zero
64 *		- Rename a spa_t
65 *		- add/remove/attach/detach devices
66 *		- Held for the duration of create/destroy/import/export
67 *
68 *	It does not need to handle recursion.  A create or destroy may
69 *	reference objects (files or zvols) in other pools, but by
70 *	definition they must have an existing reference, and will never need
71 *	to lookup a spa_t by name.
72 *
73 * spa_refcount (per-spa refcount_t protected by mutex)
74 *
75 *	This reference count keep track of any active users of the spa_t.  The
76 *	spa_t cannot be destroyed or freed while this is non-zero.  Internally,
77 *	the refcount is never really 'zero' - opening a pool implicitly keeps
78 *	some references in the DMU.  Internally we check against spa_minref, but
79 *	present the image of a zero/non-zero value to consumers.
80 *
81 * spa_config_lock[] (per-spa array of rwlocks)
82 *
83 *	This protects the spa_t from config changes, and must be held in
84 *	the following circumstances:
85 *
86 *		- RW_READER to perform I/O to the spa
87 *		- RW_WRITER to change the vdev config
88 *
89 * The locking order is fairly straightforward:
90 *
91 *		spa_namespace_lock	->	spa_refcount
92 *
93 *	The namespace lock must be acquired to increase the refcount from 0
94 *	or to check if it is zero.
95 *
96 *		spa_refcount		->	spa_config_lock[]
97 *
98 *	There must be at least one valid reference on the spa_t to acquire
99 *	the config lock.
100 *
101 *		spa_namespace_lock	->	spa_config_lock[]
102 *
103 *	The namespace lock must always be taken before the config lock.
104 *
105 *
106 * The spa_namespace_lock can be acquired directly and is globally visible.
107 *
108 * The namespace is manipulated using the following functions, all of which
109 * require the spa_namespace_lock to be held.
110 *
111 *	spa_lookup()		Lookup a spa_t by name.
112 *
113 *	spa_add()		Create a new spa_t in the namespace.
114 *
115 *	spa_remove()		Remove a spa_t from the namespace.  This also
116 *				frees up any memory associated with the spa_t.
117 *
118 *	spa_next()		Returns the next spa_t in the system, or the
119 *				first if NULL is passed.
120 *
121 *	spa_evict_all()		Shutdown and remove all spa_t structures in
122 *				the system.
123 *
124 *	spa_guid_exists()	Determine whether a pool/device guid exists.
125 *
126 * The spa_refcount is manipulated using the following functions:
127 *
128 *	spa_open_ref()		Adds a reference to the given spa_t.  Must be
129 *				called with spa_namespace_lock held if the
130 *				refcount is currently zero.
131 *
132 *	spa_close()		Remove a reference from the spa_t.  This will
133 *				not free the spa_t or remove it from the
134 *				namespace.  No locking is required.
135 *
136 *	spa_refcount_zero()	Returns true if the refcount is currently
137 *				zero.  Must be called with spa_namespace_lock
138 *				held.
139 *
140 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
141 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
142 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
143 *
144 * To read the configuration, it suffices to hold one of these locks as reader.
145 * To modify the configuration, you must hold all locks as writer.  To modify
146 * vdev state without altering the vdev tree's topology (e.g. online/offline),
147 * you must hold SCL_STATE and SCL_ZIO as writer.
148 *
149 * We use these distinct config locks to avoid recursive lock entry.
150 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
151 * block allocations (SCL_ALLOC), which may require reading space maps
152 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
153 *
154 * The spa config locks cannot be normal rwlocks because we need the
155 * ability to hand off ownership.  For example, SCL_ZIO is acquired
156 * by the issuing thread and later released by an interrupt thread.
157 * They do, however, obey the usual write-wanted semantics to prevent
158 * writer (i.e. system administrator) starvation.
159 *
160 * The lock acquisition rules are as follows:
161 *
162 * SCL_CONFIG
163 *	Protects changes to the vdev tree topology, such as vdev
164 *	add/remove/attach/detach.  Protects the dirty config list
165 *	(spa_config_dirty_list) and the set of spares and l2arc devices.
166 *
167 * SCL_STATE
168 *	Protects changes to pool state and vdev state, such as vdev
169 *	online/offline/fault/degrade/clear.  Protects the dirty state list
170 *	(spa_state_dirty_list) and global pool state (spa_state).
171 *
172 * SCL_ALLOC
173 *	Protects changes to metaslab groups and classes.
174 *	Held as reader by metaslab_alloc() and metaslab_claim().
175 *
176 * SCL_ZIO
177 *	Held by bp-level zios (those which have no io_vd upon entry)
178 *	to prevent changes to the vdev tree.  The bp-level zio implicitly
179 *	protects all of its vdev child zios, which do not hold SCL_ZIO.
180 *
181 * SCL_FREE
182 *	Protects changes to metaslab groups and classes.
183 *	Held as reader by metaslab_free().  SCL_FREE is distinct from
184 *	SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
185 *	blocks in zio_done() while another i/o that holds either
186 *	SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
187 *
188 * SCL_VDEV
189 *	Held as reader to prevent changes to the vdev tree during trivial
190 *	inquiries such as bp_get_dsize().  SCL_VDEV is distinct from the
191 *	other locks, and lower than all of them, to ensure that it's safe
192 *	to acquire regardless of caller context.
193 *
194 * In addition, the following rules apply:
195 *
196 * (a)	spa_props_lock protects pool properties, spa_config and spa_config_list.
197 *	The lock ordering is SCL_CONFIG > spa_props_lock.
198 *
199 * (b)	I/O operations on leaf vdevs.  For any zio operation that takes
200 *	an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
201 *	or zio_write_phys() -- the caller must ensure that the config cannot
202 *	cannot change in the interim, and that the vdev cannot be reopened.
203 *	SCL_STATE as reader suffices for both.
204 *
205 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
206 *
207 *	spa_vdev_enter()	Acquire the namespace lock and the config lock
208 *				for writing.
209 *
210 *	spa_vdev_exit()		Release the config lock, wait for all I/O
211 *				to complete, sync the updated configs to the
212 *				cache, and release the namespace lock.
213 *
214 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
215 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
216 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
217 *
218 * spa_rename() is also implemented within this file since is requires
219 * manipulation of the namespace.
220 */
221
222static avl_tree_t spa_namespace_avl;
223kmutex_t spa_namespace_lock;
224static kcondvar_t spa_namespace_cv;
225static int spa_active_count;
226int spa_max_replication_override = SPA_DVAS_PER_BP;
227
228static kmutex_t spa_spare_lock;
229static avl_tree_t spa_spare_avl;
230static kmutex_t spa_l2cache_lock;
231static avl_tree_t spa_l2cache_avl;
232
233kmem_cache_t *spa_buffer_pool;
234int spa_mode_global;
235
236#ifdef ZFS_DEBUG
237/* Everything except dprintf is on by default in debug builds */
238int zfs_flags = ~ZFS_DEBUG_DPRINTF;
239#else
240int zfs_flags = 0;
241#endif
242
243/*
244 * zfs_recover can be set to nonzero to attempt to recover from
245 * otherwise-fatal errors, typically caused by on-disk corruption.  When
246 * set, calls to zfs_panic_recover() will turn into warning messages.
247 */
248int zfs_recover = 0;
249
250
251/*
252 * ==========================================================================
253 * SPA config locking
254 * ==========================================================================
255 */
256static void
257spa_config_lock_init(spa_t *spa)
258{
259	for (int i = 0; i < SCL_LOCKS; i++) {
260		spa_config_lock_t *scl = &spa->spa_config_lock[i];
261		mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
262		cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
263		refcount_create(&scl->scl_count);
264		scl->scl_writer = NULL;
265		scl->scl_write_wanted = 0;
266	}
267}
268
269static void
270spa_config_lock_destroy(spa_t *spa)
271{
272	for (int i = 0; i < SCL_LOCKS; i++) {
273		spa_config_lock_t *scl = &spa->spa_config_lock[i];
274		mutex_destroy(&scl->scl_lock);
275		cv_destroy(&scl->scl_cv);
276		refcount_destroy(&scl->scl_count);
277		ASSERT(scl->scl_writer == NULL);
278		ASSERT(scl->scl_write_wanted == 0);
279	}
280}
281
282int
283spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
284{
285	for (int i = 0; i < SCL_LOCKS; i++) {
286		spa_config_lock_t *scl = &spa->spa_config_lock[i];
287		if (!(locks & (1 << i)))
288			continue;
289		mutex_enter(&scl->scl_lock);
290		if (rw == RW_READER) {
291			if (scl->scl_writer || scl->scl_write_wanted) {
292				mutex_exit(&scl->scl_lock);
293				spa_config_exit(spa, locks ^ (1 << i), tag);
294				return (0);
295			}
296		} else {
297			ASSERT(scl->scl_writer != curthread);
298			if (!refcount_is_zero(&scl->scl_count)) {
299				mutex_exit(&scl->scl_lock);
300				spa_config_exit(spa, locks ^ (1 << i), tag);
301				return (0);
302			}
303			scl->scl_writer = curthread;
304		}
305		(void) refcount_add(&scl->scl_count, tag);
306		mutex_exit(&scl->scl_lock);
307	}
308	return (1);
309}
310
311void
312spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
313{
314	int wlocks_held = 0;
315
316	for (int i = 0; i < SCL_LOCKS; i++) {
317		spa_config_lock_t *scl = &spa->spa_config_lock[i];
318		if (scl->scl_writer == curthread)
319			wlocks_held |= (1 << i);
320		if (!(locks & (1 << i)))
321			continue;
322		mutex_enter(&scl->scl_lock);
323		if (rw == RW_READER) {
324			while (scl->scl_writer || scl->scl_write_wanted) {
325				cv_wait(&scl->scl_cv, &scl->scl_lock);
326			}
327		} else {
328			ASSERT(scl->scl_writer != curthread);
329			while (!refcount_is_zero(&scl->scl_count)) {
330				scl->scl_write_wanted++;
331				cv_wait(&scl->scl_cv, &scl->scl_lock);
332				scl->scl_write_wanted--;
333			}
334			scl->scl_writer = curthread;
335		}
336		(void) refcount_add(&scl->scl_count, tag);
337		mutex_exit(&scl->scl_lock);
338	}
339	ASSERT(wlocks_held <= locks);
340}
341
342void
343spa_config_exit(spa_t *spa, int locks, void *tag)
344{
345	for (int i = SCL_LOCKS - 1; i >= 0; i--) {
346		spa_config_lock_t *scl = &spa->spa_config_lock[i];
347		if (!(locks & (1 << i)))
348			continue;
349		mutex_enter(&scl->scl_lock);
350		ASSERT(!refcount_is_zero(&scl->scl_count));
351		if (refcount_remove(&scl->scl_count, tag) == 0) {
352			ASSERT(scl->scl_writer == NULL ||
353			    scl->scl_writer == curthread);
354			scl->scl_writer = NULL;	/* OK in either case */
355			cv_broadcast(&scl->scl_cv);
356		}
357		mutex_exit(&scl->scl_lock);
358	}
359}
360
361int
362spa_config_held(spa_t *spa, int locks, krw_t rw)
363{
364	int locks_held = 0;
365
366	for (int i = 0; i < SCL_LOCKS; i++) {
367		spa_config_lock_t *scl = &spa->spa_config_lock[i];
368		if (!(locks & (1 << i)))
369			continue;
370		if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
371		    (rw == RW_WRITER && scl->scl_writer == curthread))
372			locks_held |= 1 << i;
373	}
374
375	return (locks_held);
376}
377
378/*
379 * ==========================================================================
380 * SPA namespace functions
381 * ==========================================================================
382 */
383
384/*
385 * Lookup the named spa_t in the AVL tree.  The spa_namespace_lock must be held.
386 * Returns NULL if no matching spa_t is found.
387 */
388spa_t *
389spa_lookup(const char *name)
390{
391	static spa_t search;	/* spa_t is large; don't allocate on stack */
392	spa_t *spa;
393	avl_index_t where;
394	char c;
395	char *cp;
396
397	ASSERT(MUTEX_HELD(&spa_namespace_lock));
398
399	/*
400	 * If it's a full dataset name, figure out the pool name and
401	 * just use that.
402	 */
403	cp = strpbrk(name, "/@");
404	if (cp) {
405		c = *cp;
406		*cp = '\0';
407	}
408
409	(void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
410	spa = avl_find(&spa_namespace_avl, &search, &where);
411
412	if (cp)
413		*cp = c;
414
415	return (spa);
416}
417
418/*
419 * Create an uninitialized spa_t with the given name.  Requires
420 * spa_namespace_lock.  The caller must ensure that the spa_t doesn't already
421 * exist by calling spa_lookup() first.
422 */
423spa_t *
424spa_add(const char *name, nvlist_t *config, const char *altroot)
425{
426	spa_t *spa;
427	spa_config_dirent_t *dp;
428
429	ASSERT(MUTEX_HELD(&spa_namespace_lock));
430
431	spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
432
433	mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
434	mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
435	mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
436	mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
437	mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
438	mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
439	mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
440	mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
441
442	cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
443	cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
444	cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
445
446	for (int t = 0; t < TXG_SIZE; t++)
447		bplist_init(&spa->spa_free_bplist[t]);
448	bplist_init(&spa->spa_deferred_bplist);
449
450	(void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
451	spa->spa_state = POOL_STATE_UNINITIALIZED;
452	spa->spa_freeze_txg = UINT64_MAX;
453	spa->spa_final_txg = UINT64_MAX;
454	spa->spa_load_max_txg = UINT64_MAX;
455
456	refcount_create(&spa->spa_refcount);
457	spa_config_lock_init(spa);
458
459	avl_add(&spa_namespace_avl, spa);
460
461	/*
462	 * Set the alternate root, if there is one.
463	 */
464	if (altroot) {
465		spa->spa_root = spa_strdup(altroot);
466		spa_active_count++;
467	}
468
469	/*
470	 * Every pool starts with the default cachefile
471	 */
472	list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
473	    offsetof(spa_config_dirent_t, scd_link));
474
475	dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
476	dp->scd_path = spa_strdup(spa_config_path);
477	list_insert_head(&spa->spa_config_list, dp);
478
479	if (config != NULL)
480		VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
481
482	return (spa);
483}
484
485/*
486 * Removes a spa_t from the namespace, freeing up any memory used.  Requires
487 * spa_namespace_lock.  This is called only after the spa_t has been closed and
488 * deactivated.
489 */
490void
491spa_remove(spa_t *spa)
492{
493	spa_config_dirent_t *dp;
494
495	ASSERT(MUTEX_HELD(&spa_namespace_lock));
496	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
497
498	avl_remove(&spa_namespace_avl, spa);
499	cv_broadcast(&spa_namespace_cv);
500
501	if (spa->spa_root) {
502		spa_strfree(spa->spa_root);
503		spa_active_count--;
504	}
505
506	while ((dp = list_head(&spa->spa_config_list)) != NULL) {
507		list_remove(&spa->spa_config_list, dp);
508		if (dp->scd_path != NULL)
509			spa_strfree(dp->scd_path);
510		kmem_free(dp, sizeof (spa_config_dirent_t));
511	}
512
513	list_destroy(&spa->spa_config_list);
514
515	spa_config_set(spa, NULL);
516
517	refcount_destroy(&spa->spa_refcount);
518
519	spa_config_lock_destroy(spa);
520
521	for (int t = 0; t < TXG_SIZE; t++)
522		bplist_fini(&spa->spa_free_bplist[t]);
523	bplist_fini(&spa->spa_deferred_bplist);
524
525	cv_destroy(&spa->spa_async_cv);
526	cv_destroy(&spa->spa_scrub_io_cv);
527	cv_destroy(&spa->spa_suspend_cv);
528
529	mutex_destroy(&spa->spa_async_lock);
530	mutex_destroy(&spa->spa_scrub_lock);
531	mutex_destroy(&spa->spa_errlog_lock);
532	mutex_destroy(&spa->spa_errlist_lock);
533	mutex_destroy(&spa->spa_history_lock);
534	mutex_destroy(&spa->spa_props_lock);
535	mutex_destroy(&spa->spa_suspend_lock);
536	mutex_destroy(&spa->spa_vdev_top_lock);
537
538	kmem_free(spa, sizeof (spa_t));
539}
540
541/*
542 * Given a pool, return the next pool in the namespace, or NULL if there is
543 * none.  If 'prev' is NULL, return the first pool.
544 */
545spa_t *
546spa_next(spa_t *prev)
547{
548	ASSERT(MUTEX_HELD(&spa_namespace_lock));
549
550	if (prev)
551		return (AVL_NEXT(&spa_namespace_avl, prev));
552	else
553		return (avl_first(&spa_namespace_avl));
554}
555
556/*
557 * ==========================================================================
558 * SPA refcount functions
559 * ==========================================================================
560 */
561
562/*
563 * Add a reference to the given spa_t.  Must have at least one reference, or
564 * have the namespace lock held.
565 */
566void
567spa_open_ref(spa_t *spa, void *tag)
568{
569	ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
570	    MUTEX_HELD(&spa_namespace_lock));
571	(void) refcount_add(&spa->spa_refcount, tag);
572}
573
574/*
575 * Remove a reference to the given spa_t.  Must have at least one reference, or
576 * have the namespace lock held.
577 */
578void
579spa_close(spa_t *spa, void *tag)
580{
581	ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
582	    MUTEX_HELD(&spa_namespace_lock));
583	(void) refcount_remove(&spa->spa_refcount, tag);
584}
585
586/*
587 * Check to see if the spa refcount is zero.  Must be called with
588 * spa_namespace_lock held.  We really compare against spa_minref, which is the
589 * number of references acquired when opening a pool
590 */
591boolean_t
592spa_refcount_zero(spa_t *spa)
593{
594	ASSERT(MUTEX_HELD(&spa_namespace_lock));
595
596	return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
597}
598
599/*
600 * ==========================================================================
601 * SPA spare and l2cache tracking
602 * ==========================================================================
603 */
604
605/*
606 * Hot spares and cache devices are tracked using the same code below,
607 * for 'auxiliary' devices.
608 */
609
610typedef struct spa_aux {
611	uint64_t	aux_guid;
612	uint64_t	aux_pool;
613	avl_node_t	aux_avl;
614	int		aux_count;
615} spa_aux_t;
616
617static int
618spa_aux_compare(const void *a, const void *b)
619{
620	const spa_aux_t *sa = a;
621	const spa_aux_t *sb = b;
622
623	if (sa->aux_guid < sb->aux_guid)
624		return (-1);
625	else if (sa->aux_guid > sb->aux_guid)
626		return (1);
627	else
628		return (0);
629}
630
631void
632spa_aux_add(vdev_t *vd, avl_tree_t *avl)
633{
634	avl_index_t where;
635	spa_aux_t search;
636	spa_aux_t *aux;
637
638	search.aux_guid = vd->vdev_guid;
639	if ((aux = avl_find(avl, &search, &where)) != NULL) {
640		aux->aux_count++;
641	} else {
642		aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
643		aux->aux_guid = vd->vdev_guid;
644		aux->aux_count = 1;
645		avl_insert(avl, aux, where);
646	}
647}
648
649void
650spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
651{
652	spa_aux_t search;
653	spa_aux_t *aux;
654	avl_index_t where;
655
656	search.aux_guid = vd->vdev_guid;
657	aux = avl_find(avl, &search, &where);
658
659	ASSERT(aux != NULL);
660
661	if (--aux->aux_count == 0) {
662		avl_remove(avl, aux);
663		kmem_free(aux, sizeof (spa_aux_t));
664	} else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
665		aux->aux_pool = 0ULL;
666	}
667}
668
669boolean_t
670spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
671{
672	spa_aux_t search, *found;
673
674	search.aux_guid = guid;
675	found = avl_find(avl, &search, NULL);
676
677	if (pool) {
678		if (found)
679			*pool = found->aux_pool;
680		else
681			*pool = 0ULL;
682	}
683
684	if (refcnt) {
685		if (found)
686			*refcnt = found->aux_count;
687		else
688			*refcnt = 0;
689	}
690
691	return (found != NULL);
692}
693
694void
695spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
696{
697	spa_aux_t search, *found;
698	avl_index_t where;
699
700	search.aux_guid = vd->vdev_guid;
701	found = avl_find(avl, &search, &where);
702	ASSERT(found != NULL);
703	ASSERT(found->aux_pool == 0ULL);
704
705	found->aux_pool = spa_guid(vd->vdev_spa);
706}
707
708/*
709 * Spares are tracked globally due to the following constraints:
710 *
711 * 	- A spare may be part of multiple pools.
712 * 	- A spare may be added to a pool even if it's actively in use within
713 *	  another pool.
714 * 	- A spare in use in any pool can only be the source of a replacement if
715 *	  the target is a spare in the same pool.
716 *
717 * We keep track of all spares on the system through the use of a reference
718 * counted AVL tree.  When a vdev is added as a spare, or used as a replacement
719 * spare, then we bump the reference count in the AVL tree.  In addition, we set
720 * the 'vdev_isspare' member to indicate that the device is a spare (active or
721 * inactive).  When a spare is made active (used to replace a device in the
722 * pool), we also keep track of which pool its been made a part of.
723 *
724 * The 'spa_spare_lock' protects the AVL tree.  These functions are normally
725 * called under the spa_namespace lock as part of vdev reconfiguration.  The
726 * separate spare lock exists for the status query path, which does not need to
727 * be completely consistent with respect to other vdev configuration changes.
728 */
729
730static int
731spa_spare_compare(const void *a, const void *b)
732{
733	return (spa_aux_compare(a, b));
734}
735
736void
737spa_spare_add(vdev_t *vd)
738{
739	mutex_enter(&spa_spare_lock);
740	ASSERT(!vd->vdev_isspare);
741	spa_aux_add(vd, &spa_spare_avl);
742	vd->vdev_isspare = B_TRUE;
743	mutex_exit(&spa_spare_lock);
744}
745
746void
747spa_spare_remove(vdev_t *vd)
748{
749	mutex_enter(&spa_spare_lock);
750	ASSERT(vd->vdev_isspare);
751	spa_aux_remove(vd, &spa_spare_avl);
752	vd->vdev_isspare = B_FALSE;
753	mutex_exit(&spa_spare_lock);
754}
755
756boolean_t
757spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
758{
759	boolean_t found;
760
761	mutex_enter(&spa_spare_lock);
762	found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
763	mutex_exit(&spa_spare_lock);
764
765	return (found);
766}
767
768void
769spa_spare_activate(vdev_t *vd)
770{
771	mutex_enter(&spa_spare_lock);
772	ASSERT(vd->vdev_isspare);
773	spa_aux_activate(vd, &spa_spare_avl);
774	mutex_exit(&spa_spare_lock);
775}
776
777/*
778 * Level 2 ARC devices are tracked globally for the same reasons as spares.
779 * Cache devices currently only support one pool per cache device, and so
780 * for these devices the aux reference count is currently unused beyond 1.
781 */
782
783static int
784spa_l2cache_compare(const void *a, const void *b)
785{
786	return (spa_aux_compare(a, b));
787}
788
789void
790spa_l2cache_add(vdev_t *vd)
791{
792	mutex_enter(&spa_l2cache_lock);
793	ASSERT(!vd->vdev_isl2cache);
794	spa_aux_add(vd, &spa_l2cache_avl);
795	vd->vdev_isl2cache = B_TRUE;
796	mutex_exit(&spa_l2cache_lock);
797}
798
799void
800spa_l2cache_remove(vdev_t *vd)
801{
802	mutex_enter(&spa_l2cache_lock);
803	ASSERT(vd->vdev_isl2cache);
804	spa_aux_remove(vd, &spa_l2cache_avl);
805	vd->vdev_isl2cache = B_FALSE;
806	mutex_exit(&spa_l2cache_lock);
807}
808
809boolean_t
810spa_l2cache_exists(uint64_t guid, uint64_t *pool)
811{
812	boolean_t found;
813
814	mutex_enter(&spa_l2cache_lock);
815	found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
816	mutex_exit(&spa_l2cache_lock);
817
818	return (found);
819}
820
821void
822spa_l2cache_activate(vdev_t *vd)
823{
824	mutex_enter(&spa_l2cache_lock);
825	ASSERT(vd->vdev_isl2cache);
826	spa_aux_activate(vd, &spa_l2cache_avl);
827	mutex_exit(&spa_l2cache_lock);
828}
829
830/*
831 * ==========================================================================
832 * SPA vdev locking
833 * ==========================================================================
834 */
835
836/*
837 * Lock the given spa_t for the purpose of adding or removing a vdev.
838 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
839 * It returns the next transaction group for the spa_t.
840 */
841uint64_t
842spa_vdev_enter(spa_t *spa)
843{
844	mutex_enter(&spa_namespace_lock);
845	mutex_enter(&spa->spa_vdev_top_lock);
846	return (spa_vdev_config_enter(spa));
847}
848
849/*
850 * Internal implementation for spa_vdev_enter().  Used when a vdev
851 * operation requires multiple syncs (i.e. removing a device) while
852 * keeping the spa_namespace_lock held.
853 */
854uint64_t
855spa_vdev_config_enter(spa_t *spa)
856{
857	ASSERT(MUTEX_HELD(&spa_namespace_lock));
858
859	spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
860
861	return (spa_last_synced_txg(spa) + 1);
862}
863
864/*
865 * Used in combination with spa_vdev_config_enter() to allow the syncing
866 * of multiple transactions without releasing the spa_namespace_lock.
867 */
868void
869spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
870{
871	ASSERT(MUTEX_HELD(&spa_namespace_lock));
872
873	int config_changed = B_FALSE;
874
875	ASSERT(txg > spa_last_synced_txg(spa));
876
877	spa->spa_pending_vdev = NULL;
878
879	/*
880	 * Reassess the DTLs.
881	 */
882	vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
883
884	/*
885	 * If the config changed, notify the scrub thread that it must restart.
886	 */
887	if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
888		dsl_pool_scrub_restart(spa->spa_dsl_pool);
889		config_changed = B_TRUE;
890		spa->spa_config_generation++;
891	}
892
893	/*
894	 * Verify the metaslab classes.
895	 */
896	ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
897	ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
898
899	spa_config_exit(spa, SCL_ALL, spa);
900
901	/*
902	 * Panic the system if the specified tag requires it.  This
903	 * is useful for ensuring that configurations are updated
904	 * transactionally.
905	 */
906	if (zio_injection_enabled)
907		zio_handle_panic_injection(spa, tag);
908
909	/*
910	 * Note: this txg_wait_synced() is important because it ensures
911	 * that there won't be more than one config change per txg.
912	 * This allows us to use the txg as the generation number.
913	 */
914	if (error == 0)
915		txg_wait_synced(spa->spa_dsl_pool, txg);
916
917	if (vd != NULL) {
918		ASSERT(!vd->vdev_detached || vd->vdev_dtl_smo.smo_object == 0);
919		spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
920		vdev_free(vd);
921		spa_config_exit(spa, SCL_ALL, spa);
922	}
923
924	/*
925	 * If the config changed, update the config cache.
926	 */
927	if (config_changed)
928		spa_config_sync(spa, B_FALSE, B_TRUE);
929}
930
931/*
932 * Unlock the spa_t after adding or removing a vdev.  Besides undoing the
933 * locking of spa_vdev_enter(), we also want make sure the transactions have
934 * synced to disk, and then update the global configuration cache with the new
935 * information.
936 */
937int
938spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
939{
940	spa_vdev_config_exit(spa, vd, txg, error, FTAG);
941	mutex_exit(&spa->spa_vdev_top_lock);
942	mutex_exit(&spa_namespace_lock);
943
944	return (error);
945}
946
947/*
948 * Lock the given spa_t for the purpose of changing vdev state.
949 */
950void
951spa_vdev_state_enter(spa_t *spa, int oplocks)
952{
953	int locks = SCL_STATE_ALL | oplocks;
954
955	spa_config_enter(spa, locks, spa, RW_WRITER);
956	spa->spa_vdev_locks = locks;
957}
958
959int
960spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
961{
962	if (vd != NULL || error == 0)
963		vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
964		    0, 0, B_FALSE);
965
966	if (vd != NULL) {
967		vdev_state_dirty(vd->vdev_top);
968		spa->spa_config_generation++;
969	}
970
971	ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
972	spa_config_exit(spa, spa->spa_vdev_locks, spa);
973
974	/*
975	 * If anything changed, wait for it to sync.  This ensures that,
976	 * from the system administrator's perspective, zpool(1M) commands
977	 * are synchronous.  This is important for things like zpool offline:
978	 * when the command completes, you expect no further I/O from ZFS.
979	 */
980	if (vd != NULL)
981		txg_wait_synced(spa->spa_dsl_pool, 0);
982
983	return (error);
984}
985
986/*
987 * ==========================================================================
988 * Miscellaneous functions
989 * ==========================================================================
990 */
991
992/*
993 * Rename a spa_t.
994 */
995int
996spa_rename(const char *name, const char *newname)
997{
998	spa_t *spa;
999	int err;
1000
1001	/*
1002	 * Lookup the spa_t and grab the config lock for writing.  We need to
1003	 * actually open the pool so that we can sync out the necessary labels.
1004	 * It's OK to call spa_open() with the namespace lock held because we
1005	 * allow recursive calls for other reasons.
1006	 */
1007	mutex_enter(&spa_namespace_lock);
1008	if ((err = spa_open(name, &spa, FTAG)) != 0) {
1009		mutex_exit(&spa_namespace_lock);
1010		return (err);
1011	}
1012
1013	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1014
1015	avl_remove(&spa_namespace_avl, spa);
1016	(void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1017	avl_add(&spa_namespace_avl, spa);
1018
1019	/*
1020	 * Sync all labels to disk with the new names by marking the root vdev
1021	 * dirty and waiting for it to sync.  It will pick up the new pool name
1022	 * during the sync.
1023	 */
1024	vdev_config_dirty(spa->spa_root_vdev);
1025
1026	spa_config_exit(spa, SCL_ALL, FTAG);
1027
1028	txg_wait_synced(spa->spa_dsl_pool, 0);
1029
1030	/*
1031	 * Sync the updated config cache.
1032	 */
1033	spa_config_sync(spa, B_FALSE, B_TRUE);
1034
1035	spa_close(spa, FTAG);
1036
1037	mutex_exit(&spa_namespace_lock);
1038
1039	return (0);
1040}
1041
1042
1043/*
1044 * Determine whether a pool with given pool_guid exists.  If device_guid is
1045 * non-zero, determine whether the pool exists *and* contains a device with the
1046 * specified device_guid.
1047 */
1048boolean_t
1049spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1050{
1051	spa_t *spa;
1052	avl_tree_t *t = &spa_namespace_avl;
1053
1054	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1055
1056	for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1057		if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1058			continue;
1059		if (spa->spa_root_vdev == NULL)
1060			continue;
1061		if (spa_guid(spa) == pool_guid) {
1062			if (device_guid == 0)
1063				break;
1064
1065			if (vdev_lookup_by_guid(spa->spa_root_vdev,
1066			    device_guid) != NULL)
1067				break;
1068
1069			/*
1070			 * Check any devices we may be in the process of adding.
1071			 */
1072			if (spa->spa_pending_vdev) {
1073				if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1074				    device_guid) != NULL)
1075					break;
1076			}
1077		}
1078	}
1079
1080	return (spa != NULL);
1081}
1082
1083char *
1084spa_strdup(const char *s)
1085{
1086	size_t len;
1087	char *new;
1088
1089	len = strlen(s);
1090	new = kmem_alloc(len + 1, KM_SLEEP);
1091	bcopy(s, new, len);
1092	new[len] = '\0';
1093
1094	return (new);
1095}
1096
1097void
1098spa_strfree(char *s)
1099{
1100	kmem_free(s, strlen(s) + 1);
1101}
1102
1103uint64_t
1104spa_get_random(uint64_t range)
1105{
1106	uint64_t r;
1107
1108	ASSERT(range != 0);
1109
1110	(void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1111
1112	return (r % range);
1113}
1114
1115void
1116sprintf_blkptr(char *buf, const blkptr_t *bp)
1117{
1118	char *type = dmu_ot[BP_GET_TYPE(bp)].ot_name;
1119	char *checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1120	char *compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1121
1122	SPRINTF_BLKPTR(snprintf, ' ', buf, bp, type, checksum, compress);
1123}
1124
1125void
1126spa_freeze(spa_t *spa)
1127{
1128	uint64_t freeze_txg = 0;
1129
1130	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1131	if (spa->spa_freeze_txg == UINT64_MAX) {
1132		freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1133		spa->spa_freeze_txg = freeze_txg;
1134	}
1135	spa_config_exit(spa, SCL_ALL, FTAG);
1136	if (freeze_txg != 0)
1137		txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1138}
1139
1140void
1141zfs_panic_recover(const char *fmt, ...)
1142{
1143	va_list adx;
1144
1145	va_start(adx, fmt);
1146	vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1147	va_end(adx);
1148}
1149
1150/*
1151 * ==========================================================================
1152 * Accessor functions
1153 * ==========================================================================
1154 */
1155
1156boolean_t
1157spa_shutting_down(spa_t *spa)
1158{
1159	return (spa->spa_async_suspended);
1160}
1161
1162dsl_pool_t *
1163spa_get_dsl(spa_t *spa)
1164{
1165	return (spa->spa_dsl_pool);
1166}
1167
1168blkptr_t *
1169spa_get_rootblkptr(spa_t *spa)
1170{
1171	return (&spa->spa_ubsync.ub_rootbp);
1172}
1173
1174void
1175spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1176{
1177	spa->spa_uberblock.ub_rootbp = *bp;
1178}
1179
1180void
1181spa_altroot(spa_t *spa, char *buf, size_t buflen)
1182{
1183	if (spa->spa_root == NULL)
1184		buf[0] = '\0';
1185	else
1186		(void) strncpy(buf, spa->spa_root, buflen);
1187}
1188
1189int
1190spa_sync_pass(spa_t *spa)
1191{
1192	return (spa->spa_sync_pass);
1193}
1194
1195char *
1196spa_name(spa_t *spa)
1197{
1198	return (spa->spa_name);
1199}
1200
1201uint64_t
1202spa_guid(spa_t *spa)
1203{
1204	/*
1205	 * If we fail to parse the config during spa_load(), we can go through
1206	 * the error path (which posts an ereport) and end up here with no root
1207	 * vdev.  We stash the original pool guid in 'spa_load_guid' to handle
1208	 * this case.
1209	 */
1210	if (spa->spa_root_vdev != NULL)
1211		return (spa->spa_root_vdev->vdev_guid);
1212	else
1213		return (spa->spa_load_guid);
1214}
1215
1216uint64_t
1217spa_last_synced_txg(spa_t *spa)
1218{
1219	return (spa->spa_ubsync.ub_txg);
1220}
1221
1222uint64_t
1223spa_first_txg(spa_t *spa)
1224{
1225	return (spa->spa_first_txg);
1226}
1227
1228uint64_t
1229spa_syncing_txg(spa_t *spa)
1230{
1231	return (spa->spa_syncing_txg);
1232}
1233
1234pool_state_t
1235spa_state(spa_t *spa)
1236{
1237	return (spa->spa_state);
1238}
1239
1240uint64_t
1241spa_freeze_txg(spa_t *spa)
1242{
1243	return (spa->spa_freeze_txg);
1244}
1245
1246/* ARGSUSED */
1247uint64_t
1248spa_get_asize(spa_t *spa, uint64_t lsize)
1249{
1250	/*
1251	 * The worst case is single-sector max-parity RAID-Z blocks, in which
1252	 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
1253	 * times the size; so just assume that.  Add to this the fact that
1254	 * we can have up to 3 DVAs per bp, and one more factor of 2 because
1255	 * the block may be dittoed with up to 3 DVAs by ddt_sync().
1256	 */
1257	return (lsize * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2);
1258}
1259
1260uint64_t
1261spa_get_dspace(spa_t *spa)
1262{
1263	return (spa->spa_dspace);
1264}
1265
1266void
1267spa_update_dspace(spa_t *spa)
1268{
1269	spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1270	    ddt_get_dedup_dspace(spa);
1271}
1272
1273/*
1274 * Return the failure mode that has been set to this pool. The default
1275 * behavior will be to block all I/Os when a complete failure occurs.
1276 */
1277uint8_t
1278spa_get_failmode(spa_t *spa)
1279{
1280	return (spa->spa_failmode);
1281}
1282
1283boolean_t
1284spa_suspended(spa_t *spa)
1285{
1286	return (spa->spa_suspended);
1287}
1288
1289uint64_t
1290spa_version(spa_t *spa)
1291{
1292	return (spa->spa_ubsync.ub_version);
1293}
1294
1295boolean_t
1296spa_deflate(spa_t *spa)
1297{
1298	return (spa->spa_deflate);
1299}
1300
1301metaslab_class_t *
1302spa_normal_class(spa_t *spa)
1303{
1304	return (spa->spa_normal_class);
1305}
1306
1307metaslab_class_t *
1308spa_log_class(spa_t *spa)
1309{
1310	return (spa->spa_log_class);
1311}
1312
1313int
1314spa_max_replication(spa_t *spa)
1315{
1316	/*
1317	 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1318	 * handle BPs with more than one DVA allocated.  Set our max
1319	 * replication level accordingly.
1320	 */
1321	if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1322		return (1);
1323	return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1324}
1325
1326uint64_t
1327dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1328{
1329	uint64_t asize = DVA_GET_ASIZE(dva);
1330	uint64_t dsize = asize;
1331
1332	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1333
1334	if (asize != 0 && spa->spa_deflate) {
1335		vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1336		dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1337	}
1338
1339	return (dsize);
1340}
1341
1342uint64_t
1343bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1344{
1345	uint64_t dsize = 0;
1346
1347	for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1348		dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1349
1350	return (dsize);
1351}
1352
1353uint64_t
1354bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1355{
1356	uint64_t dsize = 0;
1357
1358	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1359
1360	for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1361		dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1362
1363	spa_config_exit(spa, SCL_VDEV, FTAG);
1364
1365	return (dsize);
1366}
1367
1368/*
1369 * ==========================================================================
1370 * Initialization and Termination
1371 * ==========================================================================
1372 */
1373
1374static int
1375spa_name_compare(const void *a1, const void *a2)
1376{
1377	const spa_t *s1 = a1;
1378	const spa_t *s2 = a2;
1379	int s;
1380
1381	s = strcmp(s1->spa_name, s2->spa_name);
1382	if (s > 0)
1383		return (1);
1384	if (s < 0)
1385		return (-1);
1386	return (0);
1387}
1388
1389int
1390spa_busy(void)
1391{
1392	return (spa_active_count);
1393}
1394
1395void
1396spa_boot_init()
1397{
1398	spa_config_load();
1399}
1400
1401void
1402spa_init(int mode)
1403{
1404	mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1405	mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1406	mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1407	cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1408
1409	avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1410	    offsetof(spa_t, spa_avl));
1411
1412	avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1413	    offsetof(spa_aux_t, aux_avl));
1414
1415	avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1416	    offsetof(spa_aux_t, aux_avl));
1417
1418	spa_mode_global = mode;
1419
1420	refcount_init();
1421	unique_init();
1422	zio_init();
1423	dmu_init();
1424	zil_init();
1425	vdev_cache_stat_init();
1426	zfs_prop_init();
1427	zpool_prop_init();
1428	spa_config_load();
1429	l2arc_start();
1430}
1431
1432void
1433spa_fini(void)
1434{
1435	l2arc_stop();
1436
1437	spa_evict_all();
1438
1439	vdev_cache_stat_fini();
1440	zil_fini();
1441	dmu_fini();
1442	zio_fini();
1443	unique_fini();
1444	refcount_fini();
1445
1446	avl_destroy(&spa_namespace_avl);
1447	avl_destroy(&spa_spare_avl);
1448	avl_destroy(&spa_l2cache_avl);
1449
1450	cv_destroy(&spa_namespace_cv);
1451	mutex_destroy(&spa_namespace_lock);
1452	mutex_destroy(&spa_spare_lock);
1453	mutex_destroy(&spa_l2cache_lock);
1454}
1455
1456/*
1457 * Return whether this pool has slogs. No locking needed.
1458 * It's not a problem if the wrong answer is returned as it's only for
1459 * performance and not correctness
1460 */
1461boolean_t
1462spa_has_slogs(spa_t *spa)
1463{
1464	return (spa->spa_log_class->mc_rotor != NULL);
1465}
1466
1467spa_log_state_t
1468spa_get_log_state(spa_t *spa)
1469{
1470	return (spa->spa_log_state);
1471}
1472
1473void
1474spa_set_log_state(spa_t *spa, spa_log_state_t state)
1475{
1476	spa->spa_log_state = state;
1477}
1478
1479boolean_t
1480spa_is_root(spa_t *spa)
1481{
1482	return (spa->spa_is_root);
1483}
1484
1485boolean_t
1486spa_writeable(spa_t *spa)
1487{
1488	return (!!(spa->spa_mode & FWRITE));
1489}
1490
1491int
1492spa_mode(spa_t *spa)
1493{
1494	return (spa->spa_mode);
1495}
1496
1497uint64_t
1498spa_bootfs(spa_t *spa)
1499{
1500	return (spa->spa_bootfs);
1501}
1502
1503uint64_t
1504spa_delegation(spa_t *spa)
1505{
1506	return (spa->spa_delegation);
1507}
1508
1509objset_t *
1510spa_meta_objset(spa_t *spa)
1511{
1512	return (spa->spa_meta_objset);
1513}
1514
1515enum zio_checksum
1516spa_dedup_checksum(spa_t *spa)
1517{
1518	return (spa->spa_dedup_checksum);
1519}
1520