spa_misc.c revision ef912c8004c1fbd9aa5844e85ce9bf2ce17f9222
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 2010 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/arc.h>
47#include <sys/ddt.h>
48#include "zfs_prop.h"
49
50/*
51 * SPA locking
52 *
53 * There are four basic locks for managing spa_t structures:
54 *
55 * spa_namespace_lock (global mutex)
56 *
57 *	This lock must be acquired to do any of the following:
58 *
59 *		- Lookup a spa_t by name
60 *		- Add or remove a spa_t from the namespace
61 *		- Increase spa_refcount from non-zero
62 *		- Check if spa_refcount is zero
63 *		- Rename a spa_t
64 *		- add/remove/attach/detach devices
65 *		- Held for the duration of create/destroy/import/export
66 *
67 *	It does not need to handle recursion.  A create or destroy may
68 *	reference objects (files or zvols) in other pools, but by
69 *	definition they must have an existing reference, and will never need
70 *	to lookup a spa_t by name.
71 *
72 * spa_refcount (per-spa refcount_t protected by mutex)
73 *
74 *	This reference count keep track of any active users of the spa_t.  The
75 *	spa_t cannot be destroyed or freed while this is non-zero.  Internally,
76 *	the refcount is never really 'zero' - opening a pool implicitly keeps
77 *	some references in the DMU.  Internally we check against spa_minref, but
78 *	present the image of a zero/non-zero value to consumers.
79 *
80 * spa_config_lock[] (per-spa array of rwlocks)
81 *
82 *	This protects the spa_t from config changes, and must be held in
83 *	the following circumstances:
84 *
85 *		- RW_READER to perform I/O to the spa
86 *		- RW_WRITER to change the vdev config
87 *
88 * The locking order is fairly straightforward:
89 *
90 *		spa_namespace_lock	->	spa_refcount
91 *
92 *	The namespace lock must be acquired to increase the refcount from 0
93 *	or to check if it is zero.
94 *
95 *		spa_refcount		->	spa_config_lock[]
96 *
97 *	There must be at least one valid reference on the spa_t to acquire
98 *	the config lock.
99 *
100 *		spa_namespace_lock	->	spa_config_lock[]
101 *
102 *	The namespace lock must always be taken before the config lock.
103 *
104 *
105 * The spa_namespace_lock can be acquired directly and is globally visible.
106 *
107 * The namespace is manipulated using the following functions, all of which
108 * require the spa_namespace_lock to be held.
109 *
110 *	spa_lookup()		Lookup a spa_t by name.
111 *
112 *	spa_add()		Create a new spa_t in the namespace.
113 *
114 *	spa_remove()		Remove a spa_t from the namespace.  This also
115 *				frees up any memory associated with the spa_t.
116 *
117 *	spa_next()		Returns the next spa_t in the system, or the
118 *				first if NULL is passed.
119 *
120 *	spa_evict_all()		Shutdown and remove all spa_t structures in
121 *				the system.
122 *
123 *	spa_guid_exists()	Determine whether a pool/device guid exists.
124 *
125 * The spa_refcount is manipulated using the following functions:
126 *
127 *	spa_open_ref()		Adds a reference to the given spa_t.  Must be
128 *				called with spa_namespace_lock held if the
129 *				refcount is currently zero.
130 *
131 *	spa_close()		Remove a reference from the spa_t.  This will
132 *				not free the spa_t or remove it from the
133 *				namespace.  No locking is required.
134 *
135 *	spa_refcount_zero()	Returns true if the refcount is currently
136 *				zero.  Must be called with spa_namespace_lock
137 *				held.
138 *
139 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
140 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
141 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
142 *
143 * To read the configuration, it suffices to hold one of these locks as reader.
144 * To modify the configuration, you must hold all locks as writer.  To modify
145 * vdev state without altering the vdev tree's topology (e.g. online/offline),
146 * you must hold SCL_STATE and SCL_ZIO as writer.
147 *
148 * We use these distinct config locks to avoid recursive lock entry.
149 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
150 * block allocations (SCL_ALLOC), which may require reading space maps
151 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
152 *
153 * The spa config locks cannot be normal rwlocks because we need the
154 * ability to hand off ownership.  For example, SCL_ZIO is acquired
155 * by the issuing thread and later released by an interrupt thread.
156 * They do, however, obey the usual write-wanted semantics to prevent
157 * writer (i.e. system administrator) starvation.
158 *
159 * The lock acquisition rules are as follows:
160 *
161 * SCL_CONFIG
162 *	Protects changes to the vdev tree topology, such as vdev
163 *	add/remove/attach/detach.  Protects the dirty config list
164 *	(spa_config_dirty_list) and the set of spares and l2arc devices.
165 *
166 * SCL_STATE
167 *	Protects changes to pool state and vdev state, such as vdev
168 *	online/offline/fault/degrade/clear.  Protects the dirty state list
169 *	(spa_state_dirty_list) and global pool state (spa_state).
170 *
171 * SCL_ALLOC
172 *	Protects changes to metaslab groups and classes.
173 *	Held as reader by metaslab_alloc() and metaslab_claim().
174 *
175 * SCL_ZIO
176 *	Held by bp-level zios (those which have no io_vd upon entry)
177 *	to prevent changes to the vdev tree.  The bp-level zio implicitly
178 *	protects all of its vdev child zios, which do not hold SCL_ZIO.
179 *
180 * SCL_FREE
181 *	Protects changes to metaslab groups and classes.
182 *	Held as reader by metaslab_free().  SCL_FREE is distinct from
183 *	SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
184 *	blocks in zio_done() while another i/o that holds either
185 *	SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
186 *
187 * SCL_VDEV
188 *	Held as reader to prevent changes to the vdev tree during trivial
189 *	inquiries such as bp_get_dsize().  SCL_VDEV is distinct from the
190 *	other locks, and lower than all of them, to ensure that it's safe
191 *	to acquire regardless of caller context.
192 *
193 * In addition, the following rules apply:
194 *
195 * (a)	spa_props_lock protects pool properties, spa_config and spa_config_list.
196 *	The lock ordering is SCL_CONFIG > spa_props_lock.
197 *
198 * (b)	I/O operations on leaf vdevs.  For any zio operation that takes
199 *	an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
200 *	or zio_write_phys() -- the caller must ensure that the config cannot
201 *	cannot change in the interim, and that the vdev cannot be reopened.
202 *	SCL_STATE as reader suffices for both.
203 *
204 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
205 *
206 *	spa_vdev_enter()	Acquire the namespace lock and the config lock
207 *				for writing.
208 *
209 *	spa_vdev_exit()		Release the config lock, wait for all I/O
210 *				to complete, sync the updated configs to the
211 *				cache, and release the namespace lock.
212 *
213 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
214 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
215 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
216 *
217 * spa_rename() is also implemented within this file since is requires
218 * manipulation of the namespace.
219 */
220
221static avl_tree_t spa_namespace_avl;
222kmutex_t spa_namespace_lock;
223static kcondvar_t spa_namespace_cv;
224static int spa_active_count;
225int spa_max_replication_override = SPA_DVAS_PER_BP;
226
227static kmutex_t spa_spare_lock;
228static avl_tree_t spa_spare_avl;
229static kmutex_t spa_l2cache_lock;
230static avl_tree_t spa_l2cache_avl;
231
232kmem_cache_t *spa_buffer_pool;
233int spa_mode_global;
234
235#ifdef ZFS_DEBUG
236/* Everything except dprintf is on by default in debug builds */
237int zfs_flags = ~ZFS_DEBUG_DPRINTF;
238#else
239int zfs_flags = 0;
240#endif
241
242/*
243 * zfs_recover can be set to nonzero to attempt to recover from
244 * otherwise-fatal errors, typically caused by on-disk corruption.  When
245 * set, calls to zfs_panic_recover() will turn into warning messages.
246 */
247int zfs_recover = 0;
248
249
250/*
251 * ==========================================================================
252 * SPA config locking
253 * ==========================================================================
254 */
255static void
256spa_config_lock_init(spa_t *spa)
257{
258	for (int i = 0; i < SCL_LOCKS; i++) {
259		spa_config_lock_t *scl = &spa->spa_config_lock[i];
260		mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
261		cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
262		refcount_create(&scl->scl_count);
263		scl->scl_writer = NULL;
264		scl->scl_write_wanted = 0;
265	}
266}
267
268static void
269spa_config_lock_destroy(spa_t *spa)
270{
271	for (int i = 0; i < SCL_LOCKS; i++) {
272		spa_config_lock_t *scl = &spa->spa_config_lock[i];
273		mutex_destroy(&scl->scl_lock);
274		cv_destroy(&scl->scl_cv);
275		refcount_destroy(&scl->scl_count);
276		ASSERT(scl->scl_writer == NULL);
277		ASSERT(scl->scl_write_wanted == 0);
278	}
279}
280
281int
282spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
283{
284	for (int i = 0; i < SCL_LOCKS; i++) {
285		spa_config_lock_t *scl = &spa->spa_config_lock[i];
286		if (!(locks & (1 << i)))
287			continue;
288		mutex_enter(&scl->scl_lock);
289		if (rw == RW_READER) {
290			if (scl->scl_writer || scl->scl_write_wanted) {
291				mutex_exit(&scl->scl_lock);
292				spa_config_exit(spa, locks ^ (1 << i), tag);
293				return (0);
294			}
295		} else {
296			ASSERT(scl->scl_writer != curthread);
297			if (!refcount_is_zero(&scl->scl_count)) {
298				mutex_exit(&scl->scl_lock);
299				spa_config_exit(spa, locks ^ (1 << i), tag);
300				return (0);
301			}
302			scl->scl_writer = curthread;
303		}
304		(void) refcount_add(&scl->scl_count, tag);
305		mutex_exit(&scl->scl_lock);
306	}
307	return (1);
308}
309
310void
311spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
312{
313	int wlocks_held = 0;
314
315	for (int i = 0; i < SCL_LOCKS; i++) {
316		spa_config_lock_t *scl = &spa->spa_config_lock[i];
317		if (scl->scl_writer == curthread)
318			wlocks_held |= (1 << i);
319		if (!(locks & (1 << i)))
320			continue;
321		mutex_enter(&scl->scl_lock);
322		if (rw == RW_READER) {
323			while (scl->scl_writer || scl->scl_write_wanted) {
324				cv_wait(&scl->scl_cv, &scl->scl_lock);
325			}
326		} else {
327			ASSERT(scl->scl_writer != curthread);
328			while (!refcount_is_zero(&scl->scl_count)) {
329				scl->scl_write_wanted++;
330				cv_wait(&scl->scl_cv, &scl->scl_lock);
331				scl->scl_write_wanted--;
332			}
333			scl->scl_writer = curthread;
334		}
335		(void) refcount_add(&scl->scl_count, tag);
336		mutex_exit(&scl->scl_lock);
337	}
338	ASSERT(wlocks_held <= locks);
339}
340
341void
342spa_config_exit(spa_t *spa, int locks, void *tag)
343{
344	for (int i = SCL_LOCKS - 1; i >= 0; i--) {
345		spa_config_lock_t *scl = &spa->spa_config_lock[i];
346		if (!(locks & (1 << i)))
347			continue;
348		mutex_enter(&scl->scl_lock);
349		ASSERT(!refcount_is_zero(&scl->scl_count));
350		if (refcount_remove(&scl->scl_count, tag) == 0) {
351			ASSERT(scl->scl_writer == NULL ||
352			    scl->scl_writer == curthread);
353			scl->scl_writer = NULL;	/* OK in either case */
354			cv_broadcast(&scl->scl_cv);
355		}
356		mutex_exit(&scl->scl_lock);
357	}
358}
359
360int
361spa_config_held(spa_t *spa, int locks, krw_t rw)
362{
363	int locks_held = 0;
364
365	for (int i = 0; i < SCL_LOCKS; i++) {
366		spa_config_lock_t *scl = &spa->spa_config_lock[i];
367		if (!(locks & (1 << i)))
368			continue;
369		if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
370		    (rw == RW_WRITER && scl->scl_writer == curthread))
371			locks_held |= 1 << i;
372	}
373
374	return (locks_held);
375}
376
377/*
378 * ==========================================================================
379 * SPA namespace functions
380 * ==========================================================================
381 */
382
383/*
384 * Lookup the named spa_t in the AVL tree.  The spa_namespace_lock must be held.
385 * Returns NULL if no matching spa_t is found.
386 */
387spa_t *
388spa_lookup(const char *name)
389{
390	static spa_t search;	/* spa_t is large; don't allocate on stack */
391	spa_t *spa;
392	avl_index_t where;
393	char c;
394	char *cp;
395
396	ASSERT(MUTEX_HELD(&spa_namespace_lock));
397
398	/*
399	 * If it's a full dataset name, figure out the pool name and
400	 * just use that.
401	 */
402	cp = strpbrk(name, "/@");
403	if (cp) {
404		c = *cp;
405		*cp = '\0';
406	}
407
408	(void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
409	spa = avl_find(&spa_namespace_avl, &search, &where);
410
411	if (cp)
412		*cp = c;
413
414	return (spa);
415}
416
417/*
418 * Create an uninitialized spa_t with the given name.  Requires
419 * spa_namespace_lock.  The caller must ensure that the spa_t doesn't already
420 * exist by calling spa_lookup() first.
421 */
422spa_t *
423spa_add(const char *name, nvlist_t *config, const char *altroot)
424{
425	spa_t *spa;
426	spa_config_dirent_t *dp;
427
428	ASSERT(MUTEX_HELD(&spa_namespace_lock));
429
430	spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
431
432	mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
433	mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
434	mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
435	mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
436	mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
437	mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
438	mutex_init(&spa->spa_scrub_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_proc_cv, NULL, CV_DEFAULT, NULL);
444	cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
445	cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
446
447	for (int t = 0; t < TXG_SIZE; t++)
448		bplist_init(&spa->spa_free_bplist[t]);
449	bplist_init(&spa->spa_deferred_bplist);
450
451	(void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
452	spa->spa_state = POOL_STATE_UNINITIALIZED;
453	spa->spa_freeze_txg = UINT64_MAX;
454	spa->spa_final_txg = UINT64_MAX;
455	spa->spa_load_max_txg = UINT64_MAX;
456	spa->spa_proc = &p0;
457	spa->spa_proc_state = SPA_PROC_NONE;
458
459	refcount_create(&spa->spa_refcount);
460	spa_config_lock_init(spa);
461
462	avl_add(&spa_namespace_avl, spa);
463
464	/*
465	 * Set the alternate root, if there is one.
466	 */
467	if (altroot) {
468		spa->spa_root = spa_strdup(altroot);
469		spa_active_count++;
470	}
471
472	/*
473	 * Every pool starts with the default cachefile
474	 */
475	list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
476	    offsetof(spa_config_dirent_t, scd_link));
477
478	dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
479	dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
480	list_insert_head(&spa->spa_config_list, dp);
481
482	if (config != NULL)
483		VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
484
485	return (spa);
486}
487
488/*
489 * Removes a spa_t from the namespace, freeing up any memory used.  Requires
490 * spa_namespace_lock.  This is called only after the spa_t has been closed and
491 * deactivated.
492 */
493void
494spa_remove(spa_t *spa)
495{
496	spa_config_dirent_t *dp;
497
498	ASSERT(MUTEX_HELD(&spa_namespace_lock));
499	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
500
501	nvlist_free(spa->spa_config_splitting);
502
503	avl_remove(&spa_namespace_avl, spa);
504	cv_broadcast(&spa_namespace_cv);
505
506	if (spa->spa_root) {
507		spa_strfree(spa->spa_root);
508		spa_active_count--;
509	}
510
511	while ((dp = list_head(&spa->spa_config_list)) != NULL) {
512		list_remove(&spa->spa_config_list, dp);
513		if (dp->scd_path != NULL)
514			spa_strfree(dp->scd_path);
515		kmem_free(dp, sizeof (spa_config_dirent_t));
516	}
517
518	list_destroy(&spa->spa_config_list);
519
520	spa_config_set(spa, NULL);
521
522	refcount_destroy(&spa->spa_refcount);
523
524	spa_config_lock_destroy(spa);
525
526	for (int t = 0; t < TXG_SIZE; t++)
527		bplist_fini(&spa->spa_free_bplist[t]);
528	bplist_fini(&spa->spa_deferred_bplist);
529
530	cv_destroy(&spa->spa_async_cv);
531	cv_destroy(&spa->spa_proc_cv);
532	cv_destroy(&spa->spa_scrub_io_cv);
533	cv_destroy(&spa->spa_suspend_cv);
534
535	mutex_destroy(&spa->spa_async_lock);
536	mutex_destroy(&spa->spa_errlist_lock);
537	mutex_destroy(&spa->spa_errlog_lock);
538	mutex_destroy(&spa->spa_history_lock);
539	mutex_destroy(&spa->spa_proc_lock);
540	mutex_destroy(&spa->spa_props_lock);
541	mutex_destroy(&spa->spa_scrub_lock);
542	mutex_destroy(&spa->spa_suspend_lock);
543	mutex_destroy(&spa->spa_vdev_top_lock);
544
545	kmem_free(spa, sizeof (spa_t));
546}
547
548/*
549 * Given a pool, return the next pool in the namespace, or NULL if there is
550 * none.  If 'prev' is NULL, return the first pool.
551 */
552spa_t *
553spa_next(spa_t *prev)
554{
555	ASSERT(MUTEX_HELD(&spa_namespace_lock));
556
557	if (prev)
558		return (AVL_NEXT(&spa_namespace_avl, prev));
559	else
560		return (avl_first(&spa_namespace_avl));
561}
562
563/*
564 * ==========================================================================
565 * SPA refcount functions
566 * ==========================================================================
567 */
568
569/*
570 * Add a reference to the given spa_t.  Must have at least one reference, or
571 * have the namespace lock held.
572 */
573void
574spa_open_ref(spa_t *spa, void *tag)
575{
576	ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
577	    MUTEX_HELD(&spa_namespace_lock));
578	(void) refcount_add(&spa->spa_refcount, tag);
579}
580
581/*
582 * Remove a reference to the given spa_t.  Must have at least one reference, or
583 * have the namespace lock held.
584 */
585void
586spa_close(spa_t *spa, void *tag)
587{
588	ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
589	    MUTEX_HELD(&spa_namespace_lock));
590	(void) refcount_remove(&spa->spa_refcount, tag);
591}
592
593/*
594 * Check to see if the spa refcount is zero.  Must be called with
595 * spa_namespace_lock held.  We really compare against spa_minref, which is the
596 * number of references acquired when opening a pool
597 */
598boolean_t
599spa_refcount_zero(spa_t *spa)
600{
601	ASSERT(MUTEX_HELD(&spa_namespace_lock));
602
603	return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
604}
605
606/*
607 * ==========================================================================
608 * SPA spare and l2cache tracking
609 * ==========================================================================
610 */
611
612/*
613 * Hot spares and cache devices are tracked using the same code below,
614 * for 'auxiliary' devices.
615 */
616
617typedef struct spa_aux {
618	uint64_t	aux_guid;
619	uint64_t	aux_pool;
620	avl_node_t	aux_avl;
621	int		aux_count;
622} spa_aux_t;
623
624static int
625spa_aux_compare(const void *a, const void *b)
626{
627	const spa_aux_t *sa = a;
628	const spa_aux_t *sb = b;
629
630	if (sa->aux_guid < sb->aux_guid)
631		return (-1);
632	else if (sa->aux_guid > sb->aux_guid)
633		return (1);
634	else
635		return (0);
636}
637
638void
639spa_aux_add(vdev_t *vd, avl_tree_t *avl)
640{
641	avl_index_t where;
642	spa_aux_t search;
643	spa_aux_t *aux;
644
645	search.aux_guid = vd->vdev_guid;
646	if ((aux = avl_find(avl, &search, &where)) != NULL) {
647		aux->aux_count++;
648	} else {
649		aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
650		aux->aux_guid = vd->vdev_guid;
651		aux->aux_count = 1;
652		avl_insert(avl, aux, where);
653	}
654}
655
656void
657spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
658{
659	spa_aux_t search;
660	spa_aux_t *aux;
661	avl_index_t where;
662
663	search.aux_guid = vd->vdev_guid;
664	aux = avl_find(avl, &search, &where);
665
666	ASSERT(aux != NULL);
667
668	if (--aux->aux_count == 0) {
669		avl_remove(avl, aux);
670		kmem_free(aux, sizeof (spa_aux_t));
671	} else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
672		aux->aux_pool = 0ULL;
673	}
674}
675
676boolean_t
677spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
678{
679	spa_aux_t search, *found;
680
681	search.aux_guid = guid;
682	found = avl_find(avl, &search, NULL);
683
684	if (pool) {
685		if (found)
686			*pool = found->aux_pool;
687		else
688			*pool = 0ULL;
689	}
690
691	if (refcnt) {
692		if (found)
693			*refcnt = found->aux_count;
694		else
695			*refcnt = 0;
696	}
697
698	return (found != NULL);
699}
700
701void
702spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
703{
704	spa_aux_t search, *found;
705	avl_index_t where;
706
707	search.aux_guid = vd->vdev_guid;
708	found = avl_find(avl, &search, &where);
709	ASSERT(found != NULL);
710	ASSERT(found->aux_pool == 0ULL);
711
712	found->aux_pool = spa_guid(vd->vdev_spa);
713}
714
715/*
716 * Spares are tracked globally due to the following constraints:
717 *
718 * 	- A spare may be part of multiple pools.
719 * 	- A spare may be added to a pool even if it's actively in use within
720 *	  another pool.
721 * 	- A spare in use in any pool can only be the source of a replacement if
722 *	  the target is a spare in the same pool.
723 *
724 * We keep track of all spares on the system through the use of a reference
725 * counted AVL tree.  When a vdev is added as a spare, or used as a replacement
726 * spare, then we bump the reference count in the AVL tree.  In addition, we set
727 * the 'vdev_isspare' member to indicate that the device is a spare (active or
728 * inactive).  When a spare is made active (used to replace a device in the
729 * pool), we also keep track of which pool its been made a part of.
730 *
731 * The 'spa_spare_lock' protects the AVL tree.  These functions are normally
732 * called under the spa_namespace lock as part of vdev reconfiguration.  The
733 * separate spare lock exists for the status query path, which does not need to
734 * be completely consistent with respect to other vdev configuration changes.
735 */
736
737static int
738spa_spare_compare(const void *a, const void *b)
739{
740	return (spa_aux_compare(a, b));
741}
742
743void
744spa_spare_add(vdev_t *vd)
745{
746	mutex_enter(&spa_spare_lock);
747	ASSERT(!vd->vdev_isspare);
748	spa_aux_add(vd, &spa_spare_avl);
749	vd->vdev_isspare = B_TRUE;
750	mutex_exit(&spa_spare_lock);
751}
752
753void
754spa_spare_remove(vdev_t *vd)
755{
756	mutex_enter(&spa_spare_lock);
757	ASSERT(vd->vdev_isspare);
758	spa_aux_remove(vd, &spa_spare_avl);
759	vd->vdev_isspare = B_FALSE;
760	mutex_exit(&spa_spare_lock);
761}
762
763boolean_t
764spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
765{
766	boolean_t found;
767
768	mutex_enter(&spa_spare_lock);
769	found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
770	mutex_exit(&spa_spare_lock);
771
772	return (found);
773}
774
775void
776spa_spare_activate(vdev_t *vd)
777{
778	mutex_enter(&spa_spare_lock);
779	ASSERT(vd->vdev_isspare);
780	spa_aux_activate(vd, &spa_spare_avl);
781	mutex_exit(&spa_spare_lock);
782}
783
784/*
785 * Level 2 ARC devices are tracked globally for the same reasons as spares.
786 * Cache devices currently only support one pool per cache device, and so
787 * for these devices the aux reference count is currently unused beyond 1.
788 */
789
790static int
791spa_l2cache_compare(const void *a, const void *b)
792{
793	return (spa_aux_compare(a, b));
794}
795
796void
797spa_l2cache_add(vdev_t *vd)
798{
799	mutex_enter(&spa_l2cache_lock);
800	ASSERT(!vd->vdev_isl2cache);
801	spa_aux_add(vd, &spa_l2cache_avl);
802	vd->vdev_isl2cache = B_TRUE;
803	mutex_exit(&spa_l2cache_lock);
804}
805
806void
807spa_l2cache_remove(vdev_t *vd)
808{
809	mutex_enter(&spa_l2cache_lock);
810	ASSERT(vd->vdev_isl2cache);
811	spa_aux_remove(vd, &spa_l2cache_avl);
812	vd->vdev_isl2cache = B_FALSE;
813	mutex_exit(&spa_l2cache_lock);
814}
815
816boolean_t
817spa_l2cache_exists(uint64_t guid, uint64_t *pool)
818{
819	boolean_t found;
820
821	mutex_enter(&spa_l2cache_lock);
822	found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
823	mutex_exit(&spa_l2cache_lock);
824
825	return (found);
826}
827
828void
829spa_l2cache_activate(vdev_t *vd)
830{
831	mutex_enter(&spa_l2cache_lock);
832	ASSERT(vd->vdev_isl2cache);
833	spa_aux_activate(vd, &spa_l2cache_avl);
834	mutex_exit(&spa_l2cache_lock);
835}
836
837/*
838 * ==========================================================================
839 * SPA vdev locking
840 * ==========================================================================
841 */
842
843/*
844 * Lock the given spa_t for the purpose of adding or removing a vdev.
845 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
846 * It returns the next transaction group for the spa_t.
847 */
848uint64_t
849spa_vdev_enter(spa_t *spa)
850{
851	mutex_enter(&spa->spa_vdev_top_lock);
852	mutex_enter(&spa_namespace_lock);
853	return (spa_vdev_config_enter(spa));
854}
855
856/*
857 * Internal implementation for spa_vdev_enter().  Used when a vdev
858 * operation requires multiple syncs (i.e. removing a device) while
859 * keeping the spa_namespace_lock held.
860 */
861uint64_t
862spa_vdev_config_enter(spa_t *spa)
863{
864	ASSERT(MUTEX_HELD(&spa_namespace_lock));
865
866	spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
867
868	return (spa_last_synced_txg(spa) + 1);
869}
870
871/*
872 * Used in combination with spa_vdev_config_enter() to allow the syncing
873 * of multiple transactions without releasing the spa_namespace_lock.
874 */
875void
876spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
877{
878	ASSERT(MUTEX_HELD(&spa_namespace_lock));
879
880	int config_changed = B_FALSE;
881
882	ASSERT(txg > spa_last_synced_txg(spa));
883
884	spa->spa_pending_vdev = NULL;
885
886	/*
887	 * Reassess the DTLs.
888	 */
889	vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
890
891	/*
892	 * If the config changed, notify the scrub thread that it must restart.
893	 */
894	if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
895		dsl_pool_scrub_restart(spa->spa_dsl_pool);
896		config_changed = B_TRUE;
897		spa->spa_config_generation++;
898	}
899
900	/*
901	 * Verify the metaslab classes.
902	 */
903	ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
904	ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
905
906	spa_config_exit(spa, SCL_ALL, spa);
907
908	/*
909	 * Panic the system if the specified tag requires it.  This
910	 * is useful for ensuring that configurations are updated
911	 * transactionally.
912	 */
913	if (zio_injection_enabled)
914		zio_handle_panic_injection(spa, tag, 0);
915
916	/*
917	 * Note: this txg_wait_synced() is important because it ensures
918	 * that there won't be more than one config change per txg.
919	 * This allows us to use the txg as the generation number.
920	 */
921	if (error == 0)
922		txg_wait_synced(spa->spa_dsl_pool, txg);
923
924	if (vd != NULL) {
925		ASSERT(!vd->vdev_detached || vd->vdev_dtl_smo.smo_object == 0);
926		spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
927		vdev_free(vd);
928		spa_config_exit(spa, SCL_ALL, spa);
929	}
930
931	/*
932	 * If the config changed, update the config cache.
933	 */
934	if (config_changed)
935		spa_config_sync(spa, B_FALSE, B_TRUE);
936}
937
938/*
939 * Unlock the spa_t after adding or removing a vdev.  Besides undoing the
940 * locking of spa_vdev_enter(), we also want make sure the transactions have
941 * synced to disk, and then update the global configuration cache with the new
942 * information.
943 */
944int
945spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
946{
947	spa_vdev_config_exit(spa, vd, txg, error, FTAG);
948	mutex_exit(&spa_namespace_lock);
949	mutex_exit(&spa->spa_vdev_top_lock);
950
951	return (error);
952}
953
954/*
955 * Lock the given spa_t for the purpose of changing vdev state.
956 */
957void
958spa_vdev_state_enter(spa_t *spa, int oplocks)
959{
960	int locks = SCL_STATE_ALL | oplocks;
961
962	spa_config_enter(spa, locks, spa, RW_WRITER);
963	spa->spa_vdev_locks = locks;
964}
965
966int
967spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
968{
969	if (vd != NULL || error == 0)
970		vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
971		    0, 0, B_FALSE);
972
973	if (vd != NULL) {
974		vdev_state_dirty(vd->vdev_top);
975		spa->spa_config_generation++;
976	}
977
978	ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
979	spa_config_exit(spa, spa->spa_vdev_locks, spa);
980
981	/*
982	 * If anything changed, wait for it to sync.  This ensures that,
983	 * from the system administrator's perspective, zpool(1M) commands
984	 * are synchronous.  This is important for things like zpool offline:
985	 * when the command completes, you expect no further I/O from ZFS.
986	 */
987	if (vd != NULL)
988		txg_wait_synced(spa->spa_dsl_pool, 0);
989
990	return (error);
991}
992
993/*
994 * ==========================================================================
995 * Miscellaneous functions
996 * ==========================================================================
997 */
998
999/*
1000 * Rename a spa_t.
1001 */
1002int
1003spa_rename(const char *name, const char *newname)
1004{
1005	spa_t *spa;
1006	int err;
1007
1008	/*
1009	 * Lookup the spa_t and grab the config lock for writing.  We need to
1010	 * actually open the pool so that we can sync out the necessary labels.
1011	 * It's OK to call spa_open() with the namespace lock held because we
1012	 * allow recursive calls for other reasons.
1013	 */
1014	mutex_enter(&spa_namespace_lock);
1015	if ((err = spa_open(name, &spa, FTAG)) != 0) {
1016		mutex_exit(&spa_namespace_lock);
1017		return (err);
1018	}
1019
1020	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1021
1022	avl_remove(&spa_namespace_avl, spa);
1023	(void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1024	avl_add(&spa_namespace_avl, spa);
1025
1026	/*
1027	 * Sync all labels to disk with the new names by marking the root vdev
1028	 * dirty and waiting for it to sync.  It will pick up the new pool name
1029	 * during the sync.
1030	 */
1031	vdev_config_dirty(spa->spa_root_vdev);
1032
1033	spa_config_exit(spa, SCL_ALL, FTAG);
1034
1035	txg_wait_synced(spa->spa_dsl_pool, 0);
1036
1037	/*
1038	 * Sync the updated config cache.
1039	 */
1040	spa_config_sync(spa, B_FALSE, B_TRUE);
1041
1042	spa_close(spa, FTAG);
1043
1044	mutex_exit(&spa_namespace_lock);
1045
1046	return (0);
1047}
1048
1049
1050/*
1051 * Determine whether a pool with given pool_guid exists.  If device_guid is
1052 * non-zero, determine whether the pool exists *and* contains a device with the
1053 * specified device_guid.
1054 */
1055boolean_t
1056spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1057{
1058	spa_t *spa;
1059	avl_tree_t *t = &spa_namespace_avl;
1060
1061	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1062
1063	for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1064		if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1065			continue;
1066		if (spa->spa_root_vdev == NULL)
1067			continue;
1068		if (spa_guid(spa) == pool_guid) {
1069			if (device_guid == 0)
1070				break;
1071
1072			if (vdev_lookup_by_guid(spa->spa_root_vdev,
1073			    device_guid) != NULL)
1074				break;
1075
1076			/*
1077			 * Check any devices we may be in the process of adding.
1078			 */
1079			if (spa->spa_pending_vdev) {
1080				if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1081				    device_guid) != NULL)
1082					break;
1083			}
1084		}
1085	}
1086
1087	return (spa != NULL);
1088}
1089
1090char *
1091spa_strdup(const char *s)
1092{
1093	size_t len;
1094	char *new;
1095
1096	len = strlen(s);
1097	new = kmem_alloc(len + 1, KM_SLEEP);
1098	bcopy(s, new, len);
1099	new[len] = '\0';
1100
1101	return (new);
1102}
1103
1104void
1105spa_strfree(char *s)
1106{
1107	kmem_free(s, strlen(s) + 1);
1108}
1109
1110uint64_t
1111spa_get_random(uint64_t range)
1112{
1113	uint64_t r;
1114
1115	ASSERT(range != 0);
1116
1117	(void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1118
1119	return (r % range);
1120}
1121
1122uint64_t
1123spa_generate_guid(spa_t *spa)
1124{
1125	uint64_t guid = spa_get_random(-1ULL);
1126
1127	if (spa != NULL) {
1128		while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1129			guid = spa_get_random(-1ULL);
1130	} else {
1131		while (guid == 0 || spa_guid_exists(guid, 0))
1132			guid = spa_get_random(-1ULL);
1133	}
1134
1135	return (guid);
1136}
1137
1138void
1139sprintf_blkptr(char *buf, const blkptr_t *bp)
1140{
1141	char *type = dmu_ot[BP_GET_TYPE(bp)].ot_name;
1142	char *checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1143	char *compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1144
1145	SPRINTF_BLKPTR(snprintf, ' ', buf, bp, type, checksum, compress);
1146}
1147
1148void
1149spa_freeze(spa_t *spa)
1150{
1151	uint64_t freeze_txg = 0;
1152
1153	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1154	if (spa->spa_freeze_txg == UINT64_MAX) {
1155		freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1156		spa->spa_freeze_txg = freeze_txg;
1157	}
1158	spa_config_exit(spa, SCL_ALL, FTAG);
1159	if (freeze_txg != 0)
1160		txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1161}
1162
1163void
1164zfs_panic_recover(const char *fmt, ...)
1165{
1166	va_list adx;
1167
1168	va_start(adx, fmt);
1169	vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1170	va_end(adx);
1171}
1172
1173/*
1174 * ==========================================================================
1175 * Accessor functions
1176 * ==========================================================================
1177 */
1178
1179boolean_t
1180spa_shutting_down(spa_t *spa)
1181{
1182	return (spa->spa_async_suspended);
1183}
1184
1185dsl_pool_t *
1186spa_get_dsl(spa_t *spa)
1187{
1188	return (spa->spa_dsl_pool);
1189}
1190
1191blkptr_t *
1192spa_get_rootblkptr(spa_t *spa)
1193{
1194	return (&spa->spa_ubsync.ub_rootbp);
1195}
1196
1197void
1198spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1199{
1200	spa->spa_uberblock.ub_rootbp = *bp;
1201}
1202
1203void
1204spa_altroot(spa_t *spa, char *buf, size_t buflen)
1205{
1206	if (spa->spa_root == NULL)
1207		buf[0] = '\0';
1208	else
1209		(void) strncpy(buf, spa->spa_root, buflen);
1210}
1211
1212int
1213spa_sync_pass(spa_t *spa)
1214{
1215	return (spa->spa_sync_pass);
1216}
1217
1218char *
1219spa_name(spa_t *spa)
1220{
1221	return (spa->spa_name);
1222}
1223
1224uint64_t
1225spa_guid(spa_t *spa)
1226{
1227	/*
1228	 * If we fail to parse the config during spa_load(), we can go through
1229	 * the error path (which posts an ereport) and end up here with no root
1230	 * vdev.  We stash the original pool guid in 'spa_load_guid' to handle
1231	 * this case.
1232	 */
1233	if (spa->spa_root_vdev != NULL)
1234		return (spa->spa_root_vdev->vdev_guid);
1235	else
1236		return (spa->spa_load_guid);
1237}
1238
1239uint64_t
1240spa_last_synced_txg(spa_t *spa)
1241{
1242	return (spa->spa_ubsync.ub_txg);
1243}
1244
1245uint64_t
1246spa_first_txg(spa_t *spa)
1247{
1248	return (spa->spa_first_txg);
1249}
1250
1251uint64_t
1252spa_syncing_txg(spa_t *spa)
1253{
1254	return (spa->spa_syncing_txg);
1255}
1256
1257pool_state_t
1258spa_state(spa_t *spa)
1259{
1260	return (spa->spa_state);
1261}
1262
1263spa_load_state_t
1264spa_load_state(spa_t *spa)
1265{
1266	return (spa->spa_load_state);
1267}
1268
1269uint64_t
1270spa_freeze_txg(spa_t *spa)
1271{
1272	return (spa->spa_freeze_txg);
1273}
1274
1275/* ARGSUSED */
1276uint64_t
1277spa_get_asize(spa_t *spa, uint64_t lsize)
1278{
1279	/*
1280	 * The worst case is single-sector max-parity RAID-Z blocks, in which
1281	 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
1282	 * times the size; so just assume that.  Add to this the fact that
1283	 * we can have up to 3 DVAs per bp, and one more factor of 2 because
1284	 * the block may be dittoed with up to 3 DVAs by ddt_sync().
1285	 */
1286	return (lsize * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2);
1287}
1288
1289uint64_t
1290spa_get_dspace(spa_t *spa)
1291{
1292	return (spa->spa_dspace);
1293}
1294
1295void
1296spa_update_dspace(spa_t *spa)
1297{
1298	spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1299	    ddt_get_dedup_dspace(spa);
1300}
1301
1302/*
1303 * Return the failure mode that has been set to this pool. The default
1304 * behavior will be to block all I/Os when a complete failure occurs.
1305 */
1306uint8_t
1307spa_get_failmode(spa_t *spa)
1308{
1309	return (spa->spa_failmode);
1310}
1311
1312boolean_t
1313spa_suspended(spa_t *spa)
1314{
1315	return (spa->spa_suspended);
1316}
1317
1318uint64_t
1319spa_version(spa_t *spa)
1320{
1321	return (spa->spa_ubsync.ub_version);
1322}
1323
1324boolean_t
1325spa_deflate(spa_t *spa)
1326{
1327	return (spa->spa_deflate);
1328}
1329
1330metaslab_class_t *
1331spa_normal_class(spa_t *spa)
1332{
1333	return (spa->spa_normal_class);
1334}
1335
1336metaslab_class_t *
1337spa_log_class(spa_t *spa)
1338{
1339	return (spa->spa_log_class);
1340}
1341
1342int
1343spa_max_replication(spa_t *spa)
1344{
1345	/*
1346	 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1347	 * handle BPs with more than one DVA allocated.  Set our max
1348	 * replication level accordingly.
1349	 */
1350	if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1351		return (1);
1352	return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1353}
1354
1355uint64_t
1356dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1357{
1358	uint64_t asize = DVA_GET_ASIZE(dva);
1359	uint64_t dsize = asize;
1360
1361	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1362
1363	if (asize != 0 && spa->spa_deflate) {
1364		vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1365		dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1366	}
1367
1368	return (dsize);
1369}
1370
1371uint64_t
1372bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1373{
1374	uint64_t dsize = 0;
1375
1376	for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1377		dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1378
1379	return (dsize);
1380}
1381
1382uint64_t
1383bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1384{
1385	uint64_t dsize = 0;
1386
1387	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1388
1389	for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1390		dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1391
1392	spa_config_exit(spa, SCL_VDEV, FTAG);
1393
1394	return (dsize);
1395}
1396
1397/*
1398 * ==========================================================================
1399 * Initialization and Termination
1400 * ==========================================================================
1401 */
1402
1403static int
1404spa_name_compare(const void *a1, const void *a2)
1405{
1406	const spa_t *s1 = a1;
1407	const spa_t *s2 = a2;
1408	int s;
1409
1410	s = strcmp(s1->spa_name, s2->spa_name);
1411	if (s > 0)
1412		return (1);
1413	if (s < 0)
1414		return (-1);
1415	return (0);
1416}
1417
1418int
1419spa_busy(void)
1420{
1421	return (spa_active_count);
1422}
1423
1424void
1425spa_boot_init()
1426{
1427	spa_config_load();
1428}
1429
1430void
1431spa_init(int mode)
1432{
1433	mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1434	mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1435	mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1436	cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1437
1438	avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1439	    offsetof(spa_t, spa_avl));
1440
1441	avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1442	    offsetof(spa_aux_t, aux_avl));
1443
1444	avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1445	    offsetof(spa_aux_t, aux_avl));
1446
1447	spa_mode_global = mode;
1448
1449	refcount_init();
1450	unique_init();
1451	zio_init();
1452	dmu_init();
1453	zil_init();
1454	vdev_cache_stat_init();
1455	zfs_prop_init();
1456	zpool_prop_init();
1457	spa_config_load();
1458	l2arc_start();
1459}
1460
1461void
1462spa_fini(void)
1463{
1464	l2arc_stop();
1465
1466	spa_evict_all();
1467
1468	vdev_cache_stat_fini();
1469	zil_fini();
1470	dmu_fini();
1471	zio_fini();
1472	unique_fini();
1473	refcount_fini();
1474
1475	avl_destroy(&spa_namespace_avl);
1476	avl_destroy(&spa_spare_avl);
1477	avl_destroy(&spa_l2cache_avl);
1478
1479	cv_destroy(&spa_namespace_cv);
1480	mutex_destroy(&spa_namespace_lock);
1481	mutex_destroy(&spa_spare_lock);
1482	mutex_destroy(&spa_l2cache_lock);
1483}
1484
1485/*
1486 * Return whether this pool has slogs. No locking needed.
1487 * It's not a problem if the wrong answer is returned as it's only for
1488 * performance and not correctness
1489 */
1490boolean_t
1491spa_has_slogs(spa_t *spa)
1492{
1493	return (spa->spa_log_class->mc_rotor != NULL);
1494}
1495
1496spa_log_state_t
1497spa_get_log_state(spa_t *spa)
1498{
1499	return (spa->spa_log_state);
1500}
1501
1502void
1503spa_set_log_state(spa_t *spa, spa_log_state_t state)
1504{
1505	spa->spa_log_state = state;
1506}
1507
1508boolean_t
1509spa_is_root(spa_t *spa)
1510{
1511	return (spa->spa_is_root);
1512}
1513
1514boolean_t
1515spa_writeable(spa_t *spa)
1516{
1517	return (!!(spa->spa_mode & FWRITE));
1518}
1519
1520int
1521spa_mode(spa_t *spa)
1522{
1523	return (spa->spa_mode);
1524}
1525
1526uint64_t
1527spa_bootfs(spa_t *spa)
1528{
1529	return (spa->spa_bootfs);
1530}
1531
1532uint64_t
1533spa_delegation(spa_t *spa)
1534{
1535	return (spa->spa_delegation);
1536}
1537
1538objset_t *
1539spa_meta_objset(spa_t *spa)
1540{
1541	return (spa->spa_meta_objset);
1542}
1543
1544enum zio_checksum
1545spa_dedup_checksum(spa_t *spa)
1546{
1547	return (spa->spa_dedup_checksum);
1548}
1549