spa_misc.c revision 35a5a3587fd94b666239c157d3722745250ccbd7
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/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 = 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	avl_remove(&spa_namespace_avl, spa);
502	cv_broadcast(&spa_namespace_cv);
503
504	if (spa->spa_root) {
505		spa_strfree(spa->spa_root);
506		spa_active_count--;
507	}
508
509	while ((dp = list_head(&spa->spa_config_list)) != NULL) {
510		list_remove(&spa->spa_config_list, dp);
511		if (dp->scd_path != NULL)
512			spa_strfree(dp->scd_path);
513		kmem_free(dp, sizeof (spa_config_dirent_t));
514	}
515
516	list_destroy(&spa->spa_config_list);
517
518	spa_config_set(spa, NULL);
519
520	refcount_destroy(&spa->spa_refcount);
521
522	spa_config_lock_destroy(spa);
523
524	for (int t = 0; t < TXG_SIZE; t++)
525		bplist_fini(&spa->spa_free_bplist[t]);
526	bplist_fini(&spa->spa_deferred_bplist);
527
528	cv_destroy(&spa->spa_async_cv);
529	cv_destroy(&spa->spa_proc_cv);
530	cv_destroy(&spa->spa_scrub_io_cv);
531	cv_destroy(&spa->spa_suspend_cv);
532
533	mutex_destroy(&spa->spa_async_lock);
534	mutex_destroy(&spa->spa_errlist_lock);
535	mutex_destroy(&spa->spa_errlog_lock);
536	mutex_destroy(&spa->spa_history_lock);
537	mutex_destroy(&spa->spa_proc_lock);
538	mutex_destroy(&spa->spa_props_lock);
539	mutex_destroy(&spa->spa_scrub_lock);
540	mutex_destroy(&spa->spa_suspend_lock);
541	mutex_destroy(&spa->spa_vdev_top_lock);
542
543	kmem_free(spa, sizeof (spa_t));
544}
545
546/*
547 * Given a pool, return the next pool in the namespace, or NULL if there is
548 * none.  If 'prev' is NULL, return the first pool.
549 */
550spa_t *
551spa_next(spa_t *prev)
552{
553	ASSERT(MUTEX_HELD(&spa_namespace_lock));
554
555	if (prev)
556		return (AVL_NEXT(&spa_namespace_avl, prev));
557	else
558		return (avl_first(&spa_namespace_avl));
559}
560
561/*
562 * ==========================================================================
563 * SPA refcount functions
564 * ==========================================================================
565 */
566
567/*
568 * Add a reference to the given spa_t.  Must have at least one reference, or
569 * have the namespace lock held.
570 */
571void
572spa_open_ref(spa_t *spa, void *tag)
573{
574	ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
575	    MUTEX_HELD(&spa_namespace_lock));
576	(void) refcount_add(&spa->spa_refcount, tag);
577}
578
579/*
580 * Remove a reference to the given spa_t.  Must have at least one reference, or
581 * have the namespace lock held.
582 */
583void
584spa_close(spa_t *spa, void *tag)
585{
586	ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
587	    MUTEX_HELD(&spa_namespace_lock));
588	(void) refcount_remove(&spa->spa_refcount, tag);
589}
590
591/*
592 * Check to see if the spa refcount is zero.  Must be called with
593 * spa_namespace_lock held.  We really compare against spa_minref, which is the
594 * number of references acquired when opening a pool
595 */
596boolean_t
597spa_refcount_zero(spa_t *spa)
598{
599	ASSERT(MUTEX_HELD(&spa_namespace_lock));
600
601	return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
602}
603
604/*
605 * ==========================================================================
606 * SPA spare and l2cache tracking
607 * ==========================================================================
608 */
609
610/*
611 * Hot spares and cache devices are tracked using the same code below,
612 * for 'auxiliary' devices.
613 */
614
615typedef struct spa_aux {
616	uint64_t	aux_guid;
617	uint64_t	aux_pool;
618	avl_node_t	aux_avl;
619	int		aux_count;
620} spa_aux_t;
621
622static int
623spa_aux_compare(const void *a, const void *b)
624{
625	const spa_aux_t *sa = a;
626	const spa_aux_t *sb = b;
627
628	if (sa->aux_guid < sb->aux_guid)
629		return (-1);
630	else if (sa->aux_guid > sb->aux_guid)
631		return (1);
632	else
633		return (0);
634}
635
636void
637spa_aux_add(vdev_t *vd, avl_tree_t *avl)
638{
639	avl_index_t where;
640	spa_aux_t search;
641	spa_aux_t *aux;
642
643	search.aux_guid = vd->vdev_guid;
644	if ((aux = avl_find(avl, &search, &where)) != NULL) {
645		aux->aux_count++;
646	} else {
647		aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
648		aux->aux_guid = vd->vdev_guid;
649		aux->aux_count = 1;
650		avl_insert(avl, aux, where);
651	}
652}
653
654void
655spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
656{
657	spa_aux_t search;
658	spa_aux_t *aux;
659	avl_index_t where;
660
661	search.aux_guid = vd->vdev_guid;
662	aux = avl_find(avl, &search, &where);
663
664	ASSERT(aux != NULL);
665
666	if (--aux->aux_count == 0) {
667		avl_remove(avl, aux);
668		kmem_free(aux, sizeof (spa_aux_t));
669	} else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
670		aux->aux_pool = 0ULL;
671	}
672}
673
674boolean_t
675spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
676{
677	spa_aux_t search, *found;
678
679	search.aux_guid = guid;
680	found = avl_find(avl, &search, NULL);
681
682	if (pool) {
683		if (found)
684			*pool = found->aux_pool;
685		else
686			*pool = 0ULL;
687	}
688
689	if (refcnt) {
690		if (found)
691			*refcnt = found->aux_count;
692		else
693			*refcnt = 0;
694	}
695
696	return (found != NULL);
697}
698
699void
700spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
701{
702	spa_aux_t search, *found;
703	avl_index_t where;
704
705	search.aux_guid = vd->vdev_guid;
706	found = avl_find(avl, &search, &where);
707	ASSERT(found != NULL);
708	ASSERT(found->aux_pool == 0ULL);
709
710	found->aux_pool = spa_guid(vd->vdev_spa);
711}
712
713/*
714 * Spares are tracked globally due to the following constraints:
715 *
716 * 	- A spare may be part of multiple pools.
717 * 	- A spare may be added to a pool even if it's actively in use within
718 *	  another pool.
719 * 	- A spare in use in any pool can only be the source of a replacement if
720 *	  the target is a spare in the same pool.
721 *
722 * We keep track of all spares on the system through the use of a reference
723 * counted AVL tree.  When a vdev is added as a spare, or used as a replacement
724 * spare, then we bump the reference count in the AVL tree.  In addition, we set
725 * the 'vdev_isspare' member to indicate that the device is a spare (active or
726 * inactive).  When a spare is made active (used to replace a device in the
727 * pool), we also keep track of which pool its been made a part of.
728 *
729 * The 'spa_spare_lock' protects the AVL tree.  These functions are normally
730 * called under the spa_namespace lock as part of vdev reconfiguration.  The
731 * separate spare lock exists for the status query path, which does not need to
732 * be completely consistent with respect to other vdev configuration changes.
733 */
734
735static int
736spa_spare_compare(const void *a, const void *b)
737{
738	return (spa_aux_compare(a, b));
739}
740
741void
742spa_spare_add(vdev_t *vd)
743{
744	mutex_enter(&spa_spare_lock);
745	ASSERT(!vd->vdev_isspare);
746	spa_aux_add(vd, &spa_spare_avl);
747	vd->vdev_isspare = B_TRUE;
748	mutex_exit(&spa_spare_lock);
749}
750
751void
752spa_spare_remove(vdev_t *vd)
753{
754	mutex_enter(&spa_spare_lock);
755	ASSERT(vd->vdev_isspare);
756	spa_aux_remove(vd, &spa_spare_avl);
757	vd->vdev_isspare = B_FALSE;
758	mutex_exit(&spa_spare_lock);
759}
760
761boolean_t
762spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
763{
764	boolean_t found;
765
766	mutex_enter(&spa_spare_lock);
767	found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
768	mutex_exit(&spa_spare_lock);
769
770	return (found);
771}
772
773void
774spa_spare_activate(vdev_t *vd)
775{
776	mutex_enter(&spa_spare_lock);
777	ASSERT(vd->vdev_isspare);
778	spa_aux_activate(vd, &spa_spare_avl);
779	mutex_exit(&spa_spare_lock);
780}
781
782/*
783 * Level 2 ARC devices are tracked globally for the same reasons as spares.
784 * Cache devices currently only support one pool per cache device, and so
785 * for these devices the aux reference count is currently unused beyond 1.
786 */
787
788static int
789spa_l2cache_compare(const void *a, const void *b)
790{
791	return (spa_aux_compare(a, b));
792}
793
794void
795spa_l2cache_add(vdev_t *vd)
796{
797	mutex_enter(&spa_l2cache_lock);
798	ASSERT(!vd->vdev_isl2cache);
799	spa_aux_add(vd, &spa_l2cache_avl);
800	vd->vdev_isl2cache = B_TRUE;
801	mutex_exit(&spa_l2cache_lock);
802}
803
804void
805spa_l2cache_remove(vdev_t *vd)
806{
807	mutex_enter(&spa_l2cache_lock);
808	ASSERT(vd->vdev_isl2cache);
809	spa_aux_remove(vd, &spa_l2cache_avl);
810	vd->vdev_isl2cache = B_FALSE;
811	mutex_exit(&spa_l2cache_lock);
812}
813
814boolean_t
815spa_l2cache_exists(uint64_t guid, uint64_t *pool)
816{
817	boolean_t found;
818
819	mutex_enter(&spa_l2cache_lock);
820	found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
821	mutex_exit(&spa_l2cache_lock);
822
823	return (found);
824}
825
826void
827spa_l2cache_activate(vdev_t *vd)
828{
829	mutex_enter(&spa_l2cache_lock);
830	ASSERT(vd->vdev_isl2cache);
831	spa_aux_activate(vd, &spa_l2cache_avl);
832	mutex_exit(&spa_l2cache_lock);
833}
834
835/*
836 * ==========================================================================
837 * SPA vdev locking
838 * ==========================================================================
839 */
840
841/*
842 * Lock the given spa_t for the purpose of adding or removing a vdev.
843 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
844 * It returns the next transaction group for the spa_t.
845 */
846uint64_t
847spa_vdev_enter(spa_t *spa)
848{
849	mutex_enter(&spa->spa_vdev_top_lock);
850	mutex_enter(&spa_namespace_lock);
851	return (spa_vdev_config_enter(spa));
852}
853
854/*
855 * Internal implementation for spa_vdev_enter().  Used when a vdev
856 * operation requires multiple syncs (i.e. removing a device) while
857 * keeping the spa_namespace_lock held.
858 */
859uint64_t
860spa_vdev_config_enter(spa_t *spa)
861{
862	ASSERT(MUTEX_HELD(&spa_namespace_lock));
863
864	spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
865
866	return (spa_last_synced_txg(spa) + 1);
867}
868
869/*
870 * Used in combination with spa_vdev_config_enter() to allow the syncing
871 * of multiple transactions without releasing the spa_namespace_lock.
872 */
873void
874spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
875{
876	ASSERT(MUTEX_HELD(&spa_namespace_lock));
877
878	int config_changed = B_FALSE;
879
880	ASSERT(txg > spa_last_synced_txg(spa));
881
882	spa->spa_pending_vdev = NULL;
883
884	/*
885	 * Reassess the DTLs.
886	 */
887	vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
888
889	/*
890	 * If the config changed, notify the scrub thread that it must restart.
891	 */
892	if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
893		dsl_pool_scrub_restart(spa->spa_dsl_pool);
894		config_changed = B_TRUE;
895		spa->spa_config_generation++;
896	}
897
898	/*
899	 * Verify the metaslab classes.
900	 */
901	ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
902	ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
903
904	spa_config_exit(spa, SCL_ALL, spa);
905
906	/*
907	 * Panic the system if the specified tag requires it.  This
908	 * is useful for ensuring that configurations are updated
909	 * transactionally.
910	 */
911	if (zio_injection_enabled)
912		zio_handle_panic_injection(spa, tag);
913
914	/*
915	 * Note: this txg_wait_synced() is important because it ensures
916	 * that there won't be more than one config change per txg.
917	 * This allows us to use the txg as the generation number.
918	 */
919	if (error == 0)
920		txg_wait_synced(spa->spa_dsl_pool, txg);
921
922	if (vd != NULL) {
923		ASSERT(!vd->vdev_detached || vd->vdev_dtl_smo.smo_object == 0);
924		spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
925		vdev_free(vd);
926		spa_config_exit(spa, SCL_ALL, spa);
927	}
928
929	/*
930	 * If the config changed, update the config cache.
931	 */
932	if (config_changed)
933		spa_config_sync(spa, B_FALSE, B_TRUE);
934}
935
936/*
937 * Unlock the spa_t after adding or removing a vdev.  Besides undoing the
938 * locking of spa_vdev_enter(), we also want make sure the transactions have
939 * synced to disk, and then update the global configuration cache with the new
940 * information.
941 */
942int
943spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
944{
945	spa_vdev_config_exit(spa, vd, txg, error, FTAG);
946	mutex_exit(&spa_namespace_lock);
947	mutex_exit(&spa->spa_vdev_top_lock);
948
949	return (error);
950}
951
952/*
953 * Lock the given spa_t for the purpose of changing vdev state.
954 */
955void
956spa_vdev_state_enter(spa_t *spa, int oplocks)
957{
958	int locks = SCL_STATE_ALL | oplocks;
959
960	spa_config_enter(spa, locks, spa, RW_WRITER);
961	spa->spa_vdev_locks = locks;
962}
963
964int
965spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
966{
967	if (vd != NULL || error == 0)
968		vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
969		    0, 0, B_FALSE);
970
971	if (vd != NULL) {
972		vdev_state_dirty(vd->vdev_top);
973		spa->spa_config_generation++;
974	}
975
976	ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
977	spa_config_exit(spa, spa->spa_vdev_locks, spa);
978
979	/*
980	 * If anything changed, wait for it to sync.  This ensures that,
981	 * from the system administrator's perspective, zpool(1M) commands
982	 * are synchronous.  This is important for things like zpool offline:
983	 * when the command completes, you expect no further I/O from ZFS.
984	 */
985	if (vd != NULL)
986		txg_wait_synced(spa->spa_dsl_pool, 0);
987
988	return (error);
989}
990
991/*
992 * ==========================================================================
993 * Miscellaneous functions
994 * ==========================================================================
995 */
996
997/*
998 * Rename a spa_t.
999 */
1000int
1001spa_rename(const char *name, const char *newname)
1002{
1003	spa_t *spa;
1004	int err;
1005
1006	/*
1007	 * Lookup the spa_t and grab the config lock for writing.  We need to
1008	 * actually open the pool so that we can sync out the necessary labels.
1009	 * It's OK to call spa_open() with the namespace lock held because we
1010	 * allow recursive calls for other reasons.
1011	 */
1012	mutex_enter(&spa_namespace_lock);
1013	if ((err = spa_open(name, &spa, FTAG)) != 0) {
1014		mutex_exit(&spa_namespace_lock);
1015		return (err);
1016	}
1017
1018	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1019
1020	avl_remove(&spa_namespace_avl, spa);
1021	(void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1022	avl_add(&spa_namespace_avl, spa);
1023
1024	/*
1025	 * Sync all labels to disk with the new names by marking the root vdev
1026	 * dirty and waiting for it to sync.  It will pick up the new pool name
1027	 * during the sync.
1028	 */
1029	vdev_config_dirty(spa->spa_root_vdev);
1030
1031	spa_config_exit(spa, SCL_ALL, FTAG);
1032
1033	txg_wait_synced(spa->spa_dsl_pool, 0);
1034
1035	/*
1036	 * Sync the updated config cache.
1037	 */
1038	spa_config_sync(spa, B_FALSE, B_TRUE);
1039
1040	spa_close(spa, FTAG);
1041
1042	mutex_exit(&spa_namespace_lock);
1043
1044	return (0);
1045}
1046
1047
1048/*
1049 * Determine whether a pool with given pool_guid exists.  If device_guid is
1050 * non-zero, determine whether the pool exists *and* contains a device with the
1051 * specified device_guid.
1052 */
1053boolean_t
1054spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1055{
1056	spa_t *spa;
1057	avl_tree_t *t = &spa_namespace_avl;
1058
1059	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1060
1061	for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1062		if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1063			continue;
1064		if (spa->spa_root_vdev == NULL)
1065			continue;
1066		if (spa_guid(spa) == pool_guid) {
1067			if (device_guid == 0)
1068				break;
1069
1070			if (vdev_lookup_by_guid(spa->spa_root_vdev,
1071			    device_guid) != NULL)
1072				break;
1073
1074			/*
1075			 * Check any devices we may be in the process of adding.
1076			 */
1077			if (spa->spa_pending_vdev) {
1078				if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1079				    device_guid) != NULL)
1080					break;
1081			}
1082		}
1083	}
1084
1085	return (spa != NULL);
1086}
1087
1088char *
1089spa_strdup(const char *s)
1090{
1091	size_t len;
1092	char *new;
1093
1094	len = strlen(s);
1095	new = kmem_alloc(len + 1, KM_SLEEP);
1096	bcopy(s, new, len);
1097	new[len] = '\0';
1098
1099	return (new);
1100}
1101
1102void
1103spa_strfree(char *s)
1104{
1105	kmem_free(s, strlen(s) + 1);
1106}
1107
1108uint64_t
1109spa_get_random(uint64_t range)
1110{
1111	uint64_t r;
1112
1113	ASSERT(range != 0);
1114
1115	(void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1116
1117	return (r % range);
1118}
1119
1120void
1121sprintf_blkptr(char *buf, const blkptr_t *bp)
1122{
1123	char *type = dmu_ot[BP_GET_TYPE(bp)].ot_name;
1124	char *checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1125	char *compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1126
1127	SPRINTF_BLKPTR(snprintf, ' ', buf, bp, type, checksum, compress);
1128}
1129
1130void
1131spa_freeze(spa_t *spa)
1132{
1133	uint64_t freeze_txg = 0;
1134
1135	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1136	if (spa->spa_freeze_txg == UINT64_MAX) {
1137		freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1138		spa->spa_freeze_txg = freeze_txg;
1139	}
1140	spa_config_exit(spa, SCL_ALL, FTAG);
1141	if (freeze_txg != 0)
1142		txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1143}
1144
1145void
1146zfs_panic_recover(const char *fmt, ...)
1147{
1148	va_list adx;
1149
1150	va_start(adx, fmt);
1151	vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1152	va_end(adx);
1153}
1154
1155/*
1156 * ==========================================================================
1157 * Accessor functions
1158 * ==========================================================================
1159 */
1160
1161boolean_t
1162spa_shutting_down(spa_t *spa)
1163{
1164	return (spa->spa_async_suspended);
1165}
1166
1167dsl_pool_t *
1168spa_get_dsl(spa_t *spa)
1169{
1170	return (spa->spa_dsl_pool);
1171}
1172
1173blkptr_t *
1174spa_get_rootblkptr(spa_t *spa)
1175{
1176	return (&spa->spa_ubsync.ub_rootbp);
1177}
1178
1179void
1180spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1181{
1182	spa->spa_uberblock.ub_rootbp = *bp;
1183}
1184
1185void
1186spa_altroot(spa_t *spa, char *buf, size_t buflen)
1187{
1188	if (spa->spa_root == NULL)
1189		buf[0] = '\0';
1190	else
1191		(void) strncpy(buf, spa->spa_root, buflen);
1192}
1193
1194int
1195spa_sync_pass(spa_t *spa)
1196{
1197	return (spa->spa_sync_pass);
1198}
1199
1200char *
1201spa_name(spa_t *spa)
1202{
1203	return (spa->spa_name);
1204}
1205
1206uint64_t
1207spa_guid(spa_t *spa)
1208{
1209	/*
1210	 * If we fail to parse the config during spa_load(), we can go through
1211	 * the error path (which posts an ereport) and end up here with no root
1212	 * vdev.  We stash the original pool guid in 'spa_load_guid' to handle
1213	 * this case.
1214	 */
1215	if (spa->spa_root_vdev != NULL)
1216		return (spa->spa_root_vdev->vdev_guid);
1217	else
1218		return (spa->spa_load_guid);
1219}
1220
1221uint64_t
1222spa_last_synced_txg(spa_t *spa)
1223{
1224	return (spa->spa_ubsync.ub_txg);
1225}
1226
1227uint64_t
1228spa_first_txg(spa_t *spa)
1229{
1230	return (spa->spa_first_txg);
1231}
1232
1233uint64_t
1234spa_syncing_txg(spa_t *spa)
1235{
1236	return (spa->spa_syncing_txg);
1237}
1238
1239pool_state_t
1240spa_state(spa_t *spa)
1241{
1242	return (spa->spa_state);
1243}
1244
1245spa_load_state_t
1246spa_load_state(spa_t *spa)
1247{
1248	return (spa->spa_load_state);
1249}
1250
1251uint64_t
1252spa_freeze_txg(spa_t *spa)
1253{
1254	return (spa->spa_freeze_txg);
1255}
1256
1257/* ARGSUSED */
1258uint64_t
1259spa_get_asize(spa_t *spa, uint64_t lsize)
1260{
1261	/*
1262	 * The worst case is single-sector max-parity RAID-Z blocks, in which
1263	 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
1264	 * times the size; so just assume that.  Add to this the fact that
1265	 * we can have up to 3 DVAs per bp, and one more factor of 2 because
1266	 * the block may be dittoed with up to 3 DVAs by ddt_sync().
1267	 */
1268	return (lsize * (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2);
1269}
1270
1271uint64_t
1272spa_get_dspace(spa_t *spa)
1273{
1274	return (spa->spa_dspace);
1275}
1276
1277void
1278spa_update_dspace(spa_t *spa)
1279{
1280	spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1281	    ddt_get_dedup_dspace(spa);
1282}
1283
1284/*
1285 * Return the failure mode that has been set to this pool. The default
1286 * behavior will be to block all I/Os when a complete failure occurs.
1287 */
1288uint8_t
1289spa_get_failmode(spa_t *spa)
1290{
1291	return (spa->spa_failmode);
1292}
1293
1294boolean_t
1295spa_suspended(spa_t *spa)
1296{
1297	return (spa->spa_suspended);
1298}
1299
1300uint64_t
1301spa_version(spa_t *spa)
1302{
1303	return (spa->spa_ubsync.ub_version);
1304}
1305
1306boolean_t
1307spa_deflate(spa_t *spa)
1308{
1309	return (spa->spa_deflate);
1310}
1311
1312metaslab_class_t *
1313spa_normal_class(spa_t *spa)
1314{
1315	return (spa->spa_normal_class);
1316}
1317
1318metaslab_class_t *
1319spa_log_class(spa_t *spa)
1320{
1321	return (spa->spa_log_class);
1322}
1323
1324int
1325spa_max_replication(spa_t *spa)
1326{
1327	/*
1328	 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1329	 * handle BPs with more than one DVA allocated.  Set our max
1330	 * replication level accordingly.
1331	 */
1332	if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1333		return (1);
1334	return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1335}
1336
1337uint64_t
1338dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1339{
1340	uint64_t asize = DVA_GET_ASIZE(dva);
1341	uint64_t dsize = asize;
1342
1343	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1344
1345	if (asize != 0 && spa->spa_deflate) {
1346		vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1347		dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1348	}
1349
1350	return (dsize);
1351}
1352
1353uint64_t
1354bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1355{
1356	uint64_t dsize = 0;
1357
1358	for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1359		dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1360
1361	return (dsize);
1362}
1363
1364uint64_t
1365bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1366{
1367	uint64_t dsize = 0;
1368
1369	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1370
1371	for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1372		dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1373
1374	spa_config_exit(spa, SCL_VDEV, FTAG);
1375
1376	return (dsize);
1377}
1378
1379/*
1380 * ==========================================================================
1381 * Initialization and Termination
1382 * ==========================================================================
1383 */
1384
1385static int
1386spa_name_compare(const void *a1, const void *a2)
1387{
1388	const spa_t *s1 = a1;
1389	const spa_t *s2 = a2;
1390	int s;
1391
1392	s = strcmp(s1->spa_name, s2->spa_name);
1393	if (s > 0)
1394		return (1);
1395	if (s < 0)
1396		return (-1);
1397	return (0);
1398}
1399
1400int
1401spa_busy(void)
1402{
1403	return (spa_active_count);
1404}
1405
1406void
1407spa_boot_init()
1408{
1409	spa_config_load();
1410}
1411
1412void
1413spa_init(int mode)
1414{
1415	mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1416	mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1417	mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1418	cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1419
1420	avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1421	    offsetof(spa_t, spa_avl));
1422
1423	avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1424	    offsetof(spa_aux_t, aux_avl));
1425
1426	avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1427	    offsetof(spa_aux_t, aux_avl));
1428
1429	spa_mode_global = mode;
1430
1431	refcount_init();
1432	unique_init();
1433	zio_init();
1434	dmu_init();
1435	zil_init();
1436	vdev_cache_stat_init();
1437	zfs_prop_init();
1438	zpool_prop_init();
1439	spa_config_load();
1440	l2arc_start();
1441}
1442
1443void
1444spa_fini(void)
1445{
1446	l2arc_stop();
1447
1448	spa_evict_all();
1449
1450	vdev_cache_stat_fini();
1451	zil_fini();
1452	dmu_fini();
1453	zio_fini();
1454	unique_fini();
1455	refcount_fini();
1456
1457	avl_destroy(&spa_namespace_avl);
1458	avl_destroy(&spa_spare_avl);
1459	avl_destroy(&spa_l2cache_avl);
1460
1461	cv_destroy(&spa_namespace_cv);
1462	mutex_destroy(&spa_namespace_lock);
1463	mutex_destroy(&spa_spare_lock);
1464	mutex_destroy(&spa_l2cache_lock);
1465}
1466
1467/*
1468 * Return whether this pool has slogs. No locking needed.
1469 * It's not a problem if the wrong answer is returned as it's only for
1470 * performance and not correctness
1471 */
1472boolean_t
1473spa_has_slogs(spa_t *spa)
1474{
1475	return (spa->spa_log_class->mc_rotor != NULL);
1476}
1477
1478spa_log_state_t
1479spa_get_log_state(spa_t *spa)
1480{
1481	return (spa->spa_log_state);
1482}
1483
1484void
1485spa_set_log_state(spa_t *spa, spa_log_state_t state)
1486{
1487	spa->spa_log_state = state;
1488}
1489
1490boolean_t
1491spa_is_root(spa_t *spa)
1492{
1493	return (spa->spa_is_root);
1494}
1495
1496boolean_t
1497spa_writeable(spa_t *spa)
1498{
1499	return (!!(spa->spa_mode & FWRITE));
1500}
1501
1502int
1503spa_mode(spa_t *spa)
1504{
1505	return (spa->spa_mode);
1506}
1507
1508uint64_t
1509spa_bootfs(spa_t *spa)
1510{
1511	return (spa->spa_bootfs);
1512}
1513
1514uint64_t
1515spa_delegation(spa_t *spa)
1516{
1517	return (spa->spa_delegation);
1518}
1519
1520objset_t *
1521spa_meta_objset(spa_t *spa)
1522{
1523	return (spa->spa_meta_objset);
1524}
1525
1526enum zio_checksum
1527spa_dedup_checksum(spa_t *spa)
1528{
1529	return (spa->spa_dedup_checksum);
1530}
1531