spa_misc.c revision 78f171005391b928aaf1642b3206c534ed644332
1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
25 */
26
27#include <sys/zfs_context.h>
28#include <sys/spa_impl.h>
29#include <sys/spa_boot.h>
30#include <sys/zio.h>
31#include <sys/zio_checksum.h>
32#include <sys/zio_compress.h>
33#include <sys/dmu.h>
34#include <sys/dmu_tx.h>
35#include <sys/zap.h>
36#include <sys/zil.h>
37#include <sys/vdev_impl.h>
38#include <sys/metaslab.h>
39#include <sys/uberblock_impl.h>
40#include <sys/txg.h>
41#include <sys/avl.h>
42#include <sys/unique.h>
43#include <sys/dsl_pool.h>
44#include <sys/dsl_dir.h>
45#include <sys/dsl_prop.h>
46#include <sys/dsl_scan.h>
47#include <sys/fs/zfs.h>
48#include <sys/metaslab_impl.h>
49#include <sys/arc.h>
50#include <sys/ddt.h>
51#include "zfs_prop.h"
52#include "zfeature_common.h"
53
54/*
55 * SPA locking
56 *
57 * There are four basic locks for managing spa_t structures:
58 *
59 * spa_namespace_lock (global mutex)
60 *
61 *	This lock must be acquired to do any of the following:
62 *
63 *		- Lookup a spa_t by name
64 *		- Add or remove a spa_t from the namespace
65 *		- Increase spa_refcount from non-zero
66 *		- Check if spa_refcount is zero
67 *		- Rename a spa_t
68 *		- add/remove/attach/detach devices
69 *		- Held for the duration of create/destroy/import/export
70 *
71 *	It does not need to handle recursion.  A create or destroy may
72 *	reference objects (files or zvols) in other pools, but by
73 *	definition they must have an existing reference, and will never need
74 *	to lookup a spa_t by name.
75 *
76 * spa_refcount (per-spa refcount_t protected by mutex)
77 *
78 *	This reference count keep track of any active users of the spa_t.  The
79 *	spa_t cannot be destroyed or freed while this is non-zero.  Internally,
80 *	the refcount is never really 'zero' - opening a pool implicitly keeps
81 *	some references in the DMU.  Internally we check against spa_minref, but
82 *	present the image of a zero/non-zero value to consumers.
83 *
84 * spa_config_lock[] (per-spa array of rwlocks)
85 *
86 *	This protects the spa_t from config changes, and must be held in
87 *	the following circumstances:
88 *
89 *		- RW_READER to perform I/O to the spa
90 *		- RW_WRITER to change the vdev config
91 *
92 * The locking order is fairly straightforward:
93 *
94 *		spa_namespace_lock	->	spa_refcount
95 *
96 *	The namespace lock must be acquired to increase the refcount from 0
97 *	or to check if it is zero.
98 *
99 *		spa_refcount		->	spa_config_lock[]
100 *
101 *	There must be at least one valid reference on the spa_t to acquire
102 *	the config lock.
103 *
104 *		spa_namespace_lock	->	spa_config_lock[]
105 *
106 *	The namespace lock must always be taken before the config lock.
107 *
108 *
109 * The spa_namespace_lock can be acquired directly and is globally visible.
110 *
111 * The namespace is manipulated using the following functions, all of which
112 * require the spa_namespace_lock to be held.
113 *
114 *	spa_lookup()		Lookup a spa_t by name.
115 *
116 *	spa_add()		Create a new spa_t in the namespace.
117 *
118 *	spa_remove()		Remove a spa_t from the namespace.  This also
119 *				frees up any memory associated with the spa_t.
120 *
121 *	spa_next()		Returns the next spa_t in the system, or the
122 *				first if NULL is passed.
123 *
124 *	spa_evict_all()		Shutdown and remove all spa_t structures in
125 *				the system.
126 *
127 *	spa_guid_exists()	Determine whether a pool/device guid exists.
128 *
129 * The spa_refcount is manipulated using the following functions:
130 *
131 *	spa_open_ref()		Adds a reference to the given spa_t.  Must be
132 *				called with spa_namespace_lock held if the
133 *				refcount is currently zero.
134 *
135 *	spa_close()		Remove a reference from the spa_t.  This will
136 *				not free the spa_t or remove it from the
137 *				namespace.  No locking is required.
138 *
139 *	spa_refcount_zero()	Returns true if the refcount is currently
140 *				zero.  Must be called with spa_namespace_lock
141 *				held.
142 *
143 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
144 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
145 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
146 *
147 * To read the configuration, it suffices to hold one of these locks as reader.
148 * To modify the configuration, you must hold all locks as writer.  To modify
149 * vdev state without altering the vdev tree's topology (e.g. online/offline),
150 * you must hold SCL_STATE and SCL_ZIO as writer.
151 *
152 * We use these distinct config locks to avoid recursive lock entry.
153 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
154 * block allocations (SCL_ALLOC), which may require reading space maps
155 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
156 *
157 * The spa config locks cannot be normal rwlocks because we need the
158 * ability to hand off ownership.  For example, SCL_ZIO is acquired
159 * by the issuing thread and later released by an interrupt thread.
160 * They do, however, obey the usual write-wanted semantics to prevent
161 * writer (i.e. system administrator) starvation.
162 *
163 * The lock acquisition rules are as follows:
164 *
165 * SCL_CONFIG
166 *	Protects changes to the vdev tree topology, such as vdev
167 *	add/remove/attach/detach.  Protects the dirty config list
168 *	(spa_config_dirty_list) and the set of spares and l2arc devices.
169 *
170 * SCL_STATE
171 *	Protects changes to pool state and vdev state, such as vdev
172 *	online/offline/fault/degrade/clear.  Protects the dirty state list
173 *	(spa_state_dirty_list) and global pool state (spa_state).
174 *
175 * SCL_ALLOC
176 *	Protects changes to metaslab groups and classes.
177 *	Held as reader by metaslab_alloc() and metaslab_claim().
178 *
179 * SCL_ZIO
180 *	Held by bp-level zios (those which have no io_vd upon entry)
181 *	to prevent changes to the vdev tree.  The bp-level zio implicitly
182 *	protects all of its vdev child zios, which do not hold SCL_ZIO.
183 *
184 * SCL_FREE
185 *	Protects changes to metaslab groups and classes.
186 *	Held as reader by metaslab_free().  SCL_FREE is distinct from
187 *	SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
188 *	blocks in zio_done() while another i/o that holds either
189 *	SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
190 *
191 * SCL_VDEV
192 *	Held as reader to prevent changes to the vdev tree during trivial
193 *	inquiries such as bp_get_dsize().  SCL_VDEV is distinct from the
194 *	other locks, and lower than all of them, to ensure that it's safe
195 *	to acquire regardless of caller context.
196 *
197 * In addition, the following rules apply:
198 *
199 * (a)	spa_props_lock protects pool properties, spa_config and spa_config_list.
200 *	The lock ordering is SCL_CONFIG > spa_props_lock.
201 *
202 * (b)	I/O operations on leaf vdevs.  For any zio operation that takes
203 *	an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
204 *	or zio_write_phys() -- the caller must ensure that the config cannot
205 *	cannot change in the interim, and that the vdev cannot be reopened.
206 *	SCL_STATE as reader suffices for both.
207 *
208 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
209 *
210 *	spa_vdev_enter()	Acquire the namespace lock and the config lock
211 *				for writing.
212 *
213 *	spa_vdev_exit()		Release the config lock, wait for all I/O
214 *				to complete, sync the updated configs to the
215 *				cache, and release the namespace lock.
216 *
217 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
218 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
219 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
220 *
221 * spa_rename() is also implemented within this file since it requires
222 * manipulation of the namespace.
223 */
224
225static avl_tree_t spa_namespace_avl;
226kmutex_t spa_namespace_lock;
227static kcondvar_t spa_namespace_cv;
228static int spa_active_count;
229int spa_max_replication_override = SPA_DVAS_PER_BP;
230
231static kmutex_t spa_spare_lock;
232static avl_tree_t spa_spare_avl;
233static kmutex_t spa_l2cache_lock;
234static avl_tree_t spa_l2cache_avl;
235
236kmem_cache_t *spa_buffer_pool;
237int spa_mode_global;
238
239#ifdef ZFS_DEBUG
240/* Everything except dprintf and spa is on by default in debug builds */
241int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA);
242#else
243int zfs_flags = 0;
244#endif
245
246/*
247 * zfs_recover can be set to nonzero to attempt to recover from
248 * otherwise-fatal errors, typically caused by on-disk corruption.  When
249 * set, calls to zfs_panic_recover() will turn into warning messages.
250 */
251int zfs_recover = 0;
252
253/*
254 * Expiration time in milliseconds. This value has two meanings. First it is
255 * used to determine when the spa_deadman() logic should fire. By default the
256 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
257 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
258 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
259 * in a system panic.
260 */
261uint64_t zfs_deadman_synctime_ms = 1000000ULL;
262
263/*
264 * Check time in milliseconds. This defines the frequency at which we check
265 * for hung I/O.
266 */
267uint64_t zfs_deadman_checktime_ms = 5000ULL;
268
269/*
270 * Override the zfs deadman behavior via /etc/system. By default the
271 * deadman is enabled except on VMware and sparc deployments.
272 */
273int zfs_deadman_enabled = -1;
274
275/*
276 * The worst case is single-sector max-parity RAID-Z blocks, in which
277 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
278 * times the size; so just assume that.  Add to this the fact that
279 * we can have up to 3 DVAs per bp, and one more factor of 2 because
280 * the block may be dittoed with up to 3 DVAs by ddt_sync().  All together,
281 * the worst case is:
282 *     (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
283 */
284int spa_asize_inflation = 24;
285
286/*
287 * ==========================================================================
288 * SPA config locking
289 * ==========================================================================
290 */
291static void
292spa_config_lock_init(spa_t *spa)
293{
294	for (int i = 0; i < SCL_LOCKS; i++) {
295		spa_config_lock_t *scl = &spa->spa_config_lock[i];
296		mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
297		cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
298		refcount_create_untracked(&scl->scl_count);
299		scl->scl_writer = NULL;
300		scl->scl_write_wanted = 0;
301	}
302}
303
304static void
305spa_config_lock_destroy(spa_t *spa)
306{
307	for (int i = 0; i < SCL_LOCKS; i++) {
308		spa_config_lock_t *scl = &spa->spa_config_lock[i];
309		mutex_destroy(&scl->scl_lock);
310		cv_destroy(&scl->scl_cv);
311		refcount_destroy(&scl->scl_count);
312		ASSERT(scl->scl_writer == NULL);
313		ASSERT(scl->scl_write_wanted == 0);
314	}
315}
316
317int
318spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
319{
320	for (int i = 0; i < SCL_LOCKS; i++) {
321		spa_config_lock_t *scl = &spa->spa_config_lock[i];
322		if (!(locks & (1 << i)))
323			continue;
324		mutex_enter(&scl->scl_lock);
325		if (rw == RW_READER) {
326			if (scl->scl_writer || scl->scl_write_wanted) {
327				mutex_exit(&scl->scl_lock);
328				spa_config_exit(spa, locks ^ (1 << i), tag);
329				return (0);
330			}
331		} else {
332			ASSERT(scl->scl_writer != curthread);
333			if (!refcount_is_zero(&scl->scl_count)) {
334				mutex_exit(&scl->scl_lock);
335				spa_config_exit(spa, locks ^ (1 << i), tag);
336				return (0);
337			}
338			scl->scl_writer = curthread;
339		}
340		(void) refcount_add(&scl->scl_count, tag);
341		mutex_exit(&scl->scl_lock);
342	}
343	return (1);
344}
345
346void
347spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
348{
349	int wlocks_held = 0;
350
351	ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
352
353	for (int i = 0; i < SCL_LOCKS; i++) {
354		spa_config_lock_t *scl = &spa->spa_config_lock[i];
355		if (scl->scl_writer == curthread)
356			wlocks_held |= (1 << i);
357		if (!(locks & (1 << i)))
358			continue;
359		mutex_enter(&scl->scl_lock);
360		if (rw == RW_READER) {
361			while (scl->scl_writer || scl->scl_write_wanted) {
362				cv_wait(&scl->scl_cv, &scl->scl_lock);
363			}
364		} else {
365			ASSERT(scl->scl_writer != curthread);
366			while (!refcount_is_zero(&scl->scl_count)) {
367				scl->scl_write_wanted++;
368				cv_wait(&scl->scl_cv, &scl->scl_lock);
369				scl->scl_write_wanted--;
370			}
371			scl->scl_writer = curthread;
372		}
373		(void) refcount_add(&scl->scl_count, tag);
374		mutex_exit(&scl->scl_lock);
375	}
376	ASSERT(wlocks_held <= locks);
377}
378
379void
380spa_config_exit(spa_t *spa, int locks, void *tag)
381{
382	for (int i = SCL_LOCKS - 1; i >= 0; i--) {
383		spa_config_lock_t *scl = &spa->spa_config_lock[i];
384		if (!(locks & (1 << i)))
385			continue;
386		mutex_enter(&scl->scl_lock);
387		ASSERT(!refcount_is_zero(&scl->scl_count));
388		if (refcount_remove(&scl->scl_count, tag) == 0) {
389			ASSERT(scl->scl_writer == NULL ||
390			    scl->scl_writer == curthread);
391			scl->scl_writer = NULL;	/* OK in either case */
392			cv_broadcast(&scl->scl_cv);
393		}
394		mutex_exit(&scl->scl_lock);
395	}
396}
397
398int
399spa_config_held(spa_t *spa, int locks, krw_t rw)
400{
401	int locks_held = 0;
402
403	for (int i = 0; i < SCL_LOCKS; i++) {
404		spa_config_lock_t *scl = &spa->spa_config_lock[i];
405		if (!(locks & (1 << i)))
406			continue;
407		if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
408		    (rw == RW_WRITER && scl->scl_writer == curthread))
409			locks_held |= 1 << i;
410	}
411
412	return (locks_held);
413}
414
415/*
416 * ==========================================================================
417 * SPA namespace functions
418 * ==========================================================================
419 */
420
421/*
422 * Lookup the named spa_t in the AVL tree.  The spa_namespace_lock must be held.
423 * Returns NULL if no matching spa_t is found.
424 */
425spa_t *
426spa_lookup(const char *name)
427{
428	static spa_t search;	/* spa_t is large; don't allocate on stack */
429	spa_t *spa;
430	avl_index_t where;
431	char *cp;
432
433	ASSERT(MUTEX_HELD(&spa_namespace_lock));
434
435	(void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
436
437	/*
438	 * If it's a full dataset name, figure out the pool name and
439	 * just use that.
440	 */
441	cp = strpbrk(search.spa_name, "/@#");
442	if (cp != NULL)
443		*cp = '\0';
444
445	spa = avl_find(&spa_namespace_avl, &search, &where);
446
447	return (spa);
448}
449
450/*
451 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
452 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
453 * looking for potentially hung I/Os.
454 */
455void
456spa_deadman(void *arg)
457{
458	spa_t *spa = arg;
459
460	/*
461	 * Disable the deadman timer if the pool is suspended.
462	 */
463	if (spa_suspended(spa)) {
464		VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
465		return;
466	}
467
468	zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
469	    (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
470	    ++spa->spa_deadman_calls);
471	if (zfs_deadman_enabled)
472		vdev_deadman(spa->spa_root_vdev);
473}
474
475/*
476 * Create an uninitialized spa_t with the given name.  Requires
477 * spa_namespace_lock.  The caller must ensure that the spa_t doesn't already
478 * exist by calling spa_lookup() first.
479 */
480spa_t *
481spa_add(const char *name, nvlist_t *config, const char *altroot)
482{
483	spa_t *spa;
484	spa_config_dirent_t *dp;
485	cyc_handler_t hdlr;
486	cyc_time_t when;
487
488	ASSERT(MUTEX_HELD(&spa_namespace_lock));
489
490	spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
491
492	mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
493	mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
494	mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
495	mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
496	mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
497	mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
498	mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
499	mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
500	mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
501	mutex_init(&spa->spa_iokstat_lock, NULL, MUTEX_DEFAULT, NULL);
502
503	cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
504	cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
505	cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
506	cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
507
508	for (int t = 0; t < TXG_SIZE; t++)
509		bplist_create(&spa->spa_free_bplist[t]);
510
511	(void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
512	spa->spa_state = POOL_STATE_UNINITIALIZED;
513	spa->spa_freeze_txg = UINT64_MAX;
514	spa->spa_final_txg = UINT64_MAX;
515	spa->spa_load_max_txg = UINT64_MAX;
516	spa->spa_proc = &p0;
517	spa->spa_proc_state = SPA_PROC_NONE;
518
519	hdlr.cyh_func = spa_deadman;
520	hdlr.cyh_arg = spa;
521	hdlr.cyh_level = CY_LOW_LEVEL;
522
523	spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
524
525	/*
526	 * This determines how often we need to check for hung I/Os after
527	 * the cyclic has already fired. Since checking for hung I/Os is
528	 * an expensive operation we don't want to check too frequently.
529	 * Instead wait for 5 seconds before checking again.
530	 */
531	when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms);
532	when.cyt_when = CY_INFINITY;
533	mutex_enter(&cpu_lock);
534	spa->spa_deadman_cycid = cyclic_add(&hdlr, &when);
535	mutex_exit(&cpu_lock);
536
537	refcount_create(&spa->spa_refcount);
538	spa_config_lock_init(spa);
539
540	avl_add(&spa_namespace_avl, spa);
541
542	/*
543	 * Set the alternate root, if there is one.
544	 */
545	if (altroot) {
546		spa->spa_root = spa_strdup(altroot);
547		spa_active_count++;
548	}
549
550	/*
551	 * Every pool starts with the default cachefile
552	 */
553	list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
554	    offsetof(spa_config_dirent_t, scd_link));
555
556	dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
557	dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
558	list_insert_head(&spa->spa_config_list, dp);
559
560	VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
561	    KM_SLEEP) == 0);
562
563	if (config != NULL) {
564		nvlist_t *features;
565
566		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
567		    &features) == 0) {
568			VERIFY(nvlist_dup(features, &spa->spa_label_features,
569			    0) == 0);
570		}
571
572		VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
573	}
574
575	if (spa->spa_label_features == NULL) {
576		VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
577		    KM_SLEEP) == 0);
578	}
579
580	spa->spa_iokstat = kstat_create("zfs", 0, name,
581	    "disk", KSTAT_TYPE_IO, 1, 0);
582	if (spa->spa_iokstat) {
583		spa->spa_iokstat->ks_lock = &spa->spa_iokstat_lock;
584		kstat_install(spa->spa_iokstat);
585	}
586
587	spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0);
588
589	/*
590	 * As a pool is being created, treat all features as disabled by
591	 * setting SPA_FEATURE_DISABLED for all entries in the feature
592	 * refcount cache.
593	 */
594	for (int i = 0; i < SPA_FEATURES; i++) {
595		spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED;
596	}
597
598	return (spa);
599}
600
601/*
602 * Removes a spa_t from the namespace, freeing up any memory used.  Requires
603 * spa_namespace_lock.  This is called only after the spa_t has been closed and
604 * deactivated.
605 */
606void
607spa_remove(spa_t *spa)
608{
609	spa_config_dirent_t *dp;
610
611	ASSERT(MUTEX_HELD(&spa_namespace_lock));
612	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
613
614	nvlist_free(spa->spa_config_splitting);
615
616	avl_remove(&spa_namespace_avl, spa);
617	cv_broadcast(&spa_namespace_cv);
618
619	if (spa->spa_root) {
620		spa_strfree(spa->spa_root);
621		spa_active_count--;
622	}
623
624	while ((dp = list_head(&spa->spa_config_list)) != NULL) {
625		list_remove(&spa->spa_config_list, dp);
626		if (dp->scd_path != NULL)
627			spa_strfree(dp->scd_path);
628		kmem_free(dp, sizeof (spa_config_dirent_t));
629	}
630
631	list_destroy(&spa->spa_config_list);
632
633	nvlist_free(spa->spa_label_features);
634	nvlist_free(spa->spa_load_info);
635	spa_config_set(spa, NULL);
636
637	mutex_enter(&cpu_lock);
638	if (spa->spa_deadman_cycid != CYCLIC_NONE)
639		cyclic_remove(spa->spa_deadman_cycid);
640	mutex_exit(&cpu_lock);
641	spa->spa_deadman_cycid = CYCLIC_NONE;
642
643	refcount_destroy(&spa->spa_refcount);
644
645	spa_config_lock_destroy(spa);
646
647	kstat_delete(spa->spa_iokstat);
648	spa->spa_iokstat = NULL;
649
650	for (int t = 0; t < TXG_SIZE; t++)
651		bplist_destroy(&spa->spa_free_bplist[t]);
652
653	cv_destroy(&spa->spa_async_cv);
654	cv_destroy(&spa->spa_proc_cv);
655	cv_destroy(&spa->spa_scrub_io_cv);
656	cv_destroy(&spa->spa_suspend_cv);
657
658	mutex_destroy(&spa->spa_async_lock);
659	mutex_destroy(&spa->spa_errlist_lock);
660	mutex_destroy(&spa->spa_errlog_lock);
661	mutex_destroy(&spa->spa_history_lock);
662	mutex_destroy(&spa->spa_proc_lock);
663	mutex_destroy(&spa->spa_props_lock);
664	mutex_destroy(&spa->spa_scrub_lock);
665	mutex_destroy(&spa->spa_suspend_lock);
666	mutex_destroy(&spa->spa_vdev_top_lock);
667	mutex_destroy(&spa->spa_iokstat_lock);
668
669	kmem_free(spa, sizeof (spa_t));
670}
671
672/*
673 * Given a pool, return the next pool in the namespace, or NULL if there is
674 * none.  If 'prev' is NULL, return the first pool.
675 */
676spa_t *
677spa_next(spa_t *prev)
678{
679	ASSERT(MUTEX_HELD(&spa_namespace_lock));
680
681	if (prev)
682		return (AVL_NEXT(&spa_namespace_avl, prev));
683	else
684		return (avl_first(&spa_namespace_avl));
685}
686
687/*
688 * ==========================================================================
689 * SPA refcount functions
690 * ==========================================================================
691 */
692
693/*
694 * Add a reference to the given spa_t.  Must have at least one reference, or
695 * have the namespace lock held.
696 */
697void
698spa_open_ref(spa_t *spa, void *tag)
699{
700	ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
701	    MUTEX_HELD(&spa_namespace_lock));
702	(void) refcount_add(&spa->spa_refcount, tag);
703}
704
705/*
706 * Remove a reference to the given spa_t.  Must have at least one reference, or
707 * have the namespace lock held.
708 */
709void
710spa_close(spa_t *spa, void *tag)
711{
712	ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
713	    MUTEX_HELD(&spa_namespace_lock));
714	(void) refcount_remove(&spa->spa_refcount, tag);
715}
716
717/*
718 * Check to see if the spa refcount is zero.  Must be called with
719 * spa_namespace_lock held.  We really compare against spa_minref, which is the
720 * number of references acquired when opening a pool
721 */
722boolean_t
723spa_refcount_zero(spa_t *spa)
724{
725	ASSERT(MUTEX_HELD(&spa_namespace_lock));
726
727	return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
728}
729
730/*
731 * ==========================================================================
732 * SPA spare and l2cache tracking
733 * ==========================================================================
734 */
735
736/*
737 * Hot spares and cache devices are tracked using the same code below,
738 * for 'auxiliary' devices.
739 */
740
741typedef struct spa_aux {
742	uint64_t	aux_guid;
743	uint64_t	aux_pool;
744	avl_node_t	aux_avl;
745	int		aux_count;
746} spa_aux_t;
747
748static int
749spa_aux_compare(const void *a, const void *b)
750{
751	const spa_aux_t *sa = a;
752	const spa_aux_t *sb = b;
753
754	if (sa->aux_guid < sb->aux_guid)
755		return (-1);
756	else if (sa->aux_guid > sb->aux_guid)
757		return (1);
758	else
759		return (0);
760}
761
762void
763spa_aux_add(vdev_t *vd, avl_tree_t *avl)
764{
765	avl_index_t where;
766	spa_aux_t search;
767	spa_aux_t *aux;
768
769	search.aux_guid = vd->vdev_guid;
770	if ((aux = avl_find(avl, &search, &where)) != NULL) {
771		aux->aux_count++;
772	} else {
773		aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
774		aux->aux_guid = vd->vdev_guid;
775		aux->aux_count = 1;
776		avl_insert(avl, aux, where);
777	}
778}
779
780void
781spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
782{
783	spa_aux_t search;
784	spa_aux_t *aux;
785	avl_index_t where;
786
787	search.aux_guid = vd->vdev_guid;
788	aux = avl_find(avl, &search, &where);
789
790	ASSERT(aux != NULL);
791
792	if (--aux->aux_count == 0) {
793		avl_remove(avl, aux);
794		kmem_free(aux, sizeof (spa_aux_t));
795	} else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
796		aux->aux_pool = 0ULL;
797	}
798}
799
800boolean_t
801spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
802{
803	spa_aux_t search, *found;
804
805	search.aux_guid = guid;
806	found = avl_find(avl, &search, NULL);
807
808	if (pool) {
809		if (found)
810			*pool = found->aux_pool;
811		else
812			*pool = 0ULL;
813	}
814
815	if (refcnt) {
816		if (found)
817			*refcnt = found->aux_count;
818		else
819			*refcnt = 0;
820	}
821
822	return (found != NULL);
823}
824
825void
826spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
827{
828	spa_aux_t search, *found;
829	avl_index_t where;
830
831	search.aux_guid = vd->vdev_guid;
832	found = avl_find(avl, &search, &where);
833	ASSERT(found != NULL);
834	ASSERT(found->aux_pool == 0ULL);
835
836	found->aux_pool = spa_guid(vd->vdev_spa);
837}
838
839/*
840 * Spares are tracked globally due to the following constraints:
841 *
842 * 	- A spare may be part of multiple pools.
843 * 	- A spare may be added to a pool even if it's actively in use within
844 *	  another pool.
845 * 	- A spare in use in any pool can only be the source of a replacement if
846 *	  the target is a spare in the same pool.
847 *
848 * We keep track of all spares on the system through the use of a reference
849 * counted AVL tree.  When a vdev is added as a spare, or used as a replacement
850 * spare, then we bump the reference count in the AVL tree.  In addition, we set
851 * the 'vdev_isspare' member to indicate that the device is a spare (active or
852 * inactive).  When a spare is made active (used to replace a device in the
853 * pool), we also keep track of which pool its been made a part of.
854 *
855 * The 'spa_spare_lock' protects the AVL tree.  These functions are normally
856 * called under the spa_namespace lock as part of vdev reconfiguration.  The
857 * separate spare lock exists for the status query path, which does not need to
858 * be completely consistent with respect to other vdev configuration changes.
859 */
860
861static int
862spa_spare_compare(const void *a, const void *b)
863{
864	return (spa_aux_compare(a, b));
865}
866
867void
868spa_spare_add(vdev_t *vd)
869{
870	mutex_enter(&spa_spare_lock);
871	ASSERT(!vd->vdev_isspare);
872	spa_aux_add(vd, &spa_spare_avl);
873	vd->vdev_isspare = B_TRUE;
874	mutex_exit(&spa_spare_lock);
875}
876
877void
878spa_spare_remove(vdev_t *vd)
879{
880	mutex_enter(&spa_spare_lock);
881	ASSERT(vd->vdev_isspare);
882	spa_aux_remove(vd, &spa_spare_avl);
883	vd->vdev_isspare = B_FALSE;
884	mutex_exit(&spa_spare_lock);
885}
886
887boolean_t
888spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
889{
890	boolean_t found;
891
892	mutex_enter(&spa_spare_lock);
893	found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
894	mutex_exit(&spa_spare_lock);
895
896	return (found);
897}
898
899void
900spa_spare_activate(vdev_t *vd)
901{
902	mutex_enter(&spa_spare_lock);
903	ASSERT(vd->vdev_isspare);
904	spa_aux_activate(vd, &spa_spare_avl);
905	mutex_exit(&spa_spare_lock);
906}
907
908/*
909 * Level 2 ARC devices are tracked globally for the same reasons as spares.
910 * Cache devices currently only support one pool per cache device, and so
911 * for these devices the aux reference count is currently unused beyond 1.
912 */
913
914static int
915spa_l2cache_compare(const void *a, const void *b)
916{
917	return (spa_aux_compare(a, b));
918}
919
920void
921spa_l2cache_add(vdev_t *vd)
922{
923	mutex_enter(&spa_l2cache_lock);
924	ASSERT(!vd->vdev_isl2cache);
925	spa_aux_add(vd, &spa_l2cache_avl);
926	vd->vdev_isl2cache = B_TRUE;
927	mutex_exit(&spa_l2cache_lock);
928}
929
930void
931spa_l2cache_remove(vdev_t *vd)
932{
933	mutex_enter(&spa_l2cache_lock);
934	ASSERT(vd->vdev_isl2cache);
935	spa_aux_remove(vd, &spa_l2cache_avl);
936	vd->vdev_isl2cache = B_FALSE;
937	mutex_exit(&spa_l2cache_lock);
938}
939
940boolean_t
941spa_l2cache_exists(uint64_t guid, uint64_t *pool)
942{
943	boolean_t found;
944
945	mutex_enter(&spa_l2cache_lock);
946	found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
947	mutex_exit(&spa_l2cache_lock);
948
949	return (found);
950}
951
952void
953spa_l2cache_activate(vdev_t *vd)
954{
955	mutex_enter(&spa_l2cache_lock);
956	ASSERT(vd->vdev_isl2cache);
957	spa_aux_activate(vd, &spa_l2cache_avl);
958	mutex_exit(&spa_l2cache_lock);
959}
960
961/*
962 * ==========================================================================
963 * SPA vdev locking
964 * ==========================================================================
965 */
966
967/*
968 * Lock the given spa_t for the purpose of adding or removing a vdev.
969 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
970 * It returns the next transaction group for the spa_t.
971 */
972uint64_t
973spa_vdev_enter(spa_t *spa)
974{
975	mutex_enter(&spa->spa_vdev_top_lock);
976	mutex_enter(&spa_namespace_lock);
977	return (spa_vdev_config_enter(spa));
978}
979
980/*
981 * Internal implementation for spa_vdev_enter().  Used when a vdev
982 * operation requires multiple syncs (i.e. removing a device) while
983 * keeping the spa_namespace_lock held.
984 */
985uint64_t
986spa_vdev_config_enter(spa_t *spa)
987{
988	ASSERT(MUTEX_HELD(&spa_namespace_lock));
989
990	spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
991
992	return (spa_last_synced_txg(spa) + 1);
993}
994
995/*
996 * Used in combination with spa_vdev_config_enter() to allow the syncing
997 * of multiple transactions without releasing the spa_namespace_lock.
998 */
999void
1000spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
1001{
1002	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1003
1004	int config_changed = B_FALSE;
1005
1006	ASSERT(txg > spa_last_synced_txg(spa));
1007
1008	spa->spa_pending_vdev = NULL;
1009
1010	/*
1011	 * Reassess the DTLs.
1012	 */
1013	vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
1014
1015	if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
1016		config_changed = B_TRUE;
1017		spa->spa_config_generation++;
1018	}
1019
1020	/*
1021	 * Verify the metaslab classes.
1022	 */
1023	ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
1024	ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
1025
1026	spa_config_exit(spa, SCL_ALL, spa);
1027
1028	/*
1029	 * Panic the system if the specified tag requires it.  This
1030	 * is useful for ensuring that configurations are updated
1031	 * transactionally.
1032	 */
1033	if (zio_injection_enabled)
1034		zio_handle_panic_injection(spa, tag, 0);
1035
1036	/*
1037	 * Note: this txg_wait_synced() is important because it ensures
1038	 * that there won't be more than one config change per txg.
1039	 * This allows us to use the txg as the generation number.
1040	 */
1041	if (error == 0)
1042		txg_wait_synced(spa->spa_dsl_pool, txg);
1043
1044	if (vd != NULL) {
1045		ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
1046		spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1047		vdev_free(vd);
1048		spa_config_exit(spa, SCL_ALL, spa);
1049	}
1050
1051	/*
1052	 * If the config changed, update the config cache.
1053	 */
1054	if (config_changed)
1055		spa_config_sync(spa, B_FALSE, B_TRUE);
1056}
1057
1058/*
1059 * Unlock the spa_t after adding or removing a vdev.  Besides undoing the
1060 * locking of spa_vdev_enter(), we also want make sure the transactions have
1061 * synced to disk, and then update the global configuration cache with the new
1062 * information.
1063 */
1064int
1065spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1066{
1067	spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1068	mutex_exit(&spa_namespace_lock);
1069	mutex_exit(&spa->spa_vdev_top_lock);
1070
1071	return (error);
1072}
1073
1074/*
1075 * Lock the given spa_t for the purpose of changing vdev state.
1076 */
1077void
1078spa_vdev_state_enter(spa_t *spa, int oplocks)
1079{
1080	int locks = SCL_STATE_ALL | oplocks;
1081
1082	/*
1083	 * Root pools may need to read of the underlying devfs filesystem
1084	 * when opening up a vdev.  Unfortunately if we're holding the
1085	 * SCL_ZIO lock it will result in a deadlock when we try to issue
1086	 * the read from the root filesystem.  Instead we "prefetch"
1087	 * the associated vnodes that we need prior to opening the
1088	 * underlying devices and cache them so that we can prevent
1089	 * any I/O when we are doing the actual open.
1090	 */
1091	if (spa_is_root(spa)) {
1092		int low = locks & ~(SCL_ZIO - 1);
1093		int high = locks & ~low;
1094
1095		spa_config_enter(spa, high, spa, RW_WRITER);
1096		vdev_hold(spa->spa_root_vdev);
1097		spa_config_enter(spa, low, spa, RW_WRITER);
1098	} else {
1099		spa_config_enter(spa, locks, spa, RW_WRITER);
1100	}
1101	spa->spa_vdev_locks = locks;
1102}
1103
1104int
1105spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1106{
1107	boolean_t config_changed = B_FALSE;
1108
1109	if (vd != NULL || error == 0)
1110		vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1111		    0, 0, B_FALSE);
1112
1113	if (vd != NULL) {
1114		vdev_state_dirty(vd->vdev_top);
1115		config_changed = B_TRUE;
1116		spa->spa_config_generation++;
1117	}
1118
1119	if (spa_is_root(spa))
1120		vdev_rele(spa->spa_root_vdev);
1121
1122	ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1123	spa_config_exit(spa, spa->spa_vdev_locks, spa);
1124
1125	/*
1126	 * If anything changed, wait for it to sync.  This ensures that,
1127	 * from the system administrator's perspective, zpool(1M) commands
1128	 * are synchronous.  This is important for things like zpool offline:
1129	 * when the command completes, you expect no further I/O from ZFS.
1130	 */
1131	if (vd != NULL)
1132		txg_wait_synced(spa->spa_dsl_pool, 0);
1133
1134	/*
1135	 * If the config changed, update the config cache.
1136	 */
1137	if (config_changed) {
1138		mutex_enter(&spa_namespace_lock);
1139		spa_config_sync(spa, B_FALSE, B_TRUE);
1140		mutex_exit(&spa_namespace_lock);
1141	}
1142
1143	return (error);
1144}
1145
1146/*
1147 * ==========================================================================
1148 * Miscellaneous functions
1149 * ==========================================================================
1150 */
1151
1152void
1153spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx)
1154{
1155	if (!nvlist_exists(spa->spa_label_features, feature)) {
1156		fnvlist_add_boolean(spa->spa_label_features, feature);
1157		/*
1158		 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1159		 * dirty the vdev config because lock SCL_CONFIG is not held.
1160		 * Thankfully, in this case we don't need to dirty the config
1161		 * because it will be written out anyway when we finish
1162		 * creating the pool.
1163		 */
1164		if (tx->tx_txg != TXG_INITIAL)
1165			vdev_config_dirty(spa->spa_root_vdev);
1166	}
1167}
1168
1169void
1170spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1171{
1172	if (nvlist_remove_all(spa->spa_label_features, feature) == 0)
1173		vdev_config_dirty(spa->spa_root_vdev);
1174}
1175
1176/*
1177 * Rename a spa_t.
1178 */
1179int
1180spa_rename(const char *name, const char *newname)
1181{
1182	spa_t *spa;
1183	int err;
1184
1185	/*
1186	 * Lookup the spa_t and grab the config lock for writing.  We need to
1187	 * actually open the pool so that we can sync out the necessary labels.
1188	 * It's OK to call spa_open() with the namespace lock held because we
1189	 * allow recursive calls for other reasons.
1190	 */
1191	mutex_enter(&spa_namespace_lock);
1192	if ((err = spa_open(name, &spa, FTAG)) != 0) {
1193		mutex_exit(&spa_namespace_lock);
1194		return (err);
1195	}
1196
1197	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1198
1199	avl_remove(&spa_namespace_avl, spa);
1200	(void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1201	avl_add(&spa_namespace_avl, spa);
1202
1203	/*
1204	 * Sync all labels to disk with the new names by marking the root vdev
1205	 * dirty and waiting for it to sync.  It will pick up the new pool name
1206	 * during the sync.
1207	 */
1208	vdev_config_dirty(spa->spa_root_vdev);
1209
1210	spa_config_exit(spa, SCL_ALL, FTAG);
1211
1212	txg_wait_synced(spa->spa_dsl_pool, 0);
1213
1214	/*
1215	 * Sync the updated config cache.
1216	 */
1217	spa_config_sync(spa, B_FALSE, B_TRUE);
1218
1219	spa_close(spa, FTAG);
1220
1221	mutex_exit(&spa_namespace_lock);
1222
1223	return (0);
1224}
1225
1226/*
1227 * Return the spa_t associated with given pool_guid, if it exists.  If
1228 * device_guid is non-zero, determine whether the pool exists *and* contains
1229 * a device with the specified device_guid.
1230 */
1231spa_t *
1232spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1233{
1234	spa_t *spa;
1235	avl_tree_t *t = &spa_namespace_avl;
1236
1237	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1238
1239	for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1240		if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1241			continue;
1242		if (spa->spa_root_vdev == NULL)
1243			continue;
1244		if (spa_guid(spa) == pool_guid) {
1245			if (device_guid == 0)
1246				break;
1247
1248			if (vdev_lookup_by_guid(spa->spa_root_vdev,
1249			    device_guid) != NULL)
1250				break;
1251
1252			/*
1253			 * Check any devices we may be in the process of adding.
1254			 */
1255			if (spa->spa_pending_vdev) {
1256				if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1257				    device_guid) != NULL)
1258					break;
1259			}
1260		}
1261	}
1262
1263	return (spa);
1264}
1265
1266/*
1267 * Determine whether a pool with the given pool_guid exists.
1268 */
1269boolean_t
1270spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1271{
1272	return (spa_by_guid(pool_guid, device_guid) != NULL);
1273}
1274
1275char *
1276spa_strdup(const char *s)
1277{
1278	size_t len;
1279	char *new;
1280
1281	len = strlen(s);
1282	new = kmem_alloc(len + 1, KM_SLEEP);
1283	bcopy(s, new, len);
1284	new[len] = '\0';
1285
1286	return (new);
1287}
1288
1289void
1290spa_strfree(char *s)
1291{
1292	kmem_free(s, strlen(s) + 1);
1293}
1294
1295uint64_t
1296spa_get_random(uint64_t range)
1297{
1298	uint64_t r;
1299
1300	ASSERT(range != 0);
1301
1302	(void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1303
1304	return (r % range);
1305}
1306
1307uint64_t
1308spa_generate_guid(spa_t *spa)
1309{
1310	uint64_t guid = spa_get_random(-1ULL);
1311
1312	if (spa != NULL) {
1313		while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1314			guid = spa_get_random(-1ULL);
1315	} else {
1316		while (guid == 0 || spa_guid_exists(guid, 0))
1317			guid = spa_get_random(-1ULL);
1318	}
1319
1320	return (guid);
1321}
1322
1323void
1324snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp)
1325{
1326	char type[256];
1327	char *checksum = NULL;
1328	char *compress = NULL;
1329
1330	if (bp != NULL) {
1331		if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1332			dmu_object_byteswap_t bswap =
1333			    DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1334			(void) snprintf(type, sizeof (type), "bswap %s %s",
1335			    DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1336			    "metadata" : "data",
1337			    dmu_ot_byteswap[bswap].ob_name);
1338		} else {
1339			(void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1340			    sizeof (type));
1341		}
1342		checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1343		compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1344	}
1345
1346	SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum,
1347	    compress);
1348}
1349
1350void
1351spa_freeze(spa_t *spa)
1352{
1353	uint64_t freeze_txg = 0;
1354
1355	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1356	if (spa->spa_freeze_txg == UINT64_MAX) {
1357		freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1358		spa->spa_freeze_txg = freeze_txg;
1359	}
1360	spa_config_exit(spa, SCL_ALL, FTAG);
1361	if (freeze_txg != 0)
1362		txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1363}
1364
1365void
1366zfs_panic_recover(const char *fmt, ...)
1367{
1368	va_list adx;
1369
1370	va_start(adx, fmt);
1371	vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1372	va_end(adx);
1373}
1374
1375/*
1376 * This is a stripped-down version of strtoull, suitable only for converting
1377 * lowercase hexadecimal numbers that don't overflow.
1378 */
1379uint64_t
1380strtonum(const char *str, char **nptr)
1381{
1382	uint64_t val = 0;
1383	char c;
1384	int digit;
1385
1386	while ((c = *str) != '\0') {
1387		if (c >= '0' && c <= '9')
1388			digit = c - '0';
1389		else if (c >= 'a' && c <= 'f')
1390			digit = 10 + c - 'a';
1391		else
1392			break;
1393
1394		val *= 16;
1395		val += digit;
1396
1397		str++;
1398	}
1399
1400	if (nptr)
1401		*nptr = (char *)str;
1402
1403	return (val);
1404}
1405
1406/*
1407 * ==========================================================================
1408 * Accessor functions
1409 * ==========================================================================
1410 */
1411
1412boolean_t
1413spa_shutting_down(spa_t *spa)
1414{
1415	return (spa->spa_async_suspended);
1416}
1417
1418dsl_pool_t *
1419spa_get_dsl(spa_t *spa)
1420{
1421	return (spa->spa_dsl_pool);
1422}
1423
1424boolean_t
1425spa_is_initializing(spa_t *spa)
1426{
1427	return (spa->spa_is_initializing);
1428}
1429
1430blkptr_t *
1431spa_get_rootblkptr(spa_t *spa)
1432{
1433	return (&spa->spa_ubsync.ub_rootbp);
1434}
1435
1436void
1437spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1438{
1439	spa->spa_uberblock.ub_rootbp = *bp;
1440}
1441
1442void
1443spa_altroot(spa_t *spa, char *buf, size_t buflen)
1444{
1445	if (spa->spa_root == NULL)
1446		buf[0] = '\0';
1447	else
1448		(void) strncpy(buf, spa->spa_root, buflen);
1449}
1450
1451int
1452spa_sync_pass(spa_t *spa)
1453{
1454	return (spa->spa_sync_pass);
1455}
1456
1457char *
1458spa_name(spa_t *spa)
1459{
1460	return (spa->spa_name);
1461}
1462
1463uint64_t
1464spa_guid(spa_t *spa)
1465{
1466	dsl_pool_t *dp = spa_get_dsl(spa);
1467	uint64_t guid;
1468
1469	/*
1470	 * If we fail to parse the config during spa_load(), we can go through
1471	 * the error path (which posts an ereport) and end up here with no root
1472	 * vdev.  We stash the original pool guid in 'spa_config_guid' to handle
1473	 * this case.
1474	 */
1475	if (spa->spa_root_vdev == NULL)
1476		return (spa->spa_config_guid);
1477
1478	guid = spa->spa_last_synced_guid != 0 ?
1479	    spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1480
1481	/*
1482	 * Return the most recently synced out guid unless we're
1483	 * in syncing context.
1484	 */
1485	if (dp && dsl_pool_sync_context(dp))
1486		return (spa->spa_root_vdev->vdev_guid);
1487	else
1488		return (guid);
1489}
1490
1491uint64_t
1492spa_load_guid(spa_t *spa)
1493{
1494	/*
1495	 * This is a GUID that exists solely as a reference for the
1496	 * purposes of the arc.  It is generated at load time, and
1497	 * is never written to persistent storage.
1498	 */
1499	return (spa->spa_load_guid);
1500}
1501
1502uint64_t
1503spa_last_synced_txg(spa_t *spa)
1504{
1505	return (spa->spa_ubsync.ub_txg);
1506}
1507
1508uint64_t
1509spa_first_txg(spa_t *spa)
1510{
1511	return (spa->spa_first_txg);
1512}
1513
1514uint64_t
1515spa_syncing_txg(spa_t *spa)
1516{
1517	return (spa->spa_syncing_txg);
1518}
1519
1520pool_state_t
1521spa_state(spa_t *spa)
1522{
1523	return (spa->spa_state);
1524}
1525
1526spa_load_state_t
1527spa_load_state(spa_t *spa)
1528{
1529	return (spa->spa_load_state);
1530}
1531
1532uint64_t
1533spa_freeze_txg(spa_t *spa)
1534{
1535	return (spa->spa_freeze_txg);
1536}
1537
1538/* ARGSUSED */
1539uint64_t
1540spa_get_asize(spa_t *spa, uint64_t lsize)
1541{
1542	return (lsize * spa_asize_inflation);
1543}
1544
1545uint64_t
1546spa_get_dspace(spa_t *spa)
1547{
1548	return (spa->spa_dspace);
1549}
1550
1551void
1552spa_update_dspace(spa_t *spa)
1553{
1554	spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1555	    ddt_get_dedup_dspace(spa);
1556}
1557
1558/*
1559 * Return the failure mode that has been set to this pool. The default
1560 * behavior will be to block all I/Os when a complete failure occurs.
1561 */
1562uint8_t
1563spa_get_failmode(spa_t *spa)
1564{
1565	return (spa->spa_failmode);
1566}
1567
1568boolean_t
1569spa_suspended(spa_t *spa)
1570{
1571	return (spa->spa_suspended);
1572}
1573
1574uint64_t
1575spa_version(spa_t *spa)
1576{
1577	return (spa->spa_ubsync.ub_version);
1578}
1579
1580boolean_t
1581spa_deflate(spa_t *spa)
1582{
1583	return (spa->spa_deflate);
1584}
1585
1586metaslab_class_t *
1587spa_normal_class(spa_t *spa)
1588{
1589	return (spa->spa_normal_class);
1590}
1591
1592metaslab_class_t *
1593spa_log_class(spa_t *spa)
1594{
1595	return (spa->spa_log_class);
1596}
1597
1598int
1599spa_max_replication(spa_t *spa)
1600{
1601	/*
1602	 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1603	 * handle BPs with more than one DVA allocated.  Set our max
1604	 * replication level accordingly.
1605	 */
1606	if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1607		return (1);
1608	return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1609}
1610
1611int
1612spa_prev_software_version(spa_t *spa)
1613{
1614	return (spa->spa_prev_software_version);
1615}
1616
1617uint64_t
1618spa_deadman_synctime(spa_t *spa)
1619{
1620	return (spa->spa_deadman_synctime);
1621}
1622
1623uint64_t
1624dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1625{
1626	uint64_t asize = DVA_GET_ASIZE(dva);
1627	uint64_t dsize = asize;
1628
1629	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1630
1631	if (asize != 0 && spa->spa_deflate) {
1632		vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1633		dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1634	}
1635
1636	return (dsize);
1637}
1638
1639uint64_t
1640bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1641{
1642	uint64_t dsize = 0;
1643
1644	for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1645		dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1646
1647	return (dsize);
1648}
1649
1650uint64_t
1651bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1652{
1653	uint64_t dsize = 0;
1654
1655	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1656
1657	for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1658		dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1659
1660	spa_config_exit(spa, SCL_VDEV, FTAG);
1661
1662	return (dsize);
1663}
1664
1665/*
1666 * ==========================================================================
1667 * Initialization and Termination
1668 * ==========================================================================
1669 */
1670
1671static int
1672spa_name_compare(const void *a1, const void *a2)
1673{
1674	const spa_t *s1 = a1;
1675	const spa_t *s2 = a2;
1676	int s;
1677
1678	s = strcmp(s1->spa_name, s2->spa_name);
1679	if (s > 0)
1680		return (1);
1681	if (s < 0)
1682		return (-1);
1683	return (0);
1684}
1685
1686int
1687spa_busy(void)
1688{
1689	return (spa_active_count);
1690}
1691
1692void
1693spa_boot_init()
1694{
1695	spa_config_load();
1696}
1697
1698void
1699spa_init(int mode)
1700{
1701	mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1702	mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1703	mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1704	cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1705
1706	avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1707	    offsetof(spa_t, spa_avl));
1708
1709	avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1710	    offsetof(spa_aux_t, aux_avl));
1711
1712	avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1713	    offsetof(spa_aux_t, aux_avl));
1714
1715	spa_mode_global = mode;
1716
1717#ifdef _KERNEL
1718	spa_arch_init();
1719#else
1720	if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
1721		arc_procfd = open("/proc/self/ctl", O_WRONLY);
1722		if (arc_procfd == -1) {
1723			perror("could not enable watchpoints: "
1724			    "opening /proc/self/ctl failed: ");
1725		} else {
1726			arc_watch = B_TRUE;
1727		}
1728	}
1729#endif
1730
1731	refcount_init();
1732	unique_init();
1733	range_tree_init();
1734	zio_init();
1735	dmu_init();
1736	zil_init();
1737	vdev_cache_stat_init();
1738	zfs_prop_init();
1739	zpool_prop_init();
1740	zpool_feature_init();
1741	spa_config_load();
1742	l2arc_start();
1743}
1744
1745void
1746spa_fini(void)
1747{
1748	l2arc_stop();
1749
1750	spa_evict_all();
1751
1752	vdev_cache_stat_fini();
1753	zil_fini();
1754	dmu_fini();
1755	zio_fini();
1756	range_tree_fini();
1757	unique_fini();
1758	refcount_fini();
1759
1760	avl_destroy(&spa_namespace_avl);
1761	avl_destroy(&spa_spare_avl);
1762	avl_destroy(&spa_l2cache_avl);
1763
1764	cv_destroy(&spa_namespace_cv);
1765	mutex_destroy(&spa_namespace_lock);
1766	mutex_destroy(&spa_spare_lock);
1767	mutex_destroy(&spa_l2cache_lock);
1768}
1769
1770/*
1771 * Return whether this pool has slogs. No locking needed.
1772 * It's not a problem if the wrong answer is returned as it's only for
1773 * performance and not correctness
1774 */
1775boolean_t
1776spa_has_slogs(spa_t *spa)
1777{
1778	return (spa->spa_log_class->mc_rotor != NULL);
1779}
1780
1781spa_log_state_t
1782spa_get_log_state(spa_t *spa)
1783{
1784	return (spa->spa_log_state);
1785}
1786
1787void
1788spa_set_log_state(spa_t *spa, spa_log_state_t state)
1789{
1790	spa->spa_log_state = state;
1791}
1792
1793boolean_t
1794spa_is_root(spa_t *spa)
1795{
1796	return (spa->spa_is_root);
1797}
1798
1799boolean_t
1800spa_writeable(spa_t *spa)
1801{
1802	return (!!(spa->spa_mode & FWRITE));
1803}
1804
1805int
1806spa_mode(spa_t *spa)
1807{
1808	return (spa->spa_mode);
1809}
1810
1811uint64_t
1812spa_bootfs(spa_t *spa)
1813{
1814	return (spa->spa_bootfs);
1815}
1816
1817uint64_t
1818spa_delegation(spa_t *spa)
1819{
1820	return (spa->spa_delegation);
1821}
1822
1823objset_t *
1824spa_meta_objset(spa_t *spa)
1825{
1826	return (spa->spa_meta_objset);
1827}
1828
1829enum zio_checksum
1830spa_dedup_checksum(spa_t *spa)
1831{
1832	return (spa->spa_dedup_checksum);
1833}
1834
1835/*
1836 * Reset pool scan stat per scan pass (or reboot).
1837 */
1838void
1839spa_scan_stat_init(spa_t *spa)
1840{
1841	/* data not stored on disk */
1842	spa->spa_scan_pass_start = gethrestime_sec();
1843	spa->spa_scan_pass_exam = 0;
1844	vdev_scan_stat_init(spa->spa_root_vdev);
1845}
1846
1847/*
1848 * Get scan stats for zpool status reports
1849 */
1850int
1851spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
1852{
1853	dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
1854
1855	if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
1856		return (SET_ERROR(ENOENT));
1857	bzero(ps, sizeof (pool_scan_stat_t));
1858
1859	/* data stored on disk */
1860	ps->pss_func = scn->scn_phys.scn_func;
1861	ps->pss_start_time = scn->scn_phys.scn_start_time;
1862	ps->pss_end_time = scn->scn_phys.scn_end_time;
1863	ps->pss_to_examine = scn->scn_phys.scn_to_examine;
1864	ps->pss_examined = scn->scn_phys.scn_examined;
1865	ps->pss_to_process = scn->scn_phys.scn_to_process;
1866	ps->pss_processed = scn->scn_phys.scn_processed;
1867	ps->pss_errors = scn->scn_phys.scn_errors;
1868	ps->pss_state = scn->scn_phys.scn_state;
1869
1870	/* data not stored on disk */
1871	ps->pss_pass_start = spa->spa_scan_pass_start;
1872	ps->pss_pass_exam = spa->spa_scan_pass_exam;
1873
1874	return (0);
1875}
1876
1877boolean_t
1878spa_debug_enabled(spa_t *spa)
1879{
1880	return (spa->spa_debug);
1881}
1882