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 * This should only be used as a last resort, as it typically results
251 * in leaked space, or worse.
252 */
253int zfs_recover = 0;
254
255/*
256 * Expiration time in milliseconds. This value has two meanings. First it is
257 * used to determine when the spa_deadman() logic should fire. By default the
258 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
259 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
260 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
261 * in a system panic.
262 */
263uint64_t zfs_deadman_synctime_ms = 1000000ULL;
264
265/*
266 * Check time in milliseconds. This defines the frequency at which we check
267 * for hung I/O.
268 */
269uint64_t zfs_deadman_checktime_ms = 5000ULL;
270
271/*
272 * Override the zfs deadman behavior via /etc/system. By default the
273 * deadman is enabled except on VMware and sparc deployments.
274 */
275int zfs_deadman_enabled = -1;
276
277/*
278 * The worst case is single-sector max-parity RAID-Z blocks, in which
279 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
280 * times the size; so just assume that.  Add to this the fact that
281 * we can have up to 3 DVAs per bp, and one more factor of 2 because
282 * the block may be dittoed with up to 3 DVAs by ddt_sync().  All together,
283 * the worst case is:
284 *     (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
285 */
286int spa_asize_inflation = 24;
287
288/*
289 * ==========================================================================
290 * SPA config locking
291 * ==========================================================================
292 */
293static void
294spa_config_lock_init(spa_t *spa)
295{
296	for (int i = 0; i < SCL_LOCKS; i++) {
297		spa_config_lock_t *scl = &spa->spa_config_lock[i];
298		mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
299		cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
300		refcount_create_untracked(&scl->scl_count);
301		scl->scl_writer = NULL;
302		scl->scl_write_wanted = 0;
303	}
304}
305
306static void
307spa_config_lock_destroy(spa_t *spa)
308{
309	for (int i = 0; i < SCL_LOCKS; i++) {
310		spa_config_lock_t *scl = &spa->spa_config_lock[i];
311		mutex_destroy(&scl->scl_lock);
312		cv_destroy(&scl->scl_cv);
313		refcount_destroy(&scl->scl_count);
314		ASSERT(scl->scl_writer == NULL);
315		ASSERT(scl->scl_write_wanted == 0);
316	}
317}
318
319int
320spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
321{
322	for (int i = 0; i < SCL_LOCKS; i++) {
323		spa_config_lock_t *scl = &spa->spa_config_lock[i];
324		if (!(locks & (1 << i)))
325			continue;
326		mutex_enter(&scl->scl_lock);
327		if (rw == RW_READER) {
328			if (scl->scl_writer || scl->scl_write_wanted) {
329				mutex_exit(&scl->scl_lock);
330				spa_config_exit(spa, locks ^ (1 << i), tag);
331				return (0);
332			}
333		} else {
334			ASSERT(scl->scl_writer != curthread);
335			if (!refcount_is_zero(&scl->scl_count)) {
336				mutex_exit(&scl->scl_lock);
337				spa_config_exit(spa, locks ^ (1 << i), tag);
338				return (0);
339			}
340			scl->scl_writer = curthread;
341		}
342		(void) refcount_add(&scl->scl_count, tag);
343		mutex_exit(&scl->scl_lock);
344	}
345	return (1);
346}
347
348void
349spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
350{
351	int wlocks_held = 0;
352
353	ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
354
355	for (int i = 0; i < SCL_LOCKS; i++) {
356		spa_config_lock_t *scl = &spa->spa_config_lock[i];
357		if (scl->scl_writer == curthread)
358			wlocks_held |= (1 << i);
359		if (!(locks & (1 << i)))
360			continue;
361		mutex_enter(&scl->scl_lock);
362		if (rw == RW_READER) {
363			while (scl->scl_writer || scl->scl_write_wanted) {
364				cv_wait(&scl->scl_cv, &scl->scl_lock);
365			}
366		} else {
367			ASSERT(scl->scl_writer != curthread);
368			while (!refcount_is_zero(&scl->scl_count)) {
369				scl->scl_write_wanted++;
370				cv_wait(&scl->scl_cv, &scl->scl_lock);
371				scl->scl_write_wanted--;
372			}
373			scl->scl_writer = curthread;
374		}
375		(void) refcount_add(&scl->scl_count, tag);
376		mutex_exit(&scl->scl_lock);
377	}
378	ASSERT(wlocks_held <= locks);
379}
380
381void
382spa_config_exit(spa_t *spa, int locks, void *tag)
383{
384	for (int i = SCL_LOCKS - 1; i >= 0; i--) {
385		spa_config_lock_t *scl = &spa->spa_config_lock[i];
386		if (!(locks & (1 << i)))
387			continue;
388		mutex_enter(&scl->scl_lock);
389		ASSERT(!refcount_is_zero(&scl->scl_count));
390		if (refcount_remove(&scl->scl_count, tag) == 0) {
391			ASSERT(scl->scl_writer == NULL ||
392			    scl->scl_writer == curthread);
393			scl->scl_writer = NULL;	/* OK in either case */
394			cv_broadcast(&scl->scl_cv);
395		}
396		mutex_exit(&scl->scl_lock);
397	}
398}
399
400int
401spa_config_held(spa_t *spa, int locks, krw_t rw)
402{
403	int locks_held = 0;
404
405	for (int i = 0; i < SCL_LOCKS; i++) {
406		spa_config_lock_t *scl = &spa->spa_config_lock[i];
407		if (!(locks & (1 << i)))
408			continue;
409		if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
410		    (rw == RW_WRITER && scl->scl_writer == curthread))
411			locks_held |= 1 << i;
412	}
413
414	return (locks_held);
415}
416
417/*
418 * ==========================================================================
419 * SPA namespace functions
420 * ==========================================================================
421 */
422
423/*
424 * Lookup the named spa_t in the AVL tree.  The spa_namespace_lock must be held.
425 * Returns NULL if no matching spa_t is found.
426 */
427spa_t *
428spa_lookup(const char *name)
429{
430	static spa_t search;	/* spa_t is large; don't allocate on stack */
431	spa_t *spa;
432	avl_index_t where;
433	char *cp;
434
435	ASSERT(MUTEX_HELD(&spa_namespace_lock));
436
437	(void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
438
439	/*
440	 * If it's a full dataset name, figure out the pool name and
441	 * just use that.
442	 */
443	cp = strpbrk(search.spa_name, "/@#");
444	if (cp != NULL)
445		*cp = '\0';
446
447	spa = avl_find(&spa_namespace_avl, &search, &where);
448
449	return (spa);
450}
451
452/*
453 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
454 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
455 * looking for potentially hung I/Os.
456 */
457void
458spa_deadman(void *arg)
459{
460	spa_t *spa = arg;
461
462	/*
463	 * Disable the deadman timer if the pool is suspended.
464	 */
465	if (spa_suspended(spa)) {
466		VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
467		return;
468	}
469
470	zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
471	    (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
472	    ++spa->spa_deadman_calls);
473	if (zfs_deadman_enabled)
474		vdev_deadman(spa->spa_root_vdev);
475}
476
477/*
478 * Create an uninitialized spa_t with the given name.  Requires
479 * spa_namespace_lock.  The caller must ensure that the spa_t doesn't already
480 * exist by calling spa_lookup() first.
481 */
482spa_t *
483spa_add(const char *name, nvlist_t *config, const char *altroot)
484{
485	spa_t *spa;
486	spa_config_dirent_t *dp;
487	cyc_handler_t hdlr;
488	cyc_time_t when;
489
490	ASSERT(MUTEX_HELD(&spa_namespace_lock));
491
492	spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
493
494	mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
495	mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
496	mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
497	mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
498	mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
499	mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
500	mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
501	mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
502	mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
503	mutex_init(&spa->spa_iokstat_lock, NULL, MUTEX_DEFAULT, NULL);
504
505	cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
506	cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
507	cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
508	cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
509
510	for (int t = 0; t < TXG_SIZE; t++)
511		bplist_create(&spa->spa_free_bplist[t]);
512
513	(void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
514	spa->spa_state = POOL_STATE_UNINITIALIZED;
515	spa->spa_freeze_txg = UINT64_MAX;
516	spa->spa_final_txg = UINT64_MAX;
517	spa->spa_load_max_txg = UINT64_MAX;
518	spa->spa_proc = &p0;
519	spa->spa_proc_state = SPA_PROC_NONE;
520
521	hdlr.cyh_func = spa_deadman;
522	hdlr.cyh_arg = spa;
523	hdlr.cyh_level = CY_LOW_LEVEL;
524
525	spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
526
527	/*
528	 * This determines how often we need to check for hung I/Os after
529	 * the cyclic has already fired. Since checking for hung I/Os is
530	 * an expensive operation we don't want to check too frequently.
531	 * Instead wait for 5 seconds before checking again.
532	 */
533	when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms);
534	when.cyt_when = CY_INFINITY;
535	mutex_enter(&cpu_lock);
536	spa->spa_deadman_cycid = cyclic_add(&hdlr, &when);
537	mutex_exit(&cpu_lock);
538
539	refcount_create(&spa->spa_refcount);
540	spa_config_lock_init(spa);
541
542	avl_add(&spa_namespace_avl, spa);
543
544	/*
545	 * Set the alternate root, if there is one.
546	 */
547	if (altroot) {
548		spa->spa_root = spa_strdup(altroot);
549		spa_active_count++;
550	}
551
552	/*
553	 * Every pool starts with the default cachefile
554	 */
555	list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
556	    offsetof(spa_config_dirent_t, scd_link));
557
558	dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
559	dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
560	list_insert_head(&spa->spa_config_list, dp);
561
562	VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
563	    KM_SLEEP) == 0);
564
565	if (config != NULL) {
566		nvlist_t *features;
567
568		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
569		    &features) == 0) {
570			VERIFY(nvlist_dup(features, &spa->spa_label_features,
571			    0) == 0);
572		}
573
574		VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
575	}
576
577	if (spa->spa_label_features == NULL) {
578		VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
579		    KM_SLEEP) == 0);
580	}
581
582	spa->spa_iokstat = kstat_create("zfs", 0, name,
583	    "disk", KSTAT_TYPE_IO, 1, 0);
584	if (spa->spa_iokstat) {
585		spa->spa_iokstat->ks_lock = &spa->spa_iokstat_lock;
586		kstat_install(spa->spa_iokstat);
587	}
588
589	spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0);
590
591	/*
592	 * As a pool is being created, treat all features as disabled by
593	 * setting SPA_FEATURE_DISABLED for all entries in the feature
594	 * refcount cache.
595	 */
596	for (int i = 0; i < SPA_FEATURES; i++) {
597		spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED;
598	}
599
600	return (spa);
601}
602
603/*
604 * Removes a spa_t from the namespace, freeing up any memory used.  Requires
605 * spa_namespace_lock.  This is called only after the spa_t has been closed and
606 * deactivated.
607 */
608void
609spa_remove(spa_t *spa)
610{
611	spa_config_dirent_t *dp;
612
613	ASSERT(MUTEX_HELD(&spa_namespace_lock));
614	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
615
616	nvlist_free(spa->spa_config_splitting);
617
618	avl_remove(&spa_namespace_avl, spa);
619	cv_broadcast(&spa_namespace_cv);
620
621	if (spa->spa_root) {
622		spa_strfree(spa->spa_root);
623		spa_active_count--;
624	}
625
626	while ((dp = list_head(&spa->spa_config_list)) != NULL) {
627		list_remove(&spa->spa_config_list, dp);
628		if (dp->scd_path != NULL)
629			spa_strfree(dp->scd_path);
630		kmem_free(dp, sizeof (spa_config_dirent_t));
631	}
632
633	list_destroy(&spa->spa_config_list);
634
635	nvlist_free(spa->spa_label_features);
636	nvlist_free(spa->spa_load_info);
637	spa_config_set(spa, NULL);
638
639	mutex_enter(&cpu_lock);
640	if (spa->spa_deadman_cycid != CYCLIC_NONE)
641		cyclic_remove(spa->spa_deadman_cycid);
642	mutex_exit(&cpu_lock);
643	spa->spa_deadman_cycid = CYCLIC_NONE;
644
645	refcount_destroy(&spa->spa_refcount);
646
647	spa_config_lock_destroy(spa);
648
649	kstat_delete(spa->spa_iokstat);
650	spa->spa_iokstat = NULL;
651
652	for (int t = 0; t < TXG_SIZE; t++)
653		bplist_destroy(&spa->spa_free_bplist[t]);
654
655	cv_destroy(&spa->spa_async_cv);
656	cv_destroy(&spa->spa_proc_cv);
657	cv_destroy(&spa->spa_scrub_io_cv);
658	cv_destroy(&spa->spa_suspend_cv);
659
660	mutex_destroy(&spa->spa_async_lock);
661	mutex_destroy(&spa->spa_errlist_lock);
662	mutex_destroy(&spa->spa_errlog_lock);
663	mutex_destroy(&spa->spa_history_lock);
664	mutex_destroy(&spa->spa_proc_lock);
665	mutex_destroy(&spa->spa_props_lock);
666	mutex_destroy(&spa->spa_scrub_lock);
667	mutex_destroy(&spa->spa_suspend_lock);
668	mutex_destroy(&spa->spa_vdev_top_lock);
669	mutex_destroy(&spa->spa_iokstat_lock);
670
671	kmem_free(spa, sizeof (spa_t));
672}
673
674/*
675 * Given a pool, return the next pool in the namespace, or NULL if there is
676 * none.  If 'prev' is NULL, return the first pool.
677 */
678spa_t *
679spa_next(spa_t *prev)
680{
681	ASSERT(MUTEX_HELD(&spa_namespace_lock));
682
683	if (prev)
684		return (AVL_NEXT(&spa_namespace_avl, prev));
685	else
686		return (avl_first(&spa_namespace_avl));
687}
688
689/*
690 * ==========================================================================
691 * SPA refcount functions
692 * ==========================================================================
693 */
694
695/*
696 * Add a reference to the given spa_t.  Must have at least one reference, or
697 * have the namespace lock held.
698 */
699void
700spa_open_ref(spa_t *spa, void *tag)
701{
702	ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
703	    MUTEX_HELD(&spa_namespace_lock));
704	(void) refcount_add(&spa->spa_refcount, tag);
705}
706
707/*
708 * Remove a reference to the given spa_t.  Must have at least one reference, or
709 * have the namespace lock held.
710 */
711void
712spa_close(spa_t *spa, void *tag)
713{
714	ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
715	    MUTEX_HELD(&spa_namespace_lock));
716	(void) refcount_remove(&spa->spa_refcount, tag);
717}
718
719/*
720 * Check to see if the spa refcount is zero.  Must be called with
721 * spa_namespace_lock held.  We really compare against spa_minref, which is the
722 * number of references acquired when opening a pool
723 */
724boolean_t
725spa_refcount_zero(spa_t *spa)
726{
727	ASSERT(MUTEX_HELD(&spa_namespace_lock));
728
729	return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
730}
731
732/*
733 * ==========================================================================
734 * SPA spare and l2cache tracking
735 * ==========================================================================
736 */
737
738/*
739 * Hot spares and cache devices are tracked using the same code below,
740 * for 'auxiliary' devices.
741 */
742
743typedef struct spa_aux {
744	uint64_t	aux_guid;
745	uint64_t	aux_pool;
746	avl_node_t	aux_avl;
747	int		aux_count;
748} spa_aux_t;
749
750static int
751spa_aux_compare(const void *a, const void *b)
752{
753	const spa_aux_t *sa = a;
754	const spa_aux_t *sb = b;
755
756	if (sa->aux_guid < sb->aux_guid)
757		return (-1);
758	else if (sa->aux_guid > sb->aux_guid)
759		return (1);
760	else
761		return (0);
762}
763
764void
765spa_aux_add(vdev_t *vd, avl_tree_t *avl)
766{
767	avl_index_t where;
768	spa_aux_t search;
769	spa_aux_t *aux;
770
771	search.aux_guid = vd->vdev_guid;
772	if ((aux = avl_find(avl, &search, &where)) != NULL) {
773		aux->aux_count++;
774	} else {
775		aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
776		aux->aux_guid = vd->vdev_guid;
777		aux->aux_count = 1;
778		avl_insert(avl, aux, where);
779	}
780}
781
782void
783spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
784{
785	spa_aux_t search;
786	spa_aux_t *aux;
787	avl_index_t where;
788
789	search.aux_guid = vd->vdev_guid;
790	aux = avl_find(avl, &search, &where);
791
792	ASSERT(aux != NULL);
793
794	if (--aux->aux_count == 0) {
795		avl_remove(avl, aux);
796		kmem_free(aux, sizeof (spa_aux_t));
797	} else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
798		aux->aux_pool = 0ULL;
799	}
800}
801
802boolean_t
803spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
804{
805	spa_aux_t search, *found;
806
807	search.aux_guid = guid;
808	found = avl_find(avl, &search, NULL);
809
810	if (pool) {
811		if (found)
812			*pool = found->aux_pool;
813		else
814			*pool = 0ULL;
815	}
816
817	if (refcnt) {
818		if (found)
819			*refcnt = found->aux_count;
820		else
821			*refcnt = 0;
822	}
823
824	return (found != NULL);
825}
826
827void
828spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
829{
830	spa_aux_t search, *found;
831	avl_index_t where;
832
833	search.aux_guid = vd->vdev_guid;
834	found = avl_find(avl, &search, &where);
835	ASSERT(found != NULL);
836	ASSERT(found->aux_pool == 0ULL);
837
838	found->aux_pool = spa_guid(vd->vdev_spa);
839}
840
841/*
842 * Spares are tracked globally due to the following constraints:
843 *
844 * 	- A spare may be part of multiple pools.
845 * 	- A spare may be added to a pool even if it's actively in use within
846 *	  another pool.
847 * 	- A spare in use in any pool can only be the source of a replacement if
848 *	  the target is a spare in the same pool.
849 *
850 * We keep track of all spares on the system through the use of a reference
851 * counted AVL tree.  When a vdev is added as a spare, or used as a replacement
852 * spare, then we bump the reference count in the AVL tree.  In addition, we set
853 * the 'vdev_isspare' member to indicate that the device is a spare (active or
854 * inactive).  When a spare is made active (used to replace a device in the
855 * pool), we also keep track of which pool its been made a part of.
856 *
857 * The 'spa_spare_lock' protects the AVL tree.  These functions are normally
858 * called under the spa_namespace lock as part of vdev reconfiguration.  The
859 * separate spare lock exists for the status query path, which does not need to
860 * be completely consistent with respect to other vdev configuration changes.
861 */
862
863static int
864spa_spare_compare(const void *a, const void *b)
865{
866	return (spa_aux_compare(a, b));
867}
868
869void
870spa_spare_add(vdev_t *vd)
871{
872	mutex_enter(&spa_spare_lock);
873	ASSERT(!vd->vdev_isspare);
874	spa_aux_add(vd, &spa_spare_avl);
875	vd->vdev_isspare = B_TRUE;
876	mutex_exit(&spa_spare_lock);
877}
878
879void
880spa_spare_remove(vdev_t *vd)
881{
882	mutex_enter(&spa_spare_lock);
883	ASSERT(vd->vdev_isspare);
884	spa_aux_remove(vd, &spa_spare_avl);
885	vd->vdev_isspare = B_FALSE;
886	mutex_exit(&spa_spare_lock);
887}
888
889boolean_t
890spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
891{
892	boolean_t found;
893
894	mutex_enter(&spa_spare_lock);
895	found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
896	mutex_exit(&spa_spare_lock);
897
898	return (found);
899}
900
901void
902spa_spare_activate(vdev_t *vd)
903{
904	mutex_enter(&spa_spare_lock);
905	ASSERT(vd->vdev_isspare);
906	spa_aux_activate(vd, &spa_spare_avl);
907	mutex_exit(&spa_spare_lock);
908}
909
910/*
911 * Level 2 ARC devices are tracked globally for the same reasons as spares.
912 * Cache devices currently only support one pool per cache device, and so
913 * for these devices the aux reference count is currently unused beyond 1.
914 */
915
916static int
917spa_l2cache_compare(const void *a, const void *b)
918{
919	return (spa_aux_compare(a, b));
920}
921
922void
923spa_l2cache_add(vdev_t *vd)
924{
925	mutex_enter(&spa_l2cache_lock);
926	ASSERT(!vd->vdev_isl2cache);
927	spa_aux_add(vd, &spa_l2cache_avl);
928	vd->vdev_isl2cache = B_TRUE;
929	mutex_exit(&spa_l2cache_lock);
930}
931
932void
933spa_l2cache_remove(vdev_t *vd)
934{
935	mutex_enter(&spa_l2cache_lock);
936	ASSERT(vd->vdev_isl2cache);
937	spa_aux_remove(vd, &spa_l2cache_avl);
938	vd->vdev_isl2cache = B_FALSE;
939	mutex_exit(&spa_l2cache_lock);
940}
941
942boolean_t
943spa_l2cache_exists(uint64_t guid, uint64_t *pool)
944{
945	boolean_t found;
946
947	mutex_enter(&spa_l2cache_lock);
948	found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
949	mutex_exit(&spa_l2cache_lock);
950
951	return (found);
952}
953
954void
955spa_l2cache_activate(vdev_t *vd)
956{
957	mutex_enter(&spa_l2cache_lock);
958	ASSERT(vd->vdev_isl2cache);
959	spa_aux_activate(vd, &spa_l2cache_avl);
960	mutex_exit(&spa_l2cache_lock);
961}
962
963/*
964 * ==========================================================================
965 * SPA vdev locking
966 * ==========================================================================
967 */
968
969/*
970 * Lock the given spa_t for the purpose of adding or removing a vdev.
971 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
972 * It returns the next transaction group for the spa_t.
973 */
974uint64_t
975spa_vdev_enter(spa_t *spa)
976{
977	mutex_enter(&spa->spa_vdev_top_lock);
978	mutex_enter(&spa_namespace_lock);
979	return (spa_vdev_config_enter(spa));
980}
981
982/*
983 * Internal implementation for spa_vdev_enter().  Used when a vdev
984 * operation requires multiple syncs (i.e. removing a device) while
985 * keeping the spa_namespace_lock held.
986 */
987uint64_t
988spa_vdev_config_enter(spa_t *spa)
989{
990	ASSERT(MUTEX_HELD(&spa_namespace_lock));
991
992	spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
993
994	return (spa_last_synced_txg(spa) + 1);
995}
996
997/*
998 * Used in combination with spa_vdev_config_enter() to allow the syncing
999 * of multiple transactions without releasing the spa_namespace_lock.
1000 */
1001void
1002spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
1003{
1004	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1005
1006	int config_changed = B_FALSE;
1007
1008	ASSERT(txg > spa_last_synced_txg(spa));
1009
1010	spa->spa_pending_vdev = NULL;
1011
1012	/*
1013	 * Reassess the DTLs.
1014	 */
1015	vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
1016
1017	if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
1018		config_changed = B_TRUE;
1019		spa->spa_config_generation++;
1020	}
1021
1022	/*
1023	 * Verify the metaslab classes.
1024	 */
1025	ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
1026	ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
1027
1028	spa_config_exit(spa, SCL_ALL, spa);
1029
1030	/*
1031	 * Panic the system if the specified tag requires it.  This
1032	 * is useful for ensuring that configurations are updated
1033	 * transactionally.
1034	 */
1035	if (zio_injection_enabled)
1036		zio_handle_panic_injection(spa, tag, 0);
1037
1038	/*
1039	 * Note: this txg_wait_synced() is important because it ensures
1040	 * that there won't be more than one config change per txg.
1041	 * This allows us to use the txg as the generation number.
1042	 */
1043	if (error == 0)
1044		txg_wait_synced(spa->spa_dsl_pool, txg);
1045
1046	if (vd != NULL) {
1047		ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
1048		spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1049		vdev_free(vd);
1050		spa_config_exit(spa, SCL_ALL, spa);
1051	}
1052
1053	/*
1054	 * If the config changed, update the config cache.
1055	 */
1056	if (config_changed)
1057		spa_config_sync(spa, B_FALSE, B_TRUE);
1058}
1059
1060/*
1061 * Unlock the spa_t after adding or removing a vdev.  Besides undoing the
1062 * locking of spa_vdev_enter(), we also want make sure the transactions have
1063 * synced to disk, and then update the global configuration cache with the new
1064 * information.
1065 */
1066int
1067spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1068{
1069	spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1070	mutex_exit(&spa_namespace_lock);
1071	mutex_exit(&spa->spa_vdev_top_lock);
1072
1073	return (error);
1074}
1075
1076/*
1077 * Lock the given spa_t for the purpose of changing vdev state.
1078 */
1079void
1080spa_vdev_state_enter(spa_t *spa, int oplocks)
1081{
1082	int locks = SCL_STATE_ALL | oplocks;
1083
1084	/*
1085	 * Root pools may need to read of the underlying devfs filesystem
1086	 * when opening up a vdev.  Unfortunately if we're holding the
1087	 * SCL_ZIO lock it will result in a deadlock when we try to issue
1088	 * the read from the root filesystem.  Instead we "prefetch"
1089	 * the associated vnodes that we need prior to opening the
1090	 * underlying devices and cache them so that we can prevent
1091	 * any I/O when we are doing the actual open.
1092	 */
1093	if (spa_is_root(spa)) {
1094		int low = locks & ~(SCL_ZIO - 1);
1095		int high = locks & ~low;
1096
1097		spa_config_enter(spa, high, spa, RW_WRITER);
1098		vdev_hold(spa->spa_root_vdev);
1099		spa_config_enter(spa, low, spa, RW_WRITER);
1100	} else {
1101		spa_config_enter(spa, locks, spa, RW_WRITER);
1102	}
1103	spa->spa_vdev_locks = locks;
1104}
1105
1106int
1107spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1108{
1109	boolean_t config_changed = B_FALSE;
1110
1111	if (vd != NULL || error == 0)
1112		vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1113		    0, 0, B_FALSE);
1114
1115	if (vd != NULL) {
1116		vdev_state_dirty(vd->vdev_top);
1117		config_changed = B_TRUE;
1118		spa->spa_config_generation++;
1119	}
1120
1121	if (spa_is_root(spa))
1122		vdev_rele(spa->spa_root_vdev);
1123
1124	ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1125	spa_config_exit(spa, spa->spa_vdev_locks, spa);
1126
1127	/*
1128	 * If anything changed, wait for it to sync.  This ensures that,
1129	 * from the system administrator's perspective, zpool(1M) commands
1130	 * are synchronous.  This is important for things like zpool offline:
1131	 * when the command completes, you expect no further I/O from ZFS.
1132	 */
1133	if (vd != NULL)
1134		txg_wait_synced(spa->spa_dsl_pool, 0);
1135
1136	/*
1137	 * If the config changed, update the config cache.
1138	 */
1139	if (config_changed) {
1140		mutex_enter(&spa_namespace_lock);
1141		spa_config_sync(spa, B_FALSE, B_TRUE);
1142		mutex_exit(&spa_namespace_lock);
1143	}
1144
1145	return (error);
1146}
1147
1148/*
1149 * ==========================================================================
1150 * Miscellaneous functions
1151 * ==========================================================================
1152 */
1153
1154void
1155spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx)
1156{
1157	if (!nvlist_exists(spa->spa_label_features, feature)) {
1158		fnvlist_add_boolean(spa->spa_label_features, feature);
1159		/*
1160		 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1161		 * dirty the vdev config because lock SCL_CONFIG is not held.
1162		 * Thankfully, in this case we don't need to dirty the config
1163		 * because it will be written out anyway when we finish
1164		 * creating the pool.
1165		 */
1166		if (tx->tx_txg != TXG_INITIAL)
1167			vdev_config_dirty(spa->spa_root_vdev);
1168	}
1169}
1170
1171void
1172spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1173{
1174	if (nvlist_remove_all(spa->spa_label_features, feature) == 0)
1175		vdev_config_dirty(spa->spa_root_vdev);
1176}
1177
1178/*
1179 * Rename a spa_t.
1180 */
1181int
1182spa_rename(const char *name, const char *newname)
1183{
1184	spa_t *spa;
1185	int err;
1186
1187	/*
1188	 * Lookup the spa_t and grab the config lock for writing.  We need to
1189	 * actually open the pool so that we can sync out the necessary labels.
1190	 * It's OK to call spa_open() with the namespace lock held because we
1191	 * allow recursive calls for other reasons.
1192	 */
1193	mutex_enter(&spa_namespace_lock);
1194	if ((err = spa_open(name, &spa, FTAG)) != 0) {
1195		mutex_exit(&spa_namespace_lock);
1196		return (err);
1197	}
1198
1199	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1200
1201	avl_remove(&spa_namespace_avl, spa);
1202	(void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1203	avl_add(&spa_namespace_avl, spa);
1204
1205	/*
1206	 * Sync all labels to disk with the new names by marking the root vdev
1207	 * dirty and waiting for it to sync.  It will pick up the new pool name
1208	 * during the sync.
1209	 */
1210	vdev_config_dirty(spa->spa_root_vdev);
1211
1212	spa_config_exit(spa, SCL_ALL, FTAG);
1213
1214	txg_wait_synced(spa->spa_dsl_pool, 0);
1215
1216	/*
1217	 * Sync the updated config cache.
1218	 */
1219	spa_config_sync(spa, B_FALSE, B_TRUE);
1220
1221	spa_close(spa, FTAG);
1222
1223	mutex_exit(&spa_namespace_lock);
1224
1225	return (0);
1226}
1227
1228/*
1229 * Return the spa_t associated with given pool_guid, if it exists.  If
1230 * device_guid is non-zero, determine whether the pool exists *and* contains
1231 * a device with the specified device_guid.
1232 */
1233spa_t *
1234spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1235{
1236	spa_t *spa;
1237	avl_tree_t *t = &spa_namespace_avl;
1238
1239	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1240
1241	for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1242		if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1243			continue;
1244		if (spa->spa_root_vdev == NULL)
1245			continue;
1246		if (spa_guid(spa) == pool_guid) {
1247			if (device_guid == 0)
1248				break;
1249
1250			if (vdev_lookup_by_guid(spa->spa_root_vdev,
1251			    device_guid) != NULL)
1252				break;
1253
1254			/*
1255			 * Check any devices we may be in the process of adding.
1256			 */
1257			if (spa->spa_pending_vdev) {
1258				if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1259				    device_guid) != NULL)
1260					break;
1261			}
1262		}
1263	}
1264
1265	return (spa);
1266}
1267
1268/*
1269 * Determine whether a pool with the given pool_guid exists.
1270 */
1271boolean_t
1272spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1273{
1274	return (spa_by_guid(pool_guid, device_guid) != NULL);
1275}
1276
1277char *
1278spa_strdup(const char *s)
1279{
1280	size_t len;
1281	char *new;
1282
1283	len = strlen(s);
1284	new = kmem_alloc(len + 1, KM_SLEEP);
1285	bcopy(s, new, len);
1286	new[len] = '\0';
1287
1288	return (new);
1289}
1290
1291void
1292spa_strfree(char *s)
1293{
1294	kmem_free(s, strlen(s) + 1);
1295}
1296
1297uint64_t
1298spa_get_random(uint64_t range)
1299{
1300	uint64_t r;
1301
1302	ASSERT(range != 0);
1303
1304	(void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1305
1306	return (r % range);
1307}
1308
1309uint64_t
1310spa_generate_guid(spa_t *spa)
1311{
1312	uint64_t guid = spa_get_random(-1ULL);
1313
1314	if (spa != NULL) {
1315		while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1316			guid = spa_get_random(-1ULL);
1317	} else {
1318		while (guid == 0 || spa_guid_exists(guid, 0))
1319			guid = spa_get_random(-1ULL);
1320	}
1321
1322	return (guid);
1323}
1324
1325void
1326snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp)
1327{
1328	char type[256];
1329	char *checksum = NULL;
1330	char *compress = NULL;
1331
1332	if (bp != NULL) {
1333		if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1334			dmu_object_byteswap_t bswap =
1335			    DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1336			(void) snprintf(type, sizeof (type), "bswap %s %s",
1337			    DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1338			    "metadata" : "data",
1339			    dmu_ot_byteswap[bswap].ob_name);
1340		} else {
1341			(void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1342			    sizeof (type));
1343		}
1344		checksum = zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1345		compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1346	}
1347
1348	SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum,
1349	    compress);
1350}
1351
1352void
1353spa_freeze(spa_t *spa)
1354{
1355	uint64_t freeze_txg = 0;
1356
1357	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1358	if (spa->spa_freeze_txg == UINT64_MAX) {
1359		freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1360		spa->spa_freeze_txg = freeze_txg;
1361	}
1362	spa_config_exit(spa, SCL_ALL, FTAG);
1363	if (freeze_txg != 0)
1364		txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1365}
1366
1367void
1368zfs_panic_recover(const char *fmt, ...)
1369{
1370	va_list adx;
1371
1372	va_start(adx, fmt);
1373	vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1374	va_end(adx);
1375}
1376
1377/*
1378 * This is a stripped-down version of strtoull, suitable only for converting
1379 * lowercase hexadecimal numbers that don't overflow.
1380 */
1381uint64_t
1382strtonum(const char *str, char **nptr)
1383{
1384	uint64_t val = 0;
1385	char c;
1386	int digit;
1387
1388	while ((c = *str) != '\0') {
1389		if (c >= '0' && c <= '9')
1390			digit = c - '0';
1391		else if (c >= 'a' && c <= 'f')
1392			digit = 10 + c - 'a';
1393		else
1394			break;
1395
1396		val *= 16;
1397		val += digit;
1398
1399		str++;
1400	}
1401
1402	if (nptr)
1403		*nptr = (char *)str;
1404
1405	return (val);
1406}
1407
1408/*
1409 * ==========================================================================
1410 * Accessor functions
1411 * ==========================================================================
1412 */
1413
1414boolean_t
1415spa_shutting_down(spa_t *spa)
1416{
1417	return (spa->spa_async_suspended);
1418}
1419
1420dsl_pool_t *
1421spa_get_dsl(spa_t *spa)
1422{
1423	return (spa->spa_dsl_pool);
1424}
1425
1426boolean_t
1427spa_is_initializing(spa_t *spa)
1428{
1429	return (spa->spa_is_initializing);
1430}
1431
1432blkptr_t *
1433spa_get_rootblkptr(spa_t *spa)
1434{
1435	return (&spa->spa_ubsync.ub_rootbp);
1436}
1437
1438void
1439spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1440{
1441	spa->spa_uberblock.ub_rootbp = *bp;
1442}
1443
1444void
1445spa_altroot(spa_t *spa, char *buf, size_t buflen)
1446{
1447	if (spa->spa_root == NULL)
1448		buf[0] = '\0';
1449	else
1450		(void) strncpy(buf, spa->spa_root, buflen);
1451}
1452
1453int
1454spa_sync_pass(spa_t *spa)
1455{
1456	return (spa->spa_sync_pass);
1457}
1458
1459char *
1460spa_name(spa_t *spa)
1461{
1462	return (spa->spa_name);
1463}
1464
1465uint64_t
1466spa_guid(spa_t *spa)
1467{
1468	dsl_pool_t *dp = spa_get_dsl(spa);
1469	uint64_t guid;
1470
1471	/*
1472	 * If we fail to parse the config during spa_load(), we can go through
1473	 * the error path (which posts an ereport) and end up here with no root
1474	 * vdev.  We stash the original pool guid in 'spa_config_guid' to handle
1475	 * this case.
1476	 */
1477	if (spa->spa_root_vdev == NULL)
1478		return (spa->spa_config_guid);
1479
1480	guid = spa->spa_last_synced_guid != 0 ?
1481	    spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1482
1483	/*
1484	 * Return the most recently synced out guid unless we're
1485	 * in syncing context.
1486	 */
1487	if (dp && dsl_pool_sync_context(dp))
1488		return (spa->spa_root_vdev->vdev_guid);
1489	else
1490		return (guid);
1491}
1492
1493uint64_t
1494spa_load_guid(spa_t *spa)
1495{
1496	/*
1497	 * This is a GUID that exists solely as a reference for the
1498	 * purposes of the arc.  It is generated at load time, and
1499	 * is never written to persistent storage.
1500	 */
1501	return (spa->spa_load_guid);
1502}
1503
1504uint64_t
1505spa_last_synced_txg(spa_t *spa)
1506{
1507	return (spa->spa_ubsync.ub_txg);
1508}
1509
1510uint64_t
1511spa_first_txg(spa_t *spa)
1512{
1513	return (spa->spa_first_txg);
1514}
1515
1516uint64_t
1517spa_syncing_txg(spa_t *spa)
1518{
1519	return (spa->spa_syncing_txg);
1520}
1521
1522pool_state_t
1523spa_state(spa_t *spa)
1524{
1525	return (spa->spa_state);
1526}
1527
1528spa_load_state_t
1529spa_load_state(spa_t *spa)
1530{
1531	return (spa->spa_load_state);
1532}
1533
1534uint64_t
1535spa_freeze_txg(spa_t *spa)
1536{
1537	return (spa->spa_freeze_txg);
1538}
1539
1540/* ARGSUSED */
1541uint64_t
1542spa_get_asize(spa_t *spa, uint64_t lsize)
1543{
1544	return (lsize * spa_asize_inflation);
1545}
1546
1547uint64_t
1548spa_get_dspace(spa_t *spa)
1549{
1550	return (spa->spa_dspace);
1551}
1552
1553void
1554spa_update_dspace(spa_t *spa)
1555{
1556	spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1557	    ddt_get_dedup_dspace(spa);
1558}
1559
1560/*
1561 * Return the failure mode that has been set to this pool. The default
1562 * behavior will be to block all I/Os when a complete failure occurs.
1563 */
1564uint8_t
1565spa_get_failmode(spa_t *spa)
1566{
1567	return (spa->spa_failmode);
1568}
1569
1570boolean_t
1571spa_suspended(spa_t *spa)
1572{
1573	return (spa->spa_suspended);
1574}
1575
1576uint64_t
1577spa_version(spa_t *spa)
1578{
1579	return (spa->spa_ubsync.ub_version);
1580}
1581
1582boolean_t
1583spa_deflate(spa_t *spa)
1584{
1585	return (spa->spa_deflate);
1586}
1587
1588metaslab_class_t *
1589spa_normal_class(spa_t *spa)
1590{
1591	return (spa->spa_normal_class);
1592}
1593
1594metaslab_class_t *
1595spa_log_class(spa_t *spa)
1596{
1597	return (spa->spa_log_class);
1598}
1599
1600int
1601spa_max_replication(spa_t *spa)
1602{
1603	/*
1604	 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1605	 * handle BPs with more than one DVA allocated.  Set our max
1606	 * replication level accordingly.
1607	 */
1608	if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1609		return (1);
1610	return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1611}
1612
1613int
1614spa_prev_software_version(spa_t *spa)
1615{
1616	return (spa->spa_prev_software_version);
1617}
1618
1619uint64_t
1620spa_deadman_synctime(spa_t *spa)
1621{
1622	return (spa->spa_deadman_synctime);
1623}
1624
1625uint64_t
1626dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1627{
1628	uint64_t asize = DVA_GET_ASIZE(dva);
1629	uint64_t dsize = asize;
1630
1631	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1632
1633	if (asize != 0 && spa->spa_deflate) {
1634		vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1635		dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1636	}
1637
1638	return (dsize);
1639}
1640
1641uint64_t
1642bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1643{
1644	uint64_t dsize = 0;
1645
1646	for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1647		dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1648
1649	return (dsize);
1650}
1651
1652uint64_t
1653bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1654{
1655	uint64_t dsize = 0;
1656
1657	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1658
1659	for (int d = 0; d < SPA_DVAS_PER_BP; d++)
1660		dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1661
1662	spa_config_exit(spa, SCL_VDEV, FTAG);
1663
1664	return (dsize);
1665}
1666
1667/*
1668 * ==========================================================================
1669 * Initialization and Termination
1670 * ==========================================================================
1671 */
1672
1673static int
1674spa_name_compare(const void *a1, const void *a2)
1675{
1676	const spa_t *s1 = a1;
1677	const spa_t *s2 = a2;
1678	int s;
1679
1680	s = strcmp(s1->spa_name, s2->spa_name);
1681	if (s > 0)
1682		return (1);
1683	if (s < 0)
1684		return (-1);
1685	return (0);
1686}
1687
1688int
1689spa_busy(void)
1690{
1691	return (spa_active_count);
1692}
1693
1694void
1695spa_boot_init()
1696{
1697	spa_config_load();
1698}
1699
1700void
1701spa_init(int mode)
1702{
1703	mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1704	mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1705	mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1706	cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1707
1708	avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1709	    offsetof(spa_t, spa_avl));
1710
1711	avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1712	    offsetof(spa_aux_t, aux_avl));
1713
1714	avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1715	    offsetof(spa_aux_t, aux_avl));
1716
1717	spa_mode_global = mode;
1718
1719#ifdef _KERNEL
1720	spa_arch_init();
1721#else
1722	if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
1723		arc_procfd = open("/proc/self/ctl", O_WRONLY);
1724		if (arc_procfd == -1) {
1725			perror("could not enable watchpoints: "
1726			    "opening /proc/self/ctl failed: ");
1727		} else {
1728			arc_watch = B_TRUE;
1729		}
1730	}
1731#endif
1732
1733	refcount_init();
1734	unique_init();
1735	range_tree_init();
1736	zio_init();
1737	dmu_init();
1738	zil_init();
1739	vdev_cache_stat_init();
1740	zfs_prop_init();
1741	zpool_prop_init();
1742	zpool_feature_init();
1743	spa_config_load();
1744	l2arc_start();
1745}
1746
1747void
1748spa_fini(void)
1749{
1750	l2arc_stop();
1751
1752	spa_evict_all();
1753
1754	vdev_cache_stat_fini();
1755	zil_fini();
1756	dmu_fini();
1757	zio_fini();
1758	range_tree_fini();
1759	unique_fini();
1760	refcount_fini();
1761
1762	avl_destroy(&spa_namespace_avl);
1763	avl_destroy(&spa_spare_avl);
1764	avl_destroy(&spa_l2cache_avl);
1765
1766	cv_destroy(&spa_namespace_cv);
1767	mutex_destroy(&spa_namespace_lock);
1768	mutex_destroy(&spa_spare_lock);
1769	mutex_destroy(&spa_l2cache_lock);
1770}
1771
1772/*
1773 * Return whether this pool has slogs. No locking needed.
1774 * It's not a problem if the wrong answer is returned as it's only for
1775 * performance and not correctness
1776 */
1777boolean_t
1778spa_has_slogs(spa_t *spa)
1779{
1780	return (spa->spa_log_class->mc_rotor != NULL);
1781}
1782
1783spa_log_state_t
1784spa_get_log_state(spa_t *spa)
1785{
1786	return (spa->spa_log_state);
1787}
1788
1789void
1790spa_set_log_state(spa_t *spa, spa_log_state_t state)
1791{
1792	spa->spa_log_state = state;
1793}
1794
1795boolean_t
1796spa_is_root(spa_t *spa)
1797{
1798	return (spa->spa_is_root);
1799}
1800
1801boolean_t
1802spa_writeable(spa_t *spa)
1803{
1804	return (!!(spa->spa_mode & FWRITE));
1805}
1806
1807int
1808spa_mode(spa_t *spa)
1809{
1810	return (spa->spa_mode);
1811}
1812
1813uint64_t
1814spa_bootfs(spa_t *spa)
1815{
1816	return (spa->spa_bootfs);
1817}
1818
1819uint64_t
1820spa_delegation(spa_t *spa)
1821{
1822	return (spa->spa_delegation);
1823}
1824
1825objset_t *
1826spa_meta_objset(spa_t *spa)
1827{
1828	return (spa->spa_meta_objset);
1829}
1830
1831enum zio_checksum
1832spa_dedup_checksum(spa_t *spa)
1833{
1834	return (spa->spa_dedup_checksum);
1835}
1836
1837/*
1838 * Reset pool scan stat per scan pass (or reboot).
1839 */
1840void
1841spa_scan_stat_init(spa_t *spa)
1842{
1843	/* data not stored on disk */
1844	spa->spa_scan_pass_start = gethrestime_sec();
1845	spa->spa_scan_pass_exam = 0;
1846	vdev_scan_stat_init(spa->spa_root_vdev);
1847}
1848
1849/*
1850 * Get scan stats for zpool status reports
1851 */
1852int
1853spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
1854{
1855	dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
1856
1857	if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
1858		return (SET_ERROR(ENOENT));
1859	bzero(ps, sizeof (pool_scan_stat_t));
1860
1861	/* data stored on disk */
1862	ps->pss_func = scn->scn_phys.scn_func;
1863	ps->pss_start_time = scn->scn_phys.scn_start_time;
1864	ps->pss_end_time = scn->scn_phys.scn_end_time;
1865	ps->pss_to_examine = scn->scn_phys.scn_to_examine;
1866	ps->pss_examined = scn->scn_phys.scn_examined;
1867	ps->pss_to_process = scn->scn_phys.scn_to_process;
1868	ps->pss_processed = scn->scn_phys.scn_processed;
1869	ps->pss_errors = scn->scn_phys.scn_errors;
1870	ps->pss_state = scn->scn_phys.scn_state;
1871
1872	/* data not stored on disk */
1873	ps->pss_pass_start = spa->spa_scan_pass_start;
1874	ps->pss_pass_exam = spa->spa_scan_pass_exam;
1875
1876	return (0);
1877}
1878
1879boolean_t
1880spa_debug_enabled(spa_t *spa)
1881{
1882	return (spa->spa_debug);
1883}
1884