spa_misc.c revision 12a8814c13fbb1d6d58616cf090ea5815dc107f9
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) 2011, 2018 by Delphix. All rights reserved.
24 * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
25 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
26 * Copyright 2013 Saso Kiselkov. All rights reserved.
27 * Copyright (c) 2014 Integros [integros.com]
28 * Copyright (c) 2017 Datto Inc.
29 * Copyright (c) 2017, Intel Corporation.
30 */
31
32#include <sys/zfs_context.h>
33#include <sys/spa_impl.h>
34#include <sys/spa_boot.h>
35#include <sys/zio.h>
36#include <sys/zio_checksum.h>
37#include <sys/zio_compress.h>
38#include <sys/dmu.h>
39#include <sys/dmu_tx.h>
40#include <sys/zap.h>
41#include <sys/zil.h>
42#include <sys/vdev_impl.h>
43#include <sys/vdev_initialize.h>
44#include <sys/metaslab.h>
45#include <sys/uberblock_impl.h>
46#include <sys/txg.h>
47#include <sys/avl.h>
48#include <sys/unique.h>
49#include <sys/dsl_pool.h>
50#include <sys/dsl_dir.h>
51#include <sys/dsl_prop.h>
52#include <sys/dsl_scan.h>
53#include <sys/fs/zfs.h>
54#include <sys/metaslab_impl.h>
55#include <sys/arc.h>
56#include <sys/ddt.h>
57#include "zfs_prop.h"
58#include <sys/zfeature.h>
59
60/*
61 * SPA locking
62 *
63 * There are four basic locks for managing spa_t structures:
64 *
65 * spa_namespace_lock (global mutex)
66 *
67 *	This lock must be acquired to do any of the following:
68 *
69 *		- Lookup a spa_t by name
70 *		- Add or remove a spa_t from the namespace
71 *		- Increase spa_refcount from non-zero
72 *		- Check if spa_refcount is zero
73 *		- Rename a spa_t
74 *		- add/remove/attach/detach devices
75 *		- Held for the duration of create/destroy/import/export
76 *
77 *	It does not need to handle recursion.  A create or destroy may
78 *	reference objects (files or zvols) in other pools, but by
79 *	definition they must have an existing reference, and will never need
80 *	to lookup a spa_t by name.
81 *
82 * spa_refcount (per-spa zfs_refcount_t protected by mutex)
83 *
84 *	This reference count keep track of any active users of the spa_t.  The
85 *	spa_t cannot be destroyed or freed while this is non-zero.  Internally,
86 *	the refcount is never really 'zero' - opening a pool implicitly keeps
87 *	some references in the DMU.  Internally we check against spa_minref, but
88 *	present the image of a zero/non-zero value to consumers.
89 *
90 * spa_config_lock[] (per-spa array of rwlocks)
91 *
92 *	This protects the spa_t from config changes, and must be held in
93 *	the following circumstances:
94 *
95 *		- RW_READER to perform I/O to the spa
96 *		- RW_WRITER to change the vdev config
97 *
98 * The locking order is fairly straightforward:
99 *
100 *		spa_namespace_lock	->	spa_refcount
101 *
102 *	The namespace lock must be acquired to increase the refcount from 0
103 *	or to check if it is zero.
104 *
105 *		spa_refcount		->	spa_config_lock[]
106 *
107 *	There must be at least one valid reference on the spa_t to acquire
108 *	the config lock.
109 *
110 *		spa_namespace_lock	->	spa_config_lock[]
111 *
112 *	The namespace lock must always be taken before the config lock.
113 *
114 *
115 * The spa_namespace_lock can be acquired directly and is globally visible.
116 *
117 * The namespace is manipulated using the following functions, all of which
118 * require the spa_namespace_lock to be held.
119 *
120 *	spa_lookup()		Lookup a spa_t by name.
121 *
122 *	spa_add()		Create a new spa_t in the namespace.
123 *
124 *	spa_remove()		Remove a spa_t from the namespace.  This also
125 *				frees up any memory associated with the spa_t.
126 *
127 *	spa_next()		Returns the next spa_t in the system, or the
128 *				first if NULL is passed.
129 *
130 *	spa_evict_all()		Shutdown and remove all spa_t structures in
131 *				the system.
132 *
133 *	spa_guid_exists()	Determine whether a pool/device guid exists.
134 *
135 * The spa_refcount is manipulated using the following functions:
136 *
137 *	spa_open_ref()		Adds a reference to the given spa_t.  Must be
138 *				called with spa_namespace_lock held if the
139 *				refcount is currently zero.
140 *
141 *	spa_close()		Remove a reference from the spa_t.  This will
142 *				not free the spa_t or remove it from the
143 *				namespace.  No locking is required.
144 *
145 *	spa_refcount_zero()	Returns true if the refcount is currently
146 *				zero.  Must be called with spa_namespace_lock
147 *				held.
148 *
149 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
150 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
151 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
152 *
153 * To read the configuration, it suffices to hold one of these locks as reader.
154 * To modify the configuration, you must hold all locks as writer.  To modify
155 * vdev state without altering the vdev tree's topology (e.g. online/offline),
156 * you must hold SCL_STATE and SCL_ZIO as writer.
157 *
158 * We use these distinct config locks to avoid recursive lock entry.
159 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
160 * block allocations (SCL_ALLOC), which may require reading space maps
161 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
162 *
163 * The spa config locks cannot be normal rwlocks because we need the
164 * ability to hand off ownership.  For example, SCL_ZIO is acquired
165 * by the issuing thread and later released by an interrupt thread.
166 * They do, however, obey the usual write-wanted semantics to prevent
167 * writer (i.e. system administrator) starvation.
168 *
169 * The lock acquisition rules are as follows:
170 *
171 * SCL_CONFIG
172 *	Protects changes to the vdev tree topology, such as vdev
173 *	add/remove/attach/detach.  Protects the dirty config list
174 *	(spa_config_dirty_list) and the set of spares and l2arc devices.
175 *
176 * SCL_STATE
177 *	Protects changes to pool state and vdev state, such as vdev
178 *	online/offline/fault/degrade/clear.  Protects the dirty state list
179 *	(spa_state_dirty_list) and global pool state (spa_state).
180 *
181 * SCL_ALLOC
182 *	Protects changes to metaslab groups and classes.
183 *	Held as reader by metaslab_alloc() and metaslab_claim().
184 *
185 * SCL_ZIO
186 *	Held by bp-level zios (those which have no io_vd upon entry)
187 *	to prevent changes to the vdev tree.  The bp-level zio implicitly
188 *	protects all of its vdev child zios, which do not hold SCL_ZIO.
189 *
190 * SCL_FREE
191 *	Protects changes to metaslab groups and classes.
192 *	Held as reader by metaslab_free().  SCL_FREE is distinct from
193 *	SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
194 *	blocks in zio_done() while another i/o that holds either
195 *	SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
196 *
197 * SCL_VDEV
198 *	Held as reader to prevent changes to the vdev tree during trivial
199 *	inquiries such as bp_get_dsize().  SCL_VDEV is distinct from the
200 *	other locks, and lower than all of them, to ensure that it's safe
201 *	to acquire regardless of caller context.
202 *
203 * In addition, the following rules apply:
204 *
205 * (a)	spa_props_lock protects pool properties, spa_config and spa_config_list.
206 *	The lock ordering is SCL_CONFIG > spa_props_lock.
207 *
208 * (b)	I/O operations on leaf vdevs.  For any zio operation that takes
209 *	an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
210 *	or zio_write_phys() -- the caller must ensure that the config cannot
211 *	cannot change in the interim, and that the vdev cannot be reopened.
212 *	SCL_STATE as reader suffices for both.
213 *
214 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
215 *
216 *	spa_vdev_enter()	Acquire the namespace lock and the config lock
217 *				for writing.
218 *
219 *	spa_vdev_exit()		Release the config lock, wait for all I/O
220 *				to complete, sync the updated configs to the
221 *				cache, and release the namespace lock.
222 *
223 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
224 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
225 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
226 */
227
228static avl_tree_t spa_namespace_avl;
229kmutex_t spa_namespace_lock;
230static kcondvar_t spa_namespace_cv;
231static int spa_active_count;
232int spa_max_replication_override = SPA_DVAS_PER_BP;
233
234static kmutex_t spa_spare_lock;
235static avl_tree_t spa_spare_avl;
236static kmutex_t spa_l2cache_lock;
237static avl_tree_t spa_l2cache_avl;
238
239kmem_cache_t *spa_buffer_pool;
240int spa_mode_global;
241
242#ifdef ZFS_DEBUG
243/*
244 * Everything except dprintf, spa, and indirect_remap is on by default
245 * in debug builds.
246 */
247int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_INDIRECT_REMAP);
248#else
249int zfs_flags = 0;
250#endif
251
252/*
253 * zfs_recover can be set to nonzero to attempt to recover from
254 * otherwise-fatal errors, typically caused by on-disk corruption.  When
255 * set, calls to zfs_panic_recover() will turn into warning messages.
256 * This should only be used as a last resort, as it typically results
257 * in leaked space, or worse.
258 */
259boolean_t zfs_recover = B_FALSE;
260
261/*
262 * If destroy encounters an EIO while reading metadata (e.g. indirect
263 * blocks), space referenced by the missing metadata can not be freed.
264 * Normally this causes the background destroy to become "stalled", as
265 * it is unable to make forward progress.  While in this stalled state,
266 * all remaining space to free from the error-encountering filesystem is
267 * "temporarily leaked".  Set this flag to cause it to ignore the EIO,
268 * permanently leak the space from indirect blocks that can not be read,
269 * and continue to free everything else that it can.
270 *
271 * The default, "stalling" behavior is useful if the storage partially
272 * fails (i.e. some but not all i/os fail), and then later recovers.  In
273 * this case, we will be able to continue pool operations while it is
274 * partially failed, and when it recovers, we can continue to free the
275 * space, with no leaks.  However, note that this case is actually
276 * fairly rare.
277 *
278 * Typically pools either (a) fail completely (but perhaps temporarily,
279 * e.g. a top-level vdev going offline), or (b) have localized,
280 * permanent errors (e.g. disk returns the wrong data due to bit flip or
281 * firmware bug).  In case (a), this setting does not matter because the
282 * pool will be suspended and the sync thread will not be able to make
283 * forward progress regardless.  In case (b), because the error is
284 * permanent, the best we can do is leak the minimum amount of space,
285 * which is what setting this flag will do.  Therefore, it is reasonable
286 * for this flag to normally be set, but we chose the more conservative
287 * approach of not setting it, so that there is no possibility of
288 * leaking space in the "partial temporary" failure case.
289 */
290boolean_t zfs_free_leak_on_eio = B_FALSE;
291
292/*
293 * Expiration time in milliseconds. This value has two meanings. First it is
294 * used to determine when the spa_deadman() logic should fire. By default the
295 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
296 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
297 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
298 * in a system panic.
299 */
300uint64_t zfs_deadman_synctime_ms = 1000000ULL;
301
302/*
303 * Check time in milliseconds. This defines the frequency at which we check
304 * for hung I/O.
305 */
306uint64_t zfs_deadman_checktime_ms = 5000ULL;
307
308/*
309 * Override the zfs deadman behavior via /etc/system. By default the
310 * deadman is enabled except on VMware and sparc deployments.
311 */
312int zfs_deadman_enabled = -1;
313
314/*
315 * The worst case is single-sector max-parity RAID-Z blocks, in which
316 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
317 * times the size; so just assume that.  Add to this the fact that
318 * we can have up to 3 DVAs per bp, and one more factor of 2 because
319 * the block may be dittoed with up to 3 DVAs by ddt_sync().  All together,
320 * the worst case is:
321 *     (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
322 */
323int spa_asize_inflation = 24;
324
325/*
326 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
327 * the pool to be consumed.  This ensures that we don't run the pool
328 * completely out of space, due to unaccounted changes (e.g. to the MOS).
329 * It also limits the worst-case time to allocate space.  If we have
330 * less than this amount of free space, most ZPL operations (e.g. write,
331 * create) will return ENOSPC.
332 *
333 * Certain operations (e.g. file removal, most administrative actions) can
334 * use half the slop space.  They will only return ENOSPC if less than half
335 * the slop space is free.  Typically, once the pool has less than the slop
336 * space free, the user will use these operations to free up space in the pool.
337 * These are the operations that call dsl_pool_adjustedsize() with the netfree
338 * argument set to TRUE.
339 *
340 * Operations that are almost guaranteed to free up space in the absence of
341 * a pool checkpoint can use up to three quarters of the slop space
342 * (e.g zfs destroy).
343 *
344 * A very restricted set of operations are always permitted, regardless of
345 * the amount of free space.  These are the operations that call
346 * dsl_sync_task(ZFS_SPACE_CHECK_NONE). If these operations result in a net
347 * increase in the amount of space used, it is possible to run the pool
348 * completely out of space, causing it to be permanently read-only.
349 *
350 * Note that on very small pools, the slop space will be larger than
351 * 3.2%, in an effort to have it be at least spa_min_slop (128MB),
352 * but we never allow it to be more than half the pool size.
353 *
354 * See also the comments in zfs_space_check_t.
355 */
356int spa_slop_shift = 5;
357uint64_t spa_min_slop = 128 * 1024 * 1024;
358
359int spa_allocators = 4;
360
361/*PRINTFLIKE2*/
362void
363spa_load_failed(spa_t *spa, const char *fmt, ...)
364{
365	va_list adx;
366	char buf[256];
367
368	va_start(adx, fmt);
369	(void) vsnprintf(buf, sizeof (buf), fmt, adx);
370	va_end(adx);
371
372	zfs_dbgmsg("spa_load(%s, config %s): FAILED: %s", spa->spa_name,
373	    spa->spa_trust_config ? "trusted" : "untrusted", buf);
374}
375
376/*PRINTFLIKE2*/
377void
378spa_load_note(spa_t *spa, const char *fmt, ...)
379{
380	va_list adx;
381	char buf[256];
382
383	va_start(adx, fmt);
384	(void) vsnprintf(buf, sizeof (buf), fmt, adx);
385	va_end(adx);
386
387	zfs_dbgmsg("spa_load(%s, config %s): %s", spa->spa_name,
388	    spa->spa_trust_config ? "trusted" : "untrusted", buf);
389}
390
391/*
392 * By default dedup and user data indirects land in the special class
393 */
394int zfs_ddt_data_is_special = B_TRUE;
395int zfs_user_indirect_is_special = B_TRUE;
396
397/*
398 * The percentage of special class final space reserved for metadata only.
399 * Once we allocate 100 - zfs_special_class_metadata_reserve_pct we only
400 * let metadata into the class.
401 */
402int zfs_special_class_metadata_reserve_pct = 25;
403
404/*
405 * ==========================================================================
406 * SPA config locking
407 * ==========================================================================
408 */
409static void
410spa_config_lock_init(spa_t *spa)
411{
412	for (int i = 0; i < SCL_LOCKS; i++) {
413		spa_config_lock_t *scl = &spa->spa_config_lock[i];
414		mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
415		cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
416		zfs_refcount_create_untracked(&scl->scl_count);
417		scl->scl_writer = NULL;
418		scl->scl_write_wanted = 0;
419	}
420}
421
422static void
423spa_config_lock_destroy(spa_t *spa)
424{
425	for (int i = 0; i < SCL_LOCKS; i++) {
426		spa_config_lock_t *scl = &spa->spa_config_lock[i];
427		mutex_destroy(&scl->scl_lock);
428		cv_destroy(&scl->scl_cv);
429		zfs_refcount_destroy(&scl->scl_count);
430		ASSERT(scl->scl_writer == NULL);
431		ASSERT(scl->scl_write_wanted == 0);
432	}
433}
434
435int
436spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
437{
438	for (int i = 0; i < SCL_LOCKS; i++) {
439		spa_config_lock_t *scl = &spa->spa_config_lock[i];
440		if (!(locks & (1 << i)))
441			continue;
442		mutex_enter(&scl->scl_lock);
443		if (rw == RW_READER) {
444			if (scl->scl_writer || scl->scl_write_wanted) {
445				mutex_exit(&scl->scl_lock);
446				spa_config_exit(spa, locks & ((1 << i) - 1),
447				    tag);
448				return (0);
449			}
450		} else {
451			ASSERT(scl->scl_writer != curthread);
452			if (!zfs_refcount_is_zero(&scl->scl_count)) {
453				mutex_exit(&scl->scl_lock);
454				spa_config_exit(spa, locks & ((1 << i) - 1),
455				    tag);
456				return (0);
457			}
458			scl->scl_writer = curthread;
459		}
460		(void) zfs_refcount_add(&scl->scl_count, tag);
461		mutex_exit(&scl->scl_lock);
462	}
463	return (1);
464}
465
466void
467spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
468{
469	int wlocks_held = 0;
470
471	ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
472
473	for (int i = 0; i < SCL_LOCKS; i++) {
474		spa_config_lock_t *scl = &spa->spa_config_lock[i];
475		if (scl->scl_writer == curthread)
476			wlocks_held |= (1 << i);
477		if (!(locks & (1 << i)))
478			continue;
479		mutex_enter(&scl->scl_lock);
480		if (rw == RW_READER) {
481			while (scl->scl_writer || scl->scl_write_wanted) {
482				cv_wait(&scl->scl_cv, &scl->scl_lock);
483			}
484		} else {
485			ASSERT(scl->scl_writer != curthread);
486			while (!zfs_refcount_is_zero(&scl->scl_count)) {
487				scl->scl_write_wanted++;
488				cv_wait(&scl->scl_cv, &scl->scl_lock);
489				scl->scl_write_wanted--;
490			}
491			scl->scl_writer = curthread;
492		}
493		(void) zfs_refcount_add(&scl->scl_count, tag);
494		mutex_exit(&scl->scl_lock);
495	}
496	ASSERT3U(wlocks_held, <=, locks);
497}
498
499void
500spa_config_exit(spa_t *spa, int locks, void *tag)
501{
502	for (int i = SCL_LOCKS - 1; i >= 0; i--) {
503		spa_config_lock_t *scl = &spa->spa_config_lock[i];
504		if (!(locks & (1 << i)))
505			continue;
506		mutex_enter(&scl->scl_lock);
507		ASSERT(!zfs_refcount_is_zero(&scl->scl_count));
508		if (zfs_refcount_remove(&scl->scl_count, tag) == 0) {
509			ASSERT(scl->scl_writer == NULL ||
510			    scl->scl_writer == curthread);
511			scl->scl_writer = NULL;	/* OK in either case */
512			cv_broadcast(&scl->scl_cv);
513		}
514		mutex_exit(&scl->scl_lock);
515	}
516}
517
518int
519spa_config_held(spa_t *spa, int locks, krw_t rw)
520{
521	int locks_held = 0;
522
523	for (int i = 0; i < SCL_LOCKS; i++) {
524		spa_config_lock_t *scl = &spa->spa_config_lock[i];
525		if (!(locks & (1 << i)))
526			continue;
527		if ((rw == RW_READER &&
528		    !zfs_refcount_is_zero(&scl->scl_count)) ||
529		    (rw == RW_WRITER && scl->scl_writer == curthread))
530			locks_held |= 1 << i;
531	}
532
533	return (locks_held);
534}
535
536/*
537 * ==========================================================================
538 * SPA namespace functions
539 * ==========================================================================
540 */
541
542/*
543 * Lookup the named spa_t in the AVL tree.  The spa_namespace_lock must be held.
544 * Returns NULL if no matching spa_t is found.
545 */
546spa_t *
547spa_lookup(const char *name)
548{
549	static spa_t search;	/* spa_t is large; don't allocate on stack */
550	spa_t *spa;
551	avl_index_t where;
552	char *cp;
553
554	ASSERT(MUTEX_HELD(&spa_namespace_lock));
555
556	(void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
557
558	/*
559	 * If it's a full dataset name, figure out the pool name and
560	 * just use that.
561	 */
562	cp = strpbrk(search.spa_name, "/@#");
563	if (cp != NULL)
564		*cp = '\0';
565
566	spa = avl_find(&spa_namespace_avl, &search, &where);
567
568	return (spa);
569}
570
571/*
572 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
573 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
574 * looking for potentially hung I/Os.
575 */
576void
577spa_deadman(void *arg)
578{
579	spa_t *spa = arg;
580
581	/*
582	 * Disable the deadman timer if the pool is suspended.
583	 */
584	if (spa_suspended(spa)) {
585		VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
586		return;
587	}
588
589	zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
590	    (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
591	    ++spa->spa_deadman_calls);
592	if (zfs_deadman_enabled)
593		vdev_deadman(spa->spa_root_vdev);
594}
595
596/*
597 * Create an uninitialized spa_t with the given name.  Requires
598 * spa_namespace_lock.  The caller must ensure that the spa_t doesn't already
599 * exist by calling spa_lookup() first.
600 */
601spa_t *
602spa_add(const char *name, nvlist_t *config, const char *altroot)
603{
604	spa_t *spa;
605	spa_config_dirent_t *dp;
606	cyc_handler_t hdlr;
607	cyc_time_t when;
608
609	ASSERT(MUTEX_HELD(&spa_namespace_lock));
610
611	spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
612
613	mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
614	mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
615	mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
616	mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL);
617	mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
618	mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
619	mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
620	mutex_init(&spa->spa_cksum_tmpls_lock, NULL, MUTEX_DEFAULT, NULL);
621	mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
622	mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
623	mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
624	mutex_init(&spa->spa_iokstat_lock, NULL, MUTEX_DEFAULT, NULL);
625
626	cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
627	cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL);
628	cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
629	cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
630	cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
631
632	for (int t = 0; t < TXG_SIZE; t++)
633		bplist_create(&spa->spa_free_bplist[t]);
634
635	(void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
636	spa->spa_state = POOL_STATE_UNINITIALIZED;
637	spa->spa_freeze_txg = UINT64_MAX;
638	spa->spa_final_txg = UINT64_MAX;
639	spa->spa_load_max_txg = UINT64_MAX;
640	spa->spa_proc = &p0;
641	spa->spa_proc_state = SPA_PROC_NONE;
642	spa->spa_trust_config = B_TRUE;
643
644	hdlr.cyh_func = spa_deadman;
645	hdlr.cyh_arg = spa;
646	hdlr.cyh_level = CY_LOW_LEVEL;
647
648	spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
649
650	/*
651	 * This determines how often we need to check for hung I/Os after
652	 * the cyclic has already fired. Since checking for hung I/Os is
653	 * an expensive operation we don't want to check too frequently.
654	 * Instead wait for 5 seconds before checking again.
655	 */
656	when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms);
657	when.cyt_when = CY_INFINITY;
658	mutex_enter(&cpu_lock);
659	spa->spa_deadman_cycid = cyclic_add(&hdlr, &when);
660	mutex_exit(&cpu_lock);
661
662	zfs_refcount_create(&spa->spa_refcount);
663	spa_config_lock_init(spa);
664
665	avl_add(&spa_namespace_avl, spa);
666
667	/*
668	 * Set the alternate root, if there is one.
669	 */
670	if (altroot) {
671		spa->spa_root = spa_strdup(altroot);
672		spa_active_count++;
673	}
674
675	spa->spa_alloc_count = spa_allocators;
676	spa->spa_alloc_locks = kmem_zalloc(spa->spa_alloc_count *
677	    sizeof (kmutex_t), KM_SLEEP);
678	spa->spa_alloc_trees = kmem_zalloc(spa->spa_alloc_count *
679	    sizeof (avl_tree_t), KM_SLEEP);
680	for (int i = 0; i < spa->spa_alloc_count; i++) {
681		mutex_init(&spa->spa_alloc_locks[i], NULL, MUTEX_DEFAULT, NULL);
682		avl_create(&spa->spa_alloc_trees[i], zio_bookmark_compare,
683		    sizeof (zio_t), offsetof(zio_t, io_alloc_node));
684	}
685
686	/*
687	 * Every pool starts with the default cachefile
688	 */
689	list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
690	    offsetof(spa_config_dirent_t, scd_link));
691
692	dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
693	dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
694	list_insert_head(&spa->spa_config_list, dp);
695
696	VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
697	    KM_SLEEP) == 0);
698
699	if (config != NULL) {
700		nvlist_t *features;
701
702		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
703		    &features) == 0) {
704			VERIFY(nvlist_dup(features, &spa->spa_label_features,
705			    0) == 0);
706		}
707
708		VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
709	}
710
711	if (spa->spa_label_features == NULL) {
712		VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
713		    KM_SLEEP) == 0);
714	}
715
716	spa->spa_iokstat = kstat_create("zfs", 0, name,
717	    "disk", KSTAT_TYPE_IO, 1, 0);
718	if (spa->spa_iokstat) {
719		spa->spa_iokstat->ks_lock = &spa->spa_iokstat_lock;
720		kstat_install(spa->spa_iokstat);
721	}
722
723	spa->spa_min_ashift = INT_MAX;
724	spa->spa_max_ashift = 0;
725
726	/*
727	 * As a pool is being created, treat all features as disabled by
728	 * setting SPA_FEATURE_DISABLED for all entries in the feature
729	 * refcount cache.
730	 */
731	for (int i = 0; i < SPA_FEATURES; i++) {
732		spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED;
733	}
734
735	list_create(&spa->spa_leaf_list, sizeof (vdev_t),
736	    offsetof(vdev_t, vdev_leaf_node));
737
738	return (spa);
739}
740
741/*
742 * Removes a spa_t from the namespace, freeing up any memory used.  Requires
743 * spa_namespace_lock.  This is called only after the spa_t has been closed and
744 * deactivated.
745 */
746void
747spa_remove(spa_t *spa)
748{
749	spa_config_dirent_t *dp;
750
751	ASSERT(MUTEX_HELD(&spa_namespace_lock));
752	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
753	ASSERT3U(zfs_refcount_count(&spa->spa_refcount), ==, 0);
754
755	nvlist_free(spa->spa_config_splitting);
756
757	avl_remove(&spa_namespace_avl, spa);
758	cv_broadcast(&spa_namespace_cv);
759
760	if (spa->spa_root) {
761		spa_strfree(spa->spa_root);
762		spa_active_count--;
763	}
764
765	while ((dp = list_head(&spa->spa_config_list)) != NULL) {
766		list_remove(&spa->spa_config_list, dp);
767		if (dp->scd_path != NULL)
768			spa_strfree(dp->scd_path);
769		kmem_free(dp, sizeof (spa_config_dirent_t));
770	}
771
772	for (int i = 0; i < spa->spa_alloc_count; i++) {
773		avl_destroy(&spa->spa_alloc_trees[i]);
774		mutex_destroy(&spa->spa_alloc_locks[i]);
775	}
776	kmem_free(spa->spa_alloc_locks, spa->spa_alloc_count *
777	    sizeof (kmutex_t));
778	kmem_free(spa->spa_alloc_trees, spa->spa_alloc_count *
779	    sizeof (avl_tree_t));
780
781	list_destroy(&spa->spa_config_list);
782	list_destroy(&spa->spa_leaf_list);
783
784	nvlist_free(spa->spa_label_features);
785	nvlist_free(spa->spa_load_info);
786	spa_config_set(spa, NULL);
787
788	mutex_enter(&cpu_lock);
789	if (spa->spa_deadman_cycid != CYCLIC_NONE)
790		cyclic_remove(spa->spa_deadman_cycid);
791	mutex_exit(&cpu_lock);
792	spa->spa_deadman_cycid = CYCLIC_NONE;
793
794	zfs_refcount_destroy(&spa->spa_refcount);
795
796	spa_config_lock_destroy(spa);
797
798	kstat_delete(spa->spa_iokstat);
799	spa->spa_iokstat = NULL;
800
801	for (int t = 0; t < TXG_SIZE; t++)
802		bplist_destroy(&spa->spa_free_bplist[t]);
803
804	zio_checksum_templates_free(spa);
805
806	cv_destroy(&spa->spa_async_cv);
807	cv_destroy(&spa->spa_evicting_os_cv);
808	cv_destroy(&spa->spa_proc_cv);
809	cv_destroy(&spa->spa_scrub_io_cv);
810	cv_destroy(&spa->spa_suspend_cv);
811
812	mutex_destroy(&spa->spa_async_lock);
813	mutex_destroy(&spa->spa_errlist_lock);
814	mutex_destroy(&spa->spa_errlog_lock);
815	mutex_destroy(&spa->spa_evicting_os_lock);
816	mutex_destroy(&spa->spa_history_lock);
817	mutex_destroy(&spa->spa_proc_lock);
818	mutex_destroy(&spa->spa_props_lock);
819	mutex_destroy(&spa->spa_cksum_tmpls_lock);
820	mutex_destroy(&spa->spa_scrub_lock);
821	mutex_destroy(&spa->spa_suspend_lock);
822	mutex_destroy(&spa->spa_vdev_top_lock);
823	mutex_destroy(&spa->spa_iokstat_lock);
824
825	kmem_free(spa, sizeof (spa_t));
826}
827
828/*
829 * Given a pool, return the next pool in the namespace, or NULL if there is
830 * none.  If 'prev' is NULL, return the first pool.
831 */
832spa_t *
833spa_next(spa_t *prev)
834{
835	ASSERT(MUTEX_HELD(&spa_namespace_lock));
836
837	if (prev)
838		return (AVL_NEXT(&spa_namespace_avl, prev));
839	else
840		return (avl_first(&spa_namespace_avl));
841}
842
843/*
844 * ==========================================================================
845 * SPA refcount functions
846 * ==========================================================================
847 */
848
849/*
850 * Add a reference to the given spa_t.  Must have at least one reference, or
851 * have the namespace lock held.
852 */
853void
854spa_open_ref(spa_t *spa, void *tag)
855{
856	ASSERT(zfs_refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
857	    MUTEX_HELD(&spa_namespace_lock));
858	(void) zfs_refcount_add(&spa->spa_refcount, tag);
859}
860
861/*
862 * Remove a reference to the given spa_t.  Must have at least one reference, or
863 * have the namespace lock held.
864 */
865void
866spa_close(spa_t *spa, void *tag)
867{
868	ASSERT(zfs_refcount_count(&spa->spa_refcount) > spa->spa_minref ||
869	    MUTEX_HELD(&spa_namespace_lock));
870	(void) zfs_refcount_remove(&spa->spa_refcount, tag);
871}
872
873/*
874 * Remove a reference to the given spa_t held by a dsl dir that is
875 * being asynchronously released.  Async releases occur from a taskq
876 * performing eviction of dsl datasets and dirs.  The namespace lock
877 * isn't held and the hold by the object being evicted may contribute to
878 * spa_minref (e.g. dataset or directory released during pool export),
879 * so the asserts in spa_close() do not apply.
880 */
881void
882spa_async_close(spa_t *spa, void *tag)
883{
884	(void) zfs_refcount_remove(&spa->spa_refcount, tag);
885}
886
887/*
888 * Check to see if the spa refcount is zero.  Must be called with
889 * spa_namespace_lock held.  We really compare against spa_minref, which is the
890 * number of references acquired when opening a pool
891 */
892boolean_t
893spa_refcount_zero(spa_t *spa)
894{
895	ASSERT(MUTEX_HELD(&spa_namespace_lock));
896
897	return (zfs_refcount_count(&spa->spa_refcount) == spa->spa_minref);
898}
899
900/*
901 * ==========================================================================
902 * SPA spare and l2cache tracking
903 * ==========================================================================
904 */
905
906/*
907 * Hot spares and cache devices are tracked using the same code below,
908 * for 'auxiliary' devices.
909 */
910
911typedef struct spa_aux {
912	uint64_t	aux_guid;
913	uint64_t	aux_pool;
914	avl_node_t	aux_avl;
915	int		aux_count;
916} spa_aux_t;
917
918static inline int
919spa_aux_compare(const void *a, const void *b)
920{
921	const spa_aux_t *sa = (const spa_aux_t *)a;
922	const spa_aux_t *sb = (const spa_aux_t *)b;
923
924	return (AVL_CMP(sa->aux_guid, sb->aux_guid));
925}
926
927void
928spa_aux_add(vdev_t *vd, avl_tree_t *avl)
929{
930	avl_index_t where;
931	spa_aux_t search;
932	spa_aux_t *aux;
933
934	search.aux_guid = vd->vdev_guid;
935	if ((aux = avl_find(avl, &search, &where)) != NULL) {
936		aux->aux_count++;
937	} else {
938		aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
939		aux->aux_guid = vd->vdev_guid;
940		aux->aux_count = 1;
941		avl_insert(avl, aux, where);
942	}
943}
944
945void
946spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
947{
948	spa_aux_t search;
949	spa_aux_t *aux;
950	avl_index_t where;
951
952	search.aux_guid = vd->vdev_guid;
953	aux = avl_find(avl, &search, &where);
954
955	ASSERT(aux != NULL);
956
957	if (--aux->aux_count == 0) {
958		avl_remove(avl, aux);
959		kmem_free(aux, sizeof (spa_aux_t));
960	} else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
961		aux->aux_pool = 0ULL;
962	}
963}
964
965boolean_t
966spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
967{
968	spa_aux_t search, *found;
969
970	search.aux_guid = guid;
971	found = avl_find(avl, &search, NULL);
972
973	if (pool) {
974		if (found)
975			*pool = found->aux_pool;
976		else
977			*pool = 0ULL;
978	}
979
980	if (refcnt) {
981		if (found)
982			*refcnt = found->aux_count;
983		else
984			*refcnt = 0;
985	}
986
987	return (found != NULL);
988}
989
990void
991spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
992{
993	spa_aux_t search, *found;
994	avl_index_t where;
995
996	search.aux_guid = vd->vdev_guid;
997	found = avl_find(avl, &search, &where);
998	ASSERT(found != NULL);
999	ASSERT(found->aux_pool == 0ULL);
1000
1001	found->aux_pool = spa_guid(vd->vdev_spa);
1002}
1003
1004/*
1005 * Spares are tracked globally due to the following constraints:
1006 *
1007 *	- A spare may be part of multiple pools.
1008 *	- A spare may be added to a pool even if it's actively in use within
1009 *	  another pool.
1010 *	- A spare in use in any pool can only be the source of a replacement if
1011 *	  the target is a spare in the same pool.
1012 *
1013 * We keep track of all spares on the system through the use of a reference
1014 * counted AVL tree.  When a vdev is added as a spare, or used as a replacement
1015 * spare, then we bump the reference count in the AVL tree.  In addition, we set
1016 * the 'vdev_isspare' member to indicate that the device is a spare (active or
1017 * inactive).  When a spare is made active (used to replace a device in the
1018 * pool), we also keep track of which pool its been made a part of.
1019 *
1020 * The 'spa_spare_lock' protects the AVL tree.  These functions are normally
1021 * called under the spa_namespace lock as part of vdev reconfiguration.  The
1022 * separate spare lock exists for the status query path, which does not need to
1023 * be completely consistent with respect to other vdev configuration changes.
1024 */
1025
1026static int
1027spa_spare_compare(const void *a, const void *b)
1028{
1029	return (spa_aux_compare(a, b));
1030}
1031
1032void
1033spa_spare_add(vdev_t *vd)
1034{
1035	mutex_enter(&spa_spare_lock);
1036	ASSERT(!vd->vdev_isspare);
1037	spa_aux_add(vd, &spa_spare_avl);
1038	vd->vdev_isspare = B_TRUE;
1039	mutex_exit(&spa_spare_lock);
1040}
1041
1042void
1043spa_spare_remove(vdev_t *vd)
1044{
1045	mutex_enter(&spa_spare_lock);
1046	ASSERT(vd->vdev_isspare);
1047	spa_aux_remove(vd, &spa_spare_avl);
1048	vd->vdev_isspare = B_FALSE;
1049	mutex_exit(&spa_spare_lock);
1050}
1051
1052boolean_t
1053spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
1054{
1055	boolean_t found;
1056
1057	mutex_enter(&spa_spare_lock);
1058	found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
1059	mutex_exit(&spa_spare_lock);
1060
1061	return (found);
1062}
1063
1064void
1065spa_spare_activate(vdev_t *vd)
1066{
1067	mutex_enter(&spa_spare_lock);
1068	ASSERT(vd->vdev_isspare);
1069	spa_aux_activate(vd, &spa_spare_avl);
1070	mutex_exit(&spa_spare_lock);
1071}
1072
1073/*
1074 * Level 2 ARC devices are tracked globally for the same reasons as spares.
1075 * Cache devices currently only support one pool per cache device, and so
1076 * for these devices the aux reference count is currently unused beyond 1.
1077 */
1078
1079static int
1080spa_l2cache_compare(const void *a, const void *b)
1081{
1082	return (spa_aux_compare(a, b));
1083}
1084
1085void
1086spa_l2cache_add(vdev_t *vd)
1087{
1088	mutex_enter(&spa_l2cache_lock);
1089	ASSERT(!vd->vdev_isl2cache);
1090	spa_aux_add(vd, &spa_l2cache_avl);
1091	vd->vdev_isl2cache = B_TRUE;
1092	mutex_exit(&spa_l2cache_lock);
1093}
1094
1095void
1096spa_l2cache_remove(vdev_t *vd)
1097{
1098	mutex_enter(&spa_l2cache_lock);
1099	ASSERT(vd->vdev_isl2cache);
1100	spa_aux_remove(vd, &spa_l2cache_avl);
1101	vd->vdev_isl2cache = B_FALSE;
1102	mutex_exit(&spa_l2cache_lock);
1103}
1104
1105boolean_t
1106spa_l2cache_exists(uint64_t guid, uint64_t *pool)
1107{
1108	boolean_t found;
1109
1110	mutex_enter(&spa_l2cache_lock);
1111	found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
1112	mutex_exit(&spa_l2cache_lock);
1113
1114	return (found);
1115}
1116
1117void
1118spa_l2cache_activate(vdev_t *vd)
1119{
1120	mutex_enter(&spa_l2cache_lock);
1121	ASSERT(vd->vdev_isl2cache);
1122	spa_aux_activate(vd, &spa_l2cache_avl);
1123	mutex_exit(&spa_l2cache_lock);
1124}
1125
1126/*
1127 * ==========================================================================
1128 * SPA vdev locking
1129 * ==========================================================================
1130 */
1131
1132/*
1133 * Lock the given spa_t for the purpose of adding or removing a vdev.
1134 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1135 * It returns the next transaction group for the spa_t.
1136 */
1137uint64_t
1138spa_vdev_enter(spa_t *spa)
1139{
1140	mutex_enter(&spa->spa_vdev_top_lock);
1141	mutex_enter(&spa_namespace_lock);
1142	return (spa_vdev_config_enter(spa));
1143}
1144
1145/*
1146 * Internal implementation for spa_vdev_enter().  Used when a vdev
1147 * operation requires multiple syncs (i.e. removing a device) while
1148 * keeping the spa_namespace_lock held.
1149 */
1150uint64_t
1151spa_vdev_config_enter(spa_t *spa)
1152{
1153	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1154
1155	spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1156
1157	return (spa_last_synced_txg(spa) + 1);
1158}
1159
1160/*
1161 * Used in combination with spa_vdev_config_enter() to allow the syncing
1162 * of multiple transactions without releasing the spa_namespace_lock.
1163 */
1164void
1165spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
1166{
1167	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1168
1169	int config_changed = B_FALSE;
1170
1171	ASSERT(txg > spa_last_synced_txg(spa));
1172
1173	spa->spa_pending_vdev = NULL;
1174
1175	/*
1176	 * Reassess the DTLs.
1177	 */
1178	vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
1179
1180	if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
1181		config_changed = B_TRUE;
1182		spa->spa_config_generation++;
1183	}
1184
1185	/*
1186	 * Verify the metaslab classes.
1187	 */
1188	ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
1189	ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
1190	ASSERT(metaslab_class_validate(spa_special_class(spa)) == 0);
1191	ASSERT(metaslab_class_validate(spa_dedup_class(spa)) == 0);
1192
1193	spa_config_exit(spa, SCL_ALL, spa);
1194
1195	/*
1196	 * Panic the system if the specified tag requires it.  This
1197	 * is useful for ensuring that configurations are updated
1198	 * transactionally.
1199	 */
1200	if (zio_injection_enabled)
1201		zio_handle_panic_injection(spa, tag, 0);
1202
1203	/*
1204	 * Note: this txg_wait_synced() is important because it ensures
1205	 * that there won't be more than one config change per txg.
1206	 * This allows us to use the txg as the generation number.
1207	 */
1208	if (error == 0)
1209		txg_wait_synced(spa->spa_dsl_pool, txg);
1210
1211	if (vd != NULL) {
1212		ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
1213		if (vd->vdev_ops->vdev_op_leaf) {
1214			mutex_enter(&vd->vdev_initialize_lock);
1215			vdev_initialize_stop(vd, VDEV_INITIALIZE_CANCELED);
1216			mutex_exit(&vd->vdev_initialize_lock);
1217		}
1218
1219		spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1220		vdev_free(vd);
1221		spa_config_exit(spa, SCL_ALL, spa);
1222	}
1223
1224	/*
1225	 * If the config changed, update the config cache.
1226	 */
1227	if (config_changed)
1228		spa_write_cachefile(spa, B_FALSE, B_TRUE);
1229}
1230
1231/*
1232 * Unlock the spa_t after adding or removing a vdev.  Besides undoing the
1233 * locking of spa_vdev_enter(), we also want make sure the transactions have
1234 * synced to disk, and then update the global configuration cache with the new
1235 * information.
1236 */
1237int
1238spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1239{
1240	spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1241	mutex_exit(&spa_namespace_lock);
1242	mutex_exit(&spa->spa_vdev_top_lock);
1243
1244	return (error);
1245}
1246
1247/*
1248 * Lock the given spa_t for the purpose of changing vdev state.
1249 */
1250void
1251spa_vdev_state_enter(spa_t *spa, int oplocks)
1252{
1253	int locks = SCL_STATE_ALL | oplocks;
1254
1255	/*
1256	 * Root pools may need to read of the underlying devfs filesystem
1257	 * when opening up a vdev.  Unfortunately if we're holding the
1258	 * SCL_ZIO lock it will result in a deadlock when we try to issue
1259	 * the read from the root filesystem.  Instead we "prefetch"
1260	 * the associated vnodes that we need prior to opening the
1261	 * underlying devices and cache them so that we can prevent
1262	 * any I/O when we are doing the actual open.
1263	 */
1264	if (spa_is_root(spa)) {
1265		int low = locks & ~(SCL_ZIO - 1);
1266		int high = locks & ~low;
1267
1268		spa_config_enter(spa, high, spa, RW_WRITER);
1269		vdev_hold(spa->spa_root_vdev);
1270		spa_config_enter(spa, low, spa, RW_WRITER);
1271	} else {
1272		spa_config_enter(spa, locks, spa, RW_WRITER);
1273	}
1274	spa->spa_vdev_locks = locks;
1275}
1276
1277int
1278spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1279{
1280	boolean_t config_changed = B_FALSE;
1281
1282	if (vd != NULL || error == 0)
1283		vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1284		    0, 0, B_FALSE);
1285
1286	if (vd != NULL) {
1287		vdev_state_dirty(vd->vdev_top);
1288		config_changed = B_TRUE;
1289		spa->spa_config_generation++;
1290	}
1291
1292	if (spa_is_root(spa))
1293		vdev_rele(spa->spa_root_vdev);
1294
1295	ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1296	spa_config_exit(spa, spa->spa_vdev_locks, spa);
1297
1298	/*
1299	 * If anything changed, wait for it to sync.  This ensures that,
1300	 * from the system administrator's perspective, zpool(1M) commands
1301	 * are synchronous.  This is important for things like zpool offline:
1302	 * when the command completes, you expect no further I/O from ZFS.
1303	 */
1304	if (vd != NULL)
1305		txg_wait_synced(spa->spa_dsl_pool, 0);
1306
1307	/*
1308	 * If the config changed, update the config cache.
1309	 */
1310	if (config_changed) {
1311		mutex_enter(&spa_namespace_lock);
1312		spa_write_cachefile(spa, B_FALSE, B_TRUE);
1313		mutex_exit(&spa_namespace_lock);
1314	}
1315
1316	return (error);
1317}
1318
1319/*
1320 * ==========================================================================
1321 * Miscellaneous functions
1322 * ==========================================================================
1323 */
1324
1325void
1326spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx)
1327{
1328	if (!nvlist_exists(spa->spa_label_features, feature)) {
1329		fnvlist_add_boolean(spa->spa_label_features, feature);
1330		/*
1331		 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1332		 * dirty the vdev config because lock SCL_CONFIG is not held.
1333		 * Thankfully, in this case we don't need to dirty the config
1334		 * because it will be written out anyway when we finish
1335		 * creating the pool.
1336		 */
1337		if (tx->tx_txg != TXG_INITIAL)
1338			vdev_config_dirty(spa->spa_root_vdev);
1339	}
1340}
1341
1342void
1343spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1344{
1345	if (nvlist_remove_all(spa->spa_label_features, feature) == 0)
1346		vdev_config_dirty(spa->spa_root_vdev);
1347}
1348
1349/*
1350 * Return the spa_t associated with given pool_guid, if it exists.  If
1351 * device_guid is non-zero, determine whether the pool exists *and* contains
1352 * a device with the specified device_guid.
1353 */
1354spa_t *
1355spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1356{
1357	spa_t *spa;
1358	avl_tree_t *t = &spa_namespace_avl;
1359
1360	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1361
1362	for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1363		if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1364			continue;
1365		if (spa->spa_root_vdev == NULL)
1366			continue;
1367		if (spa_guid(spa) == pool_guid) {
1368			if (device_guid == 0)
1369				break;
1370
1371			if (vdev_lookup_by_guid(spa->spa_root_vdev,
1372			    device_guid) != NULL)
1373				break;
1374
1375			/*
1376			 * Check any devices we may be in the process of adding.
1377			 */
1378			if (spa->spa_pending_vdev) {
1379				if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1380				    device_guid) != NULL)
1381					break;
1382			}
1383		}
1384	}
1385
1386	return (spa);
1387}
1388
1389/*
1390 * Determine whether a pool with the given pool_guid exists.
1391 */
1392boolean_t
1393spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1394{
1395	return (spa_by_guid(pool_guid, device_guid) != NULL);
1396}
1397
1398char *
1399spa_strdup(const char *s)
1400{
1401	size_t len;
1402	char *new;
1403
1404	len = strlen(s);
1405	new = kmem_alloc(len + 1, KM_SLEEP);
1406	bcopy(s, new, len);
1407	new[len] = '\0';
1408
1409	return (new);
1410}
1411
1412void
1413spa_strfree(char *s)
1414{
1415	kmem_free(s, strlen(s) + 1);
1416}
1417
1418uint64_t
1419spa_get_random(uint64_t range)
1420{
1421	uint64_t r;
1422
1423	ASSERT(range != 0);
1424
1425	if (range == 1)
1426		return (0);
1427
1428	(void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1429
1430	return (r % range);
1431}
1432
1433uint64_t
1434spa_generate_guid(spa_t *spa)
1435{
1436	uint64_t guid = spa_get_random(-1ULL);
1437
1438	if (spa != NULL) {
1439		while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1440			guid = spa_get_random(-1ULL);
1441	} else {
1442		while (guid == 0 || spa_guid_exists(guid, 0))
1443			guid = spa_get_random(-1ULL);
1444	}
1445
1446	return (guid);
1447}
1448
1449void
1450snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp)
1451{
1452	char type[256];
1453	char *checksum = NULL;
1454	char *compress = NULL;
1455
1456	if (bp != NULL) {
1457		if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1458			dmu_object_byteswap_t bswap =
1459			    DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1460			(void) snprintf(type, sizeof (type), "bswap %s %s",
1461			    DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1462			    "metadata" : "data",
1463			    dmu_ot_byteswap[bswap].ob_name);
1464		} else {
1465			(void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1466			    sizeof (type));
1467		}
1468		if (!BP_IS_EMBEDDED(bp)) {
1469			checksum =
1470			    zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1471		}
1472		compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1473	}
1474
1475	SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum,
1476	    compress);
1477}
1478
1479void
1480spa_freeze(spa_t *spa)
1481{
1482	uint64_t freeze_txg = 0;
1483
1484	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1485	if (spa->spa_freeze_txg == UINT64_MAX) {
1486		freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1487		spa->spa_freeze_txg = freeze_txg;
1488	}
1489	spa_config_exit(spa, SCL_ALL, FTAG);
1490	if (freeze_txg != 0)
1491		txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1492}
1493
1494void
1495zfs_panic_recover(const char *fmt, ...)
1496{
1497	va_list adx;
1498
1499	va_start(adx, fmt);
1500	vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1501	va_end(adx);
1502}
1503
1504/*
1505 * This is a stripped-down version of strtoull, suitable only for converting
1506 * lowercase hexadecimal numbers that don't overflow.
1507 */
1508uint64_t
1509zfs_strtonum(const char *str, char **nptr)
1510{
1511	uint64_t val = 0;
1512	char c;
1513	int digit;
1514
1515	while ((c = *str) != '\0') {
1516		if (c >= '0' && c <= '9')
1517			digit = c - '0';
1518		else if (c >= 'a' && c <= 'f')
1519			digit = 10 + c - 'a';
1520		else
1521			break;
1522
1523		val *= 16;
1524		val += digit;
1525
1526		str++;
1527	}
1528
1529	if (nptr)
1530		*nptr = (char *)str;
1531
1532	return (val);
1533}
1534
1535void
1536spa_activate_allocation_classes(spa_t *spa, dmu_tx_t *tx)
1537{
1538	/*
1539	 * We bump the feature refcount for each special vdev added to the pool
1540	 */
1541	ASSERT(spa_feature_is_enabled(spa, SPA_FEATURE_ALLOCATION_CLASSES));
1542	spa_feature_incr(spa, SPA_FEATURE_ALLOCATION_CLASSES, tx);
1543}
1544
1545/*
1546 * ==========================================================================
1547 * Accessor functions
1548 * ==========================================================================
1549 */
1550
1551boolean_t
1552spa_shutting_down(spa_t *spa)
1553{
1554	return (spa->spa_async_suspended);
1555}
1556
1557dsl_pool_t *
1558spa_get_dsl(spa_t *spa)
1559{
1560	return (spa->spa_dsl_pool);
1561}
1562
1563boolean_t
1564spa_is_initializing(spa_t *spa)
1565{
1566	return (spa->spa_is_initializing);
1567}
1568
1569boolean_t
1570spa_indirect_vdevs_loaded(spa_t *spa)
1571{
1572	return (spa->spa_indirect_vdevs_loaded);
1573}
1574
1575blkptr_t *
1576spa_get_rootblkptr(spa_t *spa)
1577{
1578	return (&spa->spa_ubsync.ub_rootbp);
1579}
1580
1581void
1582spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1583{
1584	spa->spa_uberblock.ub_rootbp = *bp;
1585}
1586
1587void
1588spa_altroot(spa_t *spa, char *buf, size_t buflen)
1589{
1590	if (spa->spa_root == NULL)
1591		buf[0] = '\0';
1592	else
1593		(void) strncpy(buf, spa->spa_root, buflen);
1594}
1595
1596int
1597spa_sync_pass(spa_t *spa)
1598{
1599	return (spa->spa_sync_pass);
1600}
1601
1602char *
1603spa_name(spa_t *spa)
1604{
1605	return (spa->spa_name);
1606}
1607
1608uint64_t
1609spa_guid(spa_t *spa)
1610{
1611	dsl_pool_t *dp = spa_get_dsl(spa);
1612	uint64_t guid;
1613
1614	/*
1615	 * If we fail to parse the config during spa_load(), we can go through
1616	 * the error path (which posts an ereport) and end up here with no root
1617	 * vdev.  We stash the original pool guid in 'spa_config_guid' to handle
1618	 * this case.
1619	 */
1620	if (spa->spa_root_vdev == NULL)
1621		return (spa->spa_config_guid);
1622
1623	guid = spa->spa_last_synced_guid != 0 ?
1624	    spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1625
1626	/*
1627	 * Return the most recently synced out guid unless we're
1628	 * in syncing context.
1629	 */
1630	if (dp && dsl_pool_sync_context(dp))
1631		return (spa->spa_root_vdev->vdev_guid);
1632	else
1633		return (guid);
1634}
1635
1636uint64_t
1637spa_load_guid(spa_t *spa)
1638{
1639	/*
1640	 * This is a GUID that exists solely as a reference for the
1641	 * purposes of the arc.  It is generated at load time, and
1642	 * is never written to persistent storage.
1643	 */
1644	return (spa->spa_load_guid);
1645}
1646
1647uint64_t
1648spa_last_synced_txg(spa_t *spa)
1649{
1650	return (spa->spa_ubsync.ub_txg);
1651}
1652
1653uint64_t
1654spa_first_txg(spa_t *spa)
1655{
1656	return (spa->spa_first_txg);
1657}
1658
1659uint64_t
1660spa_syncing_txg(spa_t *spa)
1661{
1662	return (spa->spa_syncing_txg);
1663}
1664
1665/*
1666 * Return the last txg where data can be dirtied. The final txgs
1667 * will be used to just clear out any deferred frees that remain.
1668 */
1669uint64_t
1670spa_final_dirty_txg(spa_t *spa)
1671{
1672	return (spa->spa_final_txg - TXG_DEFER_SIZE);
1673}
1674
1675pool_state_t
1676spa_state(spa_t *spa)
1677{
1678	return (spa->spa_state);
1679}
1680
1681spa_load_state_t
1682spa_load_state(spa_t *spa)
1683{
1684	return (spa->spa_load_state);
1685}
1686
1687uint64_t
1688spa_freeze_txg(spa_t *spa)
1689{
1690	return (spa->spa_freeze_txg);
1691}
1692
1693/* ARGSUSED */
1694uint64_t
1695spa_get_worst_case_asize(spa_t *spa, uint64_t lsize)
1696{
1697	return (lsize * spa_asize_inflation);
1698}
1699
1700/*
1701 * Return the amount of slop space in bytes.  It is 1/32 of the pool (3.2%),
1702 * or at least 128MB, unless that would cause it to be more than half the
1703 * pool size.
1704 *
1705 * See the comment above spa_slop_shift for details.
1706 */
1707uint64_t
1708spa_get_slop_space(spa_t *spa)
1709{
1710	uint64_t space = spa_get_dspace(spa);
1711	return (MAX(space >> spa_slop_shift, MIN(space >> 1, spa_min_slop)));
1712}
1713
1714uint64_t
1715spa_get_dspace(spa_t *spa)
1716{
1717	return (spa->spa_dspace);
1718}
1719
1720uint64_t
1721spa_get_checkpoint_space(spa_t *spa)
1722{
1723	return (spa->spa_checkpoint_info.sci_dspace);
1724}
1725
1726void
1727spa_update_dspace(spa_t *spa)
1728{
1729	spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1730	    ddt_get_dedup_dspace(spa);
1731	if (spa->spa_vdev_removal != NULL) {
1732		/*
1733		 * We can't allocate from the removing device, so
1734		 * subtract its size.  This prevents the DMU/DSL from
1735		 * filling up the (now smaller) pool while we are in the
1736		 * middle of removing the device.
1737		 *
1738		 * Note that the DMU/DSL doesn't actually know or care
1739		 * how much space is allocated (it does its own tracking
1740		 * of how much space has been logically used).  So it
1741		 * doesn't matter that the data we are moving may be
1742		 * allocated twice (on the old device and the new
1743		 * device).
1744		 */
1745		spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1746		vdev_t *vd =
1747		    vdev_lookup_top(spa, spa->spa_vdev_removal->svr_vdev_id);
1748		spa->spa_dspace -= spa_deflate(spa) ?
1749		    vd->vdev_stat.vs_dspace : vd->vdev_stat.vs_space;
1750		spa_config_exit(spa, SCL_VDEV, FTAG);
1751	}
1752}
1753
1754/*
1755 * Return the failure mode that has been set to this pool. The default
1756 * behavior will be to block all I/Os when a complete failure occurs.
1757 */
1758uint8_t
1759spa_get_failmode(spa_t *spa)
1760{
1761	return (spa->spa_failmode);
1762}
1763
1764boolean_t
1765spa_suspended(spa_t *spa)
1766{
1767	return (spa->spa_suspended != ZIO_SUSPEND_NONE);
1768}
1769
1770uint64_t
1771spa_version(spa_t *spa)
1772{
1773	return (spa->spa_ubsync.ub_version);
1774}
1775
1776boolean_t
1777spa_deflate(spa_t *spa)
1778{
1779	return (spa->spa_deflate);
1780}
1781
1782metaslab_class_t *
1783spa_normal_class(spa_t *spa)
1784{
1785	return (spa->spa_normal_class);
1786}
1787
1788metaslab_class_t *
1789spa_log_class(spa_t *spa)
1790{
1791	return (spa->spa_log_class);
1792}
1793
1794metaslab_class_t *
1795spa_special_class(spa_t *spa)
1796{
1797	return (spa->spa_special_class);
1798}
1799
1800metaslab_class_t *
1801spa_dedup_class(spa_t *spa)
1802{
1803	return (spa->spa_dedup_class);
1804}
1805
1806/*
1807 * Locate an appropriate allocation class
1808 */
1809metaslab_class_t *
1810spa_preferred_class(spa_t *spa, uint64_t size, dmu_object_type_t objtype,
1811    uint_t level, uint_t special_smallblk)
1812{
1813	if (DMU_OT_IS_ZIL(objtype)) {
1814		if (spa->spa_log_class->mc_groups != 0)
1815			return (spa_log_class(spa));
1816		else
1817			return (spa_normal_class(spa));
1818	}
1819
1820	boolean_t has_special_class = spa->spa_special_class->mc_groups != 0;
1821
1822	if (DMU_OT_IS_DDT(objtype)) {
1823		if (spa->spa_dedup_class->mc_groups != 0)
1824			return (spa_dedup_class(spa));
1825		else if (has_special_class && zfs_ddt_data_is_special)
1826			return (spa_special_class(spa));
1827		else
1828			return (spa_normal_class(spa));
1829	}
1830
1831	/* Indirect blocks for user data can land in special if allowed */
1832	if (level > 0 && (DMU_OT_IS_FILE(objtype) || objtype == DMU_OT_ZVOL)) {
1833		if (has_special_class && zfs_user_indirect_is_special)
1834			return (spa_special_class(spa));
1835		else
1836			return (spa_normal_class(spa));
1837	}
1838
1839	if (DMU_OT_IS_METADATA(objtype) || level > 0) {
1840		if (has_special_class)
1841			return (spa_special_class(spa));
1842		else
1843			return (spa_normal_class(spa));
1844	}
1845
1846	/*
1847	 * Allow small file blocks in special class in some cases (like
1848	 * for the dRAID vdev feature). But always leave a reserve of
1849	 * zfs_special_class_metadata_reserve_pct exclusively for metadata.
1850	 */
1851	if (DMU_OT_IS_FILE(objtype) &&
1852	    has_special_class && size <= special_smallblk) {
1853		metaslab_class_t *special = spa_special_class(spa);
1854		uint64_t alloc = metaslab_class_get_alloc(special);
1855		uint64_t space = metaslab_class_get_space(special);
1856		uint64_t limit =
1857		    (space * (100 - zfs_special_class_metadata_reserve_pct))
1858		    / 100;
1859
1860		if (alloc < limit)
1861			return (special);
1862	}
1863
1864	return (spa_normal_class(spa));
1865}
1866
1867void
1868spa_evicting_os_register(spa_t *spa, objset_t *os)
1869{
1870	mutex_enter(&spa->spa_evicting_os_lock);
1871	list_insert_head(&spa->spa_evicting_os_list, os);
1872	mutex_exit(&spa->spa_evicting_os_lock);
1873}
1874
1875void
1876spa_evicting_os_deregister(spa_t *spa, objset_t *os)
1877{
1878	mutex_enter(&spa->spa_evicting_os_lock);
1879	list_remove(&spa->spa_evicting_os_list, os);
1880	cv_broadcast(&spa->spa_evicting_os_cv);
1881	mutex_exit(&spa->spa_evicting_os_lock);
1882}
1883
1884void
1885spa_evicting_os_wait(spa_t *spa)
1886{
1887	mutex_enter(&spa->spa_evicting_os_lock);
1888	while (!list_is_empty(&spa->spa_evicting_os_list))
1889		cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock);
1890	mutex_exit(&spa->spa_evicting_os_lock);
1891
1892	dmu_buf_user_evict_wait();
1893}
1894
1895int
1896spa_max_replication(spa_t *spa)
1897{
1898	/*
1899	 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1900	 * handle BPs with more than one DVA allocated.  Set our max
1901	 * replication level accordingly.
1902	 */
1903	if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1904		return (1);
1905	return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1906}
1907
1908int
1909spa_prev_software_version(spa_t *spa)
1910{
1911	return (spa->spa_prev_software_version);
1912}
1913
1914uint64_t
1915spa_deadman_synctime(spa_t *spa)
1916{
1917	return (spa->spa_deadman_synctime);
1918}
1919
1920uint64_t
1921dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1922{
1923	uint64_t asize = DVA_GET_ASIZE(dva);
1924	uint64_t dsize = asize;
1925
1926	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1927
1928	if (asize != 0 && spa->spa_deflate) {
1929		vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1930		dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1931	}
1932
1933	return (dsize);
1934}
1935
1936uint64_t
1937bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1938{
1939	uint64_t dsize = 0;
1940
1941	for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1942		dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1943
1944	return (dsize);
1945}
1946
1947uint64_t
1948bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1949{
1950	uint64_t dsize = 0;
1951
1952	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1953
1954	for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1955		dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1956
1957	spa_config_exit(spa, SCL_VDEV, FTAG);
1958
1959	return (dsize);
1960}
1961
1962uint64_t
1963spa_dirty_data(spa_t *spa)
1964{
1965	return (spa->spa_dsl_pool->dp_dirty_total);
1966}
1967
1968/*
1969 * ==========================================================================
1970 * Initialization and Termination
1971 * ==========================================================================
1972 */
1973
1974static int
1975spa_name_compare(const void *a1, const void *a2)
1976{
1977	const spa_t *s1 = a1;
1978	const spa_t *s2 = a2;
1979	int s;
1980
1981	s = strcmp(s1->spa_name, s2->spa_name);
1982
1983	return (AVL_ISIGN(s));
1984}
1985
1986int
1987spa_busy(void)
1988{
1989	return (spa_active_count);
1990}
1991
1992void
1993spa_boot_init()
1994{
1995	spa_config_load();
1996}
1997
1998void
1999spa_init(int mode)
2000{
2001	mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
2002	mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
2003	mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
2004	cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
2005
2006	avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
2007	    offsetof(spa_t, spa_avl));
2008
2009	avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
2010	    offsetof(spa_aux_t, aux_avl));
2011
2012	avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
2013	    offsetof(spa_aux_t, aux_avl));
2014
2015	spa_mode_global = mode;
2016
2017#ifdef _KERNEL
2018	spa_arch_init();
2019#else
2020	if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
2021		arc_procfd = open("/proc/self/ctl", O_WRONLY);
2022		if (arc_procfd == -1) {
2023			perror("could not enable watchpoints: "
2024			    "opening /proc/self/ctl failed: ");
2025		} else {
2026			arc_watch = B_TRUE;
2027		}
2028	}
2029#endif
2030
2031	zfs_refcount_init();
2032	unique_init();
2033	range_tree_init();
2034	metaslab_alloc_trace_init();
2035	zio_init();
2036	dmu_init();
2037	zil_init();
2038	vdev_cache_stat_init();
2039	vdev_mirror_stat_init();
2040	zfs_prop_init();
2041	zpool_prop_init();
2042	zpool_feature_init();
2043	spa_config_load();
2044	l2arc_start();
2045	scan_init();
2046}
2047
2048void
2049spa_fini(void)
2050{
2051	l2arc_stop();
2052
2053	spa_evict_all();
2054
2055	vdev_cache_stat_fini();
2056	vdev_mirror_stat_fini();
2057	zil_fini();
2058	dmu_fini();
2059	zio_fini();
2060	metaslab_alloc_trace_fini();
2061	range_tree_fini();
2062	unique_fini();
2063	zfs_refcount_fini();
2064	scan_fini();
2065
2066	avl_destroy(&spa_namespace_avl);
2067	avl_destroy(&spa_spare_avl);
2068	avl_destroy(&spa_l2cache_avl);
2069
2070	cv_destroy(&spa_namespace_cv);
2071	mutex_destroy(&spa_namespace_lock);
2072	mutex_destroy(&spa_spare_lock);
2073	mutex_destroy(&spa_l2cache_lock);
2074}
2075
2076/*
2077 * Return whether this pool has slogs. No locking needed.
2078 * It's not a problem if the wrong answer is returned as it's only for
2079 * performance and not correctness
2080 */
2081boolean_t
2082spa_has_slogs(spa_t *spa)
2083{
2084	return (spa->spa_log_class->mc_rotor != NULL);
2085}
2086
2087spa_log_state_t
2088spa_get_log_state(spa_t *spa)
2089{
2090	return (spa->spa_log_state);
2091}
2092
2093void
2094spa_set_log_state(spa_t *spa, spa_log_state_t state)
2095{
2096	spa->spa_log_state = state;
2097}
2098
2099boolean_t
2100spa_is_root(spa_t *spa)
2101{
2102	return (spa->spa_is_root);
2103}
2104
2105boolean_t
2106spa_writeable(spa_t *spa)
2107{
2108	return (!!(spa->spa_mode & FWRITE) && spa->spa_trust_config);
2109}
2110
2111/*
2112 * Returns true if there is a pending sync task in any of the current
2113 * syncing txg, the current quiescing txg, or the current open txg.
2114 */
2115boolean_t
2116spa_has_pending_synctask(spa_t *spa)
2117{
2118	return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks) ||
2119	    !txg_all_lists_empty(&spa->spa_dsl_pool->dp_early_sync_tasks));
2120}
2121
2122int
2123spa_mode(spa_t *spa)
2124{
2125	return (spa->spa_mode);
2126}
2127
2128uint64_t
2129spa_bootfs(spa_t *spa)
2130{
2131	return (spa->spa_bootfs);
2132}
2133
2134uint64_t
2135spa_delegation(spa_t *spa)
2136{
2137	return (spa->spa_delegation);
2138}
2139
2140objset_t *
2141spa_meta_objset(spa_t *spa)
2142{
2143	return (spa->spa_meta_objset);
2144}
2145
2146enum zio_checksum
2147spa_dedup_checksum(spa_t *spa)
2148{
2149	return (spa->spa_dedup_checksum);
2150}
2151
2152/*
2153 * Reset pool scan stat per scan pass (or reboot).
2154 */
2155void
2156spa_scan_stat_init(spa_t *spa)
2157{
2158	/* data not stored on disk */
2159	spa->spa_scan_pass_start = gethrestime_sec();
2160	if (dsl_scan_is_paused_scrub(spa->spa_dsl_pool->dp_scan))
2161		spa->spa_scan_pass_scrub_pause = spa->spa_scan_pass_start;
2162	else
2163		spa->spa_scan_pass_scrub_pause = 0;
2164	spa->spa_scan_pass_scrub_spent_paused = 0;
2165	spa->spa_scan_pass_exam = 0;
2166	spa->spa_scan_pass_issued = 0;
2167	vdev_scan_stat_init(spa->spa_root_vdev);
2168}
2169
2170/*
2171 * Get scan stats for zpool status reports
2172 */
2173int
2174spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
2175{
2176	dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
2177
2178	if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
2179		return (SET_ERROR(ENOENT));
2180	bzero(ps, sizeof (pool_scan_stat_t));
2181
2182	/* data stored on disk */
2183	ps->pss_func = scn->scn_phys.scn_func;
2184	ps->pss_state = scn->scn_phys.scn_state;
2185	ps->pss_start_time = scn->scn_phys.scn_start_time;
2186	ps->pss_end_time = scn->scn_phys.scn_end_time;
2187	ps->pss_to_examine = scn->scn_phys.scn_to_examine;
2188	ps->pss_to_process = scn->scn_phys.scn_to_process;
2189	ps->pss_processed = scn->scn_phys.scn_processed;
2190	ps->pss_errors = scn->scn_phys.scn_errors;
2191	ps->pss_examined = scn->scn_phys.scn_examined;
2192	ps->pss_issued =
2193	    scn->scn_issued_before_pass + spa->spa_scan_pass_issued;
2194	ps->pss_state = scn->scn_phys.scn_state;
2195
2196	/* data not stored on disk */
2197	ps->pss_pass_start = spa->spa_scan_pass_start;
2198	ps->pss_pass_exam = spa->spa_scan_pass_exam;
2199	ps->pss_pass_issued = spa->spa_scan_pass_issued;
2200	ps->pss_pass_scrub_pause = spa->spa_scan_pass_scrub_pause;
2201	ps->pss_pass_scrub_spent_paused = spa->spa_scan_pass_scrub_spent_paused;
2202
2203	return (0);
2204}
2205
2206int
2207spa_maxblocksize(spa_t *spa)
2208{
2209	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS))
2210		return (SPA_MAXBLOCKSIZE);
2211	else
2212		return (SPA_OLD_MAXBLOCKSIZE);
2213}
2214
2215int
2216spa_maxdnodesize(spa_t *spa)
2217{
2218	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE))
2219		return (DNODE_MAX_SIZE);
2220	else
2221		return (DNODE_MIN_SIZE);
2222}
2223
2224boolean_t
2225spa_multihost(spa_t *spa)
2226{
2227	return (spa->spa_multihost ? B_TRUE : B_FALSE);
2228}
2229
2230unsigned long
2231spa_get_hostid(void)
2232{
2233	unsigned long myhostid;
2234
2235#ifdef	_KERNEL
2236	myhostid = zone_get_hostid(NULL);
2237#else	/* _KERNEL */
2238	/*
2239	 * We're emulating the system's hostid in userland, so
2240	 * we can't use zone_get_hostid().
2241	 */
2242	(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2243#endif	/* _KERNEL */
2244
2245	return (myhostid);
2246}
2247
2248/*
2249 * Returns the txg that the last device removal completed. No indirect mappings
2250 * have been added since this txg.
2251 */
2252uint64_t
2253spa_get_last_removal_txg(spa_t *spa)
2254{
2255	uint64_t vdevid;
2256	uint64_t ret = -1ULL;
2257
2258	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2259	/*
2260	 * sr_prev_indirect_vdev is only modified while holding all the
2261	 * config locks, so it is sufficient to hold SCL_VDEV as reader when
2262	 * examining it.
2263	 */
2264	vdevid = spa->spa_removing_phys.sr_prev_indirect_vdev;
2265
2266	while (vdevid != -1ULL) {
2267		vdev_t *vd = vdev_lookup_top(spa, vdevid);
2268		vdev_indirect_births_t *vib = vd->vdev_indirect_births;
2269
2270		ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops);
2271
2272		/*
2273		 * If the removal did not remap any data, we don't care.
2274		 */
2275		if (vdev_indirect_births_count(vib) != 0) {
2276			ret = vdev_indirect_births_last_entry_txg(vib);
2277			break;
2278		}
2279
2280		vdevid = vd->vdev_indirect_config.vic_prev_indirect_vdev;
2281	}
2282	spa_config_exit(spa, SCL_VDEV, FTAG);
2283
2284	IMPLY(ret != -1ULL,
2285	    spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL));
2286
2287	return (ret);
2288}
2289
2290boolean_t
2291spa_trust_config(spa_t *spa)
2292{
2293	return (spa->spa_trust_config);
2294}
2295
2296uint64_t
2297spa_missing_tvds_allowed(spa_t *spa)
2298{
2299	return (spa->spa_missing_tvds_allowed);
2300}
2301
2302void
2303spa_set_missing_tvds(spa_t *spa, uint64_t missing)
2304{
2305	spa->spa_missing_tvds = missing;
2306}
2307
2308boolean_t
2309spa_top_vdevs_spacemap_addressable(spa_t *spa)
2310{
2311	vdev_t *rvd = spa->spa_root_vdev;
2312	for (uint64_t c = 0; c < rvd->vdev_children; c++) {
2313		if (!vdev_is_spacemap_addressable(rvd->vdev_child[c]))
2314			return (B_FALSE);
2315	}
2316	return (B_TRUE);
2317}
2318
2319boolean_t
2320spa_has_checkpoint(spa_t *spa)
2321{
2322	return (spa->spa_checkpoint_txg != 0);
2323}
2324
2325boolean_t
2326spa_importing_readonly_checkpoint(spa_t *spa)
2327{
2328	return ((spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT) &&
2329	    spa->spa_mode == FREAD);
2330}
2331
2332uint64_t
2333spa_min_claim_txg(spa_t *spa)
2334{
2335	uint64_t checkpoint_txg = spa->spa_uberblock.ub_checkpoint_txg;
2336
2337	if (checkpoint_txg != 0)
2338		return (checkpoint_txg + 1);
2339
2340	return (spa->spa_first_txg);
2341}
2342
2343/*
2344 * If there is a checkpoint, async destroys may consume more space from
2345 * the pool instead of freeing it. In an attempt to save the pool from
2346 * getting suspended when it is about to run out of space, we stop
2347 * processing async destroys.
2348 */
2349boolean_t
2350spa_suspend_async_destroy(spa_t *spa)
2351{
2352	dsl_pool_t *dp = spa_get_dsl(spa);
2353
2354	uint64_t unreserved = dsl_pool_unreserved_space(dp,
2355	    ZFS_SPACE_CHECK_EXTRA_RESERVED);
2356	uint64_t used = dsl_dir_phys(dp->dp_root_dir)->dd_used_bytes;
2357	uint64_t avail = (unreserved > used) ? (unreserved - used) : 0;
2358
2359	if (spa_has_checkpoint(spa) && avail == 0)
2360		return (B_TRUE);
2361
2362	return (B_FALSE);
2363}
2364