spa_misc.c revision 21f7c81cc1156e9202ce3412d3ecaa697c3b2222
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 */
30
31#include <sys/zfs_context.h>
32#include <sys/spa_impl.h>
33#include <sys/spa_boot.h>
34#include <sys/zio.h>
35#include <sys/zio_checksum.h>
36#include <sys/zio_compress.h>
37#include <sys/dmu.h>
38#include <sys/dmu_tx.h>
39#include <sys/zap.h>
40#include <sys/zil.h>
41#include <sys/vdev_impl.h>
42#include <sys/metaslab.h>
43#include <sys/uberblock_impl.h>
44#include <sys/txg.h>
45#include <sys/avl.h>
46#include <sys/unique.h>
47#include <sys/dsl_pool.h>
48#include <sys/dsl_dir.h>
49#include <sys/dsl_prop.h>
50#include <sys/dsl_scan.h>
51#include <sys/fs/zfs.h>
52#include <sys/metaslab_impl.h>
53#include <sys/arc.h>
54#include <sys/ddt.h>
55#include "zfs_prop.h"
56#include <sys/zfeature.h>
57
58/*
59 * SPA locking
60 *
61 * There are four basic locks for managing spa_t structures:
62 *
63 * spa_namespace_lock (global mutex)
64 *
65 *	This lock must be acquired to do any of the following:
66 *
67 *		- Lookup a spa_t by name
68 *		- Add or remove a spa_t from the namespace
69 *		- Increase spa_refcount from non-zero
70 *		- Check if spa_refcount is zero
71 *		- Rename a spa_t
72 *		- add/remove/attach/detach devices
73 *		- Held for the duration of create/destroy/import/export
74 *
75 *	It does not need to handle recursion.  A create or destroy may
76 *	reference objects (files or zvols) in other pools, but by
77 *	definition they must have an existing reference, and will never need
78 *	to lookup a spa_t by name.
79 *
80 * spa_refcount (per-spa refcount_t protected by mutex)
81 *
82 *	This reference count keep track of any active users of the spa_t.  The
83 *	spa_t cannot be destroyed or freed while this is non-zero.  Internally,
84 *	the refcount is never really 'zero' - opening a pool implicitly keeps
85 *	some references in the DMU.  Internally we check against spa_minref, but
86 *	present the image of a zero/non-zero value to consumers.
87 *
88 * spa_config_lock[] (per-spa array of rwlocks)
89 *
90 *	This protects the spa_t from config changes, and must be held in
91 *	the following circumstances:
92 *
93 *		- RW_READER to perform I/O to the spa
94 *		- RW_WRITER to change the vdev config
95 *
96 * The locking order is fairly straightforward:
97 *
98 *		spa_namespace_lock	->	spa_refcount
99 *
100 *	The namespace lock must be acquired to increase the refcount from 0
101 *	or to check if it is zero.
102 *
103 *		spa_refcount		->	spa_config_lock[]
104 *
105 *	There must be at least one valid reference on the spa_t to acquire
106 *	the config lock.
107 *
108 *		spa_namespace_lock	->	spa_config_lock[]
109 *
110 *	The namespace lock must always be taken before the config lock.
111 *
112 *
113 * The spa_namespace_lock can be acquired directly and is globally visible.
114 *
115 * The namespace is manipulated using the following functions, all of which
116 * require the spa_namespace_lock to be held.
117 *
118 *	spa_lookup()		Lookup a spa_t by name.
119 *
120 *	spa_add()		Create a new spa_t in the namespace.
121 *
122 *	spa_remove()		Remove a spa_t from the namespace.  This also
123 *				frees up any memory associated with the spa_t.
124 *
125 *	spa_next()		Returns the next spa_t in the system, or the
126 *				first if NULL is passed.
127 *
128 *	spa_evict_all()		Shutdown and remove all spa_t structures in
129 *				the system.
130 *
131 *	spa_guid_exists()	Determine whether a pool/device guid exists.
132 *
133 * The spa_refcount is manipulated using the following functions:
134 *
135 *	spa_open_ref()		Adds a reference to the given spa_t.  Must be
136 *				called with spa_namespace_lock held if the
137 *				refcount is currently zero.
138 *
139 *	spa_close()		Remove a reference from the spa_t.  This will
140 *				not free the spa_t or remove it from the
141 *				namespace.  No locking is required.
142 *
143 *	spa_refcount_zero()	Returns true if the refcount is currently
144 *				zero.  Must be called with spa_namespace_lock
145 *				held.
146 *
147 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
148 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
149 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
150 *
151 * To read the configuration, it suffices to hold one of these locks as reader.
152 * To modify the configuration, you must hold all locks as writer.  To modify
153 * vdev state without altering the vdev tree's topology (e.g. online/offline),
154 * you must hold SCL_STATE and SCL_ZIO as writer.
155 *
156 * We use these distinct config locks to avoid recursive lock entry.
157 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
158 * block allocations (SCL_ALLOC), which may require reading space maps
159 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
160 *
161 * The spa config locks cannot be normal rwlocks because we need the
162 * ability to hand off ownership.  For example, SCL_ZIO is acquired
163 * by the issuing thread and later released by an interrupt thread.
164 * They do, however, obey the usual write-wanted semantics to prevent
165 * writer (i.e. system administrator) starvation.
166 *
167 * The lock acquisition rules are as follows:
168 *
169 * SCL_CONFIG
170 *	Protects changes to the vdev tree topology, such as vdev
171 *	add/remove/attach/detach.  Protects the dirty config list
172 *	(spa_config_dirty_list) and the set of spares and l2arc devices.
173 *
174 * SCL_STATE
175 *	Protects changes to pool state and vdev state, such as vdev
176 *	online/offline/fault/degrade/clear.  Protects the dirty state list
177 *	(spa_state_dirty_list) and global pool state (spa_state).
178 *
179 * SCL_ALLOC
180 *	Protects changes to metaslab groups and classes.
181 *	Held as reader by metaslab_alloc() and metaslab_claim().
182 *
183 * SCL_ZIO
184 *	Held by bp-level zios (those which have no io_vd upon entry)
185 *	to prevent changes to the vdev tree.  The bp-level zio implicitly
186 *	protects all of its vdev child zios, which do not hold SCL_ZIO.
187 *
188 * SCL_FREE
189 *	Protects changes to metaslab groups and classes.
190 *	Held as reader by metaslab_free().  SCL_FREE is distinct from
191 *	SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
192 *	blocks in zio_done() while another i/o that holds either
193 *	SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
194 *
195 * SCL_VDEV
196 *	Held as reader to prevent changes to the vdev tree during trivial
197 *	inquiries such as bp_get_dsize().  SCL_VDEV is distinct from the
198 *	other locks, and lower than all of them, to ensure that it's safe
199 *	to acquire regardless of caller context.
200 *
201 * In addition, the following rules apply:
202 *
203 * (a)	spa_props_lock protects pool properties, spa_config and spa_config_list.
204 *	The lock ordering is SCL_CONFIG > spa_props_lock.
205 *
206 * (b)	I/O operations on leaf vdevs.  For any zio operation that takes
207 *	an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
208 *	or zio_write_phys() -- the caller must ensure that the config cannot
209 *	cannot change in the interim, and that the vdev cannot be reopened.
210 *	SCL_STATE as reader suffices for both.
211 *
212 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
213 *
214 *	spa_vdev_enter()	Acquire the namespace lock and the config lock
215 *				for writing.
216 *
217 *	spa_vdev_exit()		Release the config lock, wait for all I/O
218 *				to complete, sync the updated configs to the
219 *				cache, and release the namespace lock.
220 *
221 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
222 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
223 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
224 *
225 * spa_rename() is also implemented within this file since it requires
226 * manipulation of the namespace.
227 */
228
229static avl_tree_t spa_namespace_avl;
230kmutex_t spa_namespace_lock;
231static kcondvar_t spa_namespace_cv;
232static int spa_active_count;
233int spa_max_replication_override = SPA_DVAS_PER_BP;
234
235static kmutex_t spa_spare_lock;
236static avl_tree_t spa_spare_avl;
237static kmutex_t spa_l2cache_lock;
238static avl_tree_t spa_l2cache_avl;
239
240kmem_cache_t *spa_buffer_pool;
241int spa_mode_global;
242
243#ifdef ZFS_DEBUG
244/*
245 * Everything except dprintf, spa, and indirect_remap is on by default
246 * in debug builds.
247 */
248int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_INDIRECT_REMAP);
249#else
250int zfs_flags = 0;
251#endif
252
253/*
254 * zfs_recover can be set to nonzero to attempt to recover from
255 * otherwise-fatal errors, typically caused by on-disk corruption.  When
256 * set, calls to zfs_panic_recover() will turn into warning messages.
257 * This should only be used as a last resort, as it typically results
258 * in leaked space, or worse.
259 */
260boolean_t zfs_recover = B_FALSE;
261
262/*
263 * If destroy encounters an EIO while reading metadata (e.g. indirect
264 * blocks), space referenced by the missing metadata can not be freed.
265 * Normally this causes the background destroy to become "stalled", as
266 * it is unable to make forward progress.  While in this stalled state,
267 * all remaining space to free from the error-encountering filesystem is
268 * "temporarily leaked".  Set this flag to cause it to ignore the EIO,
269 * permanently leak the space from indirect blocks that can not be read,
270 * and continue to free everything else that it can.
271 *
272 * The default, "stalling" behavior is useful if the storage partially
273 * fails (i.e. some but not all i/os fail), and then later recovers.  In
274 * this case, we will be able to continue pool operations while it is
275 * partially failed, and when it recovers, we can continue to free the
276 * space, with no leaks.  However, note that this case is actually
277 * fairly rare.
278 *
279 * Typically pools either (a) fail completely (but perhaps temporarily,
280 * e.g. a top-level vdev going offline), or (b) have localized,
281 * permanent errors (e.g. disk returns the wrong data due to bit flip or
282 * firmware bug).  In case (a), this setting does not matter because the
283 * pool will be suspended and the sync thread will not be able to make
284 * forward progress regardless.  In case (b), because the error is
285 * permanent, the best we can do is leak the minimum amount of space,
286 * which is what setting this flag will do.  Therefore, it is reasonable
287 * for this flag to normally be set, but we chose the more conservative
288 * approach of not setting it, so that there is no possibility of
289 * leaking space in the "partial temporary" failure case.
290 */
291boolean_t zfs_free_leak_on_eio = B_FALSE;
292
293/*
294 * Expiration time in milliseconds. This value has two meanings. First it is
295 * used to determine when the spa_deadman() logic should fire. By default the
296 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
297 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
298 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
299 * in a system panic.
300 */
301uint64_t zfs_deadman_synctime_ms = 1000000ULL;
302
303/*
304 * Check time in milliseconds. This defines the frequency at which we check
305 * for hung I/O.
306 */
307uint64_t zfs_deadman_checktime_ms = 5000ULL;
308
309/*
310 * Override the zfs deadman behavior via /etc/system. By default the
311 * deadman is enabled except on VMware and sparc deployments.
312 */
313int zfs_deadman_enabled = -1;
314
315/*
316 * The worst case is single-sector max-parity RAID-Z blocks, in which
317 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
318 * times the size; so just assume that.  Add to this the fact that
319 * we can have up to 3 DVAs per bp, and one more factor of 2 because
320 * the block may be dittoed with up to 3 DVAs by ddt_sync().  All together,
321 * the worst case is:
322 *     (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
323 */
324int spa_asize_inflation = 24;
325
326/*
327 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
328 * the pool to be consumed.  This ensures that we don't run the pool
329 * completely out of space, due to unaccounted changes (e.g. to the MOS).
330 * It also limits the worst-case time to allocate space.  If we have
331 * less than this amount of free space, most ZPL operations (e.g. write,
332 * create) will return ENOSPC.
333 *
334 * Certain operations (e.g. file removal, most administrative actions) can
335 * use half the slop space.  They will only return ENOSPC if less than half
336 * the slop space is free.  Typically, once the pool has less than the slop
337 * space free, the user will use these operations to free up space in the pool.
338 * These are the operations that call dsl_pool_adjustedsize() with the netfree
339 * argument set to TRUE.
340 *
341 * Operations that are almost guaranteed to free up space in the absence of
342 * a pool checkpoint can use up to three quarters of the slop space
343 * (e.g zfs destroy).
344 *
345 * A very restricted set of operations are always permitted, regardless of
346 * the amount of free space.  These are the operations that call
347 * dsl_sync_task(ZFS_SPACE_CHECK_NONE). If these operations result in a net
348 * increase in the amount of space used, it is possible to run the pool
349 * completely out of space, causing it to be permanently read-only.
350 *
351 * Note that on very small pools, the slop space will be larger than
352 * 3.2%, in an effort to have it be at least spa_min_slop (128MB),
353 * but we never allow it to be more than half the pool size.
354 *
355 * See also the comments in zfs_space_check_t.
356 */
357int spa_slop_shift = 5;
358uint64_t spa_min_slop = 128 * 1024 * 1024;
359
360int spa_allocators = 4;
361
362/*PRINTFLIKE2*/
363void
364spa_load_failed(spa_t *spa, const char *fmt, ...)
365{
366	va_list adx;
367	char buf[256];
368
369	va_start(adx, fmt);
370	(void) vsnprintf(buf, sizeof (buf), fmt, adx);
371	va_end(adx);
372
373	zfs_dbgmsg("spa_load(%s, config %s): FAILED: %s", spa->spa_name,
374	    spa->spa_trust_config ? "trusted" : "untrusted", buf);
375}
376
377/*PRINTFLIKE2*/
378void
379spa_load_note(spa_t *spa, const char *fmt, ...)
380{
381	va_list adx;
382	char buf[256];
383
384	va_start(adx, fmt);
385	(void) vsnprintf(buf, sizeof (buf), fmt, adx);
386	va_end(adx);
387
388	zfs_dbgmsg("spa_load(%s, config %s): %s", spa->spa_name,
389	    spa->spa_trust_config ? "trusted" : "untrusted", buf);
390}
391
392/*
393 * ==========================================================================
394 * SPA config locking
395 * ==========================================================================
396 */
397static void
398spa_config_lock_init(spa_t *spa)
399{
400	for (int i = 0; i < SCL_LOCKS; i++) {
401		spa_config_lock_t *scl = &spa->spa_config_lock[i];
402		mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
403		cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
404		refcount_create_untracked(&scl->scl_count);
405		scl->scl_writer = NULL;
406		scl->scl_write_wanted = 0;
407	}
408}
409
410static void
411spa_config_lock_destroy(spa_t *spa)
412{
413	for (int i = 0; i < SCL_LOCKS; i++) {
414		spa_config_lock_t *scl = &spa->spa_config_lock[i];
415		mutex_destroy(&scl->scl_lock);
416		cv_destroy(&scl->scl_cv);
417		refcount_destroy(&scl->scl_count);
418		ASSERT(scl->scl_writer == NULL);
419		ASSERT(scl->scl_write_wanted == 0);
420	}
421}
422
423int
424spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
425{
426	for (int i = 0; i < SCL_LOCKS; i++) {
427		spa_config_lock_t *scl = &spa->spa_config_lock[i];
428		if (!(locks & (1 << i)))
429			continue;
430		mutex_enter(&scl->scl_lock);
431		if (rw == RW_READER) {
432			if (scl->scl_writer || scl->scl_write_wanted) {
433				mutex_exit(&scl->scl_lock);
434				spa_config_exit(spa, locks & ((1 << i) - 1),
435				    tag);
436				return (0);
437			}
438		} else {
439			ASSERT(scl->scl_writer != curthread);
440			if (!refcount_is_zero(&scl->scl_count)) {
441				mutex_exit(&scl->scl_lock);
442				spa_config_exit(spa, locks & ((1 << i) - 1),
443				    tag);
444				return (0);
445			}
446			scl->scl_writer = curthread;
447		}
448		(void) refcount_add(&scl->scl_count, tag);
449		mutex_exit(&scl->scl_lock);
450	}
451	return (1);
452}
453
454void
455spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
456{
457	int wlocks_held = 0;
458
459	ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
460
461	for (int i = 0; i < SCL_LOCKS; i++) {
462		spa_config_lock_t *scl = &spa->spa_config_lock[i];
463		if (scl->scl_writer == curthread)
464			wlocks_held |= (1 << i);
465		if (!(locks & (1 << i)))
466			continue;
467		mutex_enter(&scl->scl_lock);
468		if (rw == RW_READER) {
469			while (scl->scl_writer || scl->scl_write_wanted) {
470				cv_wait(&scl->scl_cv, &scl->scl_lock);
471			}
472		} else {
473			ASSERT(scl->scl_writer != curthread);
474			while (!refcount_is_zero(&scl->scl_count)) {
475				scl->scl_write_wanted++;
476				cv_wait(&scl->scl_cv, &scl->scl_lock);
477				scl->scl_write_wanted--;
478			}
479			scl->scl_writer = curthread;
480		}
481		(void) refcount_add(&scl->scl_count, tag);
482		mutex_exit(&scl->scl_lock);
483	}
484	ASSERT3U(wlocks_held, <=, locks);
485}
486
487void
488spa_config_exit(spa_t *spa, int locks, void *tag)
489{
490	for (int i = SCL_LOCKS - 1; i >= 0; i--) {
491		spa_config_lock_t *scl = &spa->spa_config_lock[i];
492		if (!(locks & (1 << i)))
493			continue;
494		mutex_enter(&scl->scl_lock);
495		ASSERT(!refcount_is_zero(&scl->scl_count));
496		if (refcount_remove(&scl->scl_count, tag) == 0) {
497			ASSERT(scl->scl_writer == NULL ||
498			    scl->scl_writer == curthread);
499			scl->scl_writer = NULL;	/* OK in either case */
500			cv_broadcast(&scl->scl_cv);
501		}
502		mutex_exit(&scl->scl_lock);
503	}
504}
505
506int
507spa_config_held(spa_t *spa, int locks, krw_t rw)
508{
509	int locks_held = 0;
510
511	for (int i = 0; i < SCL_LOCKS; i++) {
512		spa_config_lock_t *scl = &spa->spa_config_lock[i];
513		if (!(locks & (1 << i)))
514			continue;
515		if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
516		    (rw == RW_WRITER && scl->scl_writer == curthread))
517			locks_held |= 1 << i;
518	}
519
520	return (locks_held);
521}
522
523/*
524 * ==========================================================================
525 * SPA namespace functions
526 * ==========================================================================
527 */
528
529/*
530 * Lookup the named spa_t in the AVL tree.  The spa_namespace_lock must be held.
531 * Returns NULL if no matching spa_t is found.
532 */
533spa_t *
534spa_lookup(const char *name)
535{
536	static spa_t search;	/* spa_t is large; don't allocate on stack */
537	spa_t *spa;
538	avl_index_t where;
539	char *cp;
540
541	ASSERT(MUTEX_HELD(&spa_namespace_lock));
542
543	(void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
544
545	/*
546	 * If it's a full dataset name, figure out the pool name and
547	 * just use that.
548	 */
549	cp = strpbrk(search.spa_name, "/@#");
550	if (cp != NULL)
551		*cp = '\0';
552
553	spa = avl_find(&spa_namespace_avl, &search, &where);
554
555	return (spa);
556}
557
558/*
559 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
560 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
561 * looking for potentially hung I/Os.
562 */
563void
564spa_deadman(void *arg)
565{
566	spa_t *spa = arg;
567
568	/*
569	 * Disable the deadman timer if the pool is suspended.
570	 */
571	if (spa_suspended(spa)) {
572		VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
573		return;
574	}
575
576	zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
577	    (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
578	    ++spa->spa_deadman_calls);
579	if (zfs_deadman_enabled)
580		vdev_deadman(spa->spa_root_vdev);
581}
582
583/*
584 * Create an uninitialized spa_t with the given name.  Requires
585 * spa_namespace_lock.  The caller must ensure that the spa_t doesn't already
586 * exist by calling spa_lookup() first.
587 */
588spa_t *
589spa_add(const char *name, nvlist_t *config, const char *altroot)
590{
591	spa_t *spa;
592	spa_config_dirent_t *dp;
593	cyc_handler_t hdlr;
594	cyc_time_t when;
595
596	ASSERT(MUTEX_HELD(&spa_namespace_lock));
597
598	spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
599
600	mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
601	mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
602	mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
603	mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL);
604	mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
605	mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
606	mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
607	mutex_init(&spa->spa_cksum_tmpls_lock, NULL, MUTEX_DEFAULT, NULL);
608	mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
609	mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
610	mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
611	mutex_init(&spa->spa_iokstat_lock, NULL, MUTEX_DEFAULT, NULL);
612
613	cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
614	cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL);
615	cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
616	cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
617	cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
618
619	for (int t = 0; t < TXG_SIZE; t++)
620		bplist_create(&spa->spa_free_bplist[t]);
621
622	(void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
623	spa->spa_state = POOL_STATE_UNINITIALIZED;
624	spa->spa_freeze_txg = UINT64_MAX;
625	spa->spa_final_txg = UINT64_MAX;
626	spa->spa_load_max_txg = UINT64_MAX;
627	spa->spa_proc = &p0;
628	spa->spa_proc_state = SPA_PROC_NONE;
629	spa->spa_trust_config = B_TRUE;
630
631	hdlr.cyh_func = spa_deadman;
632	hdlr.cyh_arg = spa;
633	hdlr.cyh_level = CY_LOW_LEVEL;
634
635	spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
636
637	/*
638	 * This determines how often we need to check for hung I/Os after
639	 * the cyclic has already fired. Since checking for hung I/Os is
640	 * an expensive operation we don't want to check too frequently.
641	 * Instead wait for 5 seconds before checking again.
642	 */
643	when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms);
644	when.cyt_when = CY_INFINITY;
645	mutex_enter(&cpu_lock);
646	spa->spa_deadman_cycid = cyclic_add(&hdlr, &when);
647	mutex_exit(&cpu_lock);
648
649	refcount_create(&spa->spa_refcount);
650	spa_config_lock_init(spa);
651
652	avl_add(&spa_namespace_avl, spa);
653
654	/*
655	 * Set the alternate root, if there is one.
656	 */
657	if (altroot) {
658		spa->spa_root = spa_strdup(altroot);
659		spa_active_count++;
660	}
661
662	spa->spa_alloc_count = spa_allocators;
663	spa->spa_alloc_locks = kmem_zalloc(spa->spa_alloc_count *
664	    sizeof (kmutex_t), KM_SLEEP);
665	spa->spa_alloc_trees = kmem_zalloc(spa->spa_alloc_count *
666	    sizeof (avl_tree_t), KM_SLEEP);
667	for (int i = 0; i < spa->spa_alloc_count; i++) {
668		mutex_init(&spa->spa_alloc_locks[i], NULL, MUTEX_DEFAULT, NULL);
669		avl_create(&spa->spa_alloc_trees[i], zio_bookmark_compare,
670		    sizeof (zio_t), offsetof(zio_t, io_alloc_node));
671	}
672
673	/*
674	 * Every pool starts with the default cachefile
675	 */
676	list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
677	    offsetof(spa_config_dirent_t, scd_link));
678
679	dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
680	dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
681	list_insert_head(&spa->spa_config_list, dp);
682
683	VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
684	    KM_SLEEP) == 0);
685
686	if (config != NULL) {
687		nvlist_t *features;
688
689		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
690		    &features) == 0) {
691			VERIFY(nvlist_dup(features, &spa->spa_label_features,
692			    0) == 0);
693		}
694
695		VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
696	}
697
698	if (spa->spa_label_features == NULL) {
699		VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
700		    KM_SLEEP) == 0);
701	}
702
703	spa->spa_iokstat = kstat_create("zfs", 0, name,
704	    "disk", KSTAT_TYPE_IO, 1, 0);
705	if (spa->spa_iokstat) {
706		spa->spa_iokstat->ks_lock = &spa->spa_iokstat_lock;
707		kstat_install(spa->spa_iokstat);
708	}
709
710	spa->spa_min_ashift = INT_MAX;
711	spa->spa_max_ashift = 0;
712
713	/*
714	 * As a pool is being created, treat all features as disabled by
715	 * setting SPA_FEATURE_DISABLED for all entries in the feature
716	 * refcount cache.
717	 */
718	for (int i = 0; i < SPA_FEATURES; i++) {
719		spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED;
720	}
721
722	return (spa);
723}
724
725/*
726 * Removes a spa_t from the namespace, freeing up any memory used.  Requires
727 * spa_namespace_lock.  This is called only after the spa_t has been closed and
728 * deactivated.
729 */
730void
731spa_remove(spa_t *spa)
732{
733	spa_config_dirent_t *dp;
734
735	ASSERT(MUTEX_HELD(&spa_namespace_lock));
736	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
737	ASSERT3U(refcount_count(&spa->spa_refcount), ==, 0);
738
739	nvlist_free(spa->spa_config_splitting);
740
741	avl_remove(&spa_namespace_avl, spa);
742	cv_broadcast(&spa_namespace_cv);
743
744	if (spa->spa_root) {
745		spa_strfree(spa->spa_root);
746		spa_active_count--;
747	}
748
749	while ((dp = list_head(&spa->spa_config_list)) != NULL) {
750		list_remove(&spa->spa_config_list, dp);
751		if (dp->scd_path != NULL)
752			spa_strfree(dp->scd_path);
753		kmem_free(dp, sizeof (spa_config_dirent_t));
754	}
755
756	for (int i = 0; i < spa->spa_alloc_count; i++) {
757		avl_destroy(&spa->spa_alloc_trees[i]);
758		mutex_destroy(&spa->spa_alloc_locks[i]);
759	}
760	kmem_free(spa->spa_alloc_locks, spa->spa_alloc_count *
761	    sizeof (kmutex_t));
762	kmem_free(spa->spa_alloc_trees, spa->spa_alloc_count *
763	    sizeof (avl_tree_t));
764
765	list_destroy(&spa->spa_config_list);
766
767	nvlist_free(spa->spa_label_features);
768	nvlist_free(spa->spa_load_info);
769	spa_config_set(spa, NULL);
770
771	mutex_enter(&cpu_lock);
772	if (spa->spa_deadman_cycid != CYCLIC_NONE)
773		cyclic_remove(spa->spa_deadman_cycid);
774	mutex_exit(&cpu_lock);
775	spa->spa_deadman_cycid = CYCLIC_NONE;
776
777	refcount_destroy(&spa->spa_refcount);
778
779	spa_config_lock_destroy(spa);
780
781	kstat_delete(spa->spa_iokstat);
782	spa->spa_iokstat = NULL;
783
784	for (int t = 0; t < TXG_SIZE; t++)
785		bplist_destroy(&spa->spa_free_bplist[t]);
786
787	zio_checksum_templates_free(spa);
788
789	cv_destroy(&spa->spa_async_cv);
790	cv_destroy(&spa->spa_evicting_os_cv);
791	cv_destroy(&spa->spa_proc_cv);
792	cv_destroy(&spa->spa_scrub_io_cv);
793	cv_destroy(&spa->spa_suspend_cv);
794
795	mutex_destroy(&spa->spa_async_lock);
796	mutex_destroy(&spa->spa_errlist_lock);
797	mutex_destroy(&spa->spa_errlog_lock);
798	mutex_destroy(&spa->spa_evicting_os_lock);
799	mutex_destroy(&spa->spa_history_lock);
800	mutex_destroy(&spa->spa_proc_lock);
801	mutex_destroy(&spa->spa_props_lock);
802	mutex_destroy(&spa->spa_cksum_tmpls_lock);
803	mutex_destroy(&spa->spa_scrub_lock);
804	mutex_destroy(&spa->spa_suspend_lock);
805	mutex_destroy(&spa->spa_vdev_top_lock);
806	mutex_destroy(&spa->spa_iokstat_lock);
807
808	kmem_free(spa, sizeof (spa_t));
809}
810
811/*
812 * Given a pool, return the next pool in the namespace, or NULL if there is
813 * none.  If 'prev' is NULL, return the first pool.
814 */
815spa_t *
816spa_next(spa_t *prev)
817{
818	ASSERT(MUTEX_HELD(&spa_namespace_lock));
819
820	if (prev)
821		return (AVL_NEXT(&spa_namespace_avl, prev));
822	else
823		return (avl_first(&spa_namespace_avl));
824}
825
826/*
827 * ==========================================================================
828 * SPA refcount functions
829 * ==========================================================================
830 */
831
832/*
833 * Add a reference to the given spa_t.  Must have at least one reference, or
834 * have the namespace lock held.
835 */
836void
837spa_open_ref(spa_t *spa, void *tag)
838{
839	ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
840	    MUTEX_HELD(&spa_namespace_lock));
841	(void) refcount_add(&spa->spa_refcount, tag);
842}
843
844/*
845 * Remove a reference to the given spa_t.  Must have at least one reference, or
846 * have the namespace lock held.
847 */
848void
849spa_close(spa_t *spa, void *tag)
850{
851	ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
852	    MUTEX_HELD(&spa_namespace_lock));
853	(void) refcount_remove(&spa->spa_refcount, tag);
854}
855
856/*
857 * Remove a reference to the given spa_t held by a dsl dir that is
858 * being asynchronously released.  Async releases occur from a taskq
859 * performing eviction of dsl datasets and dirs.  The namespace lock
860 * isn't held and the hold by the object being evicted may contribute to
861 * spa_minref (e.g. dataset or directory released during pool export),
862 * so the asserts in spa_close() do not apply.
863 */
864void
865spa_async_close(spa_t *spa, void *tag)
866{
867	(void) refcount_remove(&spa->spa_refcount, tag);
868}
869
870/*
871 * Check to see if the spa refcount is zero.  Must be called with
872 * spa_namespace_lock held.  We really compare against spa_minref, which is the
873 * number of references acquired when opening a pool
874 */
875boolean_t
876spa_refcount_zero(spa_t *spa)
877{
878	ASSERT(MUTEX_HELD(&spa_namespace_lock));
879
880	return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
881}
882
883/*
884 * ==========================================================================
885 * SPA spare and l2cache tracking
886 * ==========================================================================
887 */
888
889/*
890 * Hot spares and cache devices are tracked using the same code below,
891 * for 'auxiliary' devices.
892 */
893
894typedef struct spa_aux {
895	uint64_t	aux_guid;
896	uint64_t	aux_pool;
897	avl_node_t	aux_avl;
898	int		aux_count;
899} spa_aux_t;
900
901static int
902spa_aux_compare(const void *a, const void *b)
903{
904	const spa_aux_t *sa = a;
905	const spa_aux_t *sb = b;
906
907	if (sa->aux_guid < sb->aux_guid)
908		return (-1);
909	else if (sa->aux_guid > sb->aux_guid)
910		return (1);
911	else
912		return (0);
913}
914
915void
916spa_aux_add(vdev_t *vd, avl_tree_t *avl)
917{
918	avl_index_t where;
919	spa_aux_t search;
920	spa_aux_t *aux;
921
922	search.aux_guid = vd->vdev_guid;
923	if ((aux = avl_find(avl, &search, &where)) != NULL) {
924		aux->aux_count++;
925	} else {
926		aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
927		aux->aux_guid = vd->vdev_guid;
928		aux->aux_count = 1;
929		avl_insert(avl, aux, where);
930	}
931}
932
933void
934spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
935{
936	spa_aux_t search;
937	spa_aux_t *aux;
938	avl_index_t where;
939
940	search.aux_guid = vd->vdev_guid;
941	aux = avl_find(avl, &search, &where);
942
943	ASSERT(aux != NULL);
944
945	if (--aux->aux_count == 0) {
946		avl_remove(avl, aux);
947		kmem_free(aux, sizeof (spa_aux_t));
948	} else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
949		aux->aux_pool = 0ULL;
950	}
951}
952
953boolean_t
954spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
955{
956	spa_aux_t search, *found;
957
958	search.aux_guid = guid;
959	found = avl_find(avl, &search, NULL);
960
961	if (pool) {
962		if (found)
963			*pool = found->aux_pool;
964		else
965			*pool = 0ULL;
966	}
967
968	if (refcnt) {
969		if (found)
970			*refcnt = found->aux_count;
971		else
972			*refcnt = 0;
973	}
974
975	return (found != NULL);
976}
977
978void
979spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
980{
981	spa_aux_t search, *found;
982	avl_index_t where;
983
984	search.aux_guid = vd->vdev_guid;
985	found = avl_find(avl, &search, &where);
986	ASSERT(found != NULL);
987	ASSERT(found->aux_pool == 0ULL);
988
989	found->aux_pool = spa_guid(vd->vdev_spa);
990}
991
992/*
993 * Spares are tracked globally due to the following constraints:
994 *
995 * 	- A spare may be part of multiple pools.
996 * 	- A spare may be added to a pool even if it's actively in use within
997 *	  another pool.
998 * 	- A spare in use in any pool can only be the source of a replacement if
999 *	  the target is a spare in the same pool.
1000 *
1001 * We keep track of all spares on the system through the use of a reference
1002 * counted AVL tree.  When a vdev is added as a spare, or used as a replacement
1003 * spare, then we bump the reference count in the AVL tree.  In addition, we set
1004 * the 'vdev_isspare' member to indicate that the device is a spare (active or
1005 * inactive).  When a spare is made active (used to replace a device in the
1006 * pool), we also keep track of which pool its been made a part of.
1007 *
1008 * The 'spa_spare_lock' protects the AVL tree.  These functions are normally
1009 * called under the spa_namespace lock as part of vdev reconfiguration.  The
1010 * separate spare lock exists for the status query path, which does not need to
1011 * be completely consistent with respect to other vdev configuration changes.
1012 */
1013
1014static int
1015spa_spare_compare(const void *a, const void *b)
1016{
1017	return (spa_aux_compare(a, b));
1018}
1019
1020void
1021spa_spare_add(vdev_t *vd)
1022{
1023	mutex_enter(&spa_spare_lock);
1024	ASSERT(!vd->vdev_isspare);
1025	spa_aux_add(vd, &spa_spare_avl);
1026	vd->vdev_isspare = B_TRUE;
1027	mutex_exit(&spa_spare_lock);
1028}
1029
1030void
1031spa_spare_remove(vdev_t *vd)
1032{
1033	mutex_enter(&spa_spare_lock);
1034	ASSERT(vd->vdev_isspare);
1035	spa_aux_remove(vd, &spa_spare_avl);
1036	vd->vdev_isspare = B_FALSE;
1037	mutex_exit(&spa_spare_lock);
1038}
1039
1040boolean_t
1041spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
1042{
1043	boolean_t found;
1044
1045	mutex_enter(&spa_spare_lock);
1046	found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
1047	mutex_exit(&spa_spare_lock);
1048
1049	return (found);
1050}
1051
1052void
1053spa_spare_activate(vdev_t *vd)
1054{
1055	mutex_enter(&spa_spare_lock);
1056	ASSERT(vd->vdev_isspare);
1057	spa_aux_activate(vd, &spa_spare_avl);
1058	mutex_exit(&spa_spare_lock);
1059}
1060
1061/*
1062 * Level 2 ARC devices are tracked globally for the same reasons as spares.
1063 * Cache devices currently only support one pool per cache device, and so
1064 * for these devices the aux reference count is currently unused beyond 1.
1065 */
1066
1067static int
1068spa_l2cache_compare(const void *a, const void *b)
1069{
1070	return (spa_aux_compare(a, b));
1071}
1072
1073void
1074spa_l2cache_add(vdev_t *vd)
1075{
1076	mutex_enter(&spa_l2cache_lock);
1077	ASSERT(!vd->vdev_isl2cache);
1078	spa_aux_add(vd, &spa_l2cache_avl);
1079	vd->vdev_isl2cache = B_TRUE;
1080	mutex_exit(&spa_l2cache_lock);
1081}
1082
1083void
1084spa_l2cache_remove(vdev_t *vd)
1085{
1086	mutex_enter(&spa_l2cache_lock);
1087	ASSERT(vd->vdev_isl2cache);
1088	spa_aux_remove(vd, &spa_l2cache_avl);
1089	vd->vdev_isl2cache = B_FALSE;
1090	mutex_exit(&spa_l2cache_lock);
1091}
1092
1093boolean_t
1094spa_l2cache_exists(uint64_t guid, uint64_t *pool)
1095{
1096	boolean_t found;
1097
1098	mutex_enter(&spa_l2cache_lock);
1099	found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
1100	mutex_exit(&spa_l2cache_lock);
1101
1102	return (found);
1103}
1104
1105void
1106spa_l2cache_activate(vdev_t *vd)
1107{
1108	mutex_enter(&spa_l2cache_lock);
1109	ASSERT(vd->vdev_isl2cache);
1110	spa_aux_activate(vd, &spa_l2cache_avl);
1111	mutex_exit(&spa_l2cache_lock);
1112}
1113
1114/*
1115 * ==========================================================================
1116 * SPA vdev locking
1117 * ==========================================================================
1118 */
1119
1120/*
1121 * Lock the given spa_t for the purpose of adding or removing a vdev.
1122 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1123 * It returns the next transaction group for the spa_t.
1124 */
1125uint64_t
1126spa_vdev_enter(spa_t *spa)
1127{
1128	mutex_enter(&spa->spa_vdev_top_lock);
1129	mutex_enter(&spa_namespace_lock);
1130	return (spa_vdev_config_enter(spa));
1131}
1132
1133/*
1134 * Internal implementation for spa_vdev_enter().  Used when a vdev
1135 * operation requires multiple syncs (i.e. removing a device) while
1136 * keeping the spa_namespace_lock held.
1137 */
1138uint64_t
1139spa_vdev_config_enter(spa_t *spa)
1140{
1141	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1142
1143	spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1144
1145	return (spa_last_synced_txg(spa) + 1);
1146}
1147
1148/*
1149 * Used in combination with spa_vdev_config_enter() to allow the syncing
1150 * of multiple transactions without releasing the spa_namespace_lock.
1151 */
1152void
1153spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
1154{
1155	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1156
1157	int config_changed = B_FALSE;
1158
1159	ASSERT(txg > spa_last_synced_txg(spa));
1160
1161	spa->spa_pending_vdev = NULL;
1162
1163	/*
1164	 * Reassess the DTLs.
1165	 */
1166	vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
1167
1168	if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
1169		config_changed = B_TRUE;
1170		spa->spa_config_generation++;
1171	}
1172
1173	/*
1174	 * Verify the metaslab classes.
1175	 */
1176	ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
1177	ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
1178
1179	spa_config_exit(spa, SCL_ALL, spa);
1180
1181	/*
1182	 * Panic the system if the specified tag requires it.  This
1183	 * is useful for ensuring that configurations are updated
1184	 * transactionally.
1185	 */
1186	if (zio_injection_enabled)
1187		zio_handle_panic_injection(spa, tag, 0);
1188
1189	/*
1190	 * Note: this txg_wait_synced() is important because it ensures
1191	 * that there won't be more than one config change per txg.
1192	 * This allows us to use the txg as the generation number.
1193	 */
1194	if (error == 0)
1195		txg_wait_synced(spa->spa_dsl_pool, txg);
1196
1197	if (vd != NULL) {
1198		ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
1199		spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1200		vdev_free(vd);
1201		spa_config_exit(spa, SCL_ALL, spa);
1202	}
1203
1204	/*
1205	 * If the config changed, update the config cache.
1206	 */
1207	if (config_changed)
1208		spa_write_cachefile(spa, B_FALSE, B_TRUE);
1209}
1210
1211/*
1212 * Unlock the spa_t after adding or removing a vdev.  Besides undoing the
1213 * locking of spa_vdev_enter(), we also want make sure the transactions have
1214 * synced to disk, and then update the global configuration cache with the new
1215 * information.
1216 */
1217int
1218spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1219{
1220	spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1221	mutex_exit(&spa_namespace_lock);
1222	mutex_exit(&spa->spa_vdev_top_lock);
1223
1224	return (error);
1225}
1226
1227/*
1228 * Lock the given spa_t for the purpose of changing vdev state.
1229 */
1230void
1231spa_vdev_state_enter(spa_t *spa, int oplocks)
1232{
1233	int locks = SCL_STATE_ALL | oplocks;
1234
1235	/*
1236	 * Root pools may need to read of the underlying devfs filesystem
1237	 * when opening up a vdev.  Unfortunately if we're holding the
1238	 * SCL_ZIO lock it will result in a deadlock when we try to issue
1239	 * the read from the root filesystem.  Instead we "prefetch"
1240	 * the associated vnodes that we need prior to opening the
1241	 * underlying devices and cache them so that we can prevent
1242	 * any I/O when we are doing the actual open.
1243	 */
1244	if (spa_is_root(spa)) {
1245		int low = locks & ~(SCL_ZIO - 1);
1246		int high = locks & ~low;
1247
1248		spa_config_enter(spa, high, spa, RW_WRITER);
1249		vdev_hold(spa->spa_root_vdev);
1250		spa_config_enter(spa, low, spa, RW_WRITER);
1251	} else {
1252		spa_config_enter(spa, locks, spa, RW_WRITER);
1253	}
1254	spa->spa_vdev_locks = locks;
1255}
1256
1257int
1258spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1259{
1260	boolean_t config_changed = B_FALSE;
1261
1262	if (vd != NULL || error == 0)
1263		vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1264		    0, 0, B_FALSE);
1265
1266	if (vd != NULL) {
1267		vdev_state_dirty(vd->vdev_top);
1268		config_changed = B_TRUE;
1269		spa->spa_config_generation++;
1270	}
1271
1272	if (spa_is_root(spa))
1273		vdev_rele(spa->spa_root_vdev);
1274
1275	ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1276	spa_config_exit(spa, spa->spa_vdev_locks, spa);
1277
1278	/*
1279	 * If anything changed, wait for it to sync.  This ensures that,
1280	 * from the system administrator's perspective, zpool(1M) commands
1281	 * are synchronous.  This is important for things like zpool offline:
1282	 * when the command completes, you expect no further I/O from ZFS.
1283	 */
1284	if (vd != NULL)
1285		txg_wait_synced(spa->spa_dsl_pool, 0);
1286
1287	/*
1288	 * If the config changed, update the config cache.
1289	 */
1290	if (config_changed) {
1291		mutex_enter(&spa_namespace_lock);
1292		spa_write_cachefile(spa, B_FALSE, B_TRUE);
1293		mutex_exit(&spa_namespace_lock);
1294	}
1295
1296	return (error);
1297}
1298
1299/*
1300 * ==========================================================================
1301 * Miscellaneous functions
1302 * ==========================================================================
1303 */
1304
1305void
1306spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx)
1307{
1308	if (!nvlist_exists(spa->spa_label_features, feature)) {
1309		fnvlist_add_boolean(spa->spa_label_features, feature);
1310		/*
1311		 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1312		 * dirty the vdev config because lock SCL_CONFIG is not held.
1313		 * Thankfully, in this case we don't need to dirty the config
1314		 * because it will be written out anyway when we finish
1315		 * creating the pool.
1316		 */
1317		if (tx->tx_txg != TXG_INITIAL)
1318			vdev_config_dirty(spa->spa_root_vdev);
1319	}
1320}
1321
1322void
1323spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1324{
1325	if (nvlist_remove_all(spa->spa_label_features, feature) == 0)
1326		vdev_config_dirty(spa->spa_root_vdev);
1327}
1328
1329/*
1330 * Rename a spa_t.
1331 */
1332int
1333spa_rename(const char *name, const char *newname)
1334{
1335	spa_t *spa;
1336	int err;
1337
1338	/*
1339	 * Lookup the spa_t and grab the config lock for writing.  We need to
1340	 * actually open the pool so that we can sync out the necessary labels.
1341	 * It's OK to call spa_open() with the namespace lock held because we
1342	 * allow recursive calls for other reasons.
1343	 */
1344	mutex_enter(&spa_namespace_lock);
1345	if ((err = spa_open(name, &spa, FTAG)) != 0) {
1346		mutex_exit(&spa_namespace_lock);
1347		return (err);
1348	}
1349
1350	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1351
1352	avl_remove(&spa_namespace_avl, spa);
1353	(void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1354	avl_add(&spa_namespace_avl, spa);
1355
1356	/*
1357	 * Sync all labels to disk with the new names by marking the root vdev
1358	 * dirty and waiting for it to sync.  It will pick up the new pool name
1359	 * during the sync.
1360	 */
1361	vdev_config_dirty(spa->spa_root_vdev);
1362
1363	spa_config_exit(spa, SCL_ALL, FTAG);
1364
1365	txg_wait_synced(spa->spa_dsl_pool, 0);
1366
1367	/*
1368	 * Sync the updated config cache.
1369	 */
1370	spa_write_cachefile(spa, B_FALSE, B_TRUE);
1371
1372	spa_close(spa, FTAG);
1373
1374	mutex_exit(&spa_namespace_lock);
1375
1376	return (0);
1377}
1378
1379/*
1380 * Return the spa_t associated with given pool_guid, if it exists.  If
1381 * device_guid is non-zero, determine whether the pool exists *and* contains
1382 * a device with the specified device_guid.
1383 */
1384spa_t *
1385spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1386{
1387	spa_t *spa;
1388	avl_tree_t *t = &spa_namespace_avl;
1389
1390	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1391
1392	for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1393		if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1394			continue;
1395		if (spa->spa_root_vdev == NULL)
1396			continue;
1397		if (spa_guid(spa) == pool_guid) {
1398			if (device_guid == 0)
1399				break;
1400
1401			if (vdev_lookup_by_guid(spa->spa_root_vdev,
1402			    device_guid) != NULL)
1403				break;
1404
1405			/*
1406			 * Check any devices we may be in the process of adding.
1407			 */
1408			if (spa->spa_pending_vdev) {
1409				if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1410				    device_guid) != NULL)
1411					break;
1412			}
1413		}
1414	}
1415
1416	return (spa);
1417}
1418
1419/*
1420 * Determine whether a pool with the given pool_guid exists.
1421 */
1422boolean_t
1423spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1424{
1425	return (spa_by_guid(pool_guid, device_guid) != NULL);
1426}
1427
1428char *
1429spa_strdup(const char *s)
1430{
1431	size_t len;
1432	char *new;
1433
1434	len = strlen(s);
1435	new = kmem_alloc(len + 1, KM_SLEEP);
1436	bcopy(s, new, len);
1437	new[len] = '\0';
1438
1439	return (new);
1440}
1441
1442void
1443spa_strfree(char *s)
1444{
1445	kmem_free(s, strlen(s) + 1);
1446}
1447
1448uint64_t
1449spa_get_random(uint64_t range)
1450{
1451	uint64_t r;
1452
1453	ASSERT(range != 0);
1454
1455	(void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1456
1457	return (r % range);
1458}
1459
1460uint64_t
1461spa_generate_guid(spa_t *spa)
1462{
1463	uint64_t guid = spa_get_random(-1ULL);
1464
1465	if (spa != NULL) {
1466		while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1467			guid = spa_get_random(-1ULL);
1468	} else {
1469		while (guid == 0 || spa_guid_exists(guid, 0))
1470			guid = spa_get_random(-1ULL);
1471	}
1472
1473	return (guid);
1474}
1475
1476void
1477snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp)
1478{
1479	char type[256];
1480	char *checksum = NULL;
1481	char *compress = NULL;
1482
1483	if (bp != NULL) {
1484		if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1485			dmu_object_byteswap_t bswap =
1486			    DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1487			(void) snprintf(type, sizeof (type), "bswap %s %s",
1488			    DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1489			    "metadata" : "data",
1490			    dmu_ot_byteswap[bswap].ob_name);
1491		} else {
1492			(void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1493			    sizeof (type));
1494		}
1495		if (!BP_IS_EMBEDDED(bp)) {
1496			checksum =
1497			    zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1498		}
1499		compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1500	}
1501
1502	SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum,
1503	    compress);
1504}
1505
1506void
1507spa_freeze(spa_t *spa)
1508{
1509	uint64_t freeze_txg = 0;
1510
1511	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1512	if (spa->spa_freeze_txg == UINT64_MAX) {
1513		freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1514		spa->spa_freeze_txg = freeze_txg;
1515	}
1516	spa_config_exit(spa, SCL_ALL, FTAG);
1517	if (freeze_txg != 0)
1518		txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1519}
1520
1521void
1522zfs_panic_recover(const char *fmt, ...)
1523{
1524	va_list adx;
1525
1526	va_start(adx, fmt);
1527	vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1528	va_end(adx);
1529}
1530
1531/*
1532 * This is a stripped-down version of strtoull, suitable only for converting
1533 * lowercase hexadecimal numbers that don't overflow.
1534 */
1535uint64_t
1536zfs_strtonum(const char *str, char **nptr)
1537{
1538	uint64_t val = 0;
1539	char c;
1540	int digit;
1541
1542	while ((c = *str) != '\0') {
1543		if (c >= '0' && c <= '9')
1544			digit = c - '0';
1545		else if (c >= 'a' && c <= 'f')
1546			digit = 10 + c - 'a';
1547		else
1548			break;
1549
1550		val *= 16;
1551		val += digit;
1552
1553		str++;
1554	}
1555
1556	if (nptr)
1557		*nptr = (char *)str;
1558
1559	return (val);
1560}
1561
1562/*
1563 * ==========================================================================
1564 * Accessor functions
1565 * ==========================================================================
1566 */
1567
1568boolean_t
1569spa_shutting_down(spa_t *spa)
1570{
1571	return (spa->spa_async_suspended);
1572}
1573
1574dsl_pool_t *
1575spa_get_dsl(spa_t *spa)
1576{
1577	return (spa->spa_dsl_pool);
1578}
1579
1580boolean_t
1581spa_is_initializing(spa_t *spa)
1582{
1583	return (spa->spa_is_initializing);
1584}
1585
1586boolean_t
1587spa_indirect_vdevs_loaded(spa_t *spa)
1588{
1589	return (spa->spa_indirect_vdevs_loaded);
1590}
1591
1592blkptr_t *
1593spa_get_rootblkptr(spa_t *spa)
1594{
1595	return (&spa->spa_ubsync.ub_rootbp);
1596}
1597
1598void
1599spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1600{
1601	spa->spa_uberblock.ub_rootbp = *bp;
1602}
1603
1604void
1605spa_altroot(spa_t *spa, char *buf, size_t buflen)
1606{
1607	if (spa->spa_root == NULL)
1608		buf[0] = '\0';
1609	else
1610		(void) strncpy(buf, spa->spa_root, buflen);
1611}
1612
1613int
1614spa_sync_pass(spa_t *spa)
1615{
1616	return (spa->spa_sync_pass);
1617}
1618
1619char *
1620spa_name(spa_t *spa)
1621{
1622	return (spa->spa_name);
1623}
1624
1625uint64_t
1626spa_guid(spa_t *spa)
1627{
1628	dsl_pool_t *dp = spa_get_dsl(spa);
1629	uint64_t guid;
1630
1631	/*
1632	 * If we fail to parse the config during spa_load(), we can go through
1633	 * the error path (which posts an ereport) and end up here with no root
1634	 * vdev.  We stash the original pool guid in 'spa_config_guid' to handle
1635	 * this case.
1636	 */
1637	if (spa->spa_root_vdev == NULL)
1638		return (spa->spa_config_guid);
1639
1640	guid = spa->spa_last_synced_guid != 0 ?
1641	    spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1642
1643	/*
1644	 * Return the most recently synced out guid unless we're
1645	 * in syncing context.
1646	 */
1647	if (dp && dsl_pool_sync_context(dp))
1648		return (spa->spa_root_vdev->vdev_guid);
1649	else
1650		return (guid);
1651}
1652
1653uint64_t
1654spa_load_guid(spa_t *spa)
1655{
1656	/*
1657	 * This is a GUID that exists solely as a reference for the
1658	 * purposes of the arc.  It is generated at load time, and
1659	 * is never written to persistent storage.
1660	 */
1661	return (spa->spa_load_guid);
1662}
1663
1664uint64_t
1665spa_last_synced_txg(spa_t *spa)
1666{
1667	return (spa->spa_ubsync.ub_txg);
1668}
1669
1670uint64_t
1671spa_first_txg(spa_t *spa)
1672{
1673	return (spa->spa_first_txg);
1674}
1675
1676uint64_t
1677spa_syncing_txg(spa_t *spa)
1678{
1679	return (spa->spa_syncing_txg);
1680}
1681
1682/*
1683 * Return the last txg where data can be dirtied. The final txgs
1684 * will be used to just clear out any deferred frees that remain.
1685 */
1686uint64_t
1687spa_final_dirty_txg(spa_t *spa)
1688{
1689	return (spa->spa_final_txg - TXG_DEFER_SIZE);
1690}
1691
1692pool_state_t
1693spa_state(spa_t *spa)
1694{
1695	return (spa->spa_state);
1696}
1697
1698spa_load_state_t
1699spa_load_state(spa_t *spa)
1700{
1701	return (spa->spa_load_state);
1702}
1703
1704uint64_t
1705spa_freeze_txg(spa_t *spa)
1706{
1707	return (spa->spa_freeze_txg);
1708}
1709
1710/* ARGSUSED */
1711uint64_t
1712spa_get_worst_case_asize(spa_t *spa, uint64_t lsize)
1713{
1714	return (lsize * spa_asize_inflation);
1715}
1716
1717/*
1718 * Return the amount of slop space in bytes.  It is 1/32 of the pool (3.2%),
1719 * or at least 128MB, unless that would cause it to be more than half the
1720 * pool size.
1721 *
1722 * See the comment above spa_slop_shift for details.
1723 */
1724uint64_t
1725spa_get_slop_space(spa_t *spa)
1726{
1727	uint64_t space = spa_get_dspace(spa);
1728	return (MAX(space >> spa_slop_shift, MIN(space >> 1, spa_min_slop)));
1729}
1730
1731uint64_t
1732spa_get_dspace(spa_t *spa)
1733{
1734	return (spa->spa_dspace);
1735}
1736
1737uint64_t
1738spa_get_checkpoint_space(spa_t *spa)
1739{
1740	return (spa->spa_checkpoint_info.sci_dspace);
1741}
1742
1743void
1744spa_update_dspace(spa_t *spa)
1745{
1746	spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1747	    ddt_get_dedup_dspace(spa);
1748	if (spa->spa_vdev_removal != NULL) {
1749		/*
1750		 * We can't allocate from the removing device, so
1751		 * subtract its size.  This prevents the DMU/DSL from
1752		 * filling up the (now smaller) pool while we are in the
1753		 * middle of removing the device.
1754		 *
1755		 * Note that the DMU/DSL doesn't actually know or care
1756		 * how much space is allocated (it does its own tracking
1757		 * of how much space has been logically used).  So it
1758		 * doesn't matter that the data we are moving may be
1759		 * allocated twice (on the old device and the new
1760		 * device).
1761		 */
1762		spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1763		vdev_t *vd =
1764		    vdev_lookup_top(spa, spa->spa_vdev_removal->svr_vdev_id);
1765		spa->spa_dspace -= spa_deflate(spa) ?
1766		    vd->vdev_stat.vs_dspace : vd->vdev_stat.vs_space;
1767		spa_config_exit(spa, SCL_VDEV, FTAG);
1768	}
1769}
1770
1771/*
1772 * Return the failure mode that has been set to this pool. The default
1773 * behavior will be to block all I/Os when a complete failure occurs.
1774 */
1775uint8_t
1776spa_get_failmode(spa_t *spa)
1777{
1778	return (spa->spa_failmode);
1779}
1780
1781boolean_t
1782spa_suspended(spa_t *spa)
1783{
1784	return (spa->spa_suspended);
1785}
1786
1787uint64_t
1788spa_version(spa_t *spa)
1789{
1790	return (spa->spa_ubsync.ub_version);
1791}
1792
1793boolean_t
1794spa_deflate(spa_t *spa)
1795{
1796	return (spa->spa_deflate);
1797}
1798
1799metaslab_class_t *
1800spa_normal_class(spa_t *spa)
1801{
1802	return (spa->spa_normal_class);
1803}
1804
1805metaslab_class_t *
1806spa_log_class(spa_t *spa)
1807{
1808	return (spa->spa_log_class);
1809}
1810
1811void
1812spa_evicting_os_register(spa_t *spa, objset_t *os)
1813{
1814	mutex_enter(&spa->spa_evicting_os_lock);
1815	list_insert_head(&spa->spa_evicting_os_list, os);
1816	mutex_exit(&spa->spa_evicting_os_lock);
1817}
1818
1819void
1820spa_evicting_os_deregister(spa_t *spa, objset_t *os)
1821{
1822	mutex_enter(&spa->spa_evicting_os_lock);
1823	list_remove(&spa->spa_evicting_os_list, os);
1824	cv_broadcast(&spa->spa_evicting_os_cv);
1825	mutex_exit(&spa->spa_evicting_os_lock);
1826}
1827
1828void
1829spa_evicting_os_wait(spa_t *spa)
1830{
1831	mutex_enter(&spa->spa_evicting_os_lock);
1832	while (!list_is_empty(&spa->spa_evicting_os_list))
1833		cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock);
1834	mutex_exit(&spa->spa_evicting_os_lock);
1835
1836	dmu_buf_user_evict_wait();
1837}
1838
1839int
1840spa_max_replication(spa_t *spa)
1841{
1842	/*
1843	 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1844	 * handle BPs with more than one DVA allocated.  Set our max
1845	 * replication level accordingly.
1846	 */
1847	if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1848		return (1);
1849	return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1850}
1851
1852int
1853spa_prev_software_version(spa_t *spa)
1854{
1855	return (spa->spa_prev_software_version);
1856}
1857
1858uint64_t
1859spa_deadman_synctime(spa_t *spa)
1860{
1861	return (spa->spa_deadman_synctime);
1862}
1863
1864uint64_t
1865dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1866{
1867	uint64_t asize = DVA_GET_ASIZE(dva);
1868	uint64_t dsize = asize;
1869
1870	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1871
1872	if (asize != 0 && spa->spa_deflate) {
1873		vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1874		dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1875	}
1876
1877	return (dsize);
1878}
1879
1880uint64_t
1881bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1882{
1883	uint64_t dsize = 0;
1884
1885	for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1886		dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1887
1888	return (dsize);
1889}
1890
1891uint64_t
1892bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1893{
1894	uint64_t dsize = 0;
1895
1896	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1897
1898	for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1899		dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1900
1901	spa_config_exit(spa, SCL_VDEV, FTAG);
1902
1903	return (dsize);
1904}
1905
1906/*
1907 * ==========================================================================
1908 * Initialization and Termination
1909 * ==========================================================================
1910 */
1911
1912static int
1913spa_name_compare(const void *a1, const void *a2)
1914{
1915	const spa_t *s1 = a1;
1916	const spa_t *s2 = a2;
1917	int s;
1918
1919	s = strcmp(s1->spa_name, s2->spa_name);
1920	if (s > 0)
1921		return (1);
1922	if (s < 0)
1923		return (-1);
1924	return (0);
1925}
1926
1927int
1928spa_busy(void)
1929{
1930	return (spa_active_count);
1931}
1932
1933void
1934spa_boot_init()
1935{
1936	spa_config_load();
1937}
1938
1939void
1940spa_init(int mode)
1941{
1942	mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1943	mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1944	mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1945	cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1946
1947	avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1948	    offsetof(spa_t, spa_avl));
1949
1950	avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1951	    offsetof(spa_aux_t, aux_avl));
1952
1953	avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1954	    offsetof(spa_aux_t, aux_avl));
1955
1956	spa_mode_global = mode;
1957
1958#ifdef _KERNEL
1959	spa_arch_init();
1960#else
1961	if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
1962		arc_procfd = open("/proc/self/ctl", O_WRONLY);
1963		if (arc_procfd == -1) {
1964			perror("could not enable watchpoints: "
1965			    "opening /proc/self/ctl failed: ");
1966		} else {
1967			arc_watch = B_TRUE;
1968		}
1969	}
1970#endif
1971
1972	refcount_init();
1973	unique_init();
1974	range_tree_init();
1975	metaslab_alloc_trace_init();
1976	zio_init();
1977	dmu_init();
1978	zil_init();
1979	vdev_cache_stat_init();
1980	zfs_prop_init();
1981	zpool_prop_init();
1982	zpool_feature_init();
1983	spa_config_load();
1984	l2arc_start();
1985}
1986
1987void
1988spa_fini(void)
1989{
1990	l2arc_stop();
1991
1992	spa_evict_all();
1993
1994	vdev_cache_stat_fini();
1995	zil_fini();
1996	dmu_fini();
1997	zio_fini();
1998	metaslab_alloc_trace_fini();
1999	range_tree_fini();
2000	unique_fini();
2001	refcount_fini();
2002
2003	avl_destroy(&spa_namespace_avl);
2004	avl_destroy(&spa_spare_avl);
2005	avl_destroy(&spa_l2cache_avl);
2006
2007	cv_destroy(&spa_namespace_cv);
2008	mutex_destroy(&spa_namespace_lock);
2009	mutex_destroy(&spa_spare_lock);
2010	mutex_destroy(&spa_l2cache_lock);
2011}
2012
2013/*
2014 * Return whether this pool has slogs. No locking needed.
2015 * It's not a problem if the wrong answer is returned as it's only for
2016 * performance and not correctness
2017 */
2018boolean_t
2019spa_has_slogs(spa_t *spa)
2020{
2021	return (spa->spa_log_class->mc_rotor != NULL);
2022}
2023
2024spa_log_state_t
2025spa_get_log_state(spa_t *spa)
2026{
2027	return (spa->spa_log_state);
2028}
2029
2030void
2031spa_set_log_state(spa_t *spa, spa_log_state_t state)
2032{
2033	spa->spa_log_state = state;
2034}
2035
2036boolean_t
2037spa_is_root(spa_t *spa)
2038{
2039	return (spa->spa_is_root);
2040}
2041
2042boolean_t
2043spa_writeable(spa_t *spa)
2044{
2045	return (!!(spa->spa_mode & FWRITE) && spa->spa_trust_config);
2046}
2047
2048/*
2049 * Returns true if there is a pending sync task in any of the current
2050 * syncing txg, the current quiescing txg, or the current open txg.
2051 */
2052boolean_t
2053spa_has_pending_synctask(spa_t *spa)
2054{
2055	return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks) ||
2056	    !txg_all_lists_empty(&spa->spa_dsl_pool->dp_early_sync_tasks));
2057}
2058
2059int
2060spa_mode(spa_t *spa)
2061{
2062	return (spa->spa_mode);
2063}
2064
2065uint64_t
2066spa_bootfs(spa_t *spa)
2067{
2068	return (spa->spa_bootfs);
2069}
2070
2071uint64_t
2072spa_delegation(spa_t *spa)
2073{
2074	return (spa->spa_delegation);
2075}
2076
2077objset_t *
2078spa_meta_objset(spa_t *spa)
2079{
2080	return (spa->spa_meta_objset);
2081}
2082
2083enum zio_checksum
2084spa_dedup_checksum(spa_t *spa)
2085{
2086	return (spa->spa_dedup_checksum);
2087}
2088
2089/*
2090 * Reset pool scan stat per scan pass (or reboot).
2091 */
2092void
2093spa_scan_stat_init(spa_t *spa)
2094{
2095	/* data not stored on disk */
2096	spa->spa_scan_pass_start = gethrestime_sec();
2097	if (dsl_scan_is_paused_scrub(spa->spa_dsl_pool->dp_scan))
2098		spa->spa_scan_pass_scrub_pause = spa->spa_scan_pass_start;
2099	else
2100		spa->spa_scan_pass_scrub_pause = 0;
2101	spa->spa_scan_pass_scrub_spent_paused = 0;
2102	spa->spa_scan_pass_exam = 0;
2103	vdev_scan_stat_init(spa->spa_root_vdev);
2104}
2105
2106/*
2107 * Get scan stats for zpool status reports
2108 */
2109int
2110spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
2111{
2112	dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
2113
2114	if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
2115		return (SET_ERROR(ENOENT));
2116	bzero(ps, sizeof (pool_scan_stat_t));
2117
2118	/* data stored on disk */
2119	ps->pss_func = scn->scn_phys.scn_func;
2120	ps->pss_start_time = scn->scn_phys.scn_start_time;
2121	ps->pss_end_time = scn->scn_phys.scn_end_time;
2122	ps->pss_to_examine = scn->scn_phys.scn_to_examine;
2123	ps->pss_examined = scn->scn_phys.scn_examined;
2124	ps->pss_to_process = scn->scn_phys.scn_to_process;
2125	ps->pss_processed = scn->scn_phys.scn_processed;
2126	ps->pss_errors = scn->scn_phys.scn_errors;
2127	ps->pss_state = scn->scn_phys.scn_state;
2128
2129	/* data not stored on disk */
2130	ps->pss_pass_start = spa->spa_scan_pass_start;
2131	ps->pss_pass_exam = spa->spa_scan_pass_exam;
2132	ps->pss_pass_scrub_pause = spa->spa_scan_pass_scrub_pause;
2133	ps->pss_pass_scrub_spent_paused = spa->spa_scan_pass_scrub_spent_paused;
2134
2135	return (0);
2136}
2137
2138int
2139spa_maxblocksize(spa_t *spa)
2140{
2141	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS))
2142		return (SPA_MAXBLOCKSIZE);
2143	else
2144		return (SPA_OLD_MAXBLOCKSIZE);
2145}
2146
2147/*
2148 * Returns the txg that the last device removal completed. No indirect mappings
2149 * have been added since this txg.
2150 */
2151uint64_t
2152spa_get_last_removal_txg(spa_t *spa)
2153{
2154	uint64_t vdevid;
2155	uint64_t ret = -1ULL;
2156
2157	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2158	/*
2159	 * sr_prev_indirect_vdev is only modified while holding all the
2160	 * config locks, so it is sufficient to hold SCL_VDEV as reader when
2161	 * examining it.
2162	 */
2163	vdevid = spa->spa_removing_phys.sr_prev_indirect_vdev;
2164
2165	while (vdevid != -1ULL) {
2166		vdev_t *vd = vdev_lookup_top(spa, vdevid);
2167		vdev_indirect_births_t *vib = vd->vdev_indirect_births;
2168
2169		ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops);
2170
2171		/*
2172		 * If the removal did not remap any data, we don't care.
2173		 */
2174		if (vdev_indirect_births_count(vib) != 0) {
2175			ret = vdev_indirect_births_last_entry_txg(vib);
2176			break;
2177		}
2178
2179		vdevid = vd->vdev_indirect_config.vic_prev_indirect_vdev;
2180	}
2181	spa_config_exit(spa, SCL_VDEV, FTAG);
2182
2183	IMPLY(ret != -1ULL,
2184	    spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL));
2185
2186	return (ret);
2187}
2188
2189boolean_t
2190spa_trust_config(spa_t *spa)
2191{
2192	return (spa->spa_trust_config);
2193}
2194
2195uint64_t
2196spa_missing_tvds_allowed(spa_t *spa)
2197{
2198	return (spa->spa_missing_tvds_allowed);
2199}
2200
2201void
2202spa_set_missing_tvds(spa_t *spa, uint64_t missing)
2203{
2204	spa->spa_missing_tvds = missing;
2205}
2206
2207boolean_t
2208spa_top_vdevs_spacemap_addressable(spa_t *spa)
2209{
2210	vdev_t *rvd = spa->spa_root_vdev;
2211	for (uint64_t c = 0; c < rvd->vdev_children; c++) {
2212		if (!vdev_is_spacemap_addressable(rvd->vdev_child[c]))
2213			return (B_FALSE);
2214	}
2215	return (B_TRUE);
2216}
2217
2218boolean_t
2219spa_has_checkpoint(spa_t *spa)
2220{
2221	return (spa->spa_checkpoint_txg != 0);
2222}
2223
2224boolean_t
2225spa_importing_readonly_checkpoint(spa_t *spa)
2226{
2227	return ((spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT) &&
2228	    spa->spa_mode == FREAD);
2229}
2230
2231uint64_t
2232spa_min_claim_txg(spa_t *spa)
2233{
2234	uint64_t checkpoint_txg = spa->spa_uberblock.ub_checkpoint_txg;
2235
2236	if (checkpoint_txg != 0)
2237		return (checkpoint_txg + 1);
2238
2239	return (spa->spa_first_txg);
2240}
2241
2242/*
2243 * If there is a checkpoint, async destroys may consume more space from
2244 * the pool instead of freeing it. In an attempt to save the pool from
2245 * getting suspended when it is about to run out of space, we stop
2246 * processing async destroys.
2247 */
2248boolean_t
2249spa_suspend_async_destroy(spa_t *spa)
2250{
2251	dsl_pool_t *dp = spa_get_dsl(spa);
2252
2253	uint64_t unreserved = dsl_pool_unreserved_space(dp,
2254	    ZFS_SPACE_CHECK_EXTRA_RESERVED);
2255	uint64_t used = dsl_dir_phys(dp->dp_root_dir)->dd_used_bytes;
2256	uint64_t avail = (unreserved > used) ? (unreserved - used) : 0;
2257
2258	if (spa_has_checkpoint(spa) && avail == 0)
2259		return (B_TRUE);
2260
2261	return (B_FALSE);
2262}
2263