spa_misc.c revision 45818ee124adeaaf947698996b4f4c722afc6d1f
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, 2015 by Delphix. All rights reserved.
24 * Copyright 2011 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 */
28
29#include <sys/zfs_context.h>
30#include <sys/spa_impl.h>
31#include <sys/spa_boot.h>
32#include <sys/zio.h>
33#include <sys/zio_checksum.h>
34#include <sys/zio_compress.h>
35#include <sys/dmu.h>
36#include <sys/dmu_tx.h>
37#include <sys/zap.h>
38#include <sys/zil.h>
39#include <sys/vdev_impl.h>
40#include <sys/metaslab.h>
41#include <sys/uberblock_impl.h>
42#include <sys/txg.h>
43#include <sys/avl.h>
44#include <sys/unique.h>
45#include <sys/dsl_pool.h>
46#include <sys/dsl_dir.h>
47#include <sys/dsl_prop.h>
48#include <sys/dsl_scan.h>
49#include <sys/fs/zfs.h>
50#include <sys/metaslab_impl.h>
51#include <sys/arc.h>
52#include <sys/ddt.h>
53#include "zfs_prop.h"
54#include <sys/zfeature.h>
55
56/*
57 * SPA locking
58 *
59 * There are four basic locks for managing spa_t structures:
60 *
61 * spa_namespace_lock (global mutex)
62 *
63 *	This lock must be acquired to do any of the following:
64 *
65 *		- Lookup a spa_t by name
66 *		- Add or remove a spa_t from the namespace
67 *		- Increase spa_refcount from non-zero
68 *		- Check if spa_refcount is zero
69 *		- Rename a spa_t
70 *		- add/remove/attach/detach devices
71 *		- Held for the duration of create/destroy/import/export
72 *
73 *	It does not need to handle recursion.  A create or destroy may
74 *	reference objects (files or zvols) in other pools, but by
75 *	definition they must have an existing reference, and will never need
76 *	to lookup a spa_t by name.
77 *
78 * spa_refcount (per-spa refcount_t protected by mutex)
79 *
80 *	This reference count keep track of any active users of the spa_t.  The
81 *	spa_t cannot be destroyed or freed while this is non-zero.  Internally,
82 *	the refcount is never really 'zero' - opening a pool implicitly keeps
83 *	some references in the DMU.  Internally we check against spa_minref, but
84 *	present the image of a zero/non-zero value to consumers.
85 *
86 * spa_config_lock[] (per-spa array of rwlocks)
87 *
88 *	This protects the spa_t from config changes, and must be held in
89 *	the following circumstances:
90 *
91 *		- RW_READER to perform I/O to the spa
92 *		- RW_WRITER to change the vdev config
93 *
94 * The locking order is fairly straightforward:
95 *
96 *		spa_namespace_lock	->	spa_refcount
97 *
98 *	The namespace lock must be acquired to increase the refcount from 0
99 *	or to check if it is zero.
100 *
101 *		spa_refcount		->	spa_config_lock[]
102 *
103 *	There must be at least one valid reference on the spa_t to acquire
104 *	the config lock.
105 *
106 *		spa_namespace_lock	->	spa_config_lock[]
107 *
108 *	The namespace lock must always be taken before the config lock.
109 *
110 *
111 * The spa_namespace_lock can be acquired directly and is globally visible.
112 *
113 * The namespace is manipulated using the following functions, all of which
114 * require the spa_namespace_lock to be held.
115 *
116 *	spa_lookup()		Lookup a spa_t by name.
117 *
118 *	spa_add()		Create a new spa_t in the namespace.
119 *
120 *	spa_remove()		Remove a spa_t from the namespace.  This also
121 *				frees up any memory associated with the spa_t.
122 *
123 *	spa_next()		Returns the next spa_t in the system, or the
124 *				first if NULL is passed.
125 *
126 *	spa_evict_all()		Shutdown and remove all spa_t structures in
127 *				the system.
128 *
129 *	spa_guid_exists()	Determine whether a pool/device guid exists.
130 *
131 * The spa_refcount is manipulated using the following functions:
132 *
133 *	spa_open_ref()		Adds a reference to the given spa_t.  Must be
134 *				called with spa_namespace_lock held if the
135 *				refcount is currently zero.
136 *
137 *	spa_close()		Remove a reference from the spa_t.  This will
138 *				not free the spa_t or remove it from the
139 *				namespace.  No locking is required.
140 *
141 *	spa_refcount_zero()	Returns true if the refcount is currently
142 *				zero.  Must be called with spa_namespace_lock
143 *				held.
144 *
145 * The spa_config_lock[] is an array of rwlocks, ordered as follows:
146 * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
147 * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
148 *
149 * To read the configuration, it suffices to hold one of these locks as reader.
150 * To modify the configuration, you must hold all locks as writer.  To modify
151 * vdev state without altering the vdev tree's topology (e.g. online/offline),
152 * you must hold SCL_STATE and SCL_ZIO as writer.
153 *
154 * We use these distinct config locks to avoid recursive lock entry.
155 * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
156 * block allocations (SCL_ALLOC), which may require reading space maps
157 * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
158 *
159 * The spa config locks cannot be normal rwlocks because we need the
160 * ability to hand off ownership.  For example, SCL_ZIO is acquired
161 * by the issuing thread and later released by an interrupt thread.
162 * They do, however, obey the usual write-wanted semantics to prevent
163 * writer (i.e. system administrator) starvation.
164 *
165 * The lock acquisition rules are as follows:
166 *
167 * SCL_CONFIG
168 *	Protects changes to the vdev tree topology, such as vdev
169 *	add/remove/attach/detach.  Protects the dirty config list
170 *	(spa_config_dirty_list) and the set of spares and l2arc devices.
171 *
172 * SCL_STATE
173 *	Protects changes to pool state and vdev state, such as vdev
174 *	online/offline/fault/degrade/clear.  Protects the dirty state list
175 *	(spa_state_dirty_list) and global pool state (spa_state).
176 *
177 * SCL_ALLOC
178 *	Protects changes to metaslab groups and classes.
179 *	Held as reader by metaslab_alloc() and metaslab_claim().
180 *
181 * SCL_ZIO
182 *	Held by bp-level zios (those which have no io_vd upon entry)
183 *	to prevent changes to the vdev tree.  The bp-level zio implicitly
184 *	protects all of its vdev child zios, which do not hold SCL_ZIO.
185 *
186 * SCL_FREE
187 *	Protects changes to metaslab groups and classes.
188 *	Held as reader by metaslab_free().  SCL_FREE is distinct from
189 *	SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
190 *	blocks in zio_done() while another i/o that holds either
191 *	SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
192 *
193 * SCL_VDEV
194 *	Held as reader to prevent changes to the vdev tree during trivial
195 *	inquiries such as bp_get_dsize().  SCL_VDEV is distinct from the
196 *	other locks, and lower than all of them, to ensure that it's safe
197 *	to acquire regardless of caller context.
198 *
199 * In addition, the following rules apply:
200 *
201 * (a)	spa_props_lock protects pool properties, spa_config and spa_config_list.
202 *	The lock ordering is SCL_CONFIG > spa_props_lock.
203 *
204 * (b)	I/O operations on leaf vdevs.  For any zio operation that takes
205 *	an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
206 *	or zio_write_phys() -- the caller must ensure that the config cannot
207 *	cannot change in the interim, and that the vdev cannot be reopened.
208 *	SCL_STATE as reader suffices for both.
209 *
210 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
211 *
212 *	spa_vdev_enter()	Acquire the namespace lock and the config lock
213 *				for writing.
214 *
215 *	spa_vdev_exit()		Release the config lock, wait for all I/O
216 *				to complete, sync the updated configs to the
217 *				cache, and release the namespace lock.
218 *
219 * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
220 * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
221 * locking is, always, based on spa_namespace_lock and spa_config_lock[].
222 *
223 * spa_rename() is also implemented within this file since it requires
224 * manipulation of the namespace.
225 */
226
227static avl_tree_t spa_namespace_avl;
228kmutex_t spa_namespace_lock;
229static kcondvar_t spa_namespace_cv;
230static int spa_active_count;
231int spa_max_replication_override = SPA_DVAS_PER_BP;
232
233static kmutex_t spa_spare_lock;
234static avl_tree_t spa_spare_avl;
235static kmutex_t spa_l2cache_lock;
236static avl_tree_t spa_l2cache_avl;
237
238kmem_cache_t *spa_buffer_pool;
239int spa_mode_global;
240
241#ifdef ZFS_DEBUG
242/* Everything except dprintf and spa is on by default in debug builds */
243int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA);
244#else
245int zfs_flags = 0;
246#endif
247
248/*
249 * zfs_recover can be set to nonzero to attempt to recover from
250 * otherwise-fatal errors, typically caused by on-disk corruption.  When
251 * set, calls to zfs_panic_recover() will turn into warning messages.
252 * This should only be used as a last resort, as it typically results
253 * in leaked space, or worse.
254 */
255boolean_t zfs_recover = B_FALSE;
256
257/*
258 * If destroy encounters an EIO while reading metadata (e.g. indirect
259 * blocks), space referenced by the missing metadata can not be freed.
260 * Normally this causes the background destroy to become "stalled", as
261 * it is unable to make forward progress.  While in this stalled state,
262 * all remaining space to free from the error-encountering filesystem is
263 * "temporarily leaked".  Set this flag to cause it to ignore the EIO,
264 * permanently leak the space from indirect blocks that can not be read,
265 * and continue to free everything else that it can.
266 *
267 * The default, "stalling" behavior is useful if the storage partially
268 * fails (i.e. some but not all i/os fail), and then later recovers.  In
269 * this case, we will be able to continue pool operations while it is
270 * partially failed, and when it recovers, we can continue to free the
271 * space, with no leaks.  However, note that this case is actually
272 * fairly rare.
273 *
274 * Typically pools either (a) fail completely (but perhaps temporarily,
275 * e.g. a top-level vdev going offline), or (b) have localized,
276 * permanent errors (e.g. disk returns the wrong data due to bit flip or
277 * firmware bug).  In case (a), this setting does not matter because the
278 * pool will be suspended and the sync thread will not be able to make
279 * forward progress regardless.  In case (b), because the error is
280 * permanent, the best we can do is leak the minimum amount of space,
281 * which is what setting this flag will do.  Therefore, it is reasonable
282 * for this flag to normally be set, but we chose the more conservative
283 * approach of not setting it, so that there is no possibility of
284 * leaking space in the "partial temporary" failure case.
285 */
286boolean_t zfs_free_leak_on_eio = B_FALSE;
287
288/*
289 * Expiration time in milliseconds. This value has two meanings. First it is
290 * used to determine when the spa_deadman() logic should fire. By default the
291 * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
292 * Secondly, the value determines if an I/O is considered "hung". Any I/O that
293 * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
294 * in a system panic.
295 */
296uint64_t zfs_deadman_synctime_ms = 1000000ULL;
297
298/*
299 * Check time in milliseconds. This defines the frequency at which we check
300 * for hung I/O.
301 */
302uint64_t zfs_deadman_checktime_ms = 5000ULL;
303
304/*
305 * Override the zfs deadman behavior via /etc/system. By default the
306 * deadman is enabled except on VMware and sparc deployments.
307 */
308int zfs_deadman_enabled = -1;
309
310/*
311 * The worst case is single-sector max-parity RAID-Z blocks, in which
312 * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
313 * times the size; so just assume that.  Add to this the fact that
314 * we can have up to 3 DVAs per bp, and one more factor of 2 because
315 * the block may be dittoed with up to 3 DVAs by ddt_sync().  All together,
316 * the worst case is:
317 *     (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
318 */
319int spa_asize_inflation = 24;
320
321/*
322 * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
323 * the pool to be consumed.  This ensures that we don't run the pool
324 * completely out of space, due to unaccounted changes (e.g. to the MOS).
325 * It also limits the worst-case time to allocate space.  If we have
326 * less than this amount of free space, most ZPL operations (e.g. write,
327 * create) will return ENOSPC.
328 *
329 * Certain operations (e.g. file removal, most administrative actions) can
330 * use half the slop space.  They will only return ENOSPC if less than half
331 * the slop space is free.  Typically, once the pool has less than the slop
332 * space free, the user will use these operations to free up space in the pool.
333 * These are the operations that call dsl_pool_adjustedsize() with the netfree
334 * argument set to TRUE.
335 *
336 * A very restricted set of operations are always permitted, regardless of
337 * the amount of free space.  These are the operations that call
338 * dsl_sync_task(ZFS_SPACE_CHECK_NONE), e.g. "zfs destroy".  If these
339 * operations result in a net increase in the amount of space used,
340 * it is possible to run the pool completely out of space, causing it to
341 * be permanently read-only.
342 *
343 * See also the comments in zfs_space_check_t.
344 */
345int spa_slop_shift = 5;
346
347/*
348 * ==========================================================================
349 * SPA config locking
350 * ==========================================================================
351 */
352static void
353spa_config_lock_init(spa_t *spa)
354{
355	for (int i = 0; i < SCL_LOCKS; i++) {
356		spa_config_lock_t *scl = &spa->spa_config_lock[i];
357		mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
358		cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
359		refcount_create_untracked(&scl->scl_count);
360		scl->scl_writer = NULL;
361		scl->scl_write_wanted = 0;
362	}
363}
364
365static void
366spa_config_lock_destroy(spa_t *spa)
367{
368	for (int i = 0; i < SCL_LOCKS; i++) {
369		spa_config_lock_t *scl = &spa->spa_config_lock[i];
370		mutex_destroy(&scl->scl_lock);
371		cv_destroy(&scl->scl_cv);
372		refcount_destroy(&scl->scl_count);
373		ASSERT(scl->scl_writer == NULL);
374		ASSERT(scl->scl_write_wanted == 0);
375	}
376}
377
378int
379spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
380{
381	for (int i = 0; i < SCL_LOCKS; i++) {
382		spa_config_lock_t *scl = &spa->spa_config_lock[i];
383		if (!(locks & (1 << i)))
384			continue;
385		mutex_enter(&scl->scl_lock);
386		if (rw == RW_READER) {
387			if (scl->scl_writer || scl->scl_write_wanted) {
388				mutex_exit(&scl->scl_lock);
389				spa_config_exit(spa, locks ^ (1 << i), tag);
390				return (0);
391			}
392		} else {
393			ASSERT(scl->scl_writer != curthread);
394			if (!refcount_is_zero(&scl->scl_count)) {
395				mutex_exit(&scl->scl_lock);
396				spa_config_exit(spa, locks ^ (1 << i), tag);
397				return (0);
398			}
399			scl->scl_writer = curthread;
400		}
401		(void) refcount_add(&scl->scl_count, tag);
402		mutex_exit(&scl->scl_lock);
403	}
404	return (1);
405}
406
407void
408spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
409{
410	int wlocks_held = 0;
411
412	ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
413
414	for (int i = 0; i < SCL_LOCKS; i++) {
415		spa_config_lock_t *scl = &spa->spa_config_lock[i];
416		if (scl->scl_writer == curthread)
417			wlocks_held |= (1 << i);
418		if (!(locks & (1 << i)))
419			continue;
420		mutex_enter(&scl->scl_lock);
421		if (rw == RW_READER) {
422			while (scl->scl_writer || scl->scl_write_wanted) {
423				cv_wait(&scl->scl_cv, &scl->scl_lock);
424			}
425		} else {
426			ASSERT(scl->scl_writer != curthread);
427			while (!refcount_is_zero(&scl->scl_count)) {
428				scl->scl_write_wanted++;
429				cv_wait(&scl->scl_cv, &scl->scl_lock);
430				scl->scl_write_wanted--;
431			}
432			scl->scl_writer = curthread;
433		}
434		(void) refcount_add(&scl->scl_count, tag);
435		mutex_exit(&scl->scl_lock);
436	}
437	ASSERT(wlocks_held <= locks);
438}
439
440void
441spa_config_exit(spa_t *spa, int locks, void *tag)
442{
443	for (int i = SCL_LOCKS - 1; i >= 0; i--) {
444		spa_config_lock_t *scl = &spa->spa_config_lock[i];
445		if (!(locks & (1 << i)))
446			continue;
447		mutex_enter(&scl->scl_lock);
448		ASSERT(!refcount_is_zero(&scl->scl_count));
449		if (refcount_remove(&scl->scl_count, tag) == 0) {
450			ASSERT(scl->scl_writer == NULL ||
451			    scl->scl_writer == curthread);
452			scl->scl_writer = NULL;	/* OK in either case */
453			cv_broadcast(&scl->scl_cv);
454		}
455		mutex_exit(&scl->scl_lock);
456	}
457}
458
459int
460spa_config_held(spa_t *spa, int locks, krw_t rw)
461{
462	int locks_held = 0;
463
464	for (int i = 0; i < SCL_LOCKS; i++) {
465		spa_config_lock_t *scl = &spa->spa_config_lock[i];
466		if (!(locks & (1 << i)))
467			continue;
468		if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
469		    (rw == RW_WRITER && scl->scl_writer == curthread))
470			locks_held |= 1 << i;
471	}
472
473	return (locks_held);
474}
475
476/*
477 * ==========================================================================
478 * SPA namespace functions
479 * ==========================================================================
480 */
481
482/*
483 * Lookup the named spa_t in the AVL tree.  The spa_namespace_lock must be held.
484 * Returns NULL if no matching spa_t is found.
485 */
486spa_t *
487spa_lookup(const char *name)
488{
489	static spa_t search;	/* spa_t is large; don't allocate on stack */
490	spa_t *spa;
491	avl_index_t where;
492	char *cp;
493
494	ASSERT(MUTEX_HELD(&spa_namespace_lock));
495
496	(void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
497
498	/*
499	 * If it's a full dataset name, figure out the pool name and
500	 * just use that.
501	 */
502	cp = strpbrk(search.spa_name, "/@#");
503	if (cp != NULL)
504		*cp = '\0';
505
506	spa = avl_find(&spa_namespace_avl, &search, &where);
507
508	return (spa);
509}
510
511/*
512 * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
513 * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
514 * looking for potentially hung I/Os.
515 */
516void
517spa_deadman(void *arg)
518{
519	spa_t *spa = arg;
520
521	/*
522	 * Disable the deadman timer if the pool is suspended.
523	 */
524	if (spa_suspended(spa)) {
525		VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
526		return;
527	}
528
529	zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
530	    (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
531	    ++spa->spa_deadman_calls);
532	if (zfs_deadman_enabled)
533		vdev_deadman(spa->spa_root_vdev);
534}
535
536/*
537 * Create an uninitialized spa_t with the given name.  Requires
538 * spa_namespace_lock.  The caller must ensure that the spa_t doesn't already
539 * exist by calling spa_lookup() first.
540 */
541spa_t *
542spa_add(const char *name, nvlist_t *config, const char *altroot)
543{
544	spa_t *spa;
545	spa_config_dirent_t *dp;
546	cyc_handler_t hdlr;
547	cyc_time_t when;
548
549	ASSERT(MUTEX_HELD(&spa_namespace_lock));
550
551	spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
552
553	mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
554	mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
555	mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
556	mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL);
557	mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
558	mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
559	mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
560	mutex_init(&spa->spa_cksum_tmpls_lock, NULL, MUTEX_DEFAULT, NULL);
561	mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
562	mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
563	mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
564	mutex_init(&spa->spa_iokstat_lock, NULL, MUTEX_DEFAULT, NULL);
565
566	cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
567	cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL);
568	cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
569	cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
570	cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
571
572	for (int t = 0; t < TXG_SIZE; t++)
573		bplist_create(&spa->spa_free_bplist[t]);
574
575	(void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
576	spa->spa_state = POOL_STATE_UNINITIALIZED;
577	spa->spa_freeze_txg = UINT64_MAX;
578	spa->spa_final_txg = UINT64_MAX;
579	spa->spa_load_max_txg = UINT64_MAX;
580	spa->spa_proc = &p0;
581	spa->spa_proc_state = SPA_PROC_NONE;
582
583	hdlr.cyh_func = spa_deadman;
584	hdlr.cyh_arg = spa;
585	hdlr.cyh_level = CY_LOW_LEVEL;
586
587	spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
588
589	/*
590	 * This determines how often we need to check for hung I/Os after
591	 * the cyclic has already fired. Since checking for hung I/Os is
592	 * an expensive operation we don't want to check too frequently.
593	 * Instead wait for 5 seconds before checking again.
594	 */
595	when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms);
596	when.cyt_when = CY_INFINITY;
597	mutex_enter(&cpu_lock);
598	spa->spa_deadman_cycid = cyclic_add(&hdlr, &when);
599	mutex_exit(&cpu_lock);
600
601	refcount_create(&spa->spa_refcount);
602	spa_config_lock_init(spa);
603
604	avl_add(&spa_namespace_avl, spa);
605
606	/*
607	 * Set the alternate root, if there is one.
608	 */
609	if (altroot) {
610		spa->spa_root = spa_strdup(altroot);
611		spa_active_count++;
612	}
613
614	/*
615	 * Every pool starts with the default cachefile
616	 */
617	list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
618	    offsetof(spa_config_dirent_t, scd_link));
619
620	dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
621	dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
622	list_insert_head(&spa->spa_config_list, dp);
623
624	VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
625	    KM_SLEEP) == 0);
626
627	if (config != NULL) {
628		nvlist_t *features;
629
630		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
631		    &features) == 0) {
632			VERIFY(nvlist_dup(features, &spa->spa_label_features,
633			    0) == 0);
634		}
635
636		VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
637	}
638
639	if (spa->spa_label_features == NULL) {
640		VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
641		    KM_SLEEP) == 0);
642	}
643
644	spa->spa_iokstat = kstat_create("zfs", 0, name,
645	    "disk", KSTAT_TYPE_IO, 1, 0);
646	if (spa->spa_iokstat) {
647		spa->spa_iokstat->ks_lock = &spa->spa_iokstat_lock;
648		kstat_install(spa->spa_iokstat);
649	}
650
651	spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0);
652
653	spa->spa_min_ashift = INT_MAX;
654	spa->spa_max_ashift = 0;
655
656	/*
657	 * As a pool is being created, treat all features as disabled by
658	 * setting SPA_FEATURE_DISABLED for all entries in the feature
659	 * refcount cache.
660	 */
661	for (int i = 0; i < SPA_FEATURES; i++) {
662		spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED;
663	}
664
665	return (spa);
666}
667
668/*
669 * Removes a spa_t from the namespace, freeing up any memory used.  Requires
670 * spa_namespace_lock.  This is called only after the spa_t has been closed and
671 * deactivated.
672 */
673void
674spa_remove(spa_t *spa)
675{
676	spa_config_dirent_t *dp;
677
678	ASSERT(MUTEX_HELD(&spa_namespace_lock));
679	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
680	ASSERT3U(refcount_count(&spa->spa_refcount), ==, 0);
681
682	nvlist_free(spa->spa_config_splitting);
683
684	avl_remove(&spa_namespace_avl, spa);
685	cv_broadcast(&spa_namespace_cv);
686
687	if (spa->spa_root) {
688		spa_strfree(spa->spa_root);
689		spa_active_count--;
690	}
691
692	while ((dp = list_head(&spa->spa_config_list)) != NULL) {
693		list_remove(&spa->spa_config_list, dp);
694		if (dp->scd_path != NULL)
695			spa_strfree(dp->scd_path);
696		kmem_free(dp, sizeof (spa_config_dirent_t));
697	}
698
699	list_destroy(&spa->spa_config_list);
700
701	nvlist_free(spa->spa_label_features);
702	nvlist_free(spa->spa_load_info);
703	spa_config_set(spa, NULL);
704
705	mutex_enter(&cpu_lock);
706	if (spa->spa_deadman_cycid != CYCLIC_NONE)
707		cyclic_remove(spa->spa_deadman_cycid);
708	mutex_exit(&cpu_lock);
709	spa->spa_deadman_cycid = CYCLIC_NONE;
710
711	refcount_destroy(&spa->spa_refcount);
712
713	spa_config_lock_destroy(spa);
714
715	kstat_delete(spa->spa_iokstat);
716	spa->spa_iokstat = NULL;
717
718	for (int t = 0; t < TXG_SIZE; t++)
719		bplist_destroy(&spa->spa_free_bplist[t]);
720
721	zio_checksum_templates_free(spa);
722
723	cv_destroy(&spa->spa_async_cv);
724	cv_destroy(&spa->spa_evicting_os_cv);
725	cv_destroy(&spa->spa_proc_cv);
726	cv_destroy(&spa->spa_scrub_io_cv);
727	cv_destroy(&spa->spa_suspend_cv);
728
729	mutex_destroy(&spa->spa_async_lock);
730	mutex_destroy(&spa->spa_errlist_lock);
731	mutex_destroy(&spa->spa_errlog_lock);
732	mutex_destroy(&spa->spa_evicting_os_lock);
733	mutex_destroy(&spa->spa_history_lock);
734	mutex_destroy(&spa->spa_proc_lock);
735	mutex_destroy(&spa->spa_props_lock);
736	mutex_destroy(&spa->spa_cksum_tmpls_lock);
737	mutex_destroy(&spa->spa_scrub_lock);
738	mutex_destroy(&spa->spa_suspend_lock);
739	mutex_destroy(&spa->spa_vdev_top_lock);
740	mutex_destroy(&spa->spa_iokstat_lock);
741
742	kmem_free(spa, sizeof (spa_t));
743}
744
745/*
746 * Given a pool, return the next pool in the namespace, or NULL if there is
747 * none.  If 'prev' is NULL, return the first pool.
748 */
749spa_t *
750spa_next(spa_t *prev)
751{
752	ASSERT(MUTEX_HELD(&spa_namespace_lock));
753
754	if (prev)
755		return (AVL_NEXT(&spa_namespace_avl, prev));
756	else
757		return (avl_first(&spa_namespace_avl));
758}
759
760/*
761 * ==========================================================================
762 * SPA refcount functions
763 * ==========================================================================
764 */
765
766/*
767 * Add a reference to the given spa_t.  Must have at least one reference, or
768 * have the namespace lock held.
769 */
770void
771spa_open_ref(spa_t *spa, void *tag)
772{
773	ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
774	    MUTEX_HELD(&spa_namespace_lock));
775	(void) refcount_add(&spa->spa_refcount, tag);
776}
777
778/*
779 * Remove a reference to the given spa_t.  Must have at least one reference, or
780 * have the namespace lock held.
781 */
782void
783spa_close(spa_t *spa, void *tag)
784{
785	ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
786	    MUTEX_HELD(&spa_namespace_lock));
787	(void) refcount_remove(&spa->spa_refcount, tag);
788}
789
790/*
791 * Remove a reference to the given spa_t held by a dsl dir that is
792 * being asynchronously released.  Async releases occur from a taskq
793 * performing eviction of dsl datasets and dirs.  The namespace lock
794 * isn't held and the hold by the object being evicted may contribute to
795 * spa_minref (e.g. dataset or directory released during pool export),
796 * so the asserts in spa_close() do not apply.
797 */
798void
799spa_async_close(spa_t *spa, void *tag)
800{
801	(void) refcount_remove(&spa->spa_refcount, tag);
802}
803
804/*
805 * Check to see if the spa refcount is zero.  Must be called with
806 * spa_namespace_lock held.  We really compare against spa_minref, which is the
807 * number of references acquired when opening a pool
808 */
809boolean_t
810spa_refcount_zero(spa_t *spa)
811{
812	ASSERT(MUTEX_HELD(&spa_namespace_lock));
813
814	return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
815}
816
817/*
818 * ==========================================================================
819 * SPA spare and l2cache tracking
820 * ==========================================================================
821 */
822
823/*
824 * Hot spares and cache devices are tracked using the same code below,
825 * for 'auxiliary' devices.
826 */
827
828typedef struct spa_aux {
829	uint64_t	aux_guid;
830	uint64_t	aux_pool;
831	avl_node_t	aux_avl;
832	int		aux_count;
833} spa_aux_t;
834
835static int
836spa_aux_compare(const void *a, const void *b)
837{
838	const spa_aux_t *sa = a;
839	const spa_aux_t *sb = b;
840
841	if (sa->aux_guid < sb->aux_guid)
842		return (-1);
843	else if (sa->aux_guid > sb->aux_guid)
844		return (1);
845	else
846		return (0);
847}
848
849void
850spa_aux_add(vdev_t *vd, avl_tree_t *avl)
851{
852	avl_index_t where;
853	spa_aux_t search;
854	spa_aux_t *aux;
855
856	search.aux_guid = vd->vdev_guid;
857	if ((aux = avl_find(avl, &search, &where)) != NULL) {
858		aux->aux_count++;
859	} else {
860		aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
861		aux->aux_guid = vd->vdev_guid;
862		aux->aux_count = 1;
863		avl_insert(avl, aux, where);
864	}
865}
866
867void
868spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
869{
870	spa_aux_t search;
871	spa_aux_t *aux;
872	avl_index_t where;
873
874	search.aux_guid = vd->vdev_guid;
875	aux = avl_find(avl, &search, &where);
876
877	ASSERT(aux != NULL);
878
879	if (--aux->aux_count == 0) {
880		avl_remove(avl, aux);
881		kmem_free(aux, sizeof (spa_aux_t));
882	} else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
883		aux->aux_pool = 0ULL;
884	}
885}
886
887boolean_t
888spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
889{
890	spa_aux_t search, *found;
891
892	search.aux_guid = guid;
893	found = avl_find(avl, &search, NULL);
894
895	if (pool) {
896		if (found)
897			*pool = found->aux_pool;
898		else
899			*pool = 0ULL;
900	}
901
902	if (refcnt) {
903		if (found)
904			*refcnt = found->aux_count;
905		else
906			*refcnt = 0;
907	}
908
909	return (found != NULL);
910}
911
912void
913spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
914{
915	spa_aux_t search, *found;
916	avl_index_t where;
917
918	search.aux_guid = vd->vdev_guid;
919	found = avl_find(avl, &search, &where);
920	ASSERT(found != NULL);
921	ASSERT(found->aux_pool == 0ULL);
922
923	found->aux_pool = spa_guid(vd->vdev_spa);
924}
925
926/*
927 * Spares are tracked globally due to the following constraints:
928 *
929 * 	- A spare may be part of multiple pools.
930 * 	- A spare may be added to a pool even if it's actively in use within
931 *	  another pool.
932 * 	- A spare in use in any pool can only be the source of a replacement if
933 *	  the target is a spare in the same pool.
934 *
935 * We keep track of all spares on the system through the use of a reference
936 * counted AVL tree.  When a vdev is added as a spare, or used as a replacement
937 * spare, then we bump the reference count in the AVL tree.  In addition, we set
938 * the 'vdev_isspare' member to indicate that the device is a spare (active or
939 * inactive).  When a spare is made active (used to replace a device in the
940 * pool), we also keep track of which pool its been made a part of.
941 *
942 * The 'spa_spare_lock' protects the AVL tree.  These functions are normally
943 * called under the spa_namespace lock as part of vdev reconfiguration.  The
944 * separate spare lock exists for the status query path, which does not need to
945 * be completely consistent with respect to other vdev configuration changes.
946 */
947
948static int
949spa_spare_compare(const void *a, const void *b)
950{
951	return (spa_aux_compare(a, b));
952}
953
954void
955spa_spare_add(vdev_t *vd)
956{
957	mutex_enter(&spa_spare_lock);
958	ASSERT(!vd->vdev_isspare);
959	spa_aux_add(vd, &spa_spare_avl);
960	vd->vdev_isspare = B_TRUE;
961	mutex_exit(&spa_spare_lock);
962}
963
964void
965spa_spare_remove(vdev_t *vd)
966{
967	mutex_enter(&spa_spare_lock);
968	ASSERT(vd->vdev_isspare);
969	spa_aux_remove(vd, &spa_spare_avl);
970	vd->vdev_isspare = B_FALSE;
971	mutex_exit(&spa_spare_lock);
972}
973
974boolean_t
975spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
976{
977	boolean_t found;
978
979	mutex_enter(&spa_spare_lock);
980	found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
981	mutex_exit(&spa_spare_lock);
982
983	return (found);
984}
985
986void
987spa_spare_activate(vdev_t *vd)
988{
989	mutex_enter(&spa_spare_lock);
990	ASSERT(vd->vdev_isspare);
991	spa_aux_activate(vd, &spa_spare_avl);
992	mutex_exit(&spa_spare_lock);
993}
994
995/*
996 * Level 2 ARC devices are tracked globally for the same reasons as spares.
997 * Cache devices currently only support one pool per cache device, and so
998 * for these devices the aux reference count is currently unused beyond 1.
999 */
1000
1001static int
1002spa_l2cache_compare(const void *a, const void *b)
1003{
1004	return (spa_aux_compare(a, b));
1005}
1006
1007void
1008spa_l2cache_add(vdev_t *vd)
1009{
1010	mutex_enter(&spa_l2cache_lock);
1011	ASSERT(!vd->vdev_isl2cache);
1012	spa_aux_add(vd, &spa_l2cache_avl);
1013	vd->vdev_isl2cache = B_TRUE;
1014	mutex_exit(&spa_l2cache_lock);
1015}
1016
1017void
1018spa_l2cache_remove(vdev_t *vd)
1019{
1020	mutex_enter(&spa_l2cache_lock);
1021	ASSERT(vd->vdev_isl2cache);
1022	spa_aux_remove(vd, &spa_l2cache_avl);
1023	vd->vdev_isl2cache = B_FALSE;
1024	mutex_exit(&spa_l2cache_lock);
1025}
1026
1027boolean_t
1028spa_l2cache_exists(uint64_t guid, uint64_t *pool)
1029{
1030	boolean_t found;
1031
1032	mutex_enter(&spa_l2cache_lock);
1033	found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
1034	mutex_exit(&spa_l2cache_lock);
1035
1036	return (found);
1037}
1038
1039void
1040spa_l2cache_activate(vdev_t *vd)
1041{
1042	mutex_enter(&spa_l2cache_lock);
1043	ASSERT(vd->vdev_isl2cache);
1044	spa_aux_activate(vd, &spa_l2cache_avl);
1045	mutex_exit(&spa_l2cache_lock);
1046}
1047
1048/*
1049 * ==========================================================================
1050 * SPA vdev locking
1051 * ==========================================================================
1052 */
1053
1054/*
1055 * Lock the given spa_t for the purpose of adding or removing a vdev.
1056 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1057 * It returns the next transaction group for the spa_t.
1058 */
1059uint64_t
1060spa_vdev_enter(spa_t *spa)
1061{
1062	mutex_enter(&spa->spa_vdev_top_lock);
1063	mutex_enter(&spa_namespace_lock);
1064	return (spa_vdev_config_enter(spa));
1065}
1066
1067/*
1068 * Internal implementation for spa_vdev_enter().  Used when a vdev
1069 * operation requires multiple syncs (i.e. removing a device) while
1070 * keeping the spa_namespace_lock held.
1071 */
1072uint64_t
1073spa_vdev_config_enter(spa_t *spa)
1074{
1075	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1076
1077	spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1078
1079	return (spa_last_synced_txg(spa) + 1);
1080}
1081
1082/*
1083 * Used in combination with spa_vdev_config_enter() to allow the syncing
1084 * of multiple transactions without releasing the spa_namespace_lock.
1085 */
1086void
1087spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
1088{
1089	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1090
1091	int config_changed = B_FALSE;
1092
1093	ASSERT(txg > spa_last_synced_txg(spa));
1094
1095	spa->spa_pending_vdev = NULL;
1096
1097	/*
1098	 * Reassess the DTLs.
1099	 */
1100	vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
1101
1102	if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
1103		config_changed = B_TRUE;
1104		spa->spa_config_generation++;
1105	}
1106
1107	/*
1108	 * Verify the metaslab classes.
1109	 */
1110	ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
1111	ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
1112
1113	spa_config_exit(spa, SCL_ALL, spa);
1114
1115	/*
1116	 * Panic the system if the specified tag requires it.  This
1117	 * is useful for ensuring that configurations are updated
1118	 * transactionally.
1119	 */
1120	if (zio_injection_enabled)
1121		zio_handle_panic_injection(spa, tag, 0);
1122
1123	/*
1124	 * Note: this txg_wait_synced() is important because it ensures
1125	 * that there won't be more than one config change per txg.
1126	 * This allows us to use the txg as the generation number.
1127	 */
1128	if (error == 0)
1129		txg_wait_synced(spa->spa_dsl_pool, txg);
1130
1131	if (vd != NULL) {
1132		ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
1133		spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1134		vdev_free(vd);
1135		spa_config_exit(spa, SCL_ALL, spa);
1136	}
1137
1138	/*
1139	 * If the config changed, update the config cache.
1140	 */
1141	if (config_changed)
1142		spa_config_sync(spa, B_FALSE, B_TRUE);
1143}
1144
1145/*
1146 * Unlock the spa_t after adding or removing a vdev.  Besides undoing the
1147 * locking of spa_vdev_enter(), we also want make sure the transactions have
1148 * synced to disk, and then update the global configuration cache with the new
1149 * information.
1150 */
1151int
1152spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1153{
1154	spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1155	mutex_exit(&spa_namespace_lock);
1156	mutex_exit(&spa->spa_vdev_top_lock);
1157
1158	return (error);
1159}
1160
1161/*
1162 * Lock the given spa_t for the purpose of changing vdev state.
1163 */
1164void
1165spa_vdev_state_enter(spa_t *spa, int oplocks)
1166{
1167	int locks = SCL_STATE_ALL | oplocks;
1168
1169	/*
1170	 * Root pools may need to read of the underlying devfs filesystem
1171	 * when opening up a vdev.  Unfortunately if we're holding the
1172	 * SCL_ZIO lock it will result in a deadlock when we try to issue
1173	 * the read from the root filesystem.  Instead we "prefetch"
1174	 * the associated vnodes that we need prior to opening the
1175	 * underlying devices and cache them so that we can prevent
1176	 * any I/O when we are doing the actual open.
1177	 */
1178	if (spa_is_root(spa)) {
1179		int low = locks & ~(SCL_ZIO - 1);
1180		int high = locks & ~low;
1181
1182		spa_config_enter(spa, high, spa, RW_WRITER);
1183		vdev_hold(spa->spa_root_vdev);
1184		spa_config_enter(spa, low, spa, RW_WRITER);
1185	} else {
1186		spa_config_enter(spa, locks, spa, RW_WRITER);
1187	}
1188	spa->spa_vdev_locks = locks;
1189}
1190
1191int
1192spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1193{
1194	boolean_t config_changed = B_FALSE;
1195
1196	if (vd != NULL || error == 0)
1197		vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1198		    0, 0, B_FALSE);
1199
1200	if (vd != NULL) {
1201		vdev_state_dirty(vd->vdev_top);
1202		config_changed = B_TRUE;
1203		spa->spa_config_generation++;
1204	}
1205
1206	if (spa_is_root(spa))
1207		vdev_rele(spa->spa_root_vdev);
1208
1209	ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1210	spa_config_exit(spa, spa->spa_vdev_locks, spa);
1211
1212	/*
1213	 * If anything changed, wait for it to sync.  This ensures that,
1214	 * from the system administrator's perspective, zpool(1M) commands
1215	 * are synchronous.  This is important for things like zpool offline:
1216	 * when the command completes, you expect no further I/O from ZFS.
1217	 */
1218	if (vd != NULL)
1219		txg_wait_synced(spa->spa_dsl_pool, 0);
1220
1221	/*
1222	 * If the config changed, update the config cache.
1223	 */
1224	if (config_changed) {
1225		mutex_enter(&spa_namespace_lock);
1226		spa_config_sync(spa, B_FALSE, B_TRUE);
1227		mutex_exit(&spa_namespace_lock);
1228	}
1229
1230	return (error);
1231}
1232
1233/*
1234 * ==========================================================================
1235 * Miscellaneous functions
1236 * ==========================================================================
1237 */
1238
1239void
1240spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx)
1241{
1242	if (!nvlist_exists(spa->spa_label_features, feature)) {
1243		fnvlist_add_boolean(spa->spa_label_features, feature);
1244		/*
1245		 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1246		 * dirty the vdev config because lock SCL_CONFIG is not held.
1247		 * Thankfully, in this case we don't need to dirty the config
1248		 * because it will be written out anyway when we finish
1249		 * creating the pool.
1250		 */
1251		if (tx->tx_txg != TXG_INITIAL)
1252			vdev_config_dirty(spa->spa_root_vdev);
1253	}
1254}
1255
1256void
1257spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1258{
1259	if (nvlist_remove_all(spa->spa_label_features, feature) == 0)
1260		vdev_config_dirty(spa->spa_root_vdev);
1261}
1262
1263/*
1264 * Rename a spa_t.
1265 */
1266int
1267spa_rename(const char *name, const char *newname)
1268{
1269	spa_t *spa;
1270	int err;
1271
1272	/*
1273	 * Lookup the spa_t and grab the config lock for writing.  We need to
1274	 * actually open the pool so that we can sync out the necessary labels.
1275	 * It's OK to call spa_open() with the namespace lock held because we
1276	 * allow recursive calls for other reasons.
1277	 */
1278	mutex_enter(&spa_namespace_lock);
1279	if ((err = spa_open(name, &spa, FTAG)) != 0) {
1280		mutex_exit(&spa_namespace_lock);
1281		return (err);
1282	}
1283
1284	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1285
1286	avl_remove(&spa_namespace_avl, spa);
1287	(void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1288	avl_add(&spa_namespace_avl, spa);
1289
1290	/*
1291	 * Sync all labels to disk with the new names by marking the root vdev
1292	 * dirty and waiting for it to sync.  It will pick up the new pool name
1293	 * during the sync.
1294	 */
1295	vdev_config_dirty(spa->spa_root_vdev);
1296
1297	spa_config_exit(spa, SCL_ALL, FTAG);
1298
1299	txg_wait_synced(spa->spa_dsl_pool, 0);
1300
1301	/*
1302	 * Sync the updated config cache.
1303	 */
1304	spa_config_sync(spa, B_FALSE, B_TRUE);
1305
1306	spa_close(spa, FTAG);
1307
1308	mutex_exit(&spa_namespace_lock);
1309
1310	return (0);
1311}
1312
1313/*
1314 * Return the spa_t associated with given pool_guid, if it exists.  If
1315 * device_guid is non-zero, determine whether the pool exists *and* contains
1316 * a device with the specified device_guid.
1317 */
1318spa_t *
1319spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1320{
1321	spa_t *spa;
1322	avl_tree_t *t = &spa_namespace_avl;
1323
1324	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1325
1326	for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1327		if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1328			continue;
1329		if (spa->spa_root_vdev == NULL)
1330			continue;
1331		if (spa_guid(spa) == pool_guid) {
1332			if (device_guid == 0)
1333				break;
1334
1335			if (vdev_lookup_by_guid(spa->spa_root_vdev,
1336			    device_guid) != NULL)
1337				break;
1338
1339			/*
1340			 * Check any devices we may be in the process of adding.
1341			 */
1342			if (spa->spa_pending_vdev) {
1343				if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1344				    device_guid) != NULL)
1345					break;
1346			}
1347		}
1348	}
1349
1350	return (spa);
1351}
1352
1353/*
1354 * Determine whether a pool with the given pool_guid exists.
1355 */
1356boolean_t
1357spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1358{
1359	return (spa_by_guid(pool_guid, device_guid) != NULL);
1360}
1361
1362char *
1363spa_strdup(const char *s)
1364{
1365	size_t len;
1366	char *new;
1367
1368	len = strlen(s);
1369	new = kmem_alloc(len + 1, KM_SLEEP);
1370	bcopy(s, new, len);
1371	new[len] = '\0';
1372
1373	return (new);
1374}
1375
1376void
1377spa_strfree(char *s)
1378{
1379	kmem_free(s, strlen(s) + 1);
1380}
1381
1382uint64_t
1383spa_get_random(uint64_t range)
1384{
1385	uint64_t r;
1386
1387	ASSERT(range != 0);
1388
1389	(void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1390
1391	return (r % range);
1392}
1393
1394uint64_t
1395spa_generate_guid(spa_t *spa)
1396{
1397	uint64_t guid = spa_get_random(-1ULL);
1398
1399	if (spa != NULL) {
1400		while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1401			guid = spa_get_random(-1ULL);
1402	} else {
1403		while (guid == 0 || spa_guid_exists(guid, 0))
1404			guid = spa_get_random(-1ULL);
1405	}
1406
1407	return (guid);
1408}
1409
1410void
1411snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp)
1412{
1413	char type[256];
1414	char *checksum = NULL;
1415	char *compress = NULL;
1416
1417	if (bp != NULL) {
1418		if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1419			dmu_object_byteswap_t bswap =
1420			    DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1421			(void) snprintf(type, sizeof (type), "bswap %s %s",
1422			    DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1423			    "metadata" : "data",
1424			    dmu_ot_byteswap[bswap].ob_name);
1425		} else {
1426			(void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1427			    sizeof (type));
1428		}
1429		if (!BP_IS_EMBEDDED(bp)) {
1430			checksum =
1431			    zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1432		}
1433		compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1434	}
1435
1436	SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum,
1437	    compress);
1438}
1439
1440void
1441spa_freeze(spa_t *spa)
1442{
1443	uint64_t freeze_txg = 0;
1444
1445	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1446	if (spa->spa_freeze_txg == UINT64_MAX) {
1447		freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1448		spa->spa_freeze_txg = freeze_txg;
1449	}
1450	spa_config_exit(spa, SCL_ALL, FTAG);
1451	if (freeze_txg != 0)
1452		txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1453}
1454
1455void
1456zfs_panic_recover(const char *fmt, ...)
1457{
1458	va_list adx;
1459
1460	va_start(adx, fmt);
1461	vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1462	va_end(adx);
1463}
1464
1465/*
1466 * This is a stripped-down version of strtoull, suitable only for converting
1467 * lowercase hexadecimal numbers that don't overflow.
1468 */
1469uint64_t
1470strtonum(const char *str, char **nptr)
1471{
1472	uint64_t val = 0;
1473	char c;
1474	int digit;
1475
1476	while ((c = *str) != '\0') {
1477		if (c >= '0' && c <= '9')
1478			digit = c - '0';
1479		else if (c >= 'a' && c <= 'f')
1480			digit = 10 + c - 'a';
1481		else
1482			break;
1483
1484		val *= 16;
1485		val += digit;
1486
1487		str++;
1488	}
1489
1490	if (nptr)
1491		*nptr = (char *)str;
1492
1493	return (val);
1494}
1495
1496/*
1497 * ==========================================================================
1498 * Accessor functions
1499 * ==========================================================================
1500 */
1501
1502boolean_t
1503spa_shutting_down(spa_t *spa)
1504{
1505	return (spa->spa_async_suspended);
1506}
1507
1508dsl_pool_t *
1509spa_get_dsl(spa_t *spa)
1510{
1511	return (spa->spa_dsl_pool);
1512}
1513
1514boolean_t
1515spa_is_initializing(spa_t *spa)
1516{
1517	return (spa->spa_is_initializing);
1518}
1519
1520blkptr_t *
1521spa_get_rootblkptr(spa_t *spa)
1522{
1523	return (&spa->spa_ubsync.ub_rootbp);
1524}
1525
1526void
1527spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1528{
1529	spa->spa_uberblock.ub_rootbp = *bp;
1530}
1531
1532void
1533spa_altroot(spa_t *spa, char *buf, size_t buflen)
1534{
1535	if (spa->spa_root == NULL)
1536		buf[0] = '\0';
1537	else
1538		(void) strncpy(buf, spa->spa_root, buflen);
1539}
1540
1541int
1542spa_sync_pass(spa_t *spa)
1543{
1544	return (spa->spa_sync_pass);
1545}
1546
1547char *
1548spa_name(spa_t *spa)
1549{
1550	return (spa->spa_name);
1551}
1552
1553uint64_t
1554spa_guid(spa_t *spa)
1555{
1556	dsl_pool_t *dp = spa_get_dsl(spa);
1557	uint64_t guid;
1558
1559	/*
1560	 * If we fail to parse the config during spa_load(), we can go through
1561	 * the error path (which posts an ereport) and end up here with no root
1562	 * vdev.  We stash the original pool guid in 'spa_config_guid' to handle
1563	 * this case.
1564	 */
1565	if (spa->spa_root_vdev == NULL)
1566		return (spa->spa_config_guid);
1567
1568	guid = spa->spa_last_synced_guid != 0 ?
1569	    spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1570
1571	/*
1572	 * Return the most recently synced out guid unless we're
1573	 * in syncing context.
1574	 */
1575	if (dp && dsl_pool_sync_context(dp))
1576		return (spa->spa_root_vdev->vdev_guid);
1577	else
1578		return (guid);
1579}
1580
1581uint64_t
1582spa_load_guid(spa_t *spa)
1583{
1584	/*
1585	 * This is a GUID that exists solely as a reference for the
1586	 * purposes of the arc.  It is generated at load time, and
1587	 * is never written to persistent storage.
1588	 */
1589	return (spa->spa_load_guid);
1590}
1591
1592uint64_t
1593spa_last_synced_txg(spa_t *spa)
1594{
1595	return (spa->spa_ubsync.ub_txg);
1596}
1597
1598uint64_t
1599spa_first_txg(spa_t *spa)
1600{
1601	return (spa->spa_first_txg);
1602}
1603
1604uint64_t
1605spa_syncing_txg(spa_t *spa)
1606{
1607	return (spa->spa_syncing_txg);
1608}
1609
1610pool_state_t
1611spa_state(spa_t *spa)
1612{
1613	return (spa->spa_state);
1614}
1615
1616spa_load_state_t
1617spa_load_state(spa_t *spa)
1618{
1619	return (spa->spa_load_state);
1620}
1621
1622uint64_t
1623spa_freeze_txg(spa_t *spa)
1624{
1625	return (spa->spa_freeze_txg);
1626}
1627
1628/* ARGSUSED */
1629uint64_t
1630spa_get_asize(spa_t *spa, uint64_t lsize)
1631{
1632	return (lsize * spa_asize_inflation);
1633}
1634
1635/*
1636 * Return the amount of slop space in bytes.  It is 1/32 of the pool (3.2%),
1637 * or at least 32MB.
1638 *
1639 * See the comment above spa_slop_shift for details.
1640 */
1641uint64_t
1642spa_get_slop_space(spa_t *spa) {
1643	uint64_t space = spa_get_dspace(spa);
1644	return (MAX(space >> spa_slop_shift, SPA_MINDEVSIZE >> 1));
1645}
1646
1647uint64_t
1648spa_get_dspace(spa_t *spa)
1649{
1650	return (spa->spa_dspace);
1651}
1652
1653void
1654spa_update_dspace(spa_t *spa)
1655{
1656	spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1657	    ddt_get_dedup_dspace(spa);
1658}
1659
1660/*
1661 * Return the failure mode that has been set to this pool. The default
1662 * behavior will be to block all I/Os when a complete failure occurs.
1663 */
1664uint8_t
1665spa_get_failmode(spa_t *spa)
1666{
1667	return (spa->spa_failmode);
1668}
1669
1670boolean_t
1671spa_suspended(spa_t *spa)
1672{
1673	return (spa->spa_suspended);
1674}
1675
1676uint64_t
1677spa_version(spa_t *spa)
1678{
1679	return (spa->spa_ubsync.ub_version);
1680}
1681
1682boolean_t
1683spa_deflate(spa_t *spa)
1684{
1685	return (spa->spa_deflate);
1686}
1687
1688metaslab_class_t *
1689spa_normal_class(spa_t *spa)
1690{
1691	return (spa->spa_normal_class);
1692}
1693
1694metaslab_class_t *
1695spa_log_class(spa_t *spa)
1696{
1697	return (spa->spa_log_class);
1698}
1699
1700void
1701spa_evicting_os_register(spa_t *spa, objset_t *os)
1702{
1703	mutex_enter(&spa->spa_evicting_os_lock);
1704	list_insert_head(&spa->spa_evicting_os_list, os);
1705	mutex_exit(&spa->spa_evicting_os_lock);
1706}
1707
1708void
1709spa_evicting_os_deregister(spa_t *spa, objset_t *os)
1710{
1711	mutex_enter(&spa->spa_evicting_os_lock);
1712	list_remove(&spa->spa_evicting_os_list, os);
1713	cv_broadcast(&spa->spa_evicting_os_cv);
1714	mutex_exit(&spa->spa_evicting_os_lock);
1715}
1716
1717void
1718spa_evicting_os_wait(spa_t *spa)
1719{
1720	mutex_enter(&spa->spa_evicting_os_lock);
1721	while (!list_is_empty(&spa->spa_evicting_os_list))
1722		cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock);
1723	mutex_exit(&spa->spa_evicting_os_lock);
1724
1725	dmu_buf_user_evict_wait();
1726}
1727
1728int
1729spa_max_replication(spa_t *spa)
1730{
1731	/*
1732	 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1733	 * handle BPs with more than one DVA allocated.  Set our max
1734	 * replication level accordingly.
1735	 */
1736	if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1737		return (1);
1738	return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1739}
1740
1741int
1742spa_prev_software_version(spa_t *spa)
1743{
1744	return (spa->spa_prev_software_version);
1745}
1746
1747uint64_t
1748spa_deadman_synctime(spa_t *spa)
1749{
1750	return (spa->spa_deadman_synctime);
1751}
1752
1753uint64_t
1754dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1755{
1756	uint64_t asize = DVA_GET_ASIZE(dva);
1757	uint64_t dsize = asize;
1758
1759	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1760
1761	if (asize != 0 && spa->spa_deflate) {
1762		vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1763		dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1764	}
1765
1766	return (dsize);
1767}
1768
1769uint64_t
1770bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1771{
1772	uint64_t dsize = 0;
1773
1774	for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1775		dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1776
1777	return (dsize);
1778}
1779
1780uint64_t
1781bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1782{
1783	uint64_t dsize = 0;
1784
1785	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1786
1787	for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1788		dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1789
1790	spa_config_exit(spa, SCL_VDEV, FTAG);
1791
1792	return (dsize);
1793}
1794
1795/*
1796 * ==========================================================================
1797 * Initialization and Termination
1798 * ==========================================================================
1799 */
1800
1801static int
1802spa_name_compare(const void *a1, const void *a2)
1803{
1804	const spa_t *s1 = a1;
1805	const spa_t *s2 = a2;
1806	int s;
1807
1808	s = strcmp(s1->spa_name, s2->spa_name);
1809	if (s > 0)
1810		return (1);
1811	if (s < 0)
1812		return (-1);
1813	return (0);
1814}
1815
1816int
1817spa_busy(void)
1818{
1819	return (spa_active_count);
1820}
1821
1822void
1823spa_boot_init()
1824{
1825	spa_config_load();
1826}
1827
1828void
1829spa_init(int mode)
1830{
1831	mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1832	mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1833	mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1834	cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1835
1836	avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1837	    offsetof(spa_t, spa_avl));
1838
1839	avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1840	    offsetof(spa_aux_t, aux_avl));
1841
1842	avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1843	    offsetof(spa_aux_t, aux_avl));
1844
1845	spa_mode_global = mode;
1846
1847#ifdef _KERNEL
1848	spa_arch_init();
1849#else
1850	if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
1851		arc_procfd = open("/proc/self/ctl", O_WRONLY);
1852		if (arc_procfd == -1) {
1853			perror("could not enable watchpoints: "
1854			    "opening /proc/self/ctl failed: ");
1855		} else {
1856			arc_watch = B_TRUE;
1857		}
1858	}
1859#endif
1860
1861	refcount_init();
1862	unique_init();
1863	range_tree_init();
1864	zio_init();
1865	dmu_init();
1866	zil_init();
1867	vdev_cache_stat_init();
1868	zfs_prop_init();
1869	zpool_prop_init();
1870	zpool_feature_init();
1871	spa_config_load();
1872	l2arc_start();
1873}
1874
1875void
1876spa_fini(void)
1877{
1878	l2arc_stop();
1879
1880	spa_evict_all();
1881
1882	vdev_cache_stat_fini();
1883	zil_fini();
1884	dmu_fini();
1885	zio_fini();
1886	range_tree_fini();
1887	unique_fini();
1888	refcount_fini();
1889
1890	avl_destroy(&spa_namespace_avl);
1891	avl_destroy(&spa_spare_avl);
1892	avl_destroy(&spa_l2cache_avl);
1893
1894	cv_destroy(&spa_namespace_cv);
1895	mutex_destroy(&spa_namespace_lock);
1896	mutex_destroy(&spa_spare_lock);
1897	mutex_destroy(&spa_l2cache_lock);
1898}
1899
1900/*
1901 * Return whether this pool has slogs. No locking needed.
1902 * It's not a problem if the wrong answer is returned as it's only for
1903 * performance and not correctness
1904 */
1905boolean_t
1906spa_has_slogs(spa_t *spa)
1907{
1908	return (spa->spa_log_class->mc_rotor != NULL);
1909}
1910
1911spa_log_state_t
1912spa_get_log_state(spa_t *spa)
1913{
1914	return (spa->spa_log_state);
1915}
1916
1917void
1918spa_set_log_state(spa_t *spa, spa_log_state_t state)
1919{
1920	spa->spa_log_state = state;
1921}
1922
1923boolean_t
1924spa_is_root(spa_t *spa)
1925{
1926	return (spa->spa_is_root);
1927}
1928
1929boolean_t
1930spa_writeable(spa_t *spa)
1931{
1932	return (!!(spa->spa_mode & FWRITE));
1933}
1934
1935/*
1936 * Returns true if there is a pending sync task in any of the current
1937 * syncing txg, the current quiescing txg, or the current open txg.
1938 */
1939boolean_t
1940spa_has_pending_synctask(spa_t *spa)
1941{
1942	return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks));
1943}
1944
1945int
1946spa_mode(spa_t *spa)
1947{
1948	return (spa->spa_mode);
1949}
1950
1951uint64_t
1952spa_bootfs(spa_t *spa)
1953{
1954	return (spa->spa_bootfs);
1955}
1956
1957uint64_t
1958spa_delegation(spa_t *spa)
1959{
1960	return (spa->spa_delegation);
1961}
1962
1963objset_t *
1964spa_meta_objset(spa_t *spa)
1965{
1966	return (spa->spa_meta_objset);
1967}
1968
1969enum zio_checksum
1970spa_dedup_checksum(spa_t *spa)
1971{
1972	return (spa->spa_dedup_checksum);
1973}
1974
1975/*
1976 * Reset pool scan stat per scan pass (or reboot).
1977 */
1978void
1979spa_scan_stat_init(spa_t *spa)
1980{
1981	/* data not stored on disk */
1982	spa->spa_scan_pass_start = gethrestime_sec();
1983	spa->spa_scan_pass_exam = 0;
1984	vdev_scan_stat_init(spa->spa_root_vdev);
1985}
1986
1987/*
1988 * Get scan stats for zpool status reports
1989 */
1990int
1991spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
1992{
1993	dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
1994
1995	if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
1996		return (SET_ERROR(ENOENT));
1997	bzero(ps, sizeof (pool_scan_stat_t));
1998
1999	/* data stored on disk */
2000	ps->pss_func = scn->scn_phys.scn_func;
2001	ps->pss_start_time = scn->scn_phys.scn_start_time;
2002	ps->pss_end_time = scn->scn_phys.scn_end_time;
2003	ps->pss_to_examine = scn->scn_phys.scn_to_examine;
2004	ps->pss_examined = scn->scn_phys.scn_examined;
2005	ps->pss_to_process = scn->scn_phys.scn_to_process;
2006	ps->pss_processed = scn->scn_phys.scn_processed;
2007	ps->pss_errors = scn->scn_phys.scn_errors;
2008	ps->pss_state = scn->scn_phys.scn_state;
2009
2010	/* data not stored on disk */
2011	ps->pss_pass_start = spa->spa_scan_pass_start;
2012	ps->pss_pass_exam = spa->spa_scan_pass_exam;
2013
2014	return (0);
2015}
2016
2017boolean_t
2018spa_debug_enabled(spa_t *spa)
2019{
2020	return (spa->spa_debug);
2021}
2022
2023int
2024spa_maxblocksize(spa_t *spa)
2025{
2026	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS))
2027		return (SPA_MAXBLOCKSIZE);
2028	else
2029		return (SPA_OLD_MAXBLOCKSIZE);
2030}
2031