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