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