xref: /illumos-gate/usr/src/uts/common/fs/zfs/spa_misc.c (revision 3a4b1be9)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
24  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
25  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
26  * Copyright 2013 Saso Kiselkov. All rights reserved.
27  * Copyright (c) 2014 Integros [integros.com]
28  * Copyright (c) 2017 Datto Inc.
29  */
30 
31 #include <sys/zfs_context.h>
32 #include <sys/spa_impl.h>
33 #include <sys/spa_boot.h>
34 #include <sys/zio.h>
35 #include <sys/zio_checksum.h>
36 #include <sys/zio_compress.h>
37 #include <sys/dmu.h>
38 #include <sys/dmu_tx.h>
39 #include <sys/zap.h>
40 #include <sys/zil.h>
41 #include <sys/vdev_impl.h>
42 #include <sys/metaslab.h>
43 #include <sys/uberblock_impl.h>
44 #include <sys/txg.h>
45 #include <sys/avl.h>
46 #include <sys/unique.h>
47 #include <sys/dsl_pool.h>
48 #include <sys/dsl_dir.h>
49 #include <sys/dsl_prop.h>
50 #include <sys/dsl_scan.h>
51 #include <sys/fs/zfs.h>
52 #include <sys/metaslab_impl.h>
53 #include <sys/arc.h>
54 #include <sys/ddt.h>
55 #include "zfs_prop.h"
56 #include <sys/zfeature.h>
57 
58 /*
59  * SPA locking
60  *
61  * There are four basic locks for managing spa_t structures:
62  *
63  * spa_namespace_lock (global mutex)
64  *
65  *	This lock must be acquired to do any of the following:
66  *
67  *		- Lookup a spa_t by name
68  *		- Add or remove a spa_t from the namespace
69  *		- Increase spa_refcount from non-zero
70  *		- Check if spa_refcount is zero
71  *		- Rename a spa_t
72  *		- add/remove/attach/detach devices
73  *		- Held for the duration of create/destroy/import/export
74  *
75  *	It does not need to handle recursion.  A create or destroy may
76  *	reference objects (files or zvols) in other pools, but by
77  *	definition they must have an existing reference, and will never need
78  *	to lookup a spa_t by name.
79  *
80  * spa_refcount (per-spa refcount_t protected by mutex)
81  *
82  *	This reference count keep track of any active users of the spa_t.  The
83  *	spa_t cannot be destroyed or freed while this is non-zero.  Internally,
84  *	the refcount is never really 'zero' - opening a pool implicitly keeps
85  *	some references in the DMU.  Internally we check against spa_minref, but
86  *	present the image of a zero/non-zero value to consumers.
87  *
88  * spa_config_lock[] (per-spa array of rwlocks)
89  *
90  *	This protects the spa_t from config changes, and must be held in
91  *	the following circumstances:
92  *
93  *		- RW_READER to perform I/O to the spa
94  *		- RW_WRITER to change the vdev config
95  *
96  * The locking order is fairly straightforward:
97  *
98  *		spa_namespace_lock	->	spa_refcount
99  *
100  *	The namespace lock must be acquired to increase the refcount from 0
101  *	or to check if it is zero.
102  *
103  *		spa_refcount		->	spa_config_lock[]
104  *
105  *	There must be at least one valid reference on the spa_t to acquire
106  *	the config lock.
107  *
108  *		spa_namespace_lock	->	spa_config_lock[]
109  *
110  *	The namespace lock must always be taken before the config lock.
111  *
112  *
113  * The spa_namespace_lock can be acquired directly and is globally visible.
114  *
115  * The namespace is manipulated using the following functions, all of which
116  * require the spa_namespace_lock to be held.
117  *
118  *	spa_lookup()		Lookup a spa_t by name.
119  *
120  *	spa_add()		Create a new spa_t in the namespace.
121  *
122  *	spa_remove()		Remove a spa_t from the namespace.  This also
123  *				frees up any memory associated with the spa_t.
124  *
125  *	spa_next()		Returns the next spa_t in the system, or the
126  *				first if NULL is passed.
127  *
128  *	spa_evict_all()		Shutdown and remove all spa_t structures in
129  *				the system.
130  *
131  *	spa_guid_exists()	Determine whether a pool/device guid exists.
132  *
133  * The spa_refcount is manipulated using the following functions:
134  *
135  *	spa_open_ref()		Adds a reference to the given spa_t.  Must be
136  *				called with spa_namespace_lock held if the
137  *				refcount is currently zero.
138  *
139  *	spa_close()		Remove a reference from the spa_t.  This will
140  *				not free the spa_t or remove it from the
141  *				namespace.  No locking is required.
142  *
143  *	spa_refcount_zero()	Returns true if the refcount is currently
144  *				zero.  Must be called with spa_namespace_lock
145  *				held.
146  *
147  * The spa_config_lock[] is an array of rwlocks, ordered as follows:
148  * SCL_CONFIG > SCL_STATE > SCL_ALLOC > SCL_ZIO > SCL_FREE > SCL_VDEV.
149  * spa_config_lock[] is manipulated with spa_config_{enter,exit,held}().
150  *
151  * To read the configuration, it suffices to hold one of these locks as reader.
152  * To modify the configuration, you must hold all locks as writer.  To modify
153  * vdev state without altering the vdev tree's topology (e.g. online/offline),
154  * you must hold SCL_STATE and SCL_ZIO as writer.
155  *
156  * We use these distinct config locks to avoid recursive lock entry.
157  * For example, spa_sync() (which holds SCL_CONFIG as reader) induces
158  * block allocations (SCL_ALLOC), which may require reading space maps
159  * from disk (dmu_read() -> zio_read() -> SCL_ZIO).
160  *
161  * The spa config locks cannot be normal rwlocks because we need the
162  * ability to hand off ownership.  For example, SCL_ZIO is acquired
163  * by the issuing thread and later released by an interrupt thread.
164  * They do, however, obey the usual write-wanted semantics to prevent
165  * writer (i.e. system administrator) starvation.
166  *
167  * The lock acquisition rules are as follows:
168  *
169  * SCL_CONFIG
170  *	Protects changes to the vdev tree topology, such as vdev
171  *	add/remove/attach/detach.  Protects the dirty config list
172  *	(spa_config_dirty_list) and the set of spares and l2arc devices.
173  *
174  * SCL_STATE
175  *	Protects changes to pool state and vdev state, such as vdev
176  *	online/offline/fault/degrade/clear.  Protects the dirty state list
177  *	(spa_state_dirty_list) and global pool state (spa_state).
178  *
179  * SCL_ALLOC
180  *	Protects changes to metaslab groups and classes.
181  *	Held as reader by metaslab_alloc() and metaslab_claim().
182  *
183  * SCL_ZIO
184  *	Held by bp-level zios (those which have no io_vd upon entry)
185  *	to prevent changes to the vdev tree.  The bp-level zio implicitly
186  *	protects all of its vdev child zios, which do not hold SCL_ZIO.
187  *
188  * SCL_FREE
189  *	Protects changes to metaslab groups and classes.
190  *	Held as reader by metaslab_free().  SCL_FREE is distinct from
191  *	SCL_ALLOC, and lower than SCL_ZIO, so that we can safely free
192  *	blocks in zio_done() while another i/o that holds either
193  *	SCL_ALLOC or SCL_ZIO is waiting for this i/o to complete.
194  *
195  * SCL_VDEV
196  *	Held as reader to prevent changes to the vdev tree during trivial
197  *	inquiries such as bp_get_dsize().  SCL_VDEV is distinct from the
198  *	other locks, and lower than all of them, to ensure that it's safe
199  *	to acquire regardless of caller context.
200  *
201  * In addition, the following rules apply:
202  *
203  * (a)	spa_props_lock protects pool properties, spa_config and spa_config_list.
204  *	The lock ordering is SCL_CONFIG > spa_props_lock.
205  *
206  * (b)	I/O operations on leaf vdevs.  For any zio operation that takes
207  *	an explicit vdev_t argument -- such as zio_ioctl(), zio_read_phys(),
208  *	or zio_write_phys() -- the caller must ensure that the config cannot
209  *	cannot change in the interim, and that the vdev cannot be reopened.
210  *	SCL_STATE as reader suffices for both.
211  *
212  * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
213  *
214  *	spa_vdev_enter()	Acquire the namespace lock and the config lock
215  *				for writing.
216  *
217  *	spa_vdev_exit()		Release the config lock, wait for all I/O
218  *				to complete, sync the updated configs to the
219  *				cache, and release the namespace lock.
220  *
221  * vdev state is protected by spa_vdev_state_enter() / spa_vdev_state_exit().
222  * Like spa_vdev_enter/exit, these are convenience wrappers -- the actual
223  * locking is, always, based on spa_namespace_lock and spa_config_lock[].
224  *
225  * spa_rename() is also implemented within this file since it requires
226  * manipulation of the namespace.
227  */
228 
229 static avl_tree_t spa_namespace_avl;
230 kmutex_t spa_namespace_lock;
231 static kcondvar_t spa_namespace_cv;
232 static int spa_active_count;
233 int spa_max_replication_override = SPA_DVAS_PER_BP;
234 
235 static kmutex_t spa_spare_lock;
236 static avl_tree_t spa_spare_avl;
237 static kmutex_t spa_l2cache_lock;
238 static avl_tree_t spa_l2cache_avl;
239 
240 kmem_cache_t *spa_buffer_pool;
241 int spa_mode_global;
242 
243 #ifdef ZFS_DEBUG
244 /*
245  * Everything except dprintf, spa, and indirect_remap is on by default
246  * in debug builds.
247  */
248 int zfs_flags = ~(ZFS_DEBUG_DPRINTF | ZFS_DEBUG_SPA | ZFS_DEBUG_INDIRECT_REMAP);
249 #else
250 int zfs_flags = 0;
251 #endif
252 
253 /*
254  * zfs_recover can be set to nonzero to attempt to recover from
255  * otherwise-fatal errors, typically caused by on-disk corruption.  When
256  * set, calls to zfs_panic_recover() will turn into warning messages.
257  * This should only be used as a last resort, as it typically results
258  * in leaked space, or worse.
259  */
260 boolean_t zfs_recover = B_FALSE;
261 
262 /*
263  * If destroy encounters an EIO while reading metadata (e.g. indirect
264  * blocks), space referenced by the missing metadata can not be freed.
265  * Normally this causes the background destroy to become "stalled", as
266  * it is unable to make forward progress.  While in this stalled state,
267  * all remaining space to free from the error-encountering filesystem is
268  * "temporarily leaked".  Set this flag to cause it to ignore the EIO,
269  * permanently leak the space from indirect blocks that can not be read,
270  * and continue to free everything else that it can.
271  *
272  * The default, "stalling" behavior is useful if the storage partially
273  * fails (i.e. some but not all i/os fail), and then later recovers.  In
274  * this case, we will be able to continue pool operations while it is
275  * partially failed, and when it recovers, we can continue to free the
276  * space, with no leaks.  However, note that this case is actually
277  * fairly rare.
278  *
279  * Typically pools either (a) fail completely (but perhaps temporarily,
280  * e.g. a top-level vdev going offline), or (b) have localized,
281  * permanent errors (e.g. disk returns the wrong data due to bit flip or
282  * firmware bug).  In case (a), this setting does not matter because the
283  * pool will be suspended and the sync thread will not be able to make
284  * forward progress regardless.  In case (b), because the error is
285  * permanent, the best we can do is leak the minimum amount of space,
286  * which is what setting this flag will do.  Therefore, it is reasonable
287  * for this flag to normally be set, but we chose the more conservative
288  * approach of not setting it, so that there is no possibility of
289  * leaking space in the "partial temporary" failure case.
290  */
291 boolean_t zfs_free_leak_on_eio = B_FALSE;
292 
293 /*
294  * Expiration time in milliseconds. This value has two meanings. First it is
295  * used to determine when the spa_deadman() logic should fire. By default the
296  * spa_deadman() will fire if spa_sync() has not completed in 1000 seconds.
297  * Secondly, the value determines if an I/O is considered "hung". Any I/O that
298  * has not completed in zfs_deadman_synctime_ms is considered "hung" resulting
299  * in a system panic.
300  */
301 uint64_t zfs_deadman_synctime_ms = 1000000ULL;
302 
303 /*
304  * Check time in milliseconds. This defines the frequency at which we check
305  * for hung I/O.
306  */
307 uint64_t zfs_deadman_checktime_ms = 5000ULL;
308 
309 /*
310  * Override the zfs deadman behavior via /etc/system. By default the
311  * deadman is enabled except on VMware and sparc deployments.
312  */
313 int zfs_deadman_enabled = -1;
314 
315 /*
316  * The worst case is single-sector max-parity RAID-Z blocks, in which
317  * case the space requirement is exactly (VDEV_RAIDZ_MAXPARITY + 1)
318  * times the size; so just assume that.  Add to this the fact that
319  * we can have up to 3 DVAs per bp, and one more factor of 2 because
320  * the block may be dittoed with up to 3 DVAs by ddt_sync().  All together,
321  * the worst case is:
322  *     (VDEV_RAIDZ_MAXPARITY + 1) * SPA_DVAS_PER_BP * 2 == 24
323  */
324 int spa_asize_inflation = 24;
325 
326 /*
327  * Normally, we don't allow the last 3.2% (1/(2^spa_slop_shift)) of space in
328  * the pool to be consumed.  This ensures that we don't run the pool
329  * completely out of space, due to unaccounted changes (e.g. to the MOS).
330  * It also limits the worst-case time to allocate space.  If we have
331  * less than this amount of free space, most ZPL operations (e.g. write,
332  * create) will return ENOSPC.
333  *
334  * Certain operations (e.g. file removal, most administrative actions) can
335  * use half the slop space.  They will only return ENOSPC if less than half
336  * the slop space is free.  Typically, once the pool has less than the slop
337  * space free, the user will use these operations to free up space in the pool.
338  * These are the operations that call dsl_pool_adjustedsize() with the netfree
339  * argument set to TRUE.
340  *
341  * Operations that are almost guaranteed to free up space in the absence of
342  * a pool checkpoint can use up to three quarters of the slop space
343  * (e.g zfs destroy).
344  *
345  * A very restricted set of operations are always permitted, regardless of
346  * the amount of free space.  These are the operations that call
347  * dsl_sync_task(ZFS_SPACE_CHECK_NONE). If these operations result in a net
348  * increase in the amount of space used, it is possible to run the pool
349  * completely out of space, causing it to be permanently read-only.
350  *
351  * Note that on very small pools, the slop space will be larger than
352  * 3.2%, in an effort to have it be at least spa_min_slop (128MB),
353  * but we never allow it to be more than half the pool size.
354  *
355  * See also the comments in zfs_space_check_t.
356  */
357 int spa_slop_shift = 5;
358 uint64_t spa_min_slop = 128 * 1024 * 1024;
359 
360 int spa_allocators = 4;
361 
362 /*PRINTFLIKE2*/
363 void
364 spa_load_failed(spa_t *spa, const char *fmt, ...)
365 {
366 	va_list adx;
367 	char buf[256];
368 
369 	va_start(adx, fmt);
370 	(void) vsnprintf(buf, sizeof (buf), fmt, adx);
371 	va_end(adx);
372 
373 	zfs_dbgmsg("spa_load(%s, config %s): FAILED: %s", spa->spa_name,
374 	    spa->spa_trust_config ? "trusted" : "untrusted", buf);
375 }
376 
377 /*PRINTFLIKE2*/
378 void
379 spa_load_note(spa_t *spa, const char *fmt, ...)
380 {
381 	va_list adx;
382 	char buf[256];
383 
384 	va_start(adx, fmt);
385 	(void) vsnprintf(buf, sizeof (buf), fmt, adx);
386 	va_end(adx);
387 
388 	zfs_dbgmsg("spa_load(%s, config %s): %s", spa->spa_name,
389 	    spa->spa_trust_config ? "trusted" : "untrusted", buf);
390 }
391 
392 /*
393  * ==========================================================================
394  * SPA config locking
395  * ==========================================================================
396  */
397 static void
398 spa_config_lock_init(spa_t *spa)
399 {
400 	for (int i = 0; i < SCL_LOCKS; i++) {
401 		spa_config_lock_t *scl = &spa->spa_config_lock[i];
402 		mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
403 		cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
404 		refcount_create_untracked(&scl->scl_count);
405 		scl->scl_writer = NULL;
406 		scl->scl_write_wanted = 0;
407 	}
408 }
409 
410 static void
411 spa_config_lock_destroy(spa_t *spa)
412 {
413 	for (int i = 0; i < SCL_LOCKS; i++) {
414 		spa_config_lock_t *scl = &spa->spa_config_lock[i];
415 		mutex_destroy(&scl->scl_lock);
416 		cv_destroy(&scl->scl_cv);
417 		refcount_destroy(&scl->scl_count);
418 		ASSERT(scl->scl_writer == NULL);
419 		ASSERT(scl->scl_write_wanted == 0);
420 	}
421 }
422 
423 int
424 spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
425 {
426 	for (int i = 0; i < SCL_LOCKS; i++) {
427 		spa_config_lock_t *scl = &spa->spa_config_lock[i];
428 		if (!(locks & (1 << i)))
429 			continue;
430 		mutex_enter(&scl->scl_lock);
431 		if (rw == RW_READER) {
432 			if (scl->scl_writer || scl->scl_write_wanted) {
433 				mutex_exit(&scl->scl_lock);
434 				spa_config_exit(spa, locks & ((1 << i) - 1),
435 				    tag);
436 				return (0);
437 			}
438 		} else {
439 			ASSERT(scl->scl_writer != curthread);
440 			if (!refcount_is_zero(&scl->scl_count)) {
441 				mutex_exit(&scl->scl_lock);
442 				spa_config_exit(spa, locks & ((1 << i) - 1),
443 				    tag);
444 				return (0);
445 			}
446 			scl->scl_writer = curthread;
447 		}
448 		(void) refcount_add(&scl->scl_count, tag);
449 		mutex_exit(&scl->scl_lock);
450 	}
451 	return (1);
452 }
453 
454 void
455 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
456 {
457 	int wlocks_held = 0;
458 
459 	ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
460 
461 	for (int i = 0; i < SCL_LOCKS; i++) {
462 		spa_config_lock_t *scl = &spa->spa_config_lock[i];
463 		if (scl->scl_writer == curthread)
464 			wlocks_held |= (1 << i);
465 		if (!(locks & (1 << i)))
466 			continue;
467 		mutex_enter(&scl->scl_lock);
468 		if (rw == RW_READER) {
469 			while (scl->scl_writer || scl->scl_write_wanted) {
470 				cv_wait(&scl->scl_cv, &scl->scl_lock);
471 			}
472 		} else {
473 			ASSERT(scl->scl_writer != curthread);
474 			while (!refcount_is_zero(&scl->scl_count)) {
475 				scl->scl_write_wanted++;
476 				cv_wait(&scl->scl_cv, &scl->scl_lock);
477 				scl->scl_write_wanted--;
478 			}
479 			scl->scl_writer = curthread;
480 		}
481 		(void) refcount_add(&scl->scl_count, tag);
482 		mutex_exit(&scl->scl_lock);
483 	}
484 	ASSERT3U(wlocks_held, <=, locks);
485 }
486 
487 void
488 spa_config_exit(spa_t *spa, int locks, void *tag)
489 {
490 	for (int i = SCL_LOCKS - 1; i >= 0; i--) {
491 		spa_config_lock_t *scl = &spa->spa_config_lock[i];
492 		if (!(locks & (1 << i)))
493 			continue;
494 		mutex_enter(&scl->scl_lock);
495 		ASSERT(!refcount_is_zero(&scl->scl_count));
496 		if (refcount_remove(&scl->scl_count, tag) == 0) {
497 			ASSERT(scl->scl_writer == NULL ||
498 			    scl->scl_writer == curthread);
499 			scl->scl_writer = NULL;	/* OK in either case */
500 			cv_broadcast(&scl->scl_cv);
501 		}
502 		mutex_exit(&scl->scl_lock);
503 	}
504 }
505 
506 int
507 spa_config_held(spa_t *spa, int locks, krw_t rw)
508 {
509 	int locks_held = 0;
510 
511 	for (int i = 0; i < SCL_LOCKS; i++) {
512 		spa_config_lock_t *scl = &spa->spa_config_lock[i];
513 		if (!(locks & (1 << i)))
514 			continue;
515 		if ((rw == RW_READER && !refcount_is_zero(&scl->scl_count)) ||
516 		    (rw == RW_WRITER && scl->scl_writer == curthread))
517 			locks_held |= 1 << i;
518 	}
519 
520 	return (locks_held);
521 }
522 
523 /*
524  * ==========================================================================
525  * SPA namespace functions
526  * ==========================================================================
527  */
528 
529 /*
530  * Lookup the named spa_t in the AVL tree.  The spa_namespace_lock must be held.
531  * Returns NULL if no matching spa_t is found.
532  */
533 spa_t *
534 spa_lookup(const char *name)
535 {
536 	static spa_t search;	/* spa_t is large; don't allocate on stack */
537 	spa_t *spa;
538 	avl_index_t where;
539 	char *cp;
540 
541 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
542 
543 	(void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
544 
545 	/*
546 	 * If it's a full dataset name, figure out the pool name and
547 	 * just use that.
548 	 */
549 	cp = strpbrk(search.spa_name, "/@#");
550 	if (cp != NULL)
551 		*cp = '\0';
552 
553 	spa = avl_find(&spa_namespace_avl, &search, &where);
554 
555 	return (spa);
556 }
557 
558 /*
559  * Fires when spa_sync has not completed within zfs_deadman_synctime_ms.
560  * If the zfs_deadman_enabled flag is set then it inspects all vdev queues
561  * looking for potentially hung I/Os.
562  */
563 void
564 spa_deadman(void *arg)
565 {
566 	spa_t *spa = arg;
567 
568 	/*
569 	 * Disable the deadman timer if the pool is suspended.
570 	 */
571 	if (spa_suspended(spa)) {
572 		VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
573 		return;
574 	}
575 
576 	zfs_dbgmsg("slow spa_sync: started %llu seconds ago, calls %llu",
577 	    (gethrtime() - spa->spa_sync_starttime) / NANOSEC,
578 	    ++spa->spa_deadman_calls);
579 	if (zfs_deadman_enabled)
580 		vdev_deadman(spa->spa_root_vdev);
581 }
582 
583 /*
584  * Create an uninitialized spa_t with the given name.  Requires
585  * spa_namespace_lock.  The caller must ensure that the spa_t doesn't already
586  * exist by calling spa_lookup() first.
587  */
588 spa_t *
589 spa_add(const char *name, nvlist_t *config, const char *altroot)
590 {
591 	spa_t *spa;
592 	spa_config_dirent_t *dp;
593 	cyc_handler_t hdlr;
594 	cyc_time_t when;
595 
596 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
597 
598 	spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
599 
600 	mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
601 	mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
602 	mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
603 	mutex_init(&spa->spa_evicting_os_lock, NULL, MUTEX_DEFAULT, NULL);
604 	mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
605 	mutex_init(&spa->spa_proc_lock, NULL, MUTEX_DEFAULT, NULL);
606 	mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
607 	mutex_init(&spa->spa_cksum_tmpls_lock, NULL, MUTEX_DEFAULT, NULL);
608 	mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
609 	mutex_init(&spa->spa_suspend_lock, NULL, MUTEX_DEFAULT, NULL);
610 	mutex_init(&spa->spa_vdev_top_lock, NULL, MUTEX_DEFAULT, NULL);
611 	mutex_init(&spa->spa_iokstat_lock, NULL, MUTEX_DEFAULT, NULL);
612 
613 	cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
614 	cv_init(&spa->spa_evicting_os_cv, NULL, CV_DEFAULT, NULL);
615 	cv_init(&spa->spa_proc_cv, NULL, CV_DEFAULT, NULL);
616 	cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
617 	cv_init(&spa->spa_suspend_cv, NULL, CV_DEFAULT, NULL);
618 
619 	for (int t = 0; t < TXG_SIZE; t++)
620 		bplist_create(&spa->spa_free_bplist[t]);
621 
622 	(void) strlcpy(spa->spa_name, name, sizeof (spa->spa_name));
623 	spa->spa_state = POOL_STATE_UNINITIALIZED;
624 	spa->spa_freeze_txg = UINT64_MAX;
625 	spa->spa_final_txg = UINT64_MAX;
626 	spa->spa_load_max_txg = UINT64_MAX;
627 	spa->spa_proc = &p0;
628 	spa->spa_proc_state = SPA_PROC_NONE;
629 	spa->spa_trust_config = B_TRUE;
630 
631 	hdlr.cyh_func = spa_deadman;
632 	hdlr.cyh_arg = spa;
633 	hdlr.cyh_level = CY_LOW_LEVEL;
634 
635 	spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime_ms);
636 
637 	/*
638 	 * This determines how often we need to check for hung I/Os after
639 	 * the cyclic has already fired. Since checking for hung I/Os is
640 	 * an expensive operation we don't want to check too frequently.
641 	 * Instead wait for 5 seconds before checking again.
642 	 */
643 	when.cyt_interval = MSEC2NSEC(zfs_deadman_checktime_ms);
644 	when.cyt_when = CY_INFINITY;
645 	mutex_enter(&cpu_lock);
646 	spa->spa_deadman_cycid = cyclic_add(&hdlr, &when);
647 	mutex_exit(&cpu_lock);
648 
649 	refcount_create(&spa->spa_refcount);
650 	spa_config_lock_init(spa);
651 
652 	avl_add(&spa_namespace_avl, spa);
653 
654 	/*
655 	 * Set the alternate root, if there is one.
656 	 */
657 	if (altroot) {
658 		spa->spa_root = spa_strdup(altroot);
659 		spa_active_count++;
660 	}
661 
662 	spa->spa_alloc_count = spa_allocators;
663 	spa->spa_alloc_locks = kmem_zalloc(spa->spa_alloc_count *
664 	    sizeof (kmutex_t), KM_SLEEP);
665 	spa->spa_alloc_trees = kmem_zalloc(spa->spa_alloc_count *
666 	    sizeof (avl_tree_t), KM_SLEEP);
667 	for (int i = 0; i < spa->spa_alloc_count; i++) {
668 		mutex_init(&spa->spa_alloc_locks[i], NULL, MUTEX_DEFAULT, NULL);
669 		avl_create(&spa->spa_alloc_trees[i], zio_bookmark_compare,
670 		    sizeof (zio_t), offsetof(zio_t, io_alloc_node));
671 	}
672 
673 	/*
674 	 * Every pool starts with the default cachefile
675 	 */
676 	list_create(&spa->spa_config_list, sizeof (spa_config_dirent_t),
677 	    offsetof(spa_config_dirent_t, scd_link));
678 
679 	dp = kmem_zalloc(sizeof (spa_config_dirent_t), KM_SLEEP);
680 	dp->scd_path = altroot ? NULL : spa_strdup(spa_config_path);
681 	list_insert_head(&spa->spa_config_list, dp);
682 
683 	VERIFY(nvlist_alloc(&spa->spa_load_info, NV_UNIQUE_NAME,
684 	    KM_SLEEP) == 0);
685 
686 	if (config != NULL) {
687 		nvlist_t *features;
688 
689 		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
690 		    &features) == 0) {
691 			VERIFY(nvlist_dup(features, &spa->spa_label_features,
692 			    0) == 0);
693 		}
694 
695 		VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
696 	}
697 
698 	if (spa->spa_label_features == NULL) {
699 		VERIFY(nvlist_alloc(&spa->spa_label_features, NV_UNIQUE_NAME,
700 		    KM_SLEEP) == 0);
701 	}
702 
703 	spa->spa_iokstat = kstat_create("zfs", 0, name,
704 	    "disk", KSTAT_TYPE_IO, 1, 0);
705 	if (spa->spa_iokstat) {
706 		spa->spa_iokstat->ks_lock = &spa->spa_iokstat_lock;
707 		kstat_install(spa->spa_iokstat);
708 	}
709 
710 	spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0);
711 
712 	spa->spa_min_ashift = INT_MAX;
713 	spa->spa_max_ashift = 0;
714 
715 	/*
716 	 * As a pool is being created, treat all features as disabled by
717 	 * setting SPA_FEATURE_DISABLED for all entries in the feature
718 	 * refcount cache.
719 	 */
720 	for (int i = 0; i < SPA_FEATURES; i++) {
721 		spa->spa_feat_refcount_cache[i] = SPA_FEATURE_DISABLED;
722 	}
723 
724 	return (spa);
725 }
726 
727 /*
728  * Removes a spa_t from the namespace, freeing up any memory used.  Requires
729  * spa_namespace_lock.  This is called only after the spa_t has been closed and
730  * deactivated.
731  */
732 void
733 spa_remove(spa_t *spa)
734 {
735 	spa_config_dirent_t *dp;
736 
737 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
738 	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
739 	ASSERT3U(refcount_count(&spa->spa_refcount), ==, 0);
740 
741 	nvlist_free(spa->spa_config_splitting);
742 
743 	avl_remove(&spa_namespace_avl, spa);
744 	cv_broadcast(&spa_namespace_cv);
745 
746 	if (spa->spa_root) {
747 		spa_strfree(spa->spa_root);
748 		spa_active_count--;
749 	}
750 
751 	while ((dp = list_head(&spa->spa_config_list)) != NULL) {
752 		list_remove(&spa->spa_config_list, dp);
753 		if (dp->scd_path != NULL)
754 			spa_strfree(dp->scd_path);
755 		kmem_free(dp, sizeof (spa_config_dirent_t));
756 	}
757 
758 	for (int i = 0; i < spa->spa_alloc_count; i++) {
759 		avl_destroy(&spa->spa_alloc_trees[i]);
760 		mutex_destroy(&spa->spa_alloc_locks[i]);
761 	}
762 	kmem_free(spa->spa_alloc_locks, spa->spa_alloc_count *
763 	    sizeof (kmutex_t));
764 	kmem_free(spa->spa_alloc_trees, spa->spa_alloc_count *
765 	    sizeof (avl_tree_t));
766 
767 	list_destroy(&spa->spa_config_list);
768 
769 	nvlist_free(spa->spa_label_features);
770 	nvlist_free(spa->spa_load_info);
771 	spa_config_set(spa, NULL);
772 
773 	mutex_enter(&cpu_lock);
774 	if (spa->spa_deadman_cycid != CYCLIC_NONE)
775 		cyclic_remove(spa->spa_deadman_cycid);
776 	mutex_exit(&cpu_lock);
777 	spa->spa_deadman_cycid = CYCLIC_NONE;
778 
779 	refcount_destroy(&spa->spa_refcount);
780 
781 	spa_config_lock_destroy(spa);
782 
783 	kstat_delete(spa->spa_iokstat);
784 	spa->spa_iokstat = NULL;
785 
786 	for (int t = 0; t < TXG_SIZE; t++)
787 		bplist_destroy(&spa->spa_free_bplist[t]);
788 
789 	zio_checksum_templates_free(spa);
790 
791 	cv_destroy(&spa->spa_async_cv);
792 	cv_destroy(&spa->spa_evicting_os_cv);
793 	cv_destroy(&spa->spa_proc_cv);
794 	cv_destroy(&spa->spa_scrub_io_cv);
795 	cv_destroy(&spa->spa_suspend_cv);
796 
797 	mutex_destroy(&spa->spa_async_lock);
798 	mutex_destroy(&spa->spa_errlist_lock);
799 	mutex_destroy(&spa->spa_errlog_lock);
800 	mutex_destroy(&spa->spa_evicting_os_lock);
801 	mutex_destroy(&spa->spa_history_lock);
802 	mutex_destroy(&spa->spa_proc_lock);
803 	mutex_destroy(&spa->spa_props_lock);
804 	mutex_destroy(&spa->spa_cksum_tmpls_lock);
805 	mutex_destroy(&spa->spa_scrub_lock);
806 	mutex_destroy(&spa->spa_suspend_lock);
807 	mutex_destroy(&spa->spa_vdev_top_lock);
808 	mutex_destroy(&spa->spa_iokstat_lock);
809 
810 	kmem_free(spa, sizeof (spa_t));
811 }
812 
813 /*
814  * Given a pool, return the next pool in the namespace, or NULL if there is
815  * none.  If 'prev' is NULL, return the first pool.
816  */
817 spa_t *
818 spa_next(spa_t *prev)
819 {
820 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
821 
822 	if (prev)
823 		return (AVL_NEXT(&spa_namespace_avl, prev));
824 	else
825 		return (avl_first(&spa_namespace_avl));
826 }
827 
828 /*
829  * ==========================================================================
830  * SPA refcount functions
831  * ==========================================================================
832  */
833 
834 /*
835  * Add a reference to the given spa_t.  Must have at least one reference, or
836  * have the namespace lock held.
837  */
838 void
839 spa_open_ref(spa_t *spa, void *tag)
840 {
841 	ASSERT(refcount_count(&spa->spa_refcount) >= spa->spa_minref ||
842 	    MUTEX_HELD(&spa_namespace_lock));
843 	(void) refcount_add(&spa->spa_refcount, tag);
844 }
845 
846 /*
847  * Remove a reference to the given spa_t.  Must have at least one reference, or
848  * have the namespace lock held.
849  */
850 void
851 spa_close(spa_t *spa, void *tag)
852 {
853 	ASSERT(refcount_count(&spa->spa_refcount) > spa->spa_minref ||
854 	    MUTEX_HELD(&spa_namespace_lock));
855 	(void) refcount_remove(&spa->spa_refcount, tag);
856 }
857 
858 /*
859  * Remove a reference to the given spa_t held by a dsl dir that is
860  * being asynchronously released.  Async releases occur from a taskq
861  * performing eviction of dsl datasets and dirs.  The namespace lock
862  * isn't held and the hold by the object being evicted may contribute to
863  * spa_minref (e.g. dataset or directory released during pool export),
864  * so the asserts in spa_close() do not apply.
865  */
866 void
867 spa_async_close(spa_t *spa, void *tag)
868 {
869 	(void) refcount_remove(&spa->spa_refcount, tag);
870 }
871 
872 /*
873  * Check to see if the spa refcount is zero.  Must be called with
874  * spa_namespace_lock held.  We really compare against spa_minref, which is the
875  * number of references acquired when opening a pool
876  */
877 boolean_t
878 spa_refcount_zero(spa_t *spa)
879 {
880 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
881 
882 	return (refcount_count(&spa->spa_refcount) == spa->spa_minref);
883 }
884 
885 /*
886  * ==========================================================================
887  * SPA spare and l2cache tracking
888  * ==========================================================================
889  */
890 
891 /*
892  * Hot spares and cache devices are tracked using the same code below,
893  * for 'auxiliary' devices.
894  */
895 
896 typedef struct spa_aux {
897 	uint64_t	aux_guid;
898 	uint64_t	aux_pool;
899 	avl_node_t	aux_avl;
900 	int		aux_count;
901 } spa_aux_t;
902 
903 static int
904 spa_aux_compare(const void *a, const void *b)
905 {
906 	const spa_aux_t *sa = a;
907 	const spa_aux_t *sb = b;
908 
909 	if (sa->aux_guid < sb->aux_guid)
910 		return (-1);
911 	else if (sa->aux_guid > sb->aux_guid)
912 		return (1);
913 	else
914 		return (0);
915 }
916 
917 void
918 spa_aux_add(vdev_t *vd, avl_tree_t *avl)
919 {
920 	avl_index_t where;
921 	spa_aux_t search;
922 	spa_aux_t *aux;
923 
924 	search.aux_guid = vd->vdev_guid;
925 	if ((aux = avl_find(avl, &search, &where)) != NULL) {
926 		aux->aux_count++;
927 	} else {
928 		aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
929 		aux->aux_guid = vd->vdev_guid;
930 		aux->aux_count = 1;
931 		avl_insert(avl, aux, where);
932 	}
933 }
934 
935 void
936 spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
937 {
938 	spa_aux_t search;
939 	spa_aux_t *aux;
940 	avl_index_t where;
941 
942 	search.aux_guid = vd->vdev_guid;
943 	aux = avl_find(avl, &search, &where);
944 
945 	ASSERT(aux != NULL);
946 
947 	if (--aux->aux_count == 0) {
948 		avl_remove(avl, aux);
949 		kmem_free(aux, sizeof (spa_aux_t));
950 	} else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
951 		aux->aux_pool = 0ULL;
952 	}
953 }
954 
955 boolean_t
956 spa_aux_exists(uint64_t guid, uint64_t *pool, int *refcnt, avl_tree_t *avl)
957 {
958 	spa_aux_t search, *found;
959 
960 	search.aux_guid = guid;
961 	found = avl_find(avl, &search, NULL);
962 
963 	if (pool) {
964 		if (found)
965 			*pool = found->aux_pool;
966 		else
967 			*pool = 0ULL;
968 	}
969 
970 	if (refcnt) {
971 		if (found)
972 			*refcnt = found->aux_count;
973 		else
974 			*refcnt = 0;
975 	}
976 
977 	return (found != NULL);
978 }
979 
980 void
981 spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
982 {
983 	spa_aux_t search, *found;
984 	avl_index_t where;
985 
986 	search.aux_guid = vd->vdev_guid;
987 	found = avl_find(avl, &search, &where);
988 	ASSERT(found != NULL);
989 	ASSERT(found->aux_pool == 0ULL);
990 
991 	found->aux_pool = spa_guid(vd->vdev_spa);
992 }
993 
994 /*
995  * Spares are tracked globally due to the following constraints:
996  *
997  * 	- A spare may be part of multiple pools.
998  * 	- A spare may be added to a pool even if it's actively in use within
999  *	  another pool.
1000  * 	- A spare in use in any pool can only be the source of a replacement if
1001  *	  the target is a spare in the same pool.
1002  *
1003  * We keep track of all spares on the system through the use of a reference
1004  * counted AVL tree.  When a vdev is added as a spare, or used as a replacement
1005  * spare, then we bump the reference count in the AVL tree.  In addition, we set
1006  * the 'vdev_isspare' member to indicate that the device is a spare (active or
1007  * inactive).  When a spare is made active (used to replace a device in the
1008  * pool), we also keep track of which pool its been made a part of.
1009  *
1010  * The 'spa_spare_lock' protects the AVL tree.  These functions are normally
1011  * called under the spa_namespace lock as part of vdev reconfiguration.  The
1012  * separate spare lock exists for the status query path, which does not need to
1013  * be completely consistent with respect to other vdev configuration changes.
1014  */
1015 
1016 static int
1017 spa_spare_compare(const void *a, const void *b)
1018 {
1019 	return (spa_aux_compare(a, b));
1020 }
1021 
1022 void
1023 spa_spare_add(vdev_t *vd)
1024 {
1025 	mutex_enter(&spa_spare_lock);
1026 	ASSERT(!vd->vdev_isspare);
1027 	spa_aux_add(vd, &spa_spare_avl);
1028 	vd->vdev_isspare = B_TRUE;
1029 	mutex_exit(&spa_spare_lock);
1030 }
1031 
1032 void
1033 spa_spare_remove(vdev_t *vd)
1034 {
1035 	mutex_enter(&spa_spare_lock);
1036 	ASSERT(vd->vdev_isspare);
1037 	spa_aux_remove(vd, &spa_spare_avl);
1038 	vd->vdev_isspare = B_FALSE;
1039 	mutex_exit(&spa_spare_lock);
1040 }
1041 
1042 boolean_t
1043 spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt)
1044 {
1045 	boolean_t found;
1046 
1047 	mutex_enter(&spa_spare_lock);
1048 	found = spa_aux_exists(guid, pool, refcnt, &spa_spare_avl);
1049 	mutex_exit(&spa_spare_lock);
1050 
1051 	return (found);
1052 }
1053 
1054 void
1055 spa_spare_activate(vdev_t *vd)
1056 {
1057 	mutex_enter(&spa_spare_lock);
1058 	ASSERT(vd->vdev_isspare);
1059 	spa_aux_activate(vd, &spa_spare_avl);
1060 	mutex_exit(&spa_spare_lock);
1061 }
1062 
1063 /*
1064  * Level 2 ARC devices are tracked globally for the same reasons as spares.
1065  * Cache devices currently only support one pool per cache device, and so
1066  * for these devices the aux reference count is currently unused beyond 1.
1067  */
1068 
1069 static int
1070 spa_l2cache_compare(const void *a, const void *b)
1071 {
1072 	return (spa_aux_compare(a, b));
1073 }
1074 
1075 void
1076 spa_l2cache_add(vdev_t *vd)
1077 {
1078 	mutex_enter(&spa_l2cache_lock);
1079 	ASSERT(!vd->vdev_isl2cache);
1080 	spa_aux_add(vd, &spa_l2cache_avl);
1081 	vd->vdev_isl2cache = B_TRUE;
1082 	mutex_exit(&spa_l2cache_lock);
1083 }
1084 
1085 void
1086 spa_l2cache_remove(vdev_t *vd)
1087 {
1088 	mutex_enter(&spa_l2cache_lock);
1089 	ASSERT(vd->vdev_isl2cache);
1090 	spa_aux_remove(vd, &spa_l2cache_avl);
1091 	vd->vdev_isl2cache = B_FALSE;
1092 	mutex_exit(&spa_l2cache_lock);
1093 }
1094 
1095 boolean_t
1096 spa_l2cache_exists(uint64_t guid, uint64_t *pool)
1097 {
1098 	boolean_t found;
1099 
1100 	mutex_enter(&spa_l2cache_lock);
1101 	found = spa_aux_exists(guid, pool, NULL, &spa_l2cache_avl);
1102 	mutex_exit(&spa_l2cache_lock);
1103 
1104 	return (found);
1105 }
1106 
1107 void
1108 spa_l2cache_activate(vdev_t *vd)
1109 {
1110 	mutex_enter(&spa_l2cache_lock);
1111 	ASSERT(vd->vdev_isl2cache);
1112 	spa_aux_activate(vd, &spa_l2cache_avl);
1113 	mutex_exit(&spa_l2cache_lock);
1114 }
1115 
1116 /*
1117  * ==========================================================================
1118  * SPA vdev locking
1119  * ==========================================================================
1120  */
1121 
1122 /*
1123  * Lock the given spa_t for the purpose of adding or removing a vdev.
1124  * Grabs the global spa_namespace_lock plus the spa config lock for writing.
1125  * It returns the next transaction group for the spa_t.
1126  */
1127 uint64_t
1128 spa_vdev_enter(spa_t *spa)
1129 {
1130 	mutex_enter(&spa->spa_vdev_top_lock);
1131 	mutex_enter(&spa_namespace_lock);
1132 	return (spa_vdev_config_enter(spa));
1133 }
1134 
1135 /*
1136  * Internal implementation for spa_vdev_enter().  Used when a vdev
1137  * operation requires multiple syncs (i.e. removing a device) while
1138  * keeping the spa_namespace_lock held.
1139  */
1140 uint64_t
1141 spa_vdev_config_enter(spa_t *spa)
1142 {
1143 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1144 
1145 	spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1146 
1147 	return (spa_last_synced_txg(spa) + 1);
1148 }
1149 
1150 /*
1151  * Used in combination with spa_vdev_config_enter() to allow the syncing
1152  * of multiple transactions without releasing the spa_namespace_lock.
1153  */
1154 void
1155 spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
1156 {
1157 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1158 
1159 	int config_changed = B_FALSE;
1160 
1161 	ASSERT(txg > spa_last_synced_txg(spa));
1162 
1163 	spa->spa_pending_vdev = NULL;
1164 
1165 	/*
1166 	 * Reassess the DTLs.
1167 	 */
1168 	vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
1169 
1170 	if (error == 0 && !list_is_empty(&spa->spa_config_dirty_list)) {
1171 		config_changed = B_TRUE;
1172 		spa->spa_config_generation++;
1173 	}
1174 
1175 	/*
1176 	 * Verify the metaslab classes.
1177 	 */
1178 	ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
1179 	ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
1180 
1181 	spa_config_exit(spa, SCL_ALL, spa);
1182 
1183 	/*
1184 	 * Panic the system if the specified tag requires it.  This
1185 	 * is useful for ensuring that configurations are updated
1186 	 * transactionally.
1187 	 */
1188 	if (zio_injection_enabled)
1189 		zio_handle_panic_injection(spa, tag, 0);
1190 
1191 	/*
1192 	 * Note: this txg_wait_synced() is important because it ensures
1193 	 * that there won't be more than one config change per txg.
1194 	 * This allows us to use the txg as the generation number.
1195 	 */
1196 	if (error == 0)
1197 		txg_wait_synced(spa->spa_dsl_pool, txg);
1198 
1199 	if (vd != NULL) {
1200 		ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
1201 		spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1202 		vdev_free(vd);
1203 		spa_config_exit(spa, SCL_ALL, spa);
1204 	}
1205 
1206 	/*
1207 	 * If the config changed, update the config cache.
1208 	 */
1209 	if (config_changed)
1210 		spa_write_cachefile(spa, B_FALSE, B_TRUE);
1211 }
1212 
1213 /*
1214  * Unlock the spa_t after adding or removing a vdev.  Besides undoing the
1215  * locking of spa_vdev_enter(), we also want make sure the transactions have
1216  * synced to disk, and then update the global configuration cache with the new
1217  * information.
1218  */
1219 int
1220 spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
1221 {
1222 	spa_vdev_config_exit(spa, vd, txg, error, FTAG);
1223 	mutex_exit(&spa_namespace_lock);
1224 	mutex_exit(&spa->spa_vdev_top_lock);
1225 
1226 	return (error);
1227 }
1228 
1229 /*
1230  * Lock the given spa_t for the purpose of changing vdev state.
1231  */
1232 void
1233 spa_vdev_state_enter(spa_t *spa, int oplocks)
1234 {
1235 	int locks = SCL_STATE_ALL | oplocks;
1236 
1237 	/*
1238 	 * Root pools may need to read of the underlying devfs filesystem
1239 	 * when opening up a vdev.  Unfortunately if we're holding the
1240 	 * SCL_ZIO lock it will result in a deadlock when we try to issue
1241 	 * the read from the root filesystem.  Instead we "prefetch"
1242 	 * the associated vnodes that we need prior to opening the
1243 	 * underlying devices and cache them so that we can prevent
1244 	 * any I/O when we are doing the actual open.
1245 	 */
1246 	if (spa_is_root(spa)) {
1247 		int low = locks & ~(SCL_ZIO - 1);
1248 		int high = locks & ~low;
1249 
1250 		spa_config_enter(spa, high, spa, RW_WRITER);
1251 		vdev_hold(spa->spa_root_vdev);
1252 		spa_config_enter(spa, low, spa, RW_WRITER);
1253 	} else {
1254 		spa_config_enter(spa, locks, spa, RW_WRITER);
1255 	}
1256 	spa->spa_vdev_locks = locks;
1257 }
1258 
1259 int
1260 spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error)
1261 {
1262 	boolean_t config_changed = B_FALSE;
1263 
1264 	if (vd != NULL || error == 0)
1265 		vdev_dtl_reassess(vd ? vd->vdev_top : spa->spa_root_vdev,
1266 		    0, 0, B_FALSE);
1267 
1268 	if (vd != NULL) {
1269 		vdev_state_dirty(vd->vdev_top);
1270 		config_changed = B_TRUE;
1271 		spa->spa_config_generation++;
1272 	}
1273 
1274 	if (spa_is_root(spa))
1275 		vdev_rele(spa->spa_root_vdev);
1276 
1277 	ASSERT3U(spa->spa_vdev_locks, >=, SCL_STATE_ALL);
1278 	spa_config_exit(spa, spa->spa_vdev_locks, spa);
1279 
1280 	/*
1281 	 * If anything changed, wait for it to sync.  This ensures that,
1282 	 * from the system administrator's perspective, zpool(1M) commands
1283 	 * are synchronous.  This is important for things like zpool offline:
1284 	 * when the command completes, you expect no further I/O from ZFS.
1285 	 */
1286 	if (vd != NULL)
1287 		txg_wait_synced(spa->spa_dsl_pool, 0);
1288 
1289 	/*
1290 	 * If the config changed, update the config cache.
1291 	 */
1292 	if (config_changed) {
1293 		mutex_enter(&spa_namespace_lock);
1294 		spa_write_cachefile(spa, B_FALSE, B_TRUE);
1295 		mutex_exit(&spa_namespace_lock);
1296 	}
1297 
1298 	return (error);
1299 }
1300 
1301 /*
1302  * ==========================================================================
1303  * Miscellaneous functions
1304  * ==========================================================================
1305  */
1306 
1307 void
1308 spa_activate_mos_feature(spa_t *spa, const char *feature, dmu_tx_t *tx)
1309 {
1310 	if (!nvlist_exists(spa->spa_label_features, feature)) {
1311 		fnvlist_add_boolean(spa->spa_label_features, feature);
1312 		/*
1313 		 * When we are creating the pool (tx_txg==TXG_INITIAL), we can't
1314 		 * dirty the vdev config because lock SCL_CONFIG is not held.
1315 		 * Thankfully, in this case we don't need to dirty the config
1316 		 * because it will be written out anyway when we finish
1317 		 * creating the pool.
1318 		 */
1319 		if (tx->tx_txg != TXG_INITIAL)
1320 			vdev_config_dirty(spa->spa_root_vdev);
1321 	}
1322 }
1323 
1324 void
1325 spa_deactivate_mos_feature(spa_t *spa, const char *feature)
1326 {
1327 	if (nvlist_remove_all(spa->spa_label_features, feature) == 0)
1328 		vdev_config_dirty(spa->spa_root_vdev);
1329 }
1330 
1331 /*
1332  * Rename a spa_t.
1333  */
1334 int
1335 spa_rename(const char *name, const char *newname)
1336 {
1337 	spa_t *spa;
1338 	int err;
1339 
1340 	/*
1341 	 * Lookup the spa_t and grab the config lock for writing.  We need to
1342 	 * actually open the pool so that we can sync out the necessary labels.
1343 	 * It's OK to call spa_open() with the namespace lock held because we
1344 	 * allow recursive calls for other reasons.
1345 	 */
1346 	mutex_enter(&spa_namespace_lock);
1347 	if ((err = spa_open(name, &spa, FTAG)) != 0) {
1348 		mutex_exit(&spa_namespace_lock);
1349 		return (err);
1350 	}
1351 
1352 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1353 
1354 	avl_remove(&spa_namespace_avl, spa);
1355 	(void) strlcpy(spa->spa_name, newname, sizeof (spa->spa_name));
1356 	avl_add(&spa_namespace_avl, spa);
1357 
1358 	/*
1359 	 * Sync all labels to disk with the new names by marking the root vdev
1360 	 * dirty and waiting for it to sync.  It will pick up the new pool name
1361 	 * during the sync.
1362 	 */
1363 	vdev_config_dirty(spa->spa_root_vdev);
1364 
1365 	spa_config_exit(spa, SCL_ALL, FTAG);
1366 
1367 	txg_wait_synced(spa->spa_dsl_pool, 0);
1368 
1369 	/*
1370 	 * Sync the updated config cache.
1371 	 */
1372 	spa_write_cachefile(spa, B_FALSE, B_TRUE);
1373 
1374 	spa_close(spa, FTAG);
1375 
1376 	mutex_exit(&spa_namespace_lock);
1377 
1378 	return (0);
1379 }
1380 
1381 /*
1382  * Return the spa_t associated with given pool_guid, if it exists.  If
1383  * device_guid is non-zero, determine whether the pool exists *and* contains
1384  * a device with the specified device_guid.
1385  */
1386 spa_t *
1387 spa_by_guid(uint64_t pool_guid, uint64_t device_guid)
1388 {
1389 	spa_t *spa;
1390 	avl_tree_t *t = &spa_namespace_avl;
1391 
1392 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1393 
1394 	for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
1395 		if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1396 			continue;
1397 		if (spa->spa_root_vdev == NULL)
1398 			continue;
1399 		if (spa_guid(spa) == pool_guid) {
1400 			if (device_guid == 0)
1401 				break;
1402 
1403 			if (vdev_lookup_by_guid(spa->spa_root_vdev,
1404 			    device_guid) != NULL)
1405 				break;
1406 
1407 			/*
1408 			 * Check any devices we may be in the process of adding.
1409 			 */
1410 			if (spa->spa_pending_vdev) {
1411 				if (vdev_lookup_by_guid(spa->spa_pending_vdev,
1412 				    device_guid) != NULL)
1413 					break;
1414 			}
1415 		}
1416 	}
1417 
1418 	return (spa);
1419 }
1420 
1421 /*
1422  * Determine whether a pool with the given pool_guid exists.
1423  */
1424 boolean_t
1425 spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
1426 {
1427 	return (spa_by_guid(pool_guid, device_guid) != NULL);
1428 }
1429 
1430 char *
1431 spa_strdup(const char *s)
1432 {
1433 	size_t len;
1434 	char *new;
1435 
1436 	len = strlen(s);
1437 	new = kmem_alloc(len + 1, KM_SLEEP);
1438 	bcopy(s, new, len);
1439 	new[len] = '\0';
1440 
1441 	return (new);
1442 }
1443 
1444 void
1445 spa_strfree(char *s)
1446 {
1447 	kmem_free(s, strlen(s) + 1);
1448 }
1449 
1450 uint64_t
1451 spa_get_random(uint64_t range)
1452 {
1453 	uint64_t r;
1454 
1455 	ASSERT(range != 0);
1456 
1457 	(void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
1458 
1459 	return (r % range);
1460 }
1461 
1462 uint64_t
1463 spa_generate_guid(spa_t *spa)
1464 {
1465 	uint64_t guid = spa_get_random(-1ULL);
1466 
1467 	if (spa != NULL) {
1468 		while (guid == 0 || spa_guid_exists(spa_guid(spa), guid))
1469 			guid = spa_get_random(-1ULL);
1470 	} else {
1471 		while (guid == 0 || spa_guid_exists(guid, 0))
1472 			guid = spa_get_random(-1ULL);
1473 	}
1474 
1475 	return (guid);
1476 }
1477 
1478 void
1479 snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp)
1480 {
1481 	char type[256];
1482 	char *checksum = NULL;
1483 	char *compress = NULL;
1484 
1485 	if (bp != NULL) {
1486 		if (BP_GET_TYPE(bp) & DMU_OT_NEWTYPE) {
1487 			dmu_object_byteswap_t bswap =
1488 			    DMU_OT_BYTESWAP(BP_GET_TYPE(bp));
1489 			(void) snprintf(type, sizeof (type), "bswap %s %s",
1490 			    DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) ?
1491 			    "metadata" : "data",
1492 			    dmu_ot_byteswap[bswap].ob_name);
1493 		} else {
1494 			(void) strlcpy(type, dmu_ot[BP_GET_TYPE(bp)].ot_name,
1495 			    sizeof (type));
1496 		}
1497 		if (!BP_IS_EMBEDDED(bp)) {
1498 			checksum =
1499 			    zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name;
1500 		}
1501 		compress = zio_compress_table[BP_GET_COMPRESS(bp)].ci_name;
1502 	}
1503 
1504 	SNPRINTF_BLKPTR(snprintf, ' ', buf, buflen, bp, type, checksum,
1505 	    compress);
1506 }
1507 
1508 void
1509 spa_freeze(spa_t *spa)
1510 {
1511 	uint64_t freeze_txg = 0;
1512 
1513 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1514 	if (spa->spa_freeze_txg == UINT64_MAX) {
1515 		freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
1516 		spa->spa_freeze_txg = freeze_txg;
1517 	}
1518 	spa_config_exit(spa, SCL_ALL, FTAG);
1519 	if (freeze_txg != 0)
1520 		txg_wait_synced(spa_get_dsl(spa), freeze_txg);
1521 }
1522 
1523 void
1524 zfs_panic_recover(const char *fmt, ...)
1525 {
1526 	va_list adx;
1527 
1528 	va_start(adx, fmt);
1529 	vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
1530 	va_end(adx);
1531 }
1532 
1533 /*
1534  * This is a stripped-down version of strtoull, suitable only for converting
1535  * lowercase hexadecimal numbers that don't overflow.
1536  */
1537 uint64_t
1538 zfs_strtonum(const char *str, char **nptr)
1539 {
1540 	uint64_t val = 0;
1541 	char c;
1542 	int digit;
1543 
1544 	while ((c = *str) != '\0') {
1545 		if (c >= '0' && c <= '9')
1546 			digit = c - '0';
1547 		else if (c >= 'a' && c <= 'f')
1548 			digit = 10 + c - 'a';
1549 		else
1550 			break;
1551 
1552 		val *= 16;
1553 		val += digit;
1554 
1555 		str++;
1556 	}
1557 
1558 	if (nptr)
1559 		*nptr = (char *)str;
1560 
1561 	return (val);
1562 }
1563 
1564 /*
1565  * ==========================================================================
1566  * Accessor functions
1567  * ==========================================================================
1568  */
1569 
1570 boolean_t
1571 spa_shutting_down(spa_t *spa)
1572 {
1573 	return (spa->spa_async_suspended);
1574 }
1575 
1576 dsl_pool_t *
1577 spa_get_dsl(spa_t *spa)
1578 {
1579 	return (spa->spa_dsl_pool);
1580 }
1581 
1582 boolean_t
1583 spa_is_initializing(spa_t *spa)
1584 {
1585 	return (spa->spa_is_initializing);
1586 }
1587 
1588 boolean_t
1589 spa_indirect_vdevs_loaded(spa_t *spa)
1590 {
1591 	return (spa->spa_indirect_vdevs_loaded);
1592 }
1593 
1594 blkptr_t *
1595 spa_get_rootblkptr(spa_t *spa)
1596 {
1597 	return (&spa->spa_ubsync.ub_rootbp);
1598 }
1599 
1600 void
1601 spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1602 {
1603 	spa->spa_uberblock.ub_rootbp = *bp;
1604 }
1605 
1606 void
1607 spa_altroot(spa_t *spa, char *buf, size_t buflen)
1608 {
1609 	if (spa->spa_root == NULL)
1610 		buf[0] = '\0';
1611 	else
1612 		(void) strncpy(buf, spa->spa_root, buflen);
1613 }
1614 
1615 int
1616 spa_sync_pass(spa_t *spa)
1617 {
1618 	return (spa->spa_sync_pass);
1619 }
1620 
1621 char *
1622 spa_name(spa_t *spa)
1623 {
1624 	return (spa->spa_name);
1625 }
1626 
1627 uint64_t
1628 spa_guid(spa_t *spa)
1629 {
1630 	dsl_pool_t *dp = spa_get_dsl(spa);
1631 	uint64_t guid;
1632 
1633 	/*
1634 	 * If we fail to parse the config during spa_load(), we can go through
1635 	 * the error path (which posts an ereport) and end up here with no root
1636 	 * vdev.  We stash the original pool guid in 'spa_config_guid' to handle
1637 	 * this case.
1638 	 */
1639 	if (spa->spa_root_vdev == NULL)
1640 		return (spa->spa_config_guid);
1641 
1642 	guid = spa->spa_last_synced_guid != 0 ?
1643 	    spa->spa_last_synced_guid : spa->spa_root_vdev->vdev_guid;
1644 
1645 	/*
1646 	 * Return the most recently synced out guid unless we're
1647 	 * in syncing context.
1648 	 */
1649 	if (dp && dsl_pool_sync_context(dp))
1650 		return (spa->spa_root_vdev->vdev_guid);
1651 	else
1652 		return (guid);
1653 }
1654 
1655 uint64_t
1656 spa_load_guid(spa_t *spa)
1657 {
1658 	/*
1659 	 * This is a GUID that exists solely as a reference for the
1660 	 * purposes of the arc.  It is generated at load time, and
1661 	 * is never written to persistent storage.
1662 	 */
1663 	return (spa->spa_load_guid);
1664 }
1665 
1666 uint64_t
1667 spa_last_synced_txg(spa_t *spa)
1668 {
1669 	return (spa->spa_ubsync.ub_txg);
1670 }
1671 
1672 uint64_t
1673 spa_first_txg(spa_t *spa)
1674 {
1675 	return (spa->spa_first_txg);
1676 }
1677 
1678 uint64_t
1679 spa_syncing_txg(spa_t *spa)
1680 {
1681 	return (spa->spa_syncing_txg);
1682 }
1683 
1684 /*
1685  * Return the last txg where data can be dirtied. The final txgs
1686  * will be used to just clear out any deferred frees that remain.
1687  */
1688 uint64_t
1689 spa_final_dirty_txg(spa_t *spa)
1690 {
1691 	return (spa->spa_final_txg - TXG_DEFER_SIZE);
1692 }
1693 
1694 pool_state_t
1695 spa_state(spa_t *spa)
1696 {
1697 	return (spa->spa_state);
1698 }
1699 
1700 spa_load_state_t
1701 spa_load_state(spa_t *spa)
1702 {
1703 	return (spa->spa_load_state);
1704 }
1705 
1706 uint64_t
1707 spa_freeze_txg(spa_t *spa)
1708 {
1709 	return (spa->spa_freeze_txg);
1710 }
1711 
1712 /* ARGSUSED */
1713 uint64_t
1714 spa_get_worst_case_asize(spa_t *spa, uint64_t lsize)
1715 {
1716 	return (lsize * spa_asize_inflation);
1717 }
1718 
1719 /*
1720  * Return the amount of slop space in bytes.  It is 1/32 of the pool (3.2%),
1721  * or at least 128MB, unless that would cause it to be more than half the
1722  * pool size.
1723  *
1724  * See the comment above spa_slop_shift for details.
1725  */
1726 uint64_t
1727 spa_get_slop_space(spa_t *spa)
1728 {
1729 	uint64_t space = spa_get_dspace(spa);
1730 	return (MAX(space >> spa_slop_shift, MIN(space >> 1, spa_min_slop)));
1731 }
1732 
1733 uint64_t
1734 spa_get_dspace(spa_t *spa)
1735 {
1736 	return (spa->spa_dspace);
1737 }
1738 
1739 uint64_t
1740 spa_get_checkpoint_space(spa_t *spa)
1741 {
1742 	return (spa->spa_checkpoint_info.sci_dspace);
1743 }
1744 
1745 void
1746 spa_update_dspace(spa_t *spa)
1747 {
1748 	spa->spa_dspace = metaslab_class_get_dspace(spa_normal_class(spa)) +
1749 	    ddt_get_dedup_dspace(spa);
1750 	if (spa->spa_vdev_removal != NULL) {
1751 		/*
1752 		 * We can't allocate from the removing device, so
1753 		 * subtract its size.  This prevents the DMU/DSL from
1754 		 * filling up the (now smaller) pool while we are in the
1755 		 * middle of removing the device.
1756 		 *
1757 		 * Note that the DMU/DSL doesn't actually know or care
1758 		 * how much space is allocated (it does its own tracking
1759 		 * of how much space has been logically used).  So it
1760 		 * doesn't matter that the data we are moving may be
1761 		 * allocated twice (on the old device and the new
1762 		 * device).
1763 		 */
1764 		spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1765 		vdev_t *vd =
1766 		    vdev_lookup_top(spa, spa->spa_vdev_removal->svr_vdev_id);
1767 		spa->spa_dspace -= spa_deflate(spa) ?
1768 		    vd->vdev_stat.vs_dspace : vd->vdev_stat.vs_space;
1769 		spa_config_exit(spa, SCL_VDEV, FTAG);
1770 	}
1771 }
1772 
1773 /*
1774  * Return the failure mode that has been set to this pool. The default
1775  * behavior will be to block all I/Os when a complete failure occurs.
1776  */
1777 uint8_t
1778 spa_get_failmode(spa_t *spa)
1779 {
1780 	return (spa->spa_failmode);
1781 }
1782 
1783 boolean_t
1784 spa_suspended(spa_t *spa)
1785 {
1786 	return (spa->spa_suspended);
1787 }
1788 
1789 uint64_t
1790 spa_version(spa_t *spa)
1791 {
1792 	return (spa->spa_ubsync.ub_version);
1793 }
1794 
1795 boolean_t
1796 spa_deflate(spa_t *spa)
1797 {
1798 	return (spa->spa_deflate);
1799 }
1800 
1801 metaslab_class_t *
1802 spa_normal_class(spa_t *spa)
1803 {
1804 	return (spa->spa_normal_class);
1805 }
1806 
1807 metaslab_class_t *
1808 spa_log_class(spa_t *spa)
1809 {
1810 	return (spa->spa_log_class);
1811 }
1812 
1813 void
1814 spa_evicting_os_register(spa_t *spa, objset_t *os)
1815 {
1816 	mutex_enter(&spa->spa_evicting_os_lock);
1817 	list_insert_head(&spa->spa_evicting_os_list, os);
1818 	mutex_exit(&spa->spa_evicting_os_lock);
1819 }
1820 
1821 void
1822 spa_evicting_os_deregister(spa_t *spa, objset_t *os)
1823 {
1824 	mutex_enter(&spa->spa_evicting_os_lock);
1825 	list_remove(&spa->spa_evicting_os_list, os);
1826 	cv_broadcast(&spa->spa_evicting_os_cv);
1827 	mutex_exit(&spa->spa_evicting_os_lock);
1828 }
1829 
1830 void
1831 spa_evicting_os_wait(spa_t *spa)
1832 {
1833 	mutex_enter(&spa->spa_evicting_os_lock);
1834 	while (!list_is_empty(&spa->spa_evicting_os_list))
1835 		cv_wait(&spa->spa_evicting_os_cv, &spa->spa_evicting_os_lock);
1836 	mutex_exit(&spa->spa_evicting_os_lock);
1837 
1838 	dmu_buf_user_evict_wait();
1839 }
1840 
1841 int
1842 spa_max_replication(spa_t *spa)
1843 {
1844 	/*
1845 	 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1846 	 * handle BPs with more than one DVA allocated.  Set our max
1847 	 * replication level accordingly.
1848 	 */
1849 	if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1850 		return (1);
1851 	return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1852 }
1853 
1854 int
1855 spa_prev_software_version(spa_t *spa)
1856 {
1857 	return (spa->spa_prev_software_version);
1858 }
1859 
1860 uint64_t
1861 spa_deadman_synctime(spa_t *spa)
1862 {
1863 	return (spa->spa_deadman_synctime);
1864 }
1865 
1866 uint64_t
1867 dva_get_dsize_sync(spa_t *spa, const dva_t *dva)
1868 {
1869 	uint64_t asize = DVA_GET_ASIZE(dva);
1870 	uint64_t dsize = asize;
1871 
1872 	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1873 
1874 	if (asize != 0 && spa->spa_deflate) {
1875 		vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva));
1876 		dsize = (asize >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1877 	}
1878 
1879 	return (dsize);
1880 }
1881 
1882 uint64_t
1883 bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp)
1884 {
1885 	uint64_t dsize = 0;
1886 
1887 	for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1888 		dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1889 
1890 	return (dsize);
1891 }
1892 
1893 uint64_t
1894 bp_get_dsize(spa_t *spa, const blkptr_t *bp)
1895 {
1896 	uint64_t dsize = 0;
1897 
1898 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
1899 
1900 	for (int d = 0; d < BP_GET_NDVAS(bp); d++)
1901 		dsize += dva_get_dsize_sync(spa, &bp->blk_dva[d]);
1902 
1903 	spa_config_exit(spa, SCL_VDEV, FTAG);
1904 
1905 	return (dsize);
1906 }
1907 
1908 /*
1909  * ==========================================================================
1910  * Initialization and Termination
1911  * ==========================================================================
1912  */
1913 
1914 static int
1915 spa_name_compare(const void *a1, const void *a2)
1916 {
1917 	const spa_t *s1 = a1;
1918 	const spa_t *s2 = a2;
1919 	int s;
1920 
1921 	s = strcmp(s1->spa_name, s2->spa_name);
1922 	if (s > 0)
1923 		return (1);
1924 	if (s < 0)
1925 		return (-1);
1926 	return (0);
1927 }
1928 
1929 int
1930 spa_busy(void)
1931 {
1932 	return (spa_active_count);
1933 }
1934 
1935 void
1936 spa_boot_init()
1937 {
1938 	spa_config_load();
1939 }
1940 
1941 void
1942 spa_init(int mode)
1943 {
1944 	mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1945 	mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1946 	mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1947 	cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1948 
1949 	avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1950 	    offsetof(spa_t, spa_avl));
1951 
1952 	avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1953 	    offsetof(spa_aux_t, aux_avl));
1954 
1955 	avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1956 	    offsetof(spa_aux_t, aux_avl));
1957 
1958 	spa_mode_global = mode;
1959 
1960 #ifdef _KERNEL
1961 	spa_arch_init();
1962 #else
1963 	if (spa_mode_global != FREAD && dprintf_find_string("watch")) {
1964 		arc_procfd = open("/proc/self/ctl", O_WRONLY);
1965 		if (arc_procfd == -1) {
1966 			perror("could not enable watchpoints: "
1967 			    "opening /proc/self/ctl failed: ");
1968 		} else {
1969 			arc_watch = B_TRUE;
1970 		}
1971 	}
1972 #endif
1973 
1974 	refcount_init();
1975 	unique_init();
1976 	range_tree_init();
1977 	metaslab_alloc_trace_init();
1978 	zio_init();
1979 	dmu_init();
1980 	zil_init();
1981 	vdev_cache_stat_init();
1982 	zfs_prop_init();
1983 	zpool_prop_init();
1984 	zpool_feature_init();
1985 	spa_config_load();
1986 	l2arc_start();
1987 }
1988 
1989 void
1990 spa_fini(void)
1991 {
1992 	l2arc_stop();
1993 
1994 	spa_evict_all();
1995 
1996 	vdev_cache_stat_fini();
1997 	zil_fini();
1998 	dmu_fini();
1999 	zio_fini();
2000 	metaslab_alloc_trace_fini();
2001 	range_tree_fini();
2002 	unique_fini();
2003 	refcount_fini();
2004 
2005 	avl_destroy(&spa_namespace_avl);
2006 	avl_destroy(&spa_spare_avl);
2007 	avl_destroy(&spa_l2cache_avl);
2008 
2009 	cv_destroy(&spa_namespace_cv);
2010 	mutex_destroy(&spa_namespace_lock);
2011 	mutex_destroy(&spa_spare_lock);
2012 	mutex_destroy(&spa_l2cache_lock);
2013 }
2014 
2015 /*
2016  * Return whether this pool has slogs. No locking needed.
2017  * It's not a problem if the wrong answer is returned as it's only for
2018  * performance and not correctness
2019  */
2020 boolean_t
2021 spa_has_slogs(spa_t *spa)
2022 {
2023 	return (spa->spa_log_class->mc_rotor != NULL);
2024 }
2025 
2026 spa_log_state_t
2027 spa_get_log_state(spa_t *spa)
2028 {
2029 	return (spa->spa_log_state);
2030 }
2031 
2032 void
2033 spa_set_log_state(spa_t *spa, spa_log_state_t state)
2034 {
2035 	spa->spa_log_state = state;
2036 }
2037 
2038 boolean_t
2039 spa_is_root(spa_t *spa)
2040 {
2041 	return (spa->spa_is_root);
2042 }
2043 
2044 boolean_t
2045 spa_writeable(spa_t *spa)
2046 {
2047 	return (!!(spa->spa_mode & FWRITE) && spa->spa_trust_config);
2048 }
2049 
2050 /*
2051  * Returns true if there is a pending sync task in any of the current
2052  * syncing txg, the current quiescing txg, or the current open txg.
2053  */
2054 boolean_t
2055 spa_has_pending_synctask(spa_t *spa)
2056 {
2057 	return (!txg_all_lists_empty(&spa->spa_dsl_pool->dp_sync_tasks) ||
2058 	    !txg_all_lists_empty(&spa->spa_dsl_pool->dp_early_sync_tasks));
2059 }
2060 
2061 int
2062 spa_mode(spa_t *spa)
2063 {
2064 	return (spa->spa_mode);
2065 }
2066 
2067 uint64_t
2068 spa_bootfs(spa_t *spa)
2069 {
2070 	return (spa->spa_bootfs);
2071 }
2072 
2073 uint64_t
2074 spa_delegation(spa_t *spa)
2075 {
2076 	return (spa->spa_delegation);
2077 }
2078 
2079 objset_t *
2080 spa_meta_objset(spa_t *spa)
2081 {
2082 	return (spa->spa_meta_objset);
2083 }
2084 
2085 enum zio_checksum
2086 spa_dedup_checksum(spa_t *spa)
2087 {
2088 	return (spa->spa_dedup_checksum);
2089 }
2090 
2091 /*
2092  * Reset pool scan stat per scan pass (or reboot).
2093  */
2094 void
2095 spa_scan_stat_init(spa_t *spa)
2096 {
2097 	/* data not stored on disk */
2098 	spa->spa_scan_pass_start = gethrestime_sec();
2099 	if (dsl_scan_is_paused_scrub(spa->spa_dsl_pool->dp_scan))
2100 		spa->spa_scan_pass_scrub_pause = spa->spa_scan_pass_start;
2101 	else
2102 		spa->spa_scan_pass_scrub_pause = 0;
2103 	spa->spa_scan_pass_scrub_spent_paused = 0;
2104 	spa->spa_scan_pass_exam = 0;
2105 	vdev_scan_stat_init(spa->spa_root_vdev);
2106 }
2107 
2108 /*
2109  * Get scan stats for zpool status reports
2110  */
2111 int
2112 spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps)
2113 {
2114 	dsl_scan_t *scn = spa->spa_dsl_pool ? spa->spa_dsl_pool->dp_scan : NULL;
2115 
2116 	if (scn == NULL || scn->scn_phys.scn_func == POOL_SCAN_NONE)
2117 		return (SET_ERROR(ENOENT));
2118 	bzero(ps, sizeof (pool_scan_stat_t));
2119 
2120 	/* data stored on disk */
2121 	ps->pss_func = scn->scn_phys.scn_func;
2122 	ps->pss_start_time = scn->scn_phys.scn_start_time;
2123 	ps->pss_end_time = scn->scn_phys.scn_end_time;
2124 	ps->pss_to_examine = scn->scn_phys.scn_to_examine;
2125 	ps->pss_examined = scn->scn_phys.scn_examined;
2126 	ps->pss_to_process = scn->scn_phys.scn_to_process;
2127 	ps->pss_processed = scn->scn_phys.scn_processed;
2128 	ps->pss_errors = scn->scn_phys.scn_errors;
2129 	ps->pss_state = scn->scn_phys.scn_state;
2130 
2131 	/* data not stored on disk */
2132 	ps->pss_pass_start = spa->spa_scan_pass_start;
2133 	ps->pss_pass_exam = spa->spa_scan_pass_exam;
2134 	ps->pss_pass_scrub_pause = spa->spa_scan_pass_scrub_pause;
2135 	ps->pss_pass_scrub_spent_paused = spa->spa_scan_pass_scrub_spent_paused;
2136 
2137 	return (0);
2138 }
2139 
2140 boolean_t
2141 spa_debug_enabled(spa_t *spa)
2142 {
2143 	return (spa->spa_debug);
2144 }
2145 
2146 int
2147 spa_maxblocksize(spa_t *spa)
2148 {
2149 	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS))
2150 		return (SPA_MAXBLOCKSIZE);
2151 	else
2152 		return (SPA_OLD_MAXBLOCKSIZE);
2153 }
2154 
2155 /*
2156  * Returns the txg that the last device removal completed. No indirect mappings
2157  * have been added since this txg.
2158  */
2159 uint64_t
2160 spa_get_last_removal_txg(spa_t *spa)
2161 {
2162 	uint64_t vdevid;
2163 	uint64_t ret = -1ULL;
2164 
2165 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2166 	/*
2167 	 * sr_prev_indirect_vdev is only modified while holding all the
2168 	 * config locks, so it is sufficient to hold SCL_VDEV as reader when
2169 	 * examining it.
2170 	 */
2171 	vdevid = spa->spa_removing_phys.sr_prev_indirect_vdev;
2172 
2173 	while (vdevid != -1ULL) {
2174 		vdev_t *vd = vdev_lookup_top(spa, vdevid);
2175 		vdev_indirect_births_t *vib = vd->vdev_indirect_births;
2176 
2177 		ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops);
2178 
2179 		/*
2180 		 * If the removal did not remap any data, we don't care.
2181 		 */
2182 		if (vdev_indirect_births_count(vib) != 0) {
2183 			ret = vdev_indirect_births_last_entry_txg(vib);
2184 			break;
2185 		}
2186 
2187 		vdevid = vd->vdev_indirect_config.vic_prev_indirect_vdev;
2188 	}
2189 	spa_config_exit(spa, SCL_VDEV, FTAG);
2190 
2191 	IMPLY(ret != -1ULL,
2192 	    spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL));
2193 
2194 	return (ret);
2195 }
2196 
2197 boolean_t
2198 spa_trust_config(spa_t *spa)
2199 {
2200 	return (spa->spa_trust_config);
2201 }
2202 
2203 uint64_t
2204 spa_missing_tvds_allowed(spa_t *spa)
2205 {
2206 	return (spa->spa_missing_tvds_allowed);
2207 }
2208 
2209 void
2210 spa_set_missing_tvds(spa_t *spa, uint64_t missing)
2211 {
2212 	spa->spa_missing_tvds = missing;
2213 }
2214 
2215 boolean_t
2216 spa_top_vdevs_spacemap_addressable(spa_t *spa)
2217 {
2218 	vdev_t *rvd = spa->spa_root_vdev;
2219 	for (uint64_t c = 0; c < rvd->vdev_children; c++) {
2220 		if (!vdev_is_spacemap_addressable(rvd->vdev_child[c]))
2221 			return (B_FALSE);
2222 	}
2223 	return (B_TRUE);
2224 }
2225 
2226 boolean_t
2227 spa_has_checkpoint(spa_t *spa)
2228 {
2229 	return (spa->spa_checkpoint_txg != 0);
2230 }
2231 
2232 boolean_t
2233 spa_importing_readonly_checkpoint(spa_t *spa)
2234 {
2235 	return ((spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT) &&
2236 	    spa->spa_mode == FREAD);
2237 }
2238 
2239 uint64_t
2240 spa_min_claim_txg(spa_t *spa)
2241 {
2242 	uint64_t checkpoint_txg = spa->spa_uberblock.ub_checkpoint_txg;
2243 
2244 	if (checkpoint_txg != 0)
2245 		return (checkpoint_txg + 1);
2246 
2247 	return (spa->spa_first_txg);
2248 }
2249 
2250 /*
2251  * If there is a checkpoint, async destroys may consume more space from
2252  * the pool instead of freeing it. In an attempt to save the pool from
2253  * getting suspended when it is about to run out of space, we stop
2254  * processing async destroys.
2255  */
2256 boolean_t
2257 spa_suspend_async_destroy(spa_t *spa)
2258 {
2259 	dsl_pool_t *dp = spa_get_dsl(spa);
2260 
2261 	uint64_t unreserved = dsl_pool_unreserved_space(dp,
2262 	    ZFS_SPACE_CHECK_EXTRA_RESERVED);
2263 	uint64_t used = dsl_dir_phys(dp->dp_root_dir)->dd_used_bytes;
2264 	uint64_t avail = (unreserved > used) ? (unreserved - used) : 0;
2265 
2266 	if (spa_has_checkpoint(spa) && avail == 0)
2267 		return (B_TRUE);
2268 
2269 	return (B_FALSE);
2270 }
2271