spa_misc.c revision e05725b117836db173257fae43fb0746eb857fb5
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 2007 Sun Microsystems, Inc.  All rights reserved.
23 * Use is subject to license terms.
24 */
25
26#pragma ident	"%Z%%M%	%I%	%E% SMI"
27
28#include <sys/zfs_context.h>
29#include <sys/spa_impl.h>
30#include <sys/zio.h>
31#include <sys/zio_checksum.h>
32#include <sys/zio_compress.h>
33#include <sys/dmu.h>
34#include <sys/dmu_tx.h>
35#include <sys/zap.h>
36#include <sys/zil.h>
37#include <sys/vdev_impl.h>
38#include <sys/metaslab.h>
39#include <sys/uberblock_impl.h>
40#include <sys/txg.h>
41#include <sys/avl.h>
42#include <sys/unique.h>
43#include <sys/dsl_pool.h>
44#include <sys/dsl_dir.h>
45#include <sys/dsl_prop.h>
46#include <sys/fs/zfs.h>
47#include <sys/metaslab_impl.h>
48#include "zfs_prop.h"
49
50/*
51 * SPA locking
52 *
53 * There are four basic locks for managing spa_t structures:
54 *
55 * spa_namespace_lock (global mutex)
56 *
57 *	This lock must be acquired to do any of the following:
58 *
59 *		- Lookup a spa_t by name
60 *		- Add or remove a spa_t from the namespace
61 *		- Increase spa_refcount from non-zero
62 *		- Check if spa_refcount is zero
63 *		- Rename a spa_t
64 *		- add/remove/attach/detach devices
65 *		- Held for the duration of create/destroy/import/export
66 *
67 *	It does not need to handle recursion.  A create or destroy may
68 *	reference objects (files or zvols) in other pools, but by
69 *	definition they must have an existing reference, and will never need
70 *	to lookup a spa_t by name.
71 *
72 * spa_refcount (per-spa refcount_t protected by mutex)
73 *
74 *	This reference count keep track of any active users of the spa_t.  The
75 *	spa_t cannot be destroyed or freed while this is non-zero.  Internally,
76 *	the refcount is never really 'zero' - opening a pool implicitly keeps
77 *	some references in the DMU.  Internally we check against SPA_MINREF, but
78 *	present the image of a zero/non-zero value to consumers.
79 *
80 * spa_config_lock (per-spa read-priority rwlock)
81 *
82 *	This protects the spa_t from config changes, and must be held in
83 *	the following circumstances:
84 *
85 *		- RW_READER to perform I/O to the spa
86 *		- RW_WRITER to change the vdev config
87 *
88 * spa_config_cache_lock (per-spa mutex)
89 *
90 *	This mutex prevents the spa_config nvlist from being updated.  No
91 *      other locks are required to obtain this lock, although implicitly you
92 *      must have the namespace lock or non-zero refcount to have any kind
93 *      of spa_t pointer at all.
94 *
95 * The locking order is fairly straightforward:
96 *
97 *		spa_namespace_lock	->	spa_refcount
98 *
99 *	The namespace lock must be acquired to increase the refcount from 0
100 *	or to check if it is zero.
101 *
102 *		spa_refcount		->	spa_config_lock
103 *
104 *	There must be at least one valid reference on the spa_t to acquire
105 *	the config lock.
106 *
107 *		spa_namespace_lock	->	spa_config_lock
108 *
109 *	The namespace lock must always be taken before the config lock.
110 *
111 *
112 * The spa_namespace_lock and spa_config_cache_lock can be acquired directly and
113 * are globally visible.
114 *
115 * The namespace is manipulated using the following functions, all which require
116 * 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 a form of rwlock.  It must be held as RW_READER
148 * to perform I/O to the pool, and as RW_WRITER to change the vdev config.
149 * The spa_config_lock is manipulated with spa_config_{enter,exit,held}().
150 *
151 * The vdev configuration is protected by spa_vdev_enter() / spa_vdev_exit().
152 *
153 *	spa_vdev_enter()	Acquire the namespace lock and the config lock
154 *				for writing.
155 *
156 *	spa_vdev_exit()		Release the config lock, wait for all I/O
157 *				to complete, sync the updated configs to the
158 *				cache, and release the namespace lock.
159 *
160 * The spa_name() function also requires either the spa_namespace_lock
161 * or the spa_config_lock, as both are needed to do a rename.  spa_rename() is
162 * also implemented within this file since is requires manipulation of the
163 * namespace.
164 */
165
166static avl_tree_t spa_namespace_avl;
167kmutex_t spa_namespace_lock;
168static kcondvar_t spa_namespace_cv;
169static int spa_active_count;
170int spa_max_replication_override = SPA_DVAS_PER_BP;
171
172static kmutex_t spa_spare_lock;
173static avl_tree_t spa_spare_avl;
174static kmutex_t spa_l2cache_lock;
175static avl_tree_t spa_l2cache_avl;
176
177kmem_cache_t *spa_buffer_pool;
178int spa_mode;
179
180#ifdef ZFS_DEBUG
181/* Everything except dprintf is on by default in debug builds */
182int zfs_flags = ~ZFS_DEBUG_DPRINTF;
183#else
184int zfs_flags = 0;
185#endif
186
187/*
188 * zfs_recover can be set to nonzero to attempt to recover from
189 * otherwise-fatal errors, typically caused by on-disk corruption.  When
190 * set, calls to zfs_panic_recover() will turn into warning messages.
191 */
192int zfs_recover = 0;
193
194#define	SPA_MINREF	5	/* spa_refcnt for an open-but-idle pool */
195
196/*
197 * ==========================================================================
198 * SPA config locking
199 * ==========================================================================
200 */
201static void
202spa_config_lock_init(spa_config_lock_t *scl)
203{
204	mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
205	scl->scl_writer = NULL;
206	cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
207	refcount_create(&scl->scl_count);
208}
209
210static void
211spa_config_lock_destroy(spa_config_lock_t *scl)
212{
213	mutex_destroy(&scl->scl_lock);
214	ASSERT(scl->scl_writer == NULL);
215	cv_destroy(&scl->scl_cv);
216	refcount_destroy(&scl->scl_count);
217}
218
219void
220spa_config_enter(spa_t *spa, krw_t rw, void *tag)
221{
222	spa_config_lock_t *scl = &spa->spa_config_lock;
223
224	mutex_enter(&scl->scl_lock);
225
226	if (rw == RW_READER) {
227		while (scl->scl_writer != NULL && scl->scl_writer != curthread)
228			cv_wait(&scl->scl_cv, &scl->scl_lock);
229	} else {
230		while (!refcount_is_zero(&scl->scl_count) &&
231		    scl->scl_writer != curthread)
232			cv_wait(&scl->scl_cv, &scl->scl_lock);
233		scl->scl_writer = curthread;
234	}
235
236	(void) refcount_add(&scl->scl_count, tag);
237
238	mutex_exit(&scl->scl_lock);
239}
240
241void
242spa_config_exit(spa_t *spa, void *tag)
243{
244	spa_config_lock_t *scl = &spa->spa_config_lock;
245
246	mutex_enter(&scl->scl_lock);
247
248	ASSERT(!refcount_is_zero(&scl->scl_count));
249
250	if (refcount_remove(&scl->scl_count, tag) == 0) {
251		cv_broadcast(&scl->scl_cv);
252		ASSERT(scl->scl_writer == NULL || scl->scl_writer == curthread);
253		scl->scl_writer = NULL;  /* OK in either case */
254	}
255
256	mutex_exit(&scl->scl_lock);
257}
258
259boolean_t
260spa_config_held(spa_t *spa, krw_t rw)
261{
262	spa_config_lock_t *scl = &spa->spa_config_lock;
263
264	if (rw == RW_READER)
265		return (!refcount_is_zero(&scl->scl_count));
266	else
267		return (scl->scl_writer == curthread);
268}
269
270/*
271 * ==========================================================================
272 * SPA namespace functions
273 * ==========================================================================
274 */
275
276/*
277 * Lookup the named spa_t in the AVL tree.  The spa_namespace_lock must be held.
278 * Returns NULL if no matching spa_t is found.
279 */
280spa_t *
281spa_lookup(const char *name)
282{
283	spa_t search, *spa;
284	avl_index_t where;
285	char c;
286	char *cp;
287
288	ASSERT(MUTEX_HELD(&spa_namespace_lock));
289
290	/*
291	 * If it's a full dataset name, figure out the pool name and
292	 * just use that.
293	 */
294	cp = strpbrk(name, "/@");
295	if (cp) {
296		c = *cp;
297		*cp = '\0';
298	}
299
300	search.spa_name = (char *)name;
301	spa = avl_find(&spa_namespace_avl, &search, &where);
302
303	if (cp)
304		*cp = c;
305
306	return (spa);
307}
308
309/*
310 * Create an uninitialized spa_t with the given name.  Requires
311 * spa_namespace_lock.  The caller must ensure that the spa_t doesn't already
312 * exist by calling spa_lookup() first.
313 */
314spa_t *
315spa_add(const char *name, const char *altroot)
316{
317	spa_t *spa;
318
319	ASSERT(MUTEX_HELD(&spa_namespace_lock));
320
321	spa = kmem_zalloc(sizeof (spa_t), KM_SLEEP);
322
323	rw_init(&spa->spa_traverse_lock, NULL, RW_DEFAULT, NULL);
324
325	mutex_init(&spa->spa_uberblock_lock, NULL, MUTEX_DEFAULT, NULL);
326	mutex_init(&spa->spa_async_lock, NULL, MUTEX_DEFAULT, NULL);
327	mutex_init(&spa->spa_config_cache_lock, NULL, MUTEX_DEFAULT, NULL);
328	mutex_init(&spa->spa_scrub_lock, NULL, MUTEX_DEFAULT, NULL);
329	mutex_init(&spa->spa_errlog_lock, NULL, MUTEX_DEFAULT, NULL);
330	mutex_init(&spa->spa_errlist_lock, NULL, MUTEX_DEFAULT, NULL);
331	mutex_init(&spa->spa_sync_bplist.bpl_lock, NULL, MUTEX_DEFAULT, NULL);
332	mutex_init(&spa->spa_history_lock, NULL, MUTEX_DEFAULT, NULL);
333	mutex_init(&spa->spa_props_lock, NULL, MUTEX_DEFAULT, NULL);
334
335	cv_init(&spa->spa_async_cv, NULL, CV_DEFAULT, NULL);
336	cv_init(&spa->spa_scrub_cv, NULL, CV_DEFAULT, NULL);
337	cv_init(&spa->spa_scrub_io_cv, NULL, CV_DEFAULT, NULL);
338
339	spa->spa_name = spa_strdup(name);
340	spa->spa_state = POOL_STATE_UNINITIALIZED;
341	spa->spa_freeze_txg = UINT64_MAX;
342	spa->spa_final_txg = UINT64_MAX;
343
344	refcount_create(&spa->spa_refcount);
345	spa_config_lock_init(&spa->spa_config_lock);
346
347	avl_add(&spa_namespace_avl, spa);
348
349	mutex_init(&spa->spa_zio_lock, NULL, MUTEX_DEFAULT, NULL);
350
351	/*
352	 * Set the alternate root, if there is one.
353	 */
354	if (altroot) {
355		spa->spa_root = spa_strdup(altroot);
356		spa_active_count++;
357	}
358
359	return (spa);
360}
361
362/*
363 * Removes a spa_t from the namespace, freeing up any memory used.  Requires
364 * spa_namespace_lock.  This is called only after the spa_t has been closed and
365 * deactivated.
366 */
367void
368spa_remove(spa_t *spa)
369{
370	ASSERT(MUTEX_HELD(&spa_namespace_lock));
371	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
372	ASSERT(spa->spa_scrub_thread == NULL);
373
374	avl_remove(&spa_namespace_avl, spa);
375	cv_broadcast(&spa_namespace_cv);
376
377	if (spa->spa_root) {
378		spa_strfree(spa->spa_root);
379		spa_active_count--;
380	}
381
382	if (spa->spa_name)
383		spa_strfree(spa->spa_name);
384
385	if (spa->spa_config_dir)
386		spa_strfree(spa->spa_config_dir);
387	if (spa->spa_config_file)
388		spa_strfree(spa->spa_config_file);
389
390	spa_config_set(spa, NULL);
391
392	refcount_destroy(&spa->spa_refcount);
393
394	spa_config_lock_destroy(&spa->spa_config_lock);
395
396	rw_destroy(&spa->spa_traverse_lock);
397
398	cv_destroy(&spa->spa_async_cv);
399	cv_destroy(&spa->spa_scrub_cv);
400	cv_destroy(&spa->spa_scrub_io_cv);
401
402	mutex_destroy(&spa->spa_uberblock_lock);
403	mutex_destroy(&spa->spa_async_lock);
404	mutex_destroy(&spa->spa_config_cache_lock);
405	mutex_destroy(&spa->spa_scrub_lock);
406	mutex_destroy(&spa->spa_errlog_lock);
407	mutex_destroy(&spa->spa_errlist_lock);
408	mutex_destroy(&spa->spa_sync_bplist.bpl_lock);
409	mutex_destroy(&spa->spa_history_lock);
410	mutex_destroy(&spa->spa_props_lock);
411	mutex_destroy(&spa->spa_zio_lock);
412
413	kmem_free(spa, sizeof (spa_t));
414}
415
416/*
417 * Given a pool, return the next pool in the namespace, or NULL if there is
418 * none.  If 'prev' is NULL, return the first pool.
419 */
420spa_t *
421spa_next(spa_t *prev)
422{
423	ASSERT(MUTEX_HELD(&spa_namespace_lock));
424
425	if (prev)
426		return (AVL_NEXT(&spa_namespace_avl, prev));
427	else
428		return (avl_first(&spa_namespace_avl));
429}
430
431/*
432 * ==========================================================================
433 * SPA refcount functions
434 * ==========================================================================
435 */
436
437/*
438 * Add a reference to the given spa_t.  Must have at least one reference, or
439 * have the namespace lock held.
440 */
441void
442spa_open_ref(spa_t *spa, void *tag)
443{
444	ASSERT(refcount_count(&spa->spa_refcount) > SPA_MINREF ||
445	    MUTEX_HELD(&spa_namespace_lock));
446
447	(void) refcount_add(&spa->spa_refcount, tag);
448}
449
450/*
451 * Remove a reference to the given spa_t.  Must have at least one reference, or
452 * have the namespace lock held.
453 */
454void
455spa_close(spa_t *spa, void *tag)
456{
457	ASSERT(refcount_count(&spa->spa_refcount) > SPA_MINREF ||
458	    MUTEX_HELD(&spa_namespace_lock));
459
460	(void) refcount_remove(&spa->spa_refcount, tag);
461}
462
463/*
464 * Check to see if the spa refcount is zero.  Must be called with
465 * spa_namespace_lock held.  We really compare against SPA_MINREF, which is the
466 * number of references acquired when opening a pool
467 */
468boolean_t
469spa_refcount_zero(spa_t *spa)
470{
471	ASSERT(MUTEX_HELD(&spa_namespace_lock));
472
473	return (refcount_count(&spa->spa_refcount) == SPA_MINREF);
474}
475
476/*
477 * ==========================================================================
478 * SPA spare and l2cache tracking
479 * ==========================================================================
480 */
481
482/*
483 * Hot spares and cache devices are tracked using the same code below,
484 * for 'auxiliary' devices.
485 */
486
487typedef struct spa_aux {
488	uint64_t	aux_guid;
489	uint64_t	aux_pool;
490	avl_node_t	aux_avl;
491	int		aux_count;
492} spa_aux_t;
493
494static int
495spa_aux_compare(const void *a, const void *b)
496{
497	const spa_aux_t *sa = a;
498	const spa_aux_t *sb = b;
499
500	if (sa->aux_guid < sb->aux_guid)
501		return (-1);
502	else if (sa->aux_guid > sb->aux_guid)
503		return (1);
504	else
505		return (0);
506}
507
508void
509spa_aux_add(vdev_t *vd, avl_tree_t *avl)
510{
511	avl_index_t where;
512	spa_aux_t search;
513	spa_aux_t *aux;
514
515	search.aux_guid = vd->vdev_guid;
516	if ((aux = avl_find(avl, &search, &where)) != NULL) {
517		aux->aux_count++;
518	} else {
519		aux = kmem_zalloc(sizeof (spa_aux_t), KM_SLEEP);
520		aux->aux_guid = vd->vdev_guid;
521		aux->aux_count = 1;
522		avl_insert(avl, aux, where);
523	}
524}
525
526void
527spa_aux_remove(vdev_t *vd, avl_tree_t *avl)
528{
529	spa_aux_t search;
530	spa_aux_t *aux;
531	avl_index_t where;
532
533	search.aux_guid = vd->vdev_guid;
534	aux = avl_find(avl, &search, &where);
535
536	ASSERT(aux != NULL);
537
538	if (--aux->aux_count == 0) {
539		avl_remove(avl, aux);
540		kmem_free(aux, sizeof (spa_aux_t));
541	} else if (aux->aux_pool == spa_guid(vd->vdev_spa)) {
542		aux->aux_pool = 0ULL;
543	}
544}
545
546boolean_t
547spa_aux_exists(uint64_t guid, uint64_t *pool, avl_tree_t *avl)
548{
549	spa_aux_t search, *found;
550	avl_index_t where;
551
552	search.aux_guid = guid;
553	found = avl_find(avl, &search, &where);
554
555	if (pool) {
556		if (found)
557			*pool = found->aux_pool;
558		else
559			*pool = 0ULL;
560	}
561
562	return (found != NULL);
563}
564
565void
566spa_aux_activate(vdev_t *vd, avl_tree_t *avl)
567{
568	spa_aux_t search, *found;
569	avl_index_t where;
570
571	search.aux_guid = vd->vdev_guid;
572	found = avl_find(avl, &search, &where);
573	ASSERT(found != NULL);
574	ASSERT(found->aux_pool == 0ULL);
575
576	found->aux_pool = spa_guid(vd->vdev_spa);
577}
578
579/*
580 * Spares are tracked globally due to the following constraints:
581 *
582 * 	- A spare may be part of multiple pools.
583 * 	- A spare may be added to a pool even if it's actively in use within
584 *	  another pool.
585 * 	- A spare in use in any pool can only be the source of a replacement if
586 *	  the target is a spare in the same pool.
587 *
588 * We keep track of all spares on the system through the use of a reference
589 * counted AVL tree.  When a vdev is added as a spare, or used as a replacement
590 * spare, then we bump the reference count in the AVL tree.  In addition, we set
591 * the 'vdev_isspare' member to indicate that the device is a spare (active or
592 * inactive).  When a spare is made active (used to replace a device in the
593 * pool), we also keep track of which pool its been made a part of.
594 *
595 * The 'spa_spare_lock' protects the AVL tree.  These functions are normally
596 * called under the spa_namespace lock as part of vdev reconfiguration.  The
597 * separate spare lock exists for the status query path, which does not need to
598 * be completely consistent with respect to other vdev configuration changes.
599 */
600
601static int
602spa_spare_compare(const void *a, const void *b)
603{
604	return (spa_aux_compare(a, b));
605}
606
607void
608spa_spare_add(vdev_t *vd)
609{
610	mutex_enter(&spa_spare_lock);
611	ASSERT(!vd->vdev_isspare);
612	spa_aux_add(vd, &spa_spare_avl);
613	vd->vdev_isspare = B_TRUE;
614	mutex_exit(&spa_spare_lock);
615}
616
617void
618spa_spare_remove(vdev_t *vd)
619{
620	mutex_enter(&spa_spare_lock);
621	ASSERT(vd->vdev_isspare);
622	spa_aux_remove(vd, &spa_spare_avl);
623	vd->vdev_isspare = B_FALSE;
624	mutex_exit(&spa_spare_lock);
625}
626
627boolean_t
628spa_spare_exists(uint64_t guid, uint64_t *pool)
629{
630	boolean_t found;
631
632	mutex_enter(&spa_spare_lock);
633	found = spa_aux_exists(guid, pool, &spa_spare_avl);
634	mutex_exit(&spa_spare_lock);
635
636	return (found);
637}
638
639void
640spa_spare_activate(vdev_t *vd)
641{
642	mutex_enter(&spa_spare_lock);
643	ASSERT(vd->vdev_isspare);
644	spa_aux_activate(vd, &spa_spare_avl);
645	mutex_exit(&spa_spare_lock);
646}
647
648/*
649 * Level 2 ARC devices are tracked globally for the same reasons as spares.
650 * Cache devices currently only support one pool per cache device, and so
651 * for these devices the aux reference count is currently unused beyond 1.
652 */
653
654static int
655spa_l2cache_compare(const void *a, const void *b)
656{
657	return (spa_aux_compare(a, b));
658}
659
660void
661spa_l2cache_add(vdev_t *vd)
662{
663	mutex_enter(&spa_l2cache_lock);
664	ASSERT(!vd->vdev_isl2cache);
665	spa_aux_add(vd, &spa_l2cache_avl);
666	vd->vdev_isl2cache = B_TRUE;
667	mutex_exit(&spa_l2cache_lock);
668}
669
670void
671spa_l2cache_remove(vdev_t *vd)
672{
673	mutex_enter(&spa_l2cache_lock);
674	ASSERT(vd->vdev_isl2cache);
675	spa_aux_remove(vd, &spa_l2cache_avl);
676	vd->vdev_isl2cache = B_FALSE;
677	mutex_exit(&spa_l2cache_lock);
678}
679
680boolean_t
681spa_l2cache_exists(uint64_t guid, uint64_t *pool)
682{
683	boolean_t found;
684
685	mutex_enter(&spa_l2cache_lock);
686	found = spa_aux_exists(guid, pool, &spa_l2cache_avl);
687	mutex_exit(&spa_l2cache_lock);
688
689	return (found);
690}
691
692void
693spa_l2cache_activate(vdev_t *vd)
694{
695	mutex_enter(&spa_l2cache_lock);
696	ASSERT(vd->vdev_isl2cache);
697	spa_aux_activate(vd, &spa_l2cache_avl);
698	mutex_exit(&spa_l2cache_lock);
699}
700
701void
702spa_l2cache_space_update(vdev_t *vd, int64_t space, int64_t alloc)
703{
704	vdev_space_update(vd, space, alloc, B_FALSE);
705}
706
707/*
708 * ==========================================================================
709 * SPA vdev locking
710 * ==========================================================================
711 */
712
713/*
714 * Lock the given spa_t for the purpose of adding or removing a vdev.
715 * Grabs the global spa_namespace_lock plus the spa config lock for writing.
716 * It returns the next transaction group for the spa_t.
717 */
718uint64_t
719spa_vdev_enter(spa_t *spa)
720{
721	mutex_enter(&spa_namespace_lock);
722
723	/*
724	 * Suspend scrub activity while we mess with the config.  We must do
725	 * this after acquiring the namespace lock to avoid a 3-way deadlock
726	 * with spa_scrub_stop() and the scrub thread.
727	 */
728	spa_scrub_suspend(spa);
729
730	spa_config_enter(spa, RW_WRITER, spa);
731
732	return (spa_last_synced_txg(spa) + 1);
733}
734
735/*
736 * Unlock the spa_t after adding or removing a vdev.  Besides undoing the
737 * locking of spa_vdev_enter(), we also want make sure the transactions have
738 * synced to disk, and then update the global configuration cache with the new
739 * information.
740 */
741int
742spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error)
743{
744	int config_changed = B_FALSE;
745
746	ASSERT(txg > spa_last_synced_txg(spa));
747
748	/*
749	 * Reassess the DTLs.
750	 */
751	vdev_dtl_reassess(spa->spa_root_vdev, 0, 0, B_FALSE);
752
753	/*
754	 * If the config changed, notify the scrub thread that it must restart.
755	 */
756	if (error == 0 && !list_is_empty(&spa->spa_dirty_list)) {
757		config_changed = B_TRUE;
758		spa_scrub_restart(spa, txg);
759	}
760
761	spa_config_exit(spa, spa);
762
763	/*
764	 * Allow scrubbing to resume.
765	 */
766	spa_scrub_resume(spa);
767
768	/*
769	 * Note: this txg_wait_synced() is important because it ensures
770	 * that there won't be more than one config change per txg.
771	 * This allows us to use the txg as the generation number.
772	 */
773	if (error == 0)
774		txg_wait_synced(spa->spa_dsl_pool, txg);
775
776	if (vd != NULL) {
777		ASSERT(!vd->vdev_detached || vd->vdev_dtl.smo_object == 0);
778		vdev_free(vd);
779	}
780
781	/*
782	 * If the config changed, update the config cache.
783	 */
784	if (config_changed)
785		spa_config_sync();
786
787	mutex_exit(&spa_namespace_lock);
788
789	return (error);
790}
791
792/*
793 * ==========================================================================
794 * Miscellaneous functions
795 * ==========================================================================
796 */
797
798/*
799 * Rename a spa_t.
800 */
801int
802spa_rename(const char *name, const char *newname)
803{
804	spa_t *spa;
805	int err;
806
807	/*
808	 * Lookup the spa_t and grab the config lock for writing.  We need to
809	 * actually open the pool so that we can sync out the necessary labels.
810	 * It's OK to call spa_open() with the namespace lock held because we
811	 * allow recursive calls for other reasons.
812	 */
813	mutex_enter(&spa_namespace_lock);
814	if ((err = spa_open(name, &spa, FTAG)) != 0) {
815		mutex_exit(&spa_namespace_lock);
816		return (err);
817	}
818
819	spa_config_enter(spa, RW_WRITER, FTAG);
820
821	avl_remove(&spa_namespace_avl, spa);
822	spa_strfree(spa->spa_name);
823	spa->spa_name = spa_strdup(newname);
824	avl_add(&spa_namespace_avl, spa);
825
826	/*
827	 * Sync all labels to disk with the new names by marking the root vdev
828	 * dirty and waiting for it to sync.  It will pick up the new pool name
829	 * during the sync.
830	 */
831	vdev_config_dirty(spa->spa_root_vdev);
832
833	spa_config_exit(spa, FTAG);
834
835	txg_wait_synced(spa->spa_dsl_pool, 0);
836
837	/*
838	 * Sync the updated config cache.
839	 */
840	spa_config_sync();
841
842	spa_close(spa, FTAG);
843
844	mutex_exit(&spa_namespace_lock);
845
846	return (0);
847}
848
849
850/*
851 * Determine whether a pool with given pool_guid exists.  If device_guid is
852 * non-zero, determine whether the pool exists *and* contains a device with the
853 * specified device_guid.
854 */
855boolean_t
856spa_guid_exists(uint64_t pool_guid, uint64_t device_guid)
857{
858	spa_t *spa;
859	avl_tree_t *t = &spa_namespace_avl;
860
861	ASSERT(MUTEX_HELD(&spa_namespace_lock));
862
863	for (spa = avl_first(t); spa != NULL; spa = AVL_NEXT(t, spa)) {
864		if (spa->spa_state == POOL_STATE_UNINITIALIZED)
865			continue;
866		if (spa->spa_root_vdev == NULL)
867			continue;
868		if (spa_guid(spa) == pool_guid) {
869			if (device_guid == 0)
870				break;
871
872			if (vdev_lookup_by_guid(spa->spa_root_vdev,
873			    device_guid) != NULL)
874				break;
875
876			/*
877			 * Check any devices we may be in the process of adding.
878			 */
879			if (spa->spa_pending_vdev) {
880				if (vdev_lookup_by_guid(spa->spa_pending_vdev,
881				    device_guid) != NULL)
882					break;
883			}
884		}
885	}
886
887	return (spa != NULL);
888}
889
890char *
891spa_strdup(const char *s)
892{
893	size_t len;
894	char *new;
895
896	len = strlen(s);
897	new = kmem_alloc(len + 1, KM_SLEEP);
898	bcopy(s, new, len);
899	new[len] = '\0';
900
901	return (new);
902}
903
904void
905spa_strfree(char *s)
906{
907	kmem_free(s, strlen(s) + 1);
908}
909
910uint64_t
911spa_get_random(uint64_t range)
912{
913	uint64_t r;
914
915	ASSERT(range != 0);
916
917	(void) random_get_pseudo_bytes((void *)&r, sizeof (uint64_t));
918
919	return (r % range);
920}
921
922void
923sprintf_blkptr(char *buf, int len, const blkptr_t *bp)
924{
925	int d;
926
927	if (bp == NULL) {
928		(void) snprintf(buf, len, "<NULL>");
929		return;
930	}
931
932	if (BP_IS_HOLE(bp)) {
933		(void) snprintf(buf, len, "<hole>");
934		return;
935	}
936
937	(void) snprintf(buf, len, "[L%llu %s] %llxL/%llxP ",
938	    (u_longlong_t)BP_GET_LEVEL(bp),
939	    dmu_ot[BP_GET_TYPE(bp)].ot_name,
940	    (u_longlong_t)BP_GET_LSIZE(bp),
941	    (u_longlong_t)BP_GET_PSIZE(bp));
942
943	for (d = 0; d < BP_GET_NDVAS(bp); d++) {
944		const dva_t *dva = &bp->blk_dva[d];
945		(void) snprintf(buf + strlen(buf), len - strlen(buf),
946		    "DVA[%d]=<%llu:%llx:%llx> ", d,
947		    (u_longlong_t)DVA_GET_VDEV(dva),
948		    (u_longlong_t)DVA_GET_OFFSET(dva),
949		    (u_longlong_t)DVA_GET_ASIZE(dva));
950	}
951
952	(void) snprintf(buf + strlen(buf), len - strlen(buf),
953	    "%s %s %s %s birth=%llu fill=%llu cksum=%llx:%llx:%llx:%llx",
954	    zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_name,
955	    zio_compress_table[BP_GET_COMPRESS(bp)].ci_name,
956	    BP_GET_BYTEORDER(bp) == 0 ? "BE" : "LE",
957	    BP_IS_GANG(bp) ? "gang" : "contiguous",
958	    (u_longlong_t)bp->blk_birth,
959	    (u_longlong_t)bp->blk_fill,
960	    (u_longlong_t)bp->blk_cksum.zc_word[0],
961	    (u_longlong_t)bp->blk_cksum.zc_word[1],
962	    (u_longlong_t)bp->blk_cksum.zc_word[2],
963	    (u_longlong_t)bp->blk_cksum.zc_word[3]);
964}
965
966void
967spa_freeze(spa_t *spa)
968{
969	uint64_t freeze_txg = 0;
970
971	spa_config_enter(spa, RW_WRITER, FTAG);
972	if (spa->spa_freeze_txg == UINT64_MAX) {
973		freeze_txg = spa_last_synced_txg(spa) + TXG_SIZE;
974		spa->spa_freeze_txg = freeze_txg;
975	}
976	spa_config_exit(spa, FTAG);
977	if (freeze_txg != 0)
978		txg_wait_synced(spa_get_dsl(spa), freeze_txg);
979}
980
981void
982zfs_panic_recover(const char *fmt, ...)
983{
984	va_list adx;
985
986	va_start(adx, fmt);
987	vcmn_err(zfs_recover ? CE_WARN : CE_PANIC, fmt, adx);
988	va_end(adx);
989}
990
991/*
992 * ==========================================================================
993 * Accessor functions
994 * ==========================================================================
995 */
996
997krwlock_t *
998spa_traverse_rwlock(spa_t *spa)
999{
1000	return (&spa->spa_traverse_lock);
1001}
1002
1003int
1004spa_traverse_wanted(spa_t *spa)
1005{
1006	return (spa->spa_traverse_wanted);
1007}
1008
1009dsl_pool_t *
1010spa_get_dsl(spa_t *spa)
1011{
1012	return (spa->spa_dsl_pool);
1013}
1014
1015blkptr_t *
1016spa_get_rootblkptr(spa_t *spa)
1017{
1018	return (&spa->spa_ubsync.ub_rootbp);
1019}
1020
1021void
1022spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp)
1023{
1024	spa->spa_uberblock.ub_rootbp = *bp;
1025}
1026
1027void
1028spa_altroot(spa_t *spa, char *buf, size_t buflen)
1029{
1030	if (spa->spa_root == NULL)
1031		buf[0] = '\0';
1032	else
1033		(void) strncpy(buf, spa->spa_root, buflen);
1034}
1035
1036int
1037spa_sync_pass(spa_t *spa)
1038{
1039	return (spa->spa_sync_pass);
1040}
1041
1042char *
1043spa_name(spa_t *spa)
1044{
1045	/*
1046	 * Accessing the name requires holding either the namespace lock or the
1047	 * config lock, both of which are required to do a rename.
1048	 */
1049	ASSERT(MUTEX_HELD(&spa_namespace_lock) ||
1050	    spa_config_held(spa, RW_READER));
1051
1052	return (spa->spa_name);
1053}
1054
1055uint64_t
1056spa_guid(spa_t *spa)
1057{
1058	/*
1059	 * If we fail to parse the config during spa_load(), we can go through
1060	 * the error path (which posts an ereport) and end up here with no root
1061	 * vdev.  We stash the original pool guid in 'spa_load_guid' to handle
1062	 * this case.
1063	 */
1064	if (spa->spa_root_vdev != NULL)
1065		return (spa->spa_root_vdev->vdev_guid);
1066	else
1067		return (spa->spa_load_guid);
1068}
1069
1070uint64_t
1071spa_last_synced_txg(spa_t *spa)
1072{
1073	return (spa->spa_ubsync.ub_txg);
1074}
1075
1076uint64_t
1077spa_first_txg(spa_t *spa)
1078{
1079	return (spa->spa_first_txg);
1080}
1081
1082int
1083spa_state(spa_t *spa)
1084{
1085	return (spa->spa_state);
1086}
1087
1088uint64_t
1089spa_freeze_txg(spa_t *spa)
1090{
1091	return (spa->spa_freeze_txg);
1092}
1093
1094/*
1095 * Return how much space is allocated in the pool (ie. sum of all asize)
1096 */
1097uint64_t
1098spa_get_alloc(spa_t *spa)
1099{
1100	return (spa->spa_root_vdev->vdev_stat.vs_alloc);
1101}
1102
1103/*
1104 * Return how much (raid-z inflated) space there is in the pool.
1105 */
1106uint64_t
1107spa_get_space(spa_t *spa)
1108{
1109	return (spa->spa_root_vdev->vdev_stat.vs_space);
1110}
1111
1112/*
1113 * Return the amount of raid-z-deflated space in the pool.
1114 */
1115uint64_t
1116spa_get_dspace(spa_t *spa)
1117{
1118	if (spa->spa_deflate)
1119		return (spa->spa_root_vdev->vdev_stat.vs_dspace);
1120	else
1121		return (spa->spa_root_vdev->vdev_stat.vs_space);
1122}
1123
1124/* ARGSUSED */
1125uint64_t
1126spa_get_asize(spa_t *spa, uint64_t lsize)
1127{
1128	/*
1129	 * For now, the worst case is 512-byte RAID-Z blocks, in which
1130	 * case the space requirement is exactly 2x; so just assume that.
1131	 * Add to this the fact that we can have up to 3 DVAs per bp, and
1132	 * we have to multiply by a total of 6x.
1133	 */
1134	return (lsize * 6);
1135}
1136
1137/*
1138 * Return the failure mode that has been set to this pool. The default
1139 * behavior will be to block all I/Os when a complete failure occurs.
1140 */
1141uint8_t
1142spa_get_failmode(spa_t *spa)
1143{
1144	return (spa->spa_failmode);
1145}
1146
1147uint64_t
1148spa_version(spa_t *spa)
1149{
1150	return (spa->spa_ubsync.ub_version);
1151}
1152
1153int
1154spa_max_replication(spa_t *spa)
1155{
1156	/*
1157	 * As of SPA_VERSION == SPA_VERSION_DITTO_BLOCKS, we are able to
1158	 * handle BPs with more than one DVA allocated.  Set our max
1159	 * replication level accordingly.
1160	 */
1161	if (spa_version(spa) < SPA_VERSION_DITTO_BLOCKS)
1162		return (1);
1163	return (MIN(SPA_DVAS_PER_BP, spa_max_replication_override));
1164}
1165
1166uint64_t
1167bp_get_dasize(spa_t *spa, const blkptr_t *bp)
1168{
1169	int sz = 0, i;
1170
1171	if (!spa->spa_deflate)
1172		return (BP_GET_ASIZE(bp));
1173
1174	spa_config_enter(spa, RW_READER, FTAG);
1175	for (i = 0; i < SPA_DVAS_PER_BP; i++) {
1176		vdev_t *vd =
1177		    vdev_lookup_top(spa, DVA_GET_VDEV(&bp->blk_dva[i]));
1178		if (vd)
1179			sz += (DVA_GET_ASIZE(&bp->blk_dva[i]) >>
1180			    SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio;
1181	}
1182	spa_config_exit(spa, FTAG);
1183	return (sz);
1184}
1185
1186/*
1187 * ==========================================================================
1188 * Initialization and Termination
1189 * ==========================================================================
1190 */
1191
1192static int
1193spa_name_compare(const void *a1, const void *a2)
1194{
1195	const spa_t *s1 = a1;
1196	const spa_t *s2 = a2;
1197	int s;
1198
1199	s = strcmp(s1->spa_name, s2->spa_name);
1200	if (s > 0)
1201		return (1);
1202	if (s < 0)
1203		return (-1);
1204	return (0);
1205}
1206
1207int
1208spa_busy(void)
1209{
1210	return (spa_active_count);
1211}
1212
1213void
1214spa_init(int mode)
1215{
1216	mutex_init(&spa_namespace_lock, NULL, MUTEX_DEFAULT, NULL);
1217	mutex_init(&spa_spare_lock, NULL, MUTEX_DEFAULT, NULL);
1218	mutex_init(&spa_l2cache_lock, NULL, MUTEX_DEFAULT, NULL);
1219	cv_init(&spa_namespace_cv, NULL, CV_DEFAULT, NULL);
1220
1221	avl_create(&spa_namespace_avl, spa_name_compare, sizeof (spa_t),
1222	    offsetof(spa_t, spa_avl));
1223
1224	avl_create(&spa_spare_avl, spa_spare_compare, sizeof (spa_aux_t),
1225	    offsetof(spa_aux_t, aux_avl));
1226
1227	avl_create(&spa_l2cache_avl, spa_l2cache_compare, sizeof (spa_aux_t),
1228	    offsetof(spa_aux_t, aux_avl));
1229
1230	spa_mode = mode;
1231
1232	refcount_init();
1233	unique_init();
1234	zio_init();
1235	dmu_init();
1236	zil_init();
1237	zfs_prop_init();
1238	zpool_prop_init();
1239	spa_config_load();
1240}
1241
1242void
1243spa_fini(void)
1244{
1245	spa_evict_all();
1246
1247	zil_fini();
1248	dmu_fini();
1249	zio_fini();
1250	unique_fini();
1251	refcount_fini();
1252
1253	avl_destroy(&spa_namespace_avl);
1254	avl_destroy(&spa_spare_avl);
1255	avl_destroy(&spa_l2cache_avl);
1256
1257	cv_destroy(&spa_namespace_cv);
1258	mutex_destroy(&spa_namespace_lock);
1259	mutex_destroy(&spa_spare_lock);
1260	mutex_destroy(&spa_l2cache_lock);
1261}
1262
1263/*
1264 * Return whether this pool has slogs. No locking needed.
1265 * It's not a problem if the wrong answer is returned as it's only for
1266 * performance and not correctness
1267 */
1268boolean_t
1269spa_has_slogs(spa_t *spa)
1270{
1271	return (spa->spa_log_class->mc_rotor != NULL);
1272}
1273