spa.c revision 0373e76b3c3643df49ef3483e0f293fdea61d8c6
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 2006 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/*
29 * This file contains all the routines used when modifying on-disk SPA state.
30 * This includes opening, importing, destroying, exporting a pool, and syncing a
31 * pool.
32 */
33
34#include <sys/zfs_context.h>
35#include <sys/fm/fs/zfs.h>
36#include <sys/spa_impl.h>
37#include <sys/zio.h>
38#include <sys/zio_checksum.h>
39#include <sys/zio_compress.h>
40#include <sys/dmu.h>
41#include <sys/dmu_tx.h>
42#include <sys/zap.h>
43#include <sys/zil.h>
44#include <sys/vdev_impl.h>
45#include <sys/metaslab.h>
46#include <sys/uberblock_impl.h>
47#include <sys/txg.h>
48#include <sys/avl.h>
49#include <sys/dmu_traverse.h>
50#include <sys/unique.h>
51#include <sys/dsl_pool.h>
52#include <sys/dsl_dir.h>
53#include <sys/dsl_prop.h>
54#include <sys/fs/zfs.h>
55#include <sys/callb.h>
56
57/*
58 * ==========================================================================
59 * SPA state manipulation (open/create/destroy/import/export)
60 * ==========================================================================
61 */
62
63static int
64spa_error_entry_compare(const void *a, const void *b)
65{
66	spa_error_entry_t *sa = (spa_error_entry_t *)a;
67	spa_error_entry_t *sb = (spa_error_entry_t *)b;
68	int ret;
69
70	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
71	    sizeof (zbookmark_t));
72
73	if (ret < 0)
74		return (-1);
75	else if (ret > 0)
76		return (1);
77	else
78		return (0);
79}
80
81/*
82 * Utility function which retrieves copies of the current logs and
83 * re-initializes them in the process.
84 */
85void
86spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
87{
88	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
89
90	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
91	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
92
93	avl_create(&spa->spa_errlist_scrub,
94	    spa_error_entry_compare, sizeof (spa_error_entry_t),
95	    offsetof(spa_error_entry_t, se_avl));
96	avl_create(&spa->spa_errlist_last,
97	    spa_error_entry_compare, sizeof (spa_error_entry_t),
98	    offsetof(spa_error_entry_t, se_avl));
99}
100
101/*
102 * Activate an uninitialized pool.
103 */
104static void
105spa_activate(spa_t *spa)
106{
107	int t;
108
109	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
110
111	spa->spa_state = POOL_STATE_ACTIVE;
112
113	spa->spa_normal_class = metaslab_class_create();
114
115	for (t = 0; t < ZIO_TYPES; t++) {
116		spa->spa_zio_issue_taskq[t] = taskq_create("spa_zio_issue",
117		    8, maxclsyspri, 50, INT_MAX,
118		    TASKQ_PREPOPULATE);
119		spa->spa_zio_intr_taskq[t] = taskq_create("spa_zio_intr",
120		    8, maxclsyspri, 50, INT_MAX,
121		    TASKQ_PREPOPULATE);
122	}
123
124	rw_init(&spa->spa_traverse_lock, NULL, RW_DEFAULT, NULL);
125
126	list_create(&spa->spa_dirty_list, sizeof (vdev_t),
127	    offsetof(vdev_t, vdev_dirty_node));
128
129	txg_list_create(&spa->spa_vdev_txg_list,
130	    offsetof(struct vdev, vdev_txg_node));
131
132	avl_create(&spa->spa_errlist_scrub,
133	    spa_error_entry_compare, sizeof (spa_error_entry_t),
134	    offsetof(spa_error_entry_t, se_avl));
135	avl_create(&spa->spa_errlist_last,
136	    spa_error_entry_compare, sizeof (spa_error_entry_t),
137	    offsetof(spa_error_entry_t, se_avl));
138}
139
140/*
141 * Opposite of spa_activate().
142 */
143static void
144spa_deactivate(spa_t *spa)
145{
146	int t;
147
148	ASSERT(spa->spa_sync_on == B_FALSE);
149	ASSERT(spa->spa_dsl_pool == NULL);
150	ASSERT(spa->spa_root_vdev == NULL);
151
152	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
153
154	txg_list_destroy(&spa->spa_vdev_txg_list);
155
156	list_destroy(&spa->spa_dirty_list);
157
158	rw_destroy(&spa->spa_traverse_lock);
159
160	for (t = 0; t < ZIO_TYPES; t++) {
161		taskq_destroy(spa->spa_zio_issue_taskq[t]);
162		taskq_destroy(spa->spa_zio_intr_taskq[t]);
163		spa->spa_zio_issue_taskq[t] = NULL;
164		spa->spa_zio_intr_taskq[t] = NULL;
165	}
166
167	metaslab_class_destroy(spa->spa_normal_class);
168	spa->spa_normal_class = NULL;
169
170	/*
171	 * If this was part of an import or the open otherwise failed, we may
172	 * still have errors left in the queues.  Empty them just in case.
173	 */
174	spa_errlog_drain(spa);
175
176	avl_destroy(&spa->spa_errlist_scrub);
177	avl_destroy(&spa->spa_errlist_last);
178
179	spa->spa_state = POOL_STATE_UNINITIALIZED;
180}
181
182/*
183 * Verify a pool configuration, and construct the vdev tree appropriately.  This
184 * will create all the necessary vdevs in the appropriate layout, with each vdev
185 * in the CLOSED state.  This will prep the pool before open/creation/import.
186 * All vdev validation is done by the vdev_alloc() routine.
187 */
188static vdev_t *
189spa_config_parse(spa_t *spa, nvlist_t *nv, vdev_t *parent, uint_t id, int atype)
190{
191	nvlist_t **child;
192	uint_t c, children;
193	vdev_t *vd;
194
195	if ((vd = vdev_alloc(spa, nv, parent, id, atype)) == NULL)
196		return (NULL);
197
198	if (vd->vdev_ops->vdev_op_leaf)
199		return (vd);
200
201	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
202	    &child, &children) != 0) {
203		vdev_free(vd);
204		return (NULL);
205	}
206
207	for (c = 0; c < children; c++) {
208		if (spa_config_parse(spa, child[c], vd, c, atype) == NULL) {
209			vdev_free(vd);
210			return (NULL);
211		}
212	}
213
214	return (vd);
215}
216
217/*
218 * Opposite of spa_load().
219 */
220static void
221spa_unload(spa_t *spa)
222{
223	/*
224	 * Stop async tasks.
225	 */
226	spa_async_suspend(spa);
227
228	/*
229	 * Stop syncing.
230	 */
231	if (spa->spa_sync_on) {
232		txg_sync_stop(spa->spa_dsl_pool);
233		spa->spa_sync_on = B_FALSE;
234	}
235
236	/*
237	 * Wait for any outstanding prefetch I/O to complete.
238	 */
239	spa_config_enter(spa, RW_WRITER, FTAG);
240	spa_config_exit(spa, FTAG);
241
242	/*
243	 * Close the dsl pool.
244	 */
245	if (spa->spa_dsl_pool) {
246		dsl_pool_close(spa->spa_dsl_pool);
247		spa->spa_dsl_pool = NULL;
248	}
249
250	/*
251	 * Close all vdevs.
252	 */
253	if (spa->spa_root_vdev)
254		vdev_free(spa->spa_root_vdev);
255	ASSERT(spa->spa_root_vdev == NULL);
256
257	spa->spa_async_suspended = 0;
258}
259
260/*
261 * Load an existing storage pool, using the pool's builtin spa_config as a
262 * source of configuration information.
263 */
264static int
265spa_load(spa_t *spa, nvlist_t *config, spa_load_state_t state, int mosconfig)
266{
267	int error = 0;
268	nvlist_t *nvroot = NULL;
269	vdev_t *rvd;
270	uberblock_t *ub = &spa->spa_uberblock;
271	uint64_t config_cache_txg = spa->spa_config_txg;
272	uint64_t pool_guid;
273	zio_t *zio;
274
275	spa->spa_load_state = state;
276
277	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) ||
278	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid) ||
279	    (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
280	    &spa->spa_config_txg) && mosconfig)) {
281		error = EINVAL;
282		goto out;
283	}
284
285	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
286	    spa_guid_exists(pool_guid, 0)) {
287		error = EEXIST;
288		goto out;
289	}
290
291	/*
292	 * Parse the configuration into a vdev tree.
293	 */
294	spa_config_enter(spa, RW_WRITER, FTAG);
295	rvd = spa_config_parse(spa, nvroot, NULL, 0, VDEV_ALLOC_LOAD);
296	spa_config_exit(spa, FTAG);
297
298	if (rvd == NULL) {
299		error = EINVAL;
300		goto out;
301	}
302
303	ASSERT(spa->spa_root_vdev == rvd);
304	ASSERT(spa_guid(spa) == pool_guid);
305
306	/*
307	 * Try to open all vdevs, loading each label in the process.
308	 */
309	if (vdev_open(rvd) != 0) {
310		error = ENXIO;
311		goto out;
312	}
313
314	/*
315	 * Find the best uberblock.
316	 */
317	bzero(ub, sizeof (uberblock_t));
318
319	zio = zio_root(spa, NULL, NULL,
320	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
321	vdev_uberblock_load(zio, rvd, ub);
322	error = zio_wait(zio);
323
324	/*
325	 * If we weren't able to find a single valid uberblock, return failure.
326	 */
327	if (ub->ub_txg == 0) {
328		error = ENXIO;
329		goto out;
330	}
331
332	/*
333	 * If the pool is newer than the code, we can't open it.
334	 */
335	if (ub->ub_version > UBERBLOCK_VERSION) {
336		error = ENOTSUP;
337		goto out;
338	}
339
340	/*
341	 * If the vdev guid sum doesn't match the uberblock, we have an
342	 * incomplete configuration.
343	 */
344	if (rvd->vdev_guid_sum != ub->ub_guid_sum && (mosconfig ||
345	    state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT)) {
346		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
347		    VDEV_AUX_BAD_GUID_SUM);
348		error = ENXIO;
349		goto out;
350	}
351
352	/*
353	 * Initialize internal SPA structures.
354	 */
355	spa->spa_state = POOL_STATE_ACTIVE;
356	spa->spa_ubsync = spa->spa_uberblock;
357	spa->spa_first_txg = spa_last_synced_txg(spa) + 1;
358	error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
359	if (error) {
360		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
361		    VDEV_AUX_CORRUPT_DATA);
362		goto out;
363	}
364	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
365
366	if (zap_lookup(spa->spa_meta_objset,
367	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
368	    sizeof (uint64_t), 1, &spa->spa_config_object) != 0) {
369		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
370		    VDEV_AUX_CORRUPT_DATA);
371		error = EIO;
372		goto out;
373	}
374
375	if (!mosconfig) {
376		dmu_buf_t *db;
377		char *packed = NULL;
378		size_t nvsize = 0;
379		nvlist_t *newconfig = NULL;
380
381		VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset,
382		    spa->spa_config_object, FTAG, &db));
383		nvsize = *(uint64_t *)db->db_data;
384		dmu_buf_rele(db, FTAG);
385
386		packed = kmem_alloc(nvsize, KM_SLEEP);
387		error = dmu_read(spa->spa_meta_objset,
388		    spa->spa_config_object, 0, nvsize, packed);
389		if (error == 0)
390			error = nvlist_unpack(packed, nvsize, &newconfig, 0);
391		kmem_free(packed, nvsize);
392
393		if (error) {
394			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
395			    VDEV_AUX_CORRUPT_DATA);
396			error = EIO;
397			goto out;
398		}
399
400		spa_config_set(spa, newconfig);
401
402		spa_unload(spa);
403		spa_deactivate(spa);
404		spa_activate(spa);
405
406		return (spa_load(spa, newconfig, state, B_TRUE));
407	}
408
409	if (zap_lookup(spa->spa_meta_objset,
410	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
411	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj) != 0) {
412		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
413		    VDEV_AUX_CORRUPT_DATA);
414		error = EIO;
415		goto out;
416	}
417
418	/*
419	 * Load the persistent error log.  If we have an older pool, this will
420	 * not be present.
421	 */
422	error = zap_lookup(spa->spa_meta_objset,
423	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_LAST,
424	    sizeof (uint64_t), 1, &spa->spa_errlog_last);
425	if (error != 0 &&error != ENOENT) {
426		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
427		    VDEV_AUX_CORRUPT_DATA);
428		error = EIO;
429		goto out;
430	}
431
432	error = zap_lookup(spa->spa_meta_objset,
433	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_SCRUB,
434	    sizeof (uint64_t), 1, &spa->spa_errlog_scrub);
435	if (error != 0 && error != ENOENT) {
436		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
437		    VDEV_AUX_CORRUPT_DATA);
438		error = EIO;
439		goto out;
440	}
441
442	/*
443	 * Load the vdev state for all top level vdevs.  We need to grab the
444	 * config lock because all label I/O is done with the
445	 * ZIO_FLAG_CONFIG_HELD flag.
446	 */
447	spa_config_enter(spa, RW_READER, FTAG);
448	error = vdev_load(rvd);
449	spa_config_exit(spa, FTAG);
450
451	if (error)
452		goto out;
453
454	/*
455	 * Propagate the leaf DTLs we just loaded all the way up the tree.
456	 */
457	spa_config_enter(spa, RW_WRITER, FTAG);
458	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
459	spa_config_exit(spa, FTAG);
460
461	/*
462	 * Check the state of the root vdev.  If it can't be opened, it
463	 * indicates one or more toplevel vdevs are faulted.
464	 */
465	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
466		error = ENXIO;
467		goto out;
468	}
469
470	if ((spa_mode & FWRITE) && state != SPA_LOAD_TRYIMPORT) {
471		dmu_tx_t *tx;
472		int need_update = B_FALSE;
473		int c;
474
475		/*
476		 * Claim log blocks that haven't been committed yet.
477		 * This must all happen in a single txg.
478		 */
479		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
480		    spa_first_txg(spa));
481		dmu_objset_find(spa->spa_name, zil_claim, tx, 0);
482		dmu_tx_commit(tx);
483
484		spa->spa_sync_on = B_TRUE;
485		txg_sync_start(spa->spa_dsl_pool);
486
487		/*
488		 * Wait for all claims to sync.
489		 */
490		txg_wait_synced(spa->spa_dsl_pool, 0);
491
492		/*
493		 * If the config cache is stale, or we have uninitialized
494		 * metaslabs (see spa_vdev_add()), then update the config.
495		 */
496		if (config_cache_txg != spa->spa_config_txg ||
497		    state == SPA_LOAD_IMPORT)
498			need_update = B_TRUE;
499
500		for (c = 0; c < rvd->vdev_children; c++)
501			if (rvd->vdev_child[c]->vdev_ms_array == 0)
502				need_update = B_TRUE;
503
504		/*
505		 * Update the config cache asychronously in case we're the
506		 * root pool, in which case the config cache isn't writable yet.
507		 */
508		if (need_update)
509			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
510	}
511
512	error = 0;
513out:
514	if (error)
515		zfs_ereport_post(FM_EREPORT_ZFS_POOL, spa, NULL, NULL, 0, 0);
516	spa->spa_load_state = SPA_LOAD_NONE;
517	spa->spa_ena = 0;
518
519	return (error);
520}
521
522/*
523 * Pool Open/Import
524 *
525 * The import case is identical to an open except that the configuration is sent
526 * down from userland, instead of grabbed from the configuration cache.  For the
527 * case of an open, the pool configuration will exist in the
528 * POOL_STATE_UNITIALIZED state.
529 *
530 * The stats information (gen/count/ustats) is used to gather vdev statistics at
531 * the same time open the pool, without having to keep around the spa_t in some
532 * ambiguous state.
533 */
534static int
535spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t **config)
536{
537	spa_t *spa;
538	int error;
539	int loaded = B_FALSE;
540	int locked = B_FALSE;
541
542	*spapp = NULL;
543
544	/*
545	 * As disgusting as this is, we need to support recursive calls to this
546	 * function because dsl_dir_open() is called during spa_load(), and ends
547	 * up calling spa_open() again.  The real fix is to figure out how to
548	 * avoid dsl_dir_open() calling this in the first place.
549	 */
550	if (mutex_owner(&spa_namespace_lock) != curthread) {
551		mutex_enter(&spa_namespace_lock);
552		locked = B_TRUE;
553	}
554
555	if ((spa = spa_lookup(pool)) == NULL) {
556		if (locked)
557			mutex_exit(&spa_namespace_lock);
558		return (ENOENT);
559	}
560	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
561
562		spa_activate(spa);
563
564		error = spa_load(spa, spa->spa_config, SPA_LOAD_OPEN, B_FALSE);
565
566		if (error == EBADF) {
567			/*
568			 * If vdev_load() returns EBADF, it indicates that one
569			 * of the vdevs indicates that the pool has been
570			 * exported or destroyed.  If this is the case, the
571			 * config cache is out of sync and we should remove the
572			 * pool from the namespace.
573			 */
574			spa_unload(spa);
575			spa_deactivate(spa);
576			spa_remove(spa);
577			spa_config_sync();
578			if (locked)
579				mutex_exit(&spa_namespace_lock);
580			return (ENOENT);
581		}
582
583		if (error) {
584			/*
585			 * We can't open the pool, but we still have useful
586			 * information: the state of each vdev after the
587			 * attempted vdev_open().  Return this to the user.
588			 */
589			if (config != NULL && spa->spa_root_vdev != NULL) {
590				spa_config_enter(spa, RW_READER, FTAG);
591				*config = spa_config_generate(spa, NULL, -1ULL,
592				    B_TRUE);
593				spa_config_exit(spa, FTAG);
594			}
595			spa_unload(spa);
596			spa_deactivate(spa);
597			spa->spa_last_open_failed = B_TRUE;
598			if (locked)
599				mutex_exit(&spa_namespace_lock);
600			*spapp = NULL;
601			return (error);
602		} else {
603			zfs_post_ok(spa, NULL);
604			spa->spa_last_open_failed = B_FALSE;
605		}
606
607		loaded = B_TRUE;
608	}
609
610	spa_open_ref(spa, tag);
611	if (locked)
612		mutex_exit(&spa_namespace_lock);
613
614	*spapp = spa;
615
616	if (config != NULL) {
617		spa_config_enter(spa, RW_READER, FTAG);
618		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
619		spa_config_exit(spa, FTAG);
620	}
621
622	/*
623	 * If we just loaded the pool, resilver anything that's out of date.
624	 */
625	if (loaded && (spa_mode & FWRITE))
626		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
627
628	return (0);
629}
630
631int
632spa_open(const char *name, spa_t **spapp, void *tag)
633{
634	return (spa_open_common(name, spapp, tag, NULL));
635}
636
637/*
638 * Lookup the given spa_t, incrementing the inject count in the process,
639 * preventing it from being exported or destroyed.
640 */
641spa_t *
642spa_inject_addref(char *name)
643{
644	spa_t *spa;
645
646	mutex_enter(&spa_namespace_lock);
647	if ((spa = spa_lookup(name)) == NULL) {
648		mutex_exit(&spa_namespace_lock);
649		return (NULL);
650	}
651	spa->spa_inject_ref++;
652	mutex_exit(&spa_namespace_lock);
653
654	return (spa);
655}
656
657void
658spa_inject_delref(spa_t *spa)
659{
660	mutex_enter(&spa_namespace_lock);
661	spa->spa_inject_ref--;
662	mutex_exit(&spa_namespace_lock);
663}
664
665int
666spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
667{
668	int error;
669	spa_t *spa;
670
671	*config = NULL;
672	error = spa_open_common(name, &spa, FTAG, config);
673
674	if (spa && *config != NULL)
675		VERIFY(nvlist_add_uint64(*config, ZPOOL_CONFIG_ERRCOUNT,
676		    spa_get_errlog_size(spa)) == 0);
677
678	/*
679	 * We want to get the alternate root even for faulted pools, so we cheat
680	 * and call spa_lookup() directly.
681	 */
682	if (altroot) {
683		if (spa == NULL) {
684			mutex_enter(&spa_namespace_lock);
685			spa = spa_lookup(name);
686			if (spa)
687				spa_altroot(spa, altroot, buflen);
688			else
689				altroot[0] = '\0';
690			spa = NULL;
691			mutex_exit(&spa_namespace_lock);
692		} else {
693			spa_altroot(spa, altroot, buflen);
694		}
695	}
696
697	if (spa != NULL)
698		spa_close(spa, FTAG);
699
700	return (error);
701}
702
703/*
704 * Pool Creation
705 */
706int
707spa_create(const char *pool, nvlist_t *nvroot, const char *altroot)
708{
709	spa_t *spa;
710	vdev_t *rvd;
711	dsl_pool_t *dp;
712	dmu_tx_t *tx;
713	int c, error;
714	uint64_t txg = TXG_INITIAL;
715
716	/*
717	 * If this pool already exists, return failure.
718	 */
719	mutex_enter(&spa_namespace_lock);
720	if (spa_lookup(pool) != NULL) {
721		mutex_exit(&spa_namespace_lock);
722		return (EEXIST);
723	}
724
725	/*
726	 * Allocate a new spa_t structure.
727	 */
728	spa = spa_add(pool, altroot);
729	spa_activate(spa);
730
731	spa->spa_uberblock.ub_txg = txg - 1;
732	spa->spa_ubsync = spa->spa_uberblock;
733
734	/*
735	 * Create the root vdev.
736	 */
737	spa_config_enter(spa, RW_WRITER, FTAG);
738
739	rvd = spa_config_parse(spa, nvroot, NULL, 0, VDEV_ALLOC_ADD);
740
741	ASSERT(spa->spa_root_vdev == rvd);
742
743	if (rvd == NULL) {
744		error = EINVAL;
745	} else {
746		if ((error = vdev_create(rvd, txg)) == 0) {
747			for (c = 0; c < rvd->vdev_children; c++)
748				vdev_init(rvd->vdev_child[c], txg);
749			vdev_config_dirty(rvd);
750		}
751	}
752
753	spa_config_exit(spa, FTAG);
754
755	if (error) {
756		spa_unload(spa);
757		spa_deactivate(spa);
758		spa_remove(spa);
759		mutex_exit(&spa_namespace_lock);
760		return (error);
761	}
762
763	spa->spa_dsl_pool = dp = dsl_pool_create(spa, txg);
764	spa->spa_meta_objset = dp->dp_meta_objset;
765
766	tx = dmu_tx_create_assigned(dp, txg);
767
768	/*
769	 * Create the pool config object.
770	 */
771	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
772	    DMU_OT_PACKED_NVLIST, 1 << 14,
773	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
774
775	if (zap_add(spa->spa_meta_objset,
776	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
777	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
778		cmn_err(CE_PANIC, "failed to add pool config");
779	}
780
781	/*
782	 * Create the deferred-free bplist object.  Turn off compression
783	 * because sync-to-convergence takes longer if the blocksize
784	 * keeps changing.
785	 */
786	spa->spa_sync_bplist_obj = bplist_create(spa->spa_meta_objset,
787	    1 << 14, tx);
788	dmu_object_set_compress(spa->spa_meta_objset, spa->spa_sync_bplist_obj,
789	    ZIO_COMPRESS_OFF, tx);
790
791	if (zap_add(spa->spa_meta_objset,
792	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
793	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj, tx) != 0) {
794		cmn_err(CE_PANIC, "failed to add bplist");
795	}
796
797	dmu_tx_commit(tx);
798
799	spa->spa_sync_on = B_TRUE;
800	txg_sync_start(spa->spa_dsl_pool);
801
802	/*
803	 * We explicitly wait for the first transaction to complete so that our
804	 * bean counters are appropriately updated.
805	 */
806	txg_wait_synced(spa->spa_dsl_pool, txg);
807
808	spa_config_sync();
809
810	mutex_exit(&spa_namespace_lock);
811
812	return (0);
813}
814
815/*
816 * Import the given pool into the system.  We set up the necessary spa_t and
817 * then call spa_load() to do the dirty work.
818 */
819int
820spa_import(const char *pool, nvlist_t *config, const char *altroot)
821{
822	spa_t *spa;
823	int error;
824
825	if (!(spa_mode & FWRITE))
826		return (EROFS);
827
828	/*
829	 * If a pool with this name exists, return failure.
830	 */
831	mutex_enter(&spa_namespace_lock);
832	if (spa_lookup(pool) != NULL) {
833		mutex_exit(&spa_namespace_lock);
834		return (EEXIST);
835	}
836
837	/*
838	 * Create and initialize the spa structure.
839	 */
840	spa = spa_add(pool, altroot);
841	spa_activate(spa);
842
843	/*
844	 * Pass off the heavy lifting to spa_load().
845	 */
846	error = spa_load(spa, config, SPA_LOAD_IMPORT, B_FALSE);
847
848	if (error) {
849		spa_unload(spa);
850		spa_deactivate(spa);
851		spa_remove(spa);
852		mutex_exit(&spa_namespace_lock);
853		return (error);
854	}
855
856	/*
857	 * Update the config cache to include the newly-imported pool.
858	 */
859	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
860
861	mutex_exit(&spa_namespace_lock);
862
863	/*
864	 * Resilver anything that's out of date.
865	 */
866	if (spa_mode & FWRITE)
867		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
868
869	return (0);
870}
871
872/*
873 * This (illegal) pool name is used when temporarily importing a spa_t in order
874 * to get the vdev stats associated with the imported devices.
875 */
876#define	TRYIMPORT_NAME	"$import"
877
878nvlist_t *
879spa_tryimport(nvlist_t *tryconfig)
880{
881	nvlist_t *config = NULL;
882	char *poolname;
883	spa_t *spa;
884	uint64_t state;
885
886	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
887		return (NULL);
888
889	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
890		return (NULL);
891
892	/*
893	 * Create and initialize the spa structure.
894	 */
895	mutex_enter(&spa_namespace_lock);
896	spa = spa_add(TRYIMPORT_NAME, NULL);
897	spa_activate(spa);
898
899	/*
900	 * Pass off the heavy lifting to spa_load().
901	 */
902	(void) spa_load(spa, tryconfig, SPA_LOAD_TRYIMPORT, B_FALSE);
903
904	/*
905	 * If 'tryconfig' was at least parsable, return the current config.
906	 */
907	if (spa->spa_root_vdev != NULL) {
908		spa_config_enter(spa, RW_READER, FTAG);
909		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
910		spa_config_exit(spa, FTAG);
911		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
912		    poolname) == 0);
913		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
914		    state) == 0);
915	}
916
917	spa_unload(spa);
918	spa_deactivate(spa);
919	spa_remove(spa);
920	mutex_exit(&spa_namespace_lock);
921
922	return (config);
923}
924
925/*
926 * Pool export/destroy
927 *
928 * The act of destroying or exporting a pool is very simple.  We make sure there
929 * is no more pending I/O and any references to the pool are gone.  Then, we
930 * update the pool state and sync all the labels to disk, removing the
931 * configuration from the cache afterwards.
932 */
933static int
934spa_export_common(char *pool, int new_state)
935{
936	spa_t *spa;
937
938	if (!(spa_mode & FWRITE))
939		return (EROFS);
940
941	mutex_enter(&spa_namespace_lock);
942	if ((spa = spa_lookup(pool)) == NULL) {
943		mutex_exit(&spa_namespace_lock);
944		return (ENOENT);
945	}
946
947	/*
948	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
949	 * reacquire the namespace lock, and see if we can export.
950	 */
951	spa_open_ref(spa, FTAG);
952	mutex_exit(&spa_namespace_lock);
953	spa_async_suspend(spa);
954	mutex_enter(&spa_namespace_lock);
955	spa_close(spa, FTAG);
956
957	/*
958	 * The pool will be in core if it's openable,
959	 * in which case we can modify its state.
960	 */
961	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
962		/*
963		 * Objsets may be open only because they're dirty, so we
964		 * have to force it to sync before checking spa_refcnt.
965		 */
966		spa_scrub_suspend(spa);
967		txg_wait_synced(spa->spa_dsl_pool, 0);
968
969		/*
970		 * A pool cannot be exported or destroyed if there are active
971		 * references.  If we are resetting a pool, allow references by
972		 * fault injection handlers.
973		 */
974		if (!spa_refcount_zero(spa) ||
975		    (spa->spa_inject_ref != 0 &&
976		    new_state != POOL_STATE_UNINITIALIZED)) {
977			spa_scrub_resume(spa);
978			spa_async_resume(spa);
979			mutex_exit(&spa_namespace_lock);
980			return (EBUSY);
981		}
982
983		spa_scrub_resume(spa);
984		VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0);
985
986		/*
987		 * We want this to be reflected on every label,
988		 * so mark them all dirty.  spa_unload() will do the
989		 * final sync that pushes these changes out.
990		 */
991		if (new_state != POOL_STATE_UNINITIALIZED) {
992			spa_config_enter(spa, RW_WRITER, FTAG);
993			spa->spa_state = new_state;
994			spa->spa_final_txg = spa_last_synced_txg(spa) + 1;
995			vdev_config_dirty(spa->spa_root_vdev);
996			spa_config_exit(spa, FTAG);
997		}
998	}
999
1000	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
1001		spa_unload(spa);
1002		spa_deactivate(spa);
1003	}
1004
1005	if (new_state != POOL_STATE_UNINITIALIZED) {
1006		spa_remove(spa);
1007		spa_config_sync();
1008	}
1009	mutex_exit(&spa_namespace_lock);
1010
1011	return (0);
1012}
1013
1014/*
1015 * Destroy a storage pool.
1016 */
1017int
1018spa_destroy(char *pool)
1019{
1020	return (spa_export_common(pool, POOL_STATE_DESTROYED));
1021}
1022
1023/*
1024 * Export a storage pool.
1025 */
1026int
1027spa_export(char *pool)
1028{
1029	return (spa_export_common(pool, POOL_STATE_EXPORTED));
1030}
1031
1032/*
1033 * Similar to spa_export(), this unloads the spa_t without actually removing it
1034 * from the namespace in any way.
1035 */
1036int
1037spa_reset(char *pool)
1038{
1039	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED));
1040}
1041
1042
1043/*
1044 * ==========================================================================
1045 * Device manipulation
1046 * ==========================================================================
1047 */
1048
1049/*
1050 * Add capacity to a storage pool.
1051 */
1052int
1053spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
1054{
1055	uint64_t txg;
1056	int c, error;
1057	vdev_t *rvd = spa->spa_root_vdev;
1058	vdev_t *vd, *tvd;
1059
1060	txg = spa_vdev_enter(spa);
1061
1062	vd = spa_config_parse(spa, nvroot, NULL, 0, VDEV_ALLOC_ADD);
1063
1064	if (vd == NULL)
1065		return (spa_vdev_exit(spa, vd, txg, EINVAL));
1066
1067	if ((error = vdev_create(vd, txg)) != 0)
1068		return (spa_vdev_exit(spa, vd, txg, error));
1069
1070	/*
1071	 * Transfer each new top-level vdev from vd to rvd.
1072	 */
1073	for (c = 0; c < vd->vdev_children; c++) {
1074		tvd = vd->vdev_child[c];
1075		vdev_remove_child(vd, tvd);
1076		tvd->vdev_id = rvd->vdev_children;
1077		vdev_add_child(rvd, tvd);
1078		vdev_config_dirty(tvd);
1079	}
1080
1081	/*
1082	 * We have to be careful when adding new vdevs to an existing pool.
1083	 * If other threads start allocating from these vdevs before we
1084	 * sync the config cache, and we lose power, then upon reboot we may
1085	 * fail to open the pool because there are DVAs that the config cache
1086	 * can't translate.  Therefore, we first add the vdevs without
1087	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
1088	 * and then let spa_config_update() initialize the new metaslabs.
1089	 *
1090	 * spa_load() checks for added-but-not-initialized vdevs, so that
1091	 * if we lose power at any point in this sequence, the remaining
1092	 * steps will be completed the next time we load the pool.
1093	 */
1094	(void) spa_vdev_exit(spa, vd, txg, 0);
1095
1096	mutex_enter(&spa_namespace_lock);
1097	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
1098	mutex_exit(&spa_namespace_lock);
1099
1100	return (0);
1101}
1102
1103/*
1104 * Attach a device to a mirror.  The arguments are the path to any device
1105 * in the mirror, and the nvroot for the new device.  If the path specifies
1106 * a device that is not mirrored, we automatically insert the mirror vdev.
1107 *
1108 * If 'replacing' is specified, the new device is intended to replace the
1109 * existing device; in this case the two devices are made into their own
1110 * mirror using the 'replacing' vdev, which is functionally idendical to
1111 * the mirror vdev (it actually reuses all the same ops) but has a few
1112 * extra rules: you can't attach to it after it's been created, and upon
1113 * completion of resilvering, the first disk (the one being replaced)
1114 * is automatically detached.
1115 */
1116int
1117spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
1118{
1119	uint64_t txg, open_txg;
1120	int error;
1121	vdev_t *rvd = spa->spa_root_vdev;
1122	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
1123	vdev_ops_t *pvops = replacing ? &vdev_replacing_ops : &vdev_mirror_ops;
1124
1125	txg = spa_vdev_enter(spa);
1126
1127	oldvd = vdev_lookup_by_guid(rvd, guid);
1128
1129	if (oldvd == NULL)
1130		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
1131
1132	if (!oldvd->vdev_ops->vdev_op_leaf)
1133		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1134
1135	pvd = oldvd->vdev_parent;
1136
1137	/*
1138	 * The parent must be a mirror or the root, unless we're replacing;
1139	 * in that case, the parent can be anything but another replacing vdev.
1140	 */
1141	if (pvd->vdev_ops != &vdev_mirror_ops &&
1142	    pvd->vdev_ops != &vdev_root_ops &&
1143	    (!replacing || pvd->vdev_ops == &vdev_replacing_ops))
1144		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1145
1146	newrootvd = spa_config_parse(spa, nvroot, NULL, 0, VDEV_ALLOC_ADD);
1147
1148	if (newrootvd == NULL || newrootvd->vdev_children != 1)
1149		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
1150
1151	newvd = newrootvd->vdev_child[0];
1152
1153	if (!newvd->vdev_ops->vdev_op_leaf)
1154		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
1155
1156	if ((error = vdev_create(newrootvd, txg)) != 0)
1157		return (spa_vdev_exit(spa, newrootvd, txg, error));
1158
1159	/*
1160	 * Compare the new device size with the replaceable/attachable
1161	 * device size.
1162	 */
1163	if (newvd->vdev_psize < vdev_get_rsize(oldvd))
1164		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
1165
1166	if (newvd->vdev_ashift != oldvd->vdev_ashift && oldvd->vdev_ashift != 0)
1167		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
1168
1169	/*
1170	 * If this is an in-place replacement, update oldvd's path and devid
1171	 * to make it distinguishable from newvd, and unopenable from now on.
1172	 */
1173	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
1174		spa_strfree(oldvd->vdev_path);
1175		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
1176		    KM_SLEEP);
1177		(void) sprintf(oldvd->vdev_path, "%s/%s",
1178		    newvd->vdev_path, "old");
1179		if (oldvd->vdev_devid != NULL) {
1180			spa_strfree(oldvd->vdev_devid);
1181			oldvd->vdev_devid = NULL;
1182		}
1183	}
1184
1185	/*
1186	 * If the parent is not a mirror, or if we're replacing,
1187	 * insert the new mirror/replacing vdev above oldvd.
1188	 */
1189	if (pvd->vdev_ops != pvops)
1190		pvd = vdev_add_parent(oldvd, pvops);
1191
1192	ASSERT(pvd->vdev_top->vdev_parent == rvd);
1193	ASSERT(pvd->vdev_ops == pvops);
1194	ASSERT(oldvd->vdev_parent == pvd);
1195
1196	/*
1197	 * Extract the new device from its root and add it to pvd.
1198	 */
1199	vdev_remove_child(newrootvd, newvd);
1200	newvd->vdev_id = pvd->vdev_children;
1201	vdev_add_child(pvd, newvd);
1202
1203	/*
1204	 * If newvd is smaller than oldvd, but larger than its rsize,
1205	 * the addition of newvd may have decreased our parent's asize.
1206	 */
1207	pvd->vdev_asize = MIN(pvd->vdev_asize, newvd->vdev_asize);
1208
1209	tvd = newvd->vdev_top;
1210	ASSERT(pvd->vdev_top == tvd);
1211	ASSERT(tvd->vdev_parent == rvd);
1212
1213	vdev_config_dirty(tvd);
1214
1215	/*
1216	 * Set newvd's DTL to [TXG_INITIAL, open_txg].  It will propagate
1217	 * upward when spa_vdev_exit() calls vdev_dtl_reassess().
1218	 */
1219	open_txg = txg + TXG_CONCURRENT_STATES - 1;
1220
1221	mutex_enter(&newvd->vdev_dtl_lock);
1222	space_map_add(&newvd->vdev_dtl_map, TXG_INITIAL,
1223	    open_txg - TXG_INITIAL + 1);
1224	mutex_exit(&newvd->vdev_dtl_lock);
1225
1226	dprintf("attached %s in txg %llu\n", newvd->vdev_path, txg);
1227
1228	/*
1229	 * Mark newvd's DTL dirty in this txg.
1230	 */
1231	vdev_dirty(tvd, VDD_DTL, txg);
1232	(void) txg_list_add(&tvd->vdev_dtl_list, newvd, txg);
1233
1234	(void) spa_vdev_exit(spa, newrootvd, open_txg, 0);
1235
1236	/*
1237	 * Kick off a resilver to update newvd.
1238	 */
1239	VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
1240
1241	return (0);
1242}
1243
1244/*
1245 * Detach a device from a mirror or replacing vdev.
1246 * If 'replace_done' is specified, only detach if the parent
1247 * is a replacing vdev.
1248 */
1249int
1250spa_vdev_detach(spa_t *spa, uint64_t guid, int replace_done)
1251{
1252	uint64_t txg;
1253	int c, t, error;
1254	vdev_t *rvd = spa->spa_root_vdev;
1255	vdev_t *vd, *pvd, *cvd, *tvd;
1256
1257	txg = spa_vdev_enter(spa);
1258
1259	vd = vdev_lookup_by_guid(rvd, guid);
1260
1261	if (vd == NULL)
1262		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
1263
1264	if (!vd->vdev_ops->vdev_op_leaf)
1265		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1266
1267	pvd = vd->vdev_parent;
1268
1269	/*
1270	 * If replace_done is specified, only remove this device if it's
1271	 * the first child of a replacing vdev.
1272	 */
1273	if (replace_done &&
1274	    (vd->vdev_id != 0 || pvd->vdev_ops != &vdev_replacing_ops))
1275		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1276
1277	/*
1278	 * Only mirror and replacing vdevs support detach.
1279	 */
1280	if (pvd->vdev_ops != &vdev_replacing_ops &&
1281	    pvd->vdev_ops != &vdev_mirror_ops)
1282		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1283
1284	/*
1285	 * If there's only one replica, you can't detach it.
1286	 */
1287	if (pvd->vdev_children <= 1)
1288		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
1289
1290	/*
1291	 * If all siblings have non-empty DTLs, this device may have the only
1292	 * valid copy of the data, which means we cannot safely detach it.
1293	 *
1294	 * XXX -- as in the vdev_offline() case, we really want a more
1295	 * precise DTL check.
1296	 */
1297	for (c = 0; c < pvd->vdev_children; c++) {
1298		uint64_t dirty;
1299
1300		cvd = pvd->vdev_child[c];
1301		if (cvd == vd)
1302			continue;
1303		if (vdev_is_dead(cvd))
1304			continue;
1305		mutex_enter(&cvd->vdev_dtl_lock);
1306		dirty = cvd->vdev_dtl_map.sm_space |
1307		    cvd->vdev_dtl_scrub.sm_space;
1308		mutex_exit(&cvd->vdev_dtl_lock);
1309		if (!dirty)
1310			break;
1311	}
1312	if (c == pvd->vdev_children)
1313		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
1314
1315	/*
1316	 * Erase the disk labels so the disk can be used for other things.
1317	 * This must be done after all other error cases are handled,
1318	 * but before we disembowel vd (so we can still do I/O to it).
1319	 * But if we can't do it, don't treat the error as fatal --
1320	 * it may be that the unwritability of the disk is the reason
1321	 * it's being detached!
1322	 */
1323	error = vdev_label_init(vd, 0);
1324	if (error)
1325		dprintf("unable to erase labels on %s\n", vdev_description(vd));
1326
1327	/*
1328	 * Remove vd from its parent and compact the parent's children.
1329	 */
1330	vdev_remove_child(pvd, vd);
1331	vdev_compact_children(pvd);
1332
1333	/*
1334	 * Remember one of the remaining children so we can get tvd below.
1335	 */
1336	cvd = pvd->vdev_child[0];
1337
1338	/*
1339	 * If the parent mirror/replacing vdev only has one child,
1340	 * the parent is no longer needed.  Remove it from the tree.
1341	 */
1342	if (pvd->vdev_children == 1)
1343		vdev_remove_parent(cvd);
1344
1345	/*
1346	 * We don't set tvd until now because the parent we just removed
1347	 * may have been the previous top-level vdev.
1348	 */
1349	tvd = cvd->vdev_top;
1350	ASSERT(tvd->vdev_parent == rvd);
1351
1352	/*
1353	 * Reopen this top-level vdev to reassess health after detach.
1354	 */
1355	vdev_reopen(tvd);
1356
1357	/*
1358	 * If the device we just detached was smaller than the others,
1359	 * it may be possible to add metaslabs (i.e. grow the pool).  We ignore
1360	 * the error here because the detach still succeeded - we just weren't
1361	 * able to reinitialize the metaslabs.  This pool is in for a world of
1362	 * hurt, in any case.
1363	 */
1364	(void) vdev_metaslab_init(tvd, txg);
1365
1366	vdev_config_dirty(tvd);
1367
1368	/*
1369	 * Mark vd's DTL as dirty in this txg.
1370	 * vdev_dtl_sync() will see that vd->vdev_detached is set
1371	 * and free vd's DTL object in syncing context.
1372	 * But first make sure we're not on any *other* txg's DTL list,
1373	 * to prevent vd from being accessed after it's freed.
1374	 */
1375	vdev_dirty(tvd, VDD_DTL, txg);
1376	vd->vdev_detached = B_TRUE;
1377	for (t = 0; t < TXG_SIZE; t++)
1378		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
1379	(void) txg_list_add(&tvd->vdev_dtl_list, vd, txg);
1380
1381	dprintf("detached %s in txg %llu\n", vd->vdev_path, txg);
1382
1383	return (spa_vdev_exit(spa, vd, txg, 0));
1384}
1385
1386/*
1387 * Find any device that's done replacing, so we can detach it.
1388 */
1389static vdev_t *
1390spa_vdev_replace_done_hunt(vdev_t *vd)
1391{
1392	vdev_t *newvd, *oldvd;
1393	int c;
1394
1395	for (c = 0; c < vd->vdev_children; c++) {
1396		oldvd = spa_vdev_replace_done_hunt(vd->vdev_child[c]);
1397		if (oldvd != NULL)
1398			return (oldvd);
1399	}
1400
1401	if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
1402		oldvd = vd->vdev_child[0];
1403		newvd = vd->vdev_child[1];
1404
1405		mutex_enter(&newvd->vdev_dtl_lock);
1406		if (newvd->vdev_dtl_map.sm_space == 0 &&
1407		    newvd->vdev_dtl_scrub.sm_space == 0) {
1408			mutex_exit(&newvd->vdev_dtl_lock);
1409			return (oldvd);
1410		}
1411		mutex_exit(&newvd->vdev_dtl_lock);
1412	}
1413
1414	return (NULL);
1415}
1416
1417static void
1418spa_vdev_replace_done(spa_t *spa)
1419{
1420	vdev_t *vd;
1421	uint64_t guid;
1422
1423	spa_config_enter(spa, RW_READER, FTAG);
1424
1425	while ((vd = spa_vdev_replace_done_hunt(spa->spa_root_vdev)) != NULL) {
1426		guid = vd->vdev_guid;
1427		spa_config_exit(spa, FTAG);
1428		if (spa_vdev_detach(spa, guid, B_TRUE) != 0)
1429			return;
1430		spa_config_enter(spa, RW_READER, FTAG);
1431	}
1432
1433	spa_config_exit(spa, FTAG);
1434}
1435
1436/*
1437 * Update the stored path for this vdev.  Dirty the vdev configuration, relying
1438 * on spa_vdev_enter/exit() to synchronize the labels and cache.
1439 */
1440int
1441spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
1442{
1443	vdev_t *rvd, *vd;
1444	uint64_t txg;
1445
1446	rvd = spa->spa_root_vdev;
1447
1448	txg = spa_vdev_enter(spa);
1449
1450	if ((vd = vdev_lookup_by_guid(rvd, guid)) == NULL)
1451		return (spa_vdev_exit(spa, NULL, txg, ENOENT));
1452
1453	if (!vd->vdev_ops->vdev_op_leaf)
1454		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1455
1456	spa_strfree(vd->vdev_path);
1457	vd->vdev_path = spa_strdup(newpath);
1458
1459	vdev_config_dirty(vd->vdev_top);
1460
1461	return (spa_vdev_exit(spa, NULL, txg, 0));
1462}
1463
1464/*
1465 * ==========================================================================
1466 * SPA Scrubbing
1467 * ==========================================================================
1468 */
1469
1470void
1471spa_scrub_throttle(spa_t *spa, int direction)
1472{
1473	mutex_enter(&spa->spa_scrub_lock);
1474	spa->spa_scrub_throttled += direction;
1475	ASSERT(spa->spa_scrub_throttled >= 0);
1476	if (spa->spa_scrub_throttled == 0)
1477		cv_broadcast(&spa->spa_scrub_io_cv);
1478	mutex_exit(&spa->spa_scrub_lock);
1479}
1480
1481static void
1482spa_scrub_io_done(zio_t *zio)
1483{
1484	spa_t *spa = zio->io_spa;
1485
1486	zio_buf_free(zio->io_data, zio->io_size);
1487
1488	mutex_enter(&spa->spa_scrub_lock);
1489	if (zio->io_error && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
1490		vdev_t *vd = zio->io_vd;
1491		spa->spa_scrub_errors++;
1492		mutex_enter(&vd->vdev_stat_lock);
1493		vd->vdev_stat.vs_scrub_errors++;
1494		mutex_exit(&vd->vdev_stat_lock);
1495	}
1496	if (--spa->spa_scrub_inflight == 0) {
1497		cv_broadcast(&spa->spa_scrub_io_cv);
1498		ASSERT(spa->spa_scrub_throttled == 0);
1499	}
1500	mutex_exit(&spa->spa_scrub_lock);
1501}
1502
1503static void
1504spa_scrub_io_start(spa_t *spa, blkptr_t *bp, int priority, int flags,
1505    zbookmark_t *zb)
1506{
1507	size_t size = BP_GET_LSIZE(bp);
1508	void *data = zio_buf_alloc(size);
1509
1510	mutex_enter(&spa->spa_scrub_lock);
1511	spa->spa_scrub_inflight++;
1512	mutex_exit(&spa->spa_scrub_lock);
1513
1514	if (zb->zb_level == -1 && BP_GET_TYPE(bp) != DMU_OT_OBJSET)
1515		flags |= ZIO_FLAG_SPECULATIVE;	/* intent log block */
1516
1517	flags |= ZIO_FLAG_CANFAIL;
1518
1519	zio_nowait(zio_read(NULL, spa, bp, data, size,
1520	    spa_scrub_io_done, NULL, priority, flags, zb));
1521}
1522
1523/* ARGSUSED */
1524static int
1525spa_scrub_cb(traverse_blk_cache_t *bc, spa_t *spa, void *a)
1526{
1527	blkptr_t *bp = &bc->bc_blkptr;
1528	vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(&bp->blk_dva[0]));
1529
1530	if (bc->bc_errno || vd == NULL) {
1531		/*
1532		 * We can't scrub this block, but we can continue to scrub
1533		 * the rest of the pool.  Note the error and move along.
1534		 */
1535		mutex_enter(&spa->spa_scrub_lock);
1536		spa->spa_scrub_errors++;
1537		mutex_exit(&spa->spa_scrub_lock);
1538
1539		if (vd != NULL) {
1540			mutex_enter(&vd->vdev_stat_lock);
1541			vd->vdev_stat.vs_scrub_errors++;
1542			mutex_exit(&vd->vdev_stat_lock);
1543		}
1544
1545		return (ERESTART);
1546	}
1547
1548	ASSERT(bp->blk_birth < spa->spa_scrub_maxtxg);
1549
1550	/*
1551	 * Keep track of how much data we've examined so that
1552	 * zpool(1M) status can make useful progress reports.
1553	 */
1554	mutex_enter(&vd->vdev_stat_lock);
1555	vd->vdev_stat.vs_scrub_examined += BP_GET_ASIZE(bp);
1556	mutex_exit(&vd->vdev_stat_lock);
1557
1558	if (spa->spa_scrub_type == POOL_SCRUB_RESILVER) {
1559		if (DVA_GET_GANG(&bp->blk_dva[0])) {
1560			/*
1561			 * Gang members may be spread across multiple vdevs,
1562			 * so the best we can do is look at the pool-wide DTL.
1563			 * XXX -- it would be better to change our allocation
1564			 * policy to ensure that this can't happen.
1565			 */
1566			vd = spa->spa_root_vdev;
1567		}
1568		if (vdev_dtl_contains(&vd->vdev_dtl_map, bp->blk_birth, 1)) {
1569			spa_scrub_io_start(spa, bp, ZIO_PRIORITY_RESILVER,
1570			    ZIO_FLAG_RESILVER, &bc->bc_bookmark);
1571		}
1572	} else {
1573		spa_scrub_io_start(spa, bp, ZIO_PRIORITY_SCRUB,
1574		    ZIO_FLAG_SCRUB, &bc->bc_bookmark);
1575	}
1576
1577	return (0);
1578}
1579
1580static void
1581spa_scrub_thread(spa_t *spa)
1582{
1583	callb_cpr_t cprinfo;
1584	traverse_handle_t *th = spa->spa_scrub_th;
1585	vdev_t *rvd = spa->spa_root_vdev;
1586	pool_scrub_type_t scrub_type = spa->spa_scrub_type;
1587	int error = 0;
1588	boolean_t complete;
1589
1590	CALLB_CPR_INIT(&cprinfo, &spa->spa_scrub_lock, callb_generic_cpr, FTAG);
1591
1592	/*
1593	 * If we're restarting due to a snapshot create/delete,
1594	 * wait for that to complete.
1595	 */
1596	txg_wait_synced(spa_get_dsl(spa), 0);
1597
1598	dprintf("start %s mintxg=%llu maxtxg=%llu\n",
1599	    scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub",
1600	    spa->spa_scrub_mintxg, spa->spa_scrub_maxtxg);
1601
1602	spa_config_enter(spa, RW_WRITER, FTAG);
1603	vdev_reopen(rvd);		/* purge all vdev caches */
1604	vdev_config_dirty(rvd);		/* rewrite all disk labels */
1605	vdev_scrub_stat_update(rvd, scrub_type, B_FALSE);
1606	spa_config_exit(spa, FTAG);
1607
1608	mutex_enter(&spa->spa_scrub_lock);
1609	spa->spa_scrub_errors = 0;
1610	spa->spa_scrub_active = 1;
1611	ASSERT(spa->spa_scrub_inflight == 0);
1612	ASSERT(spa->spa_scrub_throttled == 0);
1613
1614	while (!spa->spa_scrub_stop) {
1615		CALLB_CPR_SAFE_BEGIN(&cprinfo);
1616		while (spa->spa_scrub_suspended) {
1617			spa->spa_scrub_active = 0;
1618			cv_broadcast(&spa->spa_scrub_cv);
1619			cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
1620			spa->spa_scrub_active = 1;
1621		}
1622		CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_scrub_lock);
1623
1624		if (spa->spa_scrub_restart_txg != 0)
1625			break;
1626
1627		mutex_exit(&spa->spa_scrub_lock);
1628		error = traverse_more(th);
1629		mutex_enter(&spa->spa_scrub_lock);
1630		if (error != EAGAIN)
1631			break;
1632
1633		while (spa->spa_scrub_throttled > 0)
1634			cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1635	}
1636
1637	while (spa->spa_scrub_inflight)
1638		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1639
1640	spa->spa_scrub_active = 0;
1641	cv_broadcast(&spa->spa_scrub_cv);
1642
1643	mutex_exit(&spa->spa_scrub_lock);
1644
1645	spa_config_enter(spa, RW_WRITER, FTAG);
1646
1647	mutex_enter(&spa->spa_scrub_lock);
1648
1649	/*
1650	 * Note: we check spa_scrub_restart_txg under both spa_scrub_lock
1651	 * AND the spa config lock to synchronize with any config changes
1652	 * that revise the DTLs under spa_vdev_enter() / spa_vdev_exit().
1653	 */
1654	if (spa->spa_scrub_restart_txg != 0)
1655		error = ERESTART;
1656
1657	if (spa->spa_scrub_stop)
1658		error = EINTR;
1659
1660	/*
1661	 * Even if there were uncorrectable errors, we consider the scrub
1662	 * completed.  The downside is that if there is a transient error during
1663	 * a resilver, we won't resilver the data properly to the target.  But
1664	 * if the damage is permanent (more likely) we will resilver forever,
1665	 * which isn't really acceptable.  Since there is enough information for
1666	 * the user to know what has failed and why, this seems like a more
1667	 * tractable approach.
1668	 */
1669	complete = (error == 0);
1670
1671	dprintf("end %s to maxtxg=%llu %s, traverse=%d, %llu errors, stop=%u\n",
1672	    scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub",
1673	    spa->spa_scrub_maxtxg, complete ? "done" : "FAILED",
1674	    error, spa->spa_scrub_errors, spa->spa_scrub_stop);
1675
1676	mutex_exit(&spa->spa_scrub_lock);
1677
1678	/*
1679	 * If the scrub/resilver completed, update all DTLs to reflect this.
1680	 * Whether it succeeded or not, vacate all temporary scrub DTLs.
1681	 */
1682	vdev_dtl_reassess(rvd, spa_last_synced_txg(spa) + 1,
1683	    complete ? spa->spa_scrub_maxtxg : 0, B_TRUE);
1684	vdev_scrub_stat_update(rvd, POOL_SCRUB_NONE, complete);
1685	spa_errlog_rotate(spa);
1686
1687	spa_config_exit(spa, FTAG);
1688
1689	mutex_enter(&spa->spa_scrub_lock);
1690
1691	/*
1692	 * We may have finished replacing a device.
1693	 * Let the async thread assess this and handle the detach.
1694	 */
1695	spa_async_request(spa, SPA_ASYNC_REPLACE_DONE);
1696
1697	/*
1698	 * If we were told to restart, our final act is to start a new scrub.
1699	 */
1700	if (error == ERESTART)
1701		spa_async_request(spa, scrub_type == POOL_SCRUB_RESILVER ?
1702		    SPA_ASYNC_RESILVER : SPA_ASYNC_SCRUB);
1703
1704	spa->spa_scrub_type = POOL_SCRUB_NONE;
1705	spa->spa_scrub_active = 0;
1706	spa->spa_scrub_thread = NULL;
1707	cv_broadcast(&spa->spa_scrub_cv);
1708	CALLB_CPR_EXIT(&cprinfo);	/* drops &spa->spa_scrub_lock */
1709	thread_exit();
1710}
1711
1712void
1713spa_scrub_suspend(spa_t *spa)
1714{
1715	mutex_enter(&spa->spa_scrub_lock);
1716	spa->spa_scrub_suspended++;
1717	while (spa->spa_scrub_active) {
1718		cv_broadcast(&spa->spa_scrub_cv);
1719		cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
1720	}
1721	while (spa->spa_scrub_inflight)
1722		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1723	mutex_exit(&spa->spa_scrub_lock);
1724}
1725
1726void
1727spa_scrub_resume(spa_t *spa)
1728{
1729	mutex_enter(&spa->spa_scrub_lock);
1730	ASSERT(spa->spa_scrub_suspended != 0);
1731	if (--spa->spa_scrub_suspended == 0)
1732		cv_broadcast(&spa->spa_scrub_cv);
1733	mutex_exit(&spa->spa_scrub_lock);
1734}
1735
1736void
1737spa_scrub_restart(spa_t *spa, uint64_t txg)
1738{
1739	/*
1740	 * Something happened (e.g. snapshot create/delete) that means
1741	 * we must restart any in-progress scrubs.  The itinerary will
1742	 * fix this properly.
1743	 */
1744	mutex_enter(&spa->spa_scrub_lock);
1745	spa->spa_scrub_restart_txg = txg;
1746	mutex_exit(&spa->spa_scrub_lock);
1747}
1748
1749int
1750spa_scrub(spa_t *spa, pool_scrub_type_t type, boolean_t force)
1751{
1752	space_seg_t *ss;
1753	uint64_t mintxg, maxtxg;
1754	vdev_t *rvd = spa->spa_root_vdev;
1755
1756	if ((uint_t)type >= POOL_SCRUB_TYPES)
1757		return (ENOTSUP);
1758
1759	mutex_enter(&spa->spa_scrub_lock);
1760
1761	/*
1762	 * If there's a scrub or resilver already in progress, stop it.
1763	 */
1764	while (spa->spa_scrub_thread != NULL) {
1765		/*
1766		 * Don't stop a resilver unless forced.
1767		 */
1768		if (spa->spa_scrub_type == POOL_SCRUB_RESILVER && !force) {
1769			mutex_exit(&spa->spa_scrub_lock);
1770			return (EBUSY);
1771		}
1772		spa->spa_scrub_stop = 1;
1773		cv_broadcast(&spa->spa_scrub_cv);
1774		cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
1775	}
1776
1777	/*
1778	 * Terminate the previous traverse.
1779	 */
1780	if (spa->spa_scrub_th != NULL) {
1781		traverse_fini(spa->spa_scrub_th);
1782		spa->spa_scrub_th = NULL;
1783	}
1784
1785	if (rvd == NULL) {
1786		ASSERT(spa->spa_scrub_stop == 0);
1787		ASSERT(spa->spa_scrub_type == type);
1788		ASSERT(spa->spa_scrub_restart_txg == 0);
1789		mutex_exit(&spa->spa_scrub_lock);
1790		return (0);
1791	}
1792
1793	mintxg = TXG_INITIAL - 1;
1794	maxtxg = spa_last_synced_txg(spa) + 1;
1795
1796	mutex_enter(&rvd->vdev_dtl_lock);
1797
1798	if (rvd->vdev_dtl_map.sm_space == 0) {
1799		/*
1800		 * The pool-wide DTL is empty.
1801		 * If this is a resilver, there's nothing to do.
1802		 */
1803		if (type == POOL_SCRUB_RESILVER)
1804			type = POOL_SCRUB_NONE;
1805	} else {
1806		/*
1807		 * The pool-wide DTL is non-empty.
1808		 * If this is a normal scrub, upgrade to a resilver instead.
1809		 */
1810		if (type == POOL_SCRUB_EVERYTHING)
1811			type = POOL_SCRUB_RESILVER;
1812	}
1813
1814	if (type == POOL_SCRUB_RESILVER) {
1815		/*
1816		 * Determine the resilvering boundaries.
1817		 *
1818		 * Note: (mintxg, maxtxg) is an open interval,
1819		 * i.e. mintxg and maxtxg themselves are not included.
1820		 *
1821		 * Note: for maxtxg, we MIN with spa_last_synced_txg(spa) + 1
1822		 * so we don't claim to resilver a txg that's still changing.
1823		 */
1824		ss = avl_first(&rvd->vdev_dtl_map.sm_root);
1825		mintxg = ss->ss_start - 1;
1826		ss = avl_last(&rvd->vdev_dtl_map.sm_root);
1827		maxtxg = MIN(ss->ss_end, maxtxg);
1828	}
1829
1830	mutex_exit(&rvd->vdev_dtl_lock);
1831
1832	spa->spa_scrub_stop = 0;
1833	spa->spa_scrub_type = type;
1834	spa->spa_scrub_restart_txg = 0;
1835
1836	if (type != POOL_SCRUB_NONE) {
1837		spa->spa_scrub_mintxg = mintxg;
1838		spa->spa_scrub_maxtxg = maxtxg;
1839		spa->spa_scrub_th = traverse_init(spa, spa_scrub_cb, NULL,
1840		    ADVANCE_PRE | ADVANCE_PRUNE | ADVANCE_ZIL,
1841		    ZIO_FLAG_CANFAIL);
1842		traverse_add_pool(spa->spa_scrub_th, mintxg, maxtxg);
1843		spa->spa_scrub_thread = thread_create(NULL, 0,
1844		    spa_scrub_thread, spa, 0, &p0, TS_RUN, minclsyspri);
1845	}
1846
1847	mutex_exit(&spa->spa_scrub_lock);
1848
1849	return (0);
1850}
1851
1852/*
1853 * ==========================================================================
1854 * SPA async task processing
1855 * ==========================================================================
1856 */
1857
1858static void
1859spa_async_reopen(spa_t *spa)
1860{
1861	vdev_t *rvd = spa->spa_root_vdev;
1862	vdev_t *tvd;
1863	int c;
1864
1865	spa_config_enter(spa, RW_WRITER, FTAG);
1866
1867	for (c = 0; c < rvd->vdev_children; c++) {
1868		tvd = rvd->vdev_child[c];
1869		if (tvd->vdev_reopen_wanted) {
1870			tvd->vdev_reopen_wanted = 0;
1871			vdev_reopen(tvd);
1872		}
1873	}
1874
1875	spa_config_exit(spa, FTAG);
1876}
1877
1878static void
1879spa_async_thread(spa_t *spa)
1880{
1881	int tasks;
1882
1883	ASSERT(spa->spa_sync_on);
1884
1885	mutex_enter(&spa->spa_async_lock);
1886	tasks = spa->spa_async_tasks;
1887	spa->spa_async_tasks = 0;
1888	mutex_exit(&spa->spa_async_lock);
1889
1890	/*
1891	 * See if the config needs to be updated.
1892	 */
1893	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
1894		mutex_enter(&spa_namespace_lock);
1895		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
1896		mutex_exit(&spa_namespace_lock);
1897	}
1898
1899	/*
1900	 * See if any devices need to be reopened.
1901	 */
1902	if (tasks & SPA_ASYNC_REOPEN)
1903		spa_async_reopen(spa);
1904
1905	/*
1906	 * If any devices are done replacing, detach them.
1907	 */
1908	if (tasks & SPA_ASYNC_REPLACE_DONE)
1909		spa_vdev_replace_done(spa);
1910
1911	/*
1912	 * Kick off a scrub.
1913	 */
1914	if (tasks & SPA_ASYNC_SCRUB)
1915		VERIFY(spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_TRUE) == 0);
1916
1917	/*
1918	 * Kick off a resilver.
1919	 */
1920	if (tasks & SPA_ASYNC_RESILVER)
1921		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
1922
1923	/*
1924	 * Let the world know that we're done.
1925	 */
1926	mutex_enter(&spa->spa_async_lock);
1927	spa->spa_async_thread = NULL;
1928	cv_broadcast(&spa->spa_async_cv);
1929	mutex_exit(&spa->spa_async_lock);
1930	thread_exit();
1931}
1932
1933void
1934spa_async_suspend(spa_t *spa)
1935{
1936	mutex_enter(&spa->spa_async_lock);
1937	spa->spa_async_suspended++;
1938	while (spa->spa_async_thread != NULL)
1939		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
1940	mutex_exit(&spa->spa_async_lock);
1941}
1942
1943void
1944spa_async_resume(spa_t *spa)
1945{
1946	mutex_enter(&spa->spa_async_lock);
1947	ASSERT(spa->spa_async_suspended != 0);
1948	spa->spa_async_suspended--;
1949	mutex_exit(&spa->spa_async_lock);
1950}
1951
1952static void
1953spa_async_dispatch(spa_t *spa)
1954{
1955	mutex_enter(&spa->spa_async_lock);
1956	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
1957	    spa->spa_async_thread == NULL &&
1958	    rootdir != NULL && !vn_is_readonly(rootdir))
1959		spa->spa_async_thread = thread_create(NULL, 0,
1960		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
1961	mutex_exit(&spa->spa_async_lock);
1962}
1963
1964void
1965spa_async_request(spa_t *spa, int task)
1966{
1967	mutex_enter(&spa->spa_async_lock);
1968	spa->spa_async_tasks |= task;
1969	mutex_exit(&spa->spa_async_lock);
1970}
1971
1972/*
1973 * ==========================================================================
1974 * SPA syncing routines
1975 * ==========================================================================
1976 */
1977
1978static void
1979spa_sync_deferred_frees(spa_t *spa, uint64_t txg)
1980{
1981	bplist_t *bpl = &spa->spa_sync_bplist;
1982	dmu_tx_t *tx;
1983	blkptr_t blk;
1984	uint64_t itor = 0;
1985	zio_t *zio;
1986	int error;
1987	uint8_t c = 1;
1988
1989	zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CONFIG_HELD);
1990
1991	while (bplist_iterate(bpl, &itor, &blk) == 0)
1992		zio_nowait(zio_free(zio, spa, txg, &blk, NULL, NULL));
1993
1994	error = zio_wait(zio);
1995	ASSERT3U(error, ==, 0);
1996
1997	tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1998	bplist_vacate(bpl, tx);
1999
2000	/*
2001	 * Pre-dirty the first block so we sync to convergence faster.
2002	 * (Usually only the first block is needed.)
2003	 */
2004	dmu_write(spa->spa_meta_objset, spa->spa_sync_bplist_obj, 0, 1, &c, tx);
2005	dmu_tx_commit(tx);
2006}
2007
2008static void
2009spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
2010{
2011	nvlist_t *config;
2012	char *packed = NULL;
2013	size_t nvsize = 0;
2014	dmu_buf_t *db;
2015
2016	if (list_is_empty(&spa->spa_dirty_list))
2017		return;
2018
2019	config = spa_config_generate(spa, NULL, dmu_tx_get_txg(tx), B_FALSE);
2020
2021	if (spa->spa_config_syncing)
2022		nvlist_free(spa->spa_config_syncing);
2023	spa->spa_config_syncing = config;
2024
2025	VERIFY(nvlist_size(config, &nvsize, NV_ENCODE_XDR) == 0);
2026
2027	packed = kmem_alloc(nvsize, KM_SLEEP);
2028
2029	VERIFY(nvlist_pack(config, &packed, &nvsize, NV_ENCODE_XDR,
2030	    KM_SLEEP) == 0);
2031
2032	dmu_write(spa->spa_meta_objset, spa->spa_config_object, 0, nvsize,
2033	    packed, tx);
2034
2035	kmem_free(packed, nvsize);
2036
2037	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset,
2038	    spa->spa_config_object, FTAG, &db));
2039	dmu_buf_will_dirty(db, tx);
2040	*(uint64_t *)db->db_data = nvsize;
2041	dmu_buf_rele(db, FTAG);
2042}
2043
2044/*
2045 * Sync the specified transaction group.  New blocks may be dirtied as
2046 * part of the process, so we iterate until it converges.
2047 */
2048void
2049spa_sync(spa_t *spa, uint64_t txg)
2050{
2051	dsl_pool_t *dp = spa->spa_dsl_pool;
2052	objset_t *mos = spa->spa_meta_objset;
2053	bplist_t *bpl = &spa->spa_sync_bplist;
2054	vdev_t *rvd = spa->spa_root_vdev;
2055	vdev_t *vd;
2056	dmu_tx_t *tx;
2057	int dirty_vdevs;
2058
2059	/*
2060	 * Lock out configuration changes.
2061	 */
2062	spa_config_enter(spa, RW_READER, FTAG);
2063
2064	spa->spa_syncing_txg = txg;
2065	spa->spa_sync_pass = 0;
2066
2067	VERIFY(0 == bplist_open(bpl, mos, spa->spa_sync_bplist_obj));
2068
2069	/*
2070	 * If anything has changed in this txg, push the deferred frees
2071	 * from the previous txg.  If not, leave them alone so that we
2072	 * don't generate work on an otherwise idle system.
2073	 */
2074	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
2075	    !txg_list_empty(&dp->dp_dirty_dirs, txg))
2076		spa_sync_deferred_frees(spa, txg);
2077
2078	/*
2079	 * Iterate to convergence.
2080	 */
2081	do {
2082		spa->spa_sync_pass++;
2083
2084		tx = dmu_tx_create_assigned(dp, txg);
2085		spa_sync_config_object(spa, tx);
2086		dmu_tx_commit(tx);
2087
2088		spa_errlog_sync(spa, txg);
2089
2090		dsl_pool_sync(dp, txg);
2091
2092		dirty_vdevs = 0;
2093		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) {
2094			vdev_sync(vd, txg);
2095			dirty_vdevs++;
2096		}
2097
2098		tx = dmu_tx_create_assigned(dp, txg);
2099		bplist_sync(bpl, tx);
2100		dmu_tx_commit(tx);
2101
2102	} while (dirty_vdevs);
2103
2104	bplist_close(bpl);
2105
2106	dprintf("txg %llu passes %d\n", txg, spa->spa_sync_pass);
2107
2108	/*
2109	 * Rewrite the vdev configuration (which includes the uberblock)
2110	 * to commit the transaction group.
2111	 *
2112	 * If there are any dirty vdevs, sync the uberblock to all vdevs.
2113	 * Otherwise, pick a random top-level vdev that's known to be
2114	 * visible in the config cache (see spa_vdev_add() for details).
2115	 * If the write fails, try the next vdev until we're tried them all.
2116	 */
2117	if (!list_is_empty(&spa->spa_dirty_list)) {
2118		VERIFY(vdev_config_sync(rvd, txg) == 0);
2119	} else {
2120		int children = rvd->vdev_children;
2121		int c0 = spa_get_random(children);
2122		int c;
2123
2124		for (c = 0; c < children; c++) {
2125			vd = rvd->vdev_child[(c0 + c) % children];
2126			if (vd->vdev_ms_array == 0)
2127				continue;
2128			if (vdev_config_sync(vd, txg) == 0)
2129				break;
2130		}
2131		if (c == children)
2132			VERIFY(vdev_config_sync(rvd, txg) == 0);
2133	}
2134
2135	/*
2136	 * Clear the dirty config list.
2137	 */
2138	while ((vd = list_head(&spa->spa_dirty_list)) != NULL)
2139		vdev_config_clean(vd);
2140
2141	/*
2142	 * Now that the new config has synced transactionally,
2143	 * let it become visible to the config cache.
2144	 */
2145	if (spa->spa_config_syncing != NULL) {
2146		spa_config_set(spa, spa->spa_config_syncing);
2147		spa->spa_config_txg = txg;
2148		spa->spa_config_syncing = NULL;
2149	}
2150
2151	/*
2152	 * Make a stable copy of the fully synced uberblock.
2153	 * We use this as the root for pool traversals.
2154	 */
2155	spa->spa_traverse_wanted = 1;	/* tells traverse_more() to stop */
2156
2157	spa_scrub_suspend(spa);		/* stop scrubbing and finish I/Os */
2158
2159	rw_enter(&spa->spa_traverse_lock, RW_WRITER);
2160	spa->spa_traverse_wanted = 0;
2161	spa->spa_ubsync = spa->spa_uberblock;
2162	rw_exit(&spa->spa_traverse_lock);
2163
2164	spa_scrub_resume(spa);		/* resume scrub with new ubsync */
2165
2166	/*
2167	 * Clean up the ZIL records for the synced txg.
2168	 */
2169	dsl_pool_zil_clean(dp);
2170
2171	/*
2172	 * Update usable space statistics.
2173	 */
2174	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
2175		vdev_sync_done(vd, txg);
2176
2177	/*
2178	 * It had better be the case that we didn't dirty anything
2179	 * since spa_sync_labels().
2180	 */
2181	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
2182	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
2183	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
2184	ASSERT(bpl->bpl_queue == NULL);
2185
2186	spa_config_exit(spa, FTAG);
2187
2188	/*
2189	 * If any async tasks have been requested, kick them off.
2190	 */
2191	spa_async_dispatch(spa);
2192}
2193
2194/*
2195 * Sync all pools.  We don't want to hold the namespace lock across these
2196 * operations, so we take a reference on the spa_t and drop the lock during the
2197 * sync.
2198 */
2199void
2200spa_sync_allpools(void)
2201{
2202	spa_t *spa = NULL;
2203	mutex_enter(&spa_namespace_lock);
2204	while ((spa = spa_next(spa)) != NULL) {
2205		if (spa_state(spa) != POOL_STATE_ACTIVE)
2206			continue;
2207		spa_open_ref(spa, FTAG);
2208		mutex_exit(&spa_namespace_lock);
2209		txg_wait_synced(spa_get_dsl(spa), 0);
2210		mutex_enter(&spa_namespace_lock);
2211		spa_close(spa, FTAG);
2212	}
2213	mutex_exit(&spa_namespace_lock);
2214}
2215
2216/*
2217 * ==========================================================================
2218 * Miscellaneous routines
2219 * ==========================================================================
2220 */
2221
2222/*
2223 * Remove all pools in the system.
2224 */
2225void
2226spa_evict_all(void)
2227{
2228	spa_t *spa;
2229
2230	/*
2231	 * Remove all cached state.  All pools should be closed now,
2232	 * so every spa in the AVL tree should be unreferenced.
2233	 */
2234	mutex_enter(&spa_namespace_lock);
2235	while ((spa = spa_next(NULL)) != NULL) {
2236		/*
2237		 * Stop async tasks.  The async thread may need to detach
2238		 * a device that's been replaced, which requires grabbing
2239		 * spa_namespace_lock, so we must drop it here.
2240		 */
2241		spa_open_ref(spa, FTAG);
2242		mutex_exit(&spa_namespace_lock);
2243		spa_async_suspend(spa);
2244		VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0);
2245		mutex_enter(&spa_namespace_lock);
2246		spa_close(spa, FTAG);
2247
2248		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
2249			spa_unload(spa);
2250			spa_deactivate(spa);
2251		}
2252		spa_remove(spa);
2253	}
2254	mutex_exit(&spa_namespace_lock);
2255}
2256
2257vdev_t *
2258spa_lookup_by_guid(spa_t *spa, uint64_t guid)
2259{
2260	return (vdev_lookup_by_guid(spa->spa_root_vdev, guid));
2261}
2262