spa.c revision eaca9bbd5f5d1e4e554da4c7108e8a03c8c33481
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		error = EINVAL;
280		goto out;
281	}
282
283	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
284	    &spa->spa_config_txg);
285
286	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
287	    spa_guid_exists(pool_guid, 0)) {
288		error = EEXIST;
289		goto out;
290	}
291
292	/*
293	 * Parse the configuration into a vdev tree.
294	 */
295	spa_config_enter(spa, RW_WRITER, FTAG);
296	rvd = spa_config_parse(spa, nvroot, NULL, 0, VDEV_ALLOC_LOAD);
297	spa_config_exit(spa, FTAG);
298
299	if (rvd == NULL) {
300		error = EINVAL;
301		goto out;
302	}
303
304	ASSERT(spa->spa_root_vdev == rvd);
305	ASSERT(spa_guid(spa) == pool_guid);
306
307	/*
308	 * Try to open all vdevs, loading each label in the process.
309	 */
310	if (vdev_open(rvd) != 0) {
311		error = ENXIO;
312		goto out;
313	}
314
315	/*
316	 * Find the best uberblock.
317	 */
318	bzero(ub, sizeof (uberblock_t));
319
320	zio = zio_root(spa, NULL, NULL,
321	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
322	vdev_uberblock_load(zio, rvd, ub);
323	error = zio_wait(zio);
324
325	/*
326	 * If we weren't able to find a single valid uberblock, return failure.
327	 */
328	if (ub->ub_txg == 0) {
329		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
330		    VDEV_AUX_CORRUPT_DATA);
331		error = ENXIO;
332		goto out;
333	}
334
335	/*
336	 * If the pool is newer than the code, we can't open it.
337	 */
338	if (ub->ub_version > ZFS_VERSION) {
339		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
340		    VDEV_AUX_VERSION_NEWER);
341		error = ENOTSUP;
342		goto out;
343	}
344
345	/*
346	 * If the vdev guid sum doesn't match the uberblock, we have an
347	 * incomplete configuration.
348	 */
349	if (rvd->vdev_guid_sum != ub->ub_guid_sum && mosconfig) {
350		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
351		    VDEV_AUX_BAD_GUID_SUM);
352		error = ENXIO;
353		goto out;
354	}
355
356	/*
357	 * Initialize internal SPA structures.
358	 */
359	spa->spa_state = POOL_STATE_ACTIVE;
360	spa->spa_ubsync = spa->spa_uberblock;
361	spa->spa_first_txg = spa_last_synced_txg(spa) + 1;
362	error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
363	if (error) {
364		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
365		    VDEV_AUX_CORRUPT_DATA);
366		goto out;
367	}
368	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
369
370	if (zap_lookup(spa->spa_meta_objset,
371	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
372	    sizeof (uint64_t), 1, &spa->spa_config_object) != 0) {
373		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
374		    VDEV_AUX_CORRUPT_DATA);
375		error = EIO;
376		goto out;
377	}
378
379	if (!mosconfig) {
380		dmu_buf_t *db;
381		char *packed = NULL;
382		size_t nvsize = 0;
383		nvlist_t *newconfig = NULL;
384
385		VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset,
386		    spa->spa_config_object, FTAG, &db));
387		nvsize = *(uint64_t *)db->db_data;
388		dmu_buf_rele(db, FTAG);
389
390		packed = kmem_alloc(nvsize, KM_SLEEP);
391		error = dmu_read(spa->spa_meta_objset,
392		    spa->spa_config_object, 0, nvsize, packed);
393		if (error == 0)
394			error = nvlist_unpack(packed, nvsize, &newconfig, 0);
395		kmem_free(packed, nvsize);
396
397		if (error) {
398			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
399			    VDEV_AUX_CORRUPT_DATA);
400			error = EIO;
401			goto out;
402		}
403
404		spa_config_set(spa, newconfig);
405
406		spa_unload(spa);
407		spa_deactivate(spa);
408		spa_activate(spa);
409
410		return (spa_load(spa, newconfig, state, B_TRUE));
411	}
412
413	if (zap_lookup(spa->spa_meta_objset,
414	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
415	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj) != 0) {
416		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
417		    VDEV_AUX_CORRUPT_DATA);
418		error = EIO;
419		goto out;
420	}
421
422	/*
423	 * Load the persistent error log.  If we have an older pool, this will
424	 * not be present.
425	 */
426	error = zap_lookup(spa->spa_meta_objset,
427	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_LAST,
428	    sizeof (uint64_t), 1, &spa->spa_errlog_last);
429	if (error != 0 &&error != ENOENT) {
430		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
431		    VDEV_AUX_CORRUPT_DATA);
432		error = EIO;
433		goto out;
434	}
435
436	error = zap_lookup(spa->spa_meta_objset,
437	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_SCRUB,
438	    sizeof (uint64_t), 1, &spa->spa_errlog_scrub);
439	if (error != 0 && error != ENOENT) {
440		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
441		    VDEV_AUX_CORRUPT_DATA);
442		error = EIO;
443		goto out;
444	}
445
446	/*
447	 * Load the vdev state for all top level vdevs.  We need to grab the
448	 * config lock because all label I/O is done with the
449	 * ZIO_FLAG_CONFIG_HELD flag.
450	 */
451	spa_config_enter(spa, RW_READER, FTAG);
452	error = vdev_load(rvd);
453	spa_config_exit(spa, FTAG);
454
455	if (error)
456		goto out;
457
458	/*
459	 * Propagate the leaf DTLs we just loaded all the way up the tree.
460	 */
461	spa_config_enter(spa, RW_WRITER, FTAG);
462	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
463	spa_config_exit(spa, FTAG);
464
465	/*
466	 * Check the state of the root vdev.  If it can't be opened, it
467	 * indicates one or more toplevel vdevs are faulted.
468	 */
469	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
470		error = ENXIO;
471		goto out;
472	}
473
474	if ((spa_mode & FWRITE) && state != SPA_LOAD_TRYIMPORT) {
475		dmu_tx_t *tx;
476		int need_update = B_FALSE;
477		int c;
478
479		/*
480		 * Claim log blocks that haven't been committed yet.
481		 * This must all happen in a single txg.
482		 */
483		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
484		    spa_first_txg(spa));
485		dmu_objset_find(spa->spa_name, zil_claim, tx, 0);
486		dmu_tx_commit(tx);
487
488		spa->spa_sync_on = B_TRUE;
489		txg_sync_start(spa->spa_dsl_pool);
490
491		/*
492		 * Wait for all claims to sync.
493		 */
494		txg_wait_synced(spa->spa_dsl_pool, 0);
495
496		/*
497		 * If the config cache is stale, or we have uninitialized
498		 * metaslabs (see spa_vdev_add()), then update the config.
499		 */
500		if (config_cache_txg != spa->spa_config_txg ||
501		    state == SPA_LOAD_IMPORT)
502			need_update = B_TRUE;
503
504		for (c = 0; c < rvd->vdev_children; c++)
505			if (rvd->vdev_child[c]->vdev_ms_array == 0)
506				need_update = B_TRUE;
507
508		/*
509		 * Update the config cache asychronously in case we're the
510		 * root pool, in which case the config cache isn't writable yet.
511		 */
512		if (need_update)
513			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
514	}
515
516	error = 0;
517out:
518	if (error)
519		zfs_ereport_post(FM_EREPORT_ZFS_POOL, spa, NULL, NULL, 0, 0);
520	spa->spa_load_state = SPA_LOAD_NONE;
521	spa->spa_ena = 0;
522
523	return (error);
524}
525
526/*
527 * Pool Open/Import
528 *
529 * The import case is identical to an open except that the configuration is sent
530 * down from userland, instead of grabbed from the configuration cache.  For the
531 * case of an open, the pool configuration will exist in the
532 * POOL_STATE_UNITIALIZED state.
533 *
534 * The stats information (gen/count/ustats) is used to gather vdev statistics at
535 * the same time open the pool, without having to keep around the spa_t in some
536 * ambiguous state.
537 */
538static int
539spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t **config)
540{
541	spa_t *spa;
542	int error;
543	int loaded = B_FALSE;
544	int locked = B_FALSE;
545
546	*spapp = NULL;
547
548	/*
549	 * As disgusting as this is, we need to support recursive calls to this
550	 * function because dsl_dir_open() is called during spa_load(), and ends
551	 * up calling spa_open() again.  The real fix is to figure out how to
552	 * avoid dsl_dir_open() calling this in the first place.
553	 */
554	if (mutex_owner(&spa_namespace_lock) != curthread) {
555		mutex_enter(&spa_namespace_lock);
556		locked = B_TRUE;
557	}
558
559	if ((spa = spa_lookup(pool)) == NULL) {
560		if (locked)
561			mutex_exit(&spa_namespace_lock);
562		return (ENOENT);
563	}
564	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
565
566		spa_activate(spa);
567
568		error = spa_load(spa, spa->spa_config, SPA_LOAD_OPEN, B_FALSE);
569
570		if (error == EBADF) {
571			/*
572			 * If vdev_load() returns EBADF, it indicates that one
573			 * of the vdevs indicates that the pool has been
574			 * exported or destroyed.  If this is the case, the
575			 * config cache is out of sync and we should remove the
576			 * pool from the namespace.
577			 */
578			spa_unload(spa);
579			spa_deactivate(spa);
580			spa_remove(spa);
581			spa_config_sync();
582			if (locked)
583				mutex_exit(&spa_namespace_lock);
584			return (ENOENT);
585		}
586
587		if (error) {
588			/*
589			 * We can't open the pool, but we still have useful
590			 * information: the state of each vdev after the
591			 * attempted vdev_open().  Return this to the user.
592			 */
593			if (config != NULL && spa->spa_root_vdev != NULL) {
594				spa_config_enter(spa, RW_READER, FTAG);
595				*config = spa_config_generate(spa, NULL, -1ULL,
596				    B_TRUE);
597				spa_config_exit(spa, FTAG);
598			}
599			spa_unload(spa);
600			spa_deactivate(spa);
601			spa->spa_last_open_failed = B_TRUE;
602			if (locked)
603				mutex_exit(&spa_namespace_lock);
604			*spapp = NULL;
605			return (error);
606		} else {
607			zfs_post_ok(spa, NULL);
608			spa->spa_last_open_failed = B_FALSE;
609		}
610
611		loaded = B_TRUE;
612	}
613
614	spa_open_ref(spa, tag);
615	if (locked)
616		mutex_exit(&spa_namespace_lock);
617
618	*spapp = spa;
619
620	if (config != NULL) {
621		spa_config_enter(spa, RW_READER, FTAG);
622		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
623		spa_config_exit(spa, FTAG);
624	}
625
626	/*
627	 * If we just loaded the pool, resilver anything that's out of date.
628	 */
629	if (loaded && (spa_mode & FWRITE))
630		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
631
632	return (0);
633}
634
635int
636spa_open(const char *name, spa_t **spapp, void *tag)
637{
638	return (spa_open_common(name, spapp, tag, NULL));
639}
640
641/*
642 * Lookup the given spa_t, incrementing the inject count in the process,
643 * preventing it from being exported or destroyed.
644 */
645spa_t *
646spa_inject_addref(char *name)
647{
648	spa_t *spa;
649
650	mutex_enter(&spa_namespace_lock);
651	if ((spa = spa_lookup(name)) == NULL) {
652		mutex_exit(&spa_namespace_lock);
653		return (NULL);
654	}
655	spa->spa_inject_ref++;
656	mutex_exit(&spa_namespace_lock);
657
658	return (spa);
659}
660
661void
662spa_inject_delref(spa_t *spa)
663{
664	mutex_enter(&spa_namespace_lock);
665	spa->spa_inject_ref--;
666	mutex_exit(&spa_namespace_lock);
667}
668
669int
670spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
671{
672	int error;
673	spa_t *spa;
674
675	*config = NULL;
676	error = spa_open_common(name, &spa, FTAG, config);
677
678	if (spa && *config != NULL)
679		VERIFY(nvlist_add_uint64(*config, ZPOOL_CONFIG_ERRCOUNT,
680		    spa_get_errlog_size(spa)) == 0);
681
682	/*
683	 * We want to get the alternate root even for faulted pools, so we cheat
684	 * and call spa_lookup() directly.
685	 */
686	if (altroot) {
687		if (spa == NULL) {
688			mutex_enter(&spa_namespace_lock);
689			spa = spa_lookup(name);
690			if (spa)
691				spa_altroot(spa, altroot, buflen);
692			else
693				altroot[0] = '\0';
694			spa = NULL;
695			mutex_exit(&spa_namespace_lock);
696		} else {
697			spa_altroot(spa, altroot, buflen);
698		}
699	}
700
701	if (spa != NULL)
702		spa_close(spa, FTAG);
703
704	return (error);
705}
706
707/*
708 * Pool Creation
709 */
710int
711spa_create(const char *pool, nvlist_t *nvroot, const char *altroot)
712{
713	spa_t *spa;
714	vdev_t *rvd;
715	dsl_pool_t *dp;
716	dmu_tx_t *tx;
717	int c, error;
718	uint64_t txg = TXG_INITIAL;
719
720	/*
721	 * If this pool already exists, return failure.
722	 */
723	mutex_enter(&spa_namespace_lock);
724	if (spa_lookup(pool) != NULL) {
725		mutex_exit(&spa_namespace_lock);
726		return (EEXIST);
727	}
728
729	/*
730	 * Allocate a new spa_t structure.
731	 */
732	spa = spa_add(pool, altroot);
733	spa_activate(spa);
734
735	spa->spa_uberblock.ub_txg = txg - 1;
736	spa->spa_uberblock.ub_version = ZFS_VERSION;
737	spa->spa_ubsync = spa->spa_uberblock;
738
739	/*
740	 * Create the root vdev.
741	 */
742	spa_config_enter(spa, RW_WRITER, FTAG);
743
744	rvd = spa_config_parse(spa, nvroot, NULL, 0, VDEV_ALLOC_ADD);
745
746	ASSERT(spa->spa_root_vdev == rvd);
747
748	if (rvd == NULL) {
749		error = EINVAL;
750	} else {
751		if ((error = vdev_create(rvd, txg)) == 0) {
752			for (c = 0; c < rvd->vdev_children; c++)
753				vdev_init(rvd->vdev_child[c], txg);
754			vdev_config_dirty(rvd);
755		}
756	}
757
758	spa_config_exit(spa, FTAG);
759
760	if (error) {
761		spa_unload(spa);
762		spa_deactivate(spa);
763		spa_remove(spa);
764		mutex_exit(&spa_namespace_lock);
765		return (error);
766	}
767
768	spa->spa_dsl_pool = dp = dsl_pool_create(spa, txg);
769	spa->spa_meta_objset = dp->dp_meta_objset;
770
771	tx = dmu_tx_create_assigned(dp, txg);
772
773	/*
774	 * Create the pool config object.
775	 */
776	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
777	    DMU_OT_PACKED_NVLIST, 1 << 14,
778	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
779
780	if (zap_add(spa->spa_meta_objset,
781	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
782	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
783		cmn_err(CE_PANIC, "failed to add pool config");
784	}
785
786	/*
787	 * Create the deferred-free bplist object.  Turn off compression
788	 * because sync-to-convergence takes longer if the blocksize
789	 * keeps changing.
790	 */
791	spa->spa_sync_bplist_obj = bplist_create(spa->spa_meta_objset,
792	    1 << 14, tx);
793	dmu_object_set_compress(spa->spa_meta_objset, spa->spa_sync_bplist_obj,
794	    ZIO_COMPRESS_OFF, tx);
795
796	if (zap_add(spa->spa_meta_objset,
797	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
798	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj, tx) != 0) {
799		cmn_err(CE_PANIC, "failed to add bplist");
800	}
801
802	dmu_tx_commit(tx);
803
804	spa->spa_sync_on = B_TRUE;
805	txg_sync_start(spa->spa_dsl_pool);
806
807	/*
808	 * We explicitly wait for the first transaction to complete so that our
809	 * bean counters are appropriately updated.
810	 */
811	txg_wait_synced(spa->spa_dsl_pool, txg);
812
813	spa_config_sync();
814
815	mutex_exit(&spa_namespace_lock);
816
817	return (0);
818}
819
820/*
821 * Import the given pool into the system.  We set up the necessary spa_t and
822 * then call spa_load() to do the dirty work.
823 */
824int
825spa_import(const char *pool, nvlist_t *config, const char *altroot)
826{
827	spa_t *spa;
828	int error;
829
830	if (!(spa_mode & FWRITE))
831		return (EROFS);
832
833	/*
834	 * If a pool with this name exists, return failure.
835	 */
836	mutex_enter(&spa_namespace_lock);
837	if (spa_lookup(pool) != NULL) {
838		mutex_exit(&spa_namespace_lock);
839		return (EEXIST);
840	}
841
842	/*
843	 * Create and initialize the spa structure.
844	 */
845	spa = spa_add(pool, altroot);
846	spa_activate(spa);
847
848	/*
849	 * Pass off the heavy lifting to spa_load().
850	 * Pass TRUE for mosconfig because the user-supplied config
851	 * is actually the one to trust when doing an import.
852	 */
853	error = spa_load(spa, config, SPA_LOAD_IMPORT, B_TRUE);
854
855	if (error) {
856		spa_unload(spa);
857		spa_deactivate(spa);
858		spa_remove(spa);
859		mutex_exit(&spa_namespace_lock);
860		return (error);
861	}
862
863	/*
864	 * Update the config cache to include the newly-imported pool.
865	 */
866	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
867
868	mutex_exit(&spa_namespace_lock);
869
870	/*
871	 * Resilver anything that's out of date.
872	 */
873	if (spa_mode & FWRITE)
874		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
875
876	return (0);
877}
878
879/*
880 * This (illegal) pool name is used when temporarily importing a spa_t in order
881 * to get the vdev stats associated with the imported devices.
882 */
883#define	TRYIMPORT_NAME	"$import"
884
885nvlist_t *
886spa_tryimport(nvlist_t *tryconfig)
887{
888	nvlist_t *config = NULL;
889	char *poolname;
890	spa_t *spa;
891	uint64_t state;
892
893	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
894		return (NULL);
895
896	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
897		return (NULL);
898
899	/*
900	 * Create and initialize the spa structure.
901	 */
902	mutex_enter(&spa_namespace_lock);
903	spa = spa_add(TRYIMPORT_NAME, NULL);
904	spa_activate(spa);
905
906	/*
907	 * Pass off the heavy lifting to spa_load().
908	 * Pass TRUE for mosconfig because the user-supplied config
909	 * is actually the one to trust when doing an import.
910	 */
911	(void) spa_load(spa, tryconfig, SPA_LOAD_TRYIMPORT, B_TRUE);
912
913	/*
914	 * If 'tryconfig' was at least parsable, return the current config.
915	 */
916	if (spa->spa_root_vdev != NULL) {
917		spa_config_enter(spa, RW_READER, FTAG);
918		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
919		spa_config_exit(spa, FTAG);
920		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
921		    poolname) == 0);
922		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
923		    state) == 0);
924	}
925
926	spa_unload(spa);
927	spa_deactivate(spa);
928	spa_remove(spa);
929	mutex_exit(&spa_namespace_lock);
930
931	return (config);
932}
933
934/*
935 * Pool export/destroy
936 *
937 * The act of destroying or exporting a pool is very simple.  We make sure there
938 * is no more pending I/O and any references to the pool are gone.  Then, we
939 * update the pool state and sync all the labels to disk, removing the
940 * configuration from the cache afterwards.
941 */
942static int
943spa_export_common(char *pool, int new_state)
944{
945	spa_t *spa;
946
947	if (!(spa_mode & FWRITE))
948		return (EROFS);
949
950	mutex_enter(&spa_namespace_lock);
951	if ((spa = spa_lookup(pool)) == NULL) {
952		mutex_exit(&spa_namespace_lock);
953		return (ENOENT);
954	}
955
956	/*
957	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
958	 * reacquire the namespace lock, and see if we can export.
959	 */
960	spa_open_ref(spa, FTAG);
961	mutex_exit(&spa_namespace_lock);
962	spa_async_suspend(spa);
963	mutex_enter(&spa_namespace_lock);
964	spa_close(spa, FTAG);
965
966	/*
967	 * The pool will be in core if it's openable,
968	 * in which case we can modify its state.
969	 */
970	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
971		/*
972		 * Objsets may be open only because they're dirty, so we
973		 * have to force it to sync before checking spa_refcnt.
974		 */
975		spa_scrub_suspend(spa);
976		txg_wait_synced(spa->spa_dsl_pool, 0);
977
978		/*
979		 * A pool cannot be exported or destroyed if there are active
980		 * references.  If we are resetting a pool, allow references by
981		 * fault injection handlers.
982		 */
983		if (!spa_refcount_zero(spa) ||
984		    (spa->spa_inject_ref != 0 &&
985		    new_state != POOL_STATE_UNINITIALIZED)) {
986			spa_scrub_resume(spa);
987			spa_async_resume(spa);
988			mutex_exit(&spa_namespace_lock);
989			return (EBUSY);
990		}
991
992		spa_scrub_resume(spa);
993		VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0);
994
995		/*
996		 * We want this to be reflected on every label,
997		 * so mark them all dirty.  spa_unload() will do the
998		 * final sync that pushes these changes out.
999		 */
1000		if (new_state != POOL_STATE_UNINITIALIZED) {
1001			spa_config_enter(spa, RW_WRITER, FTAG);
1002			spa->spa_state = new_state;
1003			spa->spa_final_txg = spa_last_synced_txg(spa) + 1;
1004			vdev_config_dirty(spa->spa_root_vdev);
1005			spa_config_exit(spa, FTAG);
1006		}
1007	}
1008
1009	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
1010		spa_unload(spa);
1011		spa_deactivate(spa);
1012	}
1013
1014	if (new_state != POOL_STATE_UNINITIALIZED) {
1015		spa_remove(spa);
1016		spa_config_sync();
1017	}
1018	mutex_exit(&spa_namespace_lock);
1019
1020	return (0);
1021}
1022
1023/*
1024 * Destroy a storage pool.
1025 */
1026int
1027spa_destroy(char *pool)
1028{
1029	return (spa_export_common(pool, POOL_STATE_DESTROYED));
1030}
1031
1032/*
1033 * Export a storage pool.
1034 */
1035int
1036spa_export(char *pool)
1037{
1038	return (spa_export_common(pool, POOL_STATE_EXPORTED));
1039}
1040
1041/*
1042 * Similar to spa_export(), this unloads the spa_t without actually removing it
1043 * from the namespace in any way.
1044 */
1045int
1046spa_reset(char *pool)
1047{
1048	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED));
1049}
1050
1051
1052/*
1053 * ==========================================================================
1054 * Device manipulation
1055 * ==========================================================================
1056 */
1057
1058/*
1059 * Add capacity to a storage pool.
1060 */
1061int
1062spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
1063{
1064	uint64_t txg;
1065	int c, error;
1066	vdev_t *rvd = spa->spa_root_vdev;
1067	vdev_t *vd, *tvd;
1068
1069	txg = spa_vdev_enter(spa);
1070
1071	vd = spa_config_parse(spa, nvroot, NULL, 0, VDEV_ALLOC_ADD);
1072
1073	if (vd == NULL)
1074		return (spa_vdev_exit(spa, vd, txg, EINVAL));
1075
1076	if ((error = vdev_create(vd, txg)) != 0)
1077		return (spa_vdev_exit(spa, vd, txg, error));
1078
1079	/*
1080	 * Transfer each new top-level vdev from vd to rvd.
1081	 */
1082	for (c = 0; c < vd->vdev_children; c++) {
1083		tvd = vd->vdev_child[c];
1084		vdev_remove_child(vd, tvd);
1085		tvd->vdev_id = rvd->vdev_children;
1086		vdev_add_child(rvd, tvd);
1087		vdev_config_dirty(tvd);
1088	}
1089
1090	/*
1091	 * We have to be careful when adding new vdevs to an existing pool.
1092	 * If other threads start allocating from these vdevs before we
1093	 * sync the config cache, and we lose power, then upon reboot we may
1094	 * fail to open the pool because there are DVAs that the config cache
1095	 * can't translate.  Therefore, we first add the vdevs without
1096	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
1097	 * and then let spa_config_update() initialize the new metaslabs.
1098	 *
1099	 * spa_load() checks for added-but-not-initialized vdevs, so that
1100	 * if we lose power at any point in this sequence, the remaining
1101	 * steps will be completed the next time we load the pool.
1102	 */
1103	(void) spa_vdev_exit(spa, vd, txg, 0);
1104
1105	mutex_enter(&spa_namespace_lock);
1106	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
1107	mutex_exit(&spa_namespace_lock);
1108
1109	return (0);
1110}
1111
1112/*
1113 * Attach a device to a mirror.  The arguments are the path to any device
1114 * in the mirror, and the nvroot for the new device.  If the path specifies
1115 * a device that is not mirrored, we automatically insert the mirror vdev.
1116 *
1117 * If 'replacing' is specified, the new device is intended to replace the
1118 * existing device; in this case the two devices are made into their own
1119 * mirror using the 'replacing' vdev, which is functionally idendical to
1120 * the mirror vdev (it actually reuses all the same ops) but has a few
1121 * extra rules: you can't attach to it after it's been created, and upon
1122 * completion of resilvering, the first disk (the one being replaced)
1123 * is automatically detached.
1124 */
1125int
1126spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
1127{
1128	uint64_t txg, open_txg;
1129	int error;
1130	vdev_t *rvd = spa->spa_root_vdev;
1131	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
1132	vdev_ops_t *pvops = replacing ? &vdev_replacing_ops : &vdev_mirror_ops;
1133
1134	txg = spa_vdev_enter(spa);
1135
1136	oldvd = vdev_lookup_by_guid(rvd, guid);
1137
1138	if (oldvd == NULL)
1139		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
1140
1141	if (!oldvd->vdev_ops->vdev_op_leaf)
1142		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1143
1144	pvd = oldvd->vdev_parent;
1145
1146	/*
1147	 * The parent must be a mirror or the root, unless we're replacing;
1148	 * in that case, the parent can be anything but another replacing vdev.
1149	 */
1150	if (pvd->vdev_ops != &vdev_mirror_ops &&
1151	    pvd->vdev_ops != &vdev_root_ops &&
1152	    (!replacing || pvd->vdev_ops == &vdev_replacing_ops))
1153		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1154
1155	newrootvd = spa_config_parse(spa, nvroot, NULL, 0, VDEV_ALLOC_ADD);
1156
1157	if (newrootvd == NULL || newrootvd->vdev_children != 1)
1158		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
1159
1160	newvd = newrootvd->vdev_child[0];
1161
1162	if (!newvd->vdev_ops->vdev_op_leaf)
1163		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
1164
1165	if ((error = vdev_create(newrootvd, txg)) != 0)
1166		return (spa_vdev_exit(spa, newrootvd, txg, error));
1167
1168	/*
1169	 * Compare the new device size with the replaceable/attachable
1170	 * device size.
1171	 */
1172	if (newvd->vdev_psize < vdev_get_rsize(oldvd))
1173		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
1174
1175	/*
1176	 * The new device cannot have a higher alignment requirement
1177	 * than the top-level vdev.
1178	 */
1179	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
1180		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
1181
1182	/*
1183	 * If this is an in-place replacement, update oldvd's path and devid
1184	 * to make it distinguishable from newvd, and unopenable from now on.
1185	 */
1186	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
1187		spa_strfree(oldvd->vdev_path);
1188		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
1189		    KM_SLEEP);
1190		(void) sprintf(oldvd->vdev_path, "%s/%s",
1191		    newvd->vdev_path, "old");
1192		if (oldvd->vdev_devid != NULL) {
1193			spa_strfree(oldvd->vdev_devid);
1194			oldvd->vdev_devid = NULL;
1195		}
1196	}
1197
1198	/*
1199	 * If the parent is not a mirror, or if we're replacing,
1200	 * insert the new mirror/replacing vdev above oldvd.
1201	 */
1202	if (pvd->vdev_ops != pvops)
1203		pvd = vdev_add_parent(oldvd, pvops);
1204
1205	ASSERT(pvd->vdev_top->vdev_parent == rvd);
1206	ASSERT(pvd->vdev_ops == pvops);
1207	ASSERT(oldvd->vdev_parent == pvd);
1208
1209	/*
1210	 * Extract the new device from its root and add it to pvd.
1211	 */
1212	vdev_remove_child(newrootvd, newvd);
1213	newvd->vdev_id = pvd->vdev_children;
1214	vdev_add_child(pvd, newvd);
1215
1216	/*
1217	 * If newvd is smaller than oldvd, but larger than its rsize,
1218	 * the addition of newvd may have decreased our parent's asize.
1219	 */
1220	pvd->vdev_asize = MIN(pvd->vdev_asize, newvd->vdev_asize);
1221
1222	tvd = newvd->vdev_top;
1223	ASSERT(pvd->vdev_top == tvd);
1224	ASSERT(tvd->vdev_parent == rvd);
1225
1226	vdev_config_dirty(tvd);
1227
1228	/*
1229	 * Set newvd's DTL to [TXG_INITIAL, open_txg].  It will propagate
1230	 * upward when spa_vdev_exit() calls vdev_dtl_reassess().
1231	 */
1232	open_txg = txg + TXG_CONCURRENT_STATES - 1;
1233
1234	mutex_enter(&newvd->vdev_dtl_lock);
1235	space_map_add(&newvd->vdev_dtl_map, TXG_INITIAL,
1236	    open_txg - TXG_INITIAL + 1);
1237	mutex_exit(&newvd->vdev_dtl_lock);
1238
1239	dprintf("attached %s in txg %llu\n", newvd->vdev_path, txg);
1240
1241	/*
1242	 * Mark newvd's DTL dirty in this txg.
1243	 */
1244	vdev_dirty(tvd, VDD_DTL, newvd, txg);
1245
1246	(void) spa_vdev_exit(spa, newrootvd, open_txg, 0);
1247
1248	/*
1249	 * Kick off a resilver to update newvd.
1250	 */
1251	VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
1252
1253	return (0);
1254}
1255
1256/*
1257 * Detach a device from a mirror or replacing vdev.
1258 * If 'replace_done' is specified, only detach if the parent
1259 * is a replacing vdev.
1260 */
1261int
1262spa_vdev_detach(spa_t *spa, uint64_t guid, int replace_done)
1263{
1264	uint64_t txg;
1265	int c, t, error;
1266	vdev_t *rvd = spa->spa_root_vdev;
1267	vdev_t *vd, *pvd, *cvd, *tvd;
1268
1269	txg = spa_vdev_enter(spa);
1270
1271	vd = vdev_lookup_by_guid(rvd, guid);
1272
1273	if (vd == NULL)
1274		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
1275
1276	if (!vd->vdev_ops->vdev_op_leaf)
1277		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1278
1279	pvd = vd->vdev_parent;
1280
1281	/*
1282	 * If replace_done is specified, only remove this device if it's
1283	 * the first child of a replacing vdev.
1284	 */
1285	if (replace_done &&
1286	    (vd->vdev_id != 0 || pvd->vdev_ops != &vdev_replacing_ops))
1287		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1288
1289	/*
1290	 * Only mirror and replacing vdevs support detach.
1291	 */
1292	if (pvd->vdev_ops != &vdev_replacing_ops &&
1293	    pvd->vdev_ops != &vdev_mirror_ops)
1294		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1295
1296	/*
1297	 * If there's only one replica, you can't detach it.
1298	 */
1299	if (pvd->vdev_children <= 1)
1300		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
1301
1302	/*
1303	 * If all siblings have non-empty DTLs, this device may have the only
1304	 * valid copy of the data, which means we cannot safely detach it.
1305	 *
1306	 * XXX -- as in the vdev_offline() case, we really want a more
1307	 * precise DTL check.
1308	 */
1309	for (c = 0; c < pvd->vdev_children; c++) {
1310		uint64_t dirty;
1311
1312		cvd = pvd->vdev_child[c];
1313		if (cvd == vd)
1314			continue;
1315		if (vdev_is_dead(cvd))
1316			continue;
1317		mutex_enter(&cvd->vdev_dtl_lock);
1318		dirty = cvd->vdev_dtl_map.sm_space |
1319		    cvd->vdev_dtl_scrub.sm_space;
1320		mutex_exit(&cvd->vdev_dtl_lock);
1321		if (!dirty)
1322			break;
1323	}
1324	if (c == pvd->vdev_children)
1325		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
1326
1327	/*
1328	 * Erase the disk labels so the disk can be used for other things.
1329	 * This must be done after all other error cases are handled,
1330	 * but before we disembowel vd (so we can still do I/O to it).
1331	 * But if we can't do it, don't treat the error as fatal --
1332	 * it may be that the unwritability of the disk is the reason
1333	 * it's being detached!
1334	 */
1335	error = vdev_label_init(vd, 0);
1336	if (error)
1337		dprintf("unable to erase labels on %s\n", vdev_description(vd));
1338
1339	/*
1340	 * Remove vd from its parent and compact the parent's children.
1341	 */
1342	vdev_remove_child(pvd, vd);
1343	vdev_compact_children(pvd);
1344
1345	/*
1346	 * Remember one of the remaining children so we can get tvd below.
1347	 */
1348	cvd = pvd->vdev_child[0];
1349
1350	/*
1351	 * If the parent mirror/replacing vdev only has one child,
1352	 * the parent is no longer needed.  Remove it from the tree.
1353	 */
1354	if (pvd->vdev_children == 1)
1355		vdev_remove_parent(cvd);
1356
1357	/*
1358	 * We don't set tvd until now because the parent we just removed
1359	 * may have been the previous top-level vdev.
1360	 */
1361	tvd = cvd->vdev_top;
1362	ASSERT(tvd->vdev_parent == rvd);
1363
1364	/*
1365	 * Reopen this top-level vdev to reassess health after detach.
1366	 */
1367	vdev_reopen(tvd);
1368
1369	/*
1370	 * If the device we just detached was smaller than the others,
1371	 * it may be possible to add metaslabs (i.e. grow the pool).
1372	 * vdev_metaslab_init() can't fail because the existing metaslabs
1373	 * are already in core, so there's nothing to read from disk.
1374	 */
1375	VERIFY(vdev_metaslab_init(tvd, txg) == 0);
1376
1377	vdev_config_dirty(tvd);
1378
1379	/*
1380	 * Mark vd's DTL as dirty in this txg.
1381	 * vdev_dtl_sync() will see that vd->vdev_detached is set
1382	 * and free vd's DTL object in syncing context.
1383	 * But first make sure we're not on any *other* txg's DTL list,
1384	 * to prevent vd from being accessed after it's freed.
1385	 */
1386	for (t = 0; t < TXG_SIZE; t++)
1387		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
1388	vd->vdev_detached = B_TRUE;
1389	vdev_dirty(tvd, VDD_DTL, vd, txg);
1390
1391	dprintf("detached %s in txg %llu\n", vd->vdev_path, txg);
1392
1393	return (spa_vdev_exit(spa, vd, txg, 0));
1394}
1395
1396/*
1397 * Find any device that's done replacing, so we can detach it.
1398 */
1399static vdev_t *
1400spa_vdev_replace_done_hunt(vdev_t *vd)
1401{
1402	vdev_t *newvd, *oldvd;
1403	int c;
1404
1405	for (c = 0; c < vd->vdev_children; c++) {
1406		oldvd = spa_vdev_replace_done_hunt(vd->vdev_child[c]);
1407		if (oldvd != NULL)
1408			return (oldvd);
1409	}
1410
1411	if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
1412		oldvd = vd->vdev_child[0];
1413		newvd = vd->vdev_child[1];
1414
1415		mutex_enter(&newvd->vdev_dtl_lock);
1416		if (newvd->vdev_dtl_map.sm_space == 0 &&
1417		    newvd->vdev_dtl_scrub.sm_space == 0) {
1418			mutex_exit(&newvd->vdev_dtl_lock);
1419			return (oldvd);
1420		}
1421		mutex_exit(&newvd->vdev_dtl_lock);
1422	}
1423
1424	return (NULL);
1425}
1426
1427static void
1428spa_vdev_replace_done(spa_t *spa)
1429{
1430	vdev_t *vd;
1431	uint64_t guid;
1432
1433	spa_config_enter(spa, RW_READER, FTAG);
1434
1435	while ((vd = spa_vdev_replace_done_hunt(spa->spa_root_vdev)) != NULL) {
1436		guid = vd->vdev_guid;
1437		spa_config_exit(spa, FTAG);
1438		if (spa_vdev_detach(spa, guid, B_TRUE) != 0)
1439			return;
1440		spa_config_enter(spa, RW_READER, FTAG);
1441	}
1442
1443	spa_config_exit(spa, FTAG);
1444}
1445
1446/*
1447 * Update the stored path for this vdev.  Dirty the vdev configuration, relying
1448 * on spa_vdev_enter/exit() to synchronize the labels and cache.
1449 */
1450int
1451spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
1452{
1453	vdev_t *rvd, *vd;
1454	uint64_t txg;
1455
1456	rvd = spa->spa_root_vdev;
1457
1458	txg = spa_vdev_enter(spa);
1459
1460	if ((vd = vdev_lookup_by_guid(rvd, guid)) == NULL)
1461		return (spa_vdev_exit(spa, NULL, txg, ENOENT));
1462
1463	if (!vd->vdev_ops->vdev_op_leaf)
1464		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1465
1466	spa_strfree(vd->vdev_path);
1467	vd->vdev_path = spa_strdup(newpath);
1468
1469	vdev_config_dirty(vd->vdev_top);
1470
1471	return (spa_vdev_exit(spa, NULL, txg, 0));
1472}
1473
1474/*
1475 * ==========================================================================
1476 * SPA Scrubbing
1477 * ==========================================================================
1478 */
1479
1480void
1481spa_scrub_throttle(spa_t *spa, int direction)
1482{
1483	mutex_enter(&spa->spa_scrub_lock);
1484	spa->spa_scrub_throttled += direction;
1485	ASSERT(spa->spa_scrub_throttled >= 0);
1486	if (spa->spa_scrub_throttled == 0)
1487		cv_broadcast(&spa->spa_scrub_io_cv);
1488	mutex_exit(&spa->spa_scrub_lock);
1489}
1490
1491static void
1492spa_scrub_io_done(zio_t *zio)
1493{
1494	spa_t *spa = zio->io_spa;
1495
1496	zio_buf_free(zio->io_data, zio->io_size);
1497
1498	mutex_enter(&spa->spa_scrub_lock);
1499	if (zio->io_error && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
1500		vdev_t *vd = zio->io_vd;
1501		spa->spa_scrub_errors++;
1502		mutex_enter(&vd->vdev_stat_lock);
1503		vd->vdev_stat.vs_scrub_errors++;
1504		mutex_exit(&vd->vdev_stat_lock);
1505	}
1506	if (--spa->spa_scrub_inflight == 0) {
1507		cv_broadcast(&spa->spa_scrub_io_cv);
1508		ASSERT(spa->spa_scrub_throttled == 0);
1509	}
1510	mutex_exit(&spa->spa_scrub_lock);
1511}
1512
1513static void
1514spa_scrub_io_start(spa_t *spa, blkptr_t *bp, int priority, int flags,
1515    zbookmark_t *zb)
1516{
1517	size_t size = BP_GET_LSIZE(bp);
1518	void *data = zio_buf_alloc(size);
1519
1520	mutex_enter(&spa->spa_scrub_lock);
1521	spa->spa_scrub_inflight++;
1522	mutex_exit(&spa->spa_scrub_lock);
1523
1524	if (zb->zb_level == -1 && BP_GET_TYPE(bp) != DMU_OT_OBJSET)
1525		flags |= ZIO_FLAG_SPECULATIVE;	/* intent log block */
1526
1527	flags |= ZIO_FLAG_CANFAIL;
1528
1529	zio_nowait(zio_read(NULL, spa, bp, data, size,
1530	    spa_scrub_io_done, NULL, priority, flags, zb));
1531}
1532
1533/* ARGSUSED */
1534static int
1535spa_scrub_cb(traverse_blk_cache_t *bc, spa_t *spa, void *a)
1536{
1537	blkptr_t *bp = &bc->bc_blkptr;
1538	vdev_t *vd = vdev_lookup_top(spa, DVA_GET_VDEV(&bp->blk_dva[0]));
1539
1540	if (bc->bc_errno || vd == NULL) {
1541		/*
1542		 * We can't scrub this block, but we can continue to scrub
1543		 * the rest of the pool.  Note the error and move along.
1544		 */
1545		mutex_enter(&spa->spa_scrub_lock);
1546		spa->spa_scrub_errors++;
1547		mutex_exit(&spa->spa_scrub_lock);
1548
1549		if (vd != NULL) {
1550			mutex_enter(&vd->vdev_stat_lock);
1551			vd->vdev_stat.vs_scrub_errors++;
1552			mutex_exit(&vd->vdev_stat_lock);
1553		}
1554
1555		return (ERESTART);
1556	}
1557
1558	ASSERT(bp->blk_birth < spa->spa_scrub_maxtxg);
1559
1560	/*
1561	 * Keep track of how much data we've examined so that
1562	 * zpool(1M) status can make useful progress reports.
1563	 */
1564	mutex_enter(&vd->vdev_stat_lock);
1565	vd->vdev_stat.vs_scrub_examined += BP_GET_ASIZE(bp);
1566	mutex_exit(&vd->vdev_stat_lock);
1567
1568	if (spa->spa_scrub_type == POOL_SCRUB_RESILVER) {
1569		if (DVA_GET_GANG(&bp->blk_dva[0])) {
1570			/*
1571			 * Gang members may be spread across multiple vdevs,
1572			 * so the best we can do is look at the pool-wide DTL.
1573			 * XXX -- it would be better to change our allocation
1574			 * policy to ensure that this can't happen.
1575			 */
1576			vd = spa->spa_root_vdev;
1577		}
1578		if (vdev_dtl_contains(&vd->vdev_dtl_map, bp->blk_birth, 1)) {
1579			spa_scrub_io_start(spa, bp, ZIO_PRIORITY_RESILVER,
1580			    ZIO_FLAG_RESILVER, &bc->bc_bookmark);
1581		}
1582	} else {
1583		spa_scrub_io_start(spa, bp, ZIO_PRIORITY_SCRUB,
1584		    ZIO_FLAG_SCRUB, &bc->bc_bookmark);
1585	}
1586
1587	return (0);
1588}
1589
1590static void
1591spa_scrub_thread(spa_t *spa)
1592{
1593	callb_cpr_t cprinfo;
1594	traverse_handle_t *th = spa->spa_scrub_th;
1595	vdev_t *rvd = spa->spa_root_vdev;
1596	pool_scrub_type_t scrub_type = spa->spa_scrub_type;
1597	int error = 0;
1598	boolean_t complete;
1599
1600	CALLB_CPR_INIT(&cprinfo, &spa->spa_scrub_lock, callb_generic_cpr, FTAG);
1601
1602	/*
1603	 * If we're restarting due to a snapshot create/delete,
1604	 * wait for that to complete.
1605	 */
1606	txg_wait_synced(spa_get_dsl(spa), 0);
1607
1608	dprintf("start %s mintxg=%llu maxtxg=%llu\n",
1609	    scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub",
1610	    spa->spa_scrub_mintxg, spa->spa_scrub_maxtxg);
1611
1612	spa_config_enter(spa, RW_WRITER, FTAG);
1613	vdev_reopen(rvd);		/* purge all vdev caches */
1614	vdev_config_dirty(rvd);		/* rewrite all disk labels */
1615	vdev_scrub_stat_update(rvd, scrub_type, B_FALSE);
1616	spa_config_exit(spa, FTAG);
1617
1618	mutex_enter(&spa->spa_scrub_lock);
1619	spa->spa_scrub_errors = 0;
1620	spa->spa_scrub_active = 1;
1621	ASSERT(spa->spa_scrub_inflight == 0);
1622	ASSERT(spa->spa_scrub_throttled == 0);
1623
1624	while (!spa->spa_scrub_stop) {
1625		CALLB_CPR_SAFE_BEGIN(&cprinfo);
1626		while (spa->spa_scrub_suspended) {
1627			spa->spa_scrub_active = 0;
1628			cv_broadcast(&spa->spa_scrub_cv);
1629			cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
1630			spa->spa_scrub_active = 1;
1631		}
1632		CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_scrub_lock);
1633
1634		if (spa->spa_scrub_restart_txg != 0)
1635			break;
1636
1637		mutex_exit(&spa->spa_scrub_lock);
1638		error = traverse_more(th);
1639		mutex_enter(&spa->spa_scrub_lock);
1640		if (error != EAGAIN)
1641			break;
1642
1643		while (spa->spa_scrub_throttled > 0)
1644			cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1645	}
1646
1647	while (spa->spa_scrub_inflight)
1648		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1649
1650	spa->spa_scrub_active = 0;
1651	cv_broadcast(&spa->spa_scrub_cv);
1652
1653	mutex_exit(&spa->spa_scrub_lock);
1654
1655	spa_config_enter(spa, RW_WRITER, FTAG);
1656
1657	mutex_enter(&spa->spa_scrub_lock);
1658
1659	/*
1660	 * Note: we check spa_scrub_restart_txg under both spa_scrub_lock
1661	 * AND the spa config lock to synchronize with any config changes
1662	 * that revise the DTLs under spa_vdev_enter() / spa_vdev_exit().
1663	 */
1664	if (spa->spa_scrub_restart_txg != 0)
1665		error = ERESTART;
1666
1667	if (spa->spa_scrub_stop)
1668		error = EINTR;
1669
1670	/*
1671	 * Even if there were uncorrectable errors, we consider the scrub
1672	 * completed.  The downside is that if there is a transient error during
1673	 * a resilver, we won't resilver the data properly to the target.  But
1674	 * if the damage is permanent (more likely) we will resilver forever,
1675	 * which isn't really acceptable.  Since there is enough information for
1676	 * the user to know what has failed and why, this seems like a more
1677	 * tractable approach.
1678	 */
1679	complete = (error == 0);
1680
1681	dprintf("end %s to maxtxg=%llu %s, traverse=%d, %llu errors, stop=%u\n",
1682	    scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub",
1683	    spa->spa_scrub_maxtxg, complete ? "done" : "FAILED",
1684	    error, spa->spa_scrub_errors, spa->spa_scrub_stop);
1685
1686	mutex_exit(&spa->spa_scrub_lock);
1687
1688	/*
1689	 * If the scrub/resilver completed, update all DTLs to reflect this.
1690	 * Whether it succeeded or not, vacate all temporary scrub DTLs.
1691	 */
1692	vdev_dtl_reassess(rvd, spa_last_synced_txg(spa) + 1,
1693	    complete ? spa->spa_scrub_maxtxg : 0, B_TRUE);
1694	vdev_scrub_stat_update(rvd, POOL_SCRUB_NONE, complete);
1695	spa_errlog_rotate(spa);
1696
1697	spa_config_exit(spa, FTAG);
1698
1699	mutex_enter(&spa->spa_scrub_lock);
1700
1701	/*
1702	 * We may have finished replacing a device.
1703	 * Let the async thread assess this and handle the detach.
1704	 */
1705	spa_async_request(spa, SPA_ASYNC_REPLACE_DONE);
1706
1707	/*
1708	 * If we were told to restart, our final act is to start a new scrub.
1709	 */
1710	if (error == ERESTART)
1711		spa_async_request(spa, scrub_type == POOL_SCRUB_RESILVER ?
1712		    SPA_ASYNC_RESILVER : SPA_ASYNC_SCRUB);
1713
1714	spa->spa_scrub_type = POOL_SCRUB_NONE;
1715	spa->spa_scrub_active = 0;
1716	spa->spa_scrub_thread = NULL;
1717	cv_broadcast(&spa->spa_scrub_cv);
1718	CALLB_CPR_EXIT(&cprinfo);	/* drops &spa->spa_scrub_lock */
1719	thread_exit();
1720}
1721
1722void
1723spa_scrub_suspend(spa_t *spa)
1724{
1725	mutex_enter(&spa->spa_scrub_lock);
1726	spa->spa_scrub_suspended++;
1727	while (spa->spa_scrub_active) {
1728		cv_broadcast(&spa->spa_scrub_cv);
1729		cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
1730	}
1731	while (spa->spa_scrub_inflight)
1732		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1733	mutex_exit(&spa->spa_scrub_lock);
1734}
1735
1736void
1737spa_scrub_resume(spa_t *spa)
1738{
1739	mutex_enter(&spa->spa_scrub_lock);
1740	ASSERT(spa->spa_scrub_suspended != 0);
1741	if (--spa->spa_scrub_suspended == 0)
1742		cv_broadcast(&spa->spa_scrub_cv);
1743	mutex_exit(&spa->spa_scrub_lock);
1744}
1745
1746void
1747spa_scrub_restart(spa_t *spa, uint64_t txg)
1748{
1749	/*
1750	 * Something happened (e.g. snapshot create/delete) that means
1751	 * we must restart any in-progress scrubs.  The itinerary will
1752	 * fix this properly.
1753	 */
1754	mutex_enter(&spa->spa_scrub_lock);
1755	spa->spa_scrub_restart_txg = txg;
1756	mutex_exit(&spa->spa_scrub_lock);
1757}
1758
1759int
1760spa_scrub(spa_t *spa, pool_scrub_type_t type, boolean_t force)
1761{
1762	space_seg_t *ss;
1763	uint64_t mintxg, maxtxg;
1764	vdev_t *rvd = spa->spa_root_vdev;
1765
1766	if ((uint_t)type >= POOL_SCRUB_TYPES)
1767		return (ENOTSUP);
1768
1769	mutex_enter(&spa->spa_scrub_lock);
1770
1771	/*
1772	 * If there's a scrub or resilver already in progress, stop it.
1773	 */
1774	while (spa->spa_scrub_thread != NULL) {
1775		/*
1776		 * Don't stop a resilver unless forced.
1777		 */
1778		if (spa->spa_scrub_type == POOL_SCRUB_RESILVER && !force) {
1779			mutex_exit(&spa->spa_scrub_lock);
1780			return (EBUSY);
1781		}
1782		spa->spa_scrub_stop = 1;
1783		cv_broadcast(&spa->spa_scrub_cv);
1784		cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
1785	}
1786
1787	/*
1788	 * Terminate the previous traverse.
1789	 */
1790	if (spa->spa_scrub_th != NULL) {
1791		traverse_fini(spa->spa_scrub_th);
1792		spa->spa_scrub_th = NULL;
1793	}
1794
1795	if (rvd == NULL) {
1796		ASSERT(spa->spa_scrub_stop == 0);
1797		ASSERT(spa->spa_scrub_type == type);
1798		ASSERT(spa->spa_scrub_restart_txg == 0);
1799		mutex_exit(&spa->spa_scrub_lock);
1800		return (0);
1801	}
1802
1803	mintxg = TXG_INITIAL - 1;
1804	maxtxg = spa_last_synced_txg(spa) + 1;
1805
1806	mutex_enter(&rvd->vdev_dtl_lock);
1807
1808	if (rvd->vdev_dtl_map.sm_space == 0) {
1809		/*
1810		 * The pool-wide DTL is empty.
1811		 * If this is a resilver, there's nothing to do except
1812		 * check whether any in-progress replacements have completed.
1813		 */
1814		if (type == POOL_SCRUB_RESILVER) {
1815			type = POOL_SCRUB_NONE;
1816			spa_async_request(spa, SPA_ASYNC_REPLACE_DONE);
1817		}
1818	} else {
1819		/*
1820		 * The pool-wide DTL is non-empty.
1821		 * If this is a normal scrub, upgrade to a resilver instead.
1822		 */
1823		if (type == POOL_SCRUB_EVERYTHING)
1824			type = POOL_SCRUB_RESILVER;
1825	}
1826
1827	if (type == POOL_SCRUB_RESILVER) {
1828		/*
1829		 * Determine the resilvering boundaries.
1830		 *
1831		 * Note: (mintxg, maxtxg) is an open interval,
1832		 * i.e. mintxg and maxtxg themselves are not included.
1833		 *
1834		 * Note: for maxtxg, we MIN with spa_last_synced_txg(spa) + 1
1835		 * so we don't claim to resilver a txg that's still changing.
1836		 */
1837		ss = avl_first(&rvd->vdev_dtl_map.sm_root);
1838		mintxg = ss->ss_start - 1;
1839		ss = avl_last(&rvd->vdev_dtl_map.sm_root);
1840		maxtxg = MIN(ss->ss_end, maxtxg);
1841	}
1842
1843	mutex_exit(&rvd->vdev_dtl_lock);
1844
1845	spa->spa_scrub_stop = 0;
1846	spa->spa_scrub_type = type;
1847	spa->spa_scrub_restart_txg = 0;
1848
1849	if (type != POOL_SCRUB_NONE) {
1850		spa->spa_scrub_mintxg = mintxg;
1851		spa->spa_scrub_maxtxg = maxtxg;
1852		spa->spa_scrub_th = traverse_init(spa, spa_scrub_cb, NULL,
1853		    ADVANCE_PRE | ADVANCE_PRUNE | ADVANCE_ZIL,
1854		    ZIO_FLAG_CANFAIL);
1855		traverse_add_pool(spa->spa_scrub_th, mintxg, maxtxg);
1856		spa->spa_scrub_thread = thread_create(NULL, 0,
1857		    spa_scrub_thread, spa, 0, &p0, TS_RUN, minclsyspri);
1858	}
1859
1860	mutex_exit(&spa->spa_scrub_lock);
1861
1862	return (0);
1863}
1864
1865/*
1866 * ==========================================================================
1867 * SPA async task processing
1868 * ==========================================================================
1869 */
1870
1871static void
1872spa_async_reopen(spa_t *spa)
1873{
1874	vdev_t *rvd = spa->spa_root_vdev;
1875	vdev_t *tvd;
1876	int c;
1877
1878	spa_config_enter(spa, RW_WRITER, FTAG);
1879
1880	for (c = 0; c < rvd->vdev_children; c++) {
1881		tvd = rvd->vdev_child[c];
1882		if (tvd->vdev_reopen_wanted) {
1883			tvd->vdev_reopen_wanted = 0;
1884			vdev_reopen(tvd);
1885		}
1886	}
1887
1888	spa_config_exit(spa, FTAG);
1889}
1890
1891static void
1892spa_async_thread(spa_t *spa)
1893{
1894	int tasks;
1895
1896	ASSERT(spa->spa_sync_on);
1897
1898	mutex_enter(&spa->spa_async_lock);
1899	tasks = spa->spa_async_tasks;
1900	spa->spa_async_tasks = 0;
1901	mutex_exit(&spa->spa_async_lock);
1902
1903	/*
1904	 * See if the config needs to be updated.
1905	 */
1906	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
1907		mutex_enter(&spa_namespace_lock);
1908		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
1909		mutex_exit(&spa_namespace_lock);
1910	}
1911
1912	/*
1913	 * See if any devices need to be reopened.
1914	 */
1915	if (tasks & SPA_ASYNC_REOPEN)
1916		spa_async_reopen(spa);
1917
1918	/*
1919	 * If any devices are done replacing, detach them.
1920	 */
1921	if (tasks & SPA_ASYNC_REPLACE_DONE)
1922		spa_vdev_replace_done(spa);
1923
1924	/*
1925	 * Kick off a scrub.
1926	 */
1927	if (tasks & SPA_ASYNC_SCRUB)
1928		VERIFY(spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_TRUE) == 0);
1929
1930	/*
1931	 * Kick off a resilver.
1932	 */
1933	if (tasks & SPA_ASYNC_RESILVER)
1934		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
1935
1936	/*
1937	 * Let the world know that we're done.
1938	 */
1939	mutex_enter(&spa->spa_async_lock);
1940	spa->spa_async_thread = NULL;
1941	cv_broadcast(&spa->spa_async_cv);
1942	mutex_exit(&spa->spa_async_lock);
1943	thread_exit();
1944}
1945
1946void
1947spa_async_suspend(spa_t *spa)
1948{
1949	mutex_enter(&spa->spa_async_lock);
1950	spa->spa_async_suspended++;
1951	while (spa->spa_async_thread != NULL)
1952		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
1953	mutex_exit(&spa->spa_async_lock);
1954}
1955
1956void
1957spa_async_resume(spa_t *spa)
1958{
1959	mutex_enter(&spa->spa_async_lock);
1960	ASSERT(spa->spa_async_suspended != 0);
1961	spa->spa_async_suspended--;
1962	mutex_exit(&spa->spa_async_lock);
1963}
1964
1965static void
1966spa_async_dispatch(spa_t *spa)
1967{
1968	mutex_enter(&spa->spa_async_lock);
1969	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
1970	    spa->spa_async_thread == NULL &&
1971	    rootdir != NULL && !vn_is_readonly(rootdir))
1972		spa->spa_async_thread = thread_create(NULL, 0,
1973		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
1974	mutex_exit(&spa->spa_async_lock);
1975}
1976
1977void
1978spa_async_request(spa_t *spa, int task)
1979{
1980	mutex_enter(&spa->spa_async_lock);
1981	spa->spa_async_tasks |= task;
1982	mutex_exit(&spa->spa_async_lock);
1983}
1984
1985/*
1986 * ==========================================================================
1987 * SPA syncing routines
1988 * ==========================================================================
1989 */
1990
1991static void
1992spa_sync_deferred_frees(spa_t *spa, uint64_t txg)
1993{
1994	bplist_t *bpl = &spa->spa_sync_bplist;
1995	dmu_tx_t *tx;
1996	blkptr_t blk;
1997	uint64_t itor = 0;
1998	zio_t *zio;
1999	int error;
2000	uint8_t c = 1;
2001
2002	zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CONFIG_HELD);
2003
2004	while (bplist_iterate(bpl, &itor, &blk) == 0)
2005		zio_nowait(zio_free(zio, spa, txg, &blk, NULL, NULL));
2006
2007	error = zio_wait(zio);
2008	ASSERT3U(error, ==, 0);
2009
2010	tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2011	bplist_vacate(bpl, tx);
2012
2013	/*
2014	 * Pre-dirty the first block so we sync to convergence faster.
2015	 * (Usually only the first block is needed.)
2016	 */
2017	dmu_write(spa->spa_meta_objset, spa->spa_sync_bplist_obj, 0, 1, &c, tx);
2018	dmu_tx_commit(tx);
2019}
2020
2021static void
2022spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
2023{
2024	nvlist_t *config;
2025	char *packed = NULL;
2026	size_t nvsize = 0;
2027	dmu_buf_t *db;
2028
2029	if (list_is_empty(&spa->spa_dirty_list))
2030		return;
2031
2032	config = spa_config_generate(spa, NULL, dmu_tx_get_txg(tx), B_FALSE);
2033
2034	if (spa->spa_config_syncing)
2035		nvlist_free(spa->spa_config_syncing);
2036	spa->spa_config_syncing = config;
2037
2038	VERIFY(nvlist_size(config, &nvsize, NV_ENCODE_XDR) == 0);
2039
2040	packed = kmem_alloc(nvsize, KM_SLEEP);
2041
2042	VERIFY(nvlist_pack(config, &packed, &nvsize, NV_ENCODE_XDR,
2043	    KM_SLEEP) == 0);
2044
2045	dmu_write(spa->spa_meta_objset, spa->spa_config_object, 0, nvsize,
2046	    packed, tx);
2047
2048	kmem_free(packed, nvsize);
2049
2050	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset,
2051	    spa->spa_config_object, FTAG, &db));
2052	dmu_buf_will_dirty(db, tx);
2053	*(uint64_t *)db->db_data = nvsize;
2054	dmu_buf_rele(db, FTAG);
2055}
2056
2057/*
2058 * Sync the specified transaction group.  New blocks may be dirtied as
2059 * part of the process, so we iterate until it converges.
2060 */
2061void
2062spa_sync(spa_t *spa, uint64_t txg)
2063{
2064	dsl_pool_t *dp = spa->spa_dsl_pool;
2065	objset_t *mos = spa->spa_meta_objset;
2066	bplist_t *bpl = &spa->spa_sync_bplist;
2067	vdev_t *rvd = spa->spa_root_vdev;
2068	vdev_t *vd;
2069	dmu_tx_t *tx;
2070	int dirty_vdevs;
2071
2072	/*
2073	 * Lock out configuration changes.
2074	 */
2075	spa_config_enter(spa, RW_READER, FTAG);
2076
2077	spa->spa_syncing_txg = txg;
2078	spa->spa_sync_pass = 0;
2079
2080	VERIFY(0 == bplist_open(bpl, mos, spa->spa_sync_bplist_obj));
2081
2082	/*
2083	 * If anything has changed in this txg, push the deferred frees
2084	 * from the previous txg.  If not, leave them alone so that we
2085	 * don't generate work on an otherwise idle system.
2086	 */
2087	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
2088	    !txg_list_empty(&dp->dp_dirty_dirs, txg))
2089		spa_sync_deferred_frees(spa, txg);
2090
2091	/*
2092	 * Iterate to convergence.
2093	 */
2094	do {
2095		spa->spa_sync_pass++;
2096
2097		tx = dmu_tx_create_assigned(dp, txg);
2098		spa_sync_config_object(spa, tx);
2099		dmu_tx_commit(tx);
2100
2101		spa_errlog_sync(spa, txg);
2102
2103		dsl_pool_sync(dp, txg);
2104
2105		dirty_vdevs = 0;
2106		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) {
2107			vdev_sync(vd, txg);
2108			dirty_vdevs++;
2109		}
2110
2111		tx = dmu_tx_create_assigned(dp, txg);
2112		bplist_sync(bpl, tx);
2113		dmu_tx_commit(tx);
2114
2115	} while (dirty_vdevs);
2116
2117	bplist_close(bpl);
2118
2119	dprintf("txg %llu passes %d\n", txg, spa->spa_sync_pass);
2120
2121	/*
2122	 * Rewrite the vdev configuration (which includes the uberblock)
2123	 * to commit the transaction group.
2124	 *
2125	 * If there are any dirty vdevs, sync the uberblock to all vdevs.
2126	 * Otherwise, pick a random top-level vdev that's known to be
2127	 * visible in the config cache (see spa_vdev_add() for details).
2128	 * If the write fails, try the next vdev until we're tried them all.
2129	 */
2130	if (!list_is_empty(&spa->spa_dirty_list)) {
2131		VERIFY(vdev_config_sync(rvd, txg) == 0);
2132	} else {
2133		int children = rvd->vdev_children;
2134		int c0 = spa_get_random(children);
2135		int c;
2136
2137		for (c = 0; c < children; c++) {
2138			vd = rvd->vdev_child[(c0 + c) % children];
2139			if (vd->vdev_ms_array == 0)
2140				continue;
2141			if (vdev_config_sync(vd, txg) == 0)
2142				break;
2143		}
2144		if (c == children)
2145			VERIFY(vdev_config_sync(rvd, txg) == 0);
2146	}
2147
2148	/*
2149	 * Clear the dirty config list.
2150	 */
2151	while ((vd = list_head(&spa->spa_dirty_list)) != NULL)
2152		vdev_config_clean(vd);
2153
2154	/*
2155	 * Now that the new config has synced transactionally,
2156	 * let it become visible to the config cache.
2157	 */
2158	if (spa->spa_config_syncing != NULL) {
2159		spa_config_set(spa, spa->spa_config_syncing);
2160		spa->spa_config_txg = txg;
2161		spa->spa_config_syncing = NULL;
2162	}
2163
2164	/*
2165	 * Make a stable copy of the fully synced uberblock.
2166	 * We use this as the root for pool traversals.
2167	 */
2168	spa->spa_traverse_wanted = 1;	/* tells traverse_more() to stop */
2169
2170	spa_scrub_suspend(spa);		/* stop scrubbing and finish I/Os */
2171
2172	rw_enter(&spa->spa_traverse_lock, RW_WRITER);
2173	spa->spa_traverse_wanted = 0;
2174	spa->spa_ubsync = spa->spa_uberblock;
2175	rw_exit(&spa->spa_traverse_lock);
2176
2177	spa_scrub_resume(spa);		/* resume scrub with new ubsync */
2178
2179	/*
2180	 * Clean up the ZIL records for the synced txg.
2181	 */
2182	dsl_pool_zil_clean(dp);
2183
2184	/*
2185	 * Update usable space statistics.
2186	 */
2187	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
2188		vdev_sync_done(vd, txg);
2189
2190	/*
2191	 * It had better be the case that we didn't dirty anything
2192	 * since spa_sync_labels().
2193	 */
2194	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
2195	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
2196	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
2197	ASSERT(bpl->bpl_queue == NULL);
2198
2199	spa_config_exit(spa, FTAG);
2200
2201	/*
2202	 * If any async tasks have been requested, kick them off.
2203	 */
2204	spa_async_dispatch(spa);
2205}
2206
2207/*
2208 * Sync all pools.  We don't want to hold the namespace lock across these
2209 * operations, so we take a reference on the spa_t and drop the lock during the
2210 * sync.
2211 */
2212void
2213spa_sync_allpools(void)
2214{
2215	spa_t *spa = NULL;
2216	mutex_enter(&spa_namespace_lock);
2217	while ((spa = spa_next(spa)) != NULL) {
2218		if (spa_state(spa) != POOL_STATE_ACTIVE)
2219			continue;
2220		spa_open_ref(spa, FTAG);
2221		mutex_exit(&spa_namespace_lock);
2222		txg_wait_synced(spa_get_dsl(spa), 0);
2223		mutex_enter(&spa_namespace_lock);
2224		spa_close(spa, FTAG);
2225	}
2226	mutex_exit(&spa_namespace_lock);
2227}
2228
2229/*
2230 * ==========================================================================
2231 * Miscellaneous routines
2232 * ==========================================================================
2233 */
2234
2235/*
2236 * Remove all pools in the system.
2237 */
2238void
2239spa_evict_all(void)
2240{
2241	spa_t *spa;
2242
2243	/*
2244	 * Remove all cached state.  All pools should be closed now,
2245	 * so every spa in the AVL tree should be unreferenced.
2246	 */
2247	mutex_enter(&spa_namespace_lock);
2248	while ((spa = spa_next(NULL)) != NULL) {
2249		/*
2250		 * Stop async tasks.  The async thread may need to detach
2251		 * a device that's been replaced, which requires grabbing
2252		 * spa_namespace_lock, so we must drop it here.
2253		 */
2254		spa_open_ref(spa, FTAG);
2255		mutex_exit(&spa_namespace_lock);
2256		spa_async_suspend(spa);
2257		VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0);
2258		mutex_enter(&spa_namespace_lock);
2259		spa_close(spa, FTAG);
2260
2261		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
2262			spa_unload(spa);
2263			spa_deactivate(spa);
2264		}
2265		spa_remove(spa);
2266	}
2267	mutex_exit(&spa_namespace_lock);
2268}
2269
2270vdev_t *
2271spa_lookup_by_guid(spa_t *spa, uint64_t guid)
2272{
2273	return (vdev_lookup_by_guid(spa->spa_root_vdev, guid));
2274}
2275
2276void
2277spa_upgrade(spa_t *spa)
2278{
2279	spa_config_enter(spa, RW_WRITER, FTAG);
2280
2281	/*
2282	 * This should only be called for a non-faulted pool, and since a
2283	 * future version would result in an unopenable pool, this shouldn't be
2284	 * possible.
2285	 */
2286	ASSERT(spa->spa_uberblock.ub_version <= ZFS_VERSION);
2287
2288	spa->spa_uberblock.ub_version = ZFS_VERSION;
2289	vdev_config_dirty(spa->spa_root_vdev);
2290
2291	spa_config_exit(spa, FTAG);
2292}
2293