spa.c revision 0b69c2f001a429251e2d38f25aca860396551214
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/*
23 * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
24 * Use is subject to license terms.
25 */
26
27#pragma ident	"%Z%%M%	%I%	%E% SMI"
28
29/*
30 * This file contains all the routines used when modifying on-disk SPA state.
31 * This includes opening, importing, destroying, exporting a pool, and syncing a
32 * pool.
33 */
34
35#include <sys/zfs_context.h>
36#include <sys/fm/fs/zfs.h>
37#include <sys/spa_impl.h>
38#include <sys/zio.h>
39#include <sys/zio_checksum.h>
40#include <sys/zio_compress.h>
41#include <sys/dmu.h>
42#include <sys/dmu_tx.h>
43#include <sys/zap.h>
44#include <sys/zil.h>
45#include <sys/vdev_impl.h>
46#include <sys/metaslab.h>
47#include <sys/uberblock_impl.h>
48#include <sys/txg.h>
49#include <sys/avl.h>
50#include <sys/dmu_traverse.h>
51#include <sys/unique.h>
52#include <sys/dsl_pool.h>
53#include <sys/dsl_dir.h>
54#include <sys/dsl_prop.h>
55#include <sys/fs/zfs.h>
56#include <sys/callb.h>
57
58/*
59 * ==========================================================================
60 * SPA state manipulation (open/create/destroy/import/export)
61 * ==========================================================================
62 */
63
64static int
65spa_error_entry_compare(const void *a, const void *b)
66{
67	spa_error_entry_t *sa = (spa_error_entry_t *)a;
68	spa_error_entry_t *sb = (spa_error_entry_t *)b;
69	int ret;
70
71	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
72	    sizeof (zbookmark_t));
73
74	if (ret < 0)
75		return (-1);
76	else if (ret > 0)
77		return (1);
78	else
79		return (0);
80}
81
82/*
83 * Utility function which retrieves copies of the current logs and
84 * re-initializes them in the process.
85 */
86void
87spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
88{
89	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
90
91	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
92	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
93
94	avl_create(&spa->spa_errlist_scrub,
95	    spa_error_entry_compare, sizeof (spa_error_entry_t),
96	    offsetof(spa_error_entry_t, se_avl));
97	avl_create(&spa->spa_errlist_last,
98	    spa_error_entry_compare, sizeof (spa_error_entry_t),
99	    offsetof(spa_error_entry_t, se_avl));
100}
101
102/*
103 * Activate an uninitialized pool.
104 */
105static void
106spa_activate(spa_t *spa)
107{
108	int t;
109
110	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
111
112	spa->spa_state = POOL_STATE_ACTIVE;
113
114	spa->spa_normal_class = metaslab_class_create();
115
116	for (t = 0; t < ZIO_TYPES; t++) {
117		spa->spa_zio_issue_taskq[t] = taskq_create("spa_zio_issue",
118		    8, maxclsyspri, 50, INT_MAX,
119		    TASKQ_PREPOPULATE);
120		spa->spa_zio_intr_taskq[t] = taskq_create("spa_zio_intr",
121		    8, maxclsyspri, 50, INT_MAX,
122		    TASKQ_PREPOPULATE);
123	}
124
125	rw_init(&spa->spa_traverse_lock, NULL, RW_DEFAULT, NULL);
126
127	list_create(&spa->spa_dirty_list, sizeof (vdev_t),
128	    offsetof(vdev_t, vdev_dirty_node));
129
130	txg_list_create(&spa->spa_vdev_txg_list,
131	    offsetof(struct vdev, vdev_txg_node));
132
133	avl_create(&spa->spa_errlist_scrub,
134	    spa_error_entry_compare, sizeof (spa_error_entry_t),
135	    offsetof(spa_error_entry_t, se_avl));
136	avl_create(&spa->spa_errlist_last,
137	    spa_error_entry_compare, sizeof (spa_error_entry_t),
138	    offsetof(spa_error_entry_t, se_avl));
139}
140
141/*
142 * Opposite of spa_activate().
143 */
144static void
145spa_deactivate(spa_t *spa)
146{
147	int t;
148
149	ASSERT(spa->spa_sync_on == B_FALSE);
150	ASSERT(spa->spa_dsl_pool == NULL);
151	ASSERT(spa->spa_root_vdev == NULL);
152
153	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
154
155	txg_list_destroy(&spa->spa_vdev_txg_list);
156
157	list_destroy(&spa->spa_dirty_list);
158
159	rw_destroy(&spa->spa_traverse_lock);
160
161	for (t = 0; t < ZIO_TYPES; t++) {
162		taskq_destroy(spa->spa_zio_issue_taskq[t]);
163		taskq_destroy(spa->spa_zio_intr_taskq[t]);
164		spa->spa_zio_issue_taskq[t] = NULL;
165		spa->spa_zio_intr_taskq[t] = NULL;
166	}
167
168	metaslab_class_destroy(spa->spa_normal_class);
169	spa->spa_normal_class = NULL;
170
171	/*
172	 * If this was part of an import or the open otherwise failed, we may
173	 * still have errors left in the queues.  Empty them just in case.
174	 */
175	spa_errlog_drain(spa);
176
177	avl_destroy(&spa->spa_errlist_scrub);
178	avl_destroy(&spa->spa_errlist_last);
179
180	spa->spa_state = POOL_STATE_UNINITIALIZED;
181}
182
183/*
184 * Verify a pool configuration, and construct the vdev tree appropriately.  This
185 * will create all the necessary vdevs in the appropriate layout, with each vdev
186 * in the CLOSED state.  This will prep the pool before open/creation/import.
187 * All vdev validation is done by the vdev_alloc() routine.
188 */
189static int
190spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
191    uint_t id, int atype)
192{
193	nvlist_t **child;
194	uint_t c, children;
195	int error;
196
197	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
198		return (error);
199
200	if ((*vdp)->vdev_ops->vdev_op_leaf)
201		return (0);
202
203	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
204	    &child, &children) != 0) {
205		vdev_free(*vdp);
206		*vdp = NULL;
207		return (EINVAL);
208	}
209
210	for (c = 0; c < children; c++) {
211		vdev_t *vd;
212		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
213		    atype)) != 0) {
214			vdev_free(*vdp);
215			*vdp = NULL;
216			return (error);
217		}
218	}
219
220	ASSERT(*vdp != NULL);
221
222	return (0);
223}
224
225/*
226 * Opposite of spa_load().
227 */
228static void
229spa_unload(spa_t *spa)
230{
231	int i;
232
233	/*
234	 * Stop async tasks.
235	 */
236	spa_async_suspend(spa);
237
238	/*
239	 * Stop syncing.
240	 */
241	if (spa->spa_sync_on) {
242		txg_sync_stop(spa->spa_dsl_pool);
243		spa->spa_sync_on = B_FALSE;
244	}
245
246	/*
247	 * Wait for any outstanding prefetch I/O to complete.
248	 */
249	spa_config_enter(spa, RW_WRITER, FTAG);
250	spa_config_exit(spa, FTAG);
251
252	/*
253	 * Close the dsl pool.
254	 */
255	if (spa->spa_dsl_pool) {
256		dsl_pool_close(spa->spa_dsl_pool);
257		spa->spa_dsl_pool = NULL;
258	}
259
260	/*
261	 * Close all vdevs.
262	 */
263	if (spa->spa_root_vdev)
264		vdev_free(spa->spa_root_vdev);
265	ASSERT(spa->spa_root_vdev == NULL);
266
267	for (i = 0; i < spa->spa_nspares; i++)
268		vdev_free(spa->spa_spares[i]);
269	if (spa->spa_spares) {
270		kmem_free(spa->spa_spares, spa->spa_nspares * sizeof (void *));
271		spa->spa_spares = NULL;
272	}
273	if (spa->spa_sparelist) {
274		nvlist_free(spa->spa_sparelist);
275		spa->spa_sparelist = NULL;
276	}
277
278	spa->spa_async_suspended = 0;
279}
280
281/*
282 * Load (or re-load) the current list of vdevs describing the active spares for
283 * this pool.  When this is called, we have some form of basic information in
284 * 'spa_sparelist'.  We parse this into vdevs, try to open them, and then
285 * re-generate a more complete list including status information.
286 */
287static void
288spa_load_spares(spa_t *spa)
289{
290	nvlist_t **spares;
291	uint_t nspares;
292	int i;
293
294	/*
295	 * First, close and free any existing spare vdevs.
296	 */
297	for (i = 0; i < spa->spa_nspares; i++) {
298		vdev_close(spa->spa_spares[i]);
299		vdev_free(spa->spa_spares[i]);
300	}
301	if (spa->spa_spares)
302		kmem_free(spa->spa_spares, spa->spa_nspares * sizeof (void *));
303
304	if (spa->spa_sparelist == NULL)
305		nspares = 0;
306	else
307		VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist,
308		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
309
310	spa->spa_nspares = (int)nspares;
311	spa->spa_spares = NULL;
312
313	if (nspares == 0)
314		return;
315
316	/*
317	 * Construct the array of vdevs, opening them to get status in the
318	 * process.
319	 */
320	spa->spa_spares = kmem_alloc(nspares * sizeof (void *), KM_SLEEP);
321	for (i = 0; i < spa->spa_nspares; i++) {
322		vdev_t *vd;
323
324		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
325		    VDEV_ALLOC_SPARE) == 0);
326		ASSERT(vd != NULL);
327
328		spa->spa_spares[i] = vd;
329
330		if (vdev_open(vd) != 0)
331			continue;
332
333		vd->vdev_top = vd;
334		(void) vdev_validate_spare(vd);
335	}
336
337	/*
338	 * Recompute the stashed list of spares, with status information
339	 * this time.
340	 */
341	VERIFY(nvlist_remove(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
342	    DATA_TYPE_NVLIST_ARRAY) == 0);
343
344	spares = kmem_alloc(spa->spa_nspares * sizeof (void *), KM_SLEEP);
345	for (i = 0; i < spa->spa_nspares; i++)
346		spares[i] = vdev_config_generate(spa, spa->spa_spares[i],
347		    B_TRUE, B_TRUE);
348	VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
349	    spares, spa->spa_nspares) == 0);
350	for (i = 0; i < spa->spa_nspares; i++)
351		nvlist_free(spares[i]);
352	kmem_free(spares, spa->spa_nspares * sizeof (void *));
353}
354
355static int
356load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
357{
358	dmu_buf_t *db;
359	char *packed = NULL;
360	size_t nvsize = 0;
361	int error;
362	*value = NULL;
363
364	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
365	nvsize = *(uint64_t *)db->db_data;
366	dmu_buf_rele(db, FTAG);
367
368	packed = kmem_alloc(nvsize, KM_SLEEP);
369	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed);
370	if (error == 0)
371		error = nvlist_unpack(packed, nvsize, value, 0);
372	kmem_free(packed, nvsize);
373
374	return (error);
375}
376
377/*
378 * Load an existing storage pool, using the pool's builtin spa_config as a
379 * source of configuration information.
380 */
381static int
382spa_load(spa_t *spa, nvlist_t *config, spa_load_state_t state, int mosconfig)
383{
384	int error = 0;
385	nvlist_t *nvroot = NULL;
386	vdev_t *rvd;
387	uberblock_t *ub = &spa->spa_uberblock;
388	uint64_t config_cache_txg = spa->spa_config_txg;
389	uint64_t pool_guid;
390	uint64_t version;
391	zio_t *zio;
392
393	spa->spa_load_state = state;
394
395	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) ||
396	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
397		error = EINVAL;
398		goto out;
399	}
400
401	/*
402	 * Versioning wasn't explicitly added to the label until later, so if
403	 * it's not present treat it as the initial version.
404	 */
405	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &version) != 0)
406		version = ZFS_VERSION_INITIAL;
407
408	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
409	    &spa->spa_config_txg);
410
411	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
412	    spa_guid_exists(pool_guid, 0)) {
413		error = EEXIST;
414		goto out;
415	}
416
417	spa->spa_load_guid = pool_guid;
418
419	/*
420	 * Parse the configuration into a vdev tree.  We explicitly set the
421	 * value that will be returned by spa_version() since parsing the
422	 * configuration requires knowing the version number.
423	 */
424	spa_config_enter(spa, RW_WRITER, FTAG);
425	spa->spa_ubsync.ub_version = version;
426	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_LOAD);
427	spa_config_exit(spa, FTAG);
428
429	if (error != 0)
430		goto out;
431
432	ASSERT(spa->spa_root_vdev == rvd);
433	ASSERT(spa_guid(spa) == pool_guid);
434
435	/*
436	 * Try to open all vdevs, loading each label in the process.
437	 */
438	if (vdev_open(rvd) != 0) {
439		error = ENXIO;
440		goto out;
441	}
442
443	/*
444	 * Validate the labels for all leaf vdevs.  We need to grab the config
445	 * lock because all label I/O is done with the ZIO_FLAG_CONFIG_HELD
446	 * flag.
447	 */
448	spa_config_enter(spa, RW_READER, FTAG);
449	error = vdev_validate(rvd);
450	spa_config_exit(spa, FTAG);
451
452	if (error != 0) {
453		error = EBADF;
454		goto out;
455	}
456
457	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
458		error = ENXIO;
459		goto out;
460	}
461
462	/*
463	 * Find the best uberblock.
464	 */
465	bzero(ub, sizeof (uberblock_t));
466
467	zio = zio_root(spa, NULL, NULL,
468	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
469	vdev_uberblock_load(zio, rvd, ub);
470	error = zio_wait(zio);
471
472	/*
473	 * If we weren't able to find a single valid uberblock, return failure.
474	 */
475	if (ub->ub_txg == 0) {
476		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
477		    VDEV_AUX_CORRUPT_DATA);
478		error = ENXIO;
479		goto out;
480	}
481
482	/*
483	 * If the pool is newer than the code, we can't open it.
484	 */
485	if (ub->ub_version > ZFS_VERSION) {
486		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
487		    VDEV_AUX_VERSION_NEWER);
488		error = ENOTSUP;
489		goto out;
490	}
491
492	/*
493	 * If the vdev guid sum doesn't match the uberblock, we have an
494	 * incomplete configuration.
495	 */
496	if (rvd->vdev_guid_sum != ub->ub_guid_sum && mosconfig) {
497		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
498		    VDEV_AUX_BAD_GUID_SUM);
499		error = ENXIO;
500		goto out;
501	}
502
503	/*
504	 * Initialize internal SPA structures.
505	 */
506	spa->spa_state = POOL_STATE_ACTIVE;
507	spa->spa_ubsync = spa->spa_uberblock;
508	spa->spa_first_txg = spa_last_synced_txg(spa) + 1;
509	error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
510	if (error) {
511		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
512		    VDEV_AUX_CORRUPT_DATA);
513		goto out;
514	}
515	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
516
517	if (zap_lookup(spa->spa_meta_objset,
518	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
519	    sizeof (uint64_t), 1, &spa->spa_config_object) != 0) {
520		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
521		    VDEV_AUX_CORRUPT_DATA);
522		error = EIO;
523		goto out;
524	}
525
526	if (!mosconfig) {
527		nvlist_t *newconfig;
528
529		if (load_nvlist(spa, spa->spa_config_object, &newconfig) != 0) {
530			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
531			    VDEV_AUX_CORRUPT_DATA);
532			error = EIO;
533			goto out;
534		}
535
536		spa_config_set(spa, newconfig);
537		spa_unload(spa);
538		spa_deactivate(spa);
539		spa_activate(spa);
540
541		return (spa_load(spa, newconfig, state, B_TRUE));
542	}
543
544	if (zap_lookup(spa->spa_meta_objset,
545	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
546	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj) != 0) {
547		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
548		    VDEV_AUX_CORRUPT_DATA);
549		error = EIO;
550		goto out;
551	}
552
553	/*
554	 * Load the bit that tells us to use the new accounting function
555	 * (raid-z deflation).  If we have an older pool, this will not
556	 * be present.
557	 */
558	error = zap_lookup(spa->spa_meta_objset,
559	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
560	    sizeof (uint64_t), 1, &spa->spa_deflate);
561	if (error != 0 && error != ENOENT) {
562		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
563		    VDEV_AUX_CORRUPT_DATA);
564		error = EIO;
565		goto out;
566	}
567
568	/*
569	 * Load the persistent error log.  If we have an older pool, this will
570	 * not be present.
571	 */
572	error = zap_lookup(spa->spa_meta_objset,
573	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_LAST,
574	    sizeof (uint64_t), 1, &spa->spa_errlog_last);
575	if (error != 0 && error != ENOENT) {
576		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
577		    VDEV_AUX_CORRUPT_DATA);
578		error = EIO;
579		goto out;
580	}
581
582	error = zap_lookup(spa->spa_meta_objset,
583	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_SCRUB,
584	    sizeof (uint64_t), 1, &spa->spa_errlog_scrub);
585	if (error != 0 && error != ENOENT) {
586		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
587		    VDEV_AUX_CORRUPT_DATA);
588		error = EIO;
589		goto out;
590	}
591
592	/*
593	 * Load any hot spares for this pool.
594	 */
595	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
596	    DMU_POOL_SPARES, sizeof (uint64_t), 1, &spa->spa_spares_object);
597	if (error != 0 && error != ENOENT) {
598		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
599		    VDEV_AUX_CORRUPT_DATA);
600		error = EIO;
601		goto out;
602	}
603	if (error == 0) {
604		ASSERT(spa_version(spa) >= ZFS_VERSION_SPARES);
605		if (load_nvlist(spa, spa->spa_spares_object,
606		    &spa->spa_sparelist) != 0) {
607			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
608			    VDEV_AUX_CORRUPT_DATA);
609			error = EIO;
610			goto out;
611		}
612
613		spa_config_enter(spa, RW_WRITER, FTAG);
614		spa_load_spares(spa);
615		spa_config_exit(spa, FTAG);
616	}
617
618	/*
619	 * Load the vdev state for all toplevel vdevs.
620	 */
621	vdev_load(rvd);
622
623	/*
624	 * Propagate the leaf DTLs we just loaded all the way up the tree.
625	 */
626	spa_config_enter(spa, RW_WRITER, FTAG);
627	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
628	spa_config_exit(spa, FTAG);
629
630	/*
631	 * Check the state of the root vdev.  If it can't be opened, it
632	 * indicates one or more toplevel vdevs are faulted.
633	 */
634	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
635		error = ENXIO;
636		goto out;
637	}
638
639	if ((spa_mode & FWRITE) && state != SPA_LOAD_TRYIMPORT) {
640		dmu_tx_t *tx;
641		int need_update = B_FALSE;
642		int c;
643
644		/*
645		 * Claim log blocks that haven't been committed yet.
646		 * This must all happen in a single txg.
647		 */
648		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
649		    spa_first_txg(spa));
650		(void) dmu_objset_find(spa->spa_name,
651		    zil_claim, tx, DS_FIND_CHILDREN);
652		dmu_tx_commit(tx);
653
654		spa->spa_sync_on = B_TRUE;
655		txg_sync_start(spa->spa_dsl_pool);
656
657		/*
658		 * Wait for all claims to sync.
659		 */
660		txg_wait_synced(spa->spa_dsl_pool, 0);
661
662		/*
663		 * If the config cache is stale, or we have uninitialized
664		 * metaslabs (see spa_vdev_add()), then update the config.
665		 */
666		if (config_cache_txg != spa->spa_config_txg ||
667		    state == SPA_LOAD_IMPORT)
668			need_update = B_TRUE;
669
670		for (c = 0; c < rvd->vdev_children; c++)
671			if (rvd->vdev_child[c]->vdev_ms_array == 0)
672				need_update = B_TRUE;
673
674		/*
675		 * Update the config cache asychronously in case we're the
676		 * root pool, in which case the config cache isn't writable yet.
677		 */
678		if (need_update)
679			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
680	}
681
682	error = 0;
683out:
684	if (error && error != EBADF)
685		zfs_ereport_post(FM_EREPORT_ZFS_POOL, spa, NULL, NULL, 0, 0);
686	spa->spa_load_state = SPA_LOAD_NONE;
687	spa->spa_ena = 0;
688
689	return (error);
690}
691
692/*
693 * Pool Open/Import
694 *
695 * The import case is identical to an open except that the configuration is sent
696 * down from userland, instead of grabbed from the configuration cache.  For the
697 * case of an open, the pool configuration will exist in the
698 * POOL_STATE_UNITIALIZED state.
699 *
700 * The stats information (gen/count/ustats) is used to gather vdev statistics at
701 * the same time open the pool, without having to keep around the spa_t in some
702 * ambiguous state.
703 */
704static int
705spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t **config)
706{
707	spa_t *spa;
708	int error;
709	int loaded = B_FALSE;
710	int locked = B_FALSE;
711
712	*spapp = NULL;
713
714	/*
715	 * As disgusting as this is, we need to support recursive calls to this
716	 * function because dsl_dir_open() is called during spa_load(), and ends
717	 * up calling spa_open() again.  The real fix is to figure out how to
718	 * avoid dsl_dir_open() calling this in the first place.
719	 */
720	if (mutex_owner(&spa_namespace_lock) != curthread) {
721		mutex_enter(&spa_namespace_lock);
722		locked = B_TRUE;
723	}
724
725	if ((spa = spa_lookup(pool)) == NULL) {
726		if (locked)
727			mutex_exit(&spa_namespace_lock);
728		return (ENOENT);
729	}
730	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
731
732		spa_activate(spa);
733
734		error = spa_load(spa, spa->spa_config, SPA_LOAD_OPEN, B_FALSE);
735
736		if (error == EBADF) {
737			/*
738			 * If vdev_validate() returns failure (indicated by
739			 * EBADF), it indicates that one of the vdevs indicates
740			 * that the pool has been exported or destroyed.  If
741			 * this is the case, the config cache is out of sync and
742			 * we should remove the pool from the namespace.
743			 */
744			zfs_post_ok(spa, NULL);
745			spa_unload(spa);
746			spa_deactivate(spa);
747			spa_remove(spa);
748			spa_config_sync();
749			if (locked)
750				mutex_exit(&spa_namespace_lock);
751			return (ENOENT);
752		}
753
754		if (error) {
755			/*
756			 * We can't open the pool, but we still have useful
757			 * information: the state of each vdev after the
758			 * attempted vdev_open().  Return this to the user.
759			 */
760			if (config != NULL && spa->spa_root_vdev != NULL) {
761				spa_config_enter(spa, RW_READER, FTAG);
762				*config = spa_config_generate(spa, NULL, -1ULL,
763				    B_TRUE);
764				spa_config_exit(spa, FTAG);
765			}
766			spa_unload(spa);
767			spa_deactivate(spa);
768			spa->spa_last_open_failed = B_TRUE;
769			if (locked)
770				mutex_exit(&spa_namespace_lock);
771			*spapp = NULL;
772			return (error);
773		} else {
774			zfs_post_ok(spa, NULL);
775			spa->spa_last_open_failed = B_FALSE;
776		}
777
778		loaded = B_TRUE;
779	}
780
781	spa_open_ref(spa, tag);
782	if (locked)
783		mutex_exit(&spa_namespace_lock);
784
785	*spapp = spa;
786
787	if (config != NULL) {
788		spa_config_enter(spa, RW_READER, FTAG);
789		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
790		spa_config_exit(spa, FTAG);
791	}
792
793	/*
794	 * If we just loaded the pool, resilver anything that's out of date.
795	 */
796	if (loaded && (spa_mode & FWRITE))
797		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
798
799	return (0);
800}
801
802int
803spa_open(const char *name, spa_t **spapp, void *tag)
804{
805	return (spa_open_common(name, spapp, tag, NULL));
806}
807
808/*
809 * Lookup the given spa_t, incrementing the inject count in the process,
810 * preventing it from being exported or destroyed.
811 */
812spa_t *
813spa_inject_addref(char *name)
814{
815	spa_t *spa;
816
817	mutex_enter(&spa_namespace_lock);
818	if ((spa = spa_lookup(name)) == NULL) {
819		mutex_exit(&spa_namespace_lock);
820		return (NULL);
821	}
822	spa->spa_inject_ref++;
823	mutex_exit(&spa_namespace_lock);
824
825	return (spa);
826}
827
828void
829spa_inject_delref(spa_t *spa)
830{
831	mutex_enter(&spa_namespace_lock);
832	spa->spa_inject_ref--;
833	mutex_exit(&spa_namespace_lock);
834}
835
836static void
837spa_add_spares(spa_t *spa, nvlist_t *config)
838{
839	nvlist_t **spares;
840	uint_t i, nspares;
841	nvlist_t *nvroot;
842	uint64_t guid;
843	vdev_stat_t *vs;
844	uint_t vsc;
845
846	if (spa->spa_nspares == 0)
847		return;
848
849	VERIFY(nvlist_lookup_nvlist(config,
850	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
851	VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist,
852	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
853	if (nspares != 0) {
854		VERIFY(nvlist_add_nvlist_array(nvroot,
855		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
856		VERIFY(nvlist_lookup_nvlist_array(nvroot,
857		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
858
859		/*
860		 * Go through and find any spares which have since been
861		 * repurposed as an active spare.  If this is the case, update
862		 * their status appropriately.
863		 */
864		for (i = 0; i < nspares; i++) {
865			VERIFY(nvlist_lookup_uint64(spares[i],
866			    ZPOOL_CONFIG_GUID, &guid) == 0);
867			if (spa_spare_inuse(guid)) {
868				VERIFY(nvlist_lookup_uint64_array(
869				    spares[i], ZPOOL_CONFIG_STATS,
870				    (uint64_t **)&vs, &vsc) == 0);
871				vs->vs_state = VDEV_STATE_CANT_OPEN;
872				vs->vs_aux = VDEV_AUX_SPARED;
873			}
874		}
875	}
876}
877
878int
879spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
880{
881	int error;
882	spa_t *spa;
883
884	*config = NULL;
885	error = spa_open_common(name, &spa, FTAG, config);
886
887	if (spa && *config != NULL) {
888		VERIFY(nvlist_add_uint64(*config, ZPOOL_CONFIG_ERRCOUNT,
889		    spa_get_errlog_size(spa)) == 0);
890
891		spa_add_spares(spa, *config);
892	}
893
894	/*
895	 * We want to get the alternate root even for faulted pools, so we cheat
896	 * and call spa_lookup() directly.
897	 */
898	if (altroot) {
899		if (spa == NULL) {
900			mutex_enter(&spa_namespace_lock);
901			spa = spa_lookup(name);
902			if (spa)
903				spa_altroot(spa, altroot, buflen);
904			else
905				altroot[0] = '\0';
906			spa = NULL;
907			mutex_exit(&spa_namespace_lock);
908		} else {
909			spa_altroot(spa, altroot, buflen);
910		}
911	}
912
913	if (spa != NULL)
914		spa_close(spa, FTAG);
915
916	return (error);
917}
918
919/*
920 * Validate that the 'spares' array is well formed.  We must have an array of
921 * nvlists, each which describes a valid leaf vdev.
922 */
923static int
924spa_validate_spares(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
925{
926	nvlist_t **spares;
927	uint_t i, nspares;
928	vdev_t *vd;
929	int error;
930
931	/*
932	 * It's acceptable to have no spares specified.
933	 */
934	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
935	    &spares, &nspares) != 0)
936		return (0);
937
938	if (nspares == 0)
939		return (EINVAL);
940
941	/*
942	 * Make sure the pool is formatted with a version that supports hot
943	 * spares.
944	 */
945	if (spa_version(spa) < ZFS_VERSION_SPARES)
946		return (ENOTSUP);
947
948	for (i = 0; i < nspares; i++) {
949		if ((error = spa_config_parse(spa, &vd, spares[i], NULL, 0,
950		    mode)) != 0)
951			return (error);
952
953		if (!vd->vdev_ops->vdev_op_leaf) {
954			vdev_free(vd);
955			return (EINVAL);
956		}
957
958		if ((error = vdev_open(vd)) != 0) {
959			vdev_free(vd);
960			return (error);
961		}
962
963		vd->vdev_top = vd;
964		if ((error = vdev_label_spare(vd, crtxg)) != 0) {
965			vdev_free(vd);
966			return (error);
967		}
968
969		VERIFY(nvlist_add_uint64(spares[i], ZPOOL_CONFIG_GUID,
970		    vd->vdev_guid) == 0);
971
972		vdev_free(vd);
973	}
974
975	return (0);
976}
977
978/*
979 * Pool Creation
980 */
981int
982spa_create(const char *pool, nvlist_t *nvroot, const char *altroot)
983{
984	spa_t *spa;
985	vdev_t *rvd;
986	dsl_pool_t *dp;
987	dmu_tx_t *tx;
988	int c, error = 0;
989	uint64_t txg = TXG_INITIAL;
990	nvlist_t **spares;
991	uint_t nspares;
992
993	/*
994	 * If this pool already exists, return failure.
995	 */
996	mutex_enter(&spa_namespace_lock);
997	if (spa_lookup(pool) != NULL) {
998		mutex_exit(&spa_namespace_lock);
999		return (EEXIST);
1000	}
1001
1002	/*
1003	 * Allocate a new spa_t structure.
1004	 */
1005	spa = spa_add(pool, altroot);
1006	spa_activate(spa);
1007
1008	spa->spa_uberblock.ub_txg = txg - 1;
1009	spa->spa_uberblock.ub_version = ZFS_VERSION;
1010	spa->spa_ubsync = spa->spa_uberblock;
1011
1012	/*
1013	 * Create the root vdev.
1014	 */
1015	spa_config_enter(spa, RW_WRITER, FTAG);
1016
1017	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
1018
1019	ASSERT(error != 0 || rvd != NULL);
1020	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
1021
1022	if (error == 0 && rvd->vdev_children == 0)
1023		error = EINVAL;
1024
1025	if (error == 0 &&
1026	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
1027	    (error = spa_validate_spares(spa, nvroot, txg,
1028	    VDEV_ALLOC_ADD)) == 0) {
1029		for (c = 0; c < rvd->vdev_children; c++)
1030			vdev_init(rvd->vdev_child[c], txg);
1031		vdev_config_dirty(rvd);
1032	}
1033
1034	spa_config_exit(spa, FTAG);
1035
1036	if (error != 0) {
1037		spa_unload(spa);
1038		spa_deactivate(spa);
1039		spa_remove(spa);
1040		mutex_exit(&spa_namespace_lock);
1041		return (error);
1042	}
1043
1044	/*
1045	 * Get the list of spares, if specified.
1046	 */
1047	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1048	    &spares, &nspares) == 0) {
1049		VERIFY(nvlist_alloc(&spa->spa_sparelist, NV_UNIQUE_NAME,
1050		    KM_SLEEP) == 0);
1051		VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
1052		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1053		spa_config_enter(spa, RW_WRITER, FTAG);
1054		spa_load_spares(spa);
1055		spa_config_exit(spa, FTAG);
1056		spa->spa_sync_spares = B_TRUE;
1057	}
1058
1059	spa->spa_dsl_pool = dp = dsl_pool_create(spa, txg);
1060	spa->spa_meta_objset = dp->dp_meta_objset;
1061
1062	tx = dmu_tx_create_assigned(dp, txg);
1063
1064	/*
1065	 * Create the pool config object.
1066	 */
1067	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
1068	    DMU_OT_PACKED_NVLIST, 1 << 14,
1069	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
1070
1071	if (zap_add(spa->spa_meta_objset,
1072	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
1073	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
1074		cmn_err(CE_PANIC, "failed to add pool config");
1075	}
1076
1077	/* Newly created pools are always deflated. */
1078	spa->spa_deflate = TRUE;
1079	if (zap_add(spa->spa_meta_objset,
1080	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
1081	    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
1082		cmn_err(CE_PANIC, "failed to add deflate");
1083	}
1084
1085	/*
1086	 * Create the deferred-free bplist object.  Turn off compression
1087	 * because sync-to-convergence takes longer if the blocksize
1088	 * keeps changing.
1089	 */
1090	spa->spa_sync_bplist_obj = bplist_create(spa->spa_meta_objset,
1091	    1 << 14, tx);
1092	dmu_object_set_compress(spa->spa_meta_objset, spa->spa_sync_bplist_obj,
1093	    ZIO_COMPRESS_OFF, tx);
1094
1095	if (zap_add(spa->spa_meta_objset,
1096	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
1097	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj, tx) != 0) {
1098		cmn_err(CE_PANIC, "failed to add bplist");
1099	}
1100
1101	dmu_tx_commit(tx);
1102
1103	spa->spa_sync_on = B_TRUE;
1104	txg_sync_start(spa->spa_dsl_pool);
1105
1106	/*
1107	 * We explicitly wait for the first transaction to complete so that our
1108	 * bean counters are appropriately updated.
1109	 */
1110	txg_wait_synced(spa->spa_dsl_pool, txg);
1111
1112	spa_config_sync();
1113
1114	mutex_exit(&spa_namespace_lock);
1115
1116	return (0);
1117}
1118
1119/*
1120 * Import the given pool into the system.  We set up the necessary spa_t and
1121 * then call spa_load() to do the dirty work.
1122 */
1123int
1124spa_import(const char *pool, nvlist_t *config, const char *altroot)
1125{
1126	spa_t *spa;
1127	int error;
1128	nvlist_t *nvroot;
1129	nvlist_t **spares;
1130	uint_t nspares;
1131
1132	if (!(spa_mode & FWRITE))
1133		return (EROFS);
1134
1135	/*
1136	 * If a pool with this name exists, return failure.
1137	 */
1138	mutex_enter(&spa_namespace_lock);
1139	if (spa_lookup(pool) != NULL) {
1140		mutex_exit(&spa_namespace_lock);
1141		return (EEXIST);
1142	}
1143
1144	/*
1145	 * Create and initialize the spa structure.
1146	 */
1147	spa = spa_add(pool, altroot);
1148	spa_activate(spa);
1149
1150	/*
1151	 * Pass off the heavy lifting to spa_load().
1152	 * Pass TRUE for mosconfig because the user-supplied config
1153	 * is actually the one to trust when doing an import.
1154	 */
1155	error = spa_load(spa, config, SPA_LOAD_IMPORT, B_TRUE);
1156
1157	spa_config_enter(spa, RW_WRITER, FTAG);
1158	/*
1159	 * Toss any existing sparelist, as it doesn't have any validity anymore,
1160	 * and conflicts with spa_has_spare().
1161	 */
1162	if (spa->spa_sparelist) {
1163		nvlist_free(spa->spa_sparelist);
1164		spa->spa_sparelist = NULL;
1165		spa_load_spares(spa);
1166	}
1167
1168	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1169	    &nvroot) == 0);
1170	if (error == 0)
1171		error = spa_validate_spares(spa, nvroot, -1ULL,
1172		    VDEV_ALLOC_SPARE);
1173	spa_config_exit(spa, FTAG);
1174
1175	if (error != 0) {
1176		spa_unload(spa);
1177		spa_deactivate(spa);
1178		spa_remove(spa);
1179		mutex_exit(&spa_namespace_lock);
1180		return (error);
1181	}
1182
1183	/*
1184	 * Override any spares as specified by the user, as these may have
1185	 * correct device names/devids, etc.
1186	 */
1187	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1188	    &spares, &nspares) == 0) {
1189		if (spa->spa_sparelist)
1190			VERIFY(nvlist_remove(spa->spa_sparelist,
1191			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
1192		else
1193			VERIFY(nvlist_alloc(&spa->spa_sparelist,
1194			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
1195		VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
1196		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1197		spa_config_enter(spa, RW_WRITER, FTAG);
1198		spa_load_spares(spa);
1199		spa_config_exit(spa, FTAG);
1200		spa->spa_sync_spares = B_TRUE;
1201	}
1202
1203	/*
1204	 * Update the config cache to include the newly-imported pool.
1205	 */
1206	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
1207
1208	mutex_exit(&spa_namespace_lock);
1209
1210	/*
1211	 * Resilver anything that's out of date.
1212	 */
1213	if (spa_mode & FWRITE)
1214		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
1215
1216	return (0);
1217}
1218
1219/*
1220 * This (illegal) pool name is used when temporarily importing a spa_t in order
1221 * to get the vdev stats associated with the imported devices.
1222 */
1223#define	TRYIMPORT_NAME	"$import"
1224
1225nvlist_t *
1226spa_tryimport(nvlist_t *tryconfig)
1227{
1228	nvlist_t *config = NULL;
1229	char *poolname;
1230	spa_t *spa;
1231	uint64_t state;
1232
1233	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
1234		return (NULL);
1235
1236	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
1237		return (NULL);
1238
1239	/*
1240	 * Create and initialize the spa structure.
1241	 */
1242	mutex_enter(&spa_namespace_lock);
1243	spa = spa_add(TRYIMPORT_NAME, NULL);
1244	spa_activate(spa);
1245
1246	/*
1247	 * Pass off the heavy lifting to spa_load().
1248	 * Pass TRUE for mosconfig because the user-supplied config
1249	 * is actually the one to trust when doing an import.
1250	 */
1251	(void) spa_load(spa, tryconfig, SPA_LOAD_TRYIMPORT, B_TRUE);
1252
1253	/*
1254	 * If 'tryconfig' was at least parsable, return the current config.
1255	 */
1256	if (spa->spa_root_vdev != NULL) {
1257		spa_config_enter(spa, RW_READER, FTAG);
1258		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
1259		spa_config_exit(spa, FTAG);
1260		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
1261		    poolname) == 0);
1262		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1263		    state) == 0);
1264
1265		/*
1266		 * Add the list of hot spares.
1267		 */
1268		spa_add_spares(spa, config);
1269	}
1270
1271	spa_unload(spa);
1272	spa_deactivate(spa);
1273	spa_remove(spa);
1274	mutex_exit(&spa_namespace_lock);
1275
1276	return (config);
1277}
1278
1279/*
1280 * Pool export/destroy
1281 *
1282 * The act of destroying or exporting a pool is very simple.  We make sure there
1283 * is no more pending I/O and any references to the pool are gone.  Then, we
1284 * update the pool state and sync all the labels to disk, removing the
1285 * configuration from the cache afterwards.
1286 */
1287static int
1288spa_export_common(char *pool, int new_state, nvlist_t **oldconfig)
1289{
1290	spa_t *spa;
1291
1292	if (oldconfig)
1293		*oldconfig = NULL;
1294
1295	if (!(spa_mode & FWRITE))
1296		return (EROFS);
1297
1298	mutex_enter(&spa_namespace_lock);
1299	if ((spa = spa_lookup(pool)) == NULL) {
1300		mutex_exit(&spa_namespace_lock);
1301		return (ENOENT);
1302	}
1303
1304	/*
1305	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
1306	 * reacquire the namespace lock, and see if we can export.
1307	 */
1308	spa_open_ref(spa, FTAG);
1309	mutex_exit(&spa_namespace_lock);
1310	spa_async_suspend(spa);
1311	mutex_enter(&spa_namespace_lock);
1312	spa_close(spa, FTAG);
1313
1314	/*
1315	 * The pool will be in core if it's openable,
1316	 * in which case we can modify its state.
1317	 */
1318	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
1319		/*
1320		 * Objsets may be open only because they're dirty, so we
1321		 * have to force it to sync before checking spa_refcnt.
1322		 */
1323		spa_scrub_suspend(spa);
1324		txg_wait_synced(spa->spa_dsl_pool, 0);
1325
1326		/*
1327		 * A pool cannot be exported or destroyed if there are active
1328		 * references.  If we are resetting a pool, allow references by
1329		 * fault injection handlers.
1330		 */
1331		if (!spa_refcount_zero(spa) ||
1332		    (spa->spa_inject_ref != 0 &&
1333		    new_state != POOL_STATE_UNINITIALIZED)) {
1334			spa_scrub_resume(spa);
1335			spa_async_resume(spa);
1336			mutex_exit(&spa_namespace_lock);
1337			return (EBUSY);
1338		}
1339
1340		spa_scrub_resume(spa);
1341		VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0);
1342
1343		/*
1344		 * We want this to be reflected on every label,
1345		 * so mark them all dirty.  spa_unload() will do the
1346		 * final sync that pushes these changes out.
1347		 */
1348		if (new_state != POOL_STATE_UNINITIALIZED) {
1349			spa_config_enter(spa, RW_WRITER, FTAG);
1350			spa->spa_state = new_state;
1351			spa->spa_final_txg = spa_last_synced_txg(spa) + 1;
1352			vdev_config_dirty(spa->spa_root_vdev);
1353			spa_config_exit(spa, FTAG);
1354		}
1355	}
1356
1357	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
1358		spa_unload(spa);
1359		spa_deactivate(spa);
1360	}
1361
1362	if (oldconfig && spa->spa_config)
1363		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
1364
1365	if (new_state != POOL_STATE_UNINITIALIZED) {
1366		spa_remove(spa);
1367		spa_config_sync();
1368	}
1369	mutex_exit(&spa_namespace_lock);
1370
1371	return (0);
1372}
1373
1374/*
1375 * Destroy a storage pool.
1376 */
1377int
1378spa_destroy(char *pool)
1379{
1380	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL));
1381}
1382
1383/*
1384 * Export a storage pool.
1385 */
1386int
1387spa_export(char *pool, nvlist_t **oldconfig)
1388{
1389	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig));
1390}
1391
1392/*
1393 * Similar to spa_export(), this unloads the spa_t without actually removing it
1394 * from the namespace in any way.
1395 */
1396int
1397spa_reset(char *pool)
1398{
1399	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL));
1400}
1401
1402
1403/*
1404 * ==========================================================================
1405 * Device manipulation
1406 * ==========================================================================
1407 */
1408
1409/*
1410 * Add capacity to a storage pool.
1411 */
1412int
1413spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
1414{
1415	uint64_t txg;
1416	int c, error;
1417	vdev_t *rvd = spa->spa_root_vdev;
1418	vdev_t *vd, *tvd;
1419	nvlist_t **spares;
1420	uint_t i, nspares;
1421
1422	txg = spa_vdev_enter(spa);
1423
1424	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
1425	    VDEV_ALLOC_ADD)) != 0)
1426		return (spa_vdev_exit(spa, NULL, txg, error));
1427
1428	if ((error = spa_validate_spares(spa, nvroot, txg,
1429	    VDEV_ALLOC_ADD)) != 0)
1430		return (spa_vdev_exit(spa, vd, txg, error));
1431
1432	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1433	    &spares, &nspares) != 0)
1434		nspares = 0;
1435
1436	if (vd->vdev_children == 0 && nspares == 0)
1437		return (spa_vdev_exit(spa, vd, txg, EINVAL));
1438
1439	if (vd->vdev_children != 0) {
1440		if ((error = vdev_create(vd, txg, B_FALSE)) != 0)
1441			return (spa_vdev_exit(spa, vd, txg, error));
1442
1443		/*
1444		 * Transfer each new top-level vdev from vd to rvd.
1445		 */
1446		for (c = 0; c < vd->vdev_children; c++) {
1447			tvd = vd->vdev_child[c];
1448			vdev_remove_child(vd, tvd);
1449			tvd->vdev_id = rvd->vdev_children;
1450			vdev_add_child(rvd, tvd);
1451			vdev_config_dirty(tvd);
1452		}
1453	}
1454
1455	if (nspares != 0) {
1456		if (spa->spa_sparelist != NULL) {
1457			nvlist_t **oldspares;
1458			uint_t oldnspares;
1459			nvlist_t **newspares;
1460
1461			VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist,
1462			    ZPOOL_CONFIG_SPARES, &oldspares, &oldnspares) == 0);
1463
1464			newspares = kmem_alloc(sizeof (void *) *
1465			    (nspares + oldnspares), KM_SLEEP);
1466			for (i = 0; i < oldnspares; i++)
1467				VERIFY(nvlist_dup(oldspares[i],
1468				    &newspares[i], KM_SLEEP) == 0);
1469			for (i = 0; i < nspares; i++)
1470				VERIFY(nvlist_dup(spares[i],
1471				    &newspares[i + oldnspares],
1472				    KM_SLEEP) == 0);
1473
1474			VERIFY(nvlist_remove(spa->spa_sparelist,
1475			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
1476
1477			VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
1478			    ZPOOL_CONFIG_SPARES, newspares,
1479			    nspares + oldnspares) == 0);
1480			for (i = 0; i < oldnspares + nspares; i++)
1481				nvlist_free(newspares[i]);
1482			kmem_free(newspares, (oldnspares + nspares) *
1483			    sizeof (void *));
1484		} else {
1485			VERIFY(nvlist_alloc(&spa->spa_sparelist,
1486			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
1487			VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
1488			    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1489		}
1490
1491		spa_load_spares(spa);
1492		spa->spa_sync_spares = B_TRUE;
1493	}
1494
1495	/*
1496	 * We have to be careful when adding new vdevs to an existing pool.
1497	 * If other threads start allocating from these vdevs before we
1498	 * sync the config cache, and we lose power, then upon reboot we may
1499	 * fail to open the pool because there are DVAs that the config cache
1500	 * can't translate.  Therefore, we first add the vdevs without
1501	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
1502	 * and then let spa_config_update() initialize the new metaslabs.
1503	 *
1504	 * spa_load() checks for added-but-not-initialized vdevs, so that
1505	 * if we lose power at any point in this sequence, the remaining
1506	 * steps will be completed the next time we load the pool.
1507	 */
1508	(void) spa_vdev_exit(spa, vd, txg, 0);
1509
1510	mutex_enter(&spa_namespace_lock);
1511	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
1512	mutex_exit(&spa_namespace_lock);
1513
1514	return (0);
1515}
1516
1517/*
1518 * Attach a device to a mirror.  The arguments are the path to any device
1519 * in the mirror, and the nvroot for the new device.  If the path specifies
1520 * a device that is not mirrored, we automatically insert the mirror vdev.
1521 *
1522 * If 'replacing' is specified, the new device is intended to replace the
1523 * existing device; in this case the two devices are made into their own
1524 * mirror using the 'replacing' vdev, which is functionally idendical to
1525 * the mirror vdev (it actually reuses all the same ops) but has a few
1526 * extra rules: you can't attach to it after it's been created, and upon
1527 * completion of resilvering, the first disk (the one being replaced)
1528 * is automatically detached.
1529 */
1530int
1531spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
1532{
1533	uint64_t txg, open_txg;
1534	int error;
1535	vdev_t *rvd = spa->spa_root_vdev;
1536	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
1537	vdev_ops_t *pvops;
1538
1539	txg = spa_vdev_enter(spa);
1540
1541	oldvd = vdev_lookup_by_guid(rvd, guid);
1542
1543	if (oldvd == NULL)
1544		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
1545
1546	if (!oldvd->vdev_ops->vdev_op_leaf)
1547		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1548
1549	pvd = oldvd->vdev_parent;
1550
1551	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
1552	    VDEV_ALLOC_ADD)) != 0 || newrootvd->vdev_children != 1)
1553		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
1554
1555	newvd = newrootvd->vdev_child[0];
1556
1557	if (!newvd->vdev_ops->vdev_op_leaf)
1558		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
1559
1560	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
1561		return (spa_vdev_exit(spa, newrootvd, txg, error));
1562
1563	if (!replacing) {
1564		/*
1565		 * For attach, the only allowable parent is a mirror or the root
1566		 * vdev.
1567		 */
1568		if (pvd->vdev_ops != &vdev_mirror_ops &&
1569		    pvd->vdev_ops != &vdev_root_ops)
1570			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
1571
1572		pvops = &vdev_mirror_ops;
1573	} else {
1574		/*
1575		 * Active hot spares can only be replaced by inactive hot
1576		 * spares.
1577		 */
1578		if (pvd->vdev_ops == &vdev_spare_ops &&
1579		    pvd->vdev_child[1] == oldvd &&
1580		    !spa_has_spare(spa, newvd->vdev_guid))
1581			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
1582
1583		/*
1584		 * If the source is a hot spare, and the parent isn't already a
1585		 * spare, then we want to create a new hot spare.  Otherwise, we
1586		 * want to create a replacing vdev.
1587		 */
1588		if (pvd->vdev_ops == &vdev_replacing_ops)
1589			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
1590		else if (pvd->vdev_ops != &vdev_spare_ops &&
1591		    newvd->vdev_isspare)
1592			pvops = &vdev_spare_ops;
1593		else
1594			pvops = &vdev_replacing_ops;
1595	}
1596
1597	/*
1598	 * Compare the new device size with the replaceable/attachable
1599	 * device size.
1600	 */
1601	if (newvd->vdev_psize < vdev_get_rsize(oldvd))
1602		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
1603
1604	/*
1605	 * The new device cannot have a higher alignment requirement
1606	 * than the top-level vdev.
1607	 */
1608	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
1609		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
1610
1611	/*
1612	 * If this is an in-place replacement, update oldvd's path and devid
1613	 * to make it distinguishable from newvd, and unopenable from now on.
1614	 */
1615	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
1616		spa_strfree(oldvd->vdev_path);
1617		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
1618		    KM_SLEEP);
1619		(void) sprintf(oldvd->vdev_path, "%s/%s",
1620		    newvd->vdev_path, "old");
1621		if (oldvd->vdev_devid != NULL) {
1622			spa_strfree(oldvd->vdev_devid);
1623			oldvd->vdev_devid = NULL;
1624		}
1625	}
1626
1627	/*
1628	 * If the parent is not a mirror, or if we're replacing, insert the new
1629	 * mirror/replacing/spare vdev above oldvd.
1630	 */
1631	if (pvd->vdev_ops != pvops)
1632		pvd = vdev_add_parent(oldvd, pvops);
1633
1634	ASSERT(pvd->vdev_top->vdev_parent == rvd);
1635	ASSERT(pvd->vdev_ops == pvops);
1636	ASSERT(oldvd->vdev_parent == pvd);
1637
1638	/*
1639	 * Extract the new device from its root and add it to pvd.
1640	 */
1641	vdev_remove_child(newrootvd, newvd);
1642	newvd->vdev_id = pvd->vdev_children;
1643	vdev_add_child(pvd, newvd);
1644
1645	/*
1646	 * If newvd is smaller than oldvd, but larger than its rsize,
1647	 * the addition of newvd may have decreased our parent's asize.
1648	 */
1649	pvd->vdev_asize = MIN(pvd->vdev_asize, newvd->vdev_asize);
1650
1651	tvd = newvd->vdev_top;
1652	ASSERT(pvd->vdev_top == tvd);
1653	ASSERT(tvd->vdev_parent == rvd);
1654
1655	vdev_config_dirty(tvd);
1656
1657	/*
1658	 * Set newvd's DTL to [TXG_INITIAL, open_txg].  It will propagate
1659	 * upward when spa_vdev_exit() calls vdev_dtl_reassess().
1660	 */
1661	open_txg = txg + TXG_CONCURRENT_STATES - 1;
1662
1663	mutex_enter(&newvd->vdev_dtl_lock);
1664	space_map_add(&newvd->vdev_dtl_map, TXG_INITIAL,
1665	    open_txg - TXG_INITIAL + 1);
1666	mutex_exit(&newvd->vdev_dtl_lock);
1667
1668	dprintf("attached %s in txg %llu\n", newvd->vdev_path, txg);
1669
1670	/*
1671	 * Mark newvd's DTL dirty in this txg.
1672	 */
1673	vdev_dirty(tvd, VDD_DTL, newvd, txg);
1674
1675	(void) spa_vdev_exit(spa, newrootvd, open_txg, 0);
1676
1677	/*
1678	 * Kick off a resilver to update newvd.
1679	 */
1680	VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
1681
1682	return (0);
1683}
1684
1685/*
1686 * Detach a device from a mirror or replacing vdev.
1687 * If 'replace_done' is specified, only detach if the parent
1688 * is a replacing vdev.
1689 */
1690int
1691spa_vdev_detach(spa_t *spa, uint64_t guid, int replace_done)
1692{
1693	uint64_t txg;
1694	int c, t, error;
1695	vdev_t *rvd = spa->spa_root_vdev;
1696	vdev_t *vd, *pvd, *cvd, *tvd;
1697	boolean_t unspare = B_FALSE;
1698	uint64_t unspare_guid;
1699
1700	txg = spa_vdev_enter(spa);
1701
1702	vd = vdev_lookup_by_guid(rvd, guid);
1703
1704	if (vd == NULL)
1705		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
1706
1707	if (!vd->vdev_ops->vdev_op_leaf)
1708		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1709
1710	pvd = vd->vdev_parent;
1711
1712	/*
1713	 * If replace_done is specified, only remove this device if it's
1714	 * the first child of a replacing vdev.  For the 'spare' vdev, either
1715	 * disk can be removed.
1716	 */
1717	if (replace_done) {
1718		if (pvd->vdev_ops == &vdev_replacing_ops) {
1719			if (vd->vdev_id != 0)
1720				return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1721		} else if (pvd->vdev_ops != &vdev_spare_ops) {
1722			return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1723		}
1724	}
1725
1726	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
1727	    spa_version(spa) >= ZFS_VERSION_SPARES);
1728
1729	/*
1730	 * Only mirror, replacing, and spare vdevs support detach.
1731	 */
1732	if (pvd->vdev_ops != &vdev_replacing_ops &&
1733	    pvd->vdev_ops != &vdev_mirror_ops &&
1734	    pvd->vdev_ops != &vdev_spare_ops)
1735		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1736
1737	/*
1738	 * If there's only one replica, you can't detach it.
1739	 */
1740	if (pvd->vdev_children <= 1)
1741		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
1742
1743	/*
1744	 * If all siblings have non-empty DTLs, this device may have the only
1745	 * valid copy of the data, which means we cannot safely detach it.
1746	 *
1747	 * XXX -- as in the vdev_offline() case, we really want a more
1748	 * precise DTL check.
1749	 */
1750	for (c = 0; c < pvd->vdev_children; c++) {
1751		uint64_t dirty;
1752
1753		cvd = pvd->vdev_child[c];
1754		if (cvd == vd)
1755			continue;
1756		if (vdev_is_dead(cvd))
1757			continue;
1758		mutex_enter(&cvd->vdev_dtl_lock);
1759		dirty = cvd->vdev_dtl_map.sm_space |
1760		    cvd->vdev_dtl_scrub.sm_space;
1761		mutex_exit(&cvd->vdev_dtl_lock);
1762		if (!dirty)
1763			break;
1764	}
1765
1766	/*
1767	 * If we are a replacing or spare vdev, then we can always detach the
1768	 * latter child, as that is how one cancels the operation.
1769	 */
1770	if ((pvd->vdev_ops == &vdev_mirror_ops || vd->vdev_id != 1) &&
1771	    c == pvd->vdev_children)
1772		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
1773
1774	/*
1775	 * If we are detaching the original disk from a spare, then it implies
1776	 * that the spare should become a real disk, and be removed from the
1777	 * active spare list for the pool.
1778	 */
1779	if (pvd->vdev_ops == &vdev_spare_ops &&
1780	    vd->vdev_id == 0)
1781		unspare = B_TRUE;
1782
1783	/*
1784	 * Erase the disk labels so the disk can be used for other things.
1785	 * This must be done after all other error cases are handled,
1786	 * but before we disembowel vd (so we can still do I/O to it).
1787	 * But if we can't do it, don't treat the error as fatal --
1788	 * it may be that the unwritability of the disk is the reason
1789	 * it's being detached!
1790	 */
1791	error = vdev_label_init(vd, 0, B_FALSE);
1792	if (error)
1793		dprintf("unable to erase labels on %s\n", vdev_description(vd));
1794
1795	/*
1796	 * Remove vd from its parent and compact the parent's children.
1797	 */
1798	vdev_remove_child(pvd, vd);
1799	vdev_compact_children(pvd);
1800
1801	/*
1802	 * Remember one of the remaining children so we can get tvd below.
1803	 */
1804	cvd = pvd->vdev_child[0];
1805
1806	/*
1807	 * If we need to remove the remaining child from the list of hot spares,
1808	 * do it now, marking the vdev as no longer a spare in the process.  We
1809	 * must do this before vdev_remove_parent(), because that can change the
1810	 * GUID if it creates a new toplevel GUID.
1811	 */
1812	if (unspare) {
1813		ASSERT(cvd->vdev_isspare);
1814		spa_spare_remove(cvd->vdev_guid);
1815		cvd->vdev_isspare = B_FALSE;
1816		unspare_guid = cvd->vdev_guid;
1817	}
1818
1819	/*
1820	 * If the parent mirror/replacing vdev only has one child,
1821	 * the parent is no longer needed.  Remove it from the tree.
1822	 */
1823	if (pvd->vdev_children == 1)
1824		vdev_remove_parent(cvd);
1825
1826	/*
1827	 * We don't set tvd until now because the parent we just removed
1828	 * may have been the previous top-level vdev.
1829	 */
1830	tvd = cvd->vdev_top;
1831	ASSERT(tvd->vdev_parent == rvd);
1832
1833	/*
1834	 * Reopen this top-level vdev to reassess health after detach.
1835	 */
1836	vdev_reopen(tvd);
1837
1838	/*
1839	 * If the device we just detached was smaller than the others,
1840	 * it may be possible to add metaslabs (i.e. grow the pool).
1841	 * vdev_metaslab_init() can't fail because the existing metaslabs
1842	 * are already in core, so there's nothing to read from disk.
1843	 */
1844	VERIFY(vdev_metaslab_init(tvd, txg) == 0);
1845
1846	vdev_config_dirty(tvd);
1847
1848	/*
1849	 * Mark vd's DTL as dirty in this txg.
1850	 * vdev_dtl_sync() will see that vd->vdev_detached is set
1851	 * and free vd's DTL object in syncing context.
1852	 * But first make sure we're not on any *other* txg's DTL list,
1853	 * to prevent vd from being accessed after it's freed.
1854	 */
1855	for (t = 0; t < TXG_SIZE; t++)
1856		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
1857	vd->vdev_detached = B_TRUE;
1858	vdev_dirty(tvd, VDD_DTL, vd, txg);
1859
1860	dprintf("detached %s in txg %llu\n", vd->vdev_path, txg);
1861
1862	error = spa_vdev_exit(spa, vd, txg, 0);
1863
1864	/*
1865	 * If we are supposed to remove the given vdev from the list of spares,
1866	 * iterate over all pools in the system and replace it if it's present.
1867	 */
1868	if (unspare) {
1869		spa = NULL;
1870		mutex_enter(&spa_namespace_lock);
1871		while ((spa = spa_next(spa)) != NULL) {
1872			if (spa->spa_state != POOL_STATE_ACTIVE)
1873				continue;
1874
1875			(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
1876		}
1877		mutex_exit(&spa_namespace_lock);
1878	}
1879
1880	return (error);
1881}
1882
1883/*
1884 * Remove a device from the pool.  Currently, this supports removing only hot
1885 * spares.
1886 */
1887int
1888spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
1889{
1890	vdev_t *vd;
1891	nvlist_t **spares, *nv, **newspares;
1892	uint_t i, j, nspares;
1893	int ret = 0;
1894
1895	spa_config_enter(spa, RW_WRITER, FTAG);
1896
1897	vd = spa_lookup_by_guid(spa, guid);
1898
1899	nv = NULL;
1900	if (spa->spa_spares != NULL &&
1901	    nvlist_lookup_nvlist_array(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
1902	    &spares, &nspares) == 0) {
1903		for (i = 0; i < nspares; i++) {
1904			uint64_t theguid;
1905
1906			VERIFY(nvlist_lookup_uint64(spares[i],
1907			    ZPOOL_CONFIG_GUID, &theguid) == 0);
1908			if (theguid == guid) {
1909				nv = spares[i];
1910				break;
1911			}
1912		}
1913	}
1914
1915	/*
1916	 * We only support removing a hot spare, and only if it's not currently
1917	 * in use in this pool.
1918	 */
1919	if (nv == NULL && vd == NULL) {
1920		ret = ENOENT;
1921		goto out;
1922	}
1923
1924	if (nv == NULL && vd != NULL) {
1925		ret = ENOTSUP;
1926		goto out;
1927	}
1928
1929	if (!unspare && nv != NULL && vd != NULL) {
1930		ret = EBUSY;
1931		goto out;
1932	}
1933
1934	if (nspares == 1) {
1935		newspares = NULL;
1936	} else {
1937		newspares = kmem_alloc((nspares - 1) * sizeof (void *),
1938		    KM_SLEEP);
1939		for (i = 0, j = 0; i < nspares; i++) {
1940			if (spares[i] != nv)
1941				VERIFY(nvlist_dup(spares[i],
1942				    &newspares[j++], KM_SLEEP) == 0);
1943		}
1944	}
1945
1946	VERIFY(nvlist_remove(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
1947	    DATA_TYPE_NVLIST_ARRAY) == 0);
1948	VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
1949	    newspares, nspares - 1) == 0);
1950	for (i = 0; i < nspares - 1; i++)
1951		nvlist_free(newspares[i]);
1952	kmem_free(newspares, (nspares - 1) * sizeof (void *));
1953	spa_load_spares(spa);
1954	spa->spa_sync_spares = B_TRUE;
1955
1956out:
1957	spa_config_exit(spa, FTAG);
1958
1959	return (ret);
1960}
1961
1962/*
1963 * Find any device that's done replacing, so we can detach it.
1964 */
1965static vdev_t *
1966spa_vdev_replace_done_hunt(vdev_t *vd)
1967{
1968	vdev_t *newvd, *oldvd;
1969	int c;
1970
1971	for (c = 0; c < vd->vdev_children; c++) {
1972		oldvd = spa_vdev_replace_done_hunt(vd->vdev_child[c]);
1973		if (oldvd != NULL)
1974			return (oldvd);
1975	}
1976
1977	if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
1978		oldvd = vd->vdev_child[0];
1979		newvd = vd->vdev_child[1];
1980
1981		mutex_enter(&newvd->vdev_dtl_lock);
1982		if (newvd->vdev_dtl_map.sm_space == 0 &&
1983		    newvd->vdev_dtl_scrub.sm_space == 0) {
1984			mutex_exit(&newvd->vdev_dtl_lock);
1985			return (oldvd);
1986		}
1987		mutex_exit(&newvd->vdev_dtl_lock);
1988	}
1989
1990	return (NULL);
1991}
1992
1993static void
1994spa_vdev_replace_done(spa_t *spa)
1995{
1996	vdev_t *vd;
1997	vdev_t *pvd;
1998	uint64_t guid;
1999	uint64_t pguid = 0;
2000
2001	spa_config_enter(spa, RW_READER, FTAG);
2002
2003	while ((vd = spa_vdev_replace_done_hunt(spa->spa_root_vdev)) != NULL) {
2004		guid = vd->vdev_guid;
2005		/*
2006		 * If we have just finished replacing a hot spared device, then
2007		 * we need to detach the parent's first child (the original hot
2008		 * spare) as well.
2009		 */
2010		pvd = vd->vdev_parent;
2011		if (pvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2012		    pvd->vdev_id == 0) {
2013			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
2014			ASSERT(pvd->vdev_parent->vdev_children == 2);
2015			pguid = pvd->vdev_parent->vdev_child[1]->vdev_guid;
2016		}
2017		spa_config_exit(spa, FTAG);
2018		if (spa_vdev_detach(spa, guid, B_TRUE) != 0)
2019			return;
2020		if (pguid != 0 && spa_vdev_detach(spa, pguid, B_TRUE) != 0)
2021			return;
2022		spa_config_enter(spa, RW_READER, FTAG);
2023	}
2024
2025	spa_config_exit(spa, FTAG);
2026}
2027
2028/*
2029 * Update the stored path for this vdev.  Dirty the vdev configuration, relying
2030 * on spa_vdev_enter/exit() to synchronize the labels and cache.
2031 */
2032int
2033spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
2034{
2035	vdev_t *rvd, *vd;
2036	uint64_t txg;
2037
2038	rvd = spa->spa_root_vdev;
2039
2040	txg = spa_vdev_enter(spa);
2041
2042	if ((vd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
2043		/*
2044		 * Determine if this is a reference to a hot spare.  In that
2045		 * case, update the path as stored in the spare list.
2046		 */
2047		nvlist_t **spares;
2048		uint_t i, nspares;
2049		if (spa->spa_sparelist != NULL) {
2050			VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist,
2051			    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2052			for (i = 0; i < nspares; i++) {
2053				uint64_t theguid;
2054				VERIFY(nvlist_lookup_uint64(spares[i],
2055				    ZPOOL_CONFIG_GUID, &theguid) == 0);
2056				if (theguid == guid)
2057					break;
2058			}
2059
2060			if (i == nspares)
2061				return (spa_vdev_exit(spa, NULL, txg, ENOENT));
2062
2063			VERIFY(nvlist_add_string(spares[i],
2064			    ZPOOL_CONFIG_PATH, newpath) == 0);
2065			spa_load_spares(spa);
2066			spa->spa_sync_spares = B_TRUE;
2067			return (spa_vdev_exit(spa, NULL, txg, 0));
2068		} else {
2069			return (spa_vdev_exit(spa, NULL, txg, ENOENT));
2070		}
2071	}
2072
2073	if (!vd->vdev_ops->vdev_op_leaf)
2074		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
2075
2076	spa_strfree(vd->vdev_path);
2077	vd->vdev_path = spa_strdup(newpath);
2078
2079	vdev_config_dirty(vd->vdev_top);
2080
2081	return (spa_vdev_exit(spa, NULL, txg, 0));
2082}
2083
2084/*
2085 * ==========================================================================
2086 * SPA Scrubbing
2087 * ==========================================================================
2088 */
2089
2090void
2091spa_scrub_throttle(spa_t *spa, int direction)
2092{
2093	mutex_enter(&spa->spa_scrub_lock);
2094	spa->spa_scrub_throttled += direction;
2095	ASSERT(spa->spa_scrub_throttled >= 0);
2096	if (spa->spa_scrub_throttled == 0)
2097		cv_broadcast(&spa->spa_scrub_io_cv);
2098	mutex_exit(&spa->spa_scrub_lock);
2099}
2100
2101static void
2102spa_scrub_io_done(zio_t *zio)
2103{
2104	spa_t *spa = zio->io_spa;
2105
2106	zio_buf_free(zio->io_data, zio->io_size);
2107
2108	mutex_enter(&spa->spa_scrub_lock);
2109	if (zio->io_error && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
2110		vdev_t *vd = zio->io_vd ? zio->io_vd : spa->spa_root_vdev;
2111		spa->spa_scrub_errors++;
2112		mutex_enter(&vd->vdev_stat_lock);
2113		vd->vdev_stat.vs_scrub_errors++;
2114		mutex_exit(&vd->vdev_stat_lock);
2115	}
2116	if (--spa->spa_scrub_inflight == 0) {
2117		cv_broadcast(&spa->spa_scrub_io_cv);
2118		ASSERT(spa->spa_scrub_throttled == 0);
2119	}
2120	mutex_exit(&spa->spa_scrub_lock);
2121}
2122
2123static void
2124spa_scrub_io_start(spa_t *spa, blkptr_t *bp, int priority, int flags,
2125    zbookmark_t *zb)
2126{
2127	size_t size = BP_GET_LSIZE(bp);
2128	void *data = zio_buf_alloc(size);
2129
2130	mutex_enter(&spa->spa_scrub_lock);
2131	spa->spa_scrub_inflight++;
2132	mutex_exit(&spa->spa_scrub_lock);
2133
2134	if (zb->zb_level == -1 && BP_GET_TYPE(bp) != DMU_OT_OBJSET)
2135		flags |= ZIO_FLAG_SPECULATIVE;	/* intent log block */
2136
2137	flags |= ZIO_FLAG_SCRUB_THREAD | ZIO_FLAG_CANFAIL;
2138
2139	zio_nowait(zio_read(NULL, spa, bp, data, size,
2140	    spa_scrub_io_done, NULL, priority, flags, zb));
2141}
2142
2143/* ARGSUSED */
2144static int
2145spa_scrub_cb(traverse_blk_cache_t *bc, spa_t *spa, void *a)
2146{
2147	blkptr_t *bp = &bc->bc_blkptr;
2148	vdev_t *vd = spa->spa_root_vdev;
2149	dva_t *dva = bp->blk_dva;
2150	int needs_resilver = B_FALSE;
2151	int d;
2152
2153	if (bc->bc_errno) {
2154		/*
2155		 * We can't scrub this block, but we can continue to scrub
2156		 * the rest of the pool.  Note the error and move along.
2157		 */
2158		mutex_enter(&spa->spa_scrub_lock);
2159		spa->spa_scrub_errors++;
2160		mutex_exit(&spa->spa_scrub_lock);
2161
2162		mutex_enter(&vd->vdev_stat_lock);
2163		vd->vdev_stat.vs_scrub_errors++;
2164		mutex_exit(&vd->vdev_stat_lock);
2165
2166		return (ERESTART);
2167	}
2168
2169	ASSERT(bp->blk_birth < spa->spa_scrub_maxtxg);
2170
2171	for (d = 0; d < BP_GET_NDVAS(bp); d++) {
2172		vd = vdev_lookup_top(spa, DVA_GET_VDEV(&dva[d]));
2173
2174		ASSERT(vd != NULL);
2175
2176		/*
2177		 * Keep track of how much data we've examined so that
2178		 * zpool(1M) status can make useful progress reports.
2179		 */
2180		mutex_enter(&vd->vdev_stat_lock);
2181		vd->vdev_stat.vs_scrub_examined += DVA_GET_ASIZE(&dva[d]);
2182		mutex_exit(&vd->vdev_stat_lock);
2183
2184		if (spa->spa_scrub_type == POOL_SCRUB_RESILVER) {
2185			if (DVA_GET_GANG(&dva[d])) {
2186				/*
2187				 * Gang members may be spread across multiple
2188				 * vdevs, so the best we can do is look at the
2189				 * pool-wide DTL.
2190				 * XXX -- it would be better to change our
2191				 * allocation policy to ensure that this can't
2192				 * happen.
2193				 */
2194				vd = spa->spa_root_vdev;
2195			}
2196			if (vdev_dtl_contains(&vd->vdev_dtl_map,
2197			    bp->blk_birth, 1))
2198				needs_resilver = B_TRUE;
2199		}
2200	}
2201
2202	if (spa->spa_scrub_type == POOL_SCRUB_EVERYTHING)
2203		spa_scrub_io_start(spa, bp, ZIO_PRIORITY_SCRUB,
2204		    ZIO_FLAG_SCRUB, &bc->bc_bookmark);
2205	else if (needs_resilver)
2206		spa_scrub_io_start(spa, bp, ZIO_PRIORITY_RESILVER,
2207		    ZIO_FLAG_RESILVER, &bc->bc_bookmark);
2208
2209	return (0);
2210}
2211
2212static void
2213spa_scrub_thread(spa_t *spa)
2214{
2215	callb_cpr_t cprinfo;
2216	traverse_handle_t *th = spa->spa_scrub_th;
2217	vdev_t *rvd = spa->spa_root_vdev;
2218	pool_scrub_type_t scrub_type = spa->spa_scrub_type;
2219	int error = 0;
2220	boolean_t complete;
2221
2222	CALLB_CPR_INIT(&cprinfo, &spa->spa_scrub_lock, callb_generic_cpr, FTAG);
2223
2224	/*
2225	 * If we're restarting due to a snapshot create/delete,
2226	 * wait for that to complete.
2227	 */
2228	txg_wait_synced(spa_get_dsl(spa), 0);
2229
2230	dprintf("start %s mintxg=%llu maxtxg=%llu\n",
2231	    scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub",
2232	    spa->spa_scrub_mintxg, spa->spa_scrub_maxtxg);
2233
2234	spa_config_enter(spa, RW_WRITER, FTAG);
2235	vdev_reopen(rvd);		/* purge all vdev caches */
2236	vdev_config_dirty(rvd);		/* rewrite all disk labels */
2237	vdev_scrub_stat_update(rvd, scrub_type, B_FALSE);
2238	spa_config_exit(spa, FTAG);
2239
2240	mutex_enter(&spa->spa_scrub_lock);
2241	spa->spa_scrub_errors = 0;
2242	spa->spa_scrub_active = 1;
2243	ASSERT(spa->spa_scrub_inflight == 0);
2244	ASSERT(spa->spa_scrub_throttled == 0);
2245
2246	while (!spa->spa_scrub_stop) {
2247		CALLB_CPR_SAFE_BEGIN(&cprinfo);
2248		while (spa->spa_scrub_suspended) {
2249			spa->spa_scrub_active = 0;
2250			cv_broadcast(&spa->spa_scrub_cv);
2251			cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
2252			spa->spa_scrub_active = 1;
2253		}
2254		CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_scrub_lock);
2255
2256		if (spa->spa_scrub_restart_txg != 0)
2257			break;
2258
2259		mutex_exit(&spa->spa_scrub_lock);
2260		error = traverse_more(th);
2261		mutex_enter(&spa->spa_scrub_lock);
2262		if (error != EAGAIN)
2263			break;
2264
2265		while (spa->spa_scrub_throttled > 0)
2266			cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2267	}
2268
2269	while (spa->spa_scrub_inflight)
2270		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2271
2272	spa->spa_scrub_active = 0;
2273	cv_broadcast(&spa->spa_scrub_cv);
2274
2275	mutex_exit(&spa->spa_scrub_lock);
2276
2277	spa_config_enter(spa, RW_WRITER, FTAG);
2278
2279	mutex_enter(&spa->spa_scrub_lock);
2280
2281	/*
2282	 * Note: we check spa_scrub_restart_txg under both spa_scrub_lock
2283	 * AND the spa config lock to synchronize with any config changes
2284	 * that revise the DTLs under spa_vdev_enter() / spa_vdev_exit().
2285	 */
2286	if (spa->spa_scrub_restart_txg != 0)
2287		error = ERESTART;
2288
2289	if (spa->spa_scrub_stop)
2290		error = EINTR;
2291
2292	/*
2293	 * Even if there were uncorrectable errors, we consider the scrub
2294	 * completed.  The downside is that if there is a transient error during
2295	 * a resilver, we won't resilver the data properly to the target.  But
2296	 * if the damage is permanent (more likely) we will resilver forever,
2297	 * which isn't really acceptable.  Since there is enough information for
2298	 * the user to know what has failed and why, this seems like a more
2299	 * tractable approach.
2300	 */
2301	complete = (error == 0);
2302
2303	dprintf("end %s to maxtxg=%llu %s, traverse=%d, %llu errors, stop=%u\n",
2304	    scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub",
2305	    spa->spa_scrub_maxtxg, complete ? "done" : "FAILED",
2306	    error, spa->spa_scrub_errors, spa->spa_scrub_stop);
2307
2308	mutex_exit(&spa->spa_scrub_lock);
2309
2310	/*
2311	 * If the scrub/resilver completed, update all DTLs to reflect this.
2312	 * Whether it succeeded or not, vacate all temporary scrub DTLs.
2313	 */
2314	vdev_dtl_reassess(rvd, spa_last_synced_txg(spa) + 1,
2315	    complete ? spa->spa_scrub_maxtxg : 0, B_TRUE);
2316	vdev_scrub_stat_update(rvd, POOL_SCRUB_NONE, complete);
2317	spa_errlog_rotate(spa);
2318
2319	spa_config_exit(spa, FTAG);
2320
2321	mutex_enter(&spa->spa_scrub_lock);
2322
2323	/*
2324	 * We may have finished replacing a device.
2325	 * Let the async thread assess this and handle the detach.
2326	 */
2327	spa_async_request(spa, SPA_ASYNC_REPLACE_DONE);
2328
2329	/*
2330	 * If we were told to restart, our final act is to start a new scrub.
2331	 */
2332	if (error == ERESTART)
2333		spa_async_request(spa, scrub_type == POOL_SCRUB_RESILVER ?
2334		    SPA_ASYNC_RESILVER : SPA_ASYNC_SCRUB);
2335
2336	spa->spa_scrub_type = POOL_SCRUB_NONE;
2337	spa->spa_scrub_active = 0;
2338	spa->spa_scrub_thread = NULL;
2339	cv_broadcast(&spa->spa_scrub_cv);
2340	CALLB_CPR_EXIT(&cprinfo);	/* drops &spa->spa_scrub_lock */
2341	thread_exit();
2342}
2343
2344void
2345spa_scrub_suspend(spa_t *spa)
2346{
2347	mutex_enter(&spa->spa_scrub_lock);
2348	spa->spa_scrub_suspended++;
2349	while (spa->spa_scrub_active) {
2350		cv_broadcast(&spa->spa_scrub_cv);
2351		cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
2352	}
2353	while (spa->spa_scrub_inflight)
2354		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2355	mutex_exit(&spa->spa_scrub_lock);
2356}
2357
2358void
2359spa_scrub_resume(spa_t *spa)
2360{
2361	mutex_enter(&spa->spa_scrub_lock);
2362	ASSERT(spa->spa_scrub_suspended != 0);
2363	if (--spa->spa_scrub_suspended == 0)
2364		cv_broadcast(&spa->spa_scrub_cv);
2365	mutex_exit(&spa->spa_scrub_lock);
2366}
2367
2368void
2369spa_scrub_restart(spa_t *spa, uint64_t txg)
2370{
2371	/*
2372	 * Something happened (e.g. snapshot create/delete) that means
2373	 * we must restart any in-progress scrubs.  The itinerary will
2374	 * fix this properly.
2375	 */
2376	mutex_enter(&spa->spa_scrub_lock);
2377	spa->spa_scrub_restart_txg = txg;
2378	mutex_exit(&spa->spa_scrub_lock);
2379}
2380
2381int
2382spa_scrub(spa_t *spa, pool_scrub_type_t type, boolean_t force)
2383{
2384	space_seg_t *ss;
2385	uint64_t mintxg, maxtxg;
2386	vdev_t *rvd = spa->spa_root_vdev;
2387
2388	if ((uint_t)type >= POOL_SCRUB_TYPES)
2389		return (ENOTSUP);
2390
2391	mutex_enter(&spa->spa_scrub_lock);
2392
2393	/*
2394	 * If there's a scrub or resilver already in progress, stop it.
2395	 */
2396	while (spa->spa_scrub_thread != NULL) {
2397		/*
2398		 * Don't stop a resilver unless forced.
2399		 */
2400		if (spa->spa_scrub_type == POOL_SCRUB_RESILVER && !force) {
2401			mutex_exit(&spa->spa_scrub_lock);
2402			return (EBUSY);
2403		}
2404		spa->spa_scrub_stop = 1;
2405		cv_broadcast(&spa->spa_scrub_cv);
2406		cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
2407	}
2408
2409	/*
2410	 * Terminate the previous traverse.
2411	 */
2412	if (spa->spa_scrub_th != NULL) {
2413		traverse_fini(spa->spa_scrub_th);
2414		spa->spa_scrub_th = NULL;
2415	}
2416
2417	if (rvd == NULL) {
2418		ASSERT(spa->spa_scrub_stop == 0);
2419		ASSERT(spa->spa_scrub_type == type);
2420		ASSERT(spa->spa_scrub_restart_txg == 0);
2421		mutex_exit(&spa->spa_scrub_lock);
2422		return (0);
2423	}
2424
2425	mintxg = TXG_INITIAL - 1;
2426	maxtxg = spa_last_synced_txg(spa) + 1;
2427
2428	mutex_enter(&rvd->vdev_dtl_lock);
2429
2430	if (rvd->vdev_dtl_map.sm_space == 0) {
2431		/*
2432		 * The pool-wide DTL is empty.
2433		 * If this is a resilver, there's nothing to do except
2434		 * check whether any in-progress replacements have completed.
2435		 */
2436		if (type == POOL_SCRUB_RESILVER) {
2437			type = POOL_SCRUB_NONE;
2438			spa_async_request(spa, SPA_ASYNC_REPLACE_DONE);
2439		}
2440	} else {
2441		/*
2442		 * The pool-wide DTL is non-empty.
2443		 * If this is a normal scrub, upgrade to a resilver instead.
2444		 */
2445		if (type == POOL_SCRUB_EVERYTHING)
2446			type = POOL_SCRUB_RESILVER;
2447	}
2448
2449	if (type == POOL_SCRUB_RESILVER) {
2450		/*
2451		 * Determine the resilvering boundaries.
2452		 *
2453		 * Note: (mintxg, maxtxg) is an open interval,
2454		 * i.e. mintxg and maxtxg themselves are not included.
2455		 *
2456		 * Note: for maxtxg, we MIN with spa_last_synced_txg(spa) + 1
2457		 * so we don't claim to resilver a txg that's still changing.
2458		 */
2459		ss = avl_first(&rvd->vdev_dtl_map.sm_root);
2460		mintxg = ss->ss_start - 1;
2461		ss = avl_last(&rvd->vdev_dtl_map.sm_root);
2462		maxtxg = MIN(ss->ss_end, maxtxg);
2463	}
2464
2465	mutex_exit(&rvd->vdev_dtl_lock);
2466
2467	spa->spa_scrub_stop = 0;
2468	spa->spa_scrub_type = type;
2469	spa->spa_scrub_restart_txg = 0;
2470
2471	if (type != POOL_SCRUB_NONE) {
2472		spa->spa_scrub_mintxg = mintxg;
2473		spa->spa_scrub_maxtxg = maxtxg;
2474		spa->spa_scrub_th = traverse_init(spa, spa_scrub_cb, NULL,
2475		    ADVANCE_PRE | ADVANCE_PRUNE | ADVANCE_ZIL,
2476		    ZIO_FLAG_CANFAIL);
2477		traverse_add_pool(spa->spa_scrub_th, mintxg, maxtxg);
2478		spa->spa_scrub_thread = thread_create(NULL, 0,
2479		    spa_scrub_thread, spa, 0, &p0, TS_RUN, minclsyspri);
2480	}
2481
2482	mutex_exit(&spa->spa_scrub_lock);
2483
2484	return (0);
2485}
2486
2487/*
2488 * ==========================================================================
2489 * SPA async task processing
2490 * ==========================================================================
2491 */
2492
2493static void
2494spa_async_reopen(spa_t *spa)
2495{
2496	vdev_t *rvd = spa->spa_root_vdev;
2497	vdev_t *tvd;
2498	int c;
2499
2500	spa_config_enter(spa, RW_WRITER, FTAG);
2501
2502	for (c = 0; c < rvd->vdev_children; c++) {
2503		tvd = rvd->vdev_child[c];
2504		if (tvd->vdev_reopen_wanted) {
2505			tvd->vdev_reopen_wanted = 0;
2506			vdev_reopen(tvd);
2507		}
2508	}
2509
2510	spa_config_exit(spa, FTAG);
2511}
2512
2513static void
2514spa_async_thread(spa_t *spa)
2515{
2516	int tasks;
2517
2518	ASSERT(spa->spa_sync_on);
2519
2520	mutex_enter(&spa->spa_async_lock);
2521	tasks = spa->spa_async_tasks;
2522	spa->spa_async_tasks = 0;
2523	mutex_exit(&spa->spa_async_lock);
2524
2525	/*
2526	 * See if the config needs to be updated.
2527	 */
2528	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
2529		mutex_enter(&spa_namespace_lock);
2530		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
2531		mutex_exit(&spa_namespace_lock);
2532	}
2533
2534	/*
2535	 * See if any devices need to be reopened.
2536	 */
2537	if (tasks & SPA_ASYNC_REOPEN)
2538		spa_async_reopen(spa);
2539
2540	/*
2541	 * If any devices are done replacing, detach them.
2542	 */
2543	if (tasks & SPA_ASYNC_REPLACE_DONE)
2544		spa_vdev_replace_done(spa);
2545
2546	/*
2547	 * Kick off a scrub.
2548	 */
2549	if (tasks & SPA_ASYNC_SCRUB)
2550		VERIFY(spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_TRUE) == 0);
2551
2552	/*
2553	 * Kick off a resilver.
2554	 */
2555	if (tasks & SPA_ASYNC_RESILVER)
2556		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
2557
2558	/*
2559	 * Let the world know that we're done.
2560	 */
2561	mutex_enter(&spa->spa_async_lock);
2562	spa->spa_async_thread = NULL;
2563	cv_broadcast(&spa->spa_async_cv);
2564	mutex_exit(&spa->spa_async_lock);
2565	thread_exit();
2566}
2567
2568void
2569spa_async_suspend(spa_t *spa)
2570{
2571	mutex_enter(&spa->spa_async_lock);
2572	spa->spa_async_suspended++;
2573	while (spa->spa_async_thread != NULL)
2574		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
2575	mutex_exit(&spa->spa_async_lock);
2576}
2577
2578void
2579spa_async_resume(spa_t *spa)
2580{
2581	mutex_enter(&spa->spa_async_lock);
2582	ASSERT(spa->spa_async_suspended != 0);
2583	spa->spa_async_suspended--;
2584	mutex_exit(&spa->spa_async_lock);
2585}
2586
2587static void
2588spa_async_dispatch(spa_t *spa)
2589{
2590	mutex_enter(&spa->spa_async_lock);
2591	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
2592	    spa->spa_async_thread == NULL &&
2593	    rootdir != NULL && !vn_is_readonly(rootdir))
2594		spa->spa_async_thread = thread_create(NULL, 0,
2595		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
2596	mutex_exit(&spa->spa_async_lock);
2597}
2598
2599void
2600spa_async_request(spa_t *spa, int task)
2601{
2602	mutex_enter(&spa->spa_async_lock);
2603	spa->spa_async_tasks |= task;
2604	mutex_exit(&spa->spa_async_lock);
2605}
2606
2607/*
2608 * ==========================================================================
2609 * SPA syncing routines
2610 * ==========================================================================
2611 */
2612
2613static void
2614spa_sync_deferred_frees(spa_t *spa, uint64_t txg)
2615{
2616	bplist_t *bpl = &spa->spa_sync_bplist;
2617	dmu_tx_t *tx;
2618	blkptr_t blk;
2619	uint64_t itor = 0;
2620	zio_t *zio;
2621	int error;
2622	uint8_t c = 1;
2623
2624	zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CONFIG_HELD);
2625
2626	while (bplist_iterate(bpl, &itor, &blk) == 0)
2627		zio_nowait(zio_free(zio, spa, txg, &blk, NULL, NULL));
2628
2629	error = zio_wait(zio);
2630	ASSERT3U(error, ==, 0);
2631
2632	tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2633	bplist_vacate(bpl, tx);
2634
2635	/*
2636	 * Pre-dirty the first block so we sync to convergence faster.
2637	 * (Usually only the first block is needed.)
2638	 */
2639	dmu_write(spa->spa_meta_objset, spa->spa_sync_bplist_obj, 0, 1, &c, tx);
2640	dmu_tx_commit(tx);
2641}
2642
2643static void
2644spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
2645{
2646	char *packed = NULL;
2647	size_t nvsize = 0;
2648	dmu_buf_t *db;
2649
2650	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
2651
2652	packed = kmem_alloc(nvsize, KM_SLEEP);
2653
2654	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
2655	    KM_SLEEP) == 0);
2656
2657	dmu_write(spa->spa_meta_objset, obj, 0, nvsize, packed, tx);
2658
2659	kmem_free(packed, nvsize);
2660
2661	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
2662	dmu_buf_will_dirty(db, tx);
2663	*(uint64_t *)db->db_data = nvsize;
2664	dmu_buf_rele(db, FTAG);
2665}
2666
2667static void
2668spa_sync_spares(spa_t *spa, dmu_tx_t *tx)
2669{
2670	nvlist_t *nvroot;
2671	nvlist_t **spares;
2672	int i;
2673
2674	if (!spa->spa_sync_spares)
2675		return;
2676
2677	/*
2678	 * Update the MOS nvlist describing the list of available spares.
2679	 * spa_validate_spares() will have already made sure this nvlist is
2680	 * valid and the vdevs are labelled appropriately.
2681	 */
2682	if (spa->spa_spares_object == 0) {
2683		spa->spa_spares_object = dmu_object_alloc(spa->spa_meta_objset,
2684		    DMU_OT_PACKED_NVLIST, 1 << 14,
2685		    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2686		VERIFY(zap_update(spa->spa_meta_objset,
2687		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SPARES,
2688		    sizeof (uint64_t), 1, &spa->spa_spares_object, tx) == 0);
2689	}
2690
2691	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2692	if (spa->spa_nspares == 0) {
2693		VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2694		    NULL, 0) == 0);
2695	} else {
2696		spares = kmem_alloc(spa->spa_nspares * sizeof (void *),
2697		    KM_SLEEP);
2698		for (i = 0; i < spa->spa_nspares; i++)
2699			spares[i] = vdev_config_generate(spa,
2700			    spa->spa_spares[i], B_FALSE, B_TRUE);
2701		VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2702		    spares, spa->spa_nspares) == 0);
2703		for (i = 0; i < spa->spa_nspares; i++)
2704			nvlist_free(spares[i]);
2705		kmem_free(spares, spa->spa_nspares * sizeof (void *));
2706	}
2707
2708	spa_sync_nvlist(spa, spa->spa_spares_object, nvroot, tx);
2709
2710	spa->spa_sync_spares = B_FALSE;
2711}
2712
2713static void
2714spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
2715{
2716	nvlist_t *config;
2717
2718	if (list_is_empty(&spa->spa_dirty_list))
2719		return;
2720
2721	config = spa_config_generate(spa, NULL, dmu_tx_get_txg(tx), B_FALSE);
2722
2723	if (spa->spa_config_syncing)
2724		nvlist_free(spa->spa_config_syncing);
2725	spa->spa_config_syncing = config;
2726
2727	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
2728}
2729
2730/*
2731 * Sync the specified transaction group.  New blocks may be dirtied as
2732 * part of the process, so we iterate until it converges.
2733 */
2734void
2735spa_sync(spa_t *spa, uint64_t txg)
2736{
2737	dsl_pool_t *dp = spa->spa_dsl_pool;
2738	objset_t *mos = spa->spa_meta_objset;
2739	bplist_t *bpl = &spa->spa_sync_bplist;
2740	vdev_t *rvd = spa->spa_root_vdev;
2741	vdev_t *vd;
2742	dmu_tx_t *tx;
2743	int dirty_vdevs;
2744
2745	/*
2746	 * Lock out configuration changes.
2747	 */
2748	spa_config_enter(spa, RW_READER, FTAG);
2749
2750	spa->spa_syncing_txg = txg;
2751	spa->spa_sync_pass = 0;
2752
2753	VERIFY(0 == bplist_open(bpl, mos, spa->spa_sync_bplist_obj));
2754
2755	tx = dmu_tx_create_assigned(dp, txg);
2756
2757	/*
2758	 * If we are upgrading to ZFS_VERSION_RAIDZ_DEFLATE this txg,
2759	 * set spa_deflate if we have no raid-z vdevs.
2760	 */
2761	if (spa->spa_ubsync.ub_version < ZFS_VERSION_RAIDZ_DEFLATE &&
2762	    spa->spa_uberblock.ub_version >= ZFS_VERSION_RAIDZ_DEFLATE) {
2763		int i;
2764
2765		for (i = 0; i < rvd->vdev_children; i++) {
2766			vd = rvd->vdev_child[i];
2767			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
2768				break;
2769		}
2770		if (i == rvd->vdev_children) {
2771			spa->spa_deflate = TRUE;
2772			VERIFY(0 == zap_add(spa->spa_meta_objset,
2773			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
2774			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
2775		}
2776	}
2777
2778	/*
2779	 * If anything has changed in this txg, push the deferred frees
2780	 * from the previous txg.  If not, leave them alone so that we
2781	 * don't generate work on an otherwise idle system.
2782	 */
2783	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
2784	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
2785	    !txg_list_empty(&dp->dp_sync_tasks, txg))
2786		spa_sync_deferred_frees(spa, txg);
2787
2788	/*
2789	 * Iterate to convergence.
2790	 */
2791	do {
2792		spa->spa_sync_pass++;
2793
2794		spa_sync_config_object(spa, tx);
2795		spa_sync_spares(spa, tx);
2796		spa_errlog_sync(spa, txg);
2797		dsl_pool_sync(dp, txg);
2798
2799		dirty_vdevs = 0;
2800		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) {
2801			vdev_sync(vd, txg);
2802			dirty_vdevs++;
2803		}
2804
2805		bplist_sync(bpl, tx);
2806	} while (dirty_vdevs);
2807
2808	bplist_close(bpl);
2809
2810	dprintf("txg %llu passes %d\n", txg, spa->spa_sync_pass);
2811
2812	/*
2813	 * Rewrite the vdev configuration (which includes the uberblock)
2814	 * to commit the transaction group.
2815	 *
2816	 * If there are any dirty vdevs, sync the uberblock to all vdevs.
2817	 * Otherwise, pick a random top-level vdev that's known to be
2818	 * visible in the config cache (see spa_vdev_add() for details).
2819	 * If the write fails, try the next vdev until we're tried them all.
2820	 */
2821	if (!list_is_empty(&spa->spa_dirty_list)) {
2822		VERIFY(vdev_config_sync(rvd, txg) == 0);
2823	} else {
2824		int children = rvd->vdev_children;
2825		int c0 = spa_get_random(children);
2826		int c;
2827
2828		for (c = 0; c < children; c++) {
2829			vd = rvd->vdev_child[(c0 + c) % children];
2830			if (vd->vdev_ms_array == 0)
2831				continue;
2832			if (vdev_config_sync(vd, txg) == 0)
2833				break;
2834		}
2835		if (c == children)
2836			VERIFY(vdev_config_sync(rvd, txg) == 0);
2837	}
2838
2839	dmu_tx_commit(tx);
2840
2841	/*
2842	 * Clear the dirty config list.
2843	 */
2844	while ((vd = list_head(&spa->spa_dirty_list)) != NULL)
2845		vdev_config_clean(vd);
2846
2847	/*
2848	 * Now that the new config has synced transactionally,
2849	 * let it become visible to the config cache.
2850	 */
2851	if (spa->spa_config_syncing != NULL) {
2852		spa_config_set(spa, spa->spa_config_syncing);
2853		spa->spa_config_txg = txg;
2854		spa->spa_config_syncing = NULL;
2855	}
2856
2857	/*
2858	 * Make a stable copy of the fully synced uberblock.
2859	 * We use this as the root for pool traversals.
2860	 */
2861	spa->spa_traverse_wanted = 1;	/* tells traverse_more() to stop */
2862
2863	spa_scrub_suspend(spa);		/* stop scrubbing and finish I/Os */
2864
2865	rw_enter(&spa->spa_traverse_lock, RW_WRITER);
2866	spa->spa_traverse_wanted = 0;
2867	spa->spa_ubsync = spa->spa_uberblock;
2868	rw_exit(&spa->spa_traverse_lock);
2869
2870	spa_scrub_resume(spa);		/* resume scrub with new ubsync */
2871
2872	/*
2873	 * Clean up the ZIL records for the synced txg.
2874	 */
2875	dsl_pool_zil_clean(dp);
2876
2877	/*
2878	 * Update usable space statistics.
2879	 */
2880	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
2881		vdev_sync_done(vd, txg);
2882
2883	/*
2884	 * It had better be the case that we didn't dirty anything
2885	 * since vdev_config_sync().
2886	 */
2887	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
2888	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
2889	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
2890	ASSERT(bpl->bpl_queue == NULL);
2891
2892	spa_config_exit(spa, FTAG);
2893
2894	/*
2895	 * If any async tasks have been requested, kick them off.
2896	 */
2897	spa_async_dispatch(spa);
2898}
2899
2900/*
2901 * Sync all pools.  We don't want to hold the namespace lock across these
2902 * operations, so we take a reference on the spa_t and drop the lock during the
2903 * sync.
2904 */
2905void
2906spa_sync_allpools(void)
2907{
2908	spa_t *spa = NULL;
2909	mutex_enter(&spa_namespace_lock);
2910	while ((spa = spa_next(spa)) != NULL) {
2911		if (spa_state(spa) != POOL_STATE_ACTIVE)
2912			continue;
2913		spa_open_ref(spa, FTAG);
2914		mutex_exit(&spa_namespace_lock);
2915		txg_wait_synced(spa_get_dsl(spa), 0);
2916		mutex_enter(&spa_namespace_lock);
2917		spa_close(spa, FTAG);
2918	}
2919	mutex_exit(&spa_namespace_lock);
2920}
2921
2922/*
2923 * ==========================================================================
2924 * Miscellaneous routines
2925 * ==========================================================================
2926 */
2927
2928/*
2929 * Remove all pools in the system.
2930 */
2931void
2932spa_evict_all(void)
2933{
2934	spa_t *spa;
2935
2936	/*
2937	 * Remove all cached state.  All pools should be closed now,
2938	 * so every spa in the AVL tree should be unreferenced.
2939	 */
2940	mutex_enter(&spa_namespace_lock);
2941	while ((spa = spa_next(NULL)) != NULL) {
2942		/*
2943		 * Stop async tasks.  The async thread may need to detach
2944		 * a device that's been replaced, which requires grabbing
2945		 * spa_namespace_lock, so we must drop it here.
2946		 */
2947		spa_open_ref(spa, FTAG);
2948		mutex_exit(&spa_namespace_lock);
2949		spa_async_suspend(spa);
2950		VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0);
2951		mutex_enter(&spa_namespace_lock);
2952		spa_close(spa, FTAG);
2953
2954		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
2955			spa_unload(spa);
2956			spa_deactivate(spa);
2957		}
2958		spa_remove(spa);
2959	}
2960	mutex_exit(&spa_namespace_lock);
2961}
2962
2963vdev_t *
2964spa_lookup_by_guid(spa_t *spa, uint64_t guid)
2965{
2966	return (vdev_lookup_by_guid(spa->spa_root_vdev, guid));
2967}
2968
2969void
2970spa_upgrade(spa_t *spa)
2971{
2972	spa_config_enter(spa, RW_WRITER, FTAG);
2973
2974	/*
2975	 * This should only be called for a non-faulted pool, and since a
2976	 * future version would result in an unopenable pool, this shouldn't be
2977	 * possible.
2978	 */
2979	ASSERT(spa->spa_uberblock.ub_version <= ZFS_VERSION);
2980
2981	spa->spa_uberblock.ub_version = ZFS_VERSION;
2982	vdev_config_dirty(spa->spa_root_vdev);
2983
2984	spa_config_exit(spa, FTAG);
2985
2986	txg_wait_synced(spa_get_dsl(spa), 0);
2987}
2988
2989boolean_t
2990spa_has_spare(spa_t *spa, uint64_t guid)
2991{
2992	int i;
2993
2994	for (i = 0; i < spa->spa_nspares; i++)
2995		if (spa->spa_spares[i]->vdev_guid == guid)
2996			return (B_TRUE);
2997
2998	return (B_FALSE);
2999}
3000