spa.c revision c25056de36a33f2a76f79dcf64593f731d258013
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 2007 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/dmu_objset.h>
52#include <sys/unique.h>
53#include <sys/dsl_pool.h>
54#include <sys/dsl_dataset.h>
55#include <sys/dsl_dir.h>
56#include <sys/dsl_prop.h>
57#include <sys/dsl_synctask.h>
58#include <sys/fs/zfs.h>
59#include <sys/callb.h>
60#include <sys/systeminfo.h>
61#include <sys/sunddi.h>
62
63int zio_taskq_threads = 8;
64
65/*
66 * ==========================================================================
67 * SPA state manipulation (open/create/destroy/import/export)
68 * ==========================================================================
69 */
70
71static int
72spa_error_entry_compare(const void *a, const void *b)
73{
74	spa_error_entry_t *sa = (spa_error_entry_t *)a;
75	spa_error_entry_t *sb = (spa_error_entry_t *)b;
76	int ret;
77
78	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
79	    sizeof (zbookmark_t));
80
81	if (ret < 0)
82		return (-1);
83	else if (ret > 0)
84		return (1);
85	else
86		return (0);
87}
88
89/*
90 * Utility function which retrieves copies of the current logs and
91 * re-initializes them in the process.
92 */
93void
94spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
95{
96	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
97
98	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
99	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
100
101	avl_create(&spa->spa_errlist_scrub,
102	    spa_error_entry_compare, sizeof (spa_error_entry_t),
103	    offsetof(spa_error_entry_t, se_avl));
104	avl_create(&spa->spa_errlist_last,
105	    spa_error_entry_compare, sizeof (spa_error_entry_t),
106	    offsetof(spa_error_entry_t, se_avl));
107}
108
109/*
110 * Activate an uninitialized pool.
111 */
112static void
113spa_activate(spa_t *spa)
114{
115	int t;
116
117	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
118
119	spa->spa_state = POOL_STATE_ACTIVE;
120
121	spa->spa_normal_class = metaslab_class_create();
122	spa->spa_log_class = metaslab_class_create();
123
124	for (t = 0; t < ZIO_TYPES; t++) {
125		spa->spa_zio_issue_taskq[t] = taskq_create("spa_zio_issue",
126		    zio_taskq_threads, maxclsyspri, 50, INT_MAX,
127		    TASKQ_PREPOPULATE);
128		spa->spa_zio_intr_taskq[t] = taskq_create("spa_zio_intr",
129		    zio_taskq_threads, maxclsyspri, 50, INT_MAX,
130		    TASKQ_PREPOPULATE);
131	}
132
133	list_create(&spa->spa_dirty_list, sizeof (vdev_t),
134	    offsetof(vdev_t, vdev_dirty_node));
135
136	txg_list_create(&spa->spa_vdev_txg_list,
137	    offsetof(struct vdev, vdev_txg_node));
138
139	avl_create(&spa->spa_errlist_scrub,
140	    spa_error_entry_compare, sizeof (spa_error_entry_t),
141	    offsetof(spa_error_entry_t, se_avl));
142	avl_create(&spa->spa_errlist_last,
143	    spa_error_entry_compare, sizeof (spa_error_entry_t),
144	    offsetof(spa_error_entry_t, se_avl));
145}
146
147/*
148 * Opposite of spa_activate().
149 */
150static void
151spa_deactivate(spa_t *spa)
152{
153	int t;
154
155	ASSERT(spa->spa_sync_on == B_FALSE);
156	ASSERT(spa->spa_dsl_pool == NULL);
157	ASSERT(spa->spa_root_vdev == NULL);
158
159	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
160
161	txg_list_destroy(&spa->spa_vdev_txg_list);
162
163	list_destroy(&spa->spa_dirty_list);
164
165	for (t = 0; t < ZIO_TYPES; t++) {
166		taskq_destroy(spa->spa_zio_issue_taskq[t]);
167		taskq_destroy(spa->spa_zio_intr_taskq[t]);
168		spa->spa_zio_issue_taskq[t] = NULL;
169		spa->spa_zio_intr_taskq[t] = NULL;
170	}
171
172	metaslab_class_destroy(spa->spa_normal_class);
173	spa->spa_normal_class = NULL;
174
175	metaslab_class_destroy(spa->spa_log_class);
176	spa->spa_log_class = NULL;
177
178	/*
179	 * If this was part of an import or the open otherwise failed, we may
180	 * still have errors left in the queues.  Empty them just in case.
181	 */
182	spa_errlog_drain(spa);
183
184	avl_destroy(&spa->spa_errlist_scrub);
185	avl_destroy(&spa->spa_errlist_last);
186
187	spa->spa_state = POOL_STATE_UNINITIALIZED;
188}
189
190/*
191 * Verify a pool configuration, and construct the vdev tree appropriately.  This
192 * will create all the necessary vdevs in the appropriate layout, with each vdev
193 * in the CLOSED state.  This will prep the pool before open/creation/import.
194 * All vdev validation is done by the vdev_alloc() routine.
195 */
196static int
197spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
198    uint_t id, int atype)
199{
200	nvlist_t **child;
201	uint_t c, children;
202	int error;
203
204	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
205		return (error);
206
207	if ((*vdp)->vdev_ops->vdev_op_leaf)
208		return (0);
209
210	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
211	    &child, &children) != 0) {
212		vdev_free(*vdp);
213		*vdp = NULL;
214		return (EINVAL);
215	}
216
217	for (c = 0; c < children; c++) {
218		vdev_t *vd;
219		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
220		    atype)) != 0) {
221			vdev_free(*vdp);
222			*vdp = NULL;
223			return (error);
224		}
225	}
226
227	ASSERT(*vdp != NULL);
228
229	return (0);
230}
231
232/*
233 * Opposite of spa_load().
234 */
235static void
236spa_unload(spa_t *spa)
237{
238	int i;
239
240	/*
241	 * Stop async tasks.
242	 */
243	spa_async_suspend(spa);
244
245	/*
246	 * Stop syncing.
247	 */
248	if (spa->spa_sync_on) {
249		txg_sync_stop(spa->spa_dsl_pool);
250		spa->spa_sync_on = B_FALSE;
251	}
252
253	/*
254	 * Wait for any outstanding prefetch I/O to complete.
255	 */
256	spa_config_enter(spa, RW_WRITER, FTAG);
257	spa_config_exit(spa, FTAG);
258
259	/*
260	 * Close the dsl pool.
261	 */
262	if (spa->spa_dsl_pool) {
263		dsl_pool_close(spa->spa_dsl_pool);
264		spa->spa_dsl_pool = NULL;
265	}
266
267	/*
268	 * Close all vdevs.
269	 */
270	if (spa->spa_root_vdev)
271		vdev_free(spa->spa_root_vdev);
272	ASSERT(spa->spa_root_vdev == NULL);
273
274	for (i = 0; i < spa->spa_nspares; i++)
275		vdev_free(spa->spa_spares[i]);
276	if (spa->spa_spares) {
277		kmem_free(spa->spa_spares, spa->spa_nspares * sizeof (void *));
278		spa->spa_spares = NULL;
279	}
280	if (spa->spa_sparelist) {
281		nvlist_free(spa->spa_sparelist);
282		spa->spa_sparelist = NULL;
283	}
284
285	spa->spa_async_suspended = 0;
286}
287
288/*
289 * Load (or re-load) the current list of vdevs describing the active spares for
290 * this pool.  When this is called, we have some form of basic information in
291 * 'spa_sparelist'.  We parse this into vdevs, try to open them, and then
292 * re-generate a more complete list including status information.
293 */
294static void
295spa_load_spares(spa_t *spa)
296{
297	nvlist_t **spares;
298	uint_t nspares;
299	int i;
300	vdev_t *vd, *tvd;
301
302	/*
303	 * First, close and free any existing spare vdevs.
304	 */
305	for (i = 0; i < spa->spa_nspares; i++) {
306		vd = spa->spa_spares[i];
307
308		/* Undo the call to spa_activate() below */
309		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid)) != NULL &&
310		    tvd->vdev_isspare)
311			spa_spare_remove(tvd);
312		vdev_close(vd);
313		vdev_free(vd);
314	}
315
316	if (spa->spa_spares)
317		kmem_free(spa->spa_spares, spa->spa_nspares * sizeof (void *));
318
319	if (spa->spa_sparelist == NULL)
320		nspares = 0;
321	else
322		VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist,
323		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
324
325	spa->spa_nspares = (int)nspares;
326	spa->spa_spares = NULL;
327
328	if (nspares == 0)
329		return;
330
331	/*
332	 * Construct the array of vdevs, opening them to get status in the
333	 * process.   For each spare, there is potentially two different vdev_t
334	 * structures associated with it: one in the list of spares (used only
335	 * for basic validation purposes) and one in the active vdev
336	 * configuration (if it's spared in).  During this phase we open and
337	 * validate each vdev on the spare list.  If the vdev also exists in the
338	 * active configuration, then we also mark this vdev as an active spare.
339	 */
340	spa->spa_spares = kmem_alloc(nspares * sizeof (void *), KM_SLEEP);
341	for (i = 0; i < spa->spa_nspares; i++) {
342		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
343		    VDEV_ALLOC_SPARE) == 0);
344		ASSERT(vd != NULL);
345
346		spa->spa_spares[i] = vd;
347
348		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid)) != NULL) {
349			if (!tvd->vdev_isspare)
350				spa_spare_add(tvd);
351
352			/*
353			 * We only mark the spare active if we were successfully
354			 * able to load the vdev.  Otherwise, importing a pool
355			 * with a bad active spare would result in strange
356			 * behavior, because multiple pool would think the spare
357			 * is actively in use.
358			 *
359			 * There is a vulnerability here to an equally bizarre
360			 * circumstance, where a dead active spare is later
361			 * brought back to life (onlined or otherwise).  Given
362			 * the rarity of this scenario, and the extra complexity
363			 * it adds, we ignore the possibility.
364			 */
365			if (!vdev_is_dead(tvd))
366				spa_spare_activate(tvd);
367		}
368
369		if (vdev_open(vd) != 0)
370			continue;
371
372		vd->vdev_top = vd;
373		(void) vdev_validate_spare(vd);
374	}
375
376	/*
377	 * Recompute the stashed list of spares, with status information
378	 * this time.
379	 */
380	VERIFY(nvlist_remove(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
381	    DATA_TYPE_NVLIST_ARRAY) == 0);
382
383	spares = kmem_alloc(spa->spa_nspares * sizeof (void *), KM_SLEEP);
384	for (i = 0; i < spa->spa_nspares; i++)
385		spares[i] = vdev_config_generate(spa, spa->spa_spares[i],
386		    B_TRUE, B_TRUE);
387	VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
388	    spares, spa->spa_nspares) == 0);
389	for (i = 0; i < spa->spa_nspares; i++)
390		nvlist_free(spares[i]);
391	kmem_free(spares, spa->spa_nspares * sizeof (void *));
392}
393
394static int
395load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
396{
397	dmu_buf_t *db;
398	char *packed = NULL;
399	size_t nvsize = 0;
400	int error;
401	*value = NULL;
402
403	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
404	nvsize = *(uint64_t *)db->db_data;
405	dmu_buf_rele(db, FTAG);
406
407	packed = kmem_alloc(nvsize, KM_SLEEP);
408	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed);
409	if (error == 0)
410		error = nvlist_unpack(packed, nvsize, value, 0);
411	kmem_free(packed, nvsize);
412
413	return (error);
414}
415
416/*
417 * Checks to see if the given vdev could not be opened, in which case we post a
418 * sysevent to notify the autoreplace code that the device has been removed.
419 */
420static void
421spa_check_removed(vdev_t *vd)
422{
423	int c;
424
425	for (c = 0; c < vd->vdev_children; c++)
426		spa_check_removed(vd->vdev_child[c]);
427
428	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
429		zfs_post_autoreplace(vd->vdev_spa, vd);
430		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
431	}
432}
433
434/*
435 * Load an existing storage pool, using the pool's builtin spa_config as a
436 * source of configuration information.
437 */
438static int
439spa_load(spa_t *spa, nvlist_t *config, spa_load_state_t state, int mosconfig)
440{
441	int error = 0;
442	nvlist_t *nvroot = NULL;
443	vdev_t *rvd;
444	uberblock_t *ub = &spa->spa_uberblock;
445	uint64_t config_cache_txg = spa->spa_config_txg;
446	uint64_t pool_guid;
447	uint64_t version;
448	zio_t *zio;
449	uint64_t autoreplace = 0;
450
451	spa->spa_load_state = state;
452
453	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) ||
454	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
455		error = EINVAL;
456		goto out;
457	}
458
459	/*
460	 * Versioning wasn't explicitly added to the label until later, so if
461	 * it's not present treat it as the initial version.
462	 */
463	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &version) != 0)
464		version = SPA_VERSION_INITIAL;
465
466	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
467	    &spa->spa_config_txg);
468
469	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
470	    spa_guid_exists(pool_guid, 0)) {
471		error = EEXIST;
472		goto out;
473	}
474
475	spa->spa_load_guid = pool_guid;
476
477	/*
478	 * Parse the configuration into a vdev tree.  We explicitly set the
479	 * value that will be returned by spa_version() since parsing the
480	 * configuration requires knowing the version number.
481	 */
482	spa_config_enter(spa, RW_WRITER, FTAG);
483	spa->spa_ubsync.ub_version = version;
484	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_LOAD);
485	spa_config_exit(spa, FTAG);
486
487	if (error != 0)
488		goto out;
489
490	ASSERT(spa->spa_root_vdev == rvd);
491	ASSERT(spa_guid(spa) == pool_guid);
492
493	/*
494	 * Try to open all vdevs, loading each label in the process.
495	 */
496	error = vdev_open(rvd);
497	if (error != 0)
498		goto out;
499
500	/*
501	 * Validate the labels for all leaf vdevs.  We need to grab the config
502	 * lock because all label I/O is done with the ZIO_FLAG_CONFIG_HELD
503	 * flag.
504	 */
505	spa_config_enter(spa, RW_READER, FTAG);
506	error = vdev_validate(rvd);
507	spa_config_exit(spa, FTAG);
508
509	if (error != 0)
510		goto out;
511
512	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
513		error = ENXIO;
514		goto out;
515	}
516
517	/*
518	 * Find the best uberblock.
519	 */
520	bzero(ub, sizeof (uberblock_t));
521
522	zio = zio_root(spa, NULL, NULL,
523	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
524	vdev_uberblock_load(zio, rvd, ub);
525	error = zio_wait(zio);
526
527	/*
528	 * If we weren't able to find a single valid uberblock, return failure.
529	 */
530	if (ub->ub_txg == 0) {
531		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
532		    VDEV_AUX_CORRUPT_DATA);
533		error = ENXIO;
534		goto out;
535	}
536
537	/*
538	 * If the pool is newer than the code, we can't open it.
539	 */
540	if (ub->ub_version > SPA_VERSION) {
541		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
542		    VDEV_AUX_VERSION_NEWER);
543		error = ENOTSUP;
544		goto out;
545	}
546
547	/*
548	 * If the vdev guid sum doesn't match the uberblock, we have an
549	 * incomplete configuration.
550	 */
551	if (rvd->vdev_guid_sum != ub->ub_guid_sum && mosconfig) {
552		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
553		    VDEV_AUX_BAD_GUID_SUM);
554		error = ENXIO;
555		goto out;
556	}
557
558	/*
559	 * Initialize internal SPA structures.
560	 */
561	spa->spa_state = POOL_STATE_ACTIVE;
562	spa->spa_ubsync = spa->spa_uberblock;
563	spa->spa_first_txg = spa_last_synced_txg(spa) + 1;
564	error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
565	if (error) {
566		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
567		    VDEV_AUX_CORRUPT_DATA);
568		goto out;
569	}
570	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
571
572	if (zap_lookup(spa->spa_meta_objset,
573	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
574	    sizeof (uint64_t), 1, &spa->spa_config_object) != 0) {
575		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
576		    VDEV_AUX_CORRUPT_DATA);
577		error = EIO;
578		goto out;
579	}
580
581	if (!mosconfig) {
582		nvlist_t *newconfig;
583		uint64_t hostid;
584
585		if (load_nvlist(spa, spa->spa_config_object, &newconfig) != 0) {
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		if (nvlist_lookup_uint64(newconfig, ZPOOL_CONFIG_HOSTID,
593		    &hostid) == 0) {
594			char *hostname;
595			unsigned long myhostid = 0;
596
597			VERIFY(nvlist_lookup_string(newconfig,
598			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
599
600			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
601			if (hostid != 0 && myhostid != 0 &&
602			    (unsigned long)hostid != myhostid) {
603				cmn_err(CE_WARN, "pool '%s' could not be "
604				    "loaded as it was last accessed by "
605				    "another system (host: %s hostid: 0x%lx).  "
606				    "See: http://www.sun.com/msg/ZFS-8000-EY",
607				    spa->spa_name, hostname,
608				    (unsigned long)hostid);
609				error = EBADF;
610				goto out;
611			}
612		}
613
614		spa_config_set(spa, newconfig);
615		spa_unload(spa);
616		spa_deactivate(spa);
617		spa_activate(spa);
618
619		return (spa_load(spa, newconfig, state, B_TRUE));
620	}
621
622	if (zap_lookup(spa->spa_meta_objset,
623	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
624	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj) != 0) {
625		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
626		    VDEV_AUX_CORRUPT_DATA);
627		error = EIO;
628		goto out;
629	}
630
631	/*
632	 * Load the bit that tells us to use the new accounting function
633	 * (raid-z deflation).  If we have an older pool, this will not
634	 * be present.
635	 */
636	error = zap_lookup(spa->spa_meta_objset,
637	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
638	    sizeof (uint64_t), 1, &spa->spa_deflate);
639	if (error != 0 && error != ENOENT) {
640		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
641		    VDEV_AUX_CORRUPT_DATA);
642		error = EIO;
643		goto out;
644	}
645
646	/*
647	 * Load the persistent error log.  If we have an older pool, this will
648	 * not be present.
649	 */
650	error = zap_lookup(spa->spa_meta_objset,
651	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_LAST,
652	    sizeof (uint64_t), 1, &spa->spa_errlog_last);
653	if (error != 0 && error != ENOENT) {
654		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
655		    VDEV_AUX_CORRUPT_DATA);
656		error = EIO;
657		goto out;
658	}
659
660	error = zap_lookup(spa->spa_meta_objset,
661	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_SCRUB,
662	    sizeof (uint64_t), 1, &spa->spa_errlog_scrub);
663	if (error != 0 && error != ENOENT) {
664		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
665		    VDEV_AUX_CORRUPT_DATA);
666		error = EIO;
667		goto out;
668	}
669
670	/*
671	 * Load the history object.  If we have an older pool, this
672	 * will not be present.
673	 */
674	error = zap_lookup(spa->spa_meta_objset,
675	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_HISTORY,
676	    sizeof (uint64_t), 1, &spa->spa_history);
677	if (error != 0 && error != ENOENT) {
678		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
679		    VDEV_AUX_CORRUPT_DATA);
680		error = EIO;
681		goto out;
682	}
683
684	/*
685	 * Load any hot spares for this pool.
686	 */
687	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
688	    DMU_POOL_SPARES, sizeof (uint64_t), 1, &spa->spa_spares_object);
689	if (error != 0 && error != ENOENT) {
690		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
691		    VDEV_AUX_CORRUPT_DATA);
692		error = EIO;
693		goto out;
694	}
695	if (error == 0) {
696		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
697		if (load_nvlist(spa, spa->spa_spares_object,
698		    &spa->spa_sparelist) != 0) {
699			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
700			    VDEV_AUX_CORRUPT_DATA);
701			error = EIO;
702			goto out;
703		}
704
705		spa_config_enter(spa, RW_WRITER, FTAG);
706		spa_load_spares(spa);
707		spa_config_exit(spa, FTAG);
708	}
709
710	spa->spa_delegation = zfs_prop_default_numeric(ZPOOL_PROP_DELEGATION);
711
712	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
713	    DMU_POOL_PROPS, sizeof (uint64_t), 1, &spa->spa_pool_props_object);
714
715	if (error && error != ENOENT) {
716		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
717		    VDEV_AUX_CORRUPT_DATA);
718		error = EIO;
719		goto out;
720	}
721
722	if (error == 0) {
723		(void) zap_lookup(spa->spa_meta_objset,
724		    spa->spa_pool_props_object,
725		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS),
726		    sizeof (uint64_t), 1, &spa->spa_bootfs);
727		(void) zap_lookup(spa->spa_meta_objset,
728		    spa->spa_pool_props_object,
729		    zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE),
730		    sizeof (uint64_t), 1, &autoreplace);
731		(void) zap_lookup(spa->spa_meta_objset,
732		    spa->spa_pool_props_object,
733		    zpool_prop_to_name(ZPOOL_PROP_DELEGATION),
734		    sizeof (uint64_t), 1, &spa->spa_delegation);
735	}
736
737	/*
738	 * If the 'autoreplace' property is set, then post a resource notifying
739	 * the ZFS DE that it should not issue any faults for unopenable
740	 * devices.  We also iterate over the vdevs, and post a sysevent for any
741	 * unopenable vdevs so that the normal autoreplace handler can take
742	 * over.
743	 */
744	if (autoreplace)
745		spa_check_removed(spa->spa_root_vdev);
746
747	/*
748	 * Load the vdev state for all toplevel vdevs.
749	 */
750	vdev_load(rvd);
751
752	/*
753	 * Propagate the leaf DTLs we just loaded all the way up the tree.
754	 */
755	spa_config_enter(spa, RW_WRITER, FTAG);
756	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
757	spa_config_exit(spa, FTAG);
758
759	/*
760	 * Check the state of the root vdev.  If it can't be opened, it
761	 * indicates one or more toplevel vdevs are faulted.
762	 */
763	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
764		error = ENXIO;
765		goto out;
766	}
767
768	if ((spa_mode & FWRITE) && state != SPA_LOAD_TRYIMPORT) {
769		dmu_tx_t *tx;
770		int need_update = B_FALSE;
771		int c;
772
773		/*
774		 * Claim log blocks that haven't been committed yet.
775		 * This must all happen in a single txg.
776		 */
777		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
778		    spa_first_txg(spa));
779		(void) dmu_objset_find(spa->spa_name,
780		    zil_claim, tx, DS_FIND_CHILDREN);
781		dmu_tx_commit(tx);
782
783		spa->spa_sync_on = B_TRUE;
784		txg_sync_start(spa->spa_dsl_pool);
785
786		/*
787		 * Wait for all claims to sync.
788		 */
789		txg_wait_synced(spa->spa_dsl_pool, 0);
790
791		/*
792		 * If the config cache is stale, or we have uninitialized
793		 * metaslabs (see spa_vdev_add()), then update the config.
794		 */
795		if (config_cache_txg != spa->spa_config_txg ||
796		    state == SPA_LOAD_IMPORT)
797			need_update = B_TRUE;
798
799		for (c = 0; c < rvd->vdev_children; c++)
800			if (rvd->vdev_child[c]->vdev_ms_array == 0)
801				need_update = B_TRUE;
802
803		/*
804		 * Update the config cache asychronously in case we're the
805		 * root pool, in which case the config cache isn't writable yet.
806		 */
807		if (need_update)
808			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
809	}
810
811	error = 0;
812out:
813	if (error && error != EBADF)
814		zfs_ereport_post(FM_EREPORT_ZFS_POOL, spa, NULL, NULL, 0, 0);
815	spa->spa_load_state = SPA_LOAD_NONE;
816	spa->spa_ena = 0;
817
818	return (error);
819}
820
821/*
822 * Pool Open/Import
823 *
824 * The import case is identical to an open except that the configuration is sent
825 * down from userland, instead of grabbed from the configuration cache.  For the
826 * case of an open, the pool configuration will exist in the
827 * POOL_STATE_UNINITIALIZED state.
828 *
829 * The stats information (gen/count/ustats) is used to gather vdev statistics at
830 * the same time open the pool, without having to keep around the spa_t in some
831 * ambiguous state.
832 */
833static int
834spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t **config)
835{
836	spa_t *spa;
837	int error;
838	int loaded = B_FALSE;
839	int locked = B_FALSE;
840
841	*spapp = NULL;
842
843	/*
844	 * As disgusting as this is, we need to support recursive calls to this
845	 * function because dsl_dir_open() is called during spa_load(), and ends
846	 * up calling spa_open() again.  The real fix is to figure out how to
847	 * avoid dsl_dir_open() calling this in the first place.
848	 */
849	if (mutex_owner(&spa_namespace_lock) != curthread) {
850		mutex_enter(&spa_namespace_lock);
851		locked = B_TRUE;
852	}
853
854	if ((spa = spa_lookup(pool)) == NULL) {
855		if (locked)
856			mutex_exit(&spa_namespace_lock);
857		return (ENOENT);
858	}
859	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
860
861		spa_activate(spa);
862
863		error = spa_load(spa, spa->spa_config, SPA_LOAD_OPEN, B_FALSE);
864
865		if (error == EBADF) {
866			/*
867			 * If vdev_validate() returns failure (indicated by
868			 * EBADF), it indicates that one of the vdevs indicates
869			 * that the pool has been exported or destroyed.  If
870			 * this is the case, the config cache is out of sync and
871			 * we should remove the pool from the namespace.
872			 */
873			zfs_post_ok(spa, NULL);
874			spa_unload(spa);
875			spa_deactivate(spa);
876			spa_remove(spa);
877			spa_config_sync();
878			if (locked)
879				mutex_exit(&spa_namespace_lock);
880			return (ENOENT);
881		}
882
883		if (error) {
884			/*
885			 * We can't open the pool, but we still have useful
886			 * information: the state of each vdev after the
887			 * attempted vdev_open().  Return this to the user.
888			 */
889			if (config != NULL && spa->spa_root_vdev != NULL) {
890				spa_config_enter(spa, RW_READER, FTAG);
891				*config = spa_config_generate(spa, NULL, -1ULL,
892				    B_TRUE);
893				spa_config_exit(spa, FTAG);
894			}
895			spa_unload(spa);
896			spa_deactivate(spa);
897			spa->spa_last_open_failed = B_TRUE;
898			if (locked)
899				mutex_exit(&spa_namespace_lock);
900			*spapp = NULL;
901			return (error);
902		} else {
903			zfs_post_ok(spa, NULL);
904			spa->spa_last_open_failed = B_FALSE;
905		}
906
907		loaded = B_TRUE;
908	}
909
910	spa_open_ref(spa, tag);
911
912	/*
913	 * If we just loaded the pool, resilver anything that's out of date.
914	 */
915	if (loaded && (spa_mode & FWRITE))
916		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
917
918	if (locked)
919		mutex_exit(&spa_namespace_lock);
920
921	*spapp = spa;
922
923	if (config != NULL) {
924		spa_config_enter(spa, RW_READER, FTAG);
925		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
926		spa_config_exit(spa, FTAG);
927	}
928
929	return (0);
930}
931
932int
933spa_open(const char *name, spa_t **spapp, void *tag)
934{
935	return (spa_open_common(name, spapp, tag, NULL));
936}
937
938/*
939 * Lookup the given spa_t, incrementing the inject count in the process,
940 * preventing it from being exported or destroyed.
941 */
942spa_t *
943spa_inject_addref(char *name)
944{
945	spa_t *spa;
946
947	mutex_enter(&spa_namespace_lock);
948	if ((spa = spa_lookup(name)) == NULL) {
949		mutex_exit(&spa_namespace_lock);
950		return (NULL);
951	}
952	spa->spa_inject_ref++;
953	mutex_exit(&spa_namespace_lock);
954
955	return (spa);
956}
957
958void
959spa_inject_delref(spa_t *spa)
960{
961	mutex_enter(&spa_namespace_lock);
962	spa->spa_inject_ref--;
963	mutex_exit(&spa_namespace_lock);
964}
965
966static void
967spa_add_spares(spa_t *spa, nvlist_t *config)
968{
969	nvlist_t **spares;
970	uint_t i, nspares;
971	nvlist_t *nvroot;
972	uint64_t guid;
973	vdev_stat_t *vs;
974	uint_t vsc;
975	uint64_t pool;
976
977	if (spa->spa_nspares == 0)
978		return;
979
980	VERIFY(nvlist_lookup_nvlist(config,
981	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
982	VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist,
983	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
984	if (nspares != 0) {
985		VERIFY(nvlist_add_nvlist_array(nvroot,
986		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
987		VERIFY(nvlist_lookup_nvlist_array(nvroot,
988		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
989
990		/*
991		 * Go through and find any spares which have since been
992		 * repurposed as an active spare.  If this is the case, update
993		 * their status appropriately.
994		 */
995		for (i = 0; i < nspares; i++) {
996			VERIFY(nvlist_lookup_uint64(spares[i],
997			    ZPOOL_CONFIG_GUID, &guid) == 0);
998			if (spa_spare_exists(guid, &pool) && pool != 0ULL) {
999				VERIFY(nvlist_lookup_uint64_array(
1000				    spares[i], ZPOOL_CONFIG_STATS,
1001				    (uint64_t **)&vs, &vsc) == 0);
1002				vs->vs_state = VDEV_STATE_CANT_OPEN;
1003				vs->vs_aux = VDEV_AUX_SPARED;
1004			}
1005		}
1006	}
1007}
1008
1009int
1010spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
1011{
1012	int error;
1013	spa_t *spa;
1014
1015	*config = NULL;
1016	error = spa_open_common(name, &spa, FTAG, config);
1017
1018	if (spa && *config != NULL) {
1019		VERIFY(nvlist_add_uint64(*config, ZPOOL_CONFIG_ERRCOUNT,
1020		    spa_get_errlog_size(spa)) == 0);
1021
1022		spa_add_spares(spa, *config);
1023	}
1024
1025	/*
1026	 * We want to get the alternate root even for faulted pools, so we cheat
1027	 * and call spa_lookup() directly.
1028	 */
1029	if (altroot) {
1030		if (spa == NULL) {
1031			mutex_enter(&spa_namespace_lock);
1032			spa = spa_lookup(name);
1033			if (spa)
1034				spa_altroot(spa, altroot, buflen);
1035			else
1036				altroot[0] = '\0';
1037			spa = NULL;
1038			mutex_exit(&spa_namespace_lock);
1039		} else {
1040			spa_altroot(spa, altroot, buflen);
1041		}
1042	}
1043
1044	if (spa != NULL)
1045		spa_close(spa, FTAG);
1046
1047	return (error);
1048}
1049
1050/*
1051 * Validate that the 'spares' array is well formed.  We must have an array of
1052 * nvlists, each which describes a valid leaf vdev.  If this is an import (mode
1053 * is VDEV_ALLOC_SPARE), then we allow corrupted spares to be specified, as long
1054 * as they are well-formed.
1055 */
1056static int
1057spa_validate_spares(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
1058{
1059	nvlist_t **spares;
1060	uint_t i, nspares;
1061	vdev_t *vd;
1062	int error;
1063
1064	/*
1065	 * It's acceptable to have no spares specified.
1066	 */
1067	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1068	    &spares, &nspares) != 0)
1069		return (0);
1070
1071	if (nspares == 0)
1072		return (EINVAL);
1073
1074	/*
1075	 * Make sure the pool is formatted with a version that supports hot
1076	 * spares.
1077	 */
1078	if (spa_version(spa) < SPA_VERSION_SPARES)
1079		return (ENOTSUP);
1080
1081	/*
1082	 * Set the pending spare list so we correctly handle device in-use
1083	 * checking.
1084	 */
1085	spa->spa_pending_spares = spares;
1086	spa->spa_pending_nspares = nspares;
1087
1088	for (i = 0; i < nspares; i++) {
1089		if ((error = spa_config_parse(spa, &vd, spares[i], NULL, 0,
1090		    mode)) != 0)
1091			goto out;
1092
1093		if (!vd->vdev_ops->vdev_op_leaf) {
1094			vdev_free(vd);
1095			error = EINVAL;
1096			goto out;
1097		}
1098
1099		vd->vdev_top = vd;
1100
1101		if ((error = vdev_open(vd)) == 0 &&
1102		    (error = vdev_label_init(vd, crtxg,
1103		    VDEV_LABEL_SPARE)) == 0) {
1104			VERIFY(nvlist_add_uint64(spares[i], ZPOOL_CONFIG_GUID,
1105			    vd->vdev_guid) == 0);
1106		}
1107
1108		vdev_free(vd);
1109
1110		if (error && mode != VDEV_ALLOC_SPARE)
1111			goto out;
1112		else
1113			error = 0;
1114	}
1115
1116out:
1117	spa->spa_pending_spares = NULL;
1118	spa->spa_pending_nspares = 0;
1119	return (error);
1120}
1121
1122/*
1123 * Pool Creation
1124 */
1125int
1126spa_create(const char *pool, nvlist_t *nvroot, const char *altroot,
1127    const char *history_str)
1128{
1129	spa_t *spa;
1130	vdev_t *rvd;
1131	dsl_pool_t *dp;
1132	dmu_tx_t *tx;
1133	int c, error = 0;
1134	uint64_t txg = TXG_INITIAL;
1135	nvlist_t **spares;
1136	uint_t nspares;
1137
1138	/*
1139	 * If this pool already exists, return failure.
1140	 */
1141	mutex_enter(&spa_namespace_lock);
1142	if (spa_lookup(pool) != NULL) {
1143		mutex_exit(&spa_namespace_lock);
1144		return (EEXIST);
1145	}
1146
1147	/*
1148	 * Allocate a new spa_t structure.
1149	 */
1150	spa = spa_add(pool, altroot);
1151	spa_activate(spa);
1152
1153	spa->spa_uberblock.ub_txg = txg - 1;
1154	spa->spa_uberblock.ub_version = SPA_VERSION;
1155	spa->spa_ubsync = spa->spa_uberblock;
1156
1157	/*
1158	 * Create the root vdev.
1159	 */
1160	spa_config_enter(spa, RW_WRITER, FTAG);
1161
1162	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
1163
1164	ASSERT(error != 0 || rvd != NULL);
1165	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
1166
1167	if (error == 0 && rvd->vdev_children == 0)
1168		error = EINVAL;
1169
1170	if (error == 0 &&
1171	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
1172	    (error = spa_validate_spares(spa, nvroot, txg,
1173	    VDEV_ALLOC_ADD)) == 0) {
1174		for (c = 0; c < rvd->vdev_children; c++)
1175			vdev_init(rvd->vdev_child[c], txg);
1176		vdev_config_dirty(rvd);
1177	}
1178
1179	spa_config_exit(spa, FTAG);
1180
1181	if (error != 0) {
1182		spa_unload(spa);
1183		spa_deactivate(spa);
1184		spa_remove(spa);
1185		mutex_exit(&spa_namespace_lock);
1186		return (error);
1187	}
1188
1189	/*
1190	 * Get the list of spares, if specified.
1191	 */
1192	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1193	    &spares, &nspares) == 0) {
1194		VERIFY(nvlist_alloc(&spa->spa_sparelist, NV_UNIQUE_NAME,
1195		    KM_SLEEP) == 0);
1196		VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
1197		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1198		spa_config_enter(spa, RW_WRITER, FTAG);
1199		spa_load_spares(spa);
1200		spa_config_exit(spa, FTAG);
1201		spa->spa_sync_spares = B_TRUE;
1202	}
1203
1204	spa->spa_dsl_pool = dp = dsl_pool_create(spa, txg);
1205	spa->spa_meta_objset = dp->dp_meta_objset;
1206
1207	tx = dmu_tx_create_assigned(dp, txg);
1208
1209	/*
1210	 * Create the pool config object.
1211	 */
1212	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
1213	    DMU_OT_PACKED_NVLIST, 1 << 14,
1214	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
1215
1216	if (zap_add(spa->spa_meta_objset,
1217	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
1218	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
1219		cmn_err(CE_PANIC, "failed to add pool config");
1220	}
1221
1222	/* Newly created pools are always deflated. */
1223	spa->spa_deflate = TRUE;
1224	if (zap_add(spa->spa_meta_objset,
1225	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
1226	    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
1227		cmn_err(CE_PANIC, "failed to add deflate");
1228	}
1229
1230	/*
1231	 * Create the deferred-free bplist object.  Turn off compression
1232	 * because sync-to-convergence takes longer if the blocksize
1233	 * keeps changing.
1234	 */
1235	spa->spa_sync_bplist_obj = bplist_create(spa->spa_meta_objset,
1236	    1 << 14, tx);
1237	dmu_object_set_compress(spa->spa_meta_objset, spa->spa_sync_bplist_obj,
1238	    ZIO_COMPRESS_OFF, tx);
1239
1240	if (zap_add(spa->spa_meta_objset,
1241	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
1242	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj, tx) != 0) {
1243		cmn_err(CE_PANIC, "failed to add bplist");
1244	}
1245
1246	/*
1247	 * Create the pool's history object.
1248	 */
1249	spa_history_create_obj(spa, tx);
1250
1251	dmu_tx_commit(tx);
1252
1253	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
1254	spa->spa_delegation = zfs_prop_default_numeric(ZPOOL_PROP_DELEGATION);
1255	spa->spa_sync_on = B_TRUE;
1256	txg_sync_start(spa->spa_dsl_pool);
1257
1258	/*
1259	 * We explicitly wait for the first transaction to complete so that our
1260	 * bean counters are appropriately updated.
1261	 */
1262	txg_wait_synced(spa->spa_dsl_pool, txg);
1263
1264	spa_config_sync();
1265
1266	if (history_str != NULL)
1267		(void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
1268
1269	mutex_exit(&spa_namespace_lock);
1270
1271	return (0);
1272}
1273
1274/*
1275 * Import the given pool into the system.  We set up the necessary spa_t and
1276 * then call spa_load() to do the dirty work.
1277 */
1278int
1279spa_import(const char *pool, nvlist_t *config, const char *altroot)
1280{
1281	spa_t *spa;
1282	int error;
1283	nvlist_t *nvroot;
1284	nvlist_t **spares;
1285	uint_t nspares;
1286
1287	/*
1288	 * If a pool with this name exists, return failure.
1289	 */
1290	mutex_enter(&spa_namespace_lock);
1291	if (spa_lookup(pool) != NULL) {
1292		mutex_exit(&spa_namespace_lock);
1293		return (EEXIST);
1294	}
1295
1296	/*
1297	 * Create and initialize the spa structure.
1298	 */
1299	spa = spa_add(pool, altroot);
1300	spa_activate(spa);
1301
1302	/*
1303	 * Pass off the heavy lifting to spa_load().
1304	 * Pass TRUE for mosconfig because the user-supplied config
1305	 * is actually the one to trust when doing an import.
1306	 */
1307	error = spa_load(spa, config, SPA_LOAD_IMPORT, B_TRUE);
1308
1309	spa_config_enter(spa, RW_WRITER, FTAG);
1310	/*
1311	 * Toss any existing sparelist, as it doesn't have any validity anymore,
1312	 * and conflicts with spa_has_spare().
1313	 */
1314	if (spa->spa_sparelist) {
1315		nvlist_free(spa->spa_sparelist);
1316		spa->spa_sparelist = NULL;
1317		spa_load_spares(spa);
1318	}
1319
1320	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1321	    &nvroot) == 0);
1322	if (error == 0)
1323		error = spa_validate_spares(spa, nvroot, -1ULL,
1324		    VDEV_ALLOC_SPARE);
1325	spa_config_exit(spa, FTAG);
1326
1327	if (error != 0) {
1328		spa_unload(spa);
1329		spa_deactivate(spa);
1330		spa_remove(spa);
1331		mutex_exit(&spa_namespace_lock);
1332		return (error);
1333	}
1334
1335	/*
1336	 * Override any spares as specified by the user, as these may have
1337	 * correct device names/devids, etc.
1338	 */
1339	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1340	    &spares, &nspares) == 0) {
1341		if (spa->spa_sparelist)
1342			VERIFY(nvlist_remove(spa->spa_sparelist,
1343			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
1344		else
1345			VERIFY(nvlist_alloc(&spa->spa_sparelist,
1346			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
1347		VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
1348		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1349		spa_config_enter(spa, RW_WRITER, FTAG);
1350		spa_load_spares(spa);
1351		spa_config_exit(spa, FTAG);
1352		spa->spa_sync_spares = B_TRUE;
1353	}
1354
1355	/*
1356	 * Update the config cache to include the newly-imported pool.
1357	 */
1358	if (spa_mode & FWRITE)
1359		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
1360
1361	/*
1362	 * Resilver anything that's out of date.
1363	 */
1364	if (spa_mode & FWRITE)
1365		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
1366
1367	mutex_exit(&spa_namespace_lock);
1368
1369	return (0);
1370}
1371
1372/*
1373 * This (illegal) pool name is used when temporarily importing a spa_t in order
1374 * to get the vdev stats associated with the imported devices.
1375 */
1376#define	TRYIMPORT_NAME	"$import"
1377
1378nvlist_t *
1379spa_tryimport(nvlist_t *tryconfig)
1380{
1381	nvlist_t *config = NULL;
1382	char *poolname;
1383	spa_t *spa;
1384	uint64_t state;
1385
1386	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
1387		return (NULL);
1388
1389	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
1390		return (NULL);
1391
1392	/*
1393	 * Create and initialize the spa structure.
1394	 */
1395	mutex_enter(&spa_namespace_lock);
1396	spa = spa_add(TRYIMPORT_NAME, NULL);
1397	spa_activate(spa);
1398
1399	/*
1400	 * Pass off the heavy lifting to spa_load().
1401	 * Pass TRUE for mosconfig because the user-supplied config
1402	 * is actually the one to trust when doing an import.
1403	 */
1404	(void) spa_load(spa, tryconfig, SPA_LOAD_TRYIMPORT, B_TRUE);
1405
1406	/*
1407	 * If 'tryconfig' was at least parsable, return the current config.
1408	 */
1409	if (spa->spa_root_vdev != NULL) {
1410		spa_config_enter(spa, RW_READER, FTAG);
1411		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
1412		spa_config_exit(spa, FTAG);
1413		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
1414		    poolname) == 0);
1415		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1416		    state) == 0);
1417		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
1418		    spa->spa_uberblock.ub_timestamp) == 0);
1419
1420		/*
1421		 * Add the list of hot spares.
1422		 */
1423		spa_add_spares(spa, config);
1424	}
1425
1426	spa_unload(spa);
1427	spa_deactivate(spa);
1428	spa_remove(spa);
1429	mutex_exit(&spa_namespace_lock);
1430
1431	return (config);
1432}
1433
1434/*
1435 * Pool export/destroy
1436 *
1437 * The act of destroying or exporting a pool is very simple.  We make sure there
1438 * is no more pending I/O and any references to the pool are gone.  Then, we
1439 * update the pool state and sync all the labels to disk, removing the
1440 * configuration from the cache afterwards.
1441 */
1442static int
1443spa_export_common(char *pool, int new_state, nvlist_t **oldconfig)
1444{
1445	spa_t *spa;
1446
1447	if (oldconfig)
1448		*oldconfig = NULL;
1449
1450	if (!(spa_mode & FWRITE))
1451		return (EROFS);
1452
1453	mutex_enter(&spa_namespace_lock);
1454	if ((spa = spa_lookup(pool)) == NULL) {
1455		mutex_exit(&spa_namespace_lock);
1456		return (ENOENT);
1457	}
1458
1459	/*
1460	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
1461	 * reacquire the namespace lock, and see if we can export.
1462	 */
1463	spa_open_ref(spa, FTAG);
1464	mutex_exit(&spa_namespace_lock);
1465	spa_async_suspend(spa);
1466	mutex_enter(&spa_namespace_lock);
1467	spa_close(spa, FTAG);
1468
1469	/*
1470	 * The pool will be in core if it's openable,
1471	 * in which case we can modify its state.
1472	 */
1473	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
1474		/*
1475		 * Objsets may be open only because they're dirty, so we
1476		 * have to force it to sync before checking spa_refcnt.
1477		 */
1478		spa_scrub_suspend(spa);
1479		txg_wait_synced(spa->spa_dsl_pool, 0);
1480
1481		/*
1482		 * A pool cannot be exported or destroyed if there are active
1483		 * references.  If we are resetting a pool, allow references by
1484		 * fault injection handlers.
1485		 */
1486		if (!spa_refcount_zero(spa) ||
1487		    (spa->spa_inject_ref != 0 &&
1488		    new_state != POOL_STATE_UNINITIALIZED)) {
1489			spa_scrub_resume(spa);
1490			spa_async_resume(spa);
1491			mutex_exit(&spa_namespace_lock);
1492			return (EBUSY);
1493		}
1494
1495		spa_scrub_resume(spa);
1496		VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0);
1497
1498		/*
1499		 * We want this to be reflected on every label,
1500		 * so mark them all dirty.  spa_unload() will do the
1501		 * final sync that pushes these changes out.
1502		 */
1503		if (new_state != POOL_STATE_UNINITIALIZED) {
1504			spa_config_enter(spa, RW_WRITER, FTAG);
1505			spa->spa_state = new_state;
1506			spa->spa_final_txg = spa_last_synced_txg(spa) + 1;
1507			vdev_config_dirty(spa->spa_root_vdev);
1508			spa_config_exit(spa, FTAG);
1509		}
1510	}
1511
1512	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
1513
1514	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
1515		spa_unload(spa);
1516		spa_deactivate(spa);
1517	}
1518
1519	if (oldconfig && spa->spa_config)
1520		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
1521
1522	if (new_state != POOL_STATE_UNINITIALIZED) {
1523		spa_remove(spa);
1524		spa_config_sync();
1525	}
1526	mutex_exit(&spa_namespace_lock);
1527
1528	return (0);
1529}
1530
1531/*
1532 * Destroy a storage pool.
1533 */
1534int
1535spa_destroy(char *pool)
1536{
1537	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL));
1538}
1539
1540/*
1541 * Export a storage pool.
1542 */
1543int
1544spa_export(char *pool, nvlist_t **oldconfig)
1545{
1546	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig));
1547}
1548
1549/*
1550 * Similar to spa_export(), this unloads the spa_t without actually removing it
1551 * from the namespace in any way.
1552 */
1553int
1554spa_reset(char *pool)
1555{
1556	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL));
1557}
1558
1559
1560/*
1561 * ==========================================================================
1562 * Device manipulation
1563 * ==========================================================================
1564 */
1565
1566/*
1567 * Add a device to a storage pool.
1568 */
1569int
1570spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
1571{
1572	uint64_t txg;
1573	int c, error;
1574	vdev_t *rvd = spa->spa_root_vdev;
1575	vdev_t *vd, *tvd;
1576	nvlist_t **spares;
1577	uint_t i, nspares;
1578
1579	txg = spa_vdev_enter(spa);
1580
1581	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
1582	    VDEV_ALLOC_ADD)) != 0)
1583		return (spa_vdev_exit(spa, NULL, txg, error));
1584
1585	spa->spa_pending_vdev = vd;
1586
1587	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1588	    &spares, &nspares) != 0)
1589		nspares = 0;
1590
1591	if (vd->vdev_children == 0 && nspares == 0) {
1592		spa->spa_pending_vdev = NULL;
1593		return (spa_vdev_exit(spa, vd, txg, EINVAL));
1594	}
1595
1596	if (vd->vdev_children != 0) {
1597		if ((error = vdev_create(vd, txg, B_FALSE)) != 0) {
1598			spa->spa_pending_vdev = NULL;
1599			return (spa_vdev_exit(spa, vd, txg, error));
1600		}
1601	}
1602
1603	/*
1604	 * We must validate the spares after checking the children.  Otherwise,
1605	 * vdev_inuse() will blindly overwrite the spare.
1606	 */
1607	if ((error = spa_validate_spares(spa, nvroot, txg,
1608	    VDEV_ALLOC_ADD)) != 0) {
1609		spa->spa_pending_vdev = NULL;
1610		return (spa_vdev_exit(spa, vd, txg, error));
1611	}
1612
1613	spa->spa_pending_vdev = NULL;
1614
1615	/*
1616	 * Transfer each new top-level vdev from vd to rvd.
1617	 */
1618	for (c = 0; c < vd->vdev_children; c++) {
1619		tvd = vd->vdev_child[c];
1620		vdev_remove_child(vd, tvd);
1621		tvd->vdev_id = rvd->vdev_children;
1622		vdev_add_child(rvd, tvd);
1623		vdev_config_dirty(tvd);
1624	}
1625
1626	if (nspares != 0) {
1627		if (spa->spa_sparelist != NULL) {
1628			nvlist_t **oldspares;
1629			uint_t oldnspares;
1630			nvlist_t **newspares;
1631
1632			VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist,
1633			    ZPOOL_CONFIG_SPARES, &oldspares, &oldnspares) == 0);
1634
1635			newspares = kmem_alloc(sizeof (void *) *
1636			    (nspares + oldnspares), KM_SLEEP);
1637			for (i = 0; i < oldnspares; i++)
1638				VERIFY(nvlist_dup(oldspares[i],
1639				    &newspares[i], KM_SLEEP) == 0);
1640			for (i = 0; i < nspares; i++)
1641				VERIFY(nvlist_dup(spares[i],
1642				    &newspares[i + oldnspares],
1643				    KM_SLEEP) == 0);
1644
1645			VERIFY(nvlist_remove(spa->spa_sparelist,
1646			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
1647
1648			VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
1649			    ZPOOL_CONFIG_SPARES, newspares,
1650			    nspares + oldnspares) == 0);
1651			for (i = 0; i < oldnspares + nspares; i++)
1652				nvlist_free(newspares[i]);
1653			kmem_free(newspares, (oldnspares + nspares) *
1654			    sizeof (void *));
1655		} else {
1656			VERIFY(nvlist_alloc(&spa->spa_sparelist,
1657			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
1658			VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
1659			    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1660		}
1661
1662		spa_load_spares(spa);
1663		spa->spa_sync_spares = B_TRUE;
1664	}
1665
1666	/*
1667	 * We have to be careful when adding new vdevs to an existing pool.
1668	 * If other threads start allocating from these vdevs before we
1669	 * sync the config cache, and we lose power, then upon reboot we may
1670	 * fail to open the pool because there are DVAs that the config cache
1671	 * can't translate.  Therefore, we first add the vdevs without
1672	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
1673	 * and then let spa_config_update() initialize the new metaslabs.
1674	 *
1675	 * spa_load() checks for added-but-not-initialized vdevs, so that
1676	 * if we lose power at any point in this sequence, the remaining
1677	 * steps will be completed the next time we load the pool.
1678	 */
1679	(void) spa_vdev_exit(spa, vd, txg, 0);
1680
1681	mutex_enter(&spa_namespace_lock);
1682	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
1683	mutex_exit(&spa_namespace_lock);
1684
1685	return (0);
1686}
1687
1688/*
1689 * Attach a device to a mirror.  The arguments are the path to any device
1690 * in the mirror, and the nvroot for the new device.  If the path specifies
1691 * a device that is not mirrored, we automatically insert the mirror vdev.
1692 *
1693 * If 'replacing' is specified, the new device is intended to replace the
1694 * existing device; in this case the two devices are made into their own
1695 * mirror using the 'replacing' vdev, which is functionally identical to
1696 * the mirror vdev (it actually reuses all the same ops) but has a few
1697 * extra rules: you can't attach to it after it's been created, and upon
1698 * completion of resilvering, the first disk (the one being replaced)
1699 * is automatically detached.
1700 */
1701int
1702spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
1703{
1704	uint64_t txg, open_txg;
1705	int error;
1706	vdev_t *rvd = spa->spa_root_vdev;
1707	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
1708	vdev_ops_t *pvops;
1709	int is_log;
1710
1711	txg = spa_vdev_enter(spa);
1712
1713	oldvd = vdev_lookup_by_guid(rvd, guid);
1714
1715	if (oldvd == NULL)
1716		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
1717
1718	if (!oldvd->vdev_ops->vdev_op_leaf)
1719		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1720
1721	pvd = oldvd->vdev_parent;
1722
1723	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
1724	    VDEV_ALLOC_ADD)) != 0)
1725		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
1726
1727	if (newrootvd->vdev_children != 1)
1728		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
1729
1730	newvd = newrootvd->vdev_child[0];
1731
1732	if (!newvd->vdev_ops->vdev_op_leaf)
1733		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
1734
1735	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
1736		return (spa_vdev_exit(spa, newrootvd, txg, error));
1737
1738	/*
1739	 * Spares can't replace logs
1740	 */
1741	is_log = oldvd->vdev_islog;
1742	if (is_log && newvd->vdev_isspare)
1743		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
1744
1745	if (!replacing) {
1746		/*
1747		 * For attach, the only allowable parent is a mirror or the root
1748		 * vdev.
1749		 */
1750		if (pvd->vdev_ops != &vdev_mirror_ops &&
1751		    pvd->vdev_ops != &vdev_root_ops)
1752			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
1753
1754		pvops = &vdev_mirror_ops;
1755	} else {
1756		/*
1757		 * Active hot spares can only be replaced by inactive hot
1758		 * spares.
1759		 */
1760		if (pvd->vdev_ops == &vdev_spare_ops &&
1761		    pvd->vdev_child[1] == oldvd &&
1762		    !spa_has_spare(spa, newvd->vdev_guid))
1763			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
1764
1765		/*
1766		 * If the source is a hot spare, and the parent isn't already a
1767		 * spare, then we want to create a new hot spare.  Otherwise, we
1768		 * want to create a replacing vdev.  The user is not allowed to
1769		 * attach to a spared vdev child unless the 'isspare' state is
1770		 * the same (spare replaces spare, non-spare replaces
1771		 * non-spare).
1772		 */
1773		if (pvd->vdev_ops == &vdev_replacing_ops)
1774			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
1775		else if (pvd->vdev_ops == &vdev_spare_ops &&
1776		    newvd->vdev_isspare != oldvd->vdev_isspare)
1777			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
1778		else if (pvd->vdev_ops != &vdev_spare_ops &&
1779		    newvd->vdev_isspare)
1780			pvops = &vdev_spare_ops;
1781		else
1782			pvops = &vdev_replacing_ops;
1783	}
1784
1785	/*
1786	 * Compare the new device size with the replaceable/attachable
1787	 * device size.
1788	 */
1789	if (newvd->vdev_psize < vdev_get_rsize(oldvd))
1790		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
1791
1792	/*
1793	 * The new device cannot have a higher alignment requirement
1794	 * than the top-level vdev.
1795	 */
1796	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
1797		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
1798
1799	/*
1800	 * If this is an in-place replacement, update oldvd's path and devid
1801	 * to make it distinguishable from newvd, and unopenable from now on.
1802	 */
1803	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
1804		spa_strfree(oldvd->vdev_path);
1805		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
1806		    KM_SLEEP);
1807		(void) sprintf(oldvd->vdev_path, "%s/%s",
1808		    newvd->vdev_path, "old");
1809		if (oldvd->vdev_devid != NULL) {
1810			spa_strfree(oldvd->vdev_devid);
1811			oldvd->vdev_devid = NULL;
1812		}
1813	}
1814
1815	/*
1816	 * If the parent is not a mirror, or if we're replacing, insert the new
1817	 * mirror/replacing/spare vdev above oldvd.
1818	 */
1819	if (pvd->vdev_ops != pvops)
1820		pvd = vdev_add_parent(oldvd, pvops);
1821
1822	ASSERT(pvd->vdev_top->vdev_parent == rvd);
1823	ASSERT(pvd->vdev_ops == pvops);
1824	ASSERT(oldvd->vdev_parent == pvd);
1825
1826	/*
1827	 * Extract the new device from its root and add it to pvd.
1828	 */
1829	vdev_remove_child(newrootvd, newvd);
1830	newvd->vdev_id = pvd->vdev_children;
1831	vdev_add_child(pvd, newvd);
1832
1833	/*
1834	 * If newvd is smaller than oldvd, but larger than its rsize,
1835	 * the addition of newvd may have decreased our parent's asize.
1836	 */
1837	pvd->vdev_asize = MIN(pvd->vdev_asize, newvd->vdev_asize);
1838
1839	tvd = newvd->vdev_top;
1840	ASSERT(pvd->vdev_top == tvd);
1841	ASSERT(tvd->vdev_parent == rvd);
1842
1843	vdev_config_dirty(tvd);
1844
1845	/*
1846	 * Set newvd's DTL to [TXG_INITIAL, open_txg].  It will propagate
1847	 * upward when spa_vdev_exit() calls vdev_dtl_reassess().
1848	 */
1849	open_txg = txg + TXG_CONCURRENT_STATES - 1;
1850
1851	mutex_enter(&newvd->vdev_dtl_lock);
1852	space_map_add(&newvd->vdev_dtl_map, TXG_INITIAL,
1853	    open_txg - TXG_INITIAL + 1);
1854	mutex_exit(&newvd->vdev_dtl_lock);
1855
1856	if (newvd->vdev_isspare)
1857		spa_spare_activate(newvd);
1858
1859	/*
1860	 * Mark newvd's DTL dirty in this txg.
1861	 */
1862	vdev_dirty(tvd, VDD_DTL, newvd, txg);
1863
1864	(void) spa_vdev_exit(spa, newrootvd, open_txg, 0);
1865
1866	/*
1867	 * Kick off a resilver to update newvd.  We need to grab the namespace
1868	 * lock because spa_scrub() needs to post a sysevent with the pool name.
1869	 */
1870	mutex_enter(&spa_namespace_lock);
1871	VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
1872	mutex_exit(&spa_namespace_lock);
1873
1874	return (0);
1875}
1876
1877/*
1878 * Detach a device from a mirror or replacing vdev.
1879 * If 'replace_done' is specified, only detach if the parent
1880 * is a replacing vdev.
1881 */
1882int
1883spa_vdev_detach(spa_t *spa, uint64_t guid, int replace_done)
1884{
1885	uint64_t txg;
1886	int c, t, error;
1887	vdev_t *rvd = spa->spa_root_vdev;
1888	vdev_t *vd, *pvd, *cvd, *tvd;
1889	boolean_t unspare = B_FALSE;
1890	uint64_t unspare_guid;
1891
1892	txg = spa_vdev_enter(spa);
1893
1894	vd = vdev_lookup_by_guid(rvd, guid);
1895
1896	if (vd == NULL)
1897		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
1898
1899	if (!vd->vdev_ops->vdev_op_leaf)
1900		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1901
1902	pvd = vd->vdev_parent;
1903
1904	/*
1905	 * If replace_done is specified, only remove this device if it's
1906	 * the first child of a replacing vdev.  For the 'spare' vdev, either
1907	 * disk can be removed.
1908	 */
1909	if (replace_done) {
1910		if (pvd->vdev_ops == &vdev_replacing_ops) {
1911			if (vd->vdev_id != 0)
1912				return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1913		} else if (pvd->vdev_ops != &vdev_spare_ops) {
1914			return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1915		}
1916	}
1917
1918	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
1919	    spa_version(spa) >= SPA_VERSION_SPARES);
1920
1921	/*
1922	 * Only mirror, replacing, and spare vdevs support detach.
1923	 */
1924	if (pvd->vdev_ops != &vdev_replacing_ops &&
1925	    pvd->vdev_ops != &vdev_mirror_ops &&
1926	    pvd->vdev_ops != &vdev_spare_ops)
1927		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
1928
1929	/*
1930	 * If there's only one replica, you can't detach it.
1931	 */
1932	if (pvd->vdev_children <= 1)
1933		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
1934
1935	/*
1936	 * If all siblings have non-empty DTLs, this device may have the only
1937	 * valid copy of the data, which means we cannot safely detach it.
1938	 *
1939	 * XXX -- as in the vdev_offline() case, we really want a more
1940	 * precise DTL check.
1941	 */
1942	for (c = 0; c < pvd->vdev_children; c++) {
1943		uint64_t dirty;
1944
1945		cvd = pvd->vdev_child[c];
1946		if (cvd == vd)
1947			continue;
1948		if (vdev_is_dead(cvd))
1949			continue;
1950		mutex_enter(&cvd->vdev_dtl_lock);
1951		dirty = cvd->vdev_dtl_map.sm_space |
1952		    cvd->vdev_dtl_scrub.sm_space;
1953		mutex_exit(&cvd->vdev_dtl_lock);
1954		if (!dirty)
1955			break;
1956	}
1957
1958	/*
1959	 * If we are a replacing or spare vdev, then we can always detach the
1960	 * latter child, as that is how one cancels the operation.
1961	 */
1962	if ((pvd->vdev_ops == &vdev_mirror_ops || vd->vdev_id != 1) &&
1963	    c == pvd->vdev_children)
1964		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
1965
1966	/*
1967	 * If we are detaching the original disk from a spare, then it implies
1968	 * that the spare should become a real disk, and be removed from the
1969	 * active spare list for the pool.
1970	 */
1971	if (pvd->vdev_ops == &vdev_spare_ops &&
1972	    vd->vdev_id == 0)
1973		unspare = B_TRUE;
1974
1975	/*
1976	 * Erase the disk labels so the disk can be used for other things.
1977	 * This must be done after all other error cases are handled,
1978	 * but before we disembowel vd (so we can still do I/O to it).
1979	 * But if we can't do it, don't treat the error as fatal --
1980	 * it may be that the unwritability of the disk is the reason
1981	 * it's being detached!
1982	 */
1983	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
1984
1985	/*
1986	 * Remove vd from its parent and compact the parent's children.
1987	 */
1988	vdev_remove_child(pvd, vd);
1989	vdev_compact_children(pvd);
1990
1991	/*
1992	 * Remember one of the remaining children so we can get tvd below.
1993	 */
1994	cvd = pvd->vdev_child[0];
1995
1996	/*
1997	 * If we need to remove the remaining child from the list of hot spares,
1998	 * do it now, marking the vdev as no longer a spare in the process.  We
1999	 * must do this before vdev_remove_parent(), because that can change the
2000	 * GUID if it creates a new toplevel GUID.
2001	 */
2002	if (unspare) {
2003		ASSERT(cvd->vdev_isspare);
2004		spa_spare_remove(cvd);
2005		unspare_guid = cvd->vdev_guid;
2006	}
2007
2008	/*
2009	 * If the parent mirror/replacing vdev only has one child,
2010	 * the parent is no longer needed.  Remove it from the tree.
2011	 */
2012	if (pvd->vdev_children == 1)
2013		vdev_remove_parent(cvd);
2014
2015	/*
2016	 * We don't set tvd until now because the parent we just removed
2017	 * may have been the previous top-level vdev.
2018	 */
2019	tvd = cvd->vdev_top;
2020	ASSERT(tvd->vdev_parent == rvd);
2021
2022	/*
2023	 * Reevaluate the parent vdev state.
2024	 */
2025	vdev_propagate_state(cvd);
2026
2027	/*
2028	 * If the device we just detached was smaller than the others, it may be
2029	 * possible to add metaslabs (i.e. grow the pool).  vdev_metaslab_init()
2030	 * can't fail because the existing metaslabs are already in core, so
2031	 * there's nothing to read from disk.
2032	 */
2033	VERIFY(vdev_metaslab_init(tvd, txg) == 0);
2034
2035	vdev_config_dirty(tvd);
2036
2037	/*
2038	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
2039	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
2040	 * But first make sure we're not on any *other* txg's DTL list, to
2041	 * prevent vd from being accessed after it's freed.
2042	 */
2043	for (t = 0; t < TXG_SIZE; t++)
2044		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
2045	vd->vdev_detached = B_TRUE;
2046	vdev_dirty(tvd, VDD_DTL, vd, txg);
2047
2048	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
2049
2050	error = spa_vdev_exit(spa, vd, txg, 0);
2051
2052	/*
2053	 * If this was the removal of the original device in a hot spare vdev,
2054	 * then we want to go through and remove the device from the hot spare
2055	 * list of every other pool.
2056	 */
2057	if (unspare) {
2058		spa = NULL;
2059		mutex_enter(&spa_namespace_lock);
2060		while ((spa = spa_next(spa)) != NULL) {
2061			if (spa->spa_state != POOL_STATE_ACTIVE)
2062				continue;
2063
2064			(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
2065		}
2066		mutex_exit(&spa_namespace_lock);
2067	}
2068
2069	return (error);
2070}
2071
2072/*
2073 * Remove a device from the pool.  Currently, this supports removing only hot
2074 * spares.
2075 */
2076int
2077spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
2078{
2079	vdev_t *vd;
2080	nvlist_t **spares, *nv, **newspares;
2081	uint_t i, j, nspares;
2082	int ret = 0;
2083
2084	spa_config_enter(spa, RW_WRITER, FTAG);
2085
2086	vd = spa_lookup_by_guid(spa, guid);
2087
2088	nv = NULL;
2089	if (spa->spa_spares != NULL &&
2090	    nvlist_lookup_nvlist_array(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
2091	    &spares, &nspares) == 0) {
2092		for (i = 0; i < nspares; i++) {
2093			uint64_t theguid;
2094
2095			VERIFY(nvlist_lookup_uint64(spares[i],
2096			    ZPOOL_CONFIG_GUID, &theguid) == 0);
2097			if (theguid == guid) {
2098				nv = spares[i];
2099				break;
2100			}
2101		}
2102	}
2103
2104	/*
2105	 * We only support removing a hot spare, and only if it's not currently
2106	 * in use in this pool.
2107	 */
2108	if (nv == NULL && vd == NULL) {
2109		ret = ENOENT;
2110		goto out;
2111	}
2112
2113	if (nv == NULL && vd != NULL) {
2114		ret = ENOTSUP;
2115		goto out;
2116	}
2117
2118	if (!unspare && nv != NULL && vd != NULL) {
2119		ret = EBUSY;
2120		goto out;
2121	}
2122
2123	if (nspares == 1) {
2124		newspares = NULL;
2125	} else {
2126		newspares = kmem_alloc((nspares - 1) * sizeof (void *),
2127		    KM_SLEEP);
2128		for (i = 0, j = 0; i < nspares; i++) {
2129			if (spares[i] != nv)
2130				VERIFY(nvlist_dup(spares[i],
2131				    &newspares[j++], KM_SLEEP) == 0);
2132		}
2133	}
2134
2135	VERIFY(nvlist_remove(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
2136	    DATA_TYPE_NVLIST_ARRAY) == 0);
2137	VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
2138	    newspares, nspares - 1) == 0);
2139	for (i = 0; i < nspares - 1; i++)
2140		nvlist_free(newspares[i]);
2141	kmem_free(newspares, (nspares - 1) * sizeof (void *));
2142	spa_load_spares(spa);
2143	spa->spa_sync_spares = B_TRUE;
2144
2145out:
2146	spa_config_exit(spa, FTAG);
2147
2148	return (ret);
2149}
2150
2151/*
2152 * Find any device that's done replacing, or a vdev marked 'unspare' that's
2153 * current spared, so we can detach it.
2154 */
2155static vdev_t *
2156spa_vdev_resilver_done_hunt(vdev_t *vd)
2157{
2158	vdev_t *newvd, *oldvd;
2159	int c;
2160
2161	for (c = 0; c < vd->vdev_children; c++) {
2162		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
2163		if (oldvd != NULL)
2164			return (oldvd);
2165	}
2166
2167	/*
2168	 * Check for a completed replacement.
2169	 */
2170	if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
2171		oldvd = vd->vdev_child[0];
2172		newvd = vd->vdev_child[1];
2173
2174		mutex_enter(&newvd->vdev_dtl_lock);
2175		if (newvd->vdev_dtl_map.sm_space == 0 &&
2176		    newvd->vdev_dtl_scrub.sm_space == 0) {
2177			mutex_exit(&newvd->vdev_dtl_lock);
2178			return (oldvd);
2179		}
2180		mutex_exit(&newvd->vdev_dtl_lock);
2181	}
2182
2183	/*
2184	 * Check for a completed resilver with the 'unspare' flag set.
2185	 */
2186	if (vd->vdev_ops == &vdev_spare_ops && vd->vdev_children == 2) {
2187		newvd = vd->vdev_child[0];
2188		oldvd = vd->vdev_child[1];
2189
2190		mutex_enter(&newvd->vdev_dtl_lock);
2191		if (newvd->vdev_unspare &&
2192		    newvd->vdev_dtl_map.sm_space == 0 &&
2193		    newvd->vdev_dtl_scrub.sm_space == 0) {
2194			newvd->vdev_unspare = 0;
2195			mutex_exit(&newvd->vdev_dtl_lock);
2196			return (oldvd);
2197		}
2198		mutex_exit(&newvd->vdev_dtl_lock);
2199	}
2200
2201	return (NULL);
2202}
2203
2204static void
2205spa_vdev_resilver_done(spa_t *spa)
2206{
2207	vdev_t *vd;
2208	vdev_t *pvd;
2209	uint64_t guid;
2210	uint64_t pguid = 0;
2211
2212	spa_config_enter(spa, RW_READER, FTAG);
2213
2214	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
2215		guid = vd->vdev_guid;
2216		/*
2217		 * If we have just finished replacing a hot spared device, then
2218		 * we need to detach the parent's first child (the original hot
2219		 * spare) as well.
2220		 */
2221		pvd = vd->vdev_parent;
2222		if (pvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2223		    pvd->vdev_id == 0) {
2224			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
2225			ASSERT(pvd->vdev_parent->vdev_children == 2);
2226			pguid = pvd->vdev_parent->vdev_child[1]->vdev_guid;
2227		}
2228		spa_config_exit(spa, FTAG);
2229		if (spa_vdev_detach(spa, guid, B_TRUE) != 0)
2230			return;
2231		if (pguid != 0 && spa_vdev_detach(spa, pguid, B_TRUE) != 0)
2232			return;
2233		spa_config_enter(spa, RW_READER, FTAG);
2234	}
2235
2236	spa_config_exit(spa, FTAG);
2237}
2238
2239/*
2240 * Update the stored path for this vdev.  Dirty the vdev configuration, relying
2241 * on spa_vdev_enter/exit() to synchronize the labels and cache.
2242 */
2243int
2244spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
2245{
2246	vdev_t *rvd, *vd;
2247	uint64_t txg;
2248
2249	rvd = spa->spa_root_vdev;
2250
2251	txg = spa_vdev_enter(spa);
2252
2253	if ((vd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
2254		/*
2255		 * Determine if this is a reference to a hot spare.  In that
2256		 * case, update the path as stored in the spare list.
2257		 */
2258		nvlist_t **spares;
2259		uint_t i, nspares;
2260		if (spa->spa_sparelist != NULL) {
2261			VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist,
2262			    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2263			for (i = 0; i < nspares; i++) {
2264				uint64_t theguid;
2265				VERIFY(nvlist_lookup_uint64(spares[i],
2266				    ZPOOL_CONFIG_GUID, &theguid) == 0);
2267				if (theguid == guid)
2268					break;
2269			}
2270
2271			if (i == nspares)
2272				return (spa_vdev_exit(spa, NULL, txg, ENOENT));
2273
2274			VERIFY(nvlist_add_string(spares[i],
2275			    ZPOOL_CONFIG_PATH, newpath) == 0);
2276			spa_load_spares(spa);
2277			spa->spa_sync_spares = B_TRUE;
2278			return (spa_vdev_exit(spa, NULL, txg, 0));
2279		} else {
2280			return (spa_vdev_exit(spa, NULL, txg, ENOENT));
2281		}
2282	}
2283
2284	if (!vd->vdev_ops->vdev_op_leaf)
2285		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
2286
2287	spa_strfree(vd->vdev_path);
2288	vd->vdev_path = spa_strdup(newpath);
2289
2290	vdev_config_dirty(vd->vdev_top);
2291
2292	return (spa_vdev_exit(spa, NULL, txg, 0));
2293}
2294
2295/*
2296 * ==========================================================================
2297 * SPA Scrubbing
2298 * ==========================================================================
2299 */
2300
2301static void
2302spa_scrub_io_done(zio_t *zio)
2303{
2304	spa_t *spa = zio->io_spa;
2305
2306	arc_data_buf_free(zio->io_data, zio->io_size);
2307
2308	mutex_enter(&spa->spa_scrub_lock);
2309	if (zio->io_error && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
2310		vdev_t *vd = zio->io_vd ? zio->io_vd : spa->spa_root_vdev;
2311		spa->spa_scrub_errors++;
2312		mutex_enter(&vd->vdev_stat_lock);
2313		vd->vdev_stat.vs_scrub_errors++;
2314		mutex_exit(&vd->vdev_stat_lock);
2315	}
2316
2317	if (--spa->spa_scrub_inflight < spa->spa_scrub_maxinflight)
2318		cv_broadcast(&spa->spa_scrub_io_cv);
2319
2320	ASSERT(spa->spa_scrub_inflight >= 0);
2321
2322	mutex_exit(&spa->spa_scrub_lock);
2323}
2324
2325static void
2326spa_scrub_io_start(spa_t *spa, blkptr_t *bp, int priority, int flags,
2327    zbookmark_t *zb)
2328{
2329	size_t size = BP_GET_LSIZE(bp);
2330	void *data;
2331
2332	mutex_enter(&spa->spa_scrub_lock);
2333	/*
2334	 * Do not give too much work to vdev(s).
2335	 */
2336	while (spa->spa_scrub_inflight >= spa->spa_scrub_maxinflight) {
2337		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2338	}
2339	spa->spa_scrub_inflight++;
2340	mutex_exit(&spa->spa_scrub_lock);
2341
2342	data = arc_data_buf_alloc(size);
2343
2344	if (zb->zb_level == -1 && BP_GET_TYPE(bp) != DMU_OT_OBJSET)
2345		flags |= ZIO_FLAG_SPECULATIVE;	/* intent log block */
2346
2347	flags |= ZIO_FLAG_SCRUB_THREAD | ZIO_FLAG_CANFAIL;
2348
2349	zio_nowait(zio_read(NULL, spa, bp, data, size,
2350	    spa_scrub_io_done, NULL, priority, flags, zb));
2351}
2352
2353/* ARGSUSED */
2354static int
2355spa_scrub_cb(traverse_blk_cache_t *bc, spa_t *spa, void *a)
2356{
2357	blkptr_t *bp = &bc->bc_blkptr;
2358	vdev_t *vd = spa->spa_root_vdev;
2359	dva_t *dva = bp->blk_dva;
2360	int needs_resilver = B_FALSE;
2361	int d;
2362
2363	if (bc->bc_errno) {
2364		/*
2365		 * We can't scrub this block, but we can continue to scrub
2366		 * the rest of the pool.  Note the error and move along.
2367		 */
2368		mutex_enter(&spa->spa_scrub_lock);
2369		spa->spa_scrub_errors++;
2370		mutex_exit(&spa->spa_scrub_lock);
2371
2372		mutex_enter(&vd->vdev_stat_lock);
2373		vd->vdev_stat.vs_scrub_errors++;
2374		mutex_exit(&vd->vdev_stat_lock);
2375
2376		return (ERESTART);
2377	}
2378
2379	ASSERT(bp->blk_birth < spa->spa_scrub_maxtxg);
2380
2381	for (d = 0; d < BP_GET_NDVAS(bp); d++) {
2382		vd = vdev_lookup_top(spa, DVA_GET_VDEV(&dva[d]));
2383
2384		ASSERT(vd != NULL);
2385
2386		/*
2387		 * Keep track of how much data we've examined so that
2388		 * zpool(1M) status can make useful progress reports.
2389		 */
2390		mutex_enter(&vd->vdev_stat_lock);
2391		vd->vdev_stat.vs_scrub_examined += DVA_GET_ASIZE(&dva[d]);
2392		mutex_exit(&vd->vdev_stat_lock);
2393
2394		if (spa->spa_scrub_type == POOL_SCRUB_RESILVER) {
2395			if (DVA_GET_GANG(&dva[d])) {
2396				/*
2397				 * Gang members may be spread across multiple
2398				 * vdevs, so the best we can do is look at the
2399				 * pool-wide DTL.
2400				 * XXX -- it would be better to change our
2401				 * allocation policy to ensure that this can't
2402				 * happen.
2403				 */
2404				vd = spa->spa_root_vdev;
2405			}
2406			if (vdev_dtl_contains(&vd->vdev_dtl_map,
2407			    bp->blk_birth, 1))
2408				needs_resilver = B_TRUE;
2409		}
2410	}
2411
2412	if (spa->spa_scrub_type == POOL_SCRUB_EVERYTHING)
2413		spa_scrub_io_start(spa, bp, ZIO_PRIORITY_SCRUB,
2414		    ZIO_FLAG_SCRUB, &bc->bc_bookmark);
2415	else if (needs_resilver)
2416		spa_scrub_io_start(spa, bp, ZIO_PRIORITY_RESILVER,
2417		    ZIO_FLAG_RESILVER, &bc->bc_bookmark);
2418
2419	return (0);
2420}
2421
2422static void
2423spa_scrub_thread(spa_t *spa)
2424{
2425	callb_cpr_t cprinfo;
2426	traverse_handle_t *th = spa->spa_scrub_th;
2427	vdev_t *rvd = spa->spa_root_vdev;
2428	pool_scrub_type_t scrub_type = spa->spa_scrub_type;
2429	int error = 0;
2430	boolean_t complete;
2431
2432	CALLB_CPR_INIT(&cprinfo, &spa->spa_scrub_lock, callb_generic_cpr, FTAG);
2433
2434	/*
2435	 * If we're restarting due to a snapshot create/delete,
2436	 * wait for that to complete.
2437	 */
2438	txg_wait_synced(spa_get_dsl(spa), 0);
2439
2440	dprintf("start %s mintxg=%llu maxtxg=%llu\n",
2441	    scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub",
2442	    spa->spa_scrub_mintxg, spa->spa_scrub_maxtxg);
2443
2444	spa_config_enter(spa, RW_WRITER, FTAG);
2445	vdev_reopen(rvd);		/* purge all vdev caches */
2446	vdev_config_dirty(rvd);		/* rewrite all disk labels */
2447	vdev_scrub_stat_update(rvd, scrub_type, B_FALSE);
2448	spa_config_exit(spa, FTAG);
2449
2450	mutex_enter(&spa->spa_scrub_lock);
2451	spa->spa_scrub_errors = 0;
2452	spa->spa_scrub_active = 1;
2453	ASSERT(spa->spa_scrub_inflight == 0);
2454
2455	while (!spa->spa_scrub_stop) {
2456		CALLB_CPR_SAFE_BEGIN(&cprinfo);
2457		while (spa->spa_scrub_suspended) {
2458			spa->spa_scrub_active = 0;
2459			cv_broadcast(&spa->spa_scrub_cv);
2460			cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
2461			spa->spa_scrub_active = 1;
2462		}
2463		CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_scrub_lock);
2464
2465		if (spa->spa_scrub_restart_txg != 0)
2466			break;
2467
2468		mutex_exit(&spa->spa_scrub_lock);
2469		error = traverse_more(th);
2470		mutex_enter(&spa->spa_scrub_lock);
2471		if (error != EAGAIN)
2472			break;
2473	}
2474
2475	while (spa->spa_scrub_inflight)
2476		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2477
2478	spa->spa_scrub_active = 0;
2479	cv_broadcast(&spa->spa_scrub_cv);
2480
2481	mutex_exit(&spa->spa_scrub_lock);
2482
2483	spa_config_enter(spa, RW_WRITER, FTAG);
2484
2485	mutex_enter(&spa->spa_scrub_lock);
2486
2487	/*
2488	 * Note: we check spa_scrub_restart_txg under both spa_scrub_lock
2489	 * AND the spa config lock to synchronize with any config changes
2490	 * that revise the DTLs under spa_vdev_enter() / spa_vdev_exit().
2491	 */
2492	if (spa->spa_scrub_restart_txg != 0)
2493		error = ERESTART;
2494
2495	if (spa->spa_scrub_stop)
2496		error = EINTR;
2497
2498	/*
2499	 * Even if there were uncorrectable errors, we consider the scrub
2500	 * completed.  The downside is that if there is a transient error during
2501	 * a resilver, we won't resilver the data properly to the target.  But
2502	 * if the damage is permanent (more likely) we will resilver forever,
2503	 * which isn't really acceptable.  Since there is enough information for
2504	 * the user to know what has failed and why, this seems like a more
2505	 * tractable approach.
2506	 */
2507	complete = (error == 0);
2508
2509	dprintf("end %s to maxtxg=%llu %s, traverse=%d, %llu errors, stop=%u\n",
2510	    scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub",
2511	    spa->spa_scrub_maxtxg, complete ? "done" : "FAILED",
2512	    error, spa->spa_scrub_errors, spa->spa_scrub_stop);
2513
2514	mutex_exit(&spa->spa_scrub_lock);
2515
2516	/*
2517	 * If the scrub/resilver completed, update all DTLs to reflect this.
2518	 * Whether it succeeded or not, vacate all temporary scrub DTLs.
2519	 */
2520	vdev_dtl_reassess(rvd, spa_last_synced_txg(spa) + 1,
2521	    complete ? spa->spa_scrub_maxtxg : 0, B_TRUE);
2522	vdev_scrub_stat_update(rvd, POOL_SCRUB_NONE, complete);
2523	spa_errlog_rotate(spa);
2524
2525	if (scrub_type == POOL_SCRUB_RESILVER && complete)
2526		spa_event_notify(spa, NULL, ESC_ZFS_RESILVER_FINISH);
2527
2528	spa_config_exit(spa, FTAG);
2529
2530	mutex_enter(&spa->spa_scrub_lock);
2531
2532	/*
2533	 * We may have finished replacing a device.
2534	 * Let the async thread assess this and handle the detach.
2535	 */
2536	spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
2537
2538	/*
2539	 * If we were told to restart, our final act is to start a new scrub.
2540	 */
2541	if (error == ERESTART)
2542		spa_async_request(spa, scrub_type == POOL_SCRUB_RESILVER ?
2543		    SPA_ASYNC_RESILVER : SPA_ASYNC_SCRUB);
2544
2545	spa->spa_scrub_type = POOL_SCRUB_NONE;
2546	spa->spa_scrub_active = 0;
2547	spa->spa_scrub_thread = NULL;
2548	cv_broadcast(&spa->spa_scrub_cv);
2549	CALLB_CPR_EXIT(&cprinfo);	/* drops &spa->spa_scrub_lock */
2550	thread_exit();
2551}
2552
2553void
2554spa_scrub_suspend(spa_t *spa)
2555{
2556	mutex_enter(&spa->spa_scrub_lock);
2557	spa->spa_scrub_suspended++;
2558	while (spa->spa_scrub_active) {
2559		cv_broadcast(&spa->spa_scrub_cv);
2560		cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
2561	}
2562	while (spa->spa_scrub_inflight)
2563		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2564	mutex_exit(&spa->spa_scrub_lock);
2565}
2566
2567void
2568spa_scrub_resume(spa_t *spa)
2569{
2570	mutex_enter(&spa->spa_scrub_lock);
2571	ASSERT(spa->spa_scrub_suspended != 0);
2572	if (--spa->spa_scrub_suspended == 0)
2573		cv_broadcast(&spa->spa_scrub_cv);
2574	mutex_exit(&spa->spa_scrub_lock);
2575}
2576
2577void
2578spa_scrub_restart(spa_t *spa, uint64_t txg)
2579{
2580	/*
2581	 * Something happened (e.g. snapshot create/delete) that means
2582	 * we must restart any in-progress scrubs.  The itinerary will
2583	 * fix this properly.
2584	 */
2585	mutex_enter(&spa->spa_scrub_lock);
2586	spa->spa_scrub_restart_txg = txg;
2587	mutex_exit(&spa->spa_scrub_lock);
2588}
2589
2590int
2591spa_scrub(spa_t *spa, pool_scrub_type_t type, boolean_t force)
2592{
2593	space_seg_t *ss;
2594	uint64_t mintxg, maxtxg;
2595	vdev_t *rvd = spa->spa_root_vdev;
2596
2597	ASSERT(MUTEX_HELD(&spa_namespace_lock));
2598	ASSERT(!spa_config_held(spa, RW_WRITER));
2599
2600	if ((uint_t)type >= POOL_SCRUB_TYPES)
2601		return (ENOTSUP);
2602
2603	mutex_enter(&spa->spa_scrub_lock);
2604
2605	/*
2606	 * If there's a scrub or resilver already in progress, stop it.
2607	 */
2608	while (spa->spa_scrub_thread != NULL) {
2609		/*
2610		 * Don't stop a resilver unless forced.
2611		 */
2612		if (spa->spa_scrub_type == POOL_SCRUB_RESILVER && !force) {
2613			mutex_exit(&spa->spa_scrub_lock);
2614			return (EBUSY);
2615		}
2616		spa->spa_scrub_stop = 1;
2617		cv_broadcast(&spa->spa_scrub_cv);
2618		cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
2619	}
2620
2621	/*
2622	 * Terminate the previous traverse.
2623	 */
2624	if (spa->spa_scrub_th != NULL) {
2625		traverse_fini(spa->spa_scrub_th);
2626		spa->spa_scrub_th = NULL;
2627	}
2628
2629	if (rvd == NULL) {
2630		ASSERT(spa->spa_scrub_stop == 0);
2631		ASSERT(spa->spa_scrub_type == type);
2632		ASSERT(spa->spa_scrub_restart_txg == 0);
2633		mutex_exit(&spa->spa_scrub_lock);
2634		return (0);
2635	}
2636
2637	mintxg = TXG_INITIAL - 1;
2638	maxtxg = spa_last_synced_txg(spa) + 1;
2639
2640	mutex_enter(&rvd->vdev_dtl_lock);
2641
2642	if (rvd->vdev_dtl_map.sm_space == 0) {
2643		/*
2644		 * The pool-wide DTL is empty.
2645		 * If this is a resilver, there's nothing to do except
2646		 * check whether any in-progress replacements have completed.
2647		 */
2648		if (type == POOL_SCRUB_RESILVER) {
2649			type = POOL_SCRUB_NONE;
2650			spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
2651		}
2652	} else {
2653		/*
2654		 * The pool-wide DTL is non-empty.
2655		 * If this is a normal scrub, upgrade to a resilver instead.
2656		 */
2657		if (type == POOL_SCRUB_EVERYTHING)
2658			type = POOL_SCRUB_RESILVER;
2659	}
2660
2661	if (type == POOL_SCRUB_RESILVER) {
2662		/*
2663		 * Determine the resilvering boundaries.
2664		 *
2665		 * Note: (mintxg, maxtxg) is an open interval,
2666		 * i.e. mintxg and maxtxg themselves are not included.
2667		 *
2668		 * Note: for maxtxg, we MIN with spa_last_synced_txg(spa) + 1
2669		 * so we don't claim to resilver a txg that's still changing.
2670		 */
2671		ss = avl_first(&rvd->vdev_dtl_map.sm_root);
2672		mintxg = ss->ss_start - 1;
2673		ss = avl_last(&rvd->vdev_dtl_map.sm_root);
2674		maxtxg = MIN(ss->ss_end, maxtxg);
2675
2676		spa_event_notify(spa, NULL, ESC_ZFS_RESILVER_START);
2677	}
2678
2679	mutex_exit(&rvd->vdev_dtl_lock);
2680
2681	spa->spa_scrub_stop = 0;
2682	spa->spa_scrub_type = type;
2683	spa->spa_scrub_restart_txg = 0;
2684
2685	if (type != POOL_SCRUB_NONE) {
2686		spa->spa_scrub_mintxg = mintxg;
2687		spa->spa_scrub_maxtxg = maxtxg;
2688		spa->spa_scrub_th = traverse_init(spa, spa_scrub_cb, NULL,
2689		    ADVANCE_PRE | ADVANCE_PRUNE | ADVANCE_ZIL,
2690		    ZIO_FLAG_CANFAIL);
2691		traverse_add_pool(spa->spa_scrub_th, mintxg, maxtxg);
2692		spa->spa_scrub_thread = thread_create(NULL, 0,
2693		    spa_scrub_thread, spa, 0, &p0, TS_RUN, minclsyspri);
2694	}
2695
2696	mutex_exit(&spa->spa_scrub_lock);
2697
2698	return (0);
2699}
2700
2701/*
2702 * ==========================================================================
2703 * SPA async task processing
2704 * ==========================================================================
2705 */
2706
2707static void
2708spa_async_remove(spa_t *spa, vdev_t *vd)
2709{
2710	vdev_t *tvd;
2711	int c;
2712
2713	for (c = 0; c < vd->vdev_children; c++) {
2714		tvd = vd->vdev_child[c];
2715		if (tvd->vdev_remove_wanted) {
2716			tvd->vdev_remove_wanted = 0;
2717			vdev_set_state(tvd, B_FALSE, VDEV_STATE_REMOVED,
2718			    VDEV_AUX_NONE);
2719			vdev_clear(spa, tvd);
2720			vdev_config_dirty(tvd->vdev_top);
2721		}
2722		spa_async_remove(spa, tvd);
2723	}
2724}
2725
2726static void
2727spa_async_thread(spa_t *spa)
2728{
2729	int tasks;
2730	uint64_t txg;
2731
2732	ASSERT(spa->spa_sync_on);
2733
2734	mutex_enter(&spa->spa_async_lock);
2735	tasks = spa->spa_async_tasks;
2736	spa->spa_async_tasks = 0;
2737	mutex_exit(&spa->spa_async_lock);
2738
2739	/*
2740	 * See if the config needs to be updated.
2741	 */
2742	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
2743		mutex_enter(&spa_namespace_lock);
2744		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
2745		mutex_exit(&spa_namespace_lock);
2746	}
2747
2748	/*
2749	 * See if any devices need to be marked REMOVED.
2750	 */
2751	if (tasks & SPA_ASYNC_REMOVE) {
2752		txg = spa_vdev_enter(spa);
2753		spa_async_remove(spa, spa->spa_root_vdev);
2754		(void) spa_vdev_exit(spa, NULL, txg, 0);
2755	}
2756
2757	/*
2758	 * If any devices are done replacing, detach them.
2759	 */
2760	if (tasks & SPA_ASYNC_RESILVER_DONE)
2761		spa_vdev_resilver_done(spa);
2762
2763	/*
2764	 * Kick off a scrub.  When starting a RESILVER scrub (or an EVERYTHING
2765	 * scrub which can become a resilver), we need to hold
2766	 * spa_namespace_lock() because the sysevent we post via
2767	 * spa_event_notify() needs to get the name of the pool.
2768	 */
2769	if (tasks & SPA_ASYNC_SCRUB) {
2770		mutex_enter(&spa_namespace_lock);
2771		VERIFY(spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_TRUE) == 0);
2772		mutex_exit(&spa_namespace_lock);
2773	}
2774
2775	/*
2776	 * Kick off a resilver.
2777	 */
2778	if (tasks & SPA_ASYNC_RESILVER) {
2779		mutex_enter(&spa_namespace_lock);
2780		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
2781		mutex_exit(&spa_namespace_lock);
2782	}
2783
2784	/*
2785	 * Let the world know that we're done.
2786	 */
2787	mutex_enter(&spa->spa_async_lock);
2788	spa->spa_async_thread = NULL;
2789	cv_broadcast(&spa->spa_async_cv);
2790	mutex_exit(&spa->spa_async_lock);
2791	thread_exit();
2792}
2793
2794void
2795spa_async_suspend(spa_t *spa)
2796{
2797	mutex_enter(&spa->spa_async_lock);
2798	spa->spa_async_suspended++;
2799	while (spa->spa_async_thread != NULL)
2800		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
2801	mutex_exit(&spa->spa_async_lock);
2802}
2803
2804void
2805spa_async_resume(spa_t *spa)
2806{
2807	mutex_enter(&spa->spa_async_lock);
2808	ASSERT(spa->spa_async_suspended != 0);
2809	spa->spa_async_suspended--;
2810	mutex_exit(&spa->spa_async_lock);
2811}
2812
2813static void
2814spa_async_dispatch(spa_t *spa)
2815{
2816	mutex_enter(&spa->spa_async_lock);
2817	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
2818	    spa->spa_async_thread == NULL &&
2819	    rootdir != NULL && !vn_is_readonly(rootdir))
2820		spa->spa_async_thread = thread_create(NULL, 0,
2821		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
2822	mutex_exit(&spa->spa_async_lock);
2823}
2824
2825void
2826spa_async_request(spa_t *spa, int task)
2827{
2828	mutex_enter(&spa->spa_async_lock);
2829	spa->spa_async_tasks |= task;
2830	mutex_exit(&spa->spa_async_lock);
2831}
2832
2833/*
2834 * ==========================================================================
2835 * SPA syncing routines
2836 * ==========================================================================
2837 */
2838
2839static void
2840spa_sync_deferred_frees(spa_t *spa, uint64_t txg)
2841{
2842	bplist_t *bpl = &spa->spa_sync_bplist;
2843	dmu_tx_t *tx;
2844	blkptr_t blk;
2845	uint64_t itor = 0;
2846	zio_t *zio;
2847	int error;
2848	uint8_t c = 1;
2849
2850	zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CONFIG_HELD);
2851
2852	while (bplist_iterate(bpl, &itor, &blk) == 0)
2853		zio_nowait(zio_free(zio, spa, txg, &blk, NULL, NULL));
2854
2855	error = zio_wait(zio);
2856	ASSERT3U(error, ==, 0);
2857
2858	tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2859	bplist_vacate(bpl, tx);
2860
2861	/*
2862	 * Pre-dirty the first block so we sync to convergence faster.
2863	 * (Usually only the first block is needed.)
2864	 */
2865	dmu_write(spa->spa_meta_objset, spa->spa_sync_bplist_obj, 0, 1, &c, tx);
2866	dmu_tx_commit(tx);
2867}
2868
2869static void
2870spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
2871{
2872	char *packed = NULL;
2873	size_t nvsize = 0;
2874	dmu_buf_t *db;
2875
2876	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
2877
2878	packed = kmem_alloc(nvsize, KM_SLEEP);
2879
2880	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
2881	    KM_SLEEP) == 0);
2882
2883	dmu_write(spa->spa_meta_objset, obj, 0, nvsize, packed, tx);
2884
2885	kmem_free(packed, nvsize);
2886
2887	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
2888	dmu_buf_will_dirty(db, tx);
2889	*(uint64_t *)db->db_data = nvsize;
2890	dmu_buf_rele(db, FTAG);
2891}
2892
2893static void
2894spa_sync_spares(spa_t *spa, dmu_tx_t *tx)
2895{
2896	nvlist_t *nvroot;
2897	nvlist_t **spares;
2898	int i;
2899
2900	if (!spa->spa_sync_spares)
2901		return;
2902
2903	/*
2904	 * Update the MOS nvlist describing the list of available spares.
2905	 * spa_validate_spares() will have already made sure this nvlist is
2906	 * valid and the vdevs are labeled appropriately.
2907	 */
2908	if (spa->spa_spares_object == 0) {
2909		spa->spa_spares_object = dmu_object_alloc(spa->spa_meta_objset,
2910		    DMU_OT_PACKED_NVLIST, 1 << 14,
2911		    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2912		VERIFY(zap_update(spa->spa_meta_objset,
2913		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SPARES,
2914		    sizeof (uint64_t), 1, &spa->spa_spares_object, tx) == 0);
2915	}
2916
2917	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2918	if (spa->spa_nspares == 0) {
2919		VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2920		    NULL, 0) == 0);
2921	} else {
2922		spares = kmem_alloc(spa->spa_nspares * sizeof (void *),
2923		    KM_SLEEP);
2924		for (i = 0; i < spa->spa_nspares; i++)
2925			spares[i] = vdev_config_generate(spa,
2926			    spa->spa_spares[i], B_FALSE, B_TRUE);
2927		VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2928		    spares, spa->spa_nspares) == 0);
2929		for (i = 0; i < spa->spa_nspares; i++)
2930			nvlist_free(spares[i]);
2931		kmem_free(spares, spa->spa_nspares * sizeof (void *));
2932	}
2933
2934	spa_sync_nvlist(spa, spa->spa_spares_object, nvroot, tx);
2935	nvlist_free(nvroot);
2936
2937	spa->spa_sync_spares = B_FALSE;
2938}
2939
2940static void
2941spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
2942{
2943	nvlist_t *config;
2944
2945	if (list_is_empty(&spa->spa_dirty_list))
2946		return;
2947
2948	config = spa_config_generate(spa, NULL, dmu_tx_get_txg(tx), B_FALSE);
2949
2950	if (spa->spa_config_syncing)
2951		nvlist_free(spa->spa_config_syncing);
2952	spa->spa_config_syncing = config;
2953
2954	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
2955}
2956
2957static void
2958spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2959{
2960	spa_t *spa = arg1;
2961	nvlist_t *nvp = arg2;
2962	nvpair_t *nvpair;
2963	objset_t *mos = spa->spa_meta_objset;
2964	uint64_t zapobj;
2965	uint64_t intval;
2966
2967	mutex_enter(&spa->spa_props_lock);
2968	if (spa->spa_pool_props_object == 0) {
2969		zapobj = zap_create(mos, DMU_OT_POOL_PROPS, DMU_OT_NONE, 0, tx);
2970		VERIFY(zapobj > 0);
2971
2972		spa->spa_pool_props_object = zapobj;
2973
2974		VERIFY(zap_update(mos, DMU_POOL_DIRECTORY_OBJECT,
2975		    DMU_POOL_PROPS, 8, 1,
2976		    &spa->spa_pool_props_object, tx) == 0);
2977	}
2978	mutex_exit(&spa->spa_props_lock);
2979
2980	nvpair = NULL;
2981	while ((nvpair = nvlist_next_nvpair(nvp, nvpair))) {
2982		switch (zpool_name_to_prop(nvpair_name(nvpair))) {
2983		case ZPOOL_PROP_DELEGATION:
2984			VERIFY(nvlist_lookup_uint64(nvp,
2985			    nvpair_name(nvpair), &intval) == 0);
2986			VERIFY(zap_update(mos,
2987			    spa->spa_pool_props_object,
2988			    nvpair_name(nvpair), 8, 1,
2989			    &intval, tx) == 0);
2990			spa->spa_delegation = intval;
2991			break;
2992		case ZPOOL_PROP_BOOTFS:
2993			VERIFY(nvlist_lookup_uint64(nvp,
2994			    nvpair_name(nvpair), &spa->spa_bootfs) == 0);
2995			intval = spa->spa_bootfs;
2996			VERIFY(zap_update(mos,
2997			    spa->spa_pool_props_object,
2998			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), 8, 1,
2999			    &intval, tx) == 0);
3000			break;
3001
3002		case ZPOOL_PROP_AUTOREPLACE:
3003			VERIFY(nvlist_lookup_uint64(nvp,
3004			    nvpair_name(nvpair), &intval) == 0);
3005			VERIFY(zap_update(mos,
3006			    spa->spa_pool_props_object,
3007			    zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE), 8, 1,
3008			    &intval, tx) == 0);
3009			break;
3010		}
3011		spa_history_internal_log(LOG_POOL_PROPSET,
3012		    spa, tx, cr, "%s %lld %s",
3013		    nvpair_name(nvpair), intval,
3014		    spa->spa_name);
3015	}
3016}
3017
3018/*
3019 * Sync the specified transaction group.  New blocks may be dirtied as
3020 * part of the process, so we iterate until it converges.
3021 */
3022void
3023spa_sync(spa_t *spa, uint64_t txg)
3024{
3025	dsl_pool_t *dp = spa->spa_dsl_pool;
3026	objset_t *mos = spa->spa_meta_objset;
3027	bplist_t *bpl = &spa->spa_sync_bplist;
3028	vdev_t *rvd = spa->spa_root_vdev;
3029	vdev_t *vd;
3030	dmu_tx_t *tx;
3031	int dirty_vdevs;
3032
3033	/*
3034	 * Lock out configuration changes.
3035	 */
3036	spa_config_enter(spa, RW_READER, FTAG);
3037
3038	spa->spa_syncing_txg = txg;
3039	spa->spa_sync_pass = 0;
3040
3041	VERIFY(0 == bplist_open(bpl, mos, spa->spa_sync_bplist_obj));
3042
3043	tx = dmu_tx_create_assigned(dp, txg);
3044
3045	/*
3046	 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
3047	 * set spa_deflate if we have no raid-z vdevs.
3048	 */
3049	if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
3050	    spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
3051		int i;
3052
3053		for (i = 0; i < rvd->vdev_children; i++) {
3054			vd = rvd->vdev_child[i];
3055			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
3056				break;
3057		}
3058		if (i == rvd->vdev_children) {
3059			spa->spa_deflate = TRUE;
3060			VERIFY(0 == zap_add(spa->spa_meta_objset,
3061			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3062			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
3063		}
3064	}
3065
3066	/*
3067	 * If anything has changed in this txg, push the deferred frees
3068	 * from the previous txg.  If not, leave them alone so that we
3069	 * don't generate work on an otherwise idle system.
3070	 */
3071	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
3072	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
3073	    !txg_list_empty(&dp->dp_sync_tasks, txg))
3074		spa_sync_deferred_frees(spa, txg);
3075
3076	/*
3077	 * Iterate to convergence.
3078	 */
3079	do {
3080		spa->spa_sync_pass++;
3081
3082		spa_sync_config_object(spa, tx);
3083		spa_sync_spares(spa, tx);
3084		spa_errlog_sync(spa, txg);
3085		dsl_pool_sync(dp, txg);
3086
3087		dirty_vdevs = 0;
3088		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) {
3089			vdev_sync(vd, txg);
3090			dirty_vdevs++;
3091		}
3092
3093		bplist_sync(bpl, tx);
3094	} while (dirty_vdevs);
3095
3096	bplist_close(bpl);
3097
3098	dprintf("txg %llu passes %d\n", txg, spa->spa_sync_pass);
3099
3100	/*
3101	 * Rewrite the vdev configuration (which includes the uberblock)
3102	 * to commit the transaction group.
3103	 *
3104	 * If there are any dirty vdevs, sync the uberblock to all vdevs.
3105	 * Otherwise, pick a random top-level vdev that's known to be
3106	 * visible in the config cache (see spa_vdev_add() for details).
3107	 * If the write fails, try the next vdev until we're tried them all.
3108	 */
3109	if (!list_is_empty(&spa->spa_dirty_list)) {
3110		VERIFY(vdev_config_sync(rvd, txg) == 0);
3111	} else {
3112		int children = rvd->vdev_children;
3113		int c0 = spa_get_random(children);
3114		int c;
3115
3116		for (c = 0; c < children; c++) {
3117			vd = rvd->vdev_child[(c0 + c) % children];
3118			if (vd->vdev_ms_array == 0)
3119				continue;
3120			if (vdev_config_sync(vd, txg) == 0)
3121				break;
3122		}
3123		if (c == children)
3124			VERIFY(vdev_config_sync(rvd, txg) == 0);
3125	}
3126
3127	dmu_tx_commit(tx);
3128
3129	/*
3130	 * Clear the dirty config list.
3131	 */
3132	while ((vd = list_head(&spa->spa_dirty_list)) != NULL)
3133		vdev_config_clean(vd);
3134
3135	/*
3136	 * Now that the new config has synced transactionally,
3137	 * let it become visible to the config cache.
3138	 */
3139	if (spa->spa_config_syncing != NULL) {
3140		spa_config_set(spa, spa->spa_config_syncing);
3141		spa->spa_config_txg = txg;
3142		spa->spa_config_syncing = NULL;
3143	}
3144
3145	/*
3146	 * Make a stable copy of the fully synced uberblock.
3147	 * We use this as the root for pool traversals.
3148	 */
3149	spa->spa_traverse_wanted = 1;	/* tells traverse_more() to stop */
3150
3151	spa_scrub_suspend(spa);		/* stop scrubbing and finish I/Os */
3152
3153	rw_enter(&spa->spa_traverse_lock, RW_WRITER);
3154	spa->spa_traverse_wanted = 0;
3155	spa->spa_ubsync = spa->spa_uberblock;
3156	rw_exit(&spa->spa_traverse_lock);
3157
3158	spa_scrub_resume(spa);		/* resume scrub with new ubsync */
3159
3160	/*
3161	 * Clean up the ZIL records for the synced txg.
3162	 */
3163	dsl_pool_zil_clean(dp);
3164
3165	/*
3166	 * Update usable space statistics.
3167	 */
3168	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
3169		vdev_sync_done(vd, txg);
3170
3171	/*
3172	 * It had better be the case that we didn't dirty anything
3173	 * since vdev_config_sync().
3174	 */
3175	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
3176	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
3177	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
3178	ASSERT(bpl->bpl_queue == NULL);
3179
3180	spa_config_exit(spa, FTAG);
3181
3182	/*
3183	 * If any async tasks have been requested, kick them off.
3184	 */
3185	spa_async_dispatch(spa);
3186}
3187
3188/*
3189 * Sync all pools.  We don't want to hold the namespace lock across these
3190 * operations, so we take a reference on the spa_t and drop the lock during the
3191 * sync.
3192 */
3193void
3194spa_sync_allpools(void)
3195{
3196	spa_t *spa = NULL;
3197	mutex_enter(&spa_namespace_lock);
3198	while ((spa = spa_next(spa)) != NULL) {
3199		if (spa_state(spa) != POOL_STATE_ACTIVE)
3200			continue;
3201		spa_open_ref(spa, FTAG);
3202		mutex_exit(&spa_namespace_lock);
3203		txg_wait_synced(spa_get_dsl(spa), 0);
3204		mutex_enter(&spa_namespace_lock);
3205		spa_close(spa, FTAG);
3206	}
3207	mutex_exit(&spa_namespace_lock);
3208}
3209
3210/*
3211 * ==========================================================================
3212 * Miscellaneous routines
3213 * ==========================================================================
3214 */
3215
3216/*
3217 * Remove all pools in the system.
3218 */
3219void
3220spa_evict_all(void)
3221{
3222	spa_t *spa;
3223
3224	/*
3225	 * Remove all cached state.  All pools should be closed now,
3226	 * so every spa in the AVL tree should be unreferenced.
3227	 */
3228	mutex_enter(&spa_namespace_lock);
3229	while ((spa = spa_next(NULL)) != NULL) {
3230		/*
3231		 * Stop async tasks.  The async thread may need to detach
3232		 * a device that's been replaced, which requires grabbing
3233		 * spa_namespace_lock, so we must drop it here.
3234		 */
3235		spa_open_ref(spa, FTAG);
3236		mutex_exit(&spa_namespace_lock);
3237		spa_async_suspend(spa);
3238		mutex_enter(&spa_namespace_lock);
3239		VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0);
3240		spa_close(spa, FTAG);
3241
3242		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3243			spa_unload(spa);
3244			spa_deactivate(spa);
3245		}
3246		spa_remove(spa);
3247	}
3248	mutex_exit(&spa_namespace_lock);
3249}
3250
3251vdev_t *
3252spa_lookup_by_guid(spa_t *spa, uint64_t guid)
3253{
3254	return (vdev_lookup_by_guid(spa->spa_root_vdev, guid));
3255}
3256
3257void
3258spa_upgrade(spa_t *spa)
3259{
3260	spa_config_enter(spa, RW_WRITER, FTAG);
3261
3262	/*
3263	 * This should only be called for a non-faulted pool, and since a
3264	 * future version would result in an unopenable pool, this shouldn't be
3265	 * possible.
3266	 */
3267	ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
3268
3269	spa->spa_uberblock.ub_version = SPA_VERSION;
3270	vdev_config_dirty(spa->spa_root_vdev);
3271
3272	spa_config_exit(spa, FTAG);
3273
3274	txg_wait_synced(spa_get_dsl(spa), 0);
3275}
3276
3277boolean_t
3278spa_has_spare(spa_t *spa, uint64_t guid)
3279{
3280	int i;
3281	uint64_t spareguid;
3282
3283	for (i = 0; i < spa->spa_nspares; i++)
3284		if (spa->spa_spares[i]->vdev_guid == guid)
3285			return (B_TRUE);
3286
3287	for (i = 0; i < spa->spa_pending_nspares; i++) {
3288		if (nvlist_lookup_uint64(spa->spa_pending_spares[i],
3289		    ZPOOL_CONFIG_GUID, &spareguid) == 0 &&
3290		    spareguid == guid)
3291			return (B_TRUE);
3292	}
3293
3294	return (B_FALSE);
3295}
3296
3297int
3298spa_set_props(spa_t *spa, nvlist_t *nvp)
3299{
3300	return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
3301	    spa, nvp, 3));
3302}
3303
3304int
3305spa_get_props(spa_t *spa, nvlist_t **nvp)
3306{
3307	zap_cursor_t zc;
3308	zap_attribute_t za;
3309	objset_t *mos = spa->spa_meta_objset;
3310	zfs_source_t src;
3311	zpool_prop_t prop;
3312	nvlist_t *propval;
3313	uint64_t value;
3314	int err;
3315
3316	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3317
3318	mutex_enter(&spa->spa_props_lock);
3319	/* If no props object, then just return empty nvlist */
3320	if (spa->spa_pool_props_object == 0) {
3321		mutex_exit(&spa->spa_props_lock);
3322		return (0);
3323	}
3324
3325	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
3326	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
3327	    zap_cursor_advance(&zc)) {
3328
3329		if ((prop = zpool_name_to_prop(za.za_name)) == ZFS_PROP_INVAL)
3330			continue;
3331
3332		VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3333		switch (za.za_integer_length) {
3334		case 8:
3335			if (zpool_prop_default_numeric(prop) ==
3336			    za.za_first_integer)
3337				src = ZFS_SRC_DEFAULT;
3338			else
3339				src = ZFS_SRC_LOCAL;
3340			value = za.za_first_integer;
3341
3342			if (prop == ZPOOL_PROP_BOOTFS) {
3343				dsl_pool_t *dp;
3344				dsl_dataset_t *ds = NULL;
3345				char strval[MAXPATHLEN];
3346
3347				dp = spa_get_dsl(spa);
3348				rw_enter(&dp->dp_config_rwlock, RW_READER);
3349				if ((err = dsl_dataset_open_obj(dp,
3350				    za.za_first_integer, NULL, DS_MODE_NONE,
3351				    FTAG, &ds)) != 0) {
3352					rw_exit(&dp->dp_config_rwlock);
3353					break;
3354				}
3355				dsl_dataset_name(ds, strval);
3356				dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
3357				rw_exit(&dp->dp_config_rwlock);
3358
3359				VERIFY(nvlist_add_uint64(propval,
3360				    ZFS_PROP_SOURCE, src) == 0);
3361				VERIFY(nvlist_add_string(propval,
3362				    ZFS_PROP_VALUE, strval) == 0);
3363			} else {
3364				VERIFY(nvlist_add_uint64(propval,
3365				    ZFS_PROP_SOURCE, src) == 0);
3366				VERIFY(nvlist_add_uint64(propval,
3367				    ZFS_PROP_VALUE, value) == 0);
3368			}
3369			VERIFY(nvlist_add_nvlist(*nvp, za.za_name,
3370			    propval) == 0);
3371			break;
3372		}
3373		nvlist_free(propval);
3374	}
3375	zap_cursor_fini(&zc);
3376	mutex_exit(&spa->spa_props_lock);
3377	if (err && err != ENOENT) {
3378		nvlist_free(*nvp);
3379		return (err);
3380	}
3381
3382	return (0);
3383}
3384
3385/*
3386 * If the bootfs property value is dsobj, clear it.
3387 */
3388void
3389spa_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
3390{
3391	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
3392		VERIFY(zap_remove(spa->spa_meta_objset,
3393		    spa->spa_pool_props_object,
3394		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
3395		spa->spa_bootfs = 0;
3396	}
3397}
3398
3399/*
3400 * Post a sysevent corresponding to the given event.  The 'name' must be one of
3401 * the event definitions in sys/sysevent/eventdefs.h.  The payload will be
3402 * filled in from the spa and (optionally) the vdev.  This doesn't do anything
3403 * in the userland libzpool, as we don't want consumers to misinterpret ztest
3404 * or zdb as real changes.
3405 */
3406void
3407spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
3408{
3409#ifdef _KERNEL
3410	sysevent_t		*ev;
3411	sysevent_attr_list_t	*attr = NULL;
3412	sysevent_value_t	value;
3413	sysevent_id_t		eid;
3414
3415	ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
3416	    SE_SLEEP);
3417
3418	value.value_type = SE_DATA_TYPE_STRING;
3419	value.value.sv_string = spa_name(spa);
3420	if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
3421		goto done;
3422
3423	value.value_type = SE_DATA_TYPE_UINT64;
3424	value.value.sv_uint64 = spa_guid(spa);
3425	if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
3426		goto done;
3427
3428	if (vd) {
3429		value.value_type = SE_DATA_TYPE_UINT64;
3430		value.value.sv_uint64 = vd->vdev_guid;
3431		if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
3432		    SE_SLEEP) != 0)
3433			goto done;
3434
3435		if (vd->vdev_path) {
3436			value.value_type = SE_DATA_TYPE_STRING;
3437			value.value.sv_string = vd->vdev_path;
3438			if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
3439			    &value, SE_SLEEP) != 0)
3440				goto done;
3441		}
3442	}
3443
3444	(void) log_sysevent(ev, SE_SLEEP, &eid);
3445
3446done:
3447	if (attr)
3448		sysevent_free_attr(attr);
3449	sysevent_free(ev);
3450#endif
3451}
3452