spa.c revision 0a4e9518a44f226be6d39383330b5b1792d2f184
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
63#include "zfs_prop.h"
64
65int zio_taskq_threads = 8;
66
67static void spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx);
68
69/*
70 * ==========================================================================
71 * SPA properties routines
72 * ==========================================================================
73 */
74
75/*
76 * Add a (source=src, propname=propval) list to an nvlist.
77 */
78static int
79spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
80    uint64_t intval, zprop_source_t src)
81{
82	const char *propname = zpool_prop_to_name(prop);
83	nvlist_t *propval;
84	int err = 0;
85
86	if (err = nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP))
87		return (err);
88
89	if (err = nvlist_add_uint64(propval, ZPROP_SOURCE, src))
90		goto out;
91
92	if (strval != NULL) {
93		if (err = nvlist_add_string(propval, ZPROP_VALUE, strval))
94			goto out;
95	} else {
96		if (err = nvlist_add_uint64(propval, ZPROP_VALUE, intval))
97			goto out;
98	}
99
100	err = nvlist_add_nvlist(nvl, propname, propval);
101out:
102	nvlist_free(propval);
103	return (err);
104}
105
106/*
107 * Get property values from the spa configuration.
108 */
109static int
110spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
111{
112	uint64_t size = spa_get_space(spa);
113	uint64_t used = spa_get_alloc(spa);
114	uint64_t cap, version;
115	zprop_source_t src = ZPROP_SRC_NONE;
116	int err;
117
118	/*
119	 * readonly properties
120	 */
121	if (err = spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa->spa_name,
122	    0, src))
123		return (err);
124
125	if (err = spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src))
126		return (err);
127
128	if (err = spa_prop_add_list(*nvp, ZPOOL_PROP_USED, NULL, used, src))
129		return (err);
130
131	if (err = spa_prop_add_list(*nvp, ZPOOL_PROP_AVAILABLE, NULL,
132	    size - used, src))
133		return (err);
134
135	cap = (size == 0) ? 0 : (used * 100 / size);
136	if (err = spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src))
137		return (err);
138
139	if (err = spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL,
140	    spa_guid(spa), src))
141		return (err);
142
143	if (err = spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
144	    spa->spa_root_vdev->vdev_state, src))
145		return (err);
146
147	/*
148	 * settable properties that are not stored in the pool property object.
149	 */
150	version = spa_version(spa);
151	if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
152		src = ZPROP_SRC_DEFAULT;
153	else
154		src = ZPROP_SRC_LOCAL;
155	if (err = spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL,
156	    version, src))
157		return (err);
158
159	if (spa->spa_root != NULL) {
160		src = ZPROP_SRC_LOCAL;
161		if (err = spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT,
162		    spa->spa_root, 0, src))
163			return (err);
164	}
165
166	if (spa->spa_temporary ==
167	    zpool_prop_default_numeric(ZPOOL_PROP_TEMPORARY))
168		src = ZPROP_SRC_DEFAULT;
169	else
170		src = ZPROP_SRC_LOCAL;
171	if (err = spa_prop_add_list(*nvp, ZPOOL_PROP_TEMPORARY, NULL,
172	    spa->spa_temporary, src))
173		return (err);
174
175	return (0);
176}
177
178/*
179 * Get zpool property values.
180 */
181int
182spa_prop_get(spa_t *spa, nvlist_t **nvp)
183{
184	zap_cursor_t zc;
185	zap_attribute_t za;
186	objset_t *mos = spa->spa_meta_objset;
187	int err;
188
189	if (err = nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP))
190		return (err);
191
192	/*
193	 * Get properties from the spa config.
194	 */
195	if (err = spa_prop_get_config(spa, nvp))
196		goto out;
197
198	mutex_enter(&spa->spa_props_lock);
199	/* If no pool property object, no more prop to get. */
200	if (spa->spa_pool_props_object == 0) {
201		mutex_exit(&spa->spa_props_lock);
202		return (0);
203	}
204
205	/*
206	 * Get properties from the MOS pool property object.
207	 */
208	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
209	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
210	    zap_cursor_advance(&zc)) {
211		uint64_t intval = 0;
212		char *strval = NULL;
213		zprop_source_t src = ZPROP_SRC_DEFAULT;
214		zpool_prop_t prop;
215
216		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
217			continue;
218
219		switch (za.za_integer_length) {
220		case 8:
221			/* integer property */
222			if (za.za_first_integer !=
223			    zpool_prop_default_numeric(prop))
224				src = ZPROP_SRC_LOCAL;
225
226			if (prop == ZPOOL_PROP_BOOTFS) {
227				dsl_pool_t *dp;
228				dsl_dataset_t *ds = NULL;
229
230				dp = spa_get_dsl(spa);
231				rw_enter(&dp->dp_config_rwlock, RW_READER);
232				if (err = dsl_dataset_open_obj(dp,
233				    za.za_first_integer, NULL, DS_MODE_NONE,
234				    FTAG, &ds)) {
235					rw_exit(&dp->dp_config_rwlock);
236					break;
237				}
238
239				strval = kmem_alloc(
240				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
241				    KM_SLEEP);
242				dsl_dataset_name(ds, strval);
243				dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
244				rw_exit(&dp->dp_config_rwlock);
245			} else {
246				strval = NULL;
247				intval = za.za_first_integer;
248			}
249
250			err = spa_prop_add_list(*nvp, prop, strval,
251			    intval, src);
252
253			if (strval != NULL)
254				kmem_free(strval,
255				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
256
257			break;
258
259		case 1:
260			/* string property */
261			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
262			err = zap_lookup(mos, spa->spa_pool_props_object,
263			    za.za_name, 1, za.za_num_integers, strval);
264			if (err) {
265				kmem_free(strval, za.za_num_integers);
266				break;
267			}
268			err = spa_prop_add_list(*nvp, prop, strval, 0, src);
269			kmem_free(strval, za.za_num_integers);
270			break;
271
272		default:
273			break;
274		}
275	}
276	zap_cursor_fini(&zc);
277	mutex_exit(&spa->spa_props_lock);
278out:
279	if (err && err != ENOENT) {
280		nvlist_free(*nvp);
281		return (err);
282	}
283
284	return (0);
285}
286
287/*
288 * Validate the given pool properties nvlist and modify the list
289 * for the property values to be set.
290 */
291static int
292spa_prop_validate(spa_t *spa, nvlist_t *props)
293{
294	nvpair_t *elem;
295	int error = 0, reset_bootfs = 0;
296	uint64_t objnum;
297
298	elem = NULL;
299	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
300		zpool_prop_t prop;
301		char *propname, *strval;
302		uint64_t intval;
303		vdev_t *rvdev;
304		char *vdev_type;
305		objset_t *os;
306
307		propname = nvpair_name(elem);
308
309		if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
310			return (EINVAL);
311
312		switch (prop) {
313		case ZPOOL_PROP_VERSION:
314			error = nvpair_value_uint64(elem, &intval);
315			if (!error &&
316			    (intval < spa_version(spa) || intval > SPA_VERSION))
317				error = EINVAL;
318			break;
319
320		case ZPOOL_PROP_DELEGATION:
321		case ZPOOL_PROP_AUTOREPLACE:
322			error = nvpair_value_uint64(elem, &intval);
323			if (!error && intval > 1)
324				error = EINVAL;
325			break;
326
327		case ZPOOL_PROP_BOOTFS:
328			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
329				error = ENOTSUP;
330				break;
331			}
332
333			/*
334			 * A bootable filesystem can not be on a RAIDZ pool
335			 * nor a striped pool with more than 1 device.
336			 */
337			rvdev = spa->spa_root_vdev;
338			vdev_type =
339			    rvdev->vdev_child[0]->vdev_ops->vdev_op_type;
340			if (rvdev->vdev_children > 1 ||
341			    strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
342			    strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
343				error = ENOTSUP;
344				break;
345			}
346
347			reset_bootfs = 1;
348
349			error = nvpair_value_string(elem, &strval);
350
351			if (!error) {
352				if (strval == NULL || strval[0] == '\0') {
353					objnum = zpool_prop_default_numeric(
354					    ZPOOL_PROP_BOOTFS);
355					break;
356				}
357
358				if (error = dmu_objset_open(strval, DMU_OST_ZFS,
359				    DS_MODE_STANDARD | DS_MODE_READONLY, &os))
360					break;
361				objnum = dmu_objset_id(os);
362				dmu_objset_close(os);
363			}
364			break;
365		case ZPOOL_PROP_FAILUREMODE:
366			error = nvpair_value_uint64(elem, &intval);
367			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
368			    intval > ZIO_FAILURE_MODE_PANIC))
369				error = EINVAL;
370
371			/*
372			 * This is a special case which only occurs when
373			 * the pool has completely failed. This allows
374			 * the user to change the in-core failmode property
375			 * without syncing it out to disk (I/Os might
376			 * currently be blocked). We do this by returning
377			 * EIO to the caller (spa_prop_set) to trick it
378			 * into thinking we encountered a property validation
379			 * error.
380			 */
381			if (!error && spa_state(spa) == POOL_STATE_IO_FAILURE) {
382				spa->spa_failmode = intval;
383				error = EIO;
384			}
385			break;
386		}
387
388		if (error)
389			break;
390	}
391
392	if (!error && reset_bootfs) {
393		error = nvlist_remove(props,
394		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
395
396		if (!error) {
397			error = nvlist_add_uint64(props,
398			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
399		}
400	}
401
402	return (error);
403}
404
405int
406spa_prop_set(spa_t *spa, nvlist_t *nvp)
407{
408	int error;
409
410	if ((error = spa_prop_validate(spa, nvp)) != 0)
411		return (error);
412
413	return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
414	    spa, nvp, 3));
415}
416
417/*
418 * If the bootfs property value is dsobj, clear it.
419 */
420void
421spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
422{
423	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
424		VERIFY(zap_remove(spa->spa_meta_objset,
425		    spa->spa_pool_props_object,
426		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
427		spa->spa_bootfs = 0;
428	}
429}
430
431/*
432 * ==========================================================================
433 * SPA state manipulation (open/create/destroy/import/export)
434 * ==========================================================================
435 */
436
437static int
438spa_error_entry_compare(const void *a, const void *b)
439{
440	spa_error_entry_t *sa = (spa_error_entry_t *)a;
441	spa_error_entry_t *sb = (spa_error_entry_t *)b;
442	int ret;
443
444	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
445	    sizeof (zbookmark_t));
446
447	if (ret < 0)
448		return (-1);
449	else if (ret > 0)
450		return (1);
451	else
452		return (0);
453}
454
455/*
456 * Utility function which retrieves copies of the current logs and
457 * re-initializes them in the process.
458 */
459void
460spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
461{
462	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
463
464	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
465	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
466
467	avl_create(&spa->spa_errlist_scrub,
468	    spa_error_entry_compare, sizeof (spa_error_entry_t),
469	    offsetof(spa_error_entry_t, se_avl));
470	avl_create(&spa->spa_errlist_last,
471	    spa_error_entry_compare, sizeof (spa_error_entry_t),
472	    offsetof(spa_error_entry_t, se_avl));
473}
474
475/*
476 * Activate an uninitialized pool.
477 */
478static void
479spa_activate(spa_t *spa)
480{
481	int t;
482
483	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
484
485	spa->spa_state = POOL_STATE_ACTIVE;
486
487	spa->spa_normal_class = metaslab_class_create();
488	spa->spa_log_class = metaslab_class_create();
489
490	for (t = 0; t < ZIO_TYPES; t++) {
491		spa->spa_zio_issue_taskq[t] = taskq_create("spa_zio_issue",
492		    zio_taskq_threads, maxclsyspri, 50, INT_MAX,
493		    TASKQ_PREPOPULATE);
494		spa->spa_zio_intr_taskq[t] = taskq_create("spa_zio_intr",
495		    zio_taskq_threads, maxclsyspri, 50, INT_MAX,
496		    TASKQ_PREPOPULATE);
497	}
498
499	list_create(&spa->spa_dirty_list, sizeof (vdev_t),
500	    offsetof(vdev_t, vdev_dirty_node));
501	list_create(&spa->spa_zio_list, sizeof (zio_t),
502	    offsetof(zio_t, zio_link_node));
503
504	txg_list_create(&spa->spa_vdev_txg_list,
505	    offsetof(struct vdev, vdev_txg_node));
506
507	avl_create(&spa->spa_errlist_scrub,
508	    spa_error_entry_compare, sizeof (spa_error_entry_t),
509	    offsetof(spa_error_entry_t, se_avl));
510	avl_create(&spa->spa_errlist_last,
511	    spa_error_entry_compare, sizeof (spa_error_entry_t),
512	    offsetof(spa_error_entry_t, se_avl));
513}
514
515/*
516 * Opposite of spa_activate().
517 */
518static void
519spa_deactivate(spa_t *spa)
520{
521	int t;
522
523	ASSERT(spa->spa_sync_on == B_FALSE);
524	ASSERT(spa->spa_dsl_pool == NULL);
525	ASSERT(spa->spa_root_vdev == NULL);
526
527	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
528
529	txg_list_destroy(&spa->spa_vdev_txg_list);
530
531	list_destroy(&spa->spa_dirty_list);
532	list_destroy(&spa->spa_zio_list);
533
534	for (t = 0; t < ZIO_TYPES; t++) {
535		taskq_destroy(spa->spa_zio_issue_taskq[t]);
536		taskq_destroy(spa->spa_zio_intr_taskq[t]);
537		spa->spa_zio_issue_taskq[t] = NULL;
538		spa->spa_zio_intr_taskq[t] = NULL;
539	}
540
541	metaslab_class_destroy(spa->spa_normal_class);
542	spa->spa_normal_class = NULL;
543
544	metaslab_class_destroy(spa->spa_log_class);
545	spa->spa_log_class = NULL;
546
547	/*
548	 * If this was part of an import or the open otherwise failed, we may
549	 * still have errors left in the queues.  Empty them just in case.
550	 */
551	spa_errlog_drain(spa);
552
553	avl_destroy(&spa->spa_errlist_scrub);
554	avl_destroy(&spa->spa_errlist_last);
555
556	spa->spa_state = POOL_STATE_UNINITIALIZED;
557}
558
559/*
560 * Verify a pool configuration, and construct the vdev tree appropriately.  This
561 * will create all the necessary vdevs in the appropriate layout, with each vdev
562 * in the CLOSED state.  This will prep the pool before open/creation/import.
563 * All vdev validation is done by the vdev_alloc() routine.
564 */
565static int
566spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
567    uint_t id, int atype)
568{
569	nvlist_t **child;
570	uint_t c, children;
571	int error;
572
573	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
574		return (error);
575
576	if ((*vdp)->vdev_ops->vdev_op_leaf)
577		return (0);
578
579	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
580	    &child, &children) != 0) {
581		vdev_free(*vdp);
582		*vdp = NULL;
583		return (EINVAL);
584	}
585
586	for (c = 0; c < children; c++) {
587		vdev_t *vd;
588		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
589		    atype)) != 0) {
590			vdev_free(*vdp);
591			*vdp = NULL;
592			return (error);
593		}
594	}
595
596	ASSERT(*vdp != NULL);
597
598	return (0);
599}
600
601/*
602 * Opposite of spa_load().
603 */
604static void
605spa_unload(spa_t *spa)
606{
607	int i;
608
609	/*
610	 * Stop async tasks.
611	 */
612	spa_async_suspend(spa);
613
614	/*
615	 * Stop syncing.
616	 */
617	if (spa->spa_sync_on) {
618		txg_sync_stop(spa->spa_dsl_pool);
619		spa->spa_sync_on = B_FALSE;
620	}
621
622	/*
623	 * Wait for any outstanding prefetch I/O to complete.
624	 */
625	spa_config_enter(spa, RW_WRITER, FTAG);
626	spa_config_exit(spa, FTAG);
627
628	/*
629	 * Close the dsl pool.
630	 */
631	if (spa->spa_dsl_pool) {
632		dsl_pool_close(spa->spa_dsl_pool);
633		spa->spa_dsl_pool = NULL;
634	}
635
636	/*
637	 * Close all vdevs.
638	 */
639	if (spa->spa_root_vdev)
640		vdev_free(spa->spa_root_vdev);
641	ASSERT(spa->spa_root_vdev == NULL);
642
643	for (i = 0; i < spa->spa_nspares; i++)
644		vdev_free(spa->spa_spares[i]);
645	if (spa->spa_spares) {
646		kmem_free(spa->spa_spares, spa->spa_nspares * sizeof (void *));
647		spa->spa_spares = NULL;
648	}
649	if (spa->spa_sparelist) {
650		nvlist_free(spa->spa_sparelist);
651		spa->spa_sparelist = NULL;
652	}
653
654	spa->spa_async_suspended = 0;
655}
656
657/*
658 * Load (or re-load) the current list of vdevs describing the active spares for
659 * this pool.  When this is called, we have some form of basic information in
660 * 'spa_sparelist'.  We parse this into vdevs, try to open them, and then
661 * re-generate a more complete list including status information.
662 */
663static void
664spa_load_spares(spa_t *spa)
665{
666	nvlist_t **spares;
667	uint_t nspares;
668	int i;
669	vdev_t *vd, *tvd;
670
671	/*
672	 * First, close and free any existing spare vdevs.
673	 */
674	for (i = 0; i < spa->spa_nspares; i++) {
675		vd = spa->spa_spares[i];
676
677		/* Undo the call to spa_activate() below */
678		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid)) != NULL &&
679		    tvd->vdev_isspare)
680			spa_spare_remove(tvd);
681		vdev_close(vd);
682		vdev_free(vd);
683	}
684
685	if (spa->spa_spares)
686		kmem_free(spa->spa_spares, spa->spa_nspares * sizeof (void *));
687
688	if (spa->spa_sparelist == NULL)
689		nspares = 0;
690	else
691		VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist,
692		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
693
694	spa->spa_nspares = (int)nspares;
695	spa->spa_spares = NULL;
696
697	if (nspares == 0)
698		return;
699
700	/*
701	 * Construct the array of vdevs, opening them to get status in the
702	 * process.   For each spare, there is potentially two different vdev_t
703	 * structures associated with it: one in the list of spares (used only
704	 * for basic validation purposes) and one in the active vdev
705	 * configuration (if it's spared in).  During this phase we open and
706	 * validate each vdev on the spare list.  If the vdev also exists in the
707	 * active configuration, then we also mark this vdev as an active spare.
708	 */
709	spa->spa_spares = kmem_alloc(nspares * sizeof (void *), KM_SLEEP);
710	for (i = 0; i < spa->spa_nspares; i++) {
711		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
712		    VDEV_ALLOC_SPARE) == 0);
713		ASSERT(vd != NULL);
714
715		spa->spa_spares[i] = vd;
716
717		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid)) != NULL) {
718			if (!tvd->vdev_isspare)
719				spa_spare_add(tvd);
720
721			/*
722			 * We only mark the spare active if we were successfully
723			 * able to load the vdev.  Otherwise, importing a pool
724			 * with a bad active spare would result in strange
725			 * behavior, because multiple pool would think the spare
726			 * is actively in use.
727			 *
728			 * There is a vulnerability here to an equally bizarre
729			 * circumstance, where a dead active spare is later
730			 * brought back to life (onlined or otherwise).  Given
731			 * the rarity of this scenario, and the extra complexity
732			 * it adds, we ignore the possibility.
733			 */
734			if (!vdev_is_dead(tvd))
735				spa_spare_activate(tvd);
736		}
737
738		if (vdev_open(vd) != 0)
739			continue;
740
741		vd->vdev_top = vd;
742		(void) vdev_validate_spare(vd);
743	}
744
745	/*
746	 * Recompute the stashed list of spares, with status information
747	 * this time.
748	 */
749	VERIFY(nvlist_remove(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
750	    DATA_TYPE_NVLIST_ARRAY) == 0);
751
752	spares = kmem_alloc(spa->spa_nspares * sizeof (void *), KM_SLEEP);
753	for (i = 0; i < spa->spa_nspares; i++)
754		spares[i] = vdev_config_generate(spa, spa->spa_spares[i],
755		    B_TRUE, B_TRUE);
756	VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
757	    spares, spa->spa_nspares) == 0);
758	for (i = 0; i < spa->spa_nspares; i++)
759		nvlist_free(spares[i]);
760	kmem_free(spares, spa->spa_nspares * sizeof (void *));
761}
762
763static int
764load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
765{
766	dmu_buf_t *db;
767	char *packed = NULL;
768	size_t nvsize = 0;
769	int error;
770	*value = NULL;
771
772	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
773	nvsize = *(uint64_t *)db->db_data;
774	dmu_buf_rele(db, FTAG);
775
776	packed = kmem_alloc(nvsize, KM_SLEEP);
777	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed);
778	if (error == 0)
779		error = nvlist_unpack(packed, nvsize, value, 0);
780	kmem_free(packed, nvsize);
781
782	return (error);
783}
784
785/*
786 * Checks to see if the given vdev could not be opened, in which case we post a
787 * sysevent to notify the autoreplace code that the device has been removed.
788 */
789static void
790spa_check_removed(vdev_t *vd)
791{
792	int c;
793
794	for (c = 0; c < vd->vdev_children; c++)
795		spa_check_removed(vd->vdev_child[c]);
796
797	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
798		zfs_post_autoreplace(vd->vdev_spa, vd);
799		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
800	}
801}
802
803/*
804 * Load an existing storage pool, using the pool's builtin spa_config as a
805 * source of configuration information.
806 */
807static int
808spa_load(spa_t *spa, nvlist_t *config, spa_load_state_t state, int mosconfig)
809{
810	int error = 0;
811	nvlist_t *nvroot = NULL;
812	vdev_t *rvd;
813	uberblock_t *ub = &spa->spa_uberblock;
814	uint64_t config_cache_txg = spa->spa_config_txg;
815	uint64_t pool_guid;
816	uint64_t version;
817	zio_t *zio;
818	uint64_t autoreplace = 0;
819
820	spa->spa_load_state = state;
821
822	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) ||
823	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
824		error = EINVAL;
825		goto out;
826	}
827
828	/*
829	 * Versioning wasn't explicitly added to the label until later, so if
830	 * it's not present treat it as the initial version.
831	 */
832	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &version) != 0)
833		version = SPA_VERSION_INITIAL;
834
835	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
836	    &spa->spa_config_txg);
837
838	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
839	    spa_guid_exists(pool_guid, 0)) {
840		error = EEXIST;
841		goto out;
842	}
843
844	spa->spa_load_guid = pool_guid;
845
846	/*
847	 * Parse the configuration into a vdev tree.  We explicitly set the
848	 * value that will be returned by spa_version() since parsing the
849	 * configuration requires knowing the version number.
850	 */
851	spa_config_enter(spa, RW_WRITER, FTAG);
852	spa->spa_ubsync.ub_version = version;
853	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_LOAD);
854	spa_config_exit(spa, FTAG);
855
856	if (error != 0)
857		goto out;
858
859	ASSERT(spa->spa_root_vdev == rvd);
860	ASSERT(spa_guid(spa) == pool_guid);
861
862	/*
863	 * Try to open all vdevs, loading each label in the process.
864	 */
865	error = vdev_open(rvd);
866	if (error != 0)
867		goto out;
868
869	/*
870	 * Validate the labels for all leaf vdevs.  We need to grab the config
871	 * lock because all label I/O is done with the ZIO_FLAG_CONFIG_HELD
872	 * flag.
873	 */
874	spa_config_enter(spa, RW_READER, FTAG);
875	error = vdev_validate(rvd);
876	spa_config_exit(spa, FTAG);
877
878	if (error != 0)
879		goto out;
880
881	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
882		error = ENXIO;
883		goto out;
884	}
885
886	/*
887	 * Find the best uberblock.
888	 */
889	bzero(ub, sizeof (uberblock_t));
890
891	zio = zio_root(spa, NULL, NULL,
892	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
893	vdev_uberblock_load(zio, rvd, ub);
894	error = zio_wait(zio);
895
896	/*
897	 * If we weren't able to find a single valid uberblock, return failure.
898	 */
899	if (ub->ub_txg == 0) {
900		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
901		    VDEV_AUX_CORRUPT_DATA);
902		error = ENXIO;
903		goto out;
904	}
905
906	/*
907	 * If the pool is newer than the code, we can't open it.
908	 */
909	if (ub->ub_version > SPA_VERSION) {
910		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
911		    VDEV_AUX_VERSION_NEWER);
912		error = ENOTSUP;
913		goto out;
914	}
915
916	/*
917	 * If the vdev guid sum doesn't match the uberblock, we have an
918	 * incomplete configuration.
919	 */
920	if (rvd->vdev_guid_sum != ub->ub_guid_sum && mosconfig) {
921		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
922		    VDEV_AUX_BAD_GUID_SUM);
923		error = ENXIO;
924		goto out;
925	}
926
927	/*
928	 * Initialize internal SPA structures.
929	 */
930	spa->spa_state = POOL_STATE_ACTIVE;
931	spa->spa_ubsync = spa->spa_uberblock;
932	spa->spa_first_txg = spa_last_synced_txg(spa) + 1;
933	error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
934	if (error) {
935		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
936		    VDEV_AUX_CORRUPT_DATA);
937		goto out;
938	}
939	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
940
941	if (zap_lookup(spa->spa_meta_objset,
942	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
943	    sizeof (uint64_t), 1, &spa->spa_config_object) != 0) {
944		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
945		    VDEV_AUX_CORRUPT_DATA);
946		error = EIO;
947		goto out;
948	}
949
950	if (!mosconfig) {
951		nvlist_t *newconfig;
952		uint64_t hostid;
953
954		if (load_nvlist(spa, spa->spa_config_object, &newconfig) != 0) {
955			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
956			    VDEV_AUX_CORRUPT_DATA);
957			error = EIO;
958			goto out;
959		}
960
961		if (nvlist_lookup_uint64(newconfig, ZPOOL_CONFIG_HOSTID,
962		    &hostid) == 0) {
963			char *hostname;
964			unsigned long myhostid = 0;
965
966			VERIFY(nvlist_lookup_string(newconfig,
967			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
968
969			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
970			if (hostid != 0 && myhostid != 0 &&
971			    (unsigned long)hostid != myhostid) {
972				cmn_err(CE_WARN, "pool '%s' could not be "
973				    "loaded as it was last accessed by "
974				    "another system (host: %s hostid: 0x%lx).  "
975				    "See: http://www.sun.com/msg/ZFS-8000-EY",
976				    spa->spa_name, hostname,
977				    (unsigned long)hostid);
978				error = EBADF;
979				goto out;
980			}
981		}
982
983		spa_config_set(spa, newconfig);
984		spa_unload(spa);
985		spa_deactivate(spa);
986		spa_activate(spa);
987
988		return (spa_load(spa, newconfig, state, B_TRUE));
989	}
990
991	if (zap_lookup(spa->spa_meta_objset,
992	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
993	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj) != 0) {
994		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
995		    VDEV_AUX_CORRUPT_DATA);
996		error = EIO;
997		goto out;
998	}
999
1000	/*
1001	 * Load the bit that tells us to use the new accounting function
1002	 * (raid-z deflation).  If we have an older pool, this will not
1003	 * be present.
1004	 */
1005	error = zap_lookup(spa->spa_meta_objset,
1006	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
1007	    sizeof (uint64_t), 1, &spa->spa_deflate);
1008	if (error != 0 && error != ENOENT) {
1009		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1010		    VDEV_AUX_CORRUPT_DATA);
1011		error = EIO;
1012		goto out;
1013	}
1014
1015	/*
1016	 * Load the persistent error log.  If we have an older pool, this will
1017	 * not be present.
1018	 */
1019	error = zap_lookup(spa->spa_meta_objset,
1020	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_LAST,
1021	    sizeof (uint64_t), 1, &spa->spa_errlog_last);
1022	if (error != 0 && error != ENOENT) {
1023		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1024		    VDEV_AUX_CORRUPT_DATA);
1025		error = EIO;
1026		goto out;
1027	}
1028
1029	error = zap_lookup(spa->spa_meta_objset,
1030	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_SCRUB,
1031	    sizeof (uint64_t), 1, &spa->spa_errlog_scrub);
1032	if (error != 0 && error != ENOENT) {
1033		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1034		    VDEV_AUX_CORRUPT_DATA);
1035		error = EIO;
1036		goto out;
1037	}
1038
1039	/*
1040	 * Load the history object.  If we have an older pool, this
1041	 * will not be present.
1042	 */
1043	error = zap_lookup(spa->spa_meta_objset,
1044	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_HISTORY,
1045	    sizeof (uint64_t), 1, &spa->spa_history);
1046	if (error != 0 && error != ENOENT) {
1047		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1048		    VDEV_AUX_CORRUPT_DATA);
1049		error = EIO;
1050		goto out;
1051	}
1052
1053	/*
1054	 * Load any hot spares for this pool.
1055	 */
1056	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1057	    DMU_POOL_SPARES, sizeof (uint64_t), 1, &spa->spa_spares_object);
1058	if (error != 0 && error != ENOENT) {
1059		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1060		    VDEV_AUX_CORRUPT_DATA);
1061		error = EIO;
1062		goto out;
1063	}
1064	if (error == 0) {
1065		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
1066		if (load_nvlist(spa, spa->spa_spares_object,
1067		    &spa->spa_sparelist) != 0) {
1068			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1069			    VDEV_AUX_CORRUPT_DATA);
1070			error = EIO;
1071			goto out;
1072		}
1073
1074		spa_config_enter(spa, RW_WRITER, FTAG);
1075		spa_load_spares(spa);
1076		spa_config_exit(spa, FTAG);
1077	}
1078
1079	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
1080
1081	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1082	    DMU_POOL_PROPS, sizeof (uint64_t), 1, &spa->spa_pool_props_object);
1083
1084	if (error && error != ENOENT) {
1085		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1086		    VDEV_AUX_CORRUPT_DATA);
1087		error = EIO;
1088		goto out;
1089	}
1090
1091	if (error == 0) {
1092		(void) zap_lookup(spa->spa_meta_objset,
1093		    spa->spa_pool_props_object,
1094		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS),
1095		    sizeof (uint64_t), 1, &spa->spa_bootfs);
1096		(void) zap_lookup(spa->spa_meta_objset,
1097		    spa->spa_pool_props_object,
1098		    zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE),
1099		    sizeof (uint64_t), 1, &autoreplace);
1100		(void) zap_lookup(spa->spa_meta_objset,
1101		    spa->spa_pool_props_object,
1102		    zpool_prop_to_name(ZPOOL_PROP_DELEGATION),
1103		    sizeof (uint64_t), 1, &spa->spa_delegation);
1104		(void) zap_lookup(spa->spa_meta_objset,
1105		    spa->spa_pool_props_object,
1106		    zpool_prop_to_name(ZPOOL_PROP_FAILUREMODE),
1107		    sizeof (uint64_t), 1, &spa->spa_failmode);
1108	}
1109
1110	/*
1111	 * If the 'autoreplace' property is set, then post a resource notifying
1112	 * the ZFS DE that it should not issue any faults for unopenable
1113	 * devices.  We also iterate over the vdevs, and post a sysevent for any
1114	 * unopenable vdevs so that the normal autoreplace handler can take
1115	 * over.
1116	 */
1117	if (autoreplace)
1118		spa_check_removed(spa->spa_root_vdev);
1119
1120	/*
1121	 * Load the vdev state for all toplevel vdevs.
1122	 */
1123	vdev_load(rvd);
1124
1125	/*
1126	 * Propagate the leaf DTLs we just loaded all the way up the tree.
1127	 */
1128	spa_config_enter(spa, RW_WRITER, FTAG);
1129	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
1130	spa_config_exit(spa, FTAG);
1131
1132	/*
1133	 * Check the state of the root vdev.  If it can't be opened, it
1134	 * indicates one or more toplevel vdevs are faulted.
1135	 */
1136	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
1137		error = ENXIO;
1138		goto out;
1139	}
1140
1141	if ((spa_mode & FWRITE) && state != SPA_LOAD_TRYIMPORT) {
1142		dmu_tx_t *tx;
1143		int need_update = B_FALSE;
1144		int c;
1145
1146		/*
1147		 * Claim log blocks that haven't been committed yet.
1148		 * This must all happen in a single txg.
1149		 */
1150		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
1151		    spa_first_txg(spa));
1152		(void) dmu_objset_find(spa->spa_name,
1153		    zil_claim, tx, DS_FIND_CHILDREN);
1154		dmu_tx_commit(tx);
1155
1156		spa->spa_sync_on = B_TRUE;
1157		txg_sync_start(spa->spa_dsl_pool);
1158
1159		/*
1160		 * Wait for all claims to sync.
1161		 */
1162		txg_wait_synced(spa->spa_dsl_pool, 0);
1163
1164		/*
1165		 * If the config cache is stale, or we have uninitialized
1166		 * metaslabs (see spa_vdev_add()), then update the config.
1167		 */
1168		if (config_cache_txg != spa->spa_config_txg ||
1169		    state == SPA_LOAD_IMPORT)
1170			need_update = B_TRUE;
1171
1172		for (c = 0; c < rvd->vdev_children; c++)
1173			if (rvd->vdev_child[c]->vdev_ms_array == 0)
1174				need_update = B_TRUE;
1175
1176		/*
1177		 * Update the config cache asychronously in case we're the
1178		 * root pool, in which case the config cache isn't writable yet.
1179		 */
1180		if (need_update)
1181			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
1182	}
1183
1184	error = 0;
1185out:
1186	if (error && error != EBADF)
1187		zfs_ereport_post(FM_EREPORT_ZFS_POOL, spa, NULL, NULL, 0, 0);
1188	spa->spa_load_state = SPA_LOAD_NONE;
1189	spa->spa_ena = 0;
1190
1191	return (error);
1192}
1193
1194/*
1195 * Pool Open/Import
1196 *
1197 * The import case is identical to an open except that the configuration is sent
1198 * down from userland, instead of grabbed from the configuration cache.  For the
1199 * case of an open, the pool configuration will exist in the
1200 * POOL_STATE_UNINITIALIZED state.
1201 *
1202 * The stats information (gen/count/ustats) is used to gather vdev statistics at
1203 * the same time open the pool, without having to keep around the spa_t in some
1204 * ambiguous state.
1205 */
1206static int
1207spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t **config)
1208{
1209	spa_t *spa;
1210	int error;
1211	int loaded = B_FALSE;
1212	int locked = B_FALSE;
1213
1214	*spapp = NULL;
1215
1216	/*
1217	 * As disgusting as this is, we need to support recursive calls to this
1218	 * function because dsl_dir_open() is called during spa_load(), and ends
1219	 * up calling spa_open() again.  The real fix is to figure out how to
1220	 * avoid dsl_dir_open() calling this in the first place.
1221	 */
1222	if (mutex_owner(&spa_namespace_lock) != curthread) {
1223		mutex_enter(&spa_namespace_lock);
1224		locked = B_TRUE;
1225	}
1226
1227	if ((spa = spa_lookup(pool)) == NULL) {
1228		if (locked)
1229			mutex_exit(&spa_namespace_lock);
1230		return (ENOENT);
1231	}
1232	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
1233
1234		spa_activate(spa);
1235
1236		error = spa_load(spa, spa->spa_config, SPA_LOAD_OPEN, B_FALSE);
1237
1238		if (error == EBADF) {
1239			/*
1240			 * If vdev_validate() returns failure (indicated by
1241			 * EBADF), it indicates that one of the vdevs indicates
1242			 * that the pool has been exported or destroyed.  If
1243			 * this is the case, the config cache is out of sync and
1244			 * we should remove the pool from the namespace.
1245			 */
1246			zfs_post_ok(spa, NULL);
1247			spa_unload(spa);
1248			spa_deactivate(spa);
1249			spa_remove(spa);
1250			spa_config_sync();
1251			if (locked)
1252				mutex_exit(&spa_namespace_lock);
1253			return (ENOENT);
1254		}
1255
1256		if (error) {
1257			/*
1258			 * We can't open the pool, but we still have useful
1259			 * information: the state of each vdev after the
1260			 * attempted vdev_open().  Return this to the user.
1261			 */
1262			if (config != NULL && spa->spa_root_vdev != NULL) {
1263				spa_config_enter(spa, RW_READER, FTAG);
1264				*config = spa_config_generate(spa, NULL, -1ULL,
1265				    B_TRUE);
1266				spa_config_exit(spa, FTAG);
1267			}
1268			spa_unload(spa);
1269			spa_deactivate(spa);
1270			spa->spa_last_open_failed = B_TRUE;
1271			if (locked)
1272				mutex_exit(&spa_namespace_lock);
1273			*spapp = NULL;
1274			return (error);
1275		} else {
1276			zfs_post_ok(spa, NULL);
1277			spa->spa_last_open_failed = B_FALSE;
1278		}
1279
1280		loaded = B_TRUE;
1281	}
1282
1283	spa_open_ref(spa, tag);
1284
1285	/*
1286	 * If we just loaded the pool, resilver anything that's out of date.
1287	 */
1288	if (loaded && (spa_mode & FWRITE))
1289		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
1290
1291	if (locked)
1292		mutex_exit(&spa_namespace_lock);
1293
1294	*spapp = spa;
1295
1296	if (config != NULL) {
1297		spa_config_enter(spa, RW_READER, FTAG);
1298		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
1299		spa_config_exit(spa, FTAG);
1300	}
1301
1302	return (0);
1303}
1304
1305int
1306spa_open(const char *name, spa_t **spapp, void *tag)
1307{
1308	return (spa_open_common(name, spapp, tag, NULL));
1309}
1310
1311/*
1312 * Lookup the given spa_t, incrementing the inject count in the process,
1313 * preventing it from being exported or destroyed.
1314 */
1315spa_t *
1316spa_inject_addref(char *name)
1317{
1318	spa_t *spa;
1319
1320	mutex_enter(&spa_namespace_lock);
1321	if ((spa = spa_lookup(name)) == NULL) {
1322		mutex_exit(&spa_namespace_lock);
1323		return (NULL);
1324	}
1325	spa->spa_inject_ref++;
1326	mutex_exit(&spa_namespace_lock);
1327
1328	return (spa);
1329}
1330
1331void
1332spa_inject_delref(spa_t *spa)
1333{
1334	mutex_enter(&spa_namespace_lock);
1335	spa->spa_inject_ref--;
1336	mutex_exit(&spa_namespace_lock);
1337}
1338
1339static void
1340spa_add_spares(spa_t *spa, nvlist_t *config)
1341{
1342	nvlist_t **spares;
1343	uint_t i, nspares;
1344	nvlist_t *nvroot;
1345	uint64_t guid;
1346	vdev_stat_t *vs;
1347	uint_t vsc;
1348	uint64_t pool;
1349
1350	if (spa->spa_nspares == 0)
1351		return;
1352
1353	VERIFY(nvlist_lookup_nvlist(config,
1354	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1355	VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist,
1356	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1357	if (nspares != 0) {
1358		VERIFY(nvlist_add_nvlist_array(nvroot,
1359		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1360		VERIFY(nvlist_lookup_nvlist_array(nvroot,
1361		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1362
1363		/*
1364		 * Go through and find any spares which have since been
1365		 * repurposed as an active spare.  If this is the case, update
1366		 * their status appropriately.
1367		 */
1368		for (i = 0; i < nspares; i++) {
1369			VERIFY(nvlist_lookup_uint64(spares[i],
1370			    ZPOOL_CONFIG_GUID, &guid) == 0);
1371			if (spa_spare_exists(guid, &pool) && pool != 0ULL) {
1372				VERIFY(nvlist_lookup_uint64_array(
1373				    spares[i], ZPOOL_CONFIG_STATS,
1374				    (uint64_t **)&vs, &vsc) == 0);
1375				vs->vs_state = VDEV_STATE_CANT_OPEN;
1376				vs->vs_aux = VDEV_AUX_SPARED;
1377			}
1378		}
1379	}
1380}
1381
1382int
1383spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
1384{
1385	int error;
1386	spa_t *spa;
1387
1388	*config = NULL;
1389	error = spa_open_common(name, &spa, FTAG, config);
1390
1391	if (spa && *config != NULL) {
1392		VERIFY(nvlist_add_uint64(*config, ZPOOL_CONFIG_ERRCOUNT,
1393		    spa_get_errlog_size(spa)) == 0);
1394
1395		spa_add_spares(spa, *config);
1396	}
1397
1398	/*
1399	 * We want to get the alternate root even for faulted pools, so we cheat
1400	 * and call spa_lookup() directly.
1401	 */
1402	if (altroot) {
1403		if (spa == NULL) {
1404			mutex_enter(&spa_namespace_lock);
1405			spa = spa_lookup(name);
1406			if (spa)
1407				spa_altroot(spa, altroot, buflen);
1408			else
1409				altroot[0] = '\0';
1410			spa = NULL;
1411			mutex_exit(&spa_namespace_lock);
1412		} else {
1413			spa_altroot(spa, altroot, buflen);
1414		}
1415	}
1416
1417	if (spa != NULL)
1418		spa_close(spa, FTAG);
1419
1420	return (error);
1421}
1422
1423/*
1424 * Validate that the 'spares' array is well formed.  We must have an array of
1425 * nvlists, each which describes a valid leaf vdev.  If this is an import (mode
1426 * is VDEV_ALLOC_SPARE), then we allow corrupted spares to be specified, as long
1427 * as they are well-formed.
1428 */
1429static int
1430spa_validate_spares(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
1431{
1432	nvlist_t **spares;
1433	uint_t i, nspares;
1434	vdev_t *vd;
1435	int error;
1436
1437	/*
1438	 * It's acceptable to have no spares specified.
1439	 */
1440	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1441	    &spares, &nspares) != 0)
1442		return (0);
1443
1444	if (nspares == 0)
1445		return (EINVAL);
1446
1447	/*
1448	 * Make sure the pool is formatted with a version that supports hot
1449	 * spares.
1450	 */
1451	if (spa_version(spa) < SPA_VERSION_SPARES)
1452		return (ENOTSUP);
1453
1454	/*
1455	 * Set the pending spare list so we correctly handle device in-use
1456	 * checking.
1457	 */
1458	spa->spa_pending_spares = spares;
1459	spa->spa_pending_nspares = nspares;
1460
1461	for (i = 0; i < nspares; i++) {
1462		if ((error = spa_config_parse(spa, &vd, spares[i], NULL, 0,
1463		    mode)) != 0)
1464			goto out;
1465
1466		if (!vd->vdev_ops->vdev_op_leaf) {
1467			vdev_free(vd);
1468			error = EINVAL;
1469			goto out;
1470		}
1471
1472		vd->vdev_top = vd;
1473
1474		if ((error = vdev_open(vd)) == 0 &&
1475		    (error = vdev_label_init(vd, crtxg,
1476		    VDEV_LABEL_SPARE)) == 0) {
1477			VERIFY(nvlist_add_uint64(spares[i], ZPOOL_CONFIG_GUID,
1478			    vd->vdev_guid) == 0);
1479		}
1480
1481		vdev_free(vd);
1482
1483		if (error && mode != VDEV_ALLOC_SPARE)
1484			goto out;
1485		else
1486			error = 0;
1487	}
1488
1489out:
1490	spa->spa_pending_spares = NULL;
1491	spa->spa_pending_nspares = 0;
1492	return (error);
1493}
1494
1495/*
1496 * Pool Creation
1497 */
1498int
1499spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
1500    const char *history_str)
1501{
1502	spa_t *spa;
1503	char *altroot = NULL;
1504	vdev_t *rvd;
1505	dsl_pool_t *dp;
1506	dmu_tx_t *tx;
1507	int c, error = 0;
1508	uint64_t txg = TXG_INITIAL;
1509	nvlist_t **spares;
1510	uint_t nspares;
1511	uint64_t version;
1512
1513	/*
1514	 * If this pool already exists, return failure.
1515	 */
1516	mutex_enter(&spa_namespace_lock);
1517	if (spa_lookup(pool) != NULL) {
1518		mutex_exit(&spa_namespace_lock);
1519		return (EEXIST);
1520	}
1521
1522	/*
1523	 * Allocate a new spa_t structure.
1524	 */
1525	(void) nvlist_lookup_string(props,
1526	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
1527	spa = spa_add(pool, altroot);
1528	spa_activate(spa);
1529
1530	spa->spa_uberblock.ub_txg = txg - 1;
1531
1532	if (props && (error = spa_prop_validate(spa, props))) {
1533		spa_unload(spa);
1534		spa_deactivate(spa);
1535		spa_remove(spa);
1536		return (error);
1537	}
1538
1539	if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
1540	    &version) != 0)
1541		version = SPA_VERSION;
1542	ASSERT(version <= SPA_VERSION);
1543	spa->spa_uberblock.ub_version = version;
1544	spa->spa_ubsync = spa->spa_uberblock;
1545
1546	/*
1547	 * Create the root vdev.
1548	 */
1549	spa_config_enter(spa, RW_WRITER, FTAG);
1550
1551	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
1552
1553	ASSERT(error != 0 || rvd != NULL);
1554	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
1555
1556	if (error == 0 && rvd->vdev_children == 0)
1557		error = EINVAL;
1558
1559	if (error == 0 &&
1560	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
1561	    (error = spa_validate_spares(spa, nvroot, txg,
1562	    VDEV_ALLOC_ADD)) == 0) {
1563		for (c = 0; c < rvd->vdev_children; c++)
1564			vdev_init(rvd->vdev_child[c], txg);
1565		vdev_config_dirty(rvd);
1566	}
1567
1568	spa_config_exit(spa, FTAG);
1569
1570	if (error != 0) {
1571		spa_unload(spa);
1572		spa_deactivate(spa);
1573		spa_remove(spa);
1574		mutex_exit(&spa_namespace_lock);
1575		return (error);
1576	}
1577
1578	/*
1579	 * Get the list of spares, if specified.
1580	 */
1581	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1582	    &spares, &nspares) == 0) {
1583		VERIFY(nvlist_alloc(&spa->spa_sparelist, NV_UNIQUE_NAME,
1584		    KM_SLEEP) == 0);
1585		VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
1586		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1587		spa_config_enter(spa, RW_WRITER, FTAG);
1588		spa_load_spares(spa);
1589		spa_config_exit(spa, FTAG);
1590		spa->spa_sync_spares = B_TRUE;
1591	}
1592
1593	spa->spa_dsl_pool = dp = dsl_pool_create(spa, txg);
1594	spa->spa_meta_objset = dp->dp_meta_objset;
1595
1596	tx = dmu_tx_create_assigned(dp, txg);
1597
1598	/*
1599	 * Create the pool config object.
1600	 */
1601	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
1602	    DMU_OT_PACKED_NVLIST, 1 << 14,
1603	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
1604
1605	if (zap_add(spa->spa_meta_objset,
1606	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
1607	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
1608		cmn_err(CE_PANIC, "failed to add pool config");
1609	}
1610
1611	/* Newly created pools with the right version are always deflated. */
1612	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
1613		spa->spa_deflate = TRUE;
1614		if (zap_add(spa->spa_meta_objset,
1615		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
1616		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
1617			cmn_err(CE_PANIC, "failed to add deflate");
1618		}
1619	}
1620
1621	/*
1622	 * Create the deferred-free bplist object.  Turn off compression
1623	 * because sync-to-convergence takes longer if the blocksize
1624	 * keeps changing.
1625	 */
1626	spa->spa_sync_bplist_obj = bplist_create(spa->spa_meta_objset,
1627	    1 << 14, tx);
1628	dmu_object_set_compress(spa->spa_meta_objset, spa->spa_sync_bplist_obj,
1629	    ZIO_COMPRESS_OFF, tx);
1630
1631	if (zap_add(spa->spa_meta_objset,
1632	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
1633	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj, tx) != 0) {
1634		cmn_err(CE_PANIC, "failed to add bplist");
1635	}
1636
1637	/*
1638	 * Create the pool's history object.
1639	 */
1640	if (version >= SPA_VERSION_ZPOOL_HISTORY)
1641		spa_history_create_obj(spa, tx);
1642
1643	/*
1644	 * Set pool properties.
1645	 */
1646	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
1647	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
1648	spa->spa_temporary = zpool_prop_default_numeric(ZPOOL_PROP_TEMPORARY);
1649	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
1650	if (props)
1651		spa_sync_props(spa, props, CRED(), tx);
1652
1653	dmu_tx_commit(tx);
1654
1655	spa->spa_sync_on = B_TRUE;
1656	txg_sync_start(spa->spa_dsl_pool);
1657
1658	/*
1659	 * We explicitly wait for the first transaction to complete so that our
1660	 * bean counters are appropriately updated.
1661	 */
1662	txg_wait_synced(spa->spa_dsl_pool, txg);
1663
1664	spa_config_sync();
1665
1666	if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
1667		(void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
1668
1669	mutex_exit(&spa_namespace_lock);
1670
1671	return (0);
1672}
1673
1674/*
1675 * Import the given pool into the system.  We set up the necessary spa_t and
1676 * then call spa_load() to do the dirty work.
1677 */
1678int
1679spa_import(const char *pool, nvlist_t *config, nvlist_t *props)
1680{
1681	spa_t *spa;
1682	char *altroot = NULL;
1683	int error;
1684	nvlist_t *nvroot;
1685	nvlist_t **spares;
1686	uint_t nspares;
1687
1688	/*
1689	 * If a pool with this name exists, return failure.
1690	 */
1691	mutex_enter(&spa_namespace_lock);
1692	if (spa_lookup(pool) != NULL) {
1693		mutex_exit(&spa_namespace_lock);
1694		return (EEXIST);
1695	}
1696
1697	/*
1698	 * Create and initialize the spa structure.
1699	 */
1700	(void) nvlist_lookup_string(props,
1701	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
1702	spa = spa_add(pool, altroot);
1703	spa_activate(spa);
1704
1705	/*
1706	 * Pass off the heavy lifting to spa_load().
1707	 * Pass TRUE for mosconfig because the user-supplied config
1708	 * is actually the one to trust when doing an import.
1709	 */
1710	error = spa_load(spa, config, SPA_LOAD_IMPORT, B_TRUE);
1711
1712	spa_config_enter(spa, RW_WRITER, FTAG);
1713	/*
1714	 * Toss any existing sparelist, as it doesn't have any validity anymore,
1715	 * and conflicts with spa_has_spare().
1716	 */
1717	if (spa->spa_sparelist) {
1718		nvlist_free(spa->spa_sparelist);
1719		spa->spa_sparelist = NULL;
1720		spa_load_spares(spa);
1721	}
1722
1723	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
1724	    &nvroot) == 0);
1725	if (error == 0) {
1726		error = spa_validate_spares(spa, nvroot, -1ULL,
1727		    VDEV_ALLOC_SPARE);
1728	}
1729	spa_config_exit(spa, FTAG);
1730
1731	if (error != 0 || (props && (error = spa_prop_set(spa, props)))) {
1732		spa_unload(spa);
1733		spa_deactivate(spa);
1734		spa_remove(spa);
1735		mutex_exit(&spa_namespace_lock);
1736		return (error);
1737	}
1738
1739	/*
1740	 * Override any spares as specified by the user, as these may have
1741	 * correct device names/devids, etc.
1742	 */
1743	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1744	    &spares, &nspares) == 0) {
1745		if (spa->spa_sparelist)
1746			VERIFY(nvlist_remove(spa->spa_sparelist,
1747			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
1748		else
1749			VERIFY(nvlist_alloc(&spa->spa_sparelist,
1750			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
1751		VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
1752		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1753		spa_config_enter(spa, RW_WRITER, FTAG);
1754		spa_load_spares(spa);
1755		spa_config_exit(spa, FTAG);
1756		spa->spa_sync_spares = B_TRUE;
1757	}
1758
1759	/*
1760	 * Update the config cache to include the newly-imported pool.
1761	 */
1762	if (spa_mode & FWRITE)
1763		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
1764
1765	/*
1766	 * Resilver anything that's out of date.
1767	 */
1768	if (spa_mode & FWRITE)
1769		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
1770
1771	mutex_exit(&spa_namespace_lock);
1772
1773	return (0);
1774}
1775
1776/*
1777 * This (illegal) pool name is used when temporarily importing a spa_t in order
1778 * to get the vdev stats associated with the imported devices.
1779 */
1780#define	TRYIMPORT_NAME	"$import"
1781
1782nvlist_t *
1783spa_tryimport(nvlist_t *tryconfig)
1784{
1785	nvlist_t *config = NULL;
1786	char *poolname;
1787	spa_t *spa;
1788	uint64_t state;
1789
1790	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
1791		return (NULL);
1792
1793	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
1794		return (NULL);
1795
1796	/*
1797	 * Create and initialize the spa structure.
1798	 */
1799	mutex_enter(&spa_namespace_lock);
1800	spa = spa_add(TRYIMPORT_NAME, NULL);
1801	spa_activate(spa);
1802
1803	/*
1804	 * Pass off the heavy lifting to spa_load().
1805	 * Pass TRUE for mosconfig because the user-supplied config
1806	 * is actually the one to trust when doing an import.
1807	 */
1808	(void) spa_load(spa, tryconfig, SPA_LOAD_TRYIMPORT, B_TRUE);
1809
1810	/*
1811	 * If 'tryconfig' was at least parsable, return the current config.
1812	 */
1813	if (spa->spa_root_vdev != NULL) {
1814		spa_config_enter(spa, RW_READER, FTAG);
1815		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
1816		spa_config_exit(spa, FTAG);
1817		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
1818		    poolname) == 0);
1819		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1820		    state) == 0);
1821		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
1822		    spa->spa_uberblock.ub_timestamp) == 0);
1823
1824		/*
1825		 * Add the list of hot spares.
1826		 */
1827		spa_add_spares(spa, config);
1828	}
1829
1830	spa_unload(spa);
1831	spa_deactivate(spa);
1832	spa_remove(spa);
1833	mutex_exit(&spa_namespace_lock);
1834
1835	return (config);
1836}
1837
1838/*
1839 * Pool export/destroy
1840 *
1841 * The act of destroying or exporting a pool is very simple.  We make sure there
1842 * is no more pending I/O and any references to the pool are gone.  Then, we
1843 * update the pool state and sync all the labels to disk, removing the
1844 * configuration from the cache afterwards.
1845 */
1846static int
1847spa_export_common(char *pool, int new_state, nvlist_t **oldconfig)
1848{
1849	spa_t *spa;
1850
1851	if (oldconfig)
1852		*oldconfig = NULL;
1853
1854	if (!(spa_mode & FWRITE))
1855		return (EROFS);
1856
1857	mutex_enter(&spa_namespace_lock);
1858	if ((spa = spa_lookup(pool)) == NULL) {
1859		mutex_exit(&spa_namespace_lock);
1860		return (ENOENT);
1861	}
1862
1863	/*
1864	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
1865	 * reacquire the namespace lock, and see if we can export.
1866	 */
1867	spa_open_ref(spa, FTAG);
1868	mutex_exit(&spa_namespace_lock);
1869	spa_async_suspend(spa);
1870	mutex_enter(&spa_namespace_lock);
1871	spa_close(spa, FTAG);
1872
1873	/*
1874	 * The pool will be in core if it's openable,
1875	 * in which case we can modify its state.
1876	 */
1877	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
1878		/*
1879		 * Objsets may be open only because they're dirty, so we
1880		 * have to force it to sync before checking spa_refcnt.
1881		 */
1882		spa_scrub_suspend(spa);
1883		txg_wait_synced(spa->spa_dsl_pool, 0);
1884
1885		/*
1886		 * A pool cannot be exported or destroyed if there are active
1887		 * references.  If we are resetting a pool, allow references by
1888		 * fault injection handlers.
1889		 */
1890		if (!spa_refcount_zero(spa) ||
1891		    (spa->spa_inject_ref != 0 &&
1892		    new_state != POOL_STATE_UNINITIALIZED)) {
1893			spa_scrub_resume(spa);
1894			spa_async_resume(spa);
1895			mutex_exit(&spa_namespace_lock);
1896			return (EBUSY);
1897		}
1898
1899		spa_scrub_resume(spa);
1900		VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0);
1901
1902		/*
1903		 * We want this to be reflected on every label,
1904		 * so mark them all dirty.  spa_unload() will do the
1905		 * final sync that pushes these changes out.
1906		 */
1907		if (new_state != POOL_STATE_UNINITIALIZED) {
1908			spa_config_enter(spa, RW_WRITER, FTAG);
1909			spa->spa_state = new_state;
1910			spa->spa_final_txg = spa_last_synced_txg(spa) + 1;
1911			vdev_config_dirty(spa->spa_root_vdev);
1912			spa_config_exit(spa, FTAG);
1913		}
1914	}
1915
1916	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
1917
1918	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
1919		spa_unload(spa);
1920		spa_deactivate(spa);
1921	}
1922
1923	if (oldconfig && spa->spa_config)
1924		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
1925
1926	if (new_state != POOL_STATE_UNINITIALIZED) {
1927		spa_remove(spa);
1928		spa_config_sync();
1929	}
1930	mutex_exit(&spa_namespace_lock);
1931
1932	return (0);
1933}
1934
1935/*
1936 * Destroy a storage pool.
1937 */
1938int
1939spa_destroy(char *pool)
1940{
1941	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL));
1942}
1943
1944/*
1945 * Export a storage pool.
1946 */
1947int
1948spa_export(char *pool, nvlist_t **oldconfig)
1949{
1950	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig));
1951}
1952
1953/*
1954 * Similar to spa_export(), this unloads the spa_t without actually removing it
1955 * from the namespace in any way.
1956 */
1957int
1958spa_reset(char *pool)
1959{
1960	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL));
1961}
1962
1963
1964/*
1965 * ==========================================================================
1966 * Device manipulation
1967 * ==========================================================================
1968 */
1969
1970/*
1971 * Add a device to a storage pool.
1972 */
1973int
1974spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
1975{
1976	uint64_t txg;
1977	int c, error;
1978	vdev_t *rvd = spa->spa_root_vdev;
1979	vdev_t *vd, *tvd;
1980	nvlist_t **spares;
1981	uint_t i, nspares;
1982
1983	txg = spa_vdev_enter(spa);
1984
1985	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
1986	    VDEV_ALLOC_ADD)) != 0)
1987		return (spa_vdev_exit(spa, NULL, txg, error));
1988
1989	spa->spa_pending_vdev = vd;
1990
1991	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1992	    &spares, &nspares) != 0)
1993		nspares = 0;
1994
1995	if (vd->vdev_children == 0 && nspares == 0) {
1996		spa->spa_pending_vdev = NULL;
1997		return (spa_vdev_exit(spa, vd, txg, EINVAL));
1998	}
1999
2000	if (vd->vdev_children != 0) {
2001		if ((error = vdev_create(vd, txg, B_FALSE)) != 0) {
2002			spa->spa_pending_vdev = NULL;
2003			return (spa_vdev_exit(spa, vd, txg, error));
2004		}
2005	}
2006
2007	/*
2008	 * We must validate the spares after checking the children.  Otherwise,
2009	 * vdev_inuse() will blindly overwrite the spare.
2010	 */
2011	if ((error = spa_validate_spares(spa, nvroot, txg,
2012	    VDEV_ALLOC_ADD)) != 0) {
2013		spa->spa_pending_vdev = NULL;
2014		return (spa_vdev_exit(spa, vd, txg, error));
2015	}
2016
2017	spa->spa_pending_vdev = NULL;
2018
2019	/*
2020	 * Transfer each new top-level vdev from vd to rvd.
2021	 */
2022	for (c = 0; c < vd->vdev_children; c++) {
2023		tvd = vd->vdev_child[c];
2024		vdev_remove_child(vd, tvd);
2025		tvd->vdev_id = rvd->vdev_children;
2026		vdev_add_child(rvd, tvd);
2027		vdev_config_dirty(tvd);
2028	}
2029
2030	if (nspares != 0) {
2031		if (spa->spa_sparelist != NULL) {
2032			nvlist_t **oldspares;
2033			uint_t oldnspares;
2034			nvlist_t **newspares;
2035
2036			VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist,
2037			    ZPOOL_CONFIG_SPARES, &oldspares, &oldnspares) == 0);
2038
2039			newspares = kmem_alloc(sizeof (void *) *
2040			    (nspares + oldnspares), KM_SLEEP);
2041			for (i = 0; i < oldnspares; i++)
2042				VERIFY(nvlist_dup(oldspares[i],
2043				    &newspares[i], KM_SLEEP) == 0);
2044			for (i = 0; i < nspares; i++)
2045				VERIFY(nvlist_dup(spares[i],
2046				    &newspares[i + oldnspares],
2047				    KM_SLEEP) == 0);
2048
2049			VERIFY(nvlist_remove(spa->spa_sparelist,
2050			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
2051
2052			VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
2053			    ZPOOL_CONFIG_SPARES, newspares,
2054			    nspares + oldnspares) == 0);
2055			for (i = 0; i < oldnspares + nspares; i++)
2056				nvlist_free(newspares[i]);
2057			kmem_free(newspares, (oldnspares + nspares) *
2058			    sizeof (void *));
2059		} else {
2060			VERIFY(nvlist_alloc(&spa->spa_sparelist,
2061			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2062			VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist,
2063			    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2064		}
2065
2066		spa_load_spares(spa);
2067		spa->spa_sync_spares = B_TRUE;
2068	}
2069
2070	/*
2071	 * We have to be careful when adding new vdevs to an existing pool.
2072	 * If other threads start allocating from these vdevs before we
2073	 * sync the config cache, and we lose power, then upon reboot we may
2074	 * fail to open the pool because there are DVAs that the config cache
2075	 * can't translate.  Therefore, we first add the vdevs without
2076	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
2077	 * and then let spa_config_update() initialize the new metaslabs.
2078	 *
2079	 * spa_load() checks for added-but-not-initialized vdevs, so that
2080	 * if we lose power at any point in this sequence, the remaining
2081	 * steps will be completed the next time we load the pool.
2082	 */
2083	(void) spa_vdev_exit(spa, vd, txg, 0);
2084
2085	mutex_enter(&spa_namespace_lock);
2086	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
2087	mutex_exit(&spa_namespace_lock);
2088
2089	return (0);
2090}
2091
2092/*
2093 * Attach a device to a mirror.  The arguments are the path to any device
2094 * in the mirror, and the nvroot for the new device.  If the path specifies
2095 * a device that is not mirrored, we automatically insert the mirror vdev.
2096 *
2097 * If 'replacing' is specified, the new device is intended to replace the
2098 * existing device; in this case the two devices are made into their own
2099 * mirror using the 'replacing' vdev, which is functionally identical to
2100 * the mirror vdev (it actually reuses all the same ops) but has a few
2101 * extra rules: you can't attach to it after it's been created, and upon
2102 * completion of resilvering, the first disk (the one being replaced)
2103 * is automatically detached.
2104 */
2105int
2106spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
2107{
2108	uint64_t txg, open_txg;
2109	int error;
2110	vdev_t *rvd = spa->spa_root_vdev;
2111	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
2112	vdev_ops_t *pvops;
2113	int is_log;
2114
2115	txg = spa_vdev_enter(spa);
2116
2117	oldvd = vdev_lookup_by_guid(rvd, guid);
2118
2119	if (oldvd == NULL)
2120		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
2121
2122	if (!oldvd->vdev_ops->vdev_op_leaf)
2123		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
2124
2125	pvd = oldvd->vdev_parent;
2126
2127	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
2128	    VDEV_ALLOC_ADD)) != 0)
2129		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
2130
2131	if (newrootvd->vdev_children != 1)
2132		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
2133
2134	newvd = newrootvd->vdev_child[0];
2135
2136	if (!newvd->vdev_ops->vdev_op_leaf)
2137		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
2138
2139	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
2140		return (spa_vdev_exit(spa, newrootvd, txg, error));
2141
2142	/*
2143	 * Spares can't replace logs
2144	 */
2145	is_log = oldvd->vdev_islog;
2146	if (is_log && newvd->vdev_isspare)
2147		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2148
2149	if (!replacing) {
2150		/*
2151		 * For attach, the only allowable parent is a mirror or the root
2152		 * vdev.
2153		 */
2154		if (pvd->vdev_ops != &vdev_mirror_ops &&
2155		    pvd->vdev_ops != &vdev_root_ops)
2156			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2157
2158		pvops = &vdev_mirror_ops;
2159	} else {
2160		/*
2161		 * Active hot spares can only be replaced by inactive hot
2162		 * spares.
2163		 */
2164		if (pvd->vdev_ops == &vdev_spare_ops &&
2165		    pvd->vdev_child[1] == oldvd &&
2166		    !spa_has_spare(spa, newvd->vdev_guid))
2167			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2168
2169		/*
2170		 * If the source is a hot spare, and the parent isn't already a
2171		 * spare, then we want to create a new hot spare.  Otherwise, we
2172		 * want to create a replacing vdev.  The user is not allowed to
2173		 * attach to a spared vdev child unless the 'isspare' state is
2174		 * the same (spare replaces spare, non-spare replaces
2175		 * non-spare).
2176		 */
2177		if (pvd->vdev_ops == &vdev_replacing_ops)
2178			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2179		else if (pvd->vdev_ops == &vdev_spare_ops &&
2180		    newvd->vdev_isspare != oldvd->vdev_isspare)
2181			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2182		else if (pvd->vdev_ops != &vdev_spare_ops &&
2183		    newvd->vdev_isspare)
2184			pvops = &vdev_spare_ops;
2185		else
2186			pvops = &vdev_replacing_ops;
2187	}
2188
2189	/*
2190	 * Compare the new device size with the replaceable/attachable
2191	 * device size.
2192	 */
2193	if (newvd->vdev_psize < vdev_get_rsize(oldvd))
2194		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
2195
2196	/*
2197	 * The new device cannot have a higher alignment requirement
2198	 * than the top-level vdev.
2199	 */
2200	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
2201		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
2202
2203	/*
2204	 * If this is an in-place replacement, update oldvd's path and devid
2205	 * to make it distinguishable from newvd, and unopenable from now on.
2206	 */
2207	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
2208		spa_strfree(oldvd->vdev_path);
2209		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
2210		    KM_SLEEP);
2211		(void) sprintf(oldvd->vdev_path, "%s/%s",
2212		    newvd->vdev_path, "old");
2213		if (oldvd->vdev_devid != NULL) {
2214			spa_strfree(oldvd->vdev_devid);
2215			oldvd->vdev_devid = NULL;
2216		}
2217	}
2218
2219	/*
2220	 * If the parent is not a mirror, or if we're replacing, insert the new
2221	 * mirror/replacing/spare vdev above oldvd.
2222	 */
2223	if (pvd->vdev_ops != pvops)
2224		pvd = vdev_add_parent(oldvd, pvops);
2225
2226	ASSERT(pvd->vdev_top->vdev_parent == rvd);
2227	ASSERT(pvd->vdev_ops == pvops);
2228	ASSERT(oldvd->vdev_parent == pvd);
2229
2230	/*
2231	 * Extract the new device from its root and add it to pvd.
2232	 */
2233	vdev_remove_child(newrootvd, newvd);
2234	newvd->vdev_id = pvd->vdev_children;
2235	vdev_add_child(pvd, newvd);
2236
2237	/*
2238	 * If newvd is smaller than oldvd, but larger than its rsize,
2239	 * the addition of newvd may have decreased our parent's asize.
2240	 */
2241	pvd->vdev_asize = MIN(pvd->vdev_asize, newvd->vdev_asize);
2242
2243	tvd = newvd->vdev_top;
2244	ASSERT(pvd->vdev_top == tvd);
2245	ASSERT(tvd->vdev_parent == rvd);
2246
2247	vdev_config_dirty(tvd);
2248
2249	/*
2250	 * Set newvd's DTL to [TXG_INITIAL, open_txg].  It will propagate
2251	 * upward when spa_vdev_exit() calls vdev_dtl_reassess().
2252	 */
2253	open_txg = txg + TXG_CONCURRENT_STATES - 1;
2254
2255	mutex_enter(&newvd->vdev_dtl_lock);
2256	space_map_add(&newvd->vdev_dtl_map, TXG_INITIAL,
2257	    open_txg - TXG_INITIAL + 1);
2258	mutex_exit(&newvd->vdev_dtl_lock);
2259
2260	if (newvd->vdev_isspare)
2261		spa_spare_activate(newvd);
2262
2263	/*
2264	 * Mark newvd's DTL dirty in this txg.
2265	 */
2266	vdev_dirty(tvd, VDD_DTL, newvd, txg);
2267
2268	(void) spa_vdev_exit(spa, newrootvd, open_txg, 0);
2269
2270	/*
2271	 * Kick off a resilver to update newvd.  We need to grab the namespace
2272	 * lock because spa_scrub() needs to post a sysevent with the pool name.
2273	 */
2274	mutex_enter(&spa_namespace_lock);
2275	VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
2276	mutex_exit(&spa_namespace_lock);
2277
2278	return (0);
2279}
2280
2281/*
2282 * Detach a device from a mirror or replacing vdev.
2283 * If 'replace_done' is specified, only detach if the parent
2284 * is a replacing vdev.
2285 */
2286int
2287spa_vdev_detach(spa_t *spa, uint64_t guid, int replace_done)
2288{
2289	uint64_t txg;
2290	int c, t, error;
2291	vdev_t *rvd = spa->spa_root_vdev;
2292	vdev_t *vd, *pvd, *cvd, *tvd;
2293	boolean_t unspare = B_FALSE;
2294	uint64_t unspare_guid;
2295
2296	txg = spa_vdev_enter(spa);
2297
2298	vd = vdev_lookup_by_guid(rvd, guid);
2299
2300	if (vd == NULL)
2301		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
2302
2303	if (!vd->vdev_ops->vdev_op_leaf)
2304		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
2305
2306	pvd = vd->vdev_parent;
2307
2308	/*
2309	 * If replace_done is specified, only remove this device if it's
2310	 * the first child of a replacing vdev.  For the 'spare' vdev, either
2311	 * disk can be removed.
2312	 */
2313	if (replace_done) {
2314		if (pvd->vdev_ops == &vdev_replacing_ops) {
2315			if (vd->vdev_id != 0)
2316				return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
2317		} else if (pvd->vdev_ops != &vdev_spare_ops) {
2318			return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
2319		}
2320	}
2321
2322	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
2323	    spa_version(spa) >= SPA_VERSION_SPARES);
2324
2325	/*
2326	 * Only mirror, replacing, and spare vdevs support detach.
2327	 */
2328	if (pvd->vdev_ops != &vdev_replacing_ops &&
2329	    pvd->vdev_ops != &vdev_mirror_ops &&
2330	    pvd->vdev_ops != &vdev_spare_ops)
2331		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
2332
2333	/*
2334	 * If there's only one replica, you can't detach it.
2335	 */
2336	if (pvd->vdev_children <= 1)
2337		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
2338
2339	/*
2340	 * If all siblings have non-empty DTLs, this device may have the only
2341	 * valid copy of the data, which means we cannot safely detach it.
2342	 *
2343	 * XXX -- as in the vdev_offline() case, we really want a more
2344	 * precise DTL check.
2345	 */
2346	for (c = 0; c < pvd->vdev_children; c++) {
2347		uint64_t dirty;
2348
2349		cvd = pvd->vdev_child[c];
2350		if (cvd == vd)
2351			continue;
2352		if (vdev_is_dead(cvd))
2353			continue;
2354		mutex_enter(&cvd->vdev_dtl_lock);
2355		dirty = cvd->vdev_dtl_map.sm_space |
2356		    cvd->vdev_dtl_scrub.sm_space;
2357		mutex_exit(&cvd->vdev_dtl_lock);
2358		if (!dirty)
2359			break;
2360	}
2361
2362	/*
2363	 * If we are a replacing or spare vdev, then we can always detach the
2364	 * latter child, as that is how one cancels the operation.
2365	 */
2366	if ((pvd->vdev_ops == &vdev_mirror_ops || vd->vdev_id != 1) &&
2367	    c == pvd->vdev_children)
2368		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
2369
2370	/*
2371	 * If we are detaching the original disk from a spare, then it implies
2372	 * that the spare should become a real disk, and be removed from the
2373	 * active spare list for the pool.
2374	 */
2375	if (pvd->vdev_ops == &vdev_spare_ops &&
2376	    vd->vdev_id == 0)
2377		unspare = B_TRUE;
2378
2379	/*
2380	 * Erase the disk labels so the disk can be used for other things.
2381	 * This must be done after all other error cases are handled,
2382	 * but before we disembowel vd (so we can still do I/O to it).
2383	 * But if we can't do it, don't treat the error as fatal --
2384	 * it may be that the unwritability of the disk is the reason
2385	 * it's being detached!
2386	 */
2387	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
2388
2389	/*
2390	 * Remove vd from its parent and compact the parent's children.
2391	 */
2392	vdev_remove_child(pvd, vd);
2393	vdev_compact_children(pvd);
2394
2395	/*
2396	 * Remember one of the remaining children so we can get tvd below.
2397	 */
2398	cvd = pvd->vdev_child[0];
2399
2400	/*
2401	 * If we need to remove the remaining child from the list of hot spares,
2402	 * do it now, marking the vdev as no longer a spare in the process.  We
2403	 * must do this before vdev_remove_parent(), because that can change the
2404	 * GUID if it creates a new toplevel GUID.
2405	 */
2406	if (unspare) {
2407		ASSERT(cvd->vdev_isspare);
2408		spa_spare_remove(cvd);
2409		unspare_guid = cvd->vdev_guid;
2410	}
2411
2412	/*
2413	 * If the parent mirror/replacing vdev only has one child,
2414	 * the parent is no longer needed.  Remove it from the tree.
2415	 */
2416	if (pvd->vdev_children == 1)
2417		vdev_remove_parent(cvd);
2418
2419	/*
2420	 * We don't set tvd until now because the parent we just removed
2421	 * may have been the previous top-level vdev.
2422	 */
2423	tvd = cvd->vdev_top;
2424	ASSERT(tvd->vdev_parent == rvd);
2425
2426	/*
2427	 * Reevaluate the parent vdev state.
2428	 */
2429	vdev_propagate_state(cvd);
2430
2431	/*
2432	 * If the device we just detached was smaller than the others, it may be
2433	 * possible to add metaslabs (i.e. grow the pool).  vdev_metaslab_init()
2434	 * can't fail because the existing metaslabs are already in core, so
2435	 * there's nothing to read from disk.
2436	 */
2437	VERIFY(vdev_metaslab_init(tvd, txg) == 0);
2438
2439	vdev_config_dirty(tvd);
2440
2441	/*
2442	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
2443	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
2444	 * But first make sure we're not on any *other* txg's DTL list, to
2445	 * prevent vd from being accessed after it's freed.
2446	 */
2447	for (t = 0; t < TXG_SIZE; t++)
2448		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
2449	vd->vdev_detached = B_TRUE;
2450	vdev_dirty(tvd, VDD_DTL, vd, txg);
2451
2452	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
2453
2454	error = spa_vdev_exit(spa, vd, txg, 0);
2455
2456	/*
2457	 * If this was the removal of the original device in a hot spare vdev,
2458	 * then we want to go through and remove the device from the hot spare
2459	 * list of every other pool.
2460	 */
2461	if (unspare) {
2462		spa = NULL;
2463		mutex_enter(&spa_namespace_lock);
2464		while ((spa = spa_next(spa)) != NULL) {
2465			if (spa->spa_state != POOL_STATE_ACTIVE)
2466				continue;
2467
2468			(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
2469		}
2470		mutex_exit(&spa_namespace_lock);
2471	}
2472
2473	return (error);
2474}
2475
2476/*
2477 * Remove a device from the pool.  Currently, this supports removing only hot
2478 * spares.
2479 */
2480int
2481spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
2482{
2483	vdev_t *vd;
2484	nvlist_t **spares, *nv, **newspares;
2485	uint_t i, j, nspares;
2486	int ret = 0;
2487
2488	spa_config_enter(spa, RW_WRITER, FTAG);
2489
2490	vd = spa_lookup_by_guid(spa, guid);
2491
2492	nv = NULL;
2493	if (spa->spa_spares != NULL &&
2494	    nvlist_lookup_nvlist_array(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
2495	    &spares, &nspares) == 0) {
2496		for (i = 0; i < nspares; i++) {
2497			uint64_t theguid;
2498
2499			VERIFY(nvlist_lookup_uint64(spares[i],
2500			    ZPOOL_CONFIG_GUID, &theguid) == 0);
2501			if (theguid == guid) {
2502				nv = spares[i];
2503				break;
2504			}
2505		}
2506	}
2507
2508	/*
2509	 * We only support removing a hot spare, and only if it's not currently
2510	 * in use in this pool.
2511	 */
2512	if (nv == NULL && vd == NULL) {
2513		ret = ENOENT;
2514		goto out;
2515	}
2516
2517	if (nv == NULL && vd != NULL) {
2518		ret = ENOTSUP;
2519		goto out;
2520	}
2521
2522	if (!unspare && nv != NULL && vd != NULL) {
2523		ret = EBUSY;
2524		goto out;
2525	}
2526
2527	if (nspares == 1) {
2528		newspares = NULL;
2529	} else {
2530		newspares = kmem_alloc((nspares - 1) * sizeof (void *),
2531		    KM_SLEEP);
2532		for (i = 0, j = 0; i < nspares; i++) {
2533			if (spares[i] != nv)
2534				VERIFY(nvlist_dup(spares[i],
2535				    &newspares[j++], KM_SLEEP) == 0);
2536		}
2537	}
2538
2539	VERIFY(nvlist_remove(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
2540	    DATA_TYPE_NVLIST_ARRAY) == 0);
2541	VERIFY(nvlist_add_nvlist_array(spa->spa_sparelist, ZPOOL_CONFIG_SPARES,
2542	    newspares, nspares - 1) == 0);
2543	for (i = 0; i < nspares - 1; i++)
2544		nvlist_free(newspares[i]);
2545	kmem_free(newspares, (nspares - 1) * sizeof (void *));
2546	spa_load_spares(spa);
2547	spa->spa_sync_spares = B_TRUE;
2548
2549out:
2550	spa_config_exit(spa, FTAG);
2551
2552	return (ret);
2553}
2554
2555/*
2556 * Find any device that's done replacing, or a vdev marked 'unspare' that's
2557 * current spared, so we can detach it.
2558 */
2559static vdev_t *
2560spa_vdev_resilver_done_hunt(vdev_t *vd)
2561{
2562	vdev_t *newvd, *oldvd;
2563	int c;
2564
2565	for (c = 0; c < vd->vdev_children; c++) {
2566		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
2567		if (oldvd != NULL)
2568			return (oldvd);
2569	}
2570
2571	/*
2572	 * Check for a completed replacement.
2573	 */
2574	if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
2575		oldvd = vd->vdev_child[0];
2576		newvd = vd->vdev_child[1];
2577
2578		mutex_enter(&newvd->vdev_dtl_lock);
2579		if (newvd->vdev_dtl_map.sm_space == 0 &&
2580		    newvd->vdev_dtl_scrub.sm_space == 0) {
2581			mutex_exit(&newvd->vdev_dtl_lock);
2582			return (oldvd);
2583		}
2584		mutex_exit(&newvd->vdev_dtl_lock);
2585	}
2586
2587	/*
2588	 * Check for a completed resilver with the 'unspare' flag set.
2589	 */
2590	if (vd->vdev_ops == &vdev_spare_ops && vd->vdev_children == 2) {
2591		newvd = vd->vdev_child[0];
2592		oldvd = vd->vdev_child[1];
2593
2594		mutex_enter(&newvd->vdev_dtl_lock);
2595		if (newvd->vdev_unspare &&
2596		    newvd->vdev_dtl_map.sm_space == 0 &&
2597		    newvd->vdev_dtl_scrub.sm_space == 0) {
2598			newvd->vdev_unspare = 0;
2599			mutex_exit(&newvd->vdev_dtl_lock);
2600			return (oldvd);
2601		}
2602		mutex_exit(&newvd->vdev_dtl_lock);
2603	}
2604
2605	return (NULL);
2606}
2607
2608static void
2609spa_vdev_resilver_done(spa_t *spa)
2610{
2611	vdev_t *vd;
2612	vdev_t *pvd;
2613	uint64_t guid;
2614	uint64_t pguid = 0;
2615
2616	spa_config_enter(spa, RW_READER, FTAG);
2617
2618	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
2619		guid = vd->vdev_guid;
2620		/*
2621		 * If we have just finished replacing a hot spared device, then
2622		 * we need to detach the parent's first child (the original hot
2623		 * spare) as well.
2624		 */
2625		pvd = vd->vdev_parent;
2626		if (pvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2627		    pvd->vdev_id == 0) {
2628			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
2629			ASSERT(pvd->vdev_parent->vdev_children == 2);
2630			pguid = pvd->vdev_parent->vdev_child[1]->vdev_guid;
2631		}
2632		spa_config_exit(spa, FTAG);
2633		if (spa_vdev_detach(spa, guid, B_TRUE) != 0)
2634			return;
2635		if (pguid != 0 && spa_vdev_detach(spa, pguid, B_TRUE) != 0)
2636			return;
2637		spa_config_enter(spa, RW_READER, FTAG);
2638	}
2639
2640	spa_config_exit(spa, FTAG);
2641}
2642
2643/*
2644 * Update the stored path for this vdev.  Dirty the vdev configuration, relying
2645 * on spa_vdev_enter/exit() to synchronize the labels and cache.
2646 */
2647int
2648spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
2649{
2650	vdev_t *rvd, *vd;
2651	uint64_t txg;
2652
2653	rvd = spa->spa_root_vdev;
2654
2655	txg = spa_vdev_enter(spa);
2656
2657	if ((vd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
2658		/*
2659		 * Determine if this is a reference to a hot spare.  In that
2660		 * case, update the path as stored in the spare list.
2661		 */
2662		nvlist_t **spares;
2663		uint_t i, nspares;
2664		if (spa->spa_sparelist != NULL) {
2665			VERIFY(nvlist_lookup_nvlist_array(spa->spa_sparelist,
2666			    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2667			for (i = 0; i < nspares; i++) {
2668				uint64_t theguid;
2669				VERIFY(nvlist_lookup_uint64(spares[i],
2670				    ZPOOL_CONFIG_GUID, &theguid) == 0);
2671				if (theguid == guid)
2672					break;
2673			}
2674
2675			if (i == nspares)
2676				return (spa_vdev_exit(spa, NULL, txg, ENOENT));
2677
2678			VERIFY(nvlist_add_string(spares[i],
2679			    ZPOOL_CONFIG_PATH, newpath) == 0);
2680			spa_load_spares(spa);
2681			spa->spa_sync_spares = B_TRUE;
2682			return (spa_vdev_exit(spa, NULL, txg, 0));
2683		} else {
2684			return (spa_vdev_exit(spa, NULL, txg, ENOENT));
2685		}
2686	}
2687
2688	if (!vd->vdev_ops->vdev_op_leaf)
2689		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
2690
2691	spa_strfree(vd->vdev_path);
2692	vd->vdev_path = spa_strdup(newpath);
2693
2694	vdev_config_dirty(vd->vdev_top);
2695
2696	return (spa_vdev_exit(spa, NULL, txg, 0));
2697}
2698
2699/*
2700 * ==========================================================================
2701 * SPA Scrubbing
2702 * ==========================================================================
2703 */
2704
2705static void
2706spa_scrub_io_done(zio_t *zio)
2707{
2708	spa_t *spa = zio->io_spa;
2709
2710	arc_data_buf_free(zio->io_data, zio->io_size);
2711
2712	mutex_enter(&spa->spa_scrub_lock);
2713	if (zio->io_error && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
2714		vdev_t *vd = zio->io_vd ? zio->io_vd : spa->spa_root_vdev;
2715		spa->spa_scrub_errors++;
2716		mutex_enter(&vd->vdev_stat_lock);
2717		vd->vdev_stat.vs_scrub_errors++;
2718		mutex_exit(&vd->vdev_stat_lock);
2719	}
2720
2721	if (--spa->spa_scrub_inflight < spa->spa_scrub_maxinflight)
2722		cv_broadcast(&spa->spa_scrub_io_cv);
2723
2724	ASSERT(spa->spa_scrub_inflight >= 0);
2725
2726	mutex_exit(&spa->spa_scrub_lock);
2727}
2728
2729static void
2730spa_scrub_io_start(spa_t *spa, blkptr_t *bp, int priority, int flags,
2731    zbookmark_t *zb)
2732{
2733	size_t size = BP_GET_LSIZE(bp);
2734	void *data;
2735
2736	mutex_enter(&spa->spa_scrub_lock);
2737	/*
2738	 * Do not give too much work to vdev(s).
2739	 */
2740	while (spa->spa_scrub_inflight >= spa->spa_scrub_maxinflight) {
2741		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2742	}
2743	spa->spa_scrub_inflight++;
2744	mutex_exit(&spa->spa_scrub_lock);
2745
2746	data = arc_data_buf_alloc(size);
2747
2748	if (zb->zb_level == -1 && BP_GET_TYPE(bp) != DMU_OT_OBJSET)
2749		flags |= ZIO_FLAG_SPECULATIVE;	/* intent log block */
2750
2751	flags |= ZIO_FLAG_SCRUB_THREAD | ZIO_FLAG_CANFAIL;
2752
2753	zio_nowait(zio_read(NULL, spa, bp, data, size,
2754	    spa_scrub_io_done, NULL, priority, flags, zb));
2755}
2756
2757/* ARGSUSED */
2758static int
2759spa_scrub_cb(traverse_blk_cache_t *bc, spa_t *spa, void *a)
2760{
2761	blkptr_t *bp = &bc->bc_blkptr;
2762	vdev_t *vd = spa->spa_root_vdev;
2763	dva_t *dva = bp->blk_dva;
2764	int needs_resilver = B_FALSE;
2765	int d;
2766
2767	if (bc->bc_errno) {
2768		/*
2769		 * We can't scrub this block, but we can continue to scrub
2770		 * the rest of the pool.  Note the error and move along.
2771		 */
2772		mutex_enter(&spa->spa_scrub_lock);
2773		spa->spa_scrub_errors++;
2774		mutex_exit(&spa->spa_scrub_lock);
2775
2776		mutex_enter(&vd->vdev_stat_lock);
2777		vd->vdev_stat.vs_scrub_errors++;
2778		mutex_exit(&vd->vdev_stat_lock);
2779
2780		return (ERESTART);
2781	}
2782
2783	ASSERT(bp->blk_birth < spa->spa_scrub_maxtxg);
2784
2785	for (d = 0; d < BP_GET_NDVAS(bp); d++) {
2786		vd = vdev_lookup_top(spa, DVA_GET_VDEV(&dva[d]));
2787
2788		ASSERT(vd != NULL);
2789
2790		/*
2791		 * Keep track of how much data we've examined so that
2792		 * zpool(1M) status can make useful progress reports.
2793		 */
2794		mutex_enter(&vd->vdev_stat_lock);
2795		vd->vdev_stat.vs_scrub_examined += DVA_GET_ASIZE(&dva[d]);
2796		mutex_exit(&vd->vdev_stat_lock);
2797
2798		if (spa->spa_scrub_type == POOL_SCRUB_RESILVER) {
2799			if (DVA_GET_GANG(&dva[d])) {
2800				/*
2801				 * Gang members may be spread across multiple
2802				 * vdevs, so the best we can do is look at the
2803				 * pool-wide DTL.
2804				 * XXX -- it would be better to change our
2805				 * allocation policy to ensure that this can't
2806				 * happen.
2807				 */
2808				vd = spa->spa_root_vdev;
2809			}
2810			if (vdev_dtl_contains(&vd->vdev_dtl_map,
2811			    bp->blk_birth, 1))
2812				needs_resilver = B_TRUE;
2813		}
2814	}
2815
2816	if (spa->spa_scrub_type == POOL_SCRUB_EVERYTHING)
2817		spa_scrub_io_start(spa, bp, ZIO_PRIORITY_SCRUB,
2818		    ZIO_FLAG_SCRUB, &bc->bc_bookmark);
2819	else if (needs_resilver)
2820		spa_scrub_io_start(spa, bp, ZIO_PRIORITY_RESILVER,
2821		    ZIO_FLAG_RESILVER, &bc->bc_bookmark);
2822
2823	return (0);
2824}
2825
2826static void
2827spa_scrub_thread(spa_t *spa)
2828{
2829	callb_cpr_t cprinfo;
2830	traverse_handle_t *th = spa->spa_scrub_th;
2831	vdev_t *rvd = spa->spa_root_vdev;
2832	pool_scrub_type_t scrub_type = spa->spa_scrub_type;
2833	int error = 0;
2834	boolean_t complete;
2835
2836	CALLB_CPR_INIT(&cprinfo, &spa->spa_scrub_lock, callb_generic_cpr, FTAG);
2837
2838	/*
2839	 * If we're restarting due to a snapshot create/delete,
2840	 * wait for that to complete.
2841	 */
2842	txg_wait_synced(spa_get_dsl(spa), 0);
2843
2844	dprintf("start %s mintxg=%llu maxtxg=%llu\n",
2845	    scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub",
2846	    spa->spa_scrub_mintxg, spa->spa_scrub_maxtxg);
2847
2848	spa_config_enter(spa, RW_WRITER, FTAG);
2849	vdev_reopen(rvd);		/* purge all vdev caches */
2850	vdev_config_dirty(rvd);		/* rewrite all disk labels */
2851	vdev_scrub_stat_update(rvd, scrub_type, B_FALSE);
2852	spa_config_exit(spa, FTAG);
2853
2854	mutex_enter(&spa->spa_scrub_lock);
2855	spa->spa_scrub_errors = 0;
2856	spa->spa_scrub_active = 1;
2857	ASSERT(spa->spa_scrub_inflight == 0);
2858
2859	while (!spa->spa_scrub_stop) {
2860		CALLB_CPR_SAFE_BEGIN(&cprinfo);
2861		while (spa->spa_scrub_suspended) {
2862			spa->spa_scrub_active = 0;
2863			cv_broadcast(&spa->spa_scrub_cv);
2864			cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
2865			spa->spa_scrub_active = 1;
2866		}
2867		CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_scrub_lock);
2868
2869		if (spa->spa_scrub_restart_txg != 0)
2870			break;
2871
2872		mutex_exit(&spa->spa_scrub_lock);
2873		error = traverse_more(th);
2874		mutex_enter(&spa->spa_scrub_lock);
2875		if (error != EAGAIN)
2876			break;
2877	}
2878
2879	while (spa->spa_scrub_inflight)
2880		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2881
2882	spa->spa_scrub_active = 0;
2883	cv_broadcast(&spa->spa_scrub_cv);
2884
2885	mutex_exit(&spa->spa_scrub_lock);
2886
2887	spa_config_enter(spa, RW_WRITER, FTAG);
2888
2889	mutex_enter(&spa->spa_scrub_lock);
2890
2891	/*
2892	 * Note: we check spa_scrub_restart_txg under both spa_scrub_lock
2893	 * AND the spa config lock to synchronize with any config changes
2894	 * that revise the DTLs under spa_vdev_enter() / spa_vdev_exit().
2895	 */
2896	if (spa->spa_scrub_restart_txg != 0)
2897		error = ERESTART;
2898
2899	if (spa->spa_scrub_stop)
2900		error = EINTR;
2901
2902	/*
2903	 * Even if there were uncorrectable errors, we consider the scrub
2904	 * completed.  The downside is that if there is a transient error during
2905	 * a resilver, we won't resilver the data properly to the target.  But
2906	 * if the damage is permanent (more likely) we will resilver forever,
2907	 * which isn't really acceptable.  Since there is enough information for
2908	 * the user to know what has failed and why, this seems like a more
2909	 * tractable approach.
2910	 */
2911	complete = (error == 0);
2912
2913	dprintf("end %s to maxtxg=%llu %s, traverse=%d, %llu errors, stop=%u\n",
2914	    scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub",
2915	    spa->spa_scrub_maxtxg, complete ? "done" : "FAILED",
2916	    error, spa->spa_scrub_errors, spa->spa_scrub_stop);
2917
2918	mutex_exit(&spa->spa_scrub_lock);
2919
2920	/*
2921	 * If the scrub/resilver completed, update all DTLs to reflect this.
2922	 * Whether it succeeded or not, vacate all temporary scrub DTLs.
2923	 */
2924	vdev_dtl_reassess(rvd, spa_last_synced_txg(spa) + 1,
2925	    complete ? spa->spa_scrub_maxtxg : 0, B_TRUE);
2926	vdev_scrub_stat_update(rvd, POOL_SCRUB_NONE, complete);
2927	spa_errlog_rotate(spa);
2928
2929	if (scrub_type == POOL_SCRUB_RESILVER && complete)
2930		spa_event_notify(spa, NULL, ESC_ZFS_RESILVER_FINISH);
2931
2932	spa_config_exit(spa, FTAG);
2933
2934	mutex_enter(&spa->spa_scrub_lock);
2935
2936	/*
2937	 * We may have finished replacing a device.
2938	 * Let the async thread assess this and handle the detach.
2939	 */
2940	spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
2941
2942	/*
2943	 * If we were told to restart, our final act is to start a new scrub.
2944	 */
2945	if (error == ERESTART)
2946		spa_async_request(spa, scrub_type == POOL_SCRUB_RESILVER ?
2947		    SPA_ASYNC_RESILVER : SPA_ASYNC_SCRUB);
2948
2949	spa->spa_scrub_type = POOL_SCRUB_NONE;
2950	spa->spa_scrub_active = 0;
2951	spa->spa_scrub_thread = NULL;
2952	cv_broadcast(&spa->spa_scrub_cv);
2953	CALLB_CPR_EXIT(&cprinfo);	/* drops &spa->spa_scrub_lock */
2954	thread_exit();
2955}
2956
2957void
2958spa_scrub_suspend(spa_t *spa)
2959{
2960	mutex_enter(&spa->spa_scrub_lock);
2961	spa->spa_scrub_suspended++;
2962	while (spa->spa_scrub_active) {
2963		cv_broadcast(&spa->spa_scrub_cv);
2964		cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
2965	}
2966	while (spa->spa_scrub_inflight)
2967		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2968	mutex_exit(&spa->spa_scrub_lock);
2969}
2970
2971void
2972spa_scrub_resume(spa_t *spa)
2973{
2974	mutex_enter(&spa->spa_scrub_lock);
2975	ASSERT(spa->spa_scrub_suspended != 0);
2976	if (--spa->spa_scrub_suspended == 0)
2977		cv_broadcast(&spa->spa_scrub_cv);
2978	mutex_exit(&spa->spa_scrub_lock);
2979}
2980
2981void
2982spa_scrub_restart(spa_t *spa, uint64_t txg)
2983{
2984	/*
2985	 * Something happened (e.g. snapshot create/delete) that means
2986	 * we must restart any in-progress scrubs.  The itinerary will
2987	 * fix this properly.
2988	 */
2989	mutex_enter(&spa->spa_scrub_lock);
2990	spa->spa_scrub_restart_txg = txg;
2991	mutex_exit(&spa->spa_scrub_lock);
2992}
2993
2994int
2995spa_scrub(spa_t *spa, pool_scrub_type_t type, boolean_t force)
2996{
2997	space_seg_t *ss;
2998	uint64_t mintxg, maxtxg;
2999	vdev_t *rvd = spa->spa_root_vdev;
3000
3001	ASSERT(MUTEX_HELD(&spa_namespace_lock));
3002	ASSERT(!spa_config_held(spa, RW_WRITER));
3003
3004	if ((uint_t)type >= POOL_SCRUB_TYPES)
3005		return (ENOTSUP);
3006
3007	mutex_enter(&spa->spa_scrub_lock);
3008
3009	/*
3010	 * If there's a scrub or resilver already in progress, stop it.
3011	 */
3012	while (spa->spa_scrub_thread != NULL) {
3013		/*
3014		 * Don't stop a resilver unless forced.
3015		 */
3016		if (spa->spa_scrub_type == POOL_SCRUB_RESILVER && !force) {
3017			mutex_exit(&spa->spa_scrub_lock);
3018			return (EBUSY);
3019		}
3020		spa->spa_scrub_stop = 1;
3021		cv_broadcast(&spa->spa_scrub_cv);
3022		cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
3023	}
3024
3025	/*
3026	 * Terminate the previous traverse.
3027	 */
3028	if (spa->spa_scrub_th != NULL) {
3029		traverse_fini(spa->spa_scrub_th);
3030		spa->spa_scrub_th = NULL;
3031	}
3032
3033	if (rvd == NULL) {
3034		ASSERT(spa->spa_scrub_stop == 0);
3035		ASSERT(spa->spa_scrub_type == type);
3036		ASSERT(spa->spa_scrub_restart_txg == 0);
3037		mutex_exit(&spa->spa_scrub_lock);
3038		return (0);
3039	}
3040
3041	mintxg = TXG_INITIAL - 1;
3042	maxtxg = spa_last_synced_txg(spa) + 1;
3043
3044	mutex_enter(&rvd->vdev_dtl_lock);
3045
3046	if (rvd->vdev_dtl_map.sm_space == 0) {
3047		/*
3048		 * The pool-wide DTL is empty.
3049		 * If this is a resilver, there's nothing to do except
3050		 * check whether any in-progress replacements have completed.
3051		 */
3052		if (type == POOL_SCRUB_RESILVER) {
3053			type = POOL_SCRUB_NONE;
3054			spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
3055		}
3056	} else {
3057		/*
3058		 * The pool-wide DTL is non-empty.
3059		 * If this is a normal scrub, upgrade to a resilver instead.
3060		 */
3061		if (type == POOL_SCRUB_EVERYTHING)
3062			type = POOL_SCRUB_RESILVER;
3063	}
3064
3065	if (type == POOL_SCRUB_RESILVER) {
3066		/*
3067		 * Determine the resilvering boundaries.
3068		 *
3069		 * Note: (mintxg, maxtxg) is an open interval,
3070		 * i.e. mintxg and maxtxg themselves are not included.
3071		 *
3072		 * Note: for maxtxg, we MIN with spa_last_synced_txg(spa) + 1
3073		 * so we don't claim to resilver a txg that's still changing.
3074		 */
3075		ss = avl_first(&rvd->vdev_dtl_map.sm_root);
3076		mintxg = ss->ss_start - 1;
3077		ss = avl_last(&rvd->vdev_dtl_map.sm_root);
3078		maxtxg = MIN(ss->ss_end, maxtxg);
3079
3080		spa_event_notify(spa, NULL, ESC_ZFS_RESILVER_START);
3081	}
3082
3083	mutex_exit(&rvd->vdev_dtl_lock);
3084
3085	spa->spa_scrub_stop = 0;
3086	spa->spa_scrub_type = type;
3087	spa->spa_scrub_restart_txg = 0;
3088
3089	if (type != POOL_SCRUB_NONE) {
3090		spa->spa_scrub_mintxg = mintxg;
3091		spa->spa_scrub_maxtxg = maxtxg;
3092		spa->spa_scrub_th = traverse_init(spa, spa_scrub_cb, NULL,
3093		    ADVANCE_PRE | ADVANCE_PRUNE | ADVANCE_ZIL,
3094		    ZIO_FLAG_CANFAIL);
3095		traverse_add_pool(spa->spa_scrub_th, mintxg, maxtxg);
3096		spa->spa_scrub_thread = thread_create(NULL, 0,
3097		    spa_scrub_thread, spa, 0, &p0, TS_RUN, minclsyspri);
3098	}
3099
3100	mutex_exit(&spa->spa_scrub_lock);
3101
3102	return (0);
3103}
3104
3105/*
3106 * ==========================================================================
3107 * SPA async task processing
3108 * ==========================================================================
3109 */
3110
3111static void
3112spa_async_remove(spa_t *spa, vdev_t *vd)
3113{
3114	vdev_t *tvd;
3115	int c;
3116
3117	for (c = 0; c < vd->vdev_children; c++) {
3118		tvd = vd->vdev_child[c];
3119		if (tvd->vdev_remove_wanted) {
3120			tvd->vdev_remove_wanted = 0;
3121			vdev_set_state(tvd, B_FALSE, VDEV_STATE_REMOVED,
3122			    VDEV_AUX_NONE);
3123			vdev_clear(spa, tvd, B_TRUE);
3124			vdev_config_dirty(tvd->vdev_top);
3125		}
3126		spa_async_remove(spa, tvd);
3127	}
3128}
3129
3130static void
3131spa_async_thread(spa_t *spa)
3132{
3133	int tasks;
3134	uint64_t txg;
3135
3136	ASSERT(spa->spa_sync_on);
3137
3138	mutex_enter(&spa->spa_async_lock);
3139	tasks = spa->spa_async_tasks;
3140	spa->spa_async_tasks = 0;
3141	mutex_exit(&spa->spa_async_lock);
3142
3143	/*
3144	 * See if the config needs to be updated.
3145	 */
3146	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
3147		mutex_enter(&spa_namespace_lock);
3148		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3149		mutex_exit(&spa_namespace_lock);
3150	}
3151
3152	/*
3153	 * See if any devices need to be marked REMOVED.
3154	 *
3155	 * XXX - We avoid doing this when we are in
3156	 * I/O failure state since spa_vdev_enter() grabs
3157	 * the namespace lock and would not be able to obtain
3158	 * the writer config lock.
3159	 */
3160	if (tasks & SPA_ASYNC_REMOVE &&
3161	    spa_state(spa) != POOL_STATE_IO_FAILURE) {
3162		txg = spa_vdev_enter(spa);
3163		spa_async_remove(spa, spa->spa_root_vdev);
3164		(void) spa_vdev_exit(spa, NULL, txg, 0);
3165	}
3166
3167	/*
3168	 * If any devices are done replacing, detach them.
3169	 */
3170	if (tasks & SPA_ASYNC_RESILVER_DONE)
3171		spa_vdev_resilver_done(spa);
3172
3173	/*
3174	 * Kick off a scrub.  When starting a RESILVER scrub (or an EVERYTHING
3175	 * scrub which can become a resilver), we need to hold
3176	 * spa_namespace_lock() because the sysevent we post via
3177	 * spa_event_notify() needs to get the name of the pool.
3178	 */
3179	if (tasks & SPA_ASYNC_SCRUB) {
3180		mutex_enter(&spa_namespace_lock);
3181		VERIFY(spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_TRUE) == 0);
3182		mutex_exit(&spa_namespace_lock);
3183	}
3184
3185	/*
3186	 * Kick off a resilver.
3187	 */
3188	if (tasks & SPA_ASYNC_RESILVER) {
3189		mutex_enter(&spa_namespace_lock);
3190		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
3191		mutex_exit(&spa_namespace_lock);
3192	}
3193
3194	/*
3195	 * Let the world know that we're done.
3196	 */
3197	mutex_enter(&spa->spa_async_lock);
3198	spa->spa_async_thread = NULL;
3199	cv_broadcast(&spa->spa_async_cv);
3200	mutex_exit(&spa->spa_async_lock);
3201	thread_exit();
3202}
3203
3204void
3205spa_async_suspend(spa_t *spa)
3206{
3207	mutex_enter(&spa->spa_async_lock);
3208	spa->spa_async_suspended++;
3209	while (spa->spa_async_thread != NULL)
3210		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
3211	mutex_exit(&spa->spa_async_lock);
3212}
3213
3214void
3215spa_async_resume(spa_t *spa)
3216{
3217	mutex_enter(&spa->spa_async_lock);
3218	ASSERT(spa->spa_async_suspended != 0);
3219	spa->spa_async_suspended--;
3220	mutex_exit(&spa->spa_async_lock);
3221}
3222
3223static void
3224spa_async_dispatch(spa_t *spa)
3225{
3226	mutex_enter(&spa->spa_async_lock);
3227	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
3228	    spa->spa_async_thread == NULL &&
3229	    rootdir != NULL && !vn_is_readonly(rootdir))
3230		spa->spa_async_thread = thread_create(NULL, 0,
3231		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
3232	mutex_exit(&spa->spa_async_lock);
3233}
3234
3235void
3236spa_async_request(spa_t *spa, int task)
3237{
3238	mutex_enter(&spa->spa_async_lock);
3239	spa->spa_async_tasks |= task;
3240	mutex_exit(&spa->spa_async_lock);
3241}
3242
3243/*
3244 * ==========================================================================
3245 * SPA syncing routines
3246 * ==========================================================================
3247 */
3248
3249static void
3250spa_sync_deferred_frees(spa_t *spa, uint64_t txg)
3251{
3252	bplist_t *bpl = &spa->spa_sync_bplist;
3253	dmu_tx_t *tx;
3254	blkptr_t blk;
3255	uint64_t itor = 0;
3256	zio_t *zio;
3257	int error;
3258	uint8_t c = 1;
3259
3260	zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CONFIG_HELD);
3261
3262	while (bplist_iterate(bpl, &itor, &blk) == 0)
3263		zio_nowait(zio_free(zio, spa, txg, &blk, NULL, NULL));
3264
3265	error = zio_wait(zio);
3266	ASSERT3U(error, ==, 0);
3267
3268	tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3269	bplist_vacate(bpl, tx);
3270
3271	/*
3272	 * Pre-dirty the first block so we sync to convergence faster.
3273	 * (Usually only the first block is needed.)
3274	 */
3275	dmu_write(spa->spa_meta_objset, spa->spa_sync_bplist_obj, 0, 1, &c, tx);
3276	dmu_tx_commit(tx);
3277}
3278
3279static void
3280spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
3281{
3282	char *packed = NULL;
3283	size_t nvsize = 0;
3284	dmu_buf_t *db;
3285
3286	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
3287
3288	packed = kmem_alloc(nvsize, KM_SLEEP);
3289
3290	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
3291	    KM_SLEEP) == 0);
3292
3293	dmu_write(spa->spa_meta_objset, obj, 0, nvsize, packed, tx);
3294
3295	kmem_free(packed, nvsize);
3296
3297	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
3298	dmu_buf_will_dirty(db, tx);
3299	*(uint64_t *)db->db_data = nvsize;
3300	dmu_buf_rele(db, FTAG);
3301}
3302
3303static void
3304spa_sync_spares(spa_t *spa, dmu_tx_t *tx)
3305{
3306	nvlist_t *nvroot;
3307	nvlist_t **spares;
3308	int i;
3309
3310	if (!spa->spa_sync_spares)
3311		return;
3312
3313	/*
3314	 * Update the MOS nvlist describing the list of available spares.
3315	 * spa_validate_spares() will have already made sure this nvlist is
3316	 * valid and the vdevs are labeled appropriately.
3317	 */
3318	if (spa->spa_spares_object == 0) {
3319		spa->spa_spares_object = dmu_object_alloc(spa->spa_meta_objset,
3320		    DMU_OT_PACKED_NVLIST, 1 << 14,
3321		    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3322		VERIFY(zap_update(spa->spa_meta_objset,
3323		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SPARES,
3324		    sizeof (uint64_t), 1, &spa->spa_spares_object, tx) == 0);
3325	}
3326
3327	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3328	if (spa->spa_nspares == 0) {
3329		VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3330		    NULL, 0) == 0);
3331	} else {
3332		spares = kmem_alloc(spa->spa_nspares * sizeof (void *),
3333		    KM_SLEEP);
3334		for (i = 0; i < spa->spa_nspares; i++)
3335			spares[i] = vdev_config_generate(spa,
3336			    spa->spa_spares[i], B_FALSE, B_TRUE);
3337		VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3338		    spares, spa->spa_nspares) == 0);
3339		for (i = 0; i < spa->spa_nspares; i++)
3340			nvlist_free(spares[i]);
3341		kmem_free(spares, spa->spa_nspares * sizeof (void *));
3342	}
3343
3344	spa_sync_nvlist(spa, spa->spa_spares_object, nvroot, tx);
3345	nvlist_free(nvroot);
3346
3347	spa->spa_sync_spares = B_FALSE;
3348}
3349
3350static void
3351spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
3352{
3353	nvlist_t *config;
3354
3355	if (list_is_empty(&spa->spa_dirty_list))
3356		return;
3357
3358	config = spa_config_generate(spa, NULL, dmu_tx_get_txg(tx), B_FALSE);
3359
3360	if (spa->spa_config_syncing)
3361		nvlist_free(spa->spa_config_syncing);
3362	spa->spa_config_syncing = config;
3363
3364	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
3365}
3366
3367/*
3368 * Set zpool properties.
3369 */
3370static void
3371spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
3372{
3373	spa_t *spa = arg1;
3374	objset_t *mos = spa->spa_meta_objset;
3375	nvlist_t *nvp = arg2;
3376	nvpair_t *elem;
3377	uint64_t intval;
3378	char *strval;
3379	zpool_prop_t prop;
3380	const char *propname;
3381	zprop_type_t proptype;
3382
3383	elem = NULL;
3384	while ((elem = nvlist_next_nvpair(nvp, elem))) {
3385		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
3386		case ZPOOL_PROP_VERSION:
3387			/*
3388			 * Only set version for non-zpool-creation cases
3389			 * (set/import). spa_create() needs special care
3390			 * for version setting.
3391			 */
3392			if (tx->tx_txg != TXG_INITIAL) {
3393				VERIFY(nvpair_value_uint64(elem,
3394				    &intval) == 0);
3395				ASSERT(intval <= SPA_VERSION);
3396				ASSERT(intval >= spa_version(spa));
3397				spa->spa_uberblock.ub_version = intval;
3398				vdev_config_dirty(spa->spa_root_vdev);
3399			}
3400			break;
3401
3402		case ZPOOL_PROP_ALTROOT:
3403			/*
3404			 * 'altroot' is a non-persistent property. It should
3405			 * have been set temporarily at creation or import time.
3406			 */
3407			ASSERT(spa->spa_root != NULL);
3408			break;
3409
3410		case ZPOOL_PROP_TEMPORARY:
3411			/*
3412			 * 'temporary' is a non-persistant property.
3413			 */
3414			VERIFY(nvpair_value_uint64(elem, &intval) == 0);
3415			spa->spa_temporary = intval;
3416			break;
3417		default:
3418			/*
3419			 * Set pool property values in the poolprops mos object.
3420			 */
3421			mutex_enter(&spa->spa_props_lock);
3422			if (spa->spa_pool_props_object == 0) {
3423				objset_t *mos = spa->spa_meta_objset;
3424
3425				VERIFY((spa->spa_pool_props_object =
3426				    zap_create(mos, DMU_OT_POOL_PROPS,
3427				    DMU_OT_NONE, 0, tx)) > 0);
3428
3429				VERIFY(zap_update(mos,
3430				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
3431				    8, 1, &spa->spa_pool_props_object, tx)
3432				    == 0);
3433			}
3434			mutex_exit(&spa->spa_props_lock);
3435
3436			/* normalize the property name */
3437			propname = zpool_prop_to_name(prop);
3438			proptype = zpool_prop_get_type(prop);
3439
3440			if (nvpair_type(elem) == DATA_TYPE_STRING) {
3441				ASSERT(proptype == PROP_TYPE_STRING);
3442				VERIFY(nvpair_value_string(elem, &strval) == 0);
3443				VERIFY(zap_update(mos,
3444				    spa->spa_pool_props_object, propname,
3445				    1, strlen(strval) + 1, strval, tx) == 0);
3446
3447			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
3448				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
3449
3450				if (proptype == PROP_TYPE_INDEX) {
3451					const char *unused;
3452					VERIFY(zpool_prop_index_to_string(
3453					    prop, intval, &unused) == 0);
3454				}
3455				VERIFY(zap_update(mos,
3456				    spa->spa_pool_props_object, propname,
3457				    8, 1, &intval, tx) == 0);
3458			} else {
3459				ASSERT(0); /* not allowed */
3460			}
3461
3462			switch (prop) {
3463			case ZPOOL_PROP_DELEGATION:
3464				spa->spa_delegation = intval;
3465				break;
3466			case ZPOOL_PROP_BOOTFS:
3467				spa->spa_bootfs = intval;
3468				break;
3469			case ZPOOL_PROP_FAILUREMODE:
3470				spa->spa_failmode = intval;
3471				break;
3472			default:
3473				break;
3474			}
3475		}
3476
3477		/* log internal history if this is not a zpool create */
3478		if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
3479		    tx->tx_txg != TXG_INITIAL) {
3480			spa_history_internal_log(LOG_POOL_PROPSET,
3481			    spa, tx, cr, "%s %lld %s",
3482			    nvpair_name(elem), intval, spa->spa_name);
3483		}
3484	}
3485}
3486
3487/*
3488 * Sync the specified transaction group.  New blocks may be dirtied as
3489 * part of the process, so we iterate until it converges.
3490 */
3491void
3492spa_sync(spa_t *spa, uint64_t txg)
3493{
3494	dsl_pool_t *dp = spa->spa_dsl_pool;
3495	objset_t *mos = spa->spa_meta_objset;
3496	bplist_t *bpl = &spa->spa_sync_bplist;
3497	vdev_t *rvd = spa->spa_root_vdev;
3498	vdev_t *vd;
3499	dmu_tx_t *tx;
3500	int dirty_vdevs;
3501
3502	/*
3503	 * Lock out configuration changes.
3504	 */
3505	spa_config_enter(spa, RW_READER, FTAG);
3506
3507	spa->spa_syncing_txg = txg;
3508	spa->spa_sync_pass = 0;
3509
3510	VERIFY(0 == bplist_open(bpl, mos, spa->spa_sync_bplist_obj));
3511
3512	tx = dmu_tx_create_assigned(dp, txg);
3513
3514	/*
3515	 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
3516	 * set spa_deflate if we have no raid-z vdevs.
3517	 */
3518	if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
3519	    spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
3520		int i;
3521
3522		for (i = 0; i < rvd->vdev_children; i++) {
3523			vd = rvd->vdev_child[i];
3524			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
3525				break;
3526		}
3527		if (i == rvd->vdev_children) {
3528			spa->spa_deflate = TRUE;
3529			VERIFY(0 == zap_add(spa->spa_meta_objset,
3530			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3531			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
3532		}
3533	}
3534
3535	/*
3536	 * If anything has changed in this txg, push the deferred frees
3537	 * from the previous txg.  If not, leave them alone so that we
3538	 * don't generate work on an otherwise idle system.
3539	 */
3540	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
3541	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
3542	    !txg_list_empty(&dp->dp_sync_tasks, txg))
3543		spa_sync_deferred_frees(spa, txg);
3544
3545	/*
3546	 * Iterate to convergence.
3547	 */
3548	do {
3549		spa->spa_sync_pass++;
3550
3551		spa_sync_config_object(spa, tx);
3552		spa_sync_spares(spa, tx);
3553		spa_errlog_sync(spa, txg);
3554		dsl_pool_sync(dp, txg);
3555
3556		dirty_vdevs = 0;
3557		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) {
3558			vdev_sync(vd, txg);
3559			dirty_vdevs++;
3560		}
3561
3562		bplist_sync(bpl, tx);
3563	} while (dirty_vdevs);
3564
3565	bplist_close(bpl);
3566
3567	dprintf("txg %llu passes %d\n", txg, spa->spa_sync_pass);
3568
3569	/*
3570	 * Rewrite the vdev configuration (which includes the uberblock)
3571	 * to commit the transaction group.
3572	 *
3573	 * If there are any dirty vdevs, sync the uberblock to all vdevs.
3574	 * Otherwise, pick a random top-level vdev that's known to be
3575	 * visible in the config cache (see spa_vdev_add() for details).
3576	 * If the write fails, try the next vdev until we're tried them all.
3577	 */
3578	if (!list_is_empty(&spa->spa_dirty_list)) {
3579		VERIFY(vdev_config_sync(rvd, txg) == 0);
3580	} else {
3581		int children = rvd->vdev_children;
3582		int c0 = spa_get_random(children);
3583		int c;
3584
3585		for (c = 0; c < children; c++) {
3586			vd = rvd->vdev_child[(c0 + c) % children];
3587			if (vd->vdev_ms_array == 0)
3588				continue;
3589			if (vdev_config_sync(vd, txg) == 0)
3590				break;
3591		}
3592		if (c == children)
3593			VERIFY(vdev_config_sync(rvd, txg) == 0);
3594	}
3595
3596	dmu_tx_commit(tx);
3597
3598	/*
3599	 * Clear the dirty config list.
3600	 */
3601	while ((vd = list_head(&spa->spa_dirty_list)) != NULL)
3602		vdev_config_clean(vd);
3603
3604	/*
3605	 * Now that the new config has synced transactionally,
3606	 * let it become visible to the config cache.
3607	 */
3608	if (spa->spa_config_syncing != NULL) {
3609		spa_config_set(spa, spa->spa_config_syncing);
3610		spa->spa_config_txg = txg;
3611		spa->spa_config_syncing = NULL;
3612	}
3613
3614	/*
3615	 * Make a stable copy of the fully synced uberblock.
3616	 * We use this as the root for pool traversals.
3617	 */
3618	spa->spa_traverse_wanted = 1;	/* tells traverse_more() to stop */
3619
3620	spa_scrub_suspend(spa);		/* stop scrubbing and finish I/Os */
3621
3622	rw_enter(&spa->spa_traverse_lock, RW_WRITER);
3623	spa->spa_traverse_wanted = 0;
3624	spa->spa_ubsync = spa->spa_uberblock;
3625	rw_exit(&spa->spa_traverse_lock);
3626
3627	spa_scrub_resume(spa);		/* resume scrub with new ubsync */
3628
3629	/*
3630	 * Clean up the ZIL records for the synced txg.
3631	 */
3632	dsl_pool_zil_clean(dp);
3633
3634	/*
3635	 * Update usable space statistics.
3636	 */
3637	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
3638		vdev_sync_done(vd, txg);
3639
3640	/*
3641	 * It had better be the case that we didn't dirty anything
3642	 * since vdev_config_sync().
3643	 */
3644	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
3645	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
3646	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
3647	ASSERT(bpl->bpl_queue == NULL);
3648
3649	spa_config_exit(spa, FTAG);
3650
3651	/*
3652	 * If any async tasks have been requested, kick them off.
3653	 */
3654	spa_async_dispatch(spa);
3655}
3656
3657/*
3658 * Sync all pools.  We don't want to hold the namespace lock across these
3659 * operations, so we take a reference on the spa_t and drop the lock during the
3660 * sync.
3661 */
3662void
3663spa_sync_allpools(void)
3664{
3665	spa_t *spa = NULL;
3666	mutex_enter(&spa_namespace_lock);
3667	while ((spa = spa_next(spa)) != NULL) {
3668		if (spa_state(spa) != POOL_STATE_ACTIVE)
3669			continue;
3670		spa_open_ref(spa, FTAG);
3671		mutex_exit(&spa_namespace_lock);
3672		txg_wait_synced(spa_get_dsl(spa), 0);
3673		mutex_enter(&spa_namespace_lock);
3674		spa_close(spa, FTAG);
3675	}
3676	mutex_exit(&spa_namespace_lock);
3677}
3678
3679/*
3680 * ==========================================================================
3681 * Miscellaneous routines
3682 * ==========================================================================
3683 */
3684
3685/*
3686 * Remove all pools in the system.
3687 */
3688void
3689spa_evict_all(void)
3690{
3691	spa_t *spa;
3692
3693	/*
3694	 * Remove all cached state.  All pools should be closed now,
3695	 * so every spa in the AVL tree should be unreferenced.
3696	 */
3697	mutex_enter(&spa_namespace_lock);
3698	while ((spa = spa_next(NULL)) != NULL) {
3699		/*
3700		 * Stop async tasks.  The async thread may need to detach
3701		 * a device that's been replaced, which requires grabbing
3702		 * spa_namespace_lock, so we must drop it here.
3703		 */
3704		spa_open_ref(spa, FTAG);
3705		mutex_exit(&spa_namespace_lock);
3706		spa_async_suspend(spa);
3707		mutex_enter(&spa_namespace_lock);
3708		VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0);
3709		spa_close(spa, FTAG);
3710
3711		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3712			spa_unload(spa);
3713			spa_deactivate(spa);
3714		}
3715		spa_remove(spa);
3716	}
3717	mutex_exit(&spa_namespace_lock);
3718}
3719
3720vdev_t *
3721spa_lookup_by_guid(spa_t *spa, uint64_t guid)
3722{
3723	return (vdev_lookup_by_guid(spa->spa_root_vdev, guid));
3724}
3725
3726void
3727spa_upgrade(spa_t *spa, uint64_t version)
3728{
3729	spa_config_enter(spa, RW_WRITER, FTAG);
3730
3731	/*
3732	 * This should only be called for a non-faulted pool, and since a
3733	 * future version would result in an unopenable pool, this shouldn't be
3734	 * possible.
3735	 */
3736	ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
3737	ASSERT(version >= spa->spa_uberblock.ub_version);
3738
3739	spa->spa_uberblock.ub_version = version;
3740	vdev_config_dirty(spa->spa_root_vdev);
3741
3742	spa_config_exit(spa, FTAG);
3743
3744	txg_wait_synced(spa_get_dsl(spa), 0);
3745}
3746
3747boolean_t
3748spa_has_spare(spa_t *spa, uint64_t guid)
3749{
3750	int i;
3751	uint64_t spareguid;
3752
3753	for (i = 0; i < spa->spa_nspares; i++)
3754		if (spa->spa_spares[i]->vdev_guid == guid)
3755			return (B_TRUE);
3756
3757	for (i = 0; i < spa->spa_pending_nspares; i++) {
3758		if (nvlist_lookup_uint64(spa->spa_pending_spares[i],
3759		    ZPOOL_CONFIG_GUID, &spareguid) == 0 &&
3760		    spareguid == guid)
3761			return (B_TRUE);
3762	}
3763
3764	return (B_FALSE);
3765}
3766
3767/*
3768 * Post a sysevent corresponding to the given event.  The 'name' must be one of
3769 * the event definitions in sys/sysevent/eventdefs.h.  The payload will be
3770 * filled in from the spa and (optionally) the vdev.  This doesn't do anything
3771 * in the userland libzpool, as we don't want consumers to misinterpret ztest
3772 * or zdb as real changes.
3773 */
3774void
3775spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
3776{
3777#ifdef _KERNEL
3778	sysevent_t		*ev;
3779	sysevent_attr_list_t	*attr = NULL;
3780	sysevent_value_t	value;
3781	sysevent_id_t		eid;
3782
3783	ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
3784	    SE_SLEEP);
3785
3786	value.value_type = SE_DATA_TYPE_STRING;
3787	value.value.sv_string = spa_name(spa);
3788	if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
3789		goto done;
3790
3791	value.value_type = SE_DATA_TYPE_UINT64;
3792	value.value.sv_uint64 = spa_guid(spa);
3793	if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
3794		goto done;
3795
3796	if (vd) {
3797		value.value_type = SE_DATA_TYPE_UINT64;
3798		value.value.sv_uint64 = vd->vdev_guid;
3799		if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
3800		    SE_SLEEP) != 0)
3801			goto done;
3802
3803		if (vd->vdev_path) {
3804			value.value_type = SE_DATA_TYPE_STRING;
3805			value.value.sv_string = vd->vdev_path;
3806			if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
3807			    &value, SE_SLEEP) != 0)
3808				goto done;
3809		}
3810	}
3811
3812	(void) log_sysevent(ev, SE_SLEEP, &eid);
3813
3814done:
3815	if (attr)
3816		sysevent_free_attr(attr);
3817	sysevent_free(ev);
3818#endif
3819}
3820