spa.c revision b24ab6762772a3f6a89393947930c7fa61306783
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 2009 Sun Microsystems, Inc.  All rights reserved.
24 * Use is subject to license terms.
25 */
26
27/*
28 * This file contains all the routines used when modifying on-disk SPA state.
29 * This includes opening, importing, destroying, exporting a pool, and syncing a
30 * pool.
31 */
32
33#include <sys/zfs_context.h>
34#include <sys/fm/fs/zfs.h>
35#include <sys/spa_impl.h>
36#include <sys/zio.h>
37#include <sys/zio_checksum.h>
38#include <sys/dmu.h>
39#include <sys/dmu_tx.h>
40#include <sys/zap.h>
41#include <sys/zil.h>
42#include <sys/ddt.h>
43#include <sys/vdev_impl.h>
44#include <sys/metaslab.h>
45#include <sys/metaslab_impl.h>
46#include <sys/uberblock_impl.h>
47#include <sys/txg.h>
48#include <sys/avl.h>
49#include <sys/dmu_traverse.h>
50#include <sys/dmu_objset.h>
51#include <sys/unique.h>
52#include <sys/dsl_pool.h>
53#include <sys/dsl_dataset.h>
54#include <sys/dsl_dir.h>
55#include <sys/dsl_prop.h>
56#include <sys/dsl_synctask.h>
57#include <sys/fs/zfs.h>
58#include <sys/arc.h>
59#include <sys/callb.h>
60#include <sys/systeminfo.h>
61#include <sys/sunddi.h>
62#include <sys/spa_boot.h>
63#include <sys/zfs_ioctl.h>
64
65#ifdef	_KERNEL
66#include <sys/zone.h>
67#include <sys/bootprops.h>
68#endif	/* _KERNEL */
69
70#include "zfs_prop.h"
71#include "zfs_comutil.h"
72
73enum zti_modes {
74	zti_mode_fixed,			/* value is # of threads (min 1) */
75	zti_mode_online_percent,	/* value is % of online CPUs */
76	zti_mode_tune,			/* fill from zio_taskq_tune_* */
77	zti_nmodes
78};
79
80#define	ZTI_THREAD_FIX(n)	{ zti_mode_fixed, (n) }
81#define	ZTI_THREAD_PCT(n)	{ zti_mode_online_percent, (n) }
82#define	ZTI_THREAD_TUNE		{ zti_mode_tune, 0 }
83
84#define	ZTI_THREAD_ONE		ZTI_THREAD_FIX(1)
85
86typedef struct zio_taskq_info {
87	const char *zti_name;
88	struct {
89		enum zti_modes zti_mode;
90		uint_t zti_value;
91	} zti_nthreads[ZIO_TASKQ_TYPES];
92} zio_taskq_info_t;
93
94static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
95				"issue",		"intr"
96};
97
98const zio_taskq_info_t zio_taskqs[ZIO_TYPES] = {
99	/*			ISSUE			INTR		*/
100	{ "spa_zio_null",	{ ZTI_THREAD_ONE,	ZTI_THREAD_ONE } },
101	{ "spa_zio_read",	{ ZTI_THREAD_FIX(8),	ZTI_THREAD_TUNE } },
102	{ "spa_zio_write",	{ ZTI_THREAD_TUNE,	ZTI_THREAD_FIX(8) } },
103	{ "spa_zio_free",	{ ZTI_THREAD_ONE,	ZTI_THREAD_ONE } },
104	{ "spa_zio_claim",	{ ZTI_THREAD_ONE,	ZTI_THREAD_ONE } },
105	{ "spa_zio_ioctl",	{ ZTI_THREAD_ONE,	ZTI_THREAD_ONE } },
106};
107
108enum zti_modes zio_taskq_tune_mode = zti_mode_online_percent;
109uint_t zio_taskq_tune_value = 80;	/* #threads = 80% of # online CPUs */
110
111static void spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx);
112static boolean_t spa_has_active_shared_spare(spa_t *spa);
113
114/*
115 * ==========================================================================
116 * SPA properties routines
117 * ==========================================================================
118 */
119
120/*
121 * Add a (source=src, propname=propval) list to an nvlist.
122 */
123static void
124spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
125    uint64_t intval, zprop_source_t src)
126{
127	const char *propname = zpool_prop_to_name(prop);
128	nvlist_t *propval;
129
130	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
131	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
132
133	if (strval != NULL)
134		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
135	else
136		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
137
138	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
139	nvlist_free(propval);
140}
141
142/*
143 * Get property values from the spa configuration.
144 */
145static void
146spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
147{
148	uint64_t size;
149	uint64_t used;
150	uint64_t cap, version;
151	zprop_source_t src = ZPROP_SRC_NONE;
152	spa_config_dirent_t *dp;
153
154	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
155
156	if (spa->spa_root_vdev != NULL) {
157		used = metaslab_class_get_alloc(spa_normal_class(spa));
158		size = metaslab_class_get_space(spa_normal_class(spa));
159		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
160		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
161		spa_prop_add_list(*nvp, ZPOOL_PROP_USED, NULL, used, src);
162		spa_prop_add_list(*nvp, ZPOOL_PROP_AVAILABLE, NULL,
163		    size - used, src);
164
165		cap = (size == 0) ? 0 : (used * 100 / size);
166		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
167
168		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
169		    ddt_get_pool_dedup_ratio(spa), src);
170
171		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
172		    spa->spa_root_vdev->vdev_state, src);
173
174		version = spa_version(spa);
175		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
176			src = ZPROP_SRC_DEFAULT;
177		else
178			src = ZPROP_SRC_LOCAL;
179		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
180	}
181
182	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
183
184	if (spa->spa_root != NULL)
185		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
186		    0, ZPROP_SRC_LOCAL);
187
188	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
189		if (dp->scd_path == NULL) {
190			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
191			    "none", 0, ZPROP_SRC_LOCAL);
192		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
193			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
194			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
195		}
196	}
197}
198
199/*
200 * Get zpool property values.
201 */
202int
203spa_prop_get(spa_t *spa, nvlist_t **nvp)
204{
205	objset_t *mos = spa->spa_meta_objset;
206	zap_cursor_t zc;
207	zap_attribute_t za;
208	int err;
209
210	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
211
212	mutex_enter(&spa->spa_props_lock);
213
214	/*
215	 * Get properties from the spa config.
216	 */
217	spa_prop_get_config(spa, nvp);
218
219	/* If no pool property object, no more prop to get. */
220	if (spa->spa_pool_props_object == 0) {
221		mutex_exit(&spa->spa_props_lock);
222		return (0);
223	}
224
225	/*
226	 * Get properties from the MOS pool property object.
227	 */
228	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
229	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
230	    zap_cursor_advance(&zc)) {
231		uint64_t intval = 0;
232		char *strval = NULL;
233		zprop_source_t src = ZPROP_SRC_DEFAULT;
234		zpool_prop_t prop;
235
236		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
237			continue;
238
239		switch (za.za_integer_length) {
240		case 8:
241			/* integer property */
242			if (za.za_first_integer !=
243			    zpool_prop_default_numeric(prop))
244				src = ZPROP_SRC_LOCAL;
245
246			if (prop == ZPOOL_PROP_BOOTFS) {
247				dsl_pool_t *dp;
248				dsl_dataset_t *ds = NULL;
249
250				dp = spa_get_dsl(spa);
251				rw_enter(&dp->dp_config_rwlock, RW_READER);
252				if (err = dsl_dataset_hold_obj(dp,
253				    za.za_first_integer, FTAG, &ds)) {
254					rw_exit(&dp->dp_config_rwlock);
255					break;
256				}
257
258				strval = kmem_alloc(
259				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
260				    KM_SLEEP);
261				dsl_dataset_name(ds, strval);
262				dsl_dataset_rele(ds, FTAG);
263				rw_exit(&dp->dp_config_rwlock);
264			} else {
265				strval = NULL;
266				intval = za.za_first_integer;
267			}
268
269			spa_prop_add_list(*nvp, prop, strval, intval, src);
270
271			if (strval != NULL)
272				kmem_free(strval,
273				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
274
275			break;
276
277		case 1:
278			/* string property */
279			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
280			err = zap_lookup(mos, spa->spa_pool_props_object,
281			    za.za_name, 1, za.za_num_integers, strval);
282			if (err) {
283				kmem_free(strval, za.za_num_integers);
284				break;
285			}
286			spa_prop_add_list(*nvp, prop, strval, 0, src);
287			kmem_free(strval, za.za_num_integers);
288			break;
289
290		default:
291			break;
292		}
293	}
294	zap_cursor_fini(&zc);
295	mutex_exit(&spa->spa_props_lock);
296out:
297	if (err && err != ENOENT) {
298		nvlist_free(*nvp);
299		*nvp = NULL;
300		return (err);
301	}
302
303	return (0);
304}
305
306/*
307 * Validate the given pool properties nvlist and modify the list
308 * for the property values to be set.
309 */
310static int
311spa_prop_validate(spa_t *spa, nvlist_t *props)
312{
313	nvpair_t *elem;
314	int error = 0, reset_bootfs = 0;
315	uint64_t objnum;
316
317	elem = NULL;
318	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
319		zpool_prop_t prop;
320		char *propname, *strval;
321		uint64_t intval;
322		objset_t *os;
323		char *slash;
324
325		propname = nvpair_name(elem);
326
327		if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
328			return (EINVAL);
329
330		switch (prop) {
331		case ZPOOL_PROP_VERSION:
332			error = nvpair_value_uint64(elem, &intval);
333			if (!error &&
334			    (intval < spa_version(spa) || intval > SPA_VERSION))
335				error = EINVAL;
336			break;
337
338		case ZPOOL_PROP_DELEGATION:
339		case ZPOOL_PROP_AUTOREPLACE:
340		case ZPOOL_PROP_LISTSNAPS:
341		case ZPOOL_PROP_AUTOEXPAND:
342			error = nvpair_value_uint64(elem, &intval);
343			if (!error && intval > 1)
344				error = EINVAL;
345			break;
346
347		case ZPOOL_PROP_BOOTFS:
348			/*
349			 * If the pool version is less than SPA_VERSION_BOOTFS,
350			 * or the pool is still being created (version == 0),
351			 * the bootfs property cannot be set.
352			 */
353			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
354				error = ENOTSUP;
355				break;
356			}
357
358			/*
359			 * Make sure the vdev config is bootable
360			 */
361			if (!vdev_is_bootable(spa->spa_root_vdev)) {
362				error = ENOTSUP;
363				break;
364			}
365
366			reset_bootfs = 1;
367
368			error = nvpair_value_string(elem, &strval);
369
370			if (!error) {
371				uint64_t compress;
372
373				if (strval == NULL || strval[0] == '\0') {
374					objnum = zpool_prop_default_numeric(
375					    ZPOOL_PROP_BOOTFS);
376					break;
377				}
378
379				if (error = dmu_objset_hold(strval, FTAG, &os))
380					break;
381
382				/* Must be ZPL and not gzip compressed. */
383
384				if (dmu_objset_type(os) != DMU_OST_ZFS) {
385					error = ENOTSUP;
386				} else if ((error = dsl_prop_get_integer(strval,
387				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
388				    &compress, NULL)) == 0 &&
389				    !BOOTFS_COMPRESS_VALID(compress)) {
390					error = ENOTSUP;
391				} else {
392					objnum = dmu_objset_id(os);
393				}
394				dmu_objset_rele(os, FTAG);
395			}
396			break;
397
398		case ZPOOL_PROP_FAILUREMODE:
399			error = nvpair_value_uint64(elem, &intval);
400			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
401			    intval > ZIO_FAILURE_MODE_PANIC))
402				error = EINVAL;
403
404			/*
405			 * This is a special case which only occurs when
406			 * the pool has completely failed. This allows
407			 * the user to change the in-core failmode property
408			 * without syncing it out to disk (I/Os might
409			 * currently be blocked). We do this by returning
410			 * EIO to the caller (spa_prop_set) to trick it
411			 * into thinking we encountered a property validation
412			 * error.
413			 */
414			if (!error && spa_suspended(spa)) {
415				spa->spa_failmode = intval;
416				error = EIO;
417			}
418			break;
419
420		case ZPOOL_PROP_CACHEFILE:
421			if ((error = nvpair_value_string(elem, &strval)) != 0)
422				break;
423
424			if (strval[0] == '\0')
425				break;
426
427			if (strcmp(strval, "none") == 0)
428				break;
429
430			if (strval[0] != '/') {
431				error = EINVAL;
432				break;
433			}
434
435			slash = strrchr(strval, '/');
436			ASSERT(slash != NULL);
437
438			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
439			    strcmp(slash, "/..") == 0)
440				error = EINVAL;
441			break;
442
443		case ZPOOL_PROP_DEDUPDITTO:
444			if (spa_version(spa) < SPA_VERSION_DEDUP)
445				error = ENOTSUP;
446			else
447				error = nvpair_value_uint64(elem, &intval);
448			if (error == 0 &&
449			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
450				error = EINVAL;
451			break;
452		}
453
454		if (error)
455			break;
456	}
457
458	if (!error && reset_bootfs) {
459		error = nvlist_remove(props,
460		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
461
462		if (!error) {
463			error = nvlist_add_uint64(props,
464			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
465		}
466	}
467
468	return (error);
469}
470
471void
472spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
473{
474	char *cachefile;
475	spa_config_dirent_t *dp;
476
477	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
478	    &cachefile) != 0)
479		return;
480
481	dp = kmem_alloc(sizeof (spa_config_dirent_t),
482	    KM_SLEEP);
483
484	if (cachefile[0] == '\0')
485		dp->scd_path = spa_strdup(spa_config_path);
486	else if (strcmp(cachefile, "none") == 0)
487		dp->scd_path = NULL;
488	else
489		dp->scd_path = spa_strdup(cachefile);
490
491	list_insert_head(&spa->spa_config_list, dp);
492	if (need_sync)
493		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
494}
495
496int
497spa_prop_set(spa_t *spa, nvlist_t *nvp)
498{
499	int error;
500	nvpair_t *elem;
501	boolean_t need_sync = B_FALSE;
502	zpool_prop_t prop;
503
504	if ((error = spa_prop_validate(spa, nvp)) != 0)
505		return (error);
506
507	elem = NULL;
508	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
509		if ((prop = zpool_name_to_prop(
510		    nvpair_name(elem))) == ZPROP_INVAL)
511			return (EINVAL);
512
513		if (prop == ZPOOL_PROP_CACHEFILE || prop == ZPOOL_PROP_ALTROOT)
514			continue;
515
516		need_sync = B_TRUE;
517		break;
518	}
519
520	if (need_sync)
521		return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
522		    spa, nvp, 3));
523	else
524		return (0);
525}
526
527/*
528 * If the bootfs property value is dsobj, clear it.
529 */
530void
531spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
532{
533	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
534		VERIFY(zap_remove(spa->spa_meta_objset,
535		    spa->spa_pool_props_object,
536		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
537		spa->spa_bootfs = 0;
538	}
539}
540
541/*
542 * ==========================================================================
543 * SPA state manipulation (open/create/destroy/import/export)
544 * ==========================================================================
545 */
546
547static int
548spa_error_entry_compare(const void *a, const void *b)
549{
550	spa_error_entry_t *sa = (spa_error_entry_t *)a;
551	spa_error_entry_t *sb = (spa_error_entry_t *)b;
552	int ret;
553
554	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
555	    sizeof (zbookmark_t));
556
557	if (ret < 0)
558		return (-1);
559	else if (ret > 0)
560		return (1);
561	else
562		return (0);
563}
564
565/*
566 * Utility function which retrieves copies of the current logs and
567 * re-initializes them in the process.
568 */
569void
570spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
571{
572	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
573
574	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
575	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
576
577	avl_create(&spa->spa_errlist_scrub,
578	    spa_error_entry_compare, sizeof (spa_error_entry_t),
579	    offsetof(spa_error_entry_t, se_avl));
580	avl_create(&spa->spa_errlist_last,
581	    spa_error_entry_compare, sizeof (spa_error_entry_t),
582	    offsetof(spa_error_entry_t, se_avl));
583}
584
585/*
586 * Activate an uninitialized pool.
587 */
588static void
589spa_activate(spa_t *spa, int mode)
590{
591	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
592
593	spa->spa_state = POOL_STATE_ACTIVE;
594	spa->spa_mode = mode;
595
596	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
597	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
598
599	for (int t = 0; t < ZIO_TYPES; t++) {
600		const zio_taskq_info_t *ztip = &zio_taskqs[t];
601		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
602			enum zti_modes mode = ztip->zti_nthreads[q].zti_mode;
603			uint_t value = ztip->zti_nthreads[q].zti_value;
604			char name[32];
605
606			(void) snprintf(name, sizeof (name),
607			    "%s_%s", ztip->zti_name, zio_taskq_types[q]);
608
609			if (mode == zti_mode_tune) {
610				mode = zio_taskq_tune_mode;
611				value = zio_taskq_tune_value;
612				if (mode == zti_mode_tune)
613					mode = zti_mode_online_percent;
614			}
615
616			switch (mode) {
617			case zti_mode_fixed:
618				ASSERT3U(value, >=, 1);
619				value = MAX(value, 1);
620
621				spa->spa_zio_taskq[t][q] = taskq_create(name,
622				    value, maxclsyspri, 50, INT_MAX,
623				    TASKQ_PREPOPULATE);
624				break;
625
626			case zti_mode_online_percent:
627				spa->spa_zio_taskq[t][q] = taskq_create(name,
628				    value, maxclsyspri, 50, INT_MAX,
629				    TASKQ_PREPOPULATE | TASKQ_THREADS_CPU_PCT);
630				break;
631
632			case zti_mode_tune:
633			default:
634				panic("unrecognized mode for "
635				    "zio_taskqs[%u]->zti_nthreads[%u] (%u:%u) "
636				    "in spa_activate()",
637				    t, q, mode, value);
638				break;
639			}
640		}
641	}
642
643	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
644	    offsetof(vdev_t, vdev_config_dirty_node));
645	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
646	    offsetof(vdev_t, vdev_state_dirty_node));
647
648	txg_list_create(&spa->spa_vdev_txg_list,
649	    offsetof(struct vdev, vdev_txg_node));
650
651	avl_create(&spa->spa_errlist_scrub,
652	    spa_error_entry_compare, sizeof (spa_error_entry_t),
653	    offsetof(spa_error_entry_t, se_avl));
654	avl_create(&spa->spa_errlist_last,
655	    spa_error_entry_compare, sizeof (spa_error_entry_t),
656	    offsetof(spa_error_entry_t, se_avl));
657}
658
659/*
660 * Opposite of spa_activate().
661 */
662static void
663spa_deactivate(spa_t *spa)
664{
665	ASSERT(spa->spa_sync_on == B_FALSE);
666	ASSERT(spa->spa_dsl_pool == NULL);
667	ASSERT(spa->spa_root_vdev == NULL);
668	ASSERT(spa->spa_async_zio_root == NULL);
669	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
670
671	txg_list_destroy(&spa->spa_vdev_txg_list);
672
673	list_destroy(&spa->spa_config_dirty_list);
674	list_destroy(&spa->spa_state_dirty_list);
675
676	for (int t = 0; t < ZIO_TYPES; t++) {
677		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
678			taskq_destroy(spa->spa_zio_taskq[t][q]);
679			spa->spa_zio_taskq[t][q] = NULL;
680		}
681	}
682
683	metaslab_class_destroy(spa->spa_normal_class);
684	spa->spa_normal_class = NULL;
685
686	metaslab_class_destroy(spa->spa_log_class);
687	spa->spa_log_class = NULL;
688
689	/*
690	 * If this was part of an import or the open otherwise failed, we may
691	 * still have errors left in the queues.  Empty them just in case.
692	 */
693	spa_errlog_drain(spa);
694
695	avl_destroy(&spa->spa_errlist_scrub);
696	avl_destroy(&spa->spa_errlist_last);
697
698	spa->spa_state = POOL_STATE_UNINITIALIZED;
699}
700
701/*
702 * Verify a pool configuration, and construct the vdev tree appropriately.  This
703 * will create all the necessary vdevs in the appropriate layout, with each vdev
704 * in the CLOSED state.  This will prep the pool before open/creation/import.
705 * All vdev validation is done by the vdev_alloc() routine.
706 */
707static int
708spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
709    uint_t id, int atype)
710{
711	nvlist_t **child;
712	uint_t children;
713	int error;
714
715	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
716		return (error);
717
718	if ((*vdp)->vdev_ops->vdev_op_leaf)
719		return (0);
720
721	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
722	    &child, &children);
723
724	if (error == ENOENT)
725		return (0);
726
727	if (error) {
728		vdev_free(*vdp);
729		*vdp = NULL;
730		return (EINVAL);
731	}
732
733	for (int c = 0; c < children; c++) {
734		vdev_t *vd;
735		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
736		    atype)) != 0) {
737			vdev_free(*vdp);
738			*vdp = NULL;
739			return (error);
740		}
741	}
742
743	ASSERT(*vdp != NULL);
744
745	return (0);
746}
747
748/*
749 * Opposite of spa_load().
750 */
751static void
752spa_unload(spa_t *spa)
753{
754	int i;
755
756	ASSERT(MUTEX_HELD(&spa_namespace_lock));
757
758	/*
759	 * Stop async tasks.
760	 */
761	spa_async_suspend(spa);
762
763	/*
764	 * Stop syncing.
765	 */
766	if (spa->spa_sync_on) {
767		txg_sync_stop(spa->spa_dsl_pool);
768		spa->spa_sync_on = B_FALSE;
769	}
770
771	/*
772	 * Wait for any outstanding async I/O to complete.
773	 */
774	if (spa->spa_async_zio_root != NULL) {
775		(void) zio_wait(spa->spa_async_zio_root);
776		spa->spa_async_zio_root = NULL;
777	}
778
779	/*
780	 * Close the dsl pool.
781	 */
782	if (spa->spa_dsl_pool) {
783		dsl_pool_close(spa->spa_dsl_pool);
784		spa->spa_dsl_pool = NULL;
785	}
786
787	ddt_unload(spa);
788
789	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
790
791	/*
792	 * Drop and purge level 2 cache
793	 */
794	spa_l2cache_drop(spa);
795
796	/*
797	 * Close all vdevs.
798	 */
799	if (spa->spa_root_vdev)
800		vdev_free(spa->spa_root_vdev);
801	ASSERT(spa->spa_root_vdev == NULL);
802
803	for (i = 0; i < spa->spa_spares.sav_count; i++)
804		vdev_free(spa->spa_spares.sav_vdevs[i]);
805	if (spa->spa_spares.sav_vdevs) {
806		kmem_free(spa->spa_spares.sav_vdevs,
807		    spa->spa_spares.sav_count * sizeof (void *));
808		spa->spa_spares.sav_vdevs = NULL;
809	}
810	if (spa->spa_spares.sav_config) {
811		nvlist_free(spa->spa_spares.sav_config);
812		spa->spa_spares.sav_config = NULL;
813	}
814	spa->spa_spares.sav_count = 0;
815
816	for (i = 0; i < spa->spa_l2cache.sav_count; i++)
817		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
818	if (spa->spa_l2cache.sav_vdevs) {
819		kmem_free(spa->spa_l2cache.sav_vdevs,
820		    spa->spa_l2cache.sav_count * sizeof (void *));
821		spa->spa_l2cache.sav_vdevs = NULL;
822	}
823	if (spa->spa_l2cache.sav_config) {
824		nvlist_free(spa->spa_l2cache.sav_config);
825		spa->spa_l2cache.sav_config = NULL;
826	}
827	spa->spa_l2cache.sav_count = 0;
828
829	spa->spa_async_suspended = 0;
830
831	spa_config_exit(spa, SCL_ALL, FTAG);
832}
833
834/*
835 * Load (or re-load) the current list of vdevs describing the active spares for
836 * this pool.  When this is called, we have some form of basic information in
837 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
838 * then re-generate a more complete list including status information.
839 */
840static void
841spa_load_spares(spa_t *spa)
842{
843	nvlist_t **spares;
844	uint_t nspares;
845	int i;
846	vdev_t *vd, *tvd;
847
848	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
849
850	/*
851	 * First, close and free any existing spare vdevs.
852	 */
853	for (i = 0; i < spa->spa_spares.sav_count; i++) {
854		vd = spa->spa_spares.sav_vdevs[i];
855
856		/* Undo the call to spa_activate() below */
857		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
858		    B_FALSE)) != NULL && tvd->vdev_isspare)
859			spa_spare_remove(tvd);
860		vdev_close(vd);
861		vdev_free(vd);
862	}
863
864	if (spa->spa_spares.sav_vdevs)
865		kmem_free(spa->spa_spares.sav_vdevs,
866		    spa->spa_spares.sav_count * sizeof (void *));
867
868	if (spa->spa_spares.sav_config == NULL)
869		nspares = 0;
870	else
871		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
872		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
873
874	spa->spa_spares.sav_count = (int)nspares;
875	spa->spa_spares.sav_vdevs = NULL;
876
877	if (nspares == 0)
878		return;
879
880	/*
881	 * Construct the array of vdevs, opening them to get status in the
882	 * process.   For each spare, there is potentially two different vdev_t
883	 * structures associated with it: one in the list of spares (used only
884	 * for basic validation purposes) and one in the active vdev
885	 * configuration (if it's spared in).  During this phase we open and
886	 * validate each vdev on the spare list.  If the vdev also exists in the
887	 * active configuration, then we also mark this vdev as an active spare.
888	 */
889	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
890	    KM_SLEEP);
891	for (i = 0; i < spa->spa_spares.sav_count; i++) {
892		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
893		    VDEV_ALLOC_SPARE) == 0);
894		ASSERT(vd != NULL);
895
896		spa->spa_spares.sav_vdevs[i] = vd;
897
898		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
899		    B_FALSE)) != NULL) {
900			if (!tvd->vdev_isspare)
901				spa_spare_add(tvd);
902
903			/*
904			 * We only mark the spare active if we were successfully
905			 * able to load the vdev.  Otherwise, importing a pool
906			 * with a bad active spare would result in strange
907			 * behavior, because multiple pool would think the spare
908			 * is actively in use.
909			 *
910			 * There is a vulnerability here to an equally bizarre
911			 * circumstance, where a dead active spare is later
912			 * brought back to life (onlined or otherwise).  Given
913			 * the rarity of this scenario, and the extra complexity
914			 * it adds, we ignore the possibility.
915			 */
916			if (!vdev_is_dead(tvd))
917				spa_spare_activate(tvd);
918		}
919
920		vd->vdev_top = vd;
921		vd->vdev_aux = &spa->spa_spares;
922
923		if (vdev_open(vd) != 0)
924			continue;
925
926		if (vdev_validate_aux(vd) == 0)
927			spa_spare_add(vd);
928	}
929
930	/*
931	 * Recompute the stashed list of spares, with status information
932	 * this time.
933	 */
934	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
935	    DATA_TYPE_NVLIST_ARRAY) == 0);
936
937	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
938	    KM_SLEEP);
939	for (i = 0; i < spa->spa_spares.sav_count; i++)
940		spares[i] = vdev_config_generate(spa,
941		    spa->spa_spares.sav_vdevs[i], B_TRUE, B_TRUE, B_FALSE);
942	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
943	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
944	for (i = 0; i < spa->spa_spares.sav_count; i++)
945		nvlist_free(spares[i]);
946	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
947}
948
949/*
950 * Load (or re-load) the current list of vdevs describing the active l2cache for
951 * this pool.  When this is called, we have some form of basic information in
952 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
953 * then re-generate a more complete list including status information.
954 * Devices which are already active have their details maintained, and are
955 * not re-opened.
956 */
957static void
958spa_load_l2cache(spa_t *spa)
959{
960	nvlist_t **l2cache;
961	uint_t nl2cache;
962	int i, j, oldnvdevs;
963	uint64_t guid;
964	vdev_t *vd, **oldvdevs, **newvdevs;
965	spa_aux_vdev_t *sav = &spa->spa_l2cache;
966
967	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
968
969	if (sav->sav_config != NULL) {
970		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
971		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
972		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
973	} else {
974		nl2cache = 0;
975	}
976
977	oldvdevs = sav->sav_vdevs;
978	oldnvdevs = sav->sav_count;
979	sav->sav_vdevs = NULL;
980	sav->sav_count = 0;
981
982	/*
983	 * Process new nvlist of vdevs.
984	 */
985	for (i = 0; i < nl2cache; i++) {
986		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
987		    &guid) == 0);
988
989		newvdevs[i] = NULL;
990		for (j = 0; j < oldnvdevs; j++) {
991			vd = oldvdevs[j];
992			if (vd != NULL && guid == vd->vdev_guid) {
993				/*
994				 * Retain previous vdev for add/remove ops.
995				 */
996				newvdevs[i] = vd;
997				oldvdevs[j] = NULL;
998				break;
999			}
1000		}
1001
1002		if (newvdevs[i] == NULL) {
1003			/*
1004			 * Create new vdev
1005			 */
1006			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1007			    VDEV_ALLOC_L2CACHE) == 0);
1008			ASSERT(vd != NULL);
1009			newvdevs[i] = vd;
1010
1011			/*
1012			 * Commit this vdev as an l2cache device,
1013			 * even if it fails to open.
1014			 */
1015			spa_l2cache_add(vd);
1016
1017			vd->vdev_top = vd;
1018			vd->vdev_aux = sav;
1019
1020			spa_l2cache_activate(vd);
1021
1022			if (vdev_open(vd) != 0)
1023				continue;
1024
1025			(void) vdev_validate_aux(vd);
1026
1027			if (!vdev_is_dead(vd))
1028				l2arc_add_vdev(spa, vd);
1029		}
1030	}
1031
1032	/*
1033	 * Purge vdevs that were dropped
1034	 */
1035	for (i = 0; i < oldnvdevs; i++) {
1036		uint64_t pool;
1037
1038		vd = oldvdevs[i];
1039		if (vd != NULL) {
1040			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1041			    pool != 0ULL && l2arc_vdev_present(vd))
1042				l2arc_remove_vdev(vd);
1043			(void) vdev_close(vd);
1044			spa_l2cache_remove(vd);
1045		}
1046	}
1047
1048	if (oldvdevs)
1049		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1050
1051	if (sav->sav_config == NULL)
1052		goto out;
1053
1054	sav->sav_vdevs = newvdevs;
1055	sav->sav_count = (int)nl2cache;
1056
1057	/*
1058	 * Recompute the stashed list of l2cache devices, with status
1059	 * information this time.
1060	 */
1061	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1062	    DATA_TYPE_NVLIST_ARRAY) == 0);
1063
1064	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1065	for (i = 0; i < sav->sav_count; i++)
1066		l2cache[i] = vdev_config_generate(spa,
1067		    sav->sav_vdevs[i], B_TRUE, B_FALSE, B_TRUE);
1068	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1069	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1070out:
1071	for (i = 0; i < sav->sav_count; i++)
1072		nvlist_free(l2cache[i]);
1073	if (sav->sav_count)
1074		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1075}
1076
1077static int
1078load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1079{
1080	dmu_buf_t *db;
1081	char *packed = NULL;
1082	size_t nvsize = 0;
1083	int error;
1084	*value = NULL;
1085
1086	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1087	nvsize = *(uint64_t *)db->db_data;
1088	dmu_buf_rele(db, FTAG);
1089
1090	packed = kmem_alloc(nvsize, KM_SLEEP);
1091	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1092	    DMU_READ_PREFETCH);
1093	if (error == 0)
1094		error = nvlist_unpack(packed, nvsize, value, 0);
1095	kmem_free(packed, nvsize);
1096
1097	return (error);
1098}
1099
1100/*
1101 * Checks to see if the given vdev could not be opened, in which case we post a
1102 * sysevent to notify the autoreplace code that the device has been removed.
1103 */
1104static void
1105spa_check_removed(vdev_t *vd)
1106{
1107	for (int c = 0; c < vd->vdev_children; c++)
1108		spa_check_removed(vd->vdev_child[c]);
1109
1110	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1111		zfs_post_autoreplace(vd->vdev_spa, vd);
1112		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1113	}
1114}
1115
1116/*
1117 * Load the slog device state from the config object since it's possible
1118 * that the label does not contain the most up-to-date information.
1119 */
1120void
1121spa_load_log_state(spa_t *spa, nvlist_t *nv)
1122{
1123	vdev_t *ovd, *rvd = spa->spa_root_vdev;
1124
1125	/*
1126	 * Load the original root vdev tree from the passed config.
1127	 */
1128	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1129	VERIFY(spa_config_parse(spa, &ovd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1130
1131	for (int c = 0; c < rvd->vdev_children; c++) {
1132		vdev_t *cvd = rvd->vdev_child[c];
1133		if (cvd->vdev_islog)
1134			vdev_load_log_state(cvd, ovd->vdev_child[c]);
1135	}
1136	vdev_free(ovd);
1137	spa_config_exit(spa, SCL_ALL, FTAG);
1138}
1139
1140/*
1141 * Check for missing log devices
1142 */
1143int
1144spa_check_logs(spa_t *spa)
1145{
1146	switch (spa->spa_log_state) {
1147	case SPA_LOG_MISSING:
1148		/* need to recheck in case slog has been restored */
1149	case SPA_LOG_UNKNOWN:
1150		if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1151		    DS_FIND_CHILDREN)) {
1152			spa->spa_log_state = SPA_LOG_MISSING;
1153			return (1);
1154		}
1155		break;
1156	}
1157	return (0);
1158}
1159
1160static void
1161spa_aux_check_removed(spa_aux_vdev_t *sav)
1162{
1163	for (int i = 0; i < sav->sav_count; i++)
1164		spa_check_removed(sav->sav_vdevs[i]);
1165}
1166
1167void
1168spa_claim_notify(zio_t *zio)
1169{
1170	spa_t *spa = zio->io_spa;
1171
1172	if (zio->io_error)
1173		return;
1174
1175	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1176	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1177		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1178	mutex_exit(&spa->spa_props_lock);
1179}
1180
1181typedef struct spa_load_error {
1182	uint64_t	sle_metadata_count;
1183	uint64_t	sle_data_count;
1184} spa_load_error_t;
1185
1186static void
1187spa_load_verify_done(zio_t *zio)
1188{
1189	blkptr_t *bp = zio->io_bp;
1190	spa_load_error_t *sle = zio->io_private;
1191	dmu_object_type_t type = BP_GET_TYPE(bp);
1192	int error = zio->io_error;
1193
1194	if (error) {
1195		if ((BP_GET_LEVEL(bp) != 0 || dmu_ot[type].ot_metadata) &&
1196		    type != DMU_OT_INTENT_LOG)
1197			atomic_add_64(&sle->sle_metadata_count, 1);
1198		else
1199			atomic_add_64(&sle->sle_data_count, 1);
1200	}
1201	zio_data_buf_free(zio->io_data, zio->io_size);
1202}
1203
1204/*ARGSUSED*/
1205static int
1206spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1207    const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1208{
1209	if (bp != NULL) {
1210		zio_t *rio = arg;
1211		size_t size = BP_GET_PSIZE(bp);
1212		void *data = zio_data_buf_alloc(size);
1213
1214		zio_nowait(zio_read(rio, spa, bp, data, size,
1215		    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1216		    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1217		    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1218	}
1219	return (0);
1220}
1221
1222static int
1223spa_load_verify(spa_t *spa)
1224{
1225	zio_t *rio;
1226	spa_load_error_t sle = { 0 };
1227	zpool_rewind_policy_t policy;
1228	boolean_t verify_ok = B_FALSE;
1229	int error;
1230
1231	rio = zio_root(spa, NULL, &sle,
1232	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1233
1234	error = traverse_pool(spa, spa_load_verify_cb, rio,
1235	    spa->spa_verify_min_txg);
1236
1237	(void) zio_wait(rio);
1238
1239	zpool_get_rewind_policy(spa->spa_config, &policy);
1240
1241	spa->spa_load_meta_errors = sle.sle_metadata_count;
1242	spa->spa_load_data_errors = sle.sle_data_count;
1243
1244	if (!error && sle.sle_metadata_count <= policy.zrp_maxmeta &&
1245	    sle.sle_data_count <= policy.zrp_maxdata) {
1246		verify_ok = B_TRUE;
1247		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1248		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1249	}
1250
1251	if (error) {
1252		if (error != ENXIO && error != EIO)
1253			error = EIO;
1254		return (error);
1255	}
1256
1257	return (verify_ok ? 0 : EIO);
1258}
1259
1260/*
1261 * Load an existing storage pool, using the pool's builtin spa_config as a
1262 * source of configuration information.
1263 */
1264static int
1265spa_load(spa_t *spa, spa_load_state_t state, int mosconfig)
1266{
1267	int error = 0;
1268	nvlist_t *nvconfig, *nvroot = NULL;
1269	vdev_t *rvd;
1270	uberblock_t *ub = &spa->spa_uberblock;
1271	uint64_t config_cache_txg = spa->spa_config_txg;
1272	uint64_t pool_guid;
1273	uint64_t version;
1274	uint64_t autoreplace = 0;
1275	int orig_mode = spa->spa_mode;
1276	char *ereport = FM_EREPORT_ZFS_POOL;
1277	nvlist_t *config = spa->spa_config;
1278
1279	/*
1280	 * If this is an untrusted config, access the pool in read-only mode.
1281	 * This prevents things like resilvering recently removed devices.
1282	 */
1283	if (!mosconfig)
1284		spa->spa_mode = FREAD;
1285
1286	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1287
1288	spa->spa_load_state = state;
1289
1290	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) ||
1291	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
1292		error = EINVAL;
1293		goto out;
1294	}
1295
1296	/*
1297	 * Versioning wasn't explicitly added to the label until later, so if
1298	 * it's not present treat it as the initial version.
1299	 */
1300	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &version) != 0)
1301		version = SPA_VERSION_INITIAL;
1302
1303	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1304	    &spa->spa_config_txg);
1305
1306	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1307	    spa_guid_exists(pool_guid, 0)) {
1308		error = EEXIST;
1309		goto out;
1310	}
1311
1312	spa->spa_load_guid = pool_guid;
1313
1314	/*
1315	 * Create "The Godfather" zio to hold all async IOs
1316	 */
1317	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1318	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1319
1320	/*
1321	 * Parse the configuration into a vdev tree.  We explicitly set the
1322	 * value that will be returned by spa_version() since parsing the
1323	 * configuration requires knowing the version number.
1324	 */
1325	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1326	spa->spa_ubsync.ub_version = version;
1327	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_LOAD);
1328	spa_config_exit(spa, SCL_ALL, FTAG);
1329
1330	if (error != 0)
1331		goto out;
1332
1333	ASSERT(spa->spa_root_vdev == rvd);
1334	ASSERT(spa_guid(spa) == pool_guid);
1335
1336	/*
1337	 * Try to open all vdevs, loading each label in the process.
1338	 */
1339	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1340	error = vdev_open(rvd);
1341	spa_config_exit(spa, SCL_ALL, FTAG);
1342	if (error != 0)
1343		goto out;
1344
1345	/*
1346	 * We need to validate the vdev labels against the configuration that
1347	 * we have in hand, which is dependent on the setting of mosconfig. If
1348	 * mosconfig is true then we're validating the vdev labels based on
1349	 * that config. Otherwise, we're validating against the cached config
1350	 * (zpool.cache) that was read when we loaded the zfs module, and then
1351	 * later we will recursively call spa_load() and validate against
1352	 * the vdev config.
1353	 */
1354	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1355	error = vdev_validate(rvd);
1356	spa_config_exit(spa, SCL_ALL, FTAG);
1357	if (error != 0)
1358		goto out;
1359
1360	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
1361		error = ENXIO;
1362		goto out;
1363	}
1364
1365	/*
1366	 * Find the best uberblock.
1367	 */
1368	vdev_uberblock_load(NULL, rvd, ub);
1369
1370	/*
1371	 * If we weren't able to find a single valid uberblock, return failure.
1372	 */
1373	if (ub->ub_txg == 0) {
1374		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1375		    VDEV_AUX_CORRUPT_DATA);
1376		error = ENXIO;
1377		goto out;
1378	}
1379
1380	/*
1381	 * If the pool is newer than the code, we can't open it.
1382	 */
1383	if (ub->ub_version > SPA_VERSION) {
1384		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1385		    VDEV_AUX_VERSION_NEWER);
1386		error = ENOTSUP;
1387		goto out;
1388	}
1389
1390	/*
1391	 * If the vdev guid sum doesn't match the uberblock, we have an
1392	 * incomplete configuration.
1393	 */
1394	if (rvd->vdev_guid_sum != ub->ub_guid_sum && mosconfig) {
1395		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1396		    VDEV_AUX_BAD_GUID_SUM);
1397		error = ENXIO;
1398		goto out;
1399	}
1400
1401	/*
1402	 * Initialize internal SPA structures.
1403	 */
1404	spa->spa_state = POOL_STATE_ACTIVE;
1405	spa->spa_ubsync = spa->spa_uberblock;
1406	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
1407	    TXG_INITIAL : spa_last_synced_txg(spa) - TXG_DEFER_SIZE;
1408	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
1409	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
1410	spa->spa_claim_max_txg = spa->spa_first_txg;
1411
1412	error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1413	if (error) {
1414		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1415		    VDEV_AUX_CORRUPT_DATA);
1416		error = EIO;
1417		goto out;
1418	}
1419	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1420
1421	if (zap_lookup(spa->spa_meta_objset,
1422	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
1423	    sizeof (uint64_t), 1, &spa->spa_config_object) != 0) {
1424		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1425		    VDEV_AUX_CORRUPT_DATA);
1426		error = EIO;
1427		goto out;
1428	}
1429
1430	if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) {
1431		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1432		    VDEV_AUX_CORRUPT_DATA);
1433		error = EIO;
1434		goto out;
1435	}
1436
1437	if (!mosconfig) {
1438		uint64_t hostid;
1439
1440		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
1441		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
1442			char *hostname;
1443			unsigned long myhostid = 0;
1444
1445			VERIFY(nvlist_lookup_string(nvconfig,
1446			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
1447
1448#ifdef	_KERNEL
1449			myhostid = zone_get_hostid(NULL);
1450#else	/* _KERNEL */
1451			/*
1452			 * We're emulating the system's hostid in userland, so
1453			 * we can't use zone_get_hostid().
1454			 */
1455			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
1456#endif	/* _KERNEL */
1457			if (hostid != 0 && myhostid != 0 &&
1458			    hostid != myhostid) {
1459				cmn_err(CE_WARN, "pool '%s' could not be "
1460				    "loaded as it was last accessed by "
1461				    "another system (host: %s hostid: 0x%lx). "
1462				    "See: http://www.sun.com/msg/ZFS-8000-EY",
1463				    spa_name(spa), hostname,
1464				    (unsigned long)hostid);
1465				error = EBADF;
1466				goto out;
1467			}
1468		}
1469
1470		spa_config_set(spa, nvconfig);
1471		spa_unload(spa);
1472		spa_deactivate(spa);
1473		spa_activate(spa, orig_mode);
1474
1475		return (spa_load(spa, state, B_TRUE));
1476	}
1477
1478	if (zap_lookup(spa->spa_meta_objset,
1479	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
1480	    sizeof (uint64_t), 1, &spa->spa_deferred_bplist_obj) != 0) {
1481		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1482		    VDEV_AUX_CORRUPT_DATA);
1483		error = EIO;
1484		goto out;
1485	}
1486
1487	/*
1488	 * Load the bit that tells us to use the new accounting function
1489	 * (raid-z deflation).  If we have an older pool, this will not
1490	 * be present.
1491	 */
1492	error = zap_lookup(spa->spa_meta_objset,
1493	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
1494	    sizeof (uint64_t), 1, &spa->spa_deflate);
1495	if (error != 0 && error != ENOENT) {
1496		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1497		    VDEV_AUX_CORRUPT_DATA);
1498		error = EIO;
1499		goto out;
1500	}
1501
1502	/*
1503	 * Load the persistent error log.  If we have an older pool, this will
1504	 * not be present.
1505	 */
1506	error = zap_lookup(spa->spa_meta_objset,
1507	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_LAST,
1508	    sizeof (uint64_t), 1, &spa->spa_errlog_last);
1509	if (error != 0 && error != ENOENT) {
1510		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1511		    VDEV_AUX_CORRUPT_DATA);
1512		error = EIO;
1513		goto out;
1514	}
1515
1516	error = zap_lookup(spa->spa_meta_objset,
1517	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_SCRUB,
1518	    sizeof (uint64_t), 1, &spa->spa_errlog_scrub);
1519	if (error != 0 && error != ENOENT) {
1520		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1521		    VDEV_AUX_CORRUPT_DATA);
1522		error = EIO;
1523		goto out;
1524	}
1525
1526	/*
1527	 * Load the history object.  If we have an older pool, this
1528	 * will not be present.
1529	 */
1530	error = zap_lookup(spa->spa_meta_objset,
1531	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_HISTORY,
1532	    sizeof (uint64_t), 1, &spa->spa_history);
1533	if (error != 0 && error != ENOENT) {
1534		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1535		    VDEV_AUX_CORRUPT_DATA);
1536		error = EIO;
1537		goto out;
1538	}
1539
1540	/*
1541	 * Load any hot spares for this pool.
1542	 */
1543	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1544	    DMU_POOL_SPARES, sizeof (uint64_t), 1, &spa->spa_spares.sav_object);
1545	if (error != 0 && error != ENOENT) {
1546		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1547		    VDEV_AUX_CORRUPT_DATA);
1548		error = EIO;
1549		goto out;
1550	}
1551	if (error == 0) {
1552		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
1553		if (load_nvlist(spa, spa->spa_spares.sav_object,
1554		    &spa->spa_spares.sav_config) != 0) {
1555			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1556			    VDEV_AUX_CORRUPT_DATA);
1557			error = EIO;
1558			goto out;
1559		}
1560
1561		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1562		spa_load_spares(spa);
1563		spa_config_exit(spa, SCL_ALL, FTAG);
1564	}
1565
1566	/*
1567	 * Load any level 2 ARC devices for this pool.
1568	 */
1569	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1570	    DMU_POOL_L2CACHE, sizeof (uint64_t), 1,
1571	    &spa->spa_l2cache.sav_object);
1572	if (error != 0 && error != ENOENT) {
1573		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1574		    VDEV_AUX_CORRUPT_DATA);
1575		error = EIO;
1576		goto out;
1577	}
1578	if (error == 0) {
1579		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
1580		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
1581		    &spa->spa_l2cache.sav_config) != 0) {
1582			vdev_set_state(rvd, B_TRUE,
1583			    VDEV_STATE_CANT_OPEN,
1584			    VDEV_AUX_CORRUPT_DATA);
1585			error = EIO;
1586			goto out;
1587		}
1588
1589		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1590		spa_load_l2cache(spa);
1591		spa_config_exit(spa, SCL_ALL, FTAG);
1592	}
1593
1594	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
1595
1596	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1597	    DMU_POOL_PROPS, sizeof (uint64_t), 1, &spa->spa_pool_props_object);
1598
1599	if (error && error != ENOENT) {
1600		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1601		    VDEV_AUX_CORRUPT_DATA);
1602		error = EIO;
1603		goto out;
1604	}
1605
1606	if (error == 0) {
1607		(void) zap_lookup(spa->spa_meta_objset,
1608		    spa->spa_pool_props_object,
1609		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS),
1610		    sizeof (uint64_t), 1, &spa->spa_bootfs);
1611		(void) zap_lookup(spa->spa_meta_objset,
1612		    spa->spa_pool_props_object,
1613		    zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE),
1614		    sizeof (uint64_t), 1, &autoreplace);
1615		spa->spa_autoreplace = (autoreplace != 0);
1616		(void) zap_lookup(spa->spa_meta_objset,
1617		    spa->spa_pool_props_object,
1618		    zpool_prop_to_name(ZPOOL_PROP_DELEGATION),
1619		    sizeof (uint64_t), 1, &spa->spa_delegation);
1620		(void) zap_lookup(spa->spa_meta_objset,
1621		    spa->spa_pool_props_object,
1622		    zpool_prop_to_name(ZPOOL_PROP_FAILUREMODE),
1623		    sizeof (uint64_t), 1, &spa->spa_failmode);
1624		(void) zap_lookup(spa->spa_meta_objset,
1625		    spa->spa_pool_props_object,
1626		    zpool_prop_to_name(ZPOOL_PROP_AUTOEXPAND),
1627		    sizeof (uint64_t), 1, &spa->spa_autoexpand);
1628		(void) zap_lookup(spa->spa_meta_objset,
1629		    spa->spa_pool_props_object,
1630		    zpool_prop_to_name(ZPOOL_PROP_DEDUPDITTO),
1631		    sizeof (uint64_t), 1, &spa->spa_dedup_ditto);
1632	}
1633
1634	/*
1635	 * If the 'autoreplace' property is set, then post a resource notifying
1636	 * the ZFS DE that it should not issue any faults for unopenable
1637	 * devices.  We also iterate over the vdevs, and post a sysevent for any
1638	 * unopenable vdevs so that the normal autoreplace handler can take
1639	 * over.
1640	 */
1641	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
1642		spa_check_removed(spa->spa_root_vdev);
1643		/*
1644		 * For the import case, this is done in spa_import(), because
1645		 * at this point we're using the spare definitions from
1646		 * the MOS config, not necessarily from the userland config.
1647		 */
1648		if (state != SPA_LOAD_IMPORT) {
1649			spa_aux_check_removed(&spa->spa_spares);
1650			spa_aux_check_removed(&spa->spa_l2cache);
1651		}
1652	}
1653
1654	/*
1655	 * Load the vdev state for all toplevel vdevs.
1656	 */
1657	vdev_load(rvd);
1658
1659	/*
1660	 * Propagate the leaf DTLs we just loaded all the way up the tree.
1661	 */
1662	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1663	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
1664	spa_config_exit(spa, SCL_ALL, FTAG);
1665
1666	/*
1667	 * Check the state of the root vdev.  If it can't be opened, it
1668	 * indicates one or more toplevel vdevs are faulted.
1669	 */
1670	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
1671		error = ENXIO;
1672		goto out;
1673	}
1674
1675	/*
1676	 * Load the DDTs (dedup tables).
1677	 */
1678	error = ddt_load(spa);
1679	if (error != 0) {
1680		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1681		    VDEV_AUX_CORRUPT_DATA);
1682		error = EIO;
1683		goto out;
1684	}
1685
1686	if (state != SPA_LOAD_TRYIMPORT) {
1687		error = spa_load_verify(spa);
1688		if (error) {
1689			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1690			    VDEV_AUX_CORRUPT_DATA);
1691			goto out;
1692		}
1693	}
1694
1695	/*
1696	 * Load the intent log state and check log integrity.
1697	 */
1698	VERIFY(nvlist_lookup_nvlist(nvconfig, ZPOOL_CONFIG_VDEV_TREE,
1699	    &nvroot) == 0);
1700	spa_load_log_state(spa, nvroot);
1701	nvlist_free(nvconfig);
1702
1703	if (spa_check_logs(spa)) {
1704		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1705		    VDEV_AUX_BAD_LOG);
1706		error = ENXIO;
1707		ereport = FM_EREPORT_ZFS_LOG_REPLAY;
1708		goto out;
1709	}
1710
1711	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
1712	    spa->spa_load_max_txg == UINT64_MAX)) {
1713		dmu_tx_t *tx;
1714		int need_update = B_FALSE;
1715
1716		ASSERT(state != SPA_LOAD_TRYIMPORT);
1717
1718		/*
1719		 * Claim log blocks that haven't been committed yet.
1720		 * This must all happen in a single txg.
1721		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
1722		 * invoked from zil_claim_log_block()'s i/o done callback.
1723		 * Price of rollback is that we abandon the log.
1724		 */
1725		spa->spa_claiming = B_TRUE;
1726
1727		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
1728		    spa_first_txg(spa));
1729		(void) dmu_objset_find(spa_name(spa),
1730		    zil_claim, tx, DS_FIND_CHILDREN);
1731		dmu_tx_commit(tx);
1732
1733		spa->spa_claiming = B_FALSE;
1734
1735		spa->spa_log_state = SPA_LOG_GOOD;
1736		spa->spa_sync_on = B_TRUE;
1737		txg_sync_start(spa->spa_dsl_pool);
1738
1739		/*
1740		 * Wait for all claims to sync.  We sync up to the highest
1741		 * claimed log block birth time so that claimed log blocks
1742		 * don't appear to be from the future.  spa_claim_max_txg
1743		 * will have been set for us by either zil_check_log_chain()
1744		 * (invoked from spa_check_logs()) or zil_claim() above.
1745		 */
1746		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
1747
1748		/*
1749		 * If the config cache is stale, or we have uninitialized
1750		 * metaslabs (see spa_vdev_add()), then update the config.
1751		 *
1752		 * If spa_load_verbatim is true, trust the current
1753		 * in-core spa_config and update the disk labels.
1754		 */
1755		if (config_cache_txg != spa->spa_config_txg ||
1756		    state == SPA_LOAD_IMPORT || spa->spa_load_verbatim ||
1757		    state == SPA_LOAD_RECOVER)
1758			need_update = B_TRUE;
1759
1760		for (int c = 0; c < rvd->vdev_children; c++)
1761			if (rvd->vdev_child[c]->vdev_ms_array == 0)
1762				need_update = B_TRUE;
1763
1764		/*
1765		 * Update the config cache asychronously in case we're the
1766		 * root pool, in which case the config cache isn't writable yet.
1767		 */
1768		if (need_update)
1769			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
1770
1771		/*
1772		 * Check all DTLs to see if anything needs resilvering.
1773		 */
1774		if (vdev_resilver_needed(rvd, NULL, NULL))
1775			spa_async_request(spa, SPA_ASYNC_RESILVER);
1776
1777		/*
1778		 * Delete any inconsistent datasets.
1779		 */
1780		(void) dmu_objset_find(spa_name(spa),
1781		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
1782
1783		/*
1784		 * Clean up any stale temporary dataset userrefs.
1785		 */
1786		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
1787	}
1788
1789	error = 0;
1790out:
1791
1792	spa->spa_minref = refcount_count(&spa->spa_refcount);
1793	if (error && error != EBADF)
1794		zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1795	spa->spa_load_state = SPA_LOAD_NONE;
1796	spa->spa_ena = 0;
1797
1798	return (error);
1799}
1800
1801static int
1802spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
1803{
1804	spa_unload(spa);
1805	spa_deactivate(spa);
1806
1807	spa->spa_load_max_txg--;
1808
1809	spa_activate(spa, spa_mode_global);
1810	spa_async_suspend(spa);
1811
1812	return (spa_load(spa, state, mosconfig));
1813}
1814
1815static int
1816spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
1817    uint64_t max_request, boolean_t extreme)
1818{
1819	nvlist_t *config = NULL;
1820	int load_error, rewind_error;
1821	uint64_t safe_rollback_txg;
1822	uint64_t min_txg;
1823
1824	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER)
1825		spa->spa_load_max_txg = spa->spa_load_txg;
1826	else
1827		spa->spa_load_max_txg = max_request;
1828
1829	load_error = rewind_error = spa_load(spa, state, mosconfig);
1830	if (load_error == 0)
1831		return (0);
1832
1833	if (spa->spa_root_vdev != NULL)
1834		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
1835
1836	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
1837	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
1838
1839	/* specific txg requested */
1840	if (spa->spa_load_max_txg != UINT64_MAX && !extreme) {
1841		nvlist_free(config);
1842		return (load_error);
1843	}
1844
1845	/* Price of rolling back is discarding txgs, including log */
1846	if (state == SPA_LOAD_RECOVER)
1847		spa->spa_log_state = SPA_LOG_CLEAR;
1848
1849	spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1850	safe_rollback_txg = spa->spa_uberblock.ub_txg - TXG_DEFER_SIZE;
1851
1852	min_txg = extreme ? TXG_INITIAL : safe_rollback_txg;
1853	while (rewind_error && (spa->spa_uberblock.ub_txg >= min_txg)) {
1854		if (spa->spa_load_max_txg < safe_rollback_txg)
1855			spa->spa_extreme_rewind = B_TRUE;
1856		rewind_error = spa_load_retry(spa, state, mosconfig);
1857	}
1858
1859	if (config)
1860		spa_rewind_data_to_nvlist(spa, config);
1861
1862	spa->spa_extreme_rewind = B_FALSE;
1863	spa->spa_load_max_txg = UINT64_MAX;
1864
1865	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
1866		spa_config_set(spa, config);
1867
1868	return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
1869}
1870
1871/*
1872 * Pool Open/Import
1873 *
1874 * The import case is identical to an open except that the configuration is sent
1875 * down from userland, instead of grabbed from the configuration cache.  For the
1876 * case of an open, the pool configuration will exist in the
1877 * POOL_STATE_UNINITIALIZED state.
1878 *
1879 * The stats information (gen/count/ustats) is used to gather vdev statistics at
1880 * the same time open the pool, without having to keep around the spa_t in some
1881 * ambiguous state.
1882 */
1883static int
1884spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
1885    nvlist_t **config)
1886{
1887	spa_t *spa;
1888	boolean_t norewind;
1889	boolean_t extreme;
1890	zpool_rewind_policy_t policy;
1891	spa_load_state_t state = SPA_LOAD_OPEN;
1892	int error;
1893	int locked = B_FALSE;
1894
1895	*spapp = NULL;
1896
1897	zpool_get_rewind_policy(nvpolicy, &policy);
1898	if (policy.zrp_request & ZPOOL_DO_REWIND)
1899		state = SPA_LOAD_RECOVER;
1900	norewind = (policy.zrp_request == ZPOOL_NO_REWIND);
1901	extreme = ((policy.zrp_request & ZPOOL_EXTREME_REWIND) != 0);
1902
1903	/*
1904	 * As disgusting as this is, we need to support recursive calls to this
1905	 * function because dsl_dir_open() is called during spa_load(), and ends
1906	 * up calling spa_open() again.  The real fix is to figure out how to
1907	 * avoid dsl_dir_open() calling this in the first place.
1908	 */
1909	if (mutex_owner(&spa_namespace_lock) != curthread) {
1910		mutex_enter(&spa_namespace_lock);
1911		locked = B_TRUE;
1912	}
1913
1914	if ((spa = spa_lookup(pool)) == NULL) {
1915		if (locked)
1916			mutex_exit(&spa_namespace_lock);
1917		return (ENOENT);
1918	}
1919
1920	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
1921
1922		spa_activate(spa, spa_mode_global);
1923
1924		if (spa->spa_last_open_failed && norewind) {
1925			if (config != NULL && spa->spa_config)
1926				VERIFY(nvlist_dup(spa->spa_config,
1927				    config, KM_SLEEP) == 0);
1928			spa_deactivate(spa);
1929			if (locked)
1930				mutex_exit(&spa_namespace_lock);
1931			return (spa->spa_last_open_failed);
1932		}
1933
1934		if (state != SPA_LOAD_RECOVER)
1935			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
1936
1937		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
1938		    extreme);
1939
1940		if (error == EBADF) {
1941			/*
1942			 * If vdev_validate() returns failure (indicated by
1943			 * EBADF), it indicates that one of the vdevs indicates
1944			 * that the pool has been exported or destroyed.  If
1945			 * this is the case, the config cache is out of sync and
1946			 * we should remove the pool from the namespace.
1947			 */
1948			spa_unload(spa);
1949			spa_deactivate(spa);
1950			spa_config_sync(spa, B_TRUE, B_TRUE);
1951			spa_remove(spa);
1952			if (locked)
1953				mutex_exit(&spa_namespace_lock);
1954			return (ENOENT);
1955		}
1956
1957		if (error) {
1958			/*
1959			 * We can't open the pool, but we still have useful
1960			 * information: the state of each vdev after the
1961			 * attempted vdev_open().  Return this to the user.
1962			 */
1963			if (config != NULL && spa->spa_config)
1964				VERIFY(nvlist_dup(spa->spa_config, config,
1965				    KM_SLEEP) == 0);
1966			spa_unload(spa);
1967			spa_deactivate(spa);
1968			spa->spa_last_open_failed = error;
1969			if (locked)
1970				mutex_exit(&spa_namespace_lock);
1971			*spapp = NULL;
1972			return (error);
1973		}
1974
1975	}
1976
1977	spa_open_ref(spa, tag);
1978
1979	spa->spa_last_open_failed = 0;
1980
1981	if (config != NULL)
1982		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
1983
1984	spa->spa_last_ubsync_txg = 0;
1985	spa->spa_load_txg = 0;
1986
1987	if (locked)
1988		mutex_exit(&spa_namespace_lock);
1989
1990	*spapp = spa;
1991
1992	return (0);
1993}
1994
1995int
1996spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
1997    nvlist_t **config)
1998{
1999	return (spa_open_common(name, spapp, tag, policy, config));
2000}
2001
2002int
2003spa_open(const char *name, spa_t **spapp, void *tag)
2004{
2005	return (spa_open_common(name, spapp, tag, NULL, NULL));
2006}
2007
2008/*
2009 * Lookup the given spa_t, incrementing the inject count in the process,
2010 * preventing it from being exported or destroyed.
2011 */
2012spa_t *
2013spa_inject_addref(char *name)
2014{
2015	spa_t *spa;
2016
2017	mutex_enter(&spa_namespace_lock);
2018	if ((spa = spa_lookup(name)) == NULL) {
2019		mutex_exit(&spa_namespace_lock);
2020		return (NULL);
2021	}
2022	spa->spa_inject_ref++;
2023	mutex_exit(&spa_namespace_lock);
2024
2025	return (spa);
2026}
2027
2028void
2029spa_inject_delref(spa_t *spa)
2030{
2031	mutex_enter(&spa_namespace_lock);
2032	spa->spa_inject_ref--;
2033	mutex_exit(&spa_namespace_lock);
2034}
2035
2036/*
2037 * Add spares device information to the nvlist.
2038 */
2039static void
2040spa_add_spares(spa_t *spa, nvlist_t *config)
2041{
2042	nvlist_t **spares;
2043	uint_t i, nspares;
2044	nvlist_t *nvroot;
2045	uint64_t guid;
2046	vdev_stat_t *vs;
2047	uint_t vsc;
2048	uint64_t pool;
2049
2050	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2051
2052	if (spa->spa_spares.sav_count == 0)
2053		return;
2054
2055	VERIFY(nvlist_lookup_nvlist(config,
2056	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2057	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2058	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2059	if (nspares != 0) {
2060		VERIFY(nvlist_add_nvlist_array(nvroot,
2061		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2062		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2063		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2064
2065		/*
2066		 * Go through and find any spares which have since been
2067		 * repurposed as an active spare.  If this is the case, update
2068		 * their status appropriately.
2069		 */
2070		for (i = 0; i < nspares; i++) {
2071			VERIFY(nvlist_lookup_uint64(spares[i],
2072			    ZPOOL_CONFIG_GUID, &guid) == 0);
2073			if (spa_spare_exists(guid, &pool, NULL) &&
2074			    pool != 0ULL) {
2075				VERIFY(nvlist_lookup_uint64_array(
2076				    spares[i], ZPOOL_CONFIG_STATS,
2077				    (uint64_t **)&vs, &vsc) == 0);
2078				vs->vs_state = VDEV_STATE_CANT_OPEN;
2079				vs->vs_aux = VDEV_AUX_SPARED;
2080			}
2081		}
2082	}
2083}
2084
2085/*
2086 * Add l2cache device information to the nvlist, including vdev stats.
2087 */
2088static void
2089spa_add_l2cache(spa_t *spa, nvlist_t *config)
2090{
2091	nvlist_t **l2cache;
2092	uint_t i, j, nl2cache;
2093	nvlist_t *nvroot;
2094	uint64_t guid;
2095	vdev_t *vd;
2096	vdev_stat_t *vs;
2097	uint_t vsc;
2098
2099	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2100
2101	if (spa->spa_l2cache.sav_count == 0)
2102		return;
2103
2104	VERIFY(nvlist_lookup_nvlist(config,
2105	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2106	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2107	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2108	if (nl2cache != 0) {
2109		VERIFY(nvlist_add_nvlist_array(nvroot,
2110		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2111		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2112		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2113
2114		/*
2115		 * Update level 2 cache device stats.
2116		 */
2117
2118		for (i = 0; i < nl2cache; i++) {
2119			VERIFY(nvlist_lookup_uint64(l2cache[i],
2120			    ZPOOL_CONFIG_GUID, &guid) == 0);
2121
2122			vd = NULL;
2123			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2124				if (guid ==
2125				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2126					vd = spa->spa_l2cache.sav_vdevs[j];
2127					break;
2128				}
2129			}
2130			ASSERT(vd != NULL);
2131
2132			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2133			    ZPOOL_CONFIG_STATS, (uint64_t **)&vs, &vsc) == 0);
2134			vdev_get_stats(vd, vs);
2135		}
2136	}
2137}
2138
2139int
2140spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2141{
2142	int error;
2143	spa_t *spa;
2144
2145	*config = NULL;
2146	error = spa_open_common(name, &spa, FTAG, NULL, config);
2147
2148	if (spa != NULL) {
2149		/*
2150		 * This still leaves a window of inconsistency where the spares
2151		 * or l2cache devices could change and the config would be
2152		 * self-inconsistent.
2153		 */
2154		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2155
2156		if (*config != NULL) {
2157			VERIFY(nvlist_add_uint64(*config,
2158			    ZPOOL_CONFIG_ERRCOUNT,
2159			    spa_get_errlog_size(spa)) == 0);
2160
2161			if (spa_suspended(spa))
2162				VERIFY(nvlist_add_uint64(*config,
2163				    ZPOOL_CONFIG_SUSPENDED,
2164				    spa->spa_failmode) == 0);
2165
2166			spa_add_spares(spa, *config);
2167			spa_add_l2cache(spa, *config);
2168		}
2169	}
2170
2171	/*
2172	 * We want to get the alternate root even for faulted pools, so we cheat
2173	 * and call spa_lookup() directly.
2174	 */
2175	if (altroot) {
2176		if (spa == NULL) {
2177			mutex_enter(&spa_namespace_lock);
2178			spa = spa_lookup(name);
2179			if (spa)
2180				spa_altroot(spa, altroot, buflen);
2181			else
2182				altroot[0] = '\0';
2183			spa = NULL;
2184			mutex_exit(&spa_namespace_lock);
2185		} else {
2186			spa_altroot(spa, altroot, buflen);
2187		}
2188	}
2189
2190	if (spa != NULL) {
2191		spa_config_exit(spa, SCL_CONFIG, FTAG);
2192		spa_close(spa, FTAG);
2193	}
2194
2195	return (error);
2196}
2197
2198/*
2199 * Validate that the auxiliary device array is well formed.  We must have an
2200 * array of nvlists, each which describes a valid leaf vdev.  If this is an
2201 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2202 * specified, as long as they are well-formed.
2203 */
2204static int
2205spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2206    spa_aux_vdev_t *sav, const char *config, uint64_t version,
2207    vdev_labeltype_t label)
2208{
2209	nvlist_t **dev;
2210	uint_t i, ndev;
2211	vdev_t *vd;
2212	int error;
2213
2214	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2215
2216	/*
2217	 * It's acceptable to have no devs specified.
2218	 */
2219	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2220		return (0);
2221
2222	if (ndev == 0)
2223		return (EINVAL);
2224
2225	/*
2226	 * Make sure the pool is formatted with a version that supports this
2227	 * device type.
2228	 */
2229	if (spa_version(spa) < version)
2230		return (ENOTSUP);
2231
2232	/*
2233	 * Set the pending device list so we correctly handle device in-use
2234	 * checking.
2235	 */
2236	sav->sav_pending = dev;
2237	sav->sav_npending = ndev;
2238
2239	for (i = 0; i < ndev; i++) {
2240		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2241		    mode)) != 0)
2242			goto out;
2243
2244		if (!vd->vdev_ops->vdev_op_leaf) {
2245			vdev_free(vd);
2246			error = EINVAL;
2247			goto out;
2248		}
2249
2250		/*
2251		 * The L2ARC currently only supports disk devices in
2252		 * kernel context.  For user-level testing, we allow it.
2253		 */
2254#ifdef _KERNEL
2255		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2256		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2257			error = ENOTBLK;
2258			goto out;
2259		}
2260#endif
2261		vd->vdev_top = vd;
2262
2263		if ((error = vdev_open(vd)) == 0 &&
2264		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
2265			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2266			    vd->vdev_guid) == 0);
2267		}
2268
2269		vdev_free(vd);
2270
2271		if (error &&
2272		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2273			goto out;
2274		else
2275			error = 0;
2276	}
2277
2278out:
2279	sav->sav_pending = NULL;
2280	sav->sav_npending = 0;
2281	return (error);
2282}
2283
2284static int
2285spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2286{
2287	int error;
2288
2289	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2290
2291	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2292	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2293	    VDEV_LABEL_SPARE)) != 0) {
2294		return (error);
2295	}
2296
2297	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2298	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2299	    VDEV_LABEL_L2CACHE));
2300}
2301
2302static void
2303spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2304    const char *config)
2305{
2306	int i;
2307
2308	if (sav->sav_config != NULL) {
2309		nvlist_t **olddevs;
2310		uint_t oldndevs;
2311		nvlist_t **newdevs;
2312
2313		/*
2314		 * Generate new dev list by concatentating with the
2315		 * current dev list.
2316		 */
2317		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2318		    &olddevs, &oldndevs) == 0);
2319
2320		newdevs = kmem_alloc(sizeof (void *) *
2321		    (ndevs + oldndevs), KM_SLEEP);
2322		for (i = 0; i < oldndevs; i++)
2323			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2324			    KM_SLEEP) == 0);
2325		for (i = 0; i < ndevs; i++)
2326			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2327			    KM_SLEEP) == 0);
2328
2329		VERIFY(nvlist_remove(sav->sav_config, config,
2330		    DATA_TYPE_NVLIST_ARRAY) == 0);
2331
2332		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2333		    config, newdevs, ndevs + oldndevs) == 0);
2334		for (i = 0; i < oldndevs + ndevs; i++)
2335			nvlist_free(newdevs[i]);
2336		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2337	} else {
2338		/*
2339		 * Generate a new dev list.
2340		 */
2341		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2342		    KM_SLEEP) == 0);
2343		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2344		    devs, ndevs) == 0);
2345	}
2346}
2347
2348/*
2349 * Stop and drop level 2 ARC devices
2350 */
2351void
2352spa_l2cache_drop(spa_t *spa)
2353{
2354	vdev_t *vd;
2355	int i;
2356	spa_aux_vdev_t *sav = &spa->spa_l2cache;
2357
2358	for (i = 0; i < sav->sav_count; i++) {
2359		uint64_t pool;
2360
2361		vd = sav->sav_vdevs[i];
2362		ASSERT(vd != NULL);
2363
2364		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2365		    pool != 0ULL && l2arc_vdev_present(vd))
2366			l2arc_remove_vdev(vd);
2367		if (vd->vdev_isl2cache)
2368			spa_l2cache_remove(vd);
2369		vdev_clear_stats(vd);
2370		(void) vdev_close(vd);
2371	}
2372}
2373
2374/*
2375 * Pool Creation
2376 */
2377int
2378spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2379    const char *history_str, nvlist_t *zplprops)
2380{
2381	spa_t *spa;
2382	char *altroot = NULL;
2383	vdev_t *rvd;
2384	dsl_pool_t *dp;
2385	dmu_tx_t *tx;
2386	int error = 0;
2387	uint64_t txg = TXG_INITIAL;
2388	nvlist_t **spares, **l2cache;
2389	uint_t nspares, nl2cache;
2390	uint64_t version;
2391
2392	/*
2393	 * If this pool already exists, return failure.
2394	 */
2395	mutex_enter(&spa_namespace_lock);
2396	if (spa_lookup(pool) != NULL) {
2397		mutex_exit(&spa_namespace_lock);
2398		return (EEXIST);
2399	}
2400
2401	/*
2402	 * Allocate a new spa_t structure.
2403	 */
2404	(void) nvlist_lookup_string(props,
2405	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2406	spa = spa_add(pool, NULL, altroot);
2407	spa_activate(spa, spa_mode_global);
2408
2409	if (props && (error = spa_prop_validate(spa, props))) {
2410		spa_deactivate(spa);
2411		spa_remove(spa);
2412		mutex_exit(&spa_namespace_lock);
2413		return (error);
2414	}
2415
2416	if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2417	    &version) != 0)
2418		version = SPA_VERSION;
2419	ASSERT(version <= SPA_VERSION);
2420
2421	spa->spa_first_txg = txg;
2422	spa->spa_uberblock.ub_txg = txg - 1;
2423	spa->spa_uberblock.ub_version = version;
2424	spa->spa_ubsync = spa->spa_uberblock;
2425
2426	/*
2427	 * Create "The Godfather" zio to hold all async IOs
2428	 */
2429	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2430	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2431
2432	/*
2433	 * Create the root vdev.
2434	 */
2435	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2436
2437	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2438
2439	ASSERT(error != 0 || rvd != NULL);
2440	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2441
2442	if (error == 0 && !zfs_allocatable_devs(nvroot))
2443		error = EINVAL;
2444
2445	if (error == 0 &&
2446	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2447	    (error = spa_validate_aux(spa, nvroot, txg,
2448	    VDEV_ALLOC_ADD)) == 0) {
2449		for (int c = 0; c < rvd->vdev_children; c++) {
2450			vdev_metaslab_set_size(rvd->vdev_child[c]);
2451			vdev_expand(rvd->vdev_child[c], txg);
2452		}
2453	}
2454
2455	spa_config_exit(spa, SCL_ALL, FTAG);
2456
2457	if (error != 0) {
2458		spa_unload(spa);
2459		spa_deactivate(spa);
2460		spa_remove(spa);
2461		mutex_exit(&spa_namespace_lock);
2462		return (error);
2463	}
2464
2465	/*
2466	 * Get the list of spares, if specified.
2467	 */
2468	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2469	    &spares, &nspares) == 0) {
2470		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
2471		    KM_SLEEP) == 0);
2472		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2473		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2474		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2475		spa_load_spares(spa);
2476		spa_config_exit(spa, SCL_ALL, FTAG);
2477		spa->spa_spares.sav_sync = B_TRUE;
2478	}
2479
2480	/*
2481	 * Get the list of level 2 cache devices, if specified.
2482	 */
2483	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2484	    &l2cache, &nl2cache) == 0) {
2485		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2486		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2487		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2488		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2489		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2490		spa_load_l2cache(spa);
2491		spa_config_exit(spa, SCL_ALL, FTAG);
2492		spa->spa_l2cache.sav_sync = B_TRUE;
2493	}
2494
2495	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
2496	spa->spa_meta_objset = dp->dp_meta_objset;
2497
2498	tx = dmu_tx_create_assigned(dp, txg);
2499
2500	/*
2501	 * Create the pool config object.
2502	 */
2503	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
2504	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
2505	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2506
2507	if (zap_add(spa->spa_meta_objset,
2508	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
2509	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
2510		cmn_err(CE_PANIC, "failed to add pool config");
2511	}
2512
2513	/* Newly created pools with the right version are always deflated. */
2514	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
2515		spa->spa_deflate = TRUE;
2516		if (zap_add(spa->spa_meta_objset,
2517		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
2518		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
2519			cmn_err(CE_PANIC, "failed to add deflate");
2520		}
2521	}
2522
2523	/*
2524	 * Create the deferred-free bplist object.  Turn off compression
2525	 * because sync-to-convergence takes longer if the blocksize
2526	 * keeps changing.
2527	 */
2528	spa->spa_deferred_bplist_obj = bplist_create(spa->spa_meta_objset,
2529	    1 << 14, tx);
2530	dmu_object_set_compress(spa->spa_meta_objset,
2531	    spa->spa_deferred_bplist_obj, ZIO_COMPRESS_OFF, tx);
2532
2533	if (zap_add(spa->spa_meta_objset,
2534	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
2535	    sizeof (uint64_t), 1, &spa->spa_deferred_bplist_obj, tx) != 0) {
2536		cmn_err(CE_PANIC, "failed to add bplist");
2537	}
2538
2539	/*
2540	 * Create the pool's history object.
2541	 */
2542	if (version >= SPA_VERSION_ZPOOL_HISTORY)
2543		spa_history_create_obj(spa, tx);
2544
2545	/*
2546	 * Set pool properties.
2547	 */
2548	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
2549	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2550	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
2551	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
2552
2553	if (props != NULL) {
2554		spa_configfile_set(spa, props, B_FALSE);
2555		spa_sync_props(spa, props, CRED(), tx);
2556	}
2557
2558	/*
2559	 * Create DDTs (dedup tables).
2560	 */
2561	ddt_create(spa);
2562
2563	dmu_tx_commit(tx);
2564
2565	spa->spa_sync_on = B_TRUE;
2566	txg_sync_start(spa->spa_dsl_pool);
2567
2568	/*
2569	 * We explicitly wait for the first transaction to complete so that our
2570	 * bean counters are appropriately updated.
2571	 */
2572	txg_wait_synced(spa->spa_dsl_pool, txg);
2573
2574	spa_config_sync(spa, B_FALSE, B_TRUE);
2575
2576	if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
2577		(void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
2578	spa_history_log_version(spa, LOG_POOL_CREATE);
2579
2580	spa->spa_minref = refcount_count(&spa->spa_refcount);
2581
2582	mutex_exit(&spa_namespace_lock);
2583
2584	return (0);
2585}
2586
2587#ifdef _KERNEL
2588/*
2589 * Get the root pool information from the root disk, then import the root pool
2590 * during the system boot up time.
2591 */
2592extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
2593
2594static nvlist_t *
2595spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
2596{
2597	nvlist_t *config;
2598	nvlist_t *nvtop, *nvroot;
2599	uint64_t pgid;
2600
2601	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
2602		return (NULL);
2603
2604	/*
2605	 * Add this top-level vdev to the child array.
2606	 */
2607	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2608	    &nvtop) == 0);
2609	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
2610	    &pgid) == 0);
2611	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
2612
2613	/*
2614	 * Put this pool's top-level vdevs into a root vdev.
2615	 */
2616	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2617	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
2618	    VDEV_TYPE_ROOT) == 0);
2619	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
2620	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
2621	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2622	    &nvtop, 1) == 0);
2623
2624	/*
2625	 * Replace the existing vdev_tree with the new root vdev in
2626	 * this pool's configuration (remove the old, add the new).
2627	 */
2628	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
2629	nvlist_free(nvroot);
2630	return (config);
2631}
2632
2633/*
2634 * Walk the vdev tree and see if we can find a device with "better"
2635 * configuration. A configuration is "better" if the label on that
2636 * device has a more recent txg.
2637 */
2638static void
2639spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
2640{
2641	for (int c = 0; c < vd->vdev_children; c++)
2642		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
2643
2644	if (vd->vdev_ops->vdev_op_leaf) {
2645		nvlist_t *label;
2646		uint64_t label_txg;
2647
2648		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
2649		    &label) != 0)
2650			return;
2651
2652		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
2653		    &label_txg) == 0);
2654
2655		/*
2656		 * Do we have a better boot device?
2657		 */
2658		if (label_txg > *txg) {
2659			*txg = label_txg;
2660			*avd = vd;
2661		}
2662		nvlist_free(label);
2663	}
2664}
2665
2666/*
2667 * Import a root pool.
2668 *
2669 * For x86. devpath_list will consist of devid and/or physpath name of
2670 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
2671 * The GRUB "findroot" command will return the vdev we should boot.
2672 *
2673 * For Sparc, devpath_list consists the physpath name of the booting device
2674 * no matter the rootpool is a single device pool or a mirrored pool.
2675 * e.g.
2676 *	"/pci@1f,0/ide@d/disk@0,0:a"
2677 */
2678int
2679spa_import_rootpool(char *devpath, char *devid)
2680{
2681	spa_t *spa;
2682	vdev_t *rvd, *bvd, *avd = NULL;
2683	nvlist_t *config, *nvtop;
2684	uint64_t guid, txg;
2685	char *pname;
2686	int error;
2687
2688	/*
2689	 * Read the label from the boot device and generate a configuration.
2690	 */
2691	config = spa_generate_rootconf(devpath, devid, &guid);
2692#if defined(_OBP) && defined(_KERNEL)
2693	if (config == NULL) {
2694		if (strstr(devpath, "/iscsi/ssd") != NULL) {
2695			/* iscsi boot */
2696			get_iscsi_bootpath_phy(devpath);
2697			config = spa_generate_rootconf(devpath, devid, &guid);
2698		}
2699	}
2700#endif
2701	if (config == NULL) {
2702		cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
2703		    devpath);
2704		return (EIO);
2705	}
2706
2707	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
2708	    &pname) == 0);
2709	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
2710
2711	mutex_enter(&spa_namespace_lock);
2712	if ((spa = spa_lookup(pname)) != NULL) {
2713		/*
2714		 * Remove the existing root pool from the namespace so that we
2715		 * can replace it with the correct config we just read in.
2716		 */
2717		spa_remove(spa);
2718	}
2719
2720	spa = spa_add(pname, config, NULL);
2721	spa->spa_is_root = B_TRUE;
2722	spa->spa_load_verbatim = B_TRUE;
2723
2724	/*
2725	 * Build up a vdev tree based on the boot device's label config.
2726	 */
2727	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2728	    &nvtop) == 0);
2729	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2730	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
2731	    VDEV_ALLOC_ROOTPOOL);
2732	spa_config_exit(spa, SCL_ALL, FTAG);
2733	if (error) {
2734		mutex_exit(&spa_namespace_lock);
2735		nvlist_free(config);
2736		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
2737		    pname);
2738		return (error);
2739	}
2740
2741	/*
2742	 * Get the boot vdev.
2743	 */
2744	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
2745		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
2746		    (u_longlong_t)guid);
2747		error = ENOENT;
2748		goto out;
2749	}
2750
2751	/*
2752	 * Determine if there is a better boot device.
2753	 */
2754	avd = bvd;
2755	spa_alt_rootvdev(rvd, &avd, &txg);
2756	if (avd != bvd) {
2757		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
2758		    "try booting from '%s'", avd->vdev_path);
2759		error = EINVAL;
2760		goto out;
2761	}
2762
2763	/*
2764	 * If the boot device is part of a spare vdev then ensure that
2765	 * we're booting off the active spare.
2766	 */
2767	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2768	    !bvd->vdev_isspare) {
2769		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
2770		    "try booting from '%s'",
2771		    bvd->vdev_parent->vdev_child[1]->vdev_path);
2772		error = EINVAL;
2773		goto out;
2774	}
2775
2776	VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
2777	error = 0;
2778	spa_history_log_version(spa, LOG_POOL_IMPORT);
2779out:
2780	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2781	vdev_free(rvd);
2782	spa_config_exit(spa, SCL_ALL, FTAG);
2783	mutex_exit(&spa_namespace_lock);
2784
2785	nvlist_free(config);
2786	return (error);
2787}
2788
2789#endif
2790
2791/*
2792 * Take a pool and insert it into the namespace as if it had been loaded at
2793 * boot.
2794 */
2795int
2796spa_import_verbatim(const char *pool, nvlist_t *config, nvlist_t *props)
2797{
2798	spa_t *spa;
2799	zpool_rewind_policy_t policy;
2800	char *altroot = NULL;
2801
2802	mutex_enter(&spa_namespace_lock);
2803	if (spa_lookup(pool) != NULL) {
2804		mutex_exit(&spa_namespace_lock);
2805		return (EEXIST);
2806	}
2807
2808	(void) nvlist_lookup_string(props,
2809	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2810	spa = spa_add(pool, config, altroot);
2811
2812	zpool_get_rewind_policy(config, &policy);
2813	spa->spa_load_max_txg = policy.zrp_txg;
2814
2815	spa->spa_load_verbatim = B_TRUE;
2816
2817	if (props != NULL)
2818		spa_configfile_set(spa, props, B_FALSE);
2819
2820	spa_config_sync(spa, B_FALSE, B_TRUE);
2821
2822	mutex_exit(&spa_namespace_lock);
2823	spa_history_log_version(spa, LOG_POOL_IMPORT);
2824
2825	return (0);
2826}
2827
2828/*
2829 * Import a non-root pool into the system.
2830 */
2831int
2832spa_import(const char *pool, nvlist_t *config, nvlist_t *props)
2833{
2834	spa_t *spa;
2835	char *altroot = NULL;
2836	spa_load_state_t state = SPA_LOAD_IMPORT;
2837	zpool_rewind_policy_t policy;
2838	int error;
2839	nvlist_t *nvroot;
2840	nvlist_t **spares, **l2cache;
2841	uint_t nspares, nl2cache;
2842
2843	/*
2844	 * If a pool with this name exists, return failure.
2845	 */
2846	mutex_enter(&spa_namespace_lock);
2847	if ((spa = spa_lookup(pool)) != NULL) {
2848		mutex_exit(&spa_namespace_lock);
2849		return (EEXIST);
2850	}
2851
2852	zpool_get_rewind_policy(config, &policy);
2853	if (policy.zrp_request & ZPOOL_DO_REWIND)
2854		state = SPA_LOAD_RECOVER;
2855
2856	/*
2857	 * Create and initialize the spa structure.
2858	 */
2859	(void) nvlist_lookup_string(props,
2860	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2861	spa = spa_add(pool, config, altroot);
2862	spa_activate(spa, spa_mode_global);
2863
2864	/*
2865	 * Don't start async tasks until we know everything is healthy.
2866	 */
2867	spa_async_suspend(spa);
2868
2869	/*
2870	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
2871	 * because the user-supplied config is actually the one to trust when
2872	 * doing an import.
2873	 */
2874	if (state != SPA_LOAD_RECOVER)
2875		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2876	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
2877	    ((policy.zrp_request & ZPOOL_EXTREME_REWIND) != 0));
2878
2879	/*
2880	 * Propagate anything learned about failing or best txgs
2881	 * back to caller
2882	 */
2883	spa_rewind_data_to_nvlist(spa, config);
2884
2885	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2886	/*
2887	 * Toss any existing sparelist, as it doesn't have any validity
2888	 * anymore, and conflicts with spa_has_spare().
2889	 */
2890	if (spa->spa_spares.sav_config) {
2891		nvlist_free(spa->spa_spares.sav_config);
2892		spa->spa_spares.sav_config = NULL;
2893		spa_load_spares(spa);
2894	}
2895	if (spa->spa_l2cache.sav_config) {
2896		nvlist_free(spa->spa_l2cache.sav_config);
2897		spa->spa_l2cache.sav_config = NULL;
2898		spa_load_l2cache(spa);
2899	}
2900
2901	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2902	    &nvroot) == 0);
2903	if (error == 0)
2904		error = spa_validate_aux(spa, nvroot, -1ULL,
2905		    VDEV_ALLOC_SPARE);
2906	if (error == 0)
2907		error = spa_validate_aux(spa, nvroot, -1ULL,
2908		    VDEV_ALLOC_L2CACHE);
2909	spa_config_exit(spa, SCL_ALL, FTAG);
2910
2911	if (props != NULL)
2912		spa_configfile_set(spa, props, B_FALSE);
2913
2914	if (error != 0 || (props && spa_writeable(spa) &&
2915	    (error = spa_prop_set(spa, props)))) {
2916		spa_unload(spa);
2917		spa_deactivate(spa);
2918		spa_remove(spa);
2919		mutex_exit(&spa_namespace_lock);
2920		return (error);
2921	}
2922
2923	spa_async_resume(spa);
2924
2925	/*
2926	 * Override any spares and level 2 cache devices as specified by
2927	 * the user, as these may have correct device names/devids, etc.
2928	 */
2929	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2930	    &spares, &nspares) == 0) {
2931		if (spa->spa_spares.sav_config)
2932			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
2933			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
2934		else
2935			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
2936			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2937		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2938		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2939		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2940		spa_load_spares(spa);
2941		spa_config_exit(spa, SCL_ALL, FTAG);
2942		spa->spa_spares.sav_sync = B_TRUE;
2943	}
2944	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2945	    &l2cache, &nl2cache) == 0) {
2946		if (spa->spa_l2cache.sav_config)
2947			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
2948			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
2949		else
2950			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2951			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2952		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2953		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2954		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2955		spa_load_l2cache(spa);
2956		spa_config_exit(spa, SCL_ALL, FTAG);
2957		spa->spa_l2cache.sav_sync = B_TRUE;
2958	}
2959
2960	/*
2961	 * Check for any removed devices.
2962	 */
2963	if (spa->spa_autoreplace) {
2964		spa_aux_check_removed(&spa->spa_spares);
2965		spa_aux_check_removed(&spa->spa_l2cache);
2966	}
2967
2968	if (spa_writeable(spa)) {
2969		/*
2970		 * Update the config cache to include the newly-imported pool.
2971		 */
2972		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
2973	}
2974
2975	/*
2976	 * It's possible that the pool was expanded while it was exported.
2977	 * We kick off an async task to handle this for us.
2978	 */
2979	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
2980
2981	mutex_exit(&spa_namespace_lock);
2982	spa_history_log_version(spa, LOG_POOL_IMPORT);
2983
2984	return (0);
2985}
2986
2987
2988/*
2989 * This (illegal) pool name is used when temporarily importing a spa_t in order
2990 * to get the vdev stats associated with the imported devices.
2991 */
2992#define	TRYIMPORT_NAME	"$import"
2993
2994nvlist_t *
2995spa_tryimport(nvlist_t *tryconfig)
2996{
2997	nvlist_t *config = NULL;
2998	char *poolname;
2999	spa_t *spa;
3000	uint64_t state;
3001	int error;
3002
3003	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3004		return (NULL);
3005
3006	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3007		return (NULL);
3008
3009	/*
3010	 * Create and initialize the spa structure.
3011	 */
3012	mutex_enter(&spa_namespace_lock);
3013	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3014	spa_activate(spa, FREAD);
3015
3016	/*
3017	 * Pass off the heavy lifting to spa_load().
3018	 * Pass TRUE for mosconfig because the user-supplied config
3019	 * is actually the one to trust when doing an import.
3020	 */
3021	error = spa_load(spa, SPA_LOAD_TRYIMPORT, B_TRUE);
3022
3023	/*
3024	 * If 'tryconfig' was at least parsable, return the current config.
3025	 */
3026	if (spa->spa_root_vdev != NULL) {
3027		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3028		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3029		    poolname) == 0);
3030		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3031		    state) == 0);
3032		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3033		    spa->spa_uberblock.ub_timestamp) == 0);
3034
3035		/*
3036		 * If the bootfs property exists on this pool then we
3037		 * copy it out so that external consumers can tell which
3038		 * pools are bootable.
3039		 */
3040		if ((!error || error == EEXIST) && spa->spa_bootfs) {
3041			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3042
3043			/*
3044			 * We have to play games with the name since the
3045			 * pool was opened as TRYIMPORT_NAME.
3046			 */
3047			if (dsl_dsobj_to_dsname(spa_name(spa),
3048			    spa->spa_bootfs, tmpname) == 0) {
3049				char *cp;
3050				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3051
3052				cp = strchr(tmpname, '/');
3053				if (cp == NULL) {
3054					(void) strlcpy(dsname, tmpname,
3055					    MAXPATHLEN);
3056				} else {
3057					(void) snprintf(dsname, MAXPATHLEN,
3058					    "%s/%s", poolname, ++cp);
3059				}
3060				VERIFY(nvlist_add_string(config,
3061				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3062				kmem_free(dsname, MAXPATHLEN);
3063			}
3064			kmem_free(tmpname, MAXPATHLEN);
3065		}
3066
3067		/*
3068		 * Add the list of hot spares and level 2 cache devices.
3069		 */
3070		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3071		spa_add_spares(spa, config);
3072		spa_add_l2cache(spa, config);
3073		spa_config_exit(spa, SCL_CONFIG, FTAG);
3074	}
3075
3076	spa_unload(spa);
3077	spa_deactivate(spa);
3078	spa_remove(spa);
3079	mutex_exit(&spa_namespace_lock);
3080
3081	return (config);
3082}
3083
3084/*
3085 * Pool export/destroy
3086 *
3087 * The act of destroying or exporting a pool is very simple.  We make sure there
3088 * is no more pending I/O and any references to the pool are gone.  Then, we
3089 * update the pool state and sync all the labels to disk, removing the
3090 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3091 * we don't sync the labels or remove the configuration cache.
3092 */
3093static int
3094spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3095    boolean_t force, boolean_t hardforce)
3096{
3097	spa_t *spa;
3098
3099	if (oldconfig)
3100		*oldconfig = NULL;
3101
3102	if (!(spa_mode_global & FWRITE))
3103		return (EROFS);
3104
3105	mutex_enter(&spa_namespace_lock);
3106	if ((spa = spa_lookup(pool)) == NULL) {
3107		mutex_exit(&spa_namespace_lock);
3108		return (ENOENT);
3109	}
3110
3111	/*
3112	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3113	 * reacquire the namespace lock, and see if we can export.
3114	 */
3115	spa_open_ref(spa, FTAG);
3116	mutex_exit(&spa_namespace_lock);
3117	spa_async_suspend(spa);
3118	mutex_enter(&spa_namespace_lock);
3119	spa_close(spa, FTAG);
3120
3121	/*
3122	 * The pool will be in core if it's openable,
3123	 * in which case we can modify its state.
3124	 */
3125	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3126		/*
3127		 * Objsets may be open only because they're dirty, so we
3128		 * have to force it to sync before checking spa_refcnt.
3129		 */
3130		txg_wait_synced(spa->spa_dsl_pool, 0);
3131
3132		/*
3133		 * A pool cannot be exported or destroyed if there are active
3134		 * references.  If we are resetting a pool, allow references by
3135		 * fault injection handlers.
3136		 */
3137		if (!spa_refcount_zero(spa) ||
3138		    (spa->spa_inject_ref != 0 &&
3139		    new_state != POOL_STATE_UNINITIALIZED)) {
3140			spa_async_resume(spa);
3141			mutex_exit(&spa_namespace_lock);
3142			return (EBUSY);
3143		}
3144
3145		/*
3146		 * A pool cannot be exported if it has an active shared spare.
3147		 * This is to prevent other pools stealing the active spare
3148		 * from an exported pool. At user's own will, such pool can
3149		 * be forcedly exported.
3150		 */
3151		if (!force && new_state == POOL_STATE_EXPORTED &&
3152		    spa_has_active_shared_spare(spa)) {
3153			spa_async_resume(spa);
3154			mutex_exit(&spa_namespace_lock);
3155			return (EXDEV);
3156		}
3157
3158		/*
3159		 * We want this to be reflected on every label,
3160		 * so mark them all dirty.  spa_unload() will do the
3161		 * final sync that pushes these changes out.
3162		 */
3163		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3164			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3165			spa->spa_state = new_state;
3166			spa->spa_final_txg = spa_last_synced_txg(spa) + 1;
3167			vdev_config_dirty(spa->spa_root_vdev);
3168			spa_config_exit(spa, SCL_ALL, FTAG);
3169		}
3170	}
3171
3172	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
3173
3174	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3175		spa_unload(spa);
3176		spa_deactivate(spa);
3177	}
3178
3179	if (oldconfig && spa->spa_config)
3180		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3181
3182	if (new_state != POOL_STATE_UNINITIALIZED) {
3183		if (!hardforce)
3184			spa_config_sync(spa, B_TRUE, B_TRUE);
3185		spa_remove(spa);
3186	}
3187	mutex_exit(&spa_namespace_lock);
3188
3189	return (0);
3190}
3191
3192/*
3193 * Destroy a storage pool.
3194 */
3195int
3196spa_destroy(char *pool)
3197{
3198	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3199	    B_FALSE, B_FALSE));
3200}
3201
3202/*
3203 * Export a storage pool.
3204 */
3205int
3206spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3207    boolean_t hardforce)
3208{
3209	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3210	    force, hardforce));
3211}
3212
3213/*
3214 * Similar to spa_export(), this unloads the spa_t without actually removing it
3215 * from the namespace in any way.
3216 */
3217int
3218spa_reset(char *pool)
3219{
3220	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3221	    B_FALSE, B_FALSE));
3222}
3223
3224/*
3225 * ==========================================================================
3226 * Device manipulation
3227 * ==========================================================================
3228 */
3229
3230/*
3231 * Add a device to a storage pool.
3232 */
3233int
3234spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3235{
3236	uint64_t txg, id;
3237	int error;
3238	vdev_t *rvd = spa->spa_root_vdev;
3239	vdev_t *vd, *tvd;
3240	nvlist_t **spares, **l2cache;
3241	uint_t nspares, nl2cache;
3242
3243	txg = spa_vdev_enter(spa);
3244
3245	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3246	    VDEV_ALLOC_ADD)) != 0)
3247		return (spa_vdev_exit(spa, NULL, txg, error));
3248
3249	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
3250
3251	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3252	    &nspares) != 0)
3253		nspares = 0;
3254
3255	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3256	    &nl2cache) != 0)
3257		nl2cache = 0;
3258
3259	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3260		return (spa_vdev_exit(spa, vd, txg, EINVAL));
3261
3262	if (vd->vdev_children != 0 &&
3263	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
3264		return (spa_vdev_exit(spa, vd, txg, error));
3265
3266	/*
3267	 * We must validate the spares and l2cache devices after checking the
3268	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
3269	 */
3270	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3271		return (spa_vdev_exit(spa, vd, txg, error));
3272
3273	/*
3274	 * Transfer each new top-level vdev from vd to rvd.
3275	 */
3276	for (int c = 0; c < vd->vdev_children; c++) {
3277
3278		/*
3279		 * Set the vdev id to the first hole, if one exists.
3280		 */
3281		for (id = 0; id < rvd->vdev_children; id++) {
3282			if (rvd->vdev_child[id]->vdev_ishole) {
3283				vdev_free(rvd->vdev_child[id]);
3284				break;
3285			}
3286		}
3287		tvd = vd->vdev_child[c];
3288		vdev_remove_child(vd, tvd);
3289		tvd->vdev_id = id;
3290		vdev_add_child(rvd, tvd);
3291		vdev_config_dirty(tvd);
3292	}
3293
3294	if (nspares != 0) {
3295		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3296		    ZPOOL_CONFIG_SPARES);
3297		spa_load_spares(spa);
3298		spa->spa_spares.sav_sync = B_TRUE;
3299	}
3300
3301	if (nl2cache != 0) {
3302		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3303		    ZPOOL_CONFIG_L2CACHE);
3304		spa_load_l2cache(spa);
3305		spa->spa_l2cache.sav_sync = B_TRUE;
3306	}
3307
3308	/*
3309	 * We have to be careful when adding new vdevs to an existing pool.
3310	 * If other threads start allocating from these vdevs before we
3311	 * sync the config cache, and we lose power, then upon reboot we may
3312	 * fail to open the pool because there are DVAs that the config cache
3313	 * can't translate.  Therefore, we first add the vdevs without
3314	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3315	 * and then let spa_config_update() initialize the new metaslabs.
3316	 *
3317	 * spa_load() checks for added-but-not-initialized vdevs, so that
3318	 * if we lose power at any point in this sequence, the remaining
3319	 * steps will be completed the next time we load the pool.
3320	 */
3321	(void) spa_vdev_exit(spa, vd, txg, 0);
3322
3323	mutex_enter(&spa_namespace_lock);
3324	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3325	mutex_exit(&spa_namespace_lock);
3326
3327	return (0);
3328}
3329
3330/*
3331 * Attach a device to a mirror.  The arguments are the path to any device
3332 * in the mirror, and the nvroot for the new device.  If the path specifies
3333 * a device that is not mirrored, we automatically insert the mirror vdev.
3334 *
3335 * If 'replacing' is specified, the new device is intended to replace the
3336 * existing device; in this case the two devices are made into their own
3337 * mirror using the 'replacing' vdev, which is functionally identical to
3338 * the mirror vdev (it actually reuses all the same ops) but has a few
3339 * extra rules: you can't attach to it after it's been created, and upon
3340 * completion of resilvering, the first disk (the one being replaced)
3341 * is automatically detached.
3342 */
3343int
3344spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3345{
3346	uint64_t txg, open_txg;
3347	vdev_t *rvd = spa->spa_root_vdev;
3348	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3349	vdev_ops_t *pvops;
3350	char *oldvdpath, *newvdpath;
3351	int newvd_isspare;
3352	int error;
3353
3354	txg = spa_vdev_enter(spa);
3355
3356	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3357
3358	if (oldvd == NULL)
3359		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3360
3361	if (!oldvd->vdev_ops->vdev_op_leaf)
3362		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3363
3364	pvd = oldvd->vdev_parent;
3365
3366	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3367	    VDEV_ALLOC_ADD)) != 0)
3368		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3369
3370	if (newrootvd->vdev_children != 1)
3371		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3372
3373	newvd = newrootvd->vdev_child[0];
3374
3375	if (!newvd->vdev_ops->vdev_op_leaf)
3376		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3377
3378	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3379		return (spa_vdev_exit(spa, newrootvd, txg, error));
3380
3381	/*
3382	 * Spares can't replace logs
3383	 */
3384	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3385		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3386
3387	if (!replacing) {
3388		/*
3389		 * For attach, the only allowable parent is a mirror or the root
3390		 * vdev.
3391		 */
3392		if (pvd->vdev_ops != &vdev_mirror_ops &&
3393		    pvd->vdev_ops != &vdev_root_ops)
3394			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3395
3396		pvops = &vdev_mirror_ops;
3397	} else {
3398		/*
3399		 * Active hot spares can only be replaced by inactive hot
3400		 * spares.
3401		 */
3402		if (pvd->vdev_ops == &vdev_spare_ops &&
3403		    pvd->vdev_child[1] == oldvd &&
3404		    !spa_has_spare(spa, newvd->vdev_guid))
3405			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3406
3407		/*
3408		 * If the source is a hot spare, and the parent isn't already a
3409		 * spare, then we want to create a new hot spare.  Otherwise, we
3410		 * want to create a replacing vdev.  The user is not allowed to
3411		 * attach to a spared vdev child unless the 'isspare' state is
3412		 * the same (spare replaces spare, non-spare replaces
3413		 * non-spare).
3414		 */
3415		if (pvd->vdev_ops == &vdev_replacing_ops)
3416			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3417		else if (pvd->vdev_ops == &vdev_spare_ops &&
3418		    newvd->vdev_isspare != oldvd->vdev_isspare)
3419			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3420		else if (pvd->vdev_ops != &vdev_spare_ops &&
3421		    newvd->vdev_isspare)
3422			pvops = &vdev_spare_ops;
3423		else
3424			pvops = &vdev_replacing_ops;
3425	}
3426
3427	/*
3428	 * Make sure the new device is big enough.
3429	 */
3430	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3431		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3432
3433	/*
3434	 * The new device cannot have a higher alignment requirement
3435	 * than the top-level vdev.
3436	 */
3437	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3438		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3439
3440	/*
3441	 * If this is an in-place replacement, update oldvd's path and devid
3442	 * to make it distinguishable from newvd, and unopenable from now on.
3443	 */
3444	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3445		spa_strfree(oldvd->vdev_path);
3446		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3447		    KM_SLEEP);
3448		(void) sprintf(oldvd->vdev_path, "%s/%s",
3449		    newvd->vdev_path, "old");
3450		if (oldvd->vdev_devid != NULL) {
3451			spa_strfree(oldvd->vdev_devid);
3452			oldvd->vdev_devid = NULL;
3453		}
3454	}
3455
3456	/*
3457	 * If the parent is not a mirror, or if we're replacing, insert the new
3458	 * mirror/replacing/spare vdev above oldvd.
3459	 */
3460	if (pvd->vdev_ops != pvops)
3461		pvd = vdev_add_parent(oldvd, pvops);
3462
3463	ASSERT(pvd->vdev_top->vdev_parent == rvd);
3464	ASSERT(pvd->vdev_ops == pvops);
3465	ASSERT(oldvd->vdev_parent == pvd);
3466
3467	/*
3468	 * Extract the new device from its root and add it to pvd.
3469	 */
3470	vdev_remove_child(newrootvd, newvd);
3471	newvd->vdev_id = pvd->vdev_children;
3472	newvd->vdev_crtxg = oldvd->vdev_crtxg;
3473	vdev_add_child(pvd, newvd);
3474
3475	tvd = newvd->vdev_top;
3476	ASSERT(pvd->vdev_top == tvd);
3477	ASSERT(tvd->vdev_parent == rvd);
3478
3479	vdev_config_dirty(tvd);
3480
3481	/*
3482	 * Set newvd's DTL to [TXG_INITIAL, open_txg].  It will propagate
3483	 * upward when spa_vdev_exit() calls vdev_dtl_reassess().
3484	 */
3485	open_txg = txg + TXG_CONCURRENT_STATES - 1;
3486
3487	vdev_dtl_dirty(newvd, DTL_MISSING,
3488	    TXG_INITIAL, open_txg - TXG_INITIAL + 1);
3489
3490	if (newvd->vdev_isspare) {
3491		spa_spare_activate(newvd);
3492		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
3493	}
3494
3495	oldvdpath = spa_strdup(oldvd->vdev_path);
3496	newvdpath = spa_strdup(newvd->vdev_path);
3497	newvd_isspare = newvd->vdev_isspare;
3498
3499	/*
3500	 * Mark newvd's DTL dirty in this txg.
3501	 */
3502	vdev_dirty(tvd, VDD_DTL, newvd, txg);
3503
3504	(void) spa_vdev_exit(spa, newrootvd, open_txg, 0);
3505
3506	spa_history_internal_log(LOG_POOL_VDEV_ATTACH, spa, NULL,
3507	    CRED(),  "%s vdev=%s %s vdev=%s",
3508	    replacing && newvd_isspare ? "spare in" :
3509	    replacing ? "replace" : "attach", newvdpath,
3510	    replacing ? "for" : "to", oldvdpath);
3511
3512	spa_strfree(oldvdpath);
3513	spa_strfree(newvdpath);
3514
3515	/*
3516	 * Kick off a resilver to update newvd.
3517	 */
3518	VERIFY3U(spa_scrub(spa, POOL_SCRUB_RESILVER), ==, 0);
3519
3520	return (0);
3521}
3522
3523/*
3524 * Detach a device from a mirror or replacing vdev.
3525 * If 'replace_done' is specified, only detach if the parent
3526 * is a replacing vdev.
3527 */
3528int
3529spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
3530{
3531	uint64_t txg;
3532	int error;
3533	vdev_t *rvd = spa->spa_root_vdev;
3534	vdev_t *vd, *pvd, *cvd, *tvd;
3535	boolean_t unspare = B_FALSE;
3536	uint64_t unspare_guid;
3537	size_t len;
3538
3539	txg = spa_vdev_enter(spa);
3540
3541	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3542
3543	if (vd == NULL)
3544		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3545
3546	if (!vd->vdev_ops->vdev_op_leaf)
3547		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3548
3549	pvd = vd->vdev_parent;
3550
3551	/*
3552	 * If the parent/child relationship is not as expected, don't do it.
3553	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
3554	 * vdev that's replacing B with C.  The user's intent in replacing
3555	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
3556	 * the replace by detaching C, the expected behavior is to end up
3557	 * M(A,B).  But suppose that right after deciding to detach C,
3558	 * the replacement of B completes.  We would have M(A,C), and then
3559	 * ask to detach C, which would leave us with just A -- not what
3560	 * the user wanted.  To prevent this, we make sure that the
3561	 * parent/child relationship hasn't changed -- in this example,
3562	 * that C's parent is still the replacing vdev R.
3563	 */
3564	if (pvd->vdev_guid != pguid && pguid != 0)
3565		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3566
3567	/*
3568	 * If replace_done is specified, only remove this device if it's
3569	 * the first child of a replacing vdev.  For the 'spare' vdev, either
3570	 * disk can be removed.
3571	 */
3572	if (replace_done) {
3573		if (pvd->vdev_ops == &vdev_replacing_ops) {
3574			if (vd->vdev_id != 0)
3575				return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3576		} else if (pvd->vdev_ops != &vdev_spare_ops) {
3577			return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3578		}
3579	}
3580
3581	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
3582	    spa_version(spa) >= SPA_VERSION_SPARES);
3583
3584	/*
3585	 * Only mirror, replacing, and spare vdevs support detach.
3586	 */
3587	if (pvd->vdev_ops != &vdev_replacing_ops &&
3588	    pvd->vdev_ops != &vdev_mirror_ops &&
3589	    pvd->vdev_ops != &vdev_spare_ops)
3590		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3591
3592	/*
3593	 * If this device has the only valid copy of some data,
3594	 * we cannot safely detach it.
3595	 */
3596	if (vdev_dtl_required(vd))
3597		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3598
3599	ASSERT(pvd->vdev_children >= 2);
3600
3601	/*
3602	 * If we are detaching the second disk from a replacing vdev, then
3603	 * check to see if we changed the original vdev's path to have "/old"
3604	 * at the end in spa_vdev_attach().  If so, undo that change now.
3605	 */
3606	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id == 1 &&
3607	    pvd->vdev_child[0]->vdev_path != NULL &&
3608	    pvd->vdev_child[1]->vdev_path != NULL) {
3609		ASSERT(pvd->vdev_child[1] == vd);
3610		cvd = pvd->vdev_child[0];
3611		len = strlen(vd->vdev_path);
3612		if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
3613		    strcmp(cvd->vdev_path + len, "/old") == 0) {
3614			spa_strfree(cvd->vdev_path);
3615			cvd->vdev_path = spa_strdup(vd->vdev_path);
3616		}
3617	}
3618
3619	/*
3620	 * If we are detaching the original disk from a spare, then it implies
3621	 * that the spare should become a real disk, and be removed from the
3622	 * active spare list for the pool.
3623	 */
3624	if (pvd->vdev_ops == &vdev_spare_ops &&
3625	    vd->vdev_id == 0 && pvd->vdev_child[1]->vdev_isspare)
3626		unspare = B_TRUE;
3627
3628	/*
3629	 * Erase the disk labels so the disk can be used for other things.
3630	 * This must be done after all other error cases are handled,
3631	 * but before we disembowel vd (so we can still do I/O to it).
3632	 * But if we can't do it, don't treat the error as fatal --
3633	 * it may be that the unwritability of the disk is the reason
3634	 * it's being detached!
3635	 */
3636	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
3637
3638	/*
3639	 * Remove vd from its parent and compact the parent's children.
3640	 */
3641	vdev_remove_child(pvd, vd);
3642	vdev_compact_children(pvd);
3643
3644	/*
3645	 * Remember one of the remaining children so we can get tvd below.
3646	 */
3647	cvd = pvd->vdev_child[0];
3648
3649	/*
3650	 * If we need to remove the remaining child from the list of hot spares,
3651	 * do it now, marking the vdev as no longer a spare in the process.
3652	 * We must do this before vdev_remove_parent(), because that can
3653	 * change the GUID if it creates a new toplevel GUID.  For a similar
3654	 * reason, we must remove the spare now, in the same txg as the detach;
3655	 * otherwise someone could attach a new sibling, change the GUID, and
3656	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
3657	 */
3658	if (unspare) {
3659		ASSERT(cvd->vdev_isspare);
3660		spa_spare_remove(cvd);
3661		unspare_guid = cvd->vdev_guid;
3662		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
3663	}
3664
3665	/*
3666	 * If the parent mirror/replacing vdev only has one child,
3667	 * the parent is no longer needed.  Remove it from the tree.
3668	 */
3669	if (pvd->vdev_children == 1)
3670		vdev_remove_parent(cvd);
3671
3672	/*
3673	 * We don't set tvd until now because the parent we just removed
3674	 * may have been the previous top-level vdev.
3675	 */
3676	tvd = cvd->vdev_top;
3677	ASSERT(tvd->vdev_parent == rvd);
3678
3679	/*
3680	 * Reevaluate the parent vdev state.
3681	 */
3682	vdev_propagate_state(cvd);
3683
3684	/*
3685	 * If the 'autoexpand' property is set on the pool then automatically
3686	 * try to expand the size of the pool. For example if the device we
3687	 * just detached was smaller than the others, it may be possible to
3688	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
3689	 * first so that we can obtain the updated sizes of the leaf vdevs.
3690	 */
3691	if (spa->spa_autoexpand) {
3692		vdev_reopen(tvd);
3693		vdev_expand(tvd, txg);
3694	}
3695
3696	vdev_config_dirty(tvd);
3697
3698	/*
3699	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
3700	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
3701	 * But first make sure we're not on any *other* txg's DTL list, to
3702	 * prevent vd from being accessed after it's freed.
3703	 */
3704	for (int t = 0; t < TXG_SIZE; t++)
3705		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
3706	vd->vdev_detached = B_TRUE;
3707	vdev_dirty(tvd, VDD_DTL, vd, txg);
3708
3709	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
3710
3711	error = spa_vdev_exit(spa, vd, txg, 0);
3712
3713	/*
3714	 * If this was the removal of the original device in a hot spare vdev,
3715	 * then we want to go through and remove the device from the hot spare
3716	 * list of every other pool.
3717	 */
3718	if (unspare) {
3719		spa_t *myspa = spa;
3720		spa = NULL;
3721		mutex_enter(&spa_namespace_lock);
3722		while ((spa = spa_next(spa)) != NULL) {
3723			if (spa->spa_state != POOL_STATE_ACTIVE)
3724				continue;
3725			if (spa == myspa)
3726				continue;
3727			spa_open_ref(spa, FTAG);
3728			mutex_exit(&spa_namespace_lock);
3729			(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
3730			mutex_enter(&spa_namespace_lock);
3731			spa_close(spa, FTAG);
3732		}
3733		mutex_exit(&spa_namespace_lock);
3734	}
3735
3736	return (error);
3737}
3738
3739static nvlist_t *
3740spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
3741{
3742	for (int i = 0; i < count; i++) {
3743		uint64_t guid;
3744
3745		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
3746		    &guid) == 0);
3747
3748		if (guid == target_guid)
3749			return (nvpp[i]);
3750	}
3751
3752	return (NULL);
3753}
3754
3755static void
3756spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
3757	nvlist_t *dev_to_remove)
3758{
3759	nvlist_t **newdev = NULL;
3760
3761	if (count > 1)
3762		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
3763
3764	for (int i = 0, j = 0; i < count; i++) {
3765		if (dev[i] == dev_to_remove)
3766			continue;
3767		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
3768	}
3769
3770	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
3771	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
3772
3773	for (int i = 0; i < count - 1; i++)
3774		nvlist_free(newdev[i]);
3775
3776	if (count > 1)
3777		kmem_free(newdev, (count - 1) * sizeof (void *));
3778}
3779
3780/*
3781 * Removing a device from the vdev namespace requires several steps
3782 * and can take a significant amount of time.  As a result we use
3783 * the spa_vdev_config_[enter/exit] functions which allow us to
3784 * grab and release the spa_config_lock while still holding the namespace
3785 * lock.  During each step the configuration is synced out.
3786 */
3787
3788/*
3789 * Initial phase of device removal - stop future allocations from this device.
3790 */
3791void
3792spa_vdev_remove_start(spa_t *spa, vdev_t *vd)
3793{
3794	metaslab_group_t *mg = vd->vdev_mg;
3795
3796	ASSERT(MUTEX_HELD(&spa_namespace_lock));
3797	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3798	ASSERT(vd == vd->vdev_top);
3799
3800	/*
3801	 * Remove our vdev from the allocatable vdevs
3802	 */
3803	if (mg)
3804		metaslab_class_remove(mg->mg_class, mg);
3805}
3806
3807/*
3808 * Evacuate the device.
3809 */
3810int
3811spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
3812{
3813	uint64_t txg;
3814	int error;
3815
3816	ASSERT(MUTEX_HELD(&spa_namespace_lock));
3817	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
3818	ASSERT(vd == vd->vdev_top);
3819
3820	/*
3821	 * Evacuate the device.  We don't hold the config lock as writer
3822	 * since we need to do I/O but we do keep the
3823	 * spa_namespace_lock held.  Once this completes the device
3824	 * should no longer have any blocks allocated on it.
3825	 */
3826	if (vd->vdev_islog) {
3827		/*
3828		 * Evacuate the device.
3829		 */
3830		if (error = dmu_objset_find(spa_name(spa),
3831		    zil_vdev_offline, NULL, DS_FIND_CHILDREN)) {
3832			uint64_t txg;
3833
3834			txg = spa_vdev_config_enter(spa);
3835			metaslab_class_add(spa->spa_log_class,
3836			    vd->vdev_mg);
3837			return (spa_vdev_exit(spa, NULL, txg, error));
3838		}
3839		txg_wait_synced(spa_get_dsl(spa), 0);
3840	}
3841
3842	/*
3843	 * Remove any remaining MOS metadata associated with the device.
3844	 */
3845	txg = spa_vdev_config_enter(spa);
3846	vd->vdev_removing = B_TRUE;
3847	vdev_dirty(vd, 0, NULL, txg);
3848	vdev_config_dirty(vd);
3849	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
3850
3851	return (0);
3852}
3853
3854/*
3855 * Complete the removal by cleaning up the namespace.
3856 */
3857void
3858spa_vdev_remove_done(spa_t *spa, vdev_t *vd)
3859{
3860	vdev_t *rvd = spa->spa_root_vdev;
3861	metaslab_group_t *mg = vd->vdev_mg;
3862	uint64_t id = vd->vdev_id;
3863	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
3864
3865	ASSERT(MUTEX_HELD(&spa_namespace_lock));
3866	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3867	ASSERT(vd == vd->vdev_top);
3868
3869	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
3870
3871	if (list_link_active(&vd->vdev_state_dirty_node))
3872		vdev_state_clean(vd);
3873	if (list_link_active(&vd->vdev_config_dirty_node))
3874		vdev_config_clean(vd);
3875
3876	vdev_free(vd);
3877
3878	/*
3879	 * It's possible that another thread is trying todo a spa_vdev_add()
3880	 * at the same time we're trying remove it. As a result the
3881	 * added vdev may not have initialized its metaslabs yet.
3882	 */
3883	if (mg != NULL)
3884		metaslab_group_destroy(mg);
3885
3886	if (last_vdev) {
3887		vdev_compact_children(rvd);
3888	} else {
3889		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
3890		vdev_add_child(rvd, vd);
3891	}
3892	vdev_config_dirty(rvd);
3893
3894	/*
3895	 * Reassess the health of our root vdev.
3896	 */
3897	vdev_reopen(rvd);
3898}
3899
3900/*
3901 * Remove a device from the pool.  Currently, this supports removing only hot
3902 * spares, slogs, and level 2 ARC devices.
3903 */
3904int
3905spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
3906{
3907	vdev_t *vd;
3908	nvlist_t **spares, **l2cache, *nv;
3909	uint64_t txg = 0;
3910	uint_t nspares, nl2cache;
3911	int error = 0;
3912	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
3913
3914	if (!locked)
3915		txg = spa_vdev_enter(spa);
3916
3917	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3918
3919	if (spa->spa_spares.sav_vdevs != NULL &&
3920	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3921	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
3922	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
3923		/*
3924		 * Only remove the hot spare if it's not currently in use
3925		 * in this pool.
3926		 */
3927		if (vd == NULL || unspare) {
3928			spa_vdev_remove_aux(spa->spa_spares.sav_config,
3929			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
3930			spa_load_spares(spa);
3931			spa->spa_spares.sav_sync = B_TRUE;
3932		} else {
3933			error = EBUSY;
3934		}
3935	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
3936	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3937	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
3938	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
3939		/*
3940		 * Cache devices can always be removed.
3941		 */
3942		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
3943		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
3944		spa_load_l2cache(spa);
3945		spa->spa_l2cache.sav_sync = B_TRUE;
3946	} else if (vd != NULL && vd->vdev_islog) {
3947		ASSERT(!locked);
3948		ASSERT(vd == vd->vdev_top);
3949
3950		/*
3951		 * XXX - Once we have bp-rewrite this should
3952		 * become the common case.
3953		 */
3954
3955		/*
3956		 * 1. Stop allocations
3957		 * 2. Evacuate the device (i.e. kill off stubby and
3958		 *    metadata) and wait for it to complete (i.e. sync).
3959		 * 3. Cleanup the vdev namespace.
3960		 */
3961		spa_vdev_remove_start(spa, vd);
3962
3963		/*
3964		 * Wait for the youngest allocations and frees to sync,
3965		 * and then wait for the deferral of those frees to finish.
3966		 */
3967		spa_vdev_config_exit(spa, NULL,
3968		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
3969
3970		if ((error = spa_vdev_remove_evacuate(spa, vd)) != 0)
3971			return (error);
3972		txg = spa_vdev_config_enter(spa);
3973
3974		spa_vdev_remove_done(spa, vd);
3975
3976	} else if (vd != NULL) {
3977		/*
3978		 * Normal vdevs cannot be removed (yet).
3979		 */
3980		error = ENOTSUP;
3981	} else {
3982		/*
3983		 * There is no vdev of any kind with the specified guid.
3984		 */
3985		error = ENOENT;
3986	}
3987
3988	if (!locked)
3989		return (spa_vdev_exit(spa, NULL, txg, error));
3990
3991	return (error);
3992}
3993
3994/*
3995 * Find any device that's done replacing, or a vdev marked 'unspare' that's
3996 * current spared, so we can detach it.
3997 */
3998static vdev_t *
3999spa_vdev_resilver_done_hunt(vdev_t *vd)
4000{
4001	vdev_t *newvd, *oldvd;
4002
4003	for (int c = 0; c < vd->vdev_children; c++) {
4004		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
4005		if (oldvd != NULL)
4006			return (oldvd);
4007	}
4008
4009	/*
4010	 * Check for a completed replacement.
4011	 */
4012	if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
4013		oldvd = vd->vdev_child[0];
4014		newvd = vd->vdev_child[1];
4015
4016		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4017		    !vdev_dtl_required(oldvd))
4018			return (oldvd);
4019	}
4020
4021	/*
4022	 * Check for a completed resilver with the 'unspare' flag set.
4023	 */
4024	if (vd->vdev_ops == &vdev_spare_ops && vd->vdev_children == 2) {
4025		newvd = vd->vdev_child[0];
4026		oldvd = vd->vdev_child[1];
4027
4028		if (newvd->vdev_unspare &&
4029		    vdev_dtl_empty(newvd, DTL_MISSING) &&
4030		    !vdev_dtl_required(oldvd)) {
4031			newvd->vdev_unspare = 0;
4032			return (oldvd);
4033		}
4034	}
4035
4036	return (NULL);
4037}
4038
4039static void
4040spa_vdev_resilver_done(spa_t *spa)
4041{
4042	vdev_t *vd, *pvd, *ppvd;
4043	uint64_t guid, sguid, pguid, ppguid;
4044
4045	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4046
4047	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4048		pvd = vd->vdev_parent;
4049		ppvd = pvd->vdev_parent;
4050		guid = vd->vdev_guid;
4051		pguid = pvd->vdev_guid;
4052		ppguid = ppvd->vdev_guid;
4053		sguid = 0;
4054		/*
4055		 * If we have just finished replacing a hot spared device, then
4056		 * we need to detach the parent's first child (the original hot
4057		 * spare) as well.
4058		 */
4059		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0) {
4060			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4061			ASSERT(ppvd->vdev_children == 2);
4062			sguid = ppvd->vdev_child[1]->vdev_guid;
4063		}
4064		spa_config_exit(spa, SCL_ALL, FTAG);
4065		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4066			return;
4067		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4068			return;
4069		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4070	}
4071
4072	spa_config_exit(spa, SCL_ALL, FTAG);
4073}
4074
4075/*
4076 * Update the stored path or FRU for this vdev.  Dirty the vdev configuration,
4077 * relying on spa_vdev_enter/exit() to synchronize the labels and cache.
4078 */
4079int
4080spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4081    boolean_t ispath)
4082{
4083	vdev_t *vd;
4084	uint64_t txg;
4085
4086	txg = spa_vdev_enter(spa);
4087
4088	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4089		return (spa_vdev_exit(spa, NULL, txg, ENOENT));
4090
4091	if (!vd->vdev_ops->vdev_op_leaf)
4092		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4093
4094	if (ispath) {
4095		spa_strfree(vd->vdev_path);
4096		vd->vdev_path = spa_strdup(value);
4097	} else {
4098		if (vd->vdev_fru != NULL)
4099			spa_strfree(vd->vdev_fru);
4100		vd->vdev_fru = spa_strdup(value);
4101	}
4102
4103	vdev_config_dirty(vd->vdev_top);
4104
4105	return (spa_vdev_exit(spa, NULL, txg, 0));
4106}
4107
4108int
4109spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
4110{
4111	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
4112}
4113
4114int
4115spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
4116{
4117	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
4118}
4119
4120/*
4121 * ==========================================================================
4122 * SPA Scrubbing
4123 * ==========================================================================
4124 */
4125
4126int
4127spa_scrub(spa_t *spa, pool_scrub_type_t type)
4128{
4129	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4130
4131	if ((uint_t)type >= POOL_SCRUB_TYPES)
4132		return (ENOTSUP);
4133
4134	/*
4135	 * If a resilver was requested, but there is no DTL on a
4136	 * writeable leaf device, we have nothing to do.
4137	 */
4138	if (type == POOL_SCRUB_RESILVER &&
4139	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
4140		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
4141		return (0);
4142	}
4143
4144	if (type == POOL_SCRUB_EVERYTHING &&
4145	    spa->spa_dsl_pool->dp_scrub_func != SCRUB_FUNC_NONE &&
4146	    spa->spa_dsl_pool->dp_scrub_isresilver)
4147		return (EBUSY);
4148
4149	if (type == POOL_SCRUB_EVERYTHING || type == POOL_SCRUB_RESILVER) {
4150		return (dsl_pool_scrub_clean(spa->spa_dsl_pool));
4151	} else if (type == POOL_SCRUB_NONE) {
4152		return (dsl_pool_scrub_cancel(spa->spa_dsl_pool));
4153	} else {
4154		return (EINVAL);
4155	}
4156}
4157
4158/*
4159 * ==========================================================================
4160 * SPA async task processing
4161 * ==========================================================================
4162 */
4163
4164static void
4165spa_async_remove(spa_t *spa, vdev_t *vd)
4166{
4167	if (vd->vdev_remove_wanted) {
4168		vd->vdev_remove_wanted = 0;
4169		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
4170
4171		/*
4172		 * We want to clear the stats, but we don't want to do a full
4173		 * vdev_clear() as that will cause us to throw away
4174		 * degraded/faulted state as well as attempt to reopen the
4175		 * device, all of which is a waste.
4176		 */
4177		vd->vdev_stat.vs_read_errors = 0;
4178		vd->vdev_stat.vs_write_errors = 0;
4179		vd->vdev_stat.vs_checksum_errors = 0;
4180
4181		vdev_state_dirty(vd->vdev_top);
4182	}
4183
4184	for (int c = 0; c < vd->vdev_children; c++)
4185		spa_async_remove(spa, vd->vdev_child[c]);
4186}
4187
4188static void
4189spa_async_probe(spa_t *spa, vdev_t *vd)
4190{
4191	if (vd->vdev_probe_wanted) {
4192		vd->vdev_probe_wanted = 0;
4193		vdev_reopen(vd);	/* vdev_open() does the actual probe */
4194	}
4195
4196	for (int c = 0; c < vd->vdev_children; c++)
4197		spa_async_probe(spa, vd->vdev_child[c]);
4198}
4199
4200static void
4201spa_async_autoexpand(spa_t *spa, vdev_t *vd)
4202{
4203	sysevent_id_t eid;
4204	nvlist_t *attr;
4205	char *physpath;
4206
4207	if (!spa->spa_autoexpand)
4208		return;
4209
4210	for (int c = 0; c < vd->vdev_children; c++) {
4211		vdev_t *cvd = vd->vdev_child[c];
4212		spa_async_autoexpand(spa, cvd);
4213	}
4214
4215	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
4216		return;
4217
4218	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
4219	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
4220
4221	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4222	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
4223
4224	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
4225	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
4226
4227	nvlist_free(attr);
4228	kmem_free(physpath, MAXPATHLEN);
4229}
4230
4231static void
4232spa_async_thread(spa_t *spa)
4233{
4234	int tasks;
4235
4236	ASSERT(spa->spa_sync_on);
4237
4238	mutex_enter(&spa->spa_async_lock);
4239	tasks = spa->spa_async_tasks;
4240	spa->spa_async_tasks = 0;
4241	mutex_exit(&spa->spa_async_lock);
4242
4243	/*
4244	 * See if the config needs to be updated.
4245	 */
4246	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
4247		uint64_t old_space, new_space;
4248
4249		mutex_enter(&spa_namespace_lock);
4250		old_space = metaslab_class_get_space(spa_normal_class(spa));
4251		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4252		new_space = metaslab_class_get_space(spa_normal_class(spa));
4253		mutex_exit(&spa_namespace_lock);
4254
4255		/*
4256		 * If the pool grew as a result of the config update,
4257		 * then log an internal history event.
4258		 */
4259		if (new_space != old_space) {
4260			spa_history_internal_log(LOG_POOL_VDEV_ONLINE,
4261			    spa, NULL, CRED(),
4262			    "pool '%s' size: %llu(+%llu)",
4263			    spa_name(spa), new_space, new_space - old_space);
4264		}
4265	}
4266
4267	/*
4268	 * See if any devices need to be marked REMOVED.
4269	 */
4270	if (tasks & SPA_ASYNC_REMOVE) {
4271		spa_vdev_state_enter(spa, SCL_NONE);
4272		spa_async_remove(spa, spa->spa_root_vdev);
4273		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
4274			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
4275		for (int i = 0; i < spa->spa_spares.sav_count; i++)
4276			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
4277		(void) spa_vdev_state_exit(spa, NULL, 0);
4278	}
4279
4280	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
4281		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4282		spa_async_autoexpand(spa, spa->spa_root_vdev);
4283		spa_config_exit(spa, SCL_CONFIG, FTAG);
4284	}
4285
4286	/*
4287	 * See if any devices need to be probed.
4288	 */
4289	if (tasks & SPA_ASYNC_PROBE) {
4290		spa_vdev_state_enter(spa, SCL_NONE);
4291		spa_async_probe(spa, spa->spa_root_vdev);
4292		(void) spa_vdev_state_exit(spa, NULL, 0);
4293	}
4294
4295	/*
4296	 * If any devices are done replacing, detach them.
4297	 */
4298	if (tasks & SPA_ASYNC_RESILVER_DONE)
4299		spa_vdev_resilver_done(spa);
4300
4301	/*
4302	 * Kick off a resilver.
4303	 */
4304	if (tasks & SPA_ASYNC_RESILVER)
4305		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER) == 0);
4306
4307	/*
4308	 * Let the world know that we're done.
4309	 */
4310	mutex_enter(&spa->spa_async_lock);
4311	spa->spa_async_thread = NULL;
4312	cv_broadcast(&spa->spa_async_cv);
4313	mutex_exit(&spa->spa_async_lock);
4314	thread_exit();
4315}
4316
4317void
4318spa_async_suspend(spa_t *spa)
4319{
4320	mutex_enter(&spa->spa_async_lock);
4321	spa->spa_async_suspended++;
4322	while (spa->spa_async_thread != NULL)
4323		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
4324	mutex_exit(&spa->spa_async_lock);
4325}
4326
4327void
4328spa_async_resume(spa_t *spa)
4329{
4330	mutex_enter(&spa->spa_async_lock);
4331	ASSERT(spa->spa_async_suspended != 0);
4332	spa->spa_async_suspended--;
4333	mutex_exit(&spa->spa_async_lock);
4334}
4335
4336static void
4337spa_async_dispatch(spa_t *spa)
4338{
4339	mutex_enter(&spa->spa_async_lock);
4340	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
4341	    spa->spa_async_thread == NULL &&
4342	    rootdir != NULL && !vn_is_readonly(rootdir))
4343		spa->spa_async_thread = thread_create(NULL, 0,
4344		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
4345	mutex_exit(&spa->spa_async_lock);
4346}
4347
4348void
4349spa_async_request(spa_t *spa, int task)
4350{
4351	mutex_enter(&spa->spa_async_lock);
4352	spa->spa_async_tasks |= task;
4353	mutex_exit(&spa->spa_async_lock);
4354}
4355
4356/*
4357 * ==========================================================================
4358 * SPA syncing routines
4359 * ==========================================================================
4360 */
4361static void
4362spa_sync_deferred_bplist(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx, uint64_t txg)
4363{
4364	blkptr_t blk;
4365	uint64_t itor = 0;
4366	uint8_t c = 1;
4367
4368	while (bplist_iterate(bpl, &itor, &blk) == 0) {
4369		ASSERT(blk.blk_birth < txg);
4370		zio_free(spa, txg, &blk);
4371	}
4372
4373	bplist_vacate(bpl, tx);
4374
4375	/*
4376	 * Pre-dirty the first block so we sync to convergence faster.
4377	 * (Usually only the first block is needed.)
4378	 */
4379	dmu_write(bpl->bpl_mos, spa->spa_deferred_bplist_obj, 0, 1, &c, tx);
4380}
4381
4382static void
4383spa_sync_free(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
4384{
4385	zio_t *zio = arg;
4386
4387	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
4388	    zio->io_flags));
4389}
4390
4391static void
4392spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
4393{
4394	char *packed = NULL;
4395	size_t bufsize;
4396	size_t nvsize = 0;
4397	dmu_buf_t *db;
4398
4399	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
4400
4401	/*
4402	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
4403	 * information.  This avoids the dbuf_will_dirty() path and
4404	 * saves us a pre-read to get data we don't actually care about.
4405	 */
4406	bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
4407	packed = kmem_alloc(bufsize, KM_SLEEP);
4408
4409	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
4410	    KM_SLEEP) == 0);
4411	bzero(packed + nvsize, bufsize - nvsize);
4412
4413	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
4414
4415	kmem_free(packed, bufsize);
4416
4417	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
4418	dmu_buf_will_dirty(db, tx);
4419	*(uint64_t *)db->db_data = nvsize;
4420	dmu_buf_rele(db, FTAG);
4421}
4422
4423static void
4424spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
4425    const char *config, const char *entry)
4426{
4427	nvlist_t *nvroot;
4428	nvlist_t **list;
4429	int i;
4430
4431	if (!sav->sav_sync)
4432		return;
4433
4434	/*
4435	 * Update the MOS nvlist describing the list of available devices.
4436	 * spa_validate_aux() will have already made sure this nvlist is
4437	 * valid and the vdevs are labeled appropriately.
4438	 */
4439	if (sav->sav_object == 0) {
4440		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
4441		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
4442		    sizeof (uint64_t), tx);
4443		VERIFY(zap_update(spa->spa_meta_objset,
4444		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
4445		    &sav->sav_object, tx) == 0);
4446	}
4447
4448	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4449	if (sav->sav_count == 0) {
4450		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
4451	} else {
4452		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
4453		for (i = 0; i < sav->sav_count; i++)
4454			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
4455			    B_FALSE, B_FALSE, B_TRUE);
4456		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
4457		    sav->sav_count) == 0);
4458		for (i = 0; i < sav->sav_count; i++)
4459			nvlist_free(list[i]);
4460		kmem_free(list, sav->sav_count * sizeof (void *));
4461	}
4462
4463	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
4464	nvlist_free(nvroot);
4465
4466	sav->sav_sync = B_FALSE;
4467}
4468
4469static void
4470spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
4471{
4472	nvlist_t *config;
4473
4474	if (list_is_empty(&spa->spa_config_dirty_list))
4475		return;
4476
4477	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4478
4479	config = spa_config_generate(spa, spa->spa_root_vdev,
4480	    dmu_tx_get_txg(tx), B_FALSE);
4481
4482	spa_config_exit(spa, SCL_STATE, FTAG);
4483
4484	if (spa->spa_config_syncing)
4485		nvlist_free(spa->spa_config_syncing);
4486	spa->spa_config_syncing = config;
4487
4488	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
4489}
4490
4491/*
4492 * Set zpool properties.
4493 */
4494static void
4495spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
4496{
4497	spa_t *spa = arg1;
4498	objset_t *mos = spa->spa_meta_objset;
4499	nvlist_t *nvp = arg2;
4500	nvpair_t *elem;
4501	uint64_t intval;
4502	char *strval;
4503	zpool_prop_t prop;
4504	const char *propname;
4505	zprop_type_t proptype;
4506
4507	mutex_enter(&spa->spa_props_lock);
4508
4509	elem = NULL;
4510	while ((elem = nvlist_next_nvpair(nvp, elem))) {
4511		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
4512		case ZPOOL_PROP_VERSION:
4513			/*
4514			 * Only set version for non-zpool-creation cases
4515			 * (set/import). spa_create() needs special care
4516			 * for version setting.
4517			 */
4518			if (tx->tx_txg != TXG_INITIAL) {
4519				VERIFY(nvpair_value_uint64(elem,
4520				    &intval) == 0);
4521				ASSERT(intval <= SPA_VERSION);
4522				ASSERT(intval >= spa_version(spa));
4523				spa->spa_uberblock.ub_version = intval;
4524				vdev_config_dirty(spa->spa_root_vdev);
4525			}
4526			break;
4527
4528		case ZPOOL_PROP_ALTROOT:
4529			/*
4530			 * 'altroot' is a non-persistent property. It should
4531			 * have been set temporarily at creation or import time.
4532			 */
4533			ASSERT(spa->spa_root != NULL);
4534			break;
4535
4536		case ZPOOL_PROP_CACHEFILE:
4537			/*
4538			 * 'cachefile' is also a non-persisitent property.
4539			 */
4540			break;
4541		default:
4542			/*
4543			 * Set pool property values in the poolprops mos object.
4544			 */
4545			if (spa->spa_pool_props_object == 0) {
4546				VERIFY((spa->spa_pool_props_object =
4547				    zap_create(mos, DMU_OT_POOL_PROPS,
4548				    DMU_OT_NONE, 0, tx)) > 0);
4549
4550				VERIFY(zap_update(mos,
4551				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
4552				    8, 1, &spa->spa_pool_props_object, tx)
4553				    == 0);
4554			}
4555
4556			/* normalize the property name */
4557			propname = zpool_prop_to_name(prop);
4558			proptype = zpool_prop_get_type(prop);
4559
4560			if (nvpair_type(elem) == DATA_TYPE_STRING) {
4561				ASSERT(proptype == PROP_TYPE_STRING);
4562				VERIFY(nvpair_value_string(elem, &strval) == 0);
4563				VERIFY(zap_update(mos,
4564				    spa->spa_pool_props_object, propname,
4565				    1, strlen(strval) + 1, strval, tx) == 0);
4566
4567			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
4568				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
4569
4570				if (proptype == PROP_TYPE_INDEX) {
4571					const char *unused;
4572					VERIFY(zpool_prop_index_to_string(
4573					    prop, intval, &unused) == 0);
4574				}
4575				VERIFY(zap_update(mos,
4576				    spa->spa_pool_props_object, propname,
4577				    8, 1, &intval, tx) == 0);
4578			} else {
4579				ASSERT(0); /* not allowed */
4580			}
4581
4582			switch (prop) {
4583			case ZPOOL_PROP_DELEGATION:
4584				spa->spa_delegation = intval;
4585				break;
4586			case ZPOOL_PROP_BOOTFS:
4587				spa->spa_bootfs = intval;
4588				break;
4589			case ZPOOL_PROP_FAILUREMODE:
4590				spa->spa_failmode = intval;
4591				break;
4592			case ZPOOL_PROP_AUTOEXPAND:
4593				spa->spa_autoexpand = intval;
4594				spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4595				break;
4596			case ZPOOL_PROP_DEDUPDITTO:
4597				spa->spa_dedup_ditto = intval;
4598				break;
4599			default:
4600				break;
4601			}
4602		}
4603
4604		/* log internal history if this is not a zpool create */
4605		if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
4606		    tx->tx_txg != TXG_INITIAL) {
4607			spa_history_internal_log(LOG_POOL_PROPSET,
4608			    spa, tx, cr, "%s %lld %s",
4609			    nvpair_name(elem), intval, spa_name(spa));
4610		}
4611	}
4612
4613	mutex_exit(&spa->spa_props_lock);
4614}
4615
4616/*
4617 * Sync the specified transaction group.  New blocks may be dirtied as
4618 * part of the process, so we iterate until it converges.
4619 */
4620void
4621spa_sync(spa_t *spa, uint64_t txg)
4622{
4623	dsl_pool_t *dp = spa->spa_dsl_pool;
4624	objset_t *mos = spa->spa_meta_objset;
4625	bplist_t *defer_bpl = &spa->spa_deferred_bplist;
4626	bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
4627	vdev_t *rvd = spa->spa_root_vdev;
4628	vdev_t *vd;
4629	dmu_tx_t *tx;
4630	int error;
4631
4632	/*
4633	 * Lock out configuration changes.
4634	 */
4635	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4636
4637	spa->spa_syncing_txg = txg;
4638	spa->spa_sync_pass = 0;
4639
4640	/*
4641	 * If there are any pending vdev state changes, convert them
4642	 * into config changes that go out with this transaction group.
4643	 */
4644	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4645	while (list_head(&spa->spa_state_dirty_list) != NULL) {
4646		/*
4647		 * We need the write lock here because, for aux vdevs,
4648		 * calling vdev_config_dirty() modifies sav_config.
4649		 * This is ugly and will become unnecessary when we
4650		 * eliminate the aux vdev wart by integrating all vdevs
4651		 * into the root vdev tree.
4652		 */
4653		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
4654		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
4655		while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
4656			vdev_state_clean(vd);
4657			vdev_config_dirty(vd);
4658		}
4659		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
4660		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
4661	}
4662	spa_config_exit(spa, SCL_STATE, FTAG);
4663
4664	VERIFY(0 == bplist_open(defer_bpl, mos, spa->spa_deferred_bplist_obj));
4665
4666	tx = dmu_tx_create_assigned(dp, txg);
4667
4668	/*
4669	 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
4670	 * set spa_deflate if we have no raid-z vdevs.
4671	 */
4672	if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
4673	    spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
4674		int i;
4675
4676		for (i = 0; i < rvd->vdev_children; i++) {
4677			vd = rvd->vdev_child[i];
4678			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
4679				break;
4680		}
4681		if (i == rvd->vdev_children) {
4682			spa->spa_deflate = TRUE;
4683			VERIFY(0 == zap_add(spa->spa_meta_objset,
4684			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
4685			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
4686		}
4687	}
4688
4689	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
4690	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
4691		dsl_pool_create_origin(dp, tx);
4692
4693		/* Keeping the origin open increases spa_minref */
4694		spa->spa_minref += 3;
4695	}
4696
4697	if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
4698	    spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
4699		dsl_pool_upgrade_clones(dp, tx);
4700	}
4701
4702	/*
4703	 * If anything has changed in this txg, push the deferred frees
4704	 * from the previous txg.  If not, leave them alone so that we
4705	 * don't generate work on an otherwise idle system.
4706	 */
4707	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
4708	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
4709	    !txg_list_empty(&dp->dp_sync_tasks, txg))
4710		spa_sync_deferred_bplist(spa, defer_bpl, tx, txg);
4711
4712	/*
4713	 * Iterate to convergence.
4714	 */
4715	do {
4716		int pass = ++spa->spa_sync_pass;
4717
4718		spa_sync_config_object(spa, tx);
4719		spa_sync_aux_dev(spa, &spa->spa_spares, tx,
4720		    ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
4721		spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
4722		    ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
4723		spa_errlog_sync(spa, txg);
4724		dsl_pool_sync(dp, txg);
4725
4726		if (pass <= SYNC_PASS_DEFERRED_FREE) {
4727			zio_t *zio = zio_root(spa, NULL, NULL, 0);
4728			bplist_sync(free_bpl, spa_sync_free, zio, tx);
4729			VERIFY(zio_wait(zio) == 0);
4730		} else {
4731			bplist_sync(free_bpl, bplist_enqueue_cb, defer_bpl, tx);
4732		}
4733
4734		ddt_sync(spa, txg);
4735
4736		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
4737			vdev_sync(vd, txg);
4738
4739	} while (dmu_objset_is_dirty(mos, txg));
4740
4741	ASSERT(free_bpl->bpl_queue == NULL);
4742
4743	bplist_close(defer_bpl);
4744
4745	/*
4746	 * Rewrite the vdev configuration (which includes the uberblock)
4747	 * to commit the transaction group.
4748	 *
4749	 * If there are no dirty vdevs, we sync the uberblock to a few
4750	 * random top-level vdevs that are known to be visible in the
4751	 * config cache (see spa_vdev_add() for a complete description).
4752	 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
4753	 */
4754	for (;;) {
4755		/*
4756		 * We hold SCL_STATE to prevent vdev open/close/etc.
4757		 * while we're attempting to write the vdev labels.
4758		 */
4759		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4760
4761		if (list_is_empty(&spa->spa_config_dirty_list)) {
4762			vdev_t *svd[SPA_DVAS_PER_BP];
4763			int svdcount = 0;
4764			int children = rvd->vdev_children;
4765			int c0 = spa_get_random(children);
4766
4767			for (int c = 0; c < children; c++) {
4768				vd = rvd->vdev_child[(c0 + c) % children];
4769				if (vd->vdev_ms_array == 0 || vd->vdev_islog)
4770					continue;
4771				svd[svdcount++] = vd;
4772				if (svdcount == SPA_DVAS_PER_BP)
4773					break;
4774			}
4775			error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
4776			if (error != 0)
4777				error = vdev_config_sync(svd, svdcount, txg,
4778				    B_TRUE);
4779		} else {
4780			error = vdev_config_sync(rvd->vdev_child,
4781			    rvd->vdev_children, txg, B_FALSE);
4782			if (error != 0)
4783				error = vdev_config_sync(rvd->vdev_child,
4784				    rvd->vdev_children, txg, B_TRUE);
4785		}
4786
4787		spa_config_exit(spa, SCL_STATE, FTAG);
4788
4789		if (error == 0)
4790			break;
4791		zio_suspend(spa, NULL);
4792		zio_resume_wait(spa);
4793	}
4794	dmu_tx_commit(tx);
4795
4796	/*
4797	 * Clear the dirty config list.
4798	 */
4799	while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
4800		vdev_config_clean(vd);
4801
4802	/*
4803	 * Now that the new config has synced transactionally,
4804	 * let it become visible to the config cache.
4805	 */
4806	if (spa->spa_config_syncing != NULL) {
4807		spa_config_set(spa, spa->spa_config_syncing);
4808		spa->spa_config_txg = txg;
4809		spa->spa_config_syncing = NULL;
4810	}
4811
4812	spa->spa_ubsync = spa->spa_uberblock;
4813
4814	dsl_pool_sync_done(dp, txg);
4815
4816	/*
4817	 * Update usable space statistics.
4818	 */
4819	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
4820		vdev_sync_done(vd, txg);
4821
4822	/*
4823	 * It had better be the case that we didn't dirty anything
4824	 * since vdev_config_sync().
4825	 */
4826	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
4827	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
4828	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
4829	ASSERT(defer_bpl->bpl_queue == NULL);
4830	ASSERT(free_bpl->bpl_queue == NULL);
4831
4832	spa->spa_sync_pass = 0;
4833
4834	spa_config_exit(spa, SCL_CONFIG, FTAG);
4835
4836	spa_handle_ignored_writes(spa);
4837
4838	/*
4839	 * If any async tasks have been requested, kick them off.
4840	 */
4841	spa_async_dispatch(spa);
4842}
4843
4844/*
4845 * Sync all pools.  We don't want to hold the namespace lock across these
4846 * operations, so we take a reference on the spa_t and drop the lock during the
4847 * sync.
4848 */
4849void
4850spa_sync_allpools(void)
4851{
4852	spa_t *spa = NULL;
4853	mutex_enter(&spa_namespace_lock);
4854	while ((spa = spa_next(spa)) != NULL) {
4855		if (spa_state(spa) != POOL_STATE_ACTIVE || spa_suspended(spa))
4856			continue;
4857		spa_open_ref(spa, FTAG);
4858		mutex_exit(&spa_namespace_lock);
4859		txg_wait_synced(spa_get_dsl(spa), 0);
4860		mutex_enter(&spa_namespace_lock);
4861		spa_close(spa, FTAG);
4862	}
4863	mutex_exit(&spa_namespace_lock);
4864}
4865
4866/*
4867 * ==========================================================================
4868 * Miscellaneous routines
4869 * ==========================================================================
4870 */
4871
4872/*
4873 * Remove all pools in the system.
4874 */
4875void
4876spa_evict_all(void)
4877{
4878	spa_t *spa;
4879
4880	/*
4881	 * Remove all cached state.  All pools should be closed now,
4882	 * so every spa in the AVL tree should be unreferenced.
4883	 */
4884	mutex_enter(&spa_namespace_lock);
4885	while ((spa = spa_next(NULL)) != NULL) {
4886		/*
4887		 * Stop async tasks.  The async thread may need to detach
4888		 * a device that's been replaced, which requires grabbing
4889		 * spa_namespace_lock, so we must drop it here.
4890		 */
4891		spa_open_ref(spa, FTAG);
4892		mutex_exit(&spa_namespace_lock);
4893		spa_async_suspend(spa);
4894		mutex_enter(&spa_namespace_lock);
4895		spa_close(spa, FTAG);
4896
4897		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4898			spa_unload(spa);
4899			spa_deactivate(spa);
4900		}
4901		spa_remove(spa);
4902	}
4903	mutex_exit(&spa_namespace_lock);
4904}
4905
4906vdev_t *
4907spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
4908{
4909	vdev_t *vd;
4910	int i;
4911
4912	if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
4913		return (vd);
4914
4915	if (aux) {
4916		for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
4917			vd = spa->spa_l2cache.sav_vdevs[i];
4918			if (vd->vdev_guid == guid)
4919				return (vd);
4920		}
4921
4922		for (i = 0; i < spa->spa_spares.sav_count; i++) {
4923			vd = spa->spa_spares.sav_vdevs[i];
4924			if (vd->vdev_guid == guid)
4925				return (vd);
4926		}
4927	}
4928
4929	return (NULL);
4930}
4931
4932void
4933spa_upgrade(spa_t *spa, uint64_t version)
4934{
4935	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4936
4937	/*
4938	 * This should only be called for a non-faulted pool, and since a
4939	 * future version would result in an unopenable pool, this shouldn't be
4940	 * possible.
4941	 */
4942	ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
4943	ASSERT(version >= spa->spa_uberblock.ub_version);
4944
4945	spa->spa_uberblock.ub_version = version;
4946	vdev_config_dirty(spa->spa_root_vdev);
4947
4948	spa_config_exit(spa, SCL_ALL, FTAG);
4949
4950	txg_wait_synced(spa_get_dsl(spa), 0);
4951}
4952
4953boolean_t
4954spa_has_spare(spa_t *spa, uint64_t guid)
4955{
4956	int i;
4957	uint64_t spareguid;
4958	spa_aux_vdev_t *sav = &spa->spa_spares;
4959
4960	for (i = 0; i < sav->sav_count; i++)
4961		if (sav->sav_vdevs[i]->vdev_guid == guid)
4962			return (B_TRUE);
4963
4964	for (i = 0; i < sav->sav_npending; i++) {
4965		if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
4966		    &spareguid) == 0 && spareguid == guid)
4967			return (B_TRUE);
4968	}
4969
4970	return (B_FALSE);
4971}
4972
4973/*
4974 * Check if a pool has an active shared spare device.
4975 * Note: reference count of an active spare is 2, as a spare and as a replace
4976 */
4977static boolean_t
4978spa_has_active_shared_spare(spa_t *spa)
4979{
4980	int i, refcnt;
4981	uint64_t pool;
4982	spa_aux_vdev_t *sav = &spa->spa_spares;
4983
4984	for (i = 0; i < sav->sav_count; i++) {
4985		if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
4986		    &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
4987		    refcnt > 2)
4988			return (B_TRUE);
4989	}
4990
4991	return (B_FALSE);
4992}
4993
4994/*
4995 * Post a sysevent corresponding to the given event.  The 'name' must be one of
4996 * the event definitions in sys/sysevent/eventdefs.h.  The payload will be
4997 * filled in from the spa and (optionally) the vdev.  This doesn't do anything
4998 * in the userland libzpool, as we don't want consumers to misinterpret ztest
4999 * or zdb as real changes.
5000 */
5001void
5002spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
5003{
5004#ifdef _KERNEL
5005	sysevent_t		*ev;
5006	sysevent_attr_list_t	*attr = NULL;
5007	sysevent_value_t	value;
5008	sysevent_id_t		eid;
5009
5010	ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
5011	    SE_SLEEP);
5012
5013	value.value_type = SE_DATA_TYPE_STRING;
5014	value.value.sv_string = spa_name(spa);
5015	if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
5016		goto done;
5017
5018	value.value_type = SE_DATA_TYPE_UINT64;
5019	value.value.sv_uint64 = spa_guid(spa);
5020	if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
5021		goto done;
5022
5023	if (vd) {
5024		value.value_type = SE_DATA_TYPE_UINT64;
5025		value.value.sv_uint64 = vd->vdev_guid;
5026		if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
5027		    SE_SLEEP) != 0)
5028			goto done;
5029
5030		if (vd->vdev_path) {
5031			value.value_type = SE_DATA_TYPE_STRING;
5032			value.value.sv_string = vd->vdev_path;
5033			if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
5034			    &value, SE_SLEEP) != 0)
5035				goto done;
5036		}
5037	}
5038
5039	if (sysevent_attach_attributes(ev, attr) != 0)
5040		goto done;
5041	attr = NULL;
5042
5043	(void) log_sysevent(ev, SE_SLEEP, &eid);
5044
5045done:
5046	if (attr)
5047		sysevent_free_attr(attr);
5048	sysevent_free(ev);
5049#endif
5050}
5051