spa.c revision a15215608b8bd90f714f6db21ee623b584607cb6
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 alloc;
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		alloc = 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_ALLOCATED, NULL, alloc, src);
162		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
163		    size - alloc, src);
164
165		cap = (size == 0) ? 0 : (alloc * 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	spa_update_dspace(spa);
1687
1688	if (state != SPA_LOAD_TRYIMPORT) {
1689		error = spa_load_verify(spa);
1690		if (error) {
1691			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1692			    VDEV_AUX_CORRUPT_DATA);
1693			goto out;
1694		}
1695	}
1696
1697	/*
1698	 * Load the intent log state and check log integrity.
1699	 */
1700	VERIFY(nvlist_lookup_nvlist(nvconfig, ZPOOL_CONFIG_VDEV_TREE,
1701	    &nvroot) == 0);
1702	spa_load_log_state(spa, nvroot);
1703	nvlist_free(nvconfig);
1704
1705	if (spa_check_logs(spa)) {
1706		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1707		    VDEV_AUX_BAD_LOG);
1708		error = ENXIO;
1709		ereport = FM_EREPORT_ZFS_LOG_REPLAY;
1710		goto out;
1711	}
1712
1713	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
1714	    spa->spa_load_max_txg == UINT64_MAX)) {
1715		dmu_tx_t *tx;
1716		int need_update = B_FALSE;
1717
1718		ASSERT(state != SPA_LOAD_TRYIMPORT);
1719
1720		/*
1721		 * Claim log blocks that haven't been committed yet.
1722		 * This must all happen in a single txg.
1723		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
1724		 * invoked from zil_claim_log_block()'s i/o done callback.
1725		 * Price of rollback is that we abandon the log.
1726		 */
1727		spa->spa_claiming = B_TRUE;
1728
1729		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
1730		    spa_first_txg(spa));
1731		(void) dmu_objset_find(spa_name(spa),
1732		    zil_claim, tx, DS_FIND_CHILDREN);
1733		dmu_tx_commit(tx);
1734
1735		spa->spa_claiming = B_FALSE;
1736
1737		spa->spa_log_state = SPA_LOG_GOOD;
1738		spa->spa_sync_on = B_TRUE;
1739		txg_sync_start(spa->spa_dsl_pool);
1740
1741		/*
1742		 * Wait for all claims to sync.  We sync up to the highest
1743		 * claimed log block birth time so that claimed log blocks
1744		 * don't appear to be from the future.  spa_claim_max_txg
1745		 * will have been set for us by either zil_check_log_chain()
1746		 * (invoked from spa_check_logs()) or zil_claim() above.
1747		 */
1748		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
1749
1750		/*
1751		 * If the config cache is stale, or we have uninitialized
1752		 * metaslabs (see spa_vdev_add()), then update the config.
1753		 *
1754		 * If spa_load_verbatim is true, trust the current
1755		 * in-core spa_config and update the disk labels.
1756		 */
1757		if (config_cache_txg != spa->spa_config_txg ||
1758		    state == SPA_LOAD_IMPORT || spa->spa_load_verbatim ||
1759		    state == SPA_LOAD_RECOVER)
1760			need_update = B_TRUE;
1761
1762		for (int c = 0; c < rvd->vdev_children; c++)
1763			if (rvd->vdev_child[c]->vdev_ms_array == 0)
1764				need_update = B_TRUE;
1765
1766		/*
1767		 * Update the config cache asychronously in case we're the
1768		 * root pool, in which case the config cache isn't writable yet.
1769		 */
1770		if (need_update)
1771			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
1772
1773		/*
1774		 * Check all DTLs to see if anything needs resilvering.
1775		 */
1776		if (vdev_resilver_needed(rvd, NULL, NULL))
1777			spa_async_request(spa, SPA_ASYNC_RESILVER);
1778
1779		/*
1780		 * Delete any inconsistent datasets.
1781		 */
1782		(void) dmu_objset_find(spa_name(spa),
1783		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
1784
1785		/*
1786		 * Clean up any stale temporary dataset userrefs.
1787		 */
1788		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
1789	}
1790
1791	error = 0;
1792out:
1793
1794	spa->spa_minref = refcount_count(&spa->spa_refcount);
1795	if (error && error != EBADF)
1796		zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1797	spa->spa_load_state = SPA_LOAD_NONE;
1798	spa->spa_ena = 0;
1799
1800	return (error);
1801}
1802
1803static int
1804spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
1805{
1806	spa_unload(spa);
1807	spa_deactivate(spa);
1808
1809	spa->spa_load_max_txg--;
1810
1811	spa_activate(spa, spa_mode_global);
1812	spa_async_suspend(spa);
1813
1814	return (spa_load(spa, state, mosconfig));
1815}
1816
1817static int
1818spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
1819    uint64_t max_request, boolean_t extreme)
1820{
1821	nvlist_t *config = NULL;
1822	int load_error, rewind_error;
1823	uint64_t safe_rollback_txg;
1824	uint64_t min_txg;
1825
1826	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER)
1827		spa->spa_load_max_txg = spa->spa_load_txg;
1828	else
1829		spa->spa_load_max_txg = max_request;
1830
1831	load_error = rewind_error = spa_load(spa, state, mosconfig);
1832	if (load_error == 0)
1833		return (0);
1834
1835	if (spa->spa_root_vdev != NULL)
1836		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
1837
1838	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
1839	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
1840
1841	/* specific txg requested */
1842	if (spa->spa_load_max_txg != UINT64_MAX && !extreme) {
1843		nvlist_free(config);
1844		return (load_error);
1845	}
1846
1847	/* Price of rolling back is discarding txgs, including log */
1848	if (state == SPA_LOAD_RECOVER)
1849		spa->spa_log_state = SPA_LOG_CLEAR;
1850
1851	spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1852	safe_rollback_txg = spa->spa_uberblock.ub_txg - TXG_DEFER_SIZE;
1853
1854	min_txg = extreme ? TXG_INITIAL : safe_rollback_txg;
1855	while (rewind_error && (spa->spa_uberblock.ub_txg >= min_txg)) {
1856		if (spa->spa_load_max_txg < safe_rollback_txg)
1857			spa->spa_extreme_rewind = B_TRUE;
1858		rewind_error = spa_load_retry(spa, state, mosconfig);
1859	}
1860
1861	if (config)
1862		spa_rewind_data_to_nvlist(spa, config);
1863
1864	spa->spa_extreme_rewind = B_FALSE;
1865	spa->spa_load_max_txg = UINT64_MAX;
1866
1867	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
1868		spa_config_set(spa, config);
1869
1870	return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
1871}
1872
1873/*
1874 * Pool Open/Import
1875 *
1876 * The import case is identical to an open except that the configuration is sent
1877 * down from userland, instead of grabbed from the configuration cache.  For the
1878 * case of an open, the pool configuration will exist in the
1879 * POOL_STATE_UNINITIALIZED state.
1880 *
1881 * The stats information (gen/count/ustats) is used to gather vdev statistics at
1882 * the same time open the pool, without having to keep around the spa_t in some
1883 * ambiguous state.
1884 */
1885static int
1886spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
1887    nvlist_t **config)
1888{
1889	spa_t *spa;
1890	boolean_t norewind;
1891	boolean_t extreme;
1892	zpool_rewind_policy_t policy;
1893	spa_load_state_t state = SPA_LOAD_OPEN;
1894	int error;
1895	int locked = B_FALSE;
1896
1897	*spapp = NULL;
1898
1899	zpool_get_rewind_policy(nvpolicy, &policy);
1900	if (policy.zrp_request & ZPOOL_DO_REWIND)
1901		state = SPA_LOAD_RECOVER;
1902	norewind = (policy.zrp_request == ZPOOL_NO_REWIND);
1903	extreme = ((policy.zrp_request & ZPOOL_EXTREME_REWIND) != 0);
1904
1905	/*
1906	 * As disgusting as this is, we need to support recursive calls to this
1907	 * function because dsl_dir_open() is called during spa_load(), and ends
1908	 * up calling spa_open() again.  The real fix is to figure out how to
1909	 * avoid dsl_dir_open() calling this in the first place.
1910	 */
1911	if (mutex_owner(&spa_namespace_lock) != curthread) {
1912		mutex_enter(&spa_namespace_lock);
1913		locked = B_TRUE;
1914	}
1915
1916	if ((spa = spa_lookup(pool)) == NULL) {
1917		if (locked)
1918			mutex_exit(&spa_namespace_lock);
1919		return (ENOENT);
1920	}
1921
1922	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
1923
1924		spa_activate(spa, spa_mode_global);
1925
1926		if (spa->spa_last_open_failed && norewind) {
1927			if (config != NULL && spa->spa_config)
1928				VERIFY(nvlist_dup(spa->spa_config,
1929				    config, KM_SLEEP) == 0);
1930			spa_deactivate(spa);
1931			if (locked)
1932				mutex_exit(&spa_namespace_lock);
1933			return (spa->spa_last_open_failed);
1934		}
1935
1936		if (state != SPA_LOAD_RECOVER)
1937			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
1938
1939		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
1940		    extreme);
1941
1942		if (error == EBADF) {
1943			/*
1944			 * If vdev_validate() returns failure (indicated by
1945			 * EBADF), it indicates that one of the vdevs indicates
1946			 * that the pool has been exported or destroyed.  If
1947			 * this is the case, the config cache is out of sync and
1948			 * we should remove the pool from the namespace.
1949			 */
1950			spa_unload(spa);
1951			spa_deactivate(spa);
1952			spa_config_sync(spa, B_TRUE, B_TRUE);
1953			spa_remove(spa);
1954			if (locked)
1955				mutex_exit(&spa_namespace_lock);
1956			return (ENOENT);
1957		}
1958
1959		if (error) {
1960			/*
1961			 * We can't open the pool, but we still have useful
1962			 * information: the state of each vdev after the
1963			 * attempted vdev_open().  Return this to the user.
1964			 */
1965			if (config != NULL && spa->spa_config)
1966				VERIFY(nvlist_dup(spa->spa_config, config,
1967				    KM_SLEEP) == 0);
1968			spa_unload(spa);
1969			spa_deactivate(spa);
1970			spa->spa_last_open_failed = error;
1971			if (locked)
1972				mutex_exit(&spa_namespace_lock);
1973			*spapp = NULL;
1974			return (error);
1975		}
1976
1977	}
1978
1979	spa_open_ref(spa, tag);
1980
1981	spa->spa_last_open_failed = 0;
1982
1983	if (config != NULL)
1984		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
1985
1986	spa->spa_last_ubsync_txg = 0;
1987	spa->spa_load_txg = 0;
1988
1989	if (locked)
1990		mutex_exit(&spa_namespace_lock);
1991
1992	*spapp = spa;
1993
1994	return (0);
1995}
1996
1997int
1998spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
1999    nvlist_t **config)
2000{
2001	return (spa_open_common(name, spapp, tag, policy, config));
2002}
2003
2004int
2005spa_open(const char *name, spa_t **spapp, void *tag)
2006{
2007	return (spa_open_common(name, spapp, tag, NULL, NULL));
2008}
2009
2010/*
2011 * Lookup the given spa_t, incrementing the inject count in the process,
2012 * preventing it from being exported or destroyed.
2013 */
2014spa_t *
2015spa_inject_addref(char *name)
2016{
2017	spa_t *spa;
2018
2019	mutex_enter(&spa_namespace_lock);
2020	if ((spa = spa_lookup(name)) == NULL) {
2021		mutex_exit(&spa_namespace_lock);
2022		return (NULL);
2023	}
2024	spa->spa_inject_ref++;
2025	mutex_exit(&spa_namespace_lock);
2026
2027	return (spa);
2028}
2029
2030void
2031spa_inject_delref(spa_t *spa)
2032{
2033	mutex_enter(&spa_namespace_lock);
2034	spa->spa_inject_ref--;
2035	mutex_exit(&spa_namespace_lock);
2036}
2037
2038/*
2039 * Add spares device information to the nvlist.
2040 */
2041static void
2042spa_add_spares(spa_t *spa, nvlist_t *config)
2043{
2044	nvlist_t **spares;
2045	uint_t i, nspares;
2046	nvlist_t *nvroot;
2047	uint64_t guid;
2048	vdev_stat_t *vs;
2049	uint_t vsc;
2050	uint64_t pool;
2051
2052	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2053
2054	if (spa->spa_spares.sav_count == 0)
2055		return;
2056
2057	VERIFY(nvlist_lookup_nvlist(config,
2058	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2059	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2060	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2061	if (nspares != 0) {
2062		VERIFY(nvlist_add_nvlist_array(nvroot,
2063		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2064		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2065		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2066
2067		/*
2068		 * Go through and find any spares which have since been
2069		 * repurposed as an active spare.  If this is the case, update
2070		 * their status appropriately.
2071		 */
2072		for (i = 0; i < nspares; i++) {
2073			VERIFY(nvlist_lookup_uint64(spares[i],
2074			    ZPOOL_CONFIG_GUID, &guid) == 0);
2075			if (spa_spare_exists(guid, &pool, NULL) &&
2076			    pool != 0ULL) {
2077				VERIFY(nvlist_lookup_uint64_array(
2078				    spares[i], ZPOOL_CONFIG_STATS,
2079				    (uint64_t **)&vs, &vsc) == 0);
2080				vs->vs_state = VDEV_STATE_CANT_OPEN;
2081				vs->vs_aux = VDEV_AUX_SPARED;
2082			}
2083		}
2084	}
2085}
2086
2087/*
2088 * Add l2cache device information to the nvlist, including vdev stats.
2089 */
2090static void
2091spa_add_l2cache(spa_t *spa, nvlist_t *config)
2092{
2093	nvlist_t **l2cache;
2094	uint_t i, j, nl2cache;
2095	nvlist_t *nvroot;
2096	uint64_t guid;
2097	vdev_t *vd;
2098	vdev_stat_t *vs;
2099	uint_t vsc;
2100
2101	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2102
2103	if (spa->spa_l2cache.sav_count == 0)
2104		return;
2105
2106	VERIFY(nvlist_lookup_nvlist(config,
2107	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2108	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2109	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2110	if (nl2cache != 0) {
2111		VERIFY(nvlist_add_nvlist_array(nvroot,
2112		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2113		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2114		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2115
2116		/*
2117		 * Update level 2 cache device stats.
2118		 */
2119
2120		for (i = 0; i < nl2cache; i++) {
2121			VERIFY(nvlist_lookup_uint64(l2cache[i],
2122			    ZPOOL_CONFIG_GUID, &guid) == 0);
2123
2124			vd = NULL;
2125			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2126				if (guid ==
2127				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2128					vd = spa->spa_l2cache.sav_vdevs[j];
2129					break;
2130				}
2131			}
2132			ASSERT(vd != NULL);
2133
2134			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2135			    ZPOOL_CONFIG_STATS, (uint64_t **)&vs, &vsc) == 0);
2136			vdev_get_stats(vd, vs);
2137		}
2138	}
2139}
2140
2141int
2142spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2143{
2144	int error;
2145	spa_t *spa;
2146
2147	*config = NULL;
2148	error = spa_open_common(name, &spa, FTAG, NULL, config);
2149
2150	if (spa != NULL) {
2151		/*
2152		 * This still leaves a window of inconsistency where the spares
2153		 * or l2cache devices could change and the config would be
2154		 * self-inconsistent.
2155		 */
2156		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2157
2158		if (*config != NULL) {
2159			VERIFY(nvlist_add_uint64(*config,
2160			    ZPOOL_CONFIG_ERRCOUNT,
2161			    spa_get_errlog_size(spa)) == 0);
2162
2163			if (spa_suspended(spa))
2164				VERIFY(nvlist_add_uint64(*config,
2165				    ZPOOL_CONFIG_SUSPENDED,
2166				    spa->spa_failmode) == 0);
2167
2168			spa_add_spares(spa, *config);
2169			spa_add_l2cache(spa, *config);
2170		}
2171	}
2172
2173	/*
2174	 * We want to get the alternate root even for faulted pools, so we cheat
2175	 * and call spa_lookup() directly.
2176	 */
2177	if (altroot) {
2178		if (spa == NULL) {
2179			mutex_enter(&spa_namespace_lock);
2180			spa = spa_lookup(name);
2181			if (spa)
2182				spa_altroot(spa, altroot, buflen);
2183			else
2184				altroot[0] = '\0';
2185			spa = NULL;
2186			mutex_exit(&spa_namespace_lock);
2187		} else {
2188			spa_altroot(spa, altroot, buflen);
2189		}
2190	}
2191
2192	if (spa != NULL) {
2193		spa_config_exit(spa, SCL_CONFIG, FTAG);
2194		spa_close(spa, FTAG);
2195	}
2196
2197	return (error);
2198}
2199
2200/*
2201 * Validate that the auxiliary device array is well formed.  We must have an
2202 * array of nvlists, each which describes a valid leaf vdev.  If this is an
2203 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2204 * specified, as long as they are well-formed.
2205 */
2206static int
2207spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2208    spa_aux_vdev_t *sav, const char *config, uint64_t version,
2209    vdev_labeltype_t label)
2210{
2211	nvlist_t **dev;
2212	uint_t i, ndev;
2213	vdev_t *vd;
2214	int error;
2215
2216	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2217
2218	/*
2219	 * It's acceptable to have no devs specified.
2220	 */
2221	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2222		return (0);
2223
2224	if (ndev == 0)
2225		return (EINVAL);
2226
2227	/*
2228	 * Make sure the pool is formatted with a version that supports this
2229	 * device type.
2230	 */
2231	if (spa_version(spa) < version)
2232		return (ENOTSUP);
2233
2234	/*
2235	 * Set the pending device list so we correctly handle device in-use
2236	 * checking.
2237	 */
2238	sav->sav_pending = dev;
2239	sav->sav_npending = ndev;
2240
2241	for (i = 0; i < ndev; i++) {
2242		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2243		    mode)) != 0)
2244			goto out;
2245
2246		if (!vd->vdev_ops->vdev_op_leaf) {
2247			vdev_free(vd);
2248			error = EINVAL;
2249			goto out;
2250		}
2251
2252		/*
2253		 * The L2ARC currently only supports disk devices in
2254		 * kernel context.  For user-level testing, we allow it.
2255		 */
2256#ifdef _KERNEL
2257		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2258		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2259			error = ENOTBLK;
2260			goto out;
2261		}
2262#endif
2263		vd->vdev_top = vd;
2264
2265		if ((error = vdev_open(vd)) == 0 &&
2266		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
2267			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2268			    vd->vdev_guid) == 0);
2269		}
2270
2271		vdev_free(vd);
2272
2273		if (error &&
2274		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2275			goto out;
2276		else
2277			error = 0;
2278	}
2279
2280out:
2281	sav->sav_pending = NULL;
2282	sav->sav_npending = 0;
2283	return (error);
2284}
2285
2286static int
2287spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2288{
2289	int error;
2290
2291	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2292
2293	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2294	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2295	    VDEV_LABEL_SPARE)) != 0) {
2296		return (error);
2297	}
2298
2299	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2300	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2301	    VDEV_LABEL_L2CACHE));
2302}
2303
2304static void
2305spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2306    const char *config)
2307{
2308	int i;
2309
2310	if (sav->sav_config != NULL) {
2311		nvlist_t **olddevs;
2312		uint_t oldndevs;
2313		nvlist_t **newdevs;
2314
2315		/*
2316		 * Generate new dev list by concatentating with the
2317		 * current dev list.
2318		 */
2319		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2320		    &olddevs, &oldndevs) == 0);
2321
2322		newdevs = kmem_alloc(sizeof (void *) *
2323		    (ndevs + oldndevs), KM_SLEEP);
2324		for (i = 0; i < oldndevs; i++)
2325			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2326			    KM_SLEEP) == 0);
2327		for (i = 0; i < ndevs; i++)
2328			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2329			    KM_SLEEP) == 0);
2330
2331		VERIFY(nvlist_remove(sav->sav_config, config,
2332		    DATA_TYPE_NVLIST_ARRAY) == 0);
2333
2334		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2335		    config, newdevs, ndevs + oldndevs) == 0);
2336		for (i = 0; i < oldndevs + ndevs; i++)
2337			nvlist_free(newdevs[i]);
2338		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2339	} else {
2340		/*
2341		 * Generate a new dev list.
2342		 */
2343		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2344		    KM_SLEEP) == 0);
2345		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2346		    devs, ndevs) == 0);
2347	}
2348}
2349
2350/*
2351 * Stop and drop level 2 ARC devices
2352 */
2353void
2354spa_l2cache_drop(spa_t *spa)
2355{
2356	vdev_t *vd;
2357	int i;
2358	spa_aux_vdev_t *sav = &spa->spa_l2cache;
2359
2360	for (i = 0; i < sav->sav_count; i++) {
2361		uint64_t pool;
2362
2363		vd = sav->sav_vdevs[i];
2364		ASSERT(vd != NULL);
2365
2366		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2367		    pool != 0ULL && l2arc_vdev_present(vd))
2368			l2arc_remove_vdev(vd);
2369		if (vd->vdev_isl2cache)
2370			spa_l2cache_remove(vd);
2371		vdev_clear_stats(vd);
2372		(void) vdev_close(vd);
2373	}
2374}
2375
2376/*
2377 * Pool Creation
2378 */
2379int
2380spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2381    const char *history_str, nvlist_t *zplprops)
2382{
2383	spa_t *spa;
2384	char *altroot = NULL;
2385	vdev_t *rvd;
2386	dsl_pool_t *dp;
2387	dmu_tx_t *tx;
2388	int error = 0;
2389	uint64_t txg = TXG_INITIAL;
2390	nvlist_t **spares, **l2cache;
2391	uint_t nspares, nl2cache;
2392	uint64_t version;
2393
2394	/*
2395	 * If this pool already exists, return failure.
2396	 */
2397	mutex_enter(&spa_namespace_lock);
2398	if (spa_lookup(pool) != NULL) {
2399		mutex_exit(&spa_namespace_lock);
2400		return (EEXIST);
2401	}
2402
2403	/*
2404	 * Allocate a new spa_t structure.
2405	 */
2406	(void) nvlist_lookup_string(props,
2407	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2408	spa = spa_add(pool, NULL, altroot);
2409	spa_activate(spa, spa_mode_global);
2410
2411	if (props && (error = spa_prop_validate(spa, props))) {
2412		spa_deactivate(spa);
2413		spa_remove(spa);
2414		mutex_exit(&spa_namespace_lock);
2415		return (error);
2416	}
2417
2418	if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2419	    &version) != 0)
2420		version = SPA_VERSION;
2421	ASSERT(version <= SPA_VERSION);
2422
2423	spa->spa_first_txg = txg;
2424	spa->spa_uberblock.ub_txg = txg - 1;
2425	spa->spa_uberblock.ub_version = version;
2426	spa->spa_ubsync = spa->spa_uberblock;
2427
2428	/*
2429	 * Create "The Godfather" zio to hold all async IOs
2430	 */
2431	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2432	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2433
2434	/*
2435	 * Create the root vdev.
2436	 */
2437	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2438
2439	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2440
2441	ASSERT(error != 0 || rvd != NULL);
2442	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2443
2444	if (error == 0 && !zfs_allocatable_devs(nvroot))
2445		error = EINVAL;
2446
2447	if (error == 0 &&
2448	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2449	    (error = spa_validate_aux(spa, nvroot, txg,
2450	    VDEV_ALLOC_ADD)) == 0) {
2451		for (int c = 0; c < rvd->vdev_children; c++) {
2452			vdev_metaslab_set_size(rvd->vdev_child[c]);
2453			vdev_expand(rvd->vdev_child[c], txg);
2454		}
2455	}
2456
2457	spa_config_exit(spa, SCL_ALL, FTAG);
2458
2459	if (error != 0) {
2460		spa_unload(spa);
2461		spa_deactivate(spa);
2462		spa_remove(spa);
2463		mutex_exit(&spa_namespace_lock);
2464		return (error);
2465	}
2466
2467	/*
2468	 * Get the list of spares, if specified.
2469	 */
2470	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2471	    &spares, &nspares) == 0) {
2472		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
2473		    KM_SLEEP) == 0);
2474		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2475		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2476		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2477		spa_load_spares(spa);
2478		spa_config_exit(spa, SCL_ALL, FTAG);
2479		spa->spa_spares.sav_sync = B_TRUE;
2480	}
2481
2482	/*
2483	 * Get the list of level 2 cache devices, if specified.
2484	 */
2485	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2486	    &l2cache, &nl2cache) == 0) {
2487		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2488		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2489		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2490		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2491		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2492		spa_load_l2cache(spa);
2493		spa_config_exit(spa, SCL_ALL, FTAG);
2494		spa->spa_l2cache.sav_sync = B_TRUE;
2495	}
2496
2497	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
2498	spa->spa_meta_objset = dp->dp_meta_objset;
2499
2500	/*
2501	 * Create DDTs (dedup tables).
2502	 */
2503	ddt_create(spa);
2504
2505	spa_update_dspace(spa);
2506
2507	tx = dmu_tx_create_assigned(dp, txg);
2508
2509	/*
2510	 * Create the pool config object.
2511	 */
2512	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
2513	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
2514	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2515
2516	if (zap_add(spa->spa_meta_objset,
2517	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
2518	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
2519		cmn_err(CE_PANIC, "failed to add pool config");
2520	}
2521
2522	/* Newly created pools with the right version are always deflated. */
2523	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
2524		spa->spa_deflate = TRUE;
2525		if (zap_add(spa->spa_meta_objset,
2526		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
2527		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
2528			cmn_err(CE_PANIC, "failed to add deflate");
2529		}
2530	}
2531
2532	/*
2533	 * Create the deferred-free bplist object.  Turn off compression
2534	 * because sync-to-convergence takes longer if the blocksize
2535	 * keeps changing.
2536	 */
2537	spa->spa_deferred_bplist_obj = bplist_create(spa->spa_meta_objset,
2538	    1 << 14, tx);
2539	dmu_object_set_compress(spa->spa_meta_objset,
2540	    spa->spa_deferred_bplist_obj, ZIO_COMPRESS_OFF, tx);
2541
2542	if (zap_add(spa->spa_meta_objset,
2543	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
2544	    sizeof (uint64_t), 1, &spa->spa_deferred_bplist_obj, tx) != 0) {
2545		cmn_err(CE_PANIC, "failed to add bplist");
2546	}
2547
2548	/*
2549	 * Create the pool's history object.
2550	 */
2551	if (version >= SPA_VERSION_ZPOOL_HISTORY)
2552		spa_history_create_obj(spa, tx);
2553
2554	/*
2555	 * Set pool properties.
2556	 */
2557	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
2558	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2559	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
2560	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
2561
2562	if (props != NULL) {
2563		spa_configfile_set(spa, props, B_FALSE);
2564		spa_sync_props(spa, props, CRED(), tx);
2565	}
2566
2567	dmu_tx_commit(tx);
2568
2569	spa->spa_sync_on = B_TRUE;
2570	txg_sync_start(spa->spa_dsl_pool);
2571
2572	/*
2573	 * We explicitly wait for the first transaction to complete so that our
2574	 * bean counters are appropriately updated.
2575	 */
2576	txg_wait_synced(spa->spa_dsl_pool, txg);
2577
2578	spa_config_sync(spa, B_FALSE, B_TRUE);
2579
2580	if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
2581		(void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
2582	spa_history_log_version(spa, LOG_POOL_CREATE);
2583
2584	spa->spa_minref = refcount_count(&spa->spa_refcount);
2585
2586	mutex_exit(&spa_namespace_lock);
2587
2588	return (0);
2589}
2590
2591#ifdef _KERNEL
2592/*
2593 * Get the root pool information from the root disk, then import the root pool
2594 * during the system boot up time.
2595 */
2596extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
2597
2598static nvlist_t *
2599spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
2600{
2601	nvlist_t *config;
2602	nvlist_t *nvtop, *nvroot;
2603	uint64_t pgid;
2604
2605	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
2606		return (NULL);
2607
2608	/*
2609	 * Add this top-level vdev to the child array.
2610	 */
2611	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2612	    &nvtop) == 0);
2613	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
2614	    &pgid) == 0);
2615	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
2616
2617	/*
2618	 * Put this pool's top-level vdevs into a root vdev.
2619	 */
2620	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2621	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
2622	    VDEV_TYPE_ROOT) == 0);
2623	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
2624	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
2625	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2626	    &nvtop, 1) == 0);
2627
2628	/*
2629	 * Replace the existing vdev_tree with the new root vdev in
2630	 * this pool's configuration (remove the old, add the new).
2631	 */
2632	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
2633	nvlist_free(nvroot);
2634	return (config);
2635}
2636
2637/*
2638 * Walk the vdev tree and see if we can find a device with "better"
2639 * configuration. A configuration is "better" if the label on that
2640 * device has a more recent txg.
2641 */
2642static void
2643spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
2644{
2645	for (int c = 0; c < vd->vdev_children; c++)
2646		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
2647
2648	if (vd->vdev_ops->vdev_op_leaf) {
2649		nvlist_t *label;
2650		uint64_t label_txg;
2651
2652		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
2653		    &label) != 0)
2654			return;
2655
2656		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
2657		    &label_txg) == 0);
2658
2659		/*
2660		 * Do we have a better boot device?
2661		 */
2662		if (label_txg > *txg) {
2663			*txg = label_txg;
2664			*avd = vd;
2665		}
2666		nvlist_free(label);
2667	}
2668}
2669
2670/*
2671 * Import a root pool.
2672 *
2673 * For x86. devpath_list will consist of devid and/or physpath name of
2674 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
2675 * The GRUB "findroot" command will return the vdev we should boot.
2676 *
2677 * For Sparc, devpath_list consists the physpath name of the booting device
2678 * no matter the rootpool is a single device pool or a mirrored pool.
2679 * e.g.
2680 *	"/pci@1f,0/ide@d/disk@0,0:a"
2681 */
2682int
2683spa_import_rootpool(char *devpath, char *devid)
2684{
2685	spa_t *spa;
2686	vdev_t *rvd, *bvd, *avd = NULL;
2687	nvlist_t *config, *nvtop;
2688	uint64_t guid, txg;
2689	char *pname;
2690	int error;
2691
2692	/*
2693	 * Read the label from the boot device and generate a configuration.
2694	 */
2695	config = spa_generate_rootconf(devpath, devid, &guid);
2696#if defined(_OBP) && defined(_KERNEL)
2697	if (config == NULL) {
2698		if (strstr(devpath, "/iscsi/ssd") != NULL) {
2699			/* iscsi boot */
2700			get_iscsi_bootpath_phy(devpath);
2701			config = spa_generate_rootconf(devpath, devid, &guid);
2702		}
2703	}
2704#endif
2705	if (config == NULL) {
2706		cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
2707		    devpath);
2708		return (EIO);
2709	}
2710
2711	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
2712	    &pname) == 0);
2713	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
2714
2715	mutex_enter(&spa_namespace_lock);
2716	if ((spa = spa_lookup(pname)) != NULL) {
2717		/*
2718		 * Remove the existing root pool from the namespace so that we
2719		 * can replace it with the correct config we just read in.
2720		 */
2721		spa_remove(spa);
2722	}
2723
2724	spa = spa_add(pname, config, NULL);
2725	spa->spa_is_root = B_TRUE;
2726	spa->spa_load_verbatim = B_TRUE;
2727
2728	/*
2729	 * Build up a vdev tree based on the boot device's label config.
2730	 */
2731	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2732	    &nvtop) == 0);
2733	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2734	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
2735	    VDEV_ALLOC_ROOTPOOL);
2736	spa_config_exit(spa, SCL_ALL, FTAG);
2737	if (error) {
2738		mutex_exit(&spa_namespace_lock);
2739		nvlist_free(config);
2740		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
2741		    pname);
2742		return (error);
2743	}
2744
2745	/*
2746	 * Get the boot vdev.
2747	 */
2748	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
2749		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
2750		    (u_longlong_t)guid);
2751		error = ENOENT;
2752		goto out;
2753	}
2754
2755	/*
2756	 * Determine if there is a better boot device.
2757	 */
2758	avd = bvd;
2759	spa_alt_rootvdev(rvd, &avd, &txg);
2760	if (avd != bvd) {
2761		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
2762		    "try booting from '%s'", avd->vdev_path);
2763		error = EINVAL;
2764		goto out;
2765	}
2766
2767	/*
2768	 * If the boot device is part of a spare vdev then ensure that
2769	 * we're booting off the active spare.
2770	 */
2771	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2772	    !bvd->vdev_isspare) {
2773		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
2774		    "try booting from '%s'",
2775		    bvd->vdev_parent->vdev_child[1]->vdev_path);
2776		error = EINVAL;
2777		goto out;
2778	}
2779
2780	error = 0;
2781	spa_history_log_version(spa, LOG_POOL_IMPORT);
2782out:
2783	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2784	vdev_free(rvd);
2785	spa_config_exit(spa, SCL_ALL, FTAG);
2786	mutex_exit(&spa_namespace_lock);
2787
2788	nvlist_free(config);
2789	return (error);
2790}
2791
2792#endif
2793
2794/*
2795 * Take a pool and insert it into the namespace as if it had been loaded at
2796 * boot.
2797 */
2798int
2799spa_import_verbatim(const char *pool, nvlist_t *config, nvlist_t *props)
2800{
2801	spa_t *spa;
2802	zpool_rewind_policy_t policy;
2803	char *altroot = NULL;
2804
2805	mutex_enter(&spa_namespace_lock);
2806	if (spa_lookup(pool) != NULL) {
2807		mutex_exit(&spa_namespace_lock);
2808		return (EEXIST);
2809	}
2810
2811	(void) nvlist_lookup_string(props,
2812	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2813	spa = spa_add(pool, config, altroot);
2814
2815	zpool_get_rewind_policy(config, &policy);
2816	spa->spa_load_max_txg = policy.zrp_txg;
2817
2818	spa->spa_load_verbatim = B_TRUE;
2819
2820	if (props != NULL)
2821		spa_configfile_set(spa, props, B_FALSE);
2822
2823	spa_config_sync(spa, B_FALSE, B_TRUE);
2824
2825	mutex_exit(&spa_namespace_lock);
2826	spa_history_log_version(spa, LOG_POOL_IMPORT);
2827
2828	return (0);
2829}
2830
2831/*
2832 * Import a non-root pool into the system.
2833 */
2834int
2835spa_import(const char *pool, nvlist_t *config, nvlist_t *props)
2836{
2837	spa_t *spa;
2838	char *altroot = NULL;
2839	spa_load_state_t state = SPA_LOAD_IMPORT;
2840	zpool_rewind_policy_t policy;
2841	int error;
2842	nvlist_t *nvroot;
2843	nvlist_t **spares, **l2cache;
2844	uint_t nspares, nl2cache;
2845
2846	/*
2847	 * If a pool with this name exists, return failure.
2848	 */
2849	mutex_enter(&spa_namespace_lock);
2850	if ((spa = spa_lookup(pool)) != NULL) {
2851		mutex_exit(&spa_namespace_lock);
2852		return (EEXIST);
2853	}
2854
2855	zpool_get_rewind_policy(config, &policy);
2856	if (policy.zrp_request & ZPOOL_DO_REWIND)
2857		state = SPA_LOAD_RECOVER;
2858
2859	/*
2860	 * Create and initialize the spa structure.
2861	 */
2862	(void) nvlist_lookup_string(props,
2863	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2864	spa = spa_add(pool, config, altroot);
2865	spa_activate(spa, spa_mode_global);
2866
2867	/*
2868	 * Don't start async tasks until we know everything is healthy.
2869	 */
2870	spa_async_suspend(spa);
2871
2872	/*
2873	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
2874	 * because the user-supplied config is actually the one to trust when
2875	 * doing an import.
2876	 */
2877	if (state != SPA_LOAD_RECOVER)
2878		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2879	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
2880	    ((policy.zrp_request & ZPOOL_EXTREME_REWIND) != 0));
2881
2882	/*
2883	 * Propagate anything learned about failing or best txgs
2884	 * back to caller
2885	 */
2886	spa_rewind_data_to_nvlist(spa, config);
2887
2888	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2889	/*
2890	 * Toss any existing sparelist, as it doesn't have any validity
2891	 * anymore, and conflicts with spa_has_spare().
2892	 */
2893	if (spa->spa_spares.sav_config) {
2894		nvlist_free(spa->spa_spares.sav_config);
2895		spa->spa_spares.sav_config = NULL;
2896		spa_load_spares(spa);
2897	}
2898	if (spa->spa_l2cache.sav_config) {
2899		nvlist_free(spa->spa_l2cache.sav_config);
2900		spa->spa_l2cache.sav_config = NULL;
2901		spa_load_l2cache(spa);
2902	}
2903
2904	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2905	    &nvroot) == 0);
2906	if (error == 0)
2907		error = spa_validate_aux(spa, nvroot, -1ULL,
2908		    VDEV_ALLOC_SPARE);
2909	if (error == 0)
2910		error = spa_validate_aux(spa, nvroot, -1ULL,
2911		    VDEV_ALLOC_L2CACHE);
2912	spa_config_exit(spa, SCL_ALL, FTAG);
2913
2914	if (props != NULL)
2915		spa_configfile_set(spa, props, B_FALSE);
2916
2917	if (error != 0 || (props && spa_writeable(spa) &&
2918	    (error = spa_prop_set(spa, props)))) {
2919		spa_unload(spa);
2920		spa_deactivate(spa);
2921		spa_remove(spa);
2922		mutex_exit(&spa_namespace_lock);
2923		return (error);
2924	}
2925
2926	spa_async_resume(spa);
2927
2928	/*
2929	 * Override any spares and level 2 cache devices as specified by
2930	 * the user, as these may have correct device names/devids, etc.
2931	 */
2932	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2933	    &spares, &nspares) == 0) {
2934		if (spa->spa_spares.sav_config)
2935			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
2936			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
2937		else
2938			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
2939			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2940		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2941		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2942		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2943		spa_load_spares(spa);
2944		spa_config_exit(spa, SCL_ALL, FTAG);
2945		spa->spa_spares.sav_sync = B_TRUE;
2946	}
2947	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2948	    &l2cache, &nl2cache) == 0) {
2949		if (spa->spa_l2cache.sav_config)
2950			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
2951			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
2952		else
2953			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2954			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2955		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2956		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2957		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2958		spa_load_l2cache(spa);
2959		spa_config_exit(spa, SCL_ALL, FTAG);
2960		spa->spa_l2cache.sav_sync = B_TRUE;
2961	}
2962
2963	/*
2964	 * Check for any removed devices.
2965	 */
2966	if (spa->spa_autoreplace) {
2967		spa_aux_check_removed(&spa->spa_spares);
2968		spa_aux_check_removed(&spa->spa_l2cache);
2969	}
2970
2971	if (spa_writeable(spa)) {
2972		/*
2973		 * Update the config cache to include the newly-imported pool.
2974		 */
2975		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
2976	}
2977
2978	/*
2979	 * It's possible that the pool was expanded while it was exported.
2980	 * We kick off an async task to handle this for us.
2981	 */
2982	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
2983
2984	mutex_exit(&spa_namespace_lock);
2985	spa_history_log_version(spa, LOG_POOL_IMPORT);
2986
2987	return (0);
2988}
2989
2990
2991/*
2992 * This (illegal) pool name is used when temporarily importing a spa_t in order
2993 * to get the vdev stats associated with the imported devices.
2994 */
2995#define	TRYIMPORT_NAME	"$import"
2996
2997nvlist_t *
2998spa_tryimport(nvlist_t *tryconfig)
2999{
3000	nvlist_t *config = NULL;
3001	char *poolname;
3002	spa_t *spa;
3003	uint64_t state;
3004	int error;
3005
3006	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3007		return (NULL);
3008
3009	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3010		return (NULL);
3011
3012	/*
3013	 * Create and initialize the spa structure.
3014	 */
3015	mutex_enter(&spa_namespace_lock);
3016	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3017	spa_activate(spa, FREAD);
3018
3019	/*
3020	 * Pass off the heavy lifting to spa_load().
3021	 * Pass TRUE for mosconfig because the user-supplied config
3022	 * is actually the one to trust when doing an import.
3023	 */
3024	error = spa_load(spa, SPA_LOAD_TRYIMPORT, B_TRUE);
3025
3026	/*
3027	 * If 'tryconfig' was at least parsable, return the current config.
3028	 */
3029	if (spa->spa_root_vdev != NULL) {
3030		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3031		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3032		    poolname) == 0);
3033		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3034		    state) == 0);
3035		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3036		    spa->spa_uberblock.ub_timestamp) == 0);
3037
3038		/*
3039		 * If the bootfs property exists on this pool then we
3040		 * copy it out so that external consumers can tell which
3041		 * pools are bootable.
3042		 */
3043		if ((!error || error == EEXIST) && spa->spa_bootfs) {
3044			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3045
3046			/*
3047			 * We have to play games with the name since the
3048			 * pool was opened as TRYIMPORT_NAME.
3049			 */
3050			if (dsl_dsobj_to_dsname(spa_name(spa),
3051			    spa->spa_bootfs, tmpname) == 0) {
3052				char *cp;
3053				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3054
3055				cp = strchr(tmpname, '/');
3056				if (cp == NULL) {
3057					(void) strlcpy(dsname, tmpname,
3058					    MAXPATHLEN);
3059				} else {
3060					(void) snprintf(dsname, MAXPATHLEN,
3061					    "%s/%s", poolname, ++cp);
3062				}
3063				VERIFY(nvlist_add_string(config,
3064				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3065				kmem_free(dsname, MAXPATHLEN);
3066			}
3067			kmem_free(tmpname, MAXPATHLEN);
3068		}
3069
3070		/*
3071		 * Add the list of hot spares and level 2 cache devices.
3072		 */
3073		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3074		spa_add_spares(spa, config);
3075		spa_add_l2cache(spa, config);
3076		spa_config_exit(spa, SCL_CONFIG, FTAG);
3077	}
3078
3079	spa_unload(spa);
3080	spa_deactivate(spa);
3081	spa_remove(spa);
3082	mutex_exit(&spa_namespace_lock);
3083
3084	return (config);
3085}
3086
3087/*
3088 * Pool export/destroy
3089 *
3090 * The act of destroying or exporting a pool is very simple.  We make sure there
3091 * is no more pending I/O and any references to the pool are gone.  Then, we
3092 * update the pool state and sync all the labels to disk, removing the
3093 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3094 * we don't sync the labels or remove the configuration cache.
3095 */
3096static int
3097spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3098    boolean_t force, boolean_t hardforce)
3099{
3100	spa_t *spa;
3101
3102	if (oldconfig)
3103		*oldconfig = NULL;
3104
3105	if (!(spa_mode_global & FWRITE))
3106		return (EROFS);
3107
3108	mutex_enter(&spa_namespace_lock);
3109	if ((spa = spa_lookup(pool)) == NULL) {
3110		mutex_exit(&spa_namespace_lock);
3111		return (ENOENT);
3112	}
3113
3114	/*
3115	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3116	 * reacquire the namespace lock, and see if we can export.
3117	 */
3118	spa_open_ref(spa, FTAG);
3119	mutex_exit(&spa_namespace_lock);
3120	spa_async_suspend(spa);
3121	mutex_enter(&spa_namespace_lock);
3122	spa_close(spa, FTAG);
3123
3124	/*
3125	 * The pool will be in core if it's openable,
3126	 * in which case we can modify its state.
3127	 */
3128	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3129		/*
3130		 * Objsets may be open only because they're dirty, so we
3131		 * have to force it to sync before checking spa_refcnt.
3132		 */
3133		txg_wait_synced(spa->spa_dsl_pool, 0);
3134
3135		/*
3136		 * A pool cannot be exported or destroyed if there are active
3137		 * references.  If we are resetting a pool, allow references by
3138		 * fault injection handlers.
3139		 */
3140		if (!spa_refcount_zero(spa) ||
3141		    (spa->spa_inject_ref != 0 &&
3142		    new_state != POOL_STATE_UNINITIALIZED)) {
3143			spa_async_resume(spa);
3144			mutex_exit(&spa_namespace_lock);
3145			return (EBUSY);
3146		}
3147
3148		/*
3149		 * A pool cannot be exported if it has an active shared spare.
3150		 * This is to prevent other pools stealing the active spare
3151		 * from an exported pool. At user's own will, such pool can
3152		 * be forcedly exported.
3153		 */
3154		if (!force && new_state == POOL_STATE_EXPORTED &&
3155		    spa_has_active_shared_spare(spa)) {
3156			spa_async_resume(spa);
3157			mutex_exit(&spa_namespace_lock);
3158			return (EXDEV);
3159		}
3160
3161		/*
3162		 * We want this to be reflected on every label,
3163		 * so mark them all dirty.  spa_unload() will do the
3164		 * final sync that pushes these changes out.
3165		 */
3166		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3167			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3168			spa->spa_state = new_state;
3169			spa->spa_final_txg = spa_last_synced_txg(spa) + 1;
3170			vdev_config_dirty(spa->spa_root_vdev);
3171			spa_config_exit(spa, SCL_ALL, FTAG);
3172		}
3173	}
3174
3175	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
3176
3177	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3178		spa_unload(spa);
3179		spa_deactivate(spa);
3180	}
3181
3182	if (oldconfig && spa->spa_config)
3183		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3184
3185	if (new_state != POOL_STATE_UNINITIALIZED) {
3186		if (!hardforce)
3187			spa_config_sync(spa, B_TRUE, B_TRUE);
3188		spa_remove(spa);
3189	}
3190	mutex_exit(&spa_namespace_lock);
3191
3192	return (0);
3193}
3194
3195/*
3196 * Destroy a storage pool.
3197 */
3198int
3199spa_destroy(char *pool)
3200{
3201	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3202	    B_FALSE, B_FALSE));
3203}
3204
3205/*
3206 * Export a storage pool.
3207 */
3208int
3209spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3210    boolean_t hardforce)
3211{
3212	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3213	    force, hardforce));
3214}
3215
3216/*
3217 * Similar to spa_export(), this unloads the spa_t without actually removing it
3218 * from the namespace in any way.
3219 */
3220int
3221spa_reset(char *pool)
3222{
3223	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3224	    B_FALSE, B_FALSE));
3225}
3226
3227/*
3228 * ==========================================================================
3229 * Device manipulation
3230 * ==========================================================================
3231 */
3232
3233/*
3234 * Add a device to a storage pool.
3235 */
3236int
3237spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3238{
3239	uint64_t txg, id;
3240	int error;
3241	vdev_t *rvd = spa->spa_root_vdev;
3242	vdev_t *vd, *tvd;
3243	nvlist_t **spares, **l2cache;
3244	uint_t nspares, nl2cache;
3245
3246	txg = spa_vdev_enter(spa);
3247
3248	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3249	    VDEV_ALLOC_ADD)) != 0)
3250		return (spa_vdev_exit(spa, NULL, txg, error));
3251
3252	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
3253
3254	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3255	    &nspares) != 0)
3256		nspares = 0;
3257
3258	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3259	    &nl2cache) != 0)
3260		nl2cache = 0;
3261
3262	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3263		return (spa_vdev_exit(spa, vd, txg, EINVAL));
3264
3265	if (vd->vdev_children != 0 &&
3266	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
3267		return (spa_vdev_exit(spa, vd, txg, error));
3268
3269	/*
3270	 * We must validate the spares and l2cache devices after checking the
3271	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
3272	 */
3273	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3274		return (spa_vdev_exit(spa, vd, txg, error));
3275
3276	/*
3277	 * Transfer each new top-level vdev from vd to rvd.
3278	 */
3279	for (int c = 0; c < vd->vdev_children; c++) {
3280
3281		/*
3282		 * Set the vdev id to the first hole, if one exists.
3283		 */
3284		for (id = 0; id < rvd->vdev_children; id++) {
3285			if (rvd->vdev_child[id]->vdev_ishole) {
3286				vdev_free(rvd->vdev_child[id]);
3287				break;
3288			}
3289		}
3290		tvd = vd->vdev_child[c];
3291		vdev_remove_child(vd, tvd);
3292		tvd->vdev_id = id;
3293		vdev_add_child(rvd, tvd);
3294		vdev_config_dirty(tvd);
3295	}
3296
3297	if (nspares != 0) {
3298		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3299		    ZPOOL_CONFIG_SPARES);
3300		spa_load_spares(spa);
3301		spa->spa_spares.sav_sync = B_TRUE;
3302	}
3303
3304	if (nl2cache != 0) {
3305		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3306		    ZPOOL_CONFIG_L2CACHE);
3307		spa_load_l2cache(spa);
3308		spa->spa_l2cache.sav_sync = B_TRUE;
3309	}
3310
3311	/*
3312	 * We have to be careful when adding new vdevs to an existing pool.
3313	 * If other threads start allocating from these vdevs before we
3314	 * sync the config cache, and we lose power, then upon reboot we may
3315	 * fail to open the pool because there are DVAs that the config cache
3316	 * can't translate.  Therefore, we first add the vdevs without
3317	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3318	 * and then let spa_config_update() initialize the new metaslabs.
3319	 *
3320	 * spa_load() checks for added-but-not-initialized vdevs, so that
3321	 * if we lose power at any point in this sequence, the remaining
3322	 * steps will be completed the next time we load the pool.
3323	 */
3324	(void) spa_vdev_exit(spa, vd, txg, 0);
3325
3326	mutex_enter(&spa_namespace_lock);
3327	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3328	mutex_exit(&spa_namespace_lock);
3329
3330	return (0);
3331}
3332
3333/*
3334 * Attach a device to a mirror.  The arguments are the path to any device
3335 * in the mirror, and the nvroot for the new device.  If the path specifies
3336 * a device that is not mirrored, we automatically insert the mirror vdev.
3337 *
3338 * If 'replacing' is specified, the new device is intended to replace the
3339 * existing device; in this case the two devices are made into their own
3340 * mirror using the 'replacing' vdev, which is functionally identical to
3341 * the mirror vdev (it actually reuses all the same ops) but has a few
3342 * extra rules: you can't attach to it after it's been created, and upon
3343 * completion of resilvering, the first disk (the one being replaced)
3344 * is automatically detached.
3345 */
3346int
3347spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3348{
3349	uint64_t txg, open_txg;
3350	vdev_t *rvd = spa->spa_root_vdev;
3351	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3352	vdev_ops_t *pvops;
3353	char *oldvdpath, *newvdpath;
3354	int newvd_isspare;
3355	int error;
3356
3357	txg = spa_vdev_enter(spa);
3358
3359	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3360
3361	if (oldvd == NULL)
3362		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3363
3364	if (!oldvd->vdev_ops->vdev_op_leaf)
3365		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3366
3367	pvd = oldvd->vdev_parent;
3368
3369	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3370	    VDEV_ALLOC_ADD)) != 0)
3371		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3372
3373	if (newrootvd->vdev_children != 1)
3374		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3375
3376	newvd = newrootvd->vdev_child[0];
3377
3378	if (!newvd->vdev_ops->vdev_op_leaf)
3379		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3380
3381	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3382		return (spa_vdev_exit(spa, newrootvd, txg, error));
3383
3384	/*
3385	 * Spares can't replace logs
3386	 */
3387	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3388		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3389
3390	if (!replacing) {
3391		/*
3392		 * For attach, the only allowable parent is a mirror or the root
3393		 * vdev.
3394		 */
3395		if (pvd->vdev_ops != &vdev_mirror_ops &&
3396		    pvd->vdev_ops != &vdev_root_ops)
3397			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3398
3399		pvops = &vdev_mirror_ops;
3400	} else {
3401		/*
3402		 * Active hot spares can only be replaced by inactive hot
3403		 * spares.
3404		 */
3405		if (pvd->vdev_ops == &vdev_spare_ops &&
3406		    pvd->vdev_child[1] == oldvd &&
3407		    !spa_has_spare(spa, newvd->vdev_guid))
3408			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3409
3410		/*
3411		 * If the source is a hot spare, and the parent isn't already a
3412		 * spare, then we want to create a new hot spare.  Otherwise, we
3413		 * want to create a replacing vdev.  The user is not allowed to
3414		 * attach to a spared vdev child unless the 'isspare' state is
3415		 * the same (spare replaces spare, non-spare replaces
3416		 * non-spare).
3417		 */
3418		if (pvd->vdev_ops == &vdev_replacing_ops)
3419			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3420		else if (pvd->vdev_ops == &vdev_spare_ops &&
3421		    newvd->vdev_isspare != oldvd->vdev_isspare)
3422			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3423		else if (pvd->vdev_ops != &vdev_spare_ops &&
3424		    newvd->vdev_isspare)
3425			pvops = &vdev_spare_ops;
3426		else
3427			pvops = &vdev_replacing_ops;
3428	}
3429
3430	/*
3431	 * Make sure the new device is big enough.
3432	 */
3433	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3434		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3435
3436	/*
3437	 * The new device cannot have a higher alignment requirement
3438	 * than the top-level vdev.
3439	 */
3440	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3441		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3442
3443	/*
3444	 * If this is an in-place replacement, update oldvd's path and devid
3445	 * to make it distinguishable from newvd, and unopenable from now on.
3446	 */
3447	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3448		spa_strfree(oldvd->vdev_path);
3449		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3450		    KM_SLEEP);
3451		(void) sprintf(oldvd->vdev_path, "%s/%s",
3452		    newvd->vdev_path, "old");
3453		if (oldvd->vdev_devid != NULL) {
3454			spa_strfree(oldvd->vdev_devid);
3455			oldvd->vdev_devid = NULL;
3456		}
3457	}
3458
3459	/*
3460	 * If the parent is not a mirror, or if we're replacing, insert the new
3461	 * mirror/replacing/spare vdev above oldvd.
3462	 */
3463	if (pvd->vdev_ops != pvops)
3464		pvd = vdev_add_parent(oldvd, pvops);
3465
3466	ASSERT(pvd->vdev_top->vdev_parent == rvd);
3467	ASSERT(pvd->vdev_ops == pvops);
3468	ASSERT(oldvd->vdev_parent == pvd);
3469
3470	/*
3471	 * Extract the new device from its root and add it to pvd.
3472	 */
3473	vdev_remove_child(newrootvd, newvd);
3474	newvd->vdev_id = pvd->vdev_children;
3475	newvd->vdev_crtxg = oldvd->vdev_crtxg;
3476	vdev_add_child(pvd, newvd);
3477
3478	tvd = newvd->vdev_top;
3479	ASSERT(pvd->vdev_top == tvd);
3480	ASSERT(tvd->vdev_parent == rvd);
3481
3482	vdev_config_dirty(tvd);
3483
3484	/*
3485	 * Set newvd's DTL to [TXG_INITIAL, open_txg].  It will propagate
3486	 * upward when spa_vdev_exit() calls vdev_dtl_reassess().
3487	 */
3488	open_txg = txg + TXG_CONCURRENT_STATES - 1;
3489
3490	vdev_dtl_dirty(newvd, DTL_MISSING,
3491	    TXG_INITIAL, open_txg - TXG_INITIAL + 1);
3492
3493	if (newvd->vdev_isspare) {
3494		spa_spare_activate(newvd);
3495		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
3496	}
3497
3498	oldvdpath = spa_strdup(oldvd->vdev_path);
3499	newvdpath = spa_strdup(newvd->vdev_path);
3500	newvd_isspare = newvd->vdev_isspare;
3501
3502	/*
3503	 * Mark newvd's DTL dirty in this txg.
3504	 */
3505	vdev_dirty(tvd, VDD_DTL, newvd, txg);
3506
3507	(void) spa_vdev_exit(spa, newrootvd, open_txg, 0);
3508
3509	spa_history_internal_log(LOG_POOL_VDEV_ATTACH, spa, NULL,
3510	    CRED(),  "%s vdev=%s %s vdev=%s",
3511	    replacing && newvd_isspare ? "spare in" :
3512	    replacing ? "replace" : "attach", newvdpath,
3513	    replacing ? "for" : "to", oldvdpath);
3514
3515	spa_strfree(oldvdpath);
3516	spa_strfree(newvdpath);
3517
3518	/*
3519	 * Kick off a resilver to update newvd.
3520	 */
3521	VERIFY3U(spa_scrub(spa, POOL_SCRUB_RESILVER), ==, 0);
3522
3523	return (0);
3524}
3525
3526/*
3527 * Detach a device from a mirror or replacing vdev.
3528 * If 'replace_done' is specified, only detach if the parent
3529 * is a replacing vdev.
3530 */
3531int
3532spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
3533{
3534	uint64_t txg;
3535	int error;
3536	vdev_t *rvd = spa->spa_root_vdev;
3537	vdev_t *vd, *pvd, *cvd, *tvd;
3538	boolean_t unspare = B_FALSE;
3539	uint64_t unspare_guid;
3540	size_t len;
3541
3542	txg = spa_vdev_enter(spa);
3543
3544	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3545
3546	if (vd == NULL)
3547		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3548
3549	if (!vd->vdev_ops->vdev_op_leaf)
3550		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3551
3552	pvd = vd->vdev_parent;
3553
3554	/*
3555	 * If the parent/child relationship is not as expected, don't do it.
3556	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
3557	 * vdev that's replacing B with C.  The user's intent in replacing
3558	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
3559	 * the replace by detaching C, the expected behavior is to end up
3560	 * M(A,B).  But suppose that right after deciding to detach C,
3561	 * the replacement of B completes.  We would have M(A,C), and then
3562	 * ask to detach C, which would leave us with just A -- not what
3563	 * the user wanted.  To prevent this, we make sure that the
3564	 * parent/child relationship hasn't changed -- in this example,
3565	 * that C's parent is still the replacing vdev R.
3566	 */
3567	if (pvd->vdev_guid != pguid && pguid != 0)
3568		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3569
3570	/*
3571	 * If replace_done is specified, only remove this device if it's
3572	 * the first child of a replacing vdev.  For the 'spare' vdev, either
3573	 * disk can be removed.
3574	 */
3575	if (replace_done) {
3576		if (pvd->vdev_ops == &vdev_replacing_ops) {
3577			if (vd->vdev_id != 0)
3578				return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3579		} else if (pvd->vdev_ops != &vdev_spare_ops) {
3580			return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3581		}
3582	}
3583
3584	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
3585	    spa_version(spa) >= SPA_VERSION_SPARES);
3586
3587	/*
3588	 * Only mirror, replacing, and spare vdevs support detach.
3589	 */
3590	if (pvd->vdev_ops != &vdev_replacing_ops &&
3591	    pvd->vdev_ops != &vdev_mirror_ops &&
3592	    pvd->vdev_ops != &vdev_spare_ops)
3593		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3594
3595	/*
3596	 * If this device has the only valid copy of some data,
3597	 * we cannot safely detach it.
3598	 */
3599	if (vdev_dtl_required(vd))
3600		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3601
3602	ASSERT(pvd->vdev_children >= 2);
3603
3604	/*
3605	 * If we are detaching the second disk from a replacing vdev, then
3606	 * check to see if we changed the original vdev's path to have "/old"
3607	 * at the end in spa_vdev_attach().  If so, undo that change now.
3608	 */
3609	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id == 1 &&
3610	    pvd->vdev_child[0]->vdev_path != NULL &&
3611	    pvd->vdev_child[1]->vdev_path != NULL) {
3612		ASSERT(pvd->vdev_child[1] == vd);
3613		cvd = pvd->vdev_child[0];
3614		len = strlen(vd->vdev_path);
3615		if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
3616		    strcmp(cvd->vdev_path + len, "/old") == 0) {
3617			spa_strfree(cvd->vdev_path);
3618			cvd->vdev_path = spa_strdup(vd->vdev_path);
3619		}
3620	}
3621
3622	/*
3623	 * If we are detaching the original disk from a spare, then it implies
3624	 * that the spare should become a real disk, and be removed from the
3625	 * active spare list for the pool.
3626	 */
3627	if (pvd->vdev_ops == &vdev_spare_ops &&
3628	    vd->vdev_id == 0 && pvd->vdev_child[1]->vdev_isspare)
3629		unspare = B_TRUE;
3630
3631	/*
3632	 * Erase the disk labels so the disk can be used for other things.
3633	 * This must be done after all other error cases are handled,
3634	 * but before we disembowel vd (so we can still do I/O to it).
3635	 * But if we can't do it, don't treat the error as fatal --
3636	 * it may be that the unwritability of the disk is the reason
3637	 * it's being detached!
3638	 */
3639	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
3640
3641	/*
3642	 * Remove vd from its parent and compact the parent's children.
3643	 */
3644	vdev_remove_child(pvd, vd);
3645	vdev_compact_children(pvd);
3646
3647	/*
3648	 * Remember one of the remaining children so we can get tvd below.
3649	 */
3650	cvd = pvd->vdev_child[0];
3651
3652	/*
3653	 * If we need to remove the remaining child from the list of hot spares,
3654	 * do it now, marking the vdev as no longer a spare in the process.
3655	 * We must do this before vdev_remove_parent(), because that can
3656	 * change the GUID if it creates a new toplevel GUID.  For a similar
3657	 * reason, we must remove the spare now, in the same txg as the detach;
3658	 * otherwise someone could attach a new sibling, change the GUID, and
3659	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
3660	 */
3661	if (unspare) {
3662		ASSERT(cvd->vdev_isspare);
3663		spa_spare_remove(cvd);
3664		unspare_guid = cvd->vdev_guid;
3665		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
3666	}
3667
3668	/*
3669	 * If the parent mirror/replacing vdev only has one child,
3670	 * the parent is no longer needed.  Remove it from the tree.
3671	 */
3672	if (pvd->vdev_children == 1)
3673		vdev_remove_parent(cvd);
3674
3675	/*
3676	 * We don't set tvd until now because the parent we just removed
3677	 * may have been the previous top-level vdev.
3678	 */
3679	tvd = cvd->vdev_top;
3680	ASSERT(tvd->vdev_parent == rvd);
3681
3682	/*
3683	 * Reevaluate the parent vdev state.
3684	 */
3685	vdev_propagate_state(cvd);
3686
3687	/*
3688	 * If the 'autoexpand' property is set on the pool then automatically
3689	 * try to expand the size of the pool. For example if the device we
3690	 * just detached was smaller than the others, it may be possible to
3691	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
3692	 * first so that we can obtain the updated sizes of the leaf vdevs.
3693	 */
3694	if (spa->spa_autoexpand) {
3695		vdev_reopen(tvd);
3696		vdev_expand(tvd, txg);
3697	}
3698
3699	vdev_config_dirty(tvd);
3700
3701	/*
3702	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
3703	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
3704	 * But first make sure we're not on any *other* txg's DTL list, to
3705	 * prevent vd from being accessed after it's freed.
3706	 */
3707	for (int t = 0; t < TXG_SIZE; t++)
3708		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
3709	vd->vdev_detached = B_TRUE;
3710	vdev_dirty(tvd, VDD_DTL, vd, txg);
3711
3712	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
3713
3714	error = spa_vdev_exit(spa, vd, txg, 0);
3715
3716	/*
3717	 * If this was the removal of the original device in a hot spare vdev,
3718	 * then we want to go through and remove the device from the hot spare
3719	 * list of every other pool.
3720	 */
3721	if (unspare) {
3722		spa_t *myspa = spa;
3723		spa = NULL;
3724		mutex_enter(&spa_namespace_lock);
3725		while ((spa = spa_next(spa)) != NULL) {
3726			if (spa->spa_state != POOL_STATE_ACTIVE)
3727				continue;
3728			if (spa == myspa)
3729				continue;
3730			spa_open_ref(spa, FTAG);
3731			mutex_exit(&spa_namespace_lock);
3732			(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
3733			mutex_enter(&spa_namespace_lock);
3734			spa_close(spa, FTAG);
3735		}
3736		mutex_exit(&spa_namespace_lock);
3737	}
3738
3739	return (error);
3740}
3741
3742static nvlist_t *
3743spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
3744{
3745	for (int i = 0; i < count; i++) {
3746		uint64_t guid;
3747
3748		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
3749		    &guid) == 0);
3750
3751		if (guid == target_guid)
3752			return (nvpp[i]);
3753	}
3754
3755	return (NULL);
3756}
3757
3758static void
3759spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
3760	nvlist_t *dev_to_remove)
3761{
3762	nvlist_t **newdev = NULL;
3763
3764	if (count > 1)
3765		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
3766
3767	for (int i = 0, j = 0; i < count; i++) {
3768		if (dev[i] == dev_to_remove)
3769			continue;
3770		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
3771	}
3772
3773	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
3774	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
3775
3776	for (int i = 0; i < count - 1; i++)
3777		nvlist_free(newdev[i]);
3778
3779	if (count > 1)
3780		kmem_free(newdev, (count - 1) * sizeof (void *));
3781}
3782
3783/*
3784 * Removing a device from the vdev namespace requires several steps
3785 * and can take a significant amount of time.  As a result we use
3786 * the spa_vdev_config_[enter/exit] functions which allow us to
3787 * grab and release the spa_config_lock while still holding the namespace
3788 * lock.  During each step the configuration is synced out.
3789 */
3790
3791/*
3792 * Evacuate the device.
3793 */
3794int
3795spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
3796{
3797	int error = 0;
3798	uint64_t txg;
3799
3800	ASSERT(MUTEX_HELD(&spa_namespace_lock));
3801	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
3802	ASSERT(vd == vd->vdev_top);
3803
3804	/*
3805	 * Evacuate the device.  We don't hold the config lock as writer
3806	 * since we need to do I/O but we do keep the
3807	 * spa_namespace_lock held.  Once this completes the device
3808	 * should no longer have any blocks allocated on it.
3809	 */
3810	if (vd->vdev_islog) {
3811		error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
3812		    NULL, DS_FIND_CHILDREN);
3813	} else {
3814		error = ENOTSUP;	/* until we have bp rewrite */
3815	}
3816
3817	txg_wait_synced(spa_get_dsl(spa), 0);
3818
3819	if (error)
3820		return (error);
3821
3822	/*
3823	 * The evacuation succeeded.  Remove any remaining MOS metadata
3824	 * associated with this vdev, and wait for these changes to sync.
3825	 */
3826	txg = spa_vdev_config_enter(spa);
3827	vd->vdev_removing = B_TRUE;
3828	vdev_dirty(vd, 0, NULL, txg);
3829	vdev_config_dirty(vd);
3830	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
3831
3832	return (0);
3833}
3834
3835/*
3836 * Complete the removal by cleaning up the namespace.
3837 */
3838void
3839spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
3840{
3841	vdev_t *rvd = spa->spa_root_vdev;
3842	uint64_t id = vd->vdev_id;
3843	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
3844
3845	ASSERT(MUTEX_HELD(&spa_namespace_lock));
3846	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3847	ASSERT(vd == vd->vdev_top);
3848
3849	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
3850
3851	if (list_link_active(&vd->vdev_state_dirty_node))
3852		vdev_state_clean(vd);
3853	if (list_link_active(&vd->vdev_config_dirty_node))
3854		vdev_config_clean(vd);
3855
3856	vdev_free(vd);
3857
3858	if (last_vdev) {
3859		vdev_compact_children(rvd);
3860	} else {
3861		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
3862		vdev_add_child(rvd, vd);
3863	}
3864	vdev_config_dirty(rvd);
3865
3866	/*
3867	 * Reassess the health of our root vdev.
3868	 */
3869	vdev_reopen(rvd);
3870}
3871
3872/*
3873 * Remove a device from the pool.  Currently, this supports removing only hot
3874 * spares, slogs, and level 2 ARC devices.
3875 */
3876int
3877spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
3878{
3879	vdev_t *vd;
3880	metaslab_group_t *mg;
3881	nvlist_t **spares, **l2cache, *nv;
3882	uint64_t txg = 0;
3883	uint_t nspares, nl2cache;
3884	int error = 0;
3885	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
3886
3887	if (!locked)
3888		txg = spa_vdev_enter(spa);
3889
3890	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3891
3892	if (spa->spa_spares.sav_vdevs != NULL &&
3893	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3894	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
3895	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
3896		/*
3897		 * Only remove the hot spare if it's not currently in use
3898		 * in this pool.
3899		 */
3900		if (vd == NULL || unspare) {
3901			spa_vdev_remove_aux(spa->spa_spares.sav_config,
3902			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
3903			spa_load_spares(spa);
3904			spa->spa_spares.sav_sync = B_TRUE;
3905		} else {
3906			error = EBUSY;
3907		}
3908	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
3909	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3910	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
3911	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
3912		/*
3913		 * Cache devices can always be removed.
3914		 */
3915		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
3916		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
3917		spa_load_l2cache(spa);
3918		spa->spa_l2cache.sav_sync = B_TRUE;
3919	} else if (vd != NULL && vd->vdev_islog) {
3920		ASSERT(!locked);
3921		ASSERT(vd == vd->vdev_top);
3922
3923		/*
3924		 * XXX - Once we have bp-rewrite this should
3925		 * become the common case.
3926		 */
3927
3928		mg = vd->vdev_mg;
3929
3930		/*
3931		 * Stop allocating from this vdev.
3932		 */
3933		metaslab_group_passivate(mg);
3934
3935		/*
3936		 * Wait for the youngest allocations and frees to sync,
3937		 * and then wait for the deferral of those frees to finish.
3938		 */
3939		spa_vdev_config_exit(spa, NULL,
3940		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
3941
3942		/*
3943		 * Attempt to evacuate the vdev.
3944		 */
3945		error = spa_vdev_remove_evacuate(spa, vd);
3946
3947		txg = spa_vdev_config_enter(spa);
3948
3949		/*
3950		 * If we couldn't evacuate the vdev, unwind.
3951		 */
3952		if (error) {
3953			metaslab_group_activate(mg);
3954			return (spa_vdev_exit(spa, NULL, txg, error));
3955		}
3956
3957		/*
3958		 * Clean up the vdev namespace.
3959		 */
3960		spa_vdev_remove_from_namespace(spa, vd);
3961
3962	} else if (vd != NULL) {
3963		/*
3964		 * Normal vdevs cannot be removed (yet).
3965		 */
3966		error = ENOTSUP;
3967	} else {
3968		/*
3969		 * There is no vdev of any kind with the specified guid.
3970		 */
3971		error = ENOENT;
3972	}
3973
3974	if (!locked)
3975		return (spa_vdev_exit(spa, NULL, txg, error));
3976
3977	return (error);
3978}
3979
3980/*
3981 * Find any device that's done replacing, or a vdev marked 'unspare' that's
3982 * current spared, so we can detach it.
3983 */
3984static vdev_t *
3985spa_vdev_resilver_done_hunt(vdev_t *vd)
3986{
3987	vdev_t *newvd, *oldvd;
3988
3989	for (int c = 0; c < vd->vdev_children; c++) {
3990		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
3991		if (oldvd != NULL)
3992			return (oldvd);
3993	}
3994
3995	/*
3996	 * Check for a completed replacement.
3997	 */
3998	if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
3999		oldvd = vd->vdev_child[0];
4000		newvd = vd->vdev_child[1];
4001
4002		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4003		    !vdev_dtl_required(oldvd))
4004			return (oldvd);
4005	}
4006
4007	/*
4008	 * Check for a completed resilver with the 'unspare' flag set.
4009	 */
4010	if (vd->vdev_ops == &vdev_spare_ops && vd->vdev_children == 2) {
4011		newvd = vd->vdev_child[0];
4012		oldvd = vd->vdev_child[1];
4013
4014		if (newvd->vdev_unspare &&
4015		    vdev_dtl_empty(newvd, DTL_MISSING) &&
4016		    !vdev_dtl_required(oldvd)) {
4017			newvd->vdev_unspare = 0;
4018			return (oldvd);
4019		}
4020	}
4021
4022	return (NULL);
4023}
4024
4025static void
4026spa_vdev_resilver_done(spa_t *spa)
4027{
4028	vdev_t *vd, *pvd, *ppvd;
4029	uint64_t guid, sguid, pguid, ppguid;
4030
4031	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4032
4033	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4034		pvd = vd->vdev_parent;
4035		ppvd = pvd->vdev_parent;
4036		guid = vd->vdev_guid;
4037		pguid = pvd->vdev_guid;
4038		ppguid = ppvd->vdev_guid;
4039		sguid = 0;
4040		/*
4041		 * If we have just finished replacing a hot spared device, then
4042		 * we need to detach the parent's first child (the original hot
4043		 * spare) as well.
4044		 */
4045		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0) {
4046			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4047			ASSERT(ppvd->vdev_children == 2);
4048			sguid = ppvd->vdev_child[1]->vdev_guid;
4049		}
4050		spa_config_exit(spa, SCL_ALL, FTAG);
4051		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4052			return;
4053		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4054			return;
4055		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4056	}
4057
4058	spa_config_exit(spa, SCL_ALL, FTAG);
4059}
4060
4061/*
4062 * Update the stored path or FRU for this vdev.  Dirty the vdev configuration,
4063 * relying on spa_vdev_enter/exit() to synchronize the labels and cache.
4064 */
4065int
4066spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4067    boolean_t ispath)
4068{
4069	vdev_t *vd;
4070	uint64_t txg;
4071
4072	txg = spa_vdev_enter(spa);
4073
4074	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4075		return (spa_vdev_exit(spa, NULL, txg, ENOENT));
4076
4077	if (!vd->vdev_ops->vdev_op_leaf)
4078		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4079
4080	if (ispath) {
4081		spa_strfree(vd->vdev_path);
4082		vd->vdev_path = spa_strdup(value);
4083	} else {
4084		if (vd->vdev_fru != NULL)
4085			spa_strfree(vd->vdev_fru);
4086		vd->vdev_fru = spa_strdup(value);
4087	}
4088
4089	vdev_config_dirty(vd->vdev_top);
4090
4091	return (spa_vdev_exit(spa, NULL, txg, 0));
4092}
4093
4094int
4095spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
4096{
4097	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
4098}
4099
4100int
4101spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
4102{
4103	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
4104}
4105
4106/*
4107 * ==========================================================================
4108 * SPA Scrubbing
4109 * ==========================================================================
4110 */
4111
4112int
4113spa_scrub(spa_t *spa, pool_scrub_type_t type)
4114{
4115	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4116
4117	if ((uint_t)type >= POOL_SCRUB_TYPES)
4118		return (ENOTSUP);
4119
4120	/*
4121	 * If a resilver was requested, but there is no DTL on a
4122	 * writeable leaf device, we have nothing to do.
4123	 */
4124	if (type == POOL_SCRUB_RESILVER &&
4125	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
4126		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
4127		return (0);
4128	}
4129
4130	if (type == POOL_SCRUB_EVERYTHING &&
4131	    spa->spa_dsl_pool->dp_scrub_func != SCRUB_FUNC_NONE &&
4132	    spa->spa_dsl_pool->dp_scrub_isresilver)
4133		return (EBUSY);
4134
4135	if (type == POOL_SCRUB_EVERYTHING || type == POOL_SCRUB_RESILVER) {
4136		return (dsl_pool_scrub_clean(spa->spa_dsl_pool));
4137	} else if (type == POOL_SCRUB_NONE) {
4138		return (dsl_pool_scrub_cancel(spa->spa_dsl_pool));
4139	} else {
4140		return (EINVAL);
4141	}
4142}
4143
4144/*
4145 * ==========================================================================
4146 * SPA async task processing
4147 * ==========================================================================
4148 */
4149
4150static void
4151spa_async_remove(spa_t *spa, vdev_t *vd)
4152{
4153	if (vd->vdev_remove_wanted) {
4154		vd->vdev_remove_wanted = 0;
4155		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
4156
4157		/*
4158		 * We want to clear the stats, but we don't want to do a full
4159		 * vdev_clear() as that will cause us to throw away
4160		 * degraded/faulted state as well as attempt to reopen the
4161		 * device, all of which is a waste.
4162		 */
4163		vd->vdev_stat.vs_read_errors = 0;
4164		vd->vdev_stat.vs_write_errors = 0;
4165		vd->vdev_stat.vs_checksum_errors = 0;
4166
4167		vdev_state_dirty(vd->vdev_top);
4168	}
4169
4170	for (int c = 0; c < vd->vdev_children; c++)
4171		spa_async_remove(spa, vd->vdev_child[c]);
4172}
4173
4174static void
4175spa_async_probe(spa_t *spa, vdev_t *vd)
4176{
4177	if (vd->vdev_probe_wanted) {
4178		vd->vdev_probe_wanted = 0;
4179		vdev_reopen(vd);	/* vdev_open() does the actual probe */
4180	}
4181
4182	for (int c = 0; c < vd->vdev_children; c++)
4183		spa_async_probe(spa, vd->vdev_child[c]);
4184}
4185
4186static void
4187spa_async_autoexpand(spa_t *spa, vdev_t *vd)
4188{
4189	sysevent_id_t eid;
4190	nvlist_t *attr;
4191	char *physpath;
4192
4193	if (!spa->spa_autoexpand)
4194		return;
4195
4196	for (int c = 0; c < vd->vdev_children; c++) {
4197		vdev_t *cvd = vd->vdev_child[c];
4198		spa_async_autoexpand(spa, cvd);
4199	}
4200
4201	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
4202		return;
4203
4204	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
4205	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
4206
4207	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4208	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
4209
4210	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
4211	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
4212
4213	nvlist_free(attr);
4214	kmem_free(physpath, MAXPATHLEN);
4215}
4216
4217static void
4218spa_async_thread(spa_t *spa)
4219{
4220	int tasks;
4221
4222	ASSERT(spa->spa_sync_on);
4223
4224	mutex_enter(&spa->spa_async_lock);
4225	tasks = spa->spa_async_tasks;
4226	spa->spa_async_tasks = 0;
4227	mutex_exit(&spa->spa_async_lock);
4228
4229	/*
4230	 * See if the config needs to be updated.
4231	 */
4232	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
4233		uint64_t old_space, new_space;
4234
4235		mutex_enter(&spa_namespace_lock);
4236		old_space = metaslab_class_get_space(spa_normal_class(spa));
4237		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4238		new_space = metaslab_class_get_space(spa_normal_class(spa));
4239		mutex_exit(&spa_namespace_lock);
4240
4241		/*
4242		 * If the pool grew as a result of the config update,
4243		 * then log an internal history event.
4244		 */
4245		if (new_space != old_space) {
4246			spa_history_internal_log(LOG_POOL_VDEV_ONLINE,
4247			    spa, NULL, CRED(),
4248			    "pool '%s' size: %llu(+%llu)",
4249			    spa_name(spa), new_space, new_space - old_space);
4250		}
4251	}
4252
4253	/*
4254	 * See if any devices need to be marked REMOVED.
4255	 */
4256	if (tasks & SPA_ASYNC_REMOVE) {
4257		spa_vdev_state_enter(spa, SCL_NONE);
4258		spa_async_remove(spa, spa->spa_root_vdev);
4259		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
4260			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
4261		for (int i = 0; i < spa->spa_spares.sav_count; i++)
4262			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
4263		(void) spa_vdev_state_exit(spa, NULL, 0);
4264	}
4265
4266	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
4267		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4268		spa_async_autoexpand(spa, spa->spa_root_vdev);
4269		spa_config_exit(spa, SCL_CONFIG, FTAG);
4270	}
4271
4272	/*
4273	 * See if any devices need to be probed.
4274	 */
4275	if (tasks & SPA_ASYNC_PROBE) {
4276		spa_vdev_state_enter(spa, SCL_NONE);
4277		spa_async_probe(spa, spa->spa_root_vdev);
4278		(void) spa_vdev_state_exit(spa, NULL, 0);
4279	}
4280
4281	/*
4282	 * If any devices are done replacing, detach them.
4283	 */
4284	if (tasks & SPA_ASYNC_RESILVER_DONE)
4285		spa_vdev_resilver_done(spa);
4286
4287	/*
4288	 * Kick off a resilver.
4289	 */
4290	if (tasks & SPA_ASYNC_RESILVER)
4291		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER) == 0);
4292
4293	/*
4294	 * Let the world know that we're done.
4295	 */
4296	mutex_enter(&spa->spa_async_lock);
4297	spa->spa_async_thread = NULL;
4298	cv_broadcast(&spa->spa_async_cv);
4299	mutex_exit(&spa->spa_async_lock);
4300	thread_exit();
4301}
4302
4303void
4304spa_async_suspend(spa_t *spa)
4305{
4306	mutex_enter(&spa->spa_async_lock);
4307	spa->spa_async_suspended++;
4308	while (spa->spa_async_thread != NULL)
4309		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
4310	mutex_exit(&spa->spa_async_lock);
4311}
4312
4313void
4314spa_async_resume(spa_t *spa)
4315{
4316	mutex_enter(&spa->spa_async_lock);
4317	ASSERT(spa->spa_async_suspended != 0);
4318	spa->spa_async_suspended--;
4319	mutex_exit(&spa->spa_async_lock);
4320}
4321
4322static void
4323spa_async_dispatch(spa_t *spa)
4324{
4325	mutex_enter(&spa->spa_async_lock);
4326	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
4327	    spa->spa_async_thread == NULL &&
4328	    rootdir != NULL && !vn_is_readonly(rootdir))
4329		spa->spa_async_thread = thread_create(NULL, 0,
4330		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
4331	mutex_exit(&spa->spa_async_lock);
4332}
4333
4334void
4335spa_async_request(spa_t *spa, int task)
4336{
4337	mutex_enter(&spa->spa_async_lock);
4338	spa->spa_async_tasks |= task;
4339	mutex_exit(&spa->spa_async_lock);
4340}
4341
4342/*
4343 * ==========================================================================
4344 * SPA syncing routines
4345 * ==========================================================================
4346 */
4347static void
4348spa_sync_deferred_bplist(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx, uint64_t txg)
4349{
4350	blkptr_t blk;
4351	uint64_t itor = 0;
4352	uint8_t c = 1;
4353
4354	while (bplist_iterate(bpl, &itor, &blk) == 0) {
4355		ASSERT(blk.blk_birth < txg);
4356		zio_free(spa, txg, &blk);
4357	}
4358
4359	bplist_vacate(bpl, tx);
4360
4361	/*
4362	 * Pre-dirty the first block so we sync to convergence faster.
4363	 * (Usually only the first block is needed.)
4364	 */
4365	dmu_write(bpl->bpl_mos, spa->spa_deferred_bplist_obj, 0, 1, &c, tx);
4366}
4367
4368static void
4369spa_sync_free(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
4370{
4371	zio_t *zio = arg;
4372
4373	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
4374	    zio->io_flags));
4375}
4376
4377static void
4378spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
4379{
4380	char *packed = NULL;
4381	size_t bufsize;
4382	size_t nvsize = 0;
4383	dmu_buf_t *db;
4384
4385	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
4386
4387	/*
4388	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
4389	 * information.  This avoids the dbuf_will_dirty() path and
4390	 * saves us a pre-read to get data we don't actually care about.
4391	 */
4392	bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
4393	packed = kmem_alloc(bufsize, KM_SLEEP);
4394
4395	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
4396	    KM_SLEEP) == 0);
4397	bzero(packed + nvsize, bufsize - nvsize);
4398
4399	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
4400
4401	kmem_free(packed, bufsize);
4402
4403	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
4404	dmu_buf_will_dirty(db, tx);
4405	*(uint64_t *)db->db_data = nvsize;
4406	dmu_buf_rele(db, FTAG);
4407}
4408
4409static void
4410spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
4411    const char *config, const char *entry)
4412{
4413	nvlist_t *nvroot;
4414	nvlist_t **list;
4415	int i;
4416
4417	if (!sav->sav_sync)
4418		return;
4419
4420	/*
4421	 * Update the MOS nvlist describing the list of available devices.
4422	 * spa_validate_aux() will have already made sure this nvlist is
4423	 * valid and the vdevs are labeled appropriately.
4424	 */
4425	if (sav->sav_object == 0) {
4426		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
4427		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
4428		    sizeof (uint64_t), tx);
4429		VERIFY(zap_update(spa->spa_meta_objset,
4430		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
4431		    &sav->sav_object, tx) == 0);
4432	}
4433
4434	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4435	if (sav->sav_count == 0) {
4436		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
4437	} else {
4438		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
4439		for (i = 0; i < sav->sav_count; i++)
4440			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
4441			    B_FALSE, B_FALSE, B_TRUE);
4442		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
4443		    sav->sav_count) == 0);
4444		for (i = 0; i < sav->sav_count; i++)
4445			nvlist_free(list[i]);
4446		kmem_free(list, sav->sav_count * sizeof (void *));
4447	}
4448
4449	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
4450	nvlist_free(nvroot);
4451
4452	sav->sav_sync = B_FALSE;
4453}
4454
4455static void
4456spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
4457{
4458	nvlist_t *config;
4459
4460	if (list_is_empty(&spa->spa_config_dirty_list))
4461		return;
4462
4463	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4464
4465	config = spa_config_generate(spa, spa->spa_root_vdev,
4466	    dmu_tx_get_txg(tx), B_FALSE);
4467
4468	spa_config_exit(spa, SCL_STATE, FTAG);
4469
4470	if (spa->spa_config_syncing)
4471		nvlist_free(spa->spa_config_syncing);
4472	spa->spa_config_syncing = config;
4473
4474	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
4475}
4476
4477/*
4478 * Set zpool properties.
4479 */
4480static void
4481spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
4482{
4483	spa_t *spa = arg1;
4484	objset_t *mos = spa->spa_meta_objset;
4485	nvlist_t *nvp = arg2;
4486	nvpair_t *elem;
4487	uint64_t intval;
4488	char *strval;
4489	zpool_prop_t prop;
4490	const char *propname;
4491	zprop_type_t proptype;
4492
4493	mutex_enter(&spa->spa_props_lock);
4494
4495	elem = NULL;
4496	while ((elem = nvlist_next_nvpair(nvp, elem))) {
4497		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
4498		case ZPOOL_PROP_VERSION:
4499			/*
4500			 * Only set version for non-zpool-creation cases
4501			 * (set/import). spa_create() needs special care
4502			 * for version setting.
4503			 */
4504			if (tx->tx_txg != TXG_INITIAL) {
4505				VERIFY(nvpair_value_uint64(elem,
4506				    &intval) == 0);
4507				ASSERT(intval <= SPA_VERSION);
4508				ASSERT(intval >= spa_version(spa));
4509				spa->spa_uberblock.ub_version = intval;
4510				vdev_config_dirty(spa->spa_root_vdev);
4511			}
4512			break;
4513
4514		case ZPOOL_PROP_ALTROOT:
4515			/*
4516			 * 'altroot' is a non-persistent property. It should
4517			 * have been set temporarily at creation or import time.
4518			 */
4519			ASSERT(spa->spa_root != NULL);
4520			break;
4521
4522		case ZPOOL_PROP_CACHEFILE:
4523			/*
4524			 * 'cachefile' is also a non-persisitent property.
4525			 */
4526			break;
4527		default:
4528			/*
4529			 * Set pool property values in the poolprops mos object.
4530			 */
4531			if (spa->spa_pool_props_object == 0) {
4532				VERIFY((spa->spa_pool_props_object =
4533				    zap_create(mos, DMU_OT_POOL_PROPS,
4534				    DMU_OT_NONE, 0, tx)) > 0);
4535
4536				VERIFY(zap_update(mos,
4537				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
4538				    8, 1, &spa->spa_pool_props_object, tx)
4539				    == 0);
4540			}
4541
4542			/* normalize the property name */
4543			propname = zpool_prop_to_name(prop);
4544			proptype = zpool_prop_get_type(prop);
4545
4546			if (nvpair_type(elem) == DATA_TYPE_STRING) {
4547				ASSERT(proptype == PROP_TYPE_STRING);
4548				VERIFY(nvpair_value_string(elem, &strval) == 0);
4549				VERIFY(zap_update(mos,
4550				    spa->spa_pool_props_object, propname,
4551				    1, strlen(strval) + 1, strval, tx) == 0);
4552
4553			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
4554				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
4555
4556				if (proptype == PROP_TYPE_INDEX) {
4557					const char *unused;
4558					VERIFY(zpool_prop_index_to_string(
4559					    prop, intval, &unused) == 0);
4560				}
4561				VERIFY(zap_update(mos,
4562				    spa->spa_pool_props_object, propname,
4563				    8, 1, &intval, tx) == 0);
4564			} else {
4565				ASSERT(0); /* not allowed */
4566			}
4567
4568			switch (prop) {
4569			case ZPOOL_PROP_DELEGATION:
4570				spa->spa_delegation = intval;
4571				break;
4572			case ZPOOL_PROP_BOOTFS:
4573				spa->spa_bootfs = intval;
4574				break;
4575			case ZPOOL_PROP_FAILUREMODE:
4576				spa->spa_failmode = intval;
4577				break;
4578			case ZPOOL_PROP_AUTOEXPAND:
4579				spa->spa_autoexpand = intval;
4580				spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4581				break;
4582			case ZPOOL_PROP_DEDUPDITTO:
4583				spa->spa_dedup_ditto = intval;
4584				break;
4585			default:
4586				break;
4587			}
4588		}
4589
4590		/* log internal history if this is not a zpool create */
4591		if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
4592		    tx->tx_txg != TXG_INITIAL) {
4593			spa_history_internal_log(LOG_POOL_PROPSET,
4594			    spa, tx, cr, "%s %lld %s",
4595			    nvpair_name(elem), intval, spa_name(spa));
4596		}
4597	}
4598
4599	mutex_exit(&spa->spa_props_lock);
4600}
4601
4602/*
4603 * Sync the specified transaction group.  New blocks may be dirtied as
4604 * part of the process, so we iterate until it converges.
4605 */
4606void
4607spa_sync(spa_t *spa, uint64_t txg)
4608{
4609	dsl_pool_t *dp = spa->spa_dsl_pool;
4610	objset_t *mos = spa->spa_meta_objset;
4611	bplist_t *defer_bpl = &spa->spa_deferred_bplist;
4612	bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
4613	vdev_t *rvd = spa->spa_root_vdev;
4614	vdev_t *vd;
4615	dmu_tx_t *tx;
4616	int error;
4617
4618	/*
4619	 * Lock out configuration changes.
4620	 */
4621	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4622
4623	spa->spa_syncing_txg = txg;
4624	spa->spa_sync_pass = 0;
4625
4626	/*
4627	 * If there are any pending vdev state changes, convert them
4628	 * into config changes that go out with this transaction group.
4629	 */
4630	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4631	while (list_head(&spa->spa_state_dirty_list) != NULL) {
4632		/*
4633		 * We need the write lock here because, for aux vdevs,
4634		 * calling vdev_config_dirty() modifies sav_config.
4635		 * This is ugly and will become unnecessary when we
4636		 * eliminate the aux vdev wart by integrating all vdevs
4637		 * into the root vdev tree.
4638		 */
4639		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
4640		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
4641		while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
4642			vdev_state_clean(vd);
4643			vdev_config_dirty(vd);
4644		}
4645		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
4646		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
4647	}
4648	spa_config_exit(spa, SCL_STATE, FTAG);
4649
4650	VERIFY(0 == bplist_open(defer_bpl, mos, spa->spa_deferred_bplist_obj));
4651
4652	tx = dmu_tx_create_assigned(dp, txg);
4653
4654	/*
4655	 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
4656	 * set spa_deflate if we have no raid-z vdevs.
4657	 */
4658	if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
4659	    spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
4660		int i;
4661
4662		for (i = 0; i < rvd->vdev_children; i++) {
4663			vd = rvd->vdev_child[i];
4664			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
4665				break;
4666		}
4667		if (i == rvd->vdev_children) {
4668			spa->spa_deflate = TRUE;
4669			VERIFY(0 == zap_add(spa->spa_meta_objset,
4670			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
4671			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
4672		}
4673	}
4674
4675	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
4676	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
4677		dsl_pool_create_origin(dp, tx);
4678
4679		/* Keeping the origin open increases spa_minref */
4680		spa->spa_minref += 3;
4681	}
4682
4683	if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
4684	    spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
4685		dsl_pool_upgrade_clones(dp, tx);
4686	}
4687
4688	/*
4689	 * If anything has changed in this txg, push the deferred frees
4690	 * from the previous txg.  If not, leave them alone so that we
4691	 * don't generate work on an otherwise idle system.
4692	 */
4693	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
4694	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
4695	    !txg_list_empty(&dp->dp_sync_tasks, txg))
4696		spa_sync_deferred_bplist(spa, defer_bpl, tx, txg);
4697
4698	/*
4699	 * Iterate to convergence.
4700	 */
4701	do {
4702		int pass = ++spa->spa_sync_pass;
4703
4704		spa_sync_config_object(spa, tx);
4705		spa_sync_aux_dev(spa, &spa->spa_spares, tx,
4706		    ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
4707		spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
4708		    ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
4709		spa_errlog_sync(spa, txg);
4710		dsl_pool_sync(dp, txg);
4711
4712		if (pass <= SYNC_PASS_DEFERRED_FREE) {
4713			zio_t *zio = zio_root(spa, NULL, NULL, 0);
4714			bplist_sync(free_bpl, spa_sync_free, zio, tx);
4715			VERIFY(zio_wait(zio) == 0);
4716		} else {
4717			bplist_sync(free_bpl, bplist_enqueue_cb, defer_bpl, tx);
4718		}
4719
4720		ddt_sync(spa, txg);
4721
4722		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
4723			vdev_sync(vd, txg);
4724
4725	} while (dmu_objset_is_dirty(mos, txg));
4726
4727	ASSERT(free_bpl->bpl_queue == NULL);
4728
4729	bplist_close(defer_bpl);
4730
4731	/*
4732	 * Rewrite the vdev configuration (which includes the uberblock)
4733	 * to commit the transaction group.
4734	 *
4735	 * If there are no dirty vdevs, we sync the uberblock to a few
4736	 * random top-level vdevs that are known to be visible in the
4737	 * config cache (see spa_vdev_add() for a complete description).
4738	 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
4739	 */
4740	for (;;) {
4741		/*
4742		 * We hold SCL_STATE to prevent vdev open/close/etc.
4743		 * while we're attempting to write the vdev labels.
4744		 */
4745		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4746
4747		if (list_is_empty(&spa->spa_config_dirty_list)) {
4748			vdev_t *svd[SPA_DVAS_PER_BP];
4749			int svdcount = 0;
4750			int children = rvd->vdev_children;
4751			int c0 = spa_get_random(children);
4752
4753			for (int c = 0; c < children; c++) {
4754				vd = rvd->vdev_child[(c0 + c) % children];
4755				if (vd->vdev_ms_array == 0 || vd->vdev_islog)
4756					continue;
4757				svd[svdcount++] = vd;
4758				if (svdcount == SPA_DVAS_PER_BP)
4759					break;
4760			}
4761			error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
4762			if (error != 0)
4763				error = vdev_config_sync(svd, svdcount, txg,
4764				    B_TRUE);
4765		} else {
4766			error = vdev_config_sync(rvd->vdev_child,
4767			    rvd->vdev_children, txg, B_FALSE);
4768			if (error != 0)
4769				error = vdev_config_sync(rvd->vdev_child,
4770				    rvd->vdev_children, txg, B_TRUE);
4771		}
4772
4773		spa_config_exit(spa, SCL_STATE, FTAG);
4774
4775		if (error == 0)
4776			break;
4777		zio_suspend(spa, NULL);
4778		zio_resume_wait(spa);
4779	}
4780	dmu_tx_commit(tx);
4781
4782	/*
4783	 * Clear the dirty config list.
4784	 */
4785	while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
4786		vdev_config_clean(vd);
4787
4788	/*
4789	 * Now that the new config has synced transactionally,
4790	 * let it become visible to the config cache.
4791	 */
4792	if (spa->spa_config_syncing != NULL) {
4793		spa_config_set(spa, spa->spa_config_syncing);
4794		spa->spa_config_txg = txg;
4795		spa->spa_config_syncing = NULL;
4796	}
4797
4798	spa->spa_ubsync = spa->spa_uberblock;
4799
4800	dsl_pool_sync_done(dp, txg);
4801
4802	/*
4803	 * Update usable space statistics.
4804	 */
4805	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
4806		vdev_sync_done(vd, txg);
4807
4808	spa_update_dspace(spa);
4809
4810	/*
4811	 * It had better be the case that we didn't dirty anything
4812	 * since vdev_config_sync().
4813	 */
4814	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
4815	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
4816	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
4817	ASSERT(defer_bpl->bpl_queue == NULL);
4818	ASSERT(free_bpl->bpl_queue == NULL);
4819
4820	spa->spa_sync_pass = 0;
4821
4822	spa_config_exit(spa, SCL_CONFIG, FTAG);
4823
4824	spa_handle_ignored_writes(spa);
4825
4826	/*
4827	 * If any async tasks have been requested, kick them off.
4828	 */
4829	spa_async_dispatch(spa);
4830}
4831
4832/*
4833 * Sync all pools.  We don't want to hold the namespace lock across these
4834 * operations, so we take a reference on the spa_t and drop the lock during the
4835 * sync.
4836 */
4837void
4838spa_sync_allpools(void)
4839{
4840	spa_t *spa = NULL;
4841	mutex_enter(&spa_namespace_lock);
4842	while ((spa = spa_next(spa)) != NULL) {
4843		if (spa_state(spa) != POOL_STATE_ACTIVE || spa_suspended(spa))
4844			continue;
4845		spa_open_ref(spa, FTAG);
4846		mutex_exit(&spa_namespace_lock);
4847		txg_wait_synced(spa_get_dsl(spa), 0);
4848		mutex_enter(&spa_namespace_lock);
4849		spa_close(spa, FTAG);
4850	}
4851	mutex_exit(&spa_namespace_lock);
4852}
4853
4854/*
4855 * ==========================================================================
4856 * Miscellaneous routines
4857 * ==========================================================================
4858 */
4859
4860/*
4861 * Remove all pools in the system.
4862 */
4863void
4864spa_evict_all(void)
4865{
4866	spa_t *spa;
4867
4868	/*
4869	 * Remove all cached state.  All pools should be closed now,
4870	 * so every spa in the AVL tree should be unreferenced.
4871	 */
4872	mutex_enter(&spa_namespace_lock);
4873	while ((spa = spa_next(NULL)) != NULL) {
4874		/*
4875		 * Stop async tasks.  The async thread may need to detach
4876		 * a device that's been replaced, which requires grabbing
4877		 * spa_namespace_lock, so we must drop it here.
4878		 */
4879		spa_open_ref(spa, FTAG);
4880		mutex_exit(&spa_namespace_lock);
4881		spa_async_suspend(spa);
4882		mutex_enter(&spa_namespace_lock);
4883		spa_close(spa, FTAG);
4884
4885		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4886			spa_unload(spa);
4887			spa_deactivate(spa);
4888		}
4889		spa_remove(spa);
4890	}
4891	mutex_exit(&spa_namespace_lock);
4892}
4893
4894vdev_t *
4895spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
4896{
4897	vdev_t *vd;
4898	int i;
4899
4900	if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
4901		return (vd);
4902
4903	if (aux) {
4904		for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
4905			vd = spa->spa_l2cache.sav_vdevs[i];
4906			if (vd->vdev_guid == guid)
4907				return (vd);
4908		}
4909
4910		for (i = 0; i < spa->spa_spares.sav_count; i++) {
4911			vd = spa->spa_spares.sav_vdevs[i];
4912			if (vd->vdev_guid == guid)
4913				return (vd);
4914		}
4915	}
4916
4917	return (NULL);
4918}
4919
4920void
4921spa_upgrade(spa_t *spa, uint64_t version)
4922{
4923	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4924
4925	/*
4926	 * This should only be called for a non-faulted pool, and since a
4927	 * future version would result in an unopenable pool, this shouldn't be
4928	 * possible.
4929	 */
4930	ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
4931	ASSERT(version >= spa->spa_uberblock.ub_version);
4932
4933	spa->spa_uberblock.ub_version = version;
4934	vdev_config_dirty(spa->spa_root_vdev);
4935
4936	spa_config_exit(spa, SCL_ALL, FTAG);
4937
4938	txg_wait_synced(spa_get_dsl(spa), 0);
4939}
4940
4941boolean_t
4942spa_has_spare(spa_t *spa, uint64_t guid)
4943{
4944	int i;
4945	uint64_t spareguid;
4946	spa_aux_vdev_t *sav = &spa->spa_spares;
4947
4948	for (i = 0; i < sav->sav_count; i++)
4949		if (sav->sav_vdevs[i]->vdev_guid == guid)
4950			return (B_TRUE);
4951
4952	for (i = 0; i < sav->sav_npending; i++) {
4953		if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
4954		    &spareguid) == 0 && spareguid == guid)
4955			return (B_TRUE);
4956	}
4957
4958	return (B_FALSE);
4959}
4960
4961/*
4962 * Check if a pool has an active shared spare device.
4963 * Note: reference count of an active spare is 2, as a spare and as a replace
4964 */
4965static boolean_t
4966spa_has_active_shared_spare(spa_t *spa)
4967{
4968	int i, refcnt;
4969	uint64_t pool;
4970	spa_aux_vdev_t *sav = &spa->spa_spares;
4971
4972	for (i = 0; i < sav->sav_count; i++) {
4973		if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
4974		    &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
4975		    refcnt > 2)
4976			return (B_TRUE);
4977	}
4978
4979	return (B_FALSE);
4980}
4981
4982/*
4983 * Post a sysevent corresponding to the given event.  The 'name' must be one of
4984 * the event definitions in sys/sysevent/eventdefs.h.  The payload will be
4985 * filled in from the spa and (optionally) the vdev.  This doesn't do anything
4986 * in the userland libzpool, as we don't want consumers to misinterpret ztest
4987 * or zdb as real changes.
4988 */
4989void
4990spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
4991{
4992#ifdef _KERNEL
4993	sysevent_t		*ev;
4994	sysevent_attr_list_t	*attr = NULL;
4995	sysevent_value_t	value;
4996	sysevent_id_t		eid;
4997
4998	ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
4999	    SE_SLEEP);
5000
5001	value.value_type = SE_DATA_TYPE_STRING;
5002	value.value.sv_string = spa_name(spa);
5003	if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
5004		goto done;
5005
5006	value.value_type = SE_DATA_TYPE_UINT64;
5007	value.value.sv_uint64 = spa_guid(spa);
5008	if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
5009		goto done;
5010
5011	if (vd) {
5012		value.value_type = SE_DATA_TYPE_UINT64;
5013		value.value.sv_uint64 = vd->vdev_guid;
5014		if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
5015		    SE_SLEEP) != 0)
5016			goto done;
5017
5018		if (vd->vdev_path) {
5019			value.value_type = SE_DATA_TYPE_STRING;
5020			value.value.sv_string = vd->vdev_path;
5021			if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
5022			    &value, SE_SLEEP) != 0)
5023				goto done;
5024		}
5025	}
5026
5027	if (sysevent_attach_attributes(ev, attr) != 0)
5028		goto done;
5029	attr = NULL;
5030
5031	(void) log_sysevent(ev, SE_SLEEP, &eid);
5032
5033done:
5034	if (attr)
5035		sysevent_free_attr(attr);
5036	sysevent_free(ev);
5037#endif
5038}
5039