spa.c revision 69962b5647e4a8b9b14998733b765925381b727e
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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2013 by Delphix. All rights reserved.
25 * Copyright 2013 Nexenta Systems, Inc.  All rights reserved.
26 */
27
28/*
29 * SPA: Storage Pool Allocator
30 *
31 * This file contains all the routines used when modifying on-disk SPA state.
32 * This includes opening, importing, destroying, exporting a pool, and syncing a
33 * pool.
34 */
35
36#include <sys/zfs_context.h>
37#include <sys/fm/fs/zfs.h>
38#include <sys/spa_impl.h>
39#include <sys/zio.h>
40#include <sys/zio_checksum.h>
41#include <sys/dmu.h>
42#include <sys/dmu_tx.h>
43#include <sys/zap.h>
44#include <sys/zil.h>
45#include <sys/ddt.h>
46#include <sys/vdev_impl.h>
47#include <sys/metaslab.h>
48#include <sys/metaslab_impl.h>
49#include <sys/uberblock_impl.h>
50#include <sys/txg.h>
51#include <sys/avl.h>
52#include <sys/dmu_traverse.h>
53#include <sys/dmu_objset.h>
54#include <sys/unique.h>
55#include <sys/dsl_pool.h>
56#include <sys/dsl_dataset.h>
57#include <sys/dsl_dir.h>
58#include <sys/dsl_prop.h>
59#include <sys/dsl_synctask.h>
60#include <sys/fs/zfs.h>
61#include <sys/arc.h>
62#include <sys/callb.h>
63#include <sys/systeminfo.h>
64#include <sys/spa_boot.h>
65#include <sys/zfs_ioctl.h>
66#include <sys/dsl_scan.h>
67#include <sys/zfeature.h>
68#include <sys/dsl_destroy.h>
69
70#ifdef	_KERNEL
71#include <sys/bootprops.h>
72#include <sys/callb.h>
73#include <sys/cpupart.h>
74#include <sys/pool.h>
75#include <sys/sysdc.h>
76#include <sys/zone.h>
77#endif	/* _KERNEL */
78
79#include "zfs_prop.h"
80#include "zfs_comutil.h"
81
82/*
83 * The interval, in seconds, at which failed configuration cache file writes
84 * should be retried.
85 */
86static int zfs_ccw_retry_interval = 300;
87
88typedef enum zti_modes {
89	ZTI_MODE_FIXED,			/* value is # of threads (min 1) */
90	ZTI_MODE_BATCH,			/* cpu-intensive; value is ignored */
91	ZTI_MODE_NULL,			/* don't create a taskq */
92	ZTI_NMODES
93} zti_modes_t;
94
95#define	ZTI_P(n, q)	{ ZTI_MODE_FIXED, (n), (q) }
96#define	ZTI_BATCH	{ ZTI_MODE_BATCH, 0, 1 }
97#define	ZTI_NULL	{ ZTI_MODE_NULL, 0, 0 }
98
99#define	ZTI_N(n)	ZTI_P(n, 1)
100#define	ZTI_ONE		ZTI_N(1)
101
102typedef struct zio_taskq_info {
103	zti_modes_t zti_mode;
104	uint_t zti_value;
105	uint_t zti_count;
106} zio_taskq_info_t;
107
108static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
109	"issue", "issue_high", "intr", "intr_high"
110};
111
112/*
113 * This table defines the taskq settings for each ZFS I/O type. When
114 * initializing a pool, we use this table to create an appropriately sized
115 * taskq. Some operations are low volume and therefore have a small, static
116 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
117 * macros. Other operations process a large amount of data; the ZTI_BATCH
118 * macro causes us to create a taskq oriented for throughput. Some operations
119 * are so high frequency and short-lived that the taskq itself can become a a
120 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
121 * additional degree of parallelism specified by the number of threads per-
122 * taskq and the number of taskqs; when dispatching an event in this case, the
123 * particular taskq is chosen at random.
124 *
125 * The different taskq priorities are to handle the different contexts (issue
126 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
127 * need to be handled with minimum delay.
128 */
129const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
130	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
131	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* NULL */
132	{ ZTI_N(8),	ZTI_NULL,	ZTI_BATCH,	ZTI_NULL }, /* READ */
133	{ ZTI_BATCH,	ZTI_N(5),	ZTI_N(8),	ZTI_N(5) }, /* WRITE */
134	{ ZTI_P(12, 8),	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* FREE */
135	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* CLAIM */
136	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* IOCTL */
137};
138
139static void spa_sync_version(void *arg, dmu_tx_t *tx);
140static void spa_sync_props(void *arg, dmu_tx_t *tx);
141static boolean_t spa_has_active_shared_spare(spa_t *spa);
142static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
143    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
144    char **ereport);
145static void spa_vdev_resilver_done(spa_t *spa);
146
147uint_t		zio_taskq_batch_pct = 75;	/* 1 thread per cpu in pset */
148id_t		zio_taskq_psrset_bind = PS_NONE;
149boolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
150uint_t		zio_taskq_basedc = 80;		/* base duty cycle */
151
152boolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
153extern int	zfs_sync_pass_deferred_free;
154
155/*
156 * This (illegal) pool name is used when temporarily importing a spa_t in order
157 * to get the vdev stats associated with the imported devices.
158 */
159#define	TRYIMPORT_NAME	"$import"
160
161/*
162 * ==========================================================================
163 * SPA properties routines
164 * ==========================================================================
165 */
166
167/*
168 * Add a (source=src, propname=propval) list to an nvlist.
169 */
170static void
171spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
172    uint64_t intval, zprop_source_t src)
173{
174	const char *propname = zpool_prop_to_name(prop);
175	nvlist_t *propval;
176
177	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
178	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
179
180	if (strval != NULL)
181		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
182	else
183		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
184
185	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
186	nvlist_free(propval);
187}
188
189/*
190 * Get property values from the spa configuration.
191 */
192static void
193spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
194{
195	vdev_t *rvd = spa->spa_root_vdev;
196	dsl_pool_t *pool = spa->spa_dsl_pool;
197	uint64_t size;
198	uint64_t alloc;
199	uint64_t space;
200	uint64_t cap, version;
201	zprop_source_t src = ZPROP_SRC_NONE;
202	spa_config_dirent_t *dp;
203
204	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
205
206	if (rvd != NULL) {
207		alloc = metaslab_class_get_alloc(spa_normal_class(spa));
208		size = metaslab_class_get_space(spa_normal_class(spa));
209		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
210		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
211		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
212		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
213		    size - alloc, src);
214
215		space = 0;
216		for (int c = 0; c < rvd->vdev_children; c++) {
217			vdev_t *tvd = rvd->vdev_child[c];
218			space += tvd->vdev_max_asize - tvd->vdev_asize;
219		}
220		spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
221		    src);
222
223		spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
224		    (spa_mode(spa) == FREAD), src);
225
226		cap = (size == 0) ? 0 : (alloc * 100 / size);
227		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
228
229		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
230		    ddt_get_pool_dedup_ratio(spa), src);
231
232		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
233		    rvd->vdev_state, src);
234
235		version = spa_version(spa);
236		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
237			src = ZPROP_SRC_DEFAULT;
238		else
239			src = ZPROP_SRC_LOCAL;
240		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
241	}
242
243	if (pool != NULL) {
244		dsl_dir_t *freedir = pool->dp_free_dir;
245
246		/*
247		 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
248		 * when opening pools before this version freedir will be NULL.
249		 */
250		if (freedir != NULL) {
251			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
252			    freedir->dd_phys->dd_used_bytes, src);
253		} else {
254			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
255			    NULL, 0, src);
256		}
257	}
258
259	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
260
261	if (spa->spa_comment != NULL) {
262		spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
263		    0, ZPROP_SRC_LOCAL);
264	}
265
266	if (spa->spa_root != NULL)
267		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
268		    0, ZPROP_SRC_LOCAL);
269
270	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
271		if (dp->scd_path == NULL) {
272			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
273			    "none", 0, ZPROP_SRC_LOCAL);
274		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
275			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
276			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
277		}
278	}
279}
280
281/*
282 * Get zpool property values.
283 */
284int
285spa_prop_get(spa_t *spa, nvlist_t **nvp)
286{
287	objset_t *mos = spa->spa_meta_objset;
288	zap_cursor_t zc;
289	zap_attribute_t za;
290	int err;
291
292	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
293
294	mutex_enter(&spa->spa_props_lock);
295
296	/*
297	 * Get properties from the spa config.
298	 */
299	spa_prop_get_config(spa, nvp);
300
301	/* If no pool property object, no more prop to get. */
302	if (mos == NULL || spa->spa_pool_props_object == 0) {
303		mutex_exit(&spa->spa_props_lock);
304		return (0);
305	}
306
307	/*
308	 * Get properties from the MOS pool property object.
309	 */
310	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
311	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
312	    zap_cursor_advance(&zc)) {
313		uint64_t intval = 0;
314		char *strval = NULL;
315		zprop_source_t src = ZPROP_SRC_DEFAULT;
316		zpool_prop_t prop;
317
318		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
319			continue;
320
321		switch (za.za_integer_length) {
322		case 8:
323			/* integer property */
324			if (za.za_first_integer !=
325			    zpool_prop_default_numeric(prop))
326				src = ZPROP_SRC_LOCAL;
327
328			if (prop == ZPOOL_PROP_BOOTFS) {
329				dsl_pool_t *dp;
330				dsl_dataset_t *ds = NULL;
331
332				dp = spa_get_dsl(spa);
333				dsl_pool_config_enter(dp, FTAG);
334				if (err = dsl_dataset_hold_obj(dp,
335				    za.za_first_integer, FTAG, &ds)) {
336					dsl_pool_config_exit(dp, FTAG);
337					break;
338				}
339
340				strval = kmem_alloc(
341				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
342				    KM_SLEEP);
343				dsl_dataset_name(ds, strval);
344				dsl_dataset_rele(ds, FTAG);
345				dsl_pool_config_exit(dp, FTAG);
346			} else {
347				strval = NULL;
348				intval = za.za_first_integer;
349			}
350
351			spa_prop_add_list(*nvp, prop, strval, intval, src);
352
353			if (strval != NULL)
354				kmem_free(strval,
355				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
356
357			break;
358
359		case 1:
360			/* string property */
361			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
362			err = zap_lookup(mos, spa->spa_pool_props_object,
363			    za.za_name, 1, za.za_num_integers, strval);
364			if (err) {
365				kmem_free(strval, za.za_num_integers);
366				break;
367			}
368			spa_prop_add_list(*nvp, prop, strval, 0, src);
369			kmem_free(strval, za.za_num_integers);
370			break;
371
372		default:
373			break;
374		}
375	}
376	zap_cursor_fini(&zc);
377	mutex_exit(&spa->spa_props_lock);
378out:
379	if (err && err != ENOENT) {
380		nvlist_free(*nvp);
381		*nvp = NULL;
382		return (err);
383	}
384
385	return (0);
386}
387
388/*
389 * Validate the given pool properties nvlist and modify the list
390 * for the property values to be set.
391 */
392static int
393spa_prop_validate(spa_t *spa, nvlist_t *props)
394{
395	nvpair_t *elem;
396	int error = 0, reset_bootfs = 0;
397	uint64_t objnum = 0;
398	boolean_t has_feature = B_FALSE;
399
400	elem = NULL;
401	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
402		uint64_t intval;
403		char *strval, *slash, *check, *fname;
404		const char *propname = nvpair_name(elem);
405		zpool_prop_t prop = zpool_name_to_prop(propname);
406
407		switch (prop) {
408		case ZPROP_INVAL:
409			if (!zpool_prop_feature(propname)) {
410				error = SET_ERROR(EINVAL);
411				break;
412			}
413
414			/*
415			 * Sanitize the input.
416			 */
417			if (nvpair_type(elem) != DATA_TYPE_UINT64) {
418				error = SET_ERROR(EINVAL);
419				break;
420			}
421
422			if (nvpair_value_uint64(elem, &intval) != 0) {
423				error = SET_ERROR(EINVAL);
424				break;
425			}
426
427			if (intval != 0) {
428				error = SET_ERROR(EINVAL);
429				break;
430			}
431
432			fname = strchr(propname, '@') + 1;
433			if (zfeature_lookup_name(fname, NULL) != 0) {
434				error = SET_ERROR(EINVAL);
435				break;
436			}
437
438			has_feature = B_TRUE;
439			break;
440
441		case ZPOOL_PROP_VERSION:
442			error = nvpair_value_uint64(elem, &intval);
443			if (!error &&
444			    (intval < spa_version(spa) ||
445			    intval > SPA_VERSION_BEFORE_FEATURES ||
446			    has_feature))
447				error = SET_ERROR(EINVAL);
448			break;
449
450		case ZPOOL_PROP_DELEGATION:
451		case ZPOOL_PROP_AUTOREPLACE:
452		case ZPOOL_PROP_LISTSNAPS:
453		case ZPOOL_PROP_AUTOEXPAND:
454			error = nvpair_value_uint64(elem, &intval);
455			if (!error && intval > 1)
456				error = SET_ERROR(EINVAL);
457			break;
458
459		case ZPOOL_PROP_BOOTFS:
460			/*
461			 * If the pool version is less than SPA_VERSION_BOOTFS,
462			 * or the pool is still being created (version == 0),
463			 * the bootfs property cannot be set.
464			 */
465			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
466				error = SET_ERROR(ENOTSUP);
467				break;
468			}
469
470			/*
471			 * Make sure the vdev config is bootable
472			 */
473			if (!vdev_is_bootable(spa->spa_root_vdev)) {
474				error = SET_ERROR(ENOTSUP);
475				break;
476			}
477
478			reset_bootfs = 1;
479
480			error = nvpair_value_string(elem, &strval);
481
482			if (!error) {
483				objset_t *os;
484				uint64_t compress;
485
486				if (strval == NULL || strval[0] == '\0') {
487					objnum = zpool_prop_default_numeric(
488					    ZPOOL_PROP_BOOTFS);
489					break;
490				}
491
492				if (error = dmu_objset_hold(strval, FTAG, &os))
493					break;
494
495				/* Must be ZPL and not gzip compressed. */
496
497				if (dmu_objset_type(os) != DMU_OST_ZFS) {
498					error = SET_ERROR(ENOTSUP);
499				} else if ((error =
500				    dsl_prop_get_int_ds(dmu_objset_ds(os),
501				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
502				    &compress)) == 0 &&
503				    !BOOTFS_COMPRESS_VALID(compress)) {
504					error = SET_ERROR(ENOTSUP);
505				} else {
506					objnum = dmu_objset_id(os);
507				}
508				dmu_objset_rele(os, FTAG);
509			}
510			break;
511
512		case ZPOOL_PROP_FAILUREMODE:
513			error = nvpair_value_uint64(elem, &intval);
514			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
515			    intval > ZIO_FAILURE_MODE_PANIC))
516				error = SET_ERROR(EINVAL);
517
518			/*
519			 * This is a special case which only occurs when
520			 * the pool has completely failed. This allows
521			 * the user to change the in-core failmode property
522			 * without syncing it out to disk (I/Os might
523			 * currently be blocked). We do this by returning
524			 * EIO to the caller (spa_prop_set) to trick it
525			 * into thinking we encountered a property validation
526			 * error.
527			 */
528			if (!error && spa_suspended(spa)) {
529				spa->spa_failmode = intval;
530				error = SET_ERROR(EIO);
531			}
532			break;
533
534		case ZPOOL_PROP_CACHEFILE:
535			if ((error = nvpair_value_string(elem, &strval)) != 0)
536				break;
537
538			if (strval[0] == '\0')
539				break;
540
541			if (strcmp(strval, "none") == 0)
542				break;
543
544			if (strval[0] != '/') {
545				error = SET_ERROR(EINVAL);
546				break;
547			}
548
549			slash = strrchr(strval, '/');
550			ASSERT(slash != NULL);
551
552			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
553			    strcmp(slash, "/..") == 0)
554				error = SET_ERROR(EINVAL);
555			break;
556
557		case ZPOOL_PROP_COMMENT:
558			if ((error = nvpair_value_string(elem, &strval)) != 0)
559				break;
560			for (check = strval; *check != '\0'; check++) {
561				/*
562				 * The kernel doesn't have an easy isprint()
563				 * check.  For this kernel check, we merely
564				 * check ASCII apart from DEL.  Fix this if
565				 * there is an easy-to-use kernel isprint().
566				 */
567				if (*check >= 0x7f) {
568					error = SET_ERROR(EINVAL);
569					break;
570				}
571				check++;
572			}
573			if (strlen(strval) > ZPROP_MAX_COMMENT)
574				error = E2BIG;
575			break;
576
577		case ZPOOL_PROP_DEDUPDITTO:
578			if (spa_version(spa) < SPA_VERSION_DEDUP)
579				error = SET_ERROR(ENOTSUP);
580			else
581				error = nvpair_value_uint64(elem, &intval);
582			if (error == 0 &&
583			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
584				error = SET_ERROR(EINVAL);
585			break;
586		}
587
588		if (error)
589			break;
590	}
591
592	if (!error && reset_bootfs) {
593		error = nvlist_remove(props,
594		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
595
596		if (!error) {
597			error = nvlist_add_uint64(props,
598			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
599		}
600	}
601
602	return (error);
603}
604
605void
606spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
607{
608	char *cachefile;
609	spa_config_dirent_t *dp;
610
611	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
612	    &cachefile) != 0)
613		return;
614
615	dp = kmem_alloc(sizeof (spa_config_dirent_t),
616	    KM_SLEEP);
617
618	if (cachefile[0] == '\0')
619		dp->scd_path = spa_strdup(spa_config_path);
620	else if (strcmp(cachefile, "none") == 0)
621		dp->scd_path = NULL;
622	else
623		dp->scd_path = spa_strdup(cachefile);
624
625	list_insert_head(&spa->spa_config_list, dp);
626	if (need_sync)
627		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
628}
629
630int
631spa_prop_set(spa_t *spa, nvlist_t *nvp)
632{
633	int error;
634	nvpair_t *elem = NULL;
635	boolean_t need_sync = B_FALSE;
636
637	if ((error = spa_prop_validate(spa, nvp)) != 0)
638		return (error);
639
640	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
641		zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
642
643		if (prop == ZPOOL_PROP_CACHEFILE ||
644		    prop == ZPOOL_PROP_ALTROOT ||
645		    prop == ZPOOL_PROP_READONLY)
646			continue;
647
648		if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
649			uint64_t ver;
650
651			if (prop == ZPOOL_PROP_VERSION) {
652				VERIFY(nvpair_value_uint64(elem, &ver) == 0);
653			} else {
654				ASSERT(zpool_prop_feature(nvpair_name(elem)));
655				ver = SPA_VERSION_FEATURES;
656				need_sync = B_TRUE;
657			}
658
659			/* Save time if the version is already set. */
660			if (ver == spa_version(spa))
661				continue;
662
663			/*
664			 * In addition to the pool directory object, we might
665			 * create the pool properties object, the features for
666			 * read object, the features for write object, or the
667			 * feature descriptions object.
668			 */
669			error = dsl_sync_task(spa->spa_name, NULL,
670			    spa_sync_version, &ver, 6);
671			if (error)
672				return (error);
673			continue;
674		}
675
676		need_sync = B_TRUE;
677		break;
678	}
679
680	if (need_sync) {
681		return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
682		    nvp, 6));
683	}
684
685	return (0);
686}
687
688/*
689 * If the bootfs property value is dsobj, clear it.
690 */
691void
692spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
693{
694	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
695		VERIFY(zap_remove(spa->spa_meta_objset,
696		    spa->spa_pool_props_object,
697		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
698		spa->spa_bootfs = 0;
699	}
700}
701
702/*ARGSUSED*/
703static int
704spa_change_guid_check(void *arg, dmu_tx_t *tx)
705{
706	uint64_t *newguid = arg;
707	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
708	vdev_t *rvd = spa->spa_root_vdev;
709	uint64_t vdev_state;
710
711	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
712	vdev_state = rvd->vdev_state;
713	spa_config_exit(spa, SCL_STATE, FTAG);
714
715	if (vdev_state != VDEV_STATE_HEALTHY)
716		return (SET_ERROR(ENXIO));
717
718	ASSERT3U(spa_guid(spa), !=, *newguid);
719
720	return (0);
721}
722
723static void
724spa_change_guid_sync(void *arg, dmu_tx_t *tx)
725{
726	uint64_t *newguid = arg;
727	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
728	uint64_t oldguid;
729	vdev_t *rvd = spa->spa_root_vdev;
730
731	oldguid = spa_guid(spa);
732
733	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
734	rvd->vdev_guid = *newguid;
735	rvd->vdev_guid_sum += (*newguid - oldguid);
736	vdev_config_dirty(rvd);
737	spa_config_exit(spa, SCL_STATE, FTAG);
738
739	spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
740	    oldguid, *newguid);
741}
742
743/*
744 * Change the GUID for the pool.  This is done so that we can later
745 * re-import a pool built from a clone of our own vdevs.  We will modify
746 * the root vdev's guid, our own pool guid, and then mark all of our
747 * vdevs dirty.  Note that we must make sure that all our vdevs are
748 * online when we do this, or else any vdevs that weren't present
749 * would be orphaned from our pool.  We are also going to issue a
750 * sysevent to update any watchers.
751 */
752int
753spa_change_guid(spa_t *spa)
754{
755	int error;
756	uint64_t guid;
757
758	mutex_enter(&spa->spa_vdev_top_lock);
759	mutex_enter(&spa_namespace_lock);
760	guid = spa_generate_guid(NULL);
761
762	error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
763	    spa_change_guid_sync, &guid, 5);
764
765	if (error == 0) {
766		spa_config_sync(spa, B_FALSE, B_TRUE);
767		spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
768	}
769
770	mutex_exit(&spa_namespace_lock);
771	mutex_exit(&spa->spa_vdev_top_lock);
772
773	return (error);
774}
775
776/*
777 * ==========================================================================
778 * SPA state manipulation (open/create/destroy/import/export)
779 * ==========================================================================
780 */
781
782static int
783spa_error_entry_compare(const void *a, const void *b)
784{
785	spa_error_entry_t *sa = (spa_error_entry_t *)a;
786	spa_error_entry_t *sb = (spa_error_entry_t *)b;
787	int ret;
788
789	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
790	    sizeof (zbookmark_t));
791
792	if (ret < 0)
793		return (-1);
794	else if (ret > 0)
795		return (1);
796	else
797		return (0);
798}
799
800/*
801 * Utility function which retrieves copies of the current logs and
802 * re-initializes them in the process.
803 */
804void
805spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
806{
807	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
808
809	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
810	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
811
812	avl_create(&spa->spa_errlist_scrub,
813	    spa_error_entry_compare, sizeof (spa_error_entry_t),
814	    offsetof(spa_error_entry_t, se_avl));
815	avl_create(&spa->spa_errlist_last,
816	    spa_error_entry_compare, sizeof (spa_error_entry_t),
817	    offsetof(spa_error_entry_t, se_avl));
818}
819
820static void
821spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
822{
823	const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
824	enum zti_modes mode = ztip->zti_mode;
825	uint_t value = ztip->zti_value;
826	uint_t count = ztip->zti_count;
827	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
828	char name[32];
829	uint_t flags = 0;
830	boolean_t batch = B_FALSE;
831
832	if (mode == ZTI_MODE_NULL) {
833		tqs->stqs_count = 0;
834		tqs->stqs_taskq = NULL;
835		return;
836	}
837
838	ASSERT3U(count, >, 0);
839
840	tqs->stqs_count = count;
841	tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
842
843	switch (mode) {
844	case ZTI_MODE_FIXED:
845		ASSERT3U(value, >=, 1);
846		value = MAX(value, 1);
847		break;
848
849	case ZTI_MODE_BATCH:
850		batch = B_TRUE;
851		flags |= TASKQ_THREADS_CPU_PCT;
852		value = zio_taskq_batch_pct;
853		break;
854
855	default:
856		panic("unrecognized mode for %s_%s taskq (%u:%u) in "
857		    "spa_activate()",
858		    zio_type_name[t], zio_taskq_types[q], mode, value);
859		break;
860	}
861
862	for (uint_t i = 0; i < count; i++) {
863		taskq_t *tq;
864
865		if (count > 1) {
866			(void) snprintf(name, sizeof (name), "%s_%s_%u",
867			    zio_type_name[t], zio_taskq_types[q], i);
868		} else {
869			(void) snprintf(name, sizeof (name), "%s_%s",
870			    zio_type_name[t], zio_taskq_types[q]);
871		}
872
873		if (zio_taskq_sysdc && spa->spa_proc != &p0) {
874			if (batch)
875				flags |= TASKQ_DC_BATCH;
876
877			tq = taskq_create_sysdc(name, value, 50, INT_MAX,
878			    spa->spa_proc, zio_taskq_basedc, flags);
879		} else {
880			pri_t pri = maxclsyspri;
881			/*
882			 * The write issue taskq can be extremely CPU
883			 * intensive.  Run it at slightly lower priority
884			 * than the other taskqs.
885			 */
886			if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
887				pri--;
888
889			tq = taskq_create_proc(name, value, pri, 50,
890			    INT_MAX, spa->spa_proc, flags);
891		}
892
893		tqs->stqs_taskq[i] = tq;
894	}
895}
896
897static void
898spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
899{
900	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
901
902	if (tqs->stqs_taskq == NULL) {
903		ASSERT0(tqs->stqs_count);
904		return;
905	}
906
907	for (uint_t i = 0; i < tqs->stqs_count; i++) {
908		ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
909		taskq_destroy(tqs->stqs_taskq[i]);
910	}
911
912	kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
913	tqs->stqs_taskq = NULL;
914}
915
916/*
917 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
918 * Note that a type may have multiple discrete taskqs to avoid lock contention
919 * on the taskq itself. In that case we choose which taskq at random by using
920 * the low bits of gethrtime().
921 */
922void
923spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
924    task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
925{
926	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
927	taskq_t *tq;
928
929	ASSERT3P(tqs->stqs_taskq, !=, NULL);
930	ASSERT3U(tqs->stqs_count, !=, 0);
931
932	if (tqs->stqs_count == 1) {
933		tq = tqs->stqs_taskq[0];
934	} else {
935		tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
936	}
937
938	taskq_dispatch_ent(tq, func, arg, flags, ent);
939}
940
941static void
942spa_create_zio_taskqs(spa_t *spa)
943{
944	for (int t = 0; t < ZIO_TYPES; t++) {
945		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
946			spa_taskqs_init(spa, t, q);
947		}
948	}
949}
950
951#ifdef _KERNEL
952static void
953spa_thread(void *arg)
954{
955	callb_cpr_t cprinfo;
956
957	spa_t *spa = arg;
958	user_t *pu = PTOU(curproc);
959
960	CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
961	    spa->spa_name);
962
963	ASSERT(curproc != &p0);
964	(void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
965	    "zpool-%s", spa->spa_name);
966	(void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
967
968	/* bind this thread to the requested psrset */
969	if (zio_taskq_psrset_bind != PS_NONE) {
970		pool_lock();
971		mutex_enter(&cpu_lock);
972		mutex_enter(&pidlock);
973		mutex_enter(&curproc->p_lock);
974
975		if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
976		    0, NULL, NULL) == 0)  {
977			curthread->t_bind_pset = zio_taskq_psrset_bind;
978		} else {
979			cmn_err(CE_WARN,
980			    "Couldn't bind process for zfs pool \"%s\" to "
981			    "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
982		}
983
984		mutex_exit(&curproc->p_lock);
985		mutex_exit(&pidlock);
986		mutex_exit(&cpu_lock);
987		pool_unlock();
988	}
989
990	if (zio_taskq_sysdc) {
991		sysdc_thread_enter(curthread, 100, 0);
992	}
993
994	spa->spa_proc = curproc;
995	spa->spa_did = curthread->t_did;
996
997	spa_create_zio_taskqs(spa);
998
999	mutex_enter(&spa->spa_proc_lock);
1000	ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1001
1002	spa->spa_proc_state = SPA_PROC_ACTIVE;
1003	cv_broadcast(&spa->spa_proc_cv);
1004
1005	CALLB_CPR_SAFE_BEGIN(&cprinfo);
1006	while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1007		cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1008	CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1009
1010	ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1011	spa->spa_proc_state = SPA_PROC_GONE;
1012	spa->spa_proc = &p0;
1013	cv_broadcast(&spa->spa_proc_cv);
1014	CALLB_CPR_EXIT(&cprinfo);	/* drops spa_proc_lock */
1015
1016	mutex_enter(&curproc->p_lock);
1017	lwp_exit();
1018}
1019#endif
1020
1021/*
1022 * Activate an uninitialized pool.
1023 */
1024static void
1025spa_activate(spa_t *spa, int mode)
1026{
1027	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1028
1029	spa->spa_state = POOL_STATE_ACTIVE;
1030	spa->spa_mode = mode;
1031
1032	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1033	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1034
1035	/* Try to create a covering process */
1036	mutex_enter(&spa->spa_proc_lock);
1037	ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1038	ASSERT(spa->spa_proc == &p0);
1039	spa->spa_did = 0;
1040
1041	/* Only create a process if we're going to be around a while. */
1042	if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1043		if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1044		    NULL, 0) == 0) {
1045			spa->spa_proc_state = SPA_PROC_CREATED;
1046			while (spa->spa_proc_state == SPA_PROC_CREATED) {
1047				cv_wait(&spa->spa_proc_cv,
1048				    &spa->spa_proc_lock);
1049			}
1050			ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1051			ASSERT(spa->spa_proc != &p0);
1052			ASSERT(spa->spa_did != 0);
1053		} else {
1054#ifdef _KERNEL
1055			cmn_err(CE_WARN,
1056			    "Couldn't create process for zfs pool \"%s\"\n",
1057			    spa->spa_name);
1058#endif
1059		}
1060	}
1061	mutex_exit(&spa->spa_proc_lock);
1062
1063	/* If we didn't create a process, we need to create our taskqs. */
1064	if (spa->spa_proc == &p0) {
1065		spa_create_zio_taskqs(spa);
1066	}
1067
1068	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1069	    offsetof(vdev_t, vdev_config_dirty_node));
1070	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1071	    offsetof(vdev_t, vdev_state_dirty_node));
1072
1073	txg_list_create(&spa->spa_vdev_txg_list,
1074	    offsetof(struct vdev, vdev_txg_node));
1075
1076	avl_create(&spa->spa_errlist_scrub,
1077	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1078	    offsetof(spa_error_entry_t, se_avl));
1079	avl_create(&spa->spa_errlist_last,
1080	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1081	    offsetof(spa_error_entry_t, se_avl));
1082}
1083
1084/*
1085 * Opposite of spa_activate().
1086 */
1087static void
1088spa_deactivate(spa_t *spa)
1089{
1090	ASSERT(spa->spa_sync_on == B_FALSE);
1091	ASSERT(spa->spa_dsl_pool == NULL);
1092	ASSERT(spa->spa_root_vdev == NULL);
1093	ASSERT(spa->spa_async_zio_root == NULL);
1094	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1095
1096	txg_list_destroy(&spa->spa_vdev_txg_list);
1097
1098	list_destroy(&spa->spa_config_dirty_list);
1099	list_destroy(&spa->spa_state_dirty_list);
1100
1101	for (int t = 0; t < ZIO_TYPES; t++) {
1102		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1103			spa_taskqs_fini(spa, t, q);
1104		}
1105	}
1106
1107	metaslab_class_destroy(spa->spa_normal_class);
1108	spa->spa_normal_class = NULL;
1109
1110	metaslab_class_destroy(spa->spa_log_class);
1111	spa->spa_log_class = NULL;
1112
1113	/*
1114	 * If this was part of an import or the open otherwise failed, we may
1115	 * still have errors left in the queues.  Empty them just in case.
1116	 */
1117	spa_errlog_drain(spa);
1118
1119	avl_destroy(&spa->spa_errlist_scrub);
1120	avl_destroy(&spa->spa_errlist_last);
1121
1122	spa->spa_state = POOL_STATE_UNINITIALIZED;
1123
1124	mutex_enter(&spa->spa_proc_lock);
1125	if (spa->spa_proc_state != SPA_PROC_NONE) {
1126		ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1127		spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1128		cv_broadcast(&spa->spa_proc_cv);
1129		while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1130			ASSERT(spa->spa_proc != &p0);
1131			cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1132		}
1133		ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1134		spa->spa_proc_state = SPA_PROC_NONE;
1135	}
1136	ASSERT(spa->spa_proc == &p0);
1137	mutex_exit(&spa->spa_proc_lock);
1138
1139	/*
1140	 * We want to make sure spa_thread() has actually exited the ZFS
1141	 * module, so that the module can't be unloaded out from underneath
1142	 * it.
1143	 */
1144	if (spa->spa_did != 0) {
1145		thread_join(spa->spa_did);
1146		spa->spa_did = 0;
1147	}
1148}
1149
1150/*
1151 * Verify a pool configuration, and construct the vdev tree appropriately.  This
1152 * will create all the necessary vdevs in the appropriate layout, with each vdev
1153 * in the CLOSED state.  This will prep the pool before open/creation/import.
1154 * All vdev validation is done by the vdev_alloc() routine.
1155 */
1156static int
1157spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1158    uint_t id, int atype)
1159{
1160	nvlist_t **child;
1161	uint_t children;
1162	int error;
1163
1164	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1165		return (error);
1166
1167	if ((*vdp)->vdev_ops->vdev_op_leaf)
1168		return (0);
1169
1170	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1171	    &child, &children);
1172
1173	if (error == ENOENT)
1174		return (0);
1175
1176	if (error) {
1177		vdev_free(*vdp);
1178		*vdp = NULL;
1179		return (SET_ERROR(EINVAL));
1180	}
1181
1182	for (int c = 0; c < children; c++) {
1183		vdev_t *vd;
1184		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1185		    atype)) != 0) {
1186			vdev_free(*vdp);
1187			*vdp = NULL;
1188			return (error);
1189		}
1190	}
1191
1192	ASSERT(*vdp != NULL);
1193
1194	return (0);
1195}
1196
1197/*
1198 * Opposite of spa_load().
1199 */
1200static void
1201spa_unload(spa_t *spa)
1202{
1203	int i;
1204
1205	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1206
1207	/*
1208	 * Stop async tasks.
1209	 */
1210	spa_async_suspend(spa);
1211
1212	/*
1213	 * Stop syncing.
1214	 */
1215	if (spa->spa_sync_on) {
1216		txg_sync_stop(spa->spa_dsl_pool);
1217		spa->spa_sync_on = B_FALSE;
1218	}
1219
1220	/*
1221	 * Wait for any outstanding async I/O to complete.
1222	 */
1223	if (spa->spa_async_zio_root != NULL) {
1224		(void) zio_wait(spa->spa_async_zio_root);
1225		spa->spa_async_zio_root = NULL;
1226	}
1227
1228	bpobj_close(&spa->spa_deferred_bpobj);
1229
1230	/*
1231	 * Close the dsl pool.
1232	 */
1233	if (spa->spa_dsl_pool) {
1234		dsl_pool_close(spa->spa_dsl_pool);
1235		spa->spa_dsl_pool = NULL;
1236		spa->spa_meta_objset = NULL;
1237	}
1238
1239	ddt_unload(spa);
1240
1241	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1242
1243	/*
1244	 * Drop and purge level 2 cache
1245	 */
1246	spa_l2cache_drop(spa);
1247
1248	/*
1249	 * Close all vdevs.
1250	 */
1251	if (spa->spa_root_vdev)
1252		vdev_free(spa->spa_root_vdev);
1253	ASSERT(spa->spa_root_vdev == NULL);
1254
1255	for (i = 0; i < spa->spa_spares.sav_count; i++)
1256		vdev_free(spa->spa_spares.sav_vdevs[i]);
1257	if (spa->spa_spares.sav_vdevs) {
1258		kmem_free(spa->spa_spares.sav_vdevs,
1259		    spa->spa_spares.sav_count * sizeof (void *));
1260		spa->spa_spares.sav_vdevs = NULL;
1261	}
1262	if (spa->spa_spares.sav_config) {
1263		nvlist_free(spa->spa_spares.sav_config);
1264		spa->spa_spares.sav_config = NULL;
1265	}
1266	spa->spa_spares.sav_count = 0;
1267
1268	for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1269		vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1270		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1271	}
1272	if (spa->spa_l2cache.sav_vdevs) {
1273		kmem_free(spa->spa_l2cache.sav_vdevs,
1274		    spa->spa_l2cache.sav_count * sizeof (void *));
1275		spa->spa_l2cache.sav_vdevs = NULL;
1276	}
1277	if (spa->spa_l2cache.sav_config) {
1278		nvlist_free(spa->spa_l2cache.sav_config);
1279		spa->spa_l2cache.sav_config = NULL;
1280	}
1281	spa->spa_l2cache.sav_count = 0;
1282
1283	spa->spa_async_suspended = 0;
1284
1285	if (spa->spa_comment != NULL) {
1286		spa_strfree(spa->spa_comment);
1287		spa->spa_comment = NULL;
1288	}
1289
1290	spa_config_exit(spa, SCL_ALL, FTAG);
1291}
1292
1293/*
1294 * Load (or re-load) the current list of vdevs describing the active spares for
1295 * this pool.  When this is called, we have some form of basic information in
1296 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1297 * then re-generate a more complete list including status information.
1298 */
1299static void
1300spa_load_spares(spa_t *spa)
1301{
1302	nvlist_t **spares;
1303	uint_t nspares;
1304	int i;
1305	vdev_t *vd, *tvd;
1306
1307	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1308
1309	/*
1310	 * First, close and free any existing spare vdevs.
1311	 */
1312	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1313		vd = spa->spa_spares.sav_vdevs[i];
1314
1315		/* Undo the call to spa_activate() below */
1316		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1317		    B_FALSE)) != NULL && tvd->vdev_isspare)
1318			spa_spare_remove(tvd);
1319		vdev_close(vd);
1320		vdev_free(vd);
1321	}
1322
1323	if (spa->spa_spares.sav_vdevs)
1324		kmem_free(spa->spa_spares.sav_vdevs,
1325		    spa->spa_spares.sav_count * sizeof (void *));
1326
1327	if (spa->spa_spares.sav_config == NULL)
1328		nspares = 0;
1329	else
1330		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1331		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1332
1333	spa->spa_spares.sav_count = (int)nspares;
1334	spa->spa_spares.sav_vdevs = NULL;
1335
1336	if (nspares == 0)
1337		return;
1338
1339	/*
1340	 * Construct the array of vdevs, opening them to get status in the
1341	 * process.   For each spare, there is potentially two different vdev_t
1342	 * structures associated with it: one in the list of spares (used only
1343	 * for basic validation purposes) and one in the active vdev
1344	 * configuration (if it's spared in).  During this phase we open and
1345	 * validate each vdev on the spare list.  If the vdev also exists in the
1346	 * active configuration, then we also mark this vdev as an active spare.
1347	 */
1348	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1349	    KM_SLEEP);
1350	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1351		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1352		    VDEV_ALLOC_SPARE) == 0);
1353		ASSERT(vd != NULL);
1354
1355		spa->spa_spares.sav_vdevs[i] = vd;
1356
1357		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1358		    B_FALSE)) != NULL) {
1359			if (!tvd->vdev_isspare)
1360				spa_spare_add(tvd);
1361
1362			/*
1363			 * We only mark the spare active if we were successfully
1364			 * able to load the vdev.  Otherwise, importing a pool
1365			 * with a bad active spare would result in strange
1366			 * behavior, because multiple pool would think the spare
1367			 * is actively in use.
1368			 *
1369			 * There is a vulnerability here to an equally bizarre
1370			 * circumstance, where a dead active spare is later
1371			 * brought back to life (onlined or otherwise).  Given
1372			 * the rarity of this scenario, and the extra complexity
1373			 * it adds, we ignore the possibility.
1374			 */
1375			if (!vdev_is_dead(tvd))
1376				spa_spare_activate(tvd);
1377		}
1378
1379		vd->vdev_top = vd;
1380		vd->vdev_aux = &spa->spa_spares;
1381
1382		if (vdev_open(vd) != 0)
1383			continue;
1384
1385		if (vdev_validate_aux(vd) == 0)
1386			spa_spare_add(vd);
1387	}
1388
1389	/*
1390	 * Recompute the stashed list of spares, with status information
1391	 * this time.
1392	 */
1393	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1394	    DATA_TYPE_NVLIST_ARRAY) == 0);
1395
1396	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1397	    KM_SLEEP);
1398	for (i = 0; i < spa->spa_spares.sav_count; i++)
1399		spares[i] = vdev_config_generate(spa,
1400		    spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1401	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1402	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1403	for (i = 0; i < spa->spa_spares.sav_count; i++)
1404		nvlist_free(spares[i]);
1405	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1406}
1407
1408/*
1409 * Load (or re-load) the current list of vdevs describing the active l2cache for
1410 * this pool.  When this is called, we have some form of basic information in
1411 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1412 * then re-generate a more complete list including status information.
1413 * Devices which are already active have their details maintained, and are
1414 * not re-opened.
1415 */
1416static void
1417spa_load_l2cache(spa_t *spa)
1418{
1419	nvlist_t **l2cache;
1420	uint_t nl2cache;
1421	int i, j, oldnvdevs;
1422	uint64_t guid;
1423	vdev_t *vd, **oldvdevs, **newvdevs;
1424	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1425
1426	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1427
1428	if (sav->sav_config != NULL) {
1429		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1430		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1431		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1432	} else {
1433		nl2cache = 0;
1434		newvdevs = NULL;
1435	}
1436
1437	oldvdevs = sav->sav_vdevs;
1438	oldnvdevs = sav->sav_count;
1439	sav->sav_vdevs = NULL;
1440	sav->sav_count = 0;
1441
1442	/*
1443	 * Process new nvlist of vdevs.
1444	 */
1445	for (i = 0; i < nl2cache; i++) {
1446		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1447		    &guid) == 0);
1448
1449		newvdevs[i] = NULL;
1450		for (j = 0; j < oldnvdevs; j++) {
1451			vd = oldvdevs[j];
1452			if (vd != NULL && guid == vd->vdev_guid) {
1453				/*
1454				 * Retain previous vdev for add/remove ops.
1455				 */
1456				newvdevs[i] = vd;
1457				oldvdevs[j] = NULL;
1458				break;
1459			}
1460		}
1461
1462		if (newvdevs[i] == NULL) {
1463			/*
1464			 * Create new vdev
1465			 */
1466			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1467			    VDEV_ALLOC_L2CACHE) == 0);
1468			ASSERT(vd != NULL);
1469			newvdevs[i] = vd;
1470
1471			/*
1472			 * Commit this vdev as an l2cache device,
1473			 * even if it fails to open.
1474			 */
1475			spa_l2cache_add(vd);
1476
1477			vd->vdev_top = vd;
1478			vd->vdev_aux = sav;
1479
1480			spa_l2cache_activate(vd);
1481
1482			if (vdev_open(vd) != 0)
1483				continue;
1484
1485			(void) vdev_validate_aux(vd);
1486
1487			if (!vdev_is_dead(vd))
1488				l2arc_add_vdev(spa, vd);
1489		}
1490	}
1491
1492	/*
1493	 * Purge vdevs that were dropped
1494	 */
1495	for (i = 0; i < oldnvdevs; i++) {
1496		uint64_t pool;
1497
1498		vd = oldvdevs[i];
1499		if (vd != NULL) {
1500			ASSERT(vd->vdev_isl2cache);
1501
1502			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1503			    pool != 0ULL && l2arc_vdev_present(vd))
1504				l2arc_remove_vdev(vd);
1505			vdev_clear_stats(vd);
1506			vdev_free(vd);
1507		}
1508	}
1509
1510	if (oldvdevs)
1511		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1512
1513	if (sav->sav_config == NULL)
1514		goto out;
1515
1516	sav->sav_vdevs = newvdevs;
1517	sav->sav_count = (int)nl2cache;
1518
1519	/*
1520	 * Recompute the stashed list of l2cache devices, with status
1521	 * information this time.
1522	 */
1523	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1524	    DATA_TYPE_NVLIST_ARRAY) == 0);
1525
1526	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1527	for (i = 0; i < sav->sav_count; i++)
1528		l2cache[i] = vdev_config_generate(spa,
1529		    sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1530	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1531	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1532out:
1533	for (i = 0; i < sav->sav_count; i++)
1534		nvlist_free(l2cache[i]);
1535	if (sav->sav_count)
1536		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1537}
1538
1539static int
1540load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1541{
1542	dmu_buf_t *db;
1543	char *packed = NULL;
1544	size_t nvsize = 0;
1545	int error;
1546	*value = NULL;
1547
1548	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1549	nvsize = *(uint64_t *)db->db_data;
1550	dmu_buf_rele(db, FTAG);
1551
1552	packed = kmem_alloc(nvsize, KM_SLEEP);
1553	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1554	    DMU_READ_PREFETCH);
1555	if (error == 0)
1556		error = nvlist_unpack(packed, nvsize, value, 0);
1557	kmem_free(packed, nvsize);
1558
1559	return (error);
1560}
1561
1562/*
1563 * Checks to see if the given vdev could not be opened, in which case we post a
1564 * sysevent to notify the autoreplace code that the device has been removed.
1565 */
1566static void
1567spa_check_removed(vdev_t *vd)
1568{
1569	for (int c = 0; c < vd->vdev_children; c++)
1570		spa_check_removed(vd->vdev_child[c]);
1571
1572	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1573	    !vd->vdev_ishole) {
1574		zfs_post_autoreplace(vd->vdev_spa, vd);
1575		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1576	}
1577}
1578
1579/*
1580 * Validate the current config against the MOS config
1581 */
1582static boolean_t
1583spa_config_valid(spa_t *spa, nvlist_t *config)
1584{
1585	vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1586	nvlist_t *nv;
1587
1588	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1589
1590	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1591	VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1592
1593	ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1594
1595	/*
1596	 * If we're doing a normal import, then build up any additional
1597	 * diagnostic information about missing devices in this config.
1598	 * We'll pass this up to the user for further processing.
1599	 */
1600	if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1601		nvlist_t **child, *nv;
1602		uint64_t idx = 0;
1603
1604		child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1605		    KM_SLEEP);
1606		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1607
1608		for (int c = 0; c < rvd->vdev_children; c++) {
1609			vdev_t *tvd = rvd->vdev_child[c];
1610			vdev_t *mtvd  = mrvd->vdev_child[c];
1611
1612			if (tvd->vdev_ops == &vdev_missing_ops &&
1613			    mtvd->vdev_ops != &vdev_missing_ops &&
1614			    mtvd->vdev_islog)
1615				child[idx++] = vdev_config_generate(spa, mtvd,
1616				    B_FALSE, 0);
1617		}
1618
1619		if (idx) {
1620			VERIFY(nvlist_add_nvlist_array(nv,
1621			    ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1622			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1623			    ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1624
1625			for (int i = 0; i < idx; i++)
1626				nvlist_free(child[i]);
1627		}
1628		nvlist_free(nv);
1629		kmem_free(child, rvd->vdev_children * sizeof (char **));
1630	}
1631
1632	/*
1633	 * Compare the root vdev tree with the information we have
1634	 * from the MOS config (mrvd). Check each top-level vdev
1635	 * with the corresponding MOS config top-level (mtvd).
1636	 */
1637	for (int c = 0; c < rvd->vdev_children; c++) {
1638		vdev_t *tvd = rvd->vdev_child[c];
1639		vdev_t *mtvd  = mrvd->vdev_child[c];
1640
1641		/*
1642		 * Resolve any "missing" vdevs in the current configuration.
1643		 * If we find that the MOS config has more accurate information
1644		 * about the top-level vdev then use that vdev instead.
1645		 */
1646		if (tvd->vdev_ops == &vdev_missing_ops &&
1647		    mtvd->vdev_ops != &vdev_missing_ops) {
1648
1649			if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1650				continue;
1651
1652			/*
1653			 * Device specific actions.
1654			 */
1655			if (mtvd->vdev_islog) {
1656				spa_set_log_state(spa, SPA_LOG_CLEAR);
1657			} else {
1658				/*
1659				 * XXX - once we have 'readonly' pool
1660				 * support we should be able to handle
1661				 * missing data devices by transitioning
1662				 * the pool to readonly.
1663				 */
1664				continue;
1665			}
1666
1667			/*
1668			 * Swap the missing vdev with the data we were
1669			 * able to obtain from the MOS config.
1670			 */
1671			vdev_remove_child(rvd, tvd);
1672			vdev_remove_child(mrvd, mtvd);
1673
1674			vdev_add_child(rvd, mtvd);
1675			vdev_add_child(mrvd, tvd);
1676
1677			spa_config_exit(spa, SCL_ALL, FTAG);
1678			vdev_load(mtvd);
1679			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1680
1681			vdev_reopen(rvd);
1682		} else if (mtvd->vdev_islog) {
1683			/*
1684			 * Load the slog device's state from the MOS config
1685			 * since it's possible that the label does not
1686			 * contain the most up-to-date information.
1687			 */
1688			vdev_load_log_state(tvd, mtvd);
1689			vdev_reopen(tvd);
1690		}
1691	}
1692	vdev_free(mrvd);
1693	spa_config_exit(spa, SCL_ALL, FTAG);
1694
1695	/*
1696	 * Ensure we were able to validate the config.
1697	 */
1698	return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1699}
1700
1701/*
1702 * Check for missing log devices
1703 */
1704static boolean_t
1705spa_check_logs(spa_t *spa)
1706{
1707	boolean_t rv = B_FALSE;
1708
1709	switch (spa->spa_log_state) {
1710	case SPA_LOG_MISSING:
1711		/* need to recheck in case slog has been restored */
1712	case SPA_LOG_UNKNOWN:
1713		rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1714		    NULL, DS_FIND_CHILDREN) != 0);
1715		if (rv)
1716			spa_set_log_state(spa, SPA_LOG_MISSING);
1717		break;
1718	}
1719	return (rv);
1720}
1721
1722static boolean_t
1723spa_passivate_log(spa_t *spa)
1724{
1725	vdev_t *rvd = spa->spa_root_vdev;
1726	boolean_t slog_found = B_FALSE;
1727
1728	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1729
1730	if (!spa_has_slogs(spa))
1731		return (B_FALSE);
1732
1733	for (int c = 0; c < rvd->vdev_children; c++) {
1734		vdev_t *tvd = rvd->vdev_child[c];
1735		metaslab_group_t *mg = tvd->vdev_mg;
1736
1737		if (tvd->vdev_islog) {
1738			metaslab_group_passivate(mg);
1739			slog_found = B_TRUE;
1740		}
1741	}
1742
1743	return (slog_found);
1744}
1745
1746static void
1747spa_activate_log(spa_t *spa)
1748{
1749	vdev_t *rvd = spa->spa_root_vdev;
1750
1751	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1752
1753	for (int c = 0; c < rvd->vdev_children; c++) {
1754		vdev_t *tvd = rvd->vdev_child[c];
1755		metaslab_group_t *mg = tvd->vdev_mg;
1756
1757		if (tvd->vdev_islog)
1758			metaslab_group_activate(mg);
1759	}
1760}
1761
1762int
1763spa_offline_log(spa_t *spa)
1764{
1765	int error;
1766
1767	error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1768	    NULL, DS_FIND_CHILDREN);
1769	if (error == 0) {
1770		/*
1771		 * We successfully offlined the log device, sync out the
1772		 * current txg so that the "stubby" block can be removed
1773		 * by zil_sync().
1774		 */
1775		txg_wait_synced(spa->spa_dsl_pool, 0);
1776	}
1777	return (error);
1778}
1779
1780static void
1781spa_aux_check_removed(spa_aux_vdev_t *sav)
1782{
1783	for (int i = 0; i < sav->sav_count; i++)
1784		spa_check_removed(sav->sav_vdevs[i]);
1785}
1786
1787void
1788spa_claim_notify(zio_t *zio)
1789{
1790	spa_t *spa = zio->io_spa;
1791
1792	if (zio->io_error)
1793		return;
1794
1795	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1796	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1797		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1798	mutex_exit(&spa->spa_props_lock);
1799}
1800
1801typedef struct spa_load_error {
1802	uint64_t	sle_meta_count;
1803	uint64_t	sle_data_count;
1804} spa_load_error_t;
1805
1806static void
1807spa_load_verify_done(zio_t *zio)
1808{
1809	blkptr_t *bp = zio->io_bp;
1810	spa_load_error_t *sle = zio->io_private;
1811	dmu_object_type_t type = BP_GET_TYPE(bp);
1812	int error = zio->io_error;
1813
1814	if (error) {
1815		if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1816		    type != DMU_OT_INTENT_LOG)
1817			atomic_add_64(&sle->sle_meta_count, 1);
1818		else
1819			atomic_add_64(&sle->sle_data_count, 1);
1820	}
1821	zio_data_buf_free(zio->io_data, zio->io_size);
1822}
1823
1824/*ARGSUSED*/
1825static int
1826spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1827    const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1828{
1829	if (bp != NULL) {
1830		zio_t *rio = arg;
1831		size_t size = BP_GET_PSIZE(bp);
1832		void *data = zio_data_buf_alloc(size);
1833
1834		zio_nowait(zio_read(rio, spa, bp, data, size,
1835		    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1836		    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1837		    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1838	}
1839	return (0);
1840}
1841
1842static int
1843spa_load_verify(spa_t *spa)
1844{
1845	zio_t *rio;
1846	spa_load_error_t sle = { 0 };
1847	zpool_rewind_policy_t policy;
1848	boolean_t verify_ok = B_FALSE;
1849	int error;
1850
1851	zpool_get_rewind_policy(spa->spa_config, &policy);
1852
1853	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1854		return (0);
1855
1856	rio = zio_root(spa, NULL, &sle,
1857	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1858
1859	error = traverse_pool(spa, spa->spa_verify_min_txg,
1860	    TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1861
1862	(void) zio_wait(rio);
1863
1864	spa->spa_load_meta_errors = sle.sle_meta_count;
1865	spa->spa_load_data_errors = sle.sle_data_count;
1866
1867	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1868	    sle.sle_data_count <= policy.zrp_maxdata) {
1869		int64_t loss = 0;
1870
1871		verify_ok = B_TRUE;
1872		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1873		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1874
1875		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1876		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1877		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1878		VERIFY(nvlist_add_int64(spa->spa_load_info,
1879		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1880		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1881		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1882	} else {
1883		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1884	}
1885
1886	if (error) {
1887		if (error != ENXIO && error != EIO)
1888			error = SET_ERROR(EIO);
1889		return (error);
1890	}
1891
1892	return (verify_ok ? 0 : EIO);
1893}
1894
1895/*
1896 * Find a value in the pool props object.
1897 */
1898static void
1899spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1900{
1901	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1902	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1903}
1904
1905/*
1906 * Find a value in the pool directory object.
1907 */
1908static int
1909spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1910{
1911	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1912	    name, sizeof (uint64_t), 1, val));
1913}
1914
1915static int
1916spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1917{
1918	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1919	return (err);
1920}
1921
1922/*
1923 * Fix up config after a partly-completed split.  This is done with the
1924 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1925 * pool have that entry in their config, but only the splitting one contains
1926 * a list of all the guids of the vdevs that are being split off.
1927 *
1928 * This function determines what to do with that list: either rejoin
1929 * all the disks to the pool, or complete the splitting process.  To attempt
1930 * the rejoin, each disk that is offlined is marked online again, and
1931 * we do a reopen() call.  If the vdev label for every disk that was
1932 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1933 * then we call vdev_split() on each disk, and complete the split.
1934 *
1935 * Otherwise we leave the config alone, with all the vdevs in place in
1936 * the original pool.
1937 */
1938static void
1939spa_try_repair(spa_t *spa, nvlist_t *config)
1940{
1941	uint_t extracted;
1942	uint64_t *glist;
1943	uint_t i, gcount;
1944	nvlist_t *nvl;
1945	vdev_t **vd;
1946	boolean_t attempt_reopen;
1947
1948	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1949		return;
1950
1951	/* check that the config is complete */
1952	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1953	    &glist, &gcount) != 0)
1954		return;
1955
1956	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1957
1958	/* attempt to online all the vdevs & validate */
1959	attempt_reopen = B_TRUE;
1960	for (i = 0; i < gcount; i++) {
1961		if (glist[i] == 0)	/* vdev is hole */
1962			continue;
1963
1964		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1965		if (vd[i] == NULL) {
1966			/*
1967			 * Don't bother attempting to reopen the disks;
1968			 * just do the split.
1969			 */
1970			attempt_reopen = B_FALSE;
1971		} else {
1972			/* attempt to re-online it */
1973			vd[i]->vdev_offline = B_FALSE;
1974		}
1975	}
1976
1977	if (attempt_reopen) {
1978		vdev_reopen(spa->spa_root_vdev);
1979
1980		/* check each device to see what state it's in */
1981		for (extracted = 0, i = 0; i < gcount; i++) {
1982			if (vd[i] != NULL &&
1983			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1984				break;
1985			++extracted;
1986		}
1987	}
1988
1989	/*
1990	 * If every disk has been moved to the new pool, or if we never
1991	 * even attempted to look at them, then we split them off for
1992	 * good.
1993	 */
1994	if (!attempt_reopen || gcount == extracted) {
1995		for (i = 0; i < gcount; i++)
1996			if (vd[i] != NULL)
1997				vdev_split(vd[i]);
1998		vdev_reopen(spa->spa_root_vdev);
1999	}
2000
2001	kmem_free(vd, gcount * sizeof (vdev_t *));
2002}
2003
2004static int
2005spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2006    boolean_t mosconfig)
2007{
2008	nvlist_t *config = spa->spa_config;
2009	char *ereport = FM_EREPORT_ZFS_POOL;
2010	char *comment;
2011	int error;
2012	uint64_t pool_guid;
2013	nvlist_t *nvl;
2014
2015	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2016		return (SET_ERROR(EINVAL));
2017
2018	ASSERT(spa->spa_comment == NULL);
2019	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2020		spa->spa_comment = spa_strdup(comment);
2021
2022	/*
2023	 * Versioning wasn't explicitly added to the label until later, so if
2024	 * it's not present treat it as the initial version.
2025	 */
2026	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2027	    &spa->spa_ubsync.ub_version) != 0)
2028		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2029
2030	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2031	    &spa->spa_config_txg);
2032
2033	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2034	    spa_guid_exists(pool_guid, 0)) {
2035		error = SET_ERROR(EEXIST);
2036	} else {
2037		spa->spa_config_guid = pool_guid;
2038
2039		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2040		    &nvl) == 0) {
2041			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2042			    KM_SLEEP) == 0);
2043		}
2044
2045		nvlist_free(spa->spa_load_info);
2046		spa->spa_load_info = fnvlist_alloc();
2047
2048		gethrestime(&spa->spa_loaded_ts);
2049		error = spa_load_impl(spa, pool_guid, config, state, type,
2050		    mosconfig, &ereport);
2051	}
2052
2053	spa->spa_minref = refcount_count(&spa->spa_refcount);
2054	if (error) {
2055		if (error != EEXIST) {
2056			spa->spa_loaded_ts.tv_sec = 0;
2057			spa->spa_loaded_ts.tv_nsec = 0;
2058		}
2059		if (error != EBADF) {
2060			zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2061		}
2062	}
2063	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2064	spa->spa_ena = 0;
2065
2066	return (error);
2067}
2068
2069/*
2070 * Load an existing storage pool, using the pool's builtin spa_config as a
2071 * source of configuration information.
2072 */
2073static int
2074spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2075    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2076    char **ereport)
2077{
2078	int error = 0;
2079	nvlist_t *nvroot = NULL;
2080	nvlist_t *label;
2081	vdev_t *rvd;
2082	uberblock_t *ub = &spa->spa_uberblock;
2083	uint64_t children, config_cache_txg = spa->spa_config_txg;
2084	int orig_mode = spa->spa_mode;
2085	int parse;
2086	uint64_t obj;
2087	boolean_t missing_feat_write = B_FALSE;
2088
2089	/*
2090	 * If this is an untrusted config, access the pool in read-only mode.
2091	 * This prevents things like resilvering recently removed devices.
2092	 */
2093	if (!mosconfig)
2094		spa->spa_mode = FREAD;
2095
2096	ASSERT(MUTEX_HELD(&spa_namespace_lock));
2097
2098	spa->spa_load_state = state;
2099
2100	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2101		return (SET_ERROR(EINVAL));
2102
2103	parse = (type == SPA_IMPORT_EXISTING ?
2104	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2105
2106	/*
2107	 * Create "The Godfather" zio to hold all async IOs
2108	 */
2109	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2110	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2111
2112	/*
2113	 * Parse the configuration into a vdev tree.  We explicitly set the
2114	 * value that will be returned by spa_version() since parsing the
2115	 * configuration requires knowing the version number.
2116	 */
2117	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2118	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2119	spa_config_exit(spa, SCL_ALL, FTAG);
2120
2121	if (error != 0)
2122		return (error);
2123
2124	ASSERT(spa->spa_root_vdev == rvd);
2125
2126	if (type != SPA_IMPORT_ASSEMBLE) {
2127		ASSERT(spa_guid(spa) == pool_guid);
2128	}
2129
2130	/*
2131	 * Try to open all vdevs, loading each label in the process.
2132	 */
2133	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2134	error = vdev_open(rvd);
2135	spa_config_exit(spa, SCL_ALL, FTAG);
2136	if (error != 0)
2137		return (error);
2138
2139	/*
2140	 * We need to validate the vdev labels against the configuration that
2141	 * we have in hand, which is dependent on the setting of mosconfig. If
2142	 * mosconfig is true then we're validating the vdev labels based on
2143	 * that config.  Otherwise, we're validating against the cached config
2144	 * (zpool.cache) that was read when we loaded the zfs module, and then
2145	 * later we will recursively call spa_load() and validate against
2146	 * the vdev config.
2147	 *
2148	 * If we're assembling a new pool that's been split off from an
2149	 * existing pool, the labels haven't yet been updated so we skip
2150	 * validation for now.
2151	 */
2152	if (type != SPA_IMPORT_ASSEMBLE) {
2153		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2154		error = vdev_validate(rvd, mosconfig);
2155		spa_config_exit(spa, SCL_ALL, FTAG);
2156
2157		if (error != 0)
2158			return (error);
2159
2160		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2161			return (SET_ERROR(ENXIO));
2162	}
2163
2164	/*
2165	 * Find the best uberblock.
2166	 */
2167	vdev_uberblock_load(rvd, ub, &label);
2168
2169	/*
2170	 * If we weren't able to find a single valid uberblock, return failure.
2171	 */
2172	if (ub->ub_txg == 0) {
2173		nvlist_free(label);
2174		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2175	}
2176
2177	/*
2178	 * If the pool has an unsupported version we can't open it.
2179	 */
2180	if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2181		nvlist_free(label);
2182		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2183	}
2184
2185	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2186		nvlist_t *features;
2187
2188		/*
2189		 * If we weren't able to find what's necessary for reading the
2190		 * MOS in the label, return failure.
2191		 */
2192		if (label == NULL || nvlist_lookup_nvlist(label,
2193		    ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2194			nvlist_free(label);
2195			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2196			    ENXIO));
2197		}
2198
2199		/*
2200		 * Update our in-core representation with the definitive values
2201		 * from the label.
2202		 */
2203		nvlist_free(spa->spa_label_features);
2204		VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2205	}
2206
2207	nvlist_free(label);
2208
2209	/*
2210	 * Look through entries in the label nvlist's features_for_read. If
2211	 * there is a feature listed there which we don't understand then we
2212	 * cannot open a pool.
2213	 */
2214	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2215		nvlist_t *unsup_feat;
2216
2217		VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2218		    0);
2219
2220		for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2221		    NULL); nvp != NULL;
2222		    nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2223			if (!zfeature_is_supported(nvpair_name(nvp))) {
2224				VERIFY(nvlist_add_string(unsup_feat,
2225				    nvpair_name(nvp), "") == 0);
2226			}
2227		}
2228
2229		if (!nvlist_empty(unsup_feat)) {
2230			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2231			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2232			nvlist_free(unsup_feat);
2233			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2234			    ENOTSUP));
2235		}
2236
2237		nvlist_free(unsup_feat);
2238	}
2239
2240	/*
2241	 * If the vdev guid sum doesn't match the uberblock, we have an
2242	 * incomplete configuration.  We first check to see if the pool
2243	 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2244	 * If it is, defer the vdev_guid_sum check till later so we
2245	 * can handle missing vdevs.
2246	 */
2247	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2248	    &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2249	    rvd->vdev_guid_sum != ub->ub_guid_sum)
2250		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2251
2252	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2253		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2254		spa_try_repair(spa, config);
2255		spa_config_exit(spa, SCL_ALL, FTAG);
2256		nvlist_free(spa->spa_config_splitting);
2257		spa->spa_config_splitting = NULL;
2258	}
2259
2260	/*
2261	 * Initialize internal SPA structures.
2262	 */
2263	spa->spa_state = POOL_STATE_ACTIVE;
2264	spa->spa_ubsync = spa->spa_uberblock;
2265	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2266	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2267	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2268	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2269	spa->spa_claim_max_txg = spa->spa_first_txg;
2270	spa->spa_prev_software_version = ub->ub_software_version;
2271
2272	error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2273	if (error)
2274		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2275	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2276
2277	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2278		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2279
2280	if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2281		boolean_t missing_feat_read = B_FALSE;
2282		nvlist_t *unsup_feat, *enabled_feat;
2283
2284		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2285		    &spa->spa_feat_for_read_obj) != 0) {
2286			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2287		}
2288
2289		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2290		    &spa->spa_feat_for_write_obj) != 0) {
2291			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2292		}
2293
2294		if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2295		    &spa->spa_feat_desc_obj) != 0) {
2296			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2297		}
2298
2299		enabled_feat = fnvlist_alloc();
2300		unsup_feat = fnvlist_alloc();
2301
2302		if (!feature_is_supported(spa->spa_meta_objset,
2303		    spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2304		    unsup_feat, enabled_feat))
2305			missing_feat_read = B_TRUE;
2306
2307		if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2308			if (!feature_is_supported(spa->spa_meta_objset,
2309			    spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2310			    unsup_feat, enabled_feat)) {
2311				missing_feat_write = B_TRUE;
2312			}
2313		}
2314
2315		fnvlist_add_nvlist(spa->spa_load_info,
2316		    ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2317
2318		if (!nvlist_empty(unsup_feat)) {
2319			fnvlist_add_nvlist(spa->spa_load_info,
2320			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2321		}
2322
2323		fnvlist_free(enabled_feat);
2324		fnvlist_free(unsup_feat);
2325
2326		if (!missing_feat_read) {
2327			fnvlist_add_boolean(spa->spa_load_info,
2328			    ZPOOL_CONFIG_CAN_RDONLY);
2329		}
2330
2331		/*
2332		 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2333		 * twofold: to determine whether the pool is available for
2334		 * import in read-write mode and (if it is not) whether the
2335		 * pool is available for import in read-only mode. If the pool
2336		 * is available for import in read-write mode, it is displayed
2337		 * as available in userland; if it is not available for import
2338		 * in read-only mode, it is displayed as unavailable in
2339		 * userland. If the pool is available for import in read-only
2340		 * mode but not read-write mode, it is displayed as unavailable
2341		 * in userland with a special note that the pool is actually
2342		 * available for open in read-only mode.
2343		 *
2344		 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2345		 * missing a feature for write, we must first determine whether
2346		 * the pool can be opened read-only before returning to
2347		 * userland in order to know whether to display the
2348		 * abovementioned note.
2349		 */
2350		if (missing_feat_read || (missing_feat_write &&
2351		    spa_writeable(spa))) {
2352			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2353			    ENOTSUP));
2354		}
2355	}
2356
2357	spa->spa_is_initializing = B_TRUE;
2358	error = dsl_pool_open(spa->spa_dsl_pool);
2359	spa->spa_is_initializing = B_FALSE;
2360	if (error != 0)
2361		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2362
2363	if (!mosconfig) {
2364		uint64_t hostid;
2365		nvlist_t *policy = NULL, *nvconfig;
2366
2367		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2368			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2369
2370		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2371		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2372			char *hostname;
2373			unsigned long myhostid = 0;
2374
2375			VERIFY(nvlist_lookup_string(nvconfig,
2376			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2377
2378#ifdef	_KERNEL
2379			myhostid = zone_get_hostid(NULL);
2380#else	/* _KERNEL */
2381			/*
2382			 * We're emulating the system's hostid in userland, so
2383			 * we can't use zone_get_hostid().
2384			 */
2385			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2386#endif	/* _KERNEL */
2387			if (hostid != 0 && myhostid != 0 &&
2388			    hostid != myhostid) {
2389				nvlist_free(nvconfig);
2390				cmn_err(CE_WARN, "pool '%s' could not be "
2391				    "loaded as it was last accessed by "
2392				    "another system (host: %s hostid: 0x%lx). "
2393				    "See: http://illumos.org/msg/ZFS-8000-EY",
2394				    spa_name(spa), hostname,
2395				    (unsigned long)hostid);
2396				return (SET_ERROR(EBADF));
2397			}
2398		}
2399		if (nvlist_lookup_nvlist(spa->spa_config,
2400		    ZPOOL_REWIND_POLICY, &policy) == 0)
2401			VERIFY(nvlist_add_nvlist(nvconfig,
2402			    ZPOOL_REWIND_POLICY, policy) == 0);
2403
2404		spa_config_set(spa, nvconfig);
2405		spa_unload(spa);
2406		spa_deactivate(spa);
2407		spa_activate(spa, orig_mode);
2408
2409		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2410	}
2411
2412	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2413		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2414	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2415	if (error != 0)
2416		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2417
2418	/*
2419	 * Load the bit that tells us to use the new accounting function
2420	 * (raid-z deflation).  If we have an older pool, this will not
2421	 * be present.
2422	 */
2423	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2424	if (error != 0 && error != ENOENT)
2425		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2426
2427	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2428	    &spa->spa_creation_version);
2429	if (error != 0 && error != ENOENT)
2430		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2431
2432	/*
2433	 * Load the persistent error log.  If we have an older pool, this will
2434	 * not be present.
2435	 */
2436	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2437	if (error != 0 && error != ENOENT)
2438		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2439
2440	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2441	    &spa->spa_errlog_scrub);
2442	if (error != 0 && error != ENOENT)
2443		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2444
2445	/*
2446	 * Load the history object.  If we have an older pool, this
2447	 * will not be present.
2448	 */
2449	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2450	if (error != 0 && error != ENOENT)
2451		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2452
2453	/*
2454	 * If we're assembling the pool from the split-off vdevs of
2455	 * an existing pool, we don't want to attach the spares & cache
2456	 * devices.
2457	 */
2458
2459	/*
2460	 * Load any hot spares for this pool.
2461	 */
2462	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2463	if (error != 0 && error != ENOENT)
2464		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2465	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2466		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2467		if (load_nvlist(spa, spa->spa_spares.sav_object,
2468		    &spa->spa_spares.sav_config) != 0)
2469			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2470
2471		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2472		spa_load_spares(spa);
2473		spa_config_exit(spa, SCL_ALL, FTAG);
2474	} else if (error == 0) {
2475		spa->spa_spares.sav_sync = B_TRUE;
2476	}
2477
2478	/*
2479	 * Load any level 2 ARC devices for this pool.
2480	 */
2481	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2482	    &spa->spa_l2cache.sav_object);
2483	if (error != 0 && error != ENOENT)
2484		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2485	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2486		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2487		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2488		    &spa->spa_l2cache.sav_config) != 0)
2489			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2490
2491		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2492		spa_load_l2cache(spa);
2493		spa_config_exit(spa, SCL_ALL, FTAG);
2494	} else if (error == 0) {
2495		spa->spa_l2cache.sav_sync = B_TRUE;
2496	}
2497
2498	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2499
2500	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2501	if (error && error != ENOENT)
2502		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2503
2504	if (error == 0) {
2505		uint64_t autoreplace;
2506
2507		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2508		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2509		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2510		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2511		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2512		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2513		    &spa->spa_dedup_ditto);
2514
2515		spa->spa_autoreplace = (autoreplace != 0);
2516	}
2517
2518	/*
2519	 * If the 'autoreplace' property is set, then post a resource notifying
2520	 * the ZFS DE that it should not issue any faults for unopenable
2521	 * devices.  We also iterate over the vdevs, and post a sysevent for any
2522	 * unopenable vdevs so that the normal autoreplace handler can take
2523	 * over.
2524	 */
2525	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2526		spa_check_removed(spa->spa_root_vdev);
2527		/*
2528		 * For the import case, this is done in spa_import(), because
2529		 * at this point we're using the spare definitions from
2530		 * the MOS config, not necessarily from the userland config.
2531		 */
2532		if (state != SPA_LOAD_IMPORT) {
2533			spa_aux_check_removed(&spa->spa_spares);
2534			spa_aux_check_removed(&spa->spa_l2cache);
2535		}
2536	}
2537
2538	/*
2539	 * Load the vdev state for all toplevel vdevs.
2540	 */
2541	vdev_load(rvd);
2542
2543	/*
2544	 * Propagate the leaf DTLs we just loaded all the way up the tree.
2545	 */
2546	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2547	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2548	spa_config_exit(spa, SCL_ALL, FTAG);
2549
2550	/*
2551	 * Load the DDTs (dedup tables).
2552	 */
2553	error = ddt_load(spa);
2554	if (error != 0)
2555		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2556
2557	spa_update_dspace(spa);
2558
2559	/*
2560	 * Validate the config, using the MOS config to fill in any
2561	 * information which might be missing.  If we fail to validate
2562	 * the config then declare the pool unfit for use. If we're
2563	 * assembling a pool from a split, the log is not transferred
2564	 * over.
2565	 */
2566	if (type != SPA_IMPORT_ASSEMBLE) {
2567		nvlist_t *nvconfig;
2568
2569		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2570			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2571
2572		if (!spa_config_valid(spa, nvconfig)) {
2573			nvlist_free(nvconfig);
2574			return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2575			    ENXIO));
2576		}
2577		nvlist_free(nvconfig);
2578
2579		/*
2580		 * Now that we've validated the config, check the state of the
2581		 * root vdev.  If it can't be opened, it indicates one or
2582		 * more toplevel vdevs are faulted.
2583		 */
2584		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2585			return (SET_ERROR(ENXIO));
2586
2587		if (spa_check_logs(spa)) {
2588			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2589			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2590		}
2591	}
2592
2593	if (missing_feat_write) {
2594		ASSERT(state == SPA_LOAD_TRYIMPORT);
2595
2596		/*
2597		 * At this point, we know that we can open the pool in
2598		 * read-only mode but not read-write mode. We now have enough
2599		 * information and can return to userland.
2600		 */
2601		return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2602	}
2603
2604	/*
2605	 * We've successfully opened the pool, verify that we're ready
2606	 * to start pushing transactions.
2607	 */
2608	if (state != SPA_LOAD_TRYIMPORT) {
2609		if (error = spa_load_verify(spa))
2610			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2611			    error));
2612	}
2613
2614	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2615	    spa->spa_load_max_txg == UINT64_MAX)) {
2616		dmu_tx_t *tx;
2617		int need_update = B_FALSE;
2618
2619		ASSERT(state != SPA_LOAD_TRYIMPORT);
2620
2621		/*
2622		 * Claim log blocks that haven't been committed yet.
2623		 * This must all happen in a single txg.
2624		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2625		 * invoked from zil_claim_log_block()'s i/o done callback.
2626		 * Price of rollback is that we abandon the log.
2627		 */
2628		spa->spa_claiming = B_TRUE;
2629
2630		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2631		    spa_first_txg(spa));
2632		(void) dmu_objset_find(spa_name(spa),
2633		    zil_claim, tx, DS_FIND_CHILDREN);
2634		dmu_tx_commit(tx);
2635
2636		spa->spa_claiming = B_FALSE;
2637
2638		spa_set_log_state(spa, SPA_LOG_GOOD);
2639		spa->spa_sync_on = B_TRUE;
2640		txg_sync_start(spa->spa_dsl_pool);
2641
2642		/*
2643		 * Wait for all claims to sync.  We sync up to the highest
2644		 * claimed log block birth time so that claimed log blocks
2645		 * don't appear to be from the future.  spa_claim_max_txg
2646		 * will have been set for us by either zil_check_log_chain()
2647		 * (invoked from spa_check_logs()) or zil_claim() above.
2648		 */
2649		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2650
2651		/*
2652		 * If the config cache is stale, or we have uninitialized
2653		 * metaslabs (see spa_vdev_add()), then update the config.
2654		 *
2655		 * If this is a verbatim import, trust the current
2656		 * in-core spa_config and update the disk labels.
2657		 */
2658		if (config_cache_txg != spa->spa_config_txg ||
2659		    state == SPA_LOAD_IMPORT ||
2660		    state == SPA_LOAD_RECOVER ||
2661		    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2662			need_update = B_TRUE;
2663
2664		for (int c = 0; c < rvd->vdev_children; c++)
2665			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2666				need_update = B_TRUE;
2667
2668		/*
2669		 * Update the config cache asychronously in case we're the
2670		 * root pool, in which case the config cache isn't writable yet.
2671		 */
2672		if (need_update)
2673			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2674
2675		/*
2676		 * Check all DTLs to see if anything needs resilvering.
2677		 */
2678		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2679		    vdev_resilver_needed(rvd, NULL, NULL))
2680			spa_async_request(spa, SPA_ASYNC_RESILVER);
2681
2682		/*
2683		 * Log the fact that we booted up (so that we can detect if
2684		 * we rebooted in the middle of an operation).
2685		 */
2686		spa_history_log_version(spa, "open");
2687
2688		/*
2689		 * Delete any inconsistent datasets.
2690		 */
2691		(void) dmu_objset_find(spa_name(spa),
2692		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2693
2694		/*
2695		 * Clean up any stale temporary dataset userrefs.
2696		 */
2697		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2698	}
2699
2700	return (0);
2701}
2702
2703static int
2704spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2705{
2706	int mode = spa->spa_mode;
2707
2708	spa_unload(spa);
2709	spa_deactivate(spa);
2710
2711	spa->spa_load_max_txg--;
2712
2713	spa_activate(spa, mode);
2714	spa_async_suspend(spa);
2715
2716	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2717}
2718
2719/*
2720 * If spa_load() fails this function will try loading prior txg's. If
2721 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2722 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2723 * function will not rewind the pool and will return the same error as
2724 * spa_load().
2725 */
2726static int
2727spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2728    uint64_t max_request, int rewind_flags)
2729{
2730	nvlist_t *loadinfo = NULL;
2731	nvlist_t *config = NULL;
2732	int load_error, rewind_error;
2733	uint64_t safe_rewind_txg;
2734	uint64_t min_txg;
2735
2736	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2737		spa->spa_load_max_txg = spa->spa_load_txg;
2738		spa_set_log_state(spa, SPA_LOG_CLEAR);
2739	} else {
2740		spa->spa_load_max_txg = max_request;
2741	}
2742
2743	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2744	    mosconfig);
2745	if (load_error == 0)
2746		return (0);
2747
2748	if (spa->spa_root_vdev != NULL)
2749		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2750
2751	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2752	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2753
2754	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2755		nvlist_free(config);
2756		return (load_error);
2757	}
2758
2759	if (state == SPA_LOAD_RECOVER) {
2760		/* Price of rolling back is discarding txgs, including log */
2761		spa_set_log_state(spa, SPA_LOG_CLEAR);
2762	} else {
2763		/*
2764		 * If we aren't rolling back save the load info from our first
2765		 * import attempt so that we can restore it after attempting
2766		 * to rewind.
2767		 */
2768		loadinfo = spa->spa_load_info;
2769		spa->spa_load_info = fnvlist_alloc();
2770	}
2771
2772	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2773	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2774	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2775	    TXG_INITIAL : safe_rewind_txg;
2776
2777	/*
2778	 * Continue as long as we're finding errors, we're still within
2779	 * the acceptable rewind range, and we're still finding uberblocks
2780	 */
2781	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2782	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2783		if (spa->spa_load_max_txg < safe_rewind_txg)
2784			spa->spa_extreme_rewind = B_TRUE;
2785		rewind_error = spa_load_retry(spa, state, mosconfig);
2786	}
2787
2788	spa->spa_extreme_rewind = B_FALSE;
2789	spa->spa_load_max_txg = UINT64_MAX;
2790
2791	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2792		spa_config_set(spa, config);
2793
2794	if (state == SPA_LOAD_RECOVER) {
2795		ASSERT3P(loadinfo, ==, NULL);
2796		return (rewind_error);
2797	} else {
2798		/* Store the rewind info as part of the initial load info */
2799		fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2800		    spa->spa_load_info);
2801
2802		/* Restore the initial load info */
2803		fnvlist_free(spa->spa_load_info);
2804		spa->spa_load_info = loadinfo;
2805
2806		return (load_error);
2807	}
2808}
2809
2810/*
2811 * Pool Open/Import
2812 *
2813 * The import case is identical to an open except that the configuration is sent
2814 * down from userland, instead of grabbed from the configuration cache.  For the
2815 * case of an open, the pool configuration will exist in the
2816 * POOL_STATE_UNINITIALIZED state.
2817 *
2818 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2819 * the same time open the pool, without having to keep around the spa_t in some
2820 * ambiguous state.
2821 */
2822static int
2823spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2824    nvlist_t **config)
2825{
2826	spa_t *spa;
2827	spa_load_state_t state = SPA_LOAD_OPEN;
2828	int error;
2829	int locked = B_FALSE;
2830
2831	*spapp = NULL;
2832
2833	/*
2834	 * As disgusting as this is, we need to support recursive calls to this
2835	 * function because dsl_dir_open() is called during spa_load(), and ends
2836	 * up calling spa_open() again.  The real fix is to figure out how to
2837	 * avoid dsl_dir_open() calling this in the first place.
2838	 */
2839	if (mutex_owner(&spa_namespace_lock) != curthread) {
2840		mutex_enter(&spa_namespace_lock);
2841		locked = B_TRUE;
2842	}
2843
2844	if ((spa = spa_lookup(pool)) == NULL) {
2845		if (locked)
2846			mutex_exit(&spa_namespace_lock);
2847		return (SET_ERROR(ENOENT));
2848	}
2849
2850	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2851		zpool_rewind_policy_t policy;
2852
2853		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2854		    &policy);
2855		if (policy.zrp_request & ZPOOL_DO_REWIND)
2856			state = SPA_LOAD_RECOVER;
2857
2858		spa_activate(spa, spa_mode_global);
2859
2860		if (state != SPA_LOAD_RECOVER)
2861			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2862
2863		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2864		    policy.zrp_request);
2865
2866		if (error == EBADF) {
2867			/*
2868			 * If vdev_validate() returns failure (indicated by
2869			 * EBADF), it indicates that one of the vdevs indicates
2870			 * that the pool has been exported or destroyed.  If
2871			 * this is the case, the config cache is out of sync and
2872			 * we should remove the pool from the namespace.
2873			 */
2874			spa_unload(spa);
2875			spa_deactivate(spa);
2876			spa_config_sync(spa, B_TRUE, B_TRUE);
2877			spa_remove(spa);
2878			if (locked)
2879				mutex_exit(&spa_namespace_lock);
2880			return (SET_ERROR(ENOENT));
2881		}
2882
2883		if (error) {
2884			/*
2885			 * We can't open the pool, but we still have useful
2886			 * information: the state of each vdev after the
2887			 * attempted vdev_open().  Return this to the user.
2888			 */
2889			if (config != NULL && spa->spa_config) {
2890				VERIFY(nvlist_dup(spa->spa_config, config,
2891				    KM_SLEEP) == 0);
2892				VERIFY(nvlist_add_nvlist(*config,
2893				    ZPOOL_CONFIG_LOAD_INFO,
2894				    spa->spa_load_info) == 0);
2895			}
2896			spa_unload(spa);
2897			spa_deactivate(spa);
2898			spa->spa_last_open_failed = error;
2899			if (locked)
2900				mutex_exit(&spa_namespace_lock);
2901			*spapp = NULL;
2902			return (error);
2903		}
2904	}
2905
2906	spa_open_ref(spa, tag);
2907
2908	if (config != NULL)
2909		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2910
2911	/*
2912	 * If we've recovered the pool, pass back any information we
2913	 * gathered while doing the load.
2914	 */
2915	if (state == SPA_LOAD_RECOVER) {
2916		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2917		    spa->spa_load_info) == 0);
2918	}
2919
2920	if (locked) {
2921		spa->spa_last_open_failed = 0;
2922		spa->spa_last_ubsync_txg = 0;
2923		spa->spa_load_txg = 0;
2924		mutex_exit(&spa_namespace_lock);
2925	}
2926
2927	*spapp = spa;
2928
2929	return (0);
2930}
2931
2932int
2933spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2934    nvlist_t **config)
2935{
2936	return (spa_open_common(name, spapp, tag, policy, config));
2937}
2938
2939int
2940spa_open(const char *name, spa_t **spapp, void *tag)
2941{
2942	return (spa_open_common(name, spapp, tag, NULL, NULL));
2943}
2944
2945/*
2946 * Lookup the given spa_t, incrementing the inject count in the process,
2947 * preventing it from being exported or destroyed.
2948 */
2949spa_t *
2950spa_inject_addref(char *name)
2951{
2952	spa_t *spa;
2953
2954	mutex_enter(&spa_namespace_lock);
2955	if ((spa = spa_lookup(name)) == NULL) {
2956		mutex_exit(&spa_namespace_lock);
2957		return (NULL);
2958	}
2959	spa->spa_inject_ref++;
2960	mutex_exit(&spa_namespace_lock);
2961
2962	return (spa);
2963}
2964
2965void
2966spa_inject_delref(spa_t *spa)
2967{
2968	mutex_enter(&spa_namespace_lock);
2969	spa->spa_inject_ref--;
2970	mutex_exit(&spa_namespace_lock);
2971}
2972
2973/*
2974 * Add spares device information to the nvlist.
2975 */
2976static void
2977spa_add_spares(spa_t *spa, nvlist_t *config)
2978{
2979	nvlist_t **spares;
2980	uint_t i, nspares;
2981	nvlist_t *nvroot;
2982	uint64_t guid;
2983	vdev_stat_t *vs;
2984	uint_t vsc;
2985	uint64_t pool;
2986
2987	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2988
2989	if (spa->spa_spares.sav_count == 0)
2990		return;
2991
2992	VERIFY(nvlist_lookup_nvlist(config,
2993	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2994	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2995	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2996	if (nspares != 0) {
2997		VERIFY(nvlist_add_nvlist_array(nvroot,
2998		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2999		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3000		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3001
3002		/*
3003		 * Go through and find any spares which have since been
3004		 * repurposed as an active spare.  If this is the case, update
3005		 * their status appropriately.
3006		 */
3007		for (i = 0; i < nspares; i++) {
3008			VERIFY(nvlist_lookup_uint64(spares[i],
3009			    ZPOOL_CONFIG_GUID, &guid) == 0);
3010			if (spa_spare_exists(guid, &pool, NULL) &&
3011			    pool != 0ULL) {
3012				VERIFY(nvlist_lookup_uint64_array(
3013				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
3014				    (uint64_t **)&vs, &vsc) == 0);
3015				vs->vs_state = VDEV_STATE_CANT_OPEN;
3016				vs->vs_aux = VDEV_AUX_SPARED;
3017			}
3018		}
3019	}
3020}
3021
3022/*
3023 * Add l2cache device information to the nvlist, including vdev stats.
3024 */
3025static void
3026spa_add_l2cache(spa_t *spa, nvlist_t *config)
3027{
3028	nvlist_t **l2cache;
3029	uint_t i, j, nl2cache;
3030	nvlist_t *nvroot;
3031	uint64_t guid;
3032	vdev_t *vd;
3033	vdev_stat_t *vs;
3034	uint_t vsc;
3035
3036	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3037
3038	if (spa->spa_l2cache.sav_count == 0)
3039		return;
3040
3041	VERIFY(nvlist_lookup_nvlist(config,
3042	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3043	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3044	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3045	if (nl2cache != 0) {
3046		VERIFY(nvlist_add_nvlist_array(nvroot,
3047		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3048		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3049		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3050
3051		/*
3052		 * Update level 2 cache device stats.
3053		 */
3054
3055		for (i = 0; i < nl2cache; i++) {
3056			VERIFY(nvlist_lookup_uint64(l2cache[i],
3057			    ZPOOL_CONFIG_GUID, &guid) == 0);
3058
3059			vd = NULL;
3060			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3061				if (guid ==
3062				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3063					vd = spa->spa_l2cache.sav_vdevs[j];
3064					break;
3065				}
3066			}
3067			ASSERT(vd != NULL);
3068
3069			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3070			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3071			    == 0);
3072			vdev_get_stats(vd, vs);
3073		}
3074	}
3075}
3076
3077static void
3078spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3079{
3080	nvlist_t *features;
3081	zap_cursor_t zc;
3082	zap_attribute_t za;
3083
3084	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3085	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3086
3087	if (spa->spa_feat_for_read_obj != 0) {
3088		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3089		    spa->spa_feat_for_read_obj);
3090		    zap_cursor_retrieve(&zc, &za) == 0;
3091		    zap_cursor_advance(&zc)) {
3092			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3093			    za.za_num_integers == 1);
3094			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3095			    za.za_first_integer));
3096		}
3097		zap_cursor_fini(&zc);
3098	}
3099
3100	if (spa->spa_feat_for_write_obj != 0) {
3101		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3102		    spa->spa_feat_for_write_obj);
3103		    zap_cursor_retrieve(&zc, &za) == 0;
3104		    zap_cursor_advance(&zc)) {
3105			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3106			    za.za_num_integers == 1);
3107			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3108			    za.za_first_integer));
3109		}
3110		zap_cursor_fini(&zc);
3111	}
3112
3113	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3114	    features) == 0);
3115	nvlist_free(features);
3116}
3117
3118int
3119spa_get_stats(const char *name, nvlist_t **config,
3120    char *altroot, size_t buflen)
3121{
3122	int error;
3123	spa_t *spa;
3124
3125	*config = NULL;
3126	error = spa_open_common(name, &spa, FTAG, NULL, config);
3127
3128	if (spa != NULL) {
3129		/*
3130		 * This still leaves a window of inconsistency where the spares
3131		 * or l2cache devices could change and the config would be
3132		 * self-inconsistent.
3133		 */
3134		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3135
3136		if (*config != NULL) {
3137			uint64_t loadtimes[2];
3138
3139			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3140			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3141			VERIFY(nvlist_add_uint64_array(*config,
3142			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3143
3144			VERIFY(nvlist_add_uint64(*config,
3145			    ZPOOL_CONFIG_ERRCOUNT,
3146			    spa_get_errlog_size(spa)) == 0);
3147
3148			if (spa_suspended(spa))
3149				VERIFY(nvlist_add_uint64(*config,
3150				    ZPOOL_CONFIG_SUSPENDED,
3151				    spa->spa_failmode) == 0);
3152
3153			spa_add_spares(spa, *config);
3154			spa_add_l2cache(spa, *config);
3155			spa_add_feature_stats(spa, *config);
3156		}
3157	}
3158
3159	/*
3160	 * We want to get the alternate root even for faulted pools, so we cheat
3161	 * and call spa_lookup() directly.
3162	 */
3163	if (altroot) {
3164		if (spa == NULL) {
3165			mutex_enter(&spa_namespace_lock);
3166			spa = spa_lookup(name);
3167			if (spa)
3168				spa_altroot(spa, altroot, buflen);
3169			else
3170				altroot[0] = '\0';
3171			spa = NULL;
3172			mutex_exit(&spa_namespace_lock);
3173		} else {
3174			spa_altroot(spa, altroot, buflen);
3175		}
3176	}
3177
3178	if (spa != NULL) {
3179		spa_config_exit(spa, SCL_CONFIG, FTAG);
3180		spa_close(spa, FTAG);
3181	}
3182
3183	return (error);
3184}
3185
3186/*
3187 * Validate that the auxiliary device array is well formed.  We must have an
3188 * array of nvlists, each which describes a valid leaf vdev.  If this is an
3189 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3190 * specified, as long as they are well-formed.
3191 */
3192static int
3193spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3194    spa_aux_vdev_t *sav, const char *config, uint64_t version,
3195    vdev_labeltype_t label)
3196{
3197	nvlist_t **dev;
3198	uint_t i, ndev;
3199	vdev_t *vd;
3200	int error;
3201
3202	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3203
3204	/*
3205	 * It's acceptable to have no devs specified.
3206	 */
3207	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3208		return (0);
3209
3210	if (ndev == 0)
3211		return (SET_ERROR(EINVAL));
3212
3213	/*
3214	 * Make sure the pool is formatted with a version that supports this
3215	 * device type.
3216	 */
3217	if (spa_version(spa) < version)
3218		return (SET_ERROR(ENOTSUP));
3219
3220	/*
3221	 * Set the pending device list so we correctly handle device in-use
3222	 * checking.
3223	 */
3224	sav->sav_pending = dev;
3225	sav->sav_npending = ndev;
3226
3227	for (i = 0; i < ndev; i++) {
3228		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3229		    mode)) != 0)
3230			goto out;
3231
3232		if (!vd->vdev_ops->vdev_op_leaf) {
3233			vdev_free(vd);
3234			error = SET_ERROR(EINVAL);
3235			goto out;
3236		}
3237
3238		/*
3239		 * The L2ARC currently only supports disk devices in
3240		 * kernel context.  For user-level testing, we allow it.
3241		 */
3242#ifdef _KERNEL
3243		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3244		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3245			error = SET_ERROR(ENOTBLK);
3246			vdev_free(vd);
3247			goto out;
3248		}
3249#endif
3250		vd->vdev_top = vd;
3251
3252		if ((error = vdev_open(vd)) == 0 &&
3253		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
3254			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3255			    vd->vdev_guid) == 0);
3256		}
3257
3258		vdev_free(vd);
3259
3260		if (error &&
3261		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3262			goto out;
3263		else
3264			error = 0;
3265	}
3266
3267out:
3268	sav->sav_pending = NULL;
3269	sav->sav_npending = 0;
3270	return (error);
3271}
3272
3273static int
3274spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3275{
3276	int error;
3277
3278	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3279
3280	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3281	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3282	    VDEV_LABEL_SPARE)) != 0) {
3283		return (error);
3284	}
3285
3286	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3287	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3288	    VDEV_LABEL_L2CACHE));
3289}
3290
3291static void
3292spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3293    const char *config)
3294{
3295	int i;
3296
3297	if (sav->sav_config != NULL) {
3298		nvlist_t **olddevs;
3299		uint_t oldndevs;
3300		nvlist_t **newdevs;
3301
3302		/*
3303		 * Generate new dev list by concatentating with the
3304		 * current dev list.
3305		 */
3306		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3307		    &olddevs, &oldndevs) == 0);
3308
3309		newdevs = kmem_alloc(sizeof (void *) *
3310		    (ndevs + oldndevs), KM_SLEEP);
3311		for (i = 0; i < oldndevs; i++)
3312			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3313			    KM_SLEEP) == 0);
3314		for (i = 0; i < ndevs; i++)
3315			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3316			    KM_SLEEP) == 0);
3317
3318		VERIFY(nvlist_remove(sav->sav_config, config,
3319		    DATA_TYPE_NVLIST_ARRAY) == 0);
3320
3321		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3322		    config, newdevs, ndevs + oldndevs) == 0);
3323		for (i = 0; i < oldndevs + ndevs; i++)
3324			nvlist_free(newdevs[i]);
3325		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3326	} else {
3327		/*
3328		 * Generate a new dev list.
3329		 */
3330		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3331		    KM_SLEEP) == 0);
3332		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3333		    devs, ndevs) == 0);
3334	}
3335}
3336
3337/*
3338 * Stop and drop level 2 ARC devices
3339 */
3340void
3341spa_l2cache_drop(spa_t *spa)
3342{
3343	vdev_t *vd;
3344	int i;
3345	spa_aux_vdev_t *sav = &spa->spa_l2cache;
3346
3347	for (i = 0; i < sav->sav_count; i++) {
3348		uint64_t pool;
3349
3350		vd = sav->sav_vdevs[i];
3351		ASSERT(vd != NULL);
3352
3353		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3354		    pool != 0ULL && l2arc_vdev_present(vd))
3355			l2arc_remove_vdev(vd);
3356	}
3357}
3358
3359/*
3360 * Pool Creation
3361 */
3362int
3363spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3364    nvlist_t *zplprops)
3365{
3366	spa_t *spa;
3367	char *altroot = NULL;
3368	vdev_t *rvd;
3369	dsl_pool_t *dp;
3370	dmu_tx_t *tx;
3371	int error = 0;
3372	uint64_t txg = TXG_INITIAL;
3373	nvlist_t **spares, **l2cache;
3374	uint_t nspares, nl2cache;
3375	uint64_t version, obj;
3376	boolean_t has_features;
3377
3378	/*
3379	 * If this pool already exists, return failure.
3380	 */
3381	mutex_enter(&spa_namespace_lock);
3382	if (spa_lookup(pool) != NULL) {
3383		mutex_exit(&spa_namespace_lock);
3384		return (SET_ERROR(EEXIST));
3385	}
3386
3387	/*
3388	 * Allocate a new spa_t structure.
3389	 */
3390	(void) nvlist_lookup_string(props,
3391	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3392	spa = spa_add(pool, NULL, altroot);
3393	spa_activate(spa, spa_mode_global);
3394
3395	if (props && (error = spa_prop_validate(spa, props))) {
3396		spa_deactivate(spa);
3397		spa_remove(spa);
3398		mutex_exit(&spa_namespace_lock);
3399		return (error);
3400	}
3401
3402	has_features = B_FALSE;
3403	for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3404	    elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3405		if (zpool_prop_feature(nvpair_name(elem)))
3406			has_features = B_TRUE;
3407	}
3408
3409	if (has_features || nvlist_lookup_uint64(props,
3410	    zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3411		version = SPA_VERSION;
3412	}
3413	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3414
3415	spa->spa_first_txg = txg;
3416	spa->spa_uberblock.ub_txg = txg - 1;
3417	spa->spa_uberblock.ub_version = version;
3418	spa->spa_ubsync = spa->spa_uberblock;
3419
3420	/*
3421	 * Create "The Godfather" zio to hold all async IOs
3422	 */
3423	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3424	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3425
3426	/*
3427	 * Create the root vdev.
3428	 */
3429	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3430
3431	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3432
3433	ASSERT(error != 0 || rvd != NULL);
3434	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3435
3436	if (error == 0 && !zfs_allocatable_devs(nvroot))
3437		error = SET_ERROR(EINVAL);
3438
3439	if (error == 0 &&
3440	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3441	    (error = spa_validate_aux(spa, nvroot, txg,
3442	    VDEV_ALLOC_ADD)) == 0) {
3443		for (int c = 0; c < rvd->vdev_children; c++) {
3444			vdev_metaslab_set_size(rvd->vdev_child[c]);
3445			vdev_expand(rvd->vdev_child[c], txg);
3446		}
3447	}
3448
3449	spa_config_exit(spa, SCL_ALL, FTAG);
3450
3451	if (error != 0) {
3452		spa_unload(spa);
3453		spa_deactivate(spa);
3454		spa_remove(spa);
3455		mutex_exit(&spa_namespace_lock);
3456		return (error);
3457	}
3458
3459	/*
3460	 * Get the list of spares, if specified.
3461	 */
3462	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3463	    &spares, &nspares) == 0) {
3464		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3465		    KM_SLEEP) == 0);
3466		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3467		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3468		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3469		spa_load_spares(spa);
3470		spa_config_exit(spa, SCL_ALL, FTAG);
3471		spa->spa_spares.sav_sync = B_TRUE;
3472	}
3473
3474	/*
3475	 * Get the list of level 2 cache devices, if specified.
3476	 */
3477	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3478	    &l2cache, &nl2cache) == 0) {
3479		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3480		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3481		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3482		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3483		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3484		spa_load_l2cache(spa);
3485		spa_config_exit(spa, SCL_ALL, FTAG);
3486		spa->spa_l2cache.sav_sync = B_TRUE;
3487	}
3488
3489	spa->spa_is_initializing = B_TRUE;
3490	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3491	spa->spa_meta_objset = dp->dp_meta_objset;
3492	spa->spa_is_initializing = B_FALSE;
3493
3494	/*
3495	 * Create DDTs (dedup tables).
3496	 */
3497	ddt_create(spa);
3498
3499	spa_update_dspace(spa);
3500
3501	tx = dmu_tx_create_assigned(dp, txg);
3502
3503	/*
3504	 * Create the pool config object.
3505	 */
3506	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3507	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3508	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3509
3510	if (zap_add(spa->spa_meta_objset,
3511	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3512	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3513		cmn_err(CE_PANIC, "failed to add pool config");
3514	}
3515
3516	if (spa_version(spa) >= SPA_VERSION_FEATURES)
3517		spa_feature_create_zap_objects(spa, tx);
3518
3519	if (zap_add(spa->spa_meta_objset,
3520	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3521	    sizeof (uint64_t), 1, &version, tx) != 0) {
3522		cmn_err(CE_PANIC, "failed to add pool version");
3523	}
3524
3525	/* Newly created pools with the right version are always deflated. */
3526	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3527		spa->spa_deflate = TRUE;
3528		if (zap_add(spa->spa_meta_objset,
3529		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3530		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3531			cmn_err(CE_PANIC, "failed to add deflate");
3532		}
3533	}
3534
3535	/*
3536	 * Create the deferred-free bpobj.  Turn off compression
3537	 * because sync-to-convergence takes longer if the blocksize
3538	 * keeps changing.
3539	 */
3540	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3541	dmu_object_set_compress(spa->spa_meta_objset, obj,
3542	    ZIO_COMPRESS_OFF, tx);
3543	if (zap_add(spa->spa_meta_objset,
3544	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3545	    sizeof (uint64_t), 1, &obj, tx) != 0) {
3546		cmn_err(CE_PANIC, "failed to add bpobj");
3547	}
3548	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3549	    spa->spa_meta_objset, obj));
3550
3551	/*
3552	 * Create the pool's history object.
3553	 */
3554	if (version >= SPA_VERSION_ZPOOL_HISTORY)
3555		spa_history_create_obj(spa, tx);
3556
3557	/*
3558	 * Set pool properties.
3559	 */
3560	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3561	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3562	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3563	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3564
3565	if (props != NULL) {
3566		spa_configfile_set(spa, props, B_FALSE);
3567		spa_sync_props(props, tx);
3568	}
3569
3570	dmu_tx_commit(tx);
3571
3572	spa->spa_sync_on = B_TRUE;
3573	txg_sync_start(spa->spa_dsl_pool);
3574
3575	/*
3576	 * We explicitly wait for the first transaction to complete so that our
3577	 * bean counters are appropriately updated.
3578	 */
3579	txg_wait_synced(spa->spa_dsl_pool, txg);
3580
3581	spa_config_sync(spa, B_FALSE, B_TRUE);
3582
3583	spa_history_log_version(spa, "create");
3584
3585	spa->spa_minref = refcount_count(&spa->spa_refcount);
3586
3587	mutex_exit(&spa_namespace_lock);
3588
3589	return (0);
3590}
3591
3592#ifdef _KERNEL
3593/*
3594 * Get the root pool information from the root disk, then import the root pool
3595 * during the system boot up time.
3596 */
3597extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3598
3599static nvlist_t *
3600spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3601{
3602	nvlist_t *config;
3603	nvlist_t *nvtop, *nvroot;
3604	uint64_t pgid;
3605
3606	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3607		return (NULL);
3608
3609	/*
3610	 * Add this top-level vdev to the child array.
3611	 */
3612	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3613	    &nvtop) == 0);
3614	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3615	    &pgid) == 0);
3616	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3617
3618	/*
3619	 * Put this pool's top-level vdevs into a root vdev.
3620	 */
3621	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3622	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3623	    VDEV_TYPE_ROOT) == 0);
3624	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3625	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3626	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3627	    &nvtop, 1) == 0);
3628
3629	/*
3630	 * Replace the existing vdev_tree with the new root vdev in
3631	 * this pool's configuration (remove the old, add the new).
3632	 */
3633	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3634	nvlist_free(nvroot);
3635	return (config);
3636}
3637
3638/*
3639 * Walk the vdev tree and see if we can find a device with "better"
3640 * configuration. A configuration is "better" if the label on that
3641 * device has a more recent txg.
3642 */
3643static void
3644spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3645{
3646	for (int c = 0; c < vd->vdev_children; c++)
3647		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3648
3649	if (vd->vdev_ops->vdev_op_leaf) {
3650		nvlist_t *label;
3651		uint64_t label_txg;
3652
3653		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3654		    &label) != 0)
3655			return;
3656
3657		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3658		    &label_txg) == 0);
3659
3660		/*
3661		 * Do we have a better boot device?
3662		 */
3663		if (label_txg > *txg) {
3664			*txg = label_txg;
3665			*avd = vd;
3666		}
3667		nvlist_free(label);
3668	}
3669}
3670
3671/*
3672 * Import a root pool.
3673 *
3674 * For x86. devpath_list will consist of devid and/or physpath name of
3675 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3676 * The GRUB "findroot" command will return the vdev we should boot.
3677 *
3678 * For Sparc, devpath_list consists the physpath name of the booting device
3679 * no matter the rootpool is a single device pool or a mirrored pool.
3680 * e.g.
3681 *	"/pci@1f,0/ide@d/disk@0,0:a"
3682 */
3683int
3684spa_import_rootpool(char *devpath, char *devid)
3685{
3686	spa_t *spa;
3687	vdev_t *rvd, *bvd, *avd = NULL;
3688	nvlist_t *config, *nvtop;
3689	uint64_t guid, txg;
3690	char *pname;
3691	int error;
3692
3693	/*
3694	 * Read the label from the boot device and generate a configuration.
3695	 */
3696	config = spa_generate_rootconf(devpath, devid, &guid);
3697#if defined(_OBP) && defined(_KERNEL)
3698	if (config == NULL) {
3699		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3700			/* iscsi boot */
3701			get_iscsi_bootpath_phy(devpath);
3702			config = spa_generate_rootconf(devpath, devid, &guid);
3703		}
3704	}
3705#endif
3706	if (config == NULL) {
3707		cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3708		    devpath);
3709		return (SET_ERROR(EIO));
3710	}
3711
3712	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3713	    &pname) == 0);
3714	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3715
3716	mutex_enter(&spa_namespace_lock);
3717	if ((spa = spa_lookup(pname)) != NULL) {
3718		/*
3719		 * Remove the existing root pool from the namespace so that we
3720		 * can replace it with the correct config we just read in.
3721		 */
3722		spa_remove(spa);
3723	}
3724
3725	spa = spa_add(pname, config, NULL);
3726	spa->spa_is_root = B_TRUE;
3727	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3728
3729	/*
3730	 * Build up a vdev tree based on the boot device's label config.
3731	 */
3732	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3733	    &nvtop) == 0);
3734	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3735	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3736	    VDEV_ALLOC_ROOTPOOL);
3737	spa_config_exit(spa, SCL_ALL, FTAG);
3738	if (error) {
3739		mutex_exit(&spa_namespace_lock);
3740		nvlist_free(config);
3741		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3742		    pname);
3743		return (error);
3744	}
3745
3746	/*
3747	 * Get the boot vdev.
3748	 */
3749	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3750		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3751		    (u_longlong_t)guid);
3752		error = SET_ERROR(ENOENT);
3753		goto out;
3754	}
3755
3756	/*
3757	 * Determine if there is a better boot device.
3758	 */
3759	avd = bvd;
3760	spa_alt_rootvdev(rvd, &avd, &txg);
3761	if (avd != bvd) {
3762		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3763		    "try booting from '%s'", avd->vdev_path);
3764		error = SET_ERROR(EINVAL);
3765		goto out;
3766	}
3767
3768	/*
3769	 * If the boot device is part of a spare vdev then ensure that
3770	 * we're booting off the active spare.
3771	 */
3772	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3773	    !bvd->vdev_isspare) {
3774		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3775		    "try booting from '%s'",
3776		    bvd->vdev_parent->
3777		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3778		error = SET_ERROR(EINVAL);
3779		goto out;
3780	}
3781
3782	error = 0;
3783out:
3784	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3785	vdev_free(rvd);
3786	spa_config_exit(spa, SCL_ALL, FTAG);
3787	mutex_exit(&spa_namespace_lock);
3788
3789	nvlist_free(config);
3790	return (error);
3791}
3792
3793#endif
3794
3795/*
3796 * Import a non-root pool into the system.
3797 */
3798int
3799spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3800{
3801	spa_t *spa;
3802	char *altroot = NULL;
3803	spa_load_state_t state = SPA_LOAD_IMPORT;
3804	zpool_rewind_policy_t policy;
3805	uint64_t mode = spa_mode_global;
3806	uint64_t readonly = B_FALSE;
3807	int error;
3808	nvlist_t *nvroot;
3809	nvlist_t **spares, **l2cache;
3810	uint_t nspares, nl2cache;
3811
3812	/*
3813	 * If a pool with this name exists, return failure.
3814	 */
3815	mutex_enter(&spa_namespace_lock);
3816	if (spa_lookup(pool) != NULL) {
3817		mutex_exit(&spa_namespace_lock);
3818		return (SET_ERROR(EEXIST));
3819	}
3820
3821	/*
3822	 * Create and initialize the spa structure.
3823	 */
3824	(void) nvlist_lookup_string(props,
3825	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3826	(void) nvlist_lookup_uint64(props,
3827	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3828	if (readonly)
3829		mode = FREAD;
3830	spa = spa_add(pool, config, altroot);
3831	spa->spa_import_flags = flags;
3832
3833	/*
3834	 * Verbatim import - Take a pool and insert it into the namespace
3835	 * as if it had been loaded at boot.
3836	 */
3837	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3838		if (props != NULL)
3839			spa_configfile_set(spa, props, B_FALSE);
3840
3841		spa_config_sync(spa, B_FALSE, B_TRUE);
3842
3843		mutex_exit(&spa_namespace_lock);
3844		spa_history_log_version(spa, "import");
3845
3846		return (0);
3847	}
3848
3849	spa_activate(spa, mode);
3850
3851	/*
3852	 * Don't start async tasks until we know everything is healthy.
3853	 */
3854	spa_async_suspend(spa);
3855
3856	zpool_get_rewind_policy(config, &policy);
3857	if (policy.zrp_request & ZPOOL_DO_REWIND)
3858		state = SPA_LOAD_RECOVER;
3859
3860	/*
3861	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3862	 * because the user-supplied config is actually the one to trust when
3863	 * doing an import.
3864	 */
3865	if (state != SPA_LOAD_RECOVER)
3866		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3867
3868	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3869	    policy.zrp_request);
3870
3871	/*
3872	 * Propagate anything learned while loading the pool and pass it
3873	 * back to caller (i.e. rewind info, missing devices, etc).
3874	 */
3875	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3876	    spa->spa_load_info) == 0);
3877
3878	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3879	/*
3880	 * Toss any existing sparelist, as it doesn't have any validity
3881	 * anymore, and conflicts with spa_has_spare().
3882	 */
3883	if (spa->spa_spares.sav_config) {
3884		nvlist_free(spa->spa_spares.sav_config);
3885		spa->spa_spares.sav_config = NULL;
3886		spa_load_spares(spa);
3887	}
3888	if (spa->spa_l2cache.sav_config) {
3889		nvlist_free(spa->spa_l2cache.sav_config);
3890		spa->spa_l2cache.sav_config = NULL;
3891		spa_load_l2cache(spa);
3892	}
3893
3894	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3895	    &nvroot) == 0);
3896	if (error == 0)
3897		error = spa_validate_aux(spa, nvroot, -1ULL,
3898		    VDEV_ALLOC_SPARE);
3899	if (error == 0)
3900		error = spa_validate_aux(spa, nvroot, -1ULL,
3901		    VDEV_ALLOC_L2CACHE);
3902	spa_config_exit(spa, SCL_ALL, FTAG);
3903
3904	if (props != NULL)
3905		spa_configfile_set(spa, props, B_FALSE);
3906
3907	if (error != 0 || (props && spa_writeable(spa) &&
3908	    (error = spa_prop_set(spa, props)))) {
3909		spa_unload(spa);
3910		spa_deactivate(spa);
3911		spa_remove(spa);
3912		mutex_exit(&spa_namespace_lock);
3913		return (error);
3914	}
3915
3916	spa_async_resume(spa);
3917
3918	/*
3919	 * Override any spares and level 2 cache devices as specified by
3920	 * the user, as these may have correct device names/devids, etc.
3921	 */
3922	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3923	    &spares, &nspares) == 0) {
3924		if (spa->spa_spares.sav_config)
3925			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3926			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3927		else
3928			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3929			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3930		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3931		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3932		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3933		spa_load_spares(spa);
3934		spa_config_exit(spa, SCL_ALL, FTAG);
3935		spa->spa_spares.sav_sync = B_TRUE;
3936	}
3937	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3938	    &l2cache, &nl2cache) == 0) {
3939		if (spa->spa_l2cache.sav_config)
3940			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3941			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3942		else
3943			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3944			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3945		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3946		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3947		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3948		spa_load_l2cache(spa);
3949		spa_config_exit(spa, SCL_ALL, FTAG);
3950		spa->spa_l2cache.sav_sync = B_TRUE;
3951	}
3952
3953	/*
3954	 * Check for any removed devices.
3955	 */
3956	if (spa->spa_autoreplace) {
3957		spa_aux_check_removed(&spa->spa_spares);
3958		spa_aux_check_removed(&spa->spa_l2cache);
3959	}
3960
3961	if (spa_writeable(spa)) {
3962		/*
3963		 * Update the config cache to include the newly-imported pool.
3964		 */
3965		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3966	}
3967
3968	/*
3969	 * It's possible that the pool was expanded while it was exported.
3970	 * We kick off an async task to handle this for us.
3971	 */
3972	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3973
3974	mutex_exit(&spa_namespace_lock);
3975	spa_history_log_version(spa, "import");
3976
3977	return (0);
3978}
3979
3980nvlist_t *
3981spa_tryimport(nvlist_t *tryconfig)
3982{
3983	nvlist_t *config = NULL;
3984	char *poolname;
3985	spa_t *spa;
3986	uint64_t state;
3987	int error;
3988
3989	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3990		return (NULL);
3991
3992	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3993		return (NULL);
3994
3995	/*
3996	 * Create and initialize the spa structure.
3997	 */
3998	mutex_enter(&spa_namespace_lock);
3999	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4000	spa_activate(spa, FREAD);
4001
4002	/*
4003	 * Pass off the heavy lifting to spa_load().
4004	 * Pass TRUE for mosconfig because the user-supplied config
4005	 * is actually the one to trust when doing an import.
4006	 */
4007	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4008
4009	/*
4010	 * If 'tryconfig' was at least parsable, return the current config.
4011	 */
4012	if (spa->spa_root_vdev != NULL) {
4013		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4014		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4015		    poolname) == 0);
4016		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4017		    state) == 0);
4018		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4019		    spa->spa_uberblock.ub_timestamp) == 0);
4020		VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4021		    spa->spa_load_info) == 0);
4022
4023		/*
4024		 * If the bootfs property exists on this pool then we
4025		 * copy it out so that external consumers can tell which
4026		 * pools are bootable.
4027		 */
4028		if ((!error || error == EEXIST) && spa->spa_bootfs) {
4029			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4030
4031			/*
4032			 * We have to play games with the name since the
4033			 * pool was opened as TRYIMPORT_NAME.
4034			 */
4035			if (dsl_dsobj_to_dsname(spa_name(spa),
4036			    spa->spa_bootfs, tmpname) == 0) {
4037				char *cp;
4038				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4039
4040				cp = strchr(tmpname, '/');
4041				if (cp == NULL) {
4042					(void) strlcpy(dsname, tmpname,
4043					    MAXPATHLEN);
4044				} else {
4045					(void) snprintf(dsname, MAXPATHLEN,
4046					    "%s/%s", poolname, ++cp);
4047				}
4048				VERIFY(nvlist_add_string(config,
4049				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4050				kmem_free(dsname, MAXPATHLEN);
4051			}
4052			kmem_free(tmpname, MAXPATHLEN);
4053		}
4054
4055		/*
4056		 * Add the list of hot spares and level 2 cache devices.
4057		 */
4058		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4059		spa_add_spares(spa, config);
4060		spa_add_l2cache(spa, config);
4061		spa_config_exit(spa, SCL_CONFIG, FTAG);
4062	}
4063
4064	spa_unload(spa);
4065	spa_deactivate(spa);
4066	spa_remove(spa);
4067	mutex_exit(&spa_namespace_lock);
4068
4069	return (config);
4070}
4071
4072/*
4073 * Pool export/destroy
4074 *
4075 * The act of destroying or exporting a pool is very simple.  We make sure there
4076 * is no more pending I/O and any references to the pool are gone.  Then, we
4077 * update the pool state and sync all the labels to disk, removing the
4078 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4079 * we don't sync the labels or remove the configuration cache.
4080 */
4081static int
4082spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4083    boolean_t force, boolean_t hardforce)
4084{
4085	spa_t *spa;
4086
4087	if (oldconfig)
4088		*oldconfig = NULL;
4089
4090	if (!(spa_mode_global & FWRITE))
4091		return (SET_ERROR(EROFS));
4092
4093	mutex_enter(&spa_namespace_lock);
4094	if ((spa = spa_lookup(pool)) == NULL) {
4095		mutex_exit(&spa_namespace_lock);
4096		return (SET_ERROR(ENOENT));
4097	}
4098
4099	/*
4100	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4101	 * reacquire the namespace lock, and see if we can export.
4102	 */
4103	spa_open_ref(spa, FTAG);
4104	mutex_exit(&spa_namespace_lock);
4105	spa_async_suspend(spa);
4106	mutex_enter(&spa_namespace_lock);
4107	spa_close(spa, FTAG);
4108
4109	/*
4110	 * The pool will be in core if it's openable,
4111	 * in which case we can modify its state.
4112	 */
4113	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4114		/*
4115		 * Objsets may be open only because they're dirty, so we
4116		 * have to force it to sync before checking spa_refcnt.
4117		 */
4118		txg_wait_synced(spa->spa_dsl_pool, 0);
4119
4120		/*
4121		 * A pool cannot be exported or destroyed if there are active
4122		 * references.  If we are resetting a pool, allow references by
4123		 * fault injection handlers.
4124		 */
4125		if (!spa_refcount_zero(spa) ||
4126		    (spa->spa_inject_ref != 0 &&
4127		    new_state != POOL_STATE_UNINITIALIZED)) {
4128			spa_async_resume(spa);
4129			mutex_exit(&spa_namespace_lock);
4130			return (SET_ERROR(EBUSY));
4131		}
4132
4133		/*
4134		 * A pool cannot be exported if it has an active shared spare.
4135		 * This is to prevent other pools stealing the active spare
4136		 * from an exported pool. At user's own will, such pool can
4137		 * be forcedly exported.
4138		 */
4139		if (!force && new_state == POOL_STATE_EXPORTED &&
4140		    spa_has_active_shared_spare(spa)) {
4141			spa_async_resume(spa);
4142			mutex_exit(&spa_namespace_lock);
4143			return (SET_ERROR(EXDEV));
4144		}
4145
4146		/*
4147		 * We want this to be reflected on every label,
4148		 * so mark them all dirty.  spa_unload() will do the
4149		 * final sync that pushes these changes out.
4150		 */
4151		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4152			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4153			spa->spa_state = new_state;
4154			spa->spa_final_txg = spa_last_synced_txg(spa) +
4155			    TXG_DEFER_SIZE + 1;
4156			vdev_config_dirty(spa->spa_root_vdev);
4157			spa_config_exit(spa, SCL_ALL, FTAG);
4158		}
4159	}
4160
4161	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4162
4163	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4164		spa_unload(spa);
4165		spa_deactivate(spa);
4166	}
4167
4168	if (oldconfig && spa->spa_config)
4169		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4170
4171	if (new_state != POOL_STATE_UNINITIALIZED) {
4172		if (!hardforce)
4173			spa_config_sync(spa, B_TRUE, B_TRUE);
4174		spa_remove(spa);
4175	}
4176	mutex_exit(&spa_namespace_lock);
4177
4178	return (0);
4179}
4180
4181/*
4182 * Destroy a storage pool.
4183 */
4184int
4185spa_destroy(char *pool)
4186{
4187	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4188	    B_FALSE, B_FALSE));
4189}
4190
4191/*
4192 * Export a storage pool.
4193 */
4194int
4195spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4196    boolean_t hardforce)
4197{
4198	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4199	    force, hardforce));
4200}
4201
4202/*
4203 * Similar to spa_export(), this unloads the spa_t without actually removing it
4204 * from the namespace in any way.
4205 */
4206int
4207spa_reset(char *pool)
4208{
4209	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4210	    B_FALSE, B_FALSE));
4211}
4212
4213/*
4214 * ==========================================================================
4215 * Device manipulation
4216 * ==========================================================================
4217 */
4218
4219/*
4220 * Add a device to a storage pool.
4221 */
4222int
4223spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4224{
4225	uint64_t txg, id;
4226	int error;
4227	vdev_t *rvd = spa->spa_root_vdev;
4228	vdev_t *vd, *tvd;
4229	nvlist_t **spares, **l2cache;
4230	uint_t nspares, nl2cache;
4231
4232	ASSERT(spa_writeable(spa));
4233
4234	txg = spa_vdev_enter(spa);
4235
4236	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4237	    VDEV_ALLOC_ADD)) != 0)
4238		return (spa_vdev_exit(spa, NULL, txg, error));
4239
4240	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
4241
4242	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4243	    &nspares) != 0)
4244		nspares = 0;
4245
4246	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4247	    &nl2cache) != 0)
4248		nl2cache = 0;
4249
4250	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4251		return (spa_vdev_exit(spa, vd, txg, EINVAL));
4252
4253	if (vd->vdev_children != 0 &&
4254	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
4255		return (spa_vdev_exit(spa, vd, txg, error));
4256
4257	/*
4258	 * We must validate the spares and l2cache devices after checking the
4259	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
4260	 */
4261	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4262		return (spa_vdev_exit(spa, vd, txg, error));
4263
4264	/*
4265	 * Transfer each new top-level vdev from vd to rvd.
4266	 */
4267	for (int c = 0; c < vd->vdev_children; c++) {
4268
4269		/*
4270		 * Set the vdev id to the first hole, if one exists.
4271		 */
4272		for (id = 0; id < rvd->vdev_children; id++) {
4273			if (rvd->vdev_child[id]->vdev_ishole) {
4274				vdev_free(rvd->vdev_child[id]);
4275				break;
4276			}
4277		}
4278		tvd = vd->vdev_child[c];
4279		vdev_remove_child(vd, tvd);
4280		tvd->vdev_id = id;
4281		vdev_add_child(rvd, tvd);
4282		vdev_config_dirty(tvd);
4283	}
4284
4285	if (nspares != 0) {
4286		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4287		    ZPOOL_CONFIG_SPARES);
4288		spa_load_spares(spa);
4289		spa->spa_spares.sav_sync = B_TRUE;
4290	}
4291
4292	if (nl2cache != 0) {
4293		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4294		    ZPOOL_CONFIG_L2CACHE);
4295		spa_load_l2cache(spa);
4296		spa->spa_l2cache.sav_sync = B_TRUE;
4297	}
4298
4299	/*
4300	 * We have to be careful when adding new vdevs to an existing pool.
4301	 * If other threads start allocating from these vdevs before we
4302	 * sync the config cache, and we lose power, then upon reboot we may
4303	 * fail to open the pool because there are DVAs that the config cache
4304	 * can't translate.  Therefore, we first add the vdevs without
4305	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4306	 * and then let spa_config_update() initialize the new metaslabs.
4307	 *
4308	 * spa_load() checks for added-but-not-initialized vdevs, so that
4309	 * if we lose power at any point in this sequence, the remaining
4310	 * steps will be completed the next time we load the pool.
4311	 */
4312	(void) spa_vdev_exit(spa, vd, txg, 0);
4313
4314	mutex_enter(&spa_namespace_lock);
4315	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4316	mutex_exit(&spa_namespace_lock);
4317
4318	return (0);
4319}
4320
4321/*
4322 * Attach a device to a mirror.  The arguments are the path to any device
4323 * in the mirror, and the nvroot for the new device.  If the path specifies
4324 * a device that is not mirrored, we automatically insert the mirror vdev.
4325 *
4326 * If 'replacing' is specified, the new device is intended to replace the
4327 * existing device; in this case the two devices are made into their own
4328 * mirror using the 'replacing' vdev, which is functionally identical to
4329 * the mirror vdev (it actually reuses all the same ops) but has a few
4330 * extra rules: you can't attach to it after it's been created, and upon
4331 * completion of resilvering, the first disk (the one being replaced)
4332 * is automatically detached.
4333 */
4334int
4335spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4336{
4337	uint64_t txg, dtl_max_txg;
4338	vdev_t *rvd = spa->spa_root_vdev;
4339	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4340	vdev_ops_t *pvops;
4341	char *oldvdpath, *newvdpath;
4342	int newvd_isspare;
4343	int error;
4344
4345	ASSERT(spa_writeable(spa));
4346
4347	txg = spa_vdev_enter(spa);
4348
4349	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4350
4351	if (oldvd == NULL)
4352		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4353
4354	if (!oldvd->vdev_ops->vdev_op_leaf)
4355		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4356
4357	pvd = oldvd->vdev_parent;
4358
4359	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4360	    VDEV_ALLOC_ATTACH)) != 0)
4361		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4362
4363	if (newrootvd->vdev_children != 1)
4364		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4365
4366	newvd = newrootvd->vdev_child[0];
4367
4368	if (!newvd->vdev_ops->vdev_op_leaf)
4369		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4370
4371	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4372		return (spa_vdev_exit(spa, newrootvd, txg, error));
4373
4374	/*
4375	 * Spares can't replace logs
4376	 */
4377	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4378		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4379
4380	if (!replacing) {
4381		/*
4382		 * For attach, the only allowable parent is a mirror or the root
4383		 * vdev.
4384		 */
4385		if (pvd->vdev_ops != &vdev_mirror_ops &&
4386		    pvd->vdev_ops != &vdev_root_ops)
4387			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4388
4389		pvops = &vdev_mirror_ops;
4390	} else {
4391		/*
4392		 * Active hot spares can only be replaced by inactive hot
4393		 * spares.
4394		 */
4395		if (pvd->vdev_ops == &vdev_spare_ops &&
4396		    oldvd->vdev_isspare &&
4397		    !spa_has_spare(spa, newvd->vdev_guid))
4398			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4399
4400		/*
4401		 * If the source is a hot spare, and the parent isn't already a
4402		 * spare, then we want to create a new hot spare.  Otherwise, we
4403		 * want to create a replacing vdev.  The user is not allowed to
4404		 * attach to a spared vdev child unless the 'isspare' state is
4405		 * the same (spare replaces spare, non-spare replaces
4406		 * non-spare).
4407		 */
4408		if (pvd->vdev_ops == &vdev_replacing_ops &&
4409		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4410			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4411		} else if (pvd->vdev_ops == &vdev_spare_ops &&
4412		    newvd->vdev_isspare != oldvd->vdev_isspare) {
4413			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4414		}
4415
4416		if (newvd->vdev_isspare)
4417			pvops = &vdev_spare_ops;
4418		else
4419			pvops = &vdev_replacing_ops;
4420	}
4421
4422	/*
4423	 * Make sure the new device is big enough.
4424	 */
4425	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4426		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4427
4428	/*
4429	 * The new device cannot have a higher alignment requirement
4430	 * than the top-level vdev.
4431	 */
4432	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4433		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4434
4435	/*
4436	 * If this is an in-place replacement, update oldvd's path and devid
4437	 * to make it distinguishable from newvd, and unopenable from now on.
4438	 */
4439	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4440		spa_strfree(oldvd->vdev_path);
4441		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4442		    KM_SLEEP);
4443		(void) sprintf(oldvd->vdev_path, "%s/%s",
4444		    newvd->vdev_path, "old");
4445		if (oldvd->vdev_devid != NULL) {
4446			spa_strfree(oldvd->vdev_devid);
4447			oldvd->vdev_devid = NULL;
4448		}
4449	}
4450
4451	/* mark the device being resilvered */
4452	newvd->vdev_resilver_txg = txg;
4453
4454	/*
4455	 * If the parent is not a mirror, or if we're replacing, insert the new
4456	 * mirror/replacing/spare vdev above oldvd.
4457	 */
4458	if (pvd->vdev_ops != pvops)
4459		pvd = vdev_add_parent(oldvd, pvops);
4460
4461	ASSERT(pvd->vdev_top->vdev_parent == rvd);
4462	ASSERT(pvd->vdev_ops == pvops);
4463	ASSERT(oldvd->vdev_parent == pvd);
4464
4465	/*
4466	 * Extract the new device from its root and add it to pvd.
4467	 */
4468	vdev_remove_child(newrootvd, newvd);
4469	newvd->vdev_id = pvd->vdev_children;
4470	newvd->vdev_crtxg = oldvd->vdev_crtxg;
4471	vdev_add_child(pvd, newvd);
4472
4473	tvd = newvd->vdev_top;
4474	ASSERT(pvd->vdev_top == tvd);
4475	ASSERT(tvd->vdev_parent == rvd);
4476
4477	vdev_config_dirty(tvd);
4478
4479	/*
4480	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4481	 * for any dmu_sync-ed blocks.  It will propagate upward when
4482	 * spa_vdev_exit() calls vdev_dtl_reassess().
4483	 */
4484	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4485
4486	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4487	    dtl_max_txg - TXG_INITIAL);
4488
4489	if (newvd->vdev_isspare) {
4490		spa_spare_activate(newvd);
4491		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4492	}
4493
4494	oldvdpath = spa_strdup(oldvd->vdev_path);
4495	newvdpath = spa_strdup(newvd->vdev_path);
4496	newvd_isspare = newvd->vdev_isspare;
4497
4498	/*
4499	 * Mark newvd's DTL dirty in this txg.
4500	 */
4501	vdev_dirty(tvd, VDD_DTL, newvd, txg);
4502
4503	/*
4504	 * Restart the resilver
4505	 */
4506	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4507
4508	/*
4509	 * Commit the config
4510	 */
4511	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4512
4513	spa_history_log_internal(spa, "vdev attach", NULL,
4514	    "%s vdev=%s %s vdev=%s",
4515	    replacing && newvd_isspare ? "spare in" :
4516	    replacing ? "replace" : "attach", newvdpath,
4517	    replacing ? "for" : "to", oldvdpath);
4518
4519	spa_strfree(oldvdpath);
4520	spa_strfree(newvdpath);
4521
4522	if (spa->spa_bootfs)
4523		spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4524
4525	return (0);
4526}
4527
4528/*
4529 * Detach a device from a mirror or replacing vdev.
4530 *
4531 * If 'replace_done' is specified, only detach if the parent
4532 * is a replacing vdev.
4533 */
4534int
4535spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4536{
4537	uint64_t txg;
4538	int error;
4539	vdev_t *rvd = spa->spa_root_vdev;
4540	vdev_t *vd, *pvd, *cvd, *tvd;
4541	boolean_t unspare = B_FALSE;
4542	uint64_t unspare_guid = 0;
4543	char *vdpath;
4544
4545	ASSERT(spa_writeable(spa));
4546
4547	txg = spa_vdev_enter(spa);
4548
4549	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4550
4551	if (vd == NULL)
4552		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4553
4554	if (!vd->vdev_ops->vdev_op_leaf)
4555		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4556
4557	pvd = vd->vdev_parent;
4558
4559	/*
4560	 * If the parent/child relationship is not as expected, don't do it.
4561	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4562	 * vdev that's replacing B with C.  The user's intent in replacing
4563	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4564	 * the replace by detaching C, the expected behavior is to end up
4565	 * M(A,B).  But suppose that right after deciding to detach C,
4566	 * the replacement of B completes.  We would have M(A,C), and then
4567	 * ask to detach C, which would leave us with just A -- not what
4568	 * the user wanted.  To prevent this, we make sure that the
4569	 * parent/child relationship hasn't changed -- in this example,
4570	 * that C's parent is still the replacing vdev R.
4571	 */
4572	if (pvd->vdev_guid != pguid && pguid != 0)
4573		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4574
4575	/*
4576	 * Only 'replacing' or 'spare' vdevs can be replaced.
4577	 */
4578	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4579	    pvd->vdev_ops != &vdev_spare_ops)
4580		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4581
4582	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4583	    spa_version(spa) >= SPA_VERSION_SPARES);
4584
4585	/*
4586	 * Only mirror, replacing, and spare vdevs support detach.
4587	 */
4588	if (pvd->vdev_ops != &vdev_replacing_ops &&
4589	    pvd->vdev_ops != &vdev_mirror_ops &&
4590	    pvd->vdev_ops != &vdev_spare_ops)
4591		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4592
4593	/*
4594	 * If this device has the only valid copy of some data,
4595	 * we cannot safely detach it.
4596	 */
4597	if (vdev_dtl_required(vd))
4598		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4599
4600	ASSERT(pvd->vdev_children >= 2);
4601
4602	/*
4603	 * If we are detaching the second disk from a replacing vdev, then
4604	 * check to see if we changed the original vdev's path to have "/old"
4605	 * at the end in spa_vdev_attach().  If so, undo that change now.
4606	 */
4607	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4608	    vd->vdev_path != NULL) {
4609		size_t len = strlen(vd->vdev_path);
4610
4611		for (int c = 0; c < pvd->vdev_children; c++) {
4612			cvd = pvd->vdev_child[c];
4613
4614			if (cvd == vd || cvd->vdev_path == NULL)
4615				continue;
4616
4617			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4618			    strcmp(cvd->vdev_path + len, "/old") == 0) {
4619				spa_strfree(cvd->vdev_path);
4620				cvd->vdev_path = spa_strdup(vd->vdev_path);
4621				break;
4622			}
4623		}
4624	}
4625
4626	/*
4627	 * If we are detaching the original disk from a spare, then it implies
4628	 * that the spare should become a real disk, and be removed from the
4629	 * active spare list for the pool.
4630	 */
4631	if (pvd->vdev_ops == &vdev_spare_ops &&
4632	    vd->vdev_id == 0 &&
4633	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4634		unspare = B_TRUE;
4635
4636	/*
4637	 * Erase the disk labels so the disk can be used for other things.
4638	 * This must be done after all other error cases are handled,
4639	 * but before we disembowel vd (so we can still do I/O to it).
4640	 * But if we can't do it, don't treat the error as fatal --
4641	 * it may be that the unwritability of the disk is the reason
4642	 * it's being detached!
4643	 */
4644	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4645
4646	/*
4647	 * Remove vd from its parent and compact the parent's children.
4648	 */
4649	vdev_remove_child(pvd, vd);
4650	vdev_compact_children(pvd);
4651
4652	/*
4653	 * Remember one of the remaining children so we can get tvd below.
4654	 */
4655	cvd = pvd->vdev_child[pvd->vdev_children - 1];
4656
4657	/*
4658	 * If we need to remove the remaining child from the list of hot spares,
4659	 * do it now, marking the vdev as no longer a spare in the process.
4660	 * We must do this before vdev_remove_parent(), because that can
4661	 * change the GUID if it creates a new toplevel GUID.  For a similar
4662	 * reason, we must remove the spare now, in the same txg as the detach;
4663	 * otherwise someone could attach a new sibling, change the GUID, and
4664	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4665	 */
4666	if (unspare) {
4667		ASSERT(cvd->vdev_isspare);
4668		spa_spare_remove(cvd);
4669		unspare_guid = cvd->vdev_guid;
4670		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4671		cvd->vdev_unspare = B_TRUE;
4672	}
4673
4674	/*
4675	 * If the parent mirror/replacing vdev only has one child,
4676	 * the parent is no longer needed.  Remove it from the tree.
4677	 */
4678	if (pvd->vdev_children == 1) {
4679		if (pvd->vdev_ops == &vdev_spare_ops)
4680			cvd->vdev_unspare = B_FALSE;
4681		vdev_remove_parent(cvd);
4682	}
4683
4684
4685	/*
4686	 * We don't set tvd until now because the parent we just removed
4687	 * may have been the previous top-level vdev.
4688	 */
4689	tvd = cvd->vdev_top;
4690	ASSERT(tvd->vdev_parent == rvd);
4691
4692	/*
4693	 * Reevaluate the parent vdev state.
4694	 */
4695	vdev_propagate_state(cvd);
4696
4697	/*
4698	 * If the 'autoexpand' property is set on the pool then automatically
4699	 * try to expand the size of the pool. For example if the device we
4700	 * just detached was smaller than the others, it may be possible to
4701	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4702	 * first so that we can obtain the updated sizes of the leaf vdevs.
4703	 */
4704	if (spa->spa_autoexpand) {
4705		vdev_reopen(tvd);
4706		vdev_expand(tvd, txg);
4707	}
4708
4709	vdev_config_dirty(tvd);
4710
4711	/*
4712	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4713	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4714	 * But first make sure we're not on any *other* txg's DTL list, to
4715	 * prevent vd from being accessed after it's freed.
4716	 */
4717	vdpath = spa_strdup(vd->vdev_path);
4718	for (int t = 0; t < TXG_SIZE; t++)
4719		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4720	vd->vdev_detached = B_TRUE;
4721	vdev_dirty(tvd, VDD_DTL, vd, txg);
4722
4723	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4724
4725	/* hang on to the spa before we release the lock */
4726	spa_open_ref(spa, FTAG);
4727
4728	error = spa_vdev_exit(spa, vd, txg, 0);
4729
4730	spa_history_log_internal(spa, "detach", NULL,
4731	    "vdev=%s", vdpath);
4732	spa_strfree(vdpath);
4733
4734	/*
4735	 * If this was the removal of the original device in a hot spare vdev,
4736	 * then we want to go through and remove the device from the hot spare
4737	 * list of every other pool.
4738	 */
4739	if (unspare) {
4740		spa_t *altspa = NULL;
4741
4742		mutex_enter(&spa_namespace_lock);
4743		while ((altspa = spa_next(altspa)) != NULL) {
4744			if (altspa->spa_state != POOL_STATE_ACTIVE ||
4745			    altspa == spa)
4746				continue;
4747
4748			spa_open_ref(altspa, FTAG);
4749			mutex_exit(&spa_namespace_lock);
4750			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4751			mutex_enter(&spa_namespace_lock);
4752			spa_close(altspa, FTAG);
4753		}
4754		mutex_exit(&spa_namespace_lock);
4755
4756		/* search the rest of the vdevs for spares to remove */
4757		spa_vdev_resilver_done(spa);
4758	}
4759
4760	/* all done with the spa; OK to release */
4761	mutex_enter(&spa_namespace_lock);
4762	spa_close(spa, FTAG);
4763	mutex_exit(&spa_namespace_lock);
4764
4765	return (error);
4766}
4767
4768/*
4769 * Split a set of devices from their mirrors, and create a new pool from them.
4770 */
4771int
4772spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4773    nvlist_t *props, boolean_t exp)
4774{
4775	int error = 0;
4776	uint64_t txg, *glist;
4777	spa_t *newspa;
4778	uint_t c, children, lastlog;
4779	nvlist_t **child, *nvl, *tmp;
4780	dmu_tx_t *tx;
4781	char *altroot = NULL;
4782	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4783	boolean_t activate_slog;
4784
4785	ASSERT(spa_writeable(spa));
4786
4787	txg = spa_vdev_enter(spa);
4788
4789	/* clear the log and flush everything up to now */
4790	activate_slog = spa_passivate_log(spa);
4791	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4792	error = spa_offline_log(spa);
4793	txg = spa_vdev_config_enter(spa);
4794
4795	if (activate_slog)
4796		spa_activate_log(spa);
4797
4798	if (error != 0)
4799		return (spa_vdev_exit(spa, NULL, txg, error));
4800
4801	/* check new spa name before going any further */
4802	if (spa_lookup(newname) != NULL)
4803		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4804
4805	/*
4806	 * scan through all the children to ensure they're all mirrors
4807	 */
4808	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4809	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4810	    &children) != 0)
4811		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4812
4813	/* first, check to ensure we've got the right child count */
4814	rvd = spa->spa_root_vdev;
4815	lastlog = 0;
4816	for (c = 0; c < rvd->vdev_children; c++) {
4817		vdev_t *vd = rvd->vdev_child[c];
4818
4819		/* don't count the holes & logs as children */
4820		if (vd->vdev_islog || vd->vdev_ishole) {
4821			if (lastlog == 0)
4822				lastlog = c;
4823			continue;
4824		}
4825
4826		lastlog = 0;
4827	}
4828	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4829		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4830
4831	/* next, ensure no spare or cache devices are part of the split */
4832	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4833	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4834		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4835
4836	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4837	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4838
4839	/* then, loop over each vdev and validate it */
4840	for (c = 0; c < children; c++) {
4841		uint64_t is_hole = 0;
4842
4843		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4844		    &is_hole);
4845
4846		if (is_hole != 0) {
4847			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4848			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4849				continue;
4850			} else {
4851				error = SET_ERROR(EINVAL);
4852				break;
4853			}
4854		}
4855
4856		/* which disk is going to be split? */
4857		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4858		    &glist[c]) != 0) {
4859			error = SET_ERROR(EINVAL);
4860			break;
4861		}
4862
4863		/* look it up in the spa */
4864		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4865		if (vml[c] == NULL) {
4866			error = SET_ERROR(ENODEV);
4867			break;
4868		}
4869
4870		/* make sure there's nothing stopping the split */
4871		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4872		    vml[c]->vdev_islog ||
4873		    vml[c]->vdev_ishole ||
4874		    vml[c]->vdev_isspare ||
4875		    vml[c]->vdev_isl2cache ||
4876		    !vdev_writeable(vml[c]) ||
4877		    vml[c]->vdev_children != 0 ||
4878		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4879		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4880			error = SET_ERROR(EINVAL);
4881			break;
4882		}
4883
4884		if (vdev_dtl_required(vml[c])) {
4885			error = SET_ERROR(EBUSY);
4886			break;
4887		}
4888
4889		/* we need certain info from the top level */
4890		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4891		    vml[c]->vdev_top->vdev_ms_array) == 0);
4892		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4893		    vml[c]->vdev_top->vdev_ms_shift) == 0);
4894		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4895		    vml[c]->vdev_top->vdev_asize) == 0);
4896		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4897		    vml[c]->vdev_top->vdev_ashift) == 0);
4898	}
4899
4900	if (error != 0) {
4901		kmem_free(vml, children * sizeof (vdev_t *));
4902		kmem_free(glist, children * sizeof (uint64_t));
4903		return (spa_vdev_exit(spa, NULL, txg, error));
4904	}
4905
4906	/* stop writers from using the disks */
4907	for (c = 0; c < children; c++) {
4908		if (vml[c] != NULL)
4909			vml[c]->vdev_offline = B_TRUE;
4910	}
4911	vdev_reopen(spa->spa_root_vdev);
4912
4913	/*
4914	 * Temporarily record the splitting vdevs in the spa config.  This
4915	 * will disappear once the config is regenerated.
4916	 */
4917	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4918	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4919	    glist, children) == 0);
4920	kmem_free(glist, children * sizeof (uint64_t));
4921
4922	mutex_enter(&spa->spa_props_lock);
4923	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4924	    nvl) == 0);
4925	mutex_exit(&spa->spa_props_lock);
4926	spa->spa_config_splitting = nvl;
4927	vdev_config_dirty(spa->spa_root_vdev);
4928
4929	/* configure and create the new pool */
4930	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4931	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4932	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4933	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4934	    spa_version(spa)) == 0);
4935	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4936	    spa->spa_config_txg) == 0);
4937	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4938	    spa_generate_guid(NULL)) == 0);
4939	(void) nvlist_lookup_string(props,
4940	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4941
4942	/* add the new pool to the namespace */
4943	newspa = spa_add(newname, config, altroot);
4944	newspa->spa_config_txg = spa->spa_config_txg;
4945	spa_set_log_state(newspa, SPA_LOG_CLEAR);
4946
4947	/* release the spa config lock, retaining the namespace lock */
4948	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4949
4950	if (zio_injection_enabled)
4951		zio_handle_panic_injection(spa, FTAG, 1);
4952
4953	spa_activate(newspa, spa_mode_global);
4954	spa_async_suspend(newspa);
4955
4956	/* create the new pool from the disks of the original pool */
4957	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4958	if (error)
4959		goto out;
4960
4961	/* if that worked, generate a real config for the new pool */
4962	if (newspa->spa_root_vdev != NULL) {
4963		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4964		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4965		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4966		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4967		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4968		    B_TRUE));
4969	}
4970
4971	/* set the props */
4972	if (props != NULL) {
4973		spa_configfile_set(newspa, props, B_FALSE);
4974		error = spa_prop_set(newspa, props);
4975		if (error)
4976			goto out;
4977	}
4978
4979	/* flush everything */
4980	txg = spa_vdev_config_enter(newspa);
4981	vdev_config_dirty(newspa->spa_root_vdev);
4982	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4983
4984	if (zio_injection_enabled)
4985		zio_handle_panic_injection(spa, FTAG, 2);
4986
4987	spa_async_resume(newspa);
4988
4989	/* finally, update the original pool's config */
4990	txg = spa_vdev_config_enter(spa);
4991	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4992	error = dmu_tx_assign(tx, TXG_WAIT);
4993	if (error != 0)
4994		dmu_tx_abort(tx);
4995	for (c = 0; c < children; c++) {
4996		if (vml[c] != NULL) {
4997			vdev_split(vml[c]);
4998			if (error == 0)
4999				spa_history_log_internal(spa, "detach", tx,
5000				    "vdev=%s", vml[c]->vdev_path);
5001			vdev_free(vml[c]);
5002		}
5003	}
5004	vdev_config_dirty(spa->spa_root_vdev);
5005	spa->spa_config_splitting = NULL;
5006	nvlist_free(nvl);
5007	if (error == 0)
5008		dmu_tx_commit(tx);
5009	(void) spa_vdev_exit(spa, NULL, txg, 0);
5010
5011	if (zio_injection_enabled)
5012		zio_handle_panic_injection(spa, FTAG, 3);
5013
5014	/* split is complete; log a history record */
5015	spa_history_log_internal(newspa, "split", NULL,
5016	    "from pool %s", spa_name(spa));
5017
5018	kmem_free(vml, children * sizeof (vdev_t *));
5019
5020	/* if we're not going to mount the filesystems in userland, export */
5021	if (exp)
5022		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5023		    B_FALSE, B_FALSE);
5024
5025	return (error);
5026
5027out:
5028	spa_unload(newspa);
5029	spa_deactivate(newspa);
5030	spa_remove(newspa);
5031
5032	txg = spa_vdev_config_enter(spa);
5033
5034	/* re-online all offlined disks */
5035	for (c = 0; c < children; c++) {
5036		if (vml[c] != NULL)
5037			vml[c]->vdev_offline = B_FALSE;
5038	}
5039	vdev_reopen(spa->spa_root_vdev);
5040
5041	nvlist_free(spa->spa_config_splitting);
5042	spa->spa_config_splitting = NULL;
5043	(void) spa_vdev_exit(spa, NULL, txg, error);
5044
5045	kmem_free(vml, children * sizeof (vdev_t *));
5046	return (error);
5047}
5048
5049static nvlist_t *
5050spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5051{
5052	for (int i = 0; i < count; i++) {
5053		uint64_t guid;
5054
5055		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5056		    &guid) == 0);
5057
5058		if (guid == target_guid)
5059			return (nvpp[i]);
5060	}
5061
5062	return (NULL);
5063}
5064
5065static void
5066spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5067	nvlist_t *dev_to_remove)
5068{
5069	nvlist_t **newdev = NULL;
5070
5071	if (count > 1)
5072		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5073
5074	for (int i = 0, j = 0; i < count; i++) {
5075		if (dev[i] == dev_to_remove)
5076			continue;
5077		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5078	}
5079
5080	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5081	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5082
5083	for (int i = 0; i < count - 1; i++)
5084		nvlist_free(newdev[i]);
5085
5086	if (count > 1)
5087		kmem_free(newdev, (count - 1) * sizeof (void *));
5088}
5089
5090/*
5091 * Evacuate the device.
5092 */
5093static int
5094spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5095{
5096	uint64_t txg;
5097	int error = 0;
5098
5099	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5100	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5101	ASSERT(vd == vd->vdev_top);
5102
5103	/*
5104	 * Evacuate the device.  We don't hold the config lock as writer
5105	 * since we need to do I/O but we do keep the
5106	 * spa_namespace_lock held.  Once this completes the device
5107	 * should no longer have any blocks allocated on it.
5108	 */
5109	if (vd->vdev_islog) {
5110		if (vd->vdev_stat.vs_alloc != 0)
5111			error = spa_offline_log(spa);
5112	} else {
5113		error = SET_ERROR(ENOTSUP);
5114	}
5115
5116	if (error)
5117		return (error);
5118
5119	/*
5120	 * The evacuation succeeded.  Remove any remaining MOS metadata
5121	 * associated with this vdev, and wait for these changes to sync.
5122	 */
5123	ASSERT0(vd->vdev_stat.vs_alloc);
5124	txg = spa_vdev_config_enter(spa);
5125	vd->vdev_removing = B_TRUE;
5126	vdev_dirty(vd, 0, NULL, txg);
5127	vdev_config_dirty(vd);
5128	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5129
5130	return (0);
5131}
5132
5133/*
5134 * Complete the removal by cleaning up the namespace.
5135 */
5136static void
5137spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5138{
5139	vdev_t *rvd = spa->spa_root_vdev;
5140	uint64_t id = vd->vdev_id;
5141	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5142
5143	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5144	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5145	ASSERT(vd == vd->vdev_top);
5146
5147	/*
5148	 * Only remove any devices which are empty.
5149	 */
5150	if (vd->vdev_stat.vs_alloc != 0)
5151		return;
5152
5153	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5154
5155	if (list_link_active(&vd->vdev_state_dirty_node))
5156		vdev_state_clean(vd);
5157	if (list_link_active(&vd->vdev_config_dirty_node))
5158		vdev_config_clean(vd);
5159
5160	vdev_free(vd);
5161
5162	if (last_vdev) {
5163		vdev_compact_children(rvd);
5164	} else {
5165		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5166		vdev_add_child(rvd, vd);
5167	}
5168	vdev_config_dirty(rvd);
5169
5170	/*
5171	 * Reassess the health of our root vdev.
5172	 */
5173	vdev_reopen(rvd);
5174}
5175
5176/*
5177 * Remove a device from the pool -
5178 *
5179 * Removing a device from the vdev namespace requires several steps
5180 * and can take a significant amount of time.  As a result we use
5181 * the spa_vdev_config_[enter/exit] functions which allow us to
5182 * grab and release the spa_config_lock while still holding the namespace
5183 * lock.  During each step the configuration is synced out.
5184 *
5185 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5186 * devices.
5187 */
5188int
5189spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5190{
5191	vdev_t *vd;
5192	metaslab_group_t *mg;
5193	nvlist_t **spares, **l2cache, *nv;
5194	uint64_t txg = 0;
5195	uint_t nspares, nl2cache;
5196	int error = 0;
5197	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5198
5199	ASSERT(spa_writeable(spa));
5200
5201	if (!locked)
5202		txg = spa_vdev_enter(spa);
5203
5204	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5205
5206	if (spa->spa_spares.sav_vdevs != NULL &&
5207	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5208	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5209	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5210		/*
5211		 * Only remove the hot spare if it's not currently in use
5212		 * in this pool.
5213		 */
5214		if (vd == NULL || unspare) {
5215			spa_vdev_remove_aux(spa->spa_spares.sav_config,
5216			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5217			spa_load_spares(spa);
5218			spa->spa_spares.sav_sync = B_TRUE;
5219		} else {
5220			error = SET_ERROR(EBUSY);
5221		}
5222	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
5223	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5224	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5225	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5226		/*
5227		 * Cache devices can always be removed.
5228		 */
5229		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5230		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5231		spa_load_l2cache(spa);
5232		spa->spa_l2cache.sav_sync = B_TRUE;
5233	} else if (vd != NULL && vd->vdev_islog) {
5234		ASSERT(!locked);
5235		ASSERT(vd == vd->vdev_top);
5236
5237		/*
5238		 * XXX - Once we have bp-rewrite this should
5239		 * become the common case.
5240		 */
5241
5242		mg = vd->vdev_mg;
5243
5244		/*
5245		 * Stop allocating from this vdev.
5246		 */
5247		metaslab_group_passivate(mg);
5248
5249		/*
5250		 * Wait for the youngest allocations and frees to sync,
5251		 * and then wait for the deferral of those frees to finish.
5252		 */
5253		spa_vdev_config_exit(spa, NULL,
5254		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5255
5256		/*
5257		 * Attempt to evacuate the vdev.
5258		 */
5259		error = spa_vdev_remove_evacuate(spa, vd);
5260
5261		txg = spa_vdev_config_enter(spa);
5262
5263		/*
5264		 * If we couldn't evacuate the vdev, unwind.
5265		 */
5266		if (error) {
5267			metaslab_group_activate(mg);
5268			return (spa_vdev_exit(spa, NULL, txg, error));
5269		}
5270
5271		/*
5272		 * Clean up the vdev namespace.
5273		 */
5274		spa_vdev_remove_from_namespace(spa, vd);
5275
5276	} else if (vd != NULL) {
5277		/*
5278		 * Normal vdevs cannot be removed (yet).
5279		 */
5280		error = SET_ERROR(ENOTSUP);
5281	} else {
5282		/*
5283		 * There is no vdev of any kind with the specified guid.
5284		 */
5285		error = SET_ERROR(ENOENT);
5286	}
5287
5288	if (!locked)
5289		return (spa_vdev_exit(spa, NULL, txg, error));
5290
5291	return (error);
5292}
5293
5294/*
5295 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5296 * currently spared, so we can detach it.
5297 */
5298static vdev_t *
5299spa_vdev_resilver_done_hunt(vdev_t *vd)
5300{
5301	vdev_t *newvd, *oldvd;
5302
5303	for (int c = 0; c < vd->vdev_children; c++) {
5304		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5305		if (oldvd != NULL)
5306			return (oldvd);
5307	}
5308
5309	/*
5310	 * Check for a completed replacement.  We always consider the first
5311	 * vdev in the list to be the oldest vdev, and the last one to be
5312	 * the newest (see spa_vdev_attach() for how that works).  In
5313	 * the case where the newest vdev is faulted, we will not automatically
5314	 * remove it after a resilver completes.  This is OK as it will require
5315	 * user intervention to determine which disk the admin wishes to keep.
5316	 */
5317	if (vd->vdev_ops == &vdev_replacing_ops) {
5318		ASSERT(vd->vdev_children > 1);
5319
5320		newvd = vd->vdev_child[vd->vdev_children - 1];
5321		oldvd = vd->vdev_child[0];
5322
5323		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5324		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5325		    !vdev_dtl_required(oldvd))
5326			return (oldvd);
5327	}
5328
5329	/*
5330	 * Check for a completed resilver with the 'unspare' flag set.
5331	 */
5332	if (vd->vdev_ops == &vdev_spare_ops) {
5333		vdev_t *first = vd->vdev_child[0];
5334		vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5335
5336		if (last->vdev_unspare) {
5337			oldvd = first;
5338			newvd = last;
5339		} else if (first->vdev_unspare) {
5340			oldvd = last;
5341			newvd = first;
5342		} else {
5343			oldvd = NULL;
5344		}
5345
5346		if (oldvd != NULL &&
5347		    vdev_dtl_empty(newvd, DTL_MISSING) &&
5348		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5349		    !vdev_dtl_required(oldvd))
5350			return (oldvd);
5351
5352		/*
5353		 * If there are more than two spares attached to a disk,
5354		 * and those spares are not required, then we want to
5355		 * attempt to free them up now so that they can be used
5356		 * by other pools.  Once we're back down to a single
5357		 * disk+spare, we stop removing them.
5358		 */
5359		if (vd->vdev_children > 2) {
5360			newvd = vd->vdev_child[1];
5361
5362			if (newvd->vdev_isspare && last->vdev_isspare &&
5363			    vdev_dtl_empty(last, DTL_MISSING) &&
5364			    vdev_dtl_empty(last, DTL_OUTAGE) &&
5365			    !vdev_dtl_required(newvd))
5366				return (newvd);
5367		}
5368	}
5369
5370	return (NULL);
5371}
5372
5373static void
5374spa_vdev_resilver_done(spa_t *spa)
5375{
5376	vdev_t *vd, *pvd, *ppvd;
5377	uint64_t guid, sguid, pguid, ppguid;
5378
5379	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5380
5381	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5382		pvd = vd->vdev_parent;
5383		ppvd = pvd->vdev_parent;
5384		guid = vd->vdev_guid;
5385		pguid = pvd->vdev_guid;
5386		ppguid = ppvd->vdev_guid;
5387		sguid = 0;
5388		/*
5389		 * If we have just finished replacing a hot spared device, then
5390		 * we need to detach the parent's first child (the original hot
5391		 * spare) as well.
5392		 */
5393		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5394		    ppvd->vdev_children == 2) {
5395			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5396			sguid = ppvd->vdev_child[1]->vdev_guid;
5397		}
5398		ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5399
5400		spa_config_exit(spa, SCL_ALL, FTAG);
5401		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5402			return;
5403		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5404			return;
5405		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5406	}
5407
5408	spa_config_exit(spa, SCL_ALL, FTAG);
5409}
5410
5411/*
5412 * Update the stored path or FRU for this vdev.
5413 */
5414int
5415spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5416    boolean_t ispath)
5417{
5418	vdev_t *vd;
5419	boolean_t sync = B_FALSE;
5420
5421	ASSERT(spa_writeable(spa));
5422
5423	spa_vdev_state_enter(spa, SCL_ALL);
5424
5425	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5426		return (spa_vdev_state_exit(spa, NULL, ENOENT));
5427
5428	if (!vd->vdev_ops->vdev_op_leaf)
5429		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5430
5431	if (ispath) {
5432		if (strcmp(value, vd->vdev_path) != 0) {
5433			spa_strfree(vd->vdev_path);
5434			vd->vdev_path = spa_strdup(value);
5435			sync = B_TRUE;
5436		}
5437	} else {
5438		if (vd->vdev_fru == NULL) {
5439			vd->vdev_fru = spa_strdup(value);
5440			sync = B_TRUE;
5441		} else if (strcmp(value, vd->vdev_fru) != 0) {
5442			spa_strfree(vd->vdev_fru);
5443			vd->vdev_fru = spa_strdup(value);
5444			sync = B_TRUE;
5445		}
5446	}
5447
5448	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5449}
5450
5451int
5452spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5453{
5454	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5455}
5456
5457int
5458spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5459{
5460	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5461}
5462
5463/*
5464 * ==========================================================================
5465 * SPA Scanning
5466 * ==========================================================================
5467 */
5468
5469int
5470spa_scan_stop(spa_t *spa)
5471{
5472	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5473	if (dsl_scan_resilvering(spa->spa_dsl_pool))
5474		return (SET_ERROR(EBUSY));
5475	return (dsl_scan_cancel(spa->spa_dsl_pool));
5476}
5477
5478int
5479spa_scan(spa_t *spa, pool_scan_func_t func)
5480{
5481	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5482
5483	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5484		return (SET_ERROR(ENOTSUP));
5485
5486	/*
5487	 * If a resilver was requested, but there is no DTL on a
5488	 * writeable leaf device, we have nothing to do.
5489	 */
5490	if (func == POOL_SCAN_RESILVER &&
5491	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5492		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5493		return (0);
5494	}
5495
5496	return (dsl_scan(spa->spa_dsl_pool, func));
5497}
5498
5499/*
5500 * ==========================================================================
5501 * SPA async task processing
5502 * ==========================================================================
5503 */
5504
5505static void
5506spa_async_remove(spa_t *spa, vdev_t *vd)
5507{
5508	if (vd->vdev_remove_wanted) {
5509		vd->vdev_remove_wanted = B_FALSE;
5510		vd->vdev_delayed_close = B_FALSE;
5511		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5512
5513		/*
5514		 * We want to clear the stats, but we don't want to do a full
5515		 * vdev_clear() as that will cause us to throw away
5516		 * degraded/faulted state as well as attempt to reopen the
5517		 * device, all of which is a waste.
5518		 */
5519		vd->vdev_stat.vs_read_errors = 0;
5520		vd->vdev_stat.vs_write_errors = 0;
5521		vd->vdev_stat.vs_checksum_errors = 0;
5522
5523		vdev_state_dirty(vd->vdev_top);
5524	}
5525
5526	for (int c = 0; c < vd->vdev_children; c++)
5527		spa_async_remove(spa, vd->vdev_child[c]);
5528}
5529
5530static void
5531spa_async_probe(spa_t *spa, vdev_t *vd)
5532{
5533	if (vd->vdev_probe_wanted) {
5534		vd->vdev_probe_wanted = B_FALSE;
5535		vdev_reopen(vd);	/* vdev_open() does the actual probe */
5536	}
5537
5538	for (int c = 0; c < vd->vdev_children; c++)
5539		spa_async_probe(spa, vd->vdev_child[c]);
5540}
5541
5542static void
5543spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5544{
5545	sysevent_id_t eid;
5546	nvlist_t *attr;
5547	char *physpath;
5548
5549	if (!spa->spa_autoexpand)
5550		return;
5551
5552	for (int c = 0; c < vd->vdev_children; c++) {
5553		vdev_t *cvd = vd->vdev_child[c];
5554		spa_async_autoexpand(spa, cvd);
5555	}
5556
5557	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5558		return;
5559
5560	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5561	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5562
5563	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5564	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5565
5566	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5567	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5568
5569	nvlist_free(attr);
5570	kmem_free(physpath, MAXPATHLEN);
5571}
5572
5573static void
5574spa_async_thread(spa_t *spa)
5575{
5576	int tasks;
5577
5578	ASSERT(spa->spa_sync_on);
5579
5580	mutex_enter(&spa->spa_async_lock);
5581	tasks = spa->spa_async_tasks;
5582	spa->spa_async_tasks = 0;
5583	mutex_exit(&spa->spa_async_lock);
5584
5585	/*
5586	 * See if the config needs to be updated.
5587	 */
5588	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5589		uint64_t old_space, new_space;
5590
5591		mutex_enter(&spa_namespace_lock);
5592		old_space = metaslab_class_get_space(spa_normal_class(spa));
5593		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5594		new_space = metaslab_class_get_space(spa_normal_class(spa));
5595		mutex_exit(&spa_namespace_lock);
5596
5597		/*
5598		 * If the pool grew as a result of the config update,
5599		 * then log an internal history event.
5600		 */
5601		if (new_space != old_space) {
5602			spa_history_log_internal(spa, "vdev online", NULL,
5603			    "pool '%s' size: %llu(+%llu)",
5604			    spa_name(spa), new_space, new_space - old_space);
5605		}
5606	}
5607
5608	/*
5609	 * See if any devices need to be marked REMOVED.
5610	 */
5611	if (tasks & SPA_ASYNC_REMOVE) {
5612		spa_vdev_state_enter(spa, SCL_NONE);
5613		spa_async_remove(spa, spa->spa_root_vdev);
5614		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5615			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5616		for (int i = 0; i < spa->spa_spares.sav_count; i++)
5617			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5618		(void) spa_vdev_state_exit(spa, NULL, 0);
5619	}
5620
5621	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5622		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5623		spa_async_autoexpand(spa, spa->spa_root_vdev);
5624		spa_config_exit(spa, SCL_CONFIG, FTAG);
5625	}
5626
5627	/*
5628	 * See if any devices need to be probed.
5629	 */
5630	if (tasks & SPA_ASYNC_PROBE) {
5631		spa_vdev_state_enter(spa, SCL_NONE);
5632		spa_async_probe(spa, spa->spa_root_vdev);
5633		(void) spa_vdev_state_exit(spa, NULL, 0);
5634	}
5635
5636	/*
5637	 * If any devices are done replacing, detach them.
5638	 */
5639	if (tasks & SPA_ASYNC_RESILVER_DONE)
5640		spa_vdev_resilver_done(spa);
5641
5642	/*
5643	 * Kick off a resilver.
5644	 */
5645	if (tasks & SPA_ASYNC_RESILVER)
5646		dsl_resilver_restart(spa->spa_dsl_pool, 0);
5647
5648	/*
5649	 * Let the world know that we're done.
5650	 */
5651	mutex_enter(&spa->spa_async_lock);
5652	spa->spa_async_thread = NULL;
5653	cv_broadcast(&spa->spa_async_cv);
5654	mutex_exit(&spa->spa_async_lock);
5655	thread_exit();
5656}
5657
5658void
5659spa_async_suspend(spa_t *spa)
5660{
5661	mutex_enter(&spa->spa_async_lock);
5662	spa->spa_async_suspended++;
5663	while (spa->spa_async_thread != NULL)
5664		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5665	mutex_exit(&spa->spa_async_lock);
5666}
5667
5668void
5669spa_async_resume(spa_t *spa)
5670{
5671	mutex_enter(&spa->spa_async_lock);
5672	ASSERT(spa->spa_async_suspended != 0);
5673	spa->spa_async_suspended--;
5674	mutex_exit(&spa->spa_async_lock);
5675}
5676
5677static boolean_t
5678spa_async_tasks_pending(spa_t *spa)
5679{
5680	uint_t non_config_tasks;
5681	uint_t config_task;
5682	boolean_t config_task_suspended;
5683
5684	non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5685	config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5686	if (spa->spa_ccw_fail_time == 0) {
5687		config_task_suspended = B_FALSE;
5688	} else {
5689		config_task_suspended =
5690		    (gethrtime() - spa->spa_ccw_fail_time) <
5691		    (zfs_ccw_retry_interval * NANOSEC);
5692	}
5693
5694	return (non_config_tasks || (config_task && !config_task_suspended));
5695}
5696
5697static void
5698spa_async_dispatch(spa_t *spa)
5699{
5700	mutex_enter(&spa->spa_async_lock);
5701	if (spa_async_tasks_pending(spa) &&
5702	    !spa->spa_async_suspended &&
5703	    spa->spa_async_thread == NULL &&
5704	    rootdir != NULL)
5705		spa->spa_async_thread = thread_create(NULL, 0,
5706		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5707	mutex_exit(&spa->spa_async_lock);
5708}
5709
5710void
5711spa_async_request(spa_t *spa, int task)
5712{
5713	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5714	mutex_enter(&spa->spa_async_lock);
5715	spa->spa_async_tasks |= task;
5716	mutex_exit(&spa->spa_async_lock);
5717}
5718
5719/*
5720 * ==========================================================================
5721 * SPA syncing routines
5722 * ==========================================================================
5723 */
5724
5725static int
5726bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5727{
5728	bpobj_t *bpo = arg;
5729	bpobj_enqueue(bpo, bp, tx);
5730	return (0);
5731}
5732
5733static int
5734spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5735{
5736	zio_t *zio = arg;
5737
5738	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5739	    zio->io_flags));
5740	return (0);
5741}
5742
5743/*
5744 * Note: this simple function is not inlined to make it easier to dtrace the
5745 * amount of time spent syncing frees.
5746 */
5747static void
5748spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
5749{
5750	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5751	bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
5752	VERIFY(zio_wait(zio) == 0);
5753}
5754
5755/*
5756 * Note: this simple function is not inlined to make it easier to dtrace the
5757 * amount of time spent syncing deferred frees.
5758 */
5759static void
5760spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
5761{
5762	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5763	VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
5764	    spa_free_sync_cb, zio, tx), ==, 0);
5765	VERIFY0(zio_wait(zio));
5766}
5767
5768
5769static void
5770spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5771{
5772	char *packed = NULL;
5773	size_t bufsize;
5774	size_t nvsize = 0;
5775	dmu_buf_t *db;
5776
5777	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5778
5779	/*
5780	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5781	 * information.  This avoids the dbuf_will_dirty() path and
5782	 * saves us a pre-read to get data we don't actually care about.
5783	 */
5784	bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5785	packed = kmem_alloc(bufsize, KM_SLEEP);
5786
5787	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5788	    KM_SLEEP) == 0);
5789	bzero(packed + nvsize, bufsize - nvsize);
5790
5791	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5792
5793	kmem_free(packed, bufsize);
5794
5795	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5796	dmu_buf_will_dirty(db, tx);
5797	*(uint64_t *)db->db_data = nvsize;
5798	dmu_buf_rele(db, FTAG);
5799}
5800
5801static void
5802spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5803    const char *config, const char *entry)
5804{
5805	nvlist_t *nvroot;
5806	nvlist_t **list;
5807	int i;
5808
5809	if (!sav->sav_sync)
5810		return;
5811
5812	/*
5813	 * Update the MOS nvlist describing the list of available devices.
5814	 * spa_validate_aux() will have already made sure this nvlist is
5815	 * valid and the vdevs are labeled appropriately.
5816	 */
5817	if (sav->sav_object == 0) {
5818		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5819		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5820		    sizeof (uint64_t), tx);
5821		VERIFY(zap_update(spa->spa_meta_objset,
5822		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5823		    &sav->sav_object, tx) == 0);
5824	}
5825
5826	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5827	if (sav->sav_count == 0) {
5828		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5829	} else {
5830		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5831		for (i = 0; i < sav->sav_count; i++)
5832			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5833			    B_FALSE, VDEV_CONFIG_L2CACHE);
5834		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5835		    sav->sav_count) == 0);
5836		for (i = 0; i < sav->sav_count; i++)
5837			nvlist_free(list[i]);
5838		kmem_free(list, sav->sav_count * sizeof (void *));
5839	}
5840
5841	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5842	nvlist_free(nvroot);
5843
5844	sav->sav_sync = B_FALSE;
5845}
5846
5847static void
5848spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5849{
5850	nvlist_t *config;
5851
5852	if (list_is_empty(&spa->spa_config_dirty_list))
5853		return;
5854
5855	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5856
5857	config = spa_config_generate(spa, spa->spa_root_vdev,
5858	    dmu_tx_get_txg(tx), B_FALSE);
5859
5860	/*
5861	 * If we're upgrading the spa version then make sure that
5862	 * the config object gets updated with the correct version.
5863	 */
5864	if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5865		fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5866		    spa->spa_uberblock.ub_version);
5867
5868	spa_config_exit(spa, SCL_STATE, FTAG);
5869
5870	if (spa->spa_config_syncing)
5871		nvlist_free(spa->spa_config_syncing);
5872	spa->spa_config_syncing = config;
5873
5874	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5875}
5876
5877static void
5878spa_sync_version(void *arg, dmu_tx_t *tx)
5879{
5880	uint64_t *versionp = arg;
5881	uint64_t version = *versionp;
5882	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5883
5884	/*
5885	 * Setting the version is special cased when first creating the pool.
5886	 */
5887	ASSERT(tx->tx_txg != TXG_INITIAL);
5888
5889	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5890	ASSERT(version >= spa_version(spa));
5891
5892	spa->spa_uberblock.ub_version = version;
5893	vdev_config_dirty(spa->spa_root_vdev);
5894	spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5895}
5896
5897/*
5898 * Set zpool properties.
5899 */
5900static void
5901spa_sync_props(void *arg, dmu_tx_t *tx)
5902{
5903	nvlist_t *nvp = arg;
5904	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5905	objset_t *mos = spa->spa_meta_objset;
5906	nvpair_t *elem = NULL;
5907
5908	mutex_enter(&spa->spa_props_lock);
5909
5910	while ((elem = nvlist_next_nvpair(nvp, elem))) {
5911		uint64_t intval;
5912		char *strval, *fname;
5913		zpool_prop_t prop;
5914		const char *propname;
5915		zprop_type_t proptype;
5916		zfeature_info_t *feature;
5917
5918		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5919		case ZPROP_INVAL:
5920			/*
5921			 * We checked this earlier in spa_prop_validate().
5922			 */
5923			ASSERT(zpool_prop_feature(nvpair_name(elem)));
5924
5925			fname = strchr(nvpair_name(elem), '@') + 1;
5926			VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
5927
5928			spa_feature_enable(spa, feature, tx);
5929			spa_history_log_internal(spa, "set", tx,
5930			    "%s=enabled", nvpair_name(elem));
5931			break;
5932
5933		case ZPOOL_PROP_VERSION:
5934			VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5935			/*
5936			 * The version is synced seperatly before other
5937			 * properties and should be correct by now.
5938			 */
5939			ASSERT3U(spa_version(spa), >=, intval);
5940			break;
5941
5942		case ZPOOL_PROP_ALTROOT:
5943			/*
5944			 * 'altroot' is a non-persistent property. It should
5945			 * have been set temporarily at creation or import time.
5946			 */
5947			ASSERT(spa->spa_root != NULL);
5948			break;
5949
5950		case ZPOOL_PROP_READONLY:
5951		case ZPOOL_PROP_CACHEFILE:
5952			/*
5953			 * 'readonly' and 'cachefile' are also non-persisitent
5954			 * properties.
5955			 */
5956			break;
5957		case ZPOOL_PROP_COMMENT:
5958			VERIFY(nvpair_value_string(elem, &strval) == 0);
5959			if (spa->spa_comment != NULL)
5960				spa_strfree(spa->spa_comment);
5961			spa->spa_comment = spa_strdup(strval);
5962			/*
5963			 * We need to dirty the configuration on all the vdevs
5964			 * so that their labels get updated.  It's unnecessary
5965			 * to do this for pool creation since the vdev's
5966			 * configuratoin has already been dirtied.
5967			 */
5968			if (tx->tx_txg != TXG_INITIAL)
5969				vdev_config_dirty(spa->spa_root_vdev);
5970			spa_history_log_internal(spa, "set", tx,
5971			    "%s=%s", nvpair_name(elem), strval);
5972			break;
5973		default:
5974			/*
5975			 * Set pool property values in the poolprops mos object.
5976			 */
5977			if (spa->spa_pool_props_object == 0) {
5978				spa->spa_pool_props_object =
5979				    zap_create_link(mos, DMU_OT_POOL_PROPS,
5980				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5981				    tx);
5982			}
5983
5984			/* normalize the property name */
5985			propname = zpool_prop_to_name(prop);
5986			proptype = zpool_prop_get_type(prop);
5987
5988			if (nvpair_type(elem) == DATA_TYPE_STRING) {
5989				ASSERT(proptype == PROP_TYPE_STRING);
5990				VERIFY(nvpair_value_string(elem, &strval) == 0);
5991				VERIFY(zap_update(mos,
5992				    spa->spa_pool_props_object, propname,
5993				    1, strlen(strval) + 1, strval, tx) == 0);
5994				spa_history_log_internal(spa, "set", tx,
5995				    "%s=%s", nvpair_name(elem), strval);
5996			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5997				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5998
5999				if (proptype == PROP_TYPE_INDEX) {
6000					const char *unused;
6001					VERIFY(zpool_prop_index_to_string(
6002					    prop, intval, &unused) == 0);
6003				}
6004				VERIFY(zap_update(mos,
6005				    spa->spa_pool_props_object, propname,
6006				    8, 1, &intval, tx) == 0);
6007				spa_history_log_internal(spa, "set", tx,
6008				    "%s=%lld", nvpair_name(elem), intval);
6009			} else {
6010				ASSERT(0); /* not allowed */
6011			}
6012
6013			switch (prop) {
6014			case ZPOOL_PROP_DELEGATION:
6015				spa->spa_delegation = intval;
6016				break;
6017			case ZPOOL_PROP_BOOTFS:
6018				spa->spa_bootfs = intval;
6019				break;
6020			case ZPOOL_PROP_FAILUREMODE:
6021				spa->spa_failmode = intval;
6022				break;
6023			case ZPOOL_PROP_AUTOEXPAND:
6024				spa->spa_autoexpand = intval;
6025				if (tx->tx_txg != TXG_INITIAL)
6026					spa_async_request(spa,
6027					    SPA_ASYNC_AUTOEXPAND);
6028				break;
6029			case ZPOOL_PROP_DEDUPDITTO:
6030				spa->spa_dedup_ditto = intval;
6031				break;
6032			default:
6033				break;
6034			}
6035		}
6036
6037	}
6038
6039	mutex_exit(&spa->spa_props_lock);
6040}
6041
6042/*
6043 * Perform one-time upgrade on-disk changes.  spa_version() does not
6044 * reflect the new version this txg, so there must be no changes this
6045 * txg to anything that the upgrade code depends on after it executes.
6046 * Therefore this must be called after dsl_pool_sync() does the sync
6047 * tasks.
6048 */
6049static void
6050spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6051{
6052	dsl_pool_t *dp = spa->spa_dsl_pool;
6053
6054	ASSERT(spa->spa_sync_pass == 1);
6055
6056	rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6057
6058	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6059