spa.c revision 7fdd916c474ea52896c671bbe7b56ba34a1ca132
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	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1231
1232	/*
1233	 * Close all vdevs.
1234	 */
1235	if (spa->spa_root_vdev)
1236		vdev_free(spa->spa_root_vdev);
1237	ASSERT(spa->spa_root_vdev == NULL);
1238
1239	/*
1240	 * Close the dsl pool.
1241	 */
1242	if (spa->spa_dsl_pool) {
1243		dsl_pool_close(spa->spa_dsl_pool);
1244		spa->spa_dsl_pool = NULL;
1245		spa->spa_meta_objset = NULL;
1246	}
1247
1248	ddt_unload(spa);
1249
1250
1251	/*
1252	 * Drop and purge level 2 cache
1253	 */
1254	spa_l2cache_drop(spa);
1255
1256	for (i = 0; i < spa->spa_spares.sav_count; i++)
1257		vdev_free(spa->spa_spares.sav_vdevs[i]);
1258	if (spa->spa_spares.sav_vdevs) {
1259		kmem_free(spa->spa_spares.sav_vdevs,
1260		    spa->spa_spares.sav_count * sizeof (void *));
1261		spa->spa_spares.sav_vdevs = NULL;
1262	}
1263	if (spa->spa_spares.sav_config) {
1264		nvlist_free(spa->spa_spares.sav_config);
1265		spa->spa_spares.sav_config = NULL;
1266	}
1267	spa->spa_spares.sav_count = 0;
1268
1269	for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1270		vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1271		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1272	}
1273	if (spa->spa_l2cache.sav_vdevs) {
1274		kmem_free(spa->spa_l2cache.sav_vdevs,
1275		    spa->spa_l2cache.sav_count * sizeof (void *));
1276		spa->spa_l2cache.sav_vdevs = NULL;
1277	}
1278	if (spa->spa_l2cache.sav_config) {
1279		nvlist_free(spa->spa_l2cache.sav_config);
1280		spa->spa_l2cache.sav_config = NULL;
1281	}
1282	spa->spa_l2cache.sav_count = 0;
1283
1284	spa->spa_async_suspended = 0;
1285
1286	if (spa->spa_comment != NULL) {
1287		spa_strfree(spa->spa_comment);
1288		spa->spa_comment = NULL;
1289	}
1290
1291	spa_config_exit(spa, SCL_ALL, FTAG);
1292}
1293
1294/*
1295 * Load (or re-load) the current list of vdevs describing the active spares for
1296 * this pool.  When this is called, we have some form of basic information in
1297 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1298 * then re-generate a more complete list including status information.
1299 */
1300static void
1301spa_load_spares(spa_t *spa)
1302{
1303	nvlist_t **spares;
1304	uint_t nspares;
1305	int i;
1306	vdev_t *vd, *tvd;
1307
1308	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1309
1310	/*
1311	 * First, close and free any existing spare vdevs.
1312	 */
1313	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1314		vd = spa->spa_spares.sav_vdevs[i];
1315
1316		/* Undo the call to spa_activate() below */
1317		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1318		    B_FALSE)) != NULL && tvd->vdev_isspare)
1319			spa_spare_remove(tvd);
1320		vdev_close(vd);
1321		vdev_free(vd);
1322	}
1323
1324	if (spa->spa_spares.sav_vdevs)
1325		kmem_free(spa->spa_spares.sav_vdevs,
1326		    spa->spa_spares.sav_count * sizeof (void *));
1327
1328	if (spa->spa_spares.sav_config == NULL)
1329		nspares = 0;
1330	else
1331		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1332		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1333
1334	spa->spa_spares.sav_count = (int)nspares;
1335	spa->spa_spares.sav_vdevs = NULL;
1336
1337	if (nspares == 0)
1338		return;
1339
1340	/*
1341	 * Construct the array of vdevs, opening them to get status in the
1342	 * process.   For each spare, there is potentially two different vdev_t
1343	 * structures associated with it: one in the list of spares (used only
1344	 * for basic validation purposes) and one in the active vdev
1345	 * configuration (if it's spared in).  During this phase we open and
1346	 * validate each vdev on the spare list.  If the vdev also exists in the
1347	 * active configuration, then we also mark this vdev as an active spare.
1348	 */
1349	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1350	    KM_SLEEP);
1351	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1352		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1353		    VDEV_ALLOC_SPARE) == 0);
1354		ASSERT(vd != NULL);
1355
1356		spa->spa_spares.sav_vdevs[i] = vd;
1357
1358		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1359		    B_FALSE)) != NULL) {
1360			if (!tvd->vdev_isspare)
1361				spa_spare_add(tvd);
1362
1363			/*
1364			 * We only mark the spare active if we were successfully
1365			 * able to load the vdev.  Otherwise, importing a pool
1366			 * with a bad active spare would result in strange
1367			 * behavior, because multiple pool would think the spare
1368			 * is actively in use.
1369			 *
1370			 * There is a vulnerability here to an equally bizarre
1371			 * circumstance, where a dead active spare is later
1372			 * brought back to life (onlined or otherwise).  Given
1373			 * the rarity of this scenario, and the extra complexity
1374			 * it adds, we ignore the possibility.
1375			 */
1376			if (!vdev_is_dead(tvd))
1377				spa_spare_activate(tvd);
1378		}
1379
1380		vd->vdev_top = vd;
1381		vd->vdev_aux = &spa->spa_spares;
1382
1383		if (vdev_open(vd) != 0)
1384			continue;
1385
1386		if (vdev_validate_aux(vd) == 0)
1387			spa_spare_add(vd);
1388	}
1389
1390	/*
1391	 * Recompute the stashed list of spares, with status information
1392	 * this time.
1393	 */
1394	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1395	    DATA_TYPE_NVLIST_ARRAY) == 0);
1396
1397	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1398	    KM_SLEEP);
1399	for (i = 0; i < spa->spa_spares.sav_count; i++)
1400		spares[i] = vdev_config_generate(spa,
1401		    spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1402	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1403	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1404	for (i = 0; i < spa->spa_spares.sav_count; i++)
1405		nvlist_free(spares[i]);
1406	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1407}
1408
1409/*
1410 * Load (or re-load) the current list of vdevs describing the active l2cache for
1411 * this pool.  When this is called, we have some form of basic information in
1412 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1413 * then re-generate a more complete list including status information.
1414 * Devices which are already active have their details maintained, and are
1415 * not re-opened.
1416 */
1417static void
1418spa_load_l2cache(spa_t *spa)
1419{
1420	nvlist_t **l2cache;
1421	uint_t nl2cache;
1422	int i, j, oldnvdevs;
1423	uint64_t guid;
1424	vdev_t *vd, **oldvdevs, **newvdevs;
1425	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1426
1427	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1428
1429	if (sav->sav_config != NULL) {
1430		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1431		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1432		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1433	} else {
1434		nl2cache = 0;
1435		newvdevs = NULL;
1436	}
1437
1438	oldvdevs = sav->sav_vdevs;
1439	oldnvdevs = sav->sav_count;
1440	sav->sav_vdevs = NULL;
1441	sav->sav_count = 0;
1442
1443	/*
1444	 * Process new nvlist of vdevs.
1445	 */
1446	for (i = 0; i < nl2cache; i++) {
1447		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1448		    &guid) == 0);
1449
1450		newvdevs[i] = NULL;
1451		for (j = 0; j < oldnvdevs; j++) {
1452			vd = oldvdevs[j];
1453			if (vd != NULL && guid == vd->vdev_guid) {
1454				/*
1455				 * Retain previous vdev for add/remove ops.
1456				 */
1457				newvdevs[i] = vd;
1458				oldvdevs[j] = NULL;
1459				break;
1460			}
1461		}
1462
1463		if (newvdevs[i] == NULL) {
1464			/*
1465			 * Create new vdev
1466			 */
1467			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1468			    VDEV_ALLOC_L2CACHE) == 0);
1469			ASSERT(vd != NULL);
1470			newvdevs[i] = vd;
1471
1472			/*
1473			 * Commit this vdev as an l2cache device,
1474			 * even if it fails to open.
1475			 */
1476			spa_l2cache_add(vd);
1477
1478			vd->vdev_top = vd;
1479			vd->vdev_aux = sav;
1480
1481			spa_l2cache_activate(vd);
1482
1483			if (vdev_open(vd) != 0)
1484				continue;
1485
1486			(void) vdev_validate_aux(vd);
1487
1488			if (!vdev_is_dead(vd))
1489				l2arc_add_vdev(spa, vd);
1490		}
1491	}
1492
1493	/*
1494	 * Purge vdevs that were dropped
1495	 */
1496	for (i = 0; i < oldnvdevs; i++) {
1497		uint64_t pool;
1498
1499		vd = oldvdevs[i];
1500		if (vd != NULL) {
1501			ASSERT(vd->vdev_isl2cache);
1502
1503			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1504			    pool != 0ULL && l2arc_vdev_present(vd))
1505				l2arc_remove_vdev(vd);
1506			vdev_clear_stats(vd);
1507			vdev_free(vd);
1508		}
1509	}
1510
1511	if (oldvdevs)
1512		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1513
1514	if (sav->sav_config == NULL)
1515		goto out;
1516
1517	sav->sav_vdevs = newvdevs;
1518	sav->sav_count = (int)nl2cache;
1519
1520	/*
1521	 * Recompute the stashed list of l2cache devices, with status
1522	 * information this time.
1523	 */
1524	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1525	    DATA_TYPE_NVLIST_ARRAY) == 0);
1526
1527	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1528	for (i = 0; i < sav->sav_count; i++)
1529		l2cache[i] = vdev_config_generate(spa,
1530		    sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1531	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1532	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1533out:
1534	for (i = 0; i < sav->sav_count; i++)
1535		nvlist_free(l2cache[i]);
1536	if (sav->sav_count)
1537		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1538}
1539
1540static int
1541load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1542{
1543	dmu_buf_t *db;
1544	char *packed = NULL;
1545	size_t nvsize = 0;
1546	int error;
1547	*value = NULL;
1548
1549	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1550	nvsize = *(uint64_t *)db->db_data;
1551	dmu_buf_rele(db, FTAG);
1552
1553	packed = kmem_alloc(nvsize, KM_SLEEP);
1554	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1555	    DMU_READ_PREFETCH);
1556	if (error == 0)
1557		error = nvlist_unpack(packed, nvsize, value, 0);
1558	kmem_free(packed, nvsize);
1559
1560	return (error);
1561}
1562
1563/*
1564 * Checks to see if the given vdev could not be opened, in which case we post a
1565 * sysevent to notify the autoreplace code that the device has been removed.
1566 */
1567static void
1568spa_check_removed(vdev_t *vd)
1569{
1570	for (int c = 0; c < vd->vdev_children; c++)
1571		spa_check_removed(vd->vdev_child[c]);
1572
1573	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1574	    !vd->vdev_ishole) {
1575		zfs_post_autoreplace(vd->vdev_spa, vd);
1576		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1577	}
1578}
1579
1580/*
1581 * Validate the current config against the MOS config
1582 */
1583static boolean_t
1584spa_config_valid(spa_t *spa, nvlist_t *config)
1585{
1586	vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1587	nvlist_t *nv;
1588
1589	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1590
1591	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1592	VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1593
1594	ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1595
1596	/*
1597	 * If we're doing a normal import, then build up any additional
1598	 * diagnostic information about missing devices in this config.
1599	 * We'll pass this up to the user for further processing.
1600	 */
1601	if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1602		nvlist_t **child, *nv;
1603		uint64_t idx = 0;
1604
1605		child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1606		    KM_SLEEP);
1607		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1608
1609		for (int c = 0; c < rvd->vdev_children; c++) {
1610			vdev_t *tvd = rvd->vdev_child[c];
1611			vdev_t *mtvd  = mrvd->vdev_child[c];
1612
1613			if (tvd->vdev_ops == &vdev_missing_ops &&
1614			    mtvd->vdev_ops != &vdev_missing_ops &&
1615			    mtvd->vdev_islog)
1616				child[idx++] = vdev_config_generate(spa, mtvd,
1617				    B_FALSE, 0);
1618		}
1619
1620		if (idx) {
1621			VERIFY(nvlist_add_nvlist_array(nv,
1622			    ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1623			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1624			    ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1625
1626			for (int i = 0; i < idx; i++)
1627				nvlist_free(child[i]);
1628		}
1629		nvlist_free(nv);
1630		kmem_free(child, rvd->vdev_children * sizeof (char **));
1631	}
1632
1633	/*
1634	 * Compare the root vdev tree with the information we have
1635	 * from the MOS config (mrvd). Check each top-level vdev
1636	 * with the corresponding MOS config top-level (mtvd).
1637	 */
1638	for (int c = 0; c < rvd->vdev_children; c++) {
1639		vdev_t *tvd = rvd->vdev_child[c];
1640		vdev_t *mtvd  = mrvd->vdev_child[c];
1641
1642		/*
1643		 * Resolve any "missing" vdevs in the current configuration.
1644		 * If we find that the MOS config has more accurate information
1645		 * about the top-level vdev then use that vdev instead.
1646		 */
1647		if (tvd->vdev_ops == &vdev_missing_ops &&
1648		    mtvd->vdev_ops != &vdev_missing_ops) {
1649
1650			if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1651				continue;
1652
1653			/*
1654			 * Device specific actions.
1655			 */
1656			if (mtvd->vdev_islog) {
1657				spa_set_log_state(spa, SPA_LOG_CLEAR);
1658			} else {
1659				/*
1660				 * XXX - once we have 'readonly' pool
1661				 * support we should be able to handle
1662				 * missing data devices by transitioning
1663				 * the pool to readonly.
1664				 */
1665				continue;
1666			}
1667
1668			/*
1669			 * Swap the missing vdev with the data we were
1670			 * able to obtain from the MOS config.
1671			 */
1672			vdev_remove_child(rvd, tvd);
1673			vdev_remove_child(mrvd, mtvd);
1674
1675			vdev_add_child(rvd, mtvd);
1676			vdev_add_child(mrvd, tvd);
1677
1678			spa_config_exit(spa, SCL_ALL, FTAG);
1679			vdev_load(mtvd);
1680			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1681
1682			vdev_reopen(rvd);
1683		} else if (mtvd->vdev_islog) {
1684			/*
1685			 * Load the slog device's state from the MOS config
1686			 * since it's possible that the label does not
1687			 * contain the most up-to-date information.
1688			 */
1689			vdev_load_log_state(tvd, mtvd);
1690			vdev_reopen(tvd);
1691		}
1692	}
1693	vdev_free(mrvd);
1694	spa_config_exit(spa, SCL_ALL, FTAG);
1695
1696	/*
1697	 * Ensure we were able to validate the config.
1698	 */
1699	return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1700}
1701
1702/*
1703 * Check for missing log devices
1704 */
1705static boolean_t
1706spa_check_logs(spa_t *spa)
1707{
1708	boolean_t rv = B_FALSE;
1709
1710	switch (spa->spa_log_state) {
1711	case SPA_LOG_MISSING:
1712		/* need to recheck in case slog has been restored */
1713	case SPA_LOG_UNKNOWN:
1714		rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1715		    NULL, DS_FIND_CHILDREN) != 0);
1716		if (rv)
1717			spa_set_log_state(spa, SPA_LOG_MISSING);
1718		break;
1719	}
1720	return (rv);
1721}
1722
1723static boolean_t
1724spa_passivate_log(spa_t *spa)
1725{
1726	vdev_t *rvd = spa->spa_root_vdev;
1727	boolean_t slog_found = B_FALSE;
1728
1729	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1730
1731	if (!spa_has_slogs(spa))
1732		return (B_FALSE);
1733
1734	for (int c = 0; c < rvd->vdev_children; c++) {
1735		vdev_t *tvd = rvd->vdev_child[c];
1736		metaslab_group_t *mg = tvd->vdev_mg;
1737
1738		if (tvd->vdev_islog) {
1739			metaslab_group_passivate(mg);
1740			slog_found = B_TRUE;
1741		}
1742	}
1743
1744	return (slog_found);
1745}
1746
1747static void
1748spa_activate_log(spa_t *spa)
1749{
1750	vdev_t *rvd = spa->spa_root_vdev;
1751
1752	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1753
1754	for (int c = 0; c < rvd->vdev_children; c++) {
1755		vdev_t *tvd = rvd->vdev_child[c];
1756		metaslab_group_t *mg = tvd->vdev_mg;
1757
1758		if (tvd->vdev_islog)
1759			metaslab_group_activate(mg);
1760	}
1761}
1762
1763int
1764spa_offline_log(spa_t *spa)
1765{
1766	int error;
1767
1768	error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1769	    NULL, DS_FIND_CHILDREN);
1770	if (error == 0) {
1771		/*
1772		 * We successfully offlined the log device, sync out the
1773		 * current txg so that the "stubby" block can be removed
1774		 * by zil_sync().
1775		 */
1776		txg_wait_synced(spa->spa_dsl_pool, 0);
1777	}
1778	return (error);
1779}
1780
1781static void
1782spa_aux_check_removed(spa_aux_vdev_t *sav)
1783{
1784	for (int i = 0; i < sav->sav_count; i++)
1785		spa_check_removed(sav->sav_vdevs[i]);
1786}
1787
1788void
1789spa_claim_notify(zio_t *zio)
1790{
1791	spa_t *spa = zio->io_spa;
1792
1793	if (zio->io_error)
1794		return;
1795
1796	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1797	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1798		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1799	mutex_exit(&spa->spa_props_lock);
1800}
1801
1802typedef struct spa_load_error {
1803	uint64_t	sle_meta_count;
1804	uint64_t	sle_data_count;
1805} spa_load_error_t;
1806
1807static void
1808spa_load_verify_done(zio_t *zio)
1809{
1810	blkptr_t *bp = zio->io_bp;
1811	spa_load_error_t *sle = zio->io_private;
1812	dmu_object_type_t type = BP_GET_TYPE(bp);
1813	int error = zio->io_error;
1814
1815	if (error) {
1816		if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1817		    type != DMU_OT_INTENT_LOG)
1818			atomic_add_64(&sle->sle_meta_count, 1);
1819		else
1820			atomic_add_64(&sle->sle_data_count, 1);
1821	}
1822	zio_data_buf_free(zio->io_data, zio->io_size);
1823}
1824
1825/*ARGSUSED*/
1826static int
1827spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1828    const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1829{
1830	if (bp != NULL) {
1831		zio_t *rio = arg;
1832		size_t size = BP_GET_PSIZE(bp);
1833		void *data = zio_data_buf_alloc(size);
1834
1835		zio_nowait(zio_read(rio, spa, bp, data, size,
1836		    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1837		    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1838		    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1839	}
1840	return (0);
1841}
1842
1843static int
1844spa_load_verify(spa_t *spa)
1845{
1846	zio_t *rio;
1847	spa_load_error_t sle = { 0 };
1848	zpool_rewind_policy_t policy;
1849	boolean_t verify_ok = B_FALSE;
1850	int error;
1851
1852	zpool_get_rewind_policy(spa->spa_config, &policy);
1853
1854	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1855		return (0);
1856
1857	rio = zio_root(spa, NULL, &sle,
1858	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1859
1860	error = traverse_pool(spa, spa->spa_verify_min_txg,
1861	    TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1862
1863	(void) zio_wait(rio);
1864
1865	spa->spa_load_meta_errors = sle.sle_meta_count;
1866	spa->spa_load_data_errors = sle.sle_data_count;
1867
1868	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1869	    sle.sle_data_count <= policy.zrp_maxdata) {
1870		int64_t loss = 0;
1871
1872		verify_ok = B_TRUE;
1873		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1874		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1875
1876		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1877		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1878		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1879		VERIFY(nvlist_add_int64(spa->spa_load_info,
1880		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1881		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1882		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1883	} else {
1884		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1885	}
1886
1887	if (error) {
1888		if (error != ENXIO && error != EIO)
1889			error = SET_ERROR(EIO);
1890		return (error);
1891	}
1892
1893	return (verify_ok ? 0 : EIO);
1894}
1895
1896/*
1897 * Find a value in the pool props object.
1898 */
1899static void
1900spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1901{
1902	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1903	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1904}
1905
1906/*
1907 * Find a value in the pool directory object.
1908 */
1909static int
1910spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1911{
1912	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1913	    name, sizeof (uint64_t), 1, val));
1914}
1915
1916static int
1917spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1918{
1919	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1920	return (err);
1921}
1922
1923/*
1924 * Fix up config after a partly-completed split.  This is done with the
1925 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1926 * pool have that entry in their config, but only the splitting one contains
1927 * a list of all the guids of the vdevs that are being split off.
1928 *
1929 * This function determines what to do with that list: either rejoin
1930 * all the disks to the pool, or complete the splitting process.  To attempt
1931 * the rejoin, each disk that is offlined is marked online again, and
1932 * we do a reopen() call.  If the vdev label for every disk that was
1933 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1934 * then we call vdev_split() on each disk, and complete the split.
1935 *
1936 * Otherwise we leave the config alone, with all the vdevs in place in
1937 * the original pool.
1938 */
1939static void
1940spa_try_repair(spa_t *spa, nvlist_t *config)
1941{
1942	uint_t extracted;
1943	uint64_t *glist;
1944	uint_t i, gcount;
1945	nvlist_t *nvl;
1946	vdev_t **vd;
1947	boolean_t attempt_reopen;
1948
1949	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1950		return;
1951
1952	/* check that the config is complete */
1953	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1954	    &glist, &gcount) != 0)
1955		return;
1956
1957	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1958
1959	/* attempt to online all the vdevs & validate */
1960	attempt_reopen = B_TRUE;
1961	for (i = 0; i < gcount; i++) {
1962		if (glist[i] == 0)	/* vdev is hole */
1963			continue;
1964
1965		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1966		if (vd[i] == NULL) {
1967			/*
1968			 * Don't bother attempting to reopen the disks;
1969			 * just do the split.
1970			 */
1971			attempt_reopen = B_FALSE;
1972		} else {
1973			/* attempt to re-online it */
1974			vd[i]->vdev_offline = B_FALSE;
1975		}
1976	}
1977
1978	if (attempt_reopen) {
1979		vdev_reopen(spa->spa_root_vdev);
1980
1981		/* check each device to see what state it's in */
1982		for (extracted = 0, i = 0; i < gcount; i++) {
1983			if (vd[i] != NULL &&
1984			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1985				break;
1986			++extracted;
1987		}
1988	}
1989
1990	/*
1991	 * If every disk has been moved to the new pool, or if we never
1992	 * even attempted to look at them, then we split them off for
1993	 * good.
1994	 */
1995	if (!attempt_reopen || gcount == extracted) {
1996		for (i = 0; i < gcount; i++)
1997			if (vd[i] != NULL)
1998				vdev_split(vd[i]);
1999		vdev_reopen(spa->spa_root_vdev);
2000	}
2001
2002	kmem_free(vd, gcount * sizeof (vdev_t *));
2003}
2004
2005static int
2006spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2007    boolean_t mosconfig)
2008{
2009	nvlist_t *config = spa->spa_config;
2010	char *ereport = FM_EREPORT_ZFS_POOL;
2011	char *comment;
2012	int error;
2013	uint64_t pool_guid;
2014	nvlist_t *nvl;
2015
2016	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2017		return (SET_ERROR(EINVAL));
2018
2019	ASSERT(spa->spa_comment == NULL);
2020	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2021		spa->spa_comment = spa_strdup(comment);
2022
2023	/*
2024	 * Versioning wasn't explicitly added to the label until later, so if
2025	 * it's not present treat it as the initial version.
2026	 */
2027	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2028	    &spa->spa_ubsync.ub_version) != 0)
2029		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2030
2031	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2032	    &spa->spa_config_txg);
2033
2034	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2035	    spa_guid_exists(pool_guid, 0)) {
2036		error = SET_ERROR(EEXIST);
2037	} else {
2038		spa->spa_config_guid = pool_guid;
2039
2040		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2041		    &nvl) == 0) {
2042			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2043			    KM_SLEEP) == 0);
2044		}
2045
2046		nvlist_free(spa->spa_load_info);
2047		spa->spa_load_info = fnvlist_alloc();
2048
2049		gethrestime(&spa->spa_loaded_ts);
2050		error = spa_load_impl(spa, pool_guid, config, state, type,
2051		    mosconfig, &ereport);
2052	}
2053
2054	spa->spa_minref = refcount_count(&spa->spa_refcount);
2055	if (error) {
2056		if (error != EEXIST) {
2057			spa->spa_loaded_ts.tv_sec = 0;
2058			spa->spa_loaded_ts.tv_nsec = 0;
2059		}
2060		if (error != EBADF) {
2061			zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2062		}
2063	}
2064	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2065	spa->spa_ena = 0;
2066
2067	return (error);
2068}
2069
2070/*
2071 * Load an existing storage pool, using the pool's builtin spa_config as a
2072 * source of configuration information.
2073 */
2074static int
2075spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2076    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2077    char **ereport)
2078{
2079	int error = 0;
2080	nvlist_t *nvroot = NULL;
2081	nvlist_t *label;
2082	vdev_t *rvd;
2083	uberblock_t *ub = &spa->spa_uberblock;
2084	uint64_t children, config_cache_txg = spa->spa_config_txg;
2085	int orig_mode = spa->spa_mode;
2086	int parse;
2087	uint64_t obj;
2088	boolean_t missing_feat_write = B_FALSE;
2089
2090	/*
2091	 * If this is an untrusted config, access the pool in read-only mode.
2092	 * This prevents things like resilvering recently removed devices.
2093	 */
2094	if (!mosconfig)
2095		spa->spa_mode = FREAD;
2096
2097	ASSERT(MUTEX_HELD(&spa_namespace_lock));
2098
2099	spa->spa_load_state = state;
2100
2101	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2102		return (SET_ERROR(EINVAL));
2103
2104	parse = (type == SPA_IMPORT_EXISTING ?
2105	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2106
2107	/*
2108	 * Create "The Godfather" zio to hold all async IOs
2109	 */
2110	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2111	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2112
2113	/*
2114	 * Parse the configuration into a vdev tree.  We explicitly set the
2115	 * value that will be returned by spa_version() since parsing the
2116	 * configuration requires knowing the version number.
2117	 */
2118	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2119	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2120	spa_config_exit(spa, SCL_ALL, FTAG);
2121
2122	if (error != 0)
2123		return (error);
2124
2125	ASSERT(spa->spa_root_vdev == rvd);
2126
2127	if (type != SPA_IMPORT_ASSEMBLE) {
2128		ASSERT(spa_guid(spa) == pool_guid);
2129	}
2130
2131	/*
2132	 * Try to open all vdevs, loading each label in the process.
2133	 */
2134	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2135	error = vdev_open(rvd);
2136	spa_config_exit(spa, SCL_ALL, FTAG);
2137	if (error != 0)
2138		return (error);
2139
2140	/*
2141	 * We need to validate the vdev labels against the configuration that
2142	 * we have in hand, which is dependent on the setting of mosconfig. If
2143	 * mosconfig is true then we're validating the vdev labels based on
2144	 * that config.  Otherwise, we're validating against the cached config
2145	 * (zpool.cache) that was read when we loaded the zfs module, and then
2146	 * later we will recursively call spa_load() and validate against
2147	 * the vdev config.
2148	 *
2149	 * If we're assembling a new pool that's been split off from an
2150	 * existing pool, the labels haven't yet been updated so we skip
2151	 * validation for now.
2152	 */
2153	if (type != SPA_IMPORT_ASSEMBLE) {
2154		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2155		error = vdev_validate(rvd, mosconfig);
2156		spa_config_exit(spa, SCL_ALL, FTAG);
2157
2158		if (error != 0)
2159			return (error);
2160
2161		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2162			return (SET_ERROR(ENXIO));
2163	}
2164
2165	/*
2166	 * Find the best uberblock.
2167	 */
2168	vdev_uberblock_load(rvd, ub, &label);
2169
2170	/*
2171	 * If we weren't able to find a single valid uberblock, return failure.
2172	 */
2173	if (ub->ub_txg == 0) {
2174		nvlist_free(label);
2175		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2176	}
2177
2178	/*
2179	 * If the pool has an unsupported version we can't open it.
2180	 */
2181	if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2182		nvlist_free(label);
2183		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2184	}
2185
2186	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2187		nvlist_t *features;
2188
2189		/*
2190		 * If we weren't able to find what's necessary for reading the
2191		 * MOS in the label, return failure.
2192		 */
2193		if (label == NULL || nvlist_lookup_nvlist(label,
2194		    ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2195			nvlist_free(label);
2196			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2197			    ENXIO));
2198		}
2199
2200		/*
2201		 * Update our in-core representation with the definitive values
2202		 * from the label.
2203		 */
2204		nvlist_free(spa->spa_label_features);
2205		VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2206	}
2207
2208	nvlist_free(label);
2209
2210	/*
2211	 * Look through entries in the label nvlist's features_for_read. If
2212	 * there is a feature listed there which we don't understand then we
2213	 * cannot open a pool.
2214	 */
2215	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2216		nvlist_t *unsup_feat;
2217
2218		VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2219		    0);
2220
2221		for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2222		    NULL); nvp != NULL;
2223		    nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2224			if (!zfeature_is_supported(nvpair_name(nvp))) {
2225				VERIFY(nvlist_add_string(unsup_feat,
2226				    nvpair_name(nvp), "") == 0);
2227			}
2228		}
2229
2230		if (!nvlist_empty(unsup_feat)) {
2231			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2232			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2233			nvlist_free(unsup_feat);
2234			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2235			    ENOTSUP));
2236		}
2237
2238		nvlist_free(unsup_feat);
2239	}
2240
2241	/*
2242	 * If the vdev guid sum doesn't match the uberblock, we have an
2243	 * incomplete configuration.  We first check to see if the pool
2244	 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2245	 * If it is, defer the vdev_guid_sum check till later so we
2246	 * can handle missing vdevs.
2247	 */
2248	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2249	    &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2250	    rvd->vdev_guid_sum != ub->ub_guid_sum)
2251		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2252
2253	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2254		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2255		spa_try_repair(spa, config);
2256		spa_config_exit(spa, SCL_ALL, FTAG);
2257		nvlist_free(spa->spa_config_splitting);
2258		spa->spa_config_splitting = NULL;
2259	}
2260
2261	/*
2262	 * Initialize internal SPA structures.
2263	 */
2264	spa->spa_state = POOL_STATE_ACTIVE;
2265	spa->spa_ubsync = spa->spa_uberblock;
2266	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2267	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2268	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2269	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2270	spa->spa_claim_max_txg = spa->spa_first_txg;
2271	spa->spa_prev_software_version = ub->ub_software_version;
2272
2273	error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2274	if (error)
2275		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2276	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2277
2278	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2279		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2280
2281	if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2282		boolean_t missing_feat_read = B_FALSE;
2283		nvlist_t *unsup_feat, *enabled_feat;
2284
2285		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2286		    &spa->spa_feat_for_read_obj) != 0) {
2287			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2288		}
2289
2290		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2291		    &spa->spa_feat_for_write_obj) != 0) {
2292			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2293		}
2294
2295		if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2296		    &spa->spa_feat_desc_obj) != 0) {
2297			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2298		}
2299
2300		enabled_feat = fnvlist_alloc();
2301		unsup_feat = fnvlist_alloc();
2302
2303		if (!feature_is_supported(spa->spa_meta_objset,
2304		    spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2305		    unsup_feat, enabled_feat))
2306			missing_feat_read = B_TRUE;
2307
2308		if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2309			if (!feature_is_supported(spa->spa_meta_objset,
2310			    spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2311			    unsup_feat, enabled_feat)) {
2312				missing_feat_write = B_TRUE;
2313			}
2314		}
2315
2316		fnvlist_add_nvlist(spa->spa_load_info,
2317		    ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2318
2319		if (!nvlist_empty(unsup_feat)) {
2320			fnvlist_add_nvlist(spa->spa_load_info,
2321			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2322		}
2323
2324		fnvlist_free(enabled_feat);
2325		fnvlist_free(unsup_feat);
2326
2327		if (!missing_feat_read) {
2328			fnvlist_add_boolean(spa->spa_load_info,
2329			    ZPOOL_CONFIG_CAN_RDONLY);
2330		}
2331
2332		/*
2333		 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2334		 * twofold: to determine whether the pool is available for
2335		 * import in read-write mode and (if it is not) whether the
2336		 * pool is available for import in read-only mode. If the pool
2337		 * is available for import in read-write mode, it is displayed
2338		 * as available in userland; if it is not available for import
2339		 * in read-only mode, it is displayed as unavailable in
2340		 * userland. If the pool is available for import in read-only
2341		 * mode but not read-write mode, it is displayed as unavailable
2342		 * in userland with a special note that the pool is actually
2343		 * available for open in read-only mode.
2344		 *
2345		 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2346		 * missing a feature for write, we must first determine whether
2347		 * the pool can be opened read-only before returning to
2348		 * userland in order to know whether to display the
2349		 * abovementioned note.
2350		 */
2351		if (missing_feat_read || (missing_feat_write &&
2352		    spa_writeable(spa))) {
2353			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2354			    ENOTSUP));
2355		}
2356	}
2357
2358	spa->spa_is_initializing = B_TRUE;
2359	error = dsl_pool_open(spa->spa_dsl_pool);
2360	spa->spa_is_initializing = B_FALSE;
2361	if (error != 0)
2362		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2363
2364	if (!mosconfig) {
2365		uint64_t hostid;
2366		nvlist_t *policy = NULL, *nvconfig;
2367
2368		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2369			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2370
2371		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2372		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2373			char *hostname;
2374			unsigned long myhostid = 0;
2375
2376			VERIFY(nvlist_lookup_string(nvconfig,
2377			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2378
2379#ifdef	_KERNEL
2380			myhostid = zone_get_hostid(NULL);
2381#else	/* _KERNEL */
2382			/*
2383			 * We're emulating the system's hostid in userland, so
2384			 * we can't use zone_get_hostid().
2385			 */
2386			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2387#endif	/* _KERNEL */
2388			if (hostid != 0 && myhostid != 0 &&
2389			    hostid != myhostid) {
2390				nvlist_free(nvconfig);
2391				cmn_err(CE_WARN, "pool '%s' could not be "
2392				    "loaded as it was last accessed by "
2393				    "another system (host: %s hostid: 0x%lx). "
2394				    "See: http://illumos.org/msg/ZFS-8000-EY",
2395				    spa_name(spa), hostname,
2396				    (unsigned long)hostid);
2397				return (SET_ERROR(EBADF));
2398			}
2399		}
2400		if (nvlist_lookup_nvlist(spa->spa_config,
2401		    ZPOOL_REWIND_POLICY, &policy) == 0)
2402			VERIFY(nvlist_add_nvlist(nvconfig,
2403			    ZPOOL_REWIND_POLICY, policy) == 0);
2404
2405		spa_config_set(spa, nvconfig);
2406		spa_unload(spa);
2407		spa_deactivate(spa);
2408		spa_activate(spa, orig_mode);
2409
2410		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2411	}
2412
2413	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2414		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2415	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2416	if (error != 0)
2417		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2418
2419	/*
2420	 * Load the bit that tells us to use the new accounting function
2421	 * (raid-z deflation).  If we have an older pool, this will not
2422	 * be present.
2423	 */
2424	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2425	if (error != 0 && error != ENOENT)
2426		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2427
2428	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2429	    &spa->spa_creation_version);
2430	if (error != 0 && error != ENOENT)
2431		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2432
2433	/*
2434	 * Load the persistent error log.  If we have an older pool, this will
2435	 * not be present.
2436	 */
2437	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2438	if (error != 0 && error != ENOENT)
2439		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2440
2441	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2442	    &spa->spa_errlog_scrub);
2443	if (error != 0 && error != ENOENT)
2444		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2445
2446	/*
2447	 * Load the history object.  If we have an older pool, this
2448	 * will not be present.
2449	 */
2450	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2451	if (error != 0 && error != ENOENT)
2452		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2453
2454	/*
2455	 * If we're assembling the pool from the split-off vdevs of
2456	 * an existing pool, we don't want to attach the spares & cache
2457	 * devices.
2458	 */
2459
2460	/*
2461	 * Load any hot spares for this pool.
2462	 */
2463	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2464	if (error != 0 && error != ENOENT)
2465		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2466	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2467		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2468		if (load_nvlist(spa, spa->spa_spares.sav_object,
2469		    &spa->spa_spares.sav_config) != 0)
2470			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2471
2472		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2473		spa_load_spares(spa);
2474		spa_config_exit(spa, SCL_ALL, FTAG);
2475	} else if (error == 0) {
2476		spa->spa_spares.sav_sync = B_TRUE;
2477	}
2478
2479	/*
2480	 * Load any level 2 ARC devices for this pool.
2481	 */
2482	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2483	    &spa->spa_l2cache.sav_object);
2484	if (error != 0 && error != ENOENT)
2485		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2486	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2487		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2488		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2489		    &spa->spa_l2cache.sav_config) != 0)
2490			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2491
2492		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2493		spa_load_l2cache(spa);
2494		spa_config_exit(spa, SCL_ALL, FTAG);
2495	} else if (error == 0) {
2496		spa->spa_l2cache.sav_sync = B_TRUE;
2497	}
2498
2499	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2500
2501	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2502	if (error && error != ENOENT)
2503		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2504
2505	if (error == 0) {
2506		uint64_t autoreplace;
2507
2508		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2509		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2510		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2511		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2512		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2513		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2514		    &spa->spa_dedup_ditto);
2515
2516		spa->spa_autoreplace = (autoreplace != 0);
2517	}
2518
2519	/*
2520	 * If the 'autoreplace' property is set, then post a resource notifying
2521	 * the ZFS DE that it should not issue any faults for unopenable
2522	 * devices.  We also iterate over the vdevs, and post a sysevent for any
2523	 * unopenable vdevs so that the normal autoreplace handler can take
2524	 * over.
2525	 */
2526	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2527		spa_check_removed(spa->spa_root_vdev);
2528		/*
2529		 * For the import case, this is done in spa_import(), because
2530		 * at this point we're using the spare definitions from
2531		 * the MOS config, not necessarily from the userland config.
2532		 */
2533		if (state != SPA_LOAD_IMPORT) {
2534			spa_aux_check_removed(&spa->spa_spares);
2535			spa_aux_check_removed(&spa->spa_l2cache);
2536		}
2537	}
2538
2539	/*
2540	 * Load the vdev state for all toplevel vdevs.
2541	 */
2542	vdev_load(rvd);
2543
2544	/*
2545	 * Propagate the leaf DTLs we just loaded all the way up the tree.
2546	 */
2547	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2548	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2549	spa_config_exit(spa, SCL_ALL, FTAG);
2550
2551	/*
2552	 * Load the DDTs (dedup tables).
2553	 */
2554	error = ddt_load(spa);
2555	if (error != 0)
2556		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2557
2558	spa_update_dspace(spa);
2559
2560	/*
2561	 * Validate the config, using the MOS config to fill in any
2562	 * information which might be missing.  If we fail to validate
2563	 * the config then declare the pool unfit for use. If we're
2564	 * assembling a pool from a split, the log is not transferred
2565	 * over.
2566	 */
2567	if (type != SPA_IMPORT_ASSEMBLE) {
2568		nvlist_t *nvconfig;
2569
2570		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2571			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2572
2573		if (!spa_config_valid(spa, nvconfig)) {
2574			nvlist_free(nvconfig);
2575			return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2576			    ENXIO));
2577		}
2578		nvlist_free(nvconfig);
2579
2580		/*
2581		 * Now that we've validated the config, check the state of the
2582		 * root vdev.  If it can't be opened, it indicates one or
2583		 * more toplevel vdevs are faulted.
2584		 */
2585		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2586			return (SET_ERROR(ENXIO));
2587
2588		if (spa_check_logs(spa)) {
2589			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2590			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2591		}
2592	}
2593
2594	if (missing_feat_write) {
2595		ASSERT(state == SPA_LOAD_TRYIMPORT);
2596
2597		/*
2598		 * At this point, we know that we can open the pool in
2599		 * read-only mode but not read-write mode. We now have enough
2600		 * information and can return to userland.
2601		 */
2602		return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2603	}
2604
2605	/*
2606	 * We've successfully opened the pool, verify that we're ready
2607	 * to start pushing transactions.
2608	 */
2609	if (state != SPA_LOAD_TRYIMPORT) {
2610		if (error = spa_load_verify(spa))
2611			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2612			    error));
2613	}
2614
2615	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2616	    spa->spa_load_max_txg == UINT64_MAX)) {
2617		dmu_tx_t *tx;
2618		int need_update = B_FALSE;
2619
2620		ASSERT(state != SPA_LOAD_TRYIMPORT);
2621
2622		/*
2623		 * Claim log blocks that haven't been committed yet.
2624		 * This must all happen in a single txg.
2625		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2626		 * invoked from zil_claim_log_block()'s i/o done callback.
2627		 * Price of rollback is that we abandon the log.
2628		 */
2629		spa->spa_claiming = B_TRUE;
2630
2631		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2632		    spa_first_txg(spa));
2633		(void) dmu_objset_find(spa_name(spa),
2634		    zil_claim, tx, DS_FIND_CHILDREN);
2635		dmu_tx_commit(tx);
2636
2637		spa->spa_claiming = B_FALSE;
2638
2639		spa_set_log_state(spa, SPA_LOG_GOOD);
2640		spa->spa_sync_on = B_TRUE;
2641		txg_sync_start(spa->spa_dsl_pool);
2642
2643		/*
2644		 * Wait for all claims to sync.  We sync up to the highest
2645		 * claimed log block birth time so that claimed log blocks
2646		 * don't appear to be from the future.  spa_claim_max_txg
2647		 * will have been set for us by either zil_check_log_chain()
2648		 * (invoked from spa_check_logs()) or zil_claim() above.
2649		 */
2650		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2651
2652		/*
2653		 * If the config cache is stale, or we have uninitialized
2654		 * metaslabs (see spa_vdev_add()), then update the config.
2655		 *
2656		 * If this is a verbatim import, trust the current
2657		 * in-core spa_config and update the disk labels.
2658		 */
2659		if (config_cache_txg != spa->spa_config_txg ||
2660		    state == SPA_LOAD_IMPORT ||
2661		    state == SPA_LOAD_RECOVER ||
2662		    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2663			need_update = B_TRUE;
2664
2665		for (int c = 0; c < rvd->vdev_children; c++)
2666			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2667				need_update = B_TRUE;
2668
2669		/*
2670		 * Update the config cache asychronously in case we're the
2671		 * root pool, in which case the config cache isn't writable yet.
2672		 */
2673		if (need_update)
2674			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2675
2676		/*
2677		 * Check all DTLs to see if anything needs resilvering.
2678		 */
2679		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2680		    vdev_resilver_needed(rvd, NULL, NULL))
2681			spa_async_request(spa, SPA_ASYNC_RESILVER);
2682
2683		/*
2684		 * Log the fact that we booted up (so that we can detect if
2685		 * we rebooted in the middle of an operation).
2686		 */
2687		spa_history_log_version(spa, "open");
2688
2689		/*
2690		 * Delete any inconsistent datasets.
2691		 */
2692		(void) dmu_objset_find(spa_name(spa),
2693		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2694
2695		/*
2696		 * Clean up any stale temporary dataset userrefs.
2697		 */
2698		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2699	}
2700
2701	return (0);
2702}
2703
2704static int
2705spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2706{
2707	int mode = spa->spa_mode;
2708
2709	spa_unload(spa);
2710	spa_deactivate(spa);
2711
2712	spa->spa_load_max_txg--;
2713
2714	spa_activate(spa, mode);
2715	spa_async_suspend(spa);
2716
2717	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2718}
2719
2720/*
2721 * If spa_load() fails this function will try loading prior txg's. If
2722 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2723 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2724 * function will not rewind the pool and will return the same error as
2725 * spa_load().
2726 */
2727static int
2728spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2729    uint64_t max_request, int rewind_flags)
2730{
2731	nvlist_t *loadinfo = NULL;
2732	nvlist_t *config = NULL;
2733	int load_error, rewind_error;
2734	uint64_t safe_rewind_txg;
2735	uint64_t min_txg;
2736
2737	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2738		spa->spa_load_max_txg = spa->spa_load_txg;
2739		spa_set_log_state(spa, SPA_LOG_CLEAR);
2740	} else {
2741		spa->spa_load_max_txg = max_request;
2742	}
2743
2744	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2745	    mosconfig);
2746	if (load_error == 0)
2747		return (0);
2748
2749	if (spa->spa_root_vdev != NULL)
2750		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2751
2752	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2753	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2754
2755	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2756		nvlist_free(config);
2757		return (load_error);
2758	}
2759
2760	if (state == SPA_LOAD_RECOVER) {
2761		/* Price of rolling back is discarding txgs, including log */
2762		spa_set_log_state(spa, SPA_LOG_CLEAR);
2763	} else {
2764		/*
2765		 * If we aren't rolling back save the load info from our first
2766		 * import attempt so that we can restore it after attempting
2767		 * to rewind.
2768		 */
2769		loadinfo = spa->spa_load_info;
2770		spa->spa_load_info = fnvlist_alloc();
2771	}
2772
2773	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2774	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2775	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2776	    TXG_INITIAL : safe_rewind_txg;
2777
2778	/*
2779	 * Continue as long as we're finding errors, we're still within
2780	 * the acceptable rewind range, and we're still finding uberblocks
2781	 */
2782	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2783	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2784		if (spa->spa_load_max_txg < safe_rewind_txg)
2785			spa->spa_extreme_rewind = B_TRUE;
2786		rewind_error = spa_load_retry(spa, state, mosconfig);
2787	}
2788
2789	spa->spa_extreme_rewind = B_FALSE;
2790	spa->spa_load_max_txg = UINT64_MAX;
2791
2792	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2793		spa_config_set(spa, config);
2794
2795	if (state == SPA_LOAD_RECOVER) {
2796		ASSERT3P(loadinfo, ==, NULL);
2797		return (rewind_error);
2798	} else {
2799		/* Store the rewind info as part of the initial load info */
2800		fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2801		    spa->spa_load_info);
2802
2803		/* Restore the initial load info */
2804		fnvlist_free(spa->spa_load_info);
2805		spa->spa_load_info = loadinfo;
2806
2807		return (load_error);
2808	}
2809}
2810
2811/*
2812 * Pool Open/Import
2813 *
2814 * The import case is identical to an open except that the configuration is sent
2815 * down from userland, instead of grabbed from the configuration cache.  For the
2816 * case of an open, the pool configuration will exist in the
2817 * POOL_STATE_UNINITIALIZED state.
2818 *
2819 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2820 * the same time open the pool, without having to keep around the spa_t in some
2821 * ambiguous state.
2822 */
2823static int
2824spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2825    nvlist_t **config)
2826{
2827	spa_t *spa;
2828	spa_load_state_t state = SPA_LOAD_OPEN;
2829	int error;
2830	int locked = B_FALSE;
2831
2832	*spapp = NULL;
2833
2834	/*
2835	 * As disgusting as this is, we need to support recursive calls to this
2836	 * function because dsl_dir_open() is called during spa_load(), and ends
2837	 * up calling spa_open() again.  The real fix is to figure out how to
2838	 * avoid dsl_dir_open() calling this in the first place.
2839	 */
2840	if (mutex_owner(&spa_namespace_lock) != curthread) {
2841		mutex_enter(&spa_namespace_lock);
2842		locked = B_TRUE;
2843	}
2844
2845	if ((spa = spa_lookup(pool)) == NULL) {
2846		if (locked)
2847			mutex_exit(&spa_namespace_lock);
2848		return (SET_ERROR(ENOENT));
2849	}
2850
2851	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2852		zpool_rewind_policy_t policy;
2853
2854		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2855		    &policy);
2856		if (policy.zrp_request & ZPOOL_DO_REWIND)
2857			state = SPA_LOAD_RECOVER;
2858
2859		spa_activate(spa, spa_mode_global);
2860
2861		if (state != SPA_LOAD_RECOVER)
2862			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2863
2864		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2865		    policy.zrp_request);
2866
2867		if (error == EBADF) {
2868			/*
2869			 * If vdev_validate() returns failure (indicated by
2870			 * EBADF), it indicates that one of the vdevs indicates
2871			 * that the pool has been exported or destroyed.  If
2872			 * this is the case, the config cache is out of sync and
2873			 * we should remove the pool from the namespace.
2874			 */
2875			spa_unload(spa);
2876			spa_deactivate(spa);
2877			spa_config_sync(spa, B_TRUE, B_TRUE);
2878			spa_remove(spa);
2879			if (locked)
2880				mutex_exit(&spa_namespace_lock);
2881			return (SET_ERROR(ENOENT));
2882		}
2883
2884		if (error) {
2885			/*
2886			 * We can't open the pool, but we still have useful
2887			 * information: the state of each vdev after the
2888			 * attempted vdev_open().  Return this to the user.
2889			 */
2890			if (config != NULL && spa->spa_config) {
2891				VERIFY(nvlist_dup(spa->spa_config, config,
2892				    KM_SLEEP) == 0);
2893				VERIFY(nvlist_add_nvlist(*config,
2894				    ZPOOL_CONFIG_LOAD_INFO,
2895				    spa->spa_load_info) == 0);
2896			}
2897			spa_unload(spa);
2898			spa_deactivate(spa);
2899			spa->spa_last_open_failed = error;
2900			if (locked)
2901				mutex_exit(&spa_namespace_lock);
2902			*spapp = NULL;
2903			return (error);
2904		}
2905	}
2906
2907	spa_open_ref(spa, tag);
2908
2909	if (config != NULL)
2910		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2911
2912	/*
2913	 * If we've recovered the pool, pass back any information we
2914	 * gathered while doing the load.
2915	 */
2916	if (state == SPA_LOAD_RECOVER) {
2917		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2918		    spa->spa_load_info) == 0);
2919	}
2920
2921	if (locked) {
2922		spa->spa_last_open_failed = 0;
2923		spa->spa_last_ubsync_txg = 0;
2924		spa->spa_load_txg = 0;
2925		mutex_exit(&spa_namespace_lock);
2926	}
2927
2928	*spapp = spa;
2929
2930	return (0);
2931}
2932
2933int
2934spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2935    nvlist_t **config)
2936{
2937	return (spa_open_common(name, spapp, tag, policy, config));
2938}
2939
2940int
2941spa_open(const char *name, spa_t **spapp, void *tag)
2942{
2943	return (spa_open_common(name, spapp, tag, NULL, NULL));
2944}
2945
2946/*
2947 * Lookup the given spa_t, incrementing the inject count in the process,
2948 * preventing it from being exported or destroyed.
2949 */
2950spa_t *
2951spa_inject_addref(char *name)
2952{
2953	spa_t *spa;
2954
2955	mutex_enter(&spa_namespace_lock);
2956	if ((spa = spa_lookup(name)) == NULL) {
2957		mutex_exit(&spa_namespace_lock);
2958		return (NULL);
2959	}
2960	spa->spa_inject_ref++;
2961	mutex_exit(&spa_namespace_lock);
2962
2963	return (spa);
2964}
2965
2966void
2967spa_inject_delref(spa_t *spa)
2968{
2969	mutex_enter(&spa_namespace_lock);
2970	spa->spa_inject_ref--;
2971	mutex_exit(&spa_namespace_lock);
2972}
2973
2974/*
2975 * Add spares device information to the nvlist.
2976 */
2977static void
2978spa_add_spares(spa_t *spa, nvlist_t *config)
2979{
2980	nvlist_t **spares;
2981	uint_t i, nspares;
2982	nvlist_t *nvroot;
2983	uint64_t guid;
2984	vdev_stat_t *vs;
2985	uint_t vsc;
2986	uint64_t pool;
2987
2988	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2989
2990	if (spa->spa_spares.sav_count == 0)
2991		return;
2992
2993	VERIFY(nvlist_lookup_nvlist(config,
2994	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2995	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2996	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2997	if (nspares != 0) {
2998		VERIFY(nvlist_add_nvlist_array(nvroot,
2999		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3000		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3001		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3002
3003		/*
3004		 * Go through and find any spares which have since been
3005		 * repurposed as an active spare.  If this is the case, update
3006		 * their status appropriately.
3007		 */
3008		for (i = 0; i < nspares; i++) {
3009			VERIFY(nvlist_lookup_uint64(spares[i],
3010			    ZPOOL_CONFIG_GUID, &guid) == 0);
3011			if (spa_spare_exists(guid, &pool, NULL) &&
3012			    pool != 0ULL) {
3013				VERIFY(nvlist_lookup_uint64_array(
3014				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
3015				    (uint64_t **)&vs, &vsc) == 0);
3016				vs->vs_state = VDEV_STATE_CANT_OPEN;
3017				vs->vs_aux = VDEV_AUX_SPARED;
3018			}
3019		}
3020	}
3021}
3022
3023/*
3024 * Add l2cache device information to the nvlist, including vdev stats.
3025 */
3026static void
3027spa_add_l2cache(spa_t *spa, nvlist_t *config)
3028{
3029	nvlist_t **l2cache;
3030	uint_t i, j, nl2cache;
3031	nvlist_t *nvroot;
3032	uint64_t guid;
3033	vdev_t *vd;
3034	vdev_stat_t *vs;
3035	uint_t vsc;
3036
3037	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3038
3039	if (spa->spa_l2cache.sav_count == 0)
3040		return;
3041
3042	VERIFY(nvlist_lookup_nvlist(config,
3043	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3044	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3045	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3046	if (nl2cache != 0) {
3047		VERIFY(nvlist_add_nvlist_array(nvroot,
3048		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3049		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3050		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3051
3052		/*
3053		 * Update level 2 cache device stats.
3054		 */
3055
3056		for (i = 0; i < nl2cache; i++) {
3057			VERIFY(nvlist_lookup_uint64(l2cache[i],
3058			    ZPOOL_CONFIG_GUID, &guid) == 0);
3059
3060			vd = NULL;
3061			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3062				if (guid ==
3063				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3064					vd = spa->spa_l2cache.sav_vdevs[j];
3065					break;
3066				}
3067			}
3068			ASSERT(vd != NULL);
3069
3070			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3071			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3072			    == 0);
3073			vdev_get_stats(vd, vs);
3074		}
3075	}
3076}
3077
3078static void
3079spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3080{
3081	nvlist_t *features;
3082	zap_cursor_t zc;
3083	zap_attribute_t za;
3084
3085	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3086	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3087
3088	if (spa->spa_feat_for_read_obj != 0) {
3089		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3090		    spa->spa_feat_for_read_obj);
3091		    zap_cursor_retrieve(&zc, &za) == 0;
3092		    zap_cursor_advance(&zc)) {
3093			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3094			    za.za_num_integers == 1);
3095			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3096			    za.za_first_integer));
3097		}
3098		zap_cursor_fini(&zc);
3099	}
3100
3101	if (spa->spa_feat_for_write_obj != 0) {
3102		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3103		    spa->spa_feat_for_write_obj);
3104		    zap_cursor_retrieve(&zc, &za) == 0;
3105		    zap_cursor_advance(&zc)) {
3106			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3107			    za.za_num_integers == 1);
3108			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3109			    za.za_first_integer));
3110		}
3111		zap_cursor_fini(&zc);
3112	}
3113
3114	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3115	    features) == 0);
3116	nvlist_free(features);
3117}
3118
3119int
3120spa_get_stats(const char *name, nvlist_t **config,
3121    char *altroot, size_t buflen)
3122{
3123	int error;
3124	spa_t *spa;
3125
3126	*config = NULL;
3127	error = spa_open_common(name, &spa, FTAG, NULL, config);
3128
3129	if (spa != NULL) {
3130		/*
3131		 * This still leaves a window of inconsistency where the spares
3132		 * or l2cache devices could change and the config would be
3133		 * self-inconsistent.
3134		 */
3135		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3136
3137		if (*config != NULL) {
3138			uint64_t loadtimes[2];
3139
3140			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3141			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3142			VERIFY(nvlist_add_uint64_array(*config,
3143			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3144
3145			VERIFY(nvlist_add_uint64(*config,
3146			    ZPOOL_CONFIG_ERRCOUNT,
3147			    spa_get_errlog_size(spa)) == 0);
3148
3149			if (spa_suspended(spa))
3150				VERIFY(nvlist_add_uint64(*config,
3151				    ZPOOL_CONFIG_SUSPENDED,
3152				    spa->spa_failmode) == 0);
3153
3154			spa_add_spares(spa, *config);
3155			spa_add_l2cache(spa, *config);
3156			spa_add_feature_stats(spa, *config);
3157		}
3158	}
3159
3160	/*
3161	 * We want to get the alternate root even for faulted pools, so we cheat
3162	 * and call spa_lookup() directly.
3163	 */
3164	if (altroot) {
3165		if (spa == NULL) {
3166			mutex_enter(&spa_namespace_lock);
3167			spa = spa_lookup(name);
3168			if (spa)
3169				spa_altroot(spa, altroot, buflen);
3170			else
3171				altroot[0] = '\0';
3172			spa = NULL;
3173			mutex_exit(&spa_namespace_lock);
3174		} else {
3175			spa_altroot(spa, altroot, buflen);
3176		}
3177	}
3178
3179	if (spa != NULL) {
3180		spa_config_exit(spa, SCL_CONFIG, FTAG);
3181		spa_close(spa, FTAG);
3182	}
3183
3184	return (error);
3185}
3186
3187/*
3188 * Validate that the auxiliary device array is well formed.  We must have an
3189 * array of nvlists, each which describes a valid leaf vdev.  If this is an
3190 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3191 * specified, as long as they are well-formed.
3192 */
3193static int
3194spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3195    spa_aux_vdev_t *sav, const char *config, uint64_t version,
3196    vdev_labeltype_t label)
3197{
3198	nvlist_t **dev;
3199	uint_t i, ndev;
3200	vdev_t *vd;
3201	int error;
3202
3203	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3204
3205	/*
3206	 * It's acceptable to have no devs specified.
3207	 */
3208	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3209		return (0);
3210
3211	if (ndev == 0)
3212		return (SET_ERROR(EINVAL));
3213
3214	/*
3215	 * Make sure the pool is formatted with a version that supports this
3216	 * device type.
3217	 */
3218	if (spa_version(spa) < version)
3219		return (SET_ERROR(ENOTSUP));
3220
3221	/*
3222	 * Set the pending device list so we correctly handle device in-use
3223	 * checking.
3224	 */
3225	sav->sav_pending = dev;
3226	sav->sav_npending = ndev;
3227
3228	for (i = 0; i < ndev; i++) {
3229		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3230		    mode)) != 0)
3231			goto out;
3232
3233		if (!vd->vdev_ops->vdev_op_leaf) {
3234			vdev_free(vd);
3235			error = SET_ERROR(EINVAL);
3236			goto out;
3237		}
3238
3239		/*
3240		 * The L2ARC currently only supports disk devices in
3241		 * kernel context.  For user-level testing, we allow it.
3242		 */
3243#ifdef _KERNEL
3244		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3245		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3246			error = SET_ERROR(ENOTBLK);
3247			vdev_free(vd);
3248			goto out;
3249		}
3250#endif
3251		vd->vdev_top = vd;
3252
3253		if ((error = vdev_open(vd)) == 0 &&
3254		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
3255			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3256			    vd->vdev_guid) == 0);
3257		}
3258
3259		vdev_free(vd);
3260
3261		if (error &&
3262		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3263			goto out;
3264		else
3265			error = 0;
3266	}
3267
3268out:
3269	sav->sav_pending = NULL;
3270	sav->sav_npending = 0;
3271	return (error);
3272}
3273
3274static int
3275spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3276{
3277	int error;
3278
3279	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3280
3281	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3282	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3283	    VDEV_LABEL_SPARE)) != 0) {
3284		return (error);
3285	}
3286
3287	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3288	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3289	    VDEV_LABEL_L2CACHE));
3290}
3291
3292static void
3293spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3294    const char *config)
3295{
3296	int i;
3297
3298	if (sav->sav_config != NULL) {
3299		nvlist_t **olddevs;
3300		uint_t oldndevs;
3301		nvlist_t **newdevs;
3302
3303		/*
3304		 * Generate new dev list by concatentating with the
3305		 * current dev list.
3306		 */
3307		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3308		    &olddevs, &oldndevs) == 0);
3309
3310		newdevs = kmem_alloc(sizeof (void *) *
3311		    (ndevs + oldndevs), KM_SLEEP);
3312		for (i = 0; i < oldndevs; i++)
3313			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3314			    KM_SLEEP) == 0);
3315		for (i = 0; i < ndevs; i++)
3316			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3317			    KM_SLEEP) == 0);
3318
3319		VERIFY(nvlist_remove(sav->sav_config, config,
3320		    DATA_TYPE_NVLIST_ARRAY) == 0);
3321
3322		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3323		    config, newdevs, ndevs + oldndevs) == 0);
3324		for (i = 0; i < oldndevs + ndevs; i++)
3325			nvlist_free(newdevs[i]);
3326		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3327	} else {
3328		/*
3329		 * Generate a new dev list.
3330		 */
3331		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3332		    KM_SLEEP) == 0);
3333		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3334		    devs, ndevs) == 0);
3335	}
3336}
3337
3338/*
3339 * Stop and drop level 2 ARC devices
3340 */
3341void
3342spa_l2cache_drop(spa_t *spa)
3343{
3344	vdev_t *vd;
3345	int i;
3346	spa_aux_vdev_t *sav = &spa->spa_l2cache;
3347
3348	for (i = 0; i < sav->sav_count; i++) {
3349		uint64_t pool;
3350
3351		vd = sav->sav_vdevs[i];
3352		ASSERT(vd != NULL);
3353
3354		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3355		    pool != 0ULL && l2arc_vdev_present(vd))
3356			l2arc_remove_vdev(vd);
3357	}
3358}
3359
3360/*
3361 * Pool Creation
3362 */
3363int
3364spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3365    nvlist_t *zplprops)
3366{
3367	spa_t *spa;
3368	char *altroot = NULL;
3369	vdev_t *rvd;
3370	dsl_pool_t *dp;
3371	dmu_tx_t *tx;
3372	int error = 0;
3373	uint64_t txg = TXG_INITIAL;
3374	nvlist_t **spares, **l2cache;
3375	uint_t nspares, nl2cache;
3376	uint64_t version, obj;
3377	boolean_t has_features;
3378
3379	/*
3380	 * If this pool already exists, return failure.
3381	 */
3382	mutex_enter(&spa_namespace_lock);
3383	if (spa_lookup(pool) != NULL) {
3384		mutex_exit(&spa_namespace_lock);
3385		return (SET_ERROR(EEXIST));
3386	}
3387
3388	/*
3389	 * Allocate a new spa_t structure.
3390	 */
3391	(void) nvlist_lookup_string(props,
3392	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3393	spa = spa_add(pool, NULL, altroot);
3394	spa_activate(spa, spa_mode_global);
3395
3396	if (props && (error = spa_prop_validate(spa, props))) {
3397		spa_deactivate(spa);
3398		spa_remove(spa);
3399		mutex_exit(&spa_namespace_lock);
3400		return (error);
3401	}
3402
3403	has_features = B_FALSE;
3404	for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3405	    elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3406		if (zpool_prop_feature(nvpair_name(elem)))
3407			has_features = B_TRUE;
3408	}
3409
3410	if (has_features || nvlist_lookup_uint64(props,
3411	    zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3412		version = SPA_VERSION;
3413	}
3414	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3415
3416	spa->spa_first_txg = txg;
3417	spa->spa_uberblock.ub_txg = txg - 1;
3418	spa->spa_uberblock.ub_version = version;
3419	spa->spa_ubsync = spa->spa_uberblock;
3420
3421	/*
3422	 * Create "The Godfather" zio to hold all async IOs
3423	 */
3424	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3425	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3426
3427	/*
3428	 * Create the root vdev.
3429	 */
3430	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3431
3432	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3433
3434	ASSERT(error != 0 || rvd != NULL);
3435	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3436
3437	if (error == 0 && !zfs_allocatable_devs(nvroot))
3438		error = SET_ERROR(EINVAL);
3439
3440	if (error == 0 &&
3441	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3442	    (error = spa_validate_aux(spa, nvroot, txg,
3443	    VDEV_ALLOC_ADD)) == 0) {
3444		for (int c = 0; c < rvd->vdev_children; c++) {
3445			vdev_metaslab_set_size(rvd->vdev_child[c]);
3446			vdev_expand(rvd->vdev_child[c], txg);
3447		}
3448	}
3449
3450	spa_config_exit(spa, SCL_ALL, FTAG);
3451
3452	if (error != 0) {
3453		spa_unload(spa);
3454		spa_deactivate(spa);
3455		spa_remove(spa);
3456		mutex_exit(&spa_namespace_lock);
3457		return (error);
3458	}
3459
3460	/*
3461	 * Get the list of spares, if specified.
3462	 */
3463	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3464	    &spares, &nspares) == 0) {
3465		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3466		    KM_SLEEP) == 0);
3467		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3468		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3469		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3470		spa_load_spares(spa);
3471		spa_config_exit(spa, SCL_ALL, FTAG);
3472		spa->spa_spares.sav_sync = B_TRUE;
3473	}
3474
3475	/*
3476	 * Get the list of level 2 cache devices, if specified.
3477	 */
3478	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3479	    &l2cache, &nl2cache) == 0) {
3480		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3481		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3482		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3483		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3484		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3485		spa_load_l2cache(spa);
3486		spa_config_exit(spa, SCL_ALL, FTAG);
3487		spa->spa_l2cache.sav_sync = B_TRUE;
3488	}
3489
3490	spa->spa_is_initializing = B_TRUE;
3491	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3492	spa->spa_meta_objset = dp->dp_meta_objset;
3493	spa->spa_is_initializing = B_FALSE;
3494
3495	/*
3496	 * Create DDTs (dedup tables).
3497	 */
3498	ddt_create(spa);
3499
3500	spa_update_dspace(spa);
3501
3502	tx = dmu_tx_create_assigned(dp, txg);
3503
3504	/*
3505	 * Create the pool config object.
3506	 */
3507	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3508	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3509	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3510
3511	if (zap_add(spa->spa_meta_objset,
3512	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3513	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3514		cmn_err(CE_PANIC, "failed to add pool config");
3515	}
3516
3517	if (spa_version(spa) >= SPA_VERSION_FEATURES)
3518		spa_feature_create_zap_objects(spa, tx);
3519
3520	if (zap_add(spa->spa_meta_objset,
3521	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3522	    sizeof (uint64_t), 1, &version, tx) != 0) {
3523		cmn_err(CE_PANIC, "failed to add pool version");
3524	}
3525
3526	/* Newly created pools with the right version are always deflated. */
3527	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3528		spa->spa_deflate = TRUE;
3529		if (zap_add(spa->spa_meta_objset,
3530		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3531		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3532			cmn_err(CE_PANIC, "failed to add deflate");
3533		}
3534	}
3535
3536	/*
3537	 * Create the deferred-free bpobj.  Turn off compression
3538	 * because sync-to-convergence takes longer if the blocksize
3539	 * keeps changing.
3540	 */
3541	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3542	dmu_object_set_compress(spa->spa_meta_objset, obj,
3543	    ZIO_COMPRESS_OFF, tx);
3544	if (zap_add(spa->spa_meta_objset,
3545	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3546	    sizeof (uint64_t), 1, &obj, tx) != 0) {
3547		cmn_err(CE_PANIC, "failed to add bpobj");
3548	}
3549	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3550	    spa->spa_meta_objset, obj));
3551
3552	/*
3553	 * Create the pool's history object.
3554	 */
3555	if (version >= SPA_VERSION_ZPOOL_HISTORY)
3556		spa_history_create_obj(spa, tx);
3557
3558	/*
3559	 * Set pool properties.
3560	 */
3561	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3562	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3563	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3564	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3565
3566	if (props != NULL) {
3567		spa_configfile_set(spa, props, B_FALSE);
3568		spa_sync_props(props, tx);
3569	}
3570
3571	dmu_tx_commit(tx);
3572
3573	spa->spa_sync_on = B_TRUE;
3574	txg_sync_start(spa->spa_dsl_pool);
3575
3576	/*
3577	 * We explicitly wait for the first transaction to complete so that our
3578	 * bean counters are appropriately updated.
3579	 */
3580	txg_wait_synced(spa->spa_dsl_pool, txg);
3581
3582	spa_config_sync(spa, B_FALSE, B_TRUE);
3583
3584	spa_history_log_version(spa, "create");
3585
3586	spa->spa_minref = refcount_count(&spa->spa_refcount);
3587
3588	mutex_exit(&spa_namespace_lock);
3589
3590	return (0);
3591}
3592
3593#ifdef _KERNEL
3594/*
3595 * Get the root pool information from the root disk, then import the root pool
3596 * during the system boot up time.
3597 */
3598extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3599
3600static nvlist_t *
3601spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3602{
3603	nvlist_t *config;
3604	nvlist_t *nvtop, *nvroot;
3605	uint64_t pgid;
3606
3607	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3608		return (NULL);
3609
3610	/*
3611	 * Add this top-level vdev to the child array.
3612	 */
3613	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3614	    &nvtop) == 0);
3615	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3616	    &pgid) == 0);
3617	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3618
3619	/*
3620	 * Put this pool's top-level vdevs into a root vdev.
3621	 */
3622	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3623	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3624	    VDEV_TYPE_ROOT) == 0);
3625	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3626	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3627	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3628	    &nvtop, 1) == 0);
3629
3630	/*
3631	 * Replace the existing vdev_tree with the new root vdev in
3632	 * this pool's configuration (remove the old, add the new).
3633	 */
3634	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3635	nvlist_free(nvroot);
3636	return (config);
3637}
3638
3639/*
3640 * Walk the vdev tree and see if we can find a device with "better"
3641 * configuration. A configuration is "better" if the label on that
3642 * device has a more recent txg.
3643 */
3644static void
3645spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3646{
3647	for (int c = 0; c < vd->vdev_children; c++)
3648		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3649
3650	if (vd->vdev_ops->vdev_op_leaf) {
3651		nvlist_t *label;
3652		uint64_t label_txg;
3653
3654		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3655		    &label) != 0)
3656			return;
3657
3658		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3659		    &label_txg) == 0);
3660
3661		/*
3662		 * Do we have a better boot device?
3663		 */
3664		if (label_txg > *txg) {
3665			*txg = label_txg;
3666			*avd = vd;
3667		}
3668		nvlist_free(label);
3669	}
3670}
3671
3672/*
3673 * Import a root pool.
3674 *
3675 * For x86. devpath_list will consist of devid and/or physpath name of
3676 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3677 * The GRUB "findroot" command will return the vdev we should boot.
3678 *
3679 * For Sparc, devpath_list consists the physpath name of the booting device
3680 * no matter the rootpool is a single device pool or a mirrored pool.
3681 * e.g.
3682 *	"/pci@1f,0/ide@d/disk@0,0:a"
3683 */
3684int
3685spa_import_rootpool(char *devpath, char *devid)
3686{
3687	spa_t *spa;
3688	vdev_t *rvd, *bvd, *avd = NULL;
3689	nvlist_t *config, *nvtop;
3690	uint64_t guid, txg;
3691	char *pname;
3692	int error;
3693
3694	/*
3695	 * Read the label from the boot device and generate a configuration.
3696	 */
3697	config = spa_generate_rootconf(devpath, devid, &guid);
3698#if defined(_OBP) && defined(_KERNEL)
3699	if (config == NULL) {
3700		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3701			/* iscsi boot */
3702			get_iscsi_bootpath_phy(devpath);
3703			config = spa_generate_rootconf(devpath, devid, &guid);
3704		}
3705	}
3706#endif
3707	if (config == NULL) {
3708		cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3709		    devpath);
3710		return (SET_ERROR(EIO));
3711	}
3712
3713	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3714	    &pname) == 0);
3715	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3716
3717	mutex_enter(&spa_namespace_lock);
3718	if ((spa = spa_lookup(pname)) != NULL) {
3719		/*
3720		 * Remove the existing root pool from the namespace so that we
3721		 * can replace it with the correct config we just read in.
3722		 */
3723		spa_remove(spa);
3724	}
3725
3726	spa = spa_add(pname, config, NULL);
3727	spa->spa_is_root = B_TRUE;
3728	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3729
3730	/*
3731	 * Build up a vdev tree based on the boot device's label config.
3732	 */
3733	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3734	    &nvtop) == 0);
3735	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3736	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3737	    VDEV_ALLOC_ROOTPOOL);
3738	spa_config_exit(spa, SCL_ALL, FTAG);
3739	if (error) {
3740		mutex_exit(&spa_namespace_lock);
3741		nvlist_free(config);
3742		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3743		    pname);
3744		return (error);
3745	}
3746
3747	/*
3748	 * Get the boot vdev.
3749	 */
3750	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3751		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3752		    (u_longlong_t)guid);
3753		error = SET_ERROR(ENOENT);
3754		goto out;
3755	}
3756
3757	/*
3758	 * Determine if there is a better boot device.
3759	 */
3760	avd = bvd;
3761	spa_alt_rootvdev(rvd, &avd, &txg);
3762	if (avd != bvd) {
3763		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3764		    "try booting from '%s'", avd->vdev_path);
3765		error = SET_ERROR(EINVAL);
3766		goto out;
3767	}
3768
3769	/*
3770	 * If the boot device is part of a spare vdev then ensure that
3771	 * we're booting off the active spare.
3772	 */
3773	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3774	    !bvd->vdev_isspare) {
3775		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3776		    "try booting from '%s'",
3777		    bvd->vdev_parent->
3778		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3779		error = SET_ERROR(EINVAL);
3780		goto out;
3781	}
3782
3783	error = 0;
3784out:
3785	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3786	vdev_free(rvd);
3787	spa_config_exit(spa, SCL_ALL, FTAG);
3788	mutex_exit(&spa_namespace_lock);
3789
3790	nvlist_free(config);
3791	return (error);
3792}
3793
3794#endif
3795
3796/*
3797 * Import a non-root pool into the system.
3798 */
3799int
3800spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3801{
3802	spa_t *spa;
3803	char *altroot = NULL;
3804	spa_load_state_t state = SPA_LOAD_IMPORT;
3805	zpool_rewind_policy_t policy;
3806	uint64_t mode = spa_mode_global;
3807	uint64_t readonly = B_FALSE;
3808	int error;
3809	nvlist_t *nvroot;
3810	nvlist_t **spares, **l2cache;
3811	uint_t nspares, nl2cache;
3812
3813	/*
3814	 * If a pool with this name exists, return failure.
3815	 */
3816	mutex_enter(&spa_namespace_lock);
3817	if (spa_lookup(pool) != NULL) {
3818		mutex_exit(&spa_namespace_lock);
3819		return (SET_ERROR(EEXIST));
3820	}
3821
3822	/*
3823	 * Create and initialize the spa structure.
3824	 */
3825	(void) nvlist_lookup_string(props,
3826	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3827	(void) nvlist_lookup_uint64(props,
3828	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3829	if (readonly)
3830		mode = FREAD;
3831	spa = spa_add(pool, config, altroot);
3832	spa->spa_import_flags = flags;
3833
3834	/*
3835	 * Verbatim import - Take a pool and insert it into the namespace
3836	 * as if it had been loaded at boot.
3837	 */
3838	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3839		if (props != NULL)
3840			spa_configfile_set(spa, props, B_FALSE);
3841
3842		spa_config_sync(spa, B_FALSE, B_TRUE);
3843
3844		mutex_exit(&spa_namespace_lock);
3845		return (0);
3846	}
3847
3848	spa_activate(spa, mode);
3849
3850	/*
3851	 * Don't start async tasks until we know everything is healthy.
3852	 */
3853	spa_async_suspend(spa);
3854
3855	zpool_get_rewind_policy(config, &policy);
3856	if (policy.zrp_request & ZPOOL_DO_REWIND)
3857		state = SPA_LOAD_RECOVER;
3858
3859	/*
3860	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3861	 * because the user-supplied config is actually the one to trust when
3862	 * doing an import.
3863	 */
3864	if (state != SPA_LOAD_RECOVER)
3865		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3866
3867	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3868	    policy.zrp_request);
3869
3870	/*
3871	 * Propagate anything learned while loading the pool and pass it
3872	 * back to caller (i.e. rewind info, missing devices, etc).
3873	 */
3874	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3875	    spa->spa_load_info) == 0);
3876
3877	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3878	/*
3879	 * Toss any existing sparelist, as it doesn't have any validity
3880	 * anymore, and conflicts with spa_has_spare().
3881	 */
3882	if (spa->spa_spares.sav_config) {
3883		nvlist_free(spa->spa_spares.sav_config);
3884		spa->spa_spares.sav_config = NULL;
3885		spa_load_spares(spa);
3886	}
3887	if (spa->spa_l2cache.sav_config) {
3888		nvlist_free(spa->spa_l2cache.sav_config);
3889		spa->spa_l2cache.sav_config = NULL;
3890		spa_load_l2cache(spa);
3891	}
3892
3893	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3894	    &nvroot) == 0);
3895	if (error == 0)
3896		error = spa_validate_aux(spa, nvroot, -1ULL,
3897		    VDEV_ALLOC_SPARE);
3898	if (error == 0)
3899		error = spa_validate_aux(spa, nvroot, -1ULL,
3900		    VDEV_ALLOC_L2CACHE);
3901	spa_config_exit(spa, SCL_ALL, FTAG);
3902
3903	if (props != NULL)
3904		spa_configfile_set(spa, props, B_FALSE);
3905
3906	if (error != 0 || (props && spa_writeable(spa) &&
3907	    (error = spa_prop_set(spa, props)))) {
3908		spa_unload(spa);
3909		spa_deactivate(spa);
3910		spa_remove(spa);
3911		mutex_exit(&spa_namespace_lock);
3912		return (error);
3913	}
3914
3915	spa_async_resume(spa);
3916
3917	/*
3918	 * Override any spares and level 2 cache devices as specified by
3919	 * the user, as these may have correct device names/devids, etc.
3920	 */
3921	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3922	    &spares, &nspares) == 0) {
3923		if (spa->spa_spares.sav_config)
3924			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3925			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3926		else
3927			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3928			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3929		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3930		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3931		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3932		spa_load_spares(spa);
3933		spa_config_exit(spa, SCL_ALL, FTAG);
3934		spa->spa_spares.sav_sync = B_TRUE;
3935	}
3936	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3937	    &l2cache, &nl2cache) == 0) {
3938		if (spa->spa_l2cache.sav_config)
3939			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3940			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3941		else
3942			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3943			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3944		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3945		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3946		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3947		spa_load_l2cache(spa);
3948		spa_config_exit(spa, SCL_ALL, FTAG);
3949		spa->spa_l2cache.sav_sync = B_TRUE;
3950	}
3951
3952	/*
3953	 * Check for any removed devices.
3954	 */
3955	if (spa->spa_autoreplace) {
3956		spa_aux_check_removed(&spa->spa_spares);
3957		spa_aux_check_removed(&spa->spa_l2cache);
3958	}
3959
3960	if (spa_writeable(spa)) {
3961		/*
3962		 * Update the config cache to include the newly-imported pool.
3963		 */
3964		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3965	}
3966
3967	/*
3968	 * It's possible that the pool was expanded while it was exported.
3969	 * We kick off an async task to handle this for us.
3970	 */
3971	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3972
3973	mutex_exit(&spa_namespace_lock);
3974	spa_history_log_version(spa, "import");
3975
3976	return (0);
3977}
3978
3979nvlist_t *
3980spa_tryimport(nvlist_t *tryconfig)
3981{
3982	nvlist_t *config = NULL;
3983	char *poolname;
3984	spa_t *spa;
3985	uint64_t state;
3986	int error;
3987
3988	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3989		return (NULL);
3990
3991	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3992		return (NULL);
3993
3994	/*
3995	 * Create and initialize the spa structure.
3996	 */
3997	mutex_enter(&spa_namespace_lock);
3998	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3999	spa_activate(spa, FREAD);
4000
4001	/*
4002	 * Pass off the heavy lifting to spa_load().
4003	 * Pass TRUE for mosconfig because the user-supplied config
4004	 * is actually the one to trust when doing an import.
4005	 */
4006	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4007
4008	/*
4009	 * If 'tryconfig' was at least parsable, return the current config.
4010	 */
4011	if (spa->spa_root_vdev != NULL) {
4012		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4013		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4014		    poolname) == 0);
4015		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4016		    state) == 0);
4017		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4018		    spa->spa_uberblock.ub_timestamp) == 0);
4019		VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4020		    spa->spa_load_info) == 0);
4021
4022		/*
4023		 * If the bootfs property exists on this pool then we
4024		 * copy it out so that external consumers can tell which
4025		 * pools are bootable.
4026		 */
4027		if ((!error || error == EEXIST) && spa->spa_bootfs) {
4028			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4029
4030			/*
4031			 * We have to play games with the name since the
4032			 * pool was opened as TRYIMPORT_NAME.
4033			 */
4034			if (dsl_dsobj_to_dsname(spa_name(spa),
4035			    spa->spa_bootfs, tmpname) == 0) {
4036				char *cp;
4037				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4038
4039				cp = strchr(tmpname, '/');
4040				if (cp == NULL) {
4041					(void) strlcpy(dsname, tmpname,
4042					    MAXPATHLEN);
4043				} else {
4044					(void) snprintf(dsname, MAXPATHLEN,
4045					    "%s/%s", poolname, ++cp);
4046				}
4047				VERIFY(nvlist_add_string(config,
4048				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4049				kmem_free(dsname, MAXPATHLEN);
4050			}
4051			kmem_free(tmpname, MAXPATHLEN);
4052		}
4053
4054		/*
4055		 * Add the list of hot spares and level 2 cache devices.
4056		 */
4057		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4058		spa_add_spares(spa, config);
4059		spa_add_l2cache(spa, config);
4060		spa_config_exit(spa, SCL_CONFIG, FTAG);
4061	}
4062
4063	spa_unload(spa);
4064	spa_deactivate(spa);
4065	spa_remove(spa);
4066	mutex_exit(&spa_namespace_lock);
4067
4068	return (config);
4069}
4070
4071/*
4072 * Pool export/destroy
4073 *
4074 * The act of destroying or exporting a pool is very simple.  We make sure there
4075 * is no more pending I/O and any references to the pool are gone.  Then, we
4076 * update the pool state and sync all the labels to disk, removing the
4077 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4078 * we don't sync the labels or remove the configuration cache.
4079 */
4080static int
4081spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4082    boolean_t force, boolean_t hardforce)
4083{
4084	spa_t *spa;
4085
4086	if (oldconfig)
4087		*oldconfig = NULL;
4088
4089	if (!(spa_mode_global & FWRITE))
4090		return (SET_ERROR(EROFS));
4091
4092	mutex_enter(&spa_namespace_lock);
4093	if ((spa = spa_lookup(pool)) == NULL) {
4094		mutex_exit(&spa_namespace_lock);
4095		return (SET_ERROR(ENOENT));
4096	}
4097
4098	/*
4099	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4100	 * reacquire the namespace lock, and see if we can export.
4101	 */
4102	spa_open_ref(spa, FTAG);
4103	mutex_exit(&spa_namespace_lock);
4104	spa_async_suspend(spa);
4105	mutex_enter(&spa_namespace_lock);
4106	spa_close(spa, FTAG);
4107
4108	/*
4109	 * The pool will be in core if it's openable,
4110	 * in which case we can modify its state.
4111	 */
4112	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4113		/*
4114		 * Objsets may be open only because they're dirty, so we
4115		 * have to force it to sync before checking spa_refcnt.
4116		 */
4117		txg_wait_synced(spa->spa_dsl_pool, 0);
4118
4119		/*
4120		 * A pool cannot be exported or destroyed if there are active
4121		 * references.  If we are resetting a pool, allow references by
4122		 * fault injection handlers.
4123		 */
4124		if (!spa_refcount_zero(spa) ||
4125		    (spa->spa_inject_ref != 0 &&
4126		    new_state != POOL_STATE_UNINITIALIZED)) {
4127			spa_async_resume(spa);
4128			mutex_exit(&spa_namespace_lock);
4129			return (SET_ERROR(EBUSY));
4130		}
4131
4132		/*
4133		 * A pool cannot be exported if it has an active shared spare.
4134		 * This is to prevent other pools stealing the active spare
4135		 * from an exported pool. At user's own will, such pool can
4136		 * be forcedly exported.
4137		 */
4138		if (!force && new_state == POOL_STATE_EXPORTED &&
4139		    spa_has_active_shared_spare(spa)) {
4140			spa_async_resume(spa);
4141			mutex_exit(&spa_namespace_lock);
4142			return (SET_ERROR(EXDEV));
4143		}
4144
4145		/*
4146		 * We want this to be reflected on every label,
4147		 * so mark them all dirty.  spa_unload() will do the
4148		 * final sync that pushes these changes out.
4149		 */
4150		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4151			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4152			spa->spa_state = new_state;
4153			spa->spa_final_txg = spa_last_synced_txg(spa) +
4154			    TXG_DEFER_SIZE + 1;
4155			vdev_config_dirty(spa->spa_root_vdev);
4156			spa_config_exit(spa, SCL_ALL, FTAG);
4157		}
4158	}
4159
4160	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4161
4162	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4163		spa_unload(spa);
4164		spa_deactivate(spa);
4165	}
4166
4167	if (oldconfig && spa->spa_config)
4168		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4169
4170	if (new_state != POOL_STATE_UNINITIALIZED) {
4171		if (!hardforce)
4172			spa_config_sync(spa, B_TRUE, B_TRUE);
4173		spa_remove(spa);
4174	}
4175	mutex_exit(&spa_namespace_lock);
4176
4177	return (0);
4178}
4179
4180/*
4181 * Destroy a storage pool.
4182 */
4183int
4184spa_destroy(char *pool)
4185{
4186	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4187	    B_FALSE, B_FALSE));
4188}
4189
4190/*
4191 * Export a storage pool.
4192 */
4193int
4194spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4195    boolean_t hardforce)
4196{
4197	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4198	    force, hardforce));
4199}
4200
4201/*
4202 * Similar to spa_export(), this unloads the spa_t without actually removing it
4203 * from the namespace in any way.
4204 */
4205int
4206spa_reset(char *pool)
4207{
4208	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4209	    B_FALSE, B_FALSE));
4210}
4211
4212/*
4213 * ==========================================================================
4214 * Device manipulation
4215 * ==========================================================================
4216 */
4217
4218/*
4219 * Add a device to a storage pool.
4220 */
4221int
4222spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4223{
4224	uint64_t txg, id;
4225	int error;
4226	vdev_t *rvd = spa->spa_root_vdev;
4227	vdev_t *vd, *tvd;
4228	nvlist_t **spares, **l2cache;
4229	uint_t nspares, nl2cache;
4230
4231	ASSERT(spa_writeable(spa));
4232
4233	txg = spa_vdev_enter(spa);
4234
4235	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4236	    VDEV_ALLOC_ADD)) != 0)
4237		return (spa_vdev_exit(spa, NULL, txg, error));
4238
4239	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
4240
4241	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4242	    &nspares) != 0)
4243		nspares = 0;
4244
4245	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4246	    &nl2cache) != 0)
4247		nl2cache = 0;
4248
4249	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4250		return (spa_vdev_exit(spa, vd, txg, EINVAL));
4251
4252	if (vd->vdev_children != 0 &&
4253	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
4254		return (spa_vdev_exit(spa, vd, txg, error));
4255
4256	/*
4257	 * We must validate the spares and l2cache devices after checking the
4258	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
4259	 */
4260	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4261		return (spa_vdev_exit(spa, vd, txg, error));
4262
4263	/*
4264	 * Transfer each new top-level vdev from vd to rvd.
4265	 */
4266	for (int c = 0; c < vd->vdev_children; c++) {
4267
4268		/*
4269		 * Set the vdev id to the first hole, if one exists.
4270		 */
4271		for (id = 0; id < rvd->vdev_children; id++) {
4272			if (rvd->vdev_child[id]->vdev_ishole) {
4273				vdev_free(rvd->vdev_child[id]);
4274				break;
4275			}
4276		}
4277		tvd = vd->vdev_child[c];
4278		vdev_remove_child(vd, tvd);
4279		tvd->vdev_id = id;
4280		vdev_add_child(rvd, tvd);
4281		vdev_config_dirty(tvd);
4282	}
4283
4284	if (nspares != 0) {
4285		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4286		    ZPOOL_CONFIG_SPARES);
4287		spa_load_spares(spa);
4288		spa->spa_spares.sav_sync = B_TRUE;
4289	}
4290
4291	if (nl2cache != 0) {
4292		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4293		    ZPOOL_CONFIG_L2CACHE);
4294		spa_load_l2cache(spa);
4295		spa->spa_l2cache.sav_sync = B_TRUE;
4296	}
4297
4298	/*
4299	 * We have to be careful when adding new vdevs to an existing pool.
4300	 * If other threads start allocating from these vdevs before we
4301	 * sync the config cache, and we lose power, then upon reboot we may
4302	 * fail to open the pool because there are DVAs that the config cache
4303	 * can't translate.  Therefore, we first add the vdevs without
4304	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4305	 * and then let spa_config_update() initialize the new metaslabs.
4306	 *
4307	 * spa_load() checks for added-but-not-initialized vdevs, so that
4308	 * if we lose power at any point in this sequence, the remaining
4309	 * steps will be completed the next time we load the pool.
4310	 */
4311	(void) spa_vdev_exit(spa, vd, txg, 0);
4312
4313	mutex_enter(&spa_namespace_lock);
4314	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4315	mutex_exit(&spa_namespace_lock);
4316
4317	return (0);
4318}
4319
4320/*
4321 * Attach a device to a mirror.  The arguments are the path to any device
4322 * in the mirror, and the nvroot for the new device.  If the path specifies
4323 * a device that is not mirrored, we automatically insert the mirror vdev.
4324 *
4325 * If 'replacing' is specified, the new device is intended to replace the
4326 * existing device; in this case the two devices are made into their own
4327 * mirror using the 'replacing' vdev, which is functionally identical to
4328 * the mirror vdev (it actually reuses all the same ops) but has a few
4329 * extra rules: you can't attach to it after it's been created, and upon
4330 * completion of resilvering, the first disk (the one being replaced)
4331 * is automatically detached.
4332 */
4333int
4334spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4335{
4336	uint64_t txg, dtl_max_txg;
4337	vdev_t *rvd = spa->spa_root_vdev;
4338	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4339	vdev_ops_t *pvops;
4340	char *oldvdpath, *newvdpath;
4341	int newvd_isspare;
4342	int error;
4343
4344	ASSERT(spa_writeable(spa));
4345
4346	txg = spa_vdev_enter(spa);
4347
4348	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4349
4350	if (oldvd == NULL)
4351		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4352
4353	if (!oldvd->vdev_ops->vdev_op_leaf)
4354		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4355
4356	pvd = oldvd->vdev_parent;
4357
4358	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4359	    VDEV_ALLOC_ATTACH)) != 0)
4360		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4361
4362	if (newrootvd->vdev_children != 1)
4363		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4364
4365	newvd = newrootvd->vdev_child[0];
4366
4367	if (!newvd->vdev_ops->vdev_op_leaf)
4368		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4369
4370	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4371		return (spa_vdev_exit(spa, newrootvd, txg, error));
4372
4373	/*
4374	 * Spares can't replace logs
4375	 */
4376	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4377		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4378
4379	if (!replacing) {
4380		/*
4381		 * For attach, the only allowable parent is a mirror or the root
4382		 * vdev.
4383		 */
4384		if (pvd->vdev_ops != &vdev_mirror_ops &&
4385		    pvd->vdev_ops != &vdev_root_ops)
4386			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4387
4388		pvops = &vdev_mirror_ops;
4389	} else {
4390		/*
4391		 * Active hot spares can only be replaced by inactive hot
4392		 * spares.
4393		 */
4394		if (pvd->vdev_ops == &vdev_spare_ops &&
4395		    oldvd->vdev_isspare &&
4396		    !spa_has_spare(spa, newvd->vdev_guid))
4397			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4398
4399		/*
4400		 * If the source is a hot spare, and the parent isn't already a
4401		 * spare, then we want to create a new hot spare.  Otherwise, we
4402		 * want to create a replacing vdev.  The user is not allowed to
4403		 * attach to a spared vdev child unless the 'isspare' state is
4404		 * the same (spare replaces spare, non-spare replaces
4405		 * non-spare).
4406		 */
4407		if (pvd->vdev_ops == &vdev_replacing_ops &&
4408		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4409			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4410		} else if (pvd->vdev_ops == &vdev_spare_ops &&
4411		    newvd->vdev_isspare != oldvd->vdev_isspare) {
4412			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4413		}
4414
4415		if (newvd->vdev_isspare)
4416			pvops = &vdev_spare_ops;
4417		else
4418			pvops = &vdev_replacing_ops;
4419	}
4420
4421	/*
4422	 * Make sure the new device is big enough.
4423	 */
4424	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4425		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4426
4427	/*
4428	 * The new device cannot have a higher alignment requirement
4429	 * than the top-level vdev.
4430	 */
4431	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4432		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4433
4434	/*
4435	 * If this is an in-place replacement, update oldvd's path and devid
4436	 * to make it distinguishable from newvd, and unopenable from now on.
4437	 */
4438	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4439		spa_strfree(oldvd->vdev_path);
4440		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4441		    KM_SLEEP);
4442		(void) sprintf(oldvd->vdev_path, "%s/%s",
4443		    newvd->vdev_path, "old");
4444		if (oldvd->vdev_devid != NULL) {
4445			spa_strfree(oldvd->vdev_devid);
4446			oldvd->vdev_devid = NULL;
4447		}
4448	}
4449
4450	/* mark the device being resilvered */
4451	newvd->vdev_resilver_txg = txg;
4452
4453	/*
4454	 * If the parent is not a mirror, or if we're replacing, insert the new
4455	 * mirror/replacing/spare vdev above oldvd.
4456	 */
4457	if (pvd->vdev_ops != pvops)
4458		pvd = vdev_add_parent(oldvd, pvops);
4459
4460	ASSERT(pvd->vdev_top->vdev_parent == rvd);
4461	ASSERT(pvd->vdev_ops == pvops);
4462	ASSERT(oldvd->vdev_parent == pvd);
4463
4464	/*
4465	 * Extract the new device from its root and add it to pvd.
4466	 */
4467	vdev_remove_child(newrootvd, newvd);
4468	newvd->vdev_id = pvd->vdev_children;
4469	newvd->vdev_crtxg = oldvd->vdev_crtxg;
4470	vdev_add_child(pvd, newvd);
4471
4472	tvd = newvd->vdev_top;
4473	ASSERT(pvd->vdev_top == tvd);
4474	ASSERT(tvd->vdev_parent == rvd);
4475
4476	vdev_config_dirty(tvd);
4477
4478	/*
4479	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4480	 * for any dmu_sync-ed blocks.  It will propagate upward when
4481	 * spa_vdev_exit() calls vdev_dtl_reassess().
4482	 */
4483	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4484
4485	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4486	    dtl_max_txg - TXG_INITIAL);
4487
4488	if (newvd->vdev_isspare) {
4489		spa_spare_activate(newvd);
4490		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4491	}
4492
4493	oldvdpath = spa_strdup(oldvd->vdev_path);
4494	newvdpath = spa_strdup(newvd->vdev_path);
4495	newvd_isspare = newvd->vdev_isspare;
4496
4497	/*
4498	 * Mark newvd's DTL dirty in this txg.
4499	 */
4500	vdev_dirty(tvd, VDD_DTL, newvd, txg);
4501
4502	/*
4503	 * Schedule the resilver to restart in the future. We do this to
4504	 * ensure that dmu_sync-ed blocks have been stitched into the
4505	 * respective datasets.
4506	 */
4507	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4508
4509	/*
4510	 * Commit the config
4511	 */
4512	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4513
4514	spa_history_log_internal(spa, "vdev attach", NULL,
4515	    "%s vdev=%s %s vdev=%s",
4516	    replacing && newvd_isspare ? "spare in" :
4517	    replacing ? "replace" : "attach", newvdpath,
4518	    replacing ? "for" : "to", oldvdpath);
4519
4520	spa_strfree(oldvdpath);
4521	spa_strfree(newvdpath);
4522
4523	if (spa->spa_bootfs)
4524		spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4525
4526	return (0);
4527}
4528
4529/*
4530 * Detach a device from a mirror or replacing vdev.
4531 *
4532 * If 'replace_done' is specified, only detach if the parent
4533 * is a replacing vdev.
4534 */
4535int
4536spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4537{
4538	uint64_t txg;
4539	int error;
4540	vdev_t *rvd = spa->spa_root_vdev;
4541	vdev_t *vd, *pvd, *cvd, *tvd;
4542	boolean_t unspare = B_FALSE;
4543	uint64_t unspare_guid = 0;
4544	char *vdpath;
4545
4546	ASSERT(spa_writeable(spa));
4547
4548	txg = spa_vdev_enter(spa);
4549
4550	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4551
4552	if (vd == NULL)
4553		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4554
4555	if (!vd->vdev_ops->vdev_op_leaf)
4556		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4557
4558	pvd = vd->vdev_parent;
4559
4560	/*
4561	 * If the parent/child relationship is not as expected, don't do it.
4562	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4563	 * vdev that's replacing B with C.  The user's intent in replacing
4564	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4565	 * the replace by detaching C, the expected behavior is to end up
4566	 * M(A,B).  But suppose that right after deciding to detach C,
4567	 * the replacement of B completes.  We would have M(A,C), and then
4568	 * ask to detach C, which would leave us with just A -- not what
4569	 * the user wanted.  To prevent this, we make sure that the
4570	 * parent/child relationship hasn't changed -- in this example,
4571	 * that C's parent is still the replacing vdev R.
4572	 */
4573	if (pvd->vdev_guid != pguid && pguid != 0)
4574		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4575
4576	/*
4577	 * Only 'replacing' or 'spare' vdevs can be replaced.
4578	 */
4579	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4580	    pvd->vdev_ops != &vdev_spare_ops)
4581		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4582
4583	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4584	    spa_version(spa) >= SPA_VERSION_SPARES);
4585
4586	/*
4587	 * Only mirror, replacing, and spare vdevs support detach.
4588	 */
4589	if (pvd->vdev_ops != &vdev_replacing_ops &&
4590	    pvd->vdev_ops != &vdev_mirror_ops &&
4591	    pvd->vdev_ops != &vdev_spare_ops)
4592		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4593
4594	/*
4595	 * If this device has the only valid copy of some data,
4596	 * we cannot safely detach it.
4597	 */
4598	if (vdev_dtl_required(vd))
4599		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4600
4601	ASSERT(pvd->vdev_children >= 2);
4602
4603	/*
4604	 * If we are detaching the second disk from a replacing vdev, then
4605	 * check to see if we changed the original vdev's path to have "/old"
4606	 * at the end in spa_vdev_attach().  If so, undo that change now.
4607	 */
4608	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4609	    vd->vdev_path != NULL) {
4610		size_t len = strlen(vd->vdev_path);
4611
4612		for (int c = 0; c < pvd->vdev_children; c++) {
4613			cvd = pvd->vdev_child[c];
4614
4615			if (cvd == vd || cvd->vdev_path == NULL)
4616				continue;
4617
4618			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4619			    strcmp(cvd->vdev_path + len, "/old") == 0) {
4620				spa_strfree(cvd->vdev_path);
4621				cvd->vdev_path = spa_strdup(vd->vdev_path);
4622				break;
4623			}
4624		}
4625	}
4626
4627	/*
4628	 * If we are detaching the original disk from a spare, then it implies
4629	 * that the spare should become a real disk, and be removed from the
4630	 * active spare list for the pool.
4631	 */
4632	if (pvd->vdev_ops == &vdev_spare_ops &&
4633	    vd->vdev_id == 0 &&
4634	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4635		unspare = B_TRUE;
4636
4637	/*
4638	 * Erase the disk labels so the disk can be used for other things.
4639	 * This must be done after all other error cases are handled,
4640	 * but before we disembowel vd (so we can still do I/O to it).
4641	 * But if we can't do it, don't treat the error as fatal --
4642	 * it may be that the unwritability of the disk is the reason
4643	 * it's being detached!
4644	 */
4645	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4646
4647	/*
4648	 * Remove vd from its parent and compact the parent's children.
4649	 */
4650	vdev_remove_child(pvd, vd);
4651	vdev_compact_children(pvd);
4652
4653	/*
4654	 * Remember one of the remaining children so we can get tvd below.
4655	 */
4656	cvd = pvd->vdev_child[pvd->vdev_children - 1];
4657
4658	/*
4659	 * If we need to remove the remaining child from the list of hot spares,
4660	 * do it now, marking the vdev as no longer a spare in the process.
4661	 * We must do this before vdev_remove_parent(), because that can
4662	 * change the GUID if it creates a new toplevel GUID.  For a similar
4663	 * reason, we must remove the spare now, in the same txg as the detach;
4664	 * otherwise someone could attach a new sibling, change the GUID, and
4665	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4666	 */
4667	if (unspare) {
4668		ASSERT(cvd->vdev_isspare);
4669		spa_spare_remove(cvd);
4670		unspare_guid = cvd->vdev_guid;
4671		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4672		cvd->vdev_unspare = B_TRUE;
4673	}
4674
4675	/*
4676	 * If the parent mirror/replacing vdev only has one child,
4677	 * the parent is no longer needed.  Remove it from the tree.
4678	 */
4679	if (pvd->vdev_children == 1) {
4680		if (pvd->vdev_ops == &vdev_spare_ops)
4681			cvd->vdev_unspare = B_FALSE;
4682		vdev_remove_parent(cvd);
4683	}
4684
4685
4686	/*
4687	 * We don't set tvd until now because the parent we just removed
4688	 * may have been the previous top-level vdev.
4689	 */
4690	tvd = cvd->vdev_top;
4691	ASSERT(tvd->vdev_parent == rvd);
4692
4693	/*
4694	 * Reevaluate the parent vdev state.
4695	 */
4696	vdev_propagate_state(cvd);
4697
4698	/*
4699	 * If the 'autoexpand' property is set on the pool then automatically
4700	 * try to expand the size of the pool. For example if the device we
4701	 * just detached was smaller than the others, it may be possible to
4702	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4703	 * first so that we can obtain the updated sizes of the leaf vdevs.
4704	 */
4705	if (spa->spa_autoexpand) {
4706		vdev_reopen(tvd);
4707		vdev_expand(tvd, txg);
4708	}
4709
4710	vdev_config_dirty(tvd);
4711
4712	/*
4713	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4714	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4715	 * But first make sure we're not on any *other* txg's DTL list, to
4716	 * prevent vd from being accessed after it's freed.
4717	 */
4718	vdpath = spa_strdup(vd->vdev_path);
4719	for (int t = 0; t < TXG_SIZE; t++)
4720		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4721	vd->vdev_detached = B_TRUE;
4722	vdev_dirty(tvd, VDD_DTL, vd, txg);
4723
4724	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4725
4726	/* hang on to the spa before we release the lock */
4727	spa_open_ref(spa, FTAG);
4728
4729	error = spa_vdev_exit(spa, vd, txg, 0);
4730
4731	spa_history_log_internal(spa, "detach", NULL,
4732	    "vdev=%s", vdpath);
4733	spa_strfree(vdpath);
4734
4735	/*
4736	 * If this was the removal of the original device in a hot spare vdev,
4737	 * then we want to go through and remove the device from the hot spare
4738	 * list of every other pool.
4739	 */
4740	if (unspare) {
4741		spa_t *altspa = NULL;
4742
4743		mutex_enter(&spa_namespace_lock);
4744		while ((altspa = spa_next(altspa)) != NULL) {
4745			if (altspa->spa_state != POOL_STATE_ACTIVE ||
4746			    altspa == spa)
4747				continue;
4748
4749			spa_open_ref(altspa, FTAG);
4750			mutex_exit(&spa_namespace_lock);
4751			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4752			mutex_enter(&spa_namespace_lock);
4753			spa_close(altspa, FTAG);
4754		}
4755		mutex_exit(&spa_namespace_lock);
4756
4757		/* search the rest of the vdevs for spares to remove */
4758		spa_vdev_resilver_done(spa);
4759	}
4760
4761	/* all done with the spa; OK to release */
4762	mutex_enter(&spa_namespace_lock);
4763	spa_close(spa, FTAG);
4764	mutex_exit(&spa_namespace_lock);
4765
4766	return (error);
4767}
4768
4769/*
4770 * Split a set of devices from their mirrors, and create a new pool from them.
4771 */
4772int
4773spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4774    nvlist_t *props, boolean_t exp)
4775{
4776	int error = 0;
4777	uint64_t txg, *glist;
4778	spa_t *newspa;
4779	uint_t c, children, lastlog;
4780	nvlist_t **child, *nvl, *tmp;
4781	dmu_tx_t *tx;
4782	char *altroot = NULL;
4783	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4784	boolean_t activate_slog;
4785
4786	ASSERT(spa_writeable(spa));
4787
4788	txg = spa_vdev_enter(spa);
4789
4790	/* clear the log and flush everything up to now */
4791	activate_slog = spa_passivate_log(spa);
4792	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4793	error = spa_offline_log(spa);
4794	txg = spa_vdev_config_enter(spa);
4795
4796	if (activate_slog)
4797		spa_activate_log(spa);
4798
4799	if (error != 0)
4800		return (spa_vdev_exit(spa, NULL, txg, error));
4801
4802	/* check new spa name before going any further */
4803	if (spa_lookup(newname) != NULL)
4804		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4805
4806	/*
4807	 * scan through all the children to ensure they're all mirrors
4808	 */
4809	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4810	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4811	    &children) != 0)
4812		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4813
4814	/* first, check to ensure we've got the right child count */
4815	rvd = spa->spa_root_vdev;
4816	lastlog = 0;
4817	for (c = 0; c < rvd->vdev_children; c++) {
4818		vdev_t *vd = rvd->vdev_child[c];
4819
4820		/* don't count the holes & logs as children */
4821		if (vd->vdev_islog || vd->vdev_ishole) {
4822			if (lastlog == 0)
4823				lastlog = c;
4824			continue;
4825		}
4826
4827		lastlog = 0;
4828	}
4829	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4830		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4831
4832	/* next, ensure no spare or cache devices are part of the split */
4833	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4834	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4835		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4836
4837	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4838	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4839
4840	/* then, loop over each vdev and validate it */
4841	for (c = 0; c < children; c++) {
4842		uint64_t is_hole = 0;
4843
4844		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4845		    &is_hole);
4846
4847		if (is_hole != 0) {
4848			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4849			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4850				continue;
4851			} else {
4852				error = SET_ERROR(EINVAL);
4853				break;
4854			}
4855		}
4856
4857		/* which disk is going to be split? */
4858		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4859		    &glist[c]) != 0) {
4860			error = SET_ERROR(EINVAL);
4861			break;
4862		}
4863
4864		/* look it up in the spa */
4865		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4866		if (vml[c] == NULL) {
4867			error = SET_ERROR(ENODEV);
4868			break;
4869		}
4870
4871		/* make sure there's nothing stopping the split */
4872		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4873		    vml[c]->vdev_islog ||
4874		    vml[c]->vdev_ishole ||
4875		    vml[c]->vdev_isspare ||
4876		    vml[c]->vdev_isl2cache ||
4877		    !vdev_writeable(vml[c]) ||
4878		    vml[c]->vdev_children != 0 ||
4879		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4880		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4881			error = SET_ERROR(EINVAL);
4882			break;
4883		}
4884
4885		if (vdev_dtl_required(vml[c])) {
4886			error = SET_ERROR(EBUSY);
4887			break;
4888		}
4889
4890		/* we need certain info from the top level */
4891		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4892		    vml[c]->vdev_top->vdev_ms_array) == 0);
4893		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4894		    vml[c]->vdev_top->vdev_ms_shift) == 0);
4895		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4896		    vml[c]->vdev_top->vdev_asize) == 0);
4897		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4898		    vml[c]->vdev_top->vdev_ashift) == 0);
4899	}
4900
4901	if (error != 0) {
4902		kmem_free(vml, children * sizeof (vdev_t *));
4903		kmem_free(glist, children * sizeof (uint64_t));
4904		return (spa_vdev_exit(spa, NULL, txg, error));
4905	}
4906
4907	/* stop writers from using the disks */
4908	for (c = 0; c < children; c++) {
4909		if (vml[c] != NULL)
4910			vml[c]->vdev_offline = B_TRUE;
4911	}
4912	vdev_reopen(spa->spa_root_vdev);
4913
4914	/*
4915	 * Temporarily record the splitting vdevs in the spa config.  This
4916	 * will disappear once the config is regenerated.
4917	 */
4918	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4919	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4920	    glist, children) == 0);
4921	kmem_free(glist, children * sizeof (uint64_t));
4922
4923	mutex_enter(&spa->spa_props_lock);
4924	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4925	    nvl) == 0);
4926	mutex_exit(&spa->spa_props_lock);
4927	spa->spa_config_splitting = nvl;
4928	vdev_config_dirty(spa->spa_root_vdev);
4929
4930	/* configure and create the new pool */
4931	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4932	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4933	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4934	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4935	    spa_version(spa)) == 0);
4936	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4937	    spa->spa_config_txg) == 0);
4938	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4939	    spa_generate_guid(NULL)) == 0);
4940	(void) nvlist_lookup_string(props,
4941	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4942
4943	/* add the new pool to the namespace */
4944	newspa = spa_add(newname, config, altroot);
4945	newspa->spa_config_txg = spa->spa_config_txg;
4946	spa_set_log_state(newspa, SPA_LOG_CLEAR);
4947
4948	/* release the spa config lock, retaining the namespace lock */
4949	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4950
4951	if (zio_injection_enabled)
4952		zio_handle_panic_injection(spa, FTAG, 1);
4953
4954	spa_activate(newspa, spa_mode_global);
4955	spa_async_suspend(newspa);
4956
4957	/* create the new pool from the disks of the original pool */
4958	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4959	if (error)
4960		goto out;
4961
4962	/* if that worked, generate a real config for the new pool */
4963	if (newspa->spa_root_vdev != NULL) {
4964		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4965		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4966		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4967		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4968		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4969		    B_TRUE));
4970	}
4971
4972	/* set the props */
4973	if (props != NULL) {
4974		spa_configfile_set(newspa, props, B_FALSE);
4975		error = spa_prop_set(newspa, props);
4976		if (error)
4977			goto out;
4978	}
4979
4980	/* flush everything */
4981	txg = spa_vdev_config_enter(newspa);
4982	vdev_config_dirty(newspa->spa_root_vdev);
4983	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4984
4985	if (zio_injection_enabled)
4986		zio_handle_panic_injection(spa, FTAG, 2);
4987
4988	spa_async_resume(newspa);
4989
4990	/* finally, update the original pool's config */
4991	txg = spa_vdev_config_enter(spa);
4992	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4993	error = dmu_tx_assign(tx, TXG_WAIT);
4994	if (error != 0)
4995		dmu_tx_abort(tx);
4996	for (c = 0; c < children; c++) {
4997		if (vml[c] != NULL) {
4998			vdev_split(vml[c]);
4999			if (error == 0)
5000				spa_history_log_internal(spa, "detach", tx,
5001				    "vdev=%s", vml[c]->vdev_path);
5002			vdev_free(vml[c]);
5003		}
5004	}
5005	vdev_config_dirty(spa->spa_root_vdev);
5006	spa->spa_config_splitting = NULL;
5007	nvlist_free(nvl);
5008	if (error == 0)
5009		dmu_tx_commit(tx);
5010	(void) spa_vdev_exit(spa, NULL, txg, 0);
5011
5012	if (zio_injection_enabled)
5013		zio_handle_panic_injection(spa, FTAG, 3);
5014
5015	/* split is complete; log a history record */
5016	spa_history_log_internal(newspa, "split", NULL,
5017	    "from pool %s", spa_name(spa));
5018
5019	kmem_free(vml, children * sizeof (vdev_t *));
5020
5021	/* if we're not going to mount the filesystems in userland, export */
5022	if (exp)
5023		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5024		    B_FALSE, B_FALSE);
5025
5026	return (error);
5027
5028out:
5029	spa_unload(newspa);
5030	spa_deactivate(newspa);
5031	spa_remove(newspa);
5032
5033	txg = spa_vdev_config_enter(spa);
5034
5035	/* re-online all offlined disks */
5036	for (c = 0; c < children; c++) {
5037		if (vml[c] != NULL)
5038			vml[c]->vdev_offline = B_FALSE;
5039	}
5040	vdev_reopen(spa->spa_root_vdev);
5041
5042	nvlist_free(spa->spa_config_splitting);
5043	spa->spa_config_splitting = NULL;
5044	(void) spa_vdev_exit(spa, NULL, txg, error);
5045
5046	kmem_free(vml, children * sizeof (vdev_t *));
5047	return (error);
5048}
5049
5050static nvlist_t *
5051spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5052{
5053	for (int i = 0; i < count; i++) {
5054		uint64_t guid;
5055
5056		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5057		    &guid) == 0);
5058
5059		if (guid == target_guid)
5060			return (nvpp[i]);
5061	}
5062
5063	return (NULL);
5064}
5065
5066static void
5067spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5068	nvlist_t *dev_to_remove)
5069{
5070	nvlist_t **newdev = NULL;
5071
5072	if (count > 1)
5073		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5074
5075	for (int i = 0, j = 0; i < count; i++) {
5076		if (dev[i] == dev_to_remove)
5077			continue;
5078		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5079	}
5080
5081	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5082	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5083
5084	for (int i = 0; i < count - 1; i++)
5085		nvlist_free(newdev[i]);
5086
5087	if (count > 1)
5088		kmem_free(newdev, (count - 1) * sizeof (void *));
5089}
5090
5091/*
5092 * Evacuate the device.
5093 */
5094static int
5095spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5096{
5097	uint64_t txg;
5098	int error = 0;
5099
5100	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5101	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5102	ASSERT(vd == vd->vdev_top);
5103
5104	/*
5105	 * Evacuate the device.  We don't hold the config lock as writer
5106	 * since we need to do I/O but we do keep the
5107	 * spa_namespace_lock held.  Once this completes the device
5108	 * should no longer have any blocks allocated on it.
5109	 */
5110	if (vd->vdev_islog) {
5111		if (vd->vdev_stat.vs_alloc != 0)
5112			error = spa_offline_log(spa);
5113	} else {
5114		error = SET_ERROR(ENOTSUP);
5115	}
5116
5117	if (error)
5118		return (error);
5119
5120	/*
5121	 * The evacuation succeeded.  Remove any remaining MOS metadata
5122	 * associated with this vdev, and wait for these changes to sync.
5123	 */
5124	ASSERT0(vd->vdev_stat.vs_alloc);
5125	txg = spa_vdev_config_enter(spa);
5126	vd->vdev_removing = B_TRUE;
5127	vdev_dirty_leaves(vd, VDD_DTL, txg);
5128	vdev_config_dirty(vd);
5129	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5130
5131	return (0);
5132}
5133
5134/*
5135 * Complete the removal by cleaning up the namespace.
5136 */
5137static void
5138spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5139{
5140	vdev_t *rvd = spa->spa_root_vdev;
5141	uint64_t id = vd->vdev_id;
5142	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5143
5144	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5145	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5146	ASSERT(vd == vd->vdev_top);
5147
5148	/*
5149	 * Only remove any devices which are empty.
5150	 */
5151	if (vd->vdev_stat.vs_alloc != 0)
5152		return;
5153
5154	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5155
5156	if (list_link_active(&vd->vdev_state_dirty_node))
5157		vdev_state_clean(vd);
5158	if (list_link_active(&vd->vdev_config_dirty_node))
5159		vdev_config_clean(vd);
5160
5161	vdev_free(vd);
5162
5163	if (last_vdev) {
5164		vdev_compact_children(rvd);
5165	} else {
5166		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5167		vdev_add_child(rvd, vd);
5168	}
5169	vdev_config_dirty(rvd);
5170
5171	/*
5172	 * Reassess the health of our root vdev.
5173	 */
5174	vdev_reopen(rvd);
5175}
5176
5177/*
5178 * Remove a device from the pool -
5179 *
5180 * Removing a device from the vdev namespace requires several steps
5181 * and can take a significant amount of time.  As a result we use
5182 * the spa_vdev_config_[enter/exit] functions which allow us to
5183 * grab and release the spa_config_lock while still holding the namespace
5184 * lock.  During each step the configuration is synced out.
5185 *
5186 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5187 * devices.
5188 */
5189int
5190spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5191{
5192	vdev_t *vd;
5193	metaslab_group_t *mg;
5194	nvlist_t **spares, **l2cache, *nv;
5195	uint64_t txg = 0;
5196	uint_t nspares, nl2cache;
5197	int error = 0;
5198	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5199
5200	ASSERT(spa_writeable(spa));
5201
5202	if (!locked)
5203		txg = spa_vdev_enter(spa);
5204
5205	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5206
5207	if (spa->spa_spares.sav_vdevs != NULL &&
5208	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5209	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5210	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5211		/*
5212		 * Only remove the hot spare if it's not currently in use
5213		 * in this pool.
5214		 */
5215		if (vd == NULL || unspare) {
5216			spa_vdev_remove_aux(spa->spa_spares.sav_config,
5217			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5218			spa_load_spares(spa);
5219			spa->spa_spares.sav_sync = B_TRUE;
5220		} else {
5221			error = SET_ERROR(EBUSY);
5222		}
5223	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
5224	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5225	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5226	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5227		/*
5228		 * Cache devices can always be removed.
5229		 */
5230		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5231		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5232		spa_load_l2cache(spa);
5233		spa->spa_l2cache.sav_sync = B_TRUE;
5234	} else if (vd != NULL && vd->vdev_islog) {
5235		ASSERT(!locked);
5236		ASSERT(vd == vd->vdev_top);
5237
5238		/*
5239		 * XXX - Once we have bp-rewrite this should
5240		 * become the common case.
5241		 */
5242
5243		mg = vd->vdev_mg;
5244
5245		/*
5246		 * Stop allocating from this vdev.
5247		 */
5248		metaslab_group_passivate(mg);
5249
5250		/*
5251		 * Wait for the youngest allocations and frees to sync,
5252		 * and then wait for the deferral of those frees to finish.
5253		 */
5254		spa_vdev_config_exit(spa, NULL,
5255		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5256
5257		/*
5258		 * Attempt to evacuate the vdev.
5259		 */
5260		error = spa_vdev_remove_evacuate(spa, vd);
5261
5262		txg = spa_vdev_config_enter(spa);
5263
5264		/*
5265		 * If we couldn't evacuate the vdev, unwind.
5266		 */
5267		if (error) {
5268			metaslab_group_activate(mg);
5269			return (spa_vdev_exit(spa, NULL, txg, error));
5270		}
5271
5272		/*
5273		 * Clean up the vdev namespace.
5274		 */
5275		spa_vdev_remove_from_namespace(spa, vd);
5276
5277	} else if (vd != NULL) {
5278		/*
5279		 * Normal vdevs cannot be removed (yet).
5280		 */
5281		error = SET_ERROR(ENOTSUP);
5282	} else {
5283		/*
5284		 * There is no vdev of any kind with the specified guid.
5285		 */
5286		error = SET_ERROR(ENOENT);
5287	}
5288
5289	if (!locked)
5290		return (spa_vdev_exit(spa, NULL, txg, error));
5291
5292	return (error);
5293}
5294
5295/*
5296 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5297 * currently spared, so we can detach it.
5298 */
5299static vdev_t *
5300spa_vdev_resilver_done_hunt(vdev_t *vd)
5301{
5302	vdev_t *newvd, *oldvd;
5303
5304	for (int c = 0; c < vd->vdev_children; c++) {
5305		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5306		if (oldvd != NULL)
5307			return (oldvd);
5308	}
5309
5310	/*
5311	 * Check for a completed replacement.  We always consider the first
5312	 * vdev in the list to be the oldest vdev, and the last one to be
5313	 * the newest (see spa_vdev_attach() for how that works).  In
5314	 * the case where the newest vdev is faulted, we will not automatically
5315	 * remove it after a resilver completes.  This is OK as it will require
5316	 * user intervention to determine which disk the admin wishes to keep.
5317	 */
5318	if (vd->vdev_ops == &vdev_replacing_ops) {
5319		ASSERT(vd->vdev_children > 1);
5320
5321		newvd = vd->vdev_child[vd->vdev_children - 1];
5322		oldvd = vd->vdev_child[0];
5323
5324		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5325		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5326		    !vdev_dtl_required(oldvd))
5327			return (oldvd);
5328	}
5329
5330	/*
5331	 * Check for a completed resilver with the 'unspare' flag set.
5332	 */
5333	if (vd->vdev_ops == &vdev_spare_ops) {
5334		vdev_t *first = vd->vdev_child[0];
5335		vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5336
5337		if (last->vdev_unspare) {
5338			oldvd = first;
5339			newvd = last;
5340		} else if (first->vdev_unspare) {
5341			oldvd = last;
5342			newvd = first;
5343		} else {
5344			oldvd = NULL;
5345		}
5346
5347		if (oldvd != NULL &&
5348		    vdev_dtl_empty(newvd, DTL_MISSING) &&
5349		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5350		    !vdev_dtl_required(oldvd))
5351			return (oldvd);
5352
5353		/*
5354		 * If there are more than two spares attached to a disk,
5355		 * and those spares are not required, then we want to
5356		 * attempt to free them up now so that they can be used
5357		 * by other pools.  Once we're back down to a single
5358		 * disk+spare, we stop removing them.
5359		 */
5360		if (vd->vdev_children > 2) {
5361			newvd = vd->vdev_child[1];
5362
5363			if (newvd->vdev_isspare && last->vdev_isspare &&
5364			    vdev_dtl_empty(last, DTL_MISSING) &&
5365			    vdev_dtl_empty(last, DTL_OUTAGE) &&
5366			    !vdev_dtl_required(newvd))
5367				return (newvd);
5368		}
5369	}
5370
5371	return (NULL);
5372}
5373
5374static void
5375spa_vdev_resilver_done(spa_t *spa)
5376{
5377	vdev_t *vd, *pvd, *ppvd;
5378	uint64_t guid, sguid, pguid, ppguid;
5379
5380	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5381
5382	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5383		pvd = vd->vdev_parent;
5384		ppvd = pvd->vdev_parent;
5385		guid = vd->vdev_guid;
5386		pguid = pvd->vdev_guid;
5387		ppguid = ppvd->vdev_guid;
5388		sguid = 0;
5389		/*
5390		 * If we have just finished replacing a hot spared device, then
5391		 * we need to detach the parent's first child (the original hot
5392		 * spare) as well.
5393		 */
5394		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5395		    ppvd->vdev_children == 2) {
5396			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5397			sguid = ppvd->vdev_child[1]->vdev_guid;
5398		}
5399		ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5400
5401		spa_config_exit(spa, SCL_ALL, FTAG);
5402		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5403			return;
5404		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5405			return;
5406		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5407	}
5408
5409	spa_config_exit(spa, SCL_ALL, FTAG);
5410}
5411
5412/*
5413 * Update the stored path or FRU for this vdev.
5414 */
5415int
5416spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5417    boolean_t ispath)
5418{
5419	vdev_t *vd;
5420	boolean_t sync = B_FALSE;
5421
5422	ASSERT(spa_writeable(spa));
5423
5424	spa_vdev_state_enter(spa, SCL_ALL);
5425
5426	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5427		return (spa_vdev_state_exit(spa, NULL, ENOENT));
5428
5429	if (!vd->vdev_ops->vdev_op_leaf)
5430		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5431
5432	if (ispath) {
5433		if (strcmp(value, vd->vdev_path) != 0) {
5434			spa_strfree(vd->vdev_path);
5435			vd->vdev_path = spa_strdup(value);
5436			sync = B_TRUE;
5437		}
5438	} else {
5439		if (vd->vdev_fru == NULL) {
5440			vd->vdev_fru = spa_strdup(value);
5441			sync = B_TRUE;
5442		} else if (strcmp(value, vd->vdev_fru) != 0) {
5443			spa_strfree(vd->vdev_fru);
5444			vd->vdev_fru = spa_strdup(value);
5445			sync = B_TRUE;
5446		}
5447	}
5448
5449	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5450}
5451
5452int
5453spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5454{
5455	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5456}
5457
5458int
5459spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5460{
5461	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5462}
5463
5464/*
5465 * ==========================================================================
5466 * SPA Scanning
5467 * ==========================================================================
5468 */
5469
5470int
5471spa_scan_stop(spa_t *spa)
5472{
5473	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5474	if (dsl_scan_resilvering(spa->spa_dsl_pool))
5475		return (SET_ERROR(EBUSY));
5476	return (dsl_scan_cancel(spa->spa_dsl_pool));
5477}
5478
5479int
5480spa_scan(spa_t *spa, pool_scan_func_t func)
5481{
5482	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5483
5484	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5485		return (SET_ERROR(ENOTSUP));
5486
5487	/*
5488	 * If a resilver was requested, but there is no DTL on a
5489	 * writeable leaf device, we have nothing to do.
5490	 */
5491	if (func == POOL_SCAN_RESILVER &&
5492	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5493		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5494		return (0);
5495	}
5496
5497	return (dsl_scan(spa->spa_dsl_pool, func));
5498}
5499
5500/*
5501 * ==========================================================================
5502 * SPA async task processing
5503 * ==========================================================================
5504 */
5505
5506static void
5507spa_async_remove(spa_t *spa, vdev_t *vd)
5508{
5509	if (vd->vdev_remove_wanted) {
5510		vd->vdev_remove_wanted = B_FALSE;
5511		vd->vdev_delayed_close = B_FALSE;
5512		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5513
5514		/*
5515		 * We want to clear the stats, but we don't want to do a full
5516		 * vdev_clear() as that will cause us to throw away
5517		 * degraded/faulted state as well as attempt to reopen the
5518		 * device, all of which is a waste.
5519		 */
5520		vd->vdev_stat.vs_read_errors = 0;
5521		vd->vdev_stat.vs_write_errors = 0;
5522		vd->vdev_stat.vs_checksum_errors = 0;
5523
5524		vdev_state_dirty(vd->vdev_top);
5525	}
5526
5527	for (int c = 0; c < vd->vdev_children; c++)
5528		spa_async_remove(spa, vd->vdev_child[c]);
5529}
5530
5531static void
5532spa_async_probe(spa_t *spa, vdev_t *vd)
5533{
5534	if (vd->vdev_probe_wanted) {
5535		vd->vdev_probe_wanted = B_FALSE;
5536		vdev_reopen(vd);	/* vdev_open() does the actual probe */
5537	}
5538
5539	for (int c = 0; c < vd->vdev_children; c++)
5540		spa_async_probe(spa, vd->vdev_child[c]);
5541}
5542
5543static void
5544spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5545{
5546	sysevent_id_t eid;
5547	nvlist_t *attr;
5548	char *physpath;
5549
5550	if (!spa->spa_autoexpand)
5551		return;
5552
5553	for (int c = 0; c < vd->vdev_children; c++) {
5554		vdev_t *cvd = vd->vdev_child[c];
5555		spa_async_autoexpand(spa, cvd);
5556	}
5557
5558	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5559		return;
5560
5561	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5562	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5563
5564	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5565	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5566
5567	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5568	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5569
5570	nvlist_free(attr);
5571	kmem_free(physpath, MAXPATHLEN);
5572}
5573
5574static void
5575spa_async_thread(spa_t *spa)
5576{
5577	int tasks;
5578
5579	ASSERT(spa->spa_sync_on);
5580
5581	mutex_enter(&spa->spa_async_lock);
5582	tasks = spa->spa_async_tasks;
5583	spa->spa_async_tasks = 0;
5584	mutex_exit(&spa->spa_async_lock);
5585
5586	/*
5587	 * See if the config needs to be updated.
5588	 */
5589	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5590		uint64_t old_space, new_space;
5591
5592		mutex_enter(&spa_namespace_lock);
5593		old_space = metaslab_class_get_space(spa_normal_class(spa));
5594		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5595		new_space = metaslab_class_get_space(spa_normal_class(spa));
5596		mutex_exit(&spa_namespace_lock);
5597
5598		/*
5599		 * If the pool grew as a result of the config update,
5600		 * then log an internal history event.
5601		 */
5602		if (new_space != old_space) {
5603			spa_history_log_internal(spa, "vdev online", NULL,
5604			    "pool '%s' size: %llu(+%llu)",
5605			    spa_name(spa), new_space, new_space - old_space);
5606		}
5607	}
5608
5609	/*
5610	 * See if any devices need to be marked REMOVED.
5611	 */
5612	if (tasks & SPA_ASYNC_REMOVE) {
5613		spa_vdev_state_enter(spa, SCL_NONE);
5614		spa_async_remove(spa, spa->spa_root_vdev);
5615		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5616			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5617		for (int i = 0; i < spa->spa_spares.sav_count; i++)
5618			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5619		(void) spa_vdev_state_exit(spa, NULL, 0);
5620	}
5621
5622	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5623		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5624		spa_async_autoexpand(spa, spa->spa_root_vdev);
5625		spa_config_exit(spa, SCL_CONFIG, FTAG);
5626	}
5627
5628	/*
5629	 * See if any devices need to be probed.
5630	 */
5631	if (tasks & SPA_ASYNC_PROBE) {
5632		spa_vdev_state_enter(spa, SCL_NONE);
5633		spa_async_probe(spa, spa->spa_root_vdev);
5634		(void) spa_vdev_state_exit(spa, NULL, 0);
5635	}
5636
5637	/*
5638	 * If any devices are done replacing, detach them.
5639	 */
5640	if (tasks & SPA_ASYNC_RESILVER_DONE)
5641		spa_vdev_resilver_done(spa);
5642
5643	/*
5644	 * Kick off a resilver.
5645	 */
5646	if (tasks & SPA_ASYNC_RESILVER)
5647		dsl_resilver_restart(spa->spa_dsl_pool, 0);
5648
5649	/*
5650	 * Let the world know that we're done.
5651	 */
5652	mutex_enter(&spa->spa_async_lock);
5653	spa->spa_async_thread = NULL;
5654	cv_broadcast(&spa->spa_async_cv);
5655	mutex_exit(&spa->spa_async_lock);
5656	thread_exit();
5657}
5658
5659void
5660spa_async_suspend(spa_t *spa)
5661{
5662	mutex_enter(&spa->spa_async_lock);
5663	spa->spa_async_suspended++;
5664	while (spa->spa_async_thread != NULL)
5665		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5666	mutex_exit(&spa->spa_async_lock);
5667}
5668
5669void
5670spa_async_resume(spa_t *spa)
5671{
5672	mutex_enter(&spa->spa_async_lock);
5673	ASSERT(spa->spa_async_suspended != 0);
5674	spa->spa_async_suspended--;
5675	mutex_exit(&spa->spa_async_lock);
5676}
5677
5678static boolean_t
5679spa_async_tasks_pending(spa_t *spa)
5680{
5681	uint_t non_config_tasks;
5682	uint_t config_task;
5683	boolean_t config_task_suspended;
5684
5685	non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5686	config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5687	if (spa->spa_ccw_fail_time == 0) {
5688		config_task_suspended = B_FALSE;
5689	} else {
5690		config_task_suspended =
5691		    (gethrtime() - spa->spa_ccw_fail_time) <
5692		    (zfs_ccw_retry_interval * NANOSEC);
5693	}
5694
5695	return (non_config_tasks || (config_task && !config_task_suspended));
5696}
5697
5698static void
5699spa_async_dispatch(spa_t *spa)
5700{
5701	mutex_enter(&spa->spa_async_lock);
5702	if (spa_async_tasks_pending(spa) &&
5703	    !spa->spa_async_suspended &&
5704	    spa->spa_async_thread == NULL &&
5705	    rootdir != NULL)
5706		spa->spa_async_thread = thread_create(NULL, 0,
5707		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5708	mutex_exit(&spa->spa_async_lock);
5709}
5710
5711void
5712spa_async_request(spa_t *spa, int task)
5713{
5714	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5715	mutex_enter(&spa->spa_async_lock);
5716	spa->spa_async_tasks |= task;
5717	mutex_exit(&spa->spa_async_lock);
5718}
5719
5720/*
5721 * ==========================================================================
5722 * SPA syncing routines
5723 * ==========================================================================
5724 */
5725
5726static int
5727bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5728{
5729	bpobj_t *bpo = arg;
5730	bpobj_enqueue(bpo, bp, tx);
5731	return (0);
5732}
5733
5734static int
5735spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5736{
5737	zio_t *zio = arg;
5738
5739	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5740	    zio->io_flags));
5741	return (0);
5742}
5743
5744/*
5745 * Note: this simple function is not inlined to make it easier to dtrace the
5746 * amount of time spent syncing frees.
5747 */
5748static void
5749spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
5750{
5751	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5752	bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
5753	VERIFY(zio_wait(zio) == 0);
5754}
5755
5756/*
5757 * Note: this simple function is not inlined to make it easier to dtrace the
5758 * amount of time spent syncing deferred frees.
5759 */
5760static void
5761spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
5762{
5763	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5764	VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
5765	    spa_free_sync_cb, zio, tx), ==, 0);
5766	VERIFY0(zio_wait(zio));
5767}
5768
5769
5770static void
5771spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5772{
5773	char *packed = NULL;
5774	size_t bufsize;
5775	size_t nvsize = 0;
5776	dmu_buf_t *db;
5777
5778	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5779
5780	/*
5781	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5782	 * information.  This avoids the dbuf_will_dirty() path and
5783	 * saves us a pre-read to get data we don't actually care about.
5784	 */
5785	bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5786	packed = kmem_alloc(bufsize, KM_SLEEP);
5787
5788	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5789	    KM_SLEEP) == 0);
5790	bzero(packed + nvsize, bufsize - nvsize);
5791
5792	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5793
5794	kmem_free(packed, bufsize);
5795
5796	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5797	dmu_buf_will_dirty(db, tx);
5798	*(uint64_t *)db->db_data = nvsize;
5799	dmu_buf_rele(db, FTAG);
5800}
5801
5802static void
5803spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5804    const char *config, const char *entry)
5805{
5806	nvlist_t *nvroot;
5807	nvlist_t **list;
5808	int i;
5809
5810	if (!sav->sav_sync)
5811		return;
5812
5813	/*
5814	 * Update the MOS nvlist describing the list of available devices.
5815	 * spa_validate_aux() will have already made sure this nvlist is
5816	 * valid and the vdevs are labeled appropriately.
5817	 */
5818	if (sav->sav_object == 0) {
5819		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5820		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5821		    sizeof (uint64_t), tx);
5822		VERIFY(zap_update(spa->spa_meta_objset,
5823		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5824		    &sav->sav_object, tx) == 0);
5825	}
5826
5827	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5828	if (sav->sav_count == 0) {
5829		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5830	} else {
5831		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5832		for (i = 0; i < sav->sav_count; i++)
5833			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5834			    B_FALSE, VDEV_CONFIG_L2CACHE);
5835		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5836		    sav->sav_count) == 0);
5837		for (i = 0; i < sav->sav_count; i++)
5838			nvlist_free(list[i]);
5839		kmem_free(list, sav->sav_count * sizeof (void *));
5840	}
5841
5842	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5843	nvlist_free(nvroot);
5844
5845	sav->sav_sync = B_FALSE;
5846}
5847
5848static void
5849spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5850{
5851	nvlist_t *config;
5852
5853	if (list_is_empty(&spa->spa_config_dirty_list))
5854		return;
5855
5856	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5857
5858	config = spa_config_generate(spa, spa->spa_root_vdev,
5859	    dmu_tx_get_txg(tx), B_FALSE);
5860
5861	/*
5862	 * If we're upgrading the spa version then make sure that
5863	 * the config object gets updated with the correct version.
5864	 */
5865	if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5866		fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5867		    spa->spa_uberblock.ub_version);
5868
5869	spa_config_exit(spa, SCL_STATE, FTAG);
5870
5871	if (spa->spa_config_syncing)
5872		nvlist_free(spa->spa_config_syncing);
5873	spa->spa_config_syncing = config;
5874
5875	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5876}
5877
5878static void
5879spa_sync_version(void *arg, dmu_tx_t *tx)
5880{
5881	uint64_t *versionp = arg;
5882	uint64_t version = *versionp;
5883	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5884
5885	/*
5886	 * Setting the version is special cased when first creating the pool.
5887	 */
5888	ASSERT(tx->tx_txg != TXG_INITIAL);
5889
5890	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5891	ASSERT(version >= spa_version(spa));
5892
5893	spa->spa_uberblock.ub_version = version;
5894	vdev_config_dirty(spa->spa_root_vdev);
5895	spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5896}
5897
5898/*
5899 * Set zpool properties.
5900 */
5901static void
5902spa_sync_props(void *arg, dmu_tx_t *tx)
5903{
5904	nvlist_t *nvp = arg;
5905	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5906	objset_t *mos = spa->spa_meta_objset;
5907	nvpair_t *elem = NULL;
5908
5909	mutex_enter(&spa->spa_props_lock);
5910
5911	while ((elem = nvlist_next_nvpair(nvp, elem))) {
5912		uint64_t intval;
5913		char *strval, *fname;
5914		zpool_prop_t prop;
5915		const char *propname;
5916		zprop_type_t proptype;
5917		zfeature_info_t *feature;
5918
5919		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5920		case ZPROP_INVAL:
5921			/*
5922			 * We checked this earlier in spa_prop_validate().
5923			 */
5924			ASSERT(zpool_prop_feature(nvpair_name(elem)));
5925
5926			fname = strchr(nvpair_name(elem), '@') + 1;
5927			VERIFY0(zfeature_lookup_name(fname, &feature));
5928
5929			spa_feature_enable(spa, feature, tx);
5930			spa_history_log_internal(spa, "set", tx,
5931			    "%s=enabled", nvpair_name(elem));
5932			break;
5933
5934		case ZPOOL_PROP_VERSION:
5935			intval = fnvpair_value_uint64(elem);
5936			/*
5937			 * The version is synced seperatly before other
5938			 * properties and should be correct by now.
5939			 */
5940			ASSERT3U(spa_version(spa), >=, intval);
5941			break;
5942
5943		case ZPOOL_PROP_ALTROOT:
5944			/*
5945			 * 'altroot' is a non-persistent property. It should
5946			 * have been set temporarily at creation or import time.
5947			 */
5948			ASSERT(spa->spa_root != NULL);
5949			break;
5950
5951		case ZPOOL_PROP_READONLY:
5952		case ZPOOL_PROP_CACHEFILE:
5953			/*
5954			 * 'readonly' and 'cachefile' are also non-persisitent
5955			 * properties.
5956			 */
5957			break;
5958		case ZPOOL_PROP_COMMENT:
5959			strval = fnvpair_value_string(elem);
5960			if (spa->spa_comment != NULL)
5961				spa_strfree(spa->spa_comment);
5962			spa->spa_comment = spa_strdup(strval);
5963			/*
5964			 * We need to dirty the configuration on all the vdevs
5965			 * so that their labels get updated.  It's unnecessary
5966			 * to do this for pool creation since the vdev's
5967			 * configuratoin has already been dirtied.
5968			 */
5969			if (tx->tx_txg != TXG_INITIAL)
5970				vdev_config_dirty(spa->spa_root_vdev);
5971			spa_history_log_internal(spa, "set", tx,
5972			    "%s=%s", nvpair_name(elem), strval);
5973			break;
5974		default:
5975			/*
5976			 * Set pool property values in the poolprops mos object.
5977			 */
5978			if (spa->spa_pool_props_object == 0) {
5979				spa->spa_pool_props_object =
5980				    zap_create_link(mos, DMU_OT_POOL_PROPS,
5981				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5982				    tx);
5983			}
5984
5985			/* normalize the property name */
5986			propname = zpool_prop_to_name(prop);
5987			proptype = zpool_prop_get_type(prop);
5988
5989			if (nvpair_type(elem) == DATA_TYPE_STRING) {
5990				ASSERT(proptype == PROP_TYPE_STRING);
5991				strval = fnvpair_value_string(elem);
5992				VERIFY0(zap_update(mos,
5993				    spa->spa_pool_props_object, propname,
5994				    1, strlen(strval) + 1, strval, tx));
5995				spa_history_log_internal(spa, "set", tx,
5996				    "%s=%s", nvpair_name(elem), strval);
5997			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5998				intval = fnvpair_value_uint64(elem);
5999
6000				if (proptype == PROP_TYPE_INDEX) {
6001					const char *unused;
6002					VERIFY0(zpool_prop_index_to_string(
6003					    prop, intval, &unused));
6004				}
6005				VERIFY0(zap_update(mos,
6006				    spa->spa_pool_props_object, propname,
6007				    8, 1, &intval, tx));
6008				spa_history_log_internal(spa, "set", tx,
6009				    "%s=%lld", nvpair_name(elem), intval);
6010			} else {
6011				ASSERT(0); /* not allowed */
6012			}
6013
6014			switch (prop) {
6015			case ZPOOL_PROP_DELEGATION:
6016				spa->spa_delegation = intval;
6017				break;
6018			case ZPOOL_PROP_BOOTFS:
6019				spa->spa_bootfs = intval;
6020				break;
6021			case ZPOOL_PROP_FAILUREMODE:
6022				spa->spa_failmode = intval;
6023				break;
6024			case ZPOOL_PROP_AUTOEXPAND:
6025				spa->spa_autoexpand = intval;
6026				if (tx->tx_txg != TXG_INITIAL)
6027					spa_async_request(spa,
6028					    SPA_ASYNC_AUTOEXPAND);
6029				break;
6030			case ZPOOL_PROP_DEDUPDITTO:
6031				spa->spa_dedup_ditto = intval;
6032				break;
6033			default:
6034				break;
6035			}
6036		}
6037
6038	}
6039
6040	mutex_exit(&spa->spa_props_lock);
6041}
6042
6043/*
6044 * Perform one-time upgrade on-disk changes.  spa_version() does not
6045 * reflect the new version this txg, so there must be no changes this
6046 * txg to anything that the upgrade code depends on after it executes.
6047 * Therefore this must be called after dsl_pool_sync() does the sync
6048 * tasks.
6049 */
6050static void
6051spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6052{
6053	dsl_pool_t *dp = spa->spa_dsl_pool;
6054
6055	ASSERT(spa->spa_sync_pass == 1);
6056
6057	rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6058
6059	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6060	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {