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) 2011, 2014 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_P(12, 8),	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_IS_HOLE(bp)) {
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 (!spa_features_check(spa, B_FALSE,
2304		    unsup_feat, enabled_feat))
2305			missing_feat_read = B_TRUE;
2306
2307		if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2308			if (!spa_features_check(spa, B_TRUE,
2309			    unsup_feat, enabled_feat)) {
2310				missing_feat_write = B_TRUE;
2311			}
2312		}
2313
2314		fnvlist_add_nvlist(spa->spa_load_info,
2315		    ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2316
2317		if (!nvlist_empty(unsup_feat)) {
2318			fnvlist_add_nvlist(spa->spa_load_info,
2319			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2320		}
2321
2322		fnvlist_free(enabled_feat);
2323		fnvlist_free(unsup_feat);
2324
2325		if (!missing_feat_read) {
2326			fnvlist_add_boolean(spa->spa_load_info,
2327			    ZPOOL_CONFIG_CAN_RDONLY);
2328		}
2329
2330		/*
2331		 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2332		 * twofold: to determine whether the pool is available for
2333		 * import in read-write mode and (if it is not) whether the
2334		 * pool is available for import in read-only mode. If the pool
2335		 * is available for import in read-write mode, it is displayed
2336		 * as available in userland; if it is not available for import
2337		 * in read-only mode, it is displayed as unavailable in
2338		 * userland. If the pool is available for import in read-only
2339		 * mode but not read-write mode, it is displayed as unavailable
2340		 * in userland with a special note that the pool is actually
2341		 * available for open in read-only mode.
2342		 *
2343		 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2344		 * missing a feature for write, we must first determine whether
2345		 * the pool can be opened read-only before returning to
2346		 * userland in order to know whether to display the
2347		 * abovementioned note.
2348		 */
2349		if (missing_feat_read || (missing_feat_write &&
2350		    spa_writeable(spa))) {
2351			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2352			    ENOTSUP));
2353		}
2354
2355		/*
2356		 * Load refcounts for ZFS features from disk into an in-memory
2357		 * cache during SPA initialization.
2358		 */
2359		for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2360			uint64_t refcount;
2361
2362			error = feature_get_refcount_from_disk(spa,
2363			    &spa_feature_table[i], &refcount);
2364			if (error == 0) {
2365				spa->spa_feat_refcount_cache[i] = refcount;
2366			} else if (error == ENOTSUP) {
2367				spa->spa_feat_refcount_cache[i] =
2368				    SPA_FEATURE_DISABLED;
2369			} else {
2370				return (spa_vdev_err(rvd,
2371				    VDEV_AUX_CORRUPT_DATA, EIO));
2372			}
2373		}
2374	}
2375
2376	if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2377		if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2378		    &spa->spa_feat_enabled_txg_obj) != 0) {
2379			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2380		}
2381	}
2382
2383	spa->spa_is_initializing = B_TRUE;
2384	error = dsl_pool_open(spa->spa_dsl_pool);
2385	spa->spa_is_initializing = B_FALSE;
2386	if (error != 0)
2387		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2388
2389	if (!mosconfig) {
2390		uint64_t hostid;
2391		nvlist_t *policy = NULL, *nvconfig;
2392
2393		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2394			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2395
2396		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2397		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2398			char *hostname;
2399			unsigned long myhostid = 0;
2400
2401			VERIFY(nvlist_lookup_string(nvconfig,
2402			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2403
2404#ifdef	_KERNEL
2405			myhostid = zone_get_hostid(NULL);
2406#else	/* _KERNEL */
2407			/*
2408			 * We're emulating the system's hostid in userland, so
2409			 * we can't use zone_get_hostid().
2410			 */
2411			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2412#endif	/* _KERNEL */
2413			if (hostid != 0 && myhostid != 0 &&
2414			    hostid != myhostid) {
2415				nvlist_free(nvconfig);
2416				cmn_err(CE_WARN, "pool '%s' could not be "
2417				    "loaded as it was last accessed by "
2418				    "another system (host: %s hostid: 0x%lx). "
2419				    "See: http://illumos.org/msg/ZFS-8000-EY",
2420				    spa_name(spa), hostname,
2421				    (unsigned long)hostid);
2422				return (SET_ERROR(EBADF));
2423			}
2424		}
2425		if (nvlist_lookup_nvlist(spa->spa_config,
2426		    ZPOOL_REWIND_POLICY, &policy) == 0)
2427			VERIFY(nvlist_add_nvlist(nvconfig,
2428			    ZPOOL_REWIND_POLICY, policy) == 0);
2429
2430		spa_config_set(spa, nvconfig);
2431		spa_unload(spa);
2432		spa_deactivate(spa);
2433		spa_activate(spa, orig_mode);
2434
2435		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2436	}
2437
2438	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2439		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2440	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2441	if (error != 0)
2442		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2443
2444	/*
2445	 * Load the bit that tells us to use the new accounting function
2446	 * (raid-z deflation).  If we have an older pool, this will not
2447	 * be present.
2448	 */
2449	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2450	if (error != 0 && error != ENOENT)
2451		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2452
2453	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2454	    &spa->spa_creation_version);
2455	if (error != 0 && error != ENOENT)
2456		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2457
2458	/*
2459	 * Load the persistent error log.  If we have an older pool, this will
2460	 * not be present.
2461	 */
2462	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2463	if (error != 0 && error != ENOENT)
2464		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2465
2466	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2467	    &spa->spa_errlog_scrub);
2468	if (error != 0 && error != ENOENT)
2469		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2470
2471	/*
2472	 * Load the history object.  If we have an older pool, this
2473	 * will not be present.
2474	 */
2475	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2476	if (error != 0 && error != ENOENT)
2477		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2478
2479	/*
2480	 * If we're assembling the pool from the split-off vdevs of
2481	 * an existing pool, we don't want to attach the spares & cache
2482	 * devices.
2483	 */
2484
2485	/*
2486	 * Load any hot spares for this pool.
2487	 */
2488	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2489	if (error != 0 && error != ENOENT)
2490		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2491	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2492		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2493		if (load_nvlist(spa, spa->spa_spares.sav_object,
2494		    &spa->spa_spares.sav_config) != 0)
2495			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2496
2497		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2498		spa_load_spares(spa);
2499		spa_config_exit(spa, SCL_ALL, FTAG);
2500	} else if (error == 0) {
2501		spa->spa_spares.sav_sync = B_TRUE;
2502	}
2503
2504	/*
2505	 * Load any level 2 ARC devices for this pool.
2506	 */
2507	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2508	    &spa->spa_l2cache.sav_object);
2509	if (error != 0 && error != ENOENT)
2510		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2511	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2512		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2513		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2514		    &spa->spa_l2cache.sav_config) != 0)
2515			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2516
2517		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2518		spa_load_l2cache(spa);
2519		spa_config_exit(spa, SCL_ALL, FTAG);
2520	} else if (error == 0) {
2521		spa->spa_l2cache.sav_sync = B_TRUE;
2522	}
2523
2524	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2525
2526	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2527	if (error && error != ENOENT)
2528		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2529
2530	if (error == 0) {
2531		uint64_t autoreplace;
2532
2533		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2534		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2535		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2536		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2537		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2538		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2539		    &spa->spa_dedup_ditto);
2540
2541		spa->spa_autoreplace = (autoreplace != 0);
2542	}
2543
2544	/*
2545	 * If the 'autoreplace' property is set, then post a resource notifying
2546	 * the ZFS DE that it should not issue any faults for unopenable
2547	 * devices.  We also iterate over the vdevs, and post a sysevent for any
2548	 * unopenable vdevs so that the normal autoreplace handler can take
2549	 * over.
2550	 */
2551	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2552		spa_check_removed(spa->spa_root_vdev);
2553		/*
2554		 * For the import case, this is done in spa_import(), because
2555		 * at this point we're using the spare definitions from
2556		 * the MOS config, not necessarily from the userland config.
2557		 */
2558		if (state != SPA_LOAD_IMPORT) {
2559			spa_aux_check_removed(&spa->spa_spares);
2560			spa_aux_check_removed(&spa->spa_l2cache);
2561		}
2562	}
2563
2564	/*
2565	 * Load the vdev state for all toplevel vdevs.
2566	 */
2567	vdev_load(rvd);
2568
2569	/*
2570	 * Propagate the leaf DTLs we just loaded all the way up the tree.
2571	 */
2572	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2573	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2574	spa_config_exit(spa, SCL_ALL, FTAG);
2575
2576	/*
2577	 * Load the DDTs (dedup tables).
2578	 */
2579	error = ddt_load(spa);
2580	if (error != 0)
2581		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2582
2583	spa_update_dspace(spa);
2584
2585	/*
2586	 * Validate the config, using the MOS config to fill in any
2587	 * information which might be missing.  If we fail to validate
2588	 * the config then declare the pool unfit for use. If we're
2589	 * assembling a pool from a split, the log is not transferred
2590	 * over.
2591	 */
2592	if (type != SPA_IMPORT_ASSEMBLE) {
2593		nvlist_t *nvconfig;
2594
2595		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2596			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2597
2598		if (!spa_config_valid(spa, nvconfig)) {
2599			nvlist_free(nvconfig);
2600			return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2601			    ENXIO));
2602		}
2603		nvlist_free(nvconfig);
2604
2605		/*
2606		 * Now that we've validated the config, check the state of the
2607		 * root vdev.  If it can't be opened, it indicates one or
2608		 * more toplevel vdevs are faulted.
2609		 */
2610		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2611			return (SET_ERROR(ENXIO));
2612
2613		if (spa_check_logs(spa)) {
2614			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2615			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2616		}
2617	}
2618
2619	if (missing_feat_write) {
2620		ASSERT(state == SPA_LOAD_TRYIMPORT);
2621
2622		/*
2623		 * At this point, we know that we can open the pool in
2624		 * read-only mode but not read-write mode. We now have enough
2625		 * information and can return to userland.
2626		 */
2627		return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2628	}
2629
2630	/*
2631	 * We've successfully opened the pool, verify that we're ready
2632	 * to start pushing transactions.
2633	 */
2634	if (state != SPA_LOAD_TRYIMPORT) {
2635		if (error = spa_load_verify(spa))
2636			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2637			    error));
2638	}
2639
2640	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2641	    spa->spa_load_max_txg == UINT64_MAX)) {
2642		dmu_tx_t *tx;
2643		int need_update = B_FALSE;
2644
2645		ASSERT(state != SPA_LOAD_TRYIMPORT);
2646
2647		/*
2648		 * Claim log blocks that haven't been committed yet.
2649		 * This must all happen in a single txg.
2650		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2651		 * invoked from zil_claim_log_block()'s i/o done callback.
2652		 * Price of rollback is that we abandon the log.
2653		 */
2654		spa->spa_claiming = B_TRUE;
2655
2656		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2657		    spa_first_txg(spa));
2658		(void) dmu_objset_find(spa_name(spa),
2659		    zil_claim, tx, DS_FIND_CHILDREN);
2660		dmu_tx_commit(tx);
2661
2662		spa->spa_claiming = B_FALSE;
2663
2664		spa_set_log_state(spa, SPA_LOG_GOOD);
2665		spa->spa_sync_on = B_TRUE;
2666		txg_sync_start(spa->spa_dsl_pool);
2667
2668		/*
2669		 * Wait for all claims to sync.  We sync up to the highest
2670		 * claimed log block birth time so that claimed log blocks
2671		 * don't appear to be from the future.  spa_claim_max_txg
2672		 * will have been set for us by either zil_check_log_chain()
2673		 * (invoked from spa_check_logs()) or zil_claim() above.
2674		 */
2675		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2676
2677		/*
2678		 * If the config cache is stale, or we have uninitialized
2679		 * metaslabs (see spa_vdev_add()), then update the config.
2680		 *
2681		 * If this is a verbatim import, trust the current
2682		 * in-core spa_config and update the disk labels.
2683		 */
2684		if (config_cache_txg != spa->spa_config_txg ||
2685		    state == SPA_LOAD_IMPORT ||
2686		    state == SPA_LOAD_RECOVER ||
2687		    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2688			need_update = B_TRUE;
2689
2690		for (int c = 0; c < rvd->vdev_children; c++)
2691			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2692				need_update = B_TRUE;
2693
2694		/*
2695		 * Update the config cache asychronously in case we're the
2696		 * root pool, in which case the config cache isn't writable yet.
2697		 */
2698		if (need_update)
2699			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2700
2701		/*
2702		 * Check all DTLs to see if anything needs resilvering.
2703		 */
2704		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2705		    vdev_resilver_needed(rvd, NULL, NULL))
2706			spa_async_request(spa, SPA_ASYNC_RESILVER);
2707
2708		/*
2709		 * Log the fact that we booted up (so that we can detect if
2710		 * we rebooted in the middle of an operation).
2711		 */
2712		spa_history_log_version(spa, "open");
2713
2714		/*
2715		 * Delete any inconsistent datasets.
2716		 */
2717		(void) dmu_objset_find(spa_name(spa),
2718		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2719
2720		/*
2721		 * Clean up any stale temporary dataset userrefs.
2722		 */
2723		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2724	}
2725
2726	return (0);
2727}
2728
2729static int
2730spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2731{
2732	int mode = spa->spa_mode;
2733
2734	spa_unload(spa);
2735	spa_deactivate(spa);
2736
2737	spa->spa_load_max_txg--;
2738
2739	spa_activate(spa, mode);
2740	spa_async_suspend(spa);
2741
2742	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2743}
2744
2745/*
2746 * If spa_load() fails this function will try loading prior txg's. If
2747 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2748 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2749 * function will not rewind the pool and will return the same error as
2750 * spa_load().
2751 */
2752static int
2753spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2754    uint64_t max_request, int rewind_flags)
2755{
2756	nvlist_t *loadinfo = NULL;
2757	nvlist_t *config = NULL;
2758	int load_error, rewind_error;
2759	uint64_t safe_rewind_txg;
2760	uint64_t min_txg;
2761
2762	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2763		spa->spa_load_max_txg = spa->spa_load_txg;
2764		spa_set_log_state(spa, SPA_LOG_CLEAR);
2765	} else {
2766		spa->spa_load_max_txg = max_request;
2767	}
2768
2769	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2770	    mosconfig);
2771	if (load_error == 0)
2772		return (0);
2773
2774	if (spa->spa_root_vdev != NULL)
2775		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2776
2777	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2778	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2779
2780	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2781		nvlist_free(config);
2782		return (load_error);
2783	}
2784
2785	if (state == SPA_LOAD_RECOVER) {
2786		/* Price of rolling back is discarding txgs, including log */
2787		spa_set_log_state(spa, SPA_LOG_CLEAR);
2788	} else {
2789		/*
2790		 * If we aren't rolling back save the load info from our first
2791		 * import attempt so that we can restore it after attempting
2792		 * to rewind.
2793		 */
2794		loadinfo = spa->spa_load_info;
2795		spa->spa_load_info = fnvlist_alloc();
2796	}
2797
2798	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2799	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2800	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2801	    TXG_INITIAL : safe_rewind_txg;
2802
2803	/*
2804	 * Continue as long as we're finding errors, we're still within
2805	 * the acceptable rewind range, and we're still finding uberblocks
2806	 */
2807	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2808	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2809		if (spa->spa_load_max_txg < safe_rewind_txg)
2810			spa->spa_extreme_rewind = B_TRUE;
2811		rewind_error = spa_load_retry(spa, state, mosconfig);
2812	}
2813
2814	spa->spa_extreme_rewind = B_FALSE;
2815	spa->spa_load_max_txg = UINT64_MAX;
2816
2817	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2818		spa_config_set(spa, config);
2819
2820	if (state == SPA_LOAD_RECOVER) {
2821		ASSERT3P(loadinfo, ==, NULL);
2822		return (rewind_error);
2823	} else {
2824		/* Store the rewind info as part of the initial load info */
2825		fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2826		    spa->spa_load_info);
2827
2828		/* Restore the initial load info */
2829		fnvlist_free(spa->spa_load_info);
2830		spa->spa_load_info = loadinfo;
2831
2832		return (load_error);
2833	}
2834}
2835
2836/*
2837 * Pool Open/Import
2838 *
2839 * The import case is identical to an open except that the configuration is sent
2840 * down from userland, instead of grabbed from the configuration cache.  For the
2841 * case of an open, the pool configuration will exist in the
2842 * POOL_STATE_UNINITIALIZED state.
2843 *
2844 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2845 * the same time open the pool, without having to keep around the spa_t in some
2846 * ambiguous state.
2847 */
2848static int
2849spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2850    nvlist_t **config)
2851{
2852	spa_t *spa;
2853	spa_load_state_t state = SPA_LOAD_OPEN;
2854	int error;
2855	int locked = B_FALSE;
2856
2857	*spapp = NULL;
2858
2859	/*
2860	 * As disgusting as this is, we need to support recursive calls to this
2861	 * function because dsl_dir_open() is called during spa_load(), and ends
2862	 * up calling spa_open() again.  The real fix is to figure out how to
2863	 * avoid dsl_dir_open() calling this in the first place.
2864	 */
2865	if (mutex_owner(&spa_namespace_lock) != curthread) {
2866		mutex_enter(&spa_namespace_lock);
2867		locked = B_TRUE;
2868	}
2869
2870	if ((spa = spa_lookup(pool)) == NULL) {
2871		if (locked)
2872			mutex_exit(&spa_namespace_lock);
2873		return (SET_ERROR(ENOENT));
2874	}
2875
2876	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2877		zpool_rewind_policy_t policy;
2878
2879		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2880		    &policy);
2881		if (policy.zrp_request & ZPOOL_DO_REWIND)
2882			state = SPA_LOAD_RECOVER;
2883
2884		spa_activate(spa, spa_mode_global);
2885
2886		if (state != SPA_LOAD_RECOVER)
2887			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2888
2889		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2890		    policy.zrp_request);
2891
2892		if (error == EBADF) {
2893			/*
2894			 * If vdev_validate() returns failure (indicated by
2895			 * EBADF), it indicates that one of the vdevs indicates
2896			 * that the pool has been exported or destroyed.  If
2897			 * this is the case, the config cache is out of sync and
2898			 * we should remove the pool from the namespace.
2899			 */
2900			spa_unload(spa);
2901			spa_deactivate(spa);
2902			spa_config_sync(spa, B_TRUE, B_TRUE);
2903			spa_remove(spa);
2904			if (locked)
2905				mutex_exit(&spa_namespace_lock);
2906			return (SET_ERROR(ENOENT));
2907		}
2908
2909		if (error) {
2910			/*
2911			 * We can't open the pool, but we still have useful
2912			 * information: the state of each vdev after the
2913			 * attempted vdev_open().  Return this to the user.
2914			 */
2915			if (config != NULL && spa->spa_config) {
2916				VERIFY(nvlist_dup(spa->spa_config, config,
2917				    KM_SLEEP) == 0);
2918				VERIFY(nvlist_add_nvlist(*config,
2919				    ZPOOL_CONFIG_LOAD_INFO,
2920				    spa->spa_load_info) == 0);
2921			}
2922			spa_unload(spa);
2923			spa_deactivate(spa);
2924			spa->spa_last_open_failed = error;
2925			if (locked)
2926				mutex_exit(&spa_namespace_lock);
2927			*spapp = NULL;
2928			return (error);
2929		}
2930	}
2931
2932	spa_open_ref(spa, tag);
2933
2934	if (config != NULL)
2935		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2936
2937	/*
2938	 * If we've recovered the pool, pass back any information we
2939	 * gathered while doing the load.
2940	 */
2941	if (state == SPA_LOAD_RECOVER) {
2942		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2943		    spa->spa_load_info) == 0);
2944	}
2945
2946	if (locked) {
2947		spa->spa_last_open_failed = 0;
2948		spa->spa_last_ubsync_txg = 0;
2949		spa->spa_load_txg = 0;
2950		mutex_exit(&spa_namespace_lock);
2951	}
2952
2953	*spapp = spa;
2954
2955	return (0);
2956}
2957
2958int
2959spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2960    nvlist_t **config)
2961{
2962	return (spa_open_common(name, spapp, tag, policy, config));
2963}
2964
2965int
2966spa_open(const char *name, spa_t **spapp, void *tag)
2967{
2968	return (spa_open_common(name, spapp, tag, NULL, NULL));
2969}
2970
2971/*
2972 * Lookup the given spa_t, incrementing the inject count in the process,
2973 * preventing it from being exported or destroyed.
2974 */
2975spa_t *
2976spa_inject_addref(char *name)
2977{
2978	spa_t *spa;
2979
2980	mutex_enter(&spa_namespace_lock);
2981	if ((spa = spa_lookup(name)) == NULL) {
2982		mutex_exit(&spa_namespace_lock);
2983		return (NULL);
2984	}
2985	spa->spa_inject_ref++;
2986	mutex_exit(&spa_namespace_lock);
2987
2988	return (spa);
2989}
2990
2991void
2992spa_inject_delref(spa_t *spa)
2993{
2994	mutex_enter(&spa_namespace_lock);
2995	spa->spa_inject_ref--;
2996	mutex_exit(&spa_namespace_lock);
2997}
2998
2999/*
3000 * Add spares device information to the nvlist.
3001 */
3002static void
3003spa_add_spares(spa_t *spa, nvlist_t *config)
3004{
3005	nvlist_t **spares;
3006	uint_t i, nspares;
3007	nvlist_t *nvroot;
3008	uint64_t guid;
3009	vdev_stat_t *vs;
3010	uint_t vsc;
3011	uint64_t pool;
3012
3013	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3014
3015	if (spa->spa_spares.sav_count == 0)
3016		return;
3017
3018	VERIFY(nvlist_lookup_nvlist(config,
3019	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3020	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3021	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3022	if (nspares != 0) {
3023		VERIFY(nvlist_add_nvlist_array(nvroot,
3024		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3025		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3026		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3027
3028		/*
3029		 * Go through and find any spares which have since been
3030		 * repurposed as an active spare.  If this is the case, update
3031		 * their status appropriately.
3032		 */
3033		for (i = 0; i < nspares; i++) {
3034			VERIFY(nvlist_lookup_uint64(spares[i],
3035			    ZPOOL_CONFIG_GUID, &guid) == 0);
3036			if (spa_spare_exists(guid, &pool, NULL) &&
3037			    pool != 0ULL) {
3038				VERIFY(nvlist_lookup_uint64_array(
3039				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
3040				    (uint64_t **)&vs, &vsc) == 0);
3041				vs->vs_state = VDEV_STATE_CANT_OPEN;
3042				vs->vs_aux = VDEV_AUX_SPARED;
3043			}
3044		}
3045	}
3046}
3047
3048/*
3049 * Add l2cache device information to the nvlist, including vdev stats.
3050 */
3051static void
3052spa_add_l2cache(spa_t *spa, nvlist_t *config)
3053{
3054	nvlist_t **l2cache;
3055	uint_t i, j, nl2cache;
3056	nvlist_t *nvroot;
3057	uint64_t guid;
3058	vdev_t *vd;
3059	vdev_stat_t *vs;
3060	uint_t vsc;
3061
3062	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3063
3064	if (spa->spa_l2cache.sav_count == 0)
3065		return;
3066
3067	VERIFY(nvlist_lookup_nvlist(config,
3068	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3069	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3070	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3071	if (nl2cache != 0) {
3072		VERIFY(nvlist_add_nvlist_array(nvroot,
3073		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3074		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3075		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3076
3077		/*
3078		 * Update level 2 cache device stats.
3079		 */
3080
3081		for (i = 0; i < nl2cache; i++) {
3082			VERIFY(nvlist_lookup_uint64(l2cache[i],
3083			    ZPOOL_CONFIG_GUID, &guid) == 0);
3084
3085			vd = NULL;
3086			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3087				if (guid ==
3088				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3089					vd = spa->spa_l2cache.sav_vdevs[j];
3090					break;
3091				}
3092			}
3093			ASSERT(vd != NULL);
3094
3095			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3096			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3097			    == 0);
3098			vdev_get_stats(vd, vs);
3099		}
3100	}
3101}
3102
3103static void
3104spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3105{
3106	nvlist_t *features;
3107	zap_cursor_t zc;
3108	zap_attribute_t za;
3109
3110	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3111	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3112
3113	if (spa->spa_feat_for_read_obj != 0) {
3114		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3115		    spa->spa_feat_for_read_obj);
3116		    zap_cursor_retrieve(&zc, &za) == 0;
3117		    zap_cursor_advance(&zc)) {
3118			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3119			    za.za_num_integers == 1);
3120			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3121			    za.za_first_integer));
3122		}
3123		zap_cursor_fini(&zc);
3124	}
3125
3126	if (spa->spa_feat_for_write_obj != 0) {
3127		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3128		    spa->spa_feat_for_write_obj);
3129		    zap_cursor_retrieve(&zc, &za) == 0;
3130		    zap_cursor_advance(&zc)) {
3131			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3132			    za.za_num_integers == 1);
3133			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3134			    za.za_first_integer));
3135		}
3136		zap_cursor_fini(&zc);
3137	}
3138
3139	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3140	    features) == 0);
3141	nvlist_free(features);
3142}
3143
3144int
3145spa_get_stats(const char *name, nvlist_t **config,
3146    char *altroot, size_t buflen)
3147{
3148	int error;
3149	spa_t *spa;
3150
3151	*config = NULL;
3152	error = spa_open_common(name, &spa, FTAG, NULL, config);
3153
3154	if (spa != NULL) {
3155		/*
3156		 * This still leaves a window of inconsistency where the spares
3157		 * or l2cache devices could change and the config would be
3158		 * self-inconsistent.
3159		 */
3160		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3161
3162		if (*config != NULL) {
3163			uint64_t loadtimes[2];
3164
3165			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3166			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3167			VERIFY(nvlist_add_uint64_array(*config,
3168			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3169
3170			VERIFY(nvlist_add_uint64(*config,
3171			    ZPOOL_CONFIG_ERRCOUNT,
3172			    spa_get_errlog_size(spa)) == 0);
3173
3174			if (spa_suspended(spa))
3175				VERIFY(nvlist_add_uint64(*config,
3176				    ZPOOL_CONFIG_SUSPENDED,
3177				    spa->spa_failmode) == 0);
3178
3179			spa_add_spares(spa, *config);
3180			spa_add_l2cache(spa, *config);
3181			spa_add_feature_stats(spa, *config);
3182		}
3183	}
3184
3185	/*
3186	 * We want to get the alternate root even for faulted pools, so we cheat
3187	 * and call spa_lookup() directly.
3188	 */
3189	if (altroot) {
3190		if (spa == NULL) {
3191			mutex_enter(&spa_namespace_lock);
3192			spa = spa_lookup(name);
3193			if (spa)
3194				spa_altroot(spa, altroot, buflen);
3195			else
3196				altroot[0] = '\0';
3197			spa = NULL;
3198			mutex_exit(&spa_namespace_lock);
3199		} else {
3200			spa_altroot(spa, altroot, buflen);
3201		}
3202	}
3203
3204	if (spa != NULL) {
3205		spa_config_exit(spa, SCL_CONFIG, FTAG);
3206		spa_close(spa, FTAG);
3207	}
3208
3209	return (error);
3210}
3211
3212/*
3213 * Validate that the auxiliary device array is well formed.  We must have an
3214 * array of nvlists, each which describes a valid leaf vdev.  If this is an
3215 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3216 * specified, as long as they are well-formed.
3217 */
3218static int
3219spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3220    spa_aux_vdev_t *sav, const char *config, uint64_t version,
3221    vdev_labeltype_t label)
3222{
3223	nvlist_t **dev;
3224	uint_t i, ndev;
3225	vdev_t *vd;
3226	int error;
3227
3228	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3229
3230	/*
3231	 * It's acceptable to have no devs specified.
3232	 */
3233	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3234		return (0);
3235
3236	if (ndev == 0)
3237		return (SET_ERROR(EINVAL));
3238
3239	/*
3240	 * Make sure the pool is formatted with a version that supports this
3241	 * device type.
3242	 */
3243	if (spa_version(spa) < version)
3244		return (SET_ERROR(ENOTSUP));
3245
3246	/*
3247	 * Set the pending device list so we correctly handle device in-use
3248	 * checking.
3249	 */
3250	sav->sav_pending = dev;
3251	sav->sav_npending = ndev;
3252
3253	for (i = 0; i < ndev; i++) {
3254		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3255		    mode)) != 0)
3256			goto out;
3257
3258		if (!vd->vdev_ops->vdev_op_leaf) {
3259			vdev_free(vd);
3260			error = SET_ERROR(EINVAL);
3261			goto out;
3262		}
3263
3264		/*
3265		 * The L2ARC currently only supports disk devices in
3266		 * kernel context.  For user-level testing, we allow it.
3267		 */
3268#ifdef _KERNEL
3269		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3270		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3271			error = SET_ERROR(ENOTBLK);
3272			vdev_free(vd);
3273			goto out;
3274		}
3275#endif
3276		vd->vdev_top = vd;
3277
3278		if ((error = vdev_open(vd)) == 0 &&
3279		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
3280			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3281			    vd->vdev_guid) == 0);
3282		}
3283
3284		vdev_free(vd);
3285
3286		if (error &&
3287		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3288			goto out;
3289		else
3290			error = 0;
3291	}
3292
3293out:
3294	sav->sav_pending = NULL;
3295	sav->sav_npending = 0;
3296	return (error);
3297}
3298
3299static int
3300spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3301{
3302	int error;
3303
3304	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3305
3306	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3307	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3308	    VDEV_LABEL_SPARE)) != 0) {
3309		return (error);
3310	}
3311
3312	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3313	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3314	    VDEV_LABEL_L2CACHE));
3315}
3316
3317static void
3318spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3319    const char *config)
3320{
3321	int i;
3322
3323	if (sav->sav_config != NULL) {
3324		nvlist_t **olddevs;
3325		uint_t oldndevs;
3326		nvlist_t **newdevs;
3327
3328		/*
3329		 * Generate new dev list by concatentating with the
3330		 * current dev list.
3331		 */
3332		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3333		    &olddevs, &oldndevs) == 0);
3334
3335		newdevs = kmem_alloc(sizeof (void *) *
3336		    (ndevs + oldndevs), KM_SLEEP);
3337		for (i = 0; i < oldndevs; i++)
3338			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3339			    KM_SLEEP) == 0);
3340		for (i = 0; i < ndevs; i++)
3341			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3342			    KM_SLEEP) == 0);
3343
3344		VERIFY(nvlist_remove(sav->sav_config, config,
3345		    DATA_TYPE_NVLIST_ARRAY) == 0);
3346
3347		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3348		    config, newdevs, ndevs + oldndevs) == 0);
3349		for (i = 0; i < oldndevs + ndevs; i++)
3350			nvlist_free(newdevs[i]);
3351		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3352	} else {
3353		/*
3354		 * Generate a new dev list.
3355		 */
3356		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3357		    KM_SLEEP) == 0);
3358		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3359		    devs, ndevs) == 0);
3360	}
3361}
3362
3363/*
3364 * Stop and drop level 2 ARC devices
3365 */
3366void
3367spa_l2cache_drop(spa_t *spa)
3368{
3369	vdev_t *vd;
3370	int i;
3371	spa_aux_vdev_t *sav = &spa->spa_l2cache;
3372
3373	for (i = 0; i < sav->sav_count; i++) {
3374		uint64_t pool;
3375
3376		vd = sav->sav_vdevs[i];
3377		ASSERT(vd != NULL);
3378
3379		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3380		    pool != 0ULL && l2arc_vdev_present(vd))
3381			l2arc_remove_vdev(vd);
3382	}
3383}
3384
3385/*
3386 * Pool Creation
3387 */
3388int
3389spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3390    nvlist_t *zplprops)
3391{
3392	spa_t *spa;
3393	char *altroot = NULL;
3394	vdev_t *rvd;
3395	dsl_pool_t *dp;
3396	dmu_tx_t *tx;
3397	int error = 0;
3398	uint64_t txg = TXG_INITIAL;
3399	nvlist_t **spares, **l2cache;
3400	uint_t nspares, nl2cache;
3401	uint64_t version, obj;
3402	boolean_t has_features;
3403
3404	/*
3405	 * If this pool already exists, return failure.
3406	 */
3407	mutex_enter(&spa_namespace_lock);
3408	if (spa_lookup(pool) != NULL) {
3409		mutex_exit(&spa_namespace_lock);
3410		return (SET_ERROR(EEXIST));
3411	}
3412
3413	/*
3414	 * Allocate a new spa_t structure.
3415	 */
3416	(void) nvlist_lookup_string(props,
3417	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3418	spa = spa_add(pool, NULL, altroot);
3419	spa_activate(spa, spa_mode_global);
3420
3421	if (props && (error = spa_prop_validate(spa, props))) {
3422		spa_deactivate(spa);
3423		spa_remove(spa);
3424		mutex_exit(&spa_namespace_lock);
3425		return (error);
3426	}
3427
3428	has_features = B_FALSE;
3429	for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3430	    elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3431		if (zpool_prop_feature(nvpair_name(elem)))
3432			has_features = B_TRUE;
3433	}
3434
3435	if (has_features || nvlist_lookup_uint64(props,
3436	    zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3437		version = SPA_VERSION;
3438	}
3439	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3440
3441	spa->spa_first_txg = txg;
3442	spa->spa_uberblock.ub_txg = txg - 1;
3443	spa->spa_uberblock.ub_version = version;
3444	spa->spa_ubsync = spa->spa_uberblock;
3445
3446	/*
3447	 * Create "The Godfather" zio to hold all async IOs
3448	 */
3449	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3450	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3451
3452	/*
3453	 * Create the root vdev.
3454	 */
3455	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3456
3457	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3458
3459	ASSERT(error != 0 || rvd != NULL);
3460	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3461
3462	if (error == 0 && !zfs_allocatable_devs(nvroot))
3463		error = SET_ERROR(EINVAL);
3464
3465	if (error == 0 &&
3466	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3467	    (error = spa_validate_aux(spa, nvroot, txg,
3468	    VDEV_ALLOC_ADD)) == 0) {
3469		for (int c = 0; c < rvd->vdev_children; c++) {
3470			vdev_metaslab_set_size(rvd->vdev_child[c]);
3471			vdev_expand(rvd->vdev_child[c], txg);
3472		}
3473	}
3474
3475	spa_config_exit(spa, SCL_ALL, FTAG);
3476
3477	if (error != 0) {
3478		spa_unload(spa);
3479		spa_deactivate(spa);
3480		spa_remove(spa);
3481		mutex_exit(&spa_namespace_lock);
3482		return (error);
3483	}
3484
3485	/*
3486	 * Get the list of spares, if specified.
3487	 */
3488	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3489	    &spares, &nspares) == 0) {
3490		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3491		    KM_SLEEP) == 0);
3492		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3493		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3494		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3495		spa_load_spares(spa);
3496		spa_config_exit(spa, SCL_ALL, FTAG);
3497		spa->spa_spares.sav_sync = B_TRUE;
3498	}
3499
3500	/*
3501	 * Get the list of level 2 cache devices, if specified.
3502	 */
3503	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3504	    &l2cache, &nl2cache) == 0) {
3505		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3506		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3507		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3508		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3509		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3510		spa_load_l2cache(spa);
3511		spa_config_exit(spa, SCL_ALL, FTAG);
3512		spa->spa_l2cache.sav_sync = B_TRUE;
3513	}
3514
3515	spa->spa_is_initializing = B_TRUE;
3516	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3517	spa->spa_meta_objset = dp->dp_meta_objset;
3518	spa->spa_is_initializing = B_FALSE;
3519
3520	/*
3521	 * Create DDTs (dedup tables).
3522	 */
3523	ddt_create(spa);
3524
3525	spa_update_dspace(spa);
3526
3527	tx = dmu_tx_create_assigned(dp, txg);
3528
3529	/*
3530	 * Create the pool config object.
3531	 */
3532	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3533	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3534	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3535
3536	if (zap_add(spa->spa_meta_objset,
3537	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3538	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3539		cmn_err(CE_PANIC, "failed to add pool config");
3540	}
3541
3542	if (spa_version(spa) >= SPA_VERSION_FEATURES)
3543		spa_feature_create_zap_objects(spa, tx);
3544
3545	if (zap_add(spa->spa_meta_objset,
3546	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3547	    sizeof (uint64_t), 1, &version, tx) != 0) {
3548		cmn_err(CE_PANIC, "failed to add pool version");
3549	}
3550
3551	/* Newly created pools with the right version are always deflated. */
3552	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3553		spa->spa_deflate = TRUE;
3554		if (zap_add(spa->spa_meta_objset,
3555		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3556		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3557			cmn_err(CE_PANIC, "failed to add deflate");
3558		}
3559	}
3560
3561	/*
3562	 * Create the deferred-free bpobj.  Turn off compression
3563	 * because sync-to-convergence takes longer if the blocksize
3564	 * keeps changing.
3565	 */
3566	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3567	dmu_object_set_compress(spa->spa_meta_objset, obj,
3568	    ZIO_COMPRESS_OFF, tx);
3569	if (zap_add(spa->spa_meta_objset,
3570	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3571	    sizeof (uint64_t), 1, &obj, tx) != 0) {
3572		cmn_err(CE_PANIC, "failed to add bpobj");
3573	}
3574	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3575	    spa->spa_meta_objset, obj));
3576
3577	/*
3578	 * Create the pool's history object.
3579	 */
3580	if (version >= SPA_VERSION_ZPOOL_HISTORY)
3581		spa_history_create_obj(spa, tx);
3582
3583	/*
3584	 * Set pool properties.
3585	 */
3586	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3587	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3588	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3589	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3590
3591	if (props != NULL) {
3592		spa_configfile_set(spa, props, B_FALSE);
3593		spa_sync_props(props, tx);
3594	}
3595
3596	dmu_tx_commit(tx);
3597
3598	spa->spa_sync_on = B_TRUE;
3599	txg_sync_start(spa->spa_dsl_pool);
3600
3601	/*
3602	 * We explicitly wait for the first transaction to complete so that our
3603	 * bean counters are appropriately updated.
3604	 */
3605	txg_wait_synced(spa->spa_dsl_pool, txg);
3606
3607	spa_config_sync(spa, B_FALSE, B_TRUE);
3608
3609	spa_history_log_version(spa, "create");
3610
3611	spa->spa_minref = refcount_count(&spa->spa_refcount);
3612
3613	mutex_exit(&spa_namespace_lock);
3614
3615	return (0);
3616}
3617
3618#ifdef _KERNEL
3619/*
3620 * Get the root pool information from the root disk, then import the root pool
3621 * during the system boot up time.
3622 */
3623extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3624
3625static nvlist_t *
3626spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3627{
3628	nvlist_t *config;
3629	nvlist_t *nvtop, *nvroot;
3630	uint64_t pgid;
3631
3632	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3633		return (NULL);
3634
3635	/*
3636	 * Add this top-level vdev to the child array.
3637	 */
3638	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3639	    &nvtop) == 0);
3640	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3641	    &pgid) == 0);
3642	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3643
3644	/*
3645	 * Put this pool's top-level vdevs into a root vdev.
3646	 */
3647	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3648	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3649	    VDEV_TYPE_ROOT) == 0);
3650	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3651	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3652	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3653	    &nvtop, 1) == 0);
3654
3655	/*
3656	 * Replace the existing vdev_tree with the new root vdev in
3657	 * this pool's configuration (remove the old, add the new).
3658	 */
3659	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3660	nvlist_free(nvroot);
3661	return (config);
3662}
3663
3664/*
3665 * Walk the vdev tree and see if we can find a device with "better"
3666 * configuration. A configuration is "better" if the label on that
3667 * device has a more recent txg.
3668 */
3669static void
3670spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3671{
3672	for (int c = 0; c < vd->vdev_children; c++)
3673		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3674
3675	if (vd->vdev_ops->vdev_op_leaf) {
3676		nvlist_t *label;
3677		uint64_t label_txg;
3678
3679		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3680		    &label) != 0)
3681			return;
3682
3683		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3684		    &label_txg) == 0);
3685
3686		/*
3687		 * Do we have a better boot device?
3688		 */
3689		if (label_txg > *txg) {
3690			*txg = label_txg;
3691			*avd = vd;
3692		}
3693		nvlist_free(label);
3694	}
3695}
3696
3697/*
3698 * Import a root pool.
3699 *
3700 * For x86. devpath_list will consist of devid and/or physpath name of
3701 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3702 * The GRUB "findroot" command will return the vdev we should boot.
3703 *
3704 * For Sparc, devpath_list consists the physpath name of the booting device
3705 * no matter the rootpool is a single device pool or a mirrored pool.
3706 * e.g.
3707 *	"/pci@1f,0/ide@d/disk@0,0:a"
3708 */
3709int
3710spa_import_rootpool(char *devpath, char *devid)
3711{
3712	spa_t *spa;
3713	vdev_t *rvd, *bvd, *avd = NULL;
3714	nvlist_t *config, *nvtop;
3715	uint64_t guid, txg;
3716	char *pname;
3717	int error;
3718
3719	/*
3720	 * Read the label from the boot device and generate a configuration.
3721	 */
3722	config = spa_generate_rootconf(devpath, devid, &guid);
3723#if defined(_OBP) && defined(_KERNEL)
3724	if (config == NULL) {
3725		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3726			/* iscsi boot */
3727			get_iscsi_bootpath_phy(devpath);
3728			config = spa_generate_rootconf(devpath, devid, &guid);
3729		}
3730	}
3731#endif
3732	if (config == NULL) {
3733		cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3734		    devpath);
3735		return (SET_ERROR(EIO));
3736	}
3737
3738	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3739	    &pname) == 0);
3740	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3741
3742	mutex_enter(&spa_namespace_lock);
3743	if ((spa = spa_lookup(pname)) != NULL) {
3744		/*
3745		 * Remove the existing root pool from the namespace so that we
3746		 * can replace it with the correct config we just read in.
3747		 */
3748		spa_remove(spa);
3749	}
3750
3751	spa = spa_add(pname, config, NULL);
3752	spa->spa_is_root = B_TRUE;
3753	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3754
3755	/*
3756	 * Build up a vdev tree based on the boot device's label config.
3757	 */
3758	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3759	    &nvtop) == 0);
3760	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3761	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3762	    VDEV_ALLOC_ROOTPOOL);
3763	spa_config_exit(spa, SCL_ALL, FTAG);
3764	if (error) {
3765		mutex_exit(&spa_namespace_lock);
3766		nvlist_free(config);
3767		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3768		    pname);
3769		return (error);
3770	}
3771
3772	/*
3773	 * Get the boot vdev.
3774	 */
3775	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3776		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3777		    (u_longlong_t)guid);
3778		error = SET_ERROR(ENOENT);
3779		goto out;
3780	}
3781
3782	/*
3783	 * Determine if there is a better boot device.
3784	 */
3785	avd = bvd;
3786	spa_alt_rootvdev(rvd, &avd, &txg);
3787	if (avd != bvd) {
3788		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3789		    "try booting from '%s'", avd->vdev_path);
3790		error = SET_ERROR(EINVAL);
3791		goto out;
3792	}
3793
3794	/*
3795	 * If the boot device is part of a spare vdev then ensure that
3796	 * we're booting off the active spare.
3797	 */
3798	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3799	    !bvd->vdev_isspare) {
3800		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3801		    "try booting from '%s'",
3802		    bvd->vdev_parent->
3803		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3804		error = SET_ERROR(EINVAL);
3805		goto out;
3806	}
3807
3808	error = 0;
3809out:
3810	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3811	vdev_free(rvd);
3812	spa_config_exit(spa, SCL_ALL, FTAG);
3813	mutex_exit(&spa_namespace_lock);
3814
3815	nvlist_free(config);
3816	return (error);
3817}
3818
3819#endif
3820
3821/*
3822 * Import a non-root pool into the system.
3823 */
3824int
3825spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3826{
3827	spa_t *spa;
3828	char *altroot = NULL;
3829	spa_load_state_t state = SPA_LOAD_IMPORT;
3830	zpool_rewind_policy_t policy;
3831	uint64_t mode = spa_mode_global;
3832	uint64_t readonly = B_FALSE;
3833	int error;
3834	nvlist_t *nvroot;
3835	nvlist_t **spares, **l2cache;
3836	uint_t nspares, nl2cache;
3837
3838	/*
3839	 * If a pool with this name exists, return failure.
3840	 */
3841	mutex_enter(&spa_namespace_lock);
3842	if (spa_lookup(pool) != NULL) {
3843		mutex_exit(&spa_namespace_lock);
3844		return (SET_ERROR(EEXIST));
3845	}
3846
3847	/*
3848	 * Create and initialize the spa structure.
3849	 */
3850	(void) nvlist_lookup_string(props,
3851	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3852	(void) nvlist_lookup_uint64(props,
3853	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3854	if (readonly)
3855		mode = FREAD;
3856	spa = spa_add(pool, config, altroot);
3857	spa->spa_import_flags = flags;
3858
3859	/*
3860	 * Verbatim import - Take a pool and insert it into the namespace
3861	 * as if it had been loaded at boot.
3862	 */
3863	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3864		if (props != NULL)
3865			spa_configfile_set(spa, props, B_FALSE);
3866
3867		spa_config_sync(spa, B_FALSE, B_TRUE);
3868
3869		mutex_exit(&spa_namespace_lock);
3870		return (0);
3871	}
3872
3873	spa_activate(spa, mode);
3874
3875	/*
3876	 * Don't start async tasks until we know everything is healthy.
3877	 */
3878	spa_async_suspend(spa);
3879
3880	zpool_get_rewind_policy(config, &policy);
3881	if (policy.zrp_request & ZPOOL_DO_REWIND)
3882		state = SPA_LOAD_RECOVER;
3883
3884	/*
3885	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3886	 * because the user-supplied config is actually the one to trust when
3887	 * doing an import.
3888	 */
3889	if (state != SPA_LOAD_RECOVER)
3890		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3891
3892	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3893	    policy.zrp_request);
3894
3895	/*
3896	 * Propagate anything learned while loading the pool and pass it
3897	 * back to caller (i.e. rewind info, missing devices, etc).
3898	 */
3899	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3900	    spa->spa_load_info) == 0);
3901
3902	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3903	/*
3904	 * Toss any existing sparelist, as it doesn't have any validity
3905	 * anymore, and conflicts with spa_has_spare().
3906	 */
3907	if (spa->spa_spares.sav_config) {
3908		nvlist_free(spa->spa_spares.sav_config);
3909		spa->spa_spares.sav_config = NULL;
3910		spa_load_spares(spa);
3911	}
3912	if (spa->spa_l2cache.sav_config) {
3913		nvlist_free(spa->spa_l2cache.sav_config);
3914		spa->spa_l2cache.sav_config = NULL;
3915		spa_load_l2cache(spa);
3916	}
3917
3918	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3919	    &nvroot) == 0);
3920	if (error == 0)
3921		error = spa_validate_aux(spa, nvroot, -1ULL,
3922		    VDEV_ALLOC_SPARE);
3923	if (error == 0)
3924		error = spa_validate_aux(spa, nvroot, -1ULL,
3925		    VDEV_ALLOC_L2CACHE);
3926	spa_config_exit(spa, SCL_ALL, FTAG);
3927
3928	if (props != NULL)
3929		spa_configfile_set(spa, props, B_FALSE);
3930
3931	if (error != 0 || (props && spa_writeable(spa) &&
3932	    (error = spa_prop_set(spa, props)))) {
3933		spa_unload(spa);
3934		spa_deactivate(spa);
3935		spa_remove(spa);
3936		mutex_exit(&spa_namespace_lock);
3937		return (error);
3938	}
3939
3940	spa_async_resume(spa);
3941
3942	/*
3943	 * Override any spares and level 2 cache devices as specified by
3944	 * the user, as these may have correct device names/devids, etc.
3945	 */
3946	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3947	    &spares, &nspares) == 0) {
3948		if (spa->spa_spares.sav_config)
3949			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3950			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3951		else
3952			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3953			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3954		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3955		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3956		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3957		spa_load_spares(spa);
3958		spa_config_exit(spa, SCL_ALL, FTAG);
3959		spa->spa_spares.sav_sync = B_TRUE;
3960	}
3961	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3962	    &l2cache, &nl2cache) == 0) {
3963		if (spa->spa_l2cache.sav_config)
3964			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3965			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3966		else
3967			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3968			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3969		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3970		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3971		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3972		spa_load_l2cache(spa);
3973		spa_config_exit(spa, SCL_ALL, FTAG);
3974		spa->spa_l2cache.sav_sync = B_TRUE;
3975	}
3976
3977	/*
3978	 * Check for any removed devices.
3979	 */
3980	if (spa->spa_autoreplace) {
3981		spa_aux_check_removed(&spa->spa_spares);
3982		spa_aux_check_removed(&spa->spa_l2cache);
3983	}
3984
3985	if (spa_writeable(spa)) {
3986		/*
3987		 * Update the config cache to include the newly-imported pool.
3988		 */
3989		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3990	}
3991
3992	/*
3993	 * It's possible that the pool was expanded while it was exported.
3994	 * We kick off an async task to handle this for us.
3995	 */
3996	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3997
3998	mutex_exit(&spa_namespace_lock);
3999	spa_history_log_version(spa, "import");
4000
4001	return (0);
4002}
4003
4004nvlist_t *
4005spa_tryimport(nvlist_t *tryconfig)
4006{
4007	nvlist_t *config = NULL;
4008	char *poolname;
4009	spa_t *spa;
4010	uint64_t state;
4011	int error;
4012
4013	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4014		return (NULL);
4015
4016	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4017		return (NULL);
4018
4019	/*
4020	 * Create and initialize the spa structure.
4021	 */
4022	mutex_enter(&spa_namespace_lock);
4023	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4024	spa_activate(spa, FREAD);
4025
4026	/*
4027	 * Pass off the heavy lifting to spa_load().
4028	 * Pass TRUE for mosconfig because the user-supplied config
4029	 * is actually the one to trust when doing an import.
4030	 */
4031	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4032
4033	/*
4034	 * If 'tryconfig' was at least parsable, return the current config.
4035	 */
4036	if (spa->spa_root_vdev != NULL) {
4037		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4038		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4039		    poolname) == 0);
4040		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4041		    state) == 0);
4042		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4043		    spa->spa_uberblock.ub_timestamp) == 0);
4044		VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4045		    spa->spa_load_info) == 0);
4046
4047		/*
4048		 * If the bootfs property exists on this pool then we
4049		 * copy it out so that external consumers can tell which
4050		 * pools are bootable.
4051		 */
4052		if ((!error || error == EEXIST) && spa->spa_bootfs) {
4053			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4054
4055			/*
4056			 * We have to play games with the name since the
4057			 * pool was opened as TRYIMPORT_NAME.
4058			 */
4059			if (dsl_dsobj_to_dsname(spa_name(spa),
4060			    spa->spa_bootfs, tmpname) == 0) {
4061				char *cp;
4062				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4063
4064				cp = strchr(tmpname, '/');
4065				if (cp == NULL) {
4066					(void) strlcpy(dsname, tmpname,
4067					    MAXPATHLEN);
4068				} else {
4069					(void) snprintf(dsname, MAXPATHLEN,
4070					    "%s/%s", poolname, ++cp);
4071				}
4072				VERIFY(nvlist_add_string(config,
4073				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4074				kmem_free(dsname, MAXPATHLEN);
4075			}
4076			kmem_free(tmpname, MAXPATHLEN);
4077		}
4078
4079		/*
4080		 * Add the list of hot spares and level 2 cache devices.
4081		 */
4082		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4083		spa_add_spares(spa, config);
4084		spa_add_l2cache(spa, config);
4085		spa_config_exit(spa, SCL_CONFIG, FTAG);
4086	}
4087
4088	spa_unload(spa);
4089	spa_deactivate(spa);
4090	spa_remove(spa);
4091	mutex_exit(&spa_namespace_lock);
4092
4093	return (config);
4094}
4095
4096/*
4097 * Pool export/destroy
4098 *
4099 * The act of destroying or exporting a pool is very simple.  We make sure there
4100 * is no more pending I/O and any references to the pool are gone.  Then, we
4101 * update the pool state and sync all the labels to disk, removing the
4102 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4103 * we don't sync the labels or remove the configuration cache.
4104 */
4105static int
4106spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4107    boolean_t force, boolean_t hardforce)
4108{
4109	spa_t *spa;
4110
4111	if (oldconfig)
4112		*oldconfig = NULL;
4113
4114	if (!(spa_mode_global & FWRITE))
4115		return (SET_ERROR(EROFS));
4116
4117	mutex_enter(&spa_namespace_lock);
4118	if ((spa = spa_lookup(pool)) == NULL) {
4119		mutex_exit(&spa_namespace_lock);
4120		return (SET_ERROR(ENOENT));
4121	}
4122
4123	/*
4124	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4125	 * reacquire the namespace lock, and see if we can export.
4126	 */
4127	spa_open_ref(spa, FTAG);
4128	mutex_exit(&spa_namespace_lock);
4129	spa_async_suspend(spa);
4130	mutex_enter(&spa_namespace_lock);
4131	spa_close(spa, FTAG);
4132
4133	/*
4134	 * The pool will be in core if it's openable,
4135	 * in which case we can modify its state.
4136	 */
4137	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4138		/*
4139		 * Objsets may be open only because they're dirty, so we
4140		 * have to force it to sync before checking spa_refcnt.
4141		 */
4142		txg_wait_synced(spa->spa_dsl_pool, 0);
4143
4144		/*
4145		 * A pool cannot be exported or destroyed if there are active
4146		 * references.  If we are resetting a pool, allow references by
4147		 * fault injection handlers.
4148		 */
4149		if (!spa_refcount_zero(spa) ||
4150		    (spa->spa_inject_ref != 0 &&
4151		    new_state != POOL_STATE_UNINITIALIZED)) {
4152			spa_async_resume(spa);
4153			mutex_exit(&spa_namespace_lock);
4154			return (SET_ERROR(EBUSY));
4155		}
4156
4157		/*
4158		 * A pool cannot be exported if it has an active shared spare.
4159		 * This is to prevent other pools stealing the active spare
4160		 * from an exported pool. At user's own will, such pool can
4161		 * be forcedly exported.
4162		 */
4163		if (!force && new_state == POOL_STATE_EXPORTED &&
4164		    spa_has_active_shared_spare(spa)) {
4165			spa_async_resume(spa);
4166			mutex_exit(&spa_namespace_lock);
4167			return (SET_ERROR(EXDEV));
4168		}
4169
4170		/*
4171		 * We want this to be reflected on every label,
4172		 * so mark them all dirty.  spa_unload() will do the
4173		 * final sync that pushes these changes out.
4174		 */
4175		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4176			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4177			spa->spa_state = new_state;
4178			spa->spa_final_txg = spa_last_synced_txg(spa) +
4179			    TXG_DEFER_SIZE + 1;
4180			vdev_config_dirty(spa->spa_root_vdev);
4181			spa_config_exit(spa, SCL_ALL, FTAG);
4182		}
4183	}
4184
4185	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4186
4187	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4188		spa_unload(spa);
4189		spa_deactivate(spa);
4190	}
4191
4192	if (oldconfig && spa->spa_config)
4193		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4194
4195	if (new_state != POOL_STATE_UNINITIALIZED) {
4196		if (!hardforce)
4197			spa_config_sync(spa, B_TRUE, B_TRUE);
4198		spa_remove(spa);
4199	}
4200	mutex_exit(&spa_namespace_lock);
4201
4202	return (0);
4203}
4204
4205/*
4206 * Destroy a storage pool.
4207 */
4208int
4209spa_destroy(char *pool)
4210{
4211	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4212	    B_FALSE, B_FALSE));
4213}
4214
4215/*
4216 * Export a storage pool.
4217 */
4218int
4219spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4220    boolean_t hardforce)
4221{
4222	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4223	    force, hardforce));
4224}
4225
4226/*
4227 * Similar to spa_export(), this unloads the spa_t without actually removing it
4228 * from the namespace in any way.
4229 */
4230int
4231spa_reset(char *pool)
4232{
4233	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4234	    B_FALSE, B_FALSE));
4235}
4236
4237/*
4238 * ==========================================================================
4239 * Device manipulation
4240 * ==========================================================================
4241 */
4242
4243/*
4244 * Add a device to a storage pool.
4245 */
4246int
4247spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4248{
4249	uint64_t txg, id;
4250	int error;
4251	vdev_t *rvd = spa->spa_root_vdev;
4252	vdev_t *vd, *tvd;
4253	nvlist_t **spares, **l2cache;
4254	uint_t nspares, nl2cache;
4255
4256	ASSERT(spa_writeable(spa));
4257
4258	txg = spa_vdev_enter(spa);
4259
4260	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4261	    VDEV_ALLOC_ADD)) != 0)
4262		return (spa_vdev_exit(spa, NULL, txg, error));
4263
4264	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
4265
4266	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4267	    &nspares) != 0)
4268		nspares = 0;
4269
4270	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4271	    &nl2cache) != 0)
4272		nl2cache = 0;
4273
4274	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4275		return (spa_vdev_exit(spa, vd, txg, EINVAL));
4276
4277	if (vd->vdev_children != 0 &&
4278	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
4279		return (spa_vdev_exit(spa, vd, txg, error));
4280
4281	/*
4282	 * We must validate the spares and l2cache devices after checking the
4283	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
4284	 */
4285	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4286		return (spa_vdev_exit(spa, vd, txg, error));
4287
4288	/*
4289	 * Transfer each new top-level vdev from vd to rvd.
4290	 */
4291	for (int c = 0; c < vd->vdev_children; c++) {
4292
4293		/*
4294		 * Set the vdev id to the first hole, if one exists.
4295		 */
4296		for (id = 0; id < rvd->vdev_children; id++) {
4297			if (rvd->vdev_child[id]->vdev_ishole) {
4298				vdev_free(rvd->vdev_child[id]);
4299				break;
4300			}
4301		}
4302		tvd = vd->vdev_child[c];
4303		vdev_remove_child(vd, tvd);
4304		tvd->vdev_id = id;
4305		vdev_add_child(rvd, tvd);
4306		vdev_config_dirty(tvd);
4307	}
4308
4309	if (nspares != 0) {
4310		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4311		    ZPOOL_CONFIG_SPARES);
4312		spa_load_spares(spa);
4313		spa->spa_spares.sav_sync = B_TRUE;
4314	}
4315
4316	if (nl2cache != 0) {
4317		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4318		    ZPOOL_CONFIG_L2CACHE);
4319		spa_load_l2cache(spa);
4320		spa->spa_l2cache.sav_sync = B_TRUE;
4321	}
4322
4323	/*
4324	 * We have to be careful when adding new vdevs to an existing pool.
4325	 * If other threads start allocating from these vdevs before we
4326	 * sync the config cache, and we lose power, then upon reboot we may
4327	 * fail to open the pool because there are DVAs that the config cache
4328	 * can't translate.  Therefore, we first add the vdevs without
4329	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4330	 * and then let spa_config_update() initialize the new metaslabs.
4331	 *
4332	 * spa_load() checks for added-but-not-initialized vdevs, so that
4333	 * if we lose power at any point in this sequence, the remaining
4334	 * steps will be completed the next time we load the pool.
4335	 */
4336	(void) spa_vdev_exit(spa, vd, txg, 0);
4337
4338	mutex_enter(&spa_namespace_lock);
4339	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4340	mutex_exit(&spa_namespace_lock);
4341
4342	return (0);
4343}
4344
4345/*
4346 * Attach a device to a mirror.  The arguments are the path to any device
4347 * in the mirror, and the nvroot for the new device.  If the path specifies
4348 * a device that is not mirrored, we automatically insert the mirror vdev.
4349 *
4350 * If 'replacing' is specified, the new device is intended to replace the
4351 * existing device; in this case the two devices are made into their own
4352 * mirror using the 'replacing' vdev, which is functionally identical to
4353 * the mirror vdev (it actually reuses all the same ops) but has a few
4354 * extra rules: you can't attach to it after it's been created, and upon
4355 * completion of resilvering, the first disk (the one being replaced)
4356 * is automatically detached.
4357 */
4358int
4359spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4360{
4361	uint64_t txg, dtl_max_txg;
4362	vdev_t *rvd = spa->spa_root_vdev;
4363	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4364	vdev_ops_t *pvops;
4365	char *oldvdpath, *newvdpath;
4366	int newvd_isspare;
4367	int error;
4368
4369	ASSERT(spa_writeable(spa));
4370
4371	txg = spa_vdev_enter(spa);
4372
4373	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4374
4375	if (oldvd == NULL)
4376		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4377
4378	if (!oldvd->vdev_ops->vdev_op_leaf)
4379		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4380
4381	pvd = oldvd->vdev_parent;
4382
4383	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4384	    VDEV_ALLOC_ATTACH)) != 0)
4385		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4386
4387	if (newrootvd->vdev_children != 1)
4388		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4389
4390	newvd = newrootvd->vdev_child[0];
4391
4392	if (!newvd->vdev_ops->vdev_op_leaf)
4393		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4394
4395	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4396		return (spa_vdev_exit(spa, newrootvd, txg, error));
4397
4398	/*
4399	 * Spares can't replace logs
4400	 */
4401	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4402		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4403
4404	if (!replacing) {
4405		/*
4406		 * For attach, the only allowable parent is a mirror or the root
4407		 * vdev.
4408		 */
4409		if (pvd->vdev_ops != &vdev_mirror_ops &&
4410		    pvd->vdev_ops != &vdev_root_ops)
4411			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4412
4413		pvops = &vdev_mirror_ops;
4414	} else {
4415		/*
4416		 * Active hot spares can only be replaced by inactive hot
4417		 * spares.
4418		 */
4419		if (pvd->vdev_ops == &vdev_spare_ops &&
4420		    oldvd->vdev_isspare &&
4421		    !spa_has_spare(spa, newvd->vdev_guid))
4422			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4423
4424		/*
4425		 * If the source is a hot spare, and the parent isn't already a
4426		 * spare, then we want to create a new hot spare.  Otherwise, we
4427		 * want to create a replacing vdev.  The user is not allowed to
4428		 * attach to a spared vdev child unless the 'isspare' state is
4429		 * the same (spare replaces spare, non-spare replaces
4430		 * non-spare).
4431		 */
4432		if (pvd->vdev_ops == &vdev_replacing_ops &&
4433		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4434			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4435		} else if (pvd->vdev_ops == &vdev_spare_ops &&
4436		    newvd->vdev_isspare != oldvd->vdev_isspare) {
4437			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4438		}
4439
4440		if (newvd->vdev_isspare)
4441			pvops = &vdev_spare_ops;
4442		else
4443			pvops = &vdev_replacing_ops;
4444	}
4445
4446	/*
4447	 * Make sure the new device is big enough.
4448	 */
4449	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4450		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4451
4452	/*
4453	 * The new device cannot have a higher alignment requirement
4454	 * than the top-level vdev.
4455	 */
4456	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4457		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4458
4459	/*
4460	 * If this is an in-place replacement, update oldvd's path and devid
4461	 * to make it distinguishable from newvd, and unopenable from now on.
4462	 */
4463	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4464		spa_strfree(oldvd->vdev_path);
4465		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4466		    KM_SLEEP);
4467		(void) sprintf(oldvd->vdev_path, "%s/%s",
4468		    newvd->vdev_path, "old");
4469		if (oldvd->vdev_devid != NULL) {
4470			spa_strfree(oldvd->vdev_devid);
4471			oldvd->vdev_devid = NULL;
4472		}
4473	}
4474
4475	/* mark the device being resilvered */
4476	newvd->vdev_resilver_txg = txg;
4477
4478	/*
4479	 * If the parent is not a mirror, or if we're replacing, insert the new
4480	 * mirror/replacing/spare vdev above oldvd.
4481	 */
4482	if (pvd->vdev_ops != pvops)
4483		pvd = vdev_add_parent(oldvd, pvops);
4484
4485	ASSERT(pvd->vdev_top->vdev_parent == rvd);
4486	ASSERT(pvd->vdev_ops == pvops);
4487	ASSERT(oldvd->vdev_parent == pvd);
4488
4489	/*
4490	 * Extract the new device from its root and add it to pvd.
4491	 */
4492	vdev_remove_child(newrootvd, newvd);
4493	newvd->vdev_id = pvd->vdev_children;
4494	newvd->vdev_crtxg = oldvd->vdev_crtxg;
4495	vdev_add_child(pvd, newvd);
4496
4497	tvd = newvd->vdev_top;
4498	ASSERT(pvd->vdev_top == tvd);
4499	ASSERT(tvd->vdev_parent == rvd);
4500
4501	vdev_config_dirty(tvd);
4502
4503	/*
4504	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4505	 * for any dmu_sync-ed blocks.  It will propagate upward when
4506	 * spa_vdev_exit() calls vdev_dtl_reassess().
4507	 */
4508	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4509
4510	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4511	    dtl_max_txg - TXG_INITIAL);
4512
4513	if (newvd->vdev_isspare) {
4514		spa_spare_activate(newvd);
4515		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4516	}
4517
4518	oldvdpath = spa_strdup(oldvd->vdev_path);
4519	newvdpath = spa_strdup(newvd->vdev_path);
4520	newvd_isspare = newvd->vdev_isspare;
4521
4522	/*
4523	 * Mark newvd's DTL dirty in this txg.
4524	 */
4525	vdev_dirty(tvd, VDD_DTL, newvd, txg);
4526
4527	/*
4528	 * Schedule the resilver to restart in the future. We do this to
4529	 * ensure that dmu_sync-ed blocks have been stitched into the
4530	 * respective datasets.
4531	 */
4532	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4533
4534	/*
4535	 * Commit the config
4536	 */
4537	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4538
4539	spa_history_log_internal(spa, "vdev attach", NULL,
4540	    "%s vdev=%s %s vdev=%s",
4541	    replacing && newvd_isspare ? "spare in" :
4542	    replacing ? "replace" : "attach", newvdpath,
4543	    replacing ? "for" : "to", oldvdpath);
4544
4545	spa_strfree(oldvdpath);
4546	spa_strfree(newvdpath);
4547
4548	if (spa->spa_bootfs)
4549		spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4550
4551	return (0);
4552}
4553
4554/*
4555 * Detach a device from a mirror or replacing vdev.
4556 *
4557 * If 'replace_done' is specified, only detach if the parent
4558 * is a replacing vdev.
4559 */
4560int
4561spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4562{
4563	uint64_t txg;
4564	int error;
4565	vdev_t *rvd = spa->spa_root_vdev;
4566	vdev_t *vd, *pvd, *cvd, *tvd;
4567	boolean_t unspare = B_FALSE;
4568	uint64_t unspare_guid = 0;
4569	char *vdpath;
4570
4571	ASSERT(spa_writeable(spa));
4572
4573	txg = spa_vdev_enter(spa);
4574
4575	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4576
4577	if (vd == NULL)
4578		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4579
4580	if (!vd->vdev_ops->vdev_op_leaf)
4581		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4582
4583	pvd = vd->vdev_parent;
4584
4585	/*
4586	 * If the parent/child relationship is not as expected, don't do it.
4587	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4588	 * vdev that's replacing B with C.  The user's intent in replacing
4589	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4590	 * the replace by detaching C, the expected behavior is to end up
4591	 * M(A,B).  But suppose that right after deciding to detach C,
4592	 * the replacement of B completes.  We would have M(A,C), and then
4593	 * ask to detach C, which would leave us with just A -- not what
4594	 * the user wanted.  To prevent this, we make sure that the
4595	 * parent/child relationship hasn't changed -- in this example,
4596	 * that C's parent is still the replacing vdev R.
4597	 */
4598	if (pvd->vdev_guid != pguid && pguid != 0)
4599		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4600
4601	/*
4602	 * Only 'replacing' or 'spare' vdevs can be replaced.
4603	 */
4604	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4605	    pvd->vdev_ops != &vdev_spare_ops)
4606		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4607
4608	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4609	    spa_version(spa) >= SPA_VERSION_SPARES);
4610
4611	/*
4612	 * Only mirror, replacing, and spare vdevs support detach.
4613	 */
4614	if (pvd->vdev_ops != &vdev_replacing_ops &&
4615	    pvd->vdev_ops != &vdev_mirror_ops &&
4616	    pvd->vdev_ops != &vdev_spare_ops)
4617		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4618
4619	/*
4620	 * If this device has the only valid copy of some data,
4621	 * we cannot safely detach it.
4622	 */
4623	if (vdev_dtl_required(vd))
4624		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4625
4626	ASSERT(pvd->vdev_children >= 2);
4627
4628	/*
4629	 * If we are detaching the second disk from a replacing vdev, then
4630	 * check to see if we changed the original vdev's path to have "/old"
4631	 * at the end in spa_vdev_attach().  If so, undo that change now.
4632	 */
4633	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4634	    vd->vdev_path != NULL) {
4635		size_t len = strlen(vd->vdev_path);
4636
4637		for (int c = 0; c < pvd->vdev_children; c++) {
4638			cvd = pvd->vdev_child[c];
4639
4640			if (cvd == vd || cvd->vdev_path == NULL)
4641				continue;
4642
4643			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4644			    strcmp(cvd->vdev_path + len, "/old") == 0) {
4645				spa_strfree(cvd->vdev_path);
4646				cvd->vdev_path = spa_strdup(vd->vdev_path);
4647				break;
4648			}
4649		}
4650	}
4651
4652	/*
4653	 * If we are detaching the original disk from a spare, then it implies
4654	 * that the spare should become a real disk, and be removed from the
4655	 * active spare list for the pool.
4656	 */
4657	if (pvd->vdev_ops == &vdev_spare_ops &&
4658	    vd->vdev_id == 0 &&
4659	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4660		unspare = B_TRUE;
4661
4662	/*
4663	 * Erase the disk labels so the disk can be used for other things.
4664	 * This must be done after all other error cases are handled,
4665	 * but before we disembowel vd (so we can still do I/O to it).
4666	 * But if we can't do it, don't treat the error as fatal --
4667	 * it may be that the unwritability of the disk is the reason
4668	 * it's being detached!
4669	 */
4670	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4671
4672	/*
4673	 * Remove vd from its parent and compact the parent's children.
4674	 */
4675	vdev_remove_child(pvd, vd);
4676	vdev_compact_children(pvd);
4677
4678	/*
4679	 * Remember one of the remaining children so we can get tvd below.
4680	 */
4681	cvd = pvd->vdev_child[pvd->vdev_children - 1];
4682
4683	/*
4684	 * If we need to remove the remaining child from the list of hot spares,
4685	 * do it now, marking the vdev as no longer a spare in the process.
4686	 * We must do this before vdev_remove_parent(), because that can
4687	 * change the GUID if it creates a new toplevel GUID.  For a similar
4688	 * reason, we must remove the spare now, in the same txg as the detach;
4689	 * otherwise someone could attach a new sibling, change the GUID, and
4690	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4691	 */
4692	if (unspare) {
4693		ASSERT(cvd->vdev_isspare);
4694		spa_spare_remove(cvd);
4695		unspare_guid = cvd->vdev_guid;
4696		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4697		cvd->vdev_unspare = B_TRUE;
4698	}
4699
4700	/*
4701	 * If the parent mirror/replacing vdev only has one child,
4702	 * the parent is no longer needed.  Remove it from the tree.
4703	 */
4704	if (pvd->vdev_children == 1) {
4705		if (pvd->vdev_ops == &vdev_spare_ops)
4706			cvd->vdev_unspare = B_FALSE;
4707		vdev_remove_parent(cvd);
4708	}
4709
4710
4711	/*
4712	 * We don't set tvd until now because the parent we just removed
4713	 * may have been the previous top-level vdev.
4714	 */
4715	tvd = cvd->vdev_top;
4716	ASSERT(tvd->vdev_parent == rvd);
4717
4718	/*
4719	 * Reevaluate the parent vdev state.
4720	 */
4721	vdev_propagate_state(cvd);
4722
4723	/*
4724	 * If the 'autoexpand' property is set on the pool then automatically
4725	 * try to expand the size of the pool. For example if the device we
4726	 * just detached was smaller than the others, it may be possible to
4727	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4728	 * first so that we can obtain the updated sizes of the leaf vdevs.
4729	 */
4730	if (spa->spa_autoexpand) {
4731		vdev_reopen(tvd);
4732		vdev_expand(tvd, txg);
4733	}
4734
4735	vdev_config_dirty(tvd);
4736
4737	/*
4738	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4739	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4740	 * But first make sure we're not on any *other* txg's DTL list, to
4741	 * prevent vd from being accessed after it's freed.
4742	 */
4743	vdpath = spa_strdup(vd->vdev_path);
4744	for (int t = 0; t < TXG_SIZE; t++)
4745		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4746	vd->vdev_detached = B_TRUE;
4747	vdev_dirty(tvd, VDD_DTL, vd, txg);
4748
4749	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4750
4751	/* hang on to the spa before we release the lock */
4752	spa_open_ref(spa, FTAG);
4753
4754	error = spa_vdev_exit(spa, vd, txg, 0);
4755
4756	spa_history_log_internal(spa, "detach", NULL,
4757	    "vdev=%s", vdpath);
4758	spa_strfree(vdpath);
4759
4760	/*
4761	 * If this was the removal of the original device in a hot spare vdev,
4762	 * then we want to go through and remove the device from the hot spare
4763	 * list of every other pool.
4764	 */
4765	if (unspare) {
4766		spa_t *altspa = NULL;
4767
4768		mutex_enter(&spa_namespace_lock);
4769		while ((altspa = spa_next(altspa)) != NULL) {
4770			if (altspa->spa_state != POOL_STATE_ACTIVE ||
4771			    altspa == spa)
4772				continue;
4773
4774			spa_open_ref(altspa, FTAG);
4775			mutex_exit(&spa_namespace_lock);
4776			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4777			mutex_enter(&spa_namespace_lock);
4778			spa_close(altspa, FTAG);
4779		}
4780		mutex_exit(&spa_namespace_lock);
4781
4782		/* search the rest of the vdevs for spares to remove */
4783		spa_vdev_resilver_done(spa);
4784	}
4785
4786	/* all done with the spa; OK to release */
4787	mutex_enter(&spa_namespace_lock);
4788	spa_close(spa, FTAG);
4789	mutex_exit(&spa_namespace_lock);
4790
4791	return (error);
4792}
4793
4794/*
4795 * Split a set of devices from their mirrors, and create a new pool from them.
4796 */
4797int
4798spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4799    nvlist_t *props, boolean_t exp)
4800{
4801	int error = 0;
4802	uint64_t txg, *glist;
4803	spa_t *newspa;
4804	uint_t c, children, lastlog;
4805	nvlist_t **child, *nvl, *tmp;
4806	dmu_tx_t *tx;
4807	char *altroot = NULL;
4808	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4809	boolean_t activate_slog;
4810
4811	ASSERT(spa_writeable(spa));
4812
4813	txg = spa_vdev_enter(spa);
4814
4815	/* clear the log and flush everything up to now */
4816	activate_slog = spa_passivate_log(spa);
4817	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4818	error = spa_offline_log(spa);
4819	txg = spa_vdev_config_enter(spa);
4820
4821	if (activate_slog)
4822		spa_activate_log(spa);
4823
4824	if (error != 0)
4825		return (spa_vdev_exit(spa, NULL, txg, error));
4826
4827	/* check new spa name before going any further */
4828	if (spa_lookup(newname) != NULL)
4829		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4830
4831	/*
4832	 * scan through all the children to ensure they're all mirrors
4833	 */
4834	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4835	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4836	    &children) != 0)
4837		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4838
4839	/* first, check to ensure we've got the right child count */
4840	rvd = spa->spa_root_vdev;
4841	lastlog = 0;
4842	for (c = 0; c < rvd->vdev_children; c++) {
4843		vdev_t *vd = rvd->vdev_child[c];
4844
4845		/* don't count the holes & logs as children */
4846		if (vd->vdev_islog || vd->vdev_ishole) {
4847			if (lastlog == 0)
4848				lastlog = c;
4849			continue;
4850		}
4851
4852		lastlog = 0;
4853	}
4854	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4855		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4856
4857	/* next, ensure no spare or cache devices are part of the split */
4858	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4859	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4860		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4861
4862	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4863	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4864
4865	/* then, loop over each vdev and validate it */
4866	for (c = 0; c < children; c++) {
4867		uint64_t is_hole = 0;
4868
4869		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4870		    &is_hole);
4871
4872		if (is_hole != 0) {
4873			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4874			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4875				continue;
4876			} else {
4877				error = SET_ERROR(EINVAL);
4878				break;
4879			}
4880		}
4881
4882		/* which disk is going to be split? */
4883		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4884		    &glist[c]) != 0) {
4885			error = SET_ERROR(EINVAL);
4886			break;
4887		}
4888
4889		/* look it up in the spa */
4890		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4891		if (vml[c] == NULL) {
4892			error = SET_ERROR(ENODEV);
4893			break;
4894		}
4895
4896		/* make sure there's nothing stopping the split */
4897		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4898		    vml[c]->vdev_islog ||
4899		    vml[c]->vdev_ishole ||
4900		    vml[c]->vdev_isspare ||
4901		    vml[c]->vdev_isl2cache ||
4902		    !vdev_writeable(vml[c]) ||
4903		    vml[c]->vdev_children != 0 ||
4904		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4905		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4906			error = SET_ERROR(EINVAL);
4907			break;
4908		}
4909
4910		if (vdev_dtl_required(vml[c])) {
4911			error = SET_ERROR(EBUSY);
4912			break;
4913		}
4914
4915		/* we need certain info from the top level */
4916		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4917		    vml[c]->vdev_top->vdev_ms_array) == 0);
4918		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4919		    vml[c]->vdev_top->vdev_ms_shift) == 0);
4920		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4921		    vml[c]->vdev_top->vdev_asize) == 0);
4922		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4923		    vml[c]->vdev_top->vdev_ashift) == 0);
4924	}
4925
4926	if (error != 0) {
4927		kmem_free(vml, children * sizeof (vdev_t *));
4928		kmem_free(glist, children * sizeof (uint64_t));
4929		return (spa_vdev_exit(spa, NULL, txg, error));
4930	}
4931
4932	/* stop writers from using the disks */
4933	for (c = 0; c < children; c++) {
4934		if (vml[c] != NULL)
4935			vml[c]->vdev_offline = B_TRUE;
4936	}
4937	vdev_reopen(spa->spa_root_vdev);
4938
4939	/*
4940	 * Temporarily record the splitting vdevs in the spa config.  This
4941	 * will disappear once the config is regenerated.
4942	 */
4943	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4944	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4945	    glist, children) == 0);
4946	kmem_free(glist, children * sizeof (uint64_t));
4947
4948	mutex_enter(&spa->spa_props_lock);
4949	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4950	    nvl) == 0);
4951	mutex_exit(&spa->spa_props_lock);
4952	spa->spa_config_splitting = nvl;
4953	vdev_config_dirty(spa->spa_root_vdev);
4954
4955	/* configure and create the new pool */
4956	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4957	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4958	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4959	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4960	    spa_version(spa)) == 0);
4961	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4962	    spa->spa_config_txg) == 0);
4963	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4964	    spa_generate_guid(NULL)) == 0);
4965	(void) nvlist_lookup_string(props,
4966	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4967
4968	/* add the new pool to the namespace */
4969	newspa = spa_add(newname, config, altroot);
4970	newspa->spa_config_txg = spa->spa_config_txg;
4971	spa_set_log_state(newspa, SPA_LOG_CLEAR);
4972
4973	/* release the spa config lock, retaining the namespace lock */
4974	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4975
4976	if (zio_injection_enabled)
4977		zio_handle_panic_injection(spa, FTAG, 1);
4978
4979	spa_activate(newspa, spa_mode_global);
4980	spa_async_suspend(newspa);
4981
4982	/* create the new pool from the disks of the original pool */
4983	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4984	if (error)
4985		goto out;
4986
4987	/* if that worked, generate a real config for the new pool */
4988	if (newspa->spa_root_vdev != NULL) {
4989		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4990		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4991		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4992		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4993		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4994		    B_TRUE));
4995	}
4996
4997	/* set the props */
4998	if (props != NULL) {
4999		spa_configfile_set(newspa, props, B_FALSE);
5000		error = spa_prop_set(newspa, props);
5001		if (error)
5002			goto out;
5003	}
5004
5005	/* flush everything */
5006	txg = spa_vdev_config_enter(newspa);
5007	vdev_config_dirty(newspa->spa_root_vdev);
5008	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5009
5010	if (zio_injection_enabled)
5011		zio_handle_panic_injection(spa, FTAG, 2);
5012
5013	spa_async_resume(newspa);
5014
5015	/* finally, update the original pool's config */
5016	txg = spa_vdev_config_enter(spa);
5017	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5018	error = dmu_tx_assign(tx, TXG_WAIT);
5019	if (error != 0)
5020		dmu_tx_abort(tx);
5021	for (c = 0; c < children; c++) {
5022		if (vml[c] != NULL) {
5023			vdev_split(vml[c]);
5024			if (error == 0)
5025				spa_history_log_internal(spa, "detach", tx,
5026				    "vdev=%s", vml[c]->vdev_path);
5027			vdev_free(vml[c]);
5028		}
5029	}
5030	vdev_config_dirty(spa->spa_root_vdev);
5031	spa->spa_config_splitting = NULL;
5032	nvlist_free(nvl);
5033	if (error == 0)
5034		dmu_tx_commit(tx);
5035	(void) spa_vdev_exit(spa, NULL, txg, 0);
5036
5037	if (zio_injection_enabled)
5038		zio_handle_panic_injection(spa, FTAG, 3);
5039
5040	/* split is complete; log a history record */
5041	spa_history_log_internal(newspa, "split", NULL,
5042	    "from pool %s", spa_name(spa));
5043
5044	kmem_free(vml, children * sizeof (vdev_t *));
5045
5046	/* if we're not going to mount the filesystems in userland, export */
5047	if (exp)
5048		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5049		    B_FALSE, B_FALSE);
5050
5051	return (error);
5052
5053out:
5054	spa_unload(newspa);
5055	spa_deactivate(newspa);
5056	spa_remove(newspa);
5057
5058	txg = spa_vdev_config_enter(spa);
5059
5060	/* re-online all offlined disks */
5061	for (c = 0; c < children; c++) {
5062		if (vml[c] != NULL)
5063			vml[c]->vdev_offline = B_FALSE;
5064	}
5065	vdev_reopen(spa->spa_root_vdev);
5066
5067	nvlist_free(spa->spa_config_splitting);
5068	spa->spa_config_splitting = NULL;
5069	(void) spa_vdev_exit(spa, NULL, txg, error);
5070
5071	kmem_free(vml, children * sizeof (vdev_t *));
5072	return (error);
5073}
5074
5075static nvlist_t *
5076spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5077{
5078	for (int i = 0; i < count; i++) {
5079		uint64_t guid;
5080
5081		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5082		    &guid) == 0);
5083
5084		if (guid == target_guid)
5085			return (nvpp[i]);
5086	}
5087
5088	return (NULL);
5089}
5090
5091static void
5092spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5093	nvlist_t *dev_to_remove)
5094{
5095	nvlist_t **newdev = NULL;
5096
5097	if (count > 1)
5098		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5099
5100	for (int i = 0, j = 0; i < count; i++) {
5101		if (dev[i] == dev_to_remove)
5102			continue;
5103		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5104	}
5105
5106	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5107	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5108
5109	for (int i = 0; i < count - 1; i++)
5110		nvlist_free(newdev[i]);
5111
5112	if (count > 1)
5113		kmem_free(newdev, (count - 1) * sizeof (void *));
5114}
5115
5116/*
5117 * Evacuate the device.
5118 */
5119static int
5120spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5121{
5122	uint64_t txg;
5123	int error = 0;
5124
5125	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5126	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5127	ASSERT(vd == vd->vdev_top);
5128
5129	/*
5130	 * Evacuate the device.  We don't hold the config lock as writer
5131	 * since we need to do I/O but we do keep the
5132	 * spa_namespace_lock held.  Once this completes the device
5133	 * should no longer have any blocks allocated on it.
5134	 */
5135	if (vd->vdev_islog) {
5136		if (vd->vdev_stat.vs_alloc != 0)
5137			error = spa_offline_log(spa);
5138	} else {
5139		error = SET_ERROR(ENOTSUP);
5140	}
5141
5142	if (error)
5143		return (error);
5144
5145	/*
5146	 * The evacuation succeeded.  Remove any remaining MOS metadata
5147	 * associated with this vdev, and wait for these changes to sync.
5148	 */
5149	ASSERT0(vd->vdev_stat.vs_alloc);
5150	txg = spa_vdev_config_enter(spa);
5151	vd->vdev_removing = B_TRUE;
5152	vdev_dirty_leaves(vd, VDD_DTL, txg);
5153	vdev_config_dirty(vd);
5154	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5155
5156	return (0);
5157}
5158
5159/*
5160 * Complete the removal by cleaning up the namespace.
5161 */
5162static void
5163spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5164{
5165	vdev_t *rvd = spa->spa_root_vdev;
5166	uint64_t id = vd->vdev_id;
5167	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5168
5169	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5170	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5171	ASSERT(vd == vd->vdev_top);
5172
5173	/*
5174	 * Only remove any devices which are empty.
5175	 */
5176	if (vd->vdev_stat.vs_alloc != 0)
5177		return;
5178
5179	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5180
5181	if (list_link_active(&vd->vdev_state_dirty_node))
5182		vdev_state_clean(vd);
5183	if (list_link_active(&vd->vdev_config_dirty_node))
5184		vdev_config_clean(vd);
5185
5186	vdev_free(vd);
5187
5188	if (last_vdev) {
5189		vdev_compact_children(rvd);
5190	} else {
5191		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5192		vdev_add_child(rvd, vd);
5193	}
5194	vdev_config_dirty(rvd);
5195
5196	/*
5197	 * Reassess the health of our root vdev.
5198	 */
5199	vdev_reopen(rvd);
5200}
5201
5202/*
5203 * Remove a device from the pool -
5204 *
5205 * Removing a device from the vdev namespace requires several steps
5206 * and can take a significant amount of time.  As a result we use
5207 * the spa_vdev_config_[enter/exit] functions which allow us to
5208 * grab and release the spa_config_lock while still holding the namespace
5209 * lock.  During each step the configuration is synced out.
5210 *
5211 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5212 * devices.
5213 */
5214int
5215spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5216{
5217	vdev_t *vd;
5218	metaslab_group_t *mg;
5219	nvlist_t **spares, **l2cache, *nv;
5220	uint64_t txg = 0;
5221	uint_t nspares, nl2cache;
5222	int error = 0;
5223	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5224
5225	ASSERT(spa_writeable(spa));
5226
5227	if (!locked)
5228		txg = spa_vdev_enter(spa);
5229
5230	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5231
5232	if (spa->spa_spares.sav_vdevs != NULL &&
5233	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5234	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5235	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5236		/*
5237		 * Only remove the hot spare if it's not currently in use
5238		 * in this pool.
5239		 */
5240		if (vd == NULL || unspare) {
5241			spa_vdev_remove_aux(spa->spa_spares.sav_config,
5242			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5243			spa_load_spares(spa);
5244			spa->spa_spares.sav_sync = B_TRUE;
5245		} else {
5246			error = SET_ERROR(EBUSY);
5247		}
5248	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
5249	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5250	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5251	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5252		/*
5253		 * Cache devices can always be removed.
5254		 */
5255		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5256		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5257		spa_load_l2cache(spa);
5258		spa->spa_l2cache.sav_sync = B_TRUE;
5259	} else if (vd != NULL && vd->vdev_islog) {
5260		ASSERT(!locked);
5261		ASSERT(vd == vd->vdev_top);
5262
5263		/*
5264		 * XXX - Once we have bp-rewrite this should
5265		 * become the common case.
5266		 */
5267
5268		mg = vd->vdev_mg;
5269
5270		/*
5271		 * Stop allocating from this vdev.
5272		 */
5273		metaslab_group_passivate(mg);
5274
5275		/*
5276		 * Wait for the youngest allocations and frees to sync,
5277		 * and then wait for the deferral of those frees to finish.
5278		 */
5279		spa_vdev_config_exit(spa, NULL,
5280		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5281
5282		/*
5283		 * Attempt to evacuate the vdev.
5284		 */
5285		error = spa_vdev_remove_evacuate(spa, vd);
5286
5287		txg = spa_vdev_config_enter(spa);
5288
5289		/*
5290		 * If we couldn't evacuate the vdev, unwind.
5291		 */
5292		if (error) {
5293			metaslab_group_activate(mg);
5294			return (spa_vdev_exit(spa, NULL, txg, error));
5295		}
5296
5297		/*
5298		 * Clean up the vdev namespace.
5299		 */
5300		spa_vdev_remove_from_namespace(spa, vd);
5301
5302	} else if (vd != NULL) {
5303		/*
5304		 * Normal vdevs cannot be removed (yet).
5305		 */
5306		error = SET_ERROR(ENOTSUP);
5307	} else {
5308		/*
5309		 * There is no vdev of any kind with the specified guid.
5310		 */
5311		error = SET_ERROR(ENOENT);
5312	}
5313
5314	if (!locked)
5315		return (spa_vdev_exit(spa, NULL, txg, error));
5316
5317	return (error);
5318}
5319
5320/*
5321 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5322 * currently spared, so we can detach it.
5323 */
5324static vdev_t *
5325spa_vdev_resilver_done_hunt(vdev_t *vd)
5326{
5327	vdev_t *newvd, *oldvd;
5328
5329	for (int c = 0; c < vd->vdev_children; c++) {
5330		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5331		if (oldvd != NULL)
5332			return (oldvd);
5333	}
5334
5335	/*
5336	 * Check for a completed replacement.  We always consider the first
5337	 * vdev in the list to be the oldest vdev, and the last one to be
5338	 * the newest (see spa_vdev_attach() for how that works).  In
5339	 * the case where the newest vdev is faulted, we will not automatically
5340	 * remove it after a resilver completes.  This is OK as it will require
5341	 * user intervention to determine which disk the admin wishes to keep.
5342	 */
5343	if (vd->vdev_ops == &vdev_replacing_ops) {
5344		ASSERT(vd->vdev_children > 1);
5345
5346		newvd = vd->vdev_child[vd->vdev_children - 1];
5347		oldvd = vd->vdev_child[0];
5348
5349		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5350		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5351		    !vdev_dtl_required(oldvd))
5352			return (oldvd);
5353	}
5354
5355	/*
5356	 * Check for a completed resilver with the 'unspare' flag set.
5357	 */
5358	if (vd->vdev_ops == &vdev_spare_ops) {
5359		vdev_t *first = vd->vdev_child[0];
5360		vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5361
5362		if (last->vdev_unspare) {
5363			oldvd = first;
5364			newvd = last;
5365		} else if (first->vdev_unspare) {
5366			oldvd = last;
5367			newvd = first;
5368		} else {
5369			oldvd = NULL;
5370		}
5371
5372		if (oldvd != NULL &&
5373		    vdev_dtl_empty(newvd, DTL_MISSING) &&
5374		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5375		    !vdev_dtl_required(oldvd))
5376			return (oldvd);
5377
5378		/*
5379		 * If there are more than two spares attached to a disk,
5380		 * and those spares are not required, then we want to
5381		 * attempt to free them up now so that they can be used
5382		 * by other pools.  Once we're back down to a single
5383		 * disk+spare, we stop removing them.
5384		 */
5385		if (vd->vdev_children > 2) {
5386			newvd = vd->vdev_child[1];
5387
5388			if (newvd->vdev_isspare && last->vdev_isspare &&
5389			    vdev_dtl_empty(last, DTL_MISSING) &&
5390			    vdev_dtl_empty(last, DTL_OUTAGE) &&
5391			    !vdev_dtl_required(newvd))
5392				return (newvd);
5393		}
5394	}
5395
5396	return (NULL);
5397}
5398
5399static void
5400spa_vdev_resilver_done(spa_t *spa)
5401{
5402	vdev_t *vd, *pvd, *ppvd;
5403	uint64_t guid, sguid, pguid, ppguid;
5404
5405	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5406
5407	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5408		pvd = vd->vdev_parent;
5409		ppvd = pvd->vdev_parent;
5410		guid = vd->vdev_guid;
5411		pguid = pvd->vdev_guid;
5412		ppguid = ppvd->vdev_guid;
5413		sguid = 0;
5414		/*
5415		 * If we have just finished replacing a hot spared device, then
5416		 * we need to detach the parent's first child (the original hot
5417		 * spare) as well.
5418		 */
5419		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5420		    ppvd->vdev_children == 2) {
5421			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5422			sguid = ppvd->vdev_child[1]->vdev_guid;
5423		}
5424		ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5425
5426		spa_config_exit(spa, SCL_ALL, FTAG);
5427		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5428			return;
5429		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5430			return;
5431		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5432	}
5433
5434	spa_config_exit(spa, SCL_ALL, FTAG);
5435}
5436
5437/*
5438 * Update the stored path or FRU for this vdev.
5439 */
5440int
5441spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5442    boolean_t ispath)
5443{
5444	vdev_t *vd;
5445	boolean_t sync = B_FALSE;
5446
5447	ASSERT(spa_writeable(spa));
5448
5449	spa_vdev_state_enter(spa, SCL_ALL);
5450
5451	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5452		return (spa_vdev_state_exit(spa, NULL, ENOENT));
5453
5454	if (!vd->vdev_ops->vdev_op_leaf)
5455		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5456
5457	if (ispath) {
5458		if (strcmp(value, vd->vdev_path) != 0) {
5459			spa_strfree(vd->vdev_path);
5460			vd->vdev_path = spa_strdup(value);
5461			sync = B_TRUE;
5462		}
5463	} else {
5464		if (vd->vdev_fru == NULL) {
5465			vd->vdev_fru = spa_strdup(value);
5466			sync = B_TRUE;
5467		} else if (strcmp(value, vd->vdev_fru) != 0) {
5468			spa_strfree(vd->vdev_fru);
5469			vd->vdev_fru = spa_strdup(value);
5470			sync = B_TRUE;
5471		}
5472	}
5473
5474	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5475}
5476
5477int
5478spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5479{
5480	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5481}
5482
5483int
5484spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5485{
5486	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5487}
5488
5489/*
5490 * ==========================================================================
5491 * SPA Scanning
5492 * ==========================================================================
5493 */
5494
5495int
5496spa_scan_stop(spa_t *spa)
5497{
5498	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5499	if (dsl_scan_resilvering(spa->spa_dsl_pool))
5500		return (SET_ERROR(EBUSY));
5501	return (dsl_scan_cancel(spa->spa_dsl_pool));
5502}
5503
5504int
5505spa_scan(spa_t *spa, pool_scan_func_t func)
5506{
5507	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5508
5509	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5510		return (SET_ERROR(ENOTSUP));
5511
5512	/*
5513	 * If a resilver was requested, but there is no DTL on a
5514	 * writeable leaf device, we have nothing to do.
5515	 */
5516	if (func == POOL_SCAN_RESILVER &&
5517	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5518		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5519		return (0);
5520	}
5521
5522	return (dsl_scan(spa->spa_dsl_pool, func));
5523}
5524
5525/*
5526 * ==========================================================================
5527 * SPA async task processing
5528 * ==========================================================================
5529 */
5530
5531static void
5532spa_async_remove(spa_t *spa, vdev_t *vd)
5533{
5534	if (vd->vdev_remove_wanted) {
5535		vd->vdev_remove_wanted = B_FALSE;
5536		vd->vdev_delayed_close = B_FALSE;
5537		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5538
5539		/*
5540		 * We want to clear the stats, but we don't want to do a full
5541		 * vdev_clear() as that will cause us to throw away
5542		 * degraded/faulted state as well as attempt to reopen the
5543		 * device, all of which is a waste.
5544		 */
5545		vd->vdev_stat.vs_read_errors = 0;
5546		vd->vdev_stat.vs_write_errors = 0;
5547		vd->vdev_stat.vs_checksum_errors = 0;
5548
5549		vdev_state_dirty(vd->vdev_top);
5550	}
5551
5552	for (int c = 0; c < vd->vdev_children; c++)
5553		spa_async_remove(spa, vd->vdev_child[c]);
5554}
5555
5556static void
5557spa_async_probe(spa_t *spa, vdev_t *vd)
5558{
5559	if (vd->vdev_probe_wanted) {
5560		vd->vdev_probe_wanted = B_FALSE;
5561		vdev_reopen(vd);	/* vdev_open() does the actual probe */
5562	}
5563
5564	for (int c = 0; c < vd->vdev_children; c++)
5565		spa_async_probe(spa, vd->vdev_child[c]);
5566}
5567
5568static void
5569spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5570{
5571	sysevent_id_t eid;
5572	nvlist_t *attr;
5573	char *physpath;
5574
5575	if (!spa->spa_autoexpand)
5576		return;
5577
5578	for (int c = 0; c < vd->vdev_children; c++) {
5579		vdev_t *cvd = vd->vdev_child[c];
5580		spa_async_autoexpand(spa, cvd);
5581	}
5582
5583	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5584		return;
5585
5586	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5587	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5588
5589	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5590	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5591
5592	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5593	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5594
5595	nvlist_free(attr);
5596	kmem_free(physpath, MAXPATHLEN);
5597}
5598
5599static void
5600spa_async_thread(spa_t *spa)
5601{
5602	int tasks;
5603
5604	ASSERT(spa->spa_sync_on);
5605
5606	mutex_enter(&spa->spa_async_lock);
5607	tasks = spa->spa_async_tasks;
5608	spa->spa_async_tasks = 0;
5609	mutex_exit(&spa->spa_async_lock);
5610
5611	/*
5612	 * See if the config needs to be updated.
5613	 */
5614	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5615		uint64_t old_space, new_space;
5616
5617		mutex_enter(&spa_namespace_lock);
5618		old_space = metaslab_class_get_space(spa_normal_class(spa));
5619		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5620		new_space = metaslab_class_get_space(spa_normal_class(spa));
5621		mutex_exit(&spa_namespace_lock);
5622
5623		/*
5624		 * If the pool grew as a result of the config update,
5625		 * then log an internal history event.
5626		 */
5627		if (new_space != old_space) {
5628			spa_history_log_internal(spa, "vdev online", NULL,
5629			    "pool '%s' size: %llu(+%llu)",
5630			    spa_name(spa), new_space, new_space - old_space);
5631		}
5632	}
5633
5634	/*
5635	 * See if any devices need to be marked REMOVED.
5636	 */
5637	if (tasks & SPA_ASYNC_REMOVE) {
5638		spa_vdev_state_enter(spa, SCL_NONE);
5639		spa_async_remove(spa, spa->spa_root_vdev);
5640		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5641			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5642		for (int i = 0; i < spa->spa_spares.sav_count; i++)
5643			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5644		(void) spa_vdev_state_exit(spa, NULL, 0);
5645	}
5646
5647	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5648		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5649		spa_async_autoexpand(spa, spa->spa_root_vdev);
5650		spa_config_exit(spa, SCL_CONFIG, FTAG);
5651	}
5652
5653	/*
5654	 * See if any devices need to be probed.
5655	 */
5656	if (tasks & SPA_ASYNC_PROBE) {
5657		spa_vdev_state_enter(spa, SCL_NONE);
5658		spa_async_probe(spa, spa->spa_root_vdev);
5659		(void) spa_vdev_state_exit(spa, NULL, 0);
5660	}
5661
5662	/*
5663	 * If any devices are done replacing, detach them.
5664	 */
5665	if (tasks & SPA_ASYNC_RESILVER_DONE)
5666		spa_vdev_resilver_done(spa);
5667
5668	/*
5669	 * Kick off a resilver.
5670	 */
5671	if (tasks & SPA_ASYNC_RESILVER)
5672		dsl_resilver_restart(spa->spa_dsl_pool, 0);
5673
5674	/*
5675	 * Let the world know that we're done.
5676	 */
5677	mutex_enter(&spa->spa_async_lock);
5678	spa->spa_async_thread = NULL;
5679	cv_broadcast(&spa->spa_async_cv);
5680	mutex_exit(&spa->spa_async_lock);
5681	thread_exit();
5682}
5683
5684void
5685spa_async_suspend(spa_t *spa)
5686{
5687	mutex_enter(&spa->spa_async_lock);
5688	spa->spa_async_suspended++;
5689	while (spa->spa_async_thread != NULL)
5690		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5691	mutex_exit(&spa->spa_async_lock);
5692}
5693
5694void
5695spa_async_resume(spa_t *spa)
5696{
5697	mutex_enter(&spa->spa_async_lock);
5698	ASSERT(spa->spa_async_suspended != 0);
5699	spa->spa_async_suspended--;
5700	mutex_exit(&spa->spa_async_lock);
5701}
5702
5703static boolean_t
5704spa_async_tasks_pending(spa_t *spa)
5705{
5706	uint_t non_config_tasks;
5707	uint_t config_task;
5708	boolean_t config_task_suspended;
5709
5710	non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5711	config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5712	if (spa->spa_ccw_fail_time == 0) {
5713		config_task_suspended = B_FALSE;
5714	} else {
5715		config_task_suspended =
5716		    (gethrtime() - spa->spa_ccw_fail_time) <
5717		    (zfs_ccw_retry_interval * NANOSEC);
5718	}
5719
5720	return (non_config_tasks || (config_task && !config_task_suspended));
5721}
5722
5723static void
5724spa_async_dispatch(spa_t *spa)
5725{
5726	mutex_enter(&spa->spa_async_lock);
5727	if (spa_async_tasks_pending(spa) &&
5728	    !spa->spa_async_suspended &&
5729	    spa->spa_async_thread == NULL &&
5730	    rootdir != NULL)
5731		spa->spa_async_thread = thread_create(NULL, 0,
5732		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5733	mutex_exit(&spa->spa_async_lock);
5734}
5735
5736void
5737spa_async_request(spa_t *spa, int task)
5738{
5739	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5740	mutex_enter(&spa->spa_async_lock);
5741	spa->spa_async_tasks |= task;
5742	mutex_exit(&spa->spa_async_lock);
5743}
5744
5745/*
5746 * ==========================================================================
5747 * SPA syncing routines
5748 * ==========================================================================
5749 */
5750
5751static int
5752bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5753{
5754	bpobj_t *bpo = arg;
5755	bpobj_enqueue(bpo, bp, tx);
5756	return (0);
5757}
5758
5759static int
5760spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5761{
5762	zio_t *zio = arg;
5763
5764	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5765	    zio->io_flags));
5766	return (0);
5767}
5768
5769/*
5770 * Note: this simple function is not inlined to make it easier to dtrace the
5771 * amount of time spent syncing frees.
5772 */
5773static void
5774spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
5775{
5776	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5777	bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
5778	VERIFY(zio_wait(zio) == 0);
5779}
5780
5781/*
5782 * Note: this simple function is not inlined to make it easier to dtrace the
5783 * amount of time spent syncing deferred frees.
5784 */
5785static void
5786spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
5787{
5788	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5789	VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
5790	    spa_free_sync_cb, zio, tx), ==, 0);
5791	VERIFY0(zio_wait(zio));
5792}
5793
5794
5795static void
5796spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5797{
5798	char *packed = NULL;
5799	size_t bufsize;
5800	size_t nvsize = 0;
5801	dmu_buf_t *db;
5802
5803	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5804
5805	/*
5806	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5807	 * information.  This avoids the dmu_buf_will_dirty() path and
5808	 * saves us a pre-read to get data we don't actually care about.
5809	 */
5810	bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5811	packed = kmem_alloc(bufsize, KM_SLEEP);
5812
5813	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5814	    KM_SLEEP) == 0);
5815	bzero(packed + nvsize, bufsize - nvsize);
5816
5817	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5818
5819	kmem_free(packed, bufsize);
5820
5821	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5822	dmu_buf_will_dirty(db, tx);
5823	*(uint64_t *)db->db_data = nvsize;
5824	dmu_buf_rele(db, FTAG);
5825}
5826
5827static void
5828spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5829    const char *config, const char *entry)
5830{
5831	nvlist_t *nvroot;
5832	nvlist_t **list;
5833	int i;
5834
5835	if (!sav->sav_sync)
5836		return;
5837
5838	/*
5839	 * Update the MOS nvlist describing the list of available devices.
5840	 * spa_validate_aux() will have already made sure this nvlist is
5841	 * valid and the vdevs are labeled appropriately.
5842	 */
5843	if (sav->sav_object == 0) {
5844		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5845		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5846		    sizeof (uint64_t), tx);
5847		VERIFY(zap_update(spa->spa_meta_objset,
5848		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5849		    &sav->sav_object, tx) == 0);
5850	}
5851
5852	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5853	if (sav->sav_count == 0) {
5854		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5855	} else {
5856		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5857		for (i = 0; i < sav->sav_count; i++)
5858			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5859			    B_FALSE, VDEV_CONFIG_L2CACHE);
5860		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5861		    sav->sav_count) == 0);
5862		for (i = 0; i < sav->sav_count; i++)
5863			nvlist_free(list[i]);
5864		kmem_free(list, sav->sav_count * sizeof (void *));
5865	}
5866
5867	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5868	nvlist_free(nvroot);
5869
5870	sav->sav_sync = B_FALSE;
5871}
5872
5873static void
5874spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5875{
5876	nvlist_t *config;
5877
5878	if (list_is_empty(&spa->spa_config_dirty_list))
5879		return;
5880
5881	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5882
5883	config = spa_config_generate(spa, spa->spa_root_vdev,
5884	    dmu_tx_get_txg(tx), B_FALSE);
5885
5886	/*
5887	 * If we're upgrading the spa version then make sure that
5888	 * the config object gets updated with the correct version.
5889	 */
5890	if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5891		fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5892		    spa->spa_uberblock.ub_version);
5893
5894	spa_config_exit(spa, SCL_STATE, FTAG);
5895
5896	if (spa->spa_config_syncing)
5897		nvlist_free(spa->spa_config_syncing);
5898	spa->spa_config_syncing = config;
5899
5900	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5901}
5902
5903static void
5904spa_sync_version(void *arg, dmu_tx_t *tx)
5905{
5906	uint64_t *versionp = arg;
5907	uint64_t version = *versionp;
5908	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5909
5910	/*
5911	 * Setting the version is special cased when first creating the pool.
5912	 */
5913	ASSERT(tx->tx_txg != TXG_INITIAL);
5914
5915	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5916	ASSERT(version >= spa_version(spa));
5917
5918	spa->spa_uberblock.ub_version = version;
5919	vdev_config_dirty(spa->spa_root_vdev);
5920	spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5921}
5922
5923/*
5924 * Set zpool properties.
5925 */
5926static void
5927spa_sync_props(void *arg, dmu_tx_t *tx)
5928{
5929	nvlist_t *nvp = arg;
5930	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5931	objset_t *mos = spa->spa_meta_objset;
5932	nvpair_t *elem = NULL;
5933
5934	mutex_enter(&spa->spa_props_lock);
5935
5936	while ((elem = nvlist_next_nvpair(nvp, elem))) {
5937		uint64_t intval;
5938		char *strval, *fname;
5939		zpool_prop_t prop;
5940		const char *propname;
5941		zprop_type_t proptype;
5942		spa_feature_t fid;
5943
5944		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5945		case ZPROP_INVAL:
5946			/*
5947			 * We checked this earlier in spa_prop_validate().
5948			 */
5949			ASSERT(zpool_prop_feature(nvpair_name(elem)));
5950
5951			fname = strchr(nvpair_name(elem), '@') + 1;
5952			VERIFY0(zfeature_lookup_name(fname, &fid));
5953
5954			spa_feature_enable(spa, fid, tx);
5955			spa_history_log_internal(spa, "set", tx,
5956			    "%s=enabled", nvpair_name(elem));
5957			break;
5958
5959		case ZPOOL_PROP_VERSION:
5960			intval = fnvpair_value_uint64(elem);
5961			/*
5962			 * The version is synced seperatly before other
5963			 * properties and should be correct by now.
5964			 */
5965			ASSERT3U(spa_version(spa), >=, intval);
5966			break;
5967
5968		case ZPOOL_PROP_ALTROOT:
5969			/*
5970			 * 'altroot' is a non-persistent property. It should
5971			 * have been set temporarily at creation or import time.
5972			 */
5973			ASSERT(spa->spa_root != NULL);
5974			break;
5975
5976		case ZPOOL_PROP_READONLY:
5977		case ZPOOL_PROP_CACHEFILE:
5978			/*
5979			 * 'readonly' and 'cachefile' are also non-persisitent
5980			 * properties.
5981			 */
5982			break;
5983		case ZPOOL_PROP_COMMENT:
5984			strval = fnvpair_value_string(elem);
5985			if (spa->spa_comment != NULL)
5986				spa_strfree(spa->spa_comment);
5987			spa->spa_comment = spa_strdup(strval);
5988			/*
5989			 * We need to dirty the configuration on all the vdevs
5990			 * so that their labels get updated.  It's unnecessary
5991			 * to do this for pool creation since the vdev's
5992			 * configuratoin has already been dirtied.
5993			 */
5994			if (tx->tx_txg != TXG_INITIAL)
5995				vdev_config_dirty(spa->spa_root_vdev);
5996			spa_history_log_internal(spa, "set", tx,
5997			    "%s=%s", nvpair_name(elem), strval);
5998			break;
5999		default:
6000			/*
6001			 * Set pool property values in the poolprops mos object.
6002			 */
6003			if (spa->spa_pool_props_object == 0) {
6004				spa->spa_pool_props_object =
6005				    zap_create_link(mos, DMU_OT_POOL_PROPS,
6006				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6007				    tx);
6008			}
6009
6010			/* normalize the property name */
6011			propname = zpool_prop_to_name(prop);
6012			proptype = zpool_prop_get_type(prop);
6013
6014			if (nvpair_type(elem) == DATA_TYPE_STRING) {
6015				ASSERT(proptype == PROP_TYPE_STRING);
6016				strval = fnvpair_value_string(elem);
6017				VERIFY0(zap_update(mos,
6018				    spa->spa_pool_props_object, propname,
6019				    1, strlen(strval) + 1, strval, tx));
6020				spa_history_log_internal(spa, "set", tx,
6021				    "%s=%s", nvpair_name(elem), strval);
6022			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6023				intval = fnvpair_value_uint64(elem);
6024
6025				if (proptype == PROP_TYPE_INDEX) {
6026					const char *unused;
6027					VERIFY0(zpool_prop_index_to_string(
6028					    prop, intval, &unused));
6029				}
6030				VERIFY0(zap_update(mos,
6031				    spa->spa_pool_props_object, propname,
6032				    8, 1, &intval, tx));
6033				spa_history_log_internal(spa, "set", tx,
6034				    "%s=%lld", nvpair_name(elem), intval);
6035			} else {
6036				ASSERT(0); /* not allowed */
6037			}
6038
6039			switch (prop) {
6040			case ZPOOL_PROP_DELEGATION:
6041				spa->spa_delegation = intval;
6042				break;
6043			case ZPOOL_PROP_BOOTFS:
6044				spa->spa_bootfs = intval;
6045				break;
6046			case ZPOOL_PROP_FAILUREMODE:
6047				spa->spa_failmode = intval;
6048				break;
6049			case ZPOOL_PROP_AUTOEXPAND:
6050				spa->spa_autoexpand = intval;
6051				if (tx->