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 (c) 2015, Nexenta Systems, Inc.  All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 * Copyright 2013 Saso Kiselkov. All rights reserved.
28 * Copyright (c) 2014 Integros [integros.com]
29 */
30
31/*
32 * SPA: Storage Pool Allocator
33 *
34 * This file contains all the routines used when modifying on-disk SPA state.
35 * This includes opening, importing, destroying, exporting a pool, and syncing a
36 * pool.
37 */
38
39#include <sys/zfs_context.h>
40#include <sys/fm/fs/zfs.h>
41#include <sys/spa_impl.h>
42#include <sys/zio.h>
43#include <sys/zio_checksum.h>
44#include <sys/dmu.h>
45#include <sys/dmu_tx.h>
46#include <sys/zap.h>
47#include <sys/zil.h>
48#include <sys/ddt.h>
49#include <sys/vdev_impl.h>
50#include <sys/metaslab.h>
51#include <sys/metaslab_impl.h>
52#include <sys/uberblock_impl.h>
53#include <sys/txg.h>
54#include <sys/avl.h>
55#include <sys/dmu_traverse.h>
56#include <sys/dmu_objset.h>
57#include <sys/unique.h>
58#include <sys/dsl_pool.h>
59#include <sys/dsl_dataset.h>
60#include <sys/dsl_dir.h>
61#include <sys/dsl_prop.h>
62#include <sys/dsl_synctask.h>
63#include <sys/fs/zfs.h>
64#include <sys/arc.h>
65#include <sys/callb.h>
66#include <sys/systeminfo.h>
67#include <sys/spa_boot.h>
68#include <sys/zfs_ioctl.h>
69#include <sys/dsl_scan.h>
70#include <sys/zfeature.h>
71#include <sys/dsl_destroy.h>
72
73#ifdef	_KERNEL
74#include <sys/bootprops.h>
75#include <sys/callb.h>
76#include <sys/cpupart.h>
77#include <sys/pool.h>
78#include <sys/sysdc.h>
79#include <sys/zone.h>
80#endif	/* _KERNEL */
81
82#include "zfs_prop.h"
83#include "zfs_comutil.h"
84
85/*
86 * The interval, in seconds, at which failed configuration cache file writes
87 * should be retried.
88 */
89static int zfs_ccw_retry_interval = 300;
90
91typedef enum zti_modes {
92	ZTI_MODE_FIXED,			/* value is # of threads (min 1) */
93	ZTI_MODE_BATCH,			/* cpu-intensive; value is ignored */
94	ZTI_MODE_NULL,			/* don't create a taskq */
95	ZTI_NMODES
96} zti_modes_t;
97
98#define	ZTI_P(n, q)	{ ZTI_MODE_FIXED, (n), (q) }
99#define	ZTI_BATCH	{ ZTI_MODE_BATCH, 0, 1 }
100#define	ZTI_NULL	{ ZTI_MODE_NULL, 0, 0 }
101
102#define	ZTI_N(n)	ZTI_P(n, 1)
103#define	ZTI_ONE		ZTI_N(1)
104
105typedef struct zio_taskq_info {
106	zti_modes_t zti_mode;
107	uint_t zti_value;
108	uint_t zti_count;
109} zio_taskq_info_t;
110
111static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
112	"issue", "issue_high", "intr", "intr_high"
113};
114
115/*
116 * This table defines the taskq settings for each ZFS I/O type. When
117 * initializing a pool, we use this table to create an appropriately sized
118 * taskq. Some operations are low volume and therefore have a small, static
119 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
120 * macros. Other operations process a large amount of data; the ZTI_BATCH
121 * macro causes us to create a taskq oriented for throughput. Some operations
122 * are so high frequency and short-lived that the taskq itself can become a a
123 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
124 * additional degree of parallelism specified by the number of threads per-
125 * taskq and the number of taskqs; when dispatching an event in this case, the
126 * particular taskq is chosen at random.
127 *
128 * The different taskq priorities are to handle the different contexts (issue
129 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
130 * need to be handled with minimum delay.
131 */
132const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
133	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
134	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* NULL */
135	{ ZTI_N(8),	ZTI_NULL,	ZTI_P(12, 8),	ZTI_NULL }, /* READ */
136	{ ZTI_BATCH,	ZTI_N(5),	ZTI_N(8),	ZTI_N(5) }, /* WRITE */
137	{ ZTI_P(12, 8),	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* FREE */
138	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* CLAIM */
139	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* IOCTL */
140};
141
142static void spa_sync_version(void *arg, dmu_tx_t *tx);
143static void spa_sync_props(void *arg, dmu_tx_t *tx);
144static boolean_t spa_has_active_shared_spare(spa_t *spa);
145static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
146    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
147    char **ereport);
148static void spa_vdev_resilver_done(spa_t *spa);
149
150uint_t		zio_taskq_batch_pct = 75;	/* 1 thread per cpu in pset */
151id_t		zio_taskq_psrset_bind = PS_NONE;
152boolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
153uint_t		zio_taskq_basedc = 80;		/* base duty cycle */
154
155boolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
156extern int	zfs_sync_pass_deferred_free;
157
158/*
159 * This (illegal) pool name is used when temporarily importing a spa_t in order
160 * to get the vdev stats associated with the imported devices.
161 */
162#define	TRYIMPORT_NAME	"$import"
163
164/*
165 * ==========================================================================
166 * SPA properties routines
167 * ==========================================================================
168 */
169
170/*
171 * Add a (source=src, propname=propval) list to an nvlist.
172 */
173static void
174spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
175    uint64_t intval, zprop_source_t src)
176{
177	const char *propname = zpool_prop_to_name(prop);
178	nvlist_t *propval;
179
180	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
181	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
182
183	if (strval != NULL)
184		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
185	else
186		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
187
188	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
189	nvlist_free(propval);
190}
191
192/*
193 * Get property values from the spa configuration.
194 */
195static void
196spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
197{
198	vdev_t *rvd = spa->spa_root_vdev;
199	dsl_pool_t *pool = spa->spa_dsl_pool;
200	uint64_t size, alloc, cap, version;
201	zprop_source_t src = ZPROP_SRC_NONE;
202	spa_config_dirent_t *dp;
203	metaslab_class_t *mc = spa_normal_class(spa);
204
205	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
206
207	if (rvd != NULL) {
208		alloc = metaslab_class_get_alloc(spa_normal_class(spa));
209		size = metaslab_class_get_space(spa_normal_class(spa));
210		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
211		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
212		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
213		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
214		    size - alloc, src);
215
216		spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
217		    metaslab_class_fragmentation(mc), src);
218		spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
219		    metaslab_class_expandable_space(mc), src);
220		spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
221		    (spa_mode(spa) == FREAD), src);
222
223		cap = (size == 0) ? 0 : (alloc * 100 / size);
224		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
225
226		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
227		    ddt_get_pool_dedup_ratio(spa), src);
228
229		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
230		    rvd->vdev_state, src);
231
232		version = spa_version(spa);
233		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
234			src = ZPROP_SRC_DEFAULT;
235		else
236			src = ZPROP_SRC_LOCAL;
237		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
238	}
239
240	if (pool != NULL) {
241		/*
242		 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
243		 * when opening pools before this version freedir will be NULL.
244		 */
245		if (pool->dp_free_dir != NULL) {
246			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
247			    dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
248			    src);
249		} else {
250			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
251			    NULL, 0, src);
252		}
253
254		if (pool->dp_leak_dir != NULL) {
255			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
256			    dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
257			    src);
258		} else {
259			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
260			    NULL, 0, src);
261		}
262	}
263
264	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
265
266	if (spa->spa_comment != NULL) {
267		spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
268		    0, ZPROP_SRC_LOCAL);
269	}
270
271	if (spa->spa_root != NULL)
272		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
273		    0, ZPROP_SRC_LOCAL);
274
275	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
276		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
277		    MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
278	} else {
279		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
280		    SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
281	}
282
283	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
284		if (dp->scd_path == NULL) {
285			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
286			    "none", 0, ZPROP_SRC_LOCAL);
287		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
288			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
289			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
290		}
291	}
292}
293
294/*
295 * Get zpool property values.
296 */
297int
298spa_prop_get(spa_t *spa, nvlist_t **nvp)
299{
300	objset_t *mos = spa->spa_meta_objset;
301	zap_cursor_t zc;
302	zap_attribute_t za;
303	int err;
304
305	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
306
307	mutex_enter(&spa->spa_props_lock);
308
309	/*
310	 * Get properties from the spa config.
311	 */
312	spa_prop_get_config(spa, nvp);
313
314	/* If no pool property object, no more prop to get. */
315	if (mos == NULL || spa->spa_pool_props_object == 0) {
316		mutex_exit(&spa->spa_props_lock);
317		return (0);
318	}
319
320	/*
321	 * Get properties from the MOS pool property object.
322	 */
323	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
324	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
325	    zap_cursor_advance(&zc)) {
326		uint64_t intval = 0;
327		char *strval = NULL;
328		zprop_source_t src = ZPROP_SRC_DEFAULT;
329		zpool_prop_t prop;
330
331		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
332			continue;
333
334		switch (za.za_integer_length) {
335		case 8:
336			/* integer property */
337			if (za.za_first_integer !=
338			    zpool_prop_default_numeric(prop))
339				src = ZPROP_SRC_LOCAL;
340
341			if (prop == ZPOOL_PROP_BOOTFS) {
342				dsl_pool_t *dp;
343				dsl_dataset_t *ds = NULL;
344
345				dp = spa_get_dsl(spa);
346				dsl_pool_config_enter(dp, FTAG);
347				if (err = dsl_dataset_hold_obj(dp,
348				    za.za_first_integer, FTAG, &ds)) {
349					dsl_pool_config_exit(dp, FTAG);
350					break;
351				}
352
353				strval = kmem_alloc(
354				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
355				    KM_SLEEP);
356				dsl_dataset_name(ds, strval);
357				dsl_dataset_rele(ds, FTAG);
358				dsl_pool_config_exit(dp, FTAG);
359			} else {
360				strval = NULL;
361				intval = za.za_first_integer;
362			}
363
364			spa_prop_add_list(*nvp, prop, strval, intval, src);
365
366			if (strval != NULL)
367				kmem_free(strval,
368				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
369
370			break;
371
372		case 1:
373			/* string property */
374			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
375			err = zap_lookup(mos, spa->spa_pool_props_object,
376			    za.za_name, 1, za.za_num_integers, strval);
377			if (err) {
378				kmem_free(strval, za.za_num_integers);
379				break;
380			}
381			spa_prop_add_list(*nvp, prop, strval, 0, src);
382			kmem_free(strval, za.za_num_integers);
383			break;
384
385		default:
386			break;
387		}
388	}
389	zap_cursor_fini(&zc);
390	mutex_exit(&spa->spa_props_lock);
391out:
392	if (err && err != ENOENT) {
393		nvlist_free(*nvp);
394		*nvp = NULL;
395		return (err);
396	}
397
398	return (0);
399}
400
401/*
402 * Validate the given pool properties nvlist and modify the list
403 * for the property values to be set.
404 */
405static int
406spa_prop_validate(spa_t *spa, nvlist_t *props)
407{
408	nvpair_t *elem;
409	int error = 0, reset_bootfs = 0;
410	uint64_t objnum = 0;
411	boolean_t has_feature = B_FALSE;
412
413	elem = NULL;
414	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
415		uint64_t intval;
416		char *strval, *slash, *check, *fname;
417		const char *propname = nvpair_name(elem);
418		zpool_prop_t prop = zpool_name_to_prop(propname);
419
420		switch (prop) {
421		case ZPROP_INVAL:
422			if (!zpool_prop_feature(propname)) {
423				error = SET_ERROR(EINVAL);
424				break;
425			}
426
427			/*
428			 * Sanitize the input.
429			 */
430			if (nvpair_type(elem) != DATA_TYPE_UINT64) {
431				error = SET_ERROR(EINVAL);
432				break;
433			}
434
435			if (nvpair_value_uint64(elem, &intval) != 0) {
436				error = SET_ERROR(EINVAL);
437				break;
438			}
439
440			if (intval != 0) {
441				error = SET_ERROR(EINVAL);
442				break;
443			}
444
445			fname = strchr(propname, '@') + 1;
446			if (zfeature_lookup_name(fname, NULL) != 0) {
447				error = SET_ERROR(EINVAL);
448				break;
449			}
450
451			has_feature = B_TRUE;
452			break;
453
454		case ZPOOL_PROP_VERSION:
455			error = nvpair_value_uint64(elem, &intval);
456			if (!error &&
457			    (intval < spa_version(spa) ||
458			    intval > SPA_VERSION_BEFORE_FEATURES ||
459			    has_feature))
460				error = SET_ERROR(EINVAL);
461			break;
462
463		case ZPOOL_PROP_DELEGATION:
464		case ZPOOL_PROP_AUTOREPLACE:
465		case ZPOOL_PROP_LISTSNAPS:
466		case ZPOOL_PROP_AUTOEXPAND:
467			error = nvpair_value_uint64(elem, &intval);
468			if (!error && intval > 1)
469				error = SET_ERROR(EINVAL);
470			break;
471
472		case ZPOOL_PROP_BOOTFS:
473			/*
474			 * If the pool version is less than SPA_VERSION_BOOTFS,
475			 * or the pool is still being created (version == 0),
476			 * the bootfs property cannot be set.
477			 */
478			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
479				error = SET_ERROR(ENOTSUP);
480				break;
481			}
482
483			/*
484			 * Make sure the vdev config is bootable
485			 */
486			if (!vdev_is_bootable(spa->spa_root_vdev)) {
487				error = SET_ERROR(ENOTSUP);
488				break;
489			}
490
491			reset_bootfs = 1;
492
493			error = nvpair_value_string(elem, &strval);
494
495			if (!error) {
496				objset_t *os;
497				uint64_t propval;
498
499				if (strval == NULL || strval[0] == '\0') {
500					objnum = zpool_prop_default_numeric(
501					    ZPOOL_PROP_BOOTFS);
502					break;
503				}
504
505				if (error = dmu_objset_hold(strval, FTAG, &os))
506					break;
507
508				/*
509				 * Must be ZPL, and its property settings
510				 * must be supported by GRUB (compression
511				 * is not gzip, and large blocks are not used).
512				 */
513
514				if (dmu_objset_type(os) != DMU_OST_ZFS) {
515					error = SET_ERROR(ENOTSUP);
516				} else if ((error =
517				    dsl_prop_get_int_ds(dmu_objset_ds(os),
518				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
519				    &propval)) == 0 &&
520				    !BOOTFS_COMPRESS_VALID(propval)) {
521					error = SET_ERROR(ENOTSUP);
522				} else if ((error =
523				    dsl_prop_get_int_ds(dmu_objset_ds(os),
524				    zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
525				    &propval)) == 0 &&
526				    propval > SPA_OLD_MAXBLOCKSIZE) {
527					error = SET_ERROR(ENOTSUP);
528				} else {
529					objnum = dmu_objset_id(os);
530				}
531				dmu_objset_rele(os, FTAG);
532			}
533			break;
534
535		case ZPOOL_PROP_FAILUREMODE:
536			error = nvpair_value_uint64(elem, &intval);
537			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
538			    intval > ZIO_FAILURE_MODE_PANIC))
539				error = SET_ERROR(EINVAL);
540
541			/*
542			 * This is a special case which only occurs when
543			 * the pool has completely failed. This allows
544			 * the user to change the in-core failmode property
545			 * without syncing it out to disk (I/Os might
546			 * currently be blocked). We do this by returning
547			 * EIO to the caller (spa_prop_set) to trick it
548			 * into thinking we encountered a property validation
549			 * error.
550			 */
551			if (!error && spa_suspended(spa)) {
552				spa->spa_failmode = intval;
553				error = SET_ERROR(EIO);
554			}
555			break;
556
557		case ZPOOL_PROP_CACHEFILE:
558			if ((error = nvpair_value_string(elem, &strval)) != 0)
559				break;
560
561			if (strval[0] == '\0')
562				break;
563
564			if (strcmp(strval, "none") == 0)
565				break;
566
567			if (strval[0] != '/') {
568				error = SET_ERROR(EINVAL);
569				break;
570			}
571
572			slash = strrchr(strval, '/');
573			ASSERT(slash != NULL);
574
575			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
576			    strcmp(slash, "/..") == 0)
577				error = SET_ERROR(EINVAL);
578			break;
579
580		case ZPOOL_PROP_COMMENT:
581			if ((error = nvpair_value_string(elem, &strval)) != 0)
582				break;
583			for (check = strval; *check != '\0'; check++) {
584				/*
585				 * The kernel doesn't have an easy isprint()
586				 * check.  For this kernel check, we merely
587				 * check ASCII apart from DEL.  Fix this if
588				 * there is an easy-to-use kernel isprint().
589				 */
590				if (*check >= 0x7f) {
591					error = SET_ERROR(EINVAL);
592					break;
593				}
594			}
595			if (strlen(strval) > ZPROP_MAX_COMMENT)
596				error = E2BIG;
597			break;
598
599		case ZPOOL_PROP_DEDUPDITTO:
600			if (spa_version(spa) < SPA_VERSION_DEDUP)
601				error = SET_ERROR(ENOTSUP);
602			else
603				error = nvpair_value_uint64(elem, &intval);
604			if (error == 0 &&
605			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
606				error = SET_ERROR(EINVAL);
607			break;
608		}
609
610		if (error)
611			break;
612	}
613
614	if (!error && reset_bootfs) {
615		error = nvlist_remove(props,
616		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
617
618		if (!error) {
619			error = nvlist_add_uint64(props,
620			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
621		}
622	}
623
624	return (error);
625}
626
627void
628spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
629{
630	char *cachefile;
631	spa_config_dirent_t *dp;
632
633	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
634	    &cachefile) != 0)
635		return;
636
637	dp = kmem_alloc(sizeof (spa_config_dirent_t),
638	    KM_SLEEP);
639
640	if (cachefile[0] == '\0')
641		dp->scd_path = spa_strdup(spa_config_path);
642	else if (strcmp(cachefile, "none") == 0)
643		dp->scd_path = NULL;
644	else
645		dp->scd_path = spa_strdup(cachefile);
646
647	list_insert_head(&spa->spa_config_list, dp);
648	if (need_sync)
649		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
650}
651
652int
653spa_prop_set(spa_t *spa, nvlist_t *nvp)
654{
655	int error;
656	nvpair_t *elem = NULL;
657	boolean_t need_sync = B_FALSE;
658
659	if ((error = spa_prop_validate(spa, nvp)) != 0)
660		return (error);
661
662	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
663		zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
664
665		if (prop == ZPOOL_PROP_CACHEFILE ||
666		    prop == ZPOOL_PROP_ALTROOT ||
667		    prop == ZPOOL_PROP_READONLY)
668			continue;
669
670		if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
671			uint64_t ver;
672
673			if (prop == ZPOOL_PROP_VERSION) {
674				VERIFY(nvpair_value_uint64(elem, &ver) == 0);
675			} else {
676				ASSERT(zpool_prop_feature(nvpair_name(elem)));
677				ver = SPA_VERSION_FEATURES;
678				need_sync = B_TRUE;
679			}
680
681			/* Save time if the version is already set. */
682			if (ver == spa_version(spa))
683				continue;
684
685			/*
686			 * In addition to the pool directory object, we might
687			 * create the pool properties object, the features for
688			 * read object, the features for write object, or the
689			 * feature descriptions object.
690			 */
691			error = dsl_sync_task(spa->spa_name, NULL,
692			    spa_sync_version, &ver,
693			    6, ZFS_SPACE_CHECK_RESERVED);
694			if (error)
695				return (error);
696			continue;
697		}
698
699		need_sync = B_TRUE;
700		break;
701	}
702
703	if (need_sync) {
704		return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
705		    nvp, 6, ZFS_SPACE_CHECK_RESERVED));
706	}
707
708	return (0);
709}
710
711/*
712 * If the bootfs property value is dsobj, clear it.
713 */
714void
715spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
716{
717	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
718		VERIFY(zap_remove(spa->spa_meta_objset,
719		    spa->spa_pool_props_object,
720		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
721		spa->spa_bootfs = 0;
722	}
723}
724
725/*ARGSUSED*/
726static int
727spa_change_guid_check(void *arg, dmu_tx_t *tx)
728{
729	uint64_t *newguid = arg;
730	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
731	vdev_t *rvd = spa->spa_root_vdev;
732	uint64_t vdev_state;
733
734	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
735	vdev_state = rvd->vdev_state;
736	spa_config_exit(spa, SCL_STATE, FTAG);
737
738	if (vdev_state != VDEV_STATE_HEALTHY)
739		return (SET_ERROR(ENXIO));
740
741	ASSERT3U(spa_guid(spa), !=, *newguid);
742
743	return (0);
744}
745
746static void
747spa_change_guid_sync(void *arg, dmu_tx_t *tx)
748{
749	uint64_t *newguid = arg;
750	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
751	uint64_t oldguid;
752	vdev_t *rvd = spa->spa_root_vdev;
753
754	oldguid = spa_guid(spa);
755
756	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
757	rvd->vdev_guid = *newguid;
758	rvd->vdev_guid_sum += (*newguid - oldguid);
759	vdev_config_dirty(rvd);
760	spa_config_exit(spa, SCL_STATE, FTAG);
761
762	spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
763	    oldguid, *newguid);
764}
765
766/*
767 * Change the GUID for the pool.  This is done so that we can later
768 * re-import a pool built from a clone of our own vdevs.  We will modify
769 * the root vdev's guid, our own pool guid, and then mark all of our
770 * vdevs dirty.  Note that we must make sure that all our vdevs are
771 * online when we do this, or else any vdevs that weren't present
772 * would be orphaned from our pool.  We are also going to issue a
773 * sysevent to update any watchers.
774 */
775int
776spa_change_guid(spa_t *spa)
777{
778	int error;
779	uint64_t guid;
780
781	mutex_enter(&spa->spa_vdev_top_lock);
782	mutex_enter(&spa_namespace_lock);
783	guid = spa_generate_guid(NULL);
784
785	error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
786	    spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
787
788	if (error == 0) {
789		spa_config_sync(spa, B_FALSE, B_TRUE);
790		spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
791	}
792
793	mutex_exit(&spa_namespace_lock);
794	mutex_exit(&spa->spa_vdev_top_lock);
795
796	return (error);
797}
798
799/*
800 * ==========================================================================
801 * SPA state manipulation (open/create/destroy/import/export)
802 * ==========================================================================
803 */
804
805static int
806spa_error_entry_compare(const void *a, const void *b)
807{
808	spa_error_entry_t *sa = (spa_error_entry_t *)a;
809	spa_error_entry_t *sb = (spa_error_entry_t *)b;
810	int ret;
811
812	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
813	    sizeof (zbookmark_phys_t));
814
815	if (ret < 0)
816		return (-1);
817	else if (ret > 0)
818		return (1);
819	else
820		return (0);
821}
822
823/*
824 * Utility function which retrieves copies of the current logs and
825 * re-initializes them in the process.
826 */
827void
828spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
829{
830	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
831
832	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
833	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
834
835	avl_create(&spa->spa_errlist_scrub,
836	    spa_error_entry_compare, sizeof (spa_error_entry_t),
837	    offsetof(spa_error_entry_t, se_avl));
838	avl_create(&spa->spa_errlist_last,
839	    spa_error_entry_compare, sizeof (spa_error_entry_t),
840	    offsetof(spa_error_entry_t, se_avl));
841}
842
843static void
844spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
845{
846	const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
847	enum zti_modes mode = ztip->zti_mode;
848	uint_t value = ztip->zti_value;
849	uint_t count = ztip->zti_count;
850	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
851	char name[32];
852	uint_t flags = 0;
853	boolean_t batch = B_FALSE;
854
855	if (mode == ZTI_MODE_NULL) {
856		tqs->stqs_count = 0;
857		tqs->stqs_taskq = NULL;
858		return;
859	}
860
861	ASSERT3U(count, >, 0);
862
863	tqs->stqs_count = count;
864	tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
865
866	switch (mode) {
867	case ZTI_MODE_FIXED:
868		ASSERT3U(value, >=, 1);
869		value = MAX(value, 1);
870		break;
871
872	case ZTI_MODE_BATCH:
873		batch = B_TRUE;
874		flags |= TASKQ_THREADS_CPU_PCT;
875		value = zio_taskq_batch_pct;
876		break;
877
878	default:
879		panic("unrecognized mode for %s_%s taskq (%u:%u) in "
880		    "spa_activate()",
881		    zio_type_name[t], zio_taskq_types[q], mode, value);
882		break;
883	}
884
885	for (uint_t i = 0; i < count; i++) {
886		taskq_t *tq;
887
888		if (count > 1) {
889			(void) snprintf(name, sizeof (name), "%s_%s_%u",
890			    zio_type_name[t], zio_taskq_types[q], i);
891		} else {
892			(void) snprintf(name, sizeof (name), "%s_%s",
893			    zio_type_name[t], zio_taskq_types[q]);
894		}
895
896		if (zio_taskq_sysdc && spa->spa_proc != &p0) {
897			if (batch)
898				flags |= TASKQ_DC_BATCH;
899
900			tq = taskq_create_sysdc(name, value, 50, INT_MAX,
901			    spa->spa_proc, zio_taskq_basedc, flags);
902		} else {
903			pri_t pri = maxclsyspri;
904			/*
905			 * The write issue taskq can be extremely CPU
906			 * intensive.  Run it at slightly lower priority
907			 * than the other taskqs.
908			 */
909			if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
910				pri--;
911
912			tq = taskq_create_proc(name, value, pri, 50,
913			    INT_MAX, spa->spa_proc, flags);
914		}
915
916		tqs->stqs_taskq[i] = tq;
917	}
918}
919
920static void
921spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
922{
923	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
924
925	if (tqs->stqs_taskq == NULL) {
926		ASSERT0(tqs->stqs_count);
927		return;
928	}
929
930	for (uint_t i = 0; i < tqs->stqs_count; i++) {
931		ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
932		taskq_destroy(tqs->stqs_taskq[i]);
933	}
934
935	kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
936	tqs->stqs_taskq = NULL;
937}
938
939/*
940 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
941 * Note that a type may have multiple discrete taskqs to avoid lock contention
942 * on the taskq itself. In that case we choose which taskq at random by using
943 * the low bits of gethrtime().
944 */
945void
946spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
947    task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
948{
949	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
950	taskq_t *tq;
951
952	ASSERT3P(tqs->stqs_taskq, !=, NULL);
953	ASSERT3U(tqs->stqs_count, !=, 0);
954
955	if (tqs->stqs_count == 1) {
956		tq = tqs->stqs_taskq[0];
957	} else {
958		tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
959	}
960
961	taskq_dispatch_ent(tq, func, arg, flags, ent);
962}
963
964static void
965spa_create_zio_taskqs(spa_t *spa)
966{
967	for (int t = 0; t < ZIO_TYPES; t++) {
968		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
969			spa_taskqs_init(spa, t, q);
970		}
971	}
972}
973
974#ifdef _KERNEL
975static void
976spa_thread(void *arg)
977{
978	callb_cpr_t cprinfo;
979
980	spa_t *spa = arg;
981	user_t *pu = PTOU(curproc);
982
983	CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
984	    spa->spa_name);
985
986	ASSERT(curproc != &p0);
987	(void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
988	    "zpool-%s", spa->spa_name);
989	(void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
990
991	/* bind this thread to the requested psrset */
992	if (zio_taskq_psrset_bind != PS_NONE) {
993		pool_lock();
994		mutex_enter(&cpu_lock);
995		mutex_enter(&pidlock);
996		mutex_enter(&curproc->p_lock);
997
998		if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
999		    0, NULL, NULL) == 0)  {
1000			curthread->t_bind_pset = zio_taskq_psrset_bind;
1001		} else {
1002			cmn_err(CE_WARN,
1003			    "Couldn't bind process for zfs pool \"%s\" to "
1004			    "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1005		}
1006
1007		mutex_exit(&curproc->p_lock);
1008		mutex_exit(&pidlock);
1009		mutex_exit(&cpu_lock);
1010		pool_unlock();
1011	}
1012
1013	if (zio_taskq_sysdc) {
1014		sysdc_thread_enter(curthread, 100, 0);
1015	}
1016
1017	spa->spa_proc = curproc;
1018	spa->spa_did = curthread->t_did;
1019
1020	spa_create_zio_taskqs(spa);
1021
1022	mutex_enter(&spa->spa_proc_lock);
1023	ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1024
1025	spa->spa_proc_state = SPA_PROC_ACTIVE;
1026	cv_broadcast(&spa->spa_proc_cv);
1027
1028	CALLB_CPR_SAFE_BEGIN(&cprinfo);
1029	while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1030		cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1031	CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1032
1033	ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1034	spa->spa_proc_state = SPA_PROC_GONE;
1035	spa->spa_proc = &p0;
1036	cv_broadcast(&spa->spa_proc_cv);
1037	CALLB_CPR_EXIT(&cprinfo);	/* drops spa_proc_lock */
1038
1039	mutex_enter(&curproc->p_lock);
1040	lwp_exit();
1041}
1042#endif
1043
1044/*
1045 * Activate an uninitialized pool.
1046 */
1047static void
1048spa_activate(spa_t *spa, int mode)
1049{
1050	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1051
1052	spa->spa_state = POOL_STATE_ACTIVE;
1053	spa->spa_mode = mode;
1054
1055	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1056	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1057
1058	/* Try to create a covering process */
1059	mutex_enter(&spa->spa_proc_lock);
1060	ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1061	ASSERT(spa->spa_proc == &p0);
1062	spa->spa_did = 0;
1063
1064	/* Only create a process if we're going to be around a while. */
1065	if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1066		if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1067		    NULL, 0) == 0) {
1068			spa->spa_proc_state = SPA_PROC_CREATED;
1069			while (spa->spa_proc_state == SPA_PROC_CREATED) {
1070				cv_wait(&spa->spa_proc_cv,
1071				    &spa->spa_proc_lock);
1072			}
1073			ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1074			ASSERT(spa->spa_proc != &p0);
1075			ASSERT(spa->spa_did != 0);
1076		} else {
1077#ifdef _KERNEL
1078			cmn_err(CE_WARN,
1079			    "Couldn't create process for zfs pool \"%s\"\n",
1080			    spa->spa_name);
1081#endif
1082		}
1083	}
1084	mutex_exit(&spa->spa_proc_lock);
1085
1086	/* If we didn't create a process, we need to create our taskqs. */
1087	if (spa->spa_proc == &p0) {
1088		spa_create_zio_taskqs(spa);
1089	}
1090
1091	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1092	    offsetof(vdev_t, vdev_config_dirty_node));
1093	list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1094	    offsetof(objset_t, os_evicting_node));
1095	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1096	    offsetof(vdev_t, vdev_state_dirty_node));
1097
1098	txg_list_create(&spa->spa_vdev_txg_list,
1099	    offsetof(struct vdev, vdev_txg_node));
1100
1101	avl_create(&spa->spa_errlist_scrub,
1102	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1103	    offsetof(spa_error_entry_t, se_avl));
1104	avl_create(&spa->spa_errlist_last,
1105	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1106	    offsetof(spa_error_entry_t, se_avl));
1107}
1108
1109/*
1110 * Opposite of spa_activate().
1111 */
1112static void
1113spa_deactivate(spa_t *spa)
1114{
1115	ASSERT(spa->spa_sync_on == B_FALSE);
1116	ASSERT(spa->spa_dsl_pool == NULL);
1117	ASSERT(spa->spa_root_vdev == NULL);
1118	ASSERT(spa->spa_async_zio_root == NULL);
1119	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1120
1121	spa_evicting_os_wait(spa);
1122
1123	txg_list_destroy(&spa->spa_vdev_txg_list);
1124
1125	list_destroy(&spa->spa_config_dirty_list);
1126	list_destroy(&spa->spa_evicting_os_list);
1127	list_destroy(&spa->spa_state_dirty_list);
1128
1129	for (int t = 0; t < ZIO_TYPES; t++) {
1130		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1131			spa_taskqs_fini(spa, t, q);
1132		}
1133	}
1134
1135	metaslab_class_destroy(spa->spa_normal_class);
1136	spa->spa_normal_class = NULL;
1137
1138	metaslab_class_destroy(spa->spa_log_class);
1139	spa->spa_log_class = NULL;
1140
1141	/*
1142	 * If this was part of an import or the open otherwise failed, we may
1143	 * still have errors left in the queues.  Empty them just in case.
1144	 */
1145	spa_errlog_drain(spa);
1146
1147	avl_destroy(&spa->spa_errlist_scrub);
1148	avl_destroy(&spa->spa_errlist_last);
1149
1150	spa->spa_state = POOL_STATE_UNINITIALIZED;
1151
1152	mutex_enter(&spa->spa_proc_lock);
1153	if (spa->spa_proc_state != SPA_PROC_NONE) {
1154		ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1155		spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1156		cv_broadcast(&spa->spa_proc_cv);
1157		while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1158			ASSERT(spa->spa_proc != &p0);
1159			cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1160		}
1161		ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1162		spa->spa_proc_state = SPA_PROC_NONE;
1163	}
1164	ASSERT(spa->spa_proc == &p0);
1165	mutex_exit(&spa->spa_proc_lock);
1166
1167	/*
1168	 * We want to make sure spa_thread() has actually exited the ZFS
1169	 * module, so that the module can't be unloaded out from underneath
1170	 * it.
1171	 */
1172	if (spa->spa_did != 0) {
1173		thread_join(spa->spa_did);
1174		spa->spa_did = 0;
1175	}
1176}
1177
1178/*
1179 * Verify a pool configuration, and construct the vdev tree appropriately.  This
1180 * will create all the necessary vdevs in the appropriate layout, with each vdev
1181 * in the CLOSED state.  This will prep the pool before open/creation/import.
1182 * All vdev validation is done by the vdev_alloc() routine.
1183 */
1184static int
1185spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1186    uint_t id, int atype)
1187{
1188	nvlist_t **child;
1189	uint_t children;
1190	int error;
1191
1192	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1193		return (error);
1194
1195	if ((*vdp)->vdev_ops->vdev_op_leaf)
1196		return (0);
1197
1198	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1199	    &child, &children);
1200
1201	if (error == ENOENT)
1202		return (0);
1203
1204	if (error) {
1205		vdev_free(*vdp);
1206		*vdp = NULL;
1207		return (SET_ERROR(EINVAL));
1208	}
1209
1210	for (int c = 0; c < children; c++) {
1211		vdev_t *vd;
1212		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1213		    atype)) != 0) {
1214			vdev_free(*vdp);
1215			*vdp = NULL;
1216			return (error);
1217		}
1218	}
1219
1220	ASSERT(*vdp != NULL);
1221
1222	return (0);
1223}
1224
1225/*
1226 * Opposite of spa_load().
1227 */
1228static void
1229spa_unload(spa_t *spa)
1230{
1231	int i;
1232
1233	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1234
1235	/*
1236	 * Stop async tasks.
1237	 */
1238	spa_async_suspend(spa);
1239
1240	/*
1241	 * Stop syncing.
1242	 */
1243	if (spa->spa_sync_on) {
1244		txg_sync_stop(spa->spa_dsl_pool);
1245		spa->spa_sync_on = B_FALSE;
1246	}
1247
1248	/*
1249	 * Wait for any outstanding async I/O to complete.
1250	 */
1251	if (spa->spa_async_zio_root != NULL) {
1252		for (int i = 0; i < max_ncpus; i++)
1253			(void) zio_wait(spa->spa_async_zio_root[i]);
1254		kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1255		spa->spa_async_zio_root = NULL;
1256	}
1257
1258	bpobj_close(&spa->spa_deferred_bpobj);
1259
1260	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1261
1262	/*
1263	 * Close all vdevs.
1264	 */
1265	if (spa->spa_root_vdev)
1266		vdev_free(spa->spa_root_vdev);
1267	ASSERT(spa->spa_root_vdev == NULL);
1268
1269	/*
1270	 * Close the dsl pool.
1271	 */
1272	if (spa->spa_dsl_pool) {
1273		dsl_pool_close(spa->spa_dsl_pool);
1274		spa->spa_dsl_pool = NULL;
1275		spa->spa_meta_objset = NULL;
1276	}
1277
1278	ddt_unload(spa);
1279
1280
1281	/*
1282	 * Drop and purge level 2 cache
1283	 */
1284	spa_l2cache_drop(spa);
1285
1286	for (i = 0; i < spa->spa_spares.sav_count; i++)
1287		vdev_free(spa->spa_spares.sav_vdevs[i]);
1288	if (spa->spa_spares.sav_vdevs) {
1289		kmem_free(spa->spa_spares.sav_vdevs,
1290		    spa->spa_spares.sav_count * sizeof (void *));
1291		spa->spa_spares.sav_vdevs = NULL;
1292	}
1293	if (spa->spa_spares.sav_config) {
1294		nvlist_free(spa->spa_spares.sav_config);
1295		spa->spa_spares.sav_config = NULL;
1296	}
1297	spa->spa_spares.sav_count = 0;
1298
1299	for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1300		vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1301		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1302	}
1303	if (spa->spa_l2cache.sav_vdevs) {
1304		kmem_free(spa->spa_l2cache.sav_vdevs,
1305		    spa->spa_l2cache.sav_count * sizeof (void *));
1306		spa->spa_l2cache.sav_vdevs = NULL;
1307	}
1308	if (spa->spa_l2cache.sav_config) {
1309		nvlist_free(spa->spa_l2cache.sav_config);
1310		spa->spa_l2cache.sav_config = NULL;
1311	}
1312	spa->spa_l2cache.sav_count = 0;
1313
1314	spa->spa_async_suspended = 0;
1315
1316	if (spa->spa_comment != NULL) {
1317		spa_strfree(spa->spa_comment);
1318		spa->spa_comment = NULL;
1319	}
1320
1321	spa_config_exit(spa, SCL_ALL, FTAG);
1322}
1323
1324/*
1325 * Load (or re-load) the current list of vdevs describing the active spares for
1326 * this pool.  When this is called, we have some form of basic information in
1327 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1328 * then re-generate a more complete list including status information.
1329 */
1330static void
1331spa_load_spares(spa_t *spa)
1332{
1333	nvlist_t **spares;
1334	uint_t nspares;
1335	int i;
1336	vdev_t *vd, *tvd;
1337
1338	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1339
1340	/*
1341	 * First, close and free any existing spare vdevs.
1342	 */
1343	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1344		vd = spa->spa_spares.sav_vdevs[i];
1345
1346		/* Undo the call to spa_activate() below */
1347		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1348		    B_FALSE)) != NULL && tvd->vdev_isspare)
1349			spa_spare_remove(tvd);
1350		vdev_close(vd);
1351		vdev_free(vd);
1352	}
1353
1354	if (spa->spa_spares.sav_vdevs)
1355		kmem_free(spa->spa_spares.sav_vdevs,
1356		    spa->spa_spares.sav_count * sizeof (void *));
1357
1358	if (spa->spa_spares.sav_config == NULL)
1359		nspares = 0;
1360	else
1361		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1362		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1363
1364	spa->spa_spares.sav_count = (int)nspares;
1365	spa->spa_spares.sav_vdevs = NULL;
1366
1367	if (nspares == 0)
1368		return;
1369
1370	/*
1371	 * Construct the array of vdevs, opening them to get status in the
1372	 * process.   For each spare, there is potentially two different vdev_t
1373	 * structures associated with it: one in the list of spares (used only
1374	 * for basic validation purposes) and one in the active vdev
1375	 * configuration (if it's spared in).  During this phase we open and
1376	 * validate each vdev on the spare list.  If the vdev also exists in the
1377	 * active configuration, then we also mark this vdev as an active spare.
1378	 */
1379	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1380	    KM_SLEEP);
1381	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1382		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1383		    VDEV_ALLOC_SPARE) == 0);
1384		ASSERT(vd != NULL);
1385
1386		spa->spa_spares.sav_vdevs[i] = vd;
1387
1388		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1389		    B_FALSE)) != NULL) {
1390			if (!tvd->vdev_isspare)
1391				spa_spare_add(tvd);
1392
1393			/*
1394			 * We only mark the spare active if we were successfully
1395			 * able to load the vdev.  Otherwise, importing a pool
1396			 * with a bad active spare would result in strange
1397			 * behavior, because multiple pool would think the spare
1398			 * is actively in use.
1399			 *
1400			 * There is a vulnerability here to an equally bizarre
1401			 * circumstance, where a dead active spare is later
1402			 * brought back to life (onlined or otherwise).  Given
1403			 * the rarity of this scenario, and the extra complexity
1404			 * it adds, we ignore the possibility.
1405			 */
1406			if (!vdev_is_dead(tvd))
1407				spa_spare_activate(tvd);
1408		}
1409
1410		vd->vdev_top = vd;
1411		vd->vdev_aux = &spa->spa_spares;
1412
1413		if (vdev_open(vd) != 0)
1414			continue;
1415
1416		if (vdev_validate_aux(vd) == 0)
1417			spa_spare_add(vd);
1418	}
1419
1420	/*
1421	 * Recompute the stashed list of spares, with status information
1422	 * this time.
1423	 */
1424	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1425	    DATA_TYPE_NVLIST_ARRAY) == 0);
1426
1427	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1428	    KM_SLEEP);
1429	for (i = 0; i < spa->spa_spares.sav_count; i++)
1430		spares[i] = vdev_config_generate(spa,
1431		    spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1432	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1433	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1434	for (i = 0; i < spa->spa_spares.sav_count; i++)
1435		nvlist_free(spares[i]);
1436	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1437}
1438
1439/*
1440 * Load (or re-load) the current list of vdevs describing the active l2cache for
1441 * this pool.  When this is called, we have some form of basic information in
1442 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1443 * then re-generate a more complete list including status information.
1444 * Devices which are already active have their details maintained, and are
1445 * not re-opened.
1446 */
1447static void
1448spa_load_l2cache(spa_t *spa)
1449{
1450	nvlist_t **l2cache;
1451	uint_t nl2cache;
1452	int i, j, oldnvdevs;
1453	uint64_t guid;
1454	vdev_t *vd, **oldvdevs, **newvdevs;
1455	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1456
1457	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1458
1459	if (sav->sav_config != NULL) {
1460		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1461		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1462		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1463	} else {
1464		nl2cache = 0;
1465		newvdevs = NULL;
1466	}
1467
1468	oldvdevs = sav->sav_vdevs;
1469	oldnvdevs = sav->sav_count;
1470	sav->sav_vdevs = NULL;
1471	sav->sav_count = 0;
1472
1473	/*
1474	 * Process new nvlist of vdevs.
1475	 */
1476	for (i = 0; i < nl2cache; i++) {
1477		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1478		    &guid) == 0);
1479
1480		newvdevs[i] = NULL;
1481		for (j = 0; j < oldnvdevs; j++) {
1482			vd = oldvdevs[j];
1483			if (vd != NULL && guid == vd->vdev_guid) {
1484				/*
1485				 * Retain previous vdev for add/remove ops.
1486				 */
1487				newvdevs[i] = vd;
1488				oldvdevs[j] = NULL;
1489				break;
1490			}
1491		}
1492
1493		if (newvdevs[i] == NULL) {
1494			/*
1495			 * Create new vdev
1496			 */
1497			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1498			    VDEV_ALLOC_L2CACHE) == 0);
1499			ASSERT(vd != NULL);
1500			newvdevs[i] = vd;
1501
1502			/*
1503			 * Commit this vdev as an l2cache device,
1504			 * even if it fails to open.
1505			 */
1506			spa_l2cache_add(vd);
1507
1508			vd->vdev_top = vd;
1509			vd->vdev_aux = sav;
1510
1511			spa_l2cache_activate(vd);
1512
1513			if (vdev_open(vd) != 0)
1514				continue;
1515
1516			(void) vdev_validate_aux(vd);
1517
1518			if (!vdev_is_dead(vd))
1519				l2arc_add_vdev(spa, vd);
1520		}
1521	}
1522
1523	/*
1524	 * Purge vdevs that were dropped
1525	 */
1526	for (i = 0; i < oldnvdevs; i++) {
1527		uint64_t pool;
1528
1529		vd = oldvdevs[i];
1530		if (vd != NULL) {
1531			ASSERT(vd->vdev_isl2cache);
1532
1533			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1534			    pool != 0ULL && l2arc_vdev_present(vd))
1535				l2arc_remove_vdev(vd);
1536			vdev_clear_stats(vd);
1537			vdev_free(vd);
1538		}
1539	}
1540
1541	if (oldvdevs)
1542		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1543
1544	if (sav->sav_config == NULL)
1545		goto out;
1546
1547	sav->sav_vdevs = newvdevs;
1548	sav->sav_count = (int)nl2cache;
1549
1550	/*
1551	 * Recompute the stashed list of l2cache devices, with status
1552	 * information this time.
1553	 */
1554	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1555	    DATA_TYPE_NVLIST_ARRAY) == 0);
1556
1557	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1558	for (i = 0; i < sav->sav_count; i++)
1559		l2cache[i] = vdev_config_generate(spa,
1560		    sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1561	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1562	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1563out:
1564	for (i = 0; i < sav->sav_count; i++)
1565		nvlist_free(l2cache[i]);
1566	if (sav->sav_count)
1567		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1568}
1569
1570static int
1571load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1572{
1573	dmu_buf_t *db;
1574	char *packed = NULL;
1575	size_t nvsize = 0;
1576	int error;
1577	*value = NULL;
1578
1579	error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1580	if (error != 0)
1581		return (error);
1582
1583	nvsize = *(uint64_t *)db->db_data;
1584	dmu_buf_rele(db, FTAG);
1585
1586	packed = kmem_alloc(nvsize, KM_SLEEP);
1587	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1588	    DMU_READ_PREFETCH);
1589	if (error == 0)
1590		error = nvlist_unpack(packed, nvsize, value, 0);
1591	kmem_free(packed, nvsize);
1592
1593	return (error);
1594}
1595
1596/*
1597 * Checks to see if the given vdev could not be opened, in which case we post a
1598 * sysevent to notify the autoreplace code that the device has been removed.
1599 */
1600static void
1601spa_check_removed(vdev_t *vd)
1602{
1603	for (int c = 0; c < vd->vdev_children; c++)
1604		spa_check_removed(vd->vdev_child[c]);
1605
1606	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1607	    !vd->vdev_ishole) {
1608		zfs_post_autoreplace(vd->vdev_spa, vd);
1609		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1610	}
1611}
1612
1613static void
1614spa_config_valid_zaps(vdev_t *vd, vdev_t *mvd)
1615{
1616	ASSERT3U(vd->vdev_children, ==, mvd->vdev_children);
1617
1618	vd->vdev_top_zap = mvd->vdev_top_zap;
1619	vd->vdev_leaf_zap = mvd->vdev_leaf_zap;
1620
1621	for (uint64_t i = 0; i < vd->vdev_children; i++) {
1622		spa_config_valid_zaps(vd->vdev_child[i], mvd->vdev_child[i]);
1623	}
1624}
1625
1626/*
1627 * Validate the current config against the MOS config
1628 */
1629static boolean_t
1630spa_config_valid(spa_t *spa, nvlist_t *config)
1631{
1632	vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1633	nvlist_t *nv;
1634
1635	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1636
1637	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1638	VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1639
1640	ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1641
1642	/*
1643	 * If we're doing a normal import, then build up any additional
1644	 * diagnostic information about missing devices in this config.
1645	 * We'll pass this up to the user for further processing.
1646	 */
1647	if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1648		nvlist_t **child, *nv;
1649		uint64_t idx = 0;
1650
1651		child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1652		    KM_SLEEP);
1653		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1654
1655		for (int c = 0; c < rvd->vdev_children; c++) {
1656			vdev_t *tvd = rvd->vdev_child[c];
1657			vdev_t *mtvd  = mrvd->vdev_child[c];
1658
1659			if (tvd->vdev_ops == &vdev_missing_ops &&
1660			    mtvd->vdev_ops != &vdev_missing_ops &&
1661			    mtvd->vdev_islog)
1662				child[idx++] = vdev_config_generate(spa, mtvd,
1663				    B_FALSE, 0);
1664		}
1665
1666		if (idx) {
1667			VERIFY(nvlist_add_nvlist_array(nv,
1668			    ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1669			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1670			    ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1671
1672			for (int i = 0; i < idx; i++)
1673				nvlist_free(child[i]);
1674		}
1675		nvlist_free(nv);
1676		kmem_free(child, rvd->vdev_children * sizeof (char **));
1677	}
1678
1679	/*
1680	 * Compare the root vdev tree with the information we have
1681	 * from the MOS config (mrvd). Check each top-level vdev
1682	 * with the corresponding MOS config top-level (mtvd).
1683	 */
1684	for (int c = 0; c < rvd->vdev_children; c++) {
1685		vdev_t *tvd = rvd->vdev_child[c];
1686		vdev_t *mtvd  = mrvd->vdev_child[c];
1687
1688		/*
1689		 * Resolve any "missing" vdevs in the current configuration.
1690		 * If we find that the MOS config has more accurate information
1691		 * about the top-level vdev then use that vdev instead.
1692		 */
1693		if (tvd->vdev_ops == &vdev_missing_ops &&
1694		    mtvd->vdev_ops != &vdev_missing_ops) {
1695
1696			if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1697				continue;
1698
1699			/*
1700			 * Device specific actions.
1701			 */
1702			if (mtvd->vdev_islog) {
1703				spa_set_log_state(spa, SPA_LOG_CLEAR);
1704			} else {
1705				/*
1706				 * XXX - once we have 'readonly' pool
1707				 * support we should be able to handle
1708				 * missing data devices by transitioning
1709				 * the pool to readonly.
1710				 */
1711				continue;
1712			}
1713
1714			/*
1715			 * Swap the missing vdev with the data we were
1716			 * able to obtain from the MOS config.
1717			 */
1718			vdev_remove_child(rvd, tvd);
1719			vdev_remove_child(mrvd, mtvd);
1720
1721			vdev_add_child(rvd, mtvd);
1722			vdev_add_child(mrvd, tvd);
1723
1724			spa_config_exit(spa, SCL_ALL, FTAG);
1725			vdev_load(mtvd);
1726			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1727
1728			vdev_reopen(rvd);
1729		} else {
1730			if (mtvd->vdev_islog) {
1731				/*
1732				 * Load the slog device's state from the MOS
1733				 * config since it's possible that the label
1734				 * does not contain the most up-to-date
1735				 * information.
1736				 */
1737				vdev_load_log_state(tvd, mtvd);
1738				vdev_reopen(tvd);
1739			}
1740
1741			/*
1742			 * Per-vdev ZAP info is stored exclusively in the MOS.
1743			 */
1744			spa_config_valid_zaps(tvd, mtvd);
1745		}
1746	}
1747
1748	vdev_free(mrvd);
1749	spa_config_exit(spa, SCL_ALL, FTAG);
1750
1751	/*
1752	 * Ensure we were able to validate the config.
1753	 */
1754	return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1755}
1756
1757/*
1758 * Check for missing log devices
1759 */
1760static boolean_t
1761spa_check_logs(spa_t *spa)
1762{
1763	boolean_t rv = B_FALSE;
1764	dsl_pool_t *dp = spa_get_dsl(spa);
1765
1766	switch (spa->spa_log_state) {
1767	case SPA_LOG_MISSING:
1768		/* need to recheck in case slog has been restored */
1769	case SPA_LOG_UNKNOWN:
1770		rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1771		    zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
1772		if (rv)
1773			spa_set_log_state(spa, SPA_LOG_MISSING);
1774		break;
1775	}
1776	return (rv);
1777}
1778
1779static boolean_t
1780spa_passivate_log(spa_t *spa)
1781{
1782	vdev_t *rvd = spa->spa_root_vdev;
1783	boolean_t slog_found = B_FALSE;
1784
1785	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1786
1787	if (!spa_has_slogs(spa))
1788		return (B_FALSE);
1789
1790	for (int c = 0; c < rvd->vdev_children; c++) {
1791		vdev_t *tvd = rvd->vdev_child[c];
1792		metaslab_group_t *mg = tvd->vdev_mg;
1793
1794		if (tvd->vdev_islog) {
1795			metaslab_group_passivate(mg);
1796			slog_found = B_TRUE;
1797		}
1798	}
1799
1800	return (slog_found);
1801}
1802
1803static void
1804spa_activate_log(spa_t *spa)
1805{
1806	vdev_t *rvd = spa->spa_root_vdev;
1807
1808	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1809
1810	for (int c = 0; c < rvd->vdev_children; c++) {
1811		vdev_t *tvd = rvd->vdev_child[c];
1812		metaslab_group_t *mg = tvd->vdev_mg;
1813
1814		if (tvd->vdev_islog)
1815			metaslab_group_activate(mg);
1816	}
1817}
1818
1819int
1820spa_offline_log(spa_t *spa)
1821{
1822	int error;
1823
1824	error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1825	    NULL, DS_FIND_CHILDREN);
1826	if (error == 0) {
1827		/*
1828		 * We successfully offlined the log device, sync out the
1829		 * current txg so that the "stubby" block can be removed
1830		 * by zil_sync().
1831		 */
1832		txg_wait_synced(spa->spa_dsl_pool, 0);
1833	}
1834	return (error);
1835}
1836
1837static void
1838spa_aux_check_removed(spa_aux_vdev_t *sav)
1839{
1840	for (int i = 0; i < sav->sav_count; i++)
1841		spa_check_removed(sav->sav_vdevs[i]);
1842}
1843
1844void
1845spa_claim_notify(zio_t *zio)
1846{
1847	spa_t *spa = zio->io_spa;
1848
1849	if (zio->io_error)
1850		return;
1851
1852	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1853	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1854		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1855	mutex_exit(&spa->spa_props_lock);
1856}
1857
1858typedef struct spa_load_error {
1859	uint64_t	sle_meta_count;
1860	uint64_t	sle_data_count;
1861} spa_load_error_t;
1862
1863static void
1864spa_load_verify_done(zio_t *zio)
1865{
1866	blkptr_t *bp = zio->io_bp;
1867	spa_load_error_t *sle = zio->io_private;
1868	dmu_object_type_t type = BP_GET_TYPE(bp);
1869	int error = zio->io_error;
1870	spa_t *spa = zio->io_spa;
1871
1872	if (error) {
1873		if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1874		    type != DMU_OT_INTENT_LOG)
1875			atomic_inc_64(&sle->sle_meta_count);
1876		else
1877			atomic_inc_64(&sle->sle_data_count);
1878	}
1879	zio_data_buf_free(zio->io_data, zio->io_size);
1880
1881	mutex_enter(&spa->spa_scrub_lock);
1882	spa->spa_scrub_inflight--;
1883	cv_broadcast(&spa->spa_scrub_io_cv);
1884	mutex_exit(&spa->spa_scrub_lock);
1885}
1886
1887/*
1888 * Maximum number of concurrent scrub i/os to create while verifying
1889 * a pool while importing it.
1890 */
1891int spa_load_verify_maxinflight = 10000;
1892boolean_t spa_load_verify_metadata = B_TRUE;
1893boolean_t spa_load_verify_data = B_TRUE;
1894
1895/*ARGSUSED*/
1896static int
1897spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1898    const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1899{
1900	if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
1901		return (0);
1902	/*
1903	 * Note: normally this routine will not be called if
1904	 * spa_load_verify_metadata is not set.  However, it may be useful
1905	 * to manually set the flag after the traversal has begun.
1906	 */
1907	if (!spa_load_verify_metadata)
1908		return (0);
1909	if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data)
1910		return (0);
1911
1912	zio_t *rio = arg;
1913	size_t size = BP_GET_PSIZE(bp);
1914	void *data = zio_data_buf_alloc(size);
1915
1916	mutex_enter(&spa->spa_scrub_lock);
1917	while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
1918		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1919	spa->spa_scrub_inflight++;
1920	mutex_exit(&spa->spa_scrub_lock);
1921
1922	zio_nowait(zio_read(rio, spa, bp, data, size,
1923	    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1924	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1925	    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1926	return (0);
1927}
1928
1929static int
1930spa_load_verify(spa_t *spa)
1931{
1932	zio_t *rio;
1933	spa_load_error_t sle = { 0 };
1934	zpool_rewind_policy_t policy;
1935	boolean_t verify_ok = B_FALSE;
1936	int error = 0;
1937
1938	zpool_get_rewind_policy(spa->spa_config, &policy);
1939
1940	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1941		return (0);
1942
1943	rio = zio_root(spa, NULL, &sle,
1944	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1945
1946	if (spa_load_verify_metadata) {
1947		error = traverse_pool(spa, spa->spa_verify_min_txg,
1948		    TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
1949		    spa_load_verify_cb, rio);
1950	}
1951
1952	(void) zio_wait(rio);
1953
1954	spa->spa_load_meta_errors = sle.sle_meta_count;
1955	spa->spa_load_data_errors = sle.sle_data_count;
1956
1957	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1958	    sle.sle_data_count <= policy.zrp_maxdata) {
1959		int64_t loss = 0;
1960
1961		verify_ok = B_TRUE;
1962		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1963		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1964
1965		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1966		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1967		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1968		VERIFY(nvlist_add_int64(spa->spa_load_info,
1969		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1970		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1971		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1972	} else {
1973		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1974	}
1975
1976	if (error) {
1977		if (error != ENXIO && error != EIO)
1978			error = SET_ERROR(EIO);
1979		return (error);
1980	}
1981
1982	return (verify_ok ? 0 : EIO);
1983}
1984
1985/*
1986 * Find a value in the pool props object.
1987 */
1988static void
1989spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1990{
1991	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1992	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1993}
1994
1995/*
1996 * Find a value in the pool directory object.
1997 */
1998static int
1999spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
2000{
2001	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2002	    name, sizeof (uint64_t), 1, val));
2003}
2004
2005static int
2006spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2007{
2008	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2009	return (err);
2010}
2011
2012/*
2013 * Fix up config after a partly-completed split.  This is done with the
2014 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
2015 * pool have that entry in their config, but only the splitting one contains
2016 * a list of all the guids of the vdevs that are being split off.
2017 *
2018 * This function determines what to do with that list: either rejoin
2019 * all the disks to the pool, or complete the splitting process.  To attempt
2020 * the rejoin, each disk that is offlined is marked online again, and
2021 * we do a reopen() call.  If the vdev label for every disk that was
2022 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2023 * then we call vdev_split() on each disk, and complete the split.
2024 *
2025 * Otherwise we leave the config alone, with all the vdevs in place in
2026 * the original pool.
2027 */
2028static void
2029spa_try_repair(spa_t *spa, nvlist_t *config)
2030{
2031	uint_t extracted;
2032	uint64_t *glist;
2033	uint_t i, gcount;
2034	nvlist_t *nvl;
2035	vdev_t **vd;
2036	boolean_t attempt_reopen;
2037
2038	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2039		return;
2040
2041	/* check that the config is complete */
2042	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2043	    &glist, &gcount) != 0)
2044		return;
2045
2046	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2047
2048	/* attempt to online all the vdevs & validate */
2049	attempt_reopen = B_TRUE;
2050	for (i = 0; i < gcount; i++) {
2051		if (glist[i] == 0)	/* vdev is hole */
2052			continue;
2053
2054		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2055		if (vd[i] == NULL) {
2056			/*
2057			 * Don't bother attempting to reopen the disks;
2058			 * just do the split.
2059			 */
2060			attempt_reopen = B_FALSE;
2061		} else {
2062			/* attempt to re-online it */
2063			vd[i]->vdev_offline = B_FALSE;
2064		}
2065	}
2066
2067	if (attempt_reopen) {
2068		vdev_reopen(spa->spa_root_vdev);
2069
2070		/* check each device to see what state it's in */
2071		for (extracted = 0, i = 0; i < gcount; i++) {
2072			if (vd[i] != NULL &&
2073			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2074				break;
2075			++extracted;
2076		}
2077	}
2078
2079	/*
2080	 * If every disk has been moved to the new pool, or if we never
2081	 * even attempted to look at them, then we split them off for
2082	 * good.
2083	 */
2084	if (!attempt_reopen || gcount == extracted) {
2085		for (i = 0; i < gcount; i++)
2086			if (vd[i] != NULL)
2087				vdev_split(vd[i]);
2088		vdev_reopen(spa->spa_root_vdev);
2089	}
2090
2091	kmem_free(vd, gcount * sizeof (vdev_t *));
2092}
2093
2094static int
2095spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2096    boolean_t mosconfig)
2097{
2098	nvlist_t *config = spa->spa_config;
2099	char *ereport = FM_EREPORT_ZFS_POOL;
2100	char *comment;
2101	int error;
2102	uint64_t pool_guid;
2103	nvlist_t *nvl;
2104
2105	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2106		return (SET_ERROR(EINVAL));
2107
2108	ASSERT(spa->spa_comment == NULL);
2109	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2110		spa->spa_comment = spa_strdup(comment);
2111
2112	/*
2113	 * Versioning wasn't explicitly added to the label until later, so if
2114	 * it's not present treat it as the initial version.
2115	 */
2116	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2117	    &spa->spa_ubsync.ub_version) != 0)
2118		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2119
2120	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2121	    &spa->spa_config_txg);
2122
2123	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2124	    spa_guid_exists(pool_guid, 0)) {
2125		error = SET_ERROR(EEXIST);
2126	} else {
2127		spa->spa_config_guid = pool_guid;
2128
2129		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2130		    &nvl) == 0) {
2131			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2132			    KM_SLEEP) == 0);
2133		}
2134
2135		nvlist_free(spa->spa_load_info);
2136		spa->spa_load_info = fnvlist_alloc();
2137
2138		gethrestime(&spa->spa_loaded_ts);
2139		error = spa_load_impl(spa, pool_guid, config, state, type,
2140		    mosconfig, &ereport);
2141	}
2142
2143	/*
2144	 * Don't count references from objsets that are already closed
2145	 * and are making their way through the eviction process.
2146	 */
2147	spa_evicting_os_wait(spa);
2148	spa->spa_minref = refcount_count(&spa->spa_refcount);
2149	if (error) {
2150		if (error != EEXIST) {
2151			spa->spa_loaded_ts.tv_sec = 0;
2152			spa->spa_loaded_ts.tv_nsec = 0;
2153		}
2154		if (error != EBADF) {
2155			zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2156		}
2157	}
2158	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2159	spa->spa_ena = 0;
2160
2161	return (error);
2162}
2163
2164/*
2165 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2166 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2167 * spa's per-vdev ZAP list.
2168 */
2169static uint64_t
2170vdev_count_verify_zaps(vdev_t *vd)
2171{
2172	spa_t *spa = vd->vdev_spa;
2173	uint64_t total = 0;
2174	if (vd->vdev_top_zap != 0) {
2175		total++;
2176		ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2177		    spa->spa_all_vdev_zaps, vd->vdev_top_zap));
2178	}
2179	if (vd->vdev_leaf_zap != 0) {
2180		total++;
2181		ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2182		    spa->spa_all_vdev_zaps, vd->vdev_leaf_zap));
2183	}
2184
2185	for (uint64_t i = 0; i < vd->vdev_children; i++) {
2186		total += vdev_count_verify_zaps(vd->vdev_child[i]);
2187	}
2188
2189	return (total);
2190}
2191
2192/*
2193 * Load an existing storage pool, using the pool's builtin spa_config as a
2194 * source of configuration information.
2195 */
2196static int
2197spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2198    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2199    char **ereport)
2200{
2201	int error = 0;
2202	nvlist_t *nvroot = NULL;
2203	nvlist_t *label;
2204	vdev_t *rvd;
2205	uberblock_t *ub = &spa->spa_uberblock;
2206	uint64_t children, config_cache_txg = spa->spa_config_txg;
2207	int orig_mode = spa->spa_mode;
2208	int parse;
2209	uint64_t obj;
2210	boolean_t missing_feat_write = B_FALSE;
2211
2212	/*
2213	 * If this is an untrusted config, access the pool in read-only mode.
2214	 * This prevents things like resilvering recently removed devices.
2215	 */
2216	if (!mosconfig)
2217		spa->spa_mode = FREAD;
2218
2219	ASSERT(MUTEX_HELD(&spa_namespace_lock));
2220
2221	spa->spa_load_state = state;
2222
2223	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2224		return (SET_ERROR(EINVAL));
2225
2226	parse = (type == SPA_IMPORT_EXISTING ?
2227	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2228
2229	/*
2230	 * Create "The Godfather" zio to hold all async IOs
2231	 */
2232	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2233	    KM_SLEEP);
2234	for (int i = 0; i < max_ncpus; i++) {
2235		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2236		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2237		    ZIO_FLAG_GODFATHER);
2238	}
2239
2240	/*
2241	 * Parse the configuration into a vdev tree.  We explicitly set the
2242	 * value that will be returned by spa_version() since parsing the
2243	 * configuration requires knowing the version number.
2244	 */
2245	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2246	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2247	spa_config_exit(spa, SCL_ALL, FTAG);
2248
2249	if (error != 0)
2250		return (error);
2251
2252	ASSERT(spa->spa_root_vdev == rvd);
2253	ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
2254	ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
2255
2256	if (type != SPA_IMPORT_ASSEMBLE) {
2257		ASSERT(spa_guid(spa) == pool_guid);
2258	}
2259
2260	/*
2261	 * Try to open all vdevs, loading each label in the process.
2262	 */
2263	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2264	error = vdev_open(rvd);
2265	spa_config_exit(spa, SCL_ALL, FTAG);
2266	if (error != 0)
2267		return (error);
2268
2269	/*
2270	 * We need to validate the vdev labels against the configuration that
2271	 * we have in hand, which is dependent on the setting of mosconfig. If
2272	 * mosconfig is true then we're validating the vdev labels based on
2273	 * that config.  Otherwise, we're validating against the cached config
2274	 * (zpool.cache) that was read when we loaded the zfs module, and then
2275	 * later we will recursively call spa_load() and validate against
2276	 * the vdev config.
2277	 *
2278	 * If we're assembling a new pool that's been split off from an
2279	 * existing pool, the labels haven't yet been updated so we skip
2280	 * validation for now.
2281	 */
2282	if (type != SPA_IMPORT_ASSEMBLE) {
2283		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2284		error = vdev_validate(rvd, mosconfig);
2285		spa_config_exit(spa, SCL_ALL, FTAG);
2286
2287		if (error != 0)
2288			return (error);
2289
2290		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2291			return (SET_ERROR(ENXIO));
2292	}
2293
2294	/*
2295	 * Find the best uberblock.
2296	 */
2297	vdev_uberblock_load(rvd, ub, &label);
2298
2299	/*
2300	 * If we weren't able to find a single valid uberblock, return failure.
2301	 */
2302	if (ub->ub_txg == 0) {
2303		nvlist_free(label);
2304		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2305	}
2306
2307	/*
2308	 * If the pool has an unsupported version we can't open it.
2309	 */
2310	if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2311		nvlist_free(label);
2312		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2313	}
2314
2315	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2316		nvlist_t *features;
2317
2318		/*
2319		 * If we weren't able to find what's necessary for reading the
2320		 * MOS in the label, return failure.
2321		 */
2322		if (label == NULL || nvlist_lookup_nvlist(label,
2323		    ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2324			nvlist_free(label);
2325			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2326			    ENXIO));
2327		}
2328
2329		/*
2330		 * Update our in-core representation with the definitive values
2331		 * from the label.
2332		 */
2333		nvlist_free(spa->spa_label_features);
2334		VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2335	}
2336
2337	nvlist_free(label);
2338
2339	/*
2340	 * Look through entries in the label nvlist's features_for_read. If
2341	 * there is a feature listed there which we don't understand then we
2342	 * cannot open a pool.
2343	 */
2344	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2345		nvlist_t *unsup_feat;
2346
2347		VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2348		    0);
2349
2350		for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2351		    NULL); nvp != NULL;
2352		    nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2353			if (!zfeature_is_supported(nvpair_name(nvp))) {
2354				VERIFY(nvlist_add_string(unsup_feat,
2355				    nvpair_name(nvp), "") == 0);
2356			}
2357		}
2358
2359		if (!nvlist_empty(unsup_feat)) {
2360			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2361			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2362			nvlist_free(unsup_feat);
2363			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2364			    ENOTSUP));
2365		}
2366
2367		nvlist_free(unsup_feat);
2368	}
2369
2370	/*
2371	 * If the vdev guid sum doesn't match the uberblock, we have an
2372	 * incomplete configuration.  We first check to see if the pool
2373	 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2374	 * If it is, defer the vdev_guid_sum check till later so we
2375	 * can handle missing vdevs.
2376	 */
2377	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2378	    &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2379	    rvd->vdev_guid_sum != ub->ub_guid_sum)
2380		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2381
2382	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2383		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2384		spa_try_repair(spa, config);
2385		spa_config_exit(spa, SCL_ALL, FTAG);
2386		nvlist_free(spa->spa_config_splitting);
2387		spa->spa_config_splitting = NULL;
2388	}
2389
2390	/*
2391	 * Initialize internal SPA structures.
2392	 */
2393	spa->spa_state = POOL_STATE_ACTIVE;
2394	spa->spa_ubsync = spa->spa_uberblock;
2395	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2396	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2397	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2398	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2399	spa->spa_claim_max_txg = spa->spa_first_txg;
2400	spa->spa_prev_software_version = ub->ub_software_version;
2401
2402	error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2403	if (error)
2404		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2405	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2406
2407	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2408		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2409
2410	if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2411		boolean_t missing_feat_read = B_FALSE;
2412		nvlist_t *unsup_feat, *enabled_feat;
2413
2414		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2415		    &spa->spa_feat_for_read_obj) != 0) {
2416			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2417		}
2418
2419		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2420		    &spa->spa_feat_for_write_obj) != 0) {
2421			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2422		}
2423
2424		if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2425		    &spa->spa_feat_desc_obj) != 0) {
2426			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2427		}
2428
2429		enabled_feat = fnvlist_alloc();
2430		unsup_feat = fnvlist_alloc();
2431
2432		if (!spa_features_check(spa, B_FALSE,
2433		    unsup_feat, enabled_feat))
2434			missing_feat_read = B_TRUE;
2435
2436		if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2437			if (!spa_features_check(spa, B_TRUE,
2438			    unsup_feat, enabled_feat)) {
2439				missing_feat_write = B_TRUE;
2440			}
2441		}
2442
2443		fnvlist_add_nvlist(spa->spa_load_info,
2444		    ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2445
2446		if (!nvlist_empty(unsup_feat)) {
2447			fnvlist_add_nvlist(spa->spa_load_info,
2448			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2449		}
2450
2451		fnvlist_free(enabled_feat);
2452		fnvlist_free(unsup_feat);
2453
2454		if (!missing_feat_read) {
2455			fnvlist_add_boolean(spa->spa_load_info,
2456			    ZPOOL_CONFIG_CAN_RDONLY);
2457		}
2458
2459		/*
2460		 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2461		 * twofold: to determine whether the pool is available for
2462		 * import in read-write mode and (if it is not) whether the
2463		 * pool is available for import in read-only mode. If the pool
2464		 * is available for import in read-write mode, it is displayed
2465		 * as available in userland; if it is not available for import
2466		 * in read-only mode, it is displayed as unavailable in
2467		 * userland. If the pool is available for import in read-only
2468		 * mode but not read-write mode, it is displayed as unavailable
2469		 * in userland with a special note that the pool is actually
2470		 * available for open in read-only mode.
2471		 *
2472		 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2473		 * missing a feature for write, we must first determine whether
2474		 * the pool can be opened read-only before returning to
2475		 * userland in order to know whether to display the
2476		 * abovementioned note.
2477		 */
2478		if (missing_feat_read || (missing_feat_write &&
2479		    spa_writeable(spa))) {
2480			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2481			    ENOTSUP));
2482		}
2483
2484		/*
2485		 * Load refcounts for ZFS features from disk into an in-memory
2486		 * cache during SPA initialization.
2487		 */
2488		for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2489			uint64_t refcount;
2490
2491			error = feature_get_refcount_from_disk(spa,
2492			    &spa_feature_table[i], &refcount);
2493			if (error == 0) {
2494				spa->spa_feat_refcount_cache[i] = refcount;
2495			} else if (error == ENOTSUP) {
2496				spa->spa_feat_refcount_cache[i] =
2497				    SPA_FEATURE_DISABLED;
2498			} else {
2499				return (spa_vdev_err(rvd,
2500				    VDEV_AUX_CORRUPT_DATA, EIO));
2501			}
2502		}
2503	}
2504
2505	if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2506		if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2507		    &spa->spa_feat_enabled_txg_obj) != 0)
2508			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2509	}
2510
2511	spa->spa_is_initializing = B_TRUE;
2512	error = dsl_pool_open(spa->spa_dsl_pool);
2513	spa->spa_is_initializing = B_FALSE;
2514	if (error != 0)
2515		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2516
2517	if (!mosconfig) {
2518		uint64_t hostid;
2519		nvlist_t *policy = NULL, *nvconfig;
2520
2521		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2522			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2523
2524		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2525		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2526			char *hostname;
2527			unsigned long myhostid = 0;
2528
2529			VERIFY(nvlist_lookup_string(nvconfig,
2530			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2531
2532#ifdef	_KERNEL
2533			myhostid = zone_get_hostid(NULL);
2534#else	/* _KERNEL */
2535			/*
2536			 * We're emulating the system's hostid in userland, so
2537			 * we can't use zone_get_hostid().
2538			 */
2539			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2540#endif	/* _KERNEL */
2541			if (hostid != 0 && myhostid != 0 &&
2542			    hostid != myhostid) {
2543				nvlist_free(nvconfig);
2544				cmn_err(CE_WARN, "pool '%s' could not be "
2545				    "loaded as it was last accessed by "
2546				    "another system (host: %s hostid: 0x%lx). "
2547				    "See: http://illumos.org/msg/ZFS-8000-EY",
2548				    spa_name(spa), hostname,
2549				    (unsigned long)hostid);
2550				return (SET_ERROR(EBADF));
2551			}
2552		}
2553		if (nvlist_lookup_nvlist(spa->spa_config,
2554		    ZPOOL_REWIND_POLICY, &policy) == 0)
2555			VERIFY(nvlist_add_nvlist(nvconfig,
2556			    ZPOOL_REWIND_POLICY, policy) == 0);
2557
2558		spa_config_set(spa, nvconfig);
2559		spa_unload(spa);
2560		spa_deactivate(spa);
2561		spa_activate(spa, orig_mode);
2562
2563		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2564	}
2565
2566	/* Grab the secret checksum salt from the MOS. */
2567	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2568	    DMU_POOL_CHECKSUM_SALT, 1,
2569	    sizeof (spa->spa_cksum_salt.zcs_bytes),
2570	    spa->spa_cksum_salt.zcs_bytes);
2571	if (error == ENOENT) {
2572		/* Generate a new salt for subsequent use */
2573		(void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
2574		    sizeof (spa->spa_cksum_salt.zcs_bytes));
2575	} else if (error != 0) {
2576		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2577	}
2578
2579	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2580		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2581	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2582	if (error != 0)
2583		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2584
2585	/*
2586	 * Load the bit that tells us to use the new accounting function
2587	 * (raid-z deflation).  If we have an older pool, this will not
2588	 * be present.
2589	 */
2590	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2591	if (error != 0 && error != ENOENT)
2592		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2593
2594	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2595	    &spa->spa_creation_version);
2596	if (error != 0 && error != ENOENT)
2597		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2598
2599	/*
2600	 * Load the persistent error log.  If we have an older pool, this will
2601	 * not be present.
2602	 */
2603	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2604	if (error != 0 && error != ENOENT)
2605		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2606
2607	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2608	    &spa->spa_errlog_scrub);
2609	if (error != 0 && error != ENOENT)
2610		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2611
2612	/*
2613	 * Load the history object.  If we have an older pool, this
2614	 * will not be present.
2615	 */
2616	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2617	if (error != 0 && error != ENOENT)
2618		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2619
2620	/*
2621	 * Load the per-vdev ZAP map. If we have an older pool, this will not
2622	 * be present; in this case, defer its creation to a later time to
2623	 * avoid dirtying the MOS this early / out of sync context. See
2624	 * spa_sync_config_object.
2625	 */
2626
2627	/* The sentinel is only available in the MOS config. */
2628	nvlist_t *mos_config;
2629	if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0)
2630		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2631
2632	error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP,
2633	    &spa->spa_all_vdev_zaps);
2634
2635	if (error != ENOENT && error != 0) {
2636		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2637	} else if (error == 0 && !nvlist_exists(mos_config,
2638	    ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) {
2639		/*
2640		 * An older version of ZFS overwrote the sentinel value, so
2641		 * we have orphaned per-vdev ZAPs in the MOS. Defer their
2642		 * destruction to later; see spa_sync_config_object.
2643		 */
2644		spa->spa_avz_action = AVZ_ACTION_DESTROY;
2645		/*
2646		 * We're assuming that no vdevs have had their ZAPs created
2647		 * before this. Better be sure of it.
2648		 */
2649		ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
2650	}
2651	nvlist_free(mos_config);
2652
2653	/*
2654	 * If we're assembling the pool from the split-off vdevs of
2655	 * an existing pool, we don't want to attach the spares & cache
2656	 * devices.
2657	 */
2658
2659	/*
2660	 * Load any hot spares for this pool.
2661	 */
2662	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2663	if (error != 0 && error != ENOENT)
2664		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2665	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2666		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2667		if (load_nvlist(spa, spa->spa_spares.sav_object,
2668		    &spa->spa_spares.sav_config) != 0)
2669			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2670
2671		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2672		spa_load_spares(spa);
2673		spa_config_exit(spa, SCL_ALL, FTAG);
2674	} else if (error == 0) {
2675		spa->spa_spares.sav_sync = B_TRUE;
2676	}
2677
2678	/*
2679	 * Load any level 2 ARC devices for this pool.
2680	 */
2681	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2682	    &spa->spa_l2cache.sav_object);
2683	if (error != 0 && error != ENOENT)
2684		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2685	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2686		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2687		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2688		    &spa->spa_l2cache.sav_config) != 0)
2689			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2690
2691		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2692		spa_load_l2cache(spa);
2693		spa_config_exit(spa, SCL_ALL, FTAG);
2694	} else if (error == 0) {
2695		spa->spa_l2cache.sav_sync = B_TRUE;
2696	}
2697
2698	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2699
2700	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2701	if (error && error != ENOENT)
2702		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2703
2704	if (error == 0) {
2705		uint64_t autoreplace;
2706
2707		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2708		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2709		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2710		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2711		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2712		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2713		    &spa->spa_dedup_ditto);
2714
2715		spa->spa_autoreplace = (autoreplace != 0);
2716	}
2717
2718	/*
2719	 * If the 'autoreplace' property is set, then post a resource notifying
2720	 * the ZFS DE that it should not issue any faults for unopenable
2721	 * devices.  We also iterate over the vdevs, and post a sysevent for any
2722	 * unopenable vdevs so that the normal autoreplace handler can take
2723	 * over.
2724	 */
2725	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2726		spa_check_removed(spa->spa_root_vdev);
2727		/*
2728		 * For the import case, this is done in spa_import(), because
2729		 * at this point we're using the spare definitions from
2730		 * the MOS config, not necessarily from the userland config.
2731		 */
2732		if (state != SPA_LOAD_IMPORT) {
2733			spa_aux_check_removed(&spa->spa_spares);
2734			spa_aux_check_removed(&spa->spa_l2cache);
2735		}
2736	}
2737
2738	/*
2739	 * Load the vdev state for all toplevel vdevs.
2740	 */
2741	vdev_load(rvd);
2742
2743	/*
2744	 * Propagate the leaf DTLs we just loaded all the way up the tree.
2745	 */
2746	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2747	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2748	spa_config_exit(spa, SCL_ALL, FTAG);
2749
2750	/*
2751	 * Load the DDTs (dedup tables).
2752	 */
2753	error = ddt_load(spa);
2754	if (error != 0)
2755		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2756
2757	spa_update_dspace(spa);
2758
2759	/*
2760	 * Validate the config, using the MOS config to fill in any
2761	 * information which might be missing.  If we fail to validate
2762	 * the config then declare the pool unfit for use. If we're
2763	 * assembling a pool from a split, the log is not transferred
2764	 * over.
2765	 */
2766	if (type != SPA_IMPORT_ASSEMBLE) {
2767		nvlist_t *nvconfig;
2768
2769		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2770			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2771
2772		if (!spa_config_valid(spa, nvconfig)) {
2773			nvlist_free(nvconfig);
2774			return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2775			    ENXIO));
2776		}
2777		nvlist_free(nvconfig);
2778
2779		/*
2780		 * Now that we've validated the config, check the state of the
2781		 * root vdev.  If it can't be opened, it indicates one or
2782		 * more toplevel vdevs are faulted.
2783		 */
2784		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2785			return (SET_ERROR(ENXIO));
2786
2787		if (spa_writeable(spa) && spa_check_logs(spa)) {
2788			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2789			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2790		}
2791	}
2792
2793	if (missing_feat_write) {
2794		ASSERT(state == SPA_LOAD_TRYIMPORT);
2795
2796		/*
2797		 * At this point, we know that we can open the pool in
2798		 * read-only mode but not read-write mode. We now have enough
2799		 * information and can return to userland.
2800		 */
2801		return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2802	}
2803
2804	/*
2805	 * We've successfully opened the pool, verify that we're ready
2806	 * to start pushing transactions.
2807	 */
2808	if (state != SPA_LOAD_TRYIMPORT) {
2809		if (error = spa_load_verify(spa))
2810			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2811			    error));
2812	}
2813
2814	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2815	    spa->spa_load_max_txg == UINT64_MAX)) {
2816		dmu_tx_t *tx;
2817		int need_update = B_FALSE;
2818		dsl_pool_t *dp = spa_get_dsl(spa);
2819
2820		ASSERT(state != SPA_LOAD_TRYIMPORT);
2821
2822		/*
2823		 * Claim log blocks that haven't been committed yet.
2824		 * This must all happen in a single txg.
2825		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2826		 * invoked from zil_claim_log_block()'s i/o done callback.
2827		 * Price of rollback is that we abandon the log.
2828		 */
2829		spa->spa_claiming = B_TRUE;
2830
2831		tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
2832		(void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
2833		    zil_claim, tx, DS_FIND_CHILDREN);
2834		dmu_tx_commit(tx);
2835
2836		spa->spa_claiming = B_FALSE;
2837
2838		spa_set_log_state(spa, SPA_LOG_GOOD);
2839		spa->spa_sync_on = B_TRUE;
2840		txg_sync_start(spa->spa_dsl_pool);
2841
2842		/*
2843		 * Wait for all claims to sync.  We sync up to the highest
2844		 * claimed log block birth time so that claimed log blocks
2845		 * don't appear to be from the future.  spa_claim_max_txg
2846		 * will have been set for us by either zil_check_log_chain()
2847		 * (invoked from spa_check_logs()) or zil_claim() above.
2848		 */
2849		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2850
2851		/*
2852		 * If the config cache is stale, or we have uninitialized
2853		 * metaslabs (see spa_vdev_add()), then update the config.
2854		 *
2855		 * If this is a verbatim import, trust the current
2856		 * in-core spa_config and update the disk labels.
2857		 */
2858		if (config_cache_txg != spa->spa_config_txg ||
2859		    state == SPA_LOAD_IMPORT ||
2860		    state == SPA_LOAD_RECOVER ||
2861		    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2862			need_update = B_TRUE;
2863
2864		for (int c = 0; c < rvd->vdev_children; c++)
2865			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2866				need_update = B_TRUE;
2867
2868		/*
2869		 * Update the config cache asychronously in case we're the
2870		 * root pool, in which case the config cache isn't writable yet.
2871		 */
2872		if (need_update)
2873			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2874
2875		/*
2876		 * Check all DTLs to see if anything needs resilvering.
2877		 */
2878		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2879		    vdev_resilver_needed(rvd, NULL, NULL))
2880			spa_async_request(spa, SPA_ASYNC_RESILVER);
2881
2882		/*
2883		 * Log the fact that we booted up (so that we can detect if
2884		 * we rebooted in the middle of an operation).
2885		 */
2886		spa_history_log_version(spa, "open");
2887
2888		/*
2889		 * Delete any inconsistent datasets.
2890		 */
2891		(void) dmu_objset_find(spa_name(spa),
2892		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2893
2894		/*
2895		 * Clean up any stale temporary dataset userrefs.
2896		 */
2897		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2898	}
2899
2900	return (0);
2901}
2902
2903static int
2904spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2905{
2906	int mode = spa->spa_mode;
2907
2908	spa_unload(spa);
2909	spa_deactivate(spa);
2910
2911	spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
2912
2913	spa_activate(spa, mode);
2914	spa_async_suspend(spa);
2915
2916	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2917}
2918
2919/*
2920 * If spa_load() fails this function will try loading prior txg's. If
2921 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2922 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2923 * function will not rewind the pool and will return the same error as
2924 * spa_load().
2925 */
2926static int
2927spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2928    uint64_t max_request, int rewind_flags)
2929{
2930	nvlist_t *loadinfo = NULL;
2931	nvlist_t *config = NULL;
2932	int load_error, rewind_error;
2933	uint64_t safe_rewind_txg;
2934	uint64_t min_txg;
2935
2936	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2937		spa->spa_load_max_txg = spa->spa_load_txg;
2938		spa_set_log_state(spa, SPA_LOG_CLEAR);
2939	} else {
2940		spa->spa_load_max_txg = max_request;
2941		if (max_request != UINT64_MAX)
2942			spa->spa_extreme_rewind = B_TRUE;
2943	}
2944
2945	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2946	    mosconfig);
2947	if (load_error == 0)
2948		return (0);
2949
2950	if (spa->spa_root_vdev != NULL)
2951		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2952
2953	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2954	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2955
2956	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2957		nvlist_free(config);
2958		return (load_error);
2959	}
2960
2961	if (state == SPA_LOAD_RECOVER) {
2962		/* Price of rolling back is discarding txgs, including log */
2963		spa_set_log_state(spa, SPA_LOG_CLEAR);
2964	} else {
2965		/*
2966		 * If we aren't rolling back save the load info from our first
2967		 * import attempt so that we can restore it after attempting
2968		 * to rewind.
2969		 */
2970		loadinfo = spa->spa_load_info;
2971		spa->spa_load_info = fnvlist_alloc();
2972	}
2973
2974	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2975	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2976	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2977	    TXG_INITIAL : safe_rewind_txg;
2978
2979	/*
2980	 * Continue as long as we're finding errors, we're still within
2981	 * the acceptable rewind range, and we're still finding uberblocks
2982	 */
2983	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2984	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2985		if (spa->spa_load_max_txg < safe_rewind_txg)
2986			spa->spa_extreme_rewind = B_TRUE;
2987		rewind_error = spa_load_retry(spa, state, mosconfig);
2988	}
2989
2990	spa->spa_extreme_rewind = B_FALSE;
2991	spa->spa_load_max_txg = UINT64_MAX;
2992
2993	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2994		spa_config_set(spa, config);
2995
2996	if (state == SPA_LOAD_RECOVER) {
2997		ASSERT3P(loadinfo, ==, NULL);
2998		return (rewind_error);
2999	} else {
3000		/* Store the rewind info as part of the initial load info */
3001		fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
3002		    spa->spa_load_info);
3003
3004		/* Restore the initial load info */
3005		fnvlist_free(spa->spa_load_info);
3006		spa->spa_load_info = loadinfo;
3007
3008		return (load_error);
3009	}
3010}
3011
3012/*
3013 * Pool Open/Import
3014 *
3015 * The import case is identical to an open except that the configuration is sent
3016 * down from userland, instead of grabbed from the configuration cache.  For the
3017 * case of an open, the pool configuration will exist in the
3018 * POOL_STATE_UNINITIALIZED state.
3019 *
3020 * The stats information (gen/count/ustats) is used to gather vdev statistics at
3021 * the same time open the pool, without having to keep around the spa_t in some
3022 * ambiguous state.
3023 */
3024static int
3025spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
3026    nvlist_t **config)
3027{
3028	spa_t *spa;
3029	spa_load_state_t state = SPA_LOAD_OPEN;
3030	int error;
3031	int locked = B_FALSE;
3032
3033	*spapp = NULL;
3034
3035	/*
3036	 * As disgusting as this is, we need to support recursive calls to this
3037	 * function because dsl_dir_open() is called during spa_load(), and ends
3038	 * up calling spa_open() again.  The real fix is to figure out how to
3039	 * avoid dsl_dir_open() calling this in the first place.
3040	 */
3041	if (mutex_owner(&spa_namespace_lock) != curthread) {
3042		mutex_enter(&spa_namespace_lock);
3043		locked = B_TRUE;
3044	}
3045
3046	if ((spa = spa_lookup(pool)) == NULL) {
3047		if (locked)
3048			mutex_exit(&spa_namespace_lock);
3049		return (SET_ERROR(ENOENT));
3050	}
3051
3052	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
3053		zpool_rewind_policy_t policy;
3054
3055		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
3056		    &policy);
3057		if (policy.zrp_request & ZPOOL_DO_REWIND)
3058			state = SPA_LOAD_RECOVER;
3059
3060		spa_activate(spa, spa_mode_global);
3061
3062		if (state != SPA_LOAD_RECOVER)
3063			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3064
3065		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
3066		    policy.zrp_request);
3067
3068		if (error == EBADF) {
3069			/*
3070			 * If vdev_validate() returns failure (indicated by
3071			 * EBADF), it indicates that one of the vdevs indicates
3072			 * that the pool has been exported or destroyed.  If
3073			 * this is the case, the config cache is out of sync and
3074			 * we should remove the pool from the namespace.
3075			 */
3076			spa_unload(spa);
3077			spa_deactivate(spa);
3078			spa_config_sync(spa, B_TRUE, B_TRUE);
3079			spa_remove(spa);
3080			if (locked)
3081				mutex_exit(&spa_namespace_lock);
3082			return (SET_ERROR(ENOENT));
3083		}
3084
3085		if (error) {
3086			/*
3087			 * We can't open the pool, but we still have useful
3088			 * information: the state of each vdev after the
3089			 * attempted vdev_open().  Return this to the user.
3090			 */
3091			if (config != NULL && spa->spa_config) {
3092				VERIFY(nvlist_dup(spa->spa_config, config,
3093				    KM_SLEEP) == 0);
3094				VERIFY(nvlist_add_nvlist(*config,
3095				    ZPOOL_CONFIG_LOAD_INFO,
3096				    spa->spa_load_info) == 0);
3097			}
3098			spa_unload(spa);
3099			spa_deactivate(spa);
3100			spa->spa_last_open_failed = error;
3101			if (locked)
3102				mutex_exit(&spa_namespace_lock);
3103			*spapp = NULL;
3104			return (error);
3105		}
3106	}
3107
3108	spa_open_ref(spa, tag);
3109
3110	if (config != NULL)
3111		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3112
3113	/*
3114	 * If we've recovered the pool, pass back any information we
3115	 * gathered while doing the load.
3116	 */
3117	if (state == SPA_LOAD_RECOVER) {
3118		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
3119		    spa->spa_load_info) == 0);
3120	}
3121
3122	if (locked) {
3123		spa->spa_last_open_failed = 0;
3124		spa->spa_last_ubsync_txg = 0;
3125		spa->spa_load_txg = 0;
3126		mutex_exit(&spa_namespace_lock);
3127	}
3128
3129	*spapp = spa;
3130
3131	return (0);
3132}
3133
3134int
3135spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3136    nvlist_t **config)
3137{
3138	return (spa_open_common(name, spapp, tag, policy, config));
3139}
3140
3141int
3142spa_open(const char *name, spa_t **spapp, void *tag)
3143{
3144	return (spa_open_common(name, spapp, tag, NULL, NULL));
3145}
3146
3147/*
3148 * Lookup the given spa_t, incrementing the inject count in the process,
3149 * preventing it from being exported or destroyed.
3150 */
3151spa_t *
3152spa_inject_addref(char *name)
3153{
3154	spa_t *spa;
3155
3156	mutex_enter(&spa_namespace_lock);
3157	if ((spa = spa_lookup(name)) == NULL) {
3158		mutex_exit(&spa_namespace_lock);
3159		return (NULL);
3160	}
3161	spa->spa_inject_ref++;
3162	mutex_exit(&spa_namespace_lock);
3163
3164	return (spa);
3165}
3166
3167void
3168spa_inject_delref(spa_t *spa)
3169{
3170	mutex_enter(&spa_namespace_lock);
3171	spa->spa_inject_ref--;
3172	mutex_exit(&spa_namespace_lock);
3173}
3174
3175/*
3176 * Add spares device information to the nvlist.
3177 */
3178static void
3179spa_add_spares(spa_t *spa, nvlist_t *config)
3180{
3181	nvlist_t **spares;
3182	uint_t i, nspares;
3183	nvlist_t *nvroot;
3184	uint64_t guid;
3185	vdev_stat_t *vs;
3186	uint_t vsc;
3187	uint64_t pool;
3188
3189	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3190
3191	if (spa->spa_spares.sav_count == 0)
3192		return;
3193
3194	VERIFY(nvlist_lookup_nvlist(config,
3195	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3196	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3197	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3198	if (nspares != 0) {
3199		VERIFY(nvlist_add_nvlist_array(nvroot,
3200		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3201		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3202		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3203
3204		/*
3205		 * Go through and find any spares which have since been
3206		 * repurposed as an active spare.  If this is the case, update
3207		 * their status appropriately.
3208		 */
3209		for (i = 0; i < nspares; i++) {
3210			VERIFY(nvlist_lookup_uint64(spares[i],
3211			    ZPOOL_CONFIG_GUID, &guid) == 0);
3212			if (spa_spare_exists(guid, &pool, NULL) &&
3213			    pool != 0ULL) {
3214				VERIFY(nvlist_lookup_uint64_array(
3215				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
3216				    (uint64_t **)&vs, &vsc) == 0);
3217				vs->vs_state = VDEV_STATE_CANT_OPEN;
3218				vs->vs_aux = VDEV_AUX_SPARED;
3219			}
3220		}
3221	}
3222}
3223
3224/*
3225 * Add l2cache device information to the nvlist, including vdev stats.
3226 */
3227static void
3228spa_add_l2cache(spa_t *spa, nvlist_t *config)
3229{
3230	nvlist_t **l2cache;
3231	uint_t i, j, nl2cache;
3232	nvlist_t *nvroot;
3233	uint64_t guid;
3234	vdev_t *vd;
3235	vdev_stat_t *vs;
3236	uint_t vsc;
3237
3238	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3239
3240	if (spa->spa_l2cache.sav_count == 0)
3241		return;
3242
3243	VERIFY(nvlist_lookup_nvlist(config,
3244	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3245	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3246	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3247	if (nl2cache != 0) {
3248		VERIFY(nvlist_add_nvlist_array(nvroot,
3249		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3250		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3251		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3252
3253		/*
3254		 * Update level 2 cache device stats.
3255		 */
3256
3257		for (i = 0; i < nl2cache; i++) {
3258			VERIFY(nvlist_lookup_uint64(l2cache[i],
3259			    ZPOOL_CONFIG_GUID, &guid) == 0);
3260
3261			vd = NULL;
3262			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3263				if (guid ==
3264				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3265					vd = spa->spa_l2cache.sav_vdevs[j];
3266					break;
3267				}
3268			}
3269			ASSERT(vd != NULL);
3270
3271			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3272			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3273			    == 0);
3274			vdev_get_stats(vd, vs);
3275		}
3276	}
3277}
3278
3279static void
3280spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3281{
3282	nvlist_t *features;
3283	zap_cursor_t zc;
3284	zap_attribute_t za;
3285
3286	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3287	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3288
3289	if (spa->spa_feat_for_read_obj != 0) {
3290		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3291		    spa->spa_feat_for_read_obj);
3292		    zap_cursor_retrieve(&zc, &za) == 0;
3293		    zap_cursor_advance(&zc)) {
3294			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3295			    za.za_num_integers == 1);
3296			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3297			    za.za_first_integer));
3298		}
3299		zap_cursor_fini(&zc);
3300	}
3301
3302	if (spa->spa_feat_for_write_obj != 0) {
3303		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3304		    spa->spa_feat_for_write_obj);
3305		    zap_cursor_retrieve(&zc, &za) == 0;
3306		    zap_cursor_advance(&zc)) {
3307			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3308			    za.za_num_integers == 1);
3309			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3310			    za.za_first_integer));
3311		}
3312		zap_cursor_fini(&zc);
3313	}
3314
3315	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3316	    features) == 0);
3317	nvlist_free(features);
3318}
3319
3320int
3321spa_get_stats(const char *name, nvlist_t **config,
3322    char *altroot, size_t buflen)
3323{
3324	int error;
3325	spa_t *spa;
3326
3327	*config = NULL;
3328	error = spa_open_common(name, &spa, FTAG, NULL, config);
3329
3330	if (spa != NULL) {
3331		/*
3332		 * This still leaves a window of inconsistency where the spares
3333		 * or l2cache devices could change and the config would be
3334		 * self-inconsistent.
3335		 */
3336		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3337
3338		if (*config != NULL) {
3339			uint64_t loadtimes[2];
3340
3341			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3342			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3343			VERIFY(nvlist_add_uint64_array(*config,
3344			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3345
3346			VERIFY(nvlist_add_uint64(*config,
3347			    ZPOOL_CONFIG_ERRCOUNT,
3348			    spa_get_errlog_size(spa)) == 0);
3349
3350			if (spa_suspended(spa))
3351				VERIFY(nvlist_add_uint64(*config,
3352				    ZPOOL_CONFIG_SUSPENDED,
3353				    spa->spa_failmode) == 0);
3354
3355			spa_add_spares(spa, *config);
3356			spa_add_l2cache(spa, *config);
3357			spa_add_feature_stats(spa, *config);
3358		}
3359	}
3360
3361	/*
3362	 * We want to get the alternate root even for faulted pools, so we cheat
3363	 * and call spa_lookup() directly.
3364	 */
3365	if (altroot) {
3366		if (spa == NULL) {
3367			mutex_enter(&spa_namespace_lock);
3368			spa = spa_lookup(name);
3369			if (spa)
3370				spa_altroot(spa, altroot, buflen);
3371			else
3372				altroot[0] = '\0';
3373			spa = NULL;
3374			mutex_exit(&spa_namespace_lock);
3375		} else {
3376			spa_altroot(spa, altroot, buflen);
3377		}
3378	}
3379
3380	if (spa != NULL) {
3381		spa_config_exit(spa, SCL_CONFIG, FTAG);
3382		spa_close(spa, FTAG);
3383	}
3384
3385	return (error);
3386}
3387
3388/*
3389 * Validate that the auxiliary device array is well formed.  We must have an
3390 * array of nvlists, each which describes a valid leaf vdev.  If this is an
3391 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3392 * specified, as long as they are well-formed.
3393 */
3394static int
3395spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3396    spa_aux_vdev_t *sav, const char *config, uint64_t version,
3397    vdev_labeltype_t label)
3398{
3399	nvlist_t **dev;
3400	uint_t i, ndev;
3401	vdev_t *vd;
3402	int error;
3403
3404	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3405
3406	/*
3407	 * It's acceptable to have no devs specified.
3408	 */
3409	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3410		return (0);
3411
3412	if (ndev == 0)
3413		return (SET_ERROR(EINVAL));
3414
3415	/*
3416	 * Make sure the pool is formatted with a version that supports this
3417	 * device type.
3418	 */
3419	if (spa_version(spa) < version)
3420		return (SET_ERROR(ENOTSUP));
3421
3422	/*
3423	 * Set the pending device list so we correctly handle device in-use
3424	 * checking.
3425	 */
3426	sav->sav_pending = dev;
3427	sav->sav_npending = ndev;
3428
3429	for (i = 0; i < ndev; i++) {
3430		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3431		    mode)) != 0)
3432			goto out;
3433
3434		if (!vd->vdev_ops->vdev_op_leaf) {
3435			vdev_free(vd);
3436			error = SET_ERROR(EINVAL);
3437			goto out;
3438		}
3439
3440		/*
3441		 * The L2ARC currently only supports disk devices in
3442		 * kernel context.  For user-level testing, we allow it.
3443		 */
3444#ifdef _KERNEL
3445		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3446		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3447			error = SET_ERROR(ENOTBLK);
3448			vdev_free(vd);
3449			goto out;
3450		}
3451#endif
3452		vd->vdev_top = vd;
3453
3454		if ((error = vdev_open(vd)) == 0 &&
3455		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
3456			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3457			    vd->vdev_guid) == 0);
3458		}
3459
3460		vdev_free(vd);
3461
3462		if (error &&
3463		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3464			goto out;
3465		else
3466			error = 0;
3467	}
3468
3469out:
3470	sav->sav_pending = NULL;
3471	sav->sav_npending = 0;
3472	return (error);
3473}
3474
3475static int
3476spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3477{
3478	int error;
3479
3480	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3481
3482	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3483	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3484	    VDEV_LABEL_SPARE)) != 0) {
3485		return (error);
3486	}
3487
3488	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3489	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3490	    VDEV_LABEL_L2CACHE));
3491}
3492
3493static void
3494spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3495    const char *config)
3496{
3497	int i;
3498
3499	if (sav->sav_config != NULL) {
3500		nvlist_t **olddevs;
3501		uint_t oldndevs;
3502		nvlist_t **newdevs;
3503
3504		/*
3505		 * Generate new dev list by concatentating with the
3506		 * current dev list.
3507		 */
3508		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3509		    &olddevs, &oldndevs) == 0);
3510
3511		newdevs = kmem_alloc(sizeof (void *) *
3512		    (ndevs + oldndevs), KM_SLEEP);
3513		for (i = 0; i < oldndevs; i++)
3514			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3515			    KM_SLEEP) == 0);
3516		for (i = 0; i < ndevs; i++)
3517			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3518			    KM_SLEEP) == 0);
3519
3520		VERIFY(nvlist_remove(sav->sav_config, config,
3521		    DATA_TYPE_NVLIST_ARRAY) == 0);
3522
3523		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3524		    config, newdevs, ndevs + oldndevs) == 0);
3525		for (i = 0; i < oldndevs + ndevs; i++)
3526			nvlist_free(newdevs[i]);
3527		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3528	} else {
3529		/*
3530		 * Generate a new dev list.
3531		 */
3532		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3533		    KM_SLEEP) == 0);
3534		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3535		    devs, ndevs) == 0);
3536	}
3537}
3538
3539/*
3540 * Stop and drop level 2 ARC devices
3541 */
3542void
3543spa_l2cache_drop(spa_t *spa)
3544{
3545	vdev_t *vd;
3546	int i;
3547	spa_aux_vdev_t *sav = &spa->spa_l2cache;
3548
3549	for (i = 0; i < sav->sav_count; i++) {
3550		uint64_t pool;
3551
3552		vd = sav->sav_vdevs[i];
3553		ASSERT(vd != NULL);
3554
3555		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3556		    pool != 0ULL && l2arc_vdev_present(vd))
3557			l2arc_remove_vdev(vd);
3558	}
3559}
3560
3561/*
3562 * Pool Creation
3563 */
3564int
3565spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3566    nvlist_t *zplprops)
3567{
3568	spa_t *spa;
3569	char *altroot = NULL;
3570	vdev_t *rvd;
3571	dsl_pool_t *dp;
3572	dmu_tx_t *tx;
3573	int error = 0;
3574	uint64_t txg = TXG_INITIAL;
3575	nvlist_t **spares, **l2cache;
3576	uint_t nspares, nl2cache;
3577	uint64_t version, obj;
3578	boolean_t has_features;
3579
3580	/*
3581	 * If this pool already exists, return failure.
3582	 */
3583	mutex_enter(&spa_namespace_lock);
3584	if (spa_lookup(pool) != NULL) {
3585		mutex_exit(&spa_namespace_lock);
3586		return (SET_ERROR(EEXIST));
3587	}
3588
3589	/*
3590	 * Allocate a new spa_t structure.
3591	 */
3592	(void) nvlist_lookup_string(props,
3593	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3594	spa = spa_add(pool, NULL, altroot);
3595	spa_activate(spa, spa_mode_global);
3596
3597	if (props && (error = spa_prop_validate(spa, props))) {
3598		spa_deactivate(spa);
3599		spa_remove(spa);
3600		mutex_exit(&spa_namespace_lock);
3601		return (error);
3602	}
3603
3604	has_features = B_FALSE;
3605	for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3606	    elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3607		if (zpool_prop_feature(nvpair_name(elem)))
3608			has_features = B_TRUE;
3609	}
3610
3611	if (has_features || nvlist_lookup_uint64(props,
3612	    zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3613		version = SPA_VERSION;
3614	}
3615	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3616
3617	spa->spa_first_txg = txg;
3618	spa->spa_uberblock.ub_txg = txg - 1;
3619	spa->spa_uberblock.ub_version = version;
3620	spa->spa_ubsync = spa->spa_uberblock;
3621
3622	/*
3623	 * Create "The Godfather" zio to hold all async IOs
3624	 */
3625	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
3626	    KM_SLEEP);
3627	for (int i = 0; i < max_ncpus; i++) {
3628		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
3629		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
3630		    ZIO_FLAG_GODFATHER);
3631	}
3632
3633	/*
3634	 * Create the root vdev.
3635	 */
3636	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3637
3638	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3639
3640	ASSERT(error != 0 || rvd != NULL);
3641	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3642
3643	if (error == 0 && !zfs_allocatable_devs(nvroot))
3644		error = SET_ERROR(EINVAL);
3645
3646	if (error == 0 &&
3647	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3648	    (error = spa_validate_aux(spa, nvroot, txg,
3649	    VDEV_ALLOC_ADD)) == 0) {
3650		for (int c = 0; c < rvd->vdev_children; c++) {
3651			vdev_metaslab_set_size(rvd->vdev_child[c]);
3652			vdev_expand(rvd->vdev_child[c], txg);
3653		}
3654	}
3655
3656	spa_config_exit(spa, SCL_ALL, FTAG);
3657
3658	if (error != 0) {
3659		spa_unload(spa);
3660		spa_deactivate(spa);
3661		spa_remove(spa);
3662		mutex_exit(&spa_namespace_lock);
3663		return (error);
3664	}
3665
3666	/*
3667	 * Get the list of spares, if specified.
3668	 */
3669	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3670	    &spares, &nspares) == 0) {
3671		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3672		    KM_SLEEP) == 0);
3673		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3674		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3675		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3676		spa_load_spares(spa);
3677		spa_config_exit(spa, SCL_ALL, FTAG);
3678		spa->spa_spares.sav_sync = B_TRUE;
3679	}
3680
3681	/*
3682	 * Get the list of level 2 cache devices, if specified.
3683	 */
3684	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3685	    &l2cache, &nl2cache) == 0) {
3686		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3687		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3688		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3689		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3690		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3691		spa_load_l2cache(spa);
3692		spa_config_exit(spa, SCL_ALL, FTAG);
3693		spa->spa_l2cache.sav_sync = B_TRUE;
3694	}
3695
3696	spa->spa_is_initializing = B_TRUE;
3697	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3698	spa->spa_meta_objset = dp->dp_meta_objset;
3699	spa->spa_is_initializing = B_FALSE;
3700
3701	/*
3702	 * Create DDTs (dedup tables).
3703	 */
3704	ddt_create(spa);
3705
3706	spa_update_dspace(spa);
3707
3708	tx = dmu_tx_create_assigned(dp, txg);
3709
3710	/*
3711	 * Create the pool config object.
3712	 */
3713	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3714	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3715	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3716
3717	if (zap_add(spa->spa_meta_objset,
3718	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3719	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3720		cmn_err(CE_PANIC, "failed to add pool config");
3721	}
3722
3723	if (spa_version(spa) >= SPA_VERSION_FEATURES)
3724		spa_feature_create_zap_objects(spa, tx);
3725
3726	if (zap_add(spa->spa_meta_objset,
3727	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3728	    sizeof (uint64_t), 1, &version, tx) != 0) {
3729		cmn_err(CE_PANIC, "failed to add pool version");
3730	}
3731
3732	/* Newly created pools with the right version are always deflated. */
3733	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3734		spa->spa_deflate = TRUE;
3735		if (zap_add(spa->spa_meta_objset,
3736		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3737		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3738			cmn_err(CE_PANIC, "failed to add deflate");
3739		}
3740	}
3741
3742	/*
3743	 * Create the deferred-free bpobj.  Turn off compression
3744	 * because sync-to-convergence takes longer if the blocksize
3745	 * keeps changing.
3746	 */
3747	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3748	dmu_object_set_compress(spa->spa_meta_objset, obj,
3749	    ZIO_COMPRESS_OFF, tx);
3750	if (zap_add(spa->spa_meta_objset,
3751	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3752	    sizeof (uint64_t), 1, &obj, tx) != 0) {
3753		cmn_err(CE_PANIC, "failed to add bpobj");
3754	}
3755	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3756	    spa->spa_meta_objset, obj));
3757
3758	/*
3759	 * Create the pool's history object.
3760	 */
3761	if (version >= SPA_VERSION_ZPOOL_HISTORY)
3762		spa_history_create_obj(spa, tx);
3763
3764	/*
3765	 * Generate some random noise for salted checksums to operate on.
3766	 */
3767	(void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
3768	    sizeof (spa->spa_cksum_salt.zcs_bytes));
3769
3770	/*
3771	 * Set pool properties.
3772	 */
3773	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3774	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3775	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3776	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3777
3778	if (props != NULL) {
3779		spa_configfile_set(spa, props, B_FALSE);
3780		spa_sync_props(props, tx);
3781	}
3782
3783	dmu_tx_commit(tx);
3784
3785	spa->spa_sync_on = B_TRUE;
3786	txg_sync_start(spa->spa_dsl_pool);
3787
3788	/*
3789	 * We explicitly wait for the first transaction to complete so that our
3790	 * bean counters are appropriately updated.
3791	 */
3792	txg_wait_synced(spa->spa_dsl_pool, txg);
3793
3794	spa_config_sync(spa, B_FALSE, B_TRUE);
3795	spa_event_notify(spa, NULL, ESC_ZFS_POOL_CREATE);
3796
3797	spa_history_log_version(spa, "create");
3798
3799	/*
3800	 * Don't count references from objsets that are already closed
3801	 * and are making their way through the eviction process.
3802	 */
3803	spa_evicting_os_wait(spa);
3804	spa->spa_minref = refcount_count(&spa->spa_refcount);
3805
3806	mutex_exit(&spa_namespace_lock);
3807
3808	return (0);
3809}
3810
3811#ifdef _KERNEL
3812/*
3813 * Get the root pool information from the root disk, then import the root pool
3814 * during the system boot up time.
3815 */
3816extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3817
3818static nvlist_t *
3819spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3820{
3821	nvlist_t *config;
3822	nvlist_t *nvtop, *nvroot;
3823	uint64_t pgid;
3824
3825	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3826		return (NULL);
3827
3828	/*
3829	 * Add this top-level vdev to the child array.
3830	 */
3831	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3832	    &nvtop) == 0);
3833	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3834	    &pgid) == 0);
3835	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3836
3837	/*
3838	 * Put this pool's top-level vdevs into a root vdev.
3839	 */
3840	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3841	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3842	    VDEV_TYPE_ROOT) == 0);
3843	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3844	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3845	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3846	    &nvtop, 1) == 0);
3847
3848	/*
3849	 * Replace the existing vdev_tree with the new root vdev in
3850	 * this pool's configuration (remove the old, add the new).
3851	 */
3852	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3853	nvlist_free(nvroot);
3854	return (config);
3855}
3856
3857/*
3858 * Walk the vdev tree and see if we can find a device with "better"
3859 * configuration. A configuration is "better" if the label on that
3860 * device has a more recent txg.
3861 */
3862static void
3863spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3864{
3865	for (int c = 0; c < vd->vdev_children; c++)
3866		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3867
3868	if (vd->vdev_ops->vdev_op_leaf) {
3869		nvlist_t *label;
3870		uint64_t label_txg;
3871
3872		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3873		    &label) != 0)
3874			return;
3875
3876		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3877		    &label_txg) == 0);
3878
3879		/*
3880		 * Do we have a better boot device?
3881		 */
3882		if (label_txg > *txg) {
3883			*txg = label_txg;
3884			*avd = vd;
3885		}
3886		nvlist_free(label);
3887	}
3888}
3889
3890/*
3891 * Import a root pool.
3892 *
3893 * For x86. devpath_list will consist of devid and/or physpath name of
3894 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3895 * The GRUB "findroot" command will return the vdev we should boot.
3896 *
3897 * For Sparc, devpath_list consists the physpath name of the booting device
3898 * no matter the rootpool is a single device pool or a mirrored pool.
3899 * e.g.
3900 *	"/pci@1f,0/ide@d/disk@0,0:a"
3901 */
3902int
3903spa_import_rootpool(char *devpath, char *devid)
3904{
3905	spa_t *spa;
3906	vdev_t *rvd, *bvd, *avd = NULL;
3907	nvlist_t *config, *nvtop;
3908	uint64_t guid, txg;
3909	char *pname;
3910	int error;
3911
3912	/*
3913	 * Read the label from the boot device and generate a configuration.
3914	 */
3915	config = spa_generate_rootconf(devpath, devid, &guid);
3916#if defined(_OBP) && defined(_KERNEL)
3917	if (config == NULL) {
3918		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3919			/* iscsi boot */
3920			get_iscsi_bootpath_phy(devpath);
3921			config = spa_generate_rootconf(devpath, devid, &guid);
3922		}
3923	}
3924#endif
3925	if (config == NULL) {
3926		cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3927		    devpath);
3928		return (SET_ERROR(EIO));
3929	}
3930
3931	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3932	    &pname) == 0);
3933	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3934
3935	mutex_enter(&spa_namespace_lock);
3936	if ((spa = spa_lookup(pname)) != NULL) {
3937		/*
3938		 * Remove the existing root pool from the namespace so that we
3939		 * can replace it with the correct config we just read in.
3940		 */
3941		spa_remove(spa);
3942	}
3943
3944	spa = spa_add(pname, config, NULL);
3945	spa->spa_is_root = B_TRUE;
3946	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3947
3948	/*
3949	 * Build up a vdev tree based on the boot device's label config.
3950	 */
3951	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3952	    &nvtop) == 0);
3953	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3954	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3955	    VDEV_ALLOC_ROOTPOOL);
3956	spa_config_exit(spa, SCL_ALL, FTAG);
3957	if (error) {
3958		mutex_exit(&spa_namespace_lock);
3959		nvlist_free(config);
3960		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3961		    pname);
3962		return (error);
3963	}
3964
3965	/*
3966	 * Get the boot vdev.
3967	 */
3968	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3969		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3970		    (u_longlong_t)guid);
3971		error = SET_ERROR(ENOENT);
3972		goto out;
3973	}
3974
3975	/*
3976	 * Determine if there is a better boot device.
3977	 */
3978	avd = bvd;
3979	spa_alt_rootvdev(rvd, &avd, &txg);
3980	if (avd != bvd) {
3981		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3982		    "try booting from '%s'", avd->vdev_path);
3983		error = SET_ERROR(EINVAL);
3984		goto out;
3985	}
3986
3987	/*
3988	 * If the boot device is part of a spare vdev then ensure that
3989	 * we're booting off the active spare.
3990	 */
3991	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3992	    !bvd->vdev_isspare) {
3993		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3994		    "try booting from '%s'",
3995		    bvd->vdev_parent->
3996		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3997		error = SET_ERROR(EINVAL);
3998		goto out;
3999	}
4000
4001	error = 0;
4002out:
4003	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4004	vdev_free(rvd);
4005	spa_config_exit(spa, SCL_ALL, FTAG);
4006	mutex_exit(&spa_namespace_lock);
4007
4008	nvlist_free(config);
4009	return (error);
4010}
4011
4012#endif
4013
4014/*
4015 * Import a non-root pool into the system.
4016 */
4017int
4018spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
4019{
4020	spa_t *spa;
4021	char *altroot = NULL;
4022	spa_load_state_t state = SPA_LOAD_IMPORT;
4023	zpool_rewind_policy_t policy;
4024	uint64_t mode = spa_mode_global;
4025	uint64_t readonly = B_FALSE;
4026	int error;
4027	nvlist_t *nvroot;
4028	nvlist_t **spares, **l2cache;
4029	uint_t nspares, nl2cache;
4030
4031	/*
4032	 * If a pool with this name exists, return failure.
4033	 */
4034	mutex_enter(&spa_namespace_lock);
4035	if (spa_lookup(pool) != NULL) {
4036		mutex_exit(&spa_namespace_lock);
4037		return (SET_ERROR(EEXIST));
4038	}
4039
4040	/*
4041	 * Create and initialize the spa structure.
4042	 */
4043	(void) nvlist_lookup_string(props,
4044	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4045	(void) nvlist_lookup_uint64(props,
4046	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
4047	if (readonly)
4048		mode = FREAD;
4049	spa = spa_add(pool, config, altroot);
4050	spa->spa_import_flags = flags;
4051
4052	/*
4053	 * Verbatim import - Take a pool and insert it into the namespace
4054	 * as if it had been loaded at boot.
4055	 */
4056	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
4057		if (props != NULL)
4058			spa_configfile_set(spa, props, B_FALSE);
4059
4060		spa_config_sync(spa, B_FALSE, B_TRUE);
4061		spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4062
4063		mutex_exit(&spa_namespace_lock);
4064		return (0);
4065	}
4066
4067	spa_activate(spa, mode);
4068
4069	/*
4070	 * Don't start async tasks until we know everything is healthy.
4071	 */
4072	spa_async_suspend(spa);
4073
4074	zpool_get_rewind_policy(config, &policy);
4075	if (policy.zrp_request & ZPOOL_DO_REWIND)
4076		state = SPA_LOAD_RECOVER;
4077
4078	/*
4079	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
4080	 * because the user-supplied config is actually the one to trust when
4081	 * doing an import.
4082	 */
4083	if (state != SPA_LOAD_RECOVER)
4084		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4085
4086	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4087	    policy.zrp_request);
4088
4089	/*
4090	 * Propagate anything learned while loading the pool and pass it
4091	 * back to caller (i.e. rewind info, missing devices, etc).
4092	 */
4093	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4094	    spa->spa_load_info) == 0);
4095
4096	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4097	/*
4098	 * Toss any existing sparelist, as it doesn't have any validity
4099	 * anymore, and conflicts with spa_has_spare().
4100	 */
4101	if (spa->spa_spares.sav_config) {
4102		nvlist_free(spa->spa_spares.sav_config);
4103		spa->spa_spares.sav_config = NULL;
4104		spa_load_spares(spa);
4105	}
4106	if (spa->spa_l2cache.sav_config) {
4107		nvlist_free(spa->spa_l2cache.sav_config);
4108		spa->spa_l2cache.sav_config = NULL;
4109		spa_load_l2cache(spa);
4110	}
4111
4112	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4113	    &nvroot) == 0);
4114	if (error == 0)
4115		error = spa_validate_aux(spa, nvroot, -1ULL,
4116		    VDEV_ALLOC_SPARE);
4117	if (error == 0)
4118		error = spa_validate_aux(spa, nvroot, -1ULL,
4119		    VDEV_ALLOC_L2CACHE);
4120	spa_config_exit(spa, SCL_ALL, FTAG);
4121
4122	if (props != NULL)
4123		spa_configfile_set(spa, props, B_FALSE);
4124
4125	if (error != 0 || (props && spa_writeable(spa) &&
4126	    (error = spa_prop_set(spa, props)))) {
4127		spa_unload(spa);
4128		spa_deactivate(spa);
4129		spa_remove(spa);
4130		mutex_exit(&spa_namespace_lock);
4131		return (error);
4132	}
4133
4134	spa_async_resume(spa);
4135
4136	/*
4137	 * Override any spares and level 2 cache devices as specified by
4138	 * the user, as these may have correct device names/devids, etc.
4139	 */
4140	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4141	    &spares, &nspares) == 0) {
4142		if (spa->spa_spares.sav_config)
4143			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4144			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4145		else
4146			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4147			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4148		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4149		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4150		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4151		spa_load_spares(spa);
4152		spa_config_exit(spa, SCL_ALL, FTAG);
4153		spa->spa_spares.sav_sync = B_TRUE;
4154	}
4155	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4156	    &l2cache, &nl2cache) == 0) {
4157		if (spa->spa_l2cache.sav_config)
4158			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4159			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4160		else
4161			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4162			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4163		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4164		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4165		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4166		spa_load_l2cache(spa);
4167		spa_config_exit(spa, SCL_ALL, FTAG);
4168		spa->spa_l2cache.sav_sync = B_TRUE;
4169	}
4170
4171	/*
4172	 * Check for any removed devices.
4173	 */
4174	if (spa->spa_autoreplace) {
4175		spa_aux_check_removed(&spa->spa_spares);
4176		spa_aux_check_removed(&spa->spa_l2cache);
4177	}
4178
4179	if (spa_writeable(spa)) {
4180		/*
4181		 * Update the config cache to include the newly-imported pool.
4182		 */
4183		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4184	}
4185
4186	/*
4187	 * It's possible that the pool was expanded while it was exported.
4188	 * We kick off an async task to handle this for us.
4189	 */
4190	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4191
4192	spa_history_log_version(spa, "import");
4193
4194	spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4195
4196	mutex_exit(&spa_namespace_lock);
4197
4198	return (0);
4199}
4200
4201nvlist_t *
4202spa_tryimport(nvlist_t *tryconfig)
4203{
4204	nvlist_t *config = NULL;
4205	char *poolname;
4206	spa_t *spa;
4207	uint64_t state;
4208	int error;
4209
4210	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4211		return (NULL);
4212
4213	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4214		return (NULL);
4215
4216	/*
4217	 * Create and initialize the spa structure.
4218	 */
4219	mutex_enter(&spa_namespace_lock);
4220	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4221	spa_activate(spa, FREAD);
4222
4223	/*
4224	 * Pass off the heavy lifting to spa_load().
4225	 * Pass TRUE for mosconfig because the user-supplied config
4226	 * is actually the one to trust when doing an import.
4227	 */
4228	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4229
4230	/*
4231	 * If 'tryconfig' was at least parsable, return the current config.
4232	 */
4233	if (spa->spa_root_vdev != NULL) {
4234		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4235		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4236		    poolname) == 0);
4237		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4238		    state) == 0);
4239		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4240		    spa->spa_uberblock.ub_timestamp) == 0);
4241		VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4242		    spa->spa_load_info) == 0);
4243
4244		/*
4245		 * If the bootfs property exists on this pool then we
4246		 * copy it out so that external consumers can tell which
4247		 * pools are bootable.
4248		 */
4249		if ((!error || error == EEXIST) && spa->spa_bootfs) {
4250			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4251
4252			/*
4253			 * We have to play games with the name since the
4254			 * pool was opened as TRYIMPORT_NAME.
4255			 */
4256			if (dsl_dsobj_to_dsname(spa_name(spa),
4257			    spa->spa_bootfs, tmpname) == 0) {
4258				char *cp;
4259				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4260
4261				cp = strchr(tmpname, '/');
4262				if (cp == NULL) {
4263					(void) strlcpy(dsname, tmpname,
4264					    MAXPATHLEN);
4265				} else {
4266					(void) snprintf(dsname, MAXPATHLEN,
4267					    "%s/%s", poolname, ++cp);
4268				}
4269				VERIFY(nvlist_add_string(config,
4270				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4271				kmem_free(dsname, MAXPATHLEN);
4272			}
4273			kmem_free(tmpname, MAXPATHLEN);
4274		}
4275
4276		/*
4277		 * Add the list of hot spares and level 2 cache devices.
4278		 */
4279		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4280		spa_add_spares(spa, config);
4281		spa_add_l2cache(spa, config);
4282		spa_config_exit(spa, SCL_CONFIG, FTAG);
4283	}
4284
4285	spa_unload(spa);
4286	spa_deactivate(spa);
4287	spa_remove(spa);
4288	mutex_exit(&spa_namespace_lock);
4289
4290	return (config);
4291}
4292
4293/*
4294 * Pool export/destroy
4295 *
4296 * The act of destroying or exporting a pool is very simple.  We make sure there
4297 * is no more pending I/O and any references to the pool are gone.  Then, we
4298 * update the pool state and sync all the labels to disk, removing the
4299 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4300 * we don't sync the labels or remove the configuration cache.
4301 */
4302static int
4303spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4304    boolean_t force, boolean_t hardforce)
4305{
4306	spa_t *spa;
4307
4308	if (oldconfig)
4309		*oldconfig = NULL;
4310
4311	if (!(spa_mode_global & FWRITE))
4312		return (SET_ERROR(EROFS));
4313
4314	mutex_enter(&spa_namespace_lock);
4315	if ((spa = spa_lookup(pool)) == NULL) {
4316		mutex_exit(&spa_namespace_lock);
4317		return (SET_ERROR(ENOENT));
4318	}
4319
4320	/*
4321	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4322	 * reacquire the namespace lock, and see if we can export.
4323	 */
4324	spa_open_ref(spa, FTAG);
4325	mutex_exit(&spa_namespace_lock);
4326	spa_async_suspend(spa);
4327	mutex_enter(&spa_namespace_lock);
4328	spa_close(spa, FTAG);
4329
4330	/*
4331	 * The pool will be in core if it's openable,
4332	 * in which case we can modify its state.
4333	 */
4334	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4335		/*
4336		 * Objsets may be open only because they're dirty, so we
4337		 * have to force it to sync before checking spa_refcnt.
4338		 */
4339		txg_wait_synced(spa->spa_dsl_pool, 0);
4340		spa_evicting_os_wait(spa);
4341
4342		/*
4343		 * A pool cannot be exported or destroyed if there are active
4344		 * references.  If we are resetting a pool, allow references by
4345		 * fault injection handlers.
4346		 */
4347		if (!spa_refcount_zero(spa) ||
4348		    (spa->spa_inject_ref != 0 &&
4349		    new_state != POOL_STATE_UNINITIALIZED)) {
4350			spa_async_resume(spa);
4351			mutex_exit(&spa_namespace_lock);
4352			return (SET_ERROR(EBUSY));
4353		}
4354
4355		/*
4356		 * A pool cannot be exported if it has an active shared spare.
4357		 * This is to prevent other pools stealing the active spare
4358		 * from an exported pool. At user's own will, such pool can
4359		 * be forcedly exported.
4360		 */
4361		if (!force && new_state == POOL_STATE_EXPORTED &&
4362		    spa_has_active_shared_spare(spa)) {
4363			spa_async_resume(spa);
4364			mutex_exit(&spa_namespace_lock);
4365			return (SET_ERROR(EXDEV));
4366		}
4367
4368		/*
4369		 * We want this to be reflected on every label,
4370		 * so mark them all dirty.  spa_unload() will do the
4371		 * final sync that pushes these changes out.
4372		 */
4373		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4374			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4375			spa->spa_state = new_state;
4376			spa->spa_final_txg = spa_last_synced_txg(spa) +
4377			    TXG_DEFER_SIZE + 1;
4378			vdev_config_dirty(spa->spa_root_vdev);
4379			spa_config_exit(spa, SCL_ALL, FTAG);
4380		}
4381	}
4382
4383	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4384
4385	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4386		spa_unload(spa);
4387		spa_deactivate(spa);
4388	}
4389
4390	if (oldconfig && spa->spa_config)
4391		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4392
4393	if (new_state != POOL_STATE_UNINITIALIZED) {
4394		if (!hardforce)
4395			spa_config_sync(spa, B_TRUE, B_TRUE);
4396		spa_remove(spa);
4397	}
4398	mutex_exit(&spa_namespace_lock);
4399
4400	return (0);
4401}
4402
4403/*
4404 * Destroy a storage pool.
4405 */
4406int
4407spa_destroy(char *pool)
4408{
4409	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4410	    B_FALSE, B_FALSE));
4411}
4412
4413/*
4414 * Export a storage pool.
4415 */
4416int
4417spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4418    boolean_t hardforce)
4419{
4420	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4421	    force, hardforce));
4422}
4423
4424/*
4425 * Similar to spa_export(), this unloads the spa_t without actually removing it
4426 * from the namespace in any way.
4427 */
4428int
4429spa_reset(char *pool)
4430{
4431	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4432	    B_FALSE, B_FALSE));
4433}
4434
4435/*
4436 * ==========================================================================
4437 * Device manipulation
4438 * ==========================================================================
4439 */
4440
4441/*
4442 * Add a device to a storage pool.
4443 */
4444int
4445spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4446{
4447	uint64_t txg, id;
4448	int error;
4449	vdev_t *rvd = spa->spa_root_vdev;
4450	vdev_t *vd, *tvd;
4451	nvlist_t **spares, **l2cache;
4452	uint_t nspares, nl2cache;
4453
4454	ASSERT(spa_writeable(spa));
4455
4456	txg = spa_vdev_enter(spa);
4457
4458	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4459	    VDEV_ALLOC_ADD)) != 0)
4460		return (spa_vdev_exit(spa, NULL, txg, error));
4461
4462	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
4463
4464	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4465	    &nspares) != 0)
4466		nspares = 0;
4467
4468	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4469	    &nl2cache) != 0)
4470		nl2cache = 0;
4471
4472	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4473		return (spa_vdev_exit(spa, vd, txg, EINVAL));
4474
4475	if (vd->vdev_children != 0 &&
4476	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
4477		return (spa_vdev_exit(spa, vd, txg, error));
4478
4479	/*
4480	 * We must validate the spares and l2cache devices after checking the
4481	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
4482	 */
4483	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4484		return (spa_vdev_exit(spa, vd, txg, error));
4485
4486	/*
4487	 * Transfer each new top-level vdev from vd to rvd.
4488	 */
4489	for (int c = 0; c < vd->vdev_children; c++) {
4490
4491		/*
4492		 * Set the vdev id to the first hole, if one exists.
4493		 */
4494		for (id = 0; id < rvd->vdev_children; id++) {
4495			if (rvd->vdev_child[id]->vdev_ishole) {
4496				vdev_free(rvd->vdev_child[id]);
4497				break;
4498			}
4499		}
4500		tvd = vd->vdev_child[c];
4501		vdev_remove_child(vd, tvd);
4502		tvd->vdev_id = id;
4503		vdev_add_child(rvd, tvd);
4504		vdev_config_dirty(tvd);
4505	}
4506
4507	if (nspares != 0) {
4508		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4509		    ZPOOL_CONFIG_SPARES);
4510		spa_load_spares(spa);
4511		spa->spa_spares.sav_sync = B_TRUE;
4512	}
4513
4514	if (nl2cache != 0) {
4515		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4516		    ZPOOL_CONFIG_L2CACHE);
4517		spa_load_l2cache(spa);
4518		spa->spa_l2cache.sav_sync = B_TRUE;
4519	}
4520
4521	/*
4522	 * We have to be careful when adding new vdevs to an existing pool.
4523	 * If other threads start allocating from these vdevs before we
4524	 * sync the config cache, and we lose power, then upon reboot we may
4525	 * fail to open the pool because there are DVAs that the config cache
4526	 * can't translate.  Therefore, we first add the vdevs without
4527	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4528	 * and then let spa_config_update() initialize the new metaslabs.
4529	 *
4530	 * spa_load() checks for added-but-not-initialized vdevs, so that
4531	 * if we lose power at any point in this sequence, the remaining
4532	 * steps will be completed the next time we load the pool.
4533	 */
4534	(void) spa_vdev_exit(spa, vd, txg, 0);
4535
4536	mutex_enter(&spa_namespace_lock);
4537	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4538	spa_event_notify(spa, NULL, ESC_ZFS_VDEV_ADD);
4539	mutex_exit(&spa_namespace_lock);
4540
4541	return (0);
4542}
4543
4544/*
4545 * Attach a device to a mirror.  The arguments are the path to any device
4546 * in the mirror, and the nvroot for the new device.  If the path specifies
4547 * a device that is not mirrored, we automatically insert the mirror vdev.
4548 *
4549 * If 'replacing' is specified, the new device is intended to replace the
4550 * existing device; in this case the two devices are made into their own
4551 * mirror using the 'replacing' vdev, which is functionally identical to
4552 * the mirror vdev (it actually reuses all the same ops) but has a few
4553 * extra rules: you can't attach to it after it's been created, and upon
4554 * completion of resilvering, the first disk (the one being replaced)
4555 * is automatically detached.
4556 */
4557int
4558spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4559{
4560	uint64_t txg, dtl_max_txg;
4561	vdev_t *rvd = spa->spa_root_vdev;
4562	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4563	vdev_ops_t *pvops;
4564	char *oldvdpath, *newvdpath;
4565	int newvd_isspare;
4566	int error;
4567
4568	ASSERT(spa_writeable(spa));
4569
4570	txg = spa_vdev_enter(spa);
4571
4572	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4573
4574	if (oldvd == NULL)
4575		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4576
4577	if (!oldvd->vdev_ops->vdev_op_leaf)
4578		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4579
4580	pvd = oldvd->vdev_parent;
4581
4582	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4583	    VDEV_ALLOC_ATTACH)) != 0)
4584		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4585
4586	if (newrootvd->vdev_children != 1)
4587		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4588
4589	newvd = newrootvd->vdev_child[0];
4590
4591	if (!newvd->vdev_ops->vdev_op_leaf)
4592		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4593
4594	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4595		return (spa_vdev_exit(spa, newrootvd, txg, error));
4596
4597	/*
4598	 * Spares can't replace logs
4599	 */
4600	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4601		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4602
4603	if (!replacing) {
4604		/*
4605		 * For attach, the only allowable parent is a mirror or the root
4606		 * vdev.
4607		 */
4608		if (pvd->vdev_ops != &vdev_mirror_ops &&
4609		    pvd->vdev_ops != &vdev_root_ops)
4610			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4611
4612		pvops = &vdev_mirror_ops;
4613	} else {
4614		/*
4615		 * Active hot spares can only be replaced by inactive hot
4616		 * spares.
4617		 */
4618		if (pvd->vdev_ops == &vdev_spare_ops &&
4619		    oldvd->vdev_isspare &&
4620		    !spa_has_spare(spa, newvd->vdev_guid))
4621			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4622
4623		/*
4624		 * If the source is a hot spare, and the parent isn't already a
4625		 * spare, then we want to create a new hot spare.  Otherwise, we
4626		 * want to create a replacing vdev.  The user is not allowed to
4627		 * attach to a spared vdev child unless the 'isspare' state is
4628		 * the same (spare replaces spare, non-spare replaces
4629		 * non-spare).
4630		 */
4631		if (pvd->vdev_ops == &vdev_replacing_ops &&
4632		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4633			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4634		} else if (pvd->vdev_ops == &vdev_spare_ops &&
4635		    newvd->vdev_isspare != oldvd->vdev_isspare) {
4636			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4637		}
4638
4639		if (newvd->vdev_isspare)
4640			pvops = &vdev_spare_ops;
4641		else
4642			pvops = &vdev_replacing_ops;
4643	}
4644
4645	/*
4646	 * Make sure the new device is big enough.
4647	 */
4648	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4649		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4650
4651	/*
4652	 * The new device cannot have a higher alignment requirement
4653	 * than the top-level vdev.
4654	 */
4655	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4656		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4657
4658	/*
4659	 * If this is an in-place replacement, update oldvd's path and devid
4660	 * to make it distinguishable from newvd, and unopenable from now on.
4661	 */
4662	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4663		spa_strfree(oldvd->vdev_path);
4664		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4665		    KM_SLEEP);
4666		(void) sprintf(oldvd->vdev_path, "%s/%s",
4667		    newvd->vdev_path, "old");
4668		if (oldvd->vdev_devid != NULL) {
4669			spa_strfree(oldvd->vdev_devid);
4670			oldvd->vdev_devid = NULL;
4671		}
4672	}
4673
4674	/* mark the device being resilvered */
4675	newvd->vdev_resilver_txg = txg;
4676
4677	/*
4678	 * If the parent is not a mirror, or if we're replacing, insert the new
4679	 * mirror/replacing/spare vdev above oldvd.
4680	 */
4681	if (pvd->vdev_ops != pvops)
4682		pvd = vdev_add_parent(oldvd, pvops);
4683
4684	ASSERT(pvd->vdev_top->vdev_parent == rvd);
4685	ASSERT(pvd->vdev_ops == pvops);
4686	ASSERT(oldvd->vdev_parent == pvd);
4687
4688	/*
4689	 * Extract the new device from its root and add it to pvd.
4690	 */
4691	vdev_remove_child(newrootvd, newvd);
4692	newvd->vdev_id = pvd->vdev_children;
4693	newvd->vdev_crtxg = oldvd->vdev_crtxg;
4694	vdev_add_child(pvd, newvd);
4695
4696	tvd = newvd->vdev_top;
4697	ASSERT(pvd->vdev_top == tvd);
4698	ASSERT(tvd->vdev_parent == rvd);
4699
4700	vdev_config_dirty(tvd);
4701
4702	/*
4703	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4704	 * for any dmu_sync-ed blocks.  It will propagate upward when
4705	 * spa_vdev_exit() calls vdev_dtl_reassess().
4706	 */
4707	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4708
4709	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4710	    dtl_max_txg - TXG_INITIAL);
4711
4712	if (newvd->vdev_isspare) {
4713		spa_spare_activate(newvd);
4714		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4715	}
4716
4717	oldvdpath = spa_strdup(oldvd->vdev_path);
4718	newvdpath = spa_strdup(newvd->vdev_path);
4719	newvd_isspare = newvd->vdev_isspare;
4720
4721	/*
4722	 * Mark newvd's DTL dirty in this txg.
4723	 */
4724	vdev_dirty(tvd, VDD_DTL, newvd, txg);
4725
4726	/*
4727	 * Schedule the resilver to restart in the future. We do this to
4728	 * ensure that dmu_sync-ed blocks have been stitched into the
4729	 * respective datasets.
4730	 */
4731	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4732
4733	if (spa->spa_bootfs)
4734		spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4735
4736	spa_event_notify(spa, newvd, ESC_ZFS_VDEV_ATTACH);
4737
4738	/*
4739	 * Commit the config
4740	 */
4741	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4742
4743	spa_history_log_internal(spa, "vdev attach", NULL,
4744	    "%s vdev=%s %s vdev=%s",
4745	    replacing && newvd_isspare ? "spare in" :
4746	    replacing ? "replace" : "attach", newvdpath,
4747	    replacing ? "for" : "to", oldvdpath);
4748
4749	spa_strfree(oldvdpath);
4750	spa_strfree(newvdpath);
4751
4752	return (0);
4753}
4754
4755/*
4756 * Detach a device from a mirror or replacing vdev.
4757 *
4758 * If 'replace_done' is specified, only detach if the parent
4759 * is a replacing vdev.
4760 */
4761int
4762spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4763{
4764	uint64_t txg;
4765	int error;
4766	vdev_t *rvd = spa->spa_root_vdev;
4767	vdev_t *vd, *pvd, *cvd, *tvd;
4768	boolean_t unspare = B_FALSE;
4769	uint64_t unspare_guid = 0;
4770	char *vdpath;
4771
4772	ASSERT(spa_writeable(spa));
4773
4774	txg = spa_vdev_enter(spa);
4775
4776	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4777
4778	if (vd == NULL)
4779		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4780
4781	if (!vd->vdev_ops->vdev_op_leaf)
4782		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4783
4784	pvd = vd->vdev_parent;
4785
4786	/*
4787	 * If the parent/child relationship is not as expected, don't do it.
4788	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4789	 * vdev that's replacing B with C.  The user's intent in replacing
4790	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4791	 * the replace by detaching C, the expected behavior is to end up
4792	 * M(A,B).  But suppose that right after deciding to detach C,
4793	 * the replacement of B completes.  We would have M(A,C), and then
4794	 * ask to detach C, which would leave us with just A -- not what
4795	 * the user wanted.  To prevent this, we make sure that the
4796	 * parent/child relationship hasn't changed -- in this example,
4797	 * that C's parent is still the replacing vdev R.
4798	 */
4799	if (pvd->vdev_guid != pguid && pguid != 0)
4800		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4801
4802	/*
4803	 * Only 'replacing' or 'spare' vdevs can be replaced.
4804	 */
4805	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4806	    pvd->vdev_ops != &vdev_spare_ops)
4807		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4808
4809	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4810	    spa_version(spa) >= SPA_VERSION_SPARES);
4811
4812	/*
4813	 * Only mirror, replacing, and spare vdevs support detach.
4814	 */
4815	if (pvd->vdev_ops != &vdev_replacing_ops &&
4816	    pvd->vdev_ops != &vdev_mirror_ops &&
4817	    pvd->vdev_ops != &vdev_spare_ops)
4818		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4819
4820	/*
4821	 * If this device has the only valid copy of some data,
4822	 * we cannot safely detach it.
4823	 */
4824	if (vdev_dtl_required(vd))
4825		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4826
4827	ASSERT(pvd->vdev_children >= 2);
4828
4829	/*
4830	 * If we are detaching the second disk from a replacing vdev, then
4831	 * check to see if we changed the original vdev's path to have "/old"
4832	 * at the end in spa_vdev_attach().  If so, undo that change now.
4833	 */
4834	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4835	    vd->vdev_path != NULL) {
4836		size_t len = strlen(vd->vdev_path);
4837
4838		for (int c = 0; c < pvd->vdev_children; c++) {
4839			cvd = pvd->vdev_child[c];
4840
4841			if (cvd == vd || cvd->vdev_path == NULL)
4842				continue;
4843
4844			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4845			    strcmp(cvd->vdev_path + len, "/old") == 0) {
4846				spa_strfree(cvd->vdev_path);
4847				cvd->vdev_path = spa_strdup(vd->vdev_path);
4848				break;
4849			}
4850		}
4851	}
4852
4853	/*
4854	 * If we are detaching the original disk from a spare, then it implies
4855	 * that the spare should become a real disk, and be removed from the
4856	 * active spare list for the pool.
4857	 */
4858	if (pvd->vdev_ops == &vdev_spare_ops &&
4859	    vd->vdev_id == 0 &&
4860	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4861		unspare = B_TRUE;
4862
4863	/*
4864	 * Erase the disk labels so the disk can be used for other things.
4865	 * This must be done after all other error cases are handled,
4866	 * but before we disembowel vd (so we can still do I/O to it).
4867	 * But if we can't do it, don't treat the error as fatal --
4868	 * it may be that the unwritability of the disk is the reason
4869	 * it's being detached!
4870	 */
4871	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4872
4873	/*
4874	 * Remove vd from its parent and compact the parent's children.
4875	 */
4876	vdev_remove_child(pvd, vd);
4877	vdev_compact_children(pvd);
4878
4879	/*
4880	 * Remember one of the remaining children so we can get tvd below.
4881	 */
4882	cvd = pvd->vdev_child[pvd->vdev_children - 1];
4883
4884	/*
4885	 * If we need to remove the remaining child from the list of hot spares,
4886	 * do it now, marking the vdev as no longer a spare in the process.
4887	 * We must do this before vdev_remove_parent(), because that can
4888	 * change the GUID if it creates a new toplevel GUID.  For a similar
4889	 * reason, we must remove the spare now, in the same txg as the detach;
4890	 * otherwise someone could attach a new sibling, change the GUID, and
4891	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4892	 */
4893	if (unspare) {
4894		ASSERT(cvd->vdev_isspare);
4895		spa_spare_remove(cvd);
4896		unspare_guid = cvd->vdev_guid;
4897		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4898		cvd->vdev_unspare = B_TRUE;
4899	}
4900
4901	/*
4902	 * If the parent mirror/replacing vdev only has one child,
4903	 * the parent is no longer needed.  Remove it from the tree.
4904	 */
4905	if (pvd->vdev_children == 1) {
4906		if (pvd->vdev_ops == &vdev_spare_ops)
4907			cvd->vdev_unspare = B_FALSE;
4908		vdev_remove_parent(cvd);
4909	}
4910
4911
4912	/*
4913	 * We don't set tvd until now because the parent we just removed
4914	 * may have been the previous top-level vdev.
4915	 */
4916	tvd = cvd->vdev_top;
4917	ASSERT(tvd->vdev_parent == rvd);
4918
4919	/*
4920	 * Reevaluate the parent vdev state.
4921	 */
4922	vdev_propagate_state(cvd);
4923
4924	/*
4925	 * If the 'autoexpand' property is set on the pool then automatically
4926	 * try to expand the size of the pool. For example if the device we
4927	 * just detached was smaller than the others, it may be possible to
4928	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4929	 * first so that we can obtain the updated sizes of the leaf vdevs.
4930	 */
4931	if (spa->spa_autoexpand) {
4932		vdev_reopen(tvd);
4933		vdev_expand(tvd, txg);
4934	}
4935
4936	vdev_config_dirty(tvd);
4937
4938	/*
4939	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4940	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4941	 * But first make sure we're not on any *other* txg's DTL list, to
4942	 * prevent vd from being accessed after it's freed.
4943	 */
4944	vdpath = spa_strdup(vd->vdev_path);
4945	for (int t = 0; t < TXG_SIZE; t++)
4946		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4947	vd->vdev_detached = B_TRUE;
4948	vdev_dirty(tvd, VDD_DTL, vd, txg);
4949
4950	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4951
4952	/* hang on to the spa before we release the lock */
4953	spa_open_ref(spa, FTAG);
4954
4955	error = spa_vdev_exit(spa, vd, txg, 0);
4956
4957	spa_history_log_internal(spa, "detach", NULL,
4958	    "vdev=%s", vdpath);
4959	spa_strfree(vdpath);
4960
4961	/*
4962	 * If this was the removal of the original device in a hot spare vdev,
4963	 * then we want to go through and remove the device from the hot spare
4964	 * list of every other pool.
4965	 */
4966	if (unspare) {
4967		spa_t *altspa = NULL;
4968
4969		mutex_enter(&spa_namespace_lock);
4970		while ((altspa = spa_next(altspa)) != NULL) {
4971			if (altspa->spa_state != POOL_STATE_ACTIVE ||
4972			    altspa == spa)
4973				continue;
4974
4975			spa_open_ref(altspa, FTAG);
4976			mutex_exit(&spa_namespace_lock);
4977			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4978			mutex_enter(&spa_namespace_lock);
4979			spa_close(altspa, FTAG);
4980		}
4981		mutex_exit(&spa_namespace_lock);
4982
4983		/* search the rest of the vdevs for spares to remove */
4984		spa_vdev_resilver_done(spa);
4985	}
4986
4987	/* all done with the spa; OK to release */
4988	mutex_enter(&spa_namespace_lock);
4989	spa_close(spa, FTAG);
4990	mutex_exit(&spa_namespace_lock);
4991
4992	return (error);
4993}
4994
4995/*
4996 * Split a set of devices from their mirrors, and create a new pool from them.
4997 */
4998int
4999spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
5000    nvlist_t *props, boolean_t exp)
5001{
5002	int error = 0;
5003	uint64_t txg, *glist;
5004	spa_t *newspa;
5005	uint_t c, children, lastlog;
5006	nvlist_t **child, *nvl, *tmp;
5007	dmu_tx_t *tx;
5008	char *altroot = NULL;
5009	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
5010	boolean_t activate_slog;
5011
5012	ASSERT(spa_writeable(spa));
5013
5014	txg = spa_vdev_enter(spa);
5015
5016	/* clear the log and flush everything up to now */
5017	activate_slog = spa_passivate_log(spa);
5018	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5019	error = spa_offline_log(spa);
5020	txg = spa_vdev_config_enter(spa);
5021
5022	if (activate_slog)
5023		spa_activate_log(spa);
5024
5025	if (error != 0)
5026		return (spa_vdev_exit(spa, NULL, txg, error));
5027
5028	/* check new spa name before going any further */
5029	if (spa_lookup(newname) != NULL)
5030		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
5031
5032	/*
5033	 * scan through all the children to ensure they're all mirrors
5034	 */
5035	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
5036	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
5037	    &children) != 0)
5038		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5039
5040	/* first, check to ensure we've got the right child count */
5041	rvd = spa->spa_root_vdev;
5042	lastlog = 0;
5043	for (c = 0; c < rvd->vdev_children; c++) {
5044		vdev_t *vd = rvd->vdev_child[c];
5045
5046		/* don't count the holes & logs as children */
5047		if (vd->vdev_islog || vd->vdev_ishole) {
5048			if (lastlog == 0)
5049				lastlog = c;
5050			continue;
5051		}
5052
5053		lastlog = 0;
5054	}
5055	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
5056		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5057
5058	/* next, ensure no spare or cache devices are part of the split */
5059	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
5060	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
5061		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5062
5063	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5064	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5065
5066	/* then, loop over each vdev and validate it */
5067	for (c = 0; c < children; c++) {
5068		uint64_t is_hole = 0;
5069
5070		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5071		    &is_hole);
5072
5073		if (is_hole != 0) {
5074			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5075			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5076				continue;
5077			} else {
5078				error = SET_ERROR(EINVAL);
5079				break;
5080			}
5081		}
5082
5083		/* which disk is going to be split? */
5084		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5085		    &glist[c]) != 0) {
5086			error = SET_ERROR(EINVAL);
5087			break;
5088		}
5089
5090		/* look it up in the spa */
5091		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5092		if (vml[c] == NULL) {
5093			error = SET_ERROR(ENODEV);
5094			break;
5095		}
5096
5097		/* make sure there's nothing stopping the split */
5098		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5099		    vml[c]->vdev_islog ||
5100		    vml[c]->vdev_ishole ||
5101		    vml[c]->vdev_isspare ||
5102		    vml[c]->vdev_isl2cache ||
5103		    !vdev_writeable(vml[c]) ||
5104		    vml[c]->vdev_children != 0 ||
5105		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5106		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5107			error = SET_ERROR(EINVAL);
5108			break;
5109		}
5110
5111		if (vdev_dtl_required(vml[c])) {
5112			error = SET_ERROR(EBUSY);
5113			break;
5114		}
5115
5116		/* we need certain info from the top level */
5117		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5118		    vml[c]->vdev_top->vdev_ms_array) == 0);
5119		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5120		    vml[c]->vdev_top->vdev_ms_shift) == 0);
5121		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5122		    vml[c]->vdev_top->vdev_asize) == 0);
5123		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5124		    vml[c]->vdev_top->vdev_ashift) == 0);
5125
5126		/* transfer per-vdev ZAPs */
5127		ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0);
5128		VERIFY0(nvlist_add_uint64(child[c],
5129		    ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap));
5130
5131		ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0);
5132		VERIFY0(nvlist_add_uint64(child[c],
5133		    ZPOOL_CONFIG_VDEV_TOP_ZAP,
5134		    vml[c]->vdev_parent->vdev_top_zap));
5135	}
5136
5137	if (error != 0) {
5138		kmem_free(vml, children * sizeof (vdev_t *));
5139		kmem_free(glist, children * sizeof (uint64_t));
5140		return (spa_vdev_exit(spa, NULL, txg, error));
5141	}
5142
5143	/* stop writers from using the disks */
5144	for (c = 0; c < children; c++) {
5145		if (vml[c] != NULL)
5146			vml[c]->vdev_offline = B_TRUE;
5147	}
5148	vdev_reopen(spa->spa_root_vdev);
5149
5150	/*
5151	 * Temporarily record the splitting vdevs in the spa config.  This
5152	 * will disappear once the config is regenerated.
5153	 */
5154	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5155	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5156	    glist, children) == 0);
5157	kmem_free(glist, children * sizeof (uint64_t));
5158
5159	mutex_enter(&spa->spa_props_lock);
5160	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5161	    nvl) == 0);
5162	mutex_exit(&spa->spa_props_lock);
5163	spa->spa_config_splitting = nvl;
5164	vdev_config_dirty(spa->spa_root_vdev);
5165
5166	/* configure and create the new pool */
5167	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5168	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5169	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5170	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5171	    spa_version(spa)) == 0);
5172	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5173	    spa->spa_config_txg) == 0);
5174	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5175	    spa_generate_guid(NULL)) == 0);
5176	VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
5177	(void) nvlist_lookup_string(props,
5178	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5179
5180	/* add the new pool to the namespace */
5181	newspa = spa_add(newname, config, altroot);
5182	newspa->spa_avz_action = AVZ_ACTION_REBUILD;
5183	newspa->spa_config_txg = spa->spa_config_txg;
5184	spa_set_log_state(newspa, SPA_LOG_CLEAR);
5185
5186	/* release the spa config lock, retaining the namespace lock */
5187	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5188
5189	if (zio_injection_enabled)
5190		zio_handle_panic_injection(spa, FTAG, 1);
5191
5192	spa_activate(newspa, spa_mode_global);
5193	spa_async_suspend(newspa);
5194
5195	/* create the new pool from the disks of the original pool */
5196	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5197	if (error)
5198		goto out;
5199
5200	/* if that worked, generate a real config for the new pool */
5201	if (newspa->spa_root_vdev != NULL) {
5202		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5203		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
5204		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5205		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5206		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5207		    B_TRUE));
5208	}
5209
5210	/* set the props */
5211	if (props != NULL) {
5212		spa_configfile_set(newspa, props, B_FALSE);
5213		error = spa_prop_set(newspa, props);
5214		if (error)
5215			goto out;
5216	}
5217
5218	/* flush everything */
5219	txg = spa_vdev_config_enter(newspa);
5220	vdev_config_dirty(newspa->spa_root_vdev);
5221	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5222
5223	if (zio_injection_enabled)
5224		zio_handle_panic_injection(spa, FTAG, 2);
5225
5226	spa_async_resume(newspa);
5227
5228	/* finally, update the original pool's config */
5229	txg = spa_vdev_config_enter(spa);
5230	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5231	error = dmu_tx_assign(tx, TXG_WAIT);
5232	if (error != 0)
5233		dmu_tx_abort(tx);
5234	for (c = 0; c < children; c++) {
5235		if (vml[c] != NULL) {
5236			vdev_split(vml[c]);
5237			if (error == 0)
5238				spa_history_log_internal(spa, "detach", tx,
5239				    "vdev=%s", vml[c]->vdev_path);
5240
5241			vdev_free(vml[c]);
5242		}
5243	}
5244	spa->spa_avz_action = AVZ_ACTION_REBUILD;
5245	vdev_config_dirty(spa->spa_root_vdev);
5246	spa->spa_config_splitting = NULL;
5247	nvlist_free(nvl);
5248	if (error == 0)
5249		dmu_tx_commit(tx);
5250	(void) spa_vdev_exit(spa, NULL, txg, 0);
5251
5252	if (zio_injection_enabled)
5253		zio_handle_panic_injection(spa, FTAG, 3);
5254
5255	/* split is complete; log a history record */
5256	spa_history_log_internal(newspa, "split", NULL,
5257	    "from pool %s", spa_name(spa));
5258
5259	kmem_free(vml, children * sizeof (vdev_t *));
5260
5261	/* if we're not going to mount the filesystems in userland, export */
5262	if (exp)
5263		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5264		    B_FALSE, B_FALSE);
5265
5266	return (error);
5267
5268out:
5269	spa_unload(newspa);
5270	spa_deactivate(newspa);
5271	spa_remove(newspa);
5272
5273	txg = spa_vdev_config_enter(spa);
5274
5275	/* re-online all offlined disks */
5276	for (c = 0; c < children; c++) {
5277		if (vml[c] != NULL)
5278			vml[c]->vdev_offline = B_FALSE;
5279	}
5280	vdev_reopen(spa->spa_root_vdev);
5281
5282	nvlist_free(spa->spa_config_splitting);
5283	spa->spa_config_splitting = NULL;
5284	(void) spa_vdev_exit(spa, NULL, txg, error);
5285
5286	kmem_free(vml, children * sizeof (vdev_t *));
5287	return (error);
5288}
5289
5290static nvlist_t *
5291spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5292{
5293	for (int i = 0; i < count; i++) {
5294		uint64_t guid;
5295
5296		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5297		    &guid) == 0);
5298
5299		if (guid == target_guid)
5300			return (nvpp[i]);
5301	}
5302
5303	return (NULL);
5304}
5305
5306static void
5307spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5308	nvlist_t *dev_to_remove)
5309{
5310	nvlist_t **newdev = NULL;
5311
5312	if (count > 1)
5313		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5314
5315	for (int i = 0, j = 0; i < count; i++) {
5316		if (dev[i] == dev_to_remove)
5317			continue;
5318		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5319	}
5320
5321	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5322	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5323
5324	for (int i = 0; i < count - 1; i++)
5325		nvlist_free(newdev[i]);
5326
5327	if (count > 1)
5328		kmem_free(newdev, (count - 1) * sizeof (void *));
5329}
5330
5331/*
5332 * Evacuate the device.
5333 */
5334static int
5335spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5336{
5337	uint64_t txg;
5338	int error = 0;
5339
5340	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5341	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5342	ASSERT(vd == vd->vdev_top);
5343
5344	/*
5345	 * Evacuate the device.  We don't hold the config lock as writer
5346	 * since we need to do I/O but we do keep the
5347	 * spa_namespace_lock held.  Once this completes the device
5348	 * should no longer have any blocks allocated on it.
5349	 */
5350	if (vd->vdev_islog) {
5351		if (vd->vdev_stat.vs_alloc != 0)
5352			error = spa_offline_log(spa);
5353	} else {
5354		error = SET_ERROR(ENOTSUP);
5355	}
5356
5357	if (error)
5358		return (error);
5359
5360	/*
5361	 * The evacuation succeeded.  Remove any remaining MOS metadata
5362	 * associated with this vdev, and wait for these changes to sync.
5363	 */
5364	ASSERT0(vd->vdev_stat.vs_alloc);
5365	txg = spa_vdev_config_enter(spa);
5366	vd->vdev_removing = B_TRUE;
5367	vdev_dirty_leaves(vd, VDD_DTL, txg);
5368	vdev_config_dirty(vd);
5369	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5370
5371	return (0);
5372}
5373
5374/*
5375 * Complete the removal by cleaning up the namespace.
5376 */
5377static void
5378spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5379{
5380	vdev_t *rvd = spa->spa_root_vdev;
5381	uint64_t id = vd->vdev_id;
5382	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5383
5384	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5385	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5386	ASSERT(vd == vd->vdev_top);
5387
5388	/*
5389	 * Only remove any devices which are empty.
5390	 */
5391	if (vd->vdev_stat.vs_alloc != 0)
5392		return;
5393
5394	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5395
5396	if (list_link_active(&vd->vdev_state_dirty_node))
5397		vdev_state_clean(vd);
5398	if (list_link_active(&vd->vdev_config_dirty_node))
5399		vdev_config_clean(vd);
5400
5401	vdev_free(vd);
5402
5403	if (last_vdev) {
5404		vdev_compact_children(rvd);
5405	} else {
5406		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5407		vdev_add_child(rvd, vd);
5408	}
5409	vdev_config_dirty(rvd);
5410
5411	/*
5412	 * Reassess the health of our root vdev.
5413	 */
5414	vdev_reopen(rvd);
5415}
5416
5417/*
5418 * Remove a device from the pool -
5419 *
5420 * Removing a device from the vdev namespace requires several steps
5421 * and can take a significant amount of time.  As a result we use
5422 * the spa_vdev_config_[enter/exit] functions which allow us to
5423 * grab and release the spa_config_lock while still holding the namespace
5424 * lock.  During each step the configuration is synced out.
5425 *
5426 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5427 * devices.
5428 */
5429int
5430spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5431{
5432	vdev_t *vd;
5433	metaslab_group_t *mg;
5434	nvlist_t **spares, **l2cache, *nv;
5435	uint64_t txg = 0;
5436	uint_t nspares, nl2cache;
5437	int error = 0;
5438	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5439
5440	ASSERT(spa_writeable(spa));
5441
5442	if (!locked)
5443		txg = spa_vdev_enter(spa);
5444
5445	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5446
5447	if (spa->spa_spares.sav_vdevs != NULL &&
5448	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5449	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5450	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5451		/*
5452		 * Only remove the hot spare if it's not currently in use
5453		 * in this pool.
5454		 */
5455		if (vd == NULL || unspare) {
5456			spa_vdev_remove_aux(spa->spa_spares.sav_config,
5457			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5458			spa_load_spares(spa);
5459			spa->spa_spares.sav_sync = B_TRUE;
5460		} else {
5461			error = SET_ERROR(EBUSY);
5462		}
5463	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
5464	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5465	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5466	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5467		/*
5468		 * Cache devices can always be removed.
5469		 */
5470		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5471		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5472		spa_load_l2cache(spa);
5473		spa->spa_l2cache.sav_sync = B_TRUE;
5474	} else if (vd != NULL && vd->vdev_islog) {
5475		ASSERT(!locked);
5476		ASSERT(vd == vd->vdev_top);
5477
5478		mg = vd->vdev_mg;
5479
5480		/*
5481		 * Stop allocating from this vdev.
5482		 */
5483		metaslab_group_passivate(mg);
5484
5485		/*
5486		 * Wait for the youngest allocations and frees to sync,
5487		 * and then wait for the deferral of those frees to finish.
5488		 */
5489		spa_vdev_config_exit(spa, NULL,
5490		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5491
5492		/*
5493		 * Attempt to evacuate the vdev.
5494		 */
5495		error = spa_vdev_remove_evacuate(spa, vd);
5496
5497		txg = spa_vdev_config_enter(spa);
5498
5499		/*
5500		 * If we couldn't evacuate the vdev, unwind.
5501		 */
5502		if (error) {
5503			metaslab_group_activate(mg);
5504			return (spa_vdev_exit(spa, NULL, txg, error));
5505		}
5506
5507		/*
5508		 * Clean up the vdev namespace.
5509		 */
5510		spa_vdev_remove_from_namespace(spa, vd);
5511
5512	} else if (vd != NULL) {
5513		/*
5514		 * Normal vdevs cannot be removed (yet).
5515		 */
5516		error = SET_ERROR(ENOTSUP);
5517	} else {
5518		/*
5519		 * There is no vdev of any kind with the specified guid.
5520		 */
5521		error = SET_ERROR(ENOENT);
5522	}
5523
5524	if (!locked)
5525		return (spa_vdev_exit(spa, NULL, txg, error));
5526
5527	return (error);
5528}
5529
5530/*
5531 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5532 * currently spared, so we can detach it.
5533 */
5534static vdev_t *
5535spa_vdev_resilver_done_hunt(vdev_t *vd)
5536{
5537	vdev_t *newvd, *oldvd;
5538
5539	for (int c = 0; c < vd->vdev_children; c++) {
5540		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5541		if (oldvd != NULL)
5542			return (oldvd);
5543	}
5544
5545	/*
5546	 * Check for a completed replacement.  We always consider the first
5547	 * vdev in the list to be the oldest vdev, and the last one to be
5548	 * the newest (see spa_vdev_attach() for how that works).  In
5549	 * the case where the newest vdev is faulted, we will not automatically
5550	 * remove it after a resilver completes.  This is OK as it will require
5551	 * user intervention to determine which disk the admin wishes to keep.
5552	 */
5553	if (vd->vdev_ops == &vdev_replacing_ops) {
5554		ASSERT(vd->vdev_children > 1);
5555
5556		newvd = vd->vdev_child[vd->vdev_children - 1];
5557		oldvd = vd->vdev_child[0];
5558
5559		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5560		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5561		    !vdev_dtl_required(oldvd))
5562			return (oldvd);
5563	}
5564
5565	/*
5566	 * Check for a completed resilver with the 'unspare' flag set.
5567	 */
5568	if (vd->vdev_ops == &vdev_spare_ops) {
5569		vdev_t *first = vd->vdev_child[0];
5570		vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5571
5572		if (last->vdev_unspare) {
5573			oldvd = first;
5574			newvd = last;
5575		} else if (first->vdev_unspare) {
5576			oldvd = last;
5577			newvd = first;
5578		} else {
5579			oldvd = NULL;
5580		}
5581
5582		if (oldvd != NULL &&
5583		    vdev_dtl_empty(newvd, DTL_MISSING) &&
5584		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5585		    !vdev_dtl_required(oldvd))
5586			return (oldvd);
5587
5588		/*
5589		 * If there are more than two spares attached to a disk,
5590		 * and those spares are not required, then we want to
5591		 * attempt to free them up now so that they can be used
5592		 * by other pools.  Once we're back down to a single
5593		 * disk+spare, we stop removing them.
5594		 */
5595		if (vd->vdev_children > 2) {
5596			newvd = vd->vdev_child[1];
5597
5598			if (newvd->vdev_isspare && last->vdev_isspare &&
5599			    vdev_dtl_empty(last, DTL_MISSING) &&
5600			    vdev_dtl_empty(last, DTL_OUTAGE) &&
5601			    !vdev_dtl_required(newvd))
5602				return (newvd);
5603		}
5604	}
5605
5606	return (NULL);
5607}
5608
5609static void
5610spa_vdev_resilver_done(spa_t *spa)
5611{
5612	vdev_t *vd, *pvd, *ppvd;
5613	uint64_t guid, sguid, pguid, ppguid;
5614
5615	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5616
5617	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5618		pvd = vd->vdev_parent;
5619		ppvd = pvd->vdev_parent;
5620		guid = vd->vdev_guid;
5621		pguid = pvd->vdev_guid;
5622		ppguid = ppvd->vdev_guid;
5623		sguid = 0;
5624		/*
5625		 * If we have just finished replacing a hot spared device, then
5626		 * we need to detach the parent's first child (the original hot
5627		 * spare) as well.
5628		 */
5629		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5630		    ppvd->vdev_children == 2) {
5631			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5632			sguid = ppvd->vdev_child[1]->vdev_guid;
5633		}
5634		ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5635
5636		spa_config_exit(spa, SCL_ALL, FTAG);
5637		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5638			return;
5639		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5640			return;
5641		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5642	}
5643
5644	spa_config_exit(spa, SCL_ALL, FTAG);
5645}
5646
5647/*
5648 * Update the stored path or FRU for this vdev.
5649 */
5650int
5651spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5652    boolean_t ispath)
5653{
5654	vdev_t *vd;
5655	boolean_t sync = B_FALSE;
5656
5657	ASSERT(spa_writeable(spa));
5658
5659	spa_vdev_state_enter(spa, SCL_ALL);
5660
5661	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5662		return (spa_vdev_state_exit(spa, NULL, ENOENT));
5663
5664	if (!vd->vdev_ops->vdev_op_leaf)
5665		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5666
5667	if (ispath) {
5668		if (strcmp(value, vd->vdev_path) != 0) {
5669			spa_strfree(vd->vdev_path);
5670			vd->vdev_path = spa_strdup(value);
5671			sync = B_TRUE;
5672		}
5673	} else {
5674		if (vd->vdev_fru == NULL) {
5675			vd->vdev_fru = spa_strdup(value);
5676			sync = B_TRUE;
5677		} else if (strcmp(value, vd->vdev_fru) != 0) {
5678			spa_strfree(vd->vdev_fru);
5679			vd->vdev_fru = spa_strdup(value);
5680			sync = B_TRUE;
5681		}
5682	}
5683
5684	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5685}
5686
5687int
5688spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5689{
5690	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5691}
5692
5693int
5694spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5695{
5696	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5697}
5698
5699/*
5700 * ==========================================================================
5701 * SPA Scanning
5702 * ==========================================================================
5703 */
5704
5705int
5706spa_scan_stop(spa_t *spa)
5707{
5708	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5709	if (dsl_scan_resilvering(spa->spa_dsl_pool))
5710		return (SET_ERROR(EBUSY));
5711	return (dsl_scan_cancel(spa->spa_dsl_pool));
5712}
5713
5714int
5715spa_scan(spa_t *spa, pool_scan_func_t func)
5716{
5717	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5718
5719	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5720		return (SET_ERROR(ENOTSUP));
5721
5722	/*
5723	 * If a resilver was requested, but there is no DTL on a
5724	 * writeable leaf device, we have nothing to do.
5725	 */
5726	if (func == POOL_SCAN_RESILVER &&
5727	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5728		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5729		return (0);
5730	}
5731
5732	return (dsl_scan(spa->spa_dsl_pool, func));
5733}
5734
5735/*
5736 * ==========================================================================
5737 * SPA async task processing
5738 * ==========================================================================
5739 */
5740
5741static void
5742spa_async_remove(spa_t *spa, vdev_t *vd)
5743{
5744	if (vd->vdev_remove_wanted) {
5745		vd->vdev_remove_wanted = B_FALSE;
5746		vd->vdev_delayed_close = B_FALSE;
5747		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5748
5749		/*
5750		 * We want to clear the stats, but we don't want to do a full
5751		 * vdev_clear() as that will cause us to throw away
5752		 * degraded/faulted state as well as attempt to reopen the
5753		 * device, all of which is a waste.
5754		 */
5755		vd->vdev_stat.vs_read_errors = 0;
5756		vd->vdev_stat.vs_write_errors = 0;
5757		vd->vdev_stat.vs_checksum_errors = 0;
5758
5759		vdev_state_dirty(vd->vdev_top);
5760	}
5761
5762	for (int c = 0; c < vd->vdev_children; c++)
5763		spa_async_remove(spa, vd->vdev_child[c]);
5764}
5765
5766static void
5767spa_async_probe(spa_t *spa, vdev_t *vd)
5768{
5769	if (vd->vdev_probe_wanted) {
5770		vd->vdev_probe_wanted = B_FALSE;
5771		vdev_reopen(vd);	/* vdev_open() does the actual probe */
5772	}
5773
5774	for (int c = 0; c < vd->vdev_children; c++)
5775		spa_async_probe(spa, vd->vdev_child[c]);
5776}
5777
5778static void
5779spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5780{
5781	sysevent_id_t eid;
5782	nvlist_t *attr;
5783	char *physpath;
5784
5785	if (!spa->spa_autoexpand)
5786		return;
5787
5788	for (int c = 0; c < vd->vdev_children; c++) {
5789		vdev_t *cvd = vd->vdev_child[c];
5790		spa_async_autoexpand(spa, cvd);
5791	}
5792
5793	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5794		return;
5795
5796	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5797	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5798
5799	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5800	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5801
5802	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5803	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5804
5805	nvlist_free(attr);
5806	kmem_free(physpath, MAXPATHLEN);
5807}
5808
5809static void
5810spa_async_thread(spa_t *spa)
5811{
5812	int tasks;
5813
5814	ASSERT(spa->spa_sync_on);
5815
5816	mutex_enter(&spa->spa_async_lock);
5817	tasks = spa->spa_async_tasks;
5818	spa->spa_async_tasks = 0;
5819	mutex_exit(&spa->spa_async_lock);
5820
5821	/*
5822	 * See if the config needs to be updated.
5823	 */
5824	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5825		uint64_t old_space, new_space;
5826
5827		mutex_enter(&spa_namespace_lock);
5828		old_space = metaslab_class_get_space(spa_normal_class(spa));
5829		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5830		new_space = metaslab_class_get_space(spa_normal_class(spa));
5831		mutex_exit(&spa_namespace_lock);
5832
5833		/*
5834		 * If the pool grew as a result of the config update,
5835		 * then log an internal history event.
5836		 */
5837		if (new_space != old_space) {
5838			spa_history_log_internal(spa, "vdev online", NULL,
5839			    "pool '%s' size: %llu(+%llu)",
5840			    spa_name(spa), new_space, new_space - old_space);
5841		}
5842	}
5843
5844	/*
5845	 * See if any devices need to be marked REMOVED.
5846	 */
5847	if (tasks & SPA_ASYNC_REMOVE) {
5848		spa_vdev_state_enter(spa, SCL_NONE);
5849		spa_async_remove(spa, spa->spa_root_vdev);
5850		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5851			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5852		for (int i = 0; i < spa->spa_spares.sav_count; i++)
5853			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5854		(void) spa_vdev_state_exit(spa, NULL, 0);
5855	}
5856
5857	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5858		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5859		spa_async_autoexpand(spa, spa->spa_root_vdev);
5860		spa_config_exit(spa, SCL_CONFIG, FTAG);
5861	}
5862
5863	/*
5864	 * See if any devices need to be probed.
5865	 */
5866	if (tasks & SPA_ASYNC_PROBE) {
5867		spa_vdev_state_enter(spa, SCL_NONE);
5868		spa_async_probe(spa, spa->spa_root_vdev);
5869		(void) spa_vdev_state_exit(spa, NULL, 0);
5870	}
5871
5872	/*
5873	 * If any devices are done replacing, detach them.
5874	 */
5875	if (tasks & SPA_ASYNC_RESILVER_DONE)
5876		spa_vdev_resilver_done(spa);
5877
5878	/*
5879	 * Kick off a resilver.
5880	 */
5881	if (tasks & SPA_ASYNC_RESILVER)
5882		dsl_resilver_restart(spa->spa_dsl_pool, 0);
5883
5884	/*
5885	 * Let the world know that we're done.
5886	 */
5887	mutex_enter(&spa->spa_async_lock);
5888	spa->spa_async_thread = NULL;
5889	cv_broadcast(&spa->spa_async_cv);
5890	mutex_exit(&spa->spa_async_lock);
5891	thread_exit();
5892}
5893
5894void
5895spa_async_suspend(spa_t *spa)
5896{
5897	mutex_enter(&spa->spa_async_lock);
5898	spa->spa_async_suspended++;
5899	while (spa->spa_async_thread != NULL)
5900		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5901	mutex_exit(&spa->spa_async_lock);
5902}
5903
5904void
5905spa_async_resume(spa_t *spa)
5906{
5907	mutex_enter(&spa->spa_async_lock);
5908	ASSERT(spa->spa_async_suspended != 0);
5909	spa->spa_async_suspended--;
5910	mutex_exit(&spa->spa_async_lock);
5911}
5912
5913static boolean_t
5914spa_async_tasks_pending(spa_t *spa)
5915{
5916	uint_t non_config_tasks;
5917	uint_t config_task;
5918	boolean_t config_task_suspended;
5919
5920	non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5921	config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5922	if (spa->spa_ccw_fail_time == 0) {
5923		config_task_suspended = B_FALSE;
5924	} else {
5925		config_task_suspended =
5926		    (gethrtime() - spa->spa_ccw_fail_time) <
5927		    (zfs_ccw_retry_interval * NANOSEC);
5928	}
5929
5930	return (non_config_tasks || (config_task && !config_task_suspended));
5931}
5932
5933static void
5934spa_async_dispatch(spa_t *spa)
5935{
5936	mutex_enter(&spa->spa_async_lock);
5937	if (spa_async_tasks_pending(spa) &&
5938	    !spa->spa_async_suspended &&
5939	    spa->spa_async_thread == NULL &&
5940	    rootdir != NULL)
5941		spa->spa_async_thread = thread_create(NULL, 0,
5942		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5943	mutex_exit(&spa->spa_async_lock);
5944}
5945
5946void
5947spa_async_request(spa_t *spa, int task)
5948{
5949	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5950	mutex_enter(&spa->spa_async_lock);
5951	spa->spa_async_tasks |= task;
5952	mutex_exit(&spa->spa_async_lock);
5953}
5954
5955/*
5956 * ==========================================================================
5957 * SPA syncing routines
5958 * ==========================================================================
5959 */
5960
5961static int
5962bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5963{
5964	bpobj_t *bpo = arg;
5965	bpobj_enqueue(bpo, bp, tx);
5966	return (0);
5967}
5968
5969static int
5970spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5971{
5972	zio_t *zio = arg;
5973
5974	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5975	    zio->io_flags));
5976	return (0);
5977}
5978
5979/*
5980 * Note: this simple function is not inlined to make it easier to dtrace the
5981 * amount of time spent syncing frees.
5982 */
5983static void
5984spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
5985{
5986	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5987	bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
5988	VERIFY(zio_wait(zio) == 0);
5989}
5990
5991/*
5992 * Note: this simple function is not inlined to make it easier to dtrace the
5993 * amount of time spent syncing deferred frees.
5994 */
5995static void
5996spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
5997{
5998	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5999	VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
6000	    spa_free_sync_cb, zio, tx), ==, 0);
6001	VERIFY0(zio_wait(zio));
6002}
6003
6004
6005static void
6006spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
6007{
6008	char *packed = NULL;
6009	size_t bufsize;
6010	size_t nvsize = 0;
6011	dmu_buf_t *db;
6012
6013	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
6014
6015	/*
6016	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6017	 * information.  This avoids the dmu_buf_will_dirty() path and
6018	 * saves us a pre-read to get data we don't actually care about.
6019	 */
6020	bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
6021	packed = kmem_alloc(bufsize, KM_SLEEP);
6022
6023	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
6024	    KM_SLEEP) == 0);
6025	bzero(packed +