spa.c revision 2bd7a8d078223b122d65fea49bb8641f858b1409
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 */
29
30/*
31 * SPA: Storage Pool Allocator
32 *
33 * This file contains all the routines used when modifying on-disk SPA state.
34 * This includes opening, importing, destroying, exporting a pool, and syncing a
35 * pool.
36 */
37
38#include <sys/zfs_context.h>
39#include <sys/fm/fs/zfs.h>
40#include <sys/spa_impl.h>
41#include <sys/zio.h>
42#include <sys/zio_checksum.h>
43#include <sys/dmu.h>
44#include <sys/dmu_tx.h>
45#include <sys/zap.h>
46#include <sys/zil.h>
47#include <sys/ddt.h>
48#include <sys/vdev_impl.h>
49#include <sys/metaslab.h>
50#include <sys/metaslab_impl.h>
51#include <sys/uberblock_impl.h>
52#include <sys/txg.h>
53#include <sys/avl.h>
54#include <sys/dmu_traverse.h>
55#include <sys/dmu_objset.h>
56#include <sys/unique.h>
57#include <sys/dsl_pool.h>
58#include <sys/dsl_dataset.h>
59#include <sys/dsl_dir.h>
60#include <sys/dsl_prop.h>
61#include <sys/dsl_synctask.h>
62#include <sys/fs/zfs.h>
63#include <sys/arc.h>
64#include <sys/callb.h>
65#include <sys/systeminfo.h>
66#include <sys/spa_boot.h>
67#include <sys/zfs_ioctl.h>
68#include <sys/dsl_scan.h>
69#include <sys/zfeature.h>
70#include <sys/dsl_destroy.h>
71
72#ifdef	_KERNEL
73#include <sys/bootprops.h>
74#include <sys/callb.h>
75#include <sys/cpupart.h>
76#include <sys/pool.h>
77#include <sys/sysdc.h>
78#include <sys/zone.h>
79#endif	/* _KERNEL */
80
81#include "zfs_prop.h"
82#include "zfs_comutil.h"
83
84/*
85 * The interval, in seconds, at which failed configuration cache file writes
86 * should be retried.
87 */
88static int zfs_ccw_retry_interval = 300;
89
90typedef enum zti_modes {
91	ZTI_MODE_FIXED,			/* value is # of threads (min 1) */
92	ZTI_MODE_BATCH,			/* cpu-intensive; value is ignored */
93	ZTI_MODE_NULL,			/* don't create a taskq */
94	ZTI_NMODES
95} zti_modes_t;
96
97#define	ZTI_P(n, q)	{ ZTI_MODE_FIXED, (n), (q) }
98#define	ZTI_BATCH	{ ZTI_MODE_BATCH, 0, 1 }
99#define	ZTI_NULL	{ ZTI_MODE_NULL, 0, 0 }
100
101#define	ZTI_N(n)	ZTI_P(n, 1)
102#define	ZTI_ONE		ZTI_N(1)
103
104typedef struct zio_taskq_info {
105	zti_modes_t zti_mode;
106	uint_t zti_value;
107	uint_t zti_count;
108} zio_taskq_info_t;
109
110static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
111	"issue", "issue_high", "intr", "intr_high"
112};
113
114/*
115 * This table defines the taskq settings for each ZFS I/O type. When
116 * initializing a pool, we use this table to create an appropriately sized
117 * taskq. Some operations are low volume and therefore have a small, static
118 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
119 * macros. Other operations process a large amount of data; the ZTI_BATCH
120 * macro causes us to create a taskq oriented for throughput. Some operations
121 * are so high frequency and short-lived that the taskq itself can become a a
122 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
123 * additional degree of parallelism specified by the number of threads per-
124 * taskq and the number of taskqs; when dispatching an event in this case, the
125 * particular taskq is chosen at random.
126 *
127 * The different taskq priorities are to handle the different contexts (issue
128 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
129 * need to be handled with minimum delay.
130 */
131const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
132	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
133	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* NULL */
134	{ ZTI_N(8),	ZTI_NULL,	ZTI_P(12, 8),	ZTI_NULL }, /* READ */
135	{ ZTI_BATCH,	ZTI_N(5),	ZTI_N(8),	ZTI_N(5) }, /* WRITE */
136	{ ZTI_P(12, 8),	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* FREE */
137	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* CLAIM */
138	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* IOCTL */
139};
140
141static void spa_sync_version(void *arg, dmu_tx_t *tx);
142static void spa_sync_props(void *arg, dmu_tx_t *tx);
143static boolean_t spa_has_active_shared_spare(spa_t *spa);
144static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
145    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
146    char **ereport);
147static void spa_vdev_resilver_done(spa_t *spa);
148
149uint_t		zio_taskq_batch_pct = 75;	/* 1 thread per cpu in pset */
150id_t		zio_taskq_psrset_bind = PS_NONE;
151boolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
152uint_t		zio_taskq_basedc = 80;		/* base duty cycle */
153
154boolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
155extern int	zfs_sync_pass_deferred_free;
156
157/*
158 * This (illegal) pool name is used when temporarily importing a spa_t in order
159 * to get the vdev stats associated with the imported devices.
160 */
161#define	TRYIMPORT_NAME	"$import"
162
163/*
164 * ==========================================================================
165 * SPA properties routines
166 * ==========================================================================
167 */
168
169/*
170 * Add a (source=src, propname=propval) list to an nvlist.
171 */
172static void
173spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
174    uint64_t intval, zprop_source_t src)
175{
176	const char *propname = zpool_prop_to_name(prop);
177	nvlist_t *propval;
178
179	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
180	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
181
182	if (strval != NULL)
183		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
184	else
185		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
186
187	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
188	nvlist_free(propval);
189}
190
191/*
192 * Get property values from the spa configuration.
193 */
194static void
195spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
196{
197	vdev_t *rvd = spa->spa_root_vdev;
198	dsl_pool_t *pool = spa->spa_dsl_pool;
199	uint64_t size, alloc, cap, version;
200	zprop_source_t src = ZPROP_SRC_NONE;
201	spa_config_dirent_t *dp;
202	metaslab_class_t *mc = spa_normal_class(spa);
203
204	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
205
206	if (rvd != NULL) {
207		alloc = metaslab_class_get_alloc(spa_normal_class(spa));
208		size = metaslab_class_get_space(spa_normal_class(spa));
209		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
210		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
211		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
212		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
213		    size - alloc, src);
214
215		spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
216		    metaslab_class_fragmentation(mc), src);
217		spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
218		    metaslab_class_expandable_space(mc), src);
219		spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
220		    (spa_mode(spa) == FREAD), src);
221
222		cap = (size == 0) ? 0 : (alloc * 100 / size);
223		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
224
225		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
226		    ddt_get_pool_dedup_ratio(spa), src);
227
228		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
229		    rvd->vdev_state, src);
230
231		version = spa_version(spa);
232		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
233			src = ZPROP_SRC_DEFAULT;
234		else
235			src = ZPROP_SRC_LOCAL;
236		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
237	}
238
239	if (pool != NULL) {
240		/*
241		 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
242		 * when opening pools before this version freedir will be NULL.
243		 */
244		if (pool->dp_free_dir != NULL) {
245			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
246			    dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
247			    src);
248		} else {
249			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
250			    NULL, 0, src);
251		}
252
253		if (pool->dp_leak_dir != NULL) {
254			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
255			    dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
256			    src);
257		} else {
258			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
259			    NULL, 0, src);
260		}
261	}
262
263	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
264
265	if (spa->spa_comment != NULL) {
266		spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
267		    0, ZPROP_SRC_LOCAL);
268	}
269
270	if (spa->spa_root != NULL)
271		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
272		    0, ZPROP_SRC_LOCAL);
273
274	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
275		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
276		    MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
277	} else {
278		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
279		    SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
280	}
281
282	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
283		if (dp->scd_path == NULL) {
284			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
285			    "none", 0, ZPROP_SRC_LOCAL);
286		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
287			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
288			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
289		}
290	}
291}
292
293/*
294 * Get zpool property values.
295 */
296int
297spa_prop_get(spa_t *spa, nvlist_t **nvp)
298{
299	objset_t *mos = spa->spa_meta_objset;
300	zap_cursor_t zc;
301	zap_attribute_t za;
302	int err;
303
304	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
305
306	mutex_enter(&spa->spa_props_lock);
307
308	/*
309	 * Get properties from the spa config.
310	 */
311	spa_prop_get_config(spa, nvp);
312
313	/* If no pool property object, no more prop to get. */
314	if (mos == NULL || spa->spa_pool_props_object == 0) {
315		mutex_exit(&spa->spa_props_lock);
316		return (0);
317	}
318
319	/*
320	 * Get properties from the MOS pool property object.
321	 */
322	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
323	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
324	    zap_cursor_advance(&zc)) {
325		uint64_t intval = 0;
326		char *strval = NULL;
327		zprop_source_t src = ZPROP_SRC_DEFAULT;
328		zpool_prop_t prop;
329
330		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
331			continue;
332
333		switch (za.za_integer_length) {
334		case 8:
335			/* integer property */
336			if (za.za_first_integer !=
337			    zpool_prop_default_numeric(prop))
338				src = ZPROP_SRC_LOCAL;
339
340			if (prop == ZPOOL_PROP_BOOTFS) {
341				dsl_pool_t *dp;
342				dsl_dataset_t *ds = NULL;
343
344				dp = spa_get_dsl(spa);
345				dsl_pool_config_enter(dp, FTAG);
346				if (err = dsl_dataset_hold_obj(dp,
347				    za.za_first_integer, FTAG, &ds)) {
348					dsl_pool_config_exit(dp, FTAG);
349					break;
350				}
351
352				strval = kmem_alloc(
353				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
354				    KM_SLEEP);
355				dsl_dataset_name(ds, strval);
356				dsl_dataset_rele(ds, FTAG);
357				dsl_pool_config_exit(dp, FTAG);
358			} else {
359				strval = NULL;
360				intval = za.za_first_integer;
361			}
362
363			spa_prop_add_list(*nvp, prop, strval, intval, src);
364
365			if (strval != NULL)
366				kmem_free(strval,
367				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
368
369			break;
370
371		case 1:
372			/* string property */
373			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
374			err = zap_lookup(mos, spa->spa_pool_props_object,
375			    za.za_name, 1, za.za_num_integers, strval);
376			if (err) {
377				kmem_free(strval, za.za_num_integers);
378				break;
379			}
380			spa_prop_add_list(*nvp, prop, strval, 0, src);
381			kmem_free(strval, za.za_num_integers);
382			break;
383
384		default:
385			break;
386		}
387	}
388	zap_cursor_fini(&zc);
389	mutex_exit(&spa->spa_props_lock);
390out:
391	if (err && err != ENOENT) {
392		nvlist_free(*nvp);
393		*nvp = NULL;
394		return (err);
395	}
396
397	return (0);
398}
399
400/*
401 * Validate the given pool properties nvlist and modify the list
402 * for the property values to be set.
403 */
404static int
405spa_prop_validate(spa_t *spa, nvlist_t *props)
406{
407	nvpair_t *elem;
408	int error = 0, reset_bootfs = 0;
409	uint64_t objnum = 0;
410	boolean_t has_feature = B_FALSE;
411
412	elem = NULL;
413	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
414		uint64_t intval;
415		char *strval, *slash, *check, *fname;
416		const char *propname = nvpair_name(elem);
417		zpool_prop_t prop = zpool_name_to_prop(propname);
418
419		switch (prop) {
420		case ZPROP_INVAL:
421			if (!zpool_prop_feature(propname)) {
422				error = SET_ERROR(EINVAL);
423				break;
424			}
425
426			/*
427			 * Sanitize the input.
428			 */
429			if (nvpair_type(elem) != DATA_TYPE_UINT64) {
430				error = SET_ERROR(EINVAL);
431				break;
432			}
433
434			if (nvpair_value_uint64(elem, &intval) != 0) {
435				error = SET_ERROR(EINVAL);
436				break;
437			}
438
439			if (intval != 0) {
440				error = SET_ERROR(EINVAL);
441				break;
442			}
443
444			fname = strchr(propname, '@') + 1;
445			if (zfeature_lookup_name(fname, NULL) != 0) {
446				error = SET_ERROR(EINVAL);
447				break;
448			}
449
450			has_feature = B_TRUE;
451			break;
452
453		case ZPOOL_PROP_VERSION:
454			error = nvpair_value_uint64(elem, &intval);
455			if (!error &&
456			    (intval < spa_version(spa) ||
457			    intval > SPA_VERSION_BEFORE_FEATURES ||
458			    has_feature))
459				error = SET_ERROR(EINVAL);
460			break;
461
462		case ZPOOL_PROP_DELEGATION:
463		case ZPOOL_PROP_AUTOREPLACE:
464		case ZPOOL_PROP_LISTSNAPS:
465		case ZPOOL_PROP_AUTOEXPAND:
466			error = nvpair_value_uint64(elem, &intval);
467			if (!error && intval > 1)
468				error = SET_ERROR(EINVAL);
469			break;
470
471		case ZPOOL_PROP_BOOTFS:
472			/*
473			 * If the pool version is less than SPA_VERSION_BOOTFS,
474			 * or the pool is still being created (version == 0),
475			 * the bootfs property cannot be set.
476			 */
477			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
478				error = SET_ERROR(ENOTSUP);
479				break;
480			}
481
482			/*
483			 * Make sure the vdev config is bootable
484			 */
485			if (!vdev_is_bootable(spa->spa_root_vdev)) {
486				error = SET_ERROR(ENOTSUP);
487				break;
488			}
489
490			reset_bootfs = 1;
491
492			error = nvpair_value_string(elem, &strval);
493
494			if (!error) {
495				objset_t *os;
496				uint64_t propval;
497
498				if (strval == NULL || strval[0] == '\0') {
499					objnum = zpool_prop_default_numeric(
500					    ZPOOL_PROP_BOOTFS);
501					break;
502				}
503
504				if (error = dmu_objset_hold(strval, FTAG, &os))
505					break;
506
507				/*
508				 * Must be ZPL, and its property settings
509				 * must be supported by GRUB (compression
510				 * is not gzip, and large blocks are not used).
511				 */
512
513				if (dmu_objset_type(os) != DMU_OST_ZFS) {
514					error = SET_ERROR(ENOTSUP);
515				} else if ((error =
516				    dsl_prop_get_int_ds(dmu_objset_ds(os),
517				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
518				    &propval)) == 0 &&
519				    !BOOTFS_COMPRESS_VALID(propval)) {
520					error = SET_ERROR(ENOTSUP);
521				} else if ((error =
522				    dsl_prop_get_int_ds(dmu_objset_ds(os),
523				    zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
524				    &propval)) == 0 &&
525				    propval > SPA_OLD_MAXBLOCKSIZE) {
526					error = SET_ERROR(ENOTSUP);
527				} else {
528					objnum = dmu_objset_id(os);
529				}
530				dmu_objset_rele(os, FTAG);
531			}
532			break;
533
534		case ZPOOL_PROP_FAILUREMODE:
535			error = nvpair_value_uint64(elem, &intval);
536			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
537			    intval > ZIO_FAILURE_MODE_PANIC))
538				error = SET_ERROR(EINVAL);
539
540			/*
541			 * This is a special case which only occurs when
542			 * the pool has completely failed. This allows
543			 * the user to change the in-core failmode property
544			 * without syncing it out to disk (I/Os might
545			 * currently be blocked). We do this by returning
546			 * EIO to the caller (spa_prop_set) to trick it
547			 * into thinking we encountered a property validation
548			 * error.
549			 */
550			if (!error && spa_suspended(spa)) {
551				spa->spa_failmode = intval;
552				error = SET_ERROR(EIO);
553			}
554			break;
555
556		case ZPOOL_PROP_CACHEFILE:
557			if ((error = nvpair_value_string(elem, &strval)) != 0)
558				break;
559
560			if (strval[0] == '\0')
561				break;
562
563			if (strcmp(strval, "none") == 0)
564				break;
565
566			if (strval[0] != '/') {
567				error = SET_ERROR(EINVAL);
568				break;
569			}
570
571			slash = strrchr(strval, '/');
572			ASSERT(slash != NULL);
573
574			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
575			    strcmp(slash, "/..") == 0)
576				error = SET_ERROR(EINVAL);
577			break;
578
579		case ZPOOL_PROP_COMMENT:
580			if ((error = nvpair_value_string(elem, &strval)) != 0)
581				break;
582			for (check = strval; *check != '\0'; check++) {
583				/*
584				 * The kernel doesn't have an easy isprint()
585				 * check.  For this kernel check, we merely
586				 * check ASCII apart from DEL.  Fix this if
587				 * there is an easy-to-use kernel isprint().
588				 */
589				if (*check >= 0x7f) {
590					error = SET_ERROR(EINVAL);
591					break;
592				}
593			}
594			if (strlen(strval) > ZPROP_MAX_COMMENT)
595				error = E2BIG;
596			break;
597
598		case ZPOOL_PROP_DEDUPDITTO:
599			if (spa_version(spa) < SPA_VERSION_DEDUP)
600				error = SET_ERROR(ENOTSUP);
601			else
602				error = nvpair_value_uint64(elem, &intval);
603			if (error == 0 &&
604			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
605				error = SET_ERROR(EINVAL);
606			break;
607		}
608
609		if (error)
610			break;
611	}
612
613	if (!error && reset_bootfs) {
614		error = nvlist_remove(props,
615		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
616
617		if (!error) {
618			error = nvlist_add_uint64(props,
619			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
620		}
621	}
622
623	return (error);
624}
625
626void
627spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
628{
629	char *cachefile;
630	spa_config_dirent_t *dp;
631
632	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
633	    &cachefile) != 0)
634		return;
635
636	dp = kmem_alloc(sizeof (spa_config_dirent_t),
637	    KM_SLEEP);
638
639	if (cachefile[0] == '\0')
640		dp->scd_path = spa_strdup(spa_config_path);
641	else if (strcmp(cachefile, "none") == 0)
642		dp->scd_path = NULL;
643	else
644		dp->scd_path = spa_strdup(cachefile);
645
646	list_insert_head(&spa->spa_config_list, dp);
647	if (need_sync)
648		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
649}
650
651int
652spa_prop_set(spa_t *spa, nvlist_t *nvp)
653{
654	int error;
655	nvpair_t *elem = NULL;
656	boolean_t need_sync = B_FALSE;
657
658	if ((error = spa_prop_validate(spa, nvp)) != 0)
659		return (error);
660
661	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
662		zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
663
664		if (prop == ZPOOL_PROP_CACHEFILE ||
665		    prop == ZPOOL_PROP_ALTROOT ||
666		    prop == ZPOOL_PROP_READONLY)
667			continue;
668
669		if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
670			uint64_t ver;
671
672			if (prop == ZPOOL_PROP_VERSION) {
673				VERIFY(nvpair_value_uint64(elem, &ver) == 0);
674			} else {
675				ASSERT(zpool_prop_feature(nvpair_name(elem)));
676				ver = SPA_VERSION_FEATURES;
677				need_sync = B_TRUE;
678			}
679
680			/* Save time if the version is already set. */
681			if (ver == spa_version(spa))
682				continue;
683
684			/*
685			 * In addition to the pool directory object, we might
686			 * create the pool properties object, the features for
687			 * read object, the features for write object, or the
688			 * feature descriptions object.
689			 */
690			error = dsl_sync_task(spa->spa_name, NULL,
691			    spa_sync_version, &ver,
692			    6, ZFS_SPACE_CHECK_RESERVED);
693			if (error)
694				return (error);
695			continue;
696		}
697
698		need_sync = B_TRUE;
699		break;
700	}
701
702	if (need_sync) {
703		return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
704		    nvp, 6, ZFS_SPACE_CHECK_RESERVED));
705	}
706
707	return (0);
708}
709
710/*
711 * If the bootfs property value is dsobj, clear it.
712 */
713void
714spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
715{
716	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
717		VERIFY(zap_remove(spa->spa_meta_objset,
718		    spa->spa_pool_props_object,
719		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
720		spa->spa_bootfs = 0;
721	}
722}
723
724/*ARGSUSED*/
725static int
726spa_change_guid_check(void *arg, dmu_tx_t *tx)
727{
728	uint64_t *newguid = arg;
729	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
730	vdev_t *rvd = spa->spa_root_vdev;
731	uint64_t vdev_state;
732
733	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
734	vdev_state = rvd->vdev_state;
735	spa_config_exit(spa, SCL_STATE, FTAG);
736
737	if (vdev_state != VDEV_STATE_HEALTHY)
738		return (SET_ERROR(ENXIO));
739
740	ASSERT3U(spa_guid(spa), !=, *newguid);
741
742	return (0);
743}
744
745static void
746spa_change_guid_sync(void *arg, dmu_tx_t *tx)
747{
748	uint64_t *newguid = arg;
749	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
750	uint64_t oldguid;
751	vdev_t *rvd = spa->spa_root_vdev;
752
753	oldguid = spa_guid(spa);
754
755	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
756	rvd->vdev_guid = *newguid;
757	rvd->vdev_guid_sum += (*newguid - oldguid);
758	vdev_config_dirty(rvd);
759	spa_config_exit(spa, SCL_STATE, FTAG);
760
761	spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
762	    oldguid, *newguid);
763}
764
765/*
766 * Change the GUID for the pool.  This is done so that we can later
767 * re-import a pool built from a clone of our own vdevs.  We will modify
768 * the root vdev's guid, our own pool guid, and then mark all of our
769 * vdevs dirty.  Note that we must make sure that all our vdevs are
770 * online when we do this, or else any vdevs that weren't present
771 * would be orphaned from our pool.  We are also going to issue a
772 * sysevent to update any watchers.
773 */
774int
775spa_change_guid(spa_t *spa)
776{
777	int error;
778	uint64_t guid;
779
780	mutex_enter(&spa->spa_vdev_top_lock);
781	mutex_enter(&spa_namespace_lock);
782	guid = spa_generate_guid(NULL);
783
784	error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
785	    spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
786
787	if (error == 0) {
788		spa_config_sync(spa, B_FALSE, B_TRUE);
789		spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
790	}
791
792	mutex_exit(&spa_namespace_lock);
793	mutex_exit(&spa->spa_vdev_top_lock);
794
795	return (error);
796}
797
798/*
799 * ==========================================================================
800 * SPA state manipulation (open/create/destroy/import/export)
801 * ==========================================================================
802 */
803
804static int
805spa_error_entry_compare(const void *a, const void *b)
806{
807	spa_error_entry_t *sa = (spa_error_entry_t *)a;
808	spa_error_entry_t *sb = (spa_error_entry_t *)b;
809	int ret;
810
811	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
812	    sizeof (zbookmark_phys_t));
813
814	if (ret < 0)
815		return (-1);
816	else if (ret > 0)
817		return (1);
818	else
819		return (0);
820}
821
822/*
823 * Utility function which retrieves copies of the current logs and
824 * re-initializes them in the process.
825 */
826void
827spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
828{
829	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
830
831	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
832	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
833
834	avl_create(&spa->spa_errlist_scrub,
835	    spa_error_entry_compare, sizeof (spa_error_entry_t),
836	    offsetof(spa_error_entry_t, se_avl));
837	avl_create(&spa->spa_errlist_last,
838	    spa_error_entry_compare, sizeof (spa_error_entry_t),
839	    offsetof(spa_error_entry_t, se_avl));
840}
841
842static void
843spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
844{
845	const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
846	enum zti_modes mode = ztip->zti_mode;
847	uint_t value = ztip->zti_value;
848	uint_t count = ztip->zti_count;
849	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
850	char name[32];
851	uint_t flags = 0;
852	boolean_t batch = B_FALSE;
853
854	if (mode == ZTI_MODE_NULL) {
855		tqs->stqs_count = 0;
856		tqs->stqs_taskq = NULL;
857		return;
858	}
859
860	ASSERT3U(count, >, 0);
861
862	tqs->stqs_count = count;
863	tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
864
865	switch (mode) {
866	case ZTI_MODE_FIXED:
867		ASSERT3U(value, >=, 1);
868		value = MAX(value, 1);
869		break;
870
871	case ZTI_MODE_BATCH:
872		batch = B_TRUE;
873		flags |= TASKQ_THREADS_CPU_PCT;
874		value = zio_taskq_batch_pct;
875		break;
876
877	default:
878		panic("unrecognized mode for %s_%s taskq (%u:%u) in "
879		    "spa_activate()",
880		    zio_type_name[t], zio_taskq_types[q], mode, value);
881		break;
882	}
883
884	for (uint_t i = 0; i < count; i++) {
885		taskq_t *tq;
886
887		if (count > 1) {
888			(void) snprintf(name, sizeof (name), "%s_%s_%u",
889			    zio_type_name[t], zio_taskq_types[q], i);
890		} else {
891			(void) snprintf(name, sizeof (name), "%s_%s",
892			    zio_type_name[t], zio_taskq_types[q]);
893		}
894
895		if (zio_taskq_sysdc && spa->spa_proc != &p0) {
896			if (batch)
897				flags |= TASKQ_DC_BATCH;
898
899			tq = taskq_create_sysdc(name, value, 50, INT_MAX,
900			    spa->spa_proc, zio_taskq_basedc, flags);
901		} else {
902			pri_t pri = maxclsyspri;
903			/*
904			 * The write issue taskq can be extremely CPU
905			 * intensive.  Run it at slightly lower priority
906			 * than the other taskqs.
907			 */
908			if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
909				pri--;
910
911			tq = taskq_create_proc(name, value, pri, 50,
912			    INT_MAX, spa->spa_proc, flags);
913		}
914
915		tqs->stqs_taskq[i] = tq;
916	}
917}
918
919static void
920spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
921{
922	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
923
924	if (tqs->stqs_taskq == NULL) {
925		ASSERT0(tqs->stqs_count);
926		return;
927	}
928
929	for (uint_t i = 0; i < tqs->stqs_count; i++) {
930		ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
931		taskq_destroy(tqs->stqs_taskq[i]);
932	}
933
934	kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
935	tqs->stqs_taskq = NULL;
936}
937
938/*
939 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
940 * Note that a type may have multiple discrete taskqs to avoid lock contention
941 * on the taskq itself. In that case we choose which taskq at random by using
942 * the low bits of gethrtime().
943 */
944void
945spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
946    task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
947{
948	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
949	taskq_t *tq;
950
951	ASSERT3P(tqs->stqs_taskq, !=, NULL);
952	ASSERT3U(tqs->stqs_count, !=, 0);
953
954	if (tqs->stqs_count == 1) {
955		tq = tqs->stqs_taskq[0];
956	} else {
957		tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
958	}
959
960	taskq_dispatch_ent(tq, func, arg, flags, ent);
961}
962
963static void
964spa_create_zio_taskqs(spa_t *spa)
965{
966	for (int t = 0; t < ZIO_TYPES; t++) {
967		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
968			spa_taskqs_init(spa, t, q);
969		}
970	}
971}
972
973#ifdef _KERNEL
974static void
975spa_thread(void *arg)
976{
977	callb_cpr_t cprinfo;
978
979	spa_t *spa = arg;
980	user_t *pu = PTOU(curproc);
981
982	CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
983	    spa->spa_name);
984
985	ASSERT(curproc != &p0);
986	(void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
987	    "zpool-%s", spa->spa_name);
988	(void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
989
990	/* bind this thread to the requested psrset */
991	if (zio_taskq_psrset_bind != PS_NONE) {
992		pool_lock();
993		mutex_enter(&cpu_lock);
994		mutex_enter(&pidlock);
995		mutex_enter(&curproc->p_lock);
996
997		if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
998		    0, NULL, NULL) == 0)  {
999			curthread->t_bind_pset = zio_taskq_psrset_bind;
1000		} else {
1001			cmn_err(CE_WARN,
1002			    "Couldn't bind process for zfs pool \"%s\" to "
1003			    "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1004		}
1005
1006		mutex_exit(&curproc->p_lock);
1007		mutex_exit(&pidlock);
1008		mutex_exit(&cpu_lock);
1009		pool_unlock();
1010	}
1011
1012	if (zio_taskq_sysdc) {
1013		sysdc_thread_enter(curthread, 100, 0);
1014	}
1015
1016	spa->spa_proc = curproc;
1017	spa->spa_did = curthread->t_did;
1018
1019	spa_create_zio_taskqs(spa);
1020
1021	mutex_enter(&spa->spa_proc_lock);
1022	ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1023
1024	spa->spa_proc_state = SPA_PROC_ACTIVE;
1025	cv_broadcast(&spa->spa_proc_cv);
1026
1027	CALLB_CPR_SAFE_BEGIN(&cprinfo);
1028	while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1029		cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1030	CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1031
1032	ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1033	spa->spa_proc_state = SPA_PROC_GONE;
1034	spa->spa_proc = &p0;
1035	cv_broadcast(&spa->spa_proc_cv);
1036	CALLB_CPR_EXIT(&cprinfo);	/* drops spa_proc_lock */
1037
1038	mutex_enter(&curproc->p_lock);
1039	lwp_exit();
1040}
1041#endif
1042
1043/*
1044 * Activate an uninitialized pool.
1045 */
1046static void
1047spa_activate(spa_t *spa, int mode)
1048{
1049	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1050
1051	spa->spa_state = POOL_STATE_ACTIVE;
1052	spa->spa_mode = mode;
1053
1054	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1055	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1056
1057	/* Try to create a covering process */
1058	mutex_enter(&spa->spa_proc_lock);
1059	ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1060	ASSERT(spa->spa_proc == &p0);
1061	spa->spa_did = 0;
1062
1063	/* Only create a process if we're going to be around a while. */
1064	if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1065		if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1066		    NULL, 0) == 0) {
1067			spa->spa_proc_state = SPA_PROC_CREATED;
1068			while (spa->spa_proc_state == SPA_PROC_CREATED) {
1069				cv_wait(&spa->spa_proc_cv,
1070				    &spa->spa_proc_lock);
1071			}
1072			ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1073			ASSERT(spa->spa_proc != &p0);
1074			ASSERT(spa->spa_did != 0);
1075		} else {
1076#ifdef _KERNEL
1077			cmn_err(CE_WARN,
1078			    "Couldn't create process for zfs pool \"%s\"\n",
1079			    spa->spa_name);
1080#endif
1081		}
1082	}
1083	mutex_exit(&spa->spa_proc_lock);
1084
1085	/* If we didn't create a process, we need to create our taskqs. */
1086	if (spa->spa_proc == &p0) {
1087		spa_create_zio_taskqs(spa);
1088	}
1089
1090	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1091	    offsetof(vdev_t, vdev_config_dirty_node));
1092	list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1093	    offsetof(objset_t, os_evicting_node));
1094	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1095	    offsetof(vdev_t, vdev_state_dirty_node));
1096
1097	txg_list_create(&spa->spa_vdev_txg_list,
1098	    offsetof(struct vdev, vdev_txg_node));
1099
1100	avl_create(&spa->spa_errlist_scrub,
1101	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1102	    offsetof(spa_error_entry_t, se_avl));
1103	avl_create(&spa->spa_errlist_last,
1104	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1105	    offsetof(spa_error_entry_t, se_avl));
1106}
1107
1108/*
1109 * Opposite of spa_activate().
1110 */
1111static void
1112spa_deactivate(spa_t *spa)
1113{
1114	ASSERT(spa->spa_sync_on == B_FALSE);
1115	ASSERT(spa->spa_dsl_pool == NULL);
1116	ASSERT(spa->spa_root_vdev == NULL);
1117	ASSERT(spa->spa_async_zio_root == NULL);
1118	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1119
1120	spa_evicting_os_wait(spa);
1121
1122	txg_list_destroy(&spa->spa_vdev_txg_list);
1123
1124	list_destroy(&spa->spa_config_dirty_list);
1125	list_destroy(&spa->spa_evicting_os_list);
1126	list_destroy(&spa->spa_state_dirty_list);
1127
1128	for (int t = 0; t < ZIO_TYPES; t++) {
1129		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1130			spa_taskqs_fini(spa, t, q);
1131		}
1132	}
1133
1134	metaslab_class_destroy(spa->spa_normal_class);
1135	spa->spa_normal_class = NULL;
1136
1137	metaslab_class_destroy(spa->spa_log_class);
1138	spa->spa_log_class = NULL;
1139
1140	/*
1141	 * If this was part of an import or the open otherwise failed, we may
1142	 * still have errors left in the queues.  Empty them just in case.
1143	 */
1144	spa_errlog_drain(spa);
1145
1146	avl_destroy(&spa->spa_errlist_scrub);
1147	avl_destroy(&spa->spa_errlist_last);
1148
1149	spa->spa_state = POOL_STATE_UNINITIALIZED;
1150
1151	mutex_enter(&spa->spa_proc_lock);
1152	if (spa->spa_proc_state != SPA_PROC_NONE) {
1153		ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1154		spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1155		cv_broadcast(&spa->spa_proc_cv);
1156		while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1157			ASSERT(spa->spa_proc != &p0);
1158			cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1159		}
1160		ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1161		spa->spa_proc_state = SPA_PROC_NONE;
1162	}
1163	ASSERT(spa->spa_proc == &p0);
1164	mutex_exit(&spa->spa_proc_lock);
1165
1166	/*
1167	 * We want to make sure spa_thread() has actually exited the ZFS
1168	 * module, so that the module can't be unloaded out from underneath
1169	 * it.
1170	 */
1171	if (spa->spa_did != 0) {
1172		thread_join(spa->spa_did);
1173		spa->spa_did = 0;
1174	}
1175}
1176
1177/*
1178 * Verify a pool configuration, and construct the vdev tree appropriately.  This
1179 * will create all the necessary vdevs in the appropriate layout, with each vdev
1180 * in the CLOSED state.  This will prep the pool before open/creation/import.
1181 * All vdev validation is done by the vdev_alloc() routine.
1182 */
1183static int
1184spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1185    uint_t id, int atype)
1186{
1187	nvlist_t **child;
1188	uint_t children;
1189	int error;
1190
1191	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1192		return (error);
1193
1194	if ((*vdp)->vdev_ops->vdev_op_leaf)
1195		return (0);
1196
1197	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1198	    &child, &children);
1199
1200	if (error == ENOENT)
1201		return (0);
1202
1203	if (error) {
1204		vdev_free(*vdp);
1205		*vdp = NULL;
1206		return (SET_ERROR(EINVAL));
1207	}
1208
1209	for (int c = 0; c < children; c++) {
1210		vdev_t *vd;
1211		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1212		    atype)) != 0) {
1213			vdev_free(*vdp);
1214			*vdp = NULL;
1215			return (error);
1216		}
1217	}
1218
1219	ASSERT(*vdp != NULL);
1220
1221	return (0);
1222}
1223
1224/*
1225 * Opposite of spa_load().
1226 */
1227static void
1228spa_unload(spa_t *spa)
1229{
1230	int i;
1231
1232	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1233
1234	/*
1235	 * Stop async tasks.
1236	 */
1237	spa_async_suspend(spa);
1238
1239	/*
1240	 * Stop syncing.
1241	 */
1242	if (spa->spa_sync_on) {
1243		txg_sync_stop(spa->spa_dsl_pool);
1244		spa->spa_sync_on = B_FALSE;
1245	}
1246
1247	/*
1248	 * Wait for any outstanding async I/O to complete.
1249	 */
1250	if (spa->spa_async_zio_root != NULL) {
1251		for (int i = 0; i < max_ncpus; i++)
1252			(void) zio_wait(spa->spa_async_zio_root[i]);
1253		kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1254		spa->spa_async_zio_root = NULL;
1255	}
1256
1257	bpobj_close(&spa->spa_deferred_bpobj);
1258
1259	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1260
1261	/*
1262	 * Close all vdevs.
1263	 */
1264	if (spa->spa_root_vdev)
1265		vdev_free(spa->spa_root_vdev);
1266	ASSERT(spa->spa_root_vdev == NULL);
1267
1268	/*
1269	 * Close the dsl pool.
1270	 */
1271	if (spa->spa_dsl_pool) {
1272		dsl_pool_close(spa->spa_dsl_pool);
1273		spa->spa_dsl_pool = NULL;
1274		spa->spa_meta_objset = NULL;
1275	}
1276
1277	ddt_unload(spa);
1278
1279
1280	/*
1281	 * Drop and purge level 2 cache
1282	 */
1283	spa_l2cache_drop(spa);
1284
1285	for (i = 0; i < spa->spa_spares.sav_count; i++)
1286		vdev_free(spa->spa_spares.sav_vdevs[i]);
1287	if (spa->spa_spares.sav_vdevs) {
1288		kmem_free(spa->spa_spares.sav_vdevs,
1289		    spa->spa_spares.sav_count * sizeof (void *));
1290		spa->spa_spares.sav_vdevs = NULL;
1291	}
1292	if (spa->spa_spares.sav_config) {
1293		nvlist_free(spa->spa_spares.sav_config);
1294		spa->spa_spares.sav_config = NULL;
1295	}
1296	spa->spa_spares.sav_count = 0;
1297
1298	for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1299		vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1300		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1301	}
1302	if (spa->spa_l2cache.sav_vdevs) {
1303		kmem_free(spa->spa_l2cache.sav_vdevs,
1304		    spa->spa_l2cache.sav_count * sizeof (void *));
1305		spa->spa_l2cache.sav_vdevs = NULL;
1306	}
1307	if (spa->spa_l2cache.sav_config) {
1308		nvlist_free(spa->spa_l2cache.sav_config);
1309		spa->spa_l2cache.sav_config = NULL;
1310	}
1311	spa->spa_l2cache.sav_count = 0;
1312
1313	spa->spa_async_suspended = 0;
1314
1315	if (spa->spa_comment != NULL) {
1316		spa_strfree(spa->spa_comment);
1317		spa->spa_comment = NULL;
1318	}
1319
1320	spa_config_exit(spa, SCL_ALL, FTAG);
1321}
1322
1323/*
1324 * Load (or re-load) the current list of vdevs describing the active spares for
1325 * this pool.  When this is called, we have some form of basic information in
1326 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1327 * then re-generate a more complete list including status information.
1328 */
1329static void
1330spa_load_spares(spa_t *spa)
1331{
1332	nvlist_t **spares;
1333	uint_t nspares;
1334	int i;
1335	vdev_t *vd, *tvd;
1336
1337	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1338
1339	/*
1340	 * First, close and free any existing spare vdevs.
1341	 */
1342	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1343		vd = spa->spa_spares.sav_vdevs[i];
1344
1345		/* Undo the call to spa_activate() below */
1346		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1347		    B_FALSE)) != NULL && tvd->vdev_isspare)
1348			spa_spare_remove(tvd);
1349		vdev_close(vd);
1350		vdev_free(vd);
1351	}
1352
1353	if (spa->spa_spares.sav_vdevs)
1354		kmem_free(spa->spa_spares.sav_vdevs,
1355		    spa->spa_spares.sav_count * sizeof (void *));
1356
1357	if (spa->spa_spares.sav_config == NULL)
1358		nspares = 0;
1359	else
1360		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1361		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1362
1363	spa->spa_spares.sav_count = (int)nspares;
1364	spa->spa_spares.sav_vdevs = NULL;
1365
1366	if (nspares == 0)
1367		return;
1368
1369	/*
1370	 * Construct the array of vdevs, opening them to get status in the
1371	 * process.   For each spare, there is potentially two different vdev_t
1372	 * structures associated with it: one in the list of spares (used only
1373	 * for basic validation purposes) and one in the active vdev
1374	 * configuration (if it's spared in).  During this phase we open and
1375	 * validate each vdev on the spare list.  If the vdev also exists in the
1376	 * active configuration, then we also mark this vdev as an active spare.
1377	 */
1378	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1379	    KM_SLEEP);
1380	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1381		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1382		    VDEV_ALLOC_SPARE) == 0);
1383		ASSERT(vd != NULL);
1384
1385		spa->spa_spares.sav_vdevs[i] = vd;
1386
1387		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1388		    B_FALSE)) != NULL) {
1389			if (!tvd->vdev_isspare)
1390				spa_spare_add(tvd);
1391
1392			/*
1393			 * We only mark the spare active if we were successfully
1394			 * able to load the vdev.  Otherwise, importing a pool
1395			 * with a bad active spare would result in strange
1396			 * behavior, because multiple pool would think the spare
1397			 * is actively in use.
1398			 *
1399			 * There is a vulnerability here to an equally bizarre
1400			 * circumstance, where a dead active spare is later
1401			 * brought back to life (onlined or otherwise).  Given
1402			 * the rarity of this scenario, and the extra complexity
1403			 * it adds, we ignore the possibility.
1404			 */
1405			if (!vdev_is_dead(tvd))
1406				spa_spare_activate(tvd);
1407		}
1408
1409		vd->vdev_top = vd;
1410		vd->vdev_aux = &spa->spa_spares;
1411
1412		if (vdev_open(vd) != 0)
1413			continue;
1414
1415		if (vdev_validate_aux(vd) == 0)
1416			spa_spare_add(vd);
1417	}
1418
1419	/*
1420	 * Recompute the stashed list of spares, with status information
1421	 * this time.
1422	 */
1423	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1424	    DATA_TYPE_NVLIST_ARRAY) == 0);
1425
1426	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1427	    KM_SLEEP);
1428	for (i = 0; i < spa->spa_spares.sav_count; i++)
1429		spares[i] = vdev_config_generate(spa,
1430		    spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1431	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1432	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1433	for (i = 0; i < spa->spa_spares.sav_count; i++)
1434		nvlist_free(spares[i]);
1435	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1436}
1437
1438/*
1439 * Load (or re-load) the current list of vdevs describing the active l2cache for
1440 * this pool.  When this is called, we have some form of basic information in
1441 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1442 * then re-generate a more complete list including status information.
1443 * Devices which are already active have their details maintained, and are
1444 * not re-opened.
1445 */
1446static void
1447spa_load_l2cache(spa_t *spa)
1448{
1449	nvlist_t **l2cache;
1450	uint_t nl2cache;
1451	int i, j, oldnvdevs;
1452	uint64_t guid;
1453	vdev_t *vd, **oldvdevs, **newvdevs;
1454	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1455
1456	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1457
1458	if (sav->sav_config != NULL) {
1459		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1460		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1461		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1462	} else {
1463		nl2cache = 0;
1464		newvdevs = NULL;
1465	}
1466
1467	oldvdevs = sav->sav_vdevs;
1468	oldnvdevs = sav->sav_count;
1469	sav->sav_vdevs = NULL;
1470	sav->sav_count = 0;
1471
1472	/*
1473	 * Process new nvlist of vdevs.
1474	 */
1475	for (i = 0; i < nl2cache; i++) {
1476		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1477		    &guid) == 0);
1478
1479		newvdevs[i] = NULL;
1480		for (j = 0; j < oldnvdevs; j++) {
1481			vd = oldvdevs[j];
1482			if (vd != NULL && guid == vd->vdev_guid) {
1483				/*
1484				 * Retain previous vdev for add/remove ops.
1485				 */
1486				newvdevs[i] = vd;
1487				oldvdevs[j] = NULL;
1488				break;
1489			}
1490		}
1491
1492		if (newvdevs[i] == NULL) {
1493			/*
1494			 * Create new vdev
1495			 */
1496			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1497			    VDEV_ALLOC_L2CACHE) == 0);
1498			ASSERT(vd != NULL);
1499			newvdevs[i] = vd;
1500
1501			/*
1502			 * Commit this vdev as an l2cache device,
1503			 * even if it fails to open.
1504			 */
1505			spa_l2cache_add(vd);
1506
1507			vd->vdev_top = vd;
1508			vd->vdev_aux = sav;
1509
1510			spa_l2cache_activate(vd);
1511
1512			if (vdev_open(vd) != 0)
1513				continue;
1514
1515			(void) vdev_validate_aux(vd);
1516
1517			if (!vdev_is_dead(vd))
1518				l2arc_add_vdev(spa, vd);
1519		}
1520	}
1521
1522	/*
1523	 * Purge vdevs that were dropped
1524	 */
1525	for (i = 0; i < oldnvdevs; i++) {
1526		uint64_t pool;
1527
1528		vd = oldvdevs[i];
1529		if (vd != NULL) {
1530			ASSERT(vd->vdev_isl2cache);
1531
1532			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1533			    pool != 0ULL && l2arc_vdev_present(vd))
1534				l2arc_remove_vdev(vd);
1535			vdev_clear_stats(vd);
1536			vdev_free(vd);
1537		}
1538	}
1539
1540	if (oldvdevs)
1541		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1542
1543	if (sav->sav_config == NULL)
1544		goto out;
1545
1546	sav->sav_vdevs = newvdevs;
1547	sav->sav_count = (int)nl2cache;
1548
1549	/*
1550	 * Recompute the stashed list of l2cache devices, with status
1551	 * information this time.
1552	 */
1553	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1554	    DATA_TYPE_NVLIST_ARRAY) == 0);
1555
1556	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1557	for (i = 0; i < sav->sav_count; i++)
1558		l2cache[i] = vdev_config_generate(spa,
1559		    sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1560	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1561	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1562out:
1563	for (i = 0; i < sav->sav_count; i++)
1564		nvlist_free(l2cache[i]);
1565	if (sav->sav_count)
1566		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1567}
1568
1569static int
1570load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1571{
1572	dmu_buf_t *db;
1573	char *packed = NULL;
1574	size_t nvsize = 0;
1575	int error;
1576	*value = NULL;
1577
1578	error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1579	if (error != 0)
1580		return (error);
1581
1582	nvsize = *(uint64_t *)db->db_data;
1583	dmu_buf_rele(db, FTAG);
1584
1585	packed = kmem_alloc(nvsize, KM_SLEEP);
1586	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1587	    DMU_READ_PREFETCH);
1588	if (error == 0)
1589		error = nvlist_unpack(packed, nvsize, value, 0);
1590	kmem_free(packed, nvsize);
1591
1592	return (error);
1593}
1594
1595/*
1596 * Checks to see if the given vdev could not be opened, in which case we post a
1597 * sysevent to notify the autoreplace code that the device has been removed.
1598 */
1599static void
1600spa_check_removed(vdev_t *vd)
1601{
1602	for (int c = 0; c < vd->vdev_children; c++)
1603		spa_check_removed(vd->vdev_child[c]);
1604
1605	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1606	    !vd->vdev_ishole) {
1607		zfs_post_autoreplace(vd->vdev_spa, vd);
1608		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1609	}
1610}
1611
1612/*
1613 * Validate the current config against the MOS config
1614 */
1615static boolean_t
1616spa_config_valid(spa_t *spa, nvlist_t *config)
1617{
1618	vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1619	nvlist_t *nv;
1620
1621	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1622
1623	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1624	VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1625
1626	ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1627
1628	/*
1629	 * If we're doing a normal import, then build up any additional
1630	 * diagnostic information about missing devices in this config.
1631	 * We'll pass this up to the user for further processing.
1632	 */
1633	if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1634		nvlist_t **child, *nv;
1635		uint64_t idx = 0;
1636
1637		child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1638		    KM_SLEEP);
1639		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1640
1641		for (int c = 0; c < rvd->vdev_children; c++) {
1642			vdev_t *tvd = rvd->vdev_child[c];
1643			vdev_t *mtvd  = mrvd->vdev_child[c];
1644
1645			if (tvd->vdev_ops == &vdev_missing_ops &&
1646			    mtvd->vdev_ops != &vdev_missing_ops &&
1647			    mtvd->vdev_islog)
1648				child[idx++] = vdev_config_generate(spa, mtvd,
1649				    B_FALSE, 0);
1650		}
1651
1652		if (idx) {
1653			VERIFY(nvlist_add_nvlist_array(nv,
1654			    ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1655			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1656			    ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1657
1658			for (int i = 0; i < idx; i++)
1659				nvlist_free(child[i]);
1660		}
1661		nvlist_free(nv);
1662		kmem_free(child, rvd->vdev_children * sizeof (char **));
1663	}
1664
1665	/*
1666	 * Compare the root vdev tree with the information we have
1667	 * from the MOS config (mrvd). Check each top-level vdev
1668	 * with the corresponding MOS config top-level (mtvd).
1669	 */
1670	for (int c = 0; c < rvd->vdev_children; c++) {
1671		vdev_t *tvd = rvd->vdev_child[c];
1672		vdev_t *mtvd  = mrvd->vdev_child[c];
1673
1674		/*
1675		 * Resolve any "missing" vdevs in the current configuration.
1676		 * If we find that the MOS config has more accurate information
1677		 * about the top-level vdev then use that vdev instead.
1678		 */
1679		if (tvd->vdev_ops == &vdev_missing_ops &&
1680		    mtvd->vdev_ops != &vdev_missing_ops) {
1681
1682			if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1683				continue;
1684
1685			/*
1686			 * Device specific actions.
1687			 */
1688			if (mtvd->vdev_islog) {
1689				spa_set_log_state(spa, SPA_LOG_CLEAR);
1690			} else {
1691				/*
1692				 * XXX - once we have 'readonly' pool
1693				 * support we should be able to handle
1694				 * missing data devices by transitioning
1695				 * the pool to readonly.
1696				 */
1697				continue;
1698			}
1699
1700			/*
1701			 * Swap the missing vdev with the data we were
1702			 * able to obtain from the MOS config.
1703			 */
1704			vdev_remove_child(rvd, tvd);
1705			vdev_remove_child(mrvd, mtvd);
1706
1707			vdev_add_child(rvd, mtvd);
1708			vdev_add_child(mrvd, tvd);
1709
1710			spa_config_exit(spa, SCL_ALL, FTAG);
1711			vdev_load(mtvd);
1712			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1713
1714			vdev_reopen(rvd);
1715		} else if (mtvd->vdev_islog) {
1716			/*
1717			 * Load the slog device's state from the MOS config
1718			 * since it's possible that the label does not
1719			 * contain the most up-to-date information.
1720			 */
1721			vdev_load_log_state(tvd, mtvd);
1722			vdev_reopen(tvd);
1723		}
1724	}
1725	vdev_free(mrvd);
1726	spa_config_exit(spa, SCL_ALL, FTAG);
1727
1728	/*
1729	 * Ensure we were able to validate the config.
1730	 */
1731	return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1732}
1733
1734/*
1735 * Check for missing log devices
1736 */
1737static boolean_t
1738spa_check_logs(spa_t *spa)
1739{
1740	boolean_t rv = B_FALSE;
1741	dsl_pool_t *dp = spa_get_dsl(spa);
1742
1743	switch (spa->spa_log_state) {
1744	case SPA_LOG_MISSING:
1745		/* need to recheck in case slog has been restored */
1746	case SPA_LOG_UNKNOWN:
1747		rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1748		    zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
1749		if (rv)
1750			spa_set_log_state(spa, SPA_LOG_MISSING);
1751		break;
1752	}
1753	return (rv);
1754}
1755
1756static boolean_t
1757spa_passivate_log(spa_t *spa)
1758{
1759	vdev_t *rvd = spa->spa_root_vdev;
1760	boolean_t slog_found = B_FALSE;
1761
1762	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1763
1764	if (!spa_has_slogs(spa))
1765		return (B_FALSE);
1766
1767	for (int c = 0; c < rvd->vdev_children; c++) {
1768		vdev_t *tvd = rvd->vdev_child[c];
1769		metaslab_group_t *mg = tvd->vdev_mg;
1770
1771		if (tvd->vdev_islog) {
1772			metaslab_group_passivate(mg);
1773			slog_found = B_TRUE;
1774		}
1775	}
1776
1777	return (slog_found);
1778}
1779
1780static void
1781spa_activate_log(spa_t *spa)
1782{
1783	vdev_t *rvd = spa->spa_root_vdev;
1784
1785	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1786
1787	for (int c = 0; c < rvd->vdev_children; c++) {
1788		vdev_t *tvd = rvd->vdev_child[c];
1789		metaslab_group_t *mg = tvd->vdev_mg;
1790
1791		if (tvd->vdev_islog)
1792			metaslab_group_activate(mg);
1793	}
1794}
1795
1796int
1797spa_offline_log(spa_t *spa)
1798{
1799	int error;
1800
1801	error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1802	    NULL, DS_FIND_CHILDREN);
1803	if (error == 0) {
1804		/*
1805		 * We successfully offlined the log device, sync out the
1806		 * current txg so that the "stubby" block can be removed
1807		 * by zil_sync().
1808		 */
1809		txg_wait_synced(spa->spa_dsl_pool, 0);
1810	}
1811	return (error);
1812}
1813
1814static void
1815spa_aux_check_removed(spa_aux_vdev_t *sav)
1816{
1817	for (int i = 0; i < sav->sav_count; i++)
1818		spa_check_removed(sav->sav_vdevs[i]);
1819}
1820
1821void
1822spa_claim_notify(zio_t *zio)
1823{
1824	spa_t *spa = zio->io_spa;
1825
1826	if (zio->io_error)
1827		return;
1828
1829	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1830	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1831		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1832	mutex_exit(&spa->spa_props_lock);
1833}
1834
1835typedef struct spa_load_error {
1836	uint64_t	sle_meta_count;
1837	uint64_t	sle_data_count;
1838} spa_load_error_t;
1839
1840static void
1841spa_load_verify_done(zio_t *zio)
1842{
1843	blkptr_t *bp = zio->io_bp;
1844	spa_load_error_t *sle = zio->io_private;
1845	dmu_object_type_t type = BP_GET_TYPE(bp);
1846	int error = zio->io_error;
1847	spa_t *spa = zio->io_spa;
1848
1849	if (error) {
1850		if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1851		    type != DMU_OT_INTENT_LOG)
1852			atomic_inc_64(&sle->sle_meta_count);
1853		else
1854			atomic_inc_64(&sle->sle_data_count);
1855	}
1856	zio_data_buf_free(zio->io_data, zio->io_size);
1857
1858	mutex_enter(&spa->spa_scrub_lock);
1859	spa->spa_scrub_inflight--;
1860	cv_broadcast(&spa->spa_scrub_io_cv);
1861	mutex_exit(&spa->spa_scrub_lock);
1862}
1863
1864/*
1865 * Maximum number of concurrent scrub i/os to create while verifying
1866 * a pool while importing it.
1867 */
1868int spa_load_verify_maxinflight = 10000;
1869boolean_t spa_load_verify_metadata = B_TRUE;
1870boolean_t spa_load_verify_data = B_TRUE;
1871
1872/*ARGSUSED*/
1873static int
1874spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1875    const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1876{
1877	if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
1878		return (0);
1879	/*
1880	 * Note: normally this routine will not be called if
1881	 * spa_load_verify_metadata is not set.  However, it may be useful
1882	 * to manually set the flag after the traversal has begun.
1883	 */
1884	if (!spa_load_verify_metadata)
1885		return (0);
1886	if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data)
1887		return (0);
1888
1889	zio_t *rio = arg;
1890	size_t size = BP_GET_PSIZE(bp);
1891	void *data = zio_data_buf_alloc(size);
1892
1893	mutex_enter(&spa->spa_scrub_lock);
1894	while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
1895		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1896	spa->spa_scrub_inflight++;
1897	mutex_exit(&spa->spa_scrub_lock);
1898
1899	zio_nowait(zio_read(rio, spa, bp, data, size,
1900	    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1901	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1902	    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1903	return (0);
1904}
1905
1906static int
1907spa_load_verify(spa_t *spa)
1908{
1909	zio_t *rio;
1910	spa_load_error_t sle = { 0 };
1911	zpool_rewind_policy_t policy;
1912	boolean_t verify_ok = B_FALSE;
1913	int error = 0;
1914
1915	zpool_get_rewind_policy(spa->spa_config, &policy);
1916
1917	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1918		return (0);
1919
1920	rio = zio_root(spa, NULL, &sle,
1921	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1922
1923	if (spa_load_verify_metadata) {
1924		error = traverse_pool(spa, spa->spa_verify_min_txg,
1925		    TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
1926		    spa_load_verify_cb, rio);
1927	}
1928
1929	(void) zio_wait(rio);
1930
1931	spa->spa_load_meta_errors = sle.sle_meta_count;
1932	spa->spa_load_data_errors = sle.sle_data_count;
1933
1934	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1935	    sle.sle_data_count <= policy.zrp_maxdata) {
1936		int64_t loss = 0;
1937
1938		verify_ok = B_TRUE;
1939		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1940		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1941
1942		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1943		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1944		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1945		VERIFY(nvlist_add_int64(spa->spa_load_info,
1946		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1947		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1948		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1949	} else {
1950		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1951	}
1952
1953	if (error) {
1954		if (error != ENXIO && error != EIO)
1955			error = SET_ERROR(EIO);
1956		return (error);
1957	}
1958
1959	return (verify_ok ? 0 : EIO);
1960}
1961
1962/*
1963 * Find a value in the pool props object.
1964 */
1965static void
1966spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1967{
1968	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1969	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1970}
1971
1972/*
1973 * Find a value in the pool directory object.
1974 */
1975static int
1976spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1977{
1978	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1979	    name, sizeof (uint64_t), 1, val));
1980}
1981
1982static int
1983spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1984{
1985	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1986	return (err);
1987}
1988
1989/*
1990 * Fix up config after a partly-completed split.  This is done with the
1991 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1992 * pool have that entry in their config, but only the splitting one contains
1993 * a list of all the guids of the vdevs that are being split off.
1994 *
1995 * This function determines what to do with that list: either rejoin
1996 * all the disks to the pool, or complete the splitting process.  To attempt
1997 * the rejoin, each disk that is offlined is marked online again, and
1998 * we do a reopen() call.  If the vdev label for every disk that was
1999 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2000 * then we call vdev_split() on each disk, and complete the split.
2001 *
2002 * Otherwise we leave the config alone, with all the vdevs in place in
2003 * the original pool.
2004 */
2005static void
2006spa_try_repair(spa_t *spa, nvlist_t *config)
2007{
2008	uint_t extracted;
2009	uint64_t *glist;
2010	uint_t i, gcount;
2011	nvlist_t *nvl;
2012	vdev_t **vd;
2013	boolean_t attempt_reopen;
2014
2015	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2016		return;
2017
2018	/* check that the config is complete */
2019	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2020	    &glist, &gcount) != 0)
2021		return;
2022
2023	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2024
2025	/* attempt to online all the vdevs & validate */
2026	attempt_reopen = B_TRUE;
2027	for (i = 0; i < gcount; i++) {
2028		if (glist[i] == 0)	/* vdev is hole */
2029			continue;
2030
2031		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2032		if (vd[i] == NULL) {
2033			/*
2034			 * Don't bother attempting to reopen the disks;
2035			 * just do the split.
2036			 */
2037			attempt_reopen = B_FALSE;
2038		} else {
2039			/* attempt to re-online it */
2040			vd[i]->vdev_offline = B_FALSE;
2041		}
2042	}
2043
2044	if (attempt_reopen) {
2045		vdev_reopen(spa->spa_root_vdev);
2046
2047		/* check each device to see what state it's in */
2048		for (extracted = 0, i = 0; i < gcount; i++) {
2049			if (vd[i] != NULL &&
2050			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2051				break;
2052			++extracted;
2053		}
2054	}
2055
2056	/*
2057	 * If every disk has been moved to the new pool, or if we never
2058	 * even attempted to look at them, then we split them off for
2059	 * good.
2060	 */
2061	if (!attempt_reopen || gcount == extracted) {
2062		for (i = 0; i < gcount; i++)
2063			if (vd[i] != NULL)
2064				vdev_split(vd[i]);
2065		vdev_reopen(spa->spa_root_vdev);
2066	}
2067
2068	kmem_free(vd, gcount * sizeof (vdev_t *));
2069}
2070
2071static int
2072spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2073    boolean_t mosconfig)
2074{
2075	nvlist_t *config = spa->spa_config;
2076	char *ereport = FM_EREPORT_ZFS_POOL;
2077	char *comment;
2078	int error;
2079	uint64_t pool_guid;
2080	nvlist_t *nvl;
2081
2082	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2083		return (SET_ERROR(EINVAL));
2084
2085	ASSERT(spa->spa_comment == NULL);
2086	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2087		spa->spa_comment = spa_strdup(comment);
2088
2089	/*
2090	 * Versioning wasn't explicitly added to the label until later, so if
2091	 * it's not present treat it as the initial version.
2092	 */
2093	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2094	    &spa->spa_ubsync.ub_version) != 0)
2095		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2096
2097	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2098	    &spa->spa_config_txg);
2099
2100	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2101	    spa_guid_exists(pool_guid, 0)) {
2102		error = SET_ERROR(EEXIST);
2103	} else {
2104		spa->spa_config_guid = pool_guid;
2105
2106		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2107		    &nvl) == 0) {
2108			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2109			    KM_SLEEP) == 0);
2110		}
2111
2112		nvlist_free(spa->spa_load_info);
2113		spa->spa_load_info = fnvlist_alloc();
2114
2115		gethrestime(&spa->spa_loaded_ts);
2116		error = spa_load_impl(spa, pool_guid, config, state, type,
2117		    mosconfig, &ereport);
2118	}
2119
2120	/*
2121	 * Don't count references from objsets that are already closed
2122	 * and are making their way through the eviction process.
2123	 */
2124	spa_evicting_os_wait(spa);
2125	spa->spa_minref = refcount_count(&spa->spa_refcount);
2126	if (error) {
2127		if (error != EEXIST) {
2128			spa->spa_loaded_ts.tv_sec = 0;
2129			spa->spa_loaded_ts.tv_nsec = 0;
2130		}
2131		if (error != EBADF) {
2132			zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2133		}
2134	}
2135	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2136	spa->spa_ena = 0;
2137
2138	return (error);
2139}
2140
2141/*
2142 * Load an existing storage pool, using the pool's builtin spa_config as a
2143 * source of configuration information.
2144 */
2145static int
2146spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2147    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2148    char **ereport)
2149{
2150	int error = 0;
2151	nvlist_t *nvroot = NULL;
2152	nvlist_t *label;
2153	vdev_t *rvd;
2154	uberblock_t *ub = &spa->spa_uberblock;
2155	uint64_t children, config_cache_txg = spa->spa_config_txg;
2156	int orig_mode = spa->spa_mode;
2157	int parse;
2158	uint64_t obj;
2159	boolean_t missing_feat_write = B_FALSE;
2160
2161	/*
2162	 * If this is an untrusted config, access the pool in read-only mode.
2163	 * This prevents things like resilvering recently removed devices.
2164	 */
2165	if (!mosconfig)
2166		spa->spa_mode = FREAD;
2167
2168	ASSERT(MUTEX_HELD(&spa_namespace_lock));
2169
2170	spa->spa_load_state = state;
2171
2172	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2173		return (SET_ERROR(EINVAL));
2174
2175	parse = (type == SPA_IMPORT_EXISTING ?
2176	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2177
2178	/*
2179	 * Create "The Godfather" zio to hold all async IOs
2180	 */
2181	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2182	    KM_SLEEP);
2183	for (int i = 0; i < max_ncpus; i++) {
2184		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2185		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2186		    ZIO_FLAG_GODFATHER);
2187	}
2188
2189	/*
2190	 * Parse the configuration into a vdev tree.  We explicitly set the
2191	 * value that will be returned by spa_version() since parsing the
2192	 * configuration requires knowing the version number.
2193	 */
2194	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2195	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2196	spa_config_exit(spa, SCL_ALL, FTAG);
2197
2198	if (error != 0)
2199		return (error);
2200
2201	ASSERT(spa->spa_root_vdev == rvd);
2202	ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
2203	ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
2204
2205	if (type != SPA_IMPORT_ASSEMBLE) {
2206		ASSERT(spa_guid(spa) == pool_guid);
2207	}
2208
2209	/*
2210	 * Try to open all vdevs, loading each label in the process.
2211	 */
2212	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2213	error = vdev_open(rvd);
2214	spa_config_exit(spa, SCL_ALL, FTAG);
2215	if (error != 0)
2216		return (error);
2217
2218	/*
2219	 * We need to validate the vdev labels against the configuration that
2220	 * we have in hand, which is dependent on the setting of mosconfig. If
2221	 * mosconfig is true then we're validating the vdev labels based on
2222	 * that config.  Otherwise, we're validating against the cached config
2223	 * (zpool.cache) that was read when we loaded the zfs module, and then
2224	 * later we will recursively call spa_load() and validate against
2225	 * the vdev config.
2226	 *
2227	 * If we're assembling a new pool that's been split off from an
2228	 * existing pool, the labels haven't yet been updated so we skip
2229	 * validation for now.
2230	 */
2231	if (type != SPA_IMPORT_ASSEMBLE) {
2232		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2233		error = vdev_validate(rvd, mosconfig);
2234		spa_config_exit(spa, SCL_ALL, FTAG);
2235
2236		if (error != 0)
2237			return (error);
2238
2239		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2240			return (SET_ERROR(ENXIO));
2241	}
2242
2243	/*
2244	 * Find the best uberblock.
2245	 */
2246	vdev_uberblock_load(rvd, ub, &label);
2247
2248	/*
2249	 * If we weren't able to find a single valid uberblock, return failure.
2250	 */
2251	if (ub->ub_txg == 0) {
2252		nvlist_free(label);
2253		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2254	}
2255
2256	/*
2257	 * If the pool has an unsupported version we can't open it.
2258	 */
2259	if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2260		nvlist_free(label);
2261		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2262	}
2263
2264	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2265		nvlist_t *features;
2266
2267		/*
2268		 * If we weren't able to find what's necessary for reading the
2269		 * MOS in the label, return failure.
2270		 */
2271		if (label == NULL || nvlist_lookup_nvlist(label,
2272		    ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2273			nvlist_free(label);
2274			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2275			    ENXIO));
2276		}
2277
2278		/*
2279		 * Update our in-core representation with the definitive values
2280		 * from the label.
2281		 */
2282		nvlist_free(spa->spa_label_features);
2283		VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2284	}
2285
2286	nvlist_free(label);
2287
2288	/*
2289	 * Look through entries in the label nvlist's features_for_read. If
2290	 * there is a feature listed there which we don't understand then we
2291	 * cannot open a pool.
2292	 */
2293	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2294		nvlist_t *unsup_feat;
2295
2296		VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2297		    0);
2298
2299		for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2300		    NULL); nvp != NULL;
2301		    nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2302			if (!zfeature_is_supported(nvpair_name(nvp))) {
2303				VERIFY(nvlist_add_string(unsup_feat,
2304				    nvpair_name(nvp), "") == 0);
2305			}
2306		}
2307
2308		if (!nvlist_empty(unsup_feat)) {
2309			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2310			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2311			nvlist_free(unsup_feat);
2312			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2313			    ENOTSUP));
2314		}
2315
2316		nvlist_free(unsup_feat);
2317	}
2318
2319	/*
2320	 * If the vdev guid sum doesn't match the uberblock, we have an
2321	 * incomplete configuration.  We first check to see if the pool
2322	 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2323	 * If it is, defer the vdev_guid_sum check till later so we
2324	 * can handle missing vdevs.
2325	 */
2326	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2327	    &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2328	    rvd->vdev_guid_sum != ub->ub_guid_sum)
2329		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2330
2331	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2332		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2333		spa_try_repair(spa, config);
2334		spa_config_exit(spa, SCL_ALL, FTAG);
2335		nvlist_free(spa->spa_config_splitting);
2336		spa->spa_config_splitting = NULL;
2337	}
2338
2339	/*
2340	 * Initialize internal SPA structures.
2341	 */
2342	spa->spa_state = POOL_STATE_ACTIVE;
2343	spa->spa_ubsync = spa->spa_uberblock;
2344	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2345	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2346	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2347	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2348	spa->spa_claim_max_txg = spa->spa_first_txg;
2349	spa->spa_prev_software_version = ub->ub_software_version;
2350
2351	error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2352	if (error)
2353		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2354	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2355
2356	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2357		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2358
2359	if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2360		boolean_t missing_feat_read = B_FALSE;
2361		nvlist_t *unsup_feat, *enabled_feat;
2362
2363		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2364		    &spa->spa_feat_for_read_obj) != 0) {
2365			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2366		}
2367
2368		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2369		    &spa->spa_feat_for_write_obj) != 0) {
2370			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2371		}
2372
2373		if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2374		    &spa->spa_feat_desc_obj) != 0) {
2375			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2376		}
2377
2378		enabled_feat = fnvlist_alloc();
2379		unsup_feat = fnvlist_alloc();
2380
2381		if (!spa_features_check(spa, B_FALSE,
2382		    unsup_feat, enabled_feat))
2383			missing_feat_read = B_TRUE;
2384
2385		if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2386			if (!spa_features_check(spa, B_TRUE,
2387			    unsup_feat, enabled_feat)) {
2388				missing_feat_write = B_TRUE;
2389			}
2390		}
2391
2392		fnvlist_add_nvlist(spa->spa_load_info,
2393		    ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2394
2395		if (!nvlist_empty(unsup_feat)) {
2396			fnvlist_add_nvlist(spa->spa_load_info,
2397			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2398		}
2399
2400		fnvlist_free(enabled_feat);
2401		fnvlist_free(unsup_feat);
2402
2403		if (!missing_feat_read) {
2404			fnvlist_add_boolean(spa->spa_load_info,
2405			    ZPOOL_CONFIG_CAN_RDONLY);
2406		}
2407
2408		/*
2409		 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2410		 * twofold: to determine whether the pool is available for
2411		 * import in read-write mode and (if it is not) whether the
2412		 * pool is available for import in read-only mode. If the pool
2413		 * is available for import in read-write mode, it is displayed
2414		 * as available in userland; if it is not available for import
2415		 * in read-only mode, it is displayed as unavailable in
2416		 * userland. If the pool is available for import in read-only
2417		 * mode but not read-write mode, it is displayed as unavailable
2418		 * in userland with a special note that the pool is actually
2419		 * available for open in read-only mode.
2420		 *
2421		 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2422		 * missing a feature for write, we must first determine whether
2423		 * the pool can be opened read-only before returning to
2424		 * userland in order to know whether to display the
2425		 * abovementioned note.
2426		 */
2427		if (missing_feat_read || (missing_feat_write &&
2428		    spa_writeable(spa))) {
2429			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2430			    ENOTSUP));
2431		}
2432
2433		/*
2434		 * Load refcounts for ZFS features from disk into an in-memory
2435		 * cache during SPA initialization.
2436		 */
2437		for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2438			uint64_t refcount;
2439
2440			error = feature_get_refcount_from_disk(spa,
2441			    &spa_feature_table[i], &refcount);
2442			if (error == 0) {
2443				spa->spa_feat_refcount_cache[i] = refcount;
2444			} else if (error == ENOTSUP) {
2445				spa->spa_feat_refcount_cache[i] =
2446				    SPA_FEATURE_DISABLED;
2447			} else {
2448				return (spa_vdev_err(rvd,
2449				    VDEV_AUX_CORRUPT_DATA, EIO));
2450			}
2451		}
2452	}
2453
2454	if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2455		if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2456		    &spa->spa_feat_enabled_txg_obj) != 0)
2457			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2458	}
2459
2460	spa->spa_is_initializing = B_TRUE;
2461	error = dsl_pool_open(spa->spa_dsl_pool);
2462	spa->spa_is_initializing = B_FALSE;
2463	if (error != 0)
2464		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2465
2466	if (!mosconfig) {
2467		uint64_t hostid;
2468		nvlist_t *policy = NULL, *nvconfig;
2469
2470		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2471			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2472
2473		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2474		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2475			char *hostname;
2476			unsigned long myhostid = 0;
2477
2478			VERIFY(nvlist_lookup_string(nvconfig,
2479			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2480
2481#ifdef	_KERNEL
2482			myhostid = zone_get_hostid(NULL);
2483#else	/* _KERNEL */
2484			/*
2485			 * We're emulating the system's hostid in userland, so
2486			 * we can't use zone_get_hostid().
2487			 */
2488			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2489#endif	/* _KERNEL */
2490			if (hostid != 0 && myhostid != 0 &&
2491			    hostid != myhostid) {
2492				nvlist_free(nvconfig);
2493				cmn_err(CE_WARN, "pool '%s' could not be "
2494				    "loaded as it was last accessed by "
2495				    "another system (host: %s hostid: 0x%lx). "
2496				    "See: http://illumos.org/msg/ZFS-8000-EY",
2497				    spa_name(spa), hostname,
2498				    (unsigned long)hostid);
2499				return (SET_ERROR(EBADF));
2500			}
2501		}
2502		if (nvlist_lookup_nvlist(spa->spa_config,
2503		    ZPOOL_REWIND_POLICY, &policy) == 0)
2504			VERIFY(nvlist_add_nvlist(nvconfig,
2505			    ZPOOL_REWIND_POLICY, policy) == 0);
2506
2507		spa_config_set(spa, nvconfig);
2508		spa_unload(spa);
2509		spa_deactivate(spa);
2510		spa_activate(spa, orig_mode);
2511
2512		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2513	}
2514
2515	/* Grab the secret checksum salt from the MOS. */
2516	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2517	    DMU_POOL_CHECKSUM_SALT, 1,
2518	    sizeof (spa->spa_cksum_salt.zcs_bytes),
2519	    spa->spa_cksum_salt.zcs_bytes);
2520	if (error == ENOENT) {
2521		/* Generate a new salt for subsequent use */
2522		(void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
2523		    sizeof (spa->spa_cksum_salt.zcs_bytes));
2524	} else if (error != 0) {
2525		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2526	}
2527
2528	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2529		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2530	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2531	if (error != 0)
2532		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2533
2534	/*
2535	 * Load the bit that tells us to use the new accounting function
2536	 * (raid-z deflation).  If we have an older pool, this will not
2537	 * be present.
2538	 */
2539	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2540	if (error != 0 && error != ENOENT)
2541		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2542
2543	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2544	    &spa->spa_creation_version);
2545	if (error != 0 && error != ENOENT)
2546		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2547
2548	/*
2549	 * Load the persistent error log.  If we have an older pool, this will
2550	 * not be present.
2551	 */
2552	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2553	if (error != 0 && error != ENOENT)
2554		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2555
2556	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2557	    &spa->spa_errlog_scrub);
2558	if (error != 0 && error != ENOENT)
2559		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2560
2561	/*
2562	 * Load the history object.  If we have an older pool, this
2563	 * will not be present.
2564	 */
2565	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2566	if (error != 0 && error != ENOENT)
2567		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2568
2569	/*
2570	 * If we're assembling the pool from the split-off vdevs of
2571	 * an existing pool, we don't want to attach the spares & cache
2572	 * devices.
2573	 */
2574
2575	/*
2576	 * Load any hot spares for this pool.
2577	 */
2578	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2579	if (error != 0 && error != ENOENT)
2580		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2581	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2582		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2583		if (load_nvlist(spa, spa->spa_spares.sav_object,
2584		    &spa->spa_spares.sav_config) != 0)
2585			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2586
2587		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2588		spa_load_spares(spa);
2589		spa_config_exit(spa, SCL_ALL, FTAG);
2590	} else if (error == 0) {
2591		spa->spa_spares.sav_sync = B_TRUE;
2592	}
2593
2594	/*
2595	 * Load any level 2 ARC devices for this pool.
2596	 */
2597	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2598	    &spa->spa_l2cache.sav_object);
2599	if (error != 0 && error != ENOENT)
2600		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2601	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2602		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2603		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2604		    &spa->spa_l2cache.sav_config) != 0)
2605			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2606
2607		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2608		spa_load_l2cache(spa);
2609		spa_config_exit(spa, SCL_ALL, FTAG);
2610	} else if (error == 0) {
2611		spa->spa_l2cache.sav_sync = B_TRUE;
2612	}
2613
2614	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2615
2616	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2617	if (error && error != ENOENT)
2618		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2619
2620	if (error == 0) {
2621		uint64_t autoreplace;
2622
2623		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2624		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2625		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2626		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2627		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2628		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2629		    &spa->spa_dedup_ditto);
2630
2631		spa->spa_autoreplace = (autoreplace != 0);
2632	}
2633
2634	/*
2635	 * If the 'autoreplace' property is set, then post a resource notifying
2636	 * the ZFS DE that it should not issue any faults for unopenable
2637	 * devices.  We also iterate over the vdevs, and post a sysevent for any
2638	 * unopenable vdevs so that the normal autoreplace handler can take
2639	 * over.
2640	 */
2641	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2642		spa_check_removed(spa->spa_root_vdev);
2643		/*
2644		 * For the import case, this is done in spa_import(), because
2645		 * at this point we're using the spare definitions from
2646		 * the MOS config, not necessarily from the userland config.
2647		 */
2648		if (state != SPA_LOAD_IMPORT) {
2649			spa_aux_check_removed(&spa->spa_spares);
2650			spa_aux_check_removed(&spa->spa_l2cache);
2651		}
2652	}
2653
2654	/*
2655	 * Load the vdev state for all toplevel vdevs.
2656	 */
2657	vdev_load(rvd);
2658
2659	/*
2660	 * Propagate the leaf DTLs we just loaded all the way up the tree.
2661	 */
2662	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2663	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2664	spa_config_exit(spa, SCL_ALL, FTAG);
2665
2666	/*
2667	 * Load the DDTs (dedup tables).
2668	 */
2669	error = ddt_load(spa);
2670	if (error != 0)
2671		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2672
2673	spa_update_dspace(spa);
2674
2675	/*
2676	 * Validate the config, using the MOS config to fill in any
2677	 * information which might be missing.  If we fail to validate
2678	 * the config then declare the pool unfit for use. If we're
2679	 * assembling a pool from a split, the log is not transferred
2680	 * over.
2681	 */
2682	if (type != SPA_IMPORT_ASSEMBLE) {
2683		nvlist_t *nvconfig;
2684
2685		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2686			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2687
2688		if (!spa_config_valid(spa, nvconfig)) {
2689			nvlist_free(nvconfig);
2690			return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2691			    ENXIO));
2692		}
2693		nvlist_free(nvconfig);
2694
2695		/*
2696		 * Now that we've validated the config, check the state of the
2697		 * root vdev.  If it can't be opened, it indicates one or
2698		 * more toplevel vdevs are faulted.
2699		 */
2700		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2701			return (SET_ERROR(ENXIO));
2702
2703		if (spa_writeable(spa) && spa_check_logs(spa)) {
2704			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2705			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2706		}
2707	}
2708
2709	if (missing_feat_write) {
2710		ASSERT(state == SPA_LOAD_TRYIMPORT);
2711
2712		/*
2713		 * At this point, we know that we can open the pool in
2714		 * read-only mode but not read-write mode. We now have enough
2715		 * information and can return to userland.
2716		 */
2717		return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2718	}
2719
2720	/*
2721	 * We've successfully opened the pool, verify that we're ready
2722	 * to start pushing transactions.
2723	 */
2724	if (state != SPA_LOAD_TRYIMPORT) {
2725		if (error = spa_load_verify(spa))
2726			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2727			    error));
2728	}
2729
2730	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2731	    spa->spa_load_max_txg == UINT64_MAX)) {
2732		dmu_tx_t *tx;
2733		int need_update = B_FALSE;
2734		dsl_pool_t *dp = spa_get_dsl(spa);
2735
2736		ASSERT(state != SPA_LOAD_TRYIMPORT);
2737
2738		/*
2739		 * Claim log blocks that haven't been committed yet.
2740		 * This must all happen in a single txg.
2741		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2742		 * invoked from zil_claim_log_block()'s i/o done callback.
2743		 * Price of rollback is that we abandon the log.
2744		 */
2745		spa->spa_claiming = B_TRUE;
2746
2747		tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
2748		(void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
2749		    zil_claim, tx, DS_FIND_CHILDREN);
2750		dmu_tx_commit(tx);
2751
2752		spa->spa_claiming = B_FALSE;
2753
2754		spa_set_log_state(spa, SPA_LOG_GOOD);
2755		spa->spa_sync_on = B_TRUE;
2756		txg_sync_start(spa->spa_dsl_pool);
2757
2758		/*
2759		 * Wait for all claims to sync.  We sync up to the highest
2760		 * claimed log block birth time so that claimed log blocks
2761		 * don't appear to be from the future.  spa_claim_max_txg
2762		 * will have been set for us by either zil_check_log_chain()
2763		 * (invoked from spa_check_logs()) or zil_claim() above.
2764		 */
2765		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2766
2767		/*
2768		 * If the config cache is stale, or we have uninitialized
2769		 * metaslabs (see spa_vdev_add()), then update the config.
2770		 *
2771		 * If this is a verbatim import, trust the current
2772		 * in-core spa_config and update the disk labels.
2773		 */
2774		if (config_cache_txg != spa->spa_config_txg ||
2775		    state == SPA_LOAD_IMPORT ||
2776		    state == SPA_LOAD_RECOVER ||
2777		    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2778			need_update = B_TRUE;
2779
2780		for (int c = 0; c < rvd->vdev_children; c++)
2781			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2782				need_update = B_TRUE;
2783
2784		/*
2785		 * Update the config cache asychronously in case we're the
2786		 * root pool, in which case the config cache isn't writable yet.
2787		 */
2788		if (need_update)
2789			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2790
2791		/*
2792		 * Check all DTLs to see if anything needs resilvering.
2793		 */
2794		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2795		    vdev_resilver_needed(rvd, NULL, NULL))
2796			spa_async_request(spa, SPA_ASYNC_RESILVER);
2797
2798		/*
2799		 * Log the fact that we booted up (so that we can detect if
2800		 * we rebooted in the middle of an operation).
2801		 */
2802		spa_history_log_version(spa, "open");
2803
2804		/*
2805		 * Delete any inconsistent datasets.
2806		 */
2807		(void) dmu_objset_find(spa_name(spa),
2808		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2809
2810		/*
2811		 * Clean up any stale temporary dataset userrefs.
2812		 */
2813		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2814	}
2815
2816	return (0);
2817}
2818
2819static int
2820spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2821{
2822	int mode = spa->spa_mode;
2823
2824	spa_unload(spa);
2825	spa_deactivate(spa);
2826
2827	spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
2828
2829	spa_activate(spa, mode);
2830	spa_async_suspend(spa);
2831
2832	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2833}
2834
2835/*
2836 * If spa_load() fails this function will try loading prior txg's. If
2837 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2838 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2839 * function will not rewind the pool and will return the same error as
2840 * spa_load().
2841 */
2842static int
2843spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2844    uint64_t max_request, int rewind_flags)
2845{
2846	nvlist_t *loadinfo = NULL;
2847	nvlist_t *config = NULL;
2848	int load_error, rewind_error;
2849	uint64_t safe_rewind_txg;
2850	uint64_t min_txg;
2851
2852	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2853		spa->spa_load_max_txg = spa->spa_load_txg;
2854		spa_set_log_state(spa, SPA_LOG_CLEAR);
2855	} else {
2856		spa->spa_load_max_txg = max_request;
2857		if (max_request != UINT64_MAX)
2858			spa->spa_extreme_rewind = B_TRUE;
2859	}
2860
2861	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2862	    mosconfig);
2863	if (load_error == 0)
2864		return (0);
2865
2866	if (spa->spa_root_vdev != NULL)
2867		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2868
2869	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2870	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2871
2872	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2873		nvlist_free(config);
2874		return (load_error);
2875	}
2876
2877	if (state == SPA_LOAD_RECOVER) {
2878		/* Price of rolling back is discarding txgs, including log */
2879		spa_set_log_state(spa, SPA_LOG_CLEAR);
2880	} else {
2881		/*
2882		 * If we aren't rolling back save the load info from our first
2883		 * import attempt so that we can restore it after attempting
2884		 * to rewind.
2885		 */
2886		loadinfo = spa->spa_load_info;
2887		spa->spa_load_info = fnvlist_alloc();
2888	}
2889
2890	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2891	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2892	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2893	    TXG_INITIAL : safe_rewind_txg;
2894
2895	/*
2896	 * Continue as long as we're finding errors, we're still within
2897	 * the acceptable rewind range, and we're still finding uberblocks
2898	 */
2899	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2900	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2901		if (spa->spa_load_max_txg < safe_rewind_txg)
2902			spa->spa_extreme_rewind = B_TRUE;
2903		rewind_error = spa_load_retry(spa, state, mosconfig);
2904	}
2905
2906	spa->spa_extreme_rewind = B_FALSE;
2907	spa->spa_load_max_txg = UINT64_MAX;
2908
2909	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2910		spa_config_set(spa, config);
2911
2912	if (state == SPA_LOAD_RECOVER) {
2913		ASSERT3P(loadinfo, ==, NULL);
2914		return (rewind_error);
2915	} else {
2916		/* Store the rewind info as part of the initial load info */
2917		fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2918		    spa->spa_load_info);
2919
2920		/* Restore the initial load info */
2921		fnvlist_free(spa->spa_load_info);
2922		spa->spa_load_info = loadinfo;
2923
2924		return (load_error);
2925	}
2926}
2927
2928/*
2929 * Pool Open/Import
2930 *
2931 * The import case is identical to an open except that the configuration is sent
2932 * down from userland, instead of grabbed from the configuration cache.  For the
2933 * case of an open, the pool configuration will exist in the
2934 * POOL_STATE_UNINITIALIZED state.
2935 *
2936 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2937 * the same time open the pool, without having to keep around the spa_t in some
2938 * ambiguous state.
2939 */
2940static int
2941spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2942    nvlist_t **config)
2943{
2944	spa_t *spa;
2945	spa_load_state_t state = SPA_LOAD_OPEN;
2946	int error;
2947	int locked = B_FALSE;
2948
2949	*spapp = NULL;
2950
2951	/*
2952	 * As disgusting as this is, we need to support recursive calls to this
2953	 * function because dsl_dir_open() is called during spa_load(), and ends
2954	 * up calling spa_open() again.  The real fix is to figure out how to
2955	 * avoid dsl_dir_open() calling this in the first place.
2956	 */
2957	if (mutex_owner(&spa_namespace_lock) != curthread) {
2958		mutex_enter(&spa_namespace_lock);
2959		locked = B_TRUE;
2960	}
2961
2962	if ((spa = spa_lookup(pool)) == NULL) {
2963		if (locked)
2964			mutex_exit(&spa_namespace_lock);
2965		return (SET_ERROR(ENOENT));
2966	}
2967
2968	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2969		zpool_rewind_policy_t policy;
2970
2971		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2972		    &policy);
2973		if (policy.zrp_request & ZPOOL_DO_REWIND)
2974			state = SPA_LOAD_RECOVER;
2975
2976		spa_activate(spa, spa_mode_global);
2977
2978		if (state != SPA_LOAD_RECOVER)
2979			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2980
2981		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2982		    policy.zrp_request);
2983
2984		if (error == EBADF) {
2985			/*
2986			 * If vdev_validate() returns failure (indicated by
2987			 * EBADF), it indicates that one of the vdevs indicates
2988			 * that the pool has been exported or destroyed.  If
2989			 * this is the case, the config cache is out of sync and
2990			 * we should remove the pool from the namespace.
2991			 */
2992			spa_unload(spa);
2993			spa_deactivate(spa);
2994			spa_config_sync(spa, B_TRUE, B_TRUE);
2995			spa_remove(spa);
2996			if (locked)
2997				mutex_exit(&spa_namespace_lock);
2998			return (SET_ERROR(ENOENT));
2999		}
3000
3001		if (error) {
3002			/*
3003			 * We can't open the pool, but we still have useful
3004			 * information: the state of each vdev after the
3005			 * attempted vdev_open().  Return this to the user.
3006			 */
3007			if (config != NULL && spa->spa_config) {
3008				VERIFY(nvlist_dup(spa->spa_config, config,
3009				    KM_SLEEP) == 0);
3010				VERIFY(nvlist_add_nvlist(*config,
3011				    ZPOOL_CONFIG_LOAD_INFO,
3012				    spa->spa_load_info) == 0);
3013			}
3014			spa_unload(spa);
3015			spa_deactivate(spa);
3016			spa->spa_last_open_failed = error;
3017			if (locked)
3018				mutex_exit(&spa_namespace_lock);
3019			*spapp = NULL;
3020			return (error);
3021		}
3022	}
3023
3024	spa_open_ref(spa, tag);
3025
3026	if (config != NULL)
3027		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3028
3029	/*
3030	 * If we've recovered the pool, pass back any information we
3031	 * gathered while doing the load.
3032	 */
3033	if (state == SPA_LOAD_RECOVER) {
3034		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
3035		    spa->spa_load_info) == 0);
3036	}
3037
3038	if (locked) {
3039		spa->spa_last_open_failed = 0;
3040		spa->spa_last_ubsync_txg = 0;
3041		spa->spa_load_txg = 0;
3042		mutex_exit(&spa_namespace_lock);
3043	}
3044
3045	*spapp = spa;
3046
3047	return (0);
3048}
3049
3050int
3051spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3052    nvlist_t **config)
3053{
3054	return (spa_open_common(name, spapp, tag, policy, config));
3055}
3056
3057int
3058spa_open(const char *name, spa_t **spapp, void *tag)
3059{
3060	return (spa_open_common(name, spapp, tag, NULL, NULL));
3061}
3062
3063/*
3064 * Lookup the given spa_t, incrementing the inject count in the process,
3065 * preventing it from being exported or destroyed.
3066 */
3067spa_t *
3068spa_inject_addref(char *name)
3069{
3070	spa_t *spa;
3071
3072	mutex_enter(&spa_namespace_lock);
3073	if ((spa = spa_lookup(name)) == NULL) {
3074		mutex_exit(&spa_namespace_lock);
3075		return (NULL);
3076	}
3077	spa->spa_inject_ref++;
3078	mutex_exit(&spa_namespace_lock);
3079
3080	return (spa);
3081}
3082
3083void
3084spa_inject_delref(spa_t *spa)
3085{
3086	mutex_enter(&spa_namespace_lock);
3087	spa->spa_inject_ref--;
3088	mutex_exit(&spa_namespace_lock);
3089}
3090
3091/*
3092 * Add spares device information to the nvlist.
3093 */
3094static void
3095spa_add_spares(spa_t *spa, nvlist_t *config)
3096{
3097	nvlist_t **spares;
3098	uint_t i, nspares;
3099	nvlist_t *nvroot;
3100	uint64_t guid;
3101	vdev_stat_t *vs;
3102	uint_t vsc;
3103	uint64_t pool;
3104
3105	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3106
3107	if (spa->spa_spares.sav_count == 0)
3108		return;
3109
3110	VERIFY(nvlist_lookup_nvlist(config,
3111	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3112	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3113	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3114	if (nspares != 0) {
3115		VERIFY(nvlist_add_nvlist_array(nvroot,
3116		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3117		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3118		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3119
3120		/*
3121		 * Go through and find any spares which have since been
3122		 * repurposed as an active spare.  If this is the case, update
3123		 * their status appropriately.
3124		 */
3125		for (i = 0; i < nspares; i++) {
3126			VERIFY(nvlist_lookup_uint64(spares[i],
3127			    ZPOOL_CONFIG_GUID, &guid) == 0);
3128			if (spa_spare_exists(guid, &pool, NULL) &&
3129			    pool != 0ULL) {
3130				VERIFY(nvlist_lookup_uint64_array(
3131				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
3132				    (uint64_t **)&vs, &vsc) == 0);
3133				vs->vs_state = VDEV_STATE_CANT_OPEN;
3134				vs->vs_aux = VDEV_AUX_SPARED;
3135			}
3136		}
3137	}
3138}
3139
3140/*
3141 * Add l2cache device information to the nvlist, including vdev stats.
3142 */
3143static void
3144spa_add_l2cache(spa_t *spa, nvlist_t *config)
3145{
3146	nvlist_t **l2cache;
3147	uint_t i, j, nl2cache;
3148	nvlist_t *nvroot;
3149	uint64_t guid;
3150	vdev_t *vd;
3151	vdev_stat_t *vs;
3152	uint_t vsc;
3153
3154	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3155
3156	if (spa->spa_l2cache.sav_count == 0)
3157		return;
3158
3159	VERIFY(nvlist_lookup_nvlist(config,
3160	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3161	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3162	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3163	if (nl2cache != 0) {
3164		VERIFY(nvlist_add_nvlist_array(nvroot,
3165		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3166		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3167		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3168
3169		/*
3170		 * Update level 2 cache device stats.
3171		 */
3172
3173		for (i = 0; i < nl2cache; i++) {
3174			VERIFY(nvlist_lookup_uint64(l2cache[i],
3175			    ZPOOL_CONFIG_GUID, &guid) == 0);
3176
3177			vd = NULL;
3178			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3179				if (guid ==
3180				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3181					vd = spa->spa_l2cache.sav_vdevs[j];
3182					break;
3183				}
3184			}
3185			ASSERT(vd != NULL);
3186
3187			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3188			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3189			    == 0);
3190			vdev_get_stats(vd, vs);
3191		}
3192	}
3193}
3194
3195static void
3196spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3197{
3198	nvlist_t *features;
3199	zap_cursor_t zc;
3200	zap_attribute_t za;
3201
3202	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3203	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3204
3205	if (spa->spa_feat_for_read_obj != 0) {
3206		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3207		    spa->spa_feat_for_read_obj);
3208		    zap_cursor_retrieve(&zc, &za) == 0;
3209		    zap_cursor_advance(&zc)) {
3210			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3211			    za.za_num_integers == 1);
3212			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3213			    za.za_first_integer));
3214		}
3215		zap_cursor_fini(&zc);
3216	}
3217
3218	if (spa->spa_feat_for_write_obj != 0) {
3219		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3220		    spa->spa_feat_for_write_obj);
3221		    zap_cursor_retrieve(&zc, &za) == 0;
3222		    zap_cursor_advance(&zc)) {
3223			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3224			    za.za_num_integers == 1);
3225			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3226			    za.za_first_integer));
3227		}
3228		zap_cursor_fini(&zc);
3229	}
3230
3231	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3232	    features) == 0);
3233	nvlist_free(features);
3234}
3235
3236int
3237spa_get_stats(const char *name, nvlist_t **config,
3238    char *altroot, size_t buflen)
3239{
3240	int error;
3241	spa_t *spa;
3242
3243	*config = NULL;
3244	error = spa_open_common(name, &spa, FTAG, NULL, config);
3245
3246	if (spa != NULL) {
3247		/*
3248		 * This still leaves a window of inconsistency where the spares
3249		 * or l2cache devices could change and the config would be
3250		 * self-inconsistent.
3251		 */
3252		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3253
3254		if (*config != NULL) {
3255			uint64_t loadtimes[2];
3256
3257			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3258			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3259			VERIFY(nvlist_add_uint64_array(*config,
3260			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3261
3262			VERIFY(nvlist_add_uint64(*config,
3263			    ZPOOL_CONFIG_ERRCOUNT,
3264			    spa_get_errlog_size(spa)) == 0);
3265
3266			if (spa_suspended(spa))
3267				VERIFY(nvlist_add_uint64(*config,
3268				    ZPOOL_CONFIG_SUSPENDED,
3269				    spa->spa_failmode) == 0);
3270
3271			spa_add_spares(spa, *config);
3272			spa_add_l2cache(spa, *config);
3273			spa_add_feature_stats(spa, *config);
3274		}
3275	}
3276
3277	/*
3278	 * We want to get the alternate root even for faulted pools, so we cheat
3279	 * and call spa_lookup() directly.
3280	 */
3281	if (altroot) {
3282		if (spa == NULL) {
3283			mutex_enter(&spa_namespace_lock);
3284			spa = spa_lookup(name);
3285			if (spa)
3286				spa_altroot(spa, altroot, buflen);
3287			else
3288				altroot[0] = '\0';
3289			spa = NULL;
3290			mutex_exit(&spa_namespace_lock);
3291		} else {
3292			spa_altroot(spa, altroot, buflen);
3293		}
3294	}
3295
3296	if (spa != NULL) {
3297		spa_config_exit(spa, SCL_CONFIG, FTAG);
3298		spa_close(spa, FTAG);
3299	}
3300
3301	return (error);
3302}
3303
3304/*
3305 * Validate that the auxiliary device array is well formed.  We must have an
3306 * array of nvlists, each which describes a valid leaf vdev.  If this is an
3307 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3308 * specified, as long as they are well-formed.
3309 */
3310static int
3311spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3312    spa_aux_vdev_t *sav, const char *config, uint64_t version,
3313    vdev_labeltype_t label)
3314{
3315	nvlist_t **dev;
3316	uint_t i, ndev;
3317	vdev_t *vd;
3318	int error;
3319
3320	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3321
3322	/*
3323	 * It's acceptable to have no devs specified.
3324	 */
3325	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3326		return (0);
3327
3328	if (ndev == 0)
3329		return (SET_ERROR(EINVAL));
3330
3331	/*
3332	 * Make sure the pool is formatted with a version that supports this
3333	 * device type.
3334	 */
3335	if (spa_version(spa) < version)
3336		return (SET_ERROR(ENOTSUP));
3337
3338	/*
3339	 * Set the pending device list so we correctly handle device in-use
3340	 * checking.
3341	 */
3342	sav->sav_pending = dev;
3343	sav->sav_npending = ndev;
3344
3345	for (i = 0; i < ndev; i++) {
3346		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3347		    mode)) != 0)
3348			goto out;
3349
3350		if (!vd->vdev_ops->vdev_op_leaf) {
3351			vdev_free(vd);
3352			error = SET_ERROR(EINVAL);
3353			goto out;
3354		}
3355
3356		/*
3357		 * The L2ARC currently only supports disk devices in
3358		 * kernel context.  For user-level testing, we allow it.
3359		 */
3360#ifdef _KERNEL
3361		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3362		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3363			error = SET_ERROR(ENOTBLK);
3364			vdev_free(vd);
3365			goto out;
3366		}
3367#endif
3368		vd->vdev_top = vd;
3369
3370		if ((error = vdev_open(vd)) == 0 &&
3371		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
3372			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3373			    vd->vdev_guid) == 0);
3374		}
3375
3376		vdev_free(vd);
3377
3378		if (error &&
3379		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3380			goto out;
3381		else
3382			error = 0;
3383	}
3384
3385out:
3386	sav->sav_pending = NULL;
3387	sav->sav_npending = 0;
3388	return (error);
3389}
3390
3391static int
3392spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3393{
3394	int error;
3395
3396	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3397
3398	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3399	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3400	    VDEV_LABEL_SPARE)) != 0) {
3401		return (error);
3402	}
3403
3404	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3405	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3406	    VDEV_LABEL_L2CACHE));
3407}
3408
3409static void
3410spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3411    const char *config)
3412{
3413	int i;
3414
3415	if (sav->sav_config != NULL) {
3416		nvlist_t **olddevs;
3417		uint_t oldndevs;
3418		nvlist_t **newdevs;
3419
3420		/*
3421		 * Generate new dev list by concatentating with the
3422		 * current dev list.
3423		 */
3424		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3425		    &olddevs, &oldndevs) == 0);
3426
3427		newdevs = kmem_alloc(sizeof (void *) *
3428		    (ndevs + oldndevs), KM_SLEEP);
3429		for (i = 0; i < oldndevs; i++)
3430			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3431			    KM_SLEEP) == 0);
3432		for (i = 0; i < ndevs; i++)
3433			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3434			    KM_SLEEP) == 0);
3435
3436		VERIFY(nvlist_remove(sav->sav_config, config,
3437		    DATA_TYPE_NVLIST_ARRAY) == 0);
3438
3439		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3440		    config, newdevs, ndevs + oldndevs) == 0);
3441		for (i = 0; i < oldndevs + ndevs; i++)
3442			nvlist_free(newdevs[i]);
3443		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3444	} else {
3445		/*
3446		 * Generate a new dev list.
3447		 */
3448		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3449		    KM_SLEEP) == 0);
3450		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3451		    devs, ndevs) == 0);
3452	}
3453}
3454
3455/*
3456 * Stop and drop level 2 ARC devices
3457 */
3458void
3459spa_l2cache_drop(spa_t *spa)
3460{
3461	vdev_t *vd;
3462	int i;
3463	spa_aux_vdev_t *sav = &spa->spa_l2cache;
3464
3465	for (i = 0; i < sav->sav_count; i++) {
3466		uint64_t pool;
3467
3468		vd = sav->sav_vdevs[i];
3469		ASSERT(vd != NULL);
3470
3471		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3472		    pool != 0ULL && l2arc_vdev_present(vd))
3473			l2arc_remove_vdev(vd);
3474	}
3475}
3476
3477/*
3478 * Pool Creation
3479 */
3480int
3481spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3482    nvlist_t *zplprops)
3483{
3484	spa_t *spa;
3485	char *altroot = NULL;
3486	vdev_t *rvd;
3487	dsl_pool_t *dp;
3488	dmu_tx_t *tx;
3489	int error = 0;
3490	uint64_t txg = TXG_INITIAL;
3491	nvlist_t **spares, **l2cache;
3492	uint_t nspares, nl2cache;
3493	uint64_t version, obj;
3494	boolean_t has_features;
3495
3496	/*
3497	 * If this pool already exists, return failure.
3498	 */
3499	mutex_enter(&spa_namespace_lock);
3500	if (spa_lookup(pool) != NULL) {
3501		mutex_exit(&spa_namespace_lock);
3502		return (SET_ERROR(EEXIST));
3503	}
3504
3505	/*
3506	 * Allocate a new spa_t structure.
3507	 */
3508	(void) nvlist_lookup_string(props,
3509	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3510	spa = spa_add(pool, NULL, altroot);
3511	spa_activate(spa, spa_mode_global);
3512
3513	if (props && (error = spa_prop_validate(spa, props))) {
3514		spa_deactivate(spa);
3515		spa_remove(spa);
3516		mutex_exit(&spa_namespace_lock);
3517		return (error);
3518	}
3519
3520	has_features = B_FALSE;
3521	for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3522	    elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3523		if (zpool_prop_feature(nvpair_name(elem)))
3524			has_features = B_TRUE;
3525	}
3526
3527	if (has_features || nvlist_lookup_uint64(props,
3528	    zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3529		version = SPA_VERSION;
3530	}
3531	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3532
3533	spa->spa_first_txg = txg;
3534	spa->spa_uberblock.ub_txg = txg - 1;
3535	spa->spa_uberblock.ub_version = version;
3536	spa->spa_ubsync = spa->spa_uberblock;
3537
3538	/*
3539	 * Create "The Godfather" zio to hold all async IOs
3540	 */
3541	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
3542	    KM_SLEEP);
3543	for (int i = 0; i < max_ncpus; i++) {
3544		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
3545		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
3546		    ZIO_FLAG_GODFATHER);
3547	}
3548
3549	/*
3550	 * Create the root vdev.
3551	 */
3552	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3553
3554	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3555
3556	ASSERT(error != 0 || rvd != NULL);
3557	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3558
3559	if (error == 0 && !zfs_allocatable_devs(nvroot))
3560		error = SET_ERROR(EINVAL);
3561
3562	if (error == 0 &&
3563	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3564	    (error = spa_validate_aux(spa, nvroot, txg,
3565	    VDEV_ALLOC_ADD)) == 0) {
3566		for (int c = 0; c < rvd->vdev_children; c++) {
3567			vdev_metaslab_set_size(rvd->vdev_child[c]);
3568			vdev_expand(rvd->vdev_child[c], txg);
3569		}
3570	}
3571
3572	spa_config_exit(spa, SCL_ALL, FTAG);
3573
3574	if (error != 0) {
3575		spa_unload(spa);
3576		spa_deactivate(spa);
3577		spa_remove(spa);
3578		mutex_exit(&spa_namespace_lock);
3579		return (error);
3580	}
3581
3582	/*
3583	 * Get the list of spares, if specified.
3584	 */
3585	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3586	    &spares, &nspares) == 0) {
3587		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3588		    KM_SLEEP) == 0);
3589		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3590		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3591		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3592		spa_load_spares(spa);
3593		spa_config_exit(spa, SCL_ALL, FTAG);
3594		spa->spa_spares.sav_sync = B_TRUE;
3595	}
3596
3597	/*
3598	 * Get the list of level 2 cache devices, if specified.
3599	 */
3600	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3601	    &l2cache, &nl2cache) == 0) {
3602		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3603		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3604		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3605		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3606		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3607		spa_load_l2cache(spa);
3608		spa_config_exit(spa, SCL_ALL, FTAG);
3609		spa->spa_l2cache.sav_sync = B_TRUE;
3610	}
3611
3612	spa->spa_is_initializing = B_TRUE;
3613	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3614	spa->spa_meta_objset = dp->dp_meta_objset;
3615	spa->spa_is_initializing = B_FALSE;
3616
3617	/*
3618	 * Create DDTs (dedup tables).
3619	 */
3620	ddt_create(spa);
3621
3622	spa_update_dspace(spa);
3623
3624	tx = dmu_tx_create_assigned(dp, txg);
3625
3626	/*
3627	 * Create the pool config object.
3628	 */
3629	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3630	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3631	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3632
3633	if (zap_add(spa->spa_meta_objset,
3634	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3635	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3636		cmn_err(CE_PANIC, "failed to add pool config");
3637	}
3638
3639	if (spa_version(spa) >= SPA_VERSION_FEATURES)
3640		spa_feature_create_zap_objects(spa, tx);
3641
3642	if (zap_add(spa->spa_meta_objset,
3643	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3644	    sizeof (uint64_t), 1, &version, tx) != 0) {
3645		cmn_err(CE_PANIC, "failed to add pool version");
3646	}
3647
3648	/* Newly created pools with the right version are always deflated. */
3649	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3650		spa->spa_deflate = TRUE;
3651		if (zap_add(spa->spa_meta_objset,
3652		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3653		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3654			cmn_err(CE_PANIC, "failed to add deflate");
3655		}
3656	}
3657
3658	/*
3659	 * Create the deferred-free bpobj.  Turn off compression
3660	 * because sync-to-convergence takes longer if the blocksize
3661	 * keeps changing.
3662	 */
3663	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3664	dmu_object_set_compress(spa->spa_meta_objset, obj,
3665	    ZIO_COMPRESS_OFF, tx);
3666	if (zap_add(spa->spa_meta_objset,
3667	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3668	    sizeof (uint64_t), 1, &obj, tx) != 0) {
3669		cmn_err(CE_PANIC, "failed to add bpobj");
3670	}
3671	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3672	    spa->spa_meta_objset, obj));
3673
3674	/*
3675	 * Create the pool's history object.
3676	 */
3677	if (version >= SPA_VERSION_ZPOOL_HISTORY)
3678		spa_history_create_obj(spa, tx);
3679
3680	/*
3681	 * Generate some random noise for salted checksums to operate on.
3682	 */
3683	(void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
3684	    sizeof (spa->spa_cksum_salt.zcs_bytes));
3685
3686	/*
3687	 * Set pool properties.
3688	 */
3689	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3690	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3691	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3692	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3693
3694	if (props != NULL) {
3695		spa_configfile_set(spa, props, B_FALSE);
3696		spa_sync_props(props, tx);
3697	}
3698
3699	dmu_tx_commit(tx);
3700
3701	spa->spa_sync_on = B_TRUE;
3702	txg_sync_start(spa->spa_dsl_pool);
3703
3704	/*
3705	 * We explicitly wait for the first transaction to complete so that our
3706	 * bean counters are appropriately updated.
3707	 */
3708	txg_wait_synced(spa->spa_dsl_pool, txg);
3709
3710	spa_config_sync(spa, B_FALSE, B_TRUE);
3711	spa_event_notify(spa, NULL, ESC_ZFS_POOL_CREATE);
3712
3713	spa_history_log_version(spa, "create");
3714
3715	/*
3716	 * Don't count references from objsets that are already closed
3717	 * and are making their way through the eviction process.
3718	 */
3719	spa_evicting_os_wait(spa);
3720	spa->spa_minref = refcount_count(&spa->spa_refcount);
3721
3722	mutex_exit(&spa_namespace_lock);
3723
3724	return (0);
3725}
3726
3727#ifdef _KERNEL
3728/*
3729 * Get the root pool information from the root disk, then import the root pool
3730 * during the system boot up time.
3731 */
3732extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3733
3734static nvlist_t *
3735spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3736{
3737	nvlist_t *config;
3738	nvlist_t *nvtop, *nvroot;
3739	uint64_t pgid;
3740
3741	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3742		return (NULL);
3743
3744	/*
3745	 * Add this top-level vdev to the child array.
3746	 */
3747	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3748	    &nvtop) == 0);
3749	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3750	    &pgid) == 0);
3751	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3752
3753	/*
3754	 * Put this pool's top-level vdevs into a root vdev.
3755	 */
3756	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3757	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3758	    VDEV_TYPE_ROOT) == 0);
3759	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3760	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3761	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3762	    &nvtop, 1) == 0);
3763
3764	/*
3765	 * Replace the existing vdev_tree with the new root vdev in
3766	 * this pool's configuration (remove the old, add the new).
3767	 */
3768	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3769	nvlist_free(nvroot);
3770	return (config);
3771}
3772
3773/*
3774 * Walk the vdev tree and see if we can find a device with "better"
3775 * configuration. A configuration is "better" if the label on that
3776 * device has a more recent txg.
3777 */
3778static void
3779spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3780{
3781	for (int c = 0; c < vd->vdev_children; c++)
3782		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3783
3784	if (vd->vdev_ops->vdev_op_leaf) {
3785		nvlist_t *label;
3786		uint64_t label_txg;
3787
3788		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3789		    &label) != 0)
3790			return;
3791
3792		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3793		    &label_txg) == 0);
3794
3795		/*
3796		 * Do we have a better boot device?
3797		 */
3798		if (label_txg > *txg) {
3799			*txg = label_txg;
3800			*avd = vd;
3801		}
3802		nvlist_free(label);
3803	}
3804}
3805
3806/*
3807 * Import a root pool.
3808 *
3809 * For x86. devpath_list will consist of devid and/or physpath name of
3810 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3811 * The GRUB "findroot" command will return the vdev we should boot.
3812 *
3813 * For Sparc, devpath_list consists the physpath name of the booting device
3814 * no matter the rootpool is a single device pool or a mirrored pool.
3815 * e.g.
3816 *	"/pci@1f,0/ide@d/disk@0,0:a"
3817 */
3818int
3819spa_import_rootpool(char *devpath, char *devid)
3820{
3821	spa_t *spa;
3822	vdev_t *rvd, *bvd, *avd = NULL;
3823	nvlist_t *config, *nvtop;
3824	uint64_t guid, txg;
3825	char *pname;
3826	int error;
3827
3828	/*
3829	 * Read the label from the boot device and generate a configuration.
3830	 */
3831	config = spa_generate_rootconf(devpath, devid, &guid);
3832#if defined(_OBP) && defined(_KERNEL)
3833	if (config == NULL) {
3834		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3835			/* iscsi boot */
3836			get_iscsi_bootpath_phy(devpath);
3837			config = spa_generate_rootconf(devpath, devid, &guid);
3838		}
3839	}
3840#endif
3841	if (config == NULL) {
3842		cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3843		    devpath);
3844		return (SET_ERROR(EIO));
3845	}
3846
3847	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3848	    &pname) == 0);
3849	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3850
3851	mutex_enter(&spa_namespace_lock);
3852	if ((spa = spa_lookup(pname)) != NULL) {
3853		/*
3854		 * Remove the existing root pool from the namespace so that we
3855		 * can replace it with the correct config we just read in.
3856		 */
3857		spa_remove(spa);
3858	}
3859
3860	spa = spa_add(pname, config, NULL);
3861	spa->spa_is_root = B_TRUE;
3862	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3863
3864	/*
3865	 * Build up a vdev tree based on the boot device's label config.
3866	 */
3867	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3868	    &nvtop) == 0);
3869	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3870	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3871	    VDEV_ALLOC_ROOTPOOL);
3872	spa_config_exit(spa, SCL_ALL, FTAG);
3873	if (error) {
3874		mutex_exit(&spa_namespace_lock);
3875		nvlist_free(config);
3876		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3877		    pname);
3878		return (error);
3879	}
3880
3881	/*
3882	 * Get the boot vdev.
3883	 */
3884	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3885		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3886		    (u_longlong_t)guid);
3887		error = SET_ERROR(ENOENT);
3888		goto out;
3889	}
3890
3891	/*
3892	 * Determine if there is a better boot device.
3893	 */
3894	avd = bvd;
3895	spa_alt_rootvdev(rvd, &avd, &txg);
3896	if (avd != bvd) {
3897		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3898		    "try booting from '%s'", avd->vdev_path);
3899		error = SET_ERROR(EINVAL);
3900		goto out;
3901	}
3902
3903	/*
3904	 * If the boot device is part of a spare vdev then ensure that
3905	 * we're booting off the active spare.
3906	 */
3907	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3908	    !bvd->vdev_isspare) {
3909		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3910		    "try booting from '%s'",
3911		    bvd->vdev_parent->
3912		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3913		error = SET_ERROR(EINVAL);
3914		goto out;
3915	}
3916
3917	error = 0;
3918out:
3919	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3920	vdev_free(rvd);
3921	spa_config_exit(spa, SCL_ALL, FTAG);
3922	mutex_exit(&spa_namespace_lock);
3923
3924	nvlist_free(config);
3925	return (error);
3926}
3927
3928#endif
3929
3930/*
3931 * Import a non-root pool into the system.
3932 */
3933int
3934spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3935{
3936	spa_t *spa;
3937	char *altroot = NULL;
3938	spa_load_state_t state = SPA_LOAD_IMPORT;
3939	zpool_rewind_policy_t policy;
3940	uint64_t mode = spa_mode_global;
3941	uint64_t readonly = B_FALSE;
3942	int error;
3943	nvlist_t *nvroot;
3944	nvlist_t **spares, **l2cache;
3945	uint_t nspares, nl2cache;
3946
3947	/*
3948	 * If a pool with this name exists, return failure.
3949	 */
3950	mutex_enter(&spa_namespace_lock);
3951	if (spa_lookup(pool) != NULL) {
3952		mutex_exit(&spa_namespace_lock);
3953		return (SET_ERROR(EEXIST));
3954	}
3955
3956	/*
3957	 * Create and initialize the spa structure.
3958	 */
3959	(void) nvlist_lookup_string(props,
3960	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3961	(void) nvlist_lookup_uint64(props,
3962	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3963	if (readonly)
3964		mode = FREAD;
3965	spa = spa_add(pool, config, altroot);
3966	spa->spa_import_flags = flags;
3967
3968	/*
3969	 * Verbatim import - Take a pool and insert it into the namespace
3970	 * as if it had been loaded at boot.
3971	 */
3972	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3973		if (props != NULL)
3974			spa_configfile_set(spa, props, B_FALSE);
3975
3976		spa_config_sync(spa, B_FALSE, B_TRUE);
3977		spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
3978
3979		mutex_exit(&spa_namespace_lock);
3980		return (0);
3981	}
3982
3983	spa_activate(spa, mode);
3984
3985	/*
3986	 * Don't start async tasks until we know everything is healthy.
3987	 */
3988	spa_async_suspend(spa);
3989
3990	zpool_get_rewind_policy(config, &policy);
3991	if (policy.zrp_request & ZPOOL_DO_REWIND)
3992		state = SPA_LOAD_RECOVER;
3993
3994	/*
3995	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3996	 * because the user-supplied config is actually the one to trust when
3997	 * doing an import.
3998	 */
3999	if (state != SPA_LOAD_RECOVER)
4000		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4001
4002	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4003	    policy.zrp_request);
4004
4005	/*
4006	 * Propagate anything learned while loading the pool and pass it
4007	 * back to caller (i.e. rewind info, missing devices, etc).
4008	 */
4009	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4010	    spa->spa_load_info) == 0);
4011
4012	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4013	/*
4014	 * Toss any existing sparelist, as it doesn't have any validity
4015	 * anymore, and conflicts with spa_has_spare().
4016	 */
4017	if (spa->spa_spares.sav_config) {
4018		nvlist_free(spa->spa_spares.sav_config);
4019		spa->spa_spares.sav_config = NULL;
4020		spa_load_spares(spa);
4021	}
4022	if (spa->spa_l2cache.sav_config) {
4023		nvlist_free(spa->spa_l2cache.sav_config);
4024		spa->spa_l2cache.sav_config = NULL;
4025		spa_load_l2cache(spa);
4026	}
4027
4028	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4029	    &nvroot) == 0);
4030	if (error == 0)
4031		error = spa_validate_aux(spa, nvroot, -1ULL,
4032		    VDEV_ALLOC_SPARE);
4033	if (error == 0)
4034		error = spa_validate_aux(spa, nvroot, -1ULL,
4035		    VDEV_ALLOC_L2CACHE);
4036	spa_config_exit(spa, SCL_ALL, FTAG);
4037
4038	if (props != NULL)
4039		spa_configfile_set(spa, props, B_FALSE);
4040
4041	if (error != 0 || (props && spa_writeable(spa) &&
4042	    (error = spa_prop_set(spa, props)))) {
4043		spa_unload(spa);
4044		spa_deactivate(spa);
4045		spa_remove(spa);
4046		mutex_exit(&spa_namespace_lock);
4047		return (error);
4048	}
4049
4050	spa_async_resume(spa);
4051
4052	/*
4053	 * Override any spares and level 2 cache devices as specified by
4054	 * the user, as these may have correct device names/devids, etc.
4055	 */
4056	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4057	    &spares, &nspares) == 0) {
4058		if (spa->spa_spares.sav_config)
4059			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4060			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4061		else
4062			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4063			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4064		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4065		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4066		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4067		spa_load_spares(spa);
4068		spa_config_exit(spa, SCL_ALL, FTAG);
4069		spa->spa_spares.sav_sync = B_TRUE;
4070	}
4071	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4072	    &l2cache, &nl2cache) == 0) {
4073		if (spa->spa_l2cache.sav_config)
4074			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4075			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4076		else
4077			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4078			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4079		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4080		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4081		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4082		spa_load_l2cache(spa);
4083		spa_config_exit(spa, SCL_ALL, FTAG);
4084		spa->spa_l2cache.sav_sync = B_TRUE;
4085	}
4086
4087	/*
4088	 * Check for any removed devices.
4089	 */
4090	if (spa->spa_autoreplace) {
4091		spa_aux_check_removed(&spa->spa_spares);
4092		spa_aux_check_removed(&spa->spa_l2cache);
4093	}
4094
4095	if (spa_writeable(spa)) {
4096		/*
4097		 * Update the config cache to include the newly-imported pool.
4098		 */
4099		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4100	}
4101
4102	/*
4103	 * It's possible that the pool was expanded while it was exported.
4104	 * We kick off an async task to handle this for us.
4105	 */
4106	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4107
4108	spa_history_log_version(spa, "import");
4109
4110	spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4111
4112	mutex_exit(&spa_namespace_lock);
4113
4114	return (0);
4115}
4116
4117nvlist_t *
4118spa_tryimport(nvlist_t *tryconfig)
4119{
4120	nvlist_t *config = NULL;
4121	char *poolname;
4122	spa_t *spa;
4123	uint64_t state;
4124	int error;
4125
4126	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4127		return (NULL);
4128
4129	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4130		return (NULL);
4131
4132	/*
4133	 * Create and initialize the spa structure.
4134	 */
4135	mutex_enter(&spa_namespace_lock);
4136	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4137	spa_activate(spa, FREAD);
4138
4139	/*
4140	 * Pass off the heavy lifting to spa_load().
4141	 * Pass TRUE for mosconfig because the user-supplied config
4142	 * is actually the one to trust when doing an import.
4143	 */
4144	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4145
4146	/*
4147	 * If 'tryconfig' was at least parsable, return the current config.
4148	 */
4149	if (spa->spa_root_vdev != NULL) {
4150		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4151		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4152		    poolname) == 0);
4153		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4154		    state) == 0);
4155		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4156		    spa->spa_uberblock.ub_timestamp) == 0);
4157		VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4158		    spa->spa_load_info) == 0);
4159
4160		/*
4161		 * If the bootfs property exists on this pool then we
4162		 * copy it out so that external consumers can tell which
4163		 * pools are bootable.
4164		 */
4165		if ((!error || error == EEXIST) && spa->spa_bootfs) {
4166			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4167
4168			/*
4169			 * We have to play games with the name since the
4170			 * pool was opened as TRYIMPORT_NAME.
4171			 */
4172			if (dsl_dsobj_to_dsname(spa_name(spa),
4173			    spa->spa_bootfs, tmpname) == 0) {
4174				char *cp;
4175				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4176
4177				cp = strchr(tmpname, '/');
4178				if (cp == NULL) {
4179					(void) strlcpy(dsname, tmpname,
4180					    MAXPATHLEN);
4181				} else {
4182					(void) snprintf(dsname, MAXPATHLEN,
4183					    "%s/%s", poolname, ++cp);
4184				}
4185				VERIFY(nvlist_add_string(config,
4186				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4187				kmem_free(dsname, MAXPATHLEN);
4188			}
4189			kmem_free(tmpname, MAXPATHLEN);
4190		}
4191
4192		/*
4193		 * Add the list of hot spares and level 2 cache devices.
4194		 */
4195		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4196		spa_add_spares(spa, config);
4197		spa_add_l2cache(spa, config);
4198		spa_config_exit(spa, SCL_CONFIG, FTAG);
4199	}
4200
4201	spa_unload(spa);
4202	spa_deactivate(spa);
4203	spa_remove(spa);
4204	mutex_exit(&spa_namespace_lock);
4205
4206	return (config);
4207}
4208
4209/*
4210 * Pool export/destroy
4211 *
4212 * The act of destroying or exporting a pool is very simple.  We make sure there
4213 * is no more pending I/O and any references to the pool are gone.  Then, we
4214 * update the pool state and sync all the labels to disk, removing the
4215 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4216 * we don't sync the labels or remove the configuration cache.
4217 */
4218static int
4219spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4220    boolean_t force, boolean_t hardforce)
4221{
4222	spa_t *spa;
4223
4224	if (oldconfig)
4225		*oldconfig = NULL;
4226
4227	if (!(spa_mode_global & FWRITE))
4228		return (SET_ERROR(EROFS));
4229
4230	mutex_enter(&spa_namespace_lock);
4231	if ((spa = spa_lookup(pool)) == NULL) {
4232		mutex_exit(&spa_namespace_lock);
4233		return (SET_ERROR(ENOENT));
4234	}
4235
4236	/*
4237	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4238	 * reacquire the namespace lock, and see if we can export.
4239	 */
4240	spa_open_ref(spa, FTAG);
4241	mutex_exit(&spa_namespace_lock);
4242	spa_async_suspend(spa);
4243	mutex_enter(&spa_namespace_lock);
4244	spa_close(spa, FTAG);
4245
4246	/*
4247	 * The pool will be in core if it's openable,
4248	 * in which case we can modify its state.
4249	 */
4250	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4251		/*
4252		 * Objsets may be open only because they're dirty, so we
4253		 * have to force it to sync before checking spa_refcnt.
4254		 */
4255		txg_wait_synced(spa->spa_dsl_pool, 0);
4256		spa_evicting_os_wait(spa);
4257
4258		/*
4259		 * A pool cannot be exported or destroyed if there are active
4260		 * references.  If we are resetting a pool, allow references by
4261		 * fault injection handlers.
4262		 */
4263		if (!spa_refcount_zero(spa) ||
4264		    (spa->spa_inject_ref != 0 &&
4265		    new_state != POOL_STATE_UNINITIALIZED)) {
4266			spa_async_resume(spa);
4267			mutex_exit(&spa_namespace_lock);
4268			return (SET_ERROR(EBUSY));
4269		}
4270
4271		/*
4272		 * A pool cannot be exported if it has an active shared spare.
4273		 * This is to prevent other pools stealing the active spare
4274		 * from an exported pool. At user's own will, such pool can
4275		 * be forcedly exported.
4276		 */
4277		if (!force && new_state == POOL_STATE_EXPORTED &&
4278		    spa_has_active_shared_spare(spa)) {
4279			spa_async_resume(spa);
4280			mutex_exit(&spa_namespace_lock);
4281			return (SET_ERROR(EXDEV));
4282		}
4283
4284		/*
4285		 * We want this to be reflected on every label,
4286		 * so mark them all dirty.  spa_unload() will do the
4287		 * final sync that pushes these changes out.
4288		 */
4289		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4290			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4291			spa->spa_state = new_state;
4292			spa->spa_final_txg = spa_last_synced_txg(spa) +
4293			    TXG_DEFER_SIZE + 1;
4294			vdev_config_dirty(spa->spa_root_vdev);
4295			spa_config_exit(spa, SCL_ALL, FTAG);
4296		}
4297	}
4298
4299	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4300
4301	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4302		spa_unload(spa);
4303		spa_deactivate(spa);
4304	}
4305
4306	if (oldconfig && spa->spa_config)
4307		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4308
4309	if (new_state != POOL_STATE_UNINITIALIZED) {
4310		if (!hardforce)
4311			spa_config_sync(spa, B_TRUE, B_TRUE);
4312		spa_remove(spa);
4313	}
4314	mutex_exit(&spa_namespace_lock);
4315
4316	return (0);
4317}
4318
4319/*
4320 * Destroy a storage pool.
4321 */
4322int
4323spa_destroy(char *pool)
4324{
4325	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4326	    B_FALSE, B_FALSE));
4327}
4328
4329/*
4330 * Export a storage pool.
4331 */
4332int
4333spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4334    boolean_t hardforce)
4335{
4336	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4337	    force, hardforce));
4338}
4339
4340/*
4341 * Similar to spa_export(), this unloads the spa_t without actually removing it
4342 * from the namespace in any way.
4343 */
4344int
4345spa_reset(char *pool)
4346{
4347	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4348	    B_FALSE, B_FALSE));
4349}
4350
4351/*
4352 * ==========================================================================
4353 * Device manipulation
4354 * ==========================================================================
4355 */
4356
4357/*
4358 * Add a device to a storage pool.
4359 */
4360int
4361spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4362{
4363	uint64_t txg, id;
4364	int error;
4365	vdev_t *rvd = spa->spa_root_vdev;
4366	vdev_t *vd, *tvd;
4367	nvlist_t **spares, **l2cache;
4368	uint_t nspares, nl2cache;
4369
4370	ASSERT(spa_writeable(spa));
4371
4372	txg = spa_vdev_enter(spa);
4373
4374	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4375	    VDEV_ALLOC_ADD)) != 0)
4376		return (spa_vdev_exit(spa, NULL, txg, error));
4377
4378	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
4379
4380	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4381	    &nspares) != 0)
4382		nspares = 0;
4383
4384	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4385	    &nl2cache) != 0)
4386		nl2cache = 0;
4387
4388	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4389		return (spa_vdev_exit(spa, vd, txg, EINVAL));
4390
4391	if (vd->vdev_children != 0 &&
4392	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
4393		return (spa_vdev_exit(spa, vd, txg, error));
4394
4395	/*
4396	 * We must validate the spares and l2cache devices after checking the
4397	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
4398	 */
4399	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4400		return (spa_vdev_exit(spa, vd, txg, error));
4401
4402	/*
4403	 * Transfer each new top-level vdev from vd to rvd.
4404	 */
4405	for (int c = 0; c < vd->vdev_children; c++) {
4406
4407		/*
4408		 * Set the vdev id to the first hole, if one exists.
4409		 */
4410		for (id = 0; id < rvd->vdev_children; id++) {
4411			if (rvd->vdev_child[id]->vdev_ishole) {
4412				vdev_free(rvd->vdev_child[id]);
4413				break;
4414			}
4415		}
4416		tvd = vd->vdev_child[c];
4417		vdev_remove_child(vd, tvd);
4418		tvd->vdev_id = id;
4419		vdev_add_child(rvd, tvd);
4420		vdev_config_dirty(tvd);
4421	}
4422
4423	if (nspares != 0) {
4424		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4425		    ZPOOL_CONFIG_SPARES);
4426		spa_load_spares(spa);
4427		spa->spa_spares.sav_sync = B_TRUE;
4428	}
4429
4430	if (nl2cache != 0) {
4431		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4432		    ZPOOL_CONFIG_L2CACHE);
4433		spa_load_l2cache(spa);
4434		spa->spa_l2cache.sav_sync = B_TRUE;
4435	}
4436
4437	/*
4438	 * We have to be careful when adding new vdevs to an existing pool.
4439	 * If other threads start allocating from these vdevs before we
4440	 * sync the config cache, and we lose power, then upon reboot we may
4441	 * fail to open the pool because there are DVAs that the config cache
4442	 * can't translate.  Therefore, we first add the vdevs without
4443	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4444	 * and then let spa_config_update() initialize the new metaslabs.
4445	 *
4446	 * spa_load() checks for added-but-not-initialized vdevs, so that
4447	 * if we lose power at any point in this sequence, the remaining
4448	 * steps will be completed the next time we load the pool.
4449	 */
4450	(void) spa_vdev_exit(spa, vd, txg, 0);
4451
4452	mutex_enter(&spa_namespace_lock);
4453	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4454	spa_event_notify(spa, NULL, ESC_ZFS_VDEV_ADD);
4455	mutex_exit(&spa_namespace_lock);
4456
4457	return (0);
4458}
4459
4460/*
4461 * Attach a device to a mirror.  The arguments are the path to any device
4462 * in the mirror, and the nvroot for the new device.  If the path specifies
4463 * a device that is not mirrored, we automatically insert the mirror vdev.
4464 *
4465 * If 'replacing' is specified, the new device is intended to replace the
4466 * existing device; in this case the two devices are made into their own
4467 * mirror using the 'replacing' vdev, which is functionally identical to
4468 * the mirror vdev (it actually reuses all the same ops) but has a few
4469 * extra rules: you can't attach to it after it's been created, and upon
4470 * completion of resilvering, the first disk (the one being replaced)
4471 * is automatically detached.
4472 */
4473int
4474spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4475{
4476	uint64_t txg, dtl_max_txg;
4477	vdev_t *rvd = spa->spa_root_vdev;
4478	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4479	vdev_ops_t *pvops;
4480	char *oldvdpath, *newvdpath;
4481	int newvd_isspare;
4482	int error;
4483
4484	ASSERT(spa_writeable(spa));
4485
4486	txg = spa_vdev_enter(spa);
4487
4488	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4489
4490	if (oldvd == NULL)
4491		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4492
4493	if (!oldvd->vdev_ops->vdev_op_leaf)
4494		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4495
4496	pvd = oldvd->vdev_parent;
4497
4498	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4499	    VDEV_ALLOC_ATTACH)) != 0)
4500		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4501
4502	if (newrootvd->vdev_children != 1)
4503		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4504
4505	newvd = newrootvd->vdev_child[0];
4506
4507	if (!newvd->vdev_ops->vdev_op_leaf)
4508		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4509
4510	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4511		return (spa_vdev_exit(spa, newrootvd, txg, error));
4512
4513	/*
4514	 * Spares can't replace logs
4515	 */
4516	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4517		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4518
4519	if (!replacing) {
4520		/*
4521		 * For attach, the only allowable parent is a mirror or the root
4522		 * vdev.
4523		 */
4524		if (pvd->vdev_ops != &vdev_mirror_ops &&
4525		    pvd->vdev_ops != &vdev_root_ops)
4526			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4527
4528		pvops = &vdev_mirror_ops;
4529	} else {
4530		/*
4531		 * Active hot spares can only be replaced by inactive hot
4532		 * spares.
4533		 */
4534		if (pvd->vdev_ops == &vdev_spare_ops &&
4535		    oldvd->vdev_isspare &&
4536		    !spa_has_spare(spa, newvd->vdev_guid))
4537			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4538
4539		/*
4540		 * If the source is a hot spare, and the parent isn't already a
4541		 * spare, then we want to create a new hot spare.  Otherwise, we
4542		 * want to create a replacing vdev.  The user is not allowed to
4543		 * attach to a spared vdev child unless the 'isspare' state is
4544		 * the same (spare replaces spare, non-spare replaces
4545		 * non-spare).
4546		 */
4547		if (pvd->vdev_ops == &vdev_replacing_ops &&
4548		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4549			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4550		} else if (pvd->vdev_ops == &vdev_spare_ops &&
4551		    newvd->vdev_isspare != oldvd->vdev_isspare) {
4552			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4553		}
4554
4555		if (newvd->vdev_isspare)
4556			pvops = &vdev_spare_ops;
4557		else
4558			pvops = &vdev_replacing_ops;
4559	}
4560
4561	/*
4562	 * Make sure the new device is big enough.
4563	 */
4564	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4565		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4566
4567	/*
4568	 * The new device cannot have a higher alignment requirement
4569	 * than the top-level vdev.
4570	 */
4571	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4572		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4573
4574	/*
4575	 * If this is an in-place replacement, update oldvd's path and devid
4576	 * to make it distinguishable from newvd, and unopenable from now on.
4577	 */
4578	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4579		spa_strfree(oldvd->vdev_path);
4580		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4581		    KM_SLEEP);
4582		(void) sprintf(oldvd->vdev_path, "%s/%s",
4583		    newvd->vdev_path, "old");
4584		if (oldvd->vdev_devid != NULL) {
4585			spa_strfree(oldvd->vdev_devid);
4586			oldvd->vdev_devid = NULL;
4587		}
4588	}
4589
4590	/* mark the device being resilvered */
4591	newvd->vdev_resilver_txg = txg;
4592
4593	/*
4594	 * If the parent is not a mirror, or if we're replacing, insert the new
4595	 * mirror/replacing/spare vdev above oldvd.
4596	 */
4597	if (pvd->vdev_ops != pvops)
4598		pvd = vdev_add_parent(oldvd, pvops);
4599
4600	ASSERT(pvd->vdev_top->vdev_parent == rvd);
4601	ASSERT(pvd->vdev_ops == pvops);
4602	ASSERT(oldvd->vdev_parent == pvd);
4603
4604	/*
4605	 * Extract the new device from its root and add it to pvd.
4606	 */
4607	vdev_remove_child(newrootvd, newvd);
4608	newvd->vdev_id = pvd->vdev_children;
4609	newvd->vdev_crtxg = oldvd->vdev_crtxg;
4610	vdev_add_child(pvd, newvd);
4611
4612	tvd = newvd->vdev_top;
4613	ASSERT(pvd->vdev_top == tvd);
4614	ASSERT(tvd->vdev_parent == rvd);
4615
4616	vdev_config_dirty(tvd);
4617
4618	/*
4619	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4620	 * for any dmu_sync-ed blocks.  It will propagate upward when
4621	 * spa_vdev_exit() calls vdev_dtl_reassess().
4622	 */
4623	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4624
4625	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4626	    dtl_max_txg - TXG_INITIAL);
4627
4628	if (newvd->vdev_isspare) {
4629		spa_spare_activate(newvd);
4630		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4631	}
4632
4633	oldvdpath = spa_strdup(oldvd->vdev_path);
4634	newvdpath = spa_strdup(newvd->vdev_path);
4635	newvd_isspare = newvd->vdev_isspare;
4636
4637	/*
4638	 * Mark newvd's DTL dirty in this txg.
4639	 */
4640	vdev_dirty(tvd, VDD_DTL, newvd, txg);
4641
4642	/*
4643	 * Schedule the resilver to restart in the future. We do this to
4644	 * ensure that dmu_sync-ed blocks have been stitched into the
4645	 * respective datasets.
4646	 */
4647	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4648
4649	if (spa->spa_bootfs)
4650		spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4651
4652	spa_event_notify(spa, newvd, ESC_ZFS_VDEV_ATTACH);
4653
4654	/*
4655	 * Commit the config
4656	 */
4657	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4658
4659	spa_history_log_internal(spa, "vdev attach", NULL,
4660	    "%s vdev=%s %s vdev=%s",
4661	    replacing && newvd_isspare ? "spare in" :
4662	    replacing ? "replace" : "attach", newvdpath,
4663	    replacing ? "for" : "to", oldvdpath);
4664
4665	spa_strfree(oldvdpath);
4666	spa_strfree(newvdpath);
4667
4668	return (0);
4669}
4670
4671/*
4672 * Detach a device from a mirror or replacing vdev.
4673 *
4674 * If 'replace_done' is specified, only detach if the parent
4675 * is a replacing vdev.
4676 */
4677int
4678spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4679{
4680	uint64_t txg;
4681	int error;
4682	vdev_t *rvd = spa->spa_root_vdev;
4683	vdev_t *vd, *pvd, *cvd, *tvd;
4684	boolean_t unspare = B_FALSE;
4685	uint64_t unspare_guid = 0;
4686	char *vdpath;
4687
4688	ASSERT(spa_writeable(spa));
4689
4690	txg = spa_vdev_enter(spa);
4691
4692	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4693
4694	if (vd == NULL)
4695		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4696
4697	if (!vd->vdev_ops->vdev_op_leaf)
4698		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4699
4700	pvd = vd->vdev_parent;
4701
4702	/*
4703	 * If the parent/child relationship is not as expected, don't do it.
4704	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4705	 * vdev that's replacing B with C.  The user's intent in replacing
4706	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4707	 * the replace by detaching C, the expected behavior is to end up
4708	 * M(A,B).  But suppose that right after deciding to detach C,
4709	 * the replacement of B completes.  We would have M(A,C), and then
4710	 * ask to detach C, which would leave us with just A -- not what
4711	 * the user wanted.  To prevent this, we make sure that the
4712	 * parent/child relationship hasn't changed -- in this example,
4713	 * that C's parent is still the replacing vdev R.
4714	 */
4715	if (pvd->vdev_guid != pguid && pguid != 0)
4716		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4717
4718	/*
4719	 * Only 'replacing' or 'spare' vdevs can be replaced.
4720	 */
4721	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4722	    pvd->vdev_ops != &vdev_spare_ops)
4723		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4724
4725	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4726	    spa_version(spa) >= SPA_VERSION_SPARES);
4727
4728	/*
4729	 * Only mirror, replacing, and spare vdevs support detach.
4730	 */
4731	if (pvd->vdev_ops != &vdev_replacing_ops &&
4732	    pvd->vdev_ops != &vdev_mirror_ops &&
4733	    pvd->vdev_ops != &vdev_spare_ops)
4734		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4735
4736	/*
4737	 * If this device has the only valid copy of some data,
4738	 * we cannot safely detach it.
4739	 */
4740	if (vdev_dtl_required(vd))
4741		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4742
4743	ASSERT(pvd->vdev_children >= 2);
4744
4745	/*
4746	 * If we are detaching the second disk from a replacing vdev, then
4747	 * check to see if we changed the original vdev's path to have "/old"
4748	 * at the end in spa_vdev_attach().  If so, undo that change now.
4749	 */
4750	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4751	    vd->vdev_path != NULL) {
4752		size_t len = strlen(vd->vdev_path);
4753
4754		for (int c = 0; c < pvd->vdev_children; c++) {
4755			cvd = pvd->vdev_child[c];
4756
4757			if (cvd == vd || cvd->vdev_path == NULL)
4758				continue;
4759
4760			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4761			    strcmp(cvd->vdev_path + len, "/old") == 0) {
4762				spa_strfree(cvd->vdev_path);
4763				cvd->vdev_path = spa_strdup(vd->vdev_path);
4764				break;
4765			}
4766		}
4767	}
4768
4769	/*
4770	 * If we are detaching the original disk from a spare, then it implies
4771	 * that the spare should become a real disk, and be removed from the
4772	 * active spare list for the pool.
4773	 */
4774	if (pvd->vdev_ops == &vdev_spare_ops &&
4775	    vd->vdev_id == 0 &&
4776	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4777		unspare = B_TRUE;
4778
4779	/*
4780	 * Erase the disk labels so the disk can be used for other things.
4781	 * This must be done after all other error cases are handled,
4782	 * but before we disembowel vd (so we can still do I/O to it).
4783	 * But if we can't do it, don't treat the error as fatal --
4784	 * it may be that the unwritability of the disk is the reason
4785	 * it's being detached!
4786	 */
4787	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4788
4789	/*
4790	 * Remove vd from its parent and compact the parent's children.
4791	 */
4792	vdev_remove_child(pvd, vd);
4793	vdev_compact_children(pvd);
4794
4795	/*
4796	 * Remember one of the remaining children so we can get tvd below.
4797	 */
4798	cvd = pvd->vdev_child[pvd->vdev_children - 1];
4799
4800	/*
4801	 * If we need to remove the remaining child from the list of hot spares,
4802	 * do it now, marking the vdev as no longer a spare in the process.
4803	 * We must do this before vdev_remove_parent(), because that can
4804	 * change the GUID if it creates a new toplevel GUID.  For a similar
4805	 * reason, we must remove the spare now, in the same txg as the detach;
4806	 * otherwise someone could attach a new sibling, change the GUID, and
4807	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4808	 */
4809	if (unspare) {
4810		ASSERT(cvd->vdev_isspare);
4811		spa_spare_remove(cvd);
4812		unspare_guid = cvd->vdev_guid;
4813		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4814		cvd->vdev_unspare = B_TRUE;
4815	}
4816
4817	/*
4818	 * If the parent mirror/replacing vdev only has one child,
4819	 * the parent is no longer needed.  Remove it from the tree.
4820	 */
4821	if (pvd->vdev_children == 1) {
4822		if (pvd->vdev_ops == &vdev_spare_ops)
4823			cvd->vdev_unspare = B_FALSE;
4824		vdev_remove_parent(cvd);
4825	}
4826
4827
4828	/*
4829	 * We don't set tvd until now because the parent we just removed
4830	 * may have been the previous top-level vdev.
4831	 */
4832	tvd = cvd->vdev_top;
4833	ASSERT(tvd->vdev_parent == rvd);
4834
4835	/*
4836	 * Reevaluate the parent vdev state.
4837	 */
4838	vdev_propagate_state(cvd);
4839
4840	/*
4841	 * If the 'autoexpand' property is set on the pool then automatically
4842	 * try to expand the size of the pool. For example if the device we
4843	 * just detached was smaller than the others, it may be possible to
4844	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4845	 * first so that we can obtain the updated sizes of the leaf vdevs.
4846	 */
4847	if (spa->spa_autoexpand) {
4848		vdev_reopen(tvd);
4849		vdev_expand(tvd, txg);
4850	}
4851
4852	vdev_config_dirty(tvd);
4853
4854	/*
4855	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4856	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4857	 * But first make sure we're not on any *other* txg's DTL list, to
4858	 * prevent vd from being accessed after it's freed.
4859	 */
4860	vdpath = spa_strdup(vd->vdev_path);
4861	for (int t = 0; t < TXG_SIZE; t++)
4862		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4863	vd->vdev_detached = B_TRUE;
4864	vdev_dirty(tvd, VDD_DTL, vd, txg);
4865
4866	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4867
4868	/* hang on to the spa before we release the lock */
4869	spa_open_ref(spa, FTAG);
4870
4871	error = spa_vdev_exit(spa, vd, txg, 0);
4872
4873	spa_history_log_internal(spa, "detach", NULL,
4874	    "vdev=%s", vdpath);
4875	spa_strfree(vdpath);
4876
4877	/*
4878	 * If this was the removal of the original device in a hot spare vdev,
4879	 * then we want to go through and remove the device from the hot spare
4880	 * list of every other pool.
4881	 */
4882	if (unspare) {
4883		spa_t *altspa = NULL;
4884
4885		mutex_enter(&spa_namespace_lock);
4886		while ((altspa = spa_next(altspa)) != NULL) {
4887			if (altspa->spa_state != POOL_STATE_ACTIVE ||
4888			    altspa == spa)
4889				continue;
4890
4891			spa_open_ref(altspa, FTAG);
4892			mutex_exit(&spa_namespace_lock);
4893			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4894			mutex_enter(&spa_namespace_lock);
4895			spa_close(altspa, FTAG);
4896		}
4897		mutex_exit(&spa_namespace_lock);
4898
4899		/* search the rest of the vdevs for spares to remove */
4900		spa_vdev_resilver_done(spa);
4901	}
4902
4903	/* all done with the spa; OK to release */
4904	mutex_enter(&spa_namespace_lock);
4905	spa_close(spa, FTAG);
4906	mutex_exit(&spa_namespace_lock);
4907
4908	return (error);
4909}
4910
4911/*
4912 * Split a set of devices from their mirrors, and create a new pool from them.
4913 */
4914int
4915spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4916    nvlist_t *props, boolean_t exp)
4917{
4918	int error = 0;
4919	uint64_t txg, *glist;
4920	spa_t *newspa;
4921	uint_t c, children, lastlog;
4922	nvlist_t **child, *nvl, *tmp;
4923	dmu_tx_t *tx;
4924	char *altroot = NULL;
4925	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4926	boolean_t activate_slog;
4927
4928	ASSERT(spa_writeable(spa));
4929
4930	txg = spa_vdev_enter(spa);
4931
4932	/* clear the log and flush everything up to now */
4933	activate_slog = spa_passivate_log(spa);
4934	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4935	error = spa_offline_log(spa);
4936	txg = spa_vdev_config_enter(spa);
4937
4938	if (activate_slog)
4939		spa_activate_log(spa);
4940
4941	if (error != 0)
4942		return (spa_vdev_exit(spa, NULL, txg, error));
4943
4944	/* check new spa name before going any further */
4945	if (spa_lookup(newname) != NULL)
4946		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4947
4948	/*
4949	 * scan through all the children to ensure they're all mirrors
4950	 */
4951	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4952	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4953	    &children) != 0)
4954		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4955
4956	/* first, check to ensure we've got the right child count */
4957	rvd = spa->spa_root_vdev;
4958	lastlog = 0;
4959	for (c = 0; c < rvd->vdev_children; c++) {
4960		vdev_t *vd = rvd->vdev_child[c];
4961
4962		/* don't count the holes & logs as children */
4963		if (vd->vdev_islog || vd->vdev_ishole) {
4964			if (lastlog == 0)
4965				lastlog = c;
4966			continue;
4967		}
4968
4969		lastlog = 0;
4970	}
4971	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4972		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4973
4974	/* next, ensure no spare or cache devices are part of the split */
4975	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4976	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4977		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4978
4979	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4980	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4981
4982	/* then, loop over each vdev and validate it */
4983	for (c = 0; c < children; c++) {
4984		uint64_t is_hole = 0;
4985
4986		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4987		    &is_hole);
4988
4989		if (is_hole != 0) {
4990			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4991			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4992				continue;
4993			} else {
4994				error = SET_ERROR(EINVAL);
4995				break;
4996			}
4997		}
4998
4999		/* which disk is going to be split? */
5000		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5001		    &glist[c]) != 0) {
5002			error = SET_ERROR(EINVAL);
5003			break;
5004		}
5005
5006		/* look it up in the spa */
5007		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5008		if (vml[c] == NULL) {
5009			error = SET_ERROR(ENODEV);
5010			break;
5011		}
5012
5013		/* make sure there's nothing stopping the split */
5014		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5015		    vml[c]->vdev_islog ||
5016		    vml[c]->vdev_ishole ||
5017		    vml[c]->vdev_isspare ||
5018		    vml[c]->vdev_isl2cache ||
5019		    !vdev_writeable(vml[c]) ||
5020		    vml[c]->vdev_children != 0 ||
5021		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5022		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5023			error = SET_ERROR(EINVAL);
5024			break;
5025		}
5026
5027		if (vdev_dtl_required(vml[c])) {
5028			error = SET_ERROR(EBUSY);
5029			break;
5030		}
5031
5032		/* we need certain info from the top level */
5033		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5034		    vml[c]->vdev_top->vdev_ms_array) == 0);
5035		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5036		    vml[c]->vdev_top->vdev_ms_shift) == 0);
5037		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5038		    vml[c]->vdev_top->vdev_asize) == 0);
5039		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5040		    vml[c]->vdev_top->vdev_ashift) == 0);
5041	}
5042
5043	if (error != 0) {
5044		kmem_free(vml, children * sizeof (vdev_t *));
5045		kmem_free(glist, children * sizeof (uint64_t));
5046		return (spa_vdev_exit(spa, NULL, txg, error));
5047	}
5048
5049	/* stop writers from using the disks */
5050	for (c = 0; c < children; c++) {
5051		if (vml[c] != NULL)
5052			vml[c]->vdev_offline = B_TRUE;
5053	}
5054	vdev_reopen(spa->spa_root_vdev);
5055
5056	/*
5057	 * Temporarily record the splitting vdevs in the spa config.  This
5058	 * will disappear once the config is regenerated.
5059	 */
5060	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5061	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5062	    glist, children) == 0);
5063	kmem_free(glist, children * sizeof (uint64_t));
5064
5065	mutex_enter(&spa->spa_props_lock);
5066	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5067	    nvl) == 0);
5068	mutex_exit(&spa->spa_props_lock);
5069	spa->spa_config_splitting = nvl;
5070	vdev_config_dirty(spa->spa_root_vdev);
5071
5072	/* configure and create the new pool */
5073	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5074	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5075	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5076	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5077	    spa_version(spa)) == 0);
5078	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5079	    spa->spa_config_txg) == 0);
5080	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5081	    spa_generate_guid(NULL)) == 0);
5082	(void) nvlist_lookup_string(props,
5083	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5084
5085	/* add the new pool to the namespace */
5086	newspa = spa_add(newname, config, altroot);
5087	newspa->spa_config_txg = spa->spa_config_txg;
5088	spa_set_log_state(newspa, SPA_LOG_CLEAR);
5089
5090	/* release the spa config lock, retaining the namespace lock */
5091	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5092
5093	if (zio_injection_enabled)
5094		zio_handle_panic_injection(spa, FTAG, 1);
5095
5096	spa_activate(newspa, spa_mode_global);
5097	spa_async_suspend(newspa);
5098
5099	/* create the new pool from the disks of the original pool */
5100	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5101	if (error)
5102		goto out;
5103
5104	/* if that worked, generate a real config for the new pool */
5105	if (newspa->spa_root_vdev != NULL) {
5106		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5107		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
5108		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5109		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5110		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5111		    B_TRUE));
5112	}
5113
5114	/* set the props */
5115	if (props != NULL) {
5116		spa_configfile_set(newspa, props, B_FALSE);
5117		error = spa_prop_set(newspa, props);
5118		if (error)
5119			goto out;
5120	}
5121
5122	/* flush everything */
5123	txg = spa_vdev_config_enter(newspa);
5124	vdev_config_dirty(newspa->spa_root_vdev);
5125	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5126
5127	if (zio_injection_enabled)
5128		zio_handle_panic_injection(spa, FTAG, 2);
5129
5130	spa_async_resume(newspa);
5131
5132	/* finally, update the original pool's config */
5133	txg = spa_vdev_config_enter(spa);
5134	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5135	error = dmu_tx_assign(tx, TXG_WAIT);
5136	if (error != 0)
5137		dmu_tx_abort(tx);
5138	for (c = 0; c < children; c++) {
5139		if (vml[c] != NULL) {
5140			vdev_split(vml[c]);
5141			if (error == 0)
5142				spa_history_log_internal(spa, "detach", tx,
5143				    "vdev=%s", vml[c]->vdev_path);
5144			vdev_free(vml[c]);
5145		}
5146	}
5147	vdev_config_dirty(spa->spa_root_vdev);
5148	spa->spa_config_splitting = NULL;
5149	nvlist_free(nvl);
5150	if (error == 0)
5151		dmu_tx_commit(tx);
5152	(void) spa_vdev_exit(spa, NULL, txg, 0);
5153
5154	if (zio_injection_enabled)
5155		zio_handle_panic_injection(spa, FTAG, 3);
5156
5157	/* split is complete; log a history record */
5158	spa_history_log_internal(newspa, "split", NULL,
5159	    "from pool %s", spa_name(spa));
5160
5161	kmem_free(vml, children * sizeof (vdev_t *));
5162
5163	/* if we're not going to mount the filesystems in userland, export */
5164	if (exp)
5165		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5166		    B_FALSE, B_FALSE);
5167
5168	return (error);
5169
5170out:
5171	spa_unload(newspa);
5172	spa_deactivate(newspa);
5173	spa_remove(newspa);
5174
5175	txg = spa_vdev_config_enter(spa);
5176
5177	/* re-online all offlined disks */
5178	for (c = 0; c < children; c++) {
5179		if (vml[c] != NULL)
5180			vml[c]->vdev_offline = B_FALSE;
5181	}
5182	vdev_reopen(spa->spa_root_vdev);
5183
5184	nvlist_free(spa->spa_config_splitting);
5185	spa->spa_config_splitting = NULL;
5186	(void) spa_vdev_exit(spa, NULL, txg, error);
5187
5188	kmem_free(vml, children * sizeof (vdev_t *));
5189	return (error);
5190}
5191
5192static nvlist_t *
5193spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5194{
5195	for (int i = 0; i < count; i++) {
5196		uint64_t guid;
5197
5198		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5199		    &guid) == 0);
5200
5201		if (guid == target_guid)
5202			return (nvpp[i]);
5203	}
5204
5205	return (NULL);
5206}
5207
5208static void
5209spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5210	nvlist_t *dev_to_remove)
5211{
5212	nvlist_t **newdev = NULL;
5213
5214	if (count > 1)
5215		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5216
5217	for (int i = 0, j = 0; i < count; i++) {
5218		if (dev[i] == dev_to_remove)
5219			continue;
5220		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5221	}
5222
5223	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5224	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5225
5226	for (int i = 0; i < count - 1; i++)
5227		nvlist_free(newdev[i]);
5228
5229	if (count > 1)
5230		kmem_free(newdev, (count - 1) * sizeof (void *));
5231}
5232
5233/*
5234 * Evacuate the device.
5235 */
5236static int
5237spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5238{
5239	uint64_t txg;
5240	int error = 0;
5241
5242	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5243	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5244	ASSERT(vd == vd->vdev_top);
5245
5246	/*
5247	 * Evacuate the device.  We don't hold the config lock as writer
5248	 * since we need to do I/O but we do keep the
5249	 * spa_namespace_lock held.  Once this completes the device
5250	 * should no longer have any blocks allocated on it.
5251	 */
5252	if (vd->vdev_islog) {
5253		if (vd->vdev_stat.vs_alloc != 0)
5254			error = spa_offline_log(spa);
5255	} else {
5256		error = SET_ERROR(ENOTSUP);
5257	}
5258
5259	if (error)
5260		return (error);
5261
5262	/*
5263	 * The evacuation succeeded.  Remove any remaining MOS metadata
5264	 * associated with this vdev, and wait for these changes to sync.
5265	 */
5266	ASSERT0(vd->vdev_stat.vs_alloc);
5267	txg = spa_vdev_config_enter(spa);
5268	vd->vdev_removing = B_TRUE;
5269	vdev_dirty_leaves(vd, VDD_DTL, txg);
5270	vdev_config_dirty(vd);
5271	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5272
5273	return (0);
5274}
5275
5276/*
5277 * Complete the removal by cleaning up the namespace.
5278 */
5279static void
5280spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5281{
5282	vdev_t *rvd = spa->spa_root_vdev;
5283	uint64_t id = vd->vdev_id;
5284	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5285
5286	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5287	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5288	ASSERT(vd == vd->vdev_top);
5289
5290	/*
5291	 * Only remove any devices which are empty.
5292	 */
5293	if (vd->vdev_stat.vs_alloc != 0)
5294		return;
5295
5296	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5297
5298	if (list_link_active(&vd->vdev_state_dirty_node))
5299		vdev_state_clean(vd);
5300	if (list_link_active(&vd->vdev_config_dirty_node))
5301		vdev_config_clean(vd);
5302
5303	vdev_free(vd);
5304
5305	if (last_vdev) {
5306		vdev_compact_children(rvd);
5307	} else {
5308		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5309		vdev_add_child(rvd, vd);
5310	}
5311	vdev_config_dirty(rvd);
5312
5313	/*
5314	 * Reassess the health of our root vdev.
5315	 */
5316	vdev_reopen(rvd);
5317}
5318
5319/*
5320 * Remove a device from the pool -
5321 *
5322 * Removing a device from the vdev namespace requires several steps
5323 * and can take a significant amount of time.  As a result we use
5324 * the spa_vdev_config_[enter/exit] functions which allow us to
5325 * grab and release the spa_config_lock while still holding the namespace
5326 * lock.  During each step the configuration is synced out.
5327 *
5328 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5329 * devices.
5330 */
5331int
5332spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5333{
5334	vdev_t *vd;
5335	metaslab_group_t *mg;
5336	nvlist_t **spares, **l2cache, *nv;
5337	uint64_t txg = 0;
5338	uint_t nspares, nl2cache;
5339	int error = 0;
5340	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5341
5342	ASSERT(spa_writeable(spa));
5343
5344	if (!locked)
5345		txg = spa_vdev_enter(spa);
5346
5347	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5348
5349	if (spa->spa_spares.sav_vdevs != NULL &&
5350	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5351	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5352	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5353		/*
5354		 * Only remove the hot spare if it's not currently in use
5355		 * in this pool.
5356		 */
5357		if (vd == NULL || unspare) {
5358			spa_vdev_remove_aux(spa->spa_spares.sav_config,
5359			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5360			spa_load_spares(spa);
5361			spa->spa_spares.sav_sync = B_TRUE;
5362		} else {
5363			error = SET_ERROR(EBUSY);
5364		}
5365	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
5366	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5367	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5368	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5369		/*
5370		 * Cache devices can always be removed.
5371		 */
5372		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5373		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5374		spa_load_l2cache(spa);
5375		spa->spa_l2cache.sav_sync = B_TRUE;
5376	} else if (vd != NULL && vd->vdev_islog) {
5377		ASSERT(!locked);
5378		ASSERT(vd == vd->vdev_top);
5379
5380		mg = vd->vdev_mg;
5381
5382		/*
5383		 * Stop allocating from this vdev.
5384		 */
5385		metaslab_group_passivate(mg);
5386
5387		/*
5388		 * Wait for the youngest allocations and frees to sync,
5389		 * and then wait for the deferral of those frees to finish.
5390		 */
5391		spa_vdev_config_exit(spa, NULL,
5392		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5393
5394		/*
5395		 * Attempt to evacuate the vdev.
5396		 */
5397		error = spa_vdev_remove_evacuate(spa, vd);
5398
5399		txg = spa_vdev_config_enter(spa);
5400
5401		/*
5402		 * If we couldn't evacuate the vdev, unwind.
5403		 */
5404		if (error) {
5405			metaslab_group_activate(mg);
5406			return (spa_vdev_exit(spa, NULL, txg, error));
5407		}
5408
5409		/*
5410		 * Clean up the vdev namespace.
5411		 */
5412		spa_vdev_remove_from_namespace(spa, vd);
5413
5414	} else if (vd != NULL) {
5415		/*
5416		 * Normal vdevs cannot be removed (yet).
5417		 */
5418		error = SET_ERROR(ENOTSUP);
5419	} else {
5420		/*
5421		 * There is no vdev of any kind with the specified guid.
5422		 */
5423		error = SET_ERROR(ENOENT);
5424	}
5425
5426	if (!locked)
5427		return (spa_vdev_exit(spa, NULL, txg, error));
5428
5429	return (error);
5430}
5431
5432/*
5433 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5434 * currently spared, so we can detach it.
5435 */
5436static vdev_t *
5437spa_vdev_resilver_done_hunt(vdev_t *vd)
5438{
5439	vdev_t *newvd, *oldvd;
5440
5441	for (int c = 0; c < vd->vdev_children; c++) {
5442		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5443		if (oldvd != NULL)
5444			return (oldvd);
5445	}
5446
5447	/*
5448	 * Check for a completed replacement.  We always consider the first
5449	 * vdev in the list to be the oldest vdev, and the last one to be
5450	 * the newest (see spa_vdev_attach() for how that works).  In
5451	 * the case where the newest vdev is faulted, we will not automatically
5452	 * remove it after a resilver completes.  This is OK as it will require
5453	 * user intervention to determine which disk the admin wishes to keep.
5454	 */
5455	if (vd->vdev_ops == &vdev_replacing_ops) {
5456		ASSERT(vd->vdev_children > 1);
5457
5458		newvd = vd->vdev_child[vd->vdev_children - 1];
5459		oldvd = vd->vdev_child[0];
5460
5461		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5462		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5463		    !vdev_dtl_required(oldvd))
5464			return (oldvd);
5465	}
5466
5467	/*
5468	 * Check for a completed resilver with the 'unspare' flag set.
5469	 */
5470	if (vd->vdev_ops == &vdev_spare_ops) {
5471		vdev_t *first = vd->vdev_child[0];
5472		vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5473
5474		if (last->vdev_unspare) {
5475			oldvd = first;
5476			newvd = last;
5477		} else if (first->vdev_unspare) {
5478			oldvd = last;
5479			newvd = first;
5480		} else {
5481			oldvd = NULL;
5482		}
5483
5484		if (oldvd != NULL &&
5485		    vdev_dtl_empty(newvd, DTL_MISSING) &&
5486		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5487		    !vdev_dtl_required(oldvd))
5488			return (oldvd);
5489
5490		/*
5491		 * If there are more than two spares attached to a disk,
5492		 * and those spares are not required, then we want to
5493		 * attempt to free them up now so that they can be used
5494		 * by other pools.  Once we're back down to a single
5495		 * disk+spare, we stop removing them.
5496		 */
5497		if (vd->vdev_children > 2) {
5498			newvd = vd->vdev_child[1];
5499
5500			if (newvd->vdev_isspare && last->vdev_isspare &&
5501			    vdev_dtl_empty(last, DTL_MISSING) &&
5502			    vdev_dtl_empty(last, DTL_OUTAGE) &&
5503			    !vdev_dtl_required(newvd))
5504				return (newvd);
5505		}
5506	}
5507
5508	return (NULL);
5509}
5510
5511static void
5512spa_vdev_resilver_done(spa_t *spa)
5513{
5514	vdev_t *vd, *pvd, *ppvd;
5515	uint64_t guid, sguid, pguid, ppguid;
5516
5517	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5518
5519	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5520		pvd = vd->vdev_parent;
5521		ppvd = pvd->vdev_parent;
5522		guid = vd->vdev_guid;
5523		pguid = pvd->vdev_guid;
5524		ppguid = ppvd->vdev_guid;
5525		sguid = 0;
5526		/*
5527		 * If we have just finished replacing a hot spared device, then
5528		 * we need to detach the parent's first child (the original hot
5529		 * spare) as well.
5530		 */
5531		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5532		    ppvd->vdev_children == 2) {
5533			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5534			sguid = ppvd->vdev_child[1]->vdev_guid;
5535		}
5536		ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5537
5538		spa_config_exit(spa, SCL_ALL, FTAG);
5539		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5540			return;
5541		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5542			return;
5543		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5544	}
5545
5546	spa_config_exit(spa, SCL_ALL, FTAG);
5547}
5548
5549/*
5550 * Update the stored path or FRU for this vdev.
5551 */
5552int
5553spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5554    boolean_t ispath)
5555{
5556	vdev_t *vd;
5557	boolean_t sync = B_FALSE;
5558
5559	ASSERT(spa_writeable(spa));
5560
5561	spa_vdev_state_enter(spa, SCL_ALL);
5562
5563	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5564		return (spa_vdev_state_exit(spa, NULL, ENOENT));
5565
5566	if (!vd->vdev_ops->vdev_op_leaf)
5567		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5568
5569	if (ispath) {
5570		if (strcmp(value, vd->vdev_path) != 0) {
5571			spa_strfree(vd->vdev_path);
5572			vd->vdev_path = spa_strdup(value);
5573			sync = B_TRUE;
5574		}
5575	} else {
5576		if (vd->vdev_fru == NULL) {
5577			vd->vdev_fru = spa_strdup(value);
5578			sync = B_TRUE;
5579		} else if (strcmp(value, vd->vdev_fru) != 0) {
5580			spa_strfree(vd->vdev_fru);
5581			vd->vdev_fru = spa_strdup(value);
5582			sync = B_TRUE;
5583		}
5584	}
5585
5586	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5587}
5588
5589int
5590spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5591{
5592	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5593}
5594
5595int
5596spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5597{
5598	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5599}
5600
5601/*
5602 * ==========================================================================
5603 * SPA Scanning
5604 * ==========================================================================
5605 */
5606
5607int
5608spa_scan_stop(spa_t *spa)
5609{
5610	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5611	if (dsl_scan_resilvering(spa->spa_dsl_pool))
5612		return (SET_ERROR(EBUSY));
5613	return (dsl_scan_cancel(spa->spa_dsl_pool));
5614}
5615
5616int
5617spa_scan(spa_t *spa, pool_scan_func_t func)
5618{
5619	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5620
5621	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5622		return (SET_ERROR(ENOTSUP));
5623
5624	/*
5625	 * If a resilver was requested, but there is no DTL on a
5626	 * writeable leaf device, we have nothing to do.
5627	 */
5628	if (func == POOL_SCAN_RESILVER &&
5629	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5630		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5631		return (0);
5632	}
5633
5634	return (dsl_scan(spa->spa_dsl_pool, func));
5635}
5636
5637/*
5638 * ==========================================================================
5639 * SPA async task processing
5640 * ==========================================================================
5641 */
5642
5643static void
5644spa_async_remove(spa_t *spa, vdev_t *vd)
5645{
5646	if (vd->vdev_remove_wanted) {
5647		vd->vdev_remove_wanted = B_FALSE;
5648		vd->vdev_delayed_close = B_FALSE;
5649		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5650
5651		/*
5652		 * We want to clear the stats, but we don't want to do a full
5653		 * vdev_clear() as that will cause us to throw away
5654		 * degraded/faulted state as well as attempt to reopen the
5655		 * device, all of which is a waste.
5656		 */
5657		vd->vdev_stat.vs_read_errors = 0;
5658		vd->vdev_stat.vs_write_errors = 0;
5659		vd->vdev_stat.vs_checksum_errors = 0;
5660
5661		vdev_state_dirty(vd->vdev_top);
5662	}
5663
5664	for (int c = 0; c < vd->vdev_children; c++)
5665		spa_async_remove(spa, vd->vdev_child[c]);
5666}
5667
5668static void
5669spa_async_probe(spa_t *spa, vdev_t *vd)
5670{
5671	if (vd->vdev_probe_wanted) {
5672		vd->vdev_probe_wanted = B_FALSE;
5673		vdev_reopen(vd);	/* vdev_open() does the actual probe */
5674	}
5675
5676	for (int c = 0; c < vd->vdev_children; c++)
5677		spa_async_probe(spa, vd->vdev_child[c]);
5678}
5679
5680static void
5681spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5682{
5683	sysevent_id_t eid;
5684	nvlist_t *attr;
5685	char *physpath;
5686
5687	if (!spa->spa_autoexpand)
5688		return;
5689
5690	for (int c = 0; c < vd->vdev_children; c++) {
5691		vdev_t *cvd = vd->vdev_child[c];
5692		spa_async_autoexpand(spa, cvd);
5693	}
5694
5695	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5696		return;
5697
5698	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5699	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5700
5701	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5702	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5703
5704	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5705	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5706
5707	nvlist_free(attr);
5708	kmem_free(physpath, MAXPATHLEN);
5709}
5710
5711static void
5712spa_async_thread(spa_t *spa)
5713{
5714	int tasks;
5715
5716	ASSERT(spa->spa_sync_on);
5717
5718	mutex_enter(&spa->spa_async_lock);
5719	tasks = spa->spa_async_tasks;
5720	spa->spa_async_tasks = 0;
5721	mutex_exit(&spa->spa_async_lock);
5722
5723	/*
5724	 * See if the config needs to be updated.
5725	 */
5726	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5727		uint64_t old_space, new_space;
5728
5729		mutex_enter(&spa_namespace_lock);
5730		old_space = metaslab_class_get_space(spa_normal_class(spa));
5731		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5732		new_space = metaslab_class_get_space(spa_normal_class(spa));
5733		mutex_exit(&spa_namespace_lock);
5734
5735		/*
5736		 * If the pool grew as a result of the config update,
5737		 * then log an internal history event.
5738		 */
5739		if (new_space != old_space) {
5740			spa_history_log_internal(spa, "vdev online", NULL,
5741			    "pool '%s' size: %llu(+%llu)",
5742			    spa_name(spa), new_space, new_space - old_space);
5743		}
5744	}
5745
5746	/*
5747	 * See if any devices need to be marked REMOVED.
5748	 */
5749	if (tasks & SPA_ASYNC_REMOVE) {
5750		spa_vdev_state_enter(spa, SCL_NONE);
5751		spa_async_remove(spa, spa->spa_root_vdev);
5752		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5753			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5754		for (int i = 0; i < spa->spa_spares.sav_count; i++)
5755			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5756		(void) spa_vdev_state_exit(spa, NULL, 0);
5757	}
5758
5759	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5760		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5761		spa_async_autoexpand(spa, spa->spa_root_vdev);
5762		spa_config_exit(spa, SCL_CONFIG, FTAG);
5763	}
5764
5765	/*
5766	 * See if any devices need to be probed.
5767	 */
5768	if (tasks & SPA_ASYNC_PROBE) {
5769		spa_vdev_state_enter(spa, SCL_NONE);
5770		spa_async_probe(spa, spa->spa_root_vdev);
5771		(void) spa_vdev_state_exit(spa, NULL, 0);
5772	}
5773
5774	/*
5775	 * If any devices are done replacing, detach them.
5776	 */
5777	if (tasks & SPA_ASYNC_RESILVER_DONE)
5778		spa_vdev_resilver_done(spa);
5779
5780	/*
5781	 * Kick off a resilver.
5782	 */
5783	if (tasks & SPA_ASYNC_RESILVER)
5784		dsl_resilver_restart(spa->spa_dsl_pool, 0);
5785
5786	/*
5787	 * Let the world know that we're done.
5788	 */
5789	mutex_enter(&spa->spa_async_lock);
5790	spa->spa_async_thread = NULL;
5791	cv_broadcast(&spa->spa_async_cv);
5792	mutex_exit(&spa->spa_async_lock);
5793	thread_exit();
5794}
5795
5796void
5797spa_async_suspend(spa_t *spa)
5798{
5799	mutex_enter(&spa->spa_async_lock);
5800	spa->spa_async_suspended++;
5801	while (spa->spa_async_thread != NULL)
5802		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5803	mutex_exit(&spa->spa_async_lock);
5804}
5805
5806void
5807spa_async_resume(spa_t *spa)
5808{
5809	mutex_enter(&spa->spa_async_lock);
5810	ASSERT(spa->spa_async_suspended != 0);
5811	spa->spa_async_suspended--;
5812	mutex_exit(&spa->spa_async_lock);
5813}
5814
5815static boolean_t
5816spa_async_tasks_pending(spa_t *spa)
5817{
5818	uint_t non_config_tasks;
5819	uint_t config_task;
5820	boolean_t config_task_suspended;
5821
5822	non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5823	config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5824	if (spa->spa_ccw_fail_time == 0) {
5825		config_task_suspended = B_FALSE;
5826	} else {
5827		config_task_suspended =
5828		    (gethrtime() - spa->spa_ccw_fail_time) <
5829		    (zfs_ccw_retry_interval * NANOSEC);
5830	}
5831
5832	return (non_config_tasks || (config_task && !config_task_suspended));
5833}
5834
5835static void
5836spa_async_dispatch(spa_t *spa)
5837{
5838	mutex_enter(&spa->spa_async_lock);
5839	if (spa_async_tasks_pending(spa) &&
5840	    !spa->spa_async_suspended &&
5841	    spa->spa_async_thread == NULL &&
5842	    rootdir != NULL)
5843		spa->spa_async_thread = thread_create(NULL, 0,
5844		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5845	mutex_exit(&spa->spa_async_lock);
5846}
5847
5848void
5849spa_async_request(spa_t *spa, int task)
5850{
5851	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5852	mutex_enter(&spa->spa_async_lock);
5853	spa->spa_async_tasks |= task;
5854	mutex_exit(&spa->spa_async_lock);
5855}
5856
5857/*
5858 * ==========================================================================
5859 * SPA syncing routines
5860 * ==========================================================================
5861 */
5862
5863static int
5864bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5865{
5866	bpobj_t *bpo = arg;
5867	bpobj_enqueue(bpo, bp, tx);
5868	return (0);
5869}
5870
5871static int
5872spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5873{
5874	zio_t *zio = arg;
5875
5876	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5877	    zio->io_flags));
5878	return (0);
5879}
5880
5881/*
5882 * Note: this simple function is not inlined to make it easier to dtrace the
5883 * amount of time spent syncing frees.
5884 */
5885static void
5886spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
5887{
5888	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5889	bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
5890	VERIFY(zio_wait(zio) == 0);
5891}
5892
5893/*
5894 * Note: this simple function is not inlined to make it easier to dtrace the
5895 * amount of time spent syncing deferred frees.
5896 */
5897static void
5898spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
5899{
5900	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5901	VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
5902	    spa_free_sync_cb, zio, tx), ==, 0);
5903	VERIFY0(zio_wait(zio));
5904}
5905
5906
5907static void
5908spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5909{
5910	char *packed = NULL;
5911	size_t bufsize;
5912	size_t nvsize = 0;
5913	dmu_buf_t *db;
5914
5915	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5916
5917	/*
5918	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5919	 * information.  This avoids the dmu_buf_will_dirty() path and
5920	 * saves us a pre-read to get data we don't actually care about.
5921	 */
5922	bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5923	packed = kmem_alloc(bufsize, KM_SLEEP);
5924
5925	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5926	    KM_SLEEP) == 0);
5927	bzero(packed + nvsize, bufsize - nvsize);
5928
5929	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5930
5931	kmem_free(packed, bufsize);
5932
5933	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5934	dmu_buf_will_dirty(db, tx);
5935	*(uint64_t *)db->db_data = nvsize;
5936	dmu_buf_rele(db, FTAG);
5937}
5938
5939static void
5940spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5941    const char *config, const char *entry)
5942{
5943	nvlist_t *nvroot;
5944	nvlist_t **list;
5945	int i;
5946
5947	if (!sav->sav_sync)
5948		return;
5949
5950	/*
5951	 * Update the MOS nvlist describing the list of available devices.
5952	 * spa_validate_aux() will have already made sure this nvlist is
5953	 * valid and the vdevs are labeled appropriately.
5954	 */
5955	if (sav->sav_object == 0) {
5956		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5957		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5958		    sizeof (uint64_t), tx);
5959		VERIFY(zap_update(spa->spa_meta_objset,
5960		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5961		    &sav->sav_object, tx) == 0);
5962	}
5963
5964	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5965	if (sav->sav_count == 0) {
5966		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5967	} else {
5968		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5969		for (i = 0; i < sav->sav_count; i++)
5970			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5971			    B_FALSE, VDEV_CONFIG_L2CACHE);
5972		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5973		    sav->sav_count) == 0);
5974		for (i = 0; i < sav->sav_count; i++)
5975			nvlist_free(list[i]);
5976		kmem_free(list, sav->sav_count * sizeof (void *));
5977	}
5978
5979	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5980	nvlist_free(nvroot);
5981
5982	sav->sav_sync = B_FALSE;
5983}
5984
5985static void
5986spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5987{
5988	nvlist_t *config;
5989
5990	if (list_is_empty(&spa->spa_config_dirty_list))
5991		return;
5992
5993	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5994
5995	config = spa_config_generate(spa, spa->spa_root_vdev,
5996	    dmu_tx_get_txg(tx), B_FALSE);
5997
5998	/*
5999	 * If we're upgrading the spa version then make sure that
6000	 * the config object gets updated with the correct version.
6001	 */
6002	if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
6003		fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6004		    spa->spa_uberblock.ub_version);
6005
6006	spa_config_exit(spa, SCL_STATE, FTAG);
6007
6008	if (spa->spa_config_syncing)
6009		nvlist_free(spa->spa_config_syncing);
6010	spa->spa_config_syncing = config;
6011
6012	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
6013}
6014
6015static void
6016spa_sync_version(void *arg, dmu_tx_t *tx)
6017{
6018	uint64_t *versionp = arg;
6019	uint64_t version = *versionp;
6020	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6021
6022	/*
6023	 * Setting the version is special cased when first creating the pool.
6024	 */
6025	ASSERT(tx->tx_txg != TXG_INITIAL);
6026
6027	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
6028	ASSERT(version >= spa_version(spa));
6029
6030	spa->spa_uberblock.ub_version = version;
6031	vdev_config_dirty(spa->spa_root_vdev);
6032	spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6033}
6034
6035/*
6036 * Set zpool properties.
6037 */
6038static void