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 */
28
29/*
30 * SPA: Storage Pool Allocator
31 *
32 * This file contains all the routines used when modifying on-disk SPA state.
33 * This includes opening, importing, destroying, exporting a pool, and syncing a
34 * pool.
35 */
36
37#include <sys/zfs_context.h>
38#include <sys/fm/fs/zfs.h>
39#include <sys/spa_impl.h>
40#include <sys/zio.h>
41#include <sys/zio_checksum.h>
42#include <sys/dmu.h>
43#include <sys/dmu_tx.h>
44#include <sys/zap.h>
45#include <sys/zil.h>
46#include <sys/ddt.h>
47#include <sys/vdev_impl.h>
48#include <sys/metaslab.h>
49#include <sys/metaslab_impl.h>
50#include <sys/uberblock_impl.h>
51#include <sys/txg.h>
52#include <sys/avl.h>
53#include <sys/dmu_traverse.h>
54#include <sys/dmu_objset.h>
55#include <sys/unique.h>
56#include <sys/dsl_pool.h>
57#include <sys/dsl_dataset.h>
58#include <sys/dsl_dir.h>
59#include <sys/dsl_prop.h>
60#include <sys/dsl_synctask.h>
61#include <sys/fs/zfs.h>
62#include <sys/arc.h>
63#include <sys/callb.h>
64#include <sys/systeminfo.h>
65#include <sys/spa_boot.h>
66#include <sys/zfs_ioctl.h>
67#include <sys/dsl_scan.h>
68#include <sys/zfeature.h>
69#include <sys/dsl_destroy.h>
70
71#ifdef	_KERNEL
72#include <sys/bootprops.h>
73#include <sys/callb.h>
74#include <sys/cpupart.h>
75#include <sys/pool.h>
76#include <sys/sysdc.h>
77#include <sys/zone.h>
78#endif	/* _KERNEL */
79
80#include "zfs_prop.h"
81#include "zfs_comutil.h"
82
83/*
84 * The interval, in seconds, at which failed configuration cache file writes
85 * should be retried.
86 */
87static int zfs_ccw_retry_interval = 300;
88
89typedef enum zti_modes {
90	ZTI_MODE_FIXED,			/* value is # of threads (min 1) */
91	ZTI_MODE_BATCH,			/* cpu-intensive; value is ignored */
92	ZTI_MODE_NULL,			/* don't create a taskq */
93	ZTI_NMODES
94} zti_modes_t;
95
96#define	ZTI_P(n, q)	{ ZTI_MODE_FIXED, (n), (q) }
97#define	ZTI_BATCH	{ ZTI_MODE_BATCH, 0, 1 }
98#define	ZTI_NULL	{ ZTI_MODE_NULL, 0, 0 }
99
100#define	ZTI_N(n)	ZTI_P(n, 1)
101#define	ZTI_ONE		ZTI_N(1)
102
103typedef struct zio_taskq_info {
104	zti_modes_t zti_mode;
105	uint_t zti_value;
106	uint_t zti_count;
107} zio_taskq_info_t;
108
109static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
110	"issue", "issue_high", "intr", "intr_high"
111};
112
113/*
114 * This table defines the taskq settings for each ZFS I/O type. When
115 * initializing a pool, we use this table to create an appropriately sized
116 * taskq. Some operations are low volume and therefore have a small, static
117 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
118 * macros. Other operations process a large amount of data; the ZTI_BATCH
119 * macro causes us to create a taskq oriented for throughput. Some operations
120 * are so high frequency and short-lived that the taskq itself can become a a
121 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
122 * additional degree of parallelism specified by the number of threads per-
123 * taskq and the number of taskqs; when dispatching an event in this case, the
124 * particular taskq is chosen at random.
125 *
126 * The different taskq priorities are to handle the different contexts (issue
127 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
128 * need to be handled with minimum delay.
129 */
130const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
131	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
132	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* NULL */
133	{ ZTI_N(8),	ZTI_NULL,	ZTI_P(12, 8),	ZTI_NULL }, /* READ */
134	{ ZTI_BATCH,	ZTI_N(5),	ZTI_N(8),	ZTI_N(5) }, /* WRITE */
135	{ ZTI_P(12, 8),	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* FREE */
136	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* CLAIM */
137	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* IOCTL */
138};
139
140static void spa_sync_version(void *arg, dmu_tx_t *tx);
141static void spa_sync_props(void *arg, dmu_tx_t *tx);
142static boolean_t spa_has_active_shared_spare(spa_t *spa);
143static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
144    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
145    char **ereport);
146static void spa_vdev_resilver_done(spa_t *spa);
147
148uint_t		zio_taskq_batch_pct = 75;	/* 1 thread per cpu in pset */
149id_t		zio_taskq_psrset_bind = PS_NONE;
150boolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
151uint_t		zio_taskq_basedc = 80;		/* base duty cycle */
152
153boolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
154extern int	zfs_sync_pass_deferred_free;
155
156/*
157 * This (illegal) pool name is used when temporarily importing a spa_t in order
158 * to get the vdev stats associated with the imported devices.
159 */
160#define	TRYIMPORT_NAME	"$import"
161
162/*
163 * ==========================================================================
164 * SPA properties routines
165 * ==========================================================================
166 */
167
168/*
169 * Add a (source=src, propname=propval) list to an nvlist.
170 */
171static void
172spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
173    uint64_t intval, zprop_source_t src)
174{
175	const char *propname = zpool_prop_to_name(prop);
176	nvlist_t *propval;
177
178	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
179	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
180
181	if (strval != NULL)
182		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
183	else
184		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
185
186	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
187	nvlist_free(propval);
188}
189
190/*
191 * Get property values from the spa configuration.
192 */
193static void
194spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
195{
196	vdev_t *rvd = spa->spa_root_vdev;
197	dsl_pool_t *pool = spa->spa_dsl_pool;
198	uint64_t size, alloc, cap, version;
199	zprop_source_t src = ZPROP_SRC_NONE;
200	spa_config_dirent_t *dp;
201	metaslab_class_t *mc = spa_normal_class(spa);
202
203	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
204
205	if (rvd != NULL) {
206		alloc = metaslab_class_get_alloc(spa_normal_class(spa));
207		size = metaslab_class_get_space(spa_normal_class(spa));
208		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
209		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
210		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
211		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
212		    size - alloc, src);
213
214		spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
215		    metaslab_class_fragmentation(mc), src);
216		spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
217		    metaslab_class_expandable_space(mc), src);
218		spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
219		    (spa_mode(spa) == FREAD), src);
220
221		cap = (size == 0) ? 0 : (alloc * 100 / size);
222		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
223
224		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
225		    ddt_get_pool_dedup_ratio(spa), src);
226
227		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
228		    rvd->vdev_state, src);
229
230		version = spa_version(spa);
231		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
232			src = ZPROP_SRC_DEFAULT;
233		else
234			src = ZPROP_SRC_LOCAL;
235		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
236	}
237
238	if (pool != NULL) {
239		/*
240		 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
241		 * when opening pools before this version freedir will be NULL.
242		 */
243		if (pool->dp_free_dir != NULL) {
244			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
245			    dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
246			    src);
247		} else {
248			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
249			    NULL, 0, src);
250		}
251
252		if (pool->dp_leak_dir != NULL) {
253			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
254			    dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
255			    src);
256		} else {
257			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
258			    NULL, 0, src);
259		}
260	}
261
262	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
263
264	if (spa->spa_comment != NULL) {
265		spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
266		    0, ZPROP_SRC_LOCAL);
267	}
268
269	if (spa->spa_root != NULL)
270		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
271		    0, ZPROP_SRC_LOCAL);
272
273	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
274		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
275		    MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
276	} else {
277		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
278		    SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
279	}
280
281	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
282		if (dp->scd_path == NULL) {
283			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
284			    "none", 0, ZPROP_SRC_LOCAL);
285		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
286			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
287			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
288		}
289	}
290}
291
292/*
293 * Get zpool property values.
294 */
295int
296spa_prop_get(spa_t *spa, nvlist_t **nvp)
297{
298	objset_t *mos = spa->spa_meta_objset;
299	zap_cursor_t zc;
300	zap_attribute_t za;
301	int err;
302
303	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
304
305	mutex_enter(&spa->spa_props_lock);
306
307	/*
308	 * Get properties from the spa config.
309	 */
310	spa_prop_get_config(spa, nvp);
311
312	/* If no pool property object, no more prop to get. */
313	if (mos == NULL || spa->spa_pool_props_object == 0) {
314		mutex_exit(&spa->spa_props_lock);
315		return (0);
316	}
317
318	/*
319	 * Get properties from the MOS pool property object.
320	 */
321	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
322	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
323	    zap_cursor_advance(&zc)) {
324		uint64_t intval = 0;
325		char *strval = NULL;
326		zprop_source_t src = ZPROP_SRC_DEFAULT;
327		zpool_prop_t prop;
328
329		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
330			continue;
331
332		switch (za.za_integer_length) {
333		case 8:
334			/* integer property */
335			if (za.za_first_integer !=
336			    zpool_prop_default_numeric(prop))
337				src = ZPROP_SRC_LOCAL;
338
339			if (prop == ZPOOL_PROP_BOOTFS) {
340				dsl_pool_t *dp;
341				dsl_dataset_t *ds = NULL;
342
343				dp = spa_get_dsl(spa);
344				dsl_pool_config_enter(dp, FTAG);
345				if (err = dsl_dataset_hold_obj(dp,
346				    za.za_first_integer, FTAG, &ds)) {
347					dsl_pool_config_exit(dp, FTAG);
348					break;
349				}
350
351				strval = kmem_alloc(
352				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
353				    KM_SLEEP);
354				dsl_dataset_name(ds, strval);
355				dsl_dataset_rele(ds, FTAG);
356				dsl_pool_config_exit(dp, FTAG);
357			} else {
358				strval = NULL;
359				intval = za.za_first_integer;
360			}
361
362			spa_prop_add_list(*nvp, prop, strval, intval, src);
363
364			if (strval != NULL)
365				kmem_free(strval,
366				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
367
368			break;
369
370		case 1:
371			/* string property */
372			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
373			err = zap_lookup(mos, spa->spa_pool_props_object,
374			    za.za_name, 1, za.za_num_integers, strval);
375			if (err) {
376				kmem_free(strval, za.za_num_integers);
377				break;
378			}
379			spa_prop_add_list(*nvp, prop, strval, 0, src);
380			kmem_free(strval, za.za_num_integers);
381			break;
382
383		default:
384			break;
385		}
386	}
387	zap_cursor_fini(&zc);
388	mutex_exit(&spa->spa_props_lock);
389out:
390	if (err && err != ENOENT) {
391		nvlist_free(*nvp);
392		*nvp = NULL;
393		return (err);
394	}
395
396	return (0);
397}
398
399/*
400 * Validate the given pool properties nvlist and modify the list
401 * for the property values to be set.
402 */
403static int
404spa_prop_validate(spa_t *spa, nvlist_t *props)
405{
406	nvpair_t *elem;
407	int error = 0, reset_bootfs = 0;
408	uint64_t objnum = 0;
409	boolean_t has_feature = B_FALSE;
410
411	elem = NULL;
412	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
413		uint64_t intval;
414		char *strval, *slash, *check, *fname;
415		const char *propname = nvpair_name(elem);
416		zpool_prop_t prop = zpool_name_to_prop(propname);
417
418		switch (prop) {
419		case ZPROP_INVAL:
420			if (!zpool_prop_feature(propname)) {
421				error = SET_ERROR(EINVAL);
422				break;
423			}
424
425			/*
426			 * Sanitize the input.
427			 */
428			if (nvpair_type(elem) != DATA_TYPE_UINT64) {
429				error = SET_ERROR(EINVAL);
430				break;
431			}
432
433			if (nvpair_value_uint64(elem, &intval) != 0) {
434				error = SET_ERROR(EINVAL);
435				break;
436			}
437
438			if (intval != 0) {
439				error = SET_ERROR(EINVAL);
440				break;
441			}
442
443			fname = strchr(propname, '@') + 1;
444			if (zfeature_lookup_name(fname, NULL) != 0) {
445				error = SET_ERROR(EINVAL);
446				break;
447			}
448
449			has_feature = B_TRUE;
450			break;
451
452		case ZPOOL_PROP_VERSION:
453			error = nvpair_value_uint64(elem, &intval);
454			if (!error &&
455			    (intval < spa_version(spa) ||
456			    intval > SPA_VERSION_BEFORE_FEATURES ||
457			    has_feature))
458				error = SET_ERROR(EINVAL);
459			break;
460
461		case ZPOOL_PROP_DELEGATION:
462		case ZPOOL_PROP_AUTOREPLACE:
463		case ZPOOL_PROP_LISTSNAPS:
464		case ZPOOL_PROP_AUTOEXPAND:
465			error = nvpair_value_uint64(elem, &intval);
466			if (!error && intval > 1)
467				error = SET_ERROR(EINVAL);
468			break;
469
470		case ZPOOL_PROP_BOOTFS:
471			/*
472			 * If the pool version is less than SPA_VERSION_BOOTFS,
473			 * or the pool is still being created (version == 0),
474			 * the bootfs property cannot be set.
475			 */
476			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
477				error = SET_ERROR(ENOTSUP);
478				break;
479			}
480
481			/*
482			 * Make sure the vdev config is bootable
483			 */
484			if (!vdev_is_bootable(spa->spa_root_vdev)) {
485				error = SET_ERROR(ENOTSUP);
486				break;
487			}
488
489			reset_bootfs = 1;
490
491			error = nvpair_value_string(elem, &strval);
492
493			if (!error) {
494				objset_t *os;
495				uint64_t propval;
496
497				if (strval == NULL || strval[0] == '\0') {
498					objnum = zpool_prop_default_numeric(
499					    ZPOOL_PROP_BOOTFS);
500					break;
501				}
502
503				if (error = dmu_objset_hold(strval, FTAG, &os))
504					break;
505
506				/*
507				 * Must be ZPL, and its property settings
508				 * must be supported by GRUB (compression
509				 * is not gzip, and large blocks are not used).
510				 */
511
512				if (dmu_objset_type(os) != DMU_OST_ZFS) {
513					error = SET_ERROR(ENOTSUP);
514				} else if ((error =
515				    dsl_prop_get_int_ds(dmu_objset_ds(os),
516				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
517				    &propval)) == 0 &&
518				    !BOOTFS_COMPRESS_VALID(propval)) {
519					error = SET_ERROR(ENOTSUP);
520				} else if ((error =
521				    dsl_prop_get_int_ds(dmu_objset_ds(os),
522				    zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
523				    &propval)) == 0 &&
524				    propval > SPA_OLD_MAXBLOCKSIZE) {
525					error = SET_ERROR(ENOTSUP);
526				} else {
527					objnum = dmu_objset_id(os);
528				}
529				dmu_objset_rele(os, FTAG);
530			}
531			break;
532
533		case ZPOOL_PROP_FAILUREMODE:
534			error = nvpair_value_uint64(elem, &intval);
535			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
536			    intval > ZIO_FAILURE_MODE_PANIC))
537				error = SET_ERROR(EINVAL);
538
539			/*
540			 * This is a special case which only occurs when
541			 * the pool has completely failed. This allows
542			 * the user to change the in-core failmode property
543			 * without syncing it out to disk (I/Os might
544			 * currently be blocked). We do this by returning
545			 * EIO to the caller (spa_prop_set) to trick it
546			 * into thinking we encountered a property validation
547			 * error.
548			 */
549			if (!error && spa_suspended(spa)) {
550				spa->spa_failmode = intval;
551				error = SET_ERROR(EIO);
552			}
553			break;
554
555		case ZPOOL_PROP_CACHEFILE:
556			if ((error = nvpair_value_string(elem, &strval)) != 0)
557				break;
558
559			if (strval[0] == '\0')
560				break;
561
562			if (strcmp(strval, "none") == 0)
563				break;
564
565			if (strval[0] != '/') {
566				error = SET_ERROR(EINVAL);
567				break;
568			}
569
570			slash = strrchr(strval, '/');
571			ASSERT(slash != NULL);
572
573			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
574			    strcmp(slash, "/..") == 0)
575				error = SET_ERROR(EINVAL);
576			break;
577
578		case ZPOOL_PROP_COMMENT:
579			if ((error = nvpair_value_string(elem, &strval)) != 0)
580				break;
581			for (check = strval; *check != '\0'; check++) {
582				/*
583				 * The kernel doesn't have an easy isprint()
584				 * check.  For this kernel check, we merely
585				 * check ASCII apart from DEL.  Fix this if
586				 * there is an easy-to-use kernel isprint().
587				 */
588				if (*check >= 0x7f) {
589					error = SET_ERROR(EINVAL);
590					break;
591				}
592				check++;
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	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2516		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2517	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2518	if (error != 0)
2519		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2520
2521	/*
2522	 * Load the bit that tells us to use the new accounting function
2523	 * (raid-z deflation).  If we have an older pool, this will not
2524	 * be present.
2525	 */
2526	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2527	if (error != 0 && error != ENOENT)
2528		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2529
2530	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2531	    &spa->spa_creation_version);
2532	if (error != 0 && error != ENOENT)
2533		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2534
2535	/*
2536	 * Load the persistent error log.  If we have an older pool, this will
2537	 * not be present.
2538	 */
2539	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
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_ERRLOG_SCRUB,
2544	    &spa->spa_errlog_scrub);
2545	if (error != 0 && error != ENOENT)
2546		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2547
2548	/*
2549	 * Load the history object.  If we have an older pool, this
2550	 * will not be present.
2551	 */
2552	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2553	if (error != 0 && error != ENOENT)
2554		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2555
2556	/*
2557	 * If we're assembling the pool from the split-off vdevs of
2558	 * an existing pool, we don't want to attach the spares & cache
2559	 * devices.
2560	 */
2561
2562	/*
2563	 * Load any hot spares for this pool.
2564	 */
2565	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2566	if (error != 0 && error != ENOENT)
2567		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2568	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2569		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2570		if (load_nvlist(spa, spa->spa_spares.sav_object,
2571		    &spa->spa_spares.sav_config) != 0)
2572			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2573
2574		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2575		spa_load_spares(spa);
2576		spa_config_exit(spa, SCL_ALL, FTAG);
2577	} else if (error == 0) {
2578		spa->spa_spares.sav_sync = B_TRUE;
2579	}
2580
2581	/*
2582	 * Load any level 2 ARC devices for this pool.
2583	 */
2584	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2585	    &spa->spa_l2cache.sav_object);
2586	if (error != 0 && error != ENOENT)
2587		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2588	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2589		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2590		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2591		    &spa->spa_l2cache.sav_config) != 0)
2592			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2593
2594		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2595		spa_load_l2cache(spa);
2596		spa_config_exit(spa, SCL_ALL, FTAG);
2597	} else if (error == 0) {
2598		spa->spa_l2cache.sav_sync = B_TRUE;
2599	}
2600
2601	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2602
2603	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2604	if (error && error != ENOENT)
2605		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2606
2607	if (error == 0) {
2608		uint64_t autoreplace;
2609
2610		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2611		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2612		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2613		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2614		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2615		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2616		    &spa->spa_dedup_ditto);
2617
2618		spa->spa_autoreplace = (autoreplace != 0);
2619	}
2620
2621	/*
2622	 * If the 'autoreplace' property is set, then post a resource notifying
2623	 * the ZFS DE that it should not issue any faults for unopenable
2624	 * devices.  We also iterate over the vdevs, and post a sysevent for any
2625	 * unopenable vdevs so that the normal autoreplace handler can take
2626	 * over.
2627	 */
2628	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2629		spa_check_removed(spa->spa_root_vdev);
2630		/*
2631		 * For the import case, this is done in spa_import(), because
2632		 * at this point we're using the spare definitions from
2633		 * the MOS config, not necessarily from the userland config.
2634		 */
2635		if (state != SPA_LOAD_IMPORT) {
2636			spa_aux_check_removed(&spa->spa_spares);
2637			spa_aux_check_removed(&spa->spa_l2cache);
2638		}
2639	}
2640
2641	/*
2642	 * Load the vdev state for all toplevel vdevs.
2643	 */
2644	vdev_load(rvd);
2645
2646	/*
2647	 * Propagate the leaf DTLs we just loaded all the way up the tree.
2648	 */
2649	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2650	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2651	spa_config_exit(spa, SCL_ALL, FTAG);
2652
2653	/*
2654	 * Load the DDTs (dedup tables).
2655	 */
2656	error = ddt_load(spa);
2657	if (error != 0)
2658		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2659
2660	spa_update_dspace(spa);
2661
2662	/*
2663	 * Validate the config, using the MOS config to fill in any
2664	 * information which might be missing.  If we fail to validate
2665	 * the config then declare the pool unfit for use. If we're
2666	 * assembling a pool from a split, the log is not transferred
2667	 * over.
2668	 */
2669	if (type != SPA_IMPORT_ASSEMBLE) {
2670		nvlist_t *nvconfig;
2671
2672		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2673			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2674
2675		if (!spa_config_valid(spa, nvconfig)) {
2676			nvlist_free(nvconfig);
2677			return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2678			    ENXIO));
2679		}
2680		nvlist_free(nvconfig);
2681
2682		/*
2683		 * Now that we've validated the config, check the state of the
2684		 * root vdev.  If it can't be opened, it indicates one or
2685		 * more toplevel vdevs are faulted.
2686		 */
2687		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2688			return (SET_ERROR(ENXIO));
2689
2690		if (spa_writeable(spa) && spa_check_logs(spa)) {
2691			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2692			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2693		}
2694	}
2695
2696	if (missing_feat_write) {
2697		ASSERT(state == SPA_LOAD_TRYIMPORT);
2698
2699		/*
2700		 * At this point, we know that we can open the pool in
2701		 * read-only mode but not read-write mode. We now have enough
2702		 * information and can return to userland.
2703		 */
2704		return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2705	}
2706
2707	/*
2708	 * We've successfully opened the pool, verify that we're ready
2709	 * to start pushing transactions.
2710	 */
2711	if (state != SPA_LOAD_TRYIMPORT) {
2712		if (error = spa_load_verify(spa))
2713			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2714			    error));
2715	}
2716
2717	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2718	    spa->spa_load_max_txg == UINT64_MAX)) {
2719		dmu_tx_t *tx;
2720		int need_update = B_FALSE;
2721		dsl_pool_t *dp = spa_get_dsl(spa);
2722
2723		ASSERT(state != SPA_LOAD_TRYIMPORT);
2724
2725		/*
2726		 * Claim log blocks that haven't been committed yet.
2727		 * This must all happen in a single txg.
2728		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2729		 * invoked from zil_claim_log_block()'s i/o done callback.
2730		 * Price of rollback is that we abandon the log.
2731		 */
2732		spa->spa_claiming = B_TRUE;
2733
2734		tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
2735		(void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
2736		    zil_claim, tx, DS_FIND_CHILDREN);
2737		dmu_tx_commit(tx);
2738
2739		spa->spa_claiming = B_FALSE;
2740
2741		spa_set_log_state(spa, SPA_LOG_GOOD);
2742		spa->spa_sync_on = B_TRUE;
2743		txg_sync_start(spa->spa_dsl_pool);
2744
2745		/*
2746		 * Wait for all claims to sync.  We sync up to the highest
2747		 * claimed log block birth time so that claimed log blocks
2748		 * don't appear to be from the future.  spa_claim_max_txg
2749		 * will have been set for us by either zil_check_log_chain()
2750		 * (invoked from spa_check_logs()) or zil_claim() above.
2751		 */
2752		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2753
2754		/*
2755		 * If the config cache is stale, or we have uninitialized
2756		 * metaslabs (see spa_vdev_add()), then update the config.
2757		 *
2758		 * If this is a verbatim import, trust the current
2759		 * in-core spa_config and update the disk labels.
2760		 */
2761		if (config_cache_txg != spa->spa_config_txg ||
2762		    state == SPA_LOAD_IMPORT ||
2763		    state == SPA_LOAD_RECOVER ||
2764		    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2765			need_update = B_TRUE;
2766
2767		for (int c = 0; c < rvd->vdev_children; c++)
2768			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2769				need_update = B_TRUE;
2770
2771		/*
2772		 * Update the config cache asychronously in case we're the
2773		 * root pool, in which case the config cache isn't writable yet.
2774		 */
2775		if (need_update)
2776			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2777
2778		/*
2779		 * Check all DTLs to see if anything needs resilvering.
2780		 */
2781		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2782		    vdev_resilver_needed(rvd, NULL, NULL))
2783			spa_async_request(spa, SPA_ASYNC_RESILVER);
2784
2785		/*
2786		 * Log the fact that we booted up (so that we can detect if
2787		 * we rebooted in the middle of an operation).
2788		 */
2789		spa_history_log_version(spa, "open");
2790
2791		/*
2792		 * Delete any inconsistent datasets.
2793		 */
2794		(void) dmu_objset_find(spa_name(spa),
2795		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2796
2797		/*
2798		 * Clean up any stale temporary dataset userrefs.
2799		 */
2800		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2801	}
2802
2803	return (0);
2804}
2805
2806static int
2807spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2808{
2809	int mode = spa->spa_mode;
2810
2811	spa_unload(spa);
2812	spa_deactivate(spa);
2813
2814	spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
2815
2816	spa_activate(spa, mode);
2817	spa_async_suspend(spa);
2818
2819	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2820}
2821
2822/*
2823 * If spa_load() fails this function will try loading prior txg's. If
2824 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2825 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2826 * function will not rewind the pool and will return the same error as
2827 * spa_load().
2828 */
2829static int
2830spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2831    uint64_t max_request, int rewind_flags)
2832{
2833	nvlist_t *loadinfo = NULL;
2834	nvlist_t *config = NULL;
2835	int load_error, rewind_error;
2836	uint64_t safe_rewind_txg;
2837	uint64_t min_txg;
2838
2839	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2840		spa->spa_load_max_txg = spa->spa_load_txg;
2841		spa_set_log_state(spa, SPA_LOG_CLEAR);
2842	} else {
2843		spa->spa_load_max_txg = max_request;
2844		if (max_request != UINT64_MAX)
2845			spa->spa_extreme_rewind = B_TRUE;
2846	}
2847
2848	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2849	    mosconfig);
2850	if (load_error == 0)
2851		return (0);
2852
2853	if (spa->spa_root_vdev != NULL)
2854		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2855
2856	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2857	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2858
2859	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2860		nvlist_free(config);
2861		return (load_error);
2862	}
2863
2864	if (state == SPA_LOAD_RECOVER) {
2865		/* Price of rolling back is discarding txgs, including log */
2866		spa_set_log_state(spa, SPA_LOG_CLEAR);
2867	} else {
2868		/*
2869		 * If we aren't rolling back save the load info from our first
2870		 * import attempt so that we can restore it after attempting
2871		 * to rewind.
2872		 */
2873		loadinfo = spa->spa_load_info;
2874		spa->spa_load_info = fnvlist_alloc();
2875	}
2876
2877	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2878	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2879	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2880	    TXG_INITIAL : safe_rewind_txg;
2881
2882	/*
2883	 * Continue as long as we're finding errors, we're still within
2884	 * the acceptable rewind range, and we're still finding uberblocks
2885	 */
2886	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2887	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2888		if (spa->spa_load_max_txg < safe_rewind_txg)
2889			spa->spa_extreme_rewind = B_TRUE;
2890		rewind_error = spa_load_retry(spa, state, mosconfig);
2891	}
2892
2893	spa->spa_extreme_rewind = B_FALSE;
2894	spa->spa_load_max_txg = UINT64_MAX;
2895
2896	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2897		spa_config_set(spa, config);
2898
2899	if (state == SPA_LOAD_RECOVER) {
2900		ASSERT3P(loadinfo, ==, NULL);
2901		return (rewind_error);
2902	} else {
2903		/* Store the rewind info as part of the initial load info */
2904		fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2905		    spa->spa_load_info);
2906
2907		/* Restore the initial load info */
2908		fnvlist_free(spa->spa_load_info);
2909		spa->spa_load_info = loadinfo;
2910
2911		return (load_error);
2912	}
2913}
2914
2915/*
2916 * Pool Open/Import
2917 *
2918 * The import case is identical to an open except that the configuration is sent
2919 * down from userland, instead of grabbed from the configuration cache.  For the
2920 * case of an open, the pool configuration will exist in the
2921 * POOL_STATE_UNINITIALIZED state.
2922 *
2923 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2924 * the same time open the pool, without having to keep around the spa_t in some
2925 * ambiguous state.
2926 */
2927static int
2928spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2929    nvlist_t **config)
2930{
2931	spa_t *spa;
2932	spa_load_state_t state = SPA_LOAD_OPEN;
2933	int error;
2934	int locked = B_FALSE;
2935
2936	*spapp = NULL;
2937
2938	/*
2939	 * As disgusting as this is, we need to support recursive calls to this
2940	 * function because dsl_dir_open() is called during spa_load(), and ends
2941	 * up calling spa_open() again.  The real fix is to figure out how to
2942	 * avoid dsl_dir_open() calling this in the first place.
2943	 */
2944	if (mutex_owner(&spa_namespace_lock) != curthread) {
2945		mutex_enter(&spa_namespace_lock);
2946		locked = B_TRUE;
2947	}
2948
2949	if ((spa = spa_lookup(pool)) == NULL) {
2950		if (locked)
2951			mutex_exit(&spa_namespace_lock);
2952		return (SET_ERROR(ENOENT));
2953	}
2954
2955	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2956		zpool_rewind_policy_t policy;
2957
2958		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2959		    &policy);
2960		if (policy.zrp_request & ZPOOL_DO_REWIND)
2961			state = SPA_LOAD_RECOVER;
2962
2963		spa_activate(spa, spa_mode_global);
2964
2965		if (state != SPA_LOAD_RECOVER)
2966			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2967
2968		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2969		    policy.zrp_request);
2970
2971		if (error == EBADF) {
2972			/*
2973			 * If vdev_validate() returns failure (indicated by
2974			 * EBADF), it indicates that one of the vdevs indicates
2975			 * that the pool has been exported or destroyed.  If
2976			 * this is the case, the config cache is out of sync and
2977			 * we should remove the pool from the namespace.
2978			 */
2979			spa_unload(spa);
2980			spa_deactivate(spa);
2981			spa_config_sync(spa, B_TRUE, B_TRUE);
2982			spa_remove(spa);
2983			if (locked)
2984				mutex_exit(&spa_namespace_lock);
2985			return (SET_ERROR(ENOENT));
2986		}
2987
2988		if (error) {
2989			/*
2990			 * We can't open the pool, but we still have useful
2991			 * information: the state of each vdev after the
2992			 * attempted vdev_open().  Return this to the user.
2993			 */
2994			if (config != NULL && spa->spa_config) {
2995				VERIFY(nvlist_dup(spa->spa_config, config,
2996				    KM_SLEEP) == 0);
2997				VERIFY(nvlist_add_nvlist(*config,
2998				    ZPOOL_CONFIG_LOAD_INFO,
2999				    spa->spa_load_info) == 0);
3000			}
3001			spa_unload(spa);
3002			spa_deactivate(spa);
3003			spa->spa_last_open_failed = error;
3004			if (locked)
3005				mutex_exit(&spa_namespace_lock);
3006			*spapp = NULL;
3007			return (error);
3008		}
3009	}
3010
3011	spa_open_ref(spa, tag);
3012
3013	if (config != NULL)
3014		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3015
3016	/*
3017	 * If we've recovered the pool, pass back any information we
3018	 * gathered while doing the load.
3019	 */
3020	if (state == SPA_LOAD_RECOVER) {
3021		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
3022		    spa->spa_load_info) == 0);
3023	}
3024
3025	if (locked) {
3026		spa->spa_last_open_failed = 0;
3027		spa->spa_last_ubsync_txg = 0;
3028		spa->spa_load_txg = 0;
3029		mutex_exit(&spa_namespace_lock);
3030	}
3031
3032	*spapp = spa;
3033
3034	return (0);
3035}
3036
3037int
3038spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3039    nvlist_t **config)
3040{
3041	return (spa_open_common(name, spapp, tag, policy, config));
3042}
3043
3044int
3045spa_open(const char *name, spa_t **spapp, void *tag)
3046{
3047	return (spa_open_common(name, spapp, tag, NULL, NULL));
3048}
3049
3050/*
3051 * Lookup the given spa_t, incrementing the inject count in the process,
3052 * preventing it from being exported or destroyed.
3053 */
3054spa_t *
3055spa_inject_addref(char *name)
3056{
3057	spa_t *spa;
3058
3059	mutex_enter(&spa_namespace_lock);
3060	if ((spa = spa_lookup(name)) == NULL) {
3061		mutex_exit(&spa_namespace_lock);
3062		return (NULL);
3063	}
3064	spa->spa_inject_ref++;
3065	mutex_exit(&spa_namespace_lock);
3066
3067	return (spa);
3068}
3069
3070void
3071spa_inject_delref(spa_t *spa)
3072{
3073	mutex_enter(&spa_namespace_lock);
3074	spa->spa_inject_ref--;
3075	mutex_exit(&spa_namespace_lock);
3076}
3077
3078/*
3079 * Add spares device information to the nvlist.
3080 */
3081static void
3082spa_add_spares(spa_t *spa, nvlist_t *config)
3083{
3084	nvlist_t **spares;
3085	uint_t i, nspares;
3086	nvlist_t *nvroot;
3087	uint64_t guid;
3088	vdev_stat_t *vs;
3089	uint_t vsc;
3090	uint64_t pool;
3091
3092	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3093
3094	if (spa->spa_spares.sav_count == 0)
3095		return;
3096
3097	VERIFY(nvlist_lookup_nvlist(config,
3098	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3099	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3100	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3101	if (nspares != 0) {
3102		VERIFY(nvlist_add_nvlist_array(nvroot,
3103		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3104		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3105		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3106
3107		/*
3108		 * Go through and find any spares which have since been
3109		 * repurposed as an active spare.  If this is the case, update
3110		 * their status appropriately.
3111		 */
3112		for (i = 0; i < nspares; i++) {
3113			VERIFY(nvlist_lookup_uint64(spares[i],
3114			    ZPOOL_CONFIG_GUID, &guid) == 0);
3115			if (spa_spare_exists(guid, &pool, NULL) &&
3116			    pool != 0ULL) {
3117				VERIFY(nvlist_lookup_uint64_array(
3118				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
3119				    (uint64_t **)&vs, &vsc) == 0);
3120				vs->vs_state = VDEV_STATE_CANT_OPEN;
3121				vs->vs_aux = VDEV_AUX_SPARED;
3122			}
3123		}
3124	}
3125}
3126
3127/*
3128 * Add l2cache device information to the nvlist, including vdev stats.
3129 */
3130static void
3131spa_add_l2cache(spa_t *spa, nvlist_t *config)
3132{
3133	nvlist_t **l2cache;
3134	uint_t i, j, nl2cache;
3135	nvlist_t *nvroot;
3136	uint64_t guid;
3137	vdev_t *vd;
3138	vdev_stat_t *vs;
3139	uint_t vsc;
3140
3141	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3142
3143	if (spa->spa_l2cache.sav_count == 0)
3144		return;
3145
3146	VERIFY(nvlist_lookup_nvlist(config,
3147	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3148	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3149	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3150	if (nl2cache != 0) {
3151		VERIFY(nvlist_add_nvlist_array(nvroot,
3152		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3153		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3154		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3155
3156		/*
3157		 * Update level 2 cache device stats.
3158		 */
3159
3160		for (i = 0; i < nl2cache; i++) {
3161			VERIFY(nvlist_lookup_uint64(l2cache[i],
3162			    ZPOOL_CONFIG_GUID, &guid) == 0);
3163
3164			vd = NULL;
3165			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3166				if (guid ==
3167				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3168					vd = spa->spa_l2cache.sav_vdevs[j];
3169					break;
3170				}
3171			}
3172			ASSERT(vd != NULL);
3173
3174			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3175			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3176			    == 0);
3177			vdev_get_stats(vd, vs);
3178		}
3179	}
3180}
3181
3182static void
3183spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3184{
3185	nvlist_t *features;
3186	zap_cursor_t zc;
3187	zap_attribute_t za;
3188
3189	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3190	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3191
3192	if (spa->spa_feat_for_read_obj != 0) {
3193		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3194		    spa->spa_feat_for_read_obj);
3195		    zap_cursor_retrieve(&zc, &za) == 0;
3196		    zap_cursor_advance(&zc)) {
3197			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3198			    za.za_num_integers == 1);
3199			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3200			    za.za_first_integer));
3201		}
3202		zap_cursor_fini(&zc);
3203	}
3204
3205	if (spa->spa_feat_for_write_obj != 0) {
3206		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3207		    spa->spa_feat_for_write_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	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3219	    features) == 0);
3220	nvlist_free(features);
3221}
3222
3223int
3224spa_get_stats(const char *name, nvlist_t **config,
3225    char *altroot, size_t buflen)
3226{
3227	int error;
3228	spa_t *spa;
3229
3230	*config = NULL;
3231	error = spa_open_common(name, &spa, FTAG, NULL, config);
3232
3233	if (spa != NULL) {
3234		/*
3235		 * This still leaves a window of inconsistency where the spares
3236		 * or l2cache devices could change and the config would be
3237		 * self-inconsistent.
3238		 */
3239		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3240
3241		if (*config != NULL) {
3242			uint64_t loadtimes[2];
3243
3244			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3245			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3246			VERIFY(nvlist_add_uint64_array(*config,
3247			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3248
3249			VERIFY(nvlist_add_uint64(*config,
3250			    ZPOOL_CONFIG_ERRCOUNT,
3251			    spa_get_errlog_size(spa)) == 0);
3252
3253			if (spa_suspended(spa))
3254				VERIFY(nvlist_add_uint64(*config,
3255				    ZPOOL_CONFIG_SUSPENDED,
3256				    spa->spa_failmode) == 0);
3257
3258			spa_add_spares(spa, *config);
3259			spa_add_l2cache(spa, *config);
3260			spa_add_feature_stats(spa, *config);
3261		}
3262	}
3263
3264	/*
3265	 * We want to get the alternate root even for faulted pools, so we cheat
3266	 * and call spa_lookup() directly.
3267	 */
3268	if (altroot) {
3269		if (spa == NULL) {
3270			mutex_enter(&spa_namespace_lock);
3271			spa = spa_lookup(name);
3272			if (spa)
3273				spa_altroot(spa, altroot, buflen);
3274			else
3275				altroot[0] = '\0';
3276			spa = NULL;
3277			mutex_exit(&spa_namespace_lock);
3278		} else {
3279			spa_altroot(spa, altroot, buflen);
3280		}
3281	}
3282
3283	if (spa != NULL) {
3284		spa_config_exit(spa, SCL_CONFIG, FTAG);
3285		spa_close(spa, FTAG);
3286	}
3287
3288	return (error);
3289}
3290
3291/*
3292 * Validate that the auxiliary device array is well formed.  We must have an
3293 * array of nvlists, each which describes a valid leaf vdev.  If this is an
3294 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3295 * specified, as long as they are well-formed.
3296 */
3297static int
3298spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3299    spa_aux_vdev_t *sav, const char *config, uint64_t version,
3300    vdev_labeltype_t label)
3301{
3302	nvlist_t **dev;
3303	uint_t i, ndev;
3304	vdev_t *vd;
3305	int error;
3306
3307	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3308
3309	/*
3310	 * It's acceptable to have no devs specified.
3311	 */
3312	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3313		return (0);
3314
3315	if (ndev == 0)
3316		return (SET_ERROR(EINVAL));
3317
3318	/*
3319	 * Make sure the pool is formatted with a version that supports this
3320	 * device type.
3321	 */
3322	if (spa_version(spa) < version)
3323		return (SET_ERROR(ENOTSUP));
3324
3325	/*
3326	 * Set the pending device list so we correctly handle device in-use
3327	 * checking.
3328	 */
3329	sav->sav_pending = dev;
3330	sav->sav_npending = ndev;
3331
3332	for (i = 0; i < ndev; i++) {
3333		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3334		    mode)) != 0)
3335			goto out;
3336
3337		if (!vd->vdev_ops->vdev_op_leaf) {
3338			vdev_free(vd);
3339			error = SET_ERROR(EINVAL);
3340			goto out;
3341		}
3342
3343		/*
3344		 * The L2ARC currently only supports disk devices in
3345		 * kernel context.  For user-level testing, we allow it.
3346		 */
3347#ifdef _KERNEL
3348		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3349		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3350			error = SET_ERROR(ENOTBLK);
3351			vdev_free(vd);
3352			goto out;
3353		}
3354#endif
3355		vd->vdev_top = vd;
3356
3357		if ((error = vdev_open(vd)) == 0 &&
3358		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
3359			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3360			    vd->vdev_guid) == 0);
3361		}
3362
3363		vdev_free(vd);
3364
3365		if (error &&
3366		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3367			goto out;
3368		else
3369			error = 0;
3370	}
3371
3372out:
3373	sav->sav_pending = NULL;
3374	sav->sav_npending = 0;
3375	return (error);
3376}
3377
3378static int
3379spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3380{
3381	int error;
3382
3383	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3384
3385	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3386	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3387	    VDEV_LABEL_SPARE)) != 0) {
3388		return (error);
3389	}
3390
3391	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3392	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3393	    VDEV_LABEL_L2CACHE));
3394}
3395
3396static void
3397spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3398    const char *config)
3399{
3400	int i;
3401
3402	if (sav->sav_config != NULL) {
3403		nvlist_t **olddevs;
3404		uint_t oldndevs;
3405		nvlist_t **newdevs;
3406
3407		/*
3408		 * Generate new dev list by concatentating with the
3409		 * current dev list.
3410		 */
3411		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3412		    &olddevs, &oldndevs) == 0);
3413
3414		newdevs = kmem_alloc(sizeof (void *) *
3415		    (ndevs + oldndevs), KM_SLEEP);
3416		for (i = 0; i < oldndevs; i++)
3417			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3418			    KM_SLEEP) == 0);
3419		for (i = 0; i < ndevs; i++)
3420			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3421			    KM_SLEEP) == 0);
3422
3423		VERIFY(nvlist_remove(sav->sav_config, config,
3424		    DATA_TYPE_NVLIST_ARRAY) == 0);
3425
3426		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3427		    config, newdevs, ndevs + oldndevs) == 0);
3428		for (i = 0; i < oldndevs + ndevs; i++)
3429			nvlist_free(newdevs[i]);
3430		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3431	} else {
3432		/*
3433		 * Generate a new dev list.
3434		 */
3435		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3436		    KM_SLEEP) == 0);
3437		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3438		    devs, ndevs) == 0);
3439	}
3440}
3441
3442/*
3443 * Stop and drop level 2 ARC devices
3444 */
3445void
3446spa_l2cache_drop(spa_t *spa)
3447{
3448	vdev_t *vd;
3449	int i;
3450	spa_aux_vdev_t *sav = &spa->spa_l2cache;
3451
3452	for (i = 0; i < sav->sav_count; i++) {
3453		uint64_t pool;
3454
3455		vd = sav->sav_vdevs[i];
3456		ASSERT(vd != NULL);
3457
3458		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3459		    pool != 0ULL && l2arc_vdev_present(vd))
3460			l2arc_remove_vdev(vd);
3461	}
3462}
3463
3464/*
3465 * Pool Creation
3466 */
3467int
3468spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3469    nvlist_t *zplprops)
3470{
3471	spa_t *spa;
3472	char *altroot = NULL;
3473	vdev_t *rvd;
3474	dsl_pool_t *dp;
3475	dmu_tx_t *tx;
3476	int error = 0;
3477	uint64_t txg = TXG_INITIAL;
3478	nvlist_t **spares, **l2cache;
3479	uint_t nspares, nl2cache;
3480	uint64_t version, obj;
3481	boolean_t has_features;
3482
3483	/*
3484	 * If this pool already exists, return failure.
3485	 */
3486	mutex_enter(&spa_namespace_lock);
3487	if (spa_lookup(pool) != NULL) {
3488		mutex_exit(&spa_namespace_lock);
3489		return (SET_ERROR(EEXIST));
3490	}
3491
3492	/*
3493	 * Allocate a new spa_t structure.
3494	 */
3495	(void) nvlist_lookup_string(props,
3496	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3497	spa = spa_add(pool, NULL, altroot);
3498	spa_activate(spa, spa_mode_global);
3499
3500	if (props && (error = spa_prop_validate(spa, props))) {
3501		spa_deactivate(spa);
3502		spa_remove(spa);
3503		mutex_exit(&spa_namespace_lock);
3504		return (error);
3505	}
3506
3507	has_features = B_FALSE;
3508	for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3509	    elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3510		if (zpool_prop_feature(nvpair_name(elem)))
3511			has_features = B_TRUE;
3512	}
3513
3514	if (has_features || nvlist_lookup_uint64(props,
3515	    zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3516		version = SPA_VERSION;
3517	}
3518	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3519
3520	spa->spa_first_txg = txg;
3521	spa->spa_uberblock.ub_txg = txg - 1;
3522	spa->spa_uberblock.ub_version = version;
3523	spa->spa_ubsync = spa->spa_uberblock;
3524
3525	/*
3526	 * Create "The Godfather" zio to hold all async IOs
3527	 */
3528	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
3529	    KM_SLEEP);
3530	for (int i = 0; i < max_ncpus; i++) {
3531		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
3532		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
3533		    ZIO_FLAG_GODFATHER);
3534	}
3535
3536	/*
3537	 * Create the root vdev.
3538	 */
3539	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3540
3541	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3542
3543	ASSERT(error != 0 || rvd != NULL);
3544	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3545
3546	if (error == 0 && !zfs_allocatable_devs(nvroot))
3547		error = SET_ERROR(EINVAL);
3548
3549	if (error == 0 &&
3550	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3551	    (error = spa_validate_aux(spa, nvroot, txg,
3552	    VDEV_ALLOC_ADD)) == 0) {
3553		for (int c = 0; c < rvd->vdev_children; c++) {
3554			vdev_metaslab_set_size(rvd->vdev_child[c]);
3555			vdev_expand(rvd->vdev_child[c], txg);
3556		}
3557	}
3558
3559	spa_config_exit(spa, SCL_ALL, FTAG);
3560
3561	if (error != 0) {
3562		spa_unload(spa);
3563		spa_deactivate(spa);
3564		spa_remove(spa);
3565		mutex_exit(&spa_namespace_lock);
3566		return (error);
3567	}
3568
3569	/*
3570	 * Get the list of spares, if specified.
3571	 */
3572	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3573	    &spares, &nspares) == 0) {
3574		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3575		    KM_SLEEP) == 0);
3576		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3577		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3578		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3579		spa_load_spares(spa);
3580		spa_config_exit(spa, SCL_ALL, FTAG);
3581		spa->spa_spares.sav_sync = B_TRUE;
3582	}
3583
3584	/*
3585	 * Get the list of level 2 cache devices, if specified.
3586	 */
3587	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3588	    &l2cache, &nl2cache) == 0) {
3589		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3590		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3591		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3592		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3593		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3594		spa_load_l2cache(spa);
3595		spa_config_exit(spa, SCL_ALL, FTAG);
3596		spa->spa_l2cache.sav_sync = B_TRUE;
3597	}
3598
3599	spa->spa_is_initializing = B_TRUE;
3600	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3601	spa->spa_meta_objset = dp->dp_meta_objset;
3602	spa->spa_is_initializing = B_FALSE;
3603
3604	/*
3605	 * Create DDTs (dedup tables).
3606	 */
3607	ddt_create(spa);
3608
3609	spa_update_dspace(spa);
3610
3611	tx = dmu_tx_create_assigned(dp, txg);
3612
3613	/*
3614	 * Create the pool config object.
3615	 */
3616	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3617	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3618	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3619
3620	if (zap_add(spa->spa_meta_objset,
3621	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3622	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3623		cmn_err(CE_PANIC, "failed to add pool config");
3624	}
3625
3626	if (spa_version(spa) >= SPA_VERSION_FEATURES)
3627		spa_feature_create_zap_objects(spa, tx);
3628
3629	if (zap_add(spa->spa_meta_objset,
3630	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3631	    sizeof (uint64_t), 1, &version, tx) != 0) {
3632		cmn_err(CE_PANIC, "failed to add pool version");
3633	}
3634
3635	/* Newly created pools with the right version are always deflated. */
3636	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3637		spa->spa_deflate = TRUE;
3638		if (zap_add(spa->spa_meta_objset,
3639		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3640		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3641			cmn_err(CE_PANIC, "failed to add deflate");
3642		}
3643	}
3644
3645	/*
3646	 * Create the deferred-free bpobj.  Turn off compression
3647	 * because sync-to-convergence takes longer if the blocksize
3648	 * keeps changing.
3649	 */
3650	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3651	dmu_object_set_compress(spa->spa_meta_objset, obj,
3652	    ZIO_COMPRESS_OFF, tx);
3653	if (zap_add(spa->spa_meta_objset,
3654	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3655	    sizeof (uint64_t), 1, &obj, tx) != 0) {
3656		cmn_err(CE_PANIC, "failed to add bpobj");
3657	}
3658	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3659	    spa->spa_meta_objset, obj));
3660
3661	/*
3662	 * Create the pool's history object.
3663	 */
3664	if (version >= SPA_VERSION_ZPOOL_HISTORY)
3665		spa_history_create_obj(spa, tx);
3666
3667	/*
3668	 * Set pool properties.
3669	 */
3670	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3671	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3672	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3673	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3674
3675	if (props != NULL) {
3676		spa_configfile_set(spa, props, B_FALSE);
3677		spa_sync_props(props, tx);
3678	}
3679
3680	dmu_tx_commit(tx);
3681
3682	spa->spa_sync_on = B_TRUE;
3683	txg_sync_start(spa->spa_dsl_pool);
3684
3685	/*
3686	 * We explicitly wait for the first transaction to complete so that our
3687	 * bean counters are appropriately updated.
3688	 */
3689	txg_wait_synced(spa->spa_dsl_pool, txg);
3690
3691	spa_config_sync(spa, B_FALSE, B_TRUE);
3692	spa_event_notify(spa, NULL, ESC_ZFS_POOL_CREATE);
3693
3694	spa_history_log_version(spa, "create");
3695
3696	/*
3697	 * Don't count references from objsets that are already closed
3698	 * and are making their way through the eviction process.
3699	 */
3700	spa_evicting_os_wait(spa);
3701	spa->spa_minref = refcount_count(&spa->spa_refcount);
3702
3703	mutex_exit(&spa_namespace_lock);
3704
3705	return (0);
3706}
3707
3708#ifdef _KERNEL
3709/*
3710 * Get the root pool information from the root disk, then import the root pool
3711 * during the system boot up time.
3712 */
3713extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3714
3715static nvlist_t *
3716spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3717{
3718	nvlist_t *config;
3719	nvlist_t *nvtop, *nvroot;
3720	uint64_t pgid;
3721
3722	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3723		return (NULL);
3724
3725	/*
3726	 * Add this top-level vdev to the child array.
3727	 */
3728	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3729	    &nvtop) == 0);
3730	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3731	    &pgid) == 0);
3732	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3733
3734	/*
3735	 * Put this pool's top-level vdevs into a root vdev.
3736	 */
3737	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3738	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3739	    VDEV_TYPE_ROOT) == 0);
3740	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3741	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3742	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3743	    &nvtop, 1) == 0);
3744
3745	/*
3746	 * Replace the existing vdev_tree with the new root vdev in
3747	 * this pool's configuration (remove the old, add the new).
3748	 */
3749	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3750	nvlist_free(nvroot);
3751	return (config);
3752}
3753
3754/*
3755 * Walk the vdev tree and see if we can find a device with "better"
3756 * configuration. A configuration is "better" if the label on that
3757 * device has a more recent txg.
3758 */
3759static void
3760spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3761{
3762	for (int c = 0; c < vd->vdev_children; c++)
3763		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3764
3765	if (vd->vdev_ops->vdev_op_leaf) {
3766		nvlist_t *label;
3767		uint64_t label_txg;
3768
3769		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3770		    &label) != 0)
3771			return;
3772
3773		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3774		    &label_txg) == 0);
3775
3776		/*
3777		 * Do we have a better boot device?
3778		 */
3779		if (label_txg > *txg) {
3780			*txg = label_txg;
3781			*avd = vd;
3782		}
3783		nvlist_free(label);
3784	}
3785}
3786
3787/*
3788 * Import a root pool.
3789 *
3790 * For x86. devpath_list will consist of devid and/or physpath name of
3791 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3792 * The GRUB "findroot" command will return the vdev we should boot.
3793 *
3794 * For Sparc, devpath_list consists the physpath name of the booting device
3795 * no matter the rootpool is a single device pool or a mirrored pool.
3796 * e.g.
3797 *	"/pci@1f,0/ide@d/disk@0,0:a"
3798 */
3799int
3800spa_import_rootpool(char *devpath, char *devid)
3801{
3802	spa_t *spa;
3803	vdev_t *rvd, *bvd, *avd = NULL;
3804	nvlist_t *config, *nvtop;
3805	uint64_t guid, txg;
3806	char *pname;
3807	int error;
3808
3809	/*
3810	 * Read the label from the boot device and generate a configuration.
3811	 */
3812	config = spa_generate_rootconf(devpath, devid, &guid);
3813#if defined(_OBP) && defined(_KERNEL)
3814	if (config == NULL) {
3815		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3816			/* iscsi boot */
3817			get_iscsi_bootpath_phy(devpath);
3818			config = spa_generate_rootconf(devpath, devid, &guid);
3819		}
3820	}
3821#endif
3822	if (config == NULL) {
3823		cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3824		    devpath);
3825		return (SET_ERROR(EIO));
3826	}
3827
3828	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3829	    &pname) == 0);
3830	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3831
3832	mutex_enter(&spa_namespace_lock);
3833	if ((spa = spa_lookup(pname)) != NULL) {
3834		/*
3835		 * Remove the existing root pool from the namespace so that we
3836		 * can replace it with the correct config we just read in.
3837		 */
3838		spa_remove(spa);
3839	}
3840
3841	spa = spa_add(pname, config, NULL);
3842	spa->spa_is_root = B_TRUE;
3843	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3844
3845	/*
3846	 * Build up a vdev tree based on the boot device's label config.
3847	 */
3848	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3849	    &nvtop) == 0);
3850	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3851	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3852	    VDEV_ALLOC_ROOTPOOL);
3853	spa_config_exit(spa, SCL_ALL, FTAG);
3854	if (error) {
3855		mutex_exit(&spa_namespace_lock);
3856		nvlist_free(config);
3857		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3858		    pname);
3859		return (error);
3860	}
3861
3862	/*
3863	 * Get the boot vdev.
3864	 */
3865	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3866		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3867		    (u_longlong_t)guid);
3868		error = SET_ERROR(ENOENT);
3869		goto out;
3870	}
3871
3872	/*
3873	 * Determine if there is a better boot device.
3874	 */
3875	avd = bvd;
3876	spa_alt_rootvdev(rvd, &avd, &txg);
3877	if (avd != bvd) {
3878		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3879		    "try booting from '%s'", avd->vdev_path);
3880		error = SET_ERROR(EINVAL);
3881		goto out;
3882	}
3883
3884	/*
3885	 * If the boot device is part of a spare vdev then ensure that
3886	 * we're booting off the active spare.
3887	 */
3888	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3889	    !bvd->vdev_isspare) {
3890		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3891		    "try booting from '%s'",
3892		    bvd->vdev_parent->
3893		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3894		error = SET_ERROR(EINVAL);
3895		goto out;
3896	}
3897
3898	error = 0;
3899out:
3900	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3901	vdev_free(rvd);
3902	spa_config_exit(spa, SCL_ALL, FTAG);
3903	mutex_exit(&spa_namespace_lock);
3904
3905	nvlist_free(config);
3906	return (error);
3907}
3908
3909#endif
3910
3911/*
3912 * Import a non-root pool into the system.
3913 */
3914int
3915spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3916{
3917	spa_t *spa;
3918	char *altroot = NULL;
3919	spa_load_state_t state = SPA_LOAD_IMPORT;
3920	zpool_rewind_policy_t policy;
3921	uint64_t mode = spa_mode_global;
3922	uint64_t readonly = B_FALSE;
3923	int error;
3924	nvlist_t *nvroot;
3925	nvlist_t **spares, **l2cache;
3926	uint_t nspares, nl2cache;
3927
3928	/*
3929	 * If a pool with this name exists, return failure.
3930	 */
3931	mutex_enter(&spa_namespace_lock);
3932	if (spa_lookup(pool) != NULL) {
3933		mutex_exit(&spa_namespace_lock);
3934		return (SET_ERROR(EEXIST));
3935	}
3936
3937	/*
3938	 * Create and initialize the spa structure.
3939	 */
3940	(void) nvlist_lookup_string(props,
3941	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3942	(void) nvlist_lookup_uint64(props,
3943	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3944	if (readonly)
3945		mode = FREAD;
3946	spa = spa_add(pool, config, altroot);
3947	spa->spa_import_flags = flags;
3948
3949	/*
3950	 * Verbatim import - Take a pool and insert it into the namespace
3951	 * as if it had been loaded at boot.
3952	 */
3953	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3954		if (props != NULL)
3955			spa_configfile_set(spa, props, B_FALSE);
3956
3957		spa_config_sync(spa, B_FALSE, B_TRUE);
3958		spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
3959
3960		mutex_exit(&spa_namespace_lock);
3961		return (0);
3962	}
3963
3964	spa_activate(spa, mode);
3965
3966	/*
3967	 * Don't start async tasks until we know everything is healthy.
3968	 */
3969	spa_async_suspend(spa);
3970
3971	zpool_get_rewind_policy(config, &policy);
3972	if (policy.zrp_request & ZPOOL_DO_REWIND)
3973		state = SPA_LOAD_RECOVER;
3974
3975	/*
3976	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3977	 * because the user-supplied config is actually the one to trust when
3978	 * doing an import.
3979	 */
3980	if (state != SPA_LOAD_RECOVER)
3981		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3982
3983	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3984	    policy.zrp_request);
3985
3986	/*
3987	 * Propagate anything learned while loading the pool and pass it
3988	 * back to caller (i.e. rewind info, missing devices, etc).
3989	 */
3990	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3991	    spa->spa_load_info) == 0);
3992
3993	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3994	/*
3995	 * Toss any existing sparelist, as it doesn't have any validity
3996	 * anymore, and conflicts with spa_has_spare().
3997	 */
3998	if (spa->spa_spares.sav_config) {
3999		nvlist_free(spa->spa_spares.sav_config);
4000		spa->spa_spares.sav_config = NULL;
4001		spa_load_spares(spa);
4002	}
4003	if (spa->spa_l2cache.sav_config) {
4004		nvlist_free(spa->spa_l2cache.sav_config);
4005		spa->spa_l2cache.sav_config = NULL;
4006		spa_load_l2cache(spa);
4007	}
4008
4009	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4010	    &nvroot) == 0);
4011	if (error == 0)
4012		error = spa_validate_aux(spa, nvroot, -1ULL,
4013		    VDEV_ALLOC_SPARE);
4014	if (error == 0)
4015		error = spa_validate_aux(spa, nvroot, -1ULL,
4016		    VDEV_ALLOC_L2CACHE);
4017	spa_config_exit(spa, SCL_ALL, FTAG);
4018
4019	if (props != NULL)
4020		spa_configfile_set(spa, props, B_FALSE);
4021
4022	if (error != 0 || (props && spa_writeable(spa) &&
4023	    (error = spa_prop_set(spa, props)))) {
4024		spa_unload(spa);
4025		spa_deactivate(spa);
4026		spa_remove(spa);
4027		mutex_exit(&spa_namespace_lock);
4028		return (error);
4029	}
4030
4031	spa_async_resume(spa);
4032
4033	/*
4034	 * Override any spares and level 2 cache devices as specified by
4035	 * the user, as these may have correct device names/devids, etc.
4036	 */
4037	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4038	    &spares, &nspares) == 0) {
4039		if (spa->spa_spares.sav_config)
4040			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4041			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4042		else
4043			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4044			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4045		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4046		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4047		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4048		spa_load_spares(spa);
4049		spa_config_exit(spa, SCL_ALL, FTAG);
4050		spa->spa_spares.sav_sync = B_TRUE;
4051	}
4052	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4053	    &l2cache, &nl2cache) == 0) {
4054		if (spa->spa_l2cache.sav_config)
4055			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4056			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4057		else
4058			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4059			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4060		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4061		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4062		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4063		spa_load_l2cache(spa);
4064		spa_config_exit(spa, SCL_ALL, FTAG);
4065		spa->spa_l2cache.sav_sync = B_TRUE;
4066	}
4067
4068	/*
4069	 * Check for any removed devices.
4070	 */
4071	if (spa->spa_autoreplace) {
4072		spa_aux_check_removed(&spa->spa_spares);
4073		spa_aux_check_removed(&spa->spa_l2cache);
4074	}
4075
4076	if (spa_writeable(spa)) {
4077		/*
4078		 * Update the config cache to include the newly-imported pool.
4079		 */
4080		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4081	}
4082
4083	/*
4084	 * It's possible that the pool was expanded while it was exported.
4085	 * We kick off an async task to handle this for us.
4086	 */
4087	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4088
4089	spa_history_log_version(spa, "import");
4090
4091	spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4092
4093	mutex_exit(&spa_namespace_lock);
4094
4095	return (0);
4096}
4097
4098nvlist_t *
4099spa_tryimport(nvlist_t *tryconfig)
4100{
4101	nvlist_t *config = NULL;
4102	char *poolname;
4103	spa_t *spa;
4104	uint64_t state;
4105	int error;
4106
4107	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4108		return (NULL);
4109
4110	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4111		return (NULL);
4112
4113	/*
4114	 * Create and initialize the spa structure.
4115	 */
4116	mutex_enter(&spa_namespace_lock);
4117	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4118	spa_activate(spa, FREAD);
4119
4120	/*
4121	 * Pass off the heavy lifting to spa_load().
4122	 * Pass TRUE for mosconfig because the user-supplied config
4123	 * is actually the one to trust when doing an import.
4124	 */
4125	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4126
4127	/*
4128	 * If 'tryconfig' was at least parsable, return the current config.
4129	 */
4130	if (spa->spa_root_vdev != NULL) {
4131		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4132		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4133		    poolname) == 0);
4134		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4135		    state) == 0);
4136		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4137		    spa->spa_uberblock.ub_timestamp) == 0);
4138		VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4139		    spa->spa_load_info) == 0);
4140
4141		/*
4142		 * If the bootfs property exists on this pool then we
4143		 * copy it out so that external consumers can tell which
4144		 * pools are bootable.
4145		 */
4146		if ((!error || error == EEXIST) && spa->spa_bootfs) {
4147			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4148
4149			/*
4150			 * We have to play games with the name since the
4151			 * pool was opened as TRYIMPORT_NAME.
4152			 */
4153			if (dsl_dsobj_to_dsname(spa_name(spa),
4154			    spa->spa_bootfs, tmpname) == 0) {
4155				char *cp;
4156				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4157
4158				cp = strchr(tmpname, '/');
4159				if (cp == NULL) {
4160					(void) strlcpy(dsname, tmpname,
4161					    MAXPATHLEN);
4162				} else {
4163					(void) snprintf(dsname, MAXPATHLEN,
4164					    "%s/%s", poolname, ++cp);
4165				}
4166				VERIFY(nvlist_add_string(config,
4167				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4168				kmem_free(dsname, MAXPATHLEN);
4169			}
4170			kmem_free(tmpname, MAXPATHLEN);
4171		}
4172
4173		/*
4174		 * Add the list of hot spares and level 2 cache devices.
4175		 */
4176		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4177		spa_add_spares(spa, config);
4178		spa_add_l2cache(spa, config);
4179		spa_config_exit(spa, SCL_CONFIG, FTAG);
4180	}
4181
4182	spa_unload(spa);
4183	spa_deactivate(spa);
4184	spa_remove(spa);
4185	mutex_exit(&spa_namespace_lock);
4186
4187	return (config);
4188}
4189
4190/*
4191 * Pool export/destroy
4192 *
4193 * The act of destroying or exporting a pool is very simple.  We make sure there
4194 * is no more pending I/O and any references to the pool are gone.  Then, we
4195 * update the pool state and sync all the labels to disk, removing the
4196 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4197 * we don't sync the labels or remove the configuration cache.
4198 */
4199static int
4200spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4201    boolean_t force, boolean_t hardforce)
4202{
4203	spa_t *spa;
4204
4205	if (oldconfig)
4206		*oldconfig = NULL;
4207
4208	if (!(spa_mode_global & FWRITE))
4209		return (SET_ERROR(EROFS));
4210
4211	mutex_enter(&spa_namespace_lock);
4212	if ((spa = spa_lookup(pool)) == NULL) {
4213		mutex_exit(&spa_namespace_lock);
4214		return (SET_ERROR(ENOENT));
4215	}
4216
4217	/*
4218	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4219	 * reacquire the namespace lock, and see if we can export.
4220	 */
4221	spa_open_ref(spa, FTAG);
4222	mutex_exit(&spa_namespace_lock);
4223	spa_async_suspend(spa);
4224	mutex_enter(&spa_namespace_lock);
4225	spa_close(spa, FTAG);
4226
4227	/*
4228	 * The pool will be in core if it's openable,
4229	 * in which case we can modify its state.
4230	 */
4231	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4232		/*
4233		 * Objsets may be open only because they're dirty, so we
4234		 * have to force it to sync before checking spa_refcnt.
4235		 */
4236		txg_wait_synced(spa->spa_dsl_pool, 0);
4237		spa_evicting_os_wait(spa);
4238
4239		/*
4240		 * A pool cannot be exported or destroyed if there are active
4241		 * references.  If we are resetting a pool, allow references by
4242		 * fault injection handlers.
4243		 */
4244		if (!spa_refcount_zero(spa) ||
4245		    (spa->spa_inject_ref != 0 &&
4246		    new_state != POOL_STATE_UNINITIALIZED)) {
4247			spa_async_resume(spa);
4248			mutex_exit(&spa_namespace_lock);
4249			return (SET_ERROR(EBUSY));
4250		}
4251
4252		/*
4253		 * A pool cannot be exported if it has an active shared spare.
4254		 * This is to prevent other pools stealing the active spare
4255		 * from an exported pool. At user's own will, such pool can
4256		 * be forcedly exported.
4257		 */
4258		if (!force && new_state == POOL_STATE_EXPORTED &&
4259		    spa_has_active_shared_spare(spa)) {
4260			spa_async_resume(spa);
4261			mutex_exit(&spa_namespace_lock);
4262			return (SET_ERROR(EXDEV));
4263		}
4264
4265		/*
4266		 * We want this to be reflected on every label,
4267		 * so mark them all dirty.  spa_unload() will do the
4268		 * final sync that pushes these changes out.
4269		 */
4270		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4271			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4272			spa->spa_state = new_state;
4273			spa->spa_final_txg = spa_last_synced_txg(spa) +
4274			    TXG_DEFER_SIZE + 1;
4275			vdev_config_dirty(spa->spa_root_vdev);
4276			spa_config_exit(spa, SCL_ALL, FTAG);
4277		}
4278	}
4279
4280	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4281
4282	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4283		spa_unload(spa);
4284		spa_deactivate(spa);
4285	}
4286
4287	if (oldconfig && spa->spa_config)
4288		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4289
4290	if (new_state != POOL_STATE_UNINITIALIZED) {
4291		if (!hardforce)
4292			spa_config_sync(spa, B_TRUE, B_TRUE);
4293		spa_remove(spa);
4294	}
4295	mutex_exit(&spa_namespace_lock);
4296
4297	return (0);
4298}
4299
4300/*
4301 * Destroy a storage pool.
4302 */
4303int
4304spa_destroy(char *pool)
4305{
4306	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4307	    B_FALSE, B_FALSE));
4308}
4309
4310/*
4311 * Export a storage pool.
4312 */
4313int
4314spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4315    boolean_t hardforce)
4316{
4317	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4318	    force, hardforce));
4319}
4320
4321/*
4322 * Similar to spa_export(), this unloads the spa_t without actually removing it
4323 * from the namespace in any way.
4324 */
4325int
4326spa_reset(char *pool)
4327{
4328	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4329	    B_FALSE, B_FALSE));
4330}
4331
4332/*
4333 * ==========================================================================
4334 * Device manipulation
4335 * ==========================================================================
4336 */
4337
4338/*
4339 * Add a device to a storage pool.
4340 */
4341int
4342spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4343{
4344	uint64_t txg, id;
4345	int error;
4346	vdev_t *rvd = spa->spa_root_vdev;
4347	vdev_t *vd, *tvd;
4348	nvlist_t **spares, **l2cache;
4349	uint_t nspares, nl2cache;
4350
4351	ASSERT(spa_writeable(spa));
4352
4353	txg = spa_vdev_enter(spa);
4354
4355	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4356	    VDEV_ALLOC_ADD)) != 0)
4357		return (spa_vdev_exit(spa, NULL, txg, error));
4358
4359	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
4360
4361	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4362	    &nspares) != 0)
4363		nspares = 0;
4364
4365	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4366	    &nl2cache) != 0)
4367		nl2cache = 0;
4368
4369	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4370		return (spa_vdev_exit(spa, vd, txg, EINVAL));
4371
4372	if (vd->vdev_children != 0 &&
4373	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
4374		return (spa_vdev_exit(spa, vd, txg, error));
4375
4376	/*
4377	 * We must validate the spares and l2cache devices after checking the
4378	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
4379	 */
4380	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4381		return (spa_vdev_exit(spa, vd, txg, error));
4382
4383	/*
4384	 * Transfer each new top-level vdev from vd to rvd.
4385	 */
4386	for (int c = 0; c < vd->vdev_children; c++) {
4387
4388		/*
4389		 * Set the vdev id to the first hole, if one exists.
4390		 */
4391		for (id = 0; id < rvd->vdev_children; id++) {
4392			if (rvd->vdev_child[id]->vdev_ishole) {
4393				vdev_free(rvd->vdev_child[id]);
4394				break;
4395			}
4396		}
4397		tvd = vd->vdev_child[c];
4398		vdev_remove_child(vd, tvd);
4399		tvd->vdev_id = id;
4400		vdev_add_child(rvd, tvd);
4401		vdev_config_dirty(tvd);
4402	}
4403
4404	if (nspares != 0) {
4405		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4406		    ZPOOL_CONFIG_SPARES);
4407		spa_load_spares(spa);
4408		spa->spa_spares.sav_sync = B_TRUE;
4409	}
4410
4411	if (nl2cache != 0) {
4412		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4413		    ZPOOL_CONFIG_L2CACHE);
4414		spa_load_l2cache(spa);
4415		spa->spa_l2cache.sav_sync = B_TRUE;
4416	}
4417
4418	/*
4419	 * We have to be careful when adding new vdevs to an existing pool.
4420	 * If other threads start allocating from these vdevs before we
4421	 * sync the config cache, and we lose power, then upon reboot we may
4422	 * fail to open the pool because there are DVAs that the config cache
4423	 * can't translate.  Therefore, we first add the vdevs without
4424	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4425	 * and then let spa_config_update() initialize the new metaslabs.
4426	 *
4427	 * spa_load() checks for added-but-not-initialized vdevs, so that
4428	 * if we lose power at any point in this sequence, the remaining
4429	 * steps will be completed the next time we load the pool.
4430	 */
4431	(void) spa_vdev_exit(spa, vd, txg, 0);
4432
4433	mutex_enter(&spa_namespace_lock);
4434	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4435	spa_event_notify(spa, NULL, ESC_ZFS_VDEV_ADD);
4436	mutex_exit(&spa_namespace_lock);
4437
4438	return (0);
4439}
4440
4441/*
4442 * Attach a device to a mirror.  The arguments are the path to any device
4443 * in the mirror, and the nvroot for the new device.  If the path specifies
4444 * a device that is not mirrored, we automatically insert the mirror vdev.
4445 *
4446 * If 'replacing' is specified, the new device is intended to replace the
4447 * existing device; in this case the two devices are made into their own
4448 * mirror using the 'replacing' vdev, which is functionally identical to
4449 * the mirror vdev (it actually reuses all the same ops) but has a few
4450 * extra rules: you can't attach to it after it's been created, and upon
4451 * completion of resilvering, the first disk (the one being replaced)
4452 * is automatically detached.
4453 */
4454int
4455spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4456{
4457	uint64_t txg, dtl_max_txg;
4458	vdev_t *rvd = spa->spa_root_vdev;
4459	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4460	vdev_ops_t *pvops;
4461	char *oldvdpath, *newvdpath;
4462	int newvd_isspare;
4463	int error;
4464
4465	ASSERT(spa_writeable(spa));
4466
4467	txg = spa_vdev_enter(spa);
4468
4469	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4470
4471	if (oldvd == NULL)
4472		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4473
4474	if (!oldvd->vdev_ops->vdev_op_leaf)
4475		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4476
4477	pvd = oldvd->vdev_parent;
4478
4479	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4480	    VDEV_ALLOC_ATTACH)) != 0)
4481		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4482
4483	if (newrootvd->vdev_children != 1)
4484		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4485
4486	newvd = newrootvd->vdev_child[0];
4487
4488	if (!newvd->vdev_ops->vdev_op_leaf)
4489		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4490
4491	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4492		return (spa_vdev_exit(spa, newrootvd, txg, error));
4493
4494	/*
4495	 * Spares can't replace logs
4496	 */
4497	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4498		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4499
4500	if (!replacing) {
4501		/*
4502		 * For attach, the only allowable parent is a mirror or the root
4503		 * vdev.
4504		 */
4505		if (pvd->vdev_ops != &vdev_mirror_ops &&
4506		    pvd->vdev_ops != &vdev_root_ops)
4507			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4508
4509		pvops = &vdev_mirror_ops;
4510	} else {
4511		/*
4512		 * Active hot spares can only be replaced by inactive hot
4513		 * spares.
4514		 */
4515		if (pvd->vdev_ops == &vdev_spare_ops &&
4516		    oldvd->vdev_isspare &&
4517		    !spa_has_spare(spa, newvd->vdev_guid))
4518			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4519
4520		/*
4521		 * If the source is a hot spare, and the parent isn't already a
4522		 * spare, then we want to create a new hot spare.  Otherwise, we
4523		 * want to create a replacing vdev.  The user is not allowed to
4524		 * attach to a spared vdev child unless the 'isspare' state is
4525		 * the same (spare replaces spare, non-spare replaces
4526		 * non-spare).
4527		 */
4528		if (pvd->vdev_ops == &vdev_replacing_ops &&
4529		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4530			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4531		} else if (pvd->vdev_ops == &vdev_spare_ops &&
4532		    newvd->vdev_isspare != oldvd->vdev_isspare) {
4533			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4534		}
4535
4536		if (newvd->vdev_isspare)
4537			pvops = &vdev_spare_ops;
4538		else
4539			pvops = &vdev_replacing_ops;
4540	}
4541
4542	/*
4543	 * Make sure the new device is big enough.
4544	 */
4545	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4546		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4547
4548	/*
4549	 * The new device cannot have a higher alignment requirement
4550	 * than the top-level vdev.
4551	 */
4552	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4553		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4554
4555	/*
4556	 * If this is an in-place replacement, update oldvd's path and devid
4557	 * to make it distinguishable from newvd, and unopenable from now on.
4558	 */
4559	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4560		spa_strfree(oldvd->vdev_path);
4561		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4562		    KM_SLEEP);
4563		(void) sprintf(oldvd->vdev_path, "%s/%s",
4564		    newvd->vdev_path, "old");
4565		if (oldvd->vdev_devid != NULL) {
4566			spa_strfree(oldvd->vdev_devid);
4567			oldvd->vdev_devid = NULL;
4568		}
4569	}
4570
4571	/* mark the device being resilvered */
4572	newvd->vdev_resilver_txg = txg;
4573
4574	/*
4575	 * If the parent is not a mirror, or if we're replacing, insert the new
4576	 * mirror/replacing/spare vdev above oldvd.
4577	 */
4578	if (pvd->vdev_ops != pvops)
4579		pvd = vdev_add_parent(oldvd, pvops);
4580
4581	ASSERT(pvd->vdev_top->vdev_parent == rvd);
4582	ASSERT(pvd->vdev_ops == pvops);
4583	ASSERT(oldvd->vdev_parent == pvd);
4584
4585	/*
4586	 * Extract the new device from its root and add it to pvd.
4587	 */
4588	vdev_remove_child(newrootvd, newvd);
4589	newvd->vdev_id = pvd->vdev_children;
4590	newvd->vdev_crtxg = oldvd->vdev_crtxg;
4591	vdev_add_child(pvd, newvd);
4592
4593	tvd = newvd->vdev_top;
4594	ASSERT(pvd->vdev_top == tvd);
4595	ASSERT(tvd->vdev_parent == rvd);
4596
4597	vdev_config_dirty(tvd);
4598
4599	/*
4600	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4601	 * for any dmu_sync-ed blocks.  It will propagate upward when
4602	 * spa_vdev_exit() calls vdev_dtl_reassess().
4603	 */
4604	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4605
4606	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4607	    dtl_max_txg - TXG_INITIAL);
4608
4609	if (newvd->vdev_isspare) {
4610		spa_spare_activate(newvd);
4611		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4612	}
4613
4614	oldvdpath = spa_strdup(oldvd->vdev_path);
4615	newvdpath = spa_strdup(newvd->vdev_path);
4616	newvd_isspare = newvd->vdev_isspare;
4617
4618	/*
4619	 * Mark newvd's DTL dirty in this txg.
4620	 */
4621	vdev_dirty(tvd, VDD_DTL, newvd, txg);
4622
4623	/*
4624	 * Schedule the resilver to restart in the future. We do this to
4625	 * ensure that dmu_sync-ed blocks have been stitched into the
4626	 * respective datasets.
4627	 */
4628	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4629
4630	if (spa->spa_bootfs)
4631		spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4632
4633	spa_event_notify(spa, newvd, ESC_ZFS_VDEV_ATTACH);
4634
4635	/*
4636	 * Commit the config
4637	 */
4638	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4639
4640	spa_history_log_internal(spa, "vdev attach", NULL,
4641	    "%s vdev=%s %s vdev=%s",
4642	    replacing && newvd_isspare ? "spare in" :
4643	    replacing ? "replace" : "attach", newvdpath,
4644	    replacing ? "for" : "to", oldvdpath);
4645
4646	spa_strfree(oldvdpath);
4647	spa_strfree(newvdpath);
4648
4649	return (0);
4650}
4651
4652/*
4653 * Detach a device from a mirror or replacing vdev.
4654 *
4655 * If 'replace_done' is specified, only detach if the parent
4656 * is a replacing vdev.
4657 */
4658int
4659spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4660{
4661	uint64_t txg;
4662	int error;
4663	vdev_t *rvd = spa->spa_root_vdev;
4664	vdev_t *vd, *pvd, *cvd, *tvd;
4665	boolean_t unspare = B_FALSE;
4666	uint64_t unspare_guid = 0;
4667	char *vdpath;
4668
4669	ASSERT(spa_writeable(spa));
4670
4671	txg = spa_vdev_enter(spa);
4672
4673	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4674
4675	if (vd == NULL)
4676		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4677
4678	if (!vd->vdev_ops->vdev_op_leaf)
4679		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4680
4681	pvd = vd->vdev_parent;
4682
4683	/*
4684	 * If the parent/child relationship is not as expected, don't do it.
4685	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4686	 * vdev that's replacing B with C.  The user's intent in replacing
4687	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4688	 * the replace by detaching C, the expected behavior is to end up
4689	 * M(A,B).  But suppose that right after deciding to detach C,
4690	 * the replacement of B completes.  We would have M(A,C), and then
4691	 * ask to detach C, which would leave us with just A -- not what
4692	 * the user wanted.  To prevent this, we make sure that the
4693	 * parent/child relationship hasn't changed -- in this example,
4694	 * that C's parent is still the replacing vdev R.
4695	 */
4696	if (pvd->vdev_guid != pguid && pguid != 0)
4697		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4698
4699	/*
4700	 * Only 'replacing' or 'spare' vdevs can be replaced.
4701	 */
4702	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4703	    pvd->vdev_ops != &vdev_spare_ops)
4704		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4705
4706	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4707	    spa_version(spa) >= SPA_VERSION_SPARES);
4708
4709	/*
4710	 * Only mirror, replacing, and spare vdevs support detach.
4711	 */
4712	if (pvd->vdev_ops != &vdev_replacing_ops &&
4713	    pvd->vdev_ops != &vdev_mirror_ops &&
4714	    pvd->vdev_ops != &vdev_spare_ops)
4715		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4716
4717	/*
4718	 * If this device has the only valid copy of some data,
4719	 * we cannot safely detach it.
4720	 */
4721	if (vdev_dtl_required(vd))
4722		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4723
4724	ASSERT(pvd->vdev_children >= 2);
4725
4726	/*
4727	 * If we are detaching the second disk from a replacing vdev, then
4728	 * check to see if we changed the original vdev's path to have "/old"
4729	 * at the end in spa_vdev_attach().  If so, undo that change now.
4730	 */
4731	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4732	    vd->vdev_path != NULL) {
4733		size_t len = strlen(vd->vdev_path);
4734
4735		for (int c = 0; c < pvd->vdev_children; c++) {
4736			cvd = pvd->vdev_child[c];
4737
4738			if (cvd == vd || cvd->vdev_path == NULL)
4739				continue;
4740
4741			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4742			    strcmp(cvd->vdev_path + len, "/old") == 0) {
4743				spa_strfree(cvd->vdev_path);
4744				cvd->vdev_path = spa_strdup(vd->vdev_path);
4745				break;
4746			}
4747		}
4748	}
4749
4750	/*
4751	 * If we are detaching the original disk from a spare, then it implies
4752	 * that the spare should become a real disk, and be removed from the
4753	 * active spare list for the pool.
4754	 */
4755	if (pvd->vdev_ops == &vdev_spare_ops &&
4756	    vd->vdev_id == 0 &&
4757	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4758		unspare = B_TRUE;
4759
4760	/*
4761	 * Erase the disk labels so the disk can be used for other things.
4762	 * This must be done after all other error cases are handled,
4763	 * but before we disembowel vd (so we can still do I/O to it).
4764	 * But if we can't do it, don't treat the error as fatal --
4765	 * it may be that the unwritability of the disk is the reason
4766	 * it's being detached!
4767	 */
4768	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4769
4770	/*
4771	 * Remove vd from its parent and compact the parent's children.
4772	 */
4773	vdev_remove_child(pvd, vd);
4774	vdev_compact_children(pvd);
4775
4776	/*
4777	 * Remember one of the remaining children so we can get tvd below.
4778	 */
4779	cvd = pvd->vdev_child[pvd->vdev_children - 1];
4780
4781	/*
4782	 * If we need to remove the remaining child from the list of hot spares,
4783	 * do it now, marking the vdev as no longer a spare in the process.
4784	 * We must do this before vdev_remove_parent(), because that can
4785	 * change the GUID if it creates a new toplevel GUID.  For a similar
4786	 * reason, we must remove the spare now, in the same txg as the detach;
4787	 * otherwise someone could attach a new sibling, change the GUID, and
4788	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4789	 */
4790	if (unspare) {
4791		ASSERT(cvd->vdev_isspare);
4792		spa_spare_remove(cvd);
4793		unspare_guid = cvd->vdev_guid;
4794		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4795		cvd->vdev_unspare = B_TRUE;
4796	}
4797
4798	/*
4799	 * If the parent mirror/replacing vdev only has one child,
4800	 * the parent is no longer needed.  Remove it from the tree.
4801	 */
4802	if (pvd->vdev_children == 1) {
4803		if (pvd->vdev_ops == &vdev_spare_ops)
4804			cvd->vdev_unspare = B_FALSE;
4805		vdev_remove_parent(cvd);
4806	}
4807
4808
4809	/*
4810	 * We don't set tvd until now because the parent we just removed
4811	 * may have been the previous top-level vdev.
4812	 */
4813	tvd = cvd->vdev_top;
4814	ASSERT(tvd->vdev_parent == rvd);
4815
4816	/*
4817	 * Reevaluate the parent vdev state.
4818	 */
4819	vdev_propagate_state(cvd);
4820
4821	/*
4822	 * If the 'autoexpand' property is set on the pool then automatically
4823	 * try to expand the size of the pool. For example if the device we
4824	 * just detached was smaller than the others, it may be possible to
4825	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4826	 * first so that we can obtain the updated sizes of the leaf vdevs.
4827	 */
4828	if (spa->spa_autoexpand) {
4829		vdev_reopen(tvd);
4830		vdev_expand(tvd, txg);
4831	}
4832
4833	vdev_config_dirty(tvd);
4834
4835	/*
4836	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4837	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4838	 * But first make sure we're not on any *other* txg's DTL list, to
4839	 * prevent vd from being accessed after it's freed.
4840	 */
4841	vdpath = spa_strdup(vd->vdev_path);
4842	for (int t = 0; t < TXG_SIZE; t++)
4843		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4844	vd->vdev_detached = B_TRUE;
4845	vdev_dirty(tvd, VDD_DTL, vd, txg);
4846
4847	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4848
4849	/* hang on to the spa before we release the lock */
4850	spa_open_ref(spa, FTAG);
4851
4852	error = spa_vdev_exit(spa, vd, txg, 0);
4853
4854	spa_history_log_internal(spa, "detach", NULL,
4855	    "vdev=%s", vdpath);
4856	spa_strfree(vdpath);
4857
4858	/*
4859	 * If this was the removal of the original device in a hot spare vdev,
4860	 * then we want to go through and remove the device from the hot spare
4861	 * list of every other pool.
4862	 */
4863	if (unspare) {
4864		spa_t *altspa = NULL;
4865
4866		mutex_enter(&spa_namespace_lock);
4867		while ((altspa = spa_next(altspa)) != NULL) {
4868			if (altspa->spa_state != POOL_STATE_ACTIVE ||
4869			    altspa == spa)
4870				continue;
4871
4872			spa_open_ref(altspa, FTAG);
4873			mutex_exit(&spa_namespace_lock);
4874			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4875			mutex_enter(&spa_namespace_lock);
4876			spa_close(altspa, FTAG);
4877		}
4878		mutex_exit(&spa_namespace_lock);
4879
4880		/* search the rest of the vdevs for spares to remove */
4881		spa_vdev_resilver_done(spa);
4882	}
4883
4884	/* all done with the spa; OK to release */
4885	mutex_enter(&spa_namespace_lock);
4886	spa_close(spa, FTAG);
4887	mutex_exit(&spa_namespace_lock);
4888
4889	return (error);
4890}
4891
4892/*
4893 * Split a set of devices from their mirrors, and create a new pool from them.
4894 */
4895int
4896spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4897    nvlist_t *props, boolean_t exp)
4898{
4899	int error = 0;
4900	uint64_t txg, *glist;
4901	spa_t *newspa;
4902	uint_t c, children, lastlog;
4903	nvlist_t **child, *nvl, *tmp;
4904	dmu_tx_t *tx;
4905	char *altroot = NULL;
4906	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4907	boolean_t activate_slog;
4908
4909	ASSERT(spa_writeable(spa));
4910
4911	txg = spa_vdev_enter(spa);
4912
4913	/* clear the log and flush everything up to now */
4914	activate_slog = spa_passivate_log(spa);
4915	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4916	error = spa_offline_log(spa);
4917	txg = spa_vdev_config_enter(spa);
4918
4919	if (activate_slog)
4920		spa_activate_log(spa);
4921
4922	if (error != 0)
4923		return (spa_vdev_exit(spa, NULL, txg, error));
4924
4925	/* check new spa name before going any further */
4926	if (spa_lookup(newname) != NULL)
4927		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4928
4929	/*
4930	 * scan through all the children to ensure they're all mirrors
4931	 */
4932	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4933	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4934	    &children) != 0)
4935		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4936
4937	/* first, check to ensure we've got the right child count */
4938	rvd = spa->spa_root_vdev;
4939	lastlog = 0;
4940	for (c = 0; c < rvd->vdev_children; c++) {
4941		vdev_t *vd = rvd->vdev_child[c];
4942
4943		/* don't count the holes & logs as children */
4944		if (vd->vdev_islog || vd->vdev_ishole) {
4945			if (lastlog == 0)
4946				lastlog = c;
4947			continue;
4948		}
4949
4950		lastlog = 0;
4951	}
4952	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4953		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4954
4955	/* next, ensure no spare or cache devices are part of the split */
4956	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4957	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4958		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4959
4960	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4961	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4962
4963	/* then, loop over each vdev and validate it */
4964	for (c = 0; c < children; c++) {
4965		uint64_t is_hole = 0;
4966
4967		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4968		    &is_hole);
4969
4970		if (is_hole != 0) {
4971			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4972			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4973				continue;
4974			} else {
4975				error = SET_ERROR(EINVAL);
4976				break;
4977			}
4978		}
4979
4980		/* which disk is going to be split? */
4981		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4982		    &glist[c]) != 0) {
4983			error = SET_ERROR(EINVAL);
4984			break;
4985		}
4986
4987		/* look it up in the spa */
4988		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4989		if (vml[c] == NULL) {
4990			error = SET_ERROR(ENODEV);
4991			break;
4992		}
4993
4994		/* make sure there's nothing stopping the split */
4995		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4996		    vml[c]->vdev_islog ||
4997		    vml[c]->vdev_ishole ||
4998		    vml[c]->vdev_isspare ||
4999		    vml[c]->vdev_isl2cache ||
5000		    !vdev_writeable(vml[c]) ||
5001		    vml[c]->vdev_children != 0 ||
5002		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5003		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5004			error = SET_ERROR(EINVAL);
5005			break;
5006		}
5007
5008		if (vdev_dtl_required(vml[c])) {
5009			error = SET_ERROR(EBUSY);
5010			break;
5011		}
5012
5013		/* we need certain info from the top level */
5014		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5015		    vml[c]->vdev_top->vdev_ms_array) == 0);
5016		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5017		    vml[c]->vdev_top->vdev_ms_shift) == 0);
5018		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5019		    vml[c]->vdev_top->vdev_asize) == 0);
5020		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5021		    vml[c]->vdev_top->vdev_ashift) == 0);
5022	}
5023
5024	if (error != 0) {
5025		kmem_free(vml, children * sizeof (vdev_t *));
5026		kmem_free(glist, children * sizeof (uint64_t));
5027		return (spa_vdev_exit(spa, NULL, txg, error));
5028	}
5029
5030	/* stop writers from using the disks */
5031	for (c = 0; c < children; c++) {
5032		if (vml[c] != NULL)
5033			vml[c]->vdev_offline = B_TRUE;
5034	}
5035	vdev_reopen(spa->spa_root_vdev);
5036
5037	/*
5038	 * Temporarily record the splitting vdevs in the spa config.  This
5039	 * will disappear once the config is regenerated.
5040	 */
5041	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5042	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5043	    glist, children) == 0);
5044	kmem_free(glist, children * sizeof (uint64_t));
5045
5046	mutex_enter(&spa->spa_props_lock);
5047	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5048	    nvl) == 0);
5049	mutex_exit(&spa->spa_props_lock);
5050	spa->spa_config_splitting = nvl;
5051	vdev_config_dirty(spa->spa_root_vdev);
5052
5053	/* configure and create the new pool */
5054	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5055	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5056	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5057	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5058	    spa_version(spa)) == 0);
5059	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5060	    spa->spa_config_txg) == 0);
5061	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5062	    spa_generate_guid(NULL)) == 0);
5063	(void) nvlist_lookup_string(props,
5064	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5065
5066	/* add the new pool to the namespace */
5067	newspa = spa_add(newname, config, altroot);
5068	newspa->spa_config_txg = spa->spa_config_txg;
5069	spa_set_log_state(newspa, SPA_LOG_CLEAR);
5070
5071	/* release the spa config lock, retaining the namespace lock */
5072	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5073
5074	if (zio_injection_enabled)
5075		zio_handle_panic_injection(spa, FTAG, 1);
5076
5077	spa_activate(newspa, spa_mode_global);
5078	spa_async_suspend(newspa);
5079
5080	/* create the new pool from the disks of the original pool */
5081	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5082	if (error)
5083		goto out;
5084
5085	/* if that worked, generate a real config for the new pool */
5086	if (newspa->spa_root_vdev != NULL) {
5087		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5088		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
5089		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5090		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5091		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5092		    B_TRUE));
5093	}
5094
5095	/* set the props */
5096	if (props != NULL) {
5097		spa_configfile_set(newspa, props, B_FALSE);
5098		error = spa_prop_set(newspa, props);
5099		if (error)
5100			goto out;
5101	}
5102
5103	/* flush everything */
5104	txg = spa_vdev_config_enter(newspa);
5105	vdev_config_dirty(newspa->spa_root_vdev);
5106	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5107
5108	if (zio_injection_enabled)
5109		zio_handle_panic_injection(spa, FTAG, 2);
5110
5111	spa_async_resume(newspa);
5112
5113	/* finally, update the original pool's config */
5114	txg = spa_vdev_config_enter(spa);
5115	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5116	error = dmu_tx_assign(tx, TXG_WAIT);
5117	if (error != 0)
5118		dmu_tx_abort(tx);
5119	for (c = 0; c < children; c++) {
5120		if (vml[c] != NULL) {
5121			vdev_split(vml[c]);
5122			if (error == 0)
5123				spa_history_log_internal(spa, "detach", tx,
5124				    "vdev=%s", vml[c]->vdev_path);
5125			vdev_free(vml[c]);
5126		}
5127	}
5128	vdev_config_dirty(spa->spa_root_vdev);
5129	spa->spa_config_splitting = NULL;
5130	nvlist_free(nvl);
5131	if (error == 0)
5132		dmu_tx_commit(tx);
5133	(void) spa_vdev_exit(spa, NULL, txg, 0);
5134
5135	if (zio_injection_enabled)
5136		zio_handle_panic_injection(spa, FTAG, 3);
5137
5138	/* split is complete; log a history record */
5139	spa_history_log_internal(newspa, "split", NULL,
5140	    "from pool %s", spa_name(spa));
5141
5142	kmem_free(vml, children * sizeof (vdev_t *));
5143
5144	/* if we're not going to mount the filesystems in userland, export */
5145	if (exp)
5146		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5147		    B_FALSE, B_FALSE);
5148
5149	return (error);
5150
5151out:
5152	spa_unload(newspa);
5153	spa_deactivate(newspa);
5154	spa_remove(newspa);
5155
5156	txg = spa_vdev_config_enter(spa);
5157
5158	/* re-online all offlined disks */
5159	for (c = 0; c < children; c++) {
5160		if (vml[c] != NULL)
5161			vml[c]->vdev_offline = B_FALSE;
5162	}
5163	vdev_reopen(spa->spa_root_vdev);
5164
5165	nvlist_free(spa->spa_config_splitting);
5166	spa->spa_config_splitting = NULL;
5167	(void) spa_vdev_exit(spa, NULL, txg, error);
5168
5169	kmem_free(vml, children * sizeof (vdev_t *));
5170	return (error);
5171}
5172
5173static nvlist_t *
5174spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5175{
5176	for (int i = 0; i < count; i++) {
5177		uint64_t guid;
5178
5179		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5180		    &guid) == 0);
5181
5182		if (guid == target_guid)
5183			return (nvpp[i]);
5184	}
5185
5186	return (NULL);
5187}
5188
5189static void
5190spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5191	nvlist_t *dev_to_remove)
5192{
5193	nvlist_t **newdev = NULL;
5194
5195	if (count > 1)
5196		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5197
5198	for (int i = 0, j = 0; i < count; i++) {
5199		if (dev[i] == dev_to_remove)
5200			continue;
5201		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5202	}
5203
5204	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5205	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5206
5207	for (int i = 0; i < count - 1; i++)
5208		nvlist_free(newdev[i]);
5209
5210	if (count > 1)
5211		kmem_free(newdev, (count - 1) * sizeof (void *));
5212}
5213
5214/*
5215 * Evacuate the device.
5216 */
5217static int
5218spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5219{
5220	uint64_t txg;
5221	int error = 0;
5222
5223	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5224	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5225	ASSERT(vd == vd->vdev_top);
5226
5227	/*
5228	 * Evacuate the device.  We don't hold the config lock as writer
5229	 * since we need to do I/O but we do keep the
5230	 * spa_namespace_lock held.  Once this completes the device
5231	 * should no longer have any blocks allocated on it.
5232	 */
5233	if (vd->vdev_islog) {
5234		if (vd->vdev_stat.vs_alloc != 0)
5235			error = spa_offline_log(spa);
5236	} else {
5237		error = SET_ERROR(ENOTSUP);
5238	}
5239
5240	if (error)
5241		return (error);
5242
5243	/*
5244	 * The evacuation succeeded.  Remove any remaining MOS metadata
5245	 * associated with this vdev, and wait for these changes to sync.
5246	 */
5247	ASSERT0(vd->vdev_stat.vs_alloc);
5248	txg = spa_vdev_config_enter(spa);
5249	vd->vdev_removing = B_TRUE;
5250	vdev_dirty_leaves(vd, VDD_DTL, txg);
5251	vdev_config_dirty(vd);
5252	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5253
5254	return (0);
5255}
5256
5257/*
5258 * Complete the removal by cleaning up the namespace.
5259 */
5260static void
5261spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5262{
5263	vdev_t *rvd = spa->spa_root_vdev;
5264	uint64_t id = vd->vdev_id;
5265	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5266
5267	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5268	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5269	ASSERT(vd == vd->vdev_top);
5270
5271	/*
5272	 * Only remove any devices which are empty.
5273	 */
5274	if (vd->vdev_stat.vs_alloc != 0)
5275		return;
5276
5277	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5278
5279	if (list_link_active(&vd->vdev_state_dirty_node))
5280		vdev_state_clean(vd);
5281	if (list_link_active(&vd->vdev_config_dirty_node))
5282		vdev_config_clean(vd);
5283
5284	vdev_free(vd);
5285
5286	if (last_vdev) {
5287		vdev_compact_children(rvd);
5288	} else {
5289		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5290		vdev_add_child(rvd, vd);
5291	}
5292	vdev_config_dirty(rvd);
5293
5294	/*
5295	 * Reassess the health of our root vdev.
5296	 */
5297	vdev_reopen(rvd);
5298}
5299
5300/*
5301 * Remove a device from the pool -
5302 *
5303 * Removing a device from the vdev namespace requires several steps
5304 * and can take a significant amount of time.  As a result we use
5305 * the spa_vdev_config_[enter/exit] functions which allow us to
5306 * grab and release the spa_config_lock while still holding the namespace
5307 * lock.  During each step the configuration is synced out.
5308 *
5309 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5310 * devices.
5311 */
5312int
5313spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5314{
5315	vdev_t *vd;
5316	metaslab_group_t *mg;
5317	nvlist_t **spares, **l2cache, *nv;
5318	uint64_t txg = 0;
5319	uint_t nspares, nl2cache;
5320	int error = 0;
5321	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5322
5323	ASSERT(spa_writeable(spa));
5324
5325	if (!locked)
5326		txg = spa_vdev_enter(spa);
5327
5328	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5329
5330	if (spa->spa_spares.sav_vdevs != NULL &&
5331	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5332	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5333	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5334		/*
5335		 * Only remove the hot spare if it's not currently in use
5336		 * in this pool.
5337		 */
5338		if (vd == NULL || unspare) {
5339			spa_vdev_remove_aux(spa->spa_spares.sav_config,
5340			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5341			spa_load_spares(spa);
5342			spa->spa_spares.sav_sync = B_TRUE;
5343		} else {
5344			error = SET_ERROR(EBUSY);
5345		}
5346	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
5347	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5348	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5349	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5350		/*
5351		 * Cache devices can always be removed.
5352		 */
5353		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5354		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5355		spa_load_l2cache(spa);
5356		spa->spa_l2cache.sav_sync = B_TRUE;
5357	} else if (vd != NULL && vd->vdev_islog) {
5358		ASSERT(!locked);
5359		ASSERT(vd == vd->vdev_top);
5360
5361		mg = vd->vdev_mg;
5362
5363		/*
5364		 * Stop allocating from this vdev.
5365		 */
5366		metaslab_group_passivate(mg);
5367
5368		/*
5369		 * Wait for the youngest allocations and frees to sync,
5370		 * and then wait for the deferral of those frees to finish.
5371		 */
5372		spa_vdev_config_exit(spa, NULL,
5373		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5374
5375		/*
5376		 * Attempt to evacuate the vdev.
5377		 */
5378		error = spa_vdev_remove_evacuate(spa, vd);
5379
5380		txg = spa_vdev_config_enter(spa);
5381
5382		/*
5383		 * If we couldn't evacuate the vdev, unwind.
5384		 */
5385		if (error) {
5386			metaslab_group_activate(mg);
5387			return (spa_vdev_exit(spa, NULL, txg, error));
5388		}
5389
5390		/*
5391		 * Clean up the vdev namespace.
5392		 */
5393		spa_vdev_remove_from_namespace(spa, vd);
5394
5395	} else if (vd != NULL) {
5396		/*
5397		 * Normal vdevs cannot be removed (yet).
5398		 */
5399		error = SET_ERROR(ENOTSUP);
5400	} else {
5401		/*
5402		 * There is no vdev of any kind with the specified guid.
5403		 */
5404		error = SET_ERROR(ENOENT);
5405	}
5406
5407	if (!locked)
5408		return (spa_vdev_exit(spa, NULL, txg, error));
5409
5410	return (error);
5411}
5412
5413/*
5414 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5415 * currently spared, so we can detach it.
5416 */
5417static vdev_t *
5418spa_vdev_resilver_done_hunt(vdev_t *vd)
5419{
5420	vdev_t *newvd, *oldvd;
5421
5422	for (int c = 0; c < vd->vdev_children; c++) {
5423		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5424		if (oldvd != NULL)
5425			return (oldvd);
5426	}
5427
5428	/*
5429	 * Check for a completed replacement.  We always consider the first
5430	 * vdev in the list to be the oldest vdev, and the last one to be
5431	 * the newest (see spa_vdev_attach() for how that works).  In
5432	 * the case where the newest vdev is faulted, we will not automatically
5433	 * remove it after a resilver completes.  This is OK as it will require
5434	 * user intervention to determine which disk the admin wishes to keep.
5435	 */
5436	if (vd->vdev_ops == &vdev_replacing_ops) {
5437		ASSERT(vd->vdev_children > 1);
5438
5439		newvd = vd->vdev_child[vd->vdev_children - 1];
5440		oldvd = vd->vdev_child[0];
5441
5442		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5443		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5444		    !vdev_dtl_required(oldvd))
5445			return (oldvd);
5446	}
5447
5448	/*
5449	 * Check for a completed resilver with the 'unspare' flag set.
5450	 */
5451	if (vd->vdev_ops == &vdev_spare_ops) {
5452		vdev_t *first = vd->vdev_child[0];
5453		vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5454
5455		if (last->vdev_unspare) {
5456			oldvd = first;
5457			newvd = last;
5458		} else if (first->vdev_unspare) {
5459			oldvd = last;
5460			newvd = first;
5461		} else {
5462			oldvd = NULL;
5463		}
5464
5465		if (oldvd != NULL &&
5466		    vdev_dtl_empty(newvd, DTL_MISSING) &&
5467		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5468		    !vdev_dtl_required(oldvd))
5469			return (oldvd);
5470
5471		/*
5472		 * If there are more than two spares attached to a disk,
5473		 * and those spares are not required, then we want to
5474		 * attempt to free them up now so that they can be used
5475		 * by other pools.  Once we're back down to a single
5476		 * disk+spare, we stop removing them.
5477		 */
5478		if (vd->vdev_children > 2) {
5479			newvd = vd->vdev_child[1];
5480
5481			if (newvd->vdev_isspare && last->vdev_isspare &&
5482			    vdev_dtl_empty(last, DTL_MISSING) &&
5483			    vdev_dtl_empty(last, DTL_OUTAGE) &&
5484			    !vdev_dtl_required(newvd))
5485				return (newvd);
5486		}
5487	}
5488
5489	return (NULL);
5490}
5491
5492static void
5493spa_vdev_resilver_done(spa_t *spa)
5494{
5495	vdev_t *vd, *pvd, *ppvd;
5496	uint64_t guid, sguid, pguid, ppguid;
5497
5498	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5499
5500	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5501		pvd = vd->vdev_parent;
5502		ppvd = pvd->vdev_parent;
5503		guid = vd->vdev_guid;
5504		pguid = pvd->vdev_guid;
5505		ppguid = ppvd->vdev_guid;
5506		sguid = 0;
5507		/*
5508		 * If we have just finished replacing a hot spared device, then
5509		 * we need to detach the parent's first child (the original hot
5510		 * spare) as well.
5511		 */
5512		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5513		    ppvd->vdev_children == 2) {
5514			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5515			sguid = ppvd->vdev_child[1]->vdev_guid;
5516		}
5517		ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5518
5519		spa_config_exit(spa, SCL_ALL, FTAG);
5520		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5521			return;
5522		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5523			return;
5524		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5525	}
5526
5527	spa_config_exit(spa, SCL_ALL, FTAG);
5528}
5529
5530/*
5531 * Update the stored path or FRU for this vdev.
5532 */
5533int
5534spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5535    boolean_t ispath)
5536{
5537	vdev_t *vd;
5538	boolean_t sync = B_FALSE;
5539
5540	ASSERT(spa_writeable(spa));
5541
5542	spa_vdev_state_enter(spa, SCL_ALL);
5543
5544	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5545		return (spa_vdev_state_exit(spa, NULL, ENOENT));
5546
5547	if (!vd->vdev_ops->vdev_op_leaf)
5548		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5549
5550	if (ispath) {
5551		if (strcmp(value, vd->vdev_path) != 0) {
5552			spa_strfree(vd->vdev_path);
5553			vd->vdev_path = spa_strdup(value);
5554			sync = B_TRUE;
5555		}
5556	} else {
5557		if (vd->vdev_fru == NULL) {
5558			vd->vdev_fru = spa_strdup(value);
5559			sync = B_TRUE;
5560		} else if (strcmp(value, vd->vdev_fru) != 0) {
5561			spa_strfree(vd->vdev_fru);
5562			vd->vdev_fru = spa_strdup(value);
5563			sync = B_TRUE;
5564		}
5565	}
5566
5567	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5568}
5569
5570int
5571spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5572{
5573	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5574}
5575
5576int
5577spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5578{
5579	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5580}
5581
5582/*
5583 * ==========================================================================
5584 * SPA Scanning
5585 * ==========================================================================
5586 */
5587
5588int
5589spa_scan_stop(spa_t *spa)
5590{
5591	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5592	if (dsl_scan_resilvering(spa->spa_dsl_pool))
5593		return (SET_ERROR(EBUSY));
5594	return (dsl_scan_cancel(spa->spa_dsl_pool));
5595}
5596
5597int
5598spa_scan(spa_t *spa, pool_scan_func_t func)
5599{
5600	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5601
5602	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5603		return (SET_ERROR(ENOTSUP));
5604
5605	/*
5606	 * If a resilver was requested, but there is no DTL on a
5607	 * writeable leaf device, we have nothing to do.
5608	 */
5609	if (func == POOL_SCAN_RESILVER &&
5610	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5611		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5612		return (0);
5613	}
5614
5615	return (dsl_scan(spa->spa_dsl_pool, func));
5616}
5617
5618/*
5619 * ==========================================================================
5620 * SPA async task processing
5621 * ==========================================================================
5622 */
5623
5624static void
5625spa_async_remove(spa_t *spa, vdev_t *vd)
5626{
5627	if (vd->vdev_remove_wanted) {
5628		vd->vdev_remove_wanted = B_FALSE;
5629		vd->vdev_delayed_close = B_FALSE;
5630		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5631
5632		/*
5633		 * We want to clear the stats, but we don't want to do a full
5634		 * vdev_clear() as that will cause us to throw away
5635		 * degraded/faulted state as well as attempt to reopen the
5636		 * device, all of which is a waste.
5637		 */
5638		vd->vdev_stat.vs_read_errors = 0;
5639		vd->vdev_stat.vs_write_errors = 0;
5640		vd->vdev_stat.vs_checksum_errors = 0;
5641
5642		vdev_state_dirty(vd->vdev_top);
5643	}
5644
5645	for (int c = 0; c < vd->vdev_children; c++)
5646		spa_async_remove(spa, vd->vdev_child[c]);
5647}
5648
5649static void
5650spa_async_probe(spa_t *spa, vdev_t *vd)
5651{
5652	if (vd->vdev_probe_wanted) {
5653		vd->vdev_probe_wanted = B_FALSE;
5654		vdev_reopen(vd);	/* vdev_open() does the actual probe */
5655	}
5656
5657	for (int c = 0; c < vd->vdev_children; c++)
5658		spa_async_probe(spa, vd->vdev_child[c]);
5659}
5660
5661static void
5662spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5663{
5664	sysevent_id_t eid;
5665	nvlist_t *attr;
5666	char *physpath;
5667
5668	if (!spa->spa_autoexpand)
5669		return;
5670
5671	for (int c = 0; c < vd->vdev_children; c++) {
5672		vdev_t *cvd = vd->vdev_child[c];
5673		spa_async_autoexpand(spa, cvd);
5674	}
5675
5676	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5677		return;
5678
5679	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5680	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5681
5682	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5683	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5684
5685	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5686	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5687
5688	nvlist_free(attr);
5689	kmem_free(physpath, MAXPATHLEN);
5690}
5691
5692static void
5693spa_async_thread(spa_t *spa)
5694{
5695	int tasks;
5696
5697	ASSERT(spa->spa_sync_on);
5698
5699	mutex_enter(&spa->spa_async_lock);
5700	tasks = spa->spa_async_tasks;
5701	spa->spa_async_tasks = 0;
5702	mutex_exit(&spa->spa_async_lock);
5703
5704	/*
5705	 * See if the config needs to be updated.
5706	 */
5707	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5708		uint64_t old_space, new_space;
5709
5710		mutex_enter(&spa_namespace_lock);
5711		old_space = metaslab_class_get_space(spa_normal_class(spa));
5712		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5713		new_space = metaslab_class_get_space(spa_normal_class(spa));
5714		mutex_exit(&spa_namespace_lock);
5715
5716		/*
5717		 * If the pool grew as a result of the config update,
5718		 * then log an internal history event.
5719		 */
5720		if (new_space != old_space) {
5721			spa_history_log_internal(spa, "vdev online", NULL,
5722			    "pool '%s' size: %llu(+%llu)",
5723			    spa_name(spa), new_space, new_space - old_space);
5724		}
5725	}
5726
5727	/*
5728	 * See if any devices need to be marked REMOVED.
5729	 */
5730	if (tasks & SPA_ASYNC_REMOVE) {
5731		spa_vdev_state_enter(spa, SCL_NONE);
5732		spa_async_remove(spa, spa->spa_root_vdev);
5733		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5734			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5735		for (int i = 0; i < spa->spa_spares.sav_count; i++)
5736			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5737		(void) spa_vdev_state_exit(spa, NULL, 0);
5738	}
5739
5740	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5741		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5742		spa_async_autoexpand(spa, spa->spa_root_vdev);
5743		spa_config_exit(spa, SCL_CONFIG, FTAG);
5744	}
5745
5746	/*
5747	 * See if any devices need to be probed.
5748	 */
5749	if (tasks & SPA_ASYNC_PROBE) {
5750		spa_vdev_state_enter(spa, SCL_NONE);
5751		spa_async_probe(spa, spa->spa_root_vdev);
5752		(void) spa_vdev_state_exit(spa, NULL, 0);
5753	}
5754
5755	/*
5756	 * If any devices are done replacing, detach them.
5757	 */
5758	if (tasks & SPA_ASYNC_RESILVER_DONE)
5759		spa_vdev_resilver_done(spa);
5760
5761	/*
5762	 * Kick off a resilver.
5763	 */
5764	if (tasks & SPA_ASYNC_RESILVER)
5765		dsl_resilver_restart(spa->spa_dsl_pool, 0);
5766
5767	/*
5768	 * Let the world know that we're done.
5769	 */
5770	mutex_enter(&spa->spa_async_lock);
5771	spa->spa_async_thread = NULL;
5772	cv_broadcast(&spa->spa_async_cv);
5773	mutex_exit(&spa->spa_async_lock);
5774	thread_exit();
5775}
5776
5777void
5778spa_async_suspend(spa_t *spa)
5779{
5780	mutex_enter(&spa->spa_async_lock);
5781	spa->spa_async_suspended++;
5782	while (spa->spa_async_thread != NULL)
5783		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5784	mutex_exit(&spa->spa_async_lock);
5785}
5786
5787void
5788spa_async_resume(spa_t *spa)
5789{
5790	mutex_enter(&spa->spa_async_lock);
5791	ASSERT(spa->spa_async_suspended != 0);
5792	spa->spa_async_suspended--;
5793	mutex_exit(&spa->spa_async_lock);
5794}
5795
5796static boolean_t
5797spa_async_tasks_pending(spa_t *spa)
5798{
5799	uint_t non_config_tasks;
5800	uint_t config_task;
5801	boolean_t config_task_suspended;
5802
5803	non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5804	config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5805	if (spa->spa_ccw_fail_time == 0) {
5806		config_task_suspended = B_FALSE;
5807	} else {
5808		config_task_suspended =
5809		    (gethrtime() - spa->spa_ccw_fail_time) <
5810		    (zfs_ccw_retry_interval * NANOSEC);
5811	}
5812
5813	return (non_config_tasks || (config_task && !config_task_suspended));
5814}
5815
5816static void
5817spa_async_dispatch(spa_t *spa)
5818{
5819	mutex_enter(&spa->spa_async_lock);
5820	if (spa_async_tasks_pending(spa) &&
5821	    !spa->spa_async_suspended &&
5822	    spa->spa_async_thread == NULL &&
5823	    rootdir != NULL)
5824		spa->spa_async_thread = thread_create(NULL, 0,
5825		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5826	mutex_exit(&spa->spa_async_lock);
5827}
5828
5829void
5830spa_async_request(spa_t *spa, int task)
5831{
5832	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5833	mutex_enter(&spa->spa_async_lock);
5834	spa->spa_async_tasks |= task;
5835	mutex_exit(&spa->spa_async_lock);
5836}
5837
5838/*
5839 * ==========================================================================
5840 * SPA syncing routines
5841 * ==========================================================================
5842 */
5843
5844static int
5845bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5846{
5847	bpobj_t *bpo = arg;
5848	bpobj_enqueue(bpo, bp, tx);
5849	return (0);
5850}
5851
5852static int
5853spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5854{
5855	zio_t *zio = arg;
5856
5857	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5858	    zio->io_flags));
5859	return (0);
5860}
5861
5862/*
5863 * Note: this simple function is not inlined to make it easier to dtrace the
5864 * amount of time spent syncing frees.
5865 */
5866static void
5867spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
5868{
5869	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5870	bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
5871	VERIFY(zio_wait(zio) == 0);
5872}
5873
5874/*
5875 * Note: this simple function is not inlined to make it easier to dtrace the
5876 * amount of time spent syncing deferred frees.
5877 */
5878static void
5879spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
5880{
5881	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5882	VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
5883	    spa_free_sync_cb, zio, tx), ==, 0);
5884	VERIFY0(zio_wait(zio));
5885}
5886
5887
5888static void
5889spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5890{
5891	char *packed = NULL;
5892	size_t bufsize;
5893	size_t nvsize = 0;
5894	dmu_buf_t *db;
5895
5896	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5897
5898	/*
5899	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5900	 * information.  This avoids the dmu_buf_will_dirty() path and
5901	 * saves us a pre-read to get data we don't actually care about.
5902	 */
5903	bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5904	packed = kmem_alloc(bufsize, KM_SLEEP);
5905
5906	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5907	    KM_SLEEP) == 0);
5908	bzero(packed + nvsize, bufsize - nvsize);
5909
5910	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5911
5912	kmem_free(packed, bufsize);
5913
5914	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5915	dmu_buf_will_dirty(db, tx);
5916	*(uint64_t *)db->db_data = nvsize;
5917	dmu_buf_rele(db, FTAG);
5918}
5919
5920static void
5921spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5922    const char *config, const char *entry)
5923{
5924	nvlist_t *nvroot;
5925	nvlist_t **list;
5926	int i;
5927
5928	if (!sav->sav_sync)
5929		return;
5930
5931	/*
5932	 * Update the MOS nvlist describing the list of available devices.
5933	 * spa_validate_aux() will have already made sure this nvlist is
5934	 * valid and the vdevs are labeled appropriately.
5935	 */
5936	if (sav->sav_object == 0) {
5937		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5938		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5939		    sizeof (uint64_t), tx);
5940		VERIFY(zap_update(spa->spa_meta_objset,
5941		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5942		    &sav->sav_object, tx) == 0);
5943	}
5944
5945	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5946	if (sav->sav_count == 0) {
5947		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5948	} else {
5949		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5950		for (i = 0; i < sav->sav_count; i++)
5951			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5952			    B_FALSE, VDEV_CONFIG_L2CACHE);
5953		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5954		    sav->sav_count) == 0);
5955		for (i = 0; i < sav->sav_count; i++)
5956			nvlist_free(list[i]);
5957		kmem_free(list, sav->sav_count * sizeof (void *));
5958	}
5959
5960	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5961	nvlist_free(nvroot);
5962
5963	sav->sav_sync = B_FALSE;
5964}
5965
5966static void
5967spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5968{
5969	nvlist_t *config;
5970
5971	if (list_is_empty(&spa->spa_config_dirty_list))
5972		return;
5973
5974	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5975
5976	config = spa_config_generate(spa, spa->spa_root_vdev,
5977	    dmu_tx_get_txg(tx), B_FALSE);
5978
5979	/*
5980	 * If we're upgrading the spa version then make sure that
5981	 * the config object gets updated with the correct version.
5982	 */
5983	if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5984		fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5985		    spa->spa_uberblock.ub_version);
5986
5987	spa_config_exit(spa, SCL_STATE, FTAG);
5988
5989	if (spa->spa_config_syncing)
5990		nvlist_free(spa->spa_config_syncing);
5991	spa->spa_config_syncing = config;
5992
5993	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5994}
5995
5996static void
5997spa_sync_version(void *arg, dmu_tx_t *tx)
5998{
5999	uint64_t *versionp = arg;
6000	uint64_t version = *versionp;
6001	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6002
6003	/*
6004	 * Setting the version is special cased when first creating the pool.
6005	 */
6006	ASSERT(tx->tx_txg != TXG_INITIAL);
6007
6008	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
6009	ASSERT(version >= spa_version(spa));
6010
6011	spa->spa_uberblock.ub_version = version;
6012	vdev_config_dirty(spa->spa_root_vdev);
6013	spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6014}
6015
6016/*
6017 * Set zpool properties.
6018 */
6019static void
6020spa_sync_props(void *arg, dmu_tx_t *tx)
6021{
6022	nvlist_t *nvp = arg;
6023	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6024	objset_t *mos = spa->spa_meta_objset;