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) 2013, 2014, 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	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1579	nvsize = *(uint64_t *)db->db_data;
1580	dmu_buf_rele(db, FTAG);
1581
1582	packed = kmem_alloc(nvsize, KM_SLEEP);
1583	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1584	    DMU_READ_PREFETCH);
1585	if (error == 0)
1586		error = nvlist_unpack(packed, nvsize, value, 0);
1587	kmem_free(packed, nvsize);
1588
1589	return (error);
1590}
1591
1592/*
1593 * Checks to see if the given vdev could not be opened, in which case we post a
1594 * sysevent to notify the autoreplace code that the device has been removed.
1595 */
1596static void
1597spa_check_removed(vdev_t *vd)
1598{
1599	for (int c = 0; c < vd->vdev_children; c++)
1600		spa_check_removed(vd->vdev_child[c]);
1601
1602	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1603	    !vd->vdev_ishole) {
1604		zfs_post_autoreplace(vd->vdev_spa, vd);
1605		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1606	}
1607}
1608
1609/*
1610 * Validate the current config against the MOS config
1611 */
1612static boolean_t
1613spa_config_valid(spa_t *spa, nvlist_t *config)
1614{
1615	vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1616	nvlist_t *nv;
1617
1618	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1619
1620	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1621	VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1622
1623	ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1624
1625	/*
1626	 * If we're doing a normal import, then build up any additional
1627	 * diagnostic information about missing devices in this config.
1628	 * We'll pass this up to the user for further processing.
1629	 */
1630	if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1631		nvlist_t **child, *nv;
1632		uint64_t idx = 0;
1633
1634		child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1635		    KM_SLEEP);
1636		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1637
1638		for (int c = 0; c < rvd->vdev_children; c++) {
1639			vdev_t *tvd = rvd->vdev_child[c];
1640			vdev_t *mtvd  = mrvd->vdev_child[c];
1641
1642			if (tvd->vdev_ops == &vdev_missing_ops &&
1643			    mtvd->vdev_ops != &vdev_missing_ops &&
1644			    mtvd->vdev_islog)
1645				child[idx++] = vdev_config_generate(spa, mtvd,
1646				    B_FALSE, 0);
1647		}
1648
1649		if (idx) {
1650			VERIFY(nvlist_add_nvlist_array(nv,
1651			    ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1652			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1653			    ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1654
1655			for (int i = 0; i < idx; i++)
1656				nvlist_free(child[i]);
1657		}
1658		nvlist_free(nv);
1659		kmem_free(child, rvd->vdev_children * sizeof (char **));
1660	}
1661
1662	/*
1663	 * Compare the root vdev tree with the information we have
1664	 * from the MOS config (mrvd). Check each top-level vdev
1665	 * with the corresponding MOS config top-level (mtvd).
1666	 */
1667	for (int c = 0; c < rvd->vdev_children; c++) {
1668		vdev_t *tvd = rvd->vdev_child[c];
1669		vdev_t *mtvd  = mrvd->vdev_child[c];
1670
1671		/*
1672		 * Resolve any "missing" vdevs in the current configuration.
1673		 * If we find that the MOS config has more accurate information
1674		 * about the top-level vdev then use that vdev instead.
1675		 */
1676		if (tvd->vdev_ops == &vdev_missing_ops &&
1677		    mtvd->vdev_ops != &vdev_missing_ops) {
1678
1679			if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1680				continue;
1681
1682			/*
1683			 * Device specific actions.
1684			 */
1685			if (mtvd->vdev_islog) {
1686				spa_set_log_state(spa, SPA_LOG_CLEAR);
1687			} else {
1688				/*
1689				 * XXX - once we have 'readonly' pool
1690				 * support we should be able to handle
1691				 * missing data devices by transitioning
1692				 * the pool to readonly.
1693				 */
1694				continue;
1695			}
1696
1697			/*
1698			 * Swap the missing vdev with the data we were
1699			 * able to obtain from the MOS config.
1700			 */
1701			vdev_remove_child(rvd, tvd);
1702			vdev_remove_child(mrvd, mtvd);
1703
1704			vdev_add_child(rvd, mtvd);
1705			vdev_add_child(mrvd, tvd);
1706
1707			spa_config_exit(spa, SCL_ALL, FTAG);
1708			vdev_load(mtvd);
1709			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1710
1711			vdev_reopen(rvd);
1712		} else if (mtvd->vdev_islog) {
1713			/*
1714			 * Load the slog device's state from the MOS config
1715			 * since it's possible that the label does not
1716			 * contain the most up-to-date information.
1717			 */
1718			vdev_load_log_state(tvd, mtvd);
1719			vdev_reopen(tvd);
1720		}
1721	}
1722	vdev_free(mrvd);
1723	spa_config_exit(spa, SCL_ALL, FTAG);
1724
1725	/*
1726	 * Ensure we were able to validate the config.
1727	 */
1728	return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1729}
1730
1731/*
1732 * Check for missing log devices
1733 */
1734static boolean_t
1735spa_check_logs(spa_t *spa)
1736{
1737	boolean_t rv = B_FALSE;
1738
1739	switch (spa->spa_log_state) {
1740	case SPA_LOG_MISSING:
1741		/* need to recheck in case slog has been restored */
1742	case SPA_LOG_UNKNOWN:
1743		rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1744		    NULL, DS_FIND_CHILDREN) != 0);
1745		if (rv)
1746			spa_set_log_state(spa, SPA_LOG_MISSING);
1747		break;
1748	}
1749	return (rv);
1750}
1751
1752static boolean_t
1753spa_passivate_log(spa_t *spa)
1754{
1755	vdev_t *rvd = spa->spa_root_vdev;
1756	boolean_t slog_found = B_FALSE;
1757
1758	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1759
1760	if (!spa_has_slogs(spa))
1761		return (B_FALSE);
1762
1763	for (int c = 0; c < rvd->vdev_children; c++) {
1764		vdev_t *tvd = rvd->vdev_child[c];
1765		metaslab_group_t *mg = tvd->vdev_mg;
1766
1767		if (tvd->vdev_islog) {
1768			metaslab_group_passivate(mg);
1769			slog_found = B_TRUE;
1770		}
1771	}
1772
1773	return (slog_found);
1774}
1775
1776static void
1777spa_activate_log(spa_t *spa)
1778{
1779	vdev_t *rvd = spa->spa_root_vdev;
1780
1781	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1782
1783	for (int c = 0; c < rvd->vdev_children; c++) {
1784		vdev_t *tvd = rvd->vdev_child[c];
1785		metaslab_group_t *mg = tvd->vdev_mg;
1786
1787		if (tvd->vdev_islog)
1788			metaslab_group_activate(mg);
1789	}
1790}
1791
1792int
1793spa_offline_log(spa_t *spa)
1794{
1795	int error;
1796
1797	error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1798	    NULL, DS_FIND_CHILDREN);
1799	if (error == 0) {
1800		/*
1801		 * We successfully offlined the log device, sync out the
1802		 * current txg so that the "stubby" block can be removed
1803		 * by zil_sync().
1804		 */
1805		txg_wait_synced(spa->spa_dsl_pool, 0);
1806	}
1807	return (error);
1808}
1809
1810static void
1811spa_aux_check_removed(spa_aux_vdev_t *sav)
1812{
1813	for (int i = 0; i < sav->sav_count; i++)
1814		spa_check_removed(sav->sav_vdevs[i]);
1815}
1816
1817void
1818spa_claim_notify(zio_t *zio)
1819{
1820	spa_t *spa = zio->io_spa;
1821
1822	if (zio->io_error)
1823		return;
1824
1825	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1826	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1827		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1828	mutex_exit(&spa->spa_props_lock);
1829}
1830
1831typedef struct spa_load_error {
1832	uint64_t	sle_meta_count;
1833	uint64_t	sle_data_count;
1834} spa_load_error_t;
1835
1836static void
1837spa_load_verify_done(zio_t *zio)
1838{
1839	blkptr_t *bp = zio->io_bp;
1840	spa_load_error_t *sle = zio->io_private;
1841	dmu_object_type_t type = BP_GET_TYPE(bp);
1842	int error = zio->io_error;
1843	spa_t *spa = zio->io_spa;
1844
1845	if (error) {
1846		if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1847		    type != DMU_OT_INTENT_LOG)
1848			atomic_inc_64(&sle->sle_meta_count);
1849		else
1850			atomic_inc_64(&sle->sle_data_count);
1851	}
1852	zio_data_buf_free(zio->io_data, zio->io_size);
1853
1854	mutex_enter(&spa->spa_scrub_lock);
1855	spa->spa_scrub_inflight--;
1856	cv_broadcast(&spa->spa_scrub_io_cv);
1857	mutex_exit(&spa->spa_scrub_lock);
1858}
1859
1860/*
1861 * Maximum number of concurrent scrub i/os to create while verifying
1862 * a pool while importing it.
1863 */
1864int spa_load_verify_maxinflight = 10000;
1865boolean_t spa_load_verify_metadata = B_TRUE;
1866boolean_t spa_load_verify_data = B_TRUE;
1867
1868/*ARGSUSED*/
1869static int
1870spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1871    const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1872{
1873	if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
1874		return (0);
1875	/*
1876	 * Note: normally this routine will not be called if
1877	 * spa_load_verify_metadata is not set.  However, it may be useful
1878	 * to manually set the flag after the traversal has begun.
1879	 */
1880	if (!spa_load_verify_metadata)
1881		return (0);
1882	if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data)
1883		return (0);
1884
1885	zio_t *rio = arg;
1886	size_t size = BP_GET_PSIZE(bp);
1887	void *data = zio_data_buf_alloc(size);
1888
1889	mutex_enter(&spa->spa_scrub_lock);
1890	while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
1891		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1892	spa->spa_scrub_inflight++;
1893	mutex_exit(&spa->spa_scrub_lock);
1894
1895	zio_nowait(zio_read(rio, spa, bp, data, size,
1896	    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1897	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1898	    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1899	return (0);
1900}
1901
1902static int
1903spa_load_verify(spa_t *spa)
1904{
1905	zio_t *rio;
1906	spa_load_error_t sle = { 0 };
1907	zpool_rewind_policy_t policy;
1908	boolean_t verify_ok = B_FALSE;
1909	int error = 0;
1910
1911	zpool_get_rewind_policy(spa->spa_config, &policy);
1912
1913	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1914		return (0);
1915
1916	rio = zio_root(spa, NULL, &sle,
1917	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1918
1919	if (spa_load_verify_metadata) {
1920		error = traverse_pool(spa, spa->spa_verify_min_txg,
1921		    TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
1922		    spa_load_verify_cb, rio);
1923	}
1924
1925	(void) zio_wait(rio);
1926
1927	spa->spa_load_meta_errors = sle.sle_meta_count;
1928	spa->spa_load_data_errors = sle.sle_data_count;
1929
1930	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1931	    sle.sle_data_count <= policy.zrp_maxdata) {
1932		int64_t loss = 0;
1933
1934		verify_ok = B_TRUE;
1935		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1936		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1937
1938		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1939		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1940		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1941		VERIFY(nvlist_add_int64(spa->spa_load_info,
1942		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1943		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1944		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1945	} else {
1946		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1947	}
1948
1949	if (error) {
1950		if (error != ENXIO && error != EIO)
1951			error = SET_ERROR(EIO);
1952		return (error);
1953	}
1954
1955	return (verify_ok ? 0 : EIO);
1956}
1957
1958/*
1959 * Find a value in the pool props object.
1960 */
1961static void
1962spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1963{
1964	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1965	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1966}
1967
1968/*
1969 * Find a value in the pool directory object.
1970 */
1971static int
1972spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1973{
1974	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1975	    name, sizeof (uint64_t), 1, val));
1976}
1977
1978static int
1979spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1980{
1981	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1982	return (err);
1983}
1984
1985/*
1986 * Fix up config after a partly-completed split.  This is done with the
1987 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1988 * pool have that entry in their config, but only the splitting one contains
1989 * a list of all the guids of the vdevs that are being split off.
1990 *
1991 * This function determines what to do with that list: either rejoin
1992 * all the disks to the pool, or complete the splitting process.  To attempt
1993 * the rejoin, each disk that is offlined is marked online again, and
1994 * we do a reopen() call.  If the vdev label for every disk that was
1995 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1996 * then we call vdev_split() on each disk, and complete the split.
1997 *
1998 * Otherwise we leave the config alone, with all the vdevs in place in
1999 * the original pool.
2000 */
2001static void
2002spa_try_repair(spa_t *spa, nvlist_t *config)
2003{
2004	uint_t extracted;
2005	uint64_t *glist;
2006	uint_t i, gcount;
2007	nvlist_t *nvl;
2008	vdev_t **vd;
2009	boolean_t attempt_reopen;
2010
2011	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2012		return;
2013
2014	/* check that the config is complete */
2015	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2016	    &glist, &gcount) != 0)
2017		return;
2018
2019	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2020
2021	/* attempt to online all the vdevs & validate */
2022	attempt_reopen = B_TRUE;
2023	for (i = 0; i < gcount; i++) {
2024		if (glist[i] == 0)	/* vdev is hole */
2025			continue;
2026
2027		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2028		if (vd[i] == NULL) {
2029			/*
2030			 * Don't bother attempting to reopen the disks;
2031			 * just do the split.
2032			 */
2033			attempt_reopen = B_FALSE;
2034		} else {
2035			/* attempt to re-online it */
2036			vd[i]->vdev_offline = B_FALSE;
2037		}
2038	}
2039
2040	if (attempt_reopen) {
2041		vdev_reopen(spa->spa_root_vdev);
2042
2043		/* check each device to see what state it's in */
2044		for (extracted = 0, i = 0; i < gcount; i++) {
2045			if (vd[i] != NULL &&
2046			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2047				break;
2048			++extracted;
2049		}
2050	}
2051
2052	/*
2053	 * If every disk has been moved to the new pool, or if we never
2054	 * even attempted to look at them, then we split them off for
2055	 * good.
2056	 */
2057	if (!attempt_reopen || gcount == extracted) {
2058		for (i = 0; i < gcount; i++)
2059			if (vd[i] != NULL)
2060				vdev_split(vd[i]);
2061		vdev_reopen(spa->spa_root_vdev);
2062	}
2063
2064	kmem_free(vd, gcount * sizeof (vdev_t *));
2065}
2066
2067static int
2068spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2069    boolean_t mosconfig)
2070{
2071	nvlist_t *config = spa->spa_config;
2072	char *ereport = FM_EREPORT_ZFS_POOL;
2073	char *comment;
2074	int error;
2075	uint64_t pool_guid;
2076	nvlist_t *nvl;
2077
2078	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2079		return (SET_ERROR(EINVAL));
2080
2081	ASSERT(spa->spa_comment == NULL);
2082	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2083		spa->spa_comment = spa_strdup(comment);
2084
2085	/*
2086	 * Versioning wasn't explicitly added to the label until later, so if
2087	 * it's not present treat it as the initial version.
2088	 */
2089	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2090	    &spa->spa_ubsync.ub_version) != 0)
2091		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2092
2093	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2094	    &spa->spa_config_txg);
2095
2096	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2097	    spa_guid_exists(pool_guid, 0)) {
2098		error = SET_ERROR(EEXIST);
2099	} else {
2100		spa->spa_config_guid = pool_guid;
2101
2102		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2103		    &nvl) == 0) {
2104			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2105			    KM_SLEEP) == 0);
2106		}
2107
2108		nvlist_free(spa->spa_load_info);
2109		spa->spa_load_info = fnvlist_alloc();
2110
2111		gethrestime(&spa->spa_loaded_ts);
2112		error = spa_load_impl(spa, pool_guid, config, state, type,
2113		    mosconfig, &ereport);
2114	}
2115
2116	/*
2117	 * Don't count references from objsets that are already closed
2118	 * and are making their way through the eviction process.
2119	 */
2120	spa_evicting_os_wait(spa);
2121	spa->spa_minref = refcount_count(&spa->spa_refcount);
2122	if (error) {
2123		if (error != EEXIST) {
2124			spa->spa_loaded_ts.tv_sec = 0;
2125			spa->spa_loaded_ts.tv_nsec = 0;
2126		}
2127		if (error != EBADF) {
2128			zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2129		}
2130	}
2131	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2132	spa->spa_ena = 0;
2133
2134	return (error);
2135}
2136
2137/*
2138 * Load an existing storage pool, using the pool's builtin spa_config as a
2139 * source of configuration information.
2140 */
2141static int
2142spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2143    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2144    char **ereport)
2145{
2146	int error = 0;
2147	nvlist_t *nvroot = NULL;
2148	nvlist_t *label;
2149	vdev_t *rvd;
2150	uberblock_t *ub = &spa->spa_uberblock;
2151	uint64_t children, config_cache_txg = spa->spa_config_txg;
2152	int orig_mode = spa->spa_mode;
2153	int parse;
2154	uint64_t obj;
2155	boolean_t missing_feat_write = B_FALSE;
2156
2157	/*
2158	 * If this is an untrusted config, access the pool in read-only mode.
2159	 * This prevents things like resilvering recently removed devices.
2160	 */
2161	if (!mosconfig)
2162		spa->spa_mode = FREAD;
2163
2164	ASSERT(MUTEX_HELD(&spa_namespace_lock));
2165
2166	spa->spa_load_state = state;
2167
2168	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2169		return (SET_ERROR(EINVAL));
2170
2171	parse = (type == SPA_IMPORT_EXISTING ?
2172	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2173
2174	/*
2175	 * Create "The Godfather" zio to hold all async IOs
2176	 */
2177	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2178	    KM_SLEEP);
2179	for (int i = 0; i < max_ncpus; i++) {
2180		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2181		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2182		    ZIO_FLAG_GODFATHER);
2183	}
2184
2185	/*
2186	 * Parse the configuration into a vdev tree.  We explicitly set the
2187	 * value that will be returned by spa_version() since parsing the
2188	 * configuration requires knowing the version number.
2189	 */
2190	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2191	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2192	spa_config_exit(spa, SCL_ALL, FTAG);
2193
2194	if (error != 0)
2195		return (error);
2196
2197	ASSERT(spa->spa_root_vdev == rvd);
2198
2199	if (type != SPA_IMPORT_ASSEMBLE) {
2200		ASSERT(spa_guid(spa) == pool_guid);
2201	}
2202
2203	/*
2204	 * Try to open all vdevs, loading each label in the process.
2205	 */
2206	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2207	error = vdev_open(rvd);
2208	spa_config_exit(spa, SCL_ALL, FTAG);
2209	if (error != 0)
2210		return (error);
2211
2212	/*
2213	 * We need to validate the vdev labels against the configuration that
2214	 * we have in hand, which is dependent on the setting of mosconfig. If
2215	 * mosconfig is true then we're validating the vdev labels based on
2216	 * that config.  Otherwise, we're validating against the cached config
2217	 * (zpool.cache) that was read when we loaded the zfs module, and then
2218	 * later we will recursively call spa_load() and validate against
2219	 * the vdev config.
2220	 *
2221	 * If we're assembling a new pool that's been split off from an
2222	 * existing pool, the labels haven't yet been updated so we skip
2223	 * validation for now.
2224	 */
2225	if (type != SPA_IMPORT_ASSEMBLE) {
2226		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2227		error = vdev_validate(rvd, mosconfig);
2228		spa_config_exit(spa, SCL_ALL, FTAG);
2229
2230		if (error != 0)
2231			return (error);
2232
2233		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2234			return (SET_ERROR(ENXIO));
2235	}
2236
2237	/*
2238	 * Find the best uberblock.
2239	 */
2240	vdev_uberblock_load(rvd, ub, &label);
2241
2242	/*
2243	 * If we weren't able to find a single valid uberblock, return failure.
2244	 */
2245	if (ub->ub_txg == 0) {
2246		nvlist_free(label);
2247		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2248	}
2249
2250	/*
2251	 * If the pool has an unsupported version we can't open it.
2252	 */
2253	if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2254		nvlist_free(label);
2255		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2256	}
2257
2258	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2259		nvlist_t *features;
2260
2261		/*
2262		 * If we weren't able to find what's necessary for reading the
2263		 * MOS in the label, return failure.
2264		 */
2265		if (label == NULL || nvlist_lookup_nvlist(label,
2266		    ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2267			nvlist_free(label);
2268			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2269			    ENXIO));
2270		}
2271
2272		/*
2273		 * Update our in-core representation with the definitive values
2274		 * from the label.
2275		 */
2276		nvlist_free(spa->spa_label_features);
2277		VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2278	}
2279
2280	nvlist_free(label);
2281
2282	/*
2283	 * Look through entries in the label nvlist's features_for_read. If
2284	 * there is a feature listed there which we don't understand then we
2285	 * cannot open a pool.
2286	 */
2287	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2288		nvlist_t *unsup_feat;
2289
2290		VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2291		    0);
2292
2293		for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2294		    NULL); nvp != NULL;
2295		    nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2296			if (!zfeature_is_supported(nvpair_name(nvp))) {
2297				VERIFY(nvlist_add_string(unsup_feat,
2298				    nvpair_name(nvp), "") == 0);
2299			}
2300		}
2301
2302		if (!nvlist_empty(unsup_feat)) {
2303			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2304			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2305			nvlist_free(unsup_feat);
2306			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2307			    ENOTSUP));
2308		}
2309
2310		nvlist_free(unsup_feat);
2311	}
2312
2313	/*
2314	 * If the vdev guid sum doesn't match the uberblock, we have an
2315	 * incomplete configuration.  We first check to see if the pool
2316	 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2317	 * If it is, defer the vdev_guid_sum check till later so we
2318	 * can handle missing vdevs.
2319	 */
2320	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2321	    &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2322	    rvd->vdev_guid_sum != ub->ub_guid_sum)
2323		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2324
2325	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2326		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2327		spa_try_repair(spa, config);
2328		spa_config_exit(spa, SCL_ALL, FTAG);
2329		nvlist_free(spa->spa_config_splitting);
2330		spa->spa_config_splitting = NULL;
2331	}
2332
2333	/*
2334	 * Initialize internal SPA structures.
2335	 */
2336	spa->spa_state = POOL_STATE_ACTIVE;
2337	spa->spa_ubsync = spa->spa_uberblock;
2338	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2339	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2340	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2341	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2342	spa->spa_claim_max_txg = spa->spa_first_txg;
2343	spa->spa_prev_software_version = ub->ub_software_version;
2344
2345	error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2346	if (error)
2347		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2348	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2349
2350	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2351		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2352
2353	if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2354		boolean_t missing_feat_read = B_FALSE;
2355		nvlist_t *unsup_feat, *enabled_feat;
2356
2357		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2358		    &spa->spa_feat_for_read_obj) != 0) {
2359			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2360		}
2361
2362		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2363		    &spa->spa_feat_for_write_obj) != 0) {
2364			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2365		}
2366
2367		if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2368		    &spa->spa_feat_desc_obj) != 0) {
2369			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2370		}
2371
2372		enabled_feat = fnvlist_alloc();
2373		unsup_feat = fnvlist_alloc();
2374
2375		if (!spa_features_check(spa, B_FALSE,
2376		    unsup_feat, enabled_feat))
2377			missing_feat_read = B_TRUE;
2378
2379		if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2380			if (!spa_features_check(spa, B_TRUE,
2381			    unsup_feat, enabled_feat)) {
2382				missing_feat_write = B_TRUE;
2383			}
2384		}
2385
2386		fnvlist_add_nvlist(spa->spa_load_info,
2387		    ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2388
2389		if (!nvlist_empty(unsup_feat)) {
2390			fnvlist_add_nvlist(spa->spa_load_info,
2391			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2392		}
2393
2394		fnvlist_free(enabled_feat);
2395		fnvlist_free(unsup_feat);
2396
2397		if (!missing_feat_read) {
2398			fnvlist_add_boolean(spa->spa_load_info,
2399			    ZPOOL_CONFIG_CAN_RDONLY);
2400		}
2401
2402		/*
2403		 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2404		 * twofold: to determine whether the pool is available for
2405		 * import in read-write mode and (if it is not) whether the
2406		 * pool is available for import in read-only mode. If the pool
2407		 * is available for import in read-write mode, it is displayed
2408		 * as available in userland; if it is not available for import
2409		 * in read-only mode, it is displayed as unavailable in
2410		 * userland. If the pool is available for import in read-only
2411		 * mode but not read-write mode, it is displayed as unavailable
2412		 * in userland with a special note that the pool is actually
2413		 * available for open in read-only mode.
2414		 *
2415		 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2416		 * missing a feature for write, we must first determine whether
2417		 * the pool can be opened read-only before returning to
2418		 * userland in order to know whether to display the
2419		 * abovementioned note.
2420		 */
2421		if (missing_feat_read || (missing_feat_write &&
2422		    spa_writeable(spa))) {
2423			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2424			    ENOTSUP));
2425		}
2426
2427		/*
2428		 * Load refcounts for ZFS features from disk into an in-memory
2429		 * cache during SPA initialization.
2430		 */
2431		for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2432			uint64_t refcount;
2433
2434			error = feature_get_refcount_from_disk(spa,
2435			    &spa_feature_table[i], &refcount);
2436			if (error == 0) {
2437				spa->spa_feat_refcount_cache[i] = refcount;
2438			} else if (error == ENOTSUP) {
2439				spa->spa_feat_refcount_cache[i] =
2440				    SPA_FEATURE_DISABLED;
2441			} else {
2442				return (spa_vdev_err(rvd,
2443				    VDEV_AUX_CORRUPT_DATA, EIO));
2444			}
2445		}
2446	}
2447
2448	if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2449		if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2450		    &spa->spa_feat_enabled_txg_obj) != 0)
2451			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2452	}
2453
2454	spa->spa_is_initializing = B_TRUE;
2455	error = dsl_pool_open(spa->spa_dsl_pool);
2456	spa->spa_is_initializing = B_FALSE;
2457	if (error != 0)
2458		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2459
2460	if (!mosconfig) {
2461		uint64_t hostid;
2462		nvlist_t *policy = NULL, *nvconfig;
2463
2464		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2465			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2466
2467		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2468		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2469			char *hostname;
2470			unsigned long myhostid = 0;
2471
2472			VERIFY(nvlist_lookup_string(nvconfig,
2473			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2474
2475#ifdef	_KERNEL
2476			myhostid = zone_get_hostid(NULL);
2477#else	/* _KERNEL */
2478			/*
2479			 * We're emulating the system's hostid in userland, so
2480			 * we can't use zone_get_hostid().
2481			 */
2482			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2483#endif	/* _KERNEL */
2484			if (hostid != 0 && myhostid != 0 &&
2485			    hostid != myhostid) {
2486				nvlist_free(nvconfig);
2487				cmn_err(CE_WARN, "pool '%s' could not be "
2488				    "loaded as it was last accessed by "
2489				    "another system (host: %s hostid: 0x%lx). "
2490				    "See: http://illumos.org/msg/ZFS-8000-EY",
2491				    spa_name(spa), hostname,
2492				    (unsigned long)hostid);
2493				return (SET_ERROR(EBADF));
2494			}
2495		}
2496		if (nvlist_lookup_nvlist(spa->spa_config,
2497		    ZPOOL_REWIND_POLICY, &policy) == 0)
2498			VERIFY(nvlist_add_nvlist(nvconfig,
2499			    ZPOOL_REWIND_POLICY, policy) == 0);
2500
2501		spa_config_set(spa, nvconfig);
2502		spa_unload(spa);
2503		spa_deactivate(spa);
2504		spa_activate(spa, orig_mode);
2505
2506		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2507	}
2508
2509	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2510		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2511	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2512	if (error != 0)
2513		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2514
2515	/*
2516	 * Load the bit that tells us to use the new accounting function
2517	 * (raid-z deflation).  If we have an older pool, this will not
2518	 * be present.
2519	 */
2520	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2521	if (error != 0 && error != ENOENT)
2522		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2523
2524	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2525	    &spa->spa_creation_version);
2526	if (error != 0 && error != ENOENT)
2527		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2528
2529	/*
2530	 * Load the persistent error log.  If we have an older pool, this will
2531	 * not be present.
2532	 */
2533	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2534	if (error != 0 && error != ENOENT)
2535		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2536
2537	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2538	    &spa->spa_errlog_scrub);
2539	if (error != 0 && error != ENOENT)
2540		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2541
2542	/*
2543	 * Load the history object.  If we have an older pool, this
2544	 * will not be present.
2545	 */
2546	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2547	if (error != 0 && error != ENOENT)
2548		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2549
2550	/*
2551	 * If we're assembling the pool from the split-off vdevs of
2552	 * an existing pool, we don't want to attach the spares & cache
2553	 * devices.
2554	 */
2555
2556	/*
2557	 * Load any hot spares for this pool.
2558	 */
2559	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2560	if (error != 0 && error != ENOENT)
2561		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2562	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2563		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2564		if (load_nvlist(spa, spa->spa_spares.sav_object,
2565		    &spa->spa_spares.sav_config) != 0)
2566			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2567
2568		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2569		spa_load_spares(spa);
2570		spa_config_exit(spa, SCL_ALL, FTAG);
2571	} else if (error == 0) {
2572		spa->spa_spares.sav_sync = B_TRUE;
2573	}
2574
2575	/*
2576	 * Load any level 2 ARC devices for this pool.
2577	 */
2578	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2579	    &spa->spa_l2cache.sav_object);
2580	if (error != 0 && error != ENOENT)
2581		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2582	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2583		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2584		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2585		    &spa->spa_l2cache.sav_config) != 0)
2586			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2587
2588		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2589		spa_load_l2cache(spa);
2590		spa_config_exit(spa, SCL_ALL, FTAG);
2591	} else if (error == 0) {
2592		spa->spa_l2cache.sav_sync = B_TRUE;
2593	}
2594
2595	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2596
2597	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2598	if (error && error != ENOENT)
2599		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2600
2601	if (error == 0) {
2602		uint64_t autoreplace;
2603
2604		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2605		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2606		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2607		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2608		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2609		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2610		    &spa->spa_dedup_ditto);
2611
2612		spa->spa_autoreplace = (autoreplace != 0);
2613	}
2614
2615	/*
2616	 * If the 'autoreplace' property is set, then post a resource notifying
2617	 * the ZFS DE that it should not issue any faults for unopenable
2618	 * devices.  We also iterate over the vdevs, and post a sysevent for any
2619	 * unopenable vdevs so that the normal autoreplace handler can take
2620	 * over.
2621	 */
2622	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2623		spa_check_removed(spa->spa_root_vdev);
2624		/*
2625		 * For the import case, this is done in spa_import(), because
2626		 * at this point we're using the spare definitions from
2627		 * the MOS config, not necessarily from the userland config.
2628		 */
2629		if (state != SPA_LOAD_IMPORT) {
2630			spa_aux_check_removed(&spa->spa_spares);
2631			spa_aux_check_removed(&spa->spa_l2cache);
2632		}
2633	}
2634
2635	/*
2636	 * Load the vdev state for all toplevel vdevs.
2637	 */
2638	vdev_load(rvd);
2639
2640	/*
2641	 * Propagate the leaf DTLs we just loaded all the way up the tree.
2642	 */
2643	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2644	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2645	spa_config_exit(spa, SCL_ALL, FTAG);
2646
2647	/*
2648	 * Load the DDTs (dedup tables).
2649	 */
2650	error = ddt_load(spa);
2651	if (error != 0)
2652		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2653
2654	spa_update_dspace(spa);
2655
2656	/*
2657	 * Validate the config, using the MOS config to fill in any
2658	 * information which might be missing.  If we fail to validate
2659	 * the config then declare the pool unfit for use. If we're
2660	 * assembling a pool from a split, the log is not transferred
2661	 * over.
2662	 */
2663	if (type != SPA_IMPORT_ASSEMBLE) {
2664		nvlist_t *nvconfig;
2665
2666		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2667			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2668
2669		if (!spa_config_valid(spa, nvconfig)) {
2670			nvlist_free(nvconfig);
2671			return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2672			    ENXIO));
2673		}
2674		nvlist_free(nvconfig);
2675
2676		/*
2677		 * Now that we've validated the config, check the state of the
2678		 * root vdev.  If it can't be opened, it indicates one or
2679		 * more toplevel vdevs are faulted.
2680		 */
2681		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2682			return (SET_ERROR(ENXIO));
2683
2684		if (spa_check_logs(spa)) {
2685			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2686			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2687		}
2688	}
2689
2690	if (missing_feat_write) {
2691		ASSERT(state == SPA_LOAD_TRYIMPORT);
2692
2693		/*
2694		 * At this point, we know that we can open the pool in
2695		 * read-only mode but not read-write mode. We now have enough
2696		 * information and can return to userland.
2697		 */
2698		return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2699	}
2700
2701	/*
2702	 * We've successfully opened the pool, verify that we're ready
2703	 * to start pushing transactions.
2704	 */
2705	if (state != SPA_LOAD_TRYIMPORT) {
2706		if (error = spa_load_verify(spa))
2707			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2708			    error));
2709	}
2710
2711	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2712	    spa->spa_load_max_txg == UINT64_MAX)) {
2713		dmu_tx_t *tx;
2714		int need_update = B_FALSE;
2715
2716		ASSERT(state != SPA_LOAD_TRYIMPORT);
2717
2718		/*
2719		 * Claim log blocks that haven't been committed yet.
2720		 * This must all happen in a single txg.
2721		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2722		 * invoked from zil_claim_log_block()'s i/o done callback.
2723		 * Price of rollback is that we abandon the log.
2724		 */
2725		spa->spa_claiming = B_TRUE;
2726
2727		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2728		    spa_first_txg(spa));
2729		(void) dmu_objset_find(spa_name(spa),
2730		    zil_claim, tx, DS_FIND_CHILDREN);
2731		dmu_tx_commit(tx);
2732
2733		spa->spa_claiming = B_FALSE;
2734
2735		spa_set_log_state(spa, SPA_LOG_GOOD);
2736		spa->spa_sync_on = B_TRUE;
2737		txg_sync_start(spa->spa_dsl_pool);
2738
2739		/*
2740		 * Wait for all claims to sync.  We sync up to the highest
2741		 * claimed log block birth time so that claimed log blocks
2742		 * don't appear to be from the future.  spa_claim_max_txg
2743		 * will have been set for us by either zil_check_log_chain()
2744		 * (invoked from spa_check_logs()) or zil_claim() above.
2745		 */
2746		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2747
2748		/*
2749		 * If the config cache is stale, or we have uninitialized
2750		 * metaslabs (see spa_vdev_add()), then update the config.
2751		 *
2752		 * If this is a verbatim import, trust the current
2753		 * in-core spa_config and update the disk labels.
2754		 */
2755		if (config_cache_txg != spa->spa_config_txg ||
2756		    state == SPA_LOAD_IMPORT ||
2757		    state == SPA_LOAD_RECOVER ||
2758		    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2759			need_update = B_TRUE;
2760
2761		for (int c = 0; c < rvd->vdev_children; c++)
2762			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2763				need_update = B_TRUE;
2764
2765		/*
2766		 * Update the config cache asychronously in case we're the
2767		 * root pool, in which case the config cache isn't writable yet.
2768		 */
2769		if (need_update)
2770			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2771
2772		/*
2773		 * Check all DTLs to see if anything needs resilvering.
2774		 */
2775		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2776		    vdev_resilver_needed(rvd, NULL, NULL))
2777			spa_async_request(spa, SPA_ASYNC_RESILVER);
2778
2779		/*
2780		 * Log the fact that we booted up (so that we can detect if
2781		 * we rebooted in the middle of an operation).
2782		 */
2783		spa_history_log_version(spa, "open");
2784
2785		/*
2786		 * Delete any inconsistent datasets.
2787		 */
2788		(void) dmu_objset_find(spa_name(spa),
2789		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2790
2791		/*
2792		 * Clean up any stale temporary dataset userrefs.
2793		 */
2794		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2795	}
2796
2797	return (0);
2798}
2799
2800static int
2801spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2802{
2803	int mode = spa->spa_mode;
2804
2805	spa_unload(spa);
2806	spa_deactivate(spa);
2807
2808	spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
2809
2810	spa_activate(spa, mode);
2811	spa_async_suspend(spa);
2812
2813	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2814}
2815
2816/*
2817 * If spa_load() fails this function will try loading prior txg's. If
2818 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2819 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2820 * function will not rewind the pool and will return the same error as
2821 * spa_load().
2822 */
2823static int
2824spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2825    uint64_t max_request, int rewind_flags)
2826{
2827	nvlist_t *loadinfo = NULL;
2828	nvlist_t *config = NULL;
2829	int load_error, rewind_error;
2830	uint64_t safe_rewind_txg;
2831	uint64_t min_txg;
2832
2833	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2834		spa->spa_load_max_txg = spa->spa_load_txg;
2835		spa_set_log_state(spa, SPA_LOG_CLEAR);
2836	} else {
2837		spa->spa_load_max_txg = max_request;
2838		if (max_request != UINT64_MAX)
2839			spa->spa_extreme_rewind = B_TRUE;
2840	}
2841
2842	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2843	    mosconfig);
2844	if (load_error == 0)
2845		return (0);
2846
2847	if (spa->spa_root_vdev != NULL)
2848		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2849
2850	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2851	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2852
2853	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2854		nvlist_free(config);
2855		return (load_error);
2856	}
2857
2858	if (state == SPA_LOAD_RECOVER) {
2859		/* Price of rolling back is discarding txgs, including log */
2860		spa_set_log_state(spa, SPA_LOG_CLEAR);
2861	} else {
2862		/*
2863		 * If we aren't rolling back save the load info from our first
2864		 * import attempt so that we can restore it after attempting
2865		 * to rewind.
2866		 */
2867		loadinfo = spa->spa_load_info;
2868		spa->spa_load_info = fnvlist_alloc();
2869	}
2870
2871	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2872	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2873	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2874	    TXG_INITIAL : safe_rewind_txg;
2875
2876	/*
2877	 * Continue as long as we're finding errors, we're still within
2878	 * the acceptable rewind range, and we're still finding uberblocks
2879	 */
2880	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2881	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2882		if (spa->spa_load_max_txg < safe_rewind_txg)
2883			spa->spa_extreme_rewind = B_TRUE;
2884		rewind_error = spa_load_retry(spa, state, mosconfig);
2885	}
2886
2887	spa->spa_extreme_rewind = B_FALSE;
2888	spa->spa_load_max_txg = UINT64_MAX;
2889
2890	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2891		spa_config_set(spa, config);
2892
2893	if (state == SPA_LOAD_RECOVER) {
2894		ASSERT3P(loadinfo, ==, NULL);
2895		return (rewind_error);
2896	} else {
2897		/* Store the rewind info as part of the initial load info */
2898		fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2899		    spa->spa_load_info);
2900
2901		/* Restore the initial load info */
2902		fnvlist_free(spa->spa_load_info);
2903		spa->spa_load_info = loadinfo;
2904
2905		return (load_error);
2906	}
2907}
2908
2909/*
2910 * Pool Open/Import
2911 *
2912 * The import case is identical to an open except that the configuration is sent
2913 * down from userland, instead of grabbed from the configuration cache.  For the
2914 * case of an open, the pool configuration will exist in the
2915 * POOL_STATE_UNINITIALIZED state.
2916 *
2917 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2918 * the same time open the pool, without having to keep around the spa_t in some
2919 * ambiguous state.
2920 */
2921static int
2922spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2923    nvlist_t **config)
2924{
2925	spa_t *spa;
2926	spa_load_state_t state = SPA_LOAD_OPEN;
2927	int error;
2928	int locked = B_FALSE;
2929
2930	*spapp = NULL;
2931
2932	/*
2933	 * As disgusting as this is, we need to support recursive calls to this
2934	 * function because dsl_dir_open() is called during spa_load(), and ends
2935	 * up calling spa_open() again.  The real fix is to figure out how to
2936	 * avoid dsl_dir_open() calling this in the first place.
2937	 */
2938	if (mutex_owner(&spa_namespace_lock) != curthread) {
2939		mutex_enter(&spa_namespace_lock);
2940		locked = B_TRUE;
2941	}
2942
2943	if ((spa = spa_lookup(pool)) == NULL) {
2944		if (locked)
2945			mutex_exit(&spa_namespace_lock);
2946		return (SET_ERROR(ENOENT));
2947	}
2948
2949	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2950		zpool_rewind_policy_t policy;
2951
2952		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2953		    &policy);
2954		if (policy.zrp_request & ZPOOL_DO_REWIND)
2955			state = SPA_LOAD_RECOVER;
2956
2957		spa_activate(spa, spa_mode_global);
2958
2959		if (state != SPA_LOAD_RECOVER)
2960			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2961
2962		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2963		    policy.zrp_request);
2964
2965		if (error == EBADF) {
2966			/*
2967			 * If vdev_validate() returns failure (indicated by
2968			 * EBADF), it indicates that one of the vdevs indicates
2969			 * that the pool has been exported or destroyed.  If
2970			 * this is the case, the config cache is out of sync and
2971			 * we should remove the pool from the namespace.
2972			 */
2973			spa_unload(spa);
2974			spa_deactivate(spa);
2975			spa_config_sync(spa, B_TRUE, B_TRUE);
2976			spa_remove(spa);
2977			if (locked)
2978				mutex_exit(&spa_namespace_lock);
2979			return (SET_ERROR(ENOENT));
2980		}
2981
2982		if (error) {
2983			/*
2984			 * We can't open the pool, but we still have useful
2985			 * information: the state of each vdev after the
2986			 * attempted vdev_open().  Return this to the user.
2987			 */
2988			if (config != NULL && spa->spa_config) {
2989				VERIFY(nvlist_dup(spa->spa_config, config,
2990				    KM_SLEEP) == 0);
2991				VERIFY(nvlist_add_nvlist(*config,
2992				    ZPOOL_CONFIG_LOAD_INFO,
2993				    spa->spa_load_info) == 0);
2994			}
2995			spa_unload(spa);
2996			spa_deactivate(spa);
2997			spa->spa_last_open_failed = error;
2998			if (locked)
2999				mutex_exit(&spa_namespace_lock);
3000			*spapp = NULL;
3001			return (error);
3002		}
3003	}
3004
3005	spa_open_ref(spa, tag);
3006
3007	if (config != NULL)
3008		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3009
3010	/*
3011	 * If we've recovered the pool, pass back any information we
3012	 * gathered while doing the load.
3013	 */
3014	if (state == SPA_LOAD_RECOVER) {
3015		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
3016		    spa->spa_load_info) == 0);
3017	}
3018
3019	if (locked) {
3020		spa->spa_last_open_failed = 0;
3021		spa->spa_last_ubsync_txg = 0;
3022		spa->spa_load_txg = 0;
3023		mutex_exit(&spa_namespace_lock);
3024	}
3025
3026	*spapp = spa;
3027
3028	return (0);
3029}
3030
3031int
3032spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3033    nvlist_t **config)
3034{
3035	return (spa_open_common(name, spapp, tag, policy, config));
3036}
3037
3038int
3039spa_open(const char *name, spa_t **spapp, void *tag)
3040{
3041	return (spa_open_common(name, spapp, tag, NULL, NULL));
3042}
3043
3044/*
3045 * Lookup the given spa_t, incrementing the inject count in the process,
3046 * preventing it from being exported or destroyed.
3047 */
3048spa_t *
3049spa_inject_addref(char *name)
3050{
3051	spa_t *spa;
3052
3053	mutex_enter(&spa_namespace_lock);
3054	if ((spa = spa_lookup(name)) == NULL) {
3055		mutex_exit(&spa_namespace_lock);
3056		return (NULL);
3057	}
3058	spa->spa_inject_ref++;
3059	mutex_exit(&spa_namespace_lock);
3060
3061	return (spa);
3062}
3063
3064void
3065spa_inject_delref(spa_t *spa)
3066{
3067	mutex_enter(&spa_namespace_lock);
3068	spa->spa_inject_ref--;
3069	mutex_exit(&spa_namespace_lock);
3070}
3071
3072/*
3073 * Add spares device information to the nvlist.
3074 */
3075static void
3076spa_add_spares(spa_t *spa, nvlist_t *config)
3077{
3078	nvlist_t **spares;
3079	uint_t i, nspares;
3080	nvlist_t *nvroot;
3081	uint64_t guid;
3082	vdev_stat_t *vs;
3083	uint_t vsc;
3084	uint64_t pool;
3085
3086	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3087
3088	if (spa->spa_spares.sav_count == 0)
3089		return;
3090
3091	VERIFY(nvlist_lookup_nvlist(config,
3092	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3093	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3094	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3095	if (nspares != 0) {
3096		VERIFY(nvlist_add_nvlist_array(nvroot,
3097		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3098		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3099		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3100
3101		/*
3102		 * Go through and find any spares which have since been
3103		 * repurposed as an active spare.  If this is the case, update
3104		 * their status appropriately.
3105		 */
3106		for (i = 0; i < nspares; i++) {
3107			VERIFY(nvlist_lookup_uint64(spares[i],
3108			    ZPOOL_CONFIG_GUID, &guid) == 0);
3109			if (spa_spare_exists(guid, &pool, NULL) &&
3110			    pool != 0ULL) {
3111				VERIFY(nvlist_lookup_uint64_array(
3112				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
3113				    (uint64_t **)&vs, &vsc) == 0);
3114				vs->vs_state = VDEV_STATE_CANT_OPEN;
3115				vs->vs_aux = VDEV_AUX_SPARED;
3116			}
3117		}
3118	}
3119}
3120
3121/*
3122 * Add l2cache device information to the nvlist, including vdev stats.
3123 */
3124static void
3125spa_add_l2cache(spa_t *spa, nvlist_t *config)
3126{
3127	nvlist_t **l2cache;
3128	uint_t i, j, nl2cache;
3129	nvlist_t *nvroot;
3130	uint64_t guid;
3131	vdev_t *vd;
3132	vdev_stat_t *vs;
3133	uint_t vsc;
3134
3135	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3136
3137	if (spa->spa_l2cache.sav_count == 0)
3138		return;
3139
3140	VERIFY(nvlist_lookup_nvlist(config,
3141	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3142	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3143	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3144	if (nl2cache != 0) {
3145		VERIFY(nvlist_add_nvlist_array(nvroot,
3146		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3147		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3148		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3149
3150		/*
3151		 * Update level 2 cache device stats.
3152		 */
3153
3154		for (i = 0; i < nl2cache; i++) {
3155			VERIFY(nvlist_lookup_uint64(l2cache[i],
3156			    ZPOOL_CONFIG_GUID, &guid) == 0);
3157
3158			vd = NULL;
3159			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3160				if (guid ==
3161				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3162					vd = spa->spa_l2cache.sav_vdevs[j];
3163					break;
3164				}
3165			}
3166			ASSERT(vd != NULL);
3167
3168			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3169			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3170			    == 0);
3171			vdev_get_stats(vd, vs);
3172		}
3173	}
3174}
3175
3176static void
3177spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3178{
3179	nvlist_t *features;
3180	zap_cursor_t zc;
3181	zap_attribute_t za;
3182
3183	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3184	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3185
3186	if (spa->spa_feat_for_read_obj != 0) {
3187		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3188		    spa->spa_feat_for_read_obj);
3189		    zap_cursor_retrieve(&zc, &za) == 0;
3190		    zap_cursor_advance(&zc)) {
3191			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3192			    za.za_num_integers == 1);
3193			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3194			    za.za_first_integer));
3195		}
3196		zap_cursor_fini(&zc);
3197	}
3198
3199	if (spa->spa_feat_for_write_obj != 0) {
3200		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3201		    spa->spa_feat_for_write_obj);
3202		    zap_cursor_retrieve(&zc, &za) == 0;
3203		    zap_cursor_advance(&zc)) {
3204			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3205			    za.za_num_integers == 1);
3206			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3207			    za.za_first_integer));
3208		}
3209		zap_cursor_fini(&zc);
3210	}
3211
3212	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3213	    features) == 0);
3214	nvlist_free(features);
3215}
3216
3217int
3218spa_get_stats(const char *name, nvlist_t **config,
3219    char *altroot, size_t buflen)
3220{
3221	int error;
3222	spa_t *spa;
3223
3224	*config = NULL;
3225	error = spa_open_common(name, &spa, FTAG, NULL, config);
3226
3227	if (spa != NULL) {
3228		/*
3229		 * This still leaves a window of inconsistency where the spares
3230		 * or l2cache devices could change and the config would be
3231		 * self-inconsistent.
3232		 */
3233		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3234
3235		if (*config != NULL) {
3236			uint64_t loadtimes[2];
3237
3238			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3239			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3240			VERIFY(nvlist_add_uint64_array(*config,
3241			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3242
3243			VERIFY(nvlist_add_uint64(*config,
3244			    ZPOOL_CONFIG_ERRCOUNT,
3245			    spa_get_errlog_size(spa)) == 0);
3246
3247			if (spa_suspended(spa))
3248				VERIFY(nvlist_add_uint64(*config,
3249				    ZPOOL_CONFIG_SUSPENDED,
3250				    spa->spa_failmode) == 0);
3251
3252			spa_add_spares(spa, *config);
3253			spa_add_l2cache(spa, *config);
3254			spa_add_feature_stats(spa, *config);
3255		}
3256	}
3257
3258	/*
3259	 * We want to get the alternate root even for faulted pools, so we cheat
3260	 * and call spa_lookup() directly.
3261	 */
3262	if (altroot) {
3263		if (spa == NULL) {
3264			mutex_enter(&spa_namespace_lock);
3265			spa = spa_lookup(name);
3266			if (spa)
3267				spa_altroot(spa, altroot, buflen);
3268			else
3269				altroot[0] = '\0';
3270			spa = NULL;
3271			mutex_exit(&spa_namespace_lock);
3272		} else {
3273			spa_altroot(spa, altroot, buflen);
3274		}
3275	}
3276
3277	if (spa != NULL) {
3278		spa_config_exit(spa, SCL_CONFIG, FTAG);
3279		spa_close(spa, FTAG);
3280	}
3281
3282	return (error);
3283}
3284
3285/*
3286 * Validate that the auxiliary device array is well formed.  We must have an
3287 * array of nvlists, each which describes a valid leaf vdev.  If this is an
3288 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3289 * specified, as long as they are well-formed.
3290 */
3291static int
3292spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3293    spa_aux_vdev_t *sav, const char *config, uint64_t version,
3294    vdev_labeltype_t label)
3295{
3296	nvlist_t **dev;
3297	uint_t i, ndev;
3298	vdev_t *vd;
3299	int error;
3300
3301	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3302
3303	/*
3304	 * It's acceptable to have no devs specified.
3305	 */
3306	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3307		return (0);
3308
3309	if (ndev == 0)
3310		return (SET_ERROR(EINVAL));
3311
3312	/*
3313	 * Make sure the pool is formatted with a version that supports this
3314	 * device type.
3315	 */
3316	if (spa_version(spa) < version)
3317		return (SET_ERROR(ENOTSUP));
3318
3319	/*
3320	 * Set the pending device list so we correctly handle device in-use
3321	 * checking.
3322	 */
3323	sav->sav_pending = dev;
3324	sav->sav_npending = ndev;
3325
3326	for (i = 0; i < ndev; i++) {
3327		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3328		    mode)) != 0)
3329			goto out;
3330
3331		if (!vd->vdev_ops->vdev_op_leaf) {
3332			vdev_free(vd);
3333			error = SET_ERROR(EINVAL);
3334			goto out;
3335		}
3336
3337		/*
3338		 * The L2ARC currently only supports disk devices in
3339		 * kernel context.  For user-level testing, we allow it.
3340		 */
3341#ifdef _KERNEL
3342		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3343		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3344			error = SET_ERROR(ENOTBLK);
3345			vdev_free(vd);
3346			goto out;
3347		}
3348#endif
3349		vd->vdev_top = vd;
3350
3351		if ((error = vdev_open(vd)) == 0 &&
3352		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
3353			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3354			    vd->vdev_guid) == 0);
3355		}
3356
3357		vdev_free(vd);
3358
3359		if (error &&
3360		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3361			goto out;
3362		else
3363			error = 0;
3364	}
3365
3366out:
3367	sav->sav_pending = NULL;
3368	sav->sav_npending = 0;
3369	return (error);
3370}
3371
3372static int
3373spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3374{
3375	int error;
3376
3377	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3378
3379	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3380	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3381	    VDEV_LABEL_SPARE)) != 0) {
3382		return (error);
3383	}
3384
3385	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3386	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3387	    VDEV_LABEL_L2CACHE));
3388}
3389
3390static void
3391spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3392    const char *config)
3393{
3394	int i;
3395
3396	if (sav->sav_config != NULL) {
3397		nvlist_t **olddevs;
3398		uint_t oldndevs;
3399		nvlist_t **newdevs;
3400
3401		/*
3402		 * Generate new dev list by concatentating with the
3403		 * current dev list.
3404		 */
3405		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3406		    &olddevs, &oldndevs) == 0);
3407
3408		newdevs = kmem_alloc(sizeof (void *) *
3409		    (ndevs + oldndevs), KM_SLEEP);
3410		for (i = 0; i < oldndevs; i++)
3411			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3412			    KM_SLEEP) == 0);
3413		for (i = 0; i < ndevs; i++)
3414			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3415			    KM_SLEEP) == 0);
3416
3417		VERIFY(nvlist_remove(sav->sav_config, config,
3418		    DATA_TYPE_NVLIST_ARRAY) == 0);
3419
3420		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3421		    config, newdevs, ndevs + oldndevs) == 0);
3422		for (i = 0; i < oldndevs + ndevs; i++)
3423			nvlist_free(newdevs[i]);
3424		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3425	} else {
3426		/*
3427		 * Generate a new dev list.
3428		 */
3429		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3430		    KM_SLEEP) == 0);
3431		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3432		    devs, ndevs) == 0);
3433	}
3434}
3435
3436/*
3437 * Stop and drop level 2 ARC devices
3438 */
3439void
3440spa_l2cache_drop(spa_t *spa)
3441{
3442	vdev_t *vd;
3443	int i;
3444	spa_aux_vdev_t *sav = &spa->spa_l2cache;
3445
3446	for (i = 0; i < sav->sav_count; i++) {
3447		uint64_t pool;
3448
3449		vd = sav->sav_vdevs[i];
3450		ASSERT(vd != NULL);
3451
3452		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3453		    pool != 0ULL && l2arc_vdev_present(vd))
3454			l2arc_remove_vdev(vd);
3455	}
3456}
3457
3458/*
3459 * Pool Creation
3460 */
3461int
3462spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3463    nvlist_t *zplprops)
3464{
3465	spa_t *spa;
3466	char *altroot = NULL;
3467	vdev_t *rvd;
3468	dsl_pool_t *dp;
3469	dmu_tx_t *tx;
3470	int error = 0;
3471	uint64_t txg = TXG_INITIAL;
3472	nvlist_t **spares, **l2cache;
3473	uint_t nspares, nl2cache;
3474	uint64_t version, obj;
3475	boolean_t has_features;
3476
3477	/*
3478	 * If this pool already exists, return failure.
3479	 */
3480	mutex_enter(&spa_namespace_lock);
3481	if (spa_lookup(pool) != NULL) {
3482		mutex_exit(&spa_namespace_lock);
3483		return (SET_ERROR(EEXIST));
3484	}
3485
3486	/*
3487	 * Allocate a new spa_t structure.
3488	 */
3489	(void) nvlist_lookup_string(props,
3490	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3491	spa = spa_add(pool, NULL, altroot);
3492	spa_activate(spa, spa_mode_global);
3493
3494	if (props && (error = spa_prop_validate(spa, props))) {
3495		spa_deactivate(spa);
3496		spa_remove(spa);
3497		mutex_exit(&spa_namespace_lock);
3498		return (error);
3499	}
3500
3501	has_features = B_FALSE;
3502	for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3503	    elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3504		if (zpool_prop_feature(nvpair_name(elem)))
3505			has_features = B_TRUE;
3506	}
3507
3508	if (has_features || nvlist_lookup_uint64(props,
3509	    zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3510		version = SPA_VERSION;
3511	}
3512	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3513
3514	spa->spa_first_txg = txg;
3515	spa->spa_uberblock.ub_txg = txg - 1;
3516	spa->spa_uberblock.ub_version = version;
3517	spa->spa_ubsync = spa->spa_uberblock;
3518
3519	/*
3520	 * Create "The Godfather" zio to hold all async IOs
3521	 */
3522	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
3523	    KM_SLEEP);
3524	for (int i = 0; i < max_ncpus; i++) {
3525		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
3526		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
3527		    ZIO_FLAG_GODFATHER);
3528	}
3529
3530	/*
3531	 * Create the root vdev.
3532	 */
3533	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3534
3535	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3536
3537	ASSERT(error != 0 || rvd != NULL);
3538	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3539
3540	if (error == 0 && !zfs_allocatable_devs(nvroot))
3541		error = SET_ERROR(EINVAL);
3542
3543	if (error == 0 &&
3544	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3545	    (error = spa_validate_aux(spa, nvroot, txg,
3546	    VDEV_ALLOC_ADD)) == 0) {
3547		for (int c = 0; c < rvd->vdev_children; c++) {
3548			vdev_metaslab_set_size(rvd->vdev_child[c]);
3549			vdev_expand(rvd->vdev_child[c], txg);
3550		}
3551	}
3552
3553	spa_config_exit(spa, SCL_ALL, FTAG);
3554
3555	if (error != 0) {
3556		spa_unload(spa);
3557		spa_deactivate(spa);
3558		spa_remove(spa);
3559		mutex_exit(&spa_namespace_lock);
3560		return (error);
3561	}
3562
3563	/*
3564	 * Get the list of spares, if specified.
3565	 */
3566	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3567	    &spares, &nspares) == 0) {
3568		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3569		    KM_SLEEP) == 0);
3570		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3571		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3572		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3573		spa_load_spares(spa);
3574		spa_config_exit(spa, SCL_ALL, FTAG);
3575		spa->spa_spares.sav_sync = B_TRUE;
3576	}
3577
3578	/*
3579	 * Get the list of level 2 cache devices, if specified.
3580	 */
3581	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3582	    &l2cache, &nl2cache) == 0) {
3583		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3584		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3585		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3586		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3587		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3588		spa_load_l2cache(spa);
3589		spa_config_exit(spa, SCL_ALL, FTAG);
3590		spa->spa_l2cache.sav_sync = B_TRUE;
3591	}
3592
3593	spa->spa_is_initializing = B_TRUE;
3594	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3595	spa->spa_meta_objset = dp->dp_meta_objset;
3596	spa->spa_is_initializing = B_FALSE;
3597
3598	/*
3599	 * Create DDTs (dedup tables).
3600	 */
3601	ddt_create(spa);
3602
3603	spa_update_dspace(spa);
3604
3605	tx = dmu_tx_create_assigned(dp, txg);
3606
3607	/*
3608	 * Create the pool config object.
3609	 */
3610	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3611	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3612	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3613
3614	if (zap_add(spa->spa_meta_objset,
3615	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3616	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3617		cmn_err(CE_PANIC, "failed to add pool config");
3618	}
3619
3620	if (spa_version(spa) >= SPA_VERSION_FEATURES)
3621		spa_feature_create_zap_objects(spa, tx);
3622
3623	if (zap_add(spa->spa_meta_objset,
3624	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3625	    sizeof (uint64_t), 1, &version, tx) != 0) {
3626		cmn_err(CE_PANIC, "failed to add pool version");
3627	}
3628
3629	/* Newly created pools with the right version are always deflated. */
3630	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3631		spa->spa_deflate = TRUE;
3632		if (zap_add(spa->spa_meta_objset,
3633		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3634		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3635			cmn_err(CE_PANIC, "failed to add deflate");
3636		}
3637	}
3638
3639	/*
3640	 * Create the deferred-free bpobj.  Turn off compression
3641	 * because sync-to-convergence takes longer if the blocksize
3642	 * keeps changing.
3643	 */
3644	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3645	dmu_object_set_compress(spa->spa_meta_objset, obj,
3646	    ZIO_COMPRESS_OFF, tx);
3647	if (zap_add(spa->spa_meta_objset,
3648	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3649	    sizeof (uint64_t), 1, &obj, tx) != 0) {
3650		cmn_err(CE_PANIC, "failed to add bpobj");
3651	}
3652	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3653	    spa->spa_meta_objset, obj));
3654
3655	/*
3656	 * Create the pool's history object.
3657	 */
3658	if (version >= SPA_VERSION_ZPOOL_HISTORY)
3659		spa_history_create_obj(spa, tx);
3660
3661	/*
3662	 * Set pool properties.
3663	 */
3664	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3665	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3666	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3667	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3668
3669	if (props != NULL) {
3670		spa_configfile_set(spa, props, B_FALSE);
3671		spa_sync_props(props, tx);
3672	}
3673
3674	dmu_tx_commit(tx);
3675
3676	spa->spa_sync_on = B_TRUE;
3677	txg_sync_start(spa->spa_dsl_pool);
3678
3679	/*
3680	 * We explicitly wait for the first transaction to complete so that our
3681	 * bean counters are appropriately updated.
3682	 */
3683	txg_wait_synced(spa->spa_dsl_pool, txg);
3684
3685	spa_config_sync(spa, B_FALSE, B_TRUE);
3686
3687	spa_history_log_version(spa, "create");
3688
3689	/*
3690	 * Don't count references from objsets that are already closed
3691	 * and are making their way through the eviction process.
3692	 */
3693	spa_evicting_os_wait(spa);
3694	spa->spa_minref = refcount_count(&spa->spa_refcount);
3695
3696	mutex_exit(&spa_namespace_lock);
3697
3698	return (0);
3699}
3700
3701#ifdef _KERNEL
3702/*
3703 * Get the root pool information from the root disk, then import the root pool
3704 * during the system boot up time.
3705 */
3706extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3707
3708static nvlist_t *
3709spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3710{
3711	nvlist_t *config;
3712	nvlist_t *nvtop, *nvroot;
3713	uint64_t pgid;
3714
3715	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3716		return (NULL);
3717
3718	/*
3719	 * Add this top-level vdev to the child array.
3720	 */
3721	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3722	    &nvtop) == 0);
3723	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3724	    &pgid) == 0);
3725	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3726
3727	/*
3728	 * Put this pool's top-level vdevs into a root vdev.
3729	 */
3730	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3731	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3732	    VDEV_TYPE_ROOT) == 0);
3733	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3734	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3735	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3736	    &nvtop, 1) == 0);
3737
3738	/*
3739	 * Replace the existing vdev_tree with the new root vdev in
3740	 * this pool's configuration (remove the old, add the new).
3741	 */
3742	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3743	nvlist_free(nvroot);
3744	return (config);
3745}
3746
3747/*
3748 * Walk the vdev tree and see if we can find a device with "better"
3749 * configuration. A configuration is "better" if the label on that
3750 * device has a more recent txg.
3751 */
3752static void
3753spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3754{
3755	for (int c = 0; c < vd->vdev_children; c++)
3756		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3757
3758	if (vd->vdev_ops->vdev_op_leaf) {
3759		nvlist_t *label;
3760		uint64_t label_txg;
3761
3762		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3763		    &label) != 0)
3764			return;
3765
3766		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3767		    &label_txg) == 0);
3768
3769		/*
3770		 * Do we have a better boot device?
3771		 */
3772		if (label_txg > *txg) {
3773			*txg = label_txg;
3774			*avd = vd;
3775		}
3776		nvlist_free(label);
3777	}
3778}
3779
3780/*
3781 * Import a root pool.
3782 *
3783 * For x86. devpath_list will consist of devid and/or physpath name of
3784 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3785 * The GRUB "findroot" command will return the vdev we should boot.
3786 *
3787 * For Sparc, devpath_list consists the physpath name of the booting device
3788 * no matter the rootpool is a single device pool or a mirrored pool.
3789 * e.g.
3790 *	"/pci@1f,0/ide@d/disk@0,0:a"
3791 */
3792int
3793spa_import_rootpool(char *devpath, char *devid)
3794{
3795	spa_t *spa;
3796	vdev_t *rvd, *bvd, *avd = NULL;
3797	nvlist_t *config, *nvtop;
3798	uint64_t guid, txg;
3799	char *pname;
3800	int error;
3801
3802	/*
3803	 * Read the label from the boot device and generate a configuration.
3804	 */
3805	config = spa_generate_rootconf(devpath, devid, &guid);
3806#if defined(_OBP) && defined(_KERNEL)
3807	if (config == NULL) {
3808		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3809			/* iscsi boot */
3810			get_iscsi_bootpath_phy(devpath);
3811			config = spa_generate_rootconf(devpath, devid, &guid);
3812		}
3813	}
3814#endif
3815	if (config == NULL) {
3816		cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3817		    devpath);
3818		return (SET_ERROR(EIO));
3819	}
3820
3821	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3822	    &pname) == 0);
3823	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3824
3825	mutex_enter(&spa_namespace_lock);
3826	if ((spa = spa_lookup(pname)) != NULL) {
3827		/*
3828		 * Remove the existing root pool from the namespace so that we
3829		 * can replace it with the correct config we just read in.
3830		 */
3831		spa_remove(spa);
3832	}
3833
3834	spa = spa_add(pname, config, NULL);
3835	spa->spa_is_root = B_TRUE;
3836	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3837
3838	/*
3839	 * Build up a vdev tree based on the boot device's label config.
3840	 */
3841	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3842	    &nvtop) == 0);
3843	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3844	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3845	    VDEV_ALLOC_ROOTPOOL);
3846	spa_config_exit(spa, SCL_ALL, FTAG);
3847	if (error) {
3848		mutex_exit(&spa_namespace_lock);
3849		nvlist_free(config);
3850		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3851		    pname);
3852		return (error);
3853	}
3854
3855	/*
3856	 * Get the boot vdev.
3857	 */
3858	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3859		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3860		    (u_longlong_t)guid);
3861		error = SET_ERROR(ENOENT);
3862		goto out;
3863	}
3864
3865	/*
3866	 * Determine if there is a better boot device.
3867	 */
3868	avd = bvd;
3869	spa_alt_rootvdev(rvd, &avd, &txg);
3870	if (avd != bvd) {
3871		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3872		    "try booting from '%s'", avd->vdev_path);
3873		error = SET_ERROR(EINVAL);
3874		goto out;
3875	}
3876
3877	/*
3878	 * If the boot device is part of a spare vdev then ensure that
3879	 * we're booting off the active spare.
3880	 */
3881	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3882	    !bvd->vdev_isspare) {
3883		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3884		    "try booting from '%s'",
3885		    bvd->vdev_parent->
3886		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3887		error = SET_ERROR(EINVAL);
3888		goto out;
3889	}
3890
3891	error = 0;
3892out:
3893	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3894	vdev_free(rvd);
3895	spa_config_exit(spa, SCL_ALL, FTAG);
3896	mutex_exit(&spa_namespace_lock);
3897
3898	nvlist_free(config);
3899	return (error);
3900}
3901
3902#endif
3903
3904/*
3905 * Import a non-root pool into the system.
3906 */
3907int
3908spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3909{
3910	spa_t *spa;
3911	char *altroot = NULL;
3912	spa_load_state_t state = SPA_LOAD_IMPORT;
3913	zpool_rewind_policy_t policy;
3914	uint64_t mode = spa_mode_global;
3915	uint64_t readonly = B_FALSE;
3916	int error;
3917	nvlist_t *nvroot;
3918	nvlist_t **spares, **l2cache;
3919	uint_t nspares, nl2cache;
3920
3921	/*
3922	 * If a pool with this name exists, return failure.
3923	 */
3924	mutex_enter(&spa_namespace_lock);
3925	if (spa_lookup(pool) != NULL) {
3926		mutex_exit(&spa_namespace_lock);
3927		return (SET_ERROR(EEXIST));
3928	}
3929
3930	/*
3931	 * Create and initialize the spa structure.
3932	 */
3933	(void) nvlist_lookup_string(props,
3934	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3935	(void) nvlist_lookup_uint64(props,
3936	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3937	if (readonly)
3938		mode = FREAD;
3939	spa = spa_add(pool, config, altroot);
3940	spa->spa_import_flags = flags;
3941
3942	/*
3943	 * Verbatim import - Take a pool and insert it into the namespace
3944	 * as if it had been loaded at boot.
3945	 */
3946	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3947		if (props != NULL)
3948			spa_configfile_set(spa, props, B_FALSE);
3949
3950		spa_config_sync(spa, B_FALSE, B_TRUE);
3951
3952		mutex_exit(&spa_namespace_lock);
3953		return (0);
3954	}
3955
3956	spa_activate(spa, mode);
3957
3958	/*
3959	 * Don't start async tasks until we know everything is healthy.
3960	 */
3961	spa_async_suspend(spa);
3962
3963	zpool_get_rewind_policy(config, &policy);
3964	if (policy.zrp_request & ZPOOL_DO_REWIND)
3965		state = SPA_LOAD_RECOVER;
3966
3967	/*
3968	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3969	 * because the user-supplied config is actually the one to trust when
3970	 * doing an import.
3971	 */
3972	if (state != SPA_LOAD_RECOVER)
3973		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3974
3975	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3976	    policy.zrp_request);
3977
3978	/*
3979	 * Propagate anything learned while loading the pool and pass it
3980	 * back to caller (i.e. rewind info, missing devices, etc).
3981	 */
3982	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3983	    spa->spa_load_info) == 0);
3984
3985	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3986	/*
3987	 * Toss any existing sparelist, as it doesn't have any validity
3988	 * anymore, and conflicts with spa_has_spare().
3989	 */
3990	if (spa->spa_spares.sav_config) {
3991		nvlist_free(spa->spa_spares.sav_config);
3992		spa->spa_spares.sav_config = NULL;
3993		spa_load_spares(spa);
3994	}
3995	if (spa->spa_l2cache.sav_config) {
3996		nvlist_free(spa->spa_l2cache.sav_config);
3997		spa->spa_l2cache.sav_config = NULL;
3998		spa_load_l2cache(spa);
3999	}
4000
4001	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4002	    &nvroot) == 0);
4003	if (error == 0)
4004		error = spa_validate_aux(spa, nvroot, -1ULL,
4005		    VDEV_ALLOC_SPARE);
4006	if (error == 0)
4007		error = spa_validate_aux(spa, nvroot, -1ULL,
4008		    VDEV_ALLOC_L2CACHE);
4009	spa_config_exit(spa, SCL_ALL, FTAG);
4010
4011	if (props != NULL)
4012		spa_configfile_set(spa, props, B_FALSE);
4013
4014	if (error != 0 || (props && spa_writeable(spa) &&
4015	    (error = spa_prop_set(spa, props)))) {
4016		spa_unload(spa);
4017		spa_deactivate(spa);
4018		spa_remove(spa);
4019		mutex_exit(&spa_namespace_lock);
4020		return (error);
4021	}
4022
4023	spa_async_resume(spa);
4024
4025	/*
4026	 * Override any spares and level 2 cache devices as specified by
4027	 * the user, as these may have correct device names/devids, etc.
4028	 */
4029	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4030	    &spares, &nspares) == 0) {
4031		if (spa->spa_spares.sav_config)
4032			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4033			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4034		else
4035			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4036			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4037		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4038		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4039		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4040		spa_load_spares(spa);
4041		spa_config_exit(spa, SCL_ALL, FTAG);
4042		spa->spa_spares.sav_sync = B_TRUE;
4043	}
4044	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4045	    &l2cache, &nl2cache) == 0) {
4046		if (spa->spa_l2cache.sav_config)
4047			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4048			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4049		else
4050			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4051			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4052		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4053		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4054		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4055		spa_load_l2cache(spa);
4056		spa_config_exit(spa, SCL_ALL, FTAG);
4057		spa->spa_l2cache.sav_sync = B_TRUE;
4058	}
4059
4060	/*
4061	 * Check for any removed devices.
4062	 */
4063	if (spa->spa_autoreplace) {
4064		spa_aux_check_removed(&spa->spa_spares);
4065		spa_aux_check_removed(&spa->spa_l2cache);
4066	}
4067
4068	if (spa_writeable(spa)) {
4069		/*
4070		 * Update the config cache to include the newly-imported pool.
4071		 */
4072		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4073	}
4074
4075	/*
4076	 * It's possible that the pool was expanded while it was exported.
4077	 * We kick off an async task to handle this for us.
4078	 */
4079	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4080
4081	mutex_exit(&spa_namespace_lock);
4082	spa_history_log_version(spa, "import");
4083
4084	return (0);
4085}
4086
4087nvlist_t *
4088spa_tryimport(nvlist_t *tryconfig)
4089{
4090	nvlist_t *config = NULL;
4091	char *poolname;
4092	spa_t *spa;
4093	uint64_t state;
4094	int error;
4095
4096	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4097		return (NULL);
4098
4099	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4100		return (NULL);
4101
4102	/*
4103	 * Create and initialize the spa structure.
4104	 */
4105	mutex_enter(&spa_namespace_lock);
4106	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4107	spa_activate(spa, FREAD);
4108
4109	/*
4110	 * Pass off the heavy lifting to spa_load().
4111	 * Pass TRUE for mosconfig because the user-supplied config
4112	 * is actually the one to trust when doing an import.
4113	 */
4114	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4115
4116	/*
4117	 * If 'tryconfig' was at least parsable, return the current config.
4118	 */
4119	if (spa->spa_root_vdev != NULL) {
4120		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4121		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4122		    poolname) == 0);
4123		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4124		    state) == 0);
4125		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4126		    spa->spa_uberblock.ub_timestamp) == 0);
4127		VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4128		    spa->spa_load_info) == 0);
4129
4130		/*
4131		 * If the bootfs property exists on this pool then we
4132		 * copy it out so that external consumers can tell which
4133		 * pools are bootable.
4134		 */
4135		if ((!error || error == EEXIST) && spa->spa_bootfs) {
4136			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4137
4138			/*
4139			 * We have to play games with the name since the
4140			 * pool was opened as TRYIMPORT_NAME.
4141			 */
4142			if (dsl_dsobj_to_dsname(spa_name(spa),
4143			    spa->spa_bootfs, tmpname) == 0) {
4144				char *cp;
4145				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4146
4147				cp = strchr(tmpname, '/');
4148				if (cp == NULL) {
4149					(void) strlcpy(dsname, tmpname,
4150					    MAXPATHLEN);
4151				} else {
4152					(void) snprintf(dsname, MAXPATHLEN,
4153					    "%s/%s", poolname, ++cp);
4154				}
4155				VERIFY(nvlist_add_string(config,
4156				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4157				kmem_free(dsname, MAXPATHLEN);
4158			}
4159			kmem_free(tmpname, MAXPATHLEN);
4160		}
4161
4162		/*
4163		 * Add the list of hot spares and level 2 cache devices.
4164		 */
4165		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4166		spa_add_spares(spa, config);
4167		spa_add_l2cache(spa, config);
4168		spa_config_exit(spa, SCL_CONFIG, FTAG);
4169	}
4170
4171	spa_unload(spa);
4172	spa_deactivate(spa);
4173	spa_remove(spa);
4174	mutex_exit(&spa_namespace_lock);
4175
4176	return (config);
4177}
4178
4179/*
4180 * Pool export/destroy
4181 *
4182 * The act of destroying or exporting a pool is very simple.  We make sure there
4183 * is no more pending I/O and any references to the pool are gone.  Then, we
4184 * update the pool state and sync all the labels to disk, removing the
4185 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4186 * we don't sync the labels or remove the configuration cache.
4187 */
4188static int
4189spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4190    boolean_t force, boolean_t hardforce)
4191{
4192	spa_t *spa;
4193
4194	if (oldconfig)
4195		*oldconfig = NULL;
4196
4197	if (!(spa_mode_global & FWRITE))
4198		return (SET_ERROR(EROFS));
4199
4200	mutex_enter(&spa_namespace_lock);
4201	if ((spa = spa_lookup(pool)) == NULL) {
4202		mutex_exit(&spa_namespace_lock);
4203		return (SET_ERROR(ENOENT));
4204	}
4205
4206	/*
4207	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4208	 * reacquire the namespace lock, and see if we can export.
4209	 */
4210	spa_open_ref(spa, FTAG);
4211	mutex_exit(&spa_namespace_lock);
4212	spa_async_suspend(spa);
4213	mutex_enter(&spa_namespace_lock);
4214	spa_close(spa, FTAG);
4215
4216	/*
4217	 * The pool will be in core if it's openable,
4218	 * in which case we can modify its state.
4219	 */
4220	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4221		/*
4222		 * Objsets may be open only because they're dirty, so we
4223		 * have to force it to sync before checking spa_refcnt.
4224		 */
4225		txg_wait_synced(spa->spa_dsl_pool, 0);
4226		spa_evicting_os_wait(spa);
4227
4228		/*
4229		 * A pool cannot be exported or destroyed if there are active
4230		 * references.  If we are resetting a pool, allow references by
4231		 * fault injection handlers.
4232		 */
4233		if (!spa_refcount_zero(spa) ||
4234		    (spa->spa_inject_ref != 0 &&
4235		    new_state != POOL_STATE_UNINITIALIZED)) {
4236			spa_async_resume(spa);
4237			mutex_exit(&spa_namespace_lock);
4238			return (SET_ERROR(EBUSY));
4239		}
4240
4241		/*
4242		 * A pool cannot be exported if it has an active shared spare.
4243		 * This is to prevent other pools stealing the active spare
4244		 * from an exported pool. At user's own will, such pool can
4245		 * be forcedly exported.
4246		 */
4247		if (!force && new_state == POOL_STATE_EXPORTED &&
4248		    spa_has_active_shared_spare(spa)) {
4249			spa_async_resume(spa);
4250			mutex_exit(&spa_namespace_lock);
4251			return (SET_ERROR(EXDEV));
4252		}
4253
4254		/*
4255		 * We want this to be reflected on every label,
4256		 * so mark them all dirty.  spa_unload() will do the
4257		 * final sync that pushes these changes out.
4258		 */
4259		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4260			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4261			spa->spa_state = new_state;
4262			spa->spa_final_txg = spa_last_synced_txg(spa) +
4263			    TXG_DEFER_SIZE + 1;
4264			vdev_config_dirty(spa->spa_root_vdev);
4265			spa_config_exit(spa, SCL_ALL, FTAG);
4266		}
4267	}
4268
4269	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4270
4271	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4272		spa_unload(spa);
4273		spa_deactivate(spa);
4274	}
4275
4276	if (oldconfig && spa->spa_config)
4277		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4278
4279	if (new_state != POOL_STATE_UNINITIALIZED) {
4280		if (!hardforce)
4281			spa_config_sync(spa, B_TRUE, B_TRUE);
4282		spa_remove(spa);
4283	}
4284	mutex_exit(&spa_namespace_lock);
4285
4286	return (0);
4287}
4288
4289/*
4290 * Destroy a storage pool.
4291 */
4292int
4293spa_destroy(char *pool)
4294{
4295	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4296	    B_FALSE, B_FALSE));
4297}
4298
4299/*
4300 * Export a storage pool.
4301 */
4302int
4303spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4304    boolean_t hardforce)
4305{
4306	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4307	    force, hardforce));
4308}
4309
4310/*
4311 * Similar to spa_export(), this unloads the spa_t without actually removing it
4312 * from the namespace in any way.
4313 */
4314int
4315spa_reset(char *pool)
4316{
4317	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4318	    B_FALSE, B_FALSE));
4319}
4320
4321/*
4322 * ==========================================================================
4323 * Device manipulation
4324 * ==========================================================================
4325 */
4326
4327/*
4328 * Add a device to a storage pool.
4329 */
4330int
4331spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4332{
4333	uint64_t txg, id;
4334	int error;
4335	vdev_t *rvd = spa->spa_root_vdev;
4336	vdev_t *vd, *tvd;
4337	nvlist_t **spares, **l2cache;
4338	uint_t nspares, nl2cache;
4339
4340	ASSERT(spa_writeable(spa));
4341
4342	txg = spa_vdev_enter(spa);
4343
4344	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4345	    VDEV_ALLOC_ADD)) != 0)
4346		return (spa_vdev_exit(spa, NULL, txg, error));
4347
4348	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
4349
4350	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4351	    &nspares) != 0)
4352		nspares = 0;
4353
4354	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4355	    &nl2cache) != 0)
4356		nl2cache = 0;
4357
4358	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4359		return (spa_vdev_exit(spa, vd, txg, EINVAL));
4360
4361	if (vd->vdev_children != 0 &&
4362	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
4363		return (spa_vdev_exit(spa, vd, txg, error));
4364
4365	/*
4366	 * We must validate the spares and l2cache devices after checking the
4367	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
4368	 */
4369	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4370		return (spa_vdev_exit(spa, vd, txg, error));
4371
4372	/*
4373	 * Transfer each new top-level vdev from vd to rvd.
4374	 */
4375	for (int c = 0; c < vd->vdev_children; c++) {
4376
4377		/*
4378		 * Set the vdev id to the first hole, if one exists.
4379		 */
4380		for (id = 0; id < rvd->vdev_children; id++) {
4381			if (rvd->vdev_child[id]->vdev_ishole) {
4382				vdev_free(rvd->vdev_child[id]);
4383				break;
4384			}
4385		}
4386		tvd = vd->vdev_child[c];
4387		vdev_remove_child(vd, tvd);
4388		tvd->vdev_id = id;
4389		vdev_add_child(rvd, tvd);
4390		vdev_config_dirty(tvd);
4391	}
4392
4393	if (nspares != 0) {
4394		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4395		    ZPOOL_CONFIG_SPARES);
4396		spa_load_spares(spa);
4397		spa->spa_spares.sav_sync = B_TRUE;
4398	}
4399
4400	if (nl2cache != 0) {
4401		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4402		    ZPOOL_CONFIG_L2CACHE);
4403		spa_load_l2cache(spa);
4404		spa->spa_l2cache.sav_sync = B_TRUE;
4405	}
4406
4407	/*
4408	 * We have to be careful when adding new vdevs to an existing pool.
4409	 * If other threads start allocating from these vdevs before we
4410	 * sync the config cache, and we lose power, then upon reboot we may
4411	 * fail to open the pool because there are DVAs that the config cache
4412	 * can't translate.  Therefore, we first add the vdevs without
4413	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4414	 * and then let spa_config_update() initialize the new metaslabs.
4415	 *
4416	 * spa_load() checks for added-but-not-initialized vdevs, so that
4417	 * if we lose power at any point in this sequence, the remaining
4418	 * steps will be completed the next time we load the pool.
4419	 */
4420	(void) spa_vdev_exit(spa, vd, txg, 0);
4421
4422	mutex_enter(&spa_namespace_lock);
4423	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4424	mutex_exit(&spa_namespace_lock);
4425
4426	return (0);
4427}
4428
4429/*
4430 * Attach a device to a mirror.  The arguments are the path to any device
4431 * in the mirror, and the nvroot for the new device.  If the path specifies
4432 * a device that is not mirrored, we automatically insert the mirror vdev.
4433 *
4434 * If 'replacing' is specified, the new device is intended to replace the
4435 * existing device; in this case the two devices are made into their own
4436 * mirror using the 'replacing' vdev, which is functionally identical to
4437 * the mirror vdev (it actually reuses all the same ops) but has a few
4438 * extra rules: you can't attach to it after it's been created, and upon
4439 * completion of resilvering, the first disk (the one being replaced)
4440 * is automatically detached.
4441 */
4442int
4443spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4444{
4445	uint64_t txg, dtl_max_txg;
4446	vdev_t *rvd = spa->spa_root_vdev;
4447	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4448	vdev_ops_t *pvops;
4449	char *oldvdpath, *newvdpath;
4450	int newvd_isspare;
4451	int error;
4452
4453	ASSERT(spa_writeable(spa));
4454
4455	txg = spa_vdev_enter(spa);
4456
4457	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4458
4459	if (oldvd == NULL)
4460		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4461
4462	if (!oldvd->vdev_ops->vdev_op_leaf)
4463		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4464
4465	pvd = oldvd->vdev_parent;
4466
4467	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4468	    VDEV_ALLOC_ATTACH)) != 0)
4469		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4470
4471	if (newrootvd->vdev_children != 1)
4472		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4473
4474	newvd = newrootvd->vdev_child[0];
4475
4476	if (!newvd->vdev_ops->vdev_op_leaf)
4477		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4478
4479	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4480		return (spa_vdev_exit(spa, newrootvd, txg, error));
4481
4482	/*
4483	 * Spares can't replace logs
4484	 */
4485	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4486		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4487
4488	if (!replacing) {
4489		/*
4490		 * For attach, the only allowable parent is a mirror or the root
4491		 * vdev.
4492		 */
4493		if (pvd->vdev_ops != &vdev_mirror_ops &&
4494		    pvd->vdev_ops != &vdev_root_ops)
4495			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4496
4497		pvops = &vdev_mirror_ops;
4498	} else {
4499		/*
4500		 * Active hot spares can only be replaced by inactive hot
4501		 * spares.
4502		 */
4503		if (pvd->vdev_ops == &vdev_spare_ops &&
4504		    oldvd->vdev_isspare &&
4505		    !spa_has_spare(spa, newvd->vdev_guid))
4506			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4507
4508		/*
4509		 * If the source is a hot spare, and the parent isn't already a
4510		 * spare, then we want to create a new hot spare.  Otherwise, we
4511		 * want to create a replacing vdev.  The user is not allowed to
4512		 * attach to a spared vdev child unless the 'isspare' state is
4513		 * the same (spare replaces spare, non-spare replaces
4514		 * non-spare).
4515		 */
4516		if (pvd->vdev_ops == &vdev_replacing_ops &&
4517		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4518			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4519		} else if (pvd->vdev_ops == &vdev_spare_ops &&
4520		    newvd->vdev_isspare != oldvd->vdev_isspare) {
4521			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4522		}
4523
4524		if (newvd->vdev_isspare)
4525			pvops = &vdev_spare_ops;
4526		else
4527			pvops = &vdev_replacing_ops;
4528	}
4529
4530	/*
4531	 * Make sure the new device is big enough.
4532	 */
4533	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4534		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4535
4536	/*
4537	 * The new device cannot have a higher alignment requirement
4538	 * than the top-level vdev.
4539	 */
4540	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4541		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4542
4543	/*
4544	 * If this is an in-place replacement, update oldvd's path and devid
4545	 * to make it distinguishable from newvd, and unopenable from now on.
4546	 */
4547	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4548		spa_strfree(oldvd->vdev_path);
4549		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4550		    KM_SLEEP);
4551		(void) sprintf(oldvd->vdev_path, "%s/%s",
4552		    newvd->vdev_path, "old");
4553		if (oldvd->vdev_devid != NULL) {
4554			spa_strfree(oldvd->vdev_devid);
4555			oldvd->vdev_devid = NULL;
4556		}
4557	}
4558
4559	/* mark the device being resilvered */
4560	newvd->vdev_resilver_txg = txg;
4561
4562	/*
4563	 * If the parent is not a mirror, or if we're replacing, insert the new
4564	 * mirror/replacing/spare vdev above oldvd.
4565	 */
4566	if (pvd->vdev_ops != pvops)
4567		pvd = vdev_add_parent(oldvd, pvops);
4568
4569	ASSERT(pvd->vdev_top->vdev_parent == rvd);
4570	ASSERT(pvd->vdev_ops == pvops);
4571	ASSERT(oldvd->vdev_parent == pvd);
4572
4573	/*
4574	 * Extract the new device from its root and add it to pvd.
4575	 */
4576	vdev_remove_child(newrootvd, newvd);
4577	newvd->vdev_id = pvd->vdev_children;
4578	newvd->vdev_crtxg = oldvd->vdev_crtxg;
4579	vdev_add_child(pvd, newvd);
4580
4581	tvd = newvd->vdev_top;
4582	ASSERT(pvd->vdev_top == tvd);
4583	ASSERT(tvd->vdev_parent == rvd);
4584
4585	vdev_config_dirty(tvd);
4586
4587	/*
4588	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4589	 * for any dmu_sync-ed blocks.  It will propagate upward when
4590	 * spa_vdev_exit() calls vdev_dtl_reassess().
4591	 */
4592	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4593
4594	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4595	    dtl_max_txg - TXG_INITIAL);
4596
4597	if (newvd->vdev_isspare) {
4598		spa_spare_activate(newvd);
4599		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4600	}
4601
4602	oldvdpath = spa_strdup(oldvd->vdev_path);
4603	newvdpath = spa_strdup(newvd->vdev_path);
4604	newvd_isspare = newvd->vdev_isspare;
4605
4606	/*
4607	 * Mark newvd's DTL dirty in this txg.
4608	 */
4609	vdev_dirty(tvd, VDD_DTL, newvd, txg);
4610
4611	/*
4612	 * Schedule the resilver to restart in the future. We do this to
4613	 * ensure that dmu_sync-ed blocks have been stitched into the
4614	 * respective datasets.
4615	 */
4616	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4617
4618	/*
4619	 * Commit the config
4620	 */
4621	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4622
4623	spa_history_log_internal(spa, "vdev attach", NULL,
4624	    "%s vdev=%s %s vdev=%s",
4625	    replacing && newvd_isspare ? "spare in" :
4626	    replacing ? "replace" : "attach", newvdpath,
4627	    replacing ? "for" : "to", oldvdpath);
4628
4629	spa_strfree(oldvdpath);
4630	spa_strfree(newvdpath);
4631
4632	if (spa->spa_bootfs)
4633		spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4634
4635	return (0);
4636}
4637
4638/*
4639 * Detach a device from a mirror or replacing vdev.
4640 *
4641 * If 'replace_done' is specified, only detach if the parent
4642 * is a replacing vdev.
4643 */
4644int
4645spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4646{
4647	uint64_t txg;
4648	int error;
4649	vdev_t *rvd = spa->spa_root_vdev;
4650	vdev_t *vd, *pvd, *cvd, *tvd;
4651	boolean_t unspare = B_FALSE;
4652	uint64_t unspare_guid = 0;
4653	char *vdpath;
4654
4655	ASSERT(spa_writeable(spa));
4656
4657	txg = spa_vdev_enter(spa);
4658
4659	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4660
4661	if (vd == NULL)
4662		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4663
4664	if (!vd->vdev_ops->vdev_op_leaf)
4665		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4666
4667	pvd = vd->vdev_parent;
4668
4669	/*
4670	 * If the parent/child relationship is not as expected, don't do it.
4671	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4672	 * vdev that's replacing B with C.  The user's intent in replacing
4673	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4674	 * the replace by detaching C, the expected behavior is to end up
4675	 * M(A,B).  But suppose that right after deciding to detach C,
4676	 * the replacement of B completes.  We would have M(A,C), and then
4677	 * ask to detach C, which would leave us with just A -- not what
4678	 * the user wanted.  To prevent this, we make sure that the
4679	 * parent/child relationship hasn't changed -- in this example,
4680	 * that C's parent is still the replacing vdev R.
4681	 */
4682	if (pvd->vdev_guid != pguid && pguid != 0)
4683		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4684
4685	/*
4686	 * Only 'replacing' or 'spare' vdevs can be replaced.
4687	 */
4688	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4689	    pvd->vdev_ops != &vdev_spare_ops)
4690		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4691
4692	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4693	    spa_version(spa) >= SPA_VERSION_SPARES);
4694
4695	/*
4696	 * Only mirror, replacing, and spare vdevs support detach.
4697	 */
4698	if (pvd->vdev_ops != &vdev_replacing_ops &&
4699	    pvd->vdev_ops != &vdev_mirror_ops &&
4700	    pvd->vdev_ops != &vdev_spare_ops)
4701		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4702
4703	/*
4704	 * If this device has the only valid copy of some data,
4705	 * we cannot safely detach it.
4706	 */
4707	if (vdev_dtl_required(vd))
4708		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4709
4710	ASSERT(pvd->vdev_children >= 2);
4711
4712	/*
4713	 * If we are detaching the second disk from a replacing vdev, then
4714	 * check to see if we changed the original vdev's path to have "/old"
4715	 * at the end in spa_vdev_attach().  If so, undo that change now.
4716	 */
4717	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4718	    vd->vdev_path != NULL) {
4719		size_t len = strlen(vd->vdev_path);
4720
4721		for (int c = 0; c < pvd->vdev_children; c++) {
4722			cvd = pvd->vdev_child[c];
4723
4724			if (cvd == vd || cvd->vdev_path == NULL)
4725				continue;
4726
4727			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4728			    strcmp(cvd->vdev_path + len, "/old") == 0) {
4729				spa_strfree(cvd->vdev_path);
4730				cvd->vdev_path = spa_strdup(vd->vdev_path);
4731				break;
4732			}
4733		}
4734	}
4735
4736	/*
4737	 * If we are detaching the original disk from a spare, then it implies
4738	 * that the spare should become a real disk, and be removed from the
4739	 * active spare list for the pool.
4740	 */
4741	if (pvd->vdev_ops == &vdev_spare_ops &&
4742	    vd->vdev_id == 0 &&
4743	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4744		unspare = B_TRUE;
4745
4746	/*
4747	 * Erase the disk labels so the disk can be used for other things.
4748	 * This must be done after all other error cases are handled,
4749	 * but before we disembowel vd (so we can still do I/O to it).
4750	 * But if we can't do it, don't treat the error as fatal --
4751	 * it may be that the unwritability of the disk is the reason
4752	 * it's being detached!
4753	 */
4754	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4755
4756	/*
4757	 * Remove vd from its parent and compact the parent's children.
4758	 */
4759	vdev_remove_child(pvd, vd);
4760	vdev_compact_children(pvd);
4761
4762	/*
4763	 * Remember one of the remaining children so we can get tvd below.
4764	 */
4765	cvd = pvd->vdev_child[pvd->vdev_children - 1];
4766
4767	/*
4768	 * If we need to remove the remaining child from the list of hot spares,
4769	 * do it now, marking the vdev as no longer a spare in the process.
4770	 * We must do this before vdev_remove_parent(), because that can
4771	 * change the GUID if it creates a new toplevel GUID.  For a similar
4772	 * reason, we must remove the spare now, in the same txg as the detach;
4773	 * otherwise someone could attach a new sibling, change the GUID, and
4774	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4775	 */
4776	if (unspare) {
4777		ASSERT(cvd->vdev_isspare);
4778		spa_spare_remove(cvd);
4779		unspare_guid = cvd->vdev_guid;
4780		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4781		cvd->vdev_unspare = B_TRUE;
4782	}
4783
4784	/*
4785	 * If the parent mirror/replacing vdev only has one child,
4786	 * the parent is no longer needed.  Remove it from the tree.
4787	 */
4788	if (pvd->vdev_children == 1) {
4789		if (pvd->vdev_ops == &vdev_spare_ops)
4790			cvd->vdev_unspare = B_FALSE;
4791		vdev_remove_parent(cvd);
4792	}
4793
4794
4795	/*
4796	 * We don't set tvd until now because the parent we just removed
4797	 * may have been the previous top-level vdev.
4798	 */
4799	tvd = cvd->vdev_top;
4800	ASSERT(tvd->vdev_parent == rvd);
4801
4802	/*
4803	 * Reevaluate the parent vdev state.
4804	 */
4805	vdev_propagate_state(cvd);
4806
4807	/*
4808	 * If the 'autoexpand' property is set on the pool then automatically
4809	 * try to expand the size of the pool. For example if the device we
4810	 * just detached was smaller than the others, it may be possible to
4811	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4812	 * first so that we can obtain the updated sizes of the leaf vdevs.
4813	 */
4814	if (spa->spa_autoexpand) {
4815		vdev_reopen(tvd);
4816		vdev_expand(tvd, txg);
4817	}
4818
4819	vdev_config_dirty(tvd);
4820
4821	/*
4822	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4823	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4824	 * But first make sure we're not on any *other* txg's DTL list, to
4825	 * prevent vd from being accessed after it's freed.
4826	 */
4827	vdpath = spa_strdup(vd->vdev_path);
4828	for (int t = 0; t < TXG_SIZE; t++)
4829		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4830	vd->vdev_detached = B_TRUE;
4831	vdev_dirty(tvd, VDD_DTL, vd, txg);
4832
4833	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4834
4835	/* hang on to the spa before we release the lock */
4836	spa_open_ref(spa, FTAG);
4837
4838	error = spa_vdev_exit(spa, vd, txg, 0);
4839
4840	spa_history_log_internal(spa, "detach", NULL,
4841	    "vdev=%s", vdpath);
4842	spa_strfree(vdpath);
4843
4844	/*
4845	 * If this was the removal of the original device in a hot spare vdev,
4846	 * then we want to go through and remove the device from the hot spare
4847	 * list of every other pool.
4848	 */
4849	if (unspare) {
4850		spa_t *altspa = NULL;
4851
4852		mutex_enter(&spa_namespace_lock);
4853		while ((altspa = spa_next(altspa)) != NULL) {
4854			if (altspa->spa_state != POOL_STATE_ACTIVE ||
4855			    altspa == spa)
4856				continue;
4857
4858			spa_open_ref(altspa, FTAG);
4859			mutex_exit(&spa_namespace_lock);
4860			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4861			mutex_enter(&spa_namespace_lock);
4862			spa_close(altspa, FTAG);
4863		}
4864		mutex_exit(&spa_namespace_lock);
4865
4866		/* search the rest of the vdevs for spares to remove */
4867		spa_vdev_resilver_done(spa);
4868	}
4869
4870	/* all done with the spa; OK to release */
4871	mutex_enter(&spa_namespace_lock);
4872	spa_close(spa, FTAG);
4873	mutex_exit(&spa_namespace_lock);
4874
4875	return (error);
4876}
4877
4878/*
4879 * Split a set of devices from their mirrors, and create a new pool from them.
4880 */
4881int
4882spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4883    nvlist_t *props, boolean_t exp)
4884{
4885	int error = 0;
4886	uint64_t txg, *glist;
4887	spa_t *newspa;
4888	uint_t c, children, lastlog;
4889	nvlist_t **child, *nvl, *tmp;
4890	dmu_tx_t *tx;
4891	char *altroot = NULL;
4892	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4893	boolean_t activate_slog;
4894
4895	ASSERT(spa_writeable(spa));
4896
4897	txg = spa_vdev_enter(spa);
4898
4899	/* clear the log and flush everything up to now */
4900	activate_slog = spa_passivate_log(spa);
4901	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4902	error = spa_offline_log(spa);
4903	txg = spa_vdev_config_enter(spa);
4904
4905	if (activate_slog)
4906		spa_activate_log(spa);
4907
4908	if (error != 0)
4909		return (spa_vdev_exit(spa, NULL, txg, error));
4910
4911	/* check new spa name before going any further */
4912	if (spa_lookup(newname) != NULL)
4913		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4914
4915	/*
4916	 * scan through all the children to ensure they're all mirrors
4917	 */
4918	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4919	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4920	    &children) != 0)
4921		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4922
4923	/* first, check to ensure we've got the right child count */
4924	rvd = spa->spa_root_vdev;
4925	lastlog = 0;
4926	for (c = 0; c < rvd->vdev_children; c++) {
4927		vdev_t *vd = rvd->vdev_child[c];
4928
4929		/* don't count the holes & logs as children */
4930		if (vd->vdev_islog || vd->vdev_ishole) {
4931			if (lastlog == 0)
4932				lastlog = c;
4933			continue;
4934		}
4935
4936		lastlog = 0;
4937	}
4938	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4939		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4940
4941	/* next, ensure no spare or cache devices are part of the split */
4942	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4943	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4944		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4945
4946	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4947	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4948
4949	/* then, loop over each vdev and validate it */
4950	for (c = 0; c < children; c++) {
4951		uint64_t is_hole = 0;
4952
4953		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4954		    &is_hole);
4955
4956		if (is_hole != 0) {
4957			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4958			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4959				continue;
4960			} else {
4961				error = SET_ERROR(EINVAL);
4962				break;
4963			}
4964		}
4965
4966		/* which disk is going to be split? */
4967		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4968		    &glist[c]) != 0) {
4969			error = SET_ERROR(EINVAL);
4970			break;
4971		}
4972
4973		/* look it up in the spa */
4974		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4975		if (vml[c] == NULL) {
4976			error = SET_ERROR(ENODEV);
4977			break;
4978		}
4979
4980		/* make sure there's nothing stopping the split */
4981		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4982		    vml[c]->vdev_islog ||
4983		    vml[c]->vdev_ishole ||
4984		    vml[c]->vdev_isspare ||
4985		    vml[c]->vdev_isl2cache ||
4986		    !vdev_writeable(vml[c]) ||
4987		    vml[c]->vdev_children != 0 ||
4988		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4989		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4990			error = SET_ERROR(EINVAL);
4991			break;
4992		}
4993
4994		if (vdev_dtl_required(vml[c])) {
4995			error = SET_ERROR(EBUSY);
4996			break;
4997		}
4998
4999		/* we need certain info from the top level */
5000		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5001		    vml[c]->vdev_top->vdev_ms_array) == 0);
5002		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5003		    vml[c]->vdev_top->vdev_ms_shift) == 0);
5004		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5005		    vml[c]->vdev_top->vdev_asize) == 0);
5006		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5007		    vml[c]->vdev_top->vdev_ashift) == 0);
5008	}
5009
5010	if (error != 0) {
5011		kmem_free(vml, children * sizeof (vdev_t *));
5012		kmem_free(glist, children * sizeof (uint64_t));
5013		return (spa_vdev_exit(spa, NULL, txg, error));
5014	}
5015
5016	/* stop writers from using the disks */
5017	for (c = 0; c < children; c++) {
5018		if (vml[c] != NULL)
5019			vml[c]->vdev_offline = B_TRUE;
5020	}
5021	vdev_reopen(spa->spa_root_vdev);
5022
5023	/*
5024	 * Temporarily record the splitting vdevs in the spa config.  This
5025	 * will disappear once the config is regenerated.
5026	 */
5027	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5028	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5029	    glist, children) == 0);
5030	kmem_free(glist, children * sizeof (uint64_t));
5031
5032	mutex_enter(&spa->spa_props_lock);
5033	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5034	    nvl) == 0);
5035	mutex_exit(&spa->spa_props_lock);
5036	spa->spa_config_splitting = nvl;
5037	vdev_config_dirty(spa->spa_root_vdev);
5038
5039	/* configure and create the new pool */
5040	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5041	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5042	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5043	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5044	    spa_version(spa)) == 0);
5045	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5046	    spa->spa_config_txg) == 0);
5047	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5048	    spa_generate_guid(NULL)) == 0);
5049	(void) nvlist_lookup_string(props,
5050	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5051
5052	/* add the new pool to the namespace */
5053	newspa = spa_add(newname, config, altroot);
5054	newspa->spa_config_txg = spa->spa_config_txg;
5055	spa_set_log_state(newspa, SPA_LOG_CLEAR);
5056
5057	/* release the spa config lock, retaining the namespace lock */
5058	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5059
5060	if (zio_injection_enabled)
5061		zio_handle_panic_injection(spa, FTAG, 1);
5062
5063	spa_activate(newspa, spa_mode_global);
5064	spa_async_suspend(newspa);
5065
5066	/* create the new pool from the disks of the original pool */
5067	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5068	if (error)
5069		goto out;
5070
5071	/* if that worked, generate a real config for the new pool */
5072	if (newspa->spa_root_vdev != NULL) {
5073		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5074		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
5075		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5076		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5077		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5078		    B_TRUE));
5079	}
5080
5081	/* set the props */
5082	if (props != NULL) {
5083		spa_configfile_set(newspa, props, B_FALSE);
5084		error = spa_prop_set(newspa, props);
5085		if (error)
5086			goto out;
5087	}
5088
5089	/* flush everything */
5090	txg = spa_vdev_config_enter(newspa);
5091	vdev_config_dirty(newspa->spa_root_vdev);
5092	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5093
5094	if (zio_injection_enabled)
5095		zio_handle_panic_injection(spa, FTAG, 2);
5096
5097	spa_async_resume(newspa);
5098
5099	/* finally, update the original pool's config */
5100	txg = spa_vdev_config_enter(spa);
5101	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5102	error = dmu_tx_assign(tx, TXG_WAIT);
5103	if (error != 0)
5104		dmu_tx_abort(tx);
5105	for (c = 0; c < children; c++) {
5106		if (vml[c] != NULL) {
5107			vdev_split(vml[c]);
5108			if (error == 0)
5109				spa_history_log_internal(spa, "detach", tx,
5110				    "vdev=%s", vml[c]->vdev_path);
5111			vdev_free(vml[c]);
5112		}
5113	}
5114	vdev_config_dirty(spa->spa_root_vdev);
5115	spa->spa_config_splitting = NULL;
5116	nvlist_free(nvl);
5117	if (error == 0)
5118		dmu_tx_commit(tx);
5119	(void) spa_vdev_exit(spa, NULL, txg, 0);
5120
5121	if (zio_injection_enabled)
5122		zio_handle_panic_injection(spa, FTAG, 3);
5123
5124	/* split is complete; log a history record */
5125	spa_history_log_internal(newspa, "split", NULL,
5126	    "from pool %s", spa_name(spa));
5127
5128	kmem_free(vml, children * sizeof (vdev_t *));
5129
5130	/* if we're not going to mount the filesystems in userland, export */
5131	if (exp)
5132		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5133		    B_FALSE, B_FALSE);
5134
5135	return (error);
5136
5137out:
5138	spa_unload(newspa);
5139	spa_deactivate(newspa);
5140	spa_remove(newspa);
5141
5142	txg = spa_vdev_config_enter(spa);
5143
5144	/* re-online all offlined disks */
5145	for (c = 0; c < children; c++) {
5146		if (vml[c] != NULL)
5147			vml[c]->vdev_offline = B_FALSE;
5148	}
5149	vdev_reopen(spa->spa_root_vdev);
5150
5151	nvlist_free(spa->spa_config_splitting);
5152	spa->spa_config_splitting = NULL;
5153	(void) spa_vdev_exit(spa, NULL, txg, error);
5154
5155	kmem_free(vml, children * sizeof (vdev_t *));
5156	return (error);
5157}
5158
5159static nvlist_t *
5160spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5161{
5162	for (int i = 0; i < count; i++) {
5163		uint64_t guid;
5164
5165		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5166		    &guid) == 0);
5167
5168		if (guid == target_guid)
5169			return (nvpp[i]);
5170	}
5171
5172	return (NULL);
5173}
5174
5175static void
5176spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5177	nvlist_t *dev_to_remove)
5178{
5179	nvlist_t **newdev = NULL;
5180
5181	if (count > 1)
5182		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5183
5184	for (int i = 0, j = 0; i < count; i++) {
5185		if (dev[i] == dev_to_remove)
5186			continue;
5187		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5188	}
5189
5190	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5191	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5192
5193	for (int i = 0; i < count - 1; i++)
5194		nvlist_free(newdev[i]);
5195
5196	if (count > 1)
5197		kmem_free(newdev, (count - 1) * sizeof (void *));
5198}
5199
5200/*
5201 * Evacuate the device.
5202 */
5203static int
5204spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5205{
5206	uint64_t txg;
5207	int error = 0;
5208
5209	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5210	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5211	ASSERT(vd == vd->vdev_top);
5212
5213	/*
5214	 * Evacuate the device.  We don't hold the config lock as writer
5215	 * since we need to do I/O but we do keep the
5216	 * spa_namespace_lock held.  Once this completes the device
5217	 * should no longer have any blocks allocated on it.
5218	 */
5219	if (vd->vdev_islog) {
5220		if (vd->vdev_stat.vs_alloc != 0)
5221			error = spa_offline_log(spa);
5222	} else {
5223		error = SET_ERROR(ENOTSUP);
5224	}
5225
5226	if (error)
5227		return (error);
5228
5229	/*
5230	 * The evacuation succeeded.  Remove any remaining MOS metadata
5231	 * associated with this vdev, and wait for these changes to sync.
5232	 */
5233	ASSERT0(vd->vdev_stat.vs_alloc);
5234	txg = spa_vdev_config_enter(spa);
5235	vd->vdev_removing = B_TRUE;
5236	vdev_dirty_leaves(vd, VDD_DTL, txg);
5237	vdev_config_dirty(vd);
5238	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5239
5240	return (0);
5241}
5242
5243/*
5244 * Complete the removal by cleaning up the namespace.
5245 */
5246static void
5247spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5248{
5249	vdev_t *rvd = spa->spa_root_vdev;
5250	uint64_t id = vd->vdev_id;
5251	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5252
5253	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5254	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5255	ASSERT(vd == vd->vdev_top);
5256
5257	/*
5258	 * Only remove any devices which are empty.
5259	 */
5260	if (vd->vdev_stat.vs_alloc != 0)
5261		return;
5262
5263	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5264
5265	if (list_link_active(&vd->vdev_state_dirty_node))
5266		vdev_state_clean(vd);
5267	if (list_link_active(&vd->vdev_config_dirty_node))
5268		vdev_config_clean(vd);
5269
5270	vdev_free(vd);
5271
5272	if (last_vdev) {
5273		vdev_compact_children(rvd);
5274	} else {
5275		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5276		vdev_add_child(rvd, vd);
5277	}
5278	vdev_config_dirty(rvd);
5279
5280	/*
5281	 * Reassess the health of our root vdev.
5282	 */
5283	vdev_reopen(rvd);
5284}
5285
5286/*
5287 * Remove a device from the pool -
5288 *
5289 * Removing a device from the vdev namespace requires several steps
5290 * and can take a significant amount of time.  As a result we use
5291 * the spa_vdev_config_[enter/exit] functions which allow us to
5292 * grab and release the spa_config_lock while still holding the namespace
5293 * lock.  During each step the configuration is synced out.
5294 *
5295 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5296 * devices.
5297 */
5298int
5299spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5300{
5301	vdev_t *vd;
5302	metaslab_group_t *mg;
5303	nvlist_t **spares, **l2cache, *nv;
5304	uint64_t txg = 0;
5305	uint_t nspares, nl2cache;
5306	int error = 0;
5307	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5308
5309	ASSERT(spa_writeable(spa));
5310
5311	if (!locked)
5312		txg = spa_vdev_enter(spa);
5313
5314	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5315
5316	if (spa->spa_spares.sav_vdevs != NULL &&
5317	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5318	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5319	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5320		/*
5321		 * Only remove the hot spare if it's not currently in use
5322		 * in this pool.
5323		 */
5324		if (vd == NULL || unspare) {
5325			spa_vdev_remove_aux(spa->spa_spares.sav_config,
5326			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5327			spa_load_spares(spa);
5328			spa->spa_spares.sav_sync = B_TRUE;
5329		} else {
5330			error = SET_ERROR(EBUSY);
5331		}
5332	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
5333	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5334	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5335	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5336		/*
5337		 * Cache devices can always be removed.
5338		 */
5339		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5340		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5341		spa_load_l2cache(spa);
5342		spa->spa_l2cache.sav_sync = B_TRUE;
5343	} else if (vd != NULL && vd->vdev_islog) {
5344		ASSERT(!locked);
5345		ASSERT(vd == vd->vdev_top);
5346
5347		mg = vd->vdev_mg;
5348
5349		/*
5350		 * Stop allocating from this vdev.
5351		 */
5352		metaslab_group_passivate(mg);
5353
5354		/*
5355		 * Wait for the youngest allocations and frees to sync,
5356		 * and then wait for the deferral of those frees to finish.
5357		 */
5358		spa_vdev_config_exit(spa, NULL,
5359		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5360
5361		/*
5362		 * Attempt to evacuate the vdev.
5363		 */
5364		error = spa_vdev_remove_evacuate(spa, vd);
5365
5366		txg = spa_vdev_config_enter(spa);
5367
5368		/*
5369		 * If we couldn't evacuate the vdev, unwind.
5370		 */
5371		if (error) {
5372			metaslab_group_activate(mg);
5373			return (spa_vdev_exit(spa, NULL, txg, error));
5374		}
5375
5376		/*
5377		 * Clean up the vdev namespace.
5378		 */
5379		spa_vdev_remove_from_namespace(spa, vd);
5380
5381	} else if (vd != NULL) {
5382		/*
5383		 * Normal vdevs cannot be removed (yet).
5384		 */
5385		error = SET_ERROR(ENOTSUP);
5386	} else {
5387		/*
5388		 * There is no vdev of any kind with the specified guid.
5389		 */
5390		error = SET_ERROR(ENOENT);
5391	}
5392
5393	if (!locked)
5394		return (spa_vdev_exit(spa, NULL, txg, error));
5395
5396	return (error);
5397}
5398
5399/*
5400 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5401 * currently spared, so we can detach it.
5402 */
5403static vdev_t *
5404spa_vdev_resilver_done_hunt(vdev_t *vd)
5405{
5406	vdev_t *newvd, *oldvd;
5407
5408	for (int c = 0; c < vd->vdev_children; c++) {
5409		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5410		if (oldvd != NULL)
5411			return (oldvd);
5412	}
5413
5414	/*
5415	 * Check for a completed replacement.  We always consider the first
5416	 * vdev in the list to be the oldest vdev, and the last one to be
5417	 * the newest (see spa_vdev_attach() for how that works).  In
5418	 * the case where the newest vdev is faulted, we will not automatically
5419	 * remove it after a resilver completes.  This is OK as it will require
5420	 * user intervention to determine which disk the admin wishes to keep.
5421	 */
5422	if (vd->vdev_ops == &vdev_replacing_ops) {
5423		ASSERT(vd->vdev_children > 1);
5424
5425		newvd = vd->vdev_child[vd->vdev_children - 1];
5426		oldvd = vd->vdev_child[0];
5427
5428		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5429		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5430		    !vdev_dtl_required(oldvd))
5431			return (oldvd);
5432	}
5433
5434	/*
5435	 * Check for a completed resilver with the 'unspare' flag set.
5436	 */
5437	if (vd->vdev_ops == &vdev_spare_ops) {
5438		vdev_t *first = vd->vdev_child[0];
5439		vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5440
5441		if (last->vdev_unspare) {
5442			oldvd = first;
5443			newvd = last;
5444		} else if (first->vdev_unspare) {
5445			oldvd = last;
5446			newvd = first;
5447		} else {
5448			oldvd = NULL;
5449		}
5450
5451		if (oldvd != NULL &&
5452		    vdev_dtl_empty(newvd, DTL_MISSING) &&
5453		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5454		    !vdev_dtl_required(oldvd))
5455			return (oldvd);
5456
5457		/*
5458		 * If there are more than two spares attached to a disk,
5459		 * and those spares are not required, then we want to
5460		 * attempt to free them up now so that they can be used
5461		 * by other pools.  Once we're back down to a single
5462		 * disk+spare, we stop removing them.
5463		 */
5464		if (vd->vdev_children > 2) {
5465			newvd = vd->vdev_child[1];
5466
5467			if (newvd->vdev_isspare && last->vdev_isspare &&
5468			    vdev_dtl_empty(last, DTL_MISSING) &&
5469			    vdev_dtl_empty(last, DTL_OUTAGE) &&
5470			    !vdev_dtl_required(newvd))
5471				return (newvd);
5472		}
5473	}
5474
5475	return (NULL);
5476}
5477
5478static void
5479spa_vdev_resilver_done(spa_t *spa)
5480{
5481	vdev_t *vd, *pvd, *ppvd;
5482	uint64_t guid, sguid, pguid, ppguid;
5483
5484	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5485
5486	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5487		pvd = vd->vdev_parent;
5488		ppvd = pvd->vdev_parent;
5489		guid = vd->vdev_guid;
5490		pguid = pvd->vdev_guid;
5491		ppguid = ppvd->vdev_guid;
5492		sguid = 0;
5493		/*
5494		 * If we have just finished replacing a hot spared device, then
5495		 * we need to detach the parent's first child (the original hot
5496		 * spare) as well.
5497		 */
5498		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5499		    ppvd->vdev_children == 2) {
5500			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5501			sguid = ppvd->vdev_child[1]->vdev_guid;
5502		}
5503		ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5504
5505		spa_config_exit(spa, SCL_ALL, FTAG);
5506		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5507			return;
5508		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5509			return;
5510		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5511	}
5512
5513	spa_config_exit(spa, SCL_ALL, FTAG);
5514}
5515
5516/*
5517 * Update the stored path or FRU for this vdev.
5518 */
5519int
5520spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5521    boolean_t ispath)
5522{
5523	vdev_t *vd;
5524	boolean_t sync = B_FALSE;
5525
5526	ASSERT(spa_writeable(spa));
5527
5528	spa_vdev_state_enter(spa, SCL_ALL);
5529
5530	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5531		return (spa_vdev_state_exit(spa, NULL, ENOENT));
5532
5533	if (!vd->vdev_ops->vdev_op_leaf)
5534		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5535
5536	if (ispath) {
5537		if (strcmp(value, vd->vdev_path) != 0) {
5538			spa_strfree(vd->vdev_path);
5539			vd->vdev_path = spa_strdup(value);
5540			sync = B_TRUE;
5541		}
5542	} else {
5543		if (vd->vdev_fru == NULL) {
5544			vd->vdev_fru = spa_strdup(value);
5545			sync = B_TRUE;
5546		} else if (strcmp(value, vd->vdev_fru) != 0) {
5547			spa_strfree(vd->vdev_fru);
5548			vd->vdev_fru = spa_strdup(value);
5549			sync = B_TRUE;
5550		}
5551	}
5552
5553	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5554}
5555
5556int
5557spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5558{
5559	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5560}
5561
5562int
5563spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5564{
5565	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5566}
5567
5568/*
5569 * ==========================================================================
5570 * SPA Scanning
5571 * ==========================================================================
5572 */
5573
5574int
5575spa_scan_stop(spa_t *spa)
5576{
5577	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5578	if (dsl_scan_resilvering(spa->spa_dsl_pool))
5579		return (SET_ERROR(EBUSY));
5580	return (dsl_scan_cancel(spa->spa_dsl_pool));
5581}
5582
5583int
5584spa_scan(spa_t *spa, pool_scan_func_t func)
5585{
5586	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5587
5588	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5589		return (SET_ERROR(ENOTSUP));
5590
5591	/*
5592	 * If a resilver was requested, but there is no DTL on a
5593	 * writeable leaf device, we have nothing to do.
5594	 */
5595	if (func == POOL_SCAN_RESILVER &&
5596	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5597		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5598		return (0);
5599	}
5600
5601	return (dsl_scan(spa->spa_dsl_pool, func));
5602}
5603
5604/*
5605 * ==========================================================================
5606 * SPA async task processing
5607 * ==========================================================================
5608 */
5609
5610static void
5611spa_async_remove(spa_t *spa, vdev_t *vd)
5612{
5613	if (vd->vdev_remove_wanted) {
5614		vd->vdev_remove_wanted = B_FALSE;
5615		vd->vdev_delayed_close = B_FALSE;
5616		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5617
5618		/*
5619		 * We want to clear the stats, but we don't want to do a full
5620		 * vdev_clear() as that will cause us to throw away
5621		 * degraded/faulted state as well as attempt to reopen the
5622		 * device, all of which is a waste.
5623		 */
5624		vd->vdev_stat.vs_read_errors = 0;
5625		vd->vdev_stat.vs_write_errors = 0;
5626		vd->vdev_stat.vs_checksum_errors = 0;
5627
5628		vdev_state_dirty(vd->vdev_top);
5629	}
5630
5631	for (int c = 0; c < vd->vdev_children; c++)
5632		spa_async_remove(spa, vd->vdev_child[c]);
5633}
5634
5635static void
5636spa_async_probe(spa_t *spa, vdev_t *vd)
5637{
5638	if (vd->vdev_probe_wanted) {
5639		vd->vdev_probe_wanted = B_FALSE;
5640		vdev_reopen(vd);	/* vdev_open() does the actual probe */
5641	}
5642
5643	for (int c = 0; c < vd->vdev_children; c++)
5644		spa_async_probe(spa, vd->vdev_child[c]);
5645}
5646
5647static void
5648spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5649{
5650	sysevent_id_t eid;
5651	nvlist_t *attr;
5652	char *physpath;
5653
5654	if (!spa->spa_autoexpand)
5655		return;
5656
5657	for (int c = 0; c < vd->vdev_children; c++) {
5658		vdev_t *cvd = vd->vdev_child[c];
5659		spa_async_autoexpand(spa, cvd);
5660	}
5661
5662	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5663		return;
5664
5665	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5666	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5667
5668	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5669	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5670
5671	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5672	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5673
5674	nvlist_free(attr);
5675	kmem_free(physpath, MAXPATHLEN);
5676}
5677
5678static void
5679spa_async_thread(spa_t *spa)
5680{
5681	int tasks;
5682
5683	ASSERT(spa->spa_sync_on);
5684
5685	mutex_enter(&spa->spa_async_lock);
5686	tasks = spa->spa_async_tasks;
5687	spa->spa_async_tasks = 0;
5688	mutex_exit(&spa->spa_async_lock);
5689
5690	/*
5691	 * See if the config needs to be updated.
5692	 */
5693	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5694		uint64_t old_space, new_space;
5695
5696		mutex_enter(&spa_namespace_lock);
5697		old_space = metaslab_class_get_space(spa_normal_class(spa));
5698		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5699		new_space = metaslab_class_get_space(spa_normal_class(spa));
5700		mutex_exit(&spa_namespace_lock);
5701
5702		/*
5703		 * If the pool grew as a result of the config update,
5704		 * then log an internal history event.
5705		 */
5706		if (new_space != old_space) {
5707			spa_history_log_internal(spa, "vdev online", NULL,
5708			    "pool '%s' size: %llu(+%llu)",
5709			    spa_name(spa), new_space, new_space - old_space);
5710		}
5711	}
5712
5713	/*
5714	 * See if any devices need to be marked REMOVED.
5715	 */
5716	if (tasks & SPA_ASYNC_REMOVE) {
5717		spa_vdev_state_enter(spa, SCL_NONE);
5718		spa_async_remove(spa, spa->spa_root_vdev);
5719		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5720			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5721		for (int i = 0; i < spa->spa_spares.sav_count; i++)
5722			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5723		(void) spa_vdev_state_exit(spa, NULL, 0);
5724	}
5725
5726	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5727		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5728		spa_async_autoexpand(spa, spa->spa_root_vdev);
5729		spa_config_exit(spa, SCL_CONFIG, FTAG);
5730	}
5731
5732	/*
5733	 * See if any devices need to be probed.
5734	 */
5735	if (tasks & SPA_ASYNC_PROBE) {
5736		spa_vdev_state_enter(spa, SCL_NONE);
5737		spa_async_probe(spa, spa->spa_root_vdev);
5738		(void) spa_vdev_state_exit(spa, NULL, 0);
5739	}
5740
5741	/*
5742	 * If any devices are done replacing, detach them.
5743	 */
5744	if (tasks & SPA_ASYNC_RESILVER_DONE)
5745		spa_vdev_resilver_done(spa);
5746
5747	/*
5748	 * Kick off a resilver.
5749	 */
5750	if (tasks & SPA_ASYNC_RESILVER)
5751		dsl_resilver_restart(spa->spa_dsl_pool, 0);
5752
5753	/*
5754	 * Let the world know that we're done.
5755	 */
5756	mutex_enter(&spa->spa_async_lock);
5757	spa->spa_async_thread = NULL;
5758	cv_broadcast(&spa->spa_async_cv);
5759	mutex_exit(&spa->spa_async_lock);
5760	thread_exit();
5761}
5762
5763void
5764spa_async_suspend(spa_t *spa)
5765{
5766	mutex_enter(&spa->spa_async_lock);
5767	spa->spa_async_suspended++;
5768	while (spa->spa_async_thread != NULL)
5769		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5770	mutex_exit(&spa->spa_async_lock);
5771}
5772
5773void
5774spa_async_resume(spa_t *spa)
5775{
5776	mutex_enter(&spa->spa_async_lock);
5777	ASSERT(spa->spa_async_suspended != 0);
5778	spa->spa_async_suspended--;
5779	mutex_exit(&spa->spa_async_lock);
5780}
5781
5782static boolean_t
5783spa_async_tasks_pending(spa_t *spa)
5784{
5785	uint_t non_config_tasks;
5786	uint_t config_task;
5787	boolean_t config_task_suspended;
5788
5789	non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5790	config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5791	if (spa->spa_ccw_fail_time == 0) {
5792		config_task_suspended = B_FALSE;
5793	} else {
5794		config_task_suspended =
5795		    (gethrtime() - spa->spa_ccw_fail_time) <
5796		    (zfs_ccw_retry_interval * NANOSEC);
5797	}
5798
5799	return (non_config_tasks || (config_task && !config_task_suspended));
5800}
5801
5802static void
5803spa_async_dispatch(spa_t *spa)
5804{
5805	mutex_enter(&spa->spa_async_lock);
5806	if (spa_async_tasks_pending(spa) &&
5807	    !spa->spa_async_suspended &&
5808	    spa->spa_async_thread == NULL &&
5809	    rootdir != NULL)
5810		spa->spa_async_thread = thread_create(NULL, 0,
5811		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5812	mutex_exit(&spa->spa_async_lock);
5813}
5814
5815void
5816spa_async_request(spa_t *spa, int task)
5817{
5818	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5819	mutex_enter(&spa->spa_async_lock);
5820	spa->spa_async_tasks |= task;
5821	mutex_exit(&spa->spa_async_lock);
5822}
5823
5824/*
5825 * ==========================================================================
5826 * SPA syncing routines
5827 * ==========================================================================
5828 */
5829
5830static int
5831bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5832{
5833	bpobj_t *bpo = arg;
5834	bpobj_enqueue(bpo, bp, tx);
5835	return (0);
5836}
5837
5838static int
5839spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5840{
5841	zio_t *zio = arg;
5842
5843	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5844	    zio->io_flags));
5845	return (0);
5846}
5847
5848/*
5849 * Note: this simple function is not inlined to make it easier to dtrace the
5850 * amount of time spent syncing frees.
5851 */
5852static void
5853spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
5854{
5855	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5856	bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
5857	VERIFY(zio_wait(zio) == 0);
5858}
5859
5860/*
5861 * Note: this simple function is not inlined to make it easier to dtrace the
5862 * amount of time spent syncing deferred frees.
5863 */
5864static void
5865spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
5866{
5867	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5868	VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
5869	    spa_free_sync_cb, zio, tx), ==, 0);
5870	VERIFY0(zio_wait(zio));
5871}
5872
5873
5874static void
5875spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5876{
5877	char *packed = NULL;
5878	size_t bufsize;
5879	size_t nvsize = 0;
5880	dmu_buf_t *db;
5881
5882	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5883
5884	/*
5885	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5886	 * information.  This avoids the dmu_buf_will_dirty() path and
5887	 * saves us a pre-read to get data we don't actually care about.
5888	 */
5889	bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5890	packed = kmem_alloc(bufsize, KM_SLEEP);
5891
5892	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5893	    KM_SLEEP) == 0);
5894	bzero(packed + nvsize, bufsize - nvsize);
5895
5896	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5897
5898	kmem_free(packed, bufsize);
5899
5900	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5901	dmu_buf_will_dirty(db, tx);
5902	*(uint64_t *)db->db_data = nvsize;
5903	dmu_buf_rele(db, FTAG);
5904}
5905
5906static void
5907spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5908    const char *config, const char *entry)
5909{
5910	nvlist_t *nvroot;
5911	nvlist_t **list;
5912	int i;
5913
5914	if (!sav->sav_sync)
5915		return;
5916
5917	/*
5918	 * Update the MOS nvlist describing the list of available devices.
5919	 * spa_validate_aux() will have already made sure this nvlist is
5920	 * valid and the vdevs are labeled appropriately.
5921	 */
5922	if (sav->sav_object == 0) {
5923		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5924		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5925		    sizeof (uint64_t), tx);
5926		VERIFY(zap_update(spa->spa_meta_objset,
5927		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5928		    &sav->sav_object, tx) == 0);
5929	}
5930
5931	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5932	if (sav->sav_count == 0) {
5933		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5934	} else {
5935		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5936		for (i = 0; i < sav->sav_count; i++)
5937			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5938			    B_FALSE, VDEV_CONFIG_L2CACHE);
5939		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5940		    sav->sav_count) == 0);
5941		for (i = 0; i < sav->sav_count; i++)
5942			nvlist_free(list[i]);
5943		kmem_free(list, sav->sav_count * sizeof (void *));
5944	}
5945
5946	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5947	nvlist_free(nvroot);
5948
5949	sav->sav_sync = B_FALSE;
5950}
5951
5952static void
5953spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5954{
5955	nvlist_t *config;
5956
5957	if (list_is_empty(&spa->spa_config_dirty_list))
5958		return;
5959
5960	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5961
5962	config = spa_config_generate(spa, spa->spa_root_vdev,
5963	    dmu_tx_get_txg(tx), B_FALSE);
5964
5965	/*
5966	 * If we're upgrading the spa version then make sure that
5967	 * the config object gets updated with the correct version.
5968	 */
5969	if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5970		fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5971		    spa->spa_uberblock.ub_version);
5972
5973	spa_config_exit(spa, SCL_STATE, FTAG);
5974
5975	if (spa->spa_config_syncing)
5976		nvlist_free(spa->spa_config_syncing);
5977	spa->spa_config_syncing = config;
5978
5979	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5980}
5981
5982static void
5983spa_sync_version(void *arg, dmu_tx_t *tx)
5984{
5985	uint64_t *versionp = arg;
5986	uint64_t version = *versionp;
5987	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5988
5989	/*
5990	 * Setting the version is special cased when first creating the pool.
5991	 */
5992	ASSERT(tx->tx_txg != TXG_INITIAL);
5993
5994	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5995	ASSERT(version >= spa_version(spa));
5996
5997	spa->spa_uberblock.ub_version = version;
5998	vdev_config_dirty(spa->spa_root_vdev);
5999	spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6000}
6001
6002/*
6003 * Set zpool properties.
6004 */
6005static void
6006spa_sync_props(void *arg, dmu_tx_t *tx)
6007{
6008	nvlist_t *nvp = arg;
6009	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6010	objset_t *mos = spa->spa_meta_objset;
6011	nvpair_t *elem = NULL;
6012
6013	mutex_enter(&spa->spa_props_lock);
6014