spa.c revision 231aab857f14a3e7a0ed5f2879399c3cd6ae92ea
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 */
27
28/*
29 * SPA: Storage Pool Allocator
30 *
31 * This file contains all the routines used when modifying on-disk SPA state.
32 * This includes opening, importing, destroying, exporting a pool, and syncing a
33 * pool.
34 */
35
36#include <sys/zfs_context.h>
37#include <sys/fm/fs/zfs.h>
38#include <sys/spa_impl.h>
39#include <sys/zio.h>
40#include <sys/zio_checksum.h>
41#include <sys/dmu.h>
42#include <sys/dmu_tx.h>
43#include <sys/zap.h>
44#include <sys/zil.h>
45#include <sys/ddt.h>
46#include <sys/vdev_impl.h>
47#include <sys/metaslab.h>
48#include <sys/metaslab_impl.h>
49#include <sys/uberblock_impl.h>
50#include <sys/txg.h>
51#include <sys/avl.h>
52#include <sys/dmu_traverse.h>
53#include <sys/dmu_objset.h>
54#include <sys/unique.h>
55#include <sys/dsl_pool.h>
56#include <sys/dsl_dataset.h>
57#include <sys/dsl_dir.h>
58#include <sys/dsl_prop.h>
59#include <sys/dsl_synctask.h>
60#include <sys/fs/zfs.h>
61#include <sys/arc.h>
62#include <sys/callb.h>
63#include <sys/systeminfo.h>
64#include <sys/spa_boot.h>
65#include <sys/zfs_ioctl.h>
66#include <sys/dsl_scan.h>
67#include <sys/zfeature.h>
68#include <sys/dsl_destroy.h>
69
70#ifdef	_KERNEL
71#include <sys/bootprops.h>
72#include <sys/callb.h>
73#include <sys/cpupart.h>
74#include <sys/pool.h>
75#include <sys/sysdc.h>
76#include <sys/zone.h>
77#endif	/* _KERNEL */
78
79#include "zfs_prop.h"
80#include "zfs_comutil.h"
81
82/*
83 * The interval, in seconds, at which failed configuration cache file writes
84 * should be retried.
85 */
86static int zfs_ccw_retry_interval = 300;
87
88typedef enum zti_modes {
89	ZTI_MODE_FIXED,			/* value is # of threads (min 1) */
90	ZTI_MODE_BATCH,			/* cpu-intensive; value is ignored */
91	ZTI_MODE_NULL,			/* don't create a taskq */
92	ZTI_NMODES
93} zti_modes_t;
94
95#define	ZTI_P(n, q)	{ ZTI_MODE_FIXED, (n), (q) }
96#define	ZTI_BATCH	{ ZTI_MODE_BATCH, 0, 1 }
97#define	ZTI_NULL	{ ZTI_MODE_NULL, 0, 0 }
98
99#define	ZTI_N(n)	ZTI_P(n, 1)
100#define	ZTI_ONE		ZTI_N(1)
101
102typedef struct zio_taskq_info {
103	zti_modes_t zti_mode;
104	uint_t zti_value;
105	uint_t zti_count;
106} zio_taskq_info_t;
107
108static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
109	"issue", "issue_high", "intr", "intr_high"
110};
111
112/*
113 * This table defines the taskq settings for each ZFS I/O type. When
114 * initializing a pool, we use this table to create an appropriately sized
115 * taskq. Some operations are low volume and therefore have a small, static
116 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
117 * macros. Other operations process a large amount of data; the ZTI_BATCH
118 * macro causes us to create a taskq oriented for throughput. Some operations
119 * are so high frequency and short-lived that the taskq itself can become a a
120 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
121 * additional degree of parallelism specified by the number of threads per-
122 * taskq and the number of taskqs; when dispatching an event in this case, the
123 * particular taskq is chosen at random.
124 *
125 * The different taskq priorities are to handle the different contexts (issue
126 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
127 * need to be handled with minimum delay.
128 */
129const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
130	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
131	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* NULL */
132	{ ZTI_N(8),	ZTI_NULL,	ZTI_P(12, 8),	ZTI_NULL }, /* READ */
133	{ ZTI_BATCH,	ZTI_N(5),	ZTI_N(8),	ZTI_N(5) }, /* WRITE */
134	{ ZTI_P(12, 8),	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* FREE */
135	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* CLAIM */
136	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* IOCTL */
137};
138
139static void spa_sync_version(void *arg, dmu_tx_t *tx);
140static void spa_sync_props(void *arg, dmu_tx_t *tx);
141static boolean_t spa_has_active_shared_spare(spa_t *spa);
142static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
143    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
144    char **ereport);
145static void spa_vdev_resilver_done(spa_t *spa);
146
147uint_t		zio_taskq_batch_pct = 75;	/* 1 thread per cpu in pset */
148id_t		zio_taskq_psrset_bind = PS_NONE;
149boolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
150uint_t		zio_taskq_basedc = 80;		/* base duty cycle */
151
152boolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
153extern int	zfs_sync_pass_deferred_free;
154
155/*
156 * This (illegal) pool name is used when temporarily importing a spa_t in order
157 * to get the vdev stats associated with the imported devices.
158 */
159#define	TRYIMPORT_NAME	"$import"
160
161/*
162 * ==========================================================================
163 * SPA properties routines
164 * ==========================================================================
165 */
166
167/*
168 * Add a (source=src, propname=propval) list to an nvlist.
169 */
170static void
171spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
172    uint64_t intval, zprop_source_t src)
173{
174	const char *propname = zpool_prop_to_name(prop);
175	nvlist_t *propval;
176
177	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
178	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
179
180	if (strval != NULL)
181		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
182	else
183		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
184
185	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
186	nvlist_free(propval);
187}
188
189/*
190 * Get property values from the spa configuration.
191 */
192static void
193spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
194{
195	vdev_t *rvd = spa->spa_root_vdev;
196	dsl_pool_t *pool = spa->spa_dsl_pool;
197	uint64_t size, alloc, cap, version;
198	zprop_source_t src = ZPROP_SRC_NONE;
199	spa_config_dirent_t *dp;
200	metaslab_class_t *mc = spa_normal_class(spa);
201
202	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
203
204	if (rvd != NULL) {
205		alloc = metaslab_class_get_alloc(spa_normal_class(spa));
206		size = metaslab_class_get_space(spa_normal_class(spa));
207		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
208		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
209		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
210		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
211		    size - alloc, src);
212
213		spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
214		    metaslab_class_fragmentation(mc), src);
215		spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
216		    metaslab_class_expandable_space(mc), src);
217		spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
218		    (spa_mode(spa) == FREAD), src);
219
220		cap = (size == 0) ? 0 : (alloc * 100 / size);
221		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
222
223		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
224		    ddt_get_pool_dedup_ratio(spa), src);
225
226		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
227		    rvd->vdev_state, src);
228
229		version = spa_version(spa);
230		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
231			src = ZPROP_SRC_DEFAULT;
232		else
233			src = ZPROP_SRC_LOCAL;
234		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
235	}
236
237	if (pool != NULL) {
238		/*
239		 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
240		 * when opening pools before this version freedir will be NULL.
241		 */
242		if (pool->dp_free_dir != NULL) {
243			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
244			    pool->dp_free_dir->dd_phys->dd_used_bytes, src);
245		} else {
246			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
247			    NULL, 0, src);
248		}
249
250		if (pool->dp_leak_dir != NULL) {
251			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
252			    pool->dp_leak_dir->dd_phys->dd_used_bytes, src);
253		} else {
254			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
255			    NULL, 0, src);
256		}
257	}
258
259	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
260
261	if (spa->spa_comment != NULL) {
262		spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
263		    0, ZPROP_SRC_LOCAL);
264	}
265
266	if (spa->spa_root != NULL)
267		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
268		    0, ZPROP_SRC_LOCAL);
269
270	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
271		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
272		    MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
273	} else {
274		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
275		    SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
276	}
277
278	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
279		if (dp->scd_path == NULL) {
280			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
281			    "none", 0, ZPROP_SRC_LOCAL);
282		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
283			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
284			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
285		}
286	}
287}
288
289/*
290 * Get zpool property values.
291 */
292int
293spa_prop_get(spa_t *spa, nvlist_t **nvp)
294{
295	objset_t *mos = spa->spa_meta_objset;
296	zap_cursor_t zc;
297	zap_attribute_t za;
298	int err;
299
300	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
301
302	mutex_enter(&spa->spa_props_lock);
303
304	/*
305	 * Get properties from the spa config.
306	 */
307	spa_prop_get_config(spa, nvp);
308
309	/* If no pool property object, no more prop to get. */
310	if (mos == NULL || spa->spa_pool_props_object == 0) {
311		mutex_exit(&spa->spa_props_lock);
312		return (0);
313	}
314
315	/*
316	 * Get properties from the MOS pool property object.
317	 */
318	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
319	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
320	    zap_cursor_advance(&zc)) {
321		uint64_t intval = 0;
322		char *strval = NULL;
323		zprop_source_t src = ZPROP_SRC_DEFAULT;
324		zpool_prop_t prop;
325
326		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
327			continue;
328
329		switch (za.za_integer_length) {
330		case 8:
331			/* integer property */
332			if (za.za_first_integer !=
333			    zpool_prop_default_numeric(prop))
334				src = ZPROP_SRC_LOCAL;
335
336			if (prop == ZPOOL_PROP_BOOTFS) {
337				dsl_pool_t *dp;
338				dsl_dataset_t *ds = NULL;
339
340				dp = spa_get_dsl(spa);
341				dsl_pool_config_enter(dp, FTAG);
342				if (err = dsl_dataset_hold_obj(dp,
343				    za.za_first_integer, FTAG, &ds)) {
344					dsl_pool_config_exit(dp, FTAG);
345					break;
346				}
347
348				strval = kmem_alloc(
349				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
350				    KM_SLEEP);
351				dsl_dataset_name(ds, strval);
352				dsl_dataset_rele(ds, FTAG);
353				dsl_pool_config_exit(dp, FTAG);
354			} else {
355				strval = NULL;
356				intval = za.za_first_integer;
357			}
358
359			spa_prop_add_list(*nvp, prop, strval, intval, src);
360
361			if (strval != NULL)
362				kmem_free(strval,
363				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
364
365			break;
366
367		case 1:
368			/* string property */
369			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
370			err = zap_lookup(mos, spa->spa_pool_props_object,
371			    za.za_name, 1, za.za_num_integers, strval);
372			if (err) {
373				kmem_free(strval, za.za_num_integers);
374				break;
375			}
376			spa_prop_add_list(*nvp, prop, strval, 0, src);
377			kmem_free(strval, za.za_num_integers);
378			break;
379
380		default:
381			break;
382		}
383	}
384	zap_cursor_fini(&zc);
385	mutex_exit(&spa->spa_props_lock);
386out:
387	if (err && err != ENOENT) {
388		nvlist_free(*nvp);
389		*nvp = NULL;
390		return (err);
391	}
392
393	return (0);
394}
395
396/*
397 * Validate the given pool properties nvlist and modify the list
398 * for the property values to be set.
399 */
400static int
401spa_prop_validate(spa_t *spa, nvlist_t *props)
402{
403	nvpair_t *elem;
404	int error = 0, reset_bootfs = 0;
405	uint64_t objnum = 0;
406	boolean_t has_feature = B_FALSE;
407
408	elem = NULL;
409	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
410		uint64_t intval;
411		char *strval, *slash, *check, *fname;
412		const char *propname = nvpair_name(elem);
413		zpool_prop_t prop = zpool_name_to_prop(propname);
414
415		switch (prop) {
416		case ZPROP_INVAL:
417			if (!zpool_prop_feature(propname)) {
418				error = SET_ERROR(EINVAL);
419				break;
420			}
421
422			/*
423			 * Sanitize the input.
424			 */
425			if (nvpair_type(elem) != DATA_TYPE_UINT64) {
426				error = SET_ERROR(EINVAL);
427				break;
428			}
429
430			if (nvpair_value_uint64(elem, &intval) != 0) {
431				error = SET_ERROR(EINVAL);
432				break;
433			}
434
435			if (intval != 0) {
436				error = SET_ERROR(EINVAL);
437				break;
438			}
439
440			fname = strchr(propname, '@') + 1;
441			if (zfeature_lookup_name(fname, NULL) != 0) {
442				error = SET_ERROR(EINVAL);
443				break;
444			}
445
446			has_feature = B_TRUE;
447			break;
448
449		case ZPOOL_PROP_VERSION:
450			error = nvpair_value_uint64(elem, &intval);
451			if (!error &&
452			    (intval < spa_version(spa) ||
453			    intval > SPA_VERSION_BEFORE_FEATURES ||
454			    has_feature))
455				error = SET_ERROR(EINVAL);
456			break;
457
458		case ZPOOL_PROP_DELEGATION:
459		case ZPOOL_PROP_AUTOREPLACE:
460		case ZPOOL_PROP_LISTSNAPS:
461		case ZPOOL_PROP_AUTOEXPAND:
462			error = nvpair_value_uint64(elem, &intval);
463			if (!error && intval > 1)
464				error = SET_ERROR(EINVAL);
465			break;
466
467		case ZPOOL_PROP_BOOTFS:
468			/*
469			 * If the pool version is less than SPA_VERSION_BOOTFS,
470			 * or the pool is still being created (version == 0),
471			 * the bootfs property cannot be set.
472			 */
473			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
474				error = SET_ERROR(ENOTSUP);
475				break;
476			}
477
478			/*
479			 * Make sure the vdev config is bootable
480			 */
481			if (!vdev_is_bootable(spa->spa_root_vdev)) {
482				error = SET_ERROR(ENOTSUP);
483				break;
484			}
485
486			reset_bootfs = 1;
487
488			error = nvpair_value_string(elem, &strval);
489
490			if (!error) {
491				objset_t *os;
492				uint64_t propval;
493
494				if (strval == NULL || strval[0] == '\0') {
495					objnum = zpool_prop_default_numeric(
496					    ZPOOL_PROP_BOOTFS);
497					break;
498				}
499
500				if (error = dmu_objset_hold(strval, FTAG, &os))
501					break;
502
503				/*
504				 * Must be ZPL, and its property settings
505				 * must be supported by GRUB (compression
506				 * is not gzip, and large blocks are not used).
507				 */
508
509				if (dmu_objset_type(os) != DMU_OST_ZFS) {
510					error = SET_ERROR(ENOTSUP);
511				} else if ((error =
512				    dsl_prop_get_int_ds(dmu_objset_ds(os),
513				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
514				    &propval)) == 0 &&
515				    !BOOTFS_COMPRESS_VALID(propval)) {
516					error = SET_ERROR(ENOTSUP);
517				} else if ((error =
518				    dsl_prop_get_int_ds(dmu_objset_ds(os),
519				    zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
520				    &propval)) == 0 &&
521				    propval > SPA_OLD_MAXBLOCKSIZE) {
522					error = SET_ERROR(ENOTSUP);
523				} else {
524					objnum = dmu_objset_id(os);
525				}
526				dmu_objset_rele(os, FTAG);
527			}
528			break;
529
530		case ZPOOL_PROP_FAILUREMODE:
531			error = nvpair_value_uint64(elem, &intval);
532			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
533			    intval > ZIO_FAILURE_MODE_PANIC))
534				error = SET_ERROR(EINVAL);
535
536			/*
537			 * This is a special case which only occurs when
538			 * the pool has completely failed. This allows
539			 * the user to change the in-core failmode property
540			 * without syncing it out to disk (I/Os might
541			 * currently be blocked). We do this by returning
542			 * EIO to the caller (spa_prop_set) to trick it
543			 * into thinking we encountered a property validation
544			 * error.
545			 */
546			if (!error && spa_suspended(spa)) {
547				spa->spa_failmode = intval;
548				error = SET_ERROR(EIO);
549			}
550			break;
551
552		case ZPOOL_PROP_CACHEFILE:
553			if ((error = nvpair_value_string(elem, &strval)) != 0)
554				break;
555
556			if (strval[0] == '\0')
557				break;
558
559			if (strcmp(strval, "none") == 0)
560				break;
561
562			if (strval[0] != '/') {
563				error = SET_ERROR(EINVAL);
564				break;
565			}
566
567			slash = strrchr(strval, '/');
568			ASSERT(slash != NULL);
569
570			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
571			    strcmp(slash, "/..") == 0)
572				error = SET_ERROR(EINVAL);
573			break;
574
575		case ZPOOL_PROP_COMMENT:
576			if ((error = nvpair_value_string(elem, &strval)) != 0)
577				break;
578			for (check = strval; *check != '\0'; check++) {
579				/*
580				 * The kernel doesn't have an easy isprint()
581				 * check.  For this kernel check, we merely
582				 * check ASCII apart from DEL.  Fix this if
583				 * there is an easy-to-use kernel isprint().
584				 */
585				if (*check >= 0x7f) {
586					error = SET_ERROR(EINVAL);
587					break;
588				}
589				check++;
590			}
591			if (strlen(strval) > ZPROP_MAX_COMMENT)
592				error = E2BIG;
593			break;
594
595		case ZPOOL_PROP_DEDUPDITTO:
596			if (spa_version(spa) < SPA_VERSION_DEDUP)
597				error = SET_ERROR(ENOTSUP);
598			else
599				error = nvpair_value_uint64(elem, &intval);
600			if (error == 0 &&
601			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
602				error = SET_ERROR(EINVAL);
603			break;
604		}
605
606		if (error)
607			break;
608	}
609
610	if (!error && reset_bootfs) {
611		error = nvlist_remove(props,
612		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
613
614		if (!error) {
615			error = nvlist_add_uint64(props,
616			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
617		}
618	}
619
620	return (error);
621}
622
623void
624spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
625{
626	char *cachefile;
627	spa_config_dirent_t *dp;
628
629	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
630	    &cachefile) != 0)
631		return;
632
633	dp = kmem_alloc(sizeof (spa_config_dirent_t),
634	    KM_SLEEP);
635
636	if (cachefile[0] == '\0')
637		dp->scd_path = spa_strdup(spa_config_path);
638	else if (strcmp(cachefile, "none") == 0)
639		dp->scd_path = NULL;
640	else
641		dp->scd_path = spa_strdup(cachefile);
642
643	list_insert_head(&spa->spa_config_list, dp);
644	if (need_sync)
645		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
646}
647
648int
649spa_prop_set(spa_t *spa, nvlist_t *nvp)
650{
651	int error;
652	nvpair_t *elem = NULL;
653	boolean_t need_sync = B_FALSE;
654
655	if ((error = spa_prop_validate(spa, nvp)) != 0)
656		return (error);
657
658	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
659		zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
660
661		if (prop == ZPOOL_PROP_CACHEFILE ||
662		    prop == ZPOOL_PROP_ALTROOT ||
663		    prop == ZPOOL_PROP_READONLY)
664			continue;
665
666		if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
667			uint64_t ver;
668
669			if (prop == ZPOOL_PROP_VERSION) {
670				VERIFY(nvpair_value_uint64(elem, &ver) == 0);
671			} else {
672				ASSERT(zpool_prop_feature(nvpair_name(elem)));
673				ver = SPA_VERSION_FEATURES;
674				need_sync = B_TRUE;
675			}
676
677			/* Save time if the version is already set. */
678			if (ver == spa_version(spa))
679				continue;
680
681			/*
682			 * In addition to the pool directory object, we might
683			 * create the pool properties object, the features for
684			 * read object, the features for write object, or the
685			 * feature descriptions object.
686			 */
687			error = dsl_sync_task(spa->spa_name, NULL,
688			    spa_sync_version, &ver,
689			    6, ZFS_SPACE_CHECK_RESERVED);
690			if (error)
691				return (error);
692			continue;
693		}
694
695		need_sync = B_TRUE;
696		break;
697	}
698
699	if (need_sync) {
700		return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
701		    nvp, 6, ZFS_SPACE_CHECK_RESERVED));
702	}
703
704	return (0);
705}
706
707/*
708 * If the bootfs property value is dsobj, clear it.
709 */
710void
711spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
712{
713	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
714		VERIFY(zap_remove(spa->spa_meta_objset,
715		    spa->spa_pool_props_object,
716		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
717		spa->spa_bootfs = 0;
718	}
719}
720
721/*ARGSUSED*/
722static int
723spa_change_guid_check(void *arg, dmu_tx_t *tx)
724{
725	uint64_t *newguid = arg;
726	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
727	vdev_t *rvd = spa->spa_root_vdev;
728	uint64_t vdev_state;
729
730	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
731	vdev_state = rvd->vdev_state;
732	spa_config_exit(spa, SCL_STATE, FTAG);
733
734	if (vdev_state != VDEV_STATE_HEALTHY)
735		return (SET_ERROR(ENXIO));
736
737	ASSERT3U(spa_guid(spa), !=, *newguid);
738
739	return (0);
740}
741
742static void
743spa_change_guid_sync(void *arg, dmu_tx_t *tx)
744{
745	uint64_t *newguid = arg;
746	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
747	uint64_t oldguid;
748	vdev_t *rvd = spa->spa_root_vdev;
749
750	oldguid = spa_guid(spa);
751
752	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
753	rvd->vdev_guid = *newguid;
754	rvd->vdev_guid_sum += (*newguid - oldguid);
755	vdev_config_dirty(rvd);
756	spa_config_exit(spa, SCL_STATE, FTAG);
757
758	spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
759	    oldguid, *newguid);
760}
761
762/*
763 * Change the GUID for the pool.  This is done so that we can later
764 * re-import a pool built from a clone of our own vdevs.  We will modify
765 * the root vdev's guid, our own pool guid, and then mark all of our
766 * vdevs dirty.  Note that we must make sure that all our vdevs are
767 * online when we do this, or else any vdevs that weren't present
768 * would be orphaned from our pool.  We are also going to issue a
769 * sysevent to update any watchers.
770 */
771int
772spa_change_guid(spa_t *spa)
773{
774	int error;
775	uint64_t guid;
776
777	mutex_enter(&spa->spa_vdev_top_lock);
778	mutex_enter(&spa_namespace_lock);
779	guid = spa_generate_guid(NULL);
780
781	error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
782	    spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
783
784	if (error == 0) {
785		spa_config_sync(spa, B_FALSE, B_TRUE);
786		spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
787	}
788
789	mutex_exit(&spa_namespace_lock);
790	mutex_exit(&spa->spa_vdev_top_lock);
791
792	return (error);
793}
794
795/*
796 * ==========================================================================
797 * SPA state manipulation (open/create/destroy/import/export)
798 * ==========================================================================
799 */
800
801static int
802spa_error_entry_compare(const void *a, const void *b)
803{
804	spa_error_entry_t *sa = (spa_error_entry_t *)a;
805	spa_error_entry_t *sb = (spa_error_entry_t *)b;
806	int ret;
807
808	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
809	    sizeof (zbookmark_phys_t));
810
811	if (ret < 0)
812		return (-1);
813	else if (ret > 0)
814		return (1);
815	else
816		return (0);
817}
818
819/*
820 * Utility function which retrieves copies of the current logs and
821 * re-initializes them in the process.
822 */
823void
824spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
825{
826	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
827
828	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
829	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
830
831	avl_create(&spa->spa_errlist_scrub,
832	    spa_error_entry_compare, sizeof (spa_error_entry_t),
833	    offsetof(spa_error_entry_t, se_avl));
834	avl_create(&spa->spa_errlist_last,
835	    spa_error_entry_compare, sizeof (spa_error_entry_t),
836	    offsetof(spa_error_entry_t, se_avl));
837}
838
839static void
840spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
841{
842	const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
843	enum zti_modes mode = ztip->zti_mode;
844	uint_t value = ztip->zti_value;
845	uint_t count = ztip->zti_count;
846	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
847	char name[32];
848	uint_t flags = 0;
849	boolean_t batch = B_FALSE;
850
851	if (mode == ZTI_MODE_NULL) {
852		tqs->stqs_count = 0;
853		tqs->stqs_taskq = NULL;
854		return;
855	}
856
857	ASSERT3U(count, >, 0);
858
859	tqs->stqs_count = count;
860	tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
861
862	switch (mode) {
863	case ZTI_MODE_FIXED:
864		ASSERT3U(value, >=, 1);
865		value = MAX(value, 1);
866		break;
867
868	case ZTI_MODE_BATCH:
869		batch = B_TRUE;
870		flags |= TASKQ_THREADS_CPU_PCT;
871		value = zio_taskq_batch_pct;
872		break;
873
874	default:
875		panic("unrecognized mode for %s_%s taskq (%u:%u) in "
876		    "spa_activate()",
877		    zio_type_name[t], zio_taskq_types[q], mode, value);
878		break;
879	}
880
881	for (uint_t i = 0; i < count; i++) {
882		taskq_t *tq;
883
884		if (count > 1) {
885			(void) snprintf(name, sizeof (name), "%s_%s_%u",
886			    zio_type_name[t], zio_taskq_types[q], i);
887		} else {
888			(void) snprintf(name, sizeof (name), "%s_%s",
889			    zio_type_name[t], zio_taskq_types[q]);
890		}
891
892		if (zio_taskq_sysdc && spa->spa_proc != &p0) {
893			if (batch)
894				flags |= TASKQ_DC_BATCH;
895
896			tq = taskq_create_sysdc(name, value, 50, INT_MAX,
897			    spa->spa_proc, zio_taskq_basedc, flags);
898		} else {
899			pri_t pri = maxclsyspri;
900			/*
901			 * The write issue taskq can be extremely CPU
902			 * intensive.  Run it at slightly lower priority
903			 * than the other taskqs.
904			 */
905			if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
906				pri--;
907
908			tq = taskq_create_proc(name, value, pri, 50,
909			    INT_MAX, spa->spa_proc, flags);
910		}
911
912		tqs->stqs_taskq[i] = tq;
913	}
914}
915
916static void
917spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
918{
919	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
920
921	if (tqs->stqs_taskq == NULL) {
922		ASSERT0(tqs->stqs_count);
923		return;
924	}
925
926	for (uint_t i = 0; i < tqs->stqs_count; i++) {
927		ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
928		taskq_destroy(tqs->stqs_taskq[i]);
929	}
930
931	kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
932	tqs->stqs_taskq = NULL;
933}
934
935/*
936 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
937 * Note that a type may have multiple discrete taskqs to avoid lock contention
938 * on the taskq itself. In that case we choose which taskq at random by using
939 * the low bits of gethrtime().
940 */
941void
942spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
943    task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
944{
945	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
946	taskq_t *tq;
947
948	ASSERT3P(tqs->stqs_taskq, !=, NULL);
949	ASSERT3U(tqs->stqs_count, !=, 0);
950
951	if (tqs->stqs_count == 1) {
952		tq = tqs->stqs_taskq[0];
953	} else {
954		tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
955	}
956
957	taskq_dispatch_ent(tq, func, arg, flags, ent);
958}
959
960static void
961spa_create_zio_taskqs(spa_t *spa)
962{
963	for (int t = 0; t < ZIO_TYPES; t++) {
964		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
965			spa_taskqs_init(spa, t, q);
966		}
967	}
968}
969
970#ifdef _KERNEL
971static void
972spa_thread(void *arg)
973{
974	callb_cpr_t cprinfo;
975
976	spa_t *spa = arg;
977	user_t *pu = PTOU(curproc);
978
979	CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
980	    spa->spa_name);
981
982	ASSERT(curproc != &p0);
983	(void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
984	    "zpool-%s", spa->spa_name);
985	(void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
986
987	/* bind this thread to the requested psrset */
988	if (zio_taskq_psrset_bind != PS_NONE) {
989		pool_lock();
990		mutex_enter(&cpu_lock);
991		mutex_enter(&pidlock);
992		mutex_enter(&curproc->p_lock);
993
994		if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
995		    0, NULL, NULL) == 0)  {
996			curthread->t_bind_pset = zio_taskq_psrset_bind;
997		} else {
998			cmn_err(CE_WARN,
999			    "Couldn't bind process for zfs pool \"%s\" to "
1000			    "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1001		}
1002
1003		mutex_exit(&curproc->p_lock);
1004		mutex_exit(&pidlock);
1005		mutex_exit(&cpu_lock);
1006		pool_unlock();
1007	}
1008
1009	if (zio_taskq_sysdc) {
1010		sysdc_thread_enter(curthread, 100, 0);
1011	}
1012
1013	spa->spa_proc = curproc;
1014	spa->spa_did = curthread->t_did;
1015
1016	spa_create_zio_taskqs(spa);
1017
1018	mutex_enter(&spa->spa_proc_lock);
1019	ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1020
1021	spa->spa_proc_state = SPA_PROC_ACTIVE;
1022	cv_broadcast(&spa->spa_proc_cv);
1023
1024	CALLB_CPR_SAFE_BEGIN(&cprinfo);
1025	while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1026		cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1027	CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1028
1029	ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1030	spa->spa_proc_state = SPA_PROC_GONE;
1031	spa->spa_proc = &p0;
1032	cv_broadcast(&spa->spa_proc_cv);
1033	CALLB_CPR_EXIT(&cprinfo);	/* drops spa_proc_lock */
1034
1035	mutex_enter(&curproc->p_lock);
1036	lwp_exit();
1037}
1038#endif
1039
1040/*
1041 * Activate an uninitialized pool.
1042 */
1043static void
1044spa_activate(spa_t *spa, int mode)
1045{
1046	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1047
1048	spa->spa_state = POOL_STATE_ACTIVE;
1049	spa->spa_mode = mode;
1050
1051	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1052	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1053
1054	/* Try to create a covering process */
1055	mutex_enter(&spa->spa_proc_lock);
1056	ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1057	ASSERT(spa->spa_proc == &p0);
1058	spa->spa_did = 0;
1059
1060	/* Only create a process if we're going to be around a while. */
1061	if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1062		if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1063		    NULL, 0) == 0) {
1064			spa->spa_proc_state = SPA_PROC_CREATED;
1065			while (spa->spa_proc_state == SPA_PROC_CREATED) {
1066				cv_wait(&spa->spa_proc_cv,
1067				    &spa->spa_proc_lock);
1068			}
1069			ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1070			ASSERT(spa->spa_proc != &p0);
1071			ASSERT(spa->spa_did != 0);
1072		} else {
1073#ifdef _KERNEL
1074			cmn_err(CE_WARN,
1075			    "Couldn't create process for zfs pool \"%s\"\n",
1076			    spa->spa_name);
1077#endif
1078		}
1079	}
1080	mutex_exit(&spa->spa_proc_lock);
1081
1082	/* If we didn't create a process, we need to create our taskqs. */
1083	if (spa->spa_proc == &p0) {
1084		spa_create_zio_taskqs(spa);
1085	}
1086
1087	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1088	    offsetof(vdev_t, vdev_config_dirty_node));
1089	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1090	    offsetof(vdev_t, vdev_state_dirty_node));
1091
1092	txg_list_create(&spa->spa_vdev_txg_list,
1093	    offsetof(struct vdev, vdev_txg_node));
1094
1095	avl_create(&spa->spa_errlist_scrub,
1096	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1097	    offsetof(spa_error_entry_t, se_avl));
1098	avl_create(&spa->spa_errlist_last,
1099	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1100	    offsetof(spa_error_entry_t, se_avl));
1101}
1102
1103/*
1104 * Opposite of spa_activate().
1105 */
1106static void
1107spa_deactivate(spa_t *spa)
1108{
1109	ASSERT(spa->spa_sync_on == B_FALSE);
1110	ASSERT(spa->spa_dsl_pool == NULL);
1111	ASSERT(spa->spa_root_vdev == NULL);
1112	ASSERT(spa->spa_async_zio_root == NULL);
1113	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1114
1115	txg_list_destroy(&spa->spa_vdev_txg_list);
1116
1117	list_destroy(&spa->spa_config_dirty_list);
1118	list_destroy(&spa->spa_state_dirty_list);
1119
1120	for (int t = 0; t < ZIO_TYPES; t++) {
1121		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1122			spa_taskqs_fini(spa, t, q);
1123		}
1124	}
1125
1126	metaslab_class_destroy(spa->spa_normal_class);
1127	spa->spa_normal_class = NULL;
1128
1129	metaslab_class_destroy(spa->spa_log_class);
1130	spa->spa_log_class = NULL;
1131
1132	/*
1133	 * If this was part of an import or the open otherwise failed, we may
1134	 * still have errors left in the queues.  Empty them just in case.
1135	 */
1136	spa_errlog_drain(spa);
1137
1138	avl_destroy(&spa->spa_errlist_scrub);
1139	avl_destroy(&spa->spa_errlist_last);
1140
1141	spa->spa_state = POOL_STATE_UNINITIALIZED;
1142
1143	mutex_enter(&spa->spa_proc_lock);
1144	if (spa->spa_proc_state != SPA_PROC_NONE) {
1145		ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1146		spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1147		cv_broadcast(&spa->spa_proc_cv);
1148		while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1149			ASSERT(spa->spa_proc != &p0);
1150			cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1151		}
1152		ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1153		spa->spa_proc_state = SPA_PROC_NONE;
1154	}
1155	ASSERT(spa->spa_proc == &p0);
1156	mutex_exit(&spa->spa_proc_lock);
1157
1158	/*
1159	 * We want to make sure spa_thread() has actually exited the ZFS
1160	 * module, so that the module can't be unloaded out from underneath
1161	 * it.
1162	 */
1163	if (spa->spa_did != 0) {
1164		thread_join(spa->spa_did);
1165		spa->spa_did = 0;
1166	}
1167}
1168
1169/*
1170 * Verify a pool configuration, and construct the vdev tree appropriately.  This
1171 * will create all the necessary vdevs in the appropriate layout, with each vdev
1172 * in the CLOSED state.  This will prep the pool before open/creation/import.
1173 * All vdev validation is done by the vdev_alloc() routine.
1174 */
1175static int
1176spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1177    uint_t id, int atype)
1178{
1179	nvlist_t **child;
1180	uint_t children;
1181	int error;
1182
1183	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1184		return (error);
1185
1186	if ((*vdp)->vdev_ops->vdev_op_leaf)
1187		return (0);
1188
1189	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1190	    &child, &children);
1191
1192	if (error == ENOENT)
1193		return (0);
1194
1195	if (error) {
1196		vdev_free(*vdp);
1197		*vdp = NULL;
1198		return (SET_ERROR(EINVAL));
1199	}
1200
1201	for (int c = 0; c < children; c++) {
1202		vdev_t *vd;
1203		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1204		    atype)) != 0) {
1205			vdev_free(*vdp);
1206			*vdp = NULL;
1207			return (error);
1208		}
1209	}
1210
1211	ASSERT(*vdp != NULL);
1212
1213	return (0);
1214}
1215
1216/*
1217 * Opposite of spa_load().
1218 */
1219static void
1220spa_unload(spa_t *spa)
1221{
1222	int i;
1223
1224	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1225
1226	/*
1227	 * Stop async tasks.
1228	 */
1229	spa_async_suspend(spa);
1230
1231	/*
1232	 * Stop syncing.
1233	 */
1234	if (spa->spa_sync_on) {
1235		txg_sync_stop(spa->spa_dsl_pool);
1236		spa->spa_sync_on = B_FALSE;
1237	}
1238
1239	/*
1240	 * Wait for any outstanding async I/O to complete.
1241	 */
1242	if (spa->spa_async_zio_root != NULL) {
1243		for (int i = 0; i < max_ncpus; i++)
1244			(void) zio_wait(spa->spa_async_zio_root[i]);
1245		kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1246		spa->spa_async_zio_root = NULL;
1247	}
1248
1249	bpobj_close(&spa->spa_deferred_bpobj);
1250
1251	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1252
1253	/*
1254	 * Close all vdevs.
1255	 */
1256	if (spa->spa_root_vdev)
1257		vdev_free(spa->spa_root_vdev);
1258	ASSERT(spa->spa_root_vdev == NULL);
1259
1260	/*
1261	 * Close the dsl pool.
1262	 */
1263	if (spa->spa_dsl_pool) {
1264		dsl_pool_close(spa->spa_dsl_pool);
1265		spa->spa_dsl_pool = NULL;
1266		spa->spa_meta_objset = NULL;
1267	}
1268
1269	ddt_unload(spa);
1270
1271
1272	/*
1273	 * Drop and purge level 2 cache
1274	 */
1275	spa_l2cache_drop(spa);
1276
1277	for (i = 0; i < spa->spa_spares.sav_count; i++)
1278		vdev_free(spa->spa_spares.sav_vdevs[i]);
1279	if (spa->spa_spares.sav_vdevs) {
1280		kmem_free(spa->spa_spares.sav_vdevs,
1281		    spa->spa_spares.sav_count * sizeof (void *));
1282		spa->spa_spares.sav_vdevs = NULL;
1283	}
1284	if (spa->spa_spares.sav_config) {
1285		nvlist_free(spa->spa_spares.sav_config);
1286		spa->spa_spares.sav_config = NULL;
1287	}
1288	spa->spa_spares.sav_count = 0;
1289
1290	for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1291		vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1292		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1293	}
1294	if (spa->spa_l2cache.sav_vdevs) {
1295		kmem_free(spa->spa_l2cache.sav_vdevs,
1296		    spa->spa_l2cache.sav_count * sizeof (void *));
1297		spa->spa_l2cache.sav_vdevs = NULL;
1298	}
1299	if (spa->spa_l2cache.sav_config) {
1300		nvlist_free(spa->spa_l2cache.sav_config);
1301		spa->spa_l2cache.sav_config = NULL;
1302	}
1303	spa->spa_l2cache.sav_count = 0;
1304
1305	spa->spa_async_suspended = 0;
1306
1307	if (spa->spa_comment != NULL) {
1308		spa_strfree(spa->spa_comment);
1309		spa->spa_comment = NULL;
1310	}
1311
1312	spa_config_exit(spa, SCL_ALL, FTAG);
1313}
1314
1315/*
1316 * Load (or re-load) the current list of vdevs describing the active spares for
1317 * this pool.  When this is called, we have some form of basic information in
1318 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1319 * then re-generate a more complete list including status information.
1320 */
1321static void
1322spa_load_spares(spa_t *spa)
1323{
1324	nvlist_t **spares;
1325	uint_t nspares;
1326	int i;
1327	vdev_t *vd, *tvd;
1328
1329	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1330
1331	/*
1332	 * First, close and free any existing spare vdevs.
1333	 */
1334	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1335		vd = spa->spa_spares.sav_vdevs[i];
1336
1337		/* Undo the call to spa_activate() below */
1338		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1339		    B_FALSE)) != NULL && tvd->vdev_isspare)
1340			spa_spare_remove(tvd);
1341		vdev_close(vd);
1342		vdev_free(vd);
1343	}
1344
1345	if (spa->spa_spares.sav_vdevs)
1346		kmem_free(spa->spa_spares.sav_vdevs,
1347		    spa->spa_spares.sav_count * sizeof (void *));
1348
1349	if (spa->spa_spares.sav_config == NULL)
1350		nspares = 0;
1351	else
1352		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1353		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1354
1355	spa->spa_spares.sav_count = (int)nspares;
1356	spa->spa_spares.sav_vdevs = NULL;
1357
1358	if (nspares == 0)
1359		return;
1360
1361	/*
1362	 * Construct the array of vdevs, opening them to get status in the
1363	 * process.   For each spare, there is potentially two different vdev_t
1364	 * structures associated with it: one in the list of spares (used only
1365	 * for basic validation purposes) and one in the active vdev
1366	 * configuration (if it's spared in).  During this phase we open and
1367	 * validate each vdev on the spare list.  If the vdev also exists in the
1368	 * active configuration, then we also mark this vdev as an active spare.
1369	 */
1370	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1371	    KM_SLEEP);
1372	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1373		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1374		    VDEV_ALLOC_SPARE) == 0);
1375		ASSERT(vd != NULL);
1376
1377		spa->spa_spares.sav_vdevs[i] = vd;
1378
1379		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1380		    B_FALSE)) != NULL) {
1381			if (!tvd->vdev_isspare)
1382				spa_spare_add(tvd);
1383
1384			/*
1385			 * We only mark the spare active if we were successfully
1386			 * able to load the vdev.  Otherwise, importing a pool
1387			 * with a bad active spare would result in strange
1388			 * behavior, because multiple pool would think the spare
1389			 * is actively in use.
1390			 *
1391			 * There is a vulnerability here to an equally bizarre
1392			 * circumstance, where a dead active spare is later
1393			 * brought back to life (onlined or otherwise).  Given
1394			 * the rarity of this scenario, and the extra complexity
1395			 * it adds, we ignore the possibility.
1396			 */
1397			if (!vdev_is_dead(tvd))
1398				spa_spare_activate(tvd);
1399		}
1400
1401		vd->vdev_top = vd;
1402		vd->vdev_aux = &spa->spa_spares;
1403
1404		if (vdev_open(vd) != 0)
1405			continue;
1406
1407		if (vdev_validate_aux(vd) == 0)
1408			spa_spare_add(vd);
1409	}
1410
1411	/*
1412	 * Recompute the stashed list of spares, with status information
1413	 * this time.
1414	 */
1415	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1416	    DATA_TYPE_NVLIST_ARRAY) == 0);
1417
1418	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1419	    KM_SLEEP);
1420	for (i = 0; i < spa->spa_spares.sav_count; i++)
1421		spares[i] = vdev_config_generate(spa,
1422		    spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1423	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1424	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1425	for (i = 0; i < spa->spa_spares.sav_count; i++)
1426		nvlist_free(spares[i]);
1427	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1428}
1429
1430/*
1431 * Load (or re-load) the current list of vdevs describing the active l2cache for
1432 * this pool.  When this is called, we have some form of basic information in
1433 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1434 * then re-generate a more complete list including status information.
1435 * Devices which are already active have their details maintained, and are
1436 * not re-opened.
1437 */
1438static void
1439spa_load_l2cache(spa_t *spa)
1440{
1441	nvlist_t **l2cache;
1442	uint_t nl2cache;
1443	int i, j, oldnvdevs;
1444	uint64_t guid;
1445	vdev_t *vd, **oldvdevs, **newvdevs;
1446	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1447
1448	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1449
1450	if (sav->sav_config != NULL) {
1451		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1452		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1453		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1454	} else {
1455		nl2cache = 0;
1456		newvdevs = NULL;
1457	}
1458
1459	oldvdevs = sav->sav_vdevs;
1460	oldnvdevs = sav->sav_count;
1461	sav->sav_vdevs = NULL;
1462	sav->sav_count = 0;
1463
1464	/*
1465	 * Process new nvlist of vdevs.
1466	 */
1467	for (i = 0; i < nl2cache; i++) {
1468		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1469		    &guid) == 0);
1470
1471		newvdevs[i] = NULL;
1472		for (j = 0; j < oldnvdevs; j++) {
1473			vd = oldvdevs[j];
1474			if (vd != NULL && guid == vd->vdev_guid) {
1475				/*
1476				 * Retain previous vdev for add/remove ops.
1477				 */
1478				newvdevs[i] = vd;
1479				oldvdevs[j] = NULL;
1480				break;
1481			}
1482		}
1483
1484		if (newvdevs[i] == NULL) {
1485			/*
1486			 * Create new vdev
1487			 */
1488			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1489			    VDEV_ALLOC_L2CACHE) == 0);
1490			ASSERT(vd != NULL);
1491			newvdevs[i] = vd;
1492
1493			/*
1494			 * Commit this vdev as an l2cache device,
1495			 * even if it fails to open.
1496			 */
1497			spa_l2cache_add(vd);
1498
1499			vd->vdev_top = vd;
1500			vd->vdev_aux = sav;
1501
1502			spa_l2cache_activate(vd);
1503
1504			if (vdev_open(vd) != 0)
1505				continue;
1506
1507			(void) vdev_validate_aux(vd);
1508
1509			if (!vdev_is_dead(vd))
1510				l2arc_add_vdev(spa, vd);
1511		}
1512	}
1513
1514	/*
1515	 * Purge vdevs that were dropped
1516	 */
1517	for (i = 0; i < oldnvdevs; i++) {
1518		uint64_t pool;
1519
1520		vd = oldvdevs[i];
1521		if (vd != NULL) {
1522			ASSERT(vd->vdev_isl2cache);
1523
1524			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1525			    pool != 0ULL && l2arc_vdev_present(vd))
1526				l2arc_remove_vdev(vd);
1527			vdev_clear_stats(vd);
1528			vdev_free(vd);
1529		}
1530	}
1531
1532	if (oldvdevs)
1533		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1534
1535	if (sav->sav_config == NULL)
1536		goto out;
1537
1538	sav->sav_vdevs = newvdevs;
1539	sav->sav_count = (int)nl2cache;
1540
1541	/*
1542	 * Recompute the stashed list of l2cache devices, with status
1543	 * information this time.
1544	 */
1545	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1546	    DATA_TYPE_NVLIST_ARRAY) == 0);
1547
1548	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1549	for (i = 0; i < sav->sav_count; i++)
1550		l2cache[i] = vdev_config_generate(spa,
1551		    sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1552	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1553	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1554out:
1555	for (i = 0; i < sav->sav_count; i++)
1556		nvlist_free(l2cache[i]);
1557	if (sav->sav_count)
1558		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1559}
1560
1561static int
1562load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1563{
1564	dmu_buf_t *db;
1565	char *packed = NULL;
1566	size_t nvsize = 0;
1567	int error;
1568	*value = NULL;
1569
1570	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1571	nvsize = *(uint64_t *)db->db_data;
1572	dmu_buf_rele(db, FTAG);
1573
1574	packed = kmem_alloc(nvsize, KM_SLEEP);
1575	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1576	    DMU_READ_PREFETCH);
1577	if (error == 0)
1578		error = nvlist_unpack(packed, nvsize, value, 0);
1579	kmem_free(packed, nvsize);
1580
1581	return (error);
1582}
1583
1584/*
1585 * Checks to see if the given vdev could not be opened, in which case we post a
1586 * sysevent to notify the autoreplace code that the device has been removed.
1587 */
1588static void
1589spa_check_removed(vdev_t *vd)
1590{
1591	for (int c = 0; c < vd->vdev_children; c++)
1592		spa_check_removed(vd->vdev_child[c]);
1593
1594	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1595	    !vd->vdev_ishole) {
1596		zfs_post_autoreplace(vd->vdev_spa, vd);
1597		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1598	}
1599}
1600
1601/*
1602 * Validate the current config against the MOS config
1603 */
1604static boolean_t
1605spa_config_valid(spa_t *spa, nvlist_t *config)
1606{
1607	vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1608	nvlist_t *nv;
1609
1610	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1611
1612	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1613	VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1614
1615	ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1616
1617	/*
1618	 * If we're doing a normal import, then build up any additional
1619	 * diagnostic information about missing devices in this config.
1620	 * We'll pass this up to the user for further processing.
1621	 */
1622	if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1623		nvlist_t **child, *nv;
1624		uint64_t idx = 0;
1625
1626		child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1627		    KM_SLEEP);
1628		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1629
1630		for (int c = 0; c < rvd->vdev_children; c++) {
1631			vdev_t *tvd = rvd->vdev_child[c];
1632			vdev_t *mtvd  = mrvd->vdev_child[c];
1633
1634			if (tvd->vdev_ops == &vdev_missing_ops &&
1635			    mtvd->vdev_ops != &vdev_missing_ops &&
1636			    mtvd->vdev_islog)
1637				child[idx++] = vdev_config_generate(spa, mtvd,
1638				    B_FALSE, 0);
1639		}
1640
1641		if (idx) {
1642			VERIFY(nvlist_add_nvlist_array(nv,
1643			    ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1644			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1645			    ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1646
1647			for (int i = 0; i < idx; i++)
1648				nvlist_free(child[i]);
1649		}
1650		nvlist_free(nv);
1651		kmem_free(child, rvd->vdev_children * sizeof (char **));
1652	}
1653
1654	/*
1655	 * Compare the root vdev tree with the information we have
1656	 * from the MOS config (mrvd). Check each top-level vdev
1657	 * with the corresponding MOS config top-level (mtvd).
1658	 */
1659	for (int c = 0; c < rvd->vdev_children; c++) {
1660		vdev_t *tvd = rvd->vdev_child[c];
1661		vdev_t *mtvd  = mrvd->vdev_child[c];
1662
1663		/*
1664		 * Resolve any "missing" vdevs in the current configuration.
1665		 * If we find that the MOS config has more accurate information
1666		 * about the top-level vdev then use that vdev instead.
1667		 */
1668		if (tvd->vdev_ops == &vdev_missing_ops &&
1669		    mtvd->vdev_ops != &vdev_missing_ops) {
1670
1671			if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1672				continue;
1673
1674			/*
1675			 * Device specific actions.
1676			 */
1677			if (mtvd->vdev_islog) {
1678				spa_set_log_state(spa, SPA_LOG_CLEAR);
1679			} else {
1680				/*
1681				 * XXX - once we have 'readonly' pool
1682				 * support we should be able to handle
1683				 * missing data devices by transitioning
1684				 * the pool to readonly.
1685				 */
1686				continue;
1687			}
1688
1689			/*
1690			 * Swap the missing vdev with the data we were
1691			 * able to obtain from the MOS config.
1692			 */
1693			vdev_remove_child(rvd, tvd);
1694			vdev_remove_child(mrvd, mtvd);
1695
1696			vdev_add_child(rvd, mtvd);
1697			vdev_add_child(mrvd, tvd);
1698
1699			spa_config_exit(spa, SCL_ALL, FTAG);
1700			vdev_load(mtvd);
1701			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1702
1703			vdev_reopen(rvd);
1704		} else if (mtvd->vdev_islog) {
1705			/*
1706			 * Load the slog device's state from the MOS config
1707			 * since it's possible that the label does not
1708			 * contain the most up-to-date information.
1709			 */
1710			vdev_load_log_state(tvd, mtvd);
1711			vdev_reopen(tvd);
1712		}
1713	}
1714	vdev_free(mrvd);
1715	spa_config_exit(spa, SCL_ALL, FTAG);
1716
1717	/*
1718	 * Ensure we were able to validate the config.
1719	 */
1720	return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1721}
1722
1723/*
1724 * Check for missing log devices
1725 */
1726static boolean_t
1727spa_check_logs(spa_t *spa)
1728{
1729	boolean_t rv = B_FALSE;
1730
1731	switch (spa->spa_log_state) {
1732	case SPA_LOG_MISSING:
1733		/* need to recheck in case slog has been restored */
1734	case SPA_LOG_UNKNOWN:
1735		rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1736		    NULL, DS_FIND_CHILDREN) != 0);
1737		if (rv)
1738			spa_set_log_state(spa, SPA_LOG_MISSING);
1739		break;
1740	}
1741	return (rv);
1742}
1743
1744static boolean_t
1745spa_passivate_log(spa_t *spa)
1746{
1747	vdev_t *rvd = spa->spa_root_vdev;
1748	boolean_t slog_found = B_FALSE;
1749
1750	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1751
1752	if (!spa_has_slogs(spa))
1753		return (B_FALSE);
1754
1755	for (int c = 0; c < rvd->vdev_children; c++) {
1756		vdev_t *tvd = rvd->vdev_child[c];
1757		metaslab_group_t *mg = tvd->vdev_mg;
1758
1759		if (tvd->vdev_islog) {
1760			metaslab_group_passivate(mg);
1761			slog_found = B_TRUE;
1762		}
1763	}
1764
1765	return (slog_found);
1766}
1767
1768static void
1769spa_activate_log(spa_t *spa)
1770{
1771	vdev_t *rvd = spa->spa_root_vdev;
1772
1773	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1774
1775	for (int c = 0; c < rvd->vdev_children; c++) {
1776		vdev_t *tvd = rvd->vdev_child[c];
1777		metaslab_group_t *mg = tvd->vdev_mg;
1778
1779		if (tvd->vdev_islog)
1780			metaslab_group_activate(mg);
1781	}
1782}
1783
1784int
1785spa_offline_log(spa_t *spa)
1786{
1787	int error;
1788
1789	error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1790	    NULL, DS_FIND_CHILDREN);
1791	if (error == 0) {
1792		/*
1793		 * We successfully offlined the log device, sync out the
1794		 * current txg so that the "stubby" block can be removed
1795		 * by zil_sync().
1796		 */
1797		txg_wait_synced(spa->spa_dsl_pool, 0);
1798	}
1799	return (error);
1800}
1801
1802static void
1803spa_aux_check_removed(spa_aux_vdev_t *sav)
1804{
1805	for (int i = 0; i < sav->sav_count; i++)
1806		spa_check_removed(sav->sav_vdevs[i]);
1807}
1808
1809void
1810spa_claim_notify(zio_t *zio)
1811{
1812	spa_t *spa = zio->io_spa;
1813
1814	if (zio->io_error)
1815		return;
1816
1817	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1818	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1819		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1820	mutex_exit(&spa->spa_props_lock);
1821}
1822
1823typedef struct spa_load_error {
1824	uint64_t	sle_meta_count;
1825	uint64_t	sle_data_count;
1826} spa_load_error_t;
1827
1828static void
1829spa_load_verify_done(zio_t *zio)
1830{
1831	blkptr_t *bp = zio->io_bp;
1832	spa_load_error_t *sle = zio->io_private;
1833	dmu_object_type_t type = BP_GET_TYPE(bp);
1834	int error = zio->io_error;
1835	spa_t *spa = zio->io_spa;
1836
1837	if (error) {
1838		if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1839		    type != DMU_OT_INTENT_LOG)
1840			atomic_inc_64(&sle->sle_meta_count);
1841		else
1842			atomic_inc_64(&sle->sle_data_count);
1843	}
1844	zio_data_buf_free(zio->io_data, zio->io_size);
1845
1846	mutex_enter(&spa->spa_scrub_lock);
1847	spa->spa_scrub_inflight--;
1848	cv_broadcast(&spa->spa_scrub_io_cv);
1849	mutex_exit(&spa->spa_scrub_lock);
1850}
1851
1852/*
1853 * Maximum number of concurrent scrub i/os to create while verifying
1854 * a pool while importing it.
1855 */
1856int spa_load_verify_maxinflight = 10000;
1857boolean_t spa_load_verify_metadata = B_TRUE;
1858boolean_t spa_load_verify_data = B_TRUE;
1859
1860/*ARGSUSED*/
1861static int
1862spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1863    const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1864{
1865	if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
1866		return (0);
1867	/*
1868	 * Note: normally this routine will not be called if
1869	 * spa_load_verify_metadata is not set.  However, it may be useful
1870	 * to manually set the flag after the traversal has begun.
1871	 */
1872	if (!spa_load_verify_metadata)
1873		return (0);
1874	if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data)
1875		return (0);
1876
1877	zio_t *rio = arg;
1878	size_t size = BP_GET_PSIZE(bp);
1879	void *data = zio_data_buf_alloc(size);
1880
1881	mutex_enter(&spa->spa_scrub_lock);
1882	while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
1883		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1884	spa->spa_scrub_inflight++;
1885	mutex_exit(&spa->spa_scrub_lock);
1886
1887	zio_nowait(zio_read(rio, spa, bp, data, size,
1888	    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1889	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1890	    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1891	return (0);
1892}
1893
1894static int
1895spa_load_verify(spa_t *spa)
1896{
1897	zio_t *rio;
1898	spa_load_error_t sle = { 0 };
1899	zpool_rewind_policy_t policy;
1900	boolean_t verify_ok = B_FALSE;
1901	int error = 0;
1902
1903	zpool_get_rewind_policy(spa->spa_config, &policy);
1904
1905	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1906		return (0);
1907
1908	rio = zio_root(spa, NULL, &sle,
1909	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1910
1911	if (spa_load_verify_metadata) {
1912		error = traverse_pool(spa, spa->spa_verify_min_txg,
1913		    TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
1914		    spa_load_verify_cb, rio);
1915	}
1916
1917	(void) zio_wait(rio);
1918
1919	spa->spa_load_meta_errors = sle.sle_meta_count;
1920	spa->spa_load_data_errors = sle.sle_data_count;
1921
1922	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1923	    sle.sle_data_count <= policy.zrp_maxdata) {
1924		int64_t loss = 0;
1925
1926		verify_ok = B_TRUE;
1927		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1928		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1929
1930		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1931		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1932		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1933		VERIFY(nvlist_add_int64(spa->spa_load_info,
1934		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1935		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1936		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1937	} else {
1938		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1939	}
1940
1941	if (error) {
1942		if (error != ENXIO && error != EIO)
1943			error = SET_ERROR(EIO);
1944		return (error);
1945	}
1946
1947	return (verify_ok ? 0 : EIO);
1948}
1949
1950/*
1951 * Find a value in the pool props object.
1952 */
1953static void
1954spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1955{
1956	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1957	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1958}
1959
1960/*
1961 * Find a value in the pool directory object.
1962 */
1963static int
1964spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1965{
1966	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1967	    name, sizeof (uint64_t), 1, val));
1968}
1969
1970static int
1971spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1972{
1973	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1974	return (err);
1975}
1976
1977/*
1978 * Fix up config after a partly-completed split.  This is done with the
1979 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1980 * pool have that entry in their config, but only the splitting one contains
1981 * a list of all the guids of the vdevs that are being split off.
1982 *
1983 * This function determines what to do with that list: either rejoin
1984 * all the disks to the pool, or complete the splitting process.  To attempt
1985 * the rejoin, each disk that is offlined is marked online again, and
1986 * we do a reopen() call.  If the vdev label for every disk that was
1987 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1988 * then we call vdev_split() on each disk, and complete the split.
1989 *
1990 * Otherwise we leave the config alone, with all the vdevs in place in
1991 * the original pool.
1992 */
1993static void
1994spa_try_repair(spa_t *spa, nvlist_t *config)
1995{
1996	uint_t extracted;
1997	uint64_t *glist;
1998	uint_t i, gcount;
1999	nvlist_t *nvl;
2000	vdev_t **vd;
2001	boolean_t attempt_reopen;
2002
2003	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2004		return;
2005
2006	/* check that the config is complete */
2007	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2008	    &glist, &gcount) != 0)
2009		return;
2010
2011	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2012
2013	/* attempt to online all the vdevs & validate */
2014	attempt_reopen = B_TRUE;
2015	for (i = 0; i < gcount; i++) {
2016		if (glist[i] == 0)	/* vdev is hole */
2017			continue;
2018
2019		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2020		if (vd[i] == NULL) {
2021			/*
2022			 * Don't bother attempting to reopen the disks;
2023			 * just do the split.
2024			 */
2025			attempt_reopen = B_FALSE;
2026		} else {
2027			/* attempt to re-online it */
2028			vd[i]->vdev_offline = B_FALSE;
2029		}
2030	}
2031
2032	if (attempt_reopen) {
2033		vdev_reopen(spa->spa_root_vdev);
2034
2035		/* check each device to see what state it's in */
2036		for (extracted = 0, i = 0; i < gcount; i++) {
2037			if (vd[i] != NULL &&
2038			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2039				break;
2040			++extracted;
2041		}
2042	}
2043
2044	/*
2045	 * If every disk has been moved to the new pool, or if we never
2046	 * even attempted to look at them, then we split them off for
2047	 * good.
2048	 */
2049	if (!attempt_reopen || gcount == extracted) {
2050		for (i = 0; i < gcount; i++)
2051			if (vd[i] != NULL)
2052				vdev_split(vd[i]);
2053		vdev_reopen(spa->spa_root_vdev);
2054	}
2055
2056	kmem_free(vd, gcount * sizeof (vdev_t *));
2057}
2058
2059static int
2060spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2061    boolean_t mosconfig)
2062{
2063	nvlist_t *config = spa->spa_config;
2064	char *ereport = FM_EREPORT_ZFS_POOL;
2065	char *comment;
2066	int error;
2067	uint64_t pool_guid;
2068	nvlist_t *nvl;
2069
2070	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2071		return (SET_ERROR(EINVAL));
2072
2073	ASSERT(spa->spa_comment == NULL);
2074	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2075		spa->spa_comment = spa_strdup(comment);
2076
2077	/*
2078	 * Versioning wasn't explicitly added to the label until later, so if
2079	 * it's not present treat it as the initial version.
2080	 */
2081	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2082	    &spa->spa_ubsync.ub_version) != 0)
2083		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2084
2085	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2086	    &spa->spa_config_txg);
2087
2088	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2089	    spa_guid_exists(pool_guid, 0)) {
2090		error = SET_ERROR(EEXIST);
2091	} else {
2092		spa->spa_config_guid = pool_guid;
2093
2094		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2095		    &nvl) == 0) {
2096			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2097			    KM_SLEEP) == 0);
2098		}
2099
2100		nvlist_free(spa->spa_load_info);
2101		spa->spa_load_info = fnvlist_alloc();
2102
2103		gethrestime(&spa->spa_loaded_ts);
2104		error = spa_load_impl(spa, pool_guid, config, state, type,
2105		    mosconfig, &ereport);
2106	}
2107
2108	spa->spa_minref = refcount_count(&spa->spa_refcount);
2109	if (error) {
2110		if (error != EEXIST) {
2111			spa->spa_loaded_ts.tv_sec = 0;
2112			spa->spa_loaded_ts.tv_nsec = 0;
2113		}
2114		if (error != EBADF) {
2115			zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2116		}
2117	}
2118	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2119	spa->spa_ena = 0;
2120
2121	return (error);
2122}
2123
2124/*
2125 * Load an existing storage pool, using the pool's builtin spa_config as a
2126 * source of configuration information.
2127 */
2128static int
2129spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2130    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2131    char **ereport)
2132{
2133	int error = 0;
2134	nvlist_t *nvroot = NULL;
2135	nvlist_t *label;
2136	vdev_t *rvd;
2137	uberblock_t *ub = &spa->spa_uberblock;
2138	uint64_t children, config_cache_txg = spa->spa_config_txg;
2139	int orig_mode = spa->spa_mode;
2140	int parse;
2141	uint64_t obj;
2142	boolean_t missing_feat_write = B_FALSE;
2143
2144	/*
2145	 * If this is an untrusted config, access the pool in read-only mode.
2146	 * This prevents things like resilvering recently removed devices.
2147	 */
2148	if (!mosconfig)
2149		spa->spa_mode = FREAD;
2150
2151	ASSERT(MUTEX_HELD(&spa_namespace_lock));
2152
2153	spa->spa_load_state = state;
2154
2155	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2156		return (SET_ERROR(EINVAL));
2157
2158	parse = (type == SPA_IMPORT_EXISTING ?
2159	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2160
2161	/*
2162	 * Create "The Godfather" zio to hold all async IOs
2163	 */
2164	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2165	    KM_SLEEP);
2166	for (int i = 0; i < max_ncpus; i++) {
2167		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2168		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2169		    ZIO_FLAG_GODFATHER);
2170	}
2171
2172	/*
2173	 * Parse the configuration into a vdev tree.  We explicitly set the
2174	 * value that will be returned by spa_version() since parsing the
2175	 * configuration requires knowing the version number.
2176	 */
2177	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2178	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2179	spa_config_exit(spa, SCL_ALL, FTAG);
2180
2181	if (error != 0)
2182		return (error);
2183
2184	ASSERT(spa->spa_root_vdev == rvd);
2185
2186	if (type != SPA_IMPORT_ASSEMBLE) {
2187		ASSERT(spa_guid(spa) == pool_guid);
2188	}
2189
2190	/*
2191	 * Try to open all vdevs, loading each label in the process.
2192	 */
2193	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2194	error = vdev_open(rvd);
2195	spa_config_exit(spa, SCL_ALL, FTAG);
2196	if (error != 0)
2197		return (error);
2198
2199	/*
2200	 * We need to validate the vdev labels against the configuration that
2201	 * we have in hand, which is dependent on the setting of mosconfig. If
2202	 * mosconfig is true then we're validating the vdev labels based on
2203	 * that config.  Otherwise, we're validating against the cached config
2204	 * (zpool.cache) that was read when we loaded the zfs module, and then
2205	 * later we will recursively call spa_load() and validate against
2206	 * the vdev config.
2207	 *
2208	 * If we're assembling a new pool that's been split off from an
2209	 * existing pool, the labels haven't yet been updated so we skip
2210	 * validation for now.
2211	 */
2212	if (type != SPA_IMPORT_ASSEMBLE) {
2213		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2214		error = vdev_validate(rvd, mosconfig);
2215		spa_config_exit(spa, SCL_ALL, FTAG);
2216
2217		if (error != 0)
2218			return (error);
2219
2220		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2221			return (SET_ERROR(ENXIO));
2222	}
2223
2224	/*
2225	 * Find the best uberblock.
2226	 */
2227	vdev_uberblock_load(rvd, ub, &label);
2228
2229	/*
2230	 * If we weren't able to find a single valid uberblock, return failure.
2231	 */
2232	if (ub->ub_txg == 0) {
2233		nvlist_free(label);
2234		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2235	}
2236
2237	/*
2238	 * If the pool has an unsupported version we can't open it.
2239	 */
2240	if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2241		nvlist_free(label);
2242		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2243	}
2244
2245	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2246		nvlist_t *features;
2247
2248		/*
2249		 * If we weren't able to find what's necessary for reading the
2250		 * MOS in the label, return failure.
2251		 */
2252		if (label == NULL || nvlist_lookup_nvlist(label,
2253		    ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2254			nvlist_free(label);
2255			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2256			    ENXIO));
2257		}
2258
2259		/*
2260		 * Update our in-core representation with the definitive values
2261		 * from the label.
2262		 */
2263		nvlist_free(spa->spa_label_features);
2264		VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2265	}
2266
2267	nvlist_free(label);
2268
2269	/*
2270	 * Look through entries in the label nvlist's features_for_read. If
2271	 * there is a feature listed there which we don't understand then we
2272	 * cannot open a pool.
2273	 */
2274	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2275		nvlist_t *unsup_feat;
2276
2277		VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2278		    0);
2279
2280		for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2281		    NULL); nvp != NULL;
2282		    nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2283			if (!zfeature_is_supported(nvpair_name(nvp))) {
2284				VERIFY(nvlist_add_string(unsup_feat,
2285				    nvpair_name(nvp), "") == 0);
2286			}
2287		}
2288
2289		if (!nvlist_empty(unsup_feat)) {
2290			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2291			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2292			nvlist_free(unsup_feat);
2293			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2294			    ENOTSUP));
2295		}
2296
2297		nvlist_free(unsup_feat);
2298	}
2299
2300	/*
2301	 * If the vdev guid sum doesn't match the uberblock, we have an
2302	 * incomplete configuration.  We first check to see if the pool
2303	 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2304	 * If it is, defer the vdev_guid_sum check till later so we
2305	 * can handle missing vdevs.
2306	 */
2307	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2308	    &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2309	    rvd->vdev_guid_sum != ub->ub_guid_sum)
2310		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2311
2312	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2313		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2314		spa_try_repair(spa, config);
2315		spa_config_exit(spa, SCL_ALL, FTAG);
2316		nvlist_free(spa->spa_config_splitting);
2317		spa->spa_config_splitting = NULL;
2318	}
2319
2320	/*
2321	 * Initialize internal SPA structures.
2322	 */
2323	spa->spa_state = POOL_STATE_ACTIVE;
2324	spa->spa_ubsync = spa->spa_uberblock;
2325	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2326	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2327	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2328	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2329	spa->spa_claim_max_txg = spa->spa_first_txg;
2330	spa->spa_prev_software_version = ub->ub_software_version;
2331
2332	error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2333	if (error)
2334		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2335	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2336
2337	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2338		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2339
2340	if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2341		boolean_t missing_feat_read = B_FALSE;
2342		nvlist_t *unsup_feat, *enabled_feat;
2343
2344		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2345		    &spa->spa_feat_for_read_obj) != 0) {
2346			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2347		}
2348
2349		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2350		    &spa->spa_feat_for_write_obj) != 0) {
2351			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2352		}
2353
2354		if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2355		    &spa->spa_feat_desc_obj) != 0) {
2356			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2357		}
2358
2359		enabled_feat = fnvlist_alloc();
2360		unsup_feat = fnvlist_alloc();
2361
2362		if (!spa_features_check(spa, B_FALSE,
2363		    unsup_feat, enabled_feat))
2364			missing_feat_read = B_TRUE;
2365
2366		if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2367			if (!spa_features_check(spa, B_TRUE,
2368			    unsup_feat, enabled_feat)) {
2369				missing_feat_write = B_TRUE;
2370			}
2371		}
2372
2373		fnvlist_add_nvlist(spa->spa_load_info,
2374		    ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2375
2376		if (!nvlist_empty(unsup_feat)) {
2377			fnvlist_add_nvlist(spa->spa_load_info,
2378			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2379		}
2380
2381		fnvlist_free(enabled_feat);
2382		fnvlist_free(unsup_feat);
2383
2384		if (!missing_feat_read) {
2385			fnvlist_add_boolean(spa->spa_load_info,
2386			    ZPOOL_CONFIG_CAN_RDONLY);
2387		}
2388
2389		/*
2390		 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2391		 * twofold: to determine whether the pool is available for
2392		 * import in read-write mode and (if it is not) whether the
2393		 * pool is available for import in read-only mode. If the pool
2394		 * is available for import in read-write mode, it is displayed
2395		 * as available in userland; if it is not available for import
2396		 * in read-only mode, it is displayed as unavailable in
2397		 * userland. If the pool is available for import in read-only
2398		 * mode but not read-write mode, it is displayed as unavailable
2399		 * in userland with a special note that the pool is actually
2400		 * available for open in read-only mode.
2401		 *
2402		 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2403		 * missing a feature for write, we must first determine whether
2404		 * the pool can be opened read-only before returning to
2405		 * userland in order to know whether to display the
2406		 * abovementioned note.
2407		 */
2408		if (missing_feat_read || (missing_feat_write &&
2409		    spa_writeable(spa))) {
2410			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2411			    ENOTSUP));
2412		}
2413
2414		/*
2415		 * Load refcounts for ZFS features from disk into an in-memory
2416		 * cache during SPA initialization.
2417		 */
2418		for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2419			uint64_t refcount;
2420
2421			error = feature_get_refcount_from_disk(spa,
2422			    &spa_feature_table[i], &refcount);
2423			if (error == 0) {
2424				spa->spa_feat_refcount_cache[i] = refcount;
2425			} else if (error == ENOTSUP) {
2426				spa->spa_feat_refcount_cache[i] =
2427				    SPA_FEATURE_DISABLED;
2428			} else {
2429				return (spa_vdev_err(rvd,
2430				    VDEV_AUX_CORRUPT_DATA, EIO));
2431			}
2432		}
2433	}
2434
2435	if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2436		if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2437		    &spa->spa_feat_enabled_txg_obj) != 0)
2438			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2439	}
2440
2441	spa->spa_is_initializing = B_TRUE;
2442	error = dsl_pool_open(spa->spa_dsl_pool);
2443	spa->spa_is_initializing = B_FALSE;
2444	if (error != 0)
2445		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2446
2447	if (!mosconfig) {
2448		uint64_t hostid;
2449		nvlist_t *policy = NULL, *nvconfig;
2450
2451		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2452			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2453
2454		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2455		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2456			char *hostname;
2457			unsigned long myhostid = 0;
2458
2459			VERIFY(nvlist_lookup_string(nvconfig,
2460			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2461
2462#ifdef	_KERNEL
2463			myhostid = zone_get_hostid(NULL);
2464#else	/* _KERNEL */
2465			/*
2466			 * We're emulating the system's hostid in userland, so
2467			 * we can't use zone_get_hostid().
2468			 */
2469			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2470#endif	/* _KERNEL */
2471			if (hostid != 0 && myhostid != 0 &&
2472			    hostid != myhostid) {
2473				nvlist_free(nvconfig);
2474				cmn_err(CE_WARN, "pool '%s' could not be "
2475				    "loaded as it was last accessed by "
2476				    "another system (host: %s hostid: 0x%lx). "
2477				    "See: http://illumos.org/msg/ZFS-8000-EY",
2478				    spa_name(spa), hostname,
2479				    (unsigned long)hostid);
2480				return (SET_ERROR(EBADF));
2481			}
2482		}
2483		if (nvlist_lookup_nvlist(spa->spa_config,
2484		    ZPOOL_REWIND_POLICY, &policy) == 0)
2485			VERIFY(nvlist_add_nvlist(nvconfig,
2486			    ZPOOL_REWIND_POLICY, policy) == 0);
2487
2488		spa_config_set(spa, nvconfig);
2489		spa_unload(spa);
2490		spa_deactivate(spa);
2491		spa_activate(spa, orig_mode);
2492
2493		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2494	}
2495
2496	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2497		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2498	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2499	if (error != 0)
2500		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2501
2502	/*
2503	 * Load the bit that tells us to use the new accounting function
2504	 * (raid-z deflation).  If we have an older pool, this will not
2505	 * be present.
2506	 */
2507	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2508	if (error != 0 && error != ENOENT)
2509		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2510
2511	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2512	    &spa->spa_creation_version);
2513	if (error != 0 && error != ENOENT)
2514		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2515
2516	/*
2517	 * Load the persistent error log.  If we have an older pool, this will
2518	 * not be present.
2519	 */
2520	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
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_ERRLOG_SCRUB,
2525	    &spa->spa_errlog_scrub);
2526	if (error != 0 && error != ENOENT)
2527		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2528
2529	/*
2530	 * Load the history object.  If we have an older pool, this
2531	 * will not be present.
2532	 */
2533	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2534	if (error != 0 && error != ENOENT)
2535		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2536
2537	/*
2538	 * If we're assembling the pool from the split-off vdevs of
2539	 * an existing pool, we don't want to attach the spares & cache
2540	 * devices.
2541	 */
2542
2543	/*
2544	 * Load any hot spares for this pool.
2545	 */
2546	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2547	if (error != 0 && error != ENOENT)
2548		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2549	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2550		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2551		if (load_nvlist(spa, spa->spa_spares.sav_object,
2552		    &spa->spa_spares.sav_config) != 0)
2553			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2554
2555		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2556		spa_load_spares(spa);
2557		spa_config_exit(spa, SCL_ALL, FTAG);
2558	} else if (error == 0) {
2559		spa->spa_spares.sav_sync = B_TRUE;
2560	}
2561
2562	/*
2563	 * Load any level 2 ARC devices for this pool.
2564	 */
2565	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2566	    &spa->spa_l2cache.sav_object);
2567	if (error != 0 && error != ENOENT)
2568		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2569	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2570		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2571		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2572		    &spa->spa_l2cache.sav_config) != 0)
2573			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2574
2575		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2576		spa_load_l2cache(spa);
2577		spa_config_exit(spa, SCL_ALL, FTAG);
2578	} else if (error == 0) {
2579		spa->spa_l2cache.sav_sync = B_TRUE;
2580	}
2581
2582	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2583
2584	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2585	if (error && error != ENOENT)
2586		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2587
2588	if (error == 0) {
2589		uint64_t autoreplace;
2590
2591		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2592		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2593		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2594		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2595		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2596		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2597		    &spa->spa_dedup_ditto);
2598
2599		spa->spa_autoreplace = (autoreplace != 0);
2600	}
2601
2602	/*
2603	 * If the 'autoreplace' property is set, then post a resource notifying
2604	 * the ZFS DE that it should not issue any faults for unopenable
2605	 * devices.  We also iterate over the vdevs, and post a sysevent for any
2606	 * unopenable vdevs so that the normal autoreplace handler can take
2607	 * over.
2608	 */
2609	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2610		spa_check_removed(spa->spa_root_vdev);
2611		/*
2612		 * For the import case, this is done in spa_import(), because
2613		 * at this point we're using the spare definitions from
2614		 * the MOS config, not necessarily from the userland config.
2615		 */
2616		if (state != SPA_LOAD_IMPORT) {
2617			spa_aux_check_removed(&spa->spa_spares);
2618			spa_aux_check_removed(&spa->spa_l2cache);
2619		}
2620	}
2621
2622	/*
2623	 * Load the vdev state for all toplevel vdevs.
2624	 */
2625	vdev_load(rvd);
2626
2627	/*
2628	 * Propagate the leaf DTLs we just loaded all the way up the tree.
2629	 */
2630	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2631	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2632	spa_config_exit(spa, SCL_ALL, FTAG);
2633
2634	/*
2635	 * Load the DDTs (dedup tables).
2636	 */
2637	error = ddt_load(spa);
2638	if (error != 0)
2639		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2640
2641	spa_update_dspace(spa);
2642
2643	/*
2644	 * Validate the config, using the MOS config to fill in any
2645	 * information which might be missing.  If we fail to validate
2646	 * the config then declare the pool unfit for use. If we're
2647	 * assembling a pool from a split, the log is not transferred
2648	 * over.
2649	 */
2650	if (type != SPA_IMPORT_ASSEMBLE) {
2651		nvlist_t *nvconfig;
2652
2653		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2654			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2655
2656		if (!spa_config_valid(spa, nvconfig)) {
2657			nvlist_free(nvconfig);
2658			return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2659			    ENXIO));
2660		}
2661		nvlist_free(nvconfig);
2662
2663		/*
2664		 * Now that we've validated the config, check the state of the
2665		 * root vdev.  If it can't be opened, it indicates one or
2666		 * more toplevel vdevs are faulted.
2667		 */
2668		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2669			return (SET_ERROR(ENXIO));
2670
2671		if (spa_check_logs(spa)) {
2672			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2673			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2674		}
2675	}
2676
2677	if (missing_feat_write) {
2678		ASSERT(state == SPA_LOAD_TRYIMPORT);
2679
2680		/*
2681		 * At this point, we know that we can open the pool in
2682		 * read-only mode but not read-write mode. We now have enough
2683		 * information and can return to userland.
2684		 */
2685		return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2686	}
2687
2688	/*
2689	 * We've successfully opened the pool, verify that we're ready
2690	 * to start pushing transactions.
2691	 */
2692	if (state != SPA_LOAD_TRYIMPORT) {
2693		if (error = spa_load_verify(spa))
2694			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2695			    error));
2696	}
2697
2698	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2699	    spa->spa_load_max_txg == UINT64_MAX)) {
2700		dmu_tx_t *tx;
2701		int need_update = B_FALSE;
2702
2703		ASSERT(state != SPA_LOAD_TRYIMPORT);
2704
2705		/*
2706		 * Claim log blocks that haven't been committed yet.
2707		 * This must all happen in a single txg.
2708		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2709		 * invoked from zil_claim_log_block()'s i/o done callback.
2710		 * Price of rollback is that we abandon the log.
2711		 */
2712		spa->spa_claiming = B_TRUE;
2713
2714		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2715		    spa_first_txg(spa));
2716		(void) dmu_objset_find(spa_name(spa),
2717		    zil_claim, tx, DS_FIND_CHILDREN);
2718		dmu_tx_commit(tx);
2719
2720		spa->spa_claiming = B_FALSE;
2721
2722		spa_set_log_state(spa, SPA_LOG_GOOD);
2723		spa->spa_sync_on = B_TRUE;
2724		txg_sync_start(spa->spa_dsl_pool);
2725
2726		/*
2727		 * Wait for all claims to sync.  We sync up to the highest
2728		 * claimed log block birth time so that claimed log blocks
2729		 * don't appear to be from the future.  spa_claim_max_txg
2730		 * will have been set for us by either zil_check_log_chain()
2731		 * (invoked from spa_check_logs()) or zil_claim() above.
2732		 */
2733		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2734
2735		/*
2736		 * If the config cache is stale, or we have uninitialized
2737		 * metaslabs (see spa_vdev_add()), then update the config.
2738		 *
2739		 * If this is a verbatim import, trust the current
2740		 * in-core spa_config and update the disk labels.
2741		 */
2742		if (config_cache_txg != spa->spa_config_txg ||
2743		    state == SPA_LOAD_IMPORT ||
2744		    state == SPA_LOAD_RECOVER ||
2745		    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2746			need_update = B_TRUE;
2747
2748		for (int c = 0; c < rvd->vdev_children; c++)
2749			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2750				need_update = B_TRUE;
2751
2752		/*
2753		 * Update the config cache asychronously in case we're the
2754		 * root pool, in which case the config cache isn't writable yet.
2755		 */
2756		if (need_update)
2757			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2758
2759		/*
2760		 * Check all DTLs to see if anything needs resilvering.
2761		 */
2762		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2763		    vdev_resilver_needed(rvd, NULL, NULL))
2764			spa_async_request(spa, SPA_ASYNC_RESILVER);
2765
2766		/*
2767		 * Log the fact that we booted up (so that we can detect if
2768		 * we rebooted in the middle of an operation).
2769		 */
2770		spa_history_log_version(spa, "open");
2771
2772		/*
2773		 * Delete any inconsistent datasets.
2774		 */
2775		(void) dmu_objset_find(spa_name(spa),
2776		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2777
2778		/*
2779		 * Clean up any stale temporary dataset userrefs.
2780		 */
2781		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2782	}
2783
2784	return (0);
2785}
2786
2787static int
2788spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2789{
2790	int mode = spa->spa_mode;
2791
2792	spa_unload(spa);
2793	spa_deactivate(spa);
2794
2795	spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
2796
2797	spa_activate(spa, mode);
2798	spa_async_suspend(spa);
2799
2800	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2801}
2802
2803/*
2804 * If spa_load() fails this function will try loading prior txg's. If
2805 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2806 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2807 * function will not rewind the pool and will return the same error as
2808 * spa_load().
2809 */
2810static int
2811spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2812    uint64_t max_request, int rewind_flags)
2813{
2814	nvlist_t *loadinfo = NULL;
2815	nvlist_t *config = NULL;
2816	int load_error, rewind_error;
2817	uint64_t safe_rewind_txg;
2818	uint64_t min_txg;
2819
2820	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2821		spa->spa_load_max_txg = spa->spa_load_txg;
2822		spa_set_log_state(spa, SPA_LOG_CLEAR);
2823	} else {
2824		spa->spa_load_max_txg = max_request;
2825		if (max_request != UINT64_MAX)
2826			spa->spa_extreme_rewind = B_TRUE;
2827	}
2828
2829	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2830	    mosconfig);
2831	if (load_error == 0)
2832		return (0);
2833
2834	if (spa->spa_root_vdev != NULL)
2835		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2836
2837	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2838	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2839
2840	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2841		nvlist_free(config);
2842		return (load_error);
2843	}
2844
2845	if (state == SPA_LOAD_RECOVER) {
2846		/* Price of rolling back is discarding txgs, including log */
2847		spa_set_log_state(spa, SPA_LOG_CLEAR);
2848	} else {
2849		/*
2850		 * If we aren't rolling back save the load info from our first
2851		 * import attempt so that we can restore it after attempting
2852		 * to rewind.
2853		 */
2854		loadinfo = spa->spa_load_info;
2855		spa->spa_load_info = fnvlist_alloc();
2856	}
2857
2858	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2859	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2860	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2861	    TXG_INITIAL : safe_rewind_txg;
2862
2863	/*
2864	 * Continue as long as we're finding errors, we're still within
2865	 * the acceptable rewind range, and we're still finding uberblocks
2866	 */
2867	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2868	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2869		if (spa->spa_load_max_txg < safe_rewind_txg)
2870			spa->spa_extreme_rewind = B_TRUE;
2871		rewind_error = spa_load_retry(spa, state, mosconfig);
2872	}
2873
2874	spa->spa_extreme_rewind = B_FALSE;
2875	spa->spa_load_max_txg = UINT64_MAX;
2876
2877	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2878		spa_config_set(spa, config);
2879
2880	if (state == SPA_LOAD_RECOVER) {
2881		ASSERT3P(loadinfo, ==, NULL);
2882		return (rewind_error);
2883	} else {
2884		/* Store the rewind info as part of the initial load info */
2885		fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2886		    spa->spa_load_info);
2887
2888		/* Restore the initial load info */
2889		fnvlist_free(spa->spa_load_info);
2890		spa->spa_load_info = loadinfo;
2891
2892		return (load_error);
2893	}
2894}
2895
2896/*
2897 * Pool Open/Import
2898 *
2899 * The import case is identical to an open except that the configuration is sent
2900 * down from userland, instead of grabbed from the configuration cache.  For the
2901 * case of an open, the pool configuration will exist in the
2902 * POOL_STATE_UNINITIALIZED state.
2903 *
2904 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2905 * the same time open the pool, without having to keep around the spa_t in some
2906 * ambiguous state.
2907 */
2908static int
2909spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2910    nvlist_t **config)
2911{
2912	spa_t *spa;
2913	spa_load_state_t state = SPA_LOAD_OPEN;
2914	int error;
2915	int locked = B_FALSE;
2916
2917	*spapp = NULL;
2918
2919	/*
2920	 * As disgusting as this is, we need to support recursive calls to this
2921	 * function because dsl_dir_open() is called during spa_load(), and ends
2922	 * up calling spa_open() again.  The real fix is to figure out how to
2923	 * avoid dsl_dir_open() calling this in the first place.
2924	 */
2925	if (mutex_owner(&spa_namespace_lock) != curthread) {
2926		mutex_enter(&spa_namespace_lock);
2927		locked = B_TRUE;
2928	}
2929
2930	if ((spa = spa_lookup(pool)) == NULL) {
2931		if (locked)
2932			mutex_exit(&spa_namespace_lock);
2933		return (SET_ERROR(ENOENT));
2934	}
2935
2936	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2937		zpool_rewind_policy_t policy;
2938
2939		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2940		    &policy);
2941		if (policy.zrp_request & ZPOOL_DO_REWIND)
2942			state = SPA_LOAD_RECOVER;
2943
2944		spa_activate(spa, spa_mode_global);
2945
2946		if (state != SPA_LOAD_RECOVER)
2947			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2948
2949		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2950		    policy.zrp_request);
2951
2952		if (error == EBADF) {
2953			/*
2954			 * If vdev_validate() returns failure (indicated by
2955			 * EBADF), it indicates that one of the vdevs indicates
2956			 * that the pool has been exported or destroyed.  If
2957			 * this is the case, the config cache is out of sync and
2958			 * we should remove the pool from the namespace.
2959			 */
2960			spa_unload(spa);
2961			spa_deactivate(spa);
2962			spa_config_sync(spa, B_TRUE, B_TRUE);
2963			spa_remove(spa);
2964			if (locked)
2965				mutex_exit(&spa_namespace_lock);
2966			return (SET_ERROR(ENOENT));
2967		}
2968
2969		if (error) {
2970			/*
2971			 * We can't open the pool, but we still have useful
2972			 * information: the state of each vdev after the
2973			 * attempted vdev_open().  Return this to the user.
2974			 */
2975			if (config != NULL && spa->spa_config) {
2976				VERIFY(nvlist_dup(spa->spa_config, config,
2977				    KM_SLEEP) == 0);
2978				VERIFY(nvlist_add_nvlist(*config,
2979				    ZPOOL_CONFIG_LOAD_INFO,
2980				    spa->spa_load_info) == 0);
2981			}
2982			spa_unload(spa);
2983			spa_deactivate(spa);
2984			spa->spa_last_open_failed = error;
2985			if (locked)
2986				mutex_exit(&spa_namespace_lock);
2987			*spapp = NULL;
2988			return (error);
2989		}
2990	}
2991
2992	spa_open_ref(spa, tag);
2993
2994	if (config != NULL)
2995		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2996
2997	/*
2998	 * If we've recovered the pool, pass back any information we
2999	 * gathered while doing the load.
3000	 */
3001	if (state == SPA_LOAD_RECOVER) {
3002		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
3003		    spa->spa_load_info) == 0);
3004	}
3005
3006	if (locked) {
3007		spa->spa_last_open_failed = 0;
3008		spa->spa_last_ubsync_txg = 0;
3009		spa->spa_load_txg = 0;
3010		mutex_exit(&spa_namespace_lock);
3011	}
3012
3013	*spapp = spa;
3014
3015	return (0);
3016}
3017
3018int
3019spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3020    nvlist_t **config)
3021{
3022	return (spa_open_common(name, spapp, tag, policy, config));
3023}
3024
3025int
3026spa_open(const char *name, spa_t **spapp, void *tag)
3027{
3028	return (spa_open_common(name, spapp, tag, NULL, NULL));
3029}
3030
3031/*
3032 * Lookup the given spa_t, incrementing the inject count in the process,
3033 * preventing it from being exported or destroyed.
3034 */
3035spa_t *
3036spa_inject_addref(char *name)
3037{
3038	spa_t *spa;
3039
3040	mutex_enter(&spa_namespace_lock);
3041	if ((spa = spa_lookup(name)) == NULL) {
3042		mutex_exit(&spa_namespace_lock);
3043		return (NULL);
3044	}
3045	spa->spa_inject_ref++;
3046	mutex_exit(&spa_namespace_lock);
3047
3048	return (spa);
3049}
3050
3051void
3052spa_inject_delref(spa_t *spa)
3053{
3054	mutex_enter(&spa_namespace_lock);
3055	spa->spa_inject_ref--;
3056	mutex_exit(&spa_namespace_lock);
3057}
3058
3059/*
3060 * Add spares device information to the nvlist.
3061 */
3062static void
3063spa_add_spares(spa_t *spa, nvlist_t *config)
3064{
3065	nvlist_t **spares;
3066	uint_t i, nspares;
3067	nvlist_t *nvroot;
3068	uint64_t guid;
3069	vdev_stat_t *vs;
3070	uint_t vsc;
3071	uint64_t pool;
3072
3073	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3074
3075	if (spa->spa_spares.sav_count == 0)
3076		return;
3077
3078	VERIFY(nvlist_lookup_nvlist(config,
3079	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3080	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3081	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3082	if (nspares != 0) {
3083		VERIFY(nvlist_add_nvlist_array(nvroot,
3084		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3085		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3086		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3087
3088		/*
3089		 * Go through and find any spares which have since been
3090		 * repurposed as an active spare.  If this is the case, update
3091		 * their status appropriately.
3092		 */
3093		for (i = 0; i < nspares; i++) {
3094			VERIFY(nvlist_lookup_uint64(spares[i],
3095			    ZPOOL_CONFIG_GUID, &guid) == 0);
3096			if (spa_spare_exists(guid, &pool, NULL) &&
3097			    pool != 0ULL) {
3098				VERIFY(nvlist_lookup_uint64_array(
3099				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
3100				    (uint64_t **)&vs, &vsc) == 0);
3101				vs->vs_state = VDEV_STATE_CANT_OPEN;
3102				vs->vs_aux = VDEV_AUX_SPARED;
3103			}
3104		}
3105	}
3106}
3107
3108/*
3109 * Add l2cache device information to the nvlist, including vdev stats.
3110 */
3111static void
3112spa_add_l2cache(spa_t *spa, nvlist_t *config)
3113{
3114	nvlist_t **l2cache;
3115	uint_t i, j, nl2cache;
3116	nvlist_t *nvroot;
3117	uint64_t guid;
3118	vdev_t *vd;
3119	vdev_stat_t *vs;
3120	uint_t vsc;
3121
3122	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3123
3124	if (spa->spa_l2cache.sav_count == 0)
3125		return;
3126
3127	VERIFY(nvlist_lookup_nvlist(config,
3128	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3129	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3130	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3131	if (nl2cache != 0) {
3132		VERIFY(nvlist_add_nvlist_array(nvroot,
3133		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3134		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3135		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3136
3137		/*
3138		 * Update level 2 cache device stats.
3139		 */
3140
3141		for (i = 0; i < nl2cache; i++) {
3142			VERIFY(nvlist_lookup_uint64(l2cache[i],
3143			    ZPOOL_CONFIG_GUID, &guid) == 0);
3144
3145			vd = NULL;
3146			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3147				if (guid ==
3148				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3149					vd = spa->spa_l2cache.sav_vdevs[j];
3150					break;
3151				}
3152			}
3153			ASSERT(vd != NULL);
3154
3155			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3156			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3157			    == 0);
3158			vdev_get_stats(vd, vs);
3159		}
3160	}
3161}
3162
3163static void
3164spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3165{
3166	nvlist_t *features;
3167	zap_cursor_t zc;
3168	zap_attribute_t za;
3169
3170	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3171	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3172
3173	if (spa->spa_feat_for_read_obj != 0) {
3174		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3175		    spa->spa_feat_for_read_obj);
3176		    zap_cursor_retrieve(&zc, &za) == 0;
3177		    zap_cursor_advance(&zc)) {
3178			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3179			    za.za_num_integers == 1);
3180			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3181			    za.za_first_integer));
3182		}
3183		zap_cursor_fini(&zc);
3184	}
3185
3186	if (spa->spa_feat_for_write_obj != 0) {
3187		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3188		    spa->spa_feat_for_write_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	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3200	    features) == 0);
3201	nvlist_free(features);
3202}
3203
3204int
3205spa_get_stats(const char *name, nvlist_t **config,
3206    char *altroot, size_t buflen)
3207{
3208	int error;
3209	spa_t *spa;
3210
3211	*config = NULL;
3212	error = spa_open_common(name, &spa, FTAG, NULL, config);
3213
3214	if (spa != NULL) {
3215		/*
3216		 * This still leaves a window of inconsistency where the spares
3217		 * or l2cache devices could change and the config would be
3218		 * self-inconsistent.
3219		 */
3220		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3221
3222		if (*config != NULL) {
3223			uint64_t loadtimes[2];
3224
3225			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3226			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3227			VERIFY(nvlist_add_uint64_array(*config,
3228			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3229
3230			VERIFY(nvlist_add_uint64(*config,
3231			    ZPOOL_CONFIG_ERRCOUNT,
3232			    spa_get_errlog_size(spa)) == 0);
3233
3234			if (spa_suspended(spa))
3235				VERIFY(nvlist_add_uint64(*config,
3236				    ZPOOL_CONFIG_SUSPENDED,
3237				    spa->spa_failmode) == 0);
3238
3239			spa_add_spares(spa, *config);
3240			spa_add_l2cache(spa, *config);
3241			spa_add_feature_stats(spa, *config);
3242		}
3243	}
3244
3245	/*
3246	 * We want to get the alternate root even for faulted pools, so we cheat
3247	 * and call spa_lookup() directly.
3248	 */
3249	if (altroot) {
3250		if (spa == NULL) {
3251			mutex_enter(&spa_namespace_lock);
3252			spa = spa_lookup(name);
3253			if (spa)
3254				spa_altroot(spa, altroot, buflen);
3255			else
3256				altroot[0] = '\0';
3257			spa = NULL;
3258			mutex_exit(&spa_namespace_lock);
3259		} else {
3260			spa_altroot(spa, altroot, buflen);
3261		}
3262	}
3263
3264	if (spa != NULL) {
3265		spa_config_exit(spa, SCL_CONFIG, FTAG);
3266		spa_close(spa, FTAG);
3267	}
3268
3269	return (error);
3270}
3271
3272/*
3273 * Validate that the auxiliary device array is well formed.  We must have an
3274 * array of nvlists, each which describes a valid leaf vdev.  If this is an
3275 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3276 * specified, as long as they are well-formed.
3277 */
3278static int
3279spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3280    spa_aux_vdev_t *sav, const char *config, uint64_t version,
3281    vdev_labeltype_t label)
3282{
3283	nvlist_t **dev;
3284	uint_t i, ndev;
3285	vdev_t *vd;
3286	int error;
3287
3288	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3289
3290	/*
3291	 * It's acceptable to have no devs specified.
3292	 */
3293	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3294		return (0);
3295
3296	if (ndev == 0)
3297		return (SET_ERROR(EINVAL));
3298
3299	/*
3300	 * Make sure the pool is formatted with a version that supports this
3301	 * device type.
3302	 */
3303	if (spa_version(spa) < version)
3304		return (SET_ERROR(ENOTSUP));
3305
3306	/*
3307	 * Set the pending device list so we correctly handle device in-use
3308	 * checking.
3309	 */
3310	sav->sav_pending = dev;
3311	sav->sav_npending = ndev;
3312
3313	for (i = 0; i < ndev; i++) {
3314		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3315		    mode)) != 0)
3316			goto out;
3317
3318		if (!vd->vdev_ops->vdev_op_leaf) {
3319			vdev_free(vd);
3320			error = SET_ERROR(EINVAL);
3321			goto out;
3322		}
3323
3324		/*
3325		 * The L2ARC currently only supports disk devices in
3326		 * kernel context.  For user-level testing, we allow it.
3327		 */
3328#ifdef _KERNEL
3329		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3330		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3331			error = SET_ERROR(ENOTBLK);
3332			vdev_free(vd);
3333			goto out;
3334		}
3335#endif
3336		vd->vdev_top = vd;
3337
3338		if ((error = vdev_open(vd)) == 0 &&
3339		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
3340			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3341			    vd->vdev_guid) == 0);
3342		}
3343
3344		vdev_free(vd);
3345
3346		if (error &&
3347		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3348			goto out;
3349		else
3350			error = 0;
3351	}
3352
3353out:
3354	sav->sav_pending = NULL;
3355	sav->sav_npending = 0;
3356	return (error);
3357}
3358
3359static int
3360spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3361{
3362	int error;
3363
3364	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3365
3366	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3367	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3368	    VDEV_LABEL_SPARE)) != 0) {
3369		return (error);
3370	}
3371
3372	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3373	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3374	    VDEV_LABEL_L2CACHE));
3375}
3376
3377static void
3378spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3379    const char *config)
3380{
3381	int i;
3382
3383	if (sav->sav_config != NULL) {
3384		nvlist_t **olddevs;
3385		uint_t oldndevs;
3386		nvlist_t **newdevs;
3387
3388		/*
3389		 * Generate new dev list by concatentating with the
3390		 * current dev list.
3391		 */
3392		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3393		    &olddevs, &oldndevs) == 0);
3394
3395		newdevs = kmem_alloc(sizeof (void *) *
3396		    (ndevs + oldndevs), KM_SLEEP);
3397		for (i = 0; i < oldndevs; i++)
3398			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3399			    KM_SLEEP) == 0);
3400		for (i = 0; i < ndevs; i++)
3401			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3402			    KM_SLEEP) == 0);
3403
3404		VERIFY(nvlist_remove(sav->sav_config, config,
3405		    DATA_TYPE_NVLIST_ARRAY) == 0);
3406
3407		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3408		    config, newdevs, ndevs + oldndevs) == 0);
3409		for (i = 0; i < oldndevs + ndevs; i++)
3410			nvlist_free(newdevs[i]);
3411		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3412	} else {
3413		/*
3414		 * Generate a new dev list.
3415		 */
3416		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3417		    KM_SLEEP) == 0);
3418		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3419		    devs, ndevs) == 0);
3420	}
3421}
3422
3423/*
3424 * Stop and drop level 2 ARC devices
3425 */
3426void
3427spa_l2cache_drop(spa_t *spa)
3428{
3429	vdev_t *vd;
3430	int i;
3431	spa_aux_vdev_t *sav = &spa->spa_l2cache;
3432
3433	for (i = 0; i < sav->sav_count; i++) {
3434		uint64_t pool;
3435
3436		vd = sav->sav_vdevs[i];
3437		ASSERT(vd != NULL);
3438
3439		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3440		    pool != 0ULL && l2arc_vdev_present(vd))
3441			l2arc_remove_vdev(vd);
3442	}
3443}
3444
3445/*
3446 * Pool Creation
3447 */
3448int
3449spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3450    nvlist_t *zplprops)
3451{
3452	spa_t *spa;
3453	char *altroot = NULL;
3454	vdev_t *rvd;
3455	dsl_pool_t *dp;
3456	dmu_tx_t *tx;
3457	int error = 0;
3458	uint64_t txg = TXG_INITIAL;
3459	nvlist_t **spares, **l2cache;
3460	uint_t nspares, nl2cache;
3461	uint64_t version, obj;
3462	boolean_t has_features;
3463
3464	/*
3465	 * If this pool already exists, return failure.
3466	 */
3467	mutex_enter(&spa_namespace_lock);
3468	if (spa_lookup(pool) != NULL) {
3469		mutex_exit(&spa_namespace_lock);
3470		return (SET_ERROR(EEXIST));
3471	}
3472
3473	/*
3474	 * Allocate a new spa_t structure.
3475	 */
3476	(void) nvlist_lookup_string(props,
3477	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3478	spa = spa_add(pool, NULL, altroot);
3479	spa_activate(spa, spa_mode_global);
3480
3481	if (props && (error = spa_prop_validate(spa, props))) {
3482		spa_deactivate(spa);
3483		spa_remove(spa);
3484		mutex_exit(&spa_namespace_lock);
3485		return (error);
3486	}
3487
3488	has_features = B_FALSE;
3489	for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3490	    elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3491		if (zpool_prop_feature(nvpair_name(elem)))
3492			has_features = B_TRUE;
3493	}
3494
3495	if (has_features || nvlist_lookup_uint64(props,
3496	    zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3497		version = SPA_VERSION;
3498	}
3499	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3500
3501	spa->spa_first_txg = txg;
3502	spa->spa_uberblock.ub_txg = txg - 1;
3503	spa->spa_uberblock.ub_version = version;
3504	spa->spa_ubsync = spa->spa_uberblock;
3505
3506	/*
3507	 * Create "The Godfather" zio to hold all async IOs
3508	 */
3509	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
3510	    KM_SLEEP);
3511	for (int i = 0; i < max_ncpus; i++) {
3512		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
3513		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
3514		    ZIO_FLAG_GODFATHER);
3515	}
3516
3517	/*
3518	 * Create the root vdev.
3519	 */
3520	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3521
3522	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3523
3524	ASSERT(error != 0 || rvd != NULL);
3525	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3526
3527	if (error == 0 && !zfs_allocatable_devs(nvroot))
3528		error = SET_ERROR(EINVAL);
3529
3530	if (error == 0 &&
3531	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3532	    (error = spa_validate_aux(spa, nvroot, txg,
3533	    VDEV_ALLOC_ADD)) == 0) {
3534		for (int c = 0; c < rvd->vdev_children; c++) {
3535			vdev_metaslab_set_size(rvd->vdev_child[c]);
3536			vdev_expand(rvd->vdev_child[c], txg);
3537		}
3538	}
3539
3540	spa_config_exit(spa, SCL_ALL, FTAG);
3541
3542	if (error != 0) {
3543		spa_unload(spa);
3544		spa_deactivate(spa);
3545		spa_remove(spa);
3546		mutex_exit(&spa_namespace_lock);
3547		return (error);
3548	}
3549
3550	/*
3551	 * Get the list of spares, if specified.
3552	 */
3553	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3554	    &spares, &nspares) == 0) {
3555		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3556		    KM_SLEEP) == 0);
3557		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3558		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3559		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3560		spa_load_spares(spa);
3561		spa_config_exit(spa, SCL_ALL, FTAG);
3562		spa->spa_spares.sav_sync = B_TRUE;
3563	}
3564
3565	/*
3566	 * Get the list of level 2 cache devices, if specified.
3567	 */
3568	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3569	    &l2cache, &nl2cache) == 0) {
3570		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3571		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3572		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3573		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3574		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3575		spa_load_l2cache(spa);
3576		spa_config_exit(spa, SCL_ALL, FTAG);
3577		spa->spa_l2cache.sav_sync = B_TRUE;
3578	}
3579
3580	spa->spa_is_initializing = B_TRUE;
3581	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3582	spa->spa_meta_objset = dp->dp_meta_objset;
3583	spa->spa_is_initializing = B_FALSE;
3584
3585	/*
3586	 * Create DDTs (dedup tables).
3587	 */
3588	ddt_create(spa);
3589
3590	spa_update_dspace(spa);
3591
3592	tx = dmu_tx_create_assigned(dp, txg);
3593
3594	/*
3595	 * Create the pool config object.
3596	 */
3597	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3598	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3599	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3600
3601	if (zap_add(spa->spa_meta_objset,
3602	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3603	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3604		cmn_err(CE_PANIC, "failed to add pool config");
3605	}
3606
3607	if (spa_version(spa) >= SPA_VERSION_FEATURES)
3608		spa_feature_create_zap_objects(spa, tx);
3609
3610	if (zap_add(spa->spa_meta_objset,
3611	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3612	    sizeof (uint64_t), 1, &version, tx) != 0) {
3613		cmn_err(CE_PANIC, "failed to add pool version");
3614	}
3615
3616	/* Newly created pools with the right version are always deflated. */
3617	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3618		spa->spa_deflate = TRUE;
3619		if (zap_add(spa->spa_meta_objset,
3620		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3621		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3622			cmn_err(CE_PANIC, "failed to add deflate");
3623		}
3624	}
3625
3626	/*
3627	 * Create the deferred-free bpobj.  Turn off compression
3628	 * because sync-to-convergence takes longer if the blocksize
3629	 * keeps changing.
3630	 */
3631	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3632	dmu_object_set_compress(spa->spa_meta_objset, obj,
3633	    ZIO_COMPRESS_OFF, tx);
3634	if (zap_add(spa->spa_meta_objset,
3635	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3636	    sizeof (uint64_t), 1, &obj, tx) != 0) {
3637		cmn_err(CE_PANIC, "failed to add bpobj");
3638	}
3639	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3640	    spa->spa_meta_objset, obj));
3641
3642	/*
3643	 * Create the pool's history object.
3644	 */
3645	if (version >= SPA_VERSION_ZPOOL_HISTORY)
3646		spa_history_create_obj(spa, tx);
3647
3648	/*
3649	 * Set pool properties.
3650	 */
3651	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3652	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3653	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3654	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3655
3656	if (props != NULL) {
3657		spa_configfile_set(spa, props, B_FALSE);
3658		spa_sync_props(props, tx);
3659	}
3660
3661	dmu_tx_commit(tx);
3662
3663	spa->spa_sync_on = B_TRUE;
3664	txg_sync_start(spa->spa_dsl_pool);
3665
3666	/*
3667	 * We explicitly wait for the first transaction to complete so that our
3668	 * bean counters are appropriately updated.
3669	 */
3670	txg_wait_synced(spa->spa_dsl_pool, txg);
3671
3672	spa_config_sync(spa, B_FALSE, B_TRUE);
3673
3674	spa_history_log_version(spa, "create");
3675
3676	spa->spa_minref = refcount_count(&spa->spa_refcount);
3677
3678	mutex_exit(&spa_namespace_lock);
3679
3680	return (0);
3681}
3682
3683#ifdef _KERNEL
3684/*
3685 * Get the root pool information from the root disk, then import the root pool
3686 * during the system boot up time.
3687 */
3688extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3689
3690static nvlist_t *
3691spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3692{
3693	nvlist_t *config;
3694	nvlist_t *nvtop, *nvroot;
3695	uint64_t pgid;
3696
3697	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3698		return (NULL);
3699
3700	/*
3701	 * Add this top-level vdev to the child array.
3702	 */
3703	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3704	    &nvtop) == 0);
3705	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3706	    &pgid) == 0);
3707	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3708
3709	/*
3710	 * Put this pool's top-level vdevs into a root vdev.
3711	 */
3712	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3713	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3714	    VDEV_TYPE_ROOT) == 0);
3715	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3716	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3717	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3718	    &nvtop, 1) == 0);
3719
3720	/*
3721	 * Replace the existing vdev_tree with the new root vdev in
3722	 * this pool's configuration (remove the old, add the new).
3723	 */
3724	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3725	nvlist_free(nvroot);
3726	return (config);
3727}
3728
3729/*
3730 * Walk the vdev tree and see if we can find a device with "better"
3731 * configuration. A configuration is "better" if the label on that
3732 * device has a more recent txg.
3733 */
3734static void
3735spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3736{
3737	for (int c = 0; c < vd->vdev_children; c++)
3738		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3739
3740	if (vd->vdev_ops->vdev_op_leaf) {
3741		nvlist_t *label;
3742		uint64_t label_txg;
3743
3744		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3745		    &label) != 0)
3746			return;
3747
3748		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3749		    &label_txg) == 0);
3750
3751		/*
3752		 * Do we have a better boot device?
3753		 */
3754		if (label_txg > *txg) {
3755			*txg = label_txg;
3756			*avd = vd;
3757		}
3758		nvlist_free(label);
3759	}
3760}
3761
3762/*
3763 * Import a root pool.
3764 *
3765 * For x86. devpath_list will consist of devid and/or physpath name of
3766 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3767 * The GRUB "findroot" command will return the vdev we should boot.
3768 *
3769 * For Sparc, devpath_list consists the physpath name of the booting device
3770 * no matter the rootpool is a single device pool or a mirrored pool.
3771 * e.g.
3772 *	"/pci@1f,0/ide@d/disk@0,0:a"
3773 */
3774int
3775spa_import_rootpool(char *devpath, char *devid)
3776{
3777	spa_t *spa;
3778	vdev_t *rvd, *bvd, *avd = NULL;
3779	nvlist_t *config, *nvtop;
3780	uint64_t guid, txg;
3781	char *pname;
3782	int error;
3783
3784	/*
3785	 * Read the label from the boot device and generate a configuration.
3786	 */
3787	config = spa_generate_rootconf(devpath, devid, &guid);
3788#if defined(_OBP) && defined(_KERNEL)
3789	if (config == NULL) {
3790		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3791			/* iscsi boot */
3792			get_iscsi_bootpath_phy(devpath);
3793			config = spa_generate_rootconf(devpath, devid, &guid);
3794		}
3795	}
3796#endif
3797	if (config == NULL) {
3798		cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3799		    devpath);
3800		return (SET_ERROR(EIO));
3801	}
3802
3803	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3804	    &pname) == 0);
3805	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3806
3807	mutex_enter(&spa_namespace_lock);
3808	if ((spa = spa_lookup(pname)) != NULL) {
3809		/*
3810		 * Remove the existing root pool from the namespace so that we
3811		 * can replace it with the correct config we just read in.
3812		 */
3813		spa_remove(spa);
3814	}
3815
3816	spa = spa_add(pname, config, NULL);
3817	spa->spa_is_root = B_TRUE;
3818	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3819
3820	/*
3821	 * Build up a vdev tree based on the boot device's label config.
3822	 */
3823	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3824	    &nvtop) == 0);
3825	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3826	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3827	    VDEV_ALLOC_ROOTPOOL);
3828	spa_config_exit(spa, SCL_ALL, FTAG);
3829	if (error) {
3830		mutex_exit(&spa_namespace_lock);
3831		nvlist_free(config);
3832		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3833		    pname);
3834		return (error);
3835	}
3836
3837	/*
3838	 * Get the boot vdev.
3839	 */
3840	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3841		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3842		    (u_longlong_t)guid);
3843		error = SET_ERROR(ENOENT);
3844		goto out;
3845	}
3846
3847	/*
3848	 * Determine if there is a better boot device.
3849	 */
3850	avd = bvd;
3851	spa_alt_rootvdev(rvd, &avd, &txg);
3852	if (avd != bvd) {
3853		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3854		    "try booting from '%s'", avd->vdev_path);
3855		error = SET_ERROR(EINVAL);
3856		goto out;
3857	}
3858
3859	/*
3860	 * If the boot device is part of a spare vdev then ensure that
3861	 * we're booting off the active spare.
3862	 */
3863	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3864	    !bvd->vdev_isspare) {
3865		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3866		    "try booting from '%s'",
3867		    bvd->vdev_parent->
3868		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3869		error = SET_ERROR(EINVAL);
3870		goto out;
3871	}
3872
3873	error = 0;
3874out:
3875	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3876	vdev_free(rvd);
3877	spa_config_exit(spa, SCL_ALL, FTAG);
3878	mutex_exit(&spa_namespace_lock);
3879
3880	nvlist_free(config);
3881	return (error);
3882}
3883
3884#endif
3885
3886/*
3887 * Import a non-root pool into the system.
3888 */
3889int
3890spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3891{
3892	spa_t *spa;
3893	char *altroot = NULL;
3894	spa_load_state_t state = SPA_LOAD_IMPORT;
3895	zpool_rewind_policy_t policy;
3896	uint64_t mode = spa_mode_global;
3897	uint64_t readonly = B_FALSE;
3898	int error;
3899	nvlist_t *nvroot;
3900	nvlist_t **spares, **l2cache;
3901	uint_t nspares, nl2cache;
3902
3903	/*
3904	 * If a pool with this name exists, return failure.
3905	 */
3906	mutex_enter(&spa_namespace_lock);
3907	if (spa_lookup(pool) != NULL) {
3908		mutex_exit(&spa_namespace_lock);
3909		return (SET_ERROR(EEXIST));
3910	}
3911
3912	/*
3913	 * Create and initialize the spa structure.
3914	 */
3915	(void) nvlist_lookup_string(props,
3916	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3917	(void) nvlist_lookup_uint64(props,
3918	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3919	if (readonly)
3920		mode = FREAD;
3921	spa = spa_add(pool, config, altroot);
3922	spa->spa_import_flags = flags;
3923
3924	/*
3925	 * Verbatim import - Take a pool and insert it into the namespace
3926	 * as if it had been loaded at boot.
3927	 */
3928	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3929		if (props != NULL)
3930			spa_configfile_set(spa, props, B_FALSE);
3931
3932		spa_config_sync(spa, B_FALSE, B_TRUE);
3933
3934		mutex_exit(&spa_namespace_lock);
3935		return (0);
3936	}
3937
3938	spa_activate(spa, mode);
3939
3940	/*
3941	 * Don't start async tasks until we know everything is healthy.
3942	 */
3943	spa_async_suspend(spa);
3944
3945	zpool_get_rewind_policy(config, &policy);
3946	if (policy.zrp_request & ZPOOL_DO_REWIND)
3947		state = SPA_LOAD_RECOVER;
3948
3949	/*
3950	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3951	 * because the user-supplied config is actually the one to trust when
3952	 * doing an import.
3953	 */
3954	if (state != SPA_LOAD_RECOVER)
3955		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3956
3957	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3958	    policy.zrp_request);
3959
3960	/*
3961	 * Propagate anything learned while loading the pool and pass it
3962	 * back to caller (i.e. rewind info, missing devices, etc).
3963	 */
3964	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3965	    spa->spa_load_info) == 0);
3966
3967	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3968	/*
3969	 * Toss any existing sparelist, as it doesn't have any validity
3970	 * anymore, and conflicts with spa_has_spare().
3971	 */
3972	if (spa->spa_spares.sav_config) {
3973		nvlist_free(spa->spa_spares.sav_config);
3974		spa->spa_spares.sav_config = NULL;
3975		spa_load_spares(spa);
3976	}
3977	if (spa->spa_l2cache.sav_config) {
3978		nvlist_free(spa->spa_l2cache.sav_config);
3979		spa->spa_l2cache.sav_config = NULL;
3980		spa_load_l2cache(spa);
3981	}
3982
3983	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3984	    &nvroot) == 0);
3985	if (error == 0)
3986		error = spa_validate_aux(spa, nvroot, -1ULL,
3987		    VDEV_ALLOC_SPARE);
3988	if (error == 0)
3989		error = spa_validate_aux(spa, nvroot, -1ULL,
3990		    VDEV_ALLOC_L2CACHE);
3991	spa_config_exit(spa, SCL_ALL, FTAG);
3992
3993	if (props != NULL)
3994		spa_configfile_set(spa, props, B_FALSE);
3995
3996	if (error != 0 || (props && spa_writeable(spa) &&
3997	    (error = spa_prop_set(spa, props)))) {
3998		spa_unload(spa);
3999		spa_deactivate(spa);
4000		spa_remove(spa);
4001		mutex_exit(&spa_namespace_lock);
4002		return (error);
4003	}
4004
4005	spa_async_resume(spa);
4006
4007	/*
4008	 * Override any spares and level 2 cache devices as specified by
4009	 * the user, as these may have correct device names/devids, etc.
4010	 */
4011	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4012	    &spares, &nspares) == 0) {
4013		if (spa->spa_spares.sav_config)
4014			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4015			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4016		else
4017			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4018			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4019		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4020		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4021		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4022		spa_load_spares(spa);
4023		spa_config_exit(spa, SCL_ALL, FTAG);
4024		spa->spa_spares.sav_sync = B_TRUE;
4025	}
4026	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4027	    &l2cache, &nl2cache) == 0) {
4028		if (spa->spa_l2cache.sav_config)
4029			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4030			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4031		else
4032			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4033			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4034		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4035		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4036		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4037		spa_load_l2cache(spa);
4038		spa_config_exit(spa, SCL_ALL, FTAG);
4039		spa->spa_l2cache.sav_sync = B_TRUE;
4040	}
4041
4042	/*
4043	 * Check for any removed devices.
4044	 */
4045	if (spa->spa_autoreplace) {
4046		spa_aux_check_removed(&spa->spa_spares);
4047		spa_aux_check_removed(&spa->spa_l2cache);
4048	}
4049
4050	if (spa_writeable(spa)) {
4051		/*
4052		 * Update the config cache to include the newly-imported pool.
4053		 */
4054		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4055	}
4056
4057	/*
4058	 * It's possible that the pool was expanded while it was exported.
4059	 * We kick off an async task to handle this for us.
4060	 */
4061	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4062
4063	mutex_exit(&spa_namespace_lock);
4064	spa_history_log_version(spa, "import");
4065
4066	return (0);
4067}
4068
4069nvlist_t *
4070spa_tryimport(nvlist_t *tryconfig)
4071{
4072	nvlist_t *config = NULL;
4073	char *poolname;
4074	spa_t *spa;
4075	uint64_t state;
4076	int error;
4077
4078	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4079		return (NULL);
4080
4081	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4082		return (NULL);
4083
4084	/*
4085	 * Create and initialize the spa structure.
4086	 */
4087	mutex_enter(&spa_namespace_lock);
4088	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4089	spa_activate(spa, FREAD);
4090
4091	/*
4092	 * Pass off the heavy lifting to spa_load().
4093	 * Pass TRUE for mosconfig because the user-supplied config
4094	 * is actually the one to trust when doing an import.
4095	 */
4096	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4097
4098	/*
4099	 * If 'tryconfig' was at least parsable, return the current config.
4100	 */
4101	if (spa->spa_root_vdev != NULL) {
4102		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4103		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4104		    poolname) == 0);
4105		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4106		    state) == 0);
4107		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4108		    spa->spa_uberblock.ub_timestamp) == 0);
4109		VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4110		    spa->spa_load_info) == 0);
4111
4112		/*
4113		 * If the bootfs property exists on this pool then we
4114		 * copy it out so that external consumers can tell which
4115		 * pools are bootable.
4116		 */
4117		if ((!error || error == EEXIST) && spa->spa_bootfs) {
4118			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4119
4120			/*
4121			 * We have to play games with the name since the
4122			 * pool was opened as TRYIMPORT_NAME.
4123			 */
4124			if (dsl_dsobj_to_dsname(spa_name(spa),
4125			    spa->spa_bootfs, tmpname) == 0) {
4126				char *cp;
4127				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4128
4129				cp = strchr(tmpname, '/');
4130				if (cp == NULL) {
4131					(void) strlcpy(dsname, tmpname,
4132					    MAXPATHLEN);
4133				} else {
4134					(void) snprintf(dsname, MAXPATHLEN,
4135					    "%s/%s", poolname, ++cp);
4136				}
4137				VERIFY(nvlist_add_string(config,
4138				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4139				kmem_free(dsname, MAXPATHLEN);
4140			}
4141			kmem_free(tmpname, MAXPATHLEN);
4142		}
4143
4144		/*
4145		 * Add the list of hot spares and level 2 cache devices.
4146		 */
4147		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4148		spa_add_spares(spa, config);
4149		spa_add_l2cache(spa, config);
4150		spa_config_exit(spa, SCL_CONFIG, FTAG);
4151	}
4152
4153	spa_unload(spa);
4154	spa_deactivate(spa);
4155	spa_remove(spa);
4156	mutex_exit(&spa_namespace_lock);
4157
4158	return (config);
4159}
4160
4161/*
4162 * Pool export/destroy
4163 *
4164 * The act of destroying or exporting a pool is very simple.  We make sure there
4165 * is no more pending I/O and any references to the pool are gone.  Then, we
4166 * update the pool state and sync all the labels to disk, removing the
4167 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4168 * we don't sync the labels or remove the configuration cache.
4169 */
4170static int
4171spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4172    boolean_t force, boolean_t hardforce)
4173{
4174	spa_t *spa;
4175
4176	if (oldconfig)
4177		*oldconfig = NULL;
4178
4179	if (!(spa_mode_global & FWRITE))
4180		return (SET_ERROR(EROFS));
4181
4182	mutex_enter(&spa_namespace_lock);
4183	if ((spa = spa_lookup(pool)) == NULL) {
4184		mutex_exit(&spa_namespace_lock);
4185		return (SET_ERROR(ENOENT));
4186	}
4187
4188	/*
4189	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4190	 * reacquire the namespace lock, and see if we can export.
4191	 */
4192	spa_open_ref(spa, FTAG);
4193	mutex_exit(&spa_namespace_lock);
4194	spa_async_suspend(spa);
4195	mutex_enter(&spa_namespace_lock);
4196	spa_close(spa, FTAG);
4197
4198	/*
4199	 * The pool will be in core if it's openable,
4200	 * in which case we can modify its state.
4201	 */
4202	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4203		/*
4204		 * Objsets may be open only because they're dirty, so we
4205		 * have to force it to sync before checking spa_refcnt.
4206		 */
4207		txg_wait_synced(spa->spa_dsl_pool, 0);
4208
4209		/*
4210		 * A pool cannot be exported or destroyed if there are active
4211		 * references.  If we are resetting a pool, allow references by
4212		 * fault injection handlers.
4213		 */
4214		if (!spa_refcount_zero(spa) ||
4215		    (spa->spa_inject_ref != 0 &&
4216		    new_state != POOL_STATE_UNINITIALIZED)) {
4217			spa_async_resume(spa);
4218			mutex_exit(&spa_namespace_lock);
4219			return (SET_ERROR(EBUSY));
4220		}
4221
4222		/*
4223		 * A pool cannot be exported if it has an active shared spare.
4224		 * This is to prevent other pools stealing the active spare
4225		 * from an exported pool. At user's own will, such pool can
4226		 * be forcedly exported.
4227		 */
4228		if (!force && new_state == POOL_STATE_EXPORTED &&
4229		    spa_has_active_shared_spare(spa)) {
4230			spa_async_resume(spa);
4231			mutex_exit(&spa_namespace_lock);
4232			return (SET_ERROR(EXDEV));
4233		}
4234
4235		/*
4236		 * We want this to be reflected on every label,
4237		 * so mark them all dirty.  spa_unload() will do the
4238		 * final sync that pushes these changes out.
4239		 */
4240		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4241			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4242			spa->spa_state = new_state;
4243			spa->spa_final_txg = spa_last_synced_txg(spa) +
4244			    TXG_DEFER_SIZE + 1;
4245			vdev_config_dirty(spa->spa_root_vdev);
4246			spa_config_exit(spa, SCL_ALL, FTAG);
4247		}
4248	}
4249
4250	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4251
4252	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4253		spa_unload(spa);
4254		spa_deactivate(spa);
4255	}
4256
4257	if (oldconfig && spa->spa_config)
4258		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4259
4260	if (new_state != POOL_STATE_UNINITIALIZED) {
4261		if (!hardforce)
4262			spa_config_sync(spa, B_TRUE, B_TRUE);
4263		spa_remove(spa);
4264	}
4265	mutex_exit(&spa_namespace_lock);
4266
4267	return (0);
4268}
4269
4270/*
4271 * Destroy a storage pool.
4272 */
4273int
4274spa_destroy(char *pool)
4275{
4276	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4277	    B_FALSE, B_FALSE));
4278}
4279
4280/*
4281 * Export a storage pool.
4282 */
4283int
4284spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4285    boolean_t hardforce)
4286{
4287	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4288	    force, hardforce));
4289}
4290
4291/*
4292 * Similar to spa_export(), this unloads the spa_t without actually removing it
4293 * from the namespace in any way.
4294 */
4295int
4296spa_reset(char *pool)
4297{
4298	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4299	    B_FALSE, B_FALSE));
4300}
4301
4302/*
4303 * ==========================================================================
4304 * Device manipulation
4305 * ==========================================================================
4306 */
4307
4308/*
4309 * Add a device to a storage pool.
4310 */
4311int
4312spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4313{
4314	uint64_t txg, id;
4315	int error;
4316	vdev_t *rvd = spa->spa_root_vdev;
4317	vdev_t *vd, *tvd;
4318	nvlist_t **spares, **l2cache;
4319	uint_t nspares, nl2cache;
4320
4321	ASSERT(spa_writeable(spa));
4322
4323	txg = spa_vdev_enter(spa);
4324
4325	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4326	    VDEV_ALLOC_ADD)) != 0)
4327		return (spa_vdev_exit(spa, NULL, txg, error));
4328
4329	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
4330
4331	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4332	    &nspares) != 0)
4333		nspares = 0;
4334
4335	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4336	    &nl2cache) != 0)
4337		nl2cache = 0;
4338
4339	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4340		return (spa_vdev_exit(spa, vd, txg, EINVAL));
4341
4342	if (vd->vdev_children != 0 &&
4343	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
4344		return (spa_vdev_exit(spa, vd, txg, error));
4345
4346	/*
4347	 * We must validate the spares and l2cache devices after checking the
4348	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
4349	 */
4350	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4351		return (spa_vdev_exit(spa, vd, txg, error));
4352
4353	/*
4354	 * Transfer each new top-level vdev from vd to rvd.
4355	 */
4356	for (int c = 0; c < vd->vdev_children; c++) {
4357
4358		/*
4359		 * Set the vdev id to the first hole, if one exists.
4360		 */
4361		for (id = 0; id < rvd->vdev_children; id++) {
4362			if (rvd->vdev_child[id]->vdev_ishole) {
4363				vdev_free(rvd->vdev_child[id]);
4364				break;
4365			}
4366		}
4367		tvd = vd->vdev_child[c];
4368		vdev_remove_child(vd, tvd);
4369		tvd->vdev_id = id;
4370		vdev_add_child(rvd, tvd);
4371		vdev_config_dirty(tvd);
4372	}
4373
4374	if (nspares != 0) {
4375		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4376		    ZPOOL_CONFIG_SPARES);
4377		spa_load_spares(spa);
4378		spa->spa_spares.sav_sync = B_TRUE;
4379	}
4380
4381	if (nl2cache != 0) {
4382		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4383		    ZPOOL_CONFIG_L2CACHE);
4384		spa_load_l2cache(spa);
4385		spa->spa_l2cache.sav_sync = B_TRUE;
4386	}
4387
4388	/*
4389	 * We have to be careful when adding new vdevs to an existing pool.
4390	 * If other threads start allocating from these vdevs before we
4391	 * sync the config cache, and we lose power, then upon reboot we may
4392	 * fail to open the pool because there are DVAs that the config cache
4393	 * can't translate.  Therefore, we first add the vdevs without
4394	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4395	 * and then let spa_config_update() initialize the new metaslabs.
4396	 *
4397	 * spa_load() checks for added-but-not-initialized vdevs, so that
4398	 * if we lose power at any point in this sequence, the remaining
4399	 * steps will be completed the next time we load the pool.
4400	 */
4401	(void) spa_vdev_exit(spa, vd, txg, 0);
4402
4403	mutex_enter(&spa_namespace_lock);
4404	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4405	mutex_exit(&spa_namespace_lock);
4406
4407	return (0);
4408}
4409
4410/*
4411 * Attach a device to a mirror.  The arguments are the path to any device
4412 * in the mirror, and the nvroot for the new device.  If the path specifies
4413 * a device that is not mirrored, we automatically insert the mirror vdev.
4414 *
4415 * If 'replacing' is specified, the new device is intended to replace the
4416 * existing device; in this case the two devices are made into their own
4417 * mirror using the 'replacing' vdev, which is functionally identical to
4418 * the mirror vdev (it actually reuses all the same ops) but has a few
4419 * extra rules: you can't attach to it after it's been created, and upon
4420 * completion of resilvering, the first disk (the one being replaced)
4421 * is automatically detached.
4422 */
4423int
4424spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4425{
4426	uint64_t txg, dtl_max_txg;
4427	vdev_t *rvd = spa->spa_root_vdev;
4428	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4429	vdev_ops_t *pvops;
4430	char *oldvdpath, *newvdpath;
4431	int newvd_isspare;
4432	int error;
4433
4434	ASSERT(spa_writeable(spa));
4435
4436	txg = spa_vdev_enter(spa);
4437
4438	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4439
4440	if (oldvd == NULL)
4441		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4442
4443	if (!oldvd->vdev_ops->vdev_op_leaf)
4444		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4445
4446	pvd = oldvd->vdev_parent;
4447
4448	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4449	    VDEV_ALLOC_ATTACH)) != 0)
4450		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4451
4452	if (newrootvd->vdev_children != 1)
4453		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4454
4455	newvd = newrootvd->vdev_child[0];
4456
4457	if (!newvd->vdev_ops->vdev_op_leaf)
4458		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4459
4460	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4461		return (spa_vdev_exit(spa, newrootvd, txg, error));
4462
4463	/*
4464	 * Spares can't replace logs
4465	 */
4466	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4467		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4468
4469	if (!replacing) {
4470		/*
4471		 * For attach, the only allowable parent is a mirror or the root
4472		 * vdev.
4473		 */
4474		if (pvd->vdev_ops != &vdev_mirror_ops &&
4475		    pvd->vdev_ops != &vdev_root_ops)
4476			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4477
4478		pvops = &vdev_mirror_ops;
4479	} else {
4480		/*
4481		 * Active hot spares can only be replaced by inactive hot
4482		 * spares.
4483		 */
4484		if (pvd->vdev_ops == &vdev_spare_ops &&
4485		    oldvd->vdev_isspare &&
4486		    !spa_has_spare(spa, newvd->vdev_guid))
4487			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4488
4489		/*
4490		 * If the source is a hot spare, and the parent isn't already a
4491		 * spare, then we want to create a new hot spare.  Otherwise, we
4492		 * want to create a replacing vdev.  The user is not allowed to
4493		 * attach to a spared vdev child unless the 'isspare' state is
4494		 * the same (spare replaces spare, non-spare replaces
4495		 * non-spare).
4496		 */
4497		if (pvd->vdev_ops == &vdev_replacing_ops &&
4498		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4499			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4500		} else if (pvd->vdev_ops == &vdev_spare_ops &&
4501		    newvd->vdev_isspare != oldvd->vdev_isspare) {
4502			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4503		}
4504
4505		if (newvd->vdev_isspare)
4506			pvops = &vdev_spare_ops;
4507		else
4508			pvops = &vdev_replacing_ops;
4509	}
4510
4511	/*
4512	 * Make sure the new device is big enough.
4513	 */
4514	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4515		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4516
4517	/*
4518	 * The new device cannot have a higher alignment requirement
4519	 * than the top-level vdev.
4520	 */
4521	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4522		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4523
4524	/*
4525	 * If this is an in-place replacement, update oldvd's path and devid
4526	 * to make it distinguishable from newvd, and unopenable from now on.
4527	 */
4528	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4529		spa_strfree(oldvd->vdev_path);
4530		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4531		    KM_SLEEP);
4532		(void) sprintf(oldvd->vdev_path, "%s/%s",
4533		    newvd->vdev_path, "old");
4534		if (oldvd->vdev_devid != NULL) {
4535			spa_strfree(oldvd->vdev_devid);
4536			oldvd->vdev_devid = NULL;
4537		}
4538	}
4539
4540	/* mark the device being resilvered */
4541	newvd->vdev_resilver_txg = txg;
4542
4543	/*
4544	 * If the parent is not a mirror, or if we're replacing, insert the new
4545	 * mirror/replacing/spare vdev above oldvd.
4546	 */
4547	if (pvd->vdev_ops != pvops)
4548		pvd = vdev_add_parent(oldvd, pvops);
4549
4550	ASSERT(pvd->vdev_top->vdev_parent == rvd);
4551	ASSERT(pvd->vdev_ops == pvops);
4552	ASSERT(oldvd->vdev_parent == pvd);
4553
4554	/*
4555	 * Extract the new device from its root and add it to pvd.
4556	 */
4557	vdev_remove_child(newrootvd, newvd);
4558	newvd->vdev_id = pvd->vdev_children;
4559	newvd->vdev_crtxg = oldvd->vdev_crtxg;
4560	vdev_add_child(pvd, newvd);
4561
4562	tvd = newvd->vdev_top;
4563	ASSERT(pvd->vdev_top == tvd);
4564	ASSERT(tvd->vdev_parent == rvd);
4565
4566	vdev_config_dirty(tvd);
4567
4568	/*
4569	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4570	 * for any dmu_sync-ed blocks.  It will propagate upward when
4571	 * spa_vdev_exit() calls vdev_dtl_reassess().
4572	 */
4573	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4574
4575	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4576	    dtl_max_txg - TXG_INITIAL);
4577
4578	if (newvd->vdev_isspare) {
4579		spa_spare_activate(newvd);
4580		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4581	}
4582
4583	oldvdpath = spa_strdup(oldvd->vdev_path);
4584	newvdpath = spa_strdup(newvd->vdev_path);
4585	newvd_isspare = newvd->vdev_isspare;
4586
4587	/*
4588	 * Mark newvd's DTL dirty in this txg.
4589	 */
4590	vdev_dirty(tvd, VDD_DTL, newvd, txg);
4591
4592	/*
4593	 * Schedule the resilver to restart in the future. We do this to
4594	 * ensure that dmu_sync-ed blocks have been stitched into the
4595	 * respective datasets.
4596	 */
4597	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4598
4599	/*
4600	 * Commit the config
4601	 */
4602	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4603
4604	spa_history_log_internal(spa, "vdev attach", NULL,
4605	    "%s vdev=%s %s vdev=%s",
4606	    replacing && newvd_isspare ? "spare in" :
4607	    replacing ? "replace" : "attach", newvdpath,
4608	    replacing ? "for" : "to", oldvdpath);
4609
4610	spa_strfree(oldvdpath);
4611	spa_strfree(newvdpath);
4612
4613	if (spa->spa_bootfs)
4614		spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4615
4616	return (0);
4617}
4618
4619/*
4620 * Detach a device from a mirror or replacing vdev.
4621 *
4622 * If 'replace_done' is specified, only detach if the parent
4623 * is a replacing vdev.
4624 */
4625int
4626spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4627{
4628	uint64_t txg;
4629	int error;
4630	vdev_t *rvd = spa->spa_root_vdev;
4631	vdev_t *vd, *pvd, *cvd, *tvd;
4632	boolean_t unspare = B_FALSE;
4633	uint64_t unspare_guid = 0;
4634	char *vdpath;
4635
4636	ASSERT(spa_writeable(spa));
4637
4638	txg = spa_vdev_enter(spa);
4639
4640	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4641
4642	if (vd == NULL)
4643		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4644
4645	if (!vd->vdev_ops->vdev_op_leaf)
4646		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4647
4648	pvd = vd->vdev_parent;
4649
4650	/*
4651	 * If the parent/child relationship is not as expected, don't do it.
4652	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4653	 * vdev that's replacing B with C.  The user's intent in replacing
4654	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4655	 * the replace by detaching C, the expected behavior is to end up
4656	 * M(A,B).  But suppose that right after deciding to detach C,
4657	 * the replacement of B completes.  We would have M(A,C), and then
4658	 * ask to detach C, which would leave us with just A -- not what
4659	 * the user wanted.  To prevent this, we make sure that the
4660	 * parent/child relationship hasn't changed -- in this example,
4661	 * that C's parent is still the replacing vdev R.
4662	 */
4663	if (pvd->vdev_guid != pguid && pguid != 0)
4664		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4665
4666	/*
4667	 * Only 'replacing' or 'spare' vdevs can be replaced.
4668	 */
4669	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4670	    pvd->vdev_ops != &vdev_spare_ops)
4671		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4672
4673	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4674	    spa_version(spa) >= SPA_VERSION_SPARES);
4675
4676	/*
4677	 * Only mirror, replacing, and spare vdevs support detach.
4678	 */
4679	if (pvd->vdev_ops != &vdev_replacing_ops &&
4680	    pvd->vdev_ops != &vdev_mirror_ops &&
4681	    pvd->vdev_ops != &vdev_spare_ops)
4682		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4683
4684	/*
4685	 * If this device has the only valid copy of some data,
4686	 * we cannot safely detach it.
4687	 */
4688	if (vdev_dtl_required(vd))
4689		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4690
4691	ASSERT(pvd->vdev_children >= 2);
4692
4693	/*
4694	 * If we are detaching the second disk from a replacing vdev, then
4695	 * check to see if we changed the original vdev's path to have "/old"
4696	 * at the end in spa_vdev_attach().  If so, undo that change now.
4697	 */
4698	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4699	    vd->vdev_path != NULL) {
4700		size_t len = strlen(vd->vdev_path);
4701
4702		for (int c = 0; c < pvd->vdev_children; c++) {
4703			cvd = pvd->vdev_child[c];
4704
4705			if (cvd == vd || cvd->vdev_path == NULL)
4706				continue;
4707
4708			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4709			    strcmp(cvd->vdev_path + len, "/old") == 0) {
4710				spa_strfree(cvd->vdev_path);
4711				cvd->vdev_path = spa_strdup(vd->vdev_path);
4712				break;
4713			}
4714		}
4715	}
4716
4717	/*
4718	 * If we are detaching the original disk from a spare, then it implies
4719	 * that the spare should become a real disk, and be removed from the
4720	 * active spare list for the pool.
4721	 */
4722	if (pvd->vdev_ops == &vdev_spare_ops &&
4723	    vd->vdev_id == 0 &&
4724	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4725		unspare = B_TRUE;
4726
4727	/*
4728	 * Erase the disk labels so the disk can be used for other things.
4729	 * This must be done after all other error cases are handled,
4730	 * but before we disembowel vd (so we can still do I/O to it).
4731	 * But if we can't do it, don't treat the error as fatal --
4732	 * it may be that the unwritability of the disk is the reason
4733	 * it's being detached!
4734	 */
4735	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4736
4737	/*
4738	 * Remove vd from its parent and compact the parent's children.
4739	 */
4740	vdev_remove_child(pvd, vd);
4741	vdev_compact_children(pvd);
4742
4743	/*
4744	 * Remember one of the remaining children so we can get tvd below.
4745	 */
4746	cvd = pvd->vdev_child[pvd->vdev_children - 1];
4747
4748	/*
4749	 * If we need to remove the remaining child from the list of hot spares,
4750	 * do it now, marking the vdev as no longer a spare in the process.
4751	 * We must do this before vdev_remove_parent(), because that can
4752	 * change the GUID if it creates a new toplevel GUID.  For a similar
4753	 * reason, we must remove the spare now, in the same txg as the detach;
4754	 * otherwise someone could attach a new sibling, change the GUID, and
4755	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4756	 */
4757	if (unspare) {
4758		ASSERT(cvd->vdev_isspare);
4759		spa_spare_remove(cvd);
4760		unspare_guid = cvd->vdev_guid;
4761		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4762		cvd->vdev_unspare = B_TRUE;
4763	}
4764
4765	/*
4766	 * If the parent mirror/replacing vdev only has one child,
4767	 * the parent is no longer needed.  Remove it from the tree.
4768	 */
4769	if (pvd->vdev_children == 1) {
4770		if (pvd->vdev_ops == &vdev_spare_ops)
4771			cvd->vdev_unspare = B_FALSE;
4772		vdev_remove_parent(cvd);
4773	}
4774
4775
4776	/*
4777	 * We don't set tvd until now because the parent we just removed
4778	 * may have been the previous top-level vdev.
4779	 */
4780	tvd = cvd->vdev_top;
4781	ASSERT(tvd->vdev_parent == rvd);
4782
4783	/*
4784	 * Reevaluate the parent vdev state.
4785	 */
4786	vdev_propagate_state(cvd);
4787
4788	/*
4789	 * If the 'autoexpand' property is set on the pool then automatically
4790	 * try to expand the size of the pool. For example if the device we
4791	 * just detached was smaller than the others, it may be possible to
4792	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4793	 * first so that we can obtain the updated sizes of the leaf vdevs.
4794	 */
4795	if (spa->spa_autoexpand) {
4796		vdev_reopen(tvd);
4797		vdev_expand(tvd, txg);
4798	}
4799
4800	vdev_config_dirty(tvd);
4801
4802	/*
4803	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4804	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4805	 * But first make sure we're not on any *other* txg's DTL list, to
4806	 * prevent vd from being accessed after it's freed.
4807	 */
4808	vdpath = spa_strdup(vd->vdev_path);
4809	for (int t = 0; t < TXG_SIZE; t++)
4810		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4811	vd->vdev_detached = B_TRUE;
4812	vdev_dirty(tvd, VDD_DTL, vd, txg);
4813
4814	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4815
4816	/* hang on to the spa before we release the lock */
4817	spa_open_ref(spa, FTAG);
4818
4819	error = spa_vdev_exit(spa, vd, txg, 0);
4820
4821	spa_history_log_internal(spa, "detach", NULL,
4822	    "vdev=%s", vdpath);
4823	spa_strfree(vdpath);
4824
4825	/*
4826	 * If this was the removal of the original device in a hot spare vdev,
4827	 * then we want to go through and remove the device from the hot spare
4828	 * list of every other pool.
4829	 */
4830	if (unspare) {
4831		spa_t *altspa = NULL;
4832
4833		mutex_enter(&spa_namespace_lock);
4834		while ((altspa = spa_next(altspa)) != NULL) {
4835			if (altspa->spa_state != POOL_STATE_ACTIVE ||
4836			    altspa == spa)
4837				continue;
4838
4839			spa_open_ref(altspa, FTAG);
4840			mutex_exit(&spa_namespace_lock);
4841			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4842			mutex_enter(&spa_namespace_lock);
4843			spa_close(altspa, FTAG);
4844		}
4845		mutex_exit(&spa_namespace_lock);
4846
4847		/* search the rest of the vdevs for spares to remove */
4848		spa_vdev_resilver_done(spa);
4849	}
4850
4851	/* all done with the spa; OK to release */
4852	mutex_enter(&spa_namespace_lock);
4853	spa_close(spa, FTAG);
4854	mutex_exit(&spa_namespace_lock);
4855
4856	return (error);
4857}
4858
4859/*
4860 * Split a set of devices from their mirrors, and create a new pool from them.
4861 */
4862int
4863spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4864    nvlist_t *props, boolean_t exp)
4865{
4866	int error = 0;
4867	uint64_t txg, *glist;
4868	spa_t *newspa;
4869	uint_t c, children, lastlog;
4870	nvlist_t **child, *nvl, *tmp;
4871	dmu_tx_t *tx;
4872	char *altroot = NULL;
4873	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4874	boolean_t activate_slog;
4875
4876	ASSERT(spa_writeable(spa));
4877
4878	txg = spa_vdev_enter(spa);
4879
4880	/* clear the log and flush everything up to now */
4881	activate_slog = spa_passivate_log(spa);
4882	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4883	error = spa_offline_log(spa);
4884	txg = spa_vdev_config_enter(spa);
4885
4886	if (activate_slog)
4887		spa_activate_log(spa);
4888
4889	if (error != 0)
4890		return (spa_vdev_exit(spa, NULL, txg, error));
4891
4892	/* check new spa name before going any further */
4893	if (spa_lookup(newname) != NULL)
4894		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4895
4896	/*
4897	 * scan through all the children to ensure they're all mirrors
4898	 */
4899	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4900	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4901	    &children) != 0)
4902		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4903
4904	/* first, check to ensure we've got the right child count */
4905	rvd = spa->spa_root_vdev;
4906	lastlog = 0;
4907	for (c = 0; c < rvd->vdev_children; c++) {
4908		vdev_t *vd = rvd->vdev_child[c];
4909
4910		/* don't count the holes & logs as children */
4911		if (vd->vdev_islog || vd->vdev_ishole) {
4912			if (lastlog == 0)
4913				lastlog = c;
4914			continue;
4915		}
4916
4917		lastlog = 0;
4918	}
4919	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4920		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4921
4922	/* next, ensure no spare or cache devices are part of the split */
4923	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4924	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4925		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4926
4927	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4928	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4929
4930	/* then, loop over each vdev and validate it */
4931	for (c = 0; c < children; c++) {
4932		uint64_t is_hole = 0;
4933
4934		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4935		    &is_hole);
4936
4937		if (is_hole != 0) {
4938			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4939			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4940				continue;
4941			} else {
4942				error = SET_ERROR(EINVAL);
4943				break;
4944			}
4945		}
4946
4947		/* which disk is going to be split? */
4948		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4949		    &glist[c]) != 0) {
4950			error = SET_ERROR(EINVAL);
4951			break;
4952		}
4953
4954		/* look it up in the spa */
4955		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4956		if (vml[c] == NULL) {
4957			error = SET_ERROR(ENODEV);
4958			break;
4959		}
4960
4961		/* make sure there's nothing stopping the split */
4962		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4963		    vml[c]->vdev_islog ||
4964		    vml[c]->vdev_ishole ||
4965		    vml[c]->vdev_isspare ||
4966		    vml[c]->vdev_isl2cache ||
4967		    !vdev_writeable(vml[c]) ||
4968		    vml[c]->vdev_children != 0 ||
4969		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4970		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4971			error = SET_ERROR(EINVAL);
4972			break;
4973		}
4974
4975		if (vdev_dtl_required(vml[c])) {
4976			error = SET_ERROR(EBUSY);
4977			break;
4978		}
4979
4980		/* we need certain info from the top level */
4981		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4982		    vml[c]->vdev_top->vdev_ms_array) == 0);
4983		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4984		    vml[c]->vdev_top->vdev_ms_shift) == 0);
4985		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4986		    vml[c]->vdev_top->vdev_asize) == 0);
4987		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4988		    vml[c]->vdev_top->vdev_ashift) == 0);
4989	}
4990
4991	if (error != 0) {
4992		kmem_free(vml, children * sizeof (vdev_t *));
4993		kmem_free(glist, children * sizeof (uint64_t));
4994		return (spa_vdev_exit(spa, NULL, txg, error));
4995	}
4996
4997	/* stop writers from using the disks */
4998	for (c = 0; c < children; c++) {
4999		if (vml[c] != NULL)
5000			vml[c]->vdev_offline = B_TRUE;
5001	}
5002	vdev_reopen(spa->spa_root_vdev);
5003
5004	/*
5005	 * Temporarily record the splitting vdevs in the spa config.  This
5006	 * will disappear once the config is regenerated.
5007	 */
5008	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5009	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5010	    glist, children) == 0);
5011	kmem_free(glist, children * sizeof (uint64_t));
5012
5013	mutex_enter(&spa->spa_props_lock);
5014	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5015	    nvl) == 0);
5016	mutex_exit(&spa->spa_props_lock);
5017	spa->spa_config_splitting = nvl;
5018	vdev_config_dirty(spa->spa_root_vdev);
5019
5020	/* configure and create the new pool */
5021	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5022	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5023	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5024	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5025	    spa_version(spa)) == 0);
5026	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5027	    spa->spa_config_txg) == 0);
5028	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5029	    spa_generate_guid(NULL)) == 0);
5030	(void) nvlist_lookup_string(props,
5031	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5032
5033	/* add the new pool to the namespace */
5034	newspa = spa_add(newname, config, altroot);
5035	newspa->spa_config_txg = spa->spa_config_txg;
5036	spa_set_log_state(newspa, SPA_LOG_CLEAR);
5037
5038	/* release the spa config lock, retaining the namespace lock */
5039	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5040
5041	if (zio_injection_enabled)
5042		zio_handle_panic_injection(spa, FTAG, 1);
5043
5044	spa_activate(newspa, spa_mode_global);
5045	spa_async_suspend(newspa);
5046
5047	/* create the new pool from the disks of the original pool */
5048	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5049	if (error)
5050		goto out;
5051
5052	/* if that worked, generate a real config for the new pool */
5053	if (newspa->spa_root_vdev != NULL) {
5054		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5055		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
5056		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5057		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5058		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5059		    B_TRUE));
5060	}
5061
5062	/* set the props */
5063	if (props != NULL) {
5064		spa_configfile_set(newspa, props, B_FALSE);
5065		error = spa_prop_set(newspa, props);
5066		if (error)
5067			goto out;
5068	}
5069
5070	/* flush everything */
5071	txg = spa_vdev_config_enter(newspa);
5072	vdev_config_dirty(newspa->spa_root_vdev);
5073	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5074
5075	if (zio_injection_enabled)
5076		zio_handle_panic_injection(spa, FTAG, 2);
5077
5078	spa_async_resume(newspa);
5079
5080	/* finally, update the original pool's config */
5081	txg = spa_vdev_config_enter(spa);
5082	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5083	error = dmu_tx_assign(tx, TXG_WAIT);
5084	if (error != 0)
5085		dmu_tx_abort(tx);
5086	for (c = 0; c < children; c++) {
5087		if (vml[c] != NULL) {
5088			vdev_split(vml[c]);
5089			if (error == 0)
5090				spa_history_log_internal(spa, "detach", tx,
5091				    "vdev=%s", vml[c]->vdev_path);
5092			vdev_free(vml[c]);
5093		}
5094	}
5095	vdev_config_dirty(spa->spa_root_vdev);
5096	spa->spa_config_splitting = NULL;
5097	nvlist_free(nvl);
5098	if (error == 0)
5099		dmu_tx_commit(tx);
5100	(void) spa_vdev_exit(spa, NULL, txg, 0);
5101
5102	if (zio_injection_enabled)
5103		zio_handle_panic_injection(spa, FTAG, 3);
5104
5105	/* split is complete; log a history record */
5106	spa_history_log_internal(newspa, "split", NULL,
5107	    "from pool %s", spa_name(spa));
5108
5109	kmem_free(vml, children * sizeof (vdev_t *));
5110
5111	/* if we're not going to mount the filesystems in userland, export */
5112	if (exp)
5113		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5114		    B_FALSE, B_FALSE);
5115
5116	return (error);
5117
5118out:
5119	spa_unload(newspa);
5120	spa_deactivate(newspa);
5121	spa_remove(newspa);
5122
5123	txg = spa_vdev_config_enter(spa);
5124
5125	/* re-online all offlined disks */
5126	for (c = 0; c < children; c++) {
5127		if (vml[c] != NULL)
5128			vml[c]->vdev_offline = B_FALSE;
5129	}
5130	vdev_reopen(spa->spa_root_vdev);
5131
5132	nvlist_free(spa->spa_config_splitting);
5133	spa->spa_config_splitting = NULL;
5134	(void) spa_vdev_exit(spa, NULL, txg, error);
5135
5136	kmem_free(vml, children * sizeof (vdev_t *));
5137	return (error);
5138}
5139
5140static nvlist_t *
5141spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5142{
5143	for (int i = 0; i < count; i++) {
5144		uint64_t guid;
5145
5146		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5147		    &guid) == 0);
5148
5149		if (guid == target_guid)
5150			return (nvpp[i]);
5151	}
5152
5153	return (NULL);
5154}
5155
5156static void
5157spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5158	nvlist_t *dev_to_remove)
5159{
5160	nvlist_t **newdev = NULL;
5161
5162	if (count > 1)
5163		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5164
5165	for (int i = 0, j = 0; i < count; i++) {
5166		if (dev[i] == dev_to_remove)
5167			continue;
5168		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5169	}
5170
5171	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5172	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5173
5174	for (int i = 0; i < count - 1; i++)
5175		nvlist_free(newdev[i]);
5176
5177	if (count > 1)
5178		kmem_free(newdev, (count - 1) * sizeof (void *));
5179}
5180
5181/*
5182 * Evacuate the device.
5183 */
5184static int
5185spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5186{
5187	uint64_t txg;
5188	int error = 0;
5189
5190	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5191	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5192	ASSERT(vd == vd->vdev_top);
5193
5194	/*
5195	 * Evacuate the device.  We don't hold the config lock as writer
5196	 * since we need to do I/O but we do keep the
5197	 * spa_namespace_lock held.  Once this completes the device
5198	 * should no longer have any blocks allocated on it.
5199	 */
5200	if (vd->vdev_islog) {
5201		if (vd->vdev_stat.vs_alloc != 0)
5202			error = spa_offline_log(spa);
5203	} else {
5204		error = SET_ERROR(ENOTSUP);
5205	}
5206
5207	if (error)
5208		return (error);
5209
5210	/*
5211	 * The evacuation succeeded.  Remove any remaining MOS metadata
5212	 * associated with this vdev, and wait for these changes to sync.
5213	 */
5214	ASSERT0(vd->vdev_stat.vs_alloc);
5215	txg = spa_vdev_config_enter(spa);
5216	vd->vdev_removing = B_TRUE;
5217	vdev_dirty_leaves(vd, VDD_DTL, txg);
5218	vdev_config_dirty(vd);
5219	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5220
5221	return (0);
5222}
5223
5224/*
5225 * Complete the removal by cleaning up the namespace.
5226 */
5227static void
5228spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5229{
5230	vdev_t *rvd = spa->spa_root_vdev;
5231	uint64_t id = vd->vdev_id;
5232	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5233
5234	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5235	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5236	ASSERT(vd == vd->vdev_top);
5237
5238	/*
5239	 * Only remove any devices which are empty.
5240	 */
5241	if (vd->vdev_stat.vs_alloc != 0)
5242		return;
5243
5244	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5245
5246	if (list_link_active(&vd->vdev_state_dirty_node))
5247		vdev_state_clean(vd);
5248	if (list_link_active(&vd->vdev_config_dirty_node))
5249		vdev_config_clean(vd);
5250
5251	vdev_free(vd);
5252
5253	if (last_vdev) {
5254		vdev_compact_children(rvd);
5255	} else {
5256		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5257		vdev_add_child(rvd, vd);
5258	}
5259	vdev_config_dirty(rvd);
5260
5261	/*
5262	 * Reassess the health of our root vdev.
5263	 */
5264	vdev_reopen(rvd);
5265}
5266
5267/*
5268 * Remove a device from the pool -
5269 *
5270 * Removing a device from the vdev namespace requires several steps
5271 * and can take a significant amount of time.  As a result we use
5272 * the spa_vdev_config_[enter/exit] functions which allow us to
5273 * grab and release the spa_config_lock while still holding the namespace
5274 * lock.  During each step the configuration is synced out.
5275 *
5276 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5277 * devices.
5278 */
5279int
5280spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5281{
5282	vdev_t *vd;
5283	metaslab_group_t *mg;
5284	nvlist_t **spares, **l2cache, *nv;
5285	uint64_t txg = 0;
5286	uint_t nspares, nl2cache;
5287	int error = 0;
5288	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5289
5290	ASSERT(spa_writeable(spa));
5291
5292	if (!locked)
5293		txg = spa_vdev_enter(spa);
5294
5295	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5296
5297	if (spa->spa_spares.sav_vdevs != NULL &&
5298	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5299	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5300	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5301		/*
5302		 * Only remove the hot spare if it's not currently in use
5303		 * in this pool.
5304		 */
5305		if (vd == NULL || unspare) {
5306			spa_vdev_remove_aux(spa->spa_spares.sav_config,
5307			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5308			spa_load_spares(spa);
5309			spa->spa_spares.sav_sync = B_TRUE;
5310		} else {
5311			error = SET_ERROR(EBUSY);
5312		}
5313	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
5314	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5315	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5316	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5317		/*
5318		 * Cache devices can always be removed.
5319		 */
5320		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5321		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5322		spa_load_l2cache(spa);
5323		spa->spa_l2cache.sav_sync = B_TRUE;
5324	} else if (vd != NULL && vd->vdev_islog) {
5325		ASSERT(!locked);
5326		ASSERT(vd == vd->vdev_top);
5327
5328		mg = vd->vdev_mg;
5329
5330		/*
5331		 * Stop allocating from this vdev.
5332		 */
5333		metaslab_group_passivate(mg);
5334
5335		/*
5336		 * Wait for the youngest allocations and frees to sync,
5337		 * and then wait for the deferral of those frees to finish.
5338		 */
5339		spa_vdev_config_exit(spa, NULL,
5340		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5341
5342		/*
5343		 * Attempt to evacuate the vdev.
5344		 */
5345		error = spa_vdev_remove_evacuate(spa, vd);
5346
5347		txg = spa_vdev_config_enter(spa);
5348
5349		/*
5350		 * If we couldn't evacuate the vdev, unwind.
5351		 */
5352		if (error) {
5353			metaslab_group_activate(mg);
5354			return (spa_vdev_exit(spa, NULL, txg, error));
5355		}
5356
5357		/*
5358		 * Clean up the vdev namespace.
5359		 */
5360		spa_vdev_remove_from_namespace(spa, vd);
5361
5362	} else if (vd != NULL) {
5363		/*
5364		 * Normal vdevs cannot be removed (yet).
5365		 */
5366		error = SET_ERROR(ENOTSUP);
5367	} else {
5368		/*
5369		 * There is no vdev of any kind with the specified guid.
5370		 */
5371		error = SET_ERROR(ENOENT);
5372	}
5373
5374	if (!locked)
5375		return (spa_vdev_exit(spa, NULL, txg, error));
5376
5377	return (error);
5378}
5379
5380/*
5381 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5382 * currently spared, so we can detach it.
5383 */
5384static vdev_t *
5385spa_vdev_resilver_done_hunt(vdev_t *vd)
5386{
5387	vdev_t *newvd, *oldvd;
5388
5389	for (int c = 0; c < vd->vdev_children; c++) {
5390		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5391		if (oldvd != NULL)
5392			return (oldvd);
5393	}
5394
5395	/*
5396	 * Check for a completed replacement.  We always consider the first
5397	 * vdev in the list to be the oldest vdev, and the last one to be
5398	 * the newest (see spa_vdev_attach() for how that works).  In
5399	 * the case where the newest vdev is faulted, we will not automatically
5400	 * remove it after a resilver completes.  This is OK as it will require
5401	 * user intervention to determine which disk the admin wishes to keep.
5402	 */
5403	if (vd->vdev_ops == &vdev_replacing_ops) {
5404		ASSERT(vd->vdev_children > 1);
5405
5406		newvd = vd->vdev_child[vd->vdev_children - 1];
5407		oldvd = vd->vdev_child[0];
5408
5409		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5410		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5411		    !vdev_dtl_required(oldvd))
5412			return (oldvd);
5413	}
5414
5415	/*
5416	 * Check for a completed resilver with the 'unspare' flag set.
5417	 */
5418	if (vd->vdev_ops == &vdev_spare_ops) {
5419		vdev_t *first = vd->vdev_child[0];
5420		vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5421
5422		if (last->vdev_unspare) {
5423			oldvd = first;
5424			newvd = last;
5425		} else if (first->vdev_unspare) {
5426			oldvd = last;
5427			newvd = first;
5428		} else {
5429			oldvd = NULL;
5430		}
5431
5432		if (oldvd != NULL &&
5433		    vdev_dtl_empty(newvd, DTL_MISSING) &&
5434		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5435		    !vdev_dtl_required(oldvd))
5436			return (oldvd);
5437
5438		/*
5439		 * If there are more than two spares attached to a disk,
5440		 * and those spares are not required, then we want to
5441		 * attempt to free them up now so that they can be used
5442		 * by other pools.  Once we're back down to a single
5443		 * disk+spare, we stop removing them.
5444		 */
5445		if (vd->vdev_children > 2) {
5446			newvd = vd->vdev_child[1];
5447
5448			if (newvd->vdev_isspare && last->vdev_isspare &&
5449			    vdev_dtl_empty(last, DTL_MISSING) &&
5450			    vdev_dtl_empty(last, DTL_OUTAGE) &&
5451			    !vdev_dtl_required(newvd))
5452				return (newvd);
5453		}
5454	}
5455
5456	return (NULL);
5457}
5458
5459static void
5460spa_vdev_resilver_done(spa_t *spa)
5461{
5462	vdev_t *vd, *pvd, *ppvd;
5463	uint64_t guid, sguid, pguid, ppguid;
5464
5465	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5466
5467	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5468		pvd = vd->vdev_parent;
5469		ppvd = pvd->vdev_parent;
5470		guid = vd->vdev_guid;
5471		pguid = pvd->vdev_guid;
5472		ppguid = ppvd->vdev_guid;
5473		sguid = 0;
5474		/*
5475		 * If we have just finished replacing a hot spared device, then
5476		 * we need to detach the parent's first child (the original hot
5477		 * spare) as well.
5478		 */
5479		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5480		    ppvd->vdev_children == 2) {
5481			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5482			sguid = ppvd->vdev_child[1]->vdev_guid;
5483		}
5484		ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5485
5486		spa_config_exit(spa, SCL_ALL, FTAG);
5487		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5488			return;
5489		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5490			return;
5491		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5492	}
5493
5494	spa_config_exit(spa, SCL_ALL, FTAG);
5495}
5496
5497/*
5498 * Update the stored path or FRU for this vdev.
5499 */
5500int
5501spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5502    boolean_t ispath)
5503{
5504	vdev_t *vd;
5505	boolean_t sync = B_FALSE;
5506
5507	ASSERT(spa_writeable(spa));
5508
5509	spa_vdev_state_enter(spa, SCL_ALL);
5510
5511	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5512		return (spa_vdev_state_exit(spa, NULL, ENOENT));
5513
5514	if (!vd->vdev_ops->vdev_op_leaf)
5515		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5516
5517	if (ispath) {
5518		if (strcmp(value, vd->vdev_path) != 0) {
5519			spa_strfree(vd->vdev_path);
5520			vd->vdev_path = spa_strdup(value);
5521			sync = B_TRUE;
5522		}
5523	} else {
5524		if (vd->vdev_fru == NULL) {
5525			vd->vdev_fru = spa_strdup(value);
5526			sync = B_TRUE;
5527		} else if (strcmp(value, vd->vdev_fru) != 0) {
5528			spa_strfree(vd->vdev_fru);
5529			vd->vdev_fru = spa_strdup(value);
5530			sync = B_TRUE;
5531		}
5532	}
5533
5534	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5535}
5536
5537int
5538spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5539{
5540	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5541}
5542
5543int
5544spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5545{
5546	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5547}
5548
5549/*
5550 * ==========================================================================
5551 * SPA Scanning
5552 * ==========================================================================
5553 */
5554
5555int
5556spa_scan_stop(spa_t *spa)
5557{
5558	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5559	if (dsl_scan_resilvering(spa->spa_dsl_pool))
5560		return (SET_ERROR(EBUSY));
5561	return (dsl_scan_cancel(spa->spa_dsl_pool));
5562}
5563
5564int
5565spa_scan(spa_t *spa, pool_scan_func_t func)
5566{
5567	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5568
5569	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5570		return (SET_ERROR(ENOTSUP));
5571
5572	/*
5573	 * If a resilver was requested, but there is no DTL on a
5574	 * writeable leaf device, we have nothing to do.
5575	 */
5576	if (func == POOL_SCAN_RESILVER &&
5577	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5578		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5579		return (0);
5580	}
5581
5582	return (dsl_scan(spa->spa_dsl_pool, func));
5583}
5584
5585/*
5586 * ==========================================================================
5587 * SPA async task processing
5588 * ==========================================================================
5589 */
5590
5591static void
5592spa_async_remove(spa_t *spa, vdev_t *vd)
5593{
5594	if (vd->vdev_remove_wanted) {
5595		vd->vdev_remove_wanted = B_FALSE;
5596		vd->vdev_delayed_close = B_FALSE;
5597		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5598
5599		/*
5600		 * We want to clear the stats, but we don't want to do a full
5601		 * vdev_clear() as that will cause us to throw away
5602		 * degraded/faulted state as well as attempt to reopen the
5603		 * device, all of which is a waste.
5604		 */
5605		vd->vdev_stat.vs_read_errors = 0;
5606		vd->vdev_stat.vs_write_errors = 0;
5607		vd->vdev_stat.vs_checksum_errors = 0;
5608
5609		vdev_state_dirty(vd->vdev_top);
5610	}
5611
5612	for (int c = 0; c < vd->vdev_children; c++)
5613		spa_async_remove(spa, vd->vdev_child[c]);
5614}
5615
5616static void
5617spa_async_probe(spa_t *spa, vdev_t *vd)
5618{
5619	if (vd->vdev_probe_wanted) {
5620		vd->vdev_probe_wanted = B_FALSE;
5621		vdev_reopen(vd);	/* vdev_open() does the actual probe */
5622	}
5623
5624	for (int c = 0; c < vd->vdev_children; c++)
5625		spa_async_probe(spa, vd->vdev_child[c]);
5626}
5627
5628static void
5629spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5630{
5631	sysevent_id_t eid;
5632	nvlist_t *attr;
5633	char *physpath;
5634
5635	if (!spa->spa_autoexpand)
5636		return;
5637
5638	for (int c = 0; c < vd->vdev_children; c++) {
5639		vdev_t *cvd = vd->vdev_child[c];
5640		spa_async_autoexpand(spa, cvd);
5641	}
5642
5643	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5644		return;
5645
5646	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5647	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5648
5649	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5650	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5651
5652	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5653	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5654
5655	nvlist_free(attr);
5656	kmem_free(physpath, MAXPATHLEN);
5657}
5658
5659static void
5660spa_async_thread(spa_t *spa)
5661{
5662	int tasks;
5663
5664	ASSERT(spa->spa_sync_on);
5665
5666	mutex_enter(&spa->spa_async_lock);
5667	tasks = spa->spa_async_tasks;
5668	spa->spa_async_tasks = 0;
5669	mutex_exit(&spa->spa_async_lock);
5670
5671	/*
5672	 * See if the config needs to be updated.
5673	 */
5674	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5675		uint64_t old_space, new_space;
5676
5677		mutex_enter(&spa_namespace_lock);
5678		old_space = metaslab_class_get_space(spa_normal_class(spa));
5679		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5680		new_space = metaslab_class_get_space(spa_normal_class(spa));
5681		mutex_exit(&spa_namespace_lock);
5682
5683		/*
5684		 * If the pool grew as a result of the config update,
5685		 * then log an internal history event.
5686		 */
5687		if (new_space != old_space) {
5688			spa_history_log_internal(spa, "vdev online", NULL,
5689			    "pool '%s' size: %llu(+%llu)",
5690			    spa_name(spa), new_space, new_space - old_space);
5691		}
5692	}
5693
5694	/*
5695	 * See if any devices need to be marked REMOVED.
5696	 */
5697	if (tasks & SPA_ASYNC_REMOVE) {
5698		spa_vdev_state_enter(spa, SCL_NONE);
5699		spa_async_remove(spa, spa->spa_root_vdev);
5700		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5701			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5702		for (int i = 0; i < spa->spa_spares.sav_count; i++)
5703			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5704		(void) spa_vdev_state_exit(spa, NULL, 0);
5705	}
5706
5707	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5708		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5709		spa_async_autoexpand(spa, spa->spa_root_vdev);
5710		spa_config_exit(spa, SCL_CONFIG, FTAG);
5711	}
5712
5713	/*
5714	 * See if any devices need to be probed.
5715	 */
5716	if (tasks & SPA_ASYNC_PROBE) {
5717		spa_vdev_state_enter(spa, SCL_NONE);
5718		spa_async_probe(spa, spa->spa_root_vdev);
5719		(void) spa_vdev_state_exit(spa, NULL, 0);
5720	}
5721
5722	/*
5723	 * If any devices are done replacing, detach them.
5724	 */
5725	if (tasks & SPA_ASYNC_RESILVER_DONE)
5726		spa_vdev_resilver_done(spa);
5727
5728	/*
5729	 * Kick off a resilver.
5730	 */
5731	if (tasks & SPA_ASYNC_RESILVER)
5732		dsl_resilver_restart(spa->spa_dsl_pool, 0);
5733
5734	/*
5735	 * Let the world know that we're done.
5736	 */
5737	mutex_enter(&spa->spa_async_lock);
5738	spa->spa_async_thread = NULL;
5739	cv_broadcast(&spa->spa_async_cv);
5740	mutex_exit(&spa->spa_async_lock);
5741	thread_exit();
5742}
5743
5744void
5745spa_async_suspend(spa_t *spa)
5746{
5747	mutex_enter(&spa->spa_async_lock);
5748	spa->spa_async_suspended++;
5749	while (spa->spa_async_thread != NULL)
5750		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5751	mutex_exit(&spa->spa_async_lock);
5752}
5753
5754void
5755spa_async_resume(spa_t *spa)
5756{
5757	mutex_enter(&spa->spa_async_lock);
5758	ASSERT(spa->spa_async_suspended != 0);
5759	spa->spa_async_suspended--;
5760	mutex_exit(&spa->spa_async_lock);
5761}
5762
5763static boolean_t
5764spa_async_tasks_pending(spa_t *spa)
5765{
5766	uint_t non_config_tasks;
5767	uint_t config_task;
5768	boolean_t config_task_suspended;
5769
5770	non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5771	config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5772	if (spa->spa_ccw_fail_time == 0) {
5773		config_task_suspended = B_FALSE;
5774	} else {
5775		config_task_suspended =
5776		    (gethrtime() - spa->spa_ccw_fail_time) <
5777		    (zfs_ccw_retry_interval * NANOSEC);
5778	}
5779
5780	return (non_config_tasks || (config_task && !config_task_suspended));
5781}
5782
5783static void
5784spa_async_dispatch(spa_t *spa)
5785{
5786	mutex_enter(&spa->spa_async_lock);
5787	if (spa_async_tasks_pending(spa) &&
5788	    !spa->spa_async_suspended &&
5789	    spa->spa_async_thread == NULL &&
5790	    rootdir != NULL)
5791		spa->spa_async_thread = thread_create(NULL, 0,
5792		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5793	mutex_exit(&spa->spa_async_lock);
5794}
5795
5796void
5797spa_async_request(spa_t *spa, int task)
5798{
5799	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5800	mutex_enter(&spa->spa_async_lock);
5801	spa->spa_async_tasks |= task;
5802	mutex_exit(&spa->spa_async_lock);
5803}
5804
5805/*
5806 * ==========================================================================
5807 * SPA syncing routines
5808 * ==========================================================================
5809 */
5810
5811static int
5812bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5813{
5814	bpobj_t *bpo = arg;
5815	bpobj_enqueue(bpo, bp, tx);
5816	return (0);
5817}
5818
5819static int
5820spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5821{
5822	zio_t *zio = arg;
5823
5824	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5825	    zio->io_flags));
5826	return (0);
5827}
5828
5829/*
5830 * Note: this simple function is not inlined to make it easier to dtrace the
5831 * amount of time spent syncing frees.
5832 */
5833static void
5834spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
5835{
5836	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5837	bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
5838	VERIFY(zio_wait(zio) == 0);
5839}
5840
5841/*
5842 * Note: this simple function is not inlined to make it easier to dtrace the
5843 * amount of time spent syncing deferred frees.
5844 */
5845static void
5846spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
5847{
5848	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5849	VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
5850	    spa_free_sync_cb, zio, tx), ==, 0);
5851	VERIFY0(zio_wait(zio));
5852}
5853
5854
5855static void
5856spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5857{
5858	char *packed = NULL;
5859	size_t bufsize;
5860	size_t nvsize = 0;
5861	dmu_buf_t *db;
5862
5863	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5864
5865	/*
5866	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5867	 * information.  This avoids the dmu_buf_will_dirty() path and
5868	 * saves us a pre-read to get data we don't actually care about.
5869	 */
5870	bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5871	packed = kmem_alloc(bufsize, KM_SLEEP);
5872
5873	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5874	    KM_SLEEP) == 0);
5875	bzero(packed + nvsize, bufsize - nvsize);
5876
5877	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5878
5879	kmem_free(packed, bufsize);
5880
5881	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5882	dmu_buf_will_dirty(db, tx);
5883	*(uint64_t *)db->db_data = nvsize;
5884	dmu_buf_rele(db, FTAG);
5885}
5886
5887static void
5888spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5889    const char *config, const char *entry)
5890{
5891	nvlist_t *nvroot;
5892	nvlist_t **list;
5893	int i;
5894
5895	if (!sav->sav_sync)
5896		return;
5897
5898	/*
5899	 * Update the MOS nvlist describing the list of available devices.
5900	 * spa_validate_aux() will have already made sure this nvlist is
5901	 * valid and the vdevs are labeled appropriately.
5902	 */
5903	if (sav->sav_object == 0) {
5904		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5905		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5906		    sizeof (uint64_t), tx);
5907		VERIFY(zap_update(spa->spa_meta_objset,
5908		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5909		    &sav->sav_object, tx) == 0);
5910	}
5911
5912	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5913	if (sav->sav_count == 0) {
5914		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5915	} else {
5916		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5917		for (i = 0; i < sav->sav_count; i++)
5918			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5919			    B_FALSE, VDEV_CONFIG_L2CACHE);
5920		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5921		    sav->sav_count) == 0);
5922		for (i = 0; i < sav->sav_count; i++)
5923			nvlist_free(list[i]);
5924		kmem_free(list, sav->sav_count * sizeof (void *));
5925	}
5926
5927	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5928	nvlist_free(nvroot);
5929
5930	sav->sav_sync = B_FALSE;
5931}
5932
5933static void
5934spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5935{
5936	nvlist_t *config;
5937
5938	if (list_is_empty(&spa->spa_config_dirty_list))
5939		return;
5940
5941	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5942
5943	config = spa_config_generate(spa, spa->spa_root_vdev,
5944	    dmu_tx_get_txg(tx), B_FALSE);
5945
5946	/*
5947	 * If we're upgrading the spa version then make sure that
5948	 * the config object gets updated with the correct version.
5949	 */
5950	if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5951		fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5952		    spa->spa_uberblock.ub_version);
5953
5954	spa_config_exit(spa, SCL_STATE, FTAG);
5955
5956	if (spa->spa_config_syncing)
5957		nvlist_free(spa->spa_config_syncing);
5958	spa->spa_config_syncing = config;
5959
5960	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5961}
5962
5963static void
5964spa_sync_version(void *arg, dmu_tx_t *tx)
5965{
5966	uint64_t *versionp = arg;
5967	uint64_t version = *versionp;
5968	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5969
5970	/*
5971	 * Setting the version is special cased when first creating the pool.
5972	 */
5973	ASSERT(tx->tx_txg != TXG_INITIAL);
5974
5975	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5976	ASSERT(version >= spa_version(spa));
5977
5978	spa->spa_uberblock.ub_version = version;
5979	vdev_config_dirty(spa->spa_root_vdev);
5980	spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5981}
5982
5983/*
5984 * Set zpool properties.
5985 */
5986static void
5987spa_sync_props(void *arg, dmu_tx_t *tx)
5988{
5989	nvlist_t *nvp = arg;
5990	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5991	objset_t *mos = spa->spa_meta_objset;
5992	nvpair_t *elem = NULL;
5993
5994	mutex_enter(&spa->spa_props_lock);
5995
5996	while ((elem = nvlist_next_nvpair(nvp, elem))) {
5997		uint64_t intval;
5998		char *strval, *fname;
5999		zpool_prop_t prop;
6000		const char *propname;
6001		zprop_type_t proptype;
6002		spa_feature_t fid;
6003
6004		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6005		case ZPROP_INVAL:
6006			/*
6007			 * We checked this earlier in spa_prop_validate().
6008			 */
6009			ASSERT(zpool_prop_feature(nvpair_name(elem)));
6010
6011			fname = strchr(nvpair_name(elem), '@') + 1;
6012			VERIFY0(zfeature_lookup_name(fname, &fid));
6013
6014			spa_feature_enable(spa, fid, tx);
6015			spa_history_log_internal(spa, "set", tx,
6016			    "%s=enabled", nvpair_name(elem));
6017			break;
6018
6019		case ZPOOL_PROP_VERSION:
6020			intval = fnvpair_value_uint64(elem);
6021			/*
6022			 * The version is synced seperatly before other
6023			 * properties and should be correct by now.
6024			 */
6025			ASSERT3U(spa_version(spa), >=, intval);
6026			break;
6027
6028		case ZPOOL_PROP_ALTROOT:
6029			/*
6030			 * 'altroot' is a non-persistent property. It should
6031			 * have been set temporarily at creation or import time.
6032			 */
6033			ASSERT(spa->spa_root != NULL);
6034			break;
6035
6036		case ZPOOL_PROP_READONLY:
6037		case ZPOOL_PROP_CACHEFILE:
6038			/*
6039			 * 'readonly' and 'cachefile' are also non-persisitent
6040			 * properties.
6041			 */
6042			break;
6043		case ZPOOL_PROP_COMMENT:
6044			strval = fnvpair_value_string(elem);
6045			if (spa->spa_comment != NULL)
6046				spa_strfree(spa->spa_comment);
6047			spa->spa_comment = spa_strdup(strval);