spa.c revision 3b2aab18808792cbd248a12f1edf139b89833c13
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 2011 Nexenta Systems, Inc.  All rights reserved.
25 * Copyright (c) 2012 by Delphix. All rights reserved.
26 */
27
28/*
29 * This file contains all the routines used when modifying on-disk SPA state.
30 * This includes opening, importing, destroying, exporting a pool, and syncing a
31 * pool.
32 */
33
34#include <sys/zfs_context.h>
35#include <sys/fm/fs/zfs.h>
36#include <sys/spa_impl.h>
37#include <sys/zio.h>
38#include <sys/zio_checksum.h>
39#include <sys/dmu.h>
40#include <sys/dmu_tx.h>
41#include <sys/zap.h>
42#include <sys/zil.h>
43#include <sys/ddt.h>
44#include <sys/vdev_impl.h>
45#include <sys/metaslab.h>
46#include <sys/metaslab_impl.h>
47#include <sys/uberblock_impl.h>
48#include <sys/txg.h>
49#include <sys/avl.h>
50#include <sys/dmu_traverse.h>
51#include <sys/dmu_objset.h>
52#include <sys/unique.h>
53#include <sys/dsl_pool.h>
54#include <sys/dsl_dataset.h>
55#include <sys/dsl_dir.h>
56#include <sys/dsl_prop.h>
57#include <sys/dsl_synctask.h>
58#include <sys/fs/zfs.h>
59#include <sys/arc.h>
60#include <sys/callb.h>
61#include <sys/systeminfo.h>
62#include <sys/spa_boot.h>
63#include <sys/zfs_ioctl.h>
64#include <sys/dsl_scan.h>
65#include <sys/zfeature.h>
66#include <sys/dsl_destroy.h>
67
68#ifdef	_KERNEL
69#include <sys/bootprops.h>
70#include <sys/callb.h>
71#include <sys/cpupart.h>
72#include <sys/pool.h>
73#include <sys/sysdc.h>
74#include <sys/zone.h>
75#endif	/* _KERNEL */
76
77#include "zfs_prop.h"
78#include "zfs_comutil.h"
79
80typedef enum zti_modes {
81	ZTI_MODE_FIXED,			/* value is # of threads (min 1) */
82	ZTI_MODE_ONLINE_PERCENT,	/* value is % of online CPUs */
83	ZTI_MODE_BATCH,			/* cpu-intensive; value is ignored */
84	ZTI_MODE_NULL,			/* don't create a taskq */
85	ZTI_NMODES
86} zti_modes_t;
87
88#define	ZTI_P(n, q)	{ ZTI_MODE_FIXED, (n), (q) }
89#define	ZTI_PCT(n)	{ ZTI_MODE_ONLINE_PERCENT, (n), 1 }
90#define	ZTI_BATCH	{ ZTI_MODE_BATCH, 0, 1 }
91#define	ZTI_NULL	{ ZTI_MODE_NULL, 0, 0 }
92
93#define	ZTI_N(n)	ZTI_P(n, 1)
94#define	ZTI_ONE		ZTI_N(1)
95
96typedef struct zio_taskq_info {
97	zti_modes_t zti_mode;
98	uint_t zti_value;
99	uint_t zti_count;
100} zio_taskq_info_t;
101
102static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
103	"issue", "issue_high", "intr", "intr_high"
104};
105
106/*
107 * This table defines the taskq settings for each ZFS I/O type. When
108 * initializing a pool, we use this table to create an appropriately sized
109 * taskq. Some operations are low volume and therefore have a small, static
110 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
111 * macros. Other operations process a large amount of data; the ZTI_BATCH
112 * macro causes us to create a taskq oriented for throughput. Some operations
113 * are so high frequency and short-lived that the taskq itself can become a a
114 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
115 * additional degree of parallelism specified by the number of threads per-
116 * taskq and the number of taskqs; when dispatching an event in this case, the
117 * particular taskq is chosen at random.
118 *
119 * The different taskq priorities are to handle the different contexts (issue
120 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
121 * need to be handled with minimum delay.
122 */
123const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
124	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
125	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* NULL */
126	{ ZTI_N(8),	ZTI_NULL,	ZTI_BATCH,	ZTI_NULL }, /* READ */
127	{ ZTI_BATCH,	ZTI_N(5),	ZTI_N(8),	ZTI_N(5) }, /* WRITE */
128	{ ZTI_P(12, 8),	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* FREE */
129	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* CLAIM */
130	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* IOCTL */
131};
132
133static void spa_sync_version(void *arg, dmu_tx_t *tx);
134static void spa_sync_props(void *arg, dmu_tx_t *tx);
135static boolean_t spa_has_active_shared_spare(spa_t *spa);
136static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
137    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
138    char **ereport);
139static void spa_vdev_resilver_done(spa_t *spa);
140
141uint_t		zio_taskq_batch_pct = 100;	/* 1 thread per cpu in pset */
142id_t		zio_taskq_psrset_bind = PS_NONE;
143boolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
144uint_t		zio_taskq_basedc = 80;		/* base duty cycle */
145
146boolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
147extern int	zfs_sync_pass_deferred_free;
148
149/*
150 * This (illegal) pool name is used when temporarily importing a spa_t in order
151 * to get the vdev stats associated with the imported devices.
152 */
153#define	TRYIMPORT_NAME	"$import"
154
155/*
156 * ==========================================================================
157 * SPA properties routines
158 * ==========================================================================
159 */
160
161/*
162 * Add a (source=src, propname=propval) list to an nvlist.
163 */
164static void
165spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
166    uint64_t intval, zprop_source_t src)
167{
168	const char *propname = zpool_prop_to_name(prop);
169	nvlist_t *propval;
170
171	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
172	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
173
174	if (strval != NULL)
175		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
176	else
177		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
178
179	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
180	nvlist_free(propval);
181}
182
183/*
184 * Get property values from the spa configuration.
185 */
186static void
187spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
188{
189	vdev_t *rvd = spa->spa_root_vdev;
190	dsl_pool_t *pool = spa->spa_dsl_pool;
191	uint64_t size;
192	uint64_t alloc;
193	uint64_t space;
194	uint64_t cap, version;
195	zprop_source_t src = ZPROP_SRC_NONE;
196	spa_config_dirent_t *dp;
197
198	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
199
200	if (rvd != NULL) {
201		alloc = metaslab_class_get_alloc(spa_normal_class(spa));
202		size = metaslab_class_get_space(spa_normal_class(spa));
203		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
204		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
205		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
206		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
207		    size - alloc, src);
208
209		space = 0;
210		for (int c = 0; c < rvd->vdev_children; c++) {
211			vdev_t *tvd = rvd->vdev_child[c];
212			space += tvd->vdev_max_asize - tvd->vdev_asize;
213		}
214		spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
215		    src);
216
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		dsl_dir_t *freedir = pool->dp_free_dir;
239
240		/*
241		 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
242		 * when opening pools before this version freedir will be NULL.
243		 */
244		if (freedir != NULL) {
245			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
246			    freedir->dd_phys->dd_used_bytes, src);
247		} else {
248			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
249			    NULL, 0, src);
250		}
251	}
252
253	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
254
255	if (spa->spa_comment != NULL) {
256		spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
257		    0, ZPROP_SRC_LOCAL);
258	}
259
260	if (spa->spa_root != NULL)
261		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
262		    0, ZPROP_SRC_LOCAL);
263
264	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
265		if (dp->scd_path == NULL) {
266			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
267			    "none", 0, ZPROP_SRC_LOCAL);
268		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
269			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
270			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
271		}
272	}
273}
274
275/*
276 * Get zpool property values.
277 */
278int
279spa_prop_get(spa_t *spa, nvlist_t **nvp)
280{
281	objset_t *mos = spa->spa_meta_objset;
282	zap_cursor_t zc;
283	zap_attribute_t za;
284	int err;
285
286	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
287
288	mutex_enter(&spa->spa_props_lock);
289
290	/*
291	 * Get properties from the spa config.
292	 */
293	spa_prop_get_config(spa, nvp);
294
295	/* If no pool property object, no more prop to get. */
296	if (mos == NULL || spa->spa_pool_props_object == 0) {
297		mutex_exit(&spa->spa_props_lock);
298		return (0);
299	}
300
301	/*
302	 * Get properties from the MOS pool property object.
303	 */
304	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
305	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
306	    zap_cursor_advance(&zc)) {
307		uint64_t intval = 0;
308		char *strval = NULL;
309		zprop_source_t src = ZPROP_SRC_DEFAULT;
310		zpool_prop_t prop;
311
312		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
313			continue;
314
315		switch (za.za_integer_length) {
316		case 8:
317			/* integer property */
318			if (za.za_first_integer !=
319			    zpool_prop_default_numeric(prop))
320				src = ZPROP_SRC_LOCAL;
321
322			if (prop == ZPOOL_PROP_BOOTFS) {
323				dsl_pool_t *dp;
324				dsl_dataset_t *ds = NULL;
325
326				dp = spa_get_dsl(spa);
327				dsl_pool_config_enter(dp, FTAG);
328				if (err = dsl_dataset_hold_obj(dp,
329				    za.za_first_integer, FTAG, &ds)) {
330					dsl_pool_config_exit(dp, FTAG);
331					break;
332				}
333
334				strval = kmem_alloc(
335				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
336				    KM_SLEEP);
337				dsl_dataset_name(ds, strval);
338				dsl_dataset_rele(ds, FTAG);
339				dsl_pool_config_exit(dp, FTAG);
340			} else {
341				strval = NULL;
342				intval = za.za_first_integer;
343			}
344
345			spa_prop_add_list(*nvp, prop, strval, intval, src);
346
347			if (strval != NULL)
348				kmem_free(strval,
349				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
350
351			break;
352
353		case 1:
354			/* string property */
355			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
356			err = zap_lookup(mos, spa->spa_pool_props_object,
357			    za.za_name, 1, za.za_num_integers, strval);
358			if (err) {
359				kmem_free(strval, za.za_num_integers);
360				break;
361			}
362			spa_prop_add_list(*nvp, prop, strval, 0, src);
363			kmem_free(strval, za.za_num_integers);
364			break;
365
366		default:
367			break;
368		}
369	}
370	zap_cursor_fini(&zc);
371	mutex_exit(&spa->spa_props_lock);
372out:
373	if (err && err != ENOENT) {
374		nvlist_free(*nvp);
375		*nvp = NULL;
376		return (err);
377	}
378
379	return (0);
380}
381
382/*
383 * Validate the given pool properties nvlist and modify the list
384 * for the property values to be set.
385 */
386static int
387spa_prop_validate(spa_t *spa, nvlist_t *props)
388{
389	nvpair_t *elem;
390	int error = 0, reset_bootfs = 0;
391	uint64_t objnum = 0;
392	boolean_t has_feature = B_FALSE;
393
394	elem = NULL;
395	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
396		uint64_t intval;
397		char *strval, *slash, *check, *fname;
398		const char *propname = nvpair_name(elem);
399		zpool_prop_t prop = zpool_name_to_prop(propname);
400
401		switch (prop) {
402		case ZPROP_INVAL:
403			if (!zpool_prop_feature(propname)) {
404				error = EINVAL;
405				break;
406			}
407
408			/*
409			 * Sanitize the input.
410			 */
411			if (nvpair_type(elem) != DATA_TYPE_UINT64) {
412				error = EINVAL;
413				break;
414			}
415
416			if (nvpair_value_uint64(elem, &intval) != 0) {
417				error = EINVAL;
418				break;
419			}
420
421			if (intval != 0) {
422				error = EINVAL;
423				break;
424			}
425
426			fname = strchr(propname, '@') + 1;
427			if (zfeature_lookup_name(fname, NULL) != 0) {
428				error = EINVAL;
429				break;
430			}
431
432			has_feature = B_TRUE;
433			break;
434
435		case ZPOOL_PROP_VERSION:
436			error = nvpair_value_uint64(elem, &intval);
437			if (!error &&
438			    (intval < spa_version(spa) ||
439			    intval > SPA_VERSION_BEFORE_FEATURES ||
440			    has_feature))
441				error = EINVAL;
442			break;
443
444		case ZPOOL_PROP_DELEGATION:
445		case ZPOOL_PROP_AUTOREPLACE:
446		case ZPOOL_PROP_LISTSNAPS:
447		case ZPOOL_PROP_AUTOEXPAND:
448			error = nvpair_value_uint64(elem, &intval);
449			if (!error && intval > 1)
450				error = EINVAL;
451			break;
452
453		case ZPOOL_PROP_BOOTFS:
454			/*
455			 * If the pool version is less than SPA_VERSION_BOOTFS,
456			 * or the pool is still being created (version == 0),
457			 * the bootfs property cannot be set.
458			 */
459			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
460				error = ENOTSUP;
461				break;
462			}
463
464			/*
465			 * Make sure the vdev config is bootable
466			 */
467			if (!vdev_is_bootable(spa->spa_root_vdev)) {
468				error = ENOTSUP;
469				break;
470			}
471
472			reset_bootfs = 1;
473
474			error = nvpair_value_string(elem, &strval);
475
476			if (!error) {
477				objset_t *os;
478				uint64_t compress;
479
480				if (strval == NULL || strval[0] == '\0') {
481					objnum = zpool_prop_default_numeric(
482					    ZPOOL_PROP_BOOTFS);
483					break;
484				}
485
486				if (error = dmu_objset_hold(strval, FTAG, &os))
487					break;
488
489				/* Must be ZPL and not gzip compressed. */
490
491				if (dmu_objset_type(os) != DMU_OST_ZFS) {
492					error = ENOTSUP;
493				} else if ((error =
494				    dsl_prop_get_int_ds(dmu_objset_ds(os),
495				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
496				    &compress)) == 0 &&
497				    !BOOTFS_COMPRESS_VALID(compress)) {
498					error = ENOTSUP;
499				} else {
500					objnum = dmu_objset_id(os);
501				}
502				dmu_objset_rele(os, FTAG);
503			}
504			break;
505
506		case ZPOOL_PROP_FAILUREMODE:
507			error = nvpair_value_uint64(elem, &intval);
508			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
509			    intval > ZIO_FAILURE_MODE_PANIC))
510				error = EINVAL;
511
512			/*
513			 * This is a special case which only occurs when
514			 * the pool has completely failed. This allows
515			 * the user to change the in-core failmode property
516			 * without syncing it out to disk (I/Os might
517			 * currently be blocked). We do this by returning
518			 * EIO to the caller (spa_prop_set) to trick it
519			 * into thinking we encountered a property validation
520			 * error.
521			 */
522			if (!error && spa_suspended(spa)) {
523				spa->spa_failmode = intval;
524				error = EIO;
525			}
526			break;
527
528		case ZPOOL_PROP_CACHEFILE:
529			if ((error = nvpair_value_string(elem, &strval)) != 0)
530				break;
531
532			if (strval[0] == '\0')
533				break;
534
535			if (strcmp(strval, "none") == 0)
536				break;
537
538			if (strval[0] != '/') {
539				error = EINVAL;
540				break;
541			}
542
543			slash = strrchr(strval, '/');
544			ASSERT(slash != NULL);
545
546			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
547			    strcmp(slash, "/..") == 0)
548				error = EINVAL;
549			break;
550
551		case ZPOOL_PROP_COMMENT:
552			if ((error = nvpair_value_string(elem, &strval)) != 0)
553				break;
554			for (check = strval; *check != '\0'; check++) {
555				/*
556				 * The kernel doesn't have an easy isprint()
557				 * check.  For this kernel check, we merely
558				 * check ASCII apart from DEL.  Fix this if
559				 * there is an easy-to-use kernel isprint().
560				 */
561				if (*check >= 0x7f) {
562					error = EINVAL;
563					break;
564				}
565				check++;
566			}
567			if (strlen(strval) > ZPROP_MAX_COMMENT)
568				error = E2BIG;
569			break;
570
571		case ZPOOL_PROP_DEDUPDITTO:
572			if (spa_version(spa) < SPA_VERSION_DEDUP)
573				error = ENOTSUP;
574			else
575				error = nvpair_value_uint64(elem, &intval);
576			if (error == 0 &&
577			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
578				error = EINVAL;
579			break;
580		}
581
582		if (error)
583			break;
584	}
585
586	if (!error && reset_bootfs) {
587		error = nvlist_remove(props,
588		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
589
590		if (!error) {
591			error = nvlist_add_uint64(props,
592			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
593		}
594	}
595
596	return (error);
597}
598
599void
600spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
601{
602	char *cachefile;
603	spa_config_dirent_t *dp;
604
605	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
606	    &cachefile) != 0)
607		return;
608
609	dp = kmem_alloc(sizeof (spa_config_dirent_t),
610	    KM_SLEEP);
611
612	if (cachefile[0] == '\0')
613		dp->scd_path = spa_strdup(spa_config_path);
614	else if (strcmp(cachefile, "none") == 0)
615		dp->scd_path = NULL;
616	else
617		dp->scd_path = spa_strdup(cachefile);
618
619	list_insert_head(&spa->spa_config_list, dp);
620	if (need_sync)
621		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
622}
623
624int
625spa_prop_set(spa_t *spa, nvlist_t *nvp)
626{
627	int error;
628	nvpair_t *elem = NULL;
629	boolean_t need_sync = B_FALSE;
630
631	if ((error = spa_prop_validate(spa, nvp)) != 0)
632		return (error);
633
634	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
635		zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
636
637		if (prop == ZPOOL_PROP_CACHEFILE ||
638		    prop == ZPOOL_PROP_ALTROOT ||
639		    prop == ZPOOL_PROP_READONLY)
640			continue;
641
642		if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
643			uint64_t ver;
644
645			if (prop == ZPOOL_PROP_VERSION) {
646				VERIFY(nvpair_value_uint64(elem, &ver) == 0);
647			} else {
648				ASSERT(zpool_prop_feature(nvpair_name(elem)));
649				ver = SPA_VERSION_FEATURES;
650				need_sync = B_TRUE;
651			}
652
653			/* Save time if the version is already set. */
654			if (ver == spa_version(spa))
655				continue;
656
657			/*
658			 * In addition to the pool directory object, we might
659			 * create the pool properties object, the features for
660			 * read object, the features for write object, or the
661			 * feature descriptions object.
662			 */
663			error = dsl_sync_task(spa->spa_name, NULL,
664			    spa_sync_version, &ver, 6);
665			if (error)
666				return (error);
667			continue;
668		}
669
670		need_sync = B_TRUE;
671		break;
672	}
673
674	if (need_sync) {
675		return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
676		    nvp, 6));
677	}
678
679	return (0);
680}
681
682/*
683 * If the bootfs property value is dsobj, clear it.
684 */
685void
686spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
687{
688	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
689		VERIFY(zap_remove(spa->spa_meta_objset,
690		    spa->spa_pool_props_object,
691		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
692		spa->spa_bootfs = 0;
693	}
694}
695
696/*ARGSUSED*/
697static int
698spa_change_guid_check(void *arg, dmu_tx_t *tx)
699{
700	uint64_t *newguid = arg;
701	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
702	vdev_t *rvd = spa->spa_root_vdev;
703	uint64_t vdev_state;
704
705	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
706	vdev_state = rvd->vdev_state;
707	spa_config_exit(spa, SCL_STATE, FTAG);
708
709	if (vdev_state != VDEV_STATE_HEALTHY)
710		return (ENXIO);
711
712	ASSERT3U(spa_guid(spa), !=, *newguid);
713
714	return (0);
715}
716
717static void
718spa_change_guid_sync(void *arg, dmu_tx_t *tx)
719{
720	uint64_t *newguid = arg;
721	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
722	uint64_t oldguid;
723	vdev_t *rvd = spa->spa_root_vdev;
724
725	oldguid = spa_guid(spa);
726
727	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
728	rvd->vdev_guid = *newguid;
729	rvd->vdev_guid_sum += (*newguid - oldguid);
730	vdev_config_dirty(rvd);
731	spa_config_exit(spa, SCL_STATE, FTAG);
732
733	spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
734	    oldguid, *newguid);
735}
736
737/*
738 * Change the GUID for the pool.  This is done so that we can later
739 * re-import a pool built from a clone of our own vdevs.  We will modify
740 * the root vdev's guid, our own pool guid, and then mark all of our
741 * vdevs dirty.  Note that we must make sure that all our vdevs are
742 * online when we do this, or else any vdevs that weren't present
743 * would be orphaned from our pool.  We are also going to issue a
744 * sysevent to update any watchers.
745 */
746int
747spa_change_guid(spa_t *spa)
748{
749	int error;
750	uint64_t guid;
751
752	mutex_enter(&spa_namespace_lock);
753	guid = spa_generate_guid(NULL);
754
755	error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
756	    spa_change_guid_sync, &guid, 5);
757
758	if (error == 0) {
759		spa_config_sync(spa, B_FALSE, B_TRUE);
760		spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
761	}
762
763	mutex_exit(&spa_namespace_lock);
764
765	return (error);
766}
767
768/*
769 * ==========================================================================
770 * SPA state manipulation (open/create/destroy/import/export)
771 * ==========================================================================
772 */
773
774static int
775spa_error_entry_compare(const void *a, const void *b)
776{
777	spa_error_entry_t *sa = (spa_error_entry_t *)a;
778	spa_error_entry_t *sb = (spa_error_entry_t *)b;
779	int ret;
780
781	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
782	    sizeof (zbookmark_t));
783
784	if (ret < 0)
785		return (-1);
786	else if (ret > 0)
787		return (1);
788	else
789		return (0);
790}
791
792/*
793 * Utility function which retrieves copies of the current logs and
794 * re-initializes them in the process.
795 */
796void
797spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
798{
799	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
800
801	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
802	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
803
804	avl_create(&spa->spa_errlist_scrub,
805	    spa_error_entry_compare, sizeof (spa_error_entry_t),
806	    offsetof(spa_error_entry_t, se_avl));
807	avl_create(&spa->spa_errlist_last,
808	    spa_error_entry_compare, sizeof (spa_error_entry_t),
809	    offsetof(spa_error_entry_t, se_avl));
810}
811
812static void
813spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
814{
815	const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
816	enum zti_modes mode = ztip->zti_mode;
817	uint_t value = ztip->zti_value;
818	uint_t count = ztip->zti_count;
819	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
820	char name[32];
821	uint_t flags = 0;
822	boolean_t batch = B_FALSE;
823
824	if (mode == ZTI_MODE_NULL) {
825		tqs->stqs_count = 0;
826		tqs->stqs_taskq = NULL;
827		return;
828	}
829
830	ASSERT3U(count, >, 0);
831
832	tqs->stqs_count = count;
833	tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
834
835	for (uint_t i = 0; i < count; i++) {
836		taskq_t *tq;
837
838		switch (mode) {
839		case ZTI_MODE_FIXED:
840			ASSERT3U(value, >=, 1);
841			value = MAX(value, 1);
842			break;
843
844		case ZTI_MODE_BATCH:
845			batch = B_TRUE;
846			flags |= TASKQ_THREADS_CPU_PCT;
847			value = zio_taskq_batch_pct;
848			break;
849
850		case ZTI_MODE_ONLINE_PERCENT:
851			flags |= TASKQ_THREADS_CPU_PCT;
852			break;
853
854		default:
855			panic("unrecognized mode for %s_%s taskq (%u:%u) in "
856			    "spa_activate()",
857			    zio_type_name[t], zio_taskq_types[q], mode, value);
858			break;
859		}
860
861		if (count > 1) {
862			(void) snprintf(name, sizeof (name), "%s_%s_%u",
863			    zio_type_name[t], zio_taskq_types[q], i);
864		} else {
865			(void) snprintf(name, sizeof (name), "%s_%s",
866			    zio_type_name[t], zio_taskq_types[q]);
867		}
868
869		if (zio_taskq_sysdc && spa->spa_proc != &p0) {
870			if (batch)
871				flags |= TASKQ_DC_BATCH;
872
873			tq = taskq_create_sysdc(name, value, 50, INT_MAX,
874			    spa->spa_proc, zio_taskq_basedc, flags);
875		} else {
876			tq = taskq_create_proc(name, value, maxclsyspri, 50,
877			    INT_MAX, spa->spa_proc, flags);
878		}
879
880		tqs->stqs_taskq[i] = tq;
881	}
882}
883
884static void
885spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
886{
887	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
888
889	if (tqs->stqs_taskq == NULL) {
890		ASSERT0(tqs->stqs_count);
891		return;
892	}
893
894	for (uint_t i = 0; i < tqs->stqs_count; i++) {
895		ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
896		taskq_destroy(tqs->stqs_taskq[i]);
897	}
898
899	kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
900	tqs->stqs_taskq = NULL;
901}
902
903/*
904 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
905 * Note that a type may have multiple discrete taskqs to avoid lock contention
906 * on the taskq itself. In that case we choose which taskq at random by using
907 * the low bits of gethrtime().
908 */
909void
910spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
911    task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
912{
913	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
914	taskq_t *tq;
915
916	ASSERT3P(tqs->stqs_taskq, !=, NULL);
917	ASSERT3U(tqs->stqs_count, !=, 0);
918
919	if (tqs->stqs_count == 1) {
920		tq = tqs->stqs_taskq[0];
921	} else {
922		tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
923	}
924
925	taskq_dispatch_ent(tq, func, arg, flags, ent);
926}
927
928static void
929spa_create_zio_taskqs(spa_t *spa)
930{
931	for (int t = 0; t < ZIO_TYPES; t++) {
932		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
933			spa_taskqs_init(spa, t, q);
934		}
935	}
936}
937
938#ifdef _KERNEL
939static void
940spa_thread(void *arg)
941{
942	callb_cpr_t cprinfo;
943
944	spa_t *spa = arg;
945	user_t *pu = PTOU(curproc);
946
947	CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
948	    spa->spa_name);
949
950	ASSERT(curproc != &p0);
951	(void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
952	    "zpool-%s", spa->spa_name);
953	(void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
954
955	/* bind this thread to the requested psrset */
956	if (zio_taskq_psrset_bind != PS_NONE) {
957		pool_lock();
958		mutex_enter(&cpu_lock);
959		mutex_enter(&pidlock);
960		mutex_enter(&curproc->p_lock);
961
962		if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
963		    0, NULL, NULL) == 0)  {
964			curthread->t_bind_pset = zio_taskq_psrset_bind;
965		} else {
966			cmn_err(CE_WARN,
967			    "Couldn't bind process for zfs pool \"%s\" to "
968			    "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
969		}
970
971		mutex_exit(&curproc->p_lock);
972		mutex_exit(&pidlock);
973		mutex_exit(&cpu_lock);
974		pool_unlock();
975	}
976
977	if (zio_taskq_sysdc) {
978		sysdc_thread_enter(curthread, 100, 0);
979	}
980
981	spa->spa_proc = curproc;
982	spa->spa_did = curthread->t_did;
983
984	spa_create_zio_taskqs(spa);
985
986	mutex_enter(&spa->spa_proc_lock);
987	ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
988
989	spa->spa_proc_state = SPA_PROC_ACTIVE;
990	cv_broadcast(&spa->spa_proc_cv);
991
992	CALLB_CPR_SAFE_BEGIN(&cprinfo);
993	while (spa->spa_proc_state == SPA_PROC_ACTIVE)
994		cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
995	CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
996
997	ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
998	spa->spa_proc_state = SPA_PROC_GONE;
999	spa->spa_proc = &p0;
1000	cv_broadcast(&spa->spa_proc_cv);
1001	CALLB_CPR_EXIT(&cprinfo);	/* drops spa_proc_lock */
1002
1003	mutex_enter(&curproc->p_lock);
1004	lwp_exit();
1005}
1006#endif
1007
1008/*
1009 * Activate an uninitialized pool.
1010 */
1011static void
1012spa_activate(spa_t *spa, int mode)
1013{
1014	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1015
1016	spa->spa_state = POOL_STATE_ACTIVE;
1017	spa->spa_mode = mode;
1018
1019	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1020	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1021
1022	/* Try to create a covering process */
1023	mutex_enter(&spa->spa_proc_lock);
1024	ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1025	ASSERT(spa->spa_proc == &p0);
1026	spa->spa_did = 0;
1027
1028	/* Only create a process if we're going to be around a while. */
1029	if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1030		if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1031		    NULL, 0) == 0) {
1032			spa->spa_proc_state = SPA_PROC_CREATED;
1033			while (spa->spa_proc_state == SPA_PROC_CREATED) {
1034				cv_wait(&spa->spa_proc_cv,
1035				    &spa->spa_proc_lock);
1036			}
1037			ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1038			ASSERT(spa->spa_proc != &p0);
1039			ASSERT(spa->spa_did != 0);
1040		} else {
1041#ifdef _KERNEL
1042			cmn_err(CE_WARN,
1043			    "Couldn't create process for zfs pool \"%s\"\n",
1044			    spa->spa_name);
1045#endif
1046		}
1047	}
1048	mutex_exit(&spa->spa_proc_lock);
1049
1050	/* If we didn't create a process, we need to create our taskqs. */
1051	if (spa->spa_proc == &p0) {
1052		spa_create_zio_taskqs(spa);
1053	}
1054
1055	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1056	    offsetof(vdev_t, vdev_config_dirty_node));
1057	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1058	    offsetof(vdev_t, vdev_state_dirty_node));
1059
1060	txg_list_create(&spa->spa_vdev_txg_list,
1061	    offsetof(struct vdev, vdev_txg_node));
1062
1063	avl_create(&spa->spa_errlist_scrub,
1064	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1065	    offsetof(spa_error_entry_t, se_avl));
1066	avl_create(&spa->spa_errlist_last,
1067	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1068	    offsetof(spa_error_entry_t, se_avl));
1069}
1070
1071/*
1072 * Opposite of spa_activate().
1073 */
1074static void
1075spa_deactivate(spa_t *spa)
1076{
1077	ASSERT(spa->spa_sync_on == B_FALSE);
1078	ASSERT(spa->spa_dsl_pool == NULL);
1079	ASSERT(spa->spa_root_vdev == NULL);
1080	ASSERT(spa->spa_async_zio_root == NULL);
1081	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1082
1083	txg_list_destroy(&spa->spa_vdev_txg_list);
1084
1085	list_destroy(&spa->spa_config_dirty_list);
1086	list_destroy(&spa->spa_state_dirty_list);
1087
1088	for (int t = 0; t < ZIO_TYPES; t++) {
1089		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1090			spa_taskqs_fini(spa, t, q);
1091		}
1092	}
1093
1094	metaslab_class_destroy(spa->spa_normal_class);
1095	spa->spa_normal_class = NULL;
1096
1097	metaslab_class_destroy(spa->spa_log_class);
1098	spa->spa_log_class = NULL;
1099
1100	/*
1101	 * If this was part of an import or the open otherwise failed, we may
1102	 * still have errors left in the queues.  Empty them just in case.
1103	 */
1104	spa_errlog_drain(spa);
1105
1106	avl_destroy(&spa->spa_errlist_scrub);
1107	avl_destroy(&spa->spa_errlist_last);
1108
1109	spa->spa_state = POOL_STATE_UNINITIALIZED;
1110
1111	mutex_enter(&spa->spa_proc_lock);
1112	if (spa->spa_proc_state != SPA_PROC_NONE) {
1113		ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1114		spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1115		cv_broadcast(&spa->spa_proc_cv);
1116		while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1117			ASSERT(spa->spa_proc != &p0);
1118			cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1119		}
1120		ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1121		spa->spa_proc_state = SPA_PROC_NONE;
1122	}
1123	ASSERT(spa->spa_proc == &p0);
1124	mutex_exit(&spa->spa_proc_lock);
1125
1126	/*
1127	 * We want to make sure spa_thread() has actually exited the ZFS
1128	 * module, so that the module can't be unloaded out from underneath
1129	 * it.
1130	 */
1131	if (spa->spa_did != 0) {
1132		thread_join(spa->spa_did);
1133		spa->spa_did = 0;
1134	}
1135}
1136
1137/*
1138 * Verify a pool configuration, and construct the vdev tree appropriately.  This
1139 * will create all the necessary vdevs in the appropriate layout, with each vdev
1140 * in the CLOSED state.  This will prep the pool before open/creation/import.
1141 * All vdev validation is done by the vdev_alloc() routine.
1142 */
1143static int
1144spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1145    uint_t id, int atype)
1146{
1147	nvlist_t **child;
1148	uint_t children;
1149	int error;
1150
1151	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1152		return (error);
1153
1154	if ((*vdp)->vdev_ops->vdev_op_leaf)
1155		return (0);
1156
1157	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1158	    &child, &children);
1159
1160	if (error == ENOENT)
1161		return (0);
1162
1163	if (error) {
1164		vdev_free(*vdp);
1165		*vdp = NULL;
1166		return (EINVAL);
1167	}
1168
1169	for (int c = 0; c < children; c++) {
1170		vdev_t *vd;
1171		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1172		    atype)) != 0) {
1173			vdev_free(*vdp);
1174			*vdp = NULL;
1175			return (error);
1176		}
1177	}
1178
1179	ASSERT(*vdp != NULL);
1180
1181	return (0);
1182}
1183
1184/*
1185 * Opposite of spa_load().
1186 */
1187static void
1188spa_unload(spa_t *spa)
1189{
1190	int i;
1191
1192	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1193
1194	/*
1195	 * Stop async tasks.
1196	 */
1197	spa_async_suspend(spa);
1198
1199	/*
1200	 * Stop syncing.
1201	 */
1202	if (spa->spa_sync_on) {
1203		txg_sync_stop(spa->spa_dsl_pool);
1204		spa->spa_sync_on = B_FALSE;
1205	}
1206
1207	/*
1208	 * Wait for any outstanding async I/O to complete.
1209	 */
1210	if (spa->spa_async_zio_root != NULL) {
1211		(void) zio_wait(spa->spa_async_zio_root);
1212		spa->spa_async_zio_root = NULL;
1213	}
1214
1215	bpobj_close(&spa->spa_deferred_bpobj);
1216
1217	/*
1218	 * Close the dsl pool.
1219	 */
1220	if (spa->spa_dsl_pool) {
1221		dsl_pool_close(spa->spa_dsl_pool);
1222		spa->spa_dsl_pool = NULL;
1223		spa->spa_meta_objset = NULL;
1224	}
1225
1226	ddt_unload(spa);
1227
1228	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1229
1230	/*
1231	 * Drop and purge level 2 cache
1232	 */
1233	spa_l2cache_drop(spa);
1234
1235	/*
1236	 * Close all vdevs.
1237	 */
1238	if (spa->spa_root_vdev)
1239		vdev_free(spa->spa_root_vdev);
1240	ASSERT(spa->spa_root_vdev == NULL);
1241
1242	for (i = 0; i < spa->spa_spares.sav_count; i++)
1243		vdev_free(spa->spa_spares.sav_vdevs[i]);
1244	if (spa->spa_spares.sav_vdevs) {
1245		kmem_free(spa->spa_spares.sav_vdevs,
1246		    spa->spa_spares.sav_count * sizeof (void *));
1247		spa->spa_spares.sav_vdevs = NULL;
1248	}
1249	if (spa->spa_spares.sav_config) {
1250		nvlist_free(spa->spa_spares.sav_config);
1251		spa->spa_spares.sav_config = NULL;
1252	}
1253	spa->spa_spares.sav_count = 0;
1254
1255	for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1256		vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1257		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1258	}
1259	if (spa->spa_l2cache.sav_vdevs) {
1260		kmem_free(spa->spa_l2cache.sav_vdevs,
1261		    spa->spa_l2cache.sav_count * sizeof (void *));
1262		spa->spa_l2cache.sav_vdevs = NULL;
1263	}
1264	if (spa->spa_l2cache.sav_config) {
1265		nvlist_free(spa->spa_l2cache.sav_config);
1266		spa->spa_l2cache.sav_config = NULL;
1267	}
1268	spa->spa_l2cache.sav_count = 0;
1269
1270	spa->spa_async_suspended = 0;
1271
1272	if (spa->spa_comment != NULL) {
1273		spa_strfree(spa->spa_comment);
1274		spa->spa_comment = NULL;
1275	}
1276
1277	spa_config_exit(spa, SCL_ALL, FTAG);
1278}
1279
1280/*
1281 * Load (or re-load) the current list of vdevs describing the active spares for
1282 * this pool.  When this is called, we have some form of basic information in
1283 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1284 * then re-generate a more complete list including status information.
1285 */
1286static void
1287spa_load_spares(spa_t *spa)
1288{
1289	nvlist_t **spares;
1290	uint_t nspares;
1291	int i;
1292	vdev_t *vd, *tvd;
1293
1294	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1295
1296	/*
1297	 * First, close and free any existing spare vdevs.
1298	 */
1299	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1300		vd = spa->spa_spares.sav_vdevs[i];
1301
1302		/* Undo the call to spa_activate() below */
1303		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1304		    B_FALSE)) != NULL && tvd->vdev_isspare)
1305			spa_spare_remove(tvd);
1306		vdev_close(vd);
1307		vdev_free(vd);
1308	}
1309
1310	if (spa->spa_spares.sav_vdevs)
1311		kmem_free(spa->spa_spares.sav_vdevs,
1312		    spa->spa_spares.sav_count * sizeof (void *));
1313
1314	if (spa->spa_spares.sav_config == NULL)
1315		nspares = 0;
1316	else
1317		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1318		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1319
1320	spa->spa_spares.sav_count = (int)nspares;
1321	spa->spa_spares.sav_vdevs = NULL;
1322
1323	if (nspares == 0)
1324		return;
1325
1326	/*
1327	 * Construct the array of vdevs, opening them to get status in the
1328	 * process.   For each spare, there is potentially two different vdev_t
1329	 * structures associated with it: one in the list of spares (used only
1330	 * for basic validation purposes) and one in the active vdev
1331	 * configuration (if it's spared in).  During this phase we open and
1332	 * validate each vdev on the spare list.  If the vdev also exists in the
1333	 * active configuration, then we also mark this vdev as an active spare.
1334	 */
1335	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1336	    KM_SLEEP);
1337	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1338		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1339		    VDEV_ALLOC_SPARE) == 0);
1340		ASSERT(vd != NULL);
1341
1342		spa->spa_spares.sav_vdevs[i] = vd;
1343
1344		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1345		    B_FALSE)) != NULL) {
1346			if (!tvd->vdev_isspare)
1347				spa_spare_add(tvd);
1348
1349			/*
1350			 * We only mark the spare active if we were successfully
1351			 * able to load the vdev.  Otherwise, importing a pool
1352			 * with a bad active spare would result in strange
1353			 * behavior, because multiple pool would think the spare
1354			 * is actively in use.
1355			 *
1356			 * There is a vulnerability here to an equally bizarre
1357			 * circumstance, where a dead active spare is later
1358			 * brought back to life (onlined or otherwise).  Given
1359			 * the rarity of this scenario, and the extra complexity
1360			 * it adds, we ignore the possibility.
1361			 */
1362			if (!vdev_is_dead(tvd))
1363				spa_spare_activate(tvd);
1364		}
1365
1366		vd->vdev_top = vd;
1367		vd->vdev_aux = &spa->spa_spares;
1368
1369		if (vdev_open(vd) != 0)
1370			continue;
1371
1372		if (vdev_validate_aux(vd) == 0)
1373			spa_spare_add(vd);
1374	}
1375
1376	/*
1377	 * Recompute the stashed list of spares, with status information
1378	 * this time.
1379	 */
1380	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1381	    DATA_TYPE_NVLIST_ARRAY) == 0);
1382
1383	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1384	    KM_SLEEP);
1385	for (i = 0; i < spa->spa_spares.sav_count; i++)
1386		spares[i] = vdev_config_generate(spa,
1387		    spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1388	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1389	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1390	for (i = 0; i < spa->spa_spares.sav_count; i++)
1391		nvlist_free(spares[i]);
1392	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1393}
1394
1395/*
1396 * Load (or re-load) the current list of vdevs describing the active l2cache for
1397 * this pool.  When this is called, we have some form of basic information in
1398 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1399 * then re-generate a more complete list including status information.
1400 * Devices which are already active have their details maintained, and are
1401 * not re-opened.
1402 */
1403static void
1404spa_load_l2cache(spa_t *spa)
1405{
1406	nvlist_t **l2cache;
1407	uint_t nl2cache;
1408	int i, j, oldnvdevs;
1409	uint64_t guid;
1410	vdev_t *vd, **oldvdevs, **newvdevs;
1411	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1412
1413	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1414
1415	if (sav->sav_config != NULL) {
1416		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1417		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1418		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1419	} else {
1420		nl2cache = 0;
1421		newvdevs = NULL;
1422	}
1423
1424	oldvdevs = sav->sav_vdevs;
1425	oldnvdevs = sav->sav_count;
1426	sav->sav_vdevs = NULL;
1427	sav->sav_count = 0;
1428
1429	/*
1430	 * Process new nvlist of vdevs.
1431	 */
1432	for (i = 0; i < nl2cache; i++) {
1433		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1434		    &guid) == 0);
1435
1436		newvdevs[i] = NULL;
1437		for (j = 0; j < oldnvdevs; j++) {
1438			vd = oldvdevs[j];
1439			if (vd != NULL && guid == vd->vdev_guid) {
1440				/*
1441				 * Retain previous vdev for add/remove ops.
1442				 */
1443				newvdevs[i] = vd;
1444				oldvdevs[j] = NULL;
1445				break;
1446			}
1447		}
1448
1449		if (newvdevs[i] == NULL) {
1450			/*
1451			 * Create new vdev
1452			 */
1453			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1454			    VDEV_ALLOC_L2CACHE) == 0);
1455			ASSERT(vd != NULL);
1456			newvdevs[i] = vd;
1457
1458			/*
1459			 * Commit this vdev as an l2cache device,
1460			 * even if it fails to open.
1461			 */
1462			spa_l2cache_add(vd);
1463
1464			vd->vdev_top = vd;
1465			vd->vdev_aux = sav;
1466
1467			spa_l2cache_activate(vd);
1468
1469			if (vdev_open(vd) != 0)
1470				continue;
1471
1472			(void) vdev_validate_aux(vd);
1473
1474			if (!vdev_is_dead(vd))
1475				l2arc_add_vdev(spa, vd);
1476		}
1477	}
1478
1479	/*
1480	 * Purge vdevs that were dropped
1481	 */
1482	for (i = 0; i < oldnvdevs; i++) {
1483		uint64_t pool;
1484
1485		vd = oldvdevs[i];
1486		if (vd != NULL) {
1487			ASSERT(vd->vdev_isl2cache);
1488
1489			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1490			    pool != 0ULL && l2arc_vdev_present(vd))
1491				l2arc_remove_vdev(vd);
1492			vdev_clear_stats(vd);
1493			vdev_free(vd);
1494		}
1495	}
1496
1497	if (oldvdevs)
1498		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1499
1500	if (sav->sav_config == NULL)
1501		goto out;
1502
1503	sav->sav_vdevs = newvdevs;
1504	sav->sav_count = (int)nl2cache;
1505
1506	/*
1507	 * Recompute the stashed list of l2cache devices, with status
1508	 * information this time.
1509	 */
1510	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1511	    DATA_TYPE_NVLIST_ARRAY) == 0);
1512
1513	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1514	for (i = 0; i < sav->sav_count; i++)
1515		l2cache[i] = vdev_config_generate(spa,
1516		    sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1517	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1518	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1519out:
1520	for (i = 0; i < sav->sav_count; i++)
1521		nvlist_free(l2cache[i]);
1522	if (sav->sav_count)
1523		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1524}
1525
1526static int
1527load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1528{
1529	dmu_buf_t *db;
1530	char *packed = NULL;
1531	size_t nvsize = 0;
1532	int error;
1533	*value = NULL;
1534
1535	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1536	nvsize = *(uint64_t *)db->db_data;
1537	dmu_buf_rele(db, FTAG);
1538
1539	packed = kmem_alloc(nvsize, KM_SLEEP);
1540	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1541	    DMU_READ_PREFETCH);
1542	if (error == 0)
1543		error = nvlist_unpack(packed, nvsize, value, 0);
1544	kmem_free(packed, nvsize);
1545
1546	return (error);
1547}
1548
1549/*
1550 * Checks to see if the given vdev could not be opened, in which case we post a
1551 * sysevent to notify the autoreplace code that the device has been removed.
1552 */
1553static void
1554spa_check_removed(vdev_t *vd)
1555{
1556	for (int c = 0; c < vd->vdev_children; c++)
1557		spa_check_removed(vd->vdev_child[c]);
1558
1559	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1560		zfs_post_autoreplace(vd->vdev_spa, vd);
1561		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1562	}
1563}
1564
1565/*
1566 * Validate the current config against the MOS config
1567 */
1568static boolean_t
1569spa_config_valid(spa_t *spa, nvlist_t *config)
1570{
1571	vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1572	nvlist_t *nv;
1573
1574	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1575
1576	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1577	VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1578
1579	ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1580
1581	/*
1582	 * If we're doing a normal import, then build up any additional
1583	 * diagnostic information about missing devices in this config.
1584	 * We'll pass this up to the user for further processing.
1585	 */
1586	if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1587		nvlist_t **child, *nv;
1588		uint64_t idx = 0;
1589
1590		child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1591		    KM_SLEEP);
1592		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1593
1594		for (int c = 0; c < rvd->vdev_children; c++) {
1595			vdev_t *tvd = rvd->vdev_child[c];
1596			vdev_t *mtvd  = mrvd->vdev_child[c];
1597
1598			if (tvd->vdev_ops == &vdev_missing_ops &&
1599			    mtvd->vdev_ops != &vdev_missing_ops &&
1600			    mtvd->vdev_islog)
1601				child[idx++] = vdev_config_generate(spa, mtvd,
1602				    B_FALSE, 0);
1603		}
1604
1605		if (idx) {
1606			VERIFY(nvlist_add_nvlist_array(nv,
1607			    ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1608			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1609			    ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1610
1611			for (int i = 0; i < idx; i++)
1612				nvlist_free(child[i]);
1613		}
1614		nvlist_free(nv);
1615		kmem_free(child, rvd->vdev_children * sizeof (char **));
1616	}
1617
1618	/*
1619	 * Compare the root vdev tree with the information we have
1620	 * from the MOS config (mrvd). Check each top-level vdev
1621	 * with the corresponding MOS config top-level (mtvd).
1622	 */
1623	for (int c = 0; c < rvd->vdev_children; c++) {
1624		vdev_t *tvd = rvd->vdev_child[c];
1625		vdev_t *mtvd  = mrvd->vdev_child[c];
1626
1627		/*
1628		 * Resolve any "missing" vdevs in the current configuration.
1629		 * If we find that the MOS config has more accurate information
1630		 * about the top-level vdev then use that vdev instead.
1631		 */
1632		if (tvd->vdev_ops == &vdev_missing_ops &&
1633		    mtvd->vdev_ops != &vdev_missing_ops) {
1634
1635			if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1636				continue;
1637
1638			/*
1639			 * Device specific actions.
1640			 */
1641			if (mtvd->vdev_islog) {
1642				spa_set_log_state(spa, SPA_LOG_CLEAR);
1643			} else {
1644				/*
1645				 * XXX - once we have 'readonly' pool
1646				 * support we should be able to handle
1647				 * missing data devices by transitioning
1648				 * the pool to readonly.
1649				 */
1650				continue;
1651			}
1652
1653			/*
1654			 * Swap the missing vdev with the data we were
1655			 * able to obtain from the MOS config.
1656			 */
1657			vdev_remove_child(rvd, tvd);
1658			vdev_remove_child(mrvd, mtvd);
1659
1660			vdev_add_child(rvd, mtvd);
1661			vdev_add_child(mrvd, tvd);
1662
1663			spa_config_exit(spa, SCL_ALL, FTAG);
1664			vdev_load(mtvd);
1665			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1666
1667			vdev_reopen(rvd);
1668		} else if (mtvd->vdev_islog) {
1669			/*
1670			 * Load the slog device's state from the MOS config
1671			 * since it's possible that the label does not
1672			 * contain the most up-to-date information.
1673			 */
1674			vdev_load_log_state(tvd, mtvd);
1675			vdev_reopen(tvd);
1676		}
1677	}
1678	vdev_free(mrvd);
1679	spa_config_exit(spa, SCL_ALL, FTAG);
1680
1681	/*
1682	 * Ensure we were able to validate the config.
1683	 */
1684	return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1685}
1686
1687/*
1688 * Check for missing log devices
1689 */
1690static boolean_t
1691spa_check_logs(spa_t *spa)
1692{
1693	boolean_t rv = B_FALSE;
1694
1695	switch (spa->spa_log_state) {
1696	case SPA_LOG_MISSING:
1697		/* need to recheck in case slog has been restored */
1698	case SPA_LOG_UNKNOWN:
1699		rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1700		    NULL, DS_FIND_CHILDREN) != 0);
1701		if (rv)
1702			spa_set_log_state(spa, SPA_LOG_MISSING);
1703		break;
1704	}
1705	return (rv);
1706}
1707
1708static boolean_t
1709spa_passivate_log(spa_t *spa)
1710{
1711	vdev_t *rvd = spa->spa_root_vdev;
1712	boolean_t slog_found = B_FALSE;
1713
1714	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1715
1716	if (!spa_has_slogs(spa))
1717		return (B_FALSE);
1718
1719	for (int c = 0; c < rvd->vdev_children; c++) {
1720		vdev_t *tvd = rvd->vdev_child[c];
1721		metaslab_group_t *mg = tvd->vdev_mg;
1722
1723		if (tvd->vdev_islog) {
1724			metaslab_group_passivate(mg);
1725			slog_found = B_TRUE;
1726		}
1727	}
1728
1729	return (slog_found);
1730}
1731
1732static void
1733spa_activate_log(spa_t *spa)
1734{
1735	vdev_t *rvd = spa->spa_root_vdev;
1736
1737	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1738
1739	for (int c = 0; c < rvd->vdev_children; c++) {
1740		vdev_t *tvd = rvd->vdev_child[c];
1741		metaslab_group_t *mg = tvd->vdev_mg;
1742
1743		if (tvd->vdev_islog)
1744			metaslab_group_activate(mg);
1745	}
1746}
1747
1748int
1749spa_offline_log(spa_t *spa)
1750{
1751	int error;
1752
1753	error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1754	    NULL, DS_FIND_CHILDREN);
1755	if (error == 0) {
1756		/*
1757		 * We successfully offlined the log device, sync out the
1758		 * current txg so that the "stubby" block can be removed
1759		 * by zil_sync().
1760		 */
1761		txg_wait_synced(spa->spa_dsl_pool, 0);
1762	}
1763	return (error);
1764}
1765
1766static void
1767spa_aux_check_removed(spa_aux_vdev_t *sav)
1768{
1769	for (int i = 0; i < sav->sav_count; i++)
1770		spa_check_removed(sav->sav_vdevs[i]);
1771}
1772
1773void
1774spa_claim_notify(zio_t *zio)
1775{
1776	spa_t *spa = zio->io_spa;
1777
1778	if (zio->io_error)
1779		return;
1780
1781	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1782	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1783		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1784	mutex_exit(&spa->spa_props_lock);
1785}
1786
1787typedef struct spa_load_error {
1788	uint64_t	sle_meta_count;
1789	uint64_t	sle_data_count;
1790} spa_load_error_t;
1791
1792static void
1793spa_load_verify_done(zio_t *zio)
1794{
1795	blkptr_t *bp = zio->io_bp;
1796	spa_load_error_t *sle = zio->io_private;
1797	dmu_object_type_t type = BP_GET_TYPE(bp);
1798	int error = zio->io_error;
1799
1800	if (error) {
1801		if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1802		    type != DMU_OT_INTENT_LOG)
1803			atomic_add_64(&sle->sle_meta_count, 1);
1804		else
1805			atomic_add_64(&sle->sle_data_count, 1);
1806	}
1807	zio_data_buf_free(zio->io_data, zio->io_size);
1808}
1809
1810/*ARGSUSED*/
1811static int
1812spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1813    const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1814{
1815	if (bp != NULL) {
1816		zio_t *rio = arg;
1817		size_t size = BP_GET_PSIZE(bp);
1818		void *data = zio_data_buf_alloc(size);
1819
1820		zio_nowait(zio_read(rio, spa, bp, data, size,
1821		    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1822		    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1823		    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1824	}
1825	return (0);
1826}
1827
1828static int
1829spa_load_verify(spa_t *spa)
1830{
1831	zio_t *rio;
1832	spa_load_error_t sle = { 0 };
1833	zpool_rewind_policy_t policy;
1834	boolean_t verify_ok = B_FALSE;
1835	int error;
1836
1837	zpool_get_rewind_policy(spa->spa_config, &policy);
1838
1839	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1840		return (0);
1841
1842	rio = zio_root(spa, NULL, &sle,
1843	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1844
1845	error = traverse_pool(spa, spa->spa_verify_min_txg,
1846	    TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1847
1848	(void) zio_wait(rio);
1849
1850	spa->spa_load_meta_errors = sle.sle_meta_count;
1851	spa->spa_load_data_errors = sle.sle_data_count;
1852
1853	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1854	    sle.sle_data_count <= policy.zrp_maxdata) {
1855		int64_t loss = 0;
1856
1857		verify_ok = B_TRUE;
1858		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1859		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1860
1861		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1862		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1863		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1864		VERIFY(nvlist_add_int64(spa->spa_load_info,
1865		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1866		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1867		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1868	} else {
1869		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1870	}
1871
1872	if (error) {
1873		if (error != ENXIO && error != EIO)
1874			error = EIO;
1875		return (error);
1876	}
1877
1878	return (verify_ok ? 0 : EIO);
1879}
1880
1881/*
1882 * Find a value in the pool props object.
1883 */
1884static void
1885spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1886{
1887	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1888	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1889}
1890
1891/*
1892 * Find a value in the pool directory object.
1893 */
1894static int
1895spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1896{
1897	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1898	    name, sizeof (uint64_t), 1, val));
1899}
1900
1901static int
1902spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1903{
1904	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1905	return (err);
1906}
1907
1908/*
1909 * Fix up config after a partly-completed split.  This is done with the
1910 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1911 * pool have that entry in their config, but only the splitting one contains
1912 * a list of all the guids of the vdevs that are being split off.
1913 *
1914 * This function determines what to do with that list: either rejoin
1915 * all the disks to the pool, or complete the splitting process.  To attempt
1916 * the rejoin, each disk that is offlined is marked online again, and
1917 * we do a reopen() call.  If the vdev label for every disk that was
1918 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1919 * then we call vdev_split() on each disk, and complete the split.
1920 *
1921 * Otherwise we leave the config alone, with all the vdevs in place in
1922 * the original pool.
1923 */
1924static void
1925spa_try_repair(spa_t *spa, nvlist_t *config)
1926{
1927	uint_t extracted;
1928	uint64_t *glist;
1929	uint_t i, gcount;
1930	nvlist_t *nvl;
1931	vdev_t **vd;
1932	boolean_t attempt_reopen;
1933
1934	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1935		return;
1936
1937	/* check that the config is complete */
1938	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1939	    &glist, &gcount) != 0)
1940		return;
1941
1942	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1943
1944	/* attempt to online all the vdevs & validate */
1945	attempt_reopen = B_TRUE;
1946	for (i = 0; i < gcount; i++) {
1947		if (glist[i] == 0)	/* vdev is hole */
1948			continue;
1949
1950		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1951		if (vd[i] == NULL) {
1952			/*
1953			 * Don't bother attempting to reopen the disks;
1954			 * just do the split.
1955			 */
1956			attempt_reopen = B_FALSE;
1957		} else {
1958			/* attempt to re-online it */
1959			vd[i]->vdev_offline = B_FALSE;
1960		}
1961	}
1962
1963	if (attempt_reopen) {
1964		vdev_reopen(spa->spa_root_vdev);
1965
1966		/* check each device to see what state it's in */
1967		for (extracted = 0, i = 0; i < gcount; i++) {
1968			if (vd[i] != NULL &&
1969			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1970				break;
1971			++extracted;
1972		}
1973	}
1974
1975	/*
1976	 * If every disk has been moved to the new pool, or if we never
1977	 * even attempted to look at them, then we split them off for
1978	 * good.
1979	 */
1980	if (!attempt_reopen || gcount == extracted) {
1981		for (i = 0; i < gcount; i++)
1982			if (vd[i] != NULL)
1983				vdev_split(vd[i]);
1984		vdev_reopen(spa->spa_root_vdev);
1985	}
1986
1987	kmem_free(vd, gcount * sizeof (vdev_t *));
1988}
1989
1990static int
1991spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1992    boolean_t mosconfig)
1993{
1994	nvlist_t *config = spa->spa_config;
1995	char *ereport = FM_EREPORT_ZFS_POOL;
1996	char *comment;
1997	int error;
1998	uint64_t pool_guid;
1999	nvlist_t *nvl;
2000
2001	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2002		return (EINVAL);
2003
2004	ASSERT(spa->spa_comment == NULL);
2005	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2006		spa->spa_comment = spa_strdup(comment);
2007
2008	/*
2009	 * Versioning wasn't explicitly added to the label until later, so if
2010	 * it's not present treat it as the initial version.
2011	 */
2012	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2013	    &spa->spa_ubsync.ub_version) != 0)
2014		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2015
2016	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2017	    &spa->spa_config_txg);
2018
2019	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2020	    spa_guid_exists(pool_guid, 0)) {
2021		error = EEXIST;
2022	} else {
2023		spa->spa_config_guid = pool_guid;
2024
2025		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2026		    &nvl) == 0) {
2027			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2028			    KM_SLEEP) == 0);
2029		}
2030
2031		nvlist_free(spa->spa_load_info);
2032		spa->spa_load_info = fnvlist_alloc();
2033
2034		gethrestime(&spa->spa_loaded_ts);
2035		error = spa_load_impl(spa, pool_guid, config, state, type,
2036		    mosconfig, &ereport);
2037	}
2038
2039	spa->spa_minref = refcount_count(&spa->spa_refcount);
2040	if (error) {
2041		if (error != EEXIST) {
2042			spa->spa_loaded_ts.tv_sec = 0;
2043			spa->spa_loaded_ts.tv_nsec = 0;
2044		}
2045		if (error != EBADF) {
2046			zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2047		}
2048	}
2049	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2050	spa->spa_ena = 0;
2051
2052	return (error);
2053}
2054
2055/*
2056 * Load an existing storage pool, using the pool's builtin spa_config as a
2057 * source of configuration information.
2058 */
2059static int
2060spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2061    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2062    char **ereport)
2063{
2064	int error = 0;
2065	nvlist_t *nvroot = NULL;
2066	nvlist_t *label;
2067	vdev_t *rvd;
2068	uberblock_t *ub = &spa->spa_uberblock;
2069	uint64_t children, config_cache_txg = spa->spa_config_txg;
2070	int orig_mode = spa->spa_mode;
2071	int parse;
2072	uint64_t obj;
2073	boolean_t missing_feat_write = B_FALSE;
2074
2075	/*
2076	 * If this is an untrusted config, access the pool in read-only mode.
2077	 * This prevents things like resilvering recently removed devices.
2078	 */
2079	if (!mosconfig)
2080		spa->spa_mode = FREAD;
2081
2082	ASSERT(MUTEX_HELD(&spa_namespace_lock));
2083
2084	spa->spa_load_state = state;
2085
2086	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2087		return (EINVAL);
2088
2089	parse = (type == SPA_IMPORT_EXISTING ?
2090	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2091
2092	/*
2093	 * Create "The Godfather" zio to hold all async IOs
2094	 */
2095	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2096	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2097
2098	/*
2099	 * Parse the configuration into a vdev tree.  We explicitly set the
2100	 * value that will be returned by spa_version() since parsing the
2101	 * configuration requires knowing the version number.
2102	 */
2103	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2104	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2105	spa_config_exit(spa, SCL_ALL, FTAG);
2106
2107	if (error != 0)
2108		return (error);
2109
2110	ASSERT(spa->spa_root_vdev == rvd);
2111
2112	if (type != SPA_IMPORT_ASSEMBLE) {
2113		ASSERT(spa_guid(spa) == pool_guid);
2114	}
2115
2116	/*
2117	 * Try to open all vdevs, loading each label in the process.
2118	 */
2119	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2120	error = vdev_open(rvd);
2121	spa_config_exit(spa, SCL_ALL, FTAG);
2122	if (error != 0)
2123		return (error);
2124
2125	/*
2126	 * We need to validate the vdev labels against the configuration that
2127	 * we have in hand, which is dependent on the setting of mosconfig. If
2128	 * mosconfig is true then we're validating the vdev labels based on
2129	 * that config.  Otherwise, we're validating against the cached config
2130	 * (zpool.cache) that was read when we loaded the zfs module, and then
2131	 * later we will recursively call spa_load() and validate against
2132	 * the vdev config.
2133	 *
2134	 * If we're assembling a new pool that's been split off from an
2135	 * existing pool, the labels haven't yet been updated so we skip
2136	 * validation for now.
2137	 */
2138	if (type != SPA_IMPORT_ASSEMBLE) {
2139		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2140		error = vdev_validate(rvd, mosconfig);
2141		spa_config_exit(spa, SCL_ALL, FTAG);
2142
2143		if (error != 0)
2144			return (error);
2145
2146		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2147			return (ENXIO);
2148	}
2149
2150	/*
2151	 * Find the best uberblock.
2152	 */
2153	vdev_uberblock_load(rvd, ub, &label);
2154
2155	/*
2156	 * If we weren't able to find a single valid uberblock, return failure.
2157	 */
2158	if (ub->ub_txg == 0) {
2159		nvlist_free(label);
2160		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2161	}
2162
2163	/*
2164	 * If the pool has an unsupported version we can't open it.
2165	 */
2166	if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2167		nvlist_free(label);
2168		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2169	}
2170
2171	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2172		nvlist_t *features;
2173
2174		/*
2175		 * If we weren't able to find what's necessary for reading the
2176		 * MOS in the label, return failure.
2177		 */
2178		if (label == NULL || nvlist_lookup_nvlist(label,
2179		    ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2180			nvlist_free(label);
2181			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2182			    ENXIO));
2183		}
2184
2185		/*
2186		 * Update our in-core representation with the definitive values
2187		 * from the label.
2188		 */
2189		nvlist_free(spa->spa_label_features);
2190		VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2191	}
2192
2193	nvlist_free(label);
2194
2195	/*
2196	 * Look through entries in the label nvlist's features_for_read. If
2197	 * there is a feature listed there which we don't understand then we
2198	 * cannot open a pool.
2199	 */
2200	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2201		nvlist_t *unsup_feat;
2202
2203		VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2204		    0);
2205
2206		for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2207		    NULL); nvp != NULL;
2208		    nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2209			if (!zfeature_is_supported(nvpair_name(nvp))) {
2210				VERIFY(nvlist_add_string(unsup_feat,
2211				    nvpair_name(nvp), "") == 0);
2212			}
2213		}
2214
2215		if (!nvlist_empty(unsup_feat)) {
2216			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2217			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2218			nvlist_free(unsup_feat);
2219			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2220			    ENOTSUP));
2221		}
2222
2223		nvlist_free(unsup_feat);
2224	}
2225
2226	/*
2227	 * If the vdev guid sum doesn't match the uberblock, we have an
2228	 * incomplete configuration.  We first check to see if the pool
2229	 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2230	 * If it is, defer the vdev_guid_sum check till later so we
2231	 * can handle missing vdevs.
2232	 */
2233	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2234	    &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2235	    rvd->vdev_guid_sum != ub->ub_guid_sum)
2236		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2237
2238	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2239		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2240		spa_try_repair(spa, config);
2241		spa_config_exit(spa, SCL_ALL, FTAG);
2242		nvlist_free(spa->spa_config_splitting);
2243		spa->spa_config_splitting = NULL;
2244	}
2245
2246	/*
2247	 * Initialize internal SPA structures.
2248	 */
2249	spa->spa_state = POOL_STATE_ACTIVE;
2250	spa->spa_ubsync = spa->spa_uberblock;
2251	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2252	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2253	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2254	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2255	spa->spa_claim_max_txg = spa->spa_first_txg;
2256	spa->spa_prev_software_version = ub->ub_software_version;
2257
2258	error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2259	if (error)
2260		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2261	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2262
2263	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2264		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2265
2266	if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2267		boolean_t missing_feat_read = B_FALSE;
2268		nvlist_t *unsup_feat, *enabled_feat;
2269
2270		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2271		    &spa->spa_feat_for_read_obj) != 0) {
2272			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2273		}
2274
2275		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2276		    &spa->spa_feat_for_write_obj) != 0) {
2277			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2278		}
2279
2280		if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2281		    &spa->spa_feat_desc_obj) != 0) {
2282			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2283		}
2284
2285		enabled_feat = fnvlist_alloc();
2286		unsup_feat = fnvlist_alloc();
2287
2288		if (!feature_is_supported(spa->spa_meta_objset,
2289		    spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2290		    unsup_feat, enabled_feat))
2291			missing_feat_read = B_TRUE;
2292
2293		if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2294			if (!feature_is_supported(spa->spa_meta_objset,
2295			    spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2296			    unsup_feat, enabled_feat)) {
2297				missing_feat_write = B_TRUE;
2298			}
2299		}
2300
2301		fnvlist_add_nvlist(spa->spa_load_info,
2302		    ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2303
2304		if (!nvlist_empty(unsup_feat)) {
2305			fnvlist_add_nvlist(spa->spa_load_info,
2306			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2307		}
2308
2309		fnvlist_free(enabled_feat);
2310		fnvlist_free(unsup_feat);
2311
2312		if (!missing_feat_read) {
2313			fnvlist_add_boolean(spa->spa_load_info,
2314			    ZPOOL_CONFIG_CAN_RDONLY);
2315		}
2316
2317		/*
2318		 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2319		 * twofold: to determine whether the pool is available for
2320		 * import in read-write mode and (if it is not) whether the
2321		 * pool is available for import in read-only mode. If the pool
2322		 * is available for import in read-write mode, it is displayed
2323		 * as available in userland; if it is not available for import
2324		 * in read-only mode, it is displayed as unavailable in
2325		 * userland. If the pool is available for import in read-only
2326		 * mode but not read-write mode, it is displayed as unavailable
2327		 * in userland with a special note that the pool is actually
2328		 * available for open in read-only mode.
2329		 *
2330		 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2331		 * missing a feature for write, we must first determine whether
2332		 * the pool can be opened read-only before returning to
2333		 * userland in order to know whether to display the
2334		 * abovementioned note.
2335		 */
2336		if (missing_feat_read || (missing_feat_write &&
2337		    spa_writeable(spa))) {
2338			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2339			    ENOTSUP));
2340		}
2341	}
2342
2343	spa->spa_is_initializing = B_TRUE;
2344	error = dsl_pool_open(spa->spa_dsl_pool);
2345	spa->spa_is_initializing = B_FALSE;
2346	if (error != 0)
2347		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2348
2349	if (!mosconfig) {
2350		uint64_t hostid;
2351		nvlist_t *policy = NULL, *nvconfig;
2352
2353		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2354			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2355
2356		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2357		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2358			char *hostname;
2359			unsigned long myhostid = 0;
2360
2361			VERIFY(nvlist_lookup_string(nvconfig,
2362			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2363
2364#ifdef	_KERNEL
2365			myhostid = zone_get_hostid(NULL);
2366#else	/* _KERNEL */
2367			/*
2368			 * We're emulating the system's hostid in userland, so
2369			 * we can't use zone_get_hostid().
2370			 */
2371			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2372#endif	/* _KERNEL */
2373			if (hostid != 0 && myhostid != 0 &&
2374			    hostid != myhostid) {
2375				nvlist_free(nvconfig);
2376				cmn_err(CE_WARN, "pool '%s' could not be "
2377				    "loaded as it was last accessed by "
2378				    "another system (host: %s hostid: 0x%lx). "
2379				    "See: http://illumos.org/msg/ZFS-8000-EY",
2380				    spa_name(spa), hostname,
2381				    (unsigned long)hostid);
2382				return (EBADF);
2383			}
2384		}
2385		if (nvlist_lookup_nvlist(spa->spa_config,
2386		    ZPOOL_REWIND_POLICY, &policy) == 0)
2387			VERIFY(nvlist_add_nvlist(nvconfig,
2388			    ZPOOL_REWIND_POLICY, policy) == 0);
2389
2390		spa_config_set(spa, nvconfig);
2391		spa_unload(spa);
2392		spa_deactivate(spa);
2393		spa_activate(spa, orig_mode);
2394
2395		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2396	}
2397
2398	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2399		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2400	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2401	if (error != 0)
2402		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2403
2404	/*
2405	 * Load the bit that tells us to use the new accounting function
2406	 * (raid-z deflation).  If we have an older pool, this will not
2407	 * be present.
2408	 */
2409	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2410	if (error != 0 && error != ENOENT)
2411		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2412
2413	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2414	    &spa->spa_creation_version);
2415	if (error != 0 && error != ENOENT)
2416		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2417
2418	/*
2419	 * Load the persistent error log.  If we have an older pool, this will
2420	 * not be present.
2421	 */
2422	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2423	if (error != 0 && error != ENOENT)
2424		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2425
2426	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2427	    &spa->spa_errlog_scrub);
2428	if (error != 0 && error != ENOENT)
2429		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2430
2431	/*
2432	 * Load the history object.  If we have an older pool, this
2433	 * will not be present.
2434	 */
2435	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2436	if (error != 0 && error != ENOENT)
2437		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2438
2439	/*
2440	 * If we're assembling the pool from the split-off vdevs of
2441	 * an existing pool, we don't want to attach the spares & cache
2442	 * devices.
2443	 */
2444
2445	/*
2446	 * Load any hot spares for this pool.
2447	 */
2448	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2449	if (error != 0 && error != ENOENT)
2450		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2451	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2452		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2453		if (load_nvlist(spa, spa->spa_spares.sav_object,
2454		    &spa->spa_spares.sav_config) != 0)
2455			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2456
2457		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2458		spa_load_spares(spa);
2459		spa_config_exit(spa, SCL_ALL, FTAG);
2460	} else if (error == 0) {
2461		spa->spa_spares.sav_sync = B_TRUE;
2462	}
2463
2464	/*
2465	 * Load any level 2 ARC devices for this pool.
2466	 */
2467	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2468	    &spa->spa_l2cache.sav_object);
2469	if (error != 0 && error != ENOENT)
2470		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2471	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2472		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2473		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2474		    &spa->spa_l2cache.sav_config) != 0)
2475			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2476
2477		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2478		spa_load_l2cache(spa);
2479		spa_config_exit(spa, SCL_ALL, FTAG);
2480	} else if (error == 0) {
2481		spa->spa_l2cache.sav_sync = B_TRUE;
2482	}
2483
2484	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2485
2486	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2487	if (error && error != ENOENT)
2488		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2489
2490	if (error == 0) {
2491		uint64_t autoreplace;
2492
2493		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2494		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2495		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2496		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2497		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2498		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2499		    &spa->spa_dedup_ditto);
2500
2501		spa->spa_autoreplace = (autoreplace != 0);
2502	}
2503
2504	/*
2505	 * If the 'autoreplace' property is set, then post a resource notifying
2506	 * the ZFS DE that it should not issue any faults for unopenable
2507	 * devices.  We also iterate over the vdevs, and post a sysevent for any
2508	 * unopenable vdevs so that the normal autoreplace handler can take
2509	 * over.
2510	 */
2511	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2512		spa_check_removed(spa->spa_root_vdev);
2513		/*
2514		 * For the import case, this is done in spa_import(), because
2515		 * at this point we're using the spare definitions from
2516		 * the MOS config, not necessarily from the userland config.
2517		 */
2518		if (state != SPA_LOAD_IMPORT) {
2519			spa_aux_check_removed(&spa->spa_spares);
2520			spa_aux_check_removed(&spa->spa_l2cache);
2521		}
2522	}
2523
2524	/*
2525	 * Load the vdev state for all toplevel vdevs.
2526	 */
2527	vdev_load(rvd);
2528
2529	/*
2530	 * Propagate the leaf DTLs we just loaded all the way up the tree.
2531	 */
2532	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2533	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2534	spa_config_exit(spa, SCL_ALL, FTAG);
2535
2536	/*
2537	 * Load the DDTs (dedup tables).
2538	 */
2539	error = ddt_load(spa);
2540	if (error != 0)
2541		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2542
2543	spa_update_dspace(spa);
2544
2545	/*
2546	 * Validate the config, using the MOS config to fill in any
2547	 * information which might be missing.  If we fail to validate
2548	 * the config then declare the pool unfit for use. If we're
2549	 * assembling a pool from a split, the log is not transferred
2550	 * over.
2551	 */
2552	if (type != SPA_IMPORT_ASSEMBLE) {
2553		nvlist_t *nvconfig;
2554
2555		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2556			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2557
2558		if (!spa_config_valid(spa, nvconfig)) {
2559			nvlist_free(nvconfig);
2560			return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2561			    ENXIO));
2562		}
2563		nvlist_free(nvconfig);
2564
2565		/*
2566		 * Now that we've validated the config, check the state of the
2567		 * root vdev.  If it can't be opened, it indicates one or
2568		 * more toplevel vdevs are faulted.
2569		 */
2570		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2571			return (ENXIO);
2572
2573		if (spa_check_logs(spa)) {
2574			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2575			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2576		}
2577	}
2578
2579	if (missing_feat_write) {
2580		ASSERT(state == SPA_LOAD_TRYIMPORT);
2581
2582		/*
2583		 * At this point, we know that we can open the pool in
2584		 * read-only mode but not read-write mode. We now have enough
2585		 * information and can return to userland.
2586		 */
2587		return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2588	}
2589
2590	/*
2591	 * We've successfully opened the pool, verify that we're ready
2592	 * to start pushing transactions.
2593	 */
2594	if (state != SPA_LOAD_TRYIMPORT) {
2595		if (error = spa_load_verify(spa))
2596			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2597			    error));
2598	}
2599
2600	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2601	    spa->spa_load_max_txg == UINT64_MAX)) {
2602		dmu_tx_t *tx;
2603		int need_update = B_FALSE;
2604
2605		ASSERT(state != SPA_LOAD_TRYIMPORT);
2606
2607		/*
2608		 * Claim log blocks that haven't been committed yet.
2609		 * This must all happen in a single txg.
2610		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2611		 * invoked from zil_claim_log_block()'s i/o done callback.
2612		 * Price of rollback is that we abandon the log.
2613		 */
2614		spa->spa_claiming = B_TRUE;
2615
2616		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2617		    spa_first_txg(spa));
2618		(void) dmu_objset_find(spa_name(spa),
2619		    zil_claim, tx, DS_FIND_CHILDREN);
2620		dmu_tx_commit(tx);
2621
2622		spa->spa_claiming = B_FALSE;
2623
2624		spa_set_log_state(spa, SPA_LOG_GOOD);
2625		spa->spa_sync_on = B_TRUE;
2626		txg_sync_start(spa->spa_dsl_pool);
2627
2628		/*
2629		 * Wait for all claims to sync.  We sync up to the highest
2630		 * claimed log block birth time so that claimed log blocks
2631		 * don't appear to be from the future.  spa_claim_max_txg
2632		 * will have been set for us by either zil_check_log_chain()
2633		 * (invoked from spa_check_logs()) or zil_claim() above.
2634		 */
2635		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2636
2637		/*
2638		 * If the config cache is stale, or we have uninitialized
2639		 * metaslabs (see spa_vdev_add()), then update the config.
2640		 *
2641		 * If this is a verbatim import, trust the current
2642		 * in-core spa_config and update the disk labels.
2643		 */
2644		if (config_cache_txg != spa->spa_config_txg ||
2645		    state == SPA_LOAD_IMPORT ||
2646		    state == SPA_LOAD_RECOVER ||
2647		    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2648			need_update = B_TRUE;
2649
2650		for (int c = 0; c < rvd->vdev_children; c++)
2651			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2652				need_update = B_TRUE;
2653
2654		/*
2655		 * Update the config cache asychronously in case we're the
2656		 * root pool, in which case the config cache isn't writable yet.
2657		 */
2658		if (need_update)
2659			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2660
2661		/*
2662		 * Check all DTLs to see if anything needs resilvering.
2663		 */
2664		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2665		    vdev_resilver_needed(rvd, NULL, NULL))
2666			spa_async_request(spa, SPA_ASYNC_RESILVER);
2667
2668		/*
2669		 * Log the fact that we booted up (so that we can detect if
2670		 * we rebooted in the middle of an operation).
2671		 */
2672		spa_history_log_version(spa, "open");
2673
2674		/*
2675		 * Delete any inconsistent datasets.
2676		 */
2677		(void) dmu_objset_find(spa_name(spa),
2678		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2679
2680		/*
2681		 * Clean up any stale temporary dataset userrefs.
2682		 */
2683		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2684	}
2685
2686	return (0);
2687}
2688
2689static int
2690spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2691{
2692	int mode = spa->spa_mode;
2693
2694	spa_unload(spa);
2695	spa_deactivate(spa);
2696
2697	spa->spa_load_max_txg--;
2698
2699	spa_activate(spa, mode);
2700	spa_async_suspend(spa);
2701
2702	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2703}
2704
2705/*
2706 * If spa_load() fails this function will try loading prior txg's. If
2707 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2708 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2709 * function will not rewind the pool and will return the same error as
2710 * spa_load().
2711 */
2712static int
2713spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2714    uint64_t max_request, int rewind_flags)
2715{
2716	nvlist_t *loadinfo = NULL;
2717	nvlist_t *config = NULL;
2718	int load_error, rewind_error;
2719	uint64_t safe_rewind_txg;
2720	uint64_t min_txg;
2721
2722	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2723		spa->spa_load_max_txg = spa->spa_load_txg;
2724		spa_set_log_state(spa, SPA_LOG_CLEAR);
2725	} else {
2726		spa->spa_load_max_txg = max_request;
2727	}
2728
2729	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2730	    mosconfig);
2731	if (load_error == 0)
2732		return (0);
2733
2734	if (spa->spa_root_vdev != NULL)
2735		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2736
2737	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2738	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2739
2740	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2741		nvlist_free(config);
2742		return (load_error);
2743	}
2744
2745	if (state == SPA_LOAD_RECOVER) {
2746		/* Price of rolling back is discarding txgs, including log */
2747		spa_set_log_state(spa, SPA_LOG_CLEAR);
2748	} else {
2749		/*
2750		 * If we aren't rolling back save the load info from our first
2751		 * import attempt so that we can restore it after attempting
2752		 * to rewind.
2753		 */
2754		loadinfo = spa->spa_load_info;
2755		spa->spa_load_info = fnvlist_alloc();
2756	}
2757
2758	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2759	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2760	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2761	    TXG_INITIAL : safe_rewind_txg;
2762
2763	/*
2764	 * Continue as long as we're finding errors, we're still within
2765	 * the acceptable rewind range, and we're still finding uberblocks
2766	 */
2767	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2768	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2769		if (spa->spa_load_max_txg < safe_rewind_txg)
2770			spa->spa_extreme_rewind = B_TRUE;
2771		rewind_error = spa_load_retry(spa, state, mosconfig);
2772	}
2773
2774	spa->spa_extreme_rewind = B_FALSE;
2775	spa->spa_load_max_txg = UINT64_MAX;
2776
2777	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2778		spa_config_set(spa, config);
2779
2780	if (state == SPA_LOAD_RECOVER) {
2781		ASSERT3P(loadinfo, ==, NULL);
2782		return (rewind_error);
2783	} else {
2784		/* Store the rewind info as part of the initial load info */
2785		fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2786		    spa->spa_load_info);
2787
2788		/* Restore the initial load info */
2789		fnvlist_free(spa->spa_load_info);
2790		spa->spa_load_info = loadinfo;
2791
2792		return (load_error);
2793	}
2794}
2795
2796/*
2797 * Pool Open/Import
2798 *
2799 * The import case is identical to an open except that the configuration is sent
2800 * down from userland, instead of grabbed from the configuration cache.  For the
2801 * case of an open, the pool configuration will exist in the
2802 * POOL_STATE_UNINITIALIZED state.
2803 *
2804 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2805 * the same time open the pool, without having to keep around the spa_t in some
2806 * ambiguous state.
2807 */
2808static int
2809spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2810    nvlist_t **config)
2811{
2812	spa_t *spa;
2813	spa_load_state_t state = SPA_LOAD_OPEN;
2814	int error;
2815	int locked = B_FALSE;
2816
2817	*spapp = NULL;
2818
2819	/*
2820	 * As disgusting as this is, we need to support recursive calls to this
2821	 * function because dsl_dir_open() is called during spa_load(), and ends
2822	 * up calling spa_open() again.  The real fix is to figure out how to
2823	 * avoid dsl_dir_open() calling this in the first place.
2824	 */
2825	if (mutex_owner(&spa_namespace_lock) != curthread) {
2826		mutex_enter(&spa_namespace_lock);
2827		locked = B_TRUE;
2828	}
2829
2830	if ((spa = spa_lookup(pool)) == NULL) {
2831		if (locked)
2832			mutex_exit(&spa_namespace_lock);
2833		return (ENOENT);
2834	}
2835
2836	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2837		zpool_rewind_policy_t policy;
2838
2839		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2840		    &policy);
2841		if (policy.zrp_request & ZPOOL_DO_REWIND)
2842			state = SPA_LOAD_RECOVER;
2843
2844		spa_activate(spa, spa_mode_global);
2845
2846		if (state != SPA_LOAD_RECOVER)
2847			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2848
2849		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2850		    policy.zrp_request);
2851
2852		if (error == EBADF) {
2853			/*
2854			 * If vdev_validate() returns failure (indicated by
2855			 * EBADF), it indicates that one of the vdevs indicates
2856			 * that the pool has been exported or destroyed.  If
2857			 * this is the case, the config cache is out of sync and
2858			 * we should remove the pool from the namespace.
2859			 */
2860			spa_unload(spa);
2861			spa_deactivate(spa);
2862			spa_config_sync(spa, B_TRUE, B_TRUE);
2863			spa_remove(spa);
2864			if (locked)
2865				mutex_exit(&spa_namespace_lock);
2866			return (ENOENT);
2867		}
2868
2869		if (error) {
2870			/*
2871			 * We can't open the pool, but we still have useful
2872			 * information: the state of each vdev after the
2873			 * attempted vdev_open().  Return this to the user.
2874			 */
2875			if (config != NULL && spa->spa_config) {
2876				VERIFY(nvlist_dup(spa->spa_config, config,
2877				    KM_SLEEP) == 0);
2878				VERIFY(nvlist_add_nvlist(*config,
2879				    ZPOOL_CONFIG_LOAD_INFO,
2880				    spa->spa_load_info) == 0);
2881			}
2882			spa_unload(spa);
2883			spa_deactivate(spa);
2884			spa->spa_last_open_failed = error;
2885			if (locked)
2886				mutex_exit(&spa_namespace_lock);
2887			*spapp = NULL;
2888			return (error);
2889		}
2890	}
2891
2892	spa_open_ref(spa, tag);
2893
2894	if (config != NULL)
2895		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2896
2897	/*
2898	 * If we've recovered the pool, pass back any information we
2899	 * gathered while doing the load.
2900	 */
2901	if (state == SPA_LOAD_RECOVER) {
2902		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2903		    spa->spa_load_info) == 0);
2904	}
2905
2906	if (locked) {
2907		spa->spa_last_open_failed = 0;
2908		spa->spa_last_ubsync_txg = 0;
2909		spa->spa_load_txg = 0;
2910		mutex_exit(&spa_namespace_lock);
2911	}
2912
2913	*spapp = spa;
2914
2915	return (0);
2916}
2917
2918int
2919spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2920    nvlist_t **config)
2921{
2922	return (spa_open_common(name, spapp, tag, policy, config));
2923}
2924
2925int
2926spa_open(const char *name, spa_t **spapp, void *tag)
2927{
2928	return (spa_open_common(name, spapp, tag, NULL, NULL));
2929}
2930
2931/*
2932 * Lookup the given spa_t, incrementing the inject count in the process,
2933 * preventing it from being exported or destroyed.
2934 */
2935spa_t *
2936spa_inject_addref(char *name)
2937{
2938	spa_t *spa;
2939
2940	mutex_enter(&spa_namespace_lock);
2941	if ((spa = spa_lookup(name)) == NULL) {
2942		mutex_exit(&spa_namespace_lock);
2943		return (NULL);
2944	}
2945	spa->spa_inject_ref++;
2946	mutex_exit(&spa_namespace_lock);
2947
2948	return (spa);
2949}
2950
2951void
2952spa_inject_delref(spa_t *spa)
2953{
2954	mutex_enter(&spa_namespace_lock);
2955	spa->spa_inject_ref--;
2956	mutex_exit(&spa_namespace_lock);
2957}
2958
2959/*
2960 * Add spares device information to the nvlist.
2961 */
2962static void
2963spa_add_spares(spa_t *spa, nvlist_t *config)
2964{
2965	nvlist_t **spares;
2966	uint_t i, nspares;
2967	nvlist_t *nvroot;
2968	uint64_t guid;
2969	vdev_stat_t *vs;
2970	uint_t vsc;
2971	uint64_t pool;
2972
2973	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2974
2975	if (spa->spa_spares.sav_count == 0)
2976		return;
2977
2978	VERIFY(nvlist_lookup_nvlist(config,
2979	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2980	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2981	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2982	if (nspares != 0) {
2983		VERIFY(nvlist_add_nvlist_array(nvroot,
2984		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2985		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2986		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2987
2988		/*
2989		 * Go through and find any spares which have since been
2990		 * repurposed as an active spare.  If this is the case, update
2991		 * their status appropriately.
2992		 */
2993		for (i = 0; i < nspares; i++) {
2994			VERIFY(nvlist_lookup_uint64(spares[i],
2995			    ZPOOL_CONFIG_GUID, &guid) == 0);
2996			if (spa_spare_exists(guid, &pool, NULL) &&
2997			    pool != 0ULL) {
2998				VERIFY(nvlist_lookup_uint64_array(
2999				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
3000				    (uint64_t **)&vs, &vsc) == 0);
3001				vs->vs_state = VDEV_STATE_CANT_OPEN;
3002				vs->vs_aux = VDEV_AUX_SPARED;
3003			}
3004		}
3005	}
3006}
3007
3008/*
3009 * Add l2cache device information to the nvlist, including vdev stats.
3010 */
3011static void
3012spa_add_l2cache(spa_t *spa, nvlist_t *config)
3013{
3014	nvlist_t **l2cache;
3015	uint_t i, j, nl2cache;
3016	nvlist_t *nvroot;
3017	uint64_t guid;
3018	vdev_t *vd;
3019	vdev_stat_t *vs;
3020	uint_t vsc;
3021
3022	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3023
3024	if (spa->spa_l2cache.sav_count == 0)
3025		return;
3026
3027	VERIFY(nvlist_lookup_nvlist(config,
3028	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3029	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3030	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3031	if (nl2cache != 0) {
3032		VERIFY(nvlist_add_nvlist_array(nvroot,
3033		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3034		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3035		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3036
3037		/*
3038		 * Update level 2 cache device stats.
3039		 */
3040
3041		for (i = 0; i < nl2cache; i++) {
3042			VERIFY(nvlist_lookup_uint64(l2cache[i],
3043			    ZPOOL_CONFIG_GUID, &guid) == 0);
3044
3045			vd = NULL;
3046			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3047				if (guid ==
3048				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3049					vd = spa->spa_l2cache.sav_vdevs[j];
3050					break;
3051				}
3052			}
3053			ASSERT(vd != NULL);
3054
3055			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3056			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3057			    == 0);
3058			vdev_get_stats(vd, vs);
3059		}
3060	}
3061}
3062
3063static void
3064spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3065{
3066	nvlist_t *features;
3067	zap_cursor_t zc;
3068	zap_attribute_t za;
3069
3070	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3071	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3072
3073	if (spa->spa_feat_for_read_obj != 0) {
3074		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3075		    spa->spa_feat_for_read_obj);
3076		    zap_cursor_retrieve(&zc, &za) == 0;
3077		    zap_cursor_advance(&zc)) {
3078			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3079			    za.za_num_integers == 1);
3080			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3081			    za.za_first_integer));
3082		}
3083		zap_cursor_fini(&zc);
3084	}
3085
3086	if (spa->spa_feat_for_write_obj != 0) {
3087		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3088		    spa->spa_feat_for_write_obj);
3089		    zap_cursor_retrieve(&zc, &za) == 0;
3090		    zap_cursor_advance(&zc)) {
3091			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3092			    za.za_num_integers == 1);
3093			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3094			    za.za_first_integer));
3095		}
3096		zap_cursor_fini(&zc);
3097	}
3098
3099	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3100	    features) == 0);
3101	nvlist_free(features);
3102}
3103
3104int
3105spa_get_stats(const char *name, nvlist_t **config,
3106    char *altroot, size_t buflen)
3107{
3108	int error;
3109	spa_t *spa;
3110
3111	*config = NULL;
3112	error = spa_open_common(name, &spa, FTAG, NULL, config);
3113
3114	if (spa != NULL) {
3115		/*
3116		 * This still leaves a window of inconsistency where the spares
3117		 * or l2cache devices could change and the config would be
3118		 * self-inconsistent.
3119		 */
3120		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3121
3122		if (*config != NULL) {
3123			uint64_t loadtimes[2];
3124
3125			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3126			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3127			VERIFY(nvlist_add_uint64_array(*config,
3128			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3129
3130			VERIFY(nvlist_add_uint64(*config,
3131			    ZPOOL_CONFIG_ERRCOUNT,
3132			    spa_get_errlog_size(spa)) == 0);
3133
3134			if (spa_suspended(spa))
3135				VERIFY(nvlist_add_uint64(*config,
3136				    ZPOOL_CONFIG_SUSPENDED,
3137				    spa->spa_failmode) == 0);
3138
3139			spa_add_spares(spa, *config);
3140			spa_add_l2cache(spa, *config);
3141			spa_add_feature_stats(spa, *config);
3142		}
3143	}
3144
3145	/*
3146	 * We want to get the alternate root even for faulted pools, so we cheat
3147	 * and call spa_lookup() directly.
3148	 */
3149	if (altroot) {
3150		if (spa == NULL) {
3151			mutex_enter(&spa_namespace_lock);
3152			spa = spa_lookup(name);
3153			if (spa)
3154				spa_altroot(spa, altroot, buflen);
3155			else
3156				altroot[0] = '\0';
3157			spa = NULL;
3158			mutex_exit(&spa_namespace_lock);
3159		} else {
3160			spa_altroot(spa, altroot, buflen);
3161		}
3162	}
3163
3164	if (spa != NULL) {
3165		spa_config_exit(spa, SCL_CONFIG, FTAG);
3166		spa_close(spa, FTAG);
3167	}
3168
3169	return (error);
3170}
3171
3172/*
3173 * Validate that the auxiliary device array is well formed.  We must have an
3174 * array of nvlists, each which describes a valid leaf vdev.  If this is an
3175 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3176 * specified, as long as they are well-formed.
3177 */
3178static int
3179spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3180    spa_aux_vdev_t *sav, const char *config, uint64_t version,
3181    vdev_labeltype_t label)
3182{
3183	nvlist_t **dev;
3184	uint_t i, ndev;
3185	vdev_t *vd;
3186	int error;
3187
3188	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3189
3190	/*
3191	 * It's acceptable to have no devs specified.
3192	 */
3193	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3194		return (0);
3195
3196	if (ndev == 0)
3197		return (EINVAL);
3198
3199	/*
3200	 * Make sure the pool is formatted with a version that supports this
3201	 * device type.
3202	 */
3203	if (spa_version(spa) < version)
3204		return (ENOTSUP);
3205
3206	/*
3207	 * Set the pending device list so we correctly handle device in-use
3208	 * checking.
3209	 */
3210	sav->sav_pending = dev;
3211	sav->sav_npending = ndev;
3212
3213	for (i = 0; i < ndev; i++) {
3214		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3215		    mode)) != 0)
3216			goto out;
3217
3218		if (!vd->vdev_ops->vdev_op_leaf) {
3219			vdev_free(vd);
3220			error = EINVAL;
3221			goto out;
3222		}
3223
3224		/*
3225		 * The L2ARC currently only supports disk devices in
3226		 * kernel context.  For user-level testing, we allow it.
3227		 */
3228#ifdef _KERNEL
3229		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3230		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3231			error = ENOTBLK;
3232			vdev_free(vd);
3233			goto out;
3234		}
3235#endif
3236		vd->vdev_top = vd;
3237
3238		if ((error = vdev_open(vd)) == 0 &&
3239		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
3240			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3241			    vd->vdev_guid) == 0);
3242		}
3243
3244		vdev_free(vd);
3245
3246		if (error &&
3247		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3248			goto out;
3249		else
3250			error = 0;
3251	}
3252
3253out:
3254	sav->sav_pending = NULL;
3255	sav->sav_npending = 0;
3256	return (error);
3257}
3258
3259static int
3260spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3261{
3262	int error;
3263
3264	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3265
3266	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3267	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3268	    VDEV_LABEL_SPARE)) != 0) {
3269		return (error);
3270	}
3271
3272	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3273	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3274	    VDEV_LABEL_L2CACHE));
3275}
3276
3277static void
3278spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3279    const char *config)
3280{
3281	int i;
3282
3283	if (sav->sav_config != NULL) {
3284		nvlist_t **olddevs;
3285		uint_t oldndevs;
3286		nvlist_t **newdevs;
3287
3288		/*
3289		 * Generate new dev list by concatentating with the
3290		 * current dev list.
3291		 */
3292		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3293		    &olddevs, &oldndevs) == 0);
3294
3295		newdevs = kmem_alloc(sizeof (void *) *
3296		    (ndevs + oldndevs), KM_SLEEP);
3297		for (i = 0; i < oldndevs; i++)
3298			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3299			    KM_SLEEP) == 0);
3300		for (i = 0; i < ndevs; i++)
3301			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3302			    KM_SLEEP) == 0);
3303
3304		VERIFY(nvlist_remove(sav->sav_config, config,
3305		    DATA_TYPE_NVLIST_ARRAY) == 0);
3306
3307		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3308		    config, newdevs, ndevs + oldndevs) == 0);
3309		for (i = 0; i < oldndevs + ndevs; i++)
3310			nvlist_free(newdevs[i]);
3311		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3312	} else {
3313		/*
3314		 * Generate a new dev list.
3315		 */
3316		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3317		    KM_SLEEP) == 0);
3318		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3319		    devs, ndevs) == 0);
3320	}
3321}
3322
3323/*
3324 * Stop and drop level 2 ARC devices
3325 */
3326void
3327spa_l2cache_drop(spa_t *spa)
3328{
3329	vdev_t *vd;
3330	int i;
3331	spa_aux_vdev_t *sav = &spa->spa_l2cache;
3332
3333	for (i = 0; i < sav->sav_count; i++) {
3334		uint64_t pool;
3335
3336		vd = sav->sav_vdevs[i];
3337		ASSERT(vd != NULL);
3338
3339		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3340		    pool != 0ULL && l2arc_vdev_present(vd))
3341			l2arc_remove_vdev(vd);
3342	}
3343}
3344
3345/*
3346 * Pool Creation
3347 */
3348int
3349spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3350    nvlist_t *zplprops)
3351{
3352	spa_t *spa;
3353	char *altroot = NULL;
3354	vdev_t *rvd;
3355	dsl_pool_t *dp;
3356	dmu_tx_t *tx;
3357	int error = 0;
3358	uint64_t txg = TXG_INITIAL;
3359	nvlist_t **spares, **l2cache;
3360	uint_t nspares, nl2cache;
3361	uint64_t version, obj;
3362	boolean_t has_features;
3363
3364	/*
3365	 * If this pool already exists, return failure.
3366	 */
3367	mutex_enter(&spa_namespace_lock);
3368	if (spa_lookup(pool) != NULL) {
3369		mutex_exit(&spa_namespace_lock);
3370		return (EEXIST);
3371	}
3372
3373	/*
3374	 * Allocate a new spa_t structure.
3375	 */
3376	(void) nvlist_lookup_string(props,
3377	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3378	spa = spa_add(pool, NULL, altroot);
3379	spa_activate(spa, spa_mode_global);
3380
3381	if (props && (error = spa_prop_validate(spa, props))) {
3382		spa_deactivate(spa);
3383		spa_remove(spa);
3384		mutex_exit(&spa_namespace_lock);
3385		return (error);
3386	}
3387
3388	has_features = B_FALSE;
3389	for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3390	    elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3391		if (zpool_prop_feature(nvpair_name(elem)))
3392			has_features = B_TRUE;
3393	}
3394
3395	if (has_features || nvlist_lookup_uint64(props,
3396	    zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3397		version = SPA_VERSION;
3398	}
3399	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3400
3401	spa->spa_first_txg = txg;
3402	spa->spa_uberblock.ub_txg = txg - 1;
3403	spa->spa_uberblock.ub_version = version;
3404	spa->spa_ubsync = spa->spa_uberblock;
3405
3406	/*
3407	 * Create "The Godfather" zio to hold all async IOs
3408	 */
3409	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3410	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3411
3412	/*
3413	 * Create the root vdev.
3414	 */
3415	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3416
3417	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3418
3419	ASSERT(error != 0 || rvd != NULL);
3420	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3421
3422	if (error == 0 && !zfs_allocatable_devs(nvroot))
3423		error = EINVAL;
3424
3425	if (error == 0 &&
3426	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3427	    (error = spa_validate_aux(spa, nvroot, txg,
3428	    VDEV_ALLOC_ADD)) == 0) {
3429		for (int c = 0; c < rvd->vdev_children; c++) {
3430			vdev_metaslab_set_size(rvd->vdev_child[c]);
3431			vdev_expand(rvd->vdev_child[c], txg);
3432		}
3433	}
3434
3435	spa_config_exit(spa, SCL_ALL, FTAG);
3436
3437	if (error != 0) {
3438		spa_unload(spa);
3439		spa_deactivate(spa);
3440		spa_remove(spa);
3441		mutex_exit(&spa_namespace_lock);
3442		return (error);
3443	}
3444
3445	/*
3446	 * Get the list of spares, if specified.
3447	 */
3448	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3449	    &spares, &nspares) == 0) {
3450		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3451		    KM_SLEEP) == 0);
3452		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3453		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3454		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3455		spa_load_spares(spa);
3456		spa_config_exit(spa, SCL_ALL, FTAG);
3457		spa->spa_spares.sav_sync = B_TRUE;
3458	}
3459
3460	/*
3461	 * Get the list of level 2 cache devices, if specified.
3462	 */
3463	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3464	    &l2cache, &nl2cache) == 0) {
3465		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3466		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3467		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3468		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3469		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3470		spa_load_l2cache(spa);
3471		spa_config_exit(spa, SCL_ALL, FTAG);
3472		spa->spa_l2cache.sav_sync = B_TRUE;
3473	}
3474
3475	spa->spa_is_initializing = B_TRUE;
3476	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3477	spa->spa_meta_objset = dp->dp_meta_objset;
3478	spa->spa_is_initializing = B_FALSE;
3479
3480	/*
3481	 * Create DDTs (dedup tables).
3482	 */
3483	ddt_create(spa);
3484
3485	spa_update_dspace(spa);
3486
3487	tx = dmu_tx_create_assigned(dp, txg);
3488
3489	/*
3490	 * Create the pool config object.
3491	 */
3492	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3493	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3494	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3495
3496	if (zap_add(spa->spa_meta_objset,
3497	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3498	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3499		cmn_err(CE_PANIC, "failed to add pool config");
3500	}
3501
3502	if (spa_version(spa) >= SPA_VERSION_FEATURES)
3503		spa_feature_create_zap_objects(spa, tx);
3504
3505	if (zap_add(spa->spa_meta_objset,
3506	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3507	    sizeof (uint64_t), 1, &version, tx) != 0) {
3508		cmn_err(CE_PANIC, "failed to add pool version");
3509	}
3510
3511	/* Newly created pools with the right version are always deflated. */
3512	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3513		spa->spa_deflate = TRUE;
3514		if (zap_add(spa->spa_meta_objset,
3515		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3516		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3517			cmn_err(CE_PANIC, "failed to add deflate");
3518		}
3519	}
3520
3521	/*
3522	 * Create the deferred-free bpobj.  Turn off compression
3523	 * because sync-to-convergence takes longer if the blocksize
3524	 * keeps changing.
3525	 */
3526	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3527	dmu_object_set_compress(spa->spa_meta_objset, obj,
3528	    ZIO_COMPRESS_OFF, tx);
3529	if (zap_add(spa->spa_meta_objset,
3530	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3531	    sizeof (uint64_t), 1, &obj, tx) != 0) {
3532		cmn_err(CE_PANIC, "failed to add bpobj");
3533	}
3534	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3535	    spa->spa_meta_objset, obj));
3536
3537	/*
3538	 * Create the pool's history object.
3539	 */
3540	if (version >= SPA_VERSION_ZPOOL_HISTORY)
3541		spa_history_create_obj(spa, tx);
3542
3543	/*
3544	 * Set pool properties.
3545	 */
3546	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3547	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3548	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3549	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3550
3551	if (props != NULL) {
3552		spa_configfile_set(spa, props, B_FALSE);
3553		spa_sync_props(props, tx);
3554	}
3555
3556	dmu_tx_commit(tx);
3557
3558	spa->spa_sync_on = B_TRUE;
3559	txg_sync_start(spa->spa_dsl_pool);
3560
3561	/*
3562	 * We explicitly wait for the first transaction to complete so that our
3563	 * bean counters are appropriately updated.
3564	 */
3565	txg_wait_synced(spa->spa_dsl_pool, txg);
3566
3567	spa_config_sync(spa, B_FALSE, B_TRUE);
3568
3569	spa_history_log_version(spa, "create");
3570
3571	spa->spa_minref = refcount_count(&spa->spa_refcount);
3572
3573	mutex_exit(&spa_namespace_lock);
3574
3575	return (0);
3576}
3577
3578#ifdef _KERNEL
3579/*
3580 * Get the root pool information from the root disk, then import the root pool
3581 * during the system boot up time.
3582 */
3583extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3584
3585static nvlist_t *
3586spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3587{
3588	nvlist_t *config;
3589	nvlist_t *nvtop, *nvroot;
3590	uint64_t pgid;
3591
3592	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3593		return (NULL);
3594
3595	/*
3596	 * Add this top-level vdev to the child array.
3597	 */
3598	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3599	    &nvtop) == 0);
3600	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3601	    &pgid) == 0);
3602	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3603
3604	/*
3605	 * Put this pool's top-level vdevs into a root vdev.
3606	 */
3607	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3608	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3609	    VDEV_TYPE_ROOT) == 0);
3610	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3611	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3612	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3613	    &nvtop, 1) == 0);
3614
3615	/*
3616	 * Replace the existing vdev_tree with the new root vdev in
3617	 * this pool's configuration (remove the old, add the new).
3618	 */
3619	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3620	nvlist_free(nvroot);
3621	return (config);
3622}
3623
3624/*
3625 * Walk the vdev tree and see if we can find a device with "better"
3626 * configuration. A configuration is "better" if the label on that
3627 * device has a more recent txg.
3628 */
3629static void
3630spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3631{
3632	for (int c = 0; c < vd->vdev_children; c++)
3633		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3634
3635	if (vd->vdev_ops->vdev_op_leaf) {
3636		nvlist_t *label;
3637		uint64_t label_txg;
3638
3639		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3640		    &label) != 0)
3641			return;
3642
3643		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3644		    &label_txg) == 0);
3645
3646		/*
3647		 * Do we have a better boot device?
3648		 */
3649		if (label_txg > *txg) {
3650			*txg = label_txg;
3651			*avd = vd;
3652		}
3653		nvlist_free(label);
3654	}
3655}
3656
3657/*
3658 * Import a root pool.
3659 *
3660 * For x86. devpath_list will consist of devid and/or physpath name of
3661 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3662 * The GRUB "findroot" command will return the vdev we should boot.
3663 *
3664 * For Sparc, devpath_list consists the physpath name of the booting device
3665 * no matter the rootpool is a single device pool or a mirrored pool.
3666 * e.g.
3667 *	"/pci@1f,0/ide@d/disk@0,0:a"
3668 */
3669int
3670spa_import_rootpool(char *devpath, char *devid)
3671{
3672	spa_t *spa;
3673	vdev_t *rvd, *bvd, *avd = NULL;
3674	nvlist_t *config, *nvtop;
3675	uint64_t guid, txg;
3676	char *pname;
3677	int error;
3678
3679	/*
3680	 * Read the label from the boot device and generate a configuration.
3681	 */
3682	config = spa_generate_rootconf(devpath, devid, &guid);
3683#if defined(_OBP) && defined(_KERNEL)
3684	if (config == NULL) {
3685		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3686			/* iscsi boot */
3687			get_iscsi_bootpath_phy(devpath);
3688			config = spa_generate_rootconf(devpath, devid, &guid);
3689		}
3690	}
3691#endif
3692	if (config == NULL) {
3693		cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3694		    devpath);
3695		return (EIO);
3696	}
3697
3698	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3699	    &pname) == 0);
3700	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3701
3702	mutex_enter(&spa_namespace_lock);
3703	if ((spa = spa_lookup(pname)) != NULL) {
3704		/*
3705		 * Remove the existing root pool from the namespace so that we
3706		 * can replace it with the correct config we just read in.
3707		 */
3708		spa_remove(spa);
3709	}
3710
3711	spa = spa_add(pname, config, NULL);
3712	spa->spa_is_root = B_TRUE;
3713	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3714
3715	/*
3716	 * Build up a vdev tree based on the boot device's label config.
3717	 */
3718	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3719	    &nvtop) == 0);
3720	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3721	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3722	    VDEV_ALLOC_ROOTPOOL);
3723	spa_config_exit(spa, SCL_ALL, FTAG);
3724	if (error) {
3725		mutex_exit(&spa_namespace_lock);
3726		nvlist_free(config);
3727		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3728		    pname);
3729		return (error);
3730	}
3731
3732	/*
3733	 * Get the boot vdev.
3734	 */
3735	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3736		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3737		    (u_longlong_t)guid);
3738		error = ENOENT;
3739		goto out;
3740	}
3741
3742	/*
3743	 * Determine if there is a better boot device.
3744	 */
3745	avd = bvd;
3746	spa_alt_rootvdev(rvd, &avd, &txg);
3747	if (avd != bvd) {
3748		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3749		    "try booting from '%s'", avd->vdev_path);
3750		error = EINVAL;
3751		goto out;
3752	}
3753
3754	/*
3755	 * If the boot device is part of a spare vdev then ensure that
3756	 * we're booting off the active spare.
3757	 */
3758	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3759	    !bvd->vdev_isspare) {
3760		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3761		    "try booting from '%s'",
3762		    bvd->vdev_parent->
3763		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3764		error = EINVAL;
3765		goto out;
3766	}
3767
3768	error = 0;
3769out:
3770	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3771	vdev_free(rvd);
3772	spa_config_exit(spa, SCL_ALL, FTAG);
3773	mutex_exit(&spa_namespace_lock);
3774
3775	nvlist_free(config);
3776	return (error);
3777}
3778
3779#endif
3780
3781/*
3782 * Import a non-root pool into the system.
3783 */
3784int
3785spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3786{
3787	spa_t *spa;
3788	char *altroot = NULL;
3789	spa_load_state_t state = SPA_LOAD_IMPORT;
3790	zpool_rewind_policy_t policy;
3791	uint64_t mode = spa_mode_global;
3792	uint64_t readonly = B_FALSE;
3793	int error;
3794	nvlist_t *nvroot;
3795	nvlist_t **spares, **l2cache;
3796	uint_t nspares, nl2cache;
3797
3798	/*
3799	 * If a pool with this name exists, return failure.
3800	 */
3801	mutex_enter(&spa_namespace_lock);
3802	if (spa_lookup(pool) != NULL) {
3803		mutex_exit(&spa_namespace_lock);
3804		return (EEXIST);
3805	}
3806
3807	/*
3808	 * Create and initialize the spa structure.
3809	 */
3810	(void) nvlist_lookup_string(props,
3811	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3812	(void) nvlist_lookup_uint64(props,
3813	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3814	if (readonly)
3815		mode = FREAD;
3816	spa = spa_add(pool, config, altroot);
3817	spa->spa_import_flags = flags;
3818
3819	/*
3820	 * Verbatim import - Take a pool and insert it into the namespace
3821	 * as if it had been loaded at boot.
3822	 */
3823	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3824		if (props != NULL)
3825			spa_configfile_set(spa, props, B_FALSE);
3826
3827		spa_config_sync(spa, B_FALSE, B_TRUE);
3828
3829		mutex_exit(&spa_namespace_lock);
3830		spa_history_log_version(spa, "import");
3831
3832		return (0);
3833	}
3834
3835	spa_activate(spa, mode);
3836
3837	/*
3838	 * Don't start async tasks until we know everything is healthy.
3839	 */
3840	spa_async_suspend(spa);
3841
3842	zpool_get_rewind_policy(config, &policy);
3843	if (policy.zrp_request & ZPOOL_DO_REWIND)
3844		state = SPA_LOAD_RECOVER;
3845
3846	/*
3847	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3848	 * because the user-supplied config is actually the one to trust when
3849	 * doing an import.
3850	 */
3851	if (state != SPA_LOAD_RECOVER)
3852		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3853
3854	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3855	    policy.zrp_request);
3856
3857	/*
3858	 * Propagate anything learned while loading the pool and pass it
3859	 * back to caller (i.e. rewind info, missing devices, etc).
3860	 */
3861	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3862	    spa->spa_load_info) == 0);
3863
3864	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3865	/*
3866	 * Toss any existing sparelist, as it doesn't have any validity
3867	 * anymore, and conflicts with spa_has_spare().
3868	 */
3869	if (spa->spa_spares.sav_config) {
3870		nvlist_free(spa->spa_spares.sav_config);
3871		spa->spa_spares.sav_config = NULL;
3872		spa_load_spares(spa);
3873	}
3874	if (spa->spa_l2cache.sav_config) {
3875		nvlist_free(spa->spa_l2cache.sav_config);
3876		spa->spa_l2cache.sav_config = NULL;
3877		spa_load_l2cache(spa);
3878	}
3879
3880	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3881	    &nvroot) == 0);
3882	if (error == 0)
3883		error = spa_validate_aux(spa, nvroot, -1ULL,
3884		    VDEV_ALLOC_SPARE);
3885	if (error == 0)
3886		error = spa_validate_aux(spa, nvroot, -1ULL,
3887		    VDEV_ALLOC_L2CACHE);
3888	spa_config_exit(spa, SCL_ALL, FTAG);
3889
3890	if (props != NULL)
3891		spa_configfile_set(spa, props, B_FALSE);
3892
3893	if (error != 0 || (props && spa_writeable(spa) &&
3894	    (error = spa_prop_set(spa, props)))) {
3895		spa_unload(spa);
3896		spa_deactivate(spa);
3897		spa_remove(spa);
3898		mutex_exit(&spa_namespace_lock);
3899		return (error);
3900	}
3901
3902	spa_async_resume(spa);
3903
3904	/*
3905	 * Override any spares and level 2 cache devices as specified by
3906	 * the user, as these may have correct device names/devids, etc.
3907	 */
3908	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3909	    &spares, &nspares) == 0) {
3910		if (spa->spa_spares.sav_config)
3911			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3912			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3913		else
3914			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3915			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3916		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3917		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3918		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3919		spa_load_spares(spa);
3920		spa_config_exit(spa, SCL_ALL, FTAG);
3921		spa->spa_spares.sav_sync = B_TRUE;
3922	}
3923	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3924	    &l2cache, &nl2cache) == 0) {
3925		if (spa->spa_l2cache.sav_config)
3926			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3927			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3928		else
3929			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3930			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3931		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3932		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3933		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3934		spa_load_l2cache(spa);
3935		spa_config_exit(spa, SCL_ALL, FTAG);
3936		spa->spa_l2cache.sav_sync = B_TRUE;
3937	}
3938
3939	/*
3940	 * Check for any removed devices.
3941	 */
3942	if (spa->spa_autoreplace) {
3943		spa_aux_check_removed(&spa->spa_spares);
3944		spa_aux_check_removed(&spa->spa_l2cache);
3945	}
3946
3947	if (spa_writeable(spa)) {
3948		/*
3949		 * Update the config cache to include the newly-imported pool.
3950		 */
3951		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3952	}
3953
3954	/*
3955	 * It's possible that the pool was expanded while it was exported.
3956	 * We kick off an async task to handle this for us.
3957	 */
3958	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3959
3960	mutex_exit(&spa_namespace_lock);
3961	spa_history_log_version(spa, "import");
3962
3963	return (0);
3964}
3965
3966nvlist_t *
3967spa_tryimport(nvlist_t *tryconfig)
3968{
3969	nvlist_t *config = NULL;
3970	char *poolname;
3971	spa_t *spa;
3972	uint64_t state;
3973	int error;
3974
3975	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3976		return (NULL);
3977
3978	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3979		return (NULL);
3980
3981	/*
3982	 * Create and initialize the spa structure.
3983	 */
3984	mutex_enter(&spa_namespace_lock);
3985	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3986	spa_activate(spa, FREAD);
3987
3988	/*
3989	 * Pass off the heavy lifting to spa_load().
3990	 * Pass TRUE for mosconfig because the user-supplied config
3991	 * is actually the one to trust when doing an import.
3992	 */
3993	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3994
3995	/*
3996	 * If 'tryconfig' was at least parsable, return the current config.
3997	 */
3998	if (spa->spa_root_vdev != NULL) {
3999		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4000		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4001		    poolname) == 0);
4002		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4003		    state) == 0);
4004		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4005		    spa->spa_uberblock.ub_timestamp) == 0);
4006		VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4007		    spa->spa_load_info) == 0);
4008
4009		/*
4010		 * If the bootfs property exists on this pool then we
4011		 * copy it out so that external consumers can tell which
4012		 * pools are bootable.
4013		 */
4014		if ((!error || error == EEXIST) && spa->spa_bootfs) {
4015			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4016
4017			/*
4018			 * We have to play games with the name since the
4019			 * pool was opened as TRYIMPORT_NAME.
4020			 */
4021			if (dsl_dsobj_to_dsname(spa_name(spa),
4022			    spa->spa_bootfs, tmpname) == 0) {
4023				char *cp;
4024				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4025
4026				cp = strchr(tmpname, '/');
4027				if (cp == NULL) {
4028					(void) strlcpy(dsname, tmpname,
4029					    MAXPATHLEN);
4030				} else {
4031					(void) snprintf(dsname, MAXPATHLEN,
4032					    "%s/%s", poolname, ++cp);
4033				}
4034				VERIFY(nvlist_add_string(config,
4035				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4036				kmem_free(dsname, MAXPATHLEN);
4037			}
4038			kmem_free(tmpname, MAXPATHLEN);
4039		}
4040
4041		/*
4042		 * Add the list of hot spares and level 2 cache devices.
4043		 */
4044		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4045		spa_add_spares(spa, config);
4046		spa_add_l2cache(spa, config);
4047		spa_config_exit(spa, SCL_CONFIG, FTAG);
4048	}
4049
4050	spa_unload(spa);
4051	spa_deactivate(spa);
4052	spa_remove(spa);
4053	mutex_exit(&spa_namespace_lock);
4054
4055	return (config);
4056}
4057
4058/*
4059 * Pool export/destroy
4060 *
4061 * The act of destroying or exporting a pool is very simple.  We make sure there
4062 * is no more pending I/O and any references to the pool are gone.  Then, we
4063 * update the pool state and sync all the labels to disk, removing the
4064 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4065 * we don't sync the labels or remove the configuration cache.
4066 */
4067static int
4068spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4069    boolean_t force, boolean_t hardforce)
4070{
4071	spa_t *spa;
4072
4073	if (oldconfig)
4074		*oldconfig = NULL;
4075
4076	if (!(spa_mode_global & FWRITE))
4077		return (EROFS);
4078
4079	mutex_enter(&spa_namespace_lock);
4080	if ((spa = spa_lookup(pool)) == NULL) {
4081		mutex_exit(&spa_namespace_lock);
4082		return (ENOENT);
4083	}
4084
4085	/*
4086	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4087	 * reacquire the namespace lock, and see if we can export.
4088	 */
4089	spa_open_ref(spa, FTAG);
4090	mutex_exit(&spa_namespace_lock);
4091	spa_async_suspend(spa);
4092	mutex_enter(&spa_namespace_lock);
4093	spa_close(spa, FTAG);
4094
4095	/*
4096	 * The pool will be in core if it's openable,
4097	 * in which case we can modify its state.
4098	 */
4099	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4100		/*
4101		 * Objsets may be open only because they're dirty, so we
4102		 * have to force it to sync before checking spa_refcnt.
4103		 */
4104		txg_wait_synced(spa->spa_dsl_pool, 0);
4105
4106		/*
4107		 * A pool cannot be exported or destroyed if there are active
4108		 * references.  If we are resetting a pool, allow references by
4109		 * fault injection handlers.
4110		 */
4111		if (!spa_refcount_zero(spa) ||
4112		    (spa->spa_inject_ref != 0 &&
4113		    new_state != POOL_STATE_UNINITIALIZED)) {
4114			spa_async_resume(spa);
4115			mutex_exit(&spa_namespace_lock);
4116			return (EBUSY);
4117		}
4118
4119		/*
4120		 * A pool cannot be exported if it has an active shared spare.
4121		 * This is to prevent other pools stealing the active spare
4122		 * from an exported pool. At user's own will, such pool can
4123		 * be forcedly exported.
4124		 */
4125		if (!force && new_state == POOL_STATE_EXPORTED &&
4126		    spa_has_active_shared_spare(spa)) {
4127			spa_async_resume(spa);
4128			mutex_exit(&spa_namespace_lock);
4129			return (EXDEV);
4130		}
4131
4132		/*
4133		 * We want this to be reflected on every label,
4134		 * so mark them all dirty.  spa_unload() will do the
4135		 * final sync that pushes these changes out.
4136		 */
4137		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4138			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4139			spa->spa_state = new_state;
4140			spa->spa_final_txg = spa_last_synced_txg(spa) +
4141			    TXG_DEFER_SIZE + 1;
4142			vdev_config_dirty(spa->spa_root_vdev);
4143			spa_config_exit(spa, SCL_ALL, FTAG);
4144		}
4145	}
4146
4147	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4148
4149	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4150		spa_unload(spa);
4151		spa_deactivate(spa);
4152	}
4153
4154	if (oldconfig && spa->spa_config)
4155		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4156
4157	if (new_state != POOL_STATE_UNINITIALIZED) {
4158		if (!hardforce)
4159			spa_config_sync(spa, B_TRUE, B_TRUE);
4160		spa_remove(spa);
4161	}
4162	mutex_exit(&spa_namespace_lock);
4163
4164	return (0);
4165}
4166
4167/*
4168 * Destroy a storage pool.
4169 */
4170int
4171spa_destroy(char *pool)
4172{
4173	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4174	    B_FALSE, B_FALSE));
4175}
4176
4177/*
4178 * Export a storage pool.
4179 */
4180int
4181spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4182    boolean_t hardforce)
4183{
4184	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4185	    force, hardforce));
4186}
4187
4188/*
4189 * Similar to spa_export(), this unloads the spa_t without actually removing it
4190 * from the namespace in any way.
4191 */
4192int
4193spa_reset(char *pool)
4194{
4195	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4196	    B_FALSE, B_FALSE));
4197}
4198
4199/*
4200 * ==========================================================================
4201 * Device manipulation
4202 * ==========================================================================
4203 */
4204
4205/*
4206 * Add a device to a storage pool.
4207 */
4208int
4209spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4210{
4211	uint64_t txg, id;
4212	int error;
4213	vdev_t *rvd = spa->spa_root_vdev;
4214	vdev_t *vd, *tvd;
4215	nvlist_t **spares, **l2cache;
4216	uint_t nspares, nl2cache;
4217
4218	ASSERT(spa_writeable(spa));
4219
4220	txg = spa_vdev_enter(spa);
4221
4222	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4223	    VDEV_ALLOC_ADD)) != 0)
4224		return (spa_vdev_exit(spa, NULL, txg, error));
4225
4226	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
4227
4228	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4229	    &nspares) != 0)
4230		nspares = 0;
4231
4232	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4233	    &nl2cache) != 0)
4234		nl2cache = 0;
4235
4236	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4237		return (spa_vdev_exit(spa, vd, txg, EINVAL));
4238
4239	if (vd->vdev_children != 0 &&
4240	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
4241		return (spa_vdev_exit(spa, vd, txg, error));
4242
4243	/*
4244	 * We must validate the spares and l2cache devices after checking the
4245	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
4246	 */
4247	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4248		return (spa_vdev_exit(spa, vd, txg, error));
4249
4250	/*
4251	 * Transfer each new top-level vdev from vd to rvd.
4252	 */
4253	for (int c = 0; c < vd->vdev_children; c++) {
4254
4255		/*
4256		 * Set the vdev id to the first hole, if one exists.
4257		 */
4258		for (id = 0; id < rvd->vdev_children; id++) {
4259			if (rvd->vdev_child[id]->vdev_ishole) {
4260				vdev_free(rvd->vdev_child[id]);
4261				break;
4262			}
4263		}
4264		tvd = vd->vdev_child[c];
4265		vdev_remove_child(vd, tvd);
4266		tvd->vdev_id = id;
4267		vdev_add_child(rvd, tvd);
4268		vdev_config_dirty(tvd);
4269	}
4270
4271	if (nspares != 0) {
4272		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4273		    ZPOOL_CONFIG_SPARES);
4274		spa_load_spares(spa);
4275		spa->spa_spares.sav_sync = B_TRUE;
4276	}
4277
4278	if (nl2cache != 0) {
4279		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4280		    ZPOOL_CONFIG_L2CACHE);
4281		spa_load_l2cache(spa);
4282		spa->spa_l2cache.sav_sync = B_TRUE;
4283	}
4284
4285	/*
4286	 * We have to be careful when adding new vdevs to an existing pool.
4287	 * If other threads start allocating from these vdevs before we
4288	 * sync the config cache, and we lose power, then upon reboot we may
4289	 * fail to open the pool because there are DVAs that the config cache
4290	 * can't translate.  Therefore, we first add the vdevs without
4291	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4292	 * and then let spa_config_update() initialize the new metaslabs.
4293	 *
4294	 * spa_load() checks for added-but-not-initialized vdevs, so that
4295	 * if we lose power at any point in this sequence, the remaining
4296	 * steps will be completed the next time we load the pool.
4297	 */
4298	(void) spa_vdev_exit(spa, vd, txg, 0);
4299
4300	mutex_enter(&spa_namespace_lock);
4301	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4302	mutex_exit(&spa_namespace_lock);
4303
4304	return (0);
4305}
4306
4307/*
4308 * Attach a device to a mirror.  The arguments are the path to any device
4309 * in the mirror, and the nvroot for the new device.  If the path specifies
4310 * a device that is not mirrored, we automatically insert the mirror vdev.
4311 *
4312 * If 'replacing' is specified, the new device is intended to replace the
4313 * existing device; in this case the two devices are made into their own
4314 * mirror using the 'replacing' vdev, which is functionally identical to
4315 * the mirror vdev (it actually reuses all the same ops) but has a few
4316 * extra rules: you can't attach to it after it's been created, and upon
4317 * completion of resilvering, the first disk (the one being replaced)
4318 * is automatically detached.
4319 */
4320int
4321spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4322{
4323	uint64_t txg, dtl_max_txg;
4324	vdev_t *rvd = spa->spa_root_vdev;
4325	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4326	vdev_ops_t *pvops;
4327	char *oldvdpath, *newvdpath;
4328	int newvd_isspare;
4329	int error;
4330
4331	ASSERT(spa_writeable(spa));
4332
4333	txg = spa_vdev_enter(spa);
4334
4335	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4336
4337	if (oldvd == NULL)
4338		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4339
4340	if (!oldvd->vdev_ops->vdev_op_leaf)
4341		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4342
4343	pvd = oldvd->vdev_parent;
4344
4345	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4346	    VDEV_ALLOC_ATTACH)) != 0)
4347		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4348
4349	if (newrootvd->vdev_children != 1)
4350		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4351
4352	newvd = newrootvd->vdev_child[0];
4353
4354	if (!newvd->vdev_ops->vdev_op_leaf)
4355		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4356
4357	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4358		return (spa_vdev_exit(spa, newrootvd, txg, error));
4359
4360	/*
4361	 * Spares can't replace logs
4362	 */
4363	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4364		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4365
4366	if (!replacing) {
4367		/*
4368		 * For attach, the only allowable parent is a mirror or the root
4369		 * vdev.
4370		 */
4371		if (pvd->vdev_ops != &vdev_mirror_ops &&
4372		    pvd->vdev_ops != &vdev_root_ops)
4373			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4374
4375		pvops = &vdev_mirror_ops;
4376	} else {
4377		/*
4378		 * Active hot spares can only be replaced by inactive hot
4379		 * spares.
4380		 */
4381		if (pvd->vdev_ops == &vdev_spare_ops &&
4382		    oldvd->vdev_isspare &&
4383		    !spa_has_spare(spa, newvd->vdev_guid))
4384			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4385
4386		/*
4387		 * If the source is a hot spare, and the parent isn't already a
4388		 * spare, then we want to create a new hot spare.  Otherwise, we
4389		 * want to create a replacing vdev.  The user is not allowed to
4390		 * attach to a spared vdev child unless the 'isspare' state is
4391		 * the same (spare replaces spare, non-spare replaces
4392		 * non-spare).
4393		 */
4394		if (pvd->vdev_ops == &vdev_replacing_ops &&
4395		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4396			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4397		} else if (pvd->vdev_ops == &vdev_spare_ops &&
4398		    newvd->vdev_isspare != oldvd->vdev_isspare) {
4399			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4400		}
4401
4402		if (newvd->vdev_isspare)
4403			pvops = &vdev_spare_ops;
4404		else
4405			pvops = &vdev_replacing_ops;
4406	}
4407
4408	/*
4409	 * Make sure the new device is big enough.
4410	 */
4411	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4412		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4413
4414	/*
4415	 * The new device cannot have a higher alignment requirement
4416	 * than the top-level vdev.
4417	 */
4418	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4419		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4420
4421	/*
4422	 * If this is an in-place replacement, update oldvd's path and devid
4423	 * to make it distinguishable from newvd, and unopenable from now on.
4424	 */
4425	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4426		spa_strfree(oldvd->vdev_path);
4427		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4428		    KM_SLEEP);
4429		(void) sprintf(oldvd->vdev_path, "%s/%s",
4430		    newvd->vdev_path, "old");
4431		if (oldvd->vdev_devid != NULL) {
4432			spa_strfree(oldvd->vdev_devid);
4433			oldvd->vdev_devid = NULL;
4434		}
4435	}
4436
4437	/* mark the device being resilvered */
4438	newvd->vdev_resilvering = B_TRUE;
4439
4440	/*
4441	 * If the parent is not a mirror, or if we're replacing, insert the new
4442	 * mirror/replacing/spare vdev above oldvd.
4443	 */
4444	if (pvd->vdev_ops != pvops)
4445		pvd = vdev_add_parent(oldvd, pvops);
4446
4447	ASSERT(pvd->vdev_top->vdev_parent == rvd);
4448	ASSERT(pvd->vdev_ops == pvops);
4449	ASSERT(oldvd->vdev_parent == pvd);
4450
4451	/*
4452	 * Extract the new device from its root and add it to pvd.
4453	 */
4454	vdev_remove_child(newrootvd, newvd);
4455	newvd->vdev_id = pvd->vdev_children;
4456	newvd->vdev_crtxg = oldvd->vdev_crtxg;
4457	vdev_add_child(pvd, newvd);
4458
4459	tvd = newvd->vdev_top;
4460	ASSERT(pvd->vdev_top == tvd);
4461	ASSERT(tvd->vdev_parent == rvd);
4462
4463	vdev_config_dirty(tvd);
4464
4465	/*
4466	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4467	 * for any dmu_sync-ed blocks.  It will propagate upward when
4468	 * spa_vdev_exit() calls vdev_dtl_reassess().
4469	 */
4470	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4471
4472	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4473	    dtl_max_txg - TXG_INITIAL);
4474
4475	if (newvd->vdev_isspare) {
4476		spa_spare_activate(newvd);
4477		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4478	}
4479
4480	oldvdpath = spa_strdup(oldvd->vdev_path);
4481	newvdpath = spa_strdup(newvd->vdev_path);
4482	newvd_isspare = newvd->vdev_isspare;
4483
4484	/*
4485	 * Mark newvd's DTL dirty in this txg.
4486	 */
4487	vdev_dirty(tvd, VDD_DTL, newvd, txg);
4488
4489	/*
4490	 * Restart the resilver
4491	 */
4492	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4493
4494	/*
4495	 * Commit the config
4496	 */
4497	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4498
4499	spa_history_log_internal(spa, "vdev attach", NULL,
4500	    "%s vdev=%s %s vdev=%s",
4501	    replacing && newvd_isspare ? "spare in" :
4502	    replacing ? "replace" : "attach", newvdpath,
4503	    replacing ? "for" : "to", oldvdpath);
4504
4505	spa_strfree(oldvdpath);
4506	spa_strfree(newvdpath);
4507
4508	if (spa->spa_bootfs)
4509		spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4510
4511	return (0);
4512}
4513
4514/*
4515 * Detach a device from a mirror or replacing vdev.
4516 * If 'replace_done' is specified, only detach if the parent
4517 * is a replacing vdev.
4518 */
4519int
4520spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4521{
4522	uint64_t txg;
4523	int error;
4524	vdev_t *rvd = spa->spa_root_vdev;
4525	vdev_t *vd, *pvd, *cvd, *tvd;
4526	boolean_t unspare = B_FALSE;
4527	uint64_t unspare_guid = 0;
4528	char *vdpath;
4529
4530	ASSERT(spa_writeable(spa));
4531
4532	txg = spa_vdev_enter(spa);
4533
4534	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4535
4536	if (vd == NULL)
4537		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4538
4539	if (!vd->vdev_ops->vdev_op_leaf)
4540		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4541
4542	pvd = vd->vdev_parent;
4543
4544	/*
4545	 * If the parent/child relationship is not as expected, don't do it.
4546	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4547	 * vdev that's replacing B with C.  The user's intent in replacing
4548	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4549	 * the replace by detaching C, the expected behavior is to end up
4550	 * M(A,B).  But suppose that right after deciding to detach C,
4551	 * the replacement of B completes.  We would have M(A,C), and then
4552	 * ask to detach C, which would leave us with just A -- not what
4553	 * the user wanted.  To prevent this, we make sure that the
4554	 * parent/child relationship hasn't changed -- in this example,
4555	 * that C's parent is still the replacing vdev R.
4556	 */
4557	if (pvd->vdev_guid != pguid && pguid != 0)
4558		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4559
4560	/*
4561	 * Only 'replacing' or 'spare' vdevs can be replaced.
4562	 */
4563	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4564	    pvd->vdev_ops != &vdev_spare_ops)
4565		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4566
4567	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4568	    spa_version(spa) >= SPA_VERSION_SPARES);
4569
4570	/*
4571	 * Only mirror, replacing, and spare vdevs support detach.
4572	 */
4573	if (pvd->vdev_ops != &vdev_replacing_ops &&
4574	    pvd->vdev_ops != &vdev_mirror_ops &&
4575	    pvd->vdev_ops != &vdev_spare_ops)
4576		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4577
4578	/*
4579	 * If this device has the only valid copy of some data,
4580	 * we cannot safely detach it.
4581	 */
4582	if (vdev_dtl_required(vd))
4583		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4584
4585	ASSERT(pvd->vdev_children >= 2);
4586
4587	/*
4588	 * If we are detaching the second disk from a replacing vdev, then
4589	 * check to see if we changed the original vdev's path to have "/old"
4590	 * at the end in spa_vdev_attach().  If so, undo that change now.
4591	 */
4592	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4593	    vd->vdev_path != NULL) {
4594		size_t len = strlen(vd->vdev_path);
4595
4596		for (int c = 0; c < pvd->vdev_children; c++) {
4597			cvd = pvd->vdev_child[c];
4598
4599			if (cvd == vd || cvd->vdev_path == NULL)
4600				continue;
4601
4602			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4603			    strcmp(cvd->vdev_path + len, "/old") == 0) {
4604				spa_strfree(cvd->vdev_path);
4605				cvd->vdev_path = spa_strdup(vd->vdev_path);
4606				break;
4607			}
4608		}
4609	}
4610
4611	/*
4612	 * If we are detaching the original disk from a spare, then it implies
4613	 * that the spare should become a real disk, and be removed from the
4614	 * active spare list for the pool.
4615	 */
4616	if (pvd->vdev_ops == &vdev_spare_ops &&
4617	    vd->vdev_id == 0 &&
4618	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4619		unspare = B_TRUE;
4620
4621	/*
4622	 * Erase the disk labels so the disk can be used for other things.
4623	 * This must be done after all other error cases are handled,
4624	 * but before we disembowel vd (so we can still do I/O to it).
4625	 * But if we can't do it, don't treat the error as fatal --
4626	 * it may be that the unwritability of the disk is the reason
4627	 * it's being detached!
4628	 */
4629	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4630
4631	/*
4632	 * Remove vd from its parent and compact the parent's children.
4633	 */
4634	vdev_remove_child(pvd, vd);
4635	vdev_compact_children(pvd);
4636
4637	/*
4638	 * Remember one of the remaining children so we can get tvd below.
4639	 */
4640	cvd = pvd->vdev_child[pvd->vdev_children - 1];
4641
4642	/*
4643	 * If we need to remove the remaining child from the list of hot spares,
4644	 * do it now, marking the vdev as no longer a spare in the process.
4645	 * We must do this before vdev_remove_parent(), because that can
4646	 * change the GUID if it creates a new toplevel GUID.  For a similar
4647	 * reason, we must remove the spare now, in the same txg as the detach;
4648	 * otherwise someone could attach a new sibling, change the GUID, and
4649	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4650	 */
4651	if (unspare) {
4652		ASSERT(cvd->vdev_isspare);
4653		spa_spare_remove(cvd);
4654		unspare_guid = cvd->vdev_guid;
4655		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4656		cvd->vdev_unspare = B_TRUE;
4657	}
4658
4659	/*
4660	 * If the parent mirror/replacing vdev only has one child,
4661	 * the parent is no longer needed.  Remove it from the tree.
4662	 */
4663	if (pvd->vdev_children == 1) {
4664		if (pvd->vdev_ops == &vdev_spare_ops)
4665			cvd->vdev_unspare = B_FALSE;
4666		vdev_remove_parent(cvd);
4667		cvd->vdev_resilvering = B_FALSE;
4668	}
4669
4670
4671	/*
4672	 * We don't set tvd until now because the parent we just removed
4673	 * may have been the previous top-level vdev.
4674	 */
4675	tvd = cvd->vdev_top;
4676	ASSERT(tvd->vdev_parent == rvd);
4677
4678	/*
4679	 * Reevaluate the parent vdev state.
4680	 */
4681	vdev_propagate_state(cvd);
4682
4683	/*
4684	 * If the 'autoexpand' property is set on the pool then automatically
4685	 * try to expand the size of the pool. For example if the device we
4686	 * just detached was smaller than the others, it may be possible to
4687	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4688	 * first so that we can obtain the updated sizes of the leaf vdevs.
4689	 */
4690	if (spa->spa_autoexpand) {
4691		vdev_reopen(tvd);
4692		vdev_expand(tvd, txg);
4693	}
4694
4695	vdev_config_dirty(tvd);
4696
4697	/*
4698	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4699	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4700	 * But first make sure we're not on any *other* txg's DTL list, to
4701	 * prevent vd from being accessed after it's freed.
4702	 */
4703	vdpath = spa_strdup(vd->vdev_path);
4704	for (int t = 0; t < TXG_SIZE; t++)
4705		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4706	vd->vdev_detached = B_TRUE;
4707	vdev_dirty(tvd, VDD_DTL, vd, txg);
4708
4709	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4710
4711	/* hang on to the spa before we release the lock */
4712	spa_open_ref(spa, FTAG);
4713
4714	error = spa_vdev_exit(spa, vd, txg, 0);
4715
4716	spa_history_log_internal(spa, "detach", NULL,
4717	    "vdev=%s", vdpath);
4718	spa_strfree(vdpath);
4719
4720	/*
4721	 * If this was the removal of the original device in a hot spare vdev,
4722	 * then we want to go through and remove the device from the hot spare
4723	 * list of every other pool.
4724	 */
4725	if (unspare) {
4726		spa_t *altspa = NULL;
4727
4728		mutex_enter(&spa_namespace_lock);
4729		while ((altspa = spa_next(altspa)) != NULL) {
4730			if (altspa->spa_state != POOL_STATE_ACTIVE ||
4731			    altspa == spa)
4732				continue;
4733
4734			spa_open_ref(altspa, FTAG);
4735			mutex_exit(&spa_namespace_lock);
4736			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4737			mutex_enter(&spa_namespace_lock);
4738			spa_close(altspa, FTAG);
4739		}
4740		mutex_exit(&spa_namespace_lock);
4741
4742		/* search the rest of the vdevs for spares to remove */
4743		spa_vdev_resilver_done(spa);
4744	}
4745
4746	/* all done with the spa; OK to release */
4747	mutex_enter(&spa_namespace_lock);
4748	spa_close(spa, FTAG);
4749	mutex_exit(&spa_namespace_lock);
4750
4751	return (error);
4752}
4753
4754/*
4755 * Split a set of devices from their mirrors, and create a new pool from them.
4756 */
4757int
4758spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4759    nvlist_t *props, boolean_t exp)
4760{
4761	int error = 0;
4762	uint64_t txg, *glist;
4763	spa_t *newspa;
4764	uint_t c, children, lastlog;
4765	nvlist_t **child, *nvl, *tmp;
4766	dmu_tx_t *tx;
4767	char *altroot = NULL;
4768	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4769	boolean_t activate_slog;
4770
4771	ASSERT(spa_writeable(spa));
4772
4773	txg = spa_vdev_enter(spa);
4774
4775	/* clear the log and flush everything up to now */
4776	activate_slog = spa_passivate_log(spa);
4777	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4778	error = spa_offline_log(spa);
4779	txg = spa_vdev_config_enter(spa);
4780
4781	if (activate_slog)
4782		spa_activate_log(spa);
4783
4784	if (error != 0)
4785		return (spa_vdev_exit(spa, NULL, txg, error));
4786
4787	/* check new spa name before going any further */
4788	if (spa_lookup(newname) != NULL)
4789		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4790
4791	/*
4792	 * scan through all the children to ensure they're all mirrors
4793	 */
4794	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4795	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4796	    &children) != 0)
4797		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4798
4799	/* first, check to ensure we've got the right child count */
4800	rvd = spa->spa_root_vdev;
4801	lastlog = 0;
4802	for (c = 0; c < rvd->vdev_children; c++) {
4803		vdev_t *vd = rvd->vdev_child[c];
4804
4805		/* don't count the holes & logs as children */
4806		if (vd->vdev_islog || vd->vdev_ishole) {
4807			if (lastlog == 0)
4808				lastlog = c;
4809			continue;
4810		}
4811
4812		lastlog = 0;
4813	}
4814	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4815		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4816
4817	/* next, ensure no spare or cache devices are part of the split */
4818	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4819	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4820		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4821
4822	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4823	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4824
4825	/* then, loop over each vdev and validate it */
4826	for (c = 0; c < children; c++) {
4827		uint64_t is_hole = 0;
4828
4829		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4830		    &is_hole);
4831
4832		if (is_hole != 0) {
4833			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4834			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4835				continue;
4836			} else {
4837				error = EINVAL;
4838				break;
4839			}
4840		}
4841
4842		/* which disk is going to be split? */
4843		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4844		    &glist[c]) != 0) {
4845			error = EINVAL;
4846			break;
4847		}
4848
4849		/* look it up in the spa */
4850		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4851		if (vml[c] == NULL) {
4852			error = ENODEV;
4853			break;
4854		}
4855
4856		/* make sure there's nothing stopping the split */
4857		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4858		    vml[c]->vdev_islog ||
4859		    vml[c]->vdev_ishole ||
4860		    vml[c]->vdev_isspare ||
4861		    vml[c]->vdev_isl2cache ||
4862		    !vdev_writeable(vml[c]) ||
4863		    vml[c]->vdev_children != 0 ||
4864		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4865		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4866			error = EINVAL;
4867			break;
4868		}
4869
4870		if (vdev_dtl_required(vml[c])) {
4871			error = EBUSY;
4872			break;
4873		}
4874
4875		/* we need certain info from the top level */
4876		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4877		    vml[c]->vdev_top->vdev_ms_array) == 0);
4878		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4879		    vml[c]->vdev_top->vdev_ms_shift) == 0);
4880		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4881		    vml[c]->vdev_top->vdev_asize) == 0);
4882		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4883		    vml[c]->vdev_top->vdev_ashift) == 0);
4884	}
4885
4886	if (error != 0) {
4887		kmem_free(vml, children * sizeof (vdev_t *));
4888		kmem_free(glist, children * sizeof (uint64_t));
4889		return (spa_vdev_exit(spa, NULL, txg, error));
4890	}
4891
4892	/* stop writers from using the disks */
4893	for (c = 0; c < children; c++) {
4894		if (vml[c] != NULL)
4895			vml[c]->vdev_offline = B_TRUE;
4896	}
4897	vdev_reopen(spa->spa_root_vdev);
4898
4899	/*
4900	 * Temporarily record the splitting vdevs in the spa config.  This
4901	 * will disappear once the config is regenerated.
4902	 */
4903	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4904	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4905	    glist, children) == 0);
4906	kmem_free(glist, children * sizeof (uint64_t));
4907
4908	mutex_enter(&spa->spa_props_lock);
4909	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4910	    nvl) == 0);
4911	mutex_exit(&spa->spa_props_lock);
4912	spa->spa_config_splitting = nvl;
4913	vdev_config_dirty(spa->spa_root_vdev);
4914
4915	/* configure and create the new pool */
4916	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4917	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4918	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4919	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4920	    spa_version(spa)) == 0);
4921	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4922	    spa->spa_config_txg) == 0);
4923	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4924	    spa_generate_guid(NULL)) == 0);
4925	(void) nvlist_lookup_string(props,
4926	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4927
4928	/* add the new pool to the namespace */
4929	newspa = spa_add(newname, config, altroot);
4930	newspa->spa_config_txg = spa->spa_config_txg;
4931	spa_set_log_state(newspa, SPA_LOG_CLEAR);
4932
4933	/* release the spa config lock, retaining the namespace lock */
4934	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4935
4936	if (zio_injection_enabled)
4937		zio_handle_panic_injection(spa, FTAG, 1);
4938
4939	spa_activate(newspa, spa_mode_global);
4940	spa_async_suspend(newspa);
4941
4942	/* create the new pool from the disks of the original pool */
4943	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4944	if (error)
4945		goto out;
4946
4947	/* if that worked, generate a real config for the new pool */
4948	if (newspa->spa_root_vdev != NULL) {
4949		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4950		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4951		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4952		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4953		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4954		    B_TRUE));
4955	}
4956
4957	/* set the props */
4958	if (props != NULL) {
4959		spa_configfile_set(newspa, props, B_FALSE);
4960		error = spa_prop_set(newspa, props);
4961		if (error)
4962			goto out;
4963	}
4964
4965	/* flush everything */
4966	txg = spa_vdev_config_enter(newspa);
4967	vdev_config_dirty(newspa->spa_root_vdev);
4968	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4969
4970	if (zio_injection_enabled)
4971		zio_handle_panic_injection(spa, FTAG, 2);
4972
4973	spa_async_resume(newspa);
4974
4975	/* finally, update the original pool's config */
4976	txg = spa_vdev_config_enter(spa);
4977	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4978	error = dmu_tx_assign(tx, TXG_WAIT);
4979	if (error != 0)
4980		dmu_tx_abort(tx);
4981	for (c = 0; c < children; c++) {
4982		if (vml[c] != NULL) {
4983			vdev_split(vml[c]);
4984			if (error == 0)
4985				spa_history_log_internal(spa, "detach", tx,
4986				    "vdev=%s", vml[c]->vdev_path);
4987			vdev_free(vml[c]);
4988		}
4989	}
4990	vdev_config_dirty(spa->spa_root_vdev);
4991	spa->spa_config_splitting = NULL;
4992	nvlist_free(nvl);
4993	if (error == 0)
4994		dmu_tx_commit(tx);
4995	(void) spa_vdev_exit(spa, NULL, txg, 0);
4996
4997	if (zio_injection_enabled)
4998		zio_handle_panic_injection(spa, FTAG, 3);
4999
5000	/* split is complete; log a history record */
5001	spa_history_log_internal(newspa, "split", NULL,
5002	    "from pool %s", spa_name(spa));
5003
5004	kmem_free(vml, children * sizeof (vdev_t *));
5005
5006	/* if we're not going to mount the filesystems in userland, export */
5007	if (exp)
5008		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5009		    B_FALSE, B_FALSE);
5010
5011	return (error);
5012
5013out:
5014	spa_unload(newspa);
5015	spa_deactivate(newspa);
5016	spa_remove(newspa);
5017
5018	txg = spa_vdev_config_enter(spa);
5019
5020	/* re-online all offlined disks */
5021	for (c = 0; c < children; c++) {
5022		if (vml[c] != NULL)
5023			vml[c]->vdev_offline = B_FALSE;
5024	}
5025	vdev_reopen(spa->spa_root_vdev);
5026
5027	nvlist_free(spa->spa_config_splitting);
5028	spa->spa_config_splitting = NULL;
5029	(void) spa_vdev_exit(spa, NULL, txg, error);
5030
5031	kmem_free(vml, children * sizeof (vdev_t *));
5032	return (error);
5033}
5034
5035static nvlist_t *
5036spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5037{
5038	for (int i = 0; i < count; i++) {
5039		uint64_t guid;
5040
5041		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5042		    &guid) == 0);
5043
5044		if (guid == target_guid)
5045			return (nvpp[i]);
5046	}
5047
5048	return (NULL);
5049}
5050
5051static void
5052spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5053	nvlist_t *dev_to_remove)
5054{
5055	nvlist_t **newdev = NULL;
5056
5057	if (count > 1)
5058		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5059
5060	for (int i = 0, j = 0; i < count; i++) {
5061		if (dev[i] == dev_to_remove)
5062			continue;
5063		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5064	}
5065
5066	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5067	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5068
5069	for (int i = 0; i < count - 1; i++)
5070		nvlist_free(newdev[i]);
5071
5072	if (count > 1)
5073		kmem_free(newdev, (count - 1) * sizeof (void *));
5074}
5075
5076/*
5077 * Evacuate the device.
5078 */
5079static int
5080spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5081{
5082	uint64_t txg;
5083	int error = 0;
5084
5085	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5086	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5087	ASSERT(vd == vd->vdev_top);
5088
5089	/*
5090	 * Evacuate the device.  We don't hold the config lock as writer
5091	 * since we need to do I/O but we do keep the
5092	 * spa_namespace_lock held.  Once this completes the device
5093	 * should no longer have any blocks allocated on it.
5094	 */
5095	if (vd->vdev_islog) {
5096		if (vd->vdev_stat.vs_alloc != 0)
5097			error = spa_offline_log(spa);
5098	} else {
5099		error = ENOTSUP;
5100	}
5101
5102	if (error)
5103		return (error);
5104
5105	/*
5106	 * The evacuation succeeded.  Remove any remaining MOS metadata
5107	 * associated with this vdev, and wait for these changes to sync.
5108	 */
5109	ASSERT0(vd->vdev_stat.vs_alloc);
5110	txg = spa_vdev_config_enter(spa);
5111	vd->vdev_removing = B_TRUE;
5112	vdev_dirty(vd, 0, NULL, txg);
5113	vdev_config_dirty(vd);
5114	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5115
5116	return (0);
5117}
5118
5119/*
5120 * Complete the removal by cleaning up the namespace.
5121 */
5122static void
5123spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5124{
5125	vdev_t *rvd = spa->spa_root_vdev;
5126	uint64_t id = vd->vdev_id;
5127	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5128
5129	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5130	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5131	ASSERT(vd == vd->vdev_top);
5132
5133	/*
5134	 * Only remove any devices which are empty.
5135	 */
5136	if (vd->vdev_stat.vs_alloc != 0)
5137		return;
5138
5139	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5140
5141	if (list_link_active(&vd->vdev_state_dirty_node))
5142		vdev_state_clean(vd);
5143	if (list_link_active(&vd->vdev_config_dirty_node))
5144		vdev_config_clean(vd);
5145
5146	vdev_free(vd);
5147
5148	if (last_vdev) {
5149		vdev_compact_children(rvd);
5150	} else {
5151		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5152		vdev_add_child(rvd, vd);
5153	}
5154	vdev_config_dirty(rvd);
5155
5156	/*
5157	 * Reassess the health of our root vdev.
5158	 */
5159	vdev_reopen(rvd);
5160}
5161
5162/*
5163 * Remove a device from the pool -
5164 *
5165 * Removing a device from the vdev namespace requires several steps
5166 * and can take a significant amount of time.  As a result we use
5167 * the spa_vdev_config_[enter/exit] functions which allow us to
5168 * grab and release the spa_config_lock while still holding the namespace
5169 * lock.  During each step the configuration is synced out.
5170 */
5171
5172/*
5173 * Remove a device from the pool.  Currently, this supports removing only hot
5174 * spares, slogs, and level 2 ARC devices.
5175 */
5176int
5177spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5178{
5179	vdev_t *vd;
5180	metaslab_group_t *mg;
5181	nvlist_t **spares, **l2cache, *nv;
5182	uint64_t txg = 0;
5183	uint_t nspares, nl2cache;
5184	int error = 0;
5185	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5186
5187	ASSERT(spa_writeable(spa));
5188
5189	if (!locked)
5190		txg = spa_vdev_enter(spa);
5191
5192	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5193
5194	if (spa->spa_spares.sav_vdevs != NULL &&
5195	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5196	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5197	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5198		/*
5199		 * Only remove the hot spare if it's not currently in use
5200		 * in this pool.
5201		 */
5202		if (vd == NULL || unspare) {
5203			spa_vdev_remove_aux(spa->spa_spares.sav_config,
5204			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5205			spa_load_spares(spa);
5206			spa->spa_spares.sav_sync = B_TRUE;
5207		} else {
5208			error = EBUSY;
5209		}
5210	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
5211	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5212	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5213	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5214		/*
5215		 * Cache devices can always be removed.
5216		 */
5217		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5218		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5219		spa_load_l2cache(spa);
5220		spa->spa_l2cache.sav_sync = B_TRUE;
5221	} else if (vd != NULL && vd->vdev_islog) {
5222		ASSERT(!locked);
5223		ASSERT(vd == vd->vdev_top);
5224
5225		/*
5226		 * XXX - Once we have bp-rewrite this should
5227		 * become the common case.
5228		 */
5229
5230		mg = vd->vdev_mg;
5231
5232		/*
5233		 * Stop allocating from this vdev.
5234		 */
5235		metaslab_group_passivate(mg);
5236
5237		/*
5238		 * Wait for the youngest allocations and frees to sync,
5239		 * and then wait for the deferral of those frees to finish.
5240		 */
5241		spa_vdev_config_exit(spa, NULL,
5242		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5243
5244		/*
5245		 * Attempt to evacuate the vdev.
5246		 */
5247		error = spa_vdev_remove_evacuate(spa, vd);
5248
5249		txg = spa_vdev_config_enter(spa);
5250
5251		/*
5252		 * If we couldn't evacuate the vdev, unwind.
5253		 */
5254		if (error) {
5255			metaslab_group_activate(mg);
5256			return (spa_vdev_exit(spa, NULL, txg, error));
5257		}
5258
5259		/*
5260		 * Clean up the vdev namespace.
5261		 */
5262		spa_vdev_remove_from_namespace(spa, vd);
5263
5264	} else if (vd != NULL) {
5265		/*
5266		 * Normal vdevs cannot be removed (yet).
5267		 */
5268		error = ENOTSUP;
5269	} else {
5270		/*
5271		 * There is no vdev of any kind with the specified guid.
5272		 */
5273		error = ENOENT;
5274	}
5275
5276	if (!locked)
5277		return (spa_vdev_exit(spa, NULL, txg, error));
5278
5279	return (error);
5280}
5281
5282/*
5283 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5284 * current spared, so we can detach it.
5285 */
5286static vdev_t *
5287spa_vdev_resilver_done_hunt(vdev_t *vd)
5288{
5289	vdev_t *newvd, *oldvd;
5290
5291	for (int c = 0; c < vd->vdev_children; c++) {
5292		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5293		if (oldvd != NULL)
5294			return (oldvd);
5295	}
5296
5297	/*
5298	 * Check for a completed replacement.  We always consider the first
5299	 * vdev in the list to be the oldest vdev, and the last one to be
5300	 * the newest (see spa_vdev_attach() for how that works).  In
5301	 * the case where the newest vdev is faulted, we will not automatically
5302	 * remove it after a resilver completes.  This is OK as it will require
5303	 * user intervention to determine which disk the admin wishes to keep.
5304	 */
5305	if (vd->vdev_ops == &vdev_replacing_ops) {
5306		ASSERT(vd->vdev_children > 1);
5307
5308		newvd = vd->vdev_child[vd->vdev_children - 1];
5309		oldvd = vd->vdev_child[0];
5310
5311		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5312		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5313		    !vdev_dtl_required(oldvd))
5314			return (oldvd);
5315	}
5316
5317	/*
5318	 * Check for a completed resilver with the 'unspare' flag set.
5319	 */
5320	if (vd->vdev_ops == &vdev_spare_ops) {
5321		vdev_t *first = vd->vdev_child[0];
5322		vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5323
5324		if (last->vdev_unspare) {
5325			oldvd = first;
5326			newvd = last;
5327		} else if (first->vdev_unspare) {
5328			oldvd = last;
5329			newvd = first;
5330		} else {
5331			oldvd = NULL;
5332		}
5333
5334		if (oldvd != NULL &&
5335		    vdev_dtl_empty(newvd, DTL_MISSING) &&
5336		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5337		    !vdev_dtl_required(oldvd))
5338			return (oldvd);
5339
5340		/*
5341		 * If there are more than two spares attached to a disk,
5342		 * and those spares are not required, then we want to
5343		 * attempt to free them up now so that they can be used
5344		 * by other pools.  Once we're back down to a single
5345		 * disk+spare, we stop removing them.
5346		 */
5347		if (vd->vdev_children > 2) {
5348			newvd = vd->vdev_child[1];
5349
5350			if (newvd->vdev_isspare && last->vdev_isspare &&
5351			    vdev_dtl_empty(last, DTL_MISSING) &&
5352			    vdev_dtl_empty(last, DTL_OUTAGE) &&
5353			    !vdev_dtl_required(newvd))
5354				return (newvd);
5355		}
5356	}
5357
5358	return (NULL);
5359}
5360
5361static void
5362spa_vdev_resilver_done(spa_t *spa)
5363{
5364	vdev_t *vd, *pvd, *ppvd;
5365	uint64_t guid, sguid, pguid, ppguid;
5366
5367	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5368
5369	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5370		pvd = vd->vdev_parent;
5371		ppvd = pvd->vdev_parent;
5372		guid = vd->vdev_guid;
5373		pguid = pvd->vdev_guid;
5374		ppguid = ppvd->vdev_guid;
5375		sguid = 0;
5376		/*
5377		 * If we have just finished replacing a hot spared device, then
5378		 * we need to detach the parent's first child (the original hot
5379		 * spare) as well.
5380		 */
5381		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5382		    ppvd->vdev_children == 2) {
5383			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5384			sguid = ppvd->vdev_child[1]->vdev_guid;
5385		}
5386		spa_config_exit(spa, SCL_ALL, FTAG);
5387		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5388			return;
5389		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5390			return;
5391		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5392	}
5393
5394	spa_config_exit(spa, SCL_ALL, FTAG);
5395}
5396
5397/*
5398 * Update the stored path or FRU for this vdev.
5399 */
5400int
5401spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5402    boolean_t ispath)
5403{
5404	vdev_t *vd;
5405	boolean_t sync = B_FALSE;
5406
5407	ASSERT(spa_writeable(spa));
5408
5409	spa_vdev_state_enter(spa, SCL_ALL);
5410
5411	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5412		return (spa_vdev_state_exit(spa, NULL, ENOENT));
5413
5414	if (!vd->vdev_ops->vdev_op_leaf)
5415		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5416
5417	if (ispath) {
5418		if (strcmp(value, vd->vdev_path) != 0) {
5419			spa_strfree(vd->vdev_path);
5420			vd->vdev_path = spa_strdup(value);
5421			sync = B_TRUE;
5422		}
5423	} else {
5424		if (vd->vdev_fru == NULL) {
5425			vd->vdev_fru = spa_strdup(value);
5426			sync = B_TRUE;
5427		} else if (strcmp(value, vd->vdev_fru) != 0) {
5428			spa_strfree(vd->vdev_fru);
5429			vd->vdev_fru = spa_strdup(value);
5430			sync = B_TRUE;
5431		}
5432	}
5433
5434	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5435}
5436
5437int
5438spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5439{
5440	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5441}
5442
5443int
5444spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5445{
5446	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5447}
5448
5449/*
5450 * ==========================================================================
5451 * SPA Scanning
5452 * ==========================================================================
5453 */
5454
5455int
5456spa_scan_stop(spa_t *spa)
5457{
5458	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5459	if (dsl_scan_resilvering(spa->spa_dsl_pool))
5460		return (EBUSY);
5461	return (dsl_scan_cancel(spa->spa_dsl_pool));
5462}
5463
5464int
5465spa_scan(spa_t *spa, pool_scan_func_t func)
5466{
5467	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5468
5469	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5470		return (ENOTSUP);
5471
5472	/*
5473	 * If a resilver was requested, but there is no DTL on a
5474	 * writeable leaf device, we have nothing to do.
5475	 */
5476	if (func == POOL_SCAN_RESILVER &&
5477	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5478		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5479		return (0);
5480	}
5481
5482	return (dsl_scan(spa->spa_dsl_pool, func));
5483}
5484
5485/*
5486 * ==========================================================================
5487 * SPA async task processing
5488 * ==========================================================================
5489 */
5490
5491static void
5492spa_async_remove(spa_t *spa, vdev_t *vd)
5493{
5494	if (vd->vdev_remove_wanted) {
5495		vd->vdev_remove_wanted = B_FALSE;
5496		vd->vdev_delayed_close = B_FALSE;
5497		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5498
5499		/*
5500		 * We want to clear the stats, but we don't want to do a full
5501		 * vdev_clear() as that will cause us to throw away
5502		 * degraded/faulted state as well as attempt to reopen the
5503		 * device, all of which is a waste.
5504		 */
5505		vd->vdev_stat.vs_read_errors = 0;
5506		vd->vdev_stat.vs_write_errors = 0;
5507		vd->vdev_stat.vs_checksum_errors = 0;
5508
5509		vdev_state_dirty(vd->vdev_top);
5510	}
5511
5512	for (int c = 0; c < vd->vdev_children; c++)
5513		spa_async_remove(spa, vd->vdev_child[c]);
5514}
5515
5516static void
5517spa_async_probe(spa_t *spa, vdev_t *vd)
5518{
5519	if (vd->vdev_probe_wanted) {
5520		vd->vdev_probe_wanted = B_FALSE;
5521		vdev_reopen(vd);	/* vdev_open() does the actual probe */
5522	}
5523
5524	for (int c = 0; c < vd->vdev_children; c++)
5525		spa_async_probe(spa, vd->vdev_child[c]);
5526}
5527
5528static void
5529spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5530{
5531	sysevent_id_t eid;
5532	nvlist_t *attr;
5533	char *physpath;
5534
5535	if (!spa->spa_autoexpand)
5536		return;
5537
5538	for (int c = 0; c < vd->vdev_children; c++) {
5539		vdev_t *cvd = vd->vdev_child[c];
5540		spa_async_autoexpand(spa, cvd);
5541	}
5542
5543	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5544		return;
5545
5546	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5547	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5548
5549	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5550	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5551
5552	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5553	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5554
5555	nvlist_free(attr);
5556	kmem_free(physpath, MAXPATHLEN);
5557}
5558
5559static void
5560spa_async_thread(spa_t *spa)
5561{
5562	int tasks;
5563
5564	ASSERT(spa->spa_sync_on);
5565
5566	mutex_enter(&spa->spa_async_lock);
5567	tasks = spa->spa_async_tasks;
5568	spa->spa_async_tasks = 0;
5569	mutex_exit(&spa->spa_async_lock);
5570
5571	/*
5572	 * See if the config needs to be updated.
5573	 */
5574	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5575		uint64_t old_space, new_space;
5576
5577		mutex_enter(&spa_namespace_lock);
5578		old_space = metaslab_class_get_space(spa_normal_class(spa));
5579		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5580		new_space = metaslab_class_get_space(spa_normal_class(spa));
5581		mutex_exit(&spa_namespace_lock);
5582
5583		/*
5584		 * If the pool grew as a result of the config update,
5585		 * then log an internal history event.
5586		 */
5587		if (new_space != old_space) {
5588			spa_history_log_internal(spa, "vdev online", NULL,
5589			    "pool '%s' size: %llu(+%llu)",
5590			    spa_name(spa), new_space, new_space - old_space);
5591		}
5592	}
5593
5594	/*
5595	 * See if any devices need to be marked REMOVED.
5596	 */
5597	if (tasks & SPA_ASYNC_REMOVE) {
5598		spa_vdev_state_enter(spa, SCL_NONE);
5599		spa_async_remove(spa, spa->spa_root_vdev);
5600		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5601			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5602		for (int i = 0; i < spa->spa_spares.sav_count; i++)
5603			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5604		(void) spa_vdev_state_exit(spa, NULL, 0);
5605	}
5606
5607	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5608		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5609		spa_async_autoexpand(spa, spa->spa_root_vdev);
5610		spa_config_exit(spa, SCL_CONFIG, FTAG);
5611	}
5612
5613	/*
5614	 * See if any devices need to be probed.
5615	 */
5616	if (tasks & SPA_ASYNC_PROBE) {
5617		spa_vdev_state_enter(spa, SCL_NONE);
5618		spa_async_probe(spa, spa->spa_root_vdev);
5619		(void) spa_vdev_state_exit(spa, NULL, 0);
5620	}
5621
5622	/*
5623	 * If any devices are done replacing, detach them.
5624	 */
5625	if (tasks & SPA_ASYNC_RESILVER_DONE)
5626		spa_vdev_resilver_done(spa);
5627
5628	/*
5629	 * Kick off a resilver.
5630	 */
5631	if (tasks & SPA_ASYNC_RESILVER)
5632		dsl_resilver_restart(spa->spa_dsl_pool, 0);
5633
5634	/*
5635	 * Let the world know that we're done.
5636	 */
5637	mutex_enter(&spa->spa_async_lock);
5638	spa->spa_async_thread = NULL;
5639	cv_broadcast(&spa->spa_async_cv);
5640	mutex_exit(&spa->spa_async_lock);
5641	thread_exit();
5642}
5643
5644void
5645spa_async_suspend(spa_t *spa)
5646{
5647	mutex_enter(&spa->spa_async_lock);
5648	spa->spa_async_suspended++;
5649	while (spa->spa_async_thread != NULL)
5650		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5651	mutex_exit(&spa->spa_async_lock);
5652}
5653
5654void
5655spa_async_resume(spa_t *spa)
5656{
5657	mutex_enter(&spa->spa_async_lock);
5658	ASSERT(spa->spa_async_suspended != 0);
5659	spa->spa_async_suspended--;
5660	mutex_exit(&spa->spa_async_lock);
5661}
5662
5663static void
5664spa_async_dispatch(spa_t *spa)
5665{
5666	mutex_enter(&spa->spa_async_lock);
5667	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5668	    spa->spa_async_thread == NULL &&
5669	    rootdir != NULL && !vn_is_readonly(rootdir))
5670		spa->spa_async_thread = thread_create(NULL, 0,
5671		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5672	mutex_exit(&spa->spa_async_lock);
5673}
5674
5675void
5676spa_async_request(spa_t *spa, int task)
5677{
5678	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5679	mutex_enter(&spa->spa_async_lock);
5680	spa->spa_async_tasks |= task;
5681	mutex_exit(&spa->spa_async_lock);
5682}
5683
5684/*
5685 * ==========================================================================
5686 * SPA syncing routines
5687 * ==========================================================================
5688 */
5689
5690static int
5691bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5692{
5693	bpobj_t *bpo = arg;
5694	bpobj_enqueue(bpo, bp, tx);
5695	return (0);
5696}
5697
5698static int
5699spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5700{
5701	zio_t *zio = arg;
5702
5703	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5704	    zio->io_flags));
5705	return (0);
5706}
5707
5708static void
5709spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5710{
5711	char *packed = NULL;
5712	size_t bufsize;
5713	size_t nvsize = 0;
5714	dmu_buf_t *db;
5715
5716	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5717
5718	/*
5719	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5720	 * information.  This avoids the dbuf_will_dirty() path and
5721	 * saves us a pre-read to get data we don't actually care about.
5722	 */
5723	bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5724	packed = kmem_alloc(bufsize, KM_SLEEP);
5725
5726	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5727	    KM_SLEEP) == 0);
5728	bzero(packed + nvsize, bufsize - nvsize);
5729
5730	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5731
5732	kmem_free(packed, bufsize);
5733
5734	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5735	dmu_buf_will_dirty(db, tx);
5736	*(uint64_t *)db->db_data = nvsize;
5737	dmu_buf_rele(db, FTAG);
5738}
5739
5740static void
5741spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5742    const char *config, const char *entry)
5743{
5744	nvlist_t *nvroot;
5745	nvlist_t **list;
5746	int i;
5747
5748	if (!sav->sav_sync)
5749		return;
5750
5751	/*
5752	 * Update the MOS nvlist describing the list of available devices.
5753	 * spa_validate_aux() will have already made sure this nvlist is
5754	 * valid and the vdevs are labeled appropriately.
5755	 */
5756	if (sav->sav_object == 0) {
5757		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5758		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5759		    sizeof (uint64_t), tx);
5760		VERIFY(zap_update(spa->spa_meta_objset,
5761		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5762		    &sav->sav_object, tx) == 0);
5763	}
5764
5765	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5766	if (sav->sav_count == 0) {
5767		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5768	} else {
5769		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5770		for (i = 0; i < sav->sav_count; i++)
5771			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5772			    B_FALSE, VDEV_CONFIG_L2CACHE);
5773		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5774		    sav->sav_count) == 0);
5775		for (i = 0; i < sav->sav_count; i++)
5776			nvlist_free(list[i]);
5777		kmem_free(list, sav->sav_count * sizeof (void *));
5778	}
5779
5780	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5781	nvlist_free(nvroot);
5782
5783	sav->sav_sync = B_FALSE;
5784}
5785
5786static void
5787spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5788{
5789	nvlist_t *config;
5790
5791	if (list_is_empty(&spa->spa_config_dirty_list))
5792		return;
5793
5794	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5795
5796	config = spa_config_generate(spa, spa->spa_root_vdev,
5797	    dmu_tx_get_txg(tx), B_FALSE);
5798
5799	/*
5800	 * If we're upgrading the spa version then make sure that
5801	 * the config object gets updated with the correct version.
5802	 */
5803	if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5804		fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5805		    spa->spa_uberblock.ub_version);
5806
5807	spa_config_exit(spa, SCL_STATE, FTAG);
5808
5809	if (spa->spa_config_syncing)
5810		nvlist_free(spa->spa_config_syncing);
5811	spa->spa_config_syncing = config;
5812
5813	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5814}
5815
5816static void
5817spa_sync_version(void *arg, dmu_tx_t *tx)
5818{
5819	uint64_t *versionp = arg;
5820	uint64_t version = *versionp;
5821	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5822
5823	/*
5824	 * Setting the version is special cased when first creating the pool.
5825	 */
5826	ASSERT(tx->tx_txg != TXG_INITIAL);
5827
5828	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5829	ASSERT(version >= spa_version(spa));
5830
5831	spa->spa_uberblock.ub_version = version;
5832	vdev_config_dirty(spa->spa_root_vdev);
5833	spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5834}
5835
5836/*
5837 * Set zpool properties.
5838 */
5839static void
5840spa_sync_props(void *arg, dmu_tx_t *tx)
5841{
5842	nvlist_t *nvp = arg;
5843	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5844	objset_t *mos = spa->spa_meta_objset;
5845	nvpair_t *elem = NULL;
5846
5847	mutex_enter(&spa->spa_props_lock);
5848
5849	while ((elem = nvlist_next_nvpair(nvp, elem))) {
5850		uint64_t intval;
5851		char *strval, *fname;
5852		zpool_prop_t prop;
5853		const char *propname;
5854		zprop_type_t proptype;
5855		zfeature_info_t *feature;
5856
5857		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5858		case ZPROP_INVAL:
5859			/*
5860			 * We checked this earlier in spa_prop_validate().
5861			 */
5862			ASSERT(zpool_prop_feature(nvpair_name(elem)));
5863
5864			fname = strchr(nvpair_name(elem), '@') + 1;
5865			VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
5866
5867			spa_feature_enable(spa, feature, tx);
5868			spa_history_log_internal(spa, "set", tx,
5869			    "%s=enabled", nvpair_name(elem));
5870			break;
5871
5872		case ZPOOL_PROP_VERSION:
5873			VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5874			/*
5875			 * The version is synced seperatly before other
5876			 * properties and should be correct by now.
5877			 */
5878			ASSERT3U(spa_version(spa), >=, intval);
5879			break;
5880
5881		case ZPOOL_PROP_ALTROOT:
5882			/*
5883			 * 'altroot' is a non-persistent property. It should
5884			 * have been set temporarily at creation or import time.
5885			 */
5886			ASSERT(spa->spa_root != NULL);
5887			break;
5888
5889		case ZPOOL_PROP_READONLY:
5890		case ZPOOL_PROP_CACHEFILE:
5891			/*
5892			 * 'readonly' and 'cachefile' are also non-persisitent
5893			 * properties.
5894			 */
5895			break;
5896		case ZPOOL_PROP_COMMENT:
5897			VERIFY(nvpair_value_string(elem, &strval) == 0);
5898			if (spa->spa_comment != NULL)
5899				spa_strfree(spa->spa_comment);
5900			spa->spa_comment = spa_strdup(strval);
5901			/*
5902			 * We need to dirty the configuration on all the vdevs
5903			 * so that their labels get updated.  It's unnecessary
5904			 * to do this for pool creation since the vdev's
5905			 * configuratoin has already been dirtied.
5906			 */
5907			if (tx->tx_txg != TXG_INITIAL)
5908				vdev_config_dirty(spa->spa_root_vdev);
5909			spa_history_log_internal(spa, "set", tx,
5910			    "%s=%s", nvpair_name(elem), strval);
5911			break;
5912		default:
5913			/*
5914			 * Set pool property values in the poolprops mos object.
5915			 */
5916			if (spa->spa_pool_props_object == 0) {
5917				spa->spa_pool_props_object =
5918				    zap_create_link(mos, DMU_OT_POOL_PROPS,
5919				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5920				    tx);
5921			}
5922
5923			/* normalize the property name */
5924			propname = zpool_prop_to_name(prop);
5925			proptype = zpool_prop_get_type(prop);
5926
5927			if (nvpair_type(elem) == DATA_TYPE_STRING) {
5928				ASSERT(proptype == PROP_TYPE_STRING);
5929				VERIFY(nvpair_value_string(elem, &strval) == 0);
5930				VERIFY(zap_update(mos,
5931				    spa->spa_pool_props_object, propname,
5932				    1, strlen(strval) + 1, strval, tx) == 0);
5933				spa_history_log_internal(spa, "set", tx,
5934				    "%s=%s", nvpair_name(elem), strval);
5935			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5936				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5937
5938				if (proptype == PROP_TYPE_INDEX) {
5939					const char *unused;
5940					VERIFY(zpool_prop_index_to_string(
5941					    prop, intval, &unused) == 0);
5942				}
5943				VERIFY(zap_update(mos,
5944				    spa->spa_pool_props_object, propname,
5945				    8, 1, &intval, tx) == 0);
5946				spa_history_log_internal(spa, "set", tx,
5947				    "%s=%lld", nvpair_name(elem), intval);
5948			} else {
5949				ASSERT(0); /* not allowed */
5950			}
5951
5952			switch (prop) {
5953			case ZPOOL_PROP_DELEGATION:
5954				spa->spa_delegation = intval;
5955				break;
5956			case ZPOOL_PROP_BOOTFS:
5957				spa->spa_bootfs = intval;
5958				break;
5959			case ZPOOL_PROP_FAILUREMODE:
5960				spa->spa_failmode = intval;
5961				break;
5962			case ZPOOL_PROP_AUTOEXPAND:
5963				spa->spa_autoexpand = intval;
5964				if (tx->tx_txg != TXG_INITIAL)
5965					spa_async_request(spa,
5966					    SPA_ASYNC_AUTOEXPAND);
5967				break;
5968			case ZPOOL_PROP_DEDUPDITTO:
5969				spa->spa_dedup_ditto = intval;
5970				break;
5971			default:
5972				break;
5973			}
5974		}
5975
5976	}
5977
5978	mutex_exit(&spa->spa_props_lock);
5979}
5980
5981/*
5982 * Perform one-time upgrade on-disk changes.  spa_version() does not
5983 * reflect the new version this txg, so there must be no changes this
5984 * txg to anything that the upgrade code depends on after it executes.
5985 * Therefore this must be called after dsl_pool_sync() does the sync
5986 * tasks.
5987 */
5988static void
5989spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5990{
5991	dsl_pool_t *dp = spa->spa_dsl_pool;
5992
5993	ASSERT(spa->spa_sync_pass == 1);
5994
5995	rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
5996
5997	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5998	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5999		dsl_pool_create_origin(dp, tx);
6000
6001		/* Keeping the origin open increases spa_minref */
6002		spa->spa_minref += 3;
6003	}
6004
6005	if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6006	    spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6007		dsl_pool_upgrade_clones(dp, tx);
6008	}
6009
6010	if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6011	    spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6012		dsl_pool_upgrade_dir_clones(dp, tx);
6013
6014		/* Keeping the freedir open increases spa_minref */
6015		spa->spa_minref += 3;
6016	}
6017
6018	if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6019	    spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6020		spa_feature_create_zap_objects(spa, tx);
6021	}
6022	rrw_exit(&dp->dp_config_rwlock, FTAG);
6023}
6024
6025/*
6026 * Sync the specified transaction group.  New blocks may be dirtied as
6027 * part of the process, so we iterate until it converges.
6028 */
6029void
6030spa_sync(spa_t *spa, uint64_t txg)
6031{
6032	dsl_pool_t *dp = spa->spa_dsl_pool;
6033	objset_t *mos = spa->spa_meta_objset;
6034	bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
6035	bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6036	vdev_t *rvd = spa->spa_root_vdev;
6037	vdev_t *vd;
6038	dmu_tx_t *tx;
6039	int error;
6040
6041	VERIFY(spa_writeable(spa));
6042
6043	/*
6044	 * Lock out configuration changes.
6045	 */
6046	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6047
6048	spa->spa_syncing_txg = txg;
6049	spa->spa_sync_pass = 0;
6050
6051	/*
6052	 * If there are any pending vdev state changes, convert them
6053	 * into config changes that go out with this transaction group.
6054	 */
6055	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6056	while (list_head(&spa->spa_state_dirty_list) != NULL) {
6057		/*
6058		 * We need the write lock here because, for aux vdevs,
6059		 * calling vdev_config_dirty() modifies sav_config.
6060		 * This is ugly and will become unnecessary when we
6061		 * eliminate the aux vdev wart by integrating all vdevs
6062		 * into the root vdev tree.
6063		 */
6064		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6065		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6066		while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6067			vdev_state_clean(vd);
6068			vdev_config_dirty(vd);
6069		}
6070		spa_config_exit(spa,