spa.c revision efb4a871d8fd510a833bdca610528dde5ed69e42
1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012 by Delphix. All rights reserved.
25 * Copyright 2013 Nexenta Systems, Inc.  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	    !vd->vdev_ishole) {
1561		zfs_post_autoreplace(vd->vdev_spa, vd);
1562		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1563	}
1564}
1565
1566/*
1567 * Validate the current config against the MOS config
1568 */
1569static boolean_t
1570spa_config_valid(spa_t *spa, nvlist_t *config)
1571{
1572	vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1573	nvlist_t *nv;
1574
1575	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1576
1577	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1578	VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1579
1580	ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1581
1582	/*
1583	 * If we're doing a normal import, then build up any additional
1584	 * diagnostic information about missing devices in this config.
1585	 * We'll pass this up to the user for further processing.
1586	 */
1587	if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1588		nvlist_t **child, *nv;
1589		uint64_t idx = 0;
1590
1591		child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1592		    KM_SLEEP);
1593		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1594
1595		for (int c = 0; c < rvd->vdev_children; c++) {
1596			vdev_t *tvd = rvd->vdev_child[c];
1597			vdev_t *mtvd  = mrvd->vdev_child[c];
1598
1599			if (tvd->vdev_ops == &vdev_missing_ops &&
1600			    mtvd->vdev_ops != &vdev_missing_ops &&
1601			    mtvd->vdev_islog)
1602				child[idx++] = vdev_config_generate(spa, mtvd,
1603				    B_FALSE, 0);
1604		}
1605
1606		if (idx) {
1607			VERIFY(nvlist_add_nvlist_array(nv,
1608			    ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1609			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1610			    ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1611
1612			for (int i = 0; i < idx; i++)
1613				nvlist_free(child[i]);
1614		}
1615		nvlist_free(nv);
1616		kmem_free(child, rvd->vdev_children * sizeof (char **));
1617	}
1618
1619	/*
1620	 * Compare the root vdev tree with the information we have
1621	 * from the MOS config (mrvd). Check each top-level vdev
1622	 * with the corresponding MOS config top-level (mtvd).
1623	 */
1624	for (int c = 0; c < rvd->vdev_children; c++) {
1625		vdev_t *tvd = rvd->vdev_child[c];
1626		vdev_t *mtvd  = mrvd->vdev_child[c];
1627
1628		/*
1629		 * Resolve any "missing" vdevs in the current configuration.
1630		 * If we find that the MOS config has more accurate information
1631		 * about the top-level vdev then use that vdev instead.
1632		 */
1633		if (tvd->vdev_ops == &vdev_missing_ops &&
1634		    mtvd->vdev_ops != &vdev_missing_ops) {
1635
1636			if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1637				continue;
1638
1639			/*
1640			 * Device specific actions.
1641			 */
1642			if (mtvd->vdev_islog) {
1643				spa_set_log_state(spa, SPA_LOG_CLEAR);
1644			} else {
1645				/*
1646				 * XXX - once we have 'readonly' pool
1647				 * support we should be able to handle
1648				 * missing data devices by transitioning
1649				 * the pool to readonly.
1650				 */
1651				continue;
1652			}
1653
1654			/*
1655			 * Swap the missing vdev with the data we were
1656			 * able to obtain from the MOS config.
1657			 */
1658			vdev_remove_child(rvd, tvd);
1659			vdev_remove_child(mrvd, mtvd);
1660
1661			vdev_add_child(rvd, mtvd);
1662			vdev_add_child(mrvd, tvd);
1663
1664			spa_config_exit(spa, SCL_ALL, FTAG);
1665			vdev_load(mtvd);
1666			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1667
1668			vdev_reopen(rvd);
1669		} else if (mtvd->vdev_islog) {
1670			/*
1671			 * Load the slog device's state from the MOS config
1672			 * since it's possible that the label does not
1673			 * contain the most up-to-date information.
1674			 */
1675			vdev_load_log_state(tvd, mtvd);
1676			vdev_reopen(tvd);
1677		}
1678	}
1679	vdev_free(mrvd);
1680	spa_config_exit(spa, SCL_ALL, FTAG);
1681
1682	/*
1683	 * Ensure we were able to validate the config.
1684	 */
1685	return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1686}
1687
1688/*
1689 * Check for missing log devices
1690 */
1691static boolean_t
1692spa_check_logs(spa_t *spa)
1693{
1694	boolean_t rv = B_FALSE;
1695
1696	switch (spa->spa_log_state) {
1697	case SPA_LOG_MISSING:
1698		/* need to recheck in case slog has been restored */
1699	case SPA_LOG_UNKNOWN:
1700		rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1701		    NULL, DS_FIND_CHILDREN) != 0);
1702		if (rv)
1703			spa_set_log_state(spa, SPA_LOG_MISSING);
1704		break;
1705	}
1706	return (rv);
1707}
1708
1709static boolean_t
1710spa_passivate_log(spa_t *spa)
1711{
1712	vdev_t *rvd = spa->spa_root_vdev;
1713	boolean_t slog_found = B_FALSE;
1714
1715	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1716
1717	if (!spa_has_slogs(spa))
1718		return (B_FALSE);
1719
1720	for (int c = 0; c < rvd->vdev_children; c++) {
1721		vdev_t *tvd = rvd->vdev_child[c];
1722		metaslab_group_t *mg = tvd->vdev_mg;
1723
1724		if (tvd->vdev_islog) {
1725			metaslab_group_passivate(mg);
1726			slog_found = B_TRUE;
1727		}
1728	}
1729
1730	return (slog_found);
1731}
1732
1733static void
1734spa_activate_log(spa_t *spa)
1735{
1736	vdev_t *rvd = spa->spa_root_vdev;
1737
1738	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1739
1740	for (int c = 0; c < rvd->vdev_children; c++) {
1741		vdev_t *tvd = rvd->vdev_child[c];
1742		metaslab_group_t *mg = tvd->vdev_mg;
1743
1744		if (tvd->vdev_islog)
1745			metaslab_group_activate(mg);
1746	}
1747}
1748
1749int
1750spa_offline_log(spa_t *spa)
1751{
1752	int error;
1753
1754	error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1755	    NULL, DS_FIND_CHILDREN);
1756	if (error == 0) {
1757		/*
1758		 * We successfully offlined the log device, sync out the
1759		 * current txg so that the "stubby" block can be removed
1760		 * by zil_sync().
1761		 */
1762		txg_wait_synced(spa->spa_dsl_pool, 0);
1763	}
1764	return (error);
1765}
1766
1767static void
1768spa_aux_check_removed(spa_aux_vdev_t *sav)
1769{
1770	for (int i = 0; i < sav->sav_count; i++)
1771		spa_check_removed(sav->sav_vdevs[i]);
1772}
1773
1774void
1775spa_claim_notify(zio_t *zio)
1776{
1777	spa_t *spa = zio->io_spa;
1778
1779	if (zio->io_error)
1780		return;
1781
1782	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1783	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1784		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1785	mutex_exit(&spa->spa_props_lock);
1786}
1787
1788typedef struct spa_load_error {
1789	uint64_t	sle_meta_count;
1790	uint64_t	sle_data_count;
1791} spa_load_error_t;
1792
1793static void
1794spa_load_verify_done(zio_t *zio)
1795{
1796	blkptr_t *bp = zio->io_bp;
1797	spa_load_error_t *sle = zio->io_private;
1798	dmu_object_type_t type = BP_GET_TYPE(bp);
1799	int error = zio->io_error;
1800
1801	if (error) {
1802		if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1803		    type != DMU_OT_INTENT_LOG)
1804			atomic_add_64(&sle->sle_meta_count, 1);
1805		else
1806			atomic_add_64(&sle->sle_data_count, 1);
1807	}
1808	zio_data_buf_free(zio->io_data, zio->io_size);
1809}
1810
1811/*ARGSUSED*/
1812static int
1813spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1814    const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1815{
1816	if (bp != NULL) {
1817		zio_t *rio = arg;
1818		size_t size = BP_GET_PSIZE(bp);
1819		void *data = zio_data_buf_alloc(size);
1820
1821		zio_nowait(zio_read(rio, spa, bp, data, size,
1822		    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1823		    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1824		    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1825	}
1826	return (0);
1827}
1828
1829static int
1830spa_load_verify(spa_t *spa)
1831{
1832	zio_t *rio;
1833	spa_load_error_t sle = { 0 };
1834	zpool_rewind_policy_t policy;
1835	boolean_t verify_ok = B_FALSE;
1836	int error;
1837
1838	zpool_get_rewind_policy(spa->spa_config, &policy);
1839
1840	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1841		return (0);
1842
1843	rio = zio_root(spa, NULL, &sle,
1844	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1845
1846	error = traverse_pool(spa, spa->spa_verify_min_txg,
1847	    TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1848
1849	(void) zio_wait(rio);
1850
1851	spa->spa_load_meta_errors = sle.sle_meta_count;
1852	spa->spa_load_data_errors = sle.sle_data_count;
1853
1854	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1855	    sle.sle_data_count <= policy.zrp_maxdata) {
1856		int64_t loss = 0;
1857
1858		verify_ok = B_TRUE;
1859		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1860		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1861
1862		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1863		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1864		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1865		VERIFY(nvlist_add_int64(spa->spa_load_info,
1866		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1867		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1868		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1869	} else {
1870		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1871	}
1872
1873	if (error) {
1874		if (error != ENXIO && error != EIO)
1875			error = EIO;
1876		return (error);
1877	}
1878
1879	return (verify_ok ? 0 : EIO);
1880}
1881
1882/*
1883 * Find a value in the pool props object.
1884 */
1885static void
1886spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1887{
1888	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1889	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1890}
1891
1892/*
1893 * Find a value in the pool directory object.
1894 */
1895static int
1896spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1897{
1898	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1899	    name, sizeof (uint64_t), 1, val));
1900}
1901
1902static int
1903spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1904{
1905	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1906	return (err);
1907}
1908
1909/*
1910 * Fix up config after a partly-completed split.  This is done with the
1911 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1912 * pool have that entry in their config, but only the splitting one contains
1913 * a list of all the guids of the vdevs that are being split off.
1914 *
1915 * This function determines what to do with that list: either rejoin
1916 * all the disks to the pool, or complete the splitting process.  To attempt
1917 * the rejoin, each disk that is offlined is marked online again, and
1918 * we do a reopen() call.  If the vdev label for every disk that was
1919 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1920 * then we call vdev_split() on each disk, and complete the split.
1921 *
1922 * Otherwise we leave the config alone, with all the vdevs in place in
1923 * the original pool.
1924 */
1925static void
1926spa_try_repair(spa_t *spa, nvlist_t *config)
1927{
1928	uint_t extracted;
1929	uint64_t *glist;
1930	uint_t i, gcount;
1931	nvlist_t *nvl;
1932	vdev_t **vd;
1933	boolean_t attempt_reopen;
1934
1935	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1936		return;
1937
1938	/* check that the config is complete */
1939	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1940	    &glist, &gcount) != 0)
1941		return;
1942
1943	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1944
1945	/* attempt to online all the vdevs & validate */
1946	attempt_reopen = B_TRUE;
1947	for (i = 0; i < gcount; i++) {
1948		if (glist[i] == 0)	/* vdev is hole */
1949			continue;
1950
1951		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1952		if (vd[i] == NULL) {
1953			/*
1954			 * Don't bother attempting to reopen the disks;
1955			 * just do the split.
1956			 */
1957			attempt_reopen = B_FALSE;
1958		} else {
1959			/* attempt to re-online it */
1960			vd[i]->vdev_offline = B_FALSE;
1961		}
1962	}
1963
1964	if (attempt_reopen) {
1965		vdev_reopen(spa->spa_root_vdev);
1966
1967		/* check each device to see what state it's in */
1968		for (extracted = 0, i = 0; i < gcount; i++) {
1969			if (vd[i] != NULL &&
1970			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1971				break;
1972			++extracted;
1973		}
1974	}
1975
1976	/*
1977	 * If every disk has been moved to the new pool, or if we never
1978	 * even attempted to look at them, then we split them off for
1979	 * good.
1980	 */
1981	if (!attempt_reopen || gcount == extracted) {
1982		for (i = 0; i < gcount; i++)
1983			if (vd[i] != NULL)
1984				vdev_split(vd[i]);
1985		vdev_reopen(spa->spa_root_vdev);
1986	}
1987
1988	kmem_free(vd, gcount * sizeof (vdev_t *));
1989}
1990
1991static int
1992spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1993    boolean_t mosconfig)
1994{
1995	nvlist_t *config = spa->spa_config;
1996	char *ereport = FM_EREPORT_ZFS_POOL;
1997	char *comment;
1998	int error;
1999	uint64_t pool_guid;
2000	nvlist_t *nvl;
2001
2002	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2003		return (EINVAL);
2004
2005	ASSERT(spa->spa_comment == NULL);
2006	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2007		spa->spa_comment = spa_strdup(comment);
2008
2009	/*
2010	 * Versioning wasn't explicitly added to the label until later, so if
2011	 * it's not present treat it as the initial version.
2012	 */
2013	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2014	    &spa->spa_ubsync.ub_version) != 0)
2015		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2016
2017	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2018	    &spa->spa_config_txg);
2019
2020	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2021	    spa_guid_exists(pool_guid, 0)) {
2022		error = EEXIST;
2023	} else {
2024		spa->spa_config_guid = pool_guid;
2025
2026		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2027		    &nvl) == 0) {
2028			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2029			    KM_SLEEP) == 0);
2030		}
2031
2032		nvlist_free(spa->spa_load_info);
2033		spa->spa_load_info = fnvlist_alloc();
2034
2035		gethrestime(&spa->spa_loaded_ts);
2036		error = spa_load_impl(spa, pool_guid, config, state, type,
2037		    mosconfig, &ereport);
2038	}
2039
2040	spa->spa_minref = refcount_count(&spa->spa_refcount);
2041	if (error) {
2042		if (error != EEXIST) {
2043			spa->spa_loaded_ts.tv_sec = 0;
2044			spa->spa_loaded_ts.tv_nsec = 0;
2045		}
2046		if (error != EBADF) {
2047			zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2048		}
2049	}
2050	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2051	spa->spa_ena = 0;
2052
2053	return (error);
2054}
2055
2056/*
2057 * Load an existing storage pool, using the pool's builtin spa_config as a
2058 * source of configuration information.
2059 */
2060static int
2061spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2062    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2063    char **ereport)
2064{
2065	int error = 0;
2066	nvlist_t *nvroot = NULL;
2067	nvlist_t *label;
2068	vdev_t *rvd;
2069	uberblock_t *ub = &spa->spa_uberblock;
2070	uint64_t children, config_cache_txg = spa->spa_config_txg;
2071	int orig_mode = spa->spa_mode;
2072	int parse;
2073	uint64_t obj;
2074	boolean_t missing_feat_write = B_FALSE;
2075
2076	/*
2077	 * If this is an untrusted config, access the pool in read-only mode.
2078	 * This prevents things like resilvering recently removed devices.
2079	 */
2080	if (!mosconfig)
2081		spa->spa_mode = FREAD;
2082
2083	ASSERT(MUTEX_HELD(&spa_namespace_lock));
2084
2085	spa->spa_load_state = state;
2086
2087	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2088		return (EINVAL);
2089
2090	parse = (type == SPA_IMPORT_EXISTING ?
2091	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2092
2093	/*
2094	 * Create "The Godfather" zio to hold all async IOs
2095	 */
2096	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2097	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2098
2099	/*
2100	 * Parse the configuration into a vdev tree.  We explicitly set the
2101	 * value that will be returned by spa_version() since parsing the
2102	 * configuration requires knowing the version number.
2103	 */
2104	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2105	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2106	spa_config_exit(spa, SCL_ALL, FTAG);
2107
2108	if (error != 0)
2109		return (error);
2110
2111	ASSERT(spa->spa_root_vdev == rvd);
2112
2113	if (type != SPA_IMPORT_ASSEMBLE) {
2114		ASSERT(spa_guid(spa) == pool_guid);
2115	}
2116
2117	/*
2118	 * Try to open all vdevs, loading each label in the process.
2119	 */
2120	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2121	error = vdev_open(rvd);
2122	spa_config_exit(spa, SCL_ALL, FTAG);
2123	if (error != 0)
2124		return (error);
2125
2126	/*
2127	 * We need to validate the vdev labels against the configuration that
2128	 * we have in hand, which is dependent on the setting of mosconfig. If
2129	 * mosconfig is true then we're validating the vdev labels based on
2130	 * that config.  Otherwise, we're validating against the cached config
2131	 * (zpool.cache) that was read when we loaded the zfs module, and then
2132	 * later we will recursively call spa_load() and validate against
2133	 * the vdev config.
2134	 *
2135	 * If we're assembling a new pool that's been split off from an
2136	 * existing pool, the labels haven't yet been updated so we skip
2137	 * validation for now.
2138	 */
2139	if (type != SPA_IMPORT_ASSEMBLE) {
2140		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2141		error = vdev_validate(rvd, mosconfig);
2142		spa_config_exit(spa, SCL_ALL, FTAG);
2143
2144		if (error != 0)
2145			return (error);
2146
2147		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2148			return (ENXIO);
2149	}
2150
2151	/*
2152	 * Find the best uberblock.
2153	 */
2154	vdev_uberblock_load(rvd, ub, &label);
2155
2156	/*
2157	 * If we weren't able to find a single valid uberblock, return failure.
2158	 */
2159	if (ub->ub_txg == 0) {
2160		nvlist_free(label);
2161		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2162	}
2163
2164	/*
2165	 * If the pool has an unsupported version we can't open it.
2166	 */
2167	if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2168		nvlist_free(label);
2169		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2170	}
2171
2172	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2173		nvlist_t *features;
2174
2175		/*
2176		 * If we weren't able to find what's necessary for reading the
2177		 * MOS in the label, return failure.
2178		 */
2179		if (label == NULL || nvlist_lookup_nvlist(label,
2180		    ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2181			nvlist_free(label);
2182			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2183			    ENXIO));
2184		}
2185
2186		/*
2187		 * Update our in-core representation with the definitive values
2188		 * from the label.
2189		 */
2190		nvlist_free(spa->spa_label_features);
2191		VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2192	}
2193
2194	nvlist_free(label);
2195
2196	/*
2197	 * Look through entries in the label nvlist's features_for_read. If
2198	 * there is a feature listed there which we don't understand then we
2199	 * cannot open a pool.
2200	 */
2201	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2202		nvlist_t *unsup_feat;
2203
2204		VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2205		    0);
2206
2207		for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2208		    NULL); nvp != NULL;
2209		    nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2210			if (!zfeature_is_supported(nvpair_name(nvp))) {
2211				VERIFY(nvlist_add_string(unsup_feat,
2212				    nvpair_name(nvp), "") == 0);
2213			}
2214		}
2215
2216		if (!nvlist_empty(unsup_feat)) {
2217			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2218			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2219			nvlist_free(unsup_feat);
2220			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2221			    ENOTSUP));
2222		}
2223
2224		nvlist_free(unsup_feat);
2225	}
2226
2227	/*
2228	 * If the vdev guid sum doesn't match the uberblock, we have an
2229	 * incomplete configuration.  We first check to see if the pool
2230	 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2231	 * If it is, defer the vdev_guid_sum check till later so we
2232	 * can handle missing vdevs.
2233	 */
2234	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2235	    &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2236	    rvd->vdev_guid_sum != ub->ub_guid_sum)
2237		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2238
2239	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2240		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2241		spa_try_repair(spa, config);
2242		spa_config_exit(spa, SCL_ALL, FTAG);
2243		nvlist_free(spa->spa_config_splitting);
2244		spa->spa_config_splitting = NULL;
2245	}
2246
2247	/*
2248	 * Initialize internal SPA structures.
2249	 */
2250	spa->spa_state = POOL_STATE_ACTIVE;
2251	spa->spa_ubsync = spa->spa_uberblock;
2252	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2253	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2254	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2255	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2256	spa->spa_claim_max_txg = spa->spa_first_txg;
2257	spa->spa_prev_software_version = ub->ub_software_version;
2258
2259	error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2260	if (error)
2261		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2262	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2263
2264	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2265		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2266
2267	if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2268		boolean_t missing_feat_read = B_FALSE;
2269		nvlist_t *unsup_feat, *enabled_feat;
2270
2271		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2272		    &spa->spa_feat_for_read_obj) != 0) {
2273			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2274		}
2275
2276		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2277		    &spa->spa_feat_for_write_obj) != 0) {
2278			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2279		}
2280
2281		if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2282		    &spa->spa_feat_desc_obj) != 0) {
2283			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2284		}
2285
2286		enabled_feat = fnvlist_alloc();
2287		unsup_feat = fnvlist_alloc();
2288
2289		if (!feature_is_supported(spa->spa_meta_objset,
2290		    spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2291		    unsup_feat, enabled_feat))
2292			missing_feat_read = B_TRUE;
2293
2294		if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2295			if (!feature_is_supported(spa->spa_meta_objset,
2296			    spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2297			    unsup_feat, enabled_feat)) {
2298				missing_feat_write = B_TRUE;
2299			}
2300		}
2301
2302		fnvlist_add_nvlist(spa->spa_load_info,
2303		    ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2304
2305		if (!nvlist_empty(unsup_feat)) {
2306			fnvlist_add_nvlist(spa->spa_load_info,
2307			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2308		}
2309
2310		fnvlist_free(enabled_feat);
2311		fnvlist_free(unsup_feat);
2312
2313		if (!missing_feat_read) {
2314			fnvlist_add_boolean(spa->spa_load_info,
2315			    ZPOOL_CONFIG_CAN_RDONLY);
2316		}
2317
2318		/*
2319		 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2320		 * twofold: to determine whether the pool is available for
2321		 * import in read-write mode and (if it is not) whether the
2322		 * pool is available for import in read-only mode. If the pool
2323		 * is available for import in read-write mode, it is displayed
2324		 * as available in userland; if it is not available for import
2325		 * in read-only mode, it is displayed as unavailable in
2326		 * userland. If the pool is available for import in read-only
2327		 * mode but not read-write mode, it is displayed as unavailable
2328		 * in userland with a special note that the pool is actually
2329		 * available for open in read-only mode.
2330		 *
2331		 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2332		 * missing a feature for write, we must first determine whether
2333		 * the pool can be opened read-only before returning to
2334		 * userland in order to know whether to display the
2335		 * abovementioned note.
2336		 */
2337		if (missing_feat_read || (missing_feat_write &&
2338		    spa_writeable(spa))) {
2339			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2340			    ENOTSUP));
2341		}
2342	}
2343
2344	spa->spa_is_initializing = B_TRUE;
2345	error = dsl_pool_open(spa->spa_dsl_pool);
2346	spa->spa_is_initializing = B_FALSE;
2347	if (error != 0)
2348		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2349
2350	if (!mosconfig) {
2351		uint64_t hostid;
2352		nvlist_t *policy = NULL, *nvconfig;
2353
2354		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2355			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2356
2357		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2358		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2359			char *hostname;
2360			unsigned long myhostid = 0;
2361
2362			VERIFY(nvlist_lookup_string(nvconfig,
2363			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2364
2365#ifdef	_KERNEL
2366			myhostid = zone_get_hostid(NULL);
2367#else	/* _KERNEL */
2368			/*
2369			 * We're emulating the system's hostid in userland, so
2370			 * we can't use zone_get_hostid().
2371			 */
2372			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2373#endif	/* _KERNEL */
2374			if (hostid != 0 && myhostid != 0 &&
2375			    hostid != myhostid) {
2376				nvlist_free(nvconfig);
2377				cmn_err(CE_WARN, "pool '%s' could not be "
2378				    "loaded as it was last accessed by "
2379				    "another system (host: %s hostid: 0x%lx). "
2380				    "See: http://illumos.org/msg/ZFS-8000-EY",
2381				    spa_name(spa), hostname,
2382				    (unsigned long)hostid);
2383				return (EBADF);
2384			}
2385		}
2386		if (nvlist_lookup_nvlist(spa->spa_config,
2387		    ZPOOL_REWIND_POLICY, &policy) == 0)
2388			VERIFY(nvlist_add_nvlist(nvconfig,
2389			    ZPOOL_REWIND_POLICY, policy) == 0);
2390
2391		spa_config_set(spa, nvconfig);
2392		spa_unload(spa);
2393		spa_deactivate(spa);
2394		spa_activate(spa, orig_mode);
2395
2396		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2397	}
2398
2399	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2400		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2401	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2402	if (error != 0)
2403		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2404
2405	/*
2406	 * Load the bit that tells us to use the new accounting function
2407	 * (raid-z deflation).  If we have an older pool, this will not
2408	 * be present.
2409	 */
2410	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2411	if (error != 0 && error != ENOENT)
2412		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2413
2414	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2415	    &spa->spa_creation_version);
2416	if (error != 0 && error != ENOENT)
2417		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2418
2419	/*
2420	 * Load the persistent error log.  If we have an older pool, this will
2421	 * not be present.
2422	 */
2423	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2424	if (error != 0 && error != ENOENT)
2425		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2426
2427	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2428	    &spa->spa_errlog_scrub);
2429	if (error != 0 && error != ENOENT)
2430		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2431
2432	/*
2433	 * Load the history object.  If we have an older pool, this
2434	 * will not be present.
2435	 */
2436	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2437	if (error != 0 && error != ENOENT)
2438		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2439
2440	/*
2441	 * If we're assembling the pool from the split-off vdevs of
2442	 * an existing pool, we don't want to attach the spares & cache
2443	 * devices.
2444	 */
2445
2446	/*
2447	 * Load any hot spares for this pool.
2448	 */
2449	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2450	if (error != 0 && error != ENOENT)
2451		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2452	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2453		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2454		if (load_nvlist(spa, spa->spa_spares.sav_object,
2455		    &spa->spa_spares.sav_config) != 0)
2456			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2457
2458		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2459		spa_load_spares(spa);
2460		spa_config_exit(spa, SCL_ALL, FTAG);
2461	} else if (error == 0) {
2462		spa->spa_spares.sav_sync = B_TRUE;
2463	}
2464
2465	/*
2466	 * Load any level 2 ARC devices for this pool.
2467	 */
2468	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2469	    &spa->spa_l2cache.sav_object);
2470	if (error != 0 && error != ENOENT)
2471		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2472	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2473		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2474		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2475		    &spa->spa_l2cache.sav_config) != 0)
2476			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2477
2478		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2479		spa_load_l2cache(spa);
2480		spa_config_exit(spa, SCL_ALL, FTAG);
2481	} else if (error == 0) {
2482		spa->spa_l2cache.sav_sync = B_TRUE;
2483	}
2484
2485	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2486
2487	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2488	if (error && error != ENOENT)
2489		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2490
2491	if (error == 0) {
2492		uint64_t autoreplace;
2493
2494		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2495		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2496		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2497		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2498		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2499		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2500		    &spa->spa_dedup_ditto);
2501
2502		spa->spa_autoreplace = (autoreplace != 0);
2503	}
2504
2505	/*
2506	 * If the 'autoreplace' property is set, then post a resource notifying
2507	 * the ZFS DE that it should not issue any faults for unopenable
2508	 * devices.  We also iterate over the vdevs, and post a sysevent for any
2509	 * unopenable vdevs so that the normal autoreplace handler can take
2510	 * over.
2511	 */
2512	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2513		spa_check_removed(spa->spa_root_vdev);
2514		/*
2515		 * For the import case, this is done in spa_import(), because
2516		 * at this point we're using the spare definitions from
2517		 * the MOS config, not necessarily from the userland config.
2518		 */
2519		if (state != SPA_LOAD_IMPORT) {
2520			spa_aux_check_removed(&spa->spa_spares);
2521			spa_aux_check_removed(&spa->spa_l2cache);
2522		}
2523	}
2524
2525	/*
2526	 * Load the vdev state for all toplevel vdevs.
2527	 */
2528	vdev_load(rvd);
2529
2530	/*
2531	 * Propagate the leaf DTLs we just loaded all the way up the tree.
2532	 */
2533	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2534	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2535	spa_config_exit(spa, SCL_ALL, FTAG);
2536
2537	/*
2538	 * Load the DDTs (dedup tables).
2539	 */
2540	error = ddt_load(spa);
2541	if (error != 0)
2542		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2543
2544	spa_update_dspace(spa);
2545
2546	/*
2547	 * Validate the config, using the MOS config to fill in any
2548	 * information which might be missing.  If we fail to validate
2549	 * the config then declare the pool unfit for use. If we're
2550	 * assembling a pool from a split, the log is not transferred
2551	 * over.
2552	 */
2553	if (type != SPA_IMPORT_ASSEMBLE) {
2554		nvlist_t *nvconfig;
2555
2556		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2557			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2558
2559		if (!spa_config_valid(spa, nvconfig)) {
2560			nvlist_free(nvconfig);
2561			return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2562			    ENXIO));
2563		}
2564		nvlist_free(nvconfig);
2565
2566		/*
2567		 * Now that we've validated the config, check the state of the
2568		 * root vdev.  If it can't be opened, it indicates one or
2569		 * more toplevel vdevs are faulted.
2570		 */
2571		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2572			return (ENXIO);
2573
2574		if (spa_check_logs(spa)) {
2575			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2576			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2577		}
2578	}
2579
2580	if (missing_feat_write) {
2581		ASSERT(state == SPA_LOAD_TRYIMPORT);
2582
2583		/*
2584		 * At this point, we know that we can open the pool in
2585		 * read-only mode but not read-write mode. We now have enough
2586		 * information and can return to userland.
2587		 */
2588		return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2589	}
2590
2591	/*
2592	 * We've successfully opened the pool, verify that we're ready
2593	 * to start pushing transactions.
2594	 */
2595	if (state != SPA_LOAD_TRYIMPORT) {
2596		if (error = spa_load_verify(spa))
2597			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2598			    error));
2599	}
2600
2601	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2602	    spa->spa_load_max_txg == UINT64_MAX)) {
2603		dmu_tx_t *tx;
2604		int need_update = B_FALSE;
2605
2606		ASSERT(state != SPA_LOAD_TRYIMPORT);
2607
2608		/*
2609		 * Claim log blocks that haven't been committed yet.
2610		 * This must all happen in a single txg.
2611		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2612		 * invoked from zil_claim_log_block()'s i/o done callback.
2613		 * Price of rollback is that we abandon the log.
2614		 */
2615		spa->spa_claiming = B_TRUE;
2616
2617		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2618		    spa_first_txg(spa));
2619		(void) dmu_objset_find(spa_name(spa),
2620		    zil_claim, tx, DS_FIND_CHILDREN);
2621		dmu_tx_commit(tx);
2622
2623		spa->spa_claiming = B_FALSE;
2624
2625		spa_set_log_state(spa, SPA_LOG_GOOD);
2626		spa->spa_sync_on = B_TRUE;
2627		txg_sync_start(spa->spa_dsl_pool);
2628
2629		/*
2630		 * Wait for all claims to sync.  We sync up to the highest
2631		 * claimed log block birth time so that claimed log blocks
2632		 * don't appear to be from the future.  spa_claim_max_txg
2633		 * will have been set for us by either zil_check_log_chain()
2634		 * (invoked from spa_check_logs()) or zil_claim() above.
2635		 */
2636		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2637
2638		/*
2639		 * If the config cache is stale, or we have uninitialized
2640		 * metaslabs (see spa_vdev_add()), then update the config.
2641		 *
2642		 * If this is a verbatim import, trust the current
2643		 * in-core spa_config and update the disk labels.
2644		 */
2645		if (config_cache_txg != spa->spa_config_txg ||
2646		    state == SPA_LOAD_IMPORT ||
2647		    state == SPA_LOAD_RECOVER ||
2648		    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2649			need_update = B_TRUE;
2650
2651		for (int c = 0; c < rvd->vdev_children; c++)
2652			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2653				need_update = B_TRUE;
2654
2655		/*
2656		 * Update the config cache asychronously in case we're the
2657		 * root pool, in which case the config cache isn't writable yet.
2658		 */
2659		if (need_update)
2660			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2661
2662		/*
2663		 * Check all DTLs to see if anything needs resilvering.
2664		 */
2665		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2666		    vdev_resilver_needed(rvd, NULL, NULL))
2667			spa_async_request(spa, SPA_ASYNC_RESILVER);
2668
2669		/*
2670		 * Log the fact that we booted up (so that we can detect if
2671		 * we rebooted in the middle of an operation).
2672		 */
2673		spa_history_log_version(spa, "open");
2674
2675		/*
2676		 * Delete any inconsistent datasets.
2677		 */
2678		(void) dmu_objset_find(spa_name(spa),
2679		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2680
2681		/*
2682		 * Clean up any stale temporary dataset userrefs.
2683		 */
2684		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2685	}
2686
2687	return (0);
2688}
2689
2690static int
2691spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2692{
2693	int mode = spa->spa_mode;
2694
2695	spa_unload(spa);
2696	spa_deactivate(spa);
2697
2698	spa->spa_load_max_txg--;
2699
2700	spa_activate(spa, mode);
2701	spa_async_suspend(spa);
2702
2703	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2704}
2705
2706/*
2707 * If spa_load() fails this function will try loading prior txg's. If
2708 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2709 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2710 * function will not rewind the pool and will return the same error as
2711 * spa_load().
2712 */
2713static int
2714spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2715    uint64_t max_request, int rewind_flags)
2716{
2717	nvlist_t *loadinfo = NULL;
2718	nvlist_t *config = NULL;
2719	int load_error, rewind_error;
2720	uint64_t safe_rewind_txg;
2721	uint64_t min_txg;
2722
2723	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2724		spa->spa_load_max_txg = spa->spa_load_txg;
2725		spa_set_log_state(spa, SPA_LOG_CLEAR);
2726	} else {
2727		spa->spa_load_max_txg = max_request;
2728	}
2729
2730	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2731	    mosconfig);
2732	if (load_error == 0)
2733		return (0);
2734
2735	if (spa->spa_root_vdev != NULL)
2736		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2737
2738	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2739	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2740
2741	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2742		nvlist_free(config);
2743		return (load_error);
2744	}
2745
2746	if (state == SPA_LOAD_RECOVER) {
2747		/* Price of rolling back is discarding txgs, including log */
2748		spa_set_log_state(spa, SPA_LOG_CLEAR);
2749	} else {
2750		/*
2751		 * If we aren't rolling back save the load info from our first
2752		 * import attempt so that we can restore it after attempting
2753		 * to rewind.
2754		 */
2755		loadinfo = spa->spa_load_info;
2756		spa->spa_load_info = fnvlist_alloc();
2757	}
2758
2759	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2760	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2761	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2762	    TXG_INITIAL : safe_rewind_txg;
2763
2764	/*
2765	 * Continue as long as we're finding errors, we're still within
2766	 * the acceptable rewind range, and we're still finding uberblocks
2767	 */
2768	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2769	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2770		if (spa->spa_load_max_txg < safe_rewind_txg)
2771			spa->spa_extreme_rewind = B_TRUE;
2772		rewind_error = spa_load_retry(spa, state, mosconfig);
2773	}
2774
2775	spa->spa_extreme_rewind = B_FALSE;
2776	spa->spa_load_max_txg = UINT64_MAX;
2777
2778	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2779		spa_config_set(spa, config);
2780
2781	if (state == SPA_LOAD_RECOVER) {
2782		ASSERT3P(loadinfo, ==, NULL);
2783		return (rewind_error);
2784	} else {
2785		/* Store the rewind info as part of the initial load info */
2786		fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2787		    spa->spa_load_info);
2788
2789		/* Restore the initial load info */
2790		fnvlist_free(spa->spa_load_info);
2791		spa->spa_load_info = loadinfo;
2792
2793		return (load_error);
2794	}
2795}
2796
2797/*
2798 * Pool Open/Import
2799 *
2800 * The import case is identical to an open except that the configuration is sent
2801 * down from userland, instead of grabbed from the configuration cache.  For the
2802 * case of an open, the pool configuration will exist in the
2803 * POOL_STATE_UNINITIALIZED state.
2804 *
2805 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2806 * the same time open the pool, without having to keep around the spa_t in some
2807 * ambiguous state.
2808 */
2809static int
2810spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2811    nvlist_t **config)
2812{
2813	spa_t *spa;
2814	spa_load_state_t state = SPA_LOAD_OPEN;
2815	int error;
2816	int locked = B_FALSE;
2817
2818	*spapp = NULL;
2819
2820	/*
2821	 * As disgusting as this is, we need to support recursive calls to this
2822	 * function because dsl_dir_open() is called during spa_load(), and ends
2823	 * up calling spa_open() again.  The real fix is to figure out how to
2824	 * avoid dsl_dir_open() calling this in the first place.
2825	 */
2826	if (mutex_owner(&spa_namespace_lock) != curthread) {
2827		mutex_enter(&spa_namespace_lock);
2828		locked = B_TRUE;
2829	}
2830
2831	if ((spa = spa_lookup(pool)) == NULL) {
2832		if (locked)
2833			mutex_exit(&spa_namespace_lock);
2834		return (ENOENT);
2835	}
2836
2837	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2838		zpool_rewind_policy_t policy;
2839
2840		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2841		    &policy);
2842		if (policy.zrp_request & ZPOOL_DO_REWIND)
2843			state = SPA_LOAD_RECOVER;
2844
2845		spa_activate(spa, spa_mode_global);
2846
2847		if (state != SPA_LOAD_RECOVER)
2848			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2849
2850		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2851		    policy.zrp_request);
2852
2853		if (error == EBADF) {
2854			/*
2855			 * If vdev_validate() returns failure (indicated by
2856			 * EBADF), it indicates that one of the vdevs indicates
2857			 * that the pool has been exported or destroyed.  If
2858			 * this is the case, the config cache is out of sync and
2859			 * we should remove the pool from the namespace.
2860			 */
2861			spa_unload(spa);
2862			spa_deactivate(spa);
2863			spa_config_sync(spa, B_TRUE, B_TRUE);
2864			spa_remove(spa);
2865			if (locked)
2866				mutex_exit(&spa_namespace_lock);
2867			return (ENOENT);
2868		}
2869
2870		if (error) {
2871			/*
2872			 * We can't open the pool, but we still have useful
2873			 * information: the state of each vdev after the
2874			 * attempted vdev_open().  Return this to the user.
2875			 */
2876			if (config != NULL && spa->spa_config) {
2877				VERIFY(nvlist_dup(spa->spa_config, config,
2878				    KM_SLEEP) == 0);
2879				VERIFY(nvlist_add_nvlist(*config,
2880				    ZPOOL_CONFIG_LOAD_INFO,
2881				    spa->spa_load_info) == 0);
2882			}
2883			spa_unload(spa);
2884			spa_deactivate(spa);
2885			spa->spa_last_open_failed = error;
2886			if (locked)
2887				mutex_exit(&spa_namespace_lock);
2888			*spapp = NULL;
2889			return (error);
2890		}
2891	}
2892
2893	spa_open_ref(spa, tag);
2894
2895	if (config != NULL)
2896		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2897
2898	/*
2899	 * If we've recovered the pool, pass back any information we
2900	 * gathered while doing the load.
2901	 */
2902	if (state == SPA_LOAD_RECOVER) {
2903		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2904		    spa->spa_load_info) == 0);
2905	}
2906
2907	if (locked) {
2908		spa->spa_last_open_failed = 0;
2909		spa->spa_last_ubsync_txg = 0;
2910		spa->spa_load_txg = 0;
2911		mutex_exit(&spa_namespace_lock);
2912	}
2913
2914	*spapp = spa;
2915
2916	return (0);
2917}
2918
2919int
2920spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2921    nvlist_t **config)
2922{
2923	return (spa_open_common(name, spapp, tag, policy, config));
2924}
2925
2926int
2927spa_open(const char *name, spa_t **spapp, void *tag)
2928{
2929	return (spa_open_common(name, spapp, tag, NULL, NULL));
2930}
2931
2932/*
2933 * Lookup the given spa_t, incrementing the inject count in the process,
2934 * preventing it from being exported or destroyed.
2935 */
2936spa_t *
2937spa_inject_addref(char *name)
2938{
2939	spa_t *spa;
2940
2941	mutex_enter(&spa_namespace_lock);
2942	if ((spa = spa_lookup(name)) == NULL) {
2943		mutex_exit(&spa_namespace_lock);
2944		return (NULL);
2945	}
2946	spa->spa_inject_ref++;
2947	mutex_exit(&spa_namespace_lock);
2948
2949	return (spa);
2950}
2951
2952void
2953spa_inject_delref(spa_t *spa)
2954{
2955	mutex_enter(&spa_namespace_lock);
2956	spa->spa_inject_ref--;
2957	mutex_exit(&spa_namespace_lock);
2958}
2959
2960/*
2961 * Add spares device information to the nvlist.
2962 */
2963static void
2964spa_add_spares(spa_t *spa, nvlist_t *config)
2965{
2966	nvlist_t **spares;
2967	uint_t i, nspares;
2968	nvlist_t *nvroot;
2969	uint64_t guid;
2970	vdev_stat_t *vs;
2971	uint_t vsc;
2972	uint64_t pool;
2973
2974	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2975
2976	if (spa->spa_spares.sav_count == 0)
2977		return;
2978
2979	VERIFY(nvlist_lookup_nvlist(config,
2980	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2981	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2982	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2983	if (nspares != 0) {
2984		VERIFY(nvlist_add_nvlist_array(nvroot,
2985		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2986		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2987		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2988
2989		/*
2990		 * Go through and find any spares which have since been
2991		 * repurposed as an active spare.  If this is the case, update
2992		 * their status appropriately.
2993		 */
2994		for (i = 0; i < nspares; i++) {
2995			VERIFY(nvlist_lookup_uint64(spares[i],
2996			    ZPOOL_CONFIG_GUID, &guid) == 0);
2997			if (spa_spare_exists(guid, &pool, NULL) &&
2998			    pool != 0ULL) {
2999				VERIFY(nvlist_lookup_uint64_array(
3000				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
3001				    (uint64_t **)&vs, &vsc) == 0);
3002				vs->vs_state = VDEV_STATE_CANT_OPEN;
3003				vs->vs_aux = VDEV_AUX_SPARED;
3004			}
3005		}
3006	}
3007}
3008
3009/*
3010 * Add l2cache device information to the nvlist, including vdev stats.
3011 */
3012static void
3013spa_add_l2cache(spa_t *spa, nvlist_t *config)
3014{
3015	nvlist_t **l2cache;
3016	uint_t i, j, nl2cache;
3017	nvlist_t *nvroot;
3018	uint64_t guid;
3019	vdev_t *vd;
3020	vdev_stat_t *vs;
3021	uint_t vsc;
3022
3023	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3024
3025	if (spa->spa_l2cache.sav_count == 0)
3026		return;
3027
3028	VERIFY(nvlist_lookup_nvlist(config,
3029	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3030	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3031	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3032	if (nl2cache != 0) {
3033		VERIFY(nvlist_add_nvlist_array(nvroot,
3034		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3035		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3036		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3037
3038		/*
3039		 * Update level 2 cache device stats.
3040		 */
3041
3042		for (i = 0; i < nl2cache; i++) {
3043			VERIFY(nvlist_lookup_uint64(l2cache[i],
3044			    ZPOOL_CONFIG_GUID, &guid) == 0);
3045
3046			vd = NULL;
3047			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3048				if (guid ==
3049				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3050					vd = spa->spa_l2cache.sav_vdevs[j];
3051					break;
3052				}
3053			}
3054			ASSERT(vd != NULL);
3055
3056			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3057			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3058			    == 0);
3059			vdev_get_stats(vd, vs);
3060		}
3061	}
3062}
3063
3064static void
3065spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3066{
3067	nvlist_t *features;
3068	zap_cursor_t zc;
3069	zap_attribute_t za;
3070
3071	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3072	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3073
3074	if (spa->spa_feat_for_read_obj != 0) {
3075		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3076		    spa->spa_feat_for_read_obj);
3077		    zap_cursor_retrieve(&zc, &za) == 0;
3078		    zap_cursor_advance(&zc)) {
3079			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3080			    za.za_num_integers == 1);
3081			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3082			    za.za_first_integer));
3083		}
3084		zap_cursor_fini(&zc);
3085	}
3086
3087	if (spa->spa_feat_for_write_obj != 0) {
3088		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3089		    spa->spa_feat_for_write_obj);
3090		    zap_cursor_retrieve(&zc, &za) == 0;
3091		    zap_cursor_advance(&zc)) {
3092			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3093			    za.za_num_integers == 1);
3094			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3095			    za.za_first_integer));
3096		}
3097		zap_cursor_fini(&zc);
3098	}
3099
3100	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3101	    features) == 0);
3102	nvlist_free(features);
3103}
3104
3105int
3106spa_get_stats(const char *name, nvlist_t **config,
3107    char *altroot, size_t buflen)
3108{
3109	int error;
3110	spa_t *spa;
3111
3112	*config = NULL;
3113	error = spa_open_common(name, &spa, FTAG, NULL, config);
3114
3115	if (spa != NULL) {
3116		/*
3117		 * This still leaves a window of inconsistency where the spares
3118		 * or l2cache devices could change and the config would be
3119		 * self-inconsistent.
3120		 */
3121		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3122
3123		if (*config != NULL) {
3124			uint64_t loadtimes[2];
3125
3126			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3127			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3128			VERIFY(nvlist_add_uint64_array(*config,
3129			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3130
3131			VERIFY(nvlist_add_uint64(*config,
3132			    ZPOOL_CONFIG_ERRCOUNT,
3133			    spa_get_errlog_size(spa)) == 0);
3134
3135			if (spa_suspended(spa))
3136				VERIFY(nvlist_add_uint64(*config,
3137				    ZPOOL_CONFIG_SUSPENDED,
3138				    spa->spa_failmode) == 0);
3139
3140			spa_add_spares(spa, *config);
3141			spa_add_l2cache(spa, *config);
3142			spa_add_feature_stats(spa, *config);
3143		}
3144	}
3145
3146	/*
3147	 * We want to get the alternate root even for faulted pools, so we cheat
3148	 * and call spa_lookup() directly.
3149	 */
3150	if (altroot) {
3151		if (spa == NULL) {
3152			mutex_enter(&spa_namespace_lock);
3153			spa = spa_lookup(name);
3154			if (spa)
3155				spa_altroot(spa, altroot, buflen);
3156			else
3157				altroot[0] = '\0';
3158			spa = NULL;
3159			mutex_exit(&spa_namespace_lock);
3160		} else {
3161			spa_altroot(spa, altroot, buflen);
3162		}
3163	}
3164
3165	if (spa != NULL) {
3166		spa_config_exit(spa, SCL_CONFIG, FTAG);
3167		spa_close(spa, FTAG);
3168	}
3169
3170	return (error);
3171}
3172
3173/*
3174 * Validate that the auxiliary device array is well formed.  We must have an
3175 * array of nvlists, each which describes a valid leaf vdev.  If this is an
3176 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3177 * specified, as long as they are well-formed.
3178 */
3179static int
3180spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3181    spa_aux_vdev_t *sav, const char *config, uint64_t version,
3182    vdev_labeltype_t label)
3183{
3184	nvlist_t **dev;
3185	uint_t i, ndev;
3186	vdev_t *vd;
3187	int error;
3188
3189	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3190
3191	/*
3192	 * It's acceptable to have no devs specified.
3193	 */
3194	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3195		return (0);
3196
3197	if (ndev == 0)
3198		return (EINVAL);
3199
3200	/*
3201	 * Make sure the pool is formatted with a version that supports this
3202	 * device type.
3203	 */
3204	if (spa_version(spa) < version)
3205		return (ENOTSUP);
3206
3207	/*
3208	 * Set the pending device list so we correctly handle device in-use
3209	 * checking.
3210	 */
3211	sav->sav_pending = dev;
3212	sav->sav_npending = ndev;
3213
3214	for (i = 0; i < ndev; i++) {
3215		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3216		    mode)) != 0)
3217			goto out;
3218
3219		if (!vd->vdev_ops->vdev_op_leaf) {
3220			vdev_free(vd);
3221			error = EINVAL;
3222			goto out;
3223		}
3224
3225		/*
3226		 * The L2ARC currently only supports disk devices in
3227		 * kernel context.  For user-level testing, we allow it.
3228		 */
3229#ifdef _KERNEL
3230		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3231		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3232			error = ENOTBLK;
3233			vdev_free(vd);
3234			goto out;
3235		}
3236#endif
3237		vd->vdev_top = vd;
3238
3239		if ((error = vdev_open(vd)) == 0 &&
3240		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
3241			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3242			    vd->vdev_guid) == 0);
3243		}
3244
3245		vdev_free(vd);
3246
3247		if (error &&
3248		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3249			goto out;
3250		else
3251			error = 0;
3252	}
3253
3254out:
3255	sav->sav_pending = NULL;
3256	sav->sav_npending = 0;
3257	return (error);
3258}
3259
3260static int
3261spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3262{
3263	int error;
3264
3265	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3266
3267	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3268	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3269	    VDEV_LABEL_SPARE)) != 0) {
3270		return (error);
3271	}
3272
3273	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3274	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3275	    VDEV_LABEL_L2CACHE));
3276}
3277
3278static void
3279spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3280    const char *config)
3281{
3282	int i;
3283
3284	if (sav->sav_config != NULL) {
3285		nvlist_t **olddevs;
3286		uint_t oldndevs;
3287		nvlist_t **newdevs;
3288
3289		/*
3290		 * Generate new dev list by concatentating with the
3291		 * current dev list.
3292		 */
3293		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3294		    &olddevs, &oldndevs) == 0);
3295
3296		newdevs = kmem_alloc(sizeof (void *) *
3297		    (ndevs + oldndevs), KM_SLEEP);
3298		for (i = 0; i < oldndevs; i++)
3299			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3300			    KM_SLEEP) == 0);
3301		for (i = 0; i < ndevs; i++)
3302			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3303			    KM_SLEEP) == 0);
3304
3305		VERIFY(nvlist_remove(sav->sav_config, config,
3306		    DATA_TYPE_NVLIST_ARRAY) == 0);
3307
3308		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3309		    config, newdevs, ndevs + oldndevs) == 0);
3310		for (i = 0; i < oldndevs + ndevs; i++)
3311			nvlist_free(newdevs[i]);
3312		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3313	} else {
3314		/*
3315		 * Generate a new dev list.
3316		 */
3317		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3318		    KM_SLEEP) == 0);
3319		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3320		    devs, ndevs) == 0);
3321	}
3322}
3323
3324/*
3325 * Stop and drop level 2 ARC devices
3326 */
3327void
3328spa_l2cache_drop(spa_t *spa)
3329{
3330	vdev_t *vd;
3331	int i;
3332	spa_aux_vdev_t *sav = &spa->spa_l2cache;
3333
3334	for (i = 0; i < sav->sav_count; i++) {
3335		uint64_t pool;
3336
3337		vd = sav->sav_vdevs[i];
3338		ASSERT(vd != NULL);
3339
3340		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3341		    pool != 0ULL && l2arc_vdev_present(vd))
3342			l2arc_remove_vdev(vd);
3343	}
3344}
3345
3346/*
3347 * Pool Creation
3348 */
3349int
3350spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3351    nvlist_t *zplprops)
3352{
3353	spa_t *spa;
3354	char *altroot = NULL;
3355	vdev_t *rvd;
3356	dsl_pool_t *dp;
3357	dmu_tx_t *tx;
3358	int error = 0;
3359	uint64_t txg = TXG_INITIAL;
3360	nvlist_t **spares, **l2cache;
3361	uint_t nspares, nl2cache;
3362	uint64_t version, obj;
3363	boolean_t has_features;
3364
3365	/*
3366	 * If this pool already exists, return failure.
3367	 */
3368	mutex_enter(&spa_namespace_lock);
3369	if (spa_lookup(pool) != NULL) {
3370		mutex_exit(&spa_namespace_lock);
3371		return (EEXIST);
3372	}
3373
3374	/*
3375	 * Allocate a new spa_t structure.
3376	 */
3377	(void) nvlist_lookup_string(props,
3378	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3379	spa = spa_add(pool, NULL, altroot);
3380	spa_activate(spa, spa_mode_global);
3381
3382	if (props && (error = spa_prop_validate(spa, props))) {
3383		spa_deactivate(spa);
3384		spa_remove(spa);
3385		mutex_exit(&spa_namespace_lock);
3386		return (error);
3387	}
3388
3389	has_features = B_FALSE;
3390	for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3391	    elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3392		if (zpool_prop_feature(nvpair_name(elem)))
3393			has_features = B_TRUE;
3394	}
3395
3396	if (has_features || nvlist_lookup_uint64(props,
3397	    zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3398		version = SPA_VERSION;
3399	}
3400	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3401
3402	spa->spa_first_txg = txg;
3403	spa->spa_uberblock.ub_txg = txg - 1;
3404	spa->spa_uberblock.ub_version = version;
3405	spa->spa_ubsync = spa->spa_uberblock;
3406
3407	/*
3408	 * Create "The Godfather" zio to hold all async IOs
3409	 */
3410	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3411	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3412
3413	/*
3414	 * Create the root vdev.
3415	 */
3416	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3417
3418	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3419
3420	ASSERT(error != 0 || rvd != NULL);
3421	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3422
3423	if (error == 0 && !zfs_allocatable_devs(nvroot))
3424		error = EINVAL;
3425
3426	if (error == 0 &&
3427	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3428	    (error = spa_validate_aux(spa, nvroot, txg,
3429	    VDEV_ALLOC_ADD)) == 0) {
3430		for (int c = 0; c < rvd->vdev_children; c++) {
3431			vdev_metaslab_set_size(rvd->vdev_child[c]);
3432			vdev_expand(rvd->vdev_child[c], txg);
3433		}
3434	}
3435
3436	spa_config_exit(spa, SCL_ALL, FTAG);
3437
3438	if (error != 0) {
3439		spa_unload(spa);
3440		spa_deactivate(spa);
3441		spa_remove(spa);
3442		mutex_exit(&spa_namespace_lock);
3443		return (error);
3444	}
3445
3446	/*
3447	 * Get the list of spares, if specified.
3448	 */
3449	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3450	    &spares, &nspares) == 0) {
3451		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3452		    KM_SLEEP) == 0);
3453		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3454		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3455		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3456		spa_load_spares(spa);
3457		spa_config_exit(spa, SCL_ALL, FTAG);
3458		spa->spa_spares.sav_sync = B_TRUE;
3459	}
3460
3461	/*
3462	 * Get the list of level 2 cache devices, if specified.
3463	 */
3464	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3465	    &l2cache, &nl2cache) == 0) {
3466		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3467		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3468		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3469		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3470		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3471		spa_load_l2cache(spa);
3472		spa_config_exit(spa, SCL_ALL, FTAG);
3473		spa->spa_l2cache.sav_sync = B_TRUE;
3474	}
3475
3476	spa->spa_is_initializing = B_TRUE;
3477	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3478	spa->spa_meta_objset = dp->dp_meta_objset;
3479	spa->spa_is_initializing = B_FALSE;
3480
3481	/*
3482	 * Create DDTs (dedup tables).
3483	 */
3484	ddt_create(spa);
3485
3486	spa_update_dspace(spa);
3487
3488	tx = dmu_tx_create_assigned(dp, txg);
3489
3490	/*
3491	 * Create the pool config object.
3492	 */
3493	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3494	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3495	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3496
3497	if (zap_add(spa->spa_meta_objset,
3498	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3499	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3500		cmn_err(CE_PANIC, "failed to add pool config");
3501	}
3502
3503	if (spa_version(spa) >= SPA_VERSION_FEATURES)
3504		spa_feature_create_zap_objects(spa, tx);
3505
3506	if (zap_add(spa->spa_meta_objset,
3507	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3508	    sizeof (uint64_t), 1, &version, tx) != 0) {
3509		cmn_err(CE_PANIC, "failed to add pool version");
3510	}
3511
3512	/* Newly created pools with the right version are always deflated. */
3513	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3514		spa->spa_deflate = TRUE;
3515		if (zap_add(spa->spa_meta_objset,
3516		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3517		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3518			cmn_err(CE_PANIC, "failed to add deflate");
3519		}
3520	}
3521
3522	/*
3523	 * Create the deferred-free bpobj.  Turn off compression
3524	 * because sync-to-convergence takes longer if the blocksize
3525	 * keeps changing.
3526	 */
3527	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3528	dmu_object_set_compress(spa->spa_meta_objset, obj,
3529	    ZIO_COMPRESS_OFF, tx);
3530	if (zap_add(spa->spa_meta_objset,
3531	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3532	    sizeof (uint64_t), 1, &obj, tx) != 0) {
3533		cmn_err(CE_PANIC, "failed to add bpobj");
3534	}
3535	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3536	    spa->spa_meta_objset, obj));
3537
3538	/*
3539	 * Create the pool's history object.
3540	 */
3541	if (version >= SPA_VERSION_ZPOOL_HISTORY)
3542		spa_history_create_obj(spa, tx);
3543
3544	/*
3545	 * Set pool properties.
3546	 */
3547	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3548	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3549	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3550	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3551
3552	if (props != NULL) {
3553		spa_configfile_set(spa, props, B_FALSE);
3554		spa_sync_props(props, tx);
3555	}
3556
3557	dmu_tx_commit(tx);
3558
3559	spa->spa_sync_on = B_TRUE;
3560	txg_sync_start(spa->spa_dsl_pool);
3561
3562	/*
3563	 * We explicitly wait for the first transaction to complete so that our
3564	 * bean counters are appropriately updated.
3565	 */
3566	txg_wait_synced(spa->spa_dsl_pool, txg);
3567
3568	spa_config_sync(spa, B_FALSE, B_TRUE);
3569
3570	spa_history_log_version(spa, "create");
3571
3572	spa->spa_minref = refcount_count(&spa->spa_refcount);
3573
3574	mutex_exit(&spa_namespace_lock);
3575
3576	return (0);
3577}
3578
3579#ifdef _KERNEL
3580/*
3581 * Get the root pool information from the root disk, then import the root pool
3582 * during the system boot up time.
3583 */
3584extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3585
3586static nvlist_t *
3587spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3588{
3589	nvlist_t *config;
3590	nvlist_t *nvtop, *nvroot;
3591	uint64_t pgid;
3592
3593	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3594		return (NULL);
3595
3596	/*
3597	 * Add this top-level vdev to the child array.
3598	 */
3599	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3600	    &nvtop) == 0);
3601	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3602	    &pgid) == 0);
3603	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3604
3605	/*
3606	 * Put this pool's top-level vdevs into a root vdev.
3607	 */
3608	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3609	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3610	    VDEV_TYPE_ROOT) == 0);
3611	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3612	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3613	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3614	    &nvtop, 1) == 0);
3615
3616	/*
3617	 * Replace the existing vdev_tree with the new root vdev in
3618	 * this pool's configuration (remove the old, add the new).
3619	 */
3620	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3621	nvlist_free(nvroot);
3622	return (config);
3623}
3624
3625/*
3626 * Walk the vdev tree and see if we can find a device with "better"
3627 * configuration. A configuration is "better" if the label on that
3628 * device has a more recent txg.
3629 */
3630static void
3631spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3632{
3633	for (int c = 0; c < vd->vdev_children; c++)
3634		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3635
3636	if (vd->vdev_ops->vdev_op_leaf) {
3637		nvlist_t *label;
3638		uint64_t label_txg;
3639
3640		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3641		    &label) != 0)
3642			return;
3643
3644		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3645		    &label_txg) == 0);
3646
3647		/*
3648		 * Do we have a better boot device?
3649		 */
3650		if (label_txg > *txg) {
3651			*txg = label_txg;
3652			*avd = vd;
3653		}
3654		nvlist_free(label);
3655	}
3656}
3657
3658/*
3659 * Import a root pool.
3660 *
3661 * For x86. devpath_list will consist of devid and/or physpath name of
3662 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3663 * The GRUB "findroot" command will return the vdev we should boot.
3664 *
3665 * For Sparc, devpath_list consists the physpath name of the booting device
3666 * no matter the rootpool is a single device pool or a mirrored pool.
3667 * e.g.
3668 *	"/pci@1f,0/ide@d/disk@0,0:a"
3669 */
3670int
3671spa_import_rootpool(char *devpath, char *devid)
3672{
3673	spa_t *spa;
3674	vdev_t *rvd, *bvd, *avd = NULL;
3675	nvlist_t *config, *nvtop;
3676	uint64_t guid, txg;
3677	char *pname;
3678	int error;
3679
3680	/*
3681	 * Read the label from the boot device and generate a configuration.
3682	 */
3683	config = spa_generate_rootconf(devpath, devid, &guid);
3684#if defined(_OBP) && defined(_KERNEL)
3685	if (config == NULL) {
3686		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3687			/* iscsi boot */
3688			get_iscsi_bootpath_phy(devpath);
3689			config = spa_generate_rootconf(devpath, devid, &guid);
3690		}
3691	}
3692#endif
3693	if (config == NULL) {
3694		cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3695		    devpath);
3696		return (EIO);
3697	}
3698
3699	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3700	    &pname) == 0);
3701	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3702
3703	mutex_enter(&spa_namespace_lock);
3704	if ((spa = spa_lookup(pname)) != NULL) {
3705		/*
3706		 * Remove the existing root pool from the namespace so that we
3707		 * can replace it with the correct config we just read in.
3708		 */
3709		spa_remove(spa);
3710	}
3711
3712	spa = spa_add(pname, config, NULL);
3713	spa->spa_is_root = B_TRUE;
3714	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3715
3716	/*
3717	 * Build up a vdev tree based on the boot device's label config.
3718	 */
3719	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3720	    &nvtop) == 0);
3721	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3722	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3723	    VDEV_ALLOC_ROOTPOOL);
3724	spa_config_exit(spa, SCL_ALL, FTAG);
3725	if (error) {
3726		mutex_exit(&spa_namespace_lock);
3727		nvlist_free(config);
3728		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3729		    pname);
3730		return (error);
3731	}
3732
3733	/*
3734	 * Get the boot vdev.
3735	 */
3736	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3737		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3738		    (u_longlong_t)guid);
3739		error = ENOENT;
3740		goto out;
3741	}
3742
3743	/*
3744	 * Determine if there is a better boot device.
3745	 */
3746	avd = bvd;
3747	spa_alt_rootvdev(rvd, &avd, &txg);
3748	if (avd != bvd) {
3749		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3750		    "try booting from '%s'", avd->vdev_path);
3751		error = EINVAL;
3752		goto out;
3753	}
3754
3755	/*
3756	 * If the boot device is part of a spare vdev then ensure that
3757	 * we're booting off the active spare.
3758	 */
3759	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3760	    !bvd->vdev_isspare) {
3761		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3762		    "try booting from '%s'",
3763		    bvd->vdev_parent->
3764		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3765		error = EINVAL;
3766		goto out;
3767	}
3768
3769	error = 0;
3770out:
3771	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3772	vdev_free(rvd);
3773	spa_config_exit(spa, SCL_ALL, FTAG);
3774	mutex_exit(&spa_namespace_lock);
3775
3776	nvlist_free(config);
3777	return (error);
3778}
3779
3780#endif
3781
3782/*
3783 * Import a non-root pool into the system.
3784 */
3785int
3786spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3787{
3788	spa_t *spa;
3789	char *altroot = NULL;
3790	spa_load_state_t state = SPA_LOAD_IMPORT;
3791	zpool_rewind_policy_t policy;
3792	uint64_t mode = spa_mode_global;
3793	uint64_t readonly = B_FALSE;
3794	int error;
3795	nvlist_t *nvroot;
3796	nvlist_t **spares, **l2cache;
3797	uint_t nspares, nl2cache;
3798
3799	/*
3800	 * If a pool with this name exists, return failure.
3801	 */
3802	mutex_enter(&spa_namespace_lock);
3803	if (spa_lookup(pool) != NULL) {
3804		mutex_exit(&spa_namespace_lock);
3805		return (EEXIST);
3806	}
3807
3808	/*
3809	 * Create and initialize the spa structure.
3810	 */
3811	(void) nvlist_lookup_string(props,
3812	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3813	(void) nvlist_lookup_uint64(props,
3814	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3815	if (readonly)
3816		mode = FREAD;
3817	spa = spa_add(pool, config, altroot);
3818	spa->spa_import_flags = flags;
3819
3820	/*
3821	 * Verbatim import - Take a pool and insert it into the namespace
3822	 * as if it had been loaded at boot.
3823	 */
3824	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3825		if (props != NULL)
3826			spa_configfile_set(spa, props, B_FALSE);
3827
3828		spa_config_sync(spa, B_FALSE, B_TRUE);
3829
3830		mutex_exit(&spa_namespace_lock);
3831		spa_history_log_version(spa, "import");
3832
3833		return (0);
3834	}
3835
3836	spa_activate(spa, mode);
3837
3838	/*
3839	 * Don't start async tasks until we know everything is healthy.
3840	 */
3841	spa_async_suspend(spa);
3842
3843	zpool_get_rewind_policy(config, &policy);
3844	if (policy.zrp_request & ZPOOL_DO_REWIND)
3845		state = SPA_LOAD_RECOVER;
3846
3847	/*
3848	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3849	 * because the user-supplied config is actually the one to trust when
3850	 * doing an import.
3851	 */
3852	if (state != SPA_LOAD_RECOVER)
3853		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3854
3855	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3856	    policy.zrp_request);
3857
3858	/*
3859	 * Propagate anything learned while loading the pool and pass it
3860	 * back to caller (i.e. rewind info, missing devices, etc).
3861	 */
3862	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3863	    spa->spa_load_info) == 0);
3864
3865	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3866	/*
3867	 * Toss any existing sparelist, as it doesn't have any validity
3868	 * anymore, and conflicts with spa_has_spare().
3869	 */
3870	if (spa->spa_spares.sav_config) {
3871		nvlist_free(spa->spa_spares.sav_config);
3872		spa->spa_spares.sav_config = NULL;
3873		spa_load_spares(spa);
3874	}
3875	if (spa->spa_l2cache.sav_config) {
3876		nvlist_free(spa->spa_l2cache.sav_config);
3877		spa->spa_l2cache.sav_config = NULL;
3878		spa_load_l2cache(spa);
3879	}
3880
3881	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3882	    &nvroot) == 0);
3883	if (error == 0)
3884		error = spa_validate_aux(spa, nvroot, -1ULL,
3885		    VDEV_ALLOC_SPARE);
3886	if (error == 0)
3887		error = spa_validate_aux(spa, nvroot, -1ULL,
3888		    VDEV_ALLOC_L2CACHE);
3889	spa_config_exit(spa, SCL_ALL, FTAG);
3890
3891	if (props != NULL)
3892		spa_configfile_set(spa, props, B_FALSE);
3893
3894	if (error != 0 || (props && spa_writeable(spa) &&
3895	    (error = spa_prop_set(spa, props)))) {
3896		spa_unload(spa);
3897		spa_deactivate(spa);
3898		spa_remove(spa);
3899		mutex_exit(&spa_namespace_lock);
3900		return (error);
3901	}
3902
3903	spa_async_resume(spa);
3904
3905	/*
3906	 * Override any spares and level 2 cache devices as specified by
3907	 * the user, as these may have correct device names/devids, etc.
3908	 */
3909	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3910	    &spares, &nspares) == 0) {
3911		if (spa->spa_spares.sav_config)
3912			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3913			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3914		else
3915			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3916			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3917		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3918		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3919		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3920		spa_load_spares(spa);
3921		spa_config_exit(spa, SCL_ALL, FTAG);
3922		spa->spa_spares.sav_sync = B_TRUE;
3923	}
3924	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3925	    &l2cache, &nl2cache) == 0) {
3926		if (spa->spa_l2cache.sav_config)
3927			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3928			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3929		else
3930			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3931			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3932		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3933		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3934		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3935		spa_load_l2cache(spa);
3936		spa_config_exit(spa, SCL_ALL, FTAG);
3937		spa->spa_l2cache.sav_sync = B_TRUE;
3938	}
3939
3940	/*
3941	 * Check for any removed devices.
3942	 */
3943	if (spa->spa_autoreplace) {
3944		spa_aux_check_removed(&spa->spa_spares);
3945		spa_aux_check_removed(&spa->spa_l2cache);
3946	}
3947
3948	if (spa_writeable(spa)) {
3949		/*
3950		 * Update the config cache to include the newly-imported pool.
3951		 */
3952		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3953	}
3954
3955	/*
3956	 * It's possible that the pool was expanded while it was exported.
3957	 * We kick off an async task to handle this for us.
3958	 */
3959	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3960
3961	mutex_exit(&spa_namespace_lock);
3962	spa_history_log_version(spa, "import");
3963
3964	return (0);
3965}
3966
3967nvlist_t *
3968spa_tryimport(nvlist_t *tryconfig)
3969{
3970	nvlist_t *config = NULL;
3971	char *poolname;
3972	spa_t *spa;
3973	uint64_t state;
3974	int error;
3975
3976	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3977		return (NULL);
3978
3979	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3980		return (NULL);
3981
3982	/*
3983	 * Create and initialize the spa structure.
3984	 */
3985	mutex_enter(&spa_namespace_lock);
3986	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3987	spa_activate(spa, FREAD);
3988
3989	/*
3990	 * Pass off the heavy lifting to spa_load().
3991	 * Pass TRUE for mosconfig because the user-supplied config
3992	 * is actually the one to trust when doing an import.
3993	 */
3994	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3995
3996	/*
3997	 * If 'tryconfig' was at least parsable, return the current config.
3998	 */
3999	if (spa->spa_root_vdev != NULL) {
4000		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4001		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4002		    poolname) == 0);
4003		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4004		    state) == 0);
4005		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4006		    spa->spa_uberblock.ub_timestamp) == 0);
4007		VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4008		    spa->spa_load_info) == 0);
4009
4010		/*
4011		 * If the bootfs property exists on this pool then we
4012		 * copy it out so that external consumers can tell which
4013		 * pools are bootable.
4014		 */
4015		if ((!error || error == EEXIST) && spa->spa_bootfs) {
4016			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4017
4018			/*
4019			 * We have to play games with the name since the
4020			 * pool was opened as TRYIMPORT_NAME.
4021			 */
4022			if (dsl_dsobj_to_dsname(spa_name(spa),
4023			    spa->spa_bootfs, tmpname) == 0) {
4024				char *cp;
4025				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4026
4027				cp = strchr(tmpname, '/');
4028				if (cp == NULL) {
4029					(void) strlcpy(dsname, tmpname,
4030					    MAXPATHLEN);
4031				} else {
4032					(void) snprintf(dsname, MAXPATHLEN,
4033					    "%s/%s", poolname, ++cp);
4034				}
4035				VERIFY(nvlist_add_string(config,
4036				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4037				kmem_free(dsname, MAXPATHLEN);
4038			}
4039			kmem_free(tmpname, MAXPATHLEN);
4040		}
4041
4042		/*
4043		 * Add the list of hot spares and level 2 cache devices.
4044		 */
4045		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4046		spa_add_spares(spa, config);
4047		spa_add_l2cache(spa, config);
4048		spa_config_exit(spa, SCL_CONFIG, FTAG);
4049	}
4050
4051	spa_unload(spa);
4052	spa_deactivate(spa);
4053	spa_remove(spa);
4054	mutex_exit(&spa_namespace_lock);
4055
4056	return (config);
4057}
4058
4059/*
4060 * Pool export/destroy
4061 *
4062 * The act of destroying or exporting a pool is very simple.  We make sure there
4063 * is no more pending I/O and any references to the pool are gone.  Then, we
4064 * update the pool state and sync all the labels to disk, removing the
4065 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4066 * we don't sync the labels or remove the configuration cache.
4067 */
4068static int
4069spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4070    boolean_t force, boolean_t hardforce)
4071{
4072	spa_t *spa;
4073
4074	if (oldconfig)
4075		*oldconfig = NULL;
4076
4077	if (!(spa_mode_global & FWRITE))
4078		return (EROFS);
4079
4080	mutex_enter(&spa_namespace_lock);
4081	if ((spa = spa_lookup(pool)) == NULL) {
4082		mutex_exit(&spa_namespace_lock);
4083		return (ENOENT);
4084	}
4085
4086	/*
4087	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4088	 * reacquire the namespace lock, and see if we can export.
4089	 */
4090	spa_open_ref(spa, FTAG);
4091	mutex_exit(&spa_namespace_lock);
4092	spa_async_suspend(spa);
4093	mutex_enter(&spa_namespace_lock);
4094	spa_close(spa, FTAG);
4095
4096	/*
4097	 * The pool will be in core if it's openable,
4098	 * in which case we can modify its state.
4099	 */
4100	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4101		/*
4102		 * Objsets may be open only because they're dirty, so we
4103		 * have to force it to sync before checking spa_refcnt.
4104		 */
4105		txg_wait_synced(spa->spa_dsl_pool, 0);
4106
4107		/*
4108		 * A pool cannot be exported or destroyed if there are active
4109		 * references.  If we are resetting a pool, allow references by
4110		 * fault injection handlers.
4111		 */
4112		if (!spa_refcount_zero(spa) ||
4113		    (spa->spa_inject_ref != 0 &&
4114		    new_state != POOL_STATE_UNINITIALIZED)) {
4115			spa_async_resume(spa);
4116			mutex_exit(&spa_namespace_lock);
4117			return (EBUSY);
4118		}
4119
4120		/*
4121		 * A pool cannot be exported if it has an active shared spare.
4122		 * This is to prevent other pools stealing the active spare
4123		 * from an exported pool. At user's own will, such pool can
4124		 * be forcedly exported.
4125		 */
4126		if (!force && new_state == POOL_STATE_EXPORTED &&
4127		    spa_has_active_shared_spare(spa)) {
4128			spa_async_resume(spa);
4129			mutex_exit(&spa_namespace_lock);
4130			return (EXDEV);
4131		}
4132
4133		/*
4134		 * We want this to be reflected on every label,
4135		 * so mark them all dirty.  spa_unload() will do the
4136		 * final sync that pushes these changes out.
4137		 */
4138		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4139			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4140			spa->spa_state = new_state;
4141			spa->spa_final_txg = spa_last_synced_txg(spa) +
4142			    TXG_DEFER_SIZE + 1;
4143			vdev_config_dirty(spa->spa_root_vdev);
4144			spa_config_exit(spa, SCL_ALL, FTAG);
4145		}
4146	}
4147
4148	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4149
4150	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4151		spa_unload(spa);
4152		spa_deactivate(spa);
4153	}
4154
4155	if (oldconfig && spa->spa_config)
4156		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4157
4158	if (new_state != POOL_STATE_UNINITIALIZED) {
4159		if (!hardforce)
4160			spa_config_sync(spa, B_TRUE, B_TRUE);
4161		spa_remove(spa);
4162	}
4163	mutex_exit(&spa_namespace_lock);
4164
4165	return (0);
4166}
4167
4168/*
4169 * Destroy a storage pool.
4170 */
4171int
4172spa_destroy(char *pool)
4173{
4174	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4175	    B_FALSE, B_FALSE));
4176}
4177
4178/*
4179 * Export a storage pool.
4180 */
4181int
4182spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4183    boolean_t hardforce)
4184{
4185	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4186	    force, hardforce));
4187}
4188
4189/*
4190 * Similar to spa_export(), this unloads the spa_t without actually removing it
4191 * from the namespace in any way.
4192 */
4193int
4194spa_reset(char *pool)
4195{
4196	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4197	    B_FALSE, B_FALSE));
4198}
4199
4200/*
4201 * ==========================================================================
4202 * Device manipulation
4203 * ==========================================================================
4204 */
4205
4206/*
4207 * Add a device to a storage pool.
4208 */
4209int
4210spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4211{
4212	uint64_t txg, id;
4213	int error;
4214	vdev_t *rvd = spa->spa_root_vdev;
4215	vdev_t *vd, *tvd;
4216	nvlist_t **spares, **l2cache;
4217	uint_t nspares, nl2cache;
4218
4219	ASSERT(spa_writeable(spa));
4220
4221	txg = spa_vdev_enter(spa);
4222
4223	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4224	    VDEV_ALLOC_ADD)) != 0)
4225		return (spa_vdev_exit(spa, NULL, txg, error));
4226
4227	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
4228
4229	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4230	    &nspares) != 0)
4231		nspares = 0;
4232
4233	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4234	    &nl2cache) != 0)
4235		nl2cache = 0;
4236
4237	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4238		return (spa_vdev_exit(spa, vd, txg, EINVAL));
4239
4240	if (vd->vdev_children != 0 &&
4241	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
4242		return (spa_vdev_exit(spa, vd, txg, error));
4243
4244	/*
4245	 * We must validate the spares and l2cache devices after checking the
4246	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
4247	 */
4248	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4249		return (spa_vdev_exit(spa, vd, txg, error));
4250
4251	/*
4252	 * Transfer each new top-level vdev from vd to rvd.
4253	 */
4254	for (int c = 0; c < vd->vdev_children; c++) {
4255
4256		/*
4257		 * Set the vdev id to the first hole, if one exists.
4258		 */
4259		for (id = 0; id < rvd->vdev_children; id++) {
4260			if (rvd->vdev_child[id]->vdev_ishole) {
4261				vdev_free(rvd->vdev_child[id]);
4262				break;
4263			}
4264		}
4265		tvd = vd->vdev_child[c];
4266		vdev_remove_child(vd, tvd);
4267		tvd->vdev_id = id;
4268		vdev_add_child(rvd, tvd);
4269		vdev_config_dirty(tvd);
4270	}
4271
4272	if (nspares != 0) {
4273		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4274		    ZPOOL_CONFIG_SPARES);
4275		spa_load_spares(spa);
4276		spa->spa_spares.sav_sync = B_TRUE;
4277	}
4278
4279	if (nl2cache != 0) {
4280		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4281		    ZPOOL_CONFIG_L2CACHE);
4282		spa_load_l2cache(spa);
4283		spa->spa_l2cache.sav_sync = B_TRUE;
4284	}
4285
4286	/*
4287	 * We have to be careful when adding new vdevs to an existing pool.
4288	 * If other threads start allocating from these vdevs before we
4289	 * sync the config cache, and we lose power, then upon reboot we may
4290	 * fail to open the pool because there are DVAs that the config cache
4291	 * can't translate.  Therefore, we first add the vdevs without
4292	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4293	 * and then let spa_config_update() initialize the new metaslabs.
4294	 *
4295	 * spa_load() checks for added-but-not-initialized vdevs, so that
4296	 * if we lose power at any point in this sequence, the remaining
4297	 * steps will be completed the next time we load the pool.
4298	 */
4299	(void) spa_vdev_exit(spa, vd, txg, 0);
4300
4301	mutex_enter(&spa_namespace_lock);
4302	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4303	mutex_exit(&spa_namespace_lock);
4304
4305	return (0);
4306}
4307
4308/*
4309 * Attach a device to a mirror.  The arguments are the path to any device
4310 * in the mirror, and the nvroot for the new device.  If the path specifies
4311 * a device that is not mirrored, we automatically insert the mirror vdev.
4312 *
4313 * If 'replacing' is specified, the new device is intended to replace the
4314 * existing device; in this case the two devices are made into their own
4315 * mirror using the 'replacing' vdev, which is functionally identical to
4316 * the mirror vdev (it actually reuses all the same ops) but has a few
4317 * extra rules: you can't attach to it after it's been created, and upon
4318 * completion of resilvering, the first disk (the one being replaced)
4319 * is automatically detached.
4320 */
4321int
4322spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4323{
4324	uint64_t txg, dtl_max_txg;
4325	vdev_t *rvd = spa->spa_root_vdev;
4326	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4327	vdev_ops_t *pvops;
4328	char *oldvdpath, *newvdpath;
4329	int newvd_isspare;
4330	int error;
4331
4332	ASSERT(spa_writeable(spa));
4333
4334	txg = spa_vdev_enter(spa);
4335
4336	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4337
4338	if (oldvd == NULL)
4339		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4340
4341	if (!oldvd->vdev_ops->vdev_op_leaf)
4342		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4343
4344	pvd = oldvd->vdev_parent;
4345
4346	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4347	    VDEV_ALLOC_ATTACH)) != 0)
4348		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4349
4350	if (newrootvd->vdev_children != 1)
4351		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4352
4353	newvd = newrootvd->vdev_child[0];
4354
4355	if (!newvd->vdev_ops->vdev_op_leaf)
4356		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4357
4358	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4359		return (spa_vdev_exit(spa, newrootvd, txg, error));
4360
4361	/*
4362	 * Spares can't replace logs
4363	 */
4364	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4365		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4366
4367	if (!replacing) {
4368		/*
4369		 * For attach, the only allowable parent is a mirror or the root
4370		 * vdev.
4371		 */
4372		if (pvd->vdev_ops != &vdev_mirror_ops &&
4373		    pvd->vdev_ops != &vdev_root_ops)
4374			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4375
4376		pvops = &vdev_mirror_ops;
4377	} else {
4378		/*
4379		 * Active hot spares can only be replaced by inactive hot
4380		 * spares.
4381		 */
4382		if (pvd->vdev_ops == &vdev_spare_ops &&
4383		    oldvd->vdev_isspare &&
4384		    !spa_has_spare(spa, newvd->vdev_guid))
4385			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4386
4387		/*
4388		 * If the source is a hot spare, and the parent isn't already a
4389		 * spare, then we want to create a new hot spare.  Otherwise, we
4390		 * want to create a replacing vdev.  The user is not allowed to
4391		 * attach to a spared vdev child unless the 'isspare' state is
4392		 * the same (spare replaces spare, non-spare replaces
4393		 * non-spare).
4394		 */
4395		if (pvd->vdev_ops == &vdev_replacing_ops &&
4396		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4397			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4398		} else if (pvd->vdev_ops == &vdev_spare_ops &&
4399		    newvd->vdev_isspare != oldvd->vdev_isspare) {
4400			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4401		}
4402
4403		if (newvd->vdev_isspare)
4404			pvops = &vdev_spare_ops;
4405		else
4406			pvops = &vdev_replacing_ops;
4407	}
4408
4409	/*
4410	 * Make sure the new device is big enough.
4411	 */
4412	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4413		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4414
4415	/*
4416	 * The new device cannot have a higher alignment requirement
4417	 * than the top-level vdev.
4418	 */
4419	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4420		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4421
4422	/*
4423	 * If this is an in-place replacement, update oldvd's path and devid
4424	 * to make it distinguishable from newvd, and unopenable from now on.
4425	 */
4426	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4427		spa_strfree(oldvd->vdev_path);
4428		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4429		    KM_SLEEP);
4430		(void) sprintf(oldvd->vdev_path, "%s/%s",
4431		    newvd->vdev_path, "old");
4432		if (oldvd->vdev_devid != NULL) {
4433			spa_strfree(oldvd->vdev_devid);
4434			oldvd->vdev_devid = NULL;
4435		}
4436	}
4437
4438	/* mark the device being resilvered */
4439	newvd->vdev_resilvering = B_TRUE;
4440
4441	/*
4442	 * If the parent is not a mirror, or if we're replacing, insert the new
4443	 * mirror/replacing/spare vdev above oldvd.
4444	 */
4445	if (pvd->vdev_ops != pvops)
4446		pvd = vdev_add_parent(oldvd, pvops);
4447
4448	ASSERT(pvd->vdev_top->vdev_parent == rvd);
4449	ASSERT(pvd->vdev_ops == pvops);
4450	ASSERT(oldvd->vdev_parent == pvd);
4451
4452	/*
4453	 * Extract the new device from its root and add it to pvd.
4454	 */
4455	vdev_remove_child(newrootvd, newvd);
4456	newvd->vdev_id = pvd->vdev_children;
4457	newvd->vdev_crtxg = oldvd->vdev_crtxg;
4458	vdev_add_child(pvd, newvd);
4459
4460	tvd = newvd->vdev_top;
4461	ASSERT(pvd->vdev_top == tvd);
4462	ASSERT(tvd->vdev_parent == rvd);
4463
4464	vdev_config_dirty(tvd);
4465
4466	/*
4467	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4468	 * for any dmu_sync-ed blocks.  It will propagate upward when
4469	 * spa_vdev_exit() calls vdev_dtl_reassess().
4470	 */
4471	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4472
4473	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4474	    dtl_max_txg - TXG_INITIAL);
4475
4476	if (newvd->vdev_isspare) {
4477		spa_spare_activate(newvd);
4478		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4479	}
4480
4481	oldvdpath = spa_strdup(oldvd->vdev_path);
4482	newvdpath = spa_strdup(newvd->vdev_path);
4483	newvd_isspare = newvd->vdev_isspare;
4484
4485	/*
4486	 * Mark newvd's DTL dirty in this txg.
4487	 */
4488	vdev_dirty(tvd, VDD_DTL, newvd, txg);
4489
4490	/*
4491	 * Restart the resilver
4492	 */
4493	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4494
4495	/*
4496	 * Commit the config
4497	 */
4498	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4499
4500	spa_history_log_internal(spa, "vdev attach", NULL,
4501	    "%s vdev=%s %s vdev=%s",
4502	    replacing && newvd_isspare ? "spare in" :
4503	    replacing ? "replace" : "attach", newvdpath,
4504	    replacing ? "for" : "to", oldvdpath);
4505
4506	spa_strfree(oldvdpath);
4507	spa_strfree(newvdpath);
4508
4509	if (spa->spa_bootfs)
4510		spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4511
4512	return (0);
4513}
4514
4515/*
4516 * Detach a device from a mirror or replacing vdev.
4517 * If 'replace_done' is specified, only detach if the parent
4518 * is a replacing vdev.
4519 */
4520int
4521spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4522{
4523	uint64_t txg;
4524	int error;
4525	vdev_t *rvd = spa->spa_root_vdev;
4526	vdev_t *vd, *pvd, *cvd, *tvd;
4527	boolean_t unspare = B_FALSE;
4528	uint64_t unspare_guid = 0;
4529	char *vdpath;
4530
4531	ASSERT(spa_writeable(spa));
4532
4533	txg = spa_vdev_enter(spa);
4534
4535	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4536
4537	if (vd == NULL)
4538		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4539
4540	if (!vd->vdev_ops->vdev_op_leaf)
4541		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4542
4543	pvd = vd->vdev_parent;
4544
4545	/*
4546	 * If the parent/child relationship is not as expected, don't do it.
4547	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4548	 * vdev that's replacing B with C.  The user's intent in replacing
4549	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4550	 * the replace by detaching C, the expected behavior is to end up
4551	 * M(A,B).  But suppose that right after deciding to detach C,
4552	 * the replacement of B completes.  We would have M(A,C), and then
4553	 * ask to detach C, which would leave us with just A -- not what
4554	 * the user wanted.  To prevent this, we make sure that the
4555	 * parent/child relationship hasn't changed -- in this example,
4556	 * that C's parent is still the replacing vdev R.
4557	 */
4558	if (pvd->vdev_guid != pguid && pguid != 0)
4559		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4560
4561	/*
4562	 * Only 'replacing' or 'spare' vdevs can be replaced.
4563	 */
4564	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4565	    pvd->vdev_ops != &vdev_spare_ops)
4566		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4567
4568	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4569	    spa_version(spa) >= SPA_VERSION_SPARES);
4570
4571	/*
4572	 * Only mirror, replacing, and spare vdevs support detach.
4573	 */
4574	if (pvd->vdev_ops != &vdev_replacing_ops &&
4575	    pvd->vdev_ops != &vdev_mirror_ops &&
4576	    pvd->vdev_ops != &vdev_spare_ops)
4577		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4578
4579	/*
4580	 * If this device has the only valid copy of some data,
4581	 * we cannot safely detach it.
4582	 */
4583	if (vdev_dtl_required(vd))
4584		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4585
4586	ASSERT(pvd->vdev_children >= 2);
4587
4588	/*
4589	 * If we are detaching the second disk from a replacing vdev, then
4590	 * check to see if we changed the original vdev's path to have "/old"
4591	 * at the end in spa_vdev_attach().  If so, undo that change now.
4592	 */
4593	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4594	    vd->vdev_path != NULL) {
4595		size_t len = strlen(vd->vdev_path);
4596
4597		for (int c = 0; c < pvd->vdev_children; c++) {
4598			cvd = pvd->vdev_child[c];
4599
4600			if (cvd == vd || cvd->vdev_path == NULL)
4601				continue;
4602
4603			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4604			    strcmp(cvd->vdev_path + len, "/old") == 0) {
4605				spa_strfree(cvd->vdev_path);
4606				cvd->vdev_path = spa_strdup(vd->vdev_path);
4607				break;
4608			}
4609		}
4610	}
4611
4612	/*
4613	 * If we are detaching the original disk from a spare, then it implies
4614	 * that the spare should become a real disk, and be removed from the
4615	 * active spare list for the pool.
4616	 */
4617	if (pvd->vdev_ops == &vdev_spare_ops &&
4618	    vd->vdev_id == 0 &&
4619	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4620		unspare = B_TRUE;
4621
4622	/*
4623	 * Erase the disk labels so the disk can be used for other things.
4624	 * This must be done after all other error cases are handled,
4625	 * but before we disembowel vd (so we can still do I/O to it).
4626	 * But if we can't do it, don't treat the error as fatal --
4627	 * it may be that the unwritability of the disk is the reason
4628	 * it's being detached!
4629	 */
4630	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4631
4632	/*
4633	 * Remove vd from its parent and compact the parent's children.
4634	 */
4635	vdev_remove_child(pvd, vd);
4636	vdev_compact_children(pvd);
4637
4638	/*
4639	 * Remember one of the remaining children so we can get tvd below.
4640	 */
4641	cvd = pvd->vdev_child[pvd->vdev_children - 1];
4642
4643	/*
4644	 * If we need to remove the remaining child from the list of hot spares,
4645	 * do it now, marking the vdev as no longer a spare in the process.
4646	 * We must do this before vdev_remove_parent(), because that can
4647	 * change the GUID if it creates a new toplevel GUID.  For a similar
4648	 * reason, we must remove the spare now, in the same txg as the detach;
4649	 * otherwise someone could attach a new sibling, change the GUID, and
4650	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4651	 */
4652	if (unspare) {
4653		ASSERT(cvd->vdev_isspare);
4654		spa_spare_remove(cvd);
4655		unspare_guid = cvd->vdev_guid;
4656		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4657		cvd->vdev_unspare = B_TRUE;
4658	}
4659
4660	/*
4661	 * If the parent mirror/replacing vdev only has one child,
4662	 * the parent is no longer needed.  Remove it from the tree.
4663	 */
4664	if (pvd->vdev_children == 1) {
4665		if (pvd->vdev_ops == &vdev_spare_ops)
4666			cvd->vdev_unspare = B_FALSE;
4667		vdev_remove_parent(cvd);
4668		cvd->vdev_resilvering = B_FALSE;
4669	}
4670
4671
4672	/*
4673	 * We don't set tvd until now because the parent we just removed
4674	 * may have been the previous top-level vdev.
4675	 */
4676	tvd = cvd->vdev_top;
4677	ASSERT(tvd->vdev_parent == rvd);
4678
4679	/*
4680	 * Reevaluate the parent vdev state.
4681	 */
4682	vdev_propagate_state(cvd);
4683
4684	/*
4685	 * If the 'autoexpand' property is set on the pool then automatically
4686	 * try to expand the size of the pool. For example if the device we
4687	 * just detached was smaller than the others, it may be possible to
4688	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4689	 * first so that we can obtain the updated sizes of the leaf vdevs.
4690	 */
4691	if (spa->spa_autoexpand) {
4692		vdev_reopen(tvd);
4693		vdev_expand(tvd, txg);
4694	}
4695
4696	vdev_config_dirty(tvd);
4697
4698	/*
4699	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4700	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4701	 * But first make sure we're not on any *other* txg's DTL list, to
4702	 * prevent vd from being accessed after it's freed.
4703	 */
4704	vdpath = spa_strdup(vd->vdev_path);
4705	for (int t = 0; t < TXG_SIZE; t++)
4706		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4707	vd->vdev_detached = B_TRUE;
4708	vdev_dirty(tvd, VDD_DTL, vd, txg);
4709
4710	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4711
4712	/* hang on to the spa before we release the lock */
4713	spa_open_ref(spa, FTAG);
4714
4715	error = spa_vdev_exit(spa, vd, txg, 0);
4716
4717	spa_history_log_internal(spa, "detach", NULL,
4718	    "vdev=%s", vdpath);
4719	spa_strfree(vdpath);
4720
4721	/*
4722	 * If this was the removal of the original device in a hot spare vdev,
4723	 * then we want to go through and remove the device from the hot spare
4724	 * list of every other pool.
4725	 */
4726	if (unspare) {
4727		spa_t *altspa = NULL;
4728
4729		mutex_enter(&spa_namespace_lock);
4730		while ((altspa = spa_next(altspa)) != NULL) {
4731			if (altspa->spa_state != POOL_STATE_ACTIVE ||
4732			    altspa == spa)
4733				continue;
4734
4735			spa_open_ref(altspa, FTAG);
4736			mutex_exit(&spa_namespace_lock);
4737			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4738			mutex_enter(&spa_namespace_lock);
4739			spa_close(altspa, FTAG);
4740		}
4741		mutex_exit(&spa_namespace_lock);
4742
4743		/* search the rest of the vdevs for spares to remove */
4744		spa_vdev_resilver_done(spa);
4745	}
4746
4747	/* all done with the spa; OK to release */
4748	mutex_enter(&spa_namespace_lock);
4749	spa_close(spa, FTAG);
4750	mutex_exit(&spa_namespace_lock);
4751
4752	return (error);
4753}
4754
4755/*
4756 * Split a set of devices from their mirrors, and create a new pool from them.
4757 */
4758int
4759spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4760    nvlist_t *props, boolean_t exp)
4761{
4762	int error = 0;
4763	uint64_t txg, *glist;
4764	spa_t *newspa;
4765	uint_t c, children, lastlog;
4766	nvlist_t **child, *nvl, *tmp;
4767	dmu_tx_t *tx;
4768	char *altroot = NULL;
4769	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4770	boolean_t activate_slog;
4771
4772	ASSERT(spa_writeable(spa));
4773
4774	txg = spa_vdev_enter(spa);
4775
4776	/* clear the log and flush everything up to now */
4777	activate_slog = spa_passivate_log(spa);
4778	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4779	error = spa_offline_log(spa);
4780	txg = spa_vdev_config_enter(spa);
4781
4782	if (activate_slog)
4783		spa_activate_log(spa);
4784
4785	if (error != 0)
4786		return (spa_vdev_exit(spa, NULL, txg, error));
4787
4788	/* check new spa name before going any further */
4789	if (spa_lookup(newname) != NULL)
4790		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4791
4792	/*
4793	 * scan through all the children to ensure they're all mirrors
4794	 */
4795	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4796	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4797	    &children) != 0)
4798		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4799
4800	/* first, check to ensure we've got the right child count */
4801	rvd = spa->spa_root_vdev;
4802	lastlog = 0;
4803	for (c = 0; c < rvd->vdev_children; c++) {
4804		vdev_t *vd = rvd->vdev_child[c];
4805
4806		/* don't count the holes & logs as children */
4807		if (vd->vdev_islog || vd->vdev_ishole) {
4808			if (lastlog == 0)
4809				lastlog = c;
4810			continue;
4811		}
4812
4813		lastlog = 0;
4814	}
4815	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4816		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4817
4818	/* next, ensure no spare or cache devices are part of the split */
4819	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4820	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4821		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4822
4823	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4824	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4825
4826	/* then, loop over each vdev and validate it */
4827	for (c = 0; c < children; c++) {
4828		uint64_t is_hole = 0;
4829
4830		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4831		    &is_hole);
4832
4833		if (is_hole != 0) {
4834			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4835			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4836				continue;
4837			} else {
4838				error = EINVAL;
4839				break;
4840			}
4841		}
4842
4843		/* which disk is going to be split? */
4844		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4845		    &glist[c]) != 0) {
4846			error = EINVAL;
4847			break;
4848		}
4849
4850		/* look it up in the spa */
4851		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4852		if (vml[c] == NULL) {
4853			error = ENODEV;
4854			break;
4855		}
4856
4857		/* make sure there's nothing stopping the split */
4858		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4859		    vml[c]->vdev_islog ||
4860		    vml[c]->vdev_ishole ||
4861		    vml[c]->vdev_isspare ||
4862		    vml[c]->vdev_isl2cache ||
4863		    !vdev_writeable(vml[c]) ||
4864		    vml[c]->vdev_children != 0 ||
4865		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4866		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4867			error = EINVAL;
4868			break;
4869		}
4870
4871		if (vdev_dtl_required(vml[c])) {
4872			error = EBUSY;
4873			break;
4874		}
4875
4876		/* we need certain info from the top level */
4877		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4878		    vml[c]->vdev_top->vdev_ms_array) == 0);
4879		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4880		    vml[c]->vdev_top->vdev_ms_shift) == 0);
4881		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4882		    vml[c]->vdev_top->vdev_asize) == 0);
4883		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4884		    vml[c]->vdev_top->vdev_ashift) == 0);
4885	}
4886
4887	if (error != 0) {
4888		kmem_free(vml, children * sizeof (vdev_t *));
4889		kmem_free(glist, children * sizeof (uint64_t));
4890		return (spa_vdev_exit(spa, NULL, txg, error));
4891	}
4892
4893	/* stop writers from using the disks */
4894	for (c = 0; c < children; c++) {
4895		if (vml[c] != NULL)
4896			vml[c]->vdev_offline = B_TRUE;
4897	}
4898	vdev_reopen(spa->spa_root_vdev);
4899
4900	/*
4901	 * Temporarily record the splitting vdevs in the spa config.  This
4902	 * will disappear once the config is regenerated.
4903	 */
4904	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4905	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4906	    glist, children) == 0);
4907	kmem_free(glist, children * sizeof (uint64_t));
4908
4909	mutex_enter(&spa->spa_props_lock);
4910	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4911	    nvl) == 0);
4912	mutex_exit(&spa->spa_props_lock);
4913	spa->spa_config_splitting = nvl;
4914	vdev_config_dirty(spa->spa_root_vdev);
4915
4916	/* configure and create the new pool */
4917	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4918	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4919	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4920	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4921	    spa_version(spa)) == 0);
4922	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4923	    spa->spa_config_txg) == 0);
4924	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4925	    spa_generate_guid(NULL)) == 0);
4926	(void) nvlist_lookup_string(props,
4927	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4928
4929	/* add the new pool to the namespace */
4930	newspa = spa_add(newname, config, altroot);
4931	newspa->spa_config_txg = spa->spa_config_txg;
4932	spa_set_log_state(newspa, SPA_LOG_CLEAR);
4933
4934	/* release the spa config lock, retaining the namespace lock */
4935	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4936
4937	if (zio_injection_enabled)
4938		zio_handle_panic_injection(spa, FTAG, 1);
4939
4940	spa_activate(newspa, spa_mode_global);
4941	spa_async_suspend(newspa);
4942
4943	/* create the new pool from the disks of the original pool */
4944	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4945	if (error)
4946		goto out;
4947
4948	/* if that worked, generate a real config for the new pool */
4949	if (newspa->spa_root_vdev != NULL) {
4950		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4951		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4952		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4953		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4954		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4955		    B_TRUE));
4956	}
4957
4958	/* set the props */
4959	if (props != NULL) {
4960		spa_configfile_set(newspa, props, B_FALSE);
4961		error = spa_prop_set(newspa, props);
4962		if (error)
4963			goto out;
4964	}
4965
4966	/* flush everything */
4967	txg = spa_vdev_config_enter(newspa);
4968	vdev_config_dirty(newspa->spa_root_vdev);
4969	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4970
4971	if (zio_injection_enabled)
4972		zio_handle_panic_injection(spa, FTAG, 2);
4973
4974	spa_async_resume(newspa);
4975
4976	/* finally, update the original pool's config */
4977	txg = spa_vdev_config_enter(spa);
4978	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4979	error = dmu_tx_assign(tx, TXG_WAIT);
4980	if (error != 0)
4981		dmu_tx_abort(tx);
4982	for (c = 0; c < children; c++) {
4983		if (vml[c] != NULL) {
4984			vdev_split(vml[c]);
4985			if (error == 0)
4986				spa_history_log_internal(spa, "detach", tx,
4987				    "vdev=%s", vml[c]->vdev_path);
4988			vdev_free(vml[c]);
4989		}
4990	}
4991	vdev_config_dirty(spa->spa_root_vdev);
4992	spa->spa_config_splitting = NULL;
4993	nvlist_free(nvl);
4994	if (error == 0)
4995		dmu_tx_commit(tx);
4996	(void) spa_vdev_exit(spa, NULL, txg, 0);
4997
4998	if (zio_injection_enabled)
4999		zio_handle_panic_injection(spa, FTAG, 3);
5000
5001	/* split is complete; log a history record */
5002	spa_history_log_internal(newspa, "split", NULL,
5003	    "from pool %s", spa_name(spa));
5004
5005	kmem_free(vml, children * sizeof (vdev_t *));
5006
5007	/* if we're not going to mount the filesystems in userland, export */
5008	if (exp)
5009		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5010		    B_FALSE, B_FALSE);
5011
5012	return (error);
5013
5014out:
5015	spa_unload(newspa);
5016	spa_deactivate(newspa);
5017	spa_remove(newspa);
5018
5019	txg = spa_vdev_config_enter(spa);
5020
5021	/* re-online all offlined disks */
5022	for (c = 0; c < children; c++) {
5023		if (vml[c] != NULL)
5024			vml[c]->vdev_offline = B_FALSE;
5025	}
5026	vdev_reopen(spa->spa_root_vdev);
5027
5028	nvlist_free(spa->spa_config_splitting);
5029	spa->spa_config_splitting = NULL;
5030	(void) spa_vdev_exit(spa, NULL, txg, error);
5031
5032	kmem_free(vml, children * sizeof (vdev_t *));
5033	return (error);
5034}
5035
5036static nvlist_t *
5037spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5038{
5039	for (int i = 0; i < count; i++) {
5040		uint64_t guid;
5041
5042		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5043		    &guid) == 0);
5044
5045		if (guid == target_guid)
5046			return (nvpp[i]);
5047	}
5048
5049	return (NULL);
5050}
5051
5052static void
5053spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5054	nvlist_t *dev_to_remove)
5055{
5056	nvlist_t **newdev = NULL;
5057
5058	if (count > 1)
5059		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5060
5061	for (int i = 0, j = 0; i < count; i++) {
5062		if (dev[i] == dev_to_remove)
5063			continue;
5064		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5065	}
5066
5067	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5068	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5069
5070	for (int i = 0; i < count - 1; i++)
5071		nvlist_free(newdev[i]);
5072
5073	if (count > 1)
5074		kmem_free(newdev, (count - 1) * sizeof (void *));
5075}
5076
5077/*
5078 * Evacuate the device.
5079 */
5080static int
5081spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5082{
5083	uint64_t txg;
5084	int error = 0;
5085
5086	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5087	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5088	ASSERT(vd == vd->vdev_top);
5089
5090	/*
5091	 * Evacuate the device.  We don't hold the config lock as writer
5092	 * since we need to do I/O but we do keep the
5093	 * spa_namespace_lock held.  Once this completes the device
5094	 * should no longer have any blocks allocated on it.
5095	 */
5096	if (vd->vdev_islog) {
5097		if (vd->vdev_stat.vs_alloc != 0)
5098			error = spa_offline_log(spa);
5099	} else {
5100		error = ENOTSUP;
5101	}
5102
5103	if (error)
5104		return (error);
5105
5106	/*
5107	 * The evacuation succeeded.  Remove any remaining MOS metadata
5108	 * associated with this vdev, and wait for these changes to sync.
5109	 */
5110	ASSERT0(vd->vdev_stat.vs_alloc);
5111	txg = spa_vdev_config_enter(spa);
5112	vd->vdev_removing = B_TRUE;
5113	vdev_dirty(vd, 0, NULL, txg);
5114	vdev_config_dirty(vd);
5115	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5116
5117	return (0);
5118}
5119
5120/*
5121 * Complete the removal by cleaning up the namespace.
5122 */
5123static void
5124spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5125{
5126	vdev_t *rvd = spa->spa_root_vdev;
5127	uint64_t id = vd->vdev_id;
5128	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5129
5130	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5131	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5132	ASSERT(vd == vd->vdev_top);
5133
5134	/*
5135	 * Only remove any devices which are empty.
5136	 */
5137	if (vd->vdev_stat.vs_alloc != 0)
5138		return;
5139
5140	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5141
5142	if (list_link_active(&vd->vdev_state_dirty_node))
5143		vdev_state_clean(vd);
5144	if (list_link_active(&vd->vdev_config_dirty_node))
5145		vdev_config_clean(vd);
5146
5147	vdev_free(vd);
5148
5149	if (last_vdev) {
5150		vdev_compact_children(rvd);
5151	} else {
5152		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5153		vdev_add_child(rvd, vd);
5154	}
5155	vdev_config_dirty(rvd);
5156
5157	/*
5158	 * Reassess the health of our root vdev.
5159	 */
5160	vdev_reopen(rvd);
5161}
5162
5163/*
5164 * Remove a device from the pool -
5165 *
5166 * Removing a device from the vdev namespace requires several steps
5167 * and can take a significant amount of time.  As a result we use
5168 * the spa_vdev_config_[enter/exit] functions which allow us to
5169 * grab and release the spa_config_lock while still holding the namespace
5170 * lock.  During each step the configuration is synced out.
5171 */
5172
5173/*
5174 * Remove a device from the pool.  Currently, this supports removing only hot
5175 * spares, slogs, and level 2 ARC devices.
5176 */
5177int
5178spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5179{
5180	vdev_t *vd;
5181	metaslab_group_t *mg;
5182	nvlist_t **spares, **l2cache, *nv;
5183	uint64_t txg = 0;
5184	uint_t nspares, nl2cache;
5185	int error = 0;
5186	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5187
5188	ASSERT(spa_writeable(spa));
5189
5190	if (!locked)
5191		txg = spa_vdev_enter(spa);
5192
5193	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5194
5195	if (spa->spa_spares.sav_vdevs != NULL &&
5196	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5197	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5198	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5199		/*
5200		 * Only remove the hot spare if it's not currently in use
5201		 * in this pool.
5202		 */
5203		if (vd == NULL || unspare) {
5204			spa_vdev_remove_aux(spa->spa_spares.sav_config,
5205			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5206			spa_load_spares(spa);
5207			spa->spa_spares.sav_sync = B_TRUE;
5208		} else {
5209			error = EBUSY;
5210		}
5211	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
5212	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5213	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5214	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5215		/*
5216		 * Cache devices can always be removed.
5217		 */
5218		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5219		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5220		spa_load_l2cache(spa);
5221		spa->spa_l2cache.sav_sync = B_TRUE;
5222	} else if (vd != NULL && vd->vdev_islog) {
5223		ASSERT(!locked);
5224		ASSERT(vd == vd->vdev_top);
5225
5226		/*
5227		 * XXX - Once we have bp-rewrite this should
5228		 * become the common case.
5229		 */
5230
5231		mg = vd->vdev_mg;
5232
5233		/*
5234		 * Stop allocating from this vdev.
5235		 */
5236		metaslab_group_passivate(mg);
5237
5238		/*
5239		 * Wait for the youngest allocations and frees to sync,
5240		 * and then wait for the deferral of those frees to finish.
5241		 */
5242		spa_vdev_config_exit(spa, NULL,
5243		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5244
5245		/*
5246		 * Attempt to evacuate the vdev.
5247		 */
5248		error = spa_vdev_remove_evacuate(spa, vd);
5249
5250		txg = spa_vdev_config_enter(spa);
5251
5252		/*
5253		 * If we couldn't evacuate the vdev, unwind.
5254		 */
5255		if (error) {
5256			metaslab_group_activate(mg);
5257			return (spa_vdev_exit(spa, NULL, txg, error));
5258		}
5259
5260		/*
5261		 * Clean up the vdev namespace.
5262		 */
5263		spa_vdev_remove_from_namespace(spa, vd);
5264
5265	} else if (vd != NULL) {
5266		/*
5267		 * Normal vdevs cannot be removed (yet).
5268		 */
5269		error = ENOTSUP;
5270	} else {
5271		/*
5272		 * There is no vdev of any kind with the specified guid.
5273		 */
5274		error = ENOENT;
5275	}
5276
5277	if (!locked)
5278		return (spa_vdev_exit(spa, NULL, txg, error));
5279
5280	return (error);
5281}
5282
5283/*
5284 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5285 * current spared, so we can detach it.
5286 */
5287static vdev_t *
5288spa_vdev_resilver_done_hunt(vdev_t *vd)
5289{
5290	vdev_t *newvd, *oldvd;
5291
5292	for (int c = 0; c < vd->vdev_children; c++) {
5293		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5294		if (oldvd != NULL)
5295			return (oldvd);
5296	}
5297
5298	/*
5299	 * Check for a completed replacement.  We always consider the first
5300	 * vdev in the list to be the oldest vdev, and the last one to be
5301	 * the newest (see spa_vdev_attach() for how that works).  In
5302	 * the case where the newest vdev is faulted, we will not automatically
5303	 * remove it after a resilver completes.  This is OK as it will require
5304	 * user intervention to determine which disk the admin wishes to keep.
5305	 */
5306	if (vd->vdev_ops == &vdev_replacing_ops) {
5307		ASSERT(vd->vdev_children > 1);
5308
5309		newvd = vd->vdev_child[vd->vdev_children - 1];
5310		oldvd = vd->vdev_child[0];
5311
5312		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5313		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5314		    !vdev_dtl_required(oldvd))
5315			return (oldvd);
5316	}
5317
5318	/*
5319	 * Check for a completed resilver with the 'unspare' flag set.
5320	 */
5321	if (vd->vdev_ops == &vdev_spare_ops) {
5322		vdev_t *first = vd->vdev_child[0];
5323		vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5324
5325		if (last->vdev_unspare) {
5326			oldvd = first;
5327			newvd = last;
5328		} else if (first->vdev_unspare) {
5329			oldvd = last;
5330			newvd = first;
5331		} else {
5332			oldvd = NULL;
5333		}
5334
5335		if (oldvd != NULL &&
5336		    vdev_dtl_empty(newvd, DTL_MISSING) &&
5337		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5338		    !vdev_dtl_required(oldvd))
5339			return (oldvd);
5340
5341		/*
5342		 * If there are more than two spares attached to a disk,
5343		 * and those spares are not required, then we want to
5344		 * attempt to free them up now so that they can be used
5345		 * by other pools.  Once we're back down to a single
5346		 * disk+spare, we stop removing them.
5347		 */
5348		if (vd->vdev_children > 2) {
5349			newvd = vd->vdev_child[1];
5350
5351			if (newvd->vdev_isspare && last->vdev_isspare &&
5352			    vdev_dtl_empty(last, DTL_MISSING) &&
5353			    vdev_dtl_empty(last, DTL_OUTAGE) &&
5354			    !vdev_dtl_required(newvd))
5355				return (newvd);
5356		}
5357	}
5358
5359	return (NULL);
5360}
5361
5362static void
5363spa_vdev_resilver_done(spa_t *spa)
5364{
5365	vdev_t *vd, *pvd, *ppvd;
5366	uint64_t guid, sguid, pguid, ppguid;
5367
5368	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5369
5370	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5371		pvd = vd->vdev_parent;
5372		ppvd = pvd->vdev_parent;
5373		guid = vd->vdev_guid;
5374		pguid = pvd->vdev_guid;
5375		ppguid = ppvd->vdev_guid;
5376		sguid = 0;
5377		/*
5378		 * If we have just finished replacing a hot spared device, then
5379		 * we need to detach the parent's first child (the original hot
5380		 * spare) as well.
5381		 */
5382		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5383		    ppvd->vdev_children == 2) {
5384			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5385			sguid = ppvd->vdev_child[1]->vdev_guid;
5386		}
5387		spa_config_exit(spa, SCL_ALL, FTAG);
5388		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5389			return;
5390		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5391			return;
5392		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5393	}
5394
5395	spa_config_exit(spa, SCL_ALL, FTAG);
5396}
5397
5398/*
5399 * Update the stored path or FRU for this vdev.
5400 */
5401int
5402spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5403    boolean_t ispath)
5404{
5405	vdev_t *vd;
5406	boolean_t sync = B_FALSE;
5407
5408	ASSERT(spa_writeable(spa));
5409
5410	spa_vdev_state_enter(spa, SCL_ALL);
5411
5412	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5413		return (spa_vdev_state_exit(spa, NULL, ENOENT));
5414
5415	if (!vd->vdev_ops->vdev_op_leaf)
5416		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5417
5418	if (ispath) {
5419		if (strcmp(value, vd->vdev_path) != 0) {
5420			spa_strfree(vd->vdev_path);
5421			vd->vdev_path = spa_strdup(value);
5422			sync = B_TRUE;
5423		}
5424	} else {
5425		if (vd->vdev_fru == NULL) {
5426			vd->vdev_fru = spa_strdup(value);
5427			sync = B_TRUE;
5428		} else if (strcmp(value, vd->vdev_fru) != 0) {
5429			spa_strfree(vd->vdev_fru);
5430			vd->vdev_fru = spa_strdup(value);
5431			sync = B_TRUE;
5432		}
5433	}
5434
5435	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5436}
5437
5438int
5439spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5440{
5441	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5442}
5443
5444int
5445spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5446{
5447	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5448}
5449
5450/*
5451 * ==========================================================================
5452 * SPA Scanning
5453 * ==========================================================================
5454 */
5455
5456int
5457spa_scan_stop(spa_t *spa)
5458{
5459	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5460	if (dsl_scan_resilvering(spa->spa_dsl_pool))
5461		return (EBUSY);
5462	return (dsl_scan_cancel(spa->spa_dsl_pool));
5463}
5464
5465int
5466spa_scan(spa_t *spa, pool_scan_func_t func)
5467{
5468	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5469
5470	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5471		return (ENOTSUP);
5472
5473	/*
5474	 * If a resilver was requested, but there is no DTL on a
5475	 * writeable leaf device, we have nothing to do.
5476	 */
5477	if (func == POOL_SCAN_RESILVER &&
5478	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5479		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5480		return (0);
5481	}
5482
5483	return (dsl_scan(spa->spa_dsl_pool, func));
5484}
5485
5486/*
5487 * ==========================================================================
5488 * SPA async task processing
5489 * ==========================================================================
5490 */
5491
5492static void
5493spa_async_remove(spa_t *spa, vdev_t *vd)
5494{
5495	if (vd->vdev_remove_wanted) {
5496		vd->vdev_remove_wanted = B_FALSE;
5497		vd->vdev_delayed_close = B_FALSE;
5498		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5499
5500		/*
5501		 * We want to clear the stats, but we don't want to do a full
5502		 * vdev_clear() as that will cause us to throw away
5503		 * degraded/faulted state as well as attempt to reopen the
5504		 * device, all of which is a waste.
5505		 */
5506		vd->vdev_stat.vs_read_errors = 0;
5507		vd->vdev_stat.vs_write_errors = 0;
5508		vd->vdev_stat.vs_checksum_errors = 0;
5509
5510		vdev_state_dirty(vd->vdev_top);
5511	}
5512
5513	for (int c = 0; c < vd->vdev_children; c++)
5514		spa_async_remove(spa, vd->vdev_child[c]);
5515}
5516
5517static void
5518spa_async_probe(spa_t *spa, vdev_t *vd)
5519{
5520	if (vd->vdev_probe_wanted) {
5521		vd->vdev_probe_wanted = B_FALSE;
5522		vdev_reopen(vd);	/* vdev_open() does the actual probe */
5523	}
5524
5525	for (int c = 0; c < vd->vdev_children; c++)
5526		spa_async_probe(spa, vd->vdev_child[c]);
5527}
5528
5529static void
5530spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5531{
5532	sysevent_id_t eid;
5533	nvlist_t *attr;
5534	char *physpath;
5535
5536	if (!spa->spa_autoexpand)
5537		return;
5538
5539	for (int c = 0; c < vd->vdev_children; c++) {
5540		vdev_t *cvd = vd->vdev_child[c];
5541		spa_async_autoexpand(spa, cvd);
5542	}
5543
5544	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5545		return;
5546
5547	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5548	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5549
5550	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5551	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5552
5553	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5554	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5555
5556	nvlist_free(attr);
5557	kmem_free(physpath, MAXPATHLEN);
5558}
5559
5560static void
5561spa_async_thread(spa_t *spa)
5562{
5563	int tasks;
5564
5565	ASSERT(spa->spa_sync_on);
5566
5567	mutex_enter(&spa->spa_async_lock);
5568	tasks = spa->spa_async_tasks;
5569	spa->spa_async_tasks = 0;
5570	mutex_exit(&spa->spa_async_lock);
5571
5572	/*
5573	 * See if the config needs to be updated.
5574	 */
5575	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5576		uint64_t old_space, new_space;
5577
5578		mutex_enter(&spa_namespace_lock);
5579		old_space = metaslab_class_get_space(spa_normal_class(spa));
5580		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5581		new_space = metaslab_class_get_space(spa_normal_class(spa));
5582		mutex_exit(&spa_namespace_lock);
5583
5584		/*
5585		 * If the pool grew as a result of the config update,
5586		 * then log an internal history event.
5587		 */
5588		if (new_space != old_space) {
5589			spa_history_log_internal(spa, "vdev online", NULL,
5590			    "pool '%s' size: %llu(+%llu)",
5591			    spa_name(spa), new_space, new_space - old_space);
5592		}
5593	}
5594
5595	/*
5596	 * See if any devices need to be marked REMOVED.
5597	 */
5598	if (tasks & SPA_ASYNC_REMOVE) {
5599		spa_vdev_state_enter(spa, SCL_NONE);
5600		spa_async_remove(spa, spa->spa_root_vdev);
5601		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5602			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5603		for (int i = 0; i < spa->spa_spares.sav_count; i++)
5604			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5605		(void) spa_vdev_state_exit(spa, NULL, 0);
5606	}
5607
5608	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5609		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5610		spa_async_autoexpand(spa, spa->spa_root_vdev);
5611		spa_config_exit(spa, SCL_CONFIG, FTAG);
5612	}
5613
5614	/*
5615	 * See if any devices need to be probed.
5616	 */
5617	if (tasks & SPA_ASYNC_PROBE) {
5618		spa_vdev_state_enter(spa, SCL_NONE);
5619		spa_async_probe(spa, spa->spa_root_vdev);
5620		(void) spa_vdev_state_exit(spa, NULL, 0);
5621	}
5622
5623	/*
5624	 * If any devices are done replacing, detach them.
5625	 */
5626	if (tasks & SPA_ASYNC_RESILVER_DONE)
5627		spa_vdev_resilver_done(spa);
5628
5629	/*
5630	 * Kick off a resilver.
5631	 */
5632	if (tasks & SPA_ASYNC_RESILVER)
5633		dsl_resilver_restart(spa->spa_dsl_pool, 0);
5634
5635	/*
5636	 * Let the world know that we're done.
5637	 */
5638	mutex_enter(&spa->spa_async_lock);
5639	spa->spa_async_thread = NULL;
5640	cv_broadcast(&spa->spa_async_cv);
5641	mutex_exit(&spa->spa_async_lock);
5642	thread_exit();
5643}
5644
5645void
5646spa_async_suspend(spa_t *spa)
5647{
5648	mutex_enter(&spa->spa_async_lock);
5649	spa->spa_async_suspended++;
5650	while (spa->spa_async_thread != NULL)
5651		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5652	mutex_exit(&spa->spa_async_lock);
5653}
5654
5655void
5656spa_async_resume(spa_t *spa)
5657{
5658	mutex_enter(&spa->spa_async_lock);
5659	ASSERT(spa->spa_async_suspended != 0);
5660	spa->spa_async_suspended--;
5661	mutex_exit(&spa->spa_async_lock);
5662}
5663
5664static void
5665spa_async_dispatch(spa_t *spa)
5666{
5667	mutex_enter(&spa->spa_async_lock);
5668	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5669	    spa->spa_async_thread == NULL &&
5670	    rootdir != NULL && !vn_is_readonly(rootdir))
5671		spa->spa_async_thread = thread_create(NULL, 0,
5672		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5673	mutex_exit(&spa->spa_async_lock);
5674}
5675
5676void
5677spa_async_request(spa_t *spa, int task)
5678{
5679	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5680	mutex_enter(&spa->spa_async_lock);
5681	spa->spa_async_tasks |= task;
5682	mutex_exit(&spa->spa_async_lock);
5683}
5684
5685/*
5686 * ==========================================================================
5687 * SPA syncing routines
5688 * ==========================================================================
5689 */
5690
5691static int
5692bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5693{
5694	bpobj_t *bpo = arg;
5695	bpobj_enqueue(bpo, bp, tx);
5696	return (0);
5697}
5698
5699static int
5700spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5701{
5702	zio_t *zio = arg;
5703
5704	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5705	    zio->io_flags));
5706	return (0);
5707}
5708
5709static void
5710spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5711{
5712	char *packed = NULL;
5713	size_t bufsize;
5714	size_t nvsize = 0;
5715	dmu_buf_t *db;
5716
5717	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5718
5719	/*
5720	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5721	 * information.  This avoids the dbuf_will_dirty() path and
5722	 * saves us a pre-read to get data we don't actually care about.
5723	 */
5724	bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5725	packed = kmem_alloc(bufsize, KM_SLEEP);
5726
5727	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5728	    KM_SLEEP) == 0);
5729	bzero(packed + nvsize, bufsize - nvsize);
5730
5731	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5732
5733	kmem_free(packed, bufsize);
5734
5735	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5736	dmu_buf_will_dirty(db, tx);
5737	*(uint64_t *)db->db_data = nvsize;
5738	dmu_buf_rele(db, FTAG);
5739}
5740
5741static void
5742spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5743    const char *config, const char *entry)
5744{
5745	nvlist_t *nvroot;
5746	nvlist_t **list;
5747	int i;
5748
5749	if (!sav->sav_sync)
5750		return;
5751
5752	/*
5753	 * Update the MOS nvlist describing the list of available devices.
5754	 * spa_validate_aux() will have already made sure this nvlist is
5755	 * valid and the vdevs are labeled appropriately.
5756	 */
5757	if (sav->sav_object == 0) {
5758		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5759		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5760		    sizeof (uint64_t), tx);
5761		VERIFY(zap_update(spa->spa_meta_objset,
5762		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5763		    &sav->sav_object, tx) == 0);
5764	}
5765
5766	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5767	if (sav->sav_count == 0) {
5768		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5769	} else {
5770		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5771		for (i = 0; i < sav->sav_count; i++)
5772			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5773			    B_FALSE, VDEV_CONFIG_L2CACHE);
5774		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5775		    sav->sav_count) == 0);
5776		for (i = 0; i < sav->sav_count; i++)
5777			nvlist_free(list[i]);
5778		kmem_free(list, sav->sav_count * sizeof (void *));
5779	}
5780
5781	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5782	nvlist_free(nvroot);
5783
5784	sav->sav_sync = B_FALSE;
5785}
5786
5787static void
5788spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5789{
5790	nvlist_t *config;
5791
5792	if (list_is_empty(&spa->spa_config_dirty_list))
5793		return;
5794
5795	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5796
5797	config = spa_config_generate(spa, spa->spa_root_vdev,
5798	    dmu_tx_get_txg(tx), B_FALSE);
5799
5800	/*
5801	 * If we're upgrading the spa version then make sure that
5802	 * the config object gets updated with the correct version.
5803	 */
5804	if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5805		fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5806		    spa->spa_uberblock.ub_version);
5807
5808	spa_config_exit(spa, SCL_STATE, FTAG);
5809
5810	if (spa->spa_config_syncing)
5811		nvlist_free(spa->spa_config_syncing);
5812	spa->spa_config_syncing = config;
5813
5814	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5815}
5816
5817static void
5818spa_sync_version(void *arg, dmu_tx_t *tx)
5819{
5820	uint64_t *versionp = arg;
5821	uint64_t version = *versionp;
5822	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5823
5824	/*
5825	 * Setting the version is special cased when first creating the pool.
5826	 */
5827	ASSERT(tx->tx_txg != TXG_INITIAL);
5828
5829	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5830	ASSERT(version >= spa_version(spa));
5831
5832	spa->spa_uberblock.ub_version = version;
5833	vdev_config_dirty(spa->spa_root_vdev);
5834	spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5835}
5836
5837/*
5838 * Set zpool properties.
5839 */
5840static void
5841spa_sync_props(void *arg, dmu_tx_t *tx)
5842{
5843	nvlist_t *nvp = arg;
5844	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5845	objset_t *mos = spa->spa_meta_objset;
5846	nvpair_t *elem = NULL;
5847
5848	mutex_enter(&spa->spa_props_lock);
5849
5850	while ((elem = nvlist_next_nvpair(nvp, elem))) {
5851		uint64_t intval;
5852		char *strval, *fname;
5853		zpool_prop_t prop;
5854		const char *propname;
5855		zprop_type_t proptype;
5856		zfeature_info_t *feature;
5857
5858		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5859		case ZPROP_INVAL:
5860			/*
5861			 * We checked this earlier in spa_prop_validate().
5862			 */
5863			ASSERT(zpool_prop_feature(nvpair_name(elem)));
5864
5865			fname = strchr(nvpair_name(elem), '@') + 1;
5866			VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
5867
5868			spa_feature_enable(spa, feature, tx);
5869			spa_history_log_internal(spa, "set", tx,
5870			    "%s=enabled", nvpair_name(elem));
5871			break;
5872
5873		case ZPOOL_PROP_VERSION:
5874			VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5875			/*
5876			 * The version is synced seperatly before other
5877			 * properties and should be correct by now.
5878			 */
5879			ASSERT3U(spa_version(spa), >=, intval);
5880			break;
5881
5882		case ZPOOL_PROP_ALTROOT:
5883			/*
5884			 * 'altroot' is a non-persistent property. It should
5885			 * have been set temporarily at creation or import time.
5886			 */
5887			ASSERT(spa->spa_root != NULL);
5888			break;
5889
5890		case ZPOOL_PROP_READONLY:
5891		case ZPOOL_PROP_CACHEFILE:
5892			/*
5893			 * 'readonly' and 'cachefile' are also non-persisitent
5894			 * properties.
5895			 */
5896			break;
5897		case ZPOOL_PROP_COMMENT:
5898			VERIFY(nvpair_value_string(elem, &strval) == 0);
5899			if (spa->spa_comment != NULL)
5900				spa_strfree(spa->spa_comment);
5901			spa->spa_comment = spa_strdup(strval);
5902			/*
5903			 * We need to dirty the configuration on all the vdevs
5904			 * so that their labels get updated.  It's unnecessary
5905			 * to do this for pool creation since the vdev's
5906			 * configuratoin has already been dirtied.
5907			 */
5908			if (tx->tx_txg != TXG_INITIAL)
5909				vdev_config_dirty(spa->spa_root_vdev);
5910			spa_history_log_internal(spa, "set", tx,
5911			    "%s=%s", nvpair_name(elem), strval);
5912			break;
5913		default:
5914			/*
5915			 * Set pool property values in the poolprops mos object.
5916			 */
5917			if (spa->spa_pool_props_object == 0) {
5918				spa->spa_pool_props_object =
5919				    zap_create_link(mos, DMU_OT_POOL_PROPS,
5920				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5921				    tx);
5922			}
5923
5924			/* normalize the property name */
5925			propname = zpool_prop_to_name(prop);
5926			proptype = zpool_prop_get_type(prop);
5927
5928			if (nvpair_type(elem) == DATA_TYPE_STRING) {
5929				ASSERT(proptype == PROP_TYPE_STRING);
5930				VERIFY(nvpair_value_string(elem, &strval) == 0);
5931				VERIFY(zap_update(mos,
5932				    spa->spa_pool_props_object, propname,
5933				    1, strlen(strval) + 1, strval, tx) == 0);
5934				spa_history_log_internal(spa, "set", tx,
5935				    "%s=%s", nvpair_name(elem), strval);
5936			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5937				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5938
5939				if (proptype == PROP_TYPE_INDEX) {
5940					const char *unused;
5941					VERIFY(zpool_prop_index_to_string(
5942					    prop, intval, &unused) == 0);
5943				}
5944				VERIFY(zap_update(mos,
5945				    spa->spa_pool_props_object, propname,
5946				    8, 1, &intval, tx) == 0);
5947				spa_history_log_internal(spa, "set", tx,
5948				    "%s=%lld", nvpair_name(elem), intval);
5949			} else {
5950				ASSERT(0); /* not allowed */
5951			}
5952
5953			switch (prop) {
5954			case ZPOOL_PROP_DELEGATION:
5955				spa->spa_delegation = intval;
5956				break;
5957			case ZPOOL_PROP_BOOTFS:
5958				spa->spa_bootfs = intval;
5959				break;
5960			case ZPOOL_PROP_FAILUREMODE:
5961				spa->spa_failmode = intval;
5962				break;
5963			case ZPOOL_PROP_AUTOEXPAND:
5964				spa->spa_autoexpand = intval;
5965				if (tx->tx_txg != TXG_INITIAL)
5966					spa_async_request(spa,
5967					    SPA_ASYNC_AUTOEXPAND);
5968				break;
5969			case ZPOOL_PROP_DEDUPDITTO:
5970				spa->spa_dedup_ditto = intval;
5971				break;
5972			default:
5973				break;
5974			}
5975		}
5976
5977	}
5978
5979	mutex_exit(&spa->spa_props_lock);
5980}
5981
5982/*
5983 * Perform one-time upgrade on-disk changes.  spa_version() does not
5984 * reflect the new version this txg, so there must be no changes this
5985 * txg to anything that the upgrade code depends on after it executes.
5986 * Therefore this must be called after dsl_pool_sync() does the sync
5987 * tasks.
5988 */
5989static void
5990spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5991{
5992	dsl_pool_t *dp = spa->spa_dsl_pool;
5993
5994	ASSERT(spa->spa_sync_pass == 1);
5995
5996	rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
5997
5998	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5999	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6000		dsl_pool_create_origin(dp, tx);
6001
6002		/* Keeping the origin open increases spa_minref */
6003		spa->spa_minref += 3;
6004	}
6005
6006	if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6007	    spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6008		dsl_pool_upgrade_clones(dp, tx);
6009	}
6010
6011	if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6012	    spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6013		dsl_pool_upgrade_dir_clones(dp, tx);
6014
6015		/* Keeping the freedir open increases spa_minref */
6016		spa->spa_minref += 3;
6017	}
6018
6019	if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6020	    spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6021		spa_feature_create_zap_objects(spa, tx);
6022	}
6023	rrw_exit(&dp->dp_config_rwlock, FTAG);
6024}
6025
6026/*
6027 * Sync the specified transaction group.  New blocks may be dirtied as
6028 * part of the process, so we iterate until it converges.
6029 */
6030void
6031spa_sync(spa_t *spa, uint64_t txg)
6032{
6033	dsl_pool_t *dp = spa->spa_dsl_pool;
6034	objset_t *mos = spa->spa_meta_objset;
6035	bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
6036	bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6037	vdev_t *rvd = spa->spa_root_vdev;
6038	vdev_t *vd;
6039	dmu_tx_t *tx;
6040	int error;
6041
6042	VERIFY(spa_writeable(spa));
6043
6044	/*
6045	 * Lock out configuration changes.
6046	 */
6047	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6048
6049	spa->spa_syncing_txg = txg;
6050	spa->spa_sync_pass = 0;
6051
6052	/*
6053	 * If there are any pending vdev state changes, convert them
6054	 * into config changes that go out with this transaction group.
6055	 */
6056	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6057	while (list_head(&spa->spa_state_dirty_list) != NULL) {
6058		/*
6059		 * We need the write lock here because, for aux vdevs,
6060		 * calling vdev_config_dirty() modifies sav_config.
6061		 * This is ugly and will become unnecessary when we
6062		 * eliminate the aux vdev wart by integrating all vdevs
6063		 * into the root vdev tree.
6064		 */
6065		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6066		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6067		while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6068			vdev_state_clean(vd);
6069			vdev_config_dirty(vd);
6070		}
6071