spa.c revision 7d46dc6ca63a6f3f0d51aa655bfcf10cf2405a9e
1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
25 * Copyright (c) 2013, 2014, Nexenta Systems, Inc.  All rights reserved.
26 */
27
28/*
29 * SPA: Storage Pool Allocator
30 *
31 * This file contains all the routines used when modifying on-disk SPA state.
32 * This includes opening, importing, destroying, exporting a pool, and syncing a
33 * pool.
34 */
35
36#include <sys/zfs_context.h>
37#include <sys/fm/fs/zfs.h>
38#include <sys/spa_impl.h>
39#include <sys/zio.h>
40#include <sys/zio_checksum.h>
41#include <sys/dmu.h>
42#include <sys/dmu_tx.h>
43#include <sys/zap.h>
44#include <sys/zil.h>
45#include <sys/ddt.h>
46#include <sys/vdev_impl.h>
47#include <sys/metaslab.h>
48#include <sys/metaslab_impl.h>
49#include <sys/uberblock_impl.h>
50#include <sys/txg.h>
51#include <sys/avl.h>
52#include <sys/dmu_traverse.h>
53#include <sys/dmu_objset.h>
54#include <sys/unique.h>
55#include <sys/dsl_pool.h>
56#include <sys/dsl_dataset.h>
57#include <sys/dsl_dir.h>
58#include <sys/dsl_prop.h>
59#include <sys/dsl_synctask.h>
60#include <sys/fs/zfs.h>
61#include <sys/arc.h>
62#include <sys/callb.h>
63#include <sys/systeminfo.h>
64#include <sys/spa_boot.h>
65#include <sys/zfs_ioctl.h>
66#include <sys/dsl_scan.h>
67#include <sys/zfeature.h>
68#include <sys/dsl_destroy.h>
69
70#ifdef	_KERNEL
71#include <sys/bootprops.h>
72#include <sys/callb.h>
73#include <sys/cpupart.h>
74#include <sys/pool.h>
75#include <sys/sysdc.h>
76#include <sys/zone.h>
77#endif	/* _KERNEL */
78
79#include "zfs_prop.h"
80#include "zfs_comutil.h"
81
82/*
83 * The interval, in seconds, at which failed configuration cache file writes
84 * should be retried.
85 */
86static int zfs_ccw_retry_interval = 300;
87
88typedef enum zti_modes {
89	ZTI_MODE_FIXED,			/* value is # of threads (min 1) */
90	ZTI_MODE_BATCH,			/* cpu-intensive; value is ignored */
91	ZTI_MODE_NULL,			/* don't create a taskq */
92	ZTI_NMODES
93} zti_modes_t;
94
95#define	ZTI_P(n, q)	{ ZTI_MODE_FIXED, (n), (q) }
96#define	ZTI_BATCH	{ ZTI_MODE_BATCH, 0, 1 }
97#define	ZTI_NULL	{ ZTI_MODE_NULL, 0, 0 }
98
99#define	ZTI_N(n)	ZTI_P(n, 1)
100#define	ZTI_ONE		ZTI_N(1)
101
102typedef struct zio_taskq_info {
103	zti_modes_t zti_mode;
104	uint_t zti_value;
105	uint_t zti_count;
106} zio_taskq_info_t;
107
108static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
109	"issue", "issue_high", "intr", "intr_high"
110};
111
112/*
113 * This table defines the taskq settings for each ZFS I/O type. When
114 * initializing a pool, we use this table to create an appropriately sized
115 * taskq. Some operations are low volume and therefore have a small, static
116 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
117 * macros. Other operations process a large amount of data; the ZTI_BATCH
118 * macro causes us to create a taskq oriented for throughput. Some operations
119 * are so high frequency and short-lived that the taskq itself can become a a
120 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
121 * additional degree of parallelism specified by the number of threads per-
122 * taskq and the number of taskqs; when dispatching an event in this case, the
123 * particular taskq is chosen at random.
124 *
125 * The different taskq priorities are to handle the different contexts (issue
126 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
127 * need to be handled with minimum delay.
128 */
129const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
130	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
131	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* NULL */
132	{ ZTI_N(8),	ZTI_NULL,	ZTI_P(12, 8),	ZTI_NULL }, /* READ */
133	{ ZTI_BATCH,	ZTI_N(5),	ZTI_N(8),	ZTI_N(5) }, /* WRITE */
134	{ ZTI_P(12, 8),	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* FREE */
135	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* CLAIM */
136	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* IOCTL */
137};
138
139static void spa_sync_version(void *arg, dmu_tx_t *tx);
140static void spa_sync_props(void *arg, dmu_tx_t *tx);
141static boolean_t spa_has_active_shared_spare(spa_t *spa);
142static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
143    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
144    char **ereport);
145static void spa_vdev_resilver_done(spa_t *spa);
146
147uint_t		zio_taskq_batch_pct = 75;	/* 1 thread per cpu in pset */
148id_t		zio_taskq_psrset_bind = PS_NONE;
149boolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
150uint_t		zio_taskq_basedc = 80;		/* base duty cycle */
151
152boolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
153extern int	zfs_sync_pass_deferred_free;
154
155/*
156 * This (illegal) pool name is used when temporarily importing a spa_t in order
157 * to get the vdev stats associated with the imported devices.
158 */
159#define	TRYIMPORT_NAME	"$import"
160
161/*
162 * ==========================================================================
163 * SPA properties routines
164 * ==========================================================================
165 */
166
167/*
168 * Add a (source=src, propname=propval) list to an nvlist.
169 */
170static void
171spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
172    uint64_t intval, zprop_source_t src)
173{
174	const char *propname = zpool_prop_to_name(prop);
175	nvlist_t *propval;
176
177	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
178	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
179
180	if (strval != NULL)
181		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
182	else
183		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
184
185	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
186	nvlist_free(propval);
187}
188
189/*
190 * Get property values from the spa configuration.
191 */
192static void
193spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
194{
195	vdev_t *rvd = spa->spa_root_vdev;
196	dsl_pool_t *pool = spa->spa_dsl_pool;
197	uint64_t size;
198	uint64_t alloc;
199	uint64_t space;
200	uint64_t cap, version;
201	zprop_source_t src = ZPROP_SRC_NONE;
202	spa_config_dirent_t *dp;
203
204	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
205
206	if (rvd != NULL) {
207		alloc = metaslab_class_get_alloc(spa_normal_class(spa));
208		size = metaslab_class_get_space(spa_normal_class(spa));
209		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
210		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
211		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
212		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
213		    size - alloc, src);
214
215		space = 0;
216		for (int c = 0; c < rvd->vdev_children; c++) {
217			vdev_t *tvd = rvd->vdev_child[c];
218			space += tvd->vdev_max_asize - tvd->vdev_asize;
219		}
220		spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
221		    src);
222
223		spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
224		    (spa_mode(spa) == FREAD), src);
225
226		cap = (size == 0) ? 0 : (alloc * 100 / size);
227		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
228
229		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
230		    ddt_get_pool_dedup_ratio(spa), src);
231
232		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
233		    rvd->vdev_state, src);
234
235		version = spa_version(spa);
236		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
237			src = ZPROP_SRC_DEFAULT;
238		else
239			src = ZPROP_SRC_LOCAL;
240		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
241	}
242
243	if (pool != NULL) {
244		/*
245		 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
246		 * when opening pools before this version freedir will be NULL.
247		 */
248		if (pool->dp_free_dir != NULL) {
249			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
250			    pool->dp_free_dir->dd_phys->dd_used_bytes, src);
251		} else {
252			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
253			    NULL, 0, src);
254		}
255
256		if (pool->dp_leak_dir != NULL) {
257			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
258			    pool->dp_leak_dir->dd_phys->dd_used_bytes, src);
259		} else {
260			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
261			    NULL, 0, src);
262		}
263	}
264
265	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
266
267	if (spa->spa_comment != NULL) {
268		spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
269		    0, ZPROP_SRC_LOCAL);
270	}
271
272	if (spa->spa_root != NULL)
273		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
274		    0, ZPROP_SRC_LOCAL);
275
276	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
277		if (dp->scd_path == NULL) {
278			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
279			    "none", 0, ZPROP_SRC_LOCAL);
280		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
281			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
282			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
283		}
284	}
285}
286
287/*
288 * Get zpool property values.
289 */
290int
291spa_prop_get(spa_t *spa, nvlist_t **nvp)
292{
293	objset_t *mos = spa->spa_meta_objset;
294	zap_cursor_t zc;
295	zap_attribute_t za;
296	int err;
297
298	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
299
300	mutex_enter(&spa->spa_props_lock);
301
302	/*
303	 * Get properties from the spa config.
304	 */
305	spa_prop_get_config(spa, nvp);
306
307	/* If no pool property object, no more prop to get. */
308	if (mos == NULL || spa->spa_pool_props_object == 0) {
309		mutex_exit(&spa->spa_props_lock);
310		return (0);
311	}
312
313	/*
314	 * Get properties from the MOS pool property object.
315	 */
316	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
317	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
318	    zap_cursor_advance(&zc)) {
319		uint64_t intval = 0;
320		char *strval = NULL;
321		zprop_source_t src = ZPROP_SRC_DEFAULT;
322		zpool_prop_t prop;
323
324		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
325			continue;
326
327		switch (za.za_integer_length) {
328		case 8:
329			/* integer property */
330			if (za.za_first_integer !=
331			    zpool_prop_default_numeric(prop))
332				src = ZPROP_SRC_LOCAL;
333
334			if (prop == ZPOOL_PROP_BOOTFS) {
335				dsl_pool_t *dp;
336				dsl_dataset_t *ds = NULL;
337
338				dp = spa_get_dsl(spa);
339				dsl_pool_config_enter(dp, FTAG);
340				if (err = dsl_dataset_hold_obj(dp,
341				    za.za_first_integer, FTAG, &ds)) {
342					dsl_pool_config_exit(dp, FTAG);
343					break;
344				}
345
346				strval = kmem_alloc(
347				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
348				    KM_SLEEP);
349				dsl_dataset_name(ds, strval);
350				dsl_dataset_rele(ds, FTAG);
351				dsl_pool_config_exit(dp, FTAG);
352			} else {
353				strval = NULL;
354				intval = za.za_first_integer;
355			}
356
357			spa_prop_add_list(*nvp, prop, strval, intval, src);
358
359			if (strval != NULL)
360				kmem_free(strval,
361				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
362
363			break;
364
365		case 1:
366			/* string property */
367			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
368			err = zap_lookup(mos, spa->spa_pool_props_object,
369			    za.za_name, 1, za.za_num_integers, strval);
370			if (err) {
371				kmem_free(strval, za.za_num_integers);
372				break;
373			}
374			spa_prop_add_list(*nvp, prop, strval, 0, src);
375			kmem_free(strval, za.za_num_integers);
376			break;
377
378		default:
379			break;
380		}
381	}
382	zap_cursor_fini(&zc);
383	mutex_exit(&spa->spa_props_lock);
384out:
385	if (err && err != ENOENT) {
386		nvlist_free(*nvp);
387		*nvp = NULL;
388		return (err);
389	}
390
391	return (0);
392}
393
394/*
395 * Validate the given pool properties nvlist and modify the list
396 * for the property values to be set.
397 */
398static int
399spa_prop_validate(spa_t *spa, nvlist_t *props)
400{
401	nvpair_t *elem;
402	int error = 0, reset_bootfs = 0;
403	uint64_t objnum = 0;
404	boolean_t has_feature = B_FALSE;
405
406	elem = NULL;
407	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
408		uint64_t intval;
409		char *strval, *slash, *check, *fname;
410		const char *propname = nvpair_name(elem);
411		zpool_prop_t prop = zpool_name_to_prop(propname);
412
413		switch (prop) {
414		case ZPROP_INVAL:
415			if (!zpool_prop_feature(propname)) {
416				error = SET_ERROR(EINVAL);
417				break;
418			}
419
420			/*
421			 * Sanitize the input.
422			 */
423			if (nvpair_type(elem) != DATA_TYPE_UINT64) {
424				error = SET_ERROR(EINVAL);
425				break;
426			}
427
428			if (nvpair_value_uint64(elem, &intval) != 0) {
429				error = SET_ERROR(EINVAL);
430				break;
431			}
432
433			if (intval != 0) {
434				error = SET_ERROR(EINVAL);
435				break;
436			}
437
438			fname = strchr(propname, '@') + 1;
439			if (zfeature_lookup_name(fname, NULL) != 0) {
440				error = SET_ERROR(EINVAL);
441				break;
442			}
443
444			has_feature = B_TRUE;
445			break;
446
447		case ZPOOL_PROP_VERSION:
448			error = nvpair_value_uint64(elem, &intval);
449			if (!error &&
450			    (intval < spa_version(spa) ||
451			    intval > SPA_VERSION_BEFORE_FEATURES ||
452			    has_feature))
453				error = SET_ERROR(EINVAL);
454			break;
455
456		case ZPOOL_PROP_DELEGATION:
457		case ZPOOL_PROP_AUTOREPLACE:
458		case ZPOOL_PROP_LISTSNAPS:
459		case ZPOOL_PROP_AUTOEXPAND:
460			error = nvpair_value_uint64(elem, &intval);
461			if (!error && intval > 1)
462				error = SET_ERROR(EINVAL);
463			break;
464
465		case ZPOOL_PROP_BOOTFS:
466			/*
467			 * If the pool version is less than SPA_VERSION_BOOTFS,
468			 * or the pool is still being created (version == 0),
469			 * the bootfs property cannot be set.
470			 */
471			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
472				error = SET_ERROR(ENOTSUP);
473				break;
474			}
475
476			/*
477			 * Make sure the vdev config is bootable
478			 */
479			if (!vdev_is_bootable(spa->spa_root_vdev)) {
480				error = SET_ERROR(ENOTSUP);
481				break;
482			}
483
484			reset_bootfs = 1;
485
486			error = nvpair_value_string(elem, &strval);
487
488			if (!error) {
489				objset_t *os;
490				uint64_t compress;
491
492				if (strval == NULL || strval[0] == '\0') {
493					objnum = zpool_prop_default_numeric(
494					    ZPOOL_PROP_BOOTFS);
495					break;
496				}
497
498				if (error = dmu_objset_hold(strval, FTAG, &os))
499					break;
500
501				/* Must be ZPL and not gzip compressed. */
502
503				if (dmu_objset_type(os) != DMU_OST_ZFS) {
504					error = SET_ERROR(ENOTSUP);
505				} else if ((error =
506				    dsl_prop_get_int_ds(dmu_objset_ds(os),
507				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
508				    &compress)) == 0 &&
509				    !BOOTFS_COMPRESS_VALID(compress)) {
510					error = SET_ERROR(ENOTSUP);
511				} else {
512					objnum = dmu_objset_id(os);
513				}
514				dmu_objset_rele(os, FTAG);
515			}
516			break;
517
518		case ZPOOL_PROP_FAILUREMODE:
519			error = nvpair_value_uint64(elem, &intval);
520			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
521			    intval > ZIO_FAILURE_MODE_PANIC))
522				error = SET_ERROR(EINVAL);
523
524			/*
525			 * This is a special case which only occurs when
526			 * the pool has completely failed. This allows
527			 * the user to change the in-core failmode property
528			 * without syncing it out to disk (I/Os might
529			 * currently be blocked). We do this by returning
530			 * EIO to the caller (spa_prop_set) to trick it
531			 * into thinking we encountered a property validation
532			 * error.
533			 */
534			if (!error && spa_suspended(spa)) {
535				spa->spa_failmode = intval;
536				error = SET_ERROR(EIO);
537			}
538			break;
539
540		case ZPOOL_PROP_CACHEFILE:
541			if ((error = nvpair_value_string(elem, &strval)) != 0)
542				break;
543
544			if (strval[0] == '\0')
545				break;
546
547			if (strcmp(strval, "none") == 0)
548				break;
549
550			if (strval[0] != '/') {
551				error = SET_ERROR(EINVAL);
552				break;
553			}
554
555			slash = strrchr(strval, '/');
556			ASSERT(slash != NULL);
557
558			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
559			    strcmp(slash, "/..") == 0)
560				error = SET_ERROR(EINVAL);
561			break;
562
563		case ZPOOL_PROP_COMMENT:
564			if ((error = nvpair_value_string(elem, &strval)) != 0)
565				break;
566			for (check = strval; *check != '\0'; check++) {
567				/*
568				 * The kernel doesn't have an easy isprint()
569				 * check.  For this kernel check, we merely
570				 * check ASCII apart from DEL.  Fix this if
571				 * there is an easy-to-use kernel isprint().
572				 */
573				if (*check >= 0x7f) {
574					error = SET_ERROR(EINVAL);
575					break;
576				}
577				check++;
578			}
579			if (strlen(strval) > ZPROP_MAX_COMMENT)
580				error = E2BIG;
581			break;
582
583		case ZPOOL_PROP_DEDUPDITTO:
584			if (spa_version(spa) < SPA_VERSION_DEDUP)
585				error = SET_ERROR(ENOTSUP);
586			else
587				error = nvpair_value_uint64(elem, &intval);
588			if (error == 0 &&
589			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
590				error = SET_ERROR(EINVAL);
591			break;
592		}
593
594		if (error)
595			break;
596	}
597
598	if (!error && reset_bootfs) {
599		error = nvlist_remove(props,
600		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
601
602		if (!error) {
603			error = nvlist_add_uint64(props,
604			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
605		}
606	}
607
608	return (error);
609}
610
611void
612spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
613{
614	char *cachefile;
615	spa_config_dirent_t *dp;
616
617	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
618	    &cachefile) != 0)
619		return;
620
621	dp = kmem_alloc(sizeof (spa_config_dirent_t),
622	    KM_SLEEP);
623
624	if (cachefile[0] == '\0')
625		dp->scd_path = spa_strdup(spa_config_path);
626	else if (strcmp(cachefile, "none") == 0)
627		dp->scd_path = NULL;
628	else
629		dp->scd_path = spa_strdup(cachefile);
630
631	list_insert_head(&spa->spa_config_list, dp);
632	if (need_sync)
633		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
634}
635
636int
637spa_prop_set(spa_t *spa, nvlist_t *nvp)
638{
639	int error;
640	nvpair_t *elem = NULL;
641	boolean_t need_sync = B_FALSE;
642
643	if ((error = spa_prop_validate(spa, nvp)) != 0)
644		return (error);
645
646	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
647		zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
648
649		if (prop == ZPOOL_PROP_CACHEFILE ||
650		    prop == ZPOOL_PROP_ALTROOT ||
651		    prop == ZPOOL_PROP_READONLY)
652			continue;
653
654		if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
655			uint64_t ver;
656
657			if (prop == ZPOOL_PROP_VERSION) {
658				VERIFY(nvpair_value_uint64(elem, &ver) == 0);
659			} else {
660				ASSERT(zpool_prop_feature(nvpair_name(elem)));
661				ver = SPA_VERSION_FEATURES;
662				need_sync = B_TRUE;
663			}
664
665			/* Save time if the version is already set. */
666			if (ver == spa_version(spa))
667				continue;
668
669			/*
670			 * In addition to the pool directory object, we might
671			 * create the pool properties object, the features for
672			 * read object, the features for write object, or the
673			 * feature descriptions object.
674			 */
675			error = dsl_sync_task(spa->spa_name, NULL,
676			    spa_sync_version, &ver,
677			    6, ZFS_SPACE_CHECK_RESERVED);
678			if (error)
679				return (error);
680			continue;
681		}
682
683		need_sync = B_TRUE;
684		break;
685	}
686
687	if (need_sync) {
688		return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
689		    nvp, 6, ZFS_SPACE_CHECK_RESERVED));
690	}
691
692	return (0);
693}
694
695/*
696 * If the bootfs property value is dsobj, clear it.
697 */
698void
699spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
700{
701	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
702		VERIFY(zap_remove(spa->spa_meta_objset,
703		    spa->spa_pool_props_object,
704		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
705		spa->spa_bootfs = 0;
706	}
707}
708
709/*ARGSUSED*/
710static int
711spa_change_guid_check(void *arg, dmu_tx_t *tx)
712{
713	uint64_t *newguid = arg;
714	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
715	vdev_t *rvd = spa->spa_root_vdev;
716	uint64_t vdev_state;
717
718	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
719	vdev_state = rvd->vdev_state;
720	spa_config_exit(spa, SCL_STATE, FTAG);
721
722	if (vdev_state != VDEV_STATE_HEALTHY)
723		return (SET_ERROR(ENXIO));
724
725	ASSERT3U(spa_guid(spa), !=, *newguid);
726
727	return (0);
728}
729
730static void
731spa_change_guid_sync(void *arg, dmu_tx_t *tx)
732{
733	uint64_t *newguid = arg;
734	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
735	uint64_t oldguid;
736	vdev_t *rvd = spa->spa_root_vdev;
737
738	oldguid = spa_guid(spa);
739
740	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
741	rvd->vdev_guid = *newguid;
742	rvd->vdev_guid_sum += (*newguid - oldguid);
743	vdev_config_dirty(rvd);
744	spa_config_exit(spa, SCL_STATE, FTAG);
745
746	spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
747	    oldguid, *newguid);
748}
749
750/*
751 * Change the GUID for the pool.  This is done so that we can later
752 * re-import a pool built from a clone of our own vdevs.  We will modify
753 * the root vdev's guid, our own pool guid, and then mark all of our
754 * vdevs dirty.  Note that we must make sure that all our vdevs are
755 * online when we do this, or else any vdevs that weren't present
756 * would be orphaned from our pool.  We are also going to issue a
757 * sysevent to update any watchers.
758 */
759int
760spa_change_guid(spa_t *spa)
761{
762	int error;
763	uint64_t guid;
764
765	mutex_enter(&spa->spa_vdev_top_lock);
766	mutex_enter(&spa_namespace_lock);
767	guid = spa_generate_guid(NULL);
768
769	error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
770	    spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
771
772	if (error == 0) {
773		spa_config_sync(spa, B_FALSE, B_TRUE);
774		spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
775	}
776
777	mutex_exit(&spa_namespace_lock);
778	mutex_exit(&spa->spa_vdev_top_lock);
779
780	return (error);
781}
782
783/*
784 * ==========================================================================
785 * SPA state manipulation (open/create/destroy/import/export)
786 * ==========================================================================
787 */
788
789static int
790spa_error_entry_compare(const void *a, const void *b)
791{
792	spa_error_entry_t *sa = (spa_error_entry_t *)a;
793	spa_error_entry_t *sb = (spa_error_entry_t *)b;
794	int ret;
795
796	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
797	    sizeof (zbookmark_phys_t));
798
799	if (ret < 0)
800		return (-1);
801	else if (ret > 0)
802		return (1);
803	else
804		return (0);
805}
806
807/*
808 * Utility function which retrieves copies of the current logs and
809 * re-initializes them in the process.
810 */
811void
812spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
813{
814	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
815
816	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
817	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
818
819	avl_create(&spa->spa_errlist_scrub,
820	    spa_error_entry_compare, sizeof (spa_error_entry_t),
821	    offsetof(spa_error_entry_t, se_avl));
822	avl_create(&spa->spa_errlist_last,
823	    spa_error_entry_compare, sizeof (spa_error_entry_t),
824	    offsetof(spa_error_entry_t, se_avl));
825}
826
827static void
828spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
829{
830	const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
831	enum zti_modes mode = ztip->zti_mode;
832	uint_t value = ztip->zti_value;
833	uint_t count = ztip->zti_count;
834	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
835	char name[32];
836	uint_t flags = 0;
837	boolean_t batch = B_FALSE;
838
839	if (mode == ZTI_MODE_NULL) {
840		tqs->stqs_count = 0;
841		tqs->stqs_taskq = NULL;
842		return;
843	}
844
845	ASSERT3U(count, >, 0);
846
847	tqs->stqs_count = count;
848	tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
849
850	switch (mode) {
851	case ZTI_MODE_FIXED:
852		ASSERT3U(value, >=, 1);
853		value = MAX(value, 1);
854		break;
855
856	case ZTI_MODE_BATCH:
857		batch = B_TRUE;
858		flags |= TASKQ_THREADS_CPU_PCT;
859		value = zio_taskq_batch_pct;
860		break;
861
862	default:
863		panic("unrecognized mode for %s_%s taskq (%u:%u) in "
864		    "spa_activate()",
865		    zio_type_name[t], zio_taskq_types[q], mode, value);
866		break;
867	}
868
869	for (uint_t i = 0; i < count; i++) {
870		taskq_t *tq;
871
872		if (count > 1) {
873			(void) snprintf(name, sizeof (name), "%s_%s_%u",
874			    zio_type_name[t], zio_taskq_types[q], i);
875		} else {
876			(void) snprintf(name, sizeof (name), "%s_%s",
877			    zio_type_name[t], zio_taskq_types[q]);
878		}
879
880		if (zio_taskq_sysdc && spa->spa_proc != &p0) {
881			if (batch)
882				flags |= TASKQ_DC_BATCH;
883
884			tq = taskq_create_sysdc(name, value, 50, INT_MAX,
885			    spa->spa_proc, zio_taskq_basedc, flags);
886		} else {
887			pri_t pri = maxclsyspri;
888			/*
889			 * The write issue taskq can be extremely CPU
890			 * intensive.  Run it at slightly lower priority
891			 * than the other taskqs.
892			 */
893			if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
894				pri--;
895
896			tq = taskq_create_proc(name, value, pri, 50,
897			    INT_MAX, spa->spa_proc, flags);
898		}
899
900		tqs->stqs_taskq[i] = tq;
901	}
902}
903
904static void
905spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
906{
907	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
908
909	if (tqs->stqs_taskq == NULL) {
910		ASSERT0(tqs->stqs_count);
911		return;
912	}
913
914	for (uint_t i = 0; i < tqs->stqs_count; i++) {
915		ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
916		taskq_destroy(tqs->stqs_taskq[i]);
917	}
918
919	kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
920	tqs->stqs_taskq = NULL;
921}
922
923/*
924 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
925 * Note that a type may have multiple discrete taskqs to avoid lock contention
926 * on the taskq itself. In that case we choose which taskq at random by using
927 * the low bits of gethrtime().
928 */
929void
930spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
931    task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
932{
933	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
934	taskq_t *tq;
935
936	ASSERT3P(tqs->stqs_taskq, !=, NULL);
937	ASSERT3U(tqs->stqs_count, !=, 0);
938
939	if (tqs->stqs_count == 1) {
940		tq = tqs->stqs_taskq[0];
941	} else {
942		tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
943	}
944
945	taskq_dispatch_ent(tq, func, arg, flags, ent);
946}
947
948static void
949spa_create_zio_taskqs(spa_t *spa)
950{
951	for (int t = 0; t < ZIO_TYPES; t++) {
952		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
953			spa_taskqs_init(spa, t, q);
954		}
955	}
956}
957
958#ifdef _KERNEL
959static void
960spa_thread(void *arg)
961{
962	callb_cpr_t cprinfo;
963
964	spa_t *spa = arg;
965	user_t *pu = PTOU(curproc);
966
967	CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
968	    spa->spa_name);
969
970	ASSERT(curproc != &p0);
971	(void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
972	    "zpool-%s", spa->spa_name);
973	(void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
974
975	/* bind this thread to the requested psrset */
976	if (zio_taskq_psrset_bind != PS_NONE) {
977		pool_lock();
978		mutex_enter(&cpu_lock);
979		mutex_enter(&pidlock);
980		mutex_enter(&curproc->p_lock);
981
982		if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
983		    0, NULL, NULL) == 0)  {
984			curthread->t_bind_pset = zio_taskq_psrset_bind;
985		} else {
986			cmn_err(CE_WARN,
987			    "Couldn't bind process for zfs pool \"%s\" to "
988			    "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
989		}
990
991		mutex_exit(&curproc->p_lock);
992		mutex_exit(&pidlock);
993		mutex_exit(&cpu_lock);
994		pool_unlock();
995	}
996
997	if (zio_taskq_sysdc) {
998		sysdc_thread_enter(curthread, 100, 0);
999	}
1000
1001	spa->spa_proc = curproc;
1002	spa->spa_did = curthread->t_did;
1003
1004	spa_create_zio_taskqs(spa);
1005
1006	mutex_enter(&spa->spa_proc_lock);
1007	ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1008
1009	spa->spa_proc_state = SPA_PROC_ACTIVE;
1010	cv_broadcast(&spa->spa_proc_cv);
1011
1012	CALLB_CPR_SAFE_BEGIN(&cprinfo);
1013	while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1014		cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1015	CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1016
1017	ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1018	spa->spa_proc_state = SPA_PROC_GONE;
1019	spa->spa_proc = &p0;
1020	cv_broadcast(&spa->spa_proc_cv);
1021	CALLB_CPR_EXIT(&cprinfo);	/* drops spa_proc_lock */
1022
1023	mutex_enter(&curproc->p_lock);
1024	lwp_exit();
1025}
1026#endif
1027
1028/*
1029 * Activate an uninitialized pool.
1030 */
1031static void
1032spa_activate(spa_t *spa, int mode)
1033{
1034	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1035
1036	spa->spa_state = POOL_STATE_ACTIVE;
1037	spa->spa_mode = mode;
1038
1039	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1040	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1041
1042	/* Try to create a covering process */
1043	mutex_enter(&spa->spa_proc_lock);
1044	ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1045	ASSERT(spa->spa_proc == &p0);
1046	spa->spa_did = 0;
1047
1048	/* Only create a process if we're going to be around a while. */
1049	if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1050		if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1051		    NULL, 0) == 0) {
1052			spa->spa_proc_state = SPA_PROC_CREATED;
1053			while (spa->spa_proc_state == SPA_PROC_CREATED) {
1054				cv_wait(&spa->spa_proc_cv,
1055				    &spa->spa_proc_lock);
1056			}
1057			ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1058			ASSERT(spa->spa_proc != &p0);
1059			ASSERT(spa->spa_did != 0);
1060		} else {
1061#ifdef _KERNEL
1062			cmn_err(CE_WARN,
1063			    "Couldn't create process for zfs pool \"%s\"\n",
1064			    spa->spa_name);
1065#endif
1066		}
1067	}
1068	mutex_exit(&spa->spa_proc_lock);
1069
1070	/* If we didn't create a process, we need to create our taskqs. */
1071	if (spa->spa_proc == &p0) {
1072		spa_create_zio_taskqs(spa);
1073	}
1074
1075	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1076	    offsetof(vdev_t, vdev_config_dirty_node));
1077	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1078	    offsetof(vdev_t, vdev_state_dirty_node));
1079
1080	txg_list_create(&spa->spa_vdev_txg_list,
1081	    offsetof(struct vdev, vdev_txg_node));
1082
1083	avl_create(&spa->spa_errlist_scrub,
1084	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1085	    offsetof(spa_error_entry_t, se_avl));
1086	avl_create(&spa->spa_errlist_last,
1087	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1088	    offsetof(spa_error_entry_t, se_avl));
1089}
1090
1091/*
1092 * Opposite of spa_activate().
1093 */
1094static void
1095spa_deactivate(spa_t *spa)
1096{
1097	ASSERT(spa->spa_sync_on == B_FALSE);
1098	ASSERT(spa->spa_dsl_pool == NULL);
1099	ASSERT(spa->spa_root_vdev == NULL);
1100	ASSERT(spa->spa_async_zio_root == NULL);
1101	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1102
1103	txg_list_destroy(&spa->spa_vdev_txg_list);
1104
1105	list_destroy(&spa->spa_config_dirty_list);
1106	list_destroy(&spa->spa_state_dirty_list);
1107
1108	for (int t = 0; t < ZIO_TYPES; t++) {
1109		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1110			spa_taskqs_fini(spa, t, q);
1111		}
1112	}
1113
1114	metaslab_class_destroy(spa->spa_normal_class);
1115	spa->spa_normal_class = NULL;
1116
1117	metaslab_class_destroy(spa->spa_log_class);
1118	spa->spa_log_class = NULL;
1119
1120	/*
1121	 * If this was part of an import or the open otherwise failed, we may
1122	 * still have errors left in the queues.  Empty them just in case.
1123	 */
1124	spa_errlog_drain(spa);
1125
1126	avl_destroy(&spa->spa_errlist_scrub);
1127	avl_destroy(&spa->spa_errlist_last);
1128
1129	spa->spa_state = POOL_STATE_UNINITIALIZED;
1130
1131	mutex_enter(&spa->spa_proc_lock);
1132	if (spa->spa_proc_state != SPA_PROC_NONE) {
1133		ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1134		spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1135		cv_broadcast(&spa->spa_proc_cv);
1136		while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1137			ASSERT(spa->spa_proc != &p0);
1138			cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1139		}
1140		ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1141		spa->spa_proc_state = SPA_PROC_NONE;
1142	}
1143	ASSERT(spa->spa_proc == &p0);
1144	mutex_exit(&spa->spa_proc_lock);
1145
1146	/*
1147	 * We want to make sure spa_thread() has actually exited the ZFS
1148	 * module, so that the module can't be unloaded out from underneath
1149	 * it.
1150	 */
1151	if (spa->spa_did != 0) {
1152		thread_join(spa->spa_did);
1153		spa->spa_did = 0;
1154	}
1155}
1156
1157/*
1158 * Verify a pool configuration, and construct the vdev tree appropriately.  This
1159 * will create all the necessary vdevs in the appropriate layout, with each vdev
1160 * in the CLOSED state.  This will prep the pool before open/creation/import.
1161 * All vdev validation is done by the vdev_alloc() routine.
1162 */
1163static int
1164spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1165    uint_t id, int atype)
1166{
1167	nvlist_t **child;
1168	uint_t children;
1169	int error;
1170
1171	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1172		return (error);
1173
1174	if ((*vdp)->vdev_ops->vdev_op_leaf)
1175		return (0);
1176
1177	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1178	    &child, &children);
1179
1180	if (error == ENOENT)
1181		return (0);
1182
1183	if (error) {
1184		vdev_free(*vdp);
1185		*vdp = NULL;
1186		return (SET_ERROR(EINVAL));
1187	}
1188
1189	for (int c = 0; c < children; c++) {
1190		vdev_t *vd;
1191		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1192		    atype)) != 0) {
1193			vdev_free(*vdp);
1194			*vdp = NULL;
1195			return (error);
1196		}
1197	}
1198
1199	ASSERT(*vdp != NULL);
1200
1201	return (0);
1202}
1203
1204/*
1205 * Opposite of spa_load().
1206 */
1207static void
1208spa_unload(spa_t *spa)
1209{
1210	int i;
1211
1212	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1213
1214	/*
1215	 * Stop async tasks.
1216	 */
1217	spa_async_suspend(spa);
1218
1219	/*
1220	 * Stop syncing.
1221	 */
1222	if (spa->spa_sync_on) {
1223		txg_sync_stop(spa->spa_dsl_pool);
1224		spa->spa_sync_on = B_FALSE;
1225	}
1226
1227	/*
1228	 * Wait for any outstanding async I/O to complete.
1229	 */
1230	if (spa->spa_async_zio_root != NULL) {
1231		(void) zio_wait(spa->spa_async_zio_root);
1232		spa->spa_async_zio_root = NULL;
1233	}
1234
1235	bpobj_close(&spa->spa_deferred_bpobj);
1236
1237	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1238
1239	/*
1240	 * Close all vdevs.
1241	 */
1242	if (spa->spa_root_vdev)
1243		vdev_free(spa->spa_root_vdev);
1244	ASSERT(spa->spa_root_vdev == NULL);
1245
1246	/*
1247	 * Close the dsl pool.
1248	 */
1249	if (spa->spa_dsl_pool) {
1250		dsl_pool_close(spa->spa_dsl_pool);
1251		spa->spa_dsl_pool = NULL;
1252		spa->spa_meta_objset = NULL;
1253	}
1254
1255	ddt_unload(spa);
1256
1257
1258	/*
1259	 * Drop and purge level 2 cache
1260	 */
1261	spa_l2cache_drop(spa);
1262
1263	for (i = 0; i < spa->spa_spares.sav_count; i++)
1264		vdev_free(spa->spa_spares.sav_vdevs[i]);
1265	if (spa->spa_spares.sav_vdevs) {
1266		kmem_free(spa->spa_spares.sav_vdevs,
1267		    spa->spa_spares.sav_count * sizeof (void *));
1268		spa->spa_spares.sav_vdevs = NULL;
1269	}
1270	if (spa->spa_spares.sav_config) {
1271		nvlist_free(spa->spa_spares.sav_config);
1272		spa->spa_spares.sav_config = NULL;
1273	}
1274	spa->spa_spares.sav_count = 0;
1275
1276	for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1277		vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1278		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1279	}
1280	if (spa->spa_l2cache.sav_vdevs) {
1281		kmem_free(spa->spa_l2cache.sav_vdevs,
1282		    spa->spa_l2cache.sav_count * sizeof (void *));
1283		spa->spa_l2cache.sav_vdevs = NULL;
1284	}
1285	if (spa->spa_l2cache.sav_config) {
1286		nvlist_free(spa->spa_l2cache.sav_config);
1287		spa->spa_l2cache.sav_config = NULL;
1288	}
1289	spa->spa_l2cache.sav_count = 0;
1290
1291	spa->spa_async_suspended = 0;
1292
1293	if (spa->spa_comment != NULL) {
1294		spa_strfree(spa->spa_comment);
1295		spa->spa_comment = NULL;
1296	}
1297
1298	spa_config_exit(spa, SCL_ALL, FTAG);
1299}
1300
1301/*
1302 * Load (or re-load) the current list of vdevs describing the active spares for
1303 * this pool.  When this is called, we have some form of basic information in
1304 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1305 * then re-generate a more complete list including status information.
1306 */
1307static void
1308spa_load_spares(spa_t *spa)
1309{
1310	nvlist_t **spares;
1311	uint_t nspares;
1312	int i;
1313	vdev_t *vd, *tvd;
1314
1315	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1316
1317	/*
1318	 * First, close and free any existing spare vdevs.
1319	 */
1320	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1321		vd = spa->spa_spares.sav_vdevs[i];
1322
1323		/* Undo the call to spa_activate() below */
1324		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1325		    B_FALSE)) != NULL && tvd->vdev_isspare)
1326			spa_spare_remove(tvd);
1327		vdev_close(vd);
1328		vdev_free(vd);
1329	}
1330
1331	if (spa->spa_spares.sav_vdevs)
1332		kmem_free(spa->spa_spares.sav_vdevs,
1333		    spa->spa_spares.sav_count * sizeof (void *));
1334
1335	if (spa->spa_spares.sav_config == NULL)
1336		nspares = 0;
1337	else
1338		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1339		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1340
1341	spa->spa_spares.sav_count = (int)nspares;
1342	spa->spa_spares.sav_vdevs = NULL;
1343
1344	if (nspares == 0)
1345		return;
1346
1347	/*
1348	 * Construct the array of vdevs, opening them to get status in the
1349	 * process.   For each spare, there is potentially two different vdev_t
1350	 * structures associated with it: one in the list of spares (used only
1351	 * for basic validation purposes) and one in the active vdev
1352	 * configuration (if it's spared in).  During this phase we open and
1353	 * validate each vdev on the spare list.  If the vdev also exists in the
1354	 * active configuration, then we also mark this vdev as an active spare.
1355	 */
1356	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1357	    KM_SLEEP);
1358	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1359		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1360		    VDEV_ALLOC_SPARE) == 0);
1361		ASSERT(vd != NULL);
1362
1363		spa->spa_spares.sav_vdevs[i] = vd;
1364
1365		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1366		    B_FALSE)) != NULL) {
1367			if (!tvd->vdev_isspare)
1368				spa_spare_add(tvd);
1369
1370			/*
1371			 * We only mark the spare active if we were successfully
1372			 * able to load the vdev.  Otherwise, importing a pool
1373			 * with a bad active spare would result in strange
1374			 * behavior, because multiple pool would think the spare
1375			 * is actively in use.
1376			 *
1377			 * There is a vulnerability here to an equally bizarre
1378			 * circumstance, where a dead active spare is later
1379			 * brought back to life (onlined or otherwise).  Given
1380			 * the rarity of this scenario, and the extra complexity
1381			 * it adds, we ignore the possibility.
1382			 */
1383			if (!vdev_is_dead(tvd))
1384				spa_spare_activate(tvd);
1385		}
1386
1387		vd->vdev_top = vd;
1388		vd->vdev_aux = &spa->spa_spares;
1389
1390		if (vdev_open(vd) != 0)
1391			continue;
1392
1393		if (vdev_validate_aux(vd) == 0)
1394			spa_spare_add(vd);
1395	}
1396
1397	/*
1398	 * Recompute the stashed list of spares, with status information
1399	 * this time.
1400	 */
1401	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1402	    DATA_TYPE_NVLIST_ARRAY) == 0);
1403
1404	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1405	    KM_SLEEP);
1406	for (i = 0; i < spa->spa_spares.sav_count; i++)
1407		spares[i] = vdev_config_generate(spa,
1408		    spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1409	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1410	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1411	for (i = 0; i < spa->spa_spares.sav_count; i++)
1412		nvlist_free(spares[i]);
1413	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1414}
1415
1416/*
1417 * Load (or re-load) the current list of vdevs describing the active l2cache for
1418 * this pool.  When this is called, we have some form of basic information in
1419 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1420 * then re-generate a more complete list including status information.
1421 * Devices which are already active have their details maintained, and are
1422 * not re-opened.
1423 */
1424static void
1425spa_load_l2cache(spa_t *spa)
1426{
1427	nvlist_t **l2cache;
1428	uint_t nl2cache;
1429	int i, j, oldnvdevs;
1430	uint64_t guid;
1431	vdev_t *vd, **oldvdevs, **newvdevs;
1432	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1433
1434	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1435
1436	if (sav->sav_config != NULL) {
1437		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1438		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1439		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1440	} else {
1441		nl2cache = 0;
1442		newvdevs = NULL;
1443	}
1444
1445	oldvdevs = sav->sav_vdevs;
1446	oldnvdevs = sav->sav_count;
1447	sav->sav_vdevs = NULL;
1448	sav->sav_count = 0;
1449
1450	/*
1451	 * Process new nvlist of vdevs.
1452	 */
1453	for (i = 0; i < nl2cache; i++) {
1454		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1455		    &guid) == 0);
1456
1457		newvdevs[i] = NULL;
1458		for (j = 0; j < oldnvdevs; j++) {
1459			vd = oldvdevs[j];
1460			if (vd != NULL && guid == vd->vdev_guid) {
1461				/*
1462				 * Retain previous vdev for add/remove ops.
1463				 */
1464				newvdevs[i] = vd;
1465				oldvdevs[j] = NULL;
1466				break;
1467			}
1468		}
1469
1470		if (newvdevs[i] == NULL) {
1471			/*
1472			 * Create new vdev
1473			 */
1474			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1475			    VDEV_ALLOC_L2CACHE) == 0);
1476			ASSERT(vd != NULL);
1477			newvdevs[i] = vd;
1478
1479			/*
1480			 * Commit this vdev as an l2cache device,
1481			 * even if it fails to open.
1482			 */
1483			spa_l2cache_add(vd);
1484
1485			vd->vdev_top = vd;
1486			vd->vdev_aux = sav;
1487
1488			spa_l2cache_activate(vd);
1489
1490			if (vdev_open(vd) != 0)
1491				continue;
1492
1493			(void) vdev_validate_aux(vd);
1494
1495			if (!vdev_is_dead(vd))
1496				l2arc_add_vdev(spa, vd);
1497		}
1498	}
1499
1500	/*
1501	 * Purge vdevs that were dropped
1502	 */
1503	for (i = 0; i < oldnvdevs; i++) {
1504		uint64_t pool;
1505
1506		vd = oldvdevs[i];
1507		if (vd != NULL) {
1508			ASSERT(vd->vdev_isl2cache);
1509
1510			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1511			    pool != 0ULL && l2arc_vdev_present(vd))
1512				l2arc_remove_vdev(vd);
1513			vdev_clear_stats(vd);
1514			vdev_free(vd);
1515		}
1516	}
1517
1518	if (oldvdevs)
1519		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1520
1521	if (sav->sav_config == NULL)
1522		goto out;
1523
1524	sav->sav_vdevs = newvdevs;
1525	sav->sav_count = (int)nl2cache;
1526
1527	/*
1528	 * Recompute the stashed list of l2cache devices, with status
1529	 * information this time.
1530	 */
1531	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1532	    DATA_TYPE_NVLIST_ARRAY) == 0);
1533
1534	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1535	for (i = 0; i < sav->sav_count; i++)
1536		l2cache[i] = vdev_config_generate(spa,
1537		    sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1538	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1539	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1540out:
1541	for (i = 0; i < sav->sav_count; i++)
1542		nvlist_free(l2cache[i]);
1543	if (sav->sav_count)
1544		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1545}
1546
1547static int
1548load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1549{
1550	dmu_buf_t *db;
1551	char *packed = NULL;
1552	size_t nvsize = 0;
1553	int error;
1554	*value = NULL;
1555
1556	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1557	nvsize = *(uint64_t *)db->db_data;
1558	dmu_buf_rele(db, FTAG);
1559
1560	packed = kmem_alloc(nvsize, KM_SLEEP);
1561	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1562	    DMU_READ_PREFETCH);
1563	if (error == 0)
1564		error = nvlist_unpack(packed, nvsize, value, 0);
1565	kmem_free(packed, nvsize);
1566
1567	return (error);
1568}
1569
1570/*
1571 * Checks to see if the given vdev could not be opened, in which case we post a
1572 * sysevent to notify the autoreplace code that the device has been removed.
1573 */
1574static void
1575spa_check_removed(vdev_t *vd)
1576{
1577	for (int c = 0; c < vd->vdev_children; c++)
1578		spa_check_removed(vd->vdev_child[c]);
1579
1580	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1581	    !vd->vdev_ishole) {
1582		zfs_post_autoreplace(vd->vdev_spa, vd);
1583		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1584	}
1585}
1586
1587/*
1588 * Validate the current config against the MOS config
1589 */
1590static boolean_t
1591spa_config_valid(spa_t *spa, nvlist_t *config)
1592{
1593	vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1594	nvlist_t *nv;
1595
1596	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1597
1598	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1599	VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1600
1601	ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1602
1603	/*
1604	 * If we're doing a normal import, then build up any additional
1605	 * diagnostic information about missing devices in this config.
1606	 * We'll pass this up to the user for further processing.
1607	 */
1608	if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1609		nvlist_t **child, *nv;
1610		uint64_t idx = 0;
1611
1612		child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1613		    KM_SLEEP);
1614		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1615
1616		for (int c = 0; c < rvd->vdev_children; c++) {
1617			vdev_t *tvd = rvd->vdev_child[c];
1618			vdev_t *mtvd  = mrvd->vdev_child[c];
1619
1620			if (tvd->vdev_ops == &vdev_missing_ops &&
1621			    mtvd->vdev_ops != &vdev_missing_ops &&
1622			    mtvd->vdev_islog)
1623				child[idx++] = vdev_config_generate(spa, mtvd,
1624				    B_FALSE, 0);
1625		}
1626
1627		if (idx) {
1628			VERIFY(nvlist_add_nvlist_array(nv,
1629			    ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1630			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1631			    ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1632
1633			for (int i = 0; i < idx; i++)
1634				nvlist_free(child[i]);
1635		}
1636		nvlist_free(nv);
1637		kmem_free(child, rvd->vdev_children * sizeof (char **));
1638	}
1639
1640	/*
1641	 * Compare the root vdev tree with the information we have
1642	 * from the MOS config (mrvd). Check each top-level vdev
1643	 * with the corresponding MOS config top-level (mtvd).
1644	 */
1645	for (int c = 0; c < rvd->vdev_children; c++) {
1646		vdev_t *tvd = rvd->vdev_child[c];
1647		vdev_t *mtvd  = mrvd->vdev_child[c];
1648
1649		/*
1650		 * Resolve any "missing" vdevs in the current configuration.
1651		 * If we find that the MOS config has more accurate information
1652		 * about the top-level vdev then use that vdev instead.
1653		 */
1654		if (tvd->vdev_ops == &vdev_missing_ops &&
1655		    mtvd->vdev_ops != &vdev_missing_ops) {
1656
1657			if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1658				continue;
1659
1660			/*
1661			 * Device specific actions.
1662			 */
1663			if (mtvd->vdev_islog) {
1664				spa_set_log_state(spa, SPA_LOG_CLEAR);
1665			} else {
1666				/*
1667				 * XXX - once we have 'readonly' pool
1668				 * support we should be able to handle
1669				 * missing data devices by transitioning
1670				 * the pool to readonly.
1671				 */
1672				continue;
1673			}
1674
1675			/*
1676			 * Swap the missing vdev with the data we were
1677			 * able to obtain from the MOS config.
1678			 */
1679			vdev_remove_child(rvd, tvd);
1680			vdev_remove_child(mrvd, mtvd);
1681
1682			vdev_add_child(rvd, mtvd);
1683			vdev_add_child(mrvd, tvd);
1684
1685			spa_config_exit(spa, SCL_ALL, FTAG);
1686			vdev_load(mtvd);
1687			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1688
1689			vdev_reopen(rvd);
1690		} else if (mtvd->vdev_islog) {
1691			/*
1692			 * Load the slog device's state from the MOS config
1693			 * since it's possible that the label does not
1694			 * contain the most up-to-date information.
1695			 */
1696			vdev_load_log_state(tvd, mtvd);
1697			vdev_reopen(tvd);
1698		}
1699	}
1700	vdev_free(mrvd);
1701	spa_config_exit(spa, SCL_ALL, FTAG);
1702
1703	/*
1704	 * Ensure we were able to validate the config.
1705	 */
1706	return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1707}
1708
1709/*
1710 * Check for missing log devices
1711 */
1712static boolean_t
1713spa_check_logs(spa_t *spa)
1714{
1715	boolean_t rv = B_FALSE;
1716
1717	switch (spa->spa_log_state) {
1718	case SPA_LOG_MISSING:
1719		/* need to recheck in case slog has been restored */
1720	case SPA_LOG_UNKNOWN:
1721		rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1722		    NULL, DS_FIND_CHILDREN) != 0);
1723		if (rv)
1724			spa_set_log_state(spa, SPA_LOG_MISSING);
1725		break;
1726	}
1727	return (rv);
1728}
1729
1730static boolean_t
1731spa_passivate_log(spa_t *spa)
1732{
1733	vdev_t *rvd = spa->spa_root_vdev;
1734	boolean_t slog_found = B_FALSE;
1735
1736	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1737
1738	if (!spa_has_slogs(spa))
1739		return (B_FALSE);
1740
1741	for (int c = 0; c < rvd->vdev_children; c++) {
1742		vdev_t *tvd = rvd->vdev_child[c];
1743		metaslab_group_t *mg = tvd->vdev_mg;
1744
1745		if (tvd->vdev_islog) {
1746			metaslab_group_passivate(mg);
1747			slog_found = B_TRUE;
1748		}
1749	}
1750
1751	return (slog_found);
1752}
1753
1754static void
1755spa_activate_log(spa_t *spa)
1756{
1757	vdev_t *rvd = spa->spa_root_vdev;
1758
1759	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1760
1761	for (int c = 0; c < rvd->vdev_children; c++) {
1762		vdev_t *tvd = rvd->vdev_child[c];
1763		metaslab_group_t *mg = tvd->vdev_mg;
1764
1765		if (tvd->vdev_islog)
1766			metaslab_group_activate(mg);
1767	}
1768}
1769
1770int
1771spa_offline_log(spa_t *spa)
1772{
1773	int error;
1774
1775	error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1776	    NULL, DS_FIND_CHILDREN);
1777	if (error == 0) {
1778		/*
1779		 * We successfully offlined the log device, sync out the
1780		 * current txg so that the "stubby" block can be removed
1781		 * by zil_sync().
1782		 */
1783		txg_wait_synced(spa->spa_dsl_pool, 0);
1784	}
1785	return (error);
1786}
1787
1788static void
1789spa_aux_check_removed(spa_aux_vdev_t *sav)
1790{
1791	for (int i = 0; i < sav->sav_count; i++)
1792		spa_check_removed(sav->sav_vdevs[i]);
1793}
1794
1795void
1796spa_claim_notify(zio_t *zio)
1797{
1798	spa_t *spa = zio->io_spa;
1799
1800	if (zio->io_error)
1801		return;
1802
1803	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1804	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1805		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1806	mutex_exit(&spa->spa_props_lock);
1807}
1808
1809typedef struct spa_load_error {
1810	uint64_t	sle_meta_count;
1811	uint64_t	sle_data_count;
1812} spa_load_error_t;
1813
1814static void
1815spa_load_verify_done(zio_t *zio)
1816{
1817	blkptr_t *bp = zio->io_bp;
1818	spa_load_error_t *sle = zio->io_private;
1819	dmu_object_type_t type = BP_GET_TYPE(bp);
1820	int error = zio->io_error;
1821
1822	if (error) {
1823		if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1824		    type != DMU_OT_INTENT_LOG)
1825			atomic_add_64(&sle->sle_meta_count, 1);
1826		else
1827			atomic_add_64(&sle->sle_data_count, 1);
1828	}
1829	zio_data_buf_free(zio->io_data, zio->io_size);
1830}
1831
1832/*ARGSUSED*/
1833static int
1834spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1835    const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1836{
1837	if (!BP_IS_HOLE(bp) && !BP_IS_EMBEDDED(bp)) {
1838		zio_t *rio = arg;
1839		size_t size = BP_GET_PSIZE(bp);
1840		void *data = zio_data_buf_alloc(size);
1841
1842		zio_nowait(zio_read(rio, spa, bp, data, size,
1843		    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1844		    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1845		    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1846	}
1847	return (0);
1848}
1849
1850static int
1851spa_load_verify(spa_t *spa)
1852{
1853	zio_t *rio;
1854	spa_load_error_t sle = { 0 };
1855	zpool_rewind_policy_t policy;
1856	boolean_t verify_ok = B_FALSE;
1857	int error;
1858
1859	zpool_get_rewind_policy(spa->spa_config, &policy);
1860
1861	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1862		return (0);
1863
1864	rio = zio_root(spa, NULL, &sle,
1865	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1866
1867	error = traverse_pool(spa, spa->spa_verify_min_txg,
1868	    TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1869
1870	(void) zio_wait(rio);
1871
1872	spa->spa_load_meta_errors = sle.sle_meta_count;
1873	spa->spa_load_data_errors = sle.sle_data_count;
1874
1875	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1876	    sle.sle_data_count <= policy.zrp_maxdata) {
1877		int64_t loss = 0;
1878
1879		verify_ok = B_TRUE;
1880		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1881		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1882
1883		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1884		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1885		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1886		VERIFY(nvlist_add_int64(spa->spa_load_info,
1887		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1888		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1889		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1890	} else {
1891		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1892	}
1893
1894	if (error) {
1895		if (error != ENXIO && error != EIO)
1896			error = SET_ERROR(EIO);
1897		return (error);
1898	}
1899
1900	return (verify_ok ? 0 : EIO);
1901}
1902
1903/*
1904 * Find a value in the pool props object.
1905 */
1906static void
1907spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1908{
1909	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1910	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1911}
1912
1913/*
1914 * Find a value in the pool directory object.
1915 */
1916static int
1917spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1918{
1919	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1920	    name, sizeof (uint64_t), 1, val));
1921}
1922
1923static int
1924spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1925{
1926	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1927	return (err);
1928}
1929
1930/*
1931 * Fix up config after a partly-completed split.  This is done with the
1932 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1933 * pool have that entry in their config, but only the splitting one contains
1934 * a list of all the guids of the vdevs that are being split off.
1935 *
1936 * This function determines what to do with that list: either rejoin
1937 * all the disks to the pool, or complete the splitting process.  To attempt
1938 * the rejoin, each disk that is offlined is marked online again, and
1939 * we do a reopen() call.  If the vdev label for every disk that was
1940 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1941 * then we call vdev_split() on each disk, and complete the split.
1942 *
1943 * Otherwise we leave the config alone, with all the vdevs in place in
1944 * the original pool.
1945 */
1946static void
1947spa_try_repair(spa_t *spa, nvlist_t *config)
1948{
1949	uint_t extracted;
1950	uint64_t *glist;
1951	uint_t i, gcount;
1952	nvlist_t *nvl;
1953	vdev_t **vd;
1954	boolean_t attempt_reopen;
1955
1956	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1957		return;
1958
1959	/* check that the config is complete */
1960	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1961	    &glist, &gcount) != 0)
1962		return;
1963
1964	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1965
1966	/* attempt to online all the vdevs & validate */
1967	attempt_reopen = B_TRUE;
1968	for (i = 0; i < gcount; i++) {
1969		if (glist[i] == 0)	/* vdev is hole */
1970			continue;
1971
1972		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1973		if (vd[i] == NULL) {
1974			/*
1975			 * Don't bother attempting to reopen the disks;
1976			 * just do the split.
1977			 */
1978			attempt_reopen = B_FALSE;
1979		} else {
1980			/* attempt to re-online it */
1981			vd[i]->vdev_offline = B_FALSE;
1982		}
1983	}
1984
1985	if (attempt_reopen) {
1986		vdev_reopen(spa->spa_root_vdev);
1987
1988		/* check each device to see what state it's in */
1989		for (extracted = 0, i = 0; i < gcount; i++) {
1990			if (vd[i] != NULL &&
1991			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1992				break;
1993			++extracted;
1994		}
1995	}
1996
1997	/*
1998	 * If every disk has been moved to the new pool, or if we never
1999	 * even attempted to look at them, then we split them off for
2000	 * good.
2001	 */
2002	if (!attempt_reopen || gcount == extracted) {
2003		for (i = 0; i < gcount; i++)
2004			if (vd[i] != NULL)
2005				vdev_split(vd[i]);
2006		vdev_reopen(spa->spa_root_vdev);
2007	}
2008
2009	kmem_free(vd, gcount * sizeof (vdev_t *));
2010}
2011
2012static int
2013spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2014    boolean_t mosconfig)
2015{
2016	nvlist_t *config = spa->spa_config;
2017	char *ereport = FM_EREPORT_ZFS_POOL;
2018	char *comment;
2019	int error;
2020	uint64_t pool_guid;
2021	nvlist_t *nvl;
2022
2023	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2024		return (SET_ERROR(EINVAL));
2025
2026	ASSERT(spa->spa_comment == NULL);
2027	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2028		spa->spa_comment = spa_strdup(comment);
2029
2030	/*
2031	 * Versioning wasn't explicitly added to the label until later, so if
2032	 * it's not present treat it as the initial version.
2033	 */
2034	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2035	    &spa->spa_ubsync.ub_version) != 0)
2036		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2037
2038	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2039	    &spa->spa_config_txg);
2040
2041	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2042	    spa_guid_exists(pool_guid, 0)) {
2043		error = SET_ERROR(EEXIST);
2044	} else {
2045		spa->spa_config_guid = pool_guid;
2046
2047		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2048		    &nvl) == 0) {
2049			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2050			    KM_SLEEP) == 0);
2051		}
2052
2053		nvlist_free(spa->spa_load_info);
2054		spa->spa_load_info = fnvlist_alloc();
2055
2056		gethrestime(&spa->spa_loaded_ts);
2057		error = spa_load_impl(spa, pool_guid, config, state, type,
2058		    mosconfig, &ereport);
2059	}
2060
2061	spa->spa_minref = refcount_count(&spa->spa_refcount);
2062	if (error) {
2063		if (error != EEXIST) {
2064			spa->spa_loaded_ts.tv_sec = 0;
2065			spa->spa_loaded_ts.tv_nsec = 0;
2066		}
2067		if (error != EBADF) {
2068			zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2069		}
2070	}
2071	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2072	spa->spa_ena = 0;
2073
2074	return (error);
2075}
2076
2077/*
2078 * Load an existing storage pool, using the pool's builtin spa_config as a
2079 * source of configuration information.
2080 */
2081static int
2082spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2083    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2084    char **ereport)
2085{
2086	int error = 0;
2087	nvlist_t *nvroot = NULL;
2088	nvlist_t *label;
2089	vdev_t *rvd;
2090	uberblock_t *ub = &spa->spa_uberblock;
2091	uint64_t children, config_cache_txg = spa->spa_config_txg;
2092	int orig_mode = spa->spa_mode;
2093	int parse;
2094	uint64_t obj;
2095	boolean_t missing_feat_write = B_FALSE;
2096
2097	/*
2098	 * If this is an untrusted config, access the pool in read-only mode.
2099	 * This prevents things like resilvering recently removed devices.
2100	 */
2101	if (!mosconfig)
2102		spa->spa_mode = FREAD;
2103
2104	ASSERT(MUTEX_HELD(&spa_namespace_lock));
2105
2106	spa->spa_load_state = state;
2107
2108	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2109		return (SET_ERROR(EINVAL));
2110
2111	parse = (type == SPA_IMPORT_EXISTING ?
2112	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2113
2114	/*
2115	 * Create "The Godfather" zio to hold all async IOs
2116	 */
2117	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2118	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2119
2120	/*
2121	 * Parse the configuration into a vdev tree.  We explicitly set the
2122	 * value that will be returned by spa_version() since parsing the
2123	 * configuration requires knowing the version number.
2124	 */
2125	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2126	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2127	spa_config_exit(spa, SCL_ALL, FTAG);
2128
2129	if (error != 0)
2130		return (error);
2131
2132	ASSERT(spa->spa_root_vdev == rvd);
2133
2134	if (type != SPA_IMPORT_ASSEMBLE) {
2135		ASSERT(spa_guid(spa) == pool_guid);
2136	}
2137
2138	/*
2139	 * Try to open all vdevs, loading each label in the process.
2140	 */
2141	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2142	error = vdev_open(rvd);
2143	spa_config_exit(spa, SCL_ALL, FTAG);
2144	if (error != 0)
2145		return (error);
2146
2147	/*
2148	 * We need to validate the vdev labels against the configuration that
2149	 * we have in hand, which is dependent on the setting of mosconfig. If
2150	 * mosconfig is true then we're validating the vdev labels based on
2151	 * that config.  Otherwise, we're validating against the cached config
2152	 * (zpool.cache) that was read when we loaded the zfs module, and then
2153	 * later we will recursively call spa_load() and validate against
2154	 * the vdev config.
2155	 *
2156	 * If we're assembling a new pool that's been split off from an
2157	 * existing pool, the labels haven't yet been updated so we skip
2158	 * validation for now.
2159	 */
2160	if (type != SPA_IMPORT_ASSEMBLE) {
2161		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2162		error = vdev_validate(rvd, mosconfig);
2163		spa_config_exit(spa, SCL_ALL, FTAG);
2164
2165		if (error != 0)
2166			return (error);
2167
2168		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2169			return (SET_ERROR(ENXIO));
2170	}
2171
2172	/*
2173	 * Find the best uberblock.
2174	 */
2175	vdev_uberblock_load(rvd, ub, &label);
2176
2177	/*
2178	 * If we weren't able to find a single valid uberblock, return failure.
2179	 */
2180	if (ub->ub_txg == 0) {
2181		nvlist_free(label);
2182		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2183	}
2184
2185	/*
2186	 * If the pool has an unsupported version we can't open it.
2187	 */
2188	if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2189		nvlist_free(label);
2190		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2191	}
2192
2193	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2194		nvlist_t *features;
2195
2196		/*
2197		 * If we weren't able to find what's necessary for reading the
2198		 * MOS in the label, return failure.
2199		 */
2200		if (label == NULL || nvlist_lookup_nvlist(label,
2201		    ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2202			nvlist_free(label);
2203			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2204			    ENXIO));
2205		}
2206
2207		/*
2208		 * Update our in-core representation with the definitive values
2209		 * from the label.
2210		 */
2211		nvlist_free(spa->spa_label_features);
2212		VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2213	}
2214
2215	nvlist_free(label);
2216
2217	/*
2218	 * Look through entries in the label nvlist's features_for_read. If
2219	 * there is a feature listed there which we don't understand then we
2220	 * cannot open a pool.
2221	 */
2222	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2223		nvlist_t *unsup_feat;
2224
2225		VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2226		    0);
2227
2228		for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2229		    NULL); nvp != NULL;
2230		    nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2231			if (!zfeature_is_supported(nvpair_name(nvp))) {
2232				VERIFY(nvlist_add_string(unsup_feat,
2233				    nvpair_name(nvp), "") == 0);
2234			}
2235		}
2236
2237		if (!nvlist_empty(unsup_feat)) {
2238			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2239			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2240			nvlist_free(unsup_feat);
2241			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2242			    ENOTSUP));
2243		}
2244
2245		nvlist_free(unsup_feat);
2246	}
2247
2248	/*
2249	 * If the vdev guid sum doesn't match the uberblock, we have an
2250	 * incomplete configuration.  We first check to see if the pool
2251	 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2252	 * If it is, defer the vdev_guid_sum check till later so we
2253	 * can handle missing vdevs.
2254	 */
2255	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2256	    &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2257	    rvd->vdev_guid_sum != ub->ub_guid_sum)
2258		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2259
2260	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2261		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2262		spa_try_repair(spa, config);
2263		spa_config_exit(spa, SCL_ALL, FTAG);
2264		nvlist_free(spa->spa_config_splitting);
2265		spa->spa_config_splitting = NULL;
2266	}
2267
2268	/*
2269	 * Initialize internal SPA structures.
2270	 */
2271	spa->spa_state = POOL_STATE_ACTIVE;
2272	spa->spa_ubsync = spa->spa_uberblock;
2273	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2274	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2275	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2276	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2277	spa->spa_claim_max_txg = spa->spa_first_txg;
2278	spa->spa_prev_software_version = ub->ub_software_version;
2279
2280	error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2281	if (error)
2282		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2283	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2284
2285	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2286		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2287
2288	if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2289		boolean_t missing_feat_read = B_FALSE;
2290		nvlist_t *unsup_feat, *enabled_feat;
2291
2292		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2293		    &spa->spa_feat_for_read_obj) != 0) {
2294			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2295		}
2296
2297		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2298		    &spa->spa_feat_for_write_obj) != 0) {
2299			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2300		}
2301
2302		if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2303		    &spa->spa_feat_desc_obj) != 0) {
2304			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2305		}
2306
2307		enabled_feat = fnvlist_alloc();
2308		unsup_feat = fnvlist_alloc();
2309
2310		if (!spa_features_check(spa, B_FALSE,
2311		    unsup_feat, enabled_feat))
2312			missing_feat_read = B_TRUE;
2313
2314		if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2315			if (!spa_features_check(spa, B_TRUE,
2316			    unsup_feat, enabled_feat)) {
2317				missing_feat_write = B_TRUE;
2318			}
2319		}
2320
2321		fnvlist_add_nvlist(spa->spa_load_info,
2322		    ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2323
2324		if (!nvlist_empty(unsup_feat)) {
2325			fnvlist_add_nvlist(spa->spa_load_info,
2326			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2327		}
2328
2329		fnvlist_free(enabled_feat);
2330		fnvlist_free(unsup_feat);
2331
2332		if (!missing_feat_read) {
2333			fnvlist_add_boolean(spa->spa_load_info,
2334			    ZPOOL_CONFIG_CAN_RDONLY);
2335		}
2336
2337		/*
2338		 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2339		 * twofold: to determine whether the pool is available for
2340		 * import in read-write mode and (if it is not) whether the
2341		 * pool is available for import in read-only mode. If the pool
2342		 * is available for import in read-write mode, it is displayed
2343		 * as available in userland; if it is not available for import
2344		 * in read-only mode, it is displayed as unavailable in
2345		 * userland. If the pool is available for import in read-only
2346		 * mode but not read-write mode, it is displayed as unavailable
2347		 * in userland with a special note that the pool is actually
2348		 * available for open in read-only mode.
2349		 *
2350		 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2351		 * missing a feature for write, we must first determine whether
2352		 * the pool can be opened read-only before returning to
2353		 * userland in order to know whether to display the
2354		 * abovementioned note.
2355		 */
2356		if (missing_feat_read || (missing_feat_write &&
2357		    spa_writeable(spa))) {
2358			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2359			    ENOTSUP));
2360		}
2361
2362		/*
2363		 * Load refcounts for ZFS features from disk into an in-memory
2364		 * cache during SPA initialization.
2365		 */
2366		for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2367			uint64_t refcount;
2368
2369			error = feature_get_refcount_from_disk(spa,
2370			    &spa_feature_table[i], &refcount);
2371			if (error == 0) {
2372				spa->spa_feat_refcount_cache[i] = refcount;
2373			} else if (error == ENOTSUP) {
2374				spa->spa_feat_refcount_cache[i] =
2375				    SPA_FEATURE_DISABLED;
2376			} else {
2377				return (spa_vdev_err(rvd,
2378				    VDEV_AUX_CORRUPT_DATA, EIO));
2379			}
2380		}
2381	}
2382
2383	if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2384		if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2385		    &spa->spa_feat_enabled_txg_obj) != 0)
2386			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2387	}
2388
2389	spa->spa_is_initializing = B_TRUE;
2390	error = dsl_pool_open(spa->spa_dsl_pool);
2391	spa->spa_is_initializing = B_FALSE;
2392	if (error != 0)
2393		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2394
2395	if (!mosconfig) {
2396		uint64_t hostid;
2397		nvlist_t *policy = NULL, *nvconfig;
2398
2399		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2400			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2401
2402		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2403		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2404			char *hostname;
2405			unsigned long myhostid = 0;
2406
2407			VERIFY(nvlist_lookup_string(nvconfig,
2408			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2409
2410#ifdef	_KERNEL
2411			myhostid = zone_get_hostid(NULL);
2412#else	/* _KERNEL */
2413			/*
2414			 * We're emulating the system's hostid in userland, so
2415			 * we can't use zone_get_hostid().
2416			 */
2417			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2418#endif	/* _KERNEL */
2419			if (hostid != 0 && myhostid != 0 &&
2420			    hostid != myhostid) {
2421				nvlist_free(nvconfig);
2422				cmn_err(CE_WARN, "pool '%s' could not be "
2423				    "loaded as it was last accessed by "
2424				    "another system (host: %s hostid: 0x%lx). "
2425				    "See: http://illumos.org/msg/ZFS-8000-EY",
2426				    spa_name(spa), hostname,
2427				    (unsigned long)hostid);
2428				return (SET_ERROR(EBADF));
2429			}
2430		}
2431		if (nvlist_lookup_nvlist(spa->spa_config,
2432		    ZPOOL_REWIND_POLICY, &policy) == 0)
2433			VERIFY(nvlist_add_nvlist(nvconfig,
2434			    ZPOOL_REWIND_POLICY, policy) == 0);
2435
2436		spa_config_set(spa, nvconfig);
2437		spa_unload(spa);
2438		spa_deactivate(spa);
2439		spa_activate(spa, orig_mode);
2440
2441		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2442	}
2443
2444	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2445		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2446	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2447	if (error != 0)
2448		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2449
2450	/*
2451	 * Load the bit that tells us to use the new accounting function
2452	 * (raid-z deflation).  If we have an older pool, this will not
2453	 * be present.
2454	 */
2455	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2456	if (error != 0 && error != ENOENT)
2457		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2458
2459	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2460	    &spa->spa_creation_version);
2461	if (error != 0 && error != ENOENT)
2462		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2463
2464	/*
2465	 * Load the persistent error log.  If we have an older pool, this will
2466	 * not be present.
2467	 */
2468	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2469	if (error != 0 && error != ENOENT)
2470		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2471
2472	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2473	    &spa->spa_errlog_scrub);
2474	if (error != 0 && error != ENOENT)
2475		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2476
2477	/*
2478	 * Load the history object.  If we have an older pool, this
2479	 * will not be present.
2480	 */
2481	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2482	if (error != 0 && error != ENOENT)
2483		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2484
2485	/*
2486	 * If we're assembling the pool from the split-off vdevs of
2487	 * an existing pool, we don't want to attach the spares & cache
2488	 * devices.
2489	 */
2490
2491	/*
2492	 * Load any hot spares for this pool.
2493	 */
2494	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2495	if (error != 0 && error != ENOENT)
2496		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2497	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2498		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2499		if (load_nvlist(spa, spa->spa_spares.sav_object,
2500		    &spa->spa_spares.sav_config) != 0)
2501			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2502
2503		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2504		spa_load_spares(spa);
2505		spa_config_exit(spa, SCL_ALL, FTAG);
2506	} else if (error == 0) {
2507		spa->spa_spares.sav_sync = B_TRUE;
2508	}
2509
2510	/*
2511	 * Load any level 2 ARC devices for this pool.
2512	 */
2513	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2514	    &spa->spa_l2cache.sav_object);
2515	if (error != 0 && error != ENOENT)
2516		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2517	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2518		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2519		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2520		    &spa->spa_l2cache.sav_config) != 0)
2521			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2522
2523		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2524		spa_load_l2cache(spa);
2525		spa_config_exit(spa, SCL_ALL, FTAG);
2526	} else if (error == 0) {
2527		spa->spa_l2cache.sav_sync = B_TRUE;
2528	}
2529
2530	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2531
2532	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2533	if (error && error != ENOENT)
2534		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2535
2536	if (error == 0) {
2537		uint64_t autoreplace;
2538
2539		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2540		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2541		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2542		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2543		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2544		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2545		    &spa->spa_dedup_ditto);
2546
2547		spa->spa_autoreplace = (autoreplace != 0);
2548	}
2549
2550	/*
2551	 * If the 'autoreplace' property is set, then post a resource notifying
2552	 * the ZFS DE that it should not issue any faults for unopenable
2553	 * devices.  We also iterate over the vdevs, and post a sysevent for any
2554	 * unopenable vdevs so that the normal autoreplace handler can take
2555	 * over.
2556	 */
2557	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2558		spa_check_removed(spa->spa_root_vdev);
2559		/*
2560		 * For the import case, this is done in spa_import(), because
2561		 * at this point we're using the spare definitions from
2562		 * the MOS config, not necessarily from the userland config.
2563		 */
2564		if (state != SPA_LOAD_IMPORT) {
2565			spa_aux_check_removed(&spa->spa_spares);
2566			spa_aux_check_removed(&spa->spa_l2cache);
2567		}
2568	}
2569
2570	/*
2571	 * Load the vdev state for all toplevel vdevs.
2572	 */
2573	vdev_load(rvd);
2574
2575	/*
2576	 * Propagate the leaf DTLs we just loaded all the way up the tree.
2577	 */
2578	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2579	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2580	spa_config_exit(spa, SCL_ALL, FTAG);
2581
2582	/*
2583	 * Load the DDTs (dedup tables).
2584	 */
2585	error = ddt_load(spa);
2586	if (error != 0)
2587		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2588
2589	spa_update_dspace(spa);
2590
2591	/*
2592	 * Validate the config, using the MOS config to fill in any
2593	 * information which might be missing.  If we fail to validate
2594	 * the config then declare the pool unfit for use. If we're
2595	 * assembling a pool from a split, the log is not transferred
2596	 * over.
2597	 */
2598	if (type != SPA_IMPORT_ASSEMBLE) {
2599		nvlist_t *nvconfig;
2600
2601		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2602			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2603
2604		if (!spa_config_valid(spa, nvconfig)) {
2605			nvlist_free(nvconfig);
2606			return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2607			    ENXIO));
2608		}
2609		nvlist_free(nvconfig);
2610
2611		/*
2612		 * Now that we've validated the config, check the state of the
2613		 * root vdev.  If it can't be opened, it indicates one or
2614		 * more toplevel vdevs are faulted.
2615		 */
2616		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2617			return (SET_ERROR(ENXIO));
2618
2619		if (spa_check_logs(spa)) {
2620			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2621			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2622		}
2623	}
2624
2625	if (missing_feat_write) {
2626		ASSERT(state == SPA_LOAD_TRYIMPORT);
2627
2628		/*
2629		 * At this point, we know that we can open the pool in
2630		 * read-only mode but not read-write mode. We now have enough
2631		 * information and can return to userland.
2632		 */
2633		return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2634	}
2635
2636	/*
2637	 * We've successfully opened the pool, verify that we're ready
2638	 * to start pushing transactions.
2639	 */
2640	if (state != SPA_LOAD_TRYIMPORT) {
2641		if (error = spa_load_verify(spa))
2642			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2643			    error));
2644	}
2645
2646	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2647	    spa->spa_load_max_txg == UINT64_MAX)) {
2648		dmu_tx_t *tx;
2649		int need_update = B_FALSE;
2650
2651		ASSERT(state != SPA_LOAD_TRYIMPORT);
2652
2653		/*
2654		 * Claim log blocks that haven't been committed yet.
2655		 * This must all happen in a single txg.
2656		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2657		 * invoked from zil_claim_log_block()'s i/o done callback.
2658		 * Price of rollback is that we abandon the log.
2659		 */
2660		spa->spa_claiming = B_TRUE;
2661
2662		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2663		    spa_first_txg(spa));
2664		(void) dmu_objset_find(spa_name(spa),
2665		    zil_claim, tx, DS_FIND_CHILDREN);
2666		dmu_tx_commit(tx);
2667
2668		spa->spa_claiming = B_FALSE;
2669
2670		spa_set_log_state(spa, SPA_LOG_GOOD);
2671		spa->spa_sync_on = B_TRUE;
2672		txg_sync_start(spa->spa_dsl_pool);
2673
2674		/*
2675		 * Wait for all claims to sync.  We sync up to the highest
2676		 * claimed log block birth time so that claimed log blocks
2677		 * don't appear to be from the future.  spa_claim_max_txg
2678		 * will have been set for us by either zil_check_log_chain()
2679		 * (invoked from spa_check_logs()) or zil_claim() above.
2680		 */
2681		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2682
2683		/*
2684		 * If the config cache is stale, or we have uninitialized
2685		 * metaslabs (see spa_vdev_add()), then update the config.
2686		 *
2687		 * If this is a verbatim import, trust the current
2688		 * in-core spa_config and update the disk labels.
2689		 */
2690		if (config_cache_txg != spa->spa_config_txg ||
2691		    state == SPA_LOAD_IMPORT ||
2692		    state == SPA_LOAD_RECOVER ||
2693		    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2694			need_update = B_TRUE;
2695
2696		for (int c = 0; c < rvd->vdev_children; c++)
2697			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2698				need_update = B_TRUE;
2699
2700		/*
2701		 * Update the config cache asychronously in case we're the
2702		 * root pool, in which case the config cache isn't writable yet.
2703		 */
2704		if (need_update)
2705			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2706
2707		/*
2708		 * Check all DTLs to see if anything needs resilvering.
2709		 */
2710		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2711		    vdev_resilver_needed(rvd, NULL, NULL))
2712			spa_async_request(spa, SPA_ASYNC_RESILVER);
2713
2714		/*
2715		 * Log the fact that we booted up (so that we can detect if
2716		 * we rebooted in the middle of an operation).
2717		 */
2718		spa_history_log_version(spa, "open");
2719
2720		/*
2721		 * Delete any inconsistent datasets.
2722		 */
2723		(void) dmu_objset_find(spa_name(spa),
2724		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2725
2726		/*
2727		 * Clean up any stale temporary dataset userrefs.
2728		 */
2729		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2730	}
2731
2732	return (0);
2733}
2734
2735static int
2736spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2737{
2738	int mode = spa->spa_mode;
2739
2740	spa_unload(spa);
2741	spa_deactivate(spa);
2742
2743	spa->spa_load_max_txg--;
2744
2745	spa_activate(spa, mode);
2746	spa_async_suspend(spa);
2747
2748	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2749}
2750
2751/*
2752 * If spa_load() fails this function will try loading prior txg's. If
2753 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2754 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2755 * function will not rewind the pool and will return the same error as
2756 * spa_load().
2757 */
2758static int
2759spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2760    uint64_t max_request, int rewind_flags)
2761{
2762	nvlist_t *loadinfo = NULL;
2763	nvlist_t *config = NULL;
2764	int load_error, rewind_error;
2765	uint64_t safe_rewind_txg;
2766	uint64_t min_txg;
2767
2768	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2769		spa->spa_load_max_txg = spa->spa_load_txg;
2770		spa_set_log_state(spa, SPA_LOG_CLEAR);
2771	} else {
2772		spa->spa_load_max_txg = max_request;
2773	}
2774
2775	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2776	    mosconfig);
2777	if (load_error == 0)
2778		return (0);
2779
2780	if (spa->spa_root_vdev != NULL)
2781		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2782
2783	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2784	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2785
2786	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2787		nvlist_free(config);
2788		return (load_error);
2789	}
2790
2791	if (state == SPA_LOAD_RECOVER) {
2792		/* Price of rolling back is discarding txgs, including log */
2793		spa_set_log_state(spa, SPA_LOG_CLEAR);
2794	} else {
2795		/*
2796		 * If we aren't rolling back save the load info from our first
2797		 * import attempt so that we can restore it after attempting
2798		 * to rewind.
2799		 */
2800		loadinfo = spa->spa_load_info;
2801		spa->spa_load_info = fnvlist_alloc();
2802	}
2803
2804	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2805	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2806	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2807	    TXG_INITIAL : safe_rewind_txg;
2808
2809	/*
2810	 * Continue as long as we're finding errors, we're still within
2811	 * the acceptable rewind range, and we're still finding uberblocks
2812	 */
2813	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2814	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2815		if (spa->spa_load_max_txg < safe_rewind_txg)
2816			spa->spa_extreme_rewind = B_TRUE;
2817		rewind_error = spa_load_retry(spa, state, mosconfig);
2818	}
2819
2820	spa->spa_extreme_rewind = B_FALSE;
2821	spa->spa_load_max_txg = UINT64_MAX;
2822
2823	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2824		spa_config_set(spa, config);
2825
2826	if (state == SPA_LOAD_RECOVER) {
2827		ASSERT3P(loadinfo, ==, NULL);
2828		return (rewind_error);
2829	} else {
2830		/* Store the rewind info as part of the initial load info */
2831		fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2832		    spa->spa_load_info);
2833
2834		/* Restore the initial load info */
2835		fnvlist_free(spa->spa_load_info);
2836		spa->spa_load_info = loadinfo;
2837
2838		return (load_error);
2839	}
2840}
2841
2842/*
2843 * Pool Open/Import
2844 *
2845 * The import case is identical to an open except that the configuration is sent
2846 * down from userland, instead of grabbed from the configuration cache.  For the
2847 * case of an open, the pool configuration will exist in the
2848 * POOL_STATE_UNINITIALIZED state.
2849 *
2850 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2851 * the same time open the pool, without having to keep around the spa_t in some
2852 * ambiguous state.
2853 */
2854static int
2855spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2856    nvlist_t **config)
2857{
2858	spa_t *spa;
2859	spa_load_state_t state = SPA_LOAD_OPEN;
2860	int error;
2861	int locked = B_FALSE;
2862
2863	*spapp = NULL;
2864
2865	/*
2866	 * As disgusting as this is, we need to support recursive calls to this
2867	 * function because dsl_dir_open() is called during spa_load(), and ends
2868	 * up calling spa_open() again.  The real fix is to figure out how to
2869	 * avoid dsl_dir_open() calling this in the first place.
2870	 */
2871	if (mutex_owner(&spa_namespace_lock) != curthread) {
2872		mutex_enter(&spa_namespace_lock);
2873		locked = B_TRUE;
2874	}
2875
2876	if ((spa = spa_lookup(pool)) == NULL) {
2877		if (locked)
2878			mutex_exit(&spa_namespace_lock);
2879		return (SET_ERROR(ENOENT));
2880	}
2881
2882	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2883		zpool_rewind_policy_t policy;
2884
2885		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2886		    &policy);
2887		if (policy.zrp_request & ZPOOL_DO_REWIND)
2888			state = SPA_LOAD_RECOVER;
2889
2890		spa_activate(spa, spa_mode_global);
2891
2892		if (state != SPA_LOAD_RECOVER)
2893			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2894
2895		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2896		    policy.zrp_request);
2897
2898		if (error == EBADF) {
2899			/*
2900			 * If vdev_validate() returns failure (indicated by
2901			 * EBADF), it indicates that one of the vdevs indicates
2902			 * that the pool has been exported or destroyed.  If
2903			 * this is the case, the config cache is out of sync and
2904			 * we should remove the pool from the namespace.
2905			 */
2906			spa_unload(spa);
2907			spa_deactivate(spa);
2908			spa_config_sync(spa, B_TRUE, B_TRUE);
2909			spa_remove(spa);
2910			if (locked)
2911				mutex_exit(&spa_namespace_lock);
2912			return (SET_ERROR(ENOENT));
2913		}
2914
2915		if (error) {
2916			/*
2917			 * We can't open the pool, but we still have useful
2918			 * information: the state of each vdev after the
2919			 * attempted vdev_open().  Return this to the user.
2920			 */
2921			if (config != NULL && spa->spa_config) {
2922				VERIFY(nvlist_dup(spa->spa_config, config,
2923				    KM_SLEEP) == 0);
2924				VERIFY(nvlist_add_nvlist(*config,
2925				    ZPOOL_CONFIG_LOAD_INFO,
2926				    spa->spa_load_info) == 0);
2927			}
2928			spa_unload(spa);
2929			spa_deactivate(spa);
2930			spa->spa_last_open_failed = error;
2931			if (locked)
2932				mutex_exit(&spa_namespace_lock);
2933			*spapp = NULL;
2934			return (error);
2935		}
2936	}
2937
2938	spa_open_ref(spa, tag);
2939
2940	if (config != NULL)
2941		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2942
2943	/*
2944	 * If we've recovered the pool, pass back any information we
2945	 * gathered while doing the load.
2946	 */
2947	if (state == SPA_LOAD_RECOVER) {
2948		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2949		    spa->spa_load_info) == 0);
2950	}
2951
2952	if (locked) {
2953		spa->spa_last_open_failed = 0;
2954		spa->spa_last_ubsync_txg = 0;
2955		spa->spa_load_txg = 0;
2956		mutex_exit(&spa_namespace_lock);
2957	}
2958
2959	*spapp = spa;
2960
2961	return (0);
2962}
2963
2964int
2965spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2966    nvlist_t **config)
2967{
2968	return (spa_open_common(name, spapp, tag, policy, config));
2969}
2970
2971int
2972spa_open(const char *name, spa_t **spapp, void *tag)
2973{
2974	return (spa_open_common(name, spapp, tag, NULL, NULL));
2975}
2976
2977/*
2978 * Lookup the given spa_t, incrementing the inject count in the process,
2979 * preventing it from being exported or destroyed.
2980 */
2981spa_t *
2982spa_inject_addref(char *name)
2983{
2984	spa_t *spa;
2985
2986	mutex_enter(&spa_namespace_lock);
2987	if ((spa = spa_lookup(name)) == NULL) {
2988		mutex_exit(&spa_namespace_lock);
2989		return (NULL);
2990	}
2991	spa->spa_inject_ref++;
2992	mutex_exit(&spa_namespace_lock);
2993
2994	return (spa);
2995}
2996
2997void
2998spa_inject_delref(spa_t *spa)
2999{
3000	mutex_enter(&spa_namespace_lock);
3001	spa->spa_inject_ref--;
3002	mutex_exit(&spa_namespace_lock);
3003}
3004
3005/*
3006 * Add spares device information to the nvlist.
3007 */
3008static void
3009spa_add_spares(spa_t *spa, nvlist_t *config)
3010{
3011	nvlist_t **spares;
3012	uint_t i, nspares;
3013	nvlist_t *nvroot;
3014	uint64_t guid;
3015	vdev_stat_t *vs;
3016	uint_t vsc;
3017	uint64_t pool;
3018
3019	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3020
3021	if (spa->spa_spares.sav_count == 0)
3022		return;
3023
3024	VERIFY(nvlist_lookup_nvlist(config,
3025	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3026	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3027	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3028	if (nspares != 0) {
3029		VERIFY(nvlist_add_nvlist_array(nvroot,
3030		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3031		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3032		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3033
3034		/*
3035		 * Go through and find any spares which have since been
3036		 * repurposed as an active spare.  If this is the case, update
3037		 * their status appropriately.
3038		 */
3039		for (i = 0; i < nspares; i++) {
3040			VERIFY(nvlist_lookup_uint64(spares[i],
3041			    ZPOOL_CONFIG_GUID, &guid) == 0);
3042			if (spa_spare_exists(guid, &pool, NULL) &&
3043			    pool != 0ULL) {
3044				VERIFY(nvlist_lookup_uint64_array(
3045				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
3046				    (uint64_t **)&vs, &vsc) == 0);
3047				vs->vs_state = VDEV_STATE_CANT_OPEN;
3048				vs->vs_aux = VDEV_AUX_SPARED;
3049			}
3050		}
3051	}
3052}
3053
3054/*
3055 * Add l2cache device information to the nvlist, including vdev stats.
3056 */
3057static void
3058spa_add_l2cache(spa_t *spa, nvlist_t *config)
3059{
3060	nvlist_t **l2cache;
3061	uint_t i, j, nl2cache;
3062	nvlist_t *nvroot;
3063	uint64_t guid;
3064	vdev_t *vd;
3065	vdev_stat_t *vs;
3066	uint_t vsc;
3067
3068	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3069
3070	if (spa->spa_l2cache.sav_count == 0)
3071		return;
3072
3073	VERIFY(nvlist_lookup_nvlist(config,
3074	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3075	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3076	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3077	if (nl2cache != 0) {
3078		VERIFY(nvlist_add_nvlist_array(nvroot,
3079		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3080		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3081		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3082
3083		/*
3084		 * Update level 2 cache device stats.
3085		 */
3086
3087		for (i = 0; i < nl2cache; i++) {
3088			VERIFY(nvlist_lookup_uint64(l2cache[i],
3089			    ZPOOL_CONFIG_GUID, &guid) == 0);
3090
3091			vd = NULL;
3092			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3093				if (guid ==
3094				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3095					vd = spa->spa_l2cache.sav_vdevs[j];
3096					break;
3097				}
3098			}
3099			ASSERT(vd != NULL);
3100
3101			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3102			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3103			    == 0);
3104			vdev_get_stats(vd, vs);
3105		}
3106	}
3107}
3108
3109static void
3110spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3111{
3112	nvlist_t *features;
3113	zap_cursor_t zc;
3114	zap_attribute_t za;
3115
3116	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3117	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3118
3119	if (spa->spa_feat_for_read_obj != 0) {
3120		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3121		    spa->spa_feat_for_read_obj);
3122		    zap_cursor_retrieve(&zc, &za) == 0;
3123		    zap_cursor_advance(&zc)) {
3124			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3125			    za.za_num_integers == 1);
3126			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3127			    za.za_first_integer));
3128		}
3129		zap_cursor_fini(&zc);
3130	}
3131
3132	if (spa->spa_feat_for_write_obj != 0) {
3133		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3134		    spa->spa_feat_for_write_obj);
3135		    zap_cursor_retrieve(&zc, &za) == 0;
3136		    zap_cursor_advance(&zc)) {
3137			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3138			    za.za_num_integers == 1);
3139			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3140			    za.za_first_integer));
3141		}
3142		zap_cursor_fini(&zc);
3143	}
3144
3145	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3146	    features) == 0);
3147	nvlist_free(features);
3148}
3149
3150int
3151spa_get_stats(const char *name, nvlist_t **config,
3152    char *altroot, size_t buflen)
3153{
3154	int error;
3155	spa_t *spa;
3156
3157	*config = NULL;
3158	error = spa_open_common(name, &spa, FTAG, NULL, config);
3159
3160	if (spa != NULL) {
3161		/*
3162		 * This still leaves a window of inconsistency where the spares
3163		 * or l2cache devices could change and the config would be
3164		 * self-inconsistent.
3165		 */
3166		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3167
3168		if (*config != NULL) {
3169			uint64_t loadtimes[2];
3170
3171			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3172			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3173			VERIFY(nvlist_add_uint64_array(*config,
3174			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3175
3176			VERIFY(nvlist_add_uint64(*config,
3177			    ZPOOL_CONFIG_ERRCOUNT,
3178			    spa_get_errlog_size(spa)) == 0);
3179
3180			if (spa_suspended(spa))
3181				VERIFY(nvlist_add_uint64(*config,
3182				    ZPOOL_CONFIG_SUSPENDED,
3183				    spa->spa_failmode) == 0);
3184
3185			spa_add_spares(spa, *config);
3186			spa_add_l2cache(spa, *config);
3187			spa_add_feature_stats(spa, *config);
3188		}
3189	}
3190
3191	/*
3192	 * We want to get the alternate root even for faulted pools, so we cheat
3193	 * and call spa_lookup() directly.
3194	 */
3195	if (altroot) {
3196		if (spa == NULL) {
3197			mutex_enter(&spa_namespace_lock);
3198			spa = spa_lookup(name);
3199			if (spa)
3200				spa_altroot(spa, altroot, buflen);
3201			else
3202				altroot[0] = '\0';
3203			spa = NULL;
3204			mutex_exit(&spa_namespace_lock);
3205		} else {
3206			spa_altroot(spa, altroot, buflen);
3207		}
3208	}
3209
3210	if (spa != NULL) {
3211		spa_config_exit(spa, SCL_CONFIG, FTAG);
3212		spa_close(spa, FTAG);
3213	}
3214
3215	return (error);
3216}
3217
3218/*
3219 * Validate that the auxiliary device array is well formed.  We must have an
3220 * array of nvlists, each which describes a valid leaf vdev.  If this is an
3221 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3222 * specified, as long as they are well-formed.
3223 */
3224static int
3225spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3226    spa_aux_vdev_t *sav, const char *config, uint64_t version,
3227    vdev_labeltype_t label)
3228{
3229	nvlist_t **dev;
3230	uint_t i, ndev;
3231	vdev_t *vd;
3232	int error;
3233
3234	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3235
3236	/*
3237	 * It's acceptable to have no devs specified.
3238	 */
3239	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3240		return (0);
3241
3242	if (ndev == 0)
3243		return (SET_ERROR(EINVAL));
3244
3245	/*
3246	 * Make sure the pool is formatted with a version that supports this
3247	 * device type.
3248	 */
3249	if (spa_version(spa) < version)
3250		return (SET_ERROR(ENOTSUP));
3251
3252	/*
3253	 * Set the pending device list so we correctly handle device in-use
3254	 * checking.
3255	 */
3256	sav->sav_pending = dev;
3257	sav->sav_npending = ndev;
3258
3259	for (i = 0; i < ndev; i++) {
3260		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3261		    mode)) != 0)
3262			goto out;
3263
3264		if (!vd->vdev_ops->vdev_op_leaf) {
3265			vdev_free(vd);
3266			error = SET_ERROR(EINVAL);
3267			goto out;
3268		}
3269
3270		/*
3271		 * The L2ARC currently only supports disk devices in
3272		 * kernel context.  For user-level testing, we allow it.
3273		 */
3274#ifdef _KERNEL
3275		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3276		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3277			error = SET_ERROR(ENOTBLK);
3278			vdev_free(vd);
3279			goto out;
3280		}
3281#endif
3282		vd->vdev_top = vd;
3283
3284		if ((error = vdev_open(vd)) == 0 &&
3285		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
3286			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3287			    vd->vdev_guid) == 0);
3288		}
3289
3290		vdev_free(vd);
3291
3292		if (error &&
3293		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3294			goto out;
3295		else
3296			error = 0;
3297	}
3298
3299out:
3300	sav->sav_pending = NULL;
3301	sav->sav_npending = 0;
3302	return (error);
3303}
3304
3305static int
3306spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3307{
3308	int error;
3309
3310	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3311
3312	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3313	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3314	    VDEV_LABEL_SPARE)) != 0) {
3315		return (error);
3316	}
3317
3318	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3319	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3320	    VDEV_LABEL_L2CACHE));
3321}
3322
3323static void
3324spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3325    const char *config)
3326{
3327	int i;
3328
3329	if (sav->sav_config != NULL) {
3330		nvlist_t **olddevs;
3331		uint_t oldndevs;
3332		nvlist_t **newdevs;
3333
3334		/*
3335		 * Generate new dev list by concatentating with the
3336		 * current dev list.
3337		 */
3338		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3339		    &olddevs, &oldndevs) == 0);
3340
3341		newdevs = kmem_alloc(sizeof (void *) *
3342		    (ndevs + oldndevs), KM_SLEEP);
3343		for (i = 0; i < oldndevs; i++)
3344			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3345			    KM_SLEEP) == 0);
3346		for (i = 0; i < ndevs; i++)
3347			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3348			    KM_SLEEP) == 0);
3349
3350		VERIFY(nvlist_remove(sav->sav_config, config,
3351		    DATA_TYPE_NVLIST_ARRAY) == 0);
3352
3353		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3354		    config, newdevs, ndevs + oldndevs) == 0);
3355		for (i = 0; i < oldndevs + ndevs; i++)
3356			nvlist_free(newdevs[i]);
3357		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3358	} else {
3359		/*
3360		 * Generate a new dev list.
3361		 */
3362		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3363		    KM_SLEEP) == 0);
3364		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3365		    devs, ndevs) == 0);
3366	}
3367}
3368
3369/*
3370 * Stop and drop level 2 ARC devices
3371 */
3372void
3373spa_l2cache_drop(spa_t *spa)
3374{
3375	vdev_t *vd;
3376	int i;
3377	spa_aux_vdev_t *sav = &spa->spa_l2cache;
3378
3379	for (i = 0; i < sav->sav_count; i++) {
3380		uint64_t pool;
3381
3382		vd = sav->sav_vdevs[i];
3383		ASSERT(vd != NULL);
3384
3385		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3386		    pool != 0ULL && l2arc_vdev_present(vd))
3387			l2arc_remove_vdev(vd);
3388	}
3389}
3390
3391/*
3392 * Pool Creation
3393 */
3394int
3395spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3396    nvlist_t *zplprops)
3397{
3398	spa_t *spa;
3399	char *altroot = NULL;
3400	vdev_t *rvd;
3401	dsl_pool_t *dp;
3402	dmu_tx_t *tx;
3403	int error = 0;
3404	uint64_t txg = TXG_INITIAL;
3405	nvlist_t **spares, **l2cache;
3406	uint_t nspares, nl2cache;
3407	uint64_t version, obj;
3408	boolean_t has_features;
3409
3410	/*
3411	 * If this pool already exists, return failure.
3412	 */
3413	mutex_enter(&spa_namespace_lock);
3414	if (spa_lookup(pool) != NULL) {
3415		mutex_exit(&spa_namespace_lock);
3416		return (SET_ERROR(EEXIST));
3417	}
3418
3419	/*
3420	 * Allocate a new spa_t structure.
3421	 */
3422	(void) nvlist_lookup_string(props,
3423	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3424	spa = spa_add(pool, NULL, altroot);
3425	spa_activate(spa, spa_mode_global);
3426
3427	if (props && (error = spa_prop_validate(spa, props))) {
3428		spa_deactivate(spa);
3429		spa_remove(spa);
3430		mutex_exit(&spa_namespace_lock);
3431		return (error);
3432	}
3433
3434	has_features = B_FALSE;
3435	for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3436	    elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3437		if (zpool_prop_feature(nvpair_name(elem)))
3438			has_features = B_TRUE;
3439	}
3440
3441	if (has_features || nvlist_lookup_uint64(props,
3442	    zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3443		version = SPA_VERSION;
3444	}
3445	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3446
3447	spa->spa_first_txg = txg;
3448	spa->spa_uberblock.ub_txg = txg - 1;
3449	spa->spa_uberblock.ub_version = version;
3450	spa->spa_ubsync = spa->spa_uberblock;
3451
3452	/*
3453	 * Create "The Godfather" zio to hold all async IOs
3454	 */
3455	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3456	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3457
3458	/*
3459	 * Create the root vdev.
3460	 */
3461	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3462
3463	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3464
3465	ASSERT(error != 0 || rvd != NULL);
3466	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3467
3468	if (error == 0 && !zfs_allocatable_devs(nvroot))
3469		error = SET_ERROR(EINVAL);
3470
3471	if (error == 0 &&
3472	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3473	    (error = spa_validate_aux(spa, nvroot, txg,
3474	    VDEV_ALLOC_ADD)) == 0) {
3475		for (int c = 0; c < rvd->vdev_children; c++) {
3476			vdev_metaslab_set_size(rvd->vdev_child[c]);
3477			vdev_expand(rvd->vdev_child[c], txg);
3478		}
3479	}
3480
3481	spa_config_exit(spa, SCL_ALL, FTAG);
3482
3483	if (error != 0) {
3484		spa_unload(spa);
3485		spa_deactivate(spa);
3486		spa_remove(spa);
3487		mutex_exit(&spa_namespace_lock);
3488		return (error);
3489	}
3490
3491	/*
3492	 * Get the list of spares, if specified.
3493	 */
3494	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3495	    &spares, &nspares) == 0) {
3496		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3497		    KM_SLEEP) == 0);
3498		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3499		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3500		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3501		spa_load_spares(spa);
3502		spa_config_exit(spa, SCL_ALL, FTAG);
3503		spa->spa_spares.sav_sync = B_TRUE;
3504	}
3505
3506	/*
3507	 * Get the list of level 2 cache devices, if specified.
3508	 */
3509	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3510	    &l2cache, &nl2cache) == 0) {
3511		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3512		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3513		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3514		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3515		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3516		spa_load_l2cache(spa);
3517		spa_config_exit(spa, SCL_ALL, FTAG);
3518		spa->spa_l2cache.sav_sync = B_TRUE;
3519	}
3520
3521	spa->spa_is_initializing = B_TRUE;
3522	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3523	spa->spa_meta_objset = dp->dp_meta_objset;
3524	spa->spa_is_initializing = B_FALSE;
3525
3526	/*
3527	 * Create DDTs (dedup tables).
3528	 */
3529	ddt_create(spa);
3530
3531	spa_update_dspace(spa);
3532
3533	tx = dmu_tx_create_assigned(dp, txg);
3534
3535	/*
3536	 * Create the pool config object.
3537	 */
3538	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3539	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3540	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3541
3542	if (zap_add(spa->spa_meta_objset,
3543	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3544	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3545		cmn_err(CE_PANIC, "failed to add pool config");
3546	}
3547
3548	if (spa_version(spa) >= SPA_VERSION_FEATURES)
3549		spa_feature_create_zap_objects(spa, tx);
3550
3551	if (zap_add(spa->spa_meta_objset,
3552	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3553	    sizeof (uint64_t), 1, &version, tx) != 0) {
3554		cmn_err(CE_PANIC, "failed to add pool version");
3555	}
3556
3557	/* Newly created pools with the right version are always deflated. */
3558	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3559		spa->spa_deflate = TRUE;
3560		if (zap_add(spa->spa_meta_objset,
3561		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3562		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3563			cmn_err(CE_PANIC, "failed to add deflate");
3564		}
3565	}
3566
3567	/*
3568	 * Create the deferred-free bpobj.  Turn off compression
3569	 * because sync-to-convergence takes longer if the blocksize
3570	 * keeps changing.
3571	 */
3572	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3573	dmu_object_set_compress(spa->spa_meta_objset, obj,
3574	    ZIO_COMPRESS_OFF, tx);
3575	if (zap_add(spa->spa_meta_objset,
3576	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3577	    sizeof (uint64_t), 1, &obj, tx) != 0) {
3578		cmn_err(CE_PANIC, "failed to add bpobj");
3579	}
3580	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3581	    spa->spa_meta_objset, obj));
3582
3583	/*
3584	 * Create the pool's history object.
3585	 */
3586	if (version >= SPA_VERSION_ZPOOL_HISTORY)
3587		spa_history_create_obj(spa, tx);
3588
3589	/*
3590	 * Set pool properties.
3591	 */
3592	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3593	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3594	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3595	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3596
3597	if (props != NULL) {
3598		spa_configfile_set(spa, props, B_FALSE);
3599		spa_sync_props(props, tx);
3600	}
3601
3602	dmu_tx_commit(tx);
3603
3604	spa->spa_sync_on = B_TRUE;
3605	txg_sync_start(spa->spa_dsl_pool);
3606
3607	/*
3608	 * We explicitly wait for the first transaction to complete so that our
3609	 * bean counters are appropriately updated.
3610	 */
3611	txg_wait_synced(spa->spa_dsl_pool, txg);
3612
3613	spa_config_sync(spa, B_FALSE, B_TRUE);
3614
3615	spa_history_log_version(spa, "create");
3616
3617	spa->spa_minref = refcount_count(&spa->spa_refcount);
3618
3619	mutex_exit(&spa_namespace_lock);
3620
3621	return (0);
3622}
3623
3624#ifdef _KERNEL
3625/*
3626 * Get the root pool information from the root disk, then import the root pool
3627 * during the system boot up time.
3628 */
3629extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3630
3631static nvlist_t *
3632spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3633{
3634	nvlist_t *config;
3635	nvlist_t *nvtop, *nvroot;
3636	uint64_t pgid;
3637
3638	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3639		return (NULL);
3640
3641	/*
3642	 * Add this top-level vdev to the child array.
3643	 */
3644	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3645	    &nvtop) == 0);
3646	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3647	    &pgid) == 0);
3648	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3649
3650	/*
3651	 * Put this pool's top-level vdevs into a root vdev.
3652	 */
3653	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3654	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3655	    VDEV_TYPE_ROOT) == 0);
3656	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3657	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3658	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3659	    &nvtop, 1) == 0);
3660
3661	/*
3662	 * Replace the existing vdev_tree with the new root vdev in
3663	 * this pool's configuration (remove the old, add the new).
3664	 */
3665	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3666	nvlist_free(nvroot);
3667	return (config);
3668}
3669
3670/*
3671 * Walk the vdev tree and see if we can find a device with "better"
3672 * configuration. A configuration is "better" if the label on that
3673 * device has a more recent txg.
3674 */
3675static void
3676spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3677{
3678	for (int c = 0; c < vd->vdev_children; c++)
3679		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3680
3681	if (vd->vdev_ops->vdev_op_leaf) {
3682		nvlist_t *label;
3683		uint64_t label_txg;
3684
3685		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3686		    &label) != 0)
3687			return;
3688
3689		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3690		    &label_txg) == 0);
3691
3692		/*
3693		 * Do we have a better boot device?
3694		 */
3695		if (label_txg > *txg) {
3696			*txg = label_txg;
3697			*avd = vd;
3698		}
3699		nvlist_free(label);
3700	}
3701}
3702
3703/*
3704 * Import a root pool.
3705 *
3706 * For x86. devpath_list will consist of devid and/or physpath name of
3707 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3708 * The GRUB "findroot" command will return the vdev we should boot.
3709 *
3710 * For Sparc, devpath_list consists the physpath name of the booting device
3711 * no matter the rootpool is a single device pool or a mirrored pool.
3712 * e.g.
3713 *	"/pci@1f,0/ide@d/disk@0,0:a"
3714 */
3715int
3716spa_import_rootpool(char *devpath, char *devid)
3717{
3718	spa_t *spa;
3719	vdev_t *rvd, *bvd, *avd = NULL;
3720	nvlist_t *config, *nvtop;
3721	uint64_t guid, txg;
3722	char *pname;
3723	int error;
3724
3725	/*
3726	 * Read the label from the boot device and generate a configuration.
3727	 */
3728	config = spa_generate_rootconf(devpath, devid, &guid);
3729#if defined(_OBP) && defined(_KERNEL)
3730	if (config == NULL) {
3731		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3732			/* iscsi boot */
3733			get_iscsi_bootpath_phy(devpath);
3734			config = spa_generate_rootconf(devpath, devid, &guid);
3735		}
3736	}
3737#endif
3738	if (config == NULL) {
3739		cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3740		    devpath);
3741		return (SET_ERROR(EIO));
3742	}
3743
3744	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3745	    &pname) == 0);
3746	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3747
3748	mutex_enter(&spa_namespace_lock);
3749	if ((spa = spa_lookup(pname)) != NULL) {
3750		/*
3751		 * Remove the existing root pool from the namespace so that we
3752		 * can replace it with the correct config we just read in.
3753		 */
3754		spa_remove(spa);
3755	}
3756
3757	spa = spa_add(pname, config, NULL);
3758	spa->spa_is_root = B_TRUE;
3759	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3760
3761	/*
3762	 * Build up a vdev tree based on the boot device's label config.
3763	 */
3764	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3765	    &nvtop) == 0);
3766	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3767	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3768	    VDEV_ALLOC_ROOTPOOL);
3769	spa_config_exit(spa, SCL_ALL, FTAG);
3770	if (error) {
3771		mutex_exit(&spa_namespace_lock);
3772		nvlist_free(config);
3773		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3774		    pname);
3775		return (error);
3776	}
3777
3778	/*
3779	 * Get the boot vdev.
3780	 */
3781	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3782		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3783		    (u_longlong_t)guid);
3784		error = SET_ERROR(ENOENT);
3785		goto out;
3786	}
3787
3788	/*
3789	 * Determine if there is a better boot device.
3790	 */
3791	avd = bvd;
3792	spa_alt_rootvdev(rvd, &avd, &txg);
3793	if (avd != bvd) {
3794		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3795		    "try booting from '%s'", avd->vdev_path);
3796		error = SET_ERROR(EINVAL);
3797		goto out;
3798	}
3799
3800	/*
3801	 * If the boot device is part of a spare vdev then ensure that
3802	 * we're booting off the active spare.
3803	 */
3804	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3805	    !bvd->vdev_isspare) {
3806		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3807		    "try booting from '%s'",
3808		    bvd->vdev_parent->
3809		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3810		error = SET_ERROR(EINVAL);
3811		goto out;
3812	}
3813
3814	error = 0;
3815out:
3816	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3817	vdev_free(rvd);
3818	spa_config_exit(spa, SCL_ALL, FTAG);
3819	mutex_exit(&spa_namespace_lock);
3820
3821	nvlist_free(config);
3822	return (error);
3823}
3824
3825#endif
3826
3827/*
3828 * Import a non-root pool into the system.
3829 */
3830int
3831spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3832{
3833	spa_t *spa;
3834	char *altroot = NULL;
3835	spa_load_state_t state = SPA_LOAD_IMPORT;
3836	zpool_rewind_policy_t policy;
3837	uint64_t mode = spa_mode_global;
3838	uint64_t readonly = B_FALSE;
3839	int error;
3840	nvlist_t *nvroot;
3841	nvlist_t **spares, **l2cache;
3842	uint_t nspares, nl2cache;
3843
3844	/*
3845	 * If a pool with this name exists, return failure.
3846	 */
3847	mutex_enter(&spa_namespace_lock);
3848	if (spa_lookup(pool) != NULL) {
3849		mutex_exit(&spa_namespace_lock);
3850		return (SET_ERROR(EEXIST));
3851	}
3852
3853	/*
3854	 * Create and initialize the spa structure.
3855	 */
3856	(void) nvlist_lookup_string(props,
3857	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3858	(void) nvlist_lookup_uint64(props,
3859	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3860	if (readonly)
3861		mode = FREAD;
3862	spa = spa_add(pool, config, altroot);
3863	spa->spa_import_flags = flags;
3864
3865	/*
3866	 * Verbatim import - Take a pool and insert it into the namespace
3867	 * as if it had been loaded at boot.
3868	 */
3869	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3870		if (props != NULL)
3871			spa_configfile_set(spa, props, B_FALSE);
3872
3873		spa_config_sync(spa, B_FALSE, B_TRUE);
3874
3875		mutex_exit(&spa_namespace_lock);
3876		return (0);
3877	}
3878
3879	spa_activate(spa, mode);
3880
3881	/*
3882	 * Don't start async tasks until we know everything is healthy.
3883	 */
3884	spa_async_suspend(spa);
3885
3886	zpool_get_rewind_policy(config, &policy);
3887	if (policy.zrp_request & ZPOOL_DO_REWIND)
3888		state = SPA_LOAD_RECOVER;
3889
3890	/*
3891	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3892	 * because the user-supplied config is actually the one to trust when
3893	 * doing an import.
3894	 */
3895	if (state != SPA_LOAD_RECOVER)
3896		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3897
3898	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3899	    policy.zrp_request);
3900
3901	/*
3902	 * Propagate anything learned while loading the pool and pass it
3903	 * back to caller (i.e. rewind info, missing devices, etc).
3904	 */
3905	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3906	    spa->spa_load_info) == 0);
3907
3908	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3909	/*
3910	 * Toss any existing sparelist, as it doesn't have any validity
3911	 * anymore, and conflicts with spa_has_spare().
3912	 */
3913	if (spa->spa_spares.sav_config) {
3914		nvlist_free(spa->spa_spares.sav_config);
3915		spa->spa_spares.sav_config = NULL;
3916		spa_load_spares(spa);
3917	}
3918	if (spa->spa_l2cache.sav_config) {
3919		nvlist_free(spa->spa_l2cache.sav_config);
3920		spa->spa_l2cache.sav_config = NULL;
3921		spa_load_l2cache(spa);
3922	}
3923
3924	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3925	    &nvroot) == 0);
3926	if (error == 0)
3927		error = spa_validate_aux(spa, nvroot, -1ULL,
3928		    VDEV_ALLOC_SPARE);
3929	if (error == 0)
3930		error = spa_validate_aux(spa, nvroot, -1ULL,
3931		    VDEV_ALLOC_L2CACHE);
3932	spa_config_exit(spa, SCL_ALL, FTAG);
3933
3934	if (props != NULL)
3935		spa_configfile_set(spa, props, B_FALSE);
3936
3937	if (error != 0 || (props && spa_writeable(spa) &&
3938	    (error = spa_prop_set(spa, props)))) {
3939		spa_unload(spa);
3940		spa_deactivate(spa);
3941		spa_remove(spa);
3942		mutex_exit(&spa_namespace_lock);
3943		return (error);
3944	}
3945
3946	spa_async_resume(spa);
3947
3948	/*
3949	 * Override any spares and level 2 cache devices as specified by
3950	 * the user, as these may have correct device names/devids, etc.
3951	 */
3952	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3953	    &spares, &nspares) == 0) {
3954		if (spa->spa_spares.sav_config)
3955			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3956			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3957		else
3958			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3959			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3960		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3961		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3962		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3963		spa_load_spares(spa);
3964		spa_config_exit(spa, SCL_ALL, FTAG);
3965		spa->spa_spares.sav_sync = B_TRUE;
3966	}
3967	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3968	    &l2cache, &nl2cache) == 0) {
3969		if (spa->spa_l2cache.sav_config)
3970			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3971			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3972		else
3973			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3974			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3975		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3976		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3977		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3978		spa_load_l2cache(spa);
3979		spa_config_exit(spa, SCL_ALL, FTAG);
3980		spa->spa_l2cache.sav_sync = B_TRUE;
3981	}
3982
3983	/*
3984	 * Check for any removed devices.
3985	 */
3986	if (spa->spa_autoreplace) {
3987		spa_aux_check_removed(&spa->spa_spares);
3988		spa_aux_check_removed(&spa->spa_l2cache);
3989	}
3990
3991	if (spa_writeable(spa)) {
3992		/*
3993		 * Update the config cache to include the newly-imported pool.
3994		 */
3995		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3996	}
3997
3998	/*
3999	 * It's possible that the pool was expanded while it was exported.
4000	 * We kick off an async task to handle this for us.
4001	 */
4002	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4003
4004	mutex_exit(&spa_namespace_lock);
4005	spa_history_log_version(spa, "import");
4006
4007	return (0);
4008}
4009
4010nvlist_t *
4011spa_tryimport(nvlist_t *tryconfig)
4012{
4013	nvlist_t *config = NULL;
4014	char *poolname;
4015	spa_t *spa;
4016	uint64_t state;
4017	int error;
4018
4019	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4020		return (NULL);
4021
4022	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4023		return (NULL);
4024
4025	/*
4026	 * Create and initialize the spa structure.
4027	 */
4028	mutex_enter(&spa_namespace_lock);
4029	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4030	spa_activate(spa, FREAD);
4031
4032	/*
4033	 * Pass off the heavy lifting to spa_load().
4034	 * Pass TRUE for mosconfig because the user-supplied config
4035	 * is actually the one to trust when doing an import.
4036	 */
4037	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4038
4039	/*
4040	 * If 'tryconfig' was at least parsable, return the current config.
4041	 */
4042	if (spa->spa_root_vdev != NULL) {
4043		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4044		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4045		    poolname) == 0);
4046		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4047		    state) == 0);
4048		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4049		    spa->spa_uberblock.ub_timestamp) == 0);
4050		VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4051		    spa->spa_load_info) == 0);
4052
4053		/*
4054		 * If the bootfs property exists on this pool then we
4055		 * copy it out so that external consumers can tell which
4056		 * pools are bootable.
4057		 */
4058		if ((!error || error == EEXIST) && spa->spa_bootfs) {
4059			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4060
4061			/*
4062			 * We have to play games with the name since the
4063			 * pool was opened as TRYIMPORT_NAME.
4064			 */
4065			if (dsl_dsobj_to_dsname(spa_name(spa),
4066			    spa->spa_bootfs, tmpname) == 0) {
4067				char *cp;
4068				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4069
4070				cp = strchr(tmpname, '/');
4071				if (cp == NULL) {
4072					(void) strlcpy(dsname, tmpname,
4073					    MAXPATHLEN);
4074				} else {
4075					(void) snprintf(dsname, MAXPATHLEN,
4076					    "%s/%s", poolname, ++cp);
4077				}
4078				VERIFY(nvlist_add_string(config,
4079				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4080				kmem_free(dsname, MAXPATHLEN);
4081			}
4082			kmem_free(tmpname, MAXPATHLEN);
4083		}
4084
4085		/*
4086		 * Add the list of hot spares and level 2 cache devices.
4087		 */
4088		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4089		spa_add_spares(spa, config);
4090		spa_add_l2cache(spa, config);
4091		spa_config_exit(spa, SCL_CONFIG, FTAG);
4092	}
4093
4094	spa_unload(spa);
4095	spa_deactivate(spa);
4096	spa_remove(spa);
4097	mutex_exit(&spa_namespace_lock);
4098
4099	return (config);
4100}
4101
4102/*
4103 * Pool export/destroy
4104 *
4105 * The act of destroying or exporting a pool is very simple.  We make sure there
4106 * is no more pending I/O and any references to the pool are gone.  Then, we
4107 * update the pool state and sync all the labels to disk, removing the
4108 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4109 * we don't sync the labels or remove the configuration cache.
4110 */
4111static int
4112spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4113    boolean_t force, boolean_t hardforce)
4114{
4115	spa_t *spa;
4116
4117	if (oldconfig)
4118		*oldconfig = NULL;
4119
4120	if (!(spa_mode_global & FWRITE))
4121		return (SET_ERROR(EROFS));
4122
4123	mutex_enter(&spa_namespace_lock);
4124	if ((spa = spa_lookup(pool)) == NULL) {
4125		mutex_exit(&spa_namespace_lock);
4126		return (SET_ERROR(ENOENT));
4127	}
4128
4129	/*
4130	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4131	 * reacquire the namespace lock, and see if we can export.
4132	 */
4133	spa_open_ref(spa, FTAG);
4134	mutex_exit(&spa_namespace_lock);
4135	spa_async_suspend(spa);
4136	mutex_enter(&spa_namespace_lock);
4137	spa_close(spa, FTAG);
4138
4139	/*
4140	 * The pool will be in core if it's openable,
4141	 * in which case we can modify its state.
4142	 */
4143	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4144		/*
4145		 * Objsets may be open only because they're dirty, so we
4146		 * have to force it to sync before checking spa_refcnt.
4147		 */
4148		txg_wait_synced(spa->spa_dsl_pool, 0);
4149
4150		/*
4151		 * A pool cannot be exported or destroyed if there are active
4152		 * references.  If we are resetting a pool, allow references by
4153		 * fault injection handlers.
4154		 */
4155		if (!spa_refcount_zero(spa) ||
4156		    (spa->spa_inject_ref != 0 &&
4157		    new_state != POOL_STATE_UNINITIALIZED)) {
4158			spa_async_resume(spa);
4159			mutex_exit(&spa_namespace_lock);
4160			return (SET_ERROR(EBUSY));
4161		}
4162
4163		/*
4164		 * A pool cannot be exported if it has an active shared spare.
4165		 * This is to prevent other pools stealing the active spare
4166		 * from an exported pool. At user's own will, such pool can
4167		 * be forcedly exported.
4168		 */
4169		if (!force && new_state == POOL_STATE_EXPORTED &&
4170		    spa_has_active_shared_spare(spa)) {
4171			spa_async_resume(spa);
4172			mutex_exit(&spa_namespace_lock);
4173			return (SET_ERROR(EXDEV));
4174		}
4175
4176		/*
4177		 * We want this to be reflected on every label,
4178		 * so mark them all dirty.  spa_unload() will do the
4179		 * final sync that pushes these changes out.
4180		 */
4181		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4182			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4183			spa->spa_state = new_state;
4184			spa->spa_final_txg = spa_last_synced_txg(spa) +
4185			    TXG_DEFER_SIZE + 1;
4186			vdev_config_dirty(spa->spa_root_vdev);
4187			spa_config_exit(spa, SCL_ALL, FTAG);
4188		}
4189	}
4190
4191	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4192
4193	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4194		spa_unload(spa);
4195		spa_deactivate(spa);
4196	}
4197
4198	if (oldconfig && spa->spa_config)
4199		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4200
4201	if (new_state != POOL_STATE_UNINITIALIZED) {
4202		if (!hardforce)
4203			spa_config_sync(spa, B_TRUE, B_TRUE);
4204		spa_remove(spa);
4205	}
4206	mutex_exit(&spa_namespace_lock);
4207
4208	return (0);
4209}
4210
4211/*
4212 * Destroy a storage pool.
4213 */
4214int
4215spa_destroy(char *pool)
4216{
4217	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4218	    B_FALSE, B_FALSE));
4219}
4220
4221/*
4222 * Export a storage pool.
4223 */
4224int
4225spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4226    boolean_t hardforce)
4227{
4228	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4229	    force, hardforce));
4230}
4231
4232/*
4233 * Similar to spa_export(), this unloads the spa_t without actually removing it
4234 * from the namespace in any way.
4235 */
4236int
4237spa_reset(char *pool)
4238{
4239	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4240	    B_FALSE, B_FALSE));
4241}
4242
4243/*
4244 * ==========================================================================
4245 * Device manipulation
4246 * ==========================================================================
4247 */
4248
4249/*
4250 * Add a device to a storage pool.
4251 */
4252int
4253spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4254{
4255	uint64_t txg, id;
4256	int error;
4257	vdev_t *rvd = spa->spa_root_vdev;
4258	vdev_t *vd, *tvd;
4259	nvlist_t **spares, **l2cache;
4260	uint_t nspares, nl2cache;
4261
4262	ASSERT(spa_writeable(spa));
4263
4264	txg = spa_vdev_enter(spa);
4265
4266	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4267	    VDEV_ALLOC_ADD)) != 0)
4268		return (spa_vdev_exit(spa, NULL, txg, error));
4269
4270	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
4271
4272	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4273	    &nspares) != 0)
4274		nspares = 0;
4275
4276	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4277	    &nl2cache) != 0)
4278		nl2cache = 0;
4279
4280	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4281		return (spa_vdev_exit(spa, vd, txg, EINVAL));
4282
4283	if (vd->vdev_children != 0 &&
4284	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
4285		return (spa_vdev_exit(spa, vd, txg, error));
4286
4287	/*
4288	 * We must validate the spares and l2cache devices after checking the
4289	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
4290	 */
4291	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4292		return (spa_vdev_exit(spa, vd, txg, error));
4293
4294	/*
4295	 * Transfer each new top-level vdev from vd to rvd.
4296	 */
4297	for (int c = 0; c < vd->vdev_children; c++) {
4298
4299		/*
4300		 * Set the vdev id to the first hole, if one exists.
4301		 */
4302		for (id = 0; id < rvd->vdev_children; id++) {
4303			if (rvd->vdev_child[id]->vdev_ishole) {
4304				vdev_free(rvd->vdev_child[id]);
4305				break;
4306			}
4307		}
4308		tvd = vd->vdev_child[c];
4309		vdev_remove_child(vd, tvd);
4310		tvd->vdev_id = id;
4311		vdev_add_child(rvd, tvd);
4312		vdev_config_dirty(tvd);
4313	}
4314
4315	if (nspares != 0) {
4316		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4317		    ZPOOL_CONFIG_SPARES);
4318		spa_load_spares(spa);
4319		spa->spa_spares.sav_sync = B_TRUE;
4320	}
4321
4322	if (nl2cache != 0) {
4323		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4324		    ZPOOL_CONFIG_L2CACHE);
4325		spa_load_l2cache(spa);
4326		spa->spa_l2cache.sav_sync = B_TRUE;
4327	}
4328
4329	/*
4330	 * We have to be careful when adding new vdevs to an existing pool.
4331	 * If other threads start allocating from these vdevs before we
4332	 * sync the config cache, and we lose power, then upon reboot we may
4333	 * fail to open the pool because there are DVAs that the config cache
4334	 * can't translate.  Therefore, we first add the vdevs without
4335	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4336	 * and then let spa_config_update() initialize the new metaslabs.
4337	 *
4338	 * spa_load() checks for added-but-not-initialized vdevs, so that
4339	 * if we lose power at any point in this sequence, the remaining
4340	 * steps will be completed the next time we load the pool.
4341	 */
4342	(void) spa_vdev_exit(spa, vd, txg, 0);
4343
4344	mutex_enter(&spa_namespace_lock);
4345	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4346	mutex_exit(&spa_namespace_lock);
4347
4348	return (0);
4349}
4350
4351/*
4352 * Attach a device to a mirror.  The arguments are the path to any device
4353 * in the mirror, and the nvroot for the new device.  If the path specifies
4354 * a device that is not mirrored, we automatically insert the mirror vdev.
4355 *
4356 * If 'replacing' is specified, the new device is intended to replace the
4357 * existing device; in this case the two devices are made into their own
4358 * mirror using the 'replacing' vdev, which is functionally identical to
4359 * the mirror vdev (it actually reuses all the same ops) but has a few
4360 * extra rules: you can't attach to it after it's been created, and upon
4361 * completion of resilvering, the first disk (the one being replaced)
4362 * is automatically detached.
4363 */
4364int
4365spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4366{
4367	uint64_t txg, dtl_max_txg;
4368	vdev_t *rvd = spa->spa_root_vdev;
4369	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4370	vdev_ops_t *pvops;
4371	char *oldvdpath, *newvdpath;
4372	int newvd_isspare;
4373	int error;
4374
4375	ASSERT(spa_writeable(spa));
4376
4377	txg = spa_vdev_enter(spa);
4378
4379	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4380
4381	if (oldvd == NULL)
4382		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4383
4384	if (!oldvd->vdev_ops->vdev_op_leaf)
4385		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4386
4387	pvd = oldvd->vdev_parent;
4388
4389	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4390	    VDEV_ALLOC_ATTACH)) != 0)
4391		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4392
4393	if (newrootvd->vdev_children != 1)
4394		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4395
4396	newvd = newrootvd->vdev_child[0];
4397
4398	if (!newvd->vdev_ops->vdev_op_leaf)
4399		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4400
4401	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4402		return (spa_vdev_exit(spa, newrootvd, txg, error));
4403
4404	/*
4405	 * Spares can't replace logs
4406	 */
4407	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4408		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4409
4410	if (!replacing) {
4411		/*
4412		 * For attach, the only allowable parent is a mirror or the root
4413		 * vdev.
4414		 */
4415		if (pvd->vdev_ops != &vdev_mirror_ops &&
4416		    pvd->vdev_ops != &vdev_root_ops)
4417			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4418
4419		pvops = &vdev_mirror_ops;
4420	} else {
4421		/*
4422		 * Active hot spares can only be replaced by inactive hot
4423		 * spares.
4424		 */
4425		if (pvd->vdev_ops == &vdev_spare_ops &&
4426		    oldvd->vdev_isspare &&
4427		    !spa_has_spare(spa, newvd->vdev_guid))
4428			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4429
4430		/*
4431		 * If the source is a hot spare, and the parent isn't already a
4432		 * spare, then we want to create a new hot spare.  Otherwise, we
4433		 * want to create a replacing vdev.  The user is not allowed to
4434		 * attach to a spared vdev child unless the 'isspare' state is
4435		 * the same (spare replaces spare, non-spare replaces
4436		 * non-spare).
4437		 */
4438		if (pvd->vdev_ops == &vdev_replacing_ops &&
4439		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4440			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4441		} else if (pvd->vdev_ops == &vdev_spare_ops &&
4442		    newvd->vdev_isspare != oldvd->vdev_isspare) {
4443			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4444		}
4445
4446		if (newvd->vdev_isspare)
4447			pvops = &vdev_spare_ops;
4448		else
4449			pvops = &vdev_replacing_ops;
4450	}
4451
4452	/*
4453	 * Make sure the new device is big enough.
4454	 */
4455	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4456		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4457
4458	/*
4459	 * The new device cannot have a higher alignment requirement
4460	 * than the top-level vdev.
4461	 */
4462	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4463		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4464
4465	/*
4466	 * If this is an in-place replacement, update oldvd's path and devid
4467	 * to make it distinguishable from newvd, and unopenable from now on.
4468	 */
4469	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4470		spa_strfree(oldvd->vdev_path);
4471		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4472		    KM_SLEEP);
4473		(void) sprintf(oldvd->vdev_path, "%s/%s",
4474		    newvd->vdev_path, "old");
4475		if (oldvd->vdev_devid != NULL) {
4476			spa_strfree(oldvd->vdev_devid);
4477			oldvd->vdev_devid = NULL;
4478		}
4479	}
4480
4481	/* mark the device being resilvered */
4482	newvd->vdev_resilver_txg = txg;
4483
4484	/*
4485	 * If the parent is not a mirror, or if we're replacing, insert the new
4486	 * mirror/replacing/spare vdev above oldvd.
4487	 */
4488	if (pvd->vdev_ops != pvops)
4489		pvd = vdev_add_parent(oldvd, pvops);
4490
4491	ASSERT(pvd->vdev_top->vdev_parent == rvd);
4492	ASSERT(pvd->vdev_ops == pvops);
4493	ASSERT(oldvd->vdev_parent == pvd);
4494
4495	/*
4496	 * Extract the new device from its root and add it to pvd.
4497	 */
4498	vdev_remove_child(newrootvd, newvd);
4499	newvd->vdev_id = pvd->vdev_children;
4500	newvd->vdev_crtxg = oldvd->vdev_crtxg;
4501	vdev_add_child(pvd, newvd);
4502
4503	tvd = newvd->vdev_top;
4504	ASSERT(pvd->vdev_top == tvd);
4505	ASSERT(tvd->vdev_parent == rvd);
4506
4507	vdev_config_dirty(tvd);
4508
4509	/*
4510	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4511	 * for any dmu_sync-ed blocks.  It will propagate upward when
4512	 * spa_vdev_exit() calls vdev_dtl_reassess().
4513	 */
4514	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4515
4516	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4517	    dtl_max_txg - TXG_INITIAL);
4518
4519	if (newvd->vdev_isspare) {
4520		spa_spare_activate(newvd);
4521		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4522	}
4523
4524	oldvdpath = spa_strdup(oldvd->vdev_path);
4525	newvdpath = spa_strdup(newvd->vdev_path);
4526	newvd_isspare = newvd->vdev_isspare;
4527
4528	/*
4529	 * Mark newvd's DTL dirty in this txg.
4530	 */
4531	vdev_dirty(tvd, VDD_DTL, newvd, txg);
4532
4533	/*
4534	 * Schedule the resilver to restart in the future. We do this to
4535	 * ensure that dmu_sync-ed blocks have been stitched into the
4536	 * respective datasets.
4537	 */
4538	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4539
4540	/*
4541	 * Commit the config
4542	 */
4543	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4544
4545	spa_history_log_internal(spa, "vdev attach", NULL,
4546	    "%s vdev=%s %s vdev=%s",
4547	    replacing && newvd_isspare ? "spare in" :
4548	    replacing ? "replace" : "attach", newvdpath,
4549	    replacing ? "for" : "to", oldvdpath);
4550
4551	spa_strfree(oldvdpath);
4552	spa_strfree(newvdpath);
4553
4554	if (spa->spa_bootfs)
4555		spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4556
4557	return (0);
4558}
4559
4560/*
4561 * Detach a device from a mirror or replacing vdev.
4562 *
4563 * If 'replace_done' is specified, only detach if the parent
4564 * is a replacing vdev.
4565 */
4566int
4567spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4568{
4569	uint64_t txg;
4570	int error;
4571	vdev_t *rvd = spa->spa_root_vdev;
4572	vdev_t *vd, *pvd, *cvd, *tvd;
4573	boolean_t unspare = B_FALSE;
4574	uint64_t unspare_guid = 0;
4575	char *vdpath;
4576
4577	ASSERT(spa_writeable(spa));
4578
4579	txg = spa_vdev_enter(spa);
4580
4581	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4582
4583	if (vd == NULL)
4584		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4585
4586	if (!vd->vdev_ops->vdev_op_leaf)
4587		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4588
4589	pvd = vd->vdev_parent;
4590
4591	/*
4592	 * If the parent/child relationship is not as expected, don't do it.
4593	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4594	 * vdev that's replacing B with C.  The user's intent in replacing
4595	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4596	 * the replace by detaching C, the expected behavior is to end up
4597	 * M(A,B).  But suppose that right after deciding to detach C,
4598	 * the replacement of B completes.  We would have M(A,C), and then
4599	 * ask to detach C, which would leave us with just A -- not what
4600	 * the user wanted.  To prevent this, we make sure that the
4601	 * parent/child relationship hasn't changed -- in this example,
4602	 * that C's parent is still the replacing vdev R.
4603	 */
4604	if (pvd->vdev_guid != pguid && pguid != 0)
4605		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4606
4607	/*
4608	 * Only 'replacing' or 'spare' vdevs can be replaced.
4609	 */
4610	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4611	    pvd->vdev_ops != &vdev_spare_ops)
4612		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4613
4614	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4615	    spa_version(spa) >= SPA_VERSION_SPARES);
4616
4617	/*
4618	 * Only mirror, replacing, and spare vdevs support detach.
4619	 */
4620	if (pvd->vdev_ops != &vdev_replacing_ops &&
4621	    pvd->vdev_ops != &vdev_mirror_ops &&
4622	    pvd->vdev_ops != &vdev_spare_ops)
4623		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4624
4625	/*
4626	 * If this device has the only valid copy of some data,
4627	 * we cannot safely detach it.
4628	 */
4629	if (vdev_dtl_required(vd))
4630		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4631
4632	ASSERT(pvd->vdev_children >= 2);
4633
4634	/*
4635	 * If we are detaching the second disk from a replacing vdev, then
4636	 * check to see if we changed the original vdev's path to have "/old"
4637	 * at the end in spa_vdev_attach().  If so, undo that change now.
4638	 */
4639	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4640	    vd->vdev_path != NULL) {
4641		size_t len = strlen(vd->vdev_path);
4642
4643		for (int c = 0; c < pvd->vdev_children; c++) {
4644			cvd = pvd->vdev_child[c];
4645
4646			if (cvd == vd || cvd->vdev_path == NULL)
4647				continue;
4648
4649			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4650			    strcmp(cvd->vdev_path + len, "/old") == 0) {
4651				spa_strfree(cvd->vdev_path);
4652				cvd->vdev_path = spa_strdup(vd->vdev_path);
4653				break;
4654			}
4655		}
4656	}
4657
4658	/*
4659	 * If we are detaching the original disk from a spare, then it implies
4660	 * that the spare should become a real disk, and be removed from the
4661	 * active spare list for the pool.
4662	 */
4663	if (pvd->vdev_ops == &vdev_spare_ops &&
4664	    vd->vdev_id == 0 &&
4665	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4666		unspare = B_TRUE;
4667
4668	/*
4669	 * Erase the disk labels so the disk can be used for other things.
4670	 * This must be done after all other error cases are handled,
4671	 * but before we disembowel vd (so we can still do I/O to it).
4672	 * But if we can't do it, don't treat the error as fatal --
4673	 * it may be that the unwritability of the disk is the reason
4674	 * it's being detached!
4675	 */
4676	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4677
4678	/*
4679	 * Remove vd from its parent and compact the parent's children.
4680	 */
4681	vdev_remove_child(pvd, vd);
4682	vdev_compact_children(pvd);
4683
4684	/*
4685	 * Remember one of the remaining children so we can get tvd below.
4686	 */
4687	cvd = pvd->vdev_child[pvd->vdev_children - 1];
4688
4689	/*
4690	 * If we need to remove the remaining child from the list of hot spares,
4691	 * do it now, marking the vdev as no longer a spare in the process.
4692	 * We must do this before vdev_remove_parent(), because that can
4693	 * change the GUID if it creates a new toplevel GUID.  For a similar
4694	 * reason, we must remove the spare now, in the same txg as the detach;
4695	 * otherwise someone could attach a new sibling, change the GUID, and
4696	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4697	 */
4698	if (unspare) {
4699		ASSERT(cvd->vdev_isspare);
4700		spa_spare_remove(cvd);
4701		unspare_guid = cvd->vdev_guid;
4702		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4703		cvd->vdev_unspare = B_TRUE;
4704	}
4705
4706	/*
4707	 * If the parent mirror/replacing vdev only has one child,
4708	 * the parent is no longer needed.  Remove it from the tree.
4709	 */
4710	if (pvd->vdev_children == 1) {
4711		if (pvd->vdev_ops == &vdev_spare_ops)
4712			cvd->vdev_unspare = B_FALSE;
4713		vdev_remove_parent(cvd);
4714	}
4715
4716
4717	/*
4718	 * We don't set tvd until now because the parent we just removed
4719	 * may have been the previous top-level vdev.
4720	 */
4721	tvd = cvd->vdev_top;
4722	ASSERT(tvd->vdev_parent == rvd);
4723
4724	/*
4725	 * Reevaluate the parent vdev state.
4726	 */
4727	vdev_propagate_state(cvd);
4728
4729	/*
4730	 * If the 'autoexpand' property is set on the pool then automatically
4731	 * try to expand the size of the pool. For example if the device we
4732	 * just detached was smaller than the others, it may be possible to
4733	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4734	 * first so that we can obtain the updated sizes of the leaf vdevs.
4735	 */
4736	if (spa->spa_autoexpand) {
4737		vdev_reopen(tvd);
4738		vdev_expand(tvd, txg);
4739	}
4740
4741	vdev_config_dirty(tvd);
4742
4743	/*
4744	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4745	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4746	 * But first make sure we're not on any *other* txg's DTL list, to
4747	 * prevent vd from being accessed after it's freed.
4748	 */
4749	vdpath = spa_strdup(vd->vdev_path);
4750	for (int t = 0; t < TXG_SIZE; t++)
4751		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4752	vd->vdev_detached = B_TRUE;
4753	vdev_dirty(tvd, VDD_DTL, vd, txg);
4754
4755	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4756
4757	/* hang on to the spa before we release the lock */
4758	spa_open_ref(spa, FTAG);
4759
4760	error = spa_vdev_exit(spa, vd, txg, 0);
4761
4762	spa_history_log_internal(spa, "detach", NULL,
4763	    "vdev=%s", vdpath);
4764	spa_strfree(vdpath);
4765
4766	/*
4767	 * If this was the removal of the original device in a hot spare vdev,
4768	 * then we want to go through and remove the device from the hot spare
4769	 * list of every other pool.
4770	 */
4771	if (unspare) {
4772		spa_t *altspa = NULL;
4773
4774		mutex_enter(&spa_namespace_lock);
4775		while ((altspa = spa_next(altspa)) != NULL) {
4776			if (altspa->spa_state != POOL_STATE_ACTIVE ||
4777			    altspa == spa)
4778				continue;
4779
4780			spa_open_ref(altspa, FTAG);
4781			mutex_exit(&spa_namespace_lock);
4782			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4783			mutex_enter(&spa_namespace_lock);
4784			spa_close(altspa, FTAG);
4785		}
4786		mutex_exit(&spa_namespace_lock);
4787
4788		/* search the rest of the vdevs for spares to remove */
4789		spa_vdev_resilver_done(spa);
4790	}
4791
4792	/* all done with the spa; OK to release */
4793	mutex_enter(&spa_namespace_lock);
4794	spa_close(spa, FTAG);
4795	mutex_exit(&spa_namespace_lock);
4796
4797	return (error);
4798}
4799
4800/*
4801 * Split a set of devices from their mirrors, and create a new pool from them.
4802 */
4803int
4804spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4805    nvlist_t *props, boolean_t exp)
4806{
4807	int error = 0;
4808	uint64_t txg, *glist;
4809	spa_t *newspa;
4810	uint_t c, children, lastlog;
4811	nvlist_t **child, *nvl, *tmp;
4812	dmu_tx_t *tx;
4813	char *altroot = NULL;
4814	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4815	boolean_t activate_slog;
4816
4817	ASSERT(spa_writeable(spa));
4818
4819	txg = spa_vdev_enter(spa);
4820
4821	/* clear the log and flush everything up to now */
4822	activate_slog = spa_passivate_log(spa);
4823	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4824	error = spa_offline_log(spa);
4825	txg = spa_vdev_config_enter(spa);
4826
4827	if (activate_slog)
4828		spa_activate_log(spa);
4829
4830	if (error != 0)
4831		return (spa_vdev_exit(spa, NULL, txg, error));
4832
4833	/* check new spa name before going any further */
4834	if (spa_lookup(newname) != NULL)
4835		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4836
4837	/*
4838	 * scan through all the children to ensure they're all mirrors
4839	 */
4840	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4841	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4842	    &children) != 0)
4843		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4844
4845	/* first, check to ensure we've got the right child count */
4846	rvd = spa->spa_root_vdev;
4847	lastlog = 0;
4848	for (c = 0; c < rvd->vdev_children; c++) {
4849		vdev_t *vd = rvd->vdev_child[c];
4850
4851		/* don't count the holes & logs as children */
4852		if (vd->vdev_islog || vd->vdev_ishole) {
4853			if (lastlog == 0)
4854				lastlog = c;
4855			continue;
4856		}
4857
4858		lastlog = 0;
4859	}
4860	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4861		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4862
4863	/* next, ensure no spare or cache devices are part of the split */
4864	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4865	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4866		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4867
4868	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4869	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4870
4871	/* then, loop over each vdev and validate it */
4872	for (c = 0; c < children; c++) {
4873		uint64_t is_hole = 0;
4874
4875		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4876		    &is_hole);
4877
4878		if (is_hole != 0) {
4879			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4880			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4881				continue;
4882			} else {
4883				error = SET_ERROR(EINVAL);
4884				break;
4885			}
4886		}
4887
4888		/* which disk is going to be split? */
4889		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4890		    &glist[c]) != 0) {
4891			error = SET_ERROR(EINVAL);
4892			break;
4893		}
4894
4895		/* look it up in the spa */
4896		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4897		if (vml[c] == NULL) {
4898			error = SET_ERROR(ENODEV);
4899			break;
4900		}
4901
4902		/* make sure there's nothing stopping the split */
4903		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4904		    vml[c]->vdev_islog ||
4905		    vml[c]->vdev_ishole ||
4906		    vml[c]->vdev_isspare ||
4907		    vml[c]->vdev_isl2cache ||
4908		    !vdev_writeable(vml[c]) ||
4909		    vml[c]->vdev_children != 0 ||
4910		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4911		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4912			error = SET_ERROR(EINVAL);
4913			break;
4914		}
4915
4916		if (vdev_dtl_required(vml[c])) {
4917			error = SET_ERROR(EBUSY);
4918			break;
4919		}
4920
4921		/* we need certain info from the top level */
4922		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4923		    vml[c]->vdev_top->vdev_ms_array) == 0);
4924		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4925		    vml[c]->vdev_top->vdev_ms_shift) == 0);
4926		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4927		    vml[c]->vdev_top->vdev_asize) == 0);
4928		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4929		    vml[c]->vdev_top->vdev_ashift) == 0);
4930	}
4931
4932	if (error != 0) {
4933		kmem_free(vml, children * sizeof (vdev_t *));
4934		kmem_free(glist, children * sizeof (uint64_t));
4935		return (spa_vdev_exit(spa, NULL, txg, error));
4936	}
4937
4938	/* stop writers from using the disks */
4939	for (c = 0; c < children; c++) {
4940		if (vml[c] != NULL)
4941			vml[c]->vdev_offline = B_TRUE;
4942	}
4943	vdev_reopen(spa->spa_root_vdev);
4944
4945	/*
4946	 * Temporarily record the splitting vdevs in the spa config.  This
4947	 * will disappear once the config is regenerated.
4948	 */
4949	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4950	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4951	    glist, children) == 0);
4952	kmem_free(glist, children * sizeof (uint64_t));
4953
4954	mutex_enter(&spa->spa_props_lock);
4955	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4956	    nvl) == 0);
4957	mutex_exit(&spa->spa_props_lock);
4958	spa->spa_config_splitting = nvl;
4959	vdev_config_dirty(spa->spa_root_vdev);
4960
4961	/* configure and create the new pool */
4962	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4963	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4964	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4965	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4966	    spa_version(spa)) == 0);
4967	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4968	    spa->spa_config_txg) == 0);
4969	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4970	    spa_generate_guid(NULL)) == 0);
4971	(void) nvlist_lookup_string(props,
4972	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4973
4974	/* add the new pool to the namespace */
4975	newspa = spa_add(newname, config, altroot);
4976	newspa->spa_config_txg = spa->spa_config_txg;
4977	spa_set_log_state(newspa, SPA_LOG_CLEAR);
4978
4979	/* release the spa config lock, retaining the namespace lock */
4980	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4981
4982	if (zio_injection_enabled)
4983		zio_handle_panic_injection(spa, FTAG, 1);
4984
4985	spa_activate(newspa, spa_mode_global);
4986	spa_async_suspend(newspa);
4987
4988	/* create the new pool from the disks of the original pool */
4989	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4990	if (error)
4991		goto out;
4992
4993	/* if that worked, generate a real config for the new pool */
4994	if (newspa->spa_root_vdev != NULL) {
4995		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4996		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4997		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4998		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4999		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5000		    B_TRUE));
5001	}
5002
5003	/* set the props */
5004	if (props != NULL) {
5005		spa_configfile_set(newspa, props, B_FALSE);
5006		error = spa_prop_set(newspa, props);
5007		if (error)
5008			goto out;
5009	}
5010
5011	/* flush everything */
5012	txg = spa_vdev_config_enter(newspa);
5013	vdev_config_dirty(newspa->spa_root_vdev);
5014	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5015
5016	if (zio_injection_enabled)
5017		zio_handle_panic_injection(spa, FTAG, 2);
5018
5019	spa_async_resume(newspa);
5020
5021	/* finally, update the original pool's config */
5022	txg = spa_vdev_config_enter(spa);
5023	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5024	error = dmu_tx_assign(tx, TXG_WAIT);
5025	if (error != 0)
5026		dmu_tx_abort(tx);
5027	for (c = 0; c < children; c++) {
5028		if (vml[c] != NULL) {
5029			vdev_split(vml[c]);
5030			if (error == 0)
5031				spa_history_log_internal(spa, "detach", tx,
5032				    "vdev=%s", vml[c]->vdev_path);
5033			vdev_free(vml[c]);
5034		}
5035	}
5036	vdev_config_dirty(spa->spa_root_vdev);
5037	spa->spa_config_splitting = NULL;
5038	nvlist_free(nvl);
5039	if (error == 0)
5040		dmu_tx_commit(tx);
5041	(void) spa_vdev_exit(spa, NULL, txg, 0);
5042
5043	if (zio_injection_enabled)
5044		zio_handle_panic_injection(spa, FTAG, 3);
5045
5046	/* split is complete; log a history record */
5047	spa_history_log_internal(newspa, "split", NULL,
5048	    "from pool %s", spa_name(spa));
5049
5050	kmem_free(vml, children * sizeof (vdev_t *));
5051
5052	/* if we're not going to mount the filesystems in userland, export */
5053	if (exp)
5054		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5055		    B_FALSE, B_FALSE);
5056
5057	return (error);
5058
5059out:
5060	spa_unload(newspa);
5061	spa_deactivate(newspa);
5062	spa_remove(newspa);
5063
5064	txg = spa_vdev_config_enter(spa);
5065
5066	/* re-online all offlined disks */
5067	for (c = 0; c < children; c++) {
5068		if (vml[c] != NULL)
5069			vml[c]->vdev_offline = B_FALSE;
5070	}
5071	vdev_reopen(spa->spa_root_vdev);
5072
5073	nvlist_free(spa->spa_config_splitting);
5074	spa->spa_config_splitting = NULL;
5075	(void) spa_vdev_exit(spa, NULL, txg, error);
5076
5077	kmem_free(vml, children * sizeof (vdev_t *));
5078	return (error);
5079}
5080
5081static nvlist_t *
5082spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5083{
5084	for (int i = 0; i < count; i++) {
5085		uint64_t guid;
5086
5087		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5088		    &guid) == 0);
5089
5090		if (guid == target_guid)
5091			return (nvpp[i]);
5092	}
5093
5094	return (NULL);
5095}
5096
5097static void
5098spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5099	nvlist_t *dev_to_remove)
5100{
5101	nvlist_t **newdev = NULL;
5102
5103	if (count > 1)
5104		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5105
5106	for (int i = 0, j = 0; i < count; i++) {
5107		if (dev[i] == dev_to_remove)
5108			continue;
5109		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5110	}
5111
5112	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5113	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5114
5115	for (int i = 0; i < count - 1; i++)
5116		nvlist_free(newdev[i]);
5117
5118	if (count > 1)
5119		kmem_free(newdev, (count - 1) * sizeof (void *));
5120}
5121
5122/*
5123 * Evacuate the device.
5124 */
5125static int
5126spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5127{
5128	uint64_t txg;
5129	int error = 0;
5130
5131	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5132	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5133	ASSERT(vd == vd->vdev_top);
5134
5135	/*
5136	 * Evacuate the device.  We don't hold the config lock as writer
5137	 * since we need to do I/O but we do keep the
5138	 * spa_namespace_lock held.  Once this completes the device
5139	 * should no longer have any blocks allocated on it.
5140	 */
5141	if (vd->vdev_islog) {
5142		if (vd->vdev_stat.vs_alloc != 0)
5143			error = spa_offline_log(spa);
5144	} else {
5145		error = SET_ERROR(ENOTSUP);
5146	}
5147
5148	if (error)
5149		return (error);
5150
5151	/*
5152	 * The evacuation succeeded.  Remove any remaining MOS metadata
5153	 * associated with this vdev, and wait for these changes to sync.
5154	 */
5155	ASSERT0(vd->vdev_stat.vs_alloc);
5156	txg = spa_vdev_config_enter(spa);
5157	vd->vdev_removing = B_TRUE;
5158	vdev_dirty_leaves(vd, VDD_DTL, txg);
5159	vdev_config_dirty(vd);
5160	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5161
5162	return (0);
5163}
5164
5165/*
5166 * Complete the removal by cleaning up the namespace.
5167 */
5168static void
5169spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5170{
5171	vdev_t *rvd = spa->spa_root_vdev;
5172	uint64_t id = vd->vdev_id;
5173	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5174
5175	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5176	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5177	ASSERT(vd == vd->vdev_top);
5178
5179	/*
5180	 * Only remove any devices which are empty.
5181	 */
5182	if (vd->vdev_stat.vs_alloc != 0)
5183		return;
5184
5185	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5186
5187	if (list_link_active(&vd->vdev_state_dirty_node))
5188		vdev_state_clean(vd);
5189	if (list_link_active(&vd->vdev_config_dirty_node))
5190		vdev_config_clean(vd);
5191
5192	vdev_free(vd);
5193
5194	if (last_vdev) {
5195		vdev_compact_children(rvd);
5196	} else {
5197		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5198		vdev_add_child(rvd, vd);
5199	}
5200	vdev_config_dirty(rvd);
5201
5202	/*
5203	 * Reassess the health of our root vdev.
5204	 */
5205	vdev_reopen(rvd);
5206}
5207
5208/*
5209 * Remove a device from the pool -
5210 *
5211 * Removing a device from the vdev namespace requires several steps
5212 * and can take a significant amount of time.  As a result we use
5213 * the spa_vdev_config_[enter/exit] functions which allow us to
5214 * grab and release the spa_config_lock while still holding the namespace
5215 * lock.  During each step the configuration is synced out.
5216 *
5217 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5218 * devices.
5219 */
5220int
5221spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5222{
5223	vdev_t *vd;
5224	metaslab_group_t *mg;
5225	nvlist_t **spares, **l2cache, *nv;
5226	uint64_t txg = 0;
5227	uint_t nspares, nl2cache;
5228	int error = 0;
5229	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5230
5231	ASSERT(spa_writeable(spa));
5232
5233	if (!locked)
5234		txg = spa_vdev_enter(spa);
5235
5236	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5237
5238	if (spa->spa_spares.sav_vdevs != NULL &&
5239	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5240	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5241	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5242		/*
5243		 * Only remove the hot spare if it's not currently in use
5244		 * in this pool.
5245		 */
5246		if (vd == NULL || unspare) {
5247			spa_vdev_remove_aux(spa->spa_spares.sav_config,
5248			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5249			spa_load_spares(spa);
5250			spa->spa_spares.sav_sync = B_TRUE;
5251		} else {
5252			error = SET_ERROR(EBUSY);
5253		}
5254	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
5255	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5256	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5257	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5258		/*
5259		 * Cache devices can always be removed.
5260		 */
5261		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5262		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5263		spa_load_l2cache(spa);
5264		spa->spa_l2cache.sav_sync = B_TRUE;
5265	} else if (vd != NULL && vd->vdev_islog) {
5266		ASSERT(!locked);
5267		ASSERT(vd == vd->vdev_top);
5268
5269		mg = vd->vdev_mg;
5270
5271		/*
5272		 * Stop allocating from this vdev.
5273		 */
5274		metaslab_group_passivate(mg);
5275
5276		/*
5277		 * Wait for the youngest allocations and frees to sync,
5278		 * and then wait for the deferral of those frees to finish.
5279		 */
5280		spa_vdev_config_exit(spa, NULL,
5281		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5282
5283		/*
5284		 * Attempt to evacuate the vdev.
5285		 */
5286		error = spa_vdev_remove_evacuate(spa, vd);
5287
5288		txg = spa_vdev_config_enter(spa);
5289
5290		/*
5291		 * If we couldn't evacuate the vdev, unwind.
5292		 */
5293		if (error) {
5294			metaslab_group_activate(mg);
5295			return (spa_vdev_exit(spa, NULL, txg, error));
5296		}
5297
5298		/*
5299		 * Clean up the vdev namespace.
5300		 */
5301		spa_vdev_remove_from_namespace(spa, vd);
5302
5303	} else if (vd != NULL) {
5304		/*
5305		 * Normal vdevs cannot be removed (yet).
5306		 */
5307		error = SET_ERROR(ENOTSUP);
5308	} else {
5309		/*
5310		 * There is no vdev of any kind with the specified guid.
5311		 */
5312		error = SET_ERROR(ENOENT);
5313	}
5314
5315	if (!locked)
5316		return (spa_vdev_exit(spa, NULL, txg, error));
5317
5318	return (error);
5319}
5320
5321/*
5322 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5323 * currently spared, so we can detach it.
5324 */
5325static vdev_t *
5326spa_vdev_resilver_done_hunt(vdev_t *vd)
5327{
5328	vdev_t *newvd, *oldvd;
5329
5330	for (int c = 0; c < vd->vdev_children; c++) {
5331		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5332		if (oldvd != NULL)
5333			return (oldvd);
5334	}
5335
5336	/*
5337	 * Check for a completed replacement.  We always consider the first
5338	 * vdev in the list to be the oldest vdev, and the last one to be
5339	 * the newest (see spa_vdev_attach() for how that works).  In
5340	 * the case where the newest vdev is faulted, we will not automatically
5341	 * remove it after a resilver completes.  This is OK as it will require
5342	 * user intervention to determine which disk the admin wishes to keep.
5343	 */
5344	if (vd->vdev_ops == &vdev_replacing_ops) {
5345		ASSERT(vd->vdev_children > 1);
5346
5347		newvd = vd->vdev_child[vd->vdev_children - 1];
5348		oldvd = vd->vdev_child[0];
5349
5350		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5351		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5352		    !vdev_dtl_required(oldvd))
5353			return (oldvd);
5354	}
5355
5356	/*
5357	 * Check for a completed resilver with the 'unspare' flag set.
5358	 */
5359	if (vd->vdev_ops == &vdev_spare_ops) {
5360		vdev_t *first = vd->vdev_child[0];
5361		vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5362
5363		if (last->vdev_unspare) {
5364			oldvd = first;
5365			newvd = last;
5366		} else if (first->vdev_unspare) {
5367			oldvd = last;
5368			newvd = first;
5369		} else {
5370			oldvd = NULL;
5371		}
5372
5373		if (oldvd != NULL &&
5374		    vdev_dtl_empty(newvd, DTL_MISSING) &&
5375		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5376		    !vdev_dtl_required(oldvd))
5377			return (oldvd);
5378
5379		/*
5380		 * If there are more than two spares attached to a disk,
5381		 * and those spares are not required, then we want to
5382		 * attempt to free them up now so that they can be used
5383		 * by other pools.  Once we're back down to a single
5384		 * disk+spare, we stop removing them.
5385		 */
5386		if (vd->vdev_children > 2) {
5387			newvd = vd->vdev_child[1];
5388
5389			if (newvd->vdev_isspare && last->vdev_isspare &&
5390			    vdev_dtl_empty(last, DTL_MISSING) &&
5391			    vdev_dtl_empty(last, DTL_OUTAGE) &&
5392			    !vdev_dtl_required(newvd))
5393				return (newvd);
5394		}
5395	}
5396
5397	return (NULL);
5398}
5399
5400static void
5401spa_vdev_resilver_done(spa_t *spa)
5402{
5403	vdev_t *vd, *pvd, *ppvd;
5404	uint64_t guid, sguid, pguid, ppguid;
5405
5406	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5407
5408	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5409		pvd = vd->vdev_parent;
5410		ppvd = pvd->vdev_parent;
5411		guid = vd->vdev_guid;
5412		pguid = pvd->vdev_guid;
5413		ppguid = ppvd->vdev_guid;
5414		sguid = 0;
5415		/*
5416		 * If we have just finished replacing a hot spared device, then
5417		 * we need to detach the parent's first child (the original hot
5418		 * spare) as well.
5419		 */
5420		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5421		    ppvd->vdev_children == 2) {
5422			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5423			sguid = ppvd->vdev_child[1]->vdev_guid;
5424		}
5425		ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5426
5427		spa_config_exit(spa, SCL_ALL, FTAG);
5428		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5429			return;
5430		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5431			return;
5432		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5433	}
5434
5435	spa_config_exit(spa, SCL_ALL, FTAG);
5436}
5437
5438/*
5439 * Update the stored path or FRU for this vdev.
5440 */
5441int
5442spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5443    boolean_t ispath)
5444{
5445	vdev_t *vd;
5446	boolean_t sync = B_FALSE;
5447
5448	ASSERT(spa_writeable(spa));
5449
5450	spa_vdev_state_enter(spa, SCL_ALL);
5451
5452	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5453		return (spa_vdev_state_exit(spa, NULL, ENOENT));
5454
5455	if (!vd->vdev_ops->vdev_op_leaf)
5456		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5457
5458	if (ispath) {
5459		if (strcmp(value, vd->vdev_path) != 0) {
5460			spa_strfree(vd->vdev_path);
5461			vd->vdev_path = spa_strdup(value);
5462			sync = B_TRUE;
5463		}
5464	} else {
5465		if (vd->vdev_fru == NULL) {
5466			vd->vdev_fru = spa_strdup(value);
5467			sync = B_TRUE;
5468		} else if (strcmp(value, vd->vdev_fru) != 0) {
5469			spa_strfree(vd->vdev_fru);
5470			vd->vdev_fru = spa_strdup(value);
5471			sync = B_TRUE;
5472		}
5473	}
5474
5475	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5476}
5477
5478int
5479spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5480{
5481	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5482}
5483
5484int
5485spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5486{
5487	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5488}
5489
5490/*
5491 * ==========================================================================
5492 * SPA Scanning
5493 * ==========================================================================
5494 */
5495
5496int
5497spa_scan_stop(spa_t *spa)
5498{
5499	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5500	if (dsl_scan_resilvering(spa->spa_dsl_pool))
5501		return (SET_ERROR(EBUSY));
5502	return (dsl_scan_cancel(spa->spa_dsl_pool));
5503}
5504
5505int
5506spa_scan(spa_t *spa, pool_scan_func_t func)
5507{
5508	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5509
5510	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5511		return (SET_ERROR(ENOTSUP));
5512
5513	/*
5514	 * If a resilver was requested, but there is no DTL on a
5515	 * writeable leaf device, we have nothing to do.
5516	 */
5517	if (func == POOL_SCAN_RESILVER &&
5518	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5519		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5520		return (0);
5521	}
5522
5523	return (dsl_scan(spa->spa_dsl_pool, func));
5524}
5525
5526/*
5527 * ==========================================================================
5528 * SPA async task processing
5529 * ==========================================================================
5530 */
5531
5532static void
5533spa_async_remove(spa_t *spa, vdev_t *vd)
5534{
5535	if (vd->vdev_remove_wanted) {
5536		vd->vdev_remove_wanted = B_FALSE;
5537		vd->vdev_delayed_close = B_FALSE;
5538		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5539
5540		/*
5541		 * We want to clear the stats, but we don't want to do a full
5542		 * vdev_clear() as that will cause us to throw away
5543		 * degraded/faulted state as well as attempt to reopen the
5544		 * device, all of which is a waste.
5545		 */
5546		vd->vdev_stat.vs_read_errors = 0;
5547		vd->vdev_stat.vs_write_errors = 0;
5548		vd->vdev_stat.vs_checksum_errors = 0;
5549
5550		vdev_state_dirty(vd->vdev_top);
5551	}
5552
5553	for (int c = 0; c < vd->vdev_children; c++)
5554		spa_async_remove(spa, vd->vdev_child[c]);
5555}
5556
5557static void
5558spa_async_probe(spa_t *spa, vdev_t *vd)
5559{
5560	if (vd->vdev_probe_wanted) {
5561		vd->vdev_probe_wanted = B_FALSE;
5562		vdev_reopen(vd);	/* vdev_open() does the actual probe */
5563	}
5564
5565	for (int c = 0; c < vd->vdev_children; c++)
5566		spa_async_probe(spa, vd->vdev_child[c]);
5567}
5568
5569static void
5570spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5571{
5572	sysevent_id_t eid;
5573	nvlist_t *attr;
5574	char *physpath;
5575
5576	if (!spa->spa_autoexpand)
5577		return;
5578
5579	for (int c = 0; c < vd->vdev_children; c++) {
5580		vdev_t *cvd = vd->vdev_child[c];
5581		spa_async_autoexpand(spa, cvd);
5582	}
5583
5584	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5585		return;
5586
5587	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5588	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5589
5590	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5591	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5592
5593	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5594	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5595
5596	nvlist_free(attr);
5597	kmem_free(physpath, MAXPATHLEN);
5598}
5599
5600static void
5601spa_async_thread(spa_t *spa)
5602{
5603	int tasks;
5604
5605	ASSERT(spa->spa_sync_on);
5606
5607	mutex_enter(&spa->spa_async_lock);
5608	tasks = spa->spa_async_tasks;
5609	spa->spa_async_tasks = 0;
5610	mutex_exit(&spa->spa_async_lock);
5611
5612	/*
5613	 * See if the config needs to be updated.
5614	 */
5615	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5616		uint64_t old_space, new_space;
5617
5618		mutex_enter(&spa_namespace_lock);
5619		old_space = metaslab_class_get_space(spa_normal_class(spa));
5620		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5621		new_space = metaslab_class_get_space(spa_normal_class(spa));
5622		mutex_exit(&spa_namespace_lock);
5623
5624		/*
5625		 * If the pool grew as a result of the config update,
5626		 * then log an internal history event.
5627		 */
5628		if (new_space != old_space) {
5629			spa_history_log_internal(spa, "vdev online", NULL,
5630			    "pool '%s' size: %llu(+%llu)",
5631			    spa_name(spa), new_space, new_space - old_space);
5632		}
5633	}
5634
5635	/*
5636	 * See if any devices need to be marked REMOVED.
5637	 */
5638	if (tasks & SPA_ASYNC_REMOVE) {
5639		spa_vdev_state_enter(spa, SCL_NONE);
5640		spa_async_remove(spa, spa->spa_root_vdev);
5641		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5642			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5643		for (int i = 0; i < spa->spa_spares.sav_count; i++)
5644			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5645		(void) spa_vdev_state_exit(spa, NULL, 0);
5646	}
5647
5648	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5649		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5650		spa_async_autoexpand(spa, spa->spa_root_vdev);
5651		spa_config_exit(spa, SCL_CONFIG, FTAG);
5652	}
5653
5654	/*
5655	 * See if any devices need to be probed.
5656	 */
5657	if (tasks & SPA_ASYNC_PROBE) {
5658		spa_vdev_state_enter(spa, SCL_NONE);
5659		spa_async_probe(spa, spa->spa_root_vdev);
5660		(void) spa_vdev_state_exit(spa, NULL, 0);
5661	}
5662
5663	/*
5664	 * If any devices are done replacing, detach them.
5665	 */
5666	if (tasks & SPA_ASYNC_RESILVER_DONE)
5667		spa_vdev_resilver_done(spa);
5668
5669	/*
5670	 * Kick off a resilver.
5671	 */
5672	if (tasks & SPA_ASYNC_RESILVER)
5673		dsl_resilver_restart(spa->spa_dsl_pool, 0);
5674
5675	/*
5676	 * Let the world know that we're done.
5677	 */
5678	mutex_enter(&spa->spa_async_lock);
5679	spa->spa_async_thread = NULL;
5680	cv_broadcast(&spa->spa_async_cv);
5681	mutex_exit(&spa->spa_async_lock);
5682	thread_exit();
5683}
5684
5685void
5686spa_async_suspend(spa_t *spa)
5687{
5688	mutex_enter(&spa->spa_async_lock);
5689	spa->spa_async_suspended++;
5690	while (spa->spa_async_thread != NULL)
5691		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5692	mutex_exit(&spa->spa_async_lock);
5693}
5694
5695void
5696spa_async_resume(spa_t *spa)
5697{
5698	mutex_enter(&spa->spa_async_lock);
5699	ASSERT(spa->spa_async_suspended != 0);
5700	spa->spa_async_suspended--;
5701	mutex_exit(&spa->spa_async_lock);
5702}
5703
5704static boolean_t
5705spa_async_tasks_pending(spa_t *spa)
5706{
5707	uint_t non_config_tasks;
5708	uint_t config_task;
5709	boolean_t config_task_suspended;
5710
5711	non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5712	config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5713	if (spa->spa_ccw_fail_time == 0) {
5714		config_task_suspended = B_FALSE;
5715	} else {
5716		config_task_suspended =
5717		    (gethrtime() - spa->spa_ccw_fail_time) <
5718		    (zfs_ccw_retry_interval * NANOSEC);
5719	}
5720
5721	return (non_config_tasks || (config_task && !config_task_suspended));
5722}
5723
5724static void
5725spa_async_dispatch(spa_t *spa)
5726{
5727	mutex_enter(&spa->spa_async_lock);
5728	if (spa_async_tasks_pending(spa) &&
5729	    !spa->spa_async_suspended &&
5730	    spa->spa_async_thread == NULL &&
5731	    rootdir != NULL)
5732		spa->spa_async_thread = thread_create(NULL, 0,
5733		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5734	mutex_exit(&spa->spa_async_lock);
5735}
5736
5737void
5738spa_async_request(spa_t *spa, int task)
5739{
5740	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5741	mutex_enter(&spa->spa_async_lock);
5742	spa->spa_async_tasks |= task;
5743	mutex_exit(&spa->spa_async_lock);
5744}
5745
5746/*
5747 * ==========================================================================
5748 * SPA syncing routines
5749 * ==========================================================================
5750 */
5751
5752static int
5753bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5754{
5755	bpobj_t *bpo = arg;
5756	bpobj_enqueue(bpo, bp, tx);
5757	return (0);
5758}
5759
5760static int
5761spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5762{
5763	zio_t *zio = arg;
5764
5765	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5766	    zio->io_flags));
5767	return (0);
5768}
5769
5770/*
5771 * Note: this simple function is not inlined to make it easier to dtrace the
5772 * amount of time spent syncing frees.
5773 */
5774static void
5775spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
5776{
5777	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5778	bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
5779	VERIFY(zio_wait(zio) == 0);
5780}
5781
5782/*
5783 * Note: this simple function is not inlined to make it easier to dtrace the
5784 * amount of time spent syncing deferred frees.
5785 */
5786static void
5787spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
5788{
5789	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5790	VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
5791	    spa_free_sync_cb, zio, tx), ==, 0);
5792	VERIFY0(zio_wait(zio));
5793}
5794
5795
5796static void
5797spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5798{
5799	char *packed = NULL;
5800	size_t bufsize;
5801	size_t nvsize = 0;
5802	dmu_buf_t *db;
5803
5804	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5805
5806	/*
5807	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5808	 * information.  This avoids the dmu_buf_will_dirty() path and
5809	 * saves us a pre-read to get data we don't actually care about.
5810	 */
5811	bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5812	packed = kmem_alloc(bufsize, KM_SLEEP);
5813
5814	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5815	    KM_SLEEP) == 0);
5816	bzero(packed + nvsize, bufsize - nvsize);
5817
5818	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5819
5820	kmem_free(packed, bufsize);
5821
5822	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5823	dmu_buf_will_dirty(db, tx);
5824	*(uint64_t *)db->db_data = nvsize;
5825	dmu_buf_rele(db, FTAG);
5826}
5827
5828static void
5829spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5830    const char *config, const char *entry)
5831{
5832	nvlist_t *nvroot;
5833	nvlist_t **list;
5834	int i;
5835
5836	if (!sav->sav_sync)
5837		return;
5838
5839	/*
5840	 * Update the MOS nvlist describing the list of available devices.
5841	 * spa_validate_aux() will have already made sure this nvlist is
5842	 * valid and the vdevs are labeled appropriately.
5843	 */
5844	if (sav->sav_object == 0) {
5845		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5846		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5847		    sizeof (uint64_t), tx);
5848		VERIFY(zap_update(spa->spa_meta_objset,
5849		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5850		    &sav->sav_object, tx) == 0);
5851	}
5852
5853	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5854	if (sav->sav_count == 0) {
5855		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5856	} else {
5857		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5858		for (i = 0; i < sav->sav_count; i++)
5859			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5860			    B_FALSE, VDEV_CONFIG_L2CACHE);
5861		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5862		    sav->sav_count) == 0);
5863		for (i = 0; i < sav->sav_count; i++)
5864			nvlist_free(list[i]);
5865		kmem_free(list, sav->sav_count * sizeof (void *));
5866	}
5867
5868	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5869	nvlist_free(nvroot);
5870
5871	sav->sav_sync = B_FALSE;
5872}
5873
5874static void
5875spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5876{
5877	nvlist_t *config;
5878
5879	if (list_is_empty(&spa->spa_config_dirty_list))
5880		return;
5881
5882	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5883
5884	config = spa_config_generate(spa, spa->spa_root_vdev,
5885	    dmu_tx_get_txg(tx), B_FALSE);
5886
5887	/*
5888	 * If we're upgrading the spa version then make sure that
5889	 * the config object gets updated with the correct version.
5890	 */
5891	if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5892		fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5893		    spa->spa_uberblock.ub_version);
5894
5895	spa_config_exit(spa, SCL_STATE, FTAG);
5896
5897	if (spa->spa_config_syncing)
5898		nvlist_free(spa->spa_config_syncing);
5899	spa->spa_config_syncing = config;
5900
5901	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5902}
5903
5904static void
5905spa_sync_version(void *arg, dmu_tx_t *tx)
5906{
5907	uint64_t *versionp = arg;
5908	uint64_t version = *versionp;
5909	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5910
5911	/*
5912	 * Setting the version is special cased when first creating the pool.
5913	 */
5914	ASSERT(tx->tx_txg != TXG_INITIAL);
5915
5916	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5917	ASSERT(version >= spa_version(spa));
5918
5919	spa->spa_uberblock.ub_version = version;
5920	vdev_config_dirty(spa->spa_root_vdev);
5921	spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5922}
5923
5924/*
5925 * Set zpool properties.
5926 */
5927static void
5928spa_sync_props(void *arg, dmu_tx_t *tx)
5929{
5930	nvlist_t *nvp = arg;
5931	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5932	objset_t *mos = spa->spa_meta_objset;
5933	nvpair_t *elem = NULL;
5934
5935	mutex_enter(&spa->spa_props_lock);
5936
5937	while ((elem = nvlist_next_nvpair(nvp, elem))) {
5938		uint64_t intval;
5939		char *strval, *fname;
5940		zpool_prop_t prop;
5941		const char *propname;
5942		zprop_type_t proptype;
5943		spa_feature_t fid;
5944
5945		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5946		case ZPROP_INVAL:
5947			/*
5948			 * We checked this earlier in spa_prop_validate().
5949			 */
5950			ASSERT(zpool_prop_feature(nvpair_name(elem)));
5951
5952			fname = strchr(nvpair_name(elem), '@') + 1;
5953			VERIFY0(zfeature_lookup_name(fname, &fid));
5954
5955			spa_feature_enable(spa, fid, tx);
5956			spa_history_log_internal(spa, "set", tx,
5957			    "%s=enabled", nvpair_name(elem));
5958			break;
5959
5960		case ZPOOL_PROP_VERSION:
5961			intval = fnvpair_value_uint64(elem);
5962			/*
5963			 * The version is synced seperatly before other
5964			 * properties and should be correct by now.
5965			 */
5966			ASSERT3U(spa_version(spa), >=, intval);
5967			break;
5968
5969		case ZPOOL_PROP_ALTROOT:
5970			/*
5971			 * 'altroot' is a non-persistent property. It should
5972			 * have been set temporarily at creation or import time.
5973			 */
5974			ASSERT(spa->spa_root != NULL);
5975			break;
5976
5977		case ZPOOL_PROP_READONLY:
5978		case ZPOOL_PROP_CACHEFILE:
5979			/*
5980			 * 'readonly' and 'cachefile' are also non-persisitent
5981			 * properties.
5982			 */
5983			break;
5984		case ZPOOL_PROP_COMMENT:
5985			strval = fnvpair_value_string(elem);
5986			if (spa->spa_comment != NULL)
5987				spa_strfree(spa->spa_comment);
5988			spa->spa_comment = spa_strdup(strval);
5989			/*
5990			 * We need to dirty the configuration on all the vdevs
5991			 * so that their labels get updated.  It's unnecessary
5992			 * to do this for pool creation since the vdev's
5993			 * configuratoin has already been dirtied.
5994			 */
5995			if (tx->tx_txg != TXG_INITIAL)
5996				vdev_config_dirty(spa->spa_root_vdev);
5997			spa_history_log_internal(spa, "set", tx,
5998			    "%s=%s", nvpair_name(elem), strval);
5999			break;
6000		default:
6001			/*
6002			 * Set pool property values in the poolprops mos object.
6003			 */
6004			if (spa->spa_pool_props_object == 0) {
6005				spa->spa_pool_props_object =
6006				    zap_create_link(mos, DMU_OT_POOL_PROPS,
6007				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6008				    tx);
6009			}
6010
6011			/* normalize the property name */
6012			propname = zpool_prop_to_name(prop);
6013			proptype = zpool_prop_get_type(prop);
6014
6015			if (nvpair_type(elem) == DATA_TYPE_STRING) {
6016				ASSERT(proptype == PROP_TYPE_STRING);
6017				strval = fnvpair_value_string(elem);
6018				VERIFY0(zap_update(mos,
6019				    spa->spa_pool_props_object, propname,
6020				    1, strlen(strval) + 1, strval, tx));
6021				spa_history_log_internal(spa, "set", tx,
6022				    "%s=%s", nvpair_name(elem), strval);
6023			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6024				intval = fnvpair_value_uint64(elem);
6025
6026				if (proptype == PROP_TYPE_INDEX) {
6027					const char *unused;
6028					VERIFY0(zpool_prop_index_to_string(
6029					    prop, intval, &unused));
6030				}
6031				VERIFY0(zap_update(mos,
6032				    spa->spa_pool_props_object, propname,
6033				    8, 1, &intval, tx));
6034				spa_history_log_internal(spa, "set", tx,
6035				    "%s=%lld", nvpair_name(elem), intval);
6036			} else {
6037				ASSERT(0); /* not allowed */
6038			}
6039
6040			switch (prop) {
6041			case ZPOOL_PROP_DELEGATION:
6042				spa->spa_delegation = intval;
6043				break;
6044			case ZPOOL_PROP_BOOTFS:
6045				spa->spa_bootfs = intval;
6046				break;
6047			case ZPOOL_PROP_FAILUREMODE:
6048				spa->spa_failmode = intval;
6049				break;
6050			case ZPOOL_PROP_AUTOEXPAND:
6051