spa.c revision eb633035c80613ec93d62f90482837adaaf21a0a
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, 2019 by Delphix. All rights reserved.
25 * Copyright (c) 2015, Nexenta Systems, Inc.  All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 * Copyright 2013 Saso Kiselkov. All rights reserved.
28 * Copyright (c) 2014 Integros [integros.com]
29 * Copyright 2016 Toomas Soome <tsoome@me.com>
30 * Copyright 2018 Joyent, Inc.
31 * Copyright (c) 2017, Intel Corporation.
32 * Copyright (c) 2017 Datto Inc.
33 * Copyright 2018 OmniOS Community Edition (OmniOSce) Association.
34 */
35
36/*
37 * SPA: Storage Pool Allocator
38 *
39 * This file contains all the routines used when modifying on-disk SPA state.
40 * This includes opening, importing, destroying, exporting a pool, and syncing a
41 * pool.
42 */
43
44#include <sys/zfs_context.h>
45#include <sys/fm/fs/zfs.h>
46#include <sys/spa_impl.h>
47#include <sys/zio.h>
48#include <sys/zio_checksum.h>
49#include <sys/dmu.h>
50#include <sys/dmu_tx.h>
51#include <sys/zap.h>
52#include <sys/zil.h>
53#include <sys/ddt.h>
54#include <sys/vdev_impl.h>
55#include <sys/vdev_removal.h>
56#include <sys/vdev_indirect_mapping.h>
57#include <sys/vdev_indirect_births.h>
58#include <sys/vdev_initialize.h>
59#include <sys/metaslab.h>
60#include <sys/metaslab_impl.h>
61#include <sys/mmp.h>
62#include <sys/uberblock_impl.h>
63#include <sys/txg.h>
64#include <sys/avl.h>
65#include <sys/bpobj.h>
66#include <sys/dmu_traverse.h>
67#include <sys/dmu_objset.h>
68#include <sys/unique.h>
69#include <sys/dsl_pool.h>
70#include <sys/dsl_dataset.h>
71#include <sys/dsl_dir.h>
72#include <sys/dsl_prop.h>
73#include <sys/dsl_synctask.h>
74#include <sys/fs/zfs.h>
75#include <sys/arc.h>
76#include <sys/callb.h>
77#include <sys/systeminfo.h>
78#include <sys/spa_boot.h>
79#include <sys/zfs_ioctl.h>
80#include <sys/dsl_scan.h>
81#include <sys/zfeature.h>
82#include <sys/dsl_destroy.h>
83#include <sys/abd.h>
84
85#ifdef	_KERNEL
86#include <sys/bootprops.h>
87#include <sys/callb.h>
88#include <sys/cpupart.h>
89#include <sys/pool.h>
90#include <sys/sysdc.h>
91#include <sys/zone.h>
92#endif	/* _KERNEL */
93
94#include "zfs_prop.h"
95#include "zfs_comutil.h"
96
97/*
98 * The interval, in seconds, at which failed configuration cache file writes
99 * should be retried.
100 */
101int zfs_ccw_retry_interval = 300;
102
103typedef enum zti_modes {
104	ZTI_MODE_FIXED,			/* value is # of threads (min 1) */
105	ZTI_MODE_BATCH,			/* cpu-intensive; value is ignored */
106	ZTI_MODE_NULL,			/* don't create a taskq */
107	ZTI_NMODES
108} zti_modes_t;
109
110#define	ZTI_P(n, q)	{ ZTI_MODE_FIXED, (n), (q) }
111#define	ZTI_BATCH	{ ZTI_MODE_BATCH, 0, 1 }
112#define	ZTI_NULL	{ ZTI_MODE_NULL, 0, 0 }
113
114#define	ZTI_N(n)	ZTI_P(n, 1)
115#define	ZTI_ONE		ZTI_N(1)
116
117typedef struct zio_taskq_info {
118	zti_modes_t zti_mode;
119	uint_t zti_value;
120	uint_t zti_count;
121} zio_taskq_info_t;
122
123static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
124	"issue", "issue_high", "intr", "intr_high"
125};
126
127/*
128 * This table defines the taskq settings for each ZFS I/O type. When
129 * initializing a pool, we use this table to create an appropriately sized
130 * taskq. Some operations are low volume and therefore have a small, static
131 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
132 * macros. Other operations process a large amount of data; the ZTI_BATCH
133 * macro causes us to create a taskq oriented for throughput. Some operations
134 * are so high frequency and short-lived that the taskq itself can become a a
135 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
136 * additional degree of parallelism specified by the number of threads per-
137 * taskq and the number of taskqs; when dispatching an event in this case, the
138 * particular taskq is chosen at random.
139 *
140 * The different taskq priorities are to handle the different contexts (issue
141 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
142 * need to be handled with minimum delay.
143 */
144const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
145	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
146	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* NULL */
147	{ ZTI_N(8),	ZTI_NULL,	ZTI_P(12, 8),	ZTI_NULL }, /* READ */
148	{ ZTI_BATCH,	ZTI_N(5),	ZTI_N(8),	ZTI_N(5) }, /* WRITE */
149	{ ZTI_P(12, 8),	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* FREE */
150	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* CLAIM */
151	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* IOCTL */
152};
153
154static void spa_sync_version(void *arg, dmu_tx_t *tx);
155static void spa_sync_props(void *arg, dmu_tx_t *tx);
156static boolean_t spa_has_active_shared_spare(spa_t *spa);
157static int spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport);
158static void spa_vdev_resilver_done(spa_t *spa);
159
160uint_t		zio_taskq_batch_pct = 75;	/* 1 thread per cpu in pset */
161id_t		zio_taskq_psrset_bind = PS_NONE;
162boolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
163uint_t		zio_taskq_basedc = 80;		/* base duty cycle */
164
165boolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
166extern int	zfs_sync_pass_deferred_free;
167
168/*
169 * Report any spa_load_verify errors found, but do not fail spa_load.
170 * This is used by zdb to analyze non-idle pools.
171 */
172boolean_t	spa_load_verify_dryrun = B_FALSE;
173
174/*
175 * This (illegal) pool name is used when temporarily importing a spa_t in order
176 * to get the vdev stats associated with the imported devices.
177 */
178#define	TRYIMPORT_NAME	"$import"
179
180/*
181 * For debugging purposes: print out vdev tree during pool import.
182 */
183boolean_t	spa_load_print_vdev_tree = B_FALSE;
184
185/*
186 * A non-zero value for zfs_max_missing_tvds means that we allow importing
187 * pools with missing top-level vdevs. This is strictly intended for advanced
188 * pool recovery cases since missing data is almost inevitable. Pools with
189 * missing devices can only be imported read-only for safety reasons, and their
190 * fail-mode will be automatically set to "continue".
191 *
192 * With 1 missing vdev we should be able to import the pool and mount all
193 * datasets. User data that was not modified after the missing device has been
194 * added should be recoverable. This means that snapshots created prior to the
195 * addition of that device should be completely intact.
196 *
197 * With 2 missing vdevs, some datasets may fail to mount since there are
198 * dataset statistics that are stored as regular metadata. Some data might be
199 * recoverable if those vdevs were added recently.
200 *
201 * With 3 or more missing vdevs, the pool is severely damaged and MOS entries
202 * may be missing entirely. Chances of data recovery are very low. Note that
203 * there are also risks of performing an inadvertent rewind as we might be
204 * missing all the vdevs with the latest uberblocks.
205 */
206uint64_t	zfs_max_missing_tvds = 0;
207
208/*
209 * The parameters below are similar to zfs_max_missing_tvds but are only
210 * intended for a preliminary open of the pool with an untrusted config which
211 * might be incomplete or out-dated.
212 *
213 * We are more tolerant for pools opened from a cachefile since we could have
214 * an out-dated cachefile where a device removal was not registered.
215 * We could have set the limit arbitrarily high but in the case where devices
216 * are really missing we would want to return the proper error codes; we chose
217 * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available
218 * and we get a chance to retrieve the trusted config.
219 */
220uint64_t	zfs_max_missing_tvds_cachefile = SPA_DVAS_PER_BP - 1;
221
222/*
223 * In the case where config was assembled by scanning device paths (/dev/dsks
224 * by default) we are less tolerant since all the existing devices should have
225 * been detected and we want spa_load to return the right error codes.
226 */
227uint64_t	zfs_max_missing_tvds_scan = 0;
228
229/*
230 * Debugging aid that pauses spa_sync() towards the end.
231 */
232boolean_t	zfs_pause_spa_sync = B_FALSE;
233
234/*
235 * ==========================================================================
236 * SPA properties routines
237 * ==========================================================================
238 */
239
240/*
241 * Add a (source=src, propname=propval) list to an nvlist.
242 */
243static void
244spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
245    uint64_t intval, zprop_source_t src)
246{
247	const char *propname = zpool_prop_to_name(prop);
248	nvlist_t *propval;
249
250	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
251	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
252
253	if (strval != NULL)
254		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
255	else
256		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
257
258	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
259	nvlist_free(propval);
260}
261
262/*
263 * Get property values from the spa configuration.
264 */
265static void
266spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
267{
268	vdev_t *rvd = spa->spa_root_vdev;
269	dsl_pool_t *pool = spa->spa_dsl_pool;
270	uint64_t size, alloc, cap, version;
271	zprop_source_t src = ZPROP_SRC_NONE;
272	spa_config_dirent_t *dp;
273	metaslab_class_t *mc = spa_normal_class(spa);
274
275	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
276
277	if (rvd != NULL) {
278		alloc = metaslab_class_get_alloc(mc);
279		alloc += metaslab_class_get_alloc(spa_special_class(spa));
280		alloc += metaslab_class_get_alloc(spa_dedup_class(spa));
281
282		size = metaslab_class_get_space(mc);
283		size += metaslab_class_get_space(spa_special_class(spa));
284		size += metaslab_class_get_space(spa_dedup_class(spa));
285
286		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
287		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
288		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
289		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
290		    size - alloc, src);
291		spa_prop_add_list(*nvp, ZPOOL_PROP_CHECKPOINT, NULL,
292		    spa->spa_checkpoint_info.sci_dspace, src);
293
294		spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
295		    metaslab_class_fragmentation(mc), src);
296		spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
297		    metaslab_class_expandable_space(mc), src);
298		spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
299		    (spa_mode(spa) == FREAD), src);
300
301		cap = (size == 0) ? 0 : (alloc * 100 / size);
302		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
303
304		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
305		    ddt_get_pool_dedup_ratio(spa), src);
306
307		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
308		    rvd->vdev_state, src);
309
310		version = spa_version(spa);
311		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
312			src = ZPROP_SRC_DEFAULT;
313		else
314			src = ZPROP_SRC_LOCAL;
315		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
316	}
317
318	if (pool != NULL) {
319		/*
320		 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
321		 * when opening pools before this version freedir will be NULL.
322		 */
323		if (pool->dp_free_dir != NULL) {
324			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
325			    dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
326			    src);
327		} else {
328			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
329			    NULL, 0, src);
330		}
331
332		if (pool->dp_leak_dir != NULL) {
333			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
334			    dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
335			    src);
336		} else {
337			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
338			    NULL, 0, src);
339		}
340	}
341
342	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
343
344	if (spa->spa_comment != NULL) {
345		spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
346		    0, ZPROP_SRC_LOCAL);
347	}
348
349	if (spa->spa_root != NULL)
350		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
351		    0, ZPROP_SRC_LOCAL);
352
353	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
354		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
355		    MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
356	} else {
357		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
358		    SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
359	}
360
361	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE)) {
362		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL,
363		    DNODE_MAX_SIZE, ZPROP_SRC_NONE);
364	} else {
365		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL,
366		    DNODE_MIN_SIZE, ZPROP_SRC_NONE);
367	}
368
369	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
370		if (dp->scd_path == NULL) {
371			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
372			    "none", 0, ZPROP_SRC_LOCAL);
373		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
374			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
375			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
376		}
377	}
378}
379
380/*
381 * Get zpool property values.
382 */
383int
384spa_prop_get(spa_t *spa, nvlist_t **nvp)
385{
386	objset_t *mos = spa->spa_meta_objset;
387	zap_cursor_t zc;
388	zap_attribute_t za;
389	int err;
390
391	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
392
393	mutex_enter(&spa->spa_props_lock);
394
395	/*
396	 * Get properties from the spa config.
397	 */
398	spa_prop_get_config(spa, nvp);
399
400	/* If no pool property object, no more prop to get. */
401	if (mos == NULL || spa->spa_pool_props_object == 0) {
402		mutex_exit(&spa->spa_props_lock);
403		return (0);
404	}
405
406	/*
407	 * Get properties from the MOS pool property object.
408	 */
409	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
410	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
411	    zap_cursor_advance(&zc)) {
412		uint64_t intval = 0;
413		char *strval = NULL;
414		zprop_source_t src = ZPROP_SRC_DEFAULT;
415		zpool_prop_t prop;
416
417		if ((prop = zpool_name_to_prop(za.za_name)) == ZPOOL_PROP_INVAL)
418			continue;
419
420		switch (za.za_integer_length) {
421		case 8:
422			/* integer property */
423			if (za.za_first_integer !=
424			    zpool_prop_default_numeric(prop))
425				src = ZPROP_SRC_LOCAL;
426
427			if (prop == ZPOOL_PROP_BOOTFS) {
428				dsl_pool_t *dp;
429				dsl_dataset_t *ds = NULL;
430
431				dp = spa_get_dsl(spa);
432				dsl_pool_config_enter(dp, FTAG);
433				err = dsl_dataset_hold_obj(dp,
434				    za.za_first_integer, FTAG, &ds);
435				if (err != 0) {
436					dsl_pool_config_exit(dp, FTAG);
437					break;
438				}
439
440				strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN,
441				    KM_SLEEP);
442				dsl_dataset_name(ds, strval);
443				dsl_dataset_rele(ds, FTAG);
444				dsl_pool_config_exit(dp, FTAG);
445			} else {
446				strval = NULL;
447				intval = za.za_first_integer;
448			}
449
450			spa_prop_add_list(*nvp, prop, strval, intval, src);
451
452			if (strval != NULL)
453				kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN);
454
455			break;
456
457		case 1:
458			/* string property */
459			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
460			err = zap_lookup(mos, spa->spa_pool_props_object,
461			    za.za_name, 1, za.za_num_integers, strval);
462			if (err) {
463				kmem_free(strval, za.za_num_integers);
464				break;
465			}
466			spa_prop_add_list(*nvp, prop, strval, 0, src);
467			kmem_free(strval, za.za_num_integers);
468			break;
469
470		default:
471			break;
472		}
473	}
474	zap_cursor_fini(&zc);
475	mutex_exit(&spa->spa_props_lock);
476out:
477	if (err && err != ENOENT) {
478		nvlist_free(*nvp);
479		*nvp = NULL;
480		return (err);
481	}
482
483	return (0);
484}
485
486/*
487 * Validate the given pool properties nvlist and modify the list
488 * for the property values to be set.
489 */
490static int
491spa_prop_validate(spa_t *spa, nvlist_t *props)
492{
493	nvpair_t *elem;
494	int error = 0, reset_bootfs = 0;
495	uint64_t objnum = 0;
496	boolean_t has_feature = B_FALSE;
497
498	elem = NULL;
499	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
500		uint64_t intval;
501		char *strval, *slash, *check, *fname;
502		const char *propname = nvpair_name(elem);
503		zpool_prop_t prop = zpool_name_to_prop(propname);
504
505		switch (prop) {
506		case ZPOOL_PROP_INVAL:
507			if (!zpool_prop_feature(propname)) {
508				error = SET_ERROR(EINVAL);
509				break;
510			}
511
512			/*
513			 * Sanitize the input.
514			 */
515			if (nvpair_type(elem) != DATA_TYPE_UINT64) {
516				error = SET_ERROR(EINVAL);
517				break;
518			}
519
520			if (nvpair_value_uint64(elem, &intval) != 0) {
521				error = SET_ERROR(EINVAL);
522				break;
523			}
524
525			if (intval != 0) {
526				error = SET_ERROR(EINVAL);
527				break;
528			}
529
530			fname = strchr(propname, '@') + 1;
531			if (zfeature_lookup_name(fname, NULL) != 0) {
532				error = SET_ERROR(EINVAL);
533				break;
534			}
535
536			has_feature = B_TRUE;
537			break;
538
539		case ZPOOL_PROP_VERSION:
540			error = nvpair_value_uint64(elem, &intval);
541			if (!error &&
542			    (intval < spa_version(spa) ||
543			    intval > SPA_VERSION_BEFORE_FEATURES ||
544			    has_feature))
545				error = SET_ERROR(EINVAL);
546			break;
547
548		case ZPOOL_PROP_DELEGATION:
549		case ZPOOL_PROP_AUTOREPLACE:
550		case ZPOOL_PROP_LISTSNAPS:
551		case ZPOOL_PROP_AUTOEXPAND:
552			error = nvpair_value_uint64(elem, &intval);
553			if (!error && intval > 1)
554				error = SET_ERROR(EINVAL);
555			break;
556
557		case ZPOOL_PROP_MULTIHOST:
558			error = nvpair_value_uint64(elem, &intval);
559			if (!error && intval > 1)
560				error = SET_ERROR(EINVAL);
561
562			if (!error && !spa_get_hostid())
563				error = SET_ERROR(ENOTSUP);
564
565			break;
566
567		case ZPOOL_PROP_BOOTFS:
568			/*
569			 * If the pool version is less than SPA_VERSION_BOOTFS,
570			 * or the pool is still being created (version == 0),
571			 * the bootfs property cannot be set.
572			 */
573			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
574				error = SET_ERROR(ENOTSUP);
575				break;
576			}
577
578			/*
579			 * Make sure the vdev config is bootable
580			 */
581			if (!vdev_is_bootable(spa->spa_root_vdev)) {
582				error = SET_ERROR(ENOTSUP);
583				break;
584			}
585
586			reset_bootfs = 1;
587
588			error = nvpair_value_string(elem, &strval);
589
590			if (!error) {
591				objset_t *os;
592				uint64_t propval;
593
594				if (strval == NULL || strval[0] == '\0') {
595					objnum = zpool_prop_default_numeric(
596					    ZPOOL_PROP_BOOTFS);
597					break;
598				}
599
600				error = dmu_objset_hold(strval, FTAG, &os);
601				if (error != 0)
602					break;
603
604				/*
605				 * Must be ZPL, and its property settings
606				 * must be supported.
607				 */
608
609				if (dmu_objset_type(os) != DMU_OST_ZFS) {
610					error = SET_ERROR(ENOTSUP);
611				} else if ((error =
612				    dsl_prop_get_int_ds(dmu_objset_ds(os),
613				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
614				    &propval)) == 0 &&
615				    !BOOTFS_COMPRESS_VALID(propval)) {
616					error = SET_ERROR(ENOTSUP);
617				} else {
618					objnum = dmu_objset_id(os);
619				}
620				dmu_objset_rele(os, FTAG);
621			}
622			break;
623
624		case ZPOOL_PROP_FAILUREMODE:
625			error = nvpair_value_uint64(elem, &intval);
626			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
627			    intval > ZIO_FAILURE_MODE_PANIC))
628				error = SET_ERROR(EINVAL);
629
630			/*
631			 * This is a special case which only occurs when
632			 * the pool has completely failed. This allows
633			 * the user to change the in-core failmode property
634			 * without syncing it out to disk (I/Os might
635			 * currently be blocked). We do this by returning
636			 * EIO to the caller (spa_prop_set) to trick it
637			 * into thinking we encountered a property validation
638			 * error.
639			 */
640			if (!error && spa_suspended(spa)) {
641				spa->spa_failmode = intval;
642				error = SET_ERROR(EIO);
643			}
644			break;
645
646		case ZPOOL_PROP_CACHEFILE:
647			if ((error = nvpair_value_string(elem, &strval)) != 0)
648				break;
649
650			if (strval[0] == '\0')
651				break;
652
653			if (strcmp(strval, "none") == 0)
654				break;
655
656			if (strval[0] != '/') {
657				error = SET_ERROR(EINVAL);
658				break;
659			}
660
661			slash = strrchr(strval, '/');
662			ASSERT(slash != NULL);
663
664			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
665			    strcmp(slash, "/..") == 0)
666				error = SET_ERROR(EINVAL);
667			break;
668
669		case ZPOOL_PROP_COMMENT:
670			if ((error = nvpair_value_string(elem, &strval)) != 0)
671				break;
672			for (check = strval; *check != '\0'; check++) {
673				/*
674				 * The kernel doesn't have an easy isprint()
675				 * check.  For this kernel check, we merely
676				 * check ASCII apart from DEL.  Fix this if
677				 * there is an easy-to-use kernel isprint().
678				 */
679				if (*check >= 0x7f) {
680					error = SET_ERROR(EINVAL);
681					break;
682				}
683			}
684			if (strlen(strval) > ZPROP_MAX_COMMENT)
685				error = E2BIG;
686			break;
687
688		case ZPOOL_PROP_DEDUPDITTO:
689			if (spa_version(spa) < SPA_VERSION_DEDUP)
690				error = SET_ERROR(ENOTSUP);
691			else
692				error = nvpair_value_uint64(elem, &intval);
693			if (error == 0 &&
694			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
695				error = SET_ERROR(EINVAL);
696			break;
697		}
698
699		if (error)
700			break;
701	}
702
703	if (!error && reset_bootfs) {
704		error = nvlist_remove(props,
705		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
706
707		if (!error) {
708			error = nvlist_add_uint64(props,
709			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
710		}
711	}
712
713	return (error);
714}
715
716void
717spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
718{
719	char *cachefile;
720	spa_config_dirent_t *dp;
721
722	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
723	    &cachefile) != 0)
724		return;
725
726	dp = kmem_alloc(sizeof (spa_config_dirent_t),
727	    KM_SLEEP);
728
729	if (cachefile[0] == '\0')
730		dp->scd_path = spa_strdup(spa_config_path);
731	else if (strcmp(cachefile, "none") == 0)
732		dp->scd_path = NULL;
733	else
734		dp->scd_path = spa_strdup(cachefile);
735
736	list_insert_head(&spa->spa_config_list, dp);
737	if (need_sync)
738		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
739}
740
741int
742spa_prop_set(spa_t *spa, nvlist_t *nvp)
743{
744	int error;
745	nvpair_t *elem = NULL;
746	boolean_t need_sync = B_FALSE;
747
748	if ((error = spa_prop_validate(spa, nvp)) != 0)
749		return (error);
750
751	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
752		zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
753
754		if (prop == ZPOOL_PROP_CACHEFILE ||
755		    prop == ZPOOL_PROP_ALTROOT ||
756		    prop == ZPOOL_PROP_READONLY)
757			continue;
758
759		if (prop == ZPOOL_PROP_VERSION || prop == ZPOOL_PROP_INVAL) {
760			uint64_t ver;
761
762			if (prop == ZPOOL_PROP_VERSION) {
763				VERIFY(nvpair_value_uint64(elem, &ver) == 0);
764			} else {
765				ASSERT(zpool_prop_feature(nvpair_name(elem)));
766				ver = SPA_VERSION_FEATURES;
767				need_sync = B_TRUE;
768			}
769
770			/* Save time if the version is already set. */
771			if (ver == spa_version(spa))
772				continue;
773
774			/*
775			 * In addition to the pool directory object, we might
776			 * create the pool properties object, the features for
777			 * read object, the features for write object, or the
778			 * feature descriptions object.
779			 */
780			error = dsl_sync_task(spa->spa_name, NULL,
781			    spa_sync_version, &ver,
782			    6, ZFS_SPACE_CHECK_RESERVED);
783			if (error)
784				return (error);
785			continue;
786		}
787
788		need_sync = B_TRUE;
789		break;
790	}
791
792	if (need_sync) {
793		return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
794		    nvp, 6, ZFS_SPACE_CHECK_RESERVED));
795	}
796
797	return (0);
798}
799
800/*
801 * If the bootfs property value is dsobj, clear it.
802 */
803void
804spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
805{
806	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
807		VERIFY(zap_remove(spa->spa_meta_objset,
808		    spa->spa_pool_props_object,
809		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
810		spa->spa_bootfs = 0;
811	}
812}
813
814/*ARGSUSED*/
815static int
816spa_change_guid_check(void *arg, dmu_tx_t *tx)
817{
818	uint64_t *newguid = arg;
819	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
820	vdev_t *rvd = spa->spa_root_vdev;
821	uint64_t vdev_state;
822
823	if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
824		int error = (spa_has_checkpoint(spa)) ?
825		    ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
826		return (SET_ERROR(error));
827	}
828
829	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
830	vdev_state = rvd->vdev_state;
831	spa_config_exit(spa, SCL_STATE, FTAG);
832
833	if (vdev_state != VDEV_STATE_HEALTHY)
834		return (SET_ERROR(ENXIO));
835
836	ASSERT3U(spa_guid(spa), !=, *newguid);
837
838	return (0);
839}
840
841static void
842spa_change_guid_sync(void *arg, dmu_tx_t *tx)
843{
844	uint64_t *newguid = arg;
845	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
846	uint64_t oldguid;
847	vdev_t *rvd = spa->spa_root_vdev;
848
849	oldguid = spa_guid(spa);
850
851	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
852	rvd->vdev_guid = *newguid;
853	rvd->vdev_guid_sum += (*newguid - oldguid);
854	vdev_config_dirty(rvd);
855	spa_config_exit(spa, SCL_STATE, FTAG);
856
857	spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
858	    oldguid, *newguid);
859}
860
861/*
862 * Change the GUID for the pool.  This is done so that we can later
863 * re-import a pool built from a clone of our own vdevs.  We will modify
864 * the root vdev's guid, our own pool guid, and then mark all of our
865 * vdevs dirty.  Note that we must make sure that all our vdevs are
866 * online when we do this, or else any vdevs that weren't present
867 * would be orphaned from our pool.  We are also going to issue a
868 * sysevent to update any watchers.
869 */
870int
871spa_change_guid(spa_t *spa)
872{
873	int error;
874	uint64_t guid;
875
876	mutex_enter(&spa->spa_vdev_top_lock);
877	mutex_enter(&spa_namespace_lock);
878	guid = spa_generate_guid(NULL);
879
880	error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
881	    spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
882
883	if (error == 0) {
884		spa_write_cachefile(spa, B_FALSE, B_TRUE);
885		spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID);
886	}
887
888	mutex_exit(&spa_namespace_lock);
889	mutex_exit(&spa->spa_vdev_top_lock);
890
891	return (error);
892}
893
894/*
895 * ==========================================================================
896 * SPA state manipulation (open/create/destroy/import/export)
897 * ==========================================================================
898 */
899
900static int
901spa_error_entry_compare(const void *a, const void *b)
902{
903	const spa_error_entry_t *sa = (const spa_error_entry_t *)a;
904	const spa_error_entry_t *sb = (const spa_error_entry_t *)b;
905	int ret;
906
907	ret = memcmp(&sa->se_bookmark, &sb->se_bookmark,
908	    sizeof (zbookmark_phys_t));
909
910	return (AVL_ISIGN(ret));
911}
912
913/*
914 * Utility function which retrieves copies of the current logs and
915 * re-initializes them in the process.
916 */
917void
918spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
919{
920	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
921
922	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
923	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
924
925	avl_create(&spa->spa_errlist_scrub,
926	    spa_error_entry_compare, sizeof (spa_error_entry_t),
927	    offsetof(spa_error_entry_t, se_avl));
928	avl_create(&spa->spa_errlist_last,
929	    spa_error_entry_compare, sizeof (spa_error_entry_t),
930	    offsetof(spa_error_entry_t, se_avl));
931}
932
933static void
934spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
935{
936	const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
937	enum zti_modes mode = ztip->zti_mode;
938	uint_t value = ztip->zti_value;
939	uint_t count = ztip->zti_count;
940	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
941	char name[32];
942	uint_t flags = 0;
943	boolean_t batch = B_FALSE;
944
945	if (mode == ZTI_MODE_NULL) {
946		tqs->stqs_count = 0;
947		tqs->stqs_taskq = NULL;
948		return;
949	}
950
951	ASSERT3U(count, >, 0);
952
953	tqs->stqs_count = count;
954	tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
955
956	switch (mode) {
957	case ZTI_MODE_FIXED:
958		ASSERT3U(value, >=, 1);
959		value = MAX(value, 1);
960		break;
961
962	case ZTI_MODE_BATCH:
963		batch = B_TRUE;
964		flags |= TASKQ_THREADS_CPU_PCT;
965		value = zio_taskq_batch_pct;
966		break;
967
968	default:
969		panic("unrecognized mode for %s_%s taskq (%u:%u) in "
970		    "spa_activate()",
971		    zio_type_name[t], zio_taskq_types[q], mode, value);
972		break;
973	}
974
975	for (uint_t i = 0; i < count; i++) {
976		taskq_t *tq;
977
978		if (count > 1) {
979			(void) snprintf(name, sizeof (name), "%s_%s_%u",
980			    zio_type_name[t], zio_taskq_types[q], i);
981		} else {
982			(void) snprintf(name, sizeof (name), "%s_%s",
983			    zio_type_name[t], zio_taskq_types[q]);
984		}
985
986		if (zio_taskq_sysdc && spa->spa_proc != &p0) {
987			if (batch)
988				flags |= TASKQ_DC_BATCH;
989
990			tq = taskq_create_sysdc(name, value, 50, INT_MAX,
991			    spa->spa_proc, zio_taskq_basedc, flags);
992		} else {
993			pri_t pri = maxclsyspri;
994			/*
995			 * The write issue taskq can be extremely CPU
996			 * intensive.  Run it at slightly lower priority
997			 * than the other taskqs.
998			 */
999			if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
1000				pri--;
1001
1002			tq = taskq_create_proc(name, value, pri, 50,
1003			    INT_MAX, spa->spa_proc, flags);
1004		}
1005
1006		tqs->stqs_taskq[i] = tq;
1007	}
1008}
1009
1010static void
1011spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
1012{
1013	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1014
1015	if (tqs->stqs_taskq == NULL) {
1016		ASSERT0(tqs->stqs_count);
1017		return;
1018	}
1019
1020	for (uint_t i = 0; i < tqs->stqs_count; i++) {
1021		ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
1022		taskq_destroy(tqs->stqs_taskq[i]);
1023	}
1024
1025	kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
1026	tqs->stqs_taskq = NULL;
1027}
1028
1029/*
1030 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
1031 * Note that a type may have multiple discrete taskqs to avoid lock contention
1032 * on the taskq itself. In that case we choose which taskq at random by using
1033 * the low bits of gethrtime().
1034 */
1035void
1036spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
1037    task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
1038{
1039	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1040	taskq_t *tq;
1041
1042	ASSERT3P(tqs->stqs_taskq, !=, NULL);
1043	ASSERT3U(tqs->stqs_count, !=, 0);
1044
1045	if (tqs->stqs_count == 1) {
1046		tq = tqs->stqs_taskq[0];
1047	} else {
1048		tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
1049	}
1050
1051	taskq_dispatch_ent(tq, func, arg, flags, ent);
1052}
1053
1054static void
1055spa_create_zio_taskqs(spa_t *spa)
1056{
1057	for (int t = 0; t < ZIO_TYPES; t++) {
1058		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1059			spa_taskqs_init(spa, t, q);
1060		}
1061	}
1062}
1063
1064#ifdef _KERNEL
1065static void
1066spa_thread(void *arg)
1067{
1068	callb_cpr_t cprinfo;
1069
1070	spa_t *spa = arg;
1071	user_t *pu = PTOU(curproc);
1072
1073	CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
1074	    spa->spa_name);
1075
1076	ASSERT(curproc != &p0);
1077	(void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
1078	    "zpool-%s", spa->spa_name);
1079	(void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
1080
1081	/* bind this thread to the requested psrset */
1082	if (zio_taskq_psrset_bind != PS_NONE) {
1083		pool_lock();
1084		mutex_enter(&cpu_lock);
1085		mutex_enter(&pidlock);
1086		mutex_enter(&curproc->p_lock);
1087
1088		if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1089		    0, NULL, NULL) == 0)  {
1090			curthread->t_bind_pset = zio_taskq_psrset_bind;
1091		} else {
1092			cmn_err(CE_WARN,
1093			    "Couldn't bind process for zfs pool \"%s\" to "
1094			    "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1095		}
1096
1097		mutex_exit(&curproc->p_lock);
1098		mutex_exit(&pidlock);
1099		mutex_exit(&cpu_lock);
1100		pool_unlock();
1101	}
1102
1103	if (zio_taskq_sysdc) {
1104		sysdc_thread_enter(curthread, 100, 0);
1105	}
1106
1107	spa->spa_proc = curproc;
1108	spa->spa_did = curthread->t_did;
1109
1110	spa_create_zio_taskqs(spa);
1111
1112	mutex_enter(&spa->spa_proc_lock);
1113	ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1114
1115	spa->spa_proc_state = SPA_PROC_ACTIVE;
1116	cv_broadcast(&spa->spa_proc_cv);
1117
1118	CALLB_CPR_SAFE_BEGIN(&cprinfo);
1119	while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1120		cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1121	CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1122
1123	ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1124	spa->spa_proc_state = SPA_PROC_GONE;
1125	spa->spa_proc = &p0;
1126	cv_broadcast(&spa->spa_proc_cv);
1127	CALLB_CPR_EXIT(&cprinfo);	/* drops spa_proc_lock */
1128
1129	mutex_enter(&curproc->p_lock);
1130	lwp_exit();
1131}
1132#endif
1133
1134/*
1135 * Activate an uninitialized pool.
1136 */
1137static void
1138spa_activate(spa_t *spa, int mode)
1139{
1140	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1141
1142	spa->spa_state = POOL_STATE_ACTIVE;
1143	spa->spa_mode = mode;
1144
1145	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1146	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1147	spa->spa_special_class = metaslab_class_create(spa, zfs_metaslab_ops);
1148	spa->spa_dedup_class = metaslab_class_create(spa, zfs_metaslab_ops);
1149
1150	/* Try to create a covering process */
1151	mutex_enter(&spa->spa_proc_lock);
1152	ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1153	ASSERT(spa->spa_proc == &p0);
1154	spa->spa_did = 0;
1155
1156	/* Only create a process if we're going to be around a while. */
1157	if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1158		if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1159		    NULL, 0) == 0) {
1160			spa->spa_proc_state = SPA_PROC_CREATED;
1161			while (spa->spa_proc_state == SPA_PROC_CREATED) {
1162				cv_wait(&spa->spa_proc_cv,
1163				    &spa->spa_proc_lock);
1164			}
1165			ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1166			ASSERT(spa->spa_proc != &p0);
1167			ASSERT(spa->spa_did != 0);
1168		} else {
1169#ifdef _KERNEL
1170			cmn_err(CE_WARN,
1171			    "Couldn't create process for zfs pool \"%s\"\n",
1172			    spa->spa_name);
1173#endif
1174		}
1175	}
1176	mutex_exit(&spa->spa_proc_lock);
1177
1178	/* If we didn't create a process, we need to create our taskqs. */
1179	if (spa->spa_proc == &p0) {
1180		spa_create_zio_taskqs(spa);
1181	}
1182
1183	for (size_t i = 0; i < TXG_SIZE; i++) {
1184		spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL,
1185		    ZIO_FLAG_CANFAIL);
1186	}
1187
1188	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1189	    offsetof(vdev_t, vdev_config_dirty_node));
1190	list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1191	    offsetof(objset_t, os_evicting_node));
1192	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1193	    offsetof(vdev_t, vdev_state_dirty_node));
1194
1195	txg_list_create(&spa->spa_vdev_txg_list, spa,
1196	    offsetof(struct vdev, vdev_txg_node));
1197
1198	avl_create(&spa->spa_errlist_scrub,
1199	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1200	    offsetof(spa_error_entry_t, se_avl));
1201	avl_create(&spa->spa_errlist_last,
1202	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1203	    offsetof(spa_error_entry_t, se_avl));
1204
1205	spa_keystore_init(&spa->spa_keystore);
1206}
1207
1208/*
1209 * Opposite of spa_activate().
1210 */
1211static void
1212spa_deactivate(spa_t *spa)
1213{
1214	ASSERT(spa->spa_sync_on == B_FALSE);
1215	ASSERT(spa->spa_dsl_pool == NULL);
1216	ASSERT(spa->spa_root_vdev == NULL);
1217	ASSERT(spa->spa_async_zio_root == NULL);
1218	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1219
1220	spa_evicting_os_wait(spa);
1221
1222	txg_list_destroy(&spa->spa_vdev_txg_list);
1223
1224	list_destroy(&spa->spa_config_dirty_list);
1225	list_destroy(&spa->spa_evicting_os_list);
1226	list_destroy(&spa->spa_state_dirty_list);
1227
1228	for (int t = 0; t < ZIO_TYPES; t++) {
1229		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1230			spa_taskqs_fini(spa, t, q);
1231		}
1232	}
1233
1234	for (size_t i = 0; i < TXG_SIZE; i++) {
1235		ASSERT3P(spa->spa_txg_zio[i], !=, NULL);
1236		VERIFY0(zio_wait(spa->spa_txg_zio[i]));
1237		spa->spa_txg_zio[i] = NULL;
1238	}
1239
1240	metaslab_class_destroy(spa->spa_normal_class);
1241	spa->spa_normal_class = NULL;
1242
1243	metaslab_class_destroy(spa->spa_log_class);
1244	spa->spa_log_class = NULL;
1245
1246	metaslab_class_destroy(spa->spa_special_class);
1247	spa->spa_special_class = NULL;
1248
1249	metaslab_class_destroy(spa->spa_dedup_class);
1250	spa->spa_dedup_class = NULL;
1251
1252	/*
1253	 * If this was part of an import or the open otherwise failed, we may
1254	 * still have errors left in the queues.  Empty them just in case.
1255	 */
1256	spa_errlog_drain(spa);
1257	avl_destroy(&spa->spa_errlist_scrub);
1258	avl_destroy(&spa->spa_errlist_last);
1259
1260	spa_keystore_fini(&spa->spa_keystore);
1261
1262	spa->spa_state = POOL_STATE_UNINITIALIZED;
1263
1264	mutex_enter(&spa->spa_proc_lock);
1265	if (spa->spa_proc_state != SPA_PROC_NONE) {
1266		ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1267		spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1268		cv_broadcast(&spa->spa_proc_cv);
1269		while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1270			ASSERT(spa->spa_proc != &p0);
1271			cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1272		}
1273		ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1274		spa->spa_proc_state = SPA_PROC_NONE;
1275	}
1276	ASSERT(spa->spa_proc == &p0);
1277	mutex_exit(&spa->spa_proc_lock);
1278
1279	/*
1280	 * We want to make sure spa_thread() has actually exited the ZFS
1281	 * module, so that the module can't be unloaded out from underneath
1282	 * it.
1283	 */
1284	if (spa->spa_did != 0) {
1285		thread_join(spa->spa_did);
1286		spa->spa_did = 0;
1287	}
1288}
1289
1290/*
1291 * Verify a pool configuration, and construct the vdev tree appropriately.  This
1292 * will create all the necessary vdevs in the appropriate layout, with each vdev
1293 * in the CLOSED state.  This will prep the pool before open/creation/import.
1294 * All vdev validation is done by the vdev_alloc() routine.
1295 */
1296static int
1297spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1298    uint_t id, int atype)
1299{
1300	nvlist_t **child;
1301	uint_t children;
1302	int error;
1303
1304	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1305		return (error);
1306
1307	if ((*vdp)->vdev_ops->vdev_op_leaf)
1308		return (0);
1309
1310	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1311	    &child, &children);
1312
1313	if (error == ENOENT)
1314		return (0);
1315
1316	if (error) {
1317		vdev_free(*vdp);
1318		*vdp = NULL;
1319		return (SET_ERROR(EINVAL));
1320	}
1321
1322	for (int c = 0; c < children; c++) {
1323		vdev_t *vd;
1324		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1325		    atype)) != 0) {
1326			vdev_free(*vdp);
1327			*vdp = NULL;
1328			return (error);
1329		}
1330	}
1331
1332	ASSERT(*vdp != NULL);
1333
1334	return (0);
1335}
1336
1337/*
1338 * Opposite of spa_load().
1339 */
1340static void
1341spa_unload(spa_t *spa)
1342{
1343	int i;
1344
1345	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1346
1347	spa_load_note(spa, "UNLOADING");
1348
1349	/*
1350	 * Stop async tasks.
1351	 */
1352	spa_async_suspend(spa);
1353
1354	if (spa->spa_root_vdev) {
1355		vdev_initialize_stop_all(spa->spa_root_vdev,
1356		    VDEV_INITIALIZE_ACTIVE);
1357	}
1358
1359	/*
1360	 * Stop syncing.
1361	 */
1362	if (spa->spa_sync_on) {
1363		txg_sync_stop(spa->spa_dsl_pool);
1364		spa->spa_sync_on = B_FALSE;
1365	}
1366
1367	/*
1368	 * Even though vdev_free() also calls vdev_metaslab_fini, we need
1369	 * to call it earlier, before we wait for async i/o to complete.
1370	 * This ensures that there is no async metaslab prefetching, by
1371	 * calling taskq_wait(mg_taskq).
1372	 */
1373	if (spa->spa_root_vdev != NULL) {
1374		spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1375		for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++)
1376			vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]);
1377		spa_config_exit(spa, SCL_ALL, spa);
1378	}
1379
1380	if (spa->spa_mmp.mmp_thread)
1381		mmp_thread_stop(spa);
1382
1383	/*
1384	 * Wait for any outstanding async I/O to complete.
1385	 */
1386	if (spa->spa_async_zio_root != NULL) {
1387		for (int i = 0; i < max_ncpus; i++)
1388			(void) zio_wait(spa->spa_async_zio_root[i]);
1389		kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1390		spa->spa_async_zio_root = NULL;
1391	}
1392
1393	if (spa->spa_vdev_removal != NULL) {
1394		spa_vdev_removal_destroy(spa->spa_vdev_removal);
1395		spa->spa_vdev_removal = NULL;
1396	}
1397
1398	if (spa->spa_condense_zthr != NULL) {
1399		zthr_destroy(spa->spa_condense_zthr);
1400		spa->spa_condense_zthr = NULL;
1401	}
1402
1403	if (spa->spa_checkpoint_discard_zthr != NULL) {
1404		zthr_destroy(spa->spa_checkpoint_discard_zthr);
1405		spa->spa_checkpoint_discard_zthr = NULL;
1406	}
1407
1408	spa_condense_fini(spa);
1409
1410	bpobj_close(&spa->spa_deferred_bpobj);
1411
1412	spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1413
1414	/*
1415	 * Close all vdevs.
1416	 */
1417	if (spa->spa_root_vdev)
1418		vdev_free(spa->spa_root_vdev);
1419	ASSERT(spa->spa_root_vdev == NULL);
1420
1421	/*
1422	 * Close the dsl pool.
1423	 */
1424	if (spa->spa_dsl_pool) {
1425		dsl_pool_close(spa->spa_dsl_pool);
1426		spa->spa_dsl_pool = NULL;
1427		spa->spa_meta_objset = NULL;
1428	}
1429
1430	ddt_unload(spa);
1431
1432	/*
1433	 * Drop and purge level 2 cache
1434	 */
1435	spa_l2cache_drop(spa);
1436
1437	for (i = 0; i < spa->spa_spares.sav_count; i++)
1438		vdev_free(spa->spa_spares.sav_vdevs[i]);
1439	if (spa->spa_spares.sav_vdevs) {
1440		kmem_free(spa->spa_spares.sav_vdevs,
1441		    spa->spa_spares.sav_count * sizeof (void *));
1442		spa->spa_spares.sav_vdevs = NULL;
1443	}
1444	if (spa->spa_spares.sav_config) {
1445		nvlist_free(spa->spa_spares.sav_config);
1446		spa->spa_spares.sav_config = NULL;
1447	}
1448	spa->spa_spares.sav_count = 0;
1449
1450	for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1451		vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1452		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1453	}
1454	if (spa->spa_l2cache.sav_vdevs) {
1455		kmem_free(spa->spa_l2cache.sav_vdevs,
1456		    spa->spa_l2cache.sav_count * sizeof (void *));
1457		spa->spa_l2cache.sav_vdevs = NULL;
1458	}
1459	if (spa->spa_l2cache.sav_config) {
1460		nvlist_free(spa->spa_l2cache.sav_config);
1461		spa->spa_l2cache.sav_config = NULL;
1462	}
1463	spa->spa_l2cache.sav_count = 0;
1464
1465	spa->spa_async_suspended = 0;
1466
1467	spa->spa_indirect_vdevs_loaded = B_FALSE;
1468
1469	if (spa->spa_comment != NULL) {
1470		spa_strfree(spa->spa_comment);
1471		spa->spa_comment = NULL;
1472	}
1473
1474	spa_config_exit(spa, SCL_ALL, spa);
1475}
1476
1477/*
1478 * Load (or re-load) the current list of vdevs describing the active spares for
1479 * this pool.  When this is called, we have some form of basic information in
1480 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1481 * then re-generate a more complete list including status information.
1482 */
1483void
1484spa_load_spares(spa_t *spa)
1485{
1486	nvlist_t **spares;
1487	uint_t nspares;
1488	int i;
1489	vdev_t *vd, *tvd;
1490
1491#ifndef _KERNEL
1492	/*
1493	 * zdb opens both the current state of the pool and the
1494	 * checkpointed state (if present), with a different spa_t.
1495	 *
1496	 * As spare vdevs are shared among open pools, we skip loading
1497	 * them when we load the checkpointed state of the pool.
1498	 */
1499	if (!spa_writeable(spa))
1500		return;
1501#endif
1502
1503	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1504
1505	/*
1506	 * First, close and free any existing spare vdevs.
1507	 */
1508	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1509		vd = spa->spa_spares.sav_vdevs[i];
1510
1511		/* Undo the call to spa_activate() below */
1512		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1513		    B_FALSE)) != NULL && tvd->vdev_isspare)
1514			spa_spare_remove(tvd);
1515		vdev_close(vd);
1516		vdev_free(vd);
1517	}
1518
1519	if (spa->spa_spares.sav_vdevs)
1520		kmem_free(spa->spa_spares.sav_vdevs,
1521		    spa->spa_spares.sav_count * sizeof (void *));
1522
1523	if (spa->spa_spares.sav_config == NULL)
1524		nspares = 0;
1525	else
1526		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1527		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1528
1529	spa->spa_spares.sav_count = (int)nspares;
1530	spa->spa_spares.sav_vdevs = NULL;
1531
1532	if (nspares == 0)
1533		return;
1534
1535	/*
1536	 * Construct the array of vdevs, opening them to get status in the
1537	 * process.   For each spare, there is potentially two different vdev_t
1538	 * structures associated with it: one in the list of spares (used only
1539	 * for basic validation purposes) and one in the active vdev
1540	 * configuration (if it's spared in).  During this phase we open and
1541	 * validate each vdev on the spare list.  If the vdev also exists in the
1542	 * active configuration, then we also mark this vdev as an active spare.
1543	 */
1544	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1545	    KM_SLEEP);
1546	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1547		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1548		    VDEV_ALLOC_SPARE) == 0);
1549		ASSERT(vd != NULL);
1550
1551		spa->spa_spares.sav_vdevs[i] = vd;
1552
1553		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1554		    B_FALSE)) != NULL) {
1555			if (!tvd->vdev_isspare)
1556				spa_spare_add(tvd);
1557
1558			/*
1559			 * We only mark the spare active if we were successfully
1560			 * able to load the vdev.  Otherwise, importing a pool
1561			 * with a bad active spare would result in strange
1562			 * behavior, because multiple pool would think the spare
1563			 * is actively in use.
1564			 *
1565			 * There is a vulnerability here to an equally bizarre
1566			 * circumstance, where a dead active spare is later
1567			 * brought back to life (onlined or otherwise).  Given
1568			 * the rarity of this scenario, and the extra complexity
1569			 * it adds, we ignore the possibility.
1570			 */
1571			if (!vdev_is_dead(tvd))
1572				spa_spare_activate(tvd);
1573		}
1574
1575		vd->vdev_top = vd;
1576		vd->vdev_aux = &spa->spa_spares;
1577
1578		if (vdev_open(vd) != 0)
1579			continue;
1580
1581		if (vdev_validate_aux(vd) == 0)
1582			spa_spare_add(vd);
1583	}
1584
1585	/*
1586	 * Recompute the stashed list of spares, with status information
1587	 * this time.
1588	 */
1589	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1590	    DATA_TYPE_NVLIST_ARRAY) == 0);
1591
1592	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1593	    KM_SLEEP);
1594	for (i = 0; i < spa->spa_spares.sav_count; i++)
1595		spares[i] = vdev_config_generate(spa,
1596		    spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1597	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1598	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1599	for (i = 0; i < spa->spa_spares.sav_count; i++)
1600		nvlist_free(spares[i]);
1601	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1602}
1603
1604/*
1605 * Load (or re-load) the current list of vdevs describing the active l2cache for
1606 * this pool.  When this is called, we have some form of basic information in
1607 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1608 * then re-generate a more complete list including status information.
1609 * Devices which are already active have their details maintained, and are
1610 * not re-opened.
1611 */
1612void
1613spa_load_l2cache(spa_t *spa)
1614{
1615	nvlist_t **l2cache;
1616	uint_t nl2cache;
1617	int i, j, oldnvdevs;
1618	uint64_t guid;
1619	vdev_t *vd, **oldvdevs, **newvdevs;
1620	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1621
1622#ifndef _KERNEL
1623	/*
1624	 * zdb opens both the current state of the pool and the
1625	 * checkpointed state (if present), with a different spa_t.
1626	 *
1627	 * As L2 caches are part of the ARC which is shared among open
1628	 * pools, we skip loading them when we load the checkpointed
1629	 * state of the pool.
1630	 */
1631	if (!spa_writeable(spa))
1632		return;
1633#endif
1634
1635	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1636
1637	if (sav->sav_config != NULL) {
1638		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1639		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1640		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1641	} else {
1642		nl2cache = 0;
1643		newvdevs = NULL;
1644	}
1645
1646	oldvdevs = sav->sav_vdevs;
1647	oldnvdevs = sav->sav_count;
1648	sav->sav_vdevs = NULL;
1649	sav->sav_count = 0;
1650
1651	/*
1652	 * Process new nvlist of vdevs.
1653	 */
1654	for (i = 0; i < nl2cache; i++) {
1655		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1656		    &guid) == 0);
1657
1658		newvdevs[i] = NULL;
1659		for (j = 0; j < oldnvdevs; j++) {
1660			vd = oldvdevs[j];
1661			if (vd != NULL && guid == vd->vdev_guid) {
1662				/*
1663				 * Retain previous vdev for add/remove ops.
1664				 */
1665				newvdevs[i] = vd;
1666				oldvdevs[j] = NULL;
1667				break;
1668			}
1669		}
1670
1671		if (newvdevs[i] == NULL) {
1672			/*
1673			 * Create new vdev
1674			 */
1675			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1676			    VDEV_ALLOC_L2CACHE) == 0);
1677			ASSERT(vd != NULL);
1678			newvdevs[i] = vd;
1679
1680			/*
1681			 * Commit this vdev as an l2cache device,
1682			 * even if it fails to open.
1683			 */
1684			spa_l2cache_add(vd);
1685
1686			vd->vdev_top = vd;
1687			vd->vdev_aux = sav;
1688
1689			spa_l2cache_activate(vd);
1690
1691			if (vdev_open(vd) != 0)
1692				continue;
1693
1694			(void) vdev_validate_aux(vd);
1695
1696			if (!vdev_is_dead(vd))
1697				l2arc_add_vdev(spa, vd);
1698		}
1699	}
1700
1701	/*
1702	 * Purge vdevs that were dropped
1703	 */
1704	for (i = 0; i < oldnvdevs; i++) {
1705		uint64_t pool;
1706
1707		vd = oldvdevs[i];
1708		if (vd != NULL) {
1709			ASSERT(vd->vdev_isl2cache);
1710
1711			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1712			    pool != 0ULL && l2arc_vdev_present(vd))
1713				l2arc_remove_vdev(vd);
1714			vdev_clear_stats(vd);
1715			vdev_free(vd);
1716		}
1717	}
1718
1719	if (oldvdevs)
1720		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1721
1722	if (sav->sav_config == NULL)
1723		goto out;
1724
1725	sav->sav_vdevs = newvdevs;
1726	sav->sav_count = (int)nl2cache;
1727
1728	/*
1729	 * Recompute the stashed list of l2cache devices, with status
1730	 * information this time.
1731	 */
1732	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1733	    DATA_TYPE_NVLIST_ARRAY) == 0);
1734
1735	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1736	for (i = 0; i < sav->sav_count; i++)
1737		l2cache[i] = vdev_config_generate(spa,
1738		    sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1739	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1740	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1741out:
1742	for (i = 0; i < sav->sav_count; i++)
1743		nvlist_free(l2cache[i]);
1744	if (sav->sav_count)
1745		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1746}
1747
1748static int
1749load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1750{
1751	dmu_buf_t *db;
1752	char *packed = NULL;
1753	size_t nvsize = 0;
1754	int error;
1755	*value = NULL;
1756
1757	error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1758	if (error != 0)
1759		return (error);
1760
1761	nvsize = *(uint64_t *)db->db_data;
1762	dmu_buf_rele(db, FTAG);
1763
1764	packed = kmem_alloc(nvsize, KM_SLEEP);
1765	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1766	    DMU_READ_PREFETCH);
1767	if (error == 0)
1768		error = nvlist_unpack(packed, nvsize, value, 0);
1769	kmem_free(packed, nvsize);
1770
1771	return (error);
1772}
1773
1774/*
1775 * Concrete top-level vdevs that are not missing and are not logs. At every
1776 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds.
1777 */
1778static uint64_t
1779spa_healthy_core_tvds(spa_t *spa)
1780{
1781	vdev_t *rvd = spa->spa_root_vdev;
1782	uint64_t tvds = 0;
1783
1784	for (uint64_t i = 0; i < rvd->vdev_children; i++) {
1785		vdev_t *vd = rvd->vdev_child[i];
1786		if (vd->vdev_islog)
1787			continue;
1788		if (vdev_is_concrete(vd) && !vdev_is_dead(vd))
1789			tvds++;
1790	}
1791
1792	return (tvds);
1793}
1794
1795/*
1796 * Checks to see if the given vdev could not be opened, in which case we post a
1797 * sysevent to notify the autoreplace code that the device has been removed.
1798 */
1799static void
1800spa_check_removed(vdev_t *vd)
1801{
1802	for (uint64_t c = 0; c < vd->vdev_children; c++)
1803		spa_check_removed(vd->vdev_child[c]);
1804
1805	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1806	    vdev_is_concrete(vd)) {
1807		zfs_post_autoreplace(vd->vdev_spa, vd);
1808		spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK);
1809	}
1810}
1811
1812static int
1813spa_check_for_missing_logs(spa_t *spa)
1814{
1815	vdev_t *rvd = spa->spa_root_vdev;
1816
1817	/*
1818	 * If we're doing a normal import, then build up any additional
1819	 * diagnostic information about missing log devices.
1820	 * We'll pass this up to the user for further processing.
1821	 */
1822	if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1823		nvlist_t **child, *nv;
1824		uint64_t idx = 0;
1825
1826		child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1827		    KM_SLEEP);
1828		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1829
1830		for (uint64_t c = 0; c < rvd->vdev_children; c++) {
1831			vdev_t *tvd = rvd->vdev_child[c];
1832
1833			/*
1834			 * We consider a device as missing only if it failed
1835			 * to open (i.e. offline or faulted is not considered
1836			 * as missing).
1837			 */
1838			if (tvd->vdev_islog &&
1839			    tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
1840				child[idx++] = vdev_config_generate(spa, tvd,
1841				    B_FALSE, VDEV_CONFIG_MISSING);
1842			}
1843		}
1844
1845		if (idx > 0) {
1846			fnvlist_add_nvlist_array(nv,
1847			    ZPOOL_CONFIG_CHILDREN, child, idx);
1848			fnvlist_add_nvlist(spa->spa_load_info,
1849			    ZPOOL_CONFIG_MISSING_DEVICES, nv);
1850
1851			for (uint64_t i = 0; i < idx; i++)
1852				nvlist_free(child[i]);
1853		}
1854		nvlist_free(nv);
1855		kmem_free(child, rvd->vdev_children * sizeof (char **));
1856
1857		if (idx > 0) {
1858			spa_load_failed(spa, "some log devices are missing");
1859			vdev_dbgmsg_print_tree(rvd, 2);
1860			return (SET_ERROR(ENXIO));
1861		}
1862	} else {
1863		for (uint64_t c = 0; c < rvd->vdev_children; c++) {
1864			vdev_t *tvd = rvd->vdev_child[c];
1865
1866			if (tvd->vdev_islog &&
1867			    tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
1868				spa_set_log_state(spa, SPA_LOG_CLEAR);
1869				spa_load_note(spa, "some log devices are "
1870				    "missing, ZIL is dropped.");
1871				vdev_dbgmsg_print_tree(rvd, 2);
1872				break;
1873			}
1874		}
1875	}
1876
1877	return (0);
1878}
1879
1880/*
1881 * Check for missing log devices
1882 */
1883static boolean_t
1884spa_check_logs(spa_t *spa)
1885{
1886	boolean_t rv = B_FALSE;
1887	dsl_pool_t *dp = spa_get_dsl(spa);
1888
1889	switch (spa->spa_log_state) {
1890	case SPA_LOG_MISSING:
1891		/* need to recheck in case slog has been restored */
1892	case SPA_LOG_UNKNOWN:
1893		rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1894		    zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
1895		if (rv)
1896			spa_set_log_state(spa, SPA_LOG_MISSING);
1897		break;
1898	}
1899	return (rv);
1900}
1901
1902static boolean_t
1903spa_passivate_log(spa_t *spa)
1904{
1905	vdev_t *rvd = spa->spa_root_vdev;
1906	boolean_t slog_found = B_FALSE;
1907
1908	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1909
1910	if (!spa_has_slogs(spa))
1911		return (B_FALSE);
1912
1913	for (int c = 0; c < rvd->vdev_children; c++) {
1914		vdev_t *tvd = rvd->vdev_child[c];
1915		metaslab_group_t *mg = tvd->vdev_mg;
1916
1917		if (tvd->vdev_islog) {
1918			metaslab_group_passivate(mg);
1919			slog_found = B_TRUE;
1920		}
1921	}
1922
1923	return (slog_found);
1924}
1925
1926static void
1927spa_activate_log(spa_t *spa)
1928{
1929	vdev_t *rvd = spa->spa_root_vdev;
1930
1931	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1932
1933	for (int c = 0; c < rvd->vdev_children; c++) {
1934		vdev_t *tvd = rvd->vdev_child[c];
1935		metaslab_group_t *mg = tvd->vdev_mg;
1936
1937		if (tvd->vdev_islog)
1938			metaslab_group_activate(mg);
1939	}
1940}
1941
1942int
1943spa_reset_logs(spa_t *spa)
1944{
1945	int error;
1946
1947	error = dmu_objset_find(spa_name(spa), zil_reset,
1948	    NULL, DS_FIND_CHILDREN);
1949	if (error == 0) {
1950		/*
1951		 * We successfully offlined the log device, sync out the
1952		 * current txg so that the "stubby" block can be removed
1953		 * by zil_sync().
1954		 */
1955		txg_wait_synced(spa->spa_dsl_pool, 0);
1956	}
1957	return (error);
1958}
1959
1960static void
1961spa_aux_check_removed(spa_aux_vdev_t *sav)
1962{
1963	for (int i = 0; i < sav->sav_count; i++)
1964		spa_check_removed(sav->sav_vdevs[i]);
1965}
1966
1967void
1968spa_claim_notify(zio_t *zio)
1969{
1970	spa_t *spa = zio->io_spa;
1971
1972	if (zio->io_error)
1973		return;
1974
1975	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1976	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1977		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1978	mutex_exit(&spa->spa_props_lock);
1979}
1980
1981typedef struct spa_load_error {
1982	uint64_t	sle_meta_count;
1983	uint64_t	sle_data_count;
1984} spa_load_error_t;
1985
1986static void
1987spa_load_verify_done(zio_t *zio)
1988{
1989	blkptr_t *bp = zio->io_bp;
1990	spa_load_error_t *sle = zio->io_private;
1991	dmu_object_type_t type = BP_GET_TYPE(bp);
1992	int error = zio->io_error;
1993	spa_t *spa = zio->io_spa;
1994
1995	abd_free(zio->io_abd);
1996	if (error) {
1997		if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1998		    type != DMU_OT_INTENT_LOG)
1999			atomic_inc_64(&sle->sle_meta_count);
2000		else
2001			atomic_inc_64(&sle->sle_data_count);
2002	}
2003
2004	mutex_enter(&spa->spa_scrub_lock);
2005	spa->spa_load_verify_ios--;
2006	cv_broadcast(&spa->spa_scrub_io_cv);
2007	mutex_exit(&spa->spa_scrub_lock);
2008}
2009
2010/*
2011 * Maximum number of concurrent scrub i/os to create while verifying
2012 * a pool while importing it.
2013 */
2014int spa_load_verify_maxinflight = 10000;
2015boolean_t spa_load_verify_metadata = B_TRUE;
2016boolean_t spa_load_verify_data = B_TRUE;
2017
2018/*ARGSUSED*/
2019static int
2020spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
2021    const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
2022{
2023	if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
2024		return (0);
2025	/*
2026	 * Note: normally this routine will not be called if
2027	 * spa_load_verify_metadata is not set.  However, it may be useful
2028	 * to manually set the flag after the traversal has begun.
2029	 */
2030	if (!spa_load_verify_metadata)
2031		return (0);
2032	if (!BP_IS_METADATA(bp) && !spa_load_verify_data)
2033		return (0);
2034
2035	zio_t *rio = arg;
2036	size_t size = BP_GET_PSIZE(bp);
2037
2038	mutex_enter(&spa->spa_scrub_lock);
2039	while (spa->spa_load_verify_ios >= spa_load_verify_maxinflight)
2040		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2041	spa->spa_load_verify_ios++;
2042	mutex_exit(&spa->spa_scrub_lock);
2043
2044	zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size,
2045	    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
2046	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
2047	    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
2048	return (0);
2049}
2050
2051/* ARGSUSED */
2052int
2053verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
2054{
2055	if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN)
2056		return (SET_ERROR(ENAMETOOLONG));
2057
2058	return (0);
2059}
2060
2061static int
2062spa_load_verify(spa_t *spa)
2063{
2064	zio_t *rio;
2065	spa_load_error_t sle = { 0 };
2066	zpool_load_policy_t policy;
2067	boolean_t verify_ok = B_FALSE;
2068	int error = 0;
2069
2070	zpool_get_load_policy(spa->spa_config, &policy);
2071
2072	if (policy.zlp_rewind & ZPOOL_NEVER_REWIND)
2073		return (0);
2074
2075	dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
2076	error = dmu_objset_find_dp(spa->spa_dsl_pool,
2077	    spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL,
2078	    DS_FIND_CHILDREN);
2079	dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
2080	if (error != 0)
2081		return (error);
2082
2083	rio = zio_root(spa, NULL, &sle,
2084	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
2085
2086	if (spa_load_verify_metadata) {
2087		if (spa->spa_extreme_rewind) {
2088			spa_load_note(spa, "performing a complete scan of the "
2089			    "pool since extreme rewind is on. This may take "
2090			    "a very long time.\n  (spa_load_verify_data=%u, "
2091			    "spa_load_verify_metadata=%u)",
2092			    spa_load_verify_data, spa_load_verify_metadata);
2093		}
2094		error = traverse_pool(spa, spa->spa_verify_min_txg,
2095		    TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA |
2096		    TRAVERSE_NO_DECRYPT, spa_load_verify_cb, rio);
2097	}
2098
2099	(void) zio_wait(rio);
2100
2101	spa->spa_load_meta_errors = sle.sle_meta_count;
2102	spa->spa_load_data_errors = sle.sle_data_count;
2103
2104	if (sle.sle_meta_count != 0 || sle.sle_data_count != 0) {
2105		spa_load_note(spa, "spa_load_verify found %llu metadata errors "
2106		    "and %llu data errors", (u_longlong_t)sle.sle_meta_count,
2107		    (u_longlong_t)sle.sle_data_count);
2108	}
2109
2110	if (spa_load_verify_dryrun ||
2111	    (!error && sle.sle_meta_count <= policy.zlp_maxmeta &&
2112	    sle.sle_data_count <= policy.zlp_maxdata)) {
2113		int64_t loss = 0;
2114
2115		verify_ok = B_TRUE;
2116		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
2117		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
2118
2119		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
2120		VERIFY(nvlist_add_uint64(spa->spa_load_info,
2121		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
2122		VERIFY(nvlist_add_int64(spa->spa_load_info,
2123		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
2124		VERIFY(nvlist_add_uint64(spa->spa_load_info,
2125		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
2126	} else {
2127		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2128	}
2129
2130	if (spa_load_verify_dryrun)
2131		return (0);
2132
2133	if (error) {
2134		if (error != ENXIO && error != EIO)
2135			error = SET_ERROR(EIO);
2136		return (error);
2137	}
2138
2139	return (verify_ok ? 0 : EIO);
2140}
2141
2142/*
2143 * Find a value in the pool props object.
2144 */
2145static void
2146spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2147{
2148	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2149	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2150}
2151
2152/*
2153 * Find a value in the pool directory object.
2154 */
2155static int
2156spa_dir_prop(spa_t *spa, const char *name, uint64_t *val, boolean_t log_enoent)
2157{
2158	int error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2159	    name, sizeof (uint64_t), 1, val);
2160
2161	if (error != 0 && (error != ENOENT || log_enoent)) {
2162		spa_load_failed(spa, "couldn't get '%s' value in MOS directory "
2163		    "[error=%d]", name, error);
2164	}
2165
2166	return (error);
2167}
2168
2169static int
2170spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2171{
2172	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2173	return (SET_ERROR(err));
2174}
2175
2176static void
2177spa_spawn_aux_threads(spa_t *spa)
2178{
2179	ASSERT(spa_writeable(spa));
2180
2181	ASSERT(MUTEX_HELD(&spa_namespace_lock));
2182
2183	spa_start_indirect_condensing_thread(spa);
2184
2185	ASSERT3P(spa->spa_checkpoint_discard_zthr, ==, NULL);
2186	spa->spa_checkpoint_discard_zthr =
2187	    zthr_create(spa_checkpoint_discard_thread_check,
2188	    spa_checkpoint_discard_thread, spa);
2189}
2190
2191/*
2192 * Fix up config after a partly-completed split.  This is done with the
2193 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
2194 * pool have that entry in their config, but only the splitting one contains
2195 * a list of all the guids of the vdevs that are being split off.
2196 *
2197 * This function determines what to do with that list: either rejoin
2198 * all the disks to the pool, or complete the splitting process.  To attempt
2199 * the rejoin, each disk that is offlined is marked online again, and
2200 * we do a reopen() call.  If the vdev label for every disk that was
2201 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2202 * then we call vdev_split() on each disk, and complete the split.
2203 *
2204 * Otherwise we leave the config alone, with all the vdevs in place in
2205 * the original pool.
2206 */
2207static void
2208spa_try_repair(spa_t *spa, nvlist_t *config)
2209{
2210	uint_t extracted;
2211	uint64_t *glist;
2212	uint_t i, gcount;
2213	nvlist_t *nvl;
2214	vdev_t **vd;
2215	boolean_t attempt_reopen;
2216
2217	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2218		return;
2219
2220	/* check that the config is complete */
2221	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2222	    &glist, &gcount) != 0)
2223		return;
2224
2225	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2226
2227	/* attempt to online all the vdevs & validate */
2228	attempt_reopen = B_TRUE;
2229	for (i = 0; i < gcount; i++) {
2230		if (glist[i] == 0)	/* vdev is hole */
2231			continue;
2232
2233		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2234		if (vd[i] == NULL) {
2235			/*
2236			 * Don't bother attempting to reopen the disks;
2237			 * just do the split.
2238			 */
2239			attempt_reopen = B_FALSE;
2240		} else {
2241			/* attempt to re-online it */
2242			vd[i]->vdev_offline = B_FALSE;
2243		}
2244	}
2245
2246	if (attempt_reopen) {
2247		vdev_reopen(spa->spa_root_vdev);
2248
2249		/* check each device to see what state it's in */
2250		for (extracted = 0, i = 0; i < gcount; i++) {
2251			if (vd[i] != NULL &&
2252			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2253				break;
2254			++extracted;
2255		}
2256	}
2257
2258	/*
2259	 * If every disk has been moved to the new pool, or if we never
2260	 * even attempted to look at them, then we split them off for
2261	 * good.
2262	 */
2263	if (!attempt_reopen || gcount == extracted) {
2264		for (i = 0; i < gcount; i++)
2265			if (vd[i] != NULL)
2266				vdev_split(vd[i]);
2267		vdev_reopen(spa->spa_root_vdev);
2268	}
2269
2270	kmem_free(vd, gcount * sizeof (vdev_t *));
2271}
2272
2273static int
2274spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type)
2275{
2276	char *ereport = FM_EREPORT_ZFS_POOL;
2277	int error;
2278
2279	spa->spa_load_state = state;
2280
2281	gethrestime(&spa->spa_loaded_ts);
2282	error = spa_load_impl(spa, type, &ereport);
2283
2284	/*
2285	 * Don't count references from objsets that are already closed
2286	 * and are making their way through the eviction process.
2287	 */
2288	spa_evicting_os_wait(spa);
2289	spa->spa_minref = zfs_refcount_count(&spa->spa_refcount);
2290	if (error) {
2291		if (error != EEXIST) {
2292			spa->spa_loaded_ts.tv_sec = 0;
2293			spa->spa_loaded_ts.tv_nsec = 0;
2294		}
2295		if (error != EBADF) {
2296			zfs_ereport_post(ereport, spa, NULL, NULL, NULL, 0, 0);
2297		}
2298	}
2299	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2300	spa->spa_ena = 0;
2301
2302	return (error);
2303}
2304
2305/*
2306 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2307 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2308 * spa's per-vdev ZAP list.
2309 */
2310static uint64_t
2311vdev_count_verify_zaps(vdev_t *vd)
2312{
2313	spa_t *spa = vd->vdev_spa;
2314	uint64_t total = 0;
2315	if (vd->vdev_top_zap != 0) {
2316		total++;
2317		ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2318		    spa->spa_all_vdev_zaps, vd->vdev_top_zap));
2319	}
2320	if (vd->vdev_leaf_zap != 0) {
2321		total++;
2322		ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2323		    spa->spa_all_vdev_zaps, vd->vdev_leaf_zap));
2324	}
2325
2326	for (uint64_t i = 0; i < vd->vdev_children; i++) {
2327		total += vdev_count_verify_zaps(vd->vdev_child[i]);
2328	}
2329
2330	return (total);
2331}
2332
2333/*
2334 * Determine whether the activity check is required.
2335 */
2336static boolean_t
2337spa_activity_check_required(spa_t *spa, uberblock_t *ub, nvlist_t *label,
2338    nvlist_t *config)
2339{
2340	uint64_t state = 0;
2341	uint64_t hostid = 0;
2342	uint64_t tryconfig_txg = 0;
2343	uint64_t tryconfig_timestamp = 0;
2344	nvlist_t *nvinfo;
2345
2346	if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) {
2347		nvinfo = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO);
2348		(void) nvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG,
2349		    &tryconfig_txg);
2350		(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
2351		    &tryconfig_timestamp);
2352	}
2353
2354	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE, &state);
2355
2356	/*
2357	 * Disable the MMP activity check - This is used by zdb which
2358	 * is intended to be used on potentially active pools.
2359	 */
2360	if (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP)
2361		return (B_FALSE);
2362
2363	/*
2364	 * Skip the activity check when the MMP feature is disabled.
2365	 */
2366	if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay == 0)
2367		return (B_FALSE);
2368	/*
2369	 * If the tryconfig_* values are nonzero, they are the results of an
2370	 * earlier tryimport.  If they match the uberblock we just found, then
2371	 * the pool has not changed and we return false so we do not test a
2372	 * second time.
2373	 */
2374	if (tryconfig_txg && tryconfig_txg == ub->ub_txg &&
2375	    tryconfig_timestamp && tryconfig_timestamp == ub->ub_timestamp)
2376		return (B_FALSE);
2377
2378	/*
2379	 * Allow the activity check to be skipped when importing the pool
2380	 * on the same host which last imported it.  Since the hostid from
2381	 * configuration may be stale use the one read from the label.
2382	 */
2383	if (nvlist_exists(label, ZPOOL_CONFIG_HOSTID))
2384		hostid = fnvlist_lookup_uint64(label, ZPOOL_CONFIG_HOSTID);
2385
2386	if (hostid == spa_get_hostid())
2387		return (B_FALSE);
2388
2389	/*
2390	 * Skip the activity test when the pool was cleanly exported.
2391	 */
2392	if (state != POOL_STATE_ACTIVE)
2393		return (B_FALSE);
2394
2395	return (B_TRUE);
2396}
2397
2398/*
2399 * Perform the import activity check.  If the user canceled the import or
2400 * we detected activity then fail.
2401 */
2402static int
2403spa_activity_check(spa_t *spa, uberblock_t *ub, nvlist_t *config)
2404{
2405	uint64_t import_intervals = MAX(zfs_multihost_import_intervals, 1);
2406	uint64_t txg = ub->ub_txg;
2407	uint64_t timestamp = ub->ub_timestamp;
2408	uint64_t import_delay = NANOSEC;
2409	hrtime_t import_expire;
2410	nvlist_t *mmp_label = NULL;
2411	vdev_t *rvd = spa->spa_root_vdev;
2412	kcondvar_t cv;
2413	kmutex_t mtx;
2414	int error = 0;
2415
2416	cv_init(&cv, NULL, CV_DEFAULT, NULL);
2417	mutex_init(&mtx, NULL, MUTEX_DEFAULT, NULL);
2418	mutex_enter(&mtx);
2419
2420	/*
2421	 * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed
2422	 * during the earlier tryimport.  If the txg recorded there is 0 then
2423	 * the pool is known to be active on another host.
2424	 *
2425	 * Otherwise, the pool might be in use on another node.  Check for
2426	 * changes in the uberblocks on disk if necessary.
2427	 */
2428	if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) {
2429		nvlist_t *nvinfo = fnvlist_lookup_nvlist(config,
2430		    ZPOOL_CONFIG_LOAD_INFO);
2431
2432		if (nvlist_exists(nvinfo, ZPOOL_CONFIG_MMP_TXG) &&
2433		    fnvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG) == 0) {
2434			vdev_uberblock_load(rvd, ub, &mmp_label);
2435			error = SET_ERROR(EREMOTEIO);
2436			goto out;
2437		}
2438	}
2439
2440	/*
2441	 * Preferentially use the zfs_multihost_interval from the node which
2442	 * last imported the pool.  This value is stored in an MMP uberblock as.
2443	 *
2444	 * ub_mmp_delay * vdev_count_leaves() == zfs_multihost_interval
2445	 */
2446	if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay)
2447		import_delay = MAX(import_delay, import_intervals *
2448		    ub->ub_mmp_delay * MAX(vdev_count_leaves(spa), 1));
2449
2450	/* Apply a floor using the local default values. */
2451	import_delay = MAX(import_delay, import_intervals *
2452	    MSEC2NSEC(MAX(zfs_multihost_interval, MMP_MIN_INTERVAL)));
2453
2454	zfs_dbgmsg("import_delay=%llu ub_mmp_delay=%llu import_intervals=%u "
2455	    "leaves=%u", import_delay, ub->ub_mmp_delay, import_intervals,
2456	    vdev_count_leaves(spa));
2457
2458	/* Add a small random factor in case of simultaneous imports (0-25%) */
2459	import_expire = gethrtime() + import_delay +
2460	    (import_delay * spa_get_random(250) / 1000);
2461
2462	while (gethrtime() < import_expire) {
2463		vdev_uberblock_load(rvd, ub, &mmp_label);
2464
2465		if (txg != ub->ub_txg || timestamp != ub->ub_timestamp) {
2466			error = SET_ERROR(EREMOTEIO);
2467			break;
2468		}
2469
2470		if (mmp_label) {
2471			nvlist_free(mmp_label);
2472			mmp_label = NULL;
2473		}
2474
2475		error = cv_timedwait_sig(&cv, &mtx, ddi_get_lbolt() + hz);
2476		if (error != -1) {
2477			error = SET_ERROR(EINTR);
2478			break;
2479		}
2480		error = 0;
2481	}
2482
2483out:
2484	mutex_exit(&mtx);
2485	mutex_destroy(&mtx);
2486	cv_destroy(&cv);
2487
2488	/*
2489	 * If the pool is determined to be active store the status in the
2490	 * spa->spa_load_info nvlist.  If the remote hostname or hostid are
2491	 * available from configuration read from disk store them as well.
2492	 * This allows 'zpool import' to generate a more useful message.
2493	 *
2494	 * ZPOOL_CONFIG_MMP_STATE    - observed pool status (mandatory)
2495	 * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool
2496	 * ZPOOL_CONFIG_MMP_HOSTID   - hostid from the active pool
2497	 */
2498	if (error == EREMOTEIO) {
2499		char *hostname = "<unknown>";
2500		uint64_t hostid = 0;
2501
2502		if (mmp_label) {
2503			if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTNAME)) {
2504				hostname = fnvlist_lookup_string(mmp_label,
2505				    ZPOOL_CONFIG_HOSTNAME);
2506				fnvlist_add_string(spa->spa_load_info,
2507				    ZPOOL_CONFIG_MMP_HOSTNAME, hostname);
2508			}
2509
2510			if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTID)) {
2511				hostid = fnvlist_lookup_uint64(mmp_label,
2512				    ZPOOL_CONFIG_HOSTID);
2513				fnvlist_add_uint64(spa->spa_load_info,
2514				    ZPOOL_CONFIG_MMP_HOSTID, hostid);
2515			}
2516		}
2517
2518		fnvlist_add_uint64(spa->spa_load_info,
2519		    ZPOOL_CONFIG_MMP_STATE, MMP_STATE_ACTIVE);
2520		fnvlist_add_uint64(spa->spa_load_info,
2521		    ZPOOL_CONFIG_MMP_TXG, 0);
2522
2523		error = spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO);
2524	}
2525
2526	if (mmp_label)
2527		nvlist_free(mmp_label);
2528
2529	return (error);
2530}
2531
2532static int
2533spa_verify_host(spa_t *spa, nvlist_t *mos_config)
2534{
2535	uint64_t hostid;
2536	char *hostname;
2537	uint64_t myhostid = 0;
2538
2539	if (!spa_is_root(spa) && nvlist_lookup_uint64(mos_config,
2540	    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2541		hostname = fnvlist_lookup_string(mos_config,
2542		    ZPOOL_CONFIG_HOSTNAME);
2543
2544		myhostid = zone_get_hostid(NULL);
2545
2546		if (hostid != 0 && myhostid != 0 && hostid != myhostid) {
2547			cmn_err(CE_WARN, "pool '%s' could not be "
2548			    "loaded as it was last accessed by "
2549			    "another system (host: %s hostid: 0x%llx). "
2550			    "See: http://illumos.org/msg/ZFS-8000-EY",
2551			    spa_name(spa), hostname, (u_longlong_t)hostid);
2552			spa_load_failed(spa, "hostid verification failed: pool "
2553			    "last accessed by host: %s (hostid: 0x%llx)",
2554			    hostname, (u_longlong_t)hostid);
2555			return (SET_ERROR(EBADF));
2556		}
2557	}
2558
2559	return (0);
2560}
2561
2562static int
2563spa_ld_parse_config(spa_t *spa, spa_import_type_t type)
2564{
2565	int error = 0;
2566	nvlist_t *nvtree, *nvl, *config = spa->spa_config;
2567	int parse;
2568	vdev_t *rvd;
2569	uint64_t pool_guid;
2570	char *comment;
2571
2572	/*
2573	 * Versioning wasn't explicitly added to the label until later, so if
2574	 * it's not present treat it as the initial version.
2575	 */
2576	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2577	    &spa->spa_ubsync.ub_version) != 0)
2578		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2579
2580	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
2581		spa_load_failed(spa, "invalid config provided: '%s' missing",
2582		    ZPOOL_CONFIG_POOL_GUID);
2583		return (SET_ERROR(EINVAL));
2584	}
2585
2586	/*
2587	 * If we are doing an import, ensure that the pool is not already
2588	 * imported by checking if its pool guid already exists in the
2589	 * spa namespace.
2590	 *
2591	 * The only case that we allow an already imported pool to be
2592	 * imported again, is when the pool is checkpointed and we want to
2593	 * look at its checkpointed state from userland tools like zdb.
2594	 */
2595#ifdef _KERNEL
2596	if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
2597	    spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
2598	    spa_guid_exists(pool_guid, 0)) {
2599#else
2600	if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
2601	    spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
2602	    spa_guid_exists(pool_guid, 0) &&
2603	    !spa_importing_readonly_checkpoint(spa)) {
2604#endif
2605		spa_load_failed(spa, "a pool with guid %llu is already open",
2606		    (u_longlong_t)pool_guid);
2607		return (SET_ERROR(EEXIST));
2608	}
2609
2610	spa->spa_config_guid = pool_guid;
2611
2612	nvlist_free(spa->spa_load_info);
2613	spa->spa_load_info = fnvlist_alloc();
2614
2615	ASSERT(spa->spa_comment == NULL);
2616	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2617		spa->spa_comment = spa_strdup(comment);
2618
2619	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2620	    &spa->spa_config_txg);
2621
2622	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) == 0)
2623		spa->spa_config_splitting = fnvlist_dup(nvl);
2624
2625	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtree)) {
2626		spa_load_failed(spa, "invalid config provided: '%s' missing",
2627		    ZPOOL_CONFIG_VDEV_TREE);
2628		return (SET_ERROR(EINVAL));
2629	}
2630
2631	/*
2632	 * Create "The Godfather" zio to hold all async IOs
2633	 */
2634	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2635	    KM_SLEEP);
2636	for (int i = 0; i < max_ncpus; i++) {
2637		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2638		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2639		    ZIO_FLAG_GODFATHER);
2640	}
2641
2642	/*
2643	 * Parse the configuration into a vdev tree.  We explicitly set the
2644	 * value that will be returned by spa_version() since parsing the
2645	 * configuration requires knowing the version number.
2646	 */
2647	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2648	parse = (type == SPA_IMPORT_EXISTING ?
2649	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2650	error = spa_config_parse(spa, &rvd, nvtree, NULL, 0, parse);
2651	spa_config_exit(spa, SCL_ALL, FTAG);
2652
2653	if (error != 0) {
2654		spa_load_failed(spa, "unable to parse config [error=%d]",
2655		    error);
2656		return (error);
2657	}
2658
2659	ASSERT(spa->spa_root_vdev == rvd);
2660	ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
2661	ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
2662
2663	if (type != SPA_IMPORT_ASSEMBLE) {
2664		ASSERT(spa_guid(spa) == pool_guid);
2665	}
2666
2667	return (0);
2668}
2669
2670/*
2671 * Recursively open all vdevs in the vdev tree. This function is called twice:
2672 * first with the untrusted config, then with the trusted config.
2673 */
2674static int
2675spa_ld_open_vdevs(spa_t *spa)
2676{
2677	int error = 0;
2678
2679	/*
2680	 * spa_missing_tvds_allowed defines how many top-level vdevs can be
2681	 * missing/unopenable for the root vdev to be still considered openable.
2682	 */
2683	if (spa->spa_trust_config) {
2684		spa->spa_missing_tvds_allowed = zfs_max_missing_tvds;
2685	} else if (spa->spa_config_source == SPA_CONFIG_SRC_CACHEFILE) {
2686		spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_cachefile;
2687	} else if (spa->spa_config_source == SPA_CONFIG_SRC_SCAN) {
2688		spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_scan;
2689	} else {
2690		spa->spa_missing_tvds_allowed = 0;
2691	}
2692
2693	spa->spa_missing_tvds_allowed =
2694	    MAX(zfs_max_missing_tvds, spa->spa_missing_tvds_allowed);
2695
2696	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2697	error = vdev_open(spa->spa_root_vdev);
2698	spa_config_exit(spa, SCL_ALL, FTAG);
2699
2700	if (spa->spa_missing_tvds != 0) {
2701		spa_load_note(spa, "vdev tree has %lld missing top-level "
2702		    "vdevs.", (u_longlong_t)spa->spa_missing_tvds);
2703		if (spa->spa_trust_config && (spa->spa_mode & FWRITE)) {
2704			/*
2705			 * Although theoretically we could allow users to open
2706			 * incomplete pools in RW mode, we'd need to add a lot
2707			 * of extra logic (e.g. adjust pool space to account
2708			 * for missing vdevs).
2709			 * This limitation also prevents users from accidentally
2710			 * opening the pool in RW mode during data recovery and
2711			 * damaging it further.
2712			 */
2713			spa_load_note(spa, "pools with missing top-level "
2714			    "vdevs can only be opened in read-only mode.");
2715			error = SET_ERROR(ENXIO);
2716		} else {
2717			spa_load_note(spa, "current settings allow for maximum "
2718			    "%lld missing top-level vdevs at this stage.",
2719			    (u_longlong_t)spa->spa_missing_tvds_allowed);
2720		}
2721	}
2722	if (error != 0) {
2723		spa_load_failed(spa, "unable to open vdev tree [error=%d]",
2724		    error);
2725	}
2726	if (spa->spa_missing_tvds != 0 || error != 0)
2727		vdev_dbgmsg_print_tree(spa->spa_root_vdev, 2);
2728
2729	return (error);
2730}
2731
2732/*
2733 * We need to validate the vdev labels against the configuration that
2734 * we have in hand. This function is called twice: first with an untrusted
2735 * config, then with a trusted config. The validation is more strict when the
2736 * config is trusted.
2737 */
2738static int
2739spa_ld_validate_vdevs(spa_t *spa)
2740{
2741	int error = 0;
2742	vdev_t *rvd = spa->spa_root_vdev;
2743
2744	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2745	error = vdev_validate(rvd);
2746	spa_config_exit(spa, SCL_ALL, FTAG);
2747
2748	if (error != 0) {
2749		spa_load_failed(spa, "vdev_validate failed [error=%d]", error);
2750		return (error);
2751	}
2752
2753	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
2754		spa_load_failed(spa, "cannot open vdev tree after invalidating "
2755		    "some vdevs");
2756		vdev_dbgmsg_print_tree(rvd, 2);
2757		return (SET_ERROR(ENXIO));
2758	}
2759
2760	return (0);
2761}
2762
2763static void
2764spa_ld_select_uberblock_done(spa_t *spa, uberblock_t *ub)
2765{
2766	spa->spa_state = POOL_STATE_ACTIVE;
2767	spa->spa_ubsync = spa->spa_uberblock;
2768	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2769	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2770	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2771	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2772	spa->spa_claim_max_txg = spa->spa_first_txg;
2773	spa->spa_prev_software_version = ub->ub_software_version;
2774}
2775
2776static int
2777spa_ld_select_uberblock(spa_t *spa, spa_import_type_t type)
2778{
2779	vdev_t *rvd = spa->spa_root_vdev;
2780	nvlist_t *label;
2781	uberblock_t *ub = &spa->spa_uberblock;
2782	boolean_t activity_check = B_FALSE;
2783
2784	/*
2785	 * If we are opening the checkpointed state of the pool by
2786	 * rewinding to it, at this point we will have written the
2787	 * checkpointed uberblock to the vdev labels, so searching
2788	 * the labels will find the right uberblock.  However, if
2789	 * we are opening the checkpointed state read-only, we have
2790	 * not modified the labels. Therefore, we must ignore the
2791	 * labels and continue using the spa_uberblock that was set
2792	 * by spa_ld_checkpoint_rewind.
2793	 *
2794	 * Note that it would be fine to ignore the labels when
2795	 * rewinding (opening writeable) as well. However, if we
2796	 * crash just after writing the labels, we will end up
2797	 * searching the labels. Doing so in the common case means
2798	 * that this code path gets exercised normally, rather than
2799	 * just in the edge case.
2800	 */
2801	if (ub->ub_checkpoint_txg != 0 &&
2802	    spa_importing_readonly_checkpoint(spa)) {
2803		spa_ld_select_uberblock_done(spa, ub);
2804		return (0);
2805	}
2806
2807	/*
2808	 * Find the best uberblock.
2809	 */
2810	vdev_uberblock_load(rvd, ub, &label);
2811
2812	/*
2813	 * If we weren't able to find a single valid uberblock, return failure.
2814	 */
2815	if (ub->ub_txg == 0) {
2816		nvlist_free(label);
2817		spa_load_failed(spa, "no valid uberblock found");
2818		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2819	}
2820
2821	spa_load_note(spa, "using uberblock with txg=%llu",
2822	    (u_longlong_t)ub->ub_txg);
2823
2824	/*
2825	 * For pools which have the multihost property on determine if the
2826	 * pool is truly inactive and can be safely imported.  Prevent
2827	 * hosts which don't have a hostid set from importing the pool.
2828	 */
2829	activity_check = spa_activity_check_required(spa, ub, label,
2830	    spa->spa_config);
2831	if (activity_check) {
2832		if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay &&
2833		    spa_get_hostid() == 0) {
2834			nvlist_free(label);
2835			fnvlist_add_uint64(spa->spa_load_info,
2836			    ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID);
2837			return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO));
2838		}
2839
2840		int error = spa_activity_check(spa, ub, spa->spa_config);
2841		if (error) {
2842			nvlist_free(label);
2843			return (error);
2844		}
2845
2846		fnvlist_add_uint64(spa->spa_load_info,
2847		    ZPOOL_CONFIG_MMP_STATE, MMP_STATE_INACTIVE);
2848		fnvlist_add_uint64(spa->spa_load_info,
2849		    ZPOOL_CONFIG_MMP_TXG, ub->ub_txg);
2850	}
2851
2852	/*
2853	 * If the pool has an unsupported version we can't open it.
2854	 */
2855	if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2856		nvlist_free(label);
2857		spa_load_failed(spa, "version %llu is not supported",
2858		    (u_longlong_t)ub->ub_version);
2859		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2860	}
2861
2862	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2863		nvlist_t *features;
2864
2865		/*
2866		 * If we weren't able to find what's necessary for reading the
2867		 * MOS in the label, return failure.
2868		 */
2869		if (label == NULL) {
2870			spa_load_failed(spa, "label config unavailable");
2871			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2872			    ENXIO));
2873		}
2874
2875		if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_FEATURES_FOR_READ,
2876		    &features) != 0) {
2877			nvlist_free(label);
2878			spa_load_failed(spa, "invalid label: '%s' missing",
2879			    ZPOOL_CONFIG_FEATURES_FOR_READ);
2880			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2881			    ENXIO));
2882		}
2883
2884		/*
2885		 * Update our in-core representation with the definitive values
2886		 * from the label.
2887		 */
2888		nvlist_free(spa->spa_label_features);
2889		VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2890	}
2891
2892	nvlist_free(label);
2893
2894	/*
2895	 * Look through entries in the label nvlist's features_for_read. If
2896	 * there is a feature listed there which we don't understand then we
2897	 * cannot open a pool.
2898	 */
2899	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2900		nvlist_t *unsup_feat;
2901
2902		VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2903		    0);
2904
2905		for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2906		    NULL); nvp != NULL;
2907		    nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2908			if (!zfeature_is_supported(nvpair_name(nvp))) {
2909				VERIFY(nvlist_add_string(unsup_feat,
2910				    nvpair_name(nvp), "") == 0);
2911			}
2912		}
2913
2914		if (!nvlist_empty(unsup_feat)) {
2915			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2916			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2917			nvlist_free(unsup_feat);
2918			spa_load_failed(spa, "some features are unsupported");
2919			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2920			    ENOTSUP));
2921		}
2922
2923		nvlist_free(unsup_feat);
2924	}
2925
2926	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2927		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2928		spa_try_repair(spa, spa->spa_config);
2929		spa_config_exit(spa, SCL_ALL, FTAG);
2930		nvlist_free(spa->spa_config_splitting);
2931		spa->spa_config_splitting = NULL;
2932	}
2933
2934	/*
2935	 * Initialize internal SPA structures.
2936	 */
2937	spa_ld_select_uberblock_done(spa, ub);
2938
2939	return (0);
2940}
2941
2942static int
2943spa_ld_open_rootbp(spa_t *spa)
2944{
2945	int error = 0;
2946	vdev_t *rvd = spa->spa_root_vdev;
2947
2948	error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2949	if (error != 0) {
2950		spa_load_failed(spa, "unable to open rootbp in dsl_pool_init "
2951		    "[error=%d]", error);
2952		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2953	}
2954	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2955
2956	return (0);
2957}
2958
2959static int
2960spa_ld_trusted_config(spa_t *spa, spa_import_type_t type,
2961    boolean_t reloading)
2962{
2963	vdev_t *mrvd, *rvd = spa->spa_root_vdev;
2964	nvlist_t *nv, *mos_config, *policy;
2965	int error = 0, copy_error;
2966	uint64_t healthy_tvds, healthy_tvds_mos;
2967	uint64_t mos_config_txg;
2968
2969	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object, B_TRUE)
2970	    != 0)
2971		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2972
2973	/*
2974	 * If we're assembling a pool from a split, the config provided is
2975	 * already trusted so there is nothing to do.
2976	 */
2977	if (type == SPA_IMPORT_ASSEMBLE)
2978		return (0);
2979
2980	healthy_tvds = spa_healthy_core_tvds(spa);
2981
2982	if (load_nvlist(spa, spa->spa_config_object, &mos_config)
2983	    != 0) {
2984		spa_load_failed(spa, "unable to retrieve MOS config");
2985		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2986	}
2987
2988	/*
2989	 * If we are doing an open, pool owner wasn't verified yet, thus do
2990	 * the verification here.
2991	 */
2992	if (spa->spa_load_state == SPA_LOAD_OPEN) {
2993		error = spa_verify_host(spa, mos_config);
2994		if (error != 0) {
2995			nvlist_free(mos_config);
2996			return (error);
2997		}
2998	}
2999
3000	nv = fnvlist_lookup_nvlist(mos_config, ZPOOL_CONFIG_VDEV_TREE);
3001
3002	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3003
3004	/*
3005	 * Build a new vdev tree from the trusted config
3006	 */
3007	VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
3008
3009	/*
3010	 * Vdev paths in the MOS may be obsolete. If the untrusted config was
3011	 * obtained by scanning /dev/dsk, then it will have the right vdev
3012	 * paths. We update the trusted MOS config with this information.
3013	 * We first try to copy the paths with vdev_copy_path_strict, which
3014	 * succeeds only when both configs have exactly the same vdev tree.
3015	 * If that fails, we fall back to a more flexible method that has a
3016	 * best effort policy.
3017	 */
3018	copy_error = vdev_copy_path_strict(rvd, mrvd);
3019	if (copy_error != 0 || spa_load_print_vdev_tree) {
3020		spa_load_note(spa, "provided vdev tree:");
3021		vdev_dbgmsg_print_tree(rvd, 2);
3022		spa_load_note(spa, "MOS vdev tree:");
3023		vdev_dbgmsg_print_tree(mrvd, 2);
3024	}
3025	if (copy_error != 0) {
3026		spa_load_note(spa, "vdev_copy_path_strict failed, falling "
3027		    "back to vdev_copy_path_relaxed");
3028		vdev_copy_path_relaxed(rvd, mrvd);
3029	}
3030
3031	vdev_close(rvd);
3032	vdev_free(rvd);
3033	spa->spa_root_vdev = mrvd;
3034	rvd = mrvd;
3035	spa_config_exit(spa, SCL_ALL, FTAG);
3036
3037	/*
3038	 * We will use spa_config if we decide to reload the spa or if spa_load
3039	 * fails and we rewind. We must thus regenerate the config using the
3040	 * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to
3041	 * pass settings on how to load the pool and is not stored in the MOS.
3042	 * We copy it over to our new, trusted config.
3043	 */
3044	mos_config_txg = fnvlist_lookup_uint64(mos_config,
3045	    ZPOOL_CONFIG_POOL_TXG);
3046	nvlist_free(mos_config);
3047	mos_config = spa_config_generate(spa, NULL, mos_config_txg, B_FALSE);
3048	if (nvlist_lookup_nvlist(spa->spa_config, ZPOOL_LOAD_POLICY,
3049	    &policy) == 0)
3050		fnvlist_add_nvlist(mos_config, ZPOOL_LOAD_POLICY, policy);
3051	spa_config_set(spa, mos_config);
3052	spa->spa_config_source = SPA_CONFIG_SRC_MOS;
3053
3054	/*
3055	 * Now that we got the config from the MOS, we should be more strict
3056	 * in checking blkptrs and can make assumptions about the consistency
3057	 * of the vdev tree. spa_trust_config must be set to true before opening
3058	 * vdevs in order for them to be writeable.
3059	 */
3060	spa->spa_trust_config = B_TRUE;
3061
3062	/*
3063	 * Open and validate the new vdev tree
3064	 */
3065	error = spa_ld_open_vdevs(spa);
3066	if (error != 0)
3067		return (error);
3068
3069	error = spa_ld_validate_vdevs(spa);
3070	if (error != 0)
3071		return (error);
3072
3073	if (copy_error != 0 || spa_load_print_vdev_tree) {
3074		spa_load_note(spa, "final vdev tree:");
3075		vdev_dbgmsg_print_tree(rvd, 2);
3076	}
3077
3078	if (spa->spa_load_state != SPA_LOAD_TRYIMPORT &&
3079	    !spa->spa_extreme_rewind && zfs_max_missing_tvds == 0) {
3080		/*
3081		 * Sanity check to make sure that we are indeed loading the
3082		 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds
3083		 * in the config provided and they happened to be the only ones
3084		 * to have the latest uberblock, we could involuntarily perform
3085		 * an extreme rewind.
3086		 */
3087		healthy_tvds_mos = spa_healthy_core_tvds(spa);
3088		if (healthy_tvds_mos - healthy_tvds >=
3089		    SPA_SYNC_MIN_VDEVS) {
3090			spa_load_note(spa, "config provided misses too many "
3091			    "top-level vdevs compared to MOS (%lld vs %lld). ",
3092			    (u_longlong_t)healthy_tvds,
3093			    (u_longlong_t)healthy_tvds_mos);
3094			spa_load_note(spa, "vdev tree:");
3095			vdev_dbgmsg_print_tree(rvd, 2);
3096			if (reloading) {
3097				spa_load_failed(spa, "config was already "
3098				    "provided from MOS. Aborting.");
3099				return (spa_vdev_err(rvd,
3100				    VDEV_AUX_CORRUPT_DATA, EIO));
3101			}
3102			spa_load_note(spa, "spa must be reloaded using MOS "
3103			    "config");
3104			return (SET_ERROR(EAGAIN));
3105		}
3106	}
3107
3108	error = spa_check_for_missing_logs(spa);
3109	if (error != 0)
3110		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
3111
3112	if (rvd->vdev_guid_sum != spa->spa_uberblock.ub_guid_sum) {
3113		spa_load_failed(spa, "uberblock guid sum doesn't match MOS "
3114		    "guid sum (%llu != %llu)",
3115		    (u_longlong_t)spa->spa_uberblock.ub_guid_sum,
3116		    (u_longlong_t)rvd->vdev_guid_sum);
3117		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
3118		    ENXIO));
3119	}
3120
3121	return (0);
3122}
3123
3124static int
3125spa_ld_open_indirect_vdev_metadata(spa_t *spa)
3126{
3127	int error = 0;
3128	vdev_t *rvd = spa->spa_root_vdev;
3129
3130	/*
3131	 * Everything that we read before spa_remove_init() must be stored
3132	 * on concreted vdevs.  Therefore we do this as early as possible.
3133	 */
3134	error = spa_remove_init(spa);
3135	if (error != 0) {
3136		spa_load_failed(spa, "spa_remove_init failed [error=%d]",
3137		    error);
3138		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3139	}
3140
3141	/*
3142	 * Retrieve information needed to condense indirect vdev mappings.
3143	 */
3144	error = spa_condense_init(spa);
3145	if (error != 0) {
3146		spa_load_failed(spa, "spa_condense_init failed [error=%d]",
3147		    error);
3148		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
3149	}
3150
3151	return (0);
3152}
3153
3154static int
3155spa_ld_check_features(spa_t *spa, boolean_t *missing_feat_writep)
3156{
3157	int error = 0;
3158	vdev_t *rvd = spa->spa_root_vdev;
3159
3160	if (spa_version(spa) >= SPA_VERSION_FEATURES) {
3161		boolean_t missing_feat_read = B_FALSE;
3162		nvlist_t *unsup_feat, *enabled_feat;
3163
3164		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
3165		    &spa->spa_feat_for_read_obj, B_TRUE) != 0) {
3166			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3167		}
3168
3169		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
3170		    &spa->spa_feat_for_write_obj, B_TRUE) != 0) {
3171			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3172		}
3173
3174		if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
3175		    &spa->spa_feat_desc_obj, B_TRUE) != 0) {
3176			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3177		}
3178
3179		enabled_feat = fnvlist_alloc();
3180		unsup_feat = fnvlist_alloc();
3181
3182		if (!spa_features_check(spa, B_FALSE,
3183		    unsup_feat, enabled_feat))
3184			missing_feat_read = B_TRUE;
3185
3186		if (spa_writeable(spa) ||
3187		    spa->spa_load_state == SPA_LOAD_TRYIMPORT) {
3188			if (!spa_features_check(spa, B_TRUE,
3189			    unsup_feat, enabled_feat)) {
3190				*missing_feat_writep = B_TRUE;
3191			}
3192		}
3193
3194		fnvlist_add_nvlist(spa->spa_load_info,
3195		    ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
3196
3197		if (!nvlist_empty(unsup_feat)) {
3198			fnvlist_add_nvlist(spa->spa_load_info,
3199			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
3200		}
3201
3202		fnvlist_free(enabled_feat);
3203		fnvlist_free(unsup_feat);
3204
3205		if (!missing_feat_read) {
3206			fnvlist_add_boolean(spa->spa_load_info,
3207			    ZPOOL_CONFIG_CAN_RDONLY);
3208		}
3209
3210		/*
3211		 * If the state is SPA_LOAD_TRYIMPORT, our objective is
3212		 * twofold: to determine whether the pool is available for
3213		 * import in read-write mode and (if it is not) whether the
3214		 * pool is available for import in read-only mode. If the pool
3215		 * is available for import in read-write mode, it is displayed
3216		 * as available in userland; if it is not available for import
3217		 * in read-only mode, it is displayed as unavailable in
3218		 * userland. If the pool is available for import in read-only
3219		 * mode but not read-write mode, it is displayed as unavailable
3220		 * in userland with a special note that the pool is actually
3221		 * available for open in read-only mode.
3222		 *
3223		 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
3224		 * missing a feature for write, we must first determine whether
3225		 * the pool can be opened read-only before returning to
3226		 * userland in order to know whether to display the
3227		 * abovementioned note.
3228		 */
3229		if (missing_feat_read || (*missing_feat_writep &&
3230		    spa_writeable(spa))) {
3231			spa_load_failed(spa, "pool uses unsupported features");
3232			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
3233			    ENOTSUP));
3234		}
3235
3236		/*
3237		 * Load refcounts for ZFS features from disk into an in-memory
3238		 * cache during SPA initialization.
3239		 */
3240		for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
3241			uint64_t refcount;
3242
3243			error = feature_get_refcount_from_disk(spa,
3244			    &spa_feature_table[i], &refcount);
3245			if (error == 0) {
3246				spa->spa_feat_refcount_cache[i] = refcount;
3247			} else if (error == ENOTSUP) {
3248				spa->spa_feat_refcount_cache[i] =
3249				    SPA_FEATURE_DISABLED;
3250			} else {
3251				spa_load_failed(spa, "error getting refcount "
3252				    "for feature %s [error=%d]",
3253				    spa_feature_table[i].fi_guid, error);
3254				return (spa_vdev_err(rvd,
3255				    VDEV_AUX_CORRUPT_DATA, EIO));
3256			}
3257		}
3258	}
3259
3260	if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
3261		if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
3262		    &spa->spa_feat_enabled_txg_obj, B_TRUE) != 0)
3263			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3264	}
3265
3266	/*
3267	 * Encryption was added before bookmark_v2, even though bookmark_v2
3268	 * is now a dependency. If this pool has encryption enabled without
3269	 * bookmark_v2, trigger an errata message.
3270	 */
3271	if (spa_feature_is_enabled(spa, SPA_FEATURE_ENCRYPTION) &&
3272	    !spa_feature_is_enabled(spa, SPA_FEATURE_BOOKMARK_V2)) {
3273		spa->spa_errata = ZPOOL_ERRATA_ZOL_8308_ENCRYPTION;
3274	}
3275
3276	return (0);
3277}
3278
3279static int
3280spa_ld_load_special_directories(spa_t *spa)
3281{
3282	int error = 0;
3283	vdev_t *rvd = spa->spa_root_vdev;
3284
3285	spa->spa_is_initializing = B_TRUE;
3286	error = dsl_pool_open(spa->spa_dsl_pool);
3287	spa->spa_is_initializing = B_FALSE;
3288	if (error != 0) {
3289		spa_load_failed(spa, "dsl_pool_open failed [error=%d]", error);
3290		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3291	}
3292
3293	return (0);
3294}
3295
3296static int
3297spa_ld_get_props(spa_t *spa)
3298{
3299	int error = 0;
3300	uint64_t obj;
3301	vdev_t *rvd = spa->spa_root_vdev;
3302
3303	/* Grab the secret checksum salt from the MOS. */
3304	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3305	    DMU_POOL_CHECKSUM_SALT, 1,
3306	    sizeof (spa->spa_cksum_salt.zcs_bytes),
3307	    spa->spa_cksum_salt.zcs_bytes);
3308	if (error == ENOENT) {
3309		/* Generate a new salt for subsequent use */
3310		(void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
3311		    sizeof (spa->spa_cksum_salt.zcs_bytes));
3312	} else if (error != 0) {
3313		spa_load_failed(spa, "unable to retrieve checksum salt from "
3314		    "MOS [error=%d]", error);
3315		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3316	}
3317
3318	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj, B_TRUE) != 0)
3319		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3320	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
3321	if (error != 0) {
3322		spa_load_failed(spa, "error opening deferred-frees bpobj "
3323		    "[error=%d]", error);
3324		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3325	}
3326
3327	/*
3328	 * Load the bit that tells us to use the new accounting function
3329	 * (raid-z deflation).  If we have an older pool, this will not
3330	 * be present.
3331	 */
3332	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate, B_FALSE);
3333	if (error != 0 && error != ENOENT)
3334		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3335
3336	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
3337	    &spa->spa_creation_version, B_FALSE);
3338	if (error != 0 && error != ENOENT)
3339		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3340
3341	/*
3342	 * Load the persistent error log.  If we have an older pool, this will
3343	 * not be present.
3344	 */
3345	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last,
3346	    B_FALSE);
3347	if (error != 0 && error != ENOENT)
3348		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3349
3350	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
3351	    &spa->spa_errlog_scrub, B_FALSE);
3352	if (error != 0 && error != ENOENT)
3353		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3354
3355	/*
3356	 * Load the history object.  If we have an older pool, this
3357	 * will not be present.
3358	 */
3359	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history, B_FALSE);
3360	if (error != 0 && error != ENOENT)
3361		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3362
3363	/*
3364	 * Load the per-vdev ZAP map. If we have an older pool, this will not
3365	 * be present; in this case, defer its creation to a later time to
3366	 * avoid dirtying the MOS this early / out of sync context. See
3367	 * spa_sync_config_object.
3368	 */
3369
3370	/* The sentinel is only available in the MOS config. */
3371	nvlist_t *mos_config;
3372	if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) {
3373		spa_load_failed(spa, "unable to retrieve MOS config");
3374		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3375	}
3376
3377	error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP,
3378	    &spa->spa_all_vdev_zaps, B_FALSE);
3379
3380	if (error == ENOENT) {
3381		VERIFY(!nvlist_exists(mos_config,
3382		    ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
3383		spa->spa_avz_action = AVZ_ACTION_INITIALIZE;
3384		ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
3385	} else if (error != 0) {
3386		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3387	} else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) {
3388		/*
3389		 * An older version of ZFS overwrote the sentinel value, so
3390		 * we have orphaned per-vdev ZAPs in the MOS. Defer their
3391		 * destruction to later; see spa_sync_config_object.
3392		 */
3393		spa->spa_avz_action = AVZ_ACTION_DESTROY;
3394		/*
3395		 * We're assuming that no vdevs have had their ZAPs created
3396		 * before this. Better be sure of it.
3397		 */
3398		ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
3399	}
3400	nvlist_free(mos_config);
3401
3402	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3403
3404	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object,
3405	    B_FALSE);
3406	if (error && error != ENOENT)
3407		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3408
3409	if (error == 0) {
3410		uint64_t autoreplace;
3411
3412		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
3413		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
3414		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
3415		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
3416		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
3417		spa_prop_find(spa, ZPOOL_PROP_MULTIHOST, &spa->spa_multihost);
3418		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
3419		    &spa->spa_dedup_ditto);
3420
3421		spa->spa_autoreplace = (autoreplace != 0);
3422	}
3423
3424	/*
3425	 * If we are importing a pool with missing top-level vdevs,
3426	 * we enforce that the pool doesn't panic or get suspended on
3427	 * error since the likelihood of missing data is extremely high.
3428	 */
3429	if (spa->spa_missing_tvds > 0 &&
3430	    spa->spa_failmode != ZIO_FAILURE_MODE_CONTINUE &&
3431	    spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3432		spa_load_note(spa, "forcing failmode to 'continue' "
3433		    "as some top level vdevs are missing");
3434		spa->spa_failmode = ZIO_FAILURE_MODE_CONTINUE;
3435	}
3436
3437	return (0);
3438}
3439
3440static int
3441spa_ld_open_aux_vdevs(spa_t *spa, spa_import_type_t type)
3442{
3443	int error = 0;
3444	vdev_t *rvd = spa->spa_root_vdev;
3445
3446	/*
3447	 * If we're assembling the pool from the split-off vdevs of
3448	 * an existing pool, we don't want to attach the spares & cache
3449	 * devices.
3450	 */
3451
3452	/*
3453	 * Load any hot spares for this pool.
3454	 */
3455	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object,
3456	    B_FALSE);
3457	if (error != 0 && error != ENOENT)
3458		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3459	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
3460		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
3461		if (load_nvlist(spa, spa->spa_spares.sav_object,
3462		    &spa->spa_spares.sav_config) != 0) {
3463			spa_load_failed(spa, "error loading spares nvlist");
3464			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3465		}
3466
3467		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3468		spa_load_spares(spa);
3469		spa_config_exit(spa, SCL_ALL, FTAG);
3470	} else if (error == 0) {
3471		spa->spa_spares.sav_sync = B_TRUE;
3472	}
3473
3474	/*
3475	 * Load any level 2 ARC devices for this pool.
3476	 */
3477	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
3478	    &spa->spa_l2cache.sav_object, B_FALSE);
3479	if (error != 0 && error != ENOENT)
3480		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3481	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
3482		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
3483		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
3484		    &spa->spa_l2cache.sav_config) != 0) {
3485			spa_load_failed(spa, "error loading l2cache nvlist");
3486			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3487		}
3488
3489		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3490		spa_load_l2cache(spa);
3491		spa_config_exit(spa, SCL_ALL, FTAG);
3492	} else if (error == 0) {
3493		spa->spa_l2cache.sav_sync = B_TRUE;
3494	}
3495
3496	return (0);
3497}
3498
3499static int
3500spa_ld_load_vdev_metadata(spa_t *spa)
3501{
3502	int error = 0;
3503	vdev_t *rvd = spa->spa_root_vdev;
3504
3505	/*
3506	 * If the 'multihost' property is set, then never allow a pool to
3507	 * be imported when the system hostid is zero.  The exception to
3508	 * this rule is zdb which is always allowed to access pools.
3509	 */
3510	if (spa_multihost(spa) && spa_get_hostid() == 0 &&
3511	    (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP) == 0) {
3512		fnvlist_add_uint64(spa->spa_load_info,
3513		    ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID);
3514		return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO));
3515	}
3516
3517	/*
3518	 * If the 'autoreplace' property is set, then post a resource notifying
3519	 * the ZFS DE that it should not issue any faults for unopenable
3520	 * devices.  We also iterate over the vdevs, and post a sysevent for any
3521	 * unopenable vdevs so that the normal autoreplace handler can take
3522	 * over.
3523	 */
3524	if (spa->spa_autoreplace && spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3525		spa_check_removed(spa->spa_root_vdev);
3526		/*
3527		 * For the import case, this is done in spa_import(), because
3528		 * at this point we're using the spare definitions from
3529		 * the MOS config, not necessarily from the userland config.
3530		 */
3531		if (spa->spa_load_state != SPA_LOAD_IMPORT) {
3532			spa_aux_check_removed(&spa->spa_spares);
3533			spa_aux_check_removed(&spa->spa_l2cache);
3534		}
3535	}
3536
3537	/*
3538	 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc.
3539	 */
3540	error = vdev_load(rvd);
3541	if (error != 0) {
3542		spa_load_failed(spa, "vdev_load failed [error=%d]", error);
3543		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
3544	}
3545
3546	/*
3547	 * Propagate the leaf DTLs we just loaded all the way up the vdev tree.
3548	 */
3549	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3550	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
3551	spa_config_exit(spa, SCL_ALL, FTAG);
3552
3553	return (0);
3554}
3555
3556static int
3557spa_ld_load_dedup_tables(spa_t *spa)
3558{
3559	int error = 0;
3560	vdev_t *rvd = spa->spa_root_vdev;
3561
3562	error = ddt_load(spa);
3563	if (error != 0) {
3564		spa_load_failed(spa, "ddt_load failed [error=%d]", error);
3565		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3566	}
3567
3568	return (0);
3569}
3570
3571static int
3572spa_ld_verify_logs(spa_t *spa, spa_import_type_t type, char **ereport)
3573{
3574	vdev_t *rvd = spa->spa_root_vdev;
3575
3576	if (type != SPA_IMPORT_ASSEMBLE && spa_writeable(spa)) {
3577		boolean_t missing = spa_check_logs(spa);
3578		if (missing) {
3579			if (spa->spa_missing_tvds != 0) {
3580				spa_load_note(spa, "spa_check_logs failed "
3581				    "so dropping the logs");
3582			} else {
3583				*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
3584				spa_load_failed(spa, "spa_check_logs failed");
3585				return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG,
3586				    ENXIO));
3587			}
3588		}
3589	}
3590
3591	return (0);
3592}
3593
3594static int
3595spa_ld_verify_pool_data(spa_t *spa)
3596{
3597	int error = 0;
3598	vdev_t *rvd = spa->spa_root_vdev;
3599
3600	/*
3601	 * We've successfully opened the pool, verify that we're ready
3602	 * to start pushing transactions.
3603	 */
3604	if (spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3605		error = spa_load_verify(spa);
3606		if (error != 0) {
3607			spa_load_failed(spa, "spa_load_verify failed "
3608			    "[error=%d]", error);
3609			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
3610			    error));
3611		}
3612	}
3613
3614	return (0);
3615}
3616
3617static void
3618spa_ld_claim_log_blocks(spa_t *spa)
3619{
3620	dmu_tx_t *tx;
3621	dsl_pool_t *dp = spa_get_dsl(spa);
3622
3623	/*
3624	 * Claim log blocks that haven't been committed yet.
3625	 * This must all happen in a single txg.
3626	 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
3627	 * invoked from zil_claim_log_block()'s i/o done callback.
3628	 * Price of rollback is that we abandon the log.
3629	 */
3630	spa->spa_claiming = B_TRUE;
3631
3632	tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
3633	(void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
3634	    zil_claim, tx, DS_FIND_CHILDREN);
3635	dmu_tx_commit(tx);
3636
3637	spa->spa_claiming = B_FALSE;
3638
3639	spa_set_log_state(spa, SPA_LOG_GOOD);
3640}
3641
3642static void
3643spa_ld_check_for_config_update(spa_t *spa, uint64_t config_cache_txg,
3644    boolean_t update_config_cache)
3645{
3646	vdev_t *rvd = spa->spa_root_vdev;
3647	int need_update = B_FALSE;
3648
3649	/*
3650	 * If the config cache is stale, or we have uninitialized
3651	 * metaslabs (see spa_vdev_add()), then update the config.
3652	 *
3653	 * If this is a verbatim import, trust the current
3654	 * in-core spa_config and update the disk labels.
3655	 */
3656	if (update_config_cache || config_cache_txg != spa->spa_config_txg ||
3657	    spa->spa_load_state == SPA_LOAD_IMPORT ||
3658	    spa->spa_load_state == SPA_LOAD_RECOVER ||
3659	    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
3660		need_update = B_TRUE;
3661
3662	for (int c = 0; c < rvd->vdev_children; c++)
3663		if (rvd->vdev_child[c]->vdev_ms_array == 0)
3664			need_update = B_TRUE;
3665
3666	/*
3667	 * Update the config cache asychronously in case we're the
3668	 * root pool, in which case the config cache isn't writable yet.
3669	 */
3670	if (need_update)
3671		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
3672}
3673
3674static void
3675spa_ld_prepare_for_reload(spa_t *spa)
3676{
3677	int mode = spa->spa_mode;
3678	int async_suspended = spa->spa_async_suspended;
3679
3680	spa_unload(spa);
3681	spa_deactivate(spa);
3682	spa_activate(spa, mode);
3683
3684	/*
3685	 * We save the value of spa_async_suspended as it gets reset to 0 by
3686	 * spa_unload(). We want to restore it back to the original value before
3687	 * returning as we might be calling spa_async_resume() later.
3688	 */
3689	spa->spa_async_suspended = async_suspended;
3690}
3691
3692static int
3693spa_ld_read_checkpoint_txg(spa_t *spa)
3694{
3695	uberblock_t checkpoint;
3696	int error = 0;
3697
3698	ASSERT0(spa->spa_checkpoint_txg);
3699	ASSERT(MUTEX_HELD(&spa_namespace_lock));
3700
3701	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3702	    DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
3703	    sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
3704
3705	if (error == ENOENT)
3706		return (0);
3707
3708	if (error != 0)
3709		return (error);
3710
3711	ASSERT3U(checkpoint.ub_txg, !=, 0);
3712	ASSERT3U(checkpoint.ub_checkpoint_txg, !=, 0);
3713	ASSERT3U(checkpoint.ub_timestamp, !=, 0);
3714	spa->spa_checkpoint_txg = checkpoint.ub_txg;
3715	spa->spa_checkpoint_info.sci_timestamp = checkpoint.ub_timestamp;
3716
3717	return (0);
3718}
3719
3720static int
3721spa_ld_mos_init(spa_t *spa, spa_import_type_t type)
3722{
3723	int error = 0;
3724
3725	ASSERT(MUTEX_HELD(&spa_namespace_lock));
3726	ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
3727
3728	/*
3729	 * Never trust the config that is provided unless we are assembling
3730	 * a pool following a split.
3731	 * This means don't trust blkptrs and the vdev tree in general. This
3732	 * also effectively puts the spa in read-only mode since
3733	 * spa_writeable() checks for spa_trust_config to be true.
3734	 * We will later load a trusted config from the MOS.
3735	 */
3736	if (type != SPA_IMPORT_ASSEMBLE)
3737		spa->spa_trust_config = B_FALSE;
3738
3739	/*
3740	 * Parse the config provided to create a vdev tree.
3741	 */
3742	error = spa_ld_parse_config(spa, type);
3743	if (error != 0)
3744		return (error);
3745
3746	/*
3747	 * Now that we have the vdev tree, try to open each vdev. This involves
3748	 * opening the underlying physical device, retrieving its geometry and
3749	 * probing the vdev with a dummy I/O. The state of each vdev will be set
3750	 * based on the success of those operations. After this we'll be ready
3751	 * to read from the vdevs.
3752	 */
3753	error = spa_ld_open_vdevs(spa);
3754	if (error != 0)
3755		return (error);
3756
3757	/*
3758	 * Read the label of each vdev and make sure that the GUIDs stored
3759	 * there match the GUIDs in the config provided.
3760	 * If we're assembling a new pool that's been split off from an
3761	 * existing pool, the labels haven't yet been updated so we skip
3762	 * validation for now.
3763	 */
3764	if (type != SPA_IMPORT_ASSEMBLE) {
3765		error = spa_ld_validate_vdevs(spa);
3766		if (error != 0)
3767			return (error);
3768	}
3769
3770	/*
3771	 * Read all vdev labels to find the best uberblock (i.e. latest,
3772	 * unless spa_load_max_txg is set) and store it in spa_uberblock. We
3773	 * get the list of features required to read blkptrs in the MOS from
3774	 * the vdev label with the best uberblock and verify that our version
3775	 * of zfs supports them all.
3776	 */
3777	error = spa_ld_select_uberblock(spa, type);
3778	if (error != 0)
3779		return (error);
3780
3781	/*
3782	 * Pass that uberblock to the dsl_pool layer which will open the root
3783	 * blkptr. This blkptr points to the latest version of the MOS and will
3784	 * allow us to read its contents.
3785	 */
3786	error = spa_ld_open_rootbp(spa);
3787	if (error != 0)
3788		return (error);
3789
3790	return (0);
3791}
3792
3793static int
3794spa_ld_checkpoint_rewind(spa_t *spa)
3795{
3796	uberblock_t checkpoint;
3797	int error = 0;
3798
3799	ASSERT(MUTEX_HELD(&spa_namespace_lock));
3800	ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
3801
3802	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3803	    DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
3804	    sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
3805
3806	if (error != 0) {
3807		spa_load_failed(spa, "unable to retrieve checkpointed "
3808		    "uberblock from the MOS config [error=%d]", error);
3809
3810		if (error == ENOENT)
3811			error = ZFS_ERR_NO_CHECKPOINT;
3812
3813		return (error);
3814	}
3815
3816	ASSERT3U(checkpoint.ub_txg, <, spa->spa_uberblock.ub_txg);
3817	ASSERT3U(checkpoint.ub_txg, ==, checkpoint.ub_checkpoint_txg);
3818
3819	/*
3820	 * We need to update the txg and timestamp of the checkpointed
3821	 * uberblock to be higher than the latest one. This ensures that
3822	 * the checkpointed uberblock is selected if we were to close and
3823	 * reopen the pool right after we've written it in the vdev labels.
3824	 * (also see block comment in vdev_uberblock_compare)
3825	 */
3826	checkpoint.ub_txg = spa->spa_uberblock.ub_txg + 1;
3827	checkpoint.ub_timestamp = gethrestime_sec();
3828
3829	/*
3830	 * Set current uberblock to be the checkpointed uberblock.
3831	 */
3832	spa->spa_uberblock = checkpoint;
3833
3834	/*
3835	 * If we are doing a normal rewind, then the pool is open for
3836	 * writing and we sync the "updated" checkpointed uberblock to
3837	 * disk. Once this is done, we've basically rewound the whole
3838	 * pool and there is no way back.
3839	 *
3840	 * There are cases when we don't want to attempt and sync the
3841	 * checkpointed uberblock to disk because we are opening a
3842	 * pool as read-only. Specifically, verifying the checkpointed
3843	 * state with zdb, and importing the checkpointed state to get
3844	 * a "preview" of its content.
3845	 */
3846	if (spa_writeable(spa)) {
3847		vdev_t *rvd = spa->spa_root_vdev;
3848
3849		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3850		vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
3851		int svdcount = 0;
3852		int children = rvd->vdev_children;
3853		int c0 = spa_get_random(children);
3854
3855		for (int c = 0; c < children; c++) {
3856			vdev_t *vd = rvd->vdev_child[(c0 + c) % children];
3857
3858			/* Stop when revisiting the first vdev */
3859			if (c > 0 && svd[0] == vd)
3860				break;
3861
3862			if (vd->vdev_ms_array == 0 || vd->vdev_islog ||
3863			    !vdev_is_concrete(vd))
3864				continue;
3865
3866			svd[svdcount++] = vd;
3867			if (svdcount == SPA_SYNC_MIN_VDEVS)
3868				break;
3869		}
3870		error = vdev_config_sync(svd, svdcount, spa->spa_first_txg);
3871		if (error == 0)
3872			spa->spa_last_synced_guid = rvd->vdev_guid;
3873		spa_config_exit(spa, SCL_ALL, FTAG);
3874
3875		if (error != 0) {
3876			spa_load_failed(spa, "failed to write checkpointed "
3877			    "uberblock to the vdev labels [error=%d]", error);
3878			return (error);
3879		}
3880	}
3881
3882	return (0);
3883}
3884
3885static int
3886spa_ld_mos_with_trusted_config(spa_t *spa, spa_import_type_t type,
3887    boolean_t *update_config_cache)
3888{
3889	int error;
3890
3891	/*
3892	 * Parse the config for pool, open and validate vdevs,
3893	 * select an uberblock, and use that uberblock to open
3894	 * the MOS.
3895	 */
3896	error = spa_ld_mos_init(spa, type);
3897	if (error != 0)
3898		return (error);
3899
3900	/*
3901	 * Retrieve the trusted config stored in the MOS and use it to create
3902	 * a new, exact version of the vdev tree, then reopen all vdevs.
3903	 */
3904	error = spa_ld_trusted_config(spa, type, B_FALSE);
3905	if (error == EAGAIN) {
3906		if (update_config_cache != NULL)
3907			*update_config_cache = B_TRUE;
3908
3909		/*
3910		 * Redo the loading process with the trusted config if it is
3911		 * too different from the untrusted config.
3912		 */
3913		spa_ld_prepare_for_reload(spa);
3914		spa_load_note(spa, "RELOADING");
3915		error = spa_ld_mos_init(spa, type);
3916		if (error != 0)
3917			return (error);
3918
3919		error = spa_ld_trusted_config(spa, type, B_TRUE);
3920		if (error != 0)
3921			return (error);
3922
3923	} else if (error != 0) {
3924		return (error);
3925	}
3926
3927	return (0);
3928}
3929
3930/*
3931 * Load an existing storage pool, using the config provided. This config
3932 * describes which vdevs are part of the pool and is later validated against
3933 * partial configs present in each vdev's label and an entire copy of the
3934 * config stored in the MOS.
3935 */
3936static int
3937spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport)
3938{
3939	int error = 0;
3940	boolean_t missing_feat_write = B_FALSE;
3941	boolean_t checkpoint_rewind =
3942	    (spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
3943	boolean_t update_config_cache = B_FALSE;
3944
3945	ASSERT(MUTEX_HELD(&spa_namespace_lock));
3946	ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
3947
3948	spa_load_note(spa, "LOADING");
3949
3950	error = spa_ld_mos_with_trusted_config(spa, type, &update_config_cache);
3951	if (error != 0)
3952		return (error);
3953
3954	/*
3955	 * If we are rewinding to the checkpoint then we need to repeat
3956	 * everything we've done so far in this function but this time
3957	 * selecting the checkpointed uberblock and using that to open
3958	 * the MOS.
3959	 */
3960	if (checkpoint_rewind) {
3961		/*
3962		 * If we are rewinding to the checkpoint update config cache
3963		 * anyway.
3964		 */
3965		update_config_cache = B_TRUE;
3966
3967		/*
3968		 * Extract the checkpointed uberblock from the current MOS
3969		 * and use this as the pool's uberblock from now on. If the
3970		 * pool is imported as writeable we also write the checkpoint
3971		 * uberblock to the labels, making the rewind permanent.
3972		 */
3973		error = spa_ld_checkpoint_rewind(spa);
3974		if (error != 0)
3975			return (error);
3976
3977		/*
3978		 * Redo the loading process process again with the
3979		 * checkpointed uberblock.
3980		 */
3981		spa_ld_prepare_for_reload(spa);
3982		spa_load_note(spa, "LOADING checkpointed uberblock");
3983		error = spa_ld_mos_with_trusted_config(spa, type, NULL);
3984		if (error != 0)
3985			return (error);
3986	}
3987
3988	/*
3989	 * Retrieve the checkpoint txg if the pool has a checkpoint.
3990	 */
3991	error = spa_ld_read_checkpoint_txg(spa);
3992	if (error != 0)
3993		return (error);
3994
3995	/*
3996	 * Retrieve the mapping of indirect vdevs. Those vdevs were removed
3997	 * from the pool and their contents were re-mapped to other vdevs. Note
3998	 * that everything that we read before this step must have been
3999	 * rewritten on concrete vdevs after the last device removal was
4000	 * initiated. Otherwise we could be reading from indirect vdevs before
4001	 * we have loaded their mappings.
4002	 */
4003	error = spa_ld_open_indirect_vdev_metadata(spa);
4004	if (error != 0)
4005		return (error);
4006
4007	/*
4008	 * Retrieve the full list of active features from the MOS and check if
4009	 * they are all supported.
4010	 */
4011	error = spa_ld_check_features(spa, &missing_feat_write);
4012	if (error != 0)
4013		return (error);
4014
4015	/*
4016	 * Load several special directories from the MOS needed by the dsl_pool
4017	 * layer.
4018	 */
4019	error = spa_ld_load_special_directories(spa);
4020	if (error != 0)
4021		return (error);
4022
4023	/*
4024	 * Retrieve pool properties from the MOS.
4025	 */
4026	error = spa_ld_get_props(spa);
4027	if (error != 0)
4028		return (error);
4029
4030	/*
4031	 * Retrieve the list of auxiliary devices - cache devices and spares -
4032	 * and open them.
4033	 */
4034	error = spa_ld_open_aux_vdevs(spa, type);
4035	if (error != 0)
4036		return (error);
4037
4038	/*
4039	 * Load the metadata for all vdevs. Also check if unopenable devices
4040	 * should be autoreplaced.
4041	 */
4042	error = spa_ld_load_vdev_metadata(spa);
4043	if (error != 0)
4044		return (error);
4045
4046	error = spa_ld_load_dedup_tables(spa);
4047	if (error != 0)
4048		return (error);
4049
4050	/*
4051	 * Verify the logs now to make sure we don't have any unexpected errors
4052	 * when we claim log blocks later.
4053	 */
4054	error = spa_ld_verify_logs(spa, type, ereport);
4055	if (error != 0)
4056		return (error);
4057
4058	if (missing_feat_write) {
4059		ASSERT(spa->spa_load_state == SPA_LOAD_TRYIMPORT);
4060
4061		/*
4062		 * At this point, we know that we can open the pool in
4063		 * read-only mode but not read-write mode. We now have enough
4064		 * information and can return to userland.
4065		 */
4066		return (spa_vdev_err(spa->spa_root_vdev, VDEV_AUX_UNSUP_FEAT,
4067		    ENOTSUP));
4068	}
4069
4070	/*
4071	 * Traverse the last txgs to make sure the pool was left off in a safe
4072	 * state. When performing an extreme rewind, we verify the whole pool,
4073	 * which can take a very long time.
4074	 */
4075	error = spa_ld_verify_pool_data(spa);
4076	if (error != 0)
4077		return (error);
4078
4079	/*
4080	 * Calculate the deflated space for the pool. This must be done before
4081	 * we write anything to the pool because we'd need to update the space
4082	 * accounting using the deflated sizes.
4083	 */
4084	spa_update_dspace(spa);
4085
4086	/*
4087	 * We have now retrieved all the information we needed to open the
4088	 * pool. If we are importing the pool in read-write mode, a few
4089	 * additional steps must be performed to finish the import.
4090	 */
4091	if (spa_writeable(spa) && (spa->spa_load_state == SPA_LOAD_RECOVER ||
4092	    spa->spa_load_max_txg == UINT64_MAX)) {
4093		uint64_t config_cache_txg = spa->spa_config_txg;
4094
4095		ASSERT(spa->spa_load_state != SPA_LOAD_TRYIMPORT);
4096
4097		/*
4098		 * In case of a checkpoint rewind, log the original txg
4099		 * of the checkpointed uberblock.
4100		 */
4101		if (checkpoint_rewind) {
4102			spa_history_log_internal(spa, "checkpoint rewind",
4103			    NULL, "rewound state to txg=%llu",
4104			    (u_longlong_t)spa->spa_uberblock.ub_checkpoint_txg);
4105		}
4106
4107		/*
4108		 * Traverse the ZIL and claim all blocks.
4109		 */
4110		spa_ld_claim_log_blocks(spa);
4111
4112		/*
4113		 * Kick-off the syncing thread.
4114		 */
4115		spa->spa_sync_on = B_TRUE;
4116		txg_sync_start(spa->spa_dsl_pool);
4117		mmp_thread_start(spa);
4118
4119		/*
4120		 * Wait for all claims to sync.  We sync up to the highest
4121		 * claimed log block birth time so that claimed log blocks
4122		 * don't appear to be from the future.  spa_claim_max_txg
4123		 * will have been set for us by ZIL traversal operations
4124		 * performed above.
4125		 */
4126		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
4127
4128		/*
4129		 * Check if we need to request an update of the config. On the
4130		 * next sync, we would update the config stored in vdev labels
4131		 * and the cachefile (by default /etc/zfs/zpool.cache).
4132		 */
4133		spa_ld_check_for_config_update(spa, config_cache_txg,
4134		    update_config_cache);
4135
4136		/*
4137		 * Check all DTLs to see if anything needs resilvering.
4138		 */
4139		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
4140		    vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL))
4141			spa_async_request(spa, SPA_ASYNC_RESILVER);
4142
4143		/*
4144		 * Log the fact that we booted up (so that we can detect if
4145		 * we rebooted in the middle of an operation).
4146		 */
4147		spa_history_log_version(spa, "open");
4148
4149		spa_restart_removal(spa);
4150		spa_spawn_aux_threads(spa);
4151
4152		/*
4153		 * Delete any inconsistent datasets.
4154		 *
4155		 * Note:
4156		 * Since we may be issuing deletes for clones here,
4157		 * we make sure to do so after we've spawned all the
4158		 * auxiliary threads above (from which the livelist
4159		 * deletion zthr is part of).
4160		 */
4161		(void) dmu_objset_find(spa_name(spa),
4162		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
4163
4164		/*
4165		 * Clean up any stale temporary dataset userrefs.
4166		 */
4167		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
4168
4169		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4170		vdev_initialize_restart(spa->spa_root_vdev);
4171		spa_config_exit(spa, SCL_CONFIG, FTAG);
4172	}
4173
4174	spa_load_note(spa, "LOADED");
4175
4176	return (0);
4177}
4178
4179static int
4180spa_load_retry(spa_t *spa, spa_load_state_t state)
4181{
4182	int mode = spa->spa_mode;
4183
4184	spa_unload(spa);
4185	spa_deactivate(spa);
4186
4187	spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
4188
4189	spa_activate(spa, mode);
4190	spa_async_suspend(spa);
4191
4192	spa_load_note(spa, "spa_load_retry: rewind, max txg: %llu",
4193	    (u_longlong_t)spa->spa_load_max_txg);
4194
4195	return (spa_load(spa, state, SPA_IMPORT_EXISTING));
4196}
4197
4198/*
4199 * If spa_load() fails this function will try loading prior txg's. If
4200 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
4201 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
4202 * function will not rewind the pool and will return the same error as
4203 * spa_load().
4204 */
4205static int
4206spa_load_best(spa_t *spa, spa_load_state_t state, uint64_t max_request,
4207    int rewind_flags)
4208{
4209	nvlist_t *loadinfo = NULL;
4210	nvlist_t *config = NULL;
4211	int load_error, rewind_error;
4212	uint64_t safe_rewind_txg;
4213	uint64_t min_txg;
4214
4215	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
4216		spa->spa_load_max_txg = spa->spa_load_txg;
4217		spa_set_log_state(spa, SPA_LOG_CLEAR);
4218	} else {
4219		spa->spa_load_max_txg = max_request;
4220		if (max_request != UINT64_MAX)
4221			spa->spa_extreme_rewind = B_TRUE;
4222	}
4223
4224	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING);
4225	if (load_error == 0)
4226		return (0);
4227	if (load_error == ZFS_ERR_NO_CHECKPOINT) {
4228		/*
4229		 * When attempting checkpoint-rewind on a pool with no
4230		 * checkpoint, we should not attempt to load uberblocks
4231		 * from previous txgs when spa_load fails.
4232		 */
4233		ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
4234		return (load_error);
4235	}
4236
4237	if (spa->spa_root_vdev != NULL)
4238		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4239
4240	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
4241	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
4242
4243	if (rewind_flags & ZPOOL_NEVER_REWIND) {
4244		nvlist_free(config);
4245		return (load_error);
4246	}
4247
4248	if (state == SPA_LOAD_RECOVER) {
4249		/* Price of rolling back is discarding txgs, including log */
4250		spa_set_log_state(spa, SPA_LOG_CLEAR);
4251	} else {
4252		/*
4253		 * If we aren't rolling back save the load info from our first
4254		 * import attempt so that we can restore it after attempting
4255		 * to rewind.
4256		 */
4257		loadinfo = spa->spa_load_info;
4258		spa->spa_load_info = fnvlist_alloc();
4259	}
4260
4261	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
4262	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
4263	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
4264	    TXG_INITIAL : safe_rewind_txg;
4265
4266	/*
4267	 * Continue as long as we're finding errors, we're still within
4268	 * the acceptable rewind range, and we're still finding uberblocks
4269	 */
4270	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
4271	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
4272		if (spa->spa_load_max_txg < safe_rewind_txg)
4273			spa->spa_extreme_rewind = B_TRUE;
4274		rewind_error = spa_load_retry(spa, state);
4275	}
4276
4277	spa->spa_extreme_rewind = B_FALSE;
4278	spa->spa_load_max_txg = UINT64_MAX;
4279
4280	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
4281		spa_config_set(spa, config);
4282	else
4283		nvlist_free(config);
4284
4285	if (state == SPA_LOAD_RECOVER) {
4286		ASSERT3P(loadinfo, ==, NULL);
4287		return (rewind_error);
4288	} else {
4289		/* Store the rewind info as part of the initial load info */
4290		fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
4291		    spa->spa_load_info);
4292
4293		/* Restore the initial load info */
4294		fnvlist_free(spa->spa_load_info);
4295		spa->spa_load_info = loadinfo;
4296
4297		return (load_error);
4298	}
4299}
4300
4301/*
4302 * Pool Open/Import
4303 *
4304 * The import case is identical to an open except that the configuration is sent
4305 * down from userland, instead of grabbed from the configuration cache.  For the
4306 * case of an open, the pool configuration will exist in the
4307 * POOL_STATE_UNINITIALIZED state.
4308 *
4309 * The stats information (gen/count/ustats) is used to gather vdev statistics at
4310 * the same time open the pool, without having to keep around the spa_t in some
4311 * ambiguous state.
4312 */
4313static int
4314spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
4315    nvlist_t **config)
4316{
4317	spa_t *spa;
4318	spa_load_state_t state = SPA_LOAD_OPEN;
4319	int error;
4320	int locked = B_FALSE;
4321
4322	*spapp = NULL;
4323
4324	/*
4325	 * As disgusting as this is, we need to support recursive calls to this
4326	 * function because dsl_dir_open() is called during spa_load(), and ends
4327	 * up calling spa_open() again.  The real fix is to figure out how to
4328	 * avoid dsl_dir_open() calling this in the first place.
4329	 */
4330	if (mutex_owner(&spa_namespace_lock) != curthread) {
4331		mutex_enter(&spa_namespace_lock);
4332		locked = B_TRUE;
4333	}
4334
4335	if ((spa = spa_lookup(pool)) == NULL) {
4336		if (locked)
4337			mutex_exit(&spa_namespace_lock);
4338		return (SET_ERROR(ENOENT));
4339	}
4340
4341	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
4342		zpool_load_policy_t policy;
4343
4344		zpool_get_load_policy(nvpolicy ? nvpolicy : spa->spa_config,
4345		    &policy);
4346		if (policy.zlp_rewind & ZPOOL_DO_REWIND)
4347			state = SPA_LOAD_RECOVER;
4348
4349		spa_activate(spa, spa_mode_global);
4350
4351		if (state != SPA_LOAD_RECOVER)
4352			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4353		spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
4354
4355		zfs_dbgmsg("spa_open_common: opening %s", pool);
4356		error = spa_load_best(spa, state, policy.zlp_txg,
4357		    policy.zlp_rewind);
4358
4359		if (error == EBADF) {
4360			/*
4361			 * If vdev_validate() returns failure (indicated by
4362			 * EBADF), it indicates that one of the vdevs indicates
4363			 * that the pool has been exported or destroyed.  If
4364			 * this is the case, the config cache is out of sync and
4365			 * we should remove the pool from the namespace.
4366			 */
4367			spa_unload(spa);
4368			spa_deactivate(spa);
4369			spa_write_cachefile(spa, B_TRUE, B_TRUE);
4370			spa_remove(spa);
4371			if (locked)
4372				mutex_exit(&spa_namespace_lock);
4373			return (SET_ERROR(ENOENT));
4374		}
4375
4376		if (error) {
4377			/*
4378			 * We can't open the pool, but we still have useful
4379			 * information: the state of each vdev after the
4380			 * attempted vdev_open().  Return this to the user.
4381			 */
4382			if (config != NULL && spa->spa_config) {
4383				VERIFY(nvlist_dup(spa->spa_config, config,
4384				    KM_SLEEP) == 0);
4385				VERIFY(nvlist_add_nvlist(*config,
4386				    ZPOOL_CONFIG_LOAD_INFO,
4387				    spa->spa_load_info) == 0);
4388			}
4389			spa_unload(spa);
4390			spa_deactivate(spa);
4391			spa->spa_last_open_failed = error;
4392			if (locked)
4393				mutex_exit(&spa_namespace_lock);
4394			*spapp = NULL;
4395			return (error);
4396		}
4397	}
4398
4399	spa_open_ref(spa, tag);
4400
4401	if (config != NULL)
4402		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4403
4404	/*
4405	 * If we've recovered the pool, pass back any information we
4406	 * gathered while doing the load.
4407	 */
4408	if (state == SPA_LOAD_RECOVER) {
4409		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
4410		    spa->spa_load_info) == 0);
4411	}
4412
4413	if (locked) {
4414		spa->spa_last_open_failed = 0;
4415		spa->spa_last_ubsync_txg = 0;
4416		spa->spa_load_txg = 0;
4417		mutex_exit(&spa_namespace_lock);
4418	}
4419
4420	*spapp = spa;
4421
4422	return (0);
4423}
4424
4425int
4426spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
4427    nvlist_t **config)
4428{
4429	return (spa_open_common(name, spapp, tag, policy, config));
4430}
4431
4432int
4433spa_open(const char *name, spa_t **spapp, void *tag)
4434{
4435	return (spa_open_common(name, spapp, tag, NULL, NULL));
4436}
4437
4438/*
4439 * Lookup the given spa_t, incrementing the inject count in the process,
4440 * preventing it from being exported or destroyed.
4441 */
4442spa_t *
4443spa_inject_addref(char *name)
4444{
4445	spa_t *spa;
4446
4447	mutex_enter(&spa_namespace_lock);
4448	if ((spa = spa_lookup(name)) == NULL) {
4449		mutex_exit(&spa_namespace_lock);
4450		return (NULL);
4451	}
4452	spa->spa_inject_ref++;
4453	mutex_exit(&spa_namespace_lock);
4454
4455	return (spa);
4456}
4457
4458void
4459spa_inject_delref(spa_t *spa)
4460{
4461	mutex_enter(&spa_namespace_lock);
4462	spa->spa_inject_ref--;
4463	mutex_exit(&spa_namespace_lock);
4464}
4465
4466/*
4467 * Add spares device information to the nvlist.
4468 */
4469static void
4470spa_add_spares(spa_t *spa, nvlist_t *config)
4471{
4472	nvlist_t **spares;
4473	uint_t i, nspares;
4474	nvlist_t *nvroot;
4475	uint64_t guid;
4476	vdev_stat_t *vs;
4477	uint_t vsc;
4478	uint64_t pool;
4479
4480	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4481
4482	if (spa->spa_spares.sav_count == 0)
4483		return;
4484
4485	VERIFY(nvlist_lookup_nvlist(config,
4486	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
4487	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4488	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
4489	if (nspares != 0) {
4490		VERIFY(nvlist_add_nvlist_array(nvroot,
4491		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4492		VERIFY(nvlist_lookup_nvlist_array(nvroot,
4493		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
4494
4495		/*
4496		 * Go through and find any spares which have since been
4497		 * repurposed as an active spare.  If this is the case, update
4498		 * their status appropriately.
4499		 */
4500		for (i = 0; i < nspares; i++) {
4501			VERIFY(nvlist_lookup_uint64(spares[i],
4502			    ZPOOL_CONFIG_GUID, &guid) == 0);
4503			if (spa_spare_exists(guid, &pool, NULL) &&
4504			    pool != 0ULL) {
4505				VERIFY(nvlist_lookup_uint64_array(
4506				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
4507				    (uint64_t **)&vs, &vsc) == 0);
4508				vs->vs_state = VDEV_STATE_CANT_OPEN;
4509				vs->vs_aux = VDEV_AUX_SPARED;
4510			}
4511		}
4512	}
4513}
4514
4515/*
4516 * Add l2cache device information to the nvlist, including vdev stats.
4517 */
4518static void
4519spa_add_l2cache(spa_t *spa, nvlist_t *config)
4520{
4521	nvlist_t **l2cache;
4522	uint_t i, j, nl2cache;
4523	nvlist_t *nvroot;
4524	uint64_t guid;
4525	vdev_t *vd;
4526	vdev_stat_t *vs;
4527	uint_t vsc;
4528
4529	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4530
4531	if (spa->spa_l2cache.sav_count == 0)
4532		return;
4533
4534	VERIFY(nvlist_lookup_nvlist(config,
4535	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
4536	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4537	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
4538	if (nl2cache != 0) {
4539		VERIFY(nvlist_add_nvlist_array(nvroot,
4540		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4541		VERIFY(nvlist_lookup_nvlist_array(nvroot,
4542		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
4543
4544		/*
4545		 * Update level 2 cache device stats.
4546		 */
4547
4548		for (i = 0; i < nl2cache; i++) {
4549			VERIFY(nvlist_lookup_uint64(l2cache[i],
4550			    ZPOOL_CONFIG_GUID, &guid) == 0);
4551
4552			vd = NULL;
4553			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
4554				if (guid ==
4555				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
4556					vd = spa->spa_l2cache.sav_vdevs[j];
4557					break;
4558				}
4559			}
4560			ASSERT(vd != NULL);
4561
4562			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
4563			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
4564			    == 0);
4565			vdev_get_stats(vd, vs);
4566		}
4567	}
4568}
4569
4570static void
4571spa_add_feature_stats(spa_t *spa, nvlist_t *config)
4572{
4573	nvlist_t *features;
4574	zap_cursor_t zc;
4575	zap_attribute_t za;
4576
4577	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4578	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4579
4580	if (spa->spa_feat_for_read_obj != 0) {
4581		for (zap_cursor_init(&zc, spa->spa_meta_objset,
4582		    spa->spa_feat_for_read_obj);
4583		    zap_cursor_retrieve(&zc, &za) == 0;
4584		    zap_cursor_advance(&zc)) {
4585			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
4586			    za.za_num_integers == 1);
4587			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
4588			    za.za_first_integer));
4589		}
4590		zap_cursor_fini(&zc);
4591	}
4592
4593	if (spa->spa_feat_for_write_obj != 0) {
4594		for (zap_cursor_init(&zc, spa->spa_meta_objset,
4595		    spa->spa_feat_for_write_obj);
4596		    zap_cursor_retrieve(&zc, &za) == 0;
4597		    zap_cursor_advance(&zc)) {
4598			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
4599			    za.za_num_integers == 1);
4600			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
4601			    za.za_first_integer));
4602		}
4603		zap_cursor_fini(&zc);
4604	}
4605
4606	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
4607	    features) == 0);
4608	nvlist_free(features);
4609}
4610
4611int
4612spa_get_stats(const char *name, nvlist_t **config,
4613    char *altroot, size_t buflen)
4614{
4615	int error;
4616	spa_t *spa;
4617
4618	*config = NULL;
4619	error = spa_open_common(name, &spa, FTAG, NULL, config);
4620
4621	if (spa != NULL) {
4622		/*
4623		 * This still leaves a window of inconsistency where the spares
4624		 * or l2cache devices could change and the config would be
4625		 * self-inconsistent.
4626		 */
4627		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4628
4629		if (*config != NULL) {
4630			uint64_t loadtimes[2];
4631
4632			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
4633			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
4634			VERIFY(nvlist_add_uint64_array(*config,
4635			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
4636
4637			VERIFY(nvlist_add_uint64(*config,
4638			    ZPOOL_CONFIG_ERRCOUNT,
4639			    spa_get_errlog_size(spa)) == 0);
4640
4641			if (spa_suspended(spa)) {
4642				VERIFY(nvlist_add_uint64(*config,
4643				    ZPOOL_CONFIG_SUSPENDED,
4644				    spa->spa_failmode) == 0);
4645				VERIFY(nvlist_add_uint64(*config,
4646				    ZPOOL_CONFIG_SUSPENDED_REASON,
4647				    spa->spa_suspended) == 0);
4648			}
4649
4650			spa_add_spares(spa, *config);
4651			spa_add_l2cache(spa, *config);
4652			spa_add_feature_stats(spa, *config);
4653		}
4654	}
4655
4656	/*
4657	 * We want to get the alternate root even for faulted pools, so we cheat
4658	 * and call spa_lookup() directly.
4659	 */
4660	if (altroot) {
4661		if (spa == NULL) {
4662			mutex_enter(&spa_namespace_lock);
4663			spa = spa_lookup(name);
4664			if (spa)
4665				spa_altroot(spa, altroot, buflen);
4666			else
4667				altroot[0] = '\0';
4668			spa = NULL;
4669			mutex_exit(&spa_namespace_lock);
4670		} else {
4671			spa_altroot(spa, altroot, buflen);
4672		}
4673	}
4674
4675	if (spa != NULL) {
4676		spa_config_exit(spa, SCL_CONFIG, FTAG);
4677		spa_close(spa, FTAG);
4678	}
4679
4680	return (error);
4681}
4682
4683/*
4684 * Validate that the auxiliary device array is well formed.  We must have an
4685 * array of nvlists, each which describes a valid leaf vdev.  If this is an
4686 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
4687 * specified, as long as they are well-formed.
4688 */
4689static int
4690spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
4691    spa_aux_vdev_t *sav, const char *config, uint64_t version,
4692    vdev_labeltype_t label)
4693{
4694	nvlist_t **dev;
4695	uint_t i, ndev;
4696	vdev_t *vd;
4697	int error;
4698
4699	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4700
4701	/*
4702	 * It's acceptable to have no devs specified.
4703	 */
4704	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
4705		return (0);
4706
4707	if (ndev == 0)
4708		return (SET_ERROR(EINVAL));
4709
4710	/*
4711	 * Make sure the pool is formatted with a version that supports this
4712	 * device type.
4713	 */
4714	if (spa_version(spa) < version)
4715		return (SET_ERROR(ENOTSUP));
4716
4717	/*
4718	 * Set the pending device list so we correctly handle device in-use
4719	 * checking.
4720	 */
4721	sav->sav_pending = dev;
4722	sav->sav_npending = ndev;
4723
4724	for (i = 0; i < ndev; i++) {
4725		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
4726		    mode)) != 0)
4727			goto out;
4728
4729		if (!vd->vdev_ops->vdev_op_leaf) {
4730			vdev_free(vd);
4731			error = SET_ERROR(EINVAL);
4732			goto out;
4733		}
4734
4735		vd->vdev_top = vd;
4736
4737		if ((error = vdev_open(vd)) == 0 &&
4738		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
4739			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
4740			    vd->vdev_guid) == 0);
4741		}
4742
4743		vdev_free(vd);
4744
4745		if (error &&
4746		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
4747			goto out;
4748		else
4749			error = 0;
4750	}
4751
4752out:
4753	sav->sav_pending = NULL;
4754	sav->sav_npending = 0;
4755	return (error);
4756}
4757
4758static int
4759spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
4760{
4761	int error;
4762
4763	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4764
4765	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
4766	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
4767	    VDEV_LABEL_SPARE)) != 0) {
4768		return (error);
4769	}
4770
4771	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
4772	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
4773	    VDEV_LABEL_L2CACHE));
4774}
4775
4776static void
4777spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
4778    const char *config)
4779{
4780	int i;
4781
4782	if (sav->sav_config != NULL) {
4783		nvlist_t **olddevs;
4784		uint_t oldndevs;
4785		nvlist_t **newdevs;
4786
4787		/*
4788		 * Generate new dev list by concatentating with the
4789		 * current dev list.
4790		 */
4791		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
4792		    &olddevs, &oldndevs) == 0);
4793
4794		newdevs = kmem_alloc(sizeof (void *) *
4795		    (ndevs + oldndevs), KM_SLEEP);
4796		for (i = 0; i < oldndevs; i++)
4797			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
4798			    KM_SLEEP) == 0);
4799		for (i = 0; i < ndevs; i++)
4800			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
4801			    KM_SLEEP) == 0);
4802
4803		VERIFY(nvlist_remove(sav->sav_config, config,
4804		    DATA_TYPE_NVLIST_ARRAY) == 0);
4805
4806		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
4807		    config, newdevs, ndevs + oldndevs) == 0);
4808		for (i = 0; i < oldndevs + ndevs; i++)
4809			nvlist_free(newdevs[i]);
4810		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
4811	} else {
4812		/*
4813		 * Generate a new dev list.
4814		 */
4815		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
4816		    KM_SLEEP) == 0);
4817		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
4818		    devs, ndevs) == 0);
4819	}
4820}
4821
4822/*
4823 * Stop and drop level 2 ARC devices
4824 */
4825void
4826spa_l2cache_drop(spa_t *spa)
4827{
4828	vdev_t *vd;
4829	int i;
4830	spa_aux_vdev_t *sav = &spa->spa_l2cache;
4831
4832	for (i = 0; i < sav->sav_count; i++) {
4833		uint64_t pool;
4834
4835		vd = sav->sav_vdevs[i];
4836		ASSERT(vd != NULL);
4837
4838		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
4839		    pool != 0ULL && l2arc_vdev_present(vd))
4840			l2arc_remove_vdev(vd);
4841	}
4842}
4843
4844/*
4845 * Verify encryption parameters for spa creation. If we are encrypting, we must
4846 * have the encryption feature flag enabled.
4847 */
4848static int
4849spa_create_check_encryption_params(dsl_crypto_params_t *dcp,
4850    boolean_t has_encryption)
4851{
4852	if (dcp->cp_crypt != ZIO_CRYPT_OFF &&
4853	    dcp->cp_crypt != ZIO_CRYPT_INHERIT &&
4854	    !has_encryption)
4855		return (SET_ERROR(ENOTSUP));
4856
4857	return (dmu_objset_create_crypt_check(NULL, dcp, NULL));
4858}
4859
4860/*
4861 * Pool Creation
4862 */
4863int
4864spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
4865    nvlist_t *zplprops, dsl_crypto_params_t *dcp)
4866{
4867	spa_t *spa;
4868	char *altroot = NULL;
4869	vdev_t *rvd;
4870	dsl_pool_t *dp;
4871	dmu_tx_t *tx;
4872	int error = 0;
4873	uint64_t txg = TXG_INITIAL;
4874	nvlist_t **spares, **l2cache;
4875	uint_t nspares, nl2cache;
4876	uint64_t version, obj;
4877	boolean_t has_features;
4878	char *poolname;
4879	nvlist_t *nvl;
4880	boolean_t has_encryption;
4881	spa_feature_t feat;
4882	char *feat_name;
4883
4884	if (props == NULL ||
4885	    nvlist_lookup_string(props,
4886	    zpool_prop_to_name(ZPOOL_PROP_TNAME), &poolname) != 0)
4887		poolname = (char *)pool;
4888
4889	/*
4890	 * If this pool already exists, return failure.
4891	 */
4892	mutex_enter(&spa_namespace_lock);
4893	if (spa_lookup(poolname) != NULL) {
4894		mutex_exit(&spa_namespace_lock);
4895		return (SET_ERROR(EEXIST));
4896	}
4897
4898	/*
4899	 * Allocate a new spa_t structure.
4900	 */
4901	nvl = fnvlist_alloc();
4902	fnvlist_add_string(nvl, ZPOOL_CONFIG_POOL_NAME, pool);
4903	(void) nvlist_lookup_string(props,
4904	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4905	spa = spa_add(poolname, nvl, altroot);
4906	fnvlist_free(nvl);
4907	spa_activate(spa, spa_mode_global);
4908
4909	if (props && (error = spa_prop_validate(spa, props))) {
4910		spa_deactivate(spa);
4911		spa_remove(spa);
4912		mutex_exit(&spa_namespace_lock);
4913		return (error);
4914	}
4915
4916	/*
4917	 * Temporary pool names should never be written to disk.
4918	 */
4919	if (poolname != pool)
4920		spa->spa_import_flags |= ZFS_IMPORT_TEMP_NAME;
4921
4922	has_features = B_FALSE;
4923	has_encryption = B_FALSE;
4924	for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
4925	    elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
4926		if (zpool_prop_feature(nvpair_name(elem))) {
4927			has_features = B_TRUE;
4928			feat_name = strchr(nvpair_name(elem), '@') + 1;
4929			VERIFY0(zfeature_lookup_name(feat_name, &feat));
4930			if (feat == SPA_FEATURE_ENCRYPTION)
4931				has_encryption = B_TRUE;
4932		}
4933	}
4934
4935	/* verify encryption params, if they were provided */
4936	if (dcp != NULL) {
4937		error = spa_create_check_encryption_params(dcp, has_encryption);
4938		if (error != 0) {
4939			spa_deactivate(spa);
4940			spa_remove(spa);
4941			mutex_exit(&spa_namespace_lock);
4942			return (error);
4943		}
4944	}
4945
4946	if (has_features || nvlist_lookup_uint64(props,
4947	    zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
4948		version = SPA_VERSION;
4949	}
4950	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
4951
4952	spa->spa_first_txg = txg;
4953	spa->spa_uberblock.ub_txg = txg - 1;
4954	spa->spa_uberblock.ub_version = version;
4955	spa->spa_ubsync = spa->spa_uberblock;
4956	spa->spa_load_state = SPA_LOAD_CREATE;
4957	spa->spa_removing_phys.sr_state = DSS_NONE;
4958	spa->spa_removing_phys.sr_removing_vdev = -1;
4959	spa->spa_removing_phys.sr_prev_indirect_vdev = -1;
4960	spa->spa_indirect_vdevs_loaded = B_TRUE;
4961
4962	/*
4963	 * Create "The Godfather" zio to hold all async IOs
4964	 */
4965	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
4966	    KM_SLEEP);
4967	for (int i = 0; i < max_ncpus; i++) {
4968		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
4969		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
4970		    ZIO_FLAG_GODFATHER);
4971	}
4972
4973	/*
4974	 * Create the root vdev.
4975	 */
4976	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4977
4978	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
4979
4980	ASSERT(error != 0 || rvd != NULL);
4981	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
4982
4983	if (error == 0 && !zfs_allocatable_devs(nvroot))
4984		error = SET_ERROR(EINVAL);
4985
4986	if (error == 0 &&
4987	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
4988	    (error = spa_validate_aux(spa, nvroot, txg,
4989	    VDEV_ALLOC_ADD)) == 0) {
4990		/*
4991		 * instantiate the metaslab groups (this will dirty the vdevs)
4992		 * we can no longer error exit past this point
4993		 */
4994		for (int c = 0; error == 0 && c < rvd->vdev_children; c++) {
4995			vdev_t *vd = rvd->vdev_child[c];
4996
4997			vdev_metaslab_set_size(vd);
4998			vdev_expand(vd, txg);
4999		}
5000	}
5001
5002	spa_config_exit(spa, SCL_ALL, FTAG);
5003
5004	if (error != 0) {
5005		spa_unload(spa);
5006		spa_deactivate(spa);
5007		spa_remove(spa);
5008		mutex_exit(&spa_namespace_lock);
5009		return (error);
5010	}
5011
5012	/*
5013	 * Get the list of spares, if specified.
5014	 */
5015	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
5016	    &spares, &nspares) == 0) {
5017		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
5018		    KM_SLEEP) == 0);
5019		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
5020		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
5021		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5022		spa_load_spares(spa);
5023		spa_config_exit(spa, SCL_ALL, FTAG);
5024		spa->spa_spares.sav_sync = B_TRUE;
5025	}
5026
5027	/*
5028	 * Get the list of level 2 cache devices, if specified.
5029	 */
5030	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
5031	    &l2cache, &nl2cache) == 0) {
5032		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
5033		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
5034		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
5035		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
5036		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5037		spa_load_l2cache(spa);
5038		spa_config_exit(spa, SCL_ALL, FTAG);
5039		spa->spa_l2cache.sav_sync = B_TRUE;
5040	}
5041
5042	spa->spa_is_initializing = B_TRUE;
5043	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, dcp, txg);
5044	spa->spa_is_initializing = B_FALSE;
5045
5046	/*
5047	 * Create DDTs (dedup tables).
5048	 */
5049	ddt_create(spa);
5050
5051	spa_update_dspace(spa);
5052
5053	tx = dmu_tx_create_assigned(dp, txg);
5054
5055	/*
5056	 * Create the pool config object.
5057	 */
5058	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
5059	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
5060	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
5061
5062	if (zap_add(spa->spa_meta_objset,
5063	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
5064	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
5065		cmn_err(CE_PANIC, "failed to add pool config");
5066	}
5067
5068	if (zap_add(spa->spa_meta_objset,
5069	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
5070	    sizeof (uint64_t), 1, &version, tx) != 0) {
5071		cmn_err(CE_PANIC, "failed to add pool version");
5072	}
5073
5074	/* Newly created pools with the right version are always deflated. */
5075	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
5076		spa->spa_deflate = TRUE;
5077		if (zap_add(spa->spa_meta_objset,
5078		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5079		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
5080			cmn_err(CE_PANIC, "failed to add deflate");
5081		}
5082	}
5083
5084	/*
5085	 * Create the deferred-free bpobj.  Turn off compression
5086	 * because sync-to-convergence takes longer if the blocksize
5087	 * keeps changing.
5088	 */
5089	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
5090	dmu_object_set_compress(spa->spa_meta_objset, obj,
5091	    ZIO_COMPRESS_OFF, tx);
5092	if (zap_add(spa->spa_meta_objset,
5093	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
5094	    sizeof (uint64_t), 1, &obj, tx) != 0) {
5095		cmn_err(CE_PANIC, "failed to add bpobj");
5096	}
5097	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
5098	    spa->spa_meta_objset, obj));
5099
5100	/*
5101	 * Create the pool's history object.
5102	 */
5103	if (version >= SPA_VERSION_ZPOOL_HISTORY)
5104		spa_history_create_obj(spa, tx);
5105
5106	/*
5107	 * Generate some random noise for salted checksums to operate on.
5108	 */
5109	(void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
5110	    sizeof (spa->spa_cksum_salt.zcs_bytes));
5111
5112	/*
5113	 * Set pool properties.
5114	 */
5115	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
5116	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
5117	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
5118	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
5119	spa->spa_multihost = zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST);
5120
5121	if (props != NULL) {
5122		spa_configfile_set(spa, props, B_FALSE);
5123		spa_sync_props(props, tx);
5124	}
5125
5126	dmu_tx_commit(tx);
5127
5128	spa->spa_sync_on = B_TRUE;
5129	txg_sync_start(spa->spa_dsl_pool);
5130	mmp_thread_start(spa);
5131
5132	/*
5133	 * We explicitly wait for the first transaction to complete so that our
5134	 * bean counters are appropriately updated.
5135	 */
5136	txg_wait_synced(spa->spa_dsl_pool, txg);
5137
5138	spa_spawn_aux_threads(spa);
5139
5140	spa_write_cachefile(spa, B_FALSE, B_TRUE);
5141	spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE);
5142
5143	spa_history_log_version(spa, "create");
5144
5145	/*
5146	 * Don't count references from objsets that are already closed
5147	 * and are making their way through the eviction process.
5148	 */
5149	spa_evicting_os_wait(spa);
5150	spa->spa_minref = zfs_refcount_count(&spa->spa_refcount);
5151	spa->spa_load_state = SPA_LOAD_NONE;
5152
5153	mutex_exit(&spa_namespace_lock);
5154
5155	return (0);
5156}
5157
5158#ifdef _KERNEL
5159/*
5160 * Get the root pool information from the root disk, then import the root pool
5161 * during the system boot up time.
5162 */
5163extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
5164
5165static nvlist_t *
5166spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
5167{
5168	nvlist_t *config;
5169	nvlist_t *nvtop, *nvroot;
5170	uint64_t pgid;
5171
5172	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
5173		return (NULL);
5174
5175	/*
5176	 * Add this top-level vdev to the child array.
5177	 */
5178	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
5179	    &nvtop) == 0);
5180	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5181	    &pgid) == 0);
5182	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
5183
5184	/*
5185	 * Put this pool's top-level vdevs into a root vdev.
5186	 */
5187	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5188	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
5189	    VDEV_TYPE_ROOT) == 0);
5190	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
5191	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
5192	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
5193	    &nvtop, 1) == 0);
5194
5195	/*
5196	 * Replace the existing vdev_tree with the new root vdev in
5197	 * this pool's configuration (remove the old, add the new).
5198	 */
5199	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
5200	nvlist_free(nvroot);
5201	return (config);
5202}
5203
5204/*
5205 * Walk the vdev tree and see if we can find a device with "better"
5206 * configuration. A configuration is "better" if the label on that
5207 * device has a more recent txg.
5208 */
5209static void
5210spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
5211{
5212	for (int c = 0; c < vd->vdev_children; c++)
5213		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
5214
5215	if (vd->vdev_ops->vdev_op_leaf) {
5216		nvlist_t *label;
5217		uint64_t label_txg;
5218
5219		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
5220		    &label) != 0)
5221			return;
5222
5223		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
5224		    &label_txg) == 0);
5225
5226		/*
5227		 * Do we have a better boot device?
5228		 */
5229		if (label_txg > *txg) {
5230			*txg = label_txg;
5231			*avd = vd;
5232		}
5233		nvlist_free(label);
5234	}
5235}
5236
5237/*
5238 * Import a root pool.
5239 *
5240 * For x86. devpath_list will consist of devid and/or physpath name of
5241 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
5242 * The GRUB "findroot" command will return the vdev we should boot.
5243 *
5244 * For Sparc, devpath_list consists the physpath name of the booting device
5245 * no matter the rootpool is a single device pool or a mirrored pool.
5246 * e.g.
5247 *	"/pci@1f,0/ide@d/disk@0,0:a"
5248 */
5249int
5250spa_import_rootpool(char *devpath, char *devid)
5251{
5252	spa_t *spa;
5253	vdev_t *rvd, *bvd, *avd = NULL;
5254	nvlist_t *config, *nvtop;
5255	uint64_t guid, txg;
5256	char *pname;
5257	int error;
5258
5259	/*
5260	 * Read the label from the boot device and generate a configuration.
5261	 */
5262	config = spa_generate_rootconf(devpath, devid, &guid);
5263#if defined(_OBP) && defined(_KERNEL)
5264	if (config == NULL) {
5265		if (strstr(devpath, "/iscsi/ssd") != NULL) {
5266			/* iscsi boot */
5267			get_iscsi_bootpath_phy(devpath);
5268			config = spa_generate_rootconf(devpath, devid, &guid);
5269		}
5270	}
5271#endif
5272	if (config == NULL) {
5273		cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
5274		    devpath);
5275		return (SET_ERROR(EIO));
5276	}
5277
5278	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
5279	    &pname) == 0);
5280	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
5281
5282	mutex_enter(&spa_namespace_lock);
5283	if ((spa = spa_lookup(pname)) != NULL) {
5284		/*
5285		 * Remove the existing root pool from the namespace so that we
5286		 * can replace it with the correct config we just read in.
5287		 */
5288		spa_remove(spa);
5289	}
5290
5291	spa = spa_add(pname, config, NULL);
5292	spa->spa_is_root = B_TRUE;
5293	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
5294	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
5295	    &spa->spa_ubsync.ub_version) != 0)
5296		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
5297
5298	/*
5299	 * Build up a vdev tree based on the boot device's label config.
5300	 */
5301	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
5302	    &nvtop) == 0);
5303	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5304	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
5305	    VDEV_ALLOC_ROOTPOOL);
5306	spa_config_exit(spa, SCL_ALL, FTAG);
5307	if (error) {
5308		mutex_exit(&spa_namespace_lock);
5309		nvlist_free(config);
5310		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
5311		    pname);
5312		return (error);
5313	}
5314
5315	/*
5316	 * Get the boot vdev.
5317	 */
5318	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
5319		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
5320		    (u_longlong_t)guid);
5321		error = SET_ERROR(ENOENT);
5322		goto out;
5323	}
5324
5325	/*
5326	 * Determine if there is a better boot device.
5327	 */
5328	avd = bvd;
5329	spa_alt_rootvdev(rvd, &avd, &txg);
5330	if (avd != bvd) {
5331		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
5332		    "try booting from '%s'", avd->vdev_path);
5333		error = SET_ERROR(EINVAL);
5334		goto out;
5335	}
5336
5337	/*
5338	 * If the boot device is part of a spare vdev then ensure that
5339	 * we're booting off the active spare.
5340	 */
5341	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
5342	    !bvd->vdev_isspare) {
5343		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
5344		    "try booting from '%s'",
5345		    bvd->vdev_parent->
5346		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
5347		error = SET_ERROR(EINVAL);
5348		goto out;
5349	}
5350
5351	error = 0;
5352out:
5353	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5354	vdev_free(rvd);
5355	spa_config_exit(spa, SCL_ALL, FTAG);
5356	mutex_exit(&spa_namespace_lock);
5357
5358	nvlist_free(config);
5359	return (error);
5360}
5361
5362#endif
5363
5364/*
5365 * Import a non-root pool into the system.
5366 */
5367int
5368spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
5369{
5370	spa_t *spa;
5371	char *altroot = NULL;
5372	spa_load_state_t state = SPA_LOAD_IMPORT;
5373	zpool_load_policy_t policy;
5374	uint64_t mode = spa_mode_global;
5375	uint64_t readonly = B_FALSE;
5376	int error;
5377	nvlist_t *nvroot;
5378	nvlist_t **spares, **l2cache;
5379	uint_t nspares, nl2cache;
5380
5381	/*
5382	 * If a pool with this name exists, return failure.
5383	 */
5384	mutex_enter(&spa_namespace_lock);
5385	if (spa_lookup(pool) != NULL) {
5386		mutex_exit(&spa_namespace_lock);
5387		return (SET_ERROR(EEXIST));
5388	}
5389
5390	/*
5391	 * Create and initialize the spa structure.
5392	 */
5393	(void) nvlist_lookup_string(props,
5394	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5395	(void) nvlist_lookup_uint64(props,
5396	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
5397	if (readonly)
5398		mode = FREAD;
5399	spa = spa_add(pool, config, altroot);
5400	spa->spa_import_flags = flags;
5401
5402	/*
5403	 * Verbatim import - Take a pool and insert it into the namespace
5404	 * as if it had been loaded at boot.
5405	 */
5406	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
5407		if (props != NULL)
5408			spa_configfile_set(spa, props, B_FALSE);
5409
5410		spa_write_cachefile(spa, B_FALSE, B_TRUE);
5411		spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
5412		zfs_dbgmsg("spa_import: verbatim import of %s", pool);
5413		mutex_exit(&spa_namespace_lock);
5414		return (0);
5415	}
5416
5417	spa_activate(spa, mode);
5418
5419	/*
5420	 * Don't start async tasks until we know everything is healthy.
5421	 */
5422	spa_async_suspend(spa);
5423
5424	zpool_get_load_policy(config, &policy);
5425	if (policy.zlp_rewind & ZPOOL_DO_REWIND)
5426		state = SPA_LOAD_RECOVER;
5427
5428	spa->spa_config_source = SPA_CONFIG_SRC_TRYIMPORT;
5429
5430	if (state != SPA_LOAD_RECOVER) {
5431		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
5432		zfs_dbgmsg("spa_import: importing %s", pool);
5433	} else {
5434		zfs_dbgmsg("spa_import: importing %s, max_txg=%lld "
5435		    "(RECOVERY MODE)", pool, (longlong_t)policy.zlp_txg);
5436	}
5437	error = spa_load_best(spa, state, policy.zlp_txg, policy.zlp_rewind);
5438
5439	/*
5440	 * Propagate anything learned while loading the pool and pass it
5441	 * back to caller (i.e. rewind info, missing devices, etc).
5442	 */
5443	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
5444	    spa->spa_load_info) == 0);
5445
5446	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5447	/*
5448	 * Toss any existing sparelist, as it doesn't have any validity
5449	 * anymore, and conflicts with spa_has_spare().
5450	 */
5451	if (spa->spa_spares.sav_config) {
5452		nvlist_free(spa->spa_spares.sav_config);
5453		spa->spa_spares.sav_config = NULL;
5454		spa_load_spares(spa);
5455	}
5456	if (spa->spa_l2cache.sav_config) {
5457		nvlist_free(spa->spa_l2cache.sav_config);
5458		spa->spa_l2cache.sav_config = NULL;
5459		spa_load_l2cache(spa);
5460	}
5461
5462	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
5463	    &nvroot) == 0);
5464	if (error == 0)
5465		error = spa_validate_aux(spa, nvroot, -1ULL,
5466		    VDEV_ALLOC_SPARE);
5467	if (error == 0)
5468		error = spa_validate_aux(spa, nvroot, -1ULL,
5469		    VDEV_ALLOC_L2CACHE);
5470	spa_config_exit(spa, SCL_ALL, FTAG);
5471
5472	if (props != NULL)
5473		spa_configfile_set(spa, props, B_FALSE);
5474
5475	if (error != 0 || (props && spa_writeable(spa) &&
5476	    (error = spa_prop_set(spa, props)))) {
5477		spa_unload(spa);
5478		spa_deactivate(spa);
5479		spa_remove(spa);
5480		mutex_exit(&spa_namespace_lock);
5481		return (error);
5482	}
5483
5484	spa_async_resume(spa);
5485
5486	/*
5487	 * Override any spares and level 2 cache devices as specified by
5488	 * the user, as these may have correct device names/devids, etc.
5489	 */
5490	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
5491	    &spares, &nspares) == 0) {
5492		if (spa->spa_spares.sav_config)
5493			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
5494			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
5495		else
5496			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
5497			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
5498		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
5499		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
5500		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5501		spa_load_spares(spa);
5502		spa_config_exit(spa, SCL_ALL, FTAG);
5503		spa->spa_spares.sav_sync = B_TRUE;
5504	}
5505	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
5506	    &l2cache, &nl2cache) == 0) {
5507		if (spa->spa_l2cache.sav_config)
5508			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
5509			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
5510		else
5511			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
5512			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
5513		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
5514		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
5515		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5516		spa_load_l2cache(spa);
5517		spa_config_exit(spa, SCL_ALL, FTAG);
5518		spa->spa_l2cache.sav_sync = B_TRUE;
5519	}
5520
5521	/*
5522	 * Check for any removed devices.
5523	 */
5524	if (spa->spa_autoreplace) {
5525		spa_aux_check_removed(&spa->spa_spares);
5526		spa_aux_check_removed(&spa->spa_l2cache);
5527	}
5528
5529	if (spa_writeable(spa)) {
5530		/*
5531		 * Update the config cache to include the newly-imported pool.
5532		 */
5533		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5534	}
5535
5536	/*
5537	 * It's possible that the pool was expanded while it was exported.
5538	 * We kick off an async task to handle this for us.
5539	 */
5540	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
5541
5542	spa_history_log_version(spa, "import");
5543
5544	spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
5545
5546	mutex_exit(&spa_namespace_lock);
5547
5548	return (0);
5549}
5550
5551nvlist_t *
5552spa_tryimport(nvlist_t *tryconfig)
5553{
5554	nvlist_t *config = NULL;
5555	char *poolname, *cachefile;
5556	spa_t *spa;
5557	uint64_t state;
5558	int error;
5559	zpool_load_policy_t policy;
5560
5561	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
5562		return (NULL);
5563
5564	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
5565		return (NULL);
5566
5567	/*
5568	 * Create and initialize the spa structure.
5569	 */
5570	mutex_enter(&spa_namespace_lock);
5571	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
5572	spa_activate(spa, FREAD);
5573
5574	/*
5575	 * Rewind pool if a max txg was provided.
5576	 */
5577	zpool_get_load_policy(spa->spa_config, &policy);
5578	if (policy.zlp_txg != UINT64_MAX) {
5579		spa->spa_load_max_txg = policy.zlp_txg;
5580		spa->spa_extreme_rewind = B_TRUE;
5581		zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld",
5582		    poolname, (longlong_t)policy.zlp_txg);
5583	} else {
5584		zfs_dbgmsg("spa_tryimport: importing %s", poolname);
5585	}
5586
5587	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_CACHEFILE, &cachefile)
5588	    == 0) {
5589		zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile);
5590		spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
5591	} else {
5592		spa->spa_config_source = SPA_CONFIG_SRC_SCAN;
5593	}
5594
5595	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING);
5596
5597	/*
5598	 * If 'tryconfig' was at least parsable, return the current config.
5599	 */
5600	if (spa->spa_root_vdev != NULL) {
5601		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
5602		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
5603		    poolname) == 0);
5604		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5605		    state) == 0);
5606		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
5607		    spa->spa_uberblock.ub_timestamp) == 0);
5608		VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
5609		    spa->spa_load_info) == 0);
5610
5611		/*
5612		 * If the bootfs property exists on this pool then we
5613		 * copy it out so that external consumers can tell which
5614		 * pools are bootable.
5615		 */
5616		if ((!error || error == EEXIST) && spa->spa_bootfs) {
5617			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5618
5619			/*
5620			 * We have to play games with the name since the
5621			 * pool was opened as TRYIMPORT_NAME.
5622			 */
5623			if (dsl_dsobj_to_dsname(spa_name(spa),
5624			    spa->spa_bootfs, tmpname) == 0) {
5625				char *cp;
5626				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5627
5628				cp = strchr(tmpname, '/');
5629				if (cp == NULL) {
5630					(void) strlcpy(dsname, tmpname,
5631					    MAXPATHLEN);
5632				} else {
5633					(void) snprintf(dsname, MAXPATHLEN,
5634					    "%s/%s", poolname, ++cp);
5635				}
5636				VERIFY(nvlist_add_string(config,
5637				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
5638				kmem_free(dsname, MAXPATHLEN);
5639			}
5640			kmem_free(tmpname, MAXPATHLEN);
5641		}
5642
5643		/*
5644		 * Add the list of hot spares and level 2 cache devices.
5645		 */
5646		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5647		spa_add_spares(spa, config);
5648		spa_add_l2cache(spa, config);
5649		spa_config_exit(spa, SCL_CONFIG, FTAG);
5650	}
5651
5652	spa_unload(spa);
5653	spa_deactivate(spa);
5654	spa_remove(spa);
5655	mutex_exit(&spa_namespace_lock);
5656
5657	return (config);
5658}
5659
5660/*
5661 * Pool export/destroy
5662 *
5663 * The act of destroying or exporting a pool is very simple.  We make sure there
5664 * is no more pending I/O and any references to the pool are gone.  Then, we
5665 * update the pool state and sync all the labels to disk, removing the
5666 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
5667 * we don't sync the labels or remove the configuration cache.
5668 */
5669static int
5670spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
5671    boolean_t force, boolean_t hardforce)
5672{
5673	spa_t *spa;
5674
5675	if (oldconfig)
5676		*oldconfig = NULL;
5677
5678	if (!(spa_mode_global & FWRITE))
5679		return (SET_ERROR(EROFS));
5680
5681	mutex_enter(&spa_namespace_lock);
5682	if ((spa = spa_lookup(pool)) == NULL) {
5683		mutex_exit(&spa_namespace_lock);
5684		return (SET_ERROR(ENOENT));
5685	}
5686
5687	/*
5688	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
5689	 * reacquire the namespace lock, and see if we can export.
5690	 */
5691	spa_open_ref(spa, FTAG);
5692	mutex_exit(&spa_namespace_lock);
5693	spa_async_suspend(spa);
5694	mutex_enter(&spa_namespace_lock);
5695	spa_close(spa, FTAG);
5696
5697	/*
5698	 * The pool will be in core if it's openable,
5699	 * in which case we can modify its state.
5700	 */
5701	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
5702
5703		/*
5704		 * Objsets may be open only because they're dirty, so we
5705		 * have to force it to sync before checking spa_refcnt.
5706		 */
5707		txg_wait_synced(spa->spa_dsl_pool, 0);
5708		spa_evicting_os_wait(spa);
5709
5710		/*
5711		 * A pool cannot be exported or destroyed if there are active
5712		 * references.  If we are resetting a pool, allow references by
5713		 * fault injection handlers.
5714		 */
5715		if (!spa_refcount_zero(spa) ||
5716		    (spa->spa_inject_ref != 0 &&
5717		    new_state != POOL_STATE_UNINITIALIZED)) {
5718			spa_async_resume(spa);
5719			mutex_exit(&spa_namespace_lock);
5720			return (SET_ERROR(EBUSY));
5721		}
5722
5723		/*
5724		 * A pool cannot be exported if it has an active shared spare.
5725		 * This is to prevent other pools stealing the active spare
5726		 * from an exported pool. At user's own will, such pool can
5727		 * be forcedly exported.
5728		 */
5729		if (!force && new_state == POOL_STATE_EXPORTED &&
5730		    spa_has_active_shared_spare(spa)) {
5731			spa_async_resume(spa);
5732			mutex_exit(&spa_namespace_lock);
5733			return (SET_ERROR(EXDEV));
5734		}
5735
5736		/*
5737		 * We're about to export or destroy this pool. Make sure
5738		 * we stop all initializtion activity here before we
5739		 * set the spa_final_txg. This will ensure that all
5740		 * dirty data resulting from the initialization is
5741		 * committed to disk before we unload the pool.
5742		 */
5743		if (spa->spa_root_vdev != NULL) {
5744			vdev_initialize_stop_all(spa->spa_root_vdev,
5745			    VDEV_INITIALIZE_ACTIVE);
5746		}
5747
5748		/*
5749		 * We want this to be reflected on every label,
5750		 * so mark them all dirty.  spa_unload() will do the
5751		 * final sync that pushes these changes out.
5752		 */
5753		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
5754			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5755			spa->spa_state = new_state;
5756			spa->spa_final_txg = spa_last_synced_txg(spa) +
5757			    TXG_DEFER_SIZE + 1;
5758			vdev_config_dirty(spa->spa_root_vdev);
5759			spa_config_exit(spa, SCL_ALL, FTAG);
5760		}
5761	}
5762
5763	spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY);
5764
5765	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5766		spa_unload(spa);
5767		spa_deactivate(spa);
5768	}
5769
5770	if (oldconfig && spa->spa_config)
5771		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
5772
5773	if (new_state != POOL_STATE_UNINITIALIZED) {
5774		if (!hardforce)
5775			spa_write_cachefile(spa, B_TRUE, B_TRUE);
5776		spa_remove(spa);
5777	}
5778	mutex_exit(&spa_namespace_lock);
5779
5780	return (0);
5781}
5782
5783/*
5784 * Destroy a storage pool.
5785 */
5786int
5787spa_destroy(char *pool)
5788{
5789	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
5790	    B_FALSE, B_FALSE));
5791}
5792
5793/*
5794 * Export a storage pool.
5795 */
5796int
5797spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
5798    boolean_t hardforce)
5799{
5800	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
5801	    force, hardforce));
5802}
5803
5804/*
5805 * Similar to spa_export(), this unloads the spa_t without actually removing it
5806 * from the namespace in any way.
5807 */
5808int
5809spa_reset(char *pool)
5810{
5811	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
5812	    B_FALSE, B_FALSE));
5813}
5814
5815/*
5816 * ==========================================================================
5817 * Device manipulation
5818 * ==========================================================================
5819 */
5820
5821/*
5822 * Add a device to a storage pool.
5823 */
5824int
5825spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
5826{
5827	uint64_t txg, id;
5828	int error;
5829	vdev_t *rvd = spa->spa_root_vdev;
5830	vdev_t *vd, *tvd;
5831	nvlist_t **spares, **l2cache;
5832	uint_t nspares, nl2cache;
5833
5834	ASSERT(spa_writeable(spa));
5835
5836	txg = spa_vdev_enter(spa);
5837
5838	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
5839	    VDEV_ALLOC_ADD)) != 0)
5840		return (spa_vdev_exit(spa, NULL, txg, error));
5841
5842	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
5843
5844	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
5845	    &nspares) != 0)
5846		nspares = 0;
5847
5848	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
5849	    &nl2cache) != 0)
5850		nl2cache = 0;
5851
5852	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
5853		return (spa_vdev_exit(spa, vd, txg, EINVAL));
5854
5855	if (vd->vdev_children != 0 &&
5856	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
5857		return (spa_vdev_exit(spa, vd, txg, error));
5858
5859	/*
5860	 * We must validate the spares and l2cache devices after checking the
5861	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
5862	 */
5863	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
5864		return (spa_vdev_exit(spa, vd, txg, error));
5865
5866	/*
5867	 * If we are in the middle of a device removal, we can only add
5868	 * devices which match the existing devices in the pool.
5869	 * If we are in the middle of a removal, or have some indirect
5870	 * vdevs, we can not add raidz toplevels.
5871	 */
5872	if (spa->spa_vdev_removal != NULL ||
5873	    spa->spa_removing_phys.sr_prev_indirect_vdev != -1) {
5874		for (int c = 0; c < vd->vdev_children; c++) {
5875			tvd = vd->vdev_child[c];
5876			if (spa->spa_vdev_removal != NULL &&
5877			    tvd->vdev_ashift != spa->spa_max_ashift) {
5878				return (spa_vdev_exit(spa, vd, txg, EINVAL));
5879			}
5880			/* Fail if top level vdev is raidz */
5881			if (tvd->vdev_ops == &vdev_raidz_ops) {
5882				return (spa_vdev_exit(spa, vd, txg, EINVAL));
5883			}
5884			/*
5885			 * Need the top level mirror to be
5886			 * a mirror of leaf vdevs only
5887			 */
5888			if (tvd->vdev_ops == &vdev_mirror_ops) {
5889				for (uint64_t cid = 0;
5890				    cid < tvd->vdev_children; cid++) {
5891					vdev_t *cvd = tvd->vdev_child[cid];
5892					if (!cvd->vdev_ops->vdev_op_leaf) {
5893						return (spa_vdev_exit(spa, vd,
5894						    txg, EINVAL));
5895					}
5896				}
5897			}
5898		}
5899	}
5900
5901	for (int c = 0; c < vd->vdev_children; c++) {
5902
5903		/*
5904		 * Set the vdev id to the first hole, if one exists.
5905		 */
5906		for (id = 0; id < rvd->vdev_children; id++) {
5907			if (rvd->vdev_child[id]->vdev_ishole) {
5908				vdev_free(rvd->vdev_child[id]);
5909				break;
5910			}
5911		}
5912		tvd = vd->vdev_child[c];
5913		vdev_remove_child(vd, tvd);
5914		tvd->vdev_id = id;
5915		vdev_add_child(rvd, tvd);
5916		vdev_config_dirty(tvd);
5917	}
5918
5919	if (nspares != 0) {
5920		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
5921		    ZPOOL_CONFIG_SPARES);
5922		spa_load_spares(spa);
5923		spa->spa_spares.sav_sync = B_TRUE;
5924	}
5925
5926	if (nl2cache != 0) {
5927		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
5928		    ZPOOL_CONFIG_L2CACHE);
5929		spa_load_l2cache(spa);
5930		spa->spa_l2cache.sav_sync = B_TRUE;
5931	}
5932
5933	/*
5934	 * We have to be careful when adding new vdevs to an existing pool.
5935	 * If other threads start allocating from these vdevs before we
5936	 * sync the config cache, and we lose power, then upon reboot we may
5937	 * fail to open the pool because there are DVAs that the config cache
5938	 * can't translate.  Therefore, we first add the vdevs without
5939	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
5940	 * and then let spa_config_update() initialize the new metaslabs.
5941	 *
5942	 * spa_load() checks for added-but-not-initialized vdevs, so that
5943	 * if we lose power at any point in this sequence, the remaining
5944	 * steps will be completed the next time we load the pool.
5945	 */
5946	(void) spa_vdev_exit(spa, vd, txg, 0);
5947
5948	mutex_enter(&spa_namespace_lock);
5949	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5950	spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD);
5951	mutex_exit(&spa_namespace_lock);
5952
5953	return (0);
5954}
5955
5956/*
5957 * Attach a device to a mirror.  The arguments are the path to any device
5958 * in the mirror, and the nvroot for the new device.  If the path specifies
5959 * a device that is not mirrored, we automatically insert the mirror vdev.
5960 *
5961 * If 'replacing' is specified, the new device is intended to replace the
5962 * existing device; in this case the two devices are made into their own
5963 * mirror using the 'replacing' vdev, which is functionally identical to
5964 * the mirror vdev (it actually reuses all the same ops) but has a few
5965 * extra rules: you can't attach to it after it's been created, and upon
5966 * completion of resilvering, the first disk (the one being replaced)
5967 * is automatically detached.
5968 */
5969int
5970spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
5971{
5972	uint64_t txg, dtl_max_txg;
5973	vdev_t *rvd = spa->spa_root_vdev;
5974	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
5975	vdev_ops_t *pvops;
5976	char *oldvdpath, *newvdpath;
5977	int newvd_isspare;
5978	int error;
5979
5980	ASSERT(spa_writeable(spa));
5981
5982	txg = spa_vdev_enter(spa);
5983
5984	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
5985
5986	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5987	if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
5988		error = (spa_has_checkpoint(spa)) ?
5989		    ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
5990		return (spa_vdev_exit(spa, NULL, txg, error));
5991	}
5992
5993	if (spa->spa_vdev_removal != NULL)
5994		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5995
5996	if (oldvd == NULL)
5997		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
5998
5999	if (!oldvd->vdev_ops->vdev_op_leaf)
6000		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6001
6002	pvd = oldvd->vdev_parent;
6003
6004	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
6005	    VDEV_ALLOC_ATTACH)) != 0)
6006		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6007
6008	if (newrootvd->vdev_children != 1)
6009		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
6010
6011	newvd = newrootvd->vdev_child[0];
6012
6013	if (!newvd->vdev_ops->vdev_op_leaf)
6014		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
6015
6016	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
6017		return (spa_vdev_exit(spa, newrootvd, txg, error));
6018
6019	/*
6020	 * Spares can't replace logs
6021	 */
6022	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
6023		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6024
6025	if (!replacing) {
6026		/*
6027		 * For attach, the only allowable parent is a mirror or the root
6028		 * vdev.
6029		 */
6030		if (pvd->vdev_ops != &vdev_mirror_ops &&
6031		    pvd->vdev_ops != &vdev_root_ops)
6032			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6033
6034		pvops = &vdev_mirror_ops;
6035	} else {
6036		/*
6037		 * Active hot spares can only be replaced by inactive hot
6038		 * spares.
6039		 */
6040		if (pvd->vdev_ops == &vdev_spare_ops &&
6041		    oldvd->vdev_isspare &&
6042		    !spa_has_spare(spa, newvd->vdev_guid))
6043			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6044
6045		/*
6046		 * If the source is a hot spare, and the parent isn't already a
6047		 * spare, then we want to create a new hot spare.  Otherwise, we
6048		 * want to create a replacing vdev.  The user is not allowed to
6049		 * attach to a spared vdev child unless the 'isspare' state is
6050		 * the same (spare replaces spare, non-spare replaces
6051		 * non-spare).
6052		 */
6053		if (pvd->vdev_ops == &vdev_replacing_ops &&
6054		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
6055			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6056		} else if (pvd->vdev_ops == &vdev_spare_ops &&
6057		    newvd->vdev_isspare != oldvd->vdev_isspare) {
6058			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6059		}
6060
6061		if (newvd->vdev_isspare)
6062			pvops = &vdev_spare_ops;
6063		else
6064			pvops = &vdev_replacing_ops;
6065	}
6066
6067	/*
6068	 * Make sure the new device is big enough.
6069	 */
6070	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
6071		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
6072
6073	/*
6074	 * The new device cannot have a higher alignment requirement
6075	 * than the top-level vdev.
6076	 */
6077	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
6078		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
6079
6080	/*
6081	 * If this is an in-place replacement, update oldvd's path and devid
6082	 * to make it distinguishable from newvd, and unopenable from now on.
6083	 */
6084	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
6085		spa_strfree(oldvd->vdev_path);
6086		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
6087		    KM_SLEEP);
6088		(void) sprintf(oldvd->vdev_path, "%s/%s",
6089		    newvd->vdev_path, "old");
6090		if (oldvd->vdev_devid != NULL) {
6091			spa_strfree(oldvd->vdev_devid);
6092			oldvd->vdev_devid = NULL;
6093		}
6094	}
6095
6096	/* mark the device being resilvered */
6097	newvd->vdev_resilver_txg = txg;
6098
6099	/*
6100	 * If the parent is not a mirror, or if we're replacing, insert the new
6101	 * mirror/replacing/spare vdev above oldvd.
6102	 */
6103	if (pvd->vdev_ops != pvops)
6104		pvd = vdev_add_parent(oldvd, pvops);
6105
6106	ASSERT(pvd->vdev_top->vdev_parent == rvd);
6107	ASSERT(pvd->vdev_ops == pvops);
6108	ASSERT(oldvd->vdev_parent == pvd);
6109
6110	/*
6111	 * Extract the new device from its root and add it to pvd.
6112	 */
6113	vdev_remove_child(newrootvd, newvd);
6114	newvd->vdev_id = pvd->vdev_children;
6115	newvd->vdev_crtxg = oldvd->vdev_crtxg;
6116	vdev_add_child(pvd, newvd);
6117
6118	tvd = newvd->vdev_top;
6119	ASSERT(pvd->vdev_top == tvd);
6120	ASSERT(tvd->vdev_parent == rvd);
6121
6122	vdev_config_dirty(tvd);
6123
6124	/*
6125	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
6126	 * for any dmu_sync-ed blocks.  It will propagate upward when
6127	 * spa_vdev_exit() calls vdev_dtl_reassess().
6128	 */
6129	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
6130
6131	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
6132	    dtl_max_txg - TXG_INITIAL);
6133
6134	if (newvd->vdev_isspare) {
6135		spa_spare_activate(newvd);
6136		spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE);
6137	}
6138
6139	oldvdpath = spa_strdup(oldvd->vdev_path);
6140	newvdpath = spa_strdup(newvd->vdev_path);
6141	newvd_isspare = newvd->vdev_isspare;
6142
6143	/*
6144	 * Mark newvd's DTL dirty in this txg.
6145	 */
6146	vdev_dirty(tvd, VDD_DTL, newvd, txg);
6147
6148	/*
6149	 * Schedule the resilver to restart in the future. We do this to
6150	 * ensure that dmu_sync-ed blocks have been stitched into the
6151	 * respective datasets. We do not do this if resilvers have been
6152	 * deferred.
6153	 */
6154	if (dsl_scan_resilvering(spa_get_dsl(spa)) &&
6155	    spa_feature_is_enabled(spa, SPA_FEATURE_RESILVER_DEFER))
6156		vdev_set_deferred_resilver(spa, newvd);
6157	else
6158		dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
6159
6160	if (spa->spa_bootfs)
6161		spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH);
6162
6163	spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH);
6164
6165	/*
6166	 * Commit the config
6167	 */
6168	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
6169
6170	spa_history_log_internal(spa, "vdev attach", NULL,
6171	    "%s vdev=%s %s vdev=%s",
6172	    replacing && newvd_isspare ? "spare in" :
6173	    replacing ? "replace" : "attach", newvdpath,
6174	    replacing ? "for" : "to", oldvdpath);
6175
6176	spa_strfree(oldvdpath);
6177	spa_strfree(newvdpath);
6178
6179	return (0);
6180}
6181
6182/*
6183 * Detach a device from a mirror or replacing vdev.
6184 *
6185 * If 'replace_done' is specified, only detach if the parent
6186 * is a replacing vdev.
6187 */
6188int
6189spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
6190{
6191	uint64_t txg;
6192	int error;
6193	vdev_t *rvd = spa->spa_root_vdev;
6194	vdev_t *vd, *pvd, *cvd, *tvd;
6195	boolean_t unspare = B_FALSE;
6196	uint64_t unspare_guid = 0;
6197	char *vdpath;
6198
6199	ASSERT(spa_writeable(spa));
6200
6201	txg = spa_vdev_enter(spa);
6202
6203	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
6204
6205	/*
6206	 * Besides being called directly from the userland through the
6207	 * ioctl interface, spa_vdev_detach() can be potentially called
6208	 * at the end of spa_vdev_resilver_done().
6209	 *
6210	 * In the regular case, when we have a checkpoint this shouldn't
6211	 * happen as we never empty the DTLs of a vdev during the scrub