spa.c revision 54811da5ac6b517992fdc173df5d605e4e61fdc0
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, 2018 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 Datto Inc.
32 * Copyright 2018 OmniOS Community Edition (OmniOSce) Association.
33 */
34
35/*
36 * SPA: Storage Pool Allocator
37 *
38 * This file contains all the routines used when modifying on-disk SPA state.
39 * This includes opening, importing, destroying, exporting a pool, and syncing a
40 * pool.
41 */
42
43#include <sys/zfs_context.h>
44#include <sys/fm/fs/zfs.h>
45#include <sys/spa_impl.h>
46#include <sys/zio.h>
47#include <sys/zio_checksum.h>
48#include <sys/dmu.h>
49#include <sys/dmu_tx.h>
50#include <sys/zap.h>
51#include <sys/zil.h>
52#include <sys/ddt.h>
53#include <sys/vdev_impl.h>
54#include <sys/vdev_removal.h>
55#include <sys/vdev_indirect_mapping.h>
56#include <sys/vdev_indirect_births.h>
57#include <sys/vdev_initialize.h>
58#include <sys/metaslab.h>
59#include <sys/metaslab_impl.h>
60#include <sys/uberblock_impl.h>
61#include <sys/txg.h>
62#include <sys/avl.h>
63#include <sys/bpobj.h>
64#include <sys/dmu_traverse.h>
65#include <sys/dmu_objset.h>
66#include <sys/unique.h>
67#include <sys/dsl_pool.h>
68#include <sys/dsl_dataset.h>
69#include <sys/dsl_dir.h>
70#include <sys/dsl_prop.h>
71#include <sys/dsl_synctask.h>
72#include <sys/fs/zfs.h>
73#include <sys/arc.h>
74#include <sys/callb.h>
75#include <sys/systeminfo.h>
76#include <sys/spa_boot.h>
77#include <sys/zfs_ioctl.h>
78#include <sys/dsl_scan.h>
79#include <sys/zfeature.h>
80#include <sys/dsl_destroy.h>
81#include <sys/abd.h>
82
83#ifdef	_KERNEL
84#include <sys/bootprops.h>
85#include <sys/callb.h>
86#include <sys/cpupart.h>
87#include <sys/pool.h>
88#include <sys/sysdc.h>
89#include <sys/zone.h>
90#endif	/* _KERNEL */
91
92#include "zfs_prop.h"
93#include "zfs_comutil.h"
94
95/*
96 * The interval, in seconds, at which failed configuration cache file writes
97 * should be retried.
98 */
99int zfs_ccw_retry_interval = 300;
100
101typedef enum zti_modes {
102	ZTI_MODE_FIXED,			/* value is # of threads (min 1) */
103	ZTI_MODE_BATCH,			/* cpu-intensive; value is ignored */
104	ZTI_MODE_NULL,			/* don't create a taskq */
105	ZTI_NMODES
106} zti_modes_t;
107
108#define	ZTI_P(n, q)	{ ZTI_MODE_FIXED, (n), (q) }
109#define	ZTI_BATCH	{ ZTI_MODE_BATCH, 0, 1 }
110#define	ZTI_NULL	{ ZTI_MODE_NULL, 0, 0 }
111
112#define	ZTI_N(n)	ZTI_P(n, 1)
113#define	ZTI_ONE		ZTI_N(1)
114
115typedef struct zio_taskq_info {
116	zti_modes_t zti_mode;
117	uint_t zti_value;
118	uint_t zti_count;
119} zio_taskq_info_t;
120
121static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
122	"issue", "issue_high", "intr", "intr_high"
123};
124
125/*
126 * This table defines the taskq settings for each ZFS I/O type. When
127 * initializing a pool, we use this table to create an appropriately sized
128 * taskq. Some operations are low volume and therefore have a small, static
129 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
130 * macros. Other operations process a large amount of data; the ZTI_BATCH
131 * macro causes us to create a taskq oriented for throughput. Some operations
132 * are so high frequency and short-lived that the taskq itself can become a a
133 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
134 * additional degree of parallelism specified by the number of threads per-
135 * taskq and the number of taskqs; when dispatching an event in this case, the
136 * particular taskq is chosen at random.
137 *
138 * The different taskq priorities are to handle the different contexts (issue
139 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
140 * need to be handled with minimum delay.
141 */
142const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
143	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
144	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* NULL */
145	{ ZTI_N(8),	ZTI_NULL,	ZTI_P(12, 8),	ZTI_NULL }, /* READ */
146	{ ZTI_BATCH,	ZTI_N(5),	ZTI_N(8),	ZTI_N(5) }, /* WRITE */
147	{ ZTI_P(12, 8),	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* FREE */
148	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* CLAIM */
149	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* IOCTL */
150};
151
152static void spa_sync_version(void *arg, dmu_tx_t *tx);
153static void spa_sync_props(void *arg, dmu_tx_t *tx);
154static boolean_t spa_has_active_shared_spare(spa_t *spa);
155static int spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport);
156static void spa_vdev_resilver_done(spa_t *spa);
157
158uint_t		zio_taskq_batch_pct = 75;	/* 1 thread per cpu in pset */
159id_t		zio_taskq_psrset_bind = PS_NONE;
160boolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
161uint_t		zio_taskq_basedc = 80;		/* base duty cycle */
162
163boolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
164extern int	zfs_sync_pass_deferred_free;
165
166/*
167 * Report any spa_load_verify errors found, but do not fail spa_load.
168 * This is used by zdb to analyze non-idle pools.
169 */
170boolean_t	spa_load_verify_dryrun = B_FALSE;
171
172/*
173 * This (illegal) pool name is used when temporarily importing a spa_t in order
174 * to get the vdev stats associated with the imported devices.
175 */
176#define	TRYIMPORT_NAME	"$import"
177
178/*
179 * For debugging purposes: print out vdev tree during pool import.
180 */
181boolean_t	spa_load_print_vdev_tree = B_FALSE;
182
183/*
184 * A non-zero value for zfs_max_missing_tvds means that we allow importing
185 * pools with missing top-level vdevs. This is strictly intended for advanced
186 * pool recovery cases since missing data is almost inevitable. Pools with
187 * missing devices can only be imported read-only for safety reasons, and their
188 * fail-mode will be automatically set to "continue".
189 *
190 * With 1 missing vdev we should be able to import the pool and mount all
191 * datasets. User data that was not modified after the missing device has been
192 * added should be recoverable. This means that snapshots created prior to the
193 * addition of that device should be completely intact.
194 *
195 * With 2 missing vdevs, some datasets may fail to mount since there are
196 * dataset statistics that are stored as regular metadata. Some data might be
197 * recoverable if those vdevs were added recently.
198 *
199 * With 3 or more missing vdevs, the pool is severely damaged and MOS entries
200 * may be missing entirely. Chances of data recovery are very low. Note that
201 * there are also risks of performing an inadvertent rewind as we might be
202 * missing all the vdevs with the latest uberblocks.
203 */
204uint64_t	zfs_max_missing_tvds = 0;
205
206/*
207 * The parameters below are similar to zfs_max_missing_tvds but are only
208 * intended for a preliminary open of the pool with an untrusted config which
209 * might be incomplete or out-dated.
210 *
211 * We are more tolerant for pools opened from a cachefile since we could have
212 * an out-dated cachefile where a device removal was not registered.
213 * We could have set the limit arbitrarily high but in the case where devices
214 * are really missing we would want to return the proper error codes; we chose
215 * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available
216 * and we get a chance to retrieve the trusted config.
217 */
218uint64_t	zfs_max_missing_tvds_cachefile = SPA_DVAS_PER_BP - 1;
219
220/*
221 * In the case where config was assembled by scanning device paths (/dev/dsks
222 * by default) we are less tolerant since all the existing devices should have
223 * been detected and we want spa_load to return the right error codes.
224 */
225uint64_t	zfs_max_missing_tvds_scan = 0;
226
227/*
228 * Debugging aid that pauses spa_sync() towards the end.
229 */
230boolean_t	zfs_pause_spa_sync = B_FALSE;
231
232/*
233 * ==========================================================================
234 * SPA properties routines
235 * ==========================================================================
236 */
237
238/*
239 * Add a (source=src, propname=propval) list to an nvlist.
240 */
241static void
242spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
243    uint64_t intval, zprop_source_t src)
244{
245	const char *propname = zpool_prop_to_name(prop);
246	nvlist_t *propval;
247
248	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
249	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
250
251	if (strval != NULL)
252		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
253	else
254		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
255
256	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
257	nvlist_free(propval);
258}
259
260/*
261 * Get property values from the spa configuration.
262 */
263static void
264spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
265{
266	vdev_t *rvd = spa->spa_root_vdev;
267	dsl_pool_t *pool = spa->spa_dsl_pool;
268	uint64_t size, alloc, cap, version;
269	zprop_source_t src = ZPROP_SRC_NONE;
270	spa_config_dirent_t *dp;
271	metaslab_class_t *mc = spa_normal_class(spa);
272
273	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
274
275	if (rvd != NULL) {
276		alloc = metaslab_class_get_alloc(spa_normal_class(spa));
277		size = metaslab_class_get_space(spa_normal_class(spa));
278		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
279		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
280		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
281		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
282		    size - alloc, src);
283		spa_prop_add_list(*nvp, ZPOOL_PROP_CHECKPOINT, NULL,
284		    spa->spa_checkpoint_info.sci_dspace, src);
285
286		spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
287		    metaslab_class_fragmentation(mc), src);
288		spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
289		    metaslab_class_expandable_space(mc), src);
290		spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
291		    (spa_mode(spa) == FREAD), src);
292
293		cap = (size == 0) ? 0 : (alloc * 100 / size);
294		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
295
296		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
297		    ddt_get_pool_dedup_ratio(spa), src);
298
299		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
300		    rvd->vdev_state, src);
301
302		version = spa_version(spa);
303		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
304			src = ZPROP_SRC_DEFAULT;
305		else
306			src = ZPROP_SRC_LOCAL;
307		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
308	}
309
310	if (pool != NULL) {
311		/*
312		 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
313		 * when opening pools before this version freedir will be NULL.
314		 */
315		if (pool->dp_free_dir != NULL) {
316			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
317			    dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
318			    src);
319		} else {
320			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
321			    NULL, 0, src);
322		}
323
324		if (pool->dp_leak_dir != NULL) {
325			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
326			    dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
327			    src);
328		} else {
329			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
330			    NULL, 0, src);
331		}
332	}
333
334	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
335
336	if (spa->spa_comment != NULL) {
337		spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
338		    0, ZPROP_SRC_LOCAL);
339	}
340
341	if (spa->spa_root != NULL)
342		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
343		    0, ZPROP_SRC_LOCAL);
344
345	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
346		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
347		    MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
348	} else {
349		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
350		    SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
351	}
352
353	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE)) {
354		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL,
355		    DNODE_MAX_SIZE, ZPROP_SRC_NONE);
356	} else {
357		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL,
358		    DNODE_MIN_SIZE, ZPROP_SRC_NONE);
359	}
360
361	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
362		if (dp->scd_path == NULL) {
363			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
364			    "none", 0, ZPROP_SRC_LOCAL);
365		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
366			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
367			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
368		}
369	}
370}
371
372/*
373 * Get zpool property values.
374 */
375int
376spa_prop_get(spa_t *spa, nvlist_t **nvp)
377{
378	objset_t *mos = spa->spa_meta_objset;
379	zap_cursor_t zc;
380	zap_attribute_t za;
381	int err;
382
383	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
384
385	mutex_enter(&spa->spa_props_lock);
386
387	/*
388	 * Get properties from the spa config.
389	 */
390	spa_prop_get_config(spa, nvp);
391
392	/* If no pool property object, no more prop to get. */
393	if (mos == NULL || spa->spa_pool_props_object == 0) {
394		mutex_exit(&spa->spa_props_lock);
395		return (0);
396	}
397
398	/*
399	 * Get properties from the MOS pool property object.
400	 */
401	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
402	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
403	    zap_cursor_advance(&zc)) {
404		uint64_t intval = 0;
405		char *strval = NULL;
406		zprop_source_t src = ZPROP_SRC_DEFAULT;
407		zpool_prop_t prop;
408
409		if ((prop = zpool_name_to_prop(za.za_name)) == ZPOOL_PROP_INVAL)
410			continue;
411
412		switch (za.za_integer_length) {
413		case 8:
414			/* integer property */
415			if (za.za_first_integer !=
416			    zpool_prop_default_numeric(prop))
417				src = ZPROP_SRC_LOCAL;
418
419			if (prop == ZPOOL_PROP_BOOTFS) {
420				dsl_pool_t *dp;
421				dsl_dataset_t *ds = NULL;
422
423				dp = spa_get_dsl(spa);
424				dsl_pool_config_enter(dp, FTAG);
425				err = dsl_dataset_hold_obj(dp,
426				    za.za_first_integer, FTAG, &ds);
427				if (err != 0) {
428					dsl_pool_config_exit(dp, FTAG);
429					break;
430				}
431
432				strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN,
433				    KM_SLEEP);
434				dsl_dataset_name(ds, strval);
435				dsl_dataset_rele(ds, FTAG);
436				dsl_pool_config_exit(dp, FTAG);
437			} else {
438				strval = NULL;
439				intval = za.za_first_integer;
440			}
441
442			spa_prop_add_list(*nvp, prop, strval, intval, src);
443
444			if (strval != NULL)
445				kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN);
446
447			break;
448
449		case 1:
450			/* string property */
451			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
452			err = zap_lookup(mos, spa->spa_pool_props_object,
453			    za.za_name, 1, za.za_num_integers, strval);
454			if (err) {
455				kmem_free(strval, za.za_num_integers);
456				break;
457			}
458			spa_prop_add_list(*nvp, prop, strval, 0, src);
459			kmem_free(strval, za.za_num_integers);
460			break;
461
462		default:
463			break;
464		}
465	}
466	zap_cursor_fini(&zc);
467	mutex_exit(&spa->spa_props_lock);
468out:
469	if (err && err != ENOENT) {
470		nvlist_free(*nvp);
471		*nvp = NULL;
472		return (err);
473	}
474
475	return (0);
476}
477
478/*
479 * Validate the given pool properties nvlist and modify the list
480 * for the property values to be set.
481 */
482static int
483spa_prop_validate(spa_t *spa, nvlist_t *props)
484{
485	nvpair_t *elem;
486	int error = 0, reset_bootfs = 0;
487	uint64_t objnum = 0;
488	boolean_t has_feature = B_FALSE;
489
490	elem = NULL;
491	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
492		uint64_t intval;
493		char *strval, *slash, *check, *fname;
494		const char *propname = nvpair_name(elem);
495		zpool_prop_t prop = zpool_name_to_prop(propname);
496
497		switch (prop) {
498		case ZPOOL_PROP_INVAL:
499			if (!zpool_prop_feature(propname)) {
500				error = SET_ERROR(EINVAL);
501				break;
502			}
503
504			/*
505			 * Sanitize the input.
506			 */
507			if (nvpair_type(elem) != DATA_TYPE_UINT64) {
508				error = SET_ERROR(EINVAL);
509				break;
510			}
511
512			if (nvpair_value_uint64(elem, &intval) != 0) {
513				error = SET_ERROR(EINVAL);
514				break;
515			}
516
517			if (intval != 0) {
518				error = SET_ERROR(EINVAL);
519				break;
520			}
521
522			fname = strchr(propname, '@') + 1;
523			if (zfeature_lookup_name(fname, NULL) != 0) {
524				error = SET_ERROR(EINVAL);
525				break;
526			}
527
528			has_feature = B_TRUE;
529			break;
530
531		case ZPOOL_PROP_VERSION:
532			error = nvpair_value_uint64(elem, &intval);
533			if (!error &&
534			    (intval < spa_version(spa) ||
535			    intval > SPA_VERSION_BEFORE_FEATURES ||
536			    has_feature))
537				error = SET_ERROR(EINVAL);
538			break;
539
540		case ZPOOL_PROP_DELEGATION:
541		case ZPOOL_PROP_AUTOREPLACE:
542		case ZPOOL_PROP_LISTSNAPS:
543		case ZPOOL_PROP_AUTOEXPAND:
544			error = nvpair_value_uint64(elem, &intval);
545			if (!error && intval > 1)
546				error = SET_ERROR(EINVAL);
547			break;
548
549		case ZPOOL_PROP_BOOTFS:
550			/*
551			 * If the pool version is less than SPA_VERSION_BOOTFS,
552			 * or the pool is still being created (version == 0),
553			 * the bootfs property cannot be set.
554			 */
555			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
556				error = SET_ERROR(ENOTSUP);
557				break;
558			}
559
560			/*
561			 * Make sure the vdev config is bootable
562			 */
563			if (!vdev_is_bootable(spa->spa_root_vdev)) {
564				error = SET_ERROR(ENOTSUP);
565				break;
566			}
567
568			reset_bootfs = 1;
569
570			error = nvpair_value_string(elem, &strval);
571
572			if (!error) {
573				objset_t *os;
574				uint64_t propval;
575
576				if (strval == NULL || strval[0] == '\0') {
577					objnum = zpool_prop_default_numeric(
578					    ZPOOL_PROP_BOOTFS);
579					break;
580				}
581
582				error = dmu_objset_hold(strval, FTAG, &os);
583				if (error != 0)
584					break;
585
586				/*
587				 * Must be ZPL, and its property settings
588				 * must be supported.
589				 */
590
591				if (dmu_objset_type(os) != DMU_OST_ZFS) {
592					error = SET_ERROR(ENOTSUP);
593				} else if ((error =
594				    dsl_prop_get_int_ds(dmu_objset_ds(os),
595				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
596				    &propval)) == 0 &&
597				    !BOOTFS_COMPRESS_VALID(propval)) {
598					error = SET_ERROR(ENOTSUP);
599				} else {
600					objnum = dmu_objset_id(os);
601				}
602				dmu_objset_rele(os, FTAG);
603			}
604			break;
605
606		case ZPOOL_PROP_FAILUREMODE:
607			error = nvpair_value_uint64(elem, &intval);
608			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
609			    intval > ZIO_FAILURE_MODE_PANIC))
610				error = SET_ERROR(EINVAL);
611
612			/*
613			 * This is a special case which only occurs when
614			 * the pool has completely failed. This allows
615			 * the user to change the in-core failmode property
616			 * without syncing it out to disk (I/Os might
617			 * currently be blocked). We do this by returning
618			 * EIO to the caller (spa_prop_set) to trick it
619			 * into thinking we encountered a property validation
620			 * error.
621			 */
622			if (!error && spa_suspended(spa)) {
623				spa->spa_failmode = intval;
624				error = SET_ERROR(EIO);
625			}
626			break;
627
628		case ZPOOL_PROP_CACHEFILE:
629			if ((error = nvpair_value_string(elem, &strval)) != 0)
630				break;
631
632			if (strval[0] == '\0')
633				break;
634
635			if (strcmp(strval, "none") == 0)
636				break;
637
638			if (strval[0] != '/') {
639				error = SET_ERROR(EINVAL);
640				break;
641			}
642
643			slash = strrchr(strval, '/');
644			ASSERT(slash != NULL);
645
646			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
647			    strcmp(slash, "/..") == 0)
648				error = SET_ERROR(EINVAL);
649			break;
650
651		case ZPOOL_PROP_COMMENT:
652			if ((error = nvpair_value_string(elem, &strval)) != 0)
653				break;
654			for (check = strval; *check != '\0'; check++) {
655				/*
656				 * The kernel doesn't have an easy isprint()
657				 * check.  For this kernel check, we merely
658				 * check ASCII apart from DEL.  Fix this if
659				 * there is an easy-to-use kernel isprint().
660				 */
661				if (*check >= 0x7f) {
662					error = SET_ERROR(EINVAL);
663					break;
664				}
665			}
666			if (strlen(strval) > ZPROP_MAX_COMMENT)
667				error = E2BIG;
668			break;
669
670		case ZPOOL_PROP_DEDUPDITTO:
671			if (spa_version(spa) < SPA_VERSION_DEDUP)
672				error = SET_ERROR(ENOTSUP);
673			else
674				error = nvpair_value_uint64(elem, &intval);
675			if (error == 0 &&
676			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
677				error = SET_ERROR(EINVAL);
678			break;
679		}
680
681		if (error)
682			break;
683	}
684
685	if (!error && reset_bootfs) {
686		error = nvlist_remove(props,
687		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
688
689		if (!error) {
690			error = nvlist_add_uint64(props,
691			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
692		}
693	}
694
695	return (error);
696}
697
698void
699spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
700{
701	char *cachefile;
702	spa_config_dirent_t *dp;
703
704	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
705	    &cachefile) != 0)
706		return;
707
708	dp = kmem_alloc(sizeof (spa_config_dirent_t),
709	    KM_SLEEP);
710
711	if (cachefile[0] == '\0')
712		dp->scd_path = spa_strdup(spa_config_path);
713	else if (strcmp(cachefile, "none") == 0)
714		dp->scd_path = NULL;
715	else
716		dp->scd_path = spa_strdup(cachefile);
717
718	list_insert_head(&spa->spa_config_list, dp);
719	if (need_sync)
720		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
721}
722
723int
724spa_prop_set(spa_t *spa, nvlist_t *nvp)
725{
726	int error;
727	nvpair_t *elem = NULL;
728	boolean_t need_sync = B_FALSE;
729
730	if ((error = spa_prop_validate(spa, nvp)) != 0)
731		return (error);
732
733	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
734		zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
735
736		if (prop == ZPOOL_PROP_CACHEFILE ||
737		    prop == ZPOOL_PROP_ALTROOT ||
738		    prop == ZPOOL_PROP_READONLY)
739			continue;
740
741		if (prop == ZPOOL_PROP_VERSION || prop == ZPOOL_PROP_INVAL) {
742			uint64_t ver;
743
744			if (prop == ZPOOL_PROP_VERSION) {
745				VERIFY(nvpair_value_uint64(elem, &ver) == 0);
746			} else {
747				ASSERT(zpool_prop_feature(nvpair_name(elem)));
748				ver = SPA_VERSION_FEATURES;
749				need_sync = B_TRUE;
750			}
751
752			/* Save time if the version is already set. */
753			if (ver == spa_version(spa))
754				continue;
755
756			/*
757			 * In addition to the pool directory object, we might
758			 * create the pool properties object, the features for
759			 * read object, the features for write object, or the
760			 * feature descriptions object.
761			 */
762			error = dsl_sync_task(spa->spa_name, NULL,
763			    spa_sync_version, &ver,
764			    6, ZFS_SPACE_CHECK_RESERVED);
765			if (error)
766				return (error);
767			continue;
768		}
769
770		need_sync = B_TRUE;
771		break;
772	}
773
774	if (need_sync) {
775		return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
776		    nvp, 6, ZFS_SPACE_CHECK_RESERVED));
777	}
778
779	return (0);
780}
781
782/*
783 * If the bootfs property value is dsobj, clear it.
784 */
785void
786spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
787{
788	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
789		VERIFY(zap_remove(spa->spa_meta_objset,
790		    spa->spa_pool_props_object,
791		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
792		spa->spa_bootfs = 0;
793	}
794}
795
796/*ARGSUSED*/
797static int
798spa_change_guid_check(void *arg, dmu_tx_t *tx)
799{
800	uint64_t *newguid = arg;
801	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
802	vdev_t *rvd = spa->spa_root_vdev;
803	uint64_t vdev_state;
804
805	if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
806		int error = (spa_has_checkpoint(spa)) ?
807		    ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
808		return (SET_ERROR(error));
809	}
810
811	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
812	vdev_state = rvd->vdev_state;
813	spa_config_exit(spa, SCL_STATE, FTAG);
814
815	if (vdev_state != VDEV_STATE_HEALTHY)
816		return (SET_ERROR(ENXIO));
817
818	ASSERT3U(spa_guid(spa), !=, *newguid);
819
820	return (0);
821}
822
823static void
824spa_change_guid_sync(void *arg, dmu_tx_t *tx)
825{
826	uint64_t *newguid = arg;
827	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
828	uint64_t oldguid;
829	vdev_t *rvd = spa->spa_root_vdev;
830
831	oldguid = spa_guid(spa);
832
833	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
834	rvd->vdev_guid = *newguid;
835	rvd->vdev_guid_sum += (*newguid - oldguid);
836	vdev_config_dirty(rvd);
837	spa_config_exit(spa, SCL_STATE, FTAG);
838
839	spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
840	    oldguid, *newguid);
841}
842
843/*
844 * Change the GUID for the pool.  This is done so that we can later
845 * re-import a pool built from a clone of our own vdevs.  We will modify
846 * the root vdev's guid, our own pool guid, and then mark all of our
847 * vdevs dirty.  Note that we must make sure that all our vdevs are
848 * online when we do this, or else any vdevs that weren't present
849 * would be orphaned from our pool.  We are also going to issue a
850 * sysevent to update any watchers.
851 */
852int
853spa_change_guid(spa_t *spa)
854{
855	int error;
856	uint64_t guid;
857
858	mutex_enter(&spa->spa_vdev_top_lock);
859	mutex_enter(&spa_namespace_lock);
860	guid = spa_generate_guid(NULL);
861
862	error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
863	    spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
864
865	if (error == 0) {
866		spa_write_cachefile(spa, B_FALSE, B_TRUE);
867		spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID);
868	}
869
870	mutex_exit(&spa_namespace_lock);
871	mutex_exit(&spa->spa_vdev_top_lock);
872
873	return (error);
874}
875
876/*
877 * ==========================================================================
878 * SPA state manipulation (open/create/destroy/import/export)
879 * ==========================================================================
880 */
881
882static int
883spa_error_entry_compare(const void *a, const void *b)
884{
885	spa_error_entry_t *sa = (spa_error_entry_t *)a;
886	spa_error_entry_t *sb = (spa_error_entry_t *)b;
887	int ret;
888
889	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
890	    sizeof (zbookmark_phys_t));
891
892	if (ret < 0)
893		return (-1);
894	else if (ret > 0)
895		return (1);
896	else
897		return (0);
898}
899
900/*
901 * Utility function which retrieves copies of the current logs and
902 * re-initializes them in the process.
903 */
904void
905spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
906{
907	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
908
909	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
910	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
911
912	avl_create(&spa->spa_errlist_scrub,
913	    spa_error_entry_compare, sizeof (spa_error_entry_t),
914	    offsetof(spa_error_entry_t, se_avl));
915	avl_create(&spa->spa_errlist_last,
916	    spa_error_entry_compare, sizeof (spa_error_entry_t),
917	    offsetof(spa_error_entry_t, se_avl));
918}
919
920static void
921spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
922{
923	const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
924	enum zti_modes mode = ztip->zti_mode;
925	uint_t value = ztip->zti_value;
926	uint_t count = ztip->zti_count;
927	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
928	char name[32];
929	uint_t flags = 0;
930	boolean_t batch = B_FALSE;
931
932	if (mode == ZTI_MODE_NULL) {
933		tqs->stqs_count = 0;
934		tqs->stqs_taskq = NULL;
935		return;
936	}
937
938	ASSERT3U(count, >, 0);
939
940	tqs->stqs_count = count;
941	tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
942
943	switch (mode) {
944	case ZTI_MODE_FIXED:
945		ASSERT3U(value, >=, 1);
946		value = MAX(value, 1);
947		break;
948
949	case ZTI_MODE_BATCH:
950		batch = B_TRUE;
951		flags |= TASKQ_THREADS_CPU_PCT;
952		value = zio_taskq_batch_pct;
953		break;
954
955	default:
956		panic("unrecognized mode for %s_%s taskq (%u:%u) in "
957		    "spa_activate()",
958		    zio_type_name[t], zio_taskq_types[q], mode, value);
959		break;
960	}
961
962	for (uint_t i = 0; i < count; i++) {
963		taskq_t *tq;
964
965		if (count > 1) {
966			(void) snprintf(name, sizeof (name), "%s_%s_%u",
967			    zio_type_name[t], zio_taskq_types[q], i);
968		} else {
969			(void) snprintf(name, sizeof (name), "%s_%s",
970			    zio_type_name[t], zio_taskq_types[q]);
971		}
972
973		if (zio_taskq_sysdc && spa->spa_proc != &p0) {
974			if (batch)
975				flags |= TASKQ_DC_BATCH;
976
977			tq = taskq_create_sysdc(name, value, 50, INT_MAX,
978			    spa->spa_proc, zio_taskq_basedc, flags);
979		} else {
980			pri_t pri = maxclsyspri;
981			/*
982			 * The write issue taskq can be extremely CPU
983			 * intensive.  Run it at slightly lower priority
984			 * than the other taskqs.
985			 */
986			if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
987				pri--;
988
989			tq = taskq_create_proc(name, value, pri, 50,
990			    INT_MAX, spa->spa_proc, flags);
991		}
992
993		tqs->stqs_taskq[i] = tq;
994	}
995}
996
997static void
998spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
999{
1000	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1001
1002	if (tqs->stqs_taskq == NULL) {
1003		ASSERT0(tqs->stqs_count);
1004		return;
1005	}
1006
1007	for (uint_t i = 0; i < tqs->stqs_count; i++) {
1008		ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
1009		taskq_destroy(tqs->stqs_taskq[i]);
1010	}
1011
1012	kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
1013	tqs->stqs_taskq = NULL;
1014}
1015
1016/*
1017 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
1018 * Note that a type may have multiple discrete taskqs to avoid lock contention
1019 * on the taskq itself. In that case we choose which taskq at random by using
1020 * the low bits of gethrtime().
1021 */
1022void
1023spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
1024    task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
1025{
1026	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1027	taskq_t *tq;
1028
1029	ASSERT3P(tqs->stqs_taskq, !=, NULL);
1030	ASSERT3U(tqs->stqs_count, !=, 0);
1031
1032	if (tqs->stqs_count == 1) {
1033		tq = tqs->stqs_taskq[0];
1034	} else {
1035		tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
1036	}
1037
1038	taskq_dispatch_ent(tq, func, arg, flags, ent);
1039}
1040
1041static void
1042spa_create_zio_taskqs(spa_t *spa)
1043{
1044	for (int t = 0; t < ZIO_TYPES; t++) {
1045		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1046			spa_taskqs_init(spa, t, q);
1047		}
1048	}
1049}
1050
1051#ifdef _KERNEL
1052static void
1053spa_thread(void *arg)
1054{
1055	callb_cpr_t cprinfo;
1056
1057	spa_t *spa = arg;
1058	user_t *pu = PTOU(curproc);
1059
1060	CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
1061	    spa->spa_name);
1062
1063	ASSERT(curproc != &p0);
1064	(void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
1065	    "zpool-%s", spa->spa_name);
1066	(void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
1067
1068	/* bind this thread to the requested psrset */
1069	if (zio_taskq_psrset_bind != PS_NONE) {
1070		pool_lock();
1071		mutex_enter(&cpu_lock);
1072		mutex_enter(&pidlock);
1073		mutex_enter(&curproc->p_lock);
1074
1075		if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1076		    0, NULL, NULL) == 0)  {
1077			curthread->t_bind_pset = zio_taskq_psrset_bind;
1078		} else {
1079			cmn_err(CE_WARN,
1080			    "Couldn't bind process for zfs pool \"%s\" to "
1081			    "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1082		}
1083
1084		mutex_exit(&curproc->p_lock);
1085		mutex_exit(&pidlock);
1086		mutex_exit(&cpu_lock);
1087		pool_unlock();
1088	}
1089
1090	if (zio_taskq_sysdc) {
1091		sysdc_thread_enter(curthread, 100, 0);
1092	}
1093
1094	spa->spa_proc = curproc;
1095	spa->spa_did = curthread->t_did;
1096
1097	spa_create_zio_taskqs(spa);
1098
1099	mutex_enter(&spa->spa_proc_lock);
1100	ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1101
1102	spa->spa_proc_state = SPA_PROC_ACTIVE;
1103	cv_broadcast(&spa->spa_proc_cv);
1104
1105	CALLB_CPR_SAFE_BEGIN(&cprinfo);
1106	while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1107		cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1108	CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1109
1110	ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1111	spa->spa_proc_state = SPA_PROC_GONE;
1112	spa->spa_proc = &p0;
1113	cv_broadcast(&spa->spa_proc_cv);
1114	CALLB_CPR_EXIT(&cprinfo);	/* drops spa_proc_lock */
1115
1116	mutex_enter(&curproc->p_lock);
1117	lwp_exit();
1118}
1119#endif
1120
1121/*
1122 * Activate an uninitialized pool.
1123 */
1124static void
1125spa_activate(spa_t *spa, int mode)
1126{
1127	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1128
1129	spa->spa_state = POOL_STATE_ACTIVE;
1130	spa->spa_mode = mode;
1131
1132	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1133	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1134
1135	/* Try to create a covering process */
1136	mutex_enter(&spa->spa_proc_lock);
1137	ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1138	ASSERT(spa->spa_proc == &p0);
1139	spa->spa_did = 0;
1140
1141	/* Only create a process if we're going to be around a while. */
1142	if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1143		if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1144		    NULL, 0) == 0) {
1145			spa->spa_proc_state = SPA_PROC_CREATED;
1146			while (spa->spa_proc_state == SPA_PROC_CREATED) {
1147				cv_wait(&spa->spa_proc_cv,
1148				    &spa->spa_proc_lock);
1149			}
1150			ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1151			ASSERT(spa->spa_proc != &p0);
1152			ASSERT(spa->spa_did != 0);
1153		} else {
1154#ifdef _KERNEL
1155			cmn_err(CE_WARN,
1156			    "Couldn't create process for zfs pool \"%s\"\n",
1157			    spa->spa_name);
1158#endif
1159		}
1160	}
1161	mutex_exit(&spa->spa_proc_lock);
1162
1163	/* If we didn't create a process, we need to create our taskqs. */
1164	if (spa->spa_proc == &p0) {
1165		spa_create_zio_taskqs(spa);
1166	}
1167
1168	for (size_t i = 0; i < TXG_SIZE; i++) {
1169		spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL,
1170		    ZIO_FLAG_CANFAIL);
1171	}
1172
1173	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1174	    offsetof(vdev_t, vdev_config_dirty_node));
1175	list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1176	    offsetof(objset_t, os_evicting_node));
1177	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1178	    offsetof(vdev_t, vdev_state_dirty_node));
1179
1180	txg_list_create(&spa->spa_vdev_txg_list, spa,
1181	    offsetof(struct vdev, vdev_txg_node));
1182
1183	avl_create(&spa->spa_errlist_scrub,
1184	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1185	    offsetof(spa_error_entry_t, se_avl));
1186	avl_create(&spa->spa_errlist_last,
1187	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1188	    offsetof(spa_error_entry_t, se_avl));
1189}
1190
1191/*
1192 * Opposite of spa_activate().
1193 */
1194static void
1195spa_deactivate(spa_t *spa)
1196{
1197	ASSERT(spa->spa_sync_on == B_FALSE);
1198	ASSERT(spa->spa_dsl_pool == NULL);
1199	ASSERT(spa->spa_root_vdev == NULL);
1200	ASSERT(spa->spa_async_zio_root == NULL);
1201	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1202
1203	spa_evicting_os_wait(spa);
1204
1205	txg_list_destroy(&spa->spa_vdev_txg_list);
1206
1207	list_destroy(&spa->spa_config_dirty_list);
1208	list_destroy(&spa->spa_evicting_os_list);
1209	list_destroy(&spa->spa_state_dirty_list);
1210
1211	for (int t = 0; t < ZIO_TYPES; t++) {
1212		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1213			spa_taskqs_fini(spa, t, q);
1214		}
1215	}
1216
1217	for (size_t i = 0; i < TXG_SIZE; i++) {
1218		ASSERT3P(spa->spa_txg_zio[i], !=, NULL);
1219		VERIFY0(zio_wait(spa->spa_txg_zio[i]));
1220		spa->spa_txg_zio[i] = NULL;
1221	}
1222
1223	metaslab_class_destroy(spa->spa_normal_class);
1224	spa->spa_normal_class = NULL;
1225
1226	metaslab_class_destroy(spa->spa_log_class);
1227	spa->spa_log_class = NULL;
1228
1229	/*
1230	 * If this was part of an import or the open otherwise failed, we may
1231	 * still have errors left in the queues.  Empty them just in case.
1232	 */
1233	spa_errlog_drain(spa);
1234
1235	avl_destroy(&spa->spa_errlist_scrub);
1236	avl_destroy(&spa->spa_errlist_last);
1237
1238	spa->spa_state = POOL_STATE_UNINITIALIZED;
1239
1240	mutex_enter(&spa->spa_proc_lock);
1241	if (spa->spa_proc_state != SPA_PROC_NONE) {
1242		ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1243		spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1244		cv_broadcast(&spa->spa_proc_cv);
1245		while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1246			ASSERT(spa->spa_proc != &p0);
1247			cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1248		}
1249		ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1250		spa->spa_proc_state = SPA_PROC_NONE;
1251	}
1252	ASSERT(spa->spa_proc == &p0);
1253	mutex_exit(&spa->spa_proc_lock);
1254
1255	/*
1256	 * We want to make sure spa_thread() has actually exited the ZFS
1257	 * module, so that the module can't be unloaded out from underneath
1258	 * it.
1259	 */
1260	if (spa->spa_did != 0) {
1261		thread_join(spa->spa_did);
1262		spa->spa_did = 0;
1263	}
1264}
1265
1266/*
1267 * Verify a pool configuration, and construct the vdev tree appropriately.  This
1268 * will create all the necessary vdevs in the appropriate layout, with each vdev
1269 * in the CLOSED state.  This will prep the pool before open/creation/import.
1270 * All vdev validation is done by the vdev_alloc() routine.
1271 */
1272static int
1273spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1274    uint_t id, int atype)
1275{
1276	nvlist_t **child;
1277	uint_t children;
1278	int error;
1279
1280	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1281		return (error);
1282
1283	if ((*vdp)->vdev_ops->vdev_op_leaf)
1284		return (0);
1285
1286	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1287	    &child, &children);
1288
1289	if (error == ENOENT)
1290		return (0);
1291
1292	if (error) {
1293		vdev_free(*vdp);
1294		*vdp = NULL;
1295		return (SET_ERROR(EINVAL));
1296	}
1297
1298	for (int c = 0; c < children; c++) {
1299		vdev_t *vd;
1300		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1301		    atype)) != 0) {
1302			vdev_free(*vdp);
1303			*vdp = NULL;
1304			return (error);
1305		}
1306	}
1307
1308	ASSERT(*vdp != NULL);
1309
1310	return (0);
1311}
1312
1313/*
1314 * Opposite of spa_load().
1315 */
1316static void
1317spa_unload(spa_t *spa)
1318{
1319	int i;
1320
1321	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1322
1323	spa_load_note(spa, "UNLOADING");
1324
1325	/*
1326	 * Stop async tasks.
1327	 */
1328	spa_async_suspend(spa);
1329
1330	if (spa->spa_root_vdev) {
1331		vdev_initialize_stop_all(spa->spa_root_vdev,
1332		    VDEV_INITIALIZE_ACTIVE);
1333	}
1334
1335	/*
1336	 * Stop syncing.
1337	 */
1338	if (spa->spa_sync_on) {
1339		txg_sync_stop(spa->spa_dsl_pool);
1340		spa->spa_sync_on = B_FALSE;
1341	}
1342
1343	/*
1344	 * Even though vdev_free() also calls vdev_metaslab_fini, we need
1345	 * to call it earlier, before we wait for async i/o to complete.
1346	 * This ensures that there is no async metaslab prefetching, by
1347	 * calling taskq_wait(mg_taskq).
1348	 */
1349	if (spa->spa_root_vdev != NULL) {
1350		spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1351		for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++)
1352			vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]);
1353		spa_config_exit(spa, SCL_ALL, spa);
1354	}
1355
1356	/*
1357	 * Wait for any outstanding async I/O to complete.
1358	 */
1359	if (spa->spa_async_zio_root != NULL) {
1360		for (int i = 0; i < max_ncpus; i++)
1361			(void) zio_wait(spa->spa_async_zio_root[i]);
1362		kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1363		spa->spa_async_zio_root = NULL;
1364	}
1365
1366	if (spa->spa_vdev_removal != NULL) {
1367		spa_vdev_removal_destroy(spa->spa_vdev_removal);
1368		spa->spa_vdev_removal = NULL;
1369	}
1370
1371	if (spa->spa_condense_zthr != NULL) {
1372		ASSERT(!zthr_isrunning(spa->spa_condense_zthr));
1373		zthr_destroy(spa->spa_condense_zthr);
1374		spa->spa_condense_zthr = NULL;
1375	}
1376
1377	if (spa->spa_checkpoint_discard_zthr != NULL) {
1378		ASSERT(!zthr_isrunning(spa->spa_checkpoint_discard_zthr));
1379		zthr_destroy(spa->spa_checkpoint_discard_zthr);
1380		spa->spa_checkpoint_discard_zthr = NULL;
1381	}
1382
1383	spa_condense_fini(spa);
1384
1385	bpobj_close(&spa->spa_deferred_bpobj);
1386
1387	spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1388
1389	/*
1390	 * Close all vdevs.
1391	 */
1392	if (spa->spa_root_vdev)
1393		vdev_free(spa->spa_root_vdev);
1394	ASSERT(spa->spa_root_vdev == NULL);
1395
1396	/*
1397	 * Close the dsl pool.
1398	 */
1399	if (spa->spa_dsl_pool) {
1400		dsl_pool_close(spa->spa_dsl_pool);
1401		spa->spa_dsl_pool = NULL;
1402		spa->spa_meta_objset = NULL;
1403	}
1404
1405	ddt_unload(spa);
1406
1407	/*
1408	 * Drop and purge level 2 cache
1409	 */
1410	spa_l2cache_drop(spa);
1411
1412	for (i = 0; i < spa->spa_spares.sav_count; i++)
1413		vdev_free(spa->spa_spares.sav_vdevs[i]);
1414	if (spa->spa_spares.sav_vdevs) {
1415		kmem_free(spa->spa_spares.sav_vdevs,
1416		    spa->spa_spares.sav_count * sizeof (void *));
1417		spa->spa_spares.sav_vdevs = NULL;
1418	}
1419	if (spa->spa_spares.sav_config) {
1420		nvlist_free(spa->spa_spares.sav_config);
1421		spa->spa_spares.sav_config = NULL;
1422	}
1423	spa->spa_spares.sav_count = 0;
1424
1425	for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1426		vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1427		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1428	}
1429	if (spa->spa_l2cache.sav_vdevs) {
1430		kmem_free(spa->spa_l2cache.sav_vdevs,
1431		    spa->spa_l2cache.sav_count * sizeof (void *));
1432		spa->spa_l2cache.sav_vdevs = NULL;
1433	}
1434	if (spa->spa_l2cache.sav_config) {
1435		nvlist_free(spa->spa_l2cache.sav_config);
1436		spa->spa_l2cache.sav_config = NULL;
1437	}
1438	spa->spa_l2cache.sav_count = 0;
1439
1440	spa->spa_async_suspended = 0;
1441
1442	spa->spa_indirect_vdevs_loaded = B_FALSE;
1443
1444	if (spa->spa_comment != NULL) {
1445		spa_strfree(spa->spa_comment);
1446		spa->spa_comment = NULL;
1447	}
1448
1449	spa_config_exit(spa, SCL_ALL, spa);
1450}
1451
1452/*
1453 * Load (or re-load) the current list of vdevs describing the active spares for
1454 * this pool.  When this is called, we have some form of basic information in
1455 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1456 * then re-generate a more complete list including status information.
1457 */
1458void
1459spa_load_spares(spa_t *spa)
1460{
1461	nvlist_t **spares;
1462	uint_t nspares;
1463	int i;
1464	vdev_t *vd, *tvd;
1465
1466#ifndef _KERNEL
1467	/*
1468	 * zdb opens both the current state of the pool and the
1469	 * checkpointed state (if present), with a different spa_t.
1470	 *
1471	 * As spare vdevs are shared among open pools, we skip loading
1472	 * them when we load the checkpointed state of the pool.
1473	 */
1474	if (!spa_writeable(spa))
1475		return;
1476#endif
1477
1478	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1479
1480	/*
1481	 * First, close and free any existing spare vdevs.
1482	 */
1483	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1484		vd = spa->spa_spares.sav_vdevs[i];
1485
1486		/* Undo the call to spa_activate() below */
1487		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1488		    B_FALSE)) != NULL && tvd->vdev_isspare)
1489			spa_spare_remove(tvd);
1490		vdev_close(vd);
1491		vdev_free(vd);
1492	}
1493
1494	if (spa->spa_spares.sav_vdevs)
1495		kmem_free(spa->spa_spares.sav_vdevs,
1496		    spa->spa_spares.sav_count * sizeof (void *));
1497
1498	if (spa->spa_spares.sav_config == NULL)
1499		nspares = 0;
1500	else
1501		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1502		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1503
1504	spa->spa_spares.sav_count = (int)nspares;
1505	spa->spa_spares.sav_vdevs = NULL;
1506
1507	if (nspares == 0)
1508		return;
1509
1510	/*
1511	 * Construct the array of vdevs, opening them to get status in the
1512	 * process.   For each spare, there is potentially two different vdev_t
1513	 * structures associated with it: one in the list of spares (used only
1514	 * for basic validation purposes) and one in the active vdev
1515	 * configuration (if it's spared in).  During this phase we open and
1516	 * validate each vdev on the spare list.  If the vdev also exists in the
1517	 * active configuration, then we also mark this vdev as an active spare.
1518	 */
1519	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1520	    KM_SLEEP);
1521	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1522		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1523		    VDEV_ALLOC_SPARE) == 0);
1524		ASSERT(vd != NULL);
1525
1526		spa->spa_spares.sav_vdevs[i] = vd;
1527
1528		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1529		    B_FALSE)) != NULL) {
1530			if (!tvd->vdev_isspare)
1531				spa_spare_add(tvd);
1532
1533			/*
1534			 * We only mark the spare active if we were successfully
1535			 * able to load the vdev.  Otherwise, importing a pool
1536			 * with a bad active spare would result in strange
1537			 * behavior, because multiple pool would think the spare
1538			 * is actively in use.
1539			 *
1540			 * There is a vulnerability here to an equally bizarre
1541			 * circumstance, where a dead active spare is later
1542			 * brought back to life (onlined or otherwise).  Given
1543			 * the rarity of this scenario, and the extra complexity
1544			 * it adds, we ignore the possibility.
1545			 */
1546			if (!vdev_is_dead(tvd))
1547				spa_spare_activate(tvd);
1548		}
1549
1550		vd->vdev_top = vd;
1551		vd->vdev_aux = &spa->spa_spares;
1552
1553		if (vdev_open(vd) != 0)
1554			continue;
1555
1556		if (vdev_validate_aux(vd) == 0)
1557			spa_spare_add(vd);
1558	}
1559
1560	/*
1561	 * Recompute the stashed list of spares, with status information
1562	 * this time.
1563	 */
1564	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1565	    DATA_TYPE_NVLIST_ARRAY) == 0);
1566
1567	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1568	    KM_SLEEP);
1569	for (i = 0; i < spa->spa_spares.sav_count; i++)
1570		spares[i] = vdev_config_generate(spa,
1571		    spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1572	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1573	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1574	for (i = 0; i < spa->spa_spares.sav_count; i++)
1575		nvlist_free(spares[i]);
1576	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1577}
1578
1579/*
1580 * Load (or re-load) the current list of vdevs describing the active l2cache for
1581 * this pool.  When this is called, we have some form of basic information in
1582 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1583 * then re-generate a more complete list including status information.
1584 * Devices which are already active have their details maintained, and are
1585 * not re-opened.
1586 */
1587void
1588spa_load_l2cache(spa_t *spa)
1589{
1590	nvlist_t **l2cache;
1591	uint_t nl2cache;
1592	int i, j, oldnvdevs;
1593	uint64_t guid;
1594	vdev_t *vd, **oldvdevs, **newvdevs;
1595	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1596
1597#ifndef _KERNEL
1598	/*
1599	 * zdb opens both the current state of the pool and the
1600	 * checkpointed state (if present), with a different spa_t.
1601	 *
1602	 * As L2 caches are part of the ARC which is shared among open
1603	 * pools, we skip loading them when we load the checkpointed
1604	 * state of the pool.
1605	 */
1606	if (!spa_writeable(spa))
1607		return;
1608#endif
1609
1610	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1611
1612	if (sav->sav_config != NULL) {
1613		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1614		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1615		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1616	} else {
1617		nl2cache = 0;
1618		newvdevs = NULL;
1619	}
1620
1621	oldvdevs = sav->sav_vdevs;
1622	oldnvdevs = sav->sav_count;
1623	sav->sav_vdevs = NULL;
1624	sav->sav_count = 0;
1625
1626	/*
1627	 * Process new nvlist of vdevs.
1628	 */
1629	for (i = 0; i < nl2cache; i++) {
1630		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1631		    &guid) == 0);
1632
1633		newvdevs[i] = NULL;
1634		for (j = 0; j < oldnvdevs; j++) {
1635			vd = oldvdevs[j];
1636			if (vd != NULL && guid == vd->vdev_guid) {
1637				/*
1638				 * Retain previous vdev for add/remove ops.
1639				 */
1640				newvdevs[i] = vd;
1641				oldvdevs[j] = NULL;
1642				break;
1643			}
1644		}
1645
1646		if (newvdevs[i] == NULL) {
1647			/*
1648			 * Create new vdev
1649			 */
1650			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1651			    VDEV_ALLOC_L2CACHE) == 0);
1652			ASSERT(vd != NULL);
1653			newvdevs[i] = vd;
1654
1655			/*
1656			 * Commit this vdev as an l2cache device,
1657			 * even if it fails to open.
1658			 */
1659			spa_l2cache_add(vd);
1660
1661			vd->vdev_top = vd;
1662			vd->vdev_aux = sav;
1663
1664			spa_l2cache_activate(vd);
1665
1666			if (vdev_open(vd) != 0)
1667				continue;
1668
1669			(void) vdev_validate_aux(vd);
1670
1671			if (!vdev_is_dead(vd))
1672				l2arc_add_vdev(spa, vd);
1673		}
1674	}
1675
1676	/*
1677	 * Purge vdevs that were dropped
1678	 */
1679	for (i = 0; i < oldnvdevs; i++) {
1680		uint64_t pool;
1681
1682		vd = oldvdevs[i];
1683		if (vd != NULL) {
1684			ASSERT(vd->vdev_isl2cache);
1685
1686			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1687			    pool != 0ULL && l2arc_vdev_present(vd))
1688				l2arc_remove_vdev(vd);
1689			vdev_clear_stats(vd);
1690			vdev_free(vd);
1691		}
1692	}
1693
1694	if (oldvdevs)
1695		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1696
1697	if (sav->sav_config == NULL)
1698		goto out;
1699
1700	sav->sav_vdevs = newvdevs;
1701	sav->sav_count = (int)nl2cache;
1702
1703	/*
1704	 * Recompute the stashed list of l2cache devices, with status
1705	 * information this time.
1706	 */
1707	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1708	    DATA_TYPE_NVLIST_ARRAY) == 0);
1709
1710	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1711	for (i = 0; i < sav->sav_count; i++)
1712		l2cache[i] = vdev_config_generate(spa,
1713		    sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1714	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1715	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1716out:
1717	for (i = 0; i < sav->sav_count; i++)
1718		nvlist_free(l2cache[i]);
1719	if (sav->sav_count)
1720		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1721}
1722
1723static int
1724load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1725{
1726	dmu_buf_t *db;
1727	char *packed = NULL;
1728	size_t nvsize = 0;
1729	int error;
1730	*value = NULL;
1731
1732	error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1733	if (error != 0)
1734		return (error);
1735
1736	nvsize = *(uint64_t *)db->db_data;
1737	dmu_buf_rele(db, FTAG);
1738
1739	packed = kmem_alloc(nvsize, KM_SLEEP);
1740	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1741	    DMU_READ_PREFETCH);
1742	if (error == 0)
1743		error = nvlist_unpack(packed, nvsize, value, 0);
1744	kmem_free(packed, nvsize);
1745
1746	return (error);
1747}
1748
1749/*
1750 * Concrete top-level vdevs that are not missing and are not logs. At every
1751 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds.
1752 */
1753static uint64_t
1754spa_healthy_core_tvds(spa_t *spa)
1755{
1756	vdev_t *rvd = spa->spa_root_vdev;
1757	uint64_t tvds = 0;
1758
1759	for (uint64_t i = 0; i < rvd->vdev_children; i++) {
1760		vdev_t *vd = rvd->vdev_child[i];
1761		if (vd->vdev_islog)
1762			continue;
1763		if (vdev_is_concrete(vd) && !vdev_is_dead(vd))
1764			tvds++;
1765	}
1766
1767	return (tvds);
1768}
1769
1770/*
1771 * Checks to see if the given vdev could not be opened, in which case we post a
1772 * sysevent to notify the autoreplace code that the device has been removed.
1773 */
1774static void
1775spa_check_removed(vdev_t *vd)
1776{
1777	for (uint64_t c = 0; c < vd->vdev_children; c++)
1778		spa_check_removed(vd->vdev_child[c]);
1779
1780	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1781	    vdev_is_concrete(vd)) {
1782		zfs_post_autoreplace(vd->vdev_spa, vd);
1783		spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK);
1784	}
1785}
1786
1787static int
1788spa_check_for_missing_logs(spa_t *spa)
1789{
1790	vdev_t *rvd = spa->spa_root_vdev;
1791
1792	/*
1793	 * If we're doing a normal import, then build up any additional
1794	 * diagnostic information about missing log devices.
1795	 * We'll pass this up to the user for further processing.
1796	 */
1797	if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1798		nvlist_t **child, *nv;
1799		uint64_t idx = 0;
1800
1801		child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1802		    KM_SLEEP);
1803		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1804
1805		for (uint64_t c = 0; c < rvd->vdev_children; c++) {
1806			vdev_t *tvd = rvd->vdev_child[c];
1807
1808			/*
1809			 * We consider a device as missing only if it failed
1810			 * to open (i.e. offline or faulted is not considered
1811			 * as missing).
1812			 */
1813			if (tvd->vdev_islog &&
1814			    tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
1815				child[idx++] = vdev_config_generate(spa, tvd,
1816				    B_FALSE, VDEV_CONFIG_MISSING);
1817			}
1818		}
1819
1820		if (idx > 0) {
1821			fnvlist_add_nvlist_array(nv,
1822			    ZPOOL_CONFIG_CHILDREN, child, idx);
1823			fnvlist_add_nvlist(spa->spa_load_info,
1824			    ZPOOL_CONFIG_MISSING_DEVICES, nv);
1825
1826			for (uint64_t i = 0; i < idx; i++)
1827				nvlist_free(child[i]);
1828		}
1829		nvlist_free(nv);
1830		kmem_free(child, rvd->vdev_children * sizeof (char **));
1831
1832		if (idx > 0) {
1833			spa_load_failed(spa, "some log devices are missing");
1834			vdev_dbgmsg_print_tree(rvd, 2);
1835			return (SET_ERROR(ENXIO));
1836		}
1837	} else {
1838		for (uint64_t c = 0; c < rvd->vdev_children; c++) {
1839			vdev_t *tvd = rvd->vdev_child[c];
1840
1841			if (tvd->vdev_islog &&
1842			    tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
1843				spa_set_log_state(spa, SPA_LOG_CLEAR);
1844				spa_load_note(spa, "some log devices are "
1845				    "missing, ZIL is dropped.");
1846				vdev_dbgmsg_print_tree(rvd, 2);
1847				break;
1848			}
1849		}
1850	}
1851
1852	return (0);
1853}
1854
1855/*
1856 * Check for missing log devices
1857 */
1858static boolean_t
1859spa_check_logs(spa_t *spa)
1860{
1861	boolean_t rv = B_FALSE;
1862	dsl_pool_t *dp = spa_get_dsl(spa);
1863
1864	switch (spa->spa_log_state) {
1865	case SPA_LOG_MISSING:
1866		/* need to recheck in case slog has been restored */
1867	case SPA_LOG_UNKNOWN:
1868		rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1869		    zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
1870		if (rv)
1871			spa_set_log_state(spa, SPA_LOG_MISSING);
1872		break;
1873	}
1874	return (rv);
1875}
1876
1877static boolean_t
1878spa_passivate_log(spa_t *spa)
1879{
1880	vdev_t *rvd = spa->spa_root_vdev;
1881	boolean_t slog_found = B_FALSE;
1882
1883	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1884
1885	if (!spa_has_slogs(spa))
1886		return (B_FALSE);
1887
1888	for (int c = 0; c < rvd->vdev_children; c++) {
1889		vdev_t *tvd = rvd->vdev_child[c];
1890		metaslab_group_t *mg = tvd->vdev_mg;
1891
1892		if (tvd->vdev_islog) {
1893			metaslab_group_passivate(mg);
1894			slog_found = B_TRUE;
1895		}
1896	}
1897
1898	return (slog_found);
1899}
1900
1901static void
1902spa_activate_log(spa_t *spa)
1903{
1904	vdev_t *rvd = spa->spa_root_vdev;
1905
1906	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1907
1908	for (int c = 0; c < rvd->vdev_children; c++) {
1909		vdev_t *tvd = rvd->vdev_child[c];
1910		metaslab_group_t *mg = tvd->vdev_mg;
1911
1912		if (tvd->vdev_islog)
1913			metaslab_group_activate(mg);
1914	}
1915}
1916
1917int
1918spa_reset_logs(spa_t *spa)
1919{
1920	int error;
1921
1922	error = dmu_objset_find(spa_name(spa), zil_reset,
1923	    NULL, DS_FIND_CHILDREN);
1924	if (error == 0) {
1925		/*
1926		 * We successfully offlined the log device, sync out the
1927		 * current txg so that the "stubby" block can be removed
1928		 * by zil_sync().
1929		 */
1930		txg_wait_synced(spa->spa_dsl_pool, 0);
1931	}
1932	return (error);
1933}
1934
1935static void
1936spa_aux_check_removed(spa_aux_vdev_t *sav)
1937{
1938	for (int i = 0; i < sav->sav_count; i++)
1939		spa_check_removed(sav->sav_vdevs[i]);
1940}
1941
1942void
1943spa_claim_notify(zio_t *zio)
1944{
1945	spa_t *spa = zio->io_spa;
1946
1947	if (zio->io_error)
1948		return;
1949
1950	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1951	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1952		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1953	mutex_exit(&spa->spa_props_lock);
1954}
1955
1956typedef struct spa_load_error {
1957	uint64_t	sle_meta_count;
1958	uint64_t	sle_data_count;
1959} spa_load_error_t;
1960
1961static void
1962spa_load_verify_done(zio_t *zio)
1963{
1964	blkptr_t *bp = zio->io_bp;
1965	spa_load_error_t *sle = zio->io_private;
1966	dmu_object_type_t type = BP_GET_TYPE(bp);
1967	int error = zio->io_error;
1968	spa_t *spa = zio->io_spa;
1969
1970	abd_free(zio->io_abd);
1971	if (error) {
1972		if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1973		    type != DMU_OT_INTENT_LOG)
1974			atomic_inc_64(&sle->sle_meta_count);
1975		else
1976			atomic_inc_64(&sle->sle_data_count);
1977	}
1978
1979	mutex_enter(&spa->spa_scrub_lock);
1980	spa->spa_scrub_inflight--;
1981	cv_broadcast(&spa->spa_scrub_io_cv);
1982	mutex_exit(&spa->spa_scrub_lock);
1983}
1984
1985/*
1986 * Maximum number of concurrent scrub i/os to create while verifying
1987 * a pool while importing it.
1988 */
1989int spa_load_verify_maxinflight = 10000;
1990boolean_t spa_load_verify_metadata = B_TRUE;
1991boolean_t spa_load_verify_data = B_TRUE;
1992
1993/*ARGSUSED*/
1994static int
1995spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1996    const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1997{
1998	if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
1999		return (0);
2000	/*
2001	 * Note: normally this routine will not be called if
2002	 * spa_load_verify_metadata is not set.  However, it may be useful
2003	 * to manually set the flag after the traversal has begun.
2004	 */
2005	if (!spa_load_verify_metadata)
2006		return (0);
2007	if (!BP_IS_METADATA(bp) && !spa_load_verify_data)
2008		return (0);
2009
2010	zio_t *rio = arg;
2011	size_t size = BP_GET_PSIZE(bp);
2012
2013	mutex_enter(&spa->spa_scrub_lock);
2014	while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
2015		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2016	spa->spa_scrub_inflight++;
2017	mutex_exit(&spa->spa_scrub_lock);
2018
2019	zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size,
2020	    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
2021	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
2022	    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
2023	return (0);
2024}
2025
2026/* ARGSUSED */
2027int
2028verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
2029{
2030	if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN)
2031		return (SET_ERROR(ENAMETOOLONG));
2032
2033	return (0);
2034}
2035
2036static int
2037spa_load_verify(spa_t *spa)
2038{
2039	zio_t *rio;
2040	spa_load_error_t sle = { 0 };
2041	zpool_load_policy_t policy;
2042	boolean_t verify_ok = B_FALSE;
2043	int error = 0;
2044
2045	zpool_get_load_policy(spa->spa_config, &policy);
2046
2047	if (policy.zlp_rewind & ZPOOL_NEVER_REWIND)
2048		return (0);
2049
2050	dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
2051	error = dmu_objset_find_dp(spa->spa_dsl_pool,
2052	    spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL,
2053	    DS_FIND_CHILDREN);
2054	dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
2055	if (error != 0)
2056		return (error);
2057
2058	rio = zio_root(spa, NULL, &sle,
2059	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
2060
2061	if (spa_load_verify_metadata) {
2062		if (spa->spa_extreme_rewind) {
2063			spa_load_note(spa, "performing a complete scan of the "
2064			    "pool since extreme rewind is on. This may take "
2065			    "a very long time.\n  (spa_load_verify_data=%u, "
2066			    "spa_load_verify_metadata=%u)",
2067			    spa_load_verify_data, spa_load_verify_metadata);
2068		}
2069		error = traverse_pool(spa, spa->spa_verify_min_txg,
2070		    TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
2071		    spa_load_verify_cb, rio);
2072	}
2073
2074	(void) zio_wait(rio);
2075
2076	spa->spa_load_meta_errors = sle.sle_meta_count;
2077	spa->spa_load_data_errors = sle.sle_data_count;
2078
2079	if (sle.sle_meta_count != 0 || sle.sle_data_count != 0) {
2080		spa_load_note(spa, "spa_load_verify found %llu metadata errors "
2081		    "and %llu data errors", (u_longlong_t)sle.sle_meta_count,
2082		    (u_longlong_t)sle.sle_data_count);
2083	}
2084
2085	if (spa_load_verify_dryrun ||
2086	    (!error && sle.sle_meta_count <= policy.zlp_maxmeta &&
2087	    sle.sle_data_count <= policy.zlp_maxdata)) {
2088		int64_t loss = 0;
2089
2090		verify_ok = B_TRUE;
2091		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
2092		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
2093
2094		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
2095		VERIFY(nvlist_add_uint64(spa->spa_load_info,
2096		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
2097		VERIFY(nvlist_add_int64(spa->spa_load_info,
2098		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
2099		VERIFY(nvlist_add_uint64(spa->spa_load_info,
2100		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
2101	} else {
2102		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2103	}
2104
2105	if (spa_load_verify_dryrun)
2106		return (0);
2107
2108	if (error) {
2109		if (error != ENXIO && error != EIO)
2110			error = SET_ERROR(EIO);
2111		return (error);
2112	}
2113
2114	return (verify_ok ? 0 : EIO);
2115}
2116
2117/*
2118 * Find a value in the pool props object.
2119 */
2120static void
2121spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2122{
2123	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2124	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2125}
2126
2127/*
2128 * Find a value in the pool directory object.
2129 */
2130static int
2131spa_dir_prop(spa_t *spa, const char *name, uint64_t *val, boolean_t log_enoent)
2132{
2133	int error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2134	    name, sizeof (uint64_t), 1, val);
2135
2136	if (error != 0 && (error != ENOENT || log_enoent)) {
2137		spa_load_failed(spa, "couldn't get '%s' value in MOS directory "
2138		    "[error=%d]", name, error);
2139	}
2140
2141	return (error);
2142}
2143
2144static int
2145spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2146{
2147	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2148	return (SET_ERROR(err));
2149}
2150
2151static void
2152spa_spawn_aux_threads(spa_t *spa)
2153{
2154	ASSERT(spa_writeable(spa));
2155
2156	ASSERT(MUTEX_HELD(&spa_namespace_lock));
2157
2158	spa_start_indirect_condensing_thread(spa);
2159
2160	ASSERT3P(spa->spa_checkpoint_discard_zthr, ==, NULL);
2161	spa->spa_checkpoint_discard_zthr =
2162	    zthr_create(spa_checkpoint_discard_thread_check,
2163	    spa_checkpoint_discard_thread, spa);
2164}
2165
2166/*
2167 * Fix up config after a partly-completed split.  This is done with the
2168 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
2169 * pool have that entry in their config, but only the splitting one contains
2170 * a list of all the guids of the vdevs that are being split off.
2171 *
2172 * This function determines what to do with that list: either rejoin
2173 * all the disks to the pool, or complete the splitting process.  To attempt
2174 * the rejoin, each disk that is offlined is marked online again, and
2175 * we do a reopen() call.  If the vdev label for every disk that was
2176 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2177 * then we call vdev_split() on each disk, and complete the split.
2178 *
2179 * Otherwise we leave the config alone, with all the vdevs in place in
2180 * the original pool.
2181 */
2182static void
2183spa_try_repair(spa_t *spa, nvlist_t *config)
2184{
2185	uint_t extracted;
2186	uint64_t *glist;
2187	uint_t i, gcount;
2188	nvlist_t *nvl;
2189	vdev_t **vd;
2190	boolean_t attempt_reopen;
2191
2192	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2193		return;
2194
2195	/* check that the config is complete */
2196	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2197	    &glist, &gcount) != 0)
2198		return;
2199
2200	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2201
2202	/* attempt to online all the vdevs & validate */
2203	attempt_reopen = B_TRUE;
2204	for (i = 0; i < gcount; i++) {
2205		if (glist[i] == 0)	/* vdev is hole */
2206			continue;
2207
2208		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2209		if (vd[i] == NULL) {
2210			/*
2211			 * Don't bother attempting to reopen the disks;
2212			 * just do the split.
2213			 */
2214			attempt_reopen = B_FALSE;
2215		} else {
2216			/* attempt to re-online it */
2217			vd[i]->vdev_offline = B_FALSE;
2218		}
2219	}
2220
2221	if (attempt_reopen) {
2222		vdev_reopen(spa->spa_root_vdev);
2223
2224		/* check each device to see what state it's in */
2225		for (extracted = 0, i = 0; i < gcount; i++) {
2226			if (vd[i] != NULL &&
2227			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2228				break;
2229			++extracted;
2230		}
2231	}
2232
2233	/*
2234	 * If every disk has been moved to the new pool, or if we never
2235	 * even attempted to look at them, then we split them off for
2236	 * good.
2237	 */
2238	if (!attempt_reopen || gcount == extracted) {
2239		for (i = 0; i < gcount; i++)
2240			if (vd[i] != NULL)
2241				vdev_split(vd[i]);
2242		vdev_reopen(spa->spa_root_vdev);
2243	}
2244
2245	kmem_free(vd, gcount * sizeof (vdev_t *));
2246}
2247
2248static int
2249spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type)
2250{
2251	char *ereport = FM_EREPORT_ZFS_POOL;
2252	int error;
2253
2254	spa->spa_load_state = state;
2255
2256	gethrestime(&spa->spa_loaded_ts);
2257	error = spa_load_impl(spa, type, &ereport);
2258
2259	/*
2260	 * Don't count references from objsets that are already closed
2261	 * and are making their way through the eviction process.
2262	 */
2263	spa_evicting_os_wait(spa);
2264	spa->spa_minref = refcount_count(&spa->spa_refcount);
2265	if (error) {
2266		if (error != EEXIST) {
2267			spa->spa_loaded_ts.tv_sec = 0;
2268			spa->spa_loaded_ts.tv_nsec = 0;
2269		}
2270		if (error != EBADF) {
2271			zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2272		}
2273	}
2274	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2275	spa->spa_ena = 0;
2276
2277	return (error);
2278}
2279
2280/*
2281 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2282 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2283 * spa's per-vdev ZAP list.
2284 */
2285static uint64_t
2286vdev_count_verify_zaps(vdev_t *vd)
2287{
2288	spa_t *spa = vd->vdev_spa;
2289	uint64_t total = 0;
2290	if (vd->vdev_top_zap != 0) {
2291		total++;
2292		ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2293		    spa->spa_all_vdev_zaps, vd->vdev_top_zap));
2294	}
2295	if (vd->vdev_leaf_zap != 0) {
2296		total++;
2297		ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2298		    spa->spa_all_vdev_zaps, vd->vdev_leaf_zap));
2299	}
2300
2301	for (uint64_t i = 0; i < vd->vdev_children; i++) {
2302		total += vdev_count_verify_zaps(vd->vdev_child[i]);
2303	}
2304
2305	return (total);
2306}
2307
2308static int
2309spa_verify_host(spa_t *spa, nvlist_t *mos_config)
2310{
2311	uint64_t hostid;
2312	char *hostname;
2313	uint64_t myhostid = 0;
2314
2315	if (!spa_is_root(spa) && nvlist_lookup_uint64(mos_config,
2316	    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2317		hostname = fnvlist_lookup_string(mos_config,
2318		    ZPOOL_CONFIG_HOSTNAME);
2319
2320		myhostid = zone_get_hostid(NULL);
2321
2322		if (hostid != 0 && myhostid != 0 && hostid != myhostid) {
2323			cmn_err(CE_WARN, "pool '%s' could not be "
2324			    "loaded as it was last accessed by "
2325			    "another system (host: %s hostid: 0x%llx). "
2326			    "See: http://illumos.org/msg/ZFS-8000-EY",
2327			    spa_name(spa), hostname, (u_longlong_t)hostid);
2328			spa_load_failed(spa, "hostid verification failed: pool "
2329			    "last accessed by host: %s (hostid: 0x%llx)",
2330			    hostname, (u_longlong_t)hostid);
2331			return (SET_ERROR(EBADF));
2332		}
2333	}
2334
2335	return (0);
2336}
2337
2338static int
2339spa_ld_parse_config(spa_t *spa, spa_import_type_t type)
2340{
2341	int error = 0;
2342	nvlist_t *nvtree, *nvl, *config = spa->spa_config;
2343	int parse;
2344	vdev_t *rvd;
2345	uint64_t pool_guid;
2346	char *comment;
2347
2348	/*
2349	 * Versioning wasn't explicitly added to the label until later, so if
2350	 * it's not present treat it as the initial version.
2351	 */
2352	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2353	    &spa->spa_ubsync.ub_version) != 0)
2354		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2355
2356	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
2357		spa_load_failed(spa, "invalid config provided: '%s' missing",
2358		    ZPOOL_CONFIG_POOL_GUID);
2359		return (SET_ERROR(EINVAL));
2360	}
2361
2362	/*
2363	 * If we are doing an import, ensure that the pool is not already
2364	 * imported by checking if its pool guid already exists in the
2365	 * spa namespace.
2366	 *
2367	 * The only case that we allow an already imported pool to be
2368	 * imported again, is when the pool is checkpointed and we want to
2369	 * look at its checkpointed state from userland tools like zdb.
2370	 */
2371#ifdef _KERNEL
2372	if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
2373	    spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
2374	    spa_guid_exists(pool_guid, 0)) {
2375#else
2376	if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
2377	    spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
2378	    spa_guid_exists(pool_guid, 0) &&
2379	    !spa_importing_readonly_checkpoint(spa)) {
2380#endif
2381		spa_load_failed(spa, "a pool with guid %llu is already open",
2382		    (u_longlong_t)pool_guid);
2383		return (SET_ERROR(EEXIST));
2384	}
2385
2386	spa->spa_config_guid = pool_guid;
2387
2388	nvlist_free(spa->spa_load_info);
2389	spa->spa_load_info = fnvlist_alloc();
2390
2391	ASSERT(spa->spa_comment == NULL);
2392	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2393		spa->spa_comment = spa_strdup(comment);
2394
2395	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2396	    &spa->spa_config_txg);
2397
2398	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) == 0)
2399		spa->spa_config_splitting = fnvlist_dup(nvl);
2400
2401	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtree)) {
2402		spa_load_failed(spa, "invalid config provided: '%s' missing",
2403		    ZPOOL_CONFIG_VDEV_TREE);
2404		return (SET_ERROR(EINVAL));
2405	}
2406
2407	/*
2408	 * Create "The Godfather" zio to hold all async IOs
2409	 */
2410	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2411	    KM_SLEEP);
2412	for (int i = 0; i < max_ncpus; i++) {
2413		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2414		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2415		    ZIO_FLAG_GODFATHER);
2416	}
2417
2418	/*
2419	 * Parse the configuration into a vdev tree.  We explicitly set the
2420	 * value that will be returned by spa_version() since parsing the
2421	 * configuration requires knowing the version number.
2422	 */
2423	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2424	parse = (type == SPA_IMPORT_EXISTING ?
2425	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2426	error = spa_config_parse(spa, &rvd, nvtree, NULL, 0, parse);
2427	spa_config_exit(spa, SCL_ALL, FTAG);
2428
2429	if (error != 0) {
2430		spa_load_failed(spa, "unable to parse config [error=%d]",
2431		    error);
2432		return (error);
2433	}
2434
2435	ASSERT(spa->spa_root_vdev == rvd);
2436	ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
2437	ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
2438
2439	if (type != SPA_IMPORT_ASSEMBLE) {
2440		ASSERT(spa_guid(spa) == pool_guid);
2441	}
2442
2443	return (0);
2444}
2445
2446/*
2447 * Recursively open all vdevs in the vdev tree. This function is called twice:
2448 * first with the untrusted config, then with the trusted config.
2449 */
2450static int
2451spa_ld_open_vdevs(spa_t *spa)
2452{
2453	int error = 0;
2454
2455	/*
2456	 * spa_missing_tvds_allowed defines how many top-level vdevs can be
2457	 * missing/unopenable for the root vdev to be still considered openable.
2458	 */
2459	if (spa->spa_trust_config) {
2460		spa->spa_missing_tvds_allowed = zfs_max_missing_tvds;
2461	} else if (spa->spa_config_source == SPA_CONFIG_SRC_CACHEFILE) {
2462		spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_cachefile;
2463	} else if (spa->spa_config_source == SPA_CONFIG_SRC_SCAN) {
2464		spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_scan;
2465	} else {
2466		spa->spa_missing_tvds_allowed = 0;
2467	}
2468
2469	spa->spa_missing_tvds_allowed =
2470	    MAX(zfs_max_missing_tvds, spa->spa_missing_tvds_allowed);
2471
2472	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2473	error = vdev_open(spa->spa_root_vdev);
2474	spa_config_exit(spa, SCL_ALL, FTAG);
2475
2476	if (spa->spa_missing_tvds != 0) {
2477		spa_load_note(spa, "vdev tree has %lld missing top-level "
2478		    "vdevs.", (u_longlong_t)spa->spa_missing_tvds);
2479		if (spa->spa_trust_config && (spa->spa_mode & FWRITE)) {
2480			/*
2481			 * Although theoretically we could allow users to open
2482			 * incomplete pools in RW mode, we'd need to add a lot
2483			 * of extra logic (e.g. adjust pool space to account
2484			 * for missing vdevs).
2485			 * This limitation also prevents users from accidentally
2486			 * opening the pool in RW mode during data recovery and
2487			 * damaging it further.
2488			 */
2489			spa_load_note(spa, "pools with missing top-level "
2490			    "vdevs can only be opened in read-only mode.");
2491			error = SET_ERROR(ENXIO);
2492		} else {
2493			spa_load_note(spa, "current settings allow for maximum "
2494			    "%lld missing top-level vdevs at this stage.",
2495			    (u_longlong_t)spa->spa_missing_tvds_allowed);
2496		}
2497	}
2498	if (error != 0) {
2499		spa_load_failed(spa, "unable to open vdev tree [error=%d]",
2500		    error);
2501	}
2502	if (spa->spa_missing_tvds != 0 || error != 0)
2503		vdev_dbgmsg_print_tree(spa->spa_root_vdev, 2);
2504
2505	return (error);
2506}
2507
2508/*
2509 * We need to validate the vdev labels against the configuration that
2510 * we have in hand. This function is called twice: first with an untrusted
2511 * config, then with a trusted config. The validation is more strict when the
2512 * config is trusted.
2513 */
2514static int
2515spa_ld_validate_vdevs(spa_t *spa)
2516{
2517	int error = 0;
2518	vdev_t *rvd = spa->spa_root_vdev;
2519
2520	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2521	error = vdev_validate(rvd);
2522	spa_config_exit(spa, SCL_ALL, FTAG);
2523
2524	if (error != 0) {
2525		spa_load_failed(spa, "vdev_validate failed [error=%d]", error);
2526		return (error);
2527	}
2528
2529	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
2530		spa_load_failed(spa, "cannot open vdev tree after invalidating "
2531		    "some vdevs");
2532		vdev_dbgmsg_print_tree(rvd, 2);
2533		return (SET_ERROR(ENXIO));
2534	}
2535
2536	return (0);
2537}
2538
2539static void
2540spa_ld_select_uberblock_done(spa_t *spa, uberblock_t *ub)
2541{
2542	spa->spa_state = POOL_STATE_ACTIVE;
2543	spa->spa_ubsync = spa->spa_uberblock;
2544	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2545	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2546	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2547	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2548	spa->spa_claim_max_txg = spa->spa_first_txg;
2549	spa->spa_prev_software_version = ub->ub_software_version;
2550}
2551
2552static int
2553spa_ld_select_uberblock(spa_t *spa, spa_import_type_t type)
2554{
2555	vdev_t *rvd = spa->spa_root_vdev;
2556	nvlist_t *label;
2557	uberblock_t *ub = &spa->spa_uberblock;
2558
2559	/*
2560	 * If we are opening the checkpointed state of the pool by
2561	 * rewinding to it, at this point we will have written the
2562	 * checkpointed uberblock to the vdev labels, so searching
2563	 * the labels will find the right uberblock.  However, if
2564	 * we are opening the checkpointed state read-only, we have
2565	 * not modified the labels. Therefore, we must ignore the
2566	 * labels and continue using the spa_uberblock that was set
2567	 * by spa_ld_checkpoint_rewind.
2568	 *
2569	 * Note that it would be fine to ignore the labels when
2570	 * rewinding (opening writeable) as well. However, if we
2571	 * crash just after writing the labels, we will end up
2572	 * searching the labels. Doing so in the common case means
2573	 * that this code path gets exercised normally, rather than
2574	 * just in the edge case.
2575	 */
2576	if (ub->ub_checkpoint_txg != 0 &&
2577	    spa_importing_readonly_checkpoint(spa)) {
2578		spa_ld_select_uberblock_done(spa, ub);
2579		return (0);
2580	}
2581
2582	/*
2583	 * Find the best uberblock.
2584	 */
2585	vdev_uberblock_load(rvd, ub, &label);
2586
2587	/*
2588	 * If we weren't able to find a single valid uberblock, return failure.
2589	 */
2590	if (ub->ub_txg == 0) {
2591		nvlist_free(label);
2592		spa_load_failed(spa, "no valid uberblock found");
2593		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2594	}
2595
2596	spa_load_note(spa, "using uberblock with txg=%llu",
2597	    (u_longlong_t)ub->ub_txg);
2598
2599	/*
2600	 * If the pool has an unsupported version we can't open it.
2601	 */
2602	if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2603		nvlist_free(label);
2604		spa_load_failed(spa, "version %llu is not supported",
2605		    (u_longlong_t)ub->ub_version);
2606		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2607	}
2608
2609	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2610		nvlist_t *features;
2611
2612		/*
2613		 * If we weren't able to find what's necessary for reading the
2614		 * MOS in the label, return failure.
2615		 */
2616		if (label == NULL) {
2617			spa_load_failed(spa, "label config unavailable");
2618			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2619			    ENXIO));
2620		}
2621
2622		if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_FEATURES_FOR_READ,
2623		    &features) != 0) {
2624			nvlist_free(label);
2625			spa_load_failed(spa, "invalid label: '%s' missing",
2626			    ZPOOL_CONFIG_FEATURES_FOR_READ);
2627			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2628			    ENXIO));
2629		}
2630
2631		/*
2632		 * Update our in-core representation with the definitive values
2633		 * from the label.
2634		 */
2635		nvlist_free(spa->spa_label_features);
2636		VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2637	}
2638
2639	nvlist_free(label);
2640
2641	/*
2642	 * Look through entries in the label nvlist's features_for_read. If
2643	 * there is a feature listed there which we don't understand then we
2644	 * cannot open a pool.
2645	 */
2646	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2647		nvlist_t *unsup_feat;
2648
2649		VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2650		    0);
2651
2652		for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2653		    NULL); nvp != NULL;
2654		    nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2655			if (!zfeature_is_supported(nvpair_name(nvp))) {
2656				VERIFY(nvlist_add_string(unsup_feat,
2657				    nvpair_name(nvp), "") == 0);
2658			}
2659		}
2660
2661		if (!nvlist_empty(unsup_feat)) {
2662			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2663			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2664			nvlist_free(unsup_feat);
2665			spa_load_failed(spa, "some features are unsupported");
2666			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2667			    ENOTSUP));
2668		}
2669
2670		nvlist_free(unsup_feat);
2671	}
2672
2673	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2674		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2675		spa_try_repair(spa, spa->spa_config);
2676		spa_config_exit(spa, SCL_ALL, FTAG);
2677		nvlist_free(spa->spa_config_splitting);
2678		spa->spa_config_splitting = NULL;
2679	}
2680
2681	/*
2682	 * Initialize internal SPA structures.
2683	 */
2684	spa_ld_select_uberblock_done(spa, ub);
2685
2686	return (0);
2687}
2688
2689static int
2690spa_ld_open_rootbp(spa_t *spa)
2691{
2692	int error = 0;
2693	vdev_t *rvd = spa->spa_root_vdev;
2694
2695	error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2696	if (error != 0) {
2697		spa_load_failed(spa, "unable to open rootbp in dsl_pool_init "
2698		    "[error=%d]", error);
2699		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2700	}
2701	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2702
2703	return (0);
2704}
2705
2706static int
2707spa_ld_trusted_config(spa_t *spa, spa_import_type_t type,
2708    boolean_t reloading)
2709{
2710	vdev_t *mrvd, *rvd = spa->spa_root_vdev;
2711	nvlist_t *nv, *mos_config, *policy;
2712	int error = 0, copy_error;
2713	uint64_t healthy_tvds, healthy_tvds_mos;
2714	uint64_t mos_config_txg;
2715
2716	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object, B_TRUE)
2717	    != 0)
2718		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2719
2720	/*
2721	 * If we're assembling a pool from a split, the config provided is
2722	 * already trusted so there is nothing to do.
2723	 */
2724	if (type == SPA_IMPORT_ASSEMBLE)
2725		return (0);
2726
2727	healthy_tvds = spa_healthy_core_tvds(spa);
2728
2729	if (load_nvlist(spa, spa->spa_config_object, &mos_config)
2730	    != 0) {
2731		spa_load_failed(spa, "unable to retrieve MOS config");
2732		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2733	}
2734
2735	/*
2736	 * If we are doing an open, pool owner wasn't verified yet, thus do
2737	 * the verification here.
2738	 */
2739	if (spa->spa_load_state == SPA_LOAD_OPEN) {
2740		error = spa_verify_host(spa, mos_config);
2741		if (error != 0) {
2742			nvlist_free(mos_config);
2743			return (error);
2744		}
2745	}
2746
2747	nv = fnvlist_lookup_nvlist(mos_config, ZPOOL_CONFIG_VDEV_TREE);
2748
2749	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2750
2751	/*
2752	 * Build a new vdev tree from the trusted config
2753	 */
2754	VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
2755
2756	/*
2757	 * Vdev paths in the MOS may be obsolete. If the untrusted config was
2758	 * obtained by scanning /dev/dsk, then it will have the right vdev
2759	 * paths. We update the trusted MOS config with this information.
2760	 * We first try to copy the paths with vdev_copy_path_strict, which
2761	 * succeeds only when both configs have exactly the same vdev tree.
2762	 * If that fails, we fall back to a more flexible method that has a
2763	 * best effort policy.
2764	 */
2765	copy_error = vdev_copy_path_strict(rvd, mrvd);
2766	if (copy_error != 0 || spa_load_print_vdev_tree) {
2767		spa_load_note(spa, "provided vdev tree:");
2768		vdev_dbgmsg_print_tree(rvd, 2);
2769		spa_load_note(spa, "MOS vdev tree:");
2770		vdev_dbgmsg_print_tree(mrvd, 2);
2771	}
2772	if (copy_error != 0) {
2773		spa_load_note(spa, "vdev_copy_path_strict failed, falling "
2774		    "back to vdev_copy_path_relaxed");
2775		vdev_copy_path_relaxed(rvd, mrvd);
2776	}
2777
2778	vdev_close(rvd);
2779	vdev_free(rvd);
2780	spa->spa_root_vdev = mrvd;
2781	rvd = mrvd;
2782	spa_config_exit(spa, SCL_ALL, FTAG);
2783
2784	/*
2785	 * We will use spa_config if we decide to reload the spa or if spa_load
2786	 * fails and we rewind. We must thus regenerate the config using the
2787	 * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to
2788	 * pass settings on how to load the pool and is not stored in the MOS.
2789	 * We copy it over to our new, trusted config.
2790	 */
2791	mos_config_txg = fnvlist_lookup_uint64(mos_config,
2792	    ZPOOL_CONFIG_POOL_TXG);
2793	nvlist_free(mos_config);
2794	mos_config = spa_config_generate(spa, NULL, mos_config_txg, B_FALSE);
2795	if (nvlist_lookup_nvlist(spa->spa_config, ZPOOL_LOAD_POLICY,
2796	    &policy) == 0)
2797		fnvlist_add_nvlist(mos_config, ZPOOL_LOAD_POLICY, policy);
2798	spa_config_set(spa, mos_config);
2799	spa->spa_config_source = SPA_CONFIG_SRC_MOS;
2800
2801	/*
2802	 * Now that we got the config from the MOS, we should be more strict
2803	 * in checking blkptrs and can make assumptions about the consistency
2804	 * of the vdev tree. spa_trust_config must be set to true before opening
2805	 * vdevs in order for them to be writeable.
2806	 */
2807	spa->spa_trust_config = B_TRUE;
2808
2809	/*
2810	 * Open and validate the new vdev tree
2811	 */
2812	error = spa_ld_open_vdevs(spa);
2813	if (error != 0)
2814		return (error);
2815
2816	error = spa_ld_validate_vdevs(spa);
2817	if (error != 0)
2818		return (error);
2819
2820	if (copy_error != 0 || spa_load_print_vdev_tree) {
2821		spa_load_note(spa, "final vdev tree:");
2822		vdev_dbgmsg_print_tree(rvd, 2);
2823	}
2824
2825	if (spa->spa_load_state != SPA_LOAD_TRYIMPORT &&
2826	    !spa->spa_extreme_rewind && zfs_max_missing_tvds == 0) {
2827		/*
2828		 * Sanity check to make sure that we are indeed loading the
2829		 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds
2830		 * in the config provided and they happened to be the only ones
2831		 * to have the latest uberblock, we could involuntarily perform
2832		 * an extreme rewind.
2833		 */
2834		healthy_tvds_mos = spa_healthy_core_tvds(spa);
2835		if (healthy_tvds_mos - healthy_tvds >=
2836		    SPA_SYNC_MIN_VDEVS) {
2837			spa_load_note(spa, "config provided misses too many "
2838			    "top-level vdevs compared to MOS (%lld vs %lld). ",
2839			    (u_longlong_t)healthy_tvds,
2840			    (u_longlong_t)healthy_tvds_mos);
2841			spa_load_note(spa, "vdev tree:");
2842			vdev_dbgmsg_print_tree(rvd, 2);
2843			if (reloading) {
2844				spa_load_failed(spa, "config was already "
2845				    "provided from MOS. Aborting.");
2846				return (spa_vdev_err(rvd,
2847				    VDEV_AUX_CORRUPT_DATA, EIO));
2848			}
2849			spa_load_note(spa, "spa must be reloaded using MOS "
2850			    "config");
2851			return (SET_ERROR(EAGAIN));
2852		}
2853	}
2854
2855	error = spa_check_for_missing_logs(spa);
2856	if (error != 0)
2857		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2858
2859	if (rvd->vdev_guid_sum != spa->spa_uberblock.ub_guid_sum) {
2860		spa_load_failed(spa, "uberblock guid sum doesn't match MOS "
2861		    "guid sum (%llu != %llu)",
2862		    (u_longlong_t)spa->spa_uberblock.ub_guid_sum,
2863		    (u_longlong_t)rvd->vdev_guid_sum);
2864		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2865		    ENXIO));
2866	}
2867
2868	return (0);
2869}
2870
2871static int
2872spa_ld_open_indirect_vdev_metadata(spa_t *spa)
2873{
2874	int error = 0;
2875	vdev_t *rvd = spa->spa_root_vdev;
2876
2877	/*
2878	 * Everything that we read before spa_remove_init() must be stored
2879	 * on concreted vdevs.  Therefore we do this as early as possible.
2880	 */
2881	error = spa_remove_init(spa);
2882	if (error != 0) {
2883		spa_load_failed(spa, "spa_remove_init failed [error=%d]",
2884		    error);
2885		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2886	}
2887
2888	/*
2889	 * Retrieve information needed to condense indirect vdev mappings.
2890	 */
2891	error = spa_condense_init(spa);
2892	if (error != 0) {
2893		spa_load_failed(spa, "spa_condense_init failed [error=%d]",
2894		    error);
2895		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
2896	}
2897
2898	return (0);
2899}
2900
2901static int
2902spa_ld_check_features(spa_t *spa, boolean_t *missing_feat_writep)
2903{
2904	int error = 0;
2905	vdev_t *rvd = spa->spa_root_vdev;
2906
2907	if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2908		boolean_t missing_feat_read = B_FALSE;
2909		nvlist_t *unsup_feat, *enabled_feat;
2910
2911		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2912		    &spa->spa_feat_for_read_obj, B_TRUE) != 0) {
2913			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2914		}
2915
2916		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2917		    &spa->spa_feat_for_write_obj, B_TRUE) != 0) {
2918			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2919		}
2920
2921		if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2922		    &spa->spa_feat_desc_obj, B_TRUE) != 0) {
2923			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2924		}
2925
2926		enabled_feat = fnvlist_alloc();
2927		unsup_feat = fnvlist_alloc();
2928
2929		if (!spa_features_check(spa, B_FALSE,
2930		    unsup_feat, enabled_feat))
2931			missing_feat_read = B_TRUE;
2932
2933		if (spa_writeable(spa) ||
2934		    spa->spa_load_state == SPA_LOAD_TRYIMPORT) {
2935			if (!spa_features_check(spa, B_TRUE,
2936			    unsup_feat, enabled_feat)) {
2937				*missing_feat_writep = B_TRUE;
2938			}
2939		}
2940
2941		fnvlist_add_nvlist(spa->spa_load_info,
2942		    ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2943
2944		if (!nvlist_empty(unsup_feat)) {
2945			fnvlist_add_nvlist(spa->spa_load_info,
2946			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2947		}
2948
2949		fnvlist_free(enabled_feat);
2950		fnvlist_free(unsup_feat);
2951
2952		if (!missing_feat_read) {
2953			fnvlist_add_boolean(spa->spa_load_info,
2954			    ZPOOL_CONFIG_CAN_RDONLY);
2955		}
2956
2957		/*
2958		 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2959		 * twofold: to determine whether the pool is available for
2960		 * import in read-write mode and (if it is not) whether the
2961		 * pool is available for import in read-only mode. If the pool
2962		 * is available for import in read-write mode, it is displayed
2963		 * as available in userland; if it is not available for import
2964		 * in read-only mode, it is displayed as unavailable in
2965		 * userland. If the pool is available for import in read-only
2966		 * mode but not read-write mode, it is displayed as unavailable
2967		 * in userland with a special note that the pool is actually
2968		 * available for open in read-only mode.
2969		 *
2970		 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2971		 * missing a feature for write, we must first determine whether
2972		 * the pool can be opened read-only before returning to
2973		 * userland in order to know whether to display the
2974		 * abovementioned note.
2975		 */
2976		if (missing_feat_read || (*missing_feat_writep &&
2977		    spa_writeable(spa))) {
2978			spa_load_failed(spa, "pool uses unsupported features");
2979			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2980			    ENOTSUP));
2981		}
2982
2983		/*
2984		 * Load refcounts for ZFS features from disk into an in-memory
2985		 * cache during SPA initialization.
2986		 */
2987		for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2988			uint64_t refcount;
2989
2990			error = feature_get_refcount_from_disk(spa,
2991			    &spa_feature_table[i], &refcount);
2992			if (error == 0) {
2993				spa->spa_feat_refcount_cache[i] = refcount;
2994			} else if (error == ENOTSUP) {
2995				spa->spa_feat_refcount_cache[i] =
2996				    SPA_FEATURE_DISABLED;
2997			} else {
2998				spa_load_failed(spa, "error getting refcount "
2999				    "for feature %s [error=%d]",
3000				    spa_feature_table[i].fi_guid, error);
3001				return (spa_vdev_err(rvd,
3002				    VDEV_AUX_CORRUPT_DATA, EIO));
3003			}
3004		}
3005	}
3006
3007	if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
3008		if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
3009		    &spa->spa_feat_enabled_txg_obj, B_TRUE) != 0)
3010			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3011	}
3012
3013	return (0);
3014}
3015
3016static int
3017spa_ld_load_special_directories(spa_t *spa)
3018{
3019	int error = 0;
3020	vdev_t *rvd = spa->spa_root_vdev;
3021
3022	spa->spa_is_initializing = B_TRUE;
3023	error = dsl_pool_open(spa->spa_dsl_pool);
3024	spa->spa_is_initializing = B_FALSE;
3025	if (error != 0) {
3026		spa_load_failed(spa, "dsl_pool_open failed [error=%d]", error);
3027		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3028	}
3029
3030	return (0);
3031}
3032
3033static int
3034spa_ld_get_props(spa_t *spa)
3035{
3036	int error = 0;
3037	uint64_t obj;
3038	vdev_t *rvd = spa->spa_root_vdev;
3039
3040	/* Grab the secret checksum salt from the MOS. */
3041	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3042	    DMU_POOL_CHECKSUM_SALT, 1,
3043	    sizeof (spa->spa_cksum_salt.zcs_bytes),
3044	    spa->spa_cksum_salt.zcs_bytes);
3045	if (error == ENOENT) {
3046		/* Generate a new salt for subsequent use */
3047		(void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
3048		    sizeof (spa->spa_cksum_salt.zcs_bytes));
3049	} else if (error != 0) {
3050		spa_load_failed(spa, "unable to retrieve checksum salt from "
3051		    "MOS [error=%d]", error);
3052		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3053	}
3054
3055	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj, B_TRUE) != 0)
3056		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3057	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
3058	if (error != 0) {
3059		spa_load_failed(spa, "error opening deferred-frees bpobj "
3060		    "[error=%d]", error);
3061		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3062	}
3063
3064	/*
3065	 * Load the bit that tells us to use the new accounting function
3066	 * (raid-z deflation).  If we have an older pool, this will not
3067	 * be present.
3068	 */
3069	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate, B_FALSE);
3070	if (error != 0 && error != ENOENT)
3071		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3072
3073	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
3074	    &spa->spa_creation_version, B_FALSE);
3075	if (error != 0 && error != ENOENT)
3076		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3077
3078	/*
3079	 * Load the persistent error log.  If we have an older pool, this will
3080	 * not be present.
3081	 */
3082	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last,
3083	    B_FALSE);
3084	if (error != 0 && error != ENOENT)
3085		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3086
3087	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
3088	    &spa->spa_errlog_scrub, B_FALSE);
3089	if (error != 0 && error != ENOENT)
3090		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3091
3092	/*
3093	 * Load the history object.  If we have an older pool, this
3094	 * will not be present.
3095	 */
3096	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history, B_FALSE);
3097	if (error != 0 && error != ENOENT)
3098		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3099
3100	/*
3101	 * Load the per-vdev ZAP map. If we have an older pool, this will not
3102	 * be present; in this case, defer its creation to a later time to
3103	 * avoid dirtying the MOS this early / out of sync context. See
3104	 * spa_sync_config_object.
3105	 */
3106
3107	/* The sentinel is only available in the MOS config. */
3108	nvlist_t *mos_config;
3109	if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) {
3110		spa_load_failed(spa, "unable to retrieve MOS config");
3111		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3112	}
3113
3114	error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP,
3115	    &spa->spa_all_vdev_zaps, B_FALSE);
3116
3117	if (error == ENOENT) {
3118		VERIFY(!nvlist_exists(mos_config,
3119		    ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
3120		spa->spa_avz_action = AVZ_ACTION_INITIALIZE;
3121		ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
3122	} else if (error != 0) {
3123		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3124	} else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) {
3125		/*
3126		 * An older version of ZFS overwrote the sentinel value, so
3127		 * we have orphaned per-vdev ZAPs in the MOS. Defer their
3128		 * destruction to later; see spa_sync_config_object.
3129		 */
3130		spa->spa_avz_action = AVZ_ACTION_DESTROY;
3131		/*
3132		 * We're assuming that no vdevs have had their ZAPs created
3133		 * before this. Better be sure of it.
3134		 */
3135		ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
3136	}
3137	nvlist_free(mos_config);
3138
3139	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3140
3141	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object,
3142	    B_FALSE);
3143	if (error && error != ENOENT)
3144		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3145
3146	if (error == 0) {
3147		uint64_t autoreplace;
3148
3149		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
3150		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
3151		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
3152		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
3153		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
3154		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
3155		    &spa->spa_dedup_ditto);
3156
3157		spa->spa_autoreplace = (autoreplace != 0);
3158	}
3159
3160	/*
3161	 * If we are importing a pool with missing top-level vdevs,
3162	 * we enforce that the pool doesn't panic or get suspended on
3163	 * error since the likelihood of missing data is extremely high.
3164	 */
3165	if (spa->spa_missing_tvds > 0 &&
3166	    spa->spa_failmode != ZIO_FAILURE_MODE_CONTINUE &&
3167	    spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3168		spa_load_note(spa, "forcing failmode to 'continue' "
3169		    "as some top level vdevs are missing");
3170		spa->spa_failmode = ZIO_FAILURE_MODE_CONTINUE;
3171	}
3172
3173	return (0);
3174}
3175
3176static int
3177spa_ld_open_aux_vdevs(spa_t *spa, spa_import_type_t type)
3178{
3179	int error = 0;
3180	vdev_t *rvd = spa->spa_root_vdev;
3181
3182	/*
3183	 * If we're assembling the pool from the split-off vdevs of
3184	 * an existing pool, we don't want to attach the spares & cache
3185	 * devices.
3186	 */
3187
3188	/*
3189	 * Load any hot spares for this pool.
3190	 */
3191	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object,
3192	    B_FALSE);
3193	if (error != 0 && error != ENOENT)
3194		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3195	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
3196		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
3197		if (load_nvlist(spa, spa->spa_spares.sav_object,
3198		    &spa->spa_spares.sav_config) != 0) {
3199			spa_load_failed(spa, "error loading spares nvlist");
3200			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3201		}
3202
3203		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3204		spa_load_spares(spa);
3205		spa_config_exit(spa, SCL_ALL, FTAG);
3206	} else if (error == 0) {
3207		spa->spa_spares.sav_sync = B_TRUE;
3208	}
3209
3210	/*
3211	 * Load any level 2 ARC devices for this pool.
3212	 */
3213	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
3214	    &spa->spa_l2cache.sav_object, B_FALSE);
3215	if (error != 0 && error != ENOENT)
3216		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3217	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
3218		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
3219		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
3220		    &spa->spa_l2cache.sav_config) != 0) {
3221			spa_load_failed(spa, "error loading l2cache nvlist");
3222			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3223		}
3224
3225		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3226		spa_load_l2cache(spa);
3227		spa_config_exit(spa, SCL_ALL, FTAG);
3228	} else if (error == 0) {
3229		spa->spa_l2cache.sav_sync = B_TRUE;
3230	}
3231
3232	return (0);
3233}
3234
3235static int
3236spa_ld_load_vdev_metadata(spa_t *spa)
3237{
3238	int error = 0;
3239	vdev_t *rvd = spa->spa_root_vdev;
3240
3241	/*
3242	 * If the 'autoreplace' property is set, then post a resource notifying
3243	 * the ZFS DE that it should not issue any faults for unopenable
3244	 * devices.  We also iterate over the vdevs, and post a sysevent for any
3245	 * unopenable vdevs so that the normal autoreplace handler can take
3246	 * over.
3247	 */
3248	if (spa->spa_autoreplace && spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3249		spa_check_removed(spa->spa_root_vdev);
3250		/*
3251		 * For the import case, this is done in spa_import(), because
3252		 * at this point we're using the spare definitions from
3253		 * the MOS config, not necessarily from the userland config.
3254		 */
3255		if (spa->spa_load_state != SPA_LOAD_IMPORT) {
3256			spa_aux_check_removed(&spa->spa_spares);
3257			spa_aux_check_removed(&spa->spa_l2cache);
3258		}
3259	}
3260
3261	/*
3262	 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc.
3263	 */
3264	error = vdev_load(rvd);
3265	if (error != 0) {
3266		spa_load_failed(spa, "vdev_load failed [error=%d]", error);
3267		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
3268	}
3269
3270	/*
3271	 * Propagate the leaf DTLs we just loaded all the way up the vdev tree.
3272	 */
3273	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3274	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
3275	spa_config_exit(spa, SCL_ALL, FTAG);
3276
3277	return (0);
3278}
3279
3280static int
3281spa_ld_load_dedup_tables(spa_t *spa)
3282{
3283	int error = 0;
3284	vdev_t *rvd = spa->spa_root_vdev;
3285
3286	error = ddt_load(spa);
3287	if (error != 0) {
3288		spa_load_failed(spa, "ddt_load failed [error=%d]", error);
3289		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3290	}
3291
3292	return (0);
3293}
3294
3295static int
3296spa_ld_verify_logs(spa_t *spa, spa_import_type_t type, char **ereport)
3297{
3298	vdev_t *rvd = spa->spa_root_vdev;
3299
3300	if (type != SPA_IMPORT_ASSEMBLE && spa_writeable(spa)) {
3301		boolean_t missing = spa_check_logs(spa);
3302		if (missing) {
3303			if (spa->spa_missing_tvds != 0) {
3304				spa_load_note(spa, "spa_check_logs failed "
3305				    "so dropping the logs");
3306			} else {
3307				*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
3308				spa_load_failed(spa, "spa_check_logs failed");
3309				return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG,
3310				    ENXIO));
3311			}
3312		}
3313	}
3314
3315	return (0);
3316}
3317
3318static int
3319spa_ld_verify_pool_data(spa_t *spa)
3320{
3321	int error = 0;
3322	vdev_t *rvd = spa->spa_root_vdev;
3323
3324	/*
3325	 * We've successfully opened the pool, verify that we're ready
3326	 * to start pushing transactions.
3327	 */
3328	if (spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3329		error = spa_load_verify(spa);
3330		if (error != 0) {
3331			spa_load_failed(spa, "spa_load_verify failed "
3332			    "[error=%d]", error);
3333			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
3334			    error));
3335		}
3336	}
3337
3338	return (0);
3339}
3340
3341static void
3342spa_ld_claim_log_blocks(spa_t *spa)
3343{
3344	dmu_tx_t *tx;
3345	dsl_pool_t *dp = spa_get_dsl(spa);
3346
3347	/*
3348	 * Claim log blocks that haven't been committed yet.
3349	 * This must all happen in a single txg.
3350	 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
3351	 * invoked from zil_claim_log_block()'s i/o done callback.
3352	 * Price of rollback is that we abandon the log.
3353	 */
3354	spa->spa_claiming = B_TRUE;
3355
3356	tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
3357	(void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
3358	    zil_claim, tx, DS_FIND_CHILDREN);
3359	dmu_tx_commit(tx);
3360
3361	spa->spa_claiming = B_FALSE;
3362
3363	spa_set_log_state(spa, SPA_LOG_GOOD);
3364}
3365
3366static void
3367spa_ld_check_for_config_update(spa_t *spa, uint64_t config_cache_txg,
3368    boolean_t update_config_cache)
3369{
3370	vdev_t *rvd = spa->spa_root_vdev;
3371	int need_update = B_FALSE;
3372
3373	/*
3374	 * If the config cache is stale, or we have uninitialized
3375	 * metaslabs (see spa_vdev_add()), then update the config.
3376	 *
3377	 * If this is a verbatim import, trust the current
3378	 * in-core spa_config and update the disk labels.
3379	 */
3380	if (update_config_cache || config_cache_txg != spa->spa_config_txg ||
3381	    spa->spa_load_state == SPA_LOAD_IMPORT ||
3382	    spa->spa_load_state == SPA_LOAD_RECOVER ||
3383	    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
3384		need_update = B_TRUE;
3385
3386	for (int c = 0; c < rvd->vdev_children; c++)
3387		if (rvd->vdev_child[c]->vdev_ms_array == 0)
3388			need_update = B_TRUE;
3389
3390	/*
3391	 * Update the config cache asychronously in case we're the
3392	 * root pool, in which case the config cache isn't writable yet.
3393	 */
3394	if (need_update)
3395		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
3396}
3397
3398static void
3399spa_ld_prepare_for_reload(spa_t *spa)
3400{
3401	int mode = spa->spa_mode;
3402	int async_suspended = spa->spa_async_suspended;
3403
3404	spa_unload(spa);
3405	spa_deactivate(spa);
3406	spa_activate(spa, mode);
3407
3408	/*
3409	 * We save the value of spa_async_suspended as it gets reset to 0 by
3410	 * spa_unload(). We want to restore it back to the original value before
3411	 * returning as we might be calling spa_async_resume() later.
3412	 */
3413	spa->spa_async_suspended = async_suspended;
3414}
3415
3416static int
3417spa_ld_read_checkpoint_txg(spa_t *spa)
3418{
3419	uberblock_t checkpoint;
3420	int error = 0;
3421
3422	ASSERT0(spa->spa_checkpoint_txg);
3423	ASSERT(MUTEX_HELD(&spa_namespace_lock));
3424
3425	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3426	    DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
3427	    sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
3428
3429	if (error == ENOENT)
3430		return (0);
3431
3432	if (error != 0)
3433		return (error);
3434
3435	ASSERT3U(checkpoint.ub_txg, !=, 0);
3436	ASSERT3U(checkpoint.ub_checkpoint_txg, !=, 0);
3437	ASSERT3U(checkpoint.ub_timestamp, !=, 0);
3438	spa->spa_checkpoint_txg = checkpoint.ub_txg;
3439	spa->spa_checkpoint_info.sci_timestamp = checkpoint.ub_timestamp;
3440
3441	return (0);
3442}
3443
3444static int
3445spa_ld_mos_init(spa_t *spa, spa_import_type_t type)
3446{
3447	int error = 0;
3448
3449	ASSERT(MUTEX_HELD(&spa_namespace_lock));
3450	ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
3451
3452	/*
3453	 * Never trust the config that is provided unless we are assembling
3454	 * a pool following a split.
3455	 * This means don't trust blkptrs and the vdev tree in general. This
3456	 * also effectively puts the spa in read-only mode since
3457	 * spa_writeable() checks for spa_trust_config to be true.
3458	 * We will later load a trusted config from the MOS.
3459	 */
3460	if (type != SPA_IMPORT_ASSEMBLE)
3461		spa->spa_trust_config = B_FALSE;
3462
3463	/*
3464	 * Parse the config provided to create a vdev tree.
3465	 */
3466	error = spa_ld_parse_config(spa, type);
3467	if (error != 0)
3468		return (error);
3469
3470	/*
3471	 * Now that we have the vdev tree, try to open each vdev. This involves
3472	 * opening the underlying physical device, retrieving its geometry and
3473	 * probing the vdev with a dummy I/O. The state of each vdev will be set
3474	 * based on the success of those operations. After this we'll be ready
3475	 * to read from the vdevs.
3476	 */
3477	error = spa_ld_open_vdevs(spa);
3478	if (error != 0)
3479		return (error);
3480
3481	/*
3482	 * Read the label of each vdev and make sure that the GUIDs stored
3483	 * there match the GUIDs in the config provided.
3484	 * If we're assembling a new pool that's been split off from an
3485	 * existing pool, the labels haven't yet been updated so we skip
3486	 * validation for now.
3487	 */
3488	if (type != SPA_IMPORT_ASSEMBLE) {
3489		error = spa_ld_validate_vdevs(spa);
3490		if (error != 0)
3491			return (error);
3492	}
3493
3494	/*
3495	 * Read all vdev labels to find the best uberblock (i.e. latest,
3496	 * unless spa_load_max_txg is set) and store it in spa_uberblock. We
3497	 * get the list of features required to read blkptrs in the MOS from
3498	 * the vdev label with the best uberblock and verify that our version
3499	 * of zfs supports them all.
3500	 */
3501	error = spa_ld_select_uberblock(spa, type);
3502	if (error != 0)
3503		return (error);
3504
3505	/*
3506	 * Pass that uberblock to the dsl_pool layer which will open the root
3507	 * blkptr. This blkptr points to the latest version of the MOS and will
3508	 * allow us to read its contents.
3509	 */
3510	error = spa_ld_open_rootbp(spa);
3511	if (error != 0)
3512		return (error);
3513
3514	return (0);
3515}
3516
3517static int
3518spa_ld_checkpoint_rewind(spa_t *spa)
3519{
3520	uberblock_t checkpoint;
3521	int error = 0;
3522
3523	ASSERT(MUTEX_HELD(&spa_namespace_lock));
3524	ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
3525
3526	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3527	    DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
3528	    sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
3529
3530	if (error != 0) {
3531		spa_load_failed(spa, "unable to retrieve checkpointed "
3532		    "uberblock from the MOS config [error=%d]", error);
3533
3534		if (error == ENOENT)
3535			error = ZFS_ERR_NO_CHECKPOINT;
3536
3537		return (error);
3538	}
3539
3540	ASSERT3U(checkpoint.ub_txg, <, spa->spa_uberblock.ub_txg);
3541	ASSERT3U(checkpoint.ub_txg, ==, checkpoint.ub_checkpoint_txg);
3542
3543	/*
3544	 * We need to update the txg and timestamp of the checkpointed
3545	 * uberblock to be higher than the latest one. This ensures that
3546	 * the checkpointed uberblock is selected if we were to close and
3547	 * reopen the pool right after we've written it in the vdev labels.
3548	 * (also see block comment in vdev_uberblock_compare)
3549	 */
3550	checkpoint.ub_txg = spa->spa_uberblock.ub_txg + 1;
3551	checkpoint.ub_timestamp = gethrestime_sec();
3552
3553	/*
3554	 * Set current uberblock to be the checkpointed uberblock.
3555	 */
3556	spa->spa_uberblock = checkpoint;
3557
3558	/*
3559	 * If we are doing a normal rewind, then the pool is open for
3560	 * writing and we sync the "updated" checkpointed uberblock to
3561	 * disk. Once this is done, we've basically rewound the whole
3562	 * pool and there is no way back.
3563	 *
3564	 * There are cases when we don't want to attempt and sync the
3565	 * checkpointed uberblock to disk because we are opening a
3566	 * pool as read-only. Specifically, verifying the checkpointed
3567	 * state with zdb, and importing the checkpointed state to get
3568	 * a "preview" of its content.
3569	 */
3570	if (spa_writeable(spa)) {
3571		vdev_t *rvd = spa->spa_root_vdev;
3572
3573		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3574		vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
3575		int svdcount = 0;
3576		int children = rvd->vdev_children;
3577		int c0 = spa_get_random(children);
3578
3579		for (int c = 0; c < children; c++) {
3580			vdev_t *vd = rvd->vdev_child[(c0 + c) % children];
3581
3582			/* Stop when revisiting the first vdev */
3583			if (c > 0 && svd[0] == vd)
3584				break;
3585
3586			if (vd->vdev_ms_array == 0 || vd->vdev_islog ||
3587			    !vdev_is_concrete(vd))
3588				continue;
3589
3590			svd[svdcount++] = vd;
3591			if (svdcount == SPA_SYNC_MIN_VDEVS)
3592				break;
3593		}
3594		error = vdev_config_sync(svd, svdcount, spa->spa_first_txg);
3595		if (error == 0)
3596			spa->spa_last_synced_guid = rvd->vdev_guid;
3597		spa_config_exit(spa, SCL_ALL, FTAG);
3598
3599		if (error != 0) {
3600			spa_load_failed(spa, "failed to write checkpointed "
3601			    "uberblock to the vdev labels [error=%d]", error);
3602			return (error);
3603		}
3604	}
3605
3606	return (0);
3607}
3608
3609static int
3610spa_ld_mos_with_trusted_config(spa_t *spa, spa_import_type_t type,
3611    boolean_t *update_config_cache)
3612{
3613	int error;
3614
3615	/*
3616	 * Parse the config for pool, open and validate vdevs,
3617	 * select an uberblock, and use that uberblock to open
3618	 * the MOS.
3619	 */
3620	error = spa_ld_mos_init(spa, type);
3621	if (error != 0)
3622		return (error);
3623
3624	/*
3625	 * Retrieve the trusted config stored in the MOS and use it to create
3626	 * a new, exact version of the vdev tree, then reopen all vdevs.
3627	 */
3628	error = spa_ld_trusted_config(spa, type, B_FALSE);
3629	if (error == EAGAIN) {
3630		if (update_config_cache != NULL)
3631			*update_config_cache = B_TRUE;
3632
3633		/*
3634		 * Redo the loading process with the trusted config if it is
3635		 * too different from the untrusted config.
3636		 */
3637		spa_ld_prepare_for_reload(spa);
3638		spa_load_note(spa, "RELOADING");
3639		error = spa_ld_mos_init(spa, type);
3640		if (error != 0)
3641			return (error);
3642
3643		error = spa_ld_trusted_config(spa, type, B_TRUE);
3644		if (error != 0)
3645			return (error);
3646
3647	} else if (error != 0) {
3648		return (error);
3649	}
3650
3651	return (0);
3652}
3653
3654/*
3655 * Load an existing storage pool, using the config provided. This config
3656 * describes which vdevs are part of the pool and is later validated against
3657 * partial configs present in each vdev's label and an entire copy of the
3658 * config stored in the MOS.
3659 */
3660static int
3661spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport)
3662{
3663	int error = 0;
3664	boolean_t missing_feat_write = B_FALSE;
3665	boolean_t checkpoint_rewind =
3666	    (spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
3667	boolean_t update_config_cache = B_FALSE;
3668
3669	ASSERT(MUTEX_HELD(&spa_namespace_lock));
3670	ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
3671
3672	spa_load_note(spa, "LOADING");
3673
3674	error = spa_ld_mos_with_trusted_config(spa, type, &update_config_cache);
3675	if (error != 0)
3676		return (error);
3677
3678	/*
3679	 * If we are rewinding to the checkpoint then we need to repeat
3680	 * everything we've done so far in this function but this time
3681	 * selecting the checkpointed uberblock and using that to open
3682	 * the MOS.
3683	 */
3684	if (checkpoint_rewind) {
3685		/*
3686		 * If we are rewinding to the checkpoint update config cache
3687		 * anyway.
3688		 */
3689		update_config_cache = B_TRUE;
3690
3691		/*
3692		 * Extract the checkpointed uberblock from the current MOS
3693		 * and use this as the pool's uberblock from now on. If the
3694		 * pool is imported as writeable we also write the checkpoint
3695		 * uberblock to the labels, making the rewind permanent.
3696		 */
3697		error = spa_ld_checkpoint_rewind(spa);
3698		if (error != 0)
3699			return (error);
3700
3701		/*
3702		 * Redo the loading process process again with the
3703		 * checkpointed uberblock.
3704		 */
3705		spa_ld_prepare_for_reload(spa);
3706		spa_load_note(spa, "LOADING checkpointed uberblock");
3707		error = spa_ld_mos_with_trusted_config(spa, type, NULL);
3708		if (error != 0)
3709			return (error);
3710	}
3711
3712	/*
3713	 * Retrieve the checkpoint txg if the pool has a checkpoint.
3714	 */
3715	error = spa_ld_read_checkpoint_txg(spa);
3716	if (error != 0)
3717		return (error);
3718
3719	/*
3720	 * Retrieve the mapping of indirect vdevs. Those vdevs were removed
3721	 * from the pool and their contents were re-mapped to other vdevs. Note
3722	 * that everything that we read before this step must have been
3723	 * rewritten on concrete vdevs after the last device removal was
3724	 * initiated. Otherwise we could be reading from indirect vdevs before
3725	 * we have loaded their mappings.
3726	 */
3727	error = spa_ld_open_indirect_vdev_metadata(spa);
3728	if (error != 0)
3729		return (error);
3730
3731	/*
3732	 * Retrieve the full list of active features from the MOS and check if
3733	 * they are all supported.
3734	 */
3735	error = spa_ld_check_features(spa, &missing_feat_write);
3736	if (error != 0)
3737		return (error);
3738
3739	/*
3740	 * Load several special directories from the MOS needed by the dsl_pool
3741	 * layer.
3742	 */
3743	error = spa_ld_load_special_directories(spa);
3744	if (error != 0)
3745		return (error);
3746
3747	/*
3748	 * Retrieve pool properties from the MOS.
3749	 */
3750	error = spa_ld_get_props(spa);
3751	if (error != 0)
3752		return (error);
3753
3754	/*
3755	 * Retrieve the list of auxiliary devices - cache devices and spares -
3756	 * and open them.
3757	 */
3758	error = spa_ld_open_aux_vdevs(spa, type);
3759	if (error != 0)
3760		return (error);
3761
3762	/*
3763	 * Load the metadata for all vdevs. Also check if unopenable devices
3764	 * should be autoreplaced.
3765	 */
3766	error = spa_ld_load_vdev_metadata(spa);
3767	if (error != 0)
3768		return (error);
3769
3770	error = spa_ld_load_dedup_tables(spa);
3771	if (error != 0)
3772		return (error);
3773
3774	/*
3775	 * Verify the logs now to make sure we don't have any unexpected errors
3776	 * when we claim log blocks later.
3777	 */
3778	error = spa_ld_verify_logs(spa, type, ereport);
3779	if (error != 0)
3780		return (error);
3781
3782	if (missing_feat_write) {
3783		ASSERT(spa->spa_load_state == SPA_LOAD_TRYIMPORT);
3784
3785		/*
3786		 * At this point, we know that we can open the pool in
3787		 * read-only mode but not read-write mode. We now have enough
3788		 * information and can return to userland.
3789		 */
3790		return (spa_vdev_err(spa->spa_root_vdev, VDEV_AUX_UNSUP_FEAT,
3791		    ENOTSUP));
3792	}
3793
3794	/*
3795	 * Traverse the last txgs to make sure the pool was left off in a safe
3796	 * state. When performing an extreme rewind, we verify the whole pool,
3797	 * which can take a very long time.
3798	 */
3799	error = spa_ld_verify_pool_data(spa);
3800	if (error != 0)
3801		return (error);
3802
3803	/*
3804	 * Calculate the deflated space for the pool. This must be done before
3805	 * we write anything to the pool because we'd need to update the space
3806	 * accounting using the deflated sizes.
3807	 */
3808	spa_update_dspace(spa);
3809
3810	/*
3811	 * We have now retrieved all the information we needed to open the
3812	 * pool. If we are importing the pool in read-write mode, a few
3813	 * additional steps must be performed to finish the import.
3814	 */
3815	if (spa_writeable(spa) && (spa->spa_load_state == SPA_LOAD_RECOVER ||
3816	    spa->spa_load_max_txg == UINT64_MAX)) {
3817		uint64_t config_cache_txg = spa->spa_config_txg;
3818
3819		ASSERT(spa->spa_load_state != SPA_LOAD_TRYIMPORT);
3820
3821		/*
3822		 * In case of a checkpoint rewind, log the original txg
3823		 * of the checkpointed uberblock.
3824		 */
3825		if (checkpoint_rewind) {
3826			spa_history_log_internal(spa, "checkpoint rewind",
3827			    NULL, "rewound state to txg=%llu",
3828			    (u_longlong_t)spa->spa_uberblock.ub_checkpoint_txg);
3829		}
3830
3831		/*
3832		 * Traverse the ZIL and claim all blocks.
3833		 */
3834		spa_ld_claim_log_blocks(spa);
3835
3836		/*
3837		 * Kick-off the syncing thread.
3838		 */
3839		spa->spa_sync_on = B_TRUE;
3840		txg_sync_start(spa->spa_dsl_pool);
3841
3842		/*
3843		 * Wait for all claims to sync.  We sync up to the highest
3844		 * claimed log block birth time so that claimed log blocks
3845		 * don't appear to be from the future.  spa_claim_max_txg
3846		 * will have been set for us by ZIL traversal operations
3847		 * performed above.
3848		 */
3849		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
3850
3851		/*
3852		 * Check if we need to request an update of the config. On the
3853		 * next sync, we would update the config stored in vdev labels
3854		 * and the cachefile (by default /etc/zfs/zpool.cache).
3855		 */
3856		spa_ld_check_for_config_update(spa, config_cache_txg,
3857		    update_config_cache);
3858
3859		/*
3860		 * Check all DTLs to see if anything needs resilvering.
3861		 */
3862		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
3863		    vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL))
3864			spa_async_request(spa, SPA_ASYNC_RESILVER);
3865
3866		/*
3867		 * Log the fact that we booted up (so that we can detect if
3868		 * we rebooted in the middle of an operation).
3869		 */
3870		spa_history_log_version(spa, "open");
3871
3872		spa_restart_removal(spa);
3873		spa_spawn_aux_threads(spa);
3874
3875		/*
3876		 * Delete any inconsistent datasets.
3877		 *
3878		 * Note:
3879		 * Since we may be issuing deletes for clones here,
3880		 * we make sure to do so after we've spawned all the
3881		 * auxiliary threads above (from which the livelist
3882		 * deletion zthr is part of).
3883		 */
3884		(void) dmu_objset_find(spa_name(spa),
3885		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
3886
3887		/*
3888		 * Clean up any stale temporary dataset userrefs.
3889		 */
3890		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
3891
3892		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3893		vdev_initialize_restart(spa->spa_root_vdev);
3894		spa_config_exit(spa, SCL_CONFIG, FTAG);
3895	}
3896
3897	spa_load_note(spa, "LOADED");
3898
3899	return (0);
3900}
3901
3902static int
3903spa_load_retry(spa_t *spa, spa_load_state_t state)
3904{
3905	int mode = spa->spa_mode;
3906
3907	spa_unload(spa);
3908	spa_deactivate(spa);
3909
3910	spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
3911
3912	spa_activate(spa, mode);
3913	spa_async_suspend(spa);
3914
3915	spa_load_note(spa, "spa_load_retry: rewind, max txg: %llu",
3916	    (u_longlong_t)spa->spa_load_max_txg);
3917
3918	return (spa_load(spa, state, SPA_IMPORT_EXISTING));
3919}
3920
3921/*
3922 * If spa_load() fails this function will try loading prior txg's. If
3923 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
3924 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
3925 * function will not rewind the pool and will return the same error as
3926 * spa_load().
3927 */
3928static int
3929spa_load_best(spa_t *spa, spa_load_state_t state, uint64_t max_request,
3930    int rewind_flags)
3931{
3932	nvlist_t *loadinfo = NULL;
3933	nvlist_t *config = NULL;
3934	int load_error, rewind_error;
3935	uint64_t safe_rewind_txg;
3936	uint64_t min_txg;
3937
3938	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
3939		spa->spa_load_max_txg = spa->spa_load_txg;
3940		spa_set_log_state(spa, SPA_LOG_CLEAR);
3941	} else {
3942		spa->spa_load_max_txg = max_request;
3943		if (max_request != UINT64_MAX)
3944			spa->spa_extreme_rewind = B_TRUE;
3945	}
3946
3947	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING);
3948	if (load_error == 0)
3949		return (0);
3950	if (load_error == ZFS_ERR_NO_CHECKPOINT) {
3951		/*
3952		 * When attempting checkpoint-rewind on a pool with no
3953		 * checkpoint, we should not attempt to load uberblocks
3954		 * from previous txgs when spa_load fails.
3955		 */
3956		ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
3957		return (load_error);
3958	}
3959
3960	if (spa->spa_root_vdev != NULL)
3961		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3962
3963	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
3964	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
3965
3966	if (rewind_flags & ZPOOL_NEVER_REWIND) {
3967		nvlist_free(config);
3968		return (load_error);
3969	}
3970
3971	if (state == SPA_LOAD_RECOVER) {
3972		/* Price of rolling back is discarding txgs, including log */
3973		spa_set_log_state(spa, SPA_LOG_CLEAR);
3974	} else {
3975		/*
3976		 * If we aren't rolling back save the load info from our first
3977		 * import attempt so that we can restore it after attempting
3978		 * to rewind.
3979		 */
3980		loadinfo = spa->spa_load_info;
3981		spa->spa_load_info = fnvlist_alloc();
3982	}
3983
3984	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
3985	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
3986	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
3987	    TXG_INITIAL : safe_rewind_txg;
3988
3989	/*
3990	 * Continue as long as we're finding errors, we're still within
3991	 * the acceptable rewind range, and we're still finding uberblocks
3992	 */
3993	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
3994	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
3995		if (spa->spa_load_max_txg < safe_rewind_txg)
3996			spa->spa_extreme_rewind = B_TRUE;
3997		rewind_error = spa_load_retry(spa, state);
3998	}
3999
4000	spa->spa_extreme_rewind = B_FALSE;
4001	spa->spa_load_max_txg = UINT64_MAX;
4002
4003	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
4004		spa_config_set(spa, config);
4005	else
4006		nvlist_free(config);
4007
4008	if (state == SPA_LOAD_RECOVER) {
4009		ASSERT3P(loadinfo, ==, NULL);
4010		return (rewind_error);
4011	} else {
4012		/* Store the rewind info as part of the initial load info */
4013		fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
4014		    spa->spa_load_info);
4015
4016		/* Restore the initial load info */
4017		fnvlist_free(spa->spa_load_info);
4018		spa->spa_load_info = loadinfo;
4019
4020		return (load_error);
4021	}
4022}
4023
4024/*
4025 * Pool Open/Import
4026 *
4027 * The import case is identical to an open except that the configuration is sent
4028 * down from userland, instead of grabbed from the configuration cache.  For the
4029 * case of an open, the pool configuration will exist in the
4030 * POOL_STATE_UNINITIALIZED state.
4031 *
4032 * The stats information (gen/count/ustats) is used to gather vdev statistics at
4033 * the same time open the pool, without having to keep around the spa_t in some
4034 * ambiguous state.
4035 */
4036static int
4037spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
4038    nvlist_t **config)
4039{
4040	spa_t *spa;
4041	spa_load_state_t state = SPA_LOAD_OPEN;
4042	int error;
4043	int locked = B_FALSE;
4044
4045	*spapp = NULL;
4046
4047	/*
4048	 * As disgusting as this is, we need to support recursive calls to this
4049	 * function because dsl_dir_open() is called during spa_load(), and ends
4050	 * up calling spa_open() again.  The real fix is to figure out how to
4051	 * avoid dsl_dir_open() calling this in the first place.
4052	 */
4053	if (mutex_owner(&spa_namespace_lock) != curthread) {
4054		mutex_enter(&spa_namespace_lock);
4055		locked = B_TRUE;
4056	}
4057
4058	if ((spa = spa_lookup(pool)) == NULL) {
4059		if (locked)
4060			mutex_exit(&spa_namespace_lock);
4061		return (SET_ERROR(ENOENT));
4062	}
4063
4064	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
4065		zpool_load_policy_t policy;
4066
4067		zpool_get_load_policy(nvpolicy ? nvpolicy : spa->spa_config,
4068		    &policy);
4069		if (policy.zlp_rewind & ZPOOL_DO_REWIND)
4070			state = SPA_LOAD_RECOVER;
4071
4072		spa_activate(spa, spa_mode_global);
4073
4074		if (state != SPA_LOAD_RECOVER)
4075			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4076		spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
4077
4078		zfs_dbgmsg("spa_open_common: opening %s", pool);
4079		error = spa_load_best(spa, state, policy.zlp_txg,
4080		    policy.zlp_rewind);
4081
4082		if (error == EBADF) {
4083			/*
4084			 * If vdev_validate() returns failure (indicated by
4085			 * EBADF), it indicates that one of the vdevs indicates
4086			 * that the pool has been exported or destroyed.  If
4087			 * this is the case, the config cache is out of sync and
4088			 * we should remove the pool from the namespace.
4089			 */
4090			spa_unload(spa);
4091			spa_deactivate(spa);
4092			spa_write_cachefile(spa, B_TRUE, B_TRUE);
4093			spa_remove(spa);
4094			if (locked)
4095				mutex_exit(&spa_namespace_lock);
4096			return (SET_ERROR(ENOENT));
4097		}
4098
4099		if (error) {
4100			/*
4101			 * We can't open the pool, but we still have useful
4102			 * information: the state of each vdev after the
4103			 * attempted vdev_open().  Return this to the user.
4104			 */
4105			if (config != NULL && spa->spa_config) {
4106				VERIFY(nvlist_dup(spa->spa_config, config,
4107				    KM_SLEEP) == 0);
4108				VERIFY(nvlist_add_nvlist(*config,
4109				    ZPOOL_CONFIG_LOAD_INFO,
4110				    spa->spa_load_info) == 0);
4111			}
4112			spa_unload(spa);
4113			spa_deactivate(spa);
4114			spa->spa_last_open_failed = error;
4115			if (locked)
4116				mutex_exit(&spa_namespace_lock);
4117			*spapp = NULL;
4118			return (error);
4119		}
4120	}
4121
4122	spa_open_ref(spa, tag);
4123
4124	if (config != NULL)
4125		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4126
4127	/*
4128	 * If we've recovered the pool, pass back any information we
4129	 * gathered while doing the load.
4130	 */
4131	if (state == SPA_LOAD_RECOVER) {
4132		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
4133		    spa->spa_load_info) == 0);
4134	}
4135
4136	if (locked) {
4137		spa->spa_last_open_failed = 0;
4138		spa->spa_last_ubsync_txg = 0;
4139		spa->spa_load_txg = 0;
4140		mutex_exit(&spa_namespace_lock);
4141	}
4142
4143	*spapp = spa;
4144
4145	return (0);
4146}
4147
4148int
4149spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
4150    nvlist_t **config)
4151{
4152	return (spa_open_common(name, spapp, tag, policy, config));
4153}
4154
4155int
4156spa_open(const char *name, spa_t **spapp, void *tag)
4157{
4158	return (spa_open_common(name, spapp, tag, NULL, NULL));
4159}
4160
4161/*
4162 * Lookup the given spa_t, incrementing the inject count in the process,
4163 * preventing it from being exported or destroyed.
4164 */
4165spa_t *
4166spa_inject_addref(char *name)
4167{
4168	spa_t *spa;
4169
4170	mutex_enter(&spa_namespace_lock);
4171	if ((spa = spa_lookup(name)) == NULL) {
4172		mutex_exit(&spa_namespace_lock);
4173		return (NULL);
4174	}
4175	spa->spa_inject_ref++;
4176	mutex_exit(&spa_namespace_lock);
4177
4178	return (spa);
4179}
4180
4181void
4182spa_inject_delref(spa_t *spa)
4183{
4184	mutex_enter(&spa_namespace_lock);
4185	spa->spa_inject_ref--;
4186	mutex_exit(&spa_namespace_lock);
4187}
4188
4189/*
4190 * Add spares device information to the nvlist.
4191 */
4192static void
4193spa_add_spares(spa_t *spa, nvlist_t *config)
4194{
4195	nvlist_t **spares;
4196	uint_t i, nspares;
4197	nvlist_t *nvroot;
4198	uint64_t guid;
4199	vdev_stat_t *vs;
4200	uint_t vsc;
4201	uint64_t pool;
4202
4203	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4204
4205	if (spa->spa_spares.sav_count == 0)
4206		return;
4207
4208	VERIFY(nvlist_lookup_nvlist(config,
4209	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
4210	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4211	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
4212	if (nspares != 0) {
4213		VERIFY(nvlist_add_nvlist_array(nvroot,
4214		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4215		VERIFY(nvlist_lookup_nvlist_array(nvroot,
4216		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
4217
4218		/*
4219		 * Go through and find any spares which have since been
4220		 * repurposed as an active spare.  If this is the case, update
4221		 * their status appropriately.
4222		 */
4223		for (i = 0; i < nspares; i++) {
4224			VERIFY(nvlist_lookup_uint64(spares[i],
4225			    ZPOOL_CONFIG_GUID, &guid) == 0);
4226			if (spa_spare_exists(guid, &pool, NULL) &&
4227			    pool != 0ULL) {
4228				VERIFY(nvlist_lookup_uint64_array(
4229				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
4230				    (uint64_t **)&vs, &vsc) == 0);
4231				vs->vs_state = VDEV_STATE_CANT_OPEN;
4232				vs->vs_aux = VDEV_AUX_SPARED;
4233			}
4234		}
4235	}
4236}
4237
4238/*
4239 * Add l2cache device information to the nvlist, including vdev stats.
4240 */
4241static void
4242spa_add_l2cache(spa_t *spa, nvlist_t *config)
4243{
4244	nvlist_t **l2cache;
4245	uint_t i, j, nl2cache;
4246	nvlist_t *nvroot;
4247	uint64_t guid;
4248	vdev_t *vd;
4249	vdev_stat_t *vs;
4250	uint_t vsc;
4251
4252	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4253
4254	if (spa->spa_l2cache.sav_count == 0)
4255		return;
4256
4257	VERIFY(nvlist_lookup_nvlist(config,
4258	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
4259	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4260	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
4261	if (nl2cache != 0) {
4262		VERIFY(nvlist_add_nvlist_array(nvroot,
4263		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4264		VERIFY(nvlist_lookup_nvlist_array(nvroot,
4265		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
4266
4267		/*
4268		 * Update level 2 cache device stats.
4269		 */
4270
4271		for (i = 0; i < nl2cache; i++) {
4272			VERIFY(nvlist_lookup_uint64(l2cache[i],
4273			    ZPOOL_CONFIG_GUID, &guid) == 0);
4274
4275			vd = NULL;
4276			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
4277				if (guid ==
4278				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
4279					vd = spa->spa_l2cache.sav_vdevs[j];
4280					break;
4281				}
4282			}
4283			ASSERT(vd != NULL);
4284
4285			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
4286			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
4287			    == 0);
4288			vdev_get_stats(vd, vs);
4289		}
4290	}
4291}
4292
4293static void
4294spa_add_feature_stats(spa_t *spa, nvlist_t *config)
4295{
4296	nvlist_t *features;
4297	zap_cursor_t zc;
4298	zap_attribute_t za;
4299
4300	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4301	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4302
4303	if (spa->spa_feat_for_read_obj != 0) {
4304		for (zap_cursor_init(&zc, spa->spa_meta_objset,
4305		    spa->spa_feat_for_read_obj);
4306		    zap_cursor_retrieve(&zc, &za) == 0;
4307		    zap_cursor_advance(&zc)) {
4308			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
4309			    za.za_num_integers == 1);
4310			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
4311			    za.za_first_integer));
4312		}
4313		zap_cursor_fini(&zc);
4314	}
4315
4316	if (spa->spa_feat_for_write_obj != 0) {
4317		for (zap_cursor_init(&zc, spa->spa_meta_objset,
4318		    spa->spa_feat_for_write_obj);
4319		    zap_cursor_retrieve(&zc, &za) == 0;
4320		    zap_cursor_advance(&zc)) {
4321			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
4322			    za.za_num_integers == 1);
4323			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
4324			    za.za_first_integer));
4325		}
4326		zap_cursor_fini(&zc);
4327	}
4328
4329	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
4330	    features) == 0);
4331	nvlist_free(features);
4332}
4333
4334int
4335spa_get_stats(const char *name, nvlist_t **config,
4336    char *altroot, size_t buflen)
4337{
4338	int error;
4339	spa_t *spa;
4340
4341	*config = NULL;
4342	error = spa_open_common(name, &spa, FTAG, NULL, config);
4343
4344	if (spa != NULL) {
4345		/*
4346		 * This still leaves a window of inconsistency where the spares
4347		 * or l2cache devices could change and the config would be
4348		 * self-inconsistent.
4349		 */
4350		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4351
4352		if (*config != NULL) {
4353			uint64_t loadtimes[2];
4354
4355			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
4356			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
4357			VERIFY(nvlist_add_uint64_array(*config,
4358			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
4359
4360			VERIFY(nvlist_add_uint64(*config,
4361			    ZPOOL_CONFIG_ERRCOUNT,
4362			    spa_get_errlog_size(spa)) == 0);
4363
4364			if (spa_suspended(spa))
4365				VERIFY(nvlist_add_uint64(*config,
4366				    ZPOOL_CONFIG_SUSPENDED,
4367				    spa->spa_failmode) == 0);
4368
4369			spa_add_spares(spa, *config);
4370			spa_add_l2cache(spa, *config);
4371			spa_add_feature_stats(spa, *config);
4372		}
4373	}
4374
4375	/*
4376	 * We want to get the alternate root even for faulted pools, so we cheat
4377	 * and call spa_lookup() directly.
4378	 */
4379	if (altroot) {
4380		if (spa == NULL) {
4381			mutex_enter(&spa_namespace_lock);
4382			spa = spa_lookup(name);
4383			if (spa)
4384				spa_altroot(spa, altroot, buflen);
4385			else
4386				altroot[0] = '\0';
4387			spa = NULL;
4388			mutex_exit(&spa_namespace_lock);
4389		} else {
4390			spa_altroot(spa, altroot, buflen);
4391		}
4392	}
4393
4394	if (spa != NULL) {
4395		spa_config_exit(spa, SCL_CONFIG, FTAG);
4396		spa_close(spa, FTAG);
4397	}
4398
4399	return (error);
4400}
4401
4402/*
4403 * Validate that the auxiliary device array is well formed.  We must have an
4404 * array of nvlists, each which describes a valid leaf vdev.  If this is an
4405 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
4406 * specified, as long as they are well-formed.
4407 */
4408static int
4409spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
4410    spa_aux_vdev_t *sav, const char *config, uint64_t version,
4411    vdev_labeltype_t label)
4412{
4413	nvlist_t **dev;
4414	uint_t i, ndev;
4415	vdev_t *vd;
4416	int error;
4417
4418	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4419
4420	/*
4421	 * It's acceptable to have no devs specified.
4422	 */
4423	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
4424		return (0);
4425
4426	if (ndev == 0)
4427		return (SET_ERROR(EINVAL));
4428
4429	/*
4430	 * Make sure the pool is formatted with a version that supports this
4431	 * device type.
4432	 */
4433	if (spa_version(spa) < version)
4434		return (SET_ERROR(ENOTSUP));
4435
4436	/*
4437	 * Set the pending device list so we correctly handle device in-use
4438	 * checking.
4439	 */
4440	sav->sav_pending = dev;
4441	sav->sav_npending = ndev;
4442
4443	for (i = 0; i < ndev; i++) {
4444		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
4445		    mode)) != 0)
4446			goto out;
4447
4448		if (!vd->vdev_ops->vdev_op_leaf) {
4449			vdev_free(vd);
4450			error = SET_ERROR(EINVAL);
4451			goto out;
4452		}
4453
4454		/*
4455		 * The L2ARC currently only supports disk devices in
4456		 * kernel context.  For user-level testing, we allow it.
4457		 */
4458#ifdef _KERNEL
4459		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
4460		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
4461			error = SET_ERROR(ENOTBLK);
4462			vdev_free(vd);
4463			goto out;
4464		}
4465#endif
4466		vd->vdev_top = vd;
4467
4468		if ((error = vdev_open(vd)) == 0 &&
4469		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
4470			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
4471			    vd->vdev_guid) == 0);
4472		}
4473
4474		vdev_free(vd);
4475
4476		if (error &&
4477		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
4478			goto out;
4479		else
4480			error = 0;
4481	}
4482
4483out:
4484	sav->sav_pending = NULL;
4485	sav->sav_npending = 0;
4486	return (error);
4487}
4488
4489static int
4490spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
4491{
4492	int error;
4493
4494	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4495
4496	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
4497	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
4498	    VDEV_LABEL_SPARE)) != 0) {
4499		return (error);
4500	}
4501
4502	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
4503	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
4504	    VDEV_LABEL_L2CACHE));
4505}
4506
4507static void
4508spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
4509    const char *config)
4510{
4511	int i;
4512
4513	if (sav->sav_config != NULL) {
4514		nvlist_t **olddevs;
4515		uint_t oldndevs;
4516		nvlist_t **newdevs;
4517
4518		/*
4519		 * Generate new dev list by concatentating with the
4520		 * current dev list.
4521		 */
4522		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
4523		    &olddevs, &oldndevs) == 0);
4524
4525		newdevs = kmem_alloc(sizeof (void *) *
4526		    (ndevs + oldndevs), KM_SLEEP);
4527		for (i = 0; i < oldndevs; i++)
4528			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
4529			    KM_SLEEP) == 0);
4530		for (i = 0; i < ndevs; i++)
4531			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
4532			    KM_SLEEP) == 0);
4533
4534		VERIFY(nvlist_remove(sav->sav_config, config,
4535		    DATA_TYPE_NVLIST_ARRAY) == 0);
4536
4537		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
4538		    config, newdevs, ndevs + oldndevs) == 0);
4539		for (i = 0; i < oldndevs + ndevs; i++)
4540			nvlist_free(newdevs[i]);
4541		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
4542	} else {
4543		/*
4544		 * Generate a new dev list.
4545		 */
4546		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
4547		    KM_SLEEP) == 0);
4548		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
4549		    devs, ndevs) == 0);
4550	}
4551}
4552
4553/*
4554 * Stop and drop level 2 ARC devices
4555 */
4556void
4557spa_l2cache_drop(spa_t *spa)
4558{
4559	vdev_t *vd;
4560	int i;
4561	spa_aux_vdev_t *sav = &spa->spa_l2cache;
4562
4563	for (i = 0; i < sav->sav_count; i++) {
4564		uint64_t pool;
4565
4566		vd = sav->sav_vdevs[i];
4567		ASSERT(vd != NULL);
4568
4569		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
4570		    pool != 0ULL && l2arc_vdev_present(vd))
4571			l2arc_remove_vdev(vd);
4572	}
4573}
4574
4575/*
4576 * Pool Creation
4577 */
4578int
4579spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
4580    nvlist_t *zplprops)
4581{
4582	spa_t *spa;
4583	char *altroot = NULL;
4584	vdev_t *rvd;
4585	dsl_pool_t *dp;
4586	dmu_tx_t *tx;
4587	int error = 0;
4588	uint64_t txg = TXG_INITIAL;
4589	nvlist_t **spares, **l2cache;
4590	uint_t nspares, nl2cache;
4591	uint64_t version, obj;
4592	boolean_t has_features;
4593	char *poolname;
4594	nvlist_t *nvl;
4595
4596	if (nvlist_lookup_string(props,
4597	    zpool_prop_to_name(ZPOOL_PROP_TNAME), &poolname) != 0)
4598		poolname = (char *)pool;
4599
4600	/*
4601	 * If this pool already exists, return failure.
4602	 */
4603	mutex_enter(&spa_namespace_lock);
4604	if (spa_lookup(poolname) != NULL) {
4605		mutex_exit(&spa_namespace_lock);
4606		return (SET_ERROR(EEXIST));
4607	}
4608
4609	/*
4610	 * Allocate a new spa_t structure.
4611	 */
4612	nvl = fnvlist_alloc();
4613	fnvlist_add_string(nvl, ZPOOL_CONFIG_POOL_NAME, pool);
4614	(void) nvlist_lookup_string(props,
4615	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4616	spa = spa_add(poolname, nvl, altroot);
4617	fnvlist_free(nvl);
4618	spa_activate(spa, spa_mode_global);
4619
4620	if (props && (error = spa_prop_validate(spa, props))) {
4621		spa_deactivate(spa);
4622		spa_remove(spa);
4623		mutex_exit(&spa_namespace_lock);
4624		return (error);
4625	}
4626
4627	/*
4628	 * Temporary pool names should never be written to disk.
4629	 */
4630	if (poolname != pool)
4631		spa->spa_import_flags |= ZFS_IMPORT_TEMP_NAME;
4632
4633	has_features = B_FALSE;
4634	for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
4635	    elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
4636		if (zpool_prop_feature(nvpair_name(elem)))
4637			has_features = B_TRUE;
4638	}
4639
4640	if (has_features || nvlist_lookup_uint64(props,
4641	    zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
4642		version = SPA_VERSION;
4643	}
4644	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
4645
4646	spa->spa_first_txg = txg;
4647	spa->spa_uberblock.ub_txg = txg - 1;
4648	spa->spa_uberblock.ub_version = version;
4649	spa->spa_ubsync = spa->spa_uberblock;
4650	spa->spa_load_state = SPA_LOAD_CREATE;
4651	spa->spa_removing_phys.sr_state = DSS_NONE;
4652	spa->spa_removing_phys.sr_removing_vdev = -1;
4653	spa->spa_removing_phys.sr_prev_indirect_vdev = -1;
4654
4655	/*
4656	 * Create "The Godfather" zio to hold all async IOs
4657	 */
4658	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
4659	    KM_SLEEP);
4660	for (int i = 0; i < max_ncpus; i++) {
4661		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
4662		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
4663		    ZIO_FLAG_GODFATHER);
4664	}
4665
4666	/*
4667	 * Create the root vdev.
4668	 */
4669	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4670
4671	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
4672
4673	ASSERT(error != 0 || rvd != NULL);
4674	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
4675
4676	if (error == 0 && !zfs_allocatable_devs(nvroot))
4677		error = SET_ERROR(EINVAL);
4678
4679	if (error == 0 &&
4680	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
4681	    (error = spa_validate_aux(spa, nvroot, txg,
4682	    VDEV_ALLOC_ADD)) == 0) {
4683		for (int c = 0; c < rvd->vdev_children; c++) {
4684			vdev_metaslab_set_size(rvd->vdev_child[c]);
4685			vdev_expand(rvd->vdev_child[c], txg);
4686		}
4687	}
4688
4689	spa_config_exit(spa, SCL_ALL, FTAG);
4690
4691	if (error != 0) {
4692		spa_unload(spa);
4693		spa_deactivate(spa);
4694		spa_remove(spa);
4695		mutex_exit(&spa_namespace_lock);
4696		return (error);
4697	}
4698
4699	/*
4700	 * Get the list of spares, if specified.
4701	 */
4702	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4703	    &spares, &nspares) == 0) {
4704		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
4705		    KM_SLEEP) == 0);
4706		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4707		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4708		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4709		spa_load_spares(spa);
4710		spa_config_exit(spa, SCL_ALL, FTAG);
4711		spa->spa_spares.sav_sync = B_TRUE;
4712	}
4713
4714	/*
4715	 * Get the list of level 2 cache devices, if specified.
4716	 */
4717	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4718	    &l2cache, &nl2cache) == 0) {
4719		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4720		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4721		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4722		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4723		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4724		spa_load_l2cache(spa);
4725		spa_config_exit(spa, SCL_ALL, FTAG);
4726		spa->spa_l2cache.sav_sync = B_TRUE;
4727	}
4728
4729	spa->spa_is_initializing = B_TRUE;
4730	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
4731	spa->spa_meta_objset = dp->dp_meta_objset;
4732	spa->spa_is_initializing = B_FALSE;
4733
4734	/*
4735	 * Create DDTs (dedup tables).
4736	 */
4737	ddt_create(spa);
4738
4739	spa_update_dspace(spa);
4740
4741	tx = dmu_tx_create_assigned(dp, txg);
4742
4743	/*
4744	 * Create the pool config object.
4745	 */
4746	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
4747	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
4748	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
4749
4750	if (zap_add(spa->spa_meta_objset,
4751	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
4752	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
4753		cmn_err(CE_PANIC, "failed to add pool config");
4754	}
4755
4756	if (spa_version(spa) >= SPA_VERSION_FEATURES)
4757		spa_feature_create_zap_objects(spa, tx);
4758
4759	if (zap_add(spa->spa_meta_objset,
4760	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
4761	    sizeof (uint64_t), 1, &version, tx) != 0) {
4762		cmn_err(CE_PANIC, "failed to add pool version");
4763	}
4764
4765	/* Newly created pools with the right version are always deflated. */
4766	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
4767		spa->spa_deflate = TRUE;
4768		if (zap_add(spa->spa_meta_objset,
4769		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
4770		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
4771			cmn_err(CE_PANIC, "failed to add deflate");
4772		}
4773	}
4774
4775	/*
4776	 * Create the deferred-free bpobj.  Turn off compression
4777	 * because sync-to-convergence takes longer if the blocksize
4778	 * keeps changing.
4779	 */
4780	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
4781	dmu_object_set_compress(spa->spa_meta_objset, obj,
4782	    ZIO_COMPRESS_OFF, tx);
4783	if (zap_add(spa->spa_meta_objset,
4784	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
4785	    sizeof (uint64_t), 1, &obj, tx) != 0) {
4786		cmn_err(CE_PANIC, "failed to add bpobj");
4787	}
4788	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
4789	    spa->spa_meta_objset, obj));
4790
4791	/*
4792	 * Create the pool's history object.
4793	 */
4794	if (version >= SPA_VERSION_ZPOOL_HISTORY)
4795		spa_history_create_obj(spa, tx);
4796
4797	/*
4798	 * Generate some random noise for salted checksums to operate on.
4799	 */
4800	(void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
4801	    sizeof (spa->spa_cksum_salt.zcs_bytes));
4802
4803	/*
4804	 * Set pool properties.
4805	 */
4806	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
4807	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
4808	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
4809	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
4810
4811	if (props != NULL) {
4812		spa_configfile_set(spa, props, B_FALSE);
4813		spa_sync_props(props, tx);
4814	}
4815
4816	dmu_tx_commit(tx);
4817
4818	spa->spa_sync_on = B_TRUE;
4819	txg_sync_start(spa->spa_dsl_pool);
4820
4821	/*
4822	 * We explicitly wait for the first transaction to complete so that our
4823	 * bean counters are appropriately updated.
4824	 */
4825	txg_wait_synced(spa->spa_dsl_pool, txg);
4826
4827	spa_spawn_aux_threads(spa);
4828
4829	spa_write_cachefile(spa, B_FALSE, B_TRUE);
4830	spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE);
4831
4832	spa_history_log_version(spa, "create");
4833
4834	/*
4835	 * Don't count references from objsets that are already closed
4836	 * and are making their way through the eviction process.
4837	 */
4838	spa_evicting_os_wait(spa);
4839	spa->spa_minref = refcount_count(&spa->spa_refcount);
4840	spa->spa_load_state = SPA_LOAD_NONE;
4841
4842	mutex_exit(&spa_namespace_lock);
4843
4844	return (0);
4845}
4846
4847#ifdef _KERNEL
4848/*
4849 * Get the root pool information from the root disk, then import the root pool
4850 * during the system boot up time.
4851 */
4852extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
4853
4854static nvlist_t *
4855spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
4856{
4857	nvlist_t *config;
4858	nvlist_t *nvtop, *nvroot;
4859	uint64_t pgid;
4860
4861	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
4862		return (NULL);
4863
4864	/*
4865	 * Add this top-level vdev to the child array.
4866	 */
4867	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4868	    &nvtop) == 0);
4869	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4870	    &pgid) == 0);
4871	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
4872
4873	/*
4874	 * Put this pool's top-level vdevs into a root vdev.
4875	 */
4876	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4877	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
4878	    VDEV_TYPE_ROOT) == 0);
4879	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
4880	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
4881	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4882	    &nvtop, 1) == 0);
4883
4884	/*
4885	 * Replace the existing vdev_tree with the new root vdev in
4886	 * this pool's configuration (remove the old, add the new).
4887	 */
4888	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
4889	nvlist_free(nvroot);
4890	return (config);
4891}
4892
4893/*
4894 * Walk the vdev tree and see if we can find a device with "better"
4895 * configuration. A configuration is "better" if the label on that
4896 * device has a more recent txg.
4897 */
4898static void
4899spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
4900{
4901	for (int c = 0; c < vd->vdev_children; c++)
4902		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
4903
4904	if (vd->vdev_ops->vdev_op_leaf) {
4905		nvlist_t *label;
4906		uint64_t label_txg;
4907
4908		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
4909		    &label) != 0)
4910			return;
4911
4912		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
4913		    &label_txg) == 0);
4914
4915		/*
4916		 * Do we have a better boot device?
4917		 */
4918		if (label_txg > *txg) {
4919			*txg = label_txg;
4920			*avd = vd;
4921		}
4922		nvlist_free(label);
4923	}
4924}
4925
4926/*
4927 * Import a root pool.
4928 *
4929 * For x86. devpath_list will consist of devid and/or physpath name of
4930 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
4931 * The GRUB "findroot" command will return the vdev we should boot.
4932 *
4933 * For Sparc, devpath_list consists the physpath name of the booting device
4934 * no matter the rootpool is a single device pool or a mirrored pool.
4935 * e.g.
4936 *	"/pci@1f,0/ide@d/disk@0,0:a"
4937 */
4938int
4939spa_import_rootpool(char *devpath, char *devid)
4940{
4941	spa_t *spa;
4942	vdev_t *rvd, *bvd, *avd = NULL;
4943	nvlist_t *config, *nvtop;
4944	uint64_t guid, txg;
4945	char *pname;
4946	int error;
4947
4948	/*
4949	 * Read the label from the boot device and generate a configuration.
4950	 */
4951	config = spa_generate_rootconf(devpath, devid, &guid);
4952#if defined(_OBP) && defined(_KERNEL)
4953	if (config == NULL) {
4954		if (strstr(devpath, "/iscsi/ssd") != NULL) {
4955			/* iscsi boot */
4956			get_iscsi_bootpath_phy(devpath);
4957			config = spa_generate_rootconf(devpath, devid, &guid);
4958		}
4959	}
4960#endif
4961	if (config == NULL) {
4962		cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
4963		    devpath);
4964		return (SET_ERROR(EIO));
4965	}
4966
4967	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
4968	    &pname) == 0);
4969	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
4970
4971	mutex_enter(&spa_namespace_lock);
4972	if ((spa = spa_lookup(pname)) != NULL) {
4973		/*
4974		 * Remove the existing root pool from the namespace so that we
4975		 * can replace it with the correct config we just read in.
4976		 */
4977		spa_remove(spa);
4978	}
4979
4980	spa = spa_add(pname, config, NULL);
4981	spa->spa_is_root = B_TRUE;
4982	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
4983	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
4984	    &spa->spa_ubsync.ub_version) != 0)
4985		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
4986
4987	/*
4988	 * Build up a vdev tree based on the boot device's label config.
4989	 */
4990	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4991	    &nvtop) == 0);
4992	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4993	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
4994	    VDEV_ALLOC_ROOTPOOL);
4995	spa_config_exit(spa, SCL_ALL, FTAG);
4996	if (error) {
4997		mutex_exit(&spa_namespace_lock);
4998		nvlist_free(config);
4999		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
5000		    pname);
5001		return (error);
5002	}
5003
5004	/*
5005	 * Get the boot vdev.
5006	 */
5007	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
5008		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
5009		    (u_longlong_t)guid);
5010		error = SET_ERROR(ENOENT);
5011		goto out;
5012	}
5013
5014	/*
5015	 * Determine if there is a better boot device.
5016	 */
5017	avd = bvd;
5018	spa_alt_rootvdev(rvd, &avd, &txg);
5019	if (avd != bvd) {
5020		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
5021		    "try booting from '%s'", avd->vdev_path);
5022		error = SET_ERROR(EINVAL);
5023		goto out;
5024	}
5025
5026	/*
5027	 * If the boot device is part of a spare vdev then ensure that
5028	 * we're booting off the active spare.
5029	 */
5030	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
5031	    !bvd->vdev_isspare) {
5032		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
5033		    "try booting from '%s'",
5034		    bvd->vdev_parent->
5035		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
5036		error = SET_ERROR(EINVAL);
5037		goto out;
5038	}
5039
5040	error = 0;
5041out:
5042	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5043	vdev_free(rvd);
5044	spa_config_exit(spa, SCL_ALL, FTAG);
5045	mutex_exit(&spa_namespace_lock);
5046
5047	nvlist_free(config);
5048	return (error);
5049}
5050
5051#endif
5052
5053/*
5054 * Import a non-root pool into the system.
5055 */
5056int
5057spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
5058{
5059	spa_t *spa;
5060	char *altroot = NULL;
5061	spa_load_state_t state = SPA_LOAD_IMPORT;
5062	zpool_load_policy_t policy;
5063	uint64_t mode = spa_mode_global;
5064	uint64_t readonly = B_FALSE;
5065	int error;
5066	nvlist_t *nvroot;
5067	nvlist_t **spares, **l2cache;
5068	uint_t nspares, nl2cache;
5069
5070	/*
5071	 * If a pool with this name exists, return failure.
5072	 */
5073	mutex_enter(&spa_namespace_lock);
5074	if (spa_lookup(pool) != NULL) {
5075		mutex_exit(&spa_namespace_lock);
5076		return (SET_ERROR(EEXIST));
5077	}
5078
5079	/*
5080	 * Create and initialize the spa structure.
5081	 */
5082	(void) nvlist_lookup_string(props,
5083	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5084	(void) nvlist_lookup_uint64(props,
5085	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
5086	if (readonly)
5087		mode = FREAD;
5088	spa = spa_add(pool, config, altroot);
5089	spa->spa_import_flags = flags;
5090
5091	/*
5092	 * Verbatim import - Take a pool and insert it into the namespace
5093	 * as if it had been loaded at boot.
5094	 */
5095	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
5096		if (props != NULL)
5097			spa_configfile_set(spa, props, B_FALSE);
5098
5099		spa_write_cachefile(spa, B_FALSE, B_TRUE);
5100		spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
5101		zfs_dbgmsg("spa_import: verbatim import of %s", pool);
5102		mutex_exit(&spa_namespace_lock);
5103		return (0);
5104	}
5105
5106	spa_activate(spa, mode);
5107
5108	/*
5109	 * Don't start async tasks until we know everything is healthy.
5110	 */
5111	spa_async_suspend(spa);
5112
5113	zpool_get_load_policy(config, &policy);
5114	if (policy.zlp_rewind & ZPOOL_DO_REWIND)
5115		state = SPA_LOAD_RECOVER;
5116
5117	spa->spa_config_source = SPA_CONFIG_SRC_TRYIMPORT;
5118
5119	if (state != SPA_LOAD_RECOVER) {
5120		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
5121		zfs_dbgmsg("spa_import: importing %s", pool);
5122	} else {
5123		zfs_dbgmsg("spa_import: importing %s, max_txg=%lld "
5124		    "(RECOVERY MODE)", pool, (longlong_t)policy.zlp_txg);
5125	}
5126	error = spa_load_best(spa, state, policy.zlp_txg, policy.zlp_rewind);
5127
5128	/*
5129	 * Propagate anything learned while loading the pool and pass it
5130	 * back to caller (i.e. rewind info, missing devices, etc).
5131	 */
5132	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
5133	    spa->spa_load_info) == 0);
5134
5135	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5136	/*
5137	 * Toss any existing sparelist, as it doesn't have any validity
5138	 * anymore, and conflicts with spa_has_spare().
5139	 */
5140	if (spa->spa_spares.sav_config) {
5141		nvlist_free(spa->spa_spares.sav_config);
5142		spa->spa_spares.sav_config = NULL;
5143		spa_load_spares(spa);
5144	}
5145	if (spa->spa_l2cache.sav_config) {
5146		nvlist_free(spa->spa_l2cache.sav_config);
5147		spa->spa_l2cache.sav_config = NULL;
5148		spa_load_l2cache(spa);
5149	}
5150
5151	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
5152	    &nvroot) == 0);
5153	if (error == 0)
5154		error = spa_validate_aux(spa, nvroot, -1ULL,
5155		    VDEV_ALLOC_SPARE);
5156	if (error == 0)
5157		error = spa_validate_aux(spa, nvroot, -1ULL,
5158		    VDEV_ALLOC_L2CACHE);
5159	spa_config_exit(spa, SCL_ALL, FTAG);
5160
5161	if (props != NULL)
5162		spa_configfile_set(spa, props, B_FALSE);
5163
5164	if (error != 0 || (props && spa_writeable(spa) &&
5165	    (error = spa_prop_set(spa, props)))) {
5166		spa_unload(spa);
5167		spa_deactivate(spa);
5168		spa_remove(spa);
5169		mutex_exit(&spa_namespace_lock);
5170		return (error);
5171	}
5172
5173	spa_async_resume(spa);
5174
5175	/*
5176	 * Override any spares and level 2 cache devices as specified by
5177	 * the user, as these may have correct device names/devids, etc.
5178	 */
5179	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
5180	    &spares, &nspares) == 0) {
5181		if (spa->spa_spares.sav_config)
5182			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
5183			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
5184		else
5185			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
5186			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
5187		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
5188		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
5189		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5190		spa_load_spares(spa);
5191		spa_config_exit(spa, SCL_ALL, FTAG);
5192		spa->spa_spares.sav_sync = B_TRUE;
5193	}
5194	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
5195	    &l2cache, &nl2cache) == 0) {
5196		if (spa->spa_l2cache.sav_config)
5197			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
5198			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
5199		else
5200			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
5201			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
5202		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
5203		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
5204		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5205		spa_load_l2cache(spa);
5206		spa_config_exit(spa, SCL_ALL, FTAG);
5207		spa->spa_l2cache.sav_sync = B_TRUE;
5208	}
5209
5210	/*
5211	 * Check for any removed devices.
5212	 */
5213	if (spa->spa_autoreplace) {
5214		spa_aux_check_removed(&spa->spa_spares);
5215		spa_aux_check_removed(&spa->spa_l2cache);
5216	}
5217
5218	if (spa_writeable(spa)) {
5219		/*
5220		 * Update the config cache to include the newly-imported pool.
5221		 */
5222		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5223	}
5224
5225	/*
5226	 * It's possible that the pool was expanded while it was exported.
5227	 * We kick off an async task to handle this for us.
5228	 */
5229	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
5230
5231	spa_history_log_version(spa, "import");
5232
5233	spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
5234
5235	mutex_exit(&spa_namespace_lock);
5236
5237	return (0);
5238}
5239
5240nvlist_t *
5241spa_tryimport(nvlist_t *tryconfig)
5242{
5243	nvlist_t *config = NULL;
5244	char *poolname, *cachefile;
5245	spa_t *spa;
5246	uint64_t state;
5247	int error;
5248	zpool_load_policy_t policy;
5249
5250	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
5251		return (NULL);
5252
5253	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
5254		return (NULL);
5255
5256	/*
5257	 * Create and initialize the spa structure.
5258	 */
5259	mutex_enter(&spa_namespace_lock);
5260	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
5261	spa_activate(spa, FREAD);
5262
5263	/*
5264	 * Rewind pool if a max txg was provided.
5265	 */
5266	zpool_get_load_policy(spa->spa_config, &policy);
5267	if (policy.zlp_txg != UINT64_MAX) {
5268		spa->spa_load_max_txg = policy.zlp_txg;
5269		spa->spa_extreme_rewind = B_TRUE;
5270		zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld",
5271		    poolname, (longlong_t)policy.zlp_txg);
5272	} else {
5273		zfs_dbgmsg("spa_tryimport: importing %s", poolname);
5274	}
5275
5276	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_CACHEFILE, &cachefile)
5277	    == 0) {
5278		zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile);
5279		spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
5280	} else {
5281		spa->spa_config_source = SPA_CONFIG_SRC_SCAN;
5282	}
5283
5284	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING);
5285
5286	/*
5287	 * If 'tryconfig' was at least parsable, return the current config.
5288	 */
5289	if (spa->spa_root_vdev != NULL) {
5290		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
5291		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
5292		    poolname) == 0);
5293		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5294		    state) == 0);
5295		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
5296		    spa->spa_uberblock.ub_timestamp) == 0);
5297		VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
5298		    spa->spa_load_info) == 0);
5299
5300		/*
5301		 * If the bootfs property exists on this pool then we
5302		 * copy it out so that external consumers can tell which
5303		 * pools are bootable.
5304		 */
5305		if ((!error || error == EEXIST) && spa->spa_bootfs) {
5306			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5307
5308			/*
5309			 * We have to play games with the name since the
5310			 * pool was opened as TRYIMPORT_NAME.
5311			 */
5312			if (dsl_dsobj_to_dsname(spa_name(spa),
5313			    spa->spa_bootfs, tmpname) == 0) {
5314				char *cp;
5315				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5316
5317				cp = strchr(tmpname, '/');
5318				if (cp == NULL) {
5319					(void) strlcpy(dsname, tmpname,
5320					    MAXPATHLEN);
5321				} else {
5322					(void) snprintf(dsname, MAXPATHLEN,
5323					    "%s/%s", poolname, ++cp);
5324				}
5325				VERIFY(nvlist_add_string(config,
5326				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
5327				kmem_free(dsname, MAXPATHLEN);
5328			}
5329			kmem_free(tmpname, MAXPATHLEN);
5330		}
5331
5332		/*
5333		 * Add the list of hot spares and level 2 cache devices.
5334		 */
5335		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5336		spa_add_spares(spa, config);
5337		spa_add_l2cache(spa, config);
5338		spa_config_exit(spa, SCL_CONFIG, FTAG);
5339	}
5340
5341	spa_unload(spa);
5342	spa_deactivate(spa);
5343	spa_remove(spa);
5344	mutex_exit(&spa_namespace_lock);
5345
5346	return (config);
5347}
5348
5349/*
5350 * Pool export/destroy
5351 *
5352 * The act of destroying or exporting a pool is very simple.  We make sure there
5353 * is no more pending I/O and any references to the pool are gone.  Then, we
5354 * update the pool state and sync all the labels to disk, removing the
5355 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
5356 * we don't sync the labels or remove the configuration cache.
5357 */
5358static int
5359spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
5360    boolean_t force, boolean_t hardforce)
5361{
5362	spa_t *spa;
5363
5364	if (oldconfig)
5365		*oldconfig = NULL;
5366
5367	if (!(spa_mode_global & FWRITE))
5368		return (SET_ERROR(EROFS));
5369
5370	mutex_enter(&spa_namespace_lock);
5371	if ((spa = spa_lookup(pool)) == NULL) {
5372		mutex_exit(&spa_namespace_lock);
5373		return (SET_ERROR(ENOENT));
5374	}
5375
5376	/*
5377	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
5378	 * reacquire the namespace lock, and see if we can export.
5379	 */
5380	spa_open_ref(spa, FTAG);
5381	mutex_exit(&spa_namespace_lock);
5382	spa_async_suspend(spa);
5383	mutex_enter(&spa_namespace_lock);
5384	spa_close(spa, FTAG);
5385
5386	/*
5387	 * The pool will be in core if it's openable,
5388	 * in which case we can modify its state.
5389	 */
5390	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
5391
5392		/*
5393		 * Objsets may be open only because they're dirty, so we
5394		 * have to force it to sync before checking spa_refcnt.
5395		 */
5396		txg_wait_synced(spa->spa_dsl_pool, 0);
5397		spa_evicting_os_wait(spa);
5398
5399		/*
5400		 * A pool cannot be exported or destroyed if there are active
5401		 * references.  If we are resetting a pool, allow references by
5402		 * fault injection handlers.
5403		 */
5404		if (!spa_refcount_zero(spa) ||
5405		    (spa->spa_inject_ref != 0 &&
5406		    new_state != POOL_STATE_UNINITIALIZED)) {
5407			spa_async_resume(spa);
5408			mutex_exit(&spa_namespace_lock);
5409			return (SET_ERROR(EBUSY));
5410		}
5411
5412		/*
5413		 * A pool cannot be exported if it has an active shared spare.
5414		 * This is to prevent other pools stealing the active spare
5415		 * from an exported pool. At user's own will, such pool can
5416		 * be forcedly exported.
5417		 */
5418		if (!force && new_state == POOL_STATE_EXPORTED &&
5419		    spa_has_active_shared_spare(spa)) {
5420			spa_async_resume(spa);
5421			mutex_exit(&spa_namespace_lock);
5422			return (SET_ERROR(EXDEV));
5423		}
5424
5425		/*
5426		 * We're about to export or destroy this pool. Make sure
5427		 * we stop all initializtion activity here before we
5428		 * set the spa_final_txg. This will ensure that all
5429		 * dirty data resulting from the initialization is
5430		 * committed to disk before we unload the pool.
5431		 */
5432		if (spa->spa_root_vdev != NULL) {
5433			vdev_initialize_stop_all(spa->spa_root_vdev,
5434			    VDEV_INITIALIZE_ACTIVE);
5435		}
5436
5437		/*
5438		 * We want this to be reflected on every label,
5439		 * so mark them all dirty.  spa_unload() will do the
5440		 * final sync that pushes these changes out.
5441		 */
5442		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
5443			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5444			spa->spa_state = new_state;
5445			spa->spa_final_txg = spa_last_synced_txg(spa) +
5446			    TXG_DEFER_SIZE + 1;
5447			vdev_config_dirty(spa->spa_root_vdev);
5448			spa_config_exit(spa, SCL_ALL, FTAG);
5449		}
5450	}
5451
5452	spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY);
5453
5454	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5455		spa_unload(spa);
5456		spa_deactivate(spa);
5457	}
5458
5459	if (oldconfig && spa->spa_config)
5460		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
5461
5462	if (new_state != POOL_STATE_UNINITIALIZED) {
5463		if (!hardforce)
5464			spa_write_cachefile(spa, B_TRUE, B_TRUE);
5465		spa_remove(spa);
5466	}
5467	mutex_exit(&spa_namespace_lock);
5468
5469	return (0);
5470}
5471
5472/*
5473 * Destroy a storage pool.
5474 */
5475int
5476spa_destroy(char *pool)
5477{
5478	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
5479	    B_FALSE, B_FALSE));
5480}
5481
5482/*
5483 * Export a storage pool.
5484 */
5485int
5486spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
5487    boolean_t hardforce)
5488{
5489	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
5490	    force, hardforce));
5491}
5492
5493/*
5494 * Similar to spa_export(), this unloads the spa_t without actually removing it
5495 * from the namespace in any way.
5496 */
5497int
5498spa_reset(char *pool)
5499{
5500	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
5501	    B_FALSE, B_FALSE));
5502}
5503
5504/*
5505 * ==========================================================================
5506 * Device manipulation
5507 * ==========================================================================
5508 */
5509
5510/*
5511 * Add a device to a storage pool.
5512 */
5513int
5514spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
5515{
5516	uint64_t txg, id;
5517	int error;
5518	vdev_t *rvd = spa->spa_root_vdev;
5519	vdev_t *vd, *tvd;
5520	nvlist_t **spares, **l2cache;
5521	uint_t nspares, nl2cache;
5522
5523	ASSERT(spa_writeable(spa));
5524
5525	txg = spa_vdev_enter(spa);
5526
5527	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
5528	    VDEV_ALLOC_ADD)) != 0)
5529		return (spa_vdev_exit(spa, NULL, txg, error));
5530
5531	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
5532
5533	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
5534	    &nspares) != 0)
5535		nspares = 0;
5536
5537	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
5538	    &nl2cache) != 0)
5539		nl2cache = 0;
5540
5541	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
5542		return (spa_vdev_exit(spa, vd, txg, EINVAL));
5543
5544	if (vd->vdev_children != 0 &&
5545	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
5546		return (spa_vdev_exit(spa, vd, txg, error));
5547
5548	/*
5549	 * We must validate the spares and l2cache devices after checking the
5550	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
5551	 */
5552	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
5553		return (spa_vdev_exit(spa, vd, txg, error));
5554
5555	/*
5556	 * If we are in the middle of a device removal, we can only add
5557	 * devices which match the existing devices in the pool.
5558	 * If we are in the middle of a removal, or have some indirect
5559	 * vdevs, we can not add raidz toplevels.
5560	 */
5561	if (spa->spa_vdev_removal != NULL ||
5562	    spa->spa_removing_phys.sr_prev_indirect_vdev != -1) {
5563		for (int c = 0; c < vd->vdev_children; c++) {
5564			tvd = vd->vdev_child[c];
5565			if (spa->spa_vdev_removal != NULL &&
5566			    tvd->vdev_ashift != spa->spa_max_ashift) {
5567				return (spa_vdev_exit(spa, vd, txg, EINVAL));
5568			}
5569			/* Fail if top level vdev is raidz */
5570			if (tvd->vdev_ops == &vdev_raidz_ops) {
5571				return (spa_vdev_exit(spa, vd, txg, EINVAL));
5572			}
5573			/*
5574			 * Need the top level mirror to be
5575			 * a mirror of leaf vdevs only
5576			 */
5577			if (tvd->vdev_ops == &vdev_mirror_ops) {
5578				for (uint64_t cid = 0;
5579				    cid < tvd->vdev_children; cid++) {
5580					vdev_t *cvd = tvd->vdev_child[cid];
5581					if (!cvd->vdev_ops->vdev_op_leaf) {
5582						return (spa_vdev_exit(spa, vd,
5583						    txg, EINVAL));
5584					}
5585				}
5586			}
5587		}
5588	}
5589
5590	for (int c = 0; c < vd->vdev_children; c++) {
5591
5592		/*
5593		 * Set the vdev id to the first hole, if one exists.
5594		 */
5595		for (id = 0; id < rvd->vdev_children; id++) {
5596			if (rvd->vdev_child[id]->vdev_ishole) {
5597				vdev_free(rvd->vdev_child[id]);
5598				break;
5599			}
5600		}
5601		tvd = vd->vdev_child[c];
5602		vdev_remove_child(vd, tvd);
5603		tvd->vdev_id = id;
5604		vdev_add_child(rvd, tvd);
5605		vdev_config_dirty(tvd);
5606	}
5607
5608	if (nspares != 0) {
5609		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
5610		    ZPOOL_CONFIG_SPARES);
5611		spa_load_spares(spa);
5612		spa->spa_spares.sav_sync = B_TRUE;
5613	}
5614
5615	if (nl2cache != 0) {
5616		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
5617		    ZPOOL_CONFIG_L2CACHE);
5618		spa_load_l2cache(spa);
5619		spa->spa_l2cache.sav_sync = B_TRUE;
5620	}
5621
5622	/*
5623	 * We have to be careful when adding new vdevs to an existing pool.
5624	 * If other threads start allocating from these vdevs before we
5625	 * sync the config cache, and we lose power, then upon reboot we may
5626	 * fail to open the pool because there are DVAs that the config cache
5627	 * can't translate.  Therefore, we first add the vdevs without
5628	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
5629	 * and then let spa_config_update() initialize the new metaslabs.
5630	 *
5631	 * spa_load() checks for added-but-not-initialized vdevs, so that
5632	 * if we lose power at any point in this sequence, the remaining
5633	 * steps will be completed the next time we load the pool.
5634	 */
5635	(void) spa_vdev_exit(spa, vd, txg, 0);
5636
5637	mutex_enter(&spa_namespace_lock);
5638	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5639	spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD);
5640	mutex_exit(&spa_namespace_lock);
5641
5642	return (0);
5643}
5644
5645/*
5646 * Attach a device to a mirror.  The arguments are the path to any device
5647 * in the mirror, and the nvroot for the new device.  If the path specifies
5648 * a device that is not mirrored, we automatically insert the mirror vdev.
5649 *
5650 * If 'replacing' is specified, the new device is intended to replace the
5651 * existing device; in this case the two devices are made into their own
5652 * mirror using the 'replacing' vdev, which is functionally identical to
5653 * the mirror vdev (it actually reuses all the same ops) but has a few
5654 * extra rules: you can't attach to it after it's been created, and upon
5655 * completion of resilvering, the first disk (the one being replaced)
5656 * is automatically detached.
5657 */
5658int
5659spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
5660{
5661	uint64_t txg, dtl_max_txg;
5662	vdev_t *rvd = spa->spa_root_vdev;
5663	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
5664	vdev_ops_t *pvops;
5665	char *oldvdpath, *newvdpath;
5666	int newvd_isspare;
5667	int error;
5668
5669	ASSERT(spa_writeable(spa));
5670
5671	txg = spa_vdev_enter(spa);
5672
5673	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
5674
5675	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5676	if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
5677		error = (spa_has_checkpoint(spa)) ?
5678		    ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
5679		return (spa_vdev_exit(spa, NULL, txg, error));
5680	}
5681
5682	if (spa->spa_vdev_removal != NULL)
5683		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5684
5685	if (oldvd == NULL)
5686		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
5687
5688	if (!oldvd->vdev_ops->vdev_op_leaf)
5689		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5690
5691	pvd = oldvd->vdev_parent;
5692
5693	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
5694	    VDEV_ALLOC_ATTACH)) != 0)
5695		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5696
5697	if (newrootvd->vdev_children != 1)
5698		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
5699
5700	newvd = newrootvd->vdev_child[0];
5701
5702	if (!newvd->vdev_ops->vdev_op_leaf)
5703		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
5704
5705	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
5706		return (spa_vdev_exit(spa, newrootvd, txg, error));
5707
5708	/*
5709	 * Spares can't replace logs
5710	 */
5711	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
5712		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
5713
5714	if (!replacing) {
5715		/*
5716		 * For attach, the only allowable parent is a mirror or the root
5717		 * vdev.
5718		 */
5719		if (pvd->vdev_ops != &vdev_mirror_ops &&
5720		    pvd->vdev_ops != &vdev_root_ops)
5721			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
5722
5723		pvops = &vdev_mirror_ops;
5724	} else {
5725		/*
5726		 * Active hot spares can only be replaced by inactive hot
5727		 * spares.
5728		 */
5729		if (pvd->vdev_ops == &vdev_spare_ops &&
5730		    oldvd->vdev_isspare &&
5731		    !spa_has_spare(spa, newvd->vdev_guid))
5732			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
5733
5734		/*
5735		 * If the source is a hot spare, and the parent isn't already a
5736		 * spare, then we want to create a new hot spare.  Otherwise, we
5737		 * want to create a replacing vdev.  The user is not allowed to
5738		 * attach to a spared vdev child unless the 'isspare' state is
5739		 * the same (spare replaces spare, non-spare replaces
5740		 * non-spare).
5741		 */
5742		if (pvd->vdev_ops == &vdev_replacing_ops &&
5743		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
5744			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
5745		} else if (pvd->vdev_ops == &vdev_spare_ops &&
5746		    newvd->vdev_isspare != oldvd->vdev_isspare) {
5747			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
5748		}
5749
5750		if (newvd->vdev_isspare)
5751			pvops = &vdev_spare_ops;
5752		else
5753			pvops = &vdev_replacing_ops;
5754	}
5755
5756	/*
5757	 * Make sure the new device is big enough.
5758	 */
5759	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
5760		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
5761
5762	/*
5763	 * The new device cannot have a higher alignment requirement
5764	 * than the top-level vdev.
5765	 */
5766	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
5767		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
5768
5769	/*
5770	 * If this is an in-place replacement, update oldvd's path and devid
5771	 * to make it distinguishable from newvd, and unopenable from now on.
5772	 */
5773	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
5774		spa_strfree(oldvd->vdev_path);
5775		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
5776		    KM_SLEEP);
5777		(void) sprintf(oldvd->vdev_path, "%s/%s",
5778		    newvd->vdev_path, "old");
5779		if (oldvd->vdev_devid != NULL) {
5780			spa_strfree(oldvd->vdev_devid);
5781			oldvd->vdev_devid = NULL;
5782		}
5783	}
5784
5785	/* mark the device being resilvered */
5786	newvd->vdev_resilver_txg = txg;
5787
5788	/*
5789	 * If the parent is not a mirror, or if we're replacing, insert the new
5790	 * mirror/replacing/spare vdev above oldvd.
5791	 */
5792	if (pvd->vdev_ops != pvops)
5793		pvd = vdev_add_parent(oldvd, pvops);
5794
5795	ASSERT(pvd->vdev_top->vdev_parent == rvd);
5796	ASSERT(pvd->vdev_ops == pvops);
5797	ASSERT(oldvd->vdev_parent == pvd);
5798
5799	/*
5800	 * Extract the new device from its root and add it to pvd.
5801	 */
5802	vdev_remove_child(newrootvd, newvd);
5803	newvd->vdev_id = pvd->vdev_children;
5804	newvd->vdev_crtxg = oldvd->vdev_crtxg;
5805	vdev_add_child(pvd, newvd);
5806
5807	tvd = newvd->vdev_top;
5808	ASSERT(pvd->vdev_top == tvd);
5809	ASSERT(tvd->vdev_parent == rvd);
5810
5811	vdev_config_dirty(tvd);
5812
5813	/*
5814	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
5815	 * for any dmu_sync-ed blocks.  It will propagate upward when
5816	 * spa_vdev_exit() calls vdev_dtl_reassess().
5817	 */
5818	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
5819
5820	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
5821	    dtl_max_txg - TXG_INITIAL);
5822
5823	if (newvd->vdev_isspare) {
5824		spa_spare_activate(newvd);
5825		spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE);
5826	}
5827
5828	oldvdpath = spa_strdup(oldvd->vdev_path);
5829	newvdpath = spa_strdup(newvd->vdev_path);
5830	newvd_isspare = newvd->vdev_isspare;
5831
5832	/*
5833	 * Mark newvd's DTL dirty in this txg.
5834	 */
5835	vdev_dirty(tvd, VDD_DTL, newvd, txg);
5836
5837	/*
5838	 * Schedule the resilver to restart in the future. We do this to
5839	 * ensure that dmu_sync-ed blocks have been stitched into the
5840	 * respective datasets.
5841	 */
5842	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
5843
5844	if (spa->spa_bootfs)
5845		spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH);
5846
5847	spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH);
5848
5849	/*
5850	 * Commit the config
5851	 */
5852	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
5853
5854	spa_history_log_internal(spa, "vdev attach", NULL,
5855	    "%s vdev=%s %s vdev=%s",
5856	    replacing && newvd_isspare ? "spare in" :
5857	    replacing ? "replace" : "attach", newvdpath,
5858	    replacing ? "for" : "to", oldvdpath);
5859
5860	spa_strfree(oldvdpath);
5861	spa_strfree(newvdpath);
5862
5863	return (0);
5864}
5865
5866/*
5867 * Detach a device from a mirror or replacing vdev.
5868 *
5869 * If 'replace_done' is specified, only detach if the parent
5870 * is a replacing vdev.
5871 */
5872int
5873spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
5874{
5875	uint64_t txg;
5876	int error;
5877	vdev_t *rvd = spa->spa_root_vdev;
5878	vdev_t *vd, *pvd, *cvd, *tvd;
5879	boolean_t unspare = B_FALSE;
5880	uint64_t unspare_guid = 0;
5881	char *vdpath;
5882
5883	ASSERT(spa_writeable(spa));
5884
5885	txg = spa_vdev_enter(spa);
5886
5887	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5888
5889	/*
5890	 * Besides being called directly from the userland through the
5891	 * ioctl interface, spa_vdev_detach() can be potentially called
5892	 * at the end of spa_vdev_resilver_done().
5893	 *
5894	 * In the regular case, when we have a checkpoint this shouldn't
5895	 * happen as we never empty the DTLs of a vdev during the scrub
5896	 * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done()
5897	 * should never get here when we have a checkpoint.
5898	 *
5899	 * That said, even in a case when we checkpoint the pool exactly
5900	 * as spa_vdev_resilver_done() calls this function everything
5901	 * should be fine as the resilver will return right away.
5902	 */
5903	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5904	if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
5905		error = (spa_has_checkpoint(spa)) ?
5906		    ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
5907		return (spa_vdev_exit(spa, NULL, txg, error));
5908	}
5909
5910	if (vd == NULL)
5911		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
5912
5913	if (!vd->vdev_ops->vdev_op_leaf)
5914		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5915
5916	pvd = vd->vdev_parent;
5917
5918	/*
5919	 * If the parent/child relationship is not as expected, don't do it.
5920	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
5921	 * vdev that's replacing B with C.  The user's intent in replacing
5922	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
5923	 * the replace by detaching C, the expected behavior is to end up
5924	 * M(A,B).  But suppose that right after deciding to detach C,
5925	 * the replacement of B completes.  We would have M(A,C), and then
5926	 * ask to detach C, which would leave us with just A -- not what
5927	 * the user wanted.  To prevent this, we make sure that the
5928	 * parent/child relationship hasn't changed -- in this example,
5929	 * that C's parent is still the replacing vdev R.
5930	 */
5931	if (pvd->vdev_guid != pguid && pguid != 0)
5932		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5933
5934	/*
5935	 * Only 'replacing' or 'spare' vdevs can be replaced.
5936	 */
5937	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
5938	    pvd->vdev_ops != &vdev_spare_ops)
5939		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5940
5941	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
5942	    spa_version(spa) >= SPA_VERSION_SPARES);
5943
5944	/*
5945	 * Only mirror, replacing, and spare vdevs support detach.
5946	 */
5947	if (pvd->vdev_ops != &vdev_replacing_ops &&
5948	    pvd->vdev_ops != &vdev_mirror_ops &&
5949	    pvd->vdev_ops != &vdev_spare_ops)
5950		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5951
5952	/*
5953	 * If this device has the only valid copy of some data,
5954	 * we cannot safely detach it.
5955	 */
5956	if (vdev_dtl_required(vd))
5957		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5958
5959	ASSERT(pvd->vdev_children >= 2);
5960
5961	/*
5962	 * If we are detaching the second disk from a replacing vdev, then
5963	 * check to see if we changed the original vdev's path to have "/old"
5964	 * at the end in spa_vdev_attach().  If so, undo that change now.
5965	 */
5966	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
5967	    vd->vdev_path != NULL) {
5968		size_t len = strlen(vd->vdev_path);
5969
5970		for (int c = 0; c < pvd->vdev_children; c++) {
5971			cvd = pvd->vdev_child[c];
5972
5973			if (cvd == vd || cvd->vdev_path == NULL)
5974				continue;
5975
5976			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
5977			    strcmp(cvd->vdev_path + len, "/old") == 0) {
5978				spa_strfree(cvd->vdev_path);
5979				cvd->vdev_path = spa_strdup(vd->vdev_path);
5980				break;
5981			}
5982		}
5983	}
5984
5985	/*
5986	 * If we are detaching the original disk from a spare, then it implies
5987	 * that the spare should become a real disk, and be removed from the
5988	 * active spare list for the pool.
5989	 */
5990	if (pvd->vdev_ops == &vdev_spare_ops &&
5991	    vd->vdev_id == 0 &&
5992	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
5993		unspare = B_TRUE;
5994
5995	/*
5996	 * Erase the disk labels so the disk can be used for other things.
5997	 * This must be done after all other error cases are handled,
5998	 * but before we disembowel vd (so we can still do I/O to it).
5999	 * But if we can't do it, don't treat the error as fatal --
6000	 * it may be that the unwritability of the disk is the reason
6001	 * it's being detached!
6002	 */
6003	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
6004
6005	/*
6006	 * Remove vd from its parent and compact the parent's children.
6007	 */
6008	vdev_remove_child(pvd, vd);
6009	vdev_compact_children(pvd);
6010
6011	/*
6012	 * Remember one of the remaining children so we can get tvd below.
6013	 */
6014	cvd = pvd->vdev_child[pvd->vdev_children - 1];
6015
6016	/*
6017	 * If we need to remove the remaining child from the list of hot spares,
6018	 * do it now, marking the vdev as no longer a spare in the process.
6019	 * We must do this before vdev_remove_parent(), because that can
6020	 * change the GUID if it creates a new toplevel GUID.  For a similar
6021	 * reason, we must remove the spare now, in the same txg as the detach;
6022	 * otherwise someone could attach a new sibling, change the GUID, and
6023	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
6024	 */
6025	if (unspare) {
6026		ASSERT(cvd->vdev_isspare);
6027		spa_spare_remove(cvd);
6028		unspare_guid = cvd->vdev_guid;
6029		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
6030		cvd->vdev_unspare = B_TRUE;
6031	}
6032
6033	/*
6034	 * If the parent mirror/replacing vdev only has one child,
6035	 * the parent is no longer needed.  Remove it from the tree.
6036	 */
6037	if (pvd->vdev_children == 1) {
6038		if (pvd->vdev_ops == &vdev_spare_ops)
6039			cvd->vdev_unspare = B_FALSE;
6040		vdev_remove_parent(cvd);
6041	}
6042
6043
6044	/*
6045	 * We don't set tvd until now because the parent we just removed
6046	 * may have been the previous top-level vdev.
6047	 */
6048	tvd = cvd->vdev_top;
6049	ASSERT(tvd->vdev_parent == rvd);
6050
6051	/*
6052	 * Reevaluate the parent vdev state.
6053	 */
6054	vdev_propagate_state(cvd);
6055
6056	/*
6057	 * If the 'autoexpand' property is set on the pool then automatically
6058	 * try to expand the size of the pool. For example if the device we
6059	 * just detached was smaller than the others, it may be possible to
6060	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
6061	 * first so that we can obtain the updated sizes of the leaf vdevs.
6062	 */
6063	if (spa->spa_autoexpand) {
6064		vdev_reopen(tvd);
6065		vdev_expand(tvd, txg);
6066	}
6067
6068	vdev_config_dirty(tvd);
6069
6070	/*
6071	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
6072	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
6073	 * But first make sure we're not on any *other* txg's DTL list, to
6074	 * prevent vd from being accessed after it's freed.
6075	 */
6076	vdpath = spa_strdup(vd->vdev_path);
6077	for (int t = 0; t < TXG_SIZE; t++)
6078		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
6079	vd->vdev_detached = B_TRUE;
6080	vdev_dirty(tvd, VDD_DTL, vd, txg);
6081
6082	spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE);
6083
6084	/* hang on to the spa before we release the lock */
6085	spa_open_ref(spa, FTAG);
6086
6087	error = spa_vdev_exit(spa, vd, txg, 0);
6088
6089	spa_history_log_internal(spa, "detach", NULL,
6090	    "vdev=%s", vdpath);
6091	spa_strfree(vdpath);
6092
6093	/*
6094	 * If this was the removal of the original device in a hot spare vdev,
6095	 * then we want to go through and remove the device from the hot spare
6096	 * list of every other pool.
6097	 */
6098	if (unspare) {
6099		spa_t *altspa = NULL;
6100
6101		mutex_enter(&spa_namespace_lock);
6102		while ((altspa = spa_next(altspa)) != NULL) {
6103			if (altspa->spa_state != POOL_STATE_ACTIVE ||
6104			    altspa == spa)
6105				continue;
6106
6107			spa_open_ref(altspa, FTAG);
6108			mutex_exit(&spa_namespace_lock);
6109			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
6110			mutex_enter(&spa_namespace_lock);
6111			spa_close(altspa, FTAG);
6112		}
6113		mutex_exit(&spa_namespace_lock);
6114
6115		/* search the rest of the vdevs for spares to remove */
6116		spa_vdev_resilver_done(spa);
6117	}
6118
6119	/* all done with the spa; OK to release */
6120	mutex_enter(&spa_namespace_lock);
6121	spa_close(spa, FTAG);
6122	mutex_exit(&spa_namespace_lock);
6123
6124	return (error);
6125}
6126
6127int
6128spa_vdev_initialize(spa_t *spa, uint64_t guid, uint64_t cmd_type)
6129{
6130	/*
6131	 * We hold the namespace lock through the whole function
6132	 * to prevent any changes to the pool while we're starting or
6133	 * stopping initialization. The config and state locks are held so that
6134	 * we can properly assess the vdev state before we commit to
6135	 * the initializing operation.
6136	 */
6137	mutex_enter(&spa_namespace_lock);
6138	spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6139
6140	/* Look up vdev and ensure it's a leaf. */
6141	vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE);
6142	if (vd == NULL || vd->vdev_detached) {
6143		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6144		mutex_exit(&spa_namespace_lock);
6145		return (SET_ERROR(ENODEV));
6146	} else if (!vd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(vd)) {
6147		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6148		mutex_exit(&spa_namespace_lock);
6149		return (SET_ERROR(EINVAL));
6150	} else if (!vdev_writeable(vd)) {
6151		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6152		mutex_exit(&spa_namespace_lock);
6153		return (SET_ERROR(EROFS));
6154	}
6155	mutex_enter(&vd->vdev_initialize_lock);
6156	spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6157
6158	/*
6159	 * When we activate an initialize action we check to see
6160	 * if the vdev_initialize_thread is NULL. We do this instead
6161	 * of using the vdev_initialize_state since there might be
6162	 * a previous initialization process which has completed but
6163	 * the thread is not exited.
6164	 */
6165	if (cmd_type == POOL_INITIALIZE_DO &&
6166	    (vd->vdev_initialize_thread != NULL ||
6167	    vd->vdev_top->vdev_removing)) {
6168		mutex_exit(&vd->vdev_initialize_lock);
6169		mutex_exit(&spa_namespace_lock);
6170		return (SET_ERROR(EBUSY));
6171	} else if (cmd_type == POOL_INITIALIZE_CANCEL &&
6172	    (vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE &&
6173	    vd->vdev_initialize_state != VDEV_INITIALIZE_SUSPENDED)) {
6174		mutex_exit(&vd->vdev_initialize_lock);
6175		mutex_exit(&spa_namespace_lock);
6176		return (SET_ERROR(ESRCH));
6177	} else if (cmd_type == POOL_INITIALIZE_SUSPEND &&
6178	    vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE) {
6179		mutex_exit(&vd->vdev_initialize_lock);
6180		mutex_exit(&spa_namespace_lock);
6181		return (SET_ERROR(ESRCH));
6182	}
6183
6184	switch (cmd_type) {
6185	case POOL_INITIALIZE_DO:
6186		vdev_initialize(vd);
6187		break;
6188	case POOL_INITIALIZE_CANCEL:
6189		vdev_initialize_stop(vd, VDEV_INITIALIZE_CANCELED);
6190		break;
6191	case POOL_INITIALIZE_SUSPEND:
6192		vdev_initialize_stop(vd, VDEV_INITIALIZE_SUSPENDED);
6193		break;
6194	default:
6195		panic("invalid cmd_type %llu", (unsigned long long)cmd_type);
6196	}
6197	mutex_exit(&vd->vdev_initialize_lock);
6198
6199	/* Sync out the initializing state */
6200	txg_wait_synced(spa->spa_dsl_pool, 0);
6201	mutex_exit(&spa_namespace_lock);
6202
6203	return (0);
6204}
6205
6206
6207/*
6208 * Split a set of devices from their mirrors, and create a new pool from them.
6209 */
6210int
6211spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
6212    nvlist_t *props, boolean_t exp)
6213{
6214	int error = 0;
6215	uint64_t txg, *glist;
6216	spa_t *newspa;
6217	uint_t c, children, lastlog;
6218	nvlist_t **child, *nvl, *tmp;
6219	dmu_tx_t *tx;
6220	char *altroot = NULL;
6221	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
6222	boolean_t activate_slog;
6223
6224	ASSERT(spa_writeable(