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, 2017 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 */
31
32/*
33 * SPA: Storage Pool Allocator
34 *
35 * This file contains all the routines used when modifying on-disk SPA state.
36 * This includes opening, importing, destroying, exporting a pool, and syncing a
37 * pool.
38 */
39
40#include <sys/zfs_context.h>
41#include <sys/fm/fs/zfs.h>
42#include <sys/spa_impl.h>
43#include <sys/zio.h>
44#include <sys/zio_checksum.h>
45#include <sys/dmu.h>
46#include <sys/dmu_tx.h>
47#include <sys/zap.h>
48#include <sys/zil.h>
49#include <sys/ddt.h>
50#include <sys/vdev_impl.h>
51#include <sys/metaslab.h>
52#include <sys/metaslab_impl.h>
53#include <sys/uberblock_impl.h>
54#include <sys/txg.h>
55#include <sys/avl.h>
56#include <sys/dmu_traverse.h>
57#include <sys/dmu_objset.h>
58#include <sys/unique.h>
59#include <sys/dsl_pool.h>
60#include <sys/dsl_dataset.h>
61#include <sys/dsl_dir.h>
62#include <sys/dsl_prop.h>
63#include <sys/dsl_synctask.h>
64#include <sys/fs/zfs.h>
65#include <sys/arc.h>
66#include <sys/callb.h>
67#include <sys/systeminfo.h>
68#include <sys/spa_boot.h>
69#include <sys/zfs_ioctl.h>
70#include <sys/dsl_scan.h>
71#include <sys/zfeature.h>
72#include <sys/dsl_destroy.h>
73#include <sys/abd.h>
74
75#ifdef	_KERNEL
76#include <sys/bootprops.h>
77#include <sys/callb.h>
78#include <sys/cpupart.h>
79#include <sys/pool.h>
80#include <sys/sysdc.h>
81#include <sys/zone.h>
82#endif	/* _KERNEL */
83
84#include "zfs_prop.h"
85#include "zfs_comutil.h"
86
87/*
88 * The interval, in seconds, at which failed configuration cache file writes
89 * should be retried.
90 */
91static int zfs_ccw_retry_interval = 300;
92
93typedef enum zti_modes {
94	ZTI_MODE_FIXED,			/* value is # of threads (min 1) */
95	ZTI_MODE_BATCH,			/* cpu-intensive; value is ignored */
96	ZTI_MODE_NULL,			/* don't create a taskq */
97	ZTI_NMODES
98} zti_modes_t;
99
100#define	ZTI_P(n, q)	{ ZTI_MODE_FIXED, (n), (q) }
101#define	ZTI_BATCH	{ ZTI_MODE_BATCH, 0, 1 }
102#define	ZTI_NULL	{ ZTI_MODE_NULL, 0, 0 }
103
104#define	ZTI_N(n)	ZTI_P(n, 1)
105#define	ZTI_ONE		ZTI_N(1)
106
107typedef struct zio_taskq_info {
108	zti_modes_t zti_mode;
109	uint_t zti_value;
110	uint_t zti_count;
111} zio_taskq_info_t;
112
113static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
114	"issue", "issue_high", "intr", "intr_high"
115};
116
117/*
118 * This table defines the taskq settings for each ZFS I/O type. When
119 * initializing a pool, we use this table to create an appropriately sized
120 * taskq. Some operations are low volume and therefore have a small, static
121 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
122 * macros. Other operations process a large amount of data; the ZTI_BATCH
123 * macro causes us to create a taskq oriented for throughput. Some operations
124 * are so high frequency and short-lived that the taskq itself can become a a
125 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
126 * additional degree of parallelism specified by the number of threads per-
127 * taskq and the number of taskqs; when dispatching an event in this case, the
128 * particular taskq is chosen at random.
129 *
130 * The different taskq priorities are to handle the different contexts (issue
131 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
132 * need to be handled with minimum delay.
133 */
134const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
135	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
136	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* NULL */
137	{ ZTI_N(8),	ZTI_NULL,	ZTI_P(12, 8),	ZTI_NULL }, /* READ */
138	{ ZTI_BATCH,	ZTI_N(5),	ZTI_N(8),	ZTI_N(5) }, /* WRITE */
139	{ ZTI_P(12, 8),	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* FREE */
140	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* CLAIM */
141	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* IOCTL */
142};
143
144static sysevent_t *spa_event_create(spa_t *spa, vdev_t *vd, const char *name);
145static void spa_event_post(sysevent_t *ev);
146static void spa_sync_version(void *arg, dmu_tx_t *tx);
147static void spa_sync_props(void *arg, dmu_tx_t *tx);
148static boolean_t spa_has_active_shared_spare(spa_t *spa);
149static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
150    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
151    char **ereport);
152static void spa_vdev_resilver_done(spa_t *spa);
153
154uint_t		zio_taskq_batch_pct = 75;	/* 1 thread per cpu in pset */
155id_t		zio_taskq_psrset_bind = PS_NONE;
156boolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
157uint_t		zio_taskq_basedc = 80;		/* base duty cycle */
158
159boolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
160extern int	zfs_sync_pass_deferred_free;
161
162/*
163 * This (illegal) pool name is used when temporarily importing a spa_t in order
164 * to get the vdev stats associated with the imported devices.
165 */
166#define	TRYIMPORT_NAME	"$import"
167
168/*
169 * ==========================================================================
170 * SPA properties routines
171 * ==========================================================================
172 */
173
174/*
175 * Add a (source=src, propname=propval) list to an nvlist.
176 */
177static void
178spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
179    uint64_t intval, zprop_source_t src)
180{
181	const char *propname = zpool_prop_to_name(prop);
182	nvlist_t *propval;
183
184	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
185	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
186
187	if (strval != NULL)
188		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
189	else
190		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
191
192	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
193	nvlist_free(propval);
194}
195
196/*
197 * Get property values from the spa configuration.
198 */
199static void
200spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
201{
202	vdev_t *rvd = spa->spa_root_vdev;
203	dsl_pool_t *pool = spa->spa_dsl_pool;
204	uint64_t size, alloc, cap, version;
205	zprop_source_t src = ZPROP_SRC_NONE;
206	spa_config_dirent_t *dp;
207	metaslab_class_t *mc = spa_normal_class(spa);
208
209	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
210
211	if (rvd != NULL) {
212		alloc = metaslab_class_get_alloc(spa_normal_class(spa));
213		size = metaslab_class_get_space(spa_normal_class(spa));
214		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
215		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
216		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
217		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
218		    size - alloc, src);
219
220		spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
221		    metaslab_class_fragmentation(mc), src);
222		spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
223		    metaslab_class_expandable_space(mc), src);
224		spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
225		    (spa_mode(spa) == FREAD), src);
226
227		cap = (size == 0) ? 0 : (alloc * 100 / size);
228		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
229
230		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
231		    ddt_get_pool_dedup_ratio(spa), src);
232
233		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
234		    rvd->vdev_state, src);
235
236		version = spa_version(spa);
237		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
238			src = ZPROP_SRC_DEFAULT;
239		else
240			src = ZPROP_SRC_LOCAL;
241		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
242	}
243
244	if (pool != NULL) {
245		/*
246		 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
247		 * when opening pools before this version freedir will be NULL.
248		 */
249		if (pool->dp_free_dir != NULL) {
250			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
251			    dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
252			    src);
253		} else {
254			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
255			    NULL, 0, src);
256		}
257
258		if (pool->dp_leak_dir != NULL) {
259			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
260			    dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
261			    src);
262		} else {
263			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
264			    NULL, 0, src);
265		}
266	}
267
268	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
269
270	if (spa->spa_comment != NULL) {
271		spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
272		    0, ZPROP_SRC_LOCAL);
273	}
274
275	if (spa->spa_root != NULL)
276		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
277		    0, ZPROP_SRC_LOCAL);
278
279	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
280		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
281		    MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
282	} else {
283		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
284		    SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
285	}
286
287	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
288		if (dp->scd_path == NULL) {
289			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
290			    "none", 0, ZPROP_SRC_LOCAL);
291		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
292			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
293			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
294		}
295	}
296}
297
298/*
299 * Get zpool property values.
300 */
301int
302spa_prop_get(spa_t *spa, nvlist_t **nvp)
303{
304	objset_t *mos = spa->spa_meta_objset;
305	zap_cursor_t zc;
306	zap_attribute_t za;
307	int err;
308
309	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
310
311	mutex_enter(&spa->spa_props_lock);
312
313	/*
314	 * Get properties from the spa config.
315	 */
316	spa_prop_get_config(spa, nvp);
317
318	/* If no pool property object, no more prop to get. */
319	if (mos == NULL || spa->spa_pool_props_object == 0) {
320		mutex_exit(&spa->spa_props_lock);
321		return (0);
322	}
323
324	/*
325	 * Get properties from the MOS pool property object.
326	 */
327	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
328	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
329	    zap_cursor_advance(&zc)) {
330		uint64_t intval = 0;
331		char *strval = NULL;
332		zprop_source_t src = ZPROP_SRC_DEFAULT;
333		zpool_prop_t prop;
334
335		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
336			continue;
337
338		switch (za.za_integer_length) {
339		case 8:
340			/* integer property */
341			if (za.za_first_integer !=
342			    zpool_prop_default_numeric(prop))
343				src = ZPROP_SRC_LOCAL;
344
345			if (prop == ZPOOL_PROP_BOOTFS) {
346				dsl_pool_t *dp;
347				dsl_dataset_t *ds = NULL;
348
349				dp = spa_get_dsl(spa);
350				dsl_pool_config_enter(dp, FTAG);
351				if (err = dsl_dataset_hold_obj(dp,
352				    za.za_first_integer, FTAG, &ds)) {
353					dsl_pool_config_exit(dp, FTAG);
354					break;
355				}
356
357				strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN,
358				    KM_SLEEP);
359				dsl_dataset_name(ds, strval);
360				dsl_dataset_rele(ds, FTAG);
361				dsl_pool_config_exit(dp, FTAG);
362			} else {
363				strval = NULL;
364				intval = za.za_first_integer;
365			}
366
367			spa_prop_add_list(*nvp, prop, strval, intval, src);
368
369			if (strval != NULL)
370				kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN);
371
372			break;
373
374		case 1:
375			/* string property */
376			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
377			err = zap_lookup(mos, spa->spa_pool_props_object,
378			    za.za_name, 1, za.za_num_integers, strval);
379			if (err) {
380				kmem_free(strval, za.za_num_integers);
381				break;
382			}
383			spa_prop_add_list(*nvp, prop, strval, 0, src);
384			kmem_free(strval, za.za_num_integers);
385			break;
386
387		default:
388			break;
389		}
390	}
391	zap_cursor_fini(&zc);
392	mutex_exit(&spa->spa_props_lock);
393out:
394	if (err && err != ENOENT) {
395		nvlist_free(*nvp);
396		*nvp = NULL;
397		return (err);
398	}
399
400	return (0);
401}
402
403/*
404 * Validate the given pool properties nvlist and modify the list
405 * for the property values to be set.
406 */
407static int
408spa_prop_validate(spa_t *spa, nvlist_t *props)
409{
410	nvpair_t *elem;
411	int error = 0, reset_bootfs = 0;
412	uint64_t objnum = 0;
413	boolean_t has_feature = B_FALSE;
414
415	elem = NULL;
416	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
417		uint64_t intval;
418		char *strval, *slash, *check, *fname;
419		const char *propname = nvpair_name(elem);
420		zpool_prop_t prop = zpool_name_to_prop(propname);
421
422		switch (prop) {
423		case ZPROP_INVAL:
424			if (!zpool_prop_feature(propname)) {
425				error = SET_ERROR(EINVAL);
426				break;
427			}
428
429			/*
430			 * Sanitize the input.
431			 */
432			if (nvpair_type(elem) != DATA_TYPE_UINT64) {
433				error = SET_ERROR(EINVAL);
434				break;
435			}
436
437			if (nvpair_value_uint64(elem, &intval) != 0) {
438				error = SET_ERROR(EINVAL);
439				break;
440			}
441
442			if (intval != 0) {
443				error = SET_ERROR(EINVAL);
444				break;
445			}
446
447			fname = strchr(propname, '@') + 1;
448			if (zfeature_lookup_name(fname, NULL) != 0) {
449				error = SET_ERROR(EINVAL);
450				break;
451			}
452
453			has_feature = B_TRUE;
454			break;
455
456		case ZPOOL_PROP_VERSION:
457			error = nvpair_value_uint64(elem, &intval);
458			if (!error &&
459			    (intval < spa_version(spa) ||
460			    intval > SPA_VERSION_BEFORE_FEATURES ||
461			    has_feature))
462				error = SET_ERROR(EINVAL);
463			break;
464
465		case ZPOOL_PROP_DELEGATION:
466		case ZPOOL_PROP_AUTOREPLACE:
467		case ZPOOL_PROP_LISTSNAPS:
468		case ZPOOL_PROP_AUTOEXPAND:
469			error = nvpair_value_uint64(elem, &intval);
470			if (!error && intval > 1)
471				error = SET_ERROR(EINVAL);
472			break;
473
474		case ZPOOL_PROP_BOOTFS:
475			/*
476			 * If the pool version is less than SPA_VERSION_BOOTFS,
477			 * or the pool is still being created (version == 0),
478			 * the bootfs property cannot be set.
479			 */
480			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
481				error = SET_ERROR(ENOTSUP);
482				break;
483			}
484
485			/*
486			 * Make sure the vdev config is bootable
487			 */
488			if (!vdev_is_bootable(spa->spa_root_vdev)) {
489				error = SET_ERROR(ENOTSUP);
490				break;
491			}
492
493			reset_bootfs = 1;
494
495			error = nvpair_value_string(elem, &strval);
496
497			if (!error) {
498				objset_t *os;
499				uint64_t propval;
500
501				if (strval == NULL || strval[0] == '\0') {
502					objnum = zpool_prop_default_numeric(
503					    ZPOOL_PROP_BOOTFS);
504					break;
505				}
506
507				if (error = dmu_objset_hold(strval, FTAG, &os))
508					break;
509
510				/*
511				 * Must be ZPL, and its property settings
512				 * must be supported by GRUB (compression
513				 * is not gzip, and large blocks are not used).
514				 */
515
516				if (dmu_objset_type(os) != DMU_OST_ZFS) {
517					error = SET_ERROR(ENOTSUP);
518				} else if ((error =
519				    dsl_prop_get_int_ds(dmu_objset_ds(os),
520				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
521				    &propval)) == 0 &&
522				    !BOOTFS_COMPRESS_VALID(propval)) {
523					error = SET_ERROR(ENOTSUP);
524				} else {
525					objnum = dmu_objset_id(os);
526				}
527				dmu_objset_rele(os, FTAG);
528			}
529			break;
530
531		case ZPOOL_PROP_FAILUREMODE:
532			error = nvpair_value_uint64(elem, &intval);
533			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
534			    intval > ZIO_FAILURE_MODE_PANIC))
535				error = SET_ERROR(EINVAL);
536
537			/*
538			 * This is a special case which only occurs when
539			 * the pool has completely failed. This allows
540			 * the user to change the in-core failmode property
541			 * without syncing it out to disk (I/Os might
542			 * currently be blocked). We do this by returning
543			 * EIO to the caller (spa_prop_set) to trick it
544			 * into thinking we encountered a property validation
545			 * error.
546			 */
547			if (!error && spa_suspended(spa)) {
548				spa->spa_failmode = intval;
549				error = SET_ERROR(EIO);
550			}
551			break;
552
553		case ZPOOL_PROP_CACHEFILE:
554			if ((error = nvpair_value_string(elem, &strval)) != 0)
555				break;
556
557			if (strval[0] == '\0')
558				break;
559
560			if (strcmp(strval, "none") == 0)
561				break;
562
563			if (strval[0] != '/') {
564				error = SET_ERROR(EINVAL);
565				break;
566			}
567
568			slash = strrchr(strval, '/');
569			ASSERT(slash != NULL);
570
571			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
572			    strcmp(slash, "/..") == 0)
573				error = SET_ERROR(EINVAL);
574			break;
575
576		case ZPOOL_PROP_COMMENT:
577			if ((error = nvpair_value_string(elem, &strval)) != 0)
578				break;
579			for (check = strval; *check != '\0'; check++) {
580				/*
581				 * The kernel doesn't have an easy isprint()
582				 * check.  For this kernel check, we merely
583				 * check ASCII apart from DEL.  Fix this if
584				 * there is an easy-to-use kernel isprint().
585				 */
586				if (*check >= 0x7f) {
587					error = SET_ERROR(EINVAL);
588					break;
589				}
590			}
591			if (strlen(strval) > ZPROP_MAX_COMMENT)
592				error = E2BIG;
593			break;
594
595		case ZPOOL_PROP_DEDUPDITTO:
596			if (spa_version(spa) < SPA_VERSION_DEDUP)
597				error = SET_ERROR(ENOTSUP);
598			else
599				error = nvpair_value_uint64(elem, &intval);
600			if (error == 0 &&
601			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
602				error = SET_ERROR(EINVAL);
603			break;
604		}
605
606		if (error)
607			break;
608	}
609
610	if (!error && reset_bootfs) {
611		error = nvlist_remove(props,
612		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
613
614		if (!error) {
615			error = nvlist_add_uint64(props,
616			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
617		}
618	}
619
620	return (error);
621}
622
623void
624spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
625{
626	char *cachefile;
627	spa_config_dirent_t *dp;
628
629	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
630	    &cachefile) != 0)
631		return;
632
633	dp = kmem_alloc(sizeof (spa_config_dirent_t),
634	    KM_SLEEP);
635
636	if (cachefile[0] == '\0')
637		dp->scd_path = spa_strdup(spa_config_path);
638	else if (strcmp(cachefile, "none") == 0)
639		dp->scd_path = NULL;
640	else
641		dp->scd_path = spa_strdup(cachefile);
642
643	list_insert_head(&spa->spa_config_list, dp);
644	if (need_sync)
645		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
646}
647
648int
649spa_prop_set(spa_t *spa, nvlist_t *nvp)
650{
651	int error;
652	nvpair_t *elem = NULL;
653	boolean_t need_sync = B_FALSE;
654
655	if ((error = spa_prop_validate(spa, nvp)) != 0)
656		return (error);
657
658	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
659		zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
660
661		if (prop == ZPOOL_PROP_CACHEFILE ||
662		    prop == ZPOOL_PROP_ALTROOT ||
663		    prop == ZPOOL_PROP_READONLY)
664			continue;
665
666		if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
667			uint64_t ver;
668
669			if (prop == ZPOOL_PROP_VERSION) {
670				VERIFY(nvpair_value_uint64(elem, &ver) == 0);
671			} else {
672				ASSERT(zpool_prop_feature(nvpair_name(elem)));
673				ver = SPA_VERSION_FEATURES;
674				need_sync = B_TRUE;
675			}
676
677			/* Save time if the version is already set. */
678			if (ver == spa_version(spa))
679				continue;
680
681			/*
682			 * In addition to the pool directory object, we might
683			 * create the pool properties object, the features for
684			 * read object, the features for write object, or the
685			 * feature descriptions object.
686			 */
687			error = dsl_sync_task(spa->spa_name, NULL,
688			    spa_sync_version, &ver,
689			    6, ZFS_SPACE_CHECK_RESERVED);
690			if (error)
691				return (error);
692			continue;
693		}
694
695		need_sync = B_TRUE;
696		break;
697	}
698
699	if (need_sync) {
700		return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
701		    nvp, 6, ZFS_SPACE_CHECK_RESERVED));
702	}
703
704	return (0);
705}
706
707/*
708 * If the bootfs property value is dsobj, clear it.
709 */
710void
711spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
712{
713	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
714		VERIFY(zap_remove(spa->spa_meta_objset,
715		    spa->spa_pool_props_object,
716		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
717		spa->spa_bootfs = 0;
718	}
719}
720
721/*ARGSUSED*/
722static int
723spa_change_guid_check(void *arg, dmu_tx_t *tx)
724{
725	uint64_t *newguid = arg;
726	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
727	vdev_t *rvd = spa->spa_root_vdev;
728	uint64_t vdev_state;
729
730	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
731	vdev_state = rvd->vdev_state;
732	spa_config_exit(spa, SCL_STATE, FTAG);
733
734	if (vdev_state != VDEV_STATE_HEALTHY)
735		return (SET_ERROR(ENXIO));
736
737	ASSERT3U(spa_guid(spa), !=, *newguid);
738
739	return (0);
740}
741
742static void
743spa_change_guid_sync(void *arg, dmu_tx_t *tx)
744{
745	uint64_t *newguid = arg;
746	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
747	uint64_t oldguid;
748	vdev_t *rvd = spa->spa_root_vdev;
749
750	oldguid = spa_guid(spa);
751
752	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
753	rvd->vdev_guid = *newguid;
754	rvd->vdev_guid_sum += (*newguid - oldguid);
755	vdev_config_dirty(rvd);
756	spa_config_exit(spa, SCL_STATE, FTAG);
757
758	spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
759	    oldguid, *newguid);
760}
761
762/*
763 * Change the GUID for the pool.  This is done so that we can later
764 * re-import a pool built from a clone of our own vdevs.  We will modify
765 * the root vdev's guid, our own pool guid, and then mark all of our
766 * vdevs dirty.  Note that we must make sure that all our vdevs are
767 * online when we do this, or else any vdevs that weren't present
768 * would be orphaned from our pool.  We are also going to issue a
769 * sysevent to update any watchers.
770 */
771int
772spa_change_guid(spa_t *spa)
773{
774	int error;
775	uint64_t guid;
776
777	mutex_enter(&spa->spa_vdev_top_lock);
778	mutex_enter(&spa_namespace_lock);
779	guid = spa_generate_guid(NULL);
780
781	error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
782	    spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
783
784	if (error == 0) {
785		spa_config_sync(spa, B_FALSE, B_TRUE);
786		spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
787	}
788
789	mutex_exit(&spa_namespace_lock);
790	mutex_exit(&spa->spa_vdev_top_lock);
791
792	return (error);
793}
794
795/*
796 * ==========================================================================
797 * SPA state manipulation (open/create/destroy/import/export)
798 * ==========================================================================
799 */
800
801static int
802spa_error_entry_compare(const void *a, const void *b)
803{
804	spa_error_entry_t *sa = (spa_error_entry_t *)a;
805	spa_error_entry_t *sb = (spa_error_entry_t *)b;
806	int ret;
807
808	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
809	    sizeof (zbookmark_phys_t));
810
811	if (ret < 0)
812		return (-1);
813	else if (ret > 0)
814		return (1);
815	else
816		return (0);
817}
818
819/*
820 * Utility function which retrieves copies of the current logs and
821 * re-initializes them in the process.
822 */
823void
824spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
825{
826	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
827
828	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
829	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
830
831	avl_create(&spa->spa_errlist_scrub,
832	    spa_error_entry_compare, sizeof (spa_error_entry_t),
833	    offsetof(spa_error_entry_t, se_avl));
834	avl_create(&spa->spa_errlist_last,
835	    spa_error_entry_compare, sizeof (spa_error_entry_t),
836	    offsetof(spa_error_entry_t, se_avl));
837}
838
839static void
840spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
841{
842	const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
843	enum zti_modes mode = ztip->zti_mode;
844	uint_t value = ztip->zti_value;
845	uint_t count = ztip->zti_count;
846	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
847	char name[32];
848	uint_t flags = 0;
849	boolean_t batch = B_FALSE;
850
851	if (mode == ZTI_MODE_NULL) {
852		tqs->stqs_count = 0;
853		tqs->stqs_taskq = NULL;
854		return;
855	}
856
857	ASSERT3U(count, >, 0);
858
859	tqs->stqs_count = count;
860	tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
861
862	switch (mode) {
863	case ZTI_MODE_FIXED:
864		ASSERT3U(value, >=, 1);
865		value = MAX(value, 1);
866		break;
867
868	case ZTI_MODE_BATCH:
869		batch = B_TRUE;
870		flags |= TASKQ_THREADS_CPU_PCT;
871		value = zio_taskq_batch_pct;
872		break;
873
874	default:
875		panic("unrecognized mode for %s_%s taskq (%u:%u) in "
876		    "spa_activate()",
877		    zio_type_name[t], zio_taskq_types[q], mode, value);
878		break;
879	}
880
881	for (uint_t i = 0; i < count; i++) {
882		taskq_t *tq;
883
884		if (count > 1) {
885			(void) snprintf(name, sizeof (name), "%s_%s_%u",
886			    zio_type_name[t], zio_taskq_types[q], i);
887		} else {
888			(void) snprintf(name, sizeof (name), "%s_%s",
889			    zio_type_name[t], zio_taskq_types[q]);
890		}
891
892		if (zio_taskq_sysdc && spa->spa_proc != &p0) {
893			if (batch)
894				flags |= TASKQ_DC_BATCH;
895
896			tq = taskq_create_sysdc(name, value, 50, INT_MAX,
897			    spa->spa_proc, zio_taskq_basedc, flags);
898		} else {
899			pri_t pri = maxclsyspri;
900			/*
901			 * The write issue taskq can be extremely CPU
902			 * intensive.  Run it at slightly lower priority
903			 * than the other taskqs.
904			 */
905			if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
906				pri--;
907
908			tq = taskq_create_proc(name, value, pri, 50,
909			    INT_MAX, spa->spa_proc, flags);
910		}
911
912		tqs->stqs_taskq[i] = tq;
913	}
914}
915
916static void
917spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
918{
919	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
920
921	if (tqs->stqs_taskq == NULL) {
922		ASSERT0(tqs->stqs_count);
923		return;
924	}
925
926	for (uint_t i = 0; i < tqs->stqs_count; i++) {
927		ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
928		taskq_destroy(tqs->stqs_taskq[i]);
929	}
930
931	kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
932	tqs->stqs_taskq = NULL;
933}
934
935/*
936 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
937 * Note that a type may have multiple discrete taskqs to avoid lock contention
938 * on the taskq itself. In that case we choose which taskq at random by using
939 * the low bits of gethrtime().
940 */
941void
942spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
943    task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
944{
945	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
946	taskq_t *tq;
947
948	ASSERT3P(tqs->stqs_taskq, !=, NULL);
949	ASSERT3U(tqs->stqs_count, !=, 0);
950
951	if (tqs->stqs_count == 1) {
952		tq = tqs->stqs_taskq[0];
953	} else {
954		tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
955	}
956
957	taskq_dispatch_ent(tq, func, arg, flags, ent);
958}
959
960static void
961spa_create_zio_taskqs(spa_t *spa)
962{
963	for (int t = 0; t < ZIO_TYPES; t++) {
964		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
965			spa_taskqs_init(spa, t, q);
966		}
967	}
968}
969
970#ifdef _KERNEL
971static void
972spa_thread(void *arg)
973{
974	callb_cpr_t cprinfo;
975
976	spa_t *spa = arg;
977	user_t *pu = PTOU(curproc);
978
979	CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
980	    spa->spa_name);
981
982	ASSERT(curproc != &p0);
983	(void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
984	    "zpool-%s", spa->spa_name);
985	(void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
986
987	/* bind this thread to the requested psrset */
988	if (zio_taskq_psrset_bind != PS_NONE) {
989		pool_lock();
990		mutex_enter(&cpu_lock);
991		mutex_enter(&pidlock);
992		mutex_enter(&curproc->p_lock);
993
994		if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
995		    0, NULL, NULL) == 0)  {
996			curthread->t_bind_pset = zio_taskq_psrset_bind;
997		} else {
998			cmn_err(CE_WARN,
999			    "Couldn't bind process for zfs pool \"%s\" to "
1000			    "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1001		}
1002
1003		mutex_exit(&curproc->p_lock);
1004		mutex_exit(&pidlock);
1005		mutex_exit(&cpu_lock);
1006		pool_unlock();
1007	}
1008
1009	if (zio_taskq_sysdc) {
1010		sysdc_thread_enter(curthread, 100, 0);
1011	}
1012
1013	spa->spa_proc = curproc;
1014	spa->spa_did = curthread->t_did;
1015
1016	spa_create_zio_taskqs(spa);
1017
1018	mutex_enter(&spa->spa_proc_lock);
1019	ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1020
1021	spa->spa_proc_state = SPA_PROC_ACTIVE;
1022	cv_broadcast(&spa->spa_proc_cv);
1023
1024	CALLB_CPR_SAFE_BEGIN(&cprinfo);
1025	while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1026		cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1027	CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1028
1029	ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1030	spa->spa_proc_state = SPA_PROC_GONE;
1031	spa->spa_proc = &p0;
1032	cv_broadcast(&spa->spa_proc_cv);
1033	CALLB_CPR_EXIT(&cprinfo);	/* drops spa_proc_lock */
1034
1035	mutex_enter(&curproc->p_lock);
1036	lwp_exit();
1037}
1038#endif
1039
1040/*
1041 * Activate an uninitialized pool.
1042 */
1043static void
1044spa_activate(spa_t *spa, int mode)
1045{
1046	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1047
1048	spa->spa_state = POOL_STATE_ACTIVE;
1049	spa->spa_mode = mode;
1050
1051	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1052	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1053
1054	/* Try to create a covering process */
1055	mutex_enter(&spa->spa_proc_lock);
1056	ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1057	ASSERT(spa->spa_proc == &p0);
1058	spa->spa_did = 0;
1059
1060	/* Only create a process if we're going to be around a while. */
1061	if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1062		if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1063		    NULL, 0) == 0) {
1064			spa->spa_proc_state = SPA_PROC_CREATED;
1065			while (spa->spa_proc_state == SPA_PROC_CREATED) {
1066				cv_wait(&spa->spa_proc_cv,
1067				    &spa->spa_proc_lock);
1068			}
1069			ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1070			ASSERT(spa->spa_proc != &p0);
1071			ASSERT(spa->spa_did != 0);
1072		} else {
1073#ifdef _KERNEL
1074			cmn_err(CE_WARN,
1075			    "Couldn't create process for zfs pool \"%s\"\n",
1076			    spa->spa_name);
1077#endif
1078		}
1079	}
1080	mutex_exit(&spa->spa_proc_lock);
1081
1082	/* If we didn't create a process, we need to create our taskqs. */
1083	if (spa->spa_proc == &p0) {
1084		spa_create_zio_taskqs(spa);
1085	}
1086
1087	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1088	    offsetof(vdev_t, vdev_config_dirty_node));
1089	list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1090	    offsetof(objset_t, os_evicting_node));
1091	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1092	    offsetof(vdev_t, vdev_state_dirty_node));
1093
1094	txg_list_create(&spa->spa_vdev_txg_list, spa,
1095	    offsetof(struct vdev, vdev_txg_node));
1096
1097	avl_create(&spa->spa_errlist_scrub,
1098	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1099	    offsetof(spa_error_entry_t, se_avl));
1100	avl_create(&spa->spa_errlist_last,
1101	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1102	    offsetof(spa_error_entry_t, se_avl));
1103}
1104
1105/*
1106 * Opposite of spa_activate().
1107 */
1108static void
1109spa_deactivate(spa_t *spa)
1110{
1111	ASSERT(spa->spa_sync_on == B_FALSE);
1112	ASSERT(spa->spa_dsl_pool == NULL);
1113	ASSERT(spa->spa_root_vdev == NULL);
1114	ASSERT(spa->spa_async_zio_root == NULL);
1115	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1116
1117	spa_evicting_os_wait(spa);
1118
1119	txg_list_destroy(&spa->spa_vdev_txg_list);
1120
1121	list_destroy(&spa->spa_config_dirty_list);
1122	list_destroy(&spa->spa_evicting_os_list);
1123	list_destroy(&spa->spa_state_dirty_list);
1124
1125	for (int t = 0; t < ZIO_TYPES; t++) {
1126		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1127			spa_taskqs_fini(spa, t, q);
1128		}
1129	}
1130
1131	metaslab_class_destroy(spa->spa_normal_class);
1132	spa->spa_normal_class = NULL;
1133
1134	metaslab_class_destroy(spa->spa_log_class);
1135	spa->spa_log_class = NULL;
1136
1137	/*
1138	 * If this was part of an import or the open otherwise failed, we may
1139	 * still have errors left in the queues.  Empty them just in case.
1140	 */
1141	spa_errlog_drain(spa);
1142
1143	avl_destroy(&spa->spa_errlist_scrub);
1144	avl_destroy(&spa->spa_errlist_last);
1145
1146	spa->spa_state = POOL_STATE_UNINITIALIZED;
1147
1148	mutex_enter(&spa->spa_proc_lock);
1149	if (spa->spa_proc_state != SPA_PROC_NONE) {
1150		ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1151		spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1152		cv_broadcast(&spa->spa_proc_cv);
1153		while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1154			ASSERT(spa->spa_proc != &p0);
1155			cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1156		}
1157		ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1158		spa->spa_proc_state = SPA_PROC_NONE;
1159	}
1160	ASSERT(spa->spa_proc == &p0);
1161	mutex_exit(&spa->spa_proc_lock);
1162
1163	/*
1164	 * We want to make sure spa_thread() has actually exited the ZFS
1165	 * module, so that the module can't be unloaded out from underneath
1166	 * it.
1167	 */
1168	if (spa->spa_did != 0) {
1169		thread_join(spa->spa_did);
1170		spa->spa_did = 0;
1171	}
1172}
1173
1174/*
1175 * Verify a pool configuration, and construct the vdev tree appropriately.  This
1176 * will create all the necessary vdevs in the appropriate layout, with each vdev
1177 * in the CLOSED state.  This will prep the pool before open/creation/import.
1178 * All vdev validation is done by the vdev_alloc() routine.
1179 */
1180static int
1181spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1182    uint_t id, int atype)
1183{
1184	nvlist_t **child;
1185	uint_t children;
1186	int error;
1187
1188	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1189		return (error);
1190
1191	if ((*vdp)->vdev_ops->vdev_op_leaf)
1192		return (0);
1193
1194	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1195	    &child, &children);
1196
1197	if (error == ENOENT)
1198		return (0);
1199
1200	if (error) {
1201		vdev_free(*vdp);
1202		*vdp = NULL;
1203		return (SET_ERROR(EINVAL));
1204	}
1205
1206	for (int c = 0; c < children; c++) {
1207		vdev_t *vd;
1208		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1209		    atype)) != 0) {
1210			vdev_free(*vdp);
1211			*vdp = NULL;
1212			return (error);
1213		}
1214	}
1215
1216	ASSERT(*vdp != NULL);
1217
1218	return (0);
1219}
1220
1221/*
1222 * Opposite of spa_load().
1223 */
1224static void
1225spa_unload(spa_t *spa)
1226{
1227	int i;
1228
1229	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1230
1231	/*
1232	 * Stop async tasks.
1233	 */
1234	spa_async_suspend(spa);
1235
1236	/*
1237	 * Stop syncing.
1238	 */
1239	if (spa->spa_sync_on) {
1240		txg_sync_stop(spa->spa_dsl_pool);
1241		spa->spa_sync_on = B_FALSE;
1242	}
1243
1244	/*
1245	 * Even though vdev_free() also calls vdev_metaslab_fini, we need
1246	 * to call it earlier, before we wait for async i/o to complete.
1247	 * This ensures that there is no async metaslab prefetching, by
1248	 * calling taskq_wait(mg_taskq).
1249	 */
1250	if (spa->spa_root_vdev != NULL) {
1251		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1252		for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++)
1253			vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]);
1254		spa_config_exit(spa, SCL_ALL, FTAG);
1255	}
1256
1257	/*
1258	 * Wait for any outstanding async I/O to complete.
1259	 */
1260	if (spa->spa_async_zio_root != NULL) {
1261		for (int i = 0; i < max_ncpus; i++)
1262			(void) zio_wait(spa->spa_async_zio_root[i]);
1263		kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1264		spa->spa_async_zio_root = NULL;
1265	}
1266
1267	bpobj_close(&spa->spa_deferred_bpobj);
1268
1269	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1270
1271	/*
1272	 * Close all vdevs.
1273	 */
1274	if (spa->spa_root_vdev)
1275		vdev_free(spa->spa_root_vdev);
1276	ASSERT(spa->spa_root_vdev == NULL);
1277
1278	/*
1279	 * Close the dsl pool.
1280	 */
1281	if (spa->spa_dsl_pool) {
1282		dsl_pool_close(spa->spa_dsl_pool);
1283		spa->spa_dsl_pool = NULL;
1284		spa->spa_meta_objset = NULL;
1285	}
1286
1287	ddt_unload(spa);
1288
1289	/*
1290	 * Drop and purge level 2 cache
1291	 */
1292	spa_l2cache_drop(spa);
1293
1294	for (i = 0; i < spa->spa_spares.sav_count; i++)
1295		vdev_free(spa->spa_spares.sav_vdevs[i]);
1296	if (spa->spa_spares.sav_vdevs) {
1297		kmem_free(spa->spa_spares.sav_vdevs,
1298		    spa->spa_spares.sav_count * sizeof (void *));
1299		spa->spa_spares.sav_vdevs = NULL;
1300	}
1301	if (spa->spa_spares.sav_config) {
1302		nvlist_free(spa->spa_spares.sav_config);
1303		spa->spa_spares.sav_config = NULL;
1304	}
1305	spa->spa_spares.sav_count = 0;
1306
1307	for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1308		vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1309		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1310	}
1311	if (spa->spa_l2cache.sav_vdevs) {
1312		kmem_free(spa->spa_l2cache.sav_vdevs,
1313		    spa->spa_l2cache.sav_count * sizeof (void *));
1314		spa->spa_l2cache.sav_vdevs = NULL;
1315	}
1316	if (spa->spa_l2cache.sav_config) {
1317		nvlist_free(spa->spa_l2cache.sav_config);
1318		spa->spa_l2cache.sav_config = NULL;
1319	}
1320	spa->spa_l2cache.sav_count = 0;
1321
1322	spa->spa_async_suspended = 0;
1323
1324	if (spa->spa_comment != NULL) {
1325		spa_strfree(spa->spa_comment);
1326		spa->spa_comment = NULL;
1327	}
1328
1329	spa_config_exit(spa, SCL_ALL, FTAG);
1330}
1331
1332/*
1333 * Load (or re-load) the current list of vdevs describing the active spares for
1334 * this pool.  When this is called, we have some form of basic information in
1335 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1336 * then re-generate a more complete list including status information.
1337 */
1338static void
1339spa_load_spares(spa_t *spa)
1340{
1341	nvlist_t **spares;
1342	uint_t nspares;
1343	int i;
1344	vdev_t *vd, *tvd;
1345
1346	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1347
1348	/*
1349	 * First, close and free any existing spare vdevs.
1350	 */
1351	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1352		vd = spa->spa_spares.sav_vdevs[i];
1353
1354		/* Undo the call to spa_activate() below */
1355		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1356		    B_FALSE)) != NULL && tvd->vdev_isspare)
1357			spa_spare_remove(tvd);
1358		vdev_close(vd);
1359		vdev_free(vd);
1360	}
1361
1362	if (spa->spa_spares.sav_vdevs)
1363		kmem_free(spa->spa_spares.sav_vdevs,
1364		    spa->spa_spares.sav_count * sizeof (void *));
1365
1366	if (spa->spa_spares.sav_config == NULL)
1367		nspares = 0;
1368	else
1369		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1370		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1371
1372	spa->spa_spares.sav_count = (int)nspares;
1373	spa->spa_spares.sav_vdevs = NULL;
1374
1375	if (nspares == 0)
1376		return;
1377
1378	/*
1379	 * Construct the array of vdevs, opening them to get status in the
1380	 * process.   For each spare, there is potentially two different vdev_t
1381	 * structures associated with it: one in the list of spares (used only
1382	 * for basic validation purposes) and one in the active vdev
1383	 * configuration (if it's spared in).  During this phase we open and
1384	 * validate each vdev on the spare list.  If the vdev also exists in the
1385	 * active configuration, then we also mark this vdev as an active spare.
1386	 */
1387	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1388	    KM_SLEEP);
1389	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1390		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1391		    VDEV_ALLOC_SPARE) == 0);
1392		ASSERT(vd != NULL);
1393
1394		spa->spa_spares.sav_vdevs[i] = vd;
1395
1396		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1397		    B_FALSE)) != NULL) {
1398			if (!tvd->vdev_isspare)
1399				spa_spare_add(tvd);
1400
1401			/*
1402			 * We only mark the spare active if we were successfully
1403			 * able to load the vdev.  Otherwise, importing a pool
1404			 * with a bad active spare would result in strange
1405			 * behavior, because multiple pool would think the spare
1406			 * is actively in use.
1407			 *
1408			 * There is a vulnerability here to an equally bizarre
1409			 * circumstance, where a dead active spare is later
1410			 * brought back to life (onlined or otherwise).  Given
1411			 * the rarity of this scenario, and the extra complexity
1412			 * it adds, we ignore the possibility.
1413			 */
1414			if (!vdev_is_dead(tvd))
1415				spa_spare_activate(tvd);
1416		}
1417
1418		vd->vdev_top = vd;
1419		vd->vdev_aux = &spa->spa_spares;
1420
1421		if (vdev_open(vd) != 0)
1422			continue;
1423
1424		if (vdev_validate_aux(vd) == 0)
1425			spa_spare_add(vd);
1426	}
1427
1428	/*
1429	 * Recompute the stashed list of spares, with status information
1430	 * this time.
1431	 */
1432	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1433	    DATA_TYPE_NVLIST_ARRAY) == 0);
1434
1435	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1436	    KM_SLEEP);
1437	for (i = 0; i < spa->spa_spares.sav_count; i++)
1438		spares[i] = vdev_config_generate(spa,
1439		    spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1440	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1441	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1442	for (i = 0; i < spa->spa_spares.sav_count; i++)
1443		nvlist_free(spares[i]);
1444	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1445}
1446
1447/*
1448 * Load (or re-load) the current list of vdevs describing the active l2cache for
1449 * this pool.  When this is called, we have some form of basic information in
1450 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1451 * then re-generate a more complete list including status information.
1452 * Devices which are already active have their details maintained, and are
1453 * not re-opened.
1454 */
1455static void
1456spa_load_l2cache(spa_t *spa)
1457{
1458	nvlist_t **l2cache;
1459	uint_t nl2cache;
1460	int i, j, oldnvdevs;
1461	uint64_t guid;
1462	vdev_t *vd, **oldvdevs, **newvdevs;
1463	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1464
1465	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1466
1467	if (sav->sav_config != NULL) {
1468		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1469		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1470		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1471	} else {
1472		nl2cache = 0;
1473		newvdevs = NULL;
1474	}
1475
1476	oldvdevs = sav->sav_vdevs;
1477	oldnvdevs = sav->sav_count;
1478	sav->sav_vdevs = NULL;
1479	sav->sav_count = 0;
1480
1481	/*
1482	 * Process new nvlist of vdevs.
1483	 */
1484	for (i = 0; i < nl2cache; i++) {
1485		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1486		    &guid) == 0);
1487
1488		newvdevs[i] = NULL;
1489		for (j = 0; j < oldnvdevs; j++) {
1490			vd = oldvdevs[j];
1491			if (vd != NULL && guid == vd->vdev_guid) {
1492				/*
1493				 * Retain previous vdev for add/remove ops.
1494				 */
1495				newvdevs[i] = vd;
1496				oldvdevs[j] = NULL;
1497				break;
1498			}
1499		}
1500
1501		if (newvdevs[i] == NULL) {
1502			/*
1503			 * Create new vdev
1504			 */
1505			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1506			    VDEV_ALLOC_L2CACHE) == 0);
1507			ASSERT(vd != NULL);
1508			newvdevs[i] = vd;
1509
1510			/*
1511			 * Commit this vdev as an l2cache device,
1512			 * even if it fails to open.
1513			 */
1514			spa_l2cache_add(vd);
1515
1516			vd->vdev_top = vd;
1517			vd->vdev_aux = sav;
1518
1519			spa_l2cache_activate(vd);
1520
1521			if (vdev_open(vd) != 0)
1522				continue;
1523
1524			(void) vdev_validate_aux(vd);
1525
1526			if (!vdev_is_dead(vd))
1527				l2arc_add_vdev(spa, vd);
1528		}
1529	}
1530
1531	/*
1532	 * Purge vdevs that were dropped
1533	 */
1534	for (i = 0; i < oldnvdevs; i++) {
1535		uint64_t pool;
1536
1537		vd = oldvdevs[i];
1538		if (vd != NULL) {
1539			ASSERT(vd->vdev_isl2cache);
1540
1541			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1542			    pool != 0ULL && l2arc_vdev_present(vd))
1543				l2arc_remove_vdev(vd);
1544			vdev_clear_stats(vd);
1545			vdev_free(vd);
1546		}
1547	}
1548
1549	if (oldvdevs)
1550		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1551
1552	if (sav->sav_config == NULL)
1553		goto out;
1554
1555	sav->sav_vdevs = newvdevs;
1556	sav->sav_count = (int)nl2cache;
1557
1558	/*
1559	 * Recompute the stashed list of l2cache devices, with status
1560	 * information this time.
1561	 */
1562	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1563	    DATA_TYPE_NVLIST_ARRAY) == 0);
1564
1565	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1566	for (i = 0; i < sav->sav_count; i++)
1567		l2cache[i] = vdev_config_generate(spa,
1568		    sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1569	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1570	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1571out:
1572	for (i = 0; i < sav->sav_count; i++)
1573		nvlist_free(l2cache[i]);
1574	if (sav->sav_count)
1575		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1576}
1577
1578static int
1579load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1580{
1581	dmu_buf_t *db;
1582	char *packed = NULL;
1583	size_t nvsize = 0;
1584	int error;
1585	*value = NULL;
1586
1587	error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1588	if (error != 0)
1589		return (error);
1590
1591	nvsize = *(uint64_t *)db->db_data;
1592	dmu_buf_rele(db, FTAG);
1593
1594	packed = kmem_alloc(nvsize, KM_SLEEP);
1595	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1596	    DMU_READ_PREFETCH);
1597	if (error == 0)
1598		error = nvlist_unpack(packed, nvsize, value, 0);
1599	kmem_free(packed, nvsize);
1600
1601	return (error);
1602}
1603
1604/*
1605 * Checks to see if the given vdev could not be opened, in which case we post a
1606 * sysevent to notify the autoreplace code that the device has been removed.
1607 */
1608static void
1609spa_check_removed(vdev_t *vd)
1610{
1611	for (int c = 0; c < vd->vdev_children; c++)
1612		spa_check_removed(vd->vdev_child[c]);
1613
1614	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1615	    !vd->vdev_ishole) {
1616		zfs_post_autoreplace(vd->vdev_spa, vd);
1617		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1618	}
1619}
1620
1621static void
1622spa_config_valid_zaps(vdev_t *vd, vdev_t *mvd)
1623{
1624	ASSERT3U(vd->vdev_children, ==, mvd->vdev_children);
1625
1626	vd->vdev_top_zap = mvd->vdev_top_zap;
1627	vd->vdev_leaf_zap = mvd->vdev_leaf_zap;
1628
1629	for (uint64_t i = 0; i < vd->vdev_children; i++) {
1630		spa_config_valid_zaps(vd->vdev_child[i], mvd->vdev_child[i]);
1631	}
1632}
1633
1634/*
1635 * Validate the current config against the MOS config
1636 */
1637static boolean_t
1638spa_config_valid(spa_t *spa, nvlist_t *config)
1639{
1640	vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1641	nvlist_t *nv;
1642
1643	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1644
1645	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1646	VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1647
1648	ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1649
1650	/*
1651	 * If we're doing a normal import, then build up any additional
1652	 * diagnostic information about missing devices in this config.
1653	 * We'll pass this up to the user for further processing.
1654	 */
1655	if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1656		nvlist_t **child, *nv;
1657		uint64_t idx = 0;
1658
1659		child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1660		    KM_SLEEP);
1661		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1662
1663		for (int c = 0; c < rvd->vdev_children; c++) {
1664			vdev_t *tvd = rvd->vdev_child[c];
1665			vdev_t *mtvd  = mrvd->vdev_child[c];
1666
1667			if (tvd->vdev_ops == &vdev_missing_ops &&
1668			    mtvd->vdev_ops != &vdev_missing_ops &&
1669			    mtvd->vdev_islog)
1670				child[idx++] = vdev_config_generate(spa, mtvd,
1671				    B_FALSE, 0);
1672		}
1673
1674		if (idx) {
1675			VERIFY(nvlist_add_nvlist_array(nv,
1676			    ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1677			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1678			    ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1679
1680			for (int i = 0; i < idx; i++)
1681				nvlist_free(child[i]);
1682		}
1683		nvlist_free(nv);
1684		kmem_free(child, rvd->vdev_children * sizeof (char **));
1685	}
1686
1687	/*
1688	 * Compare the root vdev tree with the information we have
1689	 * from the MOS config (mrvd). Check each top-level vdev
1690	 * with the corresponding MOS config top-level (mtvd).
1691	 */
1692	for (int c = 0; c < rvd->vdev_children; c++) {
1693		vdev_t *tvd = rvd->vdev_child[c];
1694		vdev_t *mtvd  = mrvd->vdev_child[c];
1695
1696		/*
1697		 * Resolve any "missing" vdevs in the current configuration.
1698		 * If we find that the MOS config has more accurate information
1699		 * about the top-level vdev then use that vdev instead.
1700		 */
1701		if (tvd->vdev_ops == &vdev_missing_ops &&
1702		    mtvd->vdev_ops != &vdev_missing_ops) {
1703
1704			if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1705				continue;
1706
1707			/*
1708			 * Device specific actions.
1709			 */
1710			if (mtvd->vdev_islog) {
1711				spa_set_log_state(spa, SPA_LOG_CLEAR);
1712			} else {
1713				/*
1714				 * XXX - once we have 'readonly' pool
1715				 * support we should be able to handle
1716				 * missing data devices by transitioning
1717				 * the pool to readonly.
1718				 */
1719				continue;
1720			}
1721
1722			/*
1723			 * Swap the missing vdev with the data we were
1724			 * able to obtain from the MOS config.
1725			 */
1726			vdev_remove_child(rvd, tvd);
1727			vdev_remove_child(mrvd, mtvd);
1728
1729			vdev_add_child(rvd, mtvd);
1730			vdev_add_child(mrvd, tvd);
1731
1732			spa_config_exit(spa, SCL_ALL, FTAG);
1733			vdev_load(mtvd);
1734			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1735
1736			vdev_reopen(rvd);
1737		} else {
1738			if (mtvd->vdev_islog) {
1739				/*
1740				 * Load the slog device's state from the MOS
1741				 * config since it's possible that the label
1742				 * does not contain the most up-to-date
1743				 * information.
1744				 */
1745				vdev_load_log_state(tvd, mtvd);
1746				vdev_reopen(tvd);
1747			}
1748
1749			/*
1750			 * Per-vdev ZAP info is stored exclusively in the MOS.
1751			 */
1752			spa_config_valid_zaps(tvd, mtvd);
1753		}
1754	}
1755
1756	vdev_free(mrvd);
1757	spa_config_exit(spa, SCL_ALL, FTAG);
1758
1759	/*
1760	 * Ensure we were able to validate the config.
1761	 */
1762	return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1763}
1764
1765/*
1766 * Check for missing log devices
1767 */
1768static boolean_t
1769spa_check_logs(spa_t *spa)
1770{
1771	boolean_t rv = B_FALSE;
1772	dsl_pool_t *dp = spa_get_dsl(spa);
1773
1774	switch (spa->spa_log_state) {
1775	case SPA_LOG_MISSING:
1776		/* need to recheck in case slog has been restored */
1777	case SPA_LOG_UNKNOWN:
1778		rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1779		    zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
1780		if (rv)
1781			spa_set_log_state(spa, SPA_LOG_MISSING);
1782		break;
1783	}
1784	return (rv);
1785}
1786
1787static boolean_t
1788spa_passivate_log(spa_t *spa)
1789{
1790	vdev_t *rvd = spa->spa_root_vdev;
1791	boolean_t slog_found = B_FALSE;
1792
1793	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1794
1795	if (!spa_has_slogs(spa))
1796		return (B_FALSE);
1797
1798	for (int c = 0; c < rvd->vdev_children; c++) {
1799		vdev_t *tvd = rvd->vdev_child[c];
1800		metaslab_group_t *mg = tvd->vdev_mg;
1801
1802		if (tvd->vdev_islog) {
1803			metaslab_group_passivate(mg);
1804			slog_found = B_TRUE;
1805		}
1806	}
1807
1808	return (slog_found);
1809}
1810
1811static void
1812spa_activate_log(spa_t *spa)
1813{
1814	vdev_t *rvd = spa->spa_root_vdev;
1815
1816	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1817
1818	for (int c = 0; c < rvd->vdev_children; c++) {
1819		vdev_t *tvd = rvd->vdev_child[c];
1820		metaslab_group_t *mg = tvd->vdev_mg;
1821
1822		if (tvd->vdev_islog)
1823			metaslab_group_activate(mg);
1824	}
1825}
1826
1827int
1828spa_offline_log(spa_t *spa)
1829{
1830	int error;
1831
1832	error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1833	    NULL, DS_FIND_CHILDREN);
1834	if (error == 0) {
1835		/*
1836		 * We successfully offlined the log device, sync out the
1837		 * current txg so that the "stubby" block can be removed
1838		 * by zil_sync().
1839		 */
1840		txg_wait_synced(spa->spa_dsl_pool, 0);
1841	}
1842	return (error);
1843}
1844
1845static void
1846spa_aux_check_removed(spa_aux_vdev_t *sav)
1847{
1848	for (int i = 0; i < sav->sav_count; i++)
1849		spa_check_removed(sav->sav_vdevs[i]);
1850}
1851
1852void
1853spa_claim_notify(zio_t *zio)
1854{
1855	spa_t *spa = zio->io_spa;
1856
1857	if (zio->io_error)
1858		return;
1859
1860	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1861	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1862		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1863	mutex_exit(&spa->spa_props_lock);
1864}
1865
1866typedef struct spa_load_error {
1867	uint64_t	sle_meta_count;
1868	uint64_t	sle_data_count;
1869} spa_load_error_t;
1870
1871static void
1872spa_load_verify_done(zio_t *zio)
1873{
1874	blkptr_t *bp = zio->io_bp;
1875	spa_load_error_t *sle = zio->io_private;
1876	dmu_object_type_t type = BP_GET_TYPE(bp);
1877	int error = zio->io_error;
1878	spa_t *spa = zio->io_spa;
1879
1880	abd_free(zio->io_abd);
1881	if (error) {
1882		if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1883		    type != DMU_OT_INTENT_LOG)
1884			atomic_inc_64(&sle->sle_meta_count);
1885		else
1886			atomic_inc_64(&sle->sle_data_count);
1887	}
1888
1889	mutex_enter(&spa->spa_scrub_lock);
1890	spa->spa_scrub_inflight--;
1891	cv_broadcast(&spa->spa_scrub_io_cv);
1892	mutex_exit(&spa->spa_scrub_lock);
1893}
1894
1895/*
1896 * Maximum number of concurrent scrub i/os to create while verifying
1897 * a pool while importing it.
1898 */
1899int spa_load_verify_maxinflight = 10000;
1900boolean_t spa_load_verify_metadata = B_TRUE;
1901boolean_t spa_load_verify_data = B_TRUE;
1902
1903/*ARGSUSED*/
1904static int
1905spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1906    const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1907{
1908	if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
1909		return (0);
1910	/*
1911	 * Note: normally this routine will not be called if
1912	 * spa_load_verify_metadata is not set.  However, it may be useful
1913	 * to manually set the flag after the traversal has begun.
1914	 */
1915	if (!spa_load_verify_metadata)
1916		return (0);
1917	if (!BP_IS_METADATA(bp) && !spa_load_verify_data)
1918		return (0);
1919
1920	zio_t *rio = arg;
1921	size_t size = BP_GET_PSIZE(bp);
1922
1923	mutex_enter(&spa->spa_scrub_lock);
1924	while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
1925		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1926	spa->spa_scrub_inflight++;
1927	mutex_exit(&spa->spa_scrub_lock);
1928
1929	zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size,
1930	    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1931	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1932	    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1933	return (0);
1934}
1935
1936/* ARGSUSED */
1937int
1938verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1939{
1940	if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN)
1941		return (SET_ERROR(ENAMETOOLONG));
1942
1943	return (0);
1944}
1945
1946static int
1947spa_load_verify(spa_t *spa)
1948{
1949	zio_t *rio;
1950	spa_load_error_t sle = { 0 };
1951	zpool_rewind_policy_t policy;
1952	boolean_t verify_ok = B_FALSE;
1953	int error = 0;
1954
1955	zpool_get_rewind_policy(spa->spa_config, &policy);
1956
1957	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1958		return (0);
1959
1960	dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
1961	error = dmu_objset_find_dp(spa->spa_dsl_pool,
1962	    spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL,
1963	    DS_FIND_CHILDREN);
1964	dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
1965	if (error != 0)
1966		return (error);
1967
1968	rio = zio_root(spa, NULL, &sle,
1969	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1970
1971	if (spa_load_verify_metadata) {
1972		error = traverse_pool(spa, spa->spa_verify_min_txg,
1973		    TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
1974		    spa_load_verify_cb, rio);
1975	}
1976
1977	(void) zio_wait(rio);
1978
1979	spa->spa_load_meta_errors = sle.sle_meta_count;
1980	spa->spa_load_data_errors = sle.sle_data_count;
1981
1982	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1983	    sle.sle_data_count <= policy.zrp_maxdata) {
1984		int64_t loss = 0;
1985
1986		verify_ok = B_TRUE;
1987		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1988		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1989
1990		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1991		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1992		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1993		VERIFY(nvlist_add_int64(spa->spa_load_info,
1994		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1995		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1996		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1997	} else {
1998		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1999	}
2000
2001	if (error) {
2002		if (error != ENXIO && error != EIO)
2003			error = SET_ERROR(EIO);
2004		return (error);
2005	}
2006
2007	return (verify_ok ? 0 : EIO);
2008}
2009
2010/*
2011 * Find a value in the pool props object.
2012 */
2013static void
2014spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2015{
2016	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2017	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2018}
2019
2020/*
2021 * Find a value in the pool directory object.
2022 */
2023static int
2024spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
2025{
2026	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2027	    name, sizeof (uint64_t), 1, val));
2028}
2029
2030static int
2031spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2032{
2033	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2034	return (err);
2035}
2036
2037/*
2038 * Fix up config after a partly-completed split.  This is done with the
2039 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
2040 * pool have that entry in their config, but only the splitting one contains
2041 * a list of all the guids of the vdevs that are being split off.
2042 *
2043 * This function determines what to do with that list: either rejoin
2044 * all the disks to the pool, or complete the splitting process.  To attempt
2045 * the rejoin, each disk that is offlined is marked online again, and
2046 * we do a reopen() call.  If the vdev label for every disk that was
2047 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2048 * then we call vdev_split() on each disk, and complete the split.
2049 *
2050 * Otherwise we leave the config alone, with all the vdevs in place in
2051 * the original pool.
2052 */
2053static void
2054spa_try_repair(spa_t *spa, nvlist_t *config)
2055{
2056	uint_t extracted;
2057	uint64_t *glist;
2058	uint_t i, gcount;
2059	nvlist_t *nvl;
2060	vdev_t **vd;
2061	boolean_t attempt_reopen;
2062
2063	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2064		return;
2065
2066	/* check that the config is complete */
2067	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2068	    &glist, &gcount) != 0)
2069		return;
2070
2071	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2072
2073	/* attempt to online all the vdevs & validate */
2074	attempt_reopen = B_TRUE;
2075	for (i = 0; i < gcount; i++) {
2076		if (glist[i] == 0)	/* vdev is hole */
2077			continue;
2078
2079		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2080		if (vd[i] == NULL) {
2081			/*
2082			 * Don't bother attempting to reopen the disks;
2083			 * just do the split.
2084			 */
2085			attempt_reopen = B_FALSE;
2086		} else {
2087			/* attempt to re-online it */
2088			vd[i]->vdev_offline = B_FALSE;
2089		}
2090	}
2091
2092	if (attempt_reopen) {
2093		vdev_reopen(spa->spa_root_vdev);
2094
2095		/* check each device to see what state it's in */
2096		for (extracted = 0, i = 0; i < gcount; i++) {
2097			if (vd[i] != NULL &&
2098			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2099				break;
2100			++extracted;
2101		}
2102	}
2103
2104	/*
2105	 * If every disk has been moved to the new pool, or if we never
2106	 * even attempted to look at them, then we split them off for
2107	 * good.
2108	 */
2109	if (!attempt_reopen || gcount == extracted) {
2110		for (i = 0; i < gcount; i++)
2111			if (vd[i] != NULL)
2112				vdev_split(vd[i]);
2113		vdev_reopen(spa->spa_root_vdev);
2114	}
2115
2116	kmem_free(vd, gcount * sizeof (vdev_t *));
2117}
2118
2119static int
2120spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2121    boolean_t mosconfig)
2122{
2123	nvlist_t *config = spa->spa_config;
2124	char *ereport = FM_EREPORT_ZFS_POOL;
2125	char *comment;
2126	int error;
2127	uint64_t pool_guid;
2128	nvlist_t *nvl;
2129
2130	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2131		return (SET_ERROR(EINVAL));
2132
2133	ASSERT(spa->spa_comment == NULL);
2134	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2135		spa->spa_comment = spa_strdup(comment);
2136
2137	/*
2138	 * Versioning wasn't explicitly added to the label until later, so if
2139	 * it's not present treat it as the initial version.
2140	 */
2141	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2142	    &spa->spa_ubsync.ub_version) != 0)
2143		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2144
2145	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2146	    &spa->spa_config_txg);
2147
2148	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2149	    spa_guid_exists(pool_guid, 0)) {
2150		error = SET_ERROR(EEXIST);
2151	} else {
2152		spa->spa_config_guid = pool_guid;
2153
2154		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2155		    &nvl) == 0) {
2156			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2157			    KM_SLEEP) == 0);
2158		}
2159
2160		nvlist_free(spa->spa_load_info);
2161		spa->spa_load_info = fnvlist_alloc();
2162
2163		gethrestime(&spa->spa_loaded_ts);
2164		error = spa_load_impl(spa, pool_guid, config, state, type,
2165		    mosconfig, &ereport);
2166	}
2167
2168	/*
2169	 * Don't count references from objsets that are already closed
2170	 * and are making their way through the eviction process.
2171	 */
2172	spa_evicting_os_wait(spa);
2173	spa->spa_minref = refcount_count(&spa->spa_refcount);
2174	if (error) {
2175		if (error != EEXIST) {
2176			spa->spa_loaded_ts.tv_sec = 0;
2177			spa->spa_loaded_ts.tv_nsec = 0;
2178		}
2179		if (error != EBADF) {
2180			zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2181		}
2182	}
2183	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2184	spa->spa_ena = 0;
2185
2186	return (error);
2187}
2188
2189/*
2190 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2191 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2192 * spa's per-vdev ZAP list.
2193 */
2194static uint64_t
2195vdev_count_verify_zaps(vdev_t *vd)
2196{
2197	spa_t *spa = vd->vdev_spa;
2198	uint64_t total = 0;
2199	if (vd->vdev_top_zap != 0) {
2200		total++;
2201		ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2202		    spa->spa_all_vdev_zaps, vd->vdev_top_zap));
2203	}
2204	if (vd->vdev_leaf_zap != 0) {
2205		total++;
2206		ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2207		    spa->spa_all_vdev_zaps, vd->vdev_leaf_zap));
2208	}
2209
2210	for (uint64_t i = 0; i < vd->vdev_children; i++) {
2211		total += vdev_count_verify_zaps(vd->vdev_child[i]);
2212	}
2213
2214	return (total);
2215}
2216
2217/*
2218 * Load an existing storage pool, using the pool's builtin spa_config as a
2219 * source of configuration information.
2220 */
2221static int
2222spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2223    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2224    char **ereport)
2225{
2226	int error = 0;
2227	nvlist_t *nvroot = NULL;
2228	nvlist_t *label;
2229	vdev_t *rvd;
2230	uberblock_t *ub = &spa->spa_uberblock;
2231	uint64_t children, config_cache_txg = spa->spa_config_txg;
2232	int orig_mode = spa->spa_mode;
2233	int parse;
2234	uint64_t obj;
2235	boolean_t missing_feat_write = B_FALSE;
2236
2237	/*
2238	 * If this is an untrusted config, access the pool in read-only mode.
2239	 * This prevents things like resilvering recently removed devices.
2240	 */
2241	if (!mosconfig)
2242		spa->spa_mode = FREAD;
2243
2244	ASSERT(MUTEX_HELD(&spa_namespace_lock));
2245
2246	spa->spa_load_state = state;
2247
2248	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2249		return (SET_ERROR(EINVAL));
2250
2251	parse = (type == SPA_IMPORT_EXISTING ?
2252	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2253
2254	/*
2255	 * Create "The Godfather" zio to hold all async IOs
2256	 */
2257	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2258	    KM_SLEEP);
2259	for (int i = 0; i < max_ncpus; i++) {
2260		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2261		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2262		    ZIO_FLAG_GODFATHER);
2263	}
2264
2265	/*
2266	 * Parse the configuration into a vdev tree.  We explicitly set the
2267	 * value that will be returned by spa_version() since parsing the
2268	 * configuration requires knowing the version number.
2269	 */
2270	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2271	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2272	spa_config_exit(spa, SCL_ALL, FTAG);
2273
2274	if (error != 0)
2275		return (error);
2276
2277	ASSERT(spa->spa_root_vdev == rvd);
2278	ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
2279	ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
2280
2281	if (type != SPA_IMPORT_ASSEMBLE) {
2282		ASSERT(spa_guid(spa) == pool_guid);
2283	}
2284
2285	/*
2286	 * Try to open all vdevs, loading each label in the process.
2287	 */
2288	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2289	error = vdev_open(rvd);
2290	spa_config_exit(spa, SCL_ALL, FTAG);
2291	if (error != 0)
2292		return (error);
2293
2294	/*
2295	 * We need to validate the vdev labels against the configuration that
2296	 * we have in hand, which is dependent on the setting of mosconfig. If
2297	 * mosconfig is true then we're validating the vdev labels based on
2298	 * that config.  Otherwise, we're validating against the cached config
2299	 * (zpool.cache) that was read when we loaded the zfs module, and then
2300	 * later we will recursively call spa_load() and validate against
2301	 * the vdev config.
2302	 *
2303	 * If we're assembling a new pool that's been split off from an
2304	 * existing pool, the labels haven't yet been updated so we skip
2305	 * validation for now.
2306	 */
2307	if (type != SPA_IMPORT_ASSEMBLE) {
2308		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2309		error = vdev_validate(rvd, mosconfig);
2310		spa_config_exit(spa, SCL_ALL, FTAG);
2311
2312		if (error != 0)
2313			return (error);
2314
2315		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2316			return (SET_ERROR(ENXIO));
2317	}
2318
2319	/*
2320	 * Find the best uberblock.
2321	 */
2322	vdev_uberblock_load(rvd, ub, &label);
2323
2324	/*
2325	 * If we weren't able to find a single valid uberblock, return failure.
2326	 */
2327	if (ub->ub_txg == 0) {
2328		nvlist_free(label);
2329		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2330	}
2331
2332	/*
2333	 * If the pool has an unsupported version we can't open it.
2334	 */
2335	if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2336		nvlist_free(label);
2337		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2338	}
2339
2340	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2341		nvlist_t *features;
2342
2343		/*
2344		 * If we weren't able to find what's necessary for reading the
2345		 * MOS in the label, return failure.
2346		 */
2347		if (label == NULL || nvlist_lookup_nvlist(label,
2348		    ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2349			nvlist_free(label);
2350			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2351			    ENXIO));
2352		}
2353
2354		/*
2355		 * Update our in-core representation with the definitive values
2356		 * from the label.
2357		 */
2358		nvlist_free(spa->spa_label_features);
2359		VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2360	}
2361
2362	nvlist_free(label);
2363
2364	/*
2365	 * Look through entries in the label nvlist's features_for_read. If
2366	 * there is a feature listed there which we don't understand then we
2367	 * cannot open a pool.
2368	 */
2369	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2370		nvlist_t *unsup_feat;
2371
2372		VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2373		    0);
2374
2375		for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2376		    NULL); nvp != NULL;
2377		    nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2378			if (!zfeature_is_supported(nvpair_name(nvp))) {
2379				VERIFY(nvlist_add_string(unsup_feat,
2380				    nvpair_name(nvp), "") == 0);
2381			}
2382		}
2383
2384		if (!nvlist_empty(unsup_feat)) {
2385			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2386			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2387			nvlist_free(unsup_feat);
2388			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2389			    ENOTSUP));
2390		}
2391
2392		nvlist_free(unsup_feat);
2393	}
2394
2395	/*
2396	 * If the vdev guid sum doesn't match the uberblock, we have an
2397	 * incomplete configuration.  We first check to see if the pool
2398	 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2399	 * If it is, defer the vdev_guid_sum check till later so we
2400	 * can handle missing vdevs.
2401	 */
2402	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2403	    &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2404	    rvd->vdev_guid_sum != ub->ub_guid_sum)
2405		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2406
2407	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2408		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2409		spa_try_repair(spa, config);
2410		spa_config_exit(spa, SCL_ALL, FTAG);
2411		nvlist_free(spa->spa_config_splitting);
2412		spa->spa_config_splitting = NULL;
2413	}
2414
2415	/*
2416	 * Initialize internal SPA structures.
2417	 */
2418	spa->spa_state = POOL_STATE_ACTIVE;
2419	spa->spa_ubsync = spa->spa_uberblock;
2420	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2421	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2422	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2423	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2424	spa->spa_claim_max_txg = spa->spa_first_txg;
2425	spa->spa_prev_software_version = ub->ub_software_version;
2426
2427	error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2428	if (error)
2429		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2430	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2431
2432	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2433		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2434
2435	if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2436		boolean_t missing_feat_read = B_FALSE;
2437		nvlist_t *unsup_feat, *enabled_feat;
2438
2439		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2440		    &spa->spa_feat_for_read_obj) != 0) {
2441			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2442		}
2443
2444		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2445		    &spa->spa_feat_for_write_obj) != 0) {
2446			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2447		}
2448
2449		if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2450		    &spa->spa_feat_desc_obj) != 0) {
2451			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2452		}
2453
2454		enabled_feat = fnvlist_alloc();
2455		unsup_feat = fnvlist_alloc();
2456
2457		if (!spa_features_check(spa, B_FALSE,
2458		    unsup_feat, enabled_feat))
2459			missing_feat_read = B_TRUE;
2460
2461		if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2462			if (!spa_features_check(spa, B_TRUE,
2463			    unsup_feat, enabled_feat)) {
2464				missing_feat_write = B_TRUE;
2465			}
2466		}
2467
2468		fnvlist_add_nvlist(spa->spa_load_info,
2469		    ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2470
2471		if (!nvlist_empty(unsup_feat)) {
2472			fnvlist_add_nvlist(spa->spa_load_info,
2473			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2474		}
2475
2476		fnvlist_free(enabled_feat);
2477		fnvlist_free(unsup_feat);
2478
2479		if (!missing_feat_read) {
2480			fnvlist_add_boolean(spa->spa_load_info,
2481			    ZPOOL_CONFIG_CAN_RDONLY);
2482		}
2483
2484		/*
2485		 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2486		 * twofold: to determine whether the pool is available for
2487		 * import in read-write mode and (if it is not) whether the
2488		 * pool is available for import in read-only mode. If the pool
2489		 * is available for import in read-write mode, it is displayed
2490		 * as available in userland; if it is not available for import
2491		 * in read-only mode, it is displayed as unavailable in
2492		 * userland. If the pool is available for import in read-only
2493		 * mode but not read-write mode, it is displayed as unavailable
2494		 * in userland with a special note that the pool is actually
2495		 * available for open in read-only mode.
2496		 *
2497		 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2498		 * missing a feature for write, we must first determine whether
2499		 * the pool can be opened read-only before returning to
2500		 * userland in order to know whether to display the
2501		 * abovementioned note.
2502		 */
2503		if (missing_feat_read || (missing_feat_write &&
2504		    spa_writeable(spa))) {
2505			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2506			    ENOTSUP));
2507		}
2508
2509		/*
2510		 * Load refcounts for ZFS features from disk into an in-memory
2511		 * cache during SPA initialization.
2512		 */
2513		for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2514			uint64_t refcount;
2515
2516			error = feature_get_refcount_from_disk(spa,
2517			    &spa_feature_table[i], &refcount);
2518			if (error == 0) {
2519				spa->spa_feat_refcount_cache[i] = refcount;
2520			} else if (error == ENOTSUP) {
2521				spa->spa_feat_refcount_cache[i] =
2522				    SPA_FEATURE_DISABLED;
2523			} else {
2524				return (spa_vdev_err(rvd,
2525				    VDEV_AUX_CORRUPT_DATA, EIO));
2526			}
2527		}
2528	}
2529
2530	if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2531		if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2532		    &spa->spa_feat_enabled_txg_obj) != 0)
2533			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2534	}
2535
2536	spa->spa_is_initializing = B_TRUE;
2537	error = dsl_pool_open(spa->spa_dsl_pool);
2538	spa->spa_is_initializing = B_FALSE;
2539	if (error != 0)
2540		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2541
2542	if (!mosconfig) {
2543		uint64_t hostid;
2544		nvlist_t *policy = NULL, *nvconfig;
2545
2546		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2547			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2548
2549		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2550		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2551			char *hostname;
2552			unsigned long myhostid = 0;
2553
2554			VERIFY(nvlist_lookup_string(nvconfig,
2555			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2556
2557#ifdef	_KERNEL
2558			myhostid = zone_get_hostid(NULL);
2559#else	/* _KERNEL */
2560			/*
2561			 * We're emulating the system's hostid in userland, so
2562			 * we can't use zone_get_hostid().
2563			 */
2564			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2565#endif	/* _KERNEL */
2566			if (hostid != 0 && myhostid != 0 &&
2567			    hostid != myhostid) {
2568				nvlist_free(nvconfig);
2569				cmn_err(CE_WARN, "pool '%s' could not be "
2570				    "loaded as it was last accessed by "
2571				    "another system (host: %s hostid: 0x%lx). "
2572				    "See: http://illumos.org/msg/ZFS-8000-EY",
2573				    spa_name(spa), hostname,
2574				    (unsigned long)hostid);
2575				return (SET_ERROR(EBADF));
2576			}
2577		}
2578		if (nvlist_lookup_nvlist(spa->spa_config,
2579		    ZPOOL_REWIND_POLICY, &policy) == 0)
2580			VERIFY(nvlist_add_nvlist(nvconfig,
2581			    ZPOOL_REWIND_POLICY, policy) == 0);
2582
2583		spa_config_set(spa, nvconfig);
2584		spa_unload(spa);
2585		spa_deactivate(spa);
2586		spa_activate(spa, orig_mode);
2587
2588		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2589	}
2590
2591	/* Grab the secret checksum salt from the MOS. */
2592	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2593	    DMU_POOL_CHECKSUM_SALT, 1,
2594	    sizeof (spa->spa_cksum_salt.zcs_bytes),
2595	    spa->spa_cksum_salt.zcs_bytes);
2596	if (error == ENOENT) {
2597		/* Generate a new salt for subsequent use */
2598		(void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
2599		    sizeof (spa->spa_cksum_salt.zcs_bytes));
2600	} else if (error != 0) {
2601		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2602	}
2603
2604	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2605		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2606	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2607	if (error != 0)
2608		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2609
2610	/*
2611	 * Load the bit that tells us to use the new accounting function
2612	 * (raid-z deflation).  If we have an older pool, this will not
2613	 * be present.
2614	 */
2615	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2616	if (error != 0 && error != ENOENT)
2617		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2618
2619	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2620	    &spa->spa_creation_version);
2621	if (error != 0 && error != ENOENT)
2622		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2623
2624	/*
2625	 * Load the persistent error log.  If we have an older pool, this will
2626	 * not be present.
2627	 */
2628	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2629	if (error != 0 && error != ENOENT)
2630		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2631
2632	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2633	    &spa->spa_errlog_scrub);
2634	if (error != 0 && error != ENOENT)
2635		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2636
2637	/*
2638	 * Load the history object.  If we have an older pool, this
2639	 * will not be present.
2640	 */
2641	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2642	if (error != 0 && error != ENOENT)
2643		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2644
2645	/*
2646	 * Load the per-vdev ZAP map. If we have an older pool, this will not
2647	 * be present; in this case, defer its creation to a later time to
2648	 * avoid dirtying the MOS this early / out of sync context. See
2649	 * spa_sync_config_object.
2650	 */
2651
2652	/* The sentinel is only available in the MOS config. */
2653	nvlist_t *mos_config;
2654	if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0)
2655		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2656
2657	error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP,
2658	    &spa->spa_all_vdev_zaps);
2659
2660	if (error == ENOENT) {
2661		VERIFY(!nvlist_exists(mos_config,
2662		    ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
2663		spa->spa_avz_action = AVZ_ACTION_INITIALIZE;
2664		ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
2665	} else if (error != 0) {
2666		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2667	} else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) {
2668		/*
2669		 * An older version of ZFS overwrote the sentinel value, so
2670		 * we have orphaned per-vdev ZAPs in the MOS. Defer their
2671		 * destruction to later; see spa_sync_config_object.
2672		 */
2673		spa->spa_avz_action = AVZ_ACTION_DESTROY;
2674		/*
2675		 * We're assuming that no vdevs have had their ZAPs created
2676		 * before this. Better be sure of it.
2677		 */
2678		ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
2679	}
2680	nvlist_free(mos_config);
2681
2682	/*
2683	 * If we're assembling the pool from the split-off vdevs of
2684	 * an existing pool, we don't want to attach the spares & cache
2685	 * devices.
2686	 */
2687
2688	/*
2689	 * Load any hot spares for this pool.
2690	 */
2691	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2692	if (error != 0 && error != ENOENT)
2693		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2694	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2695		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2696		if (load_nvlist(spa, spa->spa_spares.sav_object,
2697		    &spa->spa_spares.sav_config) != 0)
2698			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2699
2700		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2701		spa_load_spares(spa);
2702		spa_config_exit(spa, SCL_ALL, FTAG);
2703	} else if (error == 0) {
2704		spa->spa_spares.sav_sync = B_TRUE;
2705	}
2706
2707	/*
2708	 * Load any level 2 ARC devices for this pool.
2709	 */
2710	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2711	    &spa->spa_l2cache.sav_object);
2712	if (error != 0 && error != ENOENT)
2713		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2714	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2715		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2716		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2717		    &spa->spa_l2cache.sav_config) != 0)
2718			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2719
2720		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2721		spa_load_l2cache(spa);
2722		spa_config_exit(spa, SCL_ALL, FTAG);
2723	} else if (error == 0) {
2724		spa->spa_l2cache.sav_sync = B_TRUE;
2725	}
2726
2727	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2728
2729	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2730	if (error && error != ENOENT)
2731		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2732
2733	if (error == 0) {
2734		uint64_t autoreplace;
2735
2736		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2737		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2738		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2739		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2740		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2741		spa_prop_find(spa, ZPOOL_PROP_BOOTSIZE, &spa->spa_bootsize);
2742		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2743		    &spa->spa_dedup_ditto);
2744
2745		spa->spa_autoreplace = (autoreplace != 0);
2746	}
2747
2748	/*
2749	 * If the 'autoreplace' property is set, then post a resource notifying
2750	 * the ZFS DE that it should not issue any faults for unopenable
2751	 * devices.  We also iterate over the vdevs, and post a sysevent for any
2752	 * unopenable vdevs so that the normal autoreplace handler can take
2753	 * over.
2754	 */
2755	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2756		spa_check_removed(spa->spa_root_vdev);
2757		/*
2758		 * For the import case, this is done in spa_import(), because
2759		 * at this point we're using the spare definitions from
2760		 * the MOS config, not necessarily from the userland config.
2761		 */
2762		if (state != SPA_LOAD_IMPORT) {
2763			spa_aux_check_removed(&spa->spa_spares);
2764			spa_aux_check_removed(&spa->spa_l2cache);
2765		}
2766	}
2767
2768	/*
2769	 * Load the vdev state for all toplevel vdevs.
2770	 */
2771	vdev_load(rvd);
2772
2773	/*
2774	 * Propagate the leaf DTLs we just loaded all the way up the tree.
2775	 */
2776	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2777	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2778	spa_config_exit(spa, SCL_ALL, FTAG);
2779
2780	/*
2781	 * Load the DDTs (dedup tables).
2782	 */
2783	error = ddt_load(spa);
2784	if (error != 0)
2785		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2786
2787	spa_update_dspace(spa);
2788
2789	/*
2790	 * Validate the config, using the MOS config to fill in any
2791	 * information which might be missing.  If we fail to validate
2792	 * the config then declare the pool unfit for use. If we're
2793	 * assembling a pool from a split, the log is not transferred
2794	 * over.
2795	 */
2796	if (type != SPA_IMPORT_ASSEMBLE) {
2797		nvlist_t *nvconfig;
2798
2799		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2800			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2801
2802		if (!spa_config_valid(spa, nvconfig)) {
2803			nvlist_free(nvconfig);
2804			return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2805			    ENXIO));
2806		}
2807		nvlist_free(nvconfig);
2808
2809		/*
2810		 * Now that we've validated the config, check the state of the
2811		 * root vdev.  If it can't be opened, it indicates one or
2812		 * more toplevel vdevs are faulted.
2813		 */
2814		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2815			return (SET_ERROR(ENXIO));
2816
2817		if (spa_writeable(spa) && spa_check_logs(spa)) {
2818			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2819			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2820		}
2821	}
2822
2823	if (missing_feat_write) {
2824		ASSERT(state == SPA_LOAD_TRYIMPORT);
2825
2826		/*
2827		 * At this point, we know that we can open the pool in
2828		 * read-only mode but not read-write mode. We now have enough
2829		 * information and can return to userland.
2830		 */
2831		return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2832	}
2833
2834	/*
2835	 * We've successfully opened the pool, verify that we're ready
2836	 * to start pushing transactions.
2837	 */
2838	if (state != SPA_LOAD_TRYIMPORT) {
2839		if (error = spa_load_verify(spa))
2840			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2841			    error));
2842	}
2843
2844	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2845	    spa->spa_load_max_txg == UINT64_MAX)) {
2846		dmu_tx_t *tx;
2847		int need_update = B_FALSE;
2848		dsl_pool_t *dp = spa_get_dsl(spa);
2849
2850		ASSERT(state != SPA_LOAD_TRYIMPORT);
2851
2852		/*
2853		 * Claim log blocks that haven't been committed yet.
2854		 * This must all happen in a single txg.
2855		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2856		 * invoked from zil_claim_log_block()'s i/o done callback.
2857		 * Price of rollback is that we abandon the log.
2858		 */
2859		spa->spa_claiming = B_TRUE;
2860
2861		tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
2862		(void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
2863		    zil_claim, tx, DS_FIND_CHILDREN);
2864		dmu_tx_commit(tx);
2865
2866		spa->spa_claiming = B_FALSE;
2867
2868		spa_set_log_state(spa, SPA_LOG_GOOD);
2869		spa->spa_sync_on = B_TRUE;
2870		txg_sync_start(spa->spa_dsl_pool);
2871
2872		/*
2873		 * Wait for all claims to sync.  We sync up to the highest
2874		 * claimed log block birth time so that claimed log blocks
2875		 * don't appear to be from the future.  spa_claim_max_txg
2876		 * will have been set for us by either zil_check_log_chain()
2877		 * (invoked from spa_check_logs()) or zil_claim() above.
2878		 */
2879		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2880
2881		/*
2882		 * If the config cache is stale, or we have uninitialized
2883		 * metaslabs (see spa_vdev_add()), then update the config.
2884		 *
2885		 * If this is a verbatim import, trust the current
2886		 * in-core spa_config and update the disk labels.
2887		 */
2888		if (config_cache_txg != spa->spa_config_txg ||
2889		    state == SPA_LOAD_IMPORT ||
2890		    state == SPA_LOAD_RECOVER ||
2891		    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2892			need_update = B_TRUE;
2893
2894		for (int c = 0; c < rvd->vdev_children; c++)
2895			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2896				need_update = B_TRUE;
2897
2898		/*
2899		 * Update the config cache asychronously in case we're the
2900		 * root pool, in which case the config cache isn't writable yet.
2901		 */
2902		if (need_update)
2903			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2904
2905		/*
2906		 * Check all DTLs to see if anything needs resilvering.
2907		 */
2908		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2909		    vdev_resilver_needed(rvd, NULL, NULL))
2910			spa_async_request(spa, SPA_ASYNC_RESILVER);
2911
2912		/*
2913		 * Log the fact that we booted up (so that we can detect if
2914		 * we rebooted in the middle of an operation).
2915		 */
2916		spa_history_log_version(spa, "open");
2917
2918		/*
2919		 * Delete any inconsistent datasets.
2920		 */
2921		(void) dmu_objset_find(spa_name(spa),
2922		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2923
2924		/*
2925		 * Clean up any stale temporary dataset userrefs.
2926		 */
2927		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2928	}
2929
2930	return (0);
2931}
2932
2933static int
2934spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2935{
2936	int mode = spa->spa_mode;
2937
2938	spa_unload(spa);
2939	spa_deactivate(spa);
2940
2941	spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
2942
2943	spa_activate(spa, mode);
2944	spa_async_suspend(spa);
2945
2946	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2947}
2948
2949/*
2950 * If spa_load() fails this function will try loading prior txg's. If
2951 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2952 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2953 * function will not rewind the pool and will return the same error as
2954 * spa_load().
2955 */
2956static int
2957spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2958    uint64_t max_request, int rewind_flags)
2959{
2960	nvlist_t *loadinfo = NULL;
2961	nvlist_t *config = NULL;
2962	int load_error, rewind_error;
2963	uint64_t safe_rewind_txg;
2964	uint64_t min_txg;
2965
2966	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2967		spa->spa_load_max_txg = spa->spa_load_txg;
2968		spa_set_log_state(spa, SPA_LOG_CLEAR);
2969	} else {
2970		spa->spa_load_max_txg = max_request;
2971		if (max_request != UINT64_MAX)
2972			spa->spa_extreme_rewind = B_TRUE;
2973	}
2974
2975	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2976	    mosconfig);
2977	if (load_error == 0)
2978		return (0);
2979
2980	if (spa->spa_root_vdev != NULL)
2981		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2982
2983	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2984	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2985
2986	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2987		nvlist_free(config);
2988		return (load_error);
2989	}
2990
2991	if (state == SPA_LOAD_RECOVER) {
2992		/* Price of rolling back is discarding txgs, including log */
2993		spa_set_log_state(spa, SPA_LOG_CLEAR);
2994	} else {
2995		/*
2996		 * If we aren't rolling back save the load info from our first
2997		 * import attempt so that we can restore it after attempting
2998		 * to rewind.
2999		 */
3000		loadinfo = spa->spa_load_info;
3001		spa->spa_load_info = fnvlist_alloc();
3002	}
3003
3004	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
3005	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
3006	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
3007	    TXG_INITIAL : safe_rewind_txg;
3008
3009	/*
3010	 * Continue as long as we're finding errors, we're still within
3011	 * the acceptable rewind range, and we're still finding uberblocks
3012	 */
3013	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
3014	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
3015		if (spa->spa_load_max_txg < safe_rewind_txg)
3016			spa->spa_extreme_rewind = B_TRUE;
3017		rewind_error = spa_load_retry(spa, state, mosconfig);
3018	}
3019
3020	spa->spa_extreme_rewind = B_FALSE;
3021	spa->spa_load_max_txg = UINT64_MAX;
3022
3023	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
3024		spa_config_set(spa, config);
3025
3026	if (state == SPA_LOAD_RECOVER) {
3027		ASSERT3P(loadinfo, ==, NULL);
3028		return (rewind_error);
3029	} else {
3030		/* Store the rewind info as part of the initial load info */
3031		fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
3032		    spa->spa_load_info);
3033
3034		/* Restore the initial load info */
3035		fnvlist_free(spa->spa_load_info);
3036		spa->spa_load_info = loadinfo;
3037
3038		return (load_error);
3039	}
3040}
3041
3042/*
3043 * Pool Open/Import
3044 *
3045 * The import case is identical to an open except that the configuration is sent
3046 * down from userland, instead of grabbed from the configuration cache.  For the
3047 * case of an open, the pool configuration will exist in the
3048 * POOL_STATE_UNINITIALIZED state.
3049 *
3050 * The stats information (gen/count/ustats) is used to gather vdev statistics at
3051 * the same time open the pool, without having to keep around the spa_t in some
3052 * ambiguous state.
3053 */
3054static int
3055spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
3056    nvlist_t **config)
3057{
3058	spa_t *spa;
3059	spa_load_state_t state = SPA_LOAD_OPEN;
3060	int error;
3061	int locked = B_FALSE;
3062
3063	*spapp = NULL;
3064
3065	/*
3066	 * As disgusting as this is, we need to support recursive calls to this
3067	 * function because dsl_dir_open() is called during spa_load(), and ends
3068	 * up calling spa_open() again.  The real fix is to figure out how to
3069	 * avoid dsl_dir_open() calling this in the first place.
3070	 */
3071	if (mutex_owner(&spa_namespace_lock) != curthread) {
3072		mutex_enter(&spa_namespace_lock);
3073		locked = B_TRUE;
3074	}
3075
3076	if ((spa = spa_lookup(pool)) == NULL) {
3077		if (locked)
3078			mutex_exit(&spa_namespace_lock);
3079		return (SET_ERROR(ENOENT));
3080	}
3081
3082	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
3083		zpool_rewind_policy_t policy;
3084
3085		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
3086		    &policy);
3087		if (policy.zrp_request & ZPOOL_DO_REWIND)
3088			state = SPA_LOAD_RECOVER;
3089
3090		spa_activate(spa, spa_mode_global);
3091
3092		if (state != SPA_LOAD_RECOVER)
3093			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3094
3095		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
3096		    policy.zrp_request);
3097
3098		if (error == EBADF) {
3099			/*
3100			 * If vdev_validate() returns failure (indicated by
3101			 * EBADF), it indicates that one of the vdevs indicates
3102			 * that the pool has been exported or destroyed.  If
3103			 * this is the case, the config cache is out of sync and
3104			 * we should remove the pool from the namespace.
3105			 */
3106			spa_unload(spa);
3107			spa_deactivate(spa);
3108			spa_config_sync(spa, B_TRUE, B_TRUE);
3109			spa_remove(spa);
3110			if (locked)
3111				mutex_exit(&spa_namespace_lock);
3112			return (SET_ERROR(ENOENT));
3113		}
3114
3115		if (error) {
3116			/*
3117			 * We can't open the pool, but we still have useful
3118			 * information: the state of each vdev after the
3119			 * attempted vdev_open().  Return this to the user.
3120			 */
3121			if (config != NULL && spa->spa_config) {
3122				VERIFY(nvlist_dup(spa->spa_config, config,
3123				    KM_SLEEP) == 0);
3124				VERIFY(nvlist_add_nvlist(*config,
3125				    ZPOOL_CONFIG_LOAD_INFO,
3126				    spa->spa_load_info) == 0);
3127			}
3128			spa_unload(spa);
3129			spa_deactivate(spa);
3130			spa->spa_last_open_failed = error;
3131			if (locked)
3132				mutex_exit(&spa_namespace_lock);
3133			*spapp = NULL;
3134			return (error);
3135		}
3136	}
3137
3138	spa_open_ref(spa, tag);
3139
3140	if (config != NULL)
3141		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3142
3143	/*
3144	 * If we've recovered the pool, pass back any information we
3145	 * gathered while doing the load.
3146	 */
3147	if (state == SPA_LOAD_RECOVER) {
3148		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
3149		    spa->spa_load_info) == 0);
3150	}
3151
3152	if (locked) {
3153		spa->spa_last_open_failed = 0;
3154		spa->spa_last_ubsync_txg = 0;
3155		spa->spa_load_txg = 0;
3156		mutex_exit(&spa_namespace_lock);
3157	}
3158
3159	*spapp = spa;
3160
3161	return (0);
3162}
3163
3164int
3165spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3166    nvlist_t **config)
3167{
3168	return (spa_open_common(name, spapp, tag, policy, config));
3169}
3170
3171int
3172spa_open(const char *name, spa_t **spapp, void *tag)
3173{
3174	return (spa_open_common(name, spapp, tag, NULL, NULL));
3175}
3176
3177/*
3178 * Lookup the given spa_t, incrementing the inject count in the process,
3179 * preventing it from being exported or destroyed.
3180 */
3181spa_t *
3182spa_inject_addref(char *name)
3183{
3184	spa_t *spa;
3185
3186	mutex_enter(&spa_namespace_lock);
3187	if ((spa = spa_lookup(name)) == NULL) {
3188		mutex_exit(&spa_namespace_lock);
3189		return (NULL);
3190	}
3191	spa->spa_inject_ref++;
3192	mutex_exit(&spa_namespace_lock);
3193
3194	return (spa);
3195}
3196
3197void
3198spa_inject_delref(spa_t *spa)
3199{
3200	mutex_enter(&spa_namespace_lock);
3201	spa->spa_inject_ref--;
3202	mutex_exit(&spa_namespace_lock);
3203}
3204
3205/*
3206 * Add spares device information to the nvlist.
3207 */
3208static void
3209spa_add_spares(spa_t *spa, nvlist_t *config)
3210{
3211	nvlist_t **spares;
3212	uint_t i, nspares;
3213	nvlist_t *nvroot;
3214	uint64_t guid;
3215	vdev_stat_t *vs;
3216	uint_t vsc;
3217	uint64_t pool;
3218
3219	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3220
3221	if (spa->spa_spares.sav_count == 0)
3222		return;
3223
3224	VERIFY(nvlist_lookup_nvlist(config,
3225	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3226	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3227	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3228	if (nspares != 0) {
3229		VERIFY(nvlist_add_nvlist_array(nvroot,
3230		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3231		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3232		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3233
3234		/*
3235		 * Go through and find any spares which have since been
3236		 * repurposed as an active spare.  If this is the case, update
3237		 * their status appropriately.
3238		 */
3239		for (i = 0; i < nspares; i++) {
3240			VERIFY(nvlist_lookup_uint64(spares[i],
3241			    ZPOOL_CONFIG_GUID, &guid) == 0);
3242			if (spa_spare_exists(guid, &pool, NULL) &&
3243			    pool != 0ULL) {
3244				VERIFY(nvlist_lookup_uint64_array(
3245				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
3246				    (uint64_t **)&vs, &vsc) == 0);
3247				vs->vs_state = VDEV_STATE_CANT_OPEN;
3248				vs->vs_aux = VDEV_AUX_SPARED;
3249			}
3250		}
3251	}
3252}
3253
3254/*
3255 * Add l2cache device information to the nvlist, including vdev stats.
3256 */
3257static void
3258spa_add_l2cache(spa_t *spa, nvlist_t *config)
3259{
3260	nvlist_t **l2cache;
3261	uint_t i, j, nl2cache;
3262	nvlist_t *nvroot;
3263	uint64_t guid;
3264	vdev_t *vd;
3265	vdev_stat_t *vs;
3266	uint_t vsc;
3267
3268	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3269
3270	if (spa->spa_l2cache.sav_count == 0)
3271		return;
3272
3273	VERIFY(nvlist_lookup_nvlist(config,
3274	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3275	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3276	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3277	if (nl2cache != 0) {
3278		VERIFY(nvlist_add_nvlist_array(nvroot,
3279		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3280		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3281		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3282
3283		/*
3284		 * Update level 2 cache device stats.
3285		 */
3286
3287		for (i = 0; i < nl2cache; i++) {
3288			VERIFY(nvlist_lookup_uint64(l2cache[i],
3289			    ZPOOL_CONFIG_GUID, &guid) == 0);
3290
3291			vd = NULL;
3292			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3293				if (guid ==
3294				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3295					vd = spa->spa_l2cache.sav_vdevs[j];
3296					break;
3297				}
3298			}
3299			ASSERT(vd != NULL);
3300
3301			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3302			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3303			    == 0);
3304			vdev_get_stats(vd, vs);
3305		}
3306	}
3307}
3308
3309static void
3310spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3311{
3312	nvlist_t *features;
3313	zap_cursor_t zc;
3314	zap_attribute_t za;
3315
3316	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3317	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3318
3319	if (spa->spa_feat_for_read_obj != 0) {
3320		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3321		    spa->spa_feat_for_read_obj);
3322		    zap_cursor_retrieve(&zc, &za) == 0;
3323		    zap_cursor_advance(&zc)) {
3324			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3325			    za.za_num_integers == 1);
3326			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3327			    za.za_first_integer));
3328		}
3329		zap_cursor_fini(&zc);
3330	}
3331
3332	if (spa->spa_feat_for_write_obj != 0) {
3333		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3334		    spa->spa_feat_for_write_obj);
3335		    zap_cursor_retrieve(&zc, &za) == 0;
3336		    zap_cursor_advance(&zc)) {
3337			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3338			    za.za_num_integers == 1);
3339			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3340			    za.za_first_integer));
3341		}
3342		zap_cursor_fini(&zc);
3343	}
3344
3345	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3346	    features) == 0);
3347	nvlist_free(features);
3348}
3349
3350int
3351spa_get_stats(const char *name, nvlist_t **config,
3352    char *altroot, size_t buflen)
3353{
3354	int error;
3355	spa_t *spa;
3356
3357	*config = NULL;
3358	error = spa_open_common(name, &spa, FTAG, NULL, config);
3359
3360	if (spa != NULL) {
3361		/*
3362		 * This still leaves a window of inconsistency where the spares
3363		 * or l2cache devices could change and the config would be
3364		 * self-inconsistent.
3365		 */
3366		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3367
3368		if (*config != NULL) {
3369			uint64_t loadtimes[2];
3370
3371			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3372			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3373			VERIFY(nvlist_add_uint64_array(*config,
3374			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3375
3376			VERIFY(nvlist_add_uint64(*config,
3377			    ZPOOL_CONFIG_ERRCOUNT,
3378			    spa_get_errlog_size(spa)) == 0);
3379
3380			if (spa_suspended(spa))
3381				VERIFY(nvlist_add_uint64(*config,
3382				    ZPOOL_CONFIG_SUSPENDED,
3383				    spa->spa_failmode) == 0);
3384
3385			spa_add_spares(spa, *config);
3386			spa_add_l2cache(spa, *config);
3387			spa_add_feature_stats(spa, *config);
3388		}
3389	}
3390
3391	/*
3392	 * We want to get the alternate root even for faulted pools, so we cheat
3393	 * and call spa_lookup() directly.
3394	 */
3395	if (altroot) {
3396		if (spa == NULL) {
3397			mutex_enter(&spa_namespace_lock);
3398			spa = spa_lookup(name);
3399			if (spa)
3400				spa_altroot(spa, altroot, buflen);
3401			else
3402				altroot[0] = '\0';
3403			spa = NULL;
3404			mutex_exit(&spa_namespace_lock);
3405		} else {
3406			spa_altroot(spa, altroot, buflen);
3407		}
3408	}
3409
3410	if (spa != NULL) {
3411		spa_config_exit(spa, SCL_CONFIG, FTAG);
3412		spa_close(spa, FTAG);
3413	}
3414
3415	return (error);
3416}
3417
3418/*
3419 * Validate that the auxiliary device array is well formed.  We must have an
3420 * array of nvlists, each which describes a valid leaf vdev.  If this is an
3421 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3422 * specified, as long as they are well-formed.
3423 */
3424static int
3425spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3426    spa_aux_vdev_t *sav, const char *config, uint64_t version,
3427    vdev_labeltype_t label)
3428{
3429	nvlist_t **dev;
3430	uint_t i, ndev;
3431	vdev_t *vd;
3432	int error;
3433
3434	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3435
3436	/*
3437	 * It's acceptable to have no devs specified.
3438	 */
3439	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3440		return (0);
3441
3442	if (ndev == 0)
3443		return (SET_ERROR(EINVAL));
3444
3445	/*
3446	 * Make sure the pool is formatted with a version that supports this
3447	 * device type.
3448	 */
3449	if (spa_version(spa) < version)
3450		return (SET_ERROR(ENOTSUP));
3451
3452	/*
3453	 * Set the pending device list so we correctly handle device in-use
3454	 * checking.
3455	 */
3456	sav->sav_pending = dev;
3457	sav->sav_npending = ndev;
3458
3459	for (i = 0; i < ndev; i++) {
3460		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3461		    mode)) != 0)
3462			goto out;
3463
3464		if (!vd->vdev_ops->vdev_op_leaf) {
3465			vdev_free(vd);
3466			error = SET_ERROR(EINVAL);
3467			goto out;
3468		}
3469
3470		/*
3471		 * The L2ARC currently only supports disk devices in
3472		 * kernel context.  For user-level testing, we allow it.
3473		 */
3474#ifdef _KERNEL
3475		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3476		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3477			error = SET_ERROR(ENOTBLK);
3478			vdev_free(vd);
3479			goto out;
3480		}
3481#endif
3482		vd->vdev_top = vd;
3483
3484		if ((error = vdev_open(vd)) == 0 &&
3485		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
3486			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3487			    vd->vdev_guid) == 0);
3488		}
3489
3490		vdev_free(vd);
3491
3492		if (error &&
3493		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3494			goto out;
3495		else
3496			error = 0;
3497	}
3498
3499out:
3500	sav->sav_pending = NULL;
3501	sav->sav_npending = 0;
3502	return (error);
3503}
3504
3505static int
3506spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3507{
3508	int error;
3509
3510	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3511
3512	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3513	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3514	    VDEV_LABEL_SPARE)) != 0) {
3515		return (error);
3516	}
3517
3518	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3519	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3520	    VDEV_LABEL_L2CACHE));
3521}
3522
3523static void
3524spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3525    const char *config)
3526{
3527	int i;
3528
3529	if (sav->sav_config != NULL) {
3530		nvlist_t **olddevs;
3531		uint_t oldndevs;
3532		nvlist_t **newdevs;
3533
3534		/*
3535		 * Generate new dev list by concatentating with the
3536		 * current dev list.
3537		 */
3538		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3539		    &olddevs, &oldndevs) == 0);
3540
3541		newdevs = kmem_alloc(sizeof (void *) *
3542		    (ndevs + oldndevs), KM_SLEEP);
3543		for (i = 0; i < oldndevs; i++)
3544			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3545			    KM_SLEEP) == 0);
3546		for (i = 0; i < ndevs; i++)
3547			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3548			    KM_SLEEP) == 0);
3549
3550		VERIFY(nvlist_remove(sav->sav_config, config,
3551		    DATA_TYPE_NVLIST_ARRAY) == 0);
3552
3553		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3554		    config, newdevs, ndevs + oldndevs) == 0);
3555		for (i = 0; i < oldndevs + ndevs; i++)
3556			nvlist_free(newdevs[i]);
3557		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3558	} else {
3559		/*
3560		 * Generate a new dev list.
3561		 */
3562		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3563		    KM_SLEEP) == 0);
3564		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3565		    devs, ndevs) == 0);
3566	}
3567}
3568
3569/*
3570 * Stop and drop level 2 ARC devices
3571 */
3572void
3573spa_l2cache_drop(spa_t *spa)
3574{
3575	vdev_t *vd;
3576	int i;
3577	spa_aux_vdev_t *sav = &spa->spa_l2cache;
3578
3579	for (i = 0; i < sav->sav_count; i++) {
3580		uint64_t pool;
3581
3582		vd = sav->sav_vdevs[i];
3583		ASSERT(vd != NULL);
3584
3585		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3586		    pool != 0ULL && l2arc_vdev_present(vd))
3587			l2arc_remove_vdev(vd);
3588	}
3589}
3590
3591/*
3592 * Pool Creation
3593 */
3594int
3595spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3596    nvlist_t *zplprops)
3597{
3598	spa_t *spa;
3599	char *altroot = NULL;
3600	vdev_t *rvd;
3601	dsl_pool_t *dp;
3602	dmu_tx_t *tx;
3603	int error = 0;
3604	uint64_t txg = TXG_INITIAL;
3605	nvlist_t **spares, **l2cache;
3606	uint_t nspares, nl2cache;
3607	uint64_t version, obj;
3608	boolean_t has_features;
3609
3610	/*
3611	 * If this pool already exists, return failure.
3612	 */
3613	mutex_enter(&spa_namespace_lock);
3614	if (spa_lookup(pool) != NULL) {
3615		mutex_exit(&spa_namespace_lock);
3616		return (SET_ERROR(EEXIST));
3617	}
3618
3619	/*
3620	 * Allocate a new spa_t structure.
3621	 */
3622	(void) nvlist_lookup_string(props,
3623	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3624	spa = spa_add(pool, NULL, altroot);
3625	spa_activate(spa, spa_mode_global);
3626
3627	if (props && (error = spa_prop_validate(spa, props))) {
3628		spa_deactivate(spa);
3629		spa_remove(spa);
3630		mutex_exit(&spa_namespace_lock);
3631		return (error);
3632	}
3633
3634	has_features = B_FALSE;
3635	for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3636	    elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3637		if (zpool_prop_feature(nvpair_name(elem)))
3638			has_features = B_TRUE;
3639	}
3640
3641	if (has_features || nvlist_lookup_uint64(props,
3642	    zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3643		version = SPA_VERSION;
3644	}
3645	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3646
3647	spa->spa_first_txg = txg;
3648	spa->spa_uberblock.ub_txg = txg - 1;
3649	spa->spa_uberblock.ub_version = version;
3650	spa->spa_ubsync = spa->spa_uberblock;
3651	spa->spa_load_state = SPA_LOAD_CREATE;
3652
3653	/*
3654	 * Create "The Godfather" zio to hold all async IOs
3655	 */
3656	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
3657	    KM_SLEEP);
3658	for (int i = 0; i < max_ncpus; i++) {
3659		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
3660		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
3661		    ZIO_FLAG_GODFATHER);
3662	}
3663
3664	/*
3665	 * Create the root vdev.
3666	 */
3667	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3668
3669	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3670
3671	ASSERT(error != 0 || rvd != NULL);
3672	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3673
3674	if (error == 0 && !zfs_allocatable_devs(nvroot))
3675		error = SET_ERROR(EINVAL);
3676
3677	if (error == 0 &&
3678	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3679	    (error = spa_validate_aux(spa, nvroot, txg,
3680	    VDEV_ALLOC_ADD)) == 0) {
3681		for (int c = 0; c < rvd->vdev_children; c++) {
3682			vdev_metaslab_set_size(rvd->vdev_child[c]);
3683			vdev_expand(rvd->vdev_child[c], txg);
3684		}
3685	}
3686
3687	spa_config_exit(spa, SCL_ALL, FTAG);
3688
3689	if (error != 0) {
3690		spa_unload(spa);
3691		spa_deactivate(spa);
3692		spa_remove(spa);
3693		mutex_exit(&spa_namespace_lock);
3694		return (error);
3695	}
3696
3697	/*
3698	 * Get the list of spares, if specified.
3699	 */
3700	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3701	    &spares, &nspares) == 0) {
3702		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3703		    KM_SLEEP) == 0);
3704		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3705		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3706		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3707		spa_load_spares(spa);
3708		spa_config_exit(spa, SCL_ALL, FTAG);
3709		spa->spa_spares.sav_sync = B_TRUE;
3710	}
3711
3712	/*
3713	 * Get the list of level 2 cache devices, if specified.
3714	 */
3715	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3716	    &l2cache, &nl2cache) == 0) {
3717		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3718		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3719		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3720		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3721		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3722		spa_load_l2cache(spa);
3723		spa_config_exit(spa, SCL_ALL, FTAG);
3724		spa->spa_l2cache.sav_sync = B_TRUE;
3725	}
3726
3727	spa->spa_is_initializing = B_TRUE;
3728	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3729	spa->spa_meta_objset = dp->dp_meta_objset;
3730	spa->spa_is_initializing = B_FALSE;
3731
3732	/*
3733	 * Create DDTs (dedup tables).
3734	 */
3735	ddt_create(spa);
3736
3737	spa_update_dspace(spa);
3738
3739	tx = dmu_tx_create_assigned(dp, txg);
3740
3741	/*
3742	 * Create the pool config object.
3743	 */
3744	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3745	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3746	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3747
3748	if (zap_add(spa->spa_meta_objset,
3749	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3750	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3751		cmn_err(CE_PANIC, "failed to add pool config");
3752	}
3753
3754	if (spa_version(spa) >= SPA_VERSION_FEATURES)
3755		spa_feature_create_zap_objects(spa, tx);
3756
3757	if (zap_add(spa->spa_meta_objset,
3758	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3759	    sizeof (uint64_t), 1, &version, tx) != 0) {
3760		cmn_err(CE_PANIC, "failed to add pool version");
3761	}
3762
3763	/* Newly created pools with the right version are always deflated. */
3764	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3765		spa->spa_deflate = TRUE;
3766		if (zap_add(spa->spa_meta_objset,
3767		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3768		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3769			cmn_err(CE_PANIC, "failed to add deflate");
3770		}
3771	}
3772
3773	/*
3774	 * Create the deferred-free bpobj.  Turn off compression
3775	 * because sync-to-convergence takes longer if the blocksize
3776	 * keeps changing.
3777	 */
3778	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3779	dmu_object_set_compress(spa->spa_meta_objset, obj,
3780	    ZIO_COMPRESS_OFF, tx);
3781	if (zap_add(spa->spa_meta_objset,
3782	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3783	    sizeof (uint64_t), 1, &obj, tx) != 0) {
3784		cmn_err(CE_PANIC, "failed to add bpobj");
3785	}
3786	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3787	    spa->spa_meta_objset, obj));
3788
3789	/*
3790	 * Create the pool's history object.
3791	 */
3792	if (version >= SPA_VERSION_ZPOOL_HISTORY)
3793		spa_history_create_obj(spa, tx);
3794
3795	/*
3796	 * Generate some random noise for salted checksums to operate on.
3797	 */
3798	(void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
3799	    sizeof (spa->spa_cksum_salt.zcs_bytes));
3800
3801	/*
3802	 * Set pool properties.
3803	 */
3804	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3805	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3806	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3807	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3808
3809	if (props != NULL) {
3810		spa_configfile_set(spa, props, B_FALSE);
3811		spa_sync_props(props, tx);
3812	}
3813
3814	dmu_tx_commit(tx);
3815
3816	spa->spa_sync_on = B_TRUE;
3817	txg_sync_start(spa->spa_dsl_pool);
3818
3819	/*
3820	 * We explicitly wait for the first transaction to complete so that our
3821	 * bean counters are appropriately updated.
3822	 */
3823	txg_wait_synced(spa->spa_dsl_pool, txg);
3824
3825	spa_config_sync(spa, B_FALSE, B_TRUE);
3826	spa_event_notify(spa, NULL, ESC_ZFS_POOL_CREATE);
3827
3828	spa_history_log_version(spa, "create");
3829
3830	/*
3831	 * Don't count references from objsets that are already closed
3832	 * and are making their way through the eviction process.
3833	 */
3834	spa_evicting_os_wait(spa);
3835	spa->spa_minref = refcount_count(&spa->spa_refcount);
3836	spa->spa_load_state = SPA_LOAD_NONE;
3837
3838	mutex_exit(&spa_namespace_lock);
3839
3840	return (0);
3841}
3842
3843#ifdef _KERNEL
3844/*
3845 * Get the root pool information from the root disk, then import the root pool
3846 * during the system boot up time.
3847 */
3848extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3849
3850static nvlist_t *
3851spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3852{
3853	nvlist_t *config;
3854	nvlist_t *nvtop, *nvroot;
3855	uint64_t pgid;
3856
3857	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3858		return (NULL);
3859
3860	/*
3861	 * Add this top-level vdev to the child array.
3862	 */
3863	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3864	    &nvtop) == 0);
3865	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3866	    &pgid) == 0);
3867	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3868
3869	/*
3870	 * Put this pool's top-level vdevs into a root vdev.
3871	 */
3872	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3873	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3874	    VDEV_TYPE_ROOT) == 0);
3875	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3876	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3877	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3878	    &nvtop, 1) == 0);
3879
3880	/*
3881	 * Replace the existing vdev_tree with the new root vdev in
3882	 * this pool's configuration (remove the old, add the new).
3883	 */
3884	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3885	nvlist_free(nvroot);
3886	return (config);
3887}
3888
3889/*
3890 * Walk the vdev tree and see if we can find a device with "better"
3891 * configuration. A configuration is "better" if the label on that
3892 * device has a more recent txg.
3893 */
3894static void
3895spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3896{
3897	for (int c = 0; c < vd->vdev_children; c++)
3898		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3899
3900	if (vd->vdev_ops->vdev_op_leaf) {
3901		nvlist_t *label;
3902		uint64_t label_txg;
3903
3904		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3905		    &label) != 0)
3906			return;
3907
3908		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3909		    &label_txg) == 0);
3910
3911		/*
3912		 * Do we have a better boot device?
3913		 */
3914		if (label_txg > *txg) {
3915			*txg = label_txg;
3916			*avd = vd;
3917		}
3918		nvlist_free(label);
3919	}
3920}
3921
3922/*
3923 * Import a root pool.
3924 *
3925 * For x86. devpath_list will consist of devid and/or physpath name of
3926 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3927 * The GRUB "findroot" command will return the vdev we should boot.
3928 *
3929 * For Sparc, devpath_list consists the physpath name of the booting device
3930 * no matter the rootpool is a single device pool or a mirrored pool.
3931 * e.g.
3932 *	"/pci@1f,0/ide@d/disk@0,0:a"
3933 */
3934int
3935spa_import_rootpool(char *devpath, char *devid)
3936{
3937	spa_t *spa;
3938	vdev_t *rvd, *bvd, *avd = NULL;
3939	nvlist_t *config, *nvtop;
3940	uint64_t guid, txg;
3941	char *pname;
3942	int error;
3943
3944	/*
3945	 * Read the label from the boot device and generate a configuration.
3946	 */
3947	config = spa_generate_rootconf(devpath, devid, &guid);
3948#if defined(_OBP) && defined(_KERNEL)
3949	if (config == NULL) {
3950		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3951			/* iscsi boot */
3952			get_iscsi_bootpath_phy(devpath);
3953			config = spa_generate_rootconf(devpath, devid, &guid);
3954		}
3955	}
3956#endif
3957	if (config == NULL) {
3958		cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3959		    devpath);
3960		return (SET_ERROR(EIO));
3961	}
3962
3963	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3964	    &pname) == 0);
3965	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3966
3967	mutex_enter(&spa_namespace_lock);
3968	if ((spa = spa_lookup(pname)) != NULL) {
3969		/*
3970		 * Remove the existing root pool from the namespace so that we
3971		 * can replace it with the correct config we just read in.
3972		 */
3973		spa_remove(spa);
3974	}
3975
3976	spa = spa_add(pname, config, NULL);
3977	spa->spa_is_root = B_TRUE;
3978	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3979
3980	/*
3981	 * Build up a vdev tree based on the boot device's label config.
3982	 */
3983	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3984	    &nvtop) == 0);
3985	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3986	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3987	    VDEV_ALLOC_ROOTPOOL);
3988	spa_config_exit(spa, SCL_ALL, FTAG);
3989	if (error) {
3990		mutex_exit(&spa_namespace_lock);
3991		nvlist_free(config);
3992		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3993		    pname);
3994		return (error);
3995	}
3996
3997	/*
3998	 * Get the boot vdev.
3999	 */
4000	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
4001		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
4002		    (u_longlong_t)guid);
4003		error = SET_ERROR(ENOENT);
4004		goto out;
4005	}
4006
4007	/*
4008	 * Determine if there is a better boot device.
4009	 */
4010	avd = bvd;
4011	spa_alt_rootvdev(rvd, &avd, &txg);
4012	if (avd != bvd) {
4013		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
4014		    "try booting from '%s'", avd->vdev_path);
4015		error = SET_ERROR(EINVAL);
4016		goto out;
4017	}
4018
4019	/*
4020	 * If the boot device is part of a spare vdev then ensure that
4021	 * we're booting off the active spare.
4022	 */
4023	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
4024	    !bvd->vdev_isspare) {
4025		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
4026		    "try booting from '%s'",
4027		    bvd->vdev_parent->
4028		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
4029		error = SET_ERROR(EINVAL);
4030		goto out;
4031	}
4032
4033	error = 0;
4034out:
4035	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4036	vdev_free(rvd);
4037	spa_config_exit(spa, SCL_ALL, FTAG);
4038	mutex_exit(&spa_namespace_lock);
4039
4040	nvlist_free(config);
4041	return (error);
4042}
4043
4044#endif
4045
4046/*
4047 * Import a non-root pool into the system.
4048 */
4049int
4050spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
4051{
4052	spa_t *spa;
4053	char *altroot = NULL;
4054	spa_load_state_t state = SPA_LOAD_IMPORT;
4055	zpool_rewind_policy_t policy;
4056	uint64_t mode = spa_mode_global;
4057	uint64_t readonly = B_FALSE;
4058	int error;
4059	nvlist_t *nvroot;
4060	nvlist_t **spares, **l2cache;
4061	uint_t nspares, nl2cache;
4062
4063	/*
4064	 * If a pool with this name exists, return failure.
4065	 */
4066	mutex_enter(&spa_namespace_lock);
4067	if (spa_lookup(pool) != NULL) {
4068		mutex_exit(&spa_namespace_lock);
4069		return (SET_ERROR(EEXIST));
4070	}
4071
4072	/*
4073	 * Create and initialize the spa structure.
4074	 */
4075	(void) nvlist_lookup_string(props,
4076	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4077	(void) nvlist_lookup_uint64(props,
4078	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
4079	if (readonly)
4080		mode = FREAD;
4081	spa = spa_add(pool, config, altroot);
4082	spa->spa_import_flags = flags;
4083
4084	/*
4085	 * Verbatim import - Take a pool and insert it into the namespace
4086	 * as if it had been loaded at boot.
4087	 */
4088	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
4089		if (props != NULL)
4090			spa_configfile_set(spa, props, B_FALSE);
4091
4092		spa_config_sync(spa, B_FALSE, B_TRUE);
4093		spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4094
4095		mutex_exit(&spa_namespace_lock);
4096		return (0);
4097	}
4098
4099	spa_activate(spa, mode);
4100
4101	/*
4102	 * Don't start async tasks until we know everything is healthy.
4103	 */
4104	spa_async_suspend(spa);
4105
4106	zpool_get_rewind_policy(config, &policy);
4107	if (policy.zrp_request & ZPOOL_DO_REWIND)
4108		state = SPA_LOAD_RECOVER;
4109
4110	/*
4111	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
4112	 * because the user-supplied config is actually the one to trust when
4113	 * doing an import.
4114	 */
4115	if (state != SPA_LOAD_RECOVER)
4116		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4117
4118	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4119	    policy.zrp_request);
4120
4121	/*
4122	 * Propagate anything learned while loading the pool and pass it
4123	 * back to caller (i.e. rewind info, missing devices, etc).
4124	 */
4125	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4126	    spa->spa_load_info) == 0);
4127
4128	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4129	/*
4130	 * Toss any existing sparelist, as it doesn't have any validity
4131	 * anymore, and conflicts with spa_has_spare().
4132	 */
4133	if (spa->spa_spares.sav_config) {
4134		nvlist_free(spa->spa_spares.sav_config);
4135		spa->spa_spares.sav_config = NULL;
4136		spa_load_spares(spa);
4137	}
4138	if (spa->spa_l2cache.sav_config) {
4139		nvlist_free(spa->spa_l2cache.sav_config);
4140		spa->spa_l2cache.sav_config = NULL;
4141		spa_load_l2cache(spa);
4142	}
4143
4144	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4145	    &nvroot) == 0);
4146	if (error == 0)
4147		error = spa_validate_aux(spa, nvroot, -1ULL,
4148		    VDEV_ALLOC_SPARE);
4149	if (error == 0)
4150		error = spa_validate_aux(spa, nvroot, -1ULL,
4151		    VDEV_ALLOC_L2CACHE);
4152	spa_config_exit(spa, SCL_ALL, FTAG);
4153
4154	if (props != NULL)
4155		spa_configfile_set(spa, props, B_FALSE);
4156
4157	if (error != 0 || (props && spa_writeable(spa) &&
4158	    (error = spa_prop_set(spa, props)))) {
4159		spa_unload(spa);
4160		spa_deactivate(spa);
4161		spa_remove(spa);
4162		mutex_exit(&spa_namespace_lock);
4163		return (error);
4164	}
4165
4166	spa_async_resume(spa);
4167
4168	/*
4169	 * Override any spares and level 2 cache devices as specified by
4170	 * the user, as these may have correct device names/devids, etc.
4171	 */
4172	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4173	    &spares, &nspares) == 0) {
4174		if (spa->spa_spares.sav_config)
4175			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4176			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4177		else
4178			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4179			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4180		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4181		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4182		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4183		spa_load_spares(spa);
4184		spa_config_exit(spa, SCL_ALL, FTAG);
4185		spa->spa_spares.sav_sync = B_TRUE;
4186	}
4187	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4188	    &l2cache, &nl2cache) == 0) {
4189		if (spa->spa_l2cache.sav_config)
4190			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4191			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4192		else
4193			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4194			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4195		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4196		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4197		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4198		spa_load_l2cache(spa);
4199		spa_config_exit(spa, SCL_ALL, FTAG);
4200		spa->spa_l2cache.sav_sync = B_TRUE;
4201	}
4202
4203	/*
4204	 * Check for any removed devices.
4205	 */
4206	if (spa->spa_autoreplace) {
4207		spa_aux_check_removed(&spa->spa_spares);
4208		spa_aux_check_removed(&spa->spa_l2cache);
4209	}
4210
4211	if (spa_writeable(spa)) {
4212		/*
4213		 * Update the config cache to include the newly-imported pool.
4214		 */
4215		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4216	}
4217
4218	/*
4219	 * It's possible that the pool was expanded while it was exported.
4220	 * We kick off an async task to handle this for us.
4221	 */
4222	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4223
4224	spa_history_log_version(spa, "import");
4225
4226	spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4227
4228	mutex_exit(&spa_namespace_lock);
4229
4230	return (0);
4231}
4232
4233nvlist_t *
4234spa_tryimport(nvlist_t *tryconfig)
4235{
4236	nvlist_t *config = NULL;
4237	char *poolname;
4238	spa_t *spa;
4239	uint64_t state;
4240	int error;
4241
4242	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4243		return (NULL);
4244
4245	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4246		return (NULL);
4247
4248	/*
4249	 * Create and initialize the spa structure.
4250	 */
4251	mutex_enter(&spa_namespace_lock);
4252	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4253	spa_activate(spa, FREAD);
4254
4255	/*
4256	 * Pass off the heavy lifting to spa_load().
4257	 * Pass TRUE for mosconfig because the user-supplied config
4258	 * is actually the one to trust when doing an import.
4259	 */
4260	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4261
4262	/*
4263	 * If 'tryconfig' was at least parsable, return the current config.
4264	 */
4265	if (spa->spa_root_vdev != NULL) {
4266		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4267		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4268		    poolname) == 0);
4269		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4270		    state) == 0);
4271		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4272		    spa->spa_uberblock.ub_timestamp) == 0);
4273		VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4274		    spa->spa_load_info) == 0);
4275
4276		/*
4277		 * If the bootfs property exists on this pool then we
4278		 * copy it out so that external consumers can tell which
4279		 * pools are bootable.
4280		 */
4281		if ((!error || error == EEXIST) && spa->spa_bootfs) {
4282			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4283
4284			/*
4285			 * We have to play games with the name since the
4286			 * pool was opened as TRYIMPORT_NAME.
4287			 */
4288			if (dsl_dsobj_to_dsname(spa_name(spa),
4289			    spa->spa_bootfs, tmpname) == 0) {
4290				char *cp;
4291				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4292
4293				cp = strchr(tmpname, '/');
4294				if (cp == NULL) {
4295					(void) strlcpy(dsname, tmpname,
4296					    MAXPATHLEN);
4297				} else {
4298					(void) snprintf(dsname, MAXPATHLEN,
4299					    "%s/%s", poolname, ++cp);
4300				}
4301				VERIFY(nvlist_add_string(config,
4302				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4303				kmem_free(dsname, MAXPATHLEN);
4304			}
4305			kmem_free(tmpname, MAXPATHLEN);
4306		}
4307
4308		/*
4309		 * Add the list of hot spares and level 2 cache devices.
4310		 */
4311		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4312		spa_add_spares(spa, config);
4313		spa_add_l2cache(spa, config);
4314		spa_config_exit(spa, SCL_CONFIG, FTAG);
4315	}
4316
4317	spa_unload(spa);
4318	spa_deactivate(spa);
4319	spa_remove(spa);
4320	mutex_exit(&spa_namespace_lock);
4321
4322	return (config);
4323}
4324
4325/*
4326 * Pool export/destroy
4327 *
4328 * The act of destroying or exporting a pool is very simple.  We make sure there
4329 * is no more pending I/O and any references to the pool are gone.  Then, we
4330 * update the pool state and sync all the labels to disk, removing the
4331 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4332 * we don't sync the labels or remove the configuration cache.
4333 */
4334static int
4335spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4336    boolean_t force, boolean_t hardforce)
4337{
4338	spa_t *spa;
4339
4340	if (oldconfig)
4341		*oldconfig = NULL;
4342
4343	if (!(spa_mode_global & FWRITE))
4344		return (SET_ERROR(EROFS));
4345
4346	mutex_enter(&spa_namespace_lock);
4347	if ((spa = spa_lookup(pool)) == NULL) {
4348		mutex_exit(&spa_namespace_lock);
4349		return (SET_ERROR(ENOENT));
4350	}
4351
4352	/*
4353	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4354	 * reacquire the namespace lock, and see if we can export.
4355	 */
4356	spa_open_ref(spa, FTAG);
4357	mutex_exit(&spa_namespace_lock);
4358	spa_async_suspend(spa);
4359	mutex_enter(&spa_namespace_lock);
4360	spa_close(spa, FTAG);
4361
4362	/*
4363	 * The pool will be in core if it's openable,
4364	 * in which case we can modify its state.
4365	 */
4366	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4367		/*
4368		 * Objsets may be open only because they're dirty, so we
4369		 * have to force it to sync before checking spa_refcnt.
4370		 */
4371		txg_wait_synced(spa->spa_dsl_pool, 0);
4372		spa_evicting_os_wait(spa);
4373
4374		/*
4375		 * A pool cannot be exported or destroyed if there are active
4376		 * references.  If we are resetting a pool, allow references by
4377		 * fault injection handlers.
4378		 */
4379		if (!spa_refcount_zero(spa) ||
4380		    (spa->spa_inject_ref != 0 &&
4381		    new_state != POOL_STATE_UNINITIALIZED)) {
4382			spa_async_resume(spa);
4383			mutex_exit(&spa_namespace_lock);
4384			return (SET_ERROR(EBUSY));
4385		}
4386
4387		/*
4388		 * A pool cannot be exported if it has an active shared spare.
4389		 * This is to prevent other pools stealing the active spare
4390		 * from an exported pool. At user's own will, such pool can
4391		 * be forcedly exported.
4392		 */
4393		if (!force && new_state == POOL_STATE_EXPORTED &&
4394		    spa_has_active_shared_spare(spa)) {
4395			spa_async_resume(spa);
4396			mutex_exit(&spa_namespace_lock);
4397			return (SET_ERROR(EXDEV));
4398		}
4399
4400		/*
4401		 * We want this to be reflected on every label,
4402		 * so mark them all dirty.  spa_unload() will do the
4403		 * final sync that pushes these changes out.
4404		 */
4405		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4406			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4407			spa->spa_state = new_state;
4408			spa->spa_final_txg = spa_last_synced_txg(spa) +
4409			    TXG_DEFER_SIZE + 1;
4410			vdev_config_dirty(spa->spa_root_vdev);
4411			spa_config_exit(spa, SCL_ALL, FTAG);
4412		}
4413	}
4414
4415	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4416
4417	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4418		spa_unload(spa);
4419		spa_deactivate(spa);
4420	}
4421
4422	if (oldconfig && spa->spa_config)
4423		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4424
4425	if (new_state != POOL_STATE_UNINITIALIZED) {
4426		if (!hardforce)
4427			spa_config_sync(spa, B_TRUE, B_TRUE);
4428		spa_remove(spa);
4429	}
4430	mutex_exit(&spa_namespace_lock);
4431
4432	return (0);
4433}
4434
4435/*
4436 * Destroy a storage pool.
4437 */
4438int
4439spa_destroy(char *pool)
4440{
4441	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4442	    B_FALSE, B_FALSE));
4443}
4444
4445/*
4446 * Export a storage pool.
4447 */
4448int
4449spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4450    boolean_t hardforce)
4451{
4452	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4453	    force, hardforce));
4454}
4455
4456/*
4457 * Similar to spa_export(), this unloads the spa_t without actually removing it
4458 * from the namespace in any way.
4459 */
4460int
4461spa_reset(char *pool)
4462{
4463	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4464	    B_FALSE, B_FALSE));
4465}
4466
4467/*
4468 * ==========================================================================
4469 * Device manipulation
4470 * ==========================================================================
4471 */
4472
4473/*
4474 * Add a device to a storage pool.
4475 */
4476int
4477spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4478{
4479	uint64_t txg, id;
4480	int error;
4481	vdev_t *rvd = spa->spa_root_vdev;
4482	vdev_t *vd, *tvd;
4483	nvlist_t **spares, **l2cache;
4484	uint_t nspares, nl2cache;
4485
4486	ASSERT(spa_writeable(spa));
4487
4488	txg = spa_vdev_enter(spa);
4489
4490	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4491	    VDEV_ALLOC_ADD)) != 0)
4492		return (spa_vdev_exit(spa, NULL, txg, error));
4493
4494	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
4495
4496	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4497	    &nspares) != 0)
4498		nspares = 0;
4499
4500	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4501	    &nl2cache) != 0)
4502		nl2cache = 0;
4503
4504	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4505		return (spa_vdev_exit(spa, vd, txg, EINVAL));
4506
4507	if (vd->vdev_children != 0 &&
4508	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
4509		return (spa_vdev_exit(spa, vd, txg, error));
4510
4511	/*
4512	 * We must validate the spares and l2cache devices after checking the
4513	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
4514	 */
4515	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4516		return (spa_vdev_exit(spa, vd, txg, error));
4517
4518	/*
4519	 * Transfer each new top-level vdev from vd to rvd.
4520	 */
4521	for (int c = 0; c < vd->vdev_children; c++) {
4522
4523		/*
4524		 * Set the vdev id to the first hole, if one exists.
4525		 */
4526		for (id = 0; id < rvd->vdev_children; id++) {
4527			if (rvd->vdev_child[id]->vdev_ishole) {
4528				vdev_free(rvd->vdev_child[id]);
4529				break;
4530			}
4531		}
4532		tvd = vd->vdev_child[c];
4533		vdev_remove_child(vd, tvd);
4534		tvd->vdev_id = id;
4535		vdev_add_child(rvd, tvd);
4536		vdev_config_dirty(tvd);
4537	}
4538
4539	if (nspares != 0) {
4540		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4541		    ZPOOL_CONFIG_SPARES);
4542		spa_load_spares(spa);
4543		spa->spa_spares.sav_sync = B_TRUE;
4544	}
4545
4546	if (nl2cache != 0) {
4547		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4548		    ZPOOL_CONFIG_L2CACHE);
4549		spa_load_l2cache(spa);
4550		spa->spa_l2cache.sav_sync = B_TRUE;
4551	}
4552
4553	/*
4554	 * We have to be careful when adding new vdevs to an existing pool.
4555	 * If other threads start allocating from these vdevs before we
4556	 * sync the config cache, and we lose power, then upon reboot we may
4557	 * fail to open the pool because there are DVAs that the config cache
4558	 * can't translate.  Therefore, we first add the vdevs without
4559	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4560	 * and then let spa_config_update() initialize the new metaslabs.
4561	 *
4562	 * spa_load() checks for added-but-not-initialized vdevs, so that
4563	 * if we lose power at any point in this sequence, the remaining
4564	 * steps will be completed the next time we load the pool.
4565	 */
4566	(void) spa_vdev_exit(spa, vd, txg, 0);
4567
4568	mutex_enter(&spa_namespace_lock);
4569	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4570	spa_event_notify(spa, NULL, ESC_ZFS_VDEV_ADD);
4571	mutex_exit(&spa_namespace_lock);
4572
4573	return (0);
4574}
4575
4576/*
4577 * Attach a device to a mirror.  The arguments are the path to any device
4578 * in the mirror, and the nvroot for the new device.  If the path specifies
4579 * a device that is not mirrored, we automatically insert the mirror vdev.
4580 *
4581 * If 'replacing' is specified, the new device is intended to replace the
4582 * existing device; in this case the two devices are made into their own
4583 * mirror using the 'replacing' vdev, which is functionally identical to
4584 * the mirror vdev (it actually reuses all the same ops) but has a few
4585 * extra rules: you can't attach to it after it's been created, and upon
4586 * completion of resilvering, the first disk (the one being replaced)
4587 * is automatically detached.
4588 */
4589int
4590spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4591{
4592	uint64_t txg, dtl_max_txg;
4593	vdev_t *rvd = spa->spa_root_vdev;
4594	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4595	vdev_ops_t *pvops;
4596	char *oldvdpath, *newvdpath;
4597	int newvd_isspare;
4598	int error;
4599
4600	ASSERT(spa_writeable(spa));
4601
4602	txg = spa_vdev_enter(spa);
4603
4604	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4605
4606	if (oldvd == NULL)
4607		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4608
4609	if (!oldvd->vdev_ops->vdev_op_leaf)
4610		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4611
4612	pvd = oldvd->vdev_parent;
4613
4614	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4615	    VDEV_ALLOC_ATTACH)) != 0)
4616		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4617
4618	if (newrootvd->vdev_children != 1)
4619		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4620
4621	newvd = newrootvd->vdev_child[0];
4622
4623	if (!newvd->vdev_ops->vdev_op_leaf)
4624		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4625
4626	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4627		return (spa_vdev_exit(spa, newrootvd, txg, error));
4628
4629	/*
4630	 * Spares can't replace logs
4631	 */
4632	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4633		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4634
4635	if (!replacing) {
4636		/*
4637		 * For attach, the only allowable parent is a mirror or the root
4638		 * vdev.
4639		 */
4640		if (pvd->vdev_ops != &vdev_mirror_ops &&
4641		    pvd->vdev_ops != &vdev_root_ops)
4642			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4643
4644		pvops = &vdev_mirror_ops;
4645	} else {
4646		/*
4647		 * Active hot spares can only be replaced by inactive hot
4648		 * spares.
4649		 */
4650		if (pvd->vdev_ops == &vdev_spare_ops &&
4651		    oldvd->vdev_isspare &&
4652		    !spa_has_spare(spa, newvd->vdev_guid))
4653			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4654
4655		/*
4656		 * If the source is a hot spare, and the parent isn't already a
4657		 * spare, then we want to create a new hot spare.  Otherwise, we
4658		 * want to create a replacing vdev.  The user is not allowed to
4659		 * attach to a spared vdev child unless the 'isspare' state is
4660		 * the same (spare replaces spare, non-spare replaces
4661		 * non-spare).
4662		 */
4663		if (pvd->vdev_ops == &vdev_replacing_ops &&
4664		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4665			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4666		} else if (pvd->vdev_ops == &vdev_spare_ops &&
4667		    newvd->vdev_isspare != oldvd->vdev_isspare) {
4668			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4669		}
4670
4671		if (newvd->vdev_isspare)
4672			pvops = &vdev_spare_ops;
4673		else
4674			pvops = &vdev_replacing_ops;
4675	}
4676
4677	/*
4678	 * Make sure the new device is big enough.
4679	 */
4680	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4681		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4682
4683	/*
4684	 * The new device cannot have a higher alignment requirement
4685	 * than the top-level vdev.
4686	 */
4687	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4688		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4689
4690	/*
4691	 * If this is an in-place replacement, update oldvd's path and devid
4692	 * to make it distinguishable from newvd, and unopenable from now on.
4693	 */
4694	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4695		spa_strfree(oldvd->vdev_path);
4696		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4697		    KM_SLEEP);
4698		(void) sprintf(oldvd->vdev_path, "%s/%s",
4699		    newvd->vdev_path, "old");
4700		if (oldvd->vdev_devid != NULL) {
4701			spa_strfree(oldvd->vdev_devid);
4702			oldvd->vdev_devid = NULL;
4703		}
4704	}
4705
4706	/* mark the device being resilvered */
4707	newvd->vdev_resilver_txg = txg;
4708
4709	/*
4710	 * If the parent is not a mirror, or if we're replacing, insert the new
4711	 * mirror/replacing/spare vdev above oldvd.
4712	 */
4713	if (pvd->vdev_ops != pvops)
4714		pvd = vdev_add_parent(oldvd, pvops);
4715
4716	ASSERT(pvd->vdev_top->vdev_parent == rvd);
4717	ASSERT(pvd->vdev_ops == pvops);
4718	ASSERT(oldvd->vdev_parent == pvd);
4719
4720	/*
4721	 * Extract the new device from its root and add it to pvd.
4722	 */
4723	vdev_remove_child(newrootvd, newvd);
4724	newvd->vdev_id = pvd->vdev_children;
4725	newvd->vdev_crtxg = oldvd->vdev_crtxg;
4726	vdev_add_child(pvd, newvd);
4727
4728	tvd = newvd->vdev_top;
4729	ASSERT(pvd->vdev_top == tvd);
4730	ASSERT(tvd->vdev_parent == rvd);
4731
4732	vdev_config_dirty(tvd);
4733
4734	/*
4735	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4736	 * for any dmu_sync-ed blocks.  It will propagate upward when
4737	 * spa_vdev_exit() calls vdev_dtl_reassess().
4738	 */
4739	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4740
4741	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4742	    dtl_max_txg - TXG_INITIAL);
4743
4744	if (newvd->vdev_isspare) {
4745		spa_spare_activate(newvd);
4746		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4747	}
4748
4749	oldvdpath = spa_strdup(oldvd->vdev_path);
4750	newvdpath = spa_strdup(newvd->vdev_path);
4751	newvd_isspare = newvd->vdev_isspare;
4752
4753	/*
4754	 * Mark newvd's DTL dirty in this txg.
4755	 */
4756	vdev_dirty(tvd, VDD_DTL, newvd, txg);
4757
4758	/*
4759	 * Schedule the resilver to restart in the future. We do this to
4760	 * ensure that dmu_sync-ed blocks have been stitched into the
4761	 * respective datasets.
4762	 */
4763	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4764
4765	if (spa->spa_bootfs)
4766		spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4767
4768	spa_event_notify(spa, newvd, ESC_ZFS_VDEV_ATTACH);
4769
4770	/*
4771	 * Commit the config
4772	 */
4773	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4774
4775	spa_history_log_internal(spa, "vdev attach", NULL,
4776	    "%s vdev=%s %s vdev=%s",
4777	    replacing && newvd_isspare ? "spare in" :
4778	    replacing ? "replace" : "attach", newvdpath,
4779	    replacing ? "for" : "to", oldvdpath);
4780
4781	spa_strfree(oldvdpath);
4782	spa_strfree(newvdpath);
4783
4784	return (0);
4785}
4786
4787/*
4788 * Detach a device from a mirror or replacing vdev.
4789 *
4790 * If 'replace_done' is specified, only detach if the parent
4791 * is a replacing vdev.
4792 */
4793int
4794spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4795{
4796	uint64_t txg;
4797	int error;
4798	vdev_t *rvd = spa->spa_root_vdev;
4799	vdev_t *vd, *pvd, *cvd, *tvd;
4800	boolean_t unspare = B_FALSE;
4801	uint64_t unspare_guid = 0;
4802	char *vdpath;
4803
4804	ASSERT(spa_writeable(spa));
4805
4806	txg = spa_vdev_enter(spa);
4807
4808	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4809
4810	if (vd == NULL)
4811		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4812
4813	if (!vd->vdev_ops->vdev_op_leaf)
4814		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4815
4816	pvd = vd->vdev_parent;
4817
4818	/*
4819	 * If the parent/child relationship is not as expected, don't do it.
4820	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4821	 * vdev that's replacing B with C.  The user's intent in replacing
4822	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4823	 * the replace by detaching C, the expected behavior is to end up
4824	 * M(A,B).  But suppose that right after deciding to detach C,
4825	 * the replacement of B completes.  We would have M(A,C), and then
4826	 * ask to detach C, which would leave us with just A -- not what
4827	 * the user wanted.  To prevent this, we make sure that the
4828	 * parent/child relationship hasn't changed -- in this example,
4829	 * that C's parent is still the replacing vdev R.
4830	 */
4831	if (pvd->vdev_guid != pguid && pguid != 0)
4832		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4833
4834	/*
4835	 * Only 'replacing' or 'spare' vdevs can be replaced.
4836	 */
4837	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4838	    pvd->vdev_ops != &vdev_spare_ops)
4839		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4840
4841	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4842	    spa_version(spa) >= SPA_VERSION_SPARES);
4843
4844	/*
4845	 * Only mirror, replacing, and spare vdevs support detach.
4846	 */
4847	if (pvd->vdev_ops != &vdev_replacing_ops &&
4848	    pvd->vdev_ops != &vdev_mirror_ops &&
4849	    pvd->vdev_ops != &vdev_spare_ops)
4850		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4851
4852	/*
4853	 * If this device has the only valid copy of some data,
4854	 * we cannot safely detach it.
4855	 */
4856	if (vdev_dtl_required(vd))
4857		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4858
4859	ASSERT(pvd->vdev_children >= 2);
4860
4861	/*
4862	 * If we are detaching the second disk from a replacing vdev, then
4863	 * check to see if we changed the original vdev's path to have "/old"
4864	 * at the end in spa_vdev_attach().  If so, undo that change now.
4865	 */
4866	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4867	    vd->vdev_path != NULL) {
4868		size_t len = strlen(vd->vdev_path);
4869
4870		for (int c = 0; c < pvd->vdev_children; c++) {
4871			cvd = pvd->vdev_child[c];
4872
4873			if (cvd == vd || cvd->vdev_path == NULL)
4874				continue;
4875
4876			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4877			    strcmp(cvd->vdev_path + len, "/old") == 0) {
4878				spa_strfree(cvd->vdev_path);
4879				cvd->vdev_path = spa_strdup(vd->vdev_path);
4880				break;
4881			}
4882		}
4883	}
4884
4885	/*
4886	 * If we are detaching the original disk from a spare, then it implies
4887	 * that the spare should become a real disk, and be removed from the
4888	 * active spare list for the pool.
4889	 */
4890	if (pvd->vdev_ops == &vdev_spare_ops &&
4891	    vd->vdev_id == 0 &&
4892	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4893		unspare = B_TRUE;
4894
4895	/*
4896	 * Erase the disk labels so the disk can be used for other things.
4897	 * This must be done after all other error cases are handled,
4898	 * but before we disembowel vd (so we can still do I/O to it).
4899	 * But if we can't do it, don't treat the error as fatal --
4900	 * it may be that the unwritability of the disk is the reason
4901	 * it's being detached!
4902	 */
4903	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4904
4905	/*
4906	 * Remove vd from its parent and compact the parent's children.
4907	 */
4908	vdev_remove_child(pvd, vd);
4909	vdev_compact_children(pvd);
4910
4911	/*
4912	 * Remember one of the remaining children so we can get tvd below.
4913	 */
4914	cvd = pvd->vdev_child[pvd->vdev_children - 1];
4915
4916	/*
4917	 * If we need to remove the remaining child from the list of hot spares,
4918	 * do it now, marking the vdev as no longer a spare in the process.
4919	 * We must do this before vdev_remove_parent(), because that can
4920	 * change the GUID if it creates a new toplevel GUID.  For a similar
4921	 * reason, we must remove the spare now, in the same txg as the detach;
4922	 * otherwise someone could attach a new sibling, change the GUID, and
4923	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4924	 */
4925	if (unspare) {
4926		ASSERT(cvd->vdev_isspare);
4927		spa_spare_remove(cvd);
4928		unspare_guid = cvd->vdev_guid;
4929		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4930		cvd->vdev_unspare = B_TRUE;
4931	}
4932
4933	/*
4934	 * If the parent mirror/replacing vdev only has one child,
4935	 * the parent is no longer needed.  Remove it from the tree.
4936	 */
4937	if (pvd->vdev_children == 1) {
4938		if (pvd->vdev_ops == &vdev_spare_ops)
4939			cvd->vdev_unspare = B_FALSE;
4940		vdev_remove_parent(cvd);
4941	}
4942
4943
4944	/*
4945	 * We don't set tvd until now because the parent we just removed
4946	 * may have been the previous top-level vdev.
4947	 */
4948	tvd = cvd->vdev_top;
4949	ASSERT(tvd->vdev_parent == rvd);
4950
4951	/*
4952	 * Reevaluate the parent vdev state.
4953	 */
4954	vdev_propagate_state(cvd);
4955
4956	/*
4957	 * If the 'autoexpand' property is set on the pool then automatically
4958	 * try to expand the size of the pool. For example if the device we
4959	 * just detached was smaller than the others, it may be possible to
4960	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4961	 * first so that we can obtain the updated sizes of the leaf vdevs.
4962	 */
4963	if (spa->spa_autoexpand) {
4964		vdev_reopen(tvd);
4965		vdev_expand(tvd, txg);
4966	}
4967
4968	vdev_config_dirty(tvd);
4969
4970	/*
4971	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4972	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4973	 * But first make sure we're not on any *other* txg's DTL list, to
4974	 * prevent vd from being accessed after it's freed.
4975	 */
4976	vdpath = spa_strdup(vd->vdev_path);
4977	for (int t = 0; t < TXG_SIZE; t++)
4978		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4979	vd->vdev_detached = B_TRUE;
4980	vdev_dirty(tvd, VDD_DTL, vd, txg);
4981
4982	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4983
4984	/* hang on to the spa before we release the lock */
4985	spa_open_ref(spa, FTAG);
4986
4987	error = spa_vdev_exit(spa, vd, txg, 0);
4988
4989	spa_history_log_internal(spa, "detach", NULL,
4990	    "vdev=%s", vdpath);
4991	spa_strfree(vdpath);
4992
4993	/*
4994	 * If this was the removal of the original device in a hot spare vdev,
4995	 * then we want to go through and remove the device from the hot spare
4996	 * list of every other pool.
4997	 */
4998	if (unspare) {
4999		spa_t *altspa = NULL;
5000
5001		mutex_enter(&spa_namespace_lock);
5002		while ((altspa = spa_next(altspa)) != NULL) {
5003			if (altspa->spa_state != POOL_STATE_ACTIVE ||
5004			    altspa == spa)
5005				continue;
5006
5007			spa_open_ref(altspa, FTAG);
5008			mutex_exit(&spa_namespace_lock);
5009			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
5010			mutex_enter(&spa_namespace_lock);
5011			spa_close(altspa, FTAG);
5012		}
5013		mutex_exit(&spa_namespace_lock);
5014
5015		/* search the rest of the vdevs for spares to remove */
5016		spa_vdev_resilver_done(spa);
5017	}
5018
5019	/* all done with the spa; OK to release */
5020	mutex_enter(&spa_namespace_lock);
5021	spa_close(spa, FTAG);
5022	mutex_exit(&spa_namespace_lock);
5023
5024	return (error);
5025}
5026
5027/*
5028 * Split a set of devices from their mirrors, and create a new pool from them.
5029 */
5030int
5031spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
5032    nvlist_t *props, boolean_t exp)
5033{
5034	int error = 0;
5035	uint64_t txg, *glist;
5036	spa_t *newspa;
5037	uint_t c, children, lastlog;
5038	nvlist_t **child, *nvl, *tmp;
5039	dmu_tx_t *tx;
5040	char *altroot = NULL;
5041	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
5042	boolean_t activate_slog;
5043
5044	ASSERT(spa_writeable(spa));
5045
5046	txg = spa_vdev_enter(spa);
5047
5048	/* clear the log and flush everything up to now */
5049	activate_slog = spa_passivate_log(spa);
5050	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5051	error = spa_offline_log(spa);
5052	txg = spa_vdev_config_enter(spa);
5053
5054	if (activate_slog)
5055		spa_activate_log(spa);
5056
5057	if (error != 0)
5058		return (spa_vdev_exit(spa, NULL, txg, error));
5059
5060	/* check new spa name before going any further */
5061	if (spa_lookup(newname) != NULL)
5062		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
5063
5064	/*
5065	 * scan through all the children to ensure they're all mirrors
5066	 */
5067	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
5068	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
5069	    &children) != 0)
5070		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5071
5072	/* first, check to ensure we've got the right child count */
5073	rvd = spa->spa_root_vdev;
5074	lastlog = 0;
5075	for (c = 0; c < rvd->vdev_children; c++) {
5076		vdev_t *vd = rvd->vdev_child[c];
5077
5078		/* don't count the holes & logs as children */
5079		if (vd->vdev_islog || vd->vdev_ishole) {
5080			if (lastlog == 0)
5081				lastlog = c;
5082			continue;
5083		}
5084
5085		lastlog = 0;
5086	}
5087	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
5088		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5089
5090	/* next, ensure no spare or cache devices are part of the split */
5091	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
5092	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
5093		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5094
5095	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5096	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5097
5098	/* then, loop over each vdev and validate it */
5099	for (c = 0; c < children; c++) {
5100		uint64_t is_hole = 0;
5101
5102		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5103		    &is_hole);
5104
5105		if (is_hole != 0) {
5106			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5107			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5108				continue;
5109			} else {
5110				error = SET_ERROR(EINVAL);
5111				break;
5112			}
5113		}
5114
5115		/* which disk is going to be split? */
5116		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5117		    &glist[c]) != 0) {
5118			error = SET_ERROR(EINVAL);
5119			break;
5120		}
5121
5122		/* look it up in the spa */
5123		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5124		if (vml[c] == NULL) {
5125			error = SET_ERROR(ENODEV);
5126			break;
5127		}
5128
5129		/* make sure there's nothing stopping the split */
5130		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5131		    vml[c]->vdev_islog ||
5132		    vml[c]->vdev_ishole ||
5133		    vml[c]->vdev_isspare ||
5134		    vml[c]->vdev_isl2cache ||
5135		    !vdev_writeable(vml[c]) ||
5136		    vml[c]->vdev_children != 0 ||
5137		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5138		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5139			error = SET_ERROR(EINVAL);
5140			break;
5141		}
5142
5143		if (vdev_dtl_required(vml[c])) {
5144			error = SET_ERROR(EBUSY);
5145			break;
5146		}
5147
5148		/* we need certain info from the top level */
5149		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5150		    vml[c]->vdev_top->vdev_ms_array) == 0);
5151		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5152		    vml[c]->vdev_top->vdev_ms_shift) == 0);
5153		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5154		    vml[c]->vdev_top->vdev_asize) == 0);
5155		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5156		    vml[c]->vdev_top->vdev_ashift) == 0);
5157
5158		/* transfer per-vdev ZAPs */
5159		ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0);
5160		VERIFY0(nvlist_add_uint64(child[c],
5161		    ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap));
5162
5163		ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0);
5164		VERIFY0(nvlist_add_uint64(child[c],
5165		    ZPOOL_CONFIG_VDEV_TOP_ZAP,
5166		    vml[c]->vdev_parent->vdev_top_zap));
5167	}
5168
5169	if (error != 0) {
5170		kmem_free(vml, children * sizeof (vdev_t *));
5171		kmem_free(glist, children * sizeof (uint64_t));
5172		return (spa_vdev_exit(spa, NULL, txg, error));
5173	}
5174
5175	/* stop writers from using the disks */
5176	for (c = 0; c < children; c++) {
5177		if (vml[c] != NULL)
5178			vml[c]->vdev_offline = B_TRUE;
5179	}
5180	vdev_reopen(spa->spa_root_vdev);
5181
5182	/*
5183	 * Temporarily record the splitting vdevs in the spa config.  This
5184	 * will disappear once the config is regenerated.
5185	 */
5186	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5187	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5188	    glist, children) == 0);
5189	kmem_free(glist, children * sizeof (uint64_t));
5190
5191	mutex_enter(&spa->spa_props_lock);
5192	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5193	    nvl) == 0);
5194	mutex_exit(&spa->spa_props_lock);
5195	spa->spa_config_splitting = nvl;
5196	vdev_config_dirty(spa->spa_root_vdev);
5197
5198	/* configure and create the new pool */
5199	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5200	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5201	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5202	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5203	    spa_version(spa)) == 0);
5204	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5205	    spa->spa_config_txg) == 0);
5206	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5207	    spa_generate_guid(NULL)) == 0);
5208	VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
5209	(void) nvlist_lookup_string(props,
5210	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5211
5212	/* add the new pool to the namespace */
5213	newspa = spa_add(newname, config, altroot);
5214	newspa->spa_avz_action = AVZ_ACTION_REBUILD;
5215	newspa->spa_config_txg = spa->spa_config_txg;
5216	spa_set_log_state(newspa, SPA_LOG_CLEAR);
5217
5218	/* release the spa config lock, retaining the namespace lock */
5219	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5220
5221	if (zio_injection_enabled)
5222		zio_handle_panic_injection(spa, FTAG, 1);
5223
5224	spa_activate(newspa, spa_mode_global);
5225	spa_async_suspend(newspa);
5226
5227	/* create the new pool from the disks of the original pool */
5228	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5229	if (error)
5230		goto out;
5231
5232	/* if that worked, generate a real config for the new pool */
5233	if (newspa->spa_root_vdev != NULL) {
5234		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5235		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
5236		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5237		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5238		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5239		    B_TRUE));
5240	}
5241
5242	/* set the props */
5243	if (props != NULL) {
5244		spa_configfile_set(newspa, props, B_FALSE);
5245		error = spa_prop_set(newspa, props);
5246		if (error)
5247			goto out;
5248	}
5249
5250	/* flush everything */
5251	txg = spa_vdev_config_enter(newspa);
5252	vdev_config_dirty(newspa->spa_root_vdev);
5253	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5254
5255	if (zio_injection_enabled)
5256		zio_handle_panic_injection(spa, FTAG, 2);
5257
5258	spa_async_resume(newspa);
5259
5260	/* finally, update the original pool's config */
5261	txg = spa_vdev_config_enter(spa);
5262	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5263	error = dmu_tx_assign(tx, TXG_WAIT);
5264	if (error != 0)
5265		dmu_tx_abort(tx);
5266	for (c = 0; c < children; c++) {
5267		if (vml[c] != NULL) {
5268			vdev_split(vml[c]);
5269			if (error == 0)
5270				spa_history_log_internal(spa, "detach", tx,
5271				    "vdev=%s", vml[c]->vdev_path);
5272
5273			vdev_free(vml[c]);
5274		}
5275	}
5276	spa->spa_avz_action = AVZ_ACTION_REBUILD;
5277	vdev_config_dirty(spa->spa_root_vdev);
5278	spa->spa_config_splitting = NULL;
5279	nvlist_free(nvl);
5280	if (error == 0)
5281		dmu_tx_commit(tx);
5282	(void) spa_vdev_exit(spa, NULL, txg, 0);
5283
5284	if (zio_injection_enabled)
5285		zio_handle_panic_injection(spa, FTAG, 3);
5286
5287	/* split is complete; log a history record */
5288	spa_history_log_internal(newspa, "split", NULL,
5289	    "from pool %s", spa_name(spa));
5290
5291	kmem_free(vml, children * sizeof (vdev_t *));
5292
5293	/* if we're not going to mount the filesystems in userland, export */
5294	if (exp)
5295		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5296		    B_FALSE, B_FALSE);
5297
5298	return (error);
5299
5300out:
5301	spa_unload(newspa);
5302	spa_deactivate(newspa);
5303	spa_remove(newspa);
5304
5305	txg = spa_vdev_config_enter(spa);
5306
5307	/* re-online all offlined disks */
5308	for (c = 0; c < children; c++) {
5309		if (vml[c] != NULL)
5310			vml[c]->vdev_offline = B_FALSE;
5311	}
5312	vdev_reopen(spa->spa_root_vdev);
5313
5314	nvlist_free(spa->spa_config_splitting);
5315	spa->spa_config_splitting = NULL;
5316	(void) spa_vdev_exit(spa, NULL, txg, error);
5317
5318	kmem_free(vml, children * sizeof (vdev_t *));
5319	return (error);
5320}
5321
5322static nvlist_t *
5323spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5324{
5325	for (int i = 0; i < count; i++) {
5326		uint64_t guid;
5327
5328		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5329		    &guid) == 0);
5330
5331		if (guid == target_guid)
5332			return (nvpp[i]);
5333	}
5334
5335	return (NULL);
5336}
5337
5338static void
5339spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5340    nvlist_t *dev_to_remove)
5341{
5342	nvlist_t **newdev = NULL;
5343
5344	if (count > 1)
5345		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5346
5347	for (int i = 0, j = 0; i < count; i++) {
5348		if (dev[i] == dev_to_remove)
5349			continue;
5350		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5351	}
5352
5353	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5354	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5355
5356	for (int i = 0; i < count - 1; i++)
5357		nvlist_free(newdev[i]);
5358
5359	if (count > 1)
5360		kmem_free(newdev, (count - 1) * sizeof (void *));
5361}
5362
5363/*
5364 * Evacuate the device.
5365 */
5366static int
5367spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5368{
5369	uint64_t txg;
5370	int error = 0;
5371
5372	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5373	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5374	ASSERT(vd == vd->vdev_top);
5375
5376	/*
5377	 * Evacuate the device.  We don't hold the config lock as writer
5378	 * since we need to do I/O but we do keep the
5379	 * spa_namespace_lock held.  Once this completes the device
5380	 * should no longer have any blocks allocated on it.
5381	 */
5382	if (vd->vdev_islog) {
5383		if (vd->vdev_stat.vs_alloc != 0)
5384			error = spa_offline_log(spa);
5385	} else {
5386		error = SET_ERROR(ENOTSUP);
5387	}
5388
5389	if (error)
5390		return (error);
5391
5392	/*
5393	 * The evacuation succeeded.  Remove any remaining MOS metadata
5394	 * associated with this vdev, and wait for these changes to sync.
5395	 */
5396	ASSERT0(vd->vdev_stat.vs_alloc);
5397	txg = spa_vdev_config_enter(spa);
5398	vd->vdev_removing = B_TRUE;
5399	vdev_dirty_leaves(vd, VDD_DTL, txg);
5400	vdev_config_dirty(vd);
5401	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5402
5403	return (0);
5404}
5405
5406/*
5407 * Complete the removal by cleaning up the namespace.
5408 */
5409static void
5410spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5411{
5412	vdev_t *rvd = spa->spa_root_vdev;
5413	uint64_t id = vd->vdev_id;
5414	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5415
5416	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5417	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5418	ASSERT(vd == vd->vdev_top);
5419
5420	/*
5421	 * Only remove any devices which are empty.
5422	 */
5423	if (vd->vdev_stat.vs_alloc != 0)
5424		return;
5425
5426	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5427
5428	if (list_link_active(&vd->vdev_state_dirty_node))
5429		vdev_state_clean(vd);
5430	if (list_link_active(&vd->vdev_config_dirty_node))
5431		vdev_config_clean(vd);
5432
5433	vdev_free(vd);
5434
5435	if (last_vdev) {
5436		vdev_compact_children(rvd);
5437	} else {
5438		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5439		vdev_add_child(rvd, vd);
5440	}
5441	vdev_config_dirty(rvd);
5442
5443	/*
5444	 * Reassess the health of our root vdev.
5445	 */
5446	vdev_reopen(rvd);
5447}
5448
5449/*
5450 * Remove a device from the pool -
5451 *
5452 * Removing a device from the vdev namespace requires several steps
5453 * and can take a significant amount of time.  As a result we use
5454 * the spa_vdev_config_[enter/exit] functions which allow us to
5455 * grab and release the spa_config_lock while still holding the namespace
5456 * lock.  During each step the configuration is synced out.
5457 *
5458 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5459 * devices.
5460 */
5461int
5462spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5463{
5464	vdev_t *vd;
5465	sysevent_t *ev = NULL;
5466	metaslab_group_t *mg;
5467	nvlist_t **spares, **l2cache, *nv;
5468	uint64_t txg = 0;
5469	uint_t nspares, nl2cache;
5470	int error = 0;
5471	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5472
5473	ASSERT(spa_writeable(spa));
5474
5475	if (!locked)
5476		txg = spa_vdev_enter(spa);
5477
5478	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5479
5480	if (spa->spa_spares.sav_vdevs != NULL &&
5481	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5482	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5483	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5484		/*
5485		 * Only remove the hot spare if it's not currently in use
5486		 * in this pool.
5487		 */
5488		if (vd == NULL || unspare) {
5489			if (vd == NULL)
5490				vd = spa_lookup_by_guid(spa, guid, B_TRUE);
5491			ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX);
5492			spa_vdev_remove_aux(spa->spa_spares.sav_config,
5493			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5494			spa_load_spares(spa);
5495			spa->spa_spares.sav_sync = B_TRUE;
5496		} else {
5497			error = SET_ERROR(EBUSY);
5498		}
5499	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
5500	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5501	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5502	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5503		/*
5504		 * Cache devices can always be removed.
5505		 */
5506		vd = spa_lookup_by_guid(spa, guid, B_TRUE);
5507		ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX);
5508		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5509		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5510		spa_load_l2cache(spa);
5511		spa->spa_l2cache.sav_sync = B_TRUE;
5512	} else if (vd != NULL && vd->vdev_islog) {
5513		ASSERT(!locked);
5514		ASSERT(vd == vd->vdev_top);
5515
5516		mg = vd->vdev_mg;
5517
5518		/*
5519		 * Stop allocating from this vdev.
5520		 */
5521		metaslab_group_passivate(mg);
5522
5523		/*
5524		 * Wait for the youngest allocations and frees to sync,
5525		 * and then wait for the deferral of those frees to finish.
5526		 */
5527		spa_vdev_config_exit(spa, NULL,
5528		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5529
5530		/*
5531		 * Attempt to evacuate the vdev.
5532		 */
5533		error = spa_vdev_remove_evacuate(spa, vd);
5534
5535		txg = spa_vdev_config_enter(spa);
5536
5537		/*
5538		 * If we couldn't evacuate the vdev, unwind.
5539		 */
5540		if (error) {
5541			metaslab_group_activate(mg);
5542			return (spa_vdev_exit(spa, NULL, txg, error));
5543		}
5544
5545		/*
5546		 * Clean up the vdev namespace.
5547		 */
5548		ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_DEV);
5549		spa_vdev_remove_from_namespace(spa, vd);
5550
5551	} else if (vd != NULL) {
5552		/*
5553		 * Normal vdevs cannot be removed (yet).
5554		 */
5555		error = SET_ERROR(ENOTSUP);
5556	} else {
5557		/*
5558		 * There is no vdev of any kind with the specified guid.
5559		 */
5560		error = SET_ERROR(ENOENT);
5561	}
5562
5563	if (!locked)
5564		error = spa_vdev_exit(spa, NULL, txg, error);
5565
5566	if (ev)
5567		spa_event_post(ev);
5568
5569	return (error);
5570}
5571
5572/*
5573 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5574 * currently spared, so we can detach it.
5575 */
5576static vdev_t *
5577spa_vdev_resilver_done_hunt(vdev_t *vd)
5578{
5579	vdev_t *newvd, *oldvd;
5580
5581	for (int c = 0; c < vd->vdev_children; c++) {
5582		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5583		if (oldvd != NULL)
5584			return (oldvd);
5585	}
5586
5587	/*
5588	 * Check for a completed replacement.  We always consider the first
5589	 * vdev in the list to be the oldest vdev, and the last one to be
5590	 * the newest (see spa_vdev_attach() for how that works).  In
5591	 * the case where the newest vdev is faulted, we will not automatically
5592	 * remove it after a resilver completes.  This is OK as it will require
5593	 * user intervention to determine which disk the admin wishes to keep.
5594	 */
5595	if (vd->vdev_ops == &vdev_replacing_ops) {
5596		ASSERT(vd->vdev_children > 1);
5597
5598		newvd = vd->vdev_child[vd->vdev_children - 1];
5599		oldvd = vd->vdev_child[0];
5600
5601		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5602		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5603		    !vdev_dtl_required(oldvd))
5604			return (oldvd);
5605	}
5606
5607	/*
5608	 * Check for a completed resilver with the 'unspare' flag set.
5609	 */
5610	if (vd->vdev_ops == &vdev_spare_ops) {
5611		vdev_t *first = vd->vdev_child[0];
5612		vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5613
5614		if (last->vdev_unspare) {
5615			oldvd = first;
5616			newvd = last;
5617		} else if (first->vdev_unspare) {
5618			oldvd = last;
5619			newvd = first;
5620		} else {
5621			oldvd = NULL;
5622		}
5623
5624		if (oldvd != NULL &&
5625		    vdev_dtl_empty(newvd, DTL_MISSING) &&
5626		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5627		    !vdev_dtl_required(oldvd))
5628			return (oldvd);
5629
5630		/*
5631		 * If there are more than two spares attached to a disk,
5632		 * and those spares are not required, then we want to
5633		 * attempt to free them up now so that they can be used
5634		 * by other pools.  Once we're back down to a single
5635		 * disk+spare, we stop removing them.
5636		 */
5637		if (vd->vdev_children > 2) {
5638			newvd = vd->vdev_child[1];
5639
5640			if (newvd->vdev_isspare && last->vdev_isspare &&
5641			    vdev_dtl_empty(last, DTL_MISSING) &&
5642			    vdev_dtl_empty(last, DTL_OUTAGE) &&
5643			    !vdev_dtl_required(newvd))
5644				return (newvd);
5645		}
5646	}
5647
5648	return (NULL);
5649}
5650
5651static void
5652spa_vdev_resilver_done(spa_t *spa)
5653{
5654	vdev_t *vd, *pvd, *ppvd;
5655	uint64_t guid, sguid, pguid, ppguid;
5656
5657	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5658
5659	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5660		pvd = vd->vdev_parent;
5661		ppvd = pvd->vdev_parent;
5662		guid = vd->vdev_guid;
5663		pguid = pvd->vdev_guid;
5664		ppguid = ppvd->vdev_guid;
5665		sguid = 0;
5666		/*
5667		 * If we have just finished replacing a hot spared device, then
5668		 * we need to detach the parent's first child (the original hot
5669		 * spare) as well.
5670		 */
5671		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5672		    ppvd->vdev_children == 2) {
5673			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5674			sguid = ppvd->vdev_child[1]->vdev_guid;
5675		}
5676		ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5677
5678		spa_config_exit(spa, SCL_ALL, FTAG);
5679		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5680			return;
5681		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5682			return;
5683		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5684	}
5685
5686	spa_config_exit(spa, SCL_ALL, FTAG);
5687}
5688
5689/*
5690 * Update the stored path or FRU for this vdev.
5691 */
5692int
5693spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5694    boolean_t ispath)
5695{
5696	vdev_t *vd;
5697	boolean_t sync = B_FALSE;
5698
5699	ASSERT(spa_writeable(spa));
5700
5701	spa_vdev_state_enter(spa, SCL_ALL);
5702
5703	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5704		return (spa_vdev_state_exit(spa, NULL, ENOENT));
5705
5706	if (!vd->vdev_ops->vdev_op_leaf)
5707		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5708
5709	if (ispath) {
5710		if (strcmp(value, vd->vdev_path) != 0) {
5711			spa_strfree(vd->vdev_path);
5712			vd->vdev_path = spa_strdup(value);
5713			sync = B_TRUE;
5714		}
5715	} else {
5716		if (vd->vdev_fru == NULL) {
5717			vd->vdev_fru = spa_strdup(value);
5718			sync = B_TRUE;
5719		} else if (strcmp(value, vd->vdev_fru) != 0) {
5720			spa_strfree(vd->vdev_fru);
5721			vd->vdev_fru = spa_strdup(value);
5722			sync = B_TRUE;
5723		}
5724	}
5725
5726	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5727}
5728
5729int
5730spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5731{
5732	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5733}
5734
5735int
5736spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5737{
5738	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5739}
5740
5741/*
5742 * ==========================================================================
5743 * SPA Scanning
5744 * ==========================================================================
5745 */
5746
5747int
5748spa_scan_stop(spa_t *spa)
5749{
5750	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5751	if (dsl_scan_resilvering(spa->spa_dsl_pool))
5752		return (SET_ERROR(EBUSY));
5753	return (dsl_scan_cancel(spa->spa_dsl_pool));
5754}
5755
5756int
5757spa_scan(spa_t *spa, pool_scan_func_t func)
5758{
5759	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5760
5761	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5762		return (SET_ERROR(ENOTSUP));
5763
5764	/*
5765	 * If a resilver was requested, but there is no DTL on a
5766	 * writeable leaf device, we have nothing to do.
5767	 */
5768	if (func == POOL_SCAN_RESILVER &&
5769	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5770		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5771		return (0);
5772	}
5773
5774	return (dsl_scan(spa->spa_dsl_pool, func));
5775}
5776
5777/*
5778 * ==========================================================================
5779 * SPA async task processing
5780 * ==========================================================================
5781 */
5782
5783static void
5784spa_async_remove(spa_t *spa, vdev_t *vd)
5785{
5786	if (vd->vdev_remove_wanted) {
5787		vd->vdev_remove_wanted = B_FALSE;
5788		vd->vdev_delayed_close = B_FALSE;
5789		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5790
5791		/*
5792		 * We want to clear the stats, but we don't want to do a full
5793		 * vdev_clear() as that will cause us to throw away
5794		 * degraded/faulted state as well as attempt to reopen the
5795		 * device, all of which is a waste.
5796		 */
5797		vd->vdev_stat.vs_read_errors = 0;
5798		vd->vdev_stat.vs_write_errors = 0;
5799		vd->vdev_stat.vs_checksum_errors = 0;
5800
5801		vdev_state_dirty(vd->vdev_top);
5802	}
5803
5804	for (int c = 0; c < vd->vdev_children; c++)
5805		spa_async_remove(spa, vd->vdev_child[c]);
5806}
5807
5808static void
5809spa_async_probe(spa_t *spa, vdev_t *vd)
5810{
5811	if (vd->vdev_probe_wanted) {
5812		vd->vdev_probe_wanted = B_FALSE;
5813		vdev_reopen(vd);	/* vdev_open() does the actual probe */
5814	}
5815
5816	for (int c = 0; c < vd->vdev_children; c++)
5817		spa_async_probe(spa, vd->vdev_child[c]);
5818}
5819
5820static void
5821spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5822{
5823	sysevent_id_t eid;
5824	nvlist_t *attr;
5825	char *physpath;
5826
5827	if (!spa->spa_autoexpand)
5828		return;
5829
5830	for (int c = 0; c < vd->vdev_children; c++) {
5831		vdev_t *cvd = vd->vdev_child[c];
5832		spa_async_autoexpand(spa, cvd);
5833	}
5834
5835	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5836		return;
5837
5838	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5839	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5840
5841	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5842	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5843
5844	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5845	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5846
5847	nvlist_free(attr);
5848	kmem_free(physpath, MAXPATHLEN);
5849}
5850
5851static void
5852spa_async_thread(spa_t *spa)
5853{
5854	int tasks;
5855
5856	ASSERT(spa->spa_sync_on);
5857
5858	mutex_enter(&spa->spa_async_lock);
5859	tasks = spa->spa_async_tasks;
5860	spa->spa_async_tasks = 0;
5861	mutex_exit(&spa->spa_async_lock);
5862
5863	/*
5864	 * See if the config needs to be updated.
5865	 */
5866	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5867		uint64_t old_space, new_space;
5868
5869		mutex_enter(&spa_namespace_lock);
5870		old_space = metaslab_class_get_space(spa_normal_class(spa));
5871		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5872		new_space = metaslab_class_get_space(spa_normal_class(spa));
5873		mutex_exit(&spa_namespace_lock);
5874
5875		/*
5876		 * If the pool grew as a result of the config update,
5877		 * then log an internal history event.
5878		 */
5879		if (new_space != old_space) {
5880			spa_history_log_internal(spa, "vdev online", NULL,
5881			    "pool '%s' size: %llu(+%llu)",
5882			    spa_name(spa), new_space, new_space - old_space);
5883		}
5884	}
5885
5886	/*
5887	 * See if any devices need to be marked REMOVED.
5888	 */
5889	if (tasks & SPA_ASYNC_REMOVE) {
5890		spa_vdev_state_enter(spa, SCL_NONE);
5891		spa_async_remove(spa, spa->spa_root_vdev);
5892		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5893			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5894		for (int i = 0; i < spa->spa_spares.sav_count; i++)
5895			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5896		(void) spa_vdev_state_exit(spa, NULL, 0);
5897	}
5898
5899	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5900		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5901		spa_async_autoexpand(spa, spa->spa_root_vdev);
5902		spa_config_exit(spa, SCL_CONFIG, FTAG);
5903	}
5904
5905	/*
5906	 * See if any devices need to be probed.
5907	 */
5908	if (tasks & SPA_ASYNC_PROBE) {
5909		spa_vdev_state_enter(spa, SCL_NONE);
5910		spa_async_probe(spa, spa->spa_root_vdev);
5911		(void) spa_vdev_state_exit(spa, NULL, 0);
5912	}
5913
5914	/*
5915	 * If any devices are done replacing, detach them.
5916	 */
5917	if (tasks & SPA_ASYNC_RESILVER_DONE)
5918		spa_vdev_resilver_done(spa);
5919
5920	/*
5921	 * Kick off a resilver.
5922	 */
5923	if (tasks & SPA_ASYNC_RESILVER)
5924		dsl_resilver_restart(spa->spa_dsl_pool, 0);
5925
5926	/*
5927	 * Let the world know that we're done.
5928	 */
5929	mutex_enter(&spa->spa_async_lock);
5930	spa->spa_async_thread = NULL;
5931	cv_broadcast(&spa->spa_async_cv);
5932	mutex_exit(&spa->spa_async_lock);
5933	thread_exit();
5934}
5935
5936void
5937spa_async_suspend(spa_t *spa)
5938{
5939	mutex_enter(&spa->spa_async_lock);
5940	spa->spa_async_suspended++;
5941	while (spa->spa_async_thread != NULL)
5942		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5943	mutex_exit(&spa->spa_async_lock);
5944}
5945
5946void
5947spa_async_resume(spa_t *spa)
5948{
5949	mutex_enter(&spa->spa_async_lock);
5950	ASSERT(spa->spa_async_suspended != 0);
5951	spa->spa_async_suspended--;
5952	mutex_exit(&spa->spa_async_lock);
5953}
5954
5955static boolean_t
5956spa_async_tasks_pending(spa_t *spa)
5957{
5958	uint_t non_config_tasks;
5959	uint_t config_task;
5960	boolean_t config_task_suspended;
5961
5962	non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5963	config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5964	if (spa->spa_ccw_fail_time == 0) {
5965		config_task_suspended = B_FALSE;
5966	} else {
5967		config_task_suspended =
5968		    (gethrtime() - spa->spa_ccw_fail_time) <
5969		    (zfs_ccw_retry_interval * NANOSEC);
5970	}
5971
5972	return (non_config_tasks || (config_task && !config_task_suspended));
5973}
5974
5975static void
5976spa_async_dispatch(spa_t *spa)
5977{
5978	mutex_enter(&spa->spa_async_lock);
5979	if (spa_async_tasks_pending(spa) &&
5980	    !spa->spa_async_suspended &&
5981	    spa->spa_async_thread == NULL &&
5982	    rootdir != NULL)
5983		spa->spa_async_thread = thread_create(NULL, 0,
5984		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5985	mutex_exit(&spa->spa_async_lock);
5986}
5987
5988void
5989spa_async_request(spa_t *spa, int task)
5990{
5991	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5992	mutex_enter(&spa->spa_async_lock);
5993	spa->spa_async_tasks |= task;
5994	mutex_exit(&spa->spa_async_lock);
5995}
5996
5997/*
5998 * ==========================================================================
5999 * SPA syncing routines
6000 * ==========================================================================
6001 */
6002
6003static int
6004bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6005{
6006	bpobj_t *bpo = arg;
6007	bpobj_enqueue(bpo, bp, tx);
6008	return (0);
6009}
6010
6011static int
6012spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6013{
6014	zio_t *zio = arg;
6015
6016	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
6017	    zio->io_flags));
6018	return (0);
6019}
6020
6021/*
6022 * Note: this simple function is not inlined to make it easier to dtrace the
6023 * amount of time spent syncing frees.
6024 */
6025static void
6026spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
6027{
6028	zio_t *zio = zio_root(spa, NULL, NULL, 0);
6029	bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
6030	VERIFY(zio_wait(zio) == 0);
6031}
6032
6033/*
6034 * Note: this simple function is not inlined to make it easier to dtrace the
6035 * amount of time spent syncing deferred frees.
6036 */
6037static void
6038spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
6039{
6040	zio_t *zio = zio_root(spa, NULL, NULL, 0);
6041	VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
6042	    spa_free_sync_cb, zio, tx), ==, 0);
6043	VERIFY0(zio_wait(zio));
6044}
6045
6046
6047static void
6048spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
6049{
6050	char *packed = NULL;
6051	size_t bufsize;
6052	size_t nvsize = 0;
6053	dmu_buf_t *db;
6054
6055	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
6056
6057	/*
6058	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6059	 * information.  This avoids the dmu_buf_will_dirty() path and
6060	 * saves us a pre-read to get data we don't actually care about.
6061	 */
6062	bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
6063	packed = kmem_alloc(bufsize, KM_SLEEP);
6064
6065	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
6066	    KM_SLEEP) == 0);
6067	bzero(packed + nvsize, bufsize - nvsize);
6068
6069	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
6070
6071	kmem_free(packed, bufsize);
6072
6073	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
6074	dmu_buf_will_dirty(db, tx);
6075	*(uint64_t *)db->db_data = nvsize;
6076	dmu_buf_rele(db, FTAG);
6077}
6078
6079static void
6080spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
6081    const char *config, const char *entry)
6082{
6083	nvlist_t *nvroot;
6084	nvlist_t **list;
6085	int i;
6086
6087	if (!sav->sav_sync)
6088		return;
6089
6090	/*
6091	 * Update the MOS nvlist describing the list of available devices.
6092	 * spa_validate_aux() will have already made sure this nvlist is
6093	 * valid and the vdevs are labeled appropriately.
6094	 */
6095	if (sav->sav_object == 0) {
6096		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
6097		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
6098		    sizeof (uint64_t), tx);
6099		VERIFY(zap_update(spa->spa_meta_objset,
6100		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
6101		    &sav->sav_object, tx) == 0);
6102	}
6103
6104	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6105	if (sav->sav_count == 0) {
6106		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
6107	} else {
6108		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
6109		for (i = 0; i < sav->sav_count; i++)
6110			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
6111			    B_FALSE, VDEV_CONFIG_L2CACHE);
6112		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
6113		    sav->sav_count) == 0);
6114		for (i = 0; i < sav->sav_count; i++)
6115			nvlist_free(list[i]);
6116		kmem_free(list, sav->sav_count * sizeof (void *));
6117	}
6118
6119	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
6120	nvlist_free(nvroot);
6121
6122	sav->sav_sync = B_FALSE;
6123}
6124
6125/*
6126 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
6127 * The all-vdev ZAP must be empty.
6128 */
6129static void
6130spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx)
6131{
6132	spa_t *spa = vd->vdev_spa;
6133	if (vd->vdev_top_zap != 0) {
6134		VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
6135		    vd->vdev_top_zap, tx));
6136	}
6137	if (vd->vdev_leaf_zap != 0) {
6138		VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
6139		    vd->vdev_leaf_zap, tx));
6140	}
6141	for (uint64_t i = 0; i < vd->vdev_children; i++) {
6142		spa_avz_build(vd->vdev_child[i], avz, tx);
6143	}
6144}
6145
6146static void
6147spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
6148{
6149	nvlist_t *config;
6150
6151	/*
6152	 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
6153	 * its config may not be dirty but we still need to build per-vdev ZAPs.
6154	 * Similarly, if the pool is being assembled (e.g. after a split), we
6155	 * need to rebuild the AVZ although the config may not be dirty.
6156	 */
6157	if (list_is_empty(&spa->spa_config_dirty_list) &&
6158	    spa->spa_avz_action == AVZ_ACTION_NONE)
6159		return;
6160
6161	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6162
6163	ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE ||
6164	    spa->spa_avz_action == AVZ_ACTION_INITIALIZE ||
6165	    spa->spa_all_vdev_zaps != 0);
6166
6167	if (spa->spa_avz_action == AVZ_ACTION_REBUILD) {
6168		/* Make and build the new AVZ */
6169		uint64_t new_avz = zap_create(spa->spa_meta_objset,
6170		    DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx);
6171		spa_avz_build(spa->spa_root_vdev, new_avz, tx);
6172
6173		/* Diff old AVZ with new one */
6174		zap_cursor_t zc;
6175		zap_attribute_t za;
6176
6177		for (zap_cursor_init(&zc, spa->spa_meta_objset,
6178		    spa->spa_all_vdev_zaps);
6179		    zap_cursor_retrieve(&zc, &za) == 0;
6180		    zap_cursor_advance(&zc)) {
6181			uint64_t vdzap = za.za_first_integer;
6182			if (zap_lookup_int(spa->spa_meta_objset, new_avz,
6183			    vdzap) == ENOENT) {
6184				/*
6185				 * ZAP is listed in old AVZ but not in new one;
6186				 * destroy it
6187				 */
6188				VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap,
6189				    tx));
6190			}
6191		}
6192
6193		zap_cursor_fini(&zc);
6194
6195		/* Destroy the old AVZ */
6196		VERIFY0(zap_destroy(spa->spa_meta_objset,
6197		    spa->spa_all_vdev_zaps, tx));
6198
6199		/* Replace the old AVZ in the dir obj with the new one */
6200		VERIFY0(zap_update(spa->spa_meta_objset,
6201		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP,
6202		    sizeof (new_avz), 1, &new_avz, tx));
6203
6204		spa->spa_all_vdev_zaps = new_avz;
6205	} else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) {
6206		zap_cursor_t zc;
6207		zap_attribute_t za;
6208
6209		/* Walk through the AVZ and destroy all listed ZAPs */
6210		for (zap_cursor_init(&zc, spa->spa_meta_objset,
6211		    spa->spa_all_vdev_zaps);
6212		    zap_cursor_retrieve(&zc, &za) == 0;
6213		    zap_cursor_advance(&<