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