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