spa.c revision a45f1c3c5ee26f908655e7762d7e1e2aecdfefbc
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, 2014 by Delphix. All rights reserved.
25 * Copyright (c) 2013, 2014, Nexenta Systems, Inc.  All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 */
28
29/*
30 * SPA: Storage Pool Allocator
31 *
32 * This file contains all the routines used when modifying on-disk SPA state.
33 * This includes opening, importing, destroying, exporting a pool, and syncing a
34 * pool.
35 */
36
37#include <sys/zfs_context.h>
38#include <sys/fm/fs/zfs.h>
39#include <sys/spa_impl.h>
40#include <sys/zio.h>
41#include <sys/zio_checksum.h>
42#include <sys/dmu.h>
43#include <sys/dmu_tx.h>
44#include <sys/zap.h>
45#include <sys/zil.h>
46#include <sys/ddt.h>
47#include <sys/vdev_impl.h>
48#include <sys/metaslab.h>
49#include <sys/metaslab_impl.h>
50#include <sys/uberblock_impl.h>
51#include <sys/txg.h>
52#include <sys/avl.h>
53#include <sys/dmu_traverse.h>
54#include <sys/dmu_objset.h>
55#include <sys/unique.h>
56#include <sys/dsl_pool.h>
57#include <sys/dsl_dataset.h>
58#include <sys/dsl_dir.h>
59#include <sys/dsl_prop.h>
60#include <sys/dsl_synctask.h>
61#include <sys/fs/zfs.h>
62#include <sys/arc.h>
63#include <sys/callb.h>
64#include <sys/systeminfo.h>
65#include <sys/spa_boot.h>
66#include <sys/zfs_ioctl.h>
67#include <sys/dsl_scan.h>
68#include <sys/zfeature.h>
69#include <sys/dsl_destroy.h>
70
71#ifdef	_KERNEL
72#include <sys/bootprops.h>
73#include <sys/callb.h>
74#include <sys/cpupart.h>
75#include <sys/pool.h>
76#include <sys/sysdc.h>
77#include <sys/zone.h>
78#endif	/* _KERNEL */
79
80#include "zfs_prop.h"
81#include "zfs_comutil.h"
82
83/*
84 * The interval, in seconds, at which failed configuration cache file writes
85 * should be retried.
86 */
87static int zfs_ccw_retry_interval = 300;
88
89typedef enum zti_modes {
90	ZTI_MODE_FIXED,			/* value is # of threads (min 1) */
91	ZTI_MODE_BATCH,			/* cpu-intensive; value is ignored */
92	ZTI_MODE_NULL,			/* don't create a taskq */
93	ZTI_NMODES
94} zti_modes_t;
95
96#define	ZTI_P(n, q)	{ ZTI_MODE_FIXED, (n), (q) }
97#define	ZTI_BATCH	{ ZTI_MODE_BATCH, 0, 1 }
98#define	ZTI_NULL	{ ZTI_MODE_NULL, 0, 0 }
99
100#define	ZTI_N(n)	ZTI_P(n, 1)
101#define	ZTI_ONE		ZTI_N(1)
102
103typedef struct zio_taskq_info {
104	zti_modes_t zti_mode;
105	uint_t zti_value;
106	uint_t zti_count;
107} zio_taskq_info_t;
108
109static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
110	"issue", "issue_high", "intr", "intr_high"
111};
112
113/*
114 * This table defines the taskq settings for each ZFS I/O type. When
115 * initializing a pool, we use this table to create an appropriately sized
116 * taskq. Some operations are low volume and therefore have a small, static
117 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
118 * macros. Other operations process a large amount of data; the ZTI_BATCH
119 * macro causes us to create a taskq oriented for throughput. Some operations
120 * are so high frequency and short-lived that the taskq itself can become a a
121 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
122 * additional degree of parallelism specified by the number of threads per-
123 * taskq and the number of taskqs; when dispatching an event in this case, the
124 * particular taskq is chosen at random.
125 *
126 * The different taskq priorities are to handle the different contexts (issue
127 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
128 * need to be handled with minimum delay.
129 */
130const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
131	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
132	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* NULL */
133	{ ZTI_N(8),	ZTI_NULL,	ZTI_P(12, 8),	ZTI_NULL }, /* READ */
134	{ ZTI_BATCH,	ZTI_N(5),	ZTI_N(8),	ZTI_N(5) }, /* WRITE */
135	{ ZTI_P(12, 8),	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* FREE */
136	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* CLAIM */
137	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* IOCTL */
138};
139
140static void spa_sync_version(void *arg, dmu_tx_t *tx);
141static void spa_sync_props(void *arg, dmu_tx_t *tx);
142static boolean_t spa_has_active_shared_spare(spa_t *spa);
143static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
144    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
145    char **ereport);
146static void spa_vdev_resilver_done(spa_t *spa);
147
148uint_t		zio_taskq_batch_pct = 75;	/* 1 thread per cpu in pset */
149id_t		zio_taskq_psrset_bind = PS_NONE;
150boolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
151uint_t		zio_taskq_basedc = 80;		/* base duty cycle */
152
153boolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
154extern int	zfs_sync_pass_deferred_free;
155
156/*
157 * This (illegal) pool name is used when temporarily importing a spa_t in order
158 * to get the vdev stats associated with the imported devices.
159 */
160#define	TRYIMPORT_NAME	"$import"
161
162/*
163 * ==========================================================================
164 * SPA properties routines
165 * ==========================================================================
166 */
167
168/*
169 * Add a (source=src, propname=propval) list to an nvlist.
170 */
171static void
172spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
173    uint64_t intval, zprop_source_t src)
174{
175	const char *propname = zpool_prop_to_name(prop);
176	nvlist_t *propval;
177
178	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
179	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
180
181	if (strval != NULL)
182		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
183	else
184		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
185
186	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
187	nvlist_free(propval);
188}
189
190/*
191 * Get property values from the spa configuration.
192 */
193static void
194spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
195{
196	vdev_t *rvd = spa->spa_root_vdev;
197	dsl_pool_t *pool = spa->spa_dsl_pool;
198	uint64_t size, alloc, cap, version;
199	zprop_source_t src = ZPROP_SRC_NONE;
200	spa_config_dirent_t *dp;
201	metaslab_class_t *mc = spa_normal_class(spa);
202
203	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
204
205	if (rvd != NULL) {
206		alloc = metaslab_class_get_alloc(spa_normal_class(spa));
207		size = metaslab_class_get_space(spa_normal_class(spa));
208		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
209		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
210		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
211		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
212		    size - alloc, src);
213
214		spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
215		    metaslab_class_fragmentation(mc), src);
216		spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
217		    metaslab_class_expandable_space(mc), src);
218		spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
219		    (spa_mode(spa) == FREAD), src);
220
221		cap = (size == 0) ? 0 : (alloc * 100 / size);
222		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
223
224		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
225		    ddt_get_pool_dedup_ratio(spa), src);
226
227		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
228		    rvd->vdev_state, src);
229
230		version = spa_version(spa);
231		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
232			src = ZPROP_SRC_DEFAULT;
233		else
234			src = ZPROP_SRC_LOCAL;
235		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
236	}
237
238	if (pool != NULL) {
239		/*
240		 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
241		 * when opening pools before this version freedir will be NULL.
242		 */
243		if (pool->dp_free_dir != NULL) {
244			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
245			    dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
246			    src);
247		} else {
248			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
249			    NULL, 0, src);
250		}
251
252		if (pool->dp_leak_dir != NULL) {
253			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
254			    dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
255			    src);
256		} else {
257			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
258			    NULL, 0, src);
259		}
260	}
261
262	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
263
264	if (spa->spa_comment != NULL) {
265		spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
266		    0, ZPROP_SRC_LOCAL);
267	}
268
269	if (spa->spa_root != NULL)
270		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
271		    0, ZPROP_SRC_LOCAL);
272
273	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
274		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
275		    MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
276	} else {
277		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
278		    SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
279	}
280
281	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
282		if (dp->scd_path == NULL) {
283			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
284			    "none", 0, ZPROP_SRC_LOCAL);
285		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
286			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
287			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
288		}
289	}
290}
291
292/*
293 * Get zpool property values.
294 */
295int
296spa_prop_get(spa_t *spa, nvlist_t **nvp)
297{
298	objset_t *mos = spa->spa_meta_objset;
299	zap_cursor_t zc;
300	zap_attribute_t za;
301	int err;
302
303	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
304
305	mutex_enter(&spa->spa_props_lock);
306
307	/*
308	 * Get properties from the spa config.
309	 */
310	spa_prop_get_config(spa, nvp);
311
312	/* If no pool property object, no more prop to get. */
313	if (mos == NULL || spa->spa_pool_props_object == 0) {
314		mutex_exit(&spa->spa_props_lock);
315		return (0);
316	}
317
318	/*
319	 * Get properties from the MOS pool property object.
320	 */
321	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
322	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
323	    zap_cursor_advance(&zc)) {
324		uint64_t intval = 0;
325		char *strval = NULL;
326		zprop_source_t src = ZPROP_SRC_DEFAULT;
327		zpool_prop_t prop;
328
329		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
330			continue;
331
332		switch (za.za_integer_length) {
333		case 8:
334			/* integer property */
335			if (za.za_first_integer !=
336			    zpool_prop_default_numeric(prop))
337				src = ZPROP_SRC_LOCAL;
338
339			if (prop == ZPOOL_PROP_BOOTFS) {
340				dsl_pool_t *dp;
341				dsl_dataset_t *ds = NULL;
342
343				dp = spa_get_dsl(spa);
344				dsl_pool_config_enter(dp, FTAG);
345				if (err = dsl_dataset_hold_obj(dp,
346				    za.za_first_integer, FTAG, &ds)) {
347					dsl_pool_config_exit(dp, FTAG);
348					break;
349				}
350
351				strval = kmem_alloc(
352				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
353				    KM_SLEEP);
354				dsl_dataset_name(ds, strval);
355				dsl_dataset_rele(ds, FTAG);
356				dsl_pool_config_exit(dp, FTAG);
357			} else {
358				strval = NULL;
359				intval = za.za_first_integer;
360			}
361
362			spa_prop_add_list(*nvp, prop, strval, intval, src);
363
364			if (strval != NULL)
365				kmem_free(strval,
366				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
367
368			break;
369
370		case 1:
371			/* string property */
372			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
373			err = zap_lookup(mos, spa->spa_pool_props_object,
374			    za.za_name, 1, za.za_num_integers, strval);
375			if (err) {
376				kmem_free(strval, za.za_num_integers);
377				break;
378			}
379			spa_prop_add_list(*nvp, prop, strval, 0, src);
380			kmem_free(strval, za.za_num_integers);
381			break;
382
383		default:
384			break;
385		}
386	}
387	zap_cursor_fini(&zc);
388	mutex_exit(&spa->spa_props_lock);
389out:
390	if (err && err != ENOENT) {
391		nvlist_free(*nvp);
392		*nvp = NULL;
393		return (err);
394	}
395
396	return (0);
397}
398
399/*
400 * Validate the given pool properties nvlist and modify the list
401 * for the property values to be set.
402 */
403static int
404spa_prop_validate(spa_t *spa, nvlist_t *props)
405{
406	nvpair_t *elem;
407	int error = 0, reset_bootfs = 0;
408	uint64_t objnum = 0;
409	boolean_t has_feature = B_FALSE;
410
411	elem = NULL;
412	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
413		uint64_t intval;
414		char *strval, *slash, *check, *fname;
415		const char *propname = nvpair_name(elem);
416		zpool_prop_t prop = zpool_name_to_prop(propname);
417
418		switch (prop) {
419		case ZPROP_INVAL:
420			if (!zpool_prop_feature(propname)) {
421				error = SET_ERROR(EINVAL);
422				break;
423			}
424
425			/*
426			 * Sanitize the input.
427			 */
428			if (nvpair_type(elem) != DATA_TYPE_UINT64) {
429				error = SET_ERROR(EINVAL);
430				break;
431			}
432
433			if (nvpair_value_uint64(elem, &intval) != 0) {
434				error = SET_ERROR(EINVAL);
435				break;
436			}
437
438			if (intval != 0) {
439				error = SET_ERROR(EINVAL);
440				break;
441			}
442
443			fname = strchr(propname, '@') + 1;
444			if (zfeature_lookup_name(fname, NULL) != 0) {
445				error = SET_ERROR(EINVAL);
446				break;
447			}
448
449			has_feature = B_TRUE;
450			break;
451
452		case ZPOOL_PROP_VERSION:
453			error = nvpair_value_uint64(elem, &intval);
454			if (!error &&
455			    (intval < spa_version(spa) ||
456			    intval > SPA_VERSION_BEFORE_FEATURES ||
457			    has_feature))
458				error = SET_ERROR(EINVAL);
459			break;
460
461		case ZPOOL_PROP_DELEGATION:
462		case ZPOOL_PROP_AUTOREPLACE:
463		case ZPOOL_PROP_LISTSNAPS:
464		case ZPOOL_PROP_AUTOEXPAND:
465			error = nvpair_value_uint64(elem, &intval);
466			if (!error && intval > 1)
467				error = SET_ERROR(EINVAL);
468			break;
469
470		case ZPOOL_PROP_BOOTFS:
471			/*
472			 * If the pool version is less than SPA_VERSION_BOOTFS,
473			 * or the pool is still being created (version == 0),
474			 * the bootfs property cannot be set.
475			 */
476			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
477				error = SET_ERROR(ENOTSUP);
478				break;
479			}
480
481			/*
482			 * Make sure the vdev config is bootable
483			 */
484			if (!vdev_is_bootable(spa->spa_root_vdev)) {
485				error = SET_ERROR(ENOTSUP);
486				break;
487			}
488
489			reset_bootfs = 1;
490
491			error = nvpair_value_string(elem, &strval);
492
493			if (!error) {
494				objset_t *os;
495				uint64_t propval;
496
497				if (strval == NULL || strval[0] == '\0') {
498					objnum = zpool_prop_default_numeric(
499					    ZPOOL_PROP_BOOTFS);
500					break;
501				}
502
503				if (error = dmu_objset_hold(strval, FTAG, &os))
504					break;
505
506				/*
507				 * Must be ZPL, and its property settings
508				 * must be supported by GRUB (compression
509				 * is not gzip, and large blocks are not used).
510				 */
511
512				if (dmu_objset_type(os) != DMU_OST_ZFS) {
513					error = SET_ERROR(ENOTSUP);
514				} else if ((error =
515				    dsl_prop_get_int_ds(dmu_objset_ds(os),
516				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
517				    &propval)) == 0 &&
518				    !BOOTFS_COMPRESS_VALID(propval)) {
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_RECORDSIZE),
523				    &propval)) == 0 &&
524				    propval > SPA_OLD_MAXBLOCKSIZE) {
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				check++;
593			}
594			if (strlen(strval) > ZPROP_MAX_COMMENT)
595				error = E2BIG;
596			break;
597
598		case ZPOOL_PROP_DEDUPDITTO:
599			if (spa_version(spa) < SPA_VERSION_DEDUP)
600				error = SET_ERROR(ENOTSUP);
601			else
602				error = nvpair_value_uint64(elem, &intval);
603			if (error == 0 &&
604			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
605				error = SET_ERROR(EINVAL);
606			break;
607		}
608
609		if (error)
610			break;
611	}
612
613	if (!error && reset_bootfs) {
614		error = nvlist_remove(props,
615		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
616
617		if (!error) {
618			error = nvlist_add_uint64(props,
619			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
620		}
621	}
622
623	return (error);
624}
625
626void
627spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
628{
629	char *cachefile;
630	spa_config_dirent_t *dp;
631
632	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
633	    &cachefile) != 0)
634		return;
635
636	dp = kmem_alloc(sizeof (spa_config_dirent_t),
637	    KM_SLEEP);
638
639	if (cachefile[0] == '\0')
640		dp->scd_path = spa_strdup(spa_config_path);
641	else if (strcmp(cachefile, "none") == 0)
642		dp->scd_path = NULL;
643	else
644		dp->scd_path = spa_strdup(cachefile);
645
646	list_insert_head(&spa->spa_config_list, dp);
647	if (need_sync)
648		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
649}
650
651int
652spa_prop_set(spa_t *spa, nvlist_t *nvp)
653{
654	int error;
655	nvpair_t *elem = NULL;
656	boolean_t need_sync = B_FALSE;
657
658	if ((error = spa_prop_validate(spa, nvp)) != 0)
659		return (error);
660
661	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
662		zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
663
664		if (prop == ZPOOL_PROP_CACHEFILE ||
665		    prop == ZPOOL_PROP_ALTROOT ||
666		    prop == ZPOOL_PROP_READONLY)
667			continue;
668
669		if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
670			uint64_t ver;
671
672			if (prop == ZPOOL_PROP_VERSION) {
673				VERIFY(nvpair_value_uint64(elem, &ver) == 0);
674			} else {
675				ASSERT(zpool_prop_feature(nvpair_name(elem)));
676				ver = SPA_VERSION_FEATURES;
677				need_sync = B_TRUE;
678			}
679
680			/* Save time if the version is already set. */
681			if (ver == spa_version(spa))
682				continue;
683
684			/*
685			 * In addition to the pool directory object, we might
686			 * create the pool properties object, the features for
687			 * read object, the features for write object, or the
688			 * feature descriptions object.
689			 */
690			error = dsl_sync_task(spa->spa_name, NULL,
691			    spa_sync_version, &ver,
692			    6, ZFS_SPACE_CHECK_RESERVED);
693			if (error)
694				return (error);
695			continue;
696		}
697
698		need_sync = B_TRUE;
699		break;
700	}
701
702	if (need_sync) {
703		return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
704		    nvp, 6, ZFS_SPACE_CHECK_RESERVED));
705	}
706
707	return (0);
708}
709
710/*
711 * If the bootfs property value is dsobj, clear it.
712 */
713void
714spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
715{
716	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
717		VERIFY(zap_remove(spa->spa_meta_objset,
718		    spa->spa_pool_props_object,
719		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
720		spa->spa_bootfs = 0;
721	}
722}
723
724/*ARGSUSED*/
725static int
726spa_change_guid_check(void *arg, dmu_tx_t *tx)
727{
728	uint64_t *newguid = arg;
729	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
730	vdev_t *rvd = spa->spa_root_vdev;
731	uint64_t vdev_state;
732
733	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
734	vdev_state = rvd->vdev_state;
735	spa_config_exit(spa, SCL_STATE, FTAG);
736
737	if (vdev_state != VDEV_STATE_HEALTHY)
738		return (SET_ERROR(ENXIO));
739
740	ASSERT3U(spa_guid(spa), !=, *newguid);
741
742	return (0);
743}
744
745static void
746spa_change_guid_sync(void *arg, dmu_tx_t *tx)
747{
748	uint64_t *newguid = arg;
749	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
750	uint64_t oldguid;
751	vdev_t *rvd = spa->spa_root_vdev;
752
753	oldguid = spa_guid(spa);
754
755	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
756	rvd->vdev_guid = *newguid;
757	rvd->vdev_guid_sum += (*newguid - oldguid);
758	vdev_config_dirty(rvd);
759	spa_config_exit(spa, SCL_STATE, FTAG);
760
761	spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
762	    oldguid, *newguid);
763}
764
765/*
766 * Change the GUID for the pool.  This is done so that we can later
767 * re-import a pool built from a clone of our own vdevs.  We will modify
768 * the root vdev's guid, our own pool guid, and then mark all of our
769 * vdevs dirty.  Note that we must make sure that all our vdevs are
770 * online when we do this, or else any vdevs that weren't present
771 * would be orphaned from our pool.  We are also going to issue a
772 * sysevent to update any watchers.
773 */
774int
775spa_change_guid(spa_t *spa)
776{
777	int error;
778	uint64_t guid;
779
780	mutex_enter(&spa->spa_vdev_top_lock);
781	mutex_enter(&spa_namespace_lock);
782	guid = spa_generate_guid(NULL);
783
784	error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
785	    spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
786
787	if (error == 0) {
788		spa_config_sync(spa, B_FALSE, B_TRUE);
789		spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
790	}
791
792	mutex_exit(&spa_namespace_lock);
793	mutex_exit(&spa->spa_vdev_top_lock);
794
795	return (error);
796}
797
798/*
799 * ==========================================================================
800 * SPA state manipulation (open/create/destroy/import/export)
801 * ==========================================================================
802 */
803
804static int
805spa_error_entry_compare(const void *a, const void *b)
806{
807	spa_error_entry_t *sa = (spa_error_entry_t *)a;
808	spa_error_entry_t *sb = (spa_error_entry_t *)b;
809	int ret;
810
811	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
812	    sizeof (zbookmark_phys_t));
813
814	if (ret < 0)
815		return (-1);
816	else if (ret > 0)
817		return (1);
818	else
819		return (0);
820}
821
822/*
823 * Utility function which retrieves copies of the current logs and
824 * re-initializes them in the process.
825 */
826void
827spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
828{
829	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
830
831	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
832	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
833
834	avl_create(&spa->spa_errlist_scrub,
835	    spa_error_entry_compare, sizeof (spa_error_entry_t),
836	    offsetof(spa_error_entry_t, se_avl));
837	avl_create(&spa->spa_errlist_last,
838	    spa_error_entry_compare, sizeof (spa_error_entry_t),
839	    offsetof(spa_error_entry_t, se_avl));
840}
841
842static void
843spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
844{
845	const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
846	enum zti_modes mode = ztip->zti_mode;
847	uint_t value = ztip->zti_value;
848	uint_t count = ztip->zti_count;
849	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
850	char name[32];
851	uint_t flags = 0;
852	boolean_t batch = B_FALSE;
853
854	if (mode == ZTI_MODE_NULL) {
855		tqs->stqs_count = 0;
856		tqs->stqs_taskq = NULL;
857		return;
858	}
859
860	ASSERT3U(count, >, 0);
861
862	tqs->stqs_count = count;
863	tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
864
865	switch (mode) {
866	case ZTI_MODE_FIXED:
867		ASSERT3U(value, >=, 1);
868		value = MAX(value, 1);
869		break;
870
871	case ZTI_MODE_BATCH:
872		batch = B_TRUE;
873		flags |= TASKQ_THREADS_CPU_PCT;
874		value = zio_taskq_batch_pct;
875		break;
876
877	default:
878		panic("unrecognized mode for %s_%s taskq (%u:%u) in "
879		    "spa_activate()",
880		    zio_type_name[t], zio_taskq_types[q], mode, value);
881		break;
882	}
883
884	for (uint_t i = 0; i < count; i++) {
885		taskq_t *tq;
886
887		if (count > 1) {
888			(void) snprintf(name, sizeof (name), "%s_%s_%u",
889			    zio_type_name[t], zio_taskq_types[q], i);
890		} else {
891			(void) snprintf(name, sizeof (name), "%s_%s",
892			    zio_type_name[t], zio_taskq_types[q]);
893		}
894
895		if (zio_taskq_sysdc && spa->spa_proc != &p0) {
896			if (batch)
897				flags |= TASKQ_DC_BATCH;
898
899			tq = taskq_create_sysdc(name, value, 50, INT_MAX,
900			    spa->spa_proc, zio_taskq_basedc, flags);
901		} else {
902			pri_t pri = maxclsyspri;
903			/*
904			 * The write issue taskq can be extremely CPU
905			 * intensive.  Run it at slightly lower priority
906			 * than the other taskqs.
907			 */
908			if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
909				pri--;
910
911			tq = taskq_create_proc(name, value, pri, 50,
912			    INT_MAX, spa->spa_proc, flags);
913		}
914
915		tqs->stqs_taskq[i] = tq;
916	}
917}
918
919static void
920spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
921{
922	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
923
924	if (tqs->stqs_taskq == NULL) {
925		ASSERT0(tqs->stqs_count);
926		return;
927	}
928
929	for (uint_t i = 0; i < tqs->stqs_count; i++) {
930		ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
931		taskq_destroy(tqs->stqs_taskq[i]);
932	}
933
934	kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
935	tqs->stqs_taskq = NULL;
936}
937
938/*
939 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
940 * Note that a type may have multiple discrete taskqs to avoid lock contention
941 * on the taskq itself. In that case we choose which taskq at random by using
942 * the low bits of gethrtime().
943 */
944void
945spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
946    task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
947{
948	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
949	taskq_t *tq;
950
951	ASSERT3P(tqs->stqs_taskq, !=, NULL);
952	ASSERT3U(tqs->stqs_count, !=, 0);
953
954	if (tqs->stqs_count == 1) {
955		tq = tqs->stqs_taskq[0];
956	} else {
957		tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
958	}
959
960	taskq_dispatch_ent(tq, func, arg, flags, ent);
961}
962
963static void
964spa_create_zio_taskqs(spa_t *spa)
965{
966	for (int t = 0; t < ZIO_TYPES; t++) {
967		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
968			spa_taskqs_init(spa, t, q);
969		}
970	}
971}
972
973#ifdef _KERNEL
974static void
975spa_thread(void *arg)
976{
977	callb_cpr_t cprinfo;
978
979	spa_t *spa = arg;
980	user_t *pu = PTOU(curproc);
981
982	CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
983	    spa->spa_name);
984
985	ASSERT(curproc != &p0);
986	(void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
987	    "zpool-%s", spa->spa_name);
988	(void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
989
990	/* bind this thread to the requested psrset */
991	if (zio_taskq_psrset_bind != PS_NONE) {
992		pool_lock();
993		mutex_enter(&cpu_lock);
994		mutex_enter(&pidlock);
995		mutex_enter(&curproc->p_lock);
996
997		if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
998		    0, NULL, NULL) == 0)  {
999			curthread->t_bind_pset = zio_taskq_psrset_bind;
1000		} else {
1001			cmn_err(CE_WARN,
1002			    "Couldn't bind process for zfs pool \"%s\" to "
1003			    "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1004		}
1005
1006		mutex_exit(&curproc->p_lock);
1007		mutex_exit(&pidlock);
1008		mutex_exit(&cpu_lock);
1009		pool_unlock();
1010	}
1011
1012	if (zio_taskq_sysdc) {
1013		sysdc_thread_enter(curthread, 100, 0);
1014	}
1015
1016	spa->spa_proc = curproc;
1017	spa->spa_did = curthread->t_did;
1018
1019	spa_create_zio_taskqs(spa);
1020
1021	mutex_enter(&spa->spa_proc_lock);
1022	ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1023
1024	spa->spa_proc_state = SPA_PROC_ACTIVE;
1025	cv_broadcast(&spa->spa_proc_cv);
1026
1027	CALLB_CPR_SAFE_BEGIN(&cprinfo);
1028	while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1029		cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1030	CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1031
1032	ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1033	spa->spa_proc_state = SPA_PROC_GONE;
1034	spa->spa_proc = &p0;
1035	cv_broadcast(&spa->spa_proc_cv);
1036	CALLB_CPR_EXIT(&cprinfo);	/* drops spa_proc_lock */
1037
1038	mutex_enter(&curproc->p_lock);
1039	lwp_exit();
1040}
1041#endif
1042
1043/*
1044 * Activate an uninitialized pool.
1045 */
1046static void
1047spa_activate(spa_t *spa, int mode)
1048{
1049	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1050
1051	spa->spa_state = POOL_STATE_ACTIVE;
1052	spa->spa_mode = mode;
1053
1054	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1055	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1056
1057	/* Try to create a covering process */
1058	mutex_enter(&spa->spa_proc_lock);
1059	ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1060	ASSERT(spa->spa_proc == &p0);
1061	spa->spa_did = 0;
1062
1063	/* Only create a process if we're going to be around a while. */
1064	if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1065		if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1066		    NULL, 0) == 0) {
1067			spa->spa_proc_state = SPA_PROC_CREATED;
1068			while (spa->spa_proc_state == SPA_PROC_CREATED) {
1069				cv_wait(&spa->spa_proc_cv,
1070				    &spa->spa_proc_lock);
1071			}
1072			ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1073			ASSERT(spa->spa_proc != &p0);
1074			ASSERT(spa->spa_did != 0);
1075		} else {
1076#ifdef _KERNEL
1077			cmn_err(CE_WARN,
1078			    "Couldn't create process for zfs pool \"%s\"\n",
1079			    spa->spa_name);
1080#endif
1081		}
1082	}
1083	mutex_exit(&spa->spa_proc_lock);
1084
1085	/* If we didn't create a process, we need to create our taskqs. */
1086	if (spa->spa_proc == &p0) {
1087		spa_create_zio_taskqs(spa);
1088	}
1089
1090	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1091	    offsetof(vdev_t, vdev_config_dirty_node));
1092	list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1093	    offsetof(objset_t, os_evicting_node));
1094	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1095	    offsetof(vdev_t, vdev_state_dirty_node));
1096
1097	txg_list_create(&spa->spa_vdev_txg_list,
1098	    offsetof(struct vdev, vdev_txg_node));
1099
1100	avl_create(&spa->spa_errlist_scrub,
1101	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1102	    offsetof(spa_error_entry_t, se_avl));
1103	avl_create(&spa->spa_errlist_last,
1104	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1105	    offsetof(spa_error_entry_t, se_avl));
1106}
1107
1108/*
1109 * Opposite of spa_activate().
1110 */
1111static void
1112spa_deactivate(spa_t *spa)
1113{
1114	ASSERT(spa->spa_sync_on == B_FALSE);
1115	ASSERT(spa->spa_dsl_pool == NULL);
1116	ASSERT(spa->spa_root_vdev == NULL);
1117	ASSERT(spa->spa_async_zio_root == NULL);
1118	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1119
1120	spa_evicting_os_wait(spa);
1121
1122	txg_list_destroy(&spa->spa_vdev_txg_list);
1123
1124	list_destroy(&spa->spa_config_dirty_list);
1125	list_destroy(&spa->spa_evicting_os_list);
1126	list_destroy(&spa->spa_state_dirty_list);
1127
1128	for (int t = 0; t < ZIO_TYPES; t++) {
1129		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1130			spa_taskqs_fini(spa, t, q);
1131		}
1132	}
1133
1134	metaslab_class_destroy(spa->spa_normal_class);
1135	spa->spa_normal_class = NULL;
1136
1137	metaslab_class_destroy(spa->spa_log_class);
1138	spa->spa_log_class = NULL;
1139
1140	/*
1141	 * If this was part of an import or the open otherwise failed, we may
1142	 * still have errors left in the queues.  Empty them just in case.
1143	 */
1144	spa_errlog_drain(spa);
1145
1146	avl_destroy(&spa->spa_errlist_scrub);
1147	avl_destroy(&spa->spa_errlist_last);
1148
1149	spa->spa_state = POOL_STATE_UNINITIALIZED;
1150
1151	mutex_enter(&spa->spa_proc_lock);
1152	if (spa->spa_proc_state != SPA_PROC_NONE) {
1153		ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1154		spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1155		cv_broadcast(&spa->spa_proc_cv);
1156		while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1157			ASSERT(spa->spa_proc != &p0);
1158			cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1159		}
1160		ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1161		spa->spa_proc_state = SPA_PROC_NONE;
1162	}
1163	ASSERT(spa->spa_proc == &p0);
1164	mutex_exit(&spa->spa_proc_lock);
1165
1166	/*
1167	 * We want to make sure spa_thread() has actually exited the ZFS
1168	 * module, so that the module can't be unloaded out from underneath
1169	 * it.
1170	 */
1171	if (spa->spa_did != 0) {
1172		thread_join(spa->spa_did);
1173		spa->spa_did = 0;
1174	}
1175}
1176
1177/*
1178 * Verify a pool configuration, and construct the vdev tree appropriately.  This
1179 * will create all the necessary vdevs in the appropriate layout, with each vdev
1180 * in the CLOSED state.  This will prep the pool before open/creation/import.
1181 * All vdev validation is done by the vdev_alloc() routine.
1182 */
1183static int
1184spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1185    uint_t id, int atype)
1186{
1187	nvlist_t **child;
1188	uint_t children;
1189	int error;
1190
1191	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1192		return (error);
1193
1194	if ((*vdp)->vdev_ops->vdev_op_leaf)
1195		return (0);
1196
1197	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1198	    &child, &children);
1199
1200	if (error == ENOENT)
1201		return (0);
1202
1203	if (error) {
1204		vdev_free(*vdp);
1205		*vdp = NULL;
1206		return (SET_ERROR(EINVAL));
1207	}
1208
1209	for (int c = 0; c < children; c++) {
1210		vdev_t *vd;
1211		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1212		    atype)) != 0) {
1213			vdev_free(*vdp);
1214			*vdp = NULL;
1215			return (error);
1216		}
1217	}
1218
1219	ASSERT(*vdp != NULL);
1220
1221	return (0);
1222}
1223
1224/*
1225 * Opposite of spa_load().
1226 */
1227static void
1228spa_unload(spa_t *spa)
1229{
1230	int i;
1231
1232	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1233
1234	/*
1235	 * Stop async tasks.
1236	 */
1237	spa_async_suspend(spa);
1238
1239	/*
1240	 * Stop syncing.
1241	 */
1242	if (spa->spa_sync_on) {
1243		txg_sync_stop(spa->spa_dsl_pool);
1244		spa->spa_sync_on = B_FALSE;
1245	}
1246
1247	/*
1248	 * Wait for any outstanding async I/O to complete.
1249	 */
1250	if (spa->spa_async_zio_root != NULL) {
1251		for (int i = 0; i < max_ncpus; i++)
1252			(void) zio_wait(spa->spa_async_zio_root[i]);
1253		kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1254		spa->spa_async_zio_root = NULL;
1255	}
1256
1257	bpobj_close(&spa->spa_deferred_bpobj);
1258
1259	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1260
1261	/*
1262	 * Close all vdevs.
1263	 */
1264	if (spa->spa_root_vdev)
1265		vdev_free(spa->spa_root_vdev);
1266	ASSERT(spa->spa_root_vdev == NULL);
1267
1268	/*
1269	 * Close the dsl pool.
1270	 */
1271	if (spa->spa_dsl_pool) {
1272		dsl_pool_close(spa->spa_dsl_pool);
1273		spa->spa_dsl_pool = NULL;
1274		spa->spa_meta_objset = NULL;
1275	}
1276
1277	ddt_unload(spa);
1278
1279
1280	/*
1281	 * Drop and purge level 2 cache
1282	 */
1283	spa_l2cache_drop(spa);
1284
1285	for (i = 0; i < spa->spa_spares.sav_count; i++)
1286		vdev_free(spa->spa_spares.sav_vdevs[i]);
1287	if (spa->spa_spares.sav_vdevs) {
1288		kmem_free(spa->spa_spares.sav_vdevs,
1289		    spa->spa_spares.sav_count * sizeof (void *));
1290		spa->spa_spares.sav_vdevs = NULL;
1291	}
1292	if (spa->spa_spares.sav_config) {
1293		nvlist_free(spa->spa_spares.sav_config);
1294		spa->spa_spares.sav_config = NULL;
1295	}
1296	spa->spa_spares.sav_count = 0;
1297
1298	for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1299		vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1300		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1301	}
1302	if (spa->spa_l2cache.sav_vdevs) {
1303		kmem_free(spa->spa_l2cache.sav_vdevs,
1304		    spa->spa_l2cache.sav_count * sizeof (void *));
1305		spa->spa_l2cache.sav_vdevs = NULL;
1306	}
1307	if (spa->spa_l2cache.sav_config) {
1308		nvlist_free(spa->spa_l2cache.sav_config);
1309		spa->spa_l2cache.sav_config = NULL;
1310	}
1311	spa->spa_l2cache.sav_count = 0;
1312
1313	spa->spa_async_suspended = 0;
1314
1315	if (spa->spa_comment != NULL) {
1316		spa_strfree(spa->spa_comment);
1317		spa->spa_comment = NULL;
1318	}
1319
1320	spa_config_exit(spa, SCL_ALL, FTAG);
1321}
1322
1323/*
1324 * Load (or re-load) the current list of vdevs describing the active spares for
1325 * this pool.  When this is called, we have some form of basic information in
1326 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1327 * then re-generate a more complete list including status information.
1328 */
1329static void
1330spa_load_spares(spa_t *spa)
1331{
1332	nvlist_t **spares;
1333	uint_t nspares;
1334	int i;
1335	vdev_t *vd, *tvd;
1336
1337	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1338
1339	/*
1340	 * First, close and free any existing spare vdevs.
1341	 */
1342	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1343		vd = spa->spa_spares.sav_vdevs[i];
1344
1345		/* Undo the call to spa_activate() below */
1346		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1347		    B_FALSE)) != NULL && tvd->vdev_isspare)
1348			spa_spare_remove(tvd);
1349		vdev_close(vd);
1350		vdev_free(vd);
1351	}
1352
1353	if (spa->spa_spares.sav_vdevs)
1354		kmem_free(spa->spa_spares.sav_vdevs,
1355		    spa->spa_spares.sav_count * sizeof (void *));
1356
1357	if (spa->spa_spares.sav_config == NULL)
1358		nspares = 0;
1359	else
1360		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1361		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1362
1363	spa->spa_spares.sav_count = (int)nspares;
1364	spa->spa_spares.sav_vdevs = NULL;
1365
1366	if (nspares == 0)
1367		return;
1368
1369	/*
1370	 * Construct the array of vdevs, opening them to get status in the
1371	 * process.   For each spare, there is potentially two different vdev_t
1372	 * structures associated with it: one in the list of spares (used only
1373	 * for basic validation purposes) and one in the active vdev
1374	 * configuration (if it's spared in).  During this phase we open and
1375	 * validate each vdev on the spare list.  If the vdev also exists in the
1376	 * active configuration, then we also mark this vdev as an active spare.
1377	 */
1378	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1379	    KM_SLEEP);
1380	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1381		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1382		    VDEV_ALLOC_SPARE) == 0);
1383		ASSERT(vd != NULL);
1384
1385		spa->spa_spares.sav_vdevs[i] = vd;
1386
1387		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1388		    B_FALSE)) != NULL) {
1389			if (!tvd->vdev_isspare)
1390				spa_spare_add(tvd);
1391
1392			/*
1393			 * We only mark the spare active if we were successfully
1394			 * able to load the vdev.  Otherwise, importing a pool
1395			 * with a bad active spare would result in strange
1396			 * behavior, because multiple pool would think the spare
1397			 * is actively in use.
1398			 *
1399			 * There is a vulnerability here to an equally bizarre
1400			 * circumstance, where a dead active spare is later
1401			 * brought back to life (onlined or otherwise).  Given
1402			 * the rarity of this scenario, and the extra complexity
1403			 * it adds, we ignore the possibility.
1404			 */
1405			if (!vdev_is_dead(tvd))
1406				spa_spare_activate(tvd);
1407		}
1408
1409		vd->vdev_top = vd;
1410		vd->vdev_aux = &spa->spa_spares;
1411
1412		if (vdev_open(vd) != 0)
1413			continue;
1414
1415		if (vdev_validate_aux(vd) == 0)
1416			spa_spare_add(vd);
1417	}
1418
1419	/*
1420	 * Recompute the stashed list of spares, with status information
1421	 * this time.
1422	 */
1423	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1424	    DATA_TYPE_NVLIST_ARRAY) == 0);
1425
1426	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1427	    KM_SLEEP);
1428	for (i = 0; i < spa->spa_spares.sav_count; i++)
1429		spares[i] = vdev_config_generate(spa,
1430		    spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1431	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1432	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1433	for (i = 0; i < spa->spa_spares.sav_count; i++)
1434		nvlist_free(spares[i]);
1435	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1436}
1437
1438/*
1439 * Load (or re-load) the current list of vdevs describing the active l2cache for
1440 * this pool.  When this is called, we have some form of basic information in
1441 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1442 * then re-generate a more complete list including status information.
1443 * Devices which are already active have their details maintained, and are
1444 * not re-opened.
1445 */
1446static void
1447spa_load_l2cache(spa_t *spa)
1448{
1449	nvlist_t **l2cache;
1450	uint_t nl2cache;
1451	int i, j, oldnvdevs;
1452	uint64_t guid;
1453	vdev_t *vd, **oldvdevs, **newvdevs;
1454	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1455
1456	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1457
1458	if (sav->sav_config != NULL) {
1459		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1460		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1461		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1462	} else {
1463		nl2cache = 0;
1464		newvdevs = NULL;
1465	}
1466
1467	oldvdevs = sav->sav_vdevs;
1468	oldnvdevs = sav->sav_count;
1469	sav->sav_vdevs = NULL;
1470	sav->sav_count = 0;
1471
1472	/*
1473	 * Process new nvlist of vdevs.
1474	 */
1475	for (i = 0; i < nl2cache; i++) {
1476		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1477		    &guid) == 0);
1478
1479		newvdevs[i] = NULL;
1480		for (j = 0; j < oldnvdevs; j++) {
1481			vd = oldvdevs[j];
1482			if (vd != NULL && guid == vd->vdev_guid) {
1483				/*
1484				 * Retain previous vdev for add/remove ops.
1485				 */
1486				newvdevs[i] = vd;
1487				oldvdevs[j] = NULL;
1488				break;
1489			}
1490		}
1491
1492		if (newvdevs[i] == NULL) {
1493			/*
1494			 * Create new vdev
1495			 */
1496			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1497			    VDEV_ALLOC_L2CACHE) == 0);
1498			ASSERT(vd != NULL);
1499			newvdevs[i] = vd;
1500
1501			/*
1502			 * Commit this vdev as an l2cache device,
1503			 * even if it fails to open.
1504			 */
1505			spa_l2cache_add(vd);
1506
1507			vd->vdev_top = vd;
1508			vd->vdev_aux = sav;
1509
1510			spa_l2cache_activate(vd);
1511
1512			if (vdev_open(vd) != 0)
1513				continue;
1514
1515			(void) vdev_validate_aux(vd);
1516
1517			if (!vdev_is_dead(vd))
1518				l2arc_add_vdev(spa, vd);
1519		}
1520	}
1521
1522	/*
1523	 * Purge vdevs that were dropped
1524	 */
1525	for (i = 0; i < oldnvdevs; i++) {
1526		uint64_t pool;
1527
1528		vd = oldvdevs[i];
1529		if (vd != NULL) {
1530			ASSERT(vd->vdev_isl2cache);
1531
1532			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1533			    pool != 0ULL && l2arc_vdev_present(vd))
1534				l2arc_remove_vdev(vd);
1535			vdev_clear_stats(vd);
1536			vdev_free(vd);
1537		}
1538	}
1539
1540	if (oldvdevs)
1541		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1542
1543	if (sav->sav_config == NULL)
1544		goto out;
1545
1546	sav->sav_vdevs = newvdevs;
1547	sav->sav_count = (int)nl2cache;
1548
1549	/*
1550	 * Recompute the stashed list of l2cache devices, with status
1551	 * information this time.
1552	 */
1553	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1554	    DATA_TYPE_NVLIST_ARRAY) == 0);
1555
1556	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1557	for (i = 0; i < sav->sav_count; i++)
1558		l2cache[i] = vdev_config_generate(spa,
1559		    sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1560	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1561	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1562out:
1563	for (i = 0; i < sav->sav_count; i++)
1564		nvlist_free(l2cache[i]);
1565	if (sav->sav_count)
1566		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1567}
1568
1569static int
1570load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1571{
1572	dmu_buf_t *db;
1573	char *packed = NULL;
1574	size_t nvsize = 0;
1575	int error;
1576	*value = NULL;
1577
1578	error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1579	if (error != 0)
1580		return (error);
1581
1582	nvsize = *(uint64_t *)db->db_data;
1583	dmu_buf_rele(db, FTAG);
1584
1585	packed = kmem_alloc(nvsize, KM_SLEEP);
1586	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1587	    DMU_READ_PREFETCH);
1588	if (error == 0)
1589		error = nvlist_unpack(packed, nvsize, value, 0);
1590	kmem_free(packed, nvsize);
1591
1592	return (error);
1593}
1594
1595/*
1596 * Checks to see if the given vdev could not be opened, in which case we post a
1597 * sysevent to notify the autoreplace code that the device has been removed.
1598 */
1599static void
1600spa_check_removed(vdev_t *vd)
1601{
1602	for (int c = 0; c < vd->vdev_children; c++)
1603		spa_check_removed(vd->vdev_child[c]);
1604
1605	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1606	    !vd->vdev_ishole) {
1607		zfs_post_autoreplace(vd->vdev_spa, vd);
1608		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1609	}
1610}
1611
1612/*
1613 * Validate the current config against the MOS config
1614 */
1615static boolean_t
1616spa_config_valid(spa_t *spa, nvlist_t *config)
1617{
1618	vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1619	nvlist_t *nv;
1620
1621	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1622
1623	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1624	VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1625
1626	ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1627
1628	/*
1629	 * If we're doing a normal import, then build up any additional
1630	 * diagnostic information about missing devices in this config.
1631	 * We'll pass this up to the user for further processing.
1632	 */
1633	if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1634		nvlist_t **child, *nv;
1635		uint64_t idx = 0;
1636
1637		child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1638		    KM_SLEEP);
1639		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1640
1641		for (int c = 0; c < rvd->vdev_children; c++) {
1642			vdev_t *tvd = rvd->vdev_child[c];
1643			vdev_t *mtvd  = mrvd->vdev_child[c];
1644
1645			if (tvd->vdev_ops == &vdev_missing_ops &&
1646			    mtvd->vdev_ops != &vdev_missing_ops &&
1647			    mtvd->vdev_islog)
1648				child[idx++] = vdev_config_generate(spa, mtvd,
1649				    B_FALSE, 0);
1650		}
1651
1652		if (idx) {
1653			VERIFY(nvlist_add_nvlist_array(nv,
1654			    ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1655			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1656			    ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1657
1658			for (int i = 0; i < idx; i++)
1659				nvlist_free(child[i]);
1660		}
1661		nvlist_free(nv);
1662		kmem_free(child, rvd->vdev_children * sizeof (char **));
1663	}
1664
1665	/*
1666	 * Compare the root vdev tree with the information we have
1667	 * from the MOS config (mrvd). Check each top-level vdev
1668	 * with the corresponding MOS config top-level (mtvd).
1669	 */
1670	for (int c = 0; c < rvd->vdev_children; c++) {
1671		vdev_t *tvd = rvd->vdev_child[c];
1672		vdev_t *mtvd  = mrvd->vdev_child[c];
1673
1674		/*
1675		 * Resolve any "missing" vdevs in the current configuration.
1676		 * If we find that the MOS config has more accurate information
1677		 * about the top-level vdev then use that vdev instead.
1678		 */
1679		if (tvd->vdev_ops == &vdev_missing_ops &&
1680		    mtvd->vdev_ops != &vdev_missing_ops) {
1681
1682			if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1683				continue;
1684
1685			/*
1686			 * Device specific actions.
1687			 */
1688			if (mtvd->vdev_islog) {
1689				spa_set_log_state(spa, SPA_LOG_CLEAR);
1690			} else {
1691				/*
1692				 * XXX - once we have 'readonly' pool
1693				 * support we should be able to handle
1694				 * missing data devices by transitioning
1695				 * the pool to readonly.
1696				 */
1697				continue;
1698			}
1699
1700			/*
1701			 * Swap the missing vdev with the data we were
1702			 * able to obtain from the MOS config.
1703			 */
1704			vdev_remove_child(rvd, tvd);
1705			vdev_remove_child(mrvd, mtvd);
1706
1707			vdev_add_child(rvd, mtvd);
1708			vdev_add_child(mrvd, tvd);
1709
1710			spa_config_exit(spa, SCL_ALL, FTAG);
1711			vdev_load(mtvd);
1712			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1713
1714			vdev_reopen(rvd);
1715		} else if (mtvd->vdev_islog) {
1716			/*
1717			 * Load the slog device's state from the MOS config
1718			 * since it's possible that the label does not
1719			 * contain the most up-to-date information.
1720			 */
1721			vdev_load_log_state(tvd, mtvd);
1722			vdev_reopen(tvd);
1723		}
1724	}
1725	vdev_free(mrvd);
1726	spa_config_exit(spa, SCL_ALL, FTAG);
1727
1728	/*
1729	 * Ensure we were able to validate the config.
1730	 */
1731	return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1732}
1733
1734/*
1735 * Check for missing log devices
1736 */
1737static boolean_t
1738spa_check_logs(spa_t *spa)
1739{
1740	boolean_t rv = B_FALSE;
1741
1742	switch (spa->spa_log_state) {
1743	case SPA_LOG_MISSING:
1744		/* need to recheck in case slog has been restored */
1745	case SPA_LOG_UNKNOWN:
1746		rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1747		    NULL, DS_FIND_CHILDREN) != 0);
1748		if (rv)
1749			spa_set_log_state(spa, SPA_LOG_MISSING);
1750		break;
1751	}
1752	return (rv);
1753}
1754
1755static boolean_t
1756spa_passivate_log(spa_t *spa)
1757{
1758	vdev_t *rvd = spa->spa_root_vdev;
1759	boolean_t slog_found = B_FALSE;
1760
1761	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1762
1763	if (!spa_has_slogs(spa))
1764		return (B_FALSE);
1765
1766	for (int c = 0; c < rvd->vdev_children; c++) {
1767		vdev_t *tvd = rvd->vdev_child[c];
1768		metaslab_group_t *mg = tvd->vdev_mg;
1769
1770		if (tvd->vdev_islog) {
1771			metaslab_group_passivate(mg);
1772			slog_found = B_TRUE;
1773		}
1774	}
1775
1776	return (slog_found);
1777}
1778
1779static void
1780spa_activate_log(spa_t *spa)
1781{
1782	vdev_t *rvd = spa->spa_root_vdev;
1783
1784	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1785
1786	for (int c = 0; c < rvd->vdev_children; c++) {
1787		vdev_t *tvd = rvd->vdev_child[c];
1788		metaslab_group_t *mg = tvd->vdev_mg;
1789
1790		if (tvd->vdev_islog)
1791			metaslab_group_activate(mg);
1792	}
1793}
1794
1795int
1796spa_offline_log(spa_t *spa)
1797{
1798	int error;
1799
1800	error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1801	    NULL, DS_FIND_CHILDREN);
1802	if (error == 0) {
1803		/*
1804		 * We successfully offlined the log device, sync out the
1805		 * current txg so that the "stubby" block can be removed
1806		 * by zil_sync().
1807		 */
1808		txg_wait_synced(spa->spa_dsl_pool, 0);
1809	}
1810	return (error);
1811}
1812
1813static void
1814spa_aux_check_removed(spa_aux_vdev_t *sav)
1815{
1816	for (int i = 0; i < sav->sav_count; i++)
1817		spa_check_removed(sav->sav_vdevs[i]);
1818}
1819
1820void
1821spa_claim_notify(zio_t *zio)
1822{
1823	spa_t *spa = zio->io_spa;
1824
1825	if (zio->io_error)
1826		return;
1827
1828	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1829	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1830		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1831	mutex_exit(&spa->spa_props_lock);
1832}
1833
1834typedef struct spa_load_error {
1835	uint64_t	sle_meta_count;
1836	uint64_t	sle_data_count;
1837} spa_load_error_t;
1838
1839static void
1840spa_load_verify_done(zio_t *zio)
1841{
1842	blkptr_t *bp = zio->io_bp;
1843	spa_load_error_t *sle = zio->io_private;
1844	dmu_object_type_t type = BP_GET_TYPE(bp);
1845	int error = zio->io_error;
1846	spa_t *spa = zio->io_spa;
1847
1848	if (error) {
1849		if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1850		    type != DMU_OT_INTENT_LOG)
1851			atomic_inc_64(&sle->sle_meta_count);
1852		else
1853			atomic_inc_64(&sle->sle_data_count);
1854	}
1855	zio_data_buf_free(zio->io_data, zio->io_size);
1856
1857	mutex_enter(&spa->spa_scrub_lock);
1858	spa->spa_scrub_inflight--;
1859	cv_broadcast(&spa->spa_scrub_io_cv);
1860	mutex_exit(&spa->spa_scrub_lock);
1861}
1862
1863/*
1864 * Maximum number of concurrent scrub i/os to create while verifying
1865 * a pool while importing it.
1866 */
1867int spa_load_verify_maxinflight = 10000;
1868boolean_t spa_load_verify_metadata = B_TRUE;
1869boolean_t spa_load_verify_data = B_TRUE;
1870
1871/*ARGSUSED*/
1872static int
1873spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1874    const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1875{
1876	if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
1877		return (0);
1878	/*
1879	 * Note: normally this routine will not be called if
1880	 * spa_load_verify_metadata is not set.  However, it may be useful
1881	 * to manually set the flag after the traversal has begun.
1882	 */
1883	if (!spa_load_verify_metadata)
1884		return (0);
1885	if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data)
1886		return (0);
1887
1888	zio_t *rio = arg;
1889	size_t size = BP_GET_PSIZE(bp);
1890	void *data = zio_data_buf_alloc(size);
1891
1892	mutex_enter(&spa->spa_scrub_lock);
1893	while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
1894		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1895	spa->spa_scrub_inflight++;
1896	mutex_exit(&spa->spa_scrub_lock);
1897
1898	zio_nowait(zio_read(rio, spa, bp, data, size,
1899	    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1900	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1901	    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1902	return (0);
1903}
1904
1905static int
1906spa_load_verify(spa_t *spa)
1907{
1908	zio_t *rio;
1909	spa_load_error_t sle = { 0 };
1910	zpool_rewind_policy_t policy;
1911	boolean_t verify_ok = B_FALSE;
1912	int error = 0;
1913
1914	zpool_get_rewind_policy(spa->spa_config, &policy);
1915
1916	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1917		return (0);
1918
1919	rio = zio_root(spa, NULL, &sle,
1920	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1921
1922	if (spa_load_verify_metadata) {
1923		error = traverse_pool(spa, spa->spa_verify_min_txg,
1924		    TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
1925		    spa_load_verify_cb, rio);
1926	}
1927
1928	(void) zio_wait(rio);
1929
1930	spa->spa_load_meta_errors = sle.sle_meta_count;
1931	spa->spa_load_data_errors = sle.sle_data_count;
1932
1933	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1934	    sle.sle_data_count <= policy.zrp_maxdata) {
1935		int64_t loss = 0;
1936
1937		verify_ok = B_TRUE;
1938		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1939		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1940
1941		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1942		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1943		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1944		VERIFY(nvlist_add_int64(spa->spa_load_info,
1945		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1946		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1947		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1948	} else {
1949		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1950	}
1951
1952	if (error) {
1953		if (error != ENXIO && error != EIO)
1954			error = SET_ERROR(EIO);
1955		return (error);
1956	}
1957
1958	return (verify_ok ? 0 : EIO);
1959}
1960
1961/*
1962 * Find a value in the pool props object.
1963 */
1964static void
1965spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1966{
1967	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1968	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1969}
1970
1971/*
1972 * Find a value in the pool directory object.
1973 */
1974static int
1975spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1976{
1977	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1978	    name, sizeof (uint64_t), 1, val));
1979}
1980
1981static int
1982spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1983{
1984	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1985	return (err);
1986}
1987
1988/*
1989 * Fix up config after a partly-completed split.  This is done with the
1990 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1991 * pool have that entry in their config, but only the splitting one contains
1992 * a list of all the guids of the vdevs that are being split off.
1993 *
1994 * This function determines what to do with that list: either rejoin
1995 * all the disks to the pool, or complete the splitting process.  To attempt
1996 * the rejoin, each disk that is offlined is marked online again, and
1997 * we do a reopen() call.  If the vdev label for every disk that was
1998 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1999 * then we call vdev_split() on each disk, and complete the split.
2000 *
2001 * Otherwise we leave the config alone, with all the vdevs in place in
2002 * the original pool.
2003 */
2004static void
2005spa_try_repair(spa_t *spa, nvlist_t *config)
2006{
2007	uint_t extracted;
2008	uint64_t *glist;
2009	uint_t i, gcount;
2010	nvlist_t *nvl;
2011	vdev_t **vd;
2012	boolean_t attempt_reopen;
2013
2014	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2015		return;
2016
2017	/* check that the config is complete */
2018	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2019	    &glist, &gcount) != 0)
2020		return;
2021
2022	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2023
2024	/* attempt to online all the vdevs & validate */
2025	attempt_reopen = B_TRUE;
2026	for (i = 0; i < gcount; i++) {
2027		if (glist[i] == 0)	/* vdev is hole */
2028			continue;
2029
2030		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2031		if (vd[i] == NULL) {
2032			/*
2033			 * Don't bother attempting to reopen the disks;
2034			 * just do the split.
2035			 */
2036			attempt_reopen = B_FALSE;
2037		} else {
2038			/* attempt to re-online it */
2039			vd[i]->vdev_offline = B_FALSE;
2040		}
2041	}
2042
2043	if (attempt_reopen) {
2044		vdev_reopen(spa->spa_root_vdev);
2045
2046		/* check each device to see what state it's in */
2047		for (extracted = 0, i = 0; i < gcount; i++) {
2048			if (vd[i] != NULL &&
2049			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2050				break;
2051			++extracted;
2052		}
2053	}
2054
2055	/*
2056	 * If every disk has been moved to the new pool, or if we never
2057	 * even attempted to look at them, then we split them off for
2058	 * good.
2059	 */
2060	if (!attempt_reopen || gcount == extracted) {
2061		for (i = 0; i < gcount; i++)
2062			if (vd[i] != NULL)
2063				vdev_split(vd[i]);
2064		vdev_reopen(spa->spa_root_vdev);
2065	}
2066
2067	kmem_free(vd, gcount * sizeof (vdev_t *));
2068}
2069
2070static int
2071spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2072    boolean_t mosconfig)
2073{
2074	nvlist_t *config = spa->spa_config;
2075	char *ereport = FM_EREPORT_ZFS_POOL;
2076	char *comment;
2077	int error;
2078	uint64_t pool_guid;
2079	nvlist_t *nvl;
2080
2081	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2082		return (SET_ERROR(EINVAL));
2083
2084	ASSERT(spa->spa_comment == NULL);
2085	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2086		spa->spa_comment = spa_strdup(comment);
2087
2088	/*
2089	 * Versioning wasn't explicitly added to the label until later, so if
2090	 * it's not present treat it as the initial version.
2091	 */
2092	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2093	    &spa->spa_ubsync.ub_version) != 0)
2094		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2095
2096	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2097	    &spa->spa_config_txg);
2098
2099	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2100	    spa_guid_exists(pool_guid, 0)) {
2101		error = SET_ERROR(EEXIST);
2102	} else {
2103		spa->spa_config_guid = pool_guid;
2104
2105		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2106		    &nvl) == 0) {
2107			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2108			    KM_SLEEP) == 0);
2109		}
2110
2111		nvlist_free(spa->spa_load_info);
2112		spa->spa_load_info = fnvlist_alloc();
2113
2114		gethrestime(&spa->spa_loaded_ts);
2115		error = spa_load_impl(spa, pool_guid, config, state, type,
2116		    mosconfig, &ereport);
2117	}
2118
2119	/*
2120	 * Don't count references from objsets that are already closed
2121	 * and are making their way through the eviction process.
2122	 */
2123	spa_evicting_os_wait(spa);
2124	spa->spa_minref = refcount_count(&spa->spa_refcount);
2125	if (error) {
2126		if (error != EEXIST) {
2127			spa->spa_loaded_ts.tv_sec = 0;
2128			spa->spa_loaded_ts.tv_nsec = 0;
2129		}
2130		if (error != EBADF) {
2131			zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2132		}
2133	}
2134	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2135	spa->spa_ena = 0;
2136
2137	return (error);
2138}
2139
2140/*
2141 * Load an existing storage pool, using the pool's builtin spa_config as a
2142 * source of configuration information.
2143 */
2144static int
2145spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2146    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2147    char **ereport)
2148{
2149	int error = 0;
2150	nvlist_t *nvroot = NULL;
2151	nvlist_t *label;
2152	vdev_t *rvd;
2153	uberblock_t *ub = &spa->spa_uberblock;
2154	uint64_t children, config_cache_txg = spa->spa_config_txg;
2155	int orig_mode = spa->spa_mode;
2156	int parse;
2157	uint64_t obj;
2158	boolean_t missing_feat_write = B_FALSE;
2159
2160	/*
2161	 * If this is an untrusted config, access the pool in read-only mode.
2162	 * This prevents things like resilvering recently removed devices.
2163	 */
2164	if (!mosconfig)
2165		spa->spa_mode = FREAD;
2166
2167	ASSERT(MUTEX_HELD(&spa_namespace_lock));
2168
2169	spa->spa_load_state = state;
2170
2171	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2172		return (SET_ERROR(EINVAL));
2173
2174	parse = (type == SPA_IMPORT_EXISTING ?
2175	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2176
2177	/*
2178	 * Create "The Godfather" zio to hold all async IOs
2179	 */
2180	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2181	    KM_SLEEP);
2182	for (int i = 0; i < max_ncpus; i++) {
2183		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2184		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2185		    ZIO_FLAG_GODFATHER);
2186	}
2187
2188	/*
2189	 * Parse the configuration into a vdev tree.  We explicitly set the
2190	 * value that will be returned by spa_version() since parsing the
2191	 * configuration requires knowing the version number.
2192	 */
2193	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2194	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2195	spa_config_exit(spa, SCL_ALL, FTAG);
2196
2197	if (error != 0)
2198		return (error);
2199
2200	ASSERT(spa->spa_root_vdev == rvd);
2201
2202	if (type != SPA_IMPORT_ASSEMBLE) {
2203		ASSERT(spa_guid(spa) == pool_guid);
2204	}
2205
2206	/*
2207	 * Try to open all vdevs, loading each label in the process.
2208	 */
2209	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2210	error = vdev_open(rvd);
2211	spa_config_exit(spa, SCL_ALL, FTAG);
2212	if (error != 0)
2213		return (error);
2214
2215	/*
2216	 * We need to validate the vdev labels against the configuration that
2217	 * we have in hand, which is dependent on the setting of mosconfig. If
2218	 * mosconfig is true then we're validating the vdev labels based on
2219	 * that config.  Otherwise, we're validating against the cached config
2220	 * (zpool.cache) that was read when we loaded the zfs module, and then
2221	 * later we will recursively call spa_load() and validate against
2222	 * the vdev config.
2223	 *
2224	 * If we're assembling a new pool that's been split off from an
2225	 * existing pool, the labels haven't yet been updated so we skip
2226	 * validation for now.
2227	 */
2228	if (type != SPA_IMPORT_ASSEMBLE) {
2229		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2230		error = vdev_validate(rvd, mosconfig);
2231		spa_config_exit(spa, SCL_ALL, FTAG);
2232
2233		if (error != 0)
2234			return (error);
2235
2236		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2237			return (SET_ERROR(ENXIO));
2238	}
2239
2240	/*
2241	 * Find the best uberblock.
2242	 */
2243	vdev_uberblock_load(rvd, ub, &label);
2244
2245	/*
2246	 * If we weren't able to find a single valid uberblock, return failure.
2247	 */
2248	if (ub->ub_txg == 0) {
2249		nvlist_free(label);
2250		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2251	}
2252
2253	/*
2254	 * If the pool has an unsupported version we can't open it.
2255	 */
2256	if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2257		nvlist_free(label);
2258		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2259	}
2260
2261	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2262		nvlist_t *features;
2263
2264		/*
2265		 * If we weren't able to find what's necessary for reading the
2266		 * MOS in the label, return failure.
2267		 */
2268		if (label == NULL || nvlist_lookup_nvlist(label,
2269		    ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2270			nvlist_free(label);
2271			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2272			    ENXIO));
2273		}
2274
2275		/*
2276		 * Update our in-core representation with the definitive values
2277		 * from the label.
2278		 */
2279		nvlist_free(spa->spa_label_features);
2280		VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2281	}
2282
2283	nvlist_free(label);
2284
2285	/*
2286	 * Look through entries in the label nvlist's features_for_read. If
2287	 * there is a feature listed there which we don't understand then we
2288	 * cannot open a pool.
2289	 */
2290	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2291		nvlist_t *unsup_feat;
2292
2293		VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2294		    0);
2295
2296		for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2297		    NULL); nvp != NULL;
2298		    nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2299			if (!zfeature_is_supported(nvpair_name(nvp))) {
2300				VERIFY(nvlist_add_string(unsup_feat,
2301				    nvpair_name(nvp), "") == 0);
2302			}
2303		}
2304
2305		if (!nvlist_empty(unsup_feat)) {
2306			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2307			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2308			nvlist_free(unsup_feat);
2309			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2310			    ENOTSUP));
2311		}
2312
2313		nvlist_free(unsup_feat);
2314	}
2315
2316	/*
2317	 * If the vdev guid sum doesn't match the uberblock, we have an
2318	 * incomplete configuration.  We first check to see if the pool
2319	 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2320	 * If it is, defer the vdev_guid_sum check till later so we
2321	 * can handle missing vdevs.
2322	 */
2323	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2324	    &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2325	    rvd->vdev_guid_sum != ub->ub_guid_sum)
2326		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2327
2328	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2329		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2330		spa_try_repair(spa, config);
2331		spa_config_exit(spa, SCL_ALL, FTAG);
2332		nvlist_free(spa->spa_config_splitting);
2333		spa->spa_config_splitting = NULL;
2334	}
2335
2336	/*
2337	 * Initialize internal SPA structures.
2338	 */
2339	spa->spa_state = POOL_STATE_ACTIVE;
2340	spa->spa_ubsync = spa->spa_uberblock;
2341	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2342	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2343	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2344	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2345	spa->spa_claim_max_txg = spa->spa_first_txg;
2346	spa->spa_prev_software_version = ub->ub_software_version;
2347
2348	error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2349	if (error)
2350		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2351	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2352
2353	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2354		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2355
2356	if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2357		boolean_t missing_feat_read = B_FALSE;
2358		nvlist_t *unsup_feat, *enabled_feat;
2359
2360		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2361		    &spa->spa_feat_for_read_obj) != 0) {
2362			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2363		}
2364
2365		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2366		    &spa->spa_feat_for_write_obj) != 0) {
2367			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2368		}
2369
2370		if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2371		    &spa->spa_feat_desc_obj) != 0) {
2372			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2373		}
2374
2375		enabled_feat = fnvlist_alloc();
2376		unsup_feat = fnvlist_alloc();
2377
2378		if (!spa_features_check(spa, B_FALSE,
2379		    unsup_feat, enabled_feat))
2380			missing_feat_read = B_TRUE;
2381
2382		if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2383			if (!spa_features_check(spa, B_TRUE,
2384			    unsup_feat, enabled_feat)) {
2385				missing_feat_write = B_TRUE;
2386			}
2387		}
2388
2389		fnvlist_add_nvlist(spa->spa_load_info,
2390		    ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2391
2392		if (!nvlist_empty(unsup_feat)) {
2393			fnvlist_add_nvlist(spa->spa_load_info,
2394			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2395		}
2396
2397		fnvlist_free(enabled_feat);
2398		fnvlist_free(unsup_feat);
2399
2400		if (!missing_feat_read) {
2401			fnvlist_add_boolean(spa->spa_load_info,
2402			    ZPOOL_CONFIG_CAN_RDONLY);
2403		}
2404
2405		/*
2406		 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2407		 * twofold: to determine whether the pool is available for
2408		 * import in read-write mode and (if it is not) whether the
2409		 * pool is available for import in read-only mode. If the pool
2410		 * is available for import in read-write mode, it is displayed
2411		 * as available in userland; if it is not available for import
2412		 * in read-only mode, it is displayed as unavailable in
2413		 * userland. If the pool is available for import in read-only
2414		 * mode but not read-write mode, it is displayed as unavailable
2415		 * in userland with a special note that the pool is actually
2416		 * available for open in read-only mode.
2417		 *
2418		 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2419		 * missing a feature for write, we must first determine whether
2420		 * the pool can be opened read-only before returning to
2421		 * userland in order to know whether to display the
2422		 * abovementioned note.
2423		 */
2424		if (missing_feat_read || (missing_feat_write &&
2425		    spa_writeable(spa))) {
2426			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2427			    ENOTSUP));
2428		}
2429
2430		/*
2431		 * Load refcounts for ZFS features from disk into an in-memory
2432		 * cache during SPA initialization.
2433		 */
2434		for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2435			uint64_t refcount;
2436
2437			error = feature_get_refcount_from_disk(spa,
2438			    &spa_feature_table[i], &refcount);
2439			if (error == 0) {
2440				spa->spa_feat_refcount_cache[i] = refcount;
2441			} else if (error == ENOTSUP) {
2442				spa->spa_feat_refcount_cache[i] =
2443				    SPA_FEATURE_DISABLED;
2444			} else {
2445				return (spa_vdev_err(rvd,
2446				    VDEV_AUX_CORRUPT_DATA, EIO));
2447			}
2448		}
2449	}
2450
2451	if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2452		if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2453		    &spa->spa_feat_enabled_txg_obj) != 0)
2454			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2455	}
2456
2457	spa->spa_is_initializing = B_TRUE;
2458	error = dsl_pool_open(spa->spa_dsl_pool);
2459	spa->spa_is_initializing = B_FALSE;
2460	if (error != 0)
2461		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2462
2463	if (!mosconfig) {
2464		uint64_t hostid;
2465		nvlist_t *policy = NULL, *nvconfig;
2466
2467		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2468			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2469
2470		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2471		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2472			char *hostname;
2473			unsigned long myhostid = 0;
2474
2475			VERIFY(nvlist_lookup_string(nvconfig,
2476			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2477
2478#ifdef	_KERNEL
2479			myhostid = zone_get_hostid(NULL);
2480#else	/* _KERNEL */
2481			/*
2482			 * We're emulating the system's hostid in userland, so
2483			 * we can't use zone_get_hostid().
2484			 */
2485			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2486#endif	/* _KERNEL */
2487			if (hostid != 0 && myhostid != 0 &&
2488			    hostid != myhostid) {
2489				nvlist_free(nvconfig);
2490				cmn_err(CE_WARN, "pool '%s' could not be "
2491				    "loaded as it was last accessed by "
2492				    "another system (host: %s hostid: 0x%lx). "
2493				    "See: http://illumos.org/msg/ZFS-8000-EY",
2494				    spa_name(spa), hostname,
2495				    (unsigned long)hostid);
2496				return (SET_ERROR(EBADF));
2497			}
2498		}
2499		if (nvlist_lookup_nvlist(spa->spa_config,
2500		    ZPOOL_REWIND_POLICY, &policy) == 0)
2501			VERIFY(nvlist_add_nvlist(nvconfig,
2502			    ZPOOL_REWIND_POLICY, policy) == 0);
2503
2504		spa_config_set(spa, nvconfig);
2505		spa_unload(spa);
2506		spa_deactivate(spa);
2507		spa_activate(spa, orig_mode);
2508
2509		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2510	}
2511
2512	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2513		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2514	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2515	if (error != 0)
2516		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2517
2518	/*
2519	 * Load the bit that tells us to use the new accounting function
2520	 * (raid-z deflation).  If we have an older pool, this will not
2521	 * be present.
2522	 */
2523	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2524	if (error != 0 && error != ENOENT)
2525		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2526
2527	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2528	    &spa->spa_creation_version);
2529	if (error != 0 && error != ENOENT)
2530		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2531
2532	/*
2533	 * Load the persistent error log.  If we have an older pool, this will
2534	 * not be present.
2535	 */
2536	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2537	if (error != 0 && error != ENOENT)
2538		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2539
2540	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2541	    &spa->spa_errlog_scrub);
2542	if (error != 0 && error != ENOENT)
2543		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2544
2545	/*
2546	 * Load the history object.  If we have an older pool, this
2547	 * will not be present.
2548	 */
2549	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2550	if (error != 0 && error != ENOENT)
2551		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2552
2553	/*
2554	 * If we're assembling the pool from the split-off vdevs of
2555	 * an existing pool, we don't want to attach the spares & cache
2556	 * devices.
2557	 */
2558
2559	/*
2560	 * Load any hot spares for this pool.
2561	 */
2562	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2563	if (error != 0 && error != ENOENT)
2564		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2565	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2566		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2567		if (load_nvlist(spa, spa->spa_spares.sav_object,
2568		    &spa->spa_spares.sav_config) != 0)
2569			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2570
2571		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2572		spa_load_spares(spa);
2573		spa_config_exit(spa, SCL_ALL, FTAG);
2574	} else if (error == 0) {
2575		spa->spa_spares.sav_sync = B_TRUE;
2576	}
2577
2578	/*
2579	 * Load any level 2 ARC devices for this pool.
2580	 */
2581	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2582	    &spa->spa_l2cache.sav_object);
2583	if (error != 0 && error != ENOENT)
2584		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2585	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2586		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2587		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2588		    &spa->spa_l2cache.sav_config) != 0)
2589			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2590
2591		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2592		spa_load_l2cache(spa);
2593		spa_config_exit(spa, SCL_ALL, FTAG);
2594	} else if (error == 0) {
2595		spa->spa_l2cache.sav_sync = B_TRUE;
2596	}
2597
2598	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2599
2600	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2601	if (error && error != ENOENT)
2602		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2603
2604	if (error == 0) {
2605		uint64_t autoreplace;
2606
2607		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2608		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2609		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2610		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2611		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2612		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2613		    &spa->spa_dedup_ditto);
2614
2615		spa->spa_autoreplace = (autoreplace != 0);
2616	}
2617
2618	/*
2619	 * If the 'autoreplace' property is set, then post a resource notifying
2620	 * the ZFS DE that it should not issue any faults for unopenable
2621	 * devices.  We also iterate over the vdevs, and post a sysevent for any
2622	 * unopenable vdevs so that the normal autoreplace handler can take
2623	 * over.
2624	 */
2625	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2626		spa_check_removed(spa->spa_root_vdev);
2627		/*
2628		 * For the import case, this is done in spa_import(), because
2629		 * at this point we're using the spare definitions from
2630		 * the MOS config, not necessarily from the userland config.
2631		 */
2632		if (state != SPA_LOAD_IMPORT) {
2633			spa_aux_check_removed(&spa->spa_spares);
2634			spa_aux_check_removed(&spa->spa_l2cache);
2635		}
2636	}
2637
2638	/*
2639	 * Load the vdev state for all toplevel vdevs.
2640	 */
2641	vdev_load(rvd);
2642
2643	/*
2644	 * Propagate the leaf DTLs we just loaded all the way up the tree.
2645	 */
2646	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2647	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2648	spa_config_exit(spa, SCL_ALL, FTAG);
2649
2650	/*
2651	 * Load the DDTs (dedup tables).
2652	 */
2653	error = ddt_load(spa);
2654	if (error != 0)
2655		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2656
2657	spa_update_dspace(spa);
2658
2659	/*
2660	 * Validate the config, using the MOS config to fill in any
2661	 * information which might be missing.  If we fail to validate
2662	 * the config then declare the pool unfit for use. If we're
2663	 * assembling a pool from a split, the log is not transferred
2664	 * over.
2665	 */
2666	if (type != SPA_IMPORT_ASSEMBLE) {
2667		nvlist_t *nvconfig;
2668
2669		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2670			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2671
2672		if (!spa_config_valid(spa, nvconfig)) {
2673			nvlist_free(nvconfig);
2674			return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2675			    ENXIO));
2676		}
2677		nvlist_free(nvconfig);
2678
2679		/*
2680		 * Now that we've validated the config, check the state of the
2681		 * root vdev.  If it can't be opened, it indicates one or
2682		 * more toplevel vdevs are faulted.
2683		 */
2684		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2685			return (SET_ERROR(ENXIO));
2686
2687		if (spa_check_logs(spa)) {
2688			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2689			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2690		}
2691	}
2692
2693	if (missing_feat_write) {
2694		ASSERT(state == SPA_LOAD_TRYIMPORT);
2695
2696		/*
2697		 * At this point, we know that we can open the pool in
2698		 * read-only mode but not read-write mode. We now have enough
2699		 * information and can return to userland.
2700		 */
2701		return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2702	}
2703
2704	/*
2705	 * We've successfully opened the pool, verify that we're ready
2706	 * to start pushing transactions.
2707	 */
2708	if (state != SPA_LOAD_TRYIMPORT) {
2709		if (error = spa_load_verify(spa))
2710			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2711			    error));
2712	}
2713
2714	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2715	    spa->spa_load_max_txg == UINT64_MAX)) {
2716		dmu_tx_t *tx;
2717		int need_update = B_FALSE;
2718
2719		ASSERT(state != SPA_LOAD_TRYIMPORT);
2720
2721		/*
2722		 * Claim log blocks that haven't been committed yet.
2723		 * This must all happen in a single txg.
2724		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2725		 * invoked from zil_claim_log_block()'s i/o done callback.
2726		 * Price of rollback is that we abandon the log.
2727		 */
2728		spa->spa_claiming = B_TRUE;
2729
2730		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2731		    spa_first_txg(spa));
2732		(void) dmu_objset_find(spa_name(spa),
2733		    zil_claim, tx, DS_FIND_CHILDREN);
2734		dmu_tx_commit(tx);
2735
2736		spa->spa_claiming = B_FALSE;
2737
2738		spa_set_log_state(spa, SPA_LOG_GOOD);
2739		spa->spa_sync_on = B_TRUE;
2740		txg_sync_start(spa->spa_dsl_pool);
2741
2742		/*
2743		 * Wait for all claims to sync.  We sync up to the highest
2744		 * claimed log block birth time so that claimed log blocks
2745		 * don't appear to be from the future.  spa_claim_max_txg
2746		 * will have been set for us by either zil_check_log_chain()
2747		 * (invoked from spa_check_logs()) or zil_claim() above.
2748		 */
2749		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2750
2751		/*
2752		 * If the config cache is stale, or we have uninitialized
2753		 * metaslabs (see spa_vdev_add()), then update the config.
2754		 *
2755		 * If this is a verbatim import, trust the current
2756		 * in-core spa_config and update the disk labels.
2757		 */
2758		if (config_cache_txg != spa->spa_config_txg ||
2759		    state == SPA_LOAD_IMPORT ||
2760		    state == SPA_LOAD_RECOVER ||
2761		    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2762			need_update = B_TRUE;
2763
2764		for (int c = 0; c < rvd->vdev_children; c++)
2765			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2766				need_update = B_TRUE;
2767
2768		/*
2769		 * Update the config cache asychronously in case we're the
2770		 * root pool, in which case the config cache isn't writable yet.
2771		 */
2772		if (need_update)
2773			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2774
2775		/*
2776		 * Check all DTLs to see if anything needs resilvering.
2777		 */
2778		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2779		    vdev_resilver_needed(rvd, NULL, NULL))
2780			spa_async_request(spa, SPA_ASYNC_RESILVER);
2781
2782		/*
2783		 * Log the fact that we booted up (so that we can detect if
2784		 * we rebooted in the middle of an operation).
2785		 */
2786		spa_history_log_version(spa, "open");
2787
2788		/*
2789		 * Delete any inconsistent datasets.
2790		 */
2791		(void) dmu_objset_find(spa_name(spa),
2792		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2793
2794		/*
2795		 * Clean up any stale temporary dataset userrefs.
2796		 */
2797		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2798	}
2799
2800	return (0);
2801}
2802
2803static int
2804spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2805{
2806	int mode = spa->spa_mode;
2807
2808	spa_unload(spa);
2809	spa_deactivate(spa);
2810
2811	spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
2812
2813	spa_activate(spa, mode);
2814	spa_async_suspend(spa);
2815
2816	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2817}
2818
2819/*
2820 * If spa_load() fails this function will try loading prior txg's. If
2821 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2822 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2823 * function will not rewind the pool and will return the same error as
2824 * spa_load().
2825 */
2826static int
2827spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2828    uint64_t max_request, int rewind_flags)
2829{
2830	nvlist_t *loadinfo = NULL;
2831	nvlist_t *config = NULL;
2832	int load_error, rewind_error;
2833	uint64_t safe_rewind_txg;
2834	uint64_t min_txg;
2835
2836	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2837		spa->spa_load_max_txg = spa->spa_load_txg;
2838		spa_set_log_state(spa, SPA_LOG_CLEAR);
2839	} else {
2840		spa->spa_load_max_txg = max_request;
2841		if (max_request != UINT64_MAX)
2842			spa->spa_extreme_rewind = B_TRUE;
2843	}
2844
2845	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2846	    mosconfig);
2847	if (load_error == 0)
2848		return (0);
2849
2850	if (spa->spa_root_vdev != NULL)
2851		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2852
2853	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2854	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2855
2856	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2857		nvlist_free(config);
2858		return (load_error);
2859	}
2860
2861	if (state == SPA_LOAD_RECOVER) {
2862		/* Price of rolling back is discarding txgs, including log */
2863		spa_set_log_state(spa, SPA_LOG_CLEAR);
2864	} else {
2865		/*
2866		 * If we aren't rolling back save the load info from our first
2867		 * import attempt so that we can restore it after attempting
2868		 * to rewind.
2869		 */
2870		loadinfo = spa->spa_load_info;
2871		spa->spa_load_info = fnvlist_alloc();
2872	}
2873
2874	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2875	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2876	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2877	    TXG_INITIAL : safe_rewind_txg;
2878
2879	/*
2880	 * Continue as long as we're finding errors, we're still within
2881	 * the acceptable rewind range, and we're still finding uberblocks
2882	 */
2883	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2884	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2885		if (spa->spa_load_max_txg < safe_rewind_txg)
2886			spa->spa_extreme_rewind = B_TRUE;
2887		rewind_error = spa_load_retry(spa, state, mosconfig);
2888	}
2889
2890	spa->spa_extreme_rewind = B_FALSE;
2891	spa->spa_load_max_txg = UINT64_MAX;
2892
2893	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2894		spa_config_set(spa, config);
2895
2896	if (state == SPA_LOAD_RECOVER) {
2897		ASSERT3P(loadinfo, ==, NULL);
2898		return (rewind_error);
2899	} else {
2900		/* Store the rewind info as part of the initial load info */
2901		fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2902		    spa->spa_load_info);
2903
2904		/* Restore the initial load info */
2905		fnvlist_free(spa->spa_load_info);
2906		spa->spa_load_info = loadinfo;
2907
2908		return (load_error);
2909	}
2910}
2911
2912/*
2913 * Pool Open/Import
2914 *
2915 * The import case is identical to an open except that the configuration is sent
2916 * down from userland, instead of grabbed from the configuration cache.  For the
2917 * case of an open, the pool configuration will exist in the
2918 * POOL_STATE_UNINITIALIZED state.
2919 *
2920 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2921 * the same time open the pool, without having to keep around the spa_t in some
2922 * ambiguous state.
2923 */
2924static int
2925spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2926    nvlist_t **config)
2927{
2928	spa_t *spa;
2929	spa_load_state_t state = SPA_LOAD_OPEN;
2930	int error;
2931	int locked = B_FALSE;
2932
2933	*spapp = NULL;
2934
2935	/*
2936	 * As disgusting as this is, we need to support recursive calls to this
2937	 * function because dsl_dir_open() is called during spa_load(), and ends
2938	 * up calling spa_open() again.  The real fix is to figure out how to
2939	 * avoid dsl_dir_open() calling this in the first place.
2940	 */
2941	if (mutex_owner(&spa_namespace_lock) != curthread) {
2942		mutex_enter(&spa_namespace_lock);
2943		locked = B_TRUE;
2944	}
2945
2946	if ((spa = spa_lookup(pool)) == NULL) {
2947		if (locked)
2948			mutex_exit(&spa_namespace_lock);
2949		return (SET_ERROR(ENOENT));
2950	}
2951
2952	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2953		zpool_rewind_policy_t policy;
2954
2955		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2956		    &policy);
2957		if (policy.zrp_request & ZPOOL_DO_REWIND)
2958			state = SPA_LOAD_RECOVER;
2959
2960		spa_activate(spa, spa_mode_global);
2961
2962		if (state != SPA_LOAD_RECOVER)
2963			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2964
2965		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2966		    policy.zrp_request);
2967
2968		if (error == EBADF) {
2969			/*
2970			 * If vdev_validate() returns failure (indicated by
2971			 * EBADF), it indicates that one of the vdevs indicates
2972			 * that the pool has been exported or destroyed.  If
2973			 * this is the case, the config cache is out of sync and
2974			 * we should remove the pool from the namespace.
2975			 */
2976			spa_unload(spa);
2977			spa_deactivate(spa);
2978			spa_config_sync(spa, B_TRUE, B_TRUE);
2979			spa_remove(spa);
2980			if (locked)
2981				mutex_exit(&spa_namespace_lock);
2982			return (SET_ERROR(ENOENT));
2983		}
2984
2985		if (error) {
2986			/*
2987			 * We can't open the pool, but we still have useful
2988			 * information: the state of each vdev after the
2989			 * attempted vdev_open().  Return this to the user.
2990			 */
2991			if (config != NULL && spa->spa_config) {
2992				VERIFY(nvlist_dup(spa->spa_config, config,
2993				    KM_SLEEP) == 0);
2994				VERIFY(nvlist_add_nvlist(*config,
2995				    ZPOOL_CONFIG_LOAD_INFO,
2996				    spa->spa_load_info) == 0);
2997			}
2998			spa_unload(spa);
2999			spa_deactivate(spa);
3000			spa->spa_last_open_failed = error;
3001			if (locked)
3002				mutex_exit(&spa_namespace_lock);
3003			*spapp = NULL;
3004			return (error);
3005		}
3006	}
3007
3008	spa_open_ref(spa, tag);
3009
3010	if (config != NULL)
3011		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3012
3013	/*
3014	 * If we've recovered the pool, pass back any information we
3015	 * gathered while doing the load.
3016	 */
3017	if (state == SPA_LOAD_RECOVER) {
3018		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
3019		    spa->spa_load_info) == 0);
3020	}
3021
3022	if (locked) {
3023		spa->spa_last_open_failed = 0;
3024		spa->spa_last_ubsync_txg = 0;
3025		spa->spa_load_txg = 0;
3026		mutex_exit(&spa_namespace_lock);
3027	}
3028
3029	*spapp = spa;
3030
3031	return (0);
3032}
3033
3034int
3035spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3036    nvlist_t **config)
3037{
3038	return (spa_open_common(name, spapp, tag, policy, config));
3039}
3040
3041int
3042spa_open(const char *name, spa_t **spapp, void *tag)
3043{
3044	return (spa_open_common(name, spapp, tag, NULL, NULL));
3045}
3046
3047/*
3048 * Lookup the given spa_t, incrementing the inject count in the process,
3049 * preventing it from being exported or destroyed.
3050 */
3051spa_t *
3052spa_inject_addref(char *name)
3053{
3054	spa_t *spa;
3055
3056	mutex_enter(&spa_namespace_lock);
3057	if ((spa = spa_lookup(name)) == NULL) {
3058		mutex_exit(&spa_namespace_lock);
3059		return (NULL);
3060	}
3061	spa->spa_inject_ref++;
3062	mutex_exit(&spa_namespace_lock);
3063
3064	return (spa);
3065}
3066
3067void
3068spa_inject_delref(spa_t *spa)
3069{
3070	mutex_enter(&spa_namespace_lock);
3071	spa->spa_inject_ref--;
3072	mutex_exit(&spa_namespace_lock);
3073}
3074
3075/*
3076 * Add spares device information to the nvlist.
3077 */
3078static void
3079spa_add_spares(spa_t *spa, nvlist_t *config)
3080{
3081	nvlist_t **spares;
3082	uint_t i, nspares;
3083	nvlist_t *nvroot;
3084	uint64_t guid;
3085	vdev_stat_t *vs;
3086	uint_t vsc;
3087	uint64_t pool;
3088
3089	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3090
3091	if (spa->spa_spares.sav_count == 0)
3092		return;
3093
3094	VERIFY(nvlist_lookup_nvlist(config,
3095	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3096	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3097	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3098	if (nspares != 0) {
3099		VERIFY(nvlist_add_nvlist_array(nvroot,
3100		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3101		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3102		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3103
3104		/*
3105		 * Go through and find any spares which have since been
3106		 * repurposed as an active spare.  If this is the case, update
3107		 * their status appropriately.
3108		 */
3109		for (i = 0; i < nspares; i++) {
3110			VERIFY(nvlist_lookup_uint64(spares[i],
3111			    ZPOOL_CONFIG_GUID, &guid) == 0);
3112			if (spa_spare_exists(guid, &pool, NULL) &&
3113			    pool != 0ULL) {
3114				VERIFY(nvlist_lookup_uint64_array(
3115				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
3116				    (uint64_t **)&vs, &vsc) == 0);
3117				vs->vs_state = VDEV_STATE_CANT_OPEN;
3118				vs->vs_aux = VDEV_AUX_SPARED;
3119			}
3120		}
3121	}
3122}
3123
3124/*
3125 * Add l2cache device information to the nvlist, including vdev stats.
3126 */
3127static void
3128spa_add_l2cache(spa_t *spa, nvlist_t *config)
3129{
3130	nvlist_t **l2cache;
3131	uint_t i, j, nl2cache;
3132	nvlist_t *nvroot;
3133	uint64_t guid;
3134	vdev_t *vd;
3135	vdev_stat_t *vs;
3136	uint_t vsc;
3137
3138	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3139
3140	if (spa->spa_l2cache.sav_count == 0)
3141		return;
3142
3143	VERIFY(nvlist_lookup_nvlist(config,
3144	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3145	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3146	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3147	if (nl2cache != 0) {
3148		VERIFY(nvlist_add_nvlist_array(nvroot,
3149		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3150		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3151		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3152
3153		/*
3154		 * Update level 2 cache device stats.
3155		 */
3156
3157		for (i = 0; i < nl2cache; i++) {
3158			VERIFY(nvlist_lookup_uint64(l2cache[i],
3159			    ZPOOL_CONFIG_GUID, &guid) == 0);
3160
3161			vd = NULL;
3162			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3163				if (guid ==
3164				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3165					vd = spa->spa_l2cache.sav_vdevs[j];
3166					break;
3167				}
3168			}
3169			ASSERT(vd != NULL);
3170
3171			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3172			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3173			    == 0);
3174			vdev_get_stats(vd, vs);
3175		}
3176	}
3177}
3178
3179static void
3180spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3181{
3182	nvlist_t *features;
3183	zap_cursor_t zc;
3184	zap_attribute_t za;
3185
3186	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3187	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3188
3189	if (spa->spa_feat_for_read_obj != 0) {
3190		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3191		    spa->spa_feat_for_read_obj);
3192		    zap_cursor_retrieve(&zc, &za) == 0;
3193		    zap_cursor_advance(&zc)) {
3194			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3195			    za.za_num_integers == 1);
3196			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3197			    za.za_first_integer));
3198		}
3199		zap_cursor_fini(&zc);
3200	}
3201
3202	if (spa->spa_feat_for_write_obj != 0) {
3203		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3204		    spa->spa_feat_for_write_obj);
3205		    zap_cursor_retrieve(&zc, &za) == 0;
3206		    zap_cursor_advance(&zc)) {
3207			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3208			    za.za_num_integers == 1);
3209			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3210			    za.za_first_integer));
3211		}
3212		zap_cursor_fini(&zc);
3213	}
3214
3215	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3216	    features) == 0);
3217	nvlist_free(features);
3218}
3219
3220int
3221spa_get_stats(const char *name, nvlist_t **config,
3222    char *altroot, size_t buflen)
3223{
3224	int error;
3225	spa_t *spa;
3226
3227	*config = NULL;
3228	error = spa_open_common(name, &spa, FTAG, NULL, config);
3229
3230	if (spa != NULL) {
3231		/*
3232		 * This still leaves a window of inconsistency where the spares
3233		 * or l2cache devices could change and the config would be
3234		 * self-inconsistent.
3235		 */
3236		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3237
3238		if (*config != NULL) {
3239			uint64_t loadtimes[2];
3240
3241			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3242			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3243			VERIFY(nvlist_add_uint64_array(*config,
3244			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3245
3246			VERIFY(nvlist_add_uint64(*config,
3247			    ZPOOL_CONFIG_ERRCOUNT,
3248			    spa_get_errlog_size(spa)) == 0);
3249
3250			if (spa_suspended(spa))
3251				VERIFY(nvlist_add_uint64(*config,
3252				    ZPOOL_CONFIG_SUSPENDED,
3253				    spa->spa_failmode) == 0);
3254
3255			spa_add_spares(spa, *config);
3256			spa_add_l2cache(spa, *config);
3257			spa_add_feature_stats(spa, *config);
3258		}
3259	}
3260
3261	/*
3262	 * We want to get the alternate root even for faulted pools, so we cheat
3263	 * and call spa_lookup() directly.
3264	 */
3265	if (altroot) {
3266		if (spa == NULL) {
3267			mutex_enter(&spa_namespace_lock);
3268			spa = spa_lookup(name);
3269			if (spa)
3270				spa_altroot(spa, altroot, buflen);
3271			else
3272				altroot[0] = '\0';
3273			spa = NULL;
3274			mutex_exit(&spa_namespace_lock);
3275		} else {
3276			spa_altroot(spa, altroot, buflen);
3277		}
3278	}
3279
3280	if (spa != NULL) {
3281		spa_config_exit(spa, SCL_CONFIG, FTAG);
3282		spa_close(spa, FTAG);
3283	}
3284
3285	return (error);
3286}
3287
3288/*
3289 * Validate that the auxiliary device array is well formed.  We must have an
3290 * array of nvlists, each which describes a valid leaf vdev.  If this is an
3291 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3292 * specified, as long as they are well-formed.
3293 */
3294static int
3295spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3296    spa_aux_vdev_t *sav, const char *config, uint64_t version,
3297    vdev_labeltype_t label)
3298{
3299	nvlist_t **dev;
3300	uint_t i, ndev;
3301	vdev_t *vd;
3302	int error;
3303
3304	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3305
3306	/*
3307	 * It's acceptable to have no devs specified.
3308	 */
3309	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3310		return (0);
3311
3312	if (ndev == 0)
3313		return (SET_ERROR(EINVAL));
3314
3315	/*
3316	 * Make sure the pool is formatted with a version that supports this
3317	 * device type.
3318	 */
3319	if (spa_version(spa) < version)
3320		return (SET_ERROR(ENOTSUP));
3321
3322	/*
3323	 * Set the pending device list so we correctly handle device in-use
3324	 * checking.
3325	 */
3326	sav->sav_pending = dev;
3327	sav->sav_npending = ndev;
3328
3329	for (i = 0; i < ndev; i++) {
3330		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3331		    mode)) != 0)
3332			goto out;
3333
3334		if (!vd->vdev_ops->vdev_op_leaf) {
3335			vdev_free(vd);
3336			error = SET_ERROR(EINVAL);
3337			goto out;
3338		}
3339
3340		/*
3341		 * The L2ARC currently only supports disk devices in
3342		 * kernel context.  For user-level testing, we allow it.
3343		 */
3344#ifdef _KERNEL
3345		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3346		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3347			error = SET_ERROR(ENOTBLK);
3348			vdev_free(vd);
3349			goto out;
3350		}
3351#endif
3352		vd->vdev_top = vd;
3353
3354		if ((error = vdev_open(vd)) == 0 &&
3355		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
3356			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3357			    vd->vdev_guid) == 0);
3358		}
3359
3360		vdev_free(vd);
3361
3362		if (error &&
3363		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3364			goto out;
3365		else
3366			error = 0;
3367	}
3368
3369out:
3370	sav->sav_pending = NULL;
3371	sav->sav_npending = 0;
3372	return (error);
3373}
3374
3375static int
3376spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3377{
3378	int error;
3379
3380	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3381
3382	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3383	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3384	    VDEV_LABEL_SPARE)) != 0) {
3385		return (error);
3386	}
3387
3388	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3389	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3390	    VDEV_LABEL_L2CACHE));
3391}
3392
3393static void
3394spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3395    const char *config)
3396{
3397	int i;
3398
3399	if (sav->sav_config != NULL) {
3400		nvlist_t **olddevs;
3401		uint_t oldndevs;
3402		nvlist_t **newdevs;
3403
3404		/*
3405		 * Generate new dev list by concatentating with the
3406		 * current dev list.
3407		 */
3408		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3409		    &olddevs, &oldndevs) == 0);
3410
3411		newdevs = kmem_alloc(sizeof (void *) *
3412		    (ndevs + oldndevs), KM_SLEEP);
3413		for (i = 0; i < oldndevs; i++)
3414			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3415			    KM_SLEEP) == 0);
3416		for (i = 0; i < ndevs; i++)
3417			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3418			    KM_SLEEP) == 0);
3419
3420		VERIFY(nvlist_remove(sav->sav_config, config,
3421		    DATA_TYPE_NVLIST_ARRAY) == 0);
3422
3423		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3424		    config, newdevs, ndevs + oldndevs) == 0);
3425		for (i = 0; i < oldndevs + ndevs; i++)
3426			nvlist_free(newdevs[i]);
3427		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3428	} else {
3429		/*
3430		 * Generate a new dev list.
3431		 */
3432		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3433		    KM_SLEEP) == 0);
3434		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3435		    devs, ndevs) == 0);
3436	}
3437}
3438
3439/*
3440 * Stop and drop level 2 ARC devices
3441 */
3442void
3443spa_l2cache_drop(spa_t *spa)
3444{
3445	vdev_t *vd;
3446	int i;
3447	spa_aux_vdev_t *sav = &spa->spa_l2cache;
3448
3449	for (i = 0; i < sav->sav_count; i++) {
3450		uint64_t pool;
3451
3452		vd = sav->sav_vdevs[i];
3453		ASSERT(vd != NULL);
3454
3455		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3456		    pool != 0ULL && l2arc_vdev_present(vd))
3457			l2arc_remove_vdev(vd);
3458	}
3459}
3460
3461/*
3462 * Pool Creation
3463 */
3464int
3465spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3466    nvlist_t *zplprops)
3467{
3468	spa_t *spa;
3469	char *altroot = NULL;
3470	vdev_t *rvd;
3471	dsl_pool_t *dp;
3472	dmu_tx_t *tx;
3473	int error = 0;
3474	uint64_t txg = TXG_INITIAL;
3475	nvlist_t **spares, **l2cache;
3476	uint_t nspares, nl2cache;
3477	uint64_t version, obj;
3478	boolean_t has_features;
3479
3480	/*
3481	 * If this pool already exists, return failure.
3482	 */
3483	mutex_enter(&spa_namespace_lock);
3484	if (spa_lookup(pool) != NULL) {
3485		mutex_exit(&spa_namespace_lock);
3486		return (SET_ERROR(EEXIST));
3487	}
3488
3489	/*
3490	 * Allocate a new spa_t structure.
3491	 */
3492	(void) nvlist_lookup_string(props,
3493	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3494	spa = spa_add(pool, NULL, altroot);
3495	spa_activate(spa, spa_mode_global);
3496
3497	if (props && (error = spa_prop_validate(spa, props))) {
3498		spa_deactivate(spa);
3499		spa_remove(spa);
3500		mutex_exit(&spa_namespace_lock);
3501		return (error);
3502	}
3503
3504	has_features = B_FALSE;
3505	for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3506	    elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3507		if (zpool_prop_feature(nvpair_name(elem)))
3508			has_features = B_TRUE;
3509	}
3510
3511	if (has_features || nvlist_lookup_uint64(props,
3512	    zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3513		version = SPA_VERSION;
3514	}
3515	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3516
3517	spa->spa_first_txg = txg;
3518	spa->spa_uberblock.ub_txg = txg - 1;
3519	spa->spa_uberblock.ub_version = version;
3520	spa->spa_ubsync = spa->spa_uberblock;
3521
3522	/*
3523	 * Create "The Godfather" zio to hold all async IOs
3524	 */
3525	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
3526	    KM_SLEEP);
3527	for (int i = 0; i < max_ncpus; i++) {
3528		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
3529		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
3530		    ZIO_FLAG_GODFATHER);
3531	}
3532
3533	/*
3534	 * Create the root vdev.
3535	 */
3536	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3537
3538	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3539
3540	ASSERT(error != 0 || rvd != NULL);
3541	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3542
3543	if (error == 0 && !zfs_allocatable_devs(nvroot))
3544		error = SET_ERROR(EINVAL);
3545
3546	if (error == 0 &&
3547	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3548	    (error = spa_validate_aux(spa, nvroot, txg,
3549	    VDEV_ALLOC_ADD)) == 0) {
3550		for (int c = 0; c < rvd->vdev_children; c++) {
3551			vdev_metaslab_set_size(rvd->vdev_child[c]);
3552			vdev_expand(rvd->vdev_child[c], txg);
3553		}
3554	}
3555
3556	spa_config_exit(spa, SCL_ALL, FTAG);
3557
3558	if (error != 0) {
3559		spa_unload(spa);
3560		spa_deactivate(spa);
3561		spa_remove(spa);
3562		mutex_exit(&spa_namespace_lock);
3563		return (error);
3564	}
3565
3566	/*
3567	 * Get the list of spares, if specified.
3568	 */
3569	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3570	    &spares, &nspares) == 0) {
3571		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3572		    KM_SLEEP) == 0);
3573		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3574		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3575		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3576		spa_load_spares(spa);
3577		spa_config_exit(spa, SCL_ALL, FTAG);
3578		spa->spa_spares.sav_sync = B_TRUE;
3579	}
3580
3581	/*
3582	 * Get the list of level 2 cache devices, if specified.
3583	 */
3584	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3585	    &l2cache, &nl2cache) == 0) {
3586		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3587		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3588		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3589		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3590		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3591		spa_load_l2cache(spa);
3592		spa_config_exit(spa, SCL_ALL, FTAG);
3593		spa->spa_l2cache.sav_sync = B_TRUE;
3594	}
3595
3596	spa->spa_is_initializing = B_TRUE;
3597	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3598	spa->spa_meta_objset = dp->dp_meta_objset;
3599	spa->spa_is_initializing = B_FALSE;
3600
3601	/*
3602	 * Create DDTs (dedup tables).
3603	 */
3604	ddt_create(spa);
3605
3606	spa_update_dspace(spa);
3607
3608	tx = dmu_tx_create_assigned(dp, txg);
3609
3610	/*
3611	 * Create the pool config object.
3612	 */
3613	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3614	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3615	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3616
3617	if (zap_add(spa->spa_meta_objset,
3618	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3619	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3620		cmn_err(CE_PANIC, "failed to add pool config");
3621	}
3622
3623	if (spa_version(spa) >= SPA_VERSION_FEATURES)
3624		spa_feature_create_zap_objects(spa, tx);
3625
3626	if (zap_add(spa->spa_meta_objset,
3627	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3628	    sizeof (uint64_t), 1, &version, tx) != 0) {
3629		cmn_err(CE_PANIC, "failed to add pool version");
3630	}
3631
3632	/* Newly created pools with the right version are always deflated. */
3633	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3634		spa->spa_deflate = TRUE;
3635		if (zap_add(spa->spa_meta_objset,
3636		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3637		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3638			cmn_err(CE_PANIC, "failed to add deflate");
3639		}
3640	}
3641
3642	/*
3643	 * Create the deferred-free bpobj.  Turn off compression
3644	 * because sync-to-convergence takes longer if the blocksize
3645	 * keeps changing.
3646	 */
3647	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3648	dmu_object_set_compress(spa->spa_meta_objset, obj,
3649	    ZIO_COMPRESS_OFF, tx);
3650	if (zap_add(spa->spa_meta_objset,
3651	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3652	    sizeof (uint64_t), 1, &obj, tx) != 0) {
3653		cmn_err(CE_PANIC, "failed to add bpobj");
3654	}
3655	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3656	    spa->spa_meta_objset, obj));
3657
3658	/*
3659	 * Create the pool's history object.
3660	 */
3661	if (version >= SPA_VERSION_ZPOOL_HISTORY)
3662		spa_history_create_obj(spa, tx);
3663
3664	/*
3665	 * Set pool properties.
3666	 */
3667	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3668	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3669	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3670	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3671
3672	if (props != NULL) {
3673		spa_configfile_set(spa, props, B_FALSE);
3674		spa_sync_props(props, tx);
3675	}
3676
3677	dmu_tx_commit(tx);
3678
3679	spa->spa_sync_on = B_TRUE;
3680	txg_sync_start(spa->spa_dsl_pool);
3681
3682	/*
3683	 * We explicitly wait for the first transaction to complete so that our
3684	 * bean counters are appropriately updated.
3685	 */
3686	txg_wait_synced(spa->spa_dsl_pool, txg);
3687
3688	spa_config_sync(spa, B_FALSE, B_TRUE);
3689
3690	spa_history_log_version(spa, "create");
3691
3692	/*
3693	 * Don't count references from objsets that are already closed
3694	 * and are making their way through the eviction process.
3695	 */
3696	spa_evicting_os_wait(spa);
3697	spa->spa_minref = refcount_count(&spa->spa_refcount);
3698
3699	mutex_exit(&spa_namespace_lock);
3700
3701	return (0);
3702}
3703
3704#ifdef _KERNEL
3705/*
3706 * Get the root pool information from the root disk, then import the root pool
3707 * during the system boot up time.
3708 */
3709extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3710
3711static nvlist_t *
3712spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3713{
3714	nvlist_t *config;
3715	nvlist_t *nvtop, *nvroot;
3716	uint64_t pgid;
3717
3718	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3719		return (NULL);
3720
3721	/*
3722	 * Add this top-level vdev to the child array.
3723	 */
3724	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3725	    &nvtop) == 0);
3726	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3727	    &pgid) == 0);
3728	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3729
3730	/*
3731	 * Put this pool's top-level vdevs into a root vdev.
3732	 */
3733	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3734	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3735	    VDEV_TYPE_ROOT) == 0);
3736	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3737	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3738	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3739	    &nvtop, 1) == 0);
3740
3741	/*
3742	 * Replace the existing vdev_tree with the new root vdev in
3743	 * this pool's configuration (remove the old, add the new).
3744	 */
3745	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3746	nvlist_free(nvroot);
3747	return (config);
3748}
3749
3750/*
3751 * Walk the vdev tree and see if we can find a device with "better"
3752 * configuration. A configuration is "better" if the label on that
3753 * device has a more recent txg.
3754 */
3755static void
3756spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3757{
3758	for (int c = 0; c < vd->vdev_children; c++)
3759		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3760
3761	if (vd->vdev_ops->vdev_op_leaf) {
3762		nvlist_t *label;
3763		uint64_t label_txg;
3764
3765		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3766		    &label) != 0)
3767			return;
3768
3769		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3770		    &label_txg) == 0);
3771
3772		/*
3773		 * Do we have a better boot device?
3774		 */
3775		if (label_txg > *txg) {
3776			*txg = label_txg;
3777			*avd = vd;
3778		}
3779		nvlist_free(label);
3780	}
3781}
3782
3783/*
3784 * Import a root pool.
3785 *
3786 * For x86. devpath_list will consist of devid and/or physpath name of
3787 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3788 * The GRUB "findroot" command will return the vdev we should boot.
3789 *
3790 * For Sparc, devpath_list consists the physpath name of the booting device
3791 * no matter the rootpool is a single device pool or a mirrored pool.
3792 * e.g.
3793 *	"/pci@1f,0/ide@d/disk@0,0:a"
3794 */
3795int
3796spa_import_rootpool(char *devpath, char *devid)
3797{
3798	spa_t *spa;
3799	vdev_t *rvd, *bvd, *avd = NULL;
3800	nvlist_t *config, *nvtop;
3801	uint64_t guid, txg;
3802	char *pname;
3803	int error;
3804
3805	/*
3806	 * Read the label from the boot device and generate a configuration.
3807	 */
3808	config = spa_generate_rootconf(devpath, devid, &guid);
3809#if defined(_OBP) && defined(_KERNEL)
3810	if (config == NULL) {
3811		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3812			/* iscsi boot */
3813			get_iscsi_bootpath_phy(devpath);
3814			config = spa_generate_rootconf(devpath, devid, &guid);
3815		}
3816	}
3817#endif
3818	if (config == NULL) {
3819		cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3820		    devpath);
3821		return (SET_ERROR(EIO));
3822	}
3823
3824	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3825	    &pname) == 0);
3826	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3827
3828	mutex_enter(&spa_namespace_lock);
3829	if ((spa = spa_lookup(pname)) != NULL) {
3830		/*
3831		 * Remove the existing root pool from the namespace so that we
3832		 * can replace it with the correct config we just read in.
3833		 */
3834		spa_remove(spa);
3835	}
3836
3837	spa = spa_add(pname, config, NULL);
3838	spa->spa_is_root = B_TRUE;
3839	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3840
3841	/*
3842	 * Build up a vdev tree based on the boot device's label config.
3843	 */
3844	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3845	    &nvtop) == 0);
3846	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3847	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3848	    VDEV_ALLOC_ROOTPOOL);
3849	spa_config_exit(spa, SCL_ALL, FTAG);
3850	if (error) {
3851		mutex_exit(&spa_namespace_lock);
3852		nvlist_free(config);
3853		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3854		    pname);
3855		return (error);
3856	}
3857
3858	/*
3859	 * Get the boot vdev.
3860	 */
3861	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3862		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3863		    (u_longlong_t)guid);
3864		error = SET_ERROR(ENOENT);
3865		goto out;
3866	}
3867
3868	/*
3869	 * Determine if there is a better boot device.
3870	 */
3871	avd = bvd;
3872	spa_alt_rootvdev(rvd, &avd, &txg);
3873	if (avd != bvd) {
3874		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3875		    "try booting from '%s'", avd->vdev_path);
3876		error = SET_ERROR(EINVAL);
3877		goto out;
3878	}
3879
3880	/*
3881	 * If the boot device is part of a spare vdev then ensure that
3882	 * we're booting off the active spare.
3883	 */
3884	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3885	    !bvd->vdev_isspare) {
3886		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3887		    "try booting from '%s'",
3888		    bvd->vdev_parent->
3889		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3890		error = SET_ERROR(EINVAL);
3891		goto out;
3892	}
3893
3894	error = 0;
3895out:
3896	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3897	vdev_free(rvd);
3898	spa_config_exit(spa, SCL_ALL, FTAG);
3899	mutex_exit(&spa_namespace_lock);
3900
3901	nvlist_free(config);
3902	return (error);
3903}
3904
3905#endif
3906
3907/*
3908 * Import a non-root pool into the system.
3909 */
3910int
3911spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3912{
3913	spa_t *spa;
3914	char *altroot = NULL;
3915	spa_load_state_t state = SPA_LOAD_IMPORT;
3916	zpool_rewind_policy_t policy;
3917	uint64_t mode = spa_mode_global;
3918	uint64_t readonly = B_FALSE;
3919	int error;
3920	nvlist_t *nvroot;
3921	nvlist_t **spares, **l2cache;
3922	uint_t nspares, nl2cache;
3923
3924	/*
3925	 * If a pool with this name exists, return failure.
3926	 */
3927	mutex_enter(&spa_namespace_lock);
3928	if (spa_lookup(pool) != NULL) {
3929		mutex_exit(&spa_namespace_lock);
3930		return (SET_ERROR(EEXIST));
3931	}
3932
3933	/*
3934	 * Create and initialize the spa structure.
3935	 */
3936	(void) nvlist_lookup_string(props,
3937	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3938	(void) nvlist_lookup_uint64(props,
3939	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3940	if (readonly)
3941		mode = FREAD;
3942	spa = spa_add(pool, config, altroot);
3943	spa->spa_import_flags = flags;
3944
3945	/*
3946	 * Verbatim import - Take a pool and insert it into the namespace
3947	 * as if it had been loaded at boot.
3948	 */
3949	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3950		if (props != NULL)
3951			spa_configfile_set(spa, props, B_FALSE);
3952
3953		spa_config_sync(spa, B_FALSE, B_TRUE);
3954
3955		mutex_exit(&spa_namespace_lock);
3956		return (0);
3957	}
3958
3959	spa_activate(spa, mode);
3960
3961	/*
3962	 * Don't start async tasks until we know everything is healthy.
3963	 */
3964	spa_async_suspend(spa);
3965
3966	zpool_get_rewind_policy(config, &policy);
3967	if (policy.zrp_request & ZPOOL_DO_REWIND)
3968		state = SPA_LOAD_RECOVER;
3969
3970	/*
3971	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3972	 * because the user-supplied config is actually the one to trust when
3973	 * doing an import.
3974	 */
3975	if (state != SPA_LOAD_RECOVER)
3976		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3977
3978	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3979	    policy.zrp_request);
3980
3981	/*
3982	 * Propagate anything learned while loading the pool and pass it
3983	 * back to caller (i.e. rewind info, missing devices, etc).
3984	 */
3985	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3986	    spa->spa_load_info) == 0);
3987
3988	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3989	/*
3990	 * Toss any existing sparelist, as it doesn't have any validity
3991	 * anymore, and conflicts with spa_has_spare().
3992	 */
3993	if (spa->spa_spares.sav_config) {
3994		nvlist_free(spa->spa_spares.sav_config);
3995		spa->spa_spares.sav_config = NULL;
3996		spa_load_spares(spa);
3997	}
3998	if (spa->spa_l2cache.sav_config) {
3999		nvlist_free(spa->spa_l2cache.sav_config);
4000		spa->spa_l2cache.sav_config = NULL;
4001		spa_load_l2cache(spa);
4002	}
4003
4004	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4005	    &nvroot) == 0);
4006	if (error == 0)
4007		error = spa_validate_aux(spa, nvroot, -1ULL,
4008		    VDEV_ALLOC_SPARE);
4009	if (error == 0)
4010		error = spa_validate_aux(spa, nvroot, -1ULL,
4011		    VDEV_ALLOC_L2CACHE);
4012	spa_config_exit(spa, SCL_ALL, FTAG);
4013
4014	if (props != NULL)
4015		spa_configfile_set(spa, props, B_FALSE);
4016
4017	if (error != 0 || (props && spa_writeable(spa) &&
4018	    (error = spa_prop_set(spa, props)))) {
4019		spa_unload(spa);
4020		spa_deactivate(spa);
4021		spa_remove(spa);
4022		mutex_exit(&spa_namespace_lock);
4023		return (error);
4024	}
4025
4026	spa_async_resume(spa);
4027
4028	/*
4029	 * Override any spares and level 2 cache devices as specified by
4030	 * the user, as these may have correct device names/devids, etc.
4031	 */
4032	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4033	    &spares, &nspares) == 0) {
4034		if (spa->spa_spares.sav_config)
4035			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4036			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4037		else
4038			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4039			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4040		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4041		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4042		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4043		spa_load_spares(spa);
4044		spa_config_exit(spa, SCL_ALL, FTAG);
4045		spa->spa_spares.sav_sync = B_TRUE;
4046	}
4047	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4048	    &l2cache, &nl2cache) == 0) {
4049		if (spa->spa_l2cache.sav_config)
4050			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4051			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4052		else
4053			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4054			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4055		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4056		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4057		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4058		spa_load_l2cache(spa);
4059		spa_config_exit(spa, SCL_ALL, FTAG);
4060		spa->spa_l2cache.sav_sync = B_TRUE;
4061	}
4062
4063	/*
4064	 * Check for any removed devices.
4065	 */
4066	if (spa->spa_autoreplace) {
4067		spa_aux_check_removed(&spa->spa_spares);
4068		spa_aux_check_removed(&spa->spa_l2cache);
4069	}
4070
4071	if (spa_writeable(spa)) {
4072		/*
4073		 * Update the config cache to include the newly-imported pool.
4074		 */
4075		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4076	}
4077
4078	/*
4079	 * It's possible that the pool was expanded while it was exported.
4080	 * We kick off an async task to handle this for us.
4081	 */
4082	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4083
4084	mutex_exit(&spa_namespace_lock);
4085	spa_history_log_version(spa, "import");
4086
4087	return (0);
4088}
4089
4090nvlist_t *
4091spa_tryimport(nvlist_t *tryconfig)
4092{
4093	nvlist_t *config = NULL;
4094	char *poolname;
4095	spa_t *spa;
4096	uint64_t state;
4097	int error;
4098
4099	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4100		return (NULL);
4101
4102	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4103		return (NULL);
4104
4105	/*
4106	 * Create and initialize the spa structure.
4107	 */
4108	mutex_enter(&spa_namespace_lock);
4109	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4110	spa_activate(spa, FREAD);
4111
4112	/*
4113	 * Pass off the heavy lifting to spa_load().
4114	 * Pass TRUE for mosconfig because the user-supplied config
4115	 * is actually the one to trust when doing an import.
4116	 */
4117	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4118
4119	/*
4120	 * If 'tryconfig' was at least parsable, return the current config.
4121	 */
4122	if (spa->spa_root_vdev != NULL) {
4123		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4124		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4125		    poolname) == 0);
4126		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4127		    state) == 0);
4128		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4129		    spa->spa_uberblock.ub_timestamp) == 0);
4130		VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4131		    spa->spa_load_info) == 0);
4132
4133		/*
4134		 * If the bootfs property exists on this pool then we
4135		 * copy it out so that external consumers can tell which
4136		 * pools are bootable.
4137		 */
4138		if ((!error || error == EEXIST) && spa->spa_bootfs) {
4139			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4140
4141			/*
4142			 * We have to play games with the name since the
4143			 * pool was opened as TRYIMPORT_NAME.
4144			 */
4145			if (dsl_dsobj_to_dsname(spa_name(spa),
4146			    spa->spa_bootfs, tmpname) == 0) {
4147				char *cp;
4148				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4149
4150				cp = strchr(tmpname, '/');
4151				if (cp == NULL) {
4152					(void) strlcpy(dsname, tmpname,
4153					    MAXPATHLEN);
4154				} else {
4155					(void) snprintf(dsname, MAXPATHLEN,
4156					    "%s/%s", poolname, ++cp);
4157				}
4158				VERIFY(nvlist_add_string(config,
4159				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4160				kmem_free(dsname, MAXPATHLEN);
4161			}
4162			kmem_free(tmpname, MAXPATHLEN);
4163		}
4164
4165		/*
4166		 * Add the list of hot spares and level 2 cache devices.
4167		 */
4168		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4169		spa_add_spares(spa, config);
4170		spa_add_l2cache(spa, config);
4171		spa_config_exit(spa, SCL_CONFIG, FTAG);
4172	}
4173
4174	spa_unload(spa);
4175	spa_deactivate(spa);
4176	spa_remove(spa);
4177	mutex_exit(&spa_namespace_lock);
4178
4179	return (config);
4180}
4181
4182/*
4183 * Pool export/destroy
4184 *
4185 * The act of destroying or exporting a pool is very simple.  We make sure there
4186 * is no more pending I/O and any references to the pool are gone.  Then, we
4187 * update the pool state and sync all the labels to disk, removing the
4188 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4189 * we don't sync the labels or remove the configuration cache.
4190 */
4191static int
4192spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4193    boolean_t force, boolean_t hardforce)
4194{
4195	spa_t *spa;
4196
4197	if (oldconfig)
4198		*oldconfig = NULL;
4199
4200	if (!(spa_mode_global & FWRITE))
4201		return (SET_ERROR(EROFS));
4202
4203	mutex_enter(&spa_namespace_lock);
4204	if ((spa = spa_lookup(pool)) == NULL) {
4205		mutex_exit(&spa_namespace_lock);
4206		return (SET_ERROR(ENOENT));
4207	}
4208
4209	/*
4210	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4211	 * reacquire the namespace lock, and see if we can export.
4212	 */
4213	spa_open_ref(spa, FTAG);
4214	mutex_exit(&spa_namespace_lock);
4215	spa_async_suspend(spa);
4216	mutex_enter(&spa_namespace_lock);
4217	spa_close(spa, FTAG);
4218
4219	/*
4220	 * The pool will be in core if it's openable,
4221	 * in which case we can modify its state.
4222	 */
4223	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4224		/*
4225		 * Objsets may be open only because they're dirty, so we
4226		 * have to force it to sync before checking spa_refcnt.
4227		 */
4228		txg_wait_synced(spa->spa_dsl_pool, 0);
4229		spa_evicting_os_wait(spa);
4230
4231		/*
4232		 * A pool cannot be exported or destroyed if there are active
4233		 * references.  If we are resetting a pool, allow references by
4234		 * fault injection handlers.
4235		 */
4236		if (!spa_refcount_zero(spa) ||
4237		    (spa->spa_inject_ref != 0 &&
4238		    new_state != POOL_STATE_UNINITIALIZED)) {
4239			spa_async_resume(spa);
4240			mutex_exit(&spa_namespace_lock);
4241			return (SET_ERROR(EBUSY));
4242		}
4243
4244		/*
4245		 * A pool cannot be exported if it has an active shared spare.
4246		 * This is to prevent other pools stealing the active spare
4247		 * from an exported pool. At user's own will, such pool can
4248		 * be forcedly exported.
4249		 */
4250		if (!force && new_state == POOL_STATE_EXPORTED &&
4251		    spa_has_active_shared_spare(spa)) {
4252			spa_async_resume(spa);
4253			mutex_exit(&spa_namespace_lock);
4254			return (SET_ERROR(EXDEV));
4255		}
4256
4257		/*
4258		 * We want this to be reflected on every label,
4259		 * so mark them all dirty.  spa_unload() will do the
4260		 * final sync that pushes these changes out.
4261		 */
4262		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4263			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4264			spa->spa_state = new_state;
4265			spa->spa_final_txg = spa_last_synced_txg(spa) +
4266			    TXG_DEFER_SIZE + 1;
4267			vdev_config_dirty(spa->spa_root_vdev);
4268			spa_config_exit(spa, SCL_ALL, FTAG);
4269		}
4270	}
4271
4272	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4273
4274	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4275		spa_unload(spa);
4276		spa_deactivate(spa);
4277	}
4278
4279	if (oldconfig && spa->spa_config)
4280		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4281
4282	if (new_state != POOL_STATE_UNINITIALIZED) {
4283		if (!hardforce)
4284			spa_config_sync(spa, B_TRUE, B_TRUE);
4285		spa_remove(spa);
4286	}
4287	mutex_exit(&spa_namespace_lock);
4288
4289	return (0);
4290}
4291
4292/*
4293 * Destroy a storage pool.
4294 */
4295int
4296spa_destroy(char *pool)
4297{
4298	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4299	    B_FALSE, B_FALSE));
4300}
4301
4302/*
4303 * Export a storage pool.
4304 */
4305int
4306spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4307    boolean_t hardforce)
4308{
4309	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4310	    force, hardforce));
4311}
4312
4313/*
4314 * Similar to spa_export(), this unloads the spa_t without actually removing it
4315 * from the namespace in any way.
4316 */
4317int
4318spa_reset(char *pool)
4319{
4320	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4321	    B_FALSE, B_FALSE));
4322}
4323
4324/*
4325 * ==========================================================================
4326 * Device manipulation
4327 * ==========================================================================
4328 */
4329
4330/*
4331 * Add a device to a storage pool.
4332 */
4333int
4334spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4335{
4336	uint64_t txg, id;
4337	int error;
4338	vdev_t *rvd = spa->spa_root_vdev;
4339	vdev_t *vd, *tvd;
4340	nvlist_t **spares, **l2cache;
4341	uint_t nspares, nl2cache;
4342
4343	ASSERT(spa_writeable(spa));
4344
4345	txg = spa_vdev_enter(spa);
4346
4347	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4348	    VDEV_ALLOC_ADD)) != 0)
4349		return (spa_vdev_exit(spa, NULL, txg, error));
4350
4351	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
4352
4353	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4354	    &nspares) != 0)
4355		nspares = 0;
4356
4357	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4358	    &nl2cache) != 0)
4359		nl2cache = 0;
4360
4361	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4362		return (spa_vdev_exit(spa, vd, txg, EINVAL));
4363
4364	if (vd->vdev_children != 0 &&
4365	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
4366		return (spa_vdev_exit(spa, vd, txg, error));
4367
4368	/*
4369	 * We must validate the spares and l2cache devices after checking the
4370	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
4371	 */
4372	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4373		return (spa_vdev_exit(spa, vd, txg, error));
4374
4375	/*
4376	 * Transfer each new top-level vdev from vd to rvd.
4377	 */
4378	for (int c = 0; c < vd->vdev_children; c++) {
4379
4380		/*
4381		 * Set the vdev id to the first hole, if one exists.
4382		 */
4383		for (id = 0; id < rvd->vdev_children; id++) {
4384			if (rvd->vdev_child[id]->vdev_ishole) {
4385				vdev_free(rvd->vdev_child[id]);
4386				break;
4387			}
4388		}
4389		tvd = vd->vdev_child[c];
4390		vdev_remove_child(vd, tvd);
4391		tvd->vdev_id = id;
4392		vdev_add_child(rvd, tvd);
4393		vdev_config_dirty(tvd);
4394	}
4395
4396	if (nspares != 0) {
4397		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4398		    ZPOOL_CONFIG_SPARES);
4399		spa_load_spares(spa);
4400		spa->spa_spares.sav_sync = B_TRUE;
4401	}
4402
4403	if (nl2cache != 0) {
4404		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4405		    ZPOOL_CONFIG_L2CACHE);
4406		spa_load_l2cache(spa);
4407		spa->spa_l2cache.sav_sync = B_TRUE;
4408	}
4409
4410	/*
4411	 * We have to be careful when adding new vdevs to an existing pool.
4412	 * If other threads start allocating from these vdevs before we
4413	 * sync the config cache, and we lose power, then upon reboot we may
4414	 * fail to open the pool because there are DVAs that the config cache
4415	 * can't translate.  Therefore, we first add the vdevs without
4416	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4417	 * and then let spa_config_update() initialize the new metaslabs.
4418	 *
4419	 * spa_load() checks for added-but-not-initialized vdevs, so that
4420	 * if we lose power at any point in this sequence, the remaining
4421	 * steps will be completed the next time we load the pool.
4422	 */
4423	(void) spa_vdev_exit(spa, vd, txg, 0);
4424
4425	mutex_enter(&spa_namespace_lock);
4426	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4427	mutex_exit(&spa_namespace_lock);
4428
4429	return (0);
4430}
4431
4432/*
4433 * Attach a device to a mirror.  The arguments are the path to any device
4434 * in the mirror, and the nvroot for the new device.  If the path specifies
4435 * a device that is not mirrored, we automatically insert the mirror vdev.
4436 *
4437 * If 'replacing' is specified, the new device is intended to replace the
4438 * existing device; in this case the two devices are made into their own
4439 * mirror using the 'replacing' vdev, which is functionally identical to
4440 * the mirror vdev (it actually reuses all the same ops) but has a few
4441 * extra rules: you can't attach to it after it's been created, and upon
4442 * completion of resilvering, the first disk (the one being replaced)
4443 * is automatically detached.
4444 */
4445int
4446spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4447{
4448	uint64_t txg, dtl_max_txg;
4449	vdev_t *rvd = spa->spa_root_vdev;
4450	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4451	vdev_ops_t *pvops;
4452	char *oldvdpath, *newvdpath;
4453	int newvd_isspare;
4454	int error;
4455
4456	ASSERT(spa_writeable(spa));
4457
4458	txg = spa_vdev_enter(spa);
4459
4460	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4461
4462	if (oldvd == NULL)
4463		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4464
4465	if (!oldvd->vdev_ops->vdev_op_leaf)
4466		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4467
4468	pvd = oldvd->vdev_parent;
4469
4470	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4471	    VDEV_ALLOC_ATTACH)) != 0)
4472		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4473
4474	if (newrootvd->vdev_children != 1)
4475		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4476
4477	newvd = newrootvd->vdev_child[0];
4478
4479	if (!newvd->vdev_ops->vdev_op_leaf)
4480		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4481
4482	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4483		return (spa_vdev_exit(spa, newrootvd, txg, error));
4484
4485	/*
4486	 * Spares can't replace logs
4487	 */
4488	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4489		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4490
4491	if (!replacing) {
4492		/*
4493		 * For attach, the only allowable parent is a mirror or the root
4494		 * vdev.
4495		 */
4496		if (pvd->vdev_ops != &vdev_mirror_ops &&
4497		    pvd->vdev_ops != &vdev_root_ops)
4498			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4499
4500		pvops = &vdev_mirror_ops;
4501	} else {
4502		/*
4503		 * Active hot spares can only be replaced by inactive hot
4504		 * spares.
4505		 */
4506		if (pvd->vdev_ops == &vdev_spare_ops &&
4507		    oldvd->vdev_isspare &&
4508		    !spa_has_spare(spa, newvd->vdev_guid))
4509			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4510
4511		/*
4512		 * If the source is a hot spare, and the parent isn't already a
4513		 * spare, then we want to create a new hot spare.  Otherwise, we
4514		 * want to create a replacing vdev.  The user is not allowed to
4515		 * attach to a spared vdev child unless the 'isspare' state is
4516		 * the same (spare replaces spare, non-spare replaces
4517		 * non-spare).
4518		 */
4519		if (pvd->vdev_ops == &vdev_replacing_ops &&
4520		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4521			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4522		} else if (pvd->vdev_ops == &vdev_spare_ops &&
4523		    newvd->vdev_isspare != oldvd->vdev_isspare) {
4524			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4525		}
4526
4527		if (newvd->vdev_isspare)
4528			pvops = &vdev_spare_ops;
4529		else
4530			pvops = &vdev_replacing_ops;
4531	}
4532
4533	/*
4534	 * Make sure the new device is big enough.
4535	 */
4536	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4537		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4538
4539	/*
4540	 * The new device cannot have a higher alignment requirement
4541	 * than the top-level vdev.
4542	 */
4543	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4544		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4545
4546	/*
4547	 * If this is an in-place replacement, update oldvd's path and devid
4548	 * to make it distinguishable from newvd, and unopenable from now on.
4549	 */
4550	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4551		spa_strfree(oldvd->vdev_path);
4552		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4553		    KM_SLEEP);
4554		(void) sprintf(oldvd->vdev_path, "%s/%s",
4555		    newvd->vdev_path, "old");
4556		if (oldvd->vdev_devid != NULL) {
4557			spa_strfree(oldvd->vdev_devid);
4558			oldvd->vdev_devid = NULL;
4559		}
4560	}
4561
4562	/* mark the device being resilvered */
4563	newvd->vdev_resilver_txg = txg;
4564
4565	/*
4566	 * If the parent is not a mirror, or if we're replacing, insert the new
4567	 * mirror/replacing/spare vdev above oldvd.
4568	 */
4569	if (pvd->vdev_ops != pvops)
4570		pvd = vdev_add_parent(oldvd, pvops);
4571
4572	ASSERT(pvd->vdev_top->vdev_parent == rvd);
4573	ASSERT(pvd->vdev_ops == pvops);
4574	ASSERT(oldvd->vdev_parent == pvd);
4575
4576	/*
4577	 * Extract the new device from its root and add it to pvd.
4578	 */
4579	vdev_remove_child(newrootvd, newvd);
4580	newvd->vdev_id = pvd->vdev_children;
4581	newvd->vdev_crtxg = oldvd->vdev_crtxg;
4582	vdev_add_child(pvd, newvd);
4583
4584	tvd = newvd->vdev_top;
4585	ASSERT(pvd->vdev_top == tvd);
4586	ASSERT(tvd->vdev_parent == rvd);
4587
4588	vdev_config_dirty(tvd);
4589
4590	/*
4591	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4592	 * for any dmu_sync-ed blocks.  It will propagate upward when
4593	 * spa_vdev_exit() calls vdev_dtl_reassess().
4594	 */
4595	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4596
4597	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4598	    dtl_max_txg - TXG_INITIAL);
4599
4600	if (newvd->vdev_isspare) {
4601		spa_spare_activate(newvd);
4602		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4603	}
4604
4605	oldvdpath = spa_strdup(oldvd->vdev_path);
4606	newvdpath = spa_strdup(newvd->vdev_path);
4607	newvd_isspare = newvd->vdev_isspare;
4608
4609	/*
4610	 * Mark newvd's DTL dirty in this txg.
4611	 */
4612	vdev_dirty(tvd, VDD_DTL, newvd, txg);
4613
4614	/*
4615	 * Schedule the resilver to restart in the future. We do this to
4616	 * ensure that dmu_sync-ed blocks have been stitched into the
4617	 * respective datasets.
4618	 */
4619	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4620
4621	/*
4622	 * Commit the config
4623	 */
4624	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4625
4626	spa_history_log_internal(spa, "vdev attach", NULL,
4627	    "%s vdev=%s %s vdev=%s",
4628	    replacing && newvd_isspare ? "spare in" :
4629	    replacing ? "replace" : "attach", newvdpath,
4630	    replacing ? "for" : "to", oldvdpath);
4631
4632	spa_strfree(oldvdpath);
4633	spa_strfree(newvdpath);
4634
4635	if (spa->spa_bootfs)
4636		spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4637
4638	return (0);
4639}
4640
4641/*
4642 * Detach a device from a mirror or replacing vdev.
4643 *
4644 * If 'replace_done' is specified, only detach if the parent
4645 * is a replacing vdev.
4646 */
4647int
4648spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4649{
4650	uint64_t txg;
4651	int error;
4652	vdev_t *rvd = spa->spa_root_vdev;
4653	vdev_t *vd, *pvd, *cvd, *tvd;
4654	boolean_t unspare = B_FALSE;
4655	uint64_t unspare_guid = 0;
4656	char *vdpath;
4657
4658	ASSERT(spa_writeable(spa));
4659
4660	txg = spa_vdev_enter(spa);
4661
4662	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4663
4664	if (vd == NULL)
4665		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4666
4667	if (!vd->vdev_ops->vdev_op_leaf)
4668		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4669
4670	pvd = vd->vdev_parent;
4671
4672	/*
4673	 * If the parent/child relationship is not as expected, don't do it.
4674	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4675	 * vdev that's replacing B with C.  The user's intent in replacing
4676	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4677	 * the replace by detaching C, the expected behavior is to end up
4678	 * M(A,B).  But suppose that right after deciding to detach C,
4679	 * the replacement of B completes.  We would have M(A,C), and then
4680	 * ask to detach C, which would leave us with just A -- not what
4681	 * the user wanted.  To prevent this, we make sure that the
4682	 * parent/child relationship hasn't changed -- in this example,
4683	 * that C's parent is still the replacing vdev R.
4684	 */
4685	if (pvd->vdev_guid != pguid && pguid != 0)
4686		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4687
4688	/*
4689	 * Only 'replacing' or 'spare' vdevs can be replaced.
4690	 */
4691	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4692	    pvd->vdev_ops != &vdev_spare_ops)
4693		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4694
4695	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4696	    spa_version(spa) >= SPA_VERSION_SPARES);
4697
4698	/*
4699	 * Only mirror, replacing, and spare vdevs support detach.
4700	 */
4701	if (pvd->vdev_ops != &vdev_replacing_ops &&
4702	    pvd->vdev_ops != &vdev_mirror_ops &&
4703	    pvd->vdev_ops != &vdev_spare_ops)
4704		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4705
4706	/*
4707	 * If this device has the only valid copy of some data,
4708	 * we cannot safely detach it.
4709	 */
4710	if (vdev_dtl_required(vd))
4711		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4712
4713	ASSERT(pvd->vdev_children >= 2);
4714
4715	/*
4716	 * If we are detaching the second disk from a replacing vdev, then
4717	 * check to see if we changed the original vdev's path to have "/old"
4718	 * at the end in spa_vdev_attach().  If so, undo that change now.
4719	 */
4720	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4721	    vd->vdev_path != NULL) {
4722		size_t len = strlen(vd->vdev_path);
4723
4724		for (int c = 0; c < pvd->vdev_children; c++) {
4725			cvd = pvd->vdev_child[c];
4726
4727			if (cvd == vd || cvd->vdev_path == NULL)
4728				continue;
4729
4730			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4731			    strcmp(cvd->vdev_path + len, "/old") == 0) {
4732				spa_strfree(cvd->vdev_path);
4733				cvd->vdev_path = spa_strdup(vd->vdev_path);
4734				break;
4735			}
4736		}
4737	}
4738
4739	/*
4740	 * If we are detaching the original disk from a spare, then it implies
4741	 * that the spare should become a real disk, and be removed from the
4742	 * active spare list for the pool.
4743	 */
4744	if (pvd->vdev_ops == &vdev_spare_ops &&
4745	    vd->vdev_id == 0 &&
4746	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4747		unspare = B_TRUE;
4748
4749	/*
4750	 * Erase the disk labels so the disk can be used for other things.
4751	 * This must be done after all other error cases are handled,
4752	 * but before we disembowel vd (so we can still do I/O to it).
4753	 * But if we can't do it, don't treat the error as fatal --
4754	 * it may be that the unwritability of the disk is the reason
4755	 * it's being detached!
4756	 */
4757	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4758
4759	/*
4760	 * Remove vd from its parent and compact the parent's children.
4761	 */
4762	vdev_remove_child(pvd, vd);
4763	vdev_compact_children(pvd);
4764
4765	/*
4766	 * Remember one of the remaining children so we can get tvd below.
4767	 */
4768	cvd = pvd->vdev_child[pvd->vdev_children - 1];
4769
4770	/*
4771	 * If we need to remove the remaining child from the list of hot spares,
4772	 * do it now, marking the vdev as no longer a spare in the process.
4773	 * We must do this before vdev_remove_parent(), because that can
4774	 * change the GUID if it creates a new toplevel GUID.  For a similar
4775	 * reason, we must remove the spare now, in the same txg as the detach;
4776	 * otherwise someone could attach a new sibling, change the GUID, and
4777	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4778	 */
4779	if (unspare) {
4780		ASSERT(cvd->vdev_isspare);
4781		spa_spare_remove(cvd);
4782		unspare_guid = cvd->vdev_guid;
4783		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4784		cvd->vdev_unspare = B_TRUE;
4785	}
4786
4787	/*
4788	 * If the parent mirror/replacing vdev only has one child,
4789	 * the parent is no longer needed.  Remove it from the tree.
4790	 */
4791	if (pvd->vdev_children == 1) {
4792		if (pvd->vdev_ops == &vdev_spare_ops)
4793			cvd->vdev_unspare = B_FALSE;
4794		vdev_remove_parent(cvd);
4795	}
4796
4797
4798	/*
4799	 * We don't set tvd until now because the parent we just removed
4800	 * may have been the previous top-level vdev.
4801	 */
4802	tvd = cvd->vdev_top;
4803	ASSERT(tvd->vdev_parent == rvd);
4804
4805	/*
4806	 * Reevaluate the parent vdev state.
4807	 */
4808	vdev_propagate_state(cvd);
4809
4810	/*
4811	 * If the 'autoexpand' property is set on the pool then automatically
4812	 * try to expand the size of the pool. For example if the device we
4813	 * just detached was smaller than the others, it may be possible to
4814	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4815	 * first so that we can obtain the updated sizes of the leaf vdevs.
4816	 */
4817	if (spa->spa_autoexpand) {
4818		vdev_reopen(tvd);
4819		vdev_expand(tvd, txg);
4820	}
4821
4822	vdev_config_dirty(tvd);
4823
4824	/*
4825	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4826	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4827	 * But first make sure we're not on any *other* txg's DTL list, to
4828	 * prevent vd from being accessed after it's freed.
4829	 */
4830	vdpath = spa_strdup(vd->vdev_path);
4831	for (int t = 0; t < TXG_SIZE; t++)
4832		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4833	vd->vdev_detached = B_TRUE;
4834	vdev_dirty(tvd, VDD_DTL, vd, txg);
4835
4836	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4837
4838	/* hang on to the spa before we release the lock */
4839	spa_open_ref(spa, FTAG);
4840
4841	error = spa_vdev_exit(spa, vd, txg, 0);
4842
4843	spa_history_log_internal(spa, "detach", NULL,
4844	    "vdev=%s", vdpath);
4845	spa_strfree(vdpath);
4846
4847	/*
4848	 * If this was the removal of the original device in a hot spare vdev,
4849	 * then we want to go through and remove the device from the hot spare
4850	 * list of every other pool.
4851	 */
4852	if (unspare) {
4853		spa_t *altspa = NULL;
4854
4855		mutex_enter(&spa_namespace_lock);
4856		while ((altspa = spa_next(altspa)) != NULL) {
4857			if (altspa->spa_state != POOL_STATE_ACTIVE ||
4858			    altspa == spa)
4859				continue;
4860
4861			spa_open_ref(altspa, FTAG);
4862			mutex_exit(&spa_namespace_lock);
4863			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4864			mutex_enter(&spa_namespace_lock);
4865			spa_close(altspa, FTAG);
4866		}
4867		mutex_exit(&spa_namespace_lock);
4868
4869		/* search the rest of the vdevs for spares to remove */
4870		spa_vdev_resilver_done(spa);
4871	}
4872
4873	/* all done with the spa; OK to release */
4874	mutex_enter(&spa_namespace_lock);
4875	spa_close(spa, FTAG);
4876	mutex_exit(&spa_namespace_lock);
4877
4878	return (error);
4879}
4880
4881/*
4882 * Split a set of devices from their mirrors, and create a new pool from them.
4883 */
4884int
4885spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4886    nvlist_t *props, boolean_t exp)
4887{
4888	int error = 0;
4889	uint64_t txg, *glist;
4890	spa_t *newspa;
4891	uint_t c, children, lastlog;
4892	nvlist_t **child, *nvl, *tmp;
4893	dmu_tx_t *tx;
4894	char *altroot = NULL;
4895	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4896	boolean_t activate_slog;
4897
4898	ASSERT(spa_writeable(spa));
4899
4900	txg = spa_vdev_enter(spa);
4901
4902	/* clear the log and flush everything up to now */
4903	activate_slog = spa_passivate_log(spa);
4904	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4905	error = spa_offline_log(spa);
4906	txg = spa_vdev_config_enter(spa);
4907
4908	if (activate_slog)
4909		spa_activate_log(spa);
4910
4911	if (error != 0)
4912		return (spa_vdev_exit(spa, NULL, txg, error));
4913
4914	/* check new spa name before going any further */
4915	if (spa_lookup(newname) != NULL)
4916		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4917
4918	/*
4919	 * scan through all the children to ensure they're all mirrors
4920	 */
4921	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4922	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4923	    &children) != 0)
4924		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4925
4926	/* first, check to ensure we've got the right child count */
4927	rvd = spa->spa_root_vdev;
4928	lastlog = 0;
4929	for (c = 0; c < rvd->vdev_children; c++) {
4930		vdev_t *vd = rvd->vdev_child[c];
4931
4932		/* don't count the holes & logs as children */
4933		if (vd->vdev_islog || vd->vdev_ishole) {
4934			if (lastlog == 0)
4935				lastlog = c;
4936			continue;
4937		}
4938
4939		lastlog = 0;
4940	}
4941	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4942		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4943
4944	/* next, ensure no spare or cache devices are part of the split */
4945	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4946	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4947		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4948
4949	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4950	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4951
4952	/* then, loop over each vdev and validate it */
4953	for (c = 0; c < children; c++) {
4954		uint64_t is_hole = 0;
4955
4956		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4957		    &is_hole);
4958
4959		if (is_hole != 0) {
4960			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4961			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4962				continue;
4963			} else {
4964				error = SET_ERROR(EINVAL);
4965				break;
4966			}
4967		}
4968
4969		/* which disk is going to be split? */
4970		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4971		    &glist[c]) != 0) {
4972			error = SET_ERROR(EINVAL);
4973			break;
4974		}
4975
4976		/* look it up in the spa */
4977		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4978		if (vml[c] == NULL) {
4979			error = SET_ERROR(ENODEV);
4980			break;
4981		}
4982
4983		/* make sure there's nothing stopping the split */
4984		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4985		    vml[c]->vdev_islog ||
4986		    vml[c]->vdev_ishole ||
4987		    vml[c]->vdev_isspare ||
4988		    vml[c]->vdev_isl2cache ||
4989		    !vdev_writeable(vml[c]) ||
4990		    vml[c]->vdev_children != 0 ||
4991		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4992		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4993			error = SET_ERROR(EINVAL);
4994			break;
4995		}
4996
4997		if (vdev_dtl_required(vml[c])) {
4998			error = SET_ERROR(EBUSY);
4999			break;
5000		}
5001
5002		/* we need certain info from the top level */
5003		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5004		    vml[c]->vdev_top->vdev_ms_array) == 0);
5005		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5006		    vml[c]->vdev_top->vdev_ms_shift) == 0);
5007		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5008		    vml[c]->vdev_top->vdev_asize) == 0);
5009		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5010		    vml[c]->vdev_top->vdev_ashift) == 0);
5011	}
5012
5013	if (error != 0) {
5014		kmem_free(vml, children * sizeof (vdev_t *));
5015		kmem_free(glist, children * sizeof (uint64_t));
5016		return (spa_vdev_exit(spa, NULL, txg, error));
5017	}
5018
5019	/* stop writers from using the disks */
5020	for (c = 0; c < children; c++) {
5021		if (vml[c] != NULL)
5022			vml[c]->vdev_offline = B_TRUE;
5023	}
5024	vdev_reopen(spa->spa_root_vdev);
5025
5026	/*
5027	 * Temporarily record the splitting vdevs in the spa config.  This
5028	 * will disappear once the config is regenerated.
5029	 */
5030	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5031	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5032	    glist, children) == 0);
5033	kmem_free(glist, children * sizeof (uint64_t));
5034
5035	mutex_enter(&spa->spa_props_lock);
5036	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5037	    nvl) == 0);
5038	mutex_exit(&spa->spa_props_lock);
5039	spa->spa_config_splitting = nvl;
5040	vdev_config_dirty(spa->spa_root_vdev);
5041
5042	/* configure and create the new pool */
5043	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5044	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5045	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5046	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5047	    spa_version(spa)) == 0);
5048	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5049	    spa->spa_config_txg) == 0);
5050	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5051	    spa_generate_guid(NULL)) == 0);
5052	(void) nvlist_lookup_string(props,
5053	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5054
5055	/* add the new pool to the namespace */
5056	newspa = spa_add(newname, config, altroot);
5057	newspa->spa_config_txg = spa->spa_config_txg;
5058	spa_set_log_state(newspa, SPA_LOG_CLEAR);
5059
5060	/* release the spa config lock, retaining the namespace lock */
5061	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5062
5063	if (zio_injection_enabled)
5064		zio_handle_panic_injection(spa, FTAG, 1);
5065
5066	spa_activate(newspa, spa_mode_global);
5067	spa_async_suspend(newspa);
5068
5069	/* create the new pool from the disks of the original pool */
5070	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5071	if (error)
5072		goto out;
5073
5074	/* if that worked, generate a real config for the new pool */
5075	if (newspa->spa_root_vdev != NULL) {
5076		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5077		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
5078		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5079		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5080		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5081		    B_TRUE));
5082	}
5083
5084	/* set the props */
5085	if (props != NULL) {
5086		spa_configfile_set(newspa, props, B_FALSE);
5087		error = spa_prop_set(newspa, props);
5088		if (error)
5089			goto out;
5090	}
5091
5092	/* flush everything */
5093	txg = spa_vdev_config_enter(newspa);
5094	vdev_config_dirty(newspa->spa_root_vdev);
5095	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5096
5097	if (zio_injection_enabled)
5098		zio_handle_panic_injection(spa, FTAG, 2);
5099
5100	spa_async_resume(newspa);
5101
5102	/* finally, update the original pool's config */
5103	txg = spa_vdev_config_enter(spa);
5104	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5105	error = dmu_tx_assign(tx, TXG_WAIT);
5106	if (error != 0)
5107		dmu_tx_abort(tx);
5108	for (c = 0; c < children; c++) {
5109		if (vml[c] != NULL) {
5110			vdev_split(vml[c]);
5111			if (error == 0)
5112				spa_history_log_internal(spa, "detach", tx,
5113				    "vdev=%s", vml[c]->vdev_path);
5114			vdev_free(vml[c]);
5115		}
5116	}
5117	vdev_config_dirty(spa->spa_root_vdev);
5118	spa->spa_config_splitting = NULL;
5119	nvlist_free(nvl);
5120	if (error == 0)
5121		dmu_tx_commit(tx);
5122	(void) spa_vdev_exit(spa, NULL, txg, 0);
5123
5124	if (zio_injection_enabled)
5125		zio_handle_panic_injection(spa, FTAG, 3);
5126
5127	/* split is complete; log a history record */
5128	spa_history_log_internal(newspa, "split", NULL,
5129	    "from pool %s", spa_name(spa));
5130
5131	kmem_free(vml, children * sizeof (vdev_t *));
5132
5133	/* if we're not going to mount the filesystems in userland, export */
5134	if (exp)
5135		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5136		    B_FALSE, B_FALSE);
5137
5138	return (error);
5139
5140out:
5141	spa_unload(newspa);
5142	spa_deactivate(newspa);
5143	spa_remove(newspa);
5144
5145	txg = spa_vdev_config_enter(spa);
5146
5147	/* re-online all offlined disks */
5148	for (c = 0; c < children; c++) {
5149		if (vml[c] != NULL)
5150			vml[c]->vdev_offline = B_FALSE;
5151	}
5152	vdev_reopen(spa->spa_root_vdev);
5153
5154	nvlist_free(spa->spa_config_splitting);
5155	spa->spa_config_splitting = NULL;
5156	(void) spa_vdev_exit(spa, NULL, txg, error);
5157
5158	kmem_free(vml, children * sizeof (vdev_t *));
5159	return (error);
5160}
5161
5162static nvlist_t *
5163spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5164{
5165	for (int i = 0; i < count; i++) {
5166		uint64_t guid;
5167
5168		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5169		    &guid) == 0);
5170
5171		if (guid == target_guid)
5172			return (nvpp[i]);
5173	}
5174
5175	return (NULL);
5176}
5177
5178static void
5179spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5180	nvlist_t *dev_to_remove)
5181{
5182	nvlist_t **newdev = NULL;
5183
5184	if (count > 1)
5185		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5186
5187	for (int i = 0, j = 0; i < count; i++) {
5188		if (dev[i] == dev_to_remove)
5189			continue;
5190		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5191	}
5192
5193	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5194	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5195
5196	for (int i = 0; i < count - 1; i++)
5197		nvlist_free(newdev[i]);
5198
5199	if (count > 1)
5200		kmem_free(newdev, (count - 1) * sizeof (void *));
5201}
5202
5203/*
5204 * Evacuate the device.
5205 */
5206static int
5207spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5208{
5209	uint64_t txg;
5210	int error = 0;
5211
5212	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5213	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5214	ASSERT(vd == vd->vdev_top);
5215
5216	/*
5217	 * Evacuate the device.  We don't hold the config lock as writer
5218	 * since we need to do I/O but we do keep the
5219	 * spa_namespace_lock held.  Once this completes the device
5220	 * should no longer have any blocks allocated on it.
5221	 */
5222	if (vd->vdev_islog) {
5223		if (vd->vdev_stat.vs_alloc != 0)
5224			error = spa_offline_log(spa);
5225	} else {
5226		error = SET_ERROR(ENOTSUP);
5227	}
5228
5229	if (error)
5230		return (error);
5231
5232	/*
5233	 * The evacuation succeeded.  Remove any remaining MOS metadata
5234	 * associated with this vdev, and wait for these changes to sync.
5235	 */
5236	ASSERT0(vd->vdev_stat.vs_alloc);
5237	txg = spa_vdev_config_enter(spa);
5238	vd->vdev_removing = B_TRUE;
5239	vdev_dirty_leaves(vd, VDD_DTL, txg);
5240	vdev_config_dirty(vd);
5241	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5242
5243	return (0);
5244}
5245
5246/*
5247 * Complete the removal by cleaning up the namespace.
5248 */
5249static void
5250spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5251{
5252	vdev_t *rvd = spa->spa_root_vdev;
5253	uint64_t id = vd->vdev_id;
5254	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5255
5256	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5257	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5258	ASSERT(vd == vd->vdev_top);
5259
5260	/*
5261	 * Only remove any devices which are empty.
5262	 */
5263	if (vd->vdev_stat.vs_alloc != 0)
5264		return;
5265
5266	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5267
5268	if (list_link_active(&vd->vdev_state_dirty_node))
5269		vdev_state_clean(vd);
5270	if (list_link_active(&vd->vdev_config_dirty_node))
5271		vdev_config_clean(vd);
5272
5273	vdev_free(vd);
5274
5275	if (last_vdev) {
5276		vdev_compact_children(rvd);
5277	} else {
5278		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5279		vdev_add_child(rvd, vd);
5280	}
5281	vdev_config_dirty(rvd);
5282
5283	/*
5284	 * Reassess the health of our root vdev.
5285	 */
5286	vdev_reopen(rvd);
5287}
5288
5289/*
5290 * Remove a device from the pool -
5291 *
5292 * Removing a device from the vdev namespace requires several steps
5293 * and can take a significant amount of time.  As a result we use
5294 * the spa_vdev_config_[enter/exit] functions which allow us to
5295 * grab and release the spa_config_lock while still holding the namespace
5296 * lock.  During each step the configuration is synced out.
5297 *
5298 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5299 * devices.
5300 */
5301int
5302spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5303{
5304	vdev_t *vd;
5305	metaslab_group_t *mg;
5306	nvlist_t **spares, **l2cache, *nv;
5307	uint64_t txg = 0;
5308	uint_t nspares, nl2cache;
5309	int error = 0;
5310	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5311
5312	ASSERT(spa_writeable(spa));
5313
5314	if (!locked)
5315		txg = spa_vdev_enter(spa);
5316
5317	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5318
5319	if (spa->spa_spares.sav_vdevs != NULL &&
5320	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5321	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5322	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5323		/*
5324		 * Only remove the hot spare if it's not currently in use
5325		 * in this pool.
5326		 */
5327		if (vd == NULL || unspare) {
5328			spa_vdev_remove_aux(spa->spa_spares.sav_config,
5329			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5330			spa_load_spares(spa);
5331			spa->spa_spares.sav_sync = B_TRUE;
5332		} else {
5333			error = SET_ERROR(EBUSY);
5334		}
5335	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
5336	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5337	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5338	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5339		/*
5340		 * Cache devices can always be removed.
5341		 */
5342		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5343		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5344		spa_load_l2cache(spa);
5345		spa->spa_l2cache.sav_sync = B_TRUE;
5346	} else if (vd != NULL && vd->vdev_islog) {
5347		ASSERT(!locked);
5348		ASSERT(vd == vd->vdev_top);
5349
5350		mg = vd->vdev_mg;
5351
5352		/*
5353		 * Stop allocating from this vdev.
5354		 */
5355		metaslab_group_passivate(mg);
5356
5357		/*
5358		 * Wait for the youngest allocations and frees to sync,
5359		 * and then wait for the deferral of those frees to finish.
5360		 */
5361		spa_vdev_config_exit(spa, NULL,
5362		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5363
5364		/*
5365		 * Attempt to evacuate the vdev.
5366		 */
5367		error = spa_vdev_remove_evacuate(spa, vd);
5368
5369		txg = spa_vdev_config_enter(spa);
5370
5371		/*
5372		 * If we couldn't evacuate the vdev, unwind.
5373		 */
5374		if (error) {
5375			metaslab_group_activate(mg);
5376			return (spa_vdev_exit(spa, NULL, txg, error));
5377		}
5378
5379		/*
5380		 * Clean up the vdev namespace.
5381		 */
5382		spa_vdev_remove_from_namespace(spa, vd);
5383
5384	} else if (vd != NULL) {
5385		/*
5386		 * Normal vdevs cannot be removed (yet).
5387		 */
5388		error = SET_ERROR(ENOTSUP);
5389	} else {
5390		/*
5391		 * There is no vdev of any kind with the specified guid.
5392		 */
5393		error = SET_ERROR(ENOENT);
5394	}
5395
5396	if (!locked)
5397		return (spa_vdev_exit(spa, NULL, txg, error));
5398
5399	return (error);
5400}
5401
5402/*
5403 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5404 * currently spared, so we can detach it.
5405 */
5406static vdev_t *
5407spa_vdev_resilver_done_hunt(vdev_t *vd)
5408{
5409	vdev_t *newvd, *oldvd;
5410
5411	for (int c = 0; c < vd->vdev_children; c++) {
5412		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5413		if (oldvd != NULL)
5414			return (oldvd);
5415	}
5416
5417	/*
5418	 * Check for a completed replacement.  We always consider the first
5419	 * vdev in the list to be the oldest vdev, and the last one to be
5420	 * the newest (see spa_vdev_attach() for how that works).  In
5421	 * the case where the newest vdev is faulted, we will not automatically
5422	 * remove it after a resilver completes.  This is OK as it will require
5423	 * user intervention to determine which disk the admin wishes to keep.
5424	 */
5425	if (vd->vdev_ops == &vdev_replacing_ops) {
5426		ASSERT(vd->vdev_children > 1);
5427
5428		newvd = vd->vdev_child[vd->vdev_children - 1];
5429		oldvd = vd->vdev_child[0];
5430
5431		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5432		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5433		    !vdev_dtl_required(oldvd))
5434			return (oldvd);
5435	}
5436
5437	/*
5438	 * Check for a completed resilver with the 'unspare' flag set.
5439	 */
5440	if (vd->vdev_ops == &vdev_spare_ops) {
5441		vdev_t *first = vd->vdev_child[0];
5442		vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5443
5444		if (last->vdev_unspare) {
5445			oldvd = first;
5446			newvd = last;
5447		} else if (first->vdev_unspare) {
5448			oldvd = last;
5449			newvd = first;
5450		} else {
5451			oldvd = NULL;
5452		}
5453
5454		if (oldvd != NULL &&
5455		    vdev_dtl_empty(newvd, DTL_MISSING) &&
5456		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5457		    !vdev_dtl_required(oldvd))
5458			return (oldvd);
5459
5460		/*
5461		 * If there are more than two spares attached to a disk,
5462		 * and those spares are not required, then we want to
5463		 * attempt to free them up now so that they can be used
5464		 * by other pools.  Once we're back down to a single
5465		 * disk+spare, we stop removing them.
5466		 */
5467		if (vd->vdev_children > 2) {
5468			newvd = vd->vdev_child[1];
5469
5470			if (newvd->vdev_isspare && last->vdev_isspare &&
5471			    vdev_dtl_empty(last, DTL_MISSING) &&
5472			    vdev_dtl_empty(last, DTL_OUTAGE) &&
5473			    !vdev_dtl_required(newvd))
5474				return (newvd);
5475		}
5476	}
5477
5478	return (NULL);
5479}
5480
5481static void
5482spa_vdev_resilver_done(spa_t *spa)
5483{
5484	vdev_t *vd, *pvd, *ppvd;
5485	uint64_t guid, sguid, pguid, ppguid;
5486
5487	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5488
5489	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5490		pvd = vd->vdev_parent;
5491		ppvd = pvd->vdev_parent;
5492		guid = vd->vdev_guid;
5493		pguid = pvd->vdev_guid;
5494		ppguid = ppvd->vdev_guid;
5495		sguid = 0;
5496		/*
5497		 * If we have just finished replacing a hot spared device, then
5498		 * we need to detach the parent's first child (the original hot
5499		 * spare) as well.
5500		 */
5501		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5502		    ppvd->vdev_children == 2) {
5503			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5504			sguid = ppvd->vdev_child[1]->vdev_guid;
5505		}
5506		ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5507
5508		spa_config_exit(spa, SCL_ALL, FTAG);
5509		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5510			return;
5511		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5512			return;
5513		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5514	}
5515
5516	spa_config_exit(spa, SCL_ALL, FTAG);
5517}
5518
5519/*
5520 * Update the stored path or FRU for this vdev.
5521 */
5522int
5523spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5524    boolean_t ispath)
5525{
5526	vdev_t *vd;
5527	boolean_t sync = B_FALSE;
5528
5529	ASSERT(spa_writeable(spa));
5530
5531	spa_vdev_state_enter(spa, SCL_ALL);
5532
5533	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5534		return (spa_vdev_state_exit(spa, NULL, ENOENT));
5535
5536	if (!vd->vdev_ops->vdev_op_leaf)
5537		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5538
5539	if (ispath) {
5540		if (strcmp(value, vd->vdev_path) != 0) {
5541			spa_strfree(vd->vdev_path);
5542			vd->vdev_path = spa_strdup(value);
5543			sync = B_TRUE;
5544		}
5545	} else {
5546		if (vd->vdev_fru == NULL) {
5547			vd->vdev_fru = spa_strdup(value);
5548			sync = B_TRUE;
5549		} else if (strcmp(value, vd->vdev_fru) != 0) {
5550			spa_strfree(vd->vdev_fru);
5551			vd->vdev_fru = spa_strdup(value);
5552			sync = B_TRUE;
5553		}
5554	}
5555
5556	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5557}
5558
5559int
5560spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5561{
5562	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5563}
5564
5565int
5566spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5567{
5568	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5569}
5570
5571/*
5572 * ==========================================================================
5573 * SPA Scanning
5574 * ==========================================================================
5575 */
5576
5577int
5578spa_scan_stop(spa_t *spa)
5579{
5580	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5581	if (dsl_scan_resilvering(spa->spa_dsl_pool))
5582		return (SET_ERROR(EBUSY));
5583	return (dsl_scan_cancel(spa->spa_dsl_pool));
5584}
5585
5586int
5587spa_scan(spa_t *spa, pool_scan_func_t func)
5588{
5589	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5590
5591	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5592		return (SET_ERROR(ENOTSUP));
5593
5594	/*
5595	 * If a resilver was requested, but there is no DTL on a
5596	 * writeable leaf device, we have nothing to do.
5597	 */
5598	if (func == POOL_SCAN_RESILVER &&
5599	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5600		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5601		return (0);
5602	}
5603
5604	return (dsl_scan(spa->spa_dsl_pool, func));
5605}
5606
5607/*
5608 * ==========================================================================
5609 * SPA async task processing
5610 * ==========================================================================
5611 */
5612
5613static void
5614spa_async_remove(spa_t *spa, vdev_t *vd)
5615{
5616	if (vd->vdev_remove_wanted) {
5617		vd->vdev_remove_wanted = B_FALSE;
5618		vd->vdev_delayed_close = B_FALSE;
5619		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5620
5621		/*
5622		 * We want to clear the stats, but we don't want to do a full
5623		 * vdev_clear() as that will cause us to throw away
5624		 * degraded/faulted state as well as attempt to reopen the
5625		 * device, all of which is a waste.
5626		 */
5627		vd->vdev_stat.vs_read_errors = 0;
5628		vd->vdev_stat.vs_write_errors = 0;
5629		vd->vdev_stat.vs_checksum_errors = 0;
5630
5631		vdev_state_dirty(vd->vdev_top);
5632	}
5633
5634	for (int c = 0; c < vd->vdev_children; c++)
5635		spa_async_remove(spa, vd->vdev_child[c]);
5636}
5637
5638static void
5639spa_async_probe(spa_t *spa, vdev_t *vd)
5640{
5641	if (vd->vdev_probe_wanted) {
5642		vd->vdev_probe_wanted = B_FALSE;
5643		vdev_reopen(vd);	/* vdev_open() does the actual probe */
5644	}
5645
5646	for (int c = 0; c < vd->vdev_children; c++)
5647		spa_async_probe(spa, vd->vdev_child[c]);
5648}
5649
5650static void
5651spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5652{
5653	sysevent_id_t eid;
5654	nvlist_t *attr;
5655	char *physpath;
5656
5657	if (!spa->spa_autoexpand)
5658		return;
5659
5660	for (int c = 0; c < vd->vdev_children; c++) {
5661		vdev_t *cvd = vd->vdev_child[c];
5662		spa_async_autoexpand(spa, cvd);
5663	}
5664
5665	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5666		return;
5667
5668	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5669	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5670
5671	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5672	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5673
5674	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5675	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5676
5677	nvlist_free(attr);
5678	kmem_free(physpath, MAXPATHLEN);
5679}
5680
5681static void
5682spa_async_thread(spa_t *spa)
5683{
5684	int tasks;
5685
5686	ASSERT(spa->spa_sync_on);
5687
5688	mutex_enter(&spa->spa_async_lock);
5689	tasks = spa->spa_async_tasks;
5690	spa->spa_async_tasks = 0;
5691	mutex_exit(&spa->spa_async_lock);
5692
5693	/*
5694	 * See if the config needs to be updated.
5695	 */
5696	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5697		uint64_t old_space, new_space;
5698
5699		mutex_enter(&spa_namespace_lock);
5700		old_space = metaslab_class_get_space(spa_normal_class(spa));
5701		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5702		new_space = metaslab_class_get_space(spa_normal_class(spa));
5703		mutex_exit(&spa_namespace_lock);
5704
5705		/*
5706		 * If the pool grew as a result of the config update,
5707		 * then log an internal history event.
5708		 */
5709		if (new_space != old_space) {
5710			spa_history_log_internal(spa, "vdev online", NULL,
5711			    "pool '%s' size: %llu(+%llu)",
5712			    spa_name(spa), new_space, new_space - old_space);
5713		}
5714	}
5715
5716	/*
5717	 * See if any devices need to be marked REMOVED.
5718	 */
5719	if (tasks & SPA_ASYNC_REMOVE) {
5720		spa_vdev_state_enter(spa, SCL_NONE);
5721		spa_async_remove(spa, spa->spa_root_vdev);
5722		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5723			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5724		for (int i = 0; i < spa->spa_spares.sav_count; i++)
5725			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5726		(void) spa_vdev_state_exit(spa, NULL, 0);
5727	}
5728
5729	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5730		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5731		spa_async_autoexpand(spa, spa->spa_root_vdev);
5732		spa_config_exit(spa, SCL_CONFIG, FTAG);
5733	}
5734
5735	/*
5736	 * See if any devices need to be probed.
5737	 */
5738	if (tasks & SPA_ASYNC_PROBE) {
5739		spa_vdev_state_enter(spa, SCL_NONE);
5740		spa_async_probe(spa, spa->spa_root_vdev);
5741		(void) spa_vdev_state_exit(spa, NULL, 0);
5742	}
5743
5744	/*
5745	 * If any devices are done replacing, detach them.
5746	 */
5747	if (tasks & SPA_ASYNC_RESILVER_DONE)
5748		spa_vdev_resilver_done(spa);
5749
5750	/*
5751	 * Kick off a resilver.
5752	 */
5753	if (tasks & SPA_ASYNC_RESILVER)
5754		dsl_resilver_restart(spa->spa_dsl_pool, 0);
5755
5756	/*
5757	 * Let the world know that we're done.
5758	 */
5759	mutex_enter(&spa->spa_async_lock);
5760	spa->spa_async_thread = NULL;
5761	cv_broadcast(&spa->spa_async_cv);
5762	mutex_exit(&spa->spa_async_lock);
5763	thread_exit();
5764}
5765
5766void
5767spa_async_suspend(spa_t *spa)
5768{
5769	mutex_enter(&spa->spa_async_lock);
5770	spa->spa_async_suspended++;
5771	while (spa->spa_async_thread != NULL)
5772		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5773	mutex_exit(&spa->spa_async_lock);
5774}
5775
5776void
5777spa_async_resume(spa_t *spa)
5778{
5779	mutex_enter(&spa->spa_async_lock);
5780	ASSERT(spa->spa_async_suspended != 0);
5781	spa->spa_async_suspended--;
5782	mutex_exit(&spa->spa_async_lock);
5783}
5784
5785static boolean_t
5786spa_async_tasks_pending(spa_t *spa)
5787{
5788	uint_t non_config_tasks;
5789	uint_t config_task;
5790	boolean_t config_task_suspended;
5791
5792	non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5793	config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5794	if (spa->spa_ccw_fail_time == 0) {
5795		config_task_suspended = B_FALSE;
5796	} else {
5797		config_task_suspended =
5798		    (gethrtime() - spa->spa_ccw_fail_time) <
5799		    (zfs_ccw_retry_interval * NANOSEC);
5800	}
5801
5802	return (non_config_tasks || (config_task && !config_task_suspended));
5803}
5804
5805static void
5806spa_async_dispatch(spa_t *spa)
5807{
5808	mutex_enter(&spa->spa_async_lock);
5809	if (spa_async_tasks_pending(spa) &&
5810	    !spa->spa_async_suspended &&
5811	    spa->spa_async_thread == NULL &&
5812	    rootdir != NULL)
5813		spa->spa_async_thread = thread_create(NULL, 0,
5814		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5815	mutex_exit(&spa->spa_async_lock);
5816}
5817
5818void
5819spa_async_request(spa_t *spa, int task)
5820{
5821	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5822	mutex_enter(&spa->spa_async_lock);
5823	spa->spa_async_tasks |= task;
5824	mutex_exit(&spa->spa_async_lock);
5825}
5826
5827/*
5828 * ==========================================================================
5829 * SPA syncing routines
5830 * ==========================================================================
5831 */
5832
5833static int
5834bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5835{
5836	bpobj_t *bpo = arg;
5837	bpobj_enqueue(bpo, bp, tx);
5838	return (0);
5839}
5840
5841static int
5842spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5843{
5844	zio_t *zio = arg;
5845
5846	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5847	    zio->io_flags));
5848	return (0);
5849}
5850
5851/*
5852 * Note: this simple function is not inlined to make it easier to dtrace the
5853 * amount of time spent syncing frees.
5854 */
5855static void
5856spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
5857{
5858	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5859	bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
5860	VERIFY(zio_wait(zio) == 0);
5861}
5862
5863/*
5864 * Note: this simple function is not inlined to make it easier to dtrace the
5865 * amount of time spent syncing deferred frees.
5866 */
5867static void
5868spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
5869{
5870	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5871	VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
5872	    spa_free_sync_cb, zio, tx), ==, 0);
5873	VERIFY0(zio_wait(zio));
5874}
5875
5876
5877static void
5878spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5879{
5880	char *packed = NULL;
5881	size_t bufsize;
5882	size_t nvsize = 0;
5883	dmu_buf_t *db;
5884
5885	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5886
5887	/*
5888	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5889	 * information.  This avoids the dmu_buf_will_dirty() path and
5890	 * saves us a pre-read to get data we don't actually care about.
5891	 */
5892	bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5893	packed = kmem_alloc(bufsize, KM_SLEEP);
5894
5895	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5896	    KM_SLEEP) == 0);
5897	bzero(packed + nvsize, bufsize - nvsize);
5898
5899	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5900
5901	kmem_free(packed, bufsize);
5902
5903	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5904	dmu_buf_will_dirty(db, tx);
5905	*(uint64_t *)db->db_data = nvsize;
5906	dmu_buf_rele(db, FTAG);
5907}
5908
5909static void
5910spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5911    const char *config, const char *entry)
5912{
5913	nvlist_t *nvroot;
5914	nvlist_t **list;
5915	int i;
5916
5917	if (!sav->sav_sync)
5918		return;
5919
5920	/*
5921	 * Update the MOS nvlist describing the list of available devices.
5922	 * spa_validate_aux() will have already made sure this nvlist is
5923	 * valid and the vdevs are labeled appropriately.
5924	 */
5925	if (sav->sav_object == 0) {
5926		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5927		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5928		    sizeof (uint64_t), tx);
5929		VERIFY(zap_update(spa->spa_meta_objset,
5930		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5931		    &sav->sav_object, tx) == 0);
5932	}
5933
5934	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5935	if (sav->sav_count == 0) {
5936		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5937	} else {
5938		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5939		for (i = 0; i < sav->sav_count; i++)
5940			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5941			    B_FALSE, VDEV_CONFIG_L2CACHE);
5942		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5943		    sav->sav_count) == 0);
5944		for (i = 0; i < sav->sav_count; i++)
5945			nvlist_free(list[i]);
5946		kmem_free(list, sav->sav_count * sizeof (void *));
5947	}
5948
5949	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5950	nvlist_free(nvroot);
5951
5952	sav->sav_sync = B_FALSE;
5953}
5954
5955static void
5956spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5957{
5958	nvlist_t *config;
5959
5960	if (list_is_empty(&spa->spa_config_dirty_list))
5961		return;
5962
5963	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5964
5965	config = spa_config_generate(spa, spa->spa_root_vdev,
5966	    dmu_tx_get_txg(tx), B_FALSE);
5967
5968	/*
5969	 * If we're upgrading the spa version then make sure that
5970	 * the config object gets updated with the correct version.
5971	 */
5972	if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5973		fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5974		    spa->spa_uberblock.ub_version);
5975
5976	spa_config_exit(spa, SCL_STATE, FTAG);
5977
5978	if (spa->spa_config_syncing)
5979		nvlist_free(spa->spa_config_syncing);
5980	spa->spa_config_syncing = config;
5981
5982	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5983}
5984
5985static void
5986spa_sync_version(void *arg, dmu_tx_t *tx)
5987{
5988	uint64_t *versionp = arg;
5989	uint64_t version = *versionp;
5990	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5991
5992	/*
5993	 * Setting the version is special cased when first creating the pool.
5994	 */
5995	ASSERT(tx->tx_txg != TXG_INITIAL);
5996
5997	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5998	ASSERT(version >= spa_version(spa));
5999
6000	spa->spa_uberblock.ub_version = version;
6001	vdev_config_dirty(spa->spa_root_vdev);
6002	spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6003}
6004
6005/*
6006 * Set zpool properties.
6007 */
6008static void
6009spa_sync_props(void *arg, dmu_tx_t *tx)
6010{
6011	nvlist_t *nvp = arg;
6012	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6013	objset_t *mos = spa->spa_meta_objset;
6014	nvpair_t *elem = NULL;
6015
6016	mutex_enter(&spa->spa_props_lock);
6017
6018	while ((elem = nvlist_next_nvpair(nvp, elem))) {
6019		uint64_t intval;
6020		char *strval, *fname;
6021		zpool_prop_t prop;
6022		const char *propname;
6023		zprop_type_t proptype;
6024		spa_feature_t fid;
6025
6026		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6027		case ZPROP_INVAL:
6028			/*
6029			 * We checked this earlier in spa_prop_validate().
6030			 */
6031			ASSERT(zpool_prop_feature(nvpair_name(elem)));
6032
6033			fname = strchr(nvpair_name(elem), '@') + 1;
6034			VERIFY0(zfeature_lookup_name(fname, &fid));
6035
6036			spa_feature_enable(spa, fid, tx);
6037			spa_history_log_internal(spa, "set", tx,
6038			    "%s=enabled", nvpair_name(elem));
6039			break;
6040
6041		case ZPOOL_PROP_VERSION:
6042			intval = fnvpair_value_uint64(elem);
6043			/*
6044			 * The version is synced seperatly before other
6045			 * properties and should be correct by now.
6046			 */
6047			ASSERT3U(spa_version(spa), >=, intval);