spa.c revision 12380e1e701fda28c9e9f32d01cafb54af279eb5
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	dsl_pool_t *dp = spa_get_dsl(spa);
1742
1743	switch (spa->spa_log_state) {
1744	case SPA_LOG_MISSING:
1745		/* need to recheck in case slog has been restored */
1746	case SPA_LOG_UNKNOWN:
1747		rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1748		    zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
1749		if (rv)
1750			spa_set_log_state(spa, SPA_LOG_MISSING);
1751		break;
1752	}
1753	return (rv);
1754}
1755
1756static boolean_t
1757spa_passivate_log(spa_t *spa)
1758{
1759	vdev_t *rvd = spa->spa_root_vdev;
1760	boolean_t slog_found = B_FALSE;
1761
1762	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1763
1764	if (!spa_has_slogs(spa))
1765		return (B_FALSE);
1766
1767	for (int c = 0; c < rvd->vdev_children; c++) {
1768		vdev_t *tvd = rvd->vdev_child[c];
1769		metaslab_group_t *mg = tvd->vdev_mg;
1770
1771		if (tvd->vdev_islog) {
1772			metaslab_group_passivate(mg);
1773			slog_found = B_TRUE;
1774		}
1775	}
1776
1777	return (slog_found);
1778}
1779
1780static void
1781spa_activate_log(spa_t *spa)
1782{
1783	vdev_t *rvd = spa->spa_root_vdev;
1784
1785	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1786
1787	for (int c = 0; c < rvd->vdev_children; c++) {
1788		vdev_t *tvd = rvd->vdev_child[c];
1789		metaslab_group_t *mg = tvd->vdev_mg;
1790
1791		if (tvd->vdev_islog)
1792			metaslab_group_activate(mg);
1793	}
1794}
1795
1796int
1797spa_offline_log(spa_t *spa)
1798{
1799	int error;
1800
1801	error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1802	    NULL, DS_FIND_CHILDREN);
1803	if (error == 0) {
1804		/*
1805		 * We successfully offlined the log device, sync out the
1806		 * current txg so that the "stubby" block can be removed
1807		 * by zil_sync().
1808		 */
1809		txg_wait_synced(spa->spa_dsl_pool, 0);
1810	}
1811	return (error);
1812}
1813
1814static void
1815spa_aux_check_removed(spa_aux_vdev_t *sav)
1816{
1817	for (int i = 0; i < sav->sav_count; i++)
1818		spa_check_removed(sav->sav_vdevs[i]);
1819}
1820
1821void
1822spa_claim_notify(zio_t *zio)
1823{
1824	spa_t *spa = zio->io_spa;
1825
1826	if (zio->io_error)
1827		return;
1828
1829	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1830	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1831		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1832	mutex_exit(&spa->spa_props_lock);
1833}
1834
1835typedef struct spa_load_error {
1836	uint64_t	sle_meta_count;
1837	uint64_t	sle_data_count;
1838} spa_load_error_t;
1839
1840static void
1841spa_load_verify_done(zio_t *zio)
1842{
1843	blkptr_t *bp = zio->io_bp;
1844	spa_load_error_t *sle = zio->io_private;
1845	dmu_object_type_t type = BP_GET_TYPE(bp);
1846	int error = zio->io_error;
1847	spa_t *spa = zio->io_spa;
1848
1849	if (error) {
1850		if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1851		    type != DMU_OT_INTENT_LOG)
1852			atomic_inc_64(&sle->sle_meta_count);
1853		else
1854			atomic_inc_64(&sle->sle_data_count);
1855	}
1856	zio_data_buf_free(zio->io_data, zio->io_size);
1857
1858	mutex_enter(&spa->spa_scrub_lock);
1859	spa->spa_scrub_inflight--;
1860	cv_broadcast(&spa->spa_scrub_io_cv);
1861	mutex_exit(&spa->spa_scrub_lock);
1862}
1863
1864/*
1865 * Maximum number of concurrent scrub i/os to create while verifying
1866 * a pool while importing it.
1867 */
1868int spa_load_verify_maxinflight = 10000;
1869boolean_t spa_load_verify_metadata = B_TRUE;
1870boolean_t spa_load_verify_data = B_TRUE;
1871
1872/*ARGSUSED*/
1873static int
1874spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1875    const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1876{
1877	if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
1878		return (0);
1879	/*
1880	 * Note: normally this routine will not be called if
1881	 * spa_load_verify_metadata is not set.  However, it may be useful
1882	 * to manually set the flag after the traversal has begun.
1883	 */
1884	if (!spa_load_verify_metadata)
1885		return (0);
1886	if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data)
1887		return (0);
1888
1889	zio_t *rio = arg;
1890	size_t size = BP_GET_PSIZE(bp);
1891	void *data = zio_data_buf_alloc(size);
1892
1893	mutex_enter(&spa->spa_scrub_lock);
1894	while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
1895		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1896	spa->spa_scrub_inflight++;
1897	mutex_exit(&spa->spa_scrub_lock);
1898
1899	zio_nowait(zio_read(rio, spa, bp, data, size,
1900	    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1901	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1902	    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1903	return (0);
1904}
1905
1906static int
1907spa_load_verify(spa_t *spa)
1908{
1909	zio_t *rio;
1910	spa_load_error_t sle = { 0 };
1911	zpool_rewind_policy_t policy;
1912	boolean_t verify_ok = B_FALSE;
1913	int error = 0;
1914
1915	zpool_get_rewind_policy(spa->spa_config, &policy);
1916
1917	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1918		return (0);
1919
1920	rio = zio_root(spa, NULL, &sle,
1921	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1922
1923	if (spa_load_verify_metadata) {
1924		error = traverse_pool(spa, spa->spa_verify_min_txg,
1925		    TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
1926		    spa_load_verify_cb, rio);
1927	}
1928
1929	(void) zio_wait(rio);
1930
1931	spa->spa_load_meta_errors = sle.sle_meta_count;
1932	spa->spa_load_data_errors = sle.sle_data_count;
1933
1934	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1935	    sle.sle_data_count <= policy.zrp_maxdata) {
1936		int64_t loss = 0;
1937
1938		verify_ok = B_TRUE;
1939		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1940		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1941
1942		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1943		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1944		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1945		VERIFY(nvlist_add_int64(spa->spa_load_info,
1946		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1947		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1948		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1949	} else {
1950		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1951	}
1952
1953	if (error) {
1954		if (error != ENXIO && error != EIO)
1955			error = SET_ERROR(EIO);
1956		return (error);
1957	}
1958
1959	return (verify_ok ? 0 : EIO);
1960}
1961
1962/*
1963 * Find a value in the pool props object.
1964 */
1965static void
1966spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1967{
1968	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1969	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1970}
1971
1972/*
1973 * Find a value in the pool directory object.
1974 */
1975static int
1976spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1977{
1978	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1979	    name, sizeof (uint64_t), 1, val));
1980}
1981
1982static int
1983spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1984{
1985	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1986	return (err);
1987}
1988
1989/*
1990 * Fix up config after a partly-completed split.  This is done with the
1991 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1992 * pool have that entry in their config, but only the splitting one contains
1993 * a list of all the guids of the vdevs that are being split off.
1994 *
1995 * This function determines what to do with that list: either rejoin
1996 * all the disks to the pool, or complete the splitting process.  To attempt
1997 * the rejoin, each disk that is offlined is marked online again, and
1998 * we do a reopen() call.  If the vdev label for every disk that was
1999 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2000 * then we call vdev_split() on each disk, and complete the split.
2001 *
2002 * Otherwise we leave the config alone, with all the vdevs in place in
2003 * the original pool.
2004 */
2005static void
2006spa_try_repair(spa_t *spa, nvlist_t *config)
2007{
2008	uint_t extracted;
2009	uint64_t *glist;
2010	uint_t i, gcount;
2011	nvlist_t *nvl;
2012	vdev_t **vd;
2013	boolean_t attempt_reopen;
2014
2015	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2016		return;
2017
2018	/* check that the config is complete */
2019	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2020	    &glist, &gcount) != 0)
2021		return;
2022
2023	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2024
2025	/* attempt to online all the vdevs & validate */
2026	attempt_reopen = B_TRUE;
2027	for (i = 0; i < gcount; i++) {
2028		if (glist[i] == 0)	/* vdev is hole */
2029			continue;
2030
2031		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2032		if (vd[i] == NULL) {
2033			/*
2034			 * Don't bother attempting to reopen the disks;
2035			 * just do the split.
2036			 */
2037			attempt_reopen = B_FALSE;
2038		} else {
2039			/* attempt to re-online it */
2040			vd[i]->vdev_offline = B_FALSE;
2041		}
2042	}
2043
2044	if (attempt_reopen) {
2045		vdev_reopen(spa->spa_root_vdev);
2046
2047		/* check each device to see what state it's in */
2048		for (extracted = 0, i = 0; i < gcount; i++) {
2049			if (vd[i] != NULL &&
2050			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2051				break;
2052			++extracted;
2053		}
2054	}
2055
2056	/*
2057	 * If every disk has been moved to the new pool, or if we never
2058	 * even attempted to look at them, then we split them off for
2059	 * good.
2060	 */
2061	if (!attempt_reopen || gcount == extracted) {
2062		for (i = 0; i < gcount; i++)
2063			if (vd[i] != NULL)
2064				vdev_split(vd[i]);
2065		vdev_reopen(spa->spa_root_vdev);
2066	}
2067
2068	kmem_free(vd, gcount * sizeof (vdev_t *));
2069}
2070
2071static int
2072spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2073    boolean_t mosconfig)
2074{
2075	nvlist_t *config = spa->spa_config;
2076	char *ereport = FM_EREPORT_ZFS_POOL;
2077	char *comment;
2078	int error;
2079	uint64_t pool_guid;
2080	nvlist_t *nvl;
2081
2082	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2083		return (SET_ERROR(EINVAL));
2084
2085	ASSERT(spa->spa_comment == NULL);
2086	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2087		spa->spa_comment = spa_strdup(comment);
2088
2089	/*
2090	 * Versioning wasn't explicitly added to the label until later, so if
2091	 * it's not present treat it as the initial version.
2092	 */
2093	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2094	    &spa->spa_ubsync.ub_version) != 0)
2095		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2096
2097	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2098	    &spa->spa_config_txg);
2099
2100	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2101	    spa_guid_exists(pool_guid, 0)) {
2102		error = SET_ERROR(EEXIST);
2103	} else {
2104		spa->spa_config_guid = pool_guid;
2105
2106		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2107		    &nvl) == 0) {
2108			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2109			    KM_SLEEP) == 0);
2110		}
2111
2112		nvlist_free(spa->spa_load_info);
2113		spa->spa_load_info = fnvlist_alloc();
2114
2115		gethrestime(&spa->spa_loaded_ts);
2116		error = spa_load_impl(spa, pool_guid, config, state, type,
2117		    mosconfig, &ereport);
2118	}
2119
2120	/*
2121	 * Don't count references from objsets that are already closed
2122	 * and are making their way through the eviction process.
2123	 */
2124	spa_evicting_os_wait(spa);
2125	spa->spa_minref = refcount_count(&spa->spa_refcount);
2126	if (error) {
2127		if (error != EEXIST) {
2128			spa->spa_loaded_ts.tv_sec = 0;
2129			spa->spa_loaded_ts.tv_nsec = 0;
2130		}
2131		if (error != EBADF) {
2132			zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2133		}
2134	}
2135	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2136	spa->spa_ena = 0;
2137
2138	return (error);
2139}
2140
2141/*
2142 * Load an existing storage pool, using the pool's builtin spa_config as a
2143 * source of configuration information.
2144 */
2145static int
2146spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2147    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2148    char **ereport)
2149{
2150	int error = 0;
2151	nvlist_t *nvroot = NULL;
2152	nvlist_t *label;
2153	vdev_t *rvd;
2154	uberblock_t *ub = &spa->spa_uberblock;
2155	uint64_t children, config_cache_txg = spa->spa_config_txg;
2156	int orig_mode = spa->spa_mode;
2157	int parse;
2158	uint64_t obj;
2159	boolean_t missing_feat_write = B_FALSE;
2160
2161	/*
2162	 * If this is an untrusted config, access the pool in read-only mode.
2163	 * This prevents things like resilvering recently removed devices.
2164	 */
2165	if (!mosconfig)
2166		spa->spa_mode = FREAD;
2167
2168	ASSERT(MUTEX_HELD(&spa_namespace_lock));
2169
2170	spa->spa_load_state = state;
2171
2172	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2173		return (SET_ERROR(EINVAL));
2174
2175	parse = (type == SPA_IMPORT_EXISTING ?
2176	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2177
2178	/*
2179	 * Create "The Godfather" zio to hold all async IOs
2180	 */
2181	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2182	    KM_SLEEP);
2183	for (int i = 0; i < max_ncpus; i++) {
2184		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2185		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2186		    ZIO_FLAG_GODFATHER);
2187	}
2188
2189	/*
2190	 * Parse the configuration into a vdev tree.  We explicitly set the
2191	 * value that will be returned by spa_version() since parsing the
2192	 * configuration requires knowing the version number.
2193	 */
2194	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2195	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2196	spa_config_exit(spa, SCL_ALL, FTAG);
2197
2198	if (error != 0)
2199		return (error);
2200
2201	ASSERT(spa->spa_root_vdev == rvd);
2202
2203	if (type != SPA_IMPORT_ASSEMBLE) {
2204		ASSERT(spa_guid(spa) == pool_guid);
2205	}
2206
2207	/*
2208	 * Try to open all vdevs, loading each label in the process.
2209	 */
2210	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2211	error = vdev_open(rvd);
2212	spa_config_exit(spa, SCL_ALL, FTAG);
2213	if (error != 0)
2214		return (error);
2215
2216	/*
2217	 * We need to validate the vdev labels against the configuration that
2218	 * we have in hand, which is dependent on the setting of mosconfig. If
2219	 * mosconfig is true then we're validating the vdev labels based on
2220	 * that config.  Otherwise, we're validating against the cached config
2221	 * (zpool.cache) that was read when we loaded the zfs module, and then
2222	 * later we will recursively call spa_load() and validate against
2223	 * the vdev config.
2224	 *
2225	 * If we're assembling a new pool that's been split off from an
2226	 * existing pool, the labels haven't yet been updated so we skip
2227	 * validation for now.
2228	 */
2229	if (type != SPA_IMPORT_ASSEMBLE) {
2230		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2231		error = vdev_validate(rvd, mosconfig);
2232		spa_config_exit(spa, SCL_ALL, FTAG);
2233
2234		if (error != 0)
2235			return (error);
2236
2237		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2238			return (SET_ERROR(ENXIO));
2239	}
2240
2241	/*
2242	 * Find the best uberblock.
2243	 */
2244	vdev_uberblock_load(rvd, ub, &label);
2245
2246	/*
2247	 * If we weren't able to find a single valid uberblock, return failure.
2248	 */
2249	if (ub->ub_txg == 0) {
2250		nvlist_free(label);
2251		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2252	}
2253
2254	/*
2255	 * If the pool has an unsupported version we can't open it.
2256	 */
2257	if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2258		nvlist_free(label);
2259		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2260	}
2261
2262	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2263		nvlist_t *features;
2264
2265		/*
2266		 * If we weren't able to find what's necessary for reading the
2267		 * MOS in the label, return failure.
2268		 */
2269		if (label == NULL || nvlist_lookup_nvlist(label,
2270		    ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2271			nvlist_free(label);
2272			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2273			    ENXIO));
2274		}
2275
2276		/*
2277		 * Update our in-core representation with the definitive values
2278		 * from the label.
2279		 */
2280		nvlist_free(spa->spa_label_features);
2281		VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2282	}
2283
2284	nvlist_free(label);
2285
2286	/*
2287	 * Look through entries in the label nvlist's features_for_read. If
2288	 * there is a feature listed there which we don't understand then we
2289	 * cannot open a pool.
2290	 */
2291	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2292		nvlist_t *unsup_feat;
2293
2294		VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2295		    0);
2296
2297		for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2298		    NULL); nvp != NULL;
2299		    nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2300			if (!zfeature_is_supported(nvpair_name(nvp))) {
2301				VERIFY(nvlist_add_string(unsup_feat,
2302				    nvpair_name(nvp), "") == 0);
2303			}
2304		}
2305
2306		if (!nvlist_empty(unsup_feat)) {
2307			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2308			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2309			nvlist_free(unsup_feat);
2310			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2311			    ENOTSUP));
2312		}
2313
2314		nvlist_free(unsup_feat);
2315	}
2316
2317	/*
2318	 * If the vdev guid sum doesn't match the uberblock, we have an
2319	 * incomplete configuration.  We first check to see if the pool
2320	 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2321	 * If it is, defer the vdev_guid_sum check till later so we
2322	 * can handle missing vdevs.
2323	 */
2324	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2325	    &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2326	    rvd->vdev_guid_sum != ub->ub_guid_sum)
2327		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2328
2329	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2330		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2331		spa_try_repair(spa, config);
2332		spa_config_exit(spa, SCL_ALL, FTAG);
2333		nvlist_free(spa->spa_config_splitting);
2334		spa->spa_config_splitting = NULL;
2335	}
2336
2337	/*
2338	 * Initialize internal SPA structures.
2339	 */
2340	spa->spa_state = POOL_STATE_ACTIVE;
2341	spa->spa_ubsync = spa->spa_uberblock;
2342	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2343	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2344	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2345	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2346	spa->spa_claim_max_txg = spa->spa_first_txg;
2347	spa->spa_prev_software_version = ub->ub_software_version;
2348
2349	error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2350	if (error)
2351		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2352	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2353
2354	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2355		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2356
2357	if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2358		boolean_t missing_feat_read = B_FALSE;
2359		nvlist_t *unsup_feat, *enabled_feat;
2360
2361		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2362		    &spa->spa_feat_for_read_obj) != 0) {
2363			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2364		}
2365
2366		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2367		    &spa->spa_feat_for_write_obj) != 0) {
2368			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2369		}
2370
2371		if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2372		    &spa->spa_feat_desc_obj) != 0) {
2373			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2374		}
2375
2376		enabled_feat = fnvlist_alloc();
2377		unsup_feat = fnvlist_alloc();
2378
2379		if (!spa_features_check(spa, B_FALSE,
2380		    unsup_feat, enabled_feat))
2381			missing_feat_read = B_TRUE;
2382
2383		if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2384			if (!spa_features_check(spa, B_TRUE,
2385			    unsup_feat, enabled_feat)) {
2386				missing_feat_write = B_TRUE;
2387			}
2388		}
2389
2390		fnvlist_add_nvlist(spa->spa_load_info,
2391		    ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2392
2393		if (!nvlist_empty(unsup_feat)) {
2394			fnvlist_add_nvlist(spa->spa_load_info,
2395			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2396		}
2397
2398		fnvlist_free(enabled_feat);
2399		fnvlist_free(unsup_feat);
2400
2401		if (!missing_feat_read) {
2402			fnvlist_add_boolean(spa->spa_load_info,
2403			    ZPOOL_CONFIG_CAN_RDONLY);
2404		}
2405
2406		/*
2407		 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2408		 * twofold: to determine whether the pool is available for
2409		 * import in read-write mode and (if it is not) whether the
2410		 * pool is available for import in read-only mode. If the pool
2411		 * is available for import in read-write mode, it is displayed
2412		 * as available in userland; if it is not available for import
2413		 * in read-only mode, it is displayed as unavailable in
2414		 * userland. If the pool is available for import in read-only
2415		 * mode but not read-write mode, it is displayed as unavailable
2416		 * in userland with a special note that the pool is actually
2417		 * available for open in read-only mode.
2418		 *
2419		 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2420		 * missing a feature for write, we must first determine whether
2421		 * the pool can be opened read-only before returning to
2422		 * userland in order to know whether to display the
2423		 * abovementioned note.
2424		 */
2425		if (missing_feat_read || (missing_feat_write &&
2426		    spa_writeable(spa))) {
2427			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2428			    ENOTSUP));
2429		}
2430
2431		/*
2432		 * Load refcounts for ZFS features from disk into an in-memory
2433		 * cache during SPA initialization.
2434		 */
2435		for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2436			uint64_t refcount;
2437
2438			error = feature_get_refcount_from_disk(spa,
2439			    &spa_feature_table[i], &refcount);
2440			if (error == 0) {
2441				spa->spa_feat_refcount_cache[i] = refcount;
2442			} else if (error == ENOTSUP) {
2443				spa->spa_feat_refcount_cache[i] =
2444				    SPA_FEATURE_DISABLED;
2445			} else {
2446				return (spa_vdev_err(rvd,
2447				    VDEV_AUX_CORRUPT_DATA, EIO));
2448			}
2449		}
2450	}
2451
2452	if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2453		if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2454		    &spa->spa_feat_enabled_txg_obj) != 0)
2455			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2456	}
2457
2458	spa->spa_is_initializing = B_TRUE;
2459	error = dsl_pool_open(spa->spa_dsl_pool);
2460	spa->spa_is_initializing = B_FALSE;
2461	if (error != 0)
2462		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2463
2464	if (!mosconfig) {
2465		uint64_t hostid;
2466		nvlist_t *policy = NULL, *nvconfig;
2467
2468		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2469			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2470
2471		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2472		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2473			char *hostname;
2474			unsigned long myhostid = 0;
2475
2476			VERIFY(nvlist_lookup_string(nvconfig,
2477			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2478
2479#ifdef	_KERNEL
2480			myhostid = zone_get_hostid(NULL);
2481#else	/* _KERNEL */
2482			/*
2483			 * We're emulating the system's hostid in userland, so
2484			 * we can't use zone_get_hostid().
2485			 */
2486			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2487#endif	/* _KERNEL */
2488			if (hostid != 0 && myhostid != 0 &&
2489			    hostid != myhostid) {
2490				nvlist_free(nvconfig);
2491				cmn_err(CE_WARN, "pool '%s' could not be "
2492				    "loaded as it was last accessed by "
2493				    "another system (host: %s hostid: 0x%lx). "
2494				    "See: http://illumos.org/msg/ZFS-8000-EY",
2495				    spa_name(spa), hostname,
2496				    (unsigned long)hostid);
2497				return (SET_ERROR(EBADF));
2498			}
2499		}
2500		if (nvlist_lookup_nvlist(spa->spa_config,
2501		    ZPOOL_REWIND_POLICY, &policy) == 0)
2502			VERIFY(nvlist_add_nvlist(nvconfig,
2503			    ZPOOL_REWIND_POLICY, policy) == 0);
2504
2505		spa_config_set(spa, nvconfig);
2506		spa_unload(spa);
2507		spa_deactivate(spa);
2508		spa_activate(spa, orig_mode);
2509
2510		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2511	}
2512
2513	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2514		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2515	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2516	if (error != 0)
2517		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2518
2519	/*
2520	 * Load the bit that tells us to use the new accounting function
2521	 * (raid-z deflation).  If we have an older pool, this will not
2522	 * be present.
2523	 */
2524	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2525	if (error != 0 && error != ENOENT)
2526		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2527
2528	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2529	    &spa->spa_creation_version);
2530	if (error != 0 && error != ENOENT)
2531		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2532
2533	/*
2534	 * Load the persistent error log.  If we have an older pool, this will
2535	 * not be present.
2536	 */
2537	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2538	if (error != 0 && error != ENOENT)
2539		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2540
2541	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2542	    &spa->spa_errlog_scrub);
2543	if (error != 0 && error != ENOENT)
2544		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2545
2546	/*
2547	 * Load the history object.  If we have an older pool, this
2548	 * will not be present.
2549	 */
2550	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2551	if (error != 0 && error != ENOENT)
2552		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2553
2554	/*
2555	 * If we're assembling the pool from the split-off vdevs of
2556	 * an existing pool, we don't want to attach the spares & cache
2557	 * devices.
2558	 */
2559
2560	/*
2561	 * Load any hot spares for this pool.
2562	 */
2563	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2564	if (error != 0 && error != ENOENT)
2565		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2566	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2567		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2568		if (load_nvlist(spa, spa->spa_spares.sav_object,
2569		    &spa->spa_spares.sav_config) != 0)
2570			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2571
2572		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2573		spa_load_spares(spa);
2574		spa_config_exit(spa, SCL_ALL, FTAG);
2575	} else if (error == 0) {
2576		spa->spa_spares.sav_sync = B_TRUE;
2577	}
2578
2579	/*
2580	 * Load any level 2 ARC devices for this pool.
2581	 */
2582	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2583	    &spa->spa_l2cache.sav_object);
2584	if (error != 0 && error != ENOENT)
2585		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2586	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2587		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2588		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2589		    &spa->spa_l2cache.sav_config) != 0)
2590			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2591
2592		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2593		spa_load_l2cache(spa);
2594		spa_config_exit(spa, SCL_ALL, FTAG);
2595	} else if (error == 0) {
2596		spa->spa_l2cache.sav_sync = B_TRUE;
2597	}
2598
2599	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2600
2601	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2602	if (error && error != ENOENT)
2603		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2604
2605	if (error == 0) {
2606		uint64_t autoreplace;
2607
2608		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2609		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2610		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2611		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2612		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2613		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2614		    &spa->spa_dedup_ditto);
2615
2616		spa->spa_autoreplace = (autoreplace != 0);
2617	}
2618
2619	/*
2620	 * If the 'autoreplace' property is set, then post a resource notifying
2621	 * the ZFS DE that it should not issue any faults for unopenable
2622	 * devices.  We also iterate over the vdevs, and post a sysevent for any
2623	 * unopenable vdevs so that the normal autoreplace handler can take
2624	 * over.
2625	 */
2626	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2627		spa_check_removed(spa->spa_root_vdev);
2628		/*
2629		 * For the import case, this is done in spa_import(), because
2630		 * at this point we're using the spare definitions from
2631		 * the MOS config, not necessarily from the userland config.
2632		 */
2633		if (state != SPA_LOAD_IMPORT) {
2634			spa_aux_check_removed(&spa->spa_spares);
2635			spa_aux_check_removed(&spa->spa_l2cache);
2636		}
2637	}
2638
2639	/*
2640	 * Load the vdev state for all toplevel vdevs.
2641	 */
2642	vdev_load(rvd);
2643
2644	/*
2645	 * Propagate the leaf DTLs we just loaded all the way up the tree.
2646	 */
2647	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2648	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2649	spa_config_exit(spa, SCL_ALL, FTAG);
2650
2651	/*
2652	 * Load the DDTs (dedup tables).
2653	 */
2654	error = ddt_load(spa);
2655	if (error != 0)
2656		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2657
2658	spa_update_dspace(spa);
2659
2660	/*
2661	 * Validate the config, using the MOS config to fill in any
2662	 * information which might be missing.  If we fail to validate
2663	 * the config then declare the pool unfit for use. If we're
2664	 * assembling a pool from a split, the log is not transferred
2665	 * over.
2666	 */
2667	if (type != SPA_IMPORT_ASSEMBLE) {
2668		nvlist_t *nvconfig;
2669
2670		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2671			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2672
2673		if (!spa_config_valid(spa, nvconfig)) {
2674			nvlist_free(nvconfig);
2675			return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2676			    ENXIO));
2677		}
2678		nvlist_free(nvconfig);
2679
2680		/*
2681		 * Now that we've validated the config, check the state of the
2682		 * root vdev.  If it can't be opened, it indicates one or
2683		 * more toplevel vdevs are faulted.
2684		 */
2685		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2686			return (SET_ERROR(ENXIO));
2687
2688		if (spa_writeable(spa) && spa_check_logs(spa)) {
2689			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2690			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2691		}
2692	}
2693
2694	if (missing_feat_write) {
2695		ASSERT(state == SPA_LOAD_TRYIMPORT);
2696
2697		/*
2698		 * At this point, we know that we can open the pool in
2699		 * read-only mode but not read-write mode. We now have enough
2700		 * information and can return to userland.
2701		 */
2702		return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2703	}
2704
2705	/*
2706	 * We've successfully opened the pool, verify that we're ready
2707	 * to start pushing transactions.
2708	 */
2709	if (state != SPA_LOAD_TRYIMPORT) {
2710		if (error = spa_load_verify(spa))
2711			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2712			    error));
2713	}
2714
2715	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2716	    spa->spa_load_max_txg == UINT64_MAX)) {
2717		dmu_tx_t *tx;
2718		int need_update = B_FALSE;
2719		dsl_pool_t *dp = spa_get_dsl(spa);
2720
2721		ASSERT(state != SPA_LOAD_TRYIMPORT);
2722
2723		/*
2724		 * Claim log blocks that haven't been committed yet.
2725		 * This must all happen in a single txg.
2726		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2727		 * invoked from zil_claim_log_block()'s i/o done callback.
2728		 * Price of rollback is that we abandon the log.
2729		 */
2730		spa->spa_claiming = B_TRUE;
2731
2732		tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
2733		(void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
2734		    zil_claim, tx, DS_FIND_CHILDREN);
2735		dmu_tx_commit(tx);
2736
2737		spa->spa_claiming = B_FALSE;
2738
2739		spa_set_log_state(spa, SPA_LOG_GOOD);
2740		spa->spa_sync_on = B_TRUE;
2741		txg_sync_start(spa->spa_dsl_pool);
2742
2743		/*
2744		 * Wait for all claims to sync.  We sync up to the highest
2745		 * claimed log block birth time so that claimed log blocks
2746		 * don't appear to be from the future.  spa_claim_max_txg
2747		 * will have been set for us by either zil_check_log_chain()
2748		 * (invoked from spa_check_logs()) or zil_claim() above.
2749		 */
2750		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2751
2752		/*
2753		 * If the config cache is stale, or we have uninitialized
2754		 * metaslabs (see spa_vdev_add()), then update the config.
2755		 *
2756		 * If this is a verbatim import, trust the current
2757		 * in-core spa_config and update the disk labels.
2758		 */
2759		if (config_cache_txg != spa->spa_config_txg ||
2760		    state == SPA_LOAD_IMPORT ||
2761		    state == SPA_LOAD_RECOVER ||
2762		    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2763			need_update = B_TRUE;
2764
2765		for (int c = 0; c < rvd->vdev_children; c++)
2766			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2767				need_update = B_TRUE;
2768
2769		/*
2770		 * Update the config cache asychronously in case we're the
2771		 * root pool, in which case the config cache isn't writable yet.
2772		 */
2773		if (need_update)
2774			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2775
2776		/*
2777		 * Check all DTLs to see if anything needs resilvering.
2778		 */
2779		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2780		    vdev_resilver_needed(rvd, NULL, NULL))
2781			spa_async_request(spa, SPA_ASYNC_RESILVER);
2782
2783		/*
2784		 * Log the fact that we booted up (so that we can detect if
2785		 * we rebooted in the middle of an operation).
2786		 */
2787		spa_history_log_version(spa, "open");
2788
2789		/*
2790		 * Delete any inconsistent datasets.
2791		 */
2792		(void) dmu_objset_find(spa_name(spa),
2793		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2794
2795		/*
2796		 * Clean up any stale temporary dataset userrefs.
2797		 */
2798		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2799	}
2800
2801	return (0);
2802}
2803
2804static int
2805spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2806{
2807	int mode = spa->spa_mode;
2808
2809	spa_unload(spa);
2810	spa_deactivate(spa);
2811
2812	spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
2813
2814	spa_activate(spa, mode);
2815	spa_async_suspend(spa);
2816
2817	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2818}
2819
2820/*
2821 * If spa_load() fails this function will try loading prior txg's. If
2822 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2823 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2824 * function will not rewind the pool and will return the same error as
2825 * spa_load().
2826 */
2827static int
2828spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2829    uint64_t max_request, int rewind_flags)
2830{
2831	nvlist_t *loadinfo = NULL;
2832	nvlist_t *config = NULL;
2833	int load_error, rewind_error;
2834	uint64_t safe_rewind_txg;
2835	uint64_t min_txg;
2836
2837	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2838		spa->spa_load_max_txg = spa->spa_load_txg;
2839		spa_set_log_state(spa, SPA_LOG_CLEAR);
2840	} else {
2841		spa->spa_load_max_txg = max_request;
2842		if (max_request != UINT64_MAX)
2843			spa->spa_extreme_rewind = B_TRUE;
2844	}
2845
2846	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2847	    mosconfig);
2848	if (load_error == 0)
2849		return (0);
2850
2851	if (spa->spa_root_vdev != NULL)
2852		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2853
2854	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2855	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2856
2857	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2858		nvlist_free(config);
2859		return (load_error);
2860	}
2861
2862	if (state == SPA_LOAD_RECOVER) {
2863		/* Price of rolling back is discarding txgs, including log */
2864		spa_set_log_state(spa, SPA_LOG_CLEAR);
2865	} else {
2866		/*
2867		 * If we aren't rolling back save the load info from our first
2868		 * import attempt so that we can restore it after attempting
2869		 * to rewind.
2870		 */
2871		loadinfo = spa->spa_load_info;
2872		spa->spa_load_info = fnvlist_alloc();
2873	}
2874
2875	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2876	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2877	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2878	    TXG_INITIAL : safe_rewind_txg;
2879
2880	/*
2881	 * Continue as long as we're finding errors, we're still within
2882	 * the acceptable rewind range, and we're still finding uberblocks
2883	 */
2884	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2885	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2886		if (spa->spa_load_max_txg < safe_rewind_txg)
2887			spa->spa_extreme_rewind = B_TRUE;
2888		rewind_error = spa_load_retry(spa, state, mosconfig);
2889	}
2890
2891	spa->spa_extreme_rewind = B_FALSE;
2892	spa->spa_load_max_txg = UINT64_MAX;
2893
2894	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2895		spa_config_set(spa, config);
2896
2897	if (state == SPA_LOAD_RECOVER) {
2898		ASSERT3P(loadinfo, ==, NULL);
2899		return (rewind_error);
2900	} else {
2901		/* Store the rewind info as part of the initial load info */
2902		fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2903		    spa->spa_load_info);
2904
2905		/* Restore the initial load info */
2906		fnvlist_free(spa->spa_load_info);
2907		spa->spa_load_info = loadinfo;
2908
2909		return (load_error);
2910	}
2911}
2912
2913/*
2914 * Pool Open/Import
2915 *
2916 * The import case is identical to an open except that the configuration is sent
2917 * down from userland, instead of grabbed from the configuration cache.  For the
2918 * case of an open, the pool configuration will exist in the
2919 * POOL_STATE_UNINITIALIZED state.
2920 *
2921 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2922 * the same time open the pool, without having to keep around the spa_t in some
2923 * ambiguous state.
2924 */
2925static int
2926spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2927    nvlist_t **config)
2928{
2929	spa_t *spa;
2930	spa_load_state_t state = SPA_LOAD_OPEN;
2931	int error;
2932	int locked = B_FALSE;
2933
2934	*spapp = NULL;
2935
2936	/*
2937	 * As disgusting as this is, we need to support recursive calls to this
2938	 * function because dsl_dir_open() is called during spa_load(), and ends
2939	 * up calling spa_open() again.  The real fix is to figure out how to
2940	 * avoid dsl_dir_open() calling this in the first place.
2941	 */
2942	if (mutex_owner(&spa_namespace_lock) != curthread) {
2943		mutex_enter(&spa_namespace_lock);
2944		locked = B_TRUE;
2945	}
2946
2947	if ((spa = spa_lookup(pool)) == NULL) {
2948		if (locked)
2949			mutex_exit(&spa_namespace_lock);
2950		return (SET_ERROR(ENOENT));
2951	}
2952
2953	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2954		zpool_rewind_policy_t policy;
2955
2956		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2957		    &policy);
2958		if (policy.zrp_request & ZPOOL_DO_REWIND)
2959			state = SPA_LOAD_RECOVER;
2960
2961		spa_activate(spa, spa_mode_global);
2962
2963		if (state != SPA_LOAD_RECOVER)
2964			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2965
2966		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2967		    policy.zrp_request);
2968
2969		if (error == EBADF) {
2970			/*
2971			 * If vdev_validate() returns failure (indicated by
2972			 * EBADF), it indicates that one of the vdevs indicates
2973			 * that the pool has been exported or destroyed.  If
2974			 * this is the case, the config cache is out of sync and
2975			 * we should remove the pool from the namespace.
2976			 */
2977			spa_unload(spa);
2978			spa_deactivate(spa);
2979			spa_config_sync(spa, B_TRUE, B_TRUE);
2980			spa_remove(spa);
2981			if (locked)
2982				mutex_exit(&spa_namespace_lock);
2983			return (SET_ERROR(ENOENT));
2984		}
2985
2986		if (error) {
2987			/*
2988			 * We can't open the pool, but we still have useful
2989			 * information: the state of each vdev after the
2990			 * attempted vdev_open().  Return this to the user.
2991			 */
2992			if (config != NULL && spa->spa_config) {
2993				VERIFY(nvlist_dup(spa->spa_config, config,
2994				    KM_SLEEP) == 0);
2995				VERIFY(nvlist_add_nvlist(*config,
2996				    ZPOOL_CONFIG_LOAD_INFO,
2997				    spa->spa_load_info) == 0);
2998			}
2999			spa_unload(spa);
3000			spa_deactivate(spa);
3001			spa->spa_last_open_failed = error;
3002			if (locked)
3003				mutex_exit(&spa_namespace_lock);
3004			*spapp = NULL;
3005			return (error);
3006		}
3007	}
3008
3009	spa_open_ref(spa, tag);
3010
3011	if (config != NULL)
3012		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3013
3014	/*
3015	 * If we've recovered the pool, pass back any information we
3016	 * gathered while doing the load.
3017	 */
3018	if (state == SPA_LOAD_RECOVER) {
3019		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
3020		    spa->spa_load_info) == 0);
3021	}
3022
3023	if (locked) {
3024		spa->spa_last_open_failed = 0;
3025		spa->spa_last_ubsync_txg = 0;
3026		spa->spa_load_txg = 0;
3027		mutex_exit(&spa_namespace_lock);
3028	}
3029
3030	*spapp = spa;
3031
3032	return (0);
3033}
3034
3035int
3036spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3037    nvlist_t **config)
3038{
3039	return (spa_open_common(name, spapp, tag, policy, config));
3040}
3041
3042int
3043spa_open(const char *name, spa_t **spapp, void *tag)
3044{
3045	return (spa_open_common(name, spapp, tag, NULL, NULL));
3046}
3047
3048/*
3049 * Lookup the given spa_t, incrementing the inject count in the process,
3050 * preventing it from being exported or destroyed.
3051 */
3052spa_t *
3053spa_inject_addref(char *name)
3054{
3055	spa_t *spa;
3056
3057	mutex_enter(&spa_namespace_lock);
3058	if ((spa = spa_lookup(name)) == NULL) {
3059		mutex_exit(&spa_namespace_lock);
3060		return (NULL);
3061	}
3062	spa->spa_inject_ref++;
3063	mutex_exit(&spa_namespace_lock);
3064
3065	return (spa);
3066}
3067
3068void
3069spa_inject_delref(spa_t *spa)
3070{
3071	mutex_enter(&spa_namespace_lock);
3072	spa->spa_inject_ref--;
3073	mutex_exit(&spa_namespace_lock);
3074}
3075
3076/*
3077 * Add spares device information to the nvlist.
3078 */
3079static void
3080spa_add_spares(spa_t *spa, nvlist_t *config)
3081{
3082	nvlist_t **spares;
3083	uint_t i, nspares;
3084	nvlist_t *nvroot;
3085	uint64_t guid;
3086	vdev_stat_t *vs;
3087	uint_t vsc;
3088	uint64_t pool;
3089
3090	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3091
3092	if (spa->spa_spares.sav_count == 0)
3093		return;
3094
3095	VERIFY(nvlist_lookup_nvlist(config,
3096	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3097	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3098	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3099	if (nspares != 0) {
3100		VERIFY(nvlist_add_nvlist_array(nvroot,
3101		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3102		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3103		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3104
3105		/*
3106		 * Go through and find any spares which have since been
3107		 * repurposed as an active spare.  If this is the case, update
3108		 * their status appropriately.
3109		 */
3110		for (i = 0; i < nspares; i++) {
3111			VERIFY(nvlist_lookup_uint64(spares[i],
3112			    ZPOOL_CONFIG_GUID, &guid) == 0);
3113			if (spa_spare_exists(guid, &pool, NULL) &&
3114			    pool != 0ULL) {
3115				VERIFY(nvlist_lookup_uint64_array(
3116				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
3117				    (uint64_t **)&vs, &vsc) == 0);
3118				vs->vs_state = VDEV_STATE_CANT_OPEN;
3119				vs->vs_aux = VDEV_AUX_SPARED;
3120			}
3121		}
3122	}
3123}
3124
3125/*
3126 * Add l2cache device information to the nvlist, including vdev stats.
3127 */
3128static void
3129spa_add_l2cache(spa_t *spa, nvlist_t *config)
3130{
3131	nvlist_t **l2cache;
3132	uint_t i, j, nl2cache;
3133	nvlist_t *nvroot;
3134	uint64_t guid;
3135	vdev_t *vd;
3136	vdev_stat_t *vs;
3137	uint_t vsc;
3138
3139	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3140
3141	if (spa->spa_l2cache.sav_count == 0)
3142		return;
3143
3144	VERIFY(nvlist_lookup_nvlist(config,
3145	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3146	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3147	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3148	if (nl2cache != 0) {
3149		VERIFY(nvlist_add_nvlist_array(nvroot,
3150		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3151		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3152		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3153
3154		/*
3155		 * Update level 2 cache device stats.
3156		 */
3157
3158		for (i = 0; i < nl2cache; i++) {
3159			VERIFY(nvlist_lookup_uint64(l2cache[i],
3160			    ZPOOL_CONFIG_GUID, &guid) == 0);
3161
3162			vd = NULL;
3163			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3164				if (guid ==
3165				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3166					vd = spa->spa_l2cache.sav_vdevs[j];
3167					break;
3168				}
3169			}
3170			ASSERT(vd != NULL);
3171
3172			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3173			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3174			    == 0);
3175			vdev_get_stats(vd, vs);
3176		}
3177	}
3178}
3179
3180static void
3181spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3182{
3183	nvlist_t *features;
3184	zap_cursor_t zc;
3185	zap_attribute_t za;
3186
3187	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3188	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3189
3190	if (spa->spa_feat_for_read_obj != 0) {
3191		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3192		    spa->spa_feat_for_read_obj);
3193		    zap_cursor_retrieve(&zc, &za) == 0;
3194		    zap_cursor_advance(&zc)) {
3195			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3196			    za.za_num_integers == 1);
3197			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3198			    za.za_first_integer));
3199		}
3200		zap_cursor_fini(&zc);
3201	}
3202
3203	if (spa->spa_feat_for_write_obj != 0) {
3204		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3205		    spa->spa_feat_for_write_obj);
3206		    zap_cursor_retrieve(&zc, &za) == 0;
3207		    zap_cursor_advance(&zc)) {
3208			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3209			    za.za_num_integers == 1);
3210			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3211			    za.za_first_integer));
3212		}
3213		zap_cursor_fini(&zc);
3214	}
3215
3216	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3217	    features) == 0);
3218	nvlist_free(features);
3219}
3220
3221int
3222spa_get_stats(const char *name, nvlist_t **config,
3223    char *altroot, size_t buflen)
3224{
3225	int error;
3226	spa_t *spa;
3227
3228	*config = NULL;
3229	error = spa_open_common(name, &spa, FTAG, NULL, config);
3230
3231	if (spa != NULL) {
3232		/*
3233		 * This still leaves a window of inconsistency where the spares
3234		 * or l2cache devices could change and the config would be
3235		 * self-inconsistent.
3236		 */
3237		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3238
3239		if (*config != NULL) {
3240			uint64_t loadtimes[2];
3241
3242			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3243			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3244			VERIFY(nvlist_add_uint64_array(*config,
3245			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3246
3247			VERIFY(nvlist_add_uint64(*config,
3248			    ZPOOL_CONFIG_ERRCOUNT,
3249			    spa_get_errlog_size(spa)) == 0);
3250
3251			if (spa_suspended(spa))
3252				VERIFY(nvlist_add_uint64(*config,
3253				    ZPOOL_CONFIG_SUSPENDED,
3254				    spa->spa_failmode) == 0);
3255
3256			spa_add_spares(spa, *config);
3257			spa_add_l2cache(spa, *config);
3258			spa_add_feature_stats(spa, *config);
3259		}
3260	}
3261
3262	/*
3263	 * We want to get the alternate root even for faulted pools, so we cheat
3264	 * and call spa_lookup() directly.
3265	 */
3266	if (altroot) {
3267		if (spa == NULL) {
3268			mutex_enter(&spa_namespace_lock);
3269			spa = spa_lookup(name);
3270			if (spa)
3271				spa_altroot(spa, altroot, buflen);
3272			else
3273				altroot[0] = '\0';
3274			spa = NULL;
3275			mutex_exit(&spa_namespace_lock);
3276		} else {
3277			spa_altroot(spa, altroot, buflen);
3278		}
3279	}
3280
3281	if (spa != NULL) {
3282		spa_config_exit(spa, SCL_CONFIG, FTAG);
3283		spa_close(spa, FTAG);
3284	}
3285
3286	return (error);
3287}
3288
3289/*
3290 * Validate that the auxiliary device array is well formed.  We must have an
3291 * array of nvlists, each which describes a valid leaf vdev.  If this is an
3292 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3293 * specified, as long as they are well-formed.
3294 */
3295static int
3296spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3297    spa_aux_vdev_t *sav, const char *config, uint64_t version,
3298    vdev_labeltype_t label)
3299{
3300	nvlist_t **dev;
3301	uint_t i, ndev;
3302	vdev_t *vd;
3303	int error;
3304
3305	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3306
3307	/*
3308	 * It's acceptable to have no devs specified.
3309	 */
3310	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3311		return (0);
3312
3313	if (ndev == 0)
3314		return (SET_ERROR(EINVAL));
3315
3316	/*
3317	 * Make sure the pool is formatted with a version that supports this
3318	 * device type.
3319	 */
3320	if (spa_version(spa) < version)
3321		return (SET_ERROR(ENOTSUP));
3322
3323	/*
3324	 * Set the pending device list so we correctly handle device in-use
3325	 * checking.
3326	 */
3327	sav->sav_pending = dev;
3328	sav->sav_npending = ndev;
3329
3330	for (i = 0; i < ndev; i++) {
3331		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3332		    mode)) != 0)
3333			goto out;
3334
3335		if (!vd->vdev_ops->vdev_op_leaf) {
3336			vdev_free(vd);
3337			error = SET_ERROR(EINVAL);
3338			goto out;
3339		}
3340
3341		/*
3342		 * The L2ARC currently only supports disk devices in
3343		 * kernel context.  For user-level testing, we allow it.
3344		 */
3345#ifdef _KERNEL
3346		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3347		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3348			error = SET_ERROR(ENOTBLK);
3349			vdev_free(vd);
3350			goto out;
3351		}
3352#endif
3353		vd->vdev_top = vd;
3354
3355		if ((error = vdev_open(vd)) == 0 &&
3356		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
3357			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3358			    vd->vdev_guid) == 0);
3359		}
3360
3361		vdev_free(vd);
3362
3363		if (error &&
3364		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3365			goto out;
3366		else
3367			error = 0;
3368	}
3369
3370out:
3371	sav->sav_pending = NULL;
3372	sav->sav_npending = 0;
3373	return (error);
3374}
3375
3376static int
3377spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3378{
3379	int error;
3380
3381	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3382
3383	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3384	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3385	    VDEV_LABEL_SPARE)) != 0) {
3386		return (error);
3387	}
3388
3389	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3390	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3391	    VDEV_LABEL_L2CACHE));
3392}
3393
3394static void
3395spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3396    const char *config)
3397{
3398	int i;
3399
3400	if (sav->sav_config != NULL) {
3401		nvlist_t **olddevs;
3402		uint_t oldndevs;
3403		nvlist_t **newdevs;
3404
3405		/*
3406		 * Generate new dev list by concatentating with the
3407		 * current dev list.
3408		 */
3409		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3410		    &olddevs, &oldndevs) == 0);
3411
3412		newdevs = kmem_alloc(sizeof (void *) *
3413		    (ndevs + oldndevs), KM_SLEEP);
3414		for (i = 0; i < oldndevs; i++)
3415			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3416			    KM_SLEEP) == 0);
3417		for (i = 0; i < ndevs; i++)
3418			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3419			    KM_SLEEP) == 0);
3420
3421		VERIFY(nvlist_remove(sav->sav_config, config,
3422		    DATA_TYPE_NVLIST_ARRAY) == 0);
3423
3424		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3425		    config, newdevs, ndevs + oldndevs) == 0);
3426		for (i = 0; i < oldndevs + ndevs; i++)
3427			nvlist_free(newdevs[i]);
3428		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3429	} else {
3430		/*
3431		 * Generate a new dev list.
3432		 */
3433		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3434		    KM_SLEEP) == 0);
3435		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3436		    devs, ndevs) == 0);
3437	}
3438}
3439
3440/*
3441 * Stop and drop level 2 ARC devices
3442 */
3443void
3444spa_l2cache_drop(spa_t *spa)
3445{
3446	vdev_t *vd;
3447	int i;
3448	spa_aux_vdev_t *sav = &spa->spa_l2cache;
3449
3450	for (i = 0; i < sav->sav_count; i++) {
3451		uint64_t pool;
3452
3453		vd = sav->sav_vdevs[i];
3454		ASSERT(vd != NULL);
3455
3456		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3457		    pool != 0ULL && l2arc_vdev_present(vd))
3458			l2arc_remove_vdev(vd);
3459	}
3460}
3461
3462/*
3463 * Pool Creation
3464 */
3465int
3466spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3467    nvlist_t *zplprops)
3468{
3469	spa_t *spa;
3470	char *altroot = NULL;
3471	vdev_t *rvd;
3472	dsl_pool_t *dp;
3473	dmu_tx_t *tx;
3474	int error = 0;
3475	uint64_t txg = TXG_INITIAL;
3476	nvlist_t **spares, **l2cache;
3477	uint_t nspares, nl2cache;
3478	uint64_t version, obj;
3479	boolean_t has_features;
3480
3481	/*
3482	 * If this pool already exists, return failure.
3483	 */
3484	mutex_enter(&spa_namespace_lock);
3485	if (spa_lookup(pool) != NULL) {
3486		mutex_exit(&spa_namespace_lock);
3487		return (SET_ERROR(EEXIST));
3488	}
3489
3490	/*
3491	 * Allocate a new spa_t structure.
3492	 */
3493	(void) nvlist_lookup_string(props,
3494	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3495	spa = spa_add(pool, NULL, altroot);
3496	spa_activate(spa, spa_mode_global);
3497
3498	if (props && (error = spa_prop_validate(spa, props))) {
3499		spa_deactivate(spa);
3500		spa_remove(spa);
3501		mutex_exit(&spa_namespace_lock);
3502		return (error);
3503	}
3504
3505	has_features = B_FALSE;
3506	for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3507	    elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3508		if (zpool_prop_feature(nvpair_name(elem)))
3509			has_features = B_TRUE;
3510	}
3511
3512	if (has_features || nvlist_lookup_uint64(props,
3513	    zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3514		version = SPA_VERSION;
3515	}
3516	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3517
3518	spa->spa_first_txg = txg;
3519	spa->spa_uberblock.ub_txg = txg - 1;
3520	spa->spa_uberblock.ub_version = version;
3521	spa->spa_ubsync = spa->spa_uberblock;
3522
3523	/*
3524	 * Create "The Godfather" zio to hold all async IOs
3525	 */
3526	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
3527	    KM_SLEEP);
3528	for (int i = 0; i < max_ncpus; i++) {
3529		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
3530		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
3531		    ZIO_FLAG_GODFATHER);
3532	}
3533
3534	/*
3535	 * Create the root vdev.
3536	 */
3537	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3538
3539	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3540
3541	ASSERT(error != 0 || rvd != NULL);
3542	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3543
3544	if (error == 0 && !zfs_allocatable_devs(nvroot))
3545		error = SET_ERROR(EINVAL);
3546
3547	if (error == 0 &&
3548	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3549	    (error = spa_validate_aux(spa, nvroot, txg,
3550	    VDEV_ALLOC_ADD)) == 0) {
3551		for (int c = 0; c < rvd->vdev_children; c++) {
3552			vdev_metaslab_set_size(rvd->vdev_child[c]);
3553			vdev_expand(rvd->vdev_child[c], txg);
3554		}
3555	}
3556
3557	spa_config_exit(spa, SCL_ALL, FTAG);
3558
3559	if (error != 0) {
3560		spa_unload(spa);
3561		spa_deactivate(spa);
3562		spa_remove(spa);
3563		mutex_exit(&spa_namespace_lock);
3564		return (error);
3565	}
3566
3567	/*
3568	 * Get the list of spares, if specified.
3569	 */
3570	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3571	    &spares, &nspares) == 0) {
3572		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3573		    KM_SLEEP) == 0);
3574		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3575		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3576		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3577		spa_load_spares(spa);
3578		spa_config_exit(spa, SCL_ALL, FTAG);
3579		spa->spa_spares.sav_sync = B_TRUE;
3580	}
3581
3582	/*
3583	 * Get the list of level 2 cache devices, if specified.
3584	 */
3585	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3586	    &l2cache, &nl2cache) == 0) {
3587		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3588		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3589		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3590		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3591		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3592		spa_load_l2cache(spa);
3593		spa_config_exit(spa, SCL_ALL, FTAG);
3594		spa->spa_l2cache.sav_sync = B_TRUE;
3595	}
3596
3597	spa->spa_is_initializing = B_TRUE;
3598	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3599	spa->spa_meta_objset = dp->dp_meta_objset;
3600	spa->spa_is_initializing = B_FALSE;
3601
3602	/*
3603	 * Create DDTs (dedup tables).
3604	 */
3605	ddt_create(spa);
3606
3607	spa_update_dspace(spa);
3608
3609	tx = dmu_tx_create_assigned(dp, txg);
3610
3611	/*
3612	 * Create the pool config object.
3613	 */
3614	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3615	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3616	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3617
3618	if (zap_add(spa->spa_meta_objset,
3619	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3620	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3621		cmn_err(CE_PANIC, "failed to add pool config");
3622	}
3623
3624	if (spa_version(spa) >= SPA_VERSION_FEATURES)
3625		spa_feature_create_zap_objects(spa, tx);
3626
3627	if (zap_add(spa->spa_meta_objset,
3628	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3629	    sizeof (uint64_t), 1, &version, tx) != 0) {
3630		cmn_err(CE_PANIC, "failed to add pool version");
3631	}
3632
3633	/* Newly created pools with the right version are always deflated. */
3634	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3635		spa->spa_deflate = TRUE;
3636		if (zap_add(spa->spa_meta_objset,
3637		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3638		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3639			cmn_err(CE_PANIC, "failed to add deflate");
3640		}
3641	}
3642
3643	/*
3644	 * Create the deferred-free bpobj.  Turn off compression
3645	 * because sync-to-convergence takes longer if the blocksize
3646	 * keeps changing.
3647	 */
3648	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3649	dmu_object_set_compress(spa->spa_meta_objset, obj,
3650	    ZIO_COMPRESS_OFF, tx);
3651	if (zap_add(spa->spa_meta_objset,
3652	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3653	    sizeof (uint64_t), 1, &obj, tx) != 0) {
3654		cmn_err(CE_PANIC, "failed to add bpobj");
3655	}
3656	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3657	    spa->spa_meta_objset, obj));
3658
3659	/*
3660	 * Create the pool's history object.
3661	 */
3662	if (version >= SPA_VERSION_ZPOOL_HISTORY)
3663		spa_history_create_obj(spa, tx);
3664
3665	/*
3666	 * Set pool properties.
3667	 */
3668	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3669	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3670	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3671	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3672
3673	if (props != NULL) {
3674		spa_configfile_set(spa, props, B_FALSE);
3675		spa_sync_props(props, tx);
3676	}
3677
3678	dmu_tx_commit(tx);
3679
3680	spa->spa_sync_on = B_TRUE;
3681	txg_sync_start(spa->spa_dsl_pool);
3682
3683	/*
3684	 * We explicitly wait for the first transaction to complete so that our
3685	 * bean counters are appropriately updated.
3686	 */
3687	txg_wait_synced(spa->spa_dsl_pool, txg);
3688
3689	spa_config_sync(spa, B_FALSE, B_TRUE);
3690
3691	spa_history_log_version(spa, "create");
3692
3693	/*
3694	 * Don't count references from objsets that are already closed
3695	 * and are making their way through the eviction process.
3696	 */
3697	spa_evicting_os_wait(spa);
3698	spa->spa_minref = refcount_count(&spa->spa_refcount);
3699
3700	mutex_exit(&spa_namespace_lock);
3701
3702	return (0);
3703}
3704
3705#ifdef _KERNEL
3706/*
3707 * Get the root pool information from the root disk, then import the root pool
3708 * during the system boot up time.
3709 */
3710extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3711
3712static nvlist_t *
3713spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3714{
3715	nvlist_t *config;
3716	nvlist_t *nvtop, *nvroot;
3717	uint64_t pgid;
3718
3719	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3720		return (NULL);
3721
3722	/*
3723	 * Add this top-level vdev to the child array.
3724	 */
3725	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3726	    &nvtop) == 0);
3727	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3728	    &pgid) == 0);
3729	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3730
3731	/*
3732	 * Put this pool's top-level vdevs into a root vdev.
3733	 */
3734	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3735	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3736	    VDEV_TYPE_ROOT) == 0);
3737	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3738	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3739	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3740	    &nvtop, 1) == 0);
3741
3742	/*
3743	 * Replace the existing vdev_tree with the new root vdev in
3744	 * this pool's configuration (remove the old, add the new).
3745	 */
3746	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3747	nvlist_free(nvroot);
3748	return (config);
3749}
3750
3751/*
3752 * Walk the vdev tree and see if we can find a device with "better"
3753 * configuration. A configuration is "better" if the label on that
3754 * device has a more recent txg.
3755 */
3756static void
3757spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3758{
3759	for (int c = 0; c < vd->vdev_children; c++)
3760		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3761
3762	if (vd->vdev_ops->vdev_op_leaf) {
3763		nvlist_t *label;
3764		uint64_t label_txg;
3765
3766		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3767		    &label) != 0)
3768			return;
3769
3770		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3771		    &label_txg) == 0);
3772
3773		/*
3774		 * Do we have a better boot device?
3775		 */
3776		if (label_txg > *txg) {
3777			*txg = label_txg;
3778			*avd = vd;
3779		}
3780		nvlist_free(label);
3781	}
3782}
3783
3784/*
3785 * Import a root pool.
3786 *
3787 * For x86. devpath_list will consist of devid and/or physpath name of
3788 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3789 * The GRUB "findroot" command will return the vdev we should boot.
3790 *
3791 * For Sparc, devpath_list consists the physpath name of the booting device
3792 * no matter the rootpool is a single device pool or a mirrored pool.
3793 * e.g.
3794 *	"/pci@1f,0/ide@d/disk@0,0:a"
3795 */
3796int
3797spa_import_rootpool(char *devpath, char *devid)
3798{
3799	spa_t *spa;
3800	vdev_t *rvd, *bvd, *avd = NULL;
3801	nvlist_t *config, *nvtop;
3802	uint64_t guid, txg;
3803	char *pname;
3804	int error;
3805
3806	/*
3807	 * Read the label from the boot device and generate a configuration.
3808	 */
3809	config = spa_generate_rootconf(devpath, devid, &guid);
3810#if defined(_OBP) && defined(_KERNEL)
3811	if (config == NULL) {
3812		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3813			/* iscsi boot */
3814			get_iscsi_bootpath_phy(devpath);
3815			config = spa_generate_rootconf(devpath, devid, &guid);
3816		}
3817	}
3818#endif
3819	if (config == NULL) {
3820		cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3821		    devpath);
3822		return (SET_ERROR(EIO));
3823	}
3824
3825	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3826	    &pname) == 0);
3827	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3828
3829	mutex_enter(&spa_namespace_lock);
3830	if ((spa = spa_lookup(pname)) != NULL) {
3831		/*
3832		 * Remove the existing root pool from the namespace so that we
3833		 * can replace it with the correct config we just read in.
3834		 */
3835		spa_remove(spa);
3836	}
3837
3838	spa = spa_add(pname, config, NULL);
3839	spa->spa_is_root = B_TRUE;
3840	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3841
3842	/*
3843	 * Build up a vdev tree based on the boot device's label config.
3844	 */
3845	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3846	    &nvtop) == 0);
3847	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3848	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3849	    VDEV_ALLOC_ROOTPOOL);
3850	spa_config_exit(spa, SCL_ALL, FTAG);
3851	if (error) {
3852		mutex_exit(&spa_namespace_lock);
3853		nvlist_free(config);
3854		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3855		    pname);
3856		return (error);
3857	}
3858
3859	/*
3860	 * Get the boot vdev.
3861	 */
3862	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3863		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3864		    (u_longlong_t)guid);
3865		error = SET_ERROR(ENOENT);
3866		goto out;
3867	}
3868
3869	/*
3870	 * Determine if there is a better boot device.
3871	 */
3872	avd = bvd;
3873	spa_alt_rootvdev(rvd, &avd, &txg);
3874	if (avd != bvd) {
3875		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3876		    "try booting from '%s'", avd->vdev_path);
3877		error = SET_ERROR(EINVAL);
3878		goto out;
3879	}
3880
3881	/*
3882	 * If the boot device is part of a spare vdev then ensure that
3883	 * we're booting off the active spare.
3884	 */
3885	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3886	    !bvd->vdev_isspare) {
3887		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3888		    "try booting from '%s'",
3889		    bvd->vdev_parent->
3890		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3891		error = SET_ERROR(EINVAL);
3892		goto out;
3893	}
3894
3895	error = 0;
3896out:
3897	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3898	vdev_free(rvd);
3899	spa_config_exit(spa, SCL_ALL, FTAG);
3900	mutex_exit(&spa_namespace_lock);
3901
3902	nvlist_free(config);
3903	return (error);
3904}
3905
3906#endif
3907
3908/*
3909 * Import a non-root pool into the system.
3910 */
3911int
3912spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3913{
3914	spa_t *spa;
3915	char *altroot = NULL;
3916	spa_load_state_t state = SPA_LOAD_IMPORT;
3917	zpool_rewind_policy_t policy;
3918	uint64_t mode = spa_mode_global;
3919	uint64_t readonly = B_FALSE;
3920	int error;
3921	nvlist_t *nvroot;
3922	nvlist_t **spares, **l2cache;
3923	uint_t nspares, nl2cache;
3924
3925	/*
3926	 * If a pool with this name exists, return failure.
3927	 */
3928	mutex_enter(&spa_namespace_lock);
3929	if (spa_lookup(pool) != NULL) {
3930		mutex_exit(&spa_namespace_lock);
3931		return (SET_ERROR(EEXIST));
3932	}
3933
3934	/*
3935	 * Create and initialize the spa structure.
3936	 */
3937	(void) nvlist_lookup_string(props,
3938	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3939	(void) nvlist_lookup_uint64(props,
3940	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3941	if (readonly)
3942		mode = FREAD;
3943	spa = spa_add(pool, config, altroot);
3944	spa->spa_import_flags = flags;
3945
3946	/*
3947	 * Verbatim import - Take a pool and insert it into the namespace
3948	 * as if it had been loaded at boot.
3949	 */
3950	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3951		if (props != NULL)
3952			spa_configfile_set(spa, props, B_FALSE);
3953
3954		spa_config_sync(spa, B_FALSE, B_TRUE);
3955
3956		mutex_exit(&spa_namespace_lock);
3957		return (0);
3958	}
3959
3960	spa_activate(spa, mode);
3961
3962	/*
3963	 * Don't start async tasks until we know everything is healthy.
3964	 */
3965	spa_async_suspend(spa);
3966
3967	zpool_get_rewind_policy(config, &policy);
3968	if (policy.zrp_request & ZPOOL_DO_REWIND)
3969		state = SPA_LOAD_RECOVER;
3970
3971	/*
3972	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3973	 * because the user-supplied config is actually the one to trust when
3974	 * doing an import.
3975	 */
3976	if (state != SPA_LOAD_RECOVER)
3977		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3978
3979	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3980	    policy.zrp_request);
3981
3982	/*
3983	 * Propagate anything learned while loading the pool and pass it
3984	 * back to caller (i.e. rewind info, missing devices, etc).
3985	 */
3986	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3987	    spa->spa_load_info) == 0);
3988
3989	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3990	/*
3991	 * Toss any existing sparelist, as it doesn't have any validity
3992	 * anymore, and conflicts with spa_has_spare().
3993	 */
3994	if (spa->spa_spares.sav_config) {
3995		nvlist_free(spa->spa_spares.sav_config);
3996		spa->spa_spares.sav_config = NULL;
3997		spa_load_spares(spa);
3998	}
3999	if (spa->spa_l2cache.sav_config) {
4000		nvlist_free(spa->spa_l2cache.sav_config);
4001		spa->spa_l2cache.sav_config = NULL;
4002		spa_load_l2cache(spa);
4003	}
4004
4005	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4006	    &nvroot) == 0);
4007	if (error == 0)
4008		error = spa_validate_aux(spa, nvroot, -1ULL,
4009		    VDEV_ALLOC_SPARE);
4010	if (error == 0)
4011		error = spa_validate_aux(spa, nvroot, -1ULL,
4012		    VDEV_ALLOC_L2CACHE);
4013	spa_config_exit(spa, SCL_ALL, FTAG);
4014
4015	if (props != NULL)
4016		spa_configfile_set(spa, props, B_FALSE);
4017
4018	if (error != 0 || (props && spa_writeable(spa) &&
4019	    (error = spa_prop_set(spa, props)))) {
4020		spa_unload(spa);
4021		spa_deactivate(spa);
4022		spa_remove(spa);
4023		mutex_exit(&spa_namespace_lock);
4024		return (error);
4025	}
4026
4027	spa_async_resume(spa);
4028
4029	/*
4030	 * Override any spares and level 2 cache devices as specified by
4031	 * the user, as these may have correct device names/devids, etc.
4032	 */
4033	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4034	    &spares, &nspares) == 0) {
4035		if (spa->spa_spares.sav_config)
4036			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4037			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4038		else
4039			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4040			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4041		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4042		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4043		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4044		spa_load_spares(spa);
4045		spa_config_exit(spa, SCL_ALL, FTAG);
4046		spa->spa_spares.sav_sync = B_TRUE;
4047	}
4048	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4049	    &l2cache, &nl2cache) == 0) {
4050		if (spa->spa_l2cache.sav_config)
4051			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4052			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4053		else
4054			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4055			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4056		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4057		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4058		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4059		spa_load_l2cache(spa);
4060		spa_config_exit(spa, SCL_ALL, FTAG);
4061		spa->spa_l2cache.sav_sync = B_TRUE;
4062	}
4063
4064	/*
4065	 * Check for any removed devices.
4066	 */
4067	if (spa->spa_autoreplace) {
4068		spa_aux_check_removed(&spa->spa_spares);
4069		spa_aux_check_removed(&spa->spa_l2cache);
4070	}
4071
4072	if (spa_writeable(spa)) {
4073		/*
4074		 * Update the config cache to include the newly-imported pool.
4075		 */
4076		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4077	}
4078
4079	/*
4080	 * It's possible that the pool was expanded while it was exported.
4081	 * We kick off an async task to handle this for us.
4082	 */
4083	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4084
4085	mutex_exit(&spa_namespace_lock);
4086	spa_history_log_version(spa, "import");
4087
4088	return (0);
4089}
4090
4091nvlist_t *
4092spa_tryimport(nvlist_t *tryconfig)
4093{
4094	nvlist_t *config = NULL;
4095	char *poolname;
4096	spa_t *spa;
4097	uint64_t state;
4098	int error;
4099
4100	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4101		return (NULL);
4102
4103	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4104		return (NULL);
4105
4106	/*
4107	 * Create and initialize the spa structure.
4108	 */
4109	mutex_enter(&spa_namespace_lock);
4110	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4111	spa_activate(spa, FREAD);
4112
4113	/*
4114	 * Pass off the heavy lifting to spa_load().
4115	 * Pass TRUE for mosconfig because the user-supplied config
4116	 * is actually the one to trust when doing an import.
4117	 */
4118	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4119
4120	/*
4121	 * If 'tryconfig' was at least parsable, return the current config.
4122	 */
4123	if (spa->spa_root_vdev != NULL) {
4124		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4125		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4126		    poolname) == 0);
4127		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4128		    state) == 0);
4129		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4130		    spa->spa_uberblock.ub_timestamp) == 0);
4131		VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4132		    spa->spa_load_info) == 0);
4133
4134		/*
4135		 * If the bootfs property exists on this pool then we
4136		 * copy it out so that external consumers can tell which
4137		 * pools are bootable.
4138		 */
4139		if ((!error || error == EEXIST) && spa->spa_bootfs) {
4140			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4141
4142			/*
4143			 * We have to play games with the name since the
4144			 * pool was opened as TRYIMPORT_NAME.
4145			 */
4146			if (dsl_dsobj_to_dsname(spa_name(spa),
4147			    spa->spa_bootfs, tmpname) == 0) {
4148				char *cp;
4149				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4150
4151				cp = strchr(tmpname, '/');
4152				if (cp == NULL) {
4153					(void) strlcpy(dsname, tmpname,
4154					    MAXPATHLEN);
4155				} else {
4156					(void) snprintf(dsname, MAXPATHLEN,
4157					    "%s/%s", poolname, ++cp);
4158				}
4159				VERIFY(nvlist_add_string(config,
4160				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4161				kmem_free(dsname, MAXPATHLEN);
4162			}
4163			kmem_free(tmpname, MAXPATHLEN);
4164		}
4165
4166		/*
4167		 * Add the list of hot spares and level 2 cache devices.
4168		 */
4169		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4170		spa_add_spares(spa, config);
4171		spa_add_l2cache(spa, config);
4172		spa_config_exit(spa, SCL_CONFIG, FTAG);
4173	}
4174
4175	spa_unload(spa);
4176	spa_deactivate(spa);
4177	spa_remove(spa);
4178	mutex_exit(&spa_namespace_lock);
4179
4180	return (config);
4181}
4182
4183/*
4184 * Pool export/destroy
4185 *
4186 * The act of destroying or exporting a pool is very simple.  We make sure there
4187 * is no more pending I/O and any references to the pool are gone.  Then, we
4188 * update the pool state and sync all the labels to disk, removing the
4189 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4190 * we don't sync the labels or remove the configuration cache.
4191 */
4192static int
4193spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4194    boolean_t force, boolean_t hardforce)
4195{
4196	spa_t *spa;
4197
4198	if (oldconfig)
4199		*oldconfig = NULL;
4200
4201	if (!(spa_mode_global & FWRITE))
4202		return (SET_ERROR(EROFS));
4203
4204	mutex_enter(&spa_namespace_lock);
4205	if ((spa = spa_lookup(pool)) == NULL) {
4206		mutex_exit(&spa_namespace_lock);
4207		return (SET_ERROR(ENOENT));
4208	}
4209
4210	/*
4211	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4212	 * reacquire the namespace lock, and see if we can export.
4213	 */
4214	spa_open_ref(spa, FTAG);
4215	mutex_exit(&spa_namespace_lock);
4216	spa_async_suspend(spa);
4217	mutex_enter(&spa_namespace_lock);
4218	spa_close(spa, FTAG);
4219
4220	/*
4221	 * The pool will be in core if it's openable,
4222	 * in which case we can modify its state.
4223	 */
4224	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4225		/*
4226		 * Objsets may be open only because they're dirty, so we
4227		 * have to force it to sync before checking spa_refcnt.
4228		 */
4229		txg_wait_synced(spa->spa_dsl_pool, 0);
4230		spa_evicting_os_wait(spa);
4231
4232		/*
4233		 * A pool cannot be exported or destroyed if there are active
4234		 * references.  If we are resetting a pool, allow references by
4235		 * fault injection handlers.
4236		 */
4237		if (!spa_refcount_zero(spa) ||
4238		    (spa->spa_inject_ref != 0 &&
4239		    new_state != POOL_STATE_UNINITIALIZED)) {
4240			spa_async_resume(spa);
4241			mutex_exit(&spa_namespace_lock);
4242			return (SET_ERROR(EBUSY));
4243		}
4244
4245		/*
4246		 * A pool cannot be exported if it has an active shared spare.
4247		 * This is to prevent other pools stealing the active spare
4248		 * from an exported pool. At user's own will, such pool can
4249		 * be forcedly exported.
4250		 */
4251		if (!force && new_state == POOL_STATE_EXPORTED &&
4252		    spa_has_active_shared_spare(spa)) {
4253			spa_async_resume(spa);
4254			mutex_exit(&spa_namespace_lock);
4255			return (SET_ERROR(EXDEV));
4256		}
4257
4258		/*
4259		 * We want this to be reflected on every label,
4260		 * so mark them all dirty.  spa_unload() will do the
4261		 * final sync that pushes these changes out.
4262		 */
4263		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4264			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4265			spa->spa_state = new_state;
4266			spa->spa_final_txg = spa_last_synced_txg(spa) +
4267			    TXG_DEFER_SIZE + 1;
4268			vdev_config_dirty(spa->spa_root_vdev);
4269			spa_config_exit(spa, SCL_ALL, FTAG);
4270		}
4271	}
4272
4273	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4274
4275	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4276		spa_unload(spa);
4277		spa_deactivate(spa);
4278	}
4279
4280	if (oldconfig && spa->spa_config)
4281		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4282
4283	if (new_state != POOL_STATE_UNINITIALIZED) {
4284		if (!hardforce)
4285			spa_config_sync(spa, B_TRUE, B_TRUE);
4286		spa_remove(spa);
4287	}
4288	mutex_exit(&spa_namespace_lock);
4289
4290	return (0);
4291}
4292
4293/*
4294 * Destroy a storage pool.
4295 */
4296int
4297spa_destroy(char *pool)
4298{
4299	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4300	    B_FALSE, B_FALSE));
4301}
4302
4303/*
4304 * Export a storage pool.
4305 */
4306int
4307spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4308    boolean_t hardforce)
4309{
4310	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4311	    force, hardforce));
4312}
4313
4314/*
4315 * Similar to spa_export(), this unloads the spa_t without actually removing it
4316 * from the namespace in any way.
4317 */
4318int
4319spa_reset(char *pool)
4320{
4321	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4322	    B_FALSE, B_FALSE));
4323}
4324
4325/*
4326 * ==========================================================================
4327 * Device manipulation
4328 * ==========================================================================
4329 */
4330
4331/*
4332 * Add a device to a storage pool.
4333 */
4334int
4335spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4336{
4337	uint64_t txg, id;
4338	int error;
4339	vdev_t *rvd = spa->spa_root_vdev;
4340	vdev_t *vd, *tvd;
4341	nvlist_t **spares, **l2cache;
4342	uint_t nspares, nl2cache;
4343
4344	ASSERT(spa_writeable(spa));
4345
4346	txg = spa_vdev_enter(spa);
4347
4348	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4349	    VDEV_ALLOC_ADD)) != 0)
4350		return (spa_vdev_exit(spa, NULL, txg, error));
4351
4352	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
4353
4354	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4355	    &nspares) != 0)
4356		nspares = 0;
4357
4358	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4359	    &nl2cache) != 0)
4360		nl2cache = 0;
4361
4362	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4363		return (spa_vdev_exit(spa, vd, txg, EINVAL));
4364
4365	if (vd->vdev_children != 0 &&
4366	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
4367		return (spa_vdev_exit(spa, vd, txg, error));
4368
4369	/*
4370	 * We must validate the spares and l2cache devices after checking the
4371	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
4372	 */
4373	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4374		return (spa_vdev_exit(spa, vd, txg, error));
4375
4376	/*
4377	 * Transfer each new top-level vdev from vd to rvd.
4378	 */
4379	for (int c = 0; c < vd->vdev_children; c++) {
4380
4381		/*
4382		 * Set the vdev id to the first hole, if one exists.
4383		 */
4384		for (id = 0; id < rvd->vdev_children; id++) {
4385			if (rvd->vdev_child[id]->vdev_ishole) {
4386				vdev_free(rvd->vdev_child[id]);
4387				break;
4388			}
4389		}
4390		tvd = vd->vdev_child[c];
4391		vdev_remove_child(vd, tvd);
4392		tvd->vdev_id = id;
4393		vdev_add_child(rvd, tvd);
4394		vdev_config_dirty(tvd);
4395	}
4396
4397	if (nspares != 0) {
4398		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4399		    ZPOOL_CONFIG_SPARES);
4400		spa_load_spares(spa);
4401		spa->spa_spares.sav_sync = B_TRUE;
4402	}
4403
4404	if (nl2cache != 0) {
4405		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4406		    ZPOOL_CONFIG_L2CACHE);
4407		spa_load_l2cache(spa);
4408		spa->spa_l2cache.sav_sync = B_TRUE;
4409	}
4410
4411	/*
4412	 * We have to be careful when adding new vdevs to an existing pool.
4413	 * If other threads start allocating from these vdevs before we
4414	 * sync the config cache, and we lose power, then upon reboot we may
4415	 * fail to open the pool because there are DVAs that the config cache
4416	 * can't translate.  Therefore, we first add the vdevs without
4417	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4418	 * and then let spa_config_update() initialize the new metaslabs.
4419	 *
4420	 * spa_load() checks for added-but-not-initialized vdevs, so that
4421	 * if we lose power at any point in this sequence, the remaining
4422	 * steps will be completed the next time we load the pool.
4423	 */
4424	(void) spa_vdev_exit(spa, vd, txg, 0);
4425
4426	mutex_enter(&spa_namespace_lock);
4427	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4428	mutex_exit(&spa_namespace_lock);
4429
4430	return (0);
4431}
4432
4433/*
4434 * Attach a device to a mirror.  The arguments are the path to any device
4435 * in the mirror, and the nvroot for the new device.  If the path specifies
4436 * a device that is not mirrored, we automatically insert the mirror vdev.
4437 *
4438 * If 'replacing' is specified, the new device is intended to replace the
4439 * existing device; in this case the two devices are made into their own
4440 * mirror using the 'replacing' vdev, which is functionally identical to
4441 * the mirror vdev (it actually reuses all the same ops) but has a few
4442 * extra rules: you can't attach to it after it's been created, and upon
4443 * completion of resilvering, the first disk (the one being replaced)
4444 * is automatically detached.
4445 */
4446int
4447spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4448{
4449	uint64_t txg, dtl_max_txg;
4450	vdev_t *rvd = spa->spa_root_vdev;
4451	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4452	vdev_ops_t *pvops;
4453	char *oldvdpath, *newvdpath;
4454	int newvd_isspare;
4455	int error;
4456
4457	ASSERT(spa_writeable(spa));
4458
4459	txg = spa_vdev_enter(spa);
4460
4461	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4462
4463	if (oldvd == NULL)
4464		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4465
4466	if (!oldvd->vdev_ops->vdev_op_leaf)
4467		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4468
4469	pvd = oldvd->vdev_parent;
4470
4471	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4472	    VDEV_ALLOC_ATTACH)) != 0)
4473		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4474
4475	if (newrootvd->vdev_children != 1)
4476		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4477
4478	newvd = newrootvd->vdev_child[0];
4479
4480	if (!newvd->vdev_ops->vdev_op_leaf)
4481		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4482
4483	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4484		return (spa_vdev_exit(spa, newrootvd, txg, error));
4485
4486	/*
4487	 * Spares can't replace logs
4488	 */
4489	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4490		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4491
4492	if (!replacing) {
4493		/*
4494		 * For attach, the only allowable parent is a mirror or the root
4495		 * vdev.
4496		 */
4497		if (pvd->vdev_ops != &vdev_mirror_ops &&
4498		    pvd->vdev_ops != &vdev_root_ops)
4499			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4500
4501		pvops = &vdev_mirror_ops;
4502	} else {
4503		/*
4504		 * Active hot spares can only be replaced by inactive hot
4505		 * spares.
4506		 */
4507		if (pvd->vdev_ops == &vdev_spare_ops &&
4508		    oldvd->vdev_isspare &&
4509		    !spa_has_spare(spa, newvd->vdev_guid))
4510			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4511
4512		/*
4513		 * If the source is a hot spare, and the parent isn't already a
4514		 * spare, then we want to create a new hot spare.  Otherwise, we
4515		 * want to create a replacing vdev.  The user is not allowed to
4516		 * attach to a spared vdev child unless the 'isspare' state is
4517		 * the same (spare replaces spare, non-spare replaces
4518		 * non-spare).
4519		 */
4520		if (pvd->vdev_ops == &vdev_replacing_ops &&
4521		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4522			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4523		} else if (pvd->vdev_ops == &vdev_spare_ops &&
4524		    newvd->vdev_isspare != oldvd->vdev_isspare) {
4525			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4526		}
4527
4528		if (newvd->vdev_isspare)
4529			pvops = &vdev_spare_ops;
4530		else
4531			pvops = &vdev_replacing_ops;
4532	}
4533
4534	/*
4535	 * Make sure the new device is big enough.
4536	 */
4537	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4538		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4539
4540	/*
4541	 * The new device cannot have a higher alignment requirement
4542	 * than the top-level vdev.
4543	 */
4544	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4545		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4546
4547	/*
4548	 * If this is an in-place replacement, update oldvd's path and devid
4549	 * to make it distinguishable from newvd, and unopenable from now on.
4550	 */
4551	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4552		spa_strfree(oldvd->vdev_path);
4553		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4554		    KM_SLEEP);
4555		(void) sprintf(oldvd->vdev_path, "%s/%s",
4556		    newvd->vdev_path, "old");
4557		if (oldvd->vdev_devid != NULL) {
4558			spa_strfree(oldvd->vdev_devid);
4559			oldvd->vdev_devid = NULL;
4560		}
4561	}
4562
4563	/* mark the device being resilvered */
4564	newvd->vdev_resilver_txg = txg;
4565
4566	/*
4567	 * If the parent is not a mirror, or if we're replacing, insert the new
4568	 * mirror/replacing/spare vdev above oldvd.
4569	 */
4570	if (pvd->vdev_ops != pvops)
4571		pvd = vdev_add_parent(oldvd, pvops);
4572
4573	ASSERT(pvd->vdev_top->vdev_parent == rvd);
4574	ASSERT(pvd->vdev_ops == pvops);
4575	ASSERT(oldvd->vdev_parent == pvd);
4576
4577	/*
4578	 * Extract the new device from its root and add it to pvd.
4579	 */
4580	vdev_remove_child(newrootvd, newvd);
4581	newvd->vdev_id = pvd->vdev_children;
4582	newvd->vdev_crtxg = oldvd->vdev_crtxg;
4583	vdev_add_child(pvd, newvd);
4584
4585	tvd = newvd->vdev_top;
4586	ASSERT(pvd->vdev_top == tvd);
4587	ASSERT(tvd->vdev_parent == rvd);
4588
4589	vdev_config_dirty(tvd);
4590
4591	/*
4592	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4593	 * for any dmu_sync-ed blocks.  It will propagate upward when
4594	 * spa_vdev_exit() calls vdev_dtl_reassess().
4595	 */
4596	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4597
4598	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4599	    dtl_max_txg - TXG_INITIAL);
4600
4601	if (newvd->vdev_isspare) {
4602		spa_spare_activate(newvd);
4603		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4604	}
4605
4606	oldvdpath = spa_strdup(oldvd->vdev_path);
4607	newvdpath = spa_strdup(newvd->vdev_path);
4608	newvd_isspare = newvd->vdev_isspare;
4609
4610	/*
4611	 * Mark newvd's DTL dirty in this txg.
4612	 */
4613	vdev_dirty(tvd, VDD_DTL, newvd, txg);
4614
4615	/*
4616	 * Schedule the resilver to restart in the future. We do this to
4617	 * ensure that dmu_sync-ed blocks have been stitched into the
4618	 * respective datasets.
4619	 */
4620	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4621
4622	/*
4623	 * Commit the config
4624	 */
4625	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4626
4627	spa_history_log_internal(spa, "vdev attach", NULL,
4628	    "%s vdev=%s %s vdev=%s",
4629	    replacing && newvd_isspare ? "spare in" :
4630	    replacing ? "replace" : "attach", newvdpath,
4631	    replacing ? "for" : "to", oldvdpath);
4632
4633	spa_strfree(oldvdpath);
4634	spa_strfree(newvdpath);
4635
4636	if (spa->spa_bootfs)
4637		spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4638
4639	return (0);
4640}
4641
4642/*
4643 * Detach a device from a mirror or replacing vdev.
4644 *
4645 * If 'replace_done' is specified, only detach if the parent
4646 * is a replacing vdev.
4647 */
4648int
4649spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4650{
4651	uint64_t txg;
4652	int error;
4653	vdev_t *rvd = spa->spa_root_vdev;
4654	vdev_t *vd, *pvd, *cvd, *tvd;
4655	boolean_t unspare = B_FALSE;
4656	uint64_t unspare_guid = 0;
4657	char *vdpath;
4658
4659	ASSERT(spa_writeable(spa));
4660
4661	txg = spa_vdev_enter(spa);
4662
4663	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4664
4665	if (vd == NULL)
4666		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4667
4668	if (!vd->vdev_ops->vdev_op_leaf)
4669		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4670
4671	pvd = vd->vdev_parent;
4672
4673	/*
4674	 * If the parent/child relationship is not as expected, don't do it.
4675	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4676	 * vdev that's replacing B with C.  The user's intent in replacing
4677	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4678	 * the replace by detaching C, the expected behavior is to end up
4679	 * M(A,B).  But suppose that right after deciding to detach C,
4680	 * the replacement of B completes.  We would have M(A,C), and then
4681	 * ask to detach C, which would leave us with just A -- not what
4682	 * the user wanted.  To prevent this, we make sure that the
4683	 * parent/child relationship hasn't changed -- in this example,
4684	 * that C's parent is still the replacing vdev R.
4685	 */
4686	if (pvd->vdev_guid != pguid && pguid != 0)
4687		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4688
4689	/*
4690	 * Only 'replacing' or 'spare' vdevs can be replaced.
4691	 */
4692	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4693	    pvd->vdev_ops != &vdev_spare_ops)
4694		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4695
4696	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4697	    spa_version(spa) >= SPA_VERSION_SPARES);
4698
4699	/*
4700	 * Only mirror, replacing, and spare vdevs support detach.
4701	 */
4702	if (pvd->vdev_ops != &vdev_replacing_ops &&
4703	    pvd->vdev_ops != &vdev_mirror_ops &&
4704	    pvd->vdev_ops != &vdev_spare_ops)
4705		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4706
4707	/*
4708	 * If this device has the only valid copy of some data,
4709	 * we cannot safely detach it.
4710	 */
4711	if (vdev_dtl_required(vd))
4712		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4713
4714	ASSERT(pvd->vdev_children >= 2);
4715
4716	/*
4717	 * If we are detaching the second disk from a replacing vdev, then
4718	 * check to see if we changed the original vdev's path to have "/old"
4719	 * at the end in spa_vdev_attach().  If so, undo that change now.
4720	 */
4721	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4722	    vd->vdev_path != NULL) {
4723		size_t len = strlen(vd->vdev_path);
4724
4725		for (int c = 0; c < pvd->vdev_children; c++) {
4726			cvd = pvd->vdev_child[c];
4727
4728			if (cvd == vd || cvd->vdev_path == NULL)
4729				continue;
4730
4731			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4732			    strcmp(cvd->vdev_path + len, "/old") == 0) {
4733				spa_strfree(cvd->vdev_path);
4734				cvd->vdev_path = spa_strdup(vd->vdev_path);
4735				break;
4736			}
4737		}
4738	}
4739
4740	/*
4741	 * If we are detaching the original disk from a spare, then it implies
4742	 * that the spare should become a real disk, and be removed from the
4743	 * active spare list for the pool.
4744	 */
4745	if (pvd->vdev_ops == &vdev_spare_ops &&
4746	    vd->vdev_id == 0 &&
4747	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4748		unspare = B_TRUE;
4749
4750	/*
4751	 * Erase the disk labels so the disk can be used for other things.
4752	 * This must be done after all other error cases are handled,
4753	 * but before we disembowel vd (so we can still do I/O to it).
4754	 * But if we can't do it, don't treat the error as fatal --
4755	 * it may be that the unwritability of the disk is the reason
4756	 * it's being detached!
4757	 */
4758	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4759
4760	/*
4761	 * Remove vd from its parent and compact the parent's children.
4762	 */
4763	vdev_remove_child(pvd, vd);
4764	vdev_compact_children(pvd);
4765
4766	/*
4767	 * Remember one of the remaining children so we can get tvd below.
4768	 */
4769	cvd = pvd->vdev_child[pvd->vdev_children - 1];
4770
4771	/*
4772	 * If we need to remove the remaining child from the list of hot spares,
4773	 * do it now, marking the vdev as no longer a spare in the process.
4774	 * We must do this before vdev_remove_parent(), because that can
4775	 * change the GUID if it creates a new toplevel GUID.  For a similar
4776	 * reason, we must remove the spare now, in the same txg as the detach;
4777	 * otherwise someone could attach a new sibling, change the GUID, and
4778	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4779	 */
4780	if (unspare) {
4781		ASSERT(cvd->vdev_isspare);
4782		spa_spare_remove(cvd);
4783		unspare_guid = cvd->vdev_guid;
4784		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4785		cvd->vdev_unspare = B_TRUE;
4786	}
4787
4788	/*
4789	 * If the parent mirror/replacing vdev only has one child,
4790	 * the parent is no longer needed.  Remove it from the tree.
4791	 */
4792	if (pvd->vdev_children == 1) {
4793		if (pvd->vdev_ops == &vdev_spare_ops)
4794			cvd->vdev_unspare = B_FALSE;
4795		vdev_remove_parent(cvd);
4796	}
4797
4798
4799	/*
4800	 * We don't set tvd until now because the parent we just removed
4801	 * may have been the previous top-level vdev.
4802	 */
4803	tvd = cvd->vdev_top;
4804	ASSERT(tvd->vdev_parent == rvd);
4805
4806	/*
4807	 * Reevaluate the parent vdev state.
4808	 */
4809	vdev_propagate_state(cvd);
4810
4811	/*
4812	 * If the 'autoexpand' property is set on the pool then automatically
4813	 * try to expand the size of the pool. For example if the device we
4814	 * just detached was smaller than the others, it may be possible to
4815	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4816	 * first so that we can obtain the updated sizes of the leaf vdevs.
4817	 */
4818	if (spa->spa_autoexpand) {
4819		vdev_reopen(tvd);
4820		vdev_expand(tvd, txg);
4821	}
4822
4823	vdev_config_dirty(tvd);
4824
4825	/*
4826	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4827	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4828	 * But first make sure we're not on any *other* txg's DTL list, to
4829	 * prevent vd from being accessed after it's freed.
4830	 */
4831	vdpath = spa_strdup(vd->vdev_path);
4832	for (int t = 0; t < TXG_SIZE; t++)
4833		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4834	vd->vdev_detached = B_TRUE;
4835	vdev_dirty(tvd, VDD_DTL, vd, txg);
4836
4837	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4838
4839	/* hang on to the spa before we release the lock */
4840	spa_open_ref(spa, FTAG);
4841
4842	error = spa_vdev_exit(spa, vd, txg, 0);
4843
4844	spa_history_log_internal(spa, "detach", NULL,
4845	    "vdev=%s", vdpath);
4846	spa_strfree(vdpath);
4847
4848	/*
4849	 * If this was the removal of the original device in a hot spare vdev,
4850	 * then we want to go through and remove the device from the hot spare
4851	 * list of every other pool.
4852	 */
4853	if (unspare) {
4854		spa_t *altspa = NULL;
4855
4856		mutex_enter(&spa_namespace_lock);
4857		while ((altspa = spa_next(altspa)) != NULL) {
4858			if (altspa->spa_state != POOL_STATE_ACTIVE ||
4859			    altspa == spa)
4860				continue;
4861
4862			spa_open_ref(altspa, FTAG);
4863			mutex_exit(&spa_namespace_lock);
4864			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4865			mutex_enter(&spa_namespace_lock);
4866			spa_close(altspa, FTAG);
4867		}
4868		mutex_exit(&spa_namespace_lock);
4869
4870		/* search the rest of the vdevs for spares to remove */
4871		spa_vdev_resilver_done(spa);
4872	}
4873
4874	/* all done with the spa; OK to release */
4875	mutex_enter(&spa_namespace_lock);
4876	spa_close(spa, FTAG);
4877	mutex_exit(&spa_namespace_lock);
4878
4879	return (error);
4880}
4881
4882/*
4883 * Split a set of devices from their mirrors, and create a new pool from them.
4884 */
4885int
4886spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4887    nvlist_t *props, boolean_t exp)
4888{
4889	int error = 0;
4890	uint64_t txg, *glist;
4891	spa_t *newspa;
4892	uint_t c, children, lastlog;
4893	nvlist_t **child, *nvl, *tmp;
4894	dmu_tx_t *tx;
4895	char *altroot = NULL;
4896	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4897	boolean_t activate_slog;
4898
4899	ASSERT(spa_writeable(spa));
4900
4901	txg = spa_vdev_enter(spa);
4902
4903	/* clear the log and flush everything up to now */
4904	activate_slog = spa_passivate_log(spa);
4905	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4906	error = spa_offline_log(spa);
4907	txg = spa_vdev_config_enter(spa);
4908
4909	if (activate_slog)
4910		spa_activate_log(spa);
4911
4912	if (error != 0)
4913		return (spa_vdev_exit(spa, NULL, txg, error));
4914
4915	/* check new spa name before going any further */
4916	if (spa_lookup(newname) != NULL)
4917		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4918
4919	/*
4920	 * scan through all the children to ensure they're all mirrors
4921	 */
4922	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4923	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4924	    &children) != 0)
4925		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4926
4927	/* first, check to ensure we've got the right child count */
4928	rvd = spa->spa_root_vdev;
4929	lastlog = 0;
4930	for (c = 0; c < rvd->vdev_children; c++) {
4931		vdev_t *vd = rvd->vdev_child[c];
4932
4933		/* don't count the holes & logs as children */
4934		if (vd->vdev_islog || vd->vdev_ishole) {
4935			if (lastlog == 0)
4936				lastlog = c;
4937			continue;
4938		}
4939
4940		lastlog = 0;
4941	}
4942	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4943		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4944
4945	/* next, ensure no spare or cache devices are part of the split */
4946	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4947	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4948		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4949
4950	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4951	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4952
4953	/* then, loop over each vdev and validate it */
4954	for (c = 0; c < children; c++) {
4955		uint64_t is_hole = 0;
4956
4957		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4958		    &is_hole);
4959
4960		if (is_hole != 0) {
4961			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4962			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4963				continue;
4964			} else {
4965				error = SET_ERROR(EINVAL);
4966				break;
4967			}
4968		}
4969
4970		/* which disk is going to be split? */
4971		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4972		    &glist[c]) != 0) {
4973			error = SET_ERROR(EINVAL);
4974			break;
4975		}
4976
4977		/* look it up in the spa */
4978		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4979		if (vml[c] == NULL) {
4980			error = SET_ERROR(ENODEV);
4981			break;
4982		}
4983
4984		/* make sure there's nothing stopping the split */
4985		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4986		    vml[c]->vdev_islog ||
4987		    vml[c]->vdev_ishole ||
4988		    vml[c]->vdev_isspare ||
4989		    vml[c]->vdev_isl2cache ||
4990		    !vdev_writeable(vml[c]) ||
4991		    vml[c]->vdev_children != 0 ||
4992		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4993		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4994			error = SET_ERROR(EINVAL);
4995			break;
4996		}
4997
4998		if (vdev_dtl_required(vml[c])) {
4999			error = SET_ERROR(EBUSY);
5000			break;
5001		}
5002
5003		/* we need certain info from the top level */
5004		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5005		    vml[c]->vdev_top->vdev_ms_array) == 0);
5006		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5007		    vml[c]->vdev_top->vdev_ms_shift) == 0);
5008		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5009		    vml[c]->vdev_top->vdev_asize) == 0);
5010		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5011		    vml[c]->vdev_top->vdev_ashift) == 0);
5012	}
5013
5014	if (error != 0) {
5015		kmem_free(vml, children * sizeof (vdev_t *));
5016		kmem_free(glist, children * sizeof (uint64_t));
5017		return (spa_vdev_exit(spa, NULL, txg, error));
5018	}
5019
5020	/* stop writers from using the disks */
5021	for (c = 0; c < children; c++) {
5022		if (vml[c] != NULL)
5023			vml[c]->vdev_offline = B_TRUE;
5024	}
5025	vdev_reopen(spa->spa_root_vdev);
5026
5027	/*
5028	 * Temporarily record the splitting vdevs in the spa config.  This
5029	 * will disappear once the config is regenerated.
5030	 */
5031	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5032	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5033	    glist, children) == 0);
5034	kmem_free(glist, children * sizeof (uint64_t));
5035
5036	mutex_enter(&spa->spa_props_lock);
5037	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5038	    nvl) == 0);
5039	mutex_exit(&spa->spa_props_lock);
5040	spa->spa_config_splitting = nvl;
5041	vdev_config_dirty(spa->spa_root_vdev);
5042
5043	/* configure and create the new pool */
5044	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5045	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5046	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5047	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5048	    spa_version(spa)) == 0);
5049	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5050	    spa->spa_config_txg) == 0);
5051	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5052	    spa_generate_guid(NULL)) == 0);
5053	(void) nvlist_lookup_string(props,
5054	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5055
5056	/* add the new pool to the namespace */
5057	newspa = spa_add(newname, config, altroot);
5058	newspa->spa_config_txg = spa->spa_config_txg;
5059	spa_set_log_state(newspa, SPA_LOG_CLEAR);
5060
5061	/* release the spa config lock, retaining the namespace lock */
5062	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5063
5064	if (zio_injection_enabled)
5065		zio_handle_panic_injection(spa, FTAG, 1);
5066
5067	spa_activate(newspa, spa_mode_global);
5068	spa_async_suspend(newspa);
5069
5070	/* create the new pool from the disks of the original pool */
5071	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5072	if (error)
5073		goto out;
5074
5075	/* if that worked, generate a real config for the new pool */
5076	if (newspa->spa_root_vdev != NULL) {
5077		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5078		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
5079		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5080		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5081		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5082		    B_TRUE));
5083	}
5084
5085	/* set the props */
5086	if (props != NULL) {
5087		spa_configfile_set(newspa, props, B_FALSE);
5088		error = spa_prop_set(newspa, props);
5089		if (error)
5090			goto out;
5091	}
5092
5093	/* flush everything */
5094	txg = spa_vdev_config_enter(newspa);
5095	vdev_config_dirty(newspa->spa_root_vdev);
5096	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5097
5098	if (zio_injection_enabled)
5099		zio_handle_panic_injection(spa, FTAG, 2);
5100
5101	spa_async_resume(newspa);
5102
5103	/* finally, update the original pool's config */
5104	txg = spa_vdev_config_enter(spa);
5105	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5106	error = dmu_tx_assign(tx, TXG_WAIT);
5107	if (error != 0)
5108		dmu_tx_abort(tx);
5109	for (c = 0; c < children; c++) {
5110		if (vml[c] != NULL) {
5111			vdev_split(vml[c]);
5112			if (error == 0)
5113				spa_history_log_internal(spa, "detach", tx,
5114				    "vdev=%s", vml[c]->vdev_path);
5115			vdev_free(vml[c]);
5116		}
5117	}
5118	vdev_config_dirty(spa->spa_root_vdev);
5119	spa->spa_config_splitting = NULL;
5120	nvlist_free(nvl);
5121	if (error == 0)
5122		dmu_tx_commit(tx);
5123	(void) spa_vdev_exit(spa, NULL, txg, 0);
5124
5125	if (zio_injection_enabled)
5126		zio_handle_panic_injection(spa, FTAG, 3);
5127
5128	/* split is complete; log a history record */
5129	spa_history_log_internal(newspa, "split", NULL,
5130	    "from pool %s", spa_name(spa));
5131
5132	kmem_free(vml, children * sizeof (vdev_t *));
5133
5134	/* if we're not going to mount the filesystems in userland, export */
5135	if (exp)
5136		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5137		    B_FALSE, B_FALSE);
5138
5139	return (error);
5140
5141out:
5142	spa_unload(newspa);
5143	spa_deactivate(newspa);
5144	spa_remove(newspa);
5145
5146	txg = spa_vdev_config_enter(spa);
5147
5148	/* re-online all offlined disks */
5149	for (c = 0; c < children; c++) {
5150		if (vml[c] != NULL)
5151			vml[c]->vdev_offline = B_FALSE;
5152	}
5153	vdev_reopen(spa->spa_root_vdev);
5154
5155	nvlist_free(spa->spa_config_splitting);
5156	spa->spa_config_splitting = NULL;
5157	(void) spa_vdev_exit(spa, NULL, txg, error);
5158
5159	kmem_free(vml, children * sizeof (vdev_t *));
5160	return (error);
5161}
5162
5163static nvlist_t *
5164spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5165{
5166	for (int i = 0; i < count; i++) {
5167		uint64_t guid;
5168
5169		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5170		    &guid) == 0);
5171
5172		if (guid == target_guid)
5173			return (nvpp[i]);
5174	}
5175
5176	return (NULL);
5177}
5178
5179static void
5180spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5181	nvlist_t *dev_to_remove)
5182{
5183	nvlist_t **newdev = NULL;
5184
5185	if (count > 1)
5186		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5187
5188	for (int i = 0, j = 0; i < count; i++) {
5189		if (dev[i] == dev_to_remove)
5190			continue;
5191		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5192	}
5193
5194	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5195	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5196
5197	for (int i = 0; i < count - 1; i++)
5198		nvlist_free(newdev[i]);
5199
5200	if (count > 1)
5201		kmem_free(newdev, (count - 1) * sizeof (void *));
5202}
5203
5204/*
5205 * Evacuate the device.
5206 */
5207static int
5208spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5209{
5210	uint64_t txg;
5211	int error = 0;
5212
5213	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5214	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5215	ASSERT(vd == vd->vdev_top);
5216
5217	/*
5218	 * Evacuate the device.  We don't hold the config lock as writer
5219	 * since we need to do I/O but we do keep the
5220	 * spa_namespace_lock held.  Once this completes the device
5221	 * should no longer have any blocks allocated on it.
5222	 */
5223	if (vd->vdev_islog) {
5224		if (vd->vdev_stat.vs_alloc != 0)
5225			error = spa_offline_log(spa);
5226	} else {
5227		error = SET_ERROR(ENOTSUP);
5228	}
5229
5230	if (error)
5231		return (error);
5232
5233	/*
5234	 * The evacuation succeeded.  Remove any remaining MOS metadata
5235	 * associated with this vdev, and wait for these changes to sync.
5236	 */
5237	ASSERT0(vd->vdev_stat.vs_alloc);
5238	txg = spa_vdev_config_enter(spa);
5239	vd->vdev_removing = B_TRUE;
5240	vdev_dirty_leaves(vd, VDD_DTL, txg);
5241	vdev_config_dirty(vd);
5242	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5243
5244	return (0);
5245}
5246
5247/*
5248 * Complete the removal by cleaning up the namespace.
5249 */
5250static void
5251spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5252{
5253	vdev_t *rvd = spa->spa_root_vdev;
5254	uint64_t id = vd->vdev_id;
5255	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5256
5257	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5258	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5259	ASSERT(vd == vd->vdev_top);
5260
5261	/*
5262	 * Only remove any devices which are empty.
5263	 */
5264	if (vd->vdev_stat.vs_alloc != 0)
5265		return;
5266
5267	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5268
5269	if (list_link_active(&vd->vdev_state_dirty_node))
5270		vdev_state_clean(vd);
5271	if (list_link_active(&vd->vdev_config_dirty_node))
5272		vdev_config_clean(vd);
5273
5274	vdev_free(vd);
5275
5276	if (last_vdev) {
5277		vdev_compact_children(rvd);
5278	} else {
5279		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5280		vdev_add_child(rvd, vd);
5281	}
5282	vdev_config_dirty(rvd);
5283
5284	/*
5285	 * Reassess the health of our root vdev.
5286	 */
5287	vdev_reopen(rvd);
5288}
5289
5290/*
5291 * Remove a device from the pool -
5292 *
5293 * Removing a device from the vdev namespace requires several steps
5294 * and can take a significant amount of time.  As a result we use
5295 * the spa_vdev_config_[enter/exit] functions which allow us to
5296 * grab and release the spa_config_lock while still holding the namespace
5297 * lock.  During each step the configuration is synced out.
5298 *
5299 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5300 * devices.
5301 */
5302int
5303spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5304{
5305	vdev_t *vd;
5306	metaslab_group_t *mg;
5307	nvlist_t **spares, **l2cache, *nv;
5308	uint64_t txg = 0;
5309	uint_t nspares, nl2cache;
5310	int error = 0;
5311	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5312
5313	ASSERT(spa_writeable(spa));
5314
5315	if (!locked)
5316		txg = spa_vdev_enter(spa);
5317
5318	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5319
5320	if (spa->spa_spares.sav_vdevs != NULL &&
5321	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5322	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5323	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5324		/*
5325		 * Only remove the hot spare if it's not currently in use
5326		 * in this pool.
5327		 */
5328		if (vd == NULL || unspare) {
5329			spa_vdev_remove_aux(spa->spa_spares.sav_config,
5330			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5331			spa_load_spares(spa);
5332			spa->spa_spares.sav_sync = B_TRUE;
5333		} else {
5334			error = SET_ERROR(EBUSY);
5335		}
5336	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
5337	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5338	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5339	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5340		/*
5341		 * Cache devices can always be removed.
5342		 */
5343		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5344		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5345		spa_load_l2cache(spa);
5346		spa->spa_l2cache.sav_sync = B_TRUE;
5347	} else if (vd != NULL && vd->vdev_islog) {
5348		ASSERT(!locked);
5349		ASSERT(vd == vd->vdev_top);
5350
5351		mg = vd->vdev_mg;
5352
5353		/*
5354		 * Stop allocating from this vdev.
5355		 */
5356		metaslab_group_passivate(mg);
5357
5358		/*
5359		 * Wait for the youngest allocations and frees to sync,
5360		 * and then wait for the deferral of those frees to finish.
5361		 */
5362		spa_vdev_config_exit(spa, NULL,
5363		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5364
5365		/*
5366		 * Attempt to evacuate the vdev.
5367		 */
5368		error = spa_vdev_remove_evacuate(spa, vd);
5369
5370		txg = spa_vdev_config_enter(spa);
5371
5372		/*
5373		 * If we couldn't evacuate the vdev, unwind.
5374		 */
5375		if (error) {
5376			metaslab_group_activate(mg);
5377			return (spa_vdev_exit(spa, NULL, txg, error));
5378		}
5379
5380		/*
5381		 * Clean up the vdev namespace.
5382		 */
5383		spa_vdev_remove_from_namespace(spa, vd);
5384
5385	} else if (vd != NULL) {
5386		/*
5387		 * Normal vdevs cannot be removed (yet).
5388		 */
5389		error = SET_ERROR(ENOTSUP);
5390	} else {
5391		/*
5392		 * There is no vdev of any kind with the specified guid.
5393		 */
5394		error = SET_ERROR(ENOENT);
5395	}
5396
5397	if (!locked)
5398		return (spa_vdev_exit(spa, NULL, txg, error));
5399
5400	return (error);
5401}
5402
5403/*
5404 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5405 * currently spared, so we can detach it.
5406 */
5407static vdev_t *
5408spa_vdev_resilver_done_hunt(vdev_t *vd)
5409{
5410	vdev_t *newvd, *oldvd;
5411
5412	for (int c = 0; c < vd->vdev_children; c++) {
5413		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5414		if (oldvd != NULL)
5415			return (oldvd);
5416	}
5417
5418	/*
5419	 * Check for a completed replacement.  We always consider the first
5420	 * vdev in the list to be the oldest vdev, and the last one to be
5421	 * the newest (see spa_vdev_attach() for how that works).  In
5422	 * the case where the newest vdev is faulted, we will not automatically
5423	 * remove it after a resilver completes.  This is OK as it will require
5424	 * user intervention to determine which disk the admin wishes to keep.
5425	 */
5426	if (vd->vdev_ops == &vdev_replacing_ops) {
5427		ASSERT(vd->vdev_children > 1);
5428
5429		newvd = vd->vdev_child[vd->vdev_children - 1];
5430		oldvd = vd->vdev_child[0];
5431
5432		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5433		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5434		    !vdev_dtl_required(oldvd))
5435			return (oldvd);
5436	}
5437
5438	/*
5439	 * Check for a completed resilver with the 'unspare' flag set.
5440	 */
5441	if (vd->vdev_ops == &vdev_spare_ops) {
5442		vdev_t *first = vd->vdev_child[0];
5443		vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5444
5445		if (last->vdev_unspare) {
5446			oldvd = first;
5447			newvd = last;
5448		} else if (first->vdev_unspare) {
5449			oldvd = last;
5450			newvd = first;
5451		} else {
5452			oldvd = NULL;
5453		}
5454
5455		if (oldvd != NULL &&
5456		    vdev_dtl_empty(newvd, DTL_MISSING) &&
5457		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5458		    !vdev_dtl_required(oldvd))
5459			return (oldvd);
5460
5461		/*
5462		 * If there are more than two spares attached to a disk,
5463		 * and those spares are not required, then we want to
5464		 * attempt to free them up now so that they can be used
5465		 * by other pools.  Once we're back down to a single
5466		 * disk+spare, we stop removing them.
5467		 */
5468		if (vd->vdev_children > 2) {
5469			newvd = vd->vdev_child[1];
5470
5471			if (newvd->vdev_isspare && last->vdev_isspare &&
5472			    vdev_dtl_empty(last, DTL_MISSING) &&
5473			    vdev_dtl_empty(last, DTL_OUTAGE) &&
5474			    !vdev_dtl_required(newvd))
5475				return (newvd);
5476		}
5477	}
5478
5479	return (NULL);
5480}
5481
5482static void
5483spa_vdev_resilver_done(spa_t *spa)
5484{
5485	vdev_t *vd, *pvd, *ppvd;
5486	uint64_t guid, sguid, pguid, ppguid;
5487
5488	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5489
5490	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5491		pvd = vd->vdev_parent;
5492		ppvd = pvd->vdev_parent;
5493		guid = vd->vdev_guid;
5494		pguid = pvd->vdev_guid;
5495		ppguid = ppvd->vdev_guid;
5496		sguid = 0;
5497		/*
5498		 * If we have just finished replacing a hot spared device, then
5499		 * we need to detach the parent's first child (the original hot
5500		 * spare) as well.
5501		 */
5502		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5503		    ppvd->vdev_children == 2) {
5504			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5505			sguid = ppvd->vdev_child[1]->vdev_guid;
5506		}
5507		ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5508
5509		spa_config_exit(spa, SCL_ALL, FTAG);
5510		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5511			return;
5512		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5513			return;
5514		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5515	}
5516
5517	spa_config_exit(spa, SCL_ALL, FTAG);
5518}
5519
5520/*
5521 * Update the stored path or FRU for this vdev.
5522 */
5523int
5524spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5525    boolean_t ispath)
5526{
5527	vdev_t *vd;
5528	boolean_t sync = B_FALSE;
5529
5530	ASSERT(spa_writeable(spa));
5531
5532	spa_vdev_state_enter(spa, SCL_ALL);
5533
5534	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5535		return (spa_vdev_state_exit(spa, NULL, ENOENT));
5536
5537	if (!vd->vdev_ops->vdev_op_leaf)
5538		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5539
5540	if (ispath) {
5541		if (strcmp(value, vd->vdev_path) != 0) {
5542			spa_strfree(vd->vdev_path);
5543			vd->vdev_path = spa_strdup(value);
5544			sync = B_TRUE;
5545		}
5546	} else {
5547		if (vd->vdev_fru == NULL) {
5548			vd->vdev_fru = spa_strdup(value);
5549			sync = B_TRUE;
5550		} else if (strcmp(value, vd->vdev_fru) != 0) {
5551			spa_strfree(vd->vdev_fru);
5552			vd->vdev_fru = spa_strdup(value);
5553			sync = B_TRUE;
5554		}
5555	}
5556
5557	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5558}
5559
5560int
5561spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5562{
5563	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5564}
5565
5566int
5567spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5568{
5569	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5570}
5571
5572/*
5573 * ==========================================================================
5574 * SPA Scanning
5575 * ==========================================================================
5576 */
5577
5578int
5579spa_scan_stop(spa_t *spa)
5580{
5581	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5582	if (dsl_scan_resilvering(spa->spa_dsl_pool))
5583		return (SET_ERROR(EBUSY));
5584	return (dsl_scan_cancel(spa->spa_dsl_pool));
5585}
5586
5587int
5588spa_scan(spa_t *spa, pool_scan_func_t func)
5589{
5590	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5591
5592	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5593		return (SET_ERROR(ENOTSUP));
5594
5595	/*
5596	 * If a resilver was requested, but there is no DTL on a
5597	 * writeable leaf device, we have nothing to do.
5598	 */
5599	if (func == POOL_SCAN_RESILVER &&
5600	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5601		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5602		return (0);
5603	}
5604
5605	return (dsl_scan(spa->spa_dsl_pool, func));
5606}
5607
5608/*
5609 * ==========================================================================
5610 * SPA async task processing
5611 * ==========================================================================
5612 */
5613
5614static void
5615spa_async_remove(spa_t *spa, vdev_t *vd)
5616{
5617	if (vd->vdev_remove_wanted) {
5618		vd->vdev_remove_wanted = B_FALSE;
5619		vd->vdev_delayed_close = B_FALSE;
5620		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5621
5622		/*
5623		 * We want to clear the stats, but we don't want to do a full
5624		 * vdev_clear() as that will cause us to throw away
5625		 * degraded/faulted state as well as attempt to reopen the
5626		 * device, all of which is a waste.
5627		 */
5628		vd->vdev_stat.vs_read_errors = 0;
5629		vd->vdev_stat.vs_write_errors = 0;
5630		vd->vdev_stat.vs_checksum_errors = 0;
5631
5632		vdev_state_dirty(vd->vdev_top);
5633	}
5634
5635	for (int c = 0; c < vd->vdev_children; c++)
5636		spa_async_remove(spa, vd->vdev_child[c]);
5637}
5638
5639static void
5640spa_async_probe(spa_t *spa, vdev_t *vd)
5641{
5642	if (vd->vdev_probe_wanted) {
5643		vd->vdev_probe_wanted = B_FALSE;
5644		vdev_reopen(vd);	/* vdev_open() does the actual probe */
5645	}
5646
5647	for (int c = 0; c < vd->vdev_children; c++)
5648		spa_async_probe(spa, vd->vdev_child[c]);
5649}
5650
5651static void
5652spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5653{
5654	sysevent_id_t eid;
5655	nvlist_t *attr;
5656	char *physpath;
5657
5658	if (!spa->spa_autoexpand)
5659		return;
5660
5661	for (int c = 0; c < vd->vdev_children; c++) {
5662		vdev_t *cvd = vd->vdev_child[c];
5663		spa_async_autoexpand(spa, cvd);
5664	}
5665
5666	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5667		return;
5668
5669	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5670	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5671
5672	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5673	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5674
5675	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5676	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5677
5678	nvlist_free(attr);
5679	kmem_free(physpath, MAXPATHLEN);
5680}
5681
5682static void
5683spa_async_thread(spa_t *spa)
5684{
5685	int tasks;
5686
5687	ASSERT(spa->spa_sync_on);
5688
5689	mutex_enter(&spa->spa_async_lock);
5690	tasks = spa->spa_async_tasks;
5691	spa->spa_async_tasks = 0;
5692	mutex_exit(&spa->spa_async_lock);
5693
5694	/*
5695	 * See if the config needs to be updated.
5696	 */
5697	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5698		uint64_t old_space, new_space;
5699
5700		mutex_enter(&spa_namespace_lock);
5701		old_space = metaslab_class_get_space(spa_normal_class(spa));
5702		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5703		new_space = metaslab_class_get_space(spa_normal_class(spa));
5704		mutex_exit(&spa_namespace_lock);
5705
5706		/*
5707		 * If the pool grew as a result of the config update,
5708		 * then log an internal history event.
5709		 */
5710		if (new_space != old_space) {
5711			spa_history_log_internal(spa, "vdev online", NULL,
5712			    "pool '%s' size: %llu(+%llu)",
5713			    spa_name(spa), new_space, new_space - old_space);
5714		}
5715	}
5716
5717	/*
5718	 * See if any devices need to be marked REMOVED.
5719	 */
5720	if (tasks & SPA_ASYNC_REMOVE) {
5721		spa_vdev_state_enter(spa, SCL_NONE);
5722		spa_async_remove(spa, spa->spa_root_vdev);
5723		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5724			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5725		for (int i = 0; i < spa->spa_spares.sav_count; i++)
5726			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5727		(void) spa_vdev_state_exit(spa, NULL, 0);
5728	}
5729
5730	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5731		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5732		spa_async_autoexpand(spa, spa->spa_root_vdev);
5733		spa_config_exit(spa, SCL_CONFIG, FTAG);
5734	}
5735
5736	/*
5737	 * See if any devices need to be probed.
5738	 */
5739	if (tasks & SPA_ASYNC_PROBE) {
5740		spa_vdev_state_enter(spa, SCL_NONE);
5741		spa_async_probe(spa, spa->spa_root_vdev);
5742		(void) spa_vdev_state_exit(spa, NULL, 0);
5743	}
5744
5745	/*
5746	 * If any devices are done replacing, detach them.
5747	 */
5748	if (tasks & SPA_ASYNC_RESILVER_DONE)
5749		spa_vdev_resilver_done(spa);
5750
5751	/*
5752	 * Kick off a resilver.
5753	 */
5754	if (tasks & SPA_ASYNC_RESILVER)
5755		dsl_resilver_restart(spa->spa_dsl_pool, 0);
5756
5757	/*
5758	 * Let the world know that we're done.
5759	 */
5760	mutex_enter(&spa->spa_async_lock);
5761	spa->spa_async_thread = NULL;
5762	cv_broadcast(&spa->spa_async_cv);
5763	mutex_exit(&spa->spa_async_lock);
5764	thread_exit();
5765}
5766
5767void
5768spa_async_suspend(spa_t *spa)
5769{
5770	mutex_enter(&spa->spa_async_lock);
5771	spa->spa_async_suspended++;
5772	while (spa->spa_async_thread != NULL)
5773		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5774	mutex_exit(&spa->spa_async_lock);
5775}
5776
5777void
5778spa_async_resume(spa_t *spa)
5779{
5780	mutex_enter(&spa->spa_async_lock);
5781	ASSERT(spa->spa_async_suspended != 0);
5782	spa->spa_async_suspended--;
5783	mutex_exit(&spa->spa_async_lock);
5784}
5785
5786static boolean_t
5787spa_async_tasks_pending(spa_t *spa)
5788{
5789	uint_t non_config_tasks;
5790	uint_t config_task;
5791	boolean_t config_task_suspended;
5792
5793	non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5794	config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5795	if (spa->spa_ccw_fail_time == 0) {
5796		config_task_suspended = B_FALSE;
5797	} else {
5798		config_task_suspended =
5799		    (gethrtime() - spa->spa_ccw_fail_time) <
5800		    (zfs_ccw_retry_interval * NANOSEC);
5801	}
5802
5803	return (non_config_tasks || (config_task && !config_task_suspended));
5804}
5805
5806static void
5807spa_async_dispatch(spa_t *spa)
5808{
5809	mutex_enter(&spa->spa_async_lock);
5810	if (spa_async_tasks_pending(spa) &&
5811	    !spa->spa_async_suspended &&
5812	    spa->spa_async_thread == NULL &&
5813	    rootdir != NULL)
5814		spa->spa_async_thread = thread_create(NULL, 0,
5815		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5816	mutex_exit(&spa->spa_async_lock);
5817}
5818
5819void
5820spa_async_request(spa_t *spa, int task)
5821{
5822	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5823	mutex_enter(&spa->spa_async_lock);
5824	spa->spa_async_tasks |= task;
5825	mutex_exit(&spa->spa_async_lock);
5826}
5827
5828/*
5829 * ==========================================================================
5830 * SPA syncing routines
5831 * ==========================================================================
5832 */
5833
5834static int
5835bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5836{
5837	bpobj_t *bpo = arg;
5838	bpobj_enqueue(bpo, bp, tx);
5839	return (0);
5840}
5841
5842static int
5843spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5844{
5845	zio_t *zio = arg;
5846
5847	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5848	    zio->io_flags));
5849	return (0);
5850}
5851
5852/*
5853 * Note: this simple function is not inlined to make it easier to dtrace the
5854 * amount of time spent syncing frees.
5855 */
5856static void
5857spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
5858{
5859	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5860	bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
5861	VERIFY(zio_wait(zio) == 0);
5862}
5863
5864/*
5865 * Note: this simple function is not inlined to make it easier to dtrace the
5866 * amount of time spent syncing deferred frees.
5867 */
5868static void
5869spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
5870{
5871	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5872	VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
5873	    spa_free_sync_cb, zio, tx), ==, 0);
5874	VERIFY0(zio_wait(zio));
5875}
5876
5877
5878static void
5879spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5880{
5881	char *packed = NULL;
5882	size_t bufsize;
5883	size_t nvsize = 0;
5884	dmu_buf_t *db;
5885
5886	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5887
5888	/*
5889	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5890	 * information.  This avoids the dmu_buf_will_dirty() path and
5891	 * saves us a pre-read to get data we don't actually care about.
5892	 */
5893	bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5894	packed = kmem_alloc(bufsize, KM_SLEEP);
5895
5896	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5897	    KM_SLEEP) == 0);
5898	bzero(packed + nvsize, bufsize - nvsize);
5899
5900	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5901
5902	kmem_free(packed, bufsize);
5903
5904	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5905	dmu_buf_will_dirty(db, tx);
5906	*(uint64_t *)db->db_data = nvsize;
5907	dmu_buf_rele(db, FTAG);
5908}
5909
5910static void
5911spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5912    const char *config, const char *entry)
5913{
5914	nvlist_t *nvroot;
5915	nvlist_t **list;
5916	int i;
5917
5918	if (!sav->sav_sync)
5919		return;
5920
5921	/*
5922	 * Update the MOS nvlist describing the list of available devices.
5923	 * spa_validate_aux() will have already made sure this nvlist is
5924	 * valid and the vdevs are labeled appropriately.
5925	 */
5926	if (sav->sav_object == 0) {
5927		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5928		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5929		    sizeof (uint64_t), tx);
5930		VERIFY(zap_update(spa->spa_meta_objset,
5931		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5932		    &sav->sav_object, tx) == 0);
5933	}
5934
5935	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5936	if (sav->sav_count == 0) {
5937		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5938	} else {
5939		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5940		for (i = 0; i < sav->sav_count; i++)
5941			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5942			    B_FALSE, VDEV_CONFIG_L2CACHE);
5943		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5944		    sav->sav_count) == 0);
5945		for (i = 0; i < sav->sav_count; i++)
5946			nvlist_free(list[i]);
5947		kmem_free(list, sav->sav_count * sizeof (void *));
5948	}
5949
5950	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5951	nvlist_free(nvroot);
5952
5953	sav->sav_sync = B_FALSE;
5954}
5955
5956static void
5957spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5958{
5959	nvlist_t *config;
5960
5961	if (list_is_empty(&spa->spa_config_dirty_list))
5962		return;
5963
5964	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5965
5966	config = spa_config_generate(spa, spa->spa_root_vdev,
5967	    dmu_tx_get_txg(tx), B_FALSE);
5968
5969	/*
5970	 * If we're upgrading the spa version then make sure that
5971	 * the config object gets updated with the correct version.
5972	 */
5973	if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5974		fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5975		    spa->spa_uberblock.ub_version);
5976
5977	spa_config_exit(spa, SCL_STATE, FTAG);
5978
5979	if (spa->spa_config_syncing)
5980		nvlist_free(spa->spa_config_syncing);
5981	spa->spa_config_syncing = config;
5982
5983	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5984}
5985
5986static void
5987spa_sync_version(void *arg, dmu_tx_t *tx)
5988{
5989	uint64_t *versionp = arg;
5990	uint64_t version = *versionp;
5991	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5992
5993	/*
5994	 * Setting the version is special cased when first creating the pool.
5995	 */
5996	ASSERT(tx->tx_txg != TXG_INITIAL);
5997
5998	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5999	ASSERT(version >= spa_version(spa));
6000
6001	spa->spa_uberblock.ub_version = version;
6002	vdev_config_dirty(spa->spa_root_vdev);
6003	spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6004}
6005
6006/*
6007 * Set zpool properties.
6008 */
6009static void
6010spa_sync_props(void *arg, dmu_tx_t *tx)
6011{
6012	nvlist_t *nvp = arg;
6013	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6014	objset_t *mos = spa->spa_meta_objset;
6015	nvpair_t *elem = NULL;
6016
6017	mutex_enter(&spa->spa_props_lock);
6018
6019	while ((elem = nvlist_next_nvpair(nvp, elem))) {
6020		uint64_t intval;
6021		char *strval, *fname;
6022		zpool_prop_t prop;
6023		const char *propname;
6024		zprop_type_t proptype;
6025		spa_feature_t fid;
6026
6027		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6028		case ZPROP_INVAL:
6029			/*
6030			 * We checked this earlier in spa_prop_validate().
6031			 */
6032			ASSERT(zpool_prop_feature(nvpair_name(elem)));
6033
6034			fname = strchr(nvpair_name(elem), '@') + 1;
6035			VERIFY0(zfeature_lookup_name(fname, &fid));
6036
6037			spa_feature_enable(spa, fid, tx);
6038			spa_history_log_internal(spa, "set", tx,
6039			    "%s=enabled", nvpair_name(elem));
6040			break;
6041
6042		case ZPOOL_PROP_VERSION:
6043