spa.c revision b8289d24d866c1af02d7007348f7f057693c15d3
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 */
27
28/*
29 * SPA: Storage Pool Allocator
30 *
31 * This file contains all the routines used when modifying on-disk SPA state.
32 * This includes opening, importing, destroying, exporting a pool, and syncing a
33 * pool.
34 */
35
36#include <sys/zfs_context.h>
37#include <sys/fm/fs/zfs.h>
38#include <sys/spa_impl.h>
39#include <sys/zio.h>
40#include <sys/zio_checksum.h>
41#include <sys/dmu.h>
42#include <sys/dmu_tx.h>
43#include <sys/zap.h>
44#include <sys/zil.h>
45#include <sys/ddt.h>
46#include <sys/vdev_impl.h>
47#include <sys/metaslab.h>
48#include <sys/metaslab_impl.h>
49#include <sys/uberblock_impl.h>
50#include <sys/txg.h>
51#include <sys/avl.h>
52#include <sys/dmu_traverse.h>
53#include <sys/dmu_objset.h>
54#include <sys/unique.h>
55#include <sys/dsl_pool.h>
56#include <sys/dsl_dataset.h>
57#include <sys/dsl_dir.h>
58#include <sys/dsl_prop.h>
59#include <sys/dsl_synctask.h>
60#include <sys/fs/zfs.h>
61#include <sys/arc.h>
62#include <sys/callb.h>
63#include <sys/systeminfo.h>
64#include <sys/spa_boot.h>
65#include <sys/zfs_ioctl.h>
66#include <sys/dsl_scan.h>
67#include <sys/zfeature.h>
68#include <sys/dsl_destroy.h>
69
70#ifdef	_KERNEL
71#include <sys/bootprops.h>
72#include <sys/callb.h>
73#include <sys/cpupart.h>
74#include <sys/pool.h>
75#include <sys/sysdc.h>
76#include <sys/zone.h>
77#endif	/* _KERNEL */
78
79#include "zfs_prop.h"
80#include "zfs_comutil.h"
81
82/*
83 * The interval, in seconds, at which failed configuration cache file writes
84 * should be retried.
85 */
86static int zfs_ccw_retry_interval = 300;
87
88typedef enum zti_modes {
89	ZTI_MODE_FIXED,			/* value is # of threads (min 1) */
90	ZTI_MODE_BATCH,			/* cpu-intensive; value is ignored */
91	ZTI_MODE_NULL,			/* don't create a taskq */
92	ZTI_NMODES
93} zti_modes_t;
94
95#define	ZTI_P(n, q)	{ ZTI_MODE_FIXED, (n), (q) }
96#define	ZTI_BATCH	{ ZTI_MODE_BATCH, 0, 1 }
97#define	ZTI_NULL	{ ZTI_MODE_NULL, 0, 0 }
98
99#define	ZTI_N(n)	ZTI_P(n, 1)
100#define	ZTI_ONE		ZTI_N(1)
101
102typedef struct zio_taskq_info {
103	zti_modes_t zti_mode;
104	uint_t zti_value;
105	uint_t zti_count;
106} zio_taskq_info_t;
107
108static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
109	"issue", "issue_high", "intr", "intr_high"
110};
111
112/*
113 * This table defines the taskq settings for each ZFS I/O type. When
114 * initializing a pool, we use this table to create an appropriately sized
115 * taskq. Some operations are low volume and therefore have a small, static
116 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
117 * macros. Other operations process a large amount of data; the ZTI_BATCH
118 * macro causes us to create a taskq oriented for throughput. Some operations
119 * are so high frequency and short-lived that the taskq itself can become a a
120 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
121 * additional degree of parallelism specified by the number of threads per-
122 * taskq and the number of taskqs; when dispatching an event in this case, the
123 * particular taskq is chosen at random.
124 *
125 * The different taskq priorities are to handle the different contexts (issue
126 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
127 * need to be handled with minimum delay.
128 */
129const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
130	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
131	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* NULL */
132	{ ZTI_N(8),	ZTI_NULL,	ZTI_P(12, 8),	ZTI_NULL }, /* READ */
133	{ ZTI_BATCH,	ZTI_N(5),	ZTI_N(8),	ZTI_N(5) }, /* WRITE */
134	{ ZTI_P(12, 8),	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* FREE */
135	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* CLAIM */
136	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* IOCTL */
137};
138
139static void spa_sync_version(void *arg, dmu_tx_t *tx);
140static void spa_sync_props(void *arg, dmu_tx_t *tx);
141static boolean_t spa_has_active_shared_spare(spa_t *spa);
142static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
143    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
144    char **ereport);
145static void spa_vdev_resilver_done(spa_t *spa);
146
147uint_t		zio_taskq_batch_pct = 75;	/* 1 thread per cpu in pset */
148id_t		zio_taskq_psrset_bind = PS_NONE;
149boolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
150uint_t		zio_taskq_basedc = 80;		/* base duty cycle */
151
152boolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
153extern int	zfs_sync_pass_deferred_free;
154
155/*
156 * This (illegal) pool name is used when temporarily importing a spa_t in order
157 * to get the vdev stats associated with the imported devices.
158 */
159#define	TRYIMPORT_NAME	"$import"
160
161/*
162 * ==========================================================================
163 * SPA properties routines
164 * ==========================================================================
165 */
166
167/*
168 * Add a (source=src, propname=propval) list to an nvlist.
169 */
170static void
171spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
172    uint64_t intval, zprop_source_t src)
173{
174	const char *propname = zpool_prop_to_name(prop);
175	nvlist_t *propval;
176
177	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
178	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
179
180	if (strval != NULL)
181		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
182	else
183		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
184
185	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
186	nvlist_free(propval);
187}
188
189/*
190 * Get property values from the spa configuration.
191 */
192static void
193spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
194{
195	vdev_t *rvd = spa->spa_root_vdev;
196	dsl_pool_t *pool = spa->spa_dsl_pool;
197	uint64_t size;
198	uint64_t alloc;
199	uint64_t space;
200	uint64_t cap, version;
201	zprop_source_t src = ZPROP_SRC_NONE;
202	spa_config_dirent_t *dp;
203
204	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
205
206	if (rvd != NULL) {
207		alloc = metaslab_class_get_alloc(spa_normal_class(spa));
208		size = metaslab_class_get_space(spa_normal_class(spa));
209		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
210		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
211		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
212		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
213		    size - alloc, src);
214
215		space = 0;
216		for (int c = 0; c < rvd->vdev_children; c++) {
217			vdev_t *tvd = rvd->vdev_child[c];
218			space += tvd->vdev_max_asize - tvd->vdev_asize;
219		}
220		spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
221		    src);
222
223		spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
224		    (spa_mode(spa) == FREAD), src);
225
226		cap = (size == 0) ? 0 : (alloc * 100 / size);
227		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
228
229		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
230		    ddt_get_pool_dedup_ratio(spa), src);
231
232		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
233		    rvd->vdev_state, src);
234
235		version = spa_version(spa);
236		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
237			src = ZPROP_SRC_DEFAULT;
238		else
239			src = ZPROP_SRC_LOCAL;
240		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
241	}
242
243	if (pool != NULL) {
244		/*
245		 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
246		 * when opening pools before this version freedir will be NULL.
247		 */
248		if (pool->dp_free_dir != NULL) {
249			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
250			    pool->dp_free_dir->dd_phys->dd_used_bytes, src);
251		} else {
252			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
253			    NULL, 0, src);
254		}
255
256		if (pool->dp_leak_dir != NULL) {
257			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
258			    pool->dp_leak_dir->dd_phys->dd_used_bytes, src);
259		} else {
260			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
261			    NULL, 0, src);
262		}
263	}
264
265	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
266
267	if (spa->spa_comment != NULL) {
268		spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
269		    0, ZPROP_SRC_LOCAL);
270	}
271
272	if (spa->spa_root != NULL)
273		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
274		    0, ZPROP_SRC_LOCAL);
275
276	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
277		if (dp->scd_path == NULL) {
278			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
279			    "none", 0, ZPROP_SRC_LOCAL);
280		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
281			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
282			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
283		}
284	}
285}
286
287/*
288 * Get zpool property values.
289 */
290int
291spa_prop_get(spa_t *spa, nvlist_t **nvp)
292{
293	objset_t *mos = spa->spa_meta_objset;
294	zap_cursor_t zc;
295	zap_attribute_t za;
296	int err;
297
298	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
299
300	mutex_enter(&spa->spa_props_lock);
301
302	/*
303	 * Get properties from the spa config.
304	 */
305	spa_prop_get_config(spa, nvp);
306
307	/* If no pool property object, no more prop to get. */
308	if (mos == NULL || spa->spa_pool_props_object == 0) {
309		mutex_exit(&spa->spa_props_lock);
310		return (0);
311	}
312
313	/*
314	 * Get properties from the MOS pool property object.
315	 */
316	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
317	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
318	    zap_cursor_advance(&zc)) {
319		uint64_t intval = 0;
320		char *strval = NULL;
321		zprop_source_t src = ZPROP_SRC_DEFAULT;
322		zpool_prop_t prop;
323
324		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
325			continue;
326
327		switch (za.za_integer_length) {
328		case 8:
329			/* integer property */
330			if (za.za_first_integer !=
331			    zpool_prop_default_numeric(prop))
332				src = ZPROP_SRC_LOCAL;
333
334			if (prop == ZPOOL_PROP_BOOTFS) {
335				dsl_pool_t *dp;
336				dsl_dataset_t *ds = NULL;
337
338				dp = spa_get_dsl(spa);
339				dsl_pool_config_enter(dp, FTAG);
340				if (err = dsl_dataset_hold_obj(dp,
341				    za.za_first_integer, FTAG, &ds)) {
342					dsl_pool_config_exit(dp, FTAG);
343					break;
344				}
345
346				strval = kmem_alloc(
347				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
348				    KM_SLEEP);
349				dsl_dataset_name(ds, strval);
350				dsl_dataset_rele(ds, FTAG);
351				dsl_pool_config_exit(dp, FTAG);
352			} else {
353				strval = NULL;
354				intval = za.za_first_integer;
355			}
356
357			spa_prop_add_list(*nvp, prop, strval, intval, src);
358
359			if (strval != NULL)
360				kmem_free(strval,
361				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
362
363			break;
364
365		case 1:
366			/* string property */
367			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
368			err = zap_lookup(mos, spa->spa_pool_props_object,
369			    za.za_name, 1, za.za_num_integers, strval);
370			if (err) {
371				kmem_free(strval, za.za_num_integers);
372				break;
373			}
374			spa_prop_add_list(*nvp, prop, strval, 0, src);
375			kmem_free(strval, za.za_num_integers);
376			break;
377
378		default:
379			break;
380		}
381	}
382	zap_cursor_fini(&zc);
383	mutex_exit(&spa->spa_props_lock);
384out:
385	if (err && err != ENOENT) {
386		nvlist_free(*nvp);
387		*nvp = NULL;
388		return (err);
389	}
390
391	return (0);
392}
393
394/*
395 * Validate the given pool properties nvlist and modify the list
396 * for the property values to be set.
397 */
398static int
399spa_prop_validate(spa_t *spa, nvlist_t *props)
400{
401	nvpair_t *elem;
402	int error = 0, reset_bootfs = 0;
403	uint64_t objnum = 0;
404	boolean_t has_feature = B_FALSE;
405
406	elem = NULL;
407	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
408		uint64_t intval;
409		char *strval, *slash, *check, *fname;
410		const char *propname = nvpair_name(elem);
411		zpool_prop_t prop = zpool_name_to_prop(propname);
412
413		switch (prop) {
414		case ZPROP_INVAL:
415			if (!zpool_prop_feature(propname)) {
416				error = SET_ERROR(EINVAL);
417				break;
418			}
419
420			/*
421			 * Sanitize the input.
422			 */
423			if (nvpair_type(elem) != DATA_TYPE_UINT64) {
424				error = SET_ERROR(EINVAL);
425				break;
426			}
427
428			if (nvpair_value_uint64(elem, &intval) != 0) {
429				error = SET_ERROR(EINVAL);
430				break;
431			}
432
433			if (intval != 0) {
434				error = SET_ERROR(EINVAL);
435				break;
436			}
437
438			fname = strchr(propname, '@') + 1;
439			if (zfeature_lookup_name(fname, NULL) != 0) {
440				error = SET_ERROR(EINVAL);
441				break;
442			}
443
444			has_feature = B_TRUE;
445			break;
446
447		case ZPOOL_PROP_VERSION:
448			error = nvpair_value_uint64(elem, &intval);
449			if (!error &&
450			    (intval < spa_version(spa) ||
451			    intval > SPA_VERSION_BEFORE_FEATURES ||
452			    has_feature))
453				error = SET_ERROR(EINVAL);
454			break;
455
456		case ZPOOL_PROP_DELEGATION:
457		case ZPOOL_PROP_AUTOREPLACE:
458		case ZPOOL_PROP_LISTSNAPS:
459		case ZPOOL_PROP_AUTOEXPAND:
460			error = nvpair_value_uint64(elem, &intval);
461			if (!error && intval > 1)
462				error = SET_ERROR(EINVAL);
463			break;
464
465		case ZPOOL_PROP_BOOTFS:
466			/*
467			 * If the pool version is less than SPA_VERSION_BOOTFS,
468			 * or the pool is still being created (version == 0),
469			 * the bootfs property cannot be set.
470			 */
471			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
472				error = SET_ERROR(ENOTSUP);
473				break;
474			}
475
476			/*
477			 * Make sure the vdev config is bootable
478			 */
479			if (!vdev_is_bootable(spa->spa_root_vdev)) {
480				error = SET_ERROR(ENOTSUP);
481				break;
482			}
483
484			reset_bootfs = 1;
485
486			error = nvpair_value_string(elem, &strval);
487
488			if (!error) {
489				objset_t *os;
490				uint64_t compress;
491
492				if (strval == NULL || strval[0] == '\0') {
493					objnum = zpool_prop_default_numeric(
494					    ZPOOL_PROP_BOOTFS);
495					break;
496				}
497
498				if (error = dmu_objset_hold(strval, FTAG, &os))
499					break;
500
501				/* Must be ZPL and not gzip compressed. */
502
503				if (dmu_objset_type(os) != DMU_OST_ZFS) {
504					error = SET_ERROR(ENOTSUP);
505				} else if ((error =
506				    dsl_prop_get_int_ds(dmu_objset_ds(os),
507				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
508				    &compress)) == 0 &&
509				    !BOOTFS_COMPRESS_VALID(compress)) {
510					error = SET_ERROR(ENOTSUP);
511				} else {
512					objnum = dmu_objset_id(os);
513				}
514				dmu_objset_rele(os, FTAG);
515			}
516			break;
517
518		case ZPOOL_PROP_FAILUREMODE:
519			error = nvpair_value_uint64(elem, &intval);
520			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
521			    intval > ZIO_FAILURE_MODE_PANIC))
522				error = SET_ERROR(EINVAL);
523
524			/*
525			 * This is a special case which only occurs when
526			 * the pool has completely failed. This allows
527			 * the user to change the in-core failmode property
528			 * without syncing it out to disk (I/Os might
529			 * currently be blocked). We do this by returning
530			 * EIO to the caller (spa_prop_set) to trick it
531			 * into thinking we encountered a property validation
532			 * error.
533			 */
534			if (!error && spa_suspended(spa)) {
535				spa->spa_failmode = intval;
536				error = SET_ERROR(EIO);
537			}
538			break;
539
540		case ZPOOL_PROP_CACHEFILE:
541			if ((error = nvpair_value_string(elem, &strval)) != 0)
542				break;
543
544			if (strval[0] == '\0')
545				break;
546
547			if (strcmp(strval, "none") == 0)
548				break;
549
550			if (strval[0] != '/') {
551				error = SET_ERROR(EINVAL);
552				break;
553			}
554
555			slash = strrchr(strval, '/');
556			ASSERT(slash != NULL);
557
558			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
559			    strcmp(slash, "/..") == 0)
560				error = SET_ERROR(EINVAL);
561			break;
562
563		case ZPOOL_PROP_COMMENT:
564			if ((error = nvpair_value_string(elem, &strval)) != 0)
565				break;
566			for (check = strval; *check != '\0'; check++) {
567				/*
568				 * The kernel doesn't have an easy isprint()
569				 * check.  For this kernel check, we merely
570				 * check ASCII apart from DEL.  Fix this if
571				 * there is an easy-to-use kernel isprint().
572				 */
573				if (*check >= 0x7f) {
574					error = SET_ERROR(EINVAL);
575					break;
576				}
577				check++;
578			}
579			if (strlen(strval) > ZPROP_MAX_COMMENT)
580				error = E2BIG;
581			break;
582
583		case ZPOOL_PROP_DEDUPDITTO:
584			if (spa_version(spa) < SPA_VERSION_DEDUP)
585				error = SET_ERROR(ENOTSUP);
586			else
587				error = nvpair_value_uint64(elem, &intval);
588			if (error == 0 &&
589			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
590				error = SET_ERROR(EINVAL);
591			break;
592		}
593
594		if (error)
595			break;
596	}
597
598	if (!error && reset_bootfs) {
599		error = nvlist_remove(props,
600		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
601
602		if (!error) {
603			error = nvlist_add_uint64(props,
604			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
605		}
606	}
607
608	return (error);
609}
610
611void
612spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
613{
614	char *cachefile;
615	spa_config_dirent_t *dp;
616
617	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
618	    &cachefile) != 0)
619		return;
620
621	dp = kmem_alloc(sizeof (spa_config_dirent_t),
622	    KM_SLEEP);
623
624	if (cachefile[0] == '\0')
625		dp->scd_path = spa_strdup(spa_config_path);
626	else if (strcmp(cachefile, "none") == 0)
627		dp->scd_path = NULL;
628	else
629		dp->scd_path = spa_strdup(cachefile);
630
631	list_insert_head(&spa->spa_config_list, dp);
632	if (need_sync)
633		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
634}
635
636int
637spa_prop_set(spa_t *spa, nvlist_t *nvp)
638{
639	int error;
640	nvpair_t *elem = NULL;
641	boolean_t need_sync = B_FALSE;
642
643	if ((error = spa_prop_validate(spa, nvp)) != 0)
644		return (error);
645
646	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
647		zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
648
649		if (prop == ZPOOL_PROP_CACHEFILE ||
650		    prop == ZPOOL_PROP_ALTROOT ||
651		    prop == ZPOOL_PROP_READONLY)
652			continue;
653
654		if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
655			uint64_t ver;
656
657			if (prop == ZPOOL_PROP_VERSION) {
658				VERIFY(nvpair_value_uint64(elem, &ver) == 0);
659			} else {
660				ASSERT(zpool_prop_feature(nvpair_name(elem)));
661				ver = SPA_VERSION_FEATURES;
662				need_sync = B_TRUE;
663			}
664
665			/* Save time if the version is already set. */
666			if (ver == spa_version(spa))
667				continue;
668
669			/*
670			 * In addition to the pool directory object, we might
671			 * create the pool properties object, the features for
672			 * read object, the features for write object, or the
673			 * feature descriptions object.
674			 */
675			error = dsl_sync_task(spa->spa_name, NULL,
676			    spa_sync_version, &ver, 6);
677			if (error)
678				return (error);
679			continue;
680		}
681
682		need_sync = B_TRUE;
683		break;
684	}
685
686	if (need_sync) {
687		return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
688		    nvp, 6));
689	}
690
691	return (0);
692}
693
694/*
695 * If the bootfs property value is dsobj, clear it.
696 */
697void
698spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
699{
700	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
701		VERIFY(zap_remove(spa->spa_meta_objset,
702		    spa->spa_pool_props_object,
703		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
704		spa->spa_bootfs = 0;
705	}
706}
707
708/*ARGSUSED*/
709static int
710spa_change_guid_check(void *arg, dmu_tx_t *tx)
711{
712	uint64_t *newguid = arg;
713	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
714	vdev_t *rvd = spa->spa_root_vdev;
715	uint64_t vdev_state;
716
717	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
718	vdev_state = rvd->vdev_state;
719	spa_config_exit(spa, SCL_STATE, FTAG);
720
721	if (vdev_state != VDEV_STATE_HEALTHY)
722		return (SET_ERROR(ENXIO));
723
724	ASSERT3U(spa_guid(spa), !=, *newguid);
725
726	return (0);
727}
728
729static void
730spa_change_guid_sync(void *arg, dmu_tx_t *tx)
731{
732	uint64_t *newguid = arg;
733	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
734	uint64_t oldguid;
735	vdev_t *rvd = spa->spa_root_vdev;
736
737	oldguid = spa_guid(spa);
738
739	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
740	rvd->vdev_guid = *newguid;
741	rvd->vdev_guid_sum += (*newguid - oldguid);
742	vdev_config_dirty(rvd);
743	spa_config_exit(spa, SCL_STATE, FTAG);
744
745	spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
746	    oldguid, *newguid);
747}
748
749/*
750 * Change the GUID for the pool.  This is done so that we can later
751 * re-import a pool built from a clone of our own vdevs.  We will modify
752 * the root vdev's guid, our own pool guid, and then mark all of our
753 * vdevs dirty.  Note that we must make sure that all our vdevs are
754 * online when we do this, or else any vdevs that weren't present
755 * would be orphaned from our pool.  We are also going to issue a
756 * sysevent to update any watchers.
757 */
758int
759spa_change_guid(spa_t *spa)
760{
761	int error;
762	uint64_t guid;
763
764	mutex_enter(&spa->spa_vdev_top_lock);
765	mutex_enter(&spa_namespace_lock);
766	guid = spa_generate_guid(NULL);
767
768	error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
769	    spa_change_guid_sync, &guid, 5);
770
771	if (error == 0) {
772		spa_config_sync(spa, B_FALSE, B_TRUE);
773		spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
774	}
775
776	mutex_exit(&spa_namespace_lock);
777	mutex_exit(&spa->spa_vdev_top_lock);
778
779	return (error);
780}
781
782/*
783 * ==========================================================================
784 * SPA state manipulation (open/create/destroy/import/export)
785 * ==========================================================================
786 */
787
788static int
789spa_error_entry_compare(const void *a, const void *b)
790{
791	spa_error_entry_t *sa = (spa_error_entry_t *)a;
792	spa_error_entry_t *sb = (spa_error_entry_t *)b;
793	int ret;
794
795	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
796	    sizeof (zbookmark_phys_t));
797
798	if (ret < 0)
799		return (-1);
800	else if (ret > 0)
801		return (1);
802	else
803		return (0);
804}
805
806/*
807 * Utility function which retrieves copies of the current logs and
808 * re-initializes them in the process.
809 */
810void
811spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
812{
813	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
814
815	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
816	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
817
818	avl_create(&spa->spa_errlist_scrub,
819	    spa_error_entry_compare, sizeof (spa_error_entry_t),
820	    offsetof(spa_error_entry_t, se_avl));
821	avl_create(&spa->spa_errlist_last,
822	    spa_error_entry_compare, sizeof (spa_error_entry_t),
823	    offsetof(spa_error_entry_t, se_avl));
824}
825
826static void
827spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
828{
829	const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
830	enum zti_modes mode = ztip->zti_mode;
831	uint_t value = ztip->zti_value;
832	uint_t count = ztip->zti_count;
833	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
834	char name[32];
835	uint_t flags = 0;
836	boolean_t batch = B_FALSE;
837
838	if (mode == ZTI_MODE_NULL) {
839		tqs->stqs_count = 0;
840		tqs->stqs_taskq = NULL;
841		return;
842	}
843
844	ASSERT3U(count, >, 0);
845
846	tqs->stqs_count = count;
847	tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
848
849	switch (mode) {
850	case ZTI_MODE_FIXED:
851		ASSERT3U(value, >=, 1);
852		value = MAX(value, 1);
853		break;
854
855	case ZTI_MODE_BATCH:
856		batch = B_TRUE;
857		flags |= TASKQ_THREADS_CPU_PCT;
858		value = zio_taskq_batch_pct;
859		break;
860
861	default:
862		panic("unrecognized mode for %s_%s taskq (%u:%u) in "
863		    "spa_activate()",
864		    zio_type_name[t], zio_taskq_types[q], mode, value);
865		break;
866	}
867
868	for (uint_t i = 0; i < count; i++) {
869		taskq_t *tq;
870
871		if (count > 1) {
872			(void) snprintf(name, sizeof (name), "%s_%s_%u",
873			    zio_type_name[t], zio_taskq_types[q], i);
874		} else {
875			(void) snprintf(name, sizeof (name), "%s_%s",
876			    zio_type_name[t], zio_taskq_types[q]);
877		}
878
879		if (zio_taskq_sysdc && spa->spa_proc != &p0) {
880			if (batch)
881				flags |= TASKQ_DC_BATCH;
882
883			tq = taskq_create_sysdc(name, value, 50, INT_MAX,
884			    spa->spa_proc, zio_taskq_basedc, flags);
885		} else {
886			pri_t pri = maxclsyspri;
887			/*
888			 * The write issue taskq can be extremely CPU
889			 * intensive.  Run it at slightly lower priority
890			 * than the other taskqs.
891			 */
892			if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
893				pri--;
894
895			tq = taskq_create_proc(name, value, pri, 50,
896			    INT_MAX, spa->spa_proc, flags);
897		}
898
899		tqs->stqs_taskq[i] = tq;
900	}
901}
902
903static void
904spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
905{
906	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
907
908	if (tqs->stqs_taskq == NULL) {
909		ASSERT0(tqs->stqs_count);
910		return;
911	}
912
913	for (uint_t i = 0; i < tqs->stqs_count; i++) {
914		ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
915		taskq_destroy(tqs->stqs_taskq[i]);
916	}
917
918	kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
919	tqs->stqs_taskq = NULL;
920}
921
922/*
923 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
924 * Note that a type may have multiple discrete taskqs to avoid lock contention
925 * on the taskq itself. In that case we choose which taskq at random by using
926 * the low bits of gethrtime().
927 */
928void
929spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
930    task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
931{
932	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
933	taskq_t *tq;
934
935	ASSERT3P(tqs->stqs_taskq, !=, NULL);
936	ASSERT3U(tqs->stqs_count, !=, 0);
937
938	if (tqs->stqs_count == 1) {
939		tq = tqs->stqs_taskq[0];
940	} else {
941		tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
942	}
943
944	taskq_dispatch_ent(tq, func, arg, flags, ent);
945}
946
947static void
948spa_create_zio_taskqs(spa_t *spa)
949{
950	for (int t = 0; t < ZIO_TYPES; t++) {
951		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
952			spa_taskqs_init(spa, t, q);
953		}
954	}
955}
956
957#ifdef _KERNEL
958static void
959spa_thread(void *arg)
960{
961	callb_cpr_t cprinfo;
962
963	spa_t *spa = arg;
964	user_t *pu = PTOU(curproc);
965
966	CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
967	    spa->spa_name);
968
969	ASSERT(curproc != &p0);
970	(void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
971	    "zpool-%s", spa->spa_name);
972	(void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
973
974	/* bind this thread to the requested psrset */
975	if (zio_taskq_psrset_bind != PS_NONE) {
976		pool_lock();
977		mutex_enter(&cpu_lock);
978		mutex_enter(&pidlock);
979		mutex_enter(&curproc->p_lock);
980
981		if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
982		    0, NULL, NULL) == 0)  {
983			curthread->t_bind_pset = zio_taskq_psrset_bind;
984		} else {
985			cmn_err(CE_WARN,
986			    "Couldn't bind process for zfs pool \"%s\" to "
987			    "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
988		}
989
990		mutex_exit(&curproc->p_lock);
991		mutex_exit(&pidlock);
992		mutex_exit(&cpu_lock);
993		pool_unlock();
994	}
995
996	if (zio_taskq_sysdc) {
997		sysdc_thread_enter(curthread, 100, 0);
998	}
999
1000	spa->spa_proc = curproc;
1001	spa->spa_did = curthread->t_did;
1002
1003	spa_create_zio_taskqs(spa);
1004
1005	mutex_enter(&spa->spa_proc_lock);
1006	ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1007
1008	spa->spa_proc_state = SPA_PROC_ACTIVE;
1009	cv_broadcast(&spa->spa_proc_cv);
1010
1011	CALLB_CPR_SAFE_BEGIN(&cprinfo);
1012	while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1013		cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1014	CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1015
1016	ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1017	spa->spa_proc_state = SPA_PROC_GONE;
1018	spa->spa_proc = &p0;
1019	cv_broadcast(&spa->spa_proc_cv);
1020	CALLB_CPR_EXIT(&cprinfo);	/* drops spa_proc_lock */
1021
1022	mutex_enter(&curproc->p_lock);
1023	lwp_exit();
1024}
1025#endif
1026
1027/*
1028 * Activate an uninitialized pool.
1029 */
1030static void
1031spa_activate(spa_t *spa, int mode)
1032{
1033	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1034
1035	spa->spa_state = POOL_STATE_ACTIVE;
1036	spa->spa_mode = mode;
1037
1038	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1039	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1040
1041	/* Try to create a covering process */
1042	mutex_enter(&spa->spa_proc_lock);
1043	ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1044	ASSERT(spa->spa_proc == &p0);
1045	spa->spa_did = 0;
1046
1047	/* Only create a process if we're going to be around a while. */
1048	if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1049		if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1050		    NULL, 0) == 0) {
1051			spa->spa_proc_state = SPA_PROC_CREATED;
1052			while (spa->spa_proc_state == SPA_PROC_CREATED) {
1053				cv_wait(&spa->spa_proc_cv,
1054				    &spa->spa_proc_lock);
1055			}
1056			ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1057			ASSERT(spa->spa_proc != &p0);
1058			ASSERT(spa->spa_did != 0);
1059		} else {
1060#ifdef _KERNEL
1061			cmn_err(CE_WARN,
1062			    "Couldn't create process for zfs pool \"%s\"\n",
1063			    spa->spa_name);
1064#endif
1065		}
1066	}
1067	mutex_exit(&spa->spa_proc_lock);
1068
1069	/* If we didn't create a process, we need to create our taskqs. */
1070	if (spa->spa_proc == &p0) {
1071		spa_create_zio_taskqs(spa);
1072	}
1073
1074	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1075	    offsetof(vdev_t, vdev_config_dirty_node));
1076	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1077	    offsetof(vdev_t, vdev_state_dirty_node));
1078
1079	txg_list_create(&spa->spa_vdev_txg_list,
1080	    offsetof(struct vdev, vdev_txg_node));
1081
1082	avl_create(&spa->spa_errlist_scrub,
1083	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1084	    offsetof(spa_error_entry_t, se_avl));
1085	avl_create(&spa->spa_errlist_last,
1086	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1087	    offsetof(spa_error_entry_t, se_avl));
1088}
1089
1090/*
1091 * Opposite of spa_activate().
1092 */
1093static void
1094spa_deactivate(spa_t *spa)
1095{
1096	ASSERT(spa->spa_sync_on == B_FALSE);
1097	ASSERT(spa->spa_dsl_pool == NULL);
1098	ASSERT(spa->spa_root_vdev == NULL);
1099	ASSERT(spa->spa_async_zio_root == NULL);
1100	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1101
1102	txg_list_destroy(&spa->spa_vdev_txg_list);
1103
1104	list_destroy(&spa->spa_config_dirty_list);
1105	list_destroy(&spa->spa_state_dirty_list);
1106
1107	for (int t = 0; t < ZIO_TYPES; t++) {
1108		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1109			spa_taskqs_fini(spa, t, q);
1110		}
1111	}
1112
1113	metaslab_class_destroy(spa->spa_normal_class);
1114	spa->spa_normal_class = NULL;
1115
1116	metaslab_class_destroy(spa->spa_log_class);
1117	spa->spa_log_class = NULL;
1118
1119	/*
1120	 * If this was part of an import or the open otherwise failed, we may
1121	 * still have errors left in the queues.  Empty them just in case.
1122	 */
1123	spa_errlog_drain(spa);
1124
1125	avl_destroy(&spa->spa_errlist_scrub);
1126	avl_destroy(&spa->spa_errlist_last);
1127
1128	spa->spa_state = POOL_STATE_UNINITIALIZED;
1129
1130	mutex_enter(&spa->spa_proc_lock);
1131	if (spa->spa_proc_state != SPA_PROC_NONE) {
1132		ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1133		spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1134		cv_broadcast(&spa->spa_proc_cv);
1135		while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1136			ASSERT(spa->spa_proc != &p0);
1137			cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1138		}
1139		ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1140		spa->spa_proc_state = SPA_PROC_NONE;
1141	}
1142	ASSERT(spa->spa_proc == &p0);
1143	mutex_exit(&spa->spa_proc_lock);
1144
1145	/*
1146	 * We want to make sure spa_thread() has actually exited the ZFS
1147	 * module, so that the module can't be unloaded out from underneath
1148	 * it.
1149	 */
1150	if (spa->spa_did != 0) {
1151		thread_join(spa->spa_did);
1152		spa->spa_did = 0;
1153	}
1154}
1155
1156/*
1157 * Verify a pool configuration, and construct the vdev tree appropriately.  This
1158 * will create all the necessary vdevs in the appropriate layout, with each vdev
1159 * in the CLOSED state.  This will prep the pool before open/creation/import.
1160 * All vdev validation is done by the vdev_alloc() routine.
1161 */
1162static int
1163spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1164    uint_t id, int atype)
1165{
1166	nvlist_t **child;
1167	uint_t children;
1168	int error;
1169
1170	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1171		return (error);
1172
1173	if ((*vdp)->vdev_ops->vdev_op_leaf)
1174		return (0);
1175
1176	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1177	    &child, &children);
1178
1179	if (error == ENOENT)
1180		return (0);
1181
1182	if (error) {
1183		vdev_free(*vdp);
1184		*vdp = NULL;
1185		return (SET_ERROR(EINVAL));
1186	}
1187
1188	for (int c = 0; c < children; c++) {
1189		vdev_t *vd;
1190		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1191		    atype)) != 0) {
1192			vdev_free(*vdp);
1193			*vdp = NULL;
1194			return (error);
1195		}
1196	}
1197
1198	ASSERT(*vdp != NULL);
1199
1200	return (0);
1201}
1202
1203/*
1204 * Opposite of spa_load().
1205 */
1206static void
1207spa_unload(spa_t *spa)
1208{
1209	int i;
1210
1211	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1212
1213	/*
1214	 * Stop async tasks.
1215	 */
1216	spa_async_suspend(spa);
1217
1218	/*
1219	 * Stop syncing.
1220	 */
1221	if (spa->spa_sync_on) {
1222		txg_sync_stop(spa->spa_dsl_pool);
1223		spa->spa_sync_on = B_FALSE;
1224	}
1225
1226	/*
1227	 * Wait for any outstanding async I/O to complete.
1228	 */
1229	if (spa->spa_async_zio_root != NULL) {
1230		(void) zio_wait(spa->spa_async_zio_root);
1231		spa->spa_async_zio_root = NULL;
1232	}
1233
1234	bpobj_close(&spa->spa_deferred_bpobj);
1235
1236	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1237
1238	/*
1239	 * Close all vdevs.
1240	 */
1241	if (spa->spa_root_vdev)
1242		vdev_free(spa->spa_root_vdev);
1243	ASSERT(spa->spa_root_vdev == NULL);
1244
1245	/*
1246	 * Close the dsl pool.
1247	 */
1248	if (spa->spa_dsl_pool) {
1249		dsl_pool_close(spa->spa_dsl_pool);
1250		spa->spa_dsl_pool = NULL;
1251		spa->spa_meta_objset = NULL;
1252	}
1253
1254	ddt_unload(spa);
1255
1256
1257	/*
1258	 * Drop and purge level 2 cache
1259	 */
1260	spa_l2cache_drop(spa);
1261
1262	for (i = 0; i < spa->spa_spares.sav_count; i++)
1263		vdev_free(spa->spa_spares.sav_vdevs[i]);
1264	if (spa->spa_spares.sav_vdevs) {
1265		kmem_free(spa->spa_spares.sav_vdevs,
1266		    spa->spa_spares.sav_count * sizeof (void *));
1267		spa->spa_spares.sav_vdevs = NULL;
1268	}
1269	if (spa->spa_spares.sav_config) {
1270		nvlist_free(spa->spa_spares.sav_config);
1271		spa->spa_spares.sav_config = NULL;
1272	}
1273	spa->spa_spares.sav_count = 0;
1274
1275	for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1276		vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1277		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1278	}
1279	if (spa->spa_l2cache.sav_vdevs) {
1280		kmem_free(spa->spa_l2cache.sav_vdevs,
1281		    spa->spa_l2cache.sav_count * sizeof (void *));
1282		spa->spa_l2cache.sav_vdevs = NULL;
1283	}
1284	if (spa->spa_l2cache.sav_config) {
1285		nvlist_free(spa->spa_l2cache.sav_config);
1286		spa->spa_l2cache.sav_config = NULL;
1287	}
1288	spa->spa_l2cache.sav_count = 0;
1289
1290	spa->spa_async_suspended = 0;
1291
1292	if (spa->spa_comment != NULL) {
1293		spa_strfree(spa->spa_comment);
1294		spa->spa_comment = NULL;
1295	}
1296
1297	spa_config_exit(spa, SCL_ALL, FTAG);
1298}
1299
1300/*
1301 * Load (or re-load) the current list of vdevs describing the active spares for
1302 * this pool.  When this is called, we have some form of basic information in
1303 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1304 * then re-generate a more complete list including status information.
1305 */
1306static void
1307spa_load_spares(spa_t *spa)
1308{
1309	nvlist_t **spares;
1310	uint_t nspares;
1311	int i;
1312	vdev_t *vd, *tvd;
1313
1314	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1315
1316	/*
1317	 * First, close and free any existing spare vdevs.
1318	 */
1319	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1320		vd = spa->spa_spares.sav_vdevs[i];
1321
1322		/* Undo the call to spa_activate() below */
1323		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1324		    B_FALSE)) != NULL && tvd->vdev_isspare)
1325			spa_spare_remove(tvd);
1326		vdev_close(vd);
1327		vdev_free(vd);
1328	}
1329
1330	if (spa->spa_spares.sav_vdevs)
1331		kmem_free(spa->spa_spares.sav_vdevs,
1332		    spa->spa_spares.sav_count * sizeof (void *));
1333
1334	if (spa->spa_spares.sav_config == NULL)
1335		nspares = 0;
1336	else
1337		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1338		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1339
1340	spa->spa_spares.sav_count = (int)nspares;
1341	spa->spa_spares.sav_vdevs = NULL;
1342
1343	if (nspares == 0)
1344		return;
1345
1346	/*
1347	 * Construct the array of vdevs, opening them to get status in the
1348	 * process.   For each spare, there is potentially two different vdev_t
1349	 * structures associated with it: one in the list of spares (used only
1350	 * for basic validation purposes) and one in the active vdev
1351	 * configuration (if it's spared in).  During this phase we open and
1352	 * validate each vdev on the spare list.  If the vdev also exists in the
1353	 * active configuration, then we also mark this vdev as an active spare.
1354	 */
1355	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1356	    KM_SLEEP);
1357	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1358		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1359		    VDEV_ALLOC_SPARE) == 0);
1360		ASSERT(vd != NULL);
1361
1362		spa->spa_spares.sav_vdevs[i] = vd;
1363
1364		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1365		    B_FALSE)) != NULL) {
1366			if (!tvd->vdev_isspare)
1367				spa_spare_add(tvd);
1368
1369			/*
1370			 * We only mark the spare active if we were successfully
1371			 * able to load the vdev.  Otherwise, importing a pool
1372			 * with a bad active spare would result in strange
1373			 * behavior, because multiple pool would think the spare
1374			 * is actively in use.
1375			 *
1376			 * There is a vulnerability here to an equally bizarre
1377			 * circumstance, where a dead active spare is later
1378			 * brought back to life (onlined or otherwise).  Given
1379			 * the rarity of this scenario, and the extra complexity
1380			 * it adds, we ignore the possibility.
1381			 */
1382			if (!vdev_is_dead(tvd))
1383				spa_spare_activate(tvd);
1384		}
1385
1386		vd->vdev_top = vd;
1387		vd->vdev_aux = &spa->spa_spares;
1388
1389		if (vdev_open(vd) != 0)
1390			continue;
1391
1392		if (vdev_validate_aux(vd) == 0)
1393			spa_spare_add(vd);
1394	}
1395
1396	/*
1397	 * Recompute the stashed list of spares, with status information
1398	 * this time.
1399	 */
1400	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1401	    DATA_TYPE_NVLIST_ARRAY) == 0);
1402
1403	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1404	    KM_SLEEP);
1405	for (i = 0; i < spa->spa_spares.sav_count; i++)
1406		spares[i] = vdev_config_generate(spa,
1407		    spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1408	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1409	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1410	for (i = 0; i < spa->spa_spares.sav_count; i++)
1411		nvlist_free(spares[i]);
1412	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1413}
1414
1415/*
1416 * Load (or re-load) the current list of vdevs describing the active l2cache for
1417 * this pool.  When this is called, we have some form of basic information in
1418 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1419 * then re-generate a more complete list including status information.
1420 * Devices which are already active have their details maintained, and are
1421 * not re-opened.
1422 */
1423static void
1424spa_load_l2cache(spa_t *spa)
1425{
1426	nvlist_t **l2cache;
1427	uint_t nl2cache;
1428	int i, j, oldnvdevs;
1429	uint64_t guid;
1430	vdev_t *vd, **oldvdevs, **newvdevs;
1431	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1432
1433	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1434
1435	if (sav->sav_config != NULL) {
1436		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1437		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1438		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1439	} else {
1440		nl2cache = 0;
1441		newvdevs = NULL;
1442	}
1443
1444	oldvdevs = sav->sav_vdevs;
1445	oldnvdevs = sav->sav_count;
1446	sav->sav_vdevs = NULL;
1447	sav->sav_count = 0;
1448
1449	/*
1450	 * Process new nvlist of vdevs.
1451	 */
1452	for (i = 0; i < nl2cache; i++) {
1453		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1454		    &guid) == 0);
1455
1456		newvdevs[i] = NULL;
1457		for (j = 0; j < oldnvdevs; j++) {
1458			vd = oldvdevs[j];
1459			if (vd != NULL && guid == vd->vdev_guid) {
1460				/*
1461				 * Retain previous vdev for add/remove ops.
1462				 */
1463				newvdevs[i] = vd;
1464				oldvdevs[j] = NULL;
1465				break;
1466			}
1467		}
1468
1469		if (newvdevs[i] == NULL) {
1470			/*
1471			 * Create new vdev
1472			 */
1473			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1474			    VDEV_ALLOC_L2CACHE) == 0);
1475			ASSERT(vd != NULL);
1476			newvdevs[i] = vd;
1477
1478			/*
1479			 * Commit this vdev as an l2cache device,
1480			 * even if it fails to open.
1481			 */
1482			spa_l2cache_add(vd);
1483
1484			vd->vdev_top = vd;
1485			vd->vdev_aux = sav;
1486
1487			spa_l2cache_activate(vd);
1488
1489			if (vdev_open(vd) != 0)
1490				continue;
1491
1492			(void) vdev_validate_aux(vd);
1493
1494			if (!vdev_is_dead(vd))
1495				l2arc_add_vdev(spa, vd);
1496		}
1497	}
1498
1499	/*
1500	 * Purge vdevs that were dropped
1501	 */
1502	for (i = 0; i < oldnvdevs; i++) {
1503		uint64_t pool;
1504
1505		vd = oldvdevs[i];
1506		if (vd != NULL) {
1507			ASSERT(vd->vdev_isl2cache);
1508
1509			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1510			    pool != 0ULL && l2arc_vdev_present(vd))
1511				l2arc_remove_vdev(vd);
1512			vdev_clear_stats(vd);
1513			vdev_free(vd);
1514		}
1515	}
1516
1517	if (oldvdevs)
1518		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1519
1520	if (sav->sav_config == NULL)
1521		goto out;
1522
1523	sav->sav_vdevs = newvdevs;
1524	sav->sav_count = (int)nl2cache;
1525
1526	/*
1527	 * Recompute the stashed list of l2cache devices, with status
1528	 * information this time.
1529	 */
1530	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1531	    DATA_TYPE_NVLIST_ARRAY) == 0);
1532
1533	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1534	for (i = 0; i < sav->sav_count; i++)
1535		l2cache[i] = vdev_config_generate(spa,
1536		    sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1537	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1538	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1539out:
1540	for (i = 0; i < sav->sav_count; i++)
1541		nvlist_free(l2cache[i]);
1542	if (sav->sav_count)
1543		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1544}
1545
1546static int
1547load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1548{
1549	dmu_buf_t *db;
1550	char *packed = NULL;
1551	size_t nvsize = 0;
1552	int error;
1553	*value = NULL;
1554
1555	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1556	nvsize = *(uint64_t *)db->db_data;
1557	dmu_buf_rele(db, FTAG);
1558
1559	packed = kmem_alloc(nvsize, KM_SLEEP);
1560	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1561	    DMU_READ_PREFETCH);
1562	if (error == 0)
1563		error = nvlist_unpack(packed, nvsize, value, 0);
1564	kmem_free(packed, nvsize);
1565
1566	return (error);
1567}
1568
1569/*
1570 * Checks to see if the given vdev could not be opened, in which case we post a
1571 * sysevent to notify the autoreplace code that the device has been removed.
1572 */
1573static void
1574spa_check_removed(vdev_t *vd)
1575{
1576	for (int c = 0; c < vd->vdev_children; c++)
1577		spa_check_removed(vd->vdev_child[c]);
1578
1579	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1580	    !vd->vdev_ishole) {
1581		zfs_post_autoreplace(vd->vdev_spa, vd);
1582		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1583	}
1584}
1585
1586/*
1587 * Validate the current config against the MOS config
1588 */
1589static boolean_t
1590spa_config_valid(spa_t *spa, nvlist_t *config)
1591{
1592	vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1593	nvlist_t *nv;
1594
1595	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1596
1597	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1598	VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1599
1600	ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1601
1602	/*
1603	 * If we're doing a normal import, then build up any additional
1604	 * diagnostic information about missing devices in this config.
1605	 * We'll pass this up to the user for further processing.
1606	 */
1607	if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1608		nvlist_t **child, *nv;
1609		uint64_t idx = 0;
1610
1611		child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1612		    KM_SLEEP);
1613		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1614
1615		for (int c = 0; c < rvd->vdev_children; c++) {
1616			vdev_t *tvd = rvd->vdev_child[c];
1617			vdev_t *mtvd  = mrvd->vdev_child[c];
1618
1619			if (tvd->vdev_ops == &vdev_missing_ops &&
1620			    mtvd->vdev_ops != &vdev_missing_ops &&
1621			    mtvd->vdev_islog)
1622				child[idx++] = vdev_config_generate(spa, mtvd,
1623				    B_FALSE, 0);
1624		}
1625
1626		if (idx) {
1627			VERIFY(nvlist_add_nvlist_array(nv,
1628			    ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1629			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1630			    ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1631
1632			for (int i = 0; i < idx; i++)
1633				nvlist_free(child[i]);
1634		}
1635		nvlist_free(nv);
1636		kmem_free(child, rvd->vdev_children * sizeof (char **));
1637	}
1638
1639	/*
1640	 * Compare the root vdev tree with the information we have
1641	 * from the MOS config (mrvd). Check each top-level vdev
1642	 * with the corresponding MOS config top-level (mtvd).
1643	 */
1644	for (int c = 0; c < rvd->vdev_children; c++) {
1645		vdev_t *tvd = rvd->vdev_child[c];
1646		vdev_t *mtvd  = mrvd->vdev_child[c];
1647
1648		/*
1649		 * Resolve any "missing" vdevs in the current configuration.
1650		 * If we find that the MOS config has more accurate information
1651		 * about the top-level vdev then use that vdev instead.
1652		 */
1653		if (tvd->vdev_ops == &vdev_missing_ops &&
1654		    mtvd->vdev_ops != &vdev_missing_ops) {
1655
1656			if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1657				continue;
1658
1659			/*
1660			 * Device specific actions.
1661			 */
1662			if (mtvd->vdev_islog) {
1663				spa_set_log_state(spa, SPA_LOG_CLEAR);
1664			} else {
1665				/*
1666				 * XXX - once we have 'readonly' pool
1667				 * support we should be able to handle
1668				 * missing data devices by transitioning
1669				 * the pool to readonly.
1670				 */
1671				continue;
1672			}
1673
1674			/*
1675			 * Swap the missing vdev with the data we were
1676			 * able to obtain from the MOS config.
1677			 */
1678			vdev_remove_child(rvd, tvd);
1679			vdev_remove_child(mrvd, mtvd);
1680
1681			vdev_add_child(rvd, mtvd);
1682			vdev_add_child(mrvd, tvd);
1683
1684			spa_config_exit(spa, SCL_ALL, FTAG);
1685			vdev_load(mtvd);
1686			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1687
1688			vdev_reopen(rvd);
1689		} else if (mtvd->vdev_islog) {
1690			/*
1691			 * Load the slog device's state from the MOS config
1692			 * since it's possible that the label does not
1693			 * contain the most up-to-date information.
1694			 */
1695			vdev_load_log_state(tvd, mtvd);
1696			vdev_reopen(tvd);
1697		}
1698	}
1699	vdev_free(mrvd);
1700	spa_config_exit(spa, SCL_ALL, FTAG);
1701
1702	/*
1703	 * Ensure we were able to validate the config.
1704	 */
1705	return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1706}
1707
1708/*
1709 * Check for missing log devices
1710 */
1711static boolean_t
1712spa_check_logs(spa_t *spa)
1713{
1714	boolean_t rv = B_FALSE;
1715
1716	switch (spa->spa_log_state) {
1717	case SPA_LOG_MISSING:
1718		/* need to recheck in case slog has been restored */
1719	case SPA_LOG_UNKNOWN:
1720		rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1721		    NULL, DS_FIND_CHILDREN) != 0);
1722		if (rv)
1723			spa_set_log_state(spa, SPA_LOG_MISSING);
1724		break;
1725	}
1726	return (rv);
1727}
1728
1729static boolean_t
1730spa_passivate_log(spa_t *spa)
1731{
1732	vdev_t *rvd = spa->spa_root_vdev;
1733	boolean_t slog_found = B_FALSE;
1734
1735	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1736
1737	if (!spa_has_slogs(spa))
1738		return (B_FALSE);
1739
1740	for (int c = 0; c < rvd->vdev_children; c++) {
1741		vdev_t *tvd = rvd->vdev_child[c];
1742		metaslab_group_t *mg = tvd->vdev_mg;
1743
1744		if (tvd->vdev_islog) {
1745			metaslab_group_passivate(mg);
1746			slog_found = B_TRUE;
1747		}
1748	}
1749
1750	return (slog_found);
1751}
1752
1753static void
1754spa_activate_log(spa_t *spa)
1755{
1756	vdev_t *rvd = spa->spa_root_vdev;
1757
1758	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1759
1760	for (int c = 0; c < rvd->vdev_children; c++) {
1761		vdev_t *tvd = rvd->vdev_child[c];
1762		metaslab_group_t *mg = tvd->vdev_mg;
1763
1764		if (tvd->vdev_islog)
1765			metaslab_group_activate(mg);
1766	}
1767}
1768
1769int
1770spa_offline_log(spa_t *spa)
1771{
1772	int error;
1773
1774	error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1775	    NULL, DS_FIND_CHILDREN);
1776	if (error == 0) {
1777		/*
1778		 * We successfully offlined the log device, sync out the
1779		 * current txg so that the "stubby" block can be removed
1780		 * by zil_sync().
1781		 */
1782		txg_wait_synced(spa->spa_dsl_pool, 0);
1783	}
1784	return (error);
1785}
1786
1787static void
1788spa_aux_check_removed(spa_aux_vdev_t *sav)
1789{
1790	for (int i = 0; i < sav->sav_count; i++)
1791		spa_check_removed(sav->sav_vdevs[i]);
1792}
1793
1794void
1795spa_claim_notify(zio_t *zio)
1796{
1797	spa_t *spa = zio->io_spa;
1798
1799	if (zio->io_error)
1800		return;
1801
1802	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1803	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1804		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1805	mutex_exit(&spa->spa_props_lock);
1806}
1807
1808typedef struct spa_load_error {
1809	uint64_t	sle_meta_count;
1810	uint64_t	sle_data_count;
1811} spa_load_error_t;
1812
1813static void
1814spa_load_verify_done(zio_t *zio)
1815{
1816	blkptr_t *bp = zio->io_bp;
1817	spa_load_error_t *sle = zio->io_private;
1818	dmu_object_type_t type = BP_GET_TYPE(bp);
1819	int error = zio->io_error;
1820
1821	if (error) {
1822		if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1823		    type != DMU_OT_INTENT_LOG)
1824			atomic_add_64(&sle->sle_meta_count, 1);
1825		else
1826			atomic_add_64(&sle->sle_data_count, 1);
1827	}
1828	zio_data_buf_free(zio->io_data, zio->io_size);
1829}
1830
1831/*ARGSUSED*/
1832static int
1833spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1834    const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1835{
1836	if (!BP_IS_HOLE(bp) && !BP_IS_EMBEDDED(bp)) {
1837		zio_t *rio = arg;
1838		size_t size = BP_GET_PSIZE(bp);
1839		void *data = zio_data_buf_alloc(size);
1840
1841		zio_nowait(zio_read(rio, spa, bp, data, size,
1842		    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1843		    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1844		    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1845	}
1846	return (0);
1847}
1848
1849static int
1850spa_load_verify(spa_t *spa)
1851{
1852	zio_t *rio;
1853	spa_load_error_t sle = { 0 };
1854	zpool_rewind_policy_t policy;
1855	boolean_t verify_ok = B_FALSE;
1856	int error;
1857
1858	zpool_get_rewind_policy(spa->spa_config, &policy);
1859
1860	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1861		return (0);
1862
1863	rio = zio_root(spa, NULL, &sle,
1864	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1865
1866	error = traverse_pool(spa, spa->spa_verify_min_txg,
1867	    TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1868
1869	(void) zio_wait(rio);
1870
1871	spa->spa_load_meta_errors = sle.sle_meta_count;
1872	spa->spa_load_data_errors = sle.sle_data_count;
1873
1874	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1875	    sle.sle_data_count <= policy.zrp_maxdata) {
1876		int64_t loss = 0;
1877
1878		verify_ok = B_TRUE;
1879		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1880		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1881
1882		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1883		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1884		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1885		VERIFY(nvlist_add_int64(spa->spa_load_info,
1886		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1887		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1888		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1889	} else {
1890		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1891	}
1892
1893	if (error) {
1894		if (error != ENXIO && error != EIO)
1895			error = SET_ERROR(EIO);
1896		return (error);
1897	}
1898
1899	return (verify_ok ? 0 : EIO);
1900}
1901
1902/*
1903 * Find a value in the pool props object.
1904 */
1905static void
1906spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1907{
1908	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1909	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1910}
1911
1912/*
1913 * Find a value in the pool directory object.
1914 */
1915static int
1916spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1917{
1918	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1919	    name, sizeof (uint64_t), 1, val));
1920}
1921
1922static int
1923spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1924{
1925	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1926	return (err);
1927}
1928
1929/*
1930 * Fix up config after a partly-completed split.  This is done with the
1931 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1932 * pool have that entry in their config, but only the splitting one contains
1933 * a list of all the guids of the vdevs that are being split off.
1934 *
1935 * This function determines what to do with that list: either rejoin
1936 * all the disks to the pool, or complete the splitting process.  To attempt
1937 * the rejoin, each disk that is offlined is marked online again, and
1938 * we do a reopen() call.  If the vdev label for every disk that was
1939 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1940 * then we call vdev_split() on each disk, and complete the split.
1941 *
1942 * Otherwise we leave the config alone, with all the vdevs in place in
1943 * the original pool.
1944 */
1945static void
1946spa_try_repair(spa_t *spa, nvlist_t *config)
1947{
1948	uint_t extracted;
1949	uint64_t *glist;
1950	uint_t i, gcount;
1951	nvlist_t *nvl;
1952	vdev_t **vd;
1953	boolean_t attempt_reopen;
1954
1955	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1956		return;
1957
1958	/* check that the config is complete */
1959	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1960	    &glist, &gcount) != 0)
1961		return;
1962
1963	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1964
1965	/* attempt to online all the vdevs & validate */
1966	attempt_reopen = B_TRUE;
1967	for (i = 0; i < gcount; i++) {
1968		if (glist[i] == 0)	/* vdev is hole */
1969			continue;
1970
1971		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1972		if (vd[i] == NULL) {
1973			/*
1974			 * Don't bother attempting to reopen the disks;
1975			 * just do the split.
1976			 */
1977			attempt_reopen = B_FALSE;
1978		} else {
1979			/* attempt to re-online it */
1980			vd[i]->vdev_offline = B_FALSE;
1981		}
1982	}
1983
1984	if (attempt_reopen) {
1985		vdev_reopen(spa->spa_root_vdev);
1986
1987		/* check each device to see what state it's in */
1988		for (extracted = 0, i = 0; i < gcount; i++) {
1989			if (vd[i] != NULL &&
1990			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1991				break;
1992			++extracted;
1993		}
1994	}
1995
1996	/*
1997	 * If every disk has been moved to the new pool, or if we never
1998	 * even attempted to look at them, then we split them off for
1999	 * good.
2000	 */
2001	if (!attempt_reopen || gcount == extracted) {
2002		for (i = 0; i < gcount; i++)
2003			if (vd[i] != NULL)
2004				vdev_split(vd[i]);
2005		vdev_reopen(spa->spa_root_vdev);
2006	}
2007
2008	kmem_free(vd, gcount * sizeof (vdev_t *));
2009}
2010
2011static int
2012spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2013    boolean_t mosconfig)
2014{
2015	nvlist_t *config = spa->spa_config;
2016	char *ereport = FM_EREPORT_ZFS_POOL;
2017	char *comment;
2018	int error;
2019	uint64_t pool_guid;
2020	nvlist_t *nvl;
2021
2022	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2023		return (SET_ERROR(EINVAL));
2024
2025	ASSERT(spa->spa_comment == NULL);
2026	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2027		spa->spa_comment = spa_strdup(comment);
2028
2029	/*
2030	 * Versioning wasn't explicitly added to the label until later, so if
2031	 * it's not present treat it as the initial version.
2032	 */
2033	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2034	    &spa->spa_ubsync.ub_version) != 0)
2035		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2036
2037	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2038	    &spa->spa_config_txg);
2039
2040	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2041	    spa_guid_exists(pool_guid, 0)) {
2042		error = SET_ERROR(EEXIST);
2043	} else {
2044		spa->spa_config_guid = pool_guid;
2045
2046		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2047		    &nvl) == 0) {
2048			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2049			    KM_SLEEP) == 0);
2050		}
2051
2052		nvlist_free(spa->spa_load_info);
2053		spa->spa_load_info = fnvlist_alloc();
2054
2055		gethrestime(&spa->spa_loaded_ts);
2056		error = spa_load_impl(spa, pool_guid, config, state, type,
2057		    mosconfig, &ereport);
2058	}
2059
2060	spa->spa_minref = refcount_count(&spa->spa_refcount);
2061	if (error) {
2062		if (error != EEXIST) {
2063			spa->spa_loaded_ts.tv_sec = 0;
2064			spa->spa_loaded_ts.tv_nsec = 0;
2065		}
2066		if (error != EBADF) {
2067			zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2068		}
2069	}
2070	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2071	spa->spa_ena = 0;
2072
2073	return (error);
2074}
2075
2076/*
2077 * Load an existing storage pool, using the pool's builtin spa_config as a
2078 * source of configuration information.
2079 */
2080static int
2081spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2082    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2083    char **ereport)
2084{
2085	int error = 0;
2086	nvlist_t *nvroot = NULL;
2087	nvlist_t *label;
2088	vdev_t *rvd;
2089	uberblock_t *ub = &spa->spa_uberblock;
2090	uint64_t children, config_cache_txg = spa->spa_config_txg;
2091	int orig_mode = spa->spa_mode;
2092	int parse;
2093	uint64_t obj;
2094	boolean_t missing_feat_write = B_FALSE;
2095
2096	/*
2097	 * If this is an untrusted config, access the pool in read-only mode.
2098	 * This prevents things like resilvering recently removed devices.
2099	 */
2100	if (!mosconfig)
2101		spa->spa_mode = FREAD;
2102
2103	ASSERT(MUTEX_HELD(&spa_namespace_lock));
2104
2105	spa->spa_load_state = state;
2106
2107	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2108		return (SET_ERROR(EINVAL));
2109
2110	parse = (type == SPA_IMPORT_EXISTING ?
2111	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2112
2113	/*
2114	 * Create "The Godfather" zio to hold all async IOs
2115	 */
2116	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2117	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2118
2119	/*
2120	 * Parse the configuration into a vdev tree.  We explicitly set the
2121	 * value that will be returned by spa_version() since parsing the
2122	 * configuration requires knowing the version number.
2123	 */
2124	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2125	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2126	spa_config_exit(spa, SCL_ALL, FTAG);
2127
2128	if (error != 0)
2129		return (error);
2130
2131	ASSERT(spa->spa_root_vdev == rvd);
2132
2133	if (type != SPA_IMPORT_ASSEMBLE) {
2134		ASSERT(spa_guid(spa) == pool_guid);
2135	}
2136
2137	/*
2138	 * Try to open all vdevs, loading each label in the process.
2139	 */
2140	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2141	error = vdev_open(rvd);
2142	spa_config_exit(spa, SCL_ALL, FTAG);
2143	if (error != 0)
2144		return (error);
2145
2146	/*
2147	 * We need to validate the vdev labels against the configuration that
2148	 * we have in hand, which is dependent on the setting of mosconfig. If
2149	 * mosconfig is true then we're validating the vdev labels based on
2150	 * that config.  Otherwise, we're validating against the cached config
2151	 * (zpool.cache) that was read when we loaded the zfs module, and then
2152	 * later we will recursively call spa_load() and validate against
2153	 * the vdev config.
2154	 *
2155	 * If we're assembling a new pool that's been split off from an
2156	 * existing pool, the labels haven't yet been updated so we skip
2157	 * validation for now.
2158	 */
2159	if (type != SPA_IMPORT_ASSEMBLE) {
2160		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2161		error = vdev_validate(rvd, mosconfig);
2162		spa_config_exit(spa, SCL_ALL, FTAG);
2163
2164		if (error != 0)
2165			return (error);
2166
2167		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2168			return (SET_ERROR(ENXIO));
2169	}
2170
2171	/*
2172	 * Find the best uberblock.
2173	 */
2174	vdev_uberblock_load(rvd, ub, &label);
2175
2176	/*
2177	 * If we weren't able to find a single valid uberblock, return failure.
2178	 */
2179	if (ub->ub_txg == 0) {
2180		nvlist_free(label);
2181		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2182	}
2183
2184	/*
2185	 * If the pool has an unsupported version we can't open it.
2186	 */
2187	if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2188		nvlist_free(label);
2189		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2190	}
2191
2192	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2193		nvlist_t *features;
2194
2195		/*
2196		 * If we weren't able to find what's necessary for reading the
2197		 * MOS in the label, return failure.
2198		 */
2199		if (label == NULL || nvlist_lookup_nvlist(label,
2200		    ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2201			nvlist_free(label);
2202			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2203			    ENXIO));
2204		}
2205
2206		/*
2207		 * Update our in-core representation with the definitive values
2208		 * from the label.
2209		 */
2210		nvlist_free(spa->spa_label_features);
2211		VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2212	}
2213
2214	nvlist_free(label);
2215
2216	/*
2217	 * Look through entries in the label nvlist's features_for_read. If
2218	 * there is a feature listed there which we don't understand then we
2219	 * cannot open a pool.
2220	 */
2221	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2222		nvlist_t *unsup_feat;
2223
2224		VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2225		    0);
2226
2227		for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2228		    NULL); nvp != NULL;
2229		    nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2230			if (!zfeature_is_supported(nvpair_name(nvp))) {
2231				VERIFY(nvlist_add_string(unsup_feat,
2232				    nvpair_name(nvp), "") == 0);
2233			}
2234		}
2235
2236		if (!nvlist_empty(unsup_feat)) {
2237			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2238			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2239			nvlist_free(unsup_feat);
2240			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2241			    ENOTSUP));
2242		}
2243
2244		nvlist_free(unsup_feat);
2245	}
2246
2247	/*
2248	 * If the vdev guid sum doesn't match the uberblock, we have an
2249	 * incomplete configuration.  We first check to see if the pool
2250	 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2251	 * If it is, defer the vdev_guid_sum check till later so we
2252	 * can handle missing vdevs.
2253	 */
2254	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2255	    &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2256	    rvd->vdev_guid_sum != ub->ub_guid_sum)
2257		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2258
2259	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2260		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2261		spa_try_repair(spa, config);
2262		spa_config_exit(spa, SCL_ALL, FTAG);
2263		nvlist_free(spa->spa_config_splitting);
2264		spa->spa_config_splitting = NULL;
2265	}
2266
2267	/*
2268	 * Initialize internal SPA structures.
2269	 */
2270	spa->spa_state = POOL_STATE_ACTIVE;
2271	spa->spa_ubsync = spa->spa_uberblock;
2272	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2273	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2274	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2275	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2276	spa->spa_claim_max_txg = spa->spa_first_txg;
2277	spa->spa_prev_software_version = ub->ub_software_version;
2278
2279	error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2280	if (error)
2281		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2282	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2283
2284	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2285		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2286
2287	if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2288		boolean_t missing_feat_read = B_FALSE;
2289		nvlist_t *unsup_feat, *enabled_feat;
2290
2291		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2292		    &spa->spa_feat_for_read_obj) != 0) {
2293			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2294		}
2295
2296		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2297		    &spa->spa_feat_for_write_obj) != 0) {
2298			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2299		}
2300
2301		if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2302		    &spa->spa_feat_desc_obj) != 0) {
2303			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2304		}
2305
2306		enabled_feat = fnvlist_alloc();
2307		unsup_feat = fnvlist_alloc();
2308
2309		if (!spa_features_check(spa, B_FALSE,
2310		    unsup_feat, enabled_feat))
2311			missing_feat_read = B_TRUE;
2312
2313		if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2314			if (!spa_features_check(spa, B_TRUE,
2315			    unsup_feat, enabled_feat)) {
2316				missing_feat_write = B_TRUE;
2317			}
2318		}
2319
2320		fnvlist_add_nvlist(spa->spa_load_info,
2321		    ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2322
2323		if (!nvlist_empty(unsup_feat)) {
2324			fnvlist_add_nvlist(spa->spa_load_info,
2325			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2326		}
2327
2328		fnvlist_free(enabled_feat);
2329		fnvlist_free(unsup_feat);
2330
2331		if (!missing_feat_read) {
2332			fnvlist_add_boolean(spa->spa_load_info,
2333			    ZPOOL_CONFIG_CAN_RDONLY);
2334		}
2335
2336		/*
2337		 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2338		 * twofold: to determine whether the pool is available for
2339		 * import in read-write mode and (if it is not) whether the
2340		 * pool is available for import in read-only mode. If the pool
2341		 * is available for import in read-write mode, it is displayed
2342		 * as available in userland; if it is not available for import
2343		 * in read-only mode, it is displayed as unavailable in
2344		 * userland. If the pool is available for import in read-only
2345		 * mode but not read-write mode, it is displayed as unavailable
2346		 * in userland with a special note that the pool is actually
2347		 * available for open in read-only mode.
2348		 *
2349		 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2350		 * missing a feature for write, we must first determine whether
2351		 * the pool can be opened read-only before returning to
2352		 * userland in order to know whether to display the
2353		 * abovementioned note.
2354		 */
2355		if (missing_feat_read || (missing_feat_write &&
2356		    spa_writeable(spa))) {
2357			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2358			    ENOTSUP));
2359		}
2360
2361		/*
2362		 * Load refcounts for ZFS features from disk into an in-memory
2363		 * cache during SPA initialization.
2364		 */
2365		for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2366			uint64_t refcount;
2367
2368			error = feature_get_refcount_from_disk(spa,
2369			    &spa_feature_table[i], &refcount);
2370			if (error == 0) {
2371				spa->spa_feat_refcount_cache[i] = refcount;
2372			} else if (error == ENOTSUP) {
2373				spa->spa_feat_refcount_cache[i] =
2374				    SPA_FEATURE_DISABLED;
2375			} else {
2376				return (spa_vdev_err(rvd,
2377				    VDEV_AUX_CORRUPT_DATA, EIO));
2378			}
2379		}
2380	}
2381
2382	if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2383		if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2384		    &spa->spa_feat_enabled_txg_obj) != 0)
2385			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2386	}
2387
2388	spa->spa_is_initializing = B_TRUE;
2389	error = dsl_pool_open(spa->spa_dsl_pool);
2390	spa->spa_is_initializing = B_FALSE;
2391	if (error != 0)
2392		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2393
2394	if (!mosconfig) {
2395		uint64_t hostid;
2396		nvlist_t *policy = NULL, *nvconfig;
2397
2398		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2399			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2400
2401		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2402		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2403			char *hostname;
2404			unsigned long myhostid = 0;
2405
2406			VERIFY(nvlist_lookup_string(nvconfig,
2407			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2408
2409#ifdef	_KERNEL
2410			myhostid = zone_get_hostid(NULL);
2411#else	/* _KERNEL */
2412			/*
2413			 * We're emulating the system's hostid in userland, so
2414			 * we can't use zone_get_hostid().
2415			 */
2416			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2417#endif	/* _KERNEL */
2418			if (hostid != 0 && myhostid != 0 &&
2419			    hostid != myhostid) {
2420				nvlist_free(nvconfig);
2421				cmn_err(CE_WARN, "pool '%s' could not be "
2422				    "loaded as it was last accessed by "
2423				    "another system (host: %s hostid: 0x%lx). "
2424				    "See: http://illumos.org/msg/ZFS-8000-EY",
2425				    spa_name(spa), hostname,
2426				    (unsigned long)hostid);
2427				return (SET_ERROR(EBADF));
2428			}
2429		}
2430		if (nvlist_lookup_nvlist(spa->spa_config,
2431		    ZPOOL_REWIND_POLICY, &policy) == 0)
2432			VERIFY(nvlist_add_nvlist(nvconfig,
2433			    ZPOOL_REWIND_POLICY, policy) == 0);
2434
2435		spa_config_set(spa, nvconfig);
2436		spa_unload(spa);
2437		spa_deactivate(spa);
2438		spa_activate(spa, orig_mode);
2439
2440		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2441	}
2442
2443	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2444		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2445	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2446	if (error != 0)
2447		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2448
2449	/*
2450	 * Load the bit that tells us to use the new accounting function
2451	 * (raid-z deflation).  If we have an older pool, this will not
2452	 * be present.
2453	 */
2454	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2455	if (error != 0 && error != ENOENT)
2456		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2457
2458	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2459	    &spa->spa_creation_version);
2460	if (error != 0 && error != ENOENT)
2461		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2462
2463	/*
2464	 * Load the persistent error log.  If we have an older pool, this will
2465	 * not be present.
2466	 */
2467	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2468	if (error != 0 && error != ENOENT)
2469		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2470
2471	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2472	    &spa->spa_errlog_scrub);
2473	if (error != 0 && error != ENOENT)
2474		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2475
2476	/*
2477	 * Load the history object.  If we have an older pool, this
2478	 * will not be present.
2479	 */
2480	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2481	if (error != 0 && error != ENOENT)
2482		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2483
2484	/*
2485	 * If we're assembling the pool from the split-off vdevs of
2486	 * an existing pool, we don't want to attach the spares & cache
2487	 * devices.
2488	 */
2489
2490	/*
2491	 * Load any hot spares for this pool.
2492	 */
2493	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2494	if (error != 0 && error != ENOENT)
2495		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2496	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2497		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2498		if (load_nvlist(spa, spa->spa_spares.sav_object,
2499		    &spa->spa_spares.sav_config) != 0)
2500			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2501
2502		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2503		spa_load_spares(spa);
2504		spa_config_exit(spa, SCL_ALL, FTAG);
2505	} else if (error == 0) {
2506		spa->spa_spares.sav_sync = B_TRUE;
2507	}
2508
2509	/*
2510	 * Load any level 2 ARC devices for this pool.
2511	 */
2512	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2513	    &spa->spa_l2cache.sav_object);
2514	if (error != 0 && error != ENOENT)
2515		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2516	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2517		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2518		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2519		    &spa->spa_l2cache.sav_config) != 0)
2520			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2521
2522		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2523		spa_load_l2cache(spa);
2524		spa_config_exit(spa, SCL_ALL, FTAG);
2525	} else if (error == 0) {
2526		spa->spa_l2cache.sav_sync = B_TRUE;
2527	}
2528
2529	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2530
2531	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2532	if (error && error != ENOENT)
2533		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2534
2535	if (error == 0) {
2536		uint64_t autoreplace;
2537
2538		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2539		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2540		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2541		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2542		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2543		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2544		    &spa->spa_dedup_ditto);
2545
2546		spa->spa_autoreplace = (autoreplace != 0);
2547	}
2548
2549	/*
2550	 * If the 'autoreplace' property is set, then post a resource notifying
2551	 * the ZFS DE that it should not issue any faults for unopenable
2552	 * devices.  We also iterate over the vdevs, and post a sysevent for any
2553	 * unopenable vdevs so that the normal autoreplace handler can take
2554	 * over.
2555	 */
2556	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2557		spa_check_removed(spa->spa_root_vdev);
2558		/*
2559		 * For the import case, this is done in spa_import(), because
2560		 * at this point we're using the spare definitions from
2561		 * the MOS config, not necessarily from the userland config.
2562		 */
2563		if (state != SPA_LOAD_IMPORT) {
2564			spa_aux_check_removed(&spa->spa_spares);
2565			spa_aux_check_removed(&spa->spa_l2cache);
2566		}
2567	}
2568
2569	/*
2570	 * Load the vdev state for all toplevel vdevs.
2571	 */
2572	vdev_load(rvd);
2573
2574	/*
2575	 * Propagate the leaf DTLs we just loaded all the way up the tree.
2576	 */
2577	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2578	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2579	spa_config_exit(spa, SCL_ALL, FTAG);
2580
2581	/*
2582	 * Load the DDTs (dedup tables).
2583	 */
2584	error = ddt_load(spa);
2585	if (error != 0)
2586		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2587
2588	spa_update_dspace(spa);
2589
2590	/*
2591	 * Validate the config, using the MOS config to fill in any
2592	 * information which might be missing.  If we fail to validate
2593	 * the config then declare the pool unfit for use. If we're
2594	 * assembling a pool from a split, the log is not transferred
2595	 * over.
2596	 */
2597	if (type != SPA_IMPORT_ASSEMBLE) {
2598		nvlist_t *nvconfig;
2599
2600		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2601			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2602
2603		if (!spa_config_valid(spa, nvconfig)) {
2604			nvlist_free(nvconfig);
2605			return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2606			    ENXIO));
2607		}
2608		nvlist_free(nvconfig);
2609
2610		/*
2611		 * Now that we've validated the config, check the state of the
2612		 * root vdev.  If it can't be opened, it indicates one or
2613		 * more toplevel vdevs are faulted.
2614		 */
2615		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2616			return (SET_ERROR(ENXIO));
2617
2618		if (spa_check_logs(spa)) {
2619			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2620			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2621		}
2622	}
2623
2624	if (missing_feat_write) {
2625		ASSERT(state == SPA_LOAD_TRYIMPORT);
2626
2627		/*
2628		 * At this point, we know that we can open the pool in
2629		 * read-only mode but not read-write mode. We now have enough
2630		 * information and can return to userland.
2631		 */
2632		return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2633	}
2634
2635	/*
2636	 * We've successfully opened the pool, verify that we're ready
2637	 * to start pushing transactions.
2638	 */
2639	if (state != SPA_LOAD_TRYIMPORT) {
2640		if (error = spa_load_verify(spa))
2641			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2642			    error));
2643	}
2644
2645	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2646	    spa->spa_load_max_txg == UINT64_MAX)) {
2647		dmu_tx_t *tx;
2648		int need_update = B_FALSE;
2649
2650		ASSERT(state != SPA_LOAD_TRYIMPORT);
2651
2652		/*
2653		 * Claim log blocks that haven't been committed yet.
2654		 * This must all happen in a single txg.
2655		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2656		 * invoked from zil_claim_log_block()'s i/o done callback.
2657		 * Price of rollback is that we abandon the log.
2658		 */
2659		spa->spa_claiming = B_TRUE;
2660
2661		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2662		    spa_first_txg(spa));
2663		(void) dmu_objset_find(spa_name(spa),
2664		    zil_claim, tx, DS_FIND_CHILDREN);
2665		dmu_tx_commit(tx);
2666
2667		spa->spa_claiming = B_FALSE;
2668
2669		spa_set_log_state(spa, SPA_LOG_GOOD);
2670		spa->spa_sync_on = B_TRUE;
2671		txg_sync_start(spa->spa_dsl_pool);
2672
2673		/*
2674		 * Wait for all claims to sync.  We sync up to the highest
2675		 * claimed log block birth time so that claimed log blocks
2676		 * don't appear to be from the future.  spa_claim_max_txg
2677		 * will have been set for us by either zil_check_log_chain()
2678		 * (invoked from spa_check_logs()) or zil_claim() above.
2679		 */
2680		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2681
2682		/*
2683		 * If the config cache is stale, or we have uninitialized
2684		 * metaslabs (see spa_vdev_add()), then update the config.
2685		 *
2686		 * If this is a verbatim import, trust the current
2687		 * in-core spa_config and update the disk labels.
2688		 */
2689		if (config_cache_txg != spa->spa_config_txg ||
2690		    state == SPA_LOAD_IMPORT ||
2691		    state == SPA_LOAD_RECOVER ||
2692		    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2693			need_update = B_TRUE;
2694
2695		for (int c = 0; c < rvd->vdev_children; c++)
2696			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2697				need_update = B_TRUE;
2698
2699		/*
2700		 * Update the config cache asychronously in case we're the
2701		 * root pool, in which case the config cache isn't writable yet.
2702		 */
2703		if (need_update)
2704			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2705
2706		/*
2707		 * Check all DTLs to see if anything needs resilvering.
2708		 */
2709		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2710		    vdev_resilver_needed(rvd, NULL, NULL))
2711			spa_async_request(spa, SPA_ASYNC_RESILVER);
2712
2713		/*
2714		 * Log the fact that we booted up (so that we can detect if
2715		 * we rebooted in the middle of an operation).
2716		 */
2717		spa_history_log_version(spa, "open");
2718
2719		/*
2720		 * Delete any inconsistent datasets.
2721		 */
2722		(void) dmu_objset_find(spa_name(spa),
2723		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2724
2725		/*
2726		 * Clean up any stale temporary dataset userrefs.
2727		 */
2728		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2729	}
2730
2731	return (0);
2732}
2733
2734static int
2735spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2736{
2737	int mode = spa->spa_mode;
2738
2739	spa_unload(spa);
2740	spa_deactivate(spa);
2741
2742	spa->spa_load_max_txg--;
2743
2744	spa_activate(spa, mode);
2745	spa_async_suspend(spa);
2746
2747	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2748}
2749
2750/*
2751 * If spa_load() fails this function will try loading prior txg's. If
2752 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2753 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2754 * function will not rewind the pool and will return the same error as
2755 * spa_load().
2756 */
2757static int
2758spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2759    uint64_t max_request, int rewind_flags)
2760{
2761	nvlist_t *loadinfo = NULL;
2762	nvlist_t *config = NULL;
2763	int load_error, rewind_error;
2764	uint64_t safe_rewind_txg;
2765	uint64_t min_txg;
2766
2767	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2768		spa->spa_load_max_txg = spa->spa_load_txg;
2769		spa_set_log_state(spa, SPA_LOG_CLEAR);
2770	} else {
2771		spa->spa_load_max_txg = max_request;
2772	}
2773
2774	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2775	    mosconfig);
2776	if (load_error == 0)
2777		return (0);
2778
2779	if (spa->spa_root_vdev != NULL)
2780		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2781
2782	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2783	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2784
2785	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2786		nvlist_free(config);
2787		return (load_error);
2788	}
2789
2790	if (state == SPA_LOAD_RECOVER) {
2791		/* Price of rolling back is discarding txgs, including log */
2792		spa_set_log_state(spa, SPA_LOG_CLEAR);
2793	} else {
2794		/*
2795		 * If we aren't rolling back save the load info from our first
2796		 * import attempt so that we can restore it after attempting
2797		 * to rewind.
2798		 */
2799		loadinfo = spa->spa_load_info;
2800		spa->spa_load_info = fnvlist_alloc();
2801	}
2802
2803	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2804	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2805	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2806	    TXG_INITIAL : safe_rewind_txg;
2807
2808	/*
2809	 * Continue as long as we're finding errors, we're still within
2810	 * the acceptable rewind range, and we're still finding uberblocks
2811	 */
2812	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2813	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2814		if (spa->spa_load_max_txg < safe_rewind_txg)
2815			spa->spa_extreme_rewind = B_TRUE;
2816		rewind_error = spa_load_retry(spa, state, mosconfig);
2817	}
2818
2819	spa->spa_extreme_rewind = B_FALSE;
2820	spa->spa_load_max_txg = UINT64_MAX;
2821
2822	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2823		spa_config_set(spa, config);
2824
2825	if (state == SPA_LOAD_RECOVER) {
2826		ASSERT3P(loadinfo, ==, NULL);
2827		return (rewind_error);
2828	} else {
2829		/* Store the rewind info as part of the initial load info */
2830		fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2831		    spa->spa_load_info);
2832
2833		/* Restore the initial load info */
2834		fnvlist_free(spa->spa_load_info);
2835		spa->spa_load_info = loadinfo;
2836
2837		return (load_error);
2838	}
2839}
2840
2841/*
2842 * Pool Open/Import
2843 *
2844 * The import case is identical to an open except that the configuration is sent
2845 * down from userland, instead of grabbed from the configuration cache.  For the
2846 * case of an open, the pool configuration will exist in the
2847 * POOL_STATE_UNINITIALIZED state.
2848 *
2849 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2850 * the same time open the pool, without having to keep around the spa_t in some
2851 * ambiguous state.
2852 */
2853static int
2854spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2855    nvlist_t **config)
2856{
2857	spa_t *spa;
2858	spa_load_state_t state = SPA_LOAD_OPEN;
2859	int error;
2860	int locked = B_FALSE;
2861
2862	*spapp = NULL;
2863
2864	/*
2865	 * As disgusting as this is, we need to support recursive calls to this
2866	 * function because dsl_dir_open() is called during spa_load(), and ends
2867	 * up calling spa_open() again.  The real fix is to figure out how to
2868	 * avoid dsl_dir_open() calling this in the first place.
2869	 */
2870	if (mutex_owner(&spa_namespace_lock) != curthread) {
2871		mutex_enter(&spa_namespace_lock);
2872		locked = B_TRUE;
2873	}
2874
2875	if ((spa = spa_lookup(pool)) == NULL) {
2876		if (locked)
2877			mutex_exit(&spa_namespace_lock);
2878		return (SET_ERROR(ENOENT));
2879	}
2880
2881	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2882		zpool_rewind_policy_t policy;
2883
2884		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2885		    &policy);
2886		if (policy.zrp_request & ZPOOL_DO_REWIND)
2887			state = SPA_LOAD_RECOVER;
2888
2889		spa_activate(spa, spa_mode_global);
2890
2891		if (state != SPA_LOAD_RECOVER)
2892			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2893
2894		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2895		    policy.zrp_request);
2896
2897		if (error == EBADF) {
2898			/*
2899			 * If vdev_validate() returns failure (indicated by
2900			 * EBADF), it indicates that one of the vdevs indicates
2901			 * that the pool has been exported or destroyed.  If
2902			 * this is the case, the config cache is out of sync and
2903			 * we should remove the pool from the namespace.
2904			 */
2905			spa_unload(spa);
2906			spa_deactivate(spa);
2907			spa_config_sync(spa, B_TRUE, B_TRUE);
2908			spa_remove(spa);
2909			if (locked)
2910				mutex_exit(&spa_namespace_lock);
2911			return (SET_ERROR(ENOENT));
2912		}
2913
2914		if (error) {
2915			/*
2916			 * We can't open the pool, but we still have useful
2917			 * information: the state of each vdev after the
2918			 * attempted vdev_open().  Return this to the user.
2919			 */
2920			if (config != NULL && spa->spa_config) {
2921				VERIFY(nvlist_dup(spa->spa_config, config,
2922				    KM_SLEEP) == 0);
2923				VERIFY(nvlist_add_nvlist(*config,
2924				    ZPOOL_CONFIG_LOAD_INFO,
2925				    spa->spa_load_info) == 0);
2926			}
2927			spa_unload(spa);
2928			spa_deactivate(spa);
2929			spa->spa_last_open_failed = error;
2930			if (locked)
2931				mutex_exit(&spa_namespace_lock);
2932			*spapp = NULL;
2933			return (error);
2934		}
2935	}
2936
2937	spa_open_ref(spa, tag);
2938
2939	if (config != NULL)
2940		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2941
2942	/*
2943	 * If we've recovered the pool, pass back any information we
2944	 * gathered while doing the load.
2945	 */
2946	if (state == SPA_LOAD_RECOVER) {
2947		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2948		    spa->spa_load_info) == 0);
2949	}
2950
2951	if (locked) {
2952		spa->spa_last_open_failed = 0;
2953		spa->spa_last_ubsync_txg = 0;
2954		spa->spa_load_txg = 0;
2955		mutex_exit(&spa_namespace_lock);
2956	}
2957
2958	*spapp = spa;
2959
2960	return (0);
2961}
2962
2963int
2964spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2965    nvlist_t **config)
2966{
2967	return (spa_open_common(name, spapp, tag, policy, config));
2968}
2969
2970int
2971spa_open(const char *name, spa_t **spapp, void *tag)
2972{
2973	return (spa_open_common(name, spapp, tag, NULL, NULL));
2974}
2975
2976/*
2977 * Lookup the given spa_t, incrementing the inject count in the process,
2978 * preventing it from being exported or destroyed.
2979 */
2980spa_t *
2981spa_inject_addref(char *name)
2982{
2983	spa_t *spa;
2984
2985	mutex_enter(&spa_namespace_lock);
2986	if ((spa = spa_lookup(name)) == NULL) {
2987		mutex_exit(&spa_namespace_lock);
2988		return (NULL);
2989	}
2990	spa->spa_inject_ref++;
2991	mutex_exit(&spa_namespace_lock);
2992
2993	return (spa);
2994}
2995
2996void
2997spa_inject_delref(spa_t *spa)
2998{
2999	mutex_enter(&spa_namespace_lock);
3000	spa->spa_inject_ref--;
3001	mutex_exit(&spa_namespace_lock);
3002}
3003
3004/*
3005 * Add spares device information to the nvlist.
3006 */
3007static void
3008spa_add_spares(spa_t *spa, nvlist_t *config)
3009{
3010	nvlist_t **spares;
3011	uint_t i, nspares;
3012	nvlist_t *nvroot;
3013	uint64_t guid;
3014	vdev_stat_t *vs;
3015	uint_t vsc;
3016	uint64_t pool;
3017
3018	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3019
3020	if (spa->spa_spares.sav_count == 0)
3021		return;
3022
3023	VERIFY(nvlist_lookup_nvlist(config,
3024	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3025	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3026	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3027	if (nspares != 0) {
3028		VERIFY(nvlist_add_nvlist_array(nvroot,
3029		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3030		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3031		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3032
3033		/*
3034		 * Go through and find any spares which have since been
3035		 * repurposed as an active spare.  If this is the case, update
3036		 * their status appropriately.
3037		 */
3038		for (i = 0; i < nspares; i++) {
3039			VERIFY(nvlist_lookup_uint64(spares[i],
3040			    ZPOOL_CONFIG_GUID, &guid) == 0);
3041			if (spa_spare_exists(guid, &pool, NULL) &&
3042			    pool != 0ULL) {
3043				VERIFY(nvlist_lookup_uint64_array(
3044				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
3045				    (uint64_t **)&vs, &vsc) == 0);
3046				vs->vs_state = VDEV_STATE_CANT_OPEN;
3047				vs->vs_aux = VDEV_AUX_SPARED;
3048			}
3049		}
3050	}
3051}
3052
3053/*
3054 * Add l2cache device information to the nvlist, including vdev stats.
3055 */
3056static void
3057spa_add_l2cache(spa_t *spa, nvlist_t *config)
3058{
3059	nvlist_t **l2cache;
3060	uint_t i, j, nl2cache;
3061	nvlist_t *nvroot;
3062	uint64_t guid;
3063	vdev_t *vd;
3064	vdev_stat_t *vs;
3065	uint_t vsc;
3066
3067	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3068
3069	if (spa->spa_l2cache.sav_count == 0)
3070		return;
3071
3072	VERIFY(nvlist_lookup_nvlist(config,
3073	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3074	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3075	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3076	if (nl2cache != 0) {
3077		VERIFY(nvlist_add_nvlist_array(nvroot,
3078		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3079		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3080		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3081
3082		/*
3083		 * Update level 2 cache device stats.
3084		 */
3085
3086		for (i = 0; i < nl2cache; i++) {
3087			VERIFY(nvlist_lookup_uint64(l2cache[i],
3088			    ZPOOL_CONFIG_GUID, &guid) == 0);
3089
3090			vd = NULL;
3091			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3092				if (guid ==
3093				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3094					vd = spa->spa_l2cache.sav_vdevs[j];
3095					break;
3096				}
3097			}
3098			ASSERT(vd != NULL);
3099
3100			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3101			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3102			    == 0);
3103			vdev_get_stats(vd, vs);
3104		}
3105	}
3106}
3107
3108static void
3109spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3110{
3111	nvlist_t *features;
3112	zap_cursor_t zc;
3113	zap_attribute_t za;
3114
3115	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3116	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3117
3118	if (spa->spa_feat_for_read_obj != 0) {
3119		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3120		    spa->spa_feat_for_read_obj);
3121		    zap_cursor_retrieve(&zc, &za) == 0;
3122		    zap_cursor_advance(&zc)) {
3123			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3124			    za.za_num_integers == 1);
3125			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3126			    za.za_first_integer));
3127		}
3128		zap_cursor_fini(&zc);
3129	}
3130
3131	if (spa->spa_feat_for_write_obj != 0) {
3132		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3133		    spa->spa_feat_for_write_obj);
3134		    zap_cursor_retrieve(&zc, &za) == 0;
3135		    zap_cursor_advance(&zc)) {
3136			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3137			    za.za_num_integers == 1);
3138			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3139			    za.za_first_integer));
3140		}
3141		zap_cursor_fini(&zc);
3142	}
3143
3144	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3145	    features) == 0);
3146	nvlist_free(features);
3147}
3148
3149int
3150spa_get_stats(const char *name, nvlist_t **config,
3151    char *altroot, size_t buflen)
3152{
3153	int error;
3154	spa_t *spa;
3155
3156	*config = NULL;
3157	error = spa_open_common(name, &spa, FTAG, NULL, config);
3158
3159	if (spa != NULL) {
3160		/*
3161		 * This still leaves a window of inconsistency where the spares
3162		 * or l2cache devices could change and the config would be
3163		 * self-inconsistent.
3164		 */
3165		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3166
3167		if (*config != NULL) {
3168			uint64_t loadtimes[2];
3169
3170			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3171			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3172			VERIFY(nvlist_add_uint64_array(*config,
3173			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3174
3175			VERIFY(nvlist_add_uint64(*config,
3176			    ZPOOL_CONFIG_ERRCOUNT,
3177			    spa_get_errlog_size(spa)) == 0);
3178
3179			if (spa_suspended(spa))
3180				VERIFY(nvlist_add_uint64(*config,
3181				    ZPOOL_CONFIG_SUSPENDED,
3182				    spa->spa_failmode) == 0);
3183
3184			spa_add_spares(spa, *config);
3185			spa_add_l2cache(spa, *config);
3186			spa_add_feature_stats(spa, *config);
3187		}
3188	}
3189
3190	/*
3191	 * We want to get the alternate root even for faulted pools, so we cheat
3192	 * and call spa_lookup() directly.
3193	 */
3194	if (altroot) {
3195		if (spa == NULL) {
3196			mutex_enter(&spa_namespace_lock);
3197			spa = spa_lookup(name);
3198			if (spa)
3199				spa_altroot(spa, altroot, buflen);
3200			else
3201				altroot[0] = '\0';
3202			spa = NULL;
3203			mutex_exit(&spa_namespace_lock);
3204		} else {
3205			spa_altroot(spa, altroot, buflen);
3206		}
3207	}
3208
3209	if (spa != NULL) {
3210		spa_config_exit(spa, SCL_CONFIG, FTAG);
3211		spa_close(spa, FTAG);
3212	}
3213
3214	return (error);
3215}
3216
3217/*
3218 * Validate that the auxiliary device array is well formed.  We must have an
3219 * array of nvlists, each which describes a valid leaf vdev.  If this is an
3220 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3221 * specified, as long as they are well-formed.
3222 */
3223static int
3224spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3225    spa_aux_vdev_t *sav, const char *config, uint64_t version,
3226    vdev_labeltype_t label)
3227{
3228	nvlist_t **dev;
3229	uint_t i, ndev;
3230	vdev_t *vd;
3231	int error;
3232
3233	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3234
3235	/*
3236	 * It's acceptable to have no devs specified.
3237	 */
3238	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3239		return (0);
3240
3241	if (ndev == 0)
3242		return (SET_ERROR(EINVAL));
3243
3244	/*
3245	 * Make sure the pool is formatted with a version that supports this
3246	 * device type.
3247	 */
3248	if (spa_version(spa) < version)
3249		return (SET_ERROR(ENOTSUP));
3250
3251	/*
3252	 * Set the pending device list so we correctly handle device in-use
3253	 * checking.
3254	 */
3255	sav->sav_pending = dev;
3256	sav->sav_npending = ndev;
3257
3258	for (i = 0; i < ndev; i++) {
3259		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3260		    mode)) != 0)
3261			goto out;
3262
3263		if (!vd->vdev_ops->vdev_op_leaf) {
3264			vdev_free(vd);
3265			error = SET_ERROR(EINVAL);
3266			goto out;
3267		}
3268
3269		/*
3270		 * The L2ARC currently only supports disk devices in
3271		 * kernel context.  For user-level testing, we allow it.
3272		 */
3273#ifdef _KERNEL
3274		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3275		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3276			error = SET_ERROR(ENOTBLK);
3277			vdev_free(vd);
3278			goto out;
3279		}
3280#endif
3281		vd->vdev_top = vd;
3282
3283		if ((error = vdev_open(vd)) == 0 &&
3284		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
3285			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3286			    vd->vdev_guid) == 0);
3287		}
3288
3289		vdev_free(vd);
3290
3291		if (error &&
3292		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3293			goto out;
3294		else
3295			error = 0;
3296	}
3297
3298out:
3299	sav->sav_pending = NULL;
3300	sav->sav_npending = 0;
3301	return (error);
3302}
3303
3304static int
3305spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3306{
3307	int error;
3308
3309	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3310
3311	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3312	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3313	    VDEV_LABEL_SPARE)) != 0) {
3314		return (error);
3315	}
3316
3317	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3318	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3319	    VDEV_LABEL_L2CACHE));
3320}
3321
3322static void
3323spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3324    const char *config)
3325{
3326	int i;
3327
3328	if (sav->sav_config != NULL) {
3329		nvlist_t **olddevs;
3330		uint_t oldndevs;
3331		nvlist_t **newdevs;
3332
3333		/*
3334		 * Generate new dev list by concatentating with the
3335		 * current dev list.
3336		 */
3337		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3338		    &olddevs, &oldndevs) == 0);
3339
3340		newdevs = kmem_alloc(sizeof (void *) *
3341		    (ndevs + oldndevs), KM_SLEEP);
3342		for (i = 0; i < oldndevs; i++)
3343			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3344			    KM_SLEEP) == 0);
3345		for (i = 0; i < ndevs; i++)
3346			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3347			    KM_SLEEP) == 0);
3348
3349		VERIFY(nvlist_remove(sav->sav_config, config,
3350		    DATA_TYPE_NVLIST_ARRAY) == 0);
3351
3352		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3353		    config, newdevs, ndevs + oldndevs) == 0);
3354		for (i = 0; i < oldndevs + ndevs; i++)
3355			nvlist_free(newdevs[i]);
3356		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3357	} else {
3358		/*
3359		 * Generate a new dev list.
3360		 */
3361		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3362		    KM_SLEEP) == 0);
3363		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3364		    devs, ndevs) == 0);
3365	}
3366}
3367
3368/*
3369 * Stop and drop level 2 ARC devices
3370 */
3371void
3372spa_l2cache_drop(spa_t *spa)
3373{
3374	vdev_t *vd;
3375	int i;
3376	spa_aux_vdev_t *sav = &spa->spa_l2cache;
3377
3378	for (i = 0; i < sav->sav_count; i++) {
3379		uint64_t pool;
3380
3381		vd = sav->sav_vdevs[i];
3382		ASSERT(vd != NULL);
3383
3384		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3385		    pool != 0ULL && l2arc_vdev_present(vd))
3386			l2arc_remove_vdev(vd);
3387	}
3388}
3389
3390/*
3391 * Pool Creation
3392 */
3393int
3394spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3395    nvlist_t *zplprops)
3396{
3397	spa_t *spa;
3398	char *altroot = NULL;
3399	vdev_t *rvd;
3400	dsl_pool_t *dp;
3401	dmu_tx_t *tx;
3402	int error = 0;
3403	uint64_t txg = TXG_INITIAL;
3404	nvlist_t **spares, **l2cache;
3405	uint_t nspares, nl2cache;
3406	uint64_t version, obj;
3407	boolean_t has_features;
3408
3409	/*
3410	 * If this pool already exists, return failure.
3411	 */
3412	mutex_enter(&spa_namespace_lock);
3413	if (spa_lookup(pool) != NULL) {
3414		mutex_exit(&spa_namespace_lock);
3415		return (SET_ERROR(EEXIST));
3416	}
3417
3418	/*
3419	 * Allocate a new spa_t structure.
3420	 */
3421	(void) nvlist_lookup_string(props,
3422	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3423	spa = spa_add(pool, NULL, altroot);
3424	spa_activate(spa, spa_mode_global);
3425
3426	if (props && (error = spa_prop_validate(spa, props))) {
3427		spa_deactivate(spa);
3428		spa_remove(spa);
3429		mutex_exit(&spa_namespace_lock);
3430		return (error);
3431	}
3432
3433	has_features = B_FALSE;
3434	for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3435	    elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3436		if (zpool_prop_feature(nvpair_name(elem)))
3437			has_features = B_TRUE;
3438	}
3439
3440	if (has_features || nvlist_lookup_uint64(props,
3441	    zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3442		version = SPA_VERSION;
3443	}
3444	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3445
3446	spa->spa_first_txg = txg;
3447	spa->spa_uberblock.ub_txg = txg - 1;
3448	spa->spa_uberblock.ub_version = version;
3449	spa->spa_ubsync = spa->spa_uberblock;
3450
3451	/*
3452	 * Create "The Godfather" zio to hold all async IOs
3453	 */
3454	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3455	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3456
3457	/*
3458	 * Create the root vdev.
3459	 */
3460	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3461
3462	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3463
3464	ASSERT(error != 0 || rvd != NULL);
3465	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3466
3467	if (error == 0 && !zfs_allocatable_devs(nvroot))
3468		error = SET_ERROR(EINVAL);
3469
3470	if (error == 0 &&
3471	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3472	    (error = spa_validate_aux(spa, nvroot, txg,
3473	    VDEV_ALLOC_ADD)) == 0) {
3474		for (int c = 0; c < rvd->vdev_children; c++) {
3475			vdev_metaslab_set_size(rvd->vdev_child[c]);
3476			vdev_expand(rvd->vdev_child[c], txg);
3477		}
3478	}
3479
3480	spa_config_exit(spa, SCL_ALL, FTAG);
3481
3482	if (error != 0) {
3483		spa_unload(spa);
3484		spa_deactivate(spa);
3485		spa_remove(spa);
3486		mutex_exit(&spa_namespace_lock);
3487		return (error);
3488	}
3489
3490	/*
3491	 * Get the list of spares, if specified.
3492	 */
3493	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3494	    &spares, &nspares) == 0) {
3495		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3496		    KM_SLEEP) == 0);
3497		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3498		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3499		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3500		spa_load_spares(spa);
3501		spa_config_exit(spa, SCL_ALL, FTAG);
3502		spa->spa_spares.sav_sync = B_TRUE;
3503	}
3504
3505	/*
3506	 * Get the list of level 2 cache devices, if specified.
3507	 */
3508	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3509	    &l2cache, &nl2cache) == 0) {
3510		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3511		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3512		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3513		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3514		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3515		spa_load_l2cache(spa);
3516		spa_config_exit(spa, SCL_ALL, FTAG);
3517		spa->spa_l2cache.sav_sync = B_TRUE;
3518	}
3519
3520	spa->spa_is_initializing = B_TRUE;
3521	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3522	spa->spa_meta_objset = dp->dp_meta_objset;
3523	spa->spa_is_initializing = B_FALSE;
3524
3525	/*
3526	 * Create DDTs (dedup tables).
3527	 */
3528	ddt_create(spa);
3529
3530	spa_update_dspace(spa);
3531
3532	tx = dmu_tx_create_assigned(dp, txg);
3533
3534	/*
3535	 * Create the pool config object.
3536	 */
3537	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3538	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3539	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3540
3541	if (zap_add(spa->spa_meta_objset,
3542	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3543	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3544		cmn_err(CE_PANIC, "failed to add pool config");
3545	}
3546
3547	if (spa_version(spa) >= SPA_VERSION_FEATURES)
3548		spa_feature_create_zap_objects(spa, tx);
3549
3550	if (zap_add(spa->spa_meta_objset,
3551	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3552	    sizeof (uint64_t), 1, &version, tx) != 0) {
3553		cmn_err(CE_PANIC, "failed to add pool version");
3554	}
3555
3556	/* Newly created pools with the right version are always deflated. */
3557	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3558		spa->spa_deflate = TRUE;
3559		if (zap_add(spa->spa_meta_objset,
3560		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3561		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3562			cmn_err(CE_PANIC, "failed to add deflate");
3563		}
3564	}
3565
3566	/*
3567	 * Create the deferred-free bpobj.  Turn off compression
3568	 * because sync-to-convergence takes longer if the blocksize
3569	 * keeps changing.
3570	 */
3571	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3572	dmu_object_set_compress(spa->spa_meta_objset, obj,
3573	    ZIO_COMPRESS_OFF, tx);
3574	if (zap_add(spa->spa_meta_objset,
3575	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3576	    sizeof (uint64_t), 1, &obj, tx) != 0) {
3577		cmn_err(CE_PANIC, "failed to add bpobj");
3578	}
3579	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3580	    spa->spa_meta_objset, obj));
3581
3582	/*
3583	 * Create the pool's history object.
3584	 */
3585	if (version >= SPA_VERSION_ZPOOL_HISTORY)
3586		spa_history_create_obj(spa, tx);
3587
3588	/*
3589	 * Set pool properties.
3590	 */
3591	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3592	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3593	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3594	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3595
3596	if (props != NULL) {
3597		spa_configfile_set(spa, props, B_FALSE);
3598		spa_sync_props(props, tx);
3599	}
3600
3601	dmu_tx_commit(tx);
3602
3603	spa->spa_sync_on = B_TRUE;
3604	txg_sync_start(spa->spa_dsl_pool);
3605
3606	/*
3607	 * We explicitly wait for the first transaction to complete so that our
3608	 * bean counters are appropriately updated.
3609	 */
3610	txg_wait_synced(spa->spa_dsl_pool, txg);
3611
3612	spa_config_sync(spa, B_FALSE, B_TRUE);
3613
3614	spa_history_log_version(spa, "create");
3615
3616	spa->spa_minref = refcount_count(&spa->spa_refcount);
3617
3618	mutex_exit(&spa_namespace_lock);
3619
3620	return (0);
3621}
3622
3623#ifdef _KERNEL
3624/*
3625 * Get the root pool information from the root disk, then import the root pool
3626 * during the system boot up time.
3627 */
3628extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3629
3630static nvlist_t *
3631spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3632{
3633	nvlist_t *config;
3634	nvlist_t *nvtop, *nvroot;
3635	uint64_t pgid;
3636
3637	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3638		return (NULL);
3639
3640	/*
3641	 * Add this top-level vdev to the child array.
3642	 */
3643	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3644	    &nvtop) == 0);
3645	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3646	    &pgid) == 0);
3647	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3648
3649	/*
3650	 * Put this pool's top-level vdevs into a root vdev.
3651	 */
3652	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3653	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3654	    VDEV_TYPE_ROOT) == 0);
3655	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3656	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3657	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3658	    &nvtop, 1) == 0);
3659
3660	/*
3661	 * Replace the existing vdev_tree with the new root vdev in
3662	 * this pool's configuration (remove the old, add the new).
3663	 */
3664	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3665	nvlist_free(nvroot);
3666	return (config);
3667}
3668
3669/*
3670 * Walk the vdev tree and see if we can find a device with "better"
3671 * configuration. A configuration is "better" if the label on that
3672 * device has a more recent txg.
3673 */
3674static void
3675spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3676{
3677	for (int c = 0; c < vd->vdev_children; c++)
3678		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3679
3680	if (vd->vdev_ops->vdev_op_leaf) {
3681		nvlist_t *label;
3682		uint64_t label_txg;
3683
3684		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3685		    &label) != 0)
3686			return;
3687
3688		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3689		    &label_txg) == 0);
3690
3691		/*
3692		 * Do we have a better boot device?
3693		 */
3694		if (label_txg > *txg) {
3695			*txg = label_txg;
3696			*avd = vd;
3697		}
3698		nvlist_free(label);
3699	}
3700}
3701
3702/*
3703 * Import a root pool.
3704 *
3705 * For x86. devpath_list will consist of devid and/or physpath name of
3706 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3707 * The GRUB "findroot" command will return the vdev we should boot.
3708 *
3709 * For Sparc, devpath_list consists the physpath name of the booting device
3710 * no matter the rootpool is a single device pool or a mirrored pool.
3711 * e.g.
3712 *	"/pci@1f,0/ide@d/disk@0,0:a"
3713 */
3714int
3715spa_import_rootpool(char *devpath, char *devid)
3716{
3717	spa_t *spa;
3718	vdev_t *rvd, *bvd, *avd = NULL;
3719	nvlist_t *config, *nvtop;
3720	uint64_t guid, txg;
3721	char *pname;
3722	int error;
3723
3724	/*
3725	 * Read the label from the boot device and generate a configuration.
3726	 */
3727	config = spa_generate_rootconf(devpath, devid, &guid);
3728#if defined(_OBP) && defined(_KERNEL)
3729	if (config == NULL) {
3730		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3731			/* iscsi boot */
3732			get_iscsi_bootpath_phy(devpath);
3733			config = spa_generate_rootconf(devpath, devid, &guid);
3734		}
3735	}
3736#endif
3737	if (config == NULL) {
3738		cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3739		    devpath);
3740		return (SET_ERROR(EIO));
3741	}
3742
3743	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3744	    &pname) == 0);
3745	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3746
3747	mutex_enter(&spa_namespace_lock);
3748	if ((spa = spa_lookup(pname)) != NULL) {
3749		/*
3750		 * Remove the existing root pool from the namespace so that we
3751		 * can replace it with the correct config we just read in.
3752		 */
3753		spa_remove(spa);
3754	}
3755
3756	spa = spa_add(pname, config, NULL);
3757	spa->spa_is_root = B_TRUE;
3758	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3759
3760	/*
3761	 * Build up a vdev tree based on the boot device's label config.
3762	 */
3763	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3764	    &nvtop) == 0);
3765	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3766	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3767	    VDEV_ALLOC_ROOTPOOL);
3768	spa_config_exit(spa, SCL_ALL, FTAG);
3769	if (error) {
3770		mutex_exit(&spa_namespace_lock);
3771		nvlist_free(config);
3772		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3773		    pname);
3774		return (error);
3775	}
3776
3777	/*
3778	 * Get the boot vdev.
3779	 */
3780	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3781		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3782		    (u_longlong_t)guid);
3783		error = SET_ERROR(ENOENT);
3784		goto out;
3785	}
3786
3787	/*
3788	 * Determine if there is a better boot device.
3789	 */
3790	avd = bvd;
3791	spa_alt_rootvdev(rvd, &avd, &txg);
3792	if (avd != bvd) {
3793		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3794		    "try booting from '%s'", avd->vdev_path);
3795		error = SET_ERROR(EINVAL);
3796		goto out;
3797	}
3798
3799	/*
3800	 * If the boot device is part of a spare vdev then ensure that
3801	 * we're booting off the active spare.
3802	 */
3803	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3804	    !bvd->vdev_isspare) {
3805		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3806		    "try booting from '%s'",
3807		    bvd->vdev_parent->
3808		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3809		error = SET_ERROR(EINVAL);
3810		goto out;
3811	}
3812
3813	error = 0;
3814out:
3815	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3816	vdev_free(rvd);
3817	spa_config_exit(spa, SCL_ALL, FTAG);
3818	mutex_exit(&spa_namespace_lock);
3819
3820	nvlist_free(config);
3821	return (error);
3822}
3823
3824#endif
3825
3826/*
3827 * Import a non-root pool into the system.
3828 */
3829int
3830spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3831{
3832	spa_t *spa;
3833	char *altroot = NULL;
3834	spa_load_state_t state = SPA_LOAD_IMPORT;
3835	zpool_rewind_policy_t policy;
3836	uint64_t mode = spa_mode_global;
3837	uint64_t readonly = B_FALSE;
3838	int error;
3839	nvlist_t *nvroot;
3840	nvlist_t **spares, **l2cache;
3841	uint_t nspares, nl2cache;
3842
3843	/*
3844	 * If a pool with this name exists, return failure.
3845	 */
3846	mutex_enter(&spa_namespace_lock);
3847	if (spa_lookup(pool) != NULL) {
3848		mutex_exit(&spa_namespace_lock);
3849		return (SET_ERROR(EEXIST));
3850	}
3851
3852	/*
3853	 * Create and initialize the spa structure.
3854	 */
3855	(void) nvlist_lookup_string(props,
3856	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3857	(void) nvlist_lookup_uint64(props,
3858	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3859	if (readonly)
3860		mode = FREAD;
3861	spa = spa_add(pool, config, altroot);
3862	spa->spa_import_flags = flags;
3863
3864	/*
3865	 * Verbatim import - Take a pool and insert it into the namespace
3866	 * as if it had been loaded at boot.
3867	 */
3868	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3869		if (props != NULL)
3870			spa_configfile_set(spa, props, B_FALSE);
3871
3872		spa_config_sync(spa, B_FALSE, B_TRUE);
3873
3874		mutex_exit(&spa_namespace_lock);
3875		return (0);
3876	}
3877
3878	spa_activate(spa, mode);
3879
3880	/*
3881	 * Don't start async tasks until we know everything is healthy.
3882	 */
3883	spa_async_suspend(spa);
3884
3885	zpool_get_rewind_policy(config, &policy);
3886	if (policy.zrp_request & ZPOOL_DO_REWIND)
3887		state = SPA_LOAD_RECOVER;
3888
3889	/*
3890	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3891	 * because the user-supplied config is actually the one to trust when
3892	 * doing an import.
3893	 */
3894	if (state != SPA_LOAD_RECOVER)
3895		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3896
3897	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3898	    policy.zrp_request);
3899
3900	/*
3901	 * Propagate anything learned while loading the pool and pass it
3902	 * back to caller (i.e. rewind info, missing devices, etc).
3903	 */
3904	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3905	    spa->spa_load_info) == 0);
3906
3907	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3908	/*
3909	 * Toss any existing sparelist, as it doesn't have any validity
3910	 * anymore, and conflicts with spa_has_spare().
3911	 */
3912	if (spa->spa_spares.sav_config) {
3913		nvlist_free(spa->spa_spares.sav_config);
3914		spa->spa_spares.sav_config = NULL;
3915		spa_load_spares(spa);
3916	}
3917	if (spa->spa_l2cache.sav_config) {
3918		nvlist_free(spa->spa_l2cache.sav_config);
3919		spa->spa_l2cache.sav_config = NULL;
3920		spa_load_l2cache(spa);
3921	}
3922
3923	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3924	    &nvroot) == 0);
3925	if (error == 0)
3926		error = spa_validate_aux(spa, nvroot, -1ULL,
3927		    VDEV_ALLOC_SPARE);
3928	if (error == 0)
3929		error = spa_validate_aux(spa, nvroot, -1ULL,
3930		    VDEV_ALLOC_L2CACHE);
3931	spa_config_exit(spa, SCL_ALL, FTAG);
3932
3933	if (props != NULL)
3934		spa_configfile_set(spa, props, B_FALSE);
3935
3936	if (error != 0 || (props && spa_writeable(spa) &&
3937	    (error = spa_prop_set(spa, props)))) {
3938		spa_unload(spa);
3939		spa_deactivate(spa);
3940		spa_remove(spa);
3941		mutex_exit(&spa_namespace_lock);
3942		return (error);
3943	}
3944
3945	spa_async_resume(spa);
3946
3947	/*
3948	 * Override any spares and level 2 cache devices as specified by
3949	 * the user, as these may have correct device names/devids, etc.
3950	 */
3951	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3952	    &spares, &nspares) == 0) {
3953		if (spa->spa_spares.sav_config)
3954			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3955			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3956		else
3957			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3958			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3959		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3960		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3961		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3962		spa_load_spares(spa);
3963		spa_config_exit(spa, SCL_ALL, FTAG);
3964		spa->spa_spares.sav_sync = B_TRUE;
3965	}
3966	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3967	    &l2cache, &nl2cache) == 0) {
3968		if (spa->spa_l2cache.sav_config)
3969			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3970			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3971		else
3972			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3973			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3974		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3975		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3976		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3977		spa_load_l2cache(spa);
3978		spa_config_exit(spa, SCL_ALL, FTAG);
3979		spa->spa_l2cache.sav_sync = B_TRUE;
3980	}
3981
3982	/*
3983	 * Check for any removed devices.
3984	 */
3985	if (spa->spa_autoreplace) {
3986		spa_aux_check_removed(&spa->spa_spares);
3987		spa_aux_check_removed(&spa->spa_l2cache);
3988	}
3989
3990	if (spa_writeable(spa)) {
3991		/*
3992		 * Update the config cache to include the newly-imported pool.
3993		 */
3994		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3995	}
3996
3997	/*
3998	 * It's possible that the pool was expanded while it was exported.
3999	 * We kick off an async task to handle this for us.
4000	 */
4001	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4002
4003	mutex_exit(&spa_namespace_lock);
4004	spa_history_log_version(spa, "import");
4005
4006	return (0);
4007}
4008
4009nvlist_t *
4010spa_tryimport(nvlist_t *tryconfig)
4011{
4012	nvlist_t *config = NULL;
4013	char *poolname;
4014	spa_t *spa;
4015	uint64_t state;
4016	int error;
4017
4018	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4019		return (NULL);
4020
4021	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4022		return (NULL);
4023
4024	/*
4025	 * Create and initialize the spa structure.
4026	 */
4027	mutex_enter(&spa_namespace_lock);
4028	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4029	spa_activate(spa, FREAD);
4030
4031	/*
4032	 * Pass off the heavy lifting to spa_load().
4033	 * Pass TRUE for mosconfig because the user-supplied config
4034	 * is actually the one to trust when doing an import.
4035	 */
4036	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4037
4038	/*
4039	 * If 'tryconfig' was at least parsable, return the current config.
4040	 */
4041	if (spa->spa_root_vdev != NULL) {
4042		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4043		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4044		    poolname) == 0);
4045		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4046		    state) == 0);
4047		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4048		    spa->spa_uberblock.ub_timestamp) == 0);
4049		VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4050		    spa->spa_load_info) == 0);
4051
4052		/*
4053		 * If the bootfs property exists on this pool then we
4054		 * copy it out so that external consumers can tell which
4055		 * pools are bootable.
4056		 */
4057		if ((!error || error == EEXIST) && spa->spa_bootfs) {
4058			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4059
4060			/*
4061			 * We have to play games with the name since the
4062			 * pool was opened as TRYIMPORT_NAME.
4063			 */
4064			if (dsl_dsobj_to_dsname(spa_name(spa),
4065			    spa->spa_bootfs, tmpname) == 0) {
4066				char *cp;
4067				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4068
4069				cp = strchr(tmpname, '/');
4070				if (cp == NULL) {
4071					(void) strlcpy(dsname, tmpname,
4072					    MAXPATHLEN);
4073				} else {
4074					(void) snprintf(dsname, MAXPATHLEN,
4075					    "%s/%s", poolname, ++cp);
4076				}
4077				VERIFY(nvlist_add_string(config,
4078				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4079				kmem_free(dsname, MAXPATHLEN);
4080			}
4081			kmem_free(tmpname, MAXPATHLEN);
4082		}
4083
4084		/*
4085		 * Add the list of hot spares and level 2 cache devices.
4086		 */
4087		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4088		spa_add_spares(spa, config);
4089		spa_add_l2cache(spa, config);
4090		spa_config_exit(spa, SCL_CONFIG, FTAG);
4091	}
4092
4093	spa_unload(spa);
4094	spa_deactivate(spa);
4095	spa_remove(spa);
4096	mutex_exit(&spa_namespace_lock);
4097
4098	return (config);
4099}
4100
4101/*
4102 * Pool export/destroy
4103 *
4104 * The act of destroying or exporting a pool is very simple.  We make sure there
4105 * is no more pending I/O and any references to the pool are gone.  Then, we
4106 * update the pool state and sync all the labels to disk, removing the
4107 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4108 * we don't sync the labels or remove the configuration cache.
4109 */
4110static int
4111spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4112    boolean_t force, boolean_t hardforce)
4113{
4114	spa_t *spa;
4115
4116	if (oldconfig)
4117		*oldconfig = NULL;
4118
4119	if (!(spa_mode_global & FWRITE))
4120		return (SET_ERROR(EROFS));
4121
4122	mutex_enter(&spa_namespace_lock);
4123	if ((spa = spa_lookup(pool)) == NULL) {
4124		mutex_exit(&spa_namespace_lock);
4125		return (SET_ERROR(ENOENT));
4126	}
4127
4128	/*
4129	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4130	 * reacquire the namespace lock, and see if we can export.
4131	 */
4132	spa_open_ref(spa, FTAG);
4133	mutex_exit(&spa_namespace_lock);
4134	spa_async_suspend(spa);
4135	mutex_enter(&spa_namespace_lock);
4136	spa_close(spa, FTAG);
4137
4138	/*
4139	 * The pool will be in core if it's openable,
4140	 * in which case we can modify its state.
4141	 */
4142	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4143		/*
4144		 * Objsets may be open only because they're dirty, so we
4145		 * have to force it to sync before checking spa_refcnt.
4146		 */
4147		txg_wait_synced(spa->spa_dsl_pool, 0);
4148
4149		/*
4150		 * A pool cannot be exported or destroyed if there are active
4151		 * references.  If we are resetting a pool, allow references by
4152		 * fault injection handlers.
4153		 */
4154		if (!spa_refcount_zero(spa) ||
4155		    (spa->spa_inject_ref != 0 &&
4156		    new_state != POOL_STATE_UNINITIALIZED)) {
4157			spa_async_resume(spa);
4158			mutex_exit(&spa_namespace_lock);
4159			return (SET_ERROR(EBUSY));
4160		}
4161
4162		/*
4163		 * A pool cannot be exported if it has an active shared spare.
4164		 * This is to prevent other pools stealing the active spare
4165		 * from an exported pool. At user's own will, such pool can
4166		 * be forcedly exported.
4167		 */
4168		if (!force && new_state == POOL_STATE_EXPORTED &&
4169		    spa_has_active_shared_spare(spa)) {
4170			spa_async_resume(spa);
4171			mutex_exit(&spa_namespace_lock);
4172			return (SET_ERROR(EXDEV));
4173		}
4174
4175		/*
4176		 * We want this to be reflected on every label,
4177		 * so mark them all dirty.  spa_unload() will do the
4178		 * final sync that pushes these changes out.
4179		 */
4180		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4181			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4182			spa->spa_state = new_state;
4183			spa->spa_final_txg = spa_last_synced_txg(spa) +
4184			    TXG_DEFER_SIZE + 1;
4185			vdev_config_dirty(spa->spa_root_vdev);
4186			spa_config_exit(spa, SCL_ALL, FTAG);
4187		}
4188	}
4189
4190	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4191
4192	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4193		spa_unload(spa);
4194		spa_deactivate(spa);
4195	}
4196
4197	if (oldconfig && spa->spa_config)
4198		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4199
4200	if (new_state != POOL_STATE_UNINITIALIZED) {
4201		if (!hardforce)
4202			spa_config_sync(spa, B_TRUE, B_TRUE);
4203		spa_remove(spa);
4204	}
4205	mutex_exit(&spa_namespace_lock);
4206
4207	return (0);
4208}
4209
4210/*
4211 * Destroy a storage pool.
4212 */
4213int
4214spa_destroy(char *pool)
4215{
4216	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4217	    B_FALSE, B_FALSE));
4218}
4219
4220/*
4221 * Export a storage pool.
4222 */
4223int
4224spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4225    boolean_t hardforce)
4226{
4227	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4228	    force, hardforce));
4229}
4230
4231/*
4232 * Similar to spa_export(), this unloads the spa_t without actually removing it
4233 * from the namespace in any way.
4234 */
4235int
4236spa_reset(char *pool)
4237{
4238	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4239	    B_FALSE, B_FALSE));
4240}
4241
4242/*
4243 * ==========================================================================
4244 * Device manipulation
4245 * ==========================================================================
4246 */
4247
4248/*
4249 * Add a device to a storage pool.
4250 */
4251int
4252spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4253{
4254	uint64_t txg, id;
4255	int error;
4256	vdev_t *rvd = spa->spa_root_vdev;
4257	vdev_t *vd, *tvd;
4258	nvlist_t **spares, **l2cache;
4259	uint_t nspares, nl2cache;
4260
4261	ASSERT(spa_writeable(spa));
4262
4263	txg = spa_vdev_enter(spa);
4264
4265	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4266	    VDEV_ALLOC_ADD)) != 0)
4267		return (spa_vdev_exit(spa, NULL, txg, error));
4268
4269	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
4270
4271	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4272	    &nspares) != 0)
4273		nspares = 0;
4274
4275	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4276	    &nl2cache) != 0)
4277		nl2cache = 0;
4278
4279	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4280		return (spa_vdev_exit(spa, vd, txg, EINVAL));
4281
4282	if (vd->vdev_children != 0 &&
4283	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
4284		return (spa_vdev_exit(spa, vd, txg, error));
4285
4286	/*
4287	 * We must validate the spares and l2cache devices after checking the
4288	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
4289	 */
4290	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4291		return (spa_vdev_exit(spa, vd, txg, error));
4292
4293	/*
4294	 * Transfer each new top-level vdev from vd to rvd.
4295	 */
4296	for (int c = 0; c < vd->vdev_children; c++) {
4297
4298		/*
4299		 * Set the vdev id to the first hole, if one exists.
4300		 */
4301		for (id = 0; id < rvd->vdev_children; id++) {
4302			if (rvd->vdev_child[id]->vdev_ishole) {
4303				vdev_free(rvd->vdev_child[id]);
4304				break;
4305			}
4306		}
4307		tvd = vd->vdev_child[c];
4308		vdev_remove_child(vd, tvd);
4309		tvd->vdev_id = id;
4310		vdev_add_child(rvd, tvd);
4311		vdev_config_dirty(tvd);
4312	}
4313
4314	if (nspares != 0) {
4315		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4316		    ZPOOL_CONFIG_SPARES);
4317		spa_load_spares(spa);
4318		spa->spa_spares.sav_sync = B_TRUE;
4319	}
4320
4321	if (nl2cache != 0) {
4322		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4323		    ZPOOL_CONFIG_L2CACHE);
4324		spa_load_l2cache(spa);
4325		spa->spa_l2cache.sav_sync = B_TRUE;
4326	}
4327
4328	/*
4329	 * We have to be careful when adding new vdevs to an existing pool.
4330	 * If other threads start allocating from these vdevs before we
4331	 * sync the config cache, and we lose power, then upon reboot we may
4332	 * fail to open the pool because there are DVAs that the config cache
4333	 * can't translate.  Therefore, we first add the vdevs without
4334	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4335	 * and then let spa_config_update() initialize the new metaslabs.
4336	 *
4337	 * spa_load() checks for added-but-not-initialized vdevs, so that
4338	 * if we lose power at any point in this sequence, the remaining
4339	 * steps will be completed the next time we load the pool.
4340	 */
4341	(void) spa_vdev_exit(spa, vd, txg, 0);
4342
4343	mutex_enter(&spa_namespace_lock);
4344	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4345	mutex_exit(&spa_namespace_lock);
4346
4347	return (0);
4348}
4349
4350/*
4351 * Attach a device to a mirror.  The arguments are the path to any device
4352 * in the mirror, and the nvroot for the new device.  If the path specifies
4353 * a device that is not mirrored, we automatically insert the mirror vdev.
4354 *
4355 * If 'replacing' is specified, the new device is intended to replace the
4356 * existing device; in this case the two devices are made into their own
4357 * mirror using the 'replacing' vdev, which is functionally identical to
4358 * the mirror vdev (it actually reuses all the same ops) but has a few
4359 * extra rules: you can't attach to it after it's been created, and upon
4360 * completion of resilvering, the first disk (the one being replaced)
4361 * is automatically detached.
4362 */
4363int
4364spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4365{
4366	uint64_t txg, dtl_max_txg;
4367	vdev_t *rvd = spa->spa_root_vdev;
4368	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4369	vdev_ops_t *pvops;
4370	char *oldvdpath, *newvdpath;
4371	int newvd_isspare;
4372	int error;
4373
4374	ASSERT(spa_writeable(spa));
4375
4376	txg = spa_vdev_enter(spa);
4377
4378	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4379
4380	if (oldvd == NULL)
4381		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4382
4383	if (!oldvd->vdev_ops->vdev_op_leaf)
4384		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4385
4386	pvd = oldvd->vdev_parent;
4387
4388	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4389	    VDEV_ALLOC_ATTACH)) != 0)
4390		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4391
4392	if (newrootvd->vdev_children != 1)
4393		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4394
4395	newvd = newrootvd->vdev_child[0];
4396
4397	if (!newvd->vdev_ops->vdev_op_leaf)
4398		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4399
4400	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4401		return (spa_vdev_exit(spa, newrootvd, txg, error));
4402
4403	/*
4404	 * Spares can't replace logs
4405	 */
4406	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4407		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4408
4409	if (!replacing) {
4410		/*
4411		 * For attach, the only allowable parent is a mirror or the root
4412		 * vdev.
4413		 */
4414		if (pvd->vdev_ops != &vdev_mirror_ops &&
4415		    pvd->vdev_ops != &vdev_root_ops)
4416			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4417
4418		pvops = &vdev_mirror_ops;
4419	} else {
4420		/*
4421		 * Active hot spares can only be replaced by inactive hot
4422		 * spares.
4423		 */
4424		if (pvd->vdev_ops == &vdev_spare_ops &&
4425		    oldvd->vdev_isspare &&
4426		    !spa_has_spare(spa, newvd->vdev_guid))
4427			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4428
4429		/*
4430		 * If the source is a hot spare, and the parent isn't already a
4431		 * spare, then we want to create a new hot spare.  Otherwise, we
4432		 * want to create a replacing vdev.  The user is not allowed to
4433		 * attach to a spared vdev child unless the 'isspare' state is
4434		 * the same (spare replaces spare, non-spare replaces
4435		 * non-spare).
4436		 */
4437		if (pvd->vdev_ops == &vdev_replacing_ops &&
4438		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4439			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4440		} else if (pvd->vdev_ops == &vdev_spare_ops &&
4441		    newvd->vdev_isspare != oldvd->vdev_isspare) {
4442			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4443		}
4444
4445		if (newvd->vdev_isspare)
4446			pvops = &vdev_spare_ops;
4447		else
4448			pvops = &vdev_replacing_ops;
4449	}
4450
4451	/*
4452	 * Make sure the new device is big enough.
4453	 */
4454	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4455		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4456
4457	/*
4458	 * The new device cannot have a higher alignment requirement
4459	 * than the top-level vdev.
4460	 */
4461	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4462		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4463
4464	/*
4465	 * If this is an in-place replacement, update oldvd's path and devid
4466	 * to make it distinguishable from newvd, and unopenable from now on.
4467	 */
4468	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4469		spa_strfree(oldvd->vdev_path);
4470		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4471		    KM_SLEEP);
4472		(void) sprintf(oldvd->vdev_path, "%s/%s",
4473		    newvd->vdev_path, "old");
4474		if (oldvd->vdev_devid != NULL) {
4475			spa_strfree(oldvd->vdev_devid);
4476			oldvd->vdev_devid = NULL;
4477		}
4478	}
4479
4480	/* mark the device being resilvered */
4481	newvd->vdev_resilver_txg = txg;
4482
4483	/*
4484	 * If the parent is not a mirror, or if we're replacing, insert the new
4485	 * mirror/replacing/spare vdev above oldvd.
4486	 */
4487	if (pvd->vdev_ops != pvops)
4488		pvd = vdev_add_parent(oldvd, pvops);
4489
4490	ASSERT(pvd->vdev_top->vdev_parent == rvd);
4491	ASSERT(pvd->vdev_ops == pvops);
4492	ASSERT(oldvd->vdev_parent == pvd);
4493
4494	/*
4495	 * Extract the new device from its root and add it to pvd.
4496	 */
4497	vdev_remove_child(newrootvd, newvd);
4498	newvd->vdev_id = pvd->vdev_children;
4499	newvd->vdev_crtxg = oldvd->vdev_crtxg;
4500	vdev_add_child(pvd, newvd);
4501
4502	tvd = newvd->vdev_top;
4503	ASSERT(pvd->vdev_top == tvd);
4504	ASSERT(tvd->vdev_parent == rvd);
4505
4506	vdev_config_dirty(tvd);
4507
4508	/*
4509	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4510	 * for any dmu_sync-ed blocks.  It will propagate upward when
4511	 * spa_vdev_exit() calls vdev_dtl_reassess().
4512	 */
4513	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4514
4515	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4516	    dtl_max_txg - TXG_INITIAL);
4517
4518	if (newvd->vdev_isspare) {
4519		spa_spare_activate(newvd);
4520		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4521	}
4522
4523	oldvdpath = spa_strdup(oldvd->vdev_path);
4524	newvdpath = spa_strdup(newvd->vdev_path);
4525	newvd_isspare = newvd->vdev_isspare;
4526
4527	/*
4528	 * Mark newvd's DTL dirty in this txg.
4529	 */
4530	vdev_dirty(tvd, VDD_DTL, newvd, txg);
4531
4532	/*
4533	 * Schedule the resilver to restart in the future. We do this to
4534	 * ensure that dmu_sync-ed blocks have been stitched into the
4535	 * respective datasets.
4536	 */
4537	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4538
4539	/*
4540	 * Commit the config
4541	 */
4542	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4543
4544	spa_history_log_internal(spa, "vdev attach", NULL,
4545	    "%s vdev=%s %s vdev=%s",
4546	    replacing && newvd_isspare ? "spare in" :
4547	    replacing ? "replace" : "attach", newvdpath,
4548	    replacing ? "for" : "to", oldvdpath);
4549
4550	spa_strfree(oldvdpath);
4551	spa_strfree(newvdpath);
4552
4553	if (spa->spa_bootfs)
4554		spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4555
4556	return (0);
4557}
4558
4559/*
4560 * Detach a device from a mirror or replacing vdev.
4561 *
4562 * If 'replace_done' is specified, only detach if the parent
4563 * is a replacing vdev.
4564 */
4565int
4566spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4567{
4568	uint64_t txg;
4569	int error;
4570	vdev_t *rvd = spa->spa_root_vdev;
4571	vdev_t *vd, *pvd, *cvd, *tvd;
4572	boolean_t unspare = B_FALSE;
4573	uint64_t unspare_guid = 0;
4574	char *vdpath;
4575
4576	ASSERT(spa_writeable(spa));
4577
4578	txg = spa_vdev_enter(spa);
4579
4580	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4581
4582	if (vd == NULL)
4583		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4584
4585	if (!vd->vdev_ops->vdev_op_leaf)
4586		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4587
4588	pvd = vd->vdev_parent;
4589
4590	/*
4591	 * If the parent/child relationship is not as expected, don't do it.
4592	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4593	 * vdev that's replacing B with C.  The user's intent in replacing
4594	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4595	 * the replace by detaching C, the expected behavior is to end up
4596	 * M(A,B).  But suppose that right after deciding to detach C,
4597	 * the replacement of B completes.  We would have M(A,C), and then
4598	 * ask to detach C, which would leave us with just A -- not what
4599	 * the user wanted.  To prevent this, we make sure that the
4600	 * parent/child relationship hasn't changed -- in this example,
4601	 * that C's parent is still the replacing vdev R.
4602	 */
4603	if (pvd->vdev_guid != pguid && pguid != 0)
4604		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4605
4606	/*
4607	 * Only 'replacing' or 'spare' vdevs can be replaced.
4608	 */
4609	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4610	    pvd->vdev_ops != &vdev_spare_ops)
4611		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4612
4613	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4614	    spa_version(spa) >= SPA_VERSION_SPARES);
4615
4616	/*
4617	 * Only mirror, replacing, and spare vdevs support detach.
4618	 */
4619	if (pvd->vdev_ops != &vdev_replacing_ops &&
4620	    pvd->vdev_ops != &vdev_mirror_ops &&
4621	    pvd->vdev_ops != &vdev_spare_ops)
4622		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4623
4624	/*
4625	 * If this device has the only valid copy of some data,
4626	 * we cannot safely detach it.
4627	 */
4628	if (vdev_dtl_required(vd))
4629		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4630
4631	ASSERT(pvd->vdev_children >= 2);
4632
4633	/*
4634	 * If we are detaching the second disk from a replacing vdev, then
4635	 * check to see if we changed the original vdev's path to have "/old"
4636	 * at the end in spa_vdev_attach().  If so, undo that change now.
4637	 */
4638	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4639	    vd->vdev_path != NULL) {
4640		size_t len = strlen(vd->vdev_path);
4641
4642		for (int c = 0; c < pvd->vdev_children; c++) {
4643			cvd = pvd->vdev_child[c];
4644
4645			if (cvd == vd || cvd->vdev_path == NULL)
4646				continue;
4647
4648			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4649			    strcmp(cvd->vdev_path + len, "/old") == 0) {
4650				spa_strfree(cvd->vdev_path);
4651				cvd->vdev_path = spa_strdup(vd->vdev_path);
4652				break;
4653			}
4654		}
4655	}
4656
4657	/*
4658	 * If we are detaching the original disk from a spare, then it implies
4659	 * that the spare should become a real disk, and be removed from the
4660	 * active spare list for the pool.
4661	 */
4662	if (pvd->vdev_ops == &vdev_spare_ops &&
4663	    vd->vdev_id == 0 &&
4664	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4665		unspare = B_TRUE;
4666
4667	/*
4668	 * Erase the disk labels so the disk can be used for other things.
4669	 * This must be done after all other error cases are handled,
4670	 * but before we disembowel vd (so we can still do I/O to it).
4671	 * But if we can't do it, don't treat the error as fatal --
4672	 * it may be that the unwritability of the disk is the reason
4673	 * it's being detached!
4674	 */
4675	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4676
4677	/*
4678	 * Remove vd from its parent and compact the parent's children.
4679	 */
4680	vdev_remove_child(pvd, vd);
4681	vdev_compact_children(pvd);
4682
4683	/*
4684	 * Remember one of the remaining children so we can get tvd below.
4685	 */
4686	cvd = pvd->vdev_child[pvd->vdev_children - 1];
4687
4688	/*
4689	 * If we need to remove the remaining child from the list of hot spares,
4690	 * do it now, marking the vdev as no longer a spare in the process.
4691	 * We must do this before vdev_remove_parent(), because that can
4692	 * change the GUID if it creates a new toplevel GUID.  For a similar
4693	 * reason, we must remove the spare now, in the same txg as the detach;
4694	 * otherwise someone could attach a new sibling, change the GUID, and
4695	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4696	 */
4697	if (unspare) {
4698		ASSERT(cvd->vdev_isspare);
4699		spa_spare_remove(cvd);
4700		unspare_guid = cvd->vdev_guid;
4701		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4702		cvd->vdev_unspare = B_TRUE;
4703	}
4704
4705	/*
4706	 * If the parent mirror/replacing vdev only has one child,
4707	 * the parent is no longer needed.  Remove it from the tree.
4708	 */
4709	if (pvd->vdev_children == 1) {
4710		if (pvd->vdev_ops == &vdev_spare_ops)
4711			cvd->vdev_unspare = B_FALSE;
4712		vdev_remove_parent(cvd);
4713	}
4714
4715
4716	/*
4717	 * We don't set tvd until now because the parent we just removed
4718	 * may have been the previous top-level vdev.
4719	 */
4720	tvd = cvd->vdev_top;
4721	ASSERT(tvd->vdev_parent == rvd);
4722
4723	/*
4724	 * Reevaluate the parent vdev state.
4725	 */
4726	vdev_propagate_state(cvd);
4727
4728	/*
4729	 * If the 'autoexpand' property is set on the pool then automatically
4730	 * try to expand the size of the pool. For example if the device we
4731	 * just detached was smaller than the others, it may be possible to
4732	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4733	 * first so that we can obtain the updated sizes of the leaf vdevs.
4734	 */
4735	if (spa->spa_autoexpand) {
4736		vdev_reopen(tvd);
4737		vdev_expand(tvd, txg);
4738	}
4739
4740	vdev_config_dirty(tvd);
4741
4742	/*
4743	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4744	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4745	 * But first make sure we're not on any *other* txg's DTL list, to
4746	 * prevent vd from being accessed after it's freed.
4747	 */
4748	vdpath = spa_strdup(vd->vdev_path);
4749	for (int t = 0; t < TXG_SIZE; t++)
4750		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4751	vd->vdev_detached = B_TRUE;
4752	vdev_dirty(tvd, VDD_DTL, vd, txg);
4753
4754	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4755
4756	/* hang on to the spa before we release the lock */
4757	spa_open_ref(spa, FTAG);
4758
4759	error = spa_vdev_exit(spa, vd, txg, 0);
4760
4761	spa_history_log_internal(spa, "detach", NULL,
4762	    "vdev=%s", vdpath);
4763	spa_strfree(vdpath);
4764
4765	/*
4766	 * If this was the removal of the original device in a hot spare vdev,
4767	 * then we want to go through and remove the device from the hot spare
4768	 * list of every other pool.
4769	 */
4770	if (unspare) {
4771		spa_t *altspa = NULL;
4772
4773		mutex_enter(&spa_namespace_lock);
4774		while ((altspa = spa_next(altspa)) != NULL) {
4775			if (altspa->spa_state != POOL_STATE_ACTIVE ||
4776			    altspa == spa)
4777				continue;
4778
4779			spa_open_ref(altspa, FTAG);
4780			mutex_exit(&spa_namespace_lock);
4781			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4782			mutex_enter(&spa_namespace_lock);
4783			spa_close(altspa, FTAG);
4784		}
4785		mutex_exit(&spa_namespace_lock);
4786
4787		/* search the rest of the vdevs for spares to remove */
4788		spa_vdev_resilver_done(spa);
4789	}
4790
4791	/* all done with the spa; OK to release */
4792	mutex_enter(&spa_namespace_lock);
4793	spa_close(spa, FTAG);
4794	mutex_exit(&spa_namespace_lock);
4795
4796	return (error);
4797}
4798
4799/*
4800 * Split a set of devices from their mirrors, and create a new pool from them.
4801 */
4802int
4803spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4804    nvlist_t *props, boolean_t exp)
4805{
4806	int error = 0;
4807	uint64_t txg, *glist;
4808	spa_t *newspa;
4809	uint_t c, children, lastlog;
4810	nvlist_t **child, *nvl, *tmp;
4811	dmu_tx_t *tx;
4812	char *altroot = NULL;
4813	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4814	boolean_t activate_slog;
4815
4816	ASSERT(spa_writeable(spa));
4817
4818	txg = spa_vdev_enter(spa);
4819
4820	/* clear the log and flush everything up to now */
4821	activate_slog = spa_passivate_log(spa);
4822	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4823	error = spa_offline_log(spa);
4824	txg = spa_vdev_config_enter(spa);
4825
4826	if (activate_slog)
4827		spa_activate_log(spa);
4828
4829	if (error != 0)
4830		return (spa_vdev_exit(spa, NULL, txg, error));
4831
4832	/* check new spa name before going any further */
4833	if (spa_lookup(newname) != NULL)
4834		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4835
4836	/*
4837	 * scan through all the children to ensure they're all mirrors
4838	 */
4839	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4840	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4841	    &children) != 0)
4842		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4843
4844	/* first, check to ensure we've got the right child count */
4845	rvd = spa->spa_root_vdev;
4846	lastlog = 0;
4847	for (c = 0; c < rvd->vdev_children; c++) {
4848		vdev_t *vd = rvd->vdev_child[c];
4849
4850		/* don't count the holes & logs as children */
4851		if (vd->vdev_islog || vd->vdev_ishole) {
4852			if (lastlog == 0)
4853				lastlog = c;
4854			continue;
4855		}
4856
4857		lastlog = 0;
4858	}
4859	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4860		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4861
4862	/* next, ensure no spare or cache devices are part of the split */
4863	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4864	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4865		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4866
4867	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4868	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4869
4870	/* then, loop over each vdev and validate it */
4871	for (c = 0; c < children; c++) {
4872		uint64_t is_hole = 0;
4873
4874		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4875		    &is_hole);
4876
4877		if (is_hole != 0) {
4878			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4879			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4880				continue;
4881			} else {
4882				error = SET_ERROR(EINVAL);
4883				break;
4884			}
4885		}
4886
4887		/* which disk is going to be split? */
4888		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4889		    &glist[c]) != 0) {
4890			error = SET_ERROR(EINVAL);
4891			break;
4892		}
4893
4894		/* look it up in the spa */
4895		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4896		if (vml[c] == NULL) {
4897			error = SET_ERROR(ENODEV);
4898			break;
4899		}
4900
4901		/* make sure there's nothing stopping the split */
4902		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4903		    vml[c]->vdev_islog ||
4904		    vml[c]->vdev_ishole ||
4905		    vml[c]->vdev_isspare ||
4906		    vml[c]->vdev_isl2cache ||
4907		    !vdev_writeable(vml[c]) ||
4908		    vml[c]->vdev_children != 0 ||
4909		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4910		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4911			error = SET_ERROR(EINVAL);
4912			break;
4913		}
4914
4915		if (vdev_dtl_required(vml[c])) {
4916			error = SET_ERROR(EBUSY);
4917			break;
4918		}
4919
4920		/* we need certain info from the top level */
4921		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4922		    vml[c]->vdev_top->vdev_ms_array) == 0);
4923		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4924		    vml[c]->vdev_top->vdev_ms_shift) == 0);
4925		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4926		    vml[c]->vdev_top->vdev_asize) == 0);
4927		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4928		    vml[c]->vdev_top->vdev_ashift) == 0);
4929	}
4930
4931	if (error != 0) {
4932		kmem_free(vml, children * sizeof (vdev_t *));
4933		kmem_free(glist, children * sizeof (uint64_t));
4934		return (spa_vdev_exit(spa, NULL, txg, error));
4935	}
4936
4937	/* stop writers from using the disks */
4938	for (c = 0; c < children; c++) {
4939		if (vml[c] != NULL)
4940			vml[c]->vdev_offline = B_TRUE;
4941	}
4942	vdev_reopen(spa->spa_root_vdev);
4943
4944	/*
4945	 * Temporarily record the splitting vdevs in the spa config.  This
4946	 * will disappear once the config is regenerated.
4947	 */
4948	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4949	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4950	    glist, children) == 0);
4951	kmem_free(glist, children * sizeof (uint64_t));
4952
4953	mutex_enter(&spa->spa_props_lock);
4954	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4955	    nvl) == 0);
4956	mutex_exit(&spa->spa_props_lock);
4957	spa->spa_config_splitting = nvl;
4958	vdev_config_dirty(spa->spa_root_vdev);
4959
4960	/* configure and create the new pool */
4961	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4962	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4963	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4964	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4965	    spa_version(spa)) == 0);
4966	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4967	    spa->spa_config_txg) == 0);
4968	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4969	    spa_generate_guid(NULL)) == 0);
4970	(void) nvlist_lookup_string(props,
4971	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4972
4973	/* add the new pool to the namespace */
4974	newspa = spa_add(newname, config, altroot);
4975	newspa->spa_config_txg = spa->spa_config_txg;
4976	spa_set_log_state(newspa, SPA_LOG_CLEAR);
4977
4978	/* release the spa config lock, retaining the namespace lock */
4979	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4980
4981	if (zio_injection_enabled)
4982		zio_handle_panic_injection(spa, FTAG, 1);
4983
4984	spa_activate(newspa, spa_mode_global);
4985	spa_async_suspend(newspa);
4986
4987	/* create the new pool from the disks of the original pool */
4988	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4989	if (error)
4990		goto out;
4991
4992	/* if that worked, generate a real config for the new pool */
4993	if (newspa->spa_root_vdev != NULL) {
4994		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4995		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4996		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4997		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4998		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4999		    B_TRUE));
5000	}
5001
5002	/* set the props */
5003	if (props != NULL) {
5004		spa_configfile_set(newspa, props, B_FALSE);
5005		error = spa_prop_set(newspa, props);
5006		if (error)
5007			goto out;
5008	}
5009
5010	/* flush everything */
5011	txg = spa_vdev_config_enter(newspa);
5012	vdev_config_dirty(newspa->spa_root_vdev);
5013	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5014
5015	if (zio_injection_enabled)
5016		zio_handle_panic_injection(spa, FTAG, 2);
5017
5018	spa_async_resume(newspa);
5019
5020	/* finally, update the original pool's config */
5021	txg = spa_vdev_config_enter(spa);
5022	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5023	error = dmu_tx_assign(tx, TXG_WAIT);
5024	if (error != 0)
5025		dmu_tx_abort(tx);
5026	for (c = 0; c < children; c++) {
5027		if (vml[c] != NULL) {
5028			vdev_split(vml[c]);
5029			if (error == 0)
5030				spa_history_log_internal(spa, "detach", tx,
5031				    "vdev=%s", vml[c]->vdev_path);
5032			vdev_free(vml[c]);
5033		}
5034	}
5035	vdev_config_dirty(spa->spa_root_vdev);
5036	spa->spa_config_splitting = NULL;
5037	nvlist_free(nvl);
5038	if (error == 0)
5039		dmu_tx_commit(tx);
5040	(void) spa_vdev_exit(spa, NULL, txg, 0);
5041
5042	if (zio_injection_enabled)
5043		zio_handle_panic_injection(spa, FTAG, 3);
5044
5045	/* split is complete; log a history record */
5046	spa_history_log_internal(newspa, "split", NULL,
5047	    "from pool %s", spa_name(spa));
5048
5049	kmem_free(vml, children * sizeof (vdev_t *));
5050
5051	/* if we're not going to mount the filesystems in userland, export */
5052	if (exp)
5053		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5054		    B_FALSE, B_FALSE);
5055
5056	return (error);
5057
5058out:
5059	spa_unload(newspa);
5060	spa_deactivate(newspa);
5061	spa_remove(newspa);
5062
5063	txg = spa_vdev_config_enter(spa);
5064
5065	/* re-online all offlined disks */
5066	for (c = 0; c < children; c++) {
5067		if (vml[c] != NULL)
5068			vml[c]->vdev_offline = B_FALSE;
5069	}
5070	vdev_reopen(spa->spa_root_vdev);
5071
5072	nvlist_free(spa->spa_config_splitting);
5073	spa->spa_config_splitting = NULL;
5074	(void) spa_vdev_exit(spa, NULL, txg, error);
5075
5076	kmem_free(vml, children * sizeof (vdev_t *));
5077	return (error);
5078}
5079
5080static nvlist_t *
5081spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5082{
5083	for (int i = 0; i < count; i++) {
5084		uint64_t guid;
5085
5086		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5087		    &guid) == 0);
5088
5089		if (guid == target_guid)
5090			return (nvpp[i]);
5091	}
5092
5093	return (NULL);
5094}
5095
5096static void
5097spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5098	nvlist_t *dev_to_remove)
5099{
5100	nvlist_t **newdev = NULL;
5101
5102	if (count > 1)
5103		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5104
5105	for (int i = 0, j = 0; i < count; i++) {
5106		if (dev[i] == dev_to_remove)
5107			continue;
5108		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5109	}
5110
5111	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5112	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5113
5114	for (int i = 0; i < count - 1; i++)
5115		nvlist_free(newdev[i]);
5116
5117	if (count > 1)
5118		kmem_free(newdev, (count - 1) * sizeof (void *));
5119}
5120
5121/*
5122 * Evacuate the device.
5123 */
5124static int
5125spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5126{
5127	uint64_t txg;
5128	int error = 0;
5129
5130	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5131	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5132	ASSERT(vd == vd->vdev_top);
5133
5134	/*
5135	 * Evacuate the device.  We don't hold the config lock as writer
5136	 * since we need to do I/O but we do keep the
5137	 * spa_namespace_lock held.  Once this completes the device
5138	 * should no longer have any blocks allocated on it.
5139	 */
5140	if (vd->vdev_islog) {
5141		if (vd->vdev_stat.vs_alloc != 0)
5142			error = spa_offline_log(spa);
5143	} else {
5144		error = SET_ERROR(ENOTSUP);
5145	}
5146
5147	if (error)
5148		return (error);
5149
5150	/*
5151	 * The evacuation succeeded.  Remove any remaining MOS metadata
5152	 * associated with this vdev, and wait for these changes to sync.
5153	 */
5154	ASSERT0(vd->vdev_stat.vs_alloc);
5155	txg = spa_vdev_config_enter(spa);
5156	vd->vdev_removing = B_TRUE;
5157	vdev_dirty_leaves(vd, VDD_DTL, txg);
5158	vdev_config_dirty(vd);
5159	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5160
5161	return (0);
5162}
5163
5164/*
5165 * Complete the removal by cleaning up the namespace.
5166 */
5167static void
5168spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5169{
5170	vdev_t *rvd = spa->spa_root_vdev;
5171	uint64_t id = vd->vdev_id;
5172	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5173
5174	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5175	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5176	ASSERT(vd == vd->vdev_top);
5177
5178	/*
5179	 * Only remove any devices which are empty.
5180	 */
5181	if (vd->vdev_stat.vs_alloc != 0)
5182		return;
5183
5184	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5185
5186	if (list_link_active(&vd->vdev_state_dirty_node))
5187		vdev_state_clean(vd);
5188	if (list_link_active(&vd->vdev_config_dirty_node))
5189		vdev_config_clean(vd);
5190
5191	vdev_free(vd);
5192
5193	if (last_vdev) {
5194		vdev_compact_children(rvd);
5195	} else {
5196		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5197		vdev_add_child(rvd, vd);
5198	}
5199	vdev_config_dirty(rvd);
5200
5201	/*
5202	 * Reassess the health of our root vdev.
5203	 */
5204	vdev_reopen(rvd);
5205}
5206
5207/*
5208 * Remove a device from the pool -
5209 *
5210 * Removing a device from the vdev namespace requires several steps
5211 * and can take a significant amount of time.  As a result we use
5212 * the spa_vdev_config_[enter/exit] functions which allow us to
5213 * grab and release the spa_config_lock while still holding the namespace
5214 * lock.  During each step the configuration is synced out.
5215 *
5216 * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5217 * devices.
5218 */
5219int
5220spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5221{
5222	vdev_t *vd;
5223	metaslab_group_t *mg;
5224	nvlist_t **spares, **l2cache, *nv;
5225	uint64_t txg = 0;
5226	uint_t nspares, nl2cache;
5227	int error = 0;
5228	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5229
5230	ASSERT(spa_writeable(spa));
5231
5232	if (!locked)
5233		txg = spa_vdev_enter(spa);
5234
5235	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5236
5237	if (spa->spa_spares.sav_vdevs != NULL &&
5238	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5239	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5240	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5241		/*
5242		 * Only remove the hot spare if it's not currently in use
5243		 * in this pool.
5244		 */
5245		if (vd == NULL || unspare) {
5246			spa_vdev_remove_aux(spa->spa_spares.sav_config,
5247			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5248			spa_load_spares(spa);
5249			spa->spa_spares.sav_sync = B_TRUE;
5250		} else {
5251			error = SET_ERROR(EBUSY);
5252		}
5253	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
5254	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5255	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5256	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5257		/*
5258		 * Cache devices can always be removed.
5259		 */
5260		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5261		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5262		spa_load_l2cache(spa);
5263		spa->spa_l2cache.sav_sync = B_TRUE;
5264	} else if (vd != NULL && vd->vdev_islog) {
5265		ASSERT(!locked);
5266		ASSERT(vd == vd->vdev_top);
5267
5268		mg = vd->vdev_mg;
5269
5270		/*
5271		 * Stop allocating from this vdev.
5272		 */
5273		metaslab_group_passivate(mg);
5274
5275		/*
5276		 * Wait for the youngest allocations and frees to sync,
5277		 * and then wait for the deferral of those frees to finish.
5278		 */
5279		spa_vdev_config_exit(spa, NULL,
5280		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5281
5282		/*
5283		 * Attempt to evacuate the vdev.
5284		 */
5285		error = spa_vdev_remove_evacuate(spa, vd);
5286
5287		txg = spa_vdev_config_enter(spa);
5288
5289		/*
5290		 * If we couldn't evacuate the vdev, unwind.
5291		 */
5292		if (error) {
5293			metaslab_group_activate(mg);
5294			return (spa_vdev_exit(spa, NULL, txg, error));
5295		}
5296
5297		/*
5298		 * Clean up the vdev namespace.
5299		 */
5300		spa_vdev_remove_from_namespace(spa, vd);
5301
5302	} else if (vd != NULL) {
5303		/*
5304		 * Normal vdevs cannot be removed (yet).
5305		 */
5306		error = SET_ERROR(ENOTSUP);
5307	} else {
5308		/*
5309		 * There is no vdev of any kind with the specified guid.
5310		 */
5311		error = SET_ERROR(ENOENT);
5312	}
5313
5314	if (!locked)
5315		return (spa_vdev_exit(spa, NULL, txg, error));
5316
5317	return (error);
5318}
5319
5320/*
5321 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5322 * currently spared, so we can detach it.
5323 */
5324static vdev_t *
5325spa_vdev_resilver_done_hunt(vdev_t *vd)
5326{
5327	vdev_t *newvd, *oldvd;
5328
5329	for (int c = 0; c < vd->vdev_children; c++) {
5330		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5331		if (oldvd != NULL)
5332			return (oldvd);
5333	}
5334
5335	/*
5336	 * Check for a completed replacement.  We always consider the first
5337	 * vdev in the list to be the oldest vdev, and the last one to be
5338	 * the newest (see spa_vdev_attach() for how that works).  In
5339	 * the case where the newest vdev is faulted, we will not automatically
5340	 * remove it after a resilver completes.  This is OK as it will require
5341	 * user intervention to determine which disk the admin wishes to keep.
5342	 */
5343	if (vd->vdev_ops == &vdev_replacing_ops) {
5344		ASSERT(vd->vdev_children > 1);
5345
5346		newvd = vd->vdev_child[vd->vdev_children - 1];
5347		oldvd = vd->vdev_child[0];
5348
5349		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5350		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5351		    !vdev_dtl_required(oldvd))
5352			return (oldvd);
5353	}
5354
5355	/*
5356	 * Check for a completed resilver with the 'unspare' flag set.
5357	 */
5358	if (vd->vdev_ops == &vdev_spare_ops) {
5359		vdev_t *first = vd->vdev_child[0];
5360		vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5361
5362		if (last->vdev_unspare) {
5363			oldvd = first;
5364			newvd = last;
5365		} else if (first->vdev_unspare) {
5366			oldvd = last;
5367			newvd = first;
5368		} else {
5369			oldvd = NULL;
5370		}
5371
5372		if (oldvd != NULL &&
5373		    vdev_dtl_empty(newvd, DTL_MISSING) &&
5374		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5375		    !vdev_dtl_required(oldvd))
5376			return (oldvd);
5377
5378		/*
5379		 * If there are more than two spares attached to a disk,
5380		 * and those spares are not required, then we want to
5381		 * attempt to free them up now so that they can be used
5382		 * by other pools.  Once we're back down to a single
5383		 * disk+spare, we stop removing them.
5384		 */
5385		if (vd->vdev_children > 2) {
5386			newvd = vd->vdev_child[1];
5387
5388			if (newvd->vdev_isspare && last->vdev_isspare &&
5389			    vdev_dtl_empty(last, DTL_MISSING) &&
5390			    vdev_dtl_empty(last, DTL_OUTAGE) &&
5391			    !vdev_dtl_required(newvd))
5392				return (newvd);
5393		}
5394	}
5395
5396	return (NULL);
5397}
5398
5399static void
5400spa_vdev_resilver_done(spa_t *spa)
5401{
5402	vdev_t *vd, *pvd, *ppvd;
5403	uint64_t guid, sguid, pguid, ppguid;
5404
5405	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5406
5407	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5408		pvd = vd->vdev_parent;
5409		ppvd = pvd->vdev_parent;
5410		guid = vd->vdev_guid;
5411		pguid = pvd->vdev_guid;
5412		ppguid = ppvd->vdev_guid;
5413		sguid = 0;
5414		/*
5415		 * If we have just finished replacing a hot spared device, then
5416		 * we need to detach the parent's first child (the original hot
5417		 * spare) as well.
5418		 */
5419		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5420		    ppvd->vdev_children == 2) {
5421			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5422			sguid = ppvd->vdev_child[1]->vdev_guid;
5423		}
5424		ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5425
5426		spa_config_exit(spa, SCL_ALL, FTAG);
5427		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5428			return;
5429		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5430			return;
5431		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5432	}
5433
5434	spa_config_exit(spa, SCL_ALL, FTAG);
5435}
5436
5437/*
5438 * Update the stored path or FRU for this vdev.
5439 */
5440int
5441spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5442    boolean_t ispath)
5443{
5444	vdev_t *vd;
5445	boolean_t sync = B_FALSE;
5446
5447	ASSERT(spa_writeable(spa));
5448
5449	spa_vdev_state_enter(spa, SCL_ALL);
5450
5451	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5452		return (spa_vdev_state_exit(spa, NULL, ENOENT));
5453
5454	if (!vd->vdev_ops->vdev_op_leaf)
5455		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5456
5457	if (ispath) {
5458		if (strcmp(value, vd->vdev_path) != 0) {
5459			spa_strfree(vd->vdev_path);
5460			vd->vdev_path = spa_strdup(value);
5461			sync = B_TRUE;
5462		}
5463	} else {
5464		if (vd->vdev_fru == NULL) {
5465			vd->vdev_fru = spa_strdup(value);
5466			sync = B_TRUE;
5467		} else if (strcmp(value, vd->vdev_fru) != 0) {
5468			spa_strfree(vd->vdev_fru);
5469			vd->vdev_fru = spa_strdup(value);
5470			sync = B_TRUE;
5471		}
5472	}
5473
5474	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5475}
5476
5477int
5478spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5479{
5480	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5481}
5482
5483int
5484spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5485{
5486	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5487}
5488
5489/*
5490 * ==========================================================================
5491 * SPA Scanning
5492 * ==========================================================================
5493 */
5494
5495int
5496spa_scan_stop(spa_t *spa)
5497{
5498	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5499	if (dsl_scan_resilvering(spa->spa_dsl_pool))
5500		return (SET_ERROR(EBUSY));
5501	return (dsl_scan_cancel(spa->spa_dsl_pool));
5502}
5503
5504int
5505spa_scan(spa_t *spa, pool_scan_func_t func)
5506{
5507	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5508
5509	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5510		return (SET_ERROR(ENOTSUP));
5511
5512	/*
5513	 * If a resilver was requested, but there is no DTL on a
5514	 * writeable leaf device, we have nothing to do.
5515	 */
5516	if (func == POOL_SCAN_RESILVER &&
5517	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5518		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5519		return (0);
5520	}
5521
5522	return (dsl_scan(spa->spa_dsl_pool, func));
5523}
5524
5525/*
5526 * ==========================================================================
5527 * SPA async task processing
5528 * ==========================================================================
5529 */
5530
5531static void
5532spa_async_remove(spa_t *spa, vdev_t *vd)
5533{
5534	if (vd->vdev_remove_wanted) {
5535		vd->vdev_remove_wanted = B_FALSE;
5536		vd->vdev_delayed_close = B_FALSE;
5537		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5538
5539		/*
5540		 * We want to clear the stats, but we don't want to do a full
5541		 * vdev_clear() as that will cause us to throw away
5542		 * degraded/faulted state as well as attempt to reopen the
5543		 * device, all of which is a waste.
5544		 */
5545		vd->vdev_stat.vs_read_errors = 0;
5546		vd->vdev_stat.vs_write_errors = 0;
5547		vd->vdev_stat.vs_checksum_errors = 0;
5548
5549		vdev_state_dirty(vd->vdev_top);
5550	}
5551
5552	for (int c = 0; c < vd->vdev_children; c++)
5553		spa_async_remove(spa, vd->vdev_child[c]);
5554}
5555
5556static void
5557spa_async_probe(spa_t *spa, vdev_t *vd)
5558{
5559	if (vd->vdev_probe_wanted) {
5560		vd->vdev_probe_wanted = B_FALSE;
5561		vdev_reopen(vd);	/* vdev_open() does the actual probe */
5562	}
5563
5564	for (int c = 0; c < vd->vdev_children; c++)
5565		spa_async_probe(spa, vd->vdev_child[c]);
5566}
5567
5568static void
5569spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5570{
5571	sysevent_id_t eid;
5572	nvlist_t *attr;
5573	char *physpath;
5574
5575	if (!spa->spa_autoexpand)
5576		return;
5577
5578	for (int c = 0; c < vd->vdev_children; c++) {
5579		vdev_t *cvd = vd->vdev_child[c];
5580		spa_async_autoexpand(spa, cvd);
5581	}
5582
5583	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5584		return;
5585
5586	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5587	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5588
5589	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5590	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5591
5592	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5593	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5594
5595	nvlist_free(attr);
5596	kmem_free(physpath, MAXPATHLEN);
5597}
5598
5599static void
5600spa_async_thread(spa_t *spa)
5601{
5602	int tasks;
5603
5604	ASSERT(spa->spa_sync_on);
5605
5606	mutex_enter(&spa->spa_async_lock);
5607	tasks = spa->spa_async_tasks;
5608	spa->spa_async_tasks = 0;
5609	mutex_exit(&spa->spa_async_lock);
5610
5611	/*
5612	 * See if the config needs to be updated.
5613	 */
5614	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5615		uint64_t old_space, new_space;
5616
5617		mutex_enter(&spa_namespace_lock);
5618		old_space = metaslab_class_get_space(spa_normal_class(spa));
5619		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5620		new_space = metaslab_class_get_space(spa_normal_class(spa));
5621		mutex_exit(&spa_namespace_lock);
5622
5623		/*
5624		 * If the pool grew as a result of the config update,
5625		 * then log an internal history event.
5626		 */
5627		if (new_space != old_space) {
5628			spa_history_log_internal(spa, "vdev online", NULL,
5629			    "pool '%s' size: %llu(+%llu)",
5630			    spa_name(spa), new_space, new_space - old_space);
5631		}
5632	}
5633
5634	/*
5635	 * See if any devices need to be marked REMOVED.
5636	 */
5637	if (tasks & SPA_ASYNC_REMOVE) {
5638		spa_vdev_state_enter(spa, SCL_NONE);
5639		spa_async_remove(spa, spa->spa_root_vdev);
5640		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5641			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5642		for (int i = 0; i < spa->spa_spares.sav_count; i++)
5643			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5644		(void) spa_vdev_state_exit(spa, NULL, 0);
5645	}
5646
5647	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5648		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5649		spa_async_autoexpand(spa, spa->spa_root_vdev);
5650		spa_config_exit(spa, SCL_CONFIG, FTAG);
5651	}
5652
5653	/*
5654	 * See if any devices need to be probed.
5655	 */
5656	if (tasks & SPA_ASYNC_PROBE) {
5657		spa_vdev_state_enter(spa, SCL_NONE);
5658		spa_async_probe(spa, spa->spa_root_vdev);
5659		(void) spa_vdev_state_exit(spa, NULL, 0);
5660	}
5661
5662	/*
5663	 * If any devices are done replacing, detach them.
5664	 */
5665	if (tasks & SPA_ASYNC_RESILVER_DONE)
5666		spa_vdev_resilver_done(spa);
5667
5668	/*
5669	 * Kick off a resilver.
5670	 */
5671	if (tasks & SPA_ASYNC_RESILVER)
5672		dsl_resilver_restart(spa->spa_dsl_pool, 0);
5673
5674	/*
5675	 * Let the world know that we're done.
5676	 */
5677	mutex_enter(&spa->spa_async_lock);
5678	spa->spa_async_thread = NULL;
5679	cv_broadcast(&spa->spa_async_cv);
5680	mutex_exit(&spa->spa_async_lock);
5681	thread_exit();
5682}
5683
5684void
5685spa_async_suspend(spa_t *spa)
5686{
5687	mutex_enter(&spa->spa_async_lock);
5688	spa->spa_async_suspended++;
5689	while (spa->spa_async_thread != NULL)
5690		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5691	mutex_exit(&spa->spa_async_lock);
5692}
5693
5694void
5695spa_async_resume(spa_t *spa)
5696{
5697	mutex_enter(&spa->spa_async_lock);
5698	ASSERT(spa->spa_async_suspended != 0);
5699	spa->spa_async_suspended--;
5700	mutex_exit(&spa->spa_async_lock);
5701}
5702
5703static boolean_t
5704spa_async_tasks_pending(spa_t *spa)
5705{
5706	uint_t non_config_tasks;
5707	uint_t config_task;
5708	boolean_t config_task_suspended;
5709
5710	non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5711	config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5712	if (spa->spa_ccw_fail_time == 0) {
5713		config_task_suspended = B_FALSE;
5714	} else {
5715		config_task_suspended =
5716		    (gethrtime() - spa->spa_ccw_fail_time) <
5717		    (zfs_ccw_retry_interval * NANOSEC);
5718	}
5719
5720	return (non_config_tasks || (config_task && !config_task_suspended));
5721}
5722
5723static void
5724spa_async_dispatch(spa_t *spa)
5725{
5726	mutex_enter(&spa->spa_async_lock);
5727	if (spa_async_tasks_pending(spa) &&
5728	    !spa->spa_async_suspended &&
5729	    spa->spa_async_thread == NULL &&
5730	    rootdir != NULL)
5731		spa->spa_async_thread = thread_create(NULL, 0,
5732		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5733	mutex_exit(&spa->spa_async_lock);
5734}
5735
5736void
5737spa_async_request(spa_t *spa, int task)
5738{
5739	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5740	mutex_enter(&spa->spa_async_lock);
5741	spa->spa_async_tasks |= task;
5742	mutex_exit(&spa->spa_async_lock);
5743}
5744
5745/*
5746 * ==========================================================================
5747 * SPA syncing routines
5748 * ==========================================================================
5749 */
5750
5751static int
5752bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5753{
5754	bpobj_t *bpo = arg;
5755	bpobj_enqueue(bpo, bp, tx);
5756	return (0);
5757}
5758
5759static int
5760spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5761{
5762	zio_t *zio = arg;
5763
5764	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5765	    zio->io_flags));
5766	return (0);
5767}
5768
5769/*
5770 * Note: this simple function is not inlined to make it easier to dtrace the
5771 * amount of time spent syncing frees.
5772 */
5773static void
5774spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
5775{
5776	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5777	bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
5778	VERIFY(zio_wait(zio) == 0);
5779}
5780
5781/*
5782 * Note: this simple function is not inlined to make it easier to dtrace the
5783 * amount of time spent syncing deferred frees.
5784 */
5785static void
5786spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
5787{
5788	zio_t *zio = zio_root(spa, NULL, NULL, 0);
5789	VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
5790	    spa_free_sync_cb, zio, tx), ==, 0);
5791	VERIFY0(zio_wait(zio));
5792}
5793
5794
5795static void
5796spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5797{
5798	char *packed = NULL;
5799	size_t bufsize;
5800	size_t nvsize = 0;
5801	dmu_buf_t *db;
5802
5803	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5804
5805	/*
5806	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5807	 * information.  This avoids the dmu_buf_will_dirty() path and
5808	 * saves us a pre-read to get data we don't actually care about.
5809	 */
5810	bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5811	packed = kmem_alloc(bufsize, KM_SLEEP);
5812
5813	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5814	    KM_SLEEP) == 0);
5815	bzero(packed + nvsize, bufsize - nvsize);
5816
5817	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5818
5819	kmem_free(packed, bufsize);
5820
5821	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5822	dmu_buf_will_dirty(db, tx);
5823	*(uint64_t *)db->db_data = nvsize;
5824	dmu_buf_rele(db, FTAG);
5825}
5826
5827static void
5828spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5829    const char *config, const char *entry)
5830{
5831	nvlist_t *nvroot;
5832	nvlist_t **list;
5833	int i;
5834
5835	if (!sav->sav_sync)
5836		return;
5837
5838	/*
5839	 * Update the MOS nvlist describing the list of available devices.
5840	 * spa_validate_aux() will have already made sure this nvlist is
5841	 * valid and the vdevs are labeled appropriately.
5842	 */
5843	if (sav->sav_object == 0) {
5844		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5845		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5846		    sizeof (uint64_t), tx);
5847		VERIFY(zap_update(spa->spa_meta_objset,
5848		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5849		    &sav->sav_object, tx) == 0);
5850	}
5851
5852	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5853	if (sav->sav_count == 0) {
5854		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5855	} else {
5856		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5857		for (i = 0; i < sav->sav_count; i++)
5858			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5859			    B_FALSE, VDEV_CONFIG_L2CACHE);
5860		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5861		    sav->sav_count) == 0);
5862		for (i = 0; i < sav->sav_count; i++)
5863			nvlist_free(list[i]);
5864		kmem_free(list, sav->sav_count * sizeof (void *));
5865	}
5866
5867	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5868	nvlist_free(nvroot);
5869
5870	sav->sav_sync = B_FALSE;
5871}
5872
5873static void
5874spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5875{
5876	nvlist_t *config;
5877
5878	if (list_is_empty(&spa->spa_config_dirty_list))
5879		return;
5880
5881	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5882
5883	config = spa_config_generate(spa, spa->spa_root_vdev,
5884	    dmu_tx_get_txg(tx), B_FALSE);
5885
5886	/*
5887	 * If we're upgrading the spa version then make sure that
5888	 * the config object gets updated with the correct version.
5889	 */
5890	if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5891		fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5892		    spa->spa_uberblock.ub_version);
5893
5894	spa_config_exit(spa, SCL_STATE, FTAG);
5895
5896	if (spa->spa_config_syncing)
5897		nvlist_free(spa->spa_config_syncing);
5898	spa->spa_config_syncing = config;
5899
5900	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5901}
5902
5903static void
5904spa_sync_version(void *arg, dmu_tx_t *tx)
5905{
5906	uint64_t *versionp = arg;
5907	uint64_t version = *versionp;
5908	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5909
5910	/*
5911	 * Setting the version is special cased when first creating the pool.
5912	 */
5913	ASSERT(tx->tx_txg != TXG_INITIAL);
5914
5915	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5916	ASSERT(version >= spa_version(spa));
5917
5918	spa->spa_uberblock.ub_version = version;
5919	vdev_config_dirty(spa->spa_root_vdev);
5920	spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5921}
5922
5923/*
5924 * Set zpool properties.
5925 */
5926static void
5927spa_sync_props(void *arg, dmu_tx_t *tx)
5928{
5929	nvlist_t *nvp = arg;
5930	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5931	objset_t *mos = spa->spa_meta_objset;
5932	nvpair_t *elem = NULL;
5933
5934	mutex_enter(&spa->spa_props_lock);
5935
5936	while ((elem = nvlist_next_nvpair(nvp, elem))) {
5937		uint64_t intval;
5938		char *strval, *fname;
5939		zpool_prop_t prop;
5940		const char *propname;
5941		zprop_type_t proptype;
5942		spa_feature_t fid;
5943
5944		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5945		case ZPROP_INVAL:
5946			/*
5947			 * We checked this earlier in spa_prop_validate().
5948			 */
5949			ASSERT(zpool_prop_feature(nvpair_name(elem)));
5950
5951			fname = strchr(nvpair_name(elem), '@') + 1;
5952			VERIFY0(zfeature_lookup_name(fname, &fid));
5953
5954			spa_feature_enable(spa, fid, tx);
5955			spa_history_log_internal(spa, "set", tx,
5956			    "%s=enabled", nvpair_name(elem));
5957			break;
5958
5959		case ZPOOL_PROP_VERSION:
5960			intval = fnvpair_value_uint64(elem);
5961			/*
5962			 * The version is synced seperatly before other
5963			 * properties and should be correct by now.
5964			 */
5965			ASSERT3U(spa_version(spa), >=, intval);
5966			break;
5967
5968		case ZPOOL_PROP_ALTROOT:
5969			/*
5970			 * 'altroot' is a non-persistent property. It should
5971			 * have been set temporarily at creation or import time.
5972			 */
5973			ASSERT(spa->spa_root != NULL);
5974			break;
5975
5976		case ZPOOL_PROP_READONLY:
5977		case ZPOOL_PROP_CACHEFILE:
5978			/*
5979			 * 'readonly' and 'cachefile' are also non-persisitent
5980			 * properties.
5981			 */
5982			break;
5983		case ZPOOL_PROP_COMMENT:
5984			strval = fnvpair_value_string(elem);
5985			if (spa->spa_comment != NULL)
5986				spa_strfree(spa->spa_comment);
5987			spa->spa_comment = spa_strdup(strval);
5988			/*
5989			 * We need to dirty the configuration on all the vdevs
5990			 * so that their labels get updated.  It's unnecessary
5991			 * to do this for pool creation since the vdev's
5992			 * configuratoin has already been dirtied.
5993			 */
5994			if (tx->tx_txg != TXG_INITIAL)
5995				vdev_config_dirty(spa->spa_root_vdev);
5996			spa_history_log_internal(spa, "set", tx,
5997			    "%s=%s", nvpair_name(elem), strval);
5998			break;
5999		default:
6000			/*
6001			 * Set pool property values in the poolprops mos object.
6002			 */
6003			if (spa->spa_pool_props_object == 0) {
6004				spa->spa_pool_props_object =
6005				    zap_create_link(mos, DMU_OT_POOL_PROPS,
6006				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6007				    tx);
6008			}
6009
6010			/* normalize the property name */
6011			propname = zpool_prop_to_name(prop);
6012			proptype = zpool_prop_get_type(prop);
6013
6014			if (nvpair_type(elem) == DATA_TYPE_STRING) {
6015				ASSERT(proptype == PROP_TYPE_STRING);
6016				strval = fnvpair_value_string(elem);
6017				VERIFY0(zap_update(mos,
6018				    spa->spa_pool_props_object, propname,
6019				    1, strlen(strval) + 1, strval, tx));
6020				spa_history_log_internal(spa, "set", tx,
6021				    "%s=%s", nvpair_name(elem), strval);
6022			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6023				intval = fnvpair_value_uint64(elem);
6024
6025				if (proptype == PROP_TYPE_INDEX) {
6026					const char *unused;
6027					VERIFY0(zpool_prop_index_to_string(
6028					    prop, intval, &unused));
6029				}
6030				VERIFY0(zap_update(mos,
6031				    spa->spa_pool_props_object, propname,
6032				    8, 1, &intval, tx));
6033				spa_history_log_internal(spa, "set", tx,
6034				    "%s=%lld", nvpair_name(elem), intval);
6035			} else {
6036				ASSERT(0); /* not allowed */
6037			}
6038
6039			switch (prop) {
6040			case ZPOOL_PROP_DELEGATION:
6041				spa->spa_delegation = intval;
6042				break;
6043			case ZPOOL_PROP_BOOTFS:
6044				spa->spa_bootfs = intval;
6045				break;
6046			case ZPOOL_PROP_FAILUREMODE:
6047				spa->spa_failmode = intval;
6048				break;
6049			case ZPOOL_PROP_AUTOEXPAND:
6050				spa->spa_autoexpand = intval;
6051				if (tx->