spa.c revision c8ee1847e300c992c76cf25d6c165e6ccf64a113
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 2010 Sun Microsystems, Inc.  All rights reserved.
24 * Use is subject to license terms.
25 */
26
27/*
28 * This file contains all the routines used when modifying on-disk SPA state.
29 * This includes opening, importing, destroying, exporting a pool, and syncing a
30 * pool.
31 */
32
33#include <sys/zfs_context.h>
34#include <sys/fm/fs/zfs.h>
35#include <sys/spa_impl.h>
36#include <sys/zio.h>
37#include <sys/zio_checksum.h>
38#include <sys/dmu.h>
39#include <sys/dmu_tx.h>
40#include <sys/zap.h>
41#include <sys/zil.h>
42#include <sys/ddt.h>
43#include <sys/vdev_impl.h>
44#include <sys/metaslab.h>
45#include <sys/metaslab_impl.h>
46#include <sys/uberblock_impl.h>
47#include <sys/txg.h>
48#include <sys/avl.h>
49#include <sys/dmu_traverse.h>
50#include <sys/dmu_objset.h>
51#include <sys/unique.h>
52#include <sys/dsl_pool.h>
53#include <sys/dsl_dataset.h>
54#include <sys/dsl_dir.h>
55#include <sys/dsl_prop.h>
56#include <sys/dsl_synctask.h>
57#include <sys/fs/zfs.h>
58#include <sys/arc.h>
59#include <sys/callb.h>
60#include <sys/systeminfo.h>
61#include <sys/spa_boot.h>
62#include <sys/zfs_ioctl.h>
63
64#ifdef	_KERNEL
65#include <sys/bootprops.h>
66#include <sys/callb.h>
67#include <sys/cpupart.h>
68#include <sys/pool.h>
69#include <sys/sysdc.h>
70#include <sys/zone.h>
71#endif	/* _KERNEL */
72
73#include "zfs_prop.h"
74#include "zfs_comutil.h"
75
76typedef enum zti_modes {
77	zti_mode_fixed,			/* value is # of threads (min 1) */
78	zti_mode_online_percent,	/* value is % of online CPUs */
79	zti_mode_batch,			/* cpu-intensive; value is ignored */
80	zti_mode_null,			/* don't create a taskq */
81	zti_nmodes
82} zti_modes_t;
83
84#define	ZTI_FIX(n)	{ zti_mode_fixed, (n) }
85#define	ZTI_PCT(n)	{ zti_mode_online_percent, (n) }
86#define	ZTI_BATCH	{ zti_mode_batch, 0 }
87#define	ZTI_NULL	{ zti_mode_null, 0 }
88
89#define	ZTI_ONE		ZTI_FIX(1)
90
91typedef struct zio_taskq_info {
92	enum zti_modes zti_mode;
93	uint_t zti_value;
94} zio_taskq_info_t;
95
96static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
97	"issue", "issue_high", "intr", "intr_high"
98};
99
100/*
101 * Define the taskq threads for the following I/O types:
102 * 	NULL, READ, WRITE, FREE, CLAIM, and IOCTL
103 */
104const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
105	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
106	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
107	{ ZTI_FIX(8),	ZTI_NULL,	ZTI_BATCH,	ZTI_NULL },
108	{ ZTI_BATCH,	ZTI_FIX(5),	ZTI_FIX(8),	ZTI_FIX(5) },
109	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
110	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
111	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
112};
113
114static void spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx);
115static boolean_t spa_has_active_shared_spare(spa_t *spa);
116static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
117    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
118    char **ereport);
119
120uint_t		zio_taskq_batch_pct = 100;	/* 1 thread per cpu in pset */
121id_t		zio_taskq_psrset_bind = PS_NONE;
122boolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
123uint_t		zio_taskq_basedc = 80;		/* base duty cycle */
124
125boolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
126
127/*
128 * This (illegal) pool name is used when temporarily importing a spa_t in order
129 * to get the vdev stats associated with the imported devices.
130 */
131#define	TRYIMPORT_NAME	"$import"
132
133/*
134 * ==========================================================================
135 * SPA properties routines
136 * ==========================================================================
137 */
138
139/*
140 * Add a (source=src, propname=propval) list to an nvlist.
141 */
142static void
143spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
144    uint64_t intval, zprop_source_t src)
145{
146	const char *propname = zpool_prop_to_name(prop);
147	nvlist_t *propval;
148
149	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
150	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
151
152	if (strval != NULL)
153		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
154	else
155		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
156
157	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
158	nvlist_free(propval);
159}
160
161/*
162 * Get property values from the spa configuration.
163 */
164static void
165spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
166{
167	uint64_t size;
168	uint64_t alloc;
169	uint64_t cap, version;
170	zprop_source_t src = ZPROP_SRC_NONE;
171	spa_config_dirent_t *dp;
172
173	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
174
175	if (spa->spa_root_vdev != NULL) {
176		alloc = metaslab_class_get_alloc(spa_normal_class(spa));
177		size = metaslab_class_get_space(spa_normal_class(spa));
178		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
179		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
180		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
181		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
182		    size - alloc, src);
183
184		cap = (size == 0) ? 0 : (alloc * 100 / size);
185		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
186
187		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
188		    ddt_get_pool_dedup_ratio(spa), src);
189
190		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
191		    spa->spa_root_vdev->vdev_state, src);
192
193		version = spa_version(spa);
194		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
195			src = ZPROP_SRC_DEFAULT;
196		else
197			src = ZPROP_SRC_LOCAL;
198		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
199	}
200
201	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
202
203	if (spa->spa_root != NULL)
204		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
205		    0, ZPROP_SRC_LOCAL);
206
207	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
208		if (dp->scd_path == NULL) {
209			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
210			    "none", 0, ZPROP_SRC_LOCAL);
211		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
212			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
213			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
214		}
215	}
216}
217
218/*
219 * Get zpool property values.
220 */
221int
222spa_prop_get(spa_t *spa, nvlist_t **nvp)
223{
224	objset_t *mos = spa->spa_meta_objset;
225	zap_cursor_t zc;
226	zap_attribute_t za;
227	int err;
228
229	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
230
231	mutex_enter(&spa->spa_props_lock);
232
233	/*
234	 * Get properties from the spa config.
235	 */
236	spa_prop_get_config(spa, nvp);
237
238	/* If no pool property object, no more prop to get. */
239	if (mos == NULL || spa->spa_pool_props_object == 0) {
240		mutex_exit(&spa->spa_props_lock);
241		return (0);
242	}
243
244	/*
245	 * Get properties from the MOS pool property object.
246	 */
247	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
248	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
249	    zap_cursor_advance(&zc)) {
250		uint64_t intval = 0;
251		char *strval = NULL;
252		zprop_source_t src = ZPROP_SRC_DEFAULT;
253		zpool_prop_t prop;
254
255		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
256			continue;
257
258		switch (za.za_integer_length) {
259		case 8:
260			/* integer property */
261			if (za.za_first_integer !=
262			    zpool_prop_default_numeric(prop))
263				src = ZPROP_SRC_LOCAL;
264
265			if (prop == ZPOOL_PROP_BOOTFS) {
266				dsl_pool_t *dp;
267				dsl_dataset_t *ds = NULL;
268
269				dp = spa_get_dsl(spa);
270				rw_enter(&dp->dp_config_rwlock, RW_READER);
271				if (err = dsl_dataset_hold_obj(dp,
272				    za.za_first_integer, FTAG, &ds)) {
273					rw_exit(&dp->dp_config_rwlock);
274					break;
275				}
276
277				strval = kmem_alloc(
278				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
279				    KM_SLEEP);
280				dsl_dataset_name(ds, strval);
281				dsl_dataset_rele(ds, FTAG);
282				rw_exit(&dp->dp_config_rwlock);
283			} else {
284				strval = NULL;
285				intval = za.za_first_integer;
286			}
287
288			spa_prop_add_list(*nvp, prop, strval, intval, src);
289
290			if (strval != NULL)
291				kmem_free(strval,
292				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
293
294			break;
295
296		case 1:
297			/* string property */
298			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
299			err = zap_lookup(mos, spa->spa_pool_props_object,
300			    za.za_name, 1, za.za_num_integers, strval);
301			if (err) {
302				kmem_free(strval, za.za_num_integers);
303				break;
304			}
305			spa_prop_add_list(*nvp, prop, strval, 0, src);
306			kmem_free(strval, za.za_num_integers);
307			break;
308
309		default:
310			break;
311		}
312	}
313	zap_cursor_fini(&zc);
314	mutex_exit(&spa->spa_props_lock);
315out:
316	if (err && err != ENOENT) {
317		nvlist_free(*nvp);
318		*nvp = NULL;
319		return (err);
320	}
321
322	return (0);
323}
324
325/*
326 * Validate the given pool properties nvlist and modify the list
327 * for the property values to be set.
328 */
329static int
330spa_prop_validate(spa_t *spa, nvlist_t *props)
331{
332	nvpair_t *elem;
333	int error = 0, reset_bootfs = 0;
334	uint64_t objnum;
335
336	elem = NULL;
337	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
338		zpool_prop_t prop;
339		char *propname, *strval;
340		uint64_t intval;
341		objset_t *os;
342		char *slash;
343
344		propname = nvpair_name(elem);
345
346		if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
347			return (EINVAL);
348
349		switch (prop) {
350		case ZPOOL_PROP_VERSION:
351			error = nvpair_value_uint64(elem, &intval);
352			if (!error &&
353			    (intval < spa_version(spa) || intval > SPA_VERSION))
354				error = EINVAL;
355			break;
356
357		case ZPOOL_PROP_DELEGATION:
358		case ZPOOL_PROP_AUTOREPLACE:
359		case ZPOOL_PROP_LISTSNAPS:
360		case ZPOOL_PROP_AUTOEXPAND:
361			error = nvpair_value_uint64(elem, &intval);
362			if (!error && intval > 1)
363				error = EINVAL;
364			break;
365
366		case ZPOOL_PROP_BOOTFS:
367			/*
368			 * If the pool version is less than SPA_VERSION_BOOTFS,
369			 * or the pool is still being created (version == 0),
370			 * the bootfs property cannot be set.
371			 */
372			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
373				error = ENOTSUP;
374				break;
375			}
376
377			/*
378			 * Make sure the vdev config is bootable
379			 */
380			if (!vdev_is_bootable(spa->spa_root_vdev)) {
381				error = ENOTSUP;
382				break;
383			}
384
385			reset_bootfs = 1;
386
387			error = nvpair_value_string(elem, &strval);
388
389			if (!error) {
390				uint64_t compress;
391
392				if (strval == NULL || strval[0] == '\0') {
393					objnum = zpool_prop_default_numeric(
394					    ZPOOL_PROP_BOOTFS);
395					break;
396				}
397
398				if (error = dmu_objset_hold(strval, FTAG, &os))
399					break;
400
401				/* Must be ZPL and not gzip compressed. */
402
403				if (dmu_objset_type(os) != DMU_OST_ZFS) {
404					error = ENOTSUP;
405				} else if ((error = dsl_prop_get_integer(strval,
406				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
407				    &compress, NULL)) == 0 &&
408				    !BOOTFS_COMPRESS_VALID(compress)) {
409					error = ENOTSUP;
410				} else {
411					objnum = dmu_objset_id(os);
412				}
413				dmu_objset_rele(os, FTAG);
414			}
415			break;
416
417		case ZPOOL_PROP_FAILUREMODE:
418			error = nvpair_value_uint64(elem, &intval);
419			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
420			    intval > ZIO_FAILURE_MODE_PANIC))
421				error = EINVAL;
422
423			/*
424			 * This is a special case which only occurs when
425			 * the pool has completely failed. This allows
426			 * the user to change the in-core failmode property
427			 * without syncing it out to disk (I/Os might
428			 * currently be blocked). We do this by returning
429			 * EIO to the caller (spa_prop_set) to trick it
430			 * into thinking we encountered a property validation
431			 * error.
432			 */
433			if (!error && spa_suspended(spa)) {
434				spa->spa_failmode = intval;
435				error = EIO;
436			}
437			break;
438
439		case ZPOOL_PROP_CACHEFILE:
440			if ((error = nvpair_value_string(elem, &strval)) != 0)
441				break;
442
443			if (strval[0] == '\0')
444				break;
445
446			if (strcmp(strval, "none") == 0)
447				break;
448
449			if (strval[0] != '/') {
450				error = EINVAL;
451				break;
452			}
453
454			slash = strrchr(strval, '/');
455			ASSERT(slash != NULL);
456
457			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
458			    strcmp(slash, "/..") == 0)
459				error = EINVAL;
460			break;
461
462		case ZPOOL_PROP_DEDUPDITTO:
463			if (spa_version(spa) < SPA_VERSION_DEDUP)
464				error = ENOTSUP;
465			else
466				error = nvpair_value_uint64(elem, &intval);
467			if (error == 0 &&
468			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
469				error = EINVAL;
470			break;
471		}
472
473		if (error)
474			break;
475	}
476
477	if (!error && reset_bootfs) {
478		error = nvlist_remove(props,
479		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
480
481		if (!error) {
482			error = nvlist_add_uint64(props,
483			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
484		}
485	}
486
487	return (error);
488}
489
490void
491spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
492{
493	char *cachefile;
494	spa_config_dirent_t *dp;
495
496	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
497	    &cachefile) != 0)
498		return;
499
500	dp = kmem_alloc(sizeof (spa_config_dirent_t),
501	    KM_SLEEP);
502
503	if (cachefile[0] == '\0')
504		dp->scd_path = spa_strdup(spa_config_path);
505	else if (strcmp(cachefile, "none") == 0)
506		dp->scd_path = NULL;
507	else
508		dp->scd_path = spa_strdup(cachefile);
509
510	list_insert_head(&spa->spa_config_list, dp);
511	if (need_sync)
512		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
513}
514
515int
516spa_prop_set(spa_t *spa, nvlist_t *nvp)
517{
518	int error;
519	nvpair_t *elem;
520	boolean_t need_sync = B_FALSE;
521	zpool_prop_t prop;
522
523	if ((error = spa_prop_validate(spa, nvp)) != 0)
524		return (error);
525
526	elem = NULL;
527	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
528		if ((prop = zpool_name_to_prop(
529		    nvpair_name(elem))) == ZPROP_INVAL)
530			return (EINVAL);
531
532		if (prop == ZPOOL_PROP_CACHEFILE || prop == ZPOOL_PROP_ALTROOT)
533			continue;
534
535		need_sync = B_TRUE;
536		break;
537	}
538
539	if (need_sync)
540		return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
541		    spa, nvp, 3));
542	else
543		return (0);
544}
545
546/*
547 * If the bootfs property value is dsobj, clear it.
548 */
549void
550spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
551{
552	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
553		VERIFY(zap_remove(spa->spa_meta_objset,
554		    spa->spa_pool_props_object,
555		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
556		spa->spa_bootfs = 0;
557	}
558}
559
560/*
561 * ==========================================================================
562 * SPA state manipulation (open/create/destroy/import/export)
563 * ==========================================================================
564 */
565
566static int
567spa_error_entry_compare(const void *a, const void *b)
568{
569	spa_error_entry_t *sa = (spa_error_entry_t *)a;
570	spa_error_entry_t *sb = (spa_error_entry_t *)b;
571	int ret;
572
573	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
574	    sizeof (zbookmark_t));
575
576	if (ret < 0)
577		return (-1);
578	else if (ret > 0)
579		return (1);
580	else
581		return (0);
582}
583
584/*
585 * Utility function which retrieves copies of the current logs and
586 * re-initializes them in the process.
587 */
588void
589spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
590{
591	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
592
593	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
594	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
595
596	avl_create(&spa->spa_errlist_scrub,
597	    spa_error_entry_compare, sizeof (spa_error_entry_t),
598	    offsetof(spa_error_entry_t, se_avl));
599	avl_create(&spa->spa_errlist_last,
600	    spa_error_entry_compare, sizeof (spa_error_entry_t),
601	    offsetof(spa_error_entry_t, se_avl));
602}
603
604static taskq_t *
605spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
606    uint_t value)
607{
608	uint_t flags = TASKQ_PREPOPULATE;
609	boolean_t batch = B_FALSE;
610
611	switch (mode) {
612	case zti_mode_null:
613		return (NULL);		/* no taskq needed */
614
615	case zti_mode_fixed:
616		ASSERT3U(value, >=, 1);
617		value = MAX(value, 1);
618		break;
619
620	case zti_mode_batch:
621		batch = B_TRUE;
622		flags |= TASKQ_THREADS_CPU_PCT;
623		value = zio_taskq_batch_pct;
624		break;
625
626	case zti_mode_online_percent:
627		flags |= TASKQ_THREADS_CPU_PCT;
628		break;
629
630	default:
631		panic("unrecognized mode for %s taskq (%u:%u) in "
632		    "spa_activate()",
633		    name, mode, value);
634		break;
635	}
636
637	if (zio_taskq_sysdc && spa->spa_proc != &p0) {
638		if (batch)
639			flags |= TASKQ_DC_BATCH;
640
641		return (taskq_create_sysdc(name, value, 50, INT_MAX,
642		    spa->spa_proc, zio_taskq_basedc, flags));
643	}
644	return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
645	    spa->spa_proc, flags));
646}
647
648static void
649spa_create_zio_taskqs(spa_t *spa)
650{
651	for (int t = 0; t < ZIO_TYPES; t++) {
652		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
653			const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
654			enum zti_modes mode = ztip->zti_mode;
655			uint_t value = ztip->zti_value;
656			char name[32];
657
658			(void) snprintf(name, sizeof (name),
659			    "%s_%s", zio_type_name[t], zio_taskq_types[q]);
660
661			spa->spa_zio_taskq[t][q] =
662			    spa_taskq_create(spa, name, mode, value);
663		}
664	}
665}
666
667#ifdef _KERNEL
668static void
669spa_thread(void *arg)
670{
671	callb_cpr_t cprinfo;
672
673	spa_t *spa = arg;
674	user_t *pu = PTOU(curproc);
675
676	CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
677	    spa->spa_name);
678
679	ASSERT(curproc != &p0);
680	(void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
681	    "zpool-%s", spa->spa_name);
682	(void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
683
684	/* bind this thread to the requested psrset */
685	if (zio_taskq_psrset_bind != PS_NONE) {
686		pool_lock();
687		mutex_enter(&cpu_lock);
688		mutex_enter(&pidlock);
689		mutex_enter(&curproc->p_lock);
690
691		if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
692		    0, NULL, NULL) == 0)  {
693			curthread->t_bind_pset = zio_taskq_psrset_bind;
694		} else {
695			cmn_err(CE_WARN,
696			    "Couldn't bind process for zfs pool \"%s\" to "
697			    "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
698		}
699
700		mutex_exit(&curproc->p_lock);
701		mutex_exit(&pidlock);
702		mutex_exit(&cpu_lock);
703		pool_unlock();
704	}
705
706	if (zio_taskq_sysdc) {
707		sysdc_thread_enter(curthread, 100, 0);
708	}
709
710	spa->spa_proc = curproc;
711	spa->spa_did = curthread->t_did;
712
713	spa_create_zio_taskqs(spa);
714
715	mutex_enter(&spa->spa_proc_lock);
716	ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
717
718	spa->spa_proc_state = SPA_PROC_ACTIVE;
719	cv_broadcast(&spa->spa_proc_cv);
720
721	CALLB_CPR_SAFE_BEGIN(&cprinfo);
722	while (spa->spa_proc_state == SPA_PROC_ACTIVE)
723		cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
724	CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
725
726	ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
727	spa->spa_proc_state = SPA_PROC_GONE;
728	spa->spa_proc = &p0;
729	cv_broadcast(&spa->spa_proc_cv);
730	CALLB_CPR_EXIT(&cprinfo);	/* drops spa_proc_lock */
731
732	mutex_enter(&curproc->p_lock);
733	lwp_exit();
734}
735#endif
736
737/*
738 * Activate an uninitialized pool.
739 */
740static void
741spa_activate(spa_t *spa, int mode)
742{
743	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
744
745	spa->spa_state = POOL_STATE_ACTIVE;
746	spa->spa_mode = mode;
747
748	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
749	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
750
751	/* Try to create a covering process */
752	mutex_enter(&spa->spa_proc_lock);
753	ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
754	ASSERT(spa->spa_proc == &p0);
755	spa->spa_did = 0;
756
757	/* Only create a process if we're going to be around a while. */
758	if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
759		if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
760		    NULL, 0) == 0) {
761			spa->spa_proc_state = SPA_PROC_CREATED;
762			while (spa->spa_proc_state == SPA_PROC_CREATED) {
763				cv_wait(&spa->spa_proc_cv,
764				    &spa->spa_proc_lock);
765			}
766			ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
767			ASSERT(spa->spa_proc != &p0);
768			ASSERT(spa->spa_did != 0);
769		} else {
770#ifdef _KERNEL
771			cmn_err(CE_WARN,
772			    "Couldn't create process for zfs pool \"%s\"\n",
773			    spa->spa_name);
774#endif
775		}
776	}
777	mutex_exit(&spa->spa_proc_lock);
778
779	/* If we didn't create a process, we need to create our taskqs. */
780	if (spa->spa_proc == &p0) {
781		spa_create_zio_taskqs(spa);
782	}
783
784	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
785	    offsetof(vdev_t, vdev_config_dirty_node));
786	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
787	    offsetof(vdev_t, vdev_state_dirty_node));
788
789	txg_list_create(&spa->spa_vdev_txg_list,
790	    offsetof(struct vdev, vdev_txg_node));
791
792	avl_create(&spa->spa_errlist_scrub,
793	    spa_error_entry_compare, sizeof (spa_error_entry_t),
794	    offsetof(spa_error_entry_t, se_avl));
795	avl_create(&spa->spa_errlist_last,
796	    spa_error_entry_compare, sizeof (spa_error_entry_t),
797	    offsetof(spa_error_entry_t, se_avl));
798}
799
800/*
801 * Opposite of spa_activate().
802 */
803static void
804spa_deactivate(spa_t *spa)
805{
806	ASSERT(spa->spa_sync_on == B_FALSE);
807	ASSERT(spa->spa_dsl_pool == NULL);
808	ASSERT(spa->spa_root_vdev == NULL);
809	ASSERT(spa->spa_async_zio_root == NULL);
810	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
811
812	txg_list_destroy(&spa->spa_vdev_txg_list);
813
814	list_destroy(&spa->spa_config_dirty_list);
815	list_destroy(&spa->spa_state_dirty_list);
816
817	for (int t = 0; t < ZIO_TYPES; t++) {
818		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
819			if (spa->spa_zio_taskq[t][q] != NULL)
820				taskq_destroy(spa->spa_zio_taskq[t][q]);
821			spa->spa_zio_taskq[t][q] = NULL;
822		}
823	}
824
825	metaslab_class_destroy(spa->spa_normal_class);
826	spa->spa_normal_class = NULL;
827
828	metaslab_class_destroy(spa->spa_log_class);
829	spa->spa_log_class = NULL;
830
831	/*
832	 * If this was part of an import or the open otherwise failed, we may
833	 * still have errors left in the queues.  Empty them just in case.
834	 */
835	spa_errlog_drain(spa);
836
837	avl_destroy(&spa->spa_errlist_scrub);
838	avl_destroy(&spa->spa_errlist_last);
839
840	spa->spa_state = POOL_STATE_UNINITIALIZED;
841
842	mutex_enter(&spa->spa_proc_lock);
843	if (spa->spa_proc_state != SPA_PROC_NONE) {
844		ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
845		spa->spa_proc_state = SPA_PROC_DEACTIVATE;
846		cv_broadcast(&spa->spa_proc_cv);
847		while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
848			ASSERT(spa->spa_proc != &p0);
849			cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
850		}
851		ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
852		spa->spa_proc_state = SPA_PROC_NONE;
853	}
854	ASSERT(spa->spa_proc == &p0);
855	mutex_exit(&spa->spa_proc_lock);
856
857	/*
858	 * We want to make sure spa_thread() has actually exited the ZFS
859	 * module, so that the module can't be unloaded out from underneath
860	 * it.
861	 */
862	if (spa->spa_did != 0) {
863		thread_join(spa->spa_did);
864		spa->spa_did = 0;
865	}
866}
867
868/*
869 * Verify a pool configuration, and construct the vdev tree appropriately.  This
870 * will create all the necessary vdevs in the appropriate layout, with each vdev
871 * in the CLOSED state.  This will prep the pool before open/creation/import.
872 * All vdev validation is done by the vdev_alloc() routine.
873 */
874static int
875spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
876    uint_t id, int atype)
877{
878	nvlist_t **child;
879	uint_t children;
880	int error;
881
882	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
883		return (error);
884
885	if ((*vdp)->vdev_ops->vdev_op_leaf)
886		return (0);
887
888	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
889	    &child, &children);
890
891	if (error == ENOENT)
892		return (0);
893
894	if (error) {
895		vdev_free(*vdp);
896		*vdp = NULL;
897		return (EINVAL);
898	}
899
900	for (int c = 0; c < children; c++) {
901		vdev_t *vd;
902		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
903		    atype)) != 0) {
904			vdev_free(*vdp);
905			*vdp = NULL;
906			return (error);
907		}
908	}
909
910	ASSERT(*vdp != NULL);
911
912	return (0);
913}
914
915/*
916 * Opposite of spa_load().
917 */
918static void
919spa_unload(spa_t *spa)
920{
921	int i;
922
923	ASSERT(MUTEX_HELD(&spa_namespace_lock));
924
925	/*
926	 * Stop async tasks.
927	 */
928	spa_async_suspend(spa);
929
930	/*
931	 * Stop syncing.
932	 */
933	if (spa->spa_sync_on) {
934		txg_sync_stop(spa->spa_dsl_pool);
935		spa->spa_sync_on = B_FALSE;
936	}
937
938	/*
939	 * Wait for any outstanding async I/O to complete.
940	 */
941	if (spa->spa_async_zio_root != NULL) {
942		(void) zio_wait(spa->spa_async_zio_root);
943		spa->spa_async_zio_root = NULL;
944	}
945
946	/*
947	 * Close the dsl pool.
948	 */
949	if (spa->spa_dsl_pool) {
950		dsl_pool_close(spa->spa_dsl_pool);
951		spa->spa_dsl_pool = NULL;
952		spa->spa_meta_objset = NULL;
953	}
954
955	ddt_unload(spa);
956
957	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
958
959	/*
960	 * Drop and purge level 2 cache
961	 */
962	spa_l2cache_drop(spa);
963
964	/*
965	 * Close all vdevs.
966	 */
967	if (spa->spa_root_vdev)
968		vdev_free(spa->spa_root_vdev);
969	ASSERT(spa->spa_root_vdev == NULL);
970
971	for (i = 0; i < spa->spa_spares.sav_count; i++)
972		vdev_free(spa->spa_spares.sav_vdevs[i]);
973	if (spa->spa_spares.sav_vdevs) {
974		kmem_free(spa->spa_spares.sav_vdevs,
975		    spa->spa_spares.sav_count * sizeof (void *));
976		spa->spa_spares.sav_vdevs = NULL;
977	}
978	if (spa->spa_spares.sav_config) {
979		nvlist_free(spa->spa_spares.sav_config);
980		spa->spa_spares.sav_config = NULL;
981	}
982	spa->spa_spares.sav_count = 0;
983
984	for (i = 0; i < spa->spa_l2cache.sav_count; i++)
985		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
986	if (spa->spa_l2cache.sav_vdevs) {
987		kmem_free(spa->spa_l2cache.sav_vdevs,
988		    spa->spa_l2cache.sav_count * sizeof (void *));
989		spa->spa_l2cache.sav_vdevs = NULL;
990	}
991	if (spa->spa_l2cache.sav_config) {
992		nvlist_free(spa->spa_l2cache.sav_config);
993		spa->spa_l2cache.sav_config = NULL;
994	}
995	spa->spa_l2cache.sav_count = 0;
996
997	spa->spa_async_suspended = 0;
998
999	spa_config_exit(spa, SCL_ALL, FTAG);
1000}
1001
1002/*
1003 * Load (or re-load) the current list of vdevs describing the active spares for
1004 * this pool.  When this is called, we have some form of basic information in
1005 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1006 * then re-generate a more complete list including status information.
1007 */
1008static void
1009spa_load_spares(spa_t *spa)
1010{
1011	nvlist_t **spares;
1012	uint_t nspares;
1013	int i;
1014	vdev_t *vd, *tvd;
1015
1016	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1017
1018	/*
1019	 * First, close and free any existing spare vdevs.
1020	 */
1021	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1022		vd = spa->spa_spares.sav_vdevs[i];
1023
1024		/* Undo the call to spa_activate() below */
1025		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1026		    B_FALSE)) != NULL && tvd->vdev_isspare)
1027			spa_spare_remove(tvd);
1028		vdev_close(vd);
1029		vdev_free(vd);
1030	}
1031
1032	if (spa->spa_spares.sav_vdevs)
1033		kmem_free(spa->spa_spares.sav_vdevs,
1034		    spa->spa_spares.sav_count * sizeof (void *));
1035
1036	if (spa->spa_spares.sav_config == NULL)
1037		nspares = 0;
1038	else
1039		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1040		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1041
1042	spa->spa_spares.sav_count = (int)nspares;
1043	spa->spa_spares.sav_vdevs = NULL;
1044
1045	if (nspares == 0)
1046		return;
1047
1048	/*
1049	 * Construct the array of vdevs, opening them to get status in the
1050	 * process.   For each spare, there is potentially two different vdev_t
1051	 * structures associated with it: one in the list of spares (used only
1052	 * for basic validation purposes) and one in the active vdev
1053	 * configuration (if it's spared in).  During this phase we open and
1054	 * validate each vdev on the spare list.  If the vdev also exists in the
1055	 * active configuration, then we also mark this vdev as an active spare.
1056	 */
1057	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1058	    KM_SLEEP);
1059	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1060		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1061		    VDEV_ALLOC_SPARE) == 0);
1062		ASSERT(vd != NULL);
1063
1064		spa->spa_spares.sav_vdevs[i] = vd;
1065
1066		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1067		    B_FALSE)) != NULL) {
1068			if (!tvd->vdev_isspare)
1069				spa_spare_add(tvd);
1070
1071			/*
1072			 * We only mark the spare active if we were successfully
1073			 * able to load the vdev.  Otherwise, importing a pool
1074			 * with a bad active spare would result in strange
1075			 * behavior, because multiple pool would think the spare
1076			 * is actively in use.
1077			 *
1078			 * There is a vulnerability here to an equally bizarre
1079			 * circumstance, where a dead active spare is later
1080			 * brought back to life (onlined or otherwise).  Given
1081			 * the rarity of this scenario, and the extra complexity
1082			 * it adds, we ignore the possibility.
1083			 */
1084			if (!vdev_is_dead(tvd))
1085				spa_spare_activate(tvd);
1086		}
1087
1088		vd->vdev_top = vd;
1089		vd->vdev_aux = &spa->spa_spares;
1090
1091		if (vdev_open(vd) != 0)
1092			continue;
1093
1094		if (vdev_validate_aux(vd) == 0)
1095			spa_spare_add(vd);
1096	}
1097
1098	/*
1099	 * Recompute the stashed list of spares, with status information
1100	 * this time.
1101	 */
1102	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1103	    DATA_TYPE_NVLIST_ARRAY) == 0);
1104
1105	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1106	    KM_SLEEP);
1107	for (i = 0; i < spa->spa_spares.sav_count; i++)
1108		spares[i] = vdev_config_generate(spa,
1109		    spa->spa_spares.sav_vdevs[i], B_TRUE, B_TRUE, B_FALSE);
1110	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1111	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1112	for (i = 0; i < spa->spa_spares.sav_count; i++)
1113		nvlist_free(spares[i]);
1114	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1115}
1116
1117/*
1118 * Load (or re-load) the current list of vdevs describing the active l2cache for
1119 * this pool.  When this is called, we have some form of basic information in
1120 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1121 * then re-generate a more complete list including status information.
1122 * Devices which are already active have their details maintained, and are
1123 * not re-opened.
1124 */
1125static void
1126spa_load_l2cache(spa_t *spa)
1127{
1128	nvlist_t **l2cache;
1129	uint_t nl2cache;
1130	int i, j, oldnvdevs;
1131	uint64_t guid;
1132	vdev_t *vd, **oldvdevs, **newvdevs;
1133	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1134
1135	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1136
1137	if (sav->sav_config != NULL) {
1138		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1139		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1140		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1141	} else {
1142		nl2cache = 0;
1143	}
1144
1145	oldvdevs = sav->sav_vdevs;
1146	oldnvdevs = sav->sav_count;
1147	sav->sav_vdevs = NULL;
1148	sav->sav_count = 0;
1149
1150	/*
1151	 * Process new nvlist of vdevs.
1152	 */
1153	for (i = 0; i < nl2cache; i++) {
1154		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1155		    &guid) == 0);
1156
1157		newvdevs[i] = NULL;
1158		for (j = 0; j < oldnvdevs; j++) {
1159			vd = oldvdevs[j];
1160			if (vd != NULL && guid == vd->vdev_guid) {
1161				/*
1162				 * Retain previous vdev for add/remove ops.
1163				 */
1164				newvdevs[i] = vd;
1165				oldvdevs[j] = NULL;
1166				break;
1167			}
1168		}
1169
1170		if (newvdevs[i] == NULL) {
1171			/*
1172			 * Create new vdev
1173			 */
1174			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1175			    VDEV_ALLOC_L2CACHE) == 0);
1176			ASSERT(vd != NULL);
1177			newvdevs[i] = vd;
1178
1179			/*
1180			 * Commit this vdev as an l2cache device,
1181			 * even if it fails to open.
1182			 */
1183			spa_l2cache_add(vd);
1184
1185			vd->vdev_top = vd;
1186			vd->vdev_aux = sav;
1187
1188			spa_l2cache_activate(vd);
1189
1190			if (vdev_open(vd) != 0)
1191				continue;
1192
1193			(void) vdev_validate_aux(vd);
1194
1195			if (!vdev_is_dead(vd))
1196				l2arc_add_vdev(spa, vd);
1197		}
1198	}
1199
1200	/*
1201	 * Purge vdevs that were dropped
1202	 */
1203	for (i = 0; i < oldnvdevs; i++) {
1204		uint64_t pool;
1205
1206		vd = oldvdevs[i];
1207		if (vd != NULL) {
1208			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1209			    pool != 0ULL && l2arc_vdev_present(vd))
1210				l2arc_remove_vdev(vd);
1211			(void) vdev_close(vd);
1212			spa_l2cache_remove(vd);
1213		}
1214	}
1215
1216	if (oldvdevs)
1217		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1218
1219	if (sav->sav_config == NULL)
1220		goto out;
1221
1222	sav->sav_vdevs = newvdevs;
1223	sav->sav_count = (int)nl2cache;
1224
1225	/*
1226	 * Recompute the stashed list of l2cache devices, with status
1227	 * information this time.
1228	 */
1229	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1230	    DATA_TYPE_NVLIST_ARRAY) == 0);
1231
1232	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1233	for (i = 0; i < sav->sav_count; i++)
1234		l2cache[i] = vdev_config_generate(spa,
1235		    sav->sav_vdevs[i], B_TRUE, B_FALSE, B_TRUE);
1236	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1237	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1238out:
1239	for (i = 0; i < sav->sav_count; i++)
1240		nvlist_free(l2cache[i]);
1241	if (sav->sav_count)
1242		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1243}
1244
1245static int
1246load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1247{
1248	dmu_buf_t *db;
1249	char *packed = NULL;
1250	size_t nvsize = 0;
1251	int error;
1252	*value = NULL;
1253
1254	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1255	nvsize = *(uint64_t *)db->db_data;
1256	dmu_buf_rele(db, FTAG);
1257
1258	packed = kmem_alloc(nvsize, KM_SLEEP);
1259	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1260	    DMU_READ_PREFETCH);
1261	if (error == 0)
1262		error = nvlist_unpack(packed, nvsize, value, 0);
1263	kmem_free(packed, nvsize);
1264
1265	return (error);
1266}
1267
1268/*
1269 * Checks to see if the given vdev could not be opened, in which case we post a
1270 * sysevent to notify the autoreplace code that the device has been removed.
1271 */
1272static void
1273spa_check_removed(vdev_t *vd)
1274{
1275	for (int c = 0; c < vd->vdev_children; c++)
1276		spa_check_removed(vd->vdev_child[c]);
1277
1278	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1279		zfs_post_autoreplace(vd->vdev_spa, vd);
1280		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1281	}
1282}
1283
1284/*
1285 * Load the slog device state from the config object since it's possible
1286 * that the label does not contain the most up-to-date information.
1287 */
1288void
1289spa_load_log_state(spa_t *spa, nvlist_t *nv)
1290{
1291	vdev_t *ovd, *rvd = spa->spa_root_vdev;
1292
1293	/*
1294	 * Load the original root vdev tree from the passed config.
1295	 */
1296	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1297	VERIFY(spa_config_parse(spa, &ovd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1298
1299	for (int c = 0; c < rvd->vdev_children; c++) {
1300		vdev_t *cvd = rvd->vdev_child[c];
1301		if (cvd->vdev_islog)
1302			vdev_load_log_state(cvd, ovd->vdev_child[c]);
1303	}
1304	vdev_free(ovd);
1305	spa_config_exit(spa, SCL_ALL, FTAG);
1306}
1307
1308/*
1309 * Check for missing log devices
1310 */
1311int
1312spa_check_logs(spa_t *spa)
1313{
1314	switch (spa->spa_log_state) {
1315	case SPA_LOG_MISSING:
1316		/* need to recheck in case slog has been restored */
1317	case SPA_LOG_UNKNOWN:
1318		if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1319		    DS_FIND_CHILDREN)) {
1320			spa_set_log_state(spa, SPA_LOG_MISSING);
1321			return (1);
1322		}
1323		break;
1324	}
1325	return (0);
1326}
1327
1328static boolean_t
1329spa_passivate_log(spa_t *spa)
1330{
1331	vdev_t *rvd = spa->spa_root_vdev;
1332	boolean_t slog_found = B_FALSE;
1333
1334	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1335
1336	if (!spa_has_slogs(spa))
1337		return (B_FALSE);
1338
1339	for (int c = 0; c < rvd->vdev_children; c++) {
1340		vdev_t *tvd = rvd->vdev_child[c];
1341		metaslab_group_t *mg = tvd->vdev_mg;
1342
1343		if (tvd->vdev_islog) {
1344			metaslab_group_passivate(mg);
1345			slog_found = B_TRUE;
1346		}
1347	}
1348
1349	return (slog_found);
1350}
1351
1352static void
1353spa_activate_log(spa_t *spa)
1354{
1355	vdev_t *rvd = spa->spa_root_vdev;
1356
1357	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1358
1359	for (int c = 0; c < rvd->vdev_children; c++) {
1360		vdev_t *tvd = rvd->vdev_child[c];
1361		metaslab_group_t *mg = tvd->vdev_mg;
1362
1363		if (tvd->vdev_islog)
1364			metaslab_group_activate(mg);
1365	}
1366}
1367
1368int
1369spa_offline_log(spa_t *spa)
1370{
1371	int error = 0;
1372
1373	if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1374	    NULL, DS_FIND_CHILDREN)) == 0) {
1375
1376		/*
1377		 * We successfully offlined the log device, sync out the
1378		 * current txg so that the "stubby" block can be removed
1379		 * by zil_sync().
1380		 */
1381		txg_wait_synced(spa->spa_dsl_pool, 0);
1382	}
1383	return (error);
1384}
1385
1386static void
1387spa_aux_check_removed(spa_aux_vdev_t *sav)
1388{
1389	for (int i = 0; i < sav->sav_count; i++)
1390		spa_check_removed(sav->sav_vdevs[i]);
1391}
1392
1393void
1394spa_claim_notify(zio_t *zio)
1395{
1396	spa_t *spa = zio->io_spa;
1397
1398	if (zio->io_error)
1399		return;
1400
1401	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1402	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1403		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1404	mutex_exit(&spa->spa_props_lock);
1405}
1406
1407typedef struct spa_load_error {
1408	uint64_t	sle_meta_count;
1409	uint64_t	sle_data_count;
1410} spa_load_error_t;
1411
1412static void
1413spa_load_verify_done(zio_t *zio)
1414{
1415	blkptr_t *bp = zio->io_bp;
1416	spa_load_error_t *sle = zio->io_private;
1417	dmu_object_type_t type = BP_GET_TYPE(bp);
1418	int error = zio->io_error;
1419
1420	if (error) {
1421		if ((BP_GET_LEVEL(bp) != 0 || dmu_ot[type].ot_metadata) &&
1422		    type != DMU_OT_INTENT_LOG)
1423			atomic_add_64(&sle->sle_meta_count, 1);
1424		else
1425			atomic_add_64(&sle->sle_data_count, 1);
1426	}
1427	zio_data_buf_free(zio->io_data, zio->io_size);
1428}
1429
1430/*ARGSUSED*/
1431static int
1432spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1433    const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1434{
1435	if (bp != NULL) {
1436		zio_t *rio = arg;
1437		size_t size = BP_GET_PSIZE(bp);
1438		void *data = zio_data_buf_alloc(size);
1439
1440		zio_nowait(zio_read(rio, spa, bp, data, size,
1441		    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1442		    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1443		    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1444	}
1445	return (0);
1446}
1447
1448static int
1449spa_load_verify(spa_t *spa)
1450{
1451	zio_t *rio;
1452	spa_load_error_t sle = { 0 };
1453	zpool_rewind_policy_t policy;
1454	boolean_t verify_ok = B_FALSE;
1455	int error;
1456
1457	zpool_get_rewind_policy(spa->spa_config, &policy);
1458
1459	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1460		return (0);
1461
1462	rio = zio_root(spa, NULL, &sle,
1463	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1464
1465	error = traverse_pool(spa, spa->spa_verify_min_txg,
1466	    TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1467
1468	(void) zio_wait(rio);
1469
1470	spa->spa_load_meta_errors = sle.sle_meta_count;
1471	spa->spa_load_data_errors = sle.sle_data_count;
1472
1473	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1474	    sle.sle_data_count <= policy.zrp_maxdata) {
1475		verify_ok = B_TRUE;
1476		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1477		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1478	} else {
1479		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1480	}
1481
1482	if (error) {
1483		if (error != ENXIO && error != EIO)
1484			error = EIO;
1485		return (error);
1486	}
1487
1488	return (verify_ok ? 0 : EIO);
1489}
1490
1491/*
1492 * Find a value in the pool props object.
1493 */
1494static void
1495spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1496{
1497	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1498	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1499}
1500
1501/*
1502 * Find a value in the pool directory object.
1503 */
1504static int
1505spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1506{
1507	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1508	    name, sizeof (uint64_t), 1, val));
1509}
1510
1511static int
1512spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1513{
1514	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1515	return (err);
1516}
1517
1518/*
1519 * Fix up config after a partly-completed split.  This is done with the
1520 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1521 * pool have that entry in their config, but only the splitting one contains
1522 * a list of all the guids of the vdevs that are being split off.
1523 *
1524 * This function determines what to do with that list: either rejoin
1525 * all the disks to the pool, or complete the splitting process.  To attempt
1526 * the rejoin, each disk that is offlined is marked online again, and
1527 * we do a reopen() call.  If the vdev label for every disk that was
1528 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1529 * then we call vdev_split() on each disk, and complete the split.
1530 *
1531 * Otherwise we leave the config alone, with all the vdevs in place in
1532 * the original pool.
1533 */
1534static void
1535spa_try_repair(spa_t *spa, nvlist_t *config)
1536{
1537	uint_t extracted;
1538	uint64_t *glist;
1539	uint_t i, gcount;
1540	nvlist_t *nvl;
1541	vdev_t **vd;
1542	boolean_t attempt_reopen;
1543
1544	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1545		return;
1546
1547	/* check that the config is complete */
1548	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1549	    &glist, &gcount) != 0)
1550		return;
1551
1552	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1553
1554	/* attempt to online all the vdevs & validate */
1555	attempt_reopen = B_TRUE;
1556	for (i = 0; i < gcount; i++) {
1557		if (glist[i] == 0)	/* vdev is hole */
1558			continue;
1559
1560		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1561		if (vd[i] == NULL) {
1562			/*
1563			 * Don't bother attempting to reopen the disks;
1564			 * just do the split.
1565			 */
1566			attempt_reopen = B_FALSE;
1567		} else {
1568			/* attempt to re-online it */
1569			vd[i]->vdev_offline = B_FALSE;
1570		}
1571	}
1572
1573	if (attempt_reopen) {
1574		vdev_reopen(spa->spa_root_vdev);
1575
1576		/* check each device to see what state it's in */
1577		for (extracted = 0, i = 0; i < gcount; i++) {
1578			if (vd[i] != NULL &&
1579			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1580				break;
1581			++extracted;
1582		}
1583	}
1584
1585	/*
1586	 * If every disk has been moved to the new pool, or if we never
1587	 * even attempted to look at them, then we split them off for
1588	 * good.
1589	 */
1590	if (!attempt_reopen || gcount == extracted) {
1591		for (i = 0; i < gcount; i++)
1592			if (vd[i] != NULL)
1593				vdev_split(vd[i]);
1594		vdev_reopen(spa->spa_root_vdev);
1595	}
1596
1597	kmem_free(vd, gcount * sizeof (vdev_t *));
1598}
1599
1600static int
1601spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1602    boolean_t mosconfig)
1603{
1604	nvlist_t *config = spa->spa_config;
1605	char *ereport = FM_EREPORT_ZFS_POOL;
1606	int error;
1607	uint64_t pool_guid;
1608	nvlist_t *nvl;
1609
1610	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1611		return (EINVAL);
1612
1613	/*
1614	 * Versioning wasn't explicitly added to the label until later, so if
1615	 * it's not present treat it as the initial version.
1616	 */
1617	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1618	    &spa->spa_ubsync.ub_version) != 0)
1619		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1620
1621	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1622	    &spa->spa_config_txg);
1623
1624	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1625	    spa_guid_exists(pool_guid, 0)) {
1626		error = EEXIST;
1627	} else {
1628		spa->spa_load_guid = pool_guid;
1629
1630		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1631		    &nvl) == 0) {
1632			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1633			    KM_SLEEP) == 0);
1634		}
1635
1636		error = spa_load_impl(spa, pool_guid, config, state, type,
1637		    mosconfig, &ereport);
1638	}
1639
1640	spa->spa_minref = refcount_count(&spa->spa_refcount);
1641	if (error && error != EBADF)
1642		zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1643	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1644	spa->spa_ena = 0;
1645
1646	return (error);
1647}
1648
1649/*
1650 * Load an existing storage pool, using the pool's builtin spa_config as a
1651 * source of configuration information.
1652 */
1653static int
1654spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1655    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1656    char **ereport)
1657{
1658	int error = 0;
1659	nvlist_t *nvconfig, *nvroot = NULL;
1660	vdev_t *rvd;
1661	uberblock_t *ub = &spa->spa_uberblock;
1662	uint64_t config_cache_txg = spa->spa_config_txg;
1663	int orig_mode = spa->spa_mode;
1664	int parse;
1665
1666	/*
1667	 * If this is an untrusted config, access the pool in read-only mode.
1668	 * This prevents things like resilvering recently removed devices.
1669	 */
1670	if (!mosconfig)
1671		spa->spa_mode = FREAD;
1672
1673	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1674
1675	spa->spa_load_state = state;
1676
1677	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1678		return (EINVAL);
1679
1680	parse = (type == SPA_IMPORT_EXISTING ?
1681	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1682
1683	/*
1684	 * Create "The Godfather" zio to hold all async IOs
1685	 */
1686	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1687	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1688
1689	/*
1690	 * Parse the configuration into a vdev tree.  We explicitly set the
1691	 * value that will be returned by spa_version() since parsing the
1692	 * configuration requires knowing the version number.
1693	 */
1694	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1695	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1696	spa_config_exit(spa, SCL_ALL, FTAG);
1697
1698	if (error != 0)
1699		return (error);
1700
1701	ASSERT(spa->spa_root_vdev == rvd);
1702
1703	if (type != SPA_IMPORT_ASSEMBLE) {
1704		ASSERT(spa_guid(spa) == pool_guid);
1705	}
1706
1707	/*
1708	 * Try to open all vdevs, loading each label in the process.
1709	 */
1710	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1711	error = vdev_open(rvd);
1712	spa_config_exit(spa, SCL_ALL, FTAG);
1713	if (error != 0)
1714		return (error);
1715
1716	/*
1717	 * We need to validate the vdev labels against the configuration that
1718	 * we have in hand, which is dependent on the setting of mosconfig. If
1719	 * mosconfig is true then we're validating the vdev labels based on
1720	 * that config.  Otherwise, we're validating against the cached config
1721	 * (zpool.cache) that was read when we loaded the zfs module, and then
1722	 * later we will recursively call spa_load() and validate against
1723	 * the vdev config.
1724	 *
1725	 * If we're assembling a new pool that's been split off from an
1726	 * existing pool, the labels haven't yet been updated so we skip
1727	 * validation for now.
1728	 */
1729	if (type != SPA_IMPORT_ASSEMBLE) {
1730		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1731		error = vdev_validate(rvd);
1732		spa_config_exit(spa, SCL_ALL, FTAG);
1733
1734		if (error != 0)
1735			return (error);
1736
1737		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1738			return (ENXIO);
1739	}
1740
1741	/*
1742	 * Find the best uberblock.
1743	 */
1744	vdev_uberblock_load(NULL, rvd, ub);
1745
1746	/*
1747	 * If we weren't able to find a single valid uberblock, return failure.
1748	 */
1749	if (ub->ub_txg == 0)
1750		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
1751
1752	/*
1753	 * If the pool is newer than the code, we can't open it.
1754	 */
1755	if (ub->ub_version > SPA_VERSION)
1756		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
1757
1758	/*
1759	 * If the vdev guid sum doesn't match the uberblock, we have an
1760	 * incomplete configuration.
1761	 */
1762	if (mosconfig && type != SPA_IMPORT_ASSEMBLE &&
1763	    rvd->vdev_guid_sum != ub->ub_guid_sum)
1764		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
1765
1766	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
1767		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1768		spa_try_repair(spa, config);
1769		spa_config_exit(spa, SCL_ALL, FTAG);
1770		nvlist_free(spa->spa_config_splitting);
1771		spa->spa_config_splitting = NULL;
1772	}
1773
1774	/*
1775	 * Initialize internal SPA structures.
1776	 */
1777	spa->spa_state = POOL_STATE_ACTIVE;
1778	spa->spa_ubsync = spa->spa_uberblock;
1779	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
1780	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
1781	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
1782	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
1783	spa->spa_claim_max_txg = spa->spa_first_txg;
1784
1785	error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1786	if (error)
1787		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1788	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1789
1790	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
1791		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1792
1793	if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
1794		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1795
1796	if (!mosconfig) {
1797		uint64_t hostid;
1798		nvlist_t *policy = NULL;
1799
1800		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
1801		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
1802			char *hostname;
1803			unsigned long myhostid = 0;
1804
1805			VERIFY(nvlist_lookup_string(nvconfig,
1806			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
1807
1808#ifdef	_KERNEL
1809			myhostid = zone_get_hostid(NULL);
1810#else	/* _KERNEL */
1811			/*
1812			 * We're emulating the system's hostid in userland, so
1813			 * we can't use zone_get_hostid().
1814			 */
1815			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
1816#endif	/* _KERNEL */
1817			if (hostid != 0 && myhostid != 0 &&
1818			    hostid != myhostid) {
1819				cmn_err(CE_WARN, "pool '%s' could not be "
1820				    "loaded as it was last accessed by "
1821				    "another system (host: %s hostid: 0x%lx). "
1822				    "See: http://www.sun.com/msg/ZFS-8000-EY",
1823				    spa_name(spa), hostname,
1824				    (unsigned long)hostid);
1825				return (EBADF);
1826			}
1827		}
1828		if (nvlist_lookup_nvlist(spa->spa_config,
1829		    ZPOOL_REWIND_POLICY, &policy) == 0)
1830			VERIFY(nvlist_add_nvlist(nvconfig,
1831			    ZPOOL_REWIND_POLICY, policy) == 0);
1832
1833		spa_config_set(spa, nvconfig);
1834		spa_unload(spa);
1835		spa_deactivate(spa);
1836		spa_activate(spa, orig_mode);
1837
1838		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
1839	}
1840
1841	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPLIST,
1842	    &spa->spa_deferred_bplist_obj) != 0)
1843		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1844
1845	/*
1846	 * Load the bit that tells us to use the new accounting function
1847	 * (raid-z deflation).  If we have an older pool, this will not
1848	 * be present.
1849	 */
1850	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
1851	if (error != 0 && error != ENOENT)
1852		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1853
1854	/*
1855	 * Load the persistent error log.  If we have an older pool, this will
1856	 * not be present.
1857	 */
1858	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
1859	if (error != 0 && error != ENOENT)
1860		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1861
1862	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
1863	    &spa->spa_errlog_scrub);
1864	if (error != 0 && error != ENOENT)
1865		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1866
1867	/*
1868	 * Load the history object.  If we have an older pool, this
1869	 * will not be present.
1870	 */
1871	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
1872	if (error != 0 && error != ENOENT)
1873		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1874
1875	/*
1876	 * If we're assembling the pool from the split-off vdevs of
1877	 * an existing pool, we don't want to attach the spares & cache
1878	 * devices.
1879	 */
1880
1881	/*
1882	 * Load any hot spares for this pool.
1883	 */
1884	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
1885	if (error != 0 && error != ENOENT)
1886		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1887	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
1888		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
1889		if (load_nvlist(spa, spa->spa_spares.sav_object,
1890		    &spa->spa_spares.sav_config) != 0)
1891			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1892
1893		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1894		spa_load_spares(spa);
1895		spa_config_exit(spa, SCL_ALL, FTAG);
1896	} else if (error == 0) {
1897		spa->spa_spares.sav_sync = B_TRUE;
1898	}
1899
1900	/*
1901	 * Load any level 2 ARC devices for this pool.
1902	 */
1903	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
1904	    &spa->spa_l2cache.sav_object);
1905	if (error != 0 && error != ENOENT)
1906		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1907	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
1908		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
1909		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
1910		    &spa->spa_l2cache.sav_config) != 0)
1911			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1912
1913		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1914		spa_load_l2cache(spa);
1915		spa_config_exit(spa, SCL_ALL, FTAG);
1916	} else if (error == 0) {
1917		spa->spa_l2cache.sav_sync = B_TRUE;
1918	}
1919
1920	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
1921
1922	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
1923	if (error && error != ENOENT)
1924		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1925
1926	if (error == 0) {
1927		uint64_t autoreplace;
1928
1929		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
1930		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
1931		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
1932		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
1933		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
1934		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
1935		    &spa->spa_dedup_ditto);
1936
1937		spa->spa_autoreplace = (autoreplace != 0);
1938	}
1939
1940	/*
1941	 * If the 'autoreplace' property is set, then post a resource notifying
1942	 * the ZFS DE that it should not issue any faults for unopenable
1943	 * devices.  We also iterate over the vdevs, and post a sysevent for any
1944	 * unopenable vdevs so that the normal autoreplace handler can take
1945	 * over.
1946	 */
1947	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
1948		spa_check_removed(spa->spa_root_vdev);
1949		/*
1950		 * For the import case, this is done in spa_import(), because
1951		 * at this point we're using the spare definitions from
1952		 * the MOS config, not necessarily from the userland config.
1953		 */
1954		if (state != SPA_LOAD_IMPORT) {
1955			spa_aux_check_removed(&spa->spa_spares);
1956			spa_aux_check_removed(&spa->spa_l2cache);
1957		}
1958	}
1959
1960	/*
1961	 * Load the vdev state for all toplevel vdevs.
1962	 */
1963	vdev_load(rvd);
1964
1965	/*
1966	 * Propagate the leaf DTLs we just loaded all the way up the tree.
1967	 */
1968	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1969	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
1970	spa_config_exit(spa, SCL_ALL, FTAG);
1971
1972	/*
1973	 * Check the state of the root vdev.  If it can't be opened, it
1974	 * indicates one or more toplevel vdevs are faulted.
1975	 */
1976	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1977		return (ENXIO);
1978
1979	/*
1980	 * Load the DDTs (dedup tables).
1981	 */
1982	error = ddt_load(spa);
1983	if (error != 0)
1984		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1985
1986	spa_update_dspace(spa);
1987
1988	if (state != SPA_LOAD_TRYIMPORT) {
1989		error = spa_load_verify(spa);
1990		if (error)
1991			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
1992			    error));
1993	}
1994
1995	/*
1996	 * Load the intent log state and check log integrity.  If we're
1997	 * assembling a pool from a split, the log is not transferred over.
1998	 */
1999	if (type != SPA_IMPORT_ASSEMBLE) {
2000		VERIFY(nvlist_lookup_nvlist(nvconfig, ZPOOL_CONFIG_VDEV_TREE,
2001		    &nvroot) == 0);
2002		spa_load_log_state(spa, nvroot);
2003		nvlist_free(nvconfig);
2004
2005		if (spa_check_logs(spa)) {
2006			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2007			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2008		}
2009	}
2010
2011	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2012	    spa->spa_load_max_txg == UINT64_MAX)) {
2013		dmu_tx_t *tx;
2014		int need_update = B_FALSE;
2015
2016		ASSERT(state != SPA_LOAD_TRYIMPORT);
2017
2018		/*
2019		 * Claim log blocks that haven't been committed yet.
2020		 * This must all happen in a single txg.
2021		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2022		 * invoked from zil_claim_log_block()'s i/o done callback.
2023		 * Price of rollback is that we abandon the log.
2024		 */
2025		spa->spa_claiming = B_TRUE;
2026
2027		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2028		    spa_first_txg(spa));
2029		(void) dmu_objset_find(spa_name(spa),
2030		    zil_claim, tx, DS_FIND_CHILDREN);
2031		dmu_tx_commit(tx);
2032
2033		spa->spa_claiming = B_FALSE;
2034
2035		spa_set_log_state(spa, SPA_LOG_GOOD);
2036		spa->spa_sync_on = B_TRUE;
2037		txg_sync_start(spa->spa_dsl_pool);
2038
2039		/*
2040		 * Wait for all claims to sync.  We sync up to the highest
2041		 * claimed log block birth time so that claimed log blocks
2042		 * don't appear to be from the future.  spa_claim_max_txg
2043		 * will have been set for us by either zil_check_log_chain()
2044		 * (invoked from spa_check_logs()) or zil_claim() above.
2045		 */
2046		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2047
2048		/*
2049		 * If the config cache is stale, or we have uninitialized
2050		 * metaslabs (see spa_vdev_add()), then update the config.
2051		 *
2052		 * If spa_load_verbatim is true, trust the current
2053		 * in-core spa_config and update the disk labels.
2054		 */
2055		if (config_cache_txg != spa->spa_config_txg ||
2056		    state == SPA_LOAD_IMPORT || spa->spa_load_verbatim ||
2057		    state == SPA_LOAD_RECOVER)
2058			need_update = B_TRUE;
2059
2060		for (int c = 0; c < rvd->vdev_children; c++)
2061			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2062				need_update = B_TRUE;
2063
2064		/*
2065		 * Update the config cache asychronously in case we're the
2066		 * root pool, in which case the config cache isn't writable yet.
2067		 */
2068		if (need_update)
2069			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2070
2071		/*
2072		 * Check all DTLs to see if anything needs resilvering.
2073		 */
2074		if (vdev_resilver_needed(rvd, NULL, NULL))
2075			spa_async_request(spa, SPA_ASYNC_RESILVER);
2076
2077		/*
2078		 * Delete any inconsistent datasets.
2079		 */
2080		(void) dmu_objset_find(spa_name(spa),
2081		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2082
2083		/*
2084		 * Clean up any stale temporary dataset userrefs.
2085		 */
2086		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2087	}
2088
2089	return (0);
2090}
2091
2092static int
2093spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2094{
2095	spa_unload(spa);
2096	spa_deactivate(spa);
2097
2098	spa->spa_load_max_txg--;
2099
2100	spa_activate(spa, spa_mode_global);
2101	spa_async_suspend(spa);
2102
2103	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2104}
2105
2106static int
2107spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2108    uint64_t max_request, int rewind_flags)
2109{
2110	nvlist_t *config = NULL;
2111	int load_error, rewind_error;
2112	uint64_t safe_rewind_txg;
2113	uint64_t min_txg;
2114
2115	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2116		spa->spa_load_max_txg = spa->spa_load_txg;
2117		spa_set_log_state(spa, SPA_LOG_CLEAR);
2118	} else {
2119		spa->spa_load_max_txg = max_request;
2120	}
2121
2122	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2123	    mosconfig);
2124	if (load_error == 0)
2125		return (0);
2126
2127	if (spa->spa_root_vdev != NULL)
2128		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2129
2130	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2131	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2132
2133	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2134		nvlist_free(config);
2135		return (load_error);
2136	}
2137
2138	/* Price of rolling back is discarding txgs, including log */
2139	if (state == SPA_LOAD_RECOVER)
2140		spa_set_log_state(spa, SPA_LOG_CLEAR);
2141
2142	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2143	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2144	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2145	    TXG_INITIAL : safe_rewind_txg;
2146
2147	/*
2148	 * Continue as long as we're finding errors, we're still within
2149	 * the acceptable rewind range, and we're still finding uberblocks
2150	 */
2151	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2152	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2153		if (spa->spa_load_max_txg < safe_rewind_txg)
2154			spa->spa_extreme_rewind = B_TRUE;
2155		rewind_error = spa_load_retry(spa, state, mosconfig);
2156	}
2157
2158	if (config)
2159		spa_rewind_data_to_nvlist(spa, config);
2160
2161	spa->spa_extreme_rewind = B_FALSE;
2162	spa->spa_load_max_txg = UINT64_MAX;
2163
2164	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2165		spa_config_set(spa, config);
2166
2167	return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
2168}
2169
2170/*
2171 * Pool Open/Import
2172 *
2173 * The import case is identical to an open except that the configuration is sent
2174 * down from userland, instead of grabbed from the configuration cache.  For the
2175 * case of an open, the pool configuration will exist in the
2176 * POOL_STATE_UNINITIALIZED state.
2177 *
2178 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2179 * the same time open the pool, without having to keep around the spa_t in some
2180 * ambiguous state.
2181 */
2182static int
2183spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2184    nvlist_t **config)
2185{
2186	spa_t *spa;
2187	zpool_rewind_policy_t policy;
2188	spa_load_state_t state = SPA_LOAD_OPEN;
2189	int error;
2190	int locked = B_FALSE;
2191
2192	*spapp = NULL;
2193
2194	/*
2195	 * As disgusting as this is, we need to support recursive calls to this
2196	 * function because dsl_dir_open() is called during spa_load(), and ends
2197	 * up calling spa_open() again.  The real fix is to figure out how to
2198	 * avoid dsl_dir_open() calling this in the first place.
2199	 */
2200	if (mutex_owner(&spa_namespace_lock) != curthread) {
2201		mutex_enter(&spa_namespace_lock);
2202		locked = B_TRUE;
2203	}
2204
2205	if ((spa = spa_lookup(pool)) == NULL) {
2206		if (locked)
2207			mutex_exit(&spa_namespace_lock);
2208		return (ENOENT);
2209	}
2210
2211	zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, &policy);
2212	if (policy.zrp_request & ZPOOL_DO_REWIND)
2213		state = SPA_LOAD_RECOVER;
2214
2215	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2216
2217		spa_activate(spa, spa_mode_global);
2218
2219		if (spa->spa_last_open_failed && (policy.zrp_request &
2220		    (ZPOOL_NO_REWIND | ZPOOL_NEVER_REWIND))) {
2221			if (config != NULL && spa->spa_config)
2222				VERIFY(nvlist_dup(spa->spa_config,
2223				    config, KM_SLEEP) == 0);
2224			spa_deactivate(spa);
2225			if (locked)
2226				mutex_exit(&spa_namespace_lock);
2227			return (spa->spa_last_open_failed);
2228		}
2229
2230		if (state != SPA_LOAD_RECOVER)
2231			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2232
2233		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2234		    policy.zrp_request);
2235
2236		if (error == EBADF) {
2237			/*
2238			 * If vdev_validate() returns failure (indicated by
2239			 * EBADF), it indicates that one of the vdevs indicates
2240			 * that the pool has been exported or destroyed.  If
2241			 * this is the case, the config cache is out of sync and
2242			 * we should remove the pool from the namespace.
2243			 */
2244			spa_unload(spa);
2245			spa_deactivate(spa);
2246			spa_config_sync(spa, B_TRUE, B_TRUE);
2247			spa_remove(spa);
2248			if (locked)
2249				mutex_exit(&spa_namespace_lock);
2250			return (ENOENT);
2251		}
2252
2253		if (error) {
2254			/*
2255			 * We can't open the pool, but we still have useful
2256			 * information: the state of each vdev after the
2257			 * attempted vdev_open().  Return this to the user.
2258			 */
2259			if (config != NULL && spa->spa_config)
2260				VERIFY(nvlist_dup(spa->spa_config, config,
2261				    KM_SLEEP) == 0);
2262			spa_unload(spa);
2263			spa_deactivate(spa);
2264			spa->spa_last_open_failed = error;
2265			if (locked)
2266				mutex_exit(&spa_namespace_lock);
2267			*spapp = NULL;
2268			return (error);
2269		}
2270
2271	}
2272
2273	spa_open_ref(spa, tag);
2274
2275
2276	if (config != NULL)
2277		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2278
2279	if (locked) {
2280		spa->spa_last_open_failed = 0;
2281		spa->spa_last_ubsync_txg = 0;
2282		spa->spa_load_txg = 0;
2283		mutex_exit(&spa_namespace_lock);
2284	}
2285
2286	*spapp = spa;
2287
2288	return (0);
2289}
2290
2291int
2292spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2293    nvlist_t **config)
2294{
2295	return (spa_open_common(name, spapp, tag, policy, config));
2296}
2297
2298int
2299spa_open(const char *name, spa_t **spapp, void *tag)
2300{
2301	return (spa_open_common(name, spapp, tag, NULL, NULL));
2302}
2303
2304/*
2305 * Lookup the given spa_t, incrementing the inject count in the process,
2306 * preventing it from being exported or destroyed.
2307 */
2308spa_t *
2309spa_inject_addref(char *name)
2310{
2311	spa_t *spa;
2312
2313	mutex_enter(&spa_namespace_lock);
2314	if ((spa = spa_lookup(name)) == NULL) {
2315		mutex_exit(&spa_namespace_lock);
2316		return (NULL);
2317	}
2318	spa->spa_inject_ref++;
2319	mutex_exit(&spa_namespace_lock);
2320
2321	return (spa);
2322}
2323
2324void
2325spa_inject_delref(spa_t *spa)
2326{
2327	mutex_enter(&spa_namespace_lock);
2328	spa->spa_inject_ref--;
2329	mutex_exit(&spa_namespace_lock);
2330}
2331
2332/*
2333 * Add spares device information to the nvlist.
2334 */
2335static void
2336spa_add_spares(spa_t *spa, nvlist_t *config)
2337{
2338	nvlist_t **spares;
2339	uint_t i, nspares;
2340	nvlist_t *nvroot;
2341	uint64_t guid;
2342	vdev_stat_t *vs;
2343	uint_t vsc;
2344	uint64_t pool;
2345
2346	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2347
2348	if (spa->spa_spares.sav_count == 0)
2349		return;
2350
2351	VERIFY(nvlist_lookup_nvlist(config,
2352	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2353	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2354	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2355	if (nspares != 0) {
2356		VERIFY(nvlist_add_nvlist_array(nvroot,
2357		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2358		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2359		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2360
2361		/*
2362		 * Go through and find any spares which have since been
2363		 * repurposed as an active spare.  If this is the case, update
2364		 * their status appropriately.
2365		 */
2366		for (i = 0; i < nspares; i++) {
2367			VERIFY(nvlist_lookup_uint64(spares[i],
2368			    ZPOOL_CONFIG_GUID, &guid) == 0);
2369			if (spa_spare_exists(guid, &pool, NULL) &&
2370			    pool != 0ULL) {
2371				VERIFY(nvlist_lookup_uint64_array(
2372				    spares[i], ZPOOL_CONFIG_STATS,
2373				    (uint64_t **)&vs, &vsc) == 0);
2374				vs->vs_state = VDEV_STATE_CANT_OPEN;
2375				vs->vs_aux = VDEV_AUX_SPARED;
2376			}
2377		}
2378	}
2379}
2380
2381/*
2382 * Add l2cache device information to the nvlist, including vdev stats.
2383 */
2384static void
2385spa_add_l2cache(spa_t *spa, nvlist_t *config)
2386{
2387	nvlist_t **l2cache;
2388	uint_t i, j, nl2cache;
2389	nvlist_t *nvroot;
2390	uint64_t guid;
2391	vdev_t *vd;
2392	vdev_stat_t *vs;
2393	uint_t vsc;
2394
2395	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2396
2397	if (spa->spa_l2cache.sav_count == 0)
2398		return;
2399
2400	VERIFY(nvlist_lookup_nvlist(config,
2401	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2402	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2403	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2404	if (nl2cache != 0) {
2405		VERIFY(nvlist_add_nvlist_array(nvroot,
2406		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2407		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2408		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2409
2410		/*
2411		 * Update level 2 cache device stats.
2412		 */
2413
2414		for (i = 0; i < nl2cache; i++) {
2415			VERIFY(nvlist_lookup_uint64(l2cache[i],
2416			    ZPOOL_CONFIG_GUID, &guid) == 0);
2417
2418			vd = NULL;
2419			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2420				if (guid ==
2421				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2422					vd = spa->spa_l2cache.sav_vdevs[j];
2423					break;
2424				}
2425			}
2426			ASSERT(vd != NULL);
2427
2428			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2429			    ZPOOL_CONFIG_STATS, (uint64_t **)&vs, &vsc) == 0);
2430			vdev_get_stats(vd, vs);
2431		}
2432	}
2433}
2434
2435int
2436spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2437{
2438	int error;
2439	spa_t *spa;
2440
2441	*config = NULL;
2442	error = spa_open_common(name, &spa, FTAG, NULL, config);
2443
2444	if (spa != NULL) {
2445		/*
2446		 * This still leaves a window of inconsistency where the spares
2447		 * or l2cache devices could change and the config would be
2448		 * self-inconsistent.
2449		 */
2450		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2451
2452		if (*config != NULL) {
2453			VERIFY(nvlist_add_uint64(*config,
2454			    ZPOOL_CONFIG_ERRCOUNT,
2455			    spa_get_errlog_size(spa)) == 0);
2456
2457			if (spa_suspended(spa))
2458				VERIFY(nvlist_add_uint64(*config,
2459				    ZPOOL_CONFIG_SUSPENDED,
2460				    spa->spa_failmode) == 0);
2461
2462			spa_add_spares(spa, *config);
2463			spa_add_l2cache(spa, *config);
2464		}
2465	}
2466
2467	/*
2468	 * We want to get the alternate root even for faulted pools, so we cheat
2469	 * and call spa_lookup() directly.
2470	 */
2471	if (altroot) {
2472		if (spa == NULL) {
2473			mutex_enter(&spa_namespace_lock);
2474			spa = spa_lookup(name);
2475			if (spa)
2476				spa_altroot(spa, altroot, buflen);
2477			else
2478				altroot[0] = '\0';
2479			spa = NULL;
2480			mutex_exit(&spa_namespace_lock);
2481		} else {
2482			spa_altroot(spa, altroot, buflen);
2483		}
2484	}
2485
2486	if (spa != NULL) {
2487		spa_config_exit(spa, SCL_CONFIG, FTAG);
2488		spa_close(spa, FTAG);
2489	}
2490
2491	return (error);
2492}
2493
2494/*
2495 * Validate that the auxiliary device array is well formed.  We must have an
2496 * array of nvlists, each which describes a valid leaf vdev.  If this is an
2497 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2498 * specified, as long as they are well-formed.
2499 */
2500static int
2501spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2502    spa_aux_vdev_t *sav, const char *config, uint64_t version,
2503    vdev_labeltype_t label)
2504{
2505	nvlist_t **dev;
2506	uint_t i, ndev;
2507	vdev_t *vd;
2508	int error;
2509
2510	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2511
2512	/*
2513	 * It's acceptable to have no devs specified.
2514	 */
2515	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2516		return (0);
2517
2518	if (ndev == 0)
2519		return (EINVAL);
2520
2521	/*
2522	 * Make sure the pool is formatted with a version that supports this
2523	 * device type.
2524	 */
2525	if (spa_version(spa) < version)
2526		return (ENOTSUP);
2527
2528	/*
2529	 * Set the pending device list so we correctly handle device in-use
2530	 * checking.
2531	 */
2532	sav->sav_pending = dev;
2533	sav->sav_npending = ndev;
2534
2535	for (i = 0; i < ndev; i++) {
2536		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2537		    mode)) != 0)
2538			goto out;
2539
2540		if (!vd->vdev_ops->vdev_op_leaf) {
2541			vdev_free(vd);
2542			error = EINVAL;
2543			goto out;
2544		}
2545
2546		/*
2547		 * The L2ARC currently only supports disk devices in
2548		 * kernel context.  For user-level testing, we allow it.
2549		 */
2550#ifdef _KERNEL
2551		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2552		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2553			error = ENOTBLK;
2554			goto out;
2555		}
2556#endif
2557		vd->vdev_top = vd;
2558
2559		if ((error = vdev_open(vd)) == 0 &&
2560		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
2561			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2562			    vd->vdev_guid) == 0);
2563		}
2564
2565		vdev_free(vd);
2566
2567		if (error &&
2568		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2569			goto out;
2570		else
2571			error = 0;
2572	}
2573
2574out:
2575	sav->sav_pending = NULL;
2576	sav->sav_npending = 0;
2577	return (error);
2578}
2579
2580static int
2581spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2582{
2583	int error;
2584
2585	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2586
2587	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2588	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2589	    VDEV_LABEL_SPARE)) != 0) {
2590		return (error);
2591	}
2592
2593	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2594	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2595	    VDEV_LABEL_L2CACHE));
2596}
2597
2598static void
2599spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2600    const char *config)
2601{
2602	int i;
2603
2604	if (sav->sav_config != NULL) {
2605		nvlist_t **olddevs;
2606		uint_t oldndevs;
2607		nvlist_t **newdevs;
2608
2609		/*
2610		 * Generate new dev list by concatentating with the
2611		 * current dev list.
2612		 */
2613		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2614		    &olddevs, &oldndevs) == 0);
2615
2616		newdevs = kmem_alloc(sizeof (void *) *
2617		    (ndevs + oldndevs), KM_SLEEP);
2618		for (i = 0; i < oldndevs; i++)
2619			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2620			    KM_SLEEP) == 0);
2621		for (i = 0; i < ndevs; i++)
2622			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2623			    KM_SLEEP) == 0);
2624
2625		VERIFY(nvlist_remove(sav->sav_config, config,
2626		    DATA_TYPE_NVLIST_ARRAY) == 0);
2627
2628		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2629		    config, newdevs, ndevs + oldndevs) == 0);
2630		for (i = 0; i < oldndevs + ndevs; i++)
2631			nvlist_free(newdevs[i]);
2632		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2633	} else {
2634		/*
2635		 * Generate a new dev list.
2636		 */
2637		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2638		    KM_SLEEP) == 0);
2639		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2640		    devs, ndevs) == 0);
2641	}
2642}
2643
2644/*
2645 * Stop and drop level 2 ARC devices
2646 */
2647void
2648spa_l2cache_drop(spa_t *spa)
2649{
2650	vdev_t *vd;
2651	int i;
2652	spa_aux_vdev_t *sav = &spa->spa_l2cache;
2653
2654	for (i = 0; i < sav->sav_count; i++) {
2655		uint64_t pool;
2656
2657		vd = sav->sav_vdevs[i];
2658		ASSERT(vd != NULL);
2659
2660		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2661		    pool != 0ULL && l2arc_vdev_present(vd))
2662			l2arc_remove_vdev(vd);
2663		if (vd->vdev_isl2cache)
2664			spa_l2cache_remove(vd);
2665		vdev_clear_stats(vd);
2666		(void) vdev_close(vd);
2667	}
2668}
2669
2670/*
2671 * Pool Creation
2672 */
2673int
2674spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2675    const char *history_str, nvlist_t *zplprops)
2676{
2677	spa_t *spa;
2678	char *altroot = NULL;
2679	vdev_t *rvd;
2680	dsl_pool_t *dp;
2681	dmu_tx_t *tx;
2682	int error = 0;
2683	uint64_t txg = TXG_INITIAL;
2684	nvlist_t **spares, **l2cache;
2685	uint_t nspares, nl2cache;
2686	uint64_t version;
2687
2688	/*
2689	 * If this pool already exists, return failure.
2690	 */
2691	mutex_enter(&spa_namespace_lock);
2692	if (spa_lookup(pool) != NULL) {
2693		mutex_exit(&spa_namespace_lock);
2694		return (EEXIST);
2695	}
2696
2697	/*
2698	 * Allocate a new spa_t structure.
2699	 */
2700	(void) nvlist_lookup_string(props,
2701	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2702	spa = spa_add(pool, NULL, altroot);
2703	spa_activate(spa, spa_mode_global);
2704
2705	if (props && (error = spa_prop_validate(spa, props))) {
2706		spa_deactivate(spa);
2707		spa_remove(spa);
2708		mutex_exit(&spa_namespace_lock);
2709		return (error);
2710	}
2711
2712	if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2713	    &version) != 0)
2714		version = SPA_VERSION;
2715	ASSERT(version <= SPA_VERSION);
2716
2717	spa->spa_first_txg = txg;
2718	spa->spa_uberblock.ub_txg = txg - 1;
2719	spa->spa_uberblock.ub_version = version;
2720	spa->spa_ubsync = spa->spa_uberblock;
2721
2722	/*
2723	 * Create "The Godfather" zio to hold all async IOs
2724	 */
2725	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2726	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2727
2728	/*
2729	 * Create the root vdev.
2730	 */
2731	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2732
2733	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2734
2735	ASSERT(error != 0 || rvd != NULL);
2736	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2737
2738	if (error == 0 && !zfs_allocatable_devs(nvroot))
2739		error = EINVAL;
2740
2741	if (error == 0 &&
2742	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2743	    (error = spa_validate_aux(spa, nvroot, txg,
2744	    VDEV_ALLOC_ADD)) == 0) {
2745		for (int c = 0; c < rvd->vdev_children; c++) {
2746			vdev_metaslab_set_size(rvd->vdev_child[c]);
2747			vdev_expand(rvd->vdev_child[c], txg);
2748		}
2749	}
2750
2751	spa_config_exit(spa, SCL_ALL, FTAG);
2752
2753	if (error != 0) {
2754		spa_unload(spa);
2755		spa_deactivate(spa);
2756		spa_remove(spa);
2757		mutex_exit(&spa_namespace_lock);
2758		return (error);
2759	}
2760
2761	/*
2762	 * Get the list of spares, if specified.
2763	 */
2764	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2765	    &spares, &nspares) == 0) {
2766		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
2767		    KM_SLEEP) == 0);
2768		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2769		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2770		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2771		spa_load_spares(spa);
2772		spa_config_exit(spa, SCL_ALL, FTAG);
2773		spa->spa_spares.sav_sync = B_TRUE;
2774	}
2775
2776	/*
2777	 * Get the list of level 2 cache devices, if specified.
2778	 */
2779	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2780	    &l2cache, &nl2cache) == 0) {
2781		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2782		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2783		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2784		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2785		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2786		spa_load_l2cache(spa);
2787		spa_config_exit(spa, SCL_ALL, FTAG);
2788		spa->spa_l2cache.sav_sync = B_TRUE;
2789	}
2790
2791	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
2792	spa->spa_meta_objset = dp->dp_meta_objset;
2793
2794	/*
2795	 * Create DDTs (dedup tables).
2796	 */
2797	ddt_create(spa);
2798
2799	spa_update_dspace(spa);
2800
2801	tx = dmu_tx_create_assigned(dp, txg);
2802
2803	/*
2804	 * Create the pool config object.
2805	 */
2806	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
2807	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
2808	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2809
2810	if (zap_add(spa->spa_meta_objset,
2811	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
2812	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
2813		cmn_err(CE_PANIC, "failed to add pool config");
2814	}
2815
2816	/* Newly created pools with the right version are always deflated. */
2817	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
2818		spa->spa_deflate = TRUE;
2819		if (zap_add(spa->spa_meta_objset,
2820		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
2821		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
2822			cmn_err(CE_PANIC, "failed to add deflate");
2823		}
2824	}
2825
2826	/*
2827	 * Create the deferred-free bplist object.  Turn off compression
2828	 * because sync-to-convergence takes longer if the blocksize
2829	 * keeps changing.
2830	 */
2831	spa->spa_deferred_bplist_obj = bplist_create(spa->spa_meta_objset,
2832	    1 << 14, tx);
2833	dmu_object_set_compress(spa->spa_meta_objset,
2834	    spa->spa_deferred_bplist_obj, ZIO_COMPRESS_OFF, tx);
2835
2836	if (zap_add(spa->spa_meta_objset,
2837	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
2838	    sizeof (uint64_t), 1, &spa->spa_deferred_bplist_obj, tx) != 0) {
2839		cmn_err(CE_PANIC, "failed to add bplist");
2840	}
2841
2842	/*
2843	 * Create the pool's history object.
2844	 */
2845	if (version >= SPA_VERSION_ZPOOL_HISTORY)
2846		spa_history_create_obj(spa, tx);
2847
2848	/*
2849	 * Set pool properties.
2850	 */
2851	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
2852	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2853	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
2854	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
2855
2856	if (props != NULL) {
2857		spa_configfile_set(spa, props, B_FALSE);
2858		spa_sync_props(spa, props, CRED(), tx);
2859	}
2860
2861	dmu_tx_commit(tx);
2862
2863	spa->spa_sync_on = B_TRUE;
2864	txg_sync_start(spa->spa_dsl_pool);
2865
2866	/*
2867	 * We explicitly wait for the first transaction to complete so that our
2868	 * bean counters are appropriately updated.
2869	 */
2870	txg_wait_synced(spa->spa_dsl_pool, txg);
2871
2872	spa_config_sync(spa, B_FALSE, B_TRUE);
2873
2874	if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
2875		(void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
2876	spa_history_log_version(spa, LOG_POOL_CREATE);
2877
2878	spa->spa_minref = refcount_count(&spa->spa_refcount);
2879
2880	mutex_exit(&spa_namespace_lock);
2881
2882	return (0);
2883}
2884
2885#ifdef _KERNEL
2886/*
2887 * Get the root pool information from the root disk, then import the root pool
2888 * during the system boot up time.
2889 */
2890extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
2891
2892static nvlist_t *
2893spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
2894{
2895	nvlist_t *config;
2896	nvlist_t *nvtop, *nvroot;
2897	uint64_t pgid;
2898
2899	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
2900		return (NULL);
2901
2902	/*
2903	 * Add this top-level vdev to the child array.
2904	 */
2905	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2906	    &nvtop) == 0);
2907	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
2908	    &pgid) == 0);
2909	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
2910
2911	/*
2912	 * Put this pool's top-level vdevs into a root vdev.
2913	 */
2914	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2915	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
2916	    VDEV_TYPE_ROOT) == 0);
2917	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
2918	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
2919	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2920	    &nvtop, 1) == 0);
2921
2922	/*
2923	 * Replace the existing vdev_tree with the new root vdev in
2924	 * this pool's configuration (remove the old, add the new).
2925	 */
2926	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
2927	nvlist_free(nvroot);
2928	return (config);
2929}
2930
2931/*
2932 * Walk the vdev tree and see if we can find a device with "better"
2933 * configuration. A configuration is "better" if the label on that
2934 * device has a more recent txg.
2935 */
2936static void
2937spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
2938{
2939	for (int c = 0; c < vd->vdev_children; c++)
2940		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
2941
2942	if (vd->vdev_ops->vdev_op_leaf) {
2943		nvlist_t *label;
2944		uint64_t label_txg;
2945
2946		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
2947		    &label) != 0)
2948			return;
2949
2950		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
2951		    &label_txg) == 0);
2952
2953		/*
2954		 * Do we have a better boot device?
2955		 */
2956		if (label_txg > *txg) {
2957			*txg = label_txg;
2958			*avd = vd;
2959		}
2960		nvlist_free(label);
2961	}
2962}
2963
2964/*
2965 * Import a root pool.
2966 *
2967 * For x86. devpath_list will consist of devid and/or physpath name of
2968 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
2969 * The GRUB "findroot" command will return the vdev we should boot.
2970 *
2971 * For Sparc, devpath_list consists the physpath name of the booting device
2972 * no matter the rootpool is a single device pool or a mirrored pool.
2973 * e.g.
2974 *	"/pci@1f,0/ide@d/disk@0,0:a"
2975 */
2976int
2977spa_import_rootpool(char *devpath, char *devid)
2978{
2979	spa_t *spa;
2980	vdev_t *rvd, *bvd, *avd = NULL;
2981	nvlist_t *config, *nvtop;
2982	uint64_t guid, txg;
2983	char *pname;
2984	int error;
2985
2986	/*
2987	 * Read the label from the boot device and generate a configuration.
2988	 */
2989	config = spa_generate_rootconf(devpath, devid, &guid);
2990#if defined(_OBP) && defined(_KERNEL)
2991	if (config == NULL) {
2992		if (strstr(devpath, "/iscsi/ssd") != NULL) {
2993			/* iscsi boot */
2994			get_iscsi_bootpath_phy(devpath);
2995			config = spa_generate_rootconf(devpath, devid, &guid);
2996		}
2997	}
2998#endif
2999	if (config == NULL) {
3000		cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
3001		    devpath);
3002		return (EIO);
3003	}
3004
3005	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3006	    &pname) == 0);
3007	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3008
3009	mutex_enter(&spa_namespace_lock);
3010	if ((spa = spa_lookup(pname)) != NULL) {
3011		/*
3012		 * Remove the existing root pool from the namespace so that we
3013		 * can replace it with the correct config we just read in.
3014		 */
3015		spa_remove(spa);
3016	}
3017
3018	spa = spa_add(pname, config, NULL);
3019	spa->spa_is_root = B_TRUE;
3020	spa->spa_load_verbatim = B_TRUE;
3021
3022	/*
3023	 * Build up a vdev tree based on the boot device's label config.
3024	 */
3025	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3026	    &nvtop) == 0);
3027	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3028	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3029	    VDEV_ALLOC_ROOTPOOL);
3030	spa_config_exit(spa, SCL_ALL, FTAG);
3031	if (error) {
3032		mutex_exit(&spa_namespace_lock);
3033		nvlist_free(config);
3034		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3035		    pname);
3036		return (error);
3037	}
3038
3039	/*
3040	 * Get the boot vdev.
3041	 */
3042	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3043		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3044		    (u_longlong_t)guid);
3045		error = ENOENT;
3046		goto out;
3047	}
3048
3049	/*
3050	 * Determine if there is a better boot device.
3051	 */
3052	avd = bvd;
3053	spa_alt_rootvdev(rvd, &avd, &txg);
3054	if (avd != bvd) {
3055		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3056		    "try booting from '%s'", avd->vdev_path);
3057		error = EINVAL;
3058		goto out;
3059	}
3060
3061	/*
3062	 * If the boot device is part of a spare vdev then ensure that
3063	 * we're booting off the active spare.
3064	 */
3065	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3066	    !bvd->vdev_isspare) {
3067		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3068		    "try booting from '%s'",
3069		    bvd->vdev_parent->vdev_child[1]->vdev_path);
3070		error = EINVAL;
3071		goto out;
3072	}
3073
3074	error = 0;
3075	spa_history_log_version(spa, LOG_POOL_IMPORT);
3076out:
3077	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3078	vdev_free(rvd);
3079	spa_config_exit(spa, SCL_ALL, FTAG);
3080	mutex_exit(&spa_namespace_lock);
3081
3082	nvlist_free(config);
3083	return (error);
3084}
3085
3086#endif
3087
3088/*
3089 * Take a pool and insert it into the namespace as if it had been loaded at
3090 * boot.
3091 */
3092int
3093spa_import_verbatim(const char *pool, nvlist_t *config, nvlist_t *props)
3094{
3095	spa_t *spa;
3096	char *altroot = NULL;
3097
3098	mutex_enter(&spa_namespace_lock);
3099	if (spa_lookup(pool) != NULL) {
3100		mutex_exit(&spa_namespace_lock);
3101		return (EEXIST);
3102	}
3103
3104	(void) nvlist_lookup_string(props,
3105	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3106	spa = spa_add(pool, config, altroot);
3107
3108	spa->spa_load_verbatim = B_TRUE;
3109
3110	if (props != NULL)
3111		spa_configfile_set(spa, props, B_FALSE);
3112
3113	spa_config_sync(spa, B_FALSE, B_TRUE);
3114
3115	mutex_exit(&spa_namespace_lock);
3116	spa_history_log_version(spa, LOG_POOL_IMPORT);
3117
3118	return (0);
3119}
3120
3121/*
3122 * Import a non-root pool into the system.
3123 */
3124int
3125spa_import(const char *pool, nvlist_t *config, nvlist_t *props)
3126{
3127	spa_t *spa;
3128	char *altroot = NULL;
3129	spa_load_state_t state = SPA_LOAD_IMPORT;
3130	zpool_rewind_policy_t policy;
3131	int error;
3132	nvlist_t *nvroot;
3133	nvlist_t **spares, **l2cache;
3134	uint_t nspares, nl2cache;
3135
3136	/*
3137	 * If a pool with this name exists, return failure.
3138	 */
3139	mutex_enter(&spa_namespace_lock);
3140	if (spa_lookup(pool) != NULL) {
3141		mutex_exit(&spa_namespace_lock);
3142		return (EEXIST);
3143	}
3144
3145	zpool_get_rewind_policy(config, &policy);
3146	if (policy.zrp_request & ZPOOL_DO_REWIND)
3147		state = SPA_LOAD_RECOVER;
3148
3149	/*
3150	 * Create and initialize the spa structure.
3151	 */
3152	(void) nvlist_lookup_string(props,
3153	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3154	spa = spa_add(pool, config, altroot);
3155	spa_activate(spa, spa_mode_global);
3156
3157	/*
3158	 * Don't start async tasks until we know everything is healthy.
3159	 */
3160	spa_async_suspend(spa);
3161
3162	/*
3163	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3164	 * because the user-supplied config is actually the one to trust when
3165	 * doing an import.
3166	 */
3167	if (state != SPA_LOAD_RECOVER)
3168		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3169	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3170	    policy.zrp_request);
3171
3172	/*
3173	 * Propagate anything learned about failing or best txgs
3174	 * back to caller
3175	 */
3176	spa_rewind_data_to_nvlist(spa, config);
3177
3178	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3179	/*
3180	 * Toss any existing sparelist, as it doesn't have any validity
3181	 * anymore, and conflicts with spa_has_spare().
3182	 */
3183	if (spa->spa_spares.sav_config) {
3184		nvlist_free(spa->spa_spares.sav_config);
3185		spa->spa_spares.sav_config = NULL;
3186		spa_load_spares(spa);
3187	}
3188	if (spa->spa_l2cache.sav_config) {
3189		nvlist_free(spa->spa_l2cache.sav_config);
3190		spa->spa_l2cache.sav_config = NULL;
3191		spa_load_l2cache(spa);
3192	}
3193
3194	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3195	    &nvroot) == 0);
3196	if (error == 0)
3197		error = spa_validate_aux(spa, nvroot, -1ULL,
3198		    VDEV_ALLOC_SPARE);
3199	if (error == 0)
3200		error = spa_validate_aux(spa, nvroot, -1ULL,
3201		    VDEV_ALLOC_L2CACHE);
3202	spa_config_exit(spa, SCL_ALL, FTAG);
3203
3204	if (props != NULL)
3205		spa_configfile_set(spa, props, B_FALSE);
3206
3207	if (error != 0 || (props && spa_writeable(spa) &&
3208	    (error = spa_prop_set(spa, props)))) {
3209		spa_unload(spa);
3210		spa_deactivate(spa);
3211		spa_remove(spa);
3212		mutex_exit(&spa_namespace_lock);
3213		return (error);
3214	}
3215
3216	spa_async_resume(spa);
3217
3218	/*
3219	 * Override any spares and level 2 cache devices as specified by
3220	 * the user, as these may have correct device names/devids, etc.
3221	 */
3222	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3223	    &spares, &nspares) == 0) {
3224		if (spa->spa_spares.sav_config)
3225			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3226			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3227		else
3228			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3229			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3230		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3231		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3232		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3233		spa_load_spares(spa);
3234		spa_config_exit(spa, SCL_ALL, FTAG);
3235		spa->spa_spares.sav_sync = B_TRUE;
3236	}
3237	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3238	    &l2cache, &nl2cache) == 0) {
3239		if (spa->spa_l2cache.sav_config)
3240			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3241			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3242		else
3243			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3244			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3245		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3246		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3247		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3248		spa_load_l2cache(spa);
3249		spa_config_exit(spa, SCL_ALL, FTAG);
3250		spa->spa_l2cache.sav_sync = B_TRUE;
3251	}
3252
3253	/*
3254	 * Check for any removed devices.
3255	 */
3256	if (spa->spa_autoreplace) {
3257		spa_aux_check_removed(&spa->spa_spares);
3258		spa_aux_check_removed(&spa->spa_l2cache);
3259	}
3260
3261	if (spa_writeable(spa)) {
3262		/*
3263		 * Update the config cache to include the newly-imported pool.
3264		 */
3265		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3266	}
3267
3268	/*
3269	 * It's possible that the pool was expanded while it was exported.
3270	 * We kick off an async task to handle this for us.
3271	 */
3272	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3273
3274	mutex_exit(&spa_namespace_lock);
3275	spa_history_log_version(spa, LOG_POOL_IMPORT);
3276
3277	return (0);
3278}
3279
3280nvlist_t *
3281spa_tryimport(nvlist_t *tryconfig)
3282{
3283	nvlist_t *config = NULL;
3284	char *poolname;
3285	spa_t *spa;
3286	uint64_t state;
3287	int error;
3288
3289	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3290		return (NULL);
3291
3292	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3293		return (NULL);
3294
3295	/*
3296	 * Create and initialize the spa structure.
3297	 */
3298	mutex_enter(&spa_namespace_lock);
3299	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3300	spa_activate(spa, FREAD);
3301
3302	/*
3303	 * Pass off the heavy lifting to spa_load().
3304	 * Pass TRUE for mosconfig because the user-supplied config
3305	 * is actually the one to trust when doing an import.
3306	 */
3307	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3308
3309	/*
3310	 * If 'tryconfig' was at least parsable, return the current config.
3311	 */
3312	if (spa->spa_root_vdev != NULL) {
3313		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3314		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3315		    poolname) == 0);
3316		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3317		    state) == 0);
3318		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3319		    spa->spa_uberblock.ub_timestamp) == 0);
3320
3321		/*
3322		 * If the bootfs property exists on this pool then we
3323		 * copy it out so that external consumers can tell which
3324		 * pools are bootable.
3325		 */
3326		if ((!error || error == EEXIST) && spa->spa_bootfs) {
3327			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3328
3329			/*
3330			 * We have to play games with the name since the
3331			 * pool was opened as TRYIMPORT_NAME.
3332			 */
3333			if (dsl_dsobj_to_dsname(spa_name(spa),
3334			    spa->spa_bootfs, tmpname) == 0) {
3335				char *cp;
3336				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3337
3338				cp = strchr(tmpname, '/');
3339				if (cp == NULL) {
3340					(void) strlcpy(dsname, tmpname,
3341					    MAXPATHLEN);
3342				} else {
3343					(void) snprintf(dsname, MAXPATHLEN,
3344					    "%s/%s", poolname, ++cp);
3345				}
3346				VERIFY(nvlist_add_string(config,
3347				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3348				kmem_free(dsname, MAXPATHLEN);
3349			}
3350			kmem_free(tmpname, MAXPATHLEN);
3351		}
3352
3353		/*
3354		 * Add the list of hot spares and level 2 cache devices.
3355		 */
3356		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3357		spa_add_spares(spa, config);
3358		spa_add_l2cache(spa, config);
3359		spa_config_exit(spa, SCL_CONFIG, FTAG);
3360	}
3361
3362	spa_unload(spa);
3363	spa_deactivate(spa);
3364	spa_remove(spa);
3365	mutex_exit(&spa_namespace_lock);
3366
3367	return (config);
3368}
3369
3370/*
3371 * Pool export/destroy
3372 *
3373 * The act of destroying or exporting a pool is very simple.  We make sure there
3374 * is no more pending I/O and any references to the pool are gone.  Then, we
3375 * update the pool state and sync all the labels to disk, removing the
3376 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3377 * we don't sync the labels or remove the configuration cache.
3378 */
3379static int
3380spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3381    boolean_t force, boolean_t hardforce)
3382{
3383	spa_t *spa;
3384
3385	if (oldconfig)
3386		*oldconfig = NULL;
3387
3388	if (!(spa_mode_global & FWRITE))
3389		return (EROFS);
3390
3391	mutex_enter(&spa_namespace_lock);
3392	if ((spa = spa_lookup(pool)) == NULL) {
3393		mutex_exit(&spa_namespace_lock);
3394		return (ENOENT);
3395	}
3396
3397	/*
3398	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3399	 * reacquire the namespace lock, and see if we can export.
3400	 */
3401	spa_open_ref(spa, FTAG);
3402	mutex_exit(&spa_namespace_lock);
3403	spa_async_suspend(spa);
3404	mutex_enter(&spa_namespace_lock);
3405	spa_close(spa, FTAG);
3406
3407	/*
3408	 * The pool will be in core if it's openable,
3409	 * in which case we can modify its state.
3410	 */
3411	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3412		/*
3413		 * Objsets may be open only because they're dirty, so we
3414		 * have to force it to sync before checking spa_refcnt.
3415		 */
3416		txg_wait_synced(spa->spa_dsl_pool, 0);
3417
3418		/*
3419		 * A pool cannot be exported or destroyed if there are active
3420		 * references.  If we are resetting a pool, allow references by
3421		 * fault injection handlers.
3422		 */
3423		if (!spa_refcount_zero(spa) ||
3424		    (spa->spa_inject_ref != 0 &&
3425		    new_state != POOL_STATE_UNINITIALIZED)) {
3426			spa_async_resume(spa);
3427			mutex_exit(&spa_namespace_lock);
3428			return (EBUSY);
3429		}
3430
3431		/*
3432		 * A pool cannot be exported if it has an active shared spare.
3433		 * This is to prevent other pools stealing the active spare
3434		 * from an exported pool. At user's own will, such pool can
3435		 * be forcedly exported.
3436		 */
3437		if (!force && new_state == POOL_STATE_EXPORTED &&
3438		    spa_has_active_shared_spare(spa)) {
3439			spa_async_resume(spa);
3440			mutex_exit(&spa_namespace_lock);
3441			return (EXDEV);
3442		}
3443
3444		/*
3445		 * We want this to be reflected on every label,
3446		 * so mark them all dirty.  spa_unload() will do the
3447		 * final sync that pushes these changes out.
3448		 */
3449		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3450			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3451			spa->spa_state = new_state;
3452			spa->spa_final_txg = spa_last_synced_txg(spa) + 1;
3453			vdev_config_dirty(spa->spa_root_vdev);
3454			spa_config_exit(spa, SCL_ALL, FTAG);
3455		}
3456	}
3457
3458	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
3459
3460	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3461		spa_unload(spa);
3462		spa_deactivate(spa);
3463	}
3464
3465	if (oldconfig && spa->spa_config)
3466		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3467
3468	if (new_state != POOL_STATE_UNINITIALIZED) {
3469		if (!hardforce)
3470			spa_config_sync(spa, B_TRUE, B_TRUE);
3471		spa_remove(spa);
3472	}
3473	mutex_exit(&spa_namespace_lock);
3474
3475	return (0);
3476}
3477
3478/*
3479 * Destroy a storage pool.
3480 */
3481int
3482spa_destroy(char *pool)
3483{
3484	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3485	    B_FALSE, B_FALSE));
3486}
3487
3488/*
3489 * Export a storage pool.
3490 */
3491int
3492spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3493    boolean_t hardforce)
3494{
3495	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3496	    force, hardforce));
3497}
3498
3499/*
3500 * Similar to spa_export(), this unloads the spa_t without actually removing it
3501 * from the namespace in any way.
3502 */
3503int
3504spa_reset(char *pool)
3505{
3506	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3507	    B_FALSE, B_FALSE));
3508}
3509
3510/*
3511 * ==========================================================================
3512 * Device manipulation
3513 * ==========================================================================
3514 */
3515
3516/*
3517 * Add a device to a storage pool.
3518 */
3519int
3520spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3521{
3522	uint64_t txg, id;
3523	int error;
3524	vdev_t *rvd = spa->spa_root_vdev;
3525	vdev_t *vd, *tvd;
3526	nvlist_t **spares, **l2cache;
3527	uint_t nspares, nl2cache;
3528
3529	txg = spa_vdev_enter(spa);
3530
3531	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3532	    VDEV_ALLOC_ADD)) != 0)
3533		return (spa_vdev_exit(spa, NULL, txg, error));
3534
3535	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
3536
3537	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3538	    &nspares) != 0)
3539		nspares = 0;
3540
3541	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3542	    &nl2cache) != 0)
3543		nl2cache = 0;
3544
3545	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3546		return (spa_vdev_exit(spa, vd, txg, EINVAL));
3547
3548	if (vd->vdev_children != 0 &&
3549	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
3550		return (spa_vdev_exit(spa, vd, txg, error));
3551
3552	/*
3553	 * We must validate the spares and l2cache devices after checking the
3554	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
3555	 */
3556	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3557		return (spa_vdev_exit(spa, vd, txg, error));
3558
3559	/*
3560	 * Transfer each new top-level vdev from vd to rvd.
3561	 */
3562	for (int c = 0; c < vd->vdev_children; c++) {
3563
3564		/*
3565		 * Set the vdev id to the first hole, if one exists.
3566		 */
3567		for (id = 0; id < rvd->vdev_children; id++) {
3568			if (rvd->vdev_child[id]->vdev_ishole) {
3569				vdev_free(rvd->vdev_child[id]);
3570				break;
3571			}
3572		}
3573		tvd = vd->vdev_child[c];
3574		vdev_remove_child(vd, tvd);
3575		tvd->vdev_id = id;
3576		vdev_add_child(rvd, tvd);
3577		vdev_config_dirty(tvd);
3578	}
3579
3580	if (nspares != 0) {
3581		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3582		    ZPOOL_CONFIG_SPARES);
3583		spa_load_spares(spa);
3584		spa->spa_spares.sav_sync = B_TRUE;
3585	}
3586
3587	if (nl2cache != 0) {
3588		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3589		    ZPOOL_CONFIG_L2CACHE);
3590		spa_load_l2cache(spa);
3591		spa->spa_l2cache.sav_sync = B_TRUE;
3592	}
3593
3594	/*
3595	 * We have to be careful when adding new vdevs to an existing pool.
3596	 * If other threads start allocating from these vdevs before we
3597	 * sync the config cache, and we lose power, then upon reboot we may
3598	 * fail to open the pool because there are DVAs that the config cache
3599	 * can't translate.  Therefore, we first add the vdevs without
3600	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3601	 * and then let spa_config_update() initialize the new metaslabs.
3602	 *
3603	 * spa_load() checks for added-but-not-initialized vdevs, so that
3604	 * if we lose power at any point in this sequence, the remaining
3605	 * steps will be completed the next time we load the pool.
3606	 */
3607	(void) spa_vdev_exit(spa, vd, txg, 0);
3608
3609	mutex_enter(&spa_namespace_lock);
3610	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3611	mutex_exit(&spa_namespace_lock);
3612
3613	return (0);
3614}
3615
3616/*
3617 * Attach a device to a mirror.  The arguments are the path to any device
3618 * in the mirror, and the nvroot for the new device.  If the path specifies
3619 * a device that is not mirrored, we automatically insert the mirror vdev.
3620 *
3621 * If 'replacing' is specified, the new device is intended to replace the
3622 * existing device; in this case the two devices are made into their own
3623 * mirror using the 'replacing' vdev, which is functionally identical to
3624 * the mirror vdev (it actually reuses all the same ops) but has a few
3625 * extra rules: you can't attach to it after it's been created, and upon
3626 * completion of resilvering, the first disk (the one being replaced)
3627 * is automatically detached.
3628 */
3629int
3630spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3631{
3632	uint64_t txg, open_txg;
3633	vdev_t *rvd = spa->spa_root_vdev;
3634	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3635	vdev_ops_t *pvops;
3636	char *oldvdpath, *newvdpath;
3637	int newvd_isspare;
3638	int error;
3639
3640	txg = spa_vdev_enter(spa);
3641
3642	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3643
3644	if (oldvd == NULL)
3645		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3646
3647	if (!oldvd->vdev_ops->vdev_op_leaf)
3648		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3649
3650	pvd = oldvd->vdev_parent;
3651
3652	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3653	    VDEV_ALLOC_ADD)) != 0)
3654		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3655
3656	if (newrootvd->vdev_children != 1)
3657		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3658
3659	newvd = newrootvd->vdev_child[0];
3660
3661	if (!newvd->vdev_ops->vdev_op_leaf)
3662		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3663
3664	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3665		return (spa_vdev_exit(spa, newrootvd, txg, error));
3666
3667	/*
3668	 * Spares can't replace logs
3669	 */
3670	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3671		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3672
3673	if (!replacing) {
3674		/*
3675		 * For attach, the only allowable parent is a mirror or the root
3676		 * vdev.
3677		 */
3678		if (pvd->vdev_ops != &vdev_mirror_ops &&
3679		    pvd->vdev_ops != &vdev_root_ops)
3680			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3681
3682		pvops = &vdev_mirror_ops;
3683	} else {
3684		/*
3685		 * Active hot spares can only be replaced by inactive hot
3686		 * spares.
3687		 */
3688		if (pvd->vdev_ops == &vdev_spare_ops &&
3689		    pvd->vdev_child[1] == oldvd &&
3690		    !spa_has_spare(spa, newvd->vdev_guid))
3691			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3692
3693		/*
3694		 * If the source is a hot spare, and the parent isn't already a
3695		 * spare, then we want to create a new hot spare.  Otherwise, we
3696		 * want to create a replacing vdev.  The user is not allowed to
3697		 * attach to a spared vdev child unless the 'isspare' state is
3698		 * the same (spare replaces spare, non-spare replaces
3699		 * non-spare).
3700		 */
3701		if (pvd->vdev_ops == &vdev_replacing_ops)
3702			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3703		else if (pvd->vdev_ops == &vdev_spare_ops &&
3704		    newvd->vdev_isspare != oldvd->vdev_isspare)
3705			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3706		else if (pvd->vdev_ops != &vdev_spare_ops &&
3707		    newvd->vdev_isspare)
3708			pvops = &vdev_spare_ops;
3709		else
3710			pvops = &vdev_replacing_ops;
3711	}
3712
3713	/*
3714	 * Make sure the new device is big enough.
3715	 */
3716	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3717		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3718
3719	/*
3720	 * The new device cannot have a higher alignment requirement
3721	 * than the top-level vdev.
3722	 */
3723	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3724		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3725
3726	/*
3727	 * If this is an in-place replacement, update oldvd's path and devid
3728	 * to make it distinguishable from newvd, and unopenable from now on.
3729	 */
3730	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3731		spa_strfree(oldvd->vdev_path);
3732		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3733		    KM_SLEEP);
3734		(void) sprintf(oldvd->vdev_path, "%s/%s",
3735		    newvd->vdev_path, "old");
3736		if (oldvd->vdev_devid != NULL) {
3737			spa_strfree(oldvd->vdev_devid);
3738			oldvd->vdev_devid = NULL;
3739		}
3740	}
3741
3742	/*
3743	 * If the parent is not a mirror, or if we're replacing, insert the new
3744	 * mirror/replacing/spare vdev above oldvd.
3745	 */
3746	if (pvd->vdev_ops != pvops)
3747		pvd = vdev_add_parent(oldvd, pvops);
3748
3749	ASSERT(pvd->vdev_top->vdev_parent == rvd);
3750	ASSERT(pvd->vdev_ops == pvops);
3751	ASSERT(oldvd->vdev_parent == pvd);
3752
3753	/*
3754	 * Extract the new device from its root and add it to pvd.
3755	 */
3756	vdev_remove_child(newrootvd, newvd);
3757	newvd->vdev_id = pvd->vdev_children;
3758	newvd->vdev_crtxg = oldvd->vdev_crtxg;
3759	vdev_add_child(pvd, newvd);
3760
3761	tvd = newvd->vdev_top;
3762	ASSERT(pvd->vdev_top == tvd);
3763	ASSERT(tvd->vdev_parent == rvd);
3764
3765	vdev_config_dirty(tvd);
3766
3767	/*
3768	 * Set newvd's DTL to [TXG_INITIAL, open_txg].  It will propagate
3769	 * upward when spa_vdev_exit() calls vdev_dtl_reassess().
3770	 */
3771	open_txg = txg + TXG_CONCURRENT_STATES - 1;
3772
3773	vdev_dtl_dirty(newvd, DTL_MISSING,
3774	    TXG_INITIAL, open_txg - TXG_INITIAL + 1);
3775
3776	if (newvd->vdev_isspare) {
3777		spa_spare_activate(newvd);
3778		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
3779	}
3780
3781	oldvdpath = spa_strdup(oldvd->vdev_path);
3782	newvdpath = spa_strdup(newvd->vdev_path);
3783	newvd_isspare = newvd->vdev_isspare;
3784
3785	/*
3786	 * Mark newvd's DTL dirty in this txg.
3787	 */
3788	vdev_dirty(tvd, VDD_DTL, newvd, txg);
3789
3790	(void) spa_vdev_exit(spa, newrootvd, open_txg, 0);
3791
3792	spa_history_internal_log(LOG_POOL_VDEV_ATTACH, spa, NULL,
3793	    CRED(),  "%s vdev=%s %s vdev=%s",
3794	    replacing && newvd_isspare ? "spare in" :
3795	    replacing ? "replace" : "attach", newvdpath,
3796	    replacing ? "for" : "to", oldvdpath);
3797
3798	spa_strfree(oldvdpath);
3799	spa_strfree(newvdpath);
3800
3801	/*
3802	 * Kick off a resilver to update newvd.
3803	 */
3804	VERIFY3U(spa_scrub(spa, POOL_SCRUB_RESILVER), ==, 0);
3805
3806	return (0);
3807}
3808
3809/*
3810 * Detach a device from a mirror or replacing vdev.
3811 * If 'replace_done' is specified, only detach if the parent
3812 * is a replacing vdev.
3813 */
3814int
3815spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
3816{
3817	uint64_t txg;
3818	int error;
3819	vdev_t *rvd = spa->spa_root_vdev;
3820	vdev_t *vd, *pvd, *cvd, *tvd;
3821	boolean_t unspare = B_FALSE;
3822	uint64_t unspare_guid;
3823	size_t len;
3824	char *vdpath;
3825
3826	txg = spa_vdev_enter(spa);
3827
3828	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3829
3830	if (vd == NULL)
3831		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3832
3833	if (!vd->vdev_ops->vdev_op_leaf)
3834		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3835
3836	pvd = vd->vdev_parent;
3837
3838	/*
3839	 * If the parent/child relationship is not as expected, don't do it.
3840	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
3841	 * vdev that's replacing B with C.  The user's intent in replacing
3842	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
3843	 * the replace by detaching C, the expected behavior is to end up
3844	 * M(A,B).  But suppose that right after deciding to detach C,
3845	 * the replacement of B completes.  We would have M(A,C), and then
3846	 * ask to detach C, which would leave us with just A -- not what
3847	 * the user wanted.  To prevent this, we make sure that the
3848	 * parent/child relationship hasn't changed -- in this example,
3849	 * that C's parent is still the replacing vdev R.
3850	 */
3851	if (pvd->vdev_guid != pguid && pguid != 0)
3852		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3853
3854	/*
3855	 * If replace_done is specified, only remove this device if it's
3856	 * the first child of a replacing vdev.  For the 'spare' vdev, either
3857	 * disk can be removed.
3858	 */
3859	if (replace_done) {
3860		if (pvd->vdev_ops == &vdev_replacing_ops) {
3861			if (vd->vdev_id != 0)
3862				return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3863		} else if (pvd->vdev_ops != &vdev_spare_ops) {
3864			return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3865		}
3866	}
3867
3868	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
3869	    spa_version(spa) >= SPA_VERSION_SPARES);
3870
3871	/*
3872	 * Only mirror, replacing, and spare vdevs support detach.
3873	 */
3874	if (pvd->vdev_ops != &vdev_replacing_ops &&
3875	    pvd->vdev_ops != &vdev_mirror_ops &&
3876	    pvd->vdev_ops != &vdev_spare_ops)
3877		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3878
3879	/*
3880	 * If this device has the only valid copy of some data,
3881	 * we cannot safely detach it.
3882	 */
3883	if (vdev_dtl_required(vd))
3884		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3885
3886	ASSERT(pvd->vdev_children >= 2);
3887
3888	/*
3889	 * If we are detaching the second disk from a replacing vdev, then
3890	 * check to see if we changed the original vdev's path to have "/old"
3891	 * at the end in spa_vdev_attach().  If so, undo that change now.
3892	 */
3893	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id == 1 &&
3894	    pvd->vdev_child[0]->vdev_path != NULL &&
3895	    pvd->vdev_child[1]->vdev_path != NULL) {
3896		ASSERT(pvd->vdev_child[1] == vd);
3897		cvd = pvd->vdev_child[0];
3898		len = strlen(vd->vdev_path);
3899		if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
3900		    strcmp(cvd->vdev_path + len, "/old") == 0) {
3901			spa_strfree(cvd->vdev_path);
3902			cvd->vdev_path = spa_strdup(vd->vdev_path);
3903		}
3904	}
3905
3906	/*
3907	 * If we are detaching the original disk from a spare, then it implies
3908	 * that the spare should become a real disk, and be removed from the
3909	 * active spare list for the pool.
3910	 */
3911	if (pvd->vdev_ops == &vdev_spare_ops &&
3912	    vd->vdev_id == 0 && pvd->vdev_child[1]->vdev_isspare)
3913		unspare = B_TRUE;
3914
3915	/*
3916	 * Erase the disk labels so the disk can be used for other things.
3917	 * This must be done after all other error cases are handled,
3918	 * but before we disembowel vd (so we can still do I/O to it).
3919	 * But if we can't do it, don't treat the error as fatal --
3920	 * it may be that the unwritability of the disk is the reason
3921	 * it's being detached!
3922	 */
3923	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
3924
3925	/*
3926	 * Remove vd from its parent and compact the parent's children.
3927	 */
3928	vdev_remove_child(pvd, vd);
3929	vdev_compact_children(pvd);
3930
3931	/*
3932	 * Remember one of the remaining children so we can get tvd below.
3933	 */
3934	cvd = pvd->vdev_child[0];
3935
3936	/*
3937	 * If we need to remove the remaining child from the list of hot spares,
3938	 * do it now, marking the vdev as no longer a spare in the process.
3939	 * We must do this before vdev_remove_parent(), because that can
3940	 * change the GUID if it creates a new toplevel GUID.  For a similar
3941	 * reason, we must remove the spare now, in the same txg as the detach;
3942	 * otherwise someone could attach a new sibling, change the GUID, and
3943	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
3944	 */
3945	if (unspare) {
3946		ASSERT(cvd->vdev_isspare);
3947		spa_spare_remove(cvd);
3948		unspare_guid = cvd->vdev_guid;
3949		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
3950	}
3951
3952	/*
3953	 * If the parent mirror/replacing vdev only has one child,
3954	 * the parent is no longer needed.  Remove it from the tree.
3955	 */
3956	if (pvd->vdev_children == 1)
3957		vdev_remove_parent(cvd);
3958
3959	/*
3960	 * We don't set tvd until now because the parent we just removed
3961	 * may have been the previous top-level vdev.
3962	 */
3963	tvd = cvd->vdev_top;
3964	ASSERT(tvd->vdev_parent == rvd);
3965
3966	/*
3967	 * Reevaluate the parent vdev state.
3968	 */
3969	vdev_propagate_state(cvd);
3970
3971	/*
3972	 * If the 'autoexpand' property is set on the pool then automatically
3973	 * try to expand the size of the pool. For example if the device we
3974	 * just detached was smaller than the others, it may be possible to
3975	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
3976	 * first so that we can obtain the updated sizes of the leaf vdevs.
3977	 */
3978	if (spa->spa_autoexpand) {
3979		vdev_reopen(tvd);
3980		vdev_expand(tvd, txg);
3981	}
3982
3983	vdev_config_dirty(tvd);
3984
3985	/*
3986	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
3987	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
3988	 * But first make sure we're not on any *other* txg's DTL list, to
3989	 * prevent vd from being accessed after it's freed.
3990	 */
3991	vdpath = spa_strdup(vd->vdev_path);
3992	for (int t = 0; t < TXG_SIZE; t++)
3993		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
3994	vd->vdev_detached = B_TRUE;
3995	vdev_dirty(tvd, VDD_DTL, vd, txg);
3996
3997	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
3998
3999	error = spa_vdev_exit(spa, vd, txg, 0);
4000
4001	spa_history_internal_log(LOG_POOL_VDEV_DETACH, spa, NULL, CRED(),
4002	    "vdev=%s", vdpath);
4003	spa_strfree(vdpath);
4004
4005	/*
4006	 * If this was the removal of the original device in a hot spare vdev,
4007	 * then we want to go through and remove the device from the hot spare
4008	 * list of every other pool.
4009	 */
4010	if (unspare) {
4011		spa_t *myspa = spa;
4012		spa = NULL;
4013		mutex_enter(&spa_namespace_lock);
4014		while ((spa = spa_next(spa)) != NULL) {
4015			if (spa->spa_state != POOL_STATE_ACTIVE)
4016				continue;
4017			if (spa == myspa)
4018				continue;
4019			spa_open_ref(spa, FTAG);
4020			mutex_exit(&spa_namespace_lock);
4021			(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4022			mutex_enter(&spa_namespace_lock);
4023			spa_close(spa, FTAG);
4024		}
4025		mutex_exit(&spa_namespace_lock);
4026	}
4027
4028	return (error);
4029}
4030
4031/*
4032 * Split a set of devices from their mirrors, and create a new pool from them.
4033 */
4034int
4035spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4036    nvlist_t *props, boolean_t exp)
4037{
4038	int error = 0;
4039	uint64_t txg, *glist;
4040	spa_t *newspa;
4041	uint_t c, children, lastlog;
4042	nvlist_t **child, *nvl, *tmp;
4043	dmu_tx_t *tx;
4044	char *altroot = NULL;
4045	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4046	boolean_t activate_slog;
4047
4048	if (!spa_writeable(spa))
4049		return (EROFS);
4050
4051	txg = spa_vdev_enter(spa);
4052
4053	/* clear the log and flush everything up to now */
4054	activate_slog = spa_passivate_log(spa);
4055	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4056	error = spa_offline_log(spa);
4057	txg = spa_vdev_config_enter(spa);
4058
4059	if (activate_slog)
4060		spa_activate_log(spa);
4061
4062	if (error != 0)
4063		return (spa_vdev_exit(spa, NULL, txg, error));
4064
4065	/* check new spa name before going any further */
4066	if (spa_lookup(newname) != NULL)
4067		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4068
4069	/*
4070	 * scan through all the children to ensure they're all mirrors
4071	 */
4072	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4073	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4074	    &children) != 0)
4075		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4076
4077	/* first, check to ensure we've got the right child count */
4078	rvd = spa->spa_root_vdev;
4079	lastlog = 0;
4080	for (c = 0; c < rvd->vdev_children; c++) {
4081		vdev_t *vd = rvd->vdev_child[c];
4082
4083		/* don't count the holes & logs as children */
4084		if (vd->vdev_islog || vd->vdev_ishole) {
4085			if (lastlog == 0)
4086				lastlog = c;
4087			continue;
4088		}
4089
4090		lastlog = 0;
4091	}
4092	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4093		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4094
4095	/* next, ensure no spare or cache devices are part of the split */
4096	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4097	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4098		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4099
4100	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4101	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4102
4103	/* then, loop over each vdev and validate it */
4104	for (c = 0; c < children; c++) {
4105		uint64_t is_hole = 0;
4106
4107		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4108		    &is_hole);
4109
4110		if (is_hole != 0) {
4111			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4112			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4113				continue;
4114			} else {
4115				error = EINVAL;
4116				break;
4117			}
4118		}
4119
4120		/* which disk is going to be split? */
4121		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4122		    &glist[c]) != 0) {
4123			error = EINVAL;
4124			break;
4125		}
4126
4127		/* look it up in the spa */
4128		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4129		if (vml[c] == NULL) {
4130			error = ENODEV;
4131			break;
4132		}
4133
4134		/* make sure there's nothing stopping the split */
4135		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4136		    vml[c]->vdev_islog ||
4137		    vml[c]->vdev_ishole ||
4138		    vml[c]->vdev_isspare ||
4139		    vml[c]->vdev_isl2cache ||
4140		    !vdev_writeable(vml[c]) ||
4141		    vml[c]->vdev_children != 0 ||
4142		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4143		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4144			error = EINVAL;
4145			break;
4146		}
4147
4148		if (vdev_dtl_required(vml[c])) {
4149			error = EBUSY;
4150			break;
4151		}
4152
4153		/* we need certain info from the top level */
4154		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4155		    vml[c]->vdev_top->vdev_ms_array) == 0);
4156		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4157		    vml[c]->vdev_top->vdev_ms_shift) == 0);
4158		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4159		    vml[c]->vdev_top->vdev_asize) == 0);
4160		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4161		    vml[c]->vdev_top->vdev_ashift) == 0);
4162	}
4163
4164	if (error != 0) {
4165		kmem_free(vml, children * sizeof (vdev_t *));
4166		kmem_free(glist, children * sizeof (uint64_t));
4167		return (spa_vdev_exit(spa, NULL, txg, error));
4168	}
4169
4170	/* stop writers from using the disks */
4171	for (c = 0; c < children; c++) {
4172		if (vml[c] != NULL)
4173			vml[c]->vdev_offline = B_TRUE;
4174	}
4175	vdev_reopen(spa->spa_root_vdev);
4176
4177	/*
4178	 * Temporarily record the splitting vdevs in the spa config.  This
4179	 * will disappear once the config is regenerated.
4180	 */
4181	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4182	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4183	    glist, children) == 0);
4184	kmem_free(glist, children * sizeof (uint64_t));
4185
4186	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4187	    nvl) == 0);
4188	spa->spa_config_splitting = nvl;
4189	vdev_config_dirty(spa->spa_root_vdev);
4190
4191	/* configure and create the new pool */
4192	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4193	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4194	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4195	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4196	    spa_version(spa)) == 0);
4197	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4198	    spa->spa_config_txg) == 0);
4199	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4200	    spa_generate_guid(NULL)) == 0);
4201	(void) nvlist_lookup_string(props,
4202	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4203
4204	/* add the new pool to the namespace */
4205	newspa = spa_add(newname, config, altroot);
4206	newspa->spa_config_txg = spa->spa_config_txg;
4207	spa_set_log_state(newspa, SPA_LOG_CLEAR);
4208
4209	/* release the spa config lock, retaining the namespace lock */
4210	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4211
4212	if (zio_injection_enabled)
4213		zio_handle_panic_injection(spa, FTAG, 1);
4214
4215	spa_activate(newspa, spa_mode_global);
4216	spa_async_suspend(newspa);
4217
4218	/* create the new pool from the disks of the original pool */
4219	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4220	if (error)
4221		goto out;
4222
4223	/* if that worked, generate a real config for the new pool */
4224	if (newspa->spa_root_vdev != NULL) {
4225		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4226		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4227		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4228		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4229		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4230		    B_TRUE));
4231	}
4232
4233	/* set the props */
4234	if (props != NULL) {
4235		spa_configfile_set(newspa, props, B_FALSE);
4236		error = spa_prop_set(newspa, props);
4237		if (error)
4238			goto out;
4239	}
4240
4241	/* flush everything */
4242	txg = spa_vdev_config_enter(newspa);
4243	vdev_config_dirty(newspa->spa_root_vdev);
4244	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4245
4246	if (zio_injection_enabled)
4247		zio_handle_panic_injection(spa, FTAG, 2);
4248
4249	spa_async_resume(newspa);
4250
4251	/* finally, update the original pool's config */
4252	txg = spa_vdev_config_enter(spa);
4253	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4254	error = dmu_tx_assign(tx, TXG_WAIT);
4255	if (error != 0)
4256		dmu_tx_abort(tx);
4257	for (c = 0; c < children; c++) {
4258		if (vml[c] != NULL) {
4259			vdev_split(vml[c]);
4260			if (error == 0)
4261				spa_history_internal_log(LOG_POOL_VDEV_DETACH,
4262				    spa, tx, CRED(), "vdev=%s",
4263				    vml[c]->vdev_path);
4264			vdev_free(vml[c]);
4265		}
4266	}
4267	vdev_config_dirty(spa->spa_root_vdev);
4268	spa->spa_config_splitting = NULL;
4269	nvlist_free(nvl);
4270	if (error == 0)
4271		dmu_tx_commit(tx);
4272	(void) spa_vdev_exit(spa, NULL, txg, 0);
4273
4274	if (zio_injection_enabled)
4275		zio_handle_panic_injection(spa, FTAG, 3);
4276
4277	/* split is complete; log a history record */
4278	spa_history_internal_log(LOG_POOL_SPLIT, newspa, NULL, CRED(),
4279	    "split new pool %s from pool %s", newname, spa_name(spa));
4280
4281	kmem_free(vml, children * sizeof (vdev_t *));
4282
4283	/* if we're not going to mount the filesystems in userland, export */
4284	if (exp)
4285		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4286		    B_FALSE, B_FALSE);
4287
4288	return (error);
4289
4290out:
4291	spa_unload(newspa);
4292	spa_deactivate(newspa);
4293	spa_remove(newspa);
4294
4295	txg = spa_vdev_config_enter(spa);
4296	nvlist_free(spa->spa_config_splitting);
4297	spa->spa_config_splitting = NULL;
4298	(void) spa_vdev_exit(spa, NULL, txg, error);
4299
4300	kmem_free(vml, children * sizeof (vdev_t *));
4301	return (error);
4302}
4303
4304static nvlist_t *
4305spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4306{
4307	for (int i = 0; i < count; i++) {
4308		uint64_t guid;
4309
4310		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4311		    &guid) == 0);
4312
4313		if (guid == target_guid)
4314			return (nvpp[i]);
4315	}
4316
4317	return (NULL);
4318}
4319
4320static void
4321spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4322	nvlist_t *dev_to_remove)
4323{
4324	nvlist_t **newdev = NULL;
4325
4326	if (count > 1)
4327		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4328
4329	for (int i = 0, j = 0; i < count; i++) {
4330		if (dev[i] == dev_to_remove)
4331			continue;
4332		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4333	}
4334
4335	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4336	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4337
4338	for (int i = 0; i < count - 1; i++)
4339		nvlist_free(newdev[i]);
4340
4341	if (count > 1)
4342		kmem_free(newdev, (count - 1) * sizeof (void *));
4343}
4344
4345/*
4346 * Removing a device from the vdev namespace requires several steps
4347 * and can take a significant amount of time.  As a result we use
4348 * the spa_vdev_config_[enter/exit] functions which allow us to
4349 * grab and release the spa_config_lock while still holding the namespace
4350 * lock.  During each step the configuration is synced out.
4351 */
4352
4353/*
4354 * Evacuate the device.
4355 */
4356int
4357spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4358{
4359	int error = 0;
4360	uint64_t txg;
4361
4362	ASSERT(MUTEX_HELD(&spa_namespace_lock));
4363	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4364	ASSERT(vd == vd->vdev_top);
4365
4366	/*
4367	 * Evacuate the device.  We don't hold the config lock as writer
4368	 * since we need to do I/O but we do keep the
4369	 * spa_namespace_lock held.  Once this completes the device
4370	 * should no longer have any blocks allocated on it.
4371	 */
4372	if (vd->vdev_islog) {
4373		error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
4374		    NULL, DS_FIND_CHILDREN);
4375	} else {
4376		error = ENOTSUP;	/* until we have bp rewrite */
4377	}
4378
4379	txg_wait_synced(spa_get_dsl(spa), 0);
4380
4381	if (error)
4382		return (error);
4383
4384	/*
4385	 * The evacuation succeeded.  Remove any remaining MOS metadata
4386	 * associated with this vdev, and wait for these changes to sync.
4387	 */
4388	txg = spa_vdev_config_enter(spa);
4389	vd->vdev_removing = B_TRUE;
4390	vdev_dirty(vd, 0, NULL, txg);
4391	vdev_config_dirty(vd);
4392	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4393
4394	return (0);
4395}
4396
4397/*
4398 * Complete the removal by cleaning up the namespace.
4399 */
4400void
4401spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
4402{
4403	vdev_t *rvd = spa->spa_root_vdev;
4404	uint64_t id = vd->vdev_id;
4405	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
4406
4407	ASSERT(MUTEX_HELD(&spa_namespace_lock));
4408	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4409	ASSERT(vd == vd->vdev_top);
4410
4411	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4412
4413	if (list_link_active(&vd->vdev_state_dirty_node))
4414		vdev_state_clean(vd);
4415	if (list_link_active(&vd->vdev_config_dirty_node))
4416		vdev_config_clean(vd);
4417
4418	vdev_free(vd);
4419
4420	if (last_vdev) {
4421		vdev_compact_children(rvd);
4422	} else {
4423		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
4424		vdev_add_child(rvd, vd);
4425	}
4426	vdev_config_dirty(rvd);
4427
4428	/*
4429	 * Reassess the health of our root vdev.
4430	 */
4431	vdev_reopen(rvd);
4432}
4433
4434/*
4435 * Remove a device from the pool.  Currently, this supports removing only hot
4436 * spares, slogs, and level 2 ARC devices.
4437 */
4438int
4439spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
4440{
4441	vdev_t *vd;
4442	metaslab_group_t *mg;
4443	nvlist_t **spares, **l2cache, *nv;
4444	uint64_t txg = 0;
4445	uint_t nspares, nl2cache;
4446	int error = 0;
4447	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
4448
4449	if (!locked)
4450		txg = spa_vdev_enter(spa);
4451
4452	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4453
4454	if (spa->spa_spares.sav_vdevs != NULL &&
4455	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4456	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
4457	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
4458		/*
4459		 * Only remove the hot spare if it's not currently in use
4460		 * in this pool.
4461		 */
4462		if (vd == NULL || unspare) {
4463			spa_vdev_remove_aux(spa->spa_spares.sav_config,
4464			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
4465			spa_load_spares(spa);
4466			spa->spa_spares.sav_sync = B_TRUE;
4467		} else {
4468			error = EBUSY;
4469		}
4470	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
4471	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4472	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
4473	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
4474		/*
4475		 * Cache devices can always be removed.
4476		 */
4477		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
4478		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
4479		spa_load_l2cache(spa);
4480		spa->spa_l2cache.sav_sync = B_TRUE;
4481	} else if (vd != NULL && vd->vdev_islog) {
4482		ASSERT(!locked);
4483		ASSERT(vd == vd->vdev_top);
4484
4485		/*
4486		 * XXX - Once we have bp-rewrite this should
4487		 * become the common case.
4488		 */
4489
4490		mg = vd->vdev_mg;
4491
4492		/*
4493		 * Stop allocating from this vdev.
4494		 */
4495		metaslab_group_passivate(mg);
4496
4497		/*
4498		 * Wait for the youngest allocations and frees to sync,
4499		 * and then wait for the deferral of those frees to finish.
4500		 */
4501		spa_vdev_config_exit(spa, NULL,
4502		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
4503
4504		/*
4505		 * Attempt to evacuate the vdev.
4506		 */
4507		error = spa_vdev_remove_evacuate(spa, vd);
4508
4509		txg = spa_vdev_config_enter(spa);
4510
4511		/*
4512		 * If we couldn't evacuate the vdev, unwind.
4513		 */
4514		if (error) {
4515			metaslab_group_activate(mg);
4516			return (spa_vdev_exit(spa, NULL, txg, error));
4517		}
4518
4519		/*
4520		 * Clean up the vdev namespace.
4521		 */
4522		spa_vdev_remove_from_namespace(spa, vd);
4523
4524	} else if (vd != NULL) {
4525		/*
4526		 * Normal vdevs cannot be removed (yet).
4527		 */
4528		error = ENOTSUP;
4529	} else {
4530		/*
4531		 * There is no vdev of any kind with the specified guid.
4532		 */
4533		error = ENOENT;
4534	}
4535
4536	if (!locked)
4537		return (spa_vdev_exit(spa, NULL, txg, error));
4538
4539	return (error);
4540}
4541
4542/*
4543 * Find any device that's done replacing, or a vdev marked 'unspare' that's
4544 * current spared, so we can detach it.
4545 */
4546static vdev_t *
4547spa_vdev_resilver_done_hunt(vdev_t *vd)
4548{
4549	vdev_t *newvd, *oldvd;
4550
4551	for (int c = 0; c < vd->vdev_children; c++) {
4552		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
4553		if (oldvd != NULL)
4554			return (oldvd);
4555	}
4556
4557	/*
4558	 * Check for a completed replacement.
4559	 */
4560	if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
4561		oldvd = vd->vdev_child[0];
4562		newvd = vd->vdev_child[1];
4563
4564		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4565		    !vdev_dtl_required(oldvd))
4566			return (oldvd);
4567	}
4568
4569	/*
4570	 * Check for a completed resilver with the 'unspare' flag set.
4571	 */
4572	if (vd->vdev_ops == &vdev_spare_ops && vd->vdev_children == 2) {
4573		newvd = vd->vdev_child[0];
4574		oldvd = vd->vdev_child[1];
4575
4576		if (newvd->vdev_unspare &&
4577		    vdev_dtl_empty(newvd, DTL_MISSING) &&
4578		    !vdev_dtl_required(oldvd)) {
4579			newvd->vdev_unspare = 0;
4580			return (oldvd);
4581		}
4582	}
4583
4584	return (NULL);
4585}
4586
4587static void
4588spa_vdev_resilver_done(spa_t *spa)
4589{
4590	vdev_t *vd, *pvd, *ppvd;
4591	uint64_t guid, sguid, pguid, ppguid;
4592
4593	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4594
4595	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4596		pvd = vd->vdev_parent;
4597		ppvd = pvd->vdev_parent;
4598		guid = vd->vdev_guid;
4599		pguid = pvd->vdev_guid;
4600		ppguid = ppvd->vdev_guid;
4601		sguid = 0;
4602		/*
4603		 * If we have just finished replacing a hot spared device, then
4604		 * we need to detach the parent's first child (the original hot
4605		 * spare) as well.
4606		 */
4607		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0) {
4608			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4609			ASSERT(ppvd->vdev_children == 2);
4610			sguid = ppvd->vdev_child[1]->vdev_guid;
4611		}
4612		spa_config_exit(spa, SCL_ALL, FTAG);
4613		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4614			return;
4615		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4616			return;
4617		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4618	}
4619
4620	spa_config_exit(spa, SCL_ALL, FTAG);
4621}
4622
4623/*
4624 * Update the stored path or FRU for this vdev.
4625 */
4626int
4627spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4628    boolean_t ispath)
4629{
4630	vdev_t *vd;
4631
4632	spa_vdev_state_enter(spa, SCL_ALL);
4633
4634	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4635		return (spa_vdev_state_exit(spa, NULL, ENOENT));
4636
4637	if (!vd->vdev_ops->vdev_op_leaf)
4638		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
4639
4640	if (ispath) {
4641		spa_strfree(vd->vdev_path);
4642		vd->vdev_path = spa_strdup(value);
4643	} else {
4644		if (vd->vdev_fru != NULL)
4645			spa_strfree(vd->vdev_fru);
4646		vd->vdev_fru = spa_strdup(value);
4647	}
4648
4649	return (spa_vdev_state_exit(spa, vd, 0));
4650}
4651
4652int
4653spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
4654{
4655	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
4656}
4657
4658int
4659spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
4660{
4661	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
4662}
4663
4664/*
4665 * ==========================================================================
4666 * SPA Scrubbing
4667 * ==========================================================================
4668 */
4669
4670int
4671spa_scrub(spa_t *spa, pool_scrub_type_t type)
4672{
4673	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4674
4675	if ((uint_t)type >= POOL_SCRUB_TYPES)
4676		return (ENOTSUP);
4677
4678	/*
4679	 * If a resilver was requested, but there is no DTL on a
4680	 * writeable leaf device, we have nothing to do.
4681	 */
4682	if (type == POOL_SCRUB_RESILVER &&
4683	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
4684		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
4685		return (0);
4686	}
4687
4688	if (type == POOL_SCRUB_EVERYTHING &&
4689	    spa->spa_dsl_pool->dp_scrub_func != SCRUB_FUNC_NONE &&
4690	    spa->spa_dsl_pool->dp_scrub_isresilver)
4691		return (EBUSY);
4692
4693	if (type == POOL_SCRUB_EVERYTHING || type == POOL_SCRUB_RESILVER) {
4694		return (dsl_pool_scrub_clean(spa->spa_dsl_pool));
4695	} else if (type == POOL_SCRUB_NONE) {
4696		return (dsl_pool_scrub_cancel(spa->spa_dsl_pool));
4697	} else {
4698		return (EINVAL);
4699	}
4700}
4701
4702/*
4703 * ==========================================================================
4704 * SPA async task processing
4705 * ==========================================================================
4706 */
4707
4708static void
4709spa_async_remove(spa_t *spa, vdev_t *vd)
4710{
4711	if (vd->vdev_remove_wanted) {
4712		vd->vdev_remove_wanted = 0;
4713		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
4714
4715		/*
4716		 * We want to clear the stats, but we don't want to do a full
4717		 * vdev_clear() as that will cause us to throw away
4718		 * degraded/faulted state as well as attempt to reopen the
4719		 * device, all of which is a waste.
4720		 */
4721		vd->vdev_stat.vs_read_errors = 0;
4722		vd->vdev_stat.vs_write_errors = 0;
4723		vd->vdev_stat.vs_checksum_errors = 0;
4724
4725		vdev_state_dirty(vd->vdev_top);
4726	}
4727
4728	for (int c = 0; c < vd->vdev_children; c++)
4729		spa_async_remove(spa, vd->vdev_child[c]);
4730}
4731
4732static void
4733spa_async_probe(spa_t *spa, vdev_t *vd)
4734{
4735	if (vd->vdev_probe_wanted) {
4736		vd->vdev_probe_wanted = 0;
4737		vdev_reopen(vd);	/* vdev_open() does the actual probe */
4738	}
4739
4740	for (int c = 0; c < vd->vdev_children; c++)
4741		spa_async_probe(spa, vd->vdev_child[c]);
4742}
4743
4744static void
4745spa_async_autoexpand(spa_t *spa, vdev_t *vd)
4746{
4747	sysevent_id_t eid;
4748	nvlist_t *attr;
4749	char *physpath;
4750
4751	if (!spa->spa_autoexpand)
4752		return;
4753
4754	for (int c = 0; c < vd->vdev_children; c++) {
4755		vdev_t *cvd = vd->vdev_child[c];
4756		spa_async_autoexpand(spa, cvd);
4757	}
4758
4759	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
4760		return;
4761
4762	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
4763	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
4764
4765	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4766	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
4767
4768	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
4769	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
4770
4771	nvlist_free(attr);
4772	kmem_free(physpath, MAXPATHLEN);
4773}
4774
4775static void
4776spa_async_thread(spa_t *spa)
4777{
4778	int tasks;
4779
4780	ASSERT(spa->spa_sync_on);
4781
4782	mutex_enter(&spa->spa_async_lock);
4783	tasks = spa->spa_async_tasks;
4784	spa->spa_async_tasks = 0;
4785	mutex_exit(&spa->spa_async_lock);
4786
4787	/*
4788	 * See if the config needs to be updated.
4789	 */
4790	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
4791		uint64_t old_space, new_space;
4792
4793		mutex_enter(&spa_namespace_lock);
4794		old_space = metaslab_class_get_space(spa_normal_class(spa));
4795		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4796		new_space = metaslab_class_get_space(spa_normal_class(spa));
4797		mutex_exit(&spa_namespace_lock);
4798
4799		/*
4800		 * If the pool grew as a result of the config update,
4801		 * then log an internal history event.
4802		 */
4803		if (new_space != old_space) {
4804			spa_history_internal_log(LOG_POOL_VDEV_ONLINE,
4805			    spa, NULL, CRED(),
4806			    "pool '%s' size: %llu(+%llu)",
4807			    spa_name(spa), new_space, new_space - old_space);
4808		}
4809	}
4810
4811	/*
4812	 * See if any devices need to be marked REMOVED.
4813	 */
4814	if (tasks & SPA_ASYNC_REMOVE) {
4815		spa_vdev_state_enter(spa, SCL_NONE);
4816		spa_async_remove(spa, spa->spa_root_vdev);
4817		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
4818			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
4819		for (int i = 0; i < spa->spa_spares.sav_count; i++)
4820			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
4821		(void) spa_vdev_state_exit(spa, NULL, 0);
4822	}
4823
4824	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
4825		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4826		spa_async_autoexpand(spa, spa->spa_root_vdev);
4827		spa_config_exit(spa, SCL_CONFIG, FTAG);
4828	}
4829
4830	/*
4831	 * See if any devices need to be probed.
4832	 */
4833	if (tasks & SPA_ASYNC_PROBE) {
4834		spa_vdev_state_enter(spa, SCL_NONE);
4835		spa_async_probe(spa, spa->spa_root_vdev);
4836		(void) spa_vdev_state_exit(spa, NULL, 0);
4837	}
4838
4839	/*
4840	 * If any devices are done replacing, detach them.
4841	 */
4842	if (tasks & SPA_ASYNC_RESILVER_DONE)
4843		spa_vdev_resilver_done(spa);
4844
4845	/*
4846	 * Kick off a resilver.
4847	 */
4848	if (tasks & SPA_ASYNC_RESILVER)
4849		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER) == 0);
4850
4851	/*
4852	 * Let the world know that we're done.
4853	 */
4854	mutex_enter(&spa->spa_async_lock);
4855	spa->spa_async_thread = NULL;
4856	cv_broadcast(&spa->spa_async_cv);
4857	mutex_exit(&spa->spa_async_lock);
4858	thread_exit();
4859}
4860
4861void
4862spa_async_suspend(spa_t *spa)
4863{
4864	mutex_enter(&spa->spa_async_lock);
4865	spa->spa_async_suspended++;
4866	while (spa->spa_async_thread != NULL)
4867		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
4868	mutex_exit(&spa->spa_async_lock);
4869}
4870
4871void
4872spa_async_resume(spa_t *spa)
4873{
4874	mutex_enter(&spa->spa_async_lock);
4875	ASSERT(spa->spa_async_suspended != 0);
4876	spa->spa_async_suspended--;
4877	mutex_exit(&spa->spa_async_lock);
4878}
4879
4880static void
4881spa_async_dispatch(spa_t *spa)
4882{
4883	mutex_enter(&spa->spa_async_lock);
4884	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
4885	    spa->spa_async_thread == NULL &&
4886	    rootdir != NULL && !vn_is_readonly(rootdir))
4887		spa->spa_async_thread = thread_create(NULL, 0,
4888		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
4889	mutex_exit(&spa->spa_async_lock);
4890}
4891
4892void
4893spa_async_request(spa_t *spa, int task)
4894{
4895	mutex_enter(&spa->spa_async_lock);
4896	spa->spa_async_tasks |= task;
4897	mutex_exit(&spa->spa_async_lock);
4898}
4899
4900/*
4901 * ==========================================================================
4902 * SPA syncing routines
4903 * ==========================================================================
4904 */
4905static void
4906spa_sync_deferred_bplist(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx, uint64_t txg)
4907{
4908	blkptr_t blk;
4909	uint64_t itor = 0;
4910	uint8_t c = 1;
4911
4912	while (bplist_iterate(bpl, &itor, &blk) == 0) {
4913		ASSERT(blk.blk_birth < txg);
4914		zio_free(spa, txg, &blk);
4915	}
4916
4917	bplist_vacate(bpl, tx);
4918
4919	/*
4920	 * Pre-dirty the first block so we sync to convergence faster.
4921	 * (Usually only the first block is needed.)
4922	 */
4923	dmu_write(bpl->bpl_mos, spa->spa_deferred_bplist_obj, 0, 1, &c, tx);
4924}
4925
4926static void
4927spa_sync_free(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
4928{
4929	zio_t *zio = arg;
4930
4931	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
4932	    zio->io_flags));
4933}
4934
4935static void
4936spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
4937{
4938	char *packed = NULL;
4939	size_t bufsize;
4940	size_t nvsize = 0;
4941	dmu_buf_t *db;
4942
4943	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
4944
4945	/*
4946	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
4947	 * information.  This avoids the dbuf_will_dirty() path and
4948	 * saves us a pre-read to get data we don't actually care about.
4949	 */
4950	bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
4951	packed = kmem_alloc(bufsize, KM_SLEEP);
4952
4953	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
4954	    KM_SLEEP) == 0);
4955	bzero(packed + nvsize, bufsize - nvsize);
4956
4957	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
4958
4959	kmem_free(packed, bufsize);
4960
4961	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
4962	dmu_buf_will_dirty(db, tx);
4963	*(uint64_t *)db->db_data = nvsize;
4964	dmu_buf_rele(db, FTAG);
4965}
4966
4967static void
4968spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
4969    const char *config, const char *entry)
4970{
4971	nvlist_t *nvroot;
4972	nvlist_t **list;
4973	int i;
4974
4975	if (!sav->sav_sync)
4976		return;
4977
4978	/*
4979	 * Update the MOS nvlist describing the list of available devices.
4980	 * spa_validate_aux() will have already made sure this nvlist is
4981	 * valid and the vdevs are labeled appropriately.
4982	 */
4983	if (sav->sav_object == 0) {
4984		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
4985		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
4986		    sizeof (uint64_t), tx);
4987		VERIFY(zap_update(spa->spa_meta_objset,
4988		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
4989		    &sav->sav_object, tx) == 0);
4990	}
4991
4992	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4993	if (sav->sav_count == 0) {
4994		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
4995	} else {
4996		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
4997		for (i = 0; i < sav->sav_count; i++)
4998			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
4999			    B_FALSE, B_FALSE, B_TRUE);
5000		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5001		    sav->sav_count) == 0);
5002		for (i = 0; i < sav->sav_count; i++)
5003			nvlist_free(list[i]);
5004		kmem_free(list, sav->sav_count * sizeof (void *));
5005	}
5006
5007	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5008	nvlist_free(nvroot);
5009
5010	sav->sav_sync = B_FALSE;
5011}
5012
5013static void
5014spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5015{
5016	nvlist_t *config;
5017
5018	if (list_is_empty(&spa->spa_config_dirty_list))
5019		return;
5020
5021	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5022
5023	config = spa_config_generate(spa, spa->spa_root_vdev,
5024	    dmu_tx_get_txg(tx), B_FALSE);
5025
5026	spa_config_exit(spa, SCL_STATE, FTAG);
5027
5028	if (spa->spa_config_syncing)
5029		nvlist_free(spa->spa_config_syncing);
5030	spa->spa_config_syncing = config;
5031
5032	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5033}
5034
5035/*
5036 * Set zpool properties.
5037 */
5038static void
5039spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
5040{
5041	spa_t *spa = arg1;
5042	objset_t *mos = spa->spa_meta_objset;
5043	nvlist_t *nvp = arg2;
5044	nvpair_t *elem;
5045	uint64_t intval;
5046	char *strval;
5047	zpool_prop_t prop;
5048	const char *propname;
5049	zprop_type_t proptype;
5050
5051	mutex_enter(&spa->spa_props_lock);
5052
5053	elem = NULL;
5054	while ((elem = nvlist_next_nvpair(nvp, elem))) {
5055		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5056		case ZPOOL_PROP_VERSION:
5057			/*
5058			 * Only set version for non-zpool-creation cases
5059			 * (set/import). spa_create() needs special care
5060			 * for version setting.
5061			 */
5062			if (tx->tx_txg != TXG_INITIAL) {
5063				VERIFY(nvpair_value_uint64(elem,
5064				    &intval) == 0);
5065				ASSERT(intval <= SPA_VERSION);
5066				ASSERT(intval >= spa_version(spa));
5067				spa->spa_uberblock.ub_version = intval;
5068				vdev_config_dirty(spa->spa_root_vdev);
5069			}
5070			break;
5071
5072		case ZPOOL_PROP_ALTROOT:
5073			/*
5074			 * 'altroot' is a non-persistent property. It should
5075			 * have been set temporarily at creation or import time.
5076			 */
5077			ASSERT(spa->spa_root != NULL);
5078			break;
5079
5080		case ZPOOL_PROP_CACHEFILE:
5081			/*
5082			 * 'cachefile' is also a non-persisitent property.
5083			 */
5084			break;
5085		default:
5086			/*
5087			 * Set pool property values in the poolprops mos object.
5088			 */
5089			if (spa->spa_pool_props_object == 0) {
5090				VERIFY((spa->spa_pool_props_object =
5091				    zap_create(mos, DMU_OT_POOL_PROPS,
5092				    DMU_OT_NONE, 0, tx)) > 0);
5093
5094				VERIFY(zap_update(mos,
5095				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5096				    8, 1, &spa->spa_pool_props_object, tx)
5097				    == 0);
5098			}
5099
5100			/* normalize the property name */
5101			propname = zpool_prop_to_name(prop);
5102			proptype = zpool_prop_get_type(prop);
5103
5104			if (nvpair_type(elem) == DATA_TYPE_STRING) {
5105				ASSERT(proptype == PROP_TYPE_STRING);
5106				VERIFY(nvpair_value_string(elem, &strval) == 0);
5107				VERIFY(zap_update(mos,
5108				    spa->spa_pool_props_object, propname,
5109				    1, strlen(strval) + 1, strval, tx) == 0);
5110
5111			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5112				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5113
5114				if (proptype == PROP_TYPE_INDEX) {
5115					const char *unused;
5116					VERIFY(zpool_prop_index_to_string(
5117					    prop, intval, &unused) == 0);
5118				}
5119				VERIFY(zap_update(mos,
5120				    spa->spa_pool_props_object, propname,
5121				    8, 1, &intval, tx) == 0);
5122			} else {
5123				ASSERT(0); /* not allowed */
5124			}
5125
5126			switch (prop) {
5127			case ZPOOL_PROP_DELEGATION:
5128				spa->spa_delegation = intval;
5129				break;
5130			case ZPOOL_PROP_BOOTFS:
5131				spa->spa_bootfs = intval;
5132				break;
5133			case ZPOOL_PROP_FAILUREMODE:
5134				spa->spa_failmode = intval;
5135				break;
5136			case ZPOOL_PROP_AUTOEXPAND:
5137				spa->spa_autoexpand = intval;
5138				spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
5139				break;
5140			case ZPOOL_PROP_DEDUPDITTO:
5141				spa->spa_dedup_ditto = intval;
5142				break;
5143			default:
5144				break;
5145			}
5146		}
5147
5148		/* log internal history if this is not a zpool create */
5149		if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5150		    tx->tx_txg != TXG_INITIAL) {
5151			spa_history_internal_log(LOG_POOL_PROPSET,
5152			    spa, tx, cr, "%s %lld %s",
5153			    nvpair_name(elem), intval, spa_name(spa));
5154		}
5155	}
5156
5157	mutex_exit(&spa->spa_props_lock);
5158}
5159
5160/*
5161 * Sync the specified transaction group.  New blocks may be dirtied as
5162 * part of the process, so we iterate until it converges.
5163 */
5164void
5165spa_sync(spa_t *spa, uint64_t txg)
5166{
5167	dsl_pool_t *dp = spa->spa_dsl_pool;
5168	objset_t *mos = spa->spa_meta_objset;
5169	bplist_t *defer_bpl = &spa->spa_deferred_bplist;
5170	bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5171	vdev_t *rvd = spa->spa_root_vdev;
5172	vdev_t *vd;
5173	dmu_tx_t *tx;
5174	int error;
5175
5176	/*
5177	 * Lock out configuration changes.
5178	 */
5179	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5180
5181	spa->spa_syncing_txg = txg;
5182	spa->spa_sync_pass = 0;
5183
5184	/*
5185	 * If there are any pending vdev state changes, convert them
5186	 * into config changes that go out with this transaction group.
5187	 */
5188	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5189	while (list_head(&spa->spa_state_dirty_list) != NULL) {
5190		/*
5191		 * We need the write lock here because, for aux vdevs,
5192		 * calling vdev_config_dirty() modifies sav_config.
5193		 * This is ugly and will become unnecessary when we
5194		 * eliminate the aux vdev wart by integrating all vdevs
5195		 * into the root vdev tree.
5196		 */
5197		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5198		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5199		while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5200			vdev_state_clean(vd);
5201			vdev_config_dirty(vd);
5202		}
5203		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5204		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5205	}
5206	spa_config_exit(spa, SCL_STATE, FTAG);
5207
5208	VERIFY(0 == bplist_open(defer_bpl, mos, spa->spa_deferred_bplist_obj));
5209
5210	tx = dmu_tx_create_assigned(dp, txg);
5211
5212	/*
5213	 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5214	 * set spa_deflate if we have no raid-z vdevs.
5215	 */
5216	if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5217	    spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5218		int i;
5219
5220		for (i = 0; i < rvd->vdev_children; i++) {
5221			vd = rvd->vdev_child[i];
5222			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5223				break;
5224		}
5225		if (i == rvd->vdev_children) {
5226			spa->spa_deflate = TRUE;
5227			VERIFY(0 == zap_add(spa->spa_meta_objset,
5228			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5229			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5230		}
5231	}
5232
5233	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5234	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5235		dsl_pool_create_origin(dp, tx);
5236
5237		/* Keeping the origin open increases spa_minref */
5238		spa->spa_minref += 3;
5239	}
5240
5241	if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5242	    spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5243		dsl_pool_upgrade_clones(dp, tx);
5244	}
5245
5246	/*
5247	 * If anything has changed in this txg, push the deferred frees
5248	 * from the previous txg.  If not, leave them alone so that we
5249	 * don't generate work on an otherwise idle system.
5250	 */
5251	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5252	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5253	    !txg_list_empty(&dp->dp_sync_tasks, txg))
5254		spa_sync_deferred_bplist(spa, defer_bpl, tx, txg);
5255
5256	/*
5257	 * Iterate to convergence.
5258	 */
5259	do {
5260		int pass = ++spa->spa_sync_pass;
5261
5262		spa_sync_config_object(spa, tx);
5263		spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5264		    ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5265		spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5266		    ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5267		spa_errlog_sync(spa, txg);
5268		dsl_pool_sync(dp, txg);
5269
5270		if (pass <= SYNC_PASS_DEFERRED_FREE) {
5271			zio_t *zio = zio_root(spa, NULL, NULL, 0);
5272			bplist_sync(free_bpl, spa_sync_free, zio, tx);
5273			VERIFY(zio_wait(zio) == 0);
5274		} else {
5275			bplist_sync(free_bpl, bplist_enqueue_cb, defer_bpl, tx);
5276		}
5277
5278		ddt_sync(spa, txg);
5279
5280		mutex_enter(&spa->spa_scrub_lock);
5281		while (spa->spa_scrub_inflight > 0)
5282			cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
5283		mutex_exit(&spa->spa_scrub_lock);
5284
5285		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
5286			vdev_sync(vd, txg);
5287
5288	} while (dmu_objset_is_dirty(mos, txg));
5289
5290	ASSERT(free_bpl->bpl_queue == NULL);
5291
5292	bplist_close(defer_bpl);
5293
5294	/*
5295	 * Rewrite the vdev configuration (which includes the uberblock)
5296	 * to commit the transaction group.
5297	 *
5298	 * If there are no dirty vdevs, we sync the uberblock to a few
5299	 * random top-level vdevs that are known to be visible in the
5300	 * config cache (see spa_vdev_add() for a complete description).
5301	 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
5302	 */
5303	for (;;) {
5304		/*
5305		 * We hold SCL_STATE to prevent vdev open/close/etc.
5306		 * while we're attempting to write the vdev labels.
5307		 */
5308		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5309
5310		if (list_is_empty(&spa->spa_config_dirty_list)) {
5311			vdev_t *svd[SPA_DVAS_PER_BP];
5312			int svdcount = 0;
5313			int children = rvd->vdev_children;
5314			int c0 = spa_get_random(children);
5315
5316			for (int c = 0; c < children; c++) {
5317				vd = rvd->vdev_child[(c0 + c) % children];
5318				if (vd->vdev_ms_array == 0 || vd->vdev_islog)
5319					continue;
5320				svd[svdcount++] = vd;
5321				if (svdcount == SPA_DVAS_PER_BP)
5322					break;
5323			}
5324			error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
5325			if (error != 0)
5326				error = vdev_config_sync(svd, svdcount, txg,
5327				    B_TRUE);
5328		} else {
5329			error = vdev_config_sync(rvd->vdev_child,
5330			    rvd->vdev_children, txg, B_FALSE);
5331			if (error != 0)
5332				error = vdev_config_sync(rvd->vdev_child,
5333				    rvd->vdev_children, txg, B_TRUE);
5334		}
5335
5336		spa_config_exit(spa, SCL_STATE, FTAG);
5337
5338		if (error == 0)
5339			break;
5340		zio_suspend(spa, NULL);
5341		zio_resume_wait(spa);
5342	}
5343	dmu_tx_commit(tx);
5344
5345	/*
5346	 * Clear the dirty config list.
5347	 */
5348	while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
5349		vdev_config_clean(vd);
5350
5351	/*
5352	 * Now that the new config has synced transactionally,
5353	 * let it become visible to the config cache.
5354	 */
5355	if (spa->spa_config_syncing != NULL) {
5356		spa_config_set(spa, spa->spa_config_syncing);
5357		spa->spa_config_txg = txg;
5358		spa->spa_config_syncing = NULL;
5359	}
5360
5361	spa->spa_ubsync = spa->spa_uberblock;
5362
5363	dsl_pool_sync_done(dp, txg);
5364
5365	/*
5366	 * Update usable space statistics.
5367	 */
5368	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
5369		vdev_sync_done(vd, txg);
5370
5371	spa_update_dspace(spa);
5372
5373	/*
5374	 * It had better be the case that we didn't dirty anything
5375	 * since vdev_config_sync().
5376	 */
5377	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
5378	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
5379	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
5380	ASSERT(defer_bpl->bpl_queue == NULL);
5381	ASSERT(free_bpl->bpl_queue == NULL);
5382
5383	spa->spa_sync_pass = 0;
5384
5385	spa_config_exit(spa, SCL_CONFIG, FTAG);
5386
5387	spa_handle_ignored_writes(spa);
5388
5389	/*
5390	 * If any async tasks have been requested, kick them off.
5391	 */
5392	spa_async_dispatch(spa);
5393}
5394
5395/*
5396 * Sync all pools.  We don't want to hold the namespace lock across these
5397 * operations, so we take a reference on the spa_t and drop the lock during the
5398 * sync.
5399 */
5400void
5401spa_sync_allpools(void)
5402{
5403	spa_t *spa = NULL;
5404	mutex_enter(&spa_namespace_lock);
5405	while ((spa = spa_next(spa)) != NULL) {
5406		if (spa_state(spa) != POOL_STATE_ACTIVE || spa_suspended(spa))
5407			continue;
5408		spa_open_ref(spa, FTAG);
5409		mutex_exit(&spa_namespace_lock);
5410		txg_wait_synced(spa_get_dsl(spa), 0);
5411		mutex_enter(&spa_namespace_lock);
5412		spa_close(spa, FTAG);
5413	}
5414	mutex_exit(&spa_namespace_lock);
5415}
5416
5417/*
5418 * ==========================================================================
5419 * Miscellaneous routines
5420 * ==========================================================================
5421 */
5422
5423/*
5424 * Remove all pools in the system.
5425 */
5426void
5427spa_evict_all(void)
5428{
5429	spa_t *spa;
5430
5431	/*
5432	 * Remove all cached state.  All pools should be closed now,
5433	 * so every spa in the AVL tree should be unreferenced.
5434	 */
5435	mutex_enter(&spa_namespace_lock);
5436	while ((spa = spa_next(NULL)) != NULL) {
5437		/*
5438		 * Stop async tasks.  The async thread may need to detach
5439		 * a device that's been replaced, which requires grabbing
5440		 * spa_namespace_lock, so we must drop it here.
5441		 */
5442		spa_open_ref(spa, FTAG);
5443		mutex_exit(&spa_namespace_lock);
5444		spa_async_suspend(spa);
5445		mutex_enter(&spa_namespace_lock);
5446		spa_close(spa, FTAG);
5447
5448		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5449			spa_unload(spa);
5450			spa_deactivate(spa);
5451		}
5452		spa_remove(spa);
5453	}
5454	mutex_exit(&spa_namespace_lock);
5455}
5456
5457vdev_t *
5458spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
5459{
5460	vdev_t *vd;
5461	int i;
5462
5463	if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
5464		return (vd);
5465
5466	if (aux) {
5467		for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
5468			vd = spa->spa_l2cache.sav_vdevs[i];
5469			if (vd->vdev_guid == guid)
5470				return (vd);
5471		}
5472
5473		for (i = 0; i < spa->spa_spares.sav_count; i++) {
5474			vd = spa->spa_spares.sav_vdevs[i];
5475			if (vd->vdev_guid == guid)
5476				return (vd);
5477		}
5478	}
5479
5480	return (NULL);
5481}
5482
5483void
5484spa_upgrade(spa_t *spa, uint64_t version)
5485{
5486	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5487
5488	/*
5489	 * This should only be called for a non-faulted pool, and since a
5490	 * future version would result in an unopenable pool, this shouldn't be
5491	 * possible.
5492	 */
5493	ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
5494	ASSERT(version >= spa->spa_uberblock.ub_version);
5495
5496	spa->spa_uberblock.ub_version = version;
5497	vdev_config_dirty(spa->spa_root_vdev);
5498
5499	spa_config_exit(spa, SCL_ALL, FTAG);
5500
5501	txg_wait_synced(spa_get_dsl(spa), 0);
5502}
5503
5504boolean_t
5505spa_has_spare(spa_t *spa, uint64_t guid)
5506{
5507	int i;
5508	uint64_t spareguid;
5509	spa_aux_vdev_t *sav = &spa->spa_spares;
5510
5511	for (i = 0; i < sav->sav_count; i++)
5512		if (sav->sav_vdevs[i]->vdev_guid == guid)
5513			return (B_TRUE);
5514
5515	for (i = 0; i < sav->sav_npending; i++) {
5516		if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
5517		    &spareguid) == 0 && spareguid == guid)
5518			return (B_TRUE);
5519	}
5520
5521	return (B_FALSE);
5522}
5523
5524/*
5525 * Check if a pool has an active shared spare device.
5526 * Note: reference count of an active spare is 2, as a spare and as a replace
5527 */
5528static boolean_t
5529spa_has_active_shared_spare(spa_t *spa)
5530{
5531	int i, refcnt;
5532	uint64_t pool;
5533	spa_aux_vdev_t *sav = &spa->spa_spares;
5534
5535	for (i = 0; i < sav->sav_count; i++) {
5536		if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
5537		    &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
5538		    refcnt > 2)
5539			return (B_TRUE);
5540	}
5541
5542	return (B_FALSE);
5543}
5544
5545/*
5546 * Post a sysevent corresponding to the given event.  The 'name' must be one of
5547 * the event definitions in sys/sysevent/eventdefs.h.  The payload will be
5548 * filled in from the spa and (optionally) the vdev.  This doesn't do anything
5549 * in the userland libzpool, as we don't want consumers to misinterpret ztest
5550 * or zdb as real changes.
5551 */
5552void
5553spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
5554{
5555#ifdef _KERNEL
5556	sysevent_t		*ev;
5557	sysevent_attr_list_t	*attr = NULL;
5558	sysevent_value_t	value;
5559	sysevent_id_t		eid;
5560
5561	ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
5562	    SE_SLEEP);
5563
5564	value.value_type = SE_DATA_TYPE_STRING;
5565	value.value.sv_string = spa_name(spa);
5566	if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
5567		goto done;
5568
5569	value.value_type = SE_DATA_TYPE_UINT64;
5570	value.value.sv_uint64 = spa_guid(spa);
5571	if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
5572		goto done;
5573
5574	if (vd) {
5575		value.value_type = SE_DATA_TYPE_UINT64;
5576		value.value.sv_uint64 = vd->vdev_guid;
5577		if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
5578		    SE_SLEEP) != 0)
5579			goto done;
5580
5581		if (vd->vdev_path) {
5582			value.value_type = SE_DATA_TYPE_STRING;
5583			value.value.sv_string = vd->vdev_path;
5584			if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
5585			    &value, SE_SLEEP) != 0)
5586				goto done;
5587		}
5588	}
5589
5590	if (sysevent_attach_attributes(ev, attr) != 0)
5591		goto done;
5592	attr = NULL;
5593
5594	(void) log_sysevent(ev, SE_SLEEP, &eid);
5595
5596done:
5597	if (attr)
5598		sysevent_free_attr(attr);
5599	sysevent_free(ev);
5600#endif
5601}
5602