spa.c revision afee20e4873fef88755cd47373853eb9571edf3c
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_metadata_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_metadata_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	rio = zio_root(spa, NULL, &sle,
1458	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1459
1460	error = traverse_pool(spa, spa->spa_verify_min_txg,
1461	    TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1462
1463	(void) zio_wait(rio);
1464
1465	zpool_get_rewind_policy(spa->spa_config, &policy);
1466
1467	spa->spa_load_meta_errors = sle.sle_metadata_count;
1468	spa->spa_load_data_errors = sle.sle_data_count;
1469
1470	if (!error && sle.sle_metadata_count <= policy.zrp_maxmeta &&
1471	    sle.sle_data_count <= policy.zrp_maxdata) {
1472		verify_ok = B_TRUE;
1473		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1474		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1475	} else {
1476		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1477	}
1478
1479	if (error) {
1480		if (error != ENXIO && error != EIO)
1481			error = EIO;
1482		return (error);
1483	}
1484
1485	return (verify_ok ? 0 : EIO);
1486}
1487
1488/*
1489 * Find a value in the pool props object.
1490 */
1491static void
1492spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1493{
1494	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1495	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1496}
1497
1498/*
1499 * Find a value in the pool directory object.
1500 */
1501static int
1502spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1503{
1504	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1505	    name, sizeof (uint64_t), 1, val));
1506}
1507
1508static int
1509spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1510{
1511	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1512	return (err);
1513}
1514
1515/*
1516 * Fix up config after a partly-completed split.  This is done with the
1517 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1518 * pool have that entry in their config, but only the splitting one contains
1519 * a list of all the guids of the vdevs that are being split off.
1520 *
1521 * This function determines what to do with that list: either rejoin
1522 * all the disks to the pool, or complete the splitting process.  To attempt
1523 * the rejoin, each disk that is offlined is marked online again, and
1524 * we do a reopen() call.  If the vdev label for every disk that was
1525 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1526 * then we call vdev_split() on each disk, and complete the split.
1527 *
1528 * Otherwise we leave the config alone, with all the vdevs in place in
1529 * the original pool.
1530 */
1531static void
1532spa_try_repair(spa_t *spa, nvlist_t *config)
1533{
1534	uint_t extracted;
1535	uint64_t *glist;
1536	uint_t i, gcount;
1537	nvlist_t *nvl;
1538	vdev_t **vd;
1539	boolean_t attempt_reopen;
1540
1541	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1542		return;
1543
1544	/* check that the config is complete */
1545	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1546	    &glist, &gcount) != 0)
1547		return;
1548
1549	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1550
1551	/* attempt to online all the vdevs & validate */
1552	attempt_reopen = B_TRUE;
1553	for (i = 0; i < gcount; i++) {
1554		if (glist[i] == 0)	/* vdev is hole */
1555			continue;
1556
1557		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1558		if (vd[i] == NULL) {
1559			/*
1560			 * Don't bother attempting to reopen the disks;
1561			 * just do the split.
1562			 */
1563			attempt_reopen = B_FALSE;
1564		} else {
1565			/* attempt to re-online it */
1566			vd[i]->vdev_offline = B_FALSE;
1567		}
1568	}
1569
1570	if (attempt_reopen) {
1571		vdev_reopen(spa->spa_root_vdev);
1572
1573		/* check each device to see what state it's in */
1574		for (extracted = 0, i = 0; i < gcount; i++) {
1575			if (vd[i] != NULL &&
1576			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1577				break;
1578			++extracted;
1579		}
1580	}
1581
1582	/*
1583	 * If every disk has been moved to the new pool, or if we never
1584	 * even attempted to look at them, then we split them off for
1585	 * good.
1586	 */
1587	if (!attempt_reopen || gcount == extracted) {
1588		for (i = 0; i < gcount; i++)
1589			if (vd[i] != NULL)
1590				vdev_split(vd[i]);
1591		vdev_reopen(spa->spa_root_vdev);
1592	}
1593
1594	kmem_free(vd, gcount * sizeof (vdev_t *));
1595}
1596
1597static int
1598spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1599    boolean_t mosconfig)
1600{
1601	nvlist_t *config = spa->spa_config;
1602	char *ereport = FM_EREPORT_ZFS_POOL;
1603	int error;
1604	uint64_t pool_guid;
1605	nvlist_t *nvl;
1606
1607	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1608		return (EINVAL);
1609
1610	/*
1611	 * Versioning wasn't explicitly added to the label until later, so if
1612	 * it's not present treat it as the initial version.
1613	 */
1614	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1615	    &spa->spa_ubsync.ub_version) != 0)
1616		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1617
1618	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1619	    &spa->spa_config_txg);
1620
1621	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1622	    spa_guid_exists(pool_guid, 0)) {
1623		error = EEXIST;
1624	} else {
1625		spa->spa_load_guid = pool_guid;
1626
1627		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1628		    &nvl) == 0) {
1629			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1630			    KM_SLEEP) == 0);
1631		}
1632
1633		error = spa_load_impl(spa, pool_guid, config, state, type,
1634		    mosconfig, &ereport);
1635	}
1636
1637	spa->spa_minref = refcount_count(&spa->spa_refcount);
1638	if (error && error != EBADF)
1639		zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1640	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1641	spa->spa_ena = 0;
1642
1643	return (error);
1644}
1645
1646/*
1647 * Load an existing storage pool, using the pool's builtin spa_config as a
1648 * source of configuration information.
1649 */
1650static int
1651spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1652    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1653    char **ereport)
1654{
1655	int error = 0;
1656	nvlist_t *nvconfig, *nvroot = NULL;
1657	vdev_t *rvd;
1658	uberblock_t *ub = &spa->spa_uberblock;
1659	uint64_t config_cache_txg = spa->spa_config_txg;
1660	int orig_mode = spa->spa_mode;
1661	int parse;
1662
1663	/*
1664	 * If this is an untrusted config, access the pool in read-only mode.
1665	 * This prevents things like resilvering recently removed devices.
1666	 */
1667	if (!mosconfig)
1668		spa->spa_mode = FREAD;
1669
1670	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1671
1672	spa->spa_load_state = state;
1673
1674	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1675		return (EINVAL);
1676
1677	parse = (type == SPA_IMPORT_EXISTING ?
1678	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1679
1680	/*
1681	 * Create "The Godfather" zio to hold all async IOs
1682	 */
1683	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1684	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1685
1686	/*
1687	 * Parse the configuration into a vdev tree.  We explicitly set the
1688	 * value that will be returned by spa_version() since parsing the
1689	 * configuration requires knowing the version number.
1690	 */
1691	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1692	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1693	spa_config_exit(spa, SCL_ALL, FTAG);
1694
1695	if (error != 0)
1696		return (error);
1697
1698	ASSERT(spa->spa_root_vdev == rvd);
1699
1700	if (type != SPA_IMPORT_ASSEMBLE) {
1701		ASSERT(spa_guid(spa) == pool_guid);
1702	}
1703
1704	/*
1705	 * Try to open all vdevs, loading each label in the process.
1706	 */
1707	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1708	error = vdev_open(rvd);
1709	spa_config_exit(spa, SCL_ALL, FTAG);
1710	if (error != 0)
1711		return (error);
1712
1713	/*
1714	 * We need to validate the vdev labels against the configuration that
1715	 * we have in hand, which is dependent on the setting of mosconfig. If
1716	 * mosconfig is true then we're validating the vdev labels based on
1717	 * that config.  Otherwise, we're validating against the cached config
1718	 * (zpool.cache) that was read when we loaded the zfs module, and then
1719	 * later we will recursively call spa_load() and validate against
1720	 * the vdev config.
1721	 *
1722	 * If we're assembling a new pool that's been split off from an
1723	 * existing pool, the labels haven't yet been updated so we skip
1724	 * validation for now.
1725	 */
1726	if (type != SPA_IMPORT_ASSEMBLE) {
1727		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1728		error = vdev_validate(rvd);
1729		spa_config_exit(spa, SCL_ALL, FTAG);
1730
1731		if (error != 0)
1732			return (error);
1733
1734		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1735			return (ENXIO);
1736	}
1737
1738	/*
1739	 * Find the best uberblock.
1740	 */
1741	vdev_uberblock_load(NULL, rvd, ub);
1742
1743	/*
1744	 * If we weren't able to find a single valid uberblock, return failure.
1745	 */
1746	if (ub->ub_txg == 0)
1747		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
1748
1749	/*
1750	 * If the pool is newer than the code, we can't open it.
1751	 */
1752	if (ub->ub_version > SPA_VERSION)
1753		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
1754
1755	/*
1756	 * If the vdev guid sum doesn't match the uberblock, we have an
1757	 * incomplete configuration.
1758	 */
1759	if (mosconfig && type != SPA_IMPORT_ASSEMBLE &&
1760	    rvd->vdev_guid_sum != ub->ub_guid_sum)
1761		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
1762
1763	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
1764		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1765		spa_try_repair(spa, config);
1766		spa_config_exit(spa, SCL_ALL, FTAG);
1767		nvlist_free(spa->spa_config_splitting);
1768		spa->spa_config_splitting = NULL;
1769	}
1770
1771	/*
1772	 * Initialize internal SPA structures.
1773	 */
1774	spa->spa_state = POOL_STATE_ACTIVE;
1775	spa->spa_ubsync = spa->spa_uberblock;
1776	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
1777	    TXG_INITIAL : spa_last_synced_txg(spa) - TXG_DEFER_SIZE;
1778	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
1779	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
1780	spa->spa_claim_max_txg = spa->spa_first_txg;
1781
1782	error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1783	if (error)
1784		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1785	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1786
1787	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
1788		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1789
1790	if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
1791		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1792
1793	if (!mosconfig) {
1794		uint64_t hostid;
1795
1796		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
1797		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
1798			char *hostname;
1799			unsigned long myhostid = 0;
1800
1801			VERIFY(nvlist_lookup_string(nvconfig,
1802			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
1803
1804#ifdef	_KERNEL
1805			myhostid = zone_get_hostid(NULL);
1806#else	/* _KERNEL */
1807			/*
1808			 * We're emulating the system's hostid in userland, so
1809			 * we can't use zone_get_hostid().
1810			 */
1811			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
1812#endif	/* _KERNEL */
1813			if (hostid != 0 && myhostid != 0 &&
1814			    hostid != myhostid) {
1815				cmn_err(CE_WARN, "pool '%s' could not be "
1816				    "loaded as it was last accessed by "
1817				    "another system (host: %s hostid: 0x%lx). "
1818				    "See: http://www.sun.com/msg/ZFS-8000-EY",
1819				    spa_name(spa), hostname,
1820				    (unsigned long)hostid);
1821				return (EBADF);
1822			}
1823		}
1824
1825		spa_config_set(spa, nvconfig);
1826		spa_unload(spa);
1827		spa_deactivate(spa);
1828		spa_activate(spa, orig_mode);
1829
1830		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
1831	}
1832
1833	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPLIST,
1834	    &spa->spa_deferred_bplist_obj) != 0)
1835		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1836
1837	/*
1838	 * Load the bit that tells us to use the new accounting function
1839	 * (raid-z deflation).  If we have an older pool, this will not
1840	 * be present.
1841	 */
1842	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
1843	if (error != 0 && error != ENOENT)
1844		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1845
1846	/*
1847	 * Load the persistent error log.  If we have an older pool, this will
1848	 * not be present.
1849	 */
1850	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
1851	if (error != 0 && error != ENOENT)
1852		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1853
1854	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
1855	    &spa->spa_errlog_scrub);
1856	if (error != 0 && error != ENOENT)
1857		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1858
1859	/*
1860	 * Load the history object.  If we have an older pool, this
1861	 * will not be present.
1862	 */
1863	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
1864	if (error != 0 && error != ENOENT)
1865		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1866
1867	/*
1868	 * If we're assembling the pool from the split-off vdevs of
1869	 * an existing pool, we don't want to attach the spares & cache
1870	 * devices.
1871	 */
1872
1873	/*
1874	 * Load any hot spares for this pool.
1875	 */
1876	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
1877	if (error != 0 && error != ENOENT)
1878		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1879	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
1880		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
1881		if (load_nvlist(spa, spa->spa_spares.sav_object,
1882		    &spa->spa_spares.sav_config) != 0)
1883			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1884
1885		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1886		spa_load_spares(spa);
1887		spa_config_exit(spa, SCL_ALL, FTAG);
1888	} else if (error == 0) {
1889		spa->spa_spares.sav_sync = B_TRUE;
1890	}
1891
1892	/*
1893	 * Load any level 2 ARC devices for this pool.
1894	 */
1895	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
1896	    &spa->spa_l2cache.sav_object);
1897	if (error != 0 && error != ENOENT)
1898		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1899	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
1900		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
1901		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
1902		    &spa->spa_l2cache.sav_config) != 0)
1903			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1904
1905		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1906		spa_load_l2cache(spa);
1907		spa_config_exit(spa, SCL_ALL, FTAG);
1908	} else if (error == 0) {
1909		spa->spa_l2cache.sav_sync = B_TRUE;
1910	}
1911
1912	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
1913
1914	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
1915	if (error && error != ENOENT)
1916		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1917
1918	if (error == 0) {
1919		uint64_t autoreplace;
1920
1921		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
1922		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
1923		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
1924		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
1925		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
1926		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
1927		    &spa->spa_dedup_ditto);
1928
1929		spa->spa_autoreplace = (autoreplace != 0);
1930	}
1931
1932	/*
1933	 * If the 'autoreplace' property is set, then post a resource notifying
1934	 * the ZFS DE that it should not issue any faults for unopenable
1935	 * devices.  We also iterate over the vdevs, and post a sysevent for any
1936	 * unopenable vdevs so that the normal autoreplace handler can take
1937	 * over.
1938	 */
1939	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
1940		spa_check_removed(spa->spa_root_vdev);
1941		/*
1942		 * For the import case, this is done in spa_import(), because
1943		 * at this point we're using the spare definitions from
1944		 * the MOS config, not necessarily from the userland config.
1945		 */
1946		if (state != SPA_LOAD_IMPORT) {
1947			spa_aux_check_removed(&spa->spa_spares);
1948			spa_aux_check_removed(&spa->spa_l2cache);
1949		}
1950	}
1951
1952	/*
1953	 * Load the vdev state for all toplevel vdevs.
1954	 */
1955	vdev_load(rvd);
1956
1957	/*
1958	 * Propagate the leaf DTLs we just loaded all the way up the tree.
1959	 */
1960	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1961	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
1962	spa_config_exit(spa, SCL_ALL, FTAG);
1963
1964	/*
1965	 * Check the state of the root vdev.  If it can't be opened, it
1966	 * indicates one or more toplevel vdevs are faulted.
1967	 */
1968	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1969		return (ENXIO);
1970
1971	/*
1972	 * Load the DDTs (dedup tables).
1973	 */
1974	error = ddt_load(spa);
1975	if (error != 0)
1976		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1977
1978	spa_update_dspace(spa);
1979
1980	if (state != SPA_LOAD_TRYIMPORT) {
1981		error = spa_load_verify(spa);
1982		if (error)
1983			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
1984			    error));
1985	}
1986
1987	/*
1988	 * Load the intent log state and check log integrity.  If we're
1989	 * assembling a pool from a split, the log is not transferred over.
1990	 */
1991	if (type != SPA_IMPORT_ASSEMBLE) {
1992		VERIFY(nvlist_lookup_nvlist(nvconfig, ZPOOL_CONFIG_VDEV_TREE,
1993		    &nvroot) == 0);
1994		spa_load_log_state(spa, nvroot);
1995		nvlist_free(nvconfig);
1996
1997		if (spa_check_logs(spa)) {
1998			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
1999			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2000		}
2001	}
2002
2003	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2004	    spa->spa_load_max_txg == UINT64_MAX)) {
2005		dmu_tx_t *tx;
2006		int need_update = B_FALSE;
2007
2008		ASSERT(state != SPA_LOAD_TRYIMPORT);
2009
2010		/*
2011		 * Claim log blocks that haven't been committed yet.
2012		 * This must all happen in a single txg.
2013		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2014		 * invoked from zil_claim_log_block()'s i/o done callback.
2015		 * Price of rollback is that we abandon the log.
2016		 */
2017		spa->spa_claiming = B_TRUE;
2018
2019		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2020		    spa_first_txg(spa));
2021		(void) dmu_objset_find(spa_name(spa),
2022		    zil_claim, tx, DS_FIND_CHILDREN);
2023		dmu_tx_commit(tx);
2024
2025		spa->spa_claiming = B_FALSE;
2026
2027		spa_set_log_state(spa, SPA_LOG_GOOD);
2028		spa->spa_sync_on = B_TRUE;
2029		txg_sync_start(spa->spa_dsl_pool);
2030
2031		/*
2032		 * Wait for all claims to sync.  We sync up to the highest
2033		 * claimed log block birth time so that claimed log blocks
2034		 * don't appear to be from the future.  spa_claim_max_txg
2035		 * will have been set for us by either zil_check_log_chain()
2036		 * (invoked from spa_check_logs()) or zil_claim() above.
2037		 */
2038		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2039
2040		/*
2041		 * If the config cache is stale, or we have uninitialized
2042		 * metaslabs (see spa_vdev_add()), then update the config.
2043		 *
2044		 * If spa_load_verbatim is true, trust the current
2045		 * in-core spa_config and update the disk labels.
2046		 */
2047		if (config_cache_txg != spa->spa_config_txg ||
2048		    state == SPA_LOAD_IMPORT || spa->spa_load_verbatim ||
2049		    state == SPA_LOAD_RECOVER)
2050			need_update = B_TRUE;
2051
2052		for (int c = 0; c < rvd->vdev_children; c++)
2053			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2054				need_update = B_TRUE;
2055
2056		/*
2057		 * Update the config cache asychronously in case we're the
2058		 * root pool, in which case the config cache isn't writable yet.
2059		 */
2060		if (need_update)
2061			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2062
2063		/*
2064		 * Check all DTLs to see if anything needs resilvering.
2065		 */
2066		if (vdev_resilver_needed(rvd, NULL, NULL))
2067			spa_async_request(spa, SPA_ASYNC_RESILVER);
2068
2069		/*
2070		 * Delete any inconsistent datasets.
2071		 */
2072		(void) dmu_objset_find(spa_name(spa),
2073		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2074
2075		/*
2076		 * Clean up any stale temporary dataset userrefs.
2077		 */
2078		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2079	}
2080
2081	return (0);
2082}
2083
2084static int
2085spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2086{
2087	spa_unload(spa);
2088	spa_deactivate(spa);
2089
2090	spa->spa_load_max_txg--;
2091
2092	spa_activate(spa, spa_mode_global);
2093	spa_async_suspend(spa);
2094
2095	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2096}
2097
2098static int
2099spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2100    uint64_t max_request, boolean_t extreme)
2101{
2102	nvlist_t *config = NULL;
2103	int load_error, rewind_error;
2104	uint64_t safe_rollback_txg;
2105	uint64_t min_txg;
2106
2107	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2108		spa->spa_load_max_txg = spa->spa_load_txg;
2109		spa_set_log_state(spa, SPA_LOG_CLEAR);
2110	} else {
2111		spa->spa_load_max_txg = max_request;
2112	}
2113
2114	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2115	    mosconfig);
2116	if (load_error == 0)
2117		return (0);
2118
2119	if (spa->spa_root_vdev != NULL)
2120		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2121
2122	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2123	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2124
2125	/* specific txg requested */
2126	if (spa->spa_load_max_txg != UINT64_MAX && !extreme) {
2127		nvlist_free(config);
2128		return (load_error);
2129	}
2130
2131	/* Price of rolling back is discarding txgs, including log */
2132	if (state == SPA_LOAD_RECOVER)
2133		spa_set_log_state(spa, SPA_LOG_CLEAR);
2134
2135	spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2136	safe_rollback_txg = spa->spa_uberblock.ub_txg - TXG_DEFER_SIZE;
2137
2138	min_txg = extreme ? TXG_INITIAL : safe_rollback_txg;
2139	while (rewind_error && (spa->spa_uberblock.ub_txg >= min_txg)) {
2140		if (spa->spa_load_max_txg < safe_rollback_txg)
2141			spa->spa_extreme_rewind = B_TRUE;
2142		rewind_error = spa_load_retry(spa, state, mosconfig);
2143	}
2144
2145	if (config)
2146		spa_rewind_data_to_nvlist(spa, config);
2147
2148	spa->spa_extreme_rewind = B_FALSE;
2149	spa->spa_load_max_txg = UINT64_MAX;
2150
2151	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2152		spa_config_set(spa, config);
2153
2154	return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
2155}
2156
2157/*
2158 * Pool Open/Import
2159 *
2160 * The import case is identical to an open except that the configuration is sent
2161 * down from userland, instead of grabbed from the configuration cache.  For the
2162 * case of an open, the pool configuration will exist in the
2163 * POOL_STATE_UNINITIALIZED state.
2164 *
2165 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2166 * the same time open the pool, without having to keep around the spa_t in some
2167 * ambiguous state.
2168 */
2169static int
2170spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2171    nvlist_t **config)
2172{
2173	spa_t *spa;
2174	boolean_t norewind;
2175	boolean_t extreme;
2176	zpool_rewind_policy_t policy;
2177	spa_load_state_t state = SPA_LOAD_OPEN;
2178	int error;
2179	int locked = B_FALSE;
2180
2181	*spapp = NULL;
2182
2183	zpool_get_rewind_policy(nvpolicy, &policy);
2184	if (policy.zrp_request & ZPOOL_DO_REWIND)
2185		state = SPA_LOAD_RECOVER;
2186	norewind = (policy.zrp_request == ZPOOL_NO_REWIND);
2187	extreme = ((policy.zrp_request & ZPOOL_EXTREME_REWIND) != 0);
2188
2189	/*
2190	 * As disgusting as this is, we need to support recursive calls to this
2191	 * function because dsl_dir_open() is called during spa_load(), and ends
2192	 * up calling spa_open() again.  The real fix is to figure out how to
2193	 * avoid dsl_dir_open() calling this in the first place.
2194	 */
2195	if (mutex_owner(&spa_namespace_lock) != curthread) {
2196		mutex_enter(&spa_namespace_lock);
2197		locked = B_TRUE;
2198	}
2199
2200	if ((spa = spa_lookup(pool)) == NULL) {
2201		if (locked)
2202			mutex_exit(&spa_namespace_lock);
2203		return (ENOENT);
2204	}
2205
2206	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2207
2208		spa_activate(spa, spa_mode_global);
2209
2210		if (spa->spa_last_open_failed && norewind) {
2211			if (config != NULL && spa->spa_config)
2212				VERIFY(nvlist_dup(spa->spa_config,
2213				    config, KM_SLEEP) == 0);
2214			spa_deactivate(spa);
2215			if (locked)
2216				mutex_exit(&spa_namespace_lock);
2217			return (spa->spa_last_open_failed);
2218		}
2219
2220		if (state != SPA_LOAD_RECOVER)
2221			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2222
2223		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2224		    extreme);
2225
2226		if (error == EBADF) {
2227			/*
2228			 * If vdev_validate() returns failure (indicated by
2229			 * EBADF), it indicates that one of the vdevs indicates
2230			 * that the pool has been exported or destroyed.  If
2231			 * this is the case, the config cache is out of sync and
2232			 * we should remove the pool from the namespace.
2233			 */
2234			spa_unload(spa);
2235			spa_deactivate(spa);
2236			spa_config_sync(spa, B_TRUE, B_TRUE);
2237			spa_remove(spa);
2238			if (locked)
2239				mutex_exit(&spa_namespace_lock);
2240			return (ENOENT);
2241		}
2242
2243		if (error) {
2244			/*
2245			 * We can't open the pool, but we still have useful
2246			 * information: the state of each vdev after the
2247			 * attempted vdev_open().  Return this to the user.
2248			 */
2249			if (config != NULL && spa->spa_config)
2250				VERIFY(nvlist_dup(spa->spa_config, config,
2251				    KM_SLEEP) == 0);
2252			spa_unload(spa);
2253			spa_deactivate(spa);
2254			spa->spa_last_open_failed = error;
2255			if (locked)
2256				mutex_exit(&spa_namespace_lock);
2257			*spapp = NULL;
2258			return (error);
2259		}
2260
2261	}
2262
2263	spa_open_ref(spa, tag);
2264
2265
2266	if (config != NULL)
2267		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2268
2269	if (locked) {
2270		spa->spa_last_open_failed = 0;
2271		spa->spa_last_ubsync_txg = 0;
2272		spa->spa_load_txg = 0;
2273		mutex_exit(&spa_namespace_lock);
2274	}
2275
2276	*spapp = spa;
2277
2278	return (0);
2279}
2280
2281int
2282spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2283    nvlist_t **config)
2284{
2285	return (spa_open_common(name, spapp, tag, policy, config));
2286}
2287
2288int
2289spa_open(const char *name, spa_t **spapp, void *tag)
2290{
2291	return (spa_open_common(name, spapp, tag, NULL, NULL));
2292}
2293
2294/*
2295 * Lookup the given spa_t, incrementing the inject count in the process,
2296 * preventing it from being exported or destroyed.
2297 */
2298spa_t *
2299spa_inject_addref(char *name)
2300{
2301	spa_t *spa;
2302
2303	mutex_enter(&spa_namespace_lock);
2304	if ((spa = spa_lookup(name)) == NULL) {
2305		mutex_exit(&spa_namespace_lock);
2306		return (NULL);
2307	}
2308	spa->spa_inject_ref++;
2309	mutex_exit(&spa_namespace_lock);
2310
2311	return (spa);
2312}
2313
2314void
2315spa_inject_delref(spa_t *spa)
2316{
2317	mutex_enter(&spa_namespace_lock);
2318	spa->spa_inject_ref--;
2319	mutex_exit(&spa_namespace_lock);
2320}
2321
2322/*
2323 * Add spares device information to the nvlist.
2324 */
2325static void
2326spa_add_spares(spa_t *spa, nvlist_t *config)
2327{
2328	nvlist_t **spares;
2329	uint_t i, nspares;
2330	nvlist_t *nvroot;
2331	uint64_t guid;
2332	vdev_stat_t *vs;
2333	uint_t vsc;
2334	uint64_t pool;
2335
2336	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2337
2338	if (spa->spa_spares.sav_count == 0)
2339		return;
2340
2341	VERIFY(nvlist_lookup_nvlist(config,
2342	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2343	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2344	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2345	if (nspares != 0) {
2346		VERIFY(nvlist_add_nvlist_array(nvroot,
2347		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2348		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2349		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2350
2351		/*
2352		 * Go through and find any spares which have since been
2353		 * repurposed as an active spare.  If this is the case, update
2354		 * their status appropriately.
2355		 */
2356		for (i = 0; i < nspares; i++) {
2357			VERIFY(nvlist_lookup_uint64(spares[i],
2358			    ZPOOL_CONFIG_GUID, &guid) == 0);
2359			if (spa_spare_exists(guid, &pool, NULL) &&
2360			    pool != 0ULL) {
2361				VERIFY(nvlist_lookup_uint64_array(
2362				    spares[i], ZPOOL_CONFIG_STATS,
2363				    (uint64_t **)&vs, &vsc) == 0);
2364				vs->vs_state = VDEV_STATE_CANT_OPEN;
2365				vs->vs_aux = VDEV_AUX_SPARED;
2366			}
2367		}
2368	}
2369}
2370
2371/*
2372 * Add l2cache device information to the nvlist, including vdev stats.
2373 */
2374static void
2375spa_add_l2cache(spa_t *spa, nvlist_t *config)
2376{
2377	nvlist_t **l2cache;
2378	uint_t i, j, nl2cache;
2379	nvlist_t *nvroot;
2380	uint64_t guid;
2381	vdev_t *vd;
2382	vdev_stat_t *vs;
2383	uint_t vsc;
2384
2385	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2386
2387	if (spa->spa_l2cache.sav_count == 0)
2388		return;
2389
2390	VERIFY(nvlist_lookup_nvlist(config,
2391	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2392	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2393	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2394	if (nl2cache != 0) {
2395		VERIFY(nvlist_add_nvlist_array(nvroot,
2396		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2397		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2398		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2399
2400		/*
2401		 * Update level 2 cache device stats.
2402		 */
2403
2404		for (i = 0; i < nl2cache; i++) {
2405			VERIFY(nvlist_lookup_uint64(l2cache[i],
2406			    ZPOOL_CONFIG_GUID, &guid) == 0);
2407
2408			vd = NULL;
2409			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2410				if (guid ==
2411				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2412					vd = spa->spa_l2cache.sav_vdevs[j];
2413					break;
2414				}
2415			}
2416			ASSERT(vd != NULL);
2417
2418			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2419			    ZPOOL_CONFIG_STATS, (uint64_t **)&vs, &vsc) == 0);
2420			vdev_get_stats(vd, vs);
2421		}
2422	}
2423}
2424
2425int
2426spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2427{
2428	int error;
2429	spa_t *spa;
2430
2431	*config = NULL;
2432	error = spa_open_common(name, &spa, FTAG, NULL, config);
2433
2434	if (spa != NULL) {
2435		/*
2436		 * This still leaves a window of inconsistency where the spares
2437		 * or l2cache devices could change and the config would be
2438		 * self-inconsistent.
2439		 */
2440		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2441
2442		if (*config != NULL) {
2443			VERIFY(nvlist_add_uint64(*config,
2444			    ZPOOL_CONFIG_ERRCOUNT,
2445			    spa_get_errlog_size(spa)) == 0);
2446
2447			if (spa_suspended(spa))
2448				VERIFY(nvlist_add_uint64(*config,
2449				    ZPOOL_CONFIG_SUSPENDED,
2450				    spa->spa_failmode) == 0);
2451
2452			spa_add_spares(spa, *config);
2453			spa_add_l2cache(spa, *config);
2454		}
2455	}
2456
2457	/*
2458	 * We want to get the alternate root even for faulted pools, so we cheat
2459	 * and call spa_lookup() directly.
2460	 */
2461	if (altroot) {
2462		if (spa == NULL) {
2463			mutex_enter(&spa_namespace_lock);
2464			spa = spa_lookup(name);
2465			if (spa)
2466				spa_altroot(spa, altroot, buflen);
2467			else
2468				altroot[0] = '\0';
2469			spa = NULL;
2470			mutex_exit(&spa_namespace_lock);
2471		} else {
2472			spa_altroot(spa, altroot, buflen);
2473		}
2474	}
2475
2476	if (spa != NULL) {
2477		spa_config_exit(spa, SCL_CONFIG, FTAG);
2478		spa_close(spa, FTAG);
2479	}
2480
2481	return (error);
2482}
2483
2484/*
2485 * Validate that the auxiliary device array is well formed.  We must have an
2486 * array of nvlists, each which describes a valid leaf vdev.  If this is an
2487 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2488 * specified, as long as they are well-formed.
2489 */
2490static int
2491spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2492    spa_aux_vdev_t *sav, const char *config, uint64_t version,
2493    vdev_labeltype_t label)
2494{
2495	nvlist_t **dev;
2496	uint_t i, ndev;
2497	vdev_t *vd;
2498	int error;
2499
2500	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2501
2502	/*
2503	 * It's acceptable to have no devs specified.
2504	 */
2505	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2506		return (0);
2507
2508	if (ndev == 0)
2509		return (EINVAL);
2510
2511	/*
2512	 * Make sure the pool is formatted with a version that supports this
2513	 * device type.
2514	 */
2515	if (spa_version(spa) < version)
2516		return (ENOTSUP);
2517
2518	/*
2519	 * Set the pending device list so we correctly handle device in-use
2520	 * checking.
2521	 */
2522	sav->sav_pending = dev;
2523	sav->sav_npending = ndev;
2524
2525	for (i = 0; i < ndev; i++) {
2526		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2527		    mode)) != 0)
2528			goto out;
2529
2530		if (!vd->vdev_ops->vdev_op_leaf) {
2531			vdev_free(vd);
2532			error = EINVAL;
2533			goto out;
2534		}
2535
2536		/*
2537		 * The L2ARC currently only supports disk devices in
2538		 * kernel context.  For user-level testing, we allow it.
2539		 */
2540#ifdef _KERNEL
2541		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2542		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2543			error = ENOTBLK;
2544			goto out;
2545		}
2546#endif
2547		vd->vdev_top = vd;
2548
2549		if ((error = vdev_open(vd)) == 0 &&
2550		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
2551			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2552			    vd->vdev_guid) == 0);
2553		}
2554
2555		vdev_free(vd);
2556
2557		if (error &&
2558		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2559			goto out;
2560		else
2561			error = 0;
2562	}
2563
2564out:
2565	sav->sav_pending = NULL;
2566	sav->sav_npending = 0;
2567	return (error);
2568}
2569
2570static int
2571spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2572{
2573	int error;
2574
2575	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2576
2577	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2578	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2579	    VDEV_LABEL_SPARE)) != 0) {
2580		return (error);
2581	}
2582
2583	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2584	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2585	    VDEV_LABEL_L2CACHE));
2586}
2587
2588static void
2589spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2590    const char *config)
2591{
2592	int i;
2593
2594	if (sav->sav_config != NULL) {
2595		nvlist_t **olddevs;
2596		uint_t oldndevs;
2597		nvlist_t **newdevs;
2598
2599		/*
2600		 * Generate new dev list by concatentating with the
2601		 * current dev list.
2602		 */
2603		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2604		    &olddevs, &oldndevs) == 0);
2605
2606		newdevs = kmem_alloc(sizeof (void *) *
2607		    (ndevs + oldndevs), KM_SLEEP);
2608		for (i = 0; i < oldndevs; i++)
2609			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2610			    KM_SLEEP) == 0);
2611		for (i = 0; i < ndevs; i++)
2612			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2613			    KM_SLEEP) == 0);
2614
2615		VERIFY(nvlist_remove(sav->sav_config, config,
2616		    DATA_TYPE_NVLIST_ARRAY) == 0);
2617
2618		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2619		    config, newdevs, ndevs + oldndevs) == 0);
2620		for (i = 0; i < oldndevs + ndevs; i++)
2621			nvlist_free(newdevs[i]);
2622		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2623	} else {
2624		/*
2625		 * Generate a new dev list.
2626		 */
2627		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2628		    KM_SLEEP) == 0);
2629		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2630		    devs, ndevs) == 0);
2631	}
2632}
2633
2634/*
2635 * Stop and drop level 2 ARC devices
2636 */
2637void
2638spa_l2cache_drop(spa_t *spa)
2639{
2640	vdev_t *vd;
2641	int i;
2642	spa_aux_vdev_t *sav = &spa->spa_l2cache;
2643
2644	for (i = 0; i < sav->sav_count; i++) {
2645		uint64_t pool;
2646
2647		vd = sav->sav_vdevs[i];
2648		ASSERT(vd != NULL);
2649
2650		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2651		    pool != 0ULL && l2arc_vdev_present(vd))
2652			l2arc_remove_vdev(vd);
2653		if (vd->vdev_isl2cache)
2654			spa_l2cache_remove(vd);
2655		vdev_clear_stats(vd);
2656		(void) vdev_close(vd);
2657	}
2658}
2659
2660/*
2661 * Pool Creation
2662 */
2663int
2664spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2665    const char *history_str, nvlist_t *zplprops)
2666{
2667	spa_t *spa;
2668	char *altroot = NULL;
2669	vdev_t *rvd;
2670	dsl_pool_t *dp;
2671	dmu_tx_t *tx;
2672	int error = 0;
2673	uint64_t txg = TXG_INITIAL;
2674	nvlist_t **spares, **l2cache;
2675	uint_t nspares, nl2cache;
2676	uint64_t version;
2677
2678	/*
2679	 * If this pool already exists, return failure.
2680	 */
2681	mutex_enter(&spa_namespace_lock);
2682	if (spa_lookup(pool) != NULL) {
2683		mutex_exit(&spa_namespace_lock);
2684		return (EEXIST);
2685	}
2686
2687	/*
2688	 * Allocate a new spa_t structure.
2689	 */
2690	(void) nvlist_lookup_string(props,
2691	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2692	spa = spa_add(pool, NULL, altroot);
2693	spa_activate(spa, spa_mode_global);
2694
2695	if (props && (error = spa_prop_validate(spa, props))) {
2696		spa_deactivate(spa);
2697		spa_remove(spa);
2698		mutex_exit(&spa_namespace_lock);
2699		return (error);
2700	}
2701
2702	if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2703	    &version) != 0)
2704		version = SPA_VERSION;
2705	ASSERT(version <= SPA_VERSION);
2706
2707	spa->spa_first_txg = txg;
2708	spa->spa_uberblock.ub_txg = txg - 1;
2709	spa->spa_uberblock.ub_version = version;
2710	spa->spa_ubsync = spa->spa_uberblock;
2711
2712	/*
2713	 * Create "The Godfather" zio to hold all async IOs
2714	 */
2715	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2716	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2717
2718	/*
2719	 * Create the root vdev.
2720	 */
2721	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2722
2723	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2724
2725	ASSERT(error != 0 || rvd != NULL);
2726	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2727
2728	if (error == 0 && !zfs_allocatable_devs(nvroot))
2729		error = EINVAL;
2730
2731	if (error == 0 &&
2732	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2733	    (error = spa_validate_aux(spa, nvroot, txg,
2734	    VDEV_ALLOC_ADD)) == 0) {
2735		for (int c = 0; c < rvd->vdev_children; c++) {
2736			vdev_metaslab_set_size(rvd->vdev_child[c]);
2737			vdev_expand(rvd->vdev_child[c], txg);
2738		}
2739	}
2740
2741	spa_config_exit(spa, SCL_ALL, FTAG);
2742
2743	if (error != 0) {
2744		spa_unload(spa);
2745		spa_deactivate(spa);
2746		spa_remove(spa);
2747		mutex_exit(&spa_namespace_lock);
2748		return (error);
2749	}
2750
2751	/*
2752	 * Get the list of spares, if specified.
2753	 */
2754	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2755	    &spares, &nspares) == 0) {
2756		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
2757		    KM_SLEEP) == 0);
2758		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2759		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2760		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2761		spa_load_spares(spa);
2762		spa_config_exit(spa, SCL_ALL, FTAG);
2763		spa->spa_spares.sav_sync = B_TRUE;
2764	}
2765
2766	/*
2767	 * Get the list of level 2 cache devices, if specified.
2768	 */
2769	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2770	    &l2cache, &nl2cache) == 0) {
2771		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2772		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2773		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2774		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2775		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2776		spa_load_l2cache(spa);
2777		spa_config_exit(spa, SCL_ALL, FTAG);
2778		spa->spa_l2cache.sav_sync = B_TRUE;
2779	}
2780
2781	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
2782	spa->spa_meta_objset = dp->dp_meta_objset;
2783
2784	/*
2785	 * Create DDTs (dedup tables).
2786	 */
2787	ddt_create(spa);
2788
2789	spa_update_dspace(spa);
2790
2791	tx = dmu_tx_create_assigned(dp, txg);
2792
2793	/*
2794	 * Create the pool config object.
2795	 */
2796	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
2797	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
2798	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2799
2800	if (zap_add(spa->spa_meta_objset,
2801	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
2802	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
2803		cmn_err(CE_PANIC, "failed to add pool config");
2804	}
2805
2806	/* Newly created pools with the right version are always deflated. */
2807	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
2808		spa->spa_deflate = TRUE;
2809		if (zap_add(spa->spa_meta_objset,
2810		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
2811		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
2812			cmn_err(CE_PANIC, "failed to add deflate");
2813		}
2814	}
2815
2816	/*
2817	 * Create the deferred-free bplist object.  Turn off compression
2818	 * because sync-to-convergence takes longer if the blocksize
2819	 * keeps changing.
2820	 */
2821	spa->spa_deferred_bplist_obj = bplist_create(spa->spa_meta_objset,
2822	    1 << 14, tx);
2823	dmu_object_set_compress(spa->spa_meta_objset,
2824	    spa->spa_deferred_bplist_obj, ZIO_COMPRESS_OFF, tx);
2825
2826	if (zap_add(spa->spa_meta_objset,
2827	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
2828	    sizeof (uint64_t), 1, &spa->spa_deferred_bplist_obj, tx) != 0) {
2829		cmn_err(CE_PANIC, "failed to add bplist");
2830	}
2831
2832	/*
2833	 * Create the pool's history object.
2834	 */
2835	if (version >= SPA_VERSION_ZPOOL_HISTORY)
2836		spa_history_create_obj(spa, tx);
2837
2838	/*
2839	 * Set pool properties.
2840	 */
2841	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
2842	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2843	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
2844	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
2845
2846	if (props != NULL) {
2847		spa_configfile_set(spa, props, B_FALSE);
2848		spa_sync_props(spa, props, CRED(), tx);
2849	}
2850
2851	dmu_tx_commit(tx);
2852
2853	spa->spa_sync_on = B_TRUE;
2854	txg_sync_start(spa->spa_dsl_pool);
2855
2856	/*
2857	 * We explicitly wait for the first transaction to complete so that our
2858	 * bean counters are appropriately updated.
2859	 */
2860	txg_wait_synced(spa->spa_dsl_pool, txg);
2861
2862	spa_config_sync(spa, B_FALSE, B_TRUE);
2863
2864	if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
2865		(void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
2866	spa_history_log_version(spa, LOG_POOL_CREATE);
2867
2868	spa->spa_minref = refcount_count(&spa->spa_refcount);
2869
2870	mutex_exit(&spa_namespace_lock);
2871
2872	return (0);
2873}
2874
2875#ifdef _KERNEL
2876/*
2877 * Get the root pool information from the root disk, then import the root pool
2878 * during the system boot up time.
2879 */
2880extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
2881
2882static nvlist_t *
2883spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
2884{
2885	nvlist_t *config;
2886	nvlist_t *nvtop, *nvroot;
2887	uint64_t pgid;
2888
2889	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
2890		return (NULL);
2891
2892	/*
2893	 * Add this top-level vdev to the child array.
2894	 */
2895	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2896	    &nvtop) == 0);
2897	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
2898	    &pgid) == 0);
2899	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
2900
2901	/*
2902	 * Put this pool's top-level vdevs into a root vdev.
2903	 */
2904	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2905	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
2906	    VDEV_TYPE_ROOT) == 0);
2907	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
2908	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
2909	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2910	    &nvtop, 1) == 0);
2911
2912	/*
2913	 * Replace the existing vdev_tree with the new root vdev in
2914	 * this pool's configuration (remove the old, add the new).
2915	 */
2916	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
2917	nvlist_free(nvroot);
2918	return (config);
2919}
2920
2921/*
2922 * Walk the vdev tree and see if we can find a device with "better"
2923 * configuration. A configuration is "better" if the label on that
2924 * device has a more recent txg.
2925 */
2926static void
2927spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
2928{
2929	for (int c = 0; c < vd->vdev_children; c++)
2930		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
2931
2932	if (vd->vdev_ops->vdev_op_leaf) {
2933		nvlist_t *label;
2934		uint64_t label_txg;
2935
2936		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
2937		    &label) != 0)
2938			return;
2939
2940		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
2941		    &label_txg) == 0);
2942
2943		/*
2944		 * Do we have a better boot device?
2945		 */
2946		if (label_txg > *txg) {
2947			*txg = label_txg;
2948			*avd = vd;
2949		}
2950		nvlist_free(label);
2951	}
2952}
2953
2954/*
2955 * Import a root pool.
2956 *
2957 * For x86. devpath_list will consist of devid and/or physpath name of
2958 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
2959 * The GRUB "findroot" command will return the vdev we should boot.
2960 *
2961 * For Sparc, devpath_list consists the physpath name of the booting device
2962 * no matter the rootpool is a single device pool or a mirrored pool.
2963 * e.g.
2964 *	"/pci@1f,0/ide@d/disk@0,0:a"
2965 */
2966int
2967spa_import_rootpool(char *devpath, char *devid)
2968{
2969	spa_t *spa;
2970	vdev_t *rvd, *bvd, *avd = NULL;
2971	nvlist_t *config, *nvtop;
2972	uint64_t guid, txg;
2973	char *pname;
2974	int error;
2975
2976	/*
2977	 * Read the label from the boot device and generate a configuration.
2978	 */
2979	config = spa_generate_rootconf(devpath, devid, &guid);
2980#if defined(_OBP) && defined(_KERNEL)
2981	if (config == NULL) {
2982		if (strstr(devpath, "/iscsi/ssd") != NULL) {
2983			/* iscsi boot */
2984			get_iscsi_bootpath_phy(devpath);
2985			config = spa_generate_rootconf(devpath, devid, &guid);
2986		}
2987	}
2988#endif
2989	if (config == NULL) {
2990		cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
2991		    devpath);
2992		return (EIO);
2993	}
2994
2995	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
2996	    &pname) == 0);
2997	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
2998
2999	mutex_enter(&spa_namespace_lock);
3000	if ((spa = spa_lookup(pname)) != NULL) {
3001		/*
3002		 * Remove the existing root pool from the namespace so that we
3003		 * can replace it with the correct config we just read in.
3004		 */
3005		spa_remove(spa);
3006	}
3007
3008	spa = spa_add(pname, config, NULL);
3009	spa->spa_is_root = B_TRUE;
3010	spa->spa_load_verbatim = B_TRUE;
3011
3012	/*
3013	 * Build up a vdev tree based on the boot device's label config.
3014	 */
3015	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3016	    &nvtop) == 0);
3017	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3018	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3019	    VDEV_ALLOC_ROOTPOOL);
3020	spa_config_exit(spa, SCL_ALL, FTAG);
3021	if (error) {
3022		mutex_exit(&spa_namespace_lock);
3023		nvlist_free(config);
3024		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3025		    pname);
3026		return (error);
3027	}
3028
3029	/*
3030	 * Get the boot vdev.
3031	 */
3032	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3033		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3034		    (u_longlong_t)guid);
3035		error = ENOENT;
3036		goto out;
3037	}
3038
3039	/*
3040	 * Determine if there is a better boot device.
3041	 */
3042	avd = bvd;
3043	spa_alt_rootvdev(rvd, &avd, &txg);
3044	if (avd != bvd) {
3045		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3046		    "try booting from '%s'", avd->vdev_path);
3047		error = EINVAL;
3048		goto out;
3049	}
3050
3051	/*
3052	 * If the boot device is part of a spare vdev then ensure that
3053	 * we're booting off the active spare.
3054	 */
3055	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3056	    !bvd->vdev_isspare) {
3057		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3058		    "try booting from '%s'",
3059		    bvd->vdev_parent->vdev_child[1]->vdev_path);
3060		error = EINVAL;
3061		goto out;
3062	}
3063
3064	error = 0;
3065	spa_history_log_version(spa, LOG_POOL_IMPORT);
3066out:
3067	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3068	vdev_free(rvd);
3069	spa_config_exit(spa, SCL_ALL, FTAG);
3070	mutex_exit(&spa_namespace_lock);
3071
3072	nvlist_free(config);
3073	return (error);
3074}
3075
3076#endif
3077
3078/*
3079 * Take a pool and insert it into the namespace as if it had been loaded at
3080 * boot.
3081 */
3082int
3083spa_import_verbatim(const char *pool, nvlist_t *config, nvlist_t *props)
3084{
3085	spa_t *spa;
3086	zpool_rewind_policy_t policy;
3087	char *altroot = NULL;
3088
3089	mutex_enter(&spa_namespace_lock);
3090	if (spa_lookup(pool) != NULL) {
3091		mutex_exit(&spa_namespace_lock);
3092		return (EEXIST);
3093	}
3094
3095	(void) nvlist_lookup_string(props,
3096	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3097	spa = spa_add(pool, config, altroot);
3098
3099	zpool_get_rewind_policy(config, &policy);
3100	spa->spa_load_max_txg = policy.zrp_txg;
3101
3102	spa->spa_load_verbatim = B_TRUE;
3103
3104	if (props != NULL)
3105		spa_configfile_set(spa, props, B_FALSE);
3106
3107	spa_config_sync(spa, B_FALSE, B_TRUE);
3108
3109	mutex_exit(&spa_namespace_lock);
3110	spa_history_log_version(spa, LOG_POOL_IMPORT);
3111
3112	return (0);
3113}
3114
3115/*
3116 * Import a non-root pool into the system.
3117 */
3118int
3119spa_import(const char *pool, nvlist_t *config, nvlist_t *props)
3120{
3121	spa_t *spa;
3122	char *altroot = NULL;
3123	spa_load_state_t state = SPA_LOAD_IMPORT;
3124	zpool_rewind_policy_t policy;
3125	int error;
3126	nvlist_t *nvroot;
3127	nvlist_t **spares, **l2cache;
3128	uint_t nspares, nl2cache;
3129
3130	/*
3131	 * If a pool with this name exists, return failure.
3132	 */
3133	mutex_enter(&spa_namespace_lock);
3134	if (spa_lookup(pool) != NULL) {
3135		mutex_exit(&spa_namespace_lock);
3136		return (EEXIST);
3137	}
3138
3139	zpool_get_rewind_policy(config, &policy);
3140	if (policy.zrp_request & ZPOOL_DO_REWIND)
3141		state = SPA_LOAD_RECOVER;
3142
3143	/*
3144	 * Create and initialize the spa structure.
3145	 */
3146	(void) nvlist_lookup_string(props,
3147	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3148	spa = spa_add(pool, config, altroot);
3149	spa_activate(spa, spa_mode_global);
3150
3151	/*
3152	 * Don't start async tasks until we know everything is healthy.
3153	 */
3154	spa_async_suspend(spa);
3155
3156	/*
3157	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3158	 * because the user-supplied config is actually the one to trust when
3159	 * doing an import.
3160	 */
3161	if (state != SPA_LOAD_RECOVER)
3162		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3163	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3164	    ((policy.zrp_request & ZPOOL_EXTREME_REWIND) != 0));
3165
3166	/*
3167	 * Propagate anything learned about failing or best txgs
3168	 * back to caller
3169	 */
3170	spa_rewind_data_to_nvlist(spa, config);
3171
3172	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3173	/*
3174	 * Toss any existing sparelist, as it doesn't have any validity
3175	 * anymore, and conflicts with spa_has_spare().
3176	 */
3177	if (spa->spa_spares.sav_config) {
3178		nvlist_free(spa->spa_spares.sav_config);
3179		spa->spa_spares.sav_config = NULL;
3180		spa_load_spares(spa);
3181	}
3182	if (spa->spa_l2cache.sav_config) {
3183		nvlist_free(spa->spa_l2cache.sav_config);
3184		spa->spa_l2cache.sav_config = NULL;
3185		spa_load_l2cache(spa);
3186	}
3187
3188	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3189	    &nvroot) == 0);
3190	if (error == 0)
3191		error = spa_validate_aux(spa, nvroot, -1ULL,
3192		    VDEV_ALLOC_SPARE);
3193	if (error == 0)
3194		error = spa_validate_aux(spa, nvroot, -1ULL,
3195		    VDEV_ALLOC_L2CACHE);
3196	spa_config_exit(spa, SCL_ALL, FTAG);
3197
3198	if (props != NULL)
3199		spa_configfile_set(spa, props, B_FALSE);
3200
3201	if (error != 0 || (props && spa_writeable(spa) &&
3202	    (error = spa_prop_set(spa, props)))) {
3203		spa_unload(spa);
3204		spa_deactivate(spa);
3205		spa_remove(spa);
3206		mutex_exit(&spa_namespace_lock);
3207		return (error);
3208	}
3209
3210	spa_async_resume(spa);
3211
3212	/*
3213	 * Override any spares and level 2 cache devices as specified by
3214	 * the user, as these may have correct device names/devids, etc.
3215	 */
3216	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3217	    &spares, &nspares) == 0) {
3218		if (spa->spa_spares.sav_config)
3219			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3220			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3221		else
3222			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3223			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3224		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3225		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3226		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3227		spa_load_spares(spa);
3228		spa_config_exit(spa, SCL_ALL, FTAG);
3229		spa->spa_spares.sav_sync = B_TRUE;
3230	}
3231	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3232	    &l2cache, &nl2cache) == 0) {
3233		if (spa->spa_l2cache.sav_config)
3234			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3235			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3236		else
3237			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3238			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3239		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3240		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3241		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3242		spa_load_l2cache(spa);
3243		spa_config_exit(spa, SCL_ALL, FTAG);
3244		spa->spa_l2cache.sav_sync = B_TRUE;
3245	}
3246
3247	/*
3248	 * Check for any removed devices.
3249	 */
3250	if (spa->spa_autoreplace) {
3251		spa_aux_check_removed(&spa->spa_spares);
3252		spa_aux_check_removed(&spa->spa_l2cache);
3253	}
3254
3255	if (spa_writeable(spa)) {
3256		/*
3257		 * Update the config cache to include the newly-imported pool.
3258		 */
3259		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3260	}
3261
3262	/*
3263	 * It's possible that the pool was expanded while it was exported.
3264	 * We kick off an async task to handle this for us.
3265	 */
3266	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3267
3268	mutex_exit(&spa_namespace_lock);
3269	spa_history_log_version(spa, LOG_POOL_IMPORT);
3270
3271	return (0);
3272}
3273
3274nvlist_t *
3275spa_tryimport(nvlist_t *tryconfig)
3276{
3277	nvlist_t *config = NULL;
3278	char *poolname;
3279	spa_t *spa;
3280	uint64_t state;
3281	int error;
3282
3283	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3284		return (NULL);
3285
3286	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3287		return (NULL);
3288
3289	/*
3290	 * Create and initialize the spa structure.
3291	 */
3292	mutex_enter(&spa_namespace_lock);
3293	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3294	spa_activate(spa, FREAD);
3295
3296	/*
3297	 * Pass off the heavy lifting to spa_load().
3298	 * Pass TRUE for mosconfig because the user-supplied config
3299	 * is actually the one to trust when doing an import.
3300	 */
3301	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3302
3303	/*
3304	 * If 'tryconfig' was at least parsable, return the current config.
3305	 */
3306	if (spa->spa_root_vdev != NULL) {
3307		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3308		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3309		    poolname) == 0);
3310		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3311		    state) == 0);
3312		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3313		    spa->spa_uberblock.ub_timestamp) == 0);
3314
3315		/*
3316		 * If the bootfs property exists on this pool then we
3317		 * copy it out so that external consumers can tell which
3318		 * pools are bootable.
3319		 */
3320		if ((!error || error == EEXIST) && spa->spa_bootfs) {
3321			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3322
3323			/*
3324			 * We have to play games with the name since the
3325			 * pool was opened as TRYIMPORT_NAME.
3326			 */
3327			if (dsl_dsobj_to_dsname(spa_name(spa),
3328			    spa->spa_bootfs, tmpname) == 0) {
3329				char *cp;
3330				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3331
3332				cp = strchr(tmpname, '/');
3333				if (cp == NULL) {
3334					(void) strlcpy(dsname, tmpname,
3335					    MAXPATHLEN);
3336				} else {
3337					(void) snprintf(dsname, MAXPATHLEN,
3338					    "%s/%s", poolname, ++cp);
3339				}
3340				VERIFY(nvlist_add_string(config,
3341				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3342				kmem_free(dsname, MAXPATHLEN);
3343			}
3344			kmem_free(tmpname, MAXPATHLEN);
3345		}
3346
3347		/*
3348		 * Add the list of hot spares and level 2 cache devices.
3349		 */
3350		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3351		spa_add_spares(spa, config);
3352		spa_add_l2cache(spa, config);
3353		spa_config_exit(spa, SCL_CONFIG, FTAG);
3354	}
3355
3356	spa_unload(spa);
3357	spa_deactivate(spa);
3358	spa_remove(spa);
3359	mutex_exit(&spa_namespace_lock);
3360
3361	return (config);
3362}
3363
3364/*
3365 * Pool export/destroy
3366 *
3367 * The act of destroying or exporting a pool is very simple.  We make sure there
3368 * is no more pending I/O and any references to the pool are gone.  Then, we
3369 * update the pool state and sync all the labels to disk, removing the
3370 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3371 * we don't sync the labels or remove the configuration cache.
3372 */
3373static int
3374spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3375    boolean_t force, boolean_t hardforce)
3376{
3377	spa_t *spa;
3378
3379	if (oldconfig)
3380		*oldconfig = NULL;
3381
3382	if (!(spa_mode_global & FWRITE))
3383		return (EROFS);
3384
3385	mutex_enter(&spa_namespace_lock);
3386	if ((spa = spa_lookup(pool)) == NULL) {
3387		mutex_exit(&spa_namespace_lock);
3388		return (ENOENT);
3389	}
3390
3391	/*
3392	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3393	 * reacquire the namespace lock, and see if we can export.
3394	 */
3395	spa_open_ref(spa, FTAG);
3396	mutex_exit(&spa_namespace_lock);
3397	spa_async_suspend(spa);
3398	mutex_enter(&spa_namespace_lock);
3399	spa_close(spa, FTAG);
3400
3401	/*
3402	 * The pool will be in core if it's openable,
3403	 * in which case we can modify its state.
3404	 */
3405	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3406		/*
3407		 * Objsets may be open only because they're dirty, so we
3408		 * have to force it to sync before checking spa_refcnt.
3409		 */
3410		txg_wait_synced(spa->spa_dsl_pool, 0);
3411
3412		/*
3413		 * A pool cannot be exported or destroyed if there are active
3414		 * references.  If we are resetting a pool, allow references by
3415		 * fault injection handlers.
3416		 */
3417		if (!spa_refcount_zero(spa) ||
3418		    (spa->spa_inject_ref != 0 &&
3419		    new_state != POOL_STATE_UNINITIALIZED)) {
3420			spa_async_resume(spa);
3421			mutex_exit(&spa_namespace_lock);
3422			return (EBUSY);
3423		}
3424
3425		/*
3426		 * A pool cannot be exported if it has an active shared spare.
3427		 * This is to prevent other pools stealing the active spare
3428		 * from an exported pool. At user's own will, such pool can
3429		 * be forcedly exported.
3430		 */
3431		if (!force && new_state == POOL_STATE_EXPORTED &&
3432		    spa_has_active_shared_spare(spa)) {
3433			spa_async_resume(spa);
3434			mutex_exit(&spa_namespace_lock);
3435			return (EXDEV);
3436		}
3437
3438		/*
3439		 * We want this to be reflected on every label,
3440		 * so mark them all dirty.  spa_unload() will do the
3441		 * final sync that pushes these changes out.
3442		 */
3443		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3444			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3445			spa->spa_state = new_state;
3446			spa->spa_final_txg = spa_last_synced_txg(spa) + 1;
3447			vdev_config_dirty(spa->spa_root_vdev);
3448			spa_config_exit(spa, SCL_ALL, FTAG);
3449		}
3450	}
3451
3452	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
3453
3454	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3455		spa_unload(spa);
3456		spa_deactivate(spa);
3457	}
3458
3459	if (oldconfig && spa->spa_config)
3460		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3461
3462	if (new_state != POOL_STATE_UNINITIALIZED) {
3463		if (!hardforce)
3464			spa_config_sync(spa, B_TRUE, B_TRUE);
3465		spa_remove(spa);
3466	}
3467	mutex_exit(&spa_namespace_lock);
3468
3469	return (0);
3470}
3471
3472/*
3473 * Destroy a storage pool.
3474 */
3475int
3476spa_destroy(char *pool)
3477{
3478	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3479	    B_FALSE, B_FALSE));
3480}
3481
3482/*
3483 * Export a storage pool.
3484 */
3485int
3486spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3487    boolean_t hardforce)
3488{
3489	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3490	    force, hardforce));
3491}
3492
3493/*
3494 * Similar to spa_export(), this unloads the spa_t without actually removing it
3495 * from the namespace in any way.
3496 */
3497int
3498spa_reset(char *pool)
3499{
3500	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3501	    B_FALSE, B_FALSE));
3502}
3503
3504/*
3505 * ==========================================================================
3506 * Device manipulation
3507 * ==========================================================================
3508 */
3509
3510/*
3511 * Add a device to a storage pool.
3512 */
3513int
3514spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3515{
3516	uint64_t txg, id;
3517	int error;
3518	vdev_t *rvd = spa->spa_root_vdev;
3519	vdev_t *vd, *tvd;
3520	nvlist_t **spares, **l2cache;
3521	uint_t nspares, nl2cache;
3522
3523	txg = spa_vdev_enter(spa);
3524
3525	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3526	    VDEV_ALLOC_ADD)) != 0)
3527		return (spa_vdev_exit(spa, NULL, txg, error));
3528
3529	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
3530
3531	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3532	    &nspares) != 0)
3533		nspares = 0;
3534
3535	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3536	    &nl2cache) != 0)
3537		nl2cache = 0;
3538
3539	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3540		return (spa_vdev_exit(spa, vd, txg, EINVAL));
3541
3542	if (vd->vdev_children != 0 &&
3543	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
3544		return (spa_vdev_exit(spa, vd, txg, error));
3545
3546	/*
3547	 * We must validate the spares and l2cache devices after checking the
3548	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
3549	 */
3550	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3551		return (spa_vdev_exit(spa, vd, txg, error));
3552
3553	/*
3554	 * Transfer each new top-level vdev from vd to rvd.
3555	 */
3556	for (int c = 0; c < vd->vdev_children; c++) {
3557
3558		/*
3559		 * Set the vdev id to the first hole, if one exists.
3560		 */
3561		for (id = 0; id < rvd->vdev_children; id++) {
3562			if (rvd->vdev_child[id]->vdev_ishole) {
3563				vdev_free(rvd->vdev_child[id]);
3564				break;
3565			}
3566		}
3567		tvd = vd->vdev_child[c];
3568		vdev_remove_child(vd, tvd);
3569		tvd->vdev_id = id;
3570		vdev_add_child(rvd, tvd);
3571		vdev_config_dirty(tvd);
3572	}
3573
3574	if (nspares != 0) {
3575		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3576		    ZPOOL_CONFIG_SPARES);
3577		spa_load_spares(spa);
3578		spa->spa_spares.sav_sync = B_TRUE;
3579	}
3580
3581	if (nl2cache != 0) {
3582		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3583		    ZPOOL_CONFIG_L2CACHE);
3584		spa_load_l2cache(spa);
3585		spa->spa_l2cache.sav_sync = B_TRUE;
3586	}
3587
3588	/*
3589	 * We have to be careful when adding new vdevs to an existing pool.
3590	 * If other threads start allocating from these vdevs before we
3591	 * sync the config cache, and we lose power, then upon reboot we may
3592	 * fail to open the pool because there are DVAs that the config cache
3593	 * can't translate.  Therefore, we first add the vdevs without
3594	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3595	 * and then let spa_config_update() initialize the new metaslabs.
3596	 *
3597	 * spa_load() checks for added-but-not-initialized vdevs, so that
3598	 * if we lose power at any point in this sequence, the remaining
3599	 * steps will be completed the next time we load the pool.
3600	 */
3601	(void) spa_vdev_exit(spa, vd, txg, 0);
3602
3603	mutex_enter(&spa_namespace_lock);
3604	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3605	mutex_exit(&spa_namespace_lock);
3606
3607	return (0);
3608}
3609
3610/*
3611 * Attach a device to a mirror.  The arguments are the path to any device
3612 * in the mirror, and the nvroot for the new device.  If the path specifies
3613 * a device that is not mirrored, we automatically insert the mirror vdev.
3614 *
3615 * If 'replacing' is specified, the new device is intended to replace the
3616 * existing device; in this case the two devices are made into their own
3617 * mirror using the 'replacing' vdev, which is functionally identical to
3618 * the mirror vdev (it actually reuses all the same ops) but has a few
3619 * extra rules: you can't attach to it after it's been created, and upon
3620 * completion of resilvering, the first disk (the one being replaced)
3621 * is automatically detached.
3622 */
3623int
3624spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3625{
3626	uint64_t txg, open_txg;
3627	vdev_t *rvd = spa->spa_root_vdev;
3628	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3629	vdev_ops_t *pvops;
3630	char *oldvdpath, *newvdpath;
3631	int newvd_isspare;
3632	int error;
3633
3634	txg = spa_vdev_enter(spa);
3635
3636	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3637
3638	if (oldvd == NULL)
3639		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3640
3641	if (!oldvd->vdev_ops->vdev_op_leaf)
3642		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3643
3644	pvd = oldvd->vdev_parent;
3645
3646	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3647	    VDEV_ALLOC_ADD)) != 0)
3648		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3649
3650	if (newrootvd->vdev_children != 1)
3651		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3652
3653	newvd = newrootvd->vdev_child[0];
3654
3655	if (!newvd->vdev_ops->vdev_op_leaf)
3656		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3657
3658	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3659		return (spa_vdev_exit(spa, newrootvd, txg, error));
3660
3661	/*
3662	 * Spares can't replace logs
3663	 */
3664	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3665		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3666
3667	if (!replacing) {
3668		/*
3669		 * For attach, the only allowable parent is a mirror or the root
3670		 * vdev.
3671		 */
3672		if (pvd->vdev_ops != &vdev_mirror_ops &&
3673		    pvd->vdev_ops != &vdev_root_ops)
3674			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3675
3676		pvops = &vdev_mirror_ops;
3677	} else {
3678		/*
3679		 * Active hot spares can only be replaced by inactive hot
3680		 * spares.
3681		 */
3682		if (pvd->vdev_ops == &vdev_spare_ops &&
3683		    pvd->vdev_child[1] == oldvd &&
3684		    !spa_has_spare(spa, newvd->vdev_guid))
3685			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3686
3687		/*
3688		 * If the source is a hot spare, and the parent isn't already a
3689		 * spare, then we want to create a new hot spare.  Otherwise, we
3690		 * want to create a replacing vdev.  The user is not allowed to
3691		 * attach to a spared vdev child unless the 'isspare' state is
3692		 * the same (spare replaces spare, non-spare replaces
3693		 * non-spare).
3694		 */
3695		if (pvd->vdev_ops == &vdev_replacing_ops)
3696			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3697		else if (pvd->vdev_ops == &vdev_spare_ops &&
3698		    newvd->vdev_isspare != oldvd->vdev_isspare)
3699			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3700		else if (pvd->vdev_ops != &vdev_spare_ops &&
3701		    newvd->vdev_isspare)
3702			pvops = &vdev_spare_ops;
3703		else
3704			pvops = &vdev_replacing_ops;
3705	}
3706
3707	/*
3708	 * Make sure the new device is big enough.
3709	 */
3710	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3711		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3712
3713	/*
3714	 * The new device cannot have a higher alignment requirement
3715	 * than the top-level vdev.
3716	 */
3717	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3718		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3719
3720	/*
3721	 * If this is an in-place replacement, update oldvd's path and devid
3722	 * to make it distinguishable from newvd, and unopenable from now on.
3723	 */
3724	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3725		spa_strfree(oldvd->vdev_path);
3726		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3727		    KM_SLEEP);
3728		(void) sprintf(oldvd->vdev_path, "%s/%s",
3729		    newvd->vdev_path, "old");
3730		if (oldvd->vdev_devid != NULL) {
3731			spa_strfree(oldvd->vdev_devid);
3732			oldvd->vdev_devid = NULL;
3733		}
3734	}
3735
3736	/*
3737	 * If the parent is not a mirror, or if we're replacing, insert the new
3738	 * mirror/replacing/spare vdev above oldvd.
3739	 */
3740	if (pvd->vdev_ops != pvops)
3741		pvd = vdev_add_parent(oldvd, pvops);
3742
3743	ASSERT(pvd->vdev_top->vdev_parent == rvd);
3744	ASSERT(pvd->vdev_ops == pvops);
3745	ASSERT(oldvd->vdev_parent == pvd);
3746
3747	/*
3748	 * Extract the new device from its root and add it to pvd.
3749	 */
3750	vdev_remove_child(newrootvd, newvd);
3751	newvd->vdev_id = pvd->vdev_children;
3752	newvd->vdev_crtxg = oldvd->vdev_crtxg;
3753	vdev_add_child(pvd, newvd);
3754
3755	tvd = newvd->vdev_top;
3756	ASSERT(pvd->vdev_top == tvd);
3757	ASSERT(tvd->vdev_parent == rvd);
3758
3759	vdev_config_dirty(tvd);
3760
3761	/*
3762	 * Set newvd's DTL to [TXG_INITIAL, open_txg].  It will propagate
3763	 * upward when spa_vdev_exit() calls vdev_dtl_reassess().
3764	 */
3765	open_txg = txg + TXG_CONCURRENT_STATES - 1;
3766
3767	vdev_dtl_dirty(newvd, DTL_MISSING,
3768	    TXG_INITIAL, open_txg - TXG_INITIAL + 1);
3769
3770	if (newvd->vdev_isspare) {
3771		spa_spare_activate(newvd);
3772		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
3773	}
3774
3775	oldvdpath = spa_strdup(oldvd->vdev_path);
3776	newvdpath = spa_strdup(newvd->vdev_path);
3777	newvd_isspare = newvd->vdev_isspare;
3778
3779	/*
3780	 * Mark newvd's DTL dirty in this txg.
3781	 */
3782	vdev_dirty(tvd, VDD_DTL, newvd, txg);
3783
3784	(void) spa_vdev_exit(spa, newrootvd, open_txg, 0);
3785
3786	spa_history_internal_log(LOG_POOL_VDEV_ATTACH, spa, NULL,
3787	    CRED(),  "%s vdev=%s %s vdev=%s",
3788	    replacing && newvd_isspare ? "spare in" :
3789	    replacing ? "replace" : "attach", newvdpath,
3790	    replacing ? "for" : "to", oldvdpath);
3791
3792	spa_strfree(oldvdpath);
3793	spa_strfree(newvdpath);
3794
3795	/*
3796	 * Kick off a resilver to update newvd.
3797	 */
3798	VERIFY3U(spa_scrub(spa, POOL_SCRUB_RESILVER), ==, 0);
3799
3800	return (0);
3801}
3802
3803/*
3804 * Detach a device from a mirror or replacing vdev.
3805 * If 'replace_done' is specified, only detach if the parent
3806 * is a replacing vdev.
3807 */
3808int
3809spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
3810{
3811	uint64_t txg;
3812	int error;
3813	vdev_t *rvd = spa->spa_root_vdev;
3814	vdev_t *vd, *pvd, *cvd, *tvd;
3815	boolean_t unspare = B_FALSE;
3816	uint64_t unspare_guid;
3817	size_t len;
3818	char *vdpath;
3819
3820	txg = spa_vdev_enter(spa);
3821
3822	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3823
3824	if (vd == NULL)
3825		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3826
3827	if (!vd->vdev_ops->vdev_op_leaf)
3828		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3829
3830	pvd = vd->vdev_parent;
3831
3832	/*
3833	 * If the parent/child relationship is not as expected, don't do it.
3834	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
3835	 * vdev that's replacing B with C.  The user's intent in replacing
3836	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
3837	 * the replace by detaching C, the expected behavior is to end up
3838	 * M(A,B).  But suppose that right after deciding to detach C,
3839	 * the replacement of B completes.  We would have M(A,C), and then
3840	 * ask to detach C, which would leave us with just A -- not what
3841	 * the user wanted.  To prevent this, we make sure that the
3842	 * parent/child relationship hasn't changed -- in this example,
3843	 * that C's parent is still the replacing vdev R.
3844	 */
3845	if (pvd->vdev_guid != pguid && pguid != 0)
3846		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3847
3848	/*
3849	 * If replace_done is specified, only remove this device if it's
3850	 * the first child of a replacing vdev.  For the 'spare' vdev, either
3851	 * disk can be removed.
3852	 */
3853	if (replace_done) {
3854		if (pvd->vdev_ops == &vdev_replacing_ops) {
3855			if (vd->vdev_id != 0)
3856				return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3857		} else if (pvd->vdev_ops != &vdev_spare_ops) {
3858			return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3859		}
3860	}
3861
3862	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
3863	    spa_version(spa) >= SPA_VERSION_SPARES);
3864
3865	/*
3866	 * Only mirror, replacing, and spare vdevs support detach.
3867	 */
3868	if (pvd->vdev_ops != &vdev_replacing_ops &&
3869	    pvd->vdev_ops != &vdev_mirror_ops &&
3870	    pvd->vdev_ops != &vdev_spare_ops)
3871		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3872
3873	/*
3874	 * If this device has the only valid copy of some data,
3875	 * we cannot safely detach it.
3876	 */
3877	if (vdev_dtl_required(vd))
3878		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3879
3880	ASSERT(pvd->vdev_children >= 2);
3881
3882	/*
3883	 * If we are detaching the second disk from a replacing vdev, then
3884	 * check to see if we changed the original vdev's path to have "/old"
3885	 * at the end in spa_vdev_attach().  If so, undo that change now.
3886	 */
3887	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id == 1 &&
3888	    pvd->vdev_child[0]->vdev_path != NULL &&
3889	    pvd->vdev_child[1]->vdev_path != NULL) {
3890		ASSERT(pvd->vdev_child[1] == vd);
3891		cvd = pvd->vdev_child[0];
3892		len = strlen(vd->vdev_path);
3893		if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
3894		    strcmp(cvd->vdev_path + len, "/old") == 0) {
3895			spa_strfree(cvd->vdev_path);
3896			cvd->vdev_path = spa_strdup(vd->vdev_path);
3897		}
3898	}
3899
3900	/*
3901	 * If we are detaching the original disk from a spare, then it implies
3902	 * that the spare should become a real disk, and be removed from the
3903	 * active spare list for the pool.
3904	 */
3905	if (pvd->vdev_ops == &vdev_spare_ops &&
3906	    vd->vdev_id == 0 && pvd->vdev_child[1]->vdev_isspare)
3907		unspare = B_TRUE;
3908
3909	/*
3910	 * Erase the disk labels so the disk can be used for other things.
3911	 * This must be done after all other error cases are handled,
3912	 * but before we disembowel vd (so we can still do I/O to it).
3913	 * But if we can't do it, don't treat the error as fatal --
3914	 * it may be that the unwritability of the disk is the reason
3915	 * it's being detached!
3916	 */
3917	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
3918
3919	/*
3920	 * Remove vd from its parent and compact the parent's children.
3921	 */
3922	vdev_remove_child(pvd, vd);
3923	vdev_compact_children(pvd);
3924
3925	/*
3926	 * Remember one of the remaining children so we can get tvd below.
3927	 */
3928	cvd = pvd->vdev_child[0];
3929
3930	/*
3931	 * If we need to remove the remaining child from the list of hot spares,
3932	 * do it now, marking the vdev as no longer a spare in the process.
3933	 * We must do this before vdev_remove_parent(), because that can
3934	 * change the GUID if it creates a new toplevel GUID.  For a similar
3935	 * reason, we must remove the spare now, in the same txg as the detach;
3936	 * otherwise someone could attach a new sibling, change the GUID, and
3937	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
3938	 */
3939	if (unspare) {
3940		ASSERT(cvd->vdev_isspare);
3941		spa_spare_remove(cvd);
3942		unspare_guid = cvd->vdev_guid;
3943		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
3944	}
3945
3946	/*
3947	 * If the parent mirror/replacing vdev only has one child,
3948	 * the parent is no longer needed.  Remove it from the tree.
3949	 */
3950	if (pvd->vdev_children == 1)
3951		vdev_remove_parent(cvd);
3952
3953	/*
3954	 * We don't set tvd until now because the parent we just removed
3955	 * may have been the previous top-level vdev.
3956	 */
3957	tvd = cvd->vdev_top;
3958	ASSERT(tvd->vdev_parent == rvd);
3959
3960	/*
3961	 * Reevaluate the parent vdev state.
3962	 */
3963	vdev_propagate_state(cvd);
3964
3965	/*
3966	 * If the 'autoexpand' property is set on the pool then automatically
3967	 * try to expand the size of the pool. For example if the device we
3968	 * just detached was smaller than the others, it may be possible to
3969	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
3970	 * first so that we can obtain the updated sizes of the leaf vdevs.
3971	 */
3972	if (spa->spa_autoexpand) {
3973		vdev_reopen(tvd);
3974		vdev_expand(tvd, txg);
3975	}
3976
3977	vdev_config_dirty(tvd);
3978
3979	/*
3980	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
3981	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
3982	 * But first make sure we're not on any *other* txg's DTL list, to
3983	 * prevent vd from being accessed after it's freed.
3984	 */
3985	vdpath = spa_strdup(vd->vdev_path);
3986	for (int t = 0; t < TXG_SIZE; t++)
3987		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
3988	vd->vdev_detached = B_TRUE;
3989	vdev_dirty(tvd, VDD_DTL, vd, txg);
3990
3991	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
3992
3993	error = spa_vdev_exit(spa, vd, txg, 0);
3994
3995	spa_history_internal_log(LOG_POOL_VDEV_DETACH, spa, NULL, CRED(),
3996	    "vdev=%s", vdpath);
3997	spa_strfree(vdpath);
3998
3999	/*
4000	 * If this was the removal of the original device in a hot spare vdev,
4001	 * then we want to go through and remove the device from the hot spare
4002	 * list of every other pool.
4003	 */
4004	if (unspare) {
4005		spa_t *myspa = spa;
4006		spa = NULL;
4007		mutex_enter(&spa_namespace_lock);
4008		while ((spa = spa_next(spa)) != NULL) {
4009			if (spa->spa_state != POOL_STATE_ACTIVE)
4010				continue;
4011			if (spa == myspa)
4012				continue;
4013			spa_open_ref(spa, FTAG);
4014			mutex_exit(&spa_namespace_lock);
4015			(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4016			mutex_enter(&spa_namespace_lock);
4017			spa_close(spa, FTAG);
4018		}
4019		mutex_exit(&spa_namespace_lock);
4020	}
4021
4022	return (error);
4023}
4024
4025/*
4026 * Split a set of devices from their mirrors, and create a new pool from them.
4027 */
4028int
4029spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4030    nvlist_t *props, boolean_t exp)
4031{
4032	int error = 0;
4033	uint64_t txg, *glist;
4034	spa_t *newspa;
4035	uint_t c, children, lastlog;
4036	nvlist_t **child, *nvl, *tmp;
4037	dmu_tx_t *tx;
4038	char *altroot = NULL;
4039	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4040	boolean_t activate_slog;
4041
4042	if (!spa_writeable(spa))
4043		return (EROFS);
4044
4045	txg = spa_vdev_enter(spa);
4046
4047	/* clear the log and flush everything up to now */
4048	activate_slog = spa_passivate_log(spa);
4049	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4050	error = spa_offline_log(spa);
4051	txg = spa_vdev_config_enter(spa);
4052
4053	if (activate_slog)
4054		spa_activate_log(spa);
4055
4056	if (error != 0)
4057		return (spa_vdev_exit(spa, NULL, txg, error));
4058
4059	/* check new spa name before going any further */
4060	if (spa_lookup(newname) != NULL)
4061		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4062
4063	/*
4064	 * scan through all the children to ensure they're all mirrors
4065	 */
4066	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4067	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4068	    &children) != 0)
4069		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4070
4071	/* first, check to ensure we've got the right child count */
4072	rvd = spa->spa_root_vdev;
4073	lastlog = 0;
4074	for (c = 0; c < rvd->vdev_children; c++) {
4075		vdev_t *vd = rvd->vdev_child[c];
4076
4077		/* don't count the holes & logs as children */
4078		if (vd->vdev_islog || vd->vdev_ishole) {
4079			if (lastlog == 0)
4080				lastlog = c;
4081			continue;
4082		}
4083
4084		lastlog = 0;
4085	}
4086	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4087		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4088
4089	/* next, ensure no spare or cache devices are part of the split */
4090	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4091	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4092		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4093
4094	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4095	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4096
4097	/* then, loop over each vdev and validate it */
4098	for (c = 0; c < children; c++) {
4099		uint64_t is_hole = 0;
4100
4101		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4102		    &is_hole);
4103
4104		if (is_hole != 0) {
4105			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4106			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4107				continue;
4108			} else {
4109				error = EINVAL;
4110				break;
4111			}
4112		}
4113
4114		/* which disk is going to be split? */
4115		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4116		    &glist[c]) != 0) {
4117			error = EINVAL;
4118			break;
4119		}
4120
4121		/* look it up in the spa */
4122		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4123		if (vml[c] == NULL) {
4124			error = ENODEV;
4125			break;
4126		}
4127
4128		/* make sure there's nothing stopping the split */
4129		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4130		    vml[c]->vdev_islog ||
4131		    vml[c]->vdev_ishole ||
4132		    vml[c]->vdev_isspare ||
4133		    vml[c]->vdev_isl2cache ||
4134		    !vdev_writeable(vml[c]) ||
4135		    vml[c]->vdev_children != 0 ||
4136		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4137		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4138			error = EINVAL;
4139			break;
4140		}
4141
4142		if (vdev_dtl_required(vml[c])) {
4143			error = EBUSY;
4144			break;
4145		}
4146
4147		/* we need certain info from the top level */
4148		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4149		    vml[c]->vdev_top->vdev_ms_array) == 0);
4150		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4151		    vml[c]->vdev_top->vdev_ms_shift) == 0);
4152		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4153		    vml[c]->vdev_top->vdev_asize) == 0);
4154		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4155		    vml[c]->vdev_top->vdev_ashift) == 0);
4156	}
4157
4158	if (error != 0) {
4159		kmem_free(vml, children * sizeof (vdev_t *));
4160		kmem_free(glist, children * sizeof (uint64_t));
4161		return (spa_vdev_exit(spa, NULL, txg, error));
4162	}
4163
4164	/* stop writers from using the disks */
4165	for (c = 0; c < children; c++) {
4166		if (vml[c] != NULL)
4167			vml[c]->vdev_offline = B_TRUE;
4168	}
4169	vdev_reopen(spa->spa_root_vdev);
4170
4171	/*
4172	 * Temporarily record the splitting vdevs in the spa config.  This
4173	 * will disappear once the config is regenerated.
4174	 */
4175	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4176	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4177	    glist, children) == 0);
4178	kmem_free(glist, children * sizeof (uint64_t));
4179
4180	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4181	    nvl) == 0);
4182	spa->spa_config_splitting = nvl;
4183	vdev_config_dirty(spa->spa_root_vdev);
4184
4185	/* configure and create the new pool */
4186	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4187	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4188	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4189	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4190	    spa_version(spa)) == 0);
4191	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4192	    spa->spa_config_txg) == 0);
4193	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4194	    spa_generate_guid(NULL)) == 0);
4195	(void) nvlist_lookup_string(props,
4196	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4197
4198	/* add the new pool to the namespace */
4199	newspa = spa_add(newname, config, altroot);
4200	newspa->spa_config_txg = spa->spa_config_txg;
4201	spa_set_log_state(newspa, SPA_LOG_CLEAR);
4202
4203	/* release the spa config lock, retaining the namespace lock */
4204	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4205
4206	if (zio_injection_enabled)
4207		zio_handle_panic_injection(spa, FTAG, 1);
4208
4209	spa_activate(newspa, spa_mode_global);
4210	spa_async_suspend(newspa);
4211
4212	/* create the new pool from the disks of the original pool */
4213	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4214	if (error)
4215		goto out;
4216
4217	/* if that worked, generate a real config for the new pool */
4218	if (newspa->spa_root_vdev != NULL) {
4219		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4220		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4221		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4222		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4223		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4224		    B_TRUE));
4225	}
4226
4227	/* set the props */
4228	if (props != NULL) {
4229		spa_configfile_set(newspa, props, B_FALSE);
4230		error = spa_prop_set(newspa, props);
4231		if (error)
4232			goto out;
4233	}
4234
4235	/* flush everything */
4236	txg = spa_vdev_config_enter(newspa);
4237	vdev_config_dirty(newspa->spa_root_vdev);
4238	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4239
4240	if (zio_injection_enabled)
4241		zio_handle_panic_injection(spa, FTAG, 2);
4242
4243	spa_async_resume(newspa);
4244
4245	/* finally, update the original pool's config */
4246	txg = spa_vdev_config_enter(spa);
4247	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4248	error = dmu_tx_assign(tx, TXG_WAIT);
4249	if (error != 0)
4250		dmu_tx_abort(tx);
4251	for (c = 0; c < children; c++) {
4252		if (vml[c] != NULL) {
4253			vdev_split(vml[c]);
4254			if (error == 0)
4255				spa_history_internal_log(LOG_POOL_VDEV_DETACH,
4256				    spa, tx, CRED(), "vdev=%s",
4257				    vml[c]->vdev_path);
4258			vdev_free(vml[c]);
4259		}
4260	}
4261	vdev_config_dirty(spa->spa_root_vdev);
4262	spa->spa_config_splitting = NULL;
4263	nvlist_free(nvl);
4264	if (error == 0)
4265		dmu_tx_commit(tx);
4266	(void) spa_vdev_exit(spa, NULL, txg, 0);
4267
4268	if (zio_injection_enabled)
4269		zio_handle_panic_injection(spa, FTAG, 3);
4270
4271	/* split is complete; log a history record */
4272	spa_history_internal_log(LOG_POOL_SPLIT, newspa, NULL, CRED(),
4273	    "split new pool %s from pool %s", newname, spa_name(spa));
4274
4275	kmem_free(vml, children * sizeof (vdev_t *));
4276
4277	/* if we're not going to mount the filesystems in userland, export */
4278	if (exp)
4279		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4280		    B_FALSE, B_FALSE);
4281
4282	return (error);
4283
4284out:
4285	spa_unload(newspa);
4286	spa_deactivate(newspa);
4287	spa_remove(newspa);
4288
4289	txg = spa_vdev_config_enter(spa);
4290	nvlist_free(spa->spa_config_splitting);
4291	spa->spa_config_splitting = NULL;
4292	(void) spa_vdev_exit(spa, NULL, txg, error);
4293
4294	kmem_free(vml, children * sizeof (vdev_t *));
4295	return (error);
4296}
4297
4298static nvlist_t *
4299spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4300{
4301	for (int i = 0; i < count; i++) {
4302		uint64_t guid;
4303
4304		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4305		    &guid) == 0);
4306
4307		if (guid == target_guid)
4308			return (nvpp[i]);
4309	}
4310
4311	return (NULL);
4312}
4313
4314static void
4315spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4316	nvlist_t *dev_to_remove)
4317{
4318	nvlist_t **newdev = NULL;
4319
4320	if (count > 1)
4321		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4322
4323	for (int i = 0, j = 0; i < count; i++) {
4324		if (dev[i] == dev_to_remove)
4325			continue;
4326		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4327	}
4328
4329	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4330	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4331
4332	for (int i = 0; i < count - 1; i++)
4333		nvlist_free(newdev[i]);
4334
4335	if (count > 1)
4336		kmem_free(newdev, (count - 1) * sizeof (void *));
4337}
4338
4339/*
4340 * Removing a device from the vdev namespace requires several steps
4341 * and can take a significant amount of time.  As a result we use
4342 * the spa_vdev_config_[enter/exit] functions which allow us to
4343 * grab and release the spa_config_lock while still holding the namespace
4344 * lock.  During each step the configuration is synced out.
4345 */
4346
4347/*
4348 * Evacuate the device.
4349 */
4350int
4351spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4352{
4353	int error = 0;
4354	uint64_t txg;
4355
4356	ASSERT(MUTEX_HELD(&spa_namespace_lock));
4357	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4358	ASSERT(vd == vd->vdev_top);
4359
4360	/*
4361	 * Evacuate the device.  We don't hold the config lock as writer
4362	 * since we need to do I/O but we do keep the
4363	 * spa_namespace_lock held.  Once this completes the device
4364	 * should no longer have any blocks allocated on it.
4365	 */
4366	if (vd->vdev_islog) {
4367		error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
4368		    NULL, DS_FIND_CHILDREN);
4369	} else {
4370		error = ENOTSUP;	/* until we have bp rewrite */
4371	}
4372
4373	txg_wait_synced(spa_get_dsl(spa), 0);
4374
4375	if (error)
4376		return (error);
4377
4378	/*
4379	 * The evacuation succeeded.  Remove any remaining MOS metadata
4380	 * associated with this vdev, and wait for these changes to sync.
4381	 */
4382	txg = spa_vdev_config_enter(spa);
4383	vd->vdev_removing = B_TRUE;
4384	vdev_dirty(vd, 0, NULL, txg);
4385	vdev_config_dirty(vd);
4386	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4387
4388	return (0);
4389}
4390
4391/*
4392 * Complete the removal by cleaning up the namespace.
4393 */
4394void
4395spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
4396{
4397	vdev_t *rvd = spa->spa_root_vdev;
4398	uint64_t id = vd->vdev_id;
4399	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
4400
4401	ASSERT(MUTEX_HELD(&spa_namespace_lock));
4402	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4403	ASSERT(vd == vd->vdev_top);
4404
4405	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4406
4407	if (list_link_active(&vd->vdev_state_dirty_node))
4408		vdev_state_clean(vd);
4409	if (list_link_active(&vd->vdev_config_dirty_node))
4410		vdev_config_clean(vd);
4411
4412	vdev_free(vd);
4413
4414	if (last_vdev) {
4415		vdev_compact_children(rvd);
4416	} else {
4417		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
4418		vdev_add_child(rvd, vd);
4419	}
4420	vdev_config_dirty(rvd);
4421
4422	/*
4423	 * Reassess the health of our root vdev.
4424	 */
4425	vdev_reopen(rvd);
4426}
4427
4428/*
4429 * Remove a device from the pool.  Currently, this supports removing only hot
4430 * spares, slogs, and level 2 ARC devices.
4431 */
4432int
4433spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
4434{
4435	vdev_t *vd;
4436	metaslab_group_t *mg;
4437	nvlist_t **spares, **l2cache, *nv;
4438	uint64_t txg = 0;
4439	uint_t nspares, nl2cache;
4440	int error = 0;
4441	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
4442
4443	if (!locked)
4444		txg = spa_vdev_enter(spa);
4445
4446	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4447
4448	if (spa->spa_spares.sav_vdevs != NULL &&
4449	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4450	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
4451	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
4452		/*
4453		 * Only remove the hot spare if it's not currently in use
4454		 * in this pool.
4455		 */
4456		if (vd == NULL || unspare) {
4457			spa_vdev_remove_aux(spa->spa_spares.sav_config,
4458			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
4459			spa_load_spares(spa);
4460			spa->spa_spares.sav_sync = B_TRUE;
4461		} else {
4462			error = EBUSY;
4463		}
4464	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
4465	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4466	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
4467	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
4468		/*
4469		 * Cache devices can always be removed.
4470		 */
4471		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
4472		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
4473		spa_load_l2cache(spa);
4474		spa->spa_l2cache.sav_sync = B_TRUE;
4475	} else if (vd != NULL && vd->vdev_islog) {
4476		ASSERT(!locked);
4477		ASSERT(vd == vd->vdev_top);
4478
4479		/*
4480		 * XXX - Once we have bp-rewrite this should
4481		 * become the common case.
4482		 */
4483
4484		mg = vd->vdev_mg;
4485
4486		/*
4487		 * Stop allocating from this vdev.
4488		 */
4489		metaslab_group_passivate(mg);
4490
4491		/*
4492		 * Wait for the youngest allocations and frees to sync,
4493		 * and then wait for the deferral of those frees to finish.
4494		 */
4495		spa_vdev_config_exit(spa, NULL,
4496		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
4497
4498		/*
4499		 * Attempt to evacuate the vdev.
4500		 */
4501		error = spa_vdev_remove_evacuate(spa, vd);
4502
4503		txg = spa_vdev_config_enter(spa);
4504
4505		/*
4506		 * If we couldn't evacuate the vdev, unwind.
4507		 */
4508		if (error) {
4509			metaslab_group_activate(mg);
4510			return (spa_vdev_exit(spa, NULL, txg, error));
4511		}
4512
4513		/*
4514		 * Clean up the vdev namespace.
4515		 */
4516		spa_vdev_remove_from_namespace(spa, vd);
4517
4518	} else if (vd != NULL) {
4519		/*
4520		 * Normal vdevs cannot be removed (yet).
4521		 */
4522		error = ENOTSUP;
4523	} else {
4524		/*
4525		 * There is no vdev of any kind with the specified guid.
4526		 */
4527		error = ENOENT;
4528	}
4529
4530	if (!locked)
4531		return (spa_vdev_exit(spa, NULL, txg, error));
4532
4533	return (error);
4534}
4535
4536/*
4537 * Find any device that's done replacing, or a vdev marked 'unspare' that's
4538 * current spared, so we can detach it.
4539 */
4540static vdev_t *
4541spa_vdev_resilver_done_hunt(vdev_t *vd)
4542{
4543	vdev_t *newvd, *oldvd;
4544
4545	for (int c = 0; c < vd->vdev_children; c++) {
4546		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
4547		if (oldvd != NULL)
4548			return (oldvd);
4549	}
4550
4551	/*
4552	 * Check for a completed replacement.
4553	 */
4554	if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
4555		oldvd = vd->vdev_child[0];
4556		newvd = vd->vdev_child[1];
4557
4558		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4559		    !vdev_dtl_required(oldvd))
4560			return (oldvd);
4561	}
4562
4563	/*
4564	 * Check for a completed resilver with the 'unspare' flag set.
4565	 */
4566	if (vd->vdev_ops == &vdev_spare_ops && vd->vdev_children == 2) {
4567		newvd = vd->vdev_child[0];
4568		oldvd = vd->vdev_child[1];
4569
4570		if (newvd->vdev_unspare &&
4571		    vdev_dtl_empty(newvd, DTL_MISSING) &&
4572		    !vdev_dtl_required(oldvd)) {
4573			newvd->vdev_unspare = 0;
4574			return (oldvd);
4575		}
4576	}
4577
4578	return (NULL);
4579}
4580
4581static void
4582spa_vdev_resilver_done(spa_t *spa)
4583{
4584	vdev_t *vd, *pvd, *ppvd;
4585	uint64_t guid, sguid, pguid, ppguid;
4586
4587	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4588
4589	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4590		pvd = vd->vdev_parent;
4591		ppvd = pvd->vdev_parent;
4592		guid = vd->vdev_guid;
4593		pguid = pvd->vdev_guid;
4594		ppguid = ppvd->vdev_guid;
4595		sguid = 0;
4596		/*
4597		 * If we have just finished replacing a hot spared device, then
4598		 * we need to detach the parent's first child (the original hot
4599		 * spare) as well.
4600		 */
4601		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0) {
4602			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4603			ASSERT(ppvd->vdev_children == 2);
4604			sguid = ppvd->vdev_child[1]->vdev_guid;
4605		}
4606		spa_config_exit(spa, SCL_ALL, FTAG);
4607		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4608			return;
4609		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4610			return;
4611		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4612	}
4613
4614	spa_config_exit(spa, SCL_ALL, FTAG);
4615}
4616
4617/*
4618 * Update the stored path or FRU for this vdev.
4619 */
4620int
4621spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4622    boolean_t ispath)
4623{
4624	vdev_t *vd;
4625
4626	spa_vdev_state_enter(spa, SCL_ALL);
4627
4628	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4629		return (spa_vdev_state_exit(spa, NULL, ENOENT));
4630
4631	if (!vd->vdev_ops->vdev_op_leaf)
4632		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
4633
4634	if (ispath) {
4635		spa_strfree(vd->vdev_path);
4636		vd->vdev_path = spa_strdup(value);
4637	} else {
4638		if (vd->vdev_fru != NULL)
4639			spa_strfree(vd->vdev_fru);
4640		vd->vdev_fru = spa_strdup(value);
4641	}
4642
4643	return (spa_vdev_state_exit(spa, vd, 0));
4644}
4645
4646int
4647spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
4648{
4649	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
4650}
4651
4652int
4653spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
4654{
4655	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
4656}
4657
4658/*
4659 * ==========================================================================
4660 * SPA Scrubbing
4661 * ==========================================================================
4662 */
4663
4664int
4665spa_scrub(spa_t *spa, pool_scrub_type_t type)
4666{
4667	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4668
4669	if ((uint_t)type >= POOL_SCRUB_TYPES)
4670		return (ENOTSUP);
4671
4672	/*
4673	 * If a resilver was requested, but there is no DTL on a
4674	 * writeable leaf device, we have nothing to do.
4675	 */
4676	if (type == POOL_SCRUB_RESILVER &&
4677	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
4678		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
4679		return (0);
4680	}
4681
4682	if (type == POOL_SCRUB_EVERYTHING &&
4683	    spa->spa_dsl_pool->dp_scrub_func != SCRUB_FUNC_NONE &&
4684	    spa->spa_dsl_pool->dp_scrub_isresilver)
4685		return (EBUSY);
4686
4687	if (type == POOL_SCRUB_EVERYTHING || type == POOL_SCRUB_RESILVER) {
4688		return (dsl_pool_scrub_clean(spa->spa_dsl_pool));
4689	} else if (type == POOL_SCRUB_NONE) {
4690		return (dsl_pool_scrub_cancel(spa->spa_dsl_pool));
4691	} else {
4692		return (EINVAL);
4693	}
4694}
4695
4696/*
4697 * ==========================================================================
4698 * SPA async task processing
4699 * ==========================================================================
4700 */
4701
4702static void
4703spa_async_remove(spa_t *spa, vdev_t *vd)
4704{
4705	if (vd->vdev_remove_wanted) {
4706		vd->vdev_remove_wanted = 0;
4707		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
4708
4709		/*
4710		 * We want to clear the stats, but we don't want to do a full
4711		 * vdev_clear() as that will cause us to throw away
4712		 * degraded/faulted state as well as attempt to reopen the
4713		 * device, all of which is a waste.
4714		 */
4715		vd->vdev_stat.vs_read_errors = 0;
4716		vd->vdev_stat.vs_write_errors = 0;
4717		vd->vdev_stat.vs_checksum_errors = 0;
4718
4719		vdev_state_dirty(vd->vdev_top);
4720	}
4721
4722	for (int c = 0; c < vd->vdev_children; c++)
4723		spa_async_remove(spa, vd->vdev_child[c]);
4724}
4725
4726static void
4727spa_async_probe(spa_t *spa, vdev_t *vd)
4728{
4729	if (vd->vdev_probe_wanted) {
4730		vd->vdev_probe_wanted = 0;
4731		vdev_reopen(vd);	/* vdev_open() does the actual probe */
4732	}
4733
4734	for (int c = 0; c < vd->vdev_children; c++)
4735		spa_async_probe(spa, vd->vdev_child[c]);
4736}
4737
4738static void
4739spa_async_autoexpand(spa_t *spa, vdev_t *vd)
4740{
4741	sysevent_id_t eid;
4742	nvlist_t *attr;
4743	char *physpath;
4744
4745	if (!spa->spa_autoexpand)
4746		return;
4747
4748	for (int c = 0; c < vd->vdev_children; c++) {
4749		vdev_t *cvd = vd->vdev_child[c];
4750		spa_async_autoexpand(spa, cvd);
4751	}
4752
4753	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
4754		return;
4755
4756	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
4757	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
4758
4759	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4760	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
4761
4762	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
4763	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
4764
4765	nvlist_free(attr);
4766	kmem_free(physpath, MAXPATHLEN);
4767}
4768
4769static void
4770spa_async_thread(spa_t *spa)
4771{
4772	int tasks;
4773
4774	ASSERT(spa->spa_sync_on);
4775
4776	mutex_enter(&spa->spa_async_lock);
4777	tasks = spa->spa_async_tasks;
4778	spa->spa_async_tasks = 0;
4779	mutex_exit(&spa->spa_async_lock);
4780
4781	/*
4782	 * See if the config needs to be updated.
4783	 */
4784	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
4785		uint64_t old_space, new_space;
4786
4787		mutex_enter(&spa_namespace_lock);
4788		old_space = metaslab_class_get_space(spa_normal_class(spa));
4789		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4790		new_space = metaslab_class_get_space(spa_normal_class(spa));
4791		mutex_exit(&spa_namespace_lock);
4792
4793		/*
4794		 * If the pool grew as a result of the config update,
4795		 * then log an internal history event.
4796		 */
4797		if (new_space != old_space) {
4798			spa_history_internal_log(LOG_POOL_VDEV_ONLINE,
4799			    spa, NULL, CRED(),
4800			    "pool '%s' size: %llu(+%llu)",
4801			    spa_name(spa), new_space, new_space - old_space);
4802		}
4803	}
4804
4805	/*
4806	 * See if any devices need to be marked REMOVED.
4807	 */
4808	if (tasks & SPA_ASYNC_REMOVE) {
4809		spa_vdev_state_enter(spa, SCL_NONE);
4810		spa_async_remove(spa, spa->spa_root_vdev);
4811		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
4812			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
4813		for (int i = 0; i < spa->spa_spares.sav_count; i++)
4814			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
4815		(void) spa_vdev_state_exit(spa, NULL, 0);
4816	}
4817
4818	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
4819		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4820		spa_async_autoexpand(spa, spa->spa_root_vdev);
4821		spa_config_exit(spa, SCL_CONFIG, FTAG);
4822	}
4823
4824	/*
4825	 * See if any devices need to be probed.
4826	 */
4827	if (tasks & SPA_ASYNC_PROBE) {
4828		spa_vdev_state_enter(spa, SCL_NONE);
4829		spa_async_probe(spa, spa->spa_root_vdev);
4830		(void) spa_vdev_state_exit(spa, NULL, 0);
4831	}
4832
4833	/*
4834	 * If any devices are done replacing, detach them.
4835	 */
4836	if (tasks & SPA_ASYNC_RESILVER_DONE)
4837		spa_vdev_resilver_done(spa);
4838
4839	/*
4840	 * Kick off a resilver.
4841	 */
4842	if (tasks & SPA_ASYNC_RESILVER)
4843		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER) == 0);
4844
4845	/*
4846	 * Let the world know that we're done.
4847	 */
4848	mutex_enter(&spa->spa_async_lock);
4849	spa->spa_async_thread = NULL;
4850	cv_broadcast(&spa->spa_async_cv);
4851	mutex_exit(&spa->spa_async_lock);
4852	thread_exit();
4853}
4854
4855void
4856spa_async_suspend(spa_t *spa)
4857{
4858	mutex_enter(&spa->spa_async_lock);
4859	spa->spa_async_suspended++;
4860	while (spa->spa_async_thread != NULL)
4861		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
4862	mutex_exit(&spa->spa_async_lock);
4863}
4864
4865void
4866spa_async_resume(spa_t *spa)
4867{
4868	mutex_enter(&spa->spa_async_lock);
4869	ASSERT(spa->spa_async_suspended != 0);
4870	spa->spa_async_suspended--;
4871	mutex_exit(&spa->spa_async_lock);
4872}
4873
4874static void
4875spa_async_dispatch(spa_t *spa)
4876{
4877	mutex_enter(&spa->spa_async_lock);
4878	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
4879	    spa->spa_async_thread == NULL &&
4880	    rootdir != NULL && !vn_is_readonly(rootdir))
4881		spa->spa_async_thread = thread_create(NULL, 0,
4882		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
4883	mutex_exit(&spa->spa_async_lock);
4884}
4885
4886void
4887spa_async_request(spa_t *spa, int task)
4888{
4889	mutex_enter(&spa->spa_async_lock);
4890	spa->spa_async_tasks |= task;
4891	mutex_exit(&spa->spa_async_lock);
4892}
4893
4894/*
4895 * ==========================================================================
4896 * SPA syncing routines
4897 * ==========================================================================
4898 */
4899static void
4900spa_sync_deferred_bplist(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx, uint64_t txg)
4901{
4902	blkptr_t blk;
4903	uint64_t itor = 0;
4904	uint8_t c = 1;
4905
4906	while (bplist_iterate(bpl, &itor, &blk) == 0) {
4907		ASSERT(blk.blk_birth < txg);
4908		zio_free(spa, txg, &blk);
4909	}
4910
4911	bplist_vacate(bpl, tx);
4912
4913	/*
4914	 * Pre-dirty the first block so we sync to convergence faster.
4915	 * (Usually only the first block is needed.)
4916	 */
4917	dmu_write(bpl->bpl_mos, spa->spa_deferred_bplist_obj, 0, 1, &c, tx);
4918}
4919
4920static void
4921spa_sync_free(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
4922{
4923	zio_t *zio = arg;
4924
4925	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
4926	    zio->io_flags));
4927}
4928
4929static void
4930spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
4931{
4932	char *packed = NULL;
4933	size_t bufsize;
4934	size_t nvsize = 0;
4935	dmu_buf_t *db;
4936
4937	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
4938
4939	/*
4940	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
4941	 * information.  This avoids the dbuf_will_dirty() path and
4942	 * saves us a pre-read to get data we don't actually care about.
4943	 */
4944	bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
4945	packed = kmem_alloc(bufsize, KM_SLEEP);
4946
4947	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
4948	    KM_SLEEP) == 0);
4949	bzero(packed + nvsize, bufsize - nvsize);
4950
4951	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
4952
4953	kmem_free(packed, bufsize);
4954
4955	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
4956	dmu_buf_will_dirty(db, tx);
4957	*(uint64_t *)db->db_data = nvsize;
4958	dmu_buf_rele(db, FTAG);
4959}
4960
4961static void
4962spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
4963    const char *config, const char *entry)
4964{
4965	nvlist_t *nvroot;
4966	nvlist_t **list;
4967	int i;
4968
4969	if (!sav->sav_sync)
4970		return;
4971
4972	/*
4973	 * Update the MOS nvlist describing the list of available devices.
4974	 * spa_validate_aux() will have already made sure this nvlist is
4975	 * valid and the vdevs are labeled appropriately.
4976	 */
4977	if (sav->sav_object == 0) {
4978		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
4979		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
4980		    sizeof (uint64_t), tx);
4981		VERIFY(zap_update(spa->spa_meta_objset,
4982		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
4983		    &sav->sav_object, tx) == 0);
4984	}
4985
4986	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4987	if (sav->sav_count == 0) {
4988		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
4989	} else {
4990		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
4991		for (i = 0; i < sav->sav_count; i++)
4992			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
4993			    B_FALSE, B_FALSE, B_TRUE);
4994		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
4995		    sav->sav_count) == 0);
4996		for (i = 0; i < sav->sav_count; i++)
4997			nvlist_free(list[i]);
4998		kmem_free(list, sav->sav_count * sizeof (void *));
4999	}
5000
5001	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5002	nvlist_free(nvroot);
5003
5004	sav->sav_sync = B_FALSE;
5005}
5006
5007static void
5008spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5009{
5010	nvlist_t *config;
5011
5012	if (list_is_empty(&spa->spa_config_dirty_list))
5013		return;
5014
5015	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5016
5017	config = spa_config_generate(spa, spa->spa_root_vdev,
5018	    dmu_tx_get_txg(tx), B_FALSE);
5019
5020	spa_config_exit(spa, SCL_STATE, FTAG);
5021
5022	if (spa->spa_config_syncing)
5023		nvlist_free(spa->spa_config_syncing);
5024	spa->spa_config_syncing = config;
5025
5026	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5027}
5028
5029/*
5030 * Set zpool properties.
5031 */
5032static void
5033spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
5034{
5035	spa_t *spa = arg1;
5036	objset_t *mos = spa->spa_meta_objset;
5037	nvlist_t *nvp = arg2;
5038	nvpair_t *elem;
5039	uint64_t intval;
5040	char *strval;
5041	zpool_prop_t prop;
5042	const char *propname;
5043	zprop_type_t proptype;
5044
5045	mutex_enter(&spa->spa_props_lock);
5046
5047	elem = NULL;
5048	while ((elem = nvlist_next_nvpair(nvp, elem))) {
5049		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5050		case ZPOOL_PROP_VERSION:
5051			/*
5052			 * Only set version for non-zpool-creation cases
5053			 * (set/import). spa_create() needs special care
5054			 * for version setting.
5055			 */
5056			if (tx->tx_txg != TXG_INITIAL) {
5057				VERIFY(nvpair_value_uint64(elem,
5058				    &intval) == 0);
5059				ASSERT(intval <= SPA_VERSION);
5060				ASSERT(intval >= spa_version(spa));
5061				spa->spa_uberblock.ub_version = intval;
5062				vdev_config_dirty(spa->spa_root_vdev);
5063			}
5064			break;
5065
5066		case ZPOOL_PROP_ALTROOT:
5067			/*
5068			 * 'altroot' is a non-persistent property. It should
5069			 * have been set temporarily at creation or import time.
5070			 */
5071			ASSERT(spa->spa_root != NULL);
5072			break;
5073
5074		case ZPOOL_PROP_CACHEFILE:
5075			/*
5076			 * 'cachefile' is also a non-persisitent property.
5077			 */
5078			break;
5079		default:
5080			/*
5081			 * Set pool property values in the poolprops mos object.
5082			 */
5083			if (spa->spa_pool_props_object == 0) {
5084				VERIFY((spa->spa_pool_props_object =
5085				    zap_create(mos, DMU_OT_POOL_PROPS,
5086				    DMU_OT_NONE, 0, tx)) > 0);
5087
5088				VERIFY(zap_update(mos,
5089				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5090				    8, 1, &spa->spa_pool_props_object, tx)
5091				    == 0);
5092			}
5093
5094			/* normalize the property name */
5095			propname = zpool_prop_to_name(prop);
5096			proptype = zpool_prop_get_type(prop);
5097
5098			if (nvpair_type(elem) == DATA_TYPE_STRING) {
5099				ASSERT(proptype == PROP_TYPE_STRING);
5100				VERIFY(nvpair_value_string(elem, &strval) == 0);
5101				VERIFY(zap_update(mos,
5102				    spa->spa_pool_props_object, propname,
5103				    1, strlen(strval) + 1, strval, tx) == 0);
5104
5105			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5106				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5107
5108				if (proptype == PROP_TYPE_INDEX) {
5109					const char *unused;
5110					VERIFY(zpool_prop_index_to_string(
5111					    prop, intval, &unused) == 0);
5112				}
5113				VERIFY(zap_update(mos,
5114				    spa->spa_pool_props_object, propname,
5115				    8, 1, &intval, tx) == 0);
5116			} else {
5117				ASSERT(0); /* not allowed */
5118			}
5119
5120			switch (prop) {
5121			case ZPOOL_PROP_DELEGATION:
5122				spa->spa_delegation = intval;
5123				break;
5124			case ZPOOL_PROP_BOOTFS:
5125				spa->spa_bootfs = intval;
5126				break;
5127			case ZPOOL_PROP_FAILUREMODE:
5128				spa->spa_failmode = intval;
5129				break;
5130			case ZPOOL_PROP_AUTOEXPAND:
5131				spa->spa_autoexpand = intval;
5132				spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
5133				break;
5134			case ZPOOL_PROP_DEDUPDITTO:
5135				spa->spa_dedup_ditto = intval;
5136				break;
5137			default:
5138				break;
5139			}
5140		}
5141
5142		/* log internal history if this is not a zpool create */
5143		if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5144		    tx->tx_txg != TXG_INITIAL) {
5145			spa_history_internal_log(LOG_POOL_PROPSET,
5146			    spa, tx, cr, "%s %lld %s",
5147			    nvpair_name(elem), intval, spa_name(spa));
5148		}
5149	}
5150
5151	mutex_exit(&spa->spa_props_lock);
5152}
5153
5154/*
5155 * Sync the specified transaction group.  New blocks may be dirtied as
5156 * part of the process, so we iterate until it converges.
5157 */
5158void
5159spa_sync(spa_t *spa, uint64_t txg)
5160{
5161	dsl_pool_t *dp = spa->spa_dsl_pool;
5162	objset_t *mos = spa->spa_meta_objset;
5163	bplist_t *defer_bpl = &spa->spa_deferred_bplist;
5164	bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5165	vdev_t *rvd = spa->spa_root_vdev;
5166	vdev_t *vd;
5167	dmu_tx_t *tx;
5168	int error;
5169
5170	/*
5171	 * Lock out configuration changes.
5172	 */
5173	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5174
5175	spa->spa_syncing_txg = txg;
5176	spa->spa_sync_pass = 0;
5177
5178	/*
5179	 * If there are any pending vdev state changes, convert them
5180	 * into config changes that go out with this transaction group.
5181	 */
5182	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5183	while (list_head(&spa->spa_state_dirty_list) != NULL) {
5184		/*
5185		 * We need the write lock here because, for aux vdevs,
5186		 * calling vdev_config_dirty() modifies sav_config.
5187		 * This is ugly and will become unnecessary when we
5188		 * eliminate the aux vdev wart by integrating all vdevs
5189		 * into the root vdev tree.
5190		 */
5191		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5192		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5193		while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5194			vdev_state_clean(vd);
5195			vdev_config_dirty(vd);
5196		}
5197		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5198		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5199	}
5200	spa_config_exit(spa, SCL_STATE, FTAG);
5201
5202	VERIFY(0 == bplist_open(defer_bpl, mos, spa->spa_deferred_bplist_obj));
5203
5204	tx = dmu_tx_create_assigned(dp, txg);
5205
5206	/*
5207	 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5208	 * set spa_deflate if we have no raid-z vdevs.
5209	 */
5210	if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5211	    spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5212		int i;
5213
5214		for (i = 0; i < rvd->vdev_children; i++) {
5215			vd = rvd->vdev_child[i];
5216			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5217				break;
5218		}
5219		if (i == rvd->vdev_children) {
5220			spa->spa_deflate = TRUE;
5221			VERIFY(0 == zap_add(spa->spa_meta_objset,
5222			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5223			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5224		}
5225	}
5226
5227	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5228	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5229		dsl_pool_create_origin(dp, tx);
5230
5231		/* Keeping the origin open increases spa_minref */
5232		spa->spa_minref += 3;
5233	}
5234
5235	if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5236	    spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5237		dsl_pool_upgrade_clones(dp, tx);
5238	}
5239
5240	/*
5241	 * If anything has changed in this txg, push the deferred frees
5242	 * from the previous txg.  If not, leave them alone so that we
5243	 * don't generate work on an otherwise idle system.
5244	 */
5245	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5246	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5247	    !txg_list_empty(&dp->dp_sync_tasks, txg))
5248		spa_sync_deferred_bplist(spa, defer_bpl, tx, txg);
5249
5250	/*
5251	 * Iterate to convergence.
5252	 */
5253	do {
5254		int pass = ++spa->spa_sync_pass;
5255
5256		spa_sync_config_object(spa, tx);
5257		spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5258		    ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5259		spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5260		    ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5261		spa_errlog_sync(spa, txg);
5262		dsl_pool_sync(dp, txg);
5263
5264		if (pass <= SYNC_PASS_DEFERRED_FREE) {
5265			zio_t *zio = zio_root(spa, NULL, NULL, 0);
5266			bplist_sync(free_bpl, spa_sync_free, zio, tx);
5267			VERIFY(zio_wait(zio) == 0);
5268		} else {
5269			bplist_sync(free_bpl, bplist_enqueue_cb, defer_bpl, tx);
5270		}
5271
5272		ddt_sync(spa, txg);
5273
5274		mutex_enter(&spa->spa_scrub_lock);
5275		while (spa->spa_scrub_inflight > 0)
5276			cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
5277		mutex_exit(&spa->spa_scrub_lock);
5278
5279		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
5280			vdev_sync(vd, txg);
5281
5282	} while (dmu_objset_is_dirty(mos, txg));
5283
5284	ASSERT(free_bpl->bpl_queue == NULL);
5285
5286	bplist_close(defer_bpl);
5287
5288	/*
5289	 * Rewrite the vdev configuration (which includes the uberblock)
5290	 * to commit the transaction group.
5291	 *
5292	 * If there are no dirty vdevs, we sync the uberblock to a few
5293	 * random top-level vdevs that are known to be visible in the
5294	 * config cache (see spa_vdev_add() for a complete description).
5295	 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
5296	 */
5297	for (;;) {
5298		/*
5299		 * We hold SCL_STATE to prevent vdev open/close/etc.
5300		 * while we're attempting to write the vdev labels.
5301		 */
5302		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5303
5304		if (list_is_empty(&spa->spa_config_dirty_list)) {
5305			vdev_t *svd[SPA_DVAS_PER_BP];
5306			int svdcount = 0;
5307			int children = rvd->vdev_children;
5308			int c0 = spa_get_random(children);
5309
5310			for (int c = 0; c < children; c++) {
5311				vd = rvd->vdev_child[(c0 + c) % children];
5312				if (vd->vdev_ms_array == 0 || vd->vdev_islog)
5313					continue;
5314				svd[svdcount++] = vd;
5315				if (svdcount == SPA_DVAS_PER_BP)
5316					break;
5317			}
5318			error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
5319			if (error != 0)
5320				error = vdev_config_sync(svd, svdcount, txg,
5321				    B_TRUE);
5322		} else {
5323			error = vdev_config_sync(rvd->vdev_child,
5324			    rvd->vdev_children, txg, B_FALSE);
5325			if (error != 0)
5326				error = vdev_config_sync(rvd->vdev_child,
5327				    rvd->vdev_children, txg, B_TRUE);
5328		}
5329
5330		spa_config_exit(spa, SCL_STATE, FTAG);
5331
5332		if (error == 0)
5333			break;
5334		zio_suspend(spa, NULL);
5335		zio_resume_wait(spa);
5336	}
5337	dmu_tx_commit(tx);
5338
5339	/*
5340	 * Clear the dirty config list.
5341	 */
5342	while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
5343		vdev_config_clean(vd);
5344
5345	/*
5346	 * Now that the new config has synced transactionally,
5347	 * let it become visible to the config cache.
5348	 */
5349	if (spa->spa_config_syncing != NULL) {
5350		spa_config_set(spa, spa->spa_config_syncing);
5351		spa->spa_config_txg = txg;
5352		spa->spa_config_syncing = NULL;
5353	}
5354
5355	spa->spa_ubsync = spa->spa_uberblock;
5356
5357	dsl_pool_sync_done(dp, txg);
5358
5359	/*
5360	 * Update usable space statistics.
5361	 */
5362	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
5363		vdev_sync_done(vd, txg);
5364
5365	spa_update_dspace(spa);
5366
5367	/*
5368	 * It had better be the case that we didn't dirty anything
5369	 * since vdev_config_sync().
5370	 */
5371	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
5372	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
5373	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
5374	ASSERT(defer_bpl->bpl_queue == NULL);
5375	ASSERT(free_bpl->bpl_queue == NULL);
5376
5377	spa->spa_sync_pass = 0;
5378
5379	spa_config_exit(spa, SCL_CONFIG, FTAG);
5380
5381	spa_handle_ignored_writes(spa);
5382
5383	/*
5384	 * If any async tasks have been requested, kick them off.
5385	 */
5386	spa_async_dispatch(spa);
5387}
5388
5389/*
5390 * Sync all pools.  We don't want to hold the namespace lock across these
5391 * operations, so we take a reference on the spa_t and drop the lock during the
5392 * sync.
5393 */
5394void
5395spa_sync_allpools(void)
5396{
5397	spa_t *spa = NULL;
5398	mutex_enter(&spa_namespace_lock);
5399	while ((spa = spa_next(spa)) != NULL) {
5400		if (spa_state(spa) != POOL_STATE_ACTIVE || spa_suspended(spa))
5401			continue;
5402		spa_open_ref(spa, FTAG);
5403		mutex_exit(&spa_namespace_lock);
5404		txg_wait_synced(spa_get_dsl(spa), 0);
5405		mutex_enter(&spa_namespace_lock);
5406		spa_close(spa, FTAG);
5407	}
5408	mutex_exit(&spa_namespace_lock);
5409}
5410
5411/*
5412 * ==========================================================================
5413 * Miscellaneous routines
5414 * ==========================================================================
5415 */
5416
5417/*
5418 * Remove all pools in the system.
5419 */
5420void
5421spa_evict_all(void)
5422{
5423	spa_t *spa;
5424
5425	/*
5426	 * Remove all cached state.  All pools should be closed now,
5427	 * so every spa in the AVL tree should be unreferenced.
5428	 */
5429	mutex_enter(&spa_namespace_lock);
5430	while ((spa = spa_next(NULL)) != NULL) {
5431		/*
5432		 * Stop async tasks.  The async thread may need to detach
5433		 * a device that's been replaced, which requires grabbing
5434		 * spa_namespace_lock, so we must drop it here.
5435		 */
5436		spa_open_ref(spa, FTAG);
5437		mutex_exit(&spa_namespace_lock);
5438		spa_async_suspend(spa);
5439		mutex_enter(&spa_namespace_lock);
5440		spa_close(spa, FTAG);
5441
5442		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5443			spa_unload(spa);
5444			spa_deactivate(spa);
5445		}
5446		spa_remove(spa);
5447	}
5448	mutex_exit(&spa_namespace_lock);
5449}
5450
5451vdev_t *
5452spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
5453{
5454	vdev_t *vd;
5455	int i;
5456
5457	if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
5458		return (vd);
5459
5460	if (aux) {
5461		for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
5462			vd = spa->spa_l2cache.sav_vdevs[i];
5463			if (vd->vdev_guid == guid)
5464				return (vd);
5465		}
5466
5467		for (i = 0; i < spa->spa_spares.sav_count; i++) {
5468			vd = spa->spa_spares.sav_vdevs[i];
5469			if (vd->vdev_guid == guid)
5470				return (vd);
5471		}
5472	}
5473
5474	return (NULL);
5475}
5476
5477void
5478spa_upgrade(spa_t *spa, uint64_t version)
5479{
5480	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5481
5482	/*
5483	 * This should only be called for a non-faulted pool, and since a
5484	 * future version would result in an unopenable pool, this shouldn't be
5485	 * possible.
5486	 */
5487	ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
5488	ASSERT(version >= spa->spa_uberblock.ub_version);
5489
5490	spa->spa_uberblock.ub_version = version;
5491	vdev_config_dirty(spa->spa_root_vdev);
5492
5493	spa_config_exit(spa, SCL_ALL, FTAG);
5494
5495	txg_wait_synced(spa_get_dsl(spa), 0);
5496}
5497
5498boolean_t
5499spa_has_spare(spa_t *spa, uint64_t guid)
5500{
5501	int i;
5502	uint64_t spareguid;
5503	spa_aux_vdev_t *sav = &spa->spa_spares;
5504
5505	for (i = 0; i < sav->sav_count; i++)
5506		if (sav->sav_vdevs[i]->vdev_guid == guid)
5507			return (B_TRUE);
5508
5509	for (i = 0; i < sav->sav_npending; i++) {
5510		if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
5511		    &spareguid) == 0 && spareguid == guid)
5512			return (B_TRUE);
5513	}
5514
5515	return (B_FALSE);
5516}
5517
5518/*
5519 * Check if a pool has an active shared spare device.
5520 * Note: reference count of an active spare is 2, as a spare and as a replace
5521 */
5522static boolean_t
5523spa_has_active_shared_spare(spa_t *spa)
5524{
5525	int i, refcnt;
5526	uint64_t pool;
5527	spa_aux_vdev_t *sav = &spa->spa_spares;
5528
5529	for (i = 0; i < sav->sav_count; i++) {
5530		if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
5531		    &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
5532		    refcnt > 2)
5533			return (B_TRUE);
5534	}
5535
5536	return (B_FALSE);
5537}
5538
5539/*
5540 * Post a sysevent corresponding to the given event.  The 'name' must be one of
5541 * the event definitions in sys/sysevent/eventdefs.h.  The payload will be
5542 * filled in from the spa and (optionally) the vdev.  This doesn't do anything
5543 * in the userland libzpool, as we don't want consumers to misinterpret ztest
5544 * or zdb as real changes.
5545 */
5546void
5547spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
5548{
5549#ifdef _KERNEL
5550	sysevent_t		*ev;
5551	sysevent_attr_list_t	*attr = NULL;
5552	sysevent_value_t	value;
5553	sysevent_id_t		eid;
5554
5555	ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
5556	    SE_SLEEP);
5557
5558	value.value_type = SE_DATA_TYPE_STRING;
5559	value.value.sv_string = spa_name(spa);
5560	if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
5561		goto done;
5562
5563	value.value_type = SE_DATA_TYPE_UINT64;
5564	value.value.sv_uint64 = spa_guid(spa);
5565	if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
5566		goto done;
5567
5568	if (vd) {
5569		value.value_type = SE_DATA_TYPE_UINT64;
5570		value.value.sv_uint64 = vd->vdev_guid;
5571		if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
5572		    SE_SLEEP) != 0)
5573			goto done;
5574
5575		if (vd->vdev_path) {
5576			value.value_type = SE_DATA_TYPE_STRING;
5577			value.value.sv_string = vd->vdev_path;
5578			if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
5579			    &value, SE_SLEEP) != 0)
5580				goto done;
5581		}
5582	}
5583
5584	if (sysevent_attach_attributes(ev, attr) != 0)
5585		goto done;
5586	attr = NULL;
5587
5588	(void) log_sysevent(ev, SE_SLEEP, &eid);
5589
5590done:
5591	if (attr)
5592		sysevent_free_attr(attr);
5593	sysevent_free(ev);
5594#endif
5595}
5596