spa.c revision 208044b8d19de4d89e80cb4c8c25bee5c7ab6a82
1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
24 * Use is subject to license terms.
25 */
26
27/*
28 * This file contains all the routines used when modifying on-disk SPA state.
29 * This includes opening, importing, destroying, exporting a pool, and syncing a
30 * pool.
31 */
32
33#include <sys/zfs_context.h>
34#include <sys/fm/fs/zfs.h>
35#include <sys/spa_impl.h>
36#include <sys/zio.h>
37#include <sys/zio_checksum.h>
38#include <sys/dmu.h>
39#include <sys/dmu_tx.h>
40#include <sys/zap.h>
41#include <sys/zil.h>
42#include <sys/ddt.h>
43#include <sys/vdev_impl.h>
44#include <sys/metaslab.h>
45#include <sys/metaslab_impl.h>
46#include <sys/uberblock_impl.h>
47#include <sys/txg.h>
48#include <sys/avl.h>
49#include <sys/dmu_traverse.h>
50#include <sys/dmu_objset.h>
51#include <sys/unique.h>
52#include <sys/dsl_pool.h>
53#include <sys/dsl_dataset.h>
54#include <sys/dsl_dir.h>
55#include <sys/dsl_prop.h>
56#include <sys/dsl_synctask.h>
57#include <sys/fs/zfs.h>
58#include <sys/arc.h>
59#include <sys/callb.h>
60#include <sys/systeminfo.h>
61#include <sys/spa_boot.h>
62#include <sys/zfs_ioctl.h>
63
64#ifdef	_KERNEL
65#include <sys/bootprops.h>
66#include <sys/callb.h>
67#include <sys/cpupart.h>
68#include <sys/pool.h>
69#include <sys/sysdc.h>
70#include <sys/zone.h>
71#endif	/* _KERNEL */
72
73#include "zfs_prop.h"
74#include "zfs_comutil.h"
75
76typedef enum zti_modes {
77	zti_mode_fixed,			/* value is # of threads (min 1) */
78	zti_mode_online_percent,	/* value is % of online CPUs */
79	zti_mode_batch,			/* cpu-intensive; value is ignored */
80	zti_mode_null,			/* don't create a taskq */
81	zti_nmodes
82} zti_modes_t;
83
84#define	ZTI_FIX(n)	{ zti_mode_fixed, (n) }
85#define	ZTI_PCT(n)	{ zti_mode_online_percent, (n) }
86#define	ZTI_BATCH	{ zti_mode_batch, 0 }
87#define	ZTI_NULL	{ zti_mode_null, 0 }
88
89#define	ZTI_ONE		ZTI_FIX(1)
90
91typedef struct zio_taskq_info {
92	enum zti_modes zti_mode;
93	uint_t zti_value;
94} zio_taskq_info_t;
95
96static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
97	"issue", "issue_high", "intr", "intr_high"
98};
99
100/*
101 * Define the taskq threads for the following I/O types:
102 * 	NULL, READ, WRITE, FREE, CLAIM, and IOCTL
103 */
104const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
105	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
106	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
107	{ ZTI_FIX(8),	ZTI_NULL,	ZTI_BATCH,	ZTI_NULL },
108	{ ZTI_BATCH,	ZTI_FIX(5),	ZTI_FIX(8),	ZTI_FIX(5) },
109	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
110	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
111	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
112};
113
114static void spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx);
115static boolean_t spa_has_active_shared_spare(spa_t *spa);
116static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
117    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
118    char **ereport);
119
120uint_t		zio_taskq_batch_pct = 100;	/* 1 thread per cpu in pset */
121id_t		zio_taskq_psrset_bind = PS_NONE;
122boolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
123uint_t		zio_taskq_basedc = 80;		/* base duty cycle */
124
125boolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
126
127/*
128 * This (illegal) pool name is used when temporarily importing a spa_t in order
129 * to get the vdev stats associated with the imported devices.
130 */
131#define	TRYIMPORT_NAME	"$import"
132
133/*
134 * ==========================================================================
135 * SPA properties routines
136 * ==========================================================================
137 */
138
139/*
140 * Add a (source=src, propname=propval) list to an nvlist.
141 */
142static void
143spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
144    uint64_t intval, zprop_source_t src)
145{
146	const char *propname = zpool_prop_to_name(prop);
147	nvlist_t *propval;
148
149	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
150	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
151
152	if (strval != NULL)
153		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
154	else
155		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
156
157	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
158	nvlist_free(propval);
159}
160
161/*
162 * Get property values from the spa configuration.
163 */
164static void
165spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
166{
167	uint64_t size;
168	uint64_t alloc;
169	uint64_t cap, version;
170	zprop_source_t src = ZPROP_SRC_NONE;
171	spa_config_dirent_t *dp;
172
173	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
174
175	if (spa->spa_root_vdev != NULL) {
176		alloc = metaslab_class_get_alloc(spa_normal_class(spa));
177		size = metaslab_class_get_space(spa_normal_class(spa));
178		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
179		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
180		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
181		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
182		    size - alloc, src);
183
184		cap = (size == 0) ? 0 : (alloc * 100 / size);
185		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
186
187		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
188		    ddt_get_pool_dedup_ratio(spa), src);
189
190		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
191		    spa->spa_root_vdev->vdev_state, src);
192
193		version = spa_version(spa);
194		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
195			src = ZPROP_SRC_DEFAULT;
196		else
197			src = ZPROP_SRC_LOCAL;
198		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
199	}
200
201	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
202
203	if (spa->spa_root != NULL)
204		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
205		    0, ZPROP_SRC_LOCAL);
206
207	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
208		if (dp->scd_path == NULL) {
209			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
210			    "none", 0, ZPROP_SRC_LOCAL);
211		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
212			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
213			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
214		}
215	}
216}
217
218/*
219 * Get zpool property values.
220 */
221int
222spa_prop_get(spa_t *spa, nvlist_t **nvp)
223{
224	objset_t *mos = spa->spa_meta_objset;
225	zap_cursor_t zc;
226	zap_attribute_t za;
227	int err;
228
229	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
230
231	mutex_enter(&spa->spa_props_lock);
232
233	/*
234	 * Get properties from the spa config.
235	 */
236	spa_prop_get_config(spa, nvp);
237
238	/* If no pool property object, no more prop to get. */
239	if (mos == NULL || spa->spa_pool_props_object == 0) {
240		mutex_exit(&spa->spa_props_lock);
241		return (0);
242	}
243
244	/*
245	 * Get properties from the MOS pool property object.
246	 */
247	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
248	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
249	    zap_cursor_advance(&zc)) {
250		uint64_t intval = 0;
251		char *strval = NULL;
252		zprop_source_t src = ZPROP_SRC_DEFAULT;
253		zpool_prop_t prop;
254
255		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
256			continue;
257
258		switch (za.za_integer_length) {
259		case 8:
260			/* integer property */
261			if (za.za_first_integer !=
262			    zpool_prop_default_numeric(prop))
263				src = ZPROP_SRC_LOCAL;
264
265			if (prop == ZPOOL_PROP_BOOTFS) {
266				dsl_pool_t *dp;
267				dsl_dataset_t *ds = NULL;
268
269				dp = spa_get_dsl(spa);
270				rw_enter(&dp->dp_config_rwlock, RW_READER);
271				if (err = dsl_dataset_hold_obj(dp,
272				    za.za_first_integer, FTAG, &ds)) {
273					rw_exit(&dp->dp_config_rwlock);
274					break;
275				}
276
277				strval = kmem_alloc(
278				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
279				    KM_SLEEP);
280				dsl_dataset_name(ds, strval);
281				dsl_dataset_rele(ds, FTAG);
282				rw_exit(&dp->dp_config_rwlock);
283			} else {
284				strval = NULL;
285				intval = za.za_first_integer;
286			}
287
288			spa_prop_add_list(*nvp, prop, strval, intval, src);
289
290			if (strval != NULL)
291				kmem_free(strval,
292				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
293
294			break;
295
296		case 1:
297			/* string property */
298			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
299			err = zap_lookup(mos, spa->spa_pool_props_object,
300			    za.za_name, 1, za.za_num_integers, strval);
301			if (err) {
302				kmem_free(strval, za.za_num_integers);
303				break;
304			}
305			spa_prop_add_list(*nvp, prop, strval, 0, src);
306			kmem_free(strval, za.za_num_integers);
307			break;
308
309		default:
310			break;
311		}
312	}
313	zap_cursor_fini(&zc);
314	mutex_exit(&spa->spa_props_lock);
315out:
316	if (err && err != ENOENT) {
317		nvlist_free(*nvp);
318		*nvp = NULL;
319		return (err);
320	}
321
322	return (0);
323}
324
325/*
326 * Validate the given pool properties nvlist and modify the list
327 * for the property values to be set.
328 */
329static int
330spa_prop_validate(spa_t *spa, nvlist_t *props)
331{
332	nvpair_t *elem;
333	int error = 0, reset_bootfs = 0;
334	uint64_t objnum;
335
336	elem = NULL;
337	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
338		zpool_prop_t prop;
339		char *propname, *strval;
340		uint64_t intval;
341		objset_t *os;
342		char *slash;
343
344		propname = nvpair_name(elem);
345
346		if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
347			return (EINVAL);
348
349		switch (prop) {
350		case ZPOOL_PROP_VERSION:
351			error = nvpair_value_uint64(elem, &intval);
352			if (!error &&
353			    (intval < spa_version(spa) || intval > SPA_VERSION))
354				error = EINVAL;
355			break;
356
357		case ZPOOL_PROP_DELEGATION:
358		case ZPOOL_PROP_AUTOREPLACE:
359		case ZPOOL_PROP_LISTSNAPS:
360		case ZPOOL_PROP_AUTOEXPAND:
361			error = nvpair_value_uint64(elem, &intval);
362			if (!error && intval > 1)
363				error = EINVAL;
364			break;
365
366		case ZPOOL_PROP_BOOTFS:
367			/*
368			 * If the pool version is less than SPA_VERSION_BOOTFS,
369			 * or the pool is still being created (version == 0),
370			 * the bootfs property cannot be set.
371			 */
372			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
373				error = ENOTSUP;
374				break;
375			}
376
377			/*
378			 * Make sure the vdev config is bootable
379			 */
380			if (!vdev_is_bootable(spa->spa_root_vdev)) {
381				error = ENOTSUP;
382				break;
383			}
384
385			reset_bootfs = 1;
386
387			error = nvpair_value_string(elem, &strval);
388
389			if (!error) {
390				uint64_t compress;
391
392				if (strval == NULL || strval[0] == '\0') {
393					objnum = zpool_prop_default_numeric(
394					    ZPOOL_PROP_BOOTFS);
395					break;
396				}
397
398				if (error = dmu_objset_hold(strval, FTAG, &os))
399					break;
400
401				/* Must be ZPL and not gzip compressed. */
402
403				if (dmu_objset_type(os) != DMU_OST_ZFS) {
404					error = ENOTSUP;
405				} else if ((error = dsl_prop_get_integer(strval,
406				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
407				    &compress, NULL)) == 0 &&
408				    !BOOTFS_COMPRESS_VALID(compress)) {
409					error = ENOTSUP;
410				} else {
411					objnum = dmu_objset_id(os);
412				}
413				dmu_objset_rele(os, FTAG);
414			}
415			break;
416
417		case ZPOOL_PROP_FAILUREMODE:
418			error = nvpair_value_uint64(elem, &intval);
419			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
420			    intval > ZIO_FAILURE_MODE_PANIC))
421				error = EINVAL;
422
423			/*
424			 * This is a special case which only occurs when
425			 * the pool has completely failed. This allows
426			 * the user to change the in-core failmode property
427			 * without syncing it out to disk (I/Os might
428			 * currently be blocked). We do this by returning
429			 * EIO to the caller (spa_prop_set) to trick it
430			 * into thinking we encountered a property validation
431			 * error.
432			 */
433			if (!error && spa_suspended(spa)) {
434				spa->spa_failmode = intval;
435				error = EIO;
436			}
437			break;
438
439		case ZPOOL_PROP_CACHEFILE:
440			if ((error = nvpair_value_string(elem, &strval)) != 0)
441				break;
442
443			if (strval[0] == '\0')
444				break;
445
446			if (strcmp(strval, "none") == 0)
447				break;
448
449			if (strval[0] != '/') {
450				error = EINVAL;
451				break;
452			}
453
454			slash = strrchr(strval, '/');
455			ASSERT(slash != NULL);
456
457			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
458			    strcmp(slash, "/..") == 0)
459				error = EINVAL;
460			break;
461
462		case ZPOOL_PROP_DEDUPDITTO:
463			if (spa_version(spa) < SPA_VERSION_DEDUP)
464				error = ENOTSUP;
465			else
466				error = nvpair_value_uint64(elem, &intval);
467			if (error == 0 &&
468			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
469				error = EINVAL;
470			break;
471		}
472
473		if (error)
474			break;
475	}
476
477	if (!error && reset_bootfs) {
478		error = nvlist_remove(props,
479		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
480
481		if (!error) {
482			error = nvlist_add_uint64(props,
483			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
484		}
485	}
486
487	return (error);
488}
489
490void
491spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
492{
493	char *cachefile;
494	spa_config_dirent_t *dp;
495
496	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
497	    &cachefile) != 0)
498		return;
499
500	dp = kmem_alloc(sizeof (spa_config_dirent_t),
501	    KM_SLEEP);
502
503	if (cachefile[0] == '\0')
504		dp->scd_path = spa_strdup(spa_config_path);
505	else if (strcmp(cachefile, "none") == 0)
506		dp->scd_path = NULL;
507	else
508		dp->scd_path = spa_strdup(cachefile);
509
510	list_insert_head(&spa->spa_config_list, dp);
511	if (need_sync)
512		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
513}
514
515int
516spa_prop_set(spa_t *spa, nvlist_t *nvp)
517{
518	int error;
519	nvpair_t *elem;
520	boolean_t need_sync = B_FALSE;
521	zpool_prop_t prop;
522
523	if ((error = spa_prop_validate(spa, nvp)) != 0)
524		return (error);
525
526	elem = NULL;
527	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
528		if ((prop = zpool_name_to_prop(
529		    nvpair_name(elem))) == ZPROP_INVAL)
530			return (EINVAL);
531
532		if (prop == ZPOOL_PROP_CACHEFILE || prop == ZPOOL_PROP_ALTROOT)
533			continue;
534
535		need_sync = B_TRUE;
536		break;
537	}
538
539	if (need_sync)
540		return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
541		    spa, nvp, 3));
542	else
543		return (0);
544}
545
546/*
547 * If the bootfs property value is dsobj, clear it.
548 */
549void
550spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
551{
552	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
553		VERIFY(zap_remove(spa->spa_meta_objset,
554		    spa->spa_pool_props_object,
555		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
556		spa->spa_bootfs = 0;
557	}
558}
559
560/*
561 * ==========================================================================
562 * SPA state manipulation (open/create/destroy/import/export)
563 * ==========================================================================
564 */
565
566static int
567spa_error_entry_compare(const void *a, const void *b)
568{
569	spa_error_entry_t *sa = (spa_error_entry_t *)a;
570	spa_error_entry_t *sb = (spa_error_entry_t *)b;
571	int ret;
572
573	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
574	    sizeof (zbookmark_t));
575
576	if (ret < 0)
577		return (-1);
578	else if (ret > 0)
579		return (1);
580	else
581		return (0);
582}
583
584/*
585 * Utility function which retrieves copies of the current logs and
586 * re-initializes them in the process.
587 */
588void
589spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
590{
591	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
592
593	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
594	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
595
596	avl_create(&spa->spa_errlist_scrub,
597	    spa_error_entry_compare, sizeof (spa_error_entry_t),
598	    offsetof(spa_error_entry_t, se_avl));
599	avl_create(&spa->spa_errlist_last,
600	    spa_error_entry_compare, sizeof (spa_error_entry_t),
601	    offsetof(spa_error_entry_t, se_avl));
602}
603
604static taskq_t *
605spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
606    uint_t value)
607{
608	uint_t flags = TASKQ_PREPOPULATE;
609	boolean_t batch = B_FALSE;
610
611	switch (mode) {
612	case zti_mode_null:
613		return (NULL);		/* no taskq needed */
614
615	case zti_mode_fixed:
616		ASSERT3U(value, >=, 1);
617		value = MAX(value, 1);
618		break;
619
620	case zti_mode_batch:
621		batch = B_TRUE;
622		flags |= TASKQ_THREADS_CPU_PCT;
623		value = zio_taskq_batch_pct;
624		break;
625
626	case zti_mode_online_percent:
627		flags |= TASKQ_THREADS_CPU_PCT;
628		break;
629
630	default:
631		panic("unrecognized mode for %s taskq (%u:%u) in "
632		    "spa_activate()",
633		    name, mode, value);
634		break;
635	}
636
637	if (zio_taskq_sysdc && spa->spa_proc != &p0) {
638		if (batch)
639			flags |= TASKQ_DC_BATCH;
640
641		return (taskq_create_sysdc(name, value, 50, INT_MAX,
642		    spa->spa_proc, zio_taskq_basedc, flags));
643	}
644	return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
645	    spa->spa_proc, flags));
646}
647
648static void
649spa_create_zio_taskqs(spa_t *spa)
650{
651	for (int t = 0; t < ZIO_TYPES; t++) {
652		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
653			const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
654			enum zti_modes mode = ztip->zti_mode;
655			uint_t value = ztip->zti_value;
656			char name[32];
657
658			(void) snprintf(name, sizeof (name),
659			    "%s_%s", zio_type_name[t], zio_taskq_types[q]);
660
661			spa->spa_zio_taskq[t][q] =
662			    spa_taskq_create(spa, name, mode, value);
663		}
664	}
665}
666
667#ifdef _KERNEL
668static void
669spa_thread(void *arg)
670{
671	callb_cpr_t cprinfo;
672
673	spa_t *spa = arg;
674	user_t *pu = PTOU(curproc);
675
676	CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
677	    spa->spa_name);
678
679	ASSERT(curproc != &p0);
680	(void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
681	    "zpool-%s", spa->spa_name);
682	(void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
683
684	/* bind this thread to the requested psrset */
685	if (zio_taskq_psrset_bind != PS_NONE) {
686		pool_lock();
687		mutex_enter(&cpu_lock);
688		mutex_enter(&pidlock);
689		mutex_enter(&curproc->p_lock);
690
691		if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
692		    0, NULL, NULL) == 0)  {
693			curthread->t_bind_pset = zio_taskq_psrset_bind;
694		} else {
695			cmn_err(CE_WARN,
696			    "Couldn't bind process for zfs pool \"%s\" to "
697			    "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
698		}
699
700		mutex_exit(&curproc->p_lock);
701		mutex_exit(&pidlock);
702		mutex_exit(&cpu_lock);
703		pool_unlock();
704	}
705
706	if (zio_taskq_sysdc) {
707		sysdc_thread_enter(curthread, 100, 0);
708	}
709
710	spa->spa_proc = curproc;
711	spa->spa_did = curthread->t_did;
712
713	spa_create_zio_taskqs(spa);
714
715	mutex_enter(&spa->spa_proc_lock);
716	ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
717
718	spa->spa_proc_state = SPA_PROC_ACTIVE;
719	cv_broadcast(&spa->spa_proc_cv);
720
721	CALLB_CPR_SAFE_BEGIN(&cprinfo);
722	while (spa->spa_proc_state == SPA_PROC_ACTIVE)
723		cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
724	CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
725
726	ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
727	spa->spa_proc_state = SPA_PROC_GONE;
728	spa->spa_proc = &p0;
729	cv_broadcast(&spa->spa_proc_cv);
730	CALLB_CPR_EXIT(&cprinfo);	/* drops spa_proc_lock */
731
732	mutex_enter(&curproc->p_lock);
733	lwp_exit();
734}
735#endif
736
737/*
738 * Activate an uninitialized pool.
739 */
740static void
741spa_activate(spa_t *spa, int mode)
742{
743	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
744
745	spa->spa_state = POOL_STATE_ACTIVE;
746	spa->spa_mode = mode;
747
748	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
749	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
750
751	/* Try to create a covering process */
752	mutex_enter(&spa->spa_proc_lock);
753	ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
754	ASSERT(spa->spa_proc == &p0);
755	spa->spa_did = 0;
756
757	/* Only create a process if we're going to be around a while. */
758	if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
759		if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
760		    NULL, 0) == 0) {
761			spa->spa_proc_state = SPA_PROC_CREATED;
762			while (spa->spa_proc_state == SPA_PROC_CREATED) {
763				cv_wait(&spa->spa_proc_cv,
764				    &spa->spa_proc_lock);
765			}
766			ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
767			ASSERT(spa->spa_proc != &p0);
768			ASSERT(spa->spa_did != 0);
769		} else {
770#ifdef _KERNEL
771			cmn_err(CE_WARN,
772			    "Couldn't create process for zfs pool \"%s\"\n",
773			    spa->spa_name);
774#endif
775		}
776	}
777	mutex_exit(&spa->spa_proc_lock);
778
779	/* If we didn't create a process, we need to create our taskqs. */
780	if (spa->spa_proc == &p0) {
781		spa_create_zio_taskqs(spa);
782	}
783
784	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
785	    offsetof(vdev_t, vdev_config_dirty_node));
786	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
787	    offsetof(vdev_t, vdev_state_dirty_node));
788
789	txg_list_create(&spa->spa_vdev_txg_list,
790	    offsetof(struct vdev, vdev_txg_node));
791
792	avl_create(&spa->spa_errlist_scrub,
793	    spa_error_entry_compare, sizeof (spa_error_entry_t),
794	    offsetof(spa_error_entry_t, se_avl));
795	avl_create(&spa->spa_errlist_last,
796	    spa_error_entry_compare, sizeof (spa_error_entry_t),
797	    offsetof(spa_error_entry_t, se_avl));
798}
799
800/*
801 * Opposite of spa_activate().
802 */
803static void
804spa_deactivate(spa_t *spa)
805{
806	ASSERT(spa->spa_sync_on == B_FALSE);
807	ASSERT(spa->spa_dsl_pool == NULL);
808	ASSERT(spa->spa_root_vdev == NULL);
809	ASSERT(spa->spa_async_zio_root == NULL);
810	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
811
812	txg_list_destroy(&spa->spa_vdev_txg_list);
813
814	list_destroy(&spa->spa_config_dirty_list);
815	list_destroy(&spa->spa_state_dirty_list);
816
817	for (int t = 0; t < ZIO_TYPES; t++) {
818		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
819			if (spa->spa_zio_taskq[t][q] != NULL)
820				taskq_destroy(spa->spa_zio_taskq[t][q]);
821			spa->spa_zio_taskq[t][q] = NULL;
822		}
823	}
824
825	metaslab_class_destroy(spa->spa_normal_class);
826	spa->spa_normal_class = NULL;
827
828	metaslab_class_destroy(spa->spa_log_class);
829	spa->spa_log_class = NULL;
830
831	/*
832	 * If this was part of an import or the open otherwise failed, we may
833	 * still have errors left in the queues.  Empty them just in case.
834	 */
835	spa_errlog_drain(spa);
836
837	avl_destroy(&spa->spa_errlist_scrub);
838	avl_destroy(&spa->spa_errlist_last);
839
840	spa->spa_state = POOL_STATE_UNINITIALIZED;
841
842	mutex_enter(&spa->spa_proc_lock);
843	if (spa->spa_proc_state != SPA_PROC_NONE) {
844		ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
845		spa->spa_proc_state = SPA_PROC_DEACTIVATE;
846		cv_broadcast(&spa->spa_proc_cv);
847		while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
848			ASSERT(spa->spa_proc != &p0);
849			cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
850		}
851		ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
852		spa->spa_proc_state = SPA_PROC_NONE;
853	}
854	ASSERT(spa->spa_proc == &p0);
855	mutex_exit(&spa->spa_proc_lock);
856
857	/*
858	 * We want to make sure spa_thread() has actually exited the ZFS
859	 * module, so that the module can't be unloaded out from underneath
860	 * it.
861	 */
862	if (spa->spa_did != 0) {
863		thread_join(spa->spa_did);
864		spa->spa_did = 0;
865	}
866}
867
868/*
869 * Verify a pool configuration, and construct the vdev tree appropriately.  This
870 * will create all the necessary vdevs in the appropriate layout, with each vdev
871 * in the CLOSED state.  This will prep the pool before open/creation/import.
872 * All vdev validation is done by the vdev_alloc() routine.
873 */
874static int
875spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
876    uint_t id, int atype)
877{
878	nvlist_t **child;
879	uint_t children;
880	int error;
881
882	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
883		return (error);
884
885	if ((*vdp)->vdev_ops->vdev_op_leaf)
886		return (0);
887
888	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
889	    &child, &children);
890
891	if (error == ENOENT)
892		return (0);
893
894	if (error) {
895		vdev_free(*vdp);
896		*vdp = NULL;
897		return (EINVAL);
898	}
899
900	for (int c = 0; c < children; c++) {
901		vdev_t *vd;
902		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
903		    atype)) != 0) {
904			vdev_free(*vdp);
905			*vdp = NULL;
906			return (error);
907		}
908	}
909
910	ASSERT(*vdp != NULL);
911
912	return (0);
913}
914
915/*
916 * Opposite of spa_load().
917 */
918static void
919spa_unload(spa_t *spa)
920{
921	int i;
922
923	ASSERT(MUTEX_HELD(&spa_namespace_lock));
924
925	/*
926	 * Stop async tasks.
927	 */
928	spa_async_suspend(spa);
929
930	/*
931	 * Stop syncing.
932	 */
933	if (spa->spa_sync_on) {
934		txg_sync_stop(spa->spa_dsl_pool);
935		spa->spa_sync_on = B_FALSE;
936	}
937
938	/*
939	 * Wait for any outstanding async I/O to complete.
940	 */
941	if (spa->spa_async_zio_root != NULL) {
942		(void) zio_wait(spa->spa_async_zio_root);
943		spa->spa_async_zio_root = NULL;
944	}
945
946	/*
947	 * Close the dsl pool.
948	 */
949	if (spa->spa_dsl_pool) {
950		dsl_pool_close(spa->spa_dsl_pool);
951		spa->spa_dsl_pool = NULL;
952		spa->spa_meta_objset = NULL;
953	}
954
955	ddt_unload(spa);
956
957	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
958
959	/*
960	 * Drop and purge level 2 cache
961	 */
962	spa_l2cache_drop(spa);
963
964	/*
965	 * Close all vdevs.
966	 */
967	if (spa->spa_root_vdev)
968		vdev_free(spa->spa_root_vdev);
969	ASSERT(spa->spa_root_vdev == NULL);
970
971	for (i = 0; i < spa->spa_spares.sav_count; i++)
972		vdev_free(spa->spa_spares.sav_vdevs[i]);
973	if (spa->spa_spares.sav_vdevs) {
974		kmem_free(spa->spa_spares.sav_vdevs,
975		    spa->spa_spares.sav_count * sizeof (void *));
976		spa->spa_spares.sav_vdevs = NULL;
977	}
978	if (spa->spa_spares.sav_config) {
979		nvlist_free(spa->spa_spares.sav_config);
980		spa->spa_spares.sav_config = NULL;
981	}
982	spa->spa_spares.sav_count = 0;
983
984	for (i = 0; i < spa->spa_l2cache.sav_count; i++)
985		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
986	if (spa->spa_l2cache.sav_vdevs) {
987		kmem_free(spa->spa_l2cache.sav_vdevs,
988		    spa->spa_l2cache.sav_count * sizeof (void *));
989		spa->spa_l2cache.sav_vdevs = NULL;
990	}
991	if (spa->spa_l2cache.sav_config) {
992		nvlist_free(spa->spa_l2cache.sav_config);
993		spa->spa_l2cache.sav_config = NULL;
994	}
995	spa->spa_l2cache.sav_count = 0;
996
997	spa->spa_async_suspended = 0;
998
999	spa_config_exit(spa, SCL_ALL, FTAG);
1000}
1001
1002/*
1003 * Load (or re-load) the current list of vdevs describing the active spares for
1004 * this pool.  When this is called, we have some form of basic information in
1005 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1006 * then re-generate a more complete list including status information.
1007 */
1008static void
1009spa_load_spares(spa_t *spa)
1010{
1011	nvlist_t **spares;
1012	uint_t nspares;
1013	int i;
1014	vdev_t *vd, *tvd;
1015
1016	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1017
1018	/*
1019	 * First, close and free any existing spare vdevs.
1020	 */
1021	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1022		vd = spa->spa_spares.sav_vdevs[i];
1023
1024		/* Undo the call to spa_activate() below */
1025		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1026		    B_FALSE)) != NULL && tvd->vdev_isspare)
1027			spa_spare_remove(tvd);
1028		vdev_close(vd);
1029		vdev_free(vd);
1030	}
1031
1032	if (spa->spa_spares.sav_vdevs)
1033		kmem_free(spa->spa_spares.sav_vdevs,
1034		    spa->spa_spares.sav_count * sizeof (void *));
1035
1036	if (spa->spa_spares.sav_config == NULL)
1037		nspares = 0;
1038	else
1039		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1040		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1041
1042	spa->spa_spares.sav_count = (int)nspares;
1043	spa->spa_spares.sav_vdevs = NULL;
1044
1045	if (nspares == 0)
1046		return;
1047
1048	/*
1049	 * Construct the array of vdevs, opening them to get status in the
1050	 * process.   For each spare, there is potentially two different vdev_t
1051	 * structures associated with it: one in the list of spares (used only
1052	 * for basic validation purposes) and one in the active vdev
1053	 * configuration (if it's spared in).  During this phase we open and
1054	 * validate each vdev on the spare list.  If the vdev also exists in the
1055	 * active configuration, then we also mark this vdev as an active spare.
1056	 */
1057	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1058	    KM_SLEEP);
1059	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1060		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1061		    VDEV_ALLOC_SPARE) == 0);
1062		ASSERT(vd != NULL);
1063
1064		spa->spa_spares.sav_vdevs[i] = vd;
1065
1066		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1067		    B_FALSE)) != NULL) {
1068			if (!tvd->vdev_isspare)
1069				spa_spare_add(tvd);
1070
1071			/*
1072			 * We only mark the spare active if we were successfully
1073			 * able to load the vdev.  Otherwise, importing a pool
1074			 * with a bad active spare would result in strange
1075			 * behavior, because multiple pool would think the spare
1076			 * is actively in use.
1077			 *
1078			 * There is a vulnerability here to an equally bizarre
1079			 * circumstance, where a dead active spare is later
1080			 * brought back to life (onlined or otherwise).  Given
1081			 * the rarity of this scenario, and the extra complexity
1082			 * it adds, we ignore the possibility.
1083			 */
1084			if (!vdev_is_dead(tvd))
1085				spa_spare_activate(tvd);
1086		}
1087
1088		vd->vdev_top = vd;
1089		vd->vdev_aux = &spa->spa_spares;
1090
1091		if (vdev_open(vd) != 0)
1092			continue;
1093
1094		if (vdev_validate_aux(vd) == 0)
1095			spa_spare_add(vd);
1096	}
1097
1098	/*
1099	 * Recompute the stashed list of spares, with status information
1100	 * this time.
1101	 */
1102	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1103	    DATA_TYPE_NVLIST_ARRAY) == 0);
1104
1105	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1106	    KM_SLEEP);
1107	for (i = 0; i < spa->spa_spares.sav_count; i++)
1108		spares[i] = vdev_config_generate(spa,
1109		    spa->spa_spares.sav_vdevs[i], B_TRUE, B_TRUE, B_FALSE);
1110	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1111	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1112	for (i = 0; i < spa->spa_spares.sav_count; i++)
1113		nvlist_free(spares[i]);
1114	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1115}
1116
1117/*
1118 * Load (or re-load) the current list of vdevs describing the active l2cache for
1119 * this pool.  When this is called, we have some form of basic information in
1120 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1121 * then re-generate a more complete list including status information.
1122 * Devices which are already active have their details maintained, and are
1123 * not re-opened.
1124 */
1125static void
1126spa_load_l2cache(spa_t *spa)
1127{
1128	nvlist_t **l2cache;
1129	uint_t nl2cache;
1130	int i, j, oldnvdevs;
1131	uint64_t guid;
1132	vdev_t *vd, **oldvdevs, **newvdevs;
1133	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1134
1135	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1136
1137	if (sav->sav_config != NULL) {
1138		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1139		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1140		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1141	} else {
1142		nl2cache = 0;
1143	}
1144
1145	oldvdevs = sav->sav_vdevs;
1146	oldnvdevs = sav->sav_count;
1147	sav->sav_vdevs = NULL;
1148	sav->sav_count = 0;
1149
1150	/*
1151	 * Process new nvlist of vdevs.
1152	 */
1153	for (i = 0; i < nl2cache; i++) {
1154		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1155		    &guid) == 0);
1156
1157		newvdevs[i] = NULL;
1158		for (j = 0; j < oldnvdevs; j++) {
1159			vd = oldvdevs[j];
1160			if (vd != NULL && guid == vd->vdev_guid) {
1161				/*
1162				 * Retain previous vdev for add/remove ops.
1163				 */
1164				newvdevs[i] = vd;
1165				oldvdevs[j] = NULL;
1166				break;
1167			}
1168		}
1169
1170		if (newvdevs[i] == NULL) {
1171			/*
1172			 * Create new vdev
1173			 */
1174			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1175			    VDEV_ALLOC_L2CACHE) == 0);
1176			ASSERT(vd != NULL);
1177			newvdevs[i] = vd;
1178
1179			/*
1180			 * Commit this vdev as an l2cache device,
1181			 * even if it fails to open.
1182			 */
1183			spa_l2cache_add(vd);
1184
1185			vd->vdev_top = vd;
1186			vd->vdev_aux = sav;
1187
1188			spa_l2cache_activate(vd);
1189
1190			if (vdev_open(vd) != 0)
1191				continue;
1192
1193			(void) vdev_validate_aux(vd);
1194
1195			if (!vdev_is_dead(vd))
1196				l2arc_add_vdev(spa, vd);
1197		}
1198	}
1199
1200	/*
1201	 * Purge vdevs that were dropped
1202	 */
1203	for (i = 0; i < oldnvdevs; i++) {
1204		uint64_t pool;
1205
1206		vd = oldvdevs[i];
1207		if (vd != NULL) {
1208			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1209			    pool != 0ULL && l2arc_vdev_present(vd))
1210				l2arc_remove_vdev(vd);
1211			(void) vdev_close(vd);
1212			spa_l2cache_remove(vd);
1213		}
1214	}
1215
1216	if (oldvdevs)
1217		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1218
1219	if (sav->sav_config == NULL)
1220		goto out;
1221
1222	sav->sav_vdevs = newvdevs;
1223	sav->sav_count = (int)nl2cache;
1224
1225	/*
1226	 * Recompute the stashed list of l2cache devices, with status
1227	 * information this time.
1228	 */
1229	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1230	    DATA_TYPE_NVLIST_ARRAY) == 0);
1231
1232	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1233	for (i = 0; i < sav->sav_count; i++)
1234		l2cache[i] = vdev_config_generate(spa,
1235		    sav->sav_vdevs[i], B_TRUE, B_FALSE, B_TRUE);
1236	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1237	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1238out:
1239	for (i = 0; i < sav->sav_count; i++)
1240		nvlist_free(l2cache[i]);
1241	if (sav->sav_count)
1242		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1243}
1244
1245static int
1246load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1247{
1248	dmu_buf_t *db;
1249	char *packed = NULL;
1250	size_t nvsize = 0;
1251	int error;
1252	*value = NULL;
1253
1254	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1255	nvsize = *(uint64_t *)db->db_data;
1256	dmu_buf_rele(db, FTAG);
1257
1258	packed = kmem_alloc(nvsize, KM_SLEEP);
1259	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1260	    DMU_READ_PREFETCH);
1261	if (error == 0)
1262		error = nvlist_unpack(packed, nvsize, value, 0);
1263	kmem_free(packed, nvsize);
1264
1265	return (error);
1266}
1267
1268/*
1269 * Checks to see if the given vdev could not be opened, in which case we post a
1270 * sysevent to notify the autoreplace code that the device has been removed.
1271 */
1272static void
1273spa_check_removed(vdev_t *vd)
1274{
1275	for (int c = 0; c < vd->vdev_children; c++)
1276		spa_check_removed(vd->vdev_child[c]);
1277
1278	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1279		zfs_post_autoreplace(vd->vdev_spa, vd);
1280		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1281	}
1282}
1283
1284/*
1285 * Load the slog device state from the config object since it's possible
1286 * that the label does not contain the most up-to-date information.
1287 */
1288void
1289spa_load_log_state(spa_t *spa, nvlist_t *nv)
1290{
1291	vdev_t *ovd, *rvd = spa->spa_root_vdev;
1292
1293	/*
1294	 * Load the original root vdev tree from the passed config.
1295	 */
1296	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1297	VERIFY(spa_config_parse(spa, &ovd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1298
1299	for (int c = 0; c < rvd->vdev_children; c++) {
1300		vdev_t *cvd = rvd->vdev_child[c];
1301		if (cvd->vdev_islog)
1302			vdev_load_log_state(cvd, ovd->vdev_child[c]);
1303	}
1304	vdev_free(ovd);
1305	spa_config_exit(spa, SCL_ALL, FTAG);
1306}
1307
1308/*
1309 * Check for missing log devices
1310 */
1311int
1312spa_check_logs(spa_t *spa)
1313{
1314	switch (spa->spa_log_state) {
1315	case SPA_LOG_MISSING:
1316		/* need to recheck in case slog has been restored */
1317	case SPA_LOG_UNKNOWN:
1318		if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1319		    DS_FIND_CHILDREN)) {
1320			spa_set_log_state(spa, SPA_LOG_MISSING);
1321			return (1);
1322		}
1323		break;
1324	}
1325	return (0);
1326}
1327
1328static boolean_t
1329spa_passivate_log(spa_t *spa)
1330{
1331	vdev_t *rvd = spa->spa_root_vdev;
1332	boolean_t slog_found = B_FALSE;
1333
1334	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1335
1336	if (!spa_has_slogs(spa))
1337		return (B_FALSE);
1338
1339	for (int c = 0; c < rvd->vdev_children; c++) {
1340		vdev_t *tvd = rvd->vdev_child[c];
1341		metaslab_group_t *mg = tvd->vdev_mg;
1342
1343		if (tvd->vdev_islog) {
1344			metaslab_group_passivate(mg);
1345			slog_found = B_TRUE;
1346		}
1347	}
1348
1349	return (slog_found);
1350}
1351
1352static void
1353spa_activate_log(spa_t *spa)
1354{
1355	vdev_t *rvd = spa->spa_root_vdev;
1356
1357	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1358
1359	for (int c = 0; c < rvd->vdev_children; c++) {
1360		vdev_t *tvd = rvd->vdev_child[c];
1361		metaslab_group_t *mg = tvd->vdev_mg;
1362
1363		if (tvd->vdev_islog)
1364			metaslab_group_activate(mg);
1365	}
1366}
1367
1368int
1369spa_offline_log(spa_t *spa)
1370{
1371	int error = 0;
1372
1373	if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1374	    NULL, DS_FIND_CHILDREN)) == 0) {
1375
1376		/*
1377		 * We successfully offlined the log device, sync out the
1378		 * current txg so that the "stubby" block can be removed
1379		 * by zil_sync().
1380		 */
1381		txg_wait_synced(spa->spa_dsl_pool, 0);
1382	}
1383	return (error);
1384}
1385
1386static void
1387spa_aux_check_removed(spa_aux_vdev_t *sav)
1388{
1389	for (int i = 0; i < sav->sav_count; i++)
1390		spa_check_removed(sav->sav_vdevs[i]);
1391}
1392
1393void
1394spa_claim_notify(zio_t *zio)
1395{
1396	spa_t *spa = zio->io_spa;
1397
1398	if (zio->io_error)
1399		return;
1400
1401	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1402	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1403		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1404	mutex_exit(&spa->spa_props_lock);
1405}
1406
1407typedef struct spa_load_error {
1408	uint64_t	sle_meta_count;
1409	uint64_t	sle_data_count;
1410} spa_load_error_t;
1411
1412static void
1413spa_load_verify_done(zio_t *zio)
1414{
1415	blkptr_t *bp = zio->io_bp;
1416	spa_load_error_t *sle = zio->io_private;
1417	dmu_object_type_t type = BP_GET_TYPE(bp);
1418	int error = zio->io_error;
1419
1420	if (error) {
1421		if ((BP_GET_LEVEL(bp) != 0 || dmu_ot[type].ot_metadata) &&
1422		    type != DMU_OT_INTENT_LOG)
1423			atomic_add_64(&sle->sle_meta_count, 1);
1424		else
1425			atomic_add_64(&sle->sle_data_count, 1);
1426	}
1427	zio_data_buf_free(zio->io_data, zio->io_size);
1428}
1429
1430/*ARGSUSED*/
1431static int
1432spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1433    const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1434{
1435	if (bp != NULL) {
1436		zio_t *rio = arg;
1437		size_t size = BP_GET_PSIZE(bp);
1438		void *data = zio_data_buf_alloc(size);
1439
1440		zio_nowait(zio_read(rio, spa, bp, data, size,
1441		    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1442		    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1443		    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1444	}
1445	return (0);
1446}
1447
1448static int
1449spa_load_verify(spa_t *spa)
1450{
1451	zio_t *rio;
1452	spa_load_error_t sle = { 0 };
1453	zpool_rewind_policy_t policy;
1454	boolean_t verify_ok = B_FALSE;
1455	int error;
1456
1457	zpool_get_rewind_policy(spa->spa_config, &policy);
1458
1459	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1460		return (0);
1461
1462	rio = zio_root(spa, NULL, &sle,
1463	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1464
1465	error = traverse_pool(spa, spa->spa_verify_min_txg,
1466	    TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1467
1468	(void) zio_wait(rio);
1469
1470	spa->spa_load_meta_errors = sle.sle_meta_count;
1471	spa->spa_load_data_errors = sle.sle_data_count;
1472
1473	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1474	    sle.sle_data_count <= policy.zrp_maxdata) {
1475		verify_ok = B_TRUE;
1476		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1477		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1478	} else {
1479		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1480	}
1481
1482	if (error) {
1483		if (error != ENXIO && error != EIO)
1484			error = EIO;
1485		return (error);
1486	}
1487
1488	return (verify_ok ? 0 : EIO);
1489}
1490
1491/*
1492 * Find a value in the pool props object.
1493 */
1494static void
1495spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1496{
1497	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1498	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1499}
1500
1501/*
1502 * Find a value in the pool directory object.
1503 */
1504static int
1505spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1506{
1507	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1508	    name, sizeof (uint64_t), 1, val));
1509}
1510
1511static int
1512spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1513{
1514	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1515	return (err);
1516}
1517
1518/*
1519 * Fix up config after a partly-completed split.  This is done with the
1520 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1521 * pool have that entry in their config, but only the splitting one contains
1522 * a list of all the guids of the vdevs that are being split off.
1523 *
1524 * This function determines what to do with that list: either rejoin
1525 * all the disks to the pool, or complete the splitting process.  To attempt
1526 * the rejoin, each disk that is offlined is marked online again, and
1527 * we do a reopen() call.  If the vdev label for every disk that was
1528 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1529 * then we call vdev_split() on each disk, and complete the split.
1530 *
1531 * Otherwise we leave the config alone, with all the vdevs in place in
1532 * the original pool.
1533 */
1534static void
1535spa_try_repair(spa_t *spa, nvlist_t *config)
1536{
1537	uint_t extracted;
1538	uint64_t *glist;
1539	uint_t i, gcount;
1540	nvlist_t *nvl;
1541	vdev_t **vd;
1542	boolean_t attempt_reopen;
1543
1544	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1545		return;
1546
1547	/* check that the config is complete */
1548	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1549	    &glist, &gcount) != 0)
1550		return;
1551
1552	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1553
1554	/* attempt to online all the vdevs & validate */
1555	attempt_reopen = B_TRUE;
1556	for (i = 0; i < gcount; i++) {
1557		if (glist[i] == 0)	/* vdev is hole */
1558			continue;
1559
1560		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1561		if (vd[i] == NULL) {
1562			/*
1563			 * Don't bother attempting to reopen the disks;
1564			 * just do the split.
1565			 */
1566			attempt_reopen = B_FALSE;
1567		} else {
1568			/* attempt to re-online it */
1569			vd[i]->vdev_offline = B_FALSE;
1570		}
1571	}
1572
1573	if (attempt_reopen) {
1574		vdev_reopen(spa->spa_root_vdev);
1575
1576		/* check each device to see what state it's in */
1577		for (extracted = 0, i = 0; i < gcount; i++) {
1578			if (vd[i] != NULL &&
1579			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1580				break;
1581			++extracted;
1582		}
1583	}
1584
1585	/*
1586	 * If every disk has been moved to the new pool, or if we never
1587	 * even attempted to look at them, then we split them off for
1588	 * good.
1589	 */
1590	if (!attempt_reopen || gcount == extracted) {
1591		for (i = 0; i < gcount; i++)
1592			if (vd[i] != NULL)
1593				vdev_split(vd[i]);
1594		vdev_reopen(spa->spa_root_vdev);
1595	}
1596
1597	kmem_free(vd, gcount * sizeof (vdev_t *));
1598}
1599
1600static int
1601spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1602    boolean_t mosconfig)
1603{
1604	nvlist_t *config = spa->spa_config;
1605	char *ereport = FM_EREPORT_ZFS_POOL;
1606	int error;
1607	uint64_t pool_guid;
1608	nvlist_t *nvl;
1609
1610	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1611		return (EINVAL);
1612
1613	/*
1614	 * Versioning wasn't explicitly added to the label until later, so if
1615	 * it's not present treat it as the initial version.
1616	 */
1617	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1618	    &spa->spa_ubsync.ub_version) != 0)
1619		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1620
1621	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1622	    &spa->spa_config_txg);
1623
1624	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1625	    spa_guid_exists(pool_guid, 0)) {
1626		error = EEXIST;
1627	} else {
1628		spa->spa_load_guid = pool_guid;
1629
1630		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1631		    &nvl) == 0) {
1632			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1633			    KM_SLEEP) == 0);
1634		}
1635
1636		error = spa_load_impl(spa, pool_guid, config, state, type,
1637		    mosconfig, &ereport);
1638	}
1639
1640	spa->spa_minref = refcount_count(&spa->spa_refcount);
1641	if (error && error != EBADF)
1642		zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1643	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1644	spa->spa_ena = 0;
1645
1646	return (error);
1647}
1648
1649/*
1650 * Load an existing storage pool, using the pool's builtin spa_config as a
1651 * source of configuration information.
1652 */
1653static int
1654spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1655    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1656    char **ereport)
1657{
1658	int error = 0;
1659	nvlist_t *nvroot = NULL;
1660	vdev_t *rvd;
1661	uberblock_t *ub = &spa->spa_uberblock;
1662	uint64_t config_cache_txg = spa->spa_config_txg;
1663	int orig_mode = spa->spa_mode;
1664	int parse;
1665
1666	/*
1667	 * If this is an untrusted config, access the pool in read-only mode.
1668	 * This prevents things like resilvering recently removed devices.
1669	 */
1670	if (!mosconfig)
1671		spa->spa_mode = FREAD;
1672
1673	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1674
1675	spa->spa_load_state = state;
1676
1677	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1678		return (EINVAL);
1679
1680	parse = (type == SPA_IMPORT_EXISTING ?
1681	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1682
1683	/*
1684	 * Create "The Godfather" zio to hold all async IOs
1685	 */
1686	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1687	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1688
1689	/*
1690	 * Parse the configuration into a vdev tree.  We explicitly set the
1691	 * value that will be returned by spa_version() since parsing the
1692	 * configuration requires knowing the version number.
1693	 */
1694	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1695	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1696	spa_config_exit(spa, SCL_ALL, FTAG);
1697
1698	if (error != 0)
1699		return (error);
1700
1701	ASSERT(spa->spa_root_vdev == rvd);
1702
1703	if (type != SPA_IMPORT_ASSEMBLE) {
1704		ASSERT(spa_guid(spa) == pool_guid);
1705	}
1706
1707	/*
1708	 * Try to open all vdevs, loading each label in the process.
1709	 */
1710	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1711	error = vdev_open(rvd);
1712	spa_config_exit(spa, SCL_ALL, FTAG);
1713	if (error != 0)
1714		return (error);
1715
1716	/*
1717	 * We need to validate the vdev labels against the configuration that
1718	 * we have in hand, which is dependent on the setting of mosconfig. If
1719	 * mosconfig is true then we're validating the vdev labels based on
1720	 * that config.  Otherwise, we're validating against the cached config
1721	 * (zpool.cache) that was read when we loaded the zfs module, and then
1722	 * later we will recursively call spa_load() and validate against
1723	 * the vdev config.
1724	 *
1725	 * If we're assembling a new pool that's been split off from an
1726	 * existing pool, the labels haven't yet been updated so we skip
1727	 * validation for now.
1728	 */
1729	if (type != SPA_IMPORT_ASSEMBLE) {
1730		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1731		error = vdev_validate(rvd);
1732		spa_config_exit(spa, SCL_ALL, FTAG);
1733
1734		if (error != 0)
1735			return (error);
1736
1737		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1738			return (ENXIO);
1739	}
1740
1741	/*
1742	 * Find the best uberblock.
1743	 */
1744	vdev_uberblock_load(NULL, rvd, ub);
1745
1746	/*
1747	 * If we weren't able to find a single valid uberblock, return failure.
1748	 */
1749	if (ub->ub_txg == 0)
1750		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
1751
1752	/*
1753	 * If the pool is newer than the code, we can't open it.
1754	 */
1755	if (ub->ub_version > SPA_VERSION)
1756		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
1757
1758	/*
1759	 * If the vdev guid sum doesn't match the uberblock, we have an
1760	 * incomplete configuration.
1761	 */
1762	if (mosconfig && type != SPA_IMPORT_ASSEMBLE &&
1763	    rvd->vdev_guid_sum != ub->ub_guid_sum)
1764		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
1765
1766	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
1767		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1768		spa_try_repair(spa, config);
1769		spa_config_exit(spa, SCL_ALL, FTAG);
1770		nvlist_free(spa->spa_config_splitting);
1771		spa->spa_config_splitting = NULL;
1772	}
1773
1774	/*
1775	 * Initialize internal SPA structures.
1776	 */
1777	spa->spa_state = POOL_STATE_ACTIVE;
1778	spa->spa_ubsync = spa->spa_uberblock;
1779	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
1780	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
1781	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
1782	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
1783	spa->spa_claim_max_txg = spa->spa_first_txg;
1784
1785	error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1786	if (error)
1787		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1788	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1789
1790	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
1791		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1792
1793	if (!mosconfig) {
1794		uint64_t hostid;
1795		nvlist_t *policy = NULL, *nvconfig;
1796
1797		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
1798			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1799
1800		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
1801		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
1802			char *hostname;
1803			unsigned long myhostid = 0;
1804
1805			VERIFY(nvlist_lookup_string(nvconfig,
1806			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
1807
1808#ifdef	_KERNEL
1809			myhostid = zone_get_hostid(NULL);
1810#else	/* _KERNEL */
1811			/*
1812			 * We're emulating the system's hostid in userland, so
1813			 * we can't use zone_get_hostid().
1814			 */
1815			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
1816#endif	/* _KERNEL */
1817			if (hostid != 0 && myhostid != 0 &&
1818			    hostid != myhostid) {
1819				nvlist_free(nvconfig);
1820				cmn_err(CE_WARN, "pool '%s' could not be "
1821				    "loaded as it was last accessed by "
1822				    "another system (host: %s hostid: 0x%lx). "
1823				    "See: http://www.sun.com/msg/ZFS-8000-EY",
1824				    spa_name(spa), hostname,
1825				    (unsigned long)hostid);
1826				return (EBADF);
1827			}
1828		}
1829		if (nvlist_lookup_nvlist(spa->spa_config,
1830		    ZPOOL_REWIND_POLICY, &policy) == 0)
1831			VERIFY(nvlist_add_nvlist(nvconfig,
1832			    ZPOOL_REWIND_POLICY, policy) == 0);
1833
1834		spa_config_set(spa, nvconfig);
1835		spa_unload(spa);
1836		spa_deactivate(spa);
1837		spa_activate(spa, orig_mode);
1838
1839		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
1840	}
1841
1842	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPLIST,
1843	    &spa->spa_deferred_bplist_obj) != 0)
1844		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1845
1846	/*
1847	 * Load the bit that tells us to use the new accounting function
1848	 * (raid-z deflation).  If we have an older pool, this will not
1849	 * be present.
1850	 */
1851	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
1852	if (error != 0 && error != ENOENT)
1853		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1854
1855	/*
1856	 * Load the persistent error log.  If we have an older pool, this will
1857	 * not be present.
1858	 */
1859	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
1860	if (error != 0 && error != ENOENT)
1861		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1862
1863	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
1864	    &spa->spa_errlog_scrub);
1865	if (error != 0 && error != ENOENT)
1866		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1867
1868	/*
1869	 * Load the history object.  If we have an older pool, this
1870	 * will not be present.
1871	 */
1872	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
1873	if (error != 0 && error != ENOENT)
1874		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1875
1876	/*
1877	 * If we're assembling the pool from the split-off vdevs of
1878	 * an existing pool, we don't want to attach the spares & cache
1879	 * devices.
1880	 */
1881
1882	/*
1883	 * Load any hot spares for this pool.
1884	 */
1885	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
1886	if (error != 0 && error != ENOENT)
1887		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1888	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
1889		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
1890		if (load_nvlist(spa, spa->spa_spares.sav_object,
1891		    &spa->spa_spares.sav_config) != 0)
1892			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1893
1894		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1895		spa_load_spares(spa);
1896		spa_config_exit(spa, SCL_ALL, FTAG);
1897	} else if (error == 0) {
1898		spa->spa_spares.sav_sync = B_TRUE;
1899	}
1900
1901	/*
1902	 * Load any level 2 ARC devices for this pool.
1903	 */
1904	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
1905	    &spa->spa_l2cache.sav_object);
1906	if (error != 0 && error != ENOENT)
1907		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1908	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
1909		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
1910		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
1911		    &spa->spa_l2cache.sav_config) != 0)
1912			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1913
1914		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1915		spa_load_l2cache(spa);
1916		spa_config_exit(spa, SCL_ALL, FTAG);
1917	} else if (error == 0) {
1918		spa->spa_l2cache.sav_sync = B_TRUE;
1919	}
1920
1921	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
1922
1923	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
1924	if (error && error != ENOENT)
1925		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1926
1927	if (error == 0) {
1928		uint64_t autoreplace;
1929
1930		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
1931		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
1932		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
1933		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
1934		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
1935		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
1936		    &spa->spa_dedup_ditto);
1937
1938		spa->spa_autoreplace = (autoreplace != 0);
1939	}
1940
1941	/*
1942	 * If the 'autoreplace' property is set, then post a resource notifying
1943	 * the ZFS DE that it should not issue any faults for unopenable
1944	 * devices.  We also iterate over the vdevs, and post a sysevent for any
1945	 * unopenable vdevs so that the normal autoreplace handler can take
1946	 * over.
1947	 */
1948	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
1949		spa_check_removed(spa->spa_root_vdev);
1950		/*
1951		 * For the import case, this is done in spa_import(), because
1952		 * at this point we're using the spare definitions from
1953		 * the MOS config, not necessarily from the userland config.
1954		 */
1955		if (state != SPA_LOAD_IMPORT) {
1956			spa_aux_check_removed(&spa->spa_spares);
1957			spa_aux_check_removed(&spa->spa_l2cache);
1958		}
1959	}
1960
1961	/*
1962	 * Load the vdev state for all toplevel vdevs.
1963	 */
1964	vdev_load(rvd);
1965
1966	/*
1967	 * Propagate the leaf DTLs we just loaded all the way up the tree.
1968	 */
1969	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1970	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
1971	spa_config_exit(spa, SCL_ALL, FTAG);
1972
1973	/*
1974	 * Check the state of the root vdev.  If it can't be opened, it
1975	 * indicates one or more toplevel vdevs are faulted.
1976	 */
1977	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1978		return (ENXIO);
1979
1980	/*
1981	 * Load the DDTs (dedup tables).
1982	 */
1983	error = ddt_load(spa);
1984	if (error != 0)
1985		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1986
1987	spa_update_dspace(spa);
1988
1989	if (state != SPA_LOAD_TRYIMPORT) {
1990		error = spa_load_verify(spa);
1991		if (error)
1992			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
1993			    error));
1994	}
1995
1996	/*
1997	 * Load the intent log state and check log integrity.  If we're
1998	 * assembling a pool from a split, the log is not transferred over.
1999	 */
2000	if (type != SPA_IMPORT_ASSEMBLE) {
2001		nvlist_t *nvconfig;
2002
2003		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2004			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2005
2006		VERIFY(nvlist_lookup_nvlist(nvconfig, ZPOOL_CONFIG_VDEV_TREE,
2007		    &nvroot) == 0);
2008		spa_load_log_state(spa, nvroot);
2009		nvlist_free(nvconfig);
2010
2011		if (spa_check_logs(spa)) {
2012			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2013			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2014		}
2015	}
2016
2017	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2018	    spa->spa_load_max_txg == UINT64_MAX)) {
2019		dmu_tx_t *tx;
2020		int need_update = B_FALSE;
2021
2022		ASSERT(state != SPA_LOAD_TRYIMPORT);
2023
2024		/*
2025		 * Claim log blocks that haven't been committed yet.
2026		 * This must all happen in a single txg.
2027		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2028		 * invoked from zil_claim_log_block()'s i/o done callback.
2029		 * Price of rollback is that we abandon the log.
2030		 */
2031		spa->spa_claiming = B_TRUE;
2032
2033		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2034		    spa_first_txg(spa));
2035		(void) dmu_objset_find(spa_name(spa),
2036		    zil_claim, tx, DS_FIND_CHILDREN);
2037		dmu_tx_commit(tx);
2038
2039		spa->spa_claiming = B_FALSE;
2040
2041		spa_set_log_state(spa, SPA_LOG_GOOD);
2042		spa->spa_sync_on = B_TRUE;
2043		txg_sync_start(spa->spa_dsl_pool);
2044
2045		/*
2046		 * Wait for all claims to sync.  We sync up to the highest
2047		 * claimed log block birth time so that claimed log blocks
2048		 * don't appear to be from the future.  spa_claim_max_txg
2049		 * will have been set for us by either zil_check_log_chain()
2050		 * (invoked from spa_check_logs()) or zil_claim() above.
2051		 */
2052		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2053
2054		/*
2055		 * If the config cache is stale, or we have uninitialized
2056		 * metaslabs (see spa_vdev_add()), then update the config.
2057		 *
2058		 * If spa_load_verbatim is true, trust the current
2059		 * in-core spa_config and update the disk labels.
2060		 */
2061		if (config_cache_txg != spa->spa_config_txg ||
2062		    state == SPA_LOAD_IMPORT || spa->spa_load_verbatim ||
2063		    state == SPA_LOAD_RECOVER)
2064			need_update = B_TRUE;
2065
2066		for (int c = 0; c < rvd->vdev_children; c++)
2067			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2068				need_update = B_TRUE;
2069
2070		/*
2071		 * Update the config cache asychronously in case we're the
2072		 * root pool, in which case the config cache isn't writable yet.
2073		 */
2074		if (need_update)
2075			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2076
2077		/*
2078		 * Check all DTLs to see if anything needs resilvering.
2079		 */
2080		if (vdev_resilver_needed(rvd, NULL, NULL))
2081			spa_async_request(spa, SPA_ASYNC_RESILVER);
2082
2083		/*
2084		 * Delete any inconsistent datasets.
2085		 */
2086		(void) dmu_objset_find(spa_name(spa),
2087		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2088
2089		/*
2090		 * Clean up any stale temporary dataset userrefs.
2091		 */
2092		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2093	}
2094
2095	return (0);
2096}
2097
2098static int
2099spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2100{
2101	spa_unload(spa);
2102	spa_deactivate(spa);
2103
2104	spa->spa_load_max_txg--;
2105
2106	spa_activate(spa, spa_mode_global);
2107	spa_async_suspend(spa);
2108
2109	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2110}
2111
2112static int
2113spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2114    uint64_t max_request, int rewind_flags)
2115{
2116	nvlist_t *config = NULL;
2117	int load_error, rewind_error;
2118	uint64_t safe_rewind_txg;
2119	uint64_t min_txg;
2120
2121	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2122		spa->spa_load_max_txg = spa->spa_load_txg;
2123		spa_set_log_state(spa, SPA_LOG_CLEAR);
2124	} else {
2125		spa->spa_load_max_txg = max_request;
2126	}
2127
2128	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2129	    mosconfig);
2130	if (load_error == 0)
2131		return (0);
2132
2133	if (spa->spa_root_vdev != NULL)
2134		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2135
2136	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2137	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2138
2139	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2140		nvlist_free(config);
2141		return (load_error);
2142	}
2143
2144	/* Price of rolling back is discarding txgs, including log */
2145	if (state == SPA_LOAD_RECOVER)
2146		spa_set_log_state(spa, SPA_LOG_CLEAR);
2147
2148	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2149	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2150	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2151	    TXG_INITIAL : safe_rewind_txg;
2152
2153	/*
2154	 * Continue as long as we're finding errors, we're still within
2155	 * the acceptable rewind range, and we're still finding uberblocks
2156	 */
2157	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2158	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2159		if (spa->spa_load_max_txg < safe_rewind_txg)
2160			spa->spa_extreme_rewind = B_TRUE;
2161		rewind_error = spa_load_retry(spa, state, mosconfig);
2162	}
2163
2164	if (config)
2165		spa_rewind_data_to_nvlist(spa, config);
2166
2167	spa->spa_extreme_rewind = B_FALSE;
2168	spa->spa_load_max_txg = UINT64_MAX;
2169
2170	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2171		spa_config_set(spa, config);
2172
2173	return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
2174}
2175
2176/*
2177 * Pool Open/Import
2178 *
2179 * The import case is identical to an open except that the configuration is sent
2180 * down from userland, instead of grabbed from the configuration cache.  For the
2181 * case of an open, the pool configuration will exist in the
2182 * POOL_STATE_UNINITIALIZED state.
2183 *
2184 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2185 * the same time open the pool, without having to keep around the spa_t in some
2186 * ambiguous state.
2187 */
2188static int
2189spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2190    nvlist_t **config)
2191{
2192	spa_t *spa;
2193	zpool_rewind_policy_t policy;
2194	spa_load_state_t state = SPA_LOAD_OPEN;
2195	int error;
2196	int locked = B_FALSE;
2197
2198	*spapp = NULL;
2199
2200	/*
2201	 * As disgusting as this is, we need to support recursive calls to this
2202	 * function because dsl_dir_open() is called during spa_load(), and ends
2203	 * up calling spa_open() again.  The real fix is to figure out how to
2204	 * avoid dsl_dir_open() calling this in the first place.
2205	 */
2206	if (mutex_owner(&spa_namespace_lock) != curthread) {
2207		mutex_enter(&spa_namespace_lock);
2208		locked = B_TRUE;
2209	}
2210
2211	if ((spa = spa_lookup(pool)) == NULL) {
2212		if (locked)
2213			mutex_exit(&spa_namespace_lock);
2214		return (ENOENT);
2215	}
2216
2217	zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, &policy);
2218	if (policy.zrp_request & ZPOOL_DO_REWIND)
2219		state = SPA_LOAD_RECOVER;
2220
2221	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2222
2223		spa_activate(spa, spa_mode_global);
2224
2225		if (spa->spa_last_open_failed && (policy.zrp_request &
2226		    (ZPOOL_NO_REWIND | ZPOOL_NEVER_REWIND))) {
2227			if (config != NULL && spa->spa_config)
2228				VERIFY(nvlist_dup(spa->spa_config,
2229				    config, KM_SLEEP) == 0);
2230			spa_deactivate(spa);
2231			if (locked)
2232				mutex_exit(&spa_namespace_lock);
2233			return (spa->spa_last_open_failed);
2234		}
2235
2236		if (state != SPA_LOAD_RECOVER)
2237			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2238
2239		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2240		    policy.zrp_request);
2241
2242		if (error == EBADF) {
2243			/*
2244			 * If vdev_validate() returns failure (indicated by
2245			 * EBADF), it indicates that one of the vdevs indicates
2246			 * that the pool has been exported or destroyed.  If
2247			 * this is the case, the config cache is out of sync and
2248			 * we should remove the pool from the namespace.
2249			 */
2250			spa_unload(spa);
2251			spa_deactivate(spa);
2252			spa_config_sync(spa, B_TRUE, B_TRUE);
2253			spa_remove(spa);
2254			if (locked)
2255				mutex_exit(&spa_namespace_lock);
2256			return (ENOENT);
2257		}
2258
2259		if (error) {
2260			/*
2261			 * We can't open the pool, but we still have useful
2262			 * information: the state of each vdev after the
2263			 * attempted vdev_open().  Return this to the user.
2264			 */
2265			if (config != NULL && spa->spa_config)
2266				VERIFY(nvlist_dup(spa->spa_config, config,
2267				    KM_SLEEP) == 0);
2268			spa_unload(spa);
2269			spa_deactivate(spa);
2270			spa->spa_last_open_failed = error;
2271			if (locked)
2272				mutex_exit(&spa_namespace_lock);
2273			*spapp = NULL;
2274			return (error);
2275		}
2276
2277	}
2278
2279	spa_open_ref(spa, tag);
2280
2281
2282	if (config != NULL)
2283		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2284
2285	if (locked) {
2286		spa->spa_last_open_failed = 0;
2287		spa->spa_last_ubsync_txg = 0;
2288		spa->spa_load_txg = 0;
2289		mutex_exit(&spa_namespace_lock);
2290	}
2291
2292	*spapp = spa;
2293
2294	return (0);
2295}
2296
2297int
2298spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2299    nvlist_t **config)
2300{
2301	return (spa_open_common(name, spapp, tag, policy, config));
2302}
2303
2304int
2305spa_open(const char *name, spa_t **spapp, void *tag)
2306{
2307	return (spa_open_common(name, spapp, tag, NULL, NULL));
2308}
2309
2310/*
2311 * Lookup the given spa_t, incrementing the inject count in the process,
2312 * preventing it from being exported or destroyed.
2313 */
2314spa_t *
2315spa_inject_addref(char *name)
2316{
2317	spa_t *spa;
2318
2319	mutex_enter(&spa_namespace_lock);
2320	if ((spa = spa_lookup(name)) == NULL) {
2321		mutex_exit(&spa_namespace_lock);
2322		return (NULL);
2323	}
2324	spa->spa_inject_ref++;
2325	mutex_exit(&spa_namespace_lock);
2326
2327	return (spa);
2328}
2329
2330void
2331spa_inject_delref(spa_t *spa)
2332{
2333	mutex_enter(&spa_namespace_lock);
2334	spa->spa_inject_ref--;
2335	mutex_exit(&spa_namespace_lock);
2336}
2337
2338/*
2339 * Add spares device information to the nvlist.
2340 */
2341static void
2342spa_add_spares(spa_t *spa, nvlist_t *config)
2343{
2344	nvlist_t **spares;
2345	uint_t i, nspares;
2346	nvlist_t *nvroot;
2347	uint64_t guid;
2348	vdev_stat_t *vs;
2349	uint_t vsc;
2350	uint64_t pool;
2351
2352	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2353
2354	if (spa->spa_spares.sav_count == 0)
2355		return;
2356
2357	VERIFY(nvlist_lookup_nvlist(config,
2358	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2359	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2360	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2361	if (nspares != 0) {
2362		VERIFY(nvlist_add_nvlist_array(nvroot,
2363		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2364		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2365		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2366
2367		/*
2368		 * Go through and find any spares which have since been
2369		 * repurposed as an active spare.  If this is the case, update
2370		 * their status appropriately.
2371		 */
2372		for (i = 0; i < nspares; i++) {
2373			VERIFY(nvlist_lookup_uint64(spares[i],
2374			    ZPOOL_CONFIG_GUID, &guid) == 0);
2375			if (spa_spare_exists(guid, &pool, NULL) &&
2376			    pool != 0ULL) {
2377				VERIFY(nvlist_lookup_uint64_array(
2378				    spares[i], ZPOOL_CONFIG_STATS,
2379				    (uint64_t **)&vs, &vsc) == 0);
2380				vs->vs_state = VDEV_STATE_CANT_OPEN;
2381				vs->vs_aux = VDEV_AUX_SPARED;
2382			}
2383		}
2384	}
2385}
2386
2387/*
2388 * Add l2cache device information to the nvlist, including vdev stats.
2389 */
2390static void
2391spa_add_l2cache(spa_t *spa, nvlist_t *config)
2392{
2393	nvlist_t **l2cache;
2394	uint_t i, j, nl2cache;
2395	nvlist_t *nvroot;
2396	uint64_t guid;
2397	vdev_t *vd;
2398	vdev_stat_t *vs;
2399	uint_t vsc;
2400
2401	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2402
2403	if (spa->spa_l2cache.sav_count == 0)
2404		return;
2405
2406	VERIFY(nvlist_lookup_nvlist(config,
2407	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2408	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2409	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2410	if (nl2cache != 0) {
2411		VERIFY(nvlist_add_nvlist_array(nvroot,
2412		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2413		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2414		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2415
2416		/*
2417		 * Update level 2 cache device stats.
2418		 */
2419
2420		for (i = 0; i < nl2cache; i++) {
2421			VERIFY(nvlist_lookup_uint64(l2cache[i],
2422			    ZPOOL_CONFIG_GUID, &guid) == 0);
2423
2424			vd = NULL;
2425			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2426				if (guid ==
2427				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2428					vd = spa->spa_l2cache.sav_vdevs[j];
2429					break;
2430				}
2431			}
2432			ASSERT(vd != NULL);
2433
2434			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2435			    ZPOOL_CONFIG_STATS, (uint64_t **)&vs, &vsc) == 0);
2436			vdev_get_stats(vd, vs);
2437		}
2438	}
2439}
2440
2441int
2442spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2443{
2444	int error;
2445	spa_t *spa;
2446
2447	*config = NULL;
2448	error = spa_open_common(name, &spa, FTAG, NULL, config);
2449
2450	if (spa != NULL) {
2451		/*
2452		 * This still leaves a window of inconsistency where the spares
2453		 * or l2cache devices could change and the config would be
2454		 * self-inconsistent.
2455		 */
2456		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2457
2458		if (*config != NULL) {
2459			VERIFY(nvlist_add_uint64(*config,
2460			    ZPOOL_CONFIG_ERRCOUNT,
2461			    spa_get_errlog_size(spa)) == 0);
2462
2463			if (spa_suspended(spa))
2464				VERIFY(nvlist_add_uint64(*config,
2465				    ZPOOL_CONFIG_SUSPENDED,
2466				    spa->spa_failmode) == 0);
2467
2468			spa_add_spares(spa, *config);
2469			spa_add_l2cache(spa, *config);
2470		}
2471	}
2472
2473	/*
2474	 * We want to get the alternate root even for faulted pools, so we cheat
2475	 * and call spa_lookup() directly.
2476	 */
2477	if (altroot) {
2478		if (spa == NULL) {
2479			mutex_enter(&spa_namespace_lock);
2480			spa = spa_lookup(name);
2481			if (spa)
2482				spa_altroot(spa, altroot, buflen);
2483			else
2484				altroot[0] = '\0';
2485			spa = NULL;
2486			mutex_exit(&spa_namespace_lock);
2487		} else {
2488			spa_altroot(spa, altroot, buflen);
2489		}
2490	}
2491
2492	if (spa != NULL) {
2493		spa_config_exit(spa, SCL_CONFIG, FTAG);
2494		spa_close(spa, FTAG);
2495	}
2496
2497	return (error);
2498}
2499
2500/*
2501 * Validate that the auxiliary device array is well formed.  We must have an
2502 * array of nvlists, each which describes a valid leaf vdev.  If this is an
2503 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2504 * specified, as long as they are well-formed.
2505 */
2506static int
2507spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2508    spa_aux_vdev_t *sav, const char *config, uint64_t version,
2509    vdev_labeltype_t label)
2510{
2511	nvlist_t **dev;
2512	uint_t i, ndev;
2513	vdev_t *vd;
2514	int error;
2515
2516	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2517
2518	/*
2519	 * It's acceptable to have no devs specified.
2520	 */
2521	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2522		return (0);
2523
2524	if (ndev == 0)
2525		return (EINVAL);
2526
2527	/*
2528	 * Make sure the pool is formatted with a version that supports this
2529	 * device type.
2530	 */
2531	if (spa_version(spa) < version)
2532		return (ENOTSUP);
2533
2534	/*
2535	 * Set the pending device list so we correctly handle device in-use
2536	 * checking.
2537	 */
2538	sav->sav_pending = dev;
2539	sav->sav_npending = ndev;
2540
2541	for (i = 0; i < ndev; i++) {
2542		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2543		    mode)) != 0)
2544			goto out;
2545
2546		if (!vd->vdev_ops->vdev_op_leaf) {
2547			vdev_free(vd);
2548			error = EINVAL;
2549			goto out;
2550		}
2551
2552		/*
2553		 * The L2ARC currently only supports disk devices in
2554		 * kernel context.  For user-level testing, we allow it.
2555		 */
2556#ifdef _KERNEL
2557		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2558		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2559			error = ENOTBLK;
2560			goto out;
2561		}
2562#endif
2563		vd->vdev_top = vd;
2564
2565		if ((error = vdev_open(vd)) == 0 &&
2566		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
2567			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2568			    vd->vdev_guid) == 0);
2569		}
2570
2571		vdev_free(vd);
2572
2573		if (error &&
2574		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2575			goto out;
2576		else
2577			error = 0;
2578	}
2579
2580out:
2581	sav->sav_pending = NULL;
2582	sav->sav_npending = 0;
2583	return (error);
2584}
2585
2586static int
2587spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2588{
2589	int error;
2590
2591	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2592
2593	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2594	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2595	    VDEV_LABEL_SPARE)) != 0) {
2596		return (error);
2597	}
2598
2599	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2600	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2601	    VDEV_LABEL_L2CACHE));
2602}
2603
2604static void
2605spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2606    const char *config)
2607{
2608	int i;
2609
2610	if (sav->sav_config != NULL) {
2611		nvlist_t **olddevs;
2612		uint_t oldndevs;
2613		nvlist_t **newdevs;
2614
2615		/*
2616		 * Generate new dev list by concatentating with the
2617		 * current dev list.
2618		 */
2619		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2620		    &olddevs, &oldndevs) == 0);
2621
2622		newdevs = kmem_alloc(sizeof (void *) *
2623		    (ndevs + oldndevs), KM_SLEEP);
2624		for (i = 0; i < oldndevs; i++)
2625			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2626			    KM_SLEEP) == 0);
2627		for (i = 0; i < ndevs; i++)
2628			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2629			    KM_SLEEP) == 0);
2630
2631		VERIFY(nvlist_remove(sav->sav_config, config,
2632		    DATA_TYPE_NVLIST_ARRAY) == 0);
2633
2634		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2635		    config, newdevs, ndevs + oldndevs) == 0);
2636		for (i = 0; i < oldndevs + ndevs; i++)
2637			nvlist_free(newdevs[i]);
2638		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2639	} else {
2640		/*
2641		 * Generate a new dev list.
2642		 */
2643		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2644		    KM_SLEEP) == 0);
2645		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2646		    devs, ndevs) == 0);
2647	}
2648}
2649
2650/*
2651 * Stop and drop level 2 ARC devices
2652 */
2653void
2654spa_l2cache_drop(spa_t *spa)
2655{
2656	vdev_t *vd;
2657	int i;
2658	spa_aux_vdev_t *sav = &spa->spa_l2cache;
2659
2660	for (i = 0; i < sav->sav_count; i++) {
2661		uint64_t pool;
2662
2663		vd = sav->sav_vdevs[i];
2664		ASSERT(vd != NULL);
2665
2666		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2667		    pool != 0ULL && l2arc_vdev_present(vd))
2668			l2arc_remove_vdev(vd);
2669		if (vd->vdev_isl2cache)
2670			spa_l2cache_remove(vd);
2671		vdev_clear_stats(vd);
2672		(void) vdev_close(vd);
2673	}
2674}
2675
2676/*
2677 * Pool Creation
2678 */
2679int
2680spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2681    const char *history_str, nvlist_t *zplprops)
2682{
2683	spa_t *spa;
2684	char *altroot = NULL;
2685	vdev_t *rvd;
2686	dsl_pool_t *dp;
2687	dmu_tx_t *tx;
2688	int error = 0;
2689	uint64_t txg = TXG_INITIAL;
2690	nvlist_t **spares, **l2cache;
2691	uint_t nspares, nl2cache;
2692	uint64_t version;
2693
2694	/*
2695	 * If this pool already exists, return failure.
2696	 */
2697	mutex_enter(&spa_namespace_lock);
2698	if (spa_lookup(pool) != NULL) {
2699		mutex_exit(&spa_namespace_lock);
2700		return (EEXIST);
2701	}
2702
2703	/*
2704	 * Allocate a new spa_t structure.
2705	 */
2706	(void) nvlist_lookup_string(props,
2707	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2708	spa = spa_add(pool, NULL, altroot);
2709	spa_activate(spa, spa_mode_global);
2710
2711	if (props && (error = spa_prop_validate(spa, props))) {
2712		spa_deactivate(spa);
2713		spa_remove(spa);
2714		mutex_exit(&spa_namespace_lock);
2715		return (error);
2716	}
2717
2718	if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2719	    &version) != 0)
2720		version = SPA_VERSION;
2721	ASSERT(version <= SPA_VERSION);
2722
2723	spa->spa_first_txg = txg;
2724	spa->spa_uberblock.ub_txg = txg - 1;
2725	spa->spa_uberblock.ub_version = version;
2726	spa->spa_ubsync = spa->spa_uberblock;
2727
2728	/*
2729	 * Create "The Godfather" zio to hold all async IOs
2730	 */
2731	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2732	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2733
2734	/*
2735	 * Create the root vdev.
2736	 */
2737	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2738
2739	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2740
2741	ASSERT(error != 0 || rvd != NULL);
2742	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2743
2744	if (error == 0 && !zfs_allocatable_devs(nvroot))
2745		error = EINVAL;
2746
2747	if (error == 0 &&
2748	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2749	    (error = spa_validate_aux(spa, nvroot, txg,
2750	    VDEV_ALLOC_ADD)) == 0) {
2751		for (int c = 0; c < rvd->vdev_children; c++) {
2752			vdev_metaslab_set_size(rvd->vdev_child[c]);
2753			vdev_expand(rvd->vdev_child[c], txg);
2754		}
2755	}
2756
2757	spa_config_exit(spa, SCL_ALL, FTAG);
2758
2759	if (error != 0) {
2760		spa_unload(spa);
2761		spa_deactivate(spa);
2762		spa_remove(spa);
2763		mutex_exit(&spa_namespace_lock);
2764		return (error);
2765	}
2766
2767	/*
2768	 * Get the list of spares, if specified.
2769	 */
2770	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2771	    &spares, &nspares) == 0) {
2772		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
2773		    KM_SLEEP) == 0);
2774		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2775		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2776		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2777		spa_load_spares(spa);
2778		spa_config_exit(spa, SCL_ALL, FTAG);
2779		spa->spa_spares.sav_sync = B_TRUE;
2780	}
2781
2782	/*
2783	 * Get the list of level 2 cache devices, if specified.
2784	 */
2785	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2786	    &l2cache, &nl2cache) == 0) {
2787		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2788		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2789		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2790		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2791		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2792		spa_load_l2cache(spa);
2793		spa_config_exit(spa, SCL_ALL, FTAG);
2794		spa->spa_l2cache.sav_sync = B_TRUE;
2795	}
2796
2797	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
2798	spa->spa_meta_objset = dp->dp_meta_objset;
2799
2800	/*
2801	 * Create DDTs (dedup tables).
2802	 */
2803	ddt_create(spa);
2804
2805	spa_update_dspace(spa);
2806
2807	tx = dmu_tx_create_assigned(dp, txg);
2808
2809	/*
2810	 * Create the pool config object.
2811	 */
2812	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
2813	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
2814	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2815
2816	if (zap_add(spa->spa_meta_objset,
2817	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
2818	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
2819		cmn_err(CE_PANIC, "failed to add pool config");
2820	}
2821
2822	/* Newly created pools with the right version are always deflated. */
2823	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
2824		spa->spa_deflate = TRUE;
2825		if (zap_add(spa->spa_meta_objset,
2826		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
2827		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
2828			cmn_err(CE_PANIC, "failed to add deflate");
2829		}
2830	}
2831
2832	/*
2833	 * Create the deferred-free bplist object.  Turn off compression
2834	 * because sync-to-convergence takes longer if the blocksize
2835	 * keeps changing.
2836	 */
2837	spa->spa_deferred_bplist_obj = bplist_create(spa->spa_meta_objset,
2838	    1 << 14, tx);
2839	dmu_object_set_compress(spa->spa_meta_objset,
2840	    spa->spa_deferred_bplist_obj, ZIO_COMPRESS_OFF, tx);
2841
2842	if (zap_add(spa->spa_meta_objset,
2843	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
2844	    sizeof (uint64_t), 1, &spa->spa_deferred_bplist_obj, tx) != 0) {
2845		cmn_err(CE_PANIC, "failed to add bplist");
2846	}
2847
2848	/*
2849	 * Create the pool's history object.
2850	 */
2851	if (version >= SPA_VERSION_ZPOOL_HISTORY)
2852		spa_history_create_obj(spa, tx);
2853
2854	/*
2855	 * Set pool properties.
2856	 */
2857	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
2858	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2859	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
2860	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
2861
2862	if (props != NULL) {
2863		spa_configfile_set(spa, props, B_FALSE);
2864		spa_sync_props(spa, props, CRED(), tx);
2865	}
2866
2867	dmu_tx_commit(tx);
2868
2869	spa->spa_sync_on = B_TRUE;
2870	txg_sync_start(spa->spa_dsl_pool);
2871
2872	/*
2873	 * We explicitly wait for the first transaction to complete so that our
2874	 * bean counters are appropriately updated.
2875	 */
2876	txg_wait_synced(spa->spa_dsl_pool, txg);
2877
2878	spa_config_sync(spa, B_FALSE, B_TRUE);
2879
2880	if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
2881		(void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
2882	spa_history_log_version(spa, LOG_POOL_CREATE);
2883
2884	spa->spa_minref = refcount_count(&spa->spa_refcount);
2885
2886	mutex_exit(&spa_namespace_lock);
2887
2888	return (0);
2889}
2890
2891#ifdef _KERNEL
2892/*
2893 * Get the root pool information from the root disk, then import the root pool
2894 * during the system boot up time.
2895 */
2896extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
2897
2898static nvlist_t *
2899spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
2900{
2901	nvlist_t *config;
2902	nvlist_t *nvtop, *nvroot;
2903	uint64_t pgid;
2904
2905	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
2906		return (NULL);
2907
2908	/*
2909	 * Add this top-level vdev to the child array.
2910	 */
2911	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2912	    &nvtop) == 0);
2913	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
2914	    &pgid) == 0);
2915	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
2916
2917	/*
2918	 * Put this pool's top-level vdevs into a root vdev.
2919	 */
2920	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2921	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
2922	    VDEV_TYPE_ROOT) == 0);
2923	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
2924	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
2925	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2926	    &nvtop, 1) == 0);
2927
2928	/*
2929	 * Replace the existing vdev_tree with the new root vdev in
2930	 * this pool's configuration (remove the old, add the new).
2931	 */
2932	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
2933	nvlist_free(nvroot);
2934	return (config);
2935}
2936
2937/*
2938 * Walk the vdev tree and see if we can find a device with "better"
2939 * configuration. A configuration is "better" if the label on that
2940 * device has a more recent txg.
2941 */
2942static void
2943spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
2944{
2945	for (int c = 0; c < vd->vdev_children; c++)
2946		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
2947
2948	if (vd->vdev_ops->vdev_op_leaf) {
2949		nvlist_t *label;
2950		uint64_t label_txg;
2951
2952		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
2953		    &label) != 0)
2954			return;
2955
2956		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
2957		    &label_txg) == 0);
2958
2959		/*
2960		 * Do we have a better boot device?
2961		 */
2962		if (label_txg > *txg) {
2963			*txg = label_txg;
2964			*avd = vd;
2965		}
2966		nvlist_free(label);
2967	}
2968}
2969
2970/*
2971 * Import a root pool.
2972 *
2973 * For x86. devpath_list will consist of devid and/or physpath name of
2974 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
2975 * The GRUB "findroot" command will return the vdev we should boot.
2976 *
2977 * For Sparc, devpath_list consists the physpath name of the booting device
2978 * no matter the rootpool is a single device pool or a mirrored pool.
2979 * e.g.
2980 *	"/pci@1f,0/ide@d/disk@0,0:a"
2981 */
2982int
2983spa_import_rootpool(char *devpath, char *devid)
2984{
2985	spa_t *spa;
2986	vdev_t *rvd, *bvd, *avd = NULL;
2987	nvlist_t *config, *nvtop;
2988	uint64_t guid, txg;
2989	char *pname;
2990	int error;
2991
2992	/*
2993	 * Read the label from the boot device and generate a configuration.
2994	 */
2995	config = spa_generate_rootconf(devpath, devid, &guid);
2996#if defined(_OBP) && defined(_KERNEL)
2997	if (config == NULL) {
2998		if (strstr(devpath, "/iscsi/ssd") != NULL) {
2999			/* iscsi boot */
3000			get_iscsi_bootpath_phy(devpath);
3001			config = spa_generate_rootconf(devpath, devid, &guid);
3002		}
3003	}
3004#endif
3005	if (config == NULL) {
3006		cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
3007		    devpath);
3008		return (EIO);
3009	}
3010
3011	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3012	    &pname) == 0);
3013	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3014
3015	mutex_enter(&spa_namespace_lock);
3016	if ((spa = spa_lookup(pname)) != NULL) {
3017		/*
3018		 * Remove the existing root pool from the namespace so that we
3019		 * can replace it with the correct config we just read in.
3020		 */
3021		spa_remove(spa);
3022	}
3023
3024	spa = spa_add(pname, config, NULL);
3025	spa->spa_is_root = B_TRUE;
3026	spa->spa_load_verbatim = B_TRUE;
3027
3028	/*
3029	 * Build up a vdev tree based on the boot device's label config.
3030	 */
3031	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3032	    &nvtop) == 0);
3033	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3034	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3035	    VDEV_ALLOC_ROOTPOOL);
3036	spa_config_exit(spa, SCL_ALL, FTAG);
3037	if (error) {
3038		mutex_exit(&spa_namespace_lock);
3039		nvlist_free(config);
3040		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3041		    pname);
3042		return (error);
3043	}
3044
3045	/*
3046	 * Get the boot vdev.
3047	 */
3048	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3049		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3050		    (u_longlong_t)guid);
3051		error = ENOENT;
3052		goto out;
3053	}
3054
3055	/*
3056	 * Determine if there is a better boot device.
3057	 */
3058	avd = bvd;
3059	spa_alt_rootvdev(rvd, &avd, &txg);
3060	if (avd != bvd) {
3061		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3062		    "try booting from '%s'", avd->vdev_path);
3063		error = EINVAL;
3064		goto out;
3065	}
3066
3067	/*
3068	 * If the boot device is part of a spare vdev then ensure that
3069	 * we're booting off the active spare.
3070	 */
3071	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3072	    !bvd->vdev_isspare) {
3073		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3074		    "try booting from '%s'",
3075		    bvd->vdev_parent->vdev_child[1]->vdev_path);
3076		error = EINVAL;
3077		goto out;
3078	}
3079
3080	error = 0;
3081	spa_history_log_version(spa, LOG_POOL_IMPORT);
3082out:
3083	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3084	vdev_free(rvd);
3085	spa_config_exit(spa, SCL_ALL, FTAG);
3086	mutex_exit(&spa_namespace_lock);
3087
3088	nvlist_free(config);
3089	return (error);
3090}
3091
3092#endif
3093
3094/*
3095 * Take a pool and insert it into the namespace as if it had been loaded at
3096 * boot.
3097 */
3098int
3099spa_import_verbatim(const char *pool, nvlist_t *config, nvlist_t *props)
3100{
3101	spa_t *spa;
3102	char *altroot = NULL;
3103
3104	mutex_enter(&spa_namespace_lock);
3105	if (spa_lookup(pool) != NULL) {
3106		mutex_exit(&spa_namespace_lock);
3107		return (EEXIST);
3108	}
3109
3110	(void) nvlist_lookup_string(props,
3111	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3112	spa = spa_add(pool, config, altroot);
3113
3114	spa->spa_load_verbatim = B_TRUE;
3115
3116	if (props != NULL)
3117		spa_configfile_set(spa, props, B_FALSE);
3118
3119	spa_config_sync(spa, B_FALSE, B_TRUE);
3120
3121	mutex_exit(&spa_namespace_lock);
3122	spa_history_log_version(spa, LOG_POOL_IMPORT);
3123
3124	return (0);
3125}
3126
3127/*
3128 * Import a non-root pool into the system.
3129 */
3130int
3131spa_import(const char *pool, nvlist_t *config, nvlist_t *props)
3132{
3133	spa_t *spa;
3134	char *altroot = NULL;
3135	spa_load_state_t state = SPA_LOAD_IMPORT;
3136	zpool_rewind_policy_t policy;
3137	int error;
3138	nvlist_t *nvroot;
3139	nvlist_t **spares, **l2cache;
3140	uint_t nspares, nl2cache;
3141
3142	/*
3143	 * If a pool with this name exists, return failure.
3144	 */
3145	mutex_enter(&spa_namespace_lock);
3146	if (spa_lookup(pool) != NULL) {
3147		mutex_exit(&spa_namespace_lock);
3148		return (EEXIST);
3149	}
3150
3151	zpool_get_rewind_policy(config, &policy);
3152	if (policy.zrp_request & ZPOOL_DO_REWIND)
3153		state = SPA_LOAD_RECOVER;
3154
3155	/*
3156	 * Create and initialize the spa structure.
3157	 */
3158	(void) nvlist_lookup_string(props,
3159	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3160	spa = spa_add(pool, config, altroot);
3161	spa_activate(spa, spa_mode_global);
3162
3163	/*
3164	 * Don't start async tasks until we know everything is healthy.
3165	 */
3166	spa_async_suspend(spa);
3167
3168	/*
3169	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3170	 * because the user-supplied config is actually the one to trust when
3171	 * doing an import.
3172	 */
3173	if (state != SPA_LOAD_RECOVER)
3174		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3175	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3176	    policy.zrp_request);
3177
3178	/*
3179	 * Propagate anything learned about failing or best txgs
3180	 * back to caller
3181	 */
3182	spa_rewind_data_to_nvlist(spa, config);
3183
3184	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3185	/*
3186	 * Toss any existing sparelist, as it doesn't have any validity
3187	 * anymore, and conflicts with spa_has_spare().
3188	 */
3189	if (spa->spa_spares.sav_config) {
3190		nvlist_free(spa->spa_spares.sav_config);
3191		spa->spa_spares.sav_config = NULL;
3192		spa_load_spares(spa);
3193	}
3194	if (spa->spa_l2cache.sav_config) {
3195		nvlist_free(spa->spa_l2cache.sav_config);
3196		spa->spa_l2cache.sav_config = NULL;
3197		spa_load_l2cache(spa);
3198	}
3199
3200	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3201	    &nvroot) == 0);
3202	if (error == 0)
3203		error = spa_validate_aux(spa, nvroot, -1ULL,
3204		    VDEV_ALLOC_SPARE);
3205	if (error == 0)
3206		error = spa_validate_aux(spa, nvroot, -1ULL,
3207		    VDEV_ALLOC_L2CACHE);
3208	spa_config_exit(spa, SCL_ALL, FTAG);
3209
3210	if (props != NULL)
3211		spa_configfile_set(spa, props, B_FALSE);
3212
3213	if (error != 0 || (props && spa_writeable(spa) &&
3214	    (error = spa_prop_set(spa, props)))) {
3215		spa_unload(spa);
3216		spa_deactivate(spa);
3217		spa_remove(spa);
3218		mutex_exit(&spa_namespace_lock);
3219		return (error);
3220	}
3221
3222	spa_async_resume(spa);
3223
3224	/*
3225	 * Override any spares and level 2 cache devices as specified by
3226	 * the user, as these may have correct device names/devids, etc.
3227	 */
3228	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3229	    &spares, &nspares) == 0) {
3230		if (spa->spa_spares.sav_config)
3231			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3232			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3233		else
3234			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3235			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3236		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3237		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3238		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3239		spa_load_spares(spa);
3240		spa_config_exit(spa, SCL_ALL, FTAG);
3241		spa->spa_spares.sav_sync = B_TRUE;
3242	}
3243	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3244	    &l2cache, &nl2cache) == 0) {
3245		if (spa->spa_l2cache.sav_config)
3246			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3247			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3248		else
3249			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3250			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3251		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3252		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3253		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3254		spa_load_l2cache(spa);
3255		spa_config_exit(spa, SCL_ALL, FTAG);
3256		spa->spa_l2cache.sav_sync = B_TRUE;
3257	}
3258
3259	/*
3260	 * Check for any removed devices.
3261	 */
3262	if (spa->spa_autoreplace) {
3263		spa_aux_check_removed(&spa->spa_spares);
3264		spa_aux_check_removed(&spa->spa_l2cache);
3265	}
3266
3267	if (spa_writeable(spa)) {
3268		/*
3269		 * Update the config cache to include the newly-imported pool.
3270		 */
3271		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3272	}
3273
3274	/*
3275	 * It's possible that the pool was expanded while it was exported.
3276	 * We kick off an async task to handle this for us.
3277	 */
3278	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3279
3280	mutex_exit(&spa_namespace_lock);
3281	spa_history_log_version(spa, LOG_POOL_IMPORT);
3282
3283	return (0);
3284}
3285
3286nvlist_t *
3287spa_tryimport(nvlist_t *tryconfig)
3288{
3289	nvlist_t *config = NULL;
3290	char *poolname;
3291	spa_t *spa;
3292	uint64_t state;
3293	int error;
3294
3295	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3296		return (NULL);
3297
3298	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3299		return (NULL);
3300
3301	/*
3302	 * Create and initialize the spa structure.
3303	 */
3304	mutex_enter(&spa_namespace_lock);
3305	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3306	spa_activate(spa, FREAD);
3307
3308	/*
3309	 * Pass off the heavy lifting to spa_load().
3310	 * Pass TRUE for mosconfig because the user-supplied config
3311	 * is actually the one to trust when doing an import.
3312	 */
3313	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3314
3315	/*
3316	 * If 'tryconfig' was at least parsable, return the current config.
3317	 */
3318	if (spa->spa_root_vdev != NULL) {
3319		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3320		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3321		    poolname) == 0);
3322		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3323		    state) == 0);
3324		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3325		    spa->spa_uberblock.ub_timestamp) == 0);
3326
3327		/*
3328		 * If the bootfs property exists on this pool then we
3329		 * copy it out so that external consumers can tell which
3330		 * pools are bootable.
3331		 */
3332		if ((!error || error == EEXIST) && spa->spa_bootfs) {
3333			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3334
3335			/*
3336			 * We have to play games with the name since the
3337			 * pool was opened as TRYIMPORT_NAME.
3338			 */
3339			if (dsl_dsobj_to_dsname(spa_name(spa),
3340			    spa->spa_bootfs, tmpname) == 0) {
3341				char *cp;
3342				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3343
3344				cp = strchr(tmpname, '/');
3345				if (cp == NULL) {
3346					(void) strlcpy(dsname, tmpname,
3347					    MAXPATHLEN);
3348				} else {
3349					(void) snprintf(dsname, MAXPATHLEN,
3350					    "%s/%s", poolname, ++cp);
3351				}
3352				VERIFY(nvlist_add_string(config,
3353				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3354				kmem_free(dsname, MAXPATHLEN);
3355			}
3356			kmem_free(tmpname, MAXPATHLEN);
3357		}
3358
3359		/*
3360		 * Add the list of hot spares and level 2 cache devices.
3361		 */
3362		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3363		spa_add_spares(spa, config);
3364		spa_add_l2cache(spa, config);
3365		spa_config_exit(spa, SCL_CONFIG, FTAG);
3366	}
3367
3368	spa_unload(spa);
3369	spa_deactivate(spa);
3370	spa_remove(spa);
3371	mutex_exit(&spa_namespace_lock);
3372
3373	return (config);
3374}
3375
3376/*
3377 * Pool export/destroy
3378 *
3379 * The act of destroying or exporting a pool is very simple.  We make sure there
3380 * is no more pending I/O and any references to the pool are gone.  Then, we
3381 * update the pool state and sync all the labels to disk, removing the
3382 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3383 * we don't sync the labels or remove the configuration cache.
3384 */
3385static int
3386spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3387    boolean_t force, boolean_t hardforce)
3388{
3389	spa_t *spa;
3390
3391	if (oldconfig)
3392		*oldconfig = NULL;
3393
3394	if (!(spa_mode_global & FWRITE))
3395		return (EROFS);
3396
3397	mutex_enter(&spa_namespace_lock);
3398	if ((spa = spa_lookup(pool)) == NULL) {
3399		mutex_exit(&spa_namespace_lock);
3400		return (ENOENT);
3401	}
3402
3403	/*
3404	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3405	 * reacquire the namespace lock, and see if we can export.
3406	 */
3407	spa_open_ref(spa, FTAG);
3408	mutex_exit(&spa_namespace_lock);
3409	spa_async_suspend(spa);
3410	mutex_enter(&spa_namespace_lock);
3411	spa_close(spa, FTAG);
3412
3413	/*
3414	 * The pool will be in core if it's openable,
3415	 * in which case we can modify its state.
3416	 */
3417	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3418		/*
3419		 * Objsets may be open only because they're dirty, so we
3420		 * have to force it to sync before checking spa_refcnt.
3421		 */
3422		txg_wait_synced(spa->spa_dsl_pool, 0);
3423
3424		/*
3425		 * A pool cannot be exported or destroyed if there are active
3426		 * references.  If we are resetting a pool, allow references by
3427		 * fault injection handlers.
3428		 */
3429		if (!spa_refcount_zero(spa) ||
3430		    (spa->spa_inject_ref != 0 &&
3431		    new_state != POOL_STATE_UNINITIALIZED)) {
3432			spa_async_resume(spa);
3433			mutex_exit(&spa_namespace_lock);
3434			return (EBUSY);
3435		}
3436
3437		/*
3438		 * A pool cannot be exported if it has an active shared spare.
3439		 * This is to prevent other pools stealing the active spare
3440		 * from an exported pool. At user's own will, such pool can
3441		 * be forcedly exported.
3442		 */
3443		if (!force && new_state == POOL_STATE_EXPORTED &&
3444		    spa_has_active_shared_spare(spa)) {
3445			spa_async_resume(spa);
3446			mutex_exit(&spa_namespace_lock);
3447			return (EXDEV);
3448		}
3449
3450		/*
3451		 * We want this to be reflected on every label,
3452		 * so mark them all dirty.  spa_unload() will do the
3453		 * final sync that pushes these changes out.
3454		 */
3455		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3456			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3457			spa->spa_state = new_state;
3458			spa->spa_final_txg = spa_last_synced_txg(spa) + 1;
3459			vdev_config_dirty(spa->spa_root_vdev);
3460			spa_config_exit(spa, SCL_ALL, FTAG);
3461		}
3462	}
3463
3464	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
3465
3466	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3467		spa_unload(spa);
3468		spa_deactivate(spa);
3469	}
3470
3471	if (oldconfig && spa->spa_config)
3472		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3473
3474	if (new_state != POOL_STATE_UNINITIALIZED) {
3475		if (!hardforce)
3476			spa_config_sync(spa, B_TRUE, B_TRUE);
3477		spa_remove(spa);
3478	}
3479	mutex_exit(&spa_namespace_lock);
3480
3481	return (0);
3482}
3483
3484/*
3485 * Destroy a storage pool.
3486 */
3487int
3488spa_destroy(char *pool)
3489{
3490	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3491	    B_FALSE, B_FALSE));
3492}
3493
3494/*
3495 * Export a storage pool.
3496 */
3497int
3498spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3499    boolean_t hardforce)
3500{
3501	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3502	    force, hardforce));
3503}
3504
3505/*
3506 * Similar to spa_export(), this unloads the spa_t without actually removing it
3507 * from the namespace in any way.
3508 */
3509int
3510spa_reset(char *pool)
3511{
3512	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3513	    B_FALSE, B_FALSE));
3514}
3515
3516/*
3517 * ==========================================================================
3518 * Device manipulation
3519 * ==========================================================================
3520 */
3521
3522/*
3523 * Add a device to a storage pool.
3524 */
3525int
3526spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3527{
3528	uint64_t txg, id;
3529	int error;
3530	vdev_t *rvd = spa->spa_root_vdev;
3531	vdev_t *vd, *tvd;
3532	nvlist_t **spares, **l2cache;
3533	uint_t nspares, nl2cache;
3534
3535	txg = spa_vdev_enter(spa);
3536
3537	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3538	    VDEV_ALLOC_ADD)) != 0)
3539		return (spa_vdev_exit(spa, NULL, txg, error));
3540
3541	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
3542
3543	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3544	    &nspares) != 0)
3545		nspares = 0;
3546
3547	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3548	    &nl2cache) != 0)
3549		nl2cache = 0;
3550
3551	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3552		return (spa_vdev_exit(spa, vd, txg, EINVAL));
3553
3554	if (vd->vdev_children != 0 &&
3555	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
3556		return (spa_vdev_exit(spa, vd, txg, error));
3557
3558	/*
3559	 * We must validate the spares and l2cache devices after checking the
3560	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
3561	 */
3562	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3563		return (spa_vdev_exit(spa, vd, txg, error));
3564
3565	/*
3566	 * Transfer each new top-level vdev from vd to rvd.
3567	 */
3568	for (int c = 0; c < vd->vdev_children; c++) {
3569
3570		/*
3571		 * Set the vdev id to the first hole, if one exists.
3572		 */
3573		for (id = 0; id < rvd->vdev_children; id++) {
3574			if (rvd->vdev_child[id]->vdev_ishole) {
3575				vdev_free(rvd->vdev_child[id]);
3576				break;
3577			}
3578		}
3579		tvd = vd->vdev_child[c];
3580		vdev_remove_child(vd, tvd);
3581		tvd->vdev_id = id;
3582		vdev_add_child(rvd, tvd);
3583		vdev_config_dirty(tvd);
3584	}
3585
3586	if (nspares != 0) {
3587		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3588		    ZPOOL_CONFIG_SPARES);
3589		spa_load_spares(spa);
3590		spa->spa_spares.sav_sync = B_TRUE;
3591	}
3592
3593	if (nl2cache != 0) {
3594		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3595		    ZPOOL_CONFIG_L2CACHE);
3596		spa_load_l2cache(spa);
3597		spa->spa_l2cache.sav_sync = B_TRUE;
3598	}
3599
3600	/*
3601	 * We have to be careful when adding new vdevs to an existing pool.
3602	 * If other threads start allocating from these vdevs before we
3603	 * sync the config cache, and we lose power, then upon reboot we may
3604	 * fail to open the pool because there are DVAs that the config cache
3605	 * can't translate.  Therefore, we first add the vdevs without
3606	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3607	 * and then let spa_config_update() initialize the new metaslabs.
3608	 *
3609	 * spa_load() checks for added-but-not-initialized vdevs, so that
3610	 * if we lose power at any point in this sequence, the remaining
3611	 * steps will be completed the next time we load the pool.
3612	 */
3613	(void) spa_vdev_exit(spa, vd, txg, 0);
3614
3615	mutex_enter(&spa_namespace_lock);
3616	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3617	mutex_exit(&spa_namespace_lock);
3618
3619	return (0);
3620}
3621
3622/*
3623 * Attach a device to a mirror.  The arguments are the path to any device
3624 * in the mirror, and the nvroot for the new device.  If the path specifies
3625 * a device that is not mirrored, we automatically insert the mirror vdev.
3626 *
3627 * If 'replacing' is specified, the new device is intended to replace the
3628 * existing device; in this case the two devices are made into their own
3629 * mirror using the 'replacing' vdev, which is functionally identical to
3630 * the mirror vdev (it actually reuses all the same ops) but has a few
3631 * extra rules: you can't attach to it after it's been created, and upon
3632 * completion of resilvering, the first disk (the one being replaced)
3633 * is automatically detached.
3634 */
3635int
3636spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3637{
3638	uint64_t txg, open_txg;
3639	vdev_t *rvd = spa->spa_root_vdev;
3640	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3641	vdev_ops_t *pvops;
3642	char *oldvdpath, *newvdpath;
3643	int newvd_isspare;
3644	int error;
3645
3646	txg = spa_vdev_enter(spa);
3647
3648	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3649
3650	if (oldvd == NULL)
3651		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3652
3653	if (!oldvd->vdev_ops->vdev_op_leaf)
3654		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3655
3656	pvd = oldvd->vdev_parent;
3657
3658	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3659	    VDEV_ALLOC_ADD)) != 0)
3660		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3661
3662	if (newrootvd->vdev_children != 1)
3663		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3664
3665	newvd = newrootvd->vdev_child[0];
3666
3667	if (!newvd->vdev_ops->vdev_op_leaf)
3668		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3669
3670	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3671		return (spa_vdev_exit(spa, newrootvd, txg, error));
3672
3673	/*
3674	 * Spares can't replace logs
3675	 */
3676	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3677		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3678
3679	if (!replacing) {
3680		/*
3681		 * For attach, the only allowable parent is a mirror or the root
3682		 * vdev.
3683		 */
3684		if (pvd->vdev_ops != &vdev_mirror_ops &&
3685		    pvd->vdev_ops != &vdev_root_ops)
3686			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3687
3688		pvops = &vdev_mirror_ops;
3689	} else {
3690		/*
3691		 * Active hot spares can only be replaced by inactive hot
3692		 * spares.
3693		 */
3694		if (pvd->vdev_ops == &vdev_spare_ops &&
3695		    pvd->vdev_child[1] == oldvd &&
3696		    !spa_has_spare(spa, newvd->vdev_guid))
3697			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3698
3699		/*
3700		 * If the source is a hot spare, and the parent isn't already a
3701		 * spare, then we want to create a new hot spare.  Otherwise, we
3702		 * want to create a replacing vdev.  The user is not allowed to
3703		 * attach to a spared vdev child unless the 'isspare' state is
3704		 * the same (spare replaces spare, non-spare replaces
3705		 * non-spare).
3706		 */
3707		if (pvd->vdev_ops == &vdev_replacing_ops)
3708			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3709		else if (pvd->vdev_ops == &vdev_spare_ops &&
3710		    newvd->vdev_isspare != oldvd->vdev_isspare)
3711			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3712		else if (pvd->vdev_ops != &vdev_spare_ops &&
3713		    newvd->vdev_isspare)
3714			pvops = &vdev_spare_ops;
3715		else
3716			pvops = &vdev_replacing_ops;
3717	}
3718
3719	/*
3720	 * Make sure the new device is big enough.
3721	 */
3722	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3723		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3724
3725	/*
3726	 * The new device cannot have a higher alignment requirement
3727	 * than the top-level vdev.
3728	 */
3729	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3730		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3731
3732	/*
3733	 * If this is an in-place replacement, update oldvd's path and devid
3734	 * to make it distinguishable from newvd, and unopenable from now on.
3735	 */
3736	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3737		spa_strfree(oldvd->vdev_path);
3738		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3739		    KM_SLEEP);
3740		(void) sprintf(oldvd->vdev_path, "%s/%s",
3741		    newvd->vdev_path, "old");
3742		if (oldvd->vdev_devid != NULL) {
3743			spa_strfree(oldvd->vdev_devid);
3744			oldvd->vdev_devid = NULL;
3745		}
3746	}
3747
3748	/*
3749	 * If the parent is not a mirror, or if we're replacing, insert the new
3750	 * mirror/replacing/spare vdev above oldvd.
3751	 */
3752	if (pvd->vdev_ops != pvops)
3753		pvd = vdev_add_parent(oldvd, pvops);
3754
3755	ASSERT(pvd->vdev_top->vdev_parent == rvd);
3756	ASSERT(pvd->vdev_ops == pvops);
3757	ASSERT(oldvd->vdev_parent == pvd);
3758
3759	/*
3760	 * Extract the new device from its root and add it to pvd.
3761	 */
3762	vdev_remove_child(newrootvd, newvd);
3763	newvd->vdev_id = pvd->vdev_children;
3764	newvd->vdev_crtxg = oldvd->vdev_crtxg;
3765	vdev_add_child(pvd, newvd);
3766
3767	tvd = newvd->vdev_top;
3768	ASSERT(pvd->vdev_top == tvd);
3769	ASSERT(tvd->vdev_parent == rvd);
3770
3771	vdev_config_dirty(tvd);
3772
3773	/*
3774	 * Set newvd's DTL to [TXG_INITIAL, open_txg].  It will propagate
3775	 * upward when spa_vdev_exit() calls vdev_dtl_reassess().
3776	 */
3777	open_txg = txg + TXG_CONCURRENT_STATES - 1;
3778
3779	vdev_dtl_dirty(newvd, DTL_MISSING,
3780	    TXG_INITIAL, open_txg - TXG_INITIAL + 1);
3781
3782	if (newvd->vdev_isspare) {
3783		spa_spare_activate(newvd);
3784		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
3785	}
3786
3787	oldvdpath = spa_strdup(oldvd->vdev_path);
3788	newvdpath = spa_strdup(newvd->vdev_path);
3789	newvd_isspare = newvd->vdev_isspare;
3790
3791	/*
3792	 * Mark newvd's DTL dirty in this txg.
3793	 */
3794	vdev_dirty(tvd, VDD_DTL, newvd, txg);
3795
3796	(void) spa_vdev_exit(spa, newrootvd, open_txg, 0);
3797
3798	spa_history_internal_log(LOG_POOL_VDEV_ATTACH, spa, NULL,
3799	    CRED(),  "%s vdev=%s %s vdev=%s",
3800	    replacing && newvd_isspare ? "spare in" :
3801	    replacing ? "replace" : "attach", newvdpath,
3802	    replacing ? "for" : "to", oldvdpath);
3803
3804	spa_strfree(oldvdpath);
3805	spa_strfree(newvdpath);
3806
3807	/*
3808	 * Kick off a resilver to update newvd.
3809	 */
3810	VERIFY3U(spa_scrub(spa, POOL_SCRUB_RESILVER), ==, 0);
3811
3812	return (0);
3813}
3814
3815/*
3816 * Detach a device from a mirror or replacing vdev.
3817 * If 'replace_done' is specified, only detach if the parent
3818 * is a replacing vdev.
3819 */
3820int
3821spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
3822{
3823	uint64_t txg;
3824	int error;
3825	vdev_t *rvd = spa->spa_root_vdev;
3826	vdev_t *vd, *pvd, *cvd, *tvd;
3827	boolean_t unspare = B_FALSE;
3828	uint64_t unspare_guid;
3829	size_t len;
3830	char *vdpath;
3831
3832	txg = spa_vdev_enter(spa);
3833
3834	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3835
3836	if (vd == NULL)
3837		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3838
3839	if (!vd->vdev_ops->vdev_op_leaf)
3840		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3841
3842	pvd = vd->vdev_parent;
3843
3844	/*
3845	 * If the parent/child relationship is not as expected, don't do it.
3846	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
3847	 * vdev that's replacing B with C.  The user's intent in replacing
3848	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
3849	 * the replace by detaching C, the expected behavior is to end up
3850	 * M(A,B).  But suppose that right after deciding to detach C,
3851	 * the replacement of B completes.  We would have M(A,C), and then
3852	 * ask to detach C, which would leave us with just A -- not what
3853	 * the user wanted.  To prevent this, we make sure that the
3854	 * parent/child relationship hasn't changed -- in this example,
3855	 * that C's parent is still the replacing vdev R.
3856	 */
3857	if (pvd->vdev_guid != pguid && pguid != 0)
3858		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3859
3860	/*
3861	 * If replace_done is specified, only remove this device if it's
3862	 * the first child of a replacing vdev.  For the 'spare' vdev, either
3863	 * disk can be removed.
3864	 */
3865	if (replace_done) {
3866		if (pvd->vdev_ops == &vdev_replacing_ops) {
3867			if (vd->vdev_id != 0)
3868				return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3869		} else if (pvd->vdev_ops != &vdev_spare_ops) {
3870			return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3871		}
3872	}
3873
3874	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
3875	    spa_version(spa) >= SPA_VERSION_SPARES);
3876
3877	/*
3878	 * Only mirror, replacing, and spare vdevs support detach.
3879	 */
3880	if (pvd->vdev_ops != &vdev_replacing_ops &&
3881	    pvd->vdev_ops != &vdev_mirror_ops &&
3882	    pvd->vdev_ops != &vdev_spare_ops)
3883		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3884
3885	/*
3886	 * If this device has the only valid copy of some data,
3887	 * we cannot safely detach it.
3888	 */
3889	if (vdev_dtl_required(vd))
3890		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3891
3892	ASSERT(pvd->vdev_children >= 2);
3893
3894	/*
3895	 * If we are detaching the second disk from a replacing vdev, then
3896	 * check to see if we changed the original vdev's path to have "/old"
3897	 * at the end in spa_vdev_attach().  If so, undo that change now.
3898	 */
3899	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id == 1 &&
3900	    pvd->vdev_child[0]->vdev_path != NULL &&
3901	    pvd->vdev_child[1]->vdev_path != NULL) {
3902		ASSERT(pvd->vdev_child[1] == vd);
3903		cvd = pvd->vdev_child[0];
3904		len = strlen(vd->vdev_path);
3905		if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
3906		    strcmp(cvd->vdev_path + len, "/old") == 0) {
3907			spa_strfree(cvd->vdev_path);
3908			cvd->vdev_path = spa_strdup(vd->vdev_path);
3909		}
3910	}
3911
3912	/*
3913	 * If we are detaching the original disk from a spare, then it implies
3914	 * that the spare should become a real disk, and be removed from the
3915	 * active spare list for the pool.
3916	 */
3917	if (pvd->vdev_ops == &vdev_spare_ops &&
3918	    vd->vdev_id == 0 && pvd->vdev_child[1]->vdev_isspare)
3919		unspare = B_TRUE;
3920
3921	/*
3922	 * Erase the disk labels so the disk can be used for other things.
3923	 * This must be done after all other error cases are handled,
3924	 * but before we disembowel vd (so we can still do I/O to it).
3925	 * But if we can't do it, don't treat the error as fatal --
3926	 * it may be that the unwritability of the disk is the reason
3927	 * it's being detached!
3928	 */
3929	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
3930
3931	/*
3932	 * Remove vd from its parent and compact the parent's children.
3933	 */
3934	vdev_remove_child(pvd, vd);
3935	vdev_compact_children(pvd);
3936
3937	/*
3938	 * Remember one of the remaining children so we can get tvd below.
3939	 */
3940	cvd = pvd->vdev_child[0];
3941
3942	/*
3943	 * If we need to remove the remaining child from the list of hot spares,
3944	 * do it now, marking the vdev as no longer a spare in the process.
3945	 * We must do this before vdev_remove_parent(), because that can
3946	 * change the GUID if it creates a new toplevel GUID.  For a similar
3947	 * reason, we must remove the spare now, in the same txg as the detach;
3948	 * otherwise someone could attach a new sibling, change the GUID, and
3949	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
3950	 */
3951	if (unspare) {
3952		ASSERT(cvd->vdev_isspare);
3953		spa_spare_remove(cvd);
3954		unspare_guid = cvd->vdev_guid;
3955		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
3956	}
3957
3958	/*
3959	 * If the parent mirror/replacing vdev only has one child,
3960	 * the parent is no longer needed.  Remove it from the tree.
3961	 */
3962	if (pvd->vdev_children == 1)
3963		vdev_remove_parent(cvd);
3964
3965	/*
3966	 * We don't set tvd until now because the parent we just removed
3967	 * may have been the previous top-level vdev.
3968	 */
3969	tvd = cvd->vdev_top;
3970	ASSERT(tvd->vdev_parent == rvd);
3971
3972	/*
3973	 * Reevaluate the parent vdev state.
3974	 */
3975	vdev_propagate_state(cvd);
3976
3977	/*
3978	 * If the 'autoexpand' property is set on the pool then automatically
3979	 * try to expand the size of the pool. For example if the device we
3980	 * just detached was smaller than the others, it may be possible to
3981	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
3982	 * first so that we can obtain the updated sizes of the leaf vdevs.
3983	 */
3984	if (spa->spa_autoexpand) {
3985		vdev_reopen(tvd);
3986		vdev_expand(tvd, txg);
3987	}
3988
3989	vdev_config_dirty(tvd);
3990
3991	/*
3992	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
3993	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
3994	 * But first make sure we're not on any *other* txg's DTL list, to
3995	 * prevent vd from being accessed after it's freed.
3996	 */
3997	vdpath = spa_strdup(vd->vdev_path);
3998	for (int t = 0; t < TXG_SIZE; t++)
3999		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4000	vd->vdev_detached = B_TRUE;
4001	vdev_dirty(tvd, VDD_DTL, vd, txg);
4002
4003	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4004
4005	error = spa_vdev_exit(spa, vd, txg, 0);
4006
4007	spa_history_internal_log(LOG_POOL_VDEV_DETACH, spa, NULL, CRED(),
4008	    "vdev=%s", vdpath);
4009	spa_strfree(vdpath);
4010
4011	/*
4012	 * If this was the removal of the original device in a hot spare vdev,
4013	 * then we want to go through and remove the device from the hot spare
4014	 * list of every other pool.
4015	 */
4016	if (unspare) {
4017		spa_t *myspa = spa;
4018		spa = NULL;
4019		mutex_enter(&spa_namespace_lock);
4020		while ((spa = spa_next(spa)) != NULL) {
4021			if (spa->spa_state != POOL_STATE_ACTIVE)
4022				continue;
4023			if (spa == myspa)
4024				continue;
4025			spa_open_ref(spa, FTAG);
4026			mutex_exit(&spa_namespace_lock);
4027			(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4028			mutex_enter(&spa_namespace_lock);
4029			spa_close(spa, FTAG);
4030		}
4031		mutex_exit(&spa_namespace_lock);
4032	}
4033
4034	return (error);
4035}
4036
4037/*
4038 * Split a set of devices from their mirrors, and create a new pool from them.
4039 */
4040int
4041spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4042    nvlist_t *props, boolean_t exp)
4043{
4044	int error = 0;
4045	uint64_t txg, *glist;
4046	spa_t *newspa;
4047	uint_t c, children, lastlog;
4048	nvlist_t **child, *nvl, *tmp;
4049	dmu_tx_t *tx;
4050	char *altroot = NULL;
4051	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4052	boolean_t activate_slog;
4053
4054	if (!spa_writeable(spa))
4055		return (EROFS);
4056
4057	txg = spa_vdev_enter(spa);
4058
4059	/* clear the log and flush everything up to now */
4060	activate_slog = spa_passivate_log(spa);
4061	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4062	error = spa_offline_log(spa);
4063	txg = spa_vdev_config_enter(spa);
4064
4065	if (activate_slog)
4066		spa_activate_log(spa);
4067
4068	if (error != 0)
4069		return (spa_vdev_exit(spa, NULL, txg, error));
4070
4071	/* check new spa name before going any further */
4072	if (spa_lookup(newname) != NULL)
4073		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4074
4075	/*
4076	 * scan through all the children to ensure they're all mirrors
4077	 */
4078	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4079	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4080	    &children) != 0)
4081		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4082
4083	/* first, check to ensure we've got the right child count */
4084	rvd = spa->spa_root_vdev;
4085	lastlog = 0;
4086	for (c = 0; c < rvd->vdev_children; c++) {
4087		vdev_t *vd = rvd->vdev_child[c];
4088
4089		/* don't count the holes & logs as children */
4090		if (vd->vdev_islog || vd->vdev_ishole) {
4091			if (lastlog == 0)
4092				lastlog = c;
4093			continue;
4094		}
4095
4096		lastlog = 0;
4097	}
4098	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4099		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4100
4101	/* next, ensure no spare or cache devices are part of the split */
4102	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4103	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4104		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4105
4106	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4107	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4108
4109	/* then, loop over each vdev and validate it */
4110	for (c = 0; c < children; c++) {
4111		uint64_t is_hole = 0;
4112
4113		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4114		    &is_hole);
4115
4116		if (is_hole != 0) {
4117			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4118			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4119				continue;
4120			} else {
4121				error = EINVAL;
4122				break;
4123			}
4124		}
4125
4126		/* which disk is going to be split? */
4127		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4128		    &glist[c]) != 0) {
4129			error = EINVAL;
4130			break;
4131		}
4132
4133		/* look it up in the spa */
4134		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4135		if (vml[c] == NULL) {
4136			error = ENODEV;
4137			break;
4138		}
4139
4140		/* make sure there's nothing stopping the split */
4141		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4142		    vml[c]->vdev_islog ||
4143		    vml[c]->vdev_ishole ||
4144		    vml[c]->vdev_isspare ||
4145		    vml[c]->vdev_isl2cache ||
4146		    !vdev_writeable(vml[c]) ||
4147		    vml[c]->vdev_children != 0 ||
4148		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4149		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4150			error = EINVAL;
4151			break;
4152		}
4153
4154		if (vdev_dtl_required(vml[c])) {
4155			error = EBUSY;
4156			break;
4157		}
4158
4159		/* we need certain info from the top level */
4160		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4161		    vml[c]->vdev_top->vdev_ms_array) == 0);
4162		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4163		    vml[c]->vdev_top->vdev_ms_shift) == 0);
4164		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4165		    vml[c]->vdev_top->vdev_asize) == 0);
4166		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4167		    vml[c]->vdev_top->vdev_ashift) == 0);
4168	}
4169
4170	if (error != 0) {
4171		kmem_free(vml, children * sizeof (vdev_t *));
4172		kmem_free(glist, children * sizeof (uint64_t));
4173		return (spa_vdev_exit(spa, NULL, txg, error));
4174	}
4175
4176	/* stop writers from using the disks */
4177	for (c = 0; c < children; c++) {
4178		if (vml[c] != NULL)
4179			vml[c]->vdev_offline = B_TRUE;
4180	}
4181	vdev_reopen(spa->spa_root_vdev);
4182
4183	/*
4184	 * Temporarily record the splitting vdevs in the spa config.  This
4185	 * will disappear once the config is regenerated.
4186	 */
4187	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4188	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4189	    glist, children) == 0);
4190	kmem_free(glist, children * sizeof (uint64_t));
4191
4192	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4193	    nvl) == 0);
4194	spa->spa_config_splitting = nvl;
4195	vdev_config_dirty(spa->spa_root_vdev);
4196
4197	/* configure and create the new pool */
4198	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4199	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4200	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4201	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4202	    spa_version(spa)) == 0);
4203	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4204	    spa->spa_config_txg) == 0);
4205	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4206	    spa_generate_guid(NULL)) == 0);
4207	(void) nvlist_lookup_string(props,
4208	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4209
4210	/* add the new pool to the namespace */
4211	newspa = spa_add(newname, config, altroot);
4212	newspa->spa_config_txg = spa->spa_config_txg;
4213	spa_set_log_state(newspa, SPA_LOG_CLEAR);
4214
4215	/* release the spa config lock, retaining the namespace lock */
4216	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4217
4218	if (zio_injection_enabled)
4219		zio_handle_panic_injection(spa, FTAG, 1);
4220
4221	spa_activate(newspa, spa_mode_global);
4222	spa_async_suspend(newspa);
4223
4224	/* create the new pool from the disks of the original pool */
4225	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4226	if (error)
4227		goto out;
4228
4229	/* if that worked, generate a real config for the new pool */
4230	if (newspa->spa_root_vdev != NULL) {
4231		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4232		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4233		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4234		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4235		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4236		    B_TRUE));
4237	}
4238
4239	/* set the props */
4240	if (props != NULL) {
4241		spa_configfile_set(newspa, props, B_FALSE);
4242		error = spa_prop_set(newspa, props);
4243		if (error)
4244			goto out;
4245	}
4246
4247	/* flush everything */
4248	txg = spa_vdev_config_enter(newspa);
4249	vdev_config_dirty(newspa->spa_root_vdev);
4250	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4251
4252	if (zio_injection_enabled)
4253		zio_handle_panic_injection(spa, FTAG, 2);
4254
4255	spa_async_resume(newspa);
4256
4257	/* finally, update the original pool's config */
4258	txg = spa_vdev_config_enter(spa);
4259	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4260	error = dmu_tx_assign(tx, TXG_WAIT);
4261	if (error != 0)
4262		dmu_tx_abort(tx);
4263	for (c = 0; c < children; c++) {
4264		if (vml[c] != NULL) {
4265			vdev_split(vml[c]);
4266			if (error == 0)
4267				spa_history_internal_log(LOG_POOL_VDEV_DETACH,
4268				    spa, tx, CRED(), "vdev=%s",
4269				    vml[c]->vdev_path);
4270			vdev_free(vml[c]);
4271		}
4272	}
4273	vdev_config_dirty(spa->spa_root_vdev);
4274	spa->spa_config_splitting = NULL;
4275	nvlist_free(nvl);
4276	if (error == 0)
4277		dmu_tx_commit(tx);
4278	(void) spa_vdev_exit(spa, NULL, txg, 0);
4279
4280	if (zio_injection_enabled)
4281		zio_handle_panic_injection(spa, FTAG, 3);
4282
4283	/* split is complete; log a history record */
4284	spa_history_internal_log(LOG_POOL_SPLIT, newspa, NULL, CRED(),
4285	    "split new pool %s from pool %s", newname, spa_name(spa));
4286
4287	kmem_free(vml, children * sizeof (vdev_t *));
4288
4289	/* if we're not going to mount the filesystems in userland, export */
4290	if (exp)
4291		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4292		    B_FALSE, B_FALSE);
4293
4294	return (error);
4295
4296out:
4297	spa_unload(newspa);
4298	spa_deactivate(newspa);
4299	spa_remove(newspa);
4300
4301	txg = spa_vdev_config_enter(spa);
4302	nvlist_free(spa->spa_config_splitting);
4303	spa->spa_config_splitting = NULL;
4304	(void) spa_vdev_exit(spa, NULL, txg, error);
4305
4306	kmem_free(vml, children * sizeof (vdev_t *));
4307	return (error);
4308}
4309
4310static nvlist_t *
4311spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4312{
4313	for (int i = 0; i < count; i++) {
4314		uint64_t guid;
4315
4316		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4317		    &guid) == 0);
4318
4319		if (guid == target_guid)
4320			return (nvpp[i]);
4321	}
4322
4323	return (NULL);
4324}
4325
4326static void
4327spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4328	nvlist_t *dev_to_remove)
4329{
4330	nvlist_t **newdev = NULL;
4331
4332	if (count > 1)
4333		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4334
4335	for (int i = 0, j = 0; i < count; i++) {
4336		if (dev[i] == dev_to_remove)
4337			continue;
4338		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4339	}
4340
4341	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4342	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4343
4344	for (int i = 0; i < count - 1; i++)
4345		nvlist_free(newdev[i]);
4346
4347	if (count > 1)
4348		kmem_free(newdev, (count - 1) * sizeof (void *));
4349}
4350
4351/*
4352 * Removing a device from the vdev namespace requires several steps
4353 * and can take a significant amount of time.  As a result we use
4354 * the spa_vdev_config_[enter/exit] functions which allow us to
4355 * grab and release the spa_config_lock while still holding the namespace
4356 * lock.  During each step the configuration is synced out.
4357 */
4358
4359/*
4360 * Evacuate the device.
4361 */
4362int
4363spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4364{
4365	int error = 0;
4366	uint64_t txg;
4367
4368	ASSERT(MUTEX_HELD(&spa_namespace_lock));
4369	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4370	ASSERT(vd == vd->vdev_top);
4371
4372	/*
4373	 * Evacuate the device.  We don't hold the config lock as writer
4374	 * since we need to do I/O but we do keep the
4375	 * spa_namespace_lock held.  Once this completes the device
4376	 * should no longer have any blocks allocated on it.
4377	 */
4378	if (vd->vdev_islog) {
4379		error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
4380		    NULL, DS_FIND_CHILDREN);
4381	} else {
4382		error = ENOTSUP;	/* until we have bp rewrite */
4383	}
4384
4385	txg_wait_synced(spa_get_dsl(spa), 0);
4386
4387	if (error)
4388		return (error);
4389
4390	/*
4391	 * The evacuation succeeded.  Remove any remaining MOS metadata
4392	 * associated with this vdev, and wait for these changes to sync.
4393	 */
4394	txg = spa_vdev_config_enter(spa);
4395	vd->vdev_removing = B_TRUE;
4396	vdev_dirty(vd, 0, NULL, txg);
4397	vdev_config_dirty(vd);
4398	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4399
4400	return (0);
4401}
4402
4403/*
4404 * Complete the removal by cleaning up the namespace.
4405 */
4406void
4407spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
4408{
4409	vdev_t *rvd = spa->spa_root_vdev;
4410	uint64_t id = vd->vdev_id;
4411	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
4412
4413	ASSERT(MUTEX_HELD(&spa_namespace_lock));
4414	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4415	ASSERT(vd == vd->vdev_top);
4416
4417	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4418
4419	if (list_link_active(&vd->vdev_state_dirty_node))
4420		vdev_state_clean(vd);
4421	if (list_link_active(&vd->vdev_config_dirty_node))
4422		vdev_config_clean(vd);
4423
4424	vdev_free(vd);
4425
4426	if (last_vdev) {
4427		vdev_compact_children(rvd);
4428	} else {
4429		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
4430		vdev_add_child(rvd, vd);
4431	}
4432	vdev_config_dirty(rvd);
4433
4434	/*
4435	 * Reassess the health of our root vdev.
4436	 */
4437	vdev_reopen(rvd);
4438}
4439
4440/*
4441 * Remove a device from the pool.  Currently, this supports removing only hot
4442 * spares, slogs, and level 2 ARC devices.
4443 */
4444int
4445spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
4446{
4447	vdev_t *vd;
4448	metaslab_group_t *mg;
4449	nvlist_t **spares, **l2cache, *nv;
4450	uint64_t txg = 0;
4451	uint_t nspares, nl2cache;
4452	int error = 0;
4453	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
4454
4455	if (!locked)
4456		txg = spa_vdev_enter(spa);
4457
4458	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4459
4460	if (spa->spa_spares.sav_vdevs != NULL &&
4461	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4462	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
4463	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
4464		/*
4465		 * Only remove the hot spare if it's not currently in use
4466		 * in this pool.
4467		 */
4468		if (vd == NULL || unspare) {
4469			spa_vdev_remove_aux(spa->spa_spares.sav_config,
4470			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
4471			spa_load_spares(spa);
4472			spa->spa_spares.sav_sync = B_TRUE;
4473		} else {
4474			error = EBUSY;
4475		}
4476	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
4477	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4478	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
4479	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
4480		/*
4481		 * Cache devices can always be removed.
4482		 */
4483		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
4484		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
4485		spa_load_l2cache(spa);
4486		spa->spa_l2cache.sav_sync = B_TRUE;
4487	} else if (vd != NULL && vd->vdev_islog) {
4488		ASSERT(!locked);
4489		ASSERT(vd == vd->vdev_top);
4490
4491		/*
4492		 * XXX - Once we have bp-rewrite this should
4493		 * become the common case.
4494		 */
4495
4496		mg = vd->vdev_mg;
4497
4498		/*
4499		 * Stop allocating from this vdev.
4500		 */
4501		metaslab_group_passivate(mg);
4502
4503		/*
4504		 * Wait for the youngest allocations and frees to sync,
4505		 * and then wait for the deferral of those frees to finish.
4506		 */
4507		spa_vdev_config_exit(spa, NULL,
4508		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
4509
4510		/*
4511		 * Attempt to evacuate the vdev.
4512		 */
4513		error = spa_vdev_remove_evacuate(spa, vd);
4514
4515		txg = spa_vdev_config_enter(spa);
4516
4517		/*
4518		 * If we couldn't evacuate the vdev, unwind.
4519		 */
4520		if (error) {
4521			metaslab_group_activate(mg);
4522			return (spa_vdev_exit(spa, NULL, txg, error));
4523		}
4524
4525		/*
4526		 * Clean up the vdev namespace.
4527		 */
4528		spa_vdev_remove_from_namespace(spa, vd);
4529
4530	} else if (vd != NULL) {
4531		/*
4532		 * Normal vdevs cannot be removed (yet).
4533		 */
4534		error = ENOTSUP;
4535	} else {
4536		/*
4537		 * There is no vdev of any kind with the specified guid.
4538		 */
4539		error = ENOENT;
4540	}
4541
4542	if (!locked)
4543		return (spa_vdev_exit(spa, NULL, txg, error));
4544
4545	return (error);
4546}
4547
4548/*
4549 * Find any device that's done replacing, or a vdev marked 'unspare' that's
4550 * current spared, so we can detach it.
4551 */
4552static vdev_t *
4553spa_vdev_resilver_done_hunt(vdev_t *vd)
4554{
4555	vdev_t *newvd, *oldvd;
4556
4557	for (int c = 0; c < vd->vdev_children; c++) {
4558		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
4559		if (oldvd != NULL)
4560			return (oldvd);
4561	}
4562
4563	/*
4564	 * Check for a completed replacement.
4565	 */
4566	if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
4567		oldvd = vd->vdev_child[0];
4568		newvd = vd->vdev_child[1];
4569
4570		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4571		    !vdev_dtl_required(oldvd))
4572			return (oldvd);
4573	}
4574
4575	/*
4576	 * Check for a completed resilver with the 'unspare' flag set.
4577	 */
4578	if (vd->vdev_ops == &vdev_spare_ops && vd->vdev_children == 2) {
4579		newvd = vd->vdev_child[0];
4580		oldvd = vd->vdev_child[1];
4581
4582		if (newvd->vdev_unspare &&
4583		    vdev_dtl_empty(newvd, DTL_MISSING) &&
4584		    !vdev_dtl_required(oldvd)) {
4585			newvd->vdev_unspare = 0;
4586			return (oldvd);
4587		}
4588	}
4589
4590	return (NULL);
4591}
4592
4593static void
4594spa_vdev_resilver_done(spa_t *spa)
4595{
4596	vdev_t *vd, *pvd, *ppvd;
4597	uint64_t guid, sguid, pguid, ppguid;
4598
4599	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4600
4601	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4602		pvd = vd->vdev_parent;
4603		ppvd = pvd->vdev_parent;
4604		guid = vd->vdev_guid;
4605		pguid = pvd->vdev_guid;
4606		ppguid = ppvd->vdev_guid;
4607		sguid = 0;
4608		/*
4609		 * If we have just finished replacing a hot spared device, then
4610		 * we need to detach the parent's first child (the original hot
4611		 * spare) as well.
4612		 */
4613		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0) {
4614			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4615			ASSERT(ppvd->vdev_children == 2);
4616			sguid = ppvd->vdev_child[1]->vdev_guid;
4617		}
4618		spa_config_exit(spa, SCL_ALL, FTAG);
4619		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4620			return;
4621		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4622			return;
4623		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4624	}
4625
4626	spa_config_exit(spa, SCL_ALL, FTAG);
4627}
4628
4629/*
4630 * Update the stored path or FRU for this vdev.
4631 */
4632int
4633spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4634    boolean_t ispath)
4635{
4636	vdev_t *vd;
4637	boolean_t sync = B_FALSE;
4638
4639	spa_vdev_state_enter(spa, SCL_ALL);
4640
4641	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4642		return (spa_vdev_state_exit(spa, NULL, ENOENT));
4643
4644	if (!vd->vdev_ops->vdev_op_leaf)
4645		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
4646
4647	if (ispath) {
4648		if (strcmp(value, vd->vdev_path) != 0) {
4649			spa_strfree(vd->vdev_path);
4650			vd->vdev_path = spa_strdup(value);
4651			sync = B_TRUE;
4652		}
4653	} else {
4654		if (vd->vdev_fru == NULL) {
4655			vd->vdev_fru = spa_strdup(value);
4656			sync = B_TRUE;
4657		} else if (strcmp(value, vd->vdev_fru) != 0) {
4658			spa_strfree(vd->vdev_fru);
4659			vd->vdev_fru = spa_strdup(value);
4660			sync = B_TRUE;
4661		}
4662	}
4663
4664	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
4665}
4666
4667int
4668spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
4669{
4670	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
4671}
4672
4673int
4674spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
4675{
4676	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
4677}
4678
4679/*
4680 * ==========================================================================
4681 * SPA Scrubbing
4682 * ==========================================================================
4683 */
4684
4685int
4686spa_scrub(spa_t *spa, pool_scrub_type_t type)
4687{
4688	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4689
4690	if ((uint_t)type >= POOL_SCRUB_TYPES)
4691		return (ENOTSUP);
4692
4693	/*
4694	 * If a resilver was requested, but there is no DTL on a
4695	 * writeable leaf device, we have nothing to do.
4696	 */
4697	if (type == POOL_SCRUB_RESILVER &&
4698	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
4699		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
4700		return (0);
4701	}
4702
4703	if (type == POOL_SCRUB_EVERYTHING &&
4704	    spa->spa_dsl_pool->dp_scrub_func != SCRUB_FUNC_NONE &&
4705	    spa->spa_dsl_pool->dp_scrub_isresilver)
4706		return (EBUSY);
4707
4708	if (type == POOL_SCRUB_EVERYTHING || type == POOL_SCRUB_RESILVER) {
4709		return (dsl_pool_scrub_clean(spa->spa_dsl_pool));
4710	} else if (type == POOL_SCRUB_NONE) {
4711		return (dsl_pool_scrub_cancel(spa->spa_dsl_pool));
4712	} else {
4713		return (EINVAL);
4714	}
4715}
4716
4717/*
4718 * ==========================================================================
4719 * SPA async task processing
4720 * ==========================================================================
4721 */
4722
4723static void
4724spa_async_remove(spa_t *spa, vdev_t *vd)
4725{
4726	if (vd->vdev_remove_wanted) {
4727		vd->vdev_remove_wanted = 0;
4728		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
4729
4730		/*
4731		 * We want to clear the stats, but we don't want to do a full
4732		 * vdev_clear() as that will cause us to throw away
4733		 * degraded/faulted state as well as attempt to reopen the
4734		 * device, all of which is a waste.
4735		 */
4736		vd->vdev_stat.vs_read_errors = 0;
4737		vd->vdev_stat.vs_write_errors = 0;
4738		vd->vdev_stat.vs_checksum_errors = 0;
4739
4740		vdev_state_dirty(vd->vdev_top);
4741	}
4742
4743	for (int c = 0; c < vd->vdev_children; c++)
4744		spa_async_remove(spa, vd->vdev_child[c]);
4745}
4746
4747static void
4748spa_async_probe(spa_t *spa, vdev_t *vd)
4749{
4750	if (vd->vdev_probe_wanted) {
4751		vd->vdev_probe_wanted = 0;
4752		vdev_reopen(vd);	/* vdev_open() does the actual probe */
4753	}
4754
4755	for (int c = 0; c < vd->vdev_children; c++)
4756		spa_async_probe(spa, vd->vdev_child[c]);
4757}
4758
4759static void
4760spa_async_autoexpand(spa_t *spa, vdev_t *vd)
4761{
4762	sysevent_id_t eid;
4763	nvlist_t *attr;
4764	char *physpath;
4765
4766	if (!spa->spa_autoexpand)
4767		return;
4768
4769	for (int c = 0; c < vd->vdev_children; c++) {
4770		vdev_t *cvd = vd->vdev_child[c];
4771		spa_async_autoexpand(spa, cvd);
4772	}
4773
4774	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
4775		return;
4776
4777	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
4778	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
4779
4780	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4781	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
4782
4783	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
4784	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
4785
4786	nvlist_free(attr);
4787	kmem_free(physpath, MAXPATHLEN);
4788}
4789
4790static void
4791spa_async_thread(spa_t *spa)
4792{
4793	int tasks;
4794
4795	ASSERT(spa->spa_sync_on);
4796
4797	mutex_enter(&spa->spa_async_lock);
4798	tasks = spa->spa_async_tasks;
4799	spa->spa_async_tasks = 0;
4800	mutex_exit(&spa->spa_async_lock);
4801
4802	/*
4803	 * See if the config needs to be updated.
4804	 */
4805	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
4806		uint64_t old_space, new_space;
4807
4808		mutex_enter(&spa_namespace_lock);
4809		old_space = metaslab_class_get_space(spa_normal_class(spa));
4810		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4811		new_space = metaslab_class_get_space(spa_normal_class(spa));
4812		mutex_exit(&spa_namespace_lock);
4813
4814		/*
4815		 * If the pool grew as a result of the config update,
4816		 * then log an internal history event.
4817		 */
4818		if (new_space != old_space) {
4819			spa_history_internal_log(LOG_POOL_VDEV_ONLINE,
4820			    spa, NULL, CRED(),
4821			    "pool '%s' size: %llu(+%llu)",
4822			    spa_name(spa), new_space, new_space - old_space);
4823		}
4824	}
4825
4826	/*
4827	 * See if any devices need to be marked REMOVED.
4828	 */
4829	if (tasks & SPA_ASYNC_REMOVE) {
4830		spa_vdev_state_enter(spa, SCL_NONE);
4831		spa_async_remove(spa, spa->spa_root_vdev);
4832		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
4833			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
4834		for (int i = 0; i < spa->spa_spares.sav_count; i++)
4835			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
4836		(void) spa_vdev_state_exit(spa, NULL, 0);
4837	}
4838
4839	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
4840		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4841		spa_async_autoexpand(spa, spa->spa_root_vdev);
4842		spa_config_exit(spa, SCL_CONFIG, FTAG);
4843	}
4844
4845	/*
4846	 * See if any devices need to be probed.
4847	 */
4848	if (tasks & SPA_ASYNC_PROBE) {
4849		spa_vdev_state_enter(spa, SCL_NONE);
4850		spa_async_probe(spa, spa->spa_root_vdev);
4851		(void) spa_vdev_state_exit(spa, NULL, 0);
4852	}
4853
4854	/*
4855	 * If any devices are done replacing, detach them.
4856	 */
4857	if (tasks & SPA_ASYNC_RESILVER_DONE)
4858		spa_vdev_resilver_done(spa);
4859
4860	/*
4861	 * Kick off a resilver.
4862	 */
4863	if (tasks & SPA_ASYNC_RESILVER)
4864		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER) == 0);
4865
4866	/*
4867	 * Let the world know that we're done.
4868	 */
4869	mutex_enter(&spa->spa_async_lock);
4870	spa->spa_async_thread = NULL;
4871	cv_broadcast(&spa->spa_async_cv);
4872	mutex_exit(&spa->spa_async_lock);
4873	thread_exit();
4874}
4875
4876void
4877spa_async_suspend(spa_t *spa)
4878{
4879	mutex_enter(&spa->spa_async_lock);
4880	spa->spa_async_suspended++;
4881	while (spa->spa_async_thread != NULL)
4882		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
4883	mutex_exit(&spa->spa_async_lock);
4884}
4885
4886void
4887spa_async_resume(spa_t *spa)
4888{
4889	mutex_enter(&spa->spa_async_lock);
4890	ASSERT(spa->spa_async_suspended != 0);
4891	spa->spa_async_suspended--;
4892	mutex_exit(&spa->spa_async_lock);
4893}
4894
4895static void
4896spa_async_dispatch(spa_t *spa)
4897{
4898	mutex_enter(&spa->spa_async_lock);
4899	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
4900	    spa->spa_async_thread == NULL &&
4901	    rootdir != NULL && !vn_is_readonly(rootdir))
4902		spa->spa_async_thread = thread_create(NULL, 0,
4903		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
4904	mutex_exit(&spa->spa_async_lock);
4905}
4906
4907void
4908spa_async_request(spa_t *spa, int task)
4909{
4910	mutex_enter(&spa->spa_async_lock);
4911	spa->spa_async_tasks |= task;
4912	mutex_exit(&spa->spa_async_lock);
4913}
4914
4915/*
4916 * ==========================================================================
4917 * SPA syncing routines
4918 * ==========================================================================
4919 */
4920static void
4921spa_sync_deferred_bplist(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx, uint64_t txg)
4922{
4923	blkptr_t blk;
4924	uint64_t itor = 0;
4925	uint8_t c = 1;
4926
4927	while (bplist_iterate(bpl, &itor, &blk) == 0) {
4928		ASSERT(blk.blk_birth < txg);
4929		zio_free(spa, txg, &blk);
4930	}
4931
4932	bplist_vacate(bpl, tx);
4933
4934	/*
4935	 * Pre-dirty the first block so we sync to convergence faster.
4936	 * (Usually only the first block is needed.)
4937	 */
4938	dmu_write(bpl->bpl_mos, spa->spa_deferred_bplist_obj, 0, 1, &c, tx);
4939}
4940
4941static void
4942spa_sync_free(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
4943{
4944	zio_t *zio = arg;
4945
4946	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
4947	    zio->io_flags));
4948}
4949
4950static void
4951spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
4952{
4953	char *packed = NULL;
4954	size_t bufsize;
4955	size_t nvsize = 0;
4956	dmu_buf_t *db;
4957
4958	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
4959
4960	/*
4961	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
4962	 * information.  This avoids the dbuf_will_dirty() path and
4963	 * saves us a pre-read to get data we don't actually care about.
4964	 */
4965	bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
4966	packed = kmem_alloc(bufsize, KM_SLEEP);
4967
4968	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
4969	    KM_SLEEP) == 0);
4970	bzero(packed + nvsize, bufsize - nvsize);
4971
4972	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
4973
4974	kmem_free(packed, bufsize);
4975
4976	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
4977	dmu_buf_will_dirty(db, tx);
4978	*(uint64_t *)db->db_data = nvsize;
4979	dmu_buf_rele(db, FTAG);
4980}
4981
4982static void
4983spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
4984    const char *config, const char *entry)
4985{
4986	nvlist_t *nvroot;
4987	nvlist_t **list;
4988	int i;
4989
4990	if (!sav->sav_sync)
4991		return;
4992
4993	/*
4994	 * Update the MOS nvlist describing the list of available devices.
4995	 * spa_validate_aux() will have already made sure this nvlist is
4996	 * valid and the vdevs are labeled appropriately.
4997	 */
4998	if (sav->sav_object == 0) {
4999		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5000		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5001		    sizeof (uint64_t), tx);
5002		VERIFY(zap_update(spa->spa_meta_objset,
5003		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5004		    &sav->sav_object, tx) == 0);
5005	}
5006
5007	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5008	if (sav->sav_count == 0) {
5009		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5010	} else {
5011		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5012		for (i = 0; i < sav->sav_count; i++)
5013			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5014			    B_FALSE, B_FALSE, B_TRUE);
5015		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5016		    sav->sav_count) == 0);
5017		for (i = 0; i < sav->sav_count; i++)
5018			nvlist_free(list[i]);
5019		kmem_free(list, sav->sav_count * sizeof (void *));
5020	}
5021
5022	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5023	nvlist_free(nvroot);
5024
5025	sav->sav_sync = B_FALSE;
5026}
5027
5028static void
5029spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5030{
5031	nvlist_t *config;
5032
5033	if (list_is_empty(&spa->spa_config_dirty_list))
5034		return;
5035
5036	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5037
5038	config = spa_config_generate(spa, spa->spa_root_vdev,
5039	    dmu_tx_get_txg(tx), B_FALSE);
5040
5041	spa_config_exit(spa, SCL_STATE, FTAG);
5042
5043	if (spa->spa_config_syncing)
5044		nvlist_free(spa->spa_config_syncing);
5045	spa->spa_config_syncing = config;
5046
5047	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5048}
5049
5050/*
5051 * Set zpool properties.
5052 */
5053static void
5054spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
5055{
5056	spa_t *spa = arg1;
5057	objset_t *mos = spa->spa_meta_objset;
5058	nvlist_t *nvp = arg2;
5059	nvpair_t *elem;
5060	uint64_t intval;
5061	char *strval;
5062	zpool_prop_t prop;
5063	const char *propname;
5064	zprop_type_t proptype;
5065
5066	mutex_enter(&spa->spa_props_lock);
5067
5068	elem = NULL;
5069	while ((elem = nvlist_next_nvpair(nvp, elem))) {
5070		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5071		case ZPOOL_PROP_VERSION:
5072			/*
5073			 * Only set version for non-zpool-creation cases
5074			 * (set/import). spa_create() needs special care
5075			 * for version setting.
5076			 */
5077			if (tx->tx_txg != TXG_INITIAL) {
5078				VERIFY(nvpair_value_uint64(elem,
5079				    &intval) == 0);
5080				ASSERT(intval <= SPA_VERSION);
5081				ASSERT(intval >= spa_version(spa));
5082				spa->spa_uberblock.ub_version = intval;
5083				vdev_config_dirty(spa->spa_root_vdev);
5084			}
5085			break;
5086
5087		case ZPOOL_PROP_ALTROOT:
5088			/*
5089			 * 'altroot' is a non-persistent property. It should
5090			 * have been set temporarily at creation or import time.
5091			 */
5092			ASSERT(spa->spa_root != NULL);
5093			break;
5094
5095		case ZPOOL_PROP_CACHEFILE:
5096			/*
5097			 * 'cachefile' is also a non-persisitent property.
5098			 */
5099			break;
5100		default:
5101			/*
5102			 * Set pool property values in the poolprops mos object.
5103			 */
5104			if (spa->spa_pool_props_object == 0) {
5105				VERIFY((spa->spa_pool_props_object =
5106				    zap_create(mos, DMU_OT_POOL_PROPS,
5107				    DMU_OT_NONE, 0, tx)) > 0);
5108
5109				VERIFY(zap_update(mos,
5110				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5111				    8, 1, &spa->spa_pool_props_object, tx)
5112				    == 0);
5113			}
5114
5115			/* normalize the property name */
5116			propname = zpool_prop_to_name(prop);
5117			proptype = zpool_prop_get_type(prop);
5118
5119			if (nvpair_type(elem) == DATA_TYPE_STRING) {
5120				ASSERT(proptype == PROP_TYPE_STRING);
5121				VERIFY(nvpair_value_string(elem, &strval) == 0);
5122				VERIFY(zap_update(mos,
5123				    spa->spa_pool_props_object, propname,
5124				    1, strlen(strval) + 1, strval, tx) == 0);
5125
5126			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5127				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5128
5129				if (proptype == PROP_TYPE_INDEX) {
5130					const char *unused;
5131					VERIFY(zpool_prop_index_to_string(
5132					    prop, intval, &unused) == 0);
5133				}
5134				VERIFY(zap_update(mos,
5135				    spa->spa_pool_props_object, propname,
5136				    8, 1, &intval, tx) == 0);
5137			} else {
5138				ASSERT(0); /* not allowed */
5139			}
5140
5141			switch (prop) {
5142			case ZPOOL_PROP_DELEGATION:
5143				spa->spa_delegation = intval;
5144				break;
5145			case ZPOOL_PROP_BOOTFS:
5146				spa->spa_bootfs = intval;
5147				break;
5148			case ZPOOL_PROP_FAILUREMODE:
5149				spa->spa_failmode = intval;
5150				break;
5151			case ZPOOL_PROP_AUTOEXPAND:
5152				spa->spa_autoexpand = intval;
5153				spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
5154				break;
5155			case ZPOOL_PROP_DEDUPDITTO:
5156				spa->spa_dedup_ditto = intval;
5157				break;
5158			default:
5159				break;
5160			}
5161		}
5162
5163		/* log internal history if this is not a zpool create */
5164		if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5165		    tx->tx_txg != TXG_INITIAL) {
5166			spa_history_internal_log(LOG_POOL_PROPSET,
5167			    spa, tx, cr, "%s %lld %s",
5168			    nvpair_name(elem), intval, spa_name(spa));
5169		}
5170	}
5171
5172	mutex_exit(&spa->spa_props_lock);
5173}
5174
5175/*
5176 * Sync the specified transaction group.  New blocks may be dirtied as
5177 * part of the process, so we iterate until it converges.
5178 */
5179void
5180spa_sync(spa_t *spa, uint64_t txg)
5181{
5182	dsl_pool_t *dp = spa->spa_dsl_pool;
5183	objset_t *mos = spa->spa_meta_objset;
5184	bplist_t *defer_bpl = &spa->spa_deferred_bplist;
5185	bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5186	vdev_t *rvd = spa->spa_root_vdev;
5187	vdev_t *vd;
5188	dmu_tx_t *tx;
5189	int error;
5190
5191	/*
5192	 * Lock out configuration changes.
5193	 */
5194	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5195
5196	spa->spa_syncing_txg = txg;
5197	spa->spa_sync_pass = 0;
5198
5199	/*
5200	 * If there are any pending vdev state changes, convert them
5201	 * into config changes that go out with this transaction group.
5202	 */
5203	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5204	while (list_head(&spa->spa_state_dirty_list) != NULL) {
5205		/*
5206		 * We need the write lock here because, for aux vdevs,
5207		 * calling vdev_config_dirty() modifies sav_config.
5208		 * This is ugly and will become unnecessary when we
5209		 * eliminate the aux vdev wart by integrating all vdevs
5210		 * into the root vdev tree.
5211		 */
5212		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5213		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5214		while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5215			vdev_state_clean(vd);
5216			vdev_config_dirty(vd);
5217		}
5218		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5219		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5220	}
5221	spa_config_exit(spa, SCL_STATE, FTAG);
5222
5223	VERIFY(0 == bplist_open(defer_bpl, mos, spa->spa_deferred_bplist_obj));
5224
5225	tx = dmu_tx_create_assigned(dp, txg);
5226
5227	/*
5228	 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5229	 * set spa_deflate if we have no raid-z vdevs.
5230	 */
5231	if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5232	    spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5233		int i;
5234
5235		for (i = 0; i < rvd->vdev_children; i++) {
5236			vd = rvd->vdev_child[i];
5237			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5238				break;
5239		}
5240		if (i == rvd->vdev_children) {
5241			spa->spa_deflate = TRUE;
5242			VERIFY(0 == zap_add(spa->spa_meta_objset,
5243			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5244			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5245		}
5246	}
5247
5248	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5249	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5250		dsl_pool_create_origin(dp, tx);
5251
5252		/* Keeping the origin open increases spa_minref */
5253		spa->spa_minref += 3;
5254	}
5255
5256	if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5257	    spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5258		dsl_pool_upgrade_clones(dp, tx);
5259	}
5260
5261	/*
5262	 * If anything has changed in this txg, push the deferred frees
5263	 * from the previous txg.  If not, leave them alone so that we
5264	 * don't generate work on an otherwise idle system.
5265	 */
5266	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5267	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5268	    !txg_list_empty(&dp->dp_sync_tasks, txg))
5269		spa_sync_deferred_bplist(spa, defer_bpl, tx, txg);
5270
5271	/*
5272	 * Iterate to convergence.
5273	 */
5274	do {
5275		int pass = ++spa->spa_sync_pass;
5276
5277		spa_sync_config_object(spa, tx);
5278		spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5279		    ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5280		spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5281		    ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5282		spa_errlog_sync(spa, txg);
5283		dsl_pool_sync(dp, txg);
5284
5285		if (pass <= SYNC_PASS_DEFERRED_FREE) {
5286			zio_t *zio = zio_root(spa, NULL, NULL, 0);
5287			bplist_sync(free_bpl, spa_sync_free, zio, tx);
5288			VERIFY(zio_wait(zio) == 0);
5289		} else {
5290			bplist_sync(free_bpl, bplist_enqueue_cb, defer_bpl, tx);
5291		}
5292
5293		ddt_sync(spa, txg);
5294
5295		mutex_enter(&spa->spa_scrub_lock);
5296		while (spa->spa_scrub_inflight > 0)
5297			cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
5298		mutex_exit(&spa->spa_scrub_lock);
5299
5300		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
5301			vdev_sync(vd, txg);
5302
5303	} while (dmu_objset_is_dirty(mos, txg));
5304
5305	ASSERT(free_bpl->bpl_queue == NULL);
5306
5307	bplist_close(defer_bpl);
5308
5309	/*
5310	 * Rewrite the vdev configuration (which includes the uberblock)
5311	 * to commit the transaction group.
5312	 *
5313	 * If there are no dirty vdevs, we sync the uberblock to a few
5314	 * random top-level vdevs that are known to be visible in the
5315	 * config cache (see spa_vdev_add() for a complete description).
5316	 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
5317	 */
5318	for (;;) {
5319		/*
5320		 * We hold SCL_STATE to prevent vdev open/close/etc.
5321		 * while we're attempting to write the vdev labels.
5322		 */
5323		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5324
5325		if (list_is_empty(&spa->spa_config_dirty_list)) {
5326			vdev_t *svd[SPA_DVAS_PER_BP];
5327			int svdcount = 0;
5328			int children = rvd->vdev_children;
5329			int c0 = spa_get_random(children);
5330
5331			for (int c = 0; c < children; c++) {
5332				vd = rvd->vdev_child[(c0 + c) % children];
5333				if (vd->vdev_ms_array == 0 || vd->vdev_islog)
5334					continue;
5335				svd[svdcount++] = vd;
5336				if (svdcount == SPA_DVAS_PER_BP)
5337					break;
5338			}
5339			error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
5340			if (error != 0)
5341				error = vdev_config_sync(svd, svdcount, txg,
5342				    B_TRUE);
5343		} else {
5344			error = vdev_config_sync(rvd->vdev_child,
5345			    rvd->vdev_children, txg, B_FALSE);
5346			if (error != 0)
5347				error = vdev_config_sync(rvd->vdev_child,
5348				    rvd->vdev_children, txg, B_TRUE);
5349		}
5350
5351		spa_config_exit(spa, SCL_STATE, FTAG);
5352
5353		if (error == 0)
5354			break;
5355		zio_suspend(spa, NULL);
5356		zio_resume_wait(spa);
5357	}
5358	dmu_tx_commit(tx);
5359
5360	/*
5361	 * Clear the dirty config list.
5362	 */
5363	while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
5364		vdev_config_clean(vd);
5365
5366	/*
5367	 * Now that the new config has synced transactionally,
5368	 * let it become visible to the config cache.
5369	 */
5370	if (spa->spa_config_syncing != NULL) {
5371		spa_config_set(spa, spa->spa_config_syncing);
5372		spa->spa_config_txg = txg;
5373		spa->spa_config_syncing = NULL;
5374	}
5375
5376	spa->spa_ubsync = spa->spa_uberblock;
5377
5378	dsl_pool_sync_done(dp, txg);
5379
5380	/*
5381	 * Update usable space statistics.
5382	 */
5383	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
5384		vdev_sync_done(vd, txg);
5385
5386	spa_update_dspace(spa);
5387
5388	/*
5389	 * It had better be the case that we didn't dirty anything
5390	 * since vdev_config_sync().
5391	 */
5392	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
5393	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
5394	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
5395	ASSERT(defer_bpl->bpl_queue == NULL);
5396	ASSERT(free_bpl->bpl_queue == NULL);
5397
5398	spa->spa_sync_pass = 0;
5399
5400	spa_config_exit(spa, SCL_CONFIG, FTAG);
5401
5402	spa_handle_ignored_writes(spa);
5403
5404	/*
5405	 * If any async tasks have been requested, kick them off.
5406	 */
5407	spa_async_dispatch(spa);
5408}
5409
5410/*
5411 * Sync all pools.  We don't want to hold the namespace lock across these
5412 * operations, so we take a reference on the spa_t and drop the lock during the
5413 * sync.
5414 */
5415void
5416spa_sync_allpools(void)
5417{
5418	spa_t *spa = NULL;
5419	mutex_enter(&spa_namespace_lock);
5420	while ((spa = spa_next(spa)) != NULL) {
5421		if (spa_state(spa) != POOL_STATE_ACTIVE || spa_suspended(spa))
5422			continue;
5423		spa_open_ref(spa, FTAG);
5424		mutex_exit(&spa_namespace_lock);
5425		txg_wait_synced(spa_get_dsl(spa), 0);
5426		mutex_enter(&spa_namespace_lock);
5427		spa_close(spa, FTAG);
5428	}
5429	mutex_exit(&spa_namespace_lock);
5430}
5431
5432/*
5433 * ==========================================================================
5434 * Miscellaneous routines
5435 * ==========================================================================
5436 */
5437
5438/*
5439 * Remove all pools in the system.
5440 */
5441void
5442spa_evict_all(void)
5443{
5444	spa_t *spa;
5445
5446	/*
5447	 * Remove all cached state.  All pools should be closed now,
5448	 * so every spa in the AVL tree should be unreferenced.
5449	 */
5450	mutex_enter(&spa_namespace_lock);
5451	while ((spa = spa_next(NULL)) != NULL) {
5452		/*
5453		 * Stop async tasks.  The async thread may need to detach
5454		 * a device that's been replaced, which requires grabbing
5455		 * spa_namespace_lock, so we must drop it here.
5456		 */
5457		spa_open_ref(spa, FTAG);
5458		mutex_exit(&spa_namespace_lock);
5459		spa_async_suspend(spa);
5460		mutex_enter(&spa_namespace_lock);
5461		spa_close(spa, FTAG);
5462
5463		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5464			spa_unload(spa);
5465			spa_deactivate(spa);
5466		}
5467		spa_remove(spa);
5468	}
5469	mutex_exit(&spa_namespace_lock);
5470}
5471
5472vdev_t *
5473spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
5474{
5475	vdev_t *vd;
5476	int i;
5477
5478	if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
5479		return (vd);
5480
5481	if (aux) {
5482		for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
5483			vd = spa->spa_l2cache.sav_vdevs[i];
5484			if (vd->vdev_guid == guid)
5485				return (vd);
5486		}
5487
5488		for (i = 0; i < spa->spa_spares.sav_count; i++) {
5489			vd = spa->spa_spares.sav_vdevs[i];
5490			if (vd->vdev_guid == guid)
5491				return (vd);
5492		}
5493	}
5494
5495	return (NULL);
5496}
5497
5498void
5499spa_upgrade(spa_t *spa, uint64_t version)
5500{
5501	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5502
5503	/*
5504	 * This should only be called for a non-faulted pool, and since a
5505	 * future version would result in an unopenable pool, this shouldn't be
5506	 * possible.
5507	 */
5508	ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
5509	ASSERT(version >= spa->spa_uberblock.ub_version);
5510
5511	spa->spa_uberblock.ub_version = version;
5512	vdev_config_dirty(spa->spa_root_vdev);
5513
5514	spa_config_exit(spa, SCL_ALL, FTAG);
5515
5516	txg_wait_synced(spa_get_dsl(spa), 0);
5517}
5518
5519boolean_t
5520spa_has_spare(spa_t *spa, uint64_t guid)
5521{
5522	int i;
5523	uint64_t spareguid;
5524	spa_aux_vdev_t *sav = &spa->spa_spares;
5525
5526	for (i = 0; i < sav->sav_count; i++)
5527		if (sav->sav_vdevs[i]->vdev_guid == guid)
5528			return (B_TRUE);
5529
5530	for (i = 0; i < sav->sav_npending; i++) {
5531		if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
5532		    &spareguid) == 0 && spareguid == guid)
5533			return (B_TRUE);
5534	}
5535
5536	return (B_FALSE);
5537}
5538
5539/*
5540 * Check if a pool has an active shared spare device.
5541 * Note: reference count of an active spare is 2, as a spare and as a replace
5542 */
5543static boolean_t
5544spa_has_active_shared_spare(spa_t *spa)
5545{
5546	int i, refcnt;
5547	uint64_t pool;
5548	spa_aux_vdev_t *sav = &spa->spa_spares;
5549
5550	for (i = 0; i < sav->sav_count; i++) {
5551		if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
5552		    &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
5553		    refcnt > 2)
5554			return (B_TRUE);
5555	}
5556
5557	return (B_FALSE);
5558}
5559
5560/*
5561 * Post a sysevent corresponding to the given event.  The 'name' must be one of
5562 * the event definitions in sys/sysevent/eventdefs.h.  The payload will be
5563 * filled in from the spa and (optionally) the vdev.  This doesn't do anything
5564 * in the userland libzpool, as we don't want consumers to misinterpret ztest
5565 * or zdb as real changes.
5566 */
5567void
5568spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
5569{
5570#ifdef _KERNEL
5571	sysevent_t		*ev;
5572	sysevent_attr_list_t	*attr = NULL;
5573	sysevent_value_t	value;
5574	sysevent_id_t		eid;
5575
5576	ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
5577	    SE_SLEEP);
5578
5579	value.value_type = SE_DATA_TYPE_STRING;
5580	value.value.sv_string = spa_name(spa);
5581	if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
5582		goto done;
5583
5584	value.value_type = SE_DATA_TYPE_UINT64;
5585	value.value.sv_uint64 = spa_guid(spa);
5586	if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
5587		goto done;
5588
5589	if (vd) {
5590		value.value_type = SE_DATA_TYPE_UINT64;
5591		value.value.sv_uint64 = vd->vdev_guid;
5592		if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
5593		    SE_SLEEP) != 0)
5594			goto done;
5595
5596		if (vd->vdev_path) {
5597			value.value_type = SE_DATA_TYPE_STRING;
5598			value.value.sv_string = vd->vdev_path;
5599			if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
5600			    &value, SE_SLEEP) != 0)
5601				goto done;
5602		}
5603	}
5604
5605	if (sysevent_attach_attributes(ev, attr) != 0)
5606		goto done;
5607	attr = NULL;
5608
5609	(void) log_sysevent(ev, SE_SLEEP, &eid);
5610
5611done:
5612	if (attr)
5613		sysevent_free_attr(attr);
5614	sysevent_free(ev);
5615#endif
5616}
5617