spa.c revision fb09f5aad449c97fe309678f3f604982b563a96f
1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
25 * Copyright (c) 2012 by Delphix. All rights reserved.
26 */
27
28/*
29 * This file contains all the routines used when modifying on-disk SPA state.
30 * This includes opening, importing, destroying, exporting a pool, and syncing a
31 * pool.
32 */
33
34#include <sys/zfs_context.h>
35#include <sys/fm/fs/zfs.h>
36#include <sys/spa_impl.h>
37#include <sys/zio.h>
38#include <sys/zio_checksum.h>
39#include <sys/dmu.h>
40#include <sys/dmu_tx.h>
41#include <sys/zap.h>
42#include <sys/zil.h>
43#include <sys/ddt.h>
44#include <sys/vdev_impl.h>
45#include <sys/metaslab.h>
46#include <sys/metaslab_impl.h>
47#include <sys/uberblock_impl.h>
48#include <sys/txg.h>
49#include <sys/avl.h>
50#include <sys/dmu_traverse.h>
51#include <sys/dmu_objset.h>
52#include <sys/unique.h>
53#include <sys/dsl_pool.h>
54#include <sys/dsl_dataset.h>
55#include <sys/dsl_dir.h>
56#include <sys/dsl_prop.h>
57#include <sys/dsl_synctask.h>
58#include <sys/fs/zfs.h>
59#include <sys/arc.h>
60#include <sys/callb.h>
61#include <sys/systeminfo.h>
62#include <sys/spa_boot.h>
63#include <sys/zfs_ioctl.h>
64#include <sys/dsl_scan.h>
65#include <sys/zfeature.h>
66
67#ifdef	_KERNEL
68#include <sys/bootprops.h>
69#include <sys/callb.h>
70#include <sys/cpupart.h>
71#include <sys/pool.h>
72#include <sys/sysdc.h>
73#include <sys/zone.h>
74#endif	/* _KERNEL */
75
76#include "zfs_prop.h"
77#include "zfs_comutil.h"
78
79typedef enum zti_modes {
80	zti_mode_fixed,			/* value is # of threads (min 1) */
81	zti_mode_online_percent,	/* value is % of online CPUs */
82	zti_mode_batch,			/* cpu-intensive; value is ignored */
83	zti_mode_null,			/* don't create a taskq */
84	zti_nmodes
85} zti_modes_t;
86
87#define	ZTI_FIX(n)	{ zti_mode_fixed, (n) }
88#define	ZTI_PCT(n)	{ zti_mode_online_percent, (n) }
89#define	ZTI_BATCH	{ zti_mode_batch, 0 }
90#define	ZTI_NULL	{ zti_mode_null, 0 }
91
92#define	ZTI_ONE		ZTI_FIX(1)
93
94typedef struct zio_taskq_info {
95	enum zti_modes zti_mode;
96	uint_t zti_value;
97} zio_taskq_info_t;
98
99static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
100	"issue", "issue_high", "intr", "intr_high"
101};
102
103/*
104 * Define the taskq threads for the following I/O types:
105 * 	NULL, READ, WRITE, FREE, CLAIM, and IOCTL
106 */
107const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
108	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
109	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
110	{ ZTI_FIX(8),	ZTI_NULL,	ZTI_BATCH,	ZTI_NULL },
111	{ ZTI_BATCH,	ZTI_FIX(5),	ZTI_FIX(8),	ZTI_FIX(5) },
112	{ ZTI_FIX(100),	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
113	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
114	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
115};
116
117static dsl_syncfunc_t spa_sync_version;
118static dsl_syncfunc_t spa_sync_props;
119static dsl_checkfunc_t spa_change_guid_check;
120static dsl_syncfunc_t spa_change_guid_sync;
121static boolean_t spa_has_active_shared_spare(spa_t *spa);
122static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
123    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
124    char **ereport);
125static void spa_vdev_resilver_done(spa_t *spa);
126
127uint_t		zio_taskq_batch_pct = 100;	/* 1 thread per cpu in pset */
128id_t		zio_taskq_psrset_bind = PS_NONE;
129boolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
130uint_t		zio_taskq_basedc = 80;		/* base duty cycle */
131
132boolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
133
134/*
135 * This (illegal) pool name is used when temporarily importing a spa_t in order
136 * to get the vdev stats associated with the imported devices.
137 */
138#define	TRYIMPORT_NAME	"$import"
139
140/*
141 * ==========================================================================
142 * SPA properties routines
143 * ==========================================================================
144 */
145
146/*
147 * Add a (source=src, propname=propval) list to an nvlist.
148 */
149static void
150spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
151    uint64_t intval, zprop_source_t src)
152{
153	const char *propname = zpool_prop_to_name(prop);
154	nvlist_t *propval;
155
156	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
157	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
158
159	if (strval != NULL)
160		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
161	else
162		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
163
164	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
165	nvlist_free(propval);
166}
167
168/*
169 * Get property values from the spa configuration.
170 */
171static void
172spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
173{
174	vdev_t *rvd = spa->spa_root_vdev;
175	dsl_pool_t *pool = spa->spa_dsl_pool;
176	uint64_t size;
177	uint64_t alloc;
178	uint64_t space;
179	uint64_t cap, version;
180	zprop_source_t src = ZPROP_SRC_NONE;
181	spa_config_dirent_t *dp;
182
183	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
184
185	if (rvd != NULL) {
186		alloc = metaslab_class_get_alloc(spa_normal_class(spa));
187		size = metaslab_class_get_space(spa_normal_class(spa));
188		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
189		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
190		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
191		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
192		    size - alloc, src);
193
194		space = 0;
195		for (int c = 0; c < rvd->vdev_children; c++) {
196			vdev_t *tvd = rvd->vdev_child[c];
197			space += tvd->vdev_max_asize - tvd->vdev_asize;
198		}
199		spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
200		    src);
201
202		spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
203		    (spa_mode(spa) == FREAD), src);
204
205		cap = (size == 0) ? 0 : (alloc * 100 / size);
206		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
207
208		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
209		    ddt_get_pool_dedup_ratio(spa), src);
210
211		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
212		    rvd->vdev_state, src);
213
214		version = spa_version(spa);
215		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
216			src = ZPROP_SRC_DEFAULT;
217		else
218			src = ZPROP_SRC_LOCAL;
219		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
220	}
221
222	if (pool != NULL) {
223		dsl_dir_t *freedir = pool->dp_free_dir;
224
225		/*
226		 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
227		 * when opening pools before this version freedir will be NULL.
228		 */
229		if (freedir != NULL) {
230			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
231			    freedir->dd_phys->dd_used_bytes, src);
232		} else {
233			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
234			    NULL, 0, src);
235		}
236	}
237
238	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
239
240	if (spa->spa_comment != NULL) {
241		spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
242		    0, ZPROP_SRC_LOCAL);
243	}
244
245	if (spa->spa_root != NULL)
246		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
247		    0, ZPROP_SRC_LOCAL);
248
249	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
250		if (dp->scd_path == NULL) {
251			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
252			    "none", 0, ZPROP_SRC_LOCAL);
253		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
254			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
255			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
256		}
257	}
258}
259
260/*
261 * Get zpool property values.
262 */
263int
264spa_prop_get(spa_t *spa, nvlist_t **nvp)
265{
266	objset_t *mos = spa->spa_meta_objset;
267	zap_cursor_t zc;
268	zap_attribute_t za;
269	int err;
270
271	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
272
273	mutex_enter(&spa->spa_props_lock);
274
275	/*
276	 * Get properties from the spa config.
277	 */
278	spa_prop_get_config(spa, nvp);
279
280	/* If no pool property object, no more prop to get. */
281	if (mos == NULL || spa->spa_pool_props_object == 0) {
282		mutex_exit(&spa->spa_props_lock);
283		return (0);
284	}
285
286	/*
287	 * Get properties from the MOS pool property object.
288	 */
289	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
290	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
291	    zap_cursor_advance(&zc)) {
292		uint64_t intval = 0;
293		char *strval = NULL;
294		zprop_source_t src = ZPROP_SRC_DEFAULT;
295		zpool_prop_t prop;
296
297		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
298			continue;
299
300		switch (za.za_integer_length) {
301		case 8:
302			/* integer property */
303			if (za.za_first_integer !=
304			    zpool_prop_default_numeric(prop))
305				src = ZPROP_SRC_LOCAL;
306
307			if (prop == ZPOOL_PROP_BOOTFS) {
308				dsl_pool_t *dp;
309				dsl_dataset_t *ds = NULL;
310
311				dp = spa_get_dsl(spa);
312				rw_enter(&dp->dp_config_rwlock, RW_READER);
313				if (err = dsl_dataset_hold_obj(dp,
314				    za.za_first_integer, FTAG, &ds)) {
315					rw_exit(&dp->dp_config_rwlock);
316					break;
317				}
318
319				strval = kmem_alloc(
320				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
321				    KM_SLEEP);
322				dsl_dataset_name(ds, strval);
323				dsl_dataset_rele(ds, FTAG);
324				rw_exit(&dp->dp_config_rwlock);
325			} else {
326				strval = NULL;
327				intval = za.za_first_integer;
328			}
329
330			spa_prop_add_list(*nvp, prop, strval, intval, src);
331
332			if (strval != NULL)
333				kmem_free(strval,
334				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
335
336			break;
337
338		case 1:
339			/* string property */
340			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
341			err = zap_lookup(mos, spa->spa_pool_props_object,
342			    za.za_name, 1, za.za_num_integers, strval);
343			if (err) {
344				kmem_free(strval, za.za_num_integers);
345				break;
346			}
347			spa_prop_add_list(*nvp, prop, strval, 0, src);
348			kmem_free(strval, za.za_num_integers);
349			break;
350
351		default:
352			break;
353		}
354	}
355	zap_cursor_fini(&zc);
356	mutex_exit(&spa->spa_props_lock);
357out:
358	if (err && err != ENOENT) {
359		nvlist_free(*nvp);
360		*nvp = NULL;
361		return (err);
362	}
363
364	return (0);
365}
366
367/*
368 * Validate the given pool properties nvlist and modify the list
369 * for the property values to be set.
370 */
371static int
372spa_prop_validate(spa_t *spa, nvlist_t *props)
373{
374	nvpair_t *elem;
375	int error = 0, reset_bootfs = 0;
376	uint64_t objnum;
377	boolean_t has_feature = B_FALSE;
378
379	elem = NULL;
380	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
381		uint64_t intval;
382		char *strval, *slash, *check, *fname;
383		const char *propname = nvpair_name(elem);
384		zpool_prop_t prop = zpool_name_to_prop(propname);
385
386		switch (prop) {
387		case ZPROP_INVAL:
388			if (!zpool_prop_feature(propname)) {
389				error = EINVAL;
390				break;
391			}
392
393			/*
394			 * Sanitize the input.
395			 */
396			if (nvpair_type(elem) != DATA_TYPE_UINT64) {
397				error = EINVAL;
398				break;
399			}
400
401			if (nvpair_value_uint64(elem, &intval) != 0) {
402				error = EINVAL;
403				break;
404			}
405
406			if (intval != 0) {
407				error = EINVAL;
408				break;
409			}
410
411			fname = strchr(propname, '@') + 1;
412			if (zfeature_lookup_name(fname, NULL) != 0) {
413				error = EINVAL;
414				break;
415			}
416
417			has_feature = B_TRUE;
418			break;
419
420		case ZPOOL_PROP_VERSION:
421			error = nvpair_value_uint64(elem, &intval);
422			if (!error &&
423			    (intval < spa_version(spa) ||
424			    intval > SPA_VERSION_BEFORE_FEATURES ||
425			    has_feature))
426				error = EINVAL;
427			break;
428
429		case ZPOOL_PROP_DELEGATION:
430		case ZPOOL_PROP_AUTOREPLACE:
431		case ZPOOL_PROP_LISTSNAPS:
432		case ZPOOL_PROP_AUTOEXPAND:
433			error = nvpair_value_uint64(elem, &intval);
434			if (!error && intval > 1)
435				error = EINVAL;
436			break;
437
438		case ZPOOL_PROP_BOOTFS:
439			/*
440			 * If the pool version is less than SPA_VERSION_BOOTFS,
441			 * or the pool is still being created (version == 0),
442			 * the bootfs property cannot be set.
443			 */
444			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
445				error = ENOTSUP;
446				break;
447			}
448
449			/*
450			 * Make sure the vdev config is bootable
451			 */
452			if (!vdev_is_bootable(spa->spa_root_vdev)) {
453				error = ENOTSUP;
454				break;
455			}
456
457			reset_bootfs = 1;
458
459			error = nvpair_value_string(elem, &strval);
460
461			if (!error) {
462				objset_t *os;
463				uint64_t compress;
464
465				if (strval == NULL || strval[0] == '\0') {
466					objnum = zpool_prop_default_numeric(
467					    ZPOOL_PROP_BOOTFS);
468					break;
469				}
470
471				if (error = dmu_objset_hold(strval, FTAG, &os))
472					break;
473
474				/* Must be ZPL and not gzip compressed. */
475
476				if (dmu_objset_type(os) != DMU_OST_ZFS) {
477					error = ENOTSUP;
478				} else if ((error = dsl_prop_get_integer(strval,
479				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
480				    &compress, NULL)) == 0 &&
481				    !BOOTFS_COMPRESS_VALID(compress)) {
482					error = ENOTSUP;
483				} else {
484					objnum = dmu_objset_id(os);
485				}
486				dmu_objset_rele(os, FTAG);
487			}
488			break;
489
490		case ZPOOL_PROP_FAILUREMODE:
491			error = nvpair_value_uint64(elem, &intval);
492			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
493			    intval > ZIO_FAILURE_MODE_PANIC))
494				error = EINVAL;
495
496			/*
497			 * This is a special case which only occurs when
498			 * the pool has completely failed. This allows
499			 * the user to change the in-core failmode property
500			 * without syncing it out to disk (I/Os might
501			 * currently be blocked). We do this by returning
502			 * EIO to the caller (spa_prop_set) to trick it
503			 * into thinking we encountered a property validation
504			 * error.
505			 */
506			if (!error && spa_suspended(spa)) {
507				spa->spa_failmode = intval;
508				error = EIO;
509			}
510			break;
511
512		case ZPOOL_PROP_CACHEFILE:
513			if ((error = nvpair_value_string(elem, &strval)) != 0)
514				break;
515
516			if (strval[0] == '\0')
517				break;
518
519			if (strcmp(strval, "none") == 0)
520				break;
521
522			if (strval[0] != '/') {
523				error = EINVAL;
524				break;
525			}
526
527			slash = strrchr(strval, '/');
528			ASSERT(slash != NULL);
529
530			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
531			    strcmp(slash, "/..") == 0)
532				error = EINVAL;
533			break;
534
535		case ZPOOL_PROP_COMMENT:
536			if ((error = nvpair_value_string(elem, &strval)) != 0)
537				break;
538			for (check = strval; *check != '\0'; check++) {
539				/*
540				 * The kernel doesn't have an easy isprint()
541				 * check.  For this kernel check, we merely
542				 * check ASCII apart from DEL.  Fix this if
543				 * there is an easy-to-use kernel isprint().
544				 */
545				if (*check >= 0x7f) {
546					error = EINVAL;
547					break;
548				}
549				check++;
550			}
551			if (strlen(strval) > ZPROP_MAX_COMMENT)
552				error = E2BIG;
553			break;
554
555		case ZPOOL_PROP_DEDUPDITTO:
556			if (spa_version(spa) < SPA_VERSION_DEDUP)
557				error = ENOTSUP;
558			else
559				error = nvpair_value_uint64(elem, &intval);
560			if (error == 0 &&
561			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
562				error = EINVAL;
563			break;
564		}
565
566		if (error)
567			break;
568	}
569
570	if (!error && reset_bootfs) {
571		error = nvlist_remove(props,
572		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
573
574		if (!error) {
575			error = nvlist_add_uint64(props,
576			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
577		}
578	}
579
580	return (error);
581}
582
583void
584spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
585{
586	char *cachefile;
587	spa_config_dirent_t *dp;
588
589	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
590	    &cachefile) != 0)
591		return;
592
593	dp = kmem_alloc(sizeof (spa_config_dirent_t),
594	    KM_SLEEP);
595
596	if (cachefile[0] == '\0')
597		dp->scd_path = spa_strdup(spa_config_path);
598	else if (strcmp(cachefile, "none") == 0)
599		dp->scd_path = NULL;
600	else
601		dp->scd_path = spa_strdup(cachefile);
602
603	list_insert_head(&spa->spa_config_list, dp);
604	if (need_sync)
605		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
606}
607
608int
609spa_prop_set(spa_t *spa, nvlist_t *nvp)
610{
611	int error;
612	nvpair_t *elem = NULL;
613	boolean_t need_sync = B_FALSE;
614
615	if ((error = spa_prop_validate(spa, nvp)) != 0)
616		return (error);
617
618	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
619		zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
620
621		if (prop == ZPOOL_PROP_CACHEFILE ||
622		    prop == ZPOOL_PROP_ALTROOT ||
623		    prop == ZPOOL_PROP_READONLY)
624			continue;
625
626		if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
627			uint64_t ver;
628
629			if (prop == ZPOOL_PROP_VERSION) {
630				VERIFY(nvpair_value_uint64(elem, &ver) == 0);
631			} else {
632				ASSERT(zpool_prop_feature(nvpair_name(elem)));
633				ver = SPA_VERSION_FEATURES;
634				need_sync = B_TRUE;
635			}
636
637			/* Save time if the version is already set. */
638			if (ver == spa_version(spa))
639				continue;
640
641			/*
642			 * In addition to the pool directory object, we might
643			 * create the pool properties object, the features for
644			 * read object, the features for write object, or the
645			 * feature descriptions object.
646			 */
647			error = dsl_sync_task_do(spa_get_dsl(spa), NULL,
648			    spa_sync_version, spa, &ver, 6);
649			if (error)
650				return (error);
651			continue;
652		}
653
654		need_sync = B_TRUE;
655		break;
656	}
657
658	if (need_sync) {
659		return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
660		    spa, nvp, 6));
661	}
662
663	return (0);
664}
665
666/*
667 * If the bootfs property value is dsobj, clear it.
668 */
669void
670spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
671{
672	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
673		VERIFY(zap_remove(spa->spa_meta_objset,
674		    spa->spa_pool_props_object,
675		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
676		spa->spa_bootfs = 0;
677	}
678}
679
680/*ARGSUSED*/
681static int
682spa_change_guid_check(void *arg1, void *arg2, dmu_tx_t *tx)
683{
684	spa_t *spa = arg1;
685	uint64_t *newguid = arg2;
686	vdev_t *rvd = spa->spa_root_vdev;
687	uint64_t vdev_state;
688
689	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
690	vdev_state = rvd->vdev_state;
691	spa_config_exit(spa, SCL_STATE, FTAG);
692
693	if (vdev_state != VDEV_STATE_HEALTHY)
694		return (ENXIO);
695
696	ASSERT3U(spa_guid(spa), !=, *newguid);
697
698	return (0);
699}
700
701static void
702spa_change_guid_sync(void *arg1, void *arg2, dmu_tx_t *tx)
703{
704	spa_t *spa = arg1;
705	uint64_t *newguid = arg2;
706	uint64_t oldguid;
707	vdev_t *rvd = spa->spa_root_vdev;
708
709	oldguid = spa_guid(spa);
710
711	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
712	rvd->vdev_guid = *newguid;
713	rvd->vdev_guid_sum += (*newguid - oldguid);
714	vdev_config_dirty(rvd);
715	spa_config_exit(spa, SCL_STATE, FTAG);
716
717	spa_history_log_internal(spa, "guid change", tx, "old=%lld new=%lld",
718	    oldguid, *newguid);
719}
720
721/*
722 * Change the GUID for the pool.  This is done so that we can later
723 * re-import a pool built from a clone of our own vdevs.  We will modify
724 * the root vdev's guid, our own pool guid, and then mark all of our
725 * vdevs dirty.  Note that we must make sure that all our vdevs are
726 * online when we do this, or else any vdevs that weren't present
727 * would be orphaned from our pool.  We are also going to issue a
728 * sysevent to update any watchers.
729 */
730int
731spa_change_guid(spa_t *spa)
732{
733	int error;
734	uint64_t guid;
735
736	mutex_enter(&spa_namespace_lock);
737	guid = spa_generate_guid(NULL);
738
739	error = dsl_sync_task_do(spa_get_dsl(spa), spa_change_guid_check,
740	    spa_change_guid_sync, spa, &guid, 5);
741
742	if (error == 0) {
743		spa_config_sync(spa, B_FALSE, B_TRUE);
744		spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
745	}
746
747	mutex_exit(&spa_namespace_lock);
748
749	return (error);
750}
751
752/*
753 * ==========================================================================
754 * SPA state manipulation (open/create/destroy/import/export)
755 * ==========================================================================
756 */
757
758static int
759spa_error_entry_compare(const void *a, const void *b)
760{
761	spa_error_entry_t *sa = (spa_error_entry_t *)a;
762	spa_error_entry_t *sb = (spa_error_entry_t *)b;
763	int ret;
764
765	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
766	    sizeof (zbookmark_t));
767
768	if (ret < 0)
769		return (-1);
770	else if (ret > 0)
771		return (1);
772	else
773		return (0);
774}
775
776/*
777 * Utility function which retrieves copies of the current logs and
778 * re-initializes them in the process.
779 */
780void
781spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
782{
783	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
784
785	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
786	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
787
788	avl_create(&spa->spa_errlist_scrub,
789	    spa_error_entry_compare, sizeof (spa_error_entry_t),
790	    offsetof(spa_error_entry_t, se_avl));
791	avl_create(&spa->spa_errlist_last,
792	    spa_error_entry_compare, sizeof (spa_error_entry_t),
793	    offsetof(spa_error_entry_t, se_avl));
794}
795
796static taskq_t *
797spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
798    uint_t value)
799{
800	uint_t flags = 0;
801	boolean_t batch = B_FALSE;
802
803	switch (mode) {
804	case zti_mode_null:
805		return (NULL);		/* no taskq needed */
806
807	case zti_mode_fixed:
808		ASSERT3U(value, >=, 1);
809		value = MAX(value, 1);
810		break;
811
812	case zti_mode_batch:
813		batch = B_TRUE;
814		flags |= TASKQ_THREADS_CPU_PCT;
815		value = zio_taskq_batch_pct;
816		break;
817
818	case zti_mode_online_percent:
819		flags |= TASKQ_THREADS_CPU_PCT;
820		break;
821
822	default:
823		panic("unrecognized mode for %s taskq (%u:%u) in "
824		    "spa_activate()",
825		    name, mode, value);
826		break;
827	}
828
829	if (zio_taskq_sysdc && spa->spa_proc != &p0) {
830		if (batch)
831			flags |= TASKQ_DC_BATCH;
832
833		return (taskq_create_sysdc(name, value, 50, INT_MAX,
834		    spa->spa_proc, zio_taskq_basedc, flags));
835	}
836	return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
837	    spa->spa_proc, flags));
838}
839
840static void
841spa_create_zio_taskqs(spa_t *spa)
842{
843	for (int t = 0; t < ZIO_TYPES; t++) {
844		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
845			const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
846			enum zti_modes mode = ztip->zti_mode;
847			uint_t value = ztip->zti_value;
848			char name[32];
849
850			(void) snprintf(name, sizeof (name),
851			    "%s_%s", zio_type_name[t], zio_taskq_types[q]);
852
853			spa->spa_zio_taskq[t][q] =
854			    spa_taskq_create(spa, name, mode, value);
855		}
856	}
857}
858
859#ifdef _KERNEL
860static void
861spa_thread(void *arg)
862{
863	callb_cpr_t cprinfo;
864
865	spa_t *spa = arg;
866	user_t *pu = PTOU(curproc);
867
868	CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
869	    spa->spa_name);
870
871	ASSERT(curproc != &p0);
872	(void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
873	    "zpool-%s", spa->spa_name);
874	(void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
875
876	/* bind this thread to the requested psrset */
877	if (zio_taskq_psrset_bind != PS_NONE) {
878		pool_lock();
879		mutex_enter(&cpu_lock);
880		mutex_enter(&pidlock);
881		mutex_enter(&curproc->p_lock);
882
883		if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
884		    0, NULL, NULL) == 0)  {
885			curthread->t_bind_pset = zio_taskq_psrset_bind;
886		} else {
887			cmn_err(CE_WARN,
888			    "Couldn't bind process for zfs pool \"%s\" to "
889			    "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
890		}
891
892		mutex_exit(&curproc->p_lock);
893		mutex_exit(&pidlock);
894		mutex_exit(&cpu_lock);
895		pool_unlock();
896	}
897
898	if (zio_taskq_sysdc) {
899		sysdc_thread_enter(curthread, 100, 0);
900	}
901
902	spa->spa_proc = curproc;
903	spa->spa_did = curthread->t_did;
904
905	spa_create_zio_taskqs(spa);
906
907	mutex_enter(&spa->spa_proc_lock);
908	ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
909
910	spa->spa_proc_state = SPA_PROC_ACTIVE;
911	cv_broadcast(&spa->spa_proc_cv);
912
913	CALLB_CPR_SAFE_BEGIN(&cprinfo);
914	while (spa->spa_proc_state == SPA_PROC_ACTIVE)
915		cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
916	CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
917
918	ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
919	spa->spa_proc_state = SPA_PROC_GONE;
920	spa->spa_proc = &p0;
921	cv_broadcast(&spa->spa_proc_cv);
922	CALLB_CPR_EXIT(&cprinfo);	/* drops spa_proc_lock */
923
924	mutex_enter(&curproc->p_lock);
925	lwp_exit();
926}
927#endif
928
929/*
930 * Activate an uninitialized pool.
931 */
932static void
933spa_activate(spa_t *spa, int mode)
934{
935	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
936
937	spa->spa_state = POOL_STATE_ACTIVE;
938	spa->spa_mode = mode;
939
940	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
941	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
942
943	/* Try to create a covering process */
944	mutex_enter(&spa->spa_proc_lock);
945	ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
946	ASSERT(spa->spa_proc == &p0);
947	spa->spa_did = 0;
948
949	/* Only create a process if we're going to be around a while. */
950	if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
951		if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
952		    NULL, 0) == 0) {
953			spa->spa_proc_state = SPA_PROC_CREATED;
954			while (spa->spa_proc_state == SPA_PROC_CREATED) {
955				cv_wait(&spa->spa_proc_cv,
956				    &spa->spa_proc_lock);
957			}
958			ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
959			ASSERT(spa->spa_proc != &p0);
960			ASSERT(spa->spa_did != 0);
961		} else {
962#ifdef _KERNEL
963			cmn_err(CE_WARN,
964			    "Couldn't create process for zfs pool \"%s\"\n",
965			    spa->spa_name);
966#endif
967		}
968	}
969	mutex_exit(&spa->spa_proc_lock);
970
971	/* If we didn't create a process, we need to create our taskqs. */
972	if (spa->spa_proc == &p0) {
973		spa_create_zio_taskqs(spa);
974	}
975
976	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
977	    offsetof(vdev_t, vdev_config_dirty_node));
978	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
979	    offsetof(vdev_t, vdev_state_dirty_node));
980
981	txg_list_create(&spa->spa_vdev_txg_list,
982	    offsetof(struct vdev, vdev_txg_node));
983
984	avl_create(&spa->spa_errlist_scrub,
985	    spa_error_entry_compare, sizeof (spa_error_entry_t),
986	    offsetof(spa_error_entry_t, se_avl));
987	avl_create(&spa->spa_errlist_last,
988	    spa_error_entry_compare, sizeof (spa_error_entry_t),
989	    offsetof(spa_error_entry_t, se_avl));
990}
991
992/*
993 * Opposite of spa_activate().
994 */
995static void
996spa_deactivate(spa_t *spa)
997{
998	ASSERT(spa->spa_sync_on == B_FALSE);
999	ASSERT(spa->spa_dsl_pool == NULL);
1000	ASSERT(spa->spa_root_vdev == NULL);
1001	ASSERT(spa->spa_async_zio_root == NULL);
1002	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1003
1004	txg_list_destroy(&spa->spa_vdev_txg_list);
1005
1006	list_destroy(&spa->spa_config_dirty_list);
1007	list_destroy(&spa->spa_state_dirty_list);
1008
1009	for (int t = 0; t < ZIO_TYPES; t++) {
1010		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1011			if (spa->spa_zio_taskq[t][q] != NULL)
1012				taskq_destroy(spa->spa_zio_taskq[t][q]);
1013			spa->spa_zio_taskq[t][q] = NULL;
1014		}
1015	}
1016
1017	metaslab_class_destroy(spa->spa_normal_class);
1018	spa->spa_normal_class = NULL;
1019
1020	metaslab_class_destroy(spa->spa_log_class);
1021	spa->spa_log_class = NULL;
1022
1023	/*
1024	 * If this was part of an import or the open otherwise failed, we may
1025	 * still have errors left in the queues.  Empty them just in case.
1026	 */
1027	spa_errlog_drain(spa);
1028
1029	avl_destroy(&spa->spa_errlist_scrub);
1030	avl_destroy(&spa->spa_errlist_last);
1031
1032	spa->spa_state = POOL_STATE_UNINITIALIZED;
1033
1034	mutex_enter(&spa->spa_proc_lock);
1035	if (spa->spa_proc_state != SPA_PROC_NONE) {
1036		ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1037		spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1038		cv_broadcast(&spa->spa_proc_cv);
1039		while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1040			ASSERT(spa->spa_proc != &p0);
1041			cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1042		}
1043		ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1044		spa->spa_proc_state = SPA_PROC_NONE;
1045	}
1046	ASSERT(spa->spa_proc == &p0);
1047	mutex_exit(&spa->spa_proc_lock);
1048
1049	/*
1050	 * We want to make sure spa_thread() has actually exited the ZFS
1051	 * module, so that the module can't be unloaded out from underneath
1052	 * it.
1053	 */
1054	if (spa->spa_did != 0) {
1055		thread_join(spa->spa_did);
1056		spa->spa_did = 0;
1057	}
1058}
1059
1060/*
1061 * Verify a pool configuration, and construct the vdev tree appropriately.  This
1062 * will create all the necessary vdevs in the appropriate layout, with each vdev
1063 * in the CLOSED state.  This will prep the pool before open/creation/import.
1064 * All vdev validation is done by the vdev_alloc() routine.
1065 */
1066static int
1067spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1068    uint_t id, int atype)
1069{
1070	nvlist_t **child;
1071	uint_t children;
1072	int error;
1073
1074	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1075		return (error);
1076
1077	if ((*vdp)->vdev_ops->vdev_op_leaf)
1078		return (0);
1079
1080	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1081	    &child, &children);
1082
1083	if (error == ENOENT)
1084		return (0);
1085
1086	if (error) {
1087		vdev_free(*vdp);
1088		*vdp = NULL;
1089		return (EINVAL);
1090	}
1091
1092	for (int c = 0; c < children; c++) {
1093		vdev_t *vd;
1094		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1095		    atype)) != 0) {
1096			vdev_free(*vdp);
1097			*vdp = NULL;
1098			return (error);
1099		}
1100	}
1101
1102	ASSERT(*vdp != NULL);
1103
1104	return (0);
1105}
1106
1107/*
1108 * Opposite of spa_load().
1109 */
1110static void
1111spa_unload(spa_t *spa)
1112{
1113	int i;
1114
1115	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1116
1117	/*
1118	 * Stop async tasks.
1119	 */
1120	spa_async_suspend(spa);
1121
1122	/*
1123	 * Stop syncing.
1124	 */
1125	if (spa->spa_sync_on) {
1126		txg_sync_stop(spa->spa_dsl_pool);
1127		spa->spa_sync_on = B_FALSE;
1128	}
1129
1130	/*
1131	 * Wait for any outstanding async I/O to complete.
1132	 */
1133	if (spa->spa_async_zio_root != NULL) {
1134		(void) zio_wait(spa->spa_async_zio_root);
1135		spa->spa_async_zio_root = NULL;
1136	}
1137
1138	bpobj_close(&spa->spa_deferred_bpobj);
1139
1140	/*
1141	 * Close the dsl pool.
1142	 */
1143	if (spa->spa_dsl_pool) {
1144		dsl_pool_close(spa->spa_dsl_pool);
1145		spa->spa_dsl_pool = NULL;
1146		spa->spa_meta_objset = NULL;
1147	}
1148
1149	ddt_unload(spa);
1150
1151	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1152
1153	/*
1154	 * Drop and purge level 2 cache
1155	 */
1156	spa_l2cache_drop(spa);
1157
1158	/*
1159	 * Close all vdevs.
1160	 */
1161	if (spa->spa_root_vdev)
1162		vdev_free(spa->spa_root_vdev);
1163	ASSERT(spa->spa_root_vdev == NULL);
1164
1165	for (i = 0; i < spa->spa_spares.sav_count; i++)
1166		vdev_free(spa->spa_spares.sav_vdevs[i]);
1167	if (spa->spa_spares.sav_vdevs) {
1168		kmem_free(spa->spa_spares.sav_vdevs,
1169		    spa->spa_spares.sav_count * sizeof (void *));
1170		spa->spa_spares.sav_vdevs = NULL;
1171	}
1172	if (spa->spa_spares.sav_config) {
1173		nvlist_free(spa->spa_spares.sav_config);
1174		spa->spa_spares.sav_config = NULL;
1175	}
1176	spa->spa_spares.sav_count = 0;
1177
1178	for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1179		vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1180		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1181	}
1182	if (spa->spa_l2cache.sav_vdevs) {
1183		kmem_free(spa->spa_l2cache.sav_vdevs,
1184		    spa->spa_l2cache.sav_count * sizeof (void *));
1185		spa->spa_l2cache.sav_vdevs = NULL;
1186	}
1187	if (spa->spa_l2cache.sav_config) {
1188		nvlist_free(spa->spa_l2cache.sav_config);
1189		spa->spa_l2cache.sav_config = NULL;
1190	}
1191	spa->spa_l2cache.sav_count = 0;
1192
1193	spa->spa_async_suspended = 0;
1194
1195	if (spa->spa_comment != NULL) {
1196		spa_strfree(spa->spa_comment);
1197		spa->spa_comment = NULL;
1198	}
1199
1200	spa_config_exit(spa, SCL_ALL, FTAG);
1201}
1202
1203/*
1204 * Load (or re-load) the current list of vdevs describing the active spares for
1205 * this pool.  When this is called, we have some form of basic information in
1206 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1207 * then re-generate a more complete list including status information.
1208 */
1209static void
1210spa_load_spares(spa_t *spa)
1211{
1212	nvlist_t **spares;
1213	uint_t nspares;
1214	int i;
1215	vdev_t *vd, *tvd;
1216
1217	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1218
1219	/*
1220	 * First, close and free any existing spare vdevs.
1221	 */
1222	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1223		vd = spa->spa_spares.sav_vdevs[i];
1224
1225		/* Undo the call to spa_activate() below */
1226		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1227		    B_FALSE)) != NULL && tvd->vdev_isspare)
1228			spa_spare_remove(tvd);
1229		vdev_close(vd);
1230		vdev_free(vd);
1231	}
1232
1233	if (spa->spa_spares.sav_vdevs)
1234		kmem_free(spa->spa_spares.sav_vdevs,
1235		    spa->spa_spares.sav_count * sizeof (void *));
1236
1237	if (spa->spa_spares.sav_config == NULL)
1238		nspares = 0;
1239	else
1240		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1241		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1242
1243	spa->spa_spares.sav_count = (int)nspares;
1244	spa->spa_spares.sav_vdevs = NULL;
1245
1246	if (nspares == 0)
1247		return;
1248
1249	/*
1250	 * Construct the array of vdevs, opening them to get status in the
1251	 * process.   For each spare, there is potentially two different vdev_t
1252	 * structures associated with it: one in the list of spares (used only
1253	 * for basic validation purposes) and one in the active vdev
1254	 * configuration (if it's spared in).  During this phase we open and
1255	 * validate each vdev on the spare list.  If the vdev also exists in the
1256	 * active configuration, then we also mark this vdev as an active spare.
1257	 */
1258	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1259	    KM_SLEEP);
1260	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1261		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1262		    VDEV_ALLOC_SPARE) == 0);
1263		ASSERT(vd != NULL);
1264
1265		spa->spa_spares.sav_vdevs[i] = vd;
1266
1267		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1268		    B_FALSE)) != NULL) {
1269			if (!tvd->vdev_isspare)
1270				spa_spare_add(tvd);
1271
1272			/*
1273			 * We only mark the spare active if we were successfully
1274			 * able to load the vdev.  Otherwise, importing a pool
1275			 * with a bad active spare would result in strange
1276			 * behavior, because multiple pool would think the spare
1277			 * is actively in use.
1278			 *
1279			 * There is a vulnerability here to an equally bizarre
1280			 * circumstance, where a dead active spare is later
1281			 * brought back to life (onlined or otherwise).  Given
1282			 * the rarity of this scenario, and the extra complexity
1283			 * it adds, we ignore the possibility.
1284			 */
1285			if (!vdev_is_dead(tvd))
1286				spa_spare_activate(tvd);
1287		}
1288
1289		vd->vdev_top = vd;
1290		vd->vdev_aux = &spa->spa_spares;
1291
1292		if (vdev_open(vd) != 0)
1293			continue;
1294
1295		if (vdev_validate_aux(vd) == 0)
1296			spa_spare_add(vd);
1297	}
1298
1299	/*
1300	 * Recompute the stashed list of spares, with status information
1301	 * this time.
1302	 */
1303	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1304	    DATA_TYPE_NVLIST_ARRAY) == 0);
1305
1306	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1307	    KM_SLEEP);
1308	for (i = 0; i < spa->spa_spares.sav_count; i++)
1309		spares[i] = vdev_config_generate(spa,
1310		    spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1311	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1312	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1313	for (i = 0; i < spa->spa_spares.sav_count; i++)
1314		nvlist_free(spares[i]);
1315	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1316}
1317
1318/*
1319 * Load (or re-load) the current list of vdevs describing the active l2cache for
1320 * this pool.  When this is called, we have some form of basic information in
1321 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1322 * then re-generate a more complete list including status information.
1323 * Devices which are already active have their details maintained, and are
1324 * not re-opened.
1325 */
1326static void
1327spa_load_l2cache(spa_t *spa)
1328{
1329	nvlist_t **l2cache;
1330	uint_t nl2cache;
1331	int i, j, oldnvdevs;
1332	uint64_t guid;
1333	vdev_t *vd, **oldvdevs, **newvdevs;
1334	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1335
1336	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1337
1338	if (sav->sav_config != NULL) {
1339		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1340		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1341		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1342	} else {
1343		nl2cache = 0;
1344	}
1345
1346	oldvdevs = sav->sav_vdevs;
1347	oldnvdevs = sav->sav_count;
1348	sav->sav_vdevs = NULL;
1349	sav->sav_count = 0;
1350
1351	/*
1352	 * Process new nvlist of vdevs.
1353	 */
1354	for (i = 0; i < nl2cache; i++) {
1355		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1356		    &guid) == 0);
1357
1358		newvdevs[i] = NULL;
1359		for (j = 0; j < oldnvdevs; j++) {
1360			vd = oldvdevs[j];
1361			if (vd != NULL && guid == vd->vdev_guid) {
1362				/*
1363				 * Retain previous vdev for add/remove ops.
1364				 */
1365				newvdevs[i] = vd;
1366				oldvdevs[j] = NULL;
1367				break;
1368			}
1369		}
1370
1371		if (newvdevs[i] == NULL) {
1372			/*
1373			 * Create new vdev
1374			 */
1375			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1376			    VDEV_ALLOC_L2CACHE) == 0);
1377			ASSERT(vd != NULL);
1378			newvdevs[i] = vd;
1379
1380			/*
1381			 * Commit this vdev as an l2cache device,
1382			 * even if it fails to open.
1383			 */
1384			spa_l2cache_add(vd);
1385
1386			vd->vdev_top = vd;
1387			vd->vdev_aux = sav;
1388
1389			spa_l2cache_activate(vd);
1390
1391			if (vdev_open(vd) != 0)
1392				continue;
1393
1394			(void) vdev_validate_aux(vd);
1395
1396			if (!vdev_is_dead(vd))
1397				l2arc_add_vdev(spa, vd);
1398		}
1399	}
1400
1401	/*
1402	 * Purge vdevs that were dropped
1403	 */
1404	for (i = 0; i < oldnvdevs; i++) {
1405		uint64_t pool;
1406
1407		vd = oldvdevs[i];
1408		if (vd != NULL) {
1409			ASSERT(vd->vdev_isl2cache);
1410
1411			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1412			    pool != 0ULL && l2arc_vdev_present(vd))
1413				l2arc_remove_vdev(vd);
1414			vdev_clear_stats(vd);
1415			vdev_free(vd);
1416		}
1417	}
1418
1419	if (oldvdevs)
1420		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1421
1422	if (sav->sav_config == NULL)
1423		goto out;
1424
1425	sav->sav_vdevs = newvdevs;
1426	sav->sav_count = (int)nl2cache;
1427
1428	/*
1429	 * Recompute the stashed list of l2cache devices, with status
1430	 * information this time.
1431	 */
1432	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1433	    DATA_TYPE_NVLIST_ARRAY) == 0);
1434
1435	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1436	for (i = 0; i < sav->sav_count; i++)
1437		l2cache[i] = vdev_config_generate(spa,
1438		    sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1439	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1440	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1441out:
1442	for (i = 0; i < sav->sav_count; i++)
1443		nvlist_free(l2cache[i]);
1444	if (sav->sav_count)
1445		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1446}
1447
1448static int
1449load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1450{
1451	dmu_buf_t *db;
1452	char *packed = NULL;
1453	size_t nvsize = 0;
1454	int error;
1455	*value = NULL;
1456
1457	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1458	nvsize = *(uint64_t *)db->db_data;
1459	dmu_buf_rele(db, FTAG);
1460
1461	packed = kmem_alloc(nvsize, KM_SLEEP);
1462	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1463	    DMU_READ_PREFETCH);
1464	if (error == 0)
1465		error = nvlist_unpack(packed, nvsize, value, 0);
1466	kmem_free(packed, nvsize);
1467
1468	return (error);
1469}
1470
1471/*
1472 * Checks to see if the given vdev could not be opened, in which case we post a
1473 * sysevent to notify the autoreplace code that the device has been removed.
1474 */
1475static void
1476spa_check_removed(vdev_t *vd)
1477{
1478	for (int c = 0; c < vd->vdev_children; c++)
1479		spa_check_removed(vd->vdev_child[c]);
1480
1481	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1482		zfs_post_autoreplace(vd->vdev_spa, vd);
1483		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1484	}
1485}
1486
1487/*
1488 * Validate the current config against the MOS config
1489 */
1490static boolean_t
1491spa_config_valid(spa_t *spa, nvlist_t *config)
1492{
1493	vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1494	nvlist_t *nv;
1495
1496	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1497
1498	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1499	VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1500
1501	ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1502
1503	/*
1504	 * If we're doing a normal import, then build up any additional
1505	 * diagnostic information about missing devices in this config.
1506	 * We'll pass this up to the user for further processing.
1507	 */
1508	if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1509		nvlist_t **child, *nv;
1510		uint64_t idx = 0;
1511
1512		child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1513		    KM_SLEEP);
1514		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1515
1516		for (int c = 0; c < rvd->vdev_children; c++) {
1517			vdev_t *tvd = rvd->vdev_child[c];
1518			vdev_t *mtvd  = mrvd->vdev_child[c];
1519
1520			if (tvd->vdev_ops == &vdev_missing_ops &&
1521			    mtvd->vdev_ops != &vdev_missing_ops &&
1522			    mtvd->vdev_islog)
1523				child[idx++] = vdev_config_generate(spa, mtvd,
1524				    B_FALSE, 0);
1525		}
1526
1527		if (idx) {
1528			VERIFY(nvlist_add_nvlist_array(nv,
1529			    ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1530			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1531			    ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1532
1533			for (int i = 0; i < idx; i++)
1534				nvlist_free(child[i]);
1535		}
1536		nvlist_free(nv);
1537		kmem_free(child, rvd->vdev_children * sizeof (char **));
1538	}
1539
1540	/*
1541	 * Compare the root vdev tree with the information we have
1542	 * from the MOS config (mrvd). Check each top-level vdev
1543	 * with the corresponding MOS config top-level (mtvd).
1544	 */
1545	for (int c = 0; c < rvd->vdev_children; c++) {
1546		vdev_t *tvd = rvd->vdev_child[c];
1547		vdev_t *mtvd  = mrvd->vdev_child[c];
1548
1549		/*
1550		 * Resolve any "missing" vdevs in the current configuration.
1551		 * If we find that the MOS config has more accurate information
1552		 * about the top-level vdev then use that vdev instead.
1553		 */
1554		if (tvd->vdev_ops == &vdev_missing_ops &&
1555		    mtvd->vdev_ops != &vdev_missing_ops) {
1556
1557			if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1558				continue;
1559
1560			/*
1561			 * Device specific actions.
1562			 */
1563			if (mtvd->vdev_islog) {
1564				spa_set_log_state(spa, SPA_LOG_CLEAR);
1565			} else {
1566				/*
1567				 * XXX - once we have 'readonly' pool
1568				 * support we should be able to handle
1569				 * missing data devices by transitioning
1570				 * the pool to readonly.
1571				 */
1572				continue;
1573			}
1574
1575			/*
1576			 * Swap the missing vdev with the data we were
1577			 * able to obtain from the MOS config.
1578			 */
1579			vdev_remove_child(rvd, tvd);
1580			vdev_remove_child(mrvd, mtvd);
1581
1582			vdev_add_child(rvd, mtvd);
1583			vdev_add_child(mrvd, tvd);
1584
1585			spa_config_exit(spa, SCL_ALL, FTAG);
1586			vdev_load(mtvd);
1587			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1588
1589			vdev_reopen(rvd);
1590		} else if (mtvd->vdev_islog) {
1591			/*
1592			 * Load the slog device's state from the MOS config
1593			 * since it's possible that the label does not
1594			 * contain the most up-to-date information.
1595			 */
1596			vdev_load_log_state(tvd, mtvd);
1597			vdev_reopen(tvd);
1598		}
1599	}
1600	vdev_free(mrvd);
1601	spa_config_exit(spa, SCL_ALL, FTAG);
1602
1603	/*
1604	 * Ensure we were able to validate the config.
1605	 */
1606	return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1607}
1608
1609/*
1610 * Check for missing log devices
1611 */
1612static int
1613spa_check_logs(spa_t *spa)
1614{
1615	switch (spa->spa_log_state) {
1616	case SPA_LOG_MISSING:
1617		/* need to recheck in case slog has been restored */
1618	case SPA_LOG_UNKNOWN:
1619		if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1620		    DS_FIND_CHILDREN)) {
1621			spa_set_log_state(spa, SPA_LOG_MISSING);
1622			return (1);
1623		}
1624		break;
1625	}
1626	return (0);
1627}
1628
1629static boolean_t
1630spa_passivate_log(spa_t *spa)
1631{
1632	vdev_t *rvd = spa->spa_root_vdev;
1633	boolean_t slog_found = B_FALSE;
1634
1635	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1636
1637	if (!spa_has_slogs(spa))
1638		return (B_FALSE);
1639
1640	for (int c = 0; c < rvd->vdev_children; c++) {
1641		vdev_t *tvd = rvd->vdev_child[c];
1642		metaslab_group_t *mg = tvd->vdev_mg;
1643
1644		if (tvd->vdev_islog) {
1645			metaslab_group_passivate(mg);
1646			slog_found = B_TRUE;
1647		}
1648	}
1649
1650	return (slog_found);
1651}
1652
1653static void
1654spa_activate_log(spa_t *spa)
1655{
1656	vdev_t *rvd = spa->spa_root_vdev;
1657
1658	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1659
1660	for (int c = 0; c < rvd->vdev_children; c++) {
1661		vdev_t *tvd = rvd->vdev_child[c];
1662		metaslab_group_t *mg = tvd->vdev_mg;
1663
1664		if (tvd->vdev_islog)
1665			metaslab_group_activate(mg);
1666	}
1667}
1668
1669int
1670spa_offline_log(spa_t *spa)
1671{
1672	int error = 0;
1673
1674	if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1675	    NULL, DS_FIND_CHILDREN)) == 0) {
1676
1677		/*
1678		 * We successfully offlined the log device, sync out the
1679		 * current txg so that the "stubby" block can be removed
1680		 * by zil_sync().
1681		 */
1682		txg_wait_synced(spa->spa_dsl_pool, 0);
1683	}
1684	return (error);
1685}
1686
1687static void
1688spa_aux_check_removed(spa_aux_vdev_t *sav)
1689{
1690	for (int i = 0; i < sav->sav_count; i++)
1691		spa_check_removed(sav->sav_vdevs[i]);
1692}
1693
1694void
1695spa_claim_notify(zio_t *zio)
1696{
1697	spa_t *spa = zio->io_spa;
1698
1699	if (zio->io_error)
1700		return;
1701
1702	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1703	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1704		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1705	mutex_exit(&spa->spa_props_lock);
1706}
1707
1708typedef struct spa_load_error {
1709	uint64_t	sle_meta_count;
1710	uint64_t	sle_data_count;
1711} spa_load_error_t;
1712
1713static void
1714spa_load_verify_done(zio_t *zio)
1715{
1716	blkptr_t *bp = zio->io_bp;
1717	spa_load_error_t *sle = zio->io_private;
1718	dmu_object_type_t type = BP_GET_TYPE(bp);
1719	int error = zio->io_error;
1720
1721	if (error) {
1722		if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1723		    type != DMU_OT_INTENT_LOG)
1724			atomic_add_64(&sle->sle_meta_count, 1);
1725		else
1726			atomic_add_64(&sle->sle_data_count, 1);
1727	}
1728	zio_data_buf_free(zio->io_data, zio->io_size);
1729}
1730
1731/*ARGSUSED*/
1732static int
1733spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1734    arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1735{
1736	if (bp != NULL) {
1737		zio_t *rio = arg;
1738		size_t size = BP_GET_PSIZE(bp);
1739		void *data = zio_data_buf_alloc(size);
1740
1741		zio_nowait(zio_read(rio, spa, bp, data, size,
1742		    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1743		    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1744		    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1745	}
1746	return (0);
1747}
1748
1749static int
1750spa_load_verify(spa_t *spa)
1751{
1752	zio_t *rio;
1753	spa_load_error_t sle = { 0 };
1754	zpool_rewind_policy_t policy;
1755	boolean_t verify_ok = B_FALSE;
1756	int error;
1757
1758	zpool_get_rewind_policy(spa->spa_config, &policy);
1759
1760	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1761		return (0);
1762
1763	rio = zio_root(spa, NULL, &sle,
1764	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1765
1766	error = traverse_pool(spa, spa->spa_verify_min_txg,
1767	    TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1768
1769	(void) zio_wait(rio);
1770
1771	spa->spa_load_meta_errors = sle.sle_meta_count;
1772	spa->spa_load_data_errors = sle.sle_data_count;
1773
1774	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1775	    sle.sle_data_count <= policy.zrp_maxdata) {
1776		int64_t loss = 0;
1777
1778		verify_ok = B_TRUE;
1779		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1780		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1781
1782		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1783		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1784		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1785		VERIFY(nvlist_add_int64(spa->spa_load_info,
1786		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1787		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1788		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1789	} else {
1790		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1791	}
1792
1793	if (error) {
1794		if (error != ENXIO && error != EIO)
1795			error = EIO;
1796		return (error);
1797	}
1798
1799	return (verify_ok ? 0 : EIO);
1800}
1801
1802/*
1803 * Find a value in the pool props object.
1804 */
1805static void
1806spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1807{
1808	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1809	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1810}
1811
1812/*
1813 * Find a value in the pool directory object.
1814 */
1815static int
1816spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1817{
1818	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1819	    name, sizeof (uint64_t), 1, val));
1820}
1821
1822static int
1823spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1824{
1825	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1826	return (err);
1827}
1828
1829/*
1830 * Fix up config after a partly-completed split.  This is done with the
1831 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1832 * pool have that entry in their config, but only the splitting one contains
1833 * a list of all the guids of the vdevs that are being split off.
1834 *
1835 * This function determines what to do with that list: either rejoin
1836 * all the disks to the pool, or complete the splitting process.  To attempt
1837 * the rejoin, each disk that is offlined is marked online again, and
1838 * we do a reopen() call.  If the vdev label for every disk that was
1839 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1840 * then we call vdev_split() on each disk, and complete the split.
1841 *
1842 * Otherwise we leave the config alone, with all the vdevs in place in
1843 * the original pool.
1844 */
1845static void
1846spa_try_repair(spa_t *spa, nvlist_t *config)
1847{
1848	uint_t extracted;
1849	uint64_t *glist;
1850	uint_t i, gcount;
1851	nvlist_t *nvl;
1852	vdev_t **vd;
1853	boolean_t attempt_reopen;
1854
1855	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1856		return;
1857
1858	/* check that the config is complete */
1859	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1860	    &glist, &gcount) != 0)
1861		return;
1862
1863	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1864
1865	/* attempt to online all the vdevs & validate */
1866	attempt_reopen = B_TRUE;
1867	for (i = 0; i < gcount; i++) {
1868		if (glist[i] == 0)	/* vdev is hole */
1869			continue;
1870
1871		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1872		if (vd[i] == NULL) {
1873			/*
1874			 * Don't bother attempting to reopen the disks;
1875			 * just do the split.
1876			 */
1877			attempt_reopen = B_FALSE;
1878		} else {
1879			/* attempt to re-online it */
1880			vd[i]->vdev_offline = B_FALSE;
1881		}
1882	}
1883
1884	if (attempt_reopen) {
1885		vdev_reopen(spa->spa_root_vdev);
1886
1887		/* check each device to see what state it's in */
1888		for (extracted = 0, i = 0; i < gcount; i++) {
1889			if (vd[i] != NULL &&
1890			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1891				break;
1892			++extracted;
1893		}
1894	}
1895
1896	/*
1897	 * If every disk has been moved to the new pool, or if we never
1898	 * even attempted to look at them, then we split them off for
1899	 * good.
1900	 */
1901	if (!attempt_reopen || gcount == extracted) {
1902		for (i = 0; i < gcount; i++)
1903			if (vd[i] != NULL)
1904				vdev_split(vd[i]);
1905		vdev_reopen(spa->spa_root_vdev);
1906	}
1907
1908	kmem_free(vd, gcount * sizeof (vdev_t *));
1909}
1910
1911static int
1912spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1913    boolean_t mosconfig)
1914{
1915	nvlist_t *config = spa->spa_config;
1916	char *ereport = FM_EREPORT_ZFS_POOL;
1917	char *comment;
1918	int error;
1919	uint64_t pool_guid;
1920	nvlist_t *nvl;
1921
1922	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1923		return (EINVAL);
1924
1925	ASSERT(spa->spa_comment == NULL);
1926	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
1927		spa->spa_comment = spa_strdup(comment);
1928
1929	/*
1930	 * Versioning wasn't explicitly added to the label until later, so if
1931	 * it's not present treat it as the initial version.
1932	 */
1933	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1934	    &spa->spa_ubsync.ub_version) != 0)
1935		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1936
1937	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1938	    &spa->spa_config_txg);
1939
1940	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1941	    spa_guid_exists(pool_guid, 0)) {
1942		error = EEXIST;
1943	} else {
1944		spa->spa_config_guid = pool_guid;
1945
1946		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1947		    &nvl) == 0) {
1948			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1949			    KM_SLEEP) == 0);
1950		}
1951
1952		nvlist_free(spa->spa_load_info);
1953		spa->spa_load_info = fnvlist_alloc();
1954
1955		gethrestime(&spa->spa_loaded_ts);
1956		error = spa_load_impl(spa, pool_guid, config, state, type,
1957		    mosconfig, &ereport);
1958	}
1959
1960	spa->spa_minref = refcount_count(&spa->spa_refcount);
1961	if (error) {
1962		if (error != EEXIST) {
1963			spa->spa_loaded_ts.tv_sec = 0;
1964			spa->spa_loaded_ts.tv_nsec = 0;
1965		}
1966		if (error != EBADF) {
1967			zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1968		}
1969	}
1970	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1971	spa->spa_ena = 0;
1972
1973	return (error);
1974}
1975
1976/*
1977 * Load an existing storage pool, using the pool's builtin spa_config as a
1978 * source of configuration information.
1979 */
1980static int
1981spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1982    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1983    char **ereport)
1984{
1985	int error = 0;
1986	nvlist_t *nvroot = NULL;
1987	nvlist_t *label;
1988	vdev_t *rvd;
1989	uberblock_t *ub = &spa->spa_uberblock;
1990	uint64_t children, config_cache_txg = spa->spa_config_txg;
1991	int orig_mode = spa->spa_mode;
1992	int parse;
1993	uint64_t obj;
1994	boolean_t missing_feat_write = B_FALSE;
1995
1996	/*
1997	 * If this is an untrusted config, access the pool in read-only mode.
1998	 * This prevents things like resilvering recently removed devices.
1999	 */
2000	if (!mosconfig)
2001		spa->spa_mode = FREAD;
2002
2003	ASSERT(MUTEX_HELD(&spa_namespace_lock));
2004
2005	spa->spa_load_state = state;
2006
2007	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2008		return (EINVAL);
2009
2010	parse = (type == SPA_IMPORT_EXISTING ?
2011	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2012
2013	/*
2014	 * Create "The Godfather" zio to hold all async IOs
2015	 */
2016	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2017	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2018
2019	/*
2020	 * Parse the configuration into a vdev tree.  We explicitly set the
2021	 * value that will be returned by spa_version() since parsing the
2022	 * configuration requires knowing the version number.
2023	 */
2024	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2025	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2026	spa_config_exit(spa, SCL_ALL, FTAG);
2027
2028	if (error != 0)
2029		return (error);
2030
2031	ASSERT(spa->spa_root_vdev == rvd);
2032
2033	if (type != SPA_IMPORT_ASSEMBLE) {
2034		ASSERT(spa_guid(spa) == pool_guid);
2035	}
2036
2037	/*
2038	 * Try to open all vdevs, loading each label in the process.
2039	 */
2040	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2041	error = vdev_open(rvd);
2042	spa_config_exit(spa, SCL_ALL, FTAG);
2043	if (error != 0)
2044		return (error);
2045
2046	/*
2047	 * We need to validate the vdev labels against the configuration that
2048	 * we have in hand, which is dependent on the setting of mosconfig. If
2049	 * mosconfig is true then we're validating the vdev labels based on
2050	 * that config.  Otherwise, we're validating against the cached config
2051	 * (zpool.cache) that was read when we loaded the zfs module, and then
2052	 * later we will recursively call spa_load() and validate against
2053	 * the vdev config.
2054	 *
2055	 * If we're assembling a new pool that's been split off from an
2056	 * existing pool, the labels haven't yet been updated so we skip
2057	 * validation for now.
2058	 */
2059	if (type != SPA_IMPORT_ASSEMBLE) {
2060		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2061		error = vdev_validate(rvd, mosconfig);
2062		spa_config_exit(spa, SCL_ALL, FTAG);
2063
2064		if (error != 0)
2065			return (error);
2066
2067		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2068			return (ENXIO);
2069	}
2070
2071	/*
2072	 * Find the best uberblock.
2073	 */
2074	vdev_uberblock_load(rvd, ub, &label);
2075
2076	/*
2077	 * If we weren't able to find a single valid uberblock, return failure.
2078	 */
2079	if (ub->ub_txg == 0) {
2080		nvlist_free(label);
2081		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2082	}
2083
2084	/*
2085	 * If the pool has an unsupported version we can't open it.
2086	 */
2087	if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2088		nvlist_free(label);
2089		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2090	}
2091
2092	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2093		nvlist_t *features;
2094
2095		/*
2096		 * If we weren't able to find what's necessary for reading the
2097		 * MOS in the label, return failure.
2098		 */
2099		if (label == NULL || nvlist_lookup_nvlist(label,
2100		    ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2101			nvlist_free(label);
2102			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2103			    ENXIO));
2104		}
2105
2106		/*
2107		 * Update our in-core representation with the definitive values
2108		 * from the label.
2109		 */
2110		nvlist_free(spa->spa_label_features);
2111		VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2112	}
2113
2114	nvlist_free(label);
2115
2116	/*
2117	 * Look through entries in the label nvlist's features_for_read. If
2118	 * there is a feature listed there which we don't understand then we
2119	 * cannot open a pool.
2120	 */
2121	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2122		nvlist_t *unsup_feat;
2123
2124		VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2125		    0);
2126
2127		for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2128		    NULL); nvp != NULL;
2129		    nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2130			if (!zfeature_is_supported(nvpair_name(nvp))) {
2131				VERIFY(nvlist_add_string(unsup_feat,
2132				    nvpair_name(nvp), "") == 0);
2133			}
2134		}
2135
2136		if (!nvlist_empty(unsup_feat)) {
2137			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2138			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2139			nvlist_free(unsup_feat);
2140			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2141			    ENOTSUP));
2142		}
2143
2144		nvlist_free(unsup_feat);
2145	}
2146
2147	/*
2148	 * If the vdev guid sum doesn't match the uberblock, we have an
2149	 * incomplete configuration.  We first check to see if the pool
2150	 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2151	 * If it is, defer the vdev_guid_sum check till later so we
2152	 * can handle missing vdevs.
2153	 */
2154	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2155	    &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2156	    rvd->vdev_guid_sum != ub->ub_guid_sum)
2157		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2158
2159	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2160		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2161		spa_try_repair(spa, config);
2162		spa_config_exit(spa, SCL_ALL, FTAG);
2163		nvlist_free(spa->spa_config_splitting);
2164		spa->spa_config_splitting = NULL;
2165	}
2166
2167	/*
2168	 * Initialize internal SPA structures.
2169	 */
2170	spa->spa_state = POOL_STATE_ACTIVE;
2171	spa->spa_ubsync = spa->spa_uberblock;
2172	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2173	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2174	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2175	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2176	spa->spa_claim_max_txg = spa->spa_first_txg;
2177	spa->spa_prev_software_version = ub->ub_software_version;
2178
2179	error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2180	if (error)
2181		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2182	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2183
2184	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2185		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2186
2187	if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2188		boolean_t missing_feat_read = B_FALSE;
2189		nvlist_t *unsup_feat, *enabled_feat;
2190
2191		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2192		    &spa->spa_feat_for_read_obj) != 0) {
2193			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2194		}
2195
2196		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2197		    &spa->spa_feat_for_write_obj) != 0) {
2198			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2199		}
2200
2201		if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2202		    &spa->spa_feat_desc_obj) != 0) {
2203			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2204		}
2205
2206		enabled_feat = fnvlist_alloc();
2207		unsup_feat = fnvlist_alloc();
2208
2209		if (!feature_is_supported(spa->spa_meta_objset,
2210		    spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2211		    unsup_feat, enabled_feat))
2212			missing_feat_read = B_TRUE;
2213
2214		if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2215			if (!feature_is_supported(spa->spa_meta_objset,
2216			    spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2217			    unsup_feat, enabled_feat)) {
2218				missing_feat_write = B_TRUE;
2219			}
2220		}
2221
2222		fnvlist_add_nvlist(spa->spa_load_info,
2223		    ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2224
2225		if (!nvlist_empty(unsup_feat)) {
2226			fnvlist_add_nvlist(spa->spa_load_info,
2227			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2228		}
2229
2230		fnvlist_free(enabled_feat);
2231		fnvlist_free(unsup_feat);
2232
2233		if (!missing_feat_read) {
2234			fnvlist_add_boolean(spa->spa_load_info,
2235			    ZPOOL_CONFIG_CAN_RDONLY);
2236		}
2237
2238		/*
2239		 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2240		 * twofold: to determine whether the pool is available for
2241		 * import in read-write mode and (if it is not) whether the
2242		 * pool is available for import in read-only mode. If the pool
2243		 * is available for import in read-write mode, it is displayed
2244		 * as available in userland; if it is not available for import
2245		 * in read-only mode, it is displayed as unavailable in
2246		 * userland. If the pool is available for import in read-only
2247		 * mode but not read-write mode, it is displayed as unavailable
2248		 * in userland with a special note that the pool is actually
2249		 * available for open in read-only mode.
2250		 *
2251		 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2252		 * missing a feature for write, we must first determine whether
2253		 * the pool can be opened read-only before returning to
2254		 * userland in order to know whether to display the
2255		 * abovementioned note.
2256		 */
2257		if (missing_feat_read || (missing_feat_write &&
2258		    spa_writeable(spa))) {
2259			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2260			    ENOTSUP));
2261		}
2262	}
2263
2264	spa->spa_is_initializing = B_TRUE;
2265	error = dsl_pool_open(spa->spa_dsl_pool);
2266	spa->spa_is_initializing = B_FALSE;
2267	if (error != 0)
2268		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2269
2270	if (!mosconfig) {
2271		uint64_t hostid;
2272		nvlist_t *policy = NULL, *nvconfig;
2273
2274		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2275			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2276
2277		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2278		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2279			char *hostname;
2280			unsigned long myhostid = 0;
2281
2282			VERIFY(nvlist_lookup_string(nvconfig,
2283			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2284
2285#ifdef	_KERNEL
2286			myhostid = zone_get_hostid(NULL);
2287#else	/* _KERNEL */
2288			/*
2289			 * We're emulating the system's hostid in userland, so
2290			 * we can't use zone_get_hostid().
2291			 */
2292			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2293#endif	/* _KERNEL */
2294			if (hostid != 0 && myhostid != 0 &&
2295			    hostid != myhostid) {
2296				nvlist_free(nvconfig);
2297				cmn_err(CE_WARN, "pool '%s' could not be "
2298				    "loaded as it was last accessed by "
2299				    "another system (host: %s hostid: 0x%lx). "
2300				    "See: http://illumos.org/msg/ZFS-8000-EY",
2301				    spa_name(spa), hostname,
2302				    (unsigned long)hostid);
2303				return (EBADF);
2304			}
2305		}
2306		if (nvlist_lookup_nvlist(spa->spa_config,
2307		    ZPOOL_REWIND_POLICY, &policy) == 0)
2308			VERIFY(nvlist_add_nvlist(nvconfig,
2309			    ZPOOL_REWIND_POLICY, policy) == 0);
2310
2311		spa_config_set(spa, nvconfig);
2312		spa_unload(spa);
2313		spa_deactivate(spa);
2314		spa_activate(spa, orig_mode);
2315
2316		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2317	}
2318
2319	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2320		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2321	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2322	if (error != 0)
2323		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2324
2325	/*
2326	 * Load the bit that tells us to use the new accounting function
2327	 * (raid-z deflation).  If we have an older pool, this will not
2328	 * be present.
2329	 */
2330	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2331	if (error != 0 && error != ENOENT)
2332		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2333
2334	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2335	    &spa->spa_creation_version);
2336	if (error != 0 && error != ENOENT)
2337		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2338
2339	/*
2340	 * Load the persistent error log.  If we have an older pool, this will
2341	 * not be present.
2342	 */
2343	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2344	if (error != 0 && error != ENOENT)
2345		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2346
2347	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2348	    &spa->spa_errlog_scrub);
2349	if (error != 0 && error != ENOENT)
2350		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2351
2352	/*
2353	 * Load the history object.  If we have an older pool, this
2354	 * will not be present.
2355	 */
2356	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2357	if (error != 0 && error != ENOENT)
2358		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2359
2360	/*
2361	 * If we're assembling the pool from the split-off vdevs of
2362	 * an existing pool, we don't want to attach the spares & cache
2363	 * devices.
2364	 */
2365
2366	/*
2367	 * Load any hot spares for this pool.
2368	 */
2369	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2370	if (error != 0 && error != ENOENT)
2371		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2372	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2373		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2374		if (load_nvlist(spa, spa->spa_spares.sav_object,
2375		    &spa->spa_spares.sav_config) != 0)
2376			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2377
2378		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2379		spa_load_spares(spa);
2380		spa_config_exit(spa, SCL_ALL, FTAG);
2381	} else if (error == 0) {
2382		spa->spa_spares.sav_sync = B_TRUE;
2383	}
2384
2385	/*
2386	 * Load any level 2 ARC devices for this pool.
2387	 */
2388	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2389	    &spa->spa_l2cache.sav_object);
2390	if (error != 0 && error != ENOENT)
2391		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2392	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2393		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2394		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2395		    &spa->spa_l2cache.sav_config) != 0)
2396			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2397
2398		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2399		spa_load_l2cache(spa);
2400		spa_config_exit(spa, SCL_ALL, FTAG);
2401	} else if (error == 0) {
2402		spa->spa_l2cache.sav_sync = B_TRUE;
2403	}
2404
2405	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2406
2407	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2408	if (error && error != ENOENT)
2409		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2410
2411	if (error == 0) {
2412		uint64_t autoreplace;
2413
2414		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2415		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2416		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2417		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2418		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2419		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2420		    &spa->spa_dedup_ditto);
2421
2422		spa->spa_autoreplace = (autoreplace != 0);
2423	}
2424
2425	/*
2426	 * If the 'autoreplace' property is set, then post a resource notifying
2427	 * the ZFS DE that it should not issue any faults for unopenable
2428	 * devices.  We also iterate over the vdevs, and post a sysevent for any
2429	 * unopenable vdevs so that the normal autoreplace handler can take
2430	 * over.
2431	 */
2432	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2433		spa_check_removed(spa->spa_root_vdev);
2434		/*
2435		 * For the import case, this is done in spa_import(), because
2436		 * at this point we're using the spare definitions from
2437		 * the MOS config, not necessarily from the userland config.
2438		 */
2439		if (state != SPA_LOAD_IMPORT) {
2440			spa_aux_check_removed(&spa->spa_spares);
2441			spa_aux_check_removed(&spa->spa_l2cache);
2442		}
2443	}
2444
2445	/*
2446	 * Load the vdev state for all toplevel vdevs.
2447	 */
2448	vdev_load(rvd);
2449
2450	/*
2451	 * Propagate the leaf DTLs we just loaded all the way up the tree.
2452	 */
2453	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2454	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2455	spa_config_exit(spa, SCL_ALL, FTAG);
2456
2457	/*
2458	 * Load the DDTs (dedup tables).
2459	 */
2460	error = ddt_load(spa);
2461	if (error != 0)
2462		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2463
2464	spa_update_dspace(spa);
2465
2466	/*
2467	 * Validate the config, using the MOS config to fill in any
2468	 * information which might be missing.  If we fail to validate
2469	 * the config then declare the pool unfit for use. If we're
2470	 * assembling a pool from a split, the log is not transferred
2471	 * over.
2472	 */
2473	if (type != SPA_IMPORT_ASSEMBLE) {
2474		nvlist_t *nvconfig;
2475
2476		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2477			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2478
2479		if (!spa_config_valid(spa, nvconfig)) {
2480			nvlist_free(nvconfig);
2481			return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2482			    ENXIO));
2483		}
2484		nvlist_free(nvconfig);
2485
2486		/*
2487		 * Now that we've validated the config, check the state of the
2488		 * root vdev.  If it can't be opened, it indicates one or
2489		 * more toplevel vdevs are faulted.
2490		 */
2491		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2492			return (ENXIO);
2493
2494		if (spa_check_logs(spa)) {
2495			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2496			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2497		}
2498	}
2499
2500	if (missing_feat_write) {
2501		ASSERT(state == SPA_LOAD_TRYIMPORT);
2502
2503		/*
2504		 * At this point, we know that we can open the pool in
2505		 * read-only mode but not read-write mode. We now have enough
2506		 * information and can return to userland.
2507		 */
2508		return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2509	}
2510
2511	/*
2512	 * We've successfully opened the pool, verify that we're ready
2513	 * to start pushing transactions.
2514	 */
2515	if (state != SPA_LOAD_TRYIMPORT) {
2516		if (error = spa_load_verify(spa))
2517			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2518			    error));
2519	}
2520
2521	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2522	    spa->spa_load_max_txg == UINT64_MAX)) {
2523		dmu_tx_t *tx;
2524		int need_update = B_FALSE;
2525
2526		ASSERT(state != SPA_LOAD_TRYIMPORT);
2527
2528		/*
2529		 * Claim log blocks that haven't been committed yet.
2530		 * This must all happen in a single txg.
2531		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2532		 * invoked from zil_claim_log_block()'s i/o done callback.
2533		 * Price of rollback is that we abandon the log.
2534		 */
2535		spa->spa_claiming = B_TRUE;
2536
2537		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2538		    spa_first_txg(spa));
2539		(void) dmu_objset_find(spa_name(spa),
2540		    zil_claim, tx, DS_FIND_CHILDREN);
2541		dmu_tx_commit(tx);
2542
2543		spa->spa_claiming = B_FALSE;
2544
2545		spa_set_log_state(spa, SPA_LOG_GOOD);
2546		spa->spa_sync_on = B_TRUE;
2547		txg_sync_start(spa->spa_dsl_pool);
2548
2549		/*
2550		 * Wait for all claims to sync.  We sync up to the highest
2551		 * claimed log block birth time so that claimed log blocks
2552		 * don't appear to be from the future.  spa_claim_max_txg
2553		 * will have been set for us by either zil_check_log_chain()
2554		 * (invoked from spa_check_logs()) or zil_claim() above.
2555		 */
2556		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2557
2558		/*
2559		 * If the config cache is stale, or we have uninitialized
2560		 * metaslabs (see spa_vdev_add()), then update the config.
2561		 *
2562		 * If this is a verbatim import, trust the current
2563		 * in-core spa_config and update the disk labels.
2564		 */
2565		if (config_cache_txg != spa->spa_config_txg ||
2566		    state == SPA_LOAD_IMPORT ||
2567		    state == SPA_LOAD_RECOVER ||
2568		    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2569			need_update = B_TRUE;
2570
2571		for (int c = 0; c < rvd->vdev_children; c++)
2572			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2573				need_update = B_TRUE;
2574
2575		/*
2576		 * Update the config cache asychronously in case we're the
2577		 * root pool, in which case the config cache isn't writable yet.
2578		 */
2579		if (need_update)
2580			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2581
2582		/*
2583		 * Check all DTLs to see if anything needs resilvering.
2584		 */
2585		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2586		    vdev_resilver_needed(rvd, NULL, NULL))
2587			spa_async_request(spa, SPA_ASYNC_RESILVER);
2588
2589		/*
2590		 * Log the fact that we booted up (so that we can detect if
2591		 * we rebooted in the middle of an operation).
2592		 */
2593		spa_history_log_version(spa, "open");
2594
2595		/*
2596		 * Delete any inconsistent datasets.
2597		 */
2598		(void) dmu_objset_find(spa_name(spa),
2599		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2600
2601		/*
2602		 * Clean up any stale temporary dataset userrefs.
2603		 */
2604		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2605	}
2606
2607	return (0);
2608}
2609
2610static int
2611spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2612{
2613	int mode = spa->spa_mode;
2614
2615	spa_unload(spa);
2616	spa_deactivate(spa);
2617
2618	spa->spa_load_max_txg--;
2619
2620	spa_activate(spa, mode);
2621	spa_async_suspend(spa);
2622
2623	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2624}
2625
2626/*
2627 * If spa_load() fails this function will try loading prior txg's. If
2628 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2629 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2630 * function will not rewind the pool and will return the same error as
2631 * spa_load().
2632 */
2633static int
2634spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2635    uint64_t max_request, int rewind_flags)
2636{
2637	nvlist_t *loadinfo = NULL;
2638	nvlist_t *config = NULL;
2639	int load_error, rewind_error;
2640	uint64_t safe_rewind_txg;
2641	uint64_t min_txg;
2642
2643	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2644		spa->spa_load_max_txg = spa->spa_load_txg;
2645		spa_set_log_state(spa, SPA_LOG_CLEAR);
2646	} else {
2647		spa->spa_load_max_txg = max_request;
2648	}
2649
2650	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2651	    mosconfig);
2652	if (load_error == 0)
2653		return (0);
2654
2655	if (spa->spa_root_vdev != NULL)
2656		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2657
2658	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2659	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2660
2661	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2662		nvlist_free(config);
2663		return (load_error);
2664	}
2665
2666	if (state == SPA_LOAD_RECOVER) {
2667		/* Price of rolling back is discarding txgs, including log */
2668		spa_set_log_state(spa, SPA_LOG_CLEAR);
2669	} else {
2670		/*
2671		 * If we aren't rolling back save the load info from our first
2672		 * import attempt so that we can restore it after attempting
2673		 * to rewind.
2674		 */
2675		loadinfo = spa->spa_load_info;
2676		spa->spa_load_info = fnvlist_alloc();
2677	}
2678
2679	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2680	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2681	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2682	    TXG_INITIAL : safe_rewind_txg;
2683
2684	/*
2685	 * Continue as long as we're finding errors, we're still within
2686	 * the acceptable rewind range, and we're still finding uberblocks
2687	 */
2688	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2689	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2690		if (spa->spa_load_max_txg < safe_rewind_txg)
2691			spa->spa_extreme_rewind = B_TRUE;
2692		rewind_error = spa_load_retry(spa, state, mosconfig);
2693	}
2694
2695	spa->spa_extreme_rewind = B_FALSE;
2696	spa->spa_load_max_txg = UINT64_MAX;
2697
2698	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2699		spa_config_set(spa, config);
2700
2701	if (state == SPA_LOAD_RECOVER) {
2702		ASSERT3P(loadinfo, ==, NULL);
2703		return (rewind_error);
2704	} else {
2705		/* Store the rewind info as part of the initial load info */
2706		fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2707		    spa->spa_load_info);
2708
2709		/* Restore the initial load info */
2710		fnvlist_free(spa->spa_load_info);
2711		spa->spa_load_info = loadinfo;
2712
2713		return (load_error);
2714	}
2715}
2716
2717/*
2718 * Pool Open/Import
2719 *
2720 * The import case is identical to an open except that the configuration is sent
2721 * down from userland, instead of grabbed from the configuration cache.  For the
2722 * case of an open, the pool configuration will exist in the
2723 * POOL_STATE_UNINITIALIZED state.
2724 *
2725 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2726 * the same time open the pool, without having to keep around the spa_t in some
2727 * ambiguous state.
2728 */
2729static int
2730spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2731    nvlist_t **config)
2732{
2733	spa_t *spa;
2734	spa_load_state_t state = SPA_LOAD_OPEN;
2735	int error;
2736	int locked = B_FALSE;
2737
2738	*spapp = NULL;
2739
2740	/*
2741	 * As disgusting as this is, we need to support recursive calls to this
2742	 * function because dsl_dir_open() is called during spa_load(), and ends
2743	 * up calling spa_open() again.  The real fix is to figure out how to
2744	 * avoid dsl_dir_open() calling this in the first place.
2745	 */
2746	if (mutex_owner(&spa_namespace_lock) != curthread) {
2747		mutex_enter(&spa_namespace_lock);
2748		locked = B_TRUE;
2749	}
2750
2751	if ((spa = spa_lookup(pool)) == NULL) {
2752		if (locked)
2753			mutex_exit(&spa_namespace_lock);
2754		return (ENOENT);
2755	}
2756
2757	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2758		zpool_rewind_policy_t policy;
2759
2760		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2761		    &policy);
2762		if (policy.zrp_request & ZPOOL_DO_REWIND)
2763			state = SPA_LOAD_RECOVER;
2764
2765		spa_activate(spa, spa_mode_global);
2766
2767		if (state != SPA_LOAD_RECOVER)
2768			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2769
2770		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2771		    policy.zrp_request);
2772
2773		if (error == EBADF) {
2774			/*
2775			 * If vdev_validate() returns failure (indicated by
2776			 * EBADF), it indicates that one of the vdevs indicates
2777			 * that the pool has been exported or destroyed.  If
2778			 * this is the case, the config cache is out of sync and
2779			 * we should remove the pool from the namespace.
2780			 */
2781			spa_unload(spa);
2782			spa_deactivate(spa);
2783			spa_config_sync(spa, B_TRUE, B_TRUE);
2784			spa_remove(spa);
2785			if (locked)
2786				mutex_exit(&spa_namespace_lock);
2787			return (ENOENT);
2788		}
2789
2790		if (error) {
2791			/*
2792			 * We can't open the pool, but we still have useful
2793			 * information: the state of each vdev after the
2794			 * attempted vdev_open().  Return this to the user.
2795			 */
2796			if (config != NULL && spa->spa_config) {
2797				VERIFY(nvlist_dup(spa->spa_config, config,
2798				    KM_SLEEP) == 0);
2799				VERIFY(nvlist_add_nvlist(*config,
2800				    ZPOOL_CONFIG_LOAD_INFO,
2801				    spa->spa_load_info) == 0);
2802			}
2803			spa_unload(spa);
2804			spa_deactivate(spa);
2805			spa->spa_last_open_failed = error;
2806			if (locked)
2807				mutex_exit(&spa_namespace_lock);
2808			*spapp = NULL;
2809			return (error);
2810		}
2811	}
2812
2813	spa_open_ref(spa, tag);
2814
2815	if (config != NULL)
2816		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2817
2818	/*
2819	 * If we've recovered the pool, pass back any information we
2820	 * gathered while doing the load.
2821	 */
2822	if (state == SPA_LOAD_RECOVER) {
2823		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2824		    spa->spa_load_info) == 0);
2825	}
2826
2827	if (locked) {
2828		spa->spa_last_open_failed = 0;
2829		spa->spa_last_ubsync_txg = 0;
2830		spa->spa_load_txg = 0;
2831		mutex_exit(&spa_namespace_lock);
2832	}
2833
2834	*spapp = spa;
2835
2836	return (0);
2837}
2838
2839int
2840spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2841    nvlist_t **config)
2842{
2843	return (spa_open_common(name, spapp, tag, policy, config));
2844}
2845
2846int
2847spa_open(const char *name, spa_t **spapp, void *tag)
2848{
2849	return (spa_open_common(name, spapp, tag, NULL, NULL));
2850}
2851
2852/*
2853 * Lookup the given spa_t, incrementing the inject count in the process,
2854 * preventing it from being exported or destroyed.
2855 */
2856spa_t *
2857spa_inject_addref(char *name)
2858{
2859	spa_t *spa;
2860
2861	mutex_enter(&spa_namespace_lock);
2862	if ((spa = spa_lookup(name)) == NULL) {
2863		mutex_exit(&spa_namespace_lock);
2864		return (NULL);
2865	}
2866	spa->spa_inject_ref++;
2867	mutex_exit(&spa_namespace_lock);
2868
2869	return (spa);
2870}
2871
2872void
2873spa_inject_delref(spa_t *spa)
2874{
2875	mutex_enter(&spa_namespace_lock);
2876	spa->spa_inject_ref--;
2877	mutex_exit(&spa_namespace_lock);
2878}
2879
2880/*
2881 * Add spares device information to the nvlist.
2882 */
2883static void
2884spa_add_spares(spa_t *spa, nvlist_t *config)
2885{
2886	nvlist_t **spares;
2887	uint_t i, nspares;
2888	nvlist_t *nvroot;
2889	uint64_t guid;
2890	vdev_stat_t *vs;
2891	uint_t vsc;
2892	uint64_t pool;
2893
2894	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2895
2896	if (spa->spa_spares.sav_count == 0)
2897		return;
2898
2899	VERIFY(nvlist_lookup_nvlist(config,
2900	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2901	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2902	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2903	if (nspares != 0) {
2904		VERIFY(nvlist_add_nvlist_array(nvroot,
2905		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2906		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2907		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2908
2909		/*
2910		 * Go through and find any spares which have since been
2911		 * repurposed as an active spare.  If this is the case, update
2912		 * their status appropriately.
2913		 */
2914		for (i = 0; i < nspares; i++) {
2915			VERIFY(nvlist_lookup_uint64(spares[i],
2916			    ZPOOL_CONFIG_GUID, &guid) == 0);
2917			if (spa_spare_exists(guid, &pool, NULL) &&
2918			    pool != 0ULL) {
2919				VERIFY(nvlist_lookup_uint64_array(
2920				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
2921				    (uint64_t **)&vs, &vsc) == 0);
2922				vs->vs_state = VDEV_STATE_CANT_OPEN;
2923				vs->vs_aux = VDEV_AUX_SPARED;
2924			}
2925		}
2926	}
2927}
2928
2929/*
2930 * Add l2cache device information to the nvlist, including vdev stats.
2931 */
2932static void
2933spa_add_l2cache(spa_t *spa, nvlist_t *config)
2934{
2935	nvlist_t **l2cache;
2936	uint_t i, j, nl2cache;
2937	nvlist_t *nvroot;
2938	uint64_t guid;
2939	vdev_t *vd;
2940	vdev_stat_t *vs;
2941	uint_t vsc;
2942
2943	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2944
2945	if (spa->spa_l2cache.sav_count == 0)
2946		return;
2947
2948	VERIFY(nvlist_lookup_nvlist(config,
2949	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2950	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2951	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2952	if (nl2cache != 0) {
2953		VERIFY(nvlist_add_nvlist_array(nvroot,
2954		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2955		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2956		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2957
2958		/*
2959		 * Update level 2 cache device stats.
2960		 */
2961
2962		for (i = 0; i < nl2cache; i++) {
2963			VERIFY(nvlist_lookup_uint64(l2cache[i],
2964			    ZPOOL_CONFIG_GUID, &guid) == 0);
2965
2966			vd = NULL;
2967			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2968				if (guid ==
2969				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2970					vd = spa->spa_l2cache.sav_vdevs[j];
2971					break;
2972				}
2973			}
2974			ASSERT(vd != NULL);
2975
2976			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2977			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2978			    == 0);
2979			vdev_get_stats(vd, vs);
2980		}
2981	}
2982}
2983
2984static void
2985spa_add_feature_stats(spa_t *spa, nvlist_t *config)
2986{
2987	nvlist_t *features;
2988	zap_cursor_t zc;
2989	zap_attribute_t za;
2990
2991	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2992	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2993
2994	if (spa->spa_feat_for_read_obj != 0) {
2995		for (zap_cursor_init(&zc, spa->spa_meta_objset,
2996		    spa->spa_feat_for_read_obj);
2997		    zap_cursor_retrieve(&zc, &za) == 0;
2998		    zap_cursor_advance(&zc)) {
2999			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3000			    za.za_num_integers == 1);
3001			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3002			    za.za_first_integer));
3003		}
3004		zap_cursor_fini(&zc);
3005	}
3006
3007	if (spa->spa_feat_for_write_obj != 0) {
3008		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3009		    spa->spa_feat_for_write_obj);
3010		    zap_cursor_retrieve(&zc, &za) == 0;
3011		    zap_cursor_advance(&zc)) {
3012			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3013			    za.za_num_integers == 1);
3014			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3015			    za.za_first_integer));
3016		}
3017		zap_cursor_fini(&zc);
3018	}
3019
3020	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3021	    features) == 0);
3022	nvlist_free(features);
3023}
3024
3025int
3026spa_get_stats(const char *name, nvlist_t **config,
3027    char *altroot, size_t buflen)
3028{
3029	int error;
3030	spa_t *spa;
3031
3032	*config = NULL;
3033	error = spa_open_common(name, &spa, FTAG, NULL, config);
3034
3035	if (spa != NULL) {
3036		/*
3037		 * This still leaves a window of inconsistency where the spares
3038		 * or l2cache devices could change and the config would be
3039		 * self-inconsistent.
3040		 */
3041		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3042
3043		if (*config != NULL) {
3044			uint64_t loadtimes[2];
3045
3046			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3047			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3048			VERIFY(nvlist_add_uint64_array(*config,
3049			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3050
3051			VERIFY(nvlist_add_uint64(*config,
3052			    ZPOOL_CONFIG_ERRCOUNT,
3053			    spa_get_errlog_size(spa)) == 0);
3054
3055			if (spa_suspended(spa))
3056				VERIFY(nvlist_add_uint64(*config,
3057				    ZPOOL_CONFIG_SUSPENDED,
3058				    spa->spa_failmode) == 0);
3059
3060			spa_add_spares(spa, *config);
3061			spa_add_l2cache(spa, *config);
3062			spa_add_feature_stats(spa, *config);
3063		}
3064	}
3065
3066	/*
3067	 * We want to get the alternate root even for faulted pools, so we cheat
3068	 * and call spa_lookup() directly.
3069	 */
3070	if (altroot) {
3071		if (spa == NULL) {
3072			mutex_enter(&spa_namespace_lock);
3073			spa = spa_lookup(name);
3074			if (spa)
3075				spa_altroot(spa, altroot, buflen);
3076			else
3077				altroot[0] = '\0';
3078			spa = NULL;
3079			mutex_exit(&spa_namespace_lock);
3080		} else {
3081			spa_altroot(spa, altroot, buflen);
3082		}
3083	}
3084
3085	if (spa != NULL) {
3086		spa_config_exit(spa, SCL_CONFIG, FTAG);
3087		spa_close(spa, FTAG);
3088	}
3089
3090	return (error);
3091}
3092
3093/*
3094 * Validate that the auxiliary device array is well formed.  We must have an
3095 * array of nvlists, each which describes a valid leaf vdev.  If this is an
3096 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3097 * specified, as long as they are well-formed.
3098 */
3099static int
3100spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3101    spa_aux_vdev_t *sav, const char *config, uint64_t version,
3102    vdev_labeltype_t label)
3103{
3104	nvlist_t **dev;
3105	uint_t i, ndev;
3106	vdev_t *vd;
3107	int error;
3108
3109	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3110
3111	/*
3112	 * It's acceptable to have no devs specified.
3113	 */
3114	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3115		return (0);
3116
3117	if (ndev == 0)
3118		return (EINVAL);
3119
3120	/*
3121	 * Make sure the pool is formatted with a version that supports this
3122	 * device type.
3123	 */
3124	if (spa_version(spa) < version)
3125		return (ENOTSUP);
3126
3127	/*
3128	 * Set the pending device list so we correctly handle device in-use
3129	 * checking.
3130	 */
3131	sav->sav_pending = dev;
3132	sav->sav_npending = ndev;
3133
3134	for (i = 0; i < ndev; i++) {
3135		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3136		    mode)) != 0)
3137			goto out;
3138
3139		if (!vd->vdev_ops->vdev_op_leaf) {
3140			vdev_free(vd);
3141			error = EINVAL;
3142			goto out;
3143		}
3144
3145		/*
3146		 * The L2ARC currently only supports disk devices in
3147		 * kernel context.  For user-level testing, we allow it.
3148		 */
3149#ifdef _KERNEL
3150		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3151		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3152			error = ENOTBLK;
3153			vdev_free(vd);
3154			goto out;
3155		}
3156#endif
3157		vd->vdev_top = vd;
3158
3159		if ((error = vdev_open(vd)) == 0 &&
3160		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
3161			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3162			    vd->vdev_guid) == 0);
3163		}
3164
3165		vdev_free(vd);
3166
3167		if (error &&
3168		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3169			goto out;
3170		else
3171			error = 0;
3172	}
3173
3174out:
3175	sav->sav_pending = NULL;
3176	sav->sav_npending = 0;
3177	return (error);
3178}
3179
3180static int
3181spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3182{
3183	int error;
3184
3185	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3186
3187	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3188	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3189	    VDEV_LABEL_SPARE)) != 0) {
3190		return (error);
3191	}
3192
3193	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3194	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3195	    VDEV_LABEL_L2CACHE));
3196}
3197
3198static void
3199spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3200    const char *config)
3201{
3202	int i;
3203
3204	if (sav->sav_config != NULL) {
3205		nvlist_t **olddevs;
3206		uint_t oldndevs;
3207		nvlist_t **newdevs;
3208
3209		/*
3210		 * Generate new dev list by concatentating with the
3211		 * current dev list.
3212		 */
3213		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3214		    &olddevs, &oldndevs) == 0);
3215
3216		newdevs = kmem_alloc(sizeof (void *) *
3217		    (ndevs + oldndevs), KM_SLEEP);
3218		for (i = 0; i < oldndevs; i++)
3219			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3220			    KM_SLEEP) == 0);
3221		for (i = 0; i < ndevs; i++)
3222			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3223			    KM_SLEEP) == 0);
3224
3225		VERIFY(nvlist_remove(sav->sav_config, config,
3226		    DATA_TYPE_NVLIST_ARRAY) == 0);
3227
3228		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3229		    config, newdevs, ndevs + oldndevs) == 0);
3230		for (i = 0; i < oldndevs + ndevs; i++)
3231			nvlist_free(newdevs[i]);
3232		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3233	} else {
3234		/*
3235		 * Generate a new dev list.
3236		 */
3237		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3238		    KM_SLEEP) == 0);
3239		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3240		    devs, ndevs) == 0);
3241	}
3242}
3243
3244/*
3245 * Stop and drop level 2 ARC devices
3246 */
3247void
3248spa_l2cache_drop(spa_t *spa)
3249{
3250	vdev_t *vd;
3251	int i;
3252	spa_aux_vdev_t *sav = &spa->spa_l2cache;
3253
3254	for (i = 0; i < sav->sav_count; i++) {
3255		uint64_t pool;
3256
3257		vd = sav->sav_vdevs[i];
3258		ASSERT(vd != NULL);
3259
3260		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3261		    pool != 0ULL && l2arc_vdev_present(vd))
3262			l2arc_remove_vdev(vd);
3263	}
3264}
3265
3266/*
3267 * Pool Creation
3268 */
3269int
3270spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3271    nvlist_t *zplprops)
3272{
3273	spa_t *spa;
3274	char *altroot = NULL;
3275	vdev_t *rvd;
3276	dsl_pool_t *dp;
3277	dmu_tx_t *tx;
3278	int error = 0;
3279	uint64_t txg = TXG_INITIAL;
3280	nvlist_t **spares, **l2cache;
3281	uint_t nspares, nl2cache;
3282	uint64_t version, obj;
3283	boolean_t has_features;
3284
3285	/*
3286	 * If this pool already exists, return failure.
3287	 */
3288	mutex_enter(&spa_namespace_lock);
3289	if (spa_lookup(pool) != NULL) {
3290		mutex_exit(&spa_namespace_lock);
3291		return (EEXIST);
3292	}
3293
3294	/*
3295	 * Allocate a new spa_t structure.
3296	 */
3297	(void) nvlist_lookup_string(props,
3298	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3299	spa = spa_add(pool, NULL, altroot);
3300	spa_activate(spa, spa_mode_global);
3301
3302	if (props && (error = spa_prop_validate(spa, props))) {
3303		spa_deactivate(spa);
3304		spa_remove(spa);
3305		mutex_exit(&spa_namespace_lock);
3306		return (error);
3307	}
3308
3309	has_features = B_FALSE;
3310	for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3311	    elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3312		if (zpool_prop_feature(nvpair_name(elem)))
3313			has_features = B_TRUE;
3314	}
3315
3316	if (has_features || nvlist_lookup_uint64(props,
3317	    zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3318		version = SPA_VERSION;
3319	}
3320	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3321
3322	spa->spa_first_txg = txg;
3323	spa->spa_uberblock.ub_txg = txg - 1;
3324	spa->spa_uberblock.ub_version = version;
3325	spa->spa_ubsync = spa->spa_uberblock;
3326
3327	/*
3328	 * Create "The Godfather" zio to hold all async IOs
3329	 */
3330	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3331	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3332
3333	/*
3334	 * Create the root vdev.
3335	 */
3336	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3337
3338	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3339
3340	ASSERT(error != 0 || rvd != NULL);
3341	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3342
3343	if (error == 0 && !zfs_allocatable_devs(nvroot))
3344		error = EINVAL;
3345
3346	if (error == 0 &&
3347	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3348	    (error = spa_validate_aux(spa, nvroot, txg,
3349	    VDEV_ALLOC_ADD)) == 0) {
3350		for (int c = 0; c < rvd->vdev_children; c++) {
3351			vdev_metaslab_set_size(rvd->vdev_child[c]);
3352			vdev_expand(rvd->vdev_child[c], txg);
3353		}
3354	}
3355
3356	spa_config_exit(spa, SCL_ALL, FTAG);
3357
3358	if (error != 0) {
3359		spa_unload(spa);
3360		spa_deactivate(spa);
3361		spa_remove(spa);
3362		mutex_exit(&spa_namespace_lock);
3363		return (error);
3364	}
3365
3366	/*
3367	 * Get the list of spares, if specified.
3368	 */
3369	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3370	    &spares, &nspares) == 0) {
3371		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3372		    KM_SLEEP) == 0);
3373		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3374		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3375		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3376		spa_load_spares(spa);
3377		spa_config_exit(spa, SCL_ALL, FTAG);
3378		spa->spa_spares.sav_sync = B_TRUE;
3379	}
3380
3381	/*
3382	 * Get the list of level 2 cache devices, if specified.
3383	 */
3384	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3385	    &l2cache, &nl2cache) == 0) {
3386		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3387		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3388		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3389		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3390		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3391		spa_load_l2cache(spa);
3392		spa_config_exit(spa, SCL_ALL, FTAG);
3393		spa->spa_l2cache.sav_sync = B_TRUE;
3394	}
3395
3396	spa->spa_is_initializing = B_TRUE;
3397	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3398	spa->spa_meta_objset = dp->dp_meta_objset;
3399	spa->spa_is_initializing = B_FALSE;
3400
3401	/*
3402	 * Create DDTs (dedup tables).
3403	 */
3404	ddt_create(spa);
3405
3406	spa_update_dspace(spa);
3407
3408	tx = dmu_tx_create_assigned(dp, txg);
3409
3410	/*
3411	 * Create the pool config object.
3412	 */
3413	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3414	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3415	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3416
3417	if (zap_add(spa->spa_meta_objset,
3418	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3419	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3420		cmn_err(CE_PANIC, "failed to add pool config");
3421	}
3422
3423	if (spa_version(spa) >= SPA_VERSION_FEATURES)
3424		spa_feature_create_zap_objects(spa, tx);
3425
3426	if (zap_add(spa->spa_meta_objset,
3427	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3428	    sizeof (uint64_t), 1, &version, tx) != 0) {
3429		cmn_err(CE_PANIC, "failed to add pool version");
3430	}
3431
3432	/* Newly created pools with the right version are always deflated. */
3433	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3434		spa->spa_deflate = TRUE;
3435		if (zap_add(spa->spa_meta_objset,
3436		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3437		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3438			cmn_err(CE_PANIC, "failed to add deflate");
3439		}
3440	}
3441
3442	/*
3443	 * Create the deferred-free bpobj.  Turn off compression
3444	 * because sync-to-convergence takes longer if the blocksize
3445	 * keeps changing.
3446	 */
3447	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3448	dmu_object_set_compress(spa->spa_meta_objset, obj,
3449	    ZIO_COMPRESS_OFF, tx);
3450	if (zap_add(spa->spa_meta_objset,
3451	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3452	    sizeof (uint64_t), 1, &obj, tx) != 0) {
3453		cmn_err(CE_PANIC, "failed to add bpobj");
3454	}
3455	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3456	    spa->spa_meta_objset, obj));
3457
3458	/*
3459	 * Create the pool's history object.
3460	 */
3461	if (version >= SPA_VERSION_ZPOOL_HISTORY)
3462		spa_history_create_obj(spa, tx);
3463
3464	/*
3465	 * Set pool properties.
3466	 */
3467	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3468	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3469	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3470	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3471
3472	if (props != NULL) {
3473		spa_configfile_set(spa, props, B_FALSE);
3474		spa_sync_props(spa, props, tx);
3475	}
3476
3477	dmu_tx_commit(tx);
3478
3479	spa->spa_sync_on = B_TRUE;
3480	txg_sync_start(spa->spa_dsl_pool);
3481
3482	/*
3483	 * We explicitly wait for the first transaction to complete so that our
3484	 * bean counters are appropriately updated.
3485	 */
3486	txg_wait_synced(spa->spa_dsl_pool, txg);
3487
3488	spa_config_sync(spa, B_FALSE, B_TRUE);
3489
3490	spa_history_log_version(spa, "create");
3491
3492	spa->spa_minref = refcount_count(&spa->spa_refcount);
3493
3494	mutex_exit(&spa_namespace_lock);
3495
3496	return (0);
3497}
3498
3499#ifdef _KERNEL
3500/*
3501 * Get the root pool information from the root disk, then import the root pool
3502 * during the system boot up time.
3503 */
3504extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3505
3506static nvlist_t *
3507spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3508{
3509	nvlist_t *config;
3510	nvlist_t *nvtop, *nvroot;
3511	uint64_t pgid;
3512
3513	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3514		return (NULL);
3515
3516	/*
3517	 * Add this top-level vdev to the child array.
3518	 */
3519	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3520	    &nvtop) == 0);
3521	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3522	    &pgid) == 0);
3523	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3524
3525	/*
3526	 * Put this pool's top-level vdevs into a root vdev.
3527	 */
3528	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3529	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3530	    VDEV_TYPE_ROOT) == 0);
3531	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3532	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3533	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3534	    &nvtop, 1) == 0);
3535
3536	/*
3537	 * Replace the existing vdev_tree with the new root vdev in
3538	 * this pool's configuration (remove the old, add the new).
3539	 */
3540	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3541	nvlist_free(nvroot);
3542	return (config);
3543}
3544
3545/*
3546 * Walk the vdev tree and see if we can find a device with "better"
3547 * configuration. A configuration is "better" if the label on that
3548 * device has a more recent txg.
3549 */
3550static void
3551spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3552{
3553	for (int c = 0; c < vd->vdev_children; c++)
3554		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3555
3556	if (vd->vdev_ops->vdev_op_leaf) {
3557		nvlist_t *label;
3558		uint64_t label_txg;
3559
3560		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3561		    &label) != 0)
3562			return;
3563
3564		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3565		    &label_txg) == 0);
3566
3567		/*
3568		 * Do we have a better boot device?
3569		 */
3570		if (label_txg > *txg) {
3571			*txg = label_txg;
3572			*avd = vd;
3573		}
3574		nvlist_free(label);
3575	}
3576}
3577
3578/*
3579 * Import a root pool.
3580 *
3581 * For x86. devpath_list will consist of devid and/or physpath name of
3582 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3583 * The GRUB "findroot" command will return the vdev we should boot.
3584 *
3585 * For Sparc, devpath_list consists the physpath name of the booting device
3586 * no matter the rootpool is a single device pool or a mirrored pool.
3587 * e.g.
3588 *	"/pci@1f,0/ide@d/disk@0,0:a"
3589 */
3590int
3591spa_import_rootpool(char *devpath, char *devid)
3592{
3593	spa_t *spa;
3594	vdev_t *rvd, *bvd, *avd = NULL;
3595	nvlist_t *config, *nvtop;
3596	uint64_t guid, txg;
3597	char *pname;
3598	int error;
3599
3600	/*
3601	 * Read the label from the boot device and generate a configuration.
3602	 */
3603	config = spa_generate_rootconf(devpath, devid, &guid);
3604#if defined(_OBP) && defined(_KERNEL)
3605	if (config == NULL) {
3606		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3607			/* iscsi boot */
3608			get_iscsi_bootpath_phy(devpath);
3609			config = spa_generate_rootconf(devpath, devid, &guid);
3610		}
3611	}
3612#endif
3613	if (config == NULL) {
3614		cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3615		    devpath);
3616		return (EIO);
3617	}
3618
3619	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3620	    &pname) == 0);
3621	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3622
3623	mutex_enter(&spa_namespace_lock);
3624	if ((spa = spa_lookup(pname)) != NULL) {
3625		/*
3626		 * Remove the existing root pool from the namespace so that we
3627		 * can replace it with the correct config we just read in.
3628		 */
3629		spa_remove(spa);
3630	}
3631
3632	spa = spa_add(pname, config, NULL);
3633	spa->spa_is_root = B_TRUE;
3634	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3635
3636	/*
3637	 * Build up a vdev tree based on the boot device's label config.
3638	 */
3639	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3640	    &nvtop) == 0);
3641	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3642	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3643	    VDEV_ALLOC_ROOTPOOL);
3644	spa_config_exit(spa, SCL_ALL, FTAG);
3645	if (error) {
3646		mutex_exit(&spa_namespace_lock);
3647		nvlist_free(config);
3648		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3649		    pname);
3650		return (error);
3651	}
3652
3653	/*
3654	 * Get the boot vdev.
3655	 */
3656	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3657		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3658		    (u_longlong_t)guid);
3659		error = ENOENT;
3660		goto out;
3661	}
3662
3663	/*
3664	 * Determine if there is a better boot device.
3665	 */
3666	avd = bvd;
3667	spa_alt_rootvdev(rvd, &avd, &txg);
3668	if (avd != bvd) {
3669		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3670		    "try booting from '%s'", avd->vdev_path);
3671		error = EINVAL;
3672		goto out;
3673	}
3674
3675	/*
3676	 * If the boot device is part of a spare vdev then ensure that
3677	 * we're booting off the active spare.
3678	 */
3679	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3680	    !bvd->vdev_isspare) {
3681		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3682		    "try booting from '%s'",
3683		    bvd->vdev_parent->
3684		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3685		error = EINVAL;
3686		goto out;
3687	}
3688
3689	error = 0;
3690out:
3691	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3692	vdev_free(rvd);
3693	spa_config_exit(spa, SCL_ALL, FTAG);
3694	mutex_exit(&spa_namespace_lock);
3695
3696	nvlist_free(config);
3697	return (error);
3698}
3699
3700#endif
3701
3702/*
3703 * Import a non-root pool into the system.
3704 */
3705int
3706spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3707{
3708	spa_t *spa;
3709	char *altroot = NULL;
3710	spa_load_state_t state = SPA_LOAD_IMPORT;
3711	zpool_rewind_policy_t policy;
3712	uint64_t mode = spa_mode_global;
3713	uint64_t readonly = B_FALSE;
3714	int error;
3715	nvlist_t *nvroot;
3716	nvlist_t **spares, **l2cache;
3717	uint_t nspares, nl2cache;
3718
3719	/*
3720	 * If a pool with this name exists, return failure.
3721	 */
3722	mutex_enter(&spa_namespace_lock);
3723	if (spa_lookup(pool) != NULL) {
3724		mutex_exit(&spa_namespace_lock);
3725		return (EEXIST);
3726	}
3727
3728	/*
3729	 * Create and initialize the spa structure.
3730	 */
3731	(void) nvlist_lookup_string(props,
3732	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3733	(void) nvlist_lookup_uint64(props,
3734	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3735	if (readonly)
3736		mode = FREAD;
3737	spa = spa_add(pool, config, altroot);
3738	spa->spa_import_flags = flags;
3739
3740	/*
3741	 * Verbatim import - Take a pool and insert it into the namespace
3742	 * as if it had been loaded at boot.
3743	 */
3744	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3745		if (props != NULL)
3746			spa_configfile_set(spa, props, B_FALSE);
3747
3748		spa_config_sync(spa, B_FALSE, B_TRUE);
3749
3750		mutex_exit(&spa_namespace_lock);
3751		spa_history_log_version(spa, "import");
3752
3753		return (0);
3754	}
3755
3756	spa_activate(spa, mode);
3757
3758	/*
3759	 * Don't start async tasks until we know everything is healthy.
3760	 */
3761	spa_async_suspend(spa);
3762
3763	zpool_get_rewind_policy(config, &policy);
3764	if (policy.zrp_request & ZPOOL_DO_REWIND)
3765		state = SPA_LOAD_RECOVER;
3766
3767	/*
3768	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3769	 * because the user-supplied config is actually the one to trust when
3770	 * doing an import.
3771	 */
3772	if (state != SPA_LOAD_RECOVER)
3773		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3774
3775	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3776	    policy.zrp_request);
3777
3778	/*
3779	 * Propagate anything learned while loading the pool and pass it
3780	 * back to caller (i.e. rewind info, missing devices, etc).
3781	 */
3782	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3783	    spa->spa_load_info) == 0);
3784
3785	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3786	/*
3787	 * Toss any existing sparelist, as it doesn't have any validity
3788	 * anymore, and conflicts with spa_has_spare().
3789	 */
3790	if (spa->spa_spares.sav_config) {
3791		nvlist_free(spa->spa_spares.sav_config);
3792		spa->spa_spares.sav_config = NULL;
3793		spa_load_spares(spa);
3794	}
3795	if (spa->spa_l2cache.sav_config) {
3796		nvlist_free(spa->spa_l2cache.sav_config);
3797		spa->spa_l2cache.sav_config = NULL;
3798		spa_load_l2cache(spa);
3799	}
3800
3801	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3802	    &nvroot) == 0);
3803	if (error == 0)
3804		error = spa_validate_aux(spa, nvroot, -1ULL,
3805		    VDEV_ALLOC_SPARE);
3806	if (error == 0)
3807		error = spa_validate_aux(spa, nvroot, -1ULL,
3808		    VDEV_ALLOC_L2CACHE);
3809	spa_config_exit(spa, SCL_ALL, FTAG);
3810
3811	if (props != NULL)
3812		spa_configfile_set(spa, props, B_FALSE);
3813
3814	if (error != 0 || (props && spa_writeable(spa) &&
3815	    (error = spa_prop_set(spa, props)))) {
3816		spa_unload(spa);
3817		spa_deactivate(spa);
3818		spa_remove(spa);
3819		mutex_exit(&spa_namespace_lock);
3820		return (error);
3821	}
3822
3823	spa_async_resume(spa);
3824
3825	/*
3826	 * Override any spares and level 2 cache devices as specified by
3827	 * the user, as these may have correct device names/devids, etc.
3828	 */
3829	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3830	    &spares, &nspares) == 0) {
3831		if (spa->spa_spares.sav_config)
3832			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3833			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3834		else
3835			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3836			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3837		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3838		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3839		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3840		spa_load_spares(spa);
3841		spa_config_exit(spa, SCL_ALL, FTAG);
3842		spa->spa_spares.sav_sync = B_TRUE;
3843	}
3844	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3845	    &l2cache, &nl2cache) == 0) {
3846		if (spa->spa_l2cache.sav_config)
3847			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3848			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3849		else
3850			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3851			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3852		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3853		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3854		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3855		spa_load_l2cache(spa);
3856		spa_config_exit(spa, SCL_ALL, FTAG);
3857		spa->spa_l2cache.sav_sync = B_TRUE;
3858	}
3859
3860	/*
3861	 * Check for any removed devices.
3862	 */
3863	if (spa->spa_autoreplace) {
3864		spa_aux_check_removed(&spa->spa_spares);
3865		spa_aux_check_removed(&spa->spa_l2cache);
3866	}
3867
3868	if (spa_writeable(spa)) {
3869		/*
3870		 * Update the config cache to include the newly-imported pool.
3871		 */
3872		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3873	}
3874
3875	/*
3876	 * It's possible that the pool was expanded while it was exported.
3877	 * We kick off an async task to handle this for us.
3878	 */
3879	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3880
3881	mutex_exit(&spa_namespace_lock);
3882	spa_history_log_version(spa, "import");
3883
3884	return (0);
3885}
3886
3887nvlist_t *
3888spa_tryimport(nvlist_t *tryconfig)
3889{
3890	nvlist_t *config = NULL;
3891	char *poolname;
3892	spa_t *spa;
3893	uint64_t state;
3894	int error;
3895
3896	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3897		return (NULL);
3898
3899	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3900		return (NULL);
3901
3902	/*
3903	 * Create and initialize the spa structure.
3904	 */
3905	mutex_enter(&spa_namespace_lock);
3906	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3907	spa_activate(spa, FREAD);
3908
3909	/*
3910	 * Pass off the heavy lifting to spa_load().
3911	 * Pass TRUE for mosconfig because the user-supplied config
3912	 * is actually the one to trust when doing an import.
3913	 */
3914	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3915
3916	/*
3917	 * If 'tryconfig' was at least parsable, return the current config.
3918	 */
3919	if (spa->spa_root_vdev != NULL) {
3920		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3921		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3922		    poolname) == 0);
3923		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3924		    state) == 0);
3925		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3926		    spa->spa_uberblock.ub_timestamp) == 0);
3927		VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3928		    spa->spa_load_info) == 0);
3929
3930		/*
3931		 * If the bootfs property exists on this pool then we
3932		 * copy it out so that external consumers can tell which
3933		 * pools are bootable.
3934		 */
3935		if ((!error || error == EEXIST) && spa->spa_bootfs) {
3936			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3937
3938			/*
3939			 * We have to play games with the name since the
3940			 * pool was opened as TRYIMPORT_NAME.
3941			 */
3942			if (dsl_dsobj_to_dsname(spa_name(spa),
3943			    spa->spa_bootfs, tmpname) == 0) {
3944				char *cp;
3945				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3946
3947				cp = strchr(tmpname, '/');
3948				if (cp == NULL) {
3949					(void) strlcpy(dsname, tmpname,
3950					    MAXPATHLEN);
3951				} else {
3952					(void) snprintf(dsname, MAXPATHLEN,
3953					    "%s/%s", poolname, ++cp);
3954				}
3955				VERIFY(nvlist_add_string(config,
3956				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3957				kmem_free(dsname, MAXPATHLEN);
3958			}
3959			kmem_free(tmpname, MAXPATHLEN);
3960		}
3961
3962		/*
3963		 * Add the list of hot spares and level 2 cache devices.
3964		 */
3965		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3966		spa_add_spares(spa, config);
3967		spa_add_l2cache(spa, config);
3968		spa_config_exit(spa, SCL_CONFIG, FTAG);
3969	}
3970
3971	spa_unload(spa);
3972	spa_deactivate(spa);
3973	spa_remove(spa);
3974	mutex_exit(&spa_namespace_lock);
3975
3976	return (config);
3977}
3978
3979/*
3980 * Pool export/destroy
3981 *
3982 * The act of destroying or exporting a pool is very simple.  We make sure there
3983 * is no more pending I/O and any references to the pool are gone.  Then, we
3984 * update the pool state and sync all the labels to disk, removing the
3985 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3986 * we don't sync the labels or remove the configuration cache.
3987 */
3988static int
3989spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3990    boolean_t force, boolean_t hardforce)
3991{
3992	spa_t *spa;
3993
3994	if (oldconfig)
3995		*oldconfig = NULL;
3996
3997	if (!(spa_mode_global & FWRITE))
3998		return (EROFS);
3999
4000	mutex_enter(&spa_namespace_lock);
4001	if ((spa = spa_lookup(pool)) == NULL) {
4002		mutex_exit(&spa_namespace_lock);
4003		return (ENOENT);
4004	}
4005
4006	/*
4007	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4008	 * reacquire the namespace lock, and see if we can export.
4009	 */
4010	spa_open_ref(spa, FTAG);
4011	mutex_exit(&spa_namespace_lock);
4012	spa_async_suspend(spa);
4013	mutex_enter(&spa_namespace_lock);
4014	spa_close(spa, FTAG);
4015
4016	/*
4017	 * The pool will be in core if it's openable,
4018	 * in which case we can modify its state.
4019	 */
4020	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4021		/*
4022		 * Objsets may be open only because they're dirty, so we
4023		 * have to force it to sync before checking spa_refcnt.
4024		 */
4025		txg_wait_synced(spa->spa_dsl_pool, 0);
4026
4027		/*
4028		 * A pool cannot be exported or destroyed if there are active
4029		 * references.  If we are resetting a pool, allow references by
4030		 * fault injection handlers.
4031		 */
4032		if (!spa_refcount_zero(spa) ||
4033		    (spa->spa_inject_ref != 0 &&
4034		    new_state != POOL_STATE_UNINITIALIZED)) {
4035			spa_async_resume(spa);
4036			mutex_exit(&spa_namespace_lock);
4037			return (EBUSY);
4038		}
4039
4040		/*
4041		 * A pool cannot be exported if it has an active shared spare.
4042		 * This is to prevent other pools stealing the active spare
4043		 * from an exported pool. At user's own will, such pool can
4044		 * be forcedly exported.
4045		 */
4046		if (!force && new_state == POOL_STATE_EXPORTED &&
4047		    spa_has_active_shared_spare(spa)) {
4048			spa_async_resume(spa);
4049			mutex_exit(&spa_namespace_lock);
4050			return (EXDEV);
4051		}
4052
4053		/*
4054		 * We want this to be reflected on every label,
4055		 * so mark them all dirty.  spa_unload() will do the
4056		 * final sync that pushes these changes out.
4057		 */
4058		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4059			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4060			spa->spa_state = new_state;
4061			spa->spa_final_txg = spa_last_synced_txg(spa) +
4062			    TXG_DEFER_SIZE + 1;
4063			vdev_config_dirty(spa->spa_root_vdev);
4064			spa_config_exit(spa, SCL_ALL, FTAG);
4065		}
4066	}
4067
4068	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4069
4070	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4071		spa_unload(spa);
4072		spa_deactivate(spa);
4073	}
4074
4075	if (oldconfig && spa->spa_config)
4076		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4077
4078	if (new_state != POOL_STATE_UNINITIALIZED) {
4079		if (!hardforce)
4080			spa_config_sync(spa, B_TRUE, B_TRUE);
4081		spa_remove(spa);
4082	}
4083	mutex_exit(&spa_namespace_lock);
4084
4085	return (0);
4086}
4087
4088/*
4089 * Destroy a storage pool.
4090 */
4091int
4092spa_destroy(char *pool)
4093{
4094	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4095	    B_FALSE, B_FALSE));
4096}
4097
4098/*
4099 * Export a storage pool.
4100 */
4101int
4102spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4103    boolean_t hardforce)
4104{
4105	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4106	    force, hardforce));
4107}
4108
4109/*
4110 * Similar to spa_export(), this unloads the spa_t without actually removing it
4111 * from the namespace in any way.
4112 */
4113int
4114spa_reset(char *pool)
4115{
4116	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4117	    B_FALSE, B_FALSE));
4118}
4119
4120/*
4121 * ==========================================================================
4122 * Device manipulation
4123 * ==========================================================================
4124 */
4125
4126/*
4127 * Add a device to a storage pool.
4128 */
4129int
4130spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4131{
4132	uint64_t txg, id;
4133	int error;
4134	vdev_t *rvd = spa->spa_root_vdev;
4135	vdev_t *vd, *tvd;
4136	nvlist_t **spares, **l2cache;
4137	uint_t nspares, nl2cache;
4138
4139	ASSERT(spa_writeable(spa));
4140
4141	txg = spa_vdev_enter(spa);
4142
4143	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4144	    VDEV_ALLOC_ADD)) != 0)
4145		return (spa_vdev_exit(spa, NULL, txg, error));
4146
4147	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
4148
4149	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4150	    &nspares) != 0)
4151		nspares = 0;
4152
4153	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4154	    &nl2cache) != 0)
4155		nl2cache = 0;
4156
4157	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4158		return (spa_vdev_exit(spa, vd, txg, EINVAL));
4159
4160	if (vd->vdev_children != 0 &&
4161	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
4162		return (spa_vdev_exit(spa, vd, txg, error));
4163
4164	/*
4165	 * We must validate the spares and l2cache devices after checking the
4166	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
4167	 */
4168	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4169		return (spa_vdev_exit(spa, vd, txg, error));
4170
4171	/*
4172	 * Transfer each new top-level vdev from vd to rvd.
4173	 */
4174	for (int c = 0; c < vd->vdev_children; c++) {
4175
4176		/*
4177		 * Set the vdev id to the first hole, if one exists.
4178		 */
4179		for (id = 0; id < rvd->vdev_children; id++) {
4180			if (rvd->vdev_child[id]->vdev_ishole) {
4181				vdev_free(rvd->vdev_child[id]);
4182				break;
4183			}
4184		}
4185		tvd = vd->vdev_child[c];
4186		vdev_remove_child(vd, tvd);
4187		tvd->vdev_id = id;
4188		vdev_add_child(rvd, tvd);
4189		vdev_config_dirty(tvd);
4190	}
4191
4192	if (nspares != 0) {
4193		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4194		    ZPOOL_CONFIG_SPARES);
4195		spa_load_spares(spa);
4196		spa->spa_spares.sav_sync = B_TRUE;
4197	}
4198
4199	if (nl2cache != 0) {
4200		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4201		    ZPOOL_CONFIG_L2CACHE);
4202		spa_load_l2cache(spa);
4203		spa->spa_l2cache.sav_sync = B_TRUE;
4204	}
4205
4206	/*
4207	 * We have to be careful when adding new vdevs to an existing pool.
4208	 * If other threads start allocating from these vdevs before we
4209	 * sync the config cache, and we lose power, then upon reboot we may
4210	 * fail to open the pool because there are DVAs that the config cache
4211	 * can't translate.  Therefore, we first add the vdevs without
4212	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4213	 * and then let spa_config_update() initialize the new metaslabs.
4214	 *
4215	 * spa_load() checks for added-but-not-initialized vdevs, so that
4216	 * if we lose power at any point in this sequence, the remaining
4217	 * steps will be completed the next time we load the pool.
4218	 */
4219	(void) spa_vdev_exit(spa, vd, txg, 0);
4220
4221	mutex_enter(&spa_namespace_lock);
4222	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4223	mutex_exit(&spa_namespace_lock);
4224
4225	return (0);
4226}
4227
4228/*
4229 * Attach a device to a mirror.  The arguments are the path to any device
4230 * in the mirror, and the nvroot for the new device.  If the path specifies
4231 * a device that is not mirrored, we automatically insert the mirror vdev.
4232 *
4233 * If 'replacing' is specified, the new device is intended to replace the
4234 * existing device; in this case the two devices are made into their own
4235 * mirror using the 'replacing' vdev, which is functionally identical to
4236 * the mirror vdev (it actually reuses all the same ops) but has a few
4237 * extra rules: you can't attach to it after it's been created, and upon
4238 * completion of resilvering, the first disk (the one being replaced)
4239 * is automatically detached.
4240 */
4241int
4242spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4243{
4244	uint64_t txg, dtl_max_txg;
4245	vdev_t *rvd = spa->spa_root_vdev;
4246	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4247	vdev_ops_t *pvops;
4248	char *oldvdpath, *newvdpath;
4249	int newvd_isspare;
4250	int error;
4251
4252	ASSERT(spa_writeable(spa));
4253
4254	txg = spa_vdev_enter(spa);
4255
4256	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4257
4258	if (oldvd == NULL)
4259		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4260
4261	if (!oldvd->vdev_ops->vdev_op_leaf)
4262		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4263
4264	pvd = oldvd->vdev_parent;
4265
4266	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4267	    VDEV_ALLOC_ATTACH)) != 0)
4268		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4269
4270	if (newrootvd->vdev_children != 1)
4271		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4272
4273	newvd = newrootvd->vdev_child[0];
4274
4275	if (!newvd->vdev_ops->vdev_op_leaf)
4276		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4277
4278	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4279		return (spa_vdev_exit(spa, newrootvd, txg, error));
4280
4281	/*
4282	 * Spares can't replace logs
4283	 */
4284	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4285		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4286
4287	if (!replacing) {
4288		/*
4289		 * For attach, the only allowable parent is a mirror or the root
4290		 * vdev.
4291		 */
4292		if (pvd->vdev_ops != &vdev_mirror_ops &&
4293		    pvd->vdev_ops != &vdev_root_ops)
4294			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4295
4296		pvops = &vdev_mirror_ops;
4297	} else {
4298		/*
4299		 * Active hot spares can only be replaced by inactive hot
4300		 * spares.
4301		 */
4302		if (pvd->vdev_ops == &vdev_spare_ops &&
4303		    oldvd->vdev_isspare &&
4304		    !spa_has_spare(spa, newvd->vdev_guid))
4305			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4306
4307		/*
4308		 * If the source is a hot spare, and the parent isn't already a
4309		 * spare, then we want to create a new hot spare.  Otherwise, we
4310		 * want to create a replacing vdev.  The user is not allowed to
4311		 * attach to a spared vdev child unless the 'isspare' state is
4312		 * the same (spare replaces spare, non-spare replaces
4313		 * non-spare).
4314		 */
4315		if (pvd->vdev_ops == &vdev_replacing_ops &&
4316		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4317			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4318		} else if (pvd->vdev_ops == &vdev_spare_ops &&
4319		    newvd->vdev_isspare != oldvd->vdev_isspare) {
4320			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4321		}
4322
4323		if (newvd->vdev_isspare)
4324			pvops = &vdev_spare_ops;
4325		else
4326			pvops = &vdev_replacing_ops;
4327	}
4328
4329	/*
4330	 * Make sure the new device is big enough.
4331	 */
4332	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4333		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4334
4335	/*
4336	 * The new device cannot have a higher alignment requirement
4337	 * than the top-level vdev.
4338	 */
4339	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4340		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4341
4342	/*
4343	 * If this is an in-place replacement, update oldvd's path and devid
4344	 * to make it distinguishable from newvd, and unopenable from now on.
4345	 */
4346	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4347		spa_strfree(oldvd->vdev_path);
4348		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4349		    KM_SLEEP);
4350		(void) sprintf(oldvd->vdev_path, "%s/%s",
4351		    newvd->vdev_path, "old");
4352		if (oldvd->vdev_devid != NULL) {
4353			spa_strfree(oldvd->vdev_devid);
4354			oldvd->vdev_devid = NULL;
4355		}
4356	}
4357
4358	/* mark the device being resilvered */
4359	newvd->vdev_resilvering = B_TRUE;
4360
4361	/*
4362	 * If the parent is not a mirror, or if we're replacing, insert the new
4363	 * mirror/replacing/spare vdev above oldvd.
4364	 */
4365	if (pvd->vdev_ops != pvops)
4366		pvd = vdev_add_parent(oldvd, pvops);
4367
4368	ASSERT(pvd->vdev_top->vdev_parent == rvd);
4369	ASSERT(pvd->vdev_ops == pvops);
4370	ASSERT(oldvd->vdev_parent == pvd);
4371
4372	/*
4373	 * Extract the new device from its root and add it to pvd.
4374	 */
4375	vdev_remove_child(newrootvd, newvd);
4376	newvd->vdev_id = pvd->vdev_children;
4377	newvd->vdev_crtxg = oldvd->vdev_crtxg;
4378	vdev_add_child(pvd, newvd);
4379
4380	tvd = newvd->vdev_top;
4381	ASSERT(pvd->vdev_top == tvd);
4382	ASSERT(tvd->vdev_parent == rvd);
4383
4384	vdev_config_dirty(tvd);
4385
4386	/*
4387	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4388	 * for any dmu_sync-ed blocks.  It will propagate upward when
4389	 * spa_vdev_exit() calls vdev_dtl_reassess().
4390	 */
4391	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4392
4393	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4394	    dtl_max_txg - TXG_INITIAL);
4395
4396	if (newvd->vdev_isspare) {
4397		spa_spare_activate(newvd);
4398		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4399	}
4400
4401	oldvdpath = spa_strdup(oldvd->vdev_path);
4402	newvdpath = spa_strdup(newvd->vdev_path);
4403	newvd_isspare = newvd->vdev_isspare;
4404
4405	/*
4406	 * Mark newvd's DTL dirty in this txg.
4407	 */
4408	vdev_dirty(tvd, VDD_DTL, newvd, txg);
4409
4410	/*
4411	 * Restart the resilver
4412	 */
4413	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4414
4415	/*
4416	 * Commit the config
4417	 */
4418	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4419
4420	spa_history_log_internal(spa, "vdev attach", NULL,
4421	    "%s vdev=%s %s vdev=%s",
4422	    replacing && newvd_isspare ? "spare in" :
4423	    replacing ? "replace" : "attach", newvdpath,
4424	    replacing ? "for" : "to", oldvdpath);
4425
4426	spa_strfree(oldvdpath);
4427	spa_strfree(newvdpath);
4428
4429	if (spa->spa_bootfs)
4430		spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4431
4432	return (0);
4433}
4434
4435/*
4436 * Detach a device from a mirror or replacing vdev.
4437 * If 'replace_done' is specified, only detach if the parent
4438 * is a replacing vdev.
4439 */
4440int
4441spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4442{
4443	uint64_t txg;
4444	int error;
4445	vdev_t *rvd = spa->spa_root_vdev;
4446	vdev_t *vd, *pvd, *cvd, *tvd;
4447	boolean_t unspare = B_FALSE;
4448	uint64_t unspare_guid;
4449	char *vdpath;
4450
4451	ASSERT(spa_writeable(spa));
4452
4453	txg = spa_vdev_enter(spa);
4454
4455	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4456
4457	if (vd == NULL)
4458		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4459
4460	if (!vd->vdev_ops->vdev_op_leaf)
4461		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4462
4463	pvd = vd->vdev_parent;
4464
4465	/*
4466	 * If the parent/child relationship is not as expected, don't do it.
4467	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4468	 * vdev that's replacing B with C.  The user's intent in replacing
4469	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4470	 * the replace by detaching C, the expected behavior is to end up
4471	 * M(A,B).  But suppose that right after deciding to detach C,
4472	 * the replacement of B completes.  We would have M(A,C), and then
4473	 * ask to detach C, which would leave us with just A -- not what
4474	 * the user wanted.  To prevent this, we make sure that the
4475	 * parent/child relationship hasn't changed -- in this example,
4476	 * that C's parent is still the replacing vdev R.
4477	 */
4478	if (pvd->vdev_guid != pguid && pguid != 0)
4479		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4480
4481	/*
4482	 * Only 'replacing' or 'spare' vdevs can be replaced.
4483	 */
4484	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4485	    pvd->vdev_ops != &vdev_spare_ops)
4486		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4487
4488	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4489	    spa_version(spa) >= SPA_VERSION_SPARES);
4490
4491	/*
4492	 * Only mirror, replacing, and spare vdevs support detach.
4493	 */
4494	if (pvd->vdev_ops != &vdev_replacing_ops &&
4495	    pvd->vdev_ops != &vdev_mirror_ops &&
4496	    pvd->vdev_ops != &vdev_spare_ops)
4497		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4498
4499	/*
4500	 * If this device has the only valid copy of some data,
4501	 * we cannot safely detach it.
4502	 */
4503	if (vdev_dtl_required(vd))
4504		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4505
4506	ASSERT(pvd->vdev_children >= 2);
4507
4508	/*
4509	 * If we are detaching the second disk from a replacing vdev, then
4510	 * check to see if we changed the original vdev's path to have "/old"
4511	 * at the end in spa_vdev_attach().  If so, undo that change now.
4512	 */
4513	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4514	    vd->vdev_path != NULL) {
4515		size_t len = strlen(vd->vdev_path);
4516
4517		for (int c = 0; c < pvd->vdev_children; c++) {
4518			cvd = pvd->vdev_child[c];
4519
4520			if (cvd == vd || cvd->vdev_path == NULL)
4521				continue;
4522
4523			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4524			    strcmp(cvd->vdev_path + len, "/old") == 0) {
4525				spa_strfree(cvd->vdev_path);
4526				cvd->vdev_path = spa_strdup(vd->vdev_path);
4527				break;
4528			}
4529		}
4530	}
4531
4532	/*
4533	 * If we are detaching the original disk from a spare, then it implies
4534	 * that the spare should become a real disk, and be removed from the
4535	 * active spare list for the pool.
4536	 */
4537	if (pvd->vdev_ops == &vdev_spare_ops &&
4538	    vd->vdev_id == 0 &&
4539	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4540		unspare = B_TRUE;
4541
4542	/*
4543	 * Erase the disk labels so the disk can be used for other things.
4544	 * This must be done after all other error cases are handled,
4545	 * but before we disembowel vd (so we can still do I/O to it).
4546	 * But if we can't do it, don't treat the error as fatal --
4547	 * it may be that the unwritability of the disk is the reason
4548	 * it's being detached!
4549	 */
4550	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4551
4552	/*
4553	 * Remove vd from its parent and compact the parent's children.
4554	 */
4555	vdev_remove_child(pvd, vd);
4556	vdev_compact_children(pvd);
4557
4558	/*
4559	 * Remember one of the remaining children so we can get tvd below.
4560	 */
4561	cvd = pvd->vdev_child[pvd->vdev_children - 1];
4562
4563	/*
4564	 * If we need to remove the remaining child from the list of hot spares,
4565	 * do it now, marking the vdev as no longer a spare in the process.
4566	 * We must do this before vdev_remove_parent(), because that can
4567	 * change the GUID if it creates a new toplevel GUID.  For a similar
4568	 * reason, we must remove the spare now, in the same txg as the detach;
4569	 * otherwise someone could attach a new sibling, change the GUID, and
4570	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4571	 */
4572	if (unspare) {
4573		ASSERT(cvd->vdev_isspare);
4574		spa_spare_remove(cvd);
4575		unspare_guid = cvd->vdev_guid;
4576		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4577		cvd->vdev_unspare = B_TRUE;
4578	}
4579
4580	/*
4581	 * If the parent mirror/replacing vdev only has one child,
4582	 * the parent is no longer needed.  Remove it from the tree.
4583	 */
4584	if (pvd->vdev_children == 1) {
4585		if (pvd->vdev_ops == &vdev_spare_ops)
4586			cvd->vdev_unspare = B_FALSE;
4587		vdev_remove_parent(cvd);
4588		cvd->vdev_resilvering = B_FALSE;
4589	}
4590
4591
4592	/*
4593	 * We don't set tvd until now because the parent we just removed
4594	 * may have been the previous top-level vdev.
4595	 */
4596	tvd = cvd->vdev_top;
4597	ASSERT(tvd->vdev_parent == rvd);
4598
4599	/*
4600	 * Reevaluate the parent vdev state.
4601	 */
4602	vdev_propagate_state(cvd);
4603
4604	/*
4605	 * If the 'autoexpand' property is set on the pool then automatically
4606	 * try to expand the size of the pool. For example if the device we
4607	 * just detached was smaller than the others, it may be possible to
4608	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4609	 * first so that we can obtain the updated sizes of the leaf vdevs.
4610	 */
4611	if (spa->spa_autoexpand) {
4612		vdev_reopen(tvd);
4613		vdev_expand(tvd, txg);
4614	}
4615
4616	vdev_config_dirty(tvd);
4617
4618	/*
4619	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4620	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4621	 * But first make sure we're not on any *other* txg's DTL list, to
4622	 * prevent vd from being accessed after it's freed.
4623	 */
4624	vdpath = spa_strdup(vd->vdev_path);
4625	for (int t = 0; t < TXG_SIZE; t++)
4626		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4627	vd->vdev_detached = B_TRUE;
4628	vdev_dirty(tvd, VDD_DTL, vd, txg);
4629
4630	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4631
4632	/* hang on to the spa before we release the lock */
4633	spa_open_ref(spa, FTAG);
4634
4635	error = spa_vdev_exit(spa, vd, txg, 0);
4636
4637	spa_history_log_internal(spa, "detach", NULL,
4638	    "vdev=%s", vdpath);
4639	spa_strfree(vdpath);
4640
4641	/*
4642	 * If this was the removal of the original device in a hot spare vdev,
4643	 * then we want to go through and remove the device from the hot spare
4644	 * list of every other pool.
4645	 */
4646	if (unspare) {
4647		spa_t *altspa = NULL;
4648
4649		mutex_enter(&spa_namespace_lock);
4650		while ((altspa = spa_next(altspa)) != NULL) {
4651			if (altspa->spa_state != POOL_STATE_ACTIVE ||
4652			    altspa == spa)
4653				continue;
4654
4655			spa_open_ref(altspa, FTAG);
4656			mutex_exit(&spa_namespace_lock);
4657			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4658			mutex_enter(&spa_namespace_lock);
4659			spa_close(altspa, FTAG);
4660		}
4661		mutex_exit(&spa_namespace_lock);
4662
4663		/* search the rest of the vdevs for spares to remove */
4664		spa_vdev_resilver_done(spa);
4665	}
4666
4667	/* all done with the spa; OK to release */
4668	mutex_enter(&spa_namespace_lock);
4669	spa_close(spa, FTAG);
4670	mutex_exit(&spa_namespace_lock);
4671
4672	return (error);
4673}
4674
4675/*
4676 * Split a set of devices from their mirrors, and create a new pool from them.
4677 */
4678int
4679spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4680    nvlist_t *props, boolean_t exp)
4681{
4682	int error = 0;
4683	uint64_t txg, *glist;
4684	spa_t *newspa;
4685	uint_t c, children, lastlog;
4686	nvlist_t **child, *nvl, *tmp;
4687	dmu_tx_t *tx;
4688	char *altroot = NULL;
4689	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4690	boolean_t activate_slog;
4691
4692	ASSERT(spa_writeable(spa));
4693
4694	txg = spa_vdev_enter(spa);
4695
4696	/* clear the log and flush everything up to now */
4697	activate_slog = spa_passivate_log(spa);
4698	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4699	error = spa_offline_log(spa);
4700	txg = spa_vdev_config_enter(spa);
4701
4702	if (activate_slog)
4703		spa_activate_log(spa);
4704
4705	if (error != 0)
4706		return (spa_vdev_exit(spa, NULL, txg, error));
4707
4708	/* check new spa name before going any further */
4709	if (spa_lookup(newname) != NULL)
4710		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4711
4712	/*
4713	 * scan through all the children to ensure they're all mirrors
4714	 */
4715	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4716	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4717	    &children) != 0)
4718		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4719
4720	/* first, check to ensure we've got the right child count */
4721	rvd = spa->spa_root_vdev;
4722	lastlog = 0;
4723	for (c = 0; c < rvd->vdev_children; c++) {
4724		vdev_t *vd = rvd->vdev_child[c];
4725
4726		/* don't count the holes & logs as children */
4727		if (vd->vdev_islog || vd->vdev_ishole) {
4728			if (lastlog == 0)
4729				lastlog = c;
4730			continue;
4731		}
4732
4733		lastlog = 0;
4734	}
4735	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4736		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4737
4738	/* next, ensure no spare or cache devices are part of the split */
4739	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4740	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4741		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4742
4743	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4744	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4745
4746	/* then, loop over each vdev and validate it */
4747	for (c = 0; c < children; c++) {
4748		uint64_t is_hole = 0;
4749
4750		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4751		    &is_hole);
4752
4753		if (is_hole != 0) {
4754			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4755			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4756				continue;
4757			} else {
4758				error = EINVAL;
4759				break;
4760			}
4761		}
4762
4763		/* which disk is going to be split? */
4764		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4765		    &glist[c]) != 0) {
4766			error = EINVAL;
4767			break;
4768		}
4769
4770		/* look it up in the spa */
4771		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4772		if (vml[c] == NULL) {
4773			error = ENODEV;
4774			break;
4775		}
4776
4777		/* make sure there's nothing stopping the split */
4778		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4779		    vml[c]->vdev_islog ||
4780		    vml[c]->vdev_ishole ||
4781		    vml[c]->vdev_isspare ||
4782		    vml[c]->vdev_isl2cache ||
4783		    !vdev_writeable(vml[c]) ||
4784		    vml[c]->vdev_children != 0 ||
4785		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4786		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4787			error = EINVAL;
4788			break;
4789		}
4790
4791		if (vdev_dtl_required(vml[c])) {
4792			error = EBUSY;
4793			break;
4794		}
4795
4796		/* we need certain info from the top level */
4797		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4798		    vml[c]->vdev_top->vdev_ms_array) == 0);
4799		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4800		    vml[c]->vdev_top->vdev_ms_shift) == 0);
4801		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4802		    vml[c]->vdev_top->vdev_asize) == 0);
4803		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4804		    vml[c]->vdev_top->vdev_ashift) == 0);
4805	}
4806
4807	if (error != 0) {
4808		kmem_free(vml, children * sizeof (vdev_t *));
4809		kmem_free(glist, children * sizeof (uint64_t));
4810		return (spa_vdev_exit(spa, NULL, txg, error));
4811	}
4812
4813	/* stop writers from using the disks */
4814	for (c = 0; c < children; c++) {
4815		if (vml[c] != NULL)
4816			vml[c]->vdev_offline = B_TRUE;
4817	}
4818	vdev_reopen(spa->spa_root_vdev);
4819
4820	/*
4821	 * Temporarily record the splitting vdevs in the spa config.  This
4822	 * will disappear once the config is regenerated.
4823	 */
4824	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4825	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4826	    glist, children) == 0);
4827	kmem_free(glist, children * sizeof (uint64_t));
4828
4829	mutex_enter(&spa->spa_props_lock);
4830	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4831	    nvl) == 0);
4832	mutex_exit(&spa->spa_props_lock);
4833	spa->spa_config_splitting = nvl;
4834	vdev_config_dirty(spa->spa_root_vdev);
4835
4836	/* configure and create the new pool */
4837	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4838	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4839	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4840	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4841	    spa_version(spa)) == 0);
4842	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4843	    spa->spa_config_txg) == 0);
4844	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4845	    spa_generate_guid(NULL)) == 0);
4846	(void) nvlist_lookup_string(props,
4847	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4848
4849	/* add the new pool to the namespace */
4850	newspa = spa_add(newname, config, altroot);
4851	newspa->spa_config_txg = spa->spa_config_txg;
4852	spa_set_log_state(newspa, SPA_LOG_CLEAR);
4853
4854	/* release the spa config lock, retaining the namespace lock */
4855	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4856
4857	if (zio_injection_enabled)
4858		zio_handle_panic_injection(spa, FTAG, 1);
4859
4860	spa_activate(newspa, spa_mode_global);
4861	spa_async_suspend(newspa);
4862
4863	/* create the new pool from the disks of the original pool */
4864	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4865	if (error)
4866		goto out;
4867
4868	/* if that worked, generate a real config for the new pool */
4869	if (newspa->spa_root_vdev != NULL) {
4870		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4871		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4872		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4873		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4874		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4875		    B_TRUE));
4876	}
4877
4878	/* set the props */
4879	if (props != NULL) {
4880		spa_configfile_set(newspa, props, B_FALSE);
4881		error = spa_prop_set(newspa, props);
4882		if (error)
4883			goto out;
4884	}
4885
4886	/* flush everything */
4887	txg = spa_vdev_config_enter(newspa);
4888	vdev_config_dirty(newspa->spa_root_vdev);
4889	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4890
4891	if (zio_injection_enabled)
4892		zio_handle_panic_injection(spa, FTAG, 2);
4893
4894	spa_async_resume(newspa);
4895
4896	/* finally, update the original pool's config */
4897	txg = spa_vdev_config_enter(spa);
4898	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4899	error = dmu_tx_assign(tx, TXG_WAIT);
4900	if (error != 0)
4901		dmu_tx_abort(tx);
4902	for (c = 0; c < children; c++) {
4903		if (vml[c] != NULL) {
4904			vdev_split(vml[c]);
4905			if (error == 0)
4906				spa_history_log_internal(spa, "detach", tx,
4907				    "vdev=%s", vml[c]->vdev_path);
4908			vdev_free(vml[c]);
4909		}
4910	}
4911	vdev_config_dirty(spa->spa_root_vdev);
4912	spa->spa_config_splitting = NULL;
4913	nvlist_free(nvl);
4914	if (error == 0)
4915		dmu_tx_commit(tx);
4916	(void) spa_vdev_exit(spa, NULL, txg, 0);
4917
4918	if (zio_injection_enabled)
4919		zio_handle_panic_injection(spa, FTAG, 3);
4920
4921	/* split is complete; log a history record */
4922	spa_history_log_internal(newspa, "split", NULL,
4923	    "from pool %s", spa_name(spa));
4924
4925	kmem_free(vml, children * sizeof (vdev_t *));
4926
4927	/* if we're not going to mount the filesystems in userland, export */
4928	if (exp)
4929		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4930		    B_FALSE, B_FALSE);
4931
4932	return (error);
4933
4934out:
4935	spa_unload(newspa);
4936	spa_deactivate(newspa);
4937	spa_remove(newspa);
4938
4939	txg = spa_vdev_config_enter(spa);
4940
4941	/* re-online all offlined disks */
4942	for (c = 0; c < children; c++) {
4943		if (vml[c] != NULL)
4944			vml[c]->vdev_offline = B_FALSE;
4945	}
4946	vdev_reopen(spa->spa_root_vdev);
4947
4948	nvlist_free(spa->spa_config_splitting);
4949	spa->spa_config_splitting = NULL;
4950	(void) spa_vdev_exit(spa, NULL, txg, error);
4951
4952	kmem_free(vml, children * sizeof (vdev_t *));
4953	return (error);
4954}
4955
4956static nvlist_t *
4957spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4958{
4959	for (int i = 0; i < count; i++) {
4960		uint64_t guid;
4961
4962		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4963		    &guid) == 0);
4964
4965		if (guid == target_guid)
4966			return (nvpp[i]);
4967	}
4968
4969	return (NULL);
4970}
4971
4972static void
4973spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4974	nvlist_t *dev_to_remove)
4975{
4976	nvlist_t **newdev = NULL;
4977
4978	if (count > 1)
4979		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4980
4981	for (int i = 0, j = 0; i < count; i++) {
4982		if (dev[i] == dev_to_remove)
4983			continue;
4984		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4985	}
4986
4987	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4988	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4989
4990	for (int i = 0; i < count - 1; i++)
4991		nvlist_free(newdev[i]);
4992
4993	if (count > 1)
4994		kmem_free(newdev, (count - 1) * sizeof (void *));
4995}
4996
4997/*
4998 * Evacuate the device.
4999 */
5000static int
5001spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5002{
5003	uint64_t txg;
5004	int error = 0;
5005
5006	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5007	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5008	ASSERT(vd == vd->vdev_top);
5009
5010	/*
5011	 * Evacuate the device.  We don't hold the config lock as writer
5012	 * since we need to do I/O but we do keep the
5013	 * spa_namespace_lock held.  Once this completes the device
5014	 * should no longer have any blocks allocated on it.
5015	 */
5016	if (vd->vdev_islog) {
5017		if (vd->vdev_stat.vs_alloc != 0)
5018			error = spa_offline_log(spa);
5019	} else {
5020		error = ENOTSUP;
5021	}
5022
5023	if (error)
5024		return (error);
5025
5026	/*
5027	 * The evacuation succeeded.  Remove any remaining MOS metadata
5028	 * associated with this vdev, and wait for these changes to sync.
5029	 */
5030	ASSERT0(vd->vdev_stat.vs_alloc);
5031	txg = spa_vdev_config_enter(spa);
5032	vd->vdev_removing = B_TRUE;
5033	vdev_dirty(vd, 0, NULL, txg);
5034	vdev_config_dirty(vd);
5035	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5036
5037	return (0);
5038}
5039
5040/*
5041 * Complete the removal by cleaning up the namespace.
5042 */
5043static void
5044spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5045{
5046	vdev_t *rvd = spa->spa_root_vdev;
5047	uint64_t id = vd->vdev_id;
5048	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5049
5050	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5051	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5052	ASSERT(vd == vd->vdev_top);
5053
5054	/*
5055	 * Only remove any devices which are empty.
5056	 */
5057	if (vd->vdev_stat.vs_alloc != 0)
5058		return;
5059
5060	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5061
5062	if (list_link_active(&vd->vdev_state_dirty_node))
5063		vdev_state_clean(vd);
5064	if (list_link_active(&vd->vdev_config_dirty_node))
5065		vdev_config_clean(vd);
5066
5067	vdev_free(vd);
5068
5069	if (last_vdev) {
5070		vdev_compact_children(rvd);
5071	} else {
5072		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5073		vdev_add_child(rvd, vd);
5074	}
5075	vdev_config_dirty(rvd);
5076
5077	/*
5078	 * Reassess the health of our root vdev.
5079	 */
5080	vdev_reopen(rvd);
5081}
5082
5083/*
5084 * Remove a device from the pool -
5085 *
5086 * Removing a device from the vdev namespace requires several steps
5087 * and can take a significant amount of time.  As a result we use
5088 * the spa_vdev_config_[enter/exit] functions which allow us to
5089 * grab and release the spa_config_lock while still holding the namespace
5090 * lock.  During each step the configuration is synced out.
5091 */
5092
5093/*
5094 * Remove a device from the pool.  Currently, this supports removing only hot
5095 * spares, slogs, and level 2 ARC devices.
5096 */
5097int
5098spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5099{
5100	vdev_t *vd;
5101	metaslab_group_t *mg;
5102	nvlist_t **spares, **l2cache, *nv;
5103	uint64_t txg = 0;
5104	uint_t nspares, nl2cache;
5105	int error = 0;
5106	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5107
5108	ASSERT(spa_writeable(spa));
5109
5110	if (!locked)
5111		txg = spa_vdev_enter(spa);
5112
5113	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5114
5115	if (spa->spa_spares.sav_vdevs != NULL &&
5116	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5117	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5118	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5119		/*
5120		 * Only remove the hot spare if it's not currently in use
5121		 * in this pool.
5122		 */
5123		if (vd == NULL || unspare) {
5124			spa_vdev_remove_aux(spa->spa_spares.sav_config,
5125			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5126			spa_load_spares(spa);
5127			spa->spa_spares.sav_sync = B_TRUE;
5128		} else {
5129			error = EBUSY;
5130		}
5131	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
5132	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5133	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5134	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5135		/*
5136		 * Cache devices can always be removed.
5137		 */
5138		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5139		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5140		spa_load_l2cache(spa);
5141		spa->spa_l2cache.sav_sync = B_TRUE;
5142	} else if (vd != NULL && vd->vdev_islog) {
5143		ASSERT(!locked);
5144		ASSERT(vd == vd->vdev_top);
5145
5146		/*
5147		 * XXX - Once we have bp-rewrite this should
5148		 * become the common case.
5149		 */
5150
5151		mg = vd->vdev_mg;
5152
5153		/*
5154		 * Stop allocating from this vdev.
5155		 */
5156		metaslab_group_passivate(mg);
5157
5158		/*
5159		 * Wait for the youngest allocations and frees to sync,
5160		 * and then wait for the deferral of those frees to finish.
5161		 */
5162		spa_vdev_config_exit(spa, NULL,
5163		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5164
5165		/*
5166		 * Attempt to evacuate the vdev.
5167		 */
5168		error = spa_vdev_remove_evacuate(spa, vd);
5169
5170		txg = spa_vdev_config_enter(spa);
5171
5172		/*
5173		 * If we couldn't evacuate the vdev, unwind.
5174		 */
5175		if (error) {
5176			metaslab_group_activate(mg);
5177			return (spa_vdev_exit(spa, NULL, txg, error));
5178		}
5179
5180		/*
5181		 * Clean up the vdev namespace.
5182		 */
5183		spa_vdev_remove_from_namespace(spa, vd);
5184
5185	} else if (vd != NULL) {
5186		/*
5187		 * Normal vdevs cannot be removed (yet).
5188		 */
5189		error = ENOTSUP;
5190	} else {
5191		/*
5192		 * There is no vdev of any kind with the specified guid.
5193		 */
5194		error = ENOENT;
5195	}
5196
5197	if (!locked)
5198		return (spa_vdev_exit(spa, NULL, txg, error));
5199
5200	return (error);
5201}
5202
5203/*
5204 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5205 * current spared, so we can detach it.
5206 */
5207static vdev_t *
5208spa_vdev_resilver_done_hunt(vdev_t *vd)
5209{
5210	vdev_t *newvd, *oldvd;
5211
5212	for (int c = 0; c < vd->vdev_children; c++) {
5213		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5214		if (oldvd != NULL)
5215			return (oldvd);
5216	}
5217
5218	/*
5219	 * Check for a completed replacement.  We always consider the first
5220	 * vdev in the list to be the oldest vdev, and the last one to be
5221	 * the newest (see spa_vdev_attach() for how that works).  In
5222	 * the case where the newest vdev is faulted, we will not automatically
5223	 * remove it after a resilver completes.  This is OK as it will require
5224	 * user intervention to determine which disk the admin wishes to keep.
5225	 */
5226	if (vd->vdev_ops == &vdev_replacing_ops) {
5227		ASSERT(vd->vdev_children > 1);
5228
5229		newvd = vd->vdev_child[vd->vdev_children - 1];
5230		oldvd = vd->vdev_child[0];
5231
5232		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5233		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5234		    !vdev_dtl_required(oldvd))
5235			return (oldvd);
5236	}
5237
5238	/*
5239	 * Check for a completed resilver with the 'unspare' flag set.
5240	 */
5241	if (vd->vdev_ops == &vdev_spare_ops) {
5242		vdev_t *first = vd->vdev_child[0];
5243		vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5244
5245		if (last->vdev_unspare) {
5246			oldvd = first;
5247			newvd = last;
5248		} else if (first->vdev_unspare) {
5249			oldvd = last;
5250			newvd = first;
5251		} else {
5252			oldvd = NULL;
5253		}
5254
5255		if (oldvd != NULL &&
5256		    vdev_dtl_empty(newvd, DTL_MISSING) &&
5257		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5258		    !vdev_dtl_required(oldvd))
5259			return (oldvd);
5260
5261		/*
5262		 * If there are more than two spares attached to a disk,
5263		 * and those spares are not required, then we want to
5264		 * attempt to free them up now so that they can be used
5265		 * by other pools.  Once we're back down to a single
5266		 * disk+spare, we stop removing them.
5267		 */
5268		if (vd->vdev_children > 2) {
5269			newvd = vd->vdev_child[1];
5270
5271			if (newvd->vdev_isspare && last->vdev_isspare &&
5272			    vdev_dtl_empty(last, DTL_MISSING) &&
5273			    vdev_dtl_empty(last, DTL_OUTAGE) &&
5274			    !vdev_dtl_required(newvd))
5275				return (newvd);
5276		}
5277	}
5278
5279	return (NULL);
5280}
5281
5282static void
5283spa_vdev_resilver_done(spa_t *spa)
5284{
5285	vdev_t *vd, *pvd, *ppvd;
5286	uint64_t guid, sguid, pguid, ppguid;
5287
5288	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5289
5290	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5291		pvd = vd->vdev_parent;
5292		ppvd = pvd->vdev_parent;
5293		guid = vd->vdev_guid;
5294		pguid = pvd->vdev_guid;
5295		ppguid = ppvd->vdev_guid;
5296		sguid = 0;
5297		/*
5298		 * If we have just finished replacing a hot spared device, then
5299		 * we need to detach the parent's first child (the original hot
5300		 * spare) as well.
5301		 */
5302		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5303		    ppvd->vdev_children == 2) {
5304			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5305			sguid = ppvd->vdev_child[1]->vdev_guid;
5306		}
5307		spa_config_exit(spa, SCL_ALL, FTAG);
5308		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5309			return;
5310		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5311			return;
5312		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5313	}
5314
5315	spa_config_exit(spa, SCL_ALL, FTAG);
5316}
5317
5318/*
5319 * Update the stored path or FRU for this vdev.
5320 */
5321int
5322spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5323    boolean_t ispath)
5324{
5325	vdev_t *vd;
5326	boolean_t sync = B_FALSE;
5327
5328	ASSERT(spa_writeable(spa));
5329
5330	spa_vdev_state_enter(spa, SCL_ALL);
5331
5332	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5333		return (spa_vdev_state_exit(spa, NULL, ENOENT));
5334
5335	if (!vd->vdev_ops->vdev_op_leaf)
5336		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5337
5338	if (ispath) {
5339		if (strcmp(value, vd->vdev_path) != 0) {
5340			spa_strfree(vd->vdev_path);
5341			vd->vdev_path = spa_strdup(value);
5342			sync = B_TRUE;
5343		}
5344	} else {
5345		if (vd->vdev_fru == NULL) {
5346			vd->vdev_fru = spa_strdup(value);
5347			sync = B_TRUE;
5348		} else if (strcmp(value, vd->vdev_fru) != 0) {
5349			spa_strfree(vd->vdev_fru);
5350			vd->vdev_fru = spa_strdup(value);
5351			sync = B_TRUE;
5352		}
5353	}
5354
5355	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5356}
5357
5358int
5359spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5360{
5361	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5362}
5363
5364int
5365spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5366{
5367	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5368}
5369
5370/*
5371 * ==========================================================================
5372 * SPA Scanning
5373 * ==========================================================================
5374 */
5375
5376int
5377spa_scan_stop(spa_t *spa)
5378{
5379	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5380	if (dsl_scan_resilvering(spa->spa_dsl_pool))
5381		return (EBUSY);
5382	return (dsl_scan_cancel(spa->spa_dsl_pool));
5383}
5384
5385int
5386spa_scan(spa_t *spa, pool_scan_func_t func)
5387{
5388	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5389
5390	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5391		return (ENOTSUP);
5392
5393	/*
5394	 * If a resilver was requested, but there is no DTL on a
5395	 * writeable leaf device, we have nothing to do.
5396	 */
5397	if (func == POOL_SCAN_RESILVER &&
5398	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5399		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5400		return (0);
5401	}
5402
5403	return (dsl_scan(spa->spa_dsl_pool, func));
5404}
5405
5406/*
5407 * ==========================================================================
5408 * SPA async task processing
5409 * ==========================================================================
5410 */
5411
5412static void
5413spa_async_remove(spa_t *spa, vdev_t *vd)
5414{
5415	if (vd->vdev_remove_wanted) {
5416		vd->vdev_remove_wanted = B_FALSE;
5417		vd->vdev_delayed_close = B_FALSE;
5418		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5419
5420		/*
5421		 * We want to clear the stats, but we don't want to do a full
5422		 * vdev_clear() as that will cause us to throw away
5423		 * degraded/faulted state as well as attempt to reopen the
5424		 * device, all of which is a waste.
5425		 */
5426		vd->vdev_stat.vs_read_errors = 0;
5427		vd->vdev_stat.vs_write_errors = 0;
5428		vd->vdev_stat.vs_checksum_errors = 0;
5429
5430		vdev_state_dirty(vd->vdev_top);
5431	}
5432
5433	for (int c = 0; c < vd->vdev_children; c++)
5434		spa_async_remove(spa, vd->vdev_child[c]);
5435}
5436
5437static void
5438spa_async_probe(spa_t *spa, vdev_t *vd)
5439{
5440	if (vd->vdev_probe_wanted) {
5441		vd->vdev_probe_wanted = B_FALSE;
5442		vdev_reopen(vd);	/* vdev_open() does the actual probe */
5443	}
5444
5445	for (int c = 0; c < vd->vdev_children; c++)
5446		spa_async_probe(spa, vd->vdev_child[c]);
5447}
5448
5449static void
5450spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5451{
5452	sysevent_id_t eid;
5453	nvlist_t *attr;
5454	char *physpath;
5455
5456	if (!spa->spa_autoexpand)
5457		return;
5458
5459	for (int c = 0; c < vd->vdev_children; c++) {
5460		vdev_t *cvd = vd->vdev_child[c];
5461		spa_async_autoexpand(spa, cvd);
5462	}
5463
5464	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5465		return;
5466
5467	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5468	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5469
5470	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5471	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5472
5473	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5474	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5475
5476	nvlist_free(attr);
5477	kmem_free(physpath, MAXPATHLEN);
5478}
5479
5480static void
5481spa_async_thread(spa_t *spa)
5482{
5483	int tasks;
5484
5485	ASSERT(spa->spa_sync_on);
5486
5487	mutex_enter(&spa->spa_async_lock);
5488	tasks = spa->spa_async_tasks;
5489	spa->spa_async_tasks = 0;
5490	mutex_exit(&spa->spa_async_lock);
5491
5492	/*
5493	 * See if the config needs to be updated.
5494	 */
5495	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5496		uint64_t old_space, new_space;
5497
5498		mutex_enter(&spa_namespace_lock);
5499		old_space = metaslab_class_get_space(spa_normal_class(spa));
5500		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5501		new_space = metaslab_class_get_space(spa_normal_class(spa));
5502		mutex_exit(&spa_namespace_lock);
5503
5504		/*
5505		 * If the pool grew as a result of the config update,
5506		 * then log an internal history event.
5507		 */
5508		if (new_space != old_space) {
5509			spa_history_log_internal(spa, "vdev online", NULL,
5510			    "pool '%s' size: %llu(+%llu)",
5511			    spa_name(spa), new_space, new_space - old_space);
5512		}
5513	}
5514
5515	/*
5516	 * See if any devices need to be marked REMOVED.
5517	 */
5518	if (tasks & SPA_ASYNC_REMOVE) {
5519		spa_vdev_state_enter(spa, SCL_NONE);
5520		spa_async_remove(spa, spa->spa_root_vdev);
5521		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5522			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5523		for (int i = 0; i < spa->spa_spares.sav_count; i++)
5524			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5525		(void) spa_vdev_state_exit(spa, NULL, 0);
5526	}
5527
5528	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5529		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5530		spa_async_autoexpand(spa, spa->spa_root_vdev);
5531		spa_config_exit(spa, SCL_CONFIG, FTAG);
5532	}
5533
5534	/*
5535	 * See if any devices need to be probed.
5536	 */
5537	if (tasks & SPA_ASYNC_PROBE) {
5538		spa_vdev_state_enter(spa, SCL_NONE);
5539		spa_async_probe(spa, spa->spa_root_vdev);
5540		(void) spa_vdev_state_exit(spa, NULL, 0);
5541	}
5542
5543	/*
5544	 * If any devices are done replacing, detach them.
5545	 */
5546	if (tasks & SPA_ASYNC_RESILVER_DONE)
5547		spa_vdev_resilver_done(spa);
5548
5549	/*
5550	 * Kick off a resilver.
5551	 */
5552	if (tasks & SPA_ASYNC_RESILVER)
5553		dsl_resilver_restart(spa->spa_dsl_pool, 0);
5554
5555	/*
5556	 * Let the world know that we're done.
5557	 */
5558	mutex_enter(&spa->spa_async_lock);
5559	spa->spa_async_thread = NULL;
5560	cv_broadcast(&spa->spa_async_cv);
5561	mutex_exit(&spa->spa_async_lock);
5562	thread_exit();
5563}
5564
5565void
5566spa_async_suspend(spa_t *spa)
5567{
5568	mutex_enter(&spa->spa_async_lock);
5569	spa->spa_async_suspended++;
5570	while (spa->spa_async_thread != NULL)
5571		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5572	mutex_exit(&spa->spa_async_lock);
5573}
5574
5575void
5576spa_async_resume(spa_t *spa)
5577{
5578	mutex_enter(&spa->spa_async_lock);
5579	ASSERT(spa->spa_async_suspended != 0);
5580	spa->spa_async_suspended--;
5581	mutex_exit(&spa->spa_async_lock);
5582}
5583
5584static void
5585spa_async_dispatch(spa_t *spa)
5586{
5587	mutex_enter(&spa->spa_async_lock);
5588	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5589	    spa->spa_async_thread == NULL &&
5590	    rootdir != NULL && !vn_is_readonly(rootdir))
5591		spa->spa_async_thread = thread_create(NULL, 0,
5592		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5593	mutex_exit(&spa->spa_async_lock);
5594}
5595
5596void
5597spa_async_request(spa_t *spa, int task)
5598{
5599	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5600	mutex_enter(&spa->spa_async_lock);
5601	spa->spa_async_tasks |= task;
5602	mutex_exit(&spa->spa_async_lock);
5603}
5604
5605/*
5606 * ==========================================================================
5607 * SPA syncing routines
5608 * ==========================================================================
5609 */
5610
5611static int
5612bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5613{
5614	bpobj_t *bpo = arg;
5615	bpobj_enqueue(bpo, bp, tx);
5616	return (0);
5617}
5618
5619static int
5620spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5621{
5622	zio_t *zio = arg;
5623
5624	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5625	    zio->io_flags));
5626	return (0);
5627}
5628
5629static void
5630spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5631{
5632	char *packed = NULL;
5633	size_t bufsize;
5634	size_t nvsize = 0;
5635	dmu_buf_t *db;
5636
5637	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5638
5639	/*
5640	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5641	 * information.  This avoids the dbuf_will_dirty() path and
5642	 * saves us a pre-read to get data we don't actually care about.
5643	 */
5644	bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5645	packed = kmem_alloc(bufsize, KM_SLEEP);
5646
5647	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5648	    KM_SLEEP) == 0);
5649	bzero(packed + nvsize, bufsize - nvsize);
5650
5651	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5652
5653	kmem_free(packed, bufsize);
5654
5655	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5656	dmu_buf_will_dirty(db, tx);
5657	*(uint64_t *)db->db_data = nvsize;
5658	dmu_buf_rele(db, FTAG);
5659}
5660
5661static void
5662spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5663    const char *config, const char *entry)
5664{
5665	nvlist_t *nvroot;
5666	nvlist_t **list;
5667	int i;
5668
5669	if (!sav->sav_sync)
5670		return;
5671
5672	/*
5673	 * Update the MOS nvlist describing the list of available devices.
5674	 * spa_validate_aux() will have already made sure this nvlist is
5675	 * valid and the vdevs are labeled appropriately.
5676	 */
5677	if (sav->sav_object == 0) {
5678		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5679		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5680		    sizeof (uint64_t), tx);
5681		VERIFY(zap_update(spa->spa_meta_objset,
5682		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5683		    &sav->sav_object, tx) == 0);
5684	}
5685
5686	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5687	if (sav->sav_count == 0) {
5688		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5689	} else {
5690		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5691		for (i = 0; i < sav->sav_count; i++)
5692			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5693			    B_FALSE, VDEV_CONFIG_L2CACHE);
5694		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5695		    sav->sav_count) == 0);
5696		for (i = 0; i < sav->sav_count; i++)
5697			nvlist_free(list[i]);
5698		kmem_free(list, sav->sav_count * sizeof (void *));
5699	}
5700
5701	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5702	nvlist_free(nvroot);
5703
5704	sav->sav_sync = B_FALSE;
5705}
5706
5707static void
5708spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5709{
5710	nvlist_t *config;
5711
5712	if (list_is_empty(&spa->spa_config_dirty_list))
5713		return;
5714
5715	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5716
5717	config = spa_config_generate(spa, spa->spa_root_vdev,
5718	    dmu_tx_get_txg(tx), B_FALSE);
5719
5720	spa_config_exit(spa, SCL_STATE, FTAG);
5721
5722	if (spa->spa_config_syncing)
5723		nvlist_free(spa->spa_config_syncing);
5724	spa->spa_config_syncing = config;
5725
5726	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5727}
5728
5729static void
5730spa_sync_version(void *arg1, void *arg2, dmu_tx_t *tx)
5731{
5732	spa_t *spa = arg1;
5733	uint64_t version = *(uint64_t *)arg2;
5734
5735	/*
5736	 * Setting the version is special cased when first creating the pool.
5737	 */
5738	ASSERT(tx->tx_txg != TXG_INITIAL);
5739
5740	ASSERT(version <= SPA_VERSION);
5741	ASSERT(version >= spa_version(spa));
5742
5743	spa->spa_uberblock.ub_version = version;
5744	vdev_config_dirty(spa->spa_root_vdev);
5745	spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5746}
5747
5748/*
5749 * Set zpool properties.
5750 */
5751static void
5752spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5753{
5754	spa_t *spa = arg1;
5755	objset_t *mos = spa->spa_meta_objset;
5756	nvlist_t *nvp = arg2;
5757	nvpair_t *elem = NULL;
5758
5759	mutex_enter(&spa->spa_props_lock);
5760
5761	while ((elem = nvlist_next_nvpair(nvp, elem))) {
5762		uint64_t intval;
5763		char *strval, *fname;
5764		zpool_prop_t prop;
5765		const char *propname;
5766		zprop_type_t proptype;
5767		zfeature_info_t *feature;
5768
5769		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5770		case ZPROP_INVAL:
5771			/*
5772			 * We checked this earlier in spa_prop_validate().
5773			 */
5774			ASSERT(zpool_prop_feature(nvpair_name(elem)));
5775
5776			fname = strchr(nvpair_name(elem), '@') + 1;
5777			VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
5778
5779			spa_feature_enable(spa, feature, tx);
5780			spa_history_log_internal(spa, "set", tx,
5781			    "%s=enabled", nvpair_name(elem));
5782			break;
5783
5784		case ZPOOL_PROP_VERSION:
5785			VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5786			/*
5787			 * The version is synced seperatly before other
5788			 * properties and should be correct by now.
5789			 */
5790			ASSERT3U(spa_version(spa), >=, intval);
5791			break;
5792
5793		case ZPOOL_PROP_ALTROOT:
5794			/*
5795			 * 'altroot' is a non-persistent property. It should
5796			 * have been set temporarily at creation or import time.
5797			 */
5798			ASSERT(spa->spa_root != NULL);
5799			break;
5800
5801		case ZPOOL_PROP_READONLY:
5802		case ZPOOL_PROP_CACHEFILE:
5803			/*
5804			 * 'readonly' and 'cachefile' are also non-persisitent
5805			 * properties.
5806			 */
5807			break;
5808		case ZPOOL_PROP_COMMENT:
5809			VERIFY(nvpair_value_string(elem, &strval) == 0);
5810			if (spa->spa_comment != NULL)
5811				spa_strfree(spa->spa_comment);
5812			spa->spa_comment = spa_strdup(strval);
5813			/*
5814			 * We need to dirty the configuration on all the vdevs
5815			 * so that their labels get updated.  It's unnecessary
5816			 * to do this for pool creation since the vdev's
5817			 * configuratoin has already been dirtied.
5818			 */
5819			if (tx->tx_txg != TXG_INITIAL)
5820				vdev_config_dirty(spa->spa_root_vdev);
5821			spa_history_log_internal(spa, "set", tx,
5822			    "%s=%s", nvpair_name(elem), strval);
5823			break;
5824		default:
5825			/*
5826			 * Set pool property values in the poolprops mos object.
5827			 */
5828			if (spa->spa_pool_props_object == 0) {
5829				spa->spa_pool_props_object =
5830				    zap_create_link(mos, DMU_OT_POOL_PROPS,
5831				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5832				    tx);
5833			}
5834
5835			/* normalize the property name */
5836			propname = zpool_prop_to_name(prop);
5837			proptype = zpool_prop_get_type(prop);
5838
5839			if (nvpair_type(elem) == DATA_TYPE_STRING) {
5840				ASSERT(proptype == PROP_TYPE_STRING);
5841				VERIFY(nvpair_value_string(elem, &strval) == 0);
5842				VERIFY(zap_update(mos,
5843				    spa->spa_pool_props_object, propname,
5844				    1, strlen(strval) + 1, strval, tx) == 0);
5845				spa_history_log_internal(spa, "set", tx,
5846				    "%s=%s", nvpair_name(elem), strval);
5847			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5848				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5849
5850				if (proptype == PROP_TYPE_INDEX) {
5851					const char *unused;
5852					VERIFY(zpool_prop_index_to_string(
5853					    prop, intval, &unused) == 0);
5854				}
5855				VERIFY(zap_update(mos,
5856				    spa->spa_pool_props_object, propname,
5857				    8, 1, &intval, tx) == 0);
5858				spa_history_log_internal(spa, "set", tx,
5859				    "%s=%lld", nvpair_name(elem), intval);
5860			} else {
5861				ASSERT(0); /* not allowed */
5862			}
5863
5864			switch (prop) {
5865			case ZPOOL_PROP_DELEGATION:
5866				spa->spa_delegation = intval;
5867				break;
5868			case ZPOOL_PROP_BOOTFS:
5869				spa->spa_bootfs = intval;
5870				break;
5871			case ZPOOL_PROP_FAILUREMODE:
5872				spa->spa_failmode = intval;
5873				break;
5874			case ZPOOL_PROP_AUTOEXPAND:
5875				spa->spa_autoexpand = intval;
5876				if (tx->tx_txg != TXG_INITIAL)
5877					spa_async_request(spa,
5878					    SPA_ASYNC_AUTOEXPAND);
5879				break;
5880			case ZPOOL_PROP_DEDUPDITTO:
5881				spa->spa_dedup_ditto = intval;
5882				break;
5883			default:
5884				break;
5885			}
5886		}
5887
5888	}
5889
5890	mutex_exit(&spa->spa_props_lock);
5891}
5892
5893/*
5894 * Perform one-time upgrade on-disk changes.  spa_version() does not
5895 * reflect the new version this txg, so there must be no changes this
5896 * txg to anything that the upgrade code depends on after it executes.
5897 * Therefore this must be called after dsl_pool_sync() does the sync
5898 * tasks.
5899 */
5900static void
5901spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5902{
5903	dsl_pool_t *dp = spa->spa_dsl_pool;
5904
5905	ASSERT(spa->spa_sync_pass == 1);
5906
5907	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5908	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5909		dsl_pool_create_origin(dp, tx);
5910
5911		/* Keeping the origin open increases spa_minref */
5912		spa->spa_minref += 3;
5913	}
5914
5915	if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5916	    spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5917		dsl_pool_upgrade_clones(dp, tx);
5918	}
5919
5920	if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5921	    spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5922		dsl_pool_upgrade_dir_clones(dp, tx);
5923
5924		/* Keeping the freedir open increases spa_minref */
5925		spa->spa_minref += 3;
5926	}
5927
5928	if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
5929	    spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
5930		spa_feature_create_zap_objects(spa, tx);
5931	}
5932}
5933
5934/*
5935 * Sync the specified transaction group.  New blocks may be dirtied as
5936 * part of the process, so we iterate until it converges.
5937 */
5938void
5939spa_sync(spa_t *spa, uint64_t txg)
5940{
5941	dsl_pool_t *dp = spa->spa_dsl_pool;
5942	objset_t *mos = spa->spa_meta_objset;
5943	bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5944	bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5945	vdev_t *rvd = spa->spa_root_vdev;
5946	vdev_t *vd;
5947	dmu_tx_t *tx;
5948	int error;
5949
5950	VERIFY(spa_writeable(spa));
5951
5952	/*
5953	 * Lock out configuration changes.
5954	 */
5955	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5956
5957	spa->spa_syncing_txg = txg;
5958	spa->spa_sync_pass = 0;
5959
5960	/*
5961	 * If there are any pending vdev state changes, convert them
5962	 * into config changes that go out with this transaction group.
5963	 */
5964	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5965	while (list_head(&spa->spa_state_dirty_list) != NULL) {
5966		/*
5967		 * We need the write lock here because, for aux vdevs,
5968		 * calling vdev_config_dirty() modifies sav_config.
5969		 * This is ugly and will become unnecessary when we
5970		 * eliminate the aux vdev wart by integrating all vdevs
5971		 * into the root vdev tree.
5972		 */
5973		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5974		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5975		while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5976			vdev_state_clean(vd);
5977			vdev_config_dirty(vd);
5978		}
5979		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5980		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5981	}
5982	spa_config_exit(spa, SCL_STATE, FTAG);
5983
5984	tx = dmu_tx_create_assigned(dp, txg);
5985
5986	/*
5987	 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5988	 * set spa_deflate if we have no raid-z vdevs.
5989	 */
5990	if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5991	    spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5992		int i;
5993
5994		for (i = 0; i < rvd->vdev_children; i++) {
5995			vd = rvd->vdev_child[i];
5996			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5997				break;
5998		}
5999		if (i == rvd->vdev_children) {
6000			spa->spa_deflate = TRUE;
6001			VERIFY(0 == zap_add(spa->spa_meta_objset,
6002			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6003			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6004		}
6005	}
6006
6007	/*
6008	 * If anything has changed in this txg, or if someone is waiting
6009	 * for this txg to sync (eg, spa_vdev_remove()), push the
6010	 * deferred frees from the previous txg.  If not, leave them
6011	 * alone so that we don't generate work on an otherwise idle
6012	 * system.
6013	 */
6014	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6015	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6016	    !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6017	    ((dsl_scan_active(dp->dp_scan) ||
6018	    txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6019		zio_t *zio = zio_root(spa, NULL, NULL, 0);
6020		VERIFY3U(bpobj_iterate(defer_bpo,
6021		    spa_free_sync_cb, zio, tx), ==, 0);
6022		VERIFY0(zio_wait(zio));
6023	}
6024
6025	/*
6026	 * Iterate to convergence.
6027	 */
6028	do {
6029		int pass = ++spa->spa_sync_pass;
6030
6031		spa_sync_config_object(spa, tx);
6032		spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6033		    ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6034		spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6035		    ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6036		spa_errlog_sync(spa, txg);
6037		dsl_pool_sync(dp, txg);
6038
6039		if (pass <= SYNC_PASS_DEFERRED_FREE) {
6040			zio_t *zio = zio_root(spa, NULL, NULL, 0);
6041			bplist_iterate(free_bpl, spa_free_sync_cb,
6042			    zio, tx);
6043			VERIFY(zio_wait(zio) == 0);
6044		} else {
6045			bplist_iterate(free_bpl, bpobj_enqueue_cb,
6046			    defer_bpo, tx);
6047		}
6048
6049		ddt_sync(spa, txg);
6050		dsl_scan_sync(dp, tx);
6051
6052		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6053			vdev_sync(vd, txg);
6054
6055		if (pass == 1)
6056			spa_sync_upgrades(spa, tx);
6057
6058	} while (dmu_objset_is_dirty(mos, txg));
6059
6060	/*
6061	 * Rewrite the vdev configuration (which includes the uberblock)
6062	 * to commit the transaction group.
6063	 *
6064	 * If there are no dirty vdevs, we sync the uberblock to a