spa.c revision b420f3adeb349714478d1a7813d2c0e069d41555
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 boolean_t spa_has_active_shared_spare(spa_t *spa);
120static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
121    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
122    char **ereport);
123static void spa_vdev_resilver_done(spa_t *spa);
124
125uint_t		zio_taskq_batch_pct = 100;	/* 1 thread per cpu in pset */
126id_t		zio_taskq_psrset_bind = PS_NONE;
127boolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
128uint_t		zio_taskq_basedc = 80;		/* base duty cycle */
129
130boolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
131
132/*
133 * This (illegal) pool name is used when temporarily importing a spa_t in order
134 * to get the vdev stats associated with the imported devices.
135 */
136#define	TRYIMPORT_NAME	"$import"
137
138/*
139 * ==========================================================================
140 * SPA properties routines
141 * ==========================================================================
142 */
143
144/*
145 * Add a (source=src, propname=propval) list to an nvlist.
146 */
147static void
148spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
149    uint64_t intval, zprop_source_t src)
150{
151	const char *propname = zpool_prop_to_name(prop);
152	nvlist_t *propval;
153
154	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
155	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
156
157	if (strval != NULL)
158		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
159	else
160		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
161
162	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
163	nvlist_free(propval);
164}
165
166/*
167 * Get property values from the spa configuration.
168 */
169static void
170spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
171{
172	vdev_t *rvd = spa->spa_root_vdev;
173	dsl_pool_t *pool = spa->spa_dsl_pool;
174	uint64_t size;
175	uint64_t alloc;
176	uint64_t space;
177	uint64_t cap, version;
178	zprop_source_t src = ZPROP_SRC_NONE;
179	spa_config_dirent_t *dp;
180
181	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
182
183	if (rvd != NULL) {
184		alloc = metaslab_class_get_alloc(spa_normal_class(spa));
185		size = metaslab_class_get_space(spa_normal_class(spa));
186		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
187		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
188		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
189		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
190		    size - alloc, src);
191
192		space = 0;
193		for (int c = 0; c < rvd->vdev_children; c++) {
194			vdev_t *tvd = rvd->vdev_child[c];
195			space += tvd->vdev_max_asize - tvd->vdev_asize;
196		}
197		spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
198		    src);
199
200		spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
201		    (spa_mode(spa) == FREAD), src);
202
203		cap = (size == 0) ? 0 : (alloc * 100 / size);
204		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
205
206		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
207		    ddt_get_pool_dedup_ratio(spa), src);
208
209		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
210		    rvd->vdev_state, src);
211
212		version = spa_version(spa);
213		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
214			src = ZPROP_SRC_DEFAULT;
215		else
216			src = ZPROP_SRC_LOCAL;
217		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
218	}
219
220	if (pool != NULL) {
221		dsl_dir_t *freedir = pool->dp_free_dir;
222
223		/*
224		 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
225		 * when opening pools before this version freedir will be NULL.
226		 */
227		if (freedir != NULL) {
228			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
229			    freedir->dd_phys->dd_used_bytes, src);
230		} else {
231			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
232			    NULL, 0, src);
233		}
234	}
235
236	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
237
238	if (spa->spa_comment != NULL) {
239		spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
240		    0, ZPROP_SRC_LOCAL);
241	}
242
243	if (spa->spa_root != NULL)
244		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
245		    0, ZPROP_SRC_LOCAL);
246
247	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
248		if (dp->scd_path == NULL) {
249			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
250			    "none", 0, ZPROP_SRC_LOCAL);
251		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
252			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
253			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
254		}
255	}
256}
257
258/*
259 * Get zpool property values.
260 */
261int
262spa_prop_get(spa_t *spa, nvlist_t **nvp)
263{
264	objset_t *mos = spa->spa_meta_objset;
265	zap_cursor_t zc;
266	zap_attribute_t za;
267	int err;
268
269	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
270
271	mutex_enter(&spa->spa_props_lock);
272
273	/*
274	 * Get properties from the spa config.
275	 */
276	spa_prop_get_config(spa, nvp);
277
278	/* If no pool property object, no more prop to get. */
279	if (mos == NULL || spa->spa_pool_props_object == 0) {
280		mutex_exit(&spa->spa_props_lock);
281		return (0);
282	}
283
284	/*
285	 * Get properties from the MOS pool property object.
286	 */
287	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
288	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
289	    zap_cursor_advance(&zc)) {
290		uint64_t intval = 0;
291		char *strval = NULL;
292		zprop_source_t src = ZPROP_SRC_DEFAULT;
293		zpool_prop_t prop;
294
295		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
296			continue;
297
298		switch (za.za_integer_length) {
299		case 8:
300			/* integer property */
301			if (za.za_first_integer !=
302			    zpool_prop_default_numeric(prop))
303				src = ZPROP_SRC_LOCAL;
304
305			if (prop == ZPOOL_PROP_BOOTFS) {
306				dsl_pool_t *dp;
307				dsl_dataset_t *ds = NULL;
308
309				dp = spa_get_dsl(spa);
310				rw_enter(&dp->dp_config_rwlock, RW_READER);
311				if (err = dsl_dataset_hold_obj(dp,
312				    za.za_first_integer, FTAG, &ds)) {
313					rw_exit(&dp->dp_config_rwlock);
314					break;
315				}
316
317				strval = kmem_alloc(
318				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
319				    KM_SLEEP);
320				dsl_dataset_name(ds, strval);
321				dsl_dataset_rele(ds, FTAG);
322				rw_exit(&dp->dp_config_rwlock);
323			} else {
324				strval = NULL;
325				intval = za.za_first_integer;
326			}
327
328			spa_prop_add_list(*nvp, prop, strval, intval, src);
329
330			if (strval != NULL)
331				kmem_free(strval,
332				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
333
334			break;
335
336		case 1:
337			/* string property */
338			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
339			err = zap_lookup(mos, spa->spa_pool_props_object,
340			    za.za_name, 1, za.za_num_integers, strval);
341			if (err) {
342				kmem_free(strval, za.za_num_integers);
343				break;
344			}
345			spa_prop_add_list(*nvp, prop, strval, 0, src);
346			kmem_free(strval, za.za_num_integers);
347			break;
348
349		default:
350			break;
351		}
352	}
353	zap_cursor_fini(&zc);
354	mutex_exit(&spa->spa_props_lock);
355out:
356	if (err && err != ENOENT) {
357		nvlist_free(*nvp);
358		*nvp = NULL;
359		return (err);
360	}
361
362	return (0);
363}
364
365/*
366 * Validate the given pool properties nvlist and modify the list
367 * for the property values to be set.
368 */
369static int
370spa_prop_validate(spa_t *spa, nvlist_t *props)
371{
372	nvpair_t *elem;
373	int error = 0, reset_bootfs = 0;
374	uint64_t objnum;
375	boolean_t has_feature = B_FALSE;
376
377	elem = NULL;
378	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
379		uint64_t intval;
380		char *strval, *slash, *check, *fname;
381		const char *propname = nvpair_name(elem);
382		zpool_prop_t prop = zpool_name_to_prop(propname);
383
384		switch (prop) {
385		case ZPROP_INVAL:
386			if (!zpool_prop_feature(propname)) {
387				error = EINVAL;
388				break;
389			}
390
391			/*
392			 * Sanitize the input.
393			 */
394			if (nvpair_type(elem) != DATA_TYPE_UINT64) {
395				error = EINVAL;
396				break;
397			}
398
399			if (nvpair_value_uint64(elem, &intval) != 0) {
400				error = EINVAL;
401				break;
402			}
403
404			if (intval != 0) {
405				error = EINVAL;
406				break;
407			}
408
409			fname = strchr(propname, '@') + 1;
410			if (zfeature_lookup_name(fname, NULL) != 0) {
411				error = EINVAL;
412				break;
413			}
414
415			has_feature = B_TRUE;
416			break;
417
418		case ZPOOL_PROP_VERSION:
419			error = nvpair_value_uint64(elem, &intval);
420			if (!error &&
421			    (intval < spa_version(spa) ||
422			    intval > SPA_VERSION_BEFORE_FEATURES ||
423			    has_feature))
424				error = EINVAL;
425			break;
426
427		case ZPOOL_PROP_DELEGATION:
428		case ZPOOL_PROP_AUTOREPLACE:
429		case ZPOOL_PROP_LISTSNAPS:
430		case ZPOOL_PROP_AUTOEXPAND:
431			error = nvpair_value_uint64(elem, &intval);
432			if (!error && intval > 1)
433				error = EINVAL;
434			break;
435
436		case ZPOOL_PROP_BOOTFS:
437			/*
438			 * If the pool version is less than SPA_VERSION_BOOTFS,
439			 * or the pool is still being created (version == 0),
440			 * the bootfs property cannot be set.
441			 */
442			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
443				error = ENOTSUP;
444				break;
445			}
446
447			/*
448			 * Make sure the vdev config is bootable
449			 */
450			if (!vdev_is_bootable(spa->spa_root_vdev)) {
451				error = ENOTSUP;
452				break;
453			}
454
455			reset_bootfs = 1;
456
457			error = nvpair_value_string(elem, &strval);
458
459			if (!error) {
460				objset_t *os;
461				uint64_t compress;
462
463				if (strval == NULL || strval[0] == '\0') {
464					objnum = zpool_prop_default_numeric(
465					    ZPOOL_PROP_BOOTFS);
466					break;
467				}
468
469				if (error = dmu_objset_hold(strval, FTAG, &os))
470					break;
471
472				/* Must be ZPL and not gzip compressed. */
473
474				if (dmu_objset_type(os) != DMU_OST_ZFS) {
475					error = ENOTSUP;
476				} else if ((error = dsl_prop_get_integer(strval,
477				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
478				    &compress, NULL)) == 0 &&
479				    !BOOTFS_COMPRESS_VALID(compress)) {
480					error = ENOTSUP;
481				} else {
482					objnum = dmu_objset_id(os);
483				}
484				dmu_objset_rele(os, FTAG);
485			}
486			break;
487
488		case ZPOOL_PROP_FAILUREMODE:
489			error = nvpair_value_uint64(elem, &intval);
490			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
491			    intval > ZIO_FAILURE_MODE_PANIC))
492				error = EINVAL;
493
494			/*
495			 * This is a special case which only occurs when
496			 * the pool has completely failed. This allows
497			 * the user to change the in-core failmode property
498			 * without syncing it out to disk (I/Os might
499			 * currently be blocked). We do this by returning
500			 * EIO to the caller (spa_prop_set) to trick it
501			 * into thinking we encountered a property validation
502			 * error.
503			 */
504			if (!error && spa_suspended(spa)) {
505				spa->spa_failmode = intval;
506				error = EIO;
507			}
508			break;
509
510		case ZPOOL_PROP_CACHEFILE:
511			if ((error = nvpair_value_string(elem, &strval)) != 0)
512				break;
513
514			if (strval[0] == '\0')
515				break;
516
517			if (strcmp(strval, "none") == 0)
518				break;
519
520			if (strval[0] != '/') {
521				error = EINVAL;
522				break;
523			}
524
525			slash = strrchr(strval, '/');
526			ASSERT(slash != NULL);
527
528			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
529			    strcmp(slash, "/..") == 0)
530				error = EINVAL;
531			break;
532
533		case ZPOOL_PROP_COMMENT:
534			if ((error = nvpair_value_string(elem, &strval)) != 0)
535				break;
536			for (check = strval; *check != '\0'; check++) {
537				/*
538				 * The kernel doesn't have an easy isprint()
539				 * check.  For this kernel check, we merely
540				 * check ASCII apart from DEL.  Fix this if
541				 * there is an easy-to-use kernel isprint().
542				 */
543				if (*check >= 0x7f) {
544					error = EINVAL;
545					break;
546				}
547				check++;
548			}
549			if (strlen(strval) > ZPROP_MAX_COMMENT)
550				error = E2BIG;
551			break;
552
553		case ZPOOL_PROP_DEDUPDITTO:
554			if (spa_version(spa) < SPA_VERSION_DEDUP)
555				error = ENOTSUP;
556			else
557				error = nvpair_value_uint64(elem, &intval);
558			if (error == 0 &&
559			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
560				error = EINVAL;
561			break;
562		}
563
564		if (error)
565			break;
566	}
567
568	if (!error && reset_bootfs) {
569		error = nvlist_remove(props,
570		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
571
572		if (!error) {
573			error = nvlist_add_uint64(props,
574			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
575		}
576	}
577
578	return (error);
579}
580
581void
582spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
583{
584	char *cachefile;
585	spa_config_dirent_t *dp;
586
587	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
588	    &cachefile) != 0)
589		return;
590
591	dp = kmem_alloc(sizeof (spa_config_dirent_t),
592	    KM_SLEEP);
593
594	if (cachefile[0] == '\0')
595		dp->scd_path = spa_strdup(spa_config_path);
596	else if (strcmp(cachefile, "none") == 0)
597		dp->scd_path = NULL;
598	else
599		dp->scd_path = spa_strdup(cachefile);
600
601	list_insert_head(&spa->spa_config_list, dp);
602	if (need_sync)
603		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
604}
605
606int
607spa_prop_set(spa_t *spa, nvlist_t *nvp)
608{
609	int error;
610	nvpair_t *elem = NULL;
611	boolean_t need_sync = B_FALSE;
612
613	if ((error = spa_prop_validate(spa, nvp)) != 0)
614		return (error);
615
616	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
617		zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
618
619		if (prop == ZPOOL_PROP_CACHEFILE ||
620		    prop == ZPOOL_PROP_ALTROOT ||
621		    prop == ZPOOL_PROP_READONLY)
622			continue;
623
624		if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
625			uint64_t ver;
626
627			if (prop == ZPOOL_PROP_VERSION) {
628				VERIFY(nvpair_value_uint64(elem, &ver) == 0);
629			} else {
630				ASSERT(zpool_prop_feature(nvpair_name(elem)));
631				ver = SPA_VERSION_FEATURES;
632				need_sync = B_TRUE;
633			}
634
635			/* Save time if the version is already set. */
636			if (ver == spa_version(spa))
637				continue;
638
639			/*
640			 * In addition to the pool directory object, we might
641			 * create the pool properties object, the features for
642			 * read object, the features for write object, or the
643			 * feature descriptions object.
644			 */
645			error = dsl_sync_task_do(spa_get_dsl(spa), NULL,
646			    spa_sync_version, spa, &ver, 6);
647			if (error)
648				return (error);
649			continue;
650		}
651
652		need_sync = B_TRUE;
653		break;
654	}
655
656	if (need_sync) {
657		return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
658		    spa, nvp, 6));
659	}
660
661	return (0);
662}
663
664/*
665 * If the bootfs property value is dsobj, clear it.
666 */
667void
668spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
669{
670	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
671		VERIFY(zap_remove(spa->spa_meta_objset,
672		    spa->spa_pool_props_object,
673		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
674		spa->spa_bootfs = 0;
675	}
676}
677
678/*
679 * Change the GUID for the pool.  This is done so that we can later
680 * re-import a pool built from a clone of our own vdevs.  We will modify
681 * the root vdev's guid, our own pool guid, and then mark all of our
682 * vdevs dirty.  Note that we must make sure that all our vdevs are
683 * online when we do this, or else any vdevs that weren't present
684 * would be orphaned from our pool.  We are also going to issue a
685 * sysevent to update any watchers.
686 */
687int
688spa_change_guid(spa_t *spa)
689{
690	uint64_t	oldguid, newguid;
691	uint64_t	txg;
692
693	if (!(spa_mode_global & FWRITE))
694		return (EROFS);
695
696	txg = spa_vdev_enter(spa);
697
698	if (spa->spa_root_vdev->vdev_state != VDEV_STATE_HEALTHY)
699		return (spa_vdev_exit(spa, NULL, txg, ENXIO));
700
701	oldguid = spa_guid(spa);
702	newguid = spa_generate_guid(NULL);
703	ASSERT3U(oldguid, !=, newguid);
704
705	spa->spa_root_vdev->vdev_guid = newguid;
706	spa->spa_root_vdev->vdev_guid_sum += (newguid - oldguid);
707
708	vdev_config_dirty(spa->spa_root_vdev);
709
710	spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
711
712	return (spa_vdev_exit(spa, NULL, txg, 0));
713}
714
715/*
716 * ==========================================================================
717 * SPA state manipulation (open/create/destroy/import/export)
718 * ==========================================================================
719 */
720
721static int
722spa_error_entry_compare(const void *a, const void *b)
723{
724	spa_error_entry_t *sa = (spa_error_entry_t *)a;
725	spa_error_entry_t *sb = (spa_error_entry_t *)b;
726	int ret;
727
728	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
729	    sizeof (zbookmark_t));
730
731	if (ret < 0)
732		return (-1);
733	else if (ret > 0)
734		return (1);
735	else
736		return (0);
737}
738
739/*
740 * Utility function which retrieves copies of the current logs and
741 * re-initializes them in the process.
742 */
743void
744spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
745{
746	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
747
748	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
749	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
750
751	avl_create(&spa->spa_errlist_scrub,
752	    spa_error_entry_compare, sizeof (spa_error_entry_t),
753	    offsetof(spa_error_entry_t, se_avl));
754	avl_create(&spa->spa_errlist_last,
755	    spa_error_entry_compare, sizeof (spa_error_entry_t),
756	    offsetof(spa_error_entry_t, se_avl));
757}
758
759static taskq_t *
760spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
761    uint_t value)
762{
763	uint_t flags = 0;
764	boolean_t batch = B_FALSE;
765
766	switch (mode) {
767	case zti_mode_null:
768		return (NULL);		/* no taskq needed */
769
770	case zti_mode_fixed:
771		ASSERT3U(value, >=, 1);
772		value = MAX(value, 1);
773		break;
774
775	case zti_mode_batch:
776		batch = B_TRUE;
777		flags |= TASKQ_THREADS_CPU_PCT;
778		value = zio_taskq_batch_pct;
779		break;
780
781	case zti_mode_online_percent:
782		flags |= TASKQ_THREADS_CPU_PCT;
783		break;
784
785	default:
786		panic("unrecognized mode for %s taskq (%u:%u) in "
787		    "spa_activate()",
788		    name, mode, value);
789		break;
790	}
791
792	if (zio_taskq_sysdc && spa->spa_proc != &p0) {
793		if (batch)
794			flags |= TASKQ_DC_BATCH;
795
796		return (taskq_create_sysdc(name, value, 50, INT_MAX,
797		    spa->spa_proc, zio_taskq_basedc, flags));
798	}
799	return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
800	    spa->spa_proc, flags));
801}
802
803static void
804spa_create_zio_taskqs(spa_t *spa)
805{
806	for (int t = 0; t < ZIO_TYPES; t++) {
807		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
808			const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
809			enum zti_modes mode = ztip->zti_mode;
810			uint_t value = ztip->zti_value;
811			char name[32];
812
813			(void) snprintf(name, sizeof (name),
814			    "%s_%s", zio_type_name[t], zio_taskq_types[q]);
815
816			spa->spa_zio_taskq[t][q] =
817			    spa_taskq_create(spa, name, mode, value);
818		}
819	}
820}
821
822#ifdef _KERNEL
823static void
824spa_thread(void *arg)
825{
826	callb_cpr_t cprinfo;
827
828	spa_t *spa = arg;
829	user_t *pu = PTOU(curproc);
830
831	CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
832	    spa->spa_name);
833
834	ASSERT(curproc != &p0);
835	(void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
836	    "zpool-%s", spa->spa_name);
837	(void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
838
839	/* bind this thread to the requested psrset */
840	if (zio_taskq_psrset_bind != PS_NONE) {
841		pool_lock();
842		mutex_enter(&cpu_lock);
843		mutex_enter(&pidlock);
844		mutex_enter(&curproc->p_lock);
845
846		if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
847		    0, NULL, NULL) == 0)  {
848			curthread->t_bind_pset = zio_taskq_psrset_bind;
849		} else {
850			cmn_err(CE_WARN,
851			    "Couldn't bind process for zfs pool \"%s\" to "
852			    "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
853		}
854
855		mutex_exit(&curproc->p_lock);
856		mutex_exit(&pidlock);
857		mutex_exit(&cpu_lock);
858		pool_unlock();
859	}
860
861	if (zio_taskq_sysdc) {
862		sysdc_thread_enter(curthread, 100, 0);
863	}
864
865	spa->spa_proc = curproc;
866	spa->spa_did = curthread->t_did;
867
868	spa_create_zio_taskqs(spa);
869
870	mutex_enter(&spa->spa_proc_lock);
871	ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
872
873	spa->spa_proc_state = SPA_PROC_ACTIVE;
874	cv_broadcast(&spa->spa_proc_cv);
875
876	CALLB_CPR_SAFE_BEGIN(&cprinfo);
877	while (spa->spa_proc_state == SPA_PROC_ACTIVE)
878		cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
879	CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
880
881	ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
882	spa->spa_proc_state = SPA_PROC_GONE;
883	spa->spa_proc = &p0;
884	cv_broadcast(&spa->spa_proc_cv);
885	CALLB_CPR_EXIT(&cprinfo);	/* drops spa_proc_lock */
886
887	mutex_enter(&curproc->p_lock);
888	lwp_exit();
889}
890#endif
891
892/*
893 * Activate an uninitialized pool.
894 */
895static void
896spa_activate(spa_t *spa, int mode)
897{
898	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
899
900	spa->spa_state = POOL_STATE_ACTIVE;
901	spa->spa_mode = mode;
902
903	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
904	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
905
906	/* Try to create a covering process */
907	mutex_enter(&spa->spa_proc_lock);
908	ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
909	ASSERT(spa->spa_proc == &p0);
910	spa->spa_did = 0;
911
912	/* Only create a process if we're going to be around a while. */
913	if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
914		if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
915		    NULL, 0) == 0) {
916			spa->spa_proc_state = SPA_PROC_CREATED;
917			while (spa->spa_proc_state == SPA_PROC_CREATED) {
918				cv_wait(&spa->spa_proc_cv,
919				    &spa->spa_proc_lock);
920			}
921			ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
922			ASSERT(spa->spa_proc != &p0);
923			ASSERT(spa->spa_did != 0);
924		} else {
925#ifdef _KERNEL
926			cmn_err(CE_WARN,
927			    "Couldn't create process for zfs pool \"%s\"\n",
928			    spa->spa_name);
929#endif
930		}
931	}
932	mutex_exit(&spa->spa_proc_lock);
933
934	/* If we didn't create a process, we need to create our taskqs. */
935	if (spa->spa_proc == &p0) {
936		spa_create_zio_taskqs(spa);
937	}
938
939	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
940	    offsetof(vdev_t, vdev_config_dirty_node));
941	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
942	    offsetof(vdev_t, vdev_state_dirty_node));
943
944	txg_list_create(&spa->spa_vdev_txg_list,
945	    offsetof(struct vdev, vdev_txg_node));
946
947	avl_create(&spa->spa_errlist_scrub,
948	    spa_error_entry_compare, sizeof (spa_error_entry_t),
949	    offsetof(spa_error_entry_t, se_avl));
950	avl_create(&spa->spa_errlist_last,
951	    spa_error_entry_compare, sizeof (spa_error_entry_t),
952	    offsetof(spa_error_entry_t, se_avl));
953}
954
955/*
956 * Opposite of spa_activate().
957 */
958static void
959spa_deactivate(spa_t *spa)
960{
961	ASSERT(spa->spa_sync_on == B_FALSE);
962	ASSERT(spa->spa_dsl_pool == NULL);
963	ASSERT(spa->spa_root_vdev == NULL);
964	ASSERT(spa->spa_async_zio_root == NULL);
965	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
966
967	txg_list_destroy(&spa->spa_vdev_txg_list);
968
969	list_destroy(&spa->spa_config_dirty_list);
970	list_destroy(&spa->spa_state_dirty_list);
971
972	for (int t = 0; t < ZIO_TYPES; t++) {
973		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
974			if (spa->spa_zio_taskq[t][q] != NULL)
975				taskq_destroy(spa->spa_zio_taskq[t][q]);
976			spa->spa_zio_taskq[t][q] = NULL;
977		}
978	}
979
980	metaslab_class_destroy(spa->spa_normal_class);
981	spa->spa_normal_class = NULL;
982
983	metaslab_class_destroy(spa->spa_log_class);
984	spa->spa_log_class = NULL;
985
986	/*
987	 * If this was part of an import or the open otherwise failed, we may
988	 * still have errors left in the queues.  Empty them just in case.
989	 */
990	spa_errlog_drain(spa);
991
992	avl_destroy(&spa->spa_errlist_scrub);
993	avl_destroy(&spa->spa_errlist_last);
994
995	spa->spa_state = POOL_STATE_UNINITIALIZED;
996
997	mutex_enter(&spa->spa_proc_lock);
998	if (spa->spa_proc_state != SPA_PROC_NONE) {
999		ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1000		spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1001		cv_broadcast(&spa->spa_proc_cv);
1002		while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1003			ASSERT(spa->spa_proc != &p0);
1004			cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1005		}
1006		ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1007		spa->spa_proc_state = SPA_PROC_NONE;
1008	}
1009	ASSERT(spa->spa_proc == &p0);
1010	mutex_exit(&spa->spa_proc_lock);
1011
1012	/*
1013	 * We want to make sure spa_thread() has actually exited the ZFS
1014	 * module, so that the module can't be unloaded out from underneath
1015	 * it.
1016	 */
1017	if (spa->spa_did != 0) {
1018		thread_join(spa->spa_did);
1019		spa->spa_did = 0;
1020	}
1021}
1022
1023/*
1024 * Verify a pool configuration, and construct the vdev tree appropriately.  This
1025 * will create all the necessary vdevs in the appropriate layout, with each vdev
1026 * in the CLOSED state.  This will prep the pool before open/creation/import.
1027 * All vdev validation is done by the vdev_alloc() routine.
1028 */
1029static int
1030spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1031    uint_t id, int atype)
1032{
1033	nvlist_t **child;
1034	uint_t children;
1035	int error;
1036
1037	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1038		return (error);
1039
1040	if ((*vdp)->vdev_ops->vdev_op_leaf)
1041		return (0);
1042
1043	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1044	    &child, &children);
1045
1046	if (error == ENOENT)
1047		return (0);
1048
1049	if (error) {
1050		vdev_free(*vdp);
1051		*vdp = NULL;
1052		return (EINVAL);
1053	}
1054
1055	for (int c = 0; c < children; c++) {
1056		vdev_t *vd;
1057		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1058		    atype)) != 0) {
1059			vdev_free(*vdp);
1060			*vdp = NULL;
1061			return (error);
1062		}
1063	}
1064
1065	ASSERT(*vdp != NULL);
1066
1067	return (0);
1068}
1069
1070/*
1071 * Opposite of spa_load().
1072 */
1073static void
1074spa_unload(spa_t *spa)
1075{
1076	int i;
1077
1078	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1079
1080	/*
1081	 * Stop async tasks.
1082	 */
1083	spa_async_suspend(spa);
1084
1085	/*
1086	 * Stop syncing.
1087	 */
1088	if (spa->spa_sync_on) {
1089		txg_sync_stop(spa->spa_dsl_pool);
1090		spa->spa_sync_on = B_FALSE;
1091	}
1092
1093	/*
1094	 * Wait for any outstanding async I/O to complete.
1095	 */
1096	if (spa->spa_async_zio_root != NULL) {
1097		(void) zio_wait(spa->spa_async_zio_root);
1098		spa->spa_async_zio_root = NULL;
1099	}
1100
1101	bpobj_close(&spa->spa_deferred_bpobj);
1102
1103	/*
1104	 * Close the dsl pool.
1105	 */
1106	if (spa->spa_dsl_pool) {
1107		dsl_pool_close(spa->spa_dsl_pool);
1108		spa->spa_dsl_pool = NULL;
1109		spa->spa_meta_objset = NULL;
1110	}
1111
1112	ddt_unload(spa);
1113
1114	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1115
1116	/*
1117	 * Drop and purge level 2 cache
1118	 */
1119	spa_l2cache_drop(spa);
1120
1121	/*
1122	 * Close all vdevs.
1123	 */
1124	if (spa->spa_root_vdev)
1125		vdev_free(spa->spa_root_vdev);
1126	ASSERT(spa->spa_root_vdev == NULL);
1127
1128	for (i = 0; i < spa->spa_spares.sav_count; i++)
1129		vdev_free(spa->spa_spares.sav_vdevs[i]);
1130	if (spa->spa_spares.sav_vdevs) {
1131		kmem_free(spa->spa_spares.sav_vdevs,
1132		    spa->spa_spares.sav_count * sizeof (void *));
1133		spa->spa_spares.sav_vdevs = NULL;
1134	}
1135	if (spa->spa_spares.sav_config) {
1136		nvlist_free(spa->spa_spares.sav_config);
1137		spa->spa_spares.sav_config = NULL;
1138	}
1139	spa->spa_spares.sav_count = 0;
1140
1141	for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1142		vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1143		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1144	}
1145	if (spa->spa_l2cache.sav_vdevs) {
1146		kmem_free(spa->spa_l2cache.sav_vdevs,
1147		    spa->spa_l2cache.sav_count * sizeof (void *));
1148		spa->spa_l2cache.sav_vdevs = NULL;
1149	}
1150	if (spa->spa_l2cache.sav_config) {
1151		nvlist_free(spa->spa_l2cache.sav_config);
1152		spa->spa_l2cache.sav_config = NULL;
1153	}
1154	spa->spa_l2cache.sav_count = 0;
1155
1156	spa->spa_async_suspended = 0;
1157
1158	if (spa->spa_comment != NULL) {
1159		spa_strfree(spa->spa_comment);
1160		spa->spa_comment = NULL;
1161	}
1162
1163	spa_config_exit(spa, SCL_ALL, FTAG);
1164}
1165
1166/*
1167 * Load (or re-load) the current list of vdevs describing the active spares for
1168 * this pool.  When this is called, we have some form of basic information in
1169 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1170 * then re-generate a more complete list including status information.
1171 */
1172static void
1173spa_load_spares(spa_t *spa)
1174{
1175	nvlist_t **spares;
1176	uint_t nspares;
1177	int i;
1178	vdev_t *vd, *tvd;
1179
1180	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1181
1182	/*
1183	 * First, close and free any existing spare vdevs.
1184	 */
1185	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1186		vd = spa->spa_spares.sav_vdevs[i];
1187
1188		/* Undo the call to spa_activate() below */
1189		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1190		    B_FALSE)) != NULL && tvd->vdev_isspare)
1191			spa_spare_remove(tvd);
1192		vdev_close(vd);
1193		vdev_free(vd);
1194	}
1195
1196	if (spa->spa_spares.sav_vdevs)
1197		kmem_free(spa->spa_spares.sav_vdevs,
1198		    spa->spa_spares.sav_count * sizeof (void *));
1199
1200	if (spa->spa_spares.sav_config == NULL)
1201		nspares = 0;
1202	else
1203		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1204		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1205
1206	spa->spa_spares.sav_count = (int)nspares;
1207	spa->spa_spares.sav_vdevs = NULL;
1208
1209	if (nspares == 0)
1210		return;
1211
1212	/*
1213	 * Construct the array of vdevs, opening them to get status in the
1214	 * process.   For each spare, there is potentially two different vdev_t
1215	 * structures associated with it: one in the list of spares (used only
1216	 * for basic validation purposes) and one in the active vdev
1217	 * configuration (if it's spared in).  During this phase we open and
1218	 * validate each vdev on the spare list.  If the vdev also exists in the
1219	 * active configuration, then we also mark this vdev as an active spare.
1220	 */
1221	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1222	    KM_SLEEP);
1223	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1224		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1225		    VDEV_ALLOC_SPARE) == 0);
1226		ASSERT(vd != NULL);
1227
1228		spa->spa_spares.sav_vdevs[i] = vd;
1229
1230		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1231		    B_FALSE)) != NULL) {
1232			if (!tvd->vdev_isspare)
1233				spa_spare_add(tvd);
1234
1235			/*
1236			 * We only mark the spare active if we were successfully
1237			 * able to load the vdev.  Otherwise, importing a pool
1238			 * with a bad active spare would result in strange
1239			 * behavior, because multiple pool would think the spare
1240			 * is actively in use.
1241			 *
1242			 * There is a vulnerability here to an equally bizarre
1243			 * circumstance, where a dead active spare is later
1244			 * brought back to life (onlined or otherwise).  Given
1245			 * the rarity of this scenario, and the extra complexity
1246			 * it adds, we ignore the possibility.
1247			 */
1248			if (!vdev_is_dead(tvd))
1249				spa_spare_activate(tvd);
1250		}
1251
1252		vd->vdev_top = vd;
1253		vd->vdev_aux = &spa->spa_spares;
1254
1255		if (vdev_open(vd) != 0)
1256			continue;
1257
1258		if (vdev_validate_aux(vd) == 0)
1259			spa_spare_add(vd);
1260	}
1261
1262	/*
1263	 * Recompute the stashed list of spares, with status information
1264	 * this time.
1265	 */
1266	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1267	    DATA_TYPE_NVLIST_ARRAY) == 0);
1268
1269	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1270	    KM_SLEEP);
1271	for (i = 0; i < spa->spa_spares.sav_count; i++)
1272		spares[i] = vdev_config_generate(spa,
1273		    spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1274	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1275	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1276	for (i = 0; i < spa->spa_spares.sav_count; i++)
1277		nvlist_free(spares[i]);
1278	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1279}
1280
1281/*
1282 * Load (or re-load) the current list of vdevs describing the active l2cache for
1283 * this pool.  When this is called, we have some form of basic information in
1284 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1285 * then re-generate a more complete list including status information.
1286 * Devices which are already active have their details maintained, and are
1287 * not re-opened.
1288 */
1289static void
1290spa_load_l2cache(spa_t *spa)
1291{
1292	nvlist_t **l2cache;
1293	uint_t nl2cache;
1294	int i, j, oldnvdevs;
1295	uint64_t guid;
1296	vdev_t *vd, **oldvdevs, **newvdevs;
1297	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1298
1299	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1300
1301	if (sav->sav_config != NULL) {
1302		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1303		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1304		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1305	} else {
1306		nl2cache = 0;
1307	}
1308
1309	oldvdevs = sav->sav_vdevs;
1310	oldnvdevs = sav->sav_count;
1311	sav->sav_vdevs = NULL;
1312	sav->sav_count = 0;
1313
1314	/*
1315	 * Process new nvlist of vdevs.
1316	 */
1317	for (i = 0; i < nl2cache; i++) {
1318		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1319		    &guid) == 0);
1320
1321		newvdevs[i] = NULL;
1322		for (j = 0; j < oldnvdevs; j++) {
1323			vd = oldvdevs[j];
1324			if (vd != NULL && guid == vd->vdev_guid) {
1325				/*
1326				 * Retain previous vdev for add/remove ops.
1327				 */
1328				newvdevs[i] = vd;
1329				oldvdevs[j] = NULL;
1330				break;
1331			}
1332		}
1333
1334		if (newvdevs[i] == NULL) {
1335			/*
1336			 * Create new vdev
1337			 */
1338			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1339			    VDEV_ALLOC_L2CACHE) == 0);
1340			ASSERT(vd != NULL);
1341			newvdevs[i] = vd;
1342
1343			/*
1344			 * Commit this vdev as an l2cache device,
1345			 * even if it fails to open.
1346			 */
1347			spa_l2cache_add(vd);
1348
1349			vd->vdev_top = vd;
1350			vd->vdev_aux = sav;
1351
1352			spa_l2cache_activate(vd);
1353
1354			if (vdev_open(vd) != 0)
1355				continue;
1356
1357			(void) vdev_validate_aux(vd);
1358
1359			if (!vdev_is_dead(vd))
1360				l2arc_add_vdev(spa, vd);
1361		}
1362	}
1363
1364	/*
1365	 * Purge vdevs that were dropped
1366	 */
1367	for (i = 0; i < oldnvdevs; i++) {
1368		uint64_t pool;
1369
1370		vd = oldvdevs[i];
1371		if (vd != NULL) {
1372			ASSERT(vd->vdev_isl2cache);
1373
1374			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1375			    pool != 0ULL && l2arc_vdev_present(vd))
1376				l2arc_remove_vdev(vd);
1377			vdev_clear_stats(vd);
1378			vdev_free(vd);
1379		}
1380	}
1381
1382	if (oldvdevs)
1383		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1384
1385	if (sav->sav_config == NULL)
1386		goto out;
1387
1388	sav->sav_vdevs = newvdevs;
1389	sav->sav_count = (int)nl2cache;
1390
1391	/*
1392	 * Recompute the stashed list of l2cache devices, with status
1393	 * information this time.
1394	 */
1395	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1396	    DATA_TYPE_NVLIST_ARRAY) == 0);
1397
1398	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1399	for (i = 0; i < sav->sav_count; i++)
1400		l2cache[i] = vdev_config_generate(spa,
1401		    sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1402	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1403	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1404out:
1405	for (i = 0; i < sav->sav_count; i++)
1406		nvlist_free(l2cache[i]);
1407	if (sav->sav_count)
1408		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1409}
1410
1411static int
1412load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1413{
1414	dmu_buf_t *db;
1415	char *packed = NULL;
1416	size_t nvsize = 0;
1417	int error;
1418	*value = NULL;
1419
1420	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1421	nvsize = *(uint64_t *)db->db_data;
1422	dmu_buf_rele(db, FTAG);
1423
1424	packed = kmem_alloc(nvsize, KM_SLEEP);
1425	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1426	    DMU_READ_PREFETCH);
1427	if (error == 0)
1428		error = nvlist_unpack(packed, nvsize, value, 0);
1429	kmem_free(packed, nvsize);
1430
1431	return (error);
1432}
1433
1434/*
1435 * Checks to see if the given vdev could not be opened, in which case we post a
1436 * sysevent to notify the autoreplace code that the device has been removed.
1437 */
1438static void
1439spa_check_removed(vdev_t *vd)
1440{
1441	for (int c = 0; c < vd->vdev_children; c++)
1442		spa_check_removed(vd->vdev_child[c]);
1443
1444	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1445		zfs_post_autoreplace(vd->vdev_spa, vd);
1446		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1447	}
1448}
1449
1450/*
1451 * Validate the current config against the MOS config
1452 */
1453static boolean_t
1454spa_config_valid(spa_t *spa, nvlist_t *config)
1455{
1456	vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1457	nvlist_t *nv;
1458
1459	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1460
1461	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1462	VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1463
1464	ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1465
1466	/*
1467	 * If we're doing a normal import, then build up any additional
1468	 * diagnostic information about missing devices in this config.
1469	 * We'll pass this up to the user for further processing.
1470	 */
1471	if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1472		nvlist_t **child, *nv;
1473		uint64_t idx = 0;
1474
1475		child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1476		    KM_SLEEP);
1477		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1478
1479		for (int c = 0; c < rvd->vdev_children; c++) {
1480			vdev_t *tvd = rvd->vdev_child[c];
1481			vdev_t *mtvd  = mrvd->vdev_child[c];
1482
1483			if (tvd->vdev_ops == &vdev_missing_ops &&
1484			    mtvd->vdev_ops != &vdev_missing_ops &&
1485			    mtvd->vdev_islog)
1486				child[idx++] = vdev_config_generate(spa, mtvd,
1487				    B_FALSE, 0);
1488		}
1489
1490		if (idx) {
1491			VERIFY(nvlist_add_nvlist_array(nv,
1492			    ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1493			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1494			    ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1495
1496			for (int i = 0; i < idx; i++)
1497				nvlist_free(child[i]);
1498		}
1499		nvlist_free(nv);
1500		kmem_free(child, rvd->vdev_children * sizeof (char **));
1501	}
1502
1503	/*
1504	 * Compare the root vdev tree with the information we have
1505	 * from the MOS config (mrvd). Check each top-level vdev
1506	 * with the corresponding MOS config top-level (mtvd).
1507	 */
1508	for (int c = 0; c < rvd->vdev_children; c++) {
1509		vdev_t *tvd = rvd->vdev_child[c];
1510		vdev_t *mtvd  = mrvd->vdev_child[c];
1511
1512		/*
1513		 * Resolve any "missing" vdevs in the current configuration.
1514		 * If we find that the MOS config has more accurate information
1515		 * about the top-level vdev then use that vdev instead.
1516		 */
1517		if (tvd->vdev_ops == &vdev_missing_ops &&
1518		    mtvd->vdev_ops != &vdev_missing_ops) {
1519
1520			if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1521				continue;
1522
1523			/*
1524			 * Device specific actions.
1525			 */
1526			if (mtvd->vdev_islog) {
1527				spa_set_log_state(spa, SPA_LOG_CLEAR);
1528			} else {
1529				/*
1530				 * XXX - once we have 'readonly' pool
1531				 * support we should be able to handle
1532				 * missing data devices by transitioning
1533				 * the pool to readonly.
1534				 */
1535				continue;
1536			}
1537
1538			/*
1539			 * Swap the missing vdev with the data we were
1540			 * able to obtain from the MOS config.
1541			 */
1542			vdev_remove_child(rvd, tvd);
1543			vdev_remove_child(mrvd, mtvd);
1544
1545			vdev_add_child(rvd, mtvd);
1546			vdev_add_child(mrvd, tvd);
1547
1548			spa_config_exit(spa, SCL_ALL, FTAG);
1549			vdev_load(mtvd);
1550			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1551
1552			vdev_reopen(rvd);
1553		} else if (mtvd->vdev_islog) {
1554			/*
1555			 * Load the slog device's state from the MOS config
1556			 * since it's possible that the label does not
1557			 * contain the most up-to-date information.
1558			 */
1559			vdev_load_log_state(tvd, mtvd);
1560			vdev_reopen(tvd);
1561		}
1562	}
1563	vdev_free(mrvd);
1564	spa_config_exit(spa, SCL_ALL, FTAG);
1565
1566	/*
1567	 * Ensure we were able to validate the config.
1568	 */
1569	return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1570}
1571
1572/*
1573 * Check for missing log devices
1574 */
1575static int
1576spa_check_logs(spa_t *spa)
1577{
1578	switch (spa->spa_log_state) {
1579	case SPA_LOG_MISSING:
1580		/* need to recheck in case slog has been restored */
1581	case SPA_LOG_UNKNOWN:
1582		if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1583		    DS_FIND_CHILDREN)) {
1584			spa_set_log_state(spa, SPA_LOG_MISSING);
1585			return (1);
1586		}
1587		break;
1588	}
1589	return (0);
1590}
1591
1592static boolean_t
1593spa_passivate_log(spa_t *spa)
1594{
1595	vdev_t *rvd = spa->spa_root_vdev;
1596	boolean_t slog_found = B_FALSE;
1597
1598	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1599
1600	if (!spa_has_slogs(spa))
1601		return (B_FALSE);
1602
1603	for (int c = 0; c < rvd->vdev_children; c++) {
1604		vdev_t *tvd = rvd->vdev_child[c];
1605		metaslab_group_t *mg = tvd->vdev_mg;
1606
1607		if (tvd->vdev_islog) {
1608			metaslab_group_passivate(mg);
1609			slog_found = B_TRUE;
1610		}
1611	}
1612
1613	return (slog_found);
1614}
1615
1616static void
1617spa_activate_log(spa_t *spa)
1618{
1619	vdev_t *rvd = spa->spa_root_vdev;
1620
1621	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1622
1623	for (int c = 0; c < rvd->vdev_children; c++) {
1624		vdev_t *tvd = rvd->vdev_child[c];
1625		metaslab_group_t *mg = tvd->vdev_mg;
1626
1627		if (tvd->vdev_islog)
1628			metaslab_group_activate(mg);
1629	}
1630}
1631
1632int
1633spa_offline_log(spa_t *spa)
1634{
1635	int error = 0;
1636
1637	if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1638	    NULL, DS_FIND_CHILDREN)) == 0) {
1639
1640		/*
1641		 * We successfully offlined the log device, sync out the
1642		 * current txg so that the "stubby" block can be removed
1643		 * by zil_sync().
1644		 */
1645		txg_wait_synced(spa->spa_dsl_pool, 0);
1646	}
1647	return (error);
1648}
1649
1650static void
1651spa_aux_check_removed(spa_aux_vdev_t *sav)
1652{
1653	for (int i = 0; i < sav->sav_count; i++)
1654		spa_check_removed(sav->sav_vdevs[i]);
1655}
1656
1657void
1658spa_claim_notify(zio_t *zio)
1659{
1660	spa_t *spa = zio->io_spa;
1661
1662	if (zio->io_error)
1663		return;
1664
1665	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1666	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1667		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1668	mutex_exit(&spa->spa_props_lock);
1669}
1670
1671typedef struct spa_load_error {
1672	uint64_t	sle_meta_count;
1673	uint64_t	sle_data_count;
1674} spa_load_error_t;
1675
1676static void
1677spa_load_verify_done(zio_t *zio)
1678{
1679	blkptr_t *bp = zio->io_bp;
1680	spa_load_error_t *sle = zio->io_private;
1681	dmu_object_type_t type = BP_GET_TYPE(bp);
1682	int error = zio->io_error;
1683
1684	if (error) {
1685		if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1686		    type != DMU_OT_INTENT_LOG)
1687			atomic_add_64(&sle->sle_meta_count, 1);
1688		else
1689			atomic_add_64(&sle->sle_data_count, 1);
1690	}
1691	zio_data_buf_free(zio->io_data, zio->io_size);
1692}
1693
1694/*ARGSUSED*/
1695static int
1696spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1697    arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1698{
1699	if (bp != NULL) {
1700		zio_t *rio = arg;
1701		size_t size = BP_GET_PSIZE(bp);
1702		void *data = zio_data_buf_alloc(size);
1703
1704		zio_nowait(zio_read(rio, spa, bp, data, size,
1705		    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1706		    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1707		    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1708	}
1709	return (0);
1710}
1711
1712static int
1713spa_load_verify(spa_t *spa)
1714{
1715	zio_t *rio;
1716	spa_load_error_t sle = { 0 };
1717	zpool_rewind_policy_t policy;
1718	boolean_t verify_ok = B_FALSE;
1719	int error;
1720
1721	zpool_get_rewind_policy(spa->spa_config, &policy);
1722
1723	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1724		return (0);
1725
1726	rio = zio_root(spa, NULL, &sle,
1727	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1728
1729	error = traverse_pool(spa, spa->spa_verify_min_txg,
1730	    TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1731
1732	(void) zio_wait(rio);
1733
1734	spa->spa_load_meta_errors = sle.sle_meta_count;
1735	spa->spa_load_data_errors = sle.sle_data_count;
1736
1737	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1738	    sle.sle_data_count <= policy.zrp_maxdata) {
1739		int64_t loss = 0;
1740
1741		verify_ok = B_TRUE;
1742		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1743		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1744
1745		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1746		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1747		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1748		VERIFY(nvlist_add_int64(spa->spa_load_info,
1749		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1750		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1751		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1752	} else {
1753		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1754	}
1755
1756	if (error) {
1757		if (error != ENXIO && error != EIO)
1758			error = EIO;
1759		return (error);
1760	}
1761
1762	return (verify_ok ? 0 : EIO);
1763}
1764
1765/*
1766 * Find a value in the pool props object.
1767 */
1768static void
1769spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1770{
1771	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1772	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1773}
1774
1775/*
1776 * Find a value in the pool directory object.
1777 */
1778static int
1779spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1780{
1781	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1782	    name, sizeof (uint64_t), 1, val));
1783}
1784
1785static int
1786spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1787{
1788	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1789	return (err);
1790}
1791
1792/*
1793 * Fix up config after a partly-completed split.  This is done with the
1794 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1795 * pool have that entry in their config, but only the splitting one contains
1796 * a list of all the guids of the vdevs that are being split off.
1797 *
1798 * This function determines what to do with that list: either rejoin
1799 * all the disks to the pool, or complete the splitting process.  To attempt
1800 * the rejoin, each disk that is offlined is marked online again, and
1801 * we do a reopen() call.  If the vdev label for every disk that was
1802 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1803 * then we call vdev_split() on each disk, and complete the split.
1804 *
1805 * Otherwise we leave the config alone, with all the vdevs in place in
1806 * the original pool.
1807 */
1808static void
1809spa_try_repair(spa_t *spa, nvlist_t *config)
1810{
1811	uint_t extracted;
1812	uint64_t *glist;
1813	uint_t i, gcount;
1814	nvlist_t *nvl;
1815	vdev_t **vd;
1816	boolean_t attempt_reopen;
1817
1818	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1819		return;
1820
1821	/* check that the config is complete */
1822	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1823	    &glist, &gcount) != 0)
1824		return;
1825
1826	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1827
1828	/* attempt to online all the vdevs & validate */
1829	attempt_reopen = B_TRUE;
1830	for (i = 0; i < gcount; i++) {
1831		if (glist[i] == 0)	/* vdev is hole */
1832			continue;
1833
1834		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1835		if (vd[i] == NULL) {
1836			/*
1837			 * Don't bother attempting to reopen the disks;
1838			 * just do the split.
1839			 */
1840			attempt_reopen = B_FALSE;
1841		} else {
1842			/* attempt to re-online it */
1843			vd[i]->vdev_offline = B_FALSE;
1844		}
1845	}
1846
1847	if (attempt_reopen) {
1848		vdev_reopen(spa->spa_root_vdev);
1849
1850		/* check each device to see what state it's in */
1851		for (extracted = 0, i = 0; i < gcount; i++) {
1852			if (vd[i] != NULL &&
1853			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1854				break;
1855			++extracted;
1856		}
1857	}
1858
1859	/*
1860	 * If every disk has been moved to the new pool, or if we never
1861	 * even attempted to look at them, then we split them off for
1862	 * good.
1863	 */
1864	if (!attempt_reopen || gcount == extracted) {
1865		for (i = 0; i < gcount; i++)
1866			if (vd[i] != NULL)
1867				vdev_split(vd[i]);
1868		vdev_reopen(spa->spa_root_vdev);
1869	}
1870
1871	kmem_free(vd, gcount * sizeof (vdev_t *));
1872}
1873
1874static int
1875spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1876    boolean_t mosconfig)
1877{
1878	nvlist_t *config = spa->spa_config;
1879	char *ereport = FM_EREPORT_ZFS_POOL;
1880	char *comment;
1881	int error;
1882	uint64_t pool_guid;
1883	nvlist_t *nvl;
1884
1885	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1886		return (EINVAL);
1887
1888	ASSERT(spa->spa_comment == NULL);
1889	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
1890		spa->spa_comment = spa_strdup(comment);
1891
1892	/*
1893	 * Versioning wasn't explicitly added to the label until later, so if
1894	 * it's not present treat it as the initial version.
1895	 */
1896	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1897	    &spa->spa_ubsync.ub_version) != 0)
1898		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1899
1900	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1901	    &spa->spa_config_txg);
1902
1903	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1904	    spa_guid_exists(pool_guid, 0)) {
1905		error = EEXIST;
1906	} else {
1907		spa->spa_config_guid = pool_guid;
1908
1909		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1910		    &nvl) == 0) {
1911			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1912			    KM_SLEEP) == 0);
1913		}
1914
1915		nvlist_free(spa->spa_load_info);
1916		spa->spa_load_info = fnvlist_alloc();
1917
1918		gethrestime(&spa->spa_loaded_ts);
1919		error = spa_load_impl(spa, pool_guid, config, state, type,
1920		    mosconfig, &ereport);
1921	}
1922
1923	spa->spa_minref = refcount_count(&spa->spa_refcount);
1924	if (error) {
1925		if (error != EEXIST) {
1926			spa->spa_loaded_ts.tv_sec = 0;
1927			spa->spa_loaded_ts.tv_nsec = 0;
1928		}
1929		if (error != EBADF) {
1930			zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1931		}
1932	}
1933	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1934	spa->spa_ena = 0;
1935
1936	return (error);
1937}
1938
1939/*
1940 * Load an existing storage pool, using the pool's builtin spa_config as a
1941 * source of configuration information.
1942 */
1943static int
1944spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1945    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1946    char **ereport)
1947{
1948	int error = 0;
1949	nvlist_t *nvroot = NULL;
1950	nvlist_t *label;
1951	vdev_t *rvd;
1952	uberblock_t *ub = &spa->spa_uberblock;
1953	uint64_t children, config_cache_txg = spa->spa_config_txg;
1954	int orig_mode = spa->spa_mode;
1955	int parse;
1956	uint64_t obj;
1957	boolean_t missing_feat_write = B_FALSE;
1958
1959	/*
1960	 * If this is an untrusted config, access the pool in read-only mode.
1961	 * This prevents things like resilvering recently removed devices.
1962	 */
1963	if (!mosconfig)
1964		spa->spa_mode = FREAD;
1965
1966	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1967
1968	spa->spa_load_state = state;
1969
1970	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1971		return (EINVAL);
1972
1973	parse = (type == SPA_IMPORT_EXISTING ?
1974	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1975
1976	/*
1977	 * Create "The Godfather" zio to hold all async IOs
1978	 */
1979	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1980	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1981
1982	/*
1983	 * Parse the configuration into a vdev tree.  We explicitly set the
1984	 * value that will be returned by spa_version() since parsing the
1985	 * configuration requires knowing the version number.
1986	 */
1987	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1988	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1989	spa_config_exit(spa, SCL_ALL, FTAG);
1990
1991	if (error != 0)
1992		return (error);
1993
1994	ASSERT(spa->spa_root_vdev == rvd);
1995
1996	if (type != SPA_IMPORT_ASSEMBLE) {
1997		ASSERT(spa_guid(spa) == pool_guid);
1998	}
1999
2000	/*
2001	 * Try to open all vdevs, loading each label in the process.
2002	 */
2003	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2004	error = vdev_open(rvd);
2005	spa_config_exit(spa, SCL_ALL, FTAG);
2006	if (error != 0)
2007		return (error);
2008
2009	/*
2010	 * We need to validate the vdev labels against the configuration that
2011	 * we have in hand, which is dependent on the setting of mosconfig. If
2012	 * mosconfig is true then we're validating the vdev labels based on
2013	 * that config.  Otherwise, we're validating against the cached config
2014	 * (zpool.cache) that was read when we loaded the zfs module, and then
2015	 * later we will recursively call spa_load() and validate against
2016	 * the vdev config.
2017	 *
2018	 * If we're assembling a new pool that's been split off from an
2019	 * existing pool, the labels haven't yet been updated so we skip
2020	 * validation for now.
2021	 */
2022	if (type != SPA_IMPORT_ASSEMBLE) {
2023		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2024		error = vdev_validate(rvd, mosconfig);
2025		spa_config_exit(spa, SCL_ALL, FTAG);
2026
2027		if (error != 0)
2028			return (error);
2029
2030		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2031			return (ENXIO);
2032	}
2033
2034	/*
2035	 * Find the best uberblock.
2036	 */
2037	vdev_uberblock_load(rvd, ub, &label);
2038
2039	/*
2040	 * If we weren't able to find a single valid uberblock, return failure.
2041	 */
2042	if (ub->ub_txg == 0) {
2043		nvlist_free(label);
2044		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2045	}
2046
2047	/*
2048	 * If the pool has an unsupported version we can't open it.
2049	 */
2050	if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2051		nvlist_free(label);
2052		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2053	}
2054
2055	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2056		nvlist_t *features;
2057
2058		/*
2059		 * If we weren't able to find what's necessary for reading the
2060		 * MOS in the label, return failure.
2061		 */
2062		if (label == NULL || nvlist_lookup_nvlist(label,
2063		    ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2064			nvlist_free(label);
2065			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2066			    ENXIO));
2067		}
2068
2069		/*
2070		 * Update our in-core representation with the definitive values
2071		 * from the label.
2072		 */
2073		nvlist_free(spa->spa_label_features);
2074		VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2075	}
2076
2077	nvlist_free(label);
2078
2079	/*
2080	 * Look through entries in the label nvlist's features_for_read. If
2081	 * there is a feature listed there which we don't understand then we
2082	 * cannot open a pool.
2083	 */
2084	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2085		nvlist_t *unsup_feat;
2086
2087		VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2088		    0);
2089
2090		for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2091		    NULL); nvp != NULL;
2092		    nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2093			if (!zfeature_is_supported(nvpair_name(nvp))) {
2094				VERIFY(nvlist_add_string(unsup_feat,
2095				    nvpair_name(nvp), "") == 0);
2096			}
2097		}
2098
2099		if (!nvlist_empty(unsup_feat)) {
2100			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2101			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2102			nvlist_free(unsup_feat);
2103			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2104			    ENOTSUP));
2105		}
2106
2107		nvlist_free(unsup_feat);
2108	}
2109
2110	/*
2111	 * If the vdev guid sum doesn't match the uberblock, we have an
2112	 * incomplete configuration.  We first check to see if the pool
2113	 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2114	 * If it is, defer the vdev_guid_sum check till later so we
2115	 * can handle missing vdevs.
2116	 */
2117	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2118	    &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2119	    rvd->vdev_guid_sum != ub->ub_guid_sum)
2120		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2121
2122	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2123		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2124		spa_try_repair(spa, config);
2125		spa_config_exit(spa, SCL_ALL, FTAG);
2126		nvlist_free(spa->spa_config_splitting);
2127		spa->spa_config_splitting = NULL;
2128	}
2129
2130	/*
2131	 * Initialize internal SPA structures.
2132	 */
2133	spa->spa_state = POOL_STATE_ACTIVE;
2134	spa->spa_ubsync = spa->spa_uberblock;
2135	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2136	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2137	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2138	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2139	spa->spa_claim_max_txg = spa->spa_first_txg;
2140	spa->spa_prev_software_version = ub->ub_software_version;
2141
2142	error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2143	if (error)
2144		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2145	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2146
2147	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2148		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2149
2150	if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2151		boolean_t missing_feat_read = B_FALSE;
2152		nvlist_t *unsup_feat, *enabled_feat;
2153
2154		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2155		    &spa->spa_feat_for_read_obj) != 0) {
2156			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2157		}
2158
2159		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2160		    &spa->spa_feat_for_write_obj) != 0) {
2161			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2162		}
2163
2164		if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2165		    &spa->spa_feat_desc_obj) != 0) {
2166			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2167		}
2168
2169		enabled_feat = fnvlist_alloc();
2170		unsup_feat = fnvlist_alloc();
2171
2172		if (!feature_is_supported(spa->spa_meta_objset,
2173		    spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2174		    unsup_feat, enabled_feat))
2175			missing_feat_read = B_TRUE;
2176
2177		if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2178			if (!feature_is_supported(spa->spa_meta_objset,
2179			    spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2180			    unsup_feat, enabled_feat)) {
2181				missing_feat_write = B_TRUE;
2182			}
2183		}
2184
2185		fnvlist_add_nvlist(spa->spa_load_info,
2186		    ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2187
2188		if (!nvlist_empty(unsup_feat)) {
2189			fnvlist_add_nvlist(spa->spa_load_info,
2190			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2191		}
2192
2193		fnvlist_free(enabled_feat);
2194		fnvlist_free(unsup_feat);
2195
2196		if (!missing_feat_read) {
2197			fnvlist_add_boolean(spa->spa_load_info,
2198			    ZPOOL_CONFIG_CAN_RDONLY);
2199		}
2200
2201		/*
2202		 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2203		 * twofold: to determine whether the pool is available for
2204		 * import in read-write mode and (if it is not) whether the
2205		 * pool is available for import in read-only mode. If the pool
2206		 * is available for import in read-write mode, it is displayed
2207		 * as available in userland; if it is not available for import
2208		 * in read-only mode, it is displayed as unavailable in
2209		 * userland. If the pool is available for import in read-only
2210		 * mode but not read-write mode, it is displayed as unavailable
2211		 * in userland with a special note that the pool is actually
2212		 * available for open in read-only mode.
2213		 *
2214		 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2215		 * missing a feature for write, we must first determine whether
2216		 * the pool can be opened read-only before returning to
2217		 * userland in order to know whether to display the
2218		 * abovementioned note.
2219		 */
2220		if (missing_feat_read || (missing_feat_write &&
2221		    spa_writeable(spa))) {
2222			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2223			    ENOTSUP));
2224		}
2225	}
2226
2227	spa->spa_is_initializing = B_TRUE;
2228	error = dsl_pool_open(spa->spa_dsl_pool);
2229	spa->spa_is_initializing = B_FALSE;
2230	if (error != 0)
2231		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2232
2233	if (!mosconfig) {
2234		uint64_t hostid;
2235		nvlist_t *policy = NULL, *nvconfig;
2236
2237		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2238			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2239
2240		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2241		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2242			char *hostname;
2243			unsigned long myhostid = 0;
2244
2245			VERIFY(nvlist_lookup_string(nvconfig,
2246			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2247
2248#ifdef	_KERNEL
2249			myhostid = zone_get_hostid(NULL);
2250#else	/* _KERNEL */
2251			/*
2252			 * We're emulating the system's hostid in userland, so
2253			 * we can't use zone_get_hostid().
2254			 */
2255			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2256#endif	/* _KERNEL */
2257			if (hostid != 0 && myhostid != 0 &&
2258			    hostid != myhostid) {
2259				nvlist_free(nvconfig);
2260				cmn_err(CE_WARN, "pool '%s' could not be "
2261				    "loaded as it was last accessed by "
2262				    "another system (host: %s hostid: 0x%lx). "
2263				    "See: http://illumos.org/msg/ZFS-8000-EY",
2264				    spa_name(spa), hostname,
2265				    (unsigned long)hostid);
2266				return (EBADF);
2267			}
2268		}
2269		if (nvlist_lookup_nvlist(spa->spa_config,
2270		    ZPOOL_REWIND_POLICY, &policy) == 0)
2271			VERIFY(nvlist_add_nvlist(nvconfig,
2272			    ZPOOL_REWIND_POLICY, policy) == 0);
2273
2274		spa_config_set(spa, nvconfig);
2275		spa_unload(spa);
2276		spa_deactivate(spa);
2277		spa_activate(spa, orig_mode);
2278
2279		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2280	}
2281
2282	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2283		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2284	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2285	if (error != 0)
2286		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2287
2288	/*
2289	 * Load the bit that tells us to use the new accounting function
2290	 * (raid-z deflation).  If we have an older pool, this will not
2291	 * be present.
2292	 */
2293	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2294	if (error != 0 && error != ENOENT)
2295		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2296
2297	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2298	    &spa->spa_creation_version);
2299	if (error != 0 && error != ENOENT)
2300		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2301
2302	/*
2303	 * Load the persistent error log.  If we have an older pool, this will
2304	 * not be present.
2305	 */
2306	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2307	if (error != 0 && error != ENOENT)
2308		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2309
2310	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2311	    &spa->spa_errlog_scrub);
2312	if (error != 0 && error != ENOENT)
2313		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2314
2315	/*
2316	 * Load the history object.  If we have an older pool, this
2317	 * will not be present.
2318	 */
2319	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2320	if (error != 0 && error != ENOENT)
2321		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2322
2323	/*
2324	 * If we're assembling the pool from the split-off vdevs of
2325	 * an existing pool, we don't want to attach the spares & cache
2326	 * devices.
2327	 */
2328
2329	/*
2330	 * Load any hot spares for this pool.
2331	 */
2332	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2333	if (error != 0 && error != ENOENT)
2334		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2335	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2336		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2337		if (load_nvlist(spa, spa->spa_spares.sav_object,
2338		    &spa->spa_spares.sav_config) != 0)
2339			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2340
2341		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2342		spa_load_spares(spa);
2343		spa_config_exit(spa, SCL_ALL, FTAG);
2344	} else if (error == 0) {
2345		spa->spa_spares.sav_sync = B_TRUE;
2346	}
2347
2348	/*
2349	 * Load any level 2 ARC devices for this pool.
2350	 */
2351	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2352	    &spa->spa_l2cache.sav_object);
2353	if (error != 0 && error != ENOENT)
2354		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2355	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2356		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2357		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2358		    &spa->spa_l2cache.sav_config) != 0)
2359			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2360
2361		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2362		spa_load_l2cache(spa);
2363		spa_config_exit(spa, SCL_ALL, FTAG);
2364	} else if (error == 0) {
2365		spa->spa_l2cache.sav_sync = B_TRUE;
2366	}
2367
2368	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2369
2370	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2371	if (error && error != ENOENT)
2372		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2373
2374	if (error == 0) {
2375		uint64_t autoreplace;
2376
2377		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2378		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2379		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2380		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2381		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2382		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2383		    &spa->spa_dedup_ditto);
2384
2385		spa->spa_autoreplace = (autoreplace != 0);
2386	}
2387
2388	/*
2389	 * If the 'autoreplace' property is set, then post a resource notifying
2390	 * the ZFS DE that it should not issue any faults for unopenable
2391	 * devices.  We also iterate over the vdevs, and post a sysevent for any
2392	 * unopenable vdevs so that the normal autoreplace handler can take
2393	 * over.
2394	 */
2395	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2396		spa_check_removed(spa->spa_root_vdev);
2397		/*
2398		 * For the import case, this is done in spa_import(), because
2399		 * at this point we're using the spare definitions from
2400		 * the MOS config, not necessarily from the userland config.
2401		 */
2402		if (state != SPA_LOAD_IMPORT) {
2403			spa_aux_check_removed(&spa->spa_spares);
2404			spa_aux_check_removed(&spa->spa_l2cache);
2405		}
2406	}
2407
2408	/*
2409	 * Load the vdev state for all toplevel vdevs.
2410	 */
2411	vdev_load(rvd);
2412
2413	/*
2414	 * Propagate the leaf DTLs we just loaded all the way up the tree.
2415	 */
2416	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2417	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2418	spa_config_exit(spa, SCL_ALL, FTAG);
2419
2420	/*
2421	 * Load the DDTs (dedup tables).
2422	 */
2423	error = ddt_load(spa);
2424	if (error != 0)
2425		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2426
2427	spa_update_dspace(spa);
2428
2429	/*
2430	 * Validate the config, using the MOS config to fill in any
2431	 * information which might be missing.  If we fail to validate
2432	 * the config then declare the pool unfit for use. If we're
2433	 * assembling a pool from a split, the log is not transferred
2434	 * over.
2435	 */
2436	if (type != SPA_IMPORT_ASSEMBLE) {
2437		nvlist_t *nvconfig;
2438
2439		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2440			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2441
2442		if (!spa_config_valid(spa, nvconfig)) {
2443			nvlist_free(nvconfig);
2444			return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2445			    ENXIO));
2446		}
2447		nvlist_free(nvconfig);
2448
2449		/*
2450		 * Now that we've validated the config, check the state of the
2451		 * root vdev.  If it can't be opened, it indicates one or
2452		 * more toplevel vdevs are faulted.
2453		 */
2454		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2455			return (ENXIO);
2456
2457		if (spa_check_logs(spa)) {
2458			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2459			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2460		}
2461	}
2462
2463	if (missing_feat_write) {
2464		ASSERT(state == SPA_LOAD_TRYIMPORT);
2465
2466		/*
2467		 * At this point, we know that we can open the pool in
2468		 * read-only mode but not read-write mode. We now have enough
2469		 * information and can return to userland.
2470		 */
2471		return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2472	}
2473
2474	/*
2475	 * We've successfully opened the pool, verify that we're ready
2476	 * to start pushing transactions.
2477	 */
2478	if (state != SPA_LOAD_TRYIMPORT) {
2479		if (error = spa_load_verify(spa))
2480			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2481			    error));
2482	}
2483
2484	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2485	    spa->spa_load_max_txg == UINT64_MAX)) {
2486		dmu_tx_t *tx;
2487		int need_update = B_FALSE;
2488
2489		ASSERT(state != SPA_LOAD_TRYIMPORT);
2490
2491		/*
2492		 * Claim log blocks that haven't been committed yet.
2493		 * This must all happen in a single txg.
2494		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2495		 * invoked from zil_claim_log_block()'s i/o done callback.
2496		 * Price of rollback is that we abandon the log.
2497		 */
2498		spa->spa_claiming = B_TRUE;
2499
2500		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2501		    spa_first_txg(spa));
2502		(void) dmu_objset_find(spa_name(spa),
2503		    zil_claim, tx, DS_FIND_CHILDREN);
2504		dmu_tx_commit(tx);
2505
2506		spa->spa_claiming = B_FALSE;
2507
2508		spa_set_log_state(spa, SPA_LOG_GOOD);
2509		spa->spa_sync_on = B_TRUE;
2510		txg_sync_start(spa->spa_dsl_pool);
2511
2512		/*
2513		 * Wait for all claims to sync.  We sync up to the highest
2514		 * claimed log block birth time so that claimed log blocks
2515		 * don't appear to be from the future.  spa_claim_max_txg
2516		 * will have been set for us by either zil_check_log_chain()
2517		 * (invoked from spa_check_logs()) or zil_claim() above.
2518		 */
2519		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2520
2521		/*
2522		 * If the config cache is stale, or we have uninitialized
2523		 * metaslabs (see spa_vdev_add()), then update the config.
2524		 *
2525		 * If this is a verbatim import, trust the current
2526		 * in-core spa_config and update the disk labels.
2527		 */
2528		if (config_cache_txg != spa->spa_config_txg ||
2529		    state == SPA_LOAD_IMPORT ||
2530		    state == SPA_LOAD_RECOVER ||
2531		    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2532			need_update = B_TRUE;
2533
2534		for (int c = 0; c < rvd->vdev_children; c++)
2535			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2536				need_update = B_TRUE;
2537
2538		/*
2539		 * Update the config cache asychronously in case we're the
2540		 * root pool, in which case the config cache isn't writable yet.
2541		 */
2542		if (need_update)
2543			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2544
2545		/*
2546		 * Check all DTLs to see if anything needs resilvering.
2547		 */
2548		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2549		    vdev_resilver_needed(rvd, NULL, NULL))
2550			spa_async_request(spa, SPA_ASYNC_RESILVER);
2551
2552		/*
2553		 * Log the fact that we booted up (so that we can detect if
2554		 * we rebooted in the middle of an operation).
2555		 */
2556		spa_history_log_version(spa, "open");
2557
2558		/*
2559		 * Delete any inconsistent datasets.
2560		 */
2561		(void) dmu_objset_find(spa_name(spa),
2562		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2563
2564		/*
2565		 * Clean up any stale temporary dataset userrefs.
2566		 */
2567		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2568	}
2569
2570	return (0);
2571}
2572
2573static int
2574spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2575{
2576	int mode = spa->spa_mode;
2577
2578	spa_unload(spa);
2579	spa_deactivate(spa);
2580
2581	spa->spa_load_max_txg--;
2582
2583	spa_activate(spa, mode);
2584	spa_async_suspend(spa);
2585
2586	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2587}
2588
2589/*
2590 * If spa_load() fails this function will try loading prior txg's. If
2591 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2592 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2593 * function will not rewind the pool and will return the same error as
2594 * spa_load().
2595 */
2596static int
2597spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2598    uint64_t max_request, int rewind_flags)
2599{
2600	nvlist_t *loadinfo = NULL;
2601	nvlist_t *config = NULL;
2602	int load_error, rewind_error;
2603	uint64_t safe_rewind_txg;
2604	uint64_t min_txg;
2605
2606	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2607		spa->spa_load_max_txg = spa->spa_load_txg;
2608		spa_set_log_state(spa, SPA_LOG_CLEAR);
2609	} else {
2610		spa->spa_load_max_txg = max_request;
2611	}
2612
2613	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2614	    mosconfig);
2615	if (load_error == 0)
2616		return (0);
2617
2618	if (spa->spa_root_vdev != NULL)
2619		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2620
2621	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2622	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2623
2624	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2625		nvlist_free(config);
2626		return (load_error);
2627	}
2628
2629	if (state == SPA_LOAD_RECOVER) {
2630		/* Price of rolling back is discarding txgs, including log */
2631		spa_set_log_state(spa, SPA_LOG_CLEAR);
2632	} else {
2633		/*
2634		 * If we aren't rolling back save the load info from our first
2635		 * import attempt so that we can restore it after attempting
2636		 * to rewind.
2637		 */
2638		loadinfo = spa->spa_load_info;
2639		spa->spa_load_info = fnvlist_alloc();
2640	}
2641
2642	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2643	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2644	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2645	    TXG_INITIAL : safe_rewind_txg;
2646
2647	/*
2648	 * Continue as long as we're finding errors, we're still within
2649	 * the acceptable rewind range, and we're still finding uberblocks
2650	 */
2651	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2652	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2653		if (spa->spa_load_max_txg < safe_rewind_txg)
2654			spa->spa_extreme_rewind = B_TRUE;
2655		rewind_error = spa_load_retry(spa, state, mosconfig);
2656	}
2657
2658	spa->spa_extreme_rewind = B_FALSE;
2659	spa->spa_load_max_txg = UINT64_MAX;
2660
2661	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2662		spa_config_set(spa, config);
2663
2664	if (state == SPA_LOAD_RECOVER) {
2665		ASSERT3P(loadinfo, ==, NULL);
2666		return (rewind_error);
2667	} else {
2668		/* Store the rewind info as part of the initial load info */
2669		fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2670		    spa->spa_load_info);
2671
2672		/* Restore the initial load info */
2673		fnvlist_free(spa->spa_load_info);
2674		spa->spa_load_info = loadinfo;
2675
2676		return (load_error);
2677	}
2678}
2679
2680/*
2681 * Pool Open/Import
2682 *
2683 * The import case is identical to an open except that the configuration is sent
2684 * down from userland, instead of grabbed from the configuration cache.  For the
2685 * case of an open, the pool configuration will exist in the
2686 * POOL_STATE_UNINITIALIZED state.
2687 *
2688 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2689 * the same time open the pool, without having to keep around the spa_t in some
2690 * ambiguous state.
2691 */
2692static int
2693spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2694    nvlist_t **config)
2695{
2696	spa_t *spa;
2697	spa_load_state_t state = SPA_LOAD_OPEN;
2698	int error;
2699	int locked = B_FALSE;
2700
2701	*spapp = NULL;
2702
2703	/*
2704	 * As disgusting as this is, we need to support recursive calls to this
2705	 * function because dsl_dir_open() is called during spa_load(), and ends
2706	 * up calling spa_open() again.  The real fix is to figure out how to
2707	 * avoid dsl_dir_open() calling this in the first place.
2708	 */
2709	if (mutex_owner(&spa_namespace_lock) != curthread) {
2710		mutex_enter(&spa_namespace_lock);
2711		locked = B_TRUE;
2712	}
2713
2714	if ((spa = spa_lookup(pool)) == NULL) {
2715		if (locked)
2716			mutex_exit(&spa_namespace_lock);
2717		return (ENOENT);
2718	}
2719
2720	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2721		zpool_rewind_policy_t policy;
2722
2723		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2724		    &policy);
2725		if (policy.zrp_request & ZPOOL_DO_REWIND)
2726			state = SPA_LOAD_RECOVER;
2727
2728		spa_activate(spa, spa_mode_global);
2729
2730		if (state != SPA_LOAD_RECOVER)
2731			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2732
2733		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2734		    policy.zrp_request);
2735
2736		if (error == EBADF) {
2737			/*
2738			 * If vdev_validate() returns failure (indicated by
2739			 * EBADF), it indicates that one of the vdevs indicates
2740			 * that the pool has been exported or destroyed.  If
2741			 * this is the case, the config cache is out of sync and
2742			 * we should remove the pool from the namespace.
2743			 */
2744			spa_unload(spa);
2745			spa_deactivate(spa);
2746			spa_config_sync(spa, B_TRUE, B_TRUE);
2747			spa_remove(spa);
2748			if (locked)
2749				mutex_exit(&spa_namespace_lock);
2750			return (ENOENT);
2751		}
2752
2753		if (error) {
2754			/*
2755			 * We can't open the pool, but we still have useful
2756			 * information: the state of each vdev after the
2757			 * attempted vdev_open().  Return this to the user.
2758			 */
2759			if (config != NULL && spa->spa_config) {
2760				VERIFY(nvlist_dup(spa->spa_config, config,
2761				    KM_SLEEP) == 0);
2762				VERIFY(nvlist_add_nvlist(*config,
2763				    ZPOOL_CONFIG_LOAD_INFO,
2764				    spa->spa_load_info) == 0);
2765			}
2766			spa_unload(spa);
2767			spa_deactivate(spa);
2768			spa->spa_last_open_failed = error;
2769			if (locked)
2770				mutex_exit(&spa_namespace_lock);
2771			*spapp = NULL;
2772			return (error);
2773		}
2774	}
2775
2776	spa_open_ref(spa, tag);
2777
2778	if (config != NULL)
2779		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2780
2781	/*
2782	 * If we've recovered the pool, pass back any information we
2783	 * gathered while doing the load.
2784	 */
2785	if (state == SPA_LOAD_RECOVER) {
2786		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2787		    spa->spa_load_info) == 0);
2788	}
2789
2790	if (locked) {
2791		spa->spa_last_open_failed = 0;
2792		spa->spa_last_ubsync_txg = 0;
2793		spa->spa_load_txg = 0;
2794		mutex_exit(&spa_namespace_lock);
2795	}
2796
2797	*spapp = spa;
2798
2799	return (0);
2800}
2801
2802int
2803spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2804    nvlist_t **config)
2805{
2806	return (spa_open_common(name, spapp, tag, policy, config));
2807}
2808
2809int
2810spa_open(const char *name, spa_t **spapp, void *tag)
2811{
2812	return (spa_open_common(name, spapp, tag, NULL, NULL));
2813}
2814
2815/*
2816 * Lookup the given spa_t, incrementing the inject count in the process,
2817 * preventing it from being exported or destroyed.
2818 */
2819spa_t *
2820spa_inject_addref(char *name)
2821{
2822	spa_t *spa;
2823
2824	mutex_enter(&spa_namespace_lock);
2825	if ((spa = spa_lookup(name)) == NULL) {
2826		mutex_exit(&spa_namespace_lock);
2827		return (NULL);
2828	}
2829	spa->spa_inject_ref++;
2830	mutex_exit(&spa_namespace_lock);
2831
2832	return (spa);
2833}
2834
2835void
2836spa_inject_delref(spa_t *spa)
2837{
2838	mutex_enter(&spa_namespace_lock);
2839	spa->spa_inject_ref--;
2840	mutex_exit(&spa_namespace_lock);
2841}
2842
2843/*
2844 * Add spares device information to the nvlist.
2845 */
2846static void
2847spa_add_spares(spa_t *spa, nvlist_t *config)
2848{
2849	nvlist_t **spares;
2850	uint_t i, nspares;
2851	nvlist_t *nvroot;
2852	uint64_t guid;
2853	vdev_stat_t *vs;
2854	uint_t vsc;
2855	uint64_t pool;
2856
2857	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2858
2859	if (spa->spa_spares.sav_count == 0)
2860		return;
2861
2862	VERIFY(nvlist_lookup_nvlist(config,
2863	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2864	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2865	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2866	if (nspares != 0) {
2867		VERIFY(nvlist_add_nvlist_array(nvroot,
2868		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2869		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2870		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2871
2872		/*
2873		 * Go through and find any spares which have since been
2874		 * repurposed as an active spare.  If this is the case, update
2875		 * their status appropriately.
2876		 */
2877		for (i = 0; i < nspares; i++) {
2878			VERIFY(nvlist_lookup_uint64(spares[i],
2879			    ZPOOL_CONFIG_GUID, &guid) == 0);
2880			if (spa_spare_exists(guid, &pool, NULL) &&
2881			    pool != 0ULL) {
2882				VERIFY(nvlist_lookup_uint64_array(
2883				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
2884				    (uint64_t **)&vs, &vsc) == 0);
2885				vs->vs_state = VDEV_STATE_CANT_OPEN;
2886				vs->vs_aux = VDEV_AUX_SPARED;
2887			}
2888		}
2889	}
2890}
2891
2892/*
2893 * Add l2cache device information to the nvlist, including vdev stats.
2894 */
2895static void
2896spa_add_l2cache(spa_t *spa, nvlist_t *config)
2897{
2898	nvlist_t **l2cache;
2899	uint_t i, j, nl2cache;
2900	nvlist_t *nvroot;
2901	uint64_t guid;
2902	vdev_t *vd;
2903	vdev_stat_t *vs;
2904	uint_t vsc;
2905
2906	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2907
2908	if (spa->spa_l2cache.sav_count == 0)
2909		return;
2910
2911	VERIFY(nvlist_lookup_nvlist(config,
2912	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2913	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2914	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2915	if (nl2cache != 0) {
2916		VERIFY(nvlist_add_nvlist_array(nvroot,
2917		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2918		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2919		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2920
2921		/*
2922		 * Update level 2 cache device stats.
2923		 */
2924
2925		for (i = 0; i < nl2cache; i++) {
2926			VERIFY(nvlist_lookup_uint64(l2cache[i],
2927			    ZPOOL_CONFIG_GUID, &guid) == 0);
2928
2929			vd = NULL;
2930			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2931				if (guid ==
2932				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2933					vd = spa->spa_l2cache.sav_vdevs[j];
2934					break;
2935				}
2936			}
2937			ASSERT(vd != NULL);
2938
2939			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2940			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2941			    == 0);
2942			vdev_get_stats(vd, vs);
2943		}
2944	}
2945}
2946
2947static void
2948spa_add_feature_stats(spa_t *spa, nvlist_t *config)
2949{
2950	nvlist_t *features;
2951	zap_cursor_t zc;
2952	zap_attribute_t za;
2953
2954	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2955	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2956
2957	if (spa->spa_feat_for_read_obj != 0) {
2958		for (zap_cursor_init(&zc, spa->spa_meta_objset,
2959		    spa->spa_feat_for_read_obj);
2960		    zap_cursor_retrieve(&zc, &za) == 0;
2961		    zap_cursor_advance(&zc)) {
2962			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
2963			    za.za_num_integers == 1);
2964			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
2965			    za.za_first_integer));
2966		}
2967		zap_cursor_fini(&zc);
2968	}
2969
2970	if (spa->spa_feat_for_write_obj != 0) {
2971		for (zap_cursor_init(&zc, spa->spa_meta_objset,
2972		    spa->spa_feat_for_write_obj);
2973		    zap_cursor_retrieve(&zc, &za) == 0;
2974		    zap_cursor_advance(&zc)) {
2975			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
2976			    za.za_num_integers == 1);
2977			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
2978			    za.za_first_integer));
2979		}
2980		zap_cursor_fini(&zc);
2981	}
2982
2983	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
2984	    features) == 0);
2985	nvlist_free(features);
2986}
2987
2988int
2989spa_get_stats(const char *name, nvlist_t **config,
2990    char *altroot, size_t buflen)
2991{
2992	int error;
2993	spa_t *spa;
2994
2995	*config = NULL;
2996	error = spa_open_common(name, &spa, FTAG, NULL, config);
2997
2998	if (spa != NULL) {
2999		/*
3000		 * This still leaves a window of inconsistency where the spares
3001		 * or l2cache devices could change and the config would be
3002		 * self-inconsistent.
3003		 */
3004		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3005
3006		if (*config != NULL) {
3007			uint64_t loadtimes[2];
3008
3009			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3010			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3011			VERIFY(nvlist_add_uint64_array(*config,
3012			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3013
3014			VERIFY(nvlist_add_uint64(*config,
3015			    ZPOOL_CONFIG_ERRCOUNT,
3016			    spa_get_errlog_size(spa)) == 0);
3017
3018			if (spa_suspended(spa))
3019				VERIFY(nvlist_add_uint64(*config,
3020				    ZPOOL_CONFIG_SUSPENDED,
3021				    spa->spa_failmode) == 0);
3022
3023			spa_add_spares(spa, *config);
3024			spa_add_l2cache(spa, *config);
3025			spa_add_feature_stats(spa, *config);
3026		}
3027	}
3028
3029	/*
3030	 * We want to get the alternate root even for faulted pools, so we cheat
3031	 * and call spa_lookup() directly.
3032	 */
3033	if (altroot) {
3034		if (spa == NULL) {
3035			mutex_enter(&spa_namespace_lock);
3036			spa = spa_lookup(name);
3037			if (spa)
3038				spa_altroot(spa, altroot, buflen);
3039			else
3040				altroot[0] = '\0';
3041			spa = NULL;
3042			mutex_exit(&spa_namespace_lock);
3043		} else {
3044			spa_altroot(spa, altroot, buflen);
3045		}
3046	}
3047
3048	if (spa != NULL) {
3049		spa_config_exit(spa, SCL_CONFIG, FTAG);
3050		spa_close(spa, FTAG);
3051	}
3052
3053	return (error);
3054}
3055
3056/*
3057 * Validate that the auxiliary device array is well formed.  We must have an
3058 * array of nvlists, each which describes a valid leaf vdev.  If this is an
3059 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3060 * specified, as long as they are well-formed.
3061 */
3062static int
3063spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3064    spa_aux_vdev_t *sav, const char *config, uint64_t version,
3065    vdev_labeltype_t label)
3066{
3067	nvlist_t **dev;
3068	uint_t i, ndev;
3069	vdev_t *vd;
3070	int error;
3071
3072	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3073
3074	/*
3075	 * It's acceptable to have no devs specified.
3076	 */
3077	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3078		return (0);
3079
3080	if (ndev == 0)
3081		return (EINVAL);
3082
3083	/*
3084	 * Make sure the pool is formatted with a version that supports this
3085	 * device type.
3086	 */
3087	if (spa_version(spa) < version)
3088		return (ENOTSUP);
3089
3090	/*
3091	 * Set the pending device list so we correctly handle device in-use
3092	 * checking.
3093	 */
3094	sav->sav_pending = dev;
3095	sav->sav_npending = ndev;
3096
3097	for (i = 0; i < ndev; i++) {
3098		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3099		    mode)) != 0)
3100			goto out;
3101
3102		if (!vd->vdev_ops->vdev_op_leaf) {
3103			vdev_free(vd);
3104			error = EINVAL;
3105			goto out;
3106		}
3107
3108		/*
3109		 * The L2ARC currently only supports disk devices in
3110		 * kernel context.  For user-level testing, we allow it.
3111		 */
3112#ifdef _KERNEL
3113		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3114		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3115			error = ENOTBLK;
3116			vdev_free(vd);
3117			goto out;
3118		}
3119#endif
3120		vd->vdev_top = vd;
3121
3122		if ((error = vdev_open(vd)) == 0 &&
3123		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
3124			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3125			    vd->vdev_guid) == 0);
3126		}
3127
3128		vdev_free(vd);
3129
3130		if (error &&
3131		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3132			goto out;
3133		else
3134			error = 0;
3135	}
3136
3137out:
3138	sav->sav_pending = NULL;
3139	sav->sav_npending = 0;
3140	return (error);
3141}
3142
3143static int
3144spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3145{
3146	int error;
3147
3148	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3149
3150	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3151	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3152	    VDEV_LABEL_SPARE)) != 0) {
3153		return (error);
3154	}
3155
3156	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3157	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3158	    VDEV_LABEL_L2CACHE));
3159}
3160
3161static void
3162spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3163    const char *config)
3164{
3165	int i;
3166
3167	if (sav->sav_config != NULL) {
3168		nvlist_t **olddevs;
3169		uint_t oldndevs;
3170		nvlist_t **newdevs;
3171
3172		/*
3173		 * Generate new dev list by concatentating with the
3174		 * current dev list.
3175		 */
3176		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3177		    &olddevs, &oldndevs) == 0);
3178
3179		newdevs = kmem_alloc(sizeof (void *) *
3180		    (ndevs + oldndevs), KM_SLEEP);
3181		for (i = 0; i < oldndevs; i++)
3182			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3183			    KM_SLEEP) == 0);
3184		for (i = 0; i < ndevs; i++)
3185			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3186			    KM_SLEEP) == 0);
3187
3188		VERIFY(nvlist_remove(sav->sav_config, config,
3189		    DATA_TYPE_NVLIST_ARRAY) == 0);
3190
3191		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3192		    config, newdevs, ndevs + oldndevs) == 0);
3193		for (i = 0; i < oldndevs + ndevs; i++)
3194			nvlist_free(newdevs[i]);
3195		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3196	} else {
3197		/*
3198		 * Generate a new dev list.
3199		 */
3200		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3201		    KM_SLEEP) == 0);
3202		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3203		    devs, ndevs) == 0);
3204	}
3205}
3206
3207/*
3208 * Stop and drop level 2 ARC devices
3209 */
3210void
3211spa_l2cache_drop(spa_t *spa)
3212{
3213	vdev_t *vd;
3214	int i;
3215	spa_aux_vdev_t *sav = &spa->spa_l2cache;
3216
3217	for (i = 0; i < sav->sav_count; i++) {
3218		uint64_t pool;
3219
3220		vd = sav->sav_vdevs[i];
3221		ASSERT(vd != NULL);
3222
3223		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3224		    pool != 0ULL && l2arc_vdev_present(vd))
3225			l2arc_remove_vdev(vd);
3226	}
3227}
3228
3229/*
3230 * Pool Creation
3231 */
3232int
3233spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3234    nvlist_t *zplprops)
3235{
3236	spa_t *spa;
3237	char *altroot = NULL;
3238	vdev_t *rvd;
3239	dsl_pool_t *dp;
3240	dmu_tx_t *tx;
3241	int error = 0;
3242	uint64_t txg = TXG_INITIAL;
3243	nvlist_t **spares, **l2cache;
3244	uint_t nspares, nl2cache;
3245	uint64_t version, obj;
3246	boolean_t has_features;
3247
3248	/*
3249	 * If this pool already exists, return failure.
3250	 */
3251	mutex_enter(&spa_namespace_lock);
3252	if (spa_lookup(pool) != NULL) {
3253		mutex_exit(&spa_namespace_lock);
3254		return (EEXIST);
3255	}
3256
3257	/*
3258	 * Allocate a new spa_t structure.
3259	 */
3260	(void) nvlist_lookup_string(props,
3261	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3262	spa = spa_add(pool, NULL, altroot);
3263	spa_activate(spa, spa_mode_global);
3264
3265	if (props && (error = spa_prop_validate(spa, props))) {
3266		spa_deactivate(spa);
3267		spa_remove(spa);
3268		mutex_exit(&spa_namespace_lock);
3269		return (error);
3270	}
3271
3272	has_features = B_FALSE;
3273	for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3274	    elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3275		if (zpool_prop_feature(nvpair_name(elem)))
3276			has_features = B_TRUE;
3277	}
3278
3279	if (has_features || nvlist_lookup_uint64(props,
3280	    zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3281		version = SPA_VERSION;
3282	}
3283	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3284
3285	spa->spa_first_txg = txg;
3286	spa->spa_uberblock.ub_txg = txg - 1;
3287	spa->spa_uberblock.ub_version = version;
3288	spa->spa_ubsync = spa->spa_uberblock;
3289
3290	/*
3291	 * Create "The Godfather" zio to hold all async IOs
3292	 */
3293	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3294	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3295
3296	/*
3297	 * Create the root vdev.
3298	 */
3299	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3300
3301	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3302
3303	ASSERT(error != 0 || rvd != NULL);
3304	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3305
3306	if (error == 0 && !zfs_allocatable_devs(nvroot))
3307		error = EINVAL;
3308
3309	if (error == 0 &&
3310	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3311	    (error = spa_validate_aux(spa, nvroot, txg,
3312	    VDEV_ALLOC_ADD)) == 0) {
3313		for (int c = 0; c < rvd->vdev_children; c++) {
3314			vdev_metaslab_set_size(rvd->vdev_child[c]);
3315			vdev_expand(rvd->vdev_child[c], txg);
3316		}
3317	}
3318
3319	spa_config_exit(spa, SCL_ALL, FTAG);
3320
3321	if (error != 0) {
3322		spa_unload(spa);
3323		spa_deactivate(spa);
3324		spa_remove(spa);
3325		mutex_exit(&spa_namespace_lock);
3326		return (error);
3327	}
3328
3329	/*
3330	 * Get the list of spares, if specified.
3331	 */
3332	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3333	    &spares, &nspares) == 0) {
3334		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3335		    KM_SLEEP) == 0);
3336		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3337		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3338		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3339		spa_load_spares(spa);
3340		spa_config_exit(spa, SCL_ALL, FTAG);
3341		spa->spa_spares.sav_sync = B_TRUE;
3342	}
3343
3344	/*
3345	 * Get the list of level 2 cache devices, if specified.
3346	 */
3347	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3348	    &l2cache, &nl2cache) == 0) {
3349		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3350		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3351		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3352		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3353		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3354		spa_load_l2cache(spa);
3355		spa_config_exit(spa, SCL_ALL, FTAG);
3356		spa->spa_l2cache.sav_sync = B_TRUE;
3357	}
3358
3359	spa->spa_is_initializing = B_TRUE;
3360	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3361	spa->spa_meta_objset = dp->dp_meta_objset;
3362	spa->spa_is_initializing = B_FALSE;
3363
3364	/*
3365	 * Create DDTs (dedup tables).
3366	 */
3367	ddt_create(spa);
3368
3369	spa_update_dspace(spa);
3370
3371	tx = dmu_tx_create_assigned(dp, txg);
3372
3373	/*
3374	 * Create the pool config object.
3375	 */
3376	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3377	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3378	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3379
3380	if (zap_add(spa->spa_meta_objset,
3381	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3382	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3383		cmn_err(CE_PANIC, "failed to add pool config");
3384	}
3385
3386	if (spa_version(spa) >= SPA_VERSION_FEATURES)
3387		spa_feature_create_zap_objects(spa, tx);
3388
3389	if (zap_add(spa->spa_meta_objset,
3390	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3391	    sizeof (uint64_t), 1, &version, tx) != 0) {
3392		cmn_err(CE_PANIC, "failed to add pool version");
3393	}
3394
3395	/* Newly created pools with the right version are always deflated. */
3396	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3397		spa->spa_deflate = TRUE;
3398		if (zap_add(spa->spa_meta_objset,
3399		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3400		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3401			cmn_err(CE_PANIC, "failed to add deflate");
3402		}
3403	}
3404
3405	/*
3406	 * Create the deferred-free bpobj.  Turn off compression
3407	 * because sync-to-convergence takes longer if the blocksize
3408	 * keeps changing.
3409	 */
3410	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3411	dmu_object_set_compress(spa->spa_meta_objset, obj,
3412	    ZIO_COMPRESS_OFF, tx);
3413	if (zap_add(spa->spa_meta_objset,
3414	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3415	    sizeof (uint64_t), 1, &obj, tx) != 0) {
3416		cmn_err(CE_PANIC, "failed to add bpobj");
3417	}
3418	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3419	    spa->spa_meta_objset, obj));
3420
3421	/*
3422	 * Create the pool's history object.
3423	 */
3424	if (version >= SPA_VERSION_ZPOOL_HISTORY)
3425		spa_history_create_obj(spa, tx);
3426
3427	/*
3428	 * Set pool properties.
3429	 */
3430	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3431	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3432	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3433	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3434
3435	if (props != NULL) {
3436		spa_configfile_set(spa, props, B_FALSE);
3437		spa_sync_props(spa, props, tx);
3438	}
3439
3440	dmu_tx_commit(tx);
3441
3442	spa->spa_sync_on = B_TRUE;
3443	txg_sync_start(spa->spa_dsl_pool);
3444
3445	/*
3446	 * We explicitly wait for the first transaction to complete so that our
3447	 * bean counters are appropriately updated.
3448	 */
3449	txg_wait_synced(spa->spa_dsl_pool, txg);
3450
3451	spa_config_sync(spa, B_FALSE, B_TRUE);
3452
3453	spa_history_log_version(spa, "create");
3454
3455	spa->spa_minref = refcount_count(&spa->spa_refcount);
3456
3457	mutex_exit(&spa_namespace_lock);
3458
3459	return (0);
3460}
3461
3462#ifdef _KERNEL
3463/*
3464 * Get the root pool information from the root disk, then import the root pool
3465 * during the system boot up time.
3466 */
3467extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3468
3469static nvlist_t *
3470spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3471{
3472	nvlist_t *config;
3473	nvlist_t *nvtop, *nvroot;
3474	uint64_t pgid;
3475
3476	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3477		return (NULL);
3478
3479	/*
3480	 * Add this top-level vdev to the child array.
3481	 */
3482	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3483	    &nvtop) == 0);
3484	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3485	    &pgid) == 0);
3486	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3487
3488	/*
3489	 * Put this pool's top-level vdevs into a root vdev.
3490	 */
3491	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3492	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3493	    VDEV_TYPE_ROOT) == 0);
3494	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3495	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3496	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3497	    &nvtop, 1) == 0);
3498
3499	/*
3500	 * Replace the existing vdev_tree with the new root vdev in
3501	 * this pool's configuration (remove the old, add the new).
3502	 */
3503	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3504	nvlist_free(nvroot);
3505	return (config);
3506}
3507
3508/*
3509 * Walk the vdev tree and see if we can find a device with "better"
3510 * configuration. A configuration is "better" if the label on that
3511 * device has a more recent txg.
3512 */
3513static void
3514spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3515{
3516	for (int c = 0; c < vd->vdev_children; c++)
3517		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3518
3519	if (vd->vdev_ops->vdev_op_leaf) {
3520		nvlist_t *label;
3521		uint64_t label_txg;
3522
3523		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3524		    &label) != 0)
3525			return;
3526
3527		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3528		    &label_txg) == 0);
3529
3530		/*
3531		 * Do we have a better boot device?
3532		 */
3533		if (label_txg > *txg) {
3534			*txg = label_txg;
3535			*avd = vd;
3536		}
3537		nvlist_free(label);
3538	}
3539}
3540
3541/*
3542 * Import a root pool.
3543 *
3544 * For x86. devpath_list will consist of devid and/or physpath name of
3545 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3546 * The GRUB "findroot" command will return the vdev we should boot.
3547 *
3548 * For Sparc, devpath_list consists the physpath name of the booting device
3549 * no matter the rootpool is a single device pool or a mirrored pool.
3550 * e.g.
3551 *	"/pci@1f,0/ide@d/disk@0,0:a"
3552 */
3553int
3554spa_import_rootpool(char *devpath, char *devid)
3555{
3556	spa_t *spa;
3557	vdev_t *rvd, *bvd, *avd = NULL;
3558	nvlist_t *config, *nvtop;
3559	uint64_t guid, txg;
3560	char *pname;
3561	int error;
3562
3563	/*
3564	 * Read the label from the boot device and generate a configuration.
3565	 */
3566	config = spa_generate_rootconf(devpath, devid, &guid);
3567#if defined(_OBP) && defined(_KERNEL)
3568	if (config == NULL) {
3569		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3570			/* iscsi boot */
3571			get_iscsi_bootpath_phy(devpath);
3572			config = spa_generate_rootconf(devpath, devid, &guid);
3573		}
3574	}
3575#endif
3576	if (config == NULL) {
3577		cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3578		    devpath);
3579		return (EIO);
3580	}
3581
3582	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3583	    &pname) == 0);
3584	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3585
3586	mutex_enter(&spa_namespace_lock);
3587	if ((spa = spa_lookup(pname)) != NULL) {
3588		/*
3589		 * Remove the existing root pool from the namespace so that we
3590		 * can replace it with the correct config we just read in.
3591		 */
3592		spa_remove(spa);
3593	}
3594
3595	spa = spa_add(pname, config, NULL);
3596	spa->spa_is_root = B_TRUE;
3597	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3598
3599	/*
3600	 * Build up a vdev tree based on the boot device's label config.
3601	 */
3602	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3603	    &nvtop) == 0);
3604	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3605	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3606	    VDEV_ALLOC_ROOTPOOL);
3607	spa_config_exit(spa, SCL_ALL, FTAG);
3608	if (error) {
3609		mutex_exit(&spa_namespace_lock);
3610		nvlist_free(config);
3611		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3612		    pname);
3613		return (error);
3614	}
3615
3616	/*
3617	 * Get the boot vdev.
3618	 */
3619	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3620		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3621		    (u_longlong_t)guid);
3622		error = ENOENT;
3623		goto out;
3624	}
3625
3626	/*
3627	 * Determine if there is a better boot device.
3628	 */
3629	avd = bvd;
3630	spa_alt_rootvdev(rvd, &avd, &txg);
3631	if (avd != bvd) {
3632		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3633		    "try booting from '%s'", avd->vdev_path);
3634		error = EINVAL;
3635		goto out;
3636	}
3637
3638	/*
3639	 * If the boot device is part of a spare vdev then ensure that
3640	 * we're booting off the active spare.
3641	 */
3642	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3643	    !bvd->vdev_isspare) {
3644		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3645		    "try booting from '%s'",
3646		    bvd->vdev_parent->
3647		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3648		error = EINVAL;
3649		goto out;
3650	}
3651
3652	error = 0;
3653out:
3654	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3655	vdev_free(rvd);
3656	spa_config_exit(spa, SCL_ALL, FTAG);
3657	mutex_exit(&spa_namespace_lock);
3658
3659	nvlist_free(config);
3660	return (error);
3661}
3662
3663#endif
3664
3665/*
3666 * Import a non-root pool into the system.
3667 */
3668int
3669spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3670{
3671	spa_t *spa;
3672	char *altroot = NULL;
3673	spa_load_state_t state = SPA_LOAD_IMPORT;
3674	zpool_rewind_policy_t policy;
3675	uint64_t mode = spa_mode_global;
3676	uint64_t readonly = B_FALSE;
3677	int error;
3678	nvlist_t *nvroot;
3679	nvlist_t **spares, **l2cache;
3680	uint_t nspares, nl2cache;
3681
3682	/*
3683	 * If a pool with this name exists, return failure.
3684	 */
3685	mutex_enter(&spa_namespace_lock);
3686	if (spa_lookup(pool) != NULL) {
3687		mutex_exit(&spa_namespace_lock);
3688		return (EEXIST);
3689	}
3690
3691	/*
3692	 * Create and initialize the spa structure.
3693	 */
3694	(void) nvlist_lookup_string(props,
3695	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3696	(void) nvlist_lookup_uint64(props,
3697	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3698	if (readonly)
3699		mode = FREAD;
3700	spa = spa_add(pool, config, altroot);
3701	spa->spa_import_flags = flags;
3702
3703	/*
3704	 * Verbatim import - Take a pool and insert it into the namespace
3705	 * as if it had been loaded at boot.
3706	 */
3707	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3708		if (props != NULL)
3709			spa_configfile_set(spa, props, B_FALSE);
3710
3711		spa_config_sync(spa, B_FALSE, B_TRUE);
3712
3713		mutex_exit(&spa_namespace_lock);
3714		spa_history_log_version(spa, "import");
3715
3716		return (0);
3717	}
3718
3719	spa_activate(spa, mode);
3720
3721	/*
3722	 * Don't start async tasks until we know everything is healthy.
3723	 */
3724	spa_async_suspend(spa);
3725
3726	zpool_get_rewind_policy(config, &policy);
3727	if (policy.zrp_request & ZPOOL_DO_REWIND)
3728		state = SPA_LOAD_RECOVER;
3729
3730	/*
3731	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3732	 * because the user-supplied config is actually the one to trust when
3733	 * doing an import.
3734	 */
3735	if (state != SPA_LOAD_RECOVER)
3736		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3737
3738	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3739	    policy.zrp_request);
3740
3741	/*
3742	 * Propagate anything learned while loading the pool and pass it
3743	 * back to caller (i.e. rewind info, missing devices, etc).
3744	 */
3745	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3746	    spa->spa_load_info) == 0);
3747
3748	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3749	/*
3750	 * Toss any existing sparelist, as it doesn't have any validity
3751	 * anymore, and conflicts with spa_has_spare().
3752	 */
3753	if (spa->spa_spares.sav_config) {
3754		nvlist_free(spa->spa_spares.sav_config);
3755		spa->spa_spares.sav_config = NULL;
3756		spa_load_spares(spa);
3757	}
3758	if (spa->spa_l2cache.sav_config) {
3759		nvlist_free(spa->spa_l2cache.sav_config);
3760		spa->spa_l2cache.sav_config = NULL;
3761		spa_load_l2cache(spa);
3762	}
3763
3764	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3765	    &nvroot) == 0);
3766	if (error == 0)
3767		error = spa_validate_aux(spa, nvroot, -1ULL,
3768		    VDEV_ALLOC_SPARE);
3769	if (error == 0)
3770		error = spa_validate_aux(spa, nvroot, -1ULL,
3771		    VDEV_ALLOC_L2CACHE);
3772	spa_config_exit(spa, SCL_ALL, FTAG);
3773
3774	if (props != NULL)
3775		spa_configfile_set(spa, props, B_FALSE);
3776
3777	if (error != 0 || (props && spa_writeable(spa) &&
3778	    (error = spa_prop_set(spa, props)))) {
3779		spa_unload(spa);
3780		spa_deactivate(spa);
3781		spa_remove(spa);
3782		mutex_exit(&spa_namespace_lock);
3783		return (error);
3784	}
3785
3786	spa_async_resume(spa);
3787
3788	/*
3789	 * Override any spares and level 2 cache devices as specified by
3790	 * the user, as these may have correct device names/devids, etc.
3791	 */
3792	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3793	    &spares, &nspares) == 0) {
3794		if (spa->spa_spares.sav_config)
3795			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3796			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3797		else
3798			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3799			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3800		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3801		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3802		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3803		spa_load_spares(spa);
3804		spa_config_exit(spa, SCL_ALL, FTAG);
3805		spa->spa_spares.sav_sync = B_TRUE;
3806	}
3807	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3808	    &l2cache, &nl2cache) == 0) {
3809		if (spa->spa_l2cache.sav_config)
3810			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3811			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3812		else
3813			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3814			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3815		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3816		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3817		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3818		spa_load_l2cache(spa);
3819		spa_config_exit(spa, SCL_ALL, FTAG);
3820		spa->spa_l2cache.sav_sync = B_TRUE;
3821	}
3822
3823	/*
3824	 * Check for any removed devices.
3825	 */
3826	if (spa->spa_autoreplace) {
3827		spa_aux_check_removed(&spa->spa_spares);
3828		spa_aux_check_removed(&spa->spa_l2cache);
3829	}
3830
3831	if (spa_writeable(spa)) {
3832		/*
3833		 * Update the config cache to include the newly-imported pool.
3834		 */
3835		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3836	}
3837
3838	/*
3839	 * It's possible that the pool was expanded while it was exported.
3840	 * We kick off an async task to handle this for us.
3841	 */
3842	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3843
3844	mutex_exit(&spa_namespace_lock);
3845	spa_history_log_version(spa, "import");
3846
3847	return (0);
3848}
3849
3850nvlist_t *
3851spa_tryimport(nvlist_t *tryconfig)
3852{
3853	nvlist_t *config = NULL;
3854	char *poolname;
3855	spa_t *spa;
3856	uint64_t state;
3857	int error;
3858
3859	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3860		return (NULL);
3861
3862	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3863		return (NULL);
3864
3865	/*
3866	 * Create and initialize the spa structure.
3867	 */
3868	mutex_enter(&spa_namespace_lock);
3869	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3870	spa_activate(spa, FREAD);
3871
3872	/*
3873	 * Pass off the heavy lifting to spa_load().
3874	 * Pass TRUE for mosconfig because the user-supplied config
3875	 * is actually the one to trust when doing an import.
3876	 */
3877	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3878
3879	/*
3880	 * If 'tryconfig' was at least parsable, return the current config.
3881	 */
3882	if (spa->spa_root_vdev != NULL) {
3883		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3884		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3885		    poolname) == 0);
3886		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3887		    state) == 0);
3888		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3889		    spa->spa_uberblock.ub_timestamp) == 0);
3890		VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3891		    spa->spa_load_info) == 0);
3892
3893		/*
3894		 * If the bootfs property exists on this pool then we
3895		 * copy it out so that external consumers can tell which
3896		 * pools are bootable.
3897		 */
3898		if ((!error || error == EEXIST) && spa->spa_bootfs) {
3899			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3900
3901			/*
3902			 * We have to play games with the name since the
3903			 * pool was opened as TRYIMPORT_NAME.
3904			 */
3905			if (dsl_dsobj_to_dsname(spa_name(spa),
3906			    spa->spa_bootfs, tmpname) == 0) {
3907				char *cp;
3908				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3909
3910				cp = strchr(tmpname, '/');
3911				if (cp == NULL) {
3912					(void) strlcpy(dsname, tmpname,
3913					    MAXPATHLEN);
3914				} else {
3915					(void) snprintf(dsname, MAXPATHLEN,
3916					    "%s/%s", poolname, ++cp);
3917				}
3918				VERIFY(nvlist_add_string(config,
3919				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3920				kmem_free(dsname, MAXPATHLEN);
3921			}
3922			kmem_free(tmpname, MAXPATHLEN);
3923		}
3924
3925		/*
3926		 * Add the list of hot spares and level 2 cache devices.
3927		 */
3928		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3929		spa_add_spares(spa, config);
3930		spa_add_l2cache(spa, config);
3931		spa_config_exit(spa, SCL_CONFIG, FTAG);
3932	}
3933
3934	spa_unload(spa);
3935	spa_deactivate(spa);
3936	spa_remove(spa);
3937	mutex_exit(&spa_namespace_lock);
3938
3939	return (config);
3940}
3941
3942/*
3943 * Pool export/destroy
3944 *
3945 * The act of destroying or exporting a pool is very simple.  We make sure there
3946 * is no more pending I/O and any references to the pool are gone.  Then, we
3947 * update the pool state and sync all the labels to disk, removing the
3948 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3949 * we don't sync the labels or remove the configuration cache.
3950 */
3951static int
3952spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3953    boolean_t force, boolean_t hardforce)
3954{
3955	spa_t *spa;
3956
3957	if (oldconfig)
3958		*oldconfig = NULL;
3959
3960	if (!(spa_mode_global & FWRITE))
3961		return (EROFS);
3962
3963	mutex_enter(&spa_namespace_lock);
3964	if ((spa = spa_lookup(pool)) == NULL) {
3965		mutex_exit(&spa_namespace_lock);
3966		return (ENOENT);
3967	}
3968
3969	/*
3970	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3971	 * reacquire the namespace lock, and see if we can export.
3972	 */
3973	spa_open_ref(spa, FTAG);
3974	mutex_exit(&spa_namespace_lock);
3975	spa_async_suspend(spa);
3976	mutex_enter(&spa_namespace_lock);
3977	spa_close(spa, FTAG);
3978
3979	/*
3980	 * The pool will be in core if it's openable,
3981	 * in which case we can modify its state.
3982	 */
3983	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3984		/*
3985		 * Objsets may be open only because they're dirty, so we
3986		 * have to force it to sync before checking spa_refcnt.
3987		 */
3988		txg_wait_synced(spa->spa_dsl_pool, 0);
3989
3990		/*
3991		 * A pool cannot be exported or destroyed if there are active
3992		 * references.  If we are resetting a pool, allow references by
3993		 * fault injection handlers.
3994		 */
3995		if (!spa_refcount_zero(spa) ||
3996		    (spa->spa_inject_ref != 0 &&
3997		    new_state != POOL_STATE_UNINITIALIZED)) {
3998			spa_async_resume(spa);
3999			mutex_exit(&spa_namespace_lock);
4000			return (EBUSY);
4001		}
4002
4003		/*
4004		 * A pool cannot be exported if it has an active shared spare.
4005		 * This is to prevent other pools stealing the active spare
4006		 * from an exported pool. At user's own will, such pool can
4007		 * be forcedly exported.
4008		 */
4009		if (!force && new_state == POOL_STATE_EXPORTED &&
4010		    spa_has_active_shared_spare(spa)) {
4011			spa_async_resume(spa);
4012			mutex_exit(&spa_namespace_lock);
4013			return (EXDEV);
4014		}
4015
4016		/*
4017		 * We want this to be reflected on every label,
4018		 * so mark them all dirty.  spa_unload() will do the
4019		 * final sync that pushes these changes out.
4020		 */
4021		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4022			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4023			spa->spa_state = new_state;
4024			spa->spa_final_txg = spa_last_synced_txg(spa) +
4025			    TXG_DEFER_SIZE + 1;
4026			vdev_config_dirty(spa->spa_root_vdev);
4027			spa_config_exit(spa, SCL_ALL, FTAG);
4028		}
4029	}
4030
4031	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4032
4033	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4034		spa_unload(spa);
4035		spa_deactivate(spa);
4036	}
4037
4038	if (oldconfig && spa->spa_config)
4039		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4040
4041	if (new_state != POOL_STATE_UNINITIALIZED) {
4042		if (!hardforce)
4043			spa_config_sync(spa, B_TRUE, B_TRUE);
4044		spa_remove(spa);
4045	}
4046	mutex_exit(&spa_namespace_lock);
4047
4048	return (0);
4049}
4050
4051/*
4052 * Destroy a storage pool.
4053 */
4054int
4055spa_destroy(char *pool)
4056{
4057	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4058	    B_FALSE, B_FALSE));
4059}
4060
4061/*
4062 * Export a storage pool.
4063 */
4064int
4065spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4066    boolean_t hardforce)
4067{
4068	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4069	    force, hardforce));
4070}
4071
4072/*
4073 * Similar to spa_export(), this unloads the spa_t without actually removing it
4074 * from the namespace in any way.
4075 */
4076int
4077spa_reset(char *pool)
4078{
4079	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4080	    B_FALSE, B_FALSE));
4081}
4082
4083/*
4084 * ==========================================================================
4085 * Device manipulation
4086 * ==========================================================================
4087 */
4088
4089/*
4090 * Add a device to a storage pool.
4091 */
4092int
4093spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4094{
4095	uint64_t txg, id;
4096	int error;
4097	vdev_t *rvd = spa->spa_root_vdev;
4098	vdev_t *vd, *tvd;
4099	nvlist_t **spares, **l2cache;
4100	uint_t nspares, nl2cache;
4101
4102	ASSERT(spa_writeable(spa));
4103
4104	txg = spa_vdev_enter(spa);
4105
4106	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4107	    VDEV_ALLOC_ADD)) != 0)
4108		return (spa_vdev_exit(spa, NULL, txg, error));
4109
4110	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
4111
4112	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4113	    &nspares) != 0)
4114		nspares = 0;
4115
4116	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4117	    &nl2cache) != 0)
4118		nl2cache = 0;
4119
4120	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4121		return (spa_vdev_exit(spa, vd, txg, EINVAL));
4122
4123	if (vd->vdev_children != 0 &&
4124	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
4125		return (spa_vdev_exit(spa, vd, txg, error));
4126
4127	/*
4128	 * We must validate the spares and l2cache devices after checking the
4129	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
4130	 */
4131	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4132		return (spa_vdev_exit(spa, vd, txg, error));
4133
4134	/*
4135	 * Transfer each new top-level vdev from vd to rvd.
4136	 */
4137	for (int c = 0; c < vd->vdev_children; c++) {
4138
4139		/*
4140		 * Set the vdev id to the first hole, if one exists.
4141		 */
4142		for (id = 0; id < rvd->vdev_children; id++) {
4143			if (rvd->vdev_child[id]->vdev_ishole) {
4144				vdev_free(rvd->vdev_child[id]);
4145				break;
4146			}
4147		}
4148		tvd = vd->vdev_child[c];
4149		vdev_remove_child(vd, tvd);
4150		tvd->vdev_id = id;
4151		vdev_add_child(rvd, tvd);
4152		vdev_config_dirty(tvd);
4153	}
4154
4155	if (nspares != 0) {
4156		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4157		    ZPOOL_CONFIG_SPARES);
4158		spa_load_spares(spa);
4159		spa->spa_spares.sav_sync = B_TRUE;
4160	}
4161
4162	if (nl2cache != 0) {
4163		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4164		    ZPOOL_CONFIG_L2CACHE);
4165		spa_load_l2cache(spa);
4166		spa->spa_l2cache.sav_sync = B_TRUE;
4167	}
4168
4169	/*
4170	 * We have to be careful when adding new vdevs to an existing pool.
4171	 * If other threads start allocating from these vdevs before we
4172	 * sync the config cache, and we lose power, then upon reboot we may
4173	 * fail to open the pool because there are DVAs that the config cache
4174	 * can't translate.  Therefore, we first add the vdevs without
4175	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4176	 * and then let spa_config_update() initialize the new metaslabs.
4177	 *
4178	 * spa_load() checks for added-but-not-initialized vdevs, so that
4179	 * if we lose power at any point in this sequence, the remaining
4180	 * steps will be completed the next time we load the pool.
4181	 */
4182	(void) spa_vdev_exit(spa, vd, txg, 0);
4183
4184	mutex_enter(&spa_namespace_lock);
4185	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4186	mutex_exit(&spa_namespace_lock);
4187
4188	return (0);
4189}
4190
4191/*
4192 * Attach a device to a mirror.  The arguments are the path to any device
4193 * in the mirror, and the nvroot for the new device.  If the path specifies
4194 * a device that is not mirrored, we automatically insert the mirror vdev.
4195 *
4196 * If 'replacing' is specified, the new device is intended to replace the
4197 * existing device; in this case the two devices are made into their own
4198 * mirror using the 'replacing' vdev, which is functionally identical to
4199 * the mirror vdev (it actually reuses all the same ops) but has a few
4200 * extra rules: you can't attach to it after it's been created, and upon
4201 * completion of resilvering, the first disk (the one being replaced)
4202 * is automatically detached.
4203 */
4204int
4205spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4206{
4207	uint64_t txg, dtl_max_txg;
4208	vdev_t *rvd = spa->spa_root_vdev;
4209	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4210	vdev_ops_t *pvops;
4211	char *oldvdpath, *newvdpath;
4212	int newvd_isspare;
4213	int error;
4214
4215	ASSERT(spa_writeable(spa));
4216
4217	txg = spa_vdev_enter(spa);
4218
4219	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4220
4221	if (oldvd == NULL)
4222		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4223
4224	if (!oldvd->vdev_ops->vdev_op_leaf)
4225		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4226
4227	pvd = oldvd->vdev_parent;
4228
4229	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4230	    VDEV_ALLOC_ATTACH)) != 0)
4231		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4232
4233	if (newrootvd->vdev_children != 1)
4234		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4235
4236	newvd = newrootvd->vdev_child[0];
4237
4238	if (!newvd->vdev_ops->vdev_op_leaf)
4239		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4240
4241	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4242		return (spa_vdev_exit(spa, newrootvd, txg, error));
4243
4244	/*
4245	 * Spares can't replace logs
4246	 */
4247	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4248		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4249
4250	if (!replacing) {
4251		/*
4252		 * For attach, the only allowable parent is a mirror or the root
4253		 * vdev.
4254		 */
4255		if (pvd->vdev_ops != &vdev_mirror_ops &&
4256		    pvd->vdev_ops != &vdev_root_ops)
4257			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4258
4259		pvops = &vdev_mirror_ops;
4260	} else {
4261		/*
4262		 * Active hot spares can only be replaced by inactive hot
4263		 * spares.
4264		 */
4265		if (pvd->vdev_ops == &vdev_spare_ops &&
4266		    oldvd->vdev_isspare &&
4267		    !spa_has_spare(spa, newvd->vdev_guid))
4268			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4269
4270		/*
4271		 * If the source is a hot spare, and the parent isn't already a
4272		 * spare, then we want to create a new hot spare.  Otherwise, we
4273		 * want to create a replacing vdev.  The user is not allowed to
4274		 * attach to a spared vdev child unless the 'isspare' state is
4275		 * the same (spare replaces spare, non-spare replaces
4276		 * non-spare).
4277		 */
4278		if (pvd->vdev_ops == &vdev_replacing_ops &&
4279		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4280			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4281		} else if (pvd->vdev_ops == &vdev_spare_ops &&
4282		    newvd->vdev_isspare != oldvd->vdev_isspare) {
4283			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4284		}
4285
4286		if (newvd->vdev_isspare)
4287			pvops = &vdev_spare_ops;
4288		else
4289			pvops = &vdev_replacing_ops;
4290	}
4291
4292	/*
4293	 * Make sure the new device is big enough.
4294	 */
4295	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4296		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4297
4298	/*
4299	 * The new device cannot have a higher alignment requirement
4300	 * than the top-level vdev.
4301	 */
4302	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4303		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4304
4305	/*
4306	 * If this is an in-place replacement, update oldvd's path and devid
4307	 * to make it distinguishable from newvd, and unopenable from now on.
4308	 */
4309	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4310		spa_strfree(oldvd->vdev_path);
4311		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4312		    KM_SLEEP);
4313		(void) sprintf(oldvd->vdev_path, "%s/%s",
4314		    newvd->vdev_path, "old");
4315		if (oldvd->vdev_devid != NULL) {
4316			spa_strfree(oldvd->vdev_devid);
4317			oldvd->vdev_devid = NULL;
4318		}
4319	}
4320
4321	/* mark the device being resilvered */
4322	newvd->vdev_resilvering = B_TRUE;
4323
4324	/*
4325	 * If the parent is not a mirror, or if we're replacing, insert the new
4326	 * mirror/replacing/spare vdev above oldvd.
4327	 */
4328	if (pvd->vdev_ops != pvops)
4329		pvd = vdev_add_parent(oldvd, pvops);
4330
4331	ASSERT(pvd->vdev_top->vdev_parent == rvd);
4332	ASSERT(pvd->vdev_ops == pvops);
4333	ASSERT(oldvd->vdev_parent == pvd);
4334
4335	/*
4336	 * Extract the new device from its root and add it to pvd.
4337	 */
4338	vdev_remove_child(newrootvd, newvd);
4339	newvd->vdev_id = pvd->vdev_children;
4340	newvd->vdev_crtxg = oldvd->vdev_crtxg;
4341	vdev_add_child(pvd, newvd);
4342
4343	tvd = newvd->vdev_top;
4344	ASSERT(pvd->vdev_top == tvd);
4345	ASSERT(tvd->vdev_parent == rvd);
4346
4347	vdev_config_dirty(tvd);
4348
4349	/*
4350	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4351	 * for any dmu_sync-ed blocks.  It will propagate upward when
4352	 * spa_vdev_exit() calls vdev_dtl_reassess().
4353	 */
4354	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4355
4356	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4357	    dtl_max_txg - TXG_INITIAL);
4358
4359	if (newvd->vdev_isspare) {
4360		spa_spare_activate(newvd);
4361		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4362	}
4363
4364	oldvdpath = spa_strdup(oldvd->vdev_path);
4365	newvdpath = spa_strdup(newvd->vdev_path);
4366	newvd_isspare = newvd->vdev_isspare;
4367
4368	/*
4369	 * Mark newvd's DTL dirty in this txg.
4370	 */
4371	vdev_dirty(tvd, VDD_DTL, newvd, txg);
4372
4373	/*
4374	 * Restart the resilver
4375	 */
4376	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4377
4378	/*
4379	 * Commit the config
4380	 */
4381	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4382
4383	spa_history_log_internal(spa, "vdev attach", NULL,
4384	    "%s vdev=%s %s vdev=%s",
4385	    replacing && newvd_isspare ? "spare in" :
4386	    replacing ? "replace" : "attach", newvdpath,
4387	    replacing ? "for" : "to", oldvdpath);
4388
4389	spa_strfree(oldvdpath);
4390	spa_strfree(newvdpath);
4391
4392	if (spa->spa_bootfs)
4393		spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4394
4395	return (0);
4396}
4397
4398/*
4399 * Detach a device from a mirror or replacing vdev.
4400 * If 'replace_done' is specified, only detach if the parent
4401 * is a replacing vdev.
4402 */
4403int
4404spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4405{
4406	uint64_t txg;
4407	int error;
4408	vdev_t *rvd = spa->spa_root_vdev;
4409	vdev_t *vd, *pvd, *cvd, *tvd;
4410	boolean_t unspare = B_FALSE;
4411	uint64_t unspare_guid;
4412	char *vdpath;
4413
4414	ASSERT(spa_writeable(spa));
4415
4416	txg = spa_vdev_enter(spa);
4417
4418	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4419
4420	if (vd == NULL)
4421		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4422
4423	if (!vd->vdev_ops->vdev_op_leaf)
4424		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4425
4426	pvd = vd->vdev_parent;
4427
4428	/*
4429	 * If the parent/child relationship is not as expected, don't do it.
4430	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4431	 * vdev that's replacing B with C.  The user's intent in replacing
4432	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4433	 * the replace by detaching C, the expected behavior is to end up
4434	 * M(A,B).  But suppose that right after deciding to detach C,
4435	 * the replacement of B completes.  We would have M(A,C), and then
4436	 * ask to detach C, which would leave us with just A -- not what
4437	 * the user wanted.  To prevent this, we make sure that the
4438	 * parent/child relationship hasn't changed -- in this example,
4439	 * that C's parent is still the replacing vdev R.
4440	 */
4441	if (pvd->vdev_guid != pguid && pguid != 0)
4442		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4443
4444	/*
4445	 * Only 'replacing' or 'spare' vdevs can be replaced.
4446	 */
4447	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4448	    pvd->vdev_ops != &vdev_spare_ops)
4449		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4450
4451	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4452	    spa_version(spa) >= SPA_VERSION_SPARES);
4453
4454	/*
4455	 * Only mirror, replacing, and spare vdevs support detach.
4456	 */
4457	if (pvd->vdev_ops != &vdev_replacing_ops &&
4458	    pvd->vdev_ops != &vdev_mirror_ops &&
4459	    pvd->vdev_ops != &vdev_spare_ops)
4460		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4461
4462	/*
4463	 * If this device has the only valid copy of some data,
4464	 * we cannot safely detach it.
4465	 */
4466	if (vdev_dtl_required(vd))
4467		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4468
4469	ASSERT(pvd->vdev_children >= 2);
4470
4471	/*
4472	 * If we are detaching the second disk from a replacing vdev, then
4473	 * check to see if we changed the original vdev's path to have "/old"
4474	 * at the end in spa_vdev_attach().  If so, undo that change now.
4475	 */
4476	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4477	    vd->vdev_path != NULL) {
4478		size_t len = strlen(vd->vdev_path);
4479
4480		for (int c = 0; c < pvd->vdev_children; c++) {
4481			cvd = pvd->vdev_child[c];
4482
4483			if (cvd == vd || cvd->vdev_path == NULL)
4484				continue;
4485
4486			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4487			    strcmp(cvd->vdev_path + len, "/old") == 0) {
4488				spa_strfree(cvd->vdev_path);
4489				cvd->vdev_path = spa_strdup(vd->vdev_path);
4490				break;
4491			}
4492		}
4493	}
4494
4495	/*
4496	 * If we are detaching the original disk from a spare, then it implies
4497	 * that the spare should become a real disk, and be removed from the
4498	 * active spare list for the pool.
4499	 */
4500	if (pvd->vdev_ops == &vdev_spare_ops &&
4501	    vd->vdev_id == 0 &&
4502	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4503		unspare = B_TRUE;
4504
4505	/*
4506	 * Erase the disk labels so the disk can be used for other things.
4507	 * This must be done after all other error cases are handled,
4508	 * but before we disembowel vd (so we can still do I/O to it).
4509	 * But if we can't do it, don't treat the error as fatal --
4510	 * it may be that the unwritability of the disk is the reason
4511	 * it's being detached!
4512	 */
4513	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4514
4515	/*
4516	 * Remove vd from its parent and compact the parent's children.
4517	 */
4518	vdev_remove_child(pvd, vd);
4519	vdev_compact_children(pvd);
4520
4521	/*
4522	 * Remember one of the remaining children so we can get tvd below.
4523	 */
4524	cvd = pvd->vdev_child[pvd->vdev_children - 1];
4525
4526	/*
4527	 * If we need to remove the remaining child from the list of hot spares,
4528	 * do it now, marking the vdev as no longer a spare in the process.
4529	 * We must do this before vdev_remove_parent(), because that can
4530	 * change the GUID if it creates a new toplevel GUID.  For a similar
4531	 * reason, we must remove the spare now, in the same txg as the detach;
4532	 * otherwise someone could attach a new sibling, change the GUID, and
4533	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4534	 */
4535	if (unspare) {
4536		ASSERT(cvd->vdev_isspare);
4537		spa_spare_remove(cvd);
4538		unspare_guid = cvd->vdev_guid;
4539		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4540		cvd->vdev_unspare = B_TRUE;
4541	}
4542
4543	/*
4544	 * If the parent mirror/replacing vdev only has one child,
4545	 * the parent is no longer needed.  Remove it from the tree.
4546	 */
4547	if (pvd->vdev_children == 1) {
4548		if (pvd->vdev_ops == &vdev_spare_ops)
4549			cvd->vdev_unspare = B_FALSE;
4550		vdev_remove_parent(cvd);
4551		cvd->vdev_resilvering = B_FALSE;
4552	}
4553
4554
4555	/*
4556	 * We don't set tvd until now because the parent we just removed
4557	 * may have been the previous top-level vdev.
4558	 */
4559	tvd = cvd->vdev_top;
4560	ASSERT(tvd->vdev_parent == rvd);
4561
4562	/*
4563	 * Reevaluate the parent vdev state.
4564	 */
4565	vdev_propagate_state(cvd);
4566
4567	/*
4568	 * If the 'autoexpand' property is set on the pool then automatically
4569	 * try to expand the size of the pool. For example if the device we
4570	 * just detached was smaller than the others, it may be possible to
4571	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4572	 * first so that we can obtain the updated sizes of the leaf vdevs.
4573	 */
4574	if (spa->spa_autoexpand) {
4575		vdev_reopen(tvd);
4576		vdev_expand(tvd, txg);
4577	}
4578
4579	vdev_config_dirty(tvd);
4580
4581	/*
4582	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4583	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4584	 * But first make sure we're not on any *other* txg's DTL list, to
4585	 * prevent vd from being accessed after it's freed.
4586	 */
4587	vdpath = spa_strdup(vd->vdev_path);
4588	for (int t = 0; t < TXG_SIZE; t++)
4589		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4590	vd->vdev_detached = B_TRUE;
4591	vdev_dirty(tvd, VDD_DTL, vd, txg);
4592
4593	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4594
4595	/* hang on to the spa before we release the lock */
4596	spa_open_ref(spa, FTAG);
4597
4598	error = spa_vdev_exit(spa, vd, txg, 0);
4599
4600	spa_history_log_internal(spa, "detach", NULL,
4601	    "vdev=%s", vdpath);
4602	spa_strfree(vdpath);
4603
4604	/*
4605	 * If this was the removal of the original device in a hot spare vdev,
4606	 * then we want to go through and remove the device from the hot spare
4607	 * list of every other pool.
4608	 */
4609	if (unspare) {
4610		spa_t *altspa = NULL;
4611
4612		mutex_enter(&spa_namespace_lock);
4613		while ((altspa = spa_next(altspa)) != NULL) {
4614			if (altspa->spa_state != POOL_STATE_ACTIVE ||
4615			    altspa == spa)
4616				continue;
4617
4618			spa_open_ref(altspa, FTAG);
4619			mutex_exit(&spa_namespace_lock);
4620			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4621			mutex_enter(&spa_namespace_lock);
4622			spa_close(altspa, FTAG);
4623		}
4624		mutex_exit(&spa_namespace_lock);
4625
4626		/* search the rest of the vdevs for spares to remove */
4627		spa_vdev_resilver_done(spa);
4628	}
4629
4630	/* all done with the spa; OK to release */
4631	mutex_enter(&spa_namespace_lock);
4632	spa_close(spa, FTAG);
4633	mutex_exit(&spa_namespace_lock);
4634
4635	return (error);
4636}
4637
4638/*
4639 * Split a set of devices from their mirrors, and create a new pool from them.
4640 */
4641int
4642spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4643    nvlist_t *props, boolean_t exp)
4644{
4645	int error = 0;
4646	uint64_t txg, *glist;
4647	spa_t *newspa;
4648	uint_t c, children, lastlog;
4649	nvlist_t **child, *nvl, *tmp;
4650	dmu_tx_t *tx;
4651	char *altroot = NULL;
4652	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4653	boolean_t activate_slog;
4654
4655	ASSERT(spa_writeable(spa));
4656
4657	txg = spa_vdev_enter(spa);
4658
4659	/* clear the log and flush everything up to now */
4660	activate_slog = spa_passivate_log(spa);
4661	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4662	error = spa_offline_log(spa);
4663	txg = spa_vdev_config_enter(spa);
4664
4665	if (activate_slog)
4666		spa_activate_log(spa);
4667
4668	if (error != 0)
4669		return (spa_vdev_exit(spa, NULL, txg, error));
4670
4671	/* check new spa name before going any further */
4672	if (spa_lookup(newname) != NULL)
4673		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4674
4675	/*
4676	 * scan through all the children to ensure they're all mirrors
4677	 */
4678	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4679	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4680	    &children) != 0)
4681		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4682
4683	/* first, check to ensure we've got the right child count */
4684	rvd = spa->spa_root_vdev;
4685	lastlog = 0;
4686	for (c = 0; c < rvd->vdev_children; c++) {
4687		vdev_t *vd = rvd->vdev_child[c];
4688
4689		/* don't count the holes & logs as children */
4690		if (vd->vdev_islog || vd->vdev_ishole) {
4691			if (lastlog == 0)
4692				lastlog = c;
4693			continue;
4694		}
4695
4696		lastlog = 0;
4697	}
4698	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4699		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4700
4701	/* next, ensure no spare or cache devices are part of the split */
4702	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4703	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4704		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4705
4706	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4707	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4708
4709	/* then, loop over each vdev and validate it */
4710	for (c = 0; c < children; c++) {
4711		uint64_t is_hole = 0;
4712
4713		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4714		    &is_hole);
4715
4716		if (is_hole != 0) {
4717			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4718			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4719				continue;
4720			} else {
4721				error = EINVAL;
4722				break;
4723			}
4724		}
4725
4726		/* which disk is going to be split? */
4727		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4728		    &glist[c]) != 0) {
4729			error = EINVAL;
4730			break;
4731		}
4732
4733		/* look it up in the spa */
4734		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4735		if (vml[c] == NULL) {
4736			error = ENODEV;
4737			break;
4738		}
4739
4740		/* make sure there's nothing stopping the split */
4741		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4742		    vml[c]->vdev_islog ||
4743		    vml[c]->vdev_ishole ||
4744		    vml[c]->vdev_isspare ||
4745		    vml[c]->vdev_isl2cache ||
4746		    !vdev_writeable(vml[c]) ||
4747		    vml[c]->vdev_children != 0 ||
4748		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4749		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4750			error = EINVAL;
4751			break;
4752		}
4753
4754		if (vdev_dtl_required(vml[c])) {
4755			error = EBUSY;
4756			break;
4757		}
4758
4759		/* we need certain info from the top level */
4760		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4761		    vml[c]->vdev_top->vdev_ms_array) == 0);
4762		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4763		    vml[c]->vdev_top->vdev_ms_shift) == 0);
4764		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4765		    vml[c]->vdev_top->vdev_asize) == 0);
4766		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4767		    vml[c]->vdev_top->vdev_ashift) == 0);
4768	}
4769
4770	if (error != 0) {
4771		kmem_free(vml, children * sizeof (vdev_t *));
4772		kmem_free(glist, children * sizeof (uint64_t));
4773		return (spa_vdev_exit(spa, NULL, txg, error));
4774	}
4775
4776	/* stop writers from using the disks */
4777	for (c = 0; c < children; c++) {
4778		if (vml[c] != NULL)
4779			vml[c]->vdev_offline = B_TRUE;
4780	}
4781	vdev_reopen(spa->spa_root_vdev);
4782
4783	/*
4784	 * Temporarily record the splitting vdevs in the spa config.  This
4785	 * will disappear once the config is regenerated.
4786	 */
4787	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4788	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4789	    glist, children) == 0);
4790	kmem_free(glist, children * sizeof (uint64_t));
4791
4792	mutex_enter(&spa->spa_props_lock);
4793	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4794	    nvl) == 0);
4795	mutex_exit(&spa->spa_props_lock);
4796	spa->spa_config_splitting = nvl;
4797	vdev_config_dirty(spa->spa_root_vdev);
4798
4799	/* configure and create the new pool */
4800	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4801	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4802	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4803	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4804	    spa_version(spa)) == 0);
4805	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4806	    spa->spa_config_txg) == 0);
4807	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4808	    spa_generate_guid(NULL)) == 0);
4809	(void) nvlist_lookup_string(props,
4810	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4811
4812	/* add the new pool to the namespace */
4813	newspa = spa_add(newname, config, altroot);
4814	newspa->spa_config_txg = spa->spa_config_txg;
4815	spa_set_log_state(newspa, SPA_LOG_CLEAR);
4816
4817	/* release the spa config lock, retaining the namespace lock */
4818	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4819
4820	if (zio_injection_enabled)
4821		zio_handle_panic_injection(spa, FTAG, 1);
4822
4823	spa_activate(newspa, spa_mode_global);
4824	spa_async_suspend(newspa);
4825
4826	/* create the new pool from the disks of the original pool */
4827	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4828	if (error)
4829		goto out;
4830
4831	/* if that worked, generate a real config for the new pool */
4832	if (newspa->spa_root_vdev != NULL) {
4833		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4834		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4835		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4836		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4837		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4838		    B_TRUE));
4839	}
4840
4841	/* set the props */
4842	if (props != NULL) {
4843		spa_configfile_set(newspa, props, B_FALSE);
4844		error = spa_prop_set(newspa, props);
4845		if (error)
4846			goto out;
4847	}
4848
4849	/* flush everything */
4850	txg = spa_vdev_config_enter(newspa);
4851	vdev_config_dirty(newspa->spa_root_vdev);
4852	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4853
4854	if (zio_injection_enabled)
4855		zio_handle_panic_injection(spa, FTAG, 2);
4856
4857	spa_async_resume(newspa);
4858
4859	/* finally, update the original pool's config */
4860	txg = spa_vdev_config_enter(spa);
4861	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4862	error = dmu_tx_assign(tx, TXG_WAIT);
4863	if (error != 0)
4864		dmu_tx_abort(tx);
4865	for (c = 0; c < children; c++) {
4866		if (vml[c] != NULL) {
4867			vdev_split(vml[c]);
4868			if (error == 0)
4869				spa_history_log_internal(spa, "detach", tx,
4870				    "vdev=%s", vml[c]->vdev_path);
4871			vdev_free(vml[c]);
4872		}
4873	}
4874	vdev_config_dirty(spa->spa_root_vdev);
4875	spa->spa_config_splitting = NULL;
4876	nvlist_free(nvl);
4877	if (error == 0)
4878		dmu_tx_commit(tx);
4879	(void) spa_vdev_exit(spa, NULL, txg, 0);
4880
4881	if (zio_injection_enabled)
4882		zio_handle_panic_injection(spa, FTAG, 3);
4883
4884	/* split is complete; log a history record */
4885	spa_history_log_internal(newspa, "split", NULL,
4886	    "from pool %s", spa_name(spa));
4887
4888	kmem_free(vml, children * sizeof (vdev_t *));
4889
4890	/* if we're not going to mount the filesystems in userland, export */
4891	if (exp)
4892		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4893		    B_FALSE, B_FALSE);
4894
4895	return (error);
4896
4897out:
4898	spa_unload(newspa);
4899	spa_deactivate(newspa);
4900	spa_remove(newspa);
4901
4902	txg = spa_vdev_config_enter(spa);
4903
4904	/* re-online all offlined disks */
4905	for (c = 0; c < children; c++) {
4906		if (vml[c] != NULL)
4907			vml[c]->vdev_offline = B_FALSE;
4908	}
4909	vdev_reopen(spa->spa_root_vdev);
4910
4911	nvlist_free(spa->spa_config_splitting);
4912	spa->spa_config_splitting = NULL;
4913	(void) spa_vdev_exit(spa, NULL, txg, error);
4914
4915	kmem_free(vml, children * sizeof (vdev_t *));
4916	return (error);
4917}
4918
4919static nvlist_t *
4920spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4921{
4922	for (int i = 0; i < count; i++) {
4923		uint64_t guid;
4924
4925		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4926		    &guid) == 0);
4927
4928		if (guid == target_guid)
4929			return (nvpp[i]);
4930	}
4931
4932	return (NULL);
4933}
4934
4935static void
4936spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4937	nvlist_t *dev_to_remove)
4938{
4939	nvlist_t **newdev = NULL;
4940
4941	if (count > 1)
4942		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4943
4944	for (int i = 0, j = 0; i < count; i++) {
4945		if (dev[i] == dev_to_remove)
4946			continue;
4947		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4948	}
4949
4950	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4951	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4952
4953	for (int i = 0; i < count - 1; i++)
4954		nvlist_free(newdev[i]);
4955
4956	if (count > 1)
4957		kmem_free(newdev, (count - 1) * sizeof (void *));
4958}
4959
4960/*
4961 * Evacuate the device.
4962 */
4963static int
4964spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4965{
4966	uint64_t txg;
4967	int error = 0;
4968
4969	ASSERT(MUTEX_HELD(&spa_namespace_lock));
4970	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4971	ASSERT(vd == vd->vdev_top);
4972
4973	/*
4974	 * Evacuate the device.  We don't hold the config lock as writer
4975	 * since we need to do I/O but we do keep the
4976	 * spa_namespace_lock held.  Once this completes the device
4977	 * should no longer have any blocks allocated on it.
4978	 */
4979	if (vd->vdev_islog) {
4980		if (vd->vdev_stat.vs_alloc != 0)
4981			error = spa_offline_log(spa);
4982	} else {
4983		error = ENOTSUP;
4984	}
4985
4986	if (error)
4987		return (error);
4988
4989	/*
4990	 * The evacuation succeeded.  Remove any remaining MOS metadata
4991	 * associated with this vdev, and wait for these changes to sync.
4992	 */
4993	ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
4994	txg = spa_vdev_config_enter(spa);
4995	vd->vdev_removing = B_TRUE;
4996	vdev_dirty(vd, 0, NULL, txg);
4997	vdev_config_dirty(vd);
4998	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4999
5000	return (0);
5001}
5002
5003/*
5004 * Complete the removal by cleaning up the namespace.
5005 */
5006static void
5007spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5008{
5009	vdev_t *rvd = spa->spa_root_vdev;
5010	uint64_t id = vd->vdev_id;
5011	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5012
5013	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5014	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5015	ASSERT(vd == vd->vdev_top);
5016
5017	/*
5018	 * Only remove any devices which are empty.
5019	 */
5020	if (vd->vdev_stat.vs_alloc != 0)
5021		return;
5022
5023	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5024
5025	if (list_link_active(&vd->vdev_state_dirty_node))
5026		vdev_state_clean(vd);
5027	if (list_link_active(&vd->vdev_config_dirty_node))
5028		vdev_config_clean(vd);
5029
5030	vdev_free(vd);
5031
5032	if (last_vdev) {
5033		vdev_compact_children(rvd);
5034	} else {
5035		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5036		vdev_add_child(rvd, vd);
5037	}
5038	vdev_config_dirty(rvd);
5039
5040	/*
5041	 * Reassess the health of our root vdev.
5042	 */
5043	vdev_reopen(rvd);
5044}
5045
5046/*
5047 * Remove a device from the pool -
5048 *
5049 * Removing a device from the vdev namespace requires several steps
5050 * and can take a significant amount of time.  As a result we use
5051 * the spa_vdev_config_[enter/exit] functions which allow us to
5052 * grab and release the spa_config_lock while still holding the namespace
5053 * lock.  During each step the configuration is synced out.
5054 */
5055
5056/*
5057 * Remove a device from the pool.  Currently, this supports removing only hot
5058 * spares, slogs, and level 2 ARC devices.
5059 */
5060int
5061spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5062{
5063	vdev_t *vd;
5064	metaslab_group_t *mg;
5065	nvlist_t **spares, **l2cache, *nv;
5066	uint64_t txg = 0;
5067	uint_t nspares, nl2cache;
5068	int error = 0;
5069	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5070
5071	ASSERT(spa_writeable(spa));
5072
5073	if (!locked)
5074		txg = spa_vdev_enter(spa);
5075
5076	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5077
5078	if (spa->spa_spares.sav_vdevs != NULL &&
5079	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5080	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5081	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5082		/*
5083		 * Only remove the hot spare if it's not currently in use
5084		 * in this pool.
5085		 */
5086		if (vd == NULL || unspare) {
5087			spa_vdev_remove_aux(spa->spa_spares.sav_config,
5088			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5089			spa_load_spares(spa);
5090			spa->spa_spares.sav_sync = B_TRUE;
5091		} else {
5092			error = EBUSY;
5093		}
5094	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
5095	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5096	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5097	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5098		/*
5099		 * Cache devices can always be removed.
5100		 */
5101		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5102		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5103		spa_load_l2cache(spa);
5104		spa->spa_l2cache.sav_sync = B_TRUE;
5105	} else if (vd != NULL && vd->vdev_islog) {
5106		ASSERT(!locked);
5107		ASSERT(vd == vd->vdev_top);
5108
5109		/*
5110		 * XXX - Once we have bp-rewrite this should
5111		 * become the common case.
5112		 */
5113
5114		mg = vd->vdev_mg;
5115
5116		/*
5117		 * Stop allocating from this vdev.
5118		 */
5119		metaslab_group_passivate(mg);
5120
5121		/*
5122		 * Wait for the youngest allocations and frees to sync,
5123		 * and then wait for the deferral of those frees to finish.
5124		 */
5125		spa_vdev_config_exit(spa, NULL,
5126		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5127
5128		/*
5129		 * Attempt to evacuate the vdev.
5130		 */
5131		error = spa_vdev_remove_evacuate(spa, vd);
5132
5133		txg = spa_vdev_config_enter(spa);
5134
5135		/*
5136		 * If we couldn't evacuate the vdev, unwind.
5137		 */
5138		if (error) {
5139			metaslab_group_activate(mg);
5140			return (spa_vdev_exit(spa, NULL, txg, error));
5141		}
5142
5143		/*
5144		 * Clean up the vdev namespace.
5145		 */
5146		spa_vdev_remove_from_namespace(spa, vd);
5147
5148	} else if (vd != NULL) {
5149		/*
5150		 * Normal vdevs cannot be removed (yet).
5151		 */
5152		error = ENOTSUP;
5153	} else {
5154		/*
5155		 * There is no vdev of any kind with the specified guid.
5156		 */
5157		error = ENOENT;
5158	}
5159
5160	if (!locked)
5161		return (spa_vdev_exit(spa, NULL, txg, error));
5162
5163	return (error);
5164}
5165
5166/*
5167 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5168 * current spared, so we can detach it.
5169 */
5170static vdev_t *
5171spa_vdev_resilver_done_hunt(vdev_t *vd)
5172{
5173	vdev_t *newvd, *oldvd;
5174
5175	for (int c = 0; c < vd->vdev_children; c++) {
5176		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5177		if (oldvd != NULL)
5178			return (oldvd);
5179	}
5180
5181	/*
5182	 * Check for a completed replacement.  We always consider the first
5183	 * vdev in the list to be the oldest vdev, and the last one to be
5184	 * the newest (see spa_vdev_attach() for how that works).  In
5185	 * the case where the newest vdev is faulted, we will not automatically
5186	 * remove it after a resilver completes.  This is OK as it will require
5187	 * user intervention to determine which disk the admin wishes to keep.
5188	 */
5189	if (vd->vdev_ops == &vdev_replacing_ops) {
5190		ASSERT(vd->vdev_children > 1);
5191
5192		newvd = vd->vdev_child[vd->vdev_children - 1];
5193		oldvd = vd->vdev_child[0];
5194
5195		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5196		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5197		    !vdev_dtl_required(oldvd))
5198			return (oldvd);
5199	}
5200
5201	/*
5202	 * Check for a completed resilver with the 'unspare' flag set.
5203	 */
5204	if (vd->vdev_ops == &vdev_spare_ops) {
5205		vdev_t *first = vd->vdev_child[0];
5206		vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5207
5208		if (last->vdev_unspare) {
5209			oldvd = first;
5210			newvd = last;
5211		} else if (first->vdev_unspare) {
5212			oldvd = last;
5213			newvd = first;
5214		} else {
5215			oldvd = NULL;
5216		}
5217
5218		if (oldvd != NULL &&
5219		    vdev_dtl_empty(newvd, DTL_MISSING) &&
5220		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5221		    !vdev_dtl_required(oldvd))
5222			return (oldvd);
5223
5224		/*
5225		 * If there are more than two spares attached to a disk,
5226		 * and those spares are not required, then we want to
5227		 * attempt to free them up now so that they can be used
5228		 * by other pools.  Once we're back down to a single
5229		 * disk+spare, we stop removing them.
5230		 */
5231		if (vd->vdev_children > 2) {
5232			newvd = vd->vdev_child[1];
5233
5234			if (newvd->vdev_isspare && last->vdev_isspare &&
5235			    vdev_dtl_empty(last, DTL_MISSING) &&
5236			    vdev_dtl_empty(last, DTL_OUTAGE) &&
5237			    !vdev_dtl_required(newvd))
5238				return (newvd);
5239		}
5240	}
5241
5242	return (NULL);
5243}
5244
5245static void
5246spa_vdev_resilver_done(spa_t *spa)
5247{
5248	vdev_t *vd, *pvd, *ppvd;
5249	uint64_t guid, sguid, pguid, ppguid;
5250
5251	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5252
5253	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5254		pvd = vd->vdev_parent;
5255		ppvd = pvd->vdev_parent;
5256		guid = vd->vdev_guid;
5257		pguid = pvd->vdev_guid;
5258		ppguid = ppvd->vdev_guid;
5259		sguid = 0;
5260		/*
5261		 * If we have just finished replacing a hot spared device, then
5262		 * we need to detach the parent's first child (the original hot
5263		 * spare) as well.
5264		 */
5265		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5266		    ppvd->vdev_children == 2) {
5267			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5268			sguid = ppvd->vdev_child[1]->vdev_guid;
5269		}
5270		spa_config_exit(spa, SCL_ALL, FTAG);
5271		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5272			return;
5273		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5274			return;
5275		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5276	}
5277
5278	spa_config_exit(spa, SCL_ALL, FTAG);
5279}
5280
5281/*
5282 * Update the stored path or FRU for this vdev.
5283 */
5284int
5285spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5286    boolean_t ispath)
5287{
5288	vdev_t *vd;
5289	boolean_t sync = B_FALSE;
5290
5291	ASSERT(spa_writeable(spa));
5292
5293	spa_vdev_state_enter(spa, SCL_ALL);
5294
5295	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5296		return (spa_vdev_state_exit(spa, NULL, ENOENT));
5297
5298	if (!vd->vdev_ops->vdev_op_leaf)
5299		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5300
5301	if (ispath) {
5302		if (strcmp(value, vd->vdev_path) != 0) {
5303			spa_strfree(vd->vdev_path);
5304			vd->vdev_path = spa_strdup(value);
5305			sync = B_TRUE;
5306		}
5307	} else {
5308		if (vd->vdev_fru == NULL) {
5309			vd->vdev_fru = spa_strdup(value);
5310			sync = B_TRUE;
5311		} else if (strcmp(value, vd->vdev_fru) != 0) {
5312			spa_strfree(vd->vdev_fru);
5313			vd->vdev_fru = spa_strdup(value);
5314			sync = B_TRUE;
5315		}
5316	}
5317
5318	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5319}
5320
5321int
5322spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5323{
5324	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5325}
5326
5327int
5328spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5329{
5330	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5331}
5332
5333/*
5334 * ==========================================================================
5335 * SPA Scanning
5336 * ==========================================================================
5337 */
5338
5339int
5340spa_scan_stop(spa_t *spa)
5341{
5342	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5343	if (dsl_scan_resilvering(spa->spa_dsl_pool))
5344		return (EBUSY);
5345	return (dsl_scan_cancel(spa->spa_dsl_pool));
5346}
5347
5348int
5349spa_scan(spa_t *spa, pool_scan_func_t func)
5350{
5351	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5352
5353	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5354		return (ENOTSUP);
5355
5356	/*
5357	 * If a resilver was requested, but there is no DTL on a
5358	 * writeable leaf device, we have nothing to do.
5359	 */
5360	if (func == POOL_SCAN_RESILVER &&
5361	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5362		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5363		return (0);
5364	}
5365
5366	return (dsl_scan(spa->spa_dsl_pool, func));
5367}
5368
5369/*
5370 * ==========================================================================
5371 * SPA async task processing
5372 * ==========================================================================
5373 */
5374
5375static void
5376spa_async_remove(spa_t *spa, vdev_t *vd)
5377{
5378	if (vd->vdev_remove_wanted) {
5379		vd->vdev_remove_wanted = B_FALSE;
5380		vd->vdev_delayed_close = B_FALSE;
5381		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5382
5383		/*
5384		 * We want to clear the stats, but we don't want to do a full
5385		 * vdev_clear() as that will cause us to throw away
5386		 * degraded/faulted state as well as attempt to reopen the
5387		 * device, all of which is a waste.
5388		 */
5389		vd->vdev_stat.vs_read_errors = 0;
5390		vd->vdev_stat.vs_write_errors = 0;
5391		vd->vdev_stat.vs_checksum_errors = 0;
5392
5393		vdev_state_dirty(vd->vdev_top);
5394	}
5395
5396	for (int c = 0; c < vd->vdev_children; c++)
5397		spa_async_remove(spa, vd->vdev_child[c]);
5398}
5399
5400static void
5401spa_async_probe(spa_t *spa, vdev_t *vd)
5402{
5403	if (vd->vdev_probe_wanted) {
5404		vd->vdev_probe_wanted = B_FALSE;
5405		vdev_reopen(vd);	/* vdev_open() does the actual probe */
5406	}
5407
5408	for (int c = 0; c < vd->vdev_children; c++)
5409		spa_async_probe(spa, vd->vdev_child[c]);
5410}
5411
5412static void
5413spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5414{
5415	sysevent_id_t eid;
5416	nvlist_t *attr;
5417	char *physpath;
5418
5419	if (!spa->spa_autoexpand)
5420		return;
5421
5422	for (int c = 0; c < vd->vdev_children; c++) {
5423		vdev_t *cvd = vd->vdev_child[c];
5424		spa_async_autoexpand(spa, cvd);
5425	}
5426
5427	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5428		return;
5429
5430	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5431	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5432
5433	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5434	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5435
5436	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5437	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5438
5439	nvlist_free(attr);
5440	kmem_free(physpath, MAXPATHLEN);
5441}
5442
5443static void
5444spa_async_thread(spa_t *spa)
5445{
5446	int tasks;
5447
5448	ASSERT(spa->spa_sync_on);
5449
5450	mutex_enter(&spa->spa_async_lock);
5451	tasks = spa->spa_async_tasks;
5452	spa->spa_async_tasks = 0;
5453	mutex_exit(&spa->spa_async_lock);
5454
5455	/*
5456	 * See if the config needs to be updated.
5457	 */
5458	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5459		uint64_t old_space, new_space;
5460
5461		mutex_enter(&spa_namespace_lock);
5462		old_space = metaslab_class_get_space(spa_normal_class(spa));
5463		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5464		new_space = metaslab_class_get_space(spa_normal_class(spa));
5465		mutex_exit(&spa_namespace_lock);
5466
5467		/*
5468		 * If the pool grew as a result of the config update,
5469		 * then log an internal history event.
5470		 */
5471		if (new_space != old_space) {
5472			spa_history_log_internal(spa, "vdev online", NULL,
5473			    "pool '%s' size: %llu(+%llu)",
5474			    spa_name(spa), new_space, new_space - old_space);
5475		}
5476	}
5477
5478	/*
5479	 * See if any devices need to be marked REMOVED.
5480	 */
5481	if (tasks & SPA_ASYNC_REMOVE) {
5482		spa_vdev_state_enter(spa, SCL_NONE);
5483		spa_async_remove(spa, spa->spa_root_vdev);
5484		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5485			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5486		for (int i = 0; i < spa->spa_spares.sav_count; i++)
5487			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5488		(void) spa_vdev_state_exit(spa, NULL, 0);
5489	}
5490
5491	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5492		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5493		spa_async_autoexpand(spa, spa->spa_root_vdev);
5494		spa_config_exit(spa, SCL_CONFIG, FTAG);
5495	}
5496
5497	/*
5498	 * See if any devices need to be probed.
5499	 */
5500	if (tasks & SPA_ASYNC_PROBE) {
5501		spa_vdev_state_enter(spa, SCL_NONE);
5502		spa_async_probe(spa, spa->spa_root_vdev);
5503		(void) spa_vdev_state_exit(spa, NULL, 0);
5504	}
5505
5506	/*
5507	 * If any devices are done replacing, detach them.
5508	 */
5509	if (tasks & SPA_ASYNC_RESILVER_DONE)
5510		spa_vdev_resilver_done(spa);
5511
5512	/*
5513	 * Kick off a resilver.
5514	 */
5515	if (tasks & SPA_ASYNC_RESILVER)
5516		dsl_resilver_restart(spa->spa_dsl_pool, 0);
5517
5518	/*
5519	 * Let the world know that we're done.
5520	 */
5521	mutex_enter(&spa->spa_async_lock);
5522	spa->spa_async_thread = NULL;
5523	cv_broadcast(&spa->spa_async_cv);
5524	mutex_exit(&spa->spa_async_lock);
5525	thread_exit();
5526}
5527
5528void
5529spa_async_suspend(spa_t *spa)
5530{
5531	mutex_enter(&spa->spa_async_lock);
5532	spa->spa_async_suspended++;
5533	while (spa->spa_async_thread != NULL)
5534		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5535	mutex_exit(&spa->spa_async_lock);
5536}
5537
5538void
5539spa_async_resume(spa_t *spa)
5540{
5541	mutex_enter(&spa->spa_async_lock);
5542	ASSERT(spa->spa_async_suspended != 0);
5543	spa->spa_async_suspended--;
5544	mutex_exit(&spa->spa_async_lock);
5545}
5546
5547static void
5548spa_async_dispatch(spa_t *spa)
5549{
5550	mutex_enter(&spa->spa_async_lock);
5551	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5552	    spa->spa_async_thread == NULL &&
5553	    rootdir != NULL && !vn_is_readonly(rootdir))
5554		spa->spa_async_thread = thread_create(NULL, 0,
5555		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5556	mutex_exit(&spa->spa_async_lock);
5557}
5558
5559void
5560spa_async_request(spa_t *spa, int task)
5561{
5562	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5563	mutex_enter(&spa->spa_async_lock);
5564	spa->spa_async_tasks |= task;
5565	mutex_exit(&spa->spa_async_lock);
5566}
5567
5568/*
5569 * ==========================================================================
5570 * SPA syncing routines
5571 * ==========================================================================
5572 */
5573
5574static int
5575bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5576{
5577	bpobj_t *bpo = arg;
5578	bpobj_enqueue(bpo, bp, tx);
5579	return (0);
5580}
5581
5582static int
5583spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5584{
5585	zio_t *zio = arg;
5586
5587	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5588	    zio->io_flags));
5589	return (0);
5590}
5591
5592static void
5593spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5594{
5595	char *packed = NULL;
5596	size_t bufsize;
5597	size_t nvsize = 0;
5598	dmu_buf_t *db;
5599
5600	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5601
5602	/*
5603	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5604	 * information.  This avoids the dbuf_will_dirty() path and
5605	 * saves us a pre-read to get data we don't actually care about.
5606	 */
5607	bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5608	packed = kmem_alloc(bufsize, KM_SLEEP);
5609
5610	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5611	    KM_SLEEP) == 0);
5612	bzero(packed + nvsize, bufsize - nvsize);
5613
5614	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5615
5616	kmem_free(packed, bufsize);
5617
5618	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5619	dmu_buf_will_dirty(db, tx);
5620	*(uint64_t *)db->db_data = nvsize;
5621	dmu_buf_rele(db, FTAG);
5622}
5623
5624static void
5625spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5626    const char *config, const char *entry)
5627{
5628	nvlist_t *nvroot;
5629	nvlist_t **list;
5630	int i;
5631
5632	if (!sav->sav_sync)
5633		return;
5634
5635	/*
5636	 * Update the MOS nvlist describing the list of available devices.
5637	 * spa_validate_aux() will have already made sure this nvlist is
5638	 * valid and the vdevs are labeled appropriately.
5639	 */
5640	if (sav->sav_object == 0) {
5641		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5642		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5643		    sizeof (uint64_t), tx);
5644		VERIFY(zap_update(spa->spa_meta_objset,
5645		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5646		    &sav->sav_object, tx) == 0);
5647	}
5648
5649	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5650	if (sav->sav_count == 0) {
5651		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5652	} else {
5653		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5654		for (i = 0; i < sav->sav_count; i++)
5655			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5656			    B_FALSE, VDEV_CONFIG_L2CACHE);
5657		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5658		    sav->sav_count) == 0);
5659		for (i = 0; i < sav->sav_count; i++)
5660			nvlist_free(list[i]);
5661		kmem_free(list, sav->sav_count * sizeof (void *));
5662	}
5663
5664	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5665	nvlist_free(nvroot);
5666
5667	sav->sav_sync = B_FALSE;
5668}
5669
5670static void
5671spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5672{
5673	nvlist_t *config;
5674
5675	if (list_is_empty(&spa->spa_config_dirty_list))
5676		return;
5677
5678	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5679
5680	config = spa_config_generate(spa, spa->spa_root_vdev,
5681	    dmu_tx_get_txg(tx), B_FALSE);
5682
5683	spa_config_exit(spa, SCL_STATE, FTAG);
5684
5685	if (spa->spa_config_syncing)
5686		nvlist_free(spa->spa_config_syncing);
5687	spa->spa_config_syncing = config;
5688
5689	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5690}
5691
5692static void
5693spa_sync_version(void *arg1, void *arg2, dmu_tx_t *tx)
5694{
5695	spa_t *spa = arg1;
5696	uint64_t version = *(uint64_t *)arg2;
5697
5698	/*
5699	 * Setting the version is special cased when first creating the pool.
5700	 */
5701	ASSERT(tx->tx_txg != TXG_INITIAL);
5702
5703	ASSERT(version <= SPA_VERSION);
5704	ASSERT(version >= spa_version(spa));
5705
5706	spa->spa_uberblock.ub_version = version;
5707	vdev_config_dirty(spa->spa_root_vdev);
5708	spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5709}
5710
5711/*
5712 * Set zpool properties.
5713 */
5714static void
5715spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5716{
5717	spa_t *spa = arg1;
5718	objset_t *mos = spa->spa_meta_objset;
5719	nvlist_t *nvp = arg2;
5720	nvpair_t *elem = NULL;
5721
5722	mutex_enter(&spa->spa_props_lock);
5723
5724	while ((elem = nvlist_next_nvpair(nvp, elem))) {
5725		uint64_t intval;
5726		char *strval, *fname;
5727		zpool_prop_t prop;
5728		const char *propname;
5729		zprop_type_t proptype;
5730		zfeature_info_t *feature;
5731
5732		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5733		case ZPROP_INVAL:
5734			/*
5735			 * We checked this earlier in spa_prop_validate().
5736			 */
5737			ASSERT(zpool_prop_feature(nvpair_name(elem)));
5738
5739			fname = strchr(nvpair_name(elem), '@') + 1;
5740			VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
5741
5742			spa_feature_enable(spa, feature, tx);
5743			spa_history_log_internal(spa, "set", tx,
5744			    "%s=enabled", nvpair_name(elem));
5745			break;
5746
5747		case ZPOOL_PROP_VERSION:
5748			VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5749			/*
5750			 * The version is synced seperatly before other
5751			 * properties and should be correct by now.
5752			 */
5753			ASSERT3U(spa_version(spa), >=, intval);
5754			break;
5755
5756		case ZPOOL_PROP_ALTROOT:
5757			/*
5758			 * 'altroot' is a non-persistent property. It should
5759			 * have been set temporarily at creation or import time.
5760			 */
5761			ASSERT(spa->spa_root != NULL);
5762			break;
5763
5764		case ZPOOL_PROP_READONLY:
5765		case ZPOOL_PROP_CACHEFILE:
5766			/*
5767			 * 'readonly' and 'cachefile' are also non-persisitent
5768			 * properties.
5769			 */
5770			break;
5771		case ZPOOL_PROP_COMMENT:
5772			VERIFY(nvpair_value_string(elem, &strval) == 0);
5773			if (spa->spa_comment != NULL)
5774				spa_strfree(spa->spa_comment);
5775			spa->spa_comment = spa_strdup(strval);
5776			/*
5777			 * We need to dirty the configuration on all the vdevs
5778			 * so that their labels get updated.  It's unnecessary
5779			 * to do this for pool creation since the vdev's
5780			 * configuratoin has already been dirtied.
5781			 */
5782			if (tx->tx_txg != TXG_INITIAL)
5783				vdev_config_dirty(spa->spa_root_vdev);
5784			spa_history_log_internal(spa, "set", tx,
5785			    "%s=%s", nvpair_name(elem), strval);
5786			break;
5787		default:
5788			/*
5789			 * Set pool property values in the poolprops mos object.
5790			 */
5791			if (spa->spa_pool_props_object == 0) {
5792				spa->spa_pool_props_object =
5793				    zap_create_link(mos, DMU_OT_POOL_PROPS,
5794				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5795				    tx);
5796			}
5797
5798			/* normalize the property name */
5799			propname = zpool_prop_to_name(prop);
5800			proptype = zpool_prop_get_type(prop);
5801
5802			if (nvpair_type(elem) == DATA_TYPE_STRING) {
5803				ASSERT(proptype == PROP_TYPE_STRING);
5804				VERIFY(nvpair_value_string(elem, &strval) == 0);
5805				VERIFY(zap_update(mos,
5806				    spa->spa_pool_props_object, propname,
5807				    1, strlen(strval) + 1, strval, tx) == 0);
5808				spa_history_log_internal(spa, "set", tx,
5809				    "%s=%s", nvpair_name(elem), strval);
5810			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5811				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5812
5813				if (proptype == PROP_TYPE_INDEX) {
5814					const char *unused;
5815					VERIFY(zpool_prop_index_to_string(
5816					    prop, intval, &unused) == 0);
5817				}
5818				VERIFY(zap_update(mos,
5819				    spa->spa_pool_props_object, propname,
5820				    8, 1, &intval, tx) == 0);
5821				spa_history_log_internal(spa, "set", tx,
5822				    "%s=%lld", nvpair_name(elem), intval);
5823			} else {
5824				ASSERT(0); /* not allowed */
5825			}
5826
5827			switch (prop) {
5828			case ZPOOL_PROP_DELEGATION:
5829				spa->spa_delegation = intval;
5830				break;
5831			case ZPOOL_PROP_BOOTFS:
5832				spa->spa_bootfs = intval;
5833				break;
5834			case ZPOOL_PROP_FAILUREMODE:
5835				spa->spa_failmode = intval;
5836				break;
5837			case ZPOOL_PROP_AUTOEXPAND:
5838				spa->spa_autoexpand = intval;
5839				if (tx->tx_txg != TXG_INITIAL)
5840					spa_async_request(spa,
5841					    SPA_ASYNC_AUTOEXPAND);
5842				break;
5843			case ZPOOL_PROP_DEDUPDITTO:
5844				spa->spa_dedup_ditto = intval;
5845				break;
5846			default:
5847				break;
5848			}
5849		}
5850
5851	}
5852
5853	mutex_exit(&spa->spa_props_lock);
5854}
5855
5856/*
5857 * Perform one-time upgrade on-disk changes.  spa_version() does not
5858 * reflect the new version this txg, so there must be no changes this
5859 * txg to anything that the upgrade code depends on after it executes.
5860 * Therefore this must be called after dsl_pool_sync() does the sync
5861 * tasks.
5862 */
5863static void
5864spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5865{
5866	dsl_pool_t *dp = spa->spa_dsl_pool;
5867
5868	ASSERT(spa->spa_sync_pass == 1);
5869
5870	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5871	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5872		dsl_pool_create_origin(dp, tx);
5873
5874		/* Keeping the origin open increases spa_minref */
5875		spa->spa_minref += 3;
5876	}
5877
5878	if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5879	    spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5880		dsl_pool_upgrade_clones(dp, tx);
5881	}
5882
5883	if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5884	    spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5885		dsl_pool_upgrade_dir_clones(dp, tx);
5886
5887		/* Keeping the freedir open increases spa_minref */
5888		spa->spa_minref += 3;
5889	}
5890
5891	if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
5892	    spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
5893		spa_feature_create_zap_objects(spa, tx);
5894	}
5895}
5896
5897/*
5898 * Sync the specified transaction group.  New blocks may be dirtied as
5899 * part of the process, so we iterate until it converges.
5900 */
5901void
5902spa_sync(spa_t *spa, uint64_t txg)
5903{
5904	dsl_pool_t *dp = spa->spa_dsl_pool;
5905	objset_t *mos = spa->spa_meta_objset;
5906	bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5907	bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5908	vdev_t *rvd = spa->spa_root_vdev;
5909	vdev_t *vd;
5910	dmu_tx_t *tx;
5911	int error;
5912
5913	VERIFY(spa_writeable(spa));
5914
5915	/*
5916	 * Lock out configuration changes.
5917	 */
5918	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5919
5920	spa->spa_syncing_txg = txg;
5921	spa->spa_sync_pass = 0;
5922
5923	/*
5924	 * If there are any pending vdev state changes, convert them
5925	 * into config changes that go out with this transaction group.
5926	 */
5927	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5928	while (list_head(&spa->spa_state_dirty_list) != NULL) {
5929		/*
5930		 * We need the write lock here because, for aux vdevs,
5931		 * calling vdev_config_dirty() modifies sav_config.
5932		 * This is ugly and will become unnecessary when we
5933		 * eliminate the aux vdev wart by integrating all vdevs
5934		 * into the root vdev tree.
5935		 */
5936		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5937		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5938		while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5939			vdev_state_clean(vd);
5940			vdev_config_dirty(vd);
5941		}
5942		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5943		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5944	}
5945	spa_config_exit(spa, SCL_STATE, FTAG);
5946
5947	tx = dmu_tx_create_assigned(dp, txg);
5948
5949	/*
5950	 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5951	 * set spa_deflate if we have no raid-z vdevs.
5952	 */
5953	if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5954	    spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5955		int i;
5956
5957		for (i = 0; i < rvd->vdev_children; i++) {
5958			vd = rvd->vdev_child[i];
5959			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5960				break;
5961		}
5962		if (i == rvd->vdev_children) {
5963			spa->spa_deflate = TRUE;
5964			VERIFY(0 == zap_add(spa->spa_meta_objset,
5965			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5966			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5967		}
5968	}
5969
5970	/*
5971	 * If anything has changed in this txg, or if someone is waiting
5972	 * for this txg to sync (eg, spa_vdev_remove()), push the
5973	 * deferred frees from the previous txg.  If not, leave them
5974	 * alone so that we don't generate work on an otherwise idle
5975	 * system.
5976	 */
5977	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5978	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5979	    !txg_list_empty(&dp->dp_sync_tasks, txg) ||
5980	    ((dsl_scan_active(dp->dp_scan) ||
5981	    txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
5982		zio_t *zio = zio_root(spa, NULL, NULL, 0);
5983		VERIFY3U(bpobj_iterate(defer_bpo,
5984		    spa_free_sync_cb, zio, tx), ==, 0);
5985		VERIFY3U(zio_wait(zio), ==, 0);
5986	}
5987
5988	/*
5989	 * Iterate to convergence.
5990	 */
5991	do {
5992		int pass = ++spa->spa_sync_pass;
5993
5994		spa_sync_config_object(spa, tx);
5995		spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5996		    ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5997		spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5998		    ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5999		spa_errlog_sync(spa, txg);
6000		dsl_pool_sync(dp, txg);
6001
6002		if (pass <= SYNC_PASS_DEFERRED_FREE) {
6003			zio_t *zio = zio_root(spa, NULL, NULL, 0);
6004			bplist_iterate(free_bpl, spa_free_sync_cb,
6005			    zio, tx);
6006			VERIFY(zio_wait(zio) == 0);
6007		} else {
6008			bplist_iterate(free_bpl, bpobj_enqueue_cb,
6009			    defer_bpo, tx);
6010		}
6011
6012		ddt_sync(spa, txg);
6013		dsl_scan_sync(dp, tx);
6014
6015		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6016			vdev_sync(vd, txg);
6017
6018		if (pass == 1)
6019			spa_sync_upgrades(spa, tx);
6020
6021	} while (dmu_objset_is_dirty(mos, txg));
6022
6023	/*
6024	 * Rewrite the vdev configuration (which includes the uberblock)
6025	 * to commit the transaction group.
6026	 *
6027	 * If there are no dirty vdevs, we sync the uberblock to a few
6028	 * random top-level vdevs that are known to be visible in the
6029	 * config cache (see spa_vdev_add() for a complete description).
6030	 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6031	 */
6032	for (;;) {
6033		/*
6034		 * We hold SCL_STATE to prevent vdev open/close/etc.
6035		 * while we're attempting to write the vdev labels.
6036		 */
6037		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6038
6039		if (list_is_empty(&spa->spa_config_dirty_list)) {
6040			vdev_t *svd[SPA_DVAS_PER_BP];
6041			int svdcount = 0;
6042			int children = rvd->vdev_children;
6043			int c0 = spa_get_random(children);
6044
6045			for (int c = 0; c < children; c++) {
6046				vd = rvd->vdev_child[(c0 + c) % children];
6047				if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6048					continue;
6049				svd[svdcount++] = vd;
6050				if (svdcount == SPA_DVAS_PER_BP)
6051					break;
6052			}
6053			error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6054			if (error != 0)
6055				error = vdev_config_sync(svd, svdcount, txg,
6056				    B_TRUE);
6057		} else {
6058			error = vdev_config_sync(rvd->vdev_child,
6059			    rvd->vdev_children, txg, B_FALSE);
6060			if (error != 0)
6061				error = vdev_config_sync(rvd->