spa.c revision 25345e466695fbe736faa53b8f3413d8e8f81981
1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
25 * Copyright (c) 2012 by Delphix. All rights reserved.
26 */
27
28/*
29 * This file contains all the routines used when modifying on-disk SPA state.
30 * This includes opening, importing, destroying, exporting a pool, and syncing a
31 * pool.
32 */
33
34#include <sys/zfs_context.h>
35#include <sys/fm/fs/zfs.h>
36#include <sys/spa_impl.h>
37#include <sys/zio.h>
38#include <sys/zio_checksum.h>
39#include <sys/dmu.h>
40#include <sys/dmu_tx.h>
41#include <sys/zap.h>
42#include <sys/zil.h>
43#include <sys/ddt.h>
44#include <sys/vdev_impl.h>
45#include <sys/metaslab.h>
46#include <sys/metaslab_impl.h>
47#include <sys/uberblock_impl.h>
48#include <sys/txg.h>
49#include <sys/avl.h>
50#include <sys/dmu_traverse.h>
51#include <sys/dmu_objset.h>
52#include <sys/unique.h>
53#include <sys/dsl_pool.h>
54#include <sys/dsl_dataset.h>
55#include <sys/dsl_dir.h>
56#include <sys/dsl_prop.h>
57#include <sys/dsl_synctask.h>
58#include <sys/fs/zfs.h>
59#include <sys/arc.h>
60#include <sys/callb.h>
61#include <sys/systeminfo.h>
62#include <sys/spa_boot.h>
63#include <sys/zfs_ioctl.h>
64#include <sys/dsl_scan.h>
65#include <sys/zfeature.h>
66
67#ifdef	_KERNEL
68#include <sys/bootprops.h>
69#include <sys/callb.h>
70#include <sys/cpupart.h>
71#include <sys/pool.h>
72#include <sys/sysdc.h>
73#include <sys/zone.h>
74#endif	/* _KERNEL */
75
76#include "zfs_prop.h"
77#include "zfs_comutil.h"
78
79typedef enum zti_modes {
80	zti_mode_fixed,			/* value is # of threads (min 1) */
81	zti_mode_online_percent,	/* value is % of online CPUs */
82	zti_mode_batch,			/* cpu-intensive; value is ignored */
83	zti_mode_null,			/* don't create a taskq */
84	zti_nmodes
85} zti_modes_t;
86
87#define	ZTI_FIX(n)	{ zti_mode_fixed, (n) }
88#define	ZTI_PCT(n)	{ zti_mode_online_percent, (n) }
89#define	ZTI_BATCH	{ zti_mode_batch, 0 }
90#define	ZTI_NULL	{ zti_mode_null, 0 }
91
92#define	ZTI_ONE		ZTI_FIX(1)
93
94typedef struct zio_taskq_info {
95	enum zti_modes zti_mode;
96	uint_t zti_value;
97} zio_taskq_info_t;
98
99static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
100	"issue", "issue_high", "intr", "intr_high"
101};
102
103/*
104 * Define the taskq threads for the following I/O types:
105 * 	NULL, READ, WRITE, FREE, CLAIM, and IOCTL
106 */
107const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
108	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
109	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
110	{ ZTI_FIX(8),	ZTI_NULL,	ZTI_BATCH,	ZTI_NULL },
111	{ ZTI_BATCH,	ZTI_FIX(5),	ZTI_FIX(8),	ZTI_FIX(5) },
112	{ ZTI_FIX(100),	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
113	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
114	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
115};
116
117static dsl_syncfunc_t spa_sync_version;
118static dsl_syncfunc_t spa_sync_props;
119static dsl_checkfunc_t spa_change_guid_check;
120static dsl_syncfunc_t spa_change_guid_sync;
121static boolean_t spa_has_active_shared_spare(spa_t *spa);
122static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
123    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
124    char **ereport);
125static void spa_vdev_resilver_done(spa_t *spa);
126
127uint_t		zio_taskq_batch_pct = 100;	/* 1 thread per cpu in pset */
128id_t		zio_taskq_psrset_bind = PS_NONE;
129boolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
130uint_t		zio_taskq_basedc = 80;		/* base duty cycle */
131
132boolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
133extern int	zfs_sync_pass_deferred_free;
134
135/*
136 * This (illegal) pool name is used when temporarily importing a spa_t in order
137 * to get the vdev stats associated with the imported devices.
138 */
139#define	TRYIMPORT_NAME	"$import"
140
141/*
142 * ==========================================================================
143 * SPA properties routines
144 * ==========================================================================
145 */
146
147/*
148 * Add a (source=src, propname=propval) list to an nvlist.
149 */
150static void
151spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
152    uint64_t intval, zprop_source_t src)
153{
154	const char *propname = zpool_prop_to_name(prop);
155	nvlist_t *propval;
156
157	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
158	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
159
160	if (strval != NULL)
161		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
162	else
163		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
164
165	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
166	nvlist_free(propval);
167}
168
169/*
170 * Get property values from the spa configuration.
171 */
172static void
173spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
174{
175	vdev_t *rvd = spa->spa_root_vdev;
176	dsl_pool_t *pool = spa->spa_dsl_pool;
177	uint64_t size;
178	uint64_t alloc;
179	uint64_t space;
180	uint64_t cap, version;
181	zprop_source_t src = ZPROP_SRC_NONE;
182	spa_config_dirent_t *dp;
183
184	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
185
186	if (rvd != NULL) {
187		alloc = metaslab_class_get_alloc(spa_normal_class(spa));
188		size = metaslab_class_get_space(spa_normal_class(spa));
189		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
190		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
191		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
192		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
193		    size - alloc, src);
194
195		space = 0;
196		for (int c = 0; c < rvd->vdev_children; c++) {
197			vdev_t *tvd = rvd->vdev_child[c];
198			space += tvd->vdev_max_asize - tvd->vdev_asize;
199		}
200		spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
201		    src);
202
203		spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
204		    (spa_mode(spa) == FREAD), src);
205
206		cap = (size == 0) ? 0 : (alloc * 100 / size);
207		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
208
209		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
210		    ddt_get_pool_dedup_ratio(spa), src);
211
212		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
213		    rvd->vdev_state, src);
214
215		version = spa_version(spa);
216		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
217			src = ZPROP_SRC_DEFAULT;
218		else
219			src = ZPROP_SRC_LOCAL;
220		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
221	}
222
223	if (pool != NULL) {
224		dsl_dir_t *freedir = pool->dp_free_dir;
225
226		/*
227		 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
228		 * when opening pools before this version freedir will be NULL.
229		 */
230		if (freedir != NULL) {
231			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
232			    freedir->dd_phys->dd_used_bytes, src);
233		} else {
234			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
235			    NULL, 0, src);
236		}
237	}
238
239	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
240
241	if (spa->spa_comment != NULL) {
242		spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
243		    0, ZPROP_SRC_LOCAL);
244	}
245
246	if (spa->spa_root != NULL)
247		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
248		    0, ZPROP_SRC_LOCAL);
249
250	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
251		if (dp->scd_path == NULL) {
252			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
253			    "none", 0, ZPROP_SRC_LOCAL);
254		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
255			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
256			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
257		}
258	}
259}
260
261/*
262 * Get zpool property values.
263 */
264int
265spa_prop_get(spa_t *spa, nvlist_t **nvp)
266{
267	objset_t *mos = spa->spa_meta_objset;
268	zap_cursor_t zc;
269	zap_attribute_t za;
270	int err;
271
272	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
273
274	mutex_enter(&spa->spa_props_lock);
275
276	/*
277	 * Get properties from the spa config.
278	 */
279	spa_prop_get_config(spa, nvp);
280
281	/* If no pool property object, no more prop to get. */
282	if (mos == NULL || spa->spa_pool_props_object == 0) {
283		mutex_exit(&spa->spa_props_lock);
284		return (0);
285	}
286
287	/*
288	 * Get properties from the MOS pool property object.
289	 */
290	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
291	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
292	    zap_cursor_advance(&zc)) {
293		uint64_t intval = 0;
294		char *strval = NULL;
295		zprop_source_t src = ZPROP_SRC_DEFAULT;
296		zpool_prop_t prop;
297
298		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
299			continue;
300
301		switch (za.za_integer_length) {
302		case 8:
303			/* integer property */
304			if (za.za_first_integer !=
305			    zpool_prop_default_numeric(prop))
306				src = ZPROP_SRC_LOCAL;
307
308			if (prop == ZPOOL_PROP_BOOTFS) {
309				dsl_pool_t *dp;
310				dsl_dataset_t *ds = NULL;
311
312				dp = spa_get_dsl(spa);
313				rw_enter(&dp->dp_config_rwlock, RW_READER);
314				if (err = dsl_dataset_hold_obj(dp,
315				    za.za_first_integer, FTAG, &ds)) {
316					rw_exit(&dp->dp_config_rwlock);
317					break;
318				}
319
320				strval = kmem_alloc(
321				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
322				    KM_SLEEP);
323				dsl_dataset_name(ds, strval);
324				dsl_dataset_rele(ds, FTAG);
325				rw_exit(&dp->dp_config_rwlock);
326			} else {
327				strval = NULL;
328				intval = za.za_first_integer;
329			}
330
331			spa_prop_add_list(*nvp, prop, strval, intval, src);
332
333			if (strval != NULL)
334				kmem_free(strval,
335				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
336
337			break;
338
339		case 1:
340			/* string property */
341			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
342			err = zap_lookup(mos, spa->spa_pool_props_object,
343			    za.za_name, 1, za.za_num_integers, strval);
344			if (err) {
345				kmem_free(strval, za.za_num_integers);
346				break;
347			}
348			spa_prop_add_list(*nvp, prop, strval, 0, src);
349			kmem_free(strval, za.za_num_integers);
350			break;
351
352		default:
353			break;
354		}
355	}
356	zap_cursor_fini(&zc);
357	mutex_exit(&spa->spa_props_lock);
358out:
359	if (err && err != ENOENT) {
360		nvlist_free(*nvp);
361		*nvp = NULL;
362		return (err);
363	}
364
365	return (0);
366}
367
368/*
369 * Validate the given pool properties nvlist and modify the list
370 * for the property values to be set.
371 */
372static int
373spa_prop_validate(spa_t *spa, nvlist_t *props)
374{
375	nvpair_t *elem;
376	int error = 0, reset_bootfs = 0;
377	uint64_t objnum;
378	boolean_t has_feature = B_FALSE;
379
380	elem = NULL;
381	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
382		uint64_t intval;
383		char *strval, *slash, *check, *fname;
384		const char *propname = nvpair_name(elem);
385		zpool_prop_t prop = zpool_name_to_prop(propname);
386
387		switch (prop) {
388		case ZPROP_INVAL:
389			if (!zpool_prop_feature(propname)) {
390				error = EINVAL;
391				break;
392			}
393
394			/*
395			 * Sanitize the input.
396			 */
397			if (nvpair_type(elem) != DATA_TYPE_UINT64) {
398				error = EINVAL;
399				break;
400			}
401
402			if (nvpair_value_uint64(elem, &intval) != 0) {
403				error = EINVAL;
404				break;
405			}
406
407			if (intval != 0) {
408				error = EINVAL;
409				break;
410			}
411
412			fname = strchr(propname, '@') + 1;
413			if (zfeature_lookup_name(fname, NULL) != 0) {
414				error = EINVAL;
415				break;
416			}
417
418			has_feature = B_TRUE;
419			break;
420
421		case ZPOOL_PROP_VERSION:
422			error = nvpair_value_uint64(elem, &intval);
423			if (!error &&
424			    (intval < spa_version(spa) ||
425			    intval > SPA_VERSION_BEFORE_FEATURES ||
426			    has_feature))
427				error = EINVAL;
428			break;
429
430		case ZPOOL_PROP_DELEGATION:
431		case ZPOOL_PROP_AUTOREPLACE:
432		case ZPOOL_PROP_LISTSNAPS:
433		case ZPOOL_PROP_AUTOEXPAND:
434			error = nvpair_value_uint64(elem, &intval);
435			if (!error && intval > 1)
436				error = EINVAL;
437			break;
438
439		case ZPOOL_PROP_BOOTFS:
440			/*
441			 * If the pool version is less than SPA_VERSION_BOOTFS,
442			 * or the pool is still being created (version == 0),
443			 * the bootfs property cannot be set.
444			 */
445			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
446				error = ENOTSUP;
447				break;
448			}
449
450			/*
451			 * Make sure the vdev config is bootable
452			 */
453			if (!vdev_is_bootable(spa->spa_root_vdev)) {
454				error = ENOTSUP;
455				break;
456			}
457
458			reset_bootfs = 1;
459
460			error = nvpair_value_string(elem, &strval);
461
462			if (!error) {
463				objset_t *os;
464				uint64_t compress;
465
466				if (strval == NULL || strval[0] == '\0') {
467					objnum = zpool_prop_default_numeric(
468					    ZPOOL_PROP_BOOTFS);
469					break;
470				}
471
472				if (error = dmu_objset_hold(strval, FTAG, &os))
473					break;
474
475				/* Must be ZPL and not gzip compressed. */
476
477				if (dmu_objset_type(os) != DMU_OST_ZFS) {
478					error = ENOTSUP;
479				} else if ((error = dsl_prop_get_integer(strval,
480				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
481				    &compress, NULL)) == 0 &&
482				    !BOOTFS_COMPRESS_VALID(compress)) {
483					error = ENOTSUP;
484				} else {
485					objnum = dmu_objset_id(os);
486				}
487				dmu_objset_rele(os, FTAG);
488			}
489			break;
490
491		case ZPOOL_PROP_FAILUREMODE:
492			error = nvpair_value_uint64(elem, &intval);
493			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
494			    intval > ZIO_FAILURE_MODE_PANIC))
495				error = EINVAL;
496
497			/*
498			 * This is a special case which only occurs when
499			 * the pool has completely failed. This allows
500			 * the user to change the in-core failmode property
501			 * without syncing it out to disk (I/Os might
502			 * currently be blocked). We do this by returning
503			 * EIO to the caller (spa_prop_set) to trick it
504			 * into thinking we encountered a property validation
505			 * error.
506			 */
507			if (!error && spa_suspended(spa)) {
508				spa->spa_failmode = intval;
509				error = EIO;
510			}
511			break;
512
513		case ZPOOL_PROP_CACHEFILE:
514			if ((error = nvpair_value_string(elem, &strval)) != 0)
515				break;
516
517			if (strval[0] == '\0')
518				break;
519
520			if (strcmp(strval, "none") == 0)
521				break;
522
523			if (strval[0] != '/') {
524				error = EINVAL;
525				break;
526			}
527
528			slash = strrchr(strval, '/');
529			ASSERT(slash != NULL);
530
531			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
532			    strcmp(slash, "/..") == 0)
533				error = EINVAL;
534			break;
535
536		case ZPOOL_PROP_COMMENT:
537			if ((error = nvpair_value_string(elem, &strval)) != 0)
538				break;
539			for (check = strval; *check != '\0'; check++) {
540				/*
541				 * The kernel doesn't have an easy isprint()
542				 * check.  For this kernel check, we merely
543				 * check ASCII apart from DEL.  Fix this if
544				 * there is an easy-to-use kernel isprint().
545				 */
546				if (*check >= 0x7f) {
547					error = EINVAL;
548					break;
549				}
550				check++;
551			}
552			if (strlen(strval) > ZPROP_MAX_COMMENT)
553				error = E2BIG;
554			break;
555
556		case ZPOOL_PROP_DEDUPDITTO:
557			if (spa_version(spa) < SPA_VERSION_DEDUP)
558				error = ENOTSUP;
559			else
560				error = nvpair_value_uint64(elem, &intval);
561			if (error == 0 &&
562			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
563				error = EINVAL;
564			break;
565		}
566
567		if (error)
568			break;
569	}
570
571	if (!error && reset_bootfs) {
572		error = nvlist_remove(props,
573		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
574
575		if (!error) {
576			error = nvlist_add_uint64(props,
577			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
578		}
579	}
580
581	return (error);
582}
583
584void
585spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
586{
587	char *cachefile;
588	spa_config_dirent_t *dp;
589
590	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
591	    &cachefile) != 0)
592		return;
593
594	dp = kmem_alloc(sizeof (spa_config_dirent_t),
595	    KM_SLEEP);
596
597	if (cachefile[0] == '\0')
598		dp->scd_path = spa_strdup(spa_config_path);
599	else if (strcmp(cachefile, "none") == 0)
600		dp->scd_path = NULL;
601	else
602		dp->scd_path = spa_strdup(cachefile);
603
604	list_insert_head(&spa->spa_config_list, dp);
605	if (need_sync)
606		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
607}
608
609int
610spa_prop_set(spa_t *spa, nvlist_t *nvp)
611{
612	int error;
613	nvpair_t *elem = NULL;
614	boolean_t need_sync = B_FALSE;
615
616	if ((error = spa_prop_validate(spa, nvp)) != 0)
617		return (error);
618
619	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
620		zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
621
622		if (prop == ZPOOL_PROP_CACHEFILE ||
623		    prop == ZPOOL_PROP_ALTROOT ||
624		    prop == ZPOOL_PROP_READONLY)
625			continue;
626
627		if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
628			uint64_t ver;
629
630			if (prop == ZPOOL_PROP_VERSION) {
631				VERIFY(nvpair_value_uint64(elem, &ver) == 0);
632			} else {
633				ASSERT(zpool_prop_feature(nvpair_name(elem)));
634				ver = SPA_VERSION_FEATURES;
635				need_sync = B_TRUE;
636			}
637
638			/* Save time if the version is already set. */
639			if (ver == spa_version(spa))
640				continue;
641
642			/*
643			 * In addition to the pool directory object, we might
644			 * create the pool properties object, the features for
645			 * read object, the features for write object, or the
646			 * feature descriptions object.
647			 */
648			error = dsl_sync_task_do(spa_get_dsl(spa), NULL,
649			    spa_sync_version, spa, &ver, 6);
650			if (error)
651				return (error);
652			continue;
653		}
654
655		need_sync = B_TRUE;
656		break;
657	}
658
659	if (need_sync) {
660		return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
661		    spa, nvp, 6));
662	}
663
664	return (0);
665}
666
667/*
668 * If the bootfs property value is dsobj, clear it.
669 */
670void
671spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
672{
673	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
674		VERIFY(zap_remove(spa->spa_meta_objset,
675		    spa->spa_pool_props_object,
676		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
677		spa->spa_bootfs = 0;
678	}
679}
680
681/*ARGSUSED*/
682static int
683spa_change_guid_check(void *arg1, void *arg2, dmu_tx_t *tx)
684{
685	spa_t *spa = arg1;
686	uint64_t *newguid = arg2;
687	vdev_t *rvd = spa->spa_root_vdev;
688	uint64_t vdev_state;
689
690	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
691	vdev_state = rvd->vdev_state;
692	spa_config_exit(spa, SCL_STATE, FTAG);
693
694	if (vdev_state != VDEV_STATE_HEALTHY)
695		return (ENXIO);
696
697	ASSERT3U(spa_guid(spa), !=, *newguid);
698
699	return (0);
700}
701
702static void
703spa_change_guid_sync(void *arg1, void *arg2, dmu_tx_t *tx)
704{
705	spa_t *spa = arg1;
706	uint64_t *newguid = arg2;
707	uint64_t oldguid;
708	vdev_t *rvd = spa->spa_root_vdev;
709
710	oldguid = spa_guid(spa);
711
712	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
713	rvd->vdev_guid = *newguid;
714	rvd->vdev_guid_sum += (*newguid - oldguid);
715	vdev_config_dirty(rvd);
716	spa_config_exit(spa, SCL_STATE, FTAG);
717
718	spa_history_log_internal(spa, "guid change", tx, "old=%lld new=%lld",
719	    oldguid, *newguid);
720}
721
722/*
723 * Change the GUID for the pool.  This is done so that we can later
724 * re-import a pool built from a clone of our own vdevs.  We will modify
725 * the root vdev's guid, our own pool guid, and then mark all of our
726 * vdevs dirty.  Note that we must make sure that all our vdevs are
727 * online when we do this, or else any vdevs that weren't present
728 * would be orphaned from our pool.  We are also going to issue a
729 * sysevent to update any watchers.
730 */
731int
732spa_change_guid(spa_t *spa)
733{
734	int error;
735	uint64_t guid;
736
737	mutex_enter(&spa_namespace_lock);
738	guid = spa_generate_guid(NULL);
739
740	error = dsl_sync_task_do(spa_get_dsl(spa), spa_change_guid_check,
741	    spa_change_guid_sync, spa, &guid, 5);
742
743	if (error == 0) {
744		spa_config_sync(spa, B_FALSE, B_TRUE);
745		spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
746	}
747
748	mutex_exit(&spa_namespace_lock);
749
750	return (error);
751}
752
753/*
754 * ==========================================================================
755 * SPA state manipulation (open/create/destroy/import/export)
756 * ==========================================================================
757 */
758
759static int
760spa_error_entry_compare(const void *a, const void *b)
761{
762	spa_error_entry_t *sa = (spa_error_entry_t *)a;
763	spa_error_entry_t *sb = (spa_error_entry_t *)b;
764	int ret;
765
766	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
767	    sizeof (zbookmark_t));
768
769	if (ret < 0)
770		return (-1);
771	else if (ret > 0)
772		return (1);
773	else
774		return (0);
775}
776
777/*
778 * Utility function which retrieves copies of the current logs and
779 * re-initializes them in the process.
780 */
781void
782spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
783{
784	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
785
786	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
787	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
788
789	avl_create(&spa->spa_errlist_scrub,
790	    spa_error_entry_compare, sizeof (spa_error_entry_t),
791	    offsetof(spa_error_entry_t, se_avl));
792	avl_create(&spa->spa_errlist_last,
793	    spa_error_entry_compare, sizeof (spa_error_entry_t),
794	    offsetof(spa_error_entry_t, se_avl));
795}
796
797static taskq_t *
798spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
799    uint_t value)
800{
801	uint_t flags = 0;
802	boolean_t batch = B_FALSE;
803
804	switch (mode) {
805	case zti_mode_null:
806		return (NULL);		/* no taskq needed */
807
808	case zti_mode_fixed:
809		ASSERT3U(value, >=, 1);
810		value = MAX(value, 1);
811		break;
812
813	case zti_mode_batch:
814		batch = B_TRUE;
815		flags |= TASKQ_THREADS_CPU_PCT;
816		value = zio_taskq_batch_pct;
817		break;
818
819	case zti_mode_online_percent:
820		flags |= TASKQ_THREADS_CPU_PCT;
821		break;
822
823	default:
824		panic("unrecognized mode for %s taskq (%u:%u) in "
825		    "spa_activate()",
826		    name, mode, value);
827		break;
828	}
829
830	if (zio_taskq_sysdc && spa->spa_proc != &p0) {
831		if (batch)
832			flags |= TASKQ_DC_BATCH;
833
834		return (taskq_create_sysdc(name, value, 50, INT_MAX,
835		    spa->spa_proc, zio_taskq_basedc, flags));
836	}
837	return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
838	    spa->spa_proc, flags));
839}
840
841static void
842spa_create_zio_taskqs(spa_t *spa)
843{
844	for (int t = 0; t < ZIO_TYPES; t++) {
845		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
846			const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
847			enum zti_modes mode = ztip->zti_mode;
848			uint_t value = ztip->zti_value;
849			char name[32];
850
851			(void) snprintf(name, sizeof (name),
852			    "%s_%s", zio_type_name[t], zio_taskq_types[q]);
853
854			spa->spa_zio_taskq[t][q] =
855			    spa_taskq_create(spa, name, mode, value);
856		}
857	}
858}
859
860#ifdef _KERNEL
861static void
862spa_thread(void *arg)
863{
864	callb_cpr_t cprinfo;
865
866	spa_t *spa = arg;
867	user_t *pu = PTOU(curproc);
868
869	CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
870	    spa->spa_name);
871
872	ASSERT(curproc != &p0);
873	(void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
874	    "zpool-%s", spa->spa_name);
875	(void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
876
877	/* bind this thread to the requested psrset */
878	if (zio_taskq_psrset_bind != PS_NONE) {
879		pool_lock();
880		mutex_enter(&cpu_lock);
881		mutex_enter(&pidlock);
882		mutex_enter(&curproc->p_lock);
883
884		if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
885		    0, NULL, NULL) == 0)  {
886			curthread->t_bind_pset = zio_taskq_psrset_bind;
887		} else {
888			cmn_err(CE_WARN,
889			    "Couldn't bind process for zfs pool \"%s\" to "
890			    "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
891		}
892
893		mutex_exit(&curproc->p_lock);
894		mutex_exit(&pidlock);
895		mutex_exit(&cpu_lock);
896		pool_unlock();
897	}
898
899	if (zio_taskq_sysdc) {
900		sysdc_thread_enter(curthread, 100, 0);
901	}
902
903	spa->spa_proc = curproc;
904	spa->spa_did = curthread->t_did;
905
906	spa_create_zio_taskqs(spa);
907
908	mutex_enter(&spa->spa_proc_lock);
909	ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
910
911	spa->spa_proc_state = SPA_PROC_ACTIVE;
912	cv_broadcast(&spa->spa_proc_cv);
913
914	CALLB_CPR_SAFE_BEGIN(&cprinfo);
915	while (spa->spa_proc_state == SPA_PROC_ACTIVE)
916		cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
917	CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
918
919	ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
920	spa->spa_proc_state = SPA_PROC_GONE;
921	spa->spa_proc = &p0;
922	cv_broadcast(&spa->spa_proc_cv);
923	CALLB_CPR_EXIT(&cprinfo);	/* drops spa_proc_lock */
924
925	mutex_enter(&curproc->p_lock);
926	lwp_exit();
927}
928#endif
929
930/*
931 * Activate an uninitialized pool.
932 */
933static void
934spa_activate(spa_t *spa, int mode)
935{
936	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
937
938	spa->spa_state = POOL_STATE_ACTIVE;
939	spa->spa_mode = mode;
940
941	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
942	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
943
944	/* Try to create a covering process */
945	mutex_enter(&spa->spa_proc_lock);
946	ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
947	ASSERT(spa->spa_proc == &p0);
948	spa->spa_did = 0;
949
950	/* Only create a process if we're going to be around a while. */
951	if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
952		if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
953		    NULL, 0) == 0) {
954			spa->spa_proc_state = SPA_PROC_CREATED;
955			while (spa->spa_proc_state == SPA_PROC_CREATED) {
956				cv_wait(&spa->spa_proc_cv,
957				    &spa->spa_proc_lock);
958			}
959			ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
960			ASSERT(spa->spa_proc != &p0);
961			ASSERT(spa->spa_did != 0);
962		} else {
963#ifdef _KERNEL
964			cmn_err(CE_WARN,
965			    "Couldn't create process for zfs pool \"%s\"\n",
966			    spa->spa_name);
967#endif
968		}
969	}
970	mutex_exit(&spa->spa_proc_lock);
971
972	/* If we didn't create a process, we need to create our taskqs. */
973	if (spa->spa_proc == &p0) {
974		spa_create_zio_taskqs(spa);
975	}
976
977	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
978	    offsetof(vdev_t, vdev_config_dirty_node));
979	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
980	    offsetof(vdev_t, vdev_state_dirty_node));
981
982	txg_list_create(&spa->spa_vdev_txg_list,
983	    offsetof(struct vdev, vdev_txg_node));
984
985	avl_create(&spa->spa_errlist_scrub,
986	    spa_error_entry_compare, sizeof (spa_error_entry_t),
987	    offsetof(spa_error_entry_t, se_avl));
988	avl_create(&spa->spa_errlist_last,
989	    spa_error_entry_compare, sizeof (spa_error_entry_t),
990	    offsetof(spa_error_entry_t, se_avl));
991}
992
993/*
994 * Opposite of spa_activate().
995 */
996static void
997spa_deactivate(spa_t *spa)
998{
999	ASSERT(spa->spa_sync_on == B_FALSE);
1000	ASSERT(spa->spa_dsl_pool == NULL);
1001	ASSERT(spa->spa_root_vdev == NULL);
1002	ASSERT(spa->spa_async_zio_root == NULL);
1003	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1004
1005	txg_list_destroy(&spa->spa_vdev_txg_list);
1006
1007	list_destroy(&spa->spa_config_dirty_list);
1008	list_destroy(&spa->spa_state_dirty_list);
1009
1010	for (int t = 0; t < ZIO_TYPES; t++) {
1011		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1012			if (spa->spa_zio_taskq[t][q] != NULL)
1013				taskq_destroy(spa->spa_zio_taskq[t][q]);
1014			spa->spa_zio_taskq[t][q] = NULL;
1015		}
1016	}
1017
1018	metaslab_class_destroy(spa->spa_normal_class);
1019	spa->spa_normal_class = NULL;
1020
1021	metaslab_class_destroy(spa->spa_log_class);
1022	spa->spa_log_class = NULL;
1023
1024	/*
1025	 * If this was part of an import or the open otherwise failed, we may
1026	 * still have errors left in the queues.  Empty them just in case.
1027	 */
1028	spa_errlog_drain(spa);
1029
1030	avl_destroy(&spa->spa_errlist_scrub);
1031	avl_destroy(&spa->spa_errlist_last);
1032
1033	spa->spa_state = POOL_STATE_UNINITIALIZED;
1034
1035	mutex_enter(&spa->spa_proc_lock);
1036	if (spa->spa_proc_state != SPA_PROC_NONE) {
1037		ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1038		spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1039		cv_broadcast(&spa->spa_proc_cv);
1040		while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1041			ASSERT(spa->spa_proc != &p0);
1042			cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1043		}
1044		ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1045		spa->spa_proc_state = SPA_PROC_NONE;
1046	}
1047	ASSERT(spa->spa_proc == &p0);
1048	mutex_exit(&spa->spa_proc_lock);
1049
1050	/*
1051	 * We want to make sure spa_thread() has actually exited the ZFS
1052	 * module, so that the module can't be unloaded out from underneath
1053	 * it.
1054	 */
1055	if (spa->spa_did != 0) {
1056		thread_join(spa->spa_did);
1057		spa->spa_did = 0;
1058	}
1059}
1060
1061/*
1062 * Verify a pool configuration, and construct the vdev tree appropriately.  This
1063 * will create all the necessary vdevs in the appropriate layout, with each vdev
1064 * in the CLOSED state.  This will prep the pool before open/creation/import.
1065 * All vdev validation is done by the vdev_alloc() routine.
1066 */
1067static int
1068spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1069    uint_t id, int atype)
1070{
1071	nvlist_t **child;
1072	uint_t children;
1073	int error;
1074
1075	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1076		return (error);
1077
1078	if ((*vdp)->vdev_ops->vdev_op_leaf)
1079		return (0);
1080
1081	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1082	    &child, &children);
1083
1084	if (error == ENOENT)
1085		return (0);
1086
1087	if (error) {
1088		vdev_free(*vdp);
1089		*vdp = NULL;
1090		return (EINVAL);
1091	}
1092
1093	for (int c = 0; c < children; c++) {
1094		vdev_t *vd;
1095		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1096		    atype)) != 0) {
1097			vdev_free(*vdp);
1098			*vdp = NULL;
1099			return (error);
1100		}
1101	}
1102
1103	ASSERT(*vdp != NULL);
1104
1105	return (0);
1106}
1107
1108/*
1109 * Opposite of spa_load().
1110 */
1111static void
1112spa_unload(spa_t *spa)
1113{
1114	int i;
1115
1116	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1117
1118	/*
1119	 * Stop async tasks.
1120	 */
1121	spa_async_suspend(spa);
1122
1123	/*
1124	 * Stop syncing.
1125	 */
1126	if (spa->spa_sync_on) {
1127		txg_sync_stop(spa->spa_dsl_pool);
1128		spa->spa_sync_on = B_FALSE;
1129	}
1130
1131	/*
1132	 * Wait for any outstanding async I/O to complete.
1133	 */
1134	if (spa->spa_async_zio_root != NULL) {
1135		(void) zio_wait(spa->spa_async_zio_root);
1136		spa->spa_async_zio_root = NULL;
1137	}
1138
1139	bpobj_close(&spa->spa_deferred_bpobj);
1140
1141	/*
1142	 * Close the dsl pool.
1143	 */
1144	if (spa->spa_dsl_pool) {
1145		dsl_pool_close(spa->spa_dsl_pool);
1146		spa->spa_dsl_pool = NULL;
1147		spa->spa_meta_objset = NULL;
1148	}
1149
1150	ddt_unload(spa);
1151
1152	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1153
1154	/*
1155	 * Drop and purge level 2 cache
1156	 */
1157	spa_l2cache_drop(spa);
1158
1159	/*
1160	 * Close all vdevs.
1161	 */
1162	if (spa->spa_root_vdev)
1163		vdev_free(spa->spa_root_vdev);
1164	ASSERT(spa->spa_root_vdev == NULL);
1165
1166	for (i = 0; i < spa->spa_spares.sav_count; i++)
1167		vdev_free(spa->spa_spares.sav_vdevs[i]);
1168	if (spa->spa_spares.sav_vdevs) {
1169		kmem_free(spa->spa_spares.sav_vdevs,
1170		    spa->spa_spares.sav_count * sizeof (void *));
1171		spa->spa_spares.sav_vdevs = NULL;
1172	}
1173	if (spa->spa_spares.sav_config) {
1174		nvlist_free(spa->spa_spares.sav_config);
1175		spa->spa_spares.sav_config = NULL;
1176	}
1177	spa->spa_spares.sav_count = 0;
1178
1179	for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1180		vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1181		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1182	}
1183	if (spa->spa_l2cache.sav_vdevs) {
1184		kmem_free(spa->spa_l2cache.sav_vdevs,
1185		    spa->spa_l2cache.sav_count * sizeof (void *));
1186		spa->spa_l2cache.sav_vdevs = NULL;
1187	}
1188	if (spa->spa_l2cache.sav_config) {
1189		nvlist_free(spa->spa_l2cache.sav_config);
1190		spa->spa_l2cache.sav_config = NULL;
1191	}
1192	spa->spa_l2cache.sav_count = 0;
1193
1194	spa->spa_async_suspended = 0;
1195
1196	if (spa->spa_comment != NULL) {
1197		spa_strfree(spa->spa_comment);
1198		spa->spa_comment = NULL;
1199	}
1200
1201	spa_config_exit(spa, SCL_ALL, FTAG);
1202}
1203
1204/*
1205 * Load (or re-load) the current list of vdevs describing the active spares for
1206 * this pool.  When this is called, we have some form of basic information in
1207 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1208 * then re-generate a more complete list including status information.
1209 */
1210static void
1211spa_load_spares(spa_t *spa)
1212{
1213	nvlist_t **spares;
1214	uint_t nspares;
1215	int i;
1216	vdev_t *vd, *tvd;
1217
1218	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1219
1220	/*
1221	 * First, close and free any existing spare vdevs.
1222	 */
1223	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1224		vd = spa->spa_spares.sav_vdevs[i];
1225
1226		/* Undo the call to spa_activate() below */
1227		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1228		    B_FALSE)) != NULL && tvd->vdev_isspare)
1229			spa_spare_remove(tvd);
1230		vdev_close(vd);
1231		vdev_free(vd);
1232	}
1233
1234	if (spa->spa_spares.sav_vdevs)
1235		kmem_free(spa->spa_spares.sav_vdevs,
1236		    spa->spa_spares.sav_count * sizeof (void *));
1237
1238	if (spa->spa_spares.sav_config == NULL)
1239		nspares = 0;
1240	else
1241		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1242		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1243
1244	spa->spa_spares.sav_count = (int)nspares;
1245	spa->spa_spares.sav_vdevs = NULL;
1246
1247	if (nspares == 0)
1248		return;
1249
1250	/*
1251	 * Construct the array of vdevs, opening them to get status in the
1252	 * process.   For each spare, there is potentially two different vdev_t
1253	 * structures associated with it: one in the list of spares (used only
1254	 * for basic validation purposes) and one in the active vdev
1255	 * configuration (if it's spared in).  During this phase we open and
1256	 * validate each vdev on the spare list.  If the vdev also exists in the
1257	 * active configuration, then we also mark this vdev as an active spare.
1258	 */
1259	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1260	    KM_SLEEP);
1261	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1262		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1263		    VDEV_ALLOC_SPARE) == 0);
1264		ASSERT(vd != NULL);
1265
1266		spa->spa_spares.sav_vdevs[i] = vd;
1267
1268		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1269		    B_FALSE)) != NULL) {
1270			if (!tvd->vdev_isspare)
1271				spa_spare_add(tvd);
1272
1273			/*
1274			 * We only mark the spare active if we were successfully
1275			 * able to load the vdev.  Otherwise, importing a pool
1276			 * with a bad active spare would result in strange
1277			 * behavior, because multiple pool would think the spare
1278			 * is actively in use.
1279			 *
1280			 * There is a vulnerability here to an equally bizarre
1281			 * circumstance, where a dead active spare is later
1282			 * brought back to life (onlined or otherwise).  Given
1283			 * the rarity of this scenario, and the extra complexity
1284			 * it adds, we ignore the possibility.
1285			 */
1286			if (!vdev_is_dead(tvd))
1287				spa_spare_activate(tvd);
1288		}
1289
1290		vd->vdev_top = vd;
1291		vd->vdev_aux = &spa->spa_spares;
1292
1293		if (vdev_open(vd) != 0)
1294			continue;
1295
1296		if (vdev_validate_aux(vd) == 0)
1297			spa_spare_add(vd);
1298	}
1299
1300	/*
1301	 * Recompute the stashed list of spares, with status information
1302	 * this time.
1303	 */
1304	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1305	    DATA_TYPE_NVLIST_ARRAY) == 0);
1306
1307	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1308	    KM_SLEEP);
1309	for (i = 0; i < spa->spa_spares.sav_count; i++)
1310		spares[i] = vdev_config_generate(spa,
1311		    spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1312	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1313	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1314	for (i = 0; i < spa->spa_spares.sav_count; i++)
1315		nvlist_free(spares[i]);
1316	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1317}
1318
1319/*
1320 * Load (or re-load) the current list of vdevs describing the active l2cache for
1321 * this pool.  When this is called, we have some form of basic information in
1322 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1323 * then re-generate a more complete list including status information.
1324 * Devices which are already active have their details maintained, and are
1325 * not re-opened.
1326 */
1327static void
1328spa_load_l2cache(spa_t *spa)
1329{
1330	nvlist_t **l2cache;
1331	uint_t nl2cache;
1332	int i, j, oldnvdevs;
1333	uint64_t guid;
1334	vdev_t *vd, **oldvdevs, **newvdevs;
1335	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1336
1337	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1338
1339	if (sav->sav_config != NULL) {
1340		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1341		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1342		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1343	} else {
1344		nl2cache = 0;
1345	}
1346
1347	oldvdevs = sav->sav_vdevs;
1348	oldnvdevs = sav->sav_count;
1349	sav->sav_vdevs = NULL;
1350	sav->sav_count = 0;
1351
1352	/*
1353	 * Process new nvlist of vdevs.
1354	 */
1355	for (i = 0; i < nl2cache; i++) {
1356		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1357		    &guid) == 0);
1358
1359		newvdevs[i] = NULL;
1360		for (j = 0; j < oldnvdevs; j++) {
1361			vd = oldvdevs[j];
1362			if (vd != NULL && guid == vd->vdev_guid) {
1363				/*
1364				 * Retain previous vdev for add/remove ops.
1365				 */
1366				newvdevs[i] = vd;
1367				oldvdevs[j] = NULL;
1368				break;
1369			}
1370		}
1371
1372		if (newvdevs[i] == NULL) {
1373			/*
1374			 * Create new vdev
1375			 */
1376			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1377			    VDEV_ALLOC_L2CACHE) == 0);
1378			ASSERT(vd != NULL);
1379			newvdevs[i] = vd;
1380
1381			/*
1382			 * Commit this vdev as an l2cache device,
1383			 * even if it fails to open.
1384			 */
1385			spa_l2cache_add(vd);
1386
1387			vd->vdev_top = vd;
1388			vd->vdev_aux = sav;
1389
1390			spa_l2cache_activate(vd);
1391
1392			if (vdev_open(vd) != 0)
1393				continue;
1394
1395			(void) vdev_validate_aux(vd);
1396
1397			if (!vdev_is_dead(vd))
1398				l2arc_add_vdev(spa, vd);
1399		}
1400	}
1401
1402	/*
1403	 * Purge vdevs that were dropped
1404	 */
1405	for (i = 0; i < oldnvdevs; i++) {
1406		uint64_t pool;
1407
1408		vd = oldvdevs[i];
1409		if (vd != NULL) {
1410			ASSERT(vd->vdev_isl2cache);
1411
1412			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1413			    pool != 0ULL && l2arc_vdev_present(vd))
1414				l2arc_remove_vdev(vd);
1415			vdev_clear_stats(vd);
1416			vdev_free(vd);
1417		}
1418	}
1419
1420	if (oldvdevs)
1421		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1422
1423	if (sav->sav_config == NULL)
1424		goto out;
1425
1426	sav->sav_vdevs = newvdevs;
1427	sav->sav_count = (int)nl2cache;
1428
1429	/*
1430	 * Recompute the stashed list of l2cache devices, with status
1431	 * information this time.
1432	 */
1433	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1434	    DATA_TYPE_NVLIST_ARRAY) == 0);
1435
1436	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1437	for (i = 0; i < sav->sav_count; i++)
1438		l2cache[i] = vdev_config_generate(spa,
1439		    sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1440	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1441	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1442out:
1443	for (i = 0; i < sav->sav_count; i++)
1444		nvlist_free(l2cache[i]);
1445	if (sav->sav_count)
1446		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1447}
1448
1449static int
1450load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1451{
1452	dmu_buf_t *db;
1453	char *packed = NULL;
1454	size_t nvsize = 0;
1455	int error;
1456	*value = NULL;
1457
1458	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1459	nvsize = *(uint64_t *)db->db_data;
1460	dmu_buf_rele(db, FTAG);
1461
1462	packed = kmem_alloc(nvsize, KM_SLEEP);
1463	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1464	    DMU_READ_PREFETCH);
1465	if (error == 0)
1466		error = nvlist_unpack(packed, nvsize, value, 0);
1467	kmem_free(packed, nvsize);
1468
1469	return (error);
1470}
1471
1472/*
1473 * Checks to see if the given vdev could not be opened, in which case we post a
1474 * sysevent to notify the autoreplace code that the device has been removed.
1475 */
1476static void
1477spa_check_removed(vdev_t *vd)
1478{
1479	for (int c = 0; c < vd->vdev_children; c++)
1480		spa_check_removed(vd->vdev_child[c]);
1481
1482	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1483		zfs_post_autoreplace(vd->vdev_spa, vd);
1484		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1485	}
1486}
1487
1488/*
1489 * Validate the current config against the MOS config
1490 */
1491static boolean_t
1492spa_config_valid(spa_t *spa, nvlist_t *config)
1493{
1494	vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1495	nvlist_t *nv;
1496
1497	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1498
1499	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1500	VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1501
1502	ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1503
1504	/*
1505	 * If we're doing a normal import, then build up any additional
1506	 * diagnostic information about missing devices in this config.
1507	 * We'll pass this up to the user for further processing.
1508	 */
1509	if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1510		nvlist_t **child, *nv;
1511		uint64_t idx = 0;
1512
1513		child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1514		    KM_SLEEP);
1515		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1516
1517		for (int c = 0; c < rvd->vdev_children; c++) {
1518			vdev_t *tvd = rvd->vdev_child[c];
1519			vdev_t *mtvd  = mrvd->vdev_child[c];
1520
1521			if (tvd->vdev_ops == &vdev_missing_ops &&
1522			    mtvd->vdev_ops != &vdev_missing_ops &&
1523			    mtvd->vdev_islog)
1524				child[idx++] = vdev_config_generate(spa, mtvd,
1525				    B_FALSE, 0);
1526		}
1527
1528		if (idx) {
1529			VERIFY(nvlist_add_nvlist_array(nv,
1530			    ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1531			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1532			    ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1533
1534			for (int i = 0; i < idx; i++)
1535				nvlist_free(child[i]);
1536		}
1537		nvlist_free(nv);
1538		kmem_free(child, rvd->vdev_children * sizeof (char **));
1539	}
1540
1541	/*
1542	 * Compare the root vdev tree with the information we have
1543	 * from the MOS config (mrvd). Check each top-level vdev
1544	 * with the corresponding MOS config top-level (mtvd).
1545	 */
1546	for (int c = 0; c < rvd->vdev_children; c++) {
1547		vdev_t *tvd = rvd->vdev_child[c];
1548		vdev_t *mtvd  = mrvd->vdev_child[c];
1549
1550		/*
1551		 * Resolve any "missing" vdevs in the current configuration.
1552		 * If we find that the MOS config has more accurate information
1553		 * about the top-level vdev then use that vdev instead.
1554		 */
1555		if (tvd->vdev_ops == &vdev_missing_ops &&
1556		    mtvd->vdev_ops != &vdev_missing_ops) {
1557
1558			if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1559				continue;
1560
1561			/*
1562			 * Device specific actions.
1563			 */
1564			if (mtvd->vdev_islog) {
1565				spa_set_log_state(spa, SPA_LOG_CLEAR);
1566			} else {
1567				/*
1568				 * XXX - once we have 'readonly' pool
1569				 * support we should be able to handle
1570				 * missing data devices by transitioning
1571				 * the pool to readonly.
1572				 */
1573				continue;
1574			}
1575
1576			/*
1577			 * Swap the missing vdev with the data we were
1578			 * able to obtain from the MOS config.
1579			 */
1580			vdev_remove_child(rvd, tvd);
1581			vdev_remove_child(mrvd, mtvd);
1582
1583			vdev_add_child(rvd, mtvd);
1584			vdev_add_child(mrvd, tvd);
1585
1586			spa_config_exit(spa, SCL_ALL, FTAG);
1587			vdev_load(mtvd);
1588			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1589
1590			vdev_reopen(rvd);
1591		} else if (mtvd->vdev_islog) {
1592			/*
1593			 * Load the slog device's state from the MOS config
1594			 * since it's possible that the label does not
1595			 * contain the most up-to-date information.
1596			 */
1597			vdev_load_log_state(tvd, mtvd);
1598			vdev_reopen(tvd);
1599		}
1600	}
1601	vdev_free(mrvd);
1602	spa_config_exit(spa, SCL_ALL, FTAG);
1603
1604	/*
1605	 * Ensure we were able to validate the config.
1606	 */
1607	return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1608}
1609
1610/*
1611 * Check for missing log devices
1612 */
1613static int
1614spa_check_logs(spa_t *spa)
1615{
1616	switch (spa->spa_log_state) {
1617	case SPA_LOG_MISSING:
1618		/* need to recheck in case slog has been restored */
1619	case SPA_LOG_UNKNOWN:
1620		if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1621		    DS_FIND_CHILDREN)) {
1622			spa_set_log_state(spa, SPA_LOG_MISSING);
1623			return (1);
1624		}
1625		break;
1626	}
1627	return (0);
1628}
1629
1630static boolean_t
1631spa_passivate_log(spa_t *spa)
1632{
1633	vdev_t *rvd = spa->spa_root_vdev;
1634	boolean_t slog_found = B_FALSE;
1635
1636	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1637
1638	if (!spa_has_slogs(spa))
1639		return (B_FALSE);
1640
1641	for (int c = 0; c < rvd->vdev_children; c++) {
1642		vdev_t *tvd = rvd->vdev_child[c];
1643		metaslab_group_t *mg = tvd->vdev_mg;
1644
1645		if (tvd->vdev_islog) {
1646			metaslab_group_passivate(mg);
1647			slog_found = B_TRUE;
1648		}
1649	}
1650
1651	return (slog_found);
1652}
1653
1654static void
1655spa_activate_log(spa_t *spa)
1656{
1657	vdev_t *rvd = spa->spa_root_vdev;
1658
1659	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1660
1661	for (int c = 0; c < rvd->vdev_children; c++) {
1662		vdev_t *tvd = rvd->vdev_child[c];
1663		metaslab_group_t *mg = tvd->vdev_mg;
1664
1665		if (tvd->vdev_islog)
1666			metaslab_group_activate(mg);
1667	}
1668}
1669
1670int
1671spa_offline_log(spa_t *spa)
1672{
1673	int error = 0;
1674
1675	if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1676	    NULL, DS_FIND_CHILDREN)) == 0) {
1677
1678		/*
1679		 * We successfully offlined the log device, sync out the
1680		 * current txg so that the "stubby" block can be removed
1681		 * by zil_sync().
1682		 */
1683		txg_wait_synced(spa->spa_dsl_pool, 0);
1684	}
1685	return (error);
1686}
1687
1688static void
1689spa_aux_check_removed(spa_aux_vdev_t *sav)
1690{
1691	for (int i = 0; i < sav->sav_count; i++)
1692		spa_check_removed(sav->sav_vdevs[i]);
1693}
1694
1695void
1696spa_claim_notify(zio_t *zio)
1697{
1698	spa_t *spa = zio->io_spa;
1699
1700	if (zio->io_error)
1701		return;
1702
1703	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1704	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1705		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1706	mutex_exit(&spa->spa_props_lock);
1707}
1708
1709typedef struct spa_load_error {
1710	uint64_t	sle_meta_count;
1711	uint64_t	sle_data_count;
1712} spa_load_error_t;
1713
1714static void
1715spa_load_verify_done(zio_t *zio)
1716{
1717	blkptr_t *bp = zio->io_bp;
1718	spa_load_error_t *sle = zio->io_private;
1719	dmu_object_type_t type = BP_GET_TYPE(bp);
1720	int error = zio->io_error;
1721
1722	if (error) {
1723		if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1724		    type != DMU_OT_INTENT_LOG)
1725			atomic_add_64(&sle->sle_meta_count, 1);
1726		else
1727			atomic_add_64(&sle->sle_data_count, 1);
1728	}
1729	zio_data_buf_free(zio->io_data, zio->io_size);
1730}
1731
1732/*ARGSUSED*/
1733static int
1734spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1735    arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1736{
1737	if (bp != NULL) {
1738		zio_t *rio = arg;
1739		size_t size = BP_GET_PSIZE(bp);
1740		void *data = zio_data_buf_alloc(size);
1741
1742		zio_nowait(zio_read(rio, spa, bp, data, size,
1743		    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1744		    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1745		    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1746	}
1747	return (0);
1748}
1749
1750static int
1751spa_load_verify(spa_t *spa)
1752{
1753	zio_t *rio;
1754	spa_load_error_t sle = { 0 };
1755	zpool_rewind_policy_t policy;
1756	boolean_t verify_ok = B_FALSE;
1757	int error;
1758
1759	zpool_get_rewind_policy(spa->spa_config, &policy);
1760
1761	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1762		return (0);
1763
1764	rio = zio_root(spa, NULL, &sle,
1765	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1766
1767	error = traverse_pool(spa, spa->spa_verify_min_txg,
1768	    TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1769
1770	(void) zio_wait(rio);
1771
1772	spa->spa_load_meta_errors = sle.sle_meta_count;
1773	spa->spa_load_data_errors = sle.sle_data_count;
1774
1775	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1776	    sle.sle_data_count <= policy.zrp_maxdata) {
1777		int64_t loss = 0;
1778
1779		verify_ok = B_TRUE;
1780		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1781		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1782
1783		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1784		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1785		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1786		VERIFY(nvlist_add_int64(spa->spa_load_info,
1787		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1788		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1789		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1790	} else {
1791		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1792	}
1793
1794	if (error) {
1795		if (error != ENXIO && error != EIO)
1796			error = EIO;
1797		return (error);
1798	}
1799
1800	return (verify_ok ? 0 : EIO);
1801}
1802
1803/*
1804 * Find a value in the pool props object.
1805 */
1806static void
1807spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1808{
1809	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1810	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1811}
1812
1813/*
1814 * Find a value in the pool directory object.
1815 */
1816static int
1817spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1818{
1819	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1820	    name, sizeof (uint64_t), 1, val));
1821}
1822
1823static int
1824spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1825{
1826	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1827	return (err);
1828}
1829
1830/*
1831 * Fix up config after a partly-completed split.  This is done with the
1832 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1833 * pool have that entry in their config, but only the splitting one contains
1834 * a list of all the guids of the vdevs that are being split off.
1835 *
1836 * This function determines what to do with that list: either rejoin
1837 * all the disks to the pool, or complete the splitting process.  To attempt
1838 * the rejoin, each disk that is offlined is marked online again, and
1839 * we do a reopen() call.  If the vdev label for every disk that was
1840 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1841 * then we call vdev_split() on each disk, and complete the split.
1842 *
1843 * Otherwise we leave the config alone, with all the vdevs in place in
1844 * the original pool.
1845 */
1846static void
1847spa_try_repair(spa_t *spa, nvlist_t *config)
1848{
1849	uint_t extracted;
1850	uint64_t *glist;
1851	uint_t i, gcount;
1852	nvlist_t *nvl;
1853	vdev_t **vd;
1854	boolean_t attempt_reopen;
1855
1856	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1857		return;
1858
1859	/* check that the config is complete */
1860	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1861	    &glist, &gcount) != 0)
1862		return;
1863
1864	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1865
1866	/* attempt to online all the vdevs & validate */
1867	attempt_reopen = B_TRUE;
1868	for (i = 0; i < gcount; i++) {
1869		if (glist[i] == 0)	/* vdev is hole */
1870			continue;
1871
1872		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1873		if (vd[i] == NULL) {
1874			/*
1875			 * Don't bother attempting to reopen the disks;
1876			 * just do the split.
1877			 */
1878			attempt_reopen = B_FALSE;
1879		} else {
1880			/* attempt to re-online it */
1881			vd[i]->vdev_offline = B_FALSE;
1882		}
1883	}
1884
1885	if (attempt_reopen) {
1886		vdev_reopen(spa->spa_root_vdev);
1887
1888		/* check each device to see what state it's in */
1889		for (extracted = 0, i = 0; i < gcount; i++) {
1890			if (vd[i] != NULL &&
1891			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1892				break;
1893			++extracted;
1894		}
1895	}
1896
1897	/*
1898	 * If every disk has been moved to the new pool, or if we never
1899	 * even attempted to look at them, then we split them off for
1900	 * good.
1901	 */
1902	if (!attempt_reopen || gcount == extracted) {
1903		for (i = 0; i < gcount; i++)
1904			if (vd[i] != NULL)
1905				vdev_split(vd[i]);
1906		vdev_reopen(spa->spa_root_vdev);
1907	}
1908
1909	kmem_free(vd, gcount * sizeof (vdev_t *));
1910}
1911
1912static int
1913spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1914    boolean_t mosconfig)
1915{
1916	nvlist_t *config = spa->spa_config;
1917	char *ereport = FM_EREPORT_ZFS_POOL;
1918	char *comment;
1919	int error;
1920	uint64_t pool_guid;
1921	nvlist_t *nvl;
1922
1923	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1924		return (EINVAL);
1925
1926	ASSERT(spa->spa_comment == NULL);
1927	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
1928		spa->spa_comment = spa_strdup(comment);
1929
1930	/*
1931	 * Versioning wasn't explicitly added to the label until later, so if
1932	 * it's not present treat it as the initial version.
1933	 */
1934	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1935	    &spa->spa_ubsync.ub_version) != 0)
1936		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1937
1938	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1939	    &spa->spa_config_txg);
1940
1941	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1942	    spa_guid_exists(pool_guid, 0)) {
1943		error = EEXIST;
1944	} else {
1945		spa->spa_config_guid = pool_guid;
1946
1947		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1948		    &nvl) == 0) {
1949			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1950			    KM_SLEEP) == 0);
1951		}
1952
1953		nvlist_free(spa->spa_load_info);
1954		spa->spa_load_info = fnvlist_alloc();
1955
1956		gethrestime(&spa->spa_loaded_ts);
1957		error = spa_load_impl(spa, pool_guid, config, state, type,
1958		    mosconfig, &ereport);
1959	}
1960
1961	spa->spa_minref = refcount_count(&spa->spa_refcount);
1962	if (error) {
1963		if (error != EEXIST) {
1964			spa->spa_loaded_ts.tv_sec = 0;
1965			spa->spa_loaded_ts.tv_nsec = 0;
1966		}
1967		if (error != EBADF) {
1968			zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1969		}
1970	}
1971	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1972	spa->spa_ena = 0;
1973
1974	return (error);
1975}
1976
1977/*
1978 * Load an existing storage pool, using the pool's builtin spa_config as a
1979 * source of configuration information.
1980 */
1981static int
1982spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1983    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1984    char **ereport)
1985{
1986	int error = 0;
1987	nvlist_t *nvroot = NULL;
1988	nvlist_t *label;
1989	vdev_t *rvd;
1990	uberblock_t *ub = &spa->spa_uberblock;
1991	uint64_t children, config_cache_txg = spa->spa_config_txg;
1992	int orig_mode = spa->spa_mode;
1993	int parse;
1994	uint64_t obj;
1995	boolean_t missing_feat_write = B_FALSE;
1996
1997	/*
1998	 * If this is an untrusted config, access the pool in read-only mode.
1999	 * This prevents things like resilvering recently removed devices.
2000	 */
2001	if (!mosconfig)
2002		spa->spa_mode = FREAD;
2003
2004	ASSERT(MUTEX_HELD(&spa_namespace_lock));
2005
2006	spa->spa_load_state = state;
2007
2008	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2009		return (EINVAL);
2010
2011	parse = (type == SPA_IMPORT_EXISTING ?
2012	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2013
2014	/*
2015	 * Create "The Godfather" zio to hold all async IOs
2016	 */
2017	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2018	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2019
2020	/*
2021	 * Parse the configuration into a vdev tree.  We explicitly set the
2022	 * value that will be returned by spa_version() since parsing the
2023	 * configuration requires knowing the version number.
2024	 */
2025	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2026	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2027	spa_config_exit(spa, SCL_ALL, FTAG);
2028
2029	if (error != 0)
2030		return (error);
2031
2032	ASSERT(spa->spa_root_vdev == rvd);
2033
2034	if (type != SPA_IMPORT_ASSEMBLE) {
2035		ASSERT(spa_guid(spa) == pool_guid);
2036	}
2037
2038	/*
2039	 * Try to open all vdevs, loading each label in the process.
2040	 */
2041	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2042	error = vdev_open(rvd);
2043	spa_config_exit(spa, SCL_ALL, FTAG);
2044	if (error != 0)
2045		return (error);
2046
2047	/*
2048	 * We need to validate the vdev labels against the configuration that
2049	 * we have in hand, which is dependent on the setting of mosconfig. If
2050	 * mosconfig is true then we're validating the vdev labels based on
2051	 * that config.  Otherwise, we're validating against the cached config
2052	 * (zpool.cache) that was read when we loaded the zfs module, and then
2053	 * later we will recursively call spa_load() and validate against
2054	 * the vdev config.
2055	 *
2056	 * If we're assembling a new pool that's been split off from an
2057	 * existing pool, the labels haven't yet been updated so we skip
2058	 * validation for now.
2059	 */
2060	if (type != SPA_IMPORT_ASSEMBLE) {
2061		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2062		error = vdev_validate(rvd, mosconfig);
2063		spa_config_exit(spa, SCL_ALL, FTAG);
2064
2065		if (error != 0)
2066			return (error);
2067
2068		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2069			return (ENXIO);
2070	}
2071
2072	/*
2073	 * Find the best uberblock.
2074	 */
2075	vdev_uberblock_load(rvd, ub, &label);
2076
2077	/*
2078	 * If we weren't able to find a single valid uberblock, return failure.
2079	 */
2080	if (ub->ub_txg == 0) {
2081		nvlist_free(label);
2082		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2083	}
2084
2085	/*
2086	 * If the pool has an unsupported version we can't open it.
2087	 */
2088	if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2089		nvlist_free(label);
2090		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2091	}
2092
2093	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2094		nvlist_t *features;
2095
2096		/*
2097		 * If we weren't able to find what's necessary for reading the
2098		 * MOS in the label, return failure.
2099		 */
2100		if (label == NULL || nvlist_lookup_nvlist(label,
2101		    ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2102			nvlist_free(label);
2103			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2104			    ENXIO));
2105		}
2106
2107		/*
2108		 * Update our in-core representation with the definitive values
2109		 * from the label.
2110		 */
2111		nvlist_free(spa->spa_label_features);
2112		VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2113	}
2114
2115	nvlist_free(label);
2116
2117	/*
2118	 * Look through entries in the label nvlist's features_for_read. If
2119	 * there is a feature listed there which we don't understand then we
2120	 * cannot open a pool.
2121	 */
2122	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2123		nvlist_t *unsup_feat;
2124
2125		VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2126		    0);
2127
2128		for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2129		    NULL); nvp != NULL;
2130		    nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2131			if (!zfeature_is_supported(nvpair_name(nvp))) {
2132				VERIFY(nvlist_add_string(unsup_feat,
2133				    nvpair_name(nvp), "") == 0);
2134			}
2135		}
2136
2137		if (!nvlist_empty(unsup_feat)) {
2138			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2139			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2140			nvlist_free(unsup_feat);
2141			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2142			    ENOTSUP));
2143		}
2144
2145		nvlist_free(unsup_feat);
2146	}
2147
2148	/*
2149	 * If the vdev guid sum doesn't match the uberblock, we have an
2150	 * incomplete configuration.  We first check to see if the pool
2151	 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2152	 * If it is, defer the vdev_guid_sum check till later so we
2153	 * can handle missing vdevs.
2154	 */
2155	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2156	    &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2157	    rvd->vdev_guid_sum != ub->ub_guid_sum)
2158		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2159
2160	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2161		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2162		spa_try_repair(spa, config);
2163		spa_config_exit(spa, SCL_ALL, FTAG);
2164		nvlist_free(spa->spa_config_splitting);
2165		spa->spa_config_splitting = NULL;
2166	}
2167
2168	/*
2169	 * Initialize internal SPA structures.
2170	 */
2171	spa->spa_state = POOL_STATE_ACTIVE;
2172	spa->spa_ubsync = spa->spa_uberblock;
2173	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2174	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2175	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2176	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2177	spa->spa_claim_max_txg = spa->spa_first_txg;
2178	spa->spa_prev_software_version = ub->ub_software_version;
2179
2180	error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2181	if (error)
2182		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2183	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2184
2185	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2186		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2187
2188	if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2189		boolean_t missing_feat_read = B_FALSE;
2190		nvlist_t *unsup_feat, *enabled_feat;
2191
2192		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2193		    &spa->spa_feat_for_read_obj) != 0) {
2194			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2195		}
2196
2197		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2198		    &spa->spa_feat_for_write_obj) != 0) {
2199			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2200		}
2201
2202		if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2203		    &spa->spa_feat_desc_obj) != 0) {
2204			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2205		}
2206
2207		enabled_feat = fnvlist_alloc();
2208		unsup_feat = fnvlist_alloc();
2209
2210		if (!feature_is_supported(spa->spa_meta_objset,
2211		    spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2212		    unsup_feat, enabled_feat))
2213			missing_feat_read = B_TRUE;
2214
2215		if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2216			if (!feature_is_supported(spa->spa_meta_objset,
2217			    spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2218			    unsup_feat, enabled_feat)) {
2219				missing_feat_write = B_TRUE;
2220			}
2221		}
2222
2223		fnvlist_add_nvlist(spa->spa_load_info,
2224		    ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2225
2226		if (!nvlist_empty(unsup_feat)) {
2227			fnvlist_add_nvlist(spa->spa_load_info,
2228			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2229		}
2230
2231		fnvlist_free(enabled_feat);
2232		fnvlist_free(unsup_feat);
2233
2234		if (!missing_feat_read) {
2235			fnvlist_add_boolean(spa->spa_load_info,
2236			    ZPOOL_CONFIG_CAN_RDONLY);
2237		}
2238
2239		/*
2240		 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2241		 * twofold: to determine whether the pool is available for
2242		 * import in read-write mode and (if it is not) whether the
2243		 * pool is available for import in read-only mode. If the pool
2244		 * is available for import in read-write mode, it is displayed
2245		 * as available in userland; if it is not available for import
2246		 * in read-only mode, it is displayed as unavailable in
2247		 * userland. If the pool is available for import in read-only
2248		 * mode but not read-write mode, it is displayed as unavailable
2249		 * in userland with a special note that the pool is actually
2250		 * available for open in read-only mode.
2251		 *
2252		 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2253		 * missing a feature for write, we must first determine whether
2254		 * the pool can be opened read-only before returning to
2255		 * userland in order to know whether to display the
2256		 * abovementioned note.
2257		 */
2258		if (missing_feat_read || (missing_feat_write &&
2259		    spa_writeable(spa))) {
2260			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2261			    ENOTSUP));
2262		}
2263	}
2264
2265	spa->spa_is_initializing = B_TRUE;
2266	error = dsl_pool_open(spa->spa_dsl_pool);
2267	spa->spa_is_initializing = B_FALSE;
2268	if (error != 0)
2269		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2270
2271	if (!mosconfig) {
2272		uint64_t hostid;
2273		nvlist_t *policy = NULL, *nvconfig;
2274
2275		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2276			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2277
2278		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2279		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2280			char *hostname;
2281			unsigned long myhostid = 0;
2282
2283			VERIFY(nvlist_lookup_string(nvconfig,
2284			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2285
2286#ifdef	_KERNEL
2287			myhostid = zone_get_hostid(NULL);
2288#else	/* _KERNEL */
2289			/*
2290			 * We're emulating the system's hostid in userland, so
2291			 * we can't use zone_get_hostid().
2292			 */
2293			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2294#endif	/* _KERNEL */
2295			if (hostid != 0 && myhostid != 0 &&
2296			    hostid != myhostid) {
2297				nvlist_free(nvconfig);
2298				cmn_err(CE_WARN, "pool '%s' could not be "
2299				    "loaded as it was last accessed by "
2300				    "another system (host: %s hostid: 0x%lx). "
2301				    "See: http://illumos.org/msg/ZFS-8000-EY",
2302				    spa_name(spa), hostname,
2303				    (unsigned long)hostid);
2304				return (EBADF);
2305			}
2306		}
2307		if (nvlist_lookup_nvlist(spa->spa_config,
2308		    ZPOOL_REWIND_POLICY, &policy) == 0)
2309			VERIFY(nvlist_add_nvlist(nvconfig,
2310			    ZPOOL_REWIND_POLICY, policy) == 0);
2311
2312		spa_config_set(spa, nvconfig);
2313		spa_unload(spa);
2314		spa_deactivate(spa);
2315		spa_activate(spa, orig_mode);
2316
2317		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2318	}
2319
2320	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2321		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2322	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2323	if (error != 0)
2324		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2325
2326	/*
2327	 * Load the bit that tells us to use the new accounting function
2328	 * (raid-z deflation).  If we have an older pool, this will not
2329	 * be present.
2330	 */
2331	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2332	if (error != 0 && error != ENOENT)
2333		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2334
2335	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2336	    &spa->spa_creation_version);
2337	if (error != 0 && error != ENOENT)
2338		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2339
2340	/*
2341	 * Load the persistent error log.  If we have an older pool, this will
2342	 * not be present.
2343	 */
2344	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2345	if (error != 0 && error != ENOENT)
2346		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2347
2348	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2349	    &spa->spa_errlog_scrub);
2350	if (error != 0 && error != ENOENT)
2351		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2352
2353	/*
2354	 * Load the history object.  If we have an older pool, this
2355	 * will not be present.
2356	 */
2357	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2358	if (error != 0 && error != ENOENT)
2359		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2360
2361	/*
2362	 * If we're assembling the pool from the split-off vdevs of
2363	 * an existing pool, we don't want to attach the spares & cache
2364	 * devices.
2365	 */
2366
2367	/*
2368	 * Load any hot spares for this pool.
2369	 */
2370	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2371	if (error != 0 && error != ENOENT)
2372		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2373	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2374		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2375		if (load_nvlist(spa, spa->spa_spares.sav_object,
2376		    &spa->spa_spares.sav_config) != 0)
2377			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2378
2379		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2380		spa_load_spares(spa);
2381		spa_config_exit(spa, SCL_ALL, FTAG);
2382	} else if (error == 0) {
2383		spa->spa_spares.sav_sync = B_TRUE;
2384	}
2385
2386	/*
2387	 * Load any level 2 ARC devices for this pool.
2388	 */
2389	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2390	    &spa->spa_l2cache.sav_object);
2391	if (error != 0 && error != ENOENT)
2392		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2393	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2394		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2395		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2396		    &spa->spa_l2cache.sav_config) != 0)
2397			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2398
2399		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2400		spa_load_l2cache(spa);
2401		spa_config_exit(spa, SCL_ALL, FTAG);
2402	} else if (error == 0) {
2403		spa->spa_l2cache.sav_sync = B_TRUE;
2404	}
2405
2406	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2407
2408	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2409	if (error && error != ENOENT)
2410		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2411
2412	if (error == 0) {
2413		uint64_t autoreplace;
2414
2415		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2416		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2417		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2418		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2419		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2420		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2421		    &spa->spa_dedup_ditto);
2422
2423		spa->spa_autoreplace = (autoreplace != 0);
2424	}
2425
2426	/*
2427	 * If the 'autoreplace' property is set, then post a resource notifying
2428	 * the ZFS DE that it should not issue any faults for unopenable
2429	 * devices.  We also iterate over the vdevs, and post a sysevent for any
2430	 * unopenable vdevs so that the normal autoreplace handler can take
2431	 * over.
2432	 */
2433	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2434		spa_check_removed(spa->spa_root_vdev);
2435		/*
2436		 * For the import case, this is done in spa_import(), because
2437		 * at this point we're using the spare definitions from
2438		 * the MOS config, not necessarily from the userland config.
2439		 */
2440		if (state != SPA_LOAD_IMPORT) {
2441			spa_aux_check_removed(&spa->spa_spares);
2442			spa_aux_check_removed(&spa->spa_l2cache);
2443		}
2444	}
2445
2446	/*
2447	 * Load the vdev state for all toplevel vdevs.
2448	 */
2449	vdev_load(rvd);
2450
2451	/*
2452	 * Propagate the leaf DTLs we just loaded all the way up the tree.
2453	 */
2454	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2455	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2456	spa_config_exit(spa, SCL_ALL, FTAG);
2457
2458	/*
2459	 * Load the DDTs (dedup tables).
2460	 */
2461	error = ddt_load(spa);
2462	if (error != 0)
2463		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2464
2465	spa_update_dspace(spa);
2466
2467	/*
2468	 * Validate the config, using the MOS config to fill in any
2469	 * information which might be missing.  If we fail to validate
2470	 * the config then declare the pool unfit for use. If we're
2471	 * assembling a pool from a split, the log is not transferred
2472	 * over.
2473	 */
2474	if (type != SPA_IMPORT_ASSEMBLE) {
2475		nvlist_t *nvconfig;
2476
2477		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2478			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2479
2480		if (!spa_config_valid(spa, nvconfig)) {
2481			nvlist_free(nvconfig);
2482			return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2483			    ENXIO));
2484		}
2485		nvlist_free(nvconfig);
2486
2487		/*
2488		 * Now that we've validated the config, check the state of the
2489		 * root vdev.  If it can't be opened, it indicates one or
2490		 * more toplevel vdevs are faulted.
2491		 */
2492		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2493			return (ENXIO);
2494
2495		if (spa_check_logs(spa)) {
2496			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2497			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2498		}
2499	}
2500
2501	if (missing_feat_write) {
2502		ASSERT(state == SPA_LOAD_TRYIMPORT);
2503
2504		/*
2505		 * At this point, we know that we can open the pool in
2506		 * read-only mode but not read-write mode. We now have enough
2507		 * information and can return to userland.
2508		 */
2509		return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2510	}
2511
2512	/*
2513	 * We've successfully opened the pool, verify that we're ready
2514	 * to start pushing transactions.
2515	 */
2516	if (state != SPA_LOAD_TRYIMPORT) {
2517		if (error = spa_load_verify(spa))
2518			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2519			    error));
2520	}
2521
2522	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2523	    spa->spa_load_max_txg == UINT64_MAX)) {
2524		dmu_tx_t *tx;
2525		int need_update = B_FALSE;
2526
2527		ASSERT(state != SPA_LOAD_TRYIMPORT);
2528
2529		/*
2530		 * Claim log blocks that haven't been committed yet.
2531		 * This must all happen in a single txg.
2532		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2533		 * invoked from zil_claim_log_block()'s i/o done callback.
2534		 * Price of rollback is that we abandon the log.
2535		 */
2536		spa->spa_claiming = B_TRUE;
2537
2538		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2539		    spa_first_txg(spa));
2540		(void) dmu_objset_find(spa_name(spa),
2541		    zil_claim, tx, DS_FIND_CHILDREN);
2542		dmu_tx_commit(tx);
2543
2544		spa->spa_claiming = B_FALSE;
2545
2546		spa_set_log_state(spa, SPA_LOG_GOOD);
2547		spa->spa_sync_on = B_TRUE;
2548		txg_sync_start(spa->spa_dsl_pool);
2549
2550		/*
2551		 * Wait for all claims to sync.  We sync up to the highest
2552		 * claimed log block birth time so that claimed log blocks
2553		 * don't appear to be from the future.  spa_claim_max_txg
2554		 * will have been set for us by either zil_check_log_chain()
2555		 * (invoked from spa_check_logs()) or zil_claim() above.
2556		 */
2557		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2558
2559		/*
2560		 * If the config cache is stale, or we have uninitialized
2561		 * metaslabs (see spa_vdev_add()), then update the config.
2562		 *
2563		 * If this is a verbatim import, trust the current
2564		 * in-core spa_config and update the disk labels.
2565		 */
2566		if (config_cache_txg != spa->spa_config_txg ||
2567		    state == SPA_LOAD_IMPORT ||
2568		    state == SPA_LOAD_RECOVER ||
2569		    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2570			need_update = B_TRUE;
2571
2572		for (int c = 0; c < rvd->vdev_children; c++)
2573			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2574				need_update = B_TRUE;
2575
2576		/*
2577		 * Update the config cache asychronously in case we're the
2578		 * root pool, in which case the config cache isn't writable yet.
2579		 */
2580		if (need_update)
2581			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2582
2583		/*
2584		 * Check all DTLs to see if anything needs resilvering.
2585		 */
2586		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2587		    vdev_resilver_needed(rvd, NULL, NULL))
2588			spa_async_request(spa, SPA_ASYNC_RESILVER);
2589
2590		/*
2591		 * Log the fact that we booted up (so that we can detect if
2592		 * we rebooted in the middle of an operation).
2593		 */
2594		spa_history_log_version(spa, "open");
2595
2596		/*
2597		 * Delete any inconsistent datasets.
2598		 */
2599		(void) dmu_objset_find(spa_name(spa),
2600		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2601
2602		/*
2603		 * Clean up any stale temporary dataset userrefs.
2604		 */
2605		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2606	}
2607
2608	return (0);
2609}
2610
2611static int
2612spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2613{
2614	int mode = spa->spa_mode;
2615
2616	spa_unload(spa);
2617	spa_deactivate(spa);
2618
2619	spa->spa_load_max_txg--;
2620
2621	spa_activate(spa, mode);
2622	spa_async_suspend(spa);
2623
2624	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2625}
2626
2627/*
2628 * If spa_load() fails this function will try loading prior txg's. If
2629 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2630 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2631 * function will not rewind the pool and will return the same error as
2632 * spa_load().
2633 */
2634static int
2635spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2636    uint64_t max_request, int rewind_flags)
2637{
2638	nvlist_t *loadinfo = NULL;
2639	nvlist_t *config = NULL;
2640	int load_error, rewind_error;
2641	uint64_t safe_rewind_txg;
2642	uint64_t min_txg;
2643
2644	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2645		spa->spa_load_max_txg = spa->spa_load_txg;
2646		spa_set_log_state(spa, SPA_LOG_CLEAR);
2647	} else {
2648		spa->spa_load_max_txg = max_request;
2649	}
2650
2651	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2652	    mosconfig);
2653	if (load_error == 0)
2654		return (0);
2655
2656	if (spa->spa_root_vdev != NULL)
2657		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2658
2659	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2660	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2661
2662	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2663		nvlist_free(config);
2664		return (load_error);
2665	}
2666
2667	if (state == SPA_LOAD_RECOVER) {
2668		/* Price of rolling back is discarding txgs, including log */
2669		spa_set_log_state(spa, SPA_LOG_CLEAR);
2670	} else {
2671		/*
2672		 * If we aren't rolling back save the load info from our first
2673		 * import attempt so that we can restore it after attempting
2674		 * to rewind.
2675		 */
2676		loadinfo = spa->spa_load_info;
2677		spa->spa_load_info = fnvlist_alloc();
2678	}
2679
2680	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2681	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2682	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2683	    TXG_INITIAL : safe_rewind_txg;
2684
2685	/*
2686	 * Continue as long as we're finding errors, we're still within
2687	 * the acceptable rewind range, and we're still finding uberblocks
2688	 */
2689	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2690	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2691		if (spa->spa_load_max_txg < safe_rewind_txg)
2692			spa->spa_extreme_rewind = B_TRUE;
2693		rewind_error = spa_load_retry(spa, state, mosconfig);
2694	}
2695
2696	spa->spa_extreme_rewind = B_FALSE;
2697	spa->spa_load_max_txg = UINT64_MAX;
2698
2699	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2700		spa_config_set(spa, config);
2701
2702	if (state == SPA_LOAD_RECOVER) {
2703		ASSERT3P(loadinfo, ==, NULL);
2704		return (rewind_error);
2705	} else {
2706		/* Store the rewind info as part of the initial load info */
2707		fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2708		    spa->spa_load_info);
2709
2710		/* Restore the initial load info */
2711		fnvlist_free(spa->spa_load_info);
2712		spa->spa_load_info = loadinfo;
2713
2714		return (load_error);
2715	}
2716}
2717
2718/*
2719 * Pool Open/Import
2720 *
2721 * The import case is identical to an open except that the configuration is sent
2722 * down from userland, instead of grabbed from the configuration cache.  For the
2723 * case of an open, the pool configuration will exist in the
2724 * POOL_STATE_UNINITIALIZED state.
2725 *
2726 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2727 * the same time open the pool, without having to keep around the spa_t in some
2728 * ambiguous state.
2729 */
2730static int
2731spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2732    nvlist_t **config)
2733{
2734	spa_t *spa;
2735	spa_load_state_t state = SPA_LOAD_OPEN;
2736	int error;
2737	int locked = B_FALSE;
2738
2739	*spapp = NULL;
2740
2741	/*
2742	 * As disgusting as this is, we need to support recursive calls to this
2743	 * function because dsl_dir_open() is called during spa_load(), and ends
2744	 * up calling spa_open() again.  The real fix is to figure out how to
2745	 * avoid dsl_dir_open() calling this in the first place.
2746	 */
2747	if (mutex_owner(&spa_namespace_lock) != curthread) {
2748		mutex_enter(&spa_namespace_lock);
2749		locked = B_TRUE;
2750	}
2751
2752	if ((spa = spa_lookup(pool)) == NULL) {
2753		if (locked)
2754			mutex_exit(&spa_namespace_lock);
2755		return (ENOENT);
2756	}
2757
2758	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2759		zpool_rewind_policy_t policy;
2760
2761		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2762		    &policy);
2763		if (policy.zrp_request & ZPOOL_DO_REWIND)
2764			state = SPA_LOAD_RECOVER;
2765
2766		spa_activate(spa, spa_mode_global);
2767
2768		if (state != SPA_LOAD_RECOVER)
2769			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2770
2771		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2772		    policy.zrp_request);
2773
2774		if (error == EBADF) {
2775			/*
2776			 * If vdev_validate() returns failure (indicated by
2777			 * EBADF), it indicates that one of the vdevs indicates
2778			 * that the pool has been exported or destroyed.  If
2779			 * this is the case, the config cache is out of sync and
2780			 * we should remove the pool from the namespace.
2781			 */
2782			spa_unload(spa);
2783			spa_deactivate(spa);
2784			spa_config_sync(spa, B_TRUE, B_TRUE);
2785			spa_remove(spa);
2786			if (locked)
2787				mutex_exit(&spa_namespace_lock);
2788			return (ENOENT);
2789		}
2790
2791		if (error) {
2792			/*
2793			 * We can't open the pool, but we still have useful
2794			 * information: the state of each vdev after the
2795			 * attempted vdev_open().  Return this to the user.
2796			 */
2797			if (config != NULL && spa->spa_config) {
2798				VERIFY(nvlist_dup(spa->spa_config, config,
2799				    KM_SLEEP) == 0);
2800				VERIFY(nvlist_add_nvlist(*config,
2801				    ZPOOL_CONFIG_LOAD_INFO,
2802				    spa->spa_load_info) == 0);
2803			}
2804			spa_unload(spa);
2805			spa_deactivate(spa);
2806			spa->spa_last_open_failed = error;
2807			if (locked)
2808				mutex_exit(&spa_namespace_lock);
2809			*spapp = NULL;
2810			return (error);
2811		}
2812	}
2813
2814	spa_open_ref(spa, tag);
2815
2816	if (config != NULL)
2817		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2818
2819	/*
2820	 * If we've recovered the pool, pass back any information we
2821	 * gathered while doing the load.
2822	 */
2823	if (state == SPA_LOAD_RECOVER) {
2824		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2825		    spa->spa_load_info) == 0);
2826	}
2827
2828	if (locked) {
2829		spa->spa_last_open_failed = 0;
2830		spa->spa_last_ubsync_txg = 0;
2831		spa->spa_load_txg = 0;
2832		mutex_exit(&spa_namespace_lock);
2833	}
2834
2835	*spapp = spa;
2836
2837	return (0);
2838}
2839
2840int
2841spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2842    nvlist_t **config)
2843{
2844	return (spa_open_common(name, spapp, tag, policy, config));
2845}
2846
2847int
2848spa_open(const char *name, spa_t **spapp, void *tag)
2849{
2850	return (spa_open_common(name, spapp, tag, NULL, NULL));
2851}
2852
2853/*
2854 * Lookup the given spa_t, incrementing the inject count in the process,
2855 * preventing it from being exported or destroyed.
2856 */
2857spa_t *
2858spa_inject_addref(char *name)
2859{
2860	spa_t *spa;
2861
2862	mutex_enter(&spa_namespace_lock);
2863	if ((spa = spa_lookup(name)) == NULL) {
2864		mutex_exit(&spa_namespace_lock);
2865		return (NULL);
2866	}
2867	spa->spa_inject_ref++;
2868	mutex_exit(&spa_namespace_lock);
2869
2870	return (spa);
2871}
2872
2873void
2874spa_inject_delref(spa_t *spa)
2875{
2876	mutex_enter(&spa_namespace_lock);
2877	spa->spa_inject_ref--;
2878	mutex_exit(&spa_namespace_lock);
2879}
2880
2881/*
2882 * Add spares device information to the nvlist.
2883 */
2884static void
2885spa_add_spares(spa_t *spa, nvlist_t *config)
2886{
2887	nvlist_t **spares;
2888	uint_t i, nspares;
2889	nvlist_t *nvroot;
2890	uint64_t guid;
2891	vdev_stat_t *vs;
2892	uint_t vsc;
2893	uint64_t pool;
2894
2895	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2896
2897	if (spa->spa_spares.sav_count == 0)
2898		return;
2899
2900	VERIFY(nvlist_lookup_nvlist(config,
2901	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2902	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2903	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2904	if (nspares != 0) {
2905		VERIFY(nvlist_add_nvlist_array(nvroot,
2906		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2907		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2908		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2909
2910		/*
2911		 * Go through and find any spares which have since been
2912		 * repurposed as an active spare.  If this is the case, update
2913		 * their status appropriately.
2914		 */
2915		for (i = 0; i < nspares; i++) {
2916			VERIFY(nvlist_lookup_uint64(spares[i],
2917			    ZPOOL_CONFIG_GUID, &guid) == 0);
2918			if (spa_spare_exists(guid, &pool, NULL) &&
2919			    pool != 0ULL) {
2920				VERIFY(nvlist_lookup_uint64_array(
2921				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
2922				    (uint64_t **)&vs, &vsc) == 0);
2923				vs->vs_state = VDEV_STATE_CANT_OPEN;
2924				vs->vs_aux = VDEV_AUX_SPARED;
2925			}
2926		}
2927	}
2928}
2929
2930/*
2931 * Add l2cache device information to the nvlist, including vdev stats.
2932 */
2933static void
2934spa_add_l2cache(spa_t *spa, nvlist_t *config)
2935{
2936	nvlist_t **l2cache;
2937	uint_t i, j, nl2cache;
2938	nvlist_t *nvroot;
2939	uint64_t guid;
2940	vdev_t *vd;
2941	vdev_stat_t *vs;
2942	uint_t vsc;
2943
2944	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2945
2946	if (spa->spa_l2cache.sav_count == 0)
2947		return;
2948
2949	VERIFY(nvlist_lookup_nvlist(config,
2950	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2951	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2952	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2953	if (nl2cache != 0) {
2954		VERIFY(nvlist_add_nvlist_array(nvroot,
2955		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2956		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2957		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2958
2959		/*
2960		 * Update level 2 cache device stats.
2961		 */
2962
2963		for (i = 0; i < nl2cache; i++) {
2964			VERIFY(nvlist_lookup_uint64(l2cache[i],
2965			    ZPOOL_CONFIG_GUID, &guid) == 0);
2966
2967			vd = NULL;
2968			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2969				if (guid ==
2970				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2971					vd = spa->spa_l2cache.sav_vdevs[j];
2972					break;
2973				}
2974			}
2975			ASSERT(vd != NULL);
2976
2977			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2978			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2979			    == 0);
2980			vdev_get_stats(vd, vs);
2981		}
2982	}
2983}
2984
2985static void
2986spa_add_feature_stats(spa_t *spa, nvlist_t *config)
2987{
2988	nvlist_t *features;
2989	zap_cursor_t zc;
2990	zap_attribute_t za;
2991
2992	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2993	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2994
2995	if (spa->spa_feat_for_read_obj != 0) {
2996		for (zap_cursor_init(&zc, spa->spa_meta_objset,
2997		    spa->spa_feat_for_read_obj);
2998		    zap_cursor_retrieve(&zc, &za) == 0;
2999		    zap_cursor_advance(&zc)) {
3000			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3001			    za.za_num_integers == 1);
3002			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3003			    za.za_first_integer));
3004		}
3005		zap_cursor_fini(&zc);
3006	}
3007
3008	if (spa->spa_feat_for_write_obj != 0) {
3009		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3010		    spa->spa_feat_for_write_obj);
3011		    zap_cursor_retrieve(&zc, &za) == 0;
3012		    zap_cursor_advance(&zc)) {
3013			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3014			    za.za_num_integers == 1);
3015			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3016			    za.za_first_integer));
3017		}
3018		zap_cursor_fini(&zc);
3019	}
3020
3021	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3022	    features) == 0);
3023	nvlist_free(features);
3024}
3025
3026int
3027spa_get_stats(const char *name, nvlist_t **config,
3028    char *altroot, size_t buflen)
3029{
3030	int error;
3031	spa_t *spa;
3032
3033	*config = NULL;
3034	error = spa_open_common(name, &spa, FTAG, NULL, config);
3035
3036	if (spa != NULL) {
3037		/*
3038		 * This still leaves a window of inconsistency where the spares
3039		 * or l2cache devices could change and the config would be
3040		 * self-inconsistent.
3041		 */
3042		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3043
3044		if (*config != NULL) {
3045			uint64_t loadtimes[2];
3046
3047			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3048			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3049			VERIFY(nvlist_add_uint64_array(*config,
3050			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3051
3052			VERIFY(nvlist_add_uint64(*config,
3053			    ZPOOL_CONFIG_ERRCOUNT,
3054			    spa_get_errlog_size(spa)) == 0);
3055
3056			if (spa_suspended(spa))
3057				VERIFY(nvlist_add_uint64(*config,
3058				    ZPOOL_CONFIG_SUSPENDED,
3059				    spa->spa_failmode) == 0);
3060
3061			spa_add_spares(spa, *config);
3062			spa_add_l2cache(spa, *config);
3063			spa_add_feature_stats(spa, *config);
3064		}
3065	}
3066
3067	/*
3068	 * We want to get the alternate root even for faulted pools, so we cheat
3069	 * and call spa_lookup() directly.
3070	 */
3071	if (altroot) {
3072		if (spa == NULL) {
3073			mutex_enter(&spa_namespace_lock);
3074			spa = spa_lookup(name);
3075			if (spa)
3076				spa_altroot(spa, altroot, buflen);
3077			else
3078				altroot[0] = '\0';
3079			spa = NULL;
3080			mutex_exit(&spa_namespace_lock);
3081		} else {
3082			spa_altroot(spa, altroot, buflen);
3083		}
3084	}
3085
3086	if (spa != NULL) {
3087		spa_config_exit(spa, SCL_CONFIG, FTAG);
3088		spa_close(spa, FTAG);
3089	}
3090
3091	return (error);
3092}
3093
3094/*
3095 * Validate that the auxiliary device array is well formed.  We must have an
3096 * array of nvlists, each which describes a valid leaf vdev.  If this is an
3097 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3098 * specified, as long as they are well-formed.
3099 */
3100static int
3101spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3102    spa_aux_vdev_t *sav, const char *config, uint64_t version,
3103    vdev_labeltype_t label)
3104{
3105	nvlist_t **dev;
3106	uint_t i, ndev;
3107	vdev_t *vd;
3108	int error;
3109
3110	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3111
3112	/*
3113	 * It's acceptable to have no devs specified.
3114	 */
3115	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3116		return (0);
3117
3118	if (ndev == 0)
3119		return (EINVAL);
3120
3121	/*
3122	 * Make sure the pool is formatted with a version that supports this
3123	 * device type.
3124	 */
3125	if (spa_version(spa) < version)
3126		return (ENOTSUP);
3127
3128	/*
3129	 * Set the pending device list so we correctly handle device in-use
3130	 * checking.
3131	 */
3132	sav->sav_pending = dev;
3133	sav->sav_npending = ndev;
3134
3135	for (i = 0; i < ndev; i++) {
3136		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3137		    mode)) != 0)
3138			goto out;
3139
3140		if (!vd->vdev_ops->vdev_op_leaf) {
3141			vdev_free(vd);
3142			error = EINVAL;
3143			goto out;
3144		}
3145
3146		/*
3147		 * The L2ARC currently only supports disk devices in
3148		 * kernel context.  For user-level testing, we allow it.
3149		 */
3150#ifdef _KERNEL
3151		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3152		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3153			error = ENOTBLK;
3154			vdev_free(vd);
3155			goto out;
3156		}
3157#endif
3158		vd->vdev_top = vd;
3159
3160		if ((error = vdev_open(vd)) == 0 &&
3161		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
3162			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3163			    vd->vdev_guid) == 0);
3164		}
3165
3166		vdev_free(vd);
3167
3168		if (error &&
3169		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3170			goto out;
3171		else
3172			error = 0;
3173	}
3174
3175out:
3176	sav->sav_pending = NULL;
3177	sav->sav_npending = 0;
3178	return (error);
3179}
3180
3181static int
3182spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3183{
3184	int error;
3185
3186	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3187
3188	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3189	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3190	    VDEV_LABEL_SPARE)) != 0) {
3191		return (error);
3192	}
3193
3194	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3195	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3196	    VDEV_LABEL_L2CACHE));
3197}
3198
3199static void
3200spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3201    const char *config)
3202{
3203	int i;
3204
3205	if (sav->sav_config != NULL) {
3206		nvlist_t **olddevs;
3207		uint_t oldndevs;
3208		nvlist_t **newdevs;
3209
3210		/*
3211		 * Generate new dev list by concatentating with the
3212		 * current dev list.
3213		 */
3214		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3215		    &olddevs, &oldndevs) == 0);
3216
3217		newdevs = kmem_alloc(sizeof (void *) *
3218		    (ndevs + oldndevs), KM_SLEEP);
3219		for (i = 0; i < oldndevs; i++)
3220			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3221			    KM_SLEEP) == 0);
3222		for (i = 0; i < ndevs; i++)
3223			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3224			    KM_SLEEP) == 0);
3225
3226		VERIFY(nvlist_remove(sav->sav_config, config,
3227		    DATA_TYPE_NVLIST_ARRAY) == 0);
3228
3229		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3230		    config, newdevs, ndevs + oldndevs) == 0);
3231		for (i = 0; i < oldndevs + ndevs; i++)
3232			nvlist_free(newdevs[i]);
3233		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3234	} else {
3235		/*
3236		 * Generate a new dev list.
3237		 */
3238		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3239		    KM_SLEEP) == 0);
3240		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3241		    devs, ndevs) == 0);
3242	}
3243}
3244
3245/*
3246 * Stop and drop level 2 ARC devices
3247 */
3248void
3249spa_l2cache_drop(spa_t *spa)
3250{
3251	vdev_t *vd;
3252	int i;
3253	spa_aux_vdev_t *sav = &spa->spa_l2cache;
3254
3255	for (i = 0; i < sav->sav_count; i++) {
3256		uint64_t pool;
3257
3258		vd = sav->sav_vdevs[i];
3259		ASSERT(vd != NULL);
3260
3261		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3262		    pool != 0ULL && l2arc_vdev_present(vd))
3263			l2arc_remove_vdev(vd);
3264	}
3265}
3266
3267/*
3268 * Pool Creation
3269 */
3270int
3271spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3272    nvlist_t *zplprops)
3273{
3274	spa_t *spa;
3275	char *altroot = NULL;
3276	vdev_t *rvd;
3277	dsl_pool_t *dp;
3278	dmu_tx_t *tx;
3279	int error = 0;
3280	uint64_t txg = TXG_INITIAL;
3281	nvlist_t **spares, **l2cache;
3282	uint_t nspares, nl2cache;
3283	uint64_t version, obj;
3284	boolean_t has_features;
3285
3286	/*
3287	 * If this pool already exists, return failure.
3288	 */
3289	mutex_enter(&spa_namespace_lock);
3290	if (spa_lookup(pool) != NULL) {
3291		mutex_exit(&spa_namespace_lock);
3292		return (EEXIST);
3293	}
3294
3295	/*
3296	 * Allocate a new spa_t structure.
3297	 */
3298	(void) nvlist_lookup_string(props,
3299	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3300	spa = spa_add(pool, NULL, altroot);
3301	spa_activate(spa, spa_mode_global);
3302
3303	if (props && (error = spa_prop_validate(spa, props))) {
3304		spa_deactivate(spa);
3305		spa_remove(spa);
3306		mutex_exit(&spa_namespace_lock);
3307		return (error);
3308	}
3309
3310	has_features = B_FALSE;
3311	for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3312	    elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3313		if (zpool_prop_feature(nvpair_name(elem)))
3314			has_features = B_TRUE;
3315	}
3316
3317	if (has_features || nvlist_lookup_uint64(props,
3318	    zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3319		version = SPA_VERSION;
3320	}
3321	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3322
3323	spa->spa_first_txg = txg;
3324	spa->spa_uberblock.ub_txg = txg - 1;
3325	spa->spa_uberblock.ub_version = version;
3326	spa->spa_ubsync = spa->spa_uberblock;
3327
3328	/*
3329	 * Create "The Godfather" zio to hold all async IOs
3330	 */
3331	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3332	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3333
3334	/*
3335	 * Create the root vdev.
3336	 */
3337	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3338
3339	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3340
3341	ASSERT(error != 0 || rvd != NULL);
3342	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3343
3344	if (error == 0 && !zfs_allocatable_devs(nvroot))
3345		error = EINVAL;
3346
3347	if (error == 0 &&
3348	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3349	    (error = spa_validate_aux(spa, nvroot, txg,
3350	    VDEV_ALLOC_ADD)) == 0) {
3351		for (int c = 0; c < rvd->vdev_children; c++) {
3352			vdev_metaslab_set_size(rvd->vdev_child[c]);
3353			vdev_expand(rvd->vdev_child[c], txg);
3354		}
3355	}
3356
3357	spa_config_exit(spa, SCL_ALL, FTAG);
3358
3359	if (error != 0) {
3360		spa_unload(spa);
3361		spa_deactivate(spa);
3362		spa_remove(spa);
3363		mutex_exit(&spa_namespace_lock);
3364		return (error);
3365	}
3366
3367	/*
3368	 * Get the list of spares, if specified.
3369	 */
3370	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3371	    &spares, &nspares) == 0) {
3372		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3373		    KM_SLEEP) == 0);
3374		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3375		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3376		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3377		spa_load_spares(spa);
3378		spa_config_exit(spa, SCL_ALL, FTAG);
3379		spa->spa_spares.sav_sync = B_TRUE;
3380	}
3381
3382	/*
3383	 * Get the list of level 2 cache devices, if specified.
3384	 */
3385	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3386	    &l2cache, &nl2cache) == 0) {
3387		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3388		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3389		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3390		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3391		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3392		spa_load_l2cache(spa);
3393		spa_config_exit(spa, SCL_ALL, FTAG);
3394		spa->spa_l2cache.sav_sync = B_TRUE;
3395	}
3396
3397	spa->spa_is_initializing = B_TRUE;
3398	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3399	spa->spa_meta_objset = dp->dp_meta_objset;
3400	spa->spa_is_initializing = B_FALSE;
3401
3402	/*
3403	 * Create DDTs (dedup tables).
3404	 */
3405	ddt_create(spa);
3406
3407	spa_update_dspace(spa);
3408
3409	tx = dmu_tx_create_assigned(dp, txg);
3410
3411	/*
3412	 * Create the pool config object.
3413	 */
3414	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3415	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3416	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3417
3418	if (zap_add(spa->spa_meta_objset,
3419	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3420	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3421		cmn_err(CE_PANIC, "failed to add pool config");
3422	}
3423
3424	if (spa_version(spa) >= SPA_VERSION_FEATURES)
3425		spa_feature_create_zap_objects(spa, tx);
3426
3427	if (zap_add(spa->spa_meta_objset,
3428	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3429	    sizeof (uint64_t), 1, &version, tx) != 0) {
3430		cmn_err(CE_PANIC, "failed to add pool version");
3431	}
3432
3433	/* Newly created pools with the right version are always deflated. */
3434	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3435		spa->spa_deflate = TRUE;
3436		if (zap_add(spa->spa_meta_objset,
3437		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3438		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3439			cmn_err(CE_PANIC, "failed to add deflate");
3440		}
3441	}
3442
3443	/*
3444	 * Create the deferred-free bpobj.  Turn off compression
3445	 * because sync-to-convergence takes longer if the blocksize
3446	 * keeps changing.
3447	 */
3448	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3449	dmu_object_set_compress(spa->spa_meta_objset, obj,
3450	    ZIO_COMPRESS_OFF, tx);
3451	if (zap_add(spa->spa_meta_objset,
3452	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3453	    sizeof (uint64_t), 1, &obj, tx) != 0) {
3454		cmn_err(CE_PANIC, "failed to add bpobj");
3455	}
3456	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3457	    spa->spa_meta_objset, obj));
3458
3459	/*
3460	 * Create the pool's history object.
3461	 */
3462	if (version >= SPA_VERSION_ZPOOL_HISTORY)
3463		spa_history_create_obj(spa, tx);
3464
3465	/*
3466	 * Set pool properties.
3467	 */
3468	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3469	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3470	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3471	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3472
3473	if (props != NULL) {
3474		spa_configfile_set(spa, props, B_FALSE);
3475		spa_sync_props(spa, props, tx);
3476	}
3477
3478	dmu_tx_commit(tx);
3479
3480	spa->spa_sync_on = B_TRUE;
3481	txg_sync_start(spa->spa_dsl_pool);
3482
3483	/*
3484	 * We explicitly wait for the first transaction to complete so that our
3485	 * bean counters are appropriately updated.
3486	 */
3487	txg_wait_synced(spa->spa_dsl_pool, txg);
3488
3489	spa_config_sync(spa, B_FALSE, B_TRUE);
3490
3491	spa_history_log_version(spa, "create");
3492
3493	spa->spa_minref = refcount_count(&spa->spa_refcount);
3494
3495	mutex_exit(&spa_namespace_lock);
3496
3497	return (0);
3498}
3499
3500#ifdef _KERNEL
3501/*
3502 * Get the root pool information from the root disk, then import the root pool
3503 * during the system boot up time.
3504 */
3505extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3506
3507static nvlist_t *
3508spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3509{
3510	nvlist_t *config;
3511	nvlist_t *nvtop, *nvroot;
3512	uint64_t pgid;
3513
3514	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3515		return (NULL);
3516
3517	/*
3518	 * Add this top-level vdev to the child array.
3519	 */
3520	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3521	    &nvtop) == 0);
3522	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3523	    &pgid) == 0);
3524	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3525
3526	/*
3527	 * Put this pool's top-level vdevs into a root vdev.
3528	 */
3529	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3530	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3531	    VDEV_TYPE_ROOT) == 0);
3532	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3533	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3534	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3535	    &nvtop, 1) == 0);
3536
3537	/*
3538	 * Replace the existing vdev_tree with the new root vdev in
3539	 * this pool's configuration (remove the old, add the new).
3540	 */
3541	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3542	nvlist_free(nvroot);
3543	return (config);
3544}
3545
3546/*
3547 * Walk the vdev tree and see if we can find a device with "better"
3548 * configuration. A configuration is "better" if the label on that
3549 * device has a more recent txg.
3550 */
3551static void
3552spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3553{
3554	for (int c = 0; c < vd->vdev_children; c++)
3555		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3556
3557	if (vd->vdev_ops->vdev_op_leaf) {
3558		nvlist_t *label;
3559		uint64_t label_txg;
3560
3561		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3562		    &label) != 0)
3563			return;
3564
3565		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3566		    &label_txg) == 0);
3567
3568		/*
3569		 * Do we have a better boot device?
3570		 */
3571		if (label_txg > *txg) {
3572			*txg = label_txg;
3573			*avd = vd;
3574		}
3575		nvlist_free(label);
3576	}
3577}
3578
3579/*
3580 * Import a root pool.
3581 *
3582 * For x86. devpath_list will consist of devid and/or physpath name of
3583 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3584 * The GRUB "findroot" command will return the vdev we should boot.
3585 *
3586 * For Sparc, devpath_list consists the physpath name of the booting device
3587 * no matter the rootpool is a single device pool or a mirrored pool.
3588 * e.g.
3589 *	"/pci@1f,0/ide@d/disk@0,0:a"
3590 */
3591int
3592spa_import_rootpool(char *devpath, char *devid)
3593{
3594	spa_t *spa;
3595	vdev_t *rvd, *bvd, *avd = NULL;
3596	nvlist_t *config, *nvtop;
3597	uint64_t guid, txg;
3598	char *pname;
3599	int error;
3600
3601	/*
3602	 * Read the label from the boot device and generate a configuration.
3603	 */
3604	config = spa_generate_rootconf(devpath, devid, &guid);
3605#if defined(_OBP) && defined(_KERNEL)
3606	if (config == NULL) {
3607		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3608			/* iscsi boot */
3609			get_iscsi_bootpath_phy(devpath);
3610			config = spa_generate_rootconf(devpath, devid, &guid);
3611		}
3612	}
3613#endif
3614	if (config == NULL) {
3615		cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3616		    devpath);
3617		return (EIO);
3618	}
3619
3620	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3621	    &pname) == 0);
3622	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3623
3624	mutex_enter(&spa_namespace_lock);
3625	if ((spa = spa_lookup(pname)) != NULL) {
3626		/*
3627		 * Remove the existing root pool from the namespace so that we
3628		 * can replace it with the correct config we just read in.
3629		 */
3630		spa_remove(spa);
3631	}
3632
3633	spa = spa_add(pname, config, NULL);
3634	spa->spa_is_root = B_TRUE;
3635	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3636
3637	/*
3638	 * Build up a vdev tree based on the boot device's label config.
3639	 */
3640	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3641	    &nvtop) == 0);
3642	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3643	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3644	    VDEV_ALLOC_ROOTPOOL);
3645	spa_config_exit(spa, SCL_ALL, FTAG);
3646	if (error) {
3647		mutex_exit(&spa_namespace_lock);
3648		nvlist_free(config);
3649		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3650		    pname);
3651		return (error);
3652	}
3653
3654	/*
3655	 * Get the boot vdev.
3656	 */
3657	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3658		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3659		    (u_longlong_t)guid);
3660		error = ENOENT;
3661		goto out;
3662	}
3663
3664	/*
3665	 * Determine if there is a better boot device.
3666	 */
3667	avd = bvd;
3668	spa_alt_rootvdev(rvd, &avd, &txg);
3669	if (avd != bvd) {
3670		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3671		    "try booting from '%s'", avd->vdev_path);
3672		error = EINVAL;
3673		goto out;
3674	}
3675
3676	/*
3677	 * If the boot device is part of a spare vdev then ensure that
3678	 * we're booting off the active spare.
3679	 */
3680	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3681	    !bvd->vdev_isspare) {
3682		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3683		    "try booting from '%s'",
3684		    bvd->vdev_parent->
3685		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3686		error = EINVAL;
3687		goto out;
3688	}
3689
3690	error = 0;
3691out:
3692	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3693	vdev_free(rvd);
3694	spa_config_exit(spa, SCL_ALL, FTAG);
3695	mutex_exit(&spa_namespace_lock);
3696
3697	nvlist_free(config);
3698	return (error);
3699}
3700
3701#endif
3702
3703/*
3704 * Import a non-root pool into the system.
3705 */
3706int
3707spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3708{
3709	spa_t *spa;
3710	char *altroot = NULL;
3711	spa_load_state_t state = SPA_LOAD_IMPORT;
3712	zpool_rewind_policy_t policy;
3713	uint64_t mode = spa_mode_global;
3714	uint64_t readonly = B_FALSE;
3715	int error;
3716	nvlist_t *nvroot;
3717	nvlist_t **spares, **l2cache;
3718	uint_t nspares, nl2cache;
3719
3720	/*
3721	 * If a pool with this name exists, return failure.
3722	 */
3723	mutex_enter(&spa_namespace_lock);
3724	if (spa_lookup(pool) != NULL) {
3725		mutex_exit(&spa_namespace_lock);
3726		return (EEXIST);
3727	}
3728
3729	/*
3730	 * Create and initialize the spa structure.
3731	 */
3732	(void) nvlist_lookup_string(props,
3733	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3734	(void) nvlist_lookup_uint64(props,
3735	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3736	if (readonly)
3737		mode = FREAD;
3738	spa = spa_add(pool, config, altroot);
3739	spa->spa_import_flags = flags;
3740
3741	/*
3742	 * Verbatim import - Take a pool and insert it into the namespace
3743	 * as if it had been loaded at boot.
3744	 */
3745	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3746		if (props != NULL)
3747			spa_configfile_set(spa, props, B_FALSE);
3748
3749		spa_config_sync(spa, B_FALSE, B_TRUE);
3750
3751		mutex_exit(&spa_namespace_lock);
3752		spa_history_log_version(spa, "import");
3753
3754		return (0);
3755	}
3756
3757	spa_activate(spa, mode);
3758
3759	/*
3760	 * Don't start async tasks until we know everything is healthy.
3761	 */
3762	spa_async_suspend(spa);
3763
3764	zpool_get_rewind_policy(config, &policy);
3765	if (policy.zrp_request & ZPOOL_DO_REWIND)
3766		state = SPA_LOAD_RECOVER;
3767
3768	/*
3769	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3770	 * because the user-supplied config is actually the one to trust when
3771	 * doing an import.
3772	 */
3773	if (state != SPA_LOAD_RECOVER)
3774		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3775
3776	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3777	    policy.zrp_request);
3778
3779	/*
3780	 * Propagate anything learned while loading the pool and pass it
3781	 * back to caller (i.e. rewind info, missing devices, etc).
3782	 */
3783	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3784	    spa->spa_load_info) == 0);
3785
3786	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3787	/*
3788	 * Toss any existing sparelist, as it doesn't have any validity
3789	 * anymore, and conflicts with spa_has_spare().
3790	 */
3791	if (spa->spa_spares.sav_config) {
3792		nvlist_free(spa->spa_spares.sav_config);
3793		spa->spa_spares.sav_config = NULL;
3794		spa_load_spares(spa);
3795	}
3796	if (spa->spa_l2cache.sav_config) {
3797		nvlist_free(spa->spa_l2cache.sav_config);
3798		spa->spa_l2cache.sav_config = NULL;
3799		spa_load_l2cache(spa);
3800	}
3801
3802	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3803	    &nvroot) == 0);
3804	if (error == 0)
3805		error = spa_validate_aux(spa, nvroot, -1ULL,
3806		    VDEV_ALLOC_SPARE);
3807	if (error == 0)
3808		error = spa_validate_aux(spa, nvroot, -1ULL,
3809		    VDEV_ALLOC_L2CACHE);
3810	spa_config_exit(spa, SCL_ALL, FTAG);
3811
3812	if (props != NULL)
3813		spa_configfile_set(spa, props, B_FALSE);
3814
3815	if (error != 0 || (props && spa_writeable(spa) &&
3816	    (error = spa_prop_set(spa, props)))) {
3817		spa_unload(spa);
3818		spa_deactivate(spa);
3819		spa_remove(spa);
3820		mutex_exit(&spa_namespace_lock);
3821		return (error);
3822	}
3823
3824	spa_async_resume(spa);
3825
3826	/*
3827	 * Override any spares and level 2 cache devices as specified by
3828	 * the user, as these may have correct device names/devids, etc.
3829	 */
3830	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3831	    &spares, &nspares) == 0) {
3832		if (spa->spa_spares.sav_config)
3833			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3834			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3835		else
3836			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3837			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3838		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3839		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3840		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3841		spa_load_spares(spa);
3842		spa_config_exit(spa, SCL_ALL, FTAG);
3843		spa->spa_spares.sav_sync = B_TRUE;
3844	}
3845	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3846	    &l2cache, &nl2cache) == 0) {
3847		if (spa->spa_l2cache.sav_config)
3848			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3849			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3850		else
3851			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3852			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3853		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3854		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3855		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3856		spa_load_l2cache(spa);
3857		spa_config_exit(spa, SCL_ALL, FTAG);
3858		spa->spa_l2cache.sav_sync = B_TRUE;
3859	}
3860
3861	/*
3862	 * Check for any removed devices.
3863	 */
3864	if (spa->spa_autoreplace) {
3865		spa_aux_check_removed(&spa->spa_spares);
3866		spa_aux_check_removed(&spa->spa_l2cache);
3867	}
3868
3869	if (spa_writeable(spa)) {
3870		/*
3871		 * Update the config cache to include the newly-imported pool.
3872		 */
3873		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3874	}
3875
3876	/*
3877	 * It's possible that the pool was expanded while it was exported.
3878	 * We kick off an async task to handle this for us.
3879	 */
3880	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3881
3882	mutex_exit(&spa_namespace_lock);
3883	spa_history_log_version(spa, "import");
3884
3885	return (0);
3886}
3887
3888nvlist_t *
3889spa_tryimport(nvlist_t *tryconfig)
3890{
3891	nvlist_t *config = NULL;
3892	char *poolname;
3893	spa_t *spa;
3894	uint64_t state;
3895	int error;
3896
3897	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3898		return (NULL);
3899
3900	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3901		return (NULL);
3902
3903	/*
3904	 * Create and initialize the spa structure.
3905	 */
3906	mutex_enter(&spa_namespace_lock);
3907	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3908	spa_activate(spa, FREAD);
3909
3910	/*
3911	 * Pass off the heavy lifting to spa_load().
3912	 * Pass TRUE for mosconfig because the user-supplied config
3913	 * is actually the one to trust when doing an import.
3914	 */
3915	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3916
3917	/*
3918	 * If 'tryconfig' was at least parsable, return the current config.
3919	 */
3920	if (spa->spa_root_vdev != NULL) {
3921		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3922		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3923		    poolname) == 0);
3924		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3925		    state) == 0);
3926		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3927		    spa->spa_uberblock.ub_timestamp) == 0);
3928		VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3929		    spa->spa_load_info) == 0);
3930
3931		/*
3932		 * If the bootfs property exists on this pool then we
3933		 * copy it out so that external consumers can tell which
3934		 * pools are bootable.
3935		 */
3936		if ((!error || error == EEXIST) && spa->spa_bootfs) {
3937			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3938
3939			/*
3940			 * We have to play games with the name since the
3941			 * pool was opened as TRYIMPORT_NAME.
3942			 */
3943			if (dsl_dsobj_to_dsname(spa_name(spa),
3944			    spa->spa_bootfs, tmpname) == 0) {
3945				char *cp;
3946				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3947
3948				cp = strchr(tmpname, '/');
3949				if (cp == NULL) {
3950					(void) strlcpy(dsname, tmpname,
3951					    MAXPATHLEN);
3952				} else {
3953					(void) snprintf(dsname, MAXPATHLEN,
3954					    "%s/%s", poolname, ++cp);
3955				}
3956				VERIFY(nvlist_add_string(config,
3957				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3958				kmem_free(dsname, MAXPATHLEN);
3959			}
3960			kmem_free(tmpname, MAXPATHLEN);
3961		}
3962
3963		/*
3964		 * Add the list of hot spares and level 2 cache devices.
3965		 */
3966		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3967		spa_add_spares(spa, config);
3968		spa_add_l2cache(spa, config);
3969		spa_config_exit(spa, SCL_CONFIG, FTAG);
3970	}
3971
3972	spa_unload(spa);
3973	spa_deactivate(spa);
3974	spa_remove(spa);
3975	mutex_exit(&spa_namespace_lock);
3976
3977	return (config);
3978}
3979
3980/*
3981 * Pool export/destroy
3982 *
3983 * The act of destroying or exporting a pool is very simple.  We make sure there
3984 * is no more pending I/O and any references to the pool are gone.  Then, we
3985 * update the pool state and sync all the labels to disk, removing the
3986 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3987 * we don't sync the labels or remove the configuration cache.
3988 */
3989static int
3990spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3991    boolean_t force, boolean_t hardforce)
3992{
3993	spa_t *spa;
3994
3995	if (oldconfig)
3996		*oldconfig = NULL;
3997
3998	if (!(spa_mode_global & FWRITE))
3999		return (EROFS);
4000
4001	mutex_enter(&spa_namespace_lock);
4002	if ((spa = spa_lookup(pool)) == NULL) {
4003		mutex_exit(&spa_namespace_lock);
4004		return (ENOENT);
4005	}
4006
4007	/*
4008	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4009	 * reacquire the namespace lock, and see if we can export.
4010	 */
4011	spa_open_ref(spa, FTAG);
4012	mutex_exit(&spa_namespace_lock);
4013	spa_async_suspend(spa);
4014	mutex_enter(&spa_namespace_lock);
4015	spa_close(spa, FTAG);
4016
4017	/*
4018	 * The pool will be in core if it's openable,
4019	 * in which case we can modify its state.
4020	 */
4021	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4022		/*
4023		 * Objsets may be open only because they're dirty, so we
4024		 * have to force it to sync before checking spa_refcnt.
4025		 */
4026		txg_wait_synced(spa->spa_dsl_pool, 0);
4027
4028		/*
4029		 * A pool cannot be exported or destroyed if there are active
4030		 * references.  If we are resetting a pool, allow references by
4031		 * fault injection handlers.
4032		 */
4033		if (!spa_refcount_zero(spa) ||
4034		    (spa->spa_inject_ref != 0 &&
4035		    new_state != POOL_STATE_UNINITIALIZED)) {
4036			spa_async_resume(spa);
4037			mutex_exit(&spa_namespace_lock);
4038			return (EBUSY);
4039		}
4040
4041		/*
4042		 * A pool cannot be exported if it has an active shared spare.
4043		 * This is to prevent other pools stealing the active spare
4044		 * from an exported pool. At user's own will, such pool can
4045		 * be forcedly exported.
4046		 */
4047		if (!force && new_state == POOL_STATE_EXPORTED &&
4048		    spa_has_active_shared_spare(spa)) {
4049			spa_async_resume(spa);
4050			mutex_exit(&spa_namespace_lock);
4051			return (EXDEV);
4052		}
4053
4054		/*
4055		 * We want this to be reflected on every label,
4056		 * so mark them all dirty.  spa_unload() will do the
4057		 * final sync that pushes these changes out.
4058		 */
4059		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4060			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4061			spa->spa_state = new_state;
4062			spa->spa_final_txg = spa_last_synced_txg(spa) +
4063			    TXG_DEFER_SIZE + 1;
4064			vdev_config_dirty(spa->spa_root_vdev);
4065			spa_config_exit(spa, SCL_ALL, FTAG);
4066		}
4067	}
4068
4069	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4070
4071	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4072		spa_unload(spa);
4073		spa_deactivate(spa);
4074	}
4075
4076	if (oldconfig && spa->spa_config)
4077		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4078
4079	if (new_state != POOL_STATE_UNINITIALIZED) {
4080		if (!hardforce)
4081			spa_config_sync(spa, B_TRUE, B_TRUE);
4082		spa_remove(spa);
4083	}
4084	mutex_exit(&spa_namespace_lock);
4085
4086	return (0);
4087}
4088
4089/*
4090 * Destroy a storage pool.
4091 */
4092int
4093spa_destroy(char *pool)
4094{
4095	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4096	    B_FALSE, B_FALSE));
4097}
4098
4099/*
4100 * Export a storage pool.
4101 */
4102int
4103spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4104    boolean_t hardforce)
4105{
4106	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4107	    force, hardforce));
4108}
4109
4110/*
4111 * Similar to spa_export(), this unloads the spa_t without actually removing it
4112 * from the namespace in any way.
4113 */
4114int
4115spa_reset(char *pool)
4116{
4117	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4118	    B_FALSE, B_FALSE));
4119}
4120
4121/*
4122 * ==========================================================================
4123 * Device manipulation
4124 * ==========================================================================
4125 */
4126
4127/*
4128 * Add a device to a storage pool.
4129 */
4130int
4131spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4132{
4133	uint64_t txg, id;
4134	int error;
4135	vdev_t *rvd = spa->spa_root_vdev;
4136	vdev_t *vd, *tvd;
4137	nvlist_t **spares, **l2cache;
4138	uint_t nspares, nl2cache;
4139
4140	ASSERT(spa_writeable(spa));
4141
4142	txg = spa_vdev_enter(spa);
4143
4144	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4145	    VDEV_ALLOC_ADD)) != 0)
4146		return (spa_vdev_exit(spa, NULL, txg, error));
4147
4148	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
4149
4150	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4151	    &nspares) != 0)
4152		nspares = 0;
4153
4154	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4155	    &nl2cache) != 0)
4156		nl2cache = 0;
4157
4158	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4159		return (spa_vdev_exit(spa, vd, txg, EINVAL));
4160
4161	if (vd->vdev_children != 0 &&
4162	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
4163		return (spa_vdev_exit(spa, vd, txg, error));
4164
4165	/*
4166	 * We must validate the spares and l2cache devices after checking the
4167	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
4168	 */
4169	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4170		return (spa_vdev_exit(spa, vd, txg, error));
4171
4172	/*
4173	 * Transfer each new top-level vdev from vd to rvd.
4174	 */
4175	for (int c = 0; c < vd->vdev_children; c++) {
4176
4177		/*
4178		 * Set the vdev id to the first hole, if one exists.
4179		 */
4180		for (id = 0; id < rvd->vdev_children; id++) {
4181			if (rvd->vdev_child[id]->vdev_ishole) {
4182				vdev_free(rvd->vdev_child[id]);
4183				break;
4184			}
4185		}
4186		tvd = vd->vdev_child[c];
4187		vdev_remove_child(vd, tvd);
4188		tvd->vdev_id = id;
4189		vdev_add_child(rvd, tvd);
4190		vdev_config_dirty(tvd);
4191	}
4192
4193	if (nspares != 0) {
4194		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4195		    ZPOOL_CONFIG_SPARES);
4196		spa_load_spares(spa);
4197		spa->spa_spares.sav_sync = B_TRUE;
4198	}
4199
4200	if (nl2cache != 0) {
4201		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4202		    ZPOOL_CONFIG_L2CACHE);
4203		spa_load_l2cache(spa);
4204		spa->spa_l2cache.sav_sync = B_TRUE;
4205	}
4206
4207	/*
4208	 * We have to be careful when adding new vdevs to an existing pool.
4209	 * If other threads start allocating from these vdevs before we
4210	 * sync the config cache, and we lose power, then upon reboot we may
4211	 * fail to open the pool because there are DVAs that the config cache
4212	 * can't translate.  Therefore, we first add the vdevs without
4213	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4214	 * and then let spa_config_update() initialize the new metaslabs.
4215	 *
4216	 * spa_load() checks for added-but-not-initialized vdevs, so that
4217	 * if we lose power at any point in this sequence, the remaining
4218	 * steps will be completed the next time we load the pool.
4219	 */
4220	(void) spa_vdev_exit(spa, vd, txg, 0);
4221
4222	mutex_enter(&spa_namespace_lock);
4223	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4224	mutex_exit(&spa_namespace_lock);
4225
4226	return (0);
4227}
4228
4229/*
4230 * Attach a device to a mirror.  The arguments are the path to any device
4231 * in the mirror, and the nvroot for the new device.  If the path specifies
4232 * a device that is not mirrored, we automatically insert the mirror vdev.
4233 *
4234 * If 'replacing' is specified, the new device is intended to replace the
4235 * existing device; in this case the two devices are made into their own
4236 * mirror using the 'replacing' vdev, which is functionally identical to
4237 * the mirror vdev (it actually reuses all the same ops) but has a few
4238 * extra rules: you can't attach to it after it's been created, and upon
4239 * completion of resilvering, the first disk (the one being replaced)
4240 * is automatically detached.
4241 */
4242int
4243spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4244{
4245	uint64_t txg, dtl_max_txg;
4246	vdev_t *rvd = spa->spa_root_vdev;
4247	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4248	vdev_ops_t *pvops;
4249	char *oldvdpath, *newvdpath;
4250	int newvd_isspare;
4251	int error;
4252
4253	ASSERT(spa_writeable(spa));
4254
4255	txg = spa_vdev_enter(spa);
4256
4257	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4258
4259	if (oldvd == NULL)
4260		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4261
4262	if (!oldvd->vdev_ops->vdev_op_leaf)
4263		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4264
4265	pvd = oldvd->vdev_parent;
4266
4267	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4268	    VDEV_ALLOC_ATTACH)) != 0)
4269		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4270
4271	if (newrootvd->vdev_children != 1)
4272		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4273
4274	newvd = newrootvd->vdev_child[0];
4275
4276	if (!newvd->vdev_ops->vdev_op_leaf)
4277		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4278
4279	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4280		return (spa_vdev_exit(spa, newrootvd, txg, error));
4281
4282	/*
4283	 * Spares can't replace logs
4284	 */
4285	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4286		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4287
4288	if (!replacing) {
4289		/*
4290		 * For attach, the only allowable parent is a mirror or the root
4291		 * vdev.
4292		 */
4293		if (pvd->vdev_ops != &vdev_mirror_ops &&
4294		    pvd->vdev_ops != &vdev_root_ops)
4295			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4296
4297		pvops = &vdev_mirror_ops;
4298	} else {
4299		/*
4300		 * Active hot spares can only be replaced by inactive hot
4301		 * spares.
4302		 */
4303		if (pvd->vdev_ops == &vdev_spare_ops &&
4304		    oldvd->vdev_isspare &&
4305		    !spa_has_spare(spa, newvd->vdev_guid))
4306			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4307
4308		/*
4309		 * If the source is a hot spare, and the parent isn't already a
4310		 * spare, then we want to create a new hot spare.  Otherwise, we
4311		 * want to create a replacing vdev.  The user is not allowed to
4312		 * attach to a spared vdev child unless the 'isspare' state is
4313		 * the same (spare replaces spare, non-spare replaces
4314		 * non-spare).
4315		 */
4316		if (pvd->vdev_ops == &vdev_replacing_ops &&
4317		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4318			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4319		} else if (pvd->vdev_ops == &vdev_spare_ops &&
4320		    newvd->vdev_isspare != oldvd->vdev_isspare) {
4321			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4322		}
4323
4324		if (newvd->vdev_isspare)
4325			pvops = &vdev_spare_ops;
4326		else
4327			pvops = &vdev_replacing_ops;
4328	}
4329
4330	/*
4331	 * Make sure the new device is big enough.
4332	 */
4333	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4334		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4335
4336	/*
4337	 * The new device cannot have a higher alignment requirement
4338	 * than the top-level vdev.
4339	 */
4340	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4341		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4342
4343	/*
4344	 * If this is an in-place replacement, update oldvd's path and devid
4345	 * to make it distinguishable from newvd, and unopenable from now on.
4346	 */
4347	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4348		spa_strfree(oldvd->vdev_path);
4349		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4350		    KM_SLEEP);
4351		(void) sprintf(oldvd->vdev_path, "%s/%s",
4352		    newvd->vdev_path, "old");
4353		if (oldvd->vdev_devid != NULL) {
4354			spa_strfree(oldvd->vdev_devid);
4355			oldvd->vdev_devid = NULL;
4356		}
4357	}
4358
4359	/* mark the device being resilvered */
4360	newvd->vdev_resilvering = B_TRUE;
4361
4362	/*
4363	 * If the parent is not a mirror, or if we're replacing, insert the new
4364	 * mirror/replacing/spare vdev above oldvd.
4365	 */
4366	if (pvd->vdev_ops != pvops)
4367		pvd = vdev_add_parent(oldvd, pvops);
4368
4369	ASSERT(pvd->vdev_top->vdev_parent == rvd);
4370	ASSERT(pvd->vdev_ops == pvops);
4371	ASSERT(oldvd->vdev_parent == pvd);
4372
4373	/*
4374	 * Extract the new device from its root and add it to pvd.
4375	 */
4376	vdev_remove_child(newrootvd, newvd);
4377	newvd->vdev_id = pvd->vdev_children;
4378	newvd->vdev_crtxg = oldvd->vdev_crtxg;
4379	vdev_add_child(pvd, newvd);
4380
4381	tvd = newvd->vdev_top;
4382	ASSERT(pvd->vdev_top == tvd);
4383	ASSERT(tvd->vdev_parent == rvd);
4384
4385	vdev_config_dirty(tvd);
4386
4387	/*
4388	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4389	 * for any dmu_sync-ed blocks.  It will propagate upward when
4390	 * spa_vdev_exit() calls vdev_dtl_reassess().
4391	 */
4392	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4393
4394	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4395	    dtl_max_txg - TXG_INITIAL);
4396
4397	if (newvd->vdev_isspare) {
4398		spa_spare_activate(newvd);
4399		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4400	}
4401
4402	oldvdpath = spa_strdup(oldvd->vdev_path);
4403	newvdpath = spa_strdup(newvd->vdev_path);
4404	newvd_isspare = newvd->vdev_isspare;
4405
4406	/*
4407	 * Mark newvd's DTL dirty in this txg.
4408	 */
4409	vdev_dirty(tvd, VDD_DTL, newvd, txg);
4410
4411	/*
4412	 * Restart the resilver
4413	 */
4414	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4415
4416	/*
4417	 * Commit the config
4418	 */
4419	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4420
4421	spa_history_log_internal(spa, "vdev attach", NULL,
4422	    "%s vdev=%s %s vdev=%s",
4423	    replacing && newvd_isspare ? "spare in" :
4424	    replacing ? "replace" : "attach", newvdpath,
4425	    replacing ? "for" : "to", oldvdpath);
4426
4427	spa_strfree(oldvdpath);
4428	spa_strfree(newvdpath);
4429
4430	if (spa->spa_bootfs)
4431		spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4432
4433	return (0);
4434}
4435
4436/*
4437 * Detach a device from a mirror or replacing vdev.
4438 * If 'replace_done' is specified, only detach if the parent
4439 * is a replacing vdev.
4440 */
4441int
4442spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4443{
4444	uint64_t txg;
4445	int error;
4446	vdev_t *rvd = spa->spa_root_vdev;
4447	vdev_t *vd, *pvd, *cvd, *tvd;
4448	boolean_t unspare = B_FALSE;
4449	uint64_t unspare_guid;
4450	char *vdpath;
4451
4452	ASSERT(spa_writeable(spa));
4453
4454	txg = spa_vdev_enter(spa);
4455
4456	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4457
4458	if (vd == NULL)
4459		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4460
4461	if (!vd->vdev_ops->vdev_op_leaf)
4462		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4463
4464	pvd = vd->vdev_parent;
4465
4466	/*
4467	 * If the parent/child relationship is not as expected, don't do it.
4468	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4469	 * vdev that's replacing B with C.  The user's intent in replacing
4470	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4471	 * the replace by detaching C, the expected behavior is to end up
4472	 * M(A,B).  But suppose that right after deciding to detach C,
4473	 * the replacement of B completes.  We would have M(A,C), and then
4474	 * ask to detach C, which would leave us with just A -- not what
4475	 * the user wanted.  To prevent this, we make sure that the
4476	 * parent/child relationship hasn't changed -- in this example,
4477	 * that C's parent is still the replacing vdev R.
4478	 */
4479	if (pvd->vdev_guid != pguid && pguid != 0)
4480		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4481
4482	/*
4483	 * Only 'replacing' or 'spare' vdevs can be replaced.
4484	 */
4485	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4486	    pvd->vdev_ops != &vdev_spare_ops)
4487		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4488
4489	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4490	    spa_version(spa) >= SPA_VERSION_SPARES);
4491
4492	/*
4493	 * Only mirror, replacing, and spare vdevs support detach.
4494	 */
4495	if (pvd->vdev_ops != &vdev_replacing_ops &&
4496	    pvd->vdev_ops != &vdev_mirror_ops &&
4497	    pvd->vdev_ops != &vdev_spare_ops)
4498		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4499
4500	/*
4501	 * If this device has the only valid copy of some data,
4502	 * we cannot safely detach it.
4503	 */
4504	if (vdev_dtl_required(vd))
4505		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4506
4507	ASSERT(pvd->vdev_children >= 2);
4508
4509	/*
4510	 * If we are detaching the second disk from a replacing vdev, then
4511	 * check to see if we changed the original vdev's path to have "/old"
4512	 * at the end in spa_vdev_attach().  If so, undo that change now.
4513	 */
4514	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4515	    vd->vdev_path != NULL) {
4516		size_t len = strlen(vd->vdev_path);
4517
4518		for (int c = 0; c < pvd->vdev_children; c++) {
4519			cvd = pvd->vdev_child[c];
4520
4521			if (cvd == vd || cvd->vdev_path == NULL)
4522				continue;
4523
4524			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4525			    strcmp(cvd->vdev_path + len, "/old") == 0) {
4526				spa_strfree(cvd->vdev_path);
4527				cvd->vdev_path = spa_strdup(vd->vdev_path);
4528				break;
4529			}
4530		}
4531	}
4532
4533	/*
4534	 * If we are detaching the original disk from a spare, then it implies
4535	 * that the spare should become a real disk, and be removed from the
4536	 * active spare list for the pool.
4537	 */
4538	if (pvd->vdev_ops == &vdev_spare_ops &&
4539	    vd->vdev_id == 0 &&
4540	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4541		unspare = B_TRUE;
4542
4543	/*
4544	 * Erase the disk labels so the disk can be used for other things.
4545	 * This must be done after all other error cases are handled,
4546	 * but before we disembowel vd (so we can still do I/O to it).
4547	 * But if we can't do it, don't treat the error as fatal --
4548	 * it may be that the unwritability of the disk is the reason
4549	 * it's being detached!
4550	 */
4551	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4552
4553	/*
4554	 * Remove vd from its parent and compact the parent's children.
4555	 */
4556	vdev_remove_child(pvd, vd);
4557	vdev_compact_children(pvd);
4558
4559	/*
4560	 * Remember one of the remaining children so we can get tvd below.
4561	 */
4562	cvd = pvd->vdev_child[pvd->vdev_children - 1];
4563
4564	/*
4565	 * If we need to remove the remaining child from the list of hot spares,
4566	 * do it now, marking the vdev as no longer a spare in the process.
4567	 * We must do this before vdev_remove_parent(), because that can
4568	 * change the GUID if it creates a new toplevel GUID.  For a similar
4569	 * reason, we must remove the spare now, in the same txg as the detach;
4570	 * otherwise someone could attach a new sibling, change the GUID, and
4571	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4572	 */
4573	if (unspare) {
4574		ASSERT(cvd->vdev_isspare);
4575		spa_spare_remove(cvd);
4576		unspare_guid = cvd->vdev_guid;
4577		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4578		cvd->vdev_unspare = B_TRUE;
4579	}
4580
4581	/*
4582	 * If the parent mirror/replacing vdev only has one child,
4583	 * the parent is no longer needed.  Remove it from the tree.
4584	 */
4585	if (pvd->vdev_children == 1) {
4586		if (pvd->vdev_ops == &vdev_spare_ops)
4587			cvd->vdev_unspare = B_FALSE;
4588		vdev_remove_parent(cvd);
4589		cvd->vdev_resilvering = B_FALSE;
4590	}
4591
4592
4593	/*
4594	 * We don't set tvd until now because the parent we just removed
4595	 * may have been the previous top-level vdev.
4596	 */
4597	tvd = cvd->vdev_top;
4598	ASSERT(tvd->vdev_parent == rvd);
4599
4600	/*
4601	 * Reevaluate the parent vdev state.
4602	 */
4603	vdev_propagate_state(cvd);
4604
4605	/*
4606	 * If the 'autoexpand' property is set on the pool then automatically
4607	 * try to expand the size of the pool. For example if the device we
4608	 * just detached was smaller than the others, it may be possible to
4609	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4610	 * first so that we can obtain the updated sizes of the leaf vdevs.
4611	 */
4612	if (spa->spa_autoexpand) {
4613		vdev_reopen(tvd);
4614		vdev_expand(tvd, txg);
4615	}
4616
4617	vdev_config_dirty(tvd);
4618
4619	/*
4620	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4621	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4622	 * But first make sure we're not on any *other* txg's DTL list, to
4623	 * prevent vd from being accessed after it's freed.
4624	 */
4625	vdpath = spa_strdup(vd->vdev_path);
4626	for (int t = 0; t < TXG_SIZE; t++)
4627		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4628	vd->vdev_detached = B_TRUE;
4629	vdev_dirty(tvd, VDD_DTL, vd, txg);
4630
4631	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4632
4633	/* hang on to the spa before we release the lock */
4634	spa_open_ref(spa, FTAG);
4635
4636	error = spa_vdev_exit(spa, vd, txg, 0);
4637
4638	spa_history_log_internal(spa, "detach", NULL,
4639	    "vdev=%s", vdpath);
4640	spa_strfree(vdpath);
4641
4642	/*
4643	 * If this was the removal of the original device in a hot spare vdev,
4644	 * then we want to go through and remove the device from the hot spare
4645	 * list of every other pool.
4646	 */
4647	if (unspare) {
4648		spa_t *altspa = NULL;
4649
4650		mutex_enter(&spa_namespace_lock);
4651		while ((altspa = spa_next(altspa)) != NULL) {
4652			if (altspa->spa_state != POOL_STATE_ACTIVE ||
4653			    altspa == spa)
4654				continue;
4655
4656			spa_open_ref(altspa, FTAG);
4657			mutex_exit(&spa_namespace_lock);
4658			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4659			mutex_enter(&spa_namespace_lock);
4660			spa_close(altspa, FTAG);
4661		}
4662		mutex_exit(&spa_namespace_lock);
4663
4664		/* search the rest of the vdevs for spares to remove */
4665		spa_vdev_resilver_done(spa);
4666	}
4667
4668	/* all done with the spa; OK to release */
4669	mutex_enter(&spa_namespace_lock);
4670	spa_close(spa, FTAG);
4671	mutex_exit(&spa_namespace_lock);
4672
4673	return (error);
4674}
4675
4676/*
4677 * Split a set of devices from their mirrors, and create a new pool from them.
4678 */
4679int
4680spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4681    nvlist_t *props, boolean_t exp)
4682{
4683	int error = 0;
4684	uint64_t txg, *glist;
4685	spa_t *newspa;
4686	uint_t c, children, lastlog;
4687	nvlist_t **child, *nvl, *tmp;
4688	dmu_tx_t *tx;
4689	char *altroot = NULL;
4690	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4691	boolean_t activate_slog;
4692
4693	ASSERT(spa_writeable(spa));
4694
4695	txg = spa_vdev_enter(spa);
4696
4697	/* clear the log and flush everything up to now */
4698	activate_slog = spa_passivate_log(spa);
4699	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4700	error = spa_offline_log(spa);
4701	txg = spa_vdev_config_enter(spa);
4702
4703	if (activate_slog)
4704		spa_activate_log(spa);
4705
4706	if (error != 0)
4707		return (spa_vdev_exit(spa, NULL, txg, error));
4708
4709	/* check new spa name before going any further */
4710	if (spa_lookup(newname) != NULL)
4711		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4712
4713	/*
4714	 * scan through all the children to ensure they're all mirrors
4715	 */
4716	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4717	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4718	    &children) != 0)
4719		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4720
4721	/* first, check to ensure we've got the right child count */
4722	rvd = spa->spa_root_vdev;
4723	lastlog = 0;
4724	for (c = 0; c < rvd->vdev_children; c++) {
4725		vdev_t *vd = rvd->vdev_child[c];
4726
4727		/* don't count the holes & logs as children */
4728		if (vd->vdev_islog || vd->vdev_ishole) {
4729			if (lastlog == 0)
4730				lastlog = c;
4731			continue;
4732		}
4733
4734		lastlog = 0;
4735	}
4736	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4737		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4738
4739	/* next, ensure no spare or cache devices are part of the split */
4740	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4741	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4742		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4743
4744	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4745	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4746
4747	/* then, loop over each vdev and validate it */
4748	for (c = 0; c < children; c++) {
4749		uint64_t is_hole = 0;
4750
4751		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4752		    &is_hole);
4753
4754		if (is_hole != 0) {
4755			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4756			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4757				continue;
4758			} else {
4759				error = EINVAL;
4760				break;
4761			}
4762		}
4763
4764		/* which disk is going to be split? */
4765		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4766		    &glist[c]) != 0) {
4767			error = EINVAL;
4768			break;
4769		}
4770
4771		/* look it up in the spa */
4772		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4773		if (vml[c] == NULL) {
4774			error = ENODEV;
4775			break;
4776		}
4777
4778		/* make sure there's nothing stopping the split */
4779		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4780		    vml[c]->vdev_islog ||
4781		    vml[c]->vdev_ishole ||
4782		    vml[c]->vdev_isspare ||
4783		    vml[c]->vdev_isl2cache ||
4784		    !vdev_writeable(vml[c]) ||
4785		    vml[c]->vdev_children != 0 ||
4786		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4787		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4788			error = EINVAL;
4789			break;
4790		}
4791
4792		if (vdev_dtl_required(vml[c])) {
4793			error = EBUSY;
4794			break;
4795		}
4796
4797		/* we need certain info from the top level */
4798		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4799		    vml[c]->vdev_top->vdev_ms_array) == 0);
4800		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4801		    vml[c]->vdev_top->vdev_ms_shift) == 0);
4802		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4803		    vml[c]->vdev_top->vdev_asize) == 0);
4804		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4805		    vml[c]->vdev_top->vdev_ashift) == 0);
4806	}
4807
4808	if (error != 0) {
4809		kmem_free(vml, children * sizeof (vdev_t *));
4810		kmem_free(glist, children * sizeof (uint64_t));
4811		return (spa_vdev_exit(spa, NULL, txg, error));
4812	}
4813
4814	/* stop writers from using the disks */
4815	for (c = 0; c < children; c++) {
4816		if (vml[c] != NULL)
4817			vml[c]->vdev_offline = B_TRUE;
4818	}
4819	vdev_reopen(spa->spa_root_vdev);
4820
4821	/*
4822	 * Temporarily record the splitting vdevs in the spa config.  This
4823	 * will disappear once the config is regenerated.
4824	 */
4825	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4826	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4827	    glist, children) == 0);
4828	kmem_free(glist, children * sizeof (uint64_t));
4829
4830	mutex_enter(&spa->spa_props_lock);
4831	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4832	    nvl) == 0);
4833	mutex_exit(&spa->spa_props_lock);
4834	spa->spa_config_splitting = nvl;
4835	vdev_config_dirty(spa->spa_root_vdev);
4836
4837	/* configure and create the new pool */
4838	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4839	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4840	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4841	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4842	    spa_version(spa)) == 0);
4843	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4844	    spa->spa_config_txg) == 0);
4845	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4846	    spa_generate_guid(NULL)) == 0);
4847	(void) nvlist_lookup_string(props,
4848	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4849
4850	/* add the new pool to the namespace */
4851	newspa = spa_add(newname, config, altroot);
4852	newspa->spa_config_txg = spa->spa_config_txg;
4853	spa_set_log_state(newspa, SPA_LOG_CLEAR);
4854
4855	/* release the spa config lock, retaining the namespace lock */
4856	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4857
4858	if (zio_injection_enabled)
4859		zio_handle_panic_injection(spa, FTAG, 1);
4860
4861	spa_activate(newspa, spa_mode_global);
4862	spa_async_suspend(newspa);
4863
4864	/* create the new pool from the disks of the original pool */
4865	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4866	if (error)
4867		goto out;
4868
4869	/* if that worked, generate a real config for the new pool */
4870	if (newspa->spa_root_vdev != NULL) {
4871		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4872		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4873		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4874		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4875		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4876		    B_TRUE));
4877	}
4878
4879	/* set the props */
4880	if (props != NULL) {
4881		spa_configfile_set(newspa, props, B_FALSE);
4882		error = spa_prop_set(newspa, props);
4883		if (error)
4884			goto out;
4885	}
4886
4887	/* flush everything */
4888	txg = spa_vdev_config_enter(newspa);
4889	vdev_config_dirty(newspa->spa_root_vdev);
4890	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4891
4892	if (zio_injection_enabled)
4893		zio_handle_panic_injection(spa, FTAG, 2);
4894
4895	spa_async_resume(newspa);
4896
4897	/* finally, update the original pool's config */
4898	txg = spa_vdev_config_enter(spa);
4899	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4900	error = dmu_tx_assign(tx, TXG_WAIT);
4901	if (error != 0)
4902		dmu_tx_abort(tx);
4903	for (c = 0; c < children; c++) {
4904		if (vml[c] != NULL) {
4905			vdev_split(vml[c]);
4906			if (error == 0)
4907				spa_history_log_internal(spa, "detach", tx,
4908				    "vdev=%s", vml[c]->vdev_path);
4909			vdev_free(vml[c]);
4910		}
4911	}
4912	vdev_config_dirty(spa->spa_root_vdev);
4913	spa->spa_config_splitting = NULL;
4914	nvlist_free(nvl);
4915	if (error == 0)
4916		dmu_tx_commit(tx);
4917	(void) spa_vdev_exit(spa, NULL, txg, 0);
4918
4919	if (zio_injection_enabled)
4920		zio_handle_panic_injection(spa, FTAG, 3);
4921
4922	/* split is complete; log a history record */
4923	spa_history_log_internal(newspa, "split", NULL,
4924	    "from pool %s", spa_name(spa));
4925
4926	kmem_free(vml, children * sizeof (vdev_t *));
4927
4928	/* if we're not going to mount the filesystems in userland, export */
4929	if (exp)
4930		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4931		    B_FALSE, B_FALSE);
4932
4933	return (error);
4934
4935out:
4936	spa_unload(newspa);
4937	spa_deactivate(newspa);
4938	spa_remove(newspa);
4939
4940	txg = spa_vdev_config_enter(spa);
4941
4942	/* re-online all offlined disks */
4943	for (c = 0; c < children; c++) {
4944		if (vml[c] != NULL)
4945			vml[c]->vdev_offline = B_FALSE;
4946	}
4947	vdev_reopen(spa->spa_root_vdev);
4948
4949	nvlist_free(spa->spa_config_splitting);
4950	spa->spa_config_splitting = NULL;
4951	(void) spa_vdev_exit(spa, NULL, txg, error);
4952
4953	kmem_free(vml, children * sizeof (vdev_t *));
4954	return (error);
4955}
4956
4957static nvlist_t *
4958spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4959{
4960	for (int i = 0; i < count; i++) {
4961		uint64_t guid;
4962
4963		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4964		    &guid) == 0);
4965
4966		if (guid == target_guid)
4967			return (nvpp[i]);
4968	}
4969
4970	return (NULL);
4971}
4972
4973static void
4974spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4975	nvlist_t *dev_to_remove)
4976{
4977	nvlist_t **newdev = NULL;
4978
4979	if (count > 1)
4980		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4981
4982	for (int i = 0, j = 0; i < count; i++) {
4983		if (dev[i] == dev_to_remove)
4984			continue;
4985		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4986	}
4987
4988	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4989	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4990
4991	for (int i = 0; i < count - 1; i++)
4992		nvlist_free(newdev[i]);
4993
4994	if (count > 1)
4995		kmem_free(newdev, (count - 1) * sizeof (void *));
4996}
4997
4998/*
4999 * Evacuate the device.
5000 */
5001static int
5002spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5003{
5004	uint64_t txg;
5005	int error = 0;
5006
5007	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5008	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5009	ASSERT(vd == vd->vdev_top);
5010
5011	/*
5012	 * Evacuate the device.  We don't hold the config lock as writer
5013	 * since we need to do I/O but we do keep the
5014	 * spa_namespace_lock held.  Once this completes the device
5015	 * should no longer have any blocks allocated on it.
5016	 */
5017	if (vd->vdev_islog) {
5018		if (vd->vdev_stat.vs_alloc != 0)
5019			error = spa_offline_log(spa);
5020	} else {
5021		error = ENOTSUP;
5022	}
5023
5024	if (error)
5025		return (error);
5026
5027	/*
5028	 * The evacuation succeeded.  Remove any remaining MOS metadata
5029	 * associated with this vdev, and wait for these changes to sync.
5030	 */
5031	ASSERT0(vd->vdev_stat.vs_alloc);
5032	txg = spa_vdev_config_enter(spa);
5033	vd->vdev_removing = B_TRUE;
5034	vdev_dirty(vd, 0, NULL, txg);
5035	vdev_config_dirty(vd);
5036	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5037
5038	return (0);
5039}
5040
5041/*
5042 * Complete the removal by cleaning up the namespace.
5043 */
5044static void
5045spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5046{
5047	vdev_t *rvd = spa->spa_root_vdev;
5048	uint64_t id = vd->vdev_id;
5049	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5050
5051	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5052	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5053	ASSERT(vd == vd->vdev_top);
5054
5055	/*
5056	 * Only remove any devices which are empty.
5057	 */
5058	if (vd->vdev_stat.vs_alloc != 0)
5059		return;
5060
5061	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5062
5063	if (list_link_active(&vd->vdev_state_dirty_node))
5064		vdev_state_clean(vd);
5065	if (list_link_active(&vd->vdev_config_dirty_node))
5066		vdev_config_clean(vd);
5067
5068	vdev_free(vd);
5069
5070	if (last_vdev) {
5071		vdev_compact_children(rvd);
5072	} else {
5073		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5074		vdev_add_child(rvd, vd);
5075	}
5076	vdev_config_dirty(rvd);
5077
5078	/*
5079	 * Reassess the health of our root vdev.
5080	 */
5081	vdev_reopen(rvd);
5082}
5083
5084/*
5085 * Remove a device from the pool -
5086 *
5087 * Removing a device from the vdev namespace requires several steps
5088 * and can take a significant amount of time.  As a result we use
5089 * the spa_vdev_config_[enter/exit] functions which allow us to
5090 * grab and release the spa_config_lock while still holding the namespace
5091 * lock.  During each step the configuration is synced out.
5092 */
5093
5094/*
5095 * Remove a device from the pool.  Currently, this supports removing only hot
5096 * spares, slogs, and level 2 ARC devices.
5097 */
5098int
5099spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5100{
5101	vdev_t *vd;
5102	metaslab_group_t *mg;
5103	nvlist_t **spares, **l2cache, *nv;
5104	uint64_t txg = 0;
5105	uint_t nspares, nl2cache;
5106	int error = 0;
5107	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5108
5109	ASSERT(spa_writeable(spa));
5110
5111	if (!locked)
5112		txg = spa_vdev_enter(spa);
5113
5114	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5115
5116	if (spa->spa_spares.sav_vdevs != NULL &&
5117	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5118	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5119	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5120		/*
5121		 * Only remove the hot spare if it's not currently in use
5122		 * in this pool.
5123		 */
5124		if (vd == NULL || unspare) {
5125			spa_vdev_remove_aux(spa->spa_spares.sav_config,
5126			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5127			spa_load_spares(spa);
5128			spa->spa_spares.sav_sync = B_TRUE;
5129		} else {
5130			error = EBUSY;
5131		}
5132	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
5133	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5134	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5135	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5136		/*
5137		 * Cache devices can always be removed.
5138		 */
5139		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5140		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5141		spa_load_l2cache(spa);
5142		spa->spa_l2cache.sav_sync = B_TRUE;
5143	} else if (vd != NULL && vd->vdev_islog) {
5144		ASSERT(!locked);
5145		ASSERT(vd == vd->vdev_top);
5146
5147		/*
5148		 * XXX - Once we have bp-rewrite this should
5149		 * become the common case.
5150		 */
5151
5152		mg = vd->vdev_mg;
5153
5154		/*
5155		 * Stop allocating from this vdev.
5156		 */
5157		metaslab_group_passivate(mg);
5158
5159		/*
5160		 * Wait for the youngest allocations and frees to sync,
5161		 * and then wait for the deferral of those frees to finish.
5162		 */
5163		spa_vdev_config_exit(spa, NULL,
5164		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5165
5166		/*
5167		 * Attempt to evacuate the vdev.
5168		 */
5169		error = spa_vdev_remove_evacuate(spa, vd);
5170
5171		txg = spa_vdev_config_enter(spa);
5172
5173		/*
5174		 * If we couldn't evacuate the vdev, unwind.
5175		 */
5176		if (error) {
5177			metaslab_group_activate(mg);
5178			return (spa_vdev_exit(spa, NULL, txg, error));
5179		}
5180
5181		/*
5182		 * Clean up the vdev namespace.
5183		 */
5184		spa_vdev_remove_from_namespace(spa, vd);
5185
5186	} else if (vd != NULL) {
5187		/*
5188		 * Normal vdevs cannot be removed (yet).
5189		 */
5190		error = ENOTSUP;
5191	} else {
5192		/*
5193		 * There is no vdev of any kind with the specified guid.
5194		 */
5195		error = ENOENT;
5196	}
5197
5198	if (!locked)
5199		return (spa_vdev_exit(spa, NULL, txg, error));
5200
5201	return (error);
5202}
5203
5204/*
5205 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5206 * current spared, so we can detach it.
5207 */
5208static vdev_t *
5209spa_vdev_resilver_done_hunt(vdev_t *vd)
5210{
5211	vdev_t *newvd, *oldvd;
5212
5213	for (int c = 0; c < vd->vdev_children; c++) {
5214		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5215		if (oldvd != NULL)
5216			return (oldvd);
5217	}
5218
5219	/*
5220	 * Check for a completed replacement.  We always consider the first
5221	 * vdev in the list to be the oldest vdev, and the last one to be
5222	 * the newest (see spa_vdev_attach() for how that works).  In
5223	 * the case where the newest vdev is faulted, we will not automatically
5224	 * remove it after a resilver completes.  This is OK as it will require
5225	 * user intervention to determine which disk the admin wishes to keep.
5226	 */
5227	if (vd->vdev_ops == &vdev_replacing_ops) {
5228		ASSERT(vd->vdev_children > 1);
5229
5230		newvd = vd->vdev_child[vd->vdev_children - 1];
5231		oldvd = vd->vdev_child[0];
5232
5233		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5234		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5235		    !vdev_dtl_required(oldvd))
5236			return (oldvd);
5237	}
5238
5239	/*
5240	 * Check for a completed resilver with the 'unspare' flag set.
5241	 */
5242	if (vd->vdev_ops == &vdev_spare_ops) {
5243		vdev_t *first = vd->vdev_child[0];
5244		vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5245
5246		if (last->vdev_unspare) {
5247			oldvd = first;
5248			newvd = last;
5249		} else if (first->vdev_unspare) {
5250			oldvd = last;
5251			newvd = first;
5252		} else {
5253			oldvd = NULL;
5254		}
5255
5256		if (oldvd != NULL &&
5257		    vdev_dtl_empty(newvd, DTL_MISSING) &&
5258		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5259		    !vdev_dtl_required(oldvd))
5260			return (oldvd);
5261
5262		/*
5263		 * If there are more than two spares attached to a disk,
5264		 * and those spares are not required, then we want to
5265		 * attempt to free them up now so that they can be used
5266		 * by other pools.  Once we're back down to a single
5267		 * disk+spare, we stop removing them.
5268		 */
5269		if (vd->vdev_children > 2) {
5270			newvd = vd->vdev_child[1];
5271
5272			if (newvd->vdev_isspare && last->vdev_isspare &&
5273			    vdev_dtl_empty(last, DTL_MISSING) &&
5274			    vdev_dtl_empty(last, DTL_OUTAGE) &&
5275			    !vdev_dtl_required(newvd))
5276				return (newvd);
5277		}
5278	}
5279
5280	return (NULL);
5281}
5282
5283static void
5284spa_vdev_resilver_done(spa_t *spa)
5285{
5286	vdev_t *vd, *pvd, *ppvd;
5287	uint64_t guid, sguid, pguid, ppguid;
5288
5289	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5290
5291	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5292		pvd = vd->vdev_parent;
5293		ppvd = pvd->vdev_parent;
5294		guid = vd->vdev_guid;
5295		pguid = pvd->vdev_guid;
5296		ppguid = ppvd->vdev_guid;
5297		sguid = 0;
5298		/*
5299		 * If we have just finished replacing a hot spared device, then
5300		 * we need to detach the parent's first child (the original hot
5301		 * spare) as well.
5302		 */
5303		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5304		    ppvd->vdev_children == 2) {
5305			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5306			sguid = ppvd->vdev_child[1]->vdev_guid;
5307		}
5308		spa_config_exit(spa, SCL_ALL, FTAG);
5309		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5310			return;
5311		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5312			return;
5313		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5314	}
5315
5316	spa_config_exit(spa, SCL_ALL, FTAG);
5317}
5318
5319/*
5320 * Update the stored path or FRU for this vdev.
5321 */
5322int
5323spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5324    boolean_t ispath)
5325{
5326	vdev_t *vd;
5327	boolean_t sync = B_FALSE;
5328
5329	ASSERT(spa_writeable(spa));
5330
5331	spa_vdev_state_enter(spa, SCL_ALL);
5332
5333	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5334		return (spa_vdev_state_exit(spa, NULL, ENOENT));
5335
5336	if (!vd->vdev_ops->vdev_op_leaf)
5337		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5338
5339	if (ispath) {
5340		if (strcmp(value, vd->vdev_path) != 0) {
5341			spa_strfree(vd->vdev_path);
5342			vd->vdev_path = spa_strdup(value);
5343			sync = B_TRUE;
5344		}
5345	} else {
5346		if (vd->vdev_fru == NULL) {
5347			vd->vdev_fru = spa_strdup(value);
5348			sync = B_TRUE;
5349		} else if (strcmp(value, vd->vdev_fru) != 0) {
5350			spa_strfree(vd->vdev_fru);
5351			vd->vdev_fru = spa_strdup(value);
5352			sync = B_TRUE;
5353		}
5354	}
5355
5356	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5357}
5358
5359int
5360spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5361{
5362	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5363}
5364
5365int
5366spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5367{
5368	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5369}
5370
5371/*
5372 * ==========================================================================
5373 * SPA Scanning
5374 * ==========================================================================
5375 */
5376
5377int
5378spa_scan_stop(spa_t *spa)
5379{
5380	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5381	if (dsl_scan_resilvering(spa->spa_dsl_pool))
5382		return (EBUSY);
5383	return (dsl_scan_cancel(spa->spa_dsl_pool));
5384}
5385
5386int
5387spa_scan(spa_t *spa, pool_scan_func_t func)
5388{
5389	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5390
5391	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5392		return (ENOTSUP);
5393
5394	/*
5395	 * If a resilver was requested, but there is no DTL on a
5396	 * writeable leaf device, we have nothing to do.
5397	 */
5398	if (func == POOL_SCAN_RESILVER &&
5399	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5400		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5401		return (0);
5402	}
5403
5404	return (dsl_scan(spa->spa_dsl_pool, func));
5405}
5406
5407/*
5408 * ==========================================================================
5409 * SPA async task processing
5410 * ==========================================================================
5411 */
5412
5413static void
5414spa_async_remove(spa_t *spa, vdev_t *vd)
5415{
5416	if (vd->vdev_remove_wanted) {
5417		vd->vdev_remove_wanted = B_FALSE;
5418		vd->vdev_delayed_close = B_FALSE;
5419		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5420
5421		/*
5422		 * We want to clear the stats, but we don't want to do a full
5423		 * vdev_clear() as that will cause us to throw away
5424		 * degraded/faulted state as well as attempt to reopen the
5425		 * device, all of which is a waste.
5426		 */
5427		vd->vdev_stat.vs_read_errors = 0;
5428		vd->vdev_stat.vs_write_errors = 0;
5429		vd->vdev_stat.vs_checksum_errors = 0;
5430
5431		vdev_state_dirty(vd->vdev_top);
5432	}
5433
5434	for (int c = 0; c < vd->vdev_children; c++)
5435		spa_async_remove(spa, vd->vdev_child[c]);
5436}
5437
5438static void
5439spa_async_probe(spa_t *spa, vdev_t *vd)
5440{
5441	if (vd->vdev_probe_wanted) {
5442		vd->vdev_probe_wanted = B_FALSE;
5443		vdev_reopen(vd);	/* vdev_open() does the actual probe */
5444	}
5445
5446	for (int c = 0; c < vd->vdev_children; c++)
5447		spa_async_probe(spa, vd->vdev_child[c]);
5448}
5449
5450static void
5451spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5452{
5453	sysevent_id_t eid;
5454	nvlist_t *attr;
5455	char *physpath;
5456
5457	if (!spa->spa_autoexpand)
5458		return;
5459
5460	for (int c = 0; c < vd->vdev_children; c++) {
5461		vdev_t *cvd = vd->vdev_child[c];
5462		spa_async_autoexpand(spa, cvd);
5463	}
5464
5465	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5466		return;
5467
5468	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5469	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5470
5471	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5472	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5473
5474	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5475	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5476
5477	nvlist_free(attr);
5478	kmem_free(physpath, MAXPATHLEN);
5479}
5480
5481static void
5482spa_async_thread(spa_t *spa)
5483{
5484	int tasks;
5485
5486	ASSERT(spa->spa_sync_on);
5487
5488	mutex_enter(&spa->spa_async_lock);
5489	tasks = spa->spa_async_tasks;
5490	spa->spa_async_tasks = 0;
5491	mutex_exit(&spa->spa_async_lock);
5492
5493	/*
5494	 * See if the config needs to be updated.
5495	 */
5496	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5497		uint64_t old_space, new_space;
5498
5499		mutex_enter(&spa_namespace_lock);
5500		old_space = metaslab_class_get_space(spa_normal_class(spa));
5501		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5502		new_space = metaslab_class_get_space(spa_normal_class(spa));
5503		mutex_exit(&spa_namespace_lock);
5504
5505		/*
5506		 * If the pool grew as a result of the config update,
5507		 * then log an internal history event.
5508		 */
5509		if (new_space != old_space) {
5510			spa_history_log_internal(spa, "vdev online", NULL,
5511			    "pool '%s' size: %llu(+%llu)",
5512			    spa_name(spa), new_space, new_space - old_space);
5513		}
5514	}
5515
5516	/*
5517	 * See if any devices need to be marked REMOVED.
5518	 */
5519	if (tasks & SPA_ASYNC_REMOVE) {
5520		spa_vdev_state_enter(spa, SCL_NONE);
5521		spa_async_remove(spa, spa->spa_root_vdev);
5522		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5523			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5524		for (int i = 0; i < spa->spa_spares.sav_count; i++)
5525			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5526		(void) spa_vdev_state_exit(spa, NULL, 0);
5527	}
5528
5529	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5530		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5531		spa_async_autoexpand(spa, spa->spa_root_vdev);
5532		spa_config_exit(spa, SCL_CONFIG, FTAG);
5533	}
5534
5535	/*
5536	 * See if any devices need to be probed.
5537	 */
5538	if (tasks & SPA_ASYNC_PROBE) {
5539		spa_vdev_state_enter(spa, SCL_NONE);
5540		spa_async_probe(spa, spa->spa_root_vdev);
5541		(void) spa_vdev_state_exit(spa, NULL, 0);
5542	}
5543
5544	/*
5545	 * If any devices are done replacing, detach them.
5546	 */
5547	if (tasks & SPA_ASYNC_RESILVER_DONE)
5548		spa_vdev_resilver_done(spa);
5549
5550	/*
5551	 * Kick off a resilver.
5552	 */
5553	if (tasks & SPA_ASYNC_RESILVER)
5554		dsl_resilver_restart(spa->spa_dsl_pool, 0);
5555
5556	/*
5557	 * Let the world know that we're done.
5558	 */
5559	mutex_enter(&spa->spa_async_lock);
5560	spa->spa_async_thread = NULL;
5561	cv_broadcast(&spa->spa_async_cv);
5562	mutex_exit(&spa->spa_async_lock);
5563	thread_exit();
5564}
5565
5566void
5567spa_async_suspend(spa_t *spa)
5568{
5569	mutex_enter(&spa->spa_async_lock);
5570	spa->spa_async_suspended++;
5571	while (spa->spa_async_thread != NULL)
5572		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5573	mutex_exit(&spa->spa_async_lock);
5574}
5575
5576void
5577spa_async_resume(spa_t *spa)
5578{
5579	mutex_enter(&spa->spa_async_lock);
5580	ASSERT(spa->spa_async_suspended != 0);
5581	spa->spa_async_suspended--;
5582	mutex_exit(&spa->spa_async_lock);
5583}
5584
5585static void
5586spa_async_dispatch(spa_t *spa)
5587{
5588	mutex_enter(&spa->spa_async_lock);
5589	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5590	    spa->spa_async_thread == NULL &&
5591	    rootdir != NULL && !vn_is_readonly(rootdir))
5592		spa->spa_async_thread = thread_create(NULL, 0,
5593		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5594	mutex_exit(&spa->spa_async_lock);
5595}
5596
5597void
5598spa_async_request(spa_t *spa, int task)
5599{
5600	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5601	mutex_enter(&spa->spa_async_lock);
5602	spa->spa_async_tasks |= task;
5603	mutex_exit(&spa->spa_async_lock);
5604}
5605
5606/*
5607 * ==========================================================================
5608 * SPA syncing routines
5609 * ==========================================================================
5610 */
5611
5612static int
5613bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5614{
5615	bpobj_t *bpo = arg;
5616	bpobj_enqueue(bpo, bp, tx);
5617	return (0);
5618}
5619
5620static int
5621spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5622{
5623	zio_t *zio = arg;
5624
5625	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5626	    zio->io_flags));
5627	return (0);
5628}
5629
5630static void
5631spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5632{
5633	char *packed = NULL;
5634	size_t bufsize;
5635	size_t nvsize = 0;
5636	dmu_buf_t *db;
5637
5638	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5639
5640	/*
5641	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5642	 * information.  This avoids the dbuf_will_dirty() path and
5643	 * saves us a pre-read to get data we don't actually care about.
5644	 */
5645	bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5646	packed = kmem_alloc(bufsize, KM_SLEEP);
5647
5648	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5649	    KM_SLEEP) == 0);
5650	bzero(packed + nvsize, bufsize - nvsize);
5651
5652	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5653
5654	kmem_free(packed, bufsize);
5655
5656	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5657	dmu_buf_will_dirty(db, tx);
5658	*(uint64_t *)db->db_data = nvsize;
5659	dmu_buf_rele(db, FTAG);
5660}
5661
5662static void
5663spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5664    const char *config, const char *entry)
5665{
5666	nvlist_t *nvroot;
5667	nvlist_t **list;
5668	int i;
5669
5670	if (!sav->sav_sync)
5671		return;
5672
5673	/*
5674	 * Update the MOS nvlist describing the list of available devices.
5675	 * spa_validate_aux() will have already made sure this nvlist is
5676	 * valid and the vdevs are labeled appropriately.
5677	 */
5678	if (sav->sav_object == 0) {
5679		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5680		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5681		    sizeof (uint64_t), tx);
5682		VERIFY(zap_update(spa->spa_meta_objset,
5683		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5684		    &sav->sav_object, tx) == 0);
5685	}
5686
5687	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5688	if (sav->sav_count == 0) {
5689		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5690	} else {
5691		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5692		for (i = 0; i < sav->sav_count; i++)
5693			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5694			    B_FALSE, VDEV_CONFIG_L2CACHE);
5695		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5696		    sav->sav_count) == 0);
5697		for (i = 0; i < sav->sav_count; i++)
5698			nvlist_free(list[i]);
5699		kmem_free(list, sav->sav_count * sizeof (void *));
5700	}
5701
5702	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5703	nvlist_free(nvroot);
5704
5705	sav->sav_sync = B_FALSE;
5706}
5707
5708static void
5709spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5710{
5711	nvlist_t *config;
5712
5713	if (list_is_empty(&spa->spa_config_dirty_list))
5714		return;
5715
5716	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5717
5718	config = spa_config_generate(spa, spa->spa_root_vdev,
5719	    dmu_tx_get_txg(tx), B_FALSE);
5720
5721	/*
5722	 * If we're upgrading the spa version then make sure that
5723	 * the config object gets updated with the correct version.
5724	 */
5725	if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5726		fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5727		    spa->spa_uberblock.ub_version);
5728
5729	spa_config_exit(spa, SCL_STATE, FTAG);
5730
5731	if (spa->spa_config_syncing)
5732		nvlist_free(spa->spa_config_syncing);
5733	spa->spa_config_syncing = config;
5734
5735	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5736}
5737
5738static void
5739spa_sync_version(void *arg1, void *arg2, dmu_tx_t *tx)
5740{
5741	spa_t *spa = arg1;
5742	uint64_t version = *(uint64_t *)arg2;
5743
5744	/*
5745	 * Setting the version is special cased when first creating the pool.
5746	 */
5747	ASSERT(tx->tx_txg != TXG_INITIAL);
5748
5749	ASSERT(version <= SPA_VERSION);
5750	ASSERT(version >= spa_version(spa));
5751
5752	spa->spa_uberblock.ub_version = version;
5753	vdev_config_dirty(spa->spa_root_vdev);
5754	spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5755}
5756
5757/*
5758 * Set zpool properties.
5759 */
5760static void
5761spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5762{
5763	spa_t *spa = arg1;
5764	objset_t *mos = spa->spa_meta_objset;
5765	nvlist_t *nvp = arg2;
5766	nvpair_t *elem = NULL;
5767
5768	mutex_enter(&spa->spa_props_lock);
5769
5770	while ((elem = nvlist_next_nvpair(nvp, elem))) {
5771		uint64_t intval;
5772		char *strval, *fname;
5773		zpool_prop_t prop;
5774		const char *propname;
5775		zprop_type_t proptype;
5776		zfeature_info_t *feature;
5777
5778		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5779		case ZPROP_INVAL:
5780			/*
5781			 * We checked this earlier in spa_prop_validate().
5782			 */
5783			ASSERT(zpool_prop_feature(nvpair_name(elem)));
5784
5785			fname = strchr(nvpair_name(elem), '@') + 1;
5786			VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
5787
5788			spa_feature_enable(spa, feature, tx);
5789			spa_history_log_internal(spa, "set", tx,
5790			    "%s=enabled", nvpair_name(elem));
5791			break;
5792
5793		case ZPOOL_PROP_VERSION:
5794			VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5795			/*
5796			 * The version is synced seperatly before other
5797			 * properties and should be correct by now.
5798			 */
5799			ASSERT3U(spa_version(spa), >=, intval);
5800			break;
5801
5802		case ZPOOL_PROP_ALTROOT:
5803			/*
5804			 * 'altroot' is a non-persistent property. It should
5805			 * have been set temporarily at creation or import time.
5806			 */
5807			ASSERT(spa->spa_root != NULL);
5808			break;
5809
5810		case ZPOOL_PROP_READONLY:
5811		case ZPOOL_PROP_CACHEFILE:
5812			/*
5813			 * 'readonly' and 'cachefile' are also non-persisitent
5814			 * properties.
5815			 */
5816			break;
5817		case ZPOOL_PROP_COMMENT:
5818			VERIFY(nvpair_value_string(elem, &strval) == 0);
5819			if (spa->spa_comment != NULL)
5820				spa_strfree(spa->spa_comment);
5821			spa->spa_comment = spa_strdup(strval);
5822			/*
5823			 * We need to dirty the configuration on all the vdevs
5824			 * so that their labels get updated.  It's unnecessary
5825			 * to do this for pool creation since the vdev's
5826			 * configuratoin has already been dirtied.
5827			 */
5828			if (tx->tx_txg != TXG_INITIAL)
5829				vdev_config_dirty(spa->spa_root_vdev);
5830			spa_history_log_internal(spa, "set", tx,
5831			    "%s=%s", nvpair_name(elem), strval);
5832			break;
5833		default:
5834			/*
5835			 * Set pool property values in the poolprops mos object.
5836			 */
5837			if (spa->spa_pool_props_object == 0) {
5838				spa->spa_pool_props_object =
5839				    zap_create_link(mos, DMU_OT_POOL_PROPS,
5840				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5841				    tx);
5842			}
5843
5844			/* normalize the property name */
5845			propname = zpool_prop_to_name(prop);
5846			proptype = zpool_prop_get_type(prop);
5847
5848			if (nvpair_type(elem) == DATA_TYPE_STRING) {
5849				ASSERT(proptype == PROP_TYPE_STRING);
5850				VERIFY(nvpair_value_string(elem, &strval) == 0);
5851				VERIFY(zap_update(mos,
5852				    spa->spa_pool_props_object, propname,
5853				    1, strlen(strval) + 1, strval, tx) == 0);
5854				spa_history_log_internal(spa, "set", tx,
5855				    "%s=%s", nvpair_name(elem), strval);
5856			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5857				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5858
5859				if (proptype == PROP_TYPE_INDEX) {
5860					const char *unused;
5861					VERIFY(zpool_prop_index_to_string(
5862					    prop, intval, &unused) == 0);
5863				}
5864				VERIFY(zap_update(mos,
5865				    spa->spa_pool_props_object, propname,
5866				    8, 1, &intval, tx) == 0);
5867				spa_history_log_internal(spa, "set", tx,
5868				    "%s=%lld", nvpair_name(elem), intval);
5869			} else {
5870				ASSERT(0); /* not allowed */
5871			}
5872
5873			switch (prop) {
5874			case ZPOOL_PROP_DELEGATION:
5875				spa->spa_delegation = intval;
5876				break;
5877			case ZPOOL_PROP_BOOTFS:
5878				spa->spa_bootfs = intval;
5879				break;
5880			case ZPOOL_PROP_FAILUREMODE:
5881				spa->spa_failmode = intval;
5882				break;
5883			case ZPOOL_PROP_AUTOEXPAND:
5884				spa->spa_autoexpand = intval;
5885				if (tx->tx_txg != TXG_INITIAL)
5886					spa_async_request(spa,
5887					    SPA_ASYNC_AUTOEXPAND);
5888				break;
5889			case ZPOOL_PROP_DEDUPDITTO:
5890				spa->spa_dedup_ditto = intval;
5891				break;
5892			default:
5893				break;
5894			}
5895		}
5896
5897	}
5898
5899	mutex_exit(&spa->spa_props_lock);
5900}
5901
5902/*
5903 * Perform one-time upgrade on-disk changes.  spa_version() does not
5904 * reflect the new version this txg, so there must be no changes this
5905 * txg to anything that the upgrade code depends on after it executes.
5906 * Therefore this must be called after dsl_pool_sync() does the sync
5907 * tasks.
5908 */
5909static void
5910spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5911{
5912	dsl_pool_t *dp = spa->spa_dsl_pool;
5913
5914	ASSERT(spa->spa_sync_pass == 1);
5915
5916	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5917	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5918		dsl_pool_create_origin(dp, tx);
5919
5920		/* Keeping the origin open increases spa_minref */
5921		spa->spa_minref += 3;
5922	}
5923
5924	if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5925	    spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5926		dsl_pool_upgrade_clones(dp, tx);
5927	}
5928
5929	if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5930	    spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5931		dsl_pool_upgrade_dir_clones(dp, tx);
5932
5933		/* Keeping the freedir open increases spa_minref */
5934		spa->spa_minref += 3;
5935	}
5936
5937	if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
5938	    spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
5939		spa_feature_create_zap_objects(spa, tx);
5940	}
5941}
5942
5943/*
5944 * Sync the specified transaction group.  New blocks may be dirtied as
5945 * part of the process, so we iterate until it converges.
5946 */
5947void
5948spa_sync(spa_t *spa, uint64_t txg)
5949{
5950	dsl_pool_t *dp = spa->spa_dsl_pool;
5951	objset_t *mos = spa->spa_meta_objset;
5952	bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5953	bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5954	vdev_t *rvd = spa->spa_root_vdev;
5955	vdev_t *vd;
5956	dmu_tx_t *tx;
5957	int error;
5958
5959	VERIFY(spa_writeable(spa));
5960
5961	/*
5962	 * Lock out configuration changes.
5963	 */
5964	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5965
5966	spa->spa_syncing_txg = txg;
5967	spa->spa_sync_pass = 0;
5968
5969	/*
5970	 * If there are any pending vdev state changes, convert them
5971	 * into config changes that go out with this transaction group.
5972	 */
5973	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5974	while (list_head(&spa->spa_state_dirty_list) != NULL) {
5975		/*
5976		 * We need the write lock here because, for aux vdevs,
5977		 * calling vdev_config_dirty() modifies sav_config.
5978		 * This is ugly and will become unnecessary when we
5979		 * eliminate the aux vdev wart by integrating all vdevs
5980		 * into the root vdev tree.
5981		 */
5982		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5983		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5984		while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5985			vdev_state_clean(vd);
5986			vdev_config_dirty(vd);
5987		}
5988		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5989		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5990	}
5991	spa_config_exit(spa, SCL_STATE, FTAG);
5992
5993	tx = dmu_tx_create_assigned(dp, txg);
5994
5995	spa->spa_sync_starttime = gethrtime();
5996	VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
5997	    spa->spa_sync_starttime + spa->spa_deadman_synctime));
5998
5999	/*
6000	 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6001	 * set spa_deflate if we have no raid-z vdevs.
6002	 */
6003	if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6004	    spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6005		int i;
6006
6007		for (i = 0; i < rvd->vdev_children; i++) {
6008			vd = rvd->vdev_child[i];
6009			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6010				break;
6011		}
6012		if (i == rvd->vdev_children) {
6013			spa->spa_deflate = TRUE;
6014			VERIFY(0 == zap_add(spa->spa_meta_objset,
6015			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6016			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6017		}
6018	}
6019
6020	/*
6021	 * If anything has changed in this txg, or if someone is waiting
6022	 * for this txg to sync (eg, spa_vdev_remove()), push the
6023	 * deferred frees from the previous txg.  If not, leave them
6024	 * alone so that we don't generate work on an otherwise idle
6025	 * system.
6026	 */
6027	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6028	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6029	    !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6030	    ((dsl_scan_active(dp->dp_scan) ||
6031	    txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6032		zio_t *zio = zio_root(spa, NULL, NULL, 0);
6033		VERIFY3U(bpobj_iterate(defer_bpo,
6034		    spa_free_sync_cb, zio, tx), ==, 0);
6035		VERIFY0(zio_wait(zio));
6036	}
6037
6038	/*
6039	 * Iterate to convergence.
6040	 */
6041	do {
6042		int pass = ++spa->spa_sync_pass;
6043
6044		spa_sync_config_object(spa, tx);
6045		spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6046		    ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6047		spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6048		    ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6049		spa_errlog_sync(spa, txg);
6050		dsl_pool_sync(dp, txg);
6051
6052		if (pass < zfs_sync_pass_deferred_free) {
6053			zio_t *zio = zio_root(spa, NULL, NULL, 0);
6054			bplist_iterate(free_bpl, spa_free_sync_cb,
6055			    zio, tx);
6056			VERIFY(zio_wait(zio) == 0);
6057		} else {
6058			bplist_iterate(free_bpl, bpobj_enqueue_cb,
6059