spa.c revision d5285cae913f4e01ffa0e6693a6d8ef1fbea30ba
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 = 0;
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		newvdevs = NULL;
1346	}
1347
1348	oldvdevs = sav->sav_vdevs;
1349	oldnvdevs = sav->sav_count;
1350	sav->sav_vdevs = NULL;
1351	sav->sav_count = 0;
1352
1353	/*
1354	 * Process new nvlist of vdevs.
1355	 */
1356	for (i = 0; i < nl2cache; i++) {
1357		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1358		    &guid) == 0);
1359
1360		newvdevs[i] = NULL;
1361		for (j = 0; j < oldnvdevs; j++) {
1362			vd = oldvdevs[j];
1363			if (vd != NULL && guid == vd->vdev_guid) {
1364				/*
1365				 * Retain previous vdev for add/remove ops.
1366				 */
1367				newvdevs[i] = vd;
1368				oldvdevs[j] = NULL;
1369				break;
1370			}
1371		}
1372
1373		if (newvdevs[i] == NULL) {
1374			/*
1375			 * Create new vdev
1376			 */
1377			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1378			    VDEV_ALLOC_L2CACHE) == 0);
1379			ASSERT(vd != NULL);
1380			newvdevs[i] = vd;
1381
1382			/*
1383			 * Commit this vdev as an l2cache device,
1384			 * even if it fails to open.
1385			 */
1386			spa_l2cache_add(vd);
1387
1388			vd->vdev_top = vd;
1389			vd->vdev_aux = sav;
1390
1391			spa_l2cache_activate(vd);
1392
1393			if (vdev_open(vd) != 0)
1394				continue;
1395
1396			(void) vdev_validate_aux(vd);
1397
1398			if (!vdev_is_dead(vd))
1399				l2arc_add_vdev(spa, vd);
1400		}
1401	}
1402
1403	/*
1404	 * Purge vdevs that were dropped
1405	 */
1406	for (i = 0; i < oldnvdevs; i++) {
1407		uint64_t pool;
1408
1409		vd = oldvdevs[i];
1410		if (vd != NULL) {
1411			ASSERT(vd->vdev_isl2cache);
1412
1413			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1414			    pool != 0ULL && l2arc_vdev_present(vd))
1415				l2arc_remove_vdev(vd);
1416			vdev_clear_stats(vd);
1417			vdev_free(vd);
1418		}
1419	}
1420
1421	if (oldvdevs)
1422		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1423
1424	if (sav->sav_config == NULL)
1425		goto out;
1426
1427	sav->sav_vdevs = newvdevs;
1428	sav->sav_count = (int)nl2cache;
1429
1430	/*
1431	 * Recompute the stashed list of l2cache devices, with status
1432	 * information this time.
1433	 */
1434	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1435	    DATA_TYPE_NVLIST_ARRAY) == 0);
1436
1437	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1438	for (i = 0; i < sav->sav_count; i++)
1439		l2cache[i] = vdev_config_generate(spa,
1440		    sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1441	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1442	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1443out:
1444	for (i = 0; i < sav->sav_count; i++)
1445		nvlist_free(l2cache[i]);
1446	if (sav->sav_count)
1447		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1448}
1449
1450static int
1451load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1452{
1453	dmu_buf_t *db;
1454	char *packed = NULL;
1455	size_t nvsize = 0;
1456	int error;
1457	*value = NULL;
1458
1459	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1460	nvsize = *(uint64_t *)db->db_data;
1461	dmu_buf_rele(db, FTAG);
1462
1463	packed = kmem_alloc(nvsize, KM_SLEEP);
1464	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1465	    DMU_READ_PREFETCH);
1466	if (error == 0)
1467		error = nvlist_unpack(packed, nvsize, value, 0);
1468	kmem_free(packed, nvsize);
1469
1470	return (error);
1471}
1472
1473/*
1474 * Checks to see if the given vdev could not be opened, in which case we post a
1475 * sysevent to notify the autoreplace code that the device has been removed.
1476 */
1477static void
1478spa_check_removed(vdev_t *vd)
1479{
1480	for (int c = 0; c < vd->vdev_children; c++)
1481		spa_check_removed(vd->vdev_child[c]);
1482
1483	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1484		zfs_post_autoreplace(vd->vdev_spa, vd);
1485		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1486	}
1487}
1488
1489/*
1490 * Validate the current config against the MOS config
1491 */
1492static boolean_t
1493spa_config_valid(spa_t *spa, nvlist_t *config)
1494{
1495	vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1496	nvlist_t *nv;
1497
1498	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1499
1500	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1501	VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1502
1503	ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1504
1505	/*
1506	 * If we're doing a normal import, then build up any additional
1507	 * diagnostic information about missing devices in this config.
1508	 * We'll pass this up to the user for further processing.
1509	 */
1510	if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1511		nvlist_t **child, *nv;
1512		uint64_t idx = 0;
1513
1514		child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1515		    KM_SLEEP);
1516		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1517
1518		for (int c = 0; c < rvd->vdev_children; c++) {
1519			vdev_t *tvd = rvd->vdev_child[c];
1520			vdev_t *mtvd  = mrvd->vdev_child[c];
1521
1522			if (tvd->vdev_ops == &vdev_missing_ops &&
1523			    mtvd->vdev_ops != &vdev_missing_ops &&
1524			    mtvd->vdev_islog)
1525				child[idx++] = vdev_config_generate(spa, mtvd,
1526				    B_FALSE, 0);
1527		}
1528
1529		if (idx) {
1530			VERIFY(nvlist_add_nvlist_array(nv,
1531			    ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1532			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1533			    ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1534
1535			for (int i = 0; i < idx; i++)
1536				nvlist_free(child[i]);
1537		}
1538		nvlist_free(nv);
1539		kmem_free(child, rvd->vdev_children * sizeof (char **));
1540	}
1541
1542	/*
1543	 * Compare the root vdev tree with the information we have
1544	 * from the MOS config (mrvd). Check each top-level vdev
1545	 * with the corresponding MOS config top-level (mtvd).
1546	 */
1547	for (int c = 0; c < rvd->vdev_children; c++) {
1548		vdev_t *tvd = rvd->vdev_child[c];
1549		vdev_t *mtvd  = mrvd->vdev_child[c];
1550
1551		/*
1552		 * Resolve any "missing" vdevs in the current configuration.
1553		 * If we find that the MOS config has more accurate information
1554		 * about the top-level vdev then use that vdev instead.
1555		 */
1556		if (tvd->vdev_ops == &vdev_missing_ops &&
1557		    mtvd->vdev_ops != &vdev_missing_ops) {
1558
1559			if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1560				continue;
1561
1562			/*
1563			 * Device specific actions.
1564			 */
1565			if (mtvd->vdev_islog) {
1566				spa_set_log_state(spa, SPA_LOG_CLEAR);
1567			} else {
1568				/*
1569				 * XXX - once we have 'readonly' pool
1570				 * support we should be able to handle
1571				 * missing data devices by transitioning
1572				 * the pool to readonly.
1573				 */
1574				continue;
1575			}
1576
1577			/*
1578			 * Swap the missing vdev with the data we were
1579			 * able to obtain from the MOS config.
1580			 */
1581			vdev_remove_child(rvd, tvd);
1582			vdev_remove_child(mrvd, mtvd);
1583
1584			vdev_add_child(rvd, mtvd);
1585			vdev_add_child(mrvd, tvd);
1586
1587			spa_config_exit(spa, SCL_ALL, FTAG);
1588			vdev_load(mtvd);
1589			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1590
1591			vdev_reopen(rvd);
1592		} else if (mtvd->vdev_islog) {
1593			/*
1594			 * Load the slog device's state from the MOS config
1595			 * since it's possible that the label does not
1596			 * contain the most up-to-date information.
1597			 */
1598			vdev_load_log_state(tvd, mtvd);
1599			vdev_reopen(tvd);
1600		}
1601	}
1602	vdev_free(mrvd);
1603	spa_config_exit(spa, SCL_ALL, FTAG);
1604
1605	/*
1606	 * Ensure we were able to validate the config.
1607	 */
1608	return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1609}
1610
1611/*
1612 * Check for missing log devices
1613 */
1614static int
1615spa_check_logs(spa_t *spa)
1616{
1617	switch (spa->spa_log_state) {
1618	case SPA_LOG_MISSING:
1619		/* need to recheck in case slog has been restored */
1620	case SPA_LOG_UNKNOWN:
1621		if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1622		    DS_FIND_CHILDREN)) {
1623			spa_set_log_state(spa, SPA_LOG_MISSING);
1624			return (1);
1625		}
1626		break;
1627	}
1628	return (0);
1629}
1630
1631static boolean_t
1632spa_passivate_log(spa_t *spa)
1633{
1634	vdev_t *rvd = spa->spa_root_vdev;
1635	boolean_t slog_found = B_FALSE;
1636
1637	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1638
1639	if (!spa_has_slogs(spa))
1640		return (B_FALSE);
1641
1642	for (int c = 0; c < rvd->vdev_children; c++) {
1643		vdev_t *tvd = rvd->vdev_child[c];
1644		metaslab_group_t *mg = tvd->vdev_mg;
1645
1646		if (tvd->vdev_islog) {
1647			metaslab_group_passivate(mg);
1648			slog_found = B_TRUE;
1649		}
1650	}
1651
1652	return (slog_found);
1653}
1654
1655static void
1656spa_activate_log(spa_t *spa)
1657{
1658	vdev_t *rvd = spa->spa_root_vdev;
1659
1660	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1661
1662	for (int c = 0; c < rvd->vdev_children; c++) {
1663		vdev_t *tvd = rvd->vdev_child[c];
1664		metaslab_group_t *mg = tvd->vdev_mg;
1665
1666		if (tvd->vdev_islog)
1667			metaslab_group_activate(mg);
1668	}
1669}
1670
1671int
1672spa_offline_log(spa_t *spa)
1673{
1674	int error = 0;
1675
1676	if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1677	    NULL, DS_FIND_CHILDREN)) == 0) {
1678
1679		/*
1680		 * We successfully offlined the log device, sync out the
1681		 * current txg so that the "stubby" block can be removed
1682		 * by zil_sync().
1683		 */
1684		txg_wait_synced(spa->spa_dsl_pool, 0);
1685	}
1686	return (error);
1687}
1688
1689static void
1690spa_aux_check_removed(spa_aux_vdev_t *sav)
1691{
1692	for (int i = 0; i < sav->sav_count; i++)
1693		spa_check_removed(sav->sav_vdevs[i]);
1694}
1695
1696void
1697spa_claim_notify(zio_t *zio)
1698{
1699	spa_t *spa = zio->io_spa;
1700
1701	if (zio->io_error)
1702		return;
1703
1704	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1705	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1706		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1707	mutex_exit(&spa->spa_props_lock);
1708}
1709
1710typedef struct spa_load_error {
1711	uint64_t	sle_meta_count;
1712	uint64_t	sle_data_count;
1713} spa_load_error_t;
1714
1715static void
1716spa_load_verify_done(zio_t *zio)
1717{
1718	blkptr_t *bp = zio->io_bp;
1719	spa_load_error_t *sle = zio->io_private;
1720	dmu_object_type_t type = BP_GET_TYPE(bp);
1721	int error = zio->io_error;
1722
1723	if (error) {
1724		if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1725		    type != DMU_OT_INTENT_LOG)
1726			atomic_add_64(&sle->sle_meta_count, 1);
1727		else
1728			atomic_add_64(&sle->sle_data_count, 1);
1729	}
1730	zio_data_buf_free(zio->io_data, zio->io_size);
1731}
1732
1733/*ARGSUSED*/
1734static int
1735spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1736    const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1737{
1738	if (bp != NULL) {
1739		zio_t *rio = arg;
1740		size_t size = BP_GET_PSIZE(bp);
1741		void *data = zio_data_buf_alloc(size);
1742
1743		zio_nowait(zio_read(rio, spa, bp, data, size,
1744		    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1745		    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1746		    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1747	}
1748	return (0);
1749}
1750
1751static int
1752spa_load_verify(spa_t *spa)
1753{
1754	zio_t *rio;
1755	spa_load_error_t sle = { 0 };
1756	zpool_rewind_policy_t policy;
1757	boolean_t verify_ok = B_FALSE;
1758	int error;
1759
1760	zpool_get_rewind_policy(spa->spa_config, &policy);
1761
1762	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1763		return (0);
1764
1765	rio = zio_root(spa, NULL, &sle,
1766	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1767
1768	error = traverse_pool(spa, spa->spa_verify_min_txg,
1769	    TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1770
1771	(void) zio_wait(rio);
1772
1773	spa->spa_load_meta_errors = sle.sle_meta_count;
1774	spa->spa_load_data_errors = sle.sle_data_count;
1775
1776	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1777	    sle.sle_data_count <= policy.zrp_maxdata) {
1778		int64_t loss = 0;
1779
1780		verify_ok = B_TRUE;
1781		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1782		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1783
1784		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1785		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1786		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1787		VERIFY(nvlist_add_int64(spa->spa_load_info,
1788		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1789		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1790		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1791	} else {
1792		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1793	}
1794
1795	if (error) {
1796		if (error != ENXIO && error != EIO)
1797			error = EIO;
1798		return (error);
1799	}
1800
1801	return (verify_ok ? 0 : EIO);
1802}
1803
1804/*
1805 * Find a value in the pool props object.
1806 */
1807static void
1808spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1809{
1810	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1811	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1812}
1813
1814/*
1815 * Find a value in the pool directory object.
1816 */
1817static int
1818spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1819{
1820	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1821	    name, sizeof (uint64_t), 1, val));
1822}
1823
1824static int
1825spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1826{
1827	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1828	return (err);
1829}
1830
1831/*
1832 * Fix up config after a partly-completed split.  This is done with the
1833 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1834 * pool have that entry in their config, but only the splitting one contains
1835 * a list of all the guids of the vdevs that are being split off.
1836 *
1837 * This function determines what to do with that list: either rejoin
1838 * all the disks to the pool, or complete the splitting process.  To attempt
1839 * the rejoin, each disk that is offlined is marked online again, and
1840 * we do a reopen() call.  If the vdev label for every disk that was
1841 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1842 * then we call vdev_split() on each disk, and complete the split.
1843 *
1844 * Otherwise we leave the config alone, with all the vdevs in place in
1845 * the original pool.
1846 */
1847static void
1848spa_try_repair(spa_t *spa, nvlist_t *config)
1849{
1850	uint_t extracted;
1851	uint64_t *glist;
1852	uint_t i, gcount;
1853	nvlist_t *nvl;
1854	vdev_t **vd;
1855	boolean_t attempt_reopen;
1856
1857	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1858		return;
1859
1860	/* check that the config is complete */
1861	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1862	    &glist, &gcount) != 0)
1863		return;
1864
1865	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1866
1867	/* attempt to online all the vdevs & validate */
1868	attempt_reopen = B_TRUE;
1869	for (i = 0; i < gcount; i++) {
1870		if (glist[i] == 0)	/* vdev is hole */
1871			continue;
1872
1873		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1874		if (vd[i] == NULL) {
1875			/*
1876			 * Don't bother attempting to reopen the disks;
1877			 * just do the split.
1878			 */
1879			attempt_reopen = B_FALSE;
1880		} else {
1881			/* attempt to re-online it */
1882			vd[i]->vdev_offline = B_FALSE;
1883		}
1884	}
1885
1886	if (attempt_reopen) {
1887		vdev_reopen(spa->spa_root_vdev);
1888
1889		/* check each device to see what state it's in */
1890		for (extracted = 0, i = 0; i < gcount; i++) {
1891			if (vd[i] != NULL &&
1892			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1893				break;
1894			++extracted;
1895		}
1896	}
1897
1898	/*
1899	 * If every disk has been moved to the new pool, or if we never
1900	 * even attempted to look at them, then we split them off for
1901	 * good.
1902	 */
1903	if (!attempt_reopen || gcount == extracted) {
1904		for (i = 0; i < gcount; i++)
1905			if (vd[i] != NULL)
1906				vdev_split(vd[i]);
1907		vdev_reopen(spa->spa_root_vdev);
1908	}
1909
1910	kmem_free(vd, gcount * sizeof (vdev_t *));
1911}
1912
1913static int
1914spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1915    boolean_t mosconfig)
1916{
1917	nvlist_t *config = spa->spa_config;
1918	char *ereport = FM_EREPORT_ZFS_POOL;
1919	char *comment;
1920	int error;
1921	uint64_t pool_guid;
1922	nvlist_t *nvl;
1923
1924	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1925		return (EINVAL);
1926
1927	ASSERT(spa->spa_comment == NULL);
1928	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
1929		spa->spa_comment = spa_strdup(comment);
1930
1931	/*
1932	 * Versioning wasn't explicitly added to the label until later, so if
1933	 * it's not present treat it as the initial version.
1934	 */
1935	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1936	    &spa->spa_ubsync.ub_version) != 0)
1937		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1938
1939	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1940	    &spa->spa_config_txg);
1941
1942	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1943	    spa_guid_exists(pool_guid, 0)) {
1944		error = EEXIST;
1945	} else {
1946		spa->spa_config_guid = pool_guid;
1947
1948		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1949		    &nvl) == 0) {
1950			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1951			    KM_SLEEP) == 0);
1952		}
1953
1954		nvlist_free(spa->spa_load_info);
1955		spa->spa_load_info = fnvlist_alloc();
1956
1957		gethrestime(&spa->spa_loaded_ts);
1958		error = spa_load_impl(spa, pool_guid, config, state, type,
1959		    mosconfig, &ereport);
1960	}
1961
1962	spa->spa_minref = refcount_count(&spa->spa_refcount);
1963	if (error) {
1964		if (error != EEXIST) {
1965			spa->spa_loaded_ts.tv_sec = 0;
1966			spa->spa_loaded_ts.tv_nsec = 0;
1967		}
1968		if (error != EBADF) {
1969			zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1970		}
1971	}
1972	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1973	spa->spa_ena = 0;
1974
1975	return (error);
1976}
1977
1978/*
1979 * Load an existing storage pool, using the pool's builtin spa_config as a
1980 * source of configuration information.
1981 */
1982static int
1983spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1984    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1985    char **ereport)
1986{
1987	int error = 0;
1988	nvlist_t *nvroot = NULL;
1989	nvlist_t *label;
1990	vdev_t *rvd;
1991	uberblock_t *ub = &spa->spa_uberblock;
1992	uint64_t children, config_cache_txg = spa->spa_config_txg;
1993	int orig_mode = spa->spa_mode;
1994	int parse;
1995	uint64_t obj;
1996	boolean_t missing_feat_write = B_FALSE;
1997
1998	/*
1999	 * If this is an untrusted config, access the pool in read-only mode.
2000	 * This prevents things like resilvering recently removed devices.
2001	 */
2002	if (!mosconfig)
2003		spa->spa_mode = FREAD;
2004
2005	ASSERT(MUTEX_HELD(&spa_namespace_lock));
2006
2007	spa->spa_load_state = state;
2008
2009	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2010		return (EINVAL);
2011
2012	parse = (type == SPA_IMPORT_EXISTING ?
2013	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2014
2015	/*
2016	 * Create "The Godfather" zio to hold all async IOs
2017	 */
2018	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2019	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2020
2021	/*
2022	 * Parse the configuration into a vdev tree.  We explicitly set the
2023	 * value that will be returned by spa_version() since parsing the
2024	 * configuration requires knowing the version number.
2025	 */
2026	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2027	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2028	spa_config_exit(spa, SCL_ALL, FTAG);
2029
2030	if (error != 0)
2031		return (error);
2032
2033	ASSERT(spa->spa_root_vdev == rvd);
2034
2035	if (type != SPA_IMPORT_ASSEMBLE) {
2036		ASSERT(spa_guid(spa) == pool_guid);
2037	}
2038
2039	/*
2040	 * Try to open all vdevs, loading each label in the process.
2041	 */
2042	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2043	error = vdev_open(rvd);
2044	spa_config_exit(spa, SCL_ALL, FTAG);
2045	if (error != 0)
2046		return (error);
2047
2048	/*
2049	 * We need to validate the vdev labels against the configuration that
2050	 * we have in hand, which is dependent on the setting of mosconfig. If
2051	 * mosconfig is true then we're validating the vdev labels based on
2052	 * that config.  Otherwise, we're validating against the cached config
2053	 * (zpool.cache) that was read when we loaded the zfs module, and then
2054	 * later we will recursively call spa_load() and validate against
2055	 * the vdev config.
2056	 *
2057	 * If we're assembling a new pool that's been split off from an
2058	 * existing pool, the labels haven't yet been updated so we skip
2059	 * validation for now.
2060	 */
2061	if (type != SPA_IMPORT_ASSEMBLE) {
2062		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2063		error = vdev_validate(rvd, mosconfig);
2064		spa_config_exit(spa, SCL_ALL, FTAG);
2065
2066		if (error != 0)
2067			return (error);
2068
2069		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2070			return (ENXIO);
2071	}
2072
2073	/*
2074	 * Find the best uberblock.
2075	 */
2076	vdev_uberblock_load(rvd, ub, &label);
2077
2078	/*
2079	 * If we weren't able to find a single valid uberblock, return failure.
2080	 */
2081	if (ub->ub_txg == 0) {
2082		nvlist_free(label);
2083		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2084	}
2085
2086	/*
2087	 * If the pool has an unsupported version we can't open it.
2088	 */
2089	if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2090		nvlist_free(label);
2091		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2092	}
2093
2094	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2095		nvlist_t *features;
2096
2097		/*
2098		 * If we weren't able to find what's necessary for reading the
2099		 * MOS in the label, return failure.
2100		 */
2101		if (label == NULL || nvlist_lookup_nvlist(label,
2102		    ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2103			nvlist_free(label);
2104			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2105			    ENXIO));
2106		}
2107
2108		/*
2109		 * Update our in-core representation with the definitive values
2110		 * from the label.
2111		 */
2112		nvlist_free(spa->spa_label_features);
2113		VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2114	}
2115
2116	nvlist_free(label);
2117
2118	/*
2119	 * Look through entries in the label nvlist's features_for_read. If
2120	 * there is a feature listed there which we don't understand then we
2121	 * cannot open a pool.
2122	 */
2123	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2124		nvlist_t *unsup_feat;
2125
2126		VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2127		    0);
2128
2129		for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2130		    NULL); nvp != NULL;
2131		    nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2132			if (!zfeature_is_supported(nvpair_name(nvp))) {
2133				VERIFY(nvlist_add_string(unsup_feat,
2134				    nvpair_name(nvp), "") == 0);
2135			}
2136		}
2137
2138		if (!nvlist_empty(unsup_feat)) {
2139			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2140			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2141			nvlist_free(unsup_feat);
2142			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2143			    ENOTSUP));
2144		}
2145
2146		nvlist_free(unsup_feat);
2147	}
2148
2149	/*
2150	 * If the vdev guid sum doesn't match the uberblock, we have an
2151	 * incomplete configuration.  We first check to see if the pool
2152	 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2153	 * If it is, defer the vdev_guid_sum check till later so we
2154	 * can handle missing vdevs.
2155	 */
2156	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2157	    &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2158	    rvd->vdev_guid_sum != ub->ub_guid_sum)
2159		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2160
2161	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2162		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2163		spa_try_repair(spa, config);
2164		spa_config_exit(spa, SCL_ALL, FTAG);
2165		nvlist_free(spa->spa_config_splitting);
2166		spa->spa_config_splitting = NULL;
2167	}
2168
2169	/*
2170	 * Initialize internal SPA structures.
2171	 */
2172	spa->spa_state = POOL_STATE_ACTIVE;
2173	spa->spa_ubsync = spa->spa_uberblock;
2174	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2175	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2176	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2177	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2178	spa->spa_claim_max_txg = spa->spa_first_txg;
2179	spa->spa_prev_software_version = ub->ub_software_version;
2180
2181	error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2182	if (error)
2183		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2184	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2185
2186	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2187		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2188
2189	if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2190		boolean_t missing_feat_read = B_FALSE;
2191		nvlist_t *unsup_feat, *enabled_feat;
2192
2193		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2194		    &spa->spa_feat_for_read_obj) != 0) {
2195			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2196		}
2197
2198		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2199		    &spa->spa_feat_for_write_obj) != 0) {
2200			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2201		}
2202
2203		if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2204		    &spa->spa_feat_desc_obj) != 0) {
2205			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2206		}
2207
2208		enabled_feat = fnvlist_alloc();
2209		unsup_feat = fnvlist_alloc();
2210
2211		if (!feature_is_supported(spa->spa_meta_objset,
2212		    spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2213		    unsup_feat, enabled_feat))
2214			missing_feat_read = B_TRUE;
2215
2216		if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2217			if (!feature_is_supported(spa->spa_meta_objset,
2218			    spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2219			    unsup_feat, enabled_feat)) {
2220				missing_feat_write = B_TRUE;
2221			}
2222		}
2223
2224		fnvlist_add_nvlist(spa->spa_load_info,
2225		    ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2226
2227		if (!nvlist_empty(unsup_feat)) {
2228			fnvlist_add_nvlist(spa->spa_load_info,
2229			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2230		}
2231
2232		fnvlist_free(enabled_feat);
2233		fnvlist_free(unsup_feat);
2234
2235		if (!missing_feat_read) {
2236			fnvlist_add_boolean(spa->spa_load_info,
2237			    ZPOOL_CONFIG_CAN_RDONLY);
2238		}
2239
2240		/*
2241		 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2242		 * twofold: to determine whether the pool is available for
2243		 * import in read-write mode and (if it is not) whether the
2244		 * pool is available for import in read-only mode. If the pool
2245		 * is available for import in read-write mode, it is displayed
2246		 * as available in userland; if it is not available for import
2247		 * in read-only mode, it is displayed as unavailable in
2248		 * userland. If the pool is available for import in read-only
2249		 * mode but not read-write mode, it is displayed as unavailable
2250		 * in userland with a special note that the pool is actually
2251		 * available for open in read-only mode.
2252		 *
2253		 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2254		 * missing a feature for write, we must first determine whether
2255		 * the pool can be opened read-only before returning to
2256		 * userland in order to know whether to display the
2257		 * abovementioned note.
2258		 */
2259		if (missing_feat_read || (missing_feat_write &&
2260		    spa_writeable(spa))) {
2261			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2262			    ENOTSUP));
2263		}
2264	}
2265
2266	spa->spa_is_initializing = B_TRUE;
2267	error = dsl_pool_open(spa->spa_dsl_pool);
2268	spa->spa_is_initializing = B_FALSE;
2269	if (error != 0)
2270		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2271
2272	if (!mosconfig) {
2273		uint64_t hostid;
2274		nvlist_t *policy = NULL, *nvconfig;
2275
2276		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2277			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2278
2279		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2280		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2281			char *hostname;
2282			unsigned long myhostid = 0;
2283
2284			VERIFY(nvlist_lookup_string(nvconfig,
2285			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2286
2287#ifdef	_KERNEL
2288			myhostid = zone_get_hostid(NULL);
2289#else	/* _KERNEL */
2290			/*
2291			 * We're emulating the system's hostid in userland, so
2292			 * we can't use zone_get_hostid().
2293			 */
2294			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2295#endif	/* _KERNEL */
2296			if (hostid != 0 && myhostid != 0 &&
2297			    hostid != myhostid) {
2298				nvlist_free(nvconfig);
2299				cmn_err(CE_WARN, "pool '%s' could not be "
2300				    "loaded as it was last accessed by "
2301				    "another system (host: %s hostid: 0x%lx). "
2302				    "See: http://illumos.org/msg/ZFS-8000-EY",
2303				    spa_name(spa), hostname,
2304				    (unsigned long)hostid);
2305				return (EBADF);
2306			}
2307		}
2308		if (nvlist_lookup_nvlist(spa->spa_config,
2309		    ZPOOL_REWIND_POLICY, &policy) == 0)
2310			VERIFY(nvlist_add_nvlist(nvconfig,
2311			    ZPOOL_REWIND_POLICY, policy) == 0);
2312
2313		spa_config_set(spa, nvconfig);
2314		spa_unload(spa);
2315		spa_deactivate(spa);
2316		spa_activate(spa, orig_mode);
2317
2318		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2319	}
2320
2321	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2322		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2323	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2324	if (error != 0)
2325		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2326
2327	/*
2328	 * Load the bit that tells us to use the new accounting function
2329	 * (raid-z deflation).  If we have an older pool, this will not
2330	 * be present.
2331	 */
2332	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2333	if (error != 0 && error != ENOENT)
2334		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2335
2336	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2337	    &spa->spa_creation_version);
2338	if (error != 0 && error != ENOENT)
2339		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2340
2341	/*
2342	 * Load the persistent error log.  If we have an older pool, this will
2343	 * not be present.
2344	 */
2345	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2346	if (error != 0 && error != ENOENT)
2347		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2348
2349	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2350	    &spa->spa_errlog_scrub);
2351	if (error != 0 && error != ENOENT)
2352		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2353
2354	/*
2355	 * Load the history object.  If we have an older pool, this
2356	 * will not be present.
2357	 */
2358	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2359	if (error != 0 && error != ENOENT)
2360		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2361
2362	/*
2363	 * If we're assembling the pool from the split-off vdevs of
2364	 * an existing pool, we don't want to attach the spares & cache
2365	 * devices.
2366	 */
2367
2368	/*
2369	 * Load any hot spares for this pool.
2370	 */
2371	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2372	if (error != 0 && error != ENOENT)
2373		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2374	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2375		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2376		if (load_nvlist(spa, spa->spa_spares.sav_object,
2377		    &spa->spa_spares.sav_config) != 0)
2378			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2379
2380		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2381		spa_load_spares(spa);
2382		spa_config_exit(spa, SCL_ALL, FTAG);
2383	} else if (error == 0) {
2384		spa->spa_spares.sav_sync = B_TRUE;
2385	}
2386
2387	/*
2388	 * Load any level 2 ARC devices for this pool.
2389	 */
2390	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2391	    &spa->spa_l2cache.sav_object);
2392	if (error != 0 && error != ENOENT)
2393		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2394	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2395		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2396		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2397		    &spa->spa_l2cache.sav_config) != 0)
2398			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2399
2400		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2401		spa_load_l2cache(spa);
2402		spa_config_exit(spa, SCL_ALL, FTAG);
2403	} else if (error == 0) {
2404		spa->spa_l2cache.sav_sync = B_TRUE;
2405	}
2406
2407	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2408
2409	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2410	if (error && error != ENOENT)
2411		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2412
2413	if (error == 0) {
2414		uint64_t autoreplace;
2415
2416		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2417		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2418		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2419		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2420		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2421		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2422		    &spa->spa_dedup_ditto);
2423
2424		spa->spa_autoreplace = (autoreplace != 0);
2425	}
2426
2427	/*
2428	 * If the 'autoreplace' property is set, then post a resource notifying
2429	 * the ZFS DE that it should not issue any faults for unopenable
2430	 * devices.  We also iterate over the vdevs, and post a sysevent for any
2431	 * unopenable vdevs so that the normal autoreplace handler can take
2432	 * over.
2433	 */
2434	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2435		spa_check_removed(spa->spa_root_vdev);
2436		/*
2437		 * For the import case, this is done in spa_import(), because
2438		 * at this point we're using the spare definitions from
2439		 * the MOS config, not necessarily from the userland config.
2440		 */
2441		if (state != SPA_LOAD_IMPORT) {
2442			spa_aux_check_removed(&spa->spa_spares);
2443			spa_aux_check_removed(&spa->spa_l2cache);
2444		}
2445	}
2446
2447	/*
2448	 * Load the vdev state for all toplevel vdevs.
2449	 */
2450	vdev_load(rvd);
2451
2452	/*
2453	 * Propagate the leaf DTLs we just loaded all the way up the tree.
2454	 */
2455	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2456	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2457	spa_config_exit(spa, SCL_ALL, FTAG);
2458
2459	/*
2460	 * Load the DDTs (dedup tables).
2461	 */
2462	error = ddt_load(spa);
2463	if (error != 0)
2464		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2465
2466	spa_update_dspace(spa);
2467
2468	/*
2469	 * Validate the config, using the MOS config to fill in any
2470	 * information which might be missing.  If we fail to validate
2471	 * the config then declare the pool unfit for use. If we're
2472	 * assembling a pool from a split, the log is not transferred
2473	 * over.
2474	 */
2475	if (type != SPA_IMPORT_ASSEMBLE) {
2476		nvlist_t *nvconfig;
2477
2478		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2479			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2480
2481		if (!spa_config_valid(spa, nvconfig)) {
2482			nvlist_free(nvconfig);
2483			return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2484			    ENXIO));
2485		}
2486		nvlist_free(nvconfig);
2487
2488		/*
2489		 * Now that we've validated the config, check the state of the
2490		 * root vdev.  If it can't be opened, it indicates one or
2491		 * more toplevel vdevs are faulted.
2492		 */
2493		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2494			return (ENXIO);
2495
2496		if (spa_check_logs(spa)) {
2497			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2498			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2499		}
2500	}
2501
2502	if (missing_feat_write) {
2503		ASSERT(state == SPA_LOAD_TRYIMPORT);
2504
2505		/*
2506		 * At this point, we know that we can open the pool in
2507		 * read-only mode but not read-write mode. We now have enough
2508		 * information and can return to userland.
2509		 */
2510		return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2511	}
2512
2513	/*
2514	 * We've successfully opened the pool, verify that we're ready
2515	 * to start pushing transactions.
2516	 */
2517	if (state != SPA_LOAD_TRYIMPORT) {
2518		if (error = spa_load_verify(spa))
2519			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2520			    error));
2521	}
2522
2523	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2524	    spa->spa_load_max_txg == UINT64_MAX)) {
2525		dmu_tx_t *tx;
2526		int need_update = B_FALSE;
2527
2528		ASSERT(state != SPA_LOAD_TRYIMPORT);
2529
2530		/*
2531		 * Claim log blocks that haven't been committed yet.
2532		 * This must all happen in a single txg.
2533		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2534		 * invoked from zil_claim_log_block()'s i/o done callback.
2535		 * Price of rollback is that we abandon the log.
2536		 */
2537		spa->spa_claiming = B_TRUE;
2538
2539		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2540		    spa_first_txg(spa));
2541		(void) dmu_objset_find(spa_name(spa),
2542		    zil_claim, tx, DS_FIND_CHILDREN);
2543		dmu_tx_commit(tx);
2544
2545		spa->spa_claiming = B_FALSE;
2546
2547		spa_set_log_state(spa, SPA_LOG_GOOD);
2548		spa->spa_sync_on = B_TRUE;
2549		txg_sync_start(spa->spa_dsl_pool);
2550
2551		/*
2552		 * Wait for all claims to sync.  We sync up to the highest
2553		 * claimed log block birth time so that claimed log blocks
2554		 * don't appear to be from the future.  spa_claim_max_txg
2555		 * will have been set for us by either zil_check_log_chain()
2556		 * (invoked from spa_check_logs()) or zil_claim() above.
2557		 */
2558		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2559
2560		/*
2561		 * If the config cache is stale, or we have uninitialized
2562		 * metaslabs (see spa_vdev_add()), then update the config.
2563		 *
2564		 * If this is a verbatim import, trust the current
2565		 * in-core spa_config and update the disk labels.
2566		 */
2567		if (config_cache_txg != spa->spa_config_txg ||
2568		    state == SPA_LOAD_IMPORT ||
2569		    state == SPA_LOAD_RECOVER ||
2570		    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2571			need_update = B_TRUE;
2572
2573		for (int c = 0; c < rvd->vdev_children; c++)
2574			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2575				need_update = B_TRUE;
2576
2577		/*
2578		 * Update the config cache asychronously in case we're the
2579		 * root pool, in which case the config cache isn't writable yet.
2580		 */
2581		if (need_update)
2582			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2583
2584		/*
2585		 * Check all DTLs to see if anything needs resilvering.
2586		 */
2587		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2588		    vdev_resilver_needed(rvd, NULL, NULL))
2589			spa_async_request(spa, SPA_ASYNC_RESILVER);
2590
2591		/*
2592		 * Log the fact that we booted up (so that we can detect if
2593		 * we rebooted in the middle of an operation).
2594		 */
2595		spa_history_log_version(spa, "open");
2596
2597		/*
2598		 * Delete any inconsistent datasets.
2599		 */
2600		(void) dmu_objset_find(spa_name(spa),
2601		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2602
2603		/*
2604		 * Clean up any stale temporary dataset userrefs.
2605		 */
2606		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2607	}
2608
2609	return (0);
2610}
2611
2612static int
2613spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2614{
2615	int mode = spa->spa_mode;
2616
2617	spa_unload(spa);
2618	spa_deactivate(spa);
2619
2620	spa->spa_load_max_txg--;
2621
2622	spa_activate(spa, mode);
2623	spa_async_suspend(spa);
2624
2625	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2626}
2627
2628/*
2629 * If spa_load() fails this function will try loading prior txg's. If
2630 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2631 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2632 * function will not rewind the pool and will return the same error as
2633 * spa_load().
2634 */
2635static int
2636spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2637    uint64_t max_request, int rewind_flags)
2638{
2639	nvlist_t *loadinfo = NULL;
2640	nvlist_t *config = NULL;
2641	int load_error, rewind_error;
2642	uint64_t safe_rewind_txg;
2643	uint64_t min_txg;
2644
2645	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2646		spa->spa_load_max_txg = spa->spa_load_txg;
2647		spa_set_log_state(spa, SPA_LOG_CLEAR);
2648	} else {
2649		spa->spa_load_max_txg = max_request;
2650	}
2651
2652	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2653	    mosconfig);
2654	if (load_error == 0)
2655		return (0);
2656
2657	if (spa->spa_root_vdev != NULL)
2658		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2659
2660	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2661	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2662
2663	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2664		nvlist_free(config);
2665		return (load_error);
2666	}
2667
2668	if (state == SPA_LOAD_RECOVER) {
2669		/* Price of rolling back is discarding txgs, including log */
2670		spa_set_log_state(spa, SPA_LOG_CLEAR);
2671	} else {
2672		/*
2673		 * If we aren't rolling back save the load info from our first
2674		 * import attempt so that we can restore it after attempting
2675		 * to rewind.
2676		 */
2677		loadinfo = spa->spa_load_info;
2678		spa->spa_load_info = fnvlist_alloc();
2679	}
2680
2681	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2682	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2683	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2684	    TXG_INITIAL : safe_rewind_txg;
2685
2686	/*
2687	 * Continue as long as we're finding errors, we're still within
2688	 * the acceptable rewind range, and we're still finding uberblocks
2689	 */
2690	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2691	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2692		if (spa->spa_load_max_txg < safe_rewind_txg)
2693			spa->spa_extreme_rewind = B_TRUE;
2694		rewind_error = spa_load_retry(spa, state, mosconfig);
2695	}
2696
2697	spa->spa_extreme_rewind = B_FALSE;
2698	spa->spa_load_max_txg = UINT64_MAX;
2699
2700	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2701		spa_config_set(spa, config);
2702
2703	if (state == SPA_LOAD_RECOVER) {
2704		ASSERT3P(loadinfo, ==, NULL);
2705		return (rewind_error);
2706	} else {
2707		/* Store the rewind info as part of the initial load info */
2708		fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2709		    spa->spa_load_info);
2710
2711		/* Restore the initial load info */
2712		fnvlist_free(spa->spa_load_info);
2713		spa->spa_load_info = loadinfo;
2714
2715		return (load_error);
2716	}
2717}
2718
2719/*
2720 * Pool Open/Import
2721 *
2722 * The import case is identical to an open except that the configuration is sent
2723 * down from userland, instead of grabbed from the configuration cache.  For the
2724 * case of an open, the pool configuration will exist in the
2725 * POOL_STATE_UNINITIALIZED state.
2726 *
2727 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2728 * the same time open the pool, without having to keep around the spa_t in some
2729 * ambiguous state.
2730 */
2731static int
2732spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2733    nvlist_t **config)
2734{
2735	spa_t *spa;
2736	spa_load_state_t state = SPA_LOAD_OPEN;
2737	int error;
2738	int locked = B_FALSE;
2739
2740	*spapp = NULL;
2741
2742	/*
2743	 * As disgusting as this is, we need to support recursive calls to this
2744	 * function because dsl_dir_open() is called during spa_load(), and ends
2745	 * up calling spa_open() again.  The real fix is to figure out how to
2746	 * avoid dsl_dir_open() calling this in the first place.
2747	 */
2748	if (mutex_owner(&spa_namespace_lock) != curthread) {
2749		mutex_enter(&spa_namespace_lock);
2750		locked = B_TRUE;
2751	}
2752
2753	if ((spa = spa_lookup(pool)) == NULL) {
2754		if (locked)
2755			mutex_exit(&spa_namespace_lock);
2756		return (ENOENT);
2757	}
2758
2759	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2760		zpool_rewind_policy_t policy;
2761
2762		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2763		    &policy);
2764		if (policy.zrp_request & ZPOOL_DO_REWIND)
2765			state = SPA_LOAD_RECOVER;
2766
2767		spa_activate(spa, spa_mode_global);
2768
2769		if (state != SPA_LOAD_RECOVER)
2770			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2771
2772		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2773		    policy.zrp_request);
2774
2775		if (error == EBADF) {
2776			/*
2777			 * If vdev_validate() returns failure (indicated by
2778			 * EBADF), it indicates that one of the vdevs indicates
2779			 * that the pool has been exported or destroyed.  If
2780			 * this is the case, the config cache is out of sync and
2781			 * we should remove the pool from the namespace.
2782			 */
2783			spa_unload(spa);
2784			spa_deactivate(spa);
2785			spa_config_sync(spa, B_TRUE, B_TRUE);
2786			spa_remove(spa);
2787			if (locked)
2788				mutex_exit(&spa_namespace_lock);
2789			return (ENOENT);
2790		}
2791
2792		if (error) {
2793			/*
2794			 * We can't open the pool, but we still have useful
2795			 * information: the state of each vdev after the
2796			 * attempted vdev_open().  Return this to the user.
2797			 */
2798			if (config != NULL && spa->spa_config) {
2799				VERIFY(nvlist_dup(spa->spa_config, config,
2800				    KM_SLEEP) == 0);
2801				VERIFY(nvlist_add_nvlist(*config,
2802				    ZPOOL_CONFIG_LOAD_INFO,
2803				    spa->spa_load_info) == 0);
2804			}
2805			spa_unload(spa);
2806			spa_deactivate(spa);
2807			spa->spa_last_open_failed = error;
2808			if (locked)
2809				mutex_exit(&spa_namespace_lock);
2810			*spapp = NULL;
2811			return (error);
2812		}
2813	}
2814
2815	spa_open_ref(spa, tag);
2816
2817	if (config != NULL)
2818		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2819
2820	/*
2821	 * If we've recovered the pool, pass back any information we
2822	 * gathered while doing the load.
2823	 */
2824	if (state == SPA_LOAD_RECOVER) {
2825		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2826		    spa->spa_load_info) == 0);
2827	}
2828
2829	if (locked) {
2830		spa->spa_last_open_failed = 0;
2831		spa->spa_last_ubsync_txg = 0;
2832		spa->spa_load_txg = 0;
2833		mutex_exit(&spa_namespace_lock);
2834	}
2835
2836	*spapp = spa;
2837
2838	return (0);
2839}
2840
2841int
2842spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2843    nvlist_t **config)
2844{
2845	return (spa_open_common(name, spapp, tag, policy, config));
2846}
2847
2848int
2849spa_open(const char *name, spa_t **spapp, void *tag)
2850{
2851	return (spa_open_common(name, spapp, tag, NULL, NULL));
2852}
2853
2854/*
2855 * Lookup the given spa_t, incrementing the inject count in the process,
2856 * preventing it from being exported or destroyed.
2857 */
2858spa_t *
2859spa_inject_addref(char *name)
2860{
2861	spa_t *spa;
2862
2863	mutex_enter(&spa_namespace_lock);
2864	if ((spa = spa_lookup(name)) == NULL) {
2865		mutex_exit(&spa_namespace_lock);
2866		return (NULL);
2867	}
2868	spa->spa_inject_ref++;
2869	mutex_exit(&spa_namespace_lock);
2870
2871	return (spa);
2872}
2873
2874void
2875spa_inject_delref(spa_t *spa)
2876{
2877	mutex_enter(&spa_namespace_lock);
2878	spa->spa_inject_ref--;
2879	mutex_exit(&spa_namespace_lock);
2880}
2881
2882/*
2883 * Add spares device information to the nvlist.
2884 */
2885static void
2886spa_add_spares(spa_t *spa, nvlist_t *config)
2887{
2888	nvlist_t **spares;
2889	uint_t i, nspares;
2890	nvlist_t *nvroot;
2891	uint64_t guid;
2892	vdev_stat_t *vs;
2893	uint_t vsc;
2894	uint64_t pool;
2895
2896	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2897
2898	if (spa->spa_spares.sav_count == 0)
2899		return;
2900
2901	VERIFY(nvlist_lookup_nvlist(config,
2902	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2903	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2904	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2905	if (nspares != 0) {
2906		VERIFY(nvlist_add_nvlist_array(nvroot,
2907		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2908		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2909		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2910
2911		/*
2912		 * Go through and find any spares which have since been
2913		 * repurposed as an active spare.  If this is the case, update
2914		 * their status appropriately.
2915		 */
2916		for (i = 0; i < nspares; i++) {
2917			VERIFY(nvlist_lookup_uint64(spares[i],
2918			    ZPOOL_CONFIG_GUID, &guid) == 0);
2919			if (spa_spare_exists(guid, &pool, NULL) &&
2920			    pool != 0ULL) {
2921				VERIFY(nvlist_lookup_uint64_array(
2922				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
2923				    (uint64_t **)&vs, &vsc) == 0);
2924				vs->vs_state = VDEV_STATE_CANT_OPEN;
2925				vs->vs_aux = VDEV_AUX_SPARED;
2926			}
2927		}
2928	}
2929}
2930
2931/*
2932 * Add l2cache device information to the nvlist, including vdev stats.
2933 */
2934static void
2935spa_add_l2cache(spa_t *spa, nvlist_t *config)
2936{
2937	nvlist_t **l2cache;
2938	uint_t i, j, nl2cache;
2939	nvlist_t *nvroot;
2940	uint64_t guid;
2941	vdev_t *vd;
2942	vdev_stat_t *vs;
2943	uint_t vsc;
2944
2945	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2946
2947	if (spa->spa_l2cache.sav_count == 0)
2948		return;
2949
2950	VERIFY(nvlist_lookup_nvlist(config,
2951	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2952	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2953	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2954	if (nl2cache != 0) {
2955		VERIFY(nvlist_add_nvlist_array(nvroot,
2956		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2957		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2958		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2959
2960		/*
2961		 * Update level 2 cache device stats.
2962		 */
2963
2964		for (i = 0; i < nl2cache; i++) {
2965			VERIFY(nvlist_lookup_uint64(l2cache[i],
2966			    ZPOOL_CONFIG_GUID, &guid) == 0);
2967
2968			vd = NULL;
2969			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2970				if (guid ==
2971				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2972					vd = spa->spa_l2cache.sav_vdevs[j];
2973					break;
2974				}
2975			}
2976			ASSERT(vd != NULL);
2977
2978			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2979			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2980			    == 0);
2981			vdev_get_stats(vd, vs);
2982		}
2983	}
2984}
2985
2986static void
2987spa_add_feature_stats(spa_t *spa, nvlist_t *config)
2988{
2989	nvlist_t *features;
2990	zap_cursor_t zc;
2991	zap_attribute_t za;
2992
2993	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2994	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2995
2996	if (spa->spa_feat_for_read_obj != 0) {
2997		for (zap_cursor_init(&zc, spa->spa_meta_objset,
2998		    spa->spa_feat_for_read_obj);
2999		    zap_cursor_retrieve(&zc, &za) == 0;
3000		    zap_cursor_advance(&zc)) {
3001			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3002			    za.za_num_integers == 1);
3003			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3004			    za.za_first_integer));
3005		}
3006		zap_cursor_fini(&zc);
3007	}
3008
3009	if (spa->spa_feat_for_write_obj != 0) {
3010		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3011		    spa->spa_feat_for_write_obj);
3012		    zap_cursor_retrieve(&zc, &za) == 0;
3013		    zap_cursor_advance(&zc)) {
3014			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3015			    za.za_num_integers == 1);
3016			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3017			    za.za_first_integer));
3018		}
3019		zap_cursor_fini(&zc);
3020	}
3021
3022	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3023	    features) == 0);
3024	nvlist_free(features);
3025}
3026
3027int
3028spa_get_stats(const char *name, nvlist_t **config,
3029    char *altroot, size_t buflen)
3030{
3031	int error;
3032	spa_t *spa;
3033
3034	*config = NULL;
3035	error = spa_open_common(name, &spa, FTAG, NULL, config);
3036
3037	if (spa != NULL) {
3038		/*
3039		 * This still leaves a window of inconsistency where the spares
3040		 * or l2cache devices could change and the config would be
3041		 * self-inconsistent.
3042		 */
3043		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3044
3045		if (*config != NULL) {
3046			uint64_t loadtimes[2];
3047
3048			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3049			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3050			VERIFY(nvlist_add_uint64_array(*config,
3051			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3052
3053			VERIFY(nvlist_add_uint64(*config,
3054			    ZPOOL_CONFIG_ERRCOUNT,
3055			    spa_get_errlog_size(spa)) == 0);
3056
3057			if (spa_suspended(spa))
3058				VERIFY(nvlist_add_uint64(*config,
3059				    ZPOOL_CONFIG_SUSPENDED,
3060				    spa->spa_failmode) == 0);
3061
3062			spa_add_spares(spa, *config);
3063			spa_add_l2cache(spa, *config);
3064			spa_add_feature_stats(spa, *config);
3065		}
3066	}
3067
3068	/*
3069	 * We want to get the alternate root even for faulted pools, so we cheat
3070	 * and call spa_lookup() directly.
3071	 */
3072	if (altroot) {
3073		if (spa == NULL) {
3074			mutex_enter(&spa_namespace_lock);
3075			spa = spa_lookup(name);
3076			if (spa)
3077				spa_altroot(spa, altroot, buflen);
3078			else
3079				altroot[0] = '\0';
3080			spa = NULL;
3081			mutex_exit(&spa_namespace_lock);
3082		} else {
3083			spa_altroot(spa, altroot, buflen);
3084		}
3085	}
3086
3087	if (spa != NULL) {
3088		spa_config_exit(spa, SCL_CONFIG, FTAG);
3089		spa_close(spa, FTAG);
3090	}
3091
3092	return (error);
3093}
3094
3095/*
3096 * Validate that the auxiliary device array is well formed.  We must have an
3097 * array of nvlists, each which describes a valid leaf vdev.  If this is an
3098 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3099 * specified, as long as they are well-formed.
3100 */
3101static int
3102spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3103    spa_aux_vdev_t *sav, const char *config, uint64_t version,
3104    vdev_labeltype_t label)
3105{
3106	nvlist_t **dev;
3107	uint_t i, ndev;
3108	vdev_t *vd;
3109	int error;
3110
3111	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3112
3113	/*
3114	 * It's acceptable to have no devs specified.
3115	 */
3116	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3117		return (0);
3118
3119	if (ndev == 0)
3120		return (EINVAL);
3121
3122	/*
3123	 * Make sure the pool is formatted with a version that supports this
3124	 * device type.
3125	 */
3126	if (spa_version(spa) < version)
3127		return (ENOTSUP);
3128
3129	/*
3130	 * Set the pending device list so we correctly handle device in-use
3131	 * checking.
3132	 */
3133	sav->sav_pending = dev;
3134	sav->sav_npending = ndev;
3135
3136	for (i = 0; i < ndev; i++) {
3137		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3138		    mode)) != 0)
3139			goto out;
3140
3141		if (!vd->vdev_ops->vdev_op_leaf) {
3142			vdev_free(vd);
3143			error = EINVAL;
3144			goto out;
3145		}
3146
3147		/*
3148		 * The L2ARC currently only supports disk devices in
3149		 * kernel context.  For user-level testing, we allow it.
3150		 */
3151#ifdef _KERNEL
3152		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3153		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3154			error = ENOTBLK;
3155			vdev_free(vd);
3156			goto out;
3157		}
3158#endif
3159		vd->vdev_top = vd;
3160
3161		if ((error = vdev_open(vd)) == 0 &&
3162		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
3163			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3164			    vd->vdev_guid) == 0);
3165		}
3166
3167		vdev_free(vd);
3168
3169		if (error &&
3170		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3171			goto out;
3172		else
3173			error = 0;
3174	}
3175
3176out:
3177	sav->sav_pending = NULL;
3178	sav->sav_npending = 0;
3179	return (error);
3180}
3181
3182static int
3183spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3184{
3185	int error;
3186
3187	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3188
3189	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3190	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3191	    VDEV_LABEL_SPARE)) != 0) {
3192		return (error);
3193	}
3194
3195	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3196	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3197	    VDEV_LABEL_L2CACHE));
3198}
3199
3200static void
3201spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3202    const char *config)
3203{
3204	int i;
3205
3206	if (sav->sav_config != NULL) {
3207		nvlist_t **olddevs;
3208		uint_t oldndevs;
3209		nvlist_t **newdevs;
3210
3211		/*
3212		 * Generate new dev list by concatentating with the
3213		 * current dev list.
3214		 */
3215		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3216		    &olddevs, &oldndevs) == 0);
3217
3218		newdevs = kmem_alloc(sizeof (void *) *
3219		    (ndevs + oldndevs), KM_SLEEP);
3220		for (i = 0; i < oldndevs; i++)
3221			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3222			    KM_SLEEP) == 0);
3223		for (i = 0; i < ndevs; i++)
3224			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3225			    KM_SLEEP) == 0);
3226
3227		VERIFY(nvlist_remove(sav->sav_config, config,
3228		    DATA_TYPE_NVLIST_ARRAY) == 0);
3229
3230		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3231		    config, newdevs, ndevs + oldndevs) == 0);
3232		for (i = 0; i < oldndevs + ndevs; i++)
3233			nvlist_free(newdevs[i]);
3234		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3235	} else {
3236		/*
3237		 * Generate a new dev list.
3238		 */
3239		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3240		    KM_SLEEP) == 0);
3241		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3242		    devs, ndevs) == 0);
3243	}
3244}
3245
3246/*
3247 * Stop and drop level 2 ARC devices
3248 */
3249void
3250spa_l2cache_drop(spa_t *spa)
3251{
3252	vdev_t *vd;
3253	int i;
3254	spa_aux_vdev_t *sav = &spa->spa_l2cache;
3255
3256	for (i = 0; i < sav->sav_count; i++) {
3257		uint64_t pool;
3258
3259		vd = sav->sav_vdevs[i];
3260		ASSERT(vd != NULL);
3261
3262		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3263		    pool != 0ULL && l2arc_vdev_present(vd))
3264			l2arc_remove_vdev(vd);
3265	}
3266}
3267
3268/*
3269 * Pool Creation
3270 */
3271int
3272spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3273    nvlist_t *zplprops)
3274{
3275	spa_t *spa;
3276	char *altroot = NULL;
3277	vdev_t *rvd;
3278	dsl_pool_t *dp;
3279	dmu_tx_t *tx;
3280	int error = 0;
3281	uint64_t txg = TXG_INITIAL;
3282	nvlist_t **spares, **l2cache;
3283	uint_t nspares, nl2cache;
3284	uint64_t version, obj;
3285	boolean_t has_features;
3286
3287	/*
3288	 * If this pool already exists, return failure.
3289	 */
3290	mutex_enter(&spa_namespace_lock);
3291	if (spa_lookup(pool) != NULL) {
3292		mutex_exit(&spa_namespace_lock);
3293		return (EEXIST);
3294	}
3295
3296	/*
3297	 * Allocate a new spa_t structure.
3298	 */
3299	(void) nvlist_lookup_string(props,
3300	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3301	spa = spa_add(pool, NULL, altroot);
3302	spa_activate(spa, spa_mode_global);
3303
3304	if (props && (error = spa_prop_validate(spa, props))) {
3305		spa_deactivate(spa);
3306		spa_remove(spa);
3307		mutex_exit(&spa_namespace_lock);
3308		return (error);
3309	}
3310
3311	has_features = B_FALSE;
3312	for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3313	    elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3314		if (zpool_prop_feature(nvpair_name(elem)))
3315			has_features = B_TRUE;
3316	}
3317
3318	if (has_features || nvlist_lookup_uint64(props,
3319	    zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3320		version = SPA_VERSION;
3321	}
3322	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3323
3324	spa->spa_first_txg = txg;
3325	spa->spa_uberblock.ub_txg = txg - 1;
3326	spa->spa_uberblock.ub_version = version;
3327	spa->spa_ubsync = spa->spa_uberblock;
3328
3329	/*
3330	 * Create "The Godfather" zio to hold all async IOs
3331	 */
3332	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3333	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3334
3335	/*
3336	 * Create the root vdev.
3337	 */
3338	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3339
3340	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3341
3342	ASSERT(error != 0 || rvd != NULL);
3343	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3344
3345	if (error == 0 && !zfs_allocatable_devs(nvroot))
3346		error = EINVAL;
3347
3348	if (error == 0 &&
3349	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3350	    (error = spa_validate_aux(spa, nvroot, txg,
3351	    VDEV_ALLOC_ADD)) == 0) {
3352		for (int c = 0; c < rvd->vdev_children; c++) {
3353			vdev_metaslab_set_size(rvd->vdev_child[c]);
3354			vdev_expand(rvd->vdev_child[c], txg);
3355		}
3356	}
3357
3358	spa_config_exit(spa, SCL_ALL, FTAG);
3359
3360	if (error != 0) {
3361		spa_unload(spa);
3362		spa_deactivate(spa);
3363		spa_remove(spa);
3364		mutex_exit(&spa_namespace_lock);
3365		return (error);
3366	}
3367
3368	/*
3369	 * Get the list of spares, if specified.
3370	 */
3371	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3372	    &spares, &nspares) == 0) {
3373		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3374		    KM_SLEEP) == 0);
3375		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3376		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3377		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3378		spa_load_spares(spa);
3379		spa_config_exit(spa, SCL_ALL, FTAG);
3380		spa->spa_spares.sav_sync = B_TRUE;
3381	}
3382
3383	/*
3384	 * Get the list of level 2 cache devices, if specified.
3385	 */
3386	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3387	    &l2cache, &nl2cache) == 0) {
3388		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3389		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3390		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3391		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3392		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3393		spa_load_l2cache(spa);
3394		spa_config_exit(spa, SCL_ALL, FTAG);
3395		spa->spa_l2cache.sav_sync = B_TRUE;
3396	}
3397
3398	spa->spa_is_initializing = B_TRUE;
3399	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3400	spa->spa_meta_objset = dp->dp_meta_objset;
3401	spa->spa_is_initializing = B_FALSE;
3402
3403	/*
3404	 * Create DDTs (dedup tables).
3405	 */
3406	ddt_create(spa);
3407
3408	spa_update_dspace(spa);
3409
3410	tx = dmu_tx_create_assigned(dp, txg);
3411
3412	/*
3413	 * Create the pool config object.
3414	 */
3415	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3416	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3417	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3418
3419	if (zap_add(spa->spa_meta_objset,
3420	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3421	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3422		cmn_err(CE_PANIC, "failed to add pool config");
3423	}
3424
3425	if (spa_version(spa) >= SPA_VERSION_FEATURES)
3426		spa_feature_create_zap_objects(spa, tx);
3427
3428	if (zap_add(spa->spa_meta_objset,
3429	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3430	    sizeof (uint64_t), 1, &version, tx) != 0) {
3431		cmn_err(CE_PANIC, "failed to add pool version");
3432	}
3433
3434	/* Newly created pools with the right version are always deflated. */
3435	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3436		spa->spa_deflate = TRUE;
3437		if (zap_add(spa->spa_meta_objset,
3438		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3439		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3440			cmn_err(CE_PANIC, "failed to add deflate");
3441		}
3442	}
3443
3444	/*
3445	 * Create the deferred-free bpobj.  Turn off compression
3446	 * because sync-to-convergence takes longer if the blocksize
3447	 * keeps changing.
3448	 */
3449	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3450	dmu_object_set_compress(spa->spa_meta_objset, obj,
3451	    ZIO_COMPRESS_OFF, tx);
3452	if (zap_add(spa->spa_meta_objset,
3453	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3454	    sizeof (uint64_t), 1, &obj, tx) != 0) {
3455		cmn_err(CE_PANIC, "failed to add bpobj");
3456	}
3457	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3458	    spa->spa_meta_objset, obj));
3459
3460	/*
3461	 * Create the pool's history object.
3462	 */
3463	if (version >= SPA_VERSION_ZPOOL_HISTORY)
3464		spa_history_create_obj(spa, tx);
3465
3466	/*
3467	 * Set pool properties.
3468	 */
3469	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3470	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3471	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3472	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3473
3474	if (props != NULL) {
3475		spa_configfile_set(spa, props, B_FALSE);
3476		spa_sync_props(spa, props, tx);
3477	}
3478
3479	dmu_tx_commit(tx);
3480
3481	spa->spa_sync_on = B_TRUE;
3482	txg_sync_start(spa->spa_dsl_pool);
3483
3484	/*
3485	 * We explicitly wait for the first transaction to complete so that our
3486	 * bean counters are appropriately updated.
3487	 */
3488	txg_wait_synced(spa->spa_dsl_pool, txg);
3489
3490	spa_config_sync(spa, B_FALSE, B_TRUE);
3491
3492	spa_history_log_version(spa, "create");
3493
3494	spa->spa_minref = refcount_count(&spa->spa_refcount);
3495
3496	mutex_exit(&spa_namespace_lock);
3497
3498	return (0);
3499}
3500
3501#ifdef _KERNEL
3502/*
3503 * Get the root pool information from the root disk, then import the root pool
3504 * during the system boot up time.
3505 */
3506extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3507
3508static nvlist_t *
3509spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3510{
3511	nvlist_t *config;
3512	nvlist_t *nvtop, *nvroot;
3513	uint64_t pgid;
3514
3515	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3516		return (NULL);
3517
3518	/*
3519	 * Add this top-level vdev to the child array.
3520	 */
3521	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3522	    &nvtop) == 0);
3523	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3524	    &pgid) == 0);
3525	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3526
3527	/*
3528	 * Put this pool's top-level vdevs into a root vdev.
3529	 */
3530	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3531	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3532	    VDEV_TYPE_ROOT) == 0);
3533	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3534	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3535	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3536	    &nvtop, 1) == 0);
3537
3538	/*
3539	 * Replace the existing vdev_tree with the new root vdev in
3540	 * this pool's configuration (remove the old, add the new).
3541	 */
3542	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3543	nvlist_free(nvroot);
3544	return (config);
3545}
3546
3547/*
3548 * Walk the vdev tree and see if we can find a device with "better"
3549 * configuration. A configuration is "better" if the label on that
3550 * device has a more recent txg.
3551 */
3552static void
3553spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3554{
3555	for (int c = 0; c < vd->vdev_children; c++)
3556		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3557
3558	if (vd->vdev_ops->vdev_op_leaf) {
3559		nvlist_t *label;
3560		uint64_t label_txg;
3561
3562		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3563		    &label) != 0)
3564			return;
3565
3566		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3567		    &label_txg) == 0);
3568
3569		/*
3570		 * Do we have a better boot device?
3571		 */
3572		if (label_txg > *txg) {
3573			*txg = label_txg;
3574			*avd = vd;
3575		}
3576		nvlist_free(label);
3577	}
3578}
3579
3580/*
3581 * Import a root pool.
3582 *
3583 * For x86. devpath_list will consist of devid and/or physpath name of
3584 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3585 * The GRUB "findroot" command will return the vdev we should boot.
3586 *
3587 * For Sparc, devpath_list consists the physpath name of the booting device
3588 * no matter the rootpool is a single device pool or a mirrored pool.
3589 * e.g.
3590 *	"/pci@1f,0/ide@d/disk@0,0:a"
3591 */
3592int
3593spa_import_rootpool(char *devpath, char *devid)
3594{
3595	spa_t *spa;
3596	vdev_t *rvd, *bvd, *avd = NULL;
3597	nvlist_t *config, *nvtop;
3598	uint64_t guid, txg;
3599	char *pname;
3600	int error;
3601
3602	/*
3603	 * Read the label from the boot device and generate a configuration.
3604	 */
3605	config = spa_generate_rootconf(devpath, devid, &guid);
3606#if defined(_OBP) && defined(_KERNEL)
3607	if (config == NULL) {
3608		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3609			/* iscsi boot */
3610			get_iscsi_bootpath_phy(devpath);
3611			config = spa_generate_rootconf(devpath, devid, &guid);
3612		}
3613	}
3614#endif
3615	if (config == NULL) {
3616		cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3617		    devpath);
3618		return (EIO);
3619	}
3620
3621	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3622	    &pname) == 0);
3623	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3624
3625	mutex_enter(&spa_namespace_lock);
3626	if ((spa = spa_lookup(pname)) != NULL) {
3627		/*
3628		 * Remove the existing root pool from the namespace so that we
3629		 * can replace it with the correct config we just read in.
3630		 */
3631		spa_remove(spa);
3632	}
3633
3634	spa = spa_add(pname, config, NULL);
3635	spa->spa_is_root = B_TRUE;
3636	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3637
3638	/*
3639	 * Build up a vdev tree based on the boot device's label config.
3640	 */
3641	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3642	    &nvtop) == 0);
3643	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3644	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3645	    VDEV_ALLOC_ROOTPOOL);
3646	spa_config_exit(spa, SCL_ALL, FTAG);
3647	if (error) {
3648		mutex_exit(&spa_namespace_lock);
3649		nvlist_free(config);
3650		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3651		    pname);
3652		return (error);
3653	}
3654
3655	/*
3656	 * Get the boot vdev.
3657	 */
3658	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3659		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3660		    (u_longlong_t)guid);
3661		error = ENOENT;
3662		goto out;
3663	}
3664
3665	/*
3666	 * Determine if there is a better boot device.
3667	 */
3668	avd = bvd;
3669	spa_alt_rootvdev(rvd, &avd, &txg);
3670	if (avd != bvd) {
3671		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3672		    "try booting from '%s'", avd->vdev_path);
3673		error = EINVAL;
3674		goto out;
3675	}
3676
3677	/*
3678	 * If the boot device is part of a spare vdev then ensure that
3679	 * we're booting off the active spare.
3680	 */
3681	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3682	    !bvd->vdev_isspare) {
3683		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3684		    "try booting from '%s'",
3685		    bvd->vdev_parent->
3686		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3687		error = EINVAL;
3688		goto out;
3689	}
3690
3691	error = 0;
3692out:
3693	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3694	vdev_free(rvd);
3695	spa_config_exit(spa, SCL_ALL, FTAG);
3696	mutex_exit(&spa_namespace_lock);
3697
3698	nvlist_free(config);
3699	return (error);
3700}
3701
3702#endif
3703
3704/*
3705 * Import a non-root pool into the system.
3706 */
3707int
3708spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3709{
3710	spa_t *spa;
3711	char *altroot = NULL;
3712	spa_load_state_t state = SPA_LOAD_IMPORT;
3713	zpool_rewind_policy_t policy;
3714	uint64_t mode = spa_mode_global;
3715	uint64_t readonly = B_FALSE;
3716	int error;
3717	nvlist_t *nvroot;
3718	nvlist_t **spares, **l2cache;
3719	uint_t nspares, nl2cache;
3720
3721	/*
3722	 * If a pool with this name exists, return failure.
3723	 */
3724	mutex_enter(&spa_namespace_lock);
3725	if (spa_lookup(pool) != NULL) {
3726		mutex_exit(&spa_namespace_lock);
3727		return (EEXIST);
3728	}
3729
3730	/*
3731	 * Create and initialize the spa structure.
3732	 */
3733	(void) nvlist_lookup_string(props,
3734	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3735	(void) nvlist_lookup_uint64(props,
3736	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3737	if (readonly)
3738		mode = FREAD;
3739	spa = spa_add(pool, config, altroot);
3740	spa->spa_import_flags = flags;
3741
3742	/*
3743	 * Verbatim import - Take a pool and insert it into the namespace
3744	 * as if it had been loaded at boot.
3745	 */
3746	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3747		if (props != NULL)
3748			spa_configfile_set(spa, props, B_FALSE);
3749
3750		spa_config_sync(spa, B_FALSE, B_TRUE);
3751
3752		mutex_exit(&spa_namespace_lock);
3753		spa_history_log_version(spa, "import");
3754
3755		return (0);
3756	}
3757
3758	spa_activate(spa, mode);
3759
3760	/*
3761	 * Don't start async tasks until we know everything is healthy.
3762	 */
3763	spa_async_suspend(spa);
3764
3765	zpool_get_rewind_policy(config, &policy);
3766	if (policy.zrp_request & ZPOOL_DO_REWIND)
3767		state = SPA_LOAD_RECOVER;
3768
3769	/*
3770	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3771	 * because the user-supplied config is actually the one to trust when
3772	 * doing an import.
3773	 */
3774	if (state != SPA_LOAD_RECOVER)
3775		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3776
3777	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3778	    policy.zrp_request);
3779
3780	/*
3781	 * Propagate anything learned while loading the pool and pass it
3782	 * back to caller (i.e. rewind info, missing devices, etc).
3783	 */
3784	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3785	    spa->spa_load_info) == 0);
3786
3787	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3788	/*
3789	 * Toss any existing sparelist, as it doesn't have any validity
3790	 * anymore, and conflicts with spa_has_spare().
3791	 */
3792	if (spa->spa_spares.sav_config) {
3793		nvlist_free(spa->spa_spares.sav_config);
3794		spa->spa_spares.sav_config = NULL;
3795		spa_load_spares(spa);
3796	}
3797	if (spa->spa_l2cache.sav_config) {
3798		nvlist_free(spa->spa_l2cache.sav_config);
3799		spa->spa_l2cache.sav_config = NULL;
3800		spa_load_l2cache(spa);
3801	}
3802
3803	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3804	    &nvroot) == 0);
3805	if (error == 0)
3806		error = spa_validate_aux(spa, nvroot, -1ULL,
3807		    VDEV_ALLOC_SPARE);
3808	if (error == 0)
3809		error = spa_validate_aux(spa, nvroot, -1ULL,
3810		    VDEV_ALLOC_L2CACHE);
3811	spa_config_exit(spa, SCL_ALL, FTAG);
3812
3813	if (props != NULL)
3814		spa_configfile_set(spa, props, B_FALSE);
3815
3816	if (error != 0 || (props && spa_writeable(spa) &&
3817	    (error = spa_prop_set(spa, props)))) {
3818		spa_unload(spa);
3819		spa_deactivate(spa);
3820		spa_remove(spa);
3821		mutex_exit(&spa_namespace_lock);
3822		return (error);
3823	}
3824
3825	spa_async_resume(spa);
3826
3827	/*
3828	 * Override any spares and level 2 cache devices as specified by
3829	 * the user, as these may have correct device names/devids, etc.
3830	 */
3831	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3832	    &spares, &nspares) == 0) {
3833		if (spa->spa_spares.sav_config)
3834			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3835			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3836		else
3837			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3838			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3839		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3840		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3841		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3842		spa_load_spares(spa);
3843		spa_config_exit(spa, SCL_ALL, FTAG);
3844		spa->spa_spares.sav_sync = B_TRUE;
3845	}
3846	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3847	    &l2cache, &nl2cache) == 0) {
3848		if (spa->spa_l2cache.sav_config)
3849			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3850			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3851		else
3852			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3853			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3854		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3855		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3856		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3857		spa_load_l2cache(spa);
3858		spa_config_exit(spa, SCL_ALL, FTAG);
3859		spa->spa_l2cache.sav_sync = B_TRUE;
3860	}
3861
3862	/*
3863	 * Check for any removed devices.
3864	 */
3865	if (spa->spa_autoreplace) {
3866		spa_aux_check_removed(&spa->spa_spares);
3867		spa_aux_check_removed(&spa->spa_l2cache);
3868	}
3869
3870	if (spa_writeable(spa)) {
3871		/*
3872		 * Update the config cache to include the newly-imported pool.
3873		 */
3874		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3875	}
3876
3877	/*
3878	 * It's possible that the pool was expanded while it was exported.
3879	 * We kick off an async task to handle this for us.
3880	 */
3881	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3882
3883	mutex_exit(&spa_namespace_lock);
3884	spa_history_log_version(spa, "import");
3885
3886	return (0);
3887}
3888
3889nvlist_t *
3890spa_tryimport(nvlist_t *tryconfig)
3891{
3892	nvlist_t *config = NULL;
3893	char *poolname;
3894	spa_t *spa;
3895	uint64_t state;
3896	int error;
3897
3898	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3899		return (NULL);
3900
3901	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3902		return (NULL);
3903
3904	/*
3905	 * Create and initialize the spa structure.
3906	 */
3907	mutex_enter(&spa_namespace_lock);
3908	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3909	spa_activate(spa, FREAD);
3910
3911	/*
3912	 * Pass off the heavy lifting to spa_load().
3913	 * Pass TRUE for mosconfig because the user-supplied config
3914	 * is actually the one to trust when doing an import.
3915	 */
3916	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3917
3918	/*
3919	 * If 'tryconfig' was at least parsable, return the current config.
3920	 */
3921	if (spa->spa_root_vdev != NULL) {
3922		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3923		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3924		    poolname) == 0);
3925		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3926		    state) == 0);
3927		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3928		    spa->spa_uberblock.ub_timestamp) == 0);
3929		VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3930		    spa->spa_load_info) == 0);
3931
3932		/*
3933		 * If the bootfs property exists on this pool then we
3934		 * copy it out so that external consumers can tell which
3935		 * pools are bootable.
3936		 */
3937		if ((!error || error == EEXIST) && spa->spa_bootfs) {
3938			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3939
3940			/*
3941			 * We have to play games with the name since the
3942			 * pool was opened as TRYIMPORT_NAME.
3943			 */
3944			if (dsl_dsobj_to_dsname(spa_name(spa),
3945			    spa->spa_bootfs, tmpname) == 0) {
3946				char *cp;
3947				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3948
3949				cp = strchr(tmpname, '/');
3950				if (cp == NULL) {
3951					(void) strlcpy(dsname, tmpname,
3952					    MAXPATHLEN);
3953				} else {
3954					(void) snprintf(dsname, MAXPATHLEN,
3955					    "%s/%s", poolname, ++cp);
3956				}
3957				VERIFY(nvlist_add_string(config,
3958				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3959				kmem_free(dsname, MAXPATHLEN);
3960			}
3961			kmem_free(tmpname, MAXPATHLEN);
3962		}
3963
3964		/*
3965		 * Add the list of hot spares and level 2 cache devices.
3966		 */
3967		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3968		spa_add_spares(spa, config);
3969		spa_add_l2cache(spa, config);
3970		spa_config_exit(spa, SCL_CONFIG, FTAG);
3971	}
3972
3973	spa_unload(spa);
3974	spa_deactivate(spa);
3975	spa_remove(spa);
3976	mutex_exit(&spa_namespace_lock);
3977
3978	return (config);
3979}
3980
3981/*
3982 * Pool export/destroy
3983 *
3984 * The act of destroying or exporting a pool is very simple.  We make sure there
3985 * is no more pending I/O and any references to the pool are gone.  Then, we
3986 * update the pool state and sync all the labels to disk, removing the
3987 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3988 * we don't sync the labels or remove the configuration cache.
3989 */
3990static int
3991spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3992    boolean_t force, boolean_t hardforce)
3993{
3994	spa_t *spa;
3995
3996	if (oldconfig)
3997		*oldconfig = NULL;
3998
3999	if (!(spa_mode_global & FWRITE))
4000		return (EROFS);
4001
4002	mutex_enter(&spa_namespace_lock);
4003	if ((spa = spa_lookup(pool)) == NULL) {
4004		mutex_exit(&spa_namespace_lock);
4005		return (ENOENT);
4006	}
4007
4008	/*
4009	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4010	 * reacquire the namespace lock, and see if we can export.
4011	 */
4012	spa_open_ref(spa, FTAG);
4013	mutex_exit(&spa_namespace_lock);
4014	spa_async_suspend(spa);
4015	mutex_enter(&spa_namespace_lock);
4016	spa_close(spa, FTAG);
4017
4018	/*
4019	 * The pool will be in core if it's openable,
4020	 * in which case we can modify its state.
4021	 */
4022	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4023		/*
4024		 * Objsets may be open only because they're dirty, so we
4025		 * have to force it to sync before checking spa_refcnt.
4026		 */
4027		txg_wait_synced(spa->spa_dsl_pool, 0);
4028
4029		/*
4030		 * A pool cannot be exported or destroyed if there are active
4031		 * references.  If we are resetting a pool, allow references by
4032		 * fault injection handlers.
4033		 */
4034		if (!spa_refcount_zero(spa) ||
4035		    (spa->spa_inject_ref != 0 &&
4036		    new_state != POOL_STATE_UNINITIALIZED)) {
4037			spa_async_resume(spa);
4038			mutex_exit(&spa_namespace_lock);
4039			return (EBUSY);
4040		}
4041
4042		/*
4043		 * A pool cannot be exported if it has an active shared spare.
4044		 * This is to prevent other pools stealing the active spare
4045		 * from an exported pool. At user's own will, such pool can
4046		 * be forcedly exported.
4047		 */
4048		if (!force && new_state == POOL_STATE_EXPORTED &&
4049		    spa_has_active_shared_spare(spa)) {
4050			spa_async_resume(spa);
4051			mutex_exit(&spa_namespace_lock);
4052			return (EXDEV);
4053		}
4054
4055		/*
4056		 * We want this to be reflected on every label,
4057		 * so mark them all dirty.  spa_unload() will do the
4058		 * final sync that pushes these changes out.
4059		 */
4060		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4061			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4062			spa->spa_state = new_state;
4063			spa->spa_final_txg = spa_last_synced_txg(spa) +
4064			    TXG_DEFER_SIZE + 1;
4065			vdev_config_dirty(spa->spa_root_vdev);
4066			spa_config_exit(spa, SCL_ALL, FTAG);
4067		}
4068	}
4069
4070	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4071
4072	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4073		spa_unload(spa);
4074		spa_deactivate(spa);
4075	}
4076
4077	if (oldconfig && spa->spa_config)
4078		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4079
4080	if (new_state != POOL_STATE_UNINITIALIZED) {
4081		if (!hardforce)
4082			spa_config_sync(spa, B_TRUE, B_TRUE);
4083		spa_remove(spa);
4084	}
4085	mutex_exit(&spa_namespace_lock);
4086
4087	return (0);
4088}
4089
4090/*
4091 * Destroy a storage pool.
4092 */
4093int
4094spa_destroy(char *pool)
4095{
4096	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4097	    B_FALSE, B_FALSE));
4098}
4099
4100/*
4101 * Export a storage pool.
4102 */
4103int
4104spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4105    boolean_t hardforce)
4106{
4107	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4108	    force, hardforce));
4109}
4110
4111/*
4112 * Similar to spa_export(), this unloads the spa_t without actually removing it
4113 * from the namespace in any way.
4114 */
4115int
4116spa_reset(char *pool)
4117{
4118	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4119	    B_FALSE, B_FALSE));
4120}
4121
4122/*
4123 * ==========================================================================
4124 * Device manipulation
4125 * ==========================================================================
4126 */
4127
4128/*
4129 * Add a device to a storage pool.
4130 */
4131int
4132spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4133{
4134	uint64_t txg, id;
4135	int error;
4136	vdev_t *rvd = spa->spa_root_vdev;
4137	vdev_t *vd, *tvd;
4138	nvlist_t **spares, **l2cache;
4139	uint_t nspares, nl2cache;
4140
4141	ASSERT(spa_writeable(spa));
4142
4143	txg = spa_vdev_enter(spa);
4144
4145	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4146	    VDEV_ALLOC_ADD)) != 0)
4147		return (spa_vdev_exit(spa, NULL, txg, error));
4148
4149	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
4150
4151	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4152	    &nspares) != 0)
4153		nspares = 0;
4154
4155	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4156	    &nl2cache) != 0)
4157		nl2cache = 0;
4158
4159	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4160		return (spa_vdev_exit(spa, vd, txg, EINVAL));
4161
4162	if (vd->vdev_children != 0 &&
4163	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
4164		return (spa_vdev_exit(spa, vd, txg, error));
4165
4166	/*
4167	 * We must validate the spares and l2cache devices after checking the
4168	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
4169	 */
4170	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4171		return (spa_vdev_exit(spa, vd, txg, error));
4172
4173	/*
4174	 * Transfer each new top-level vdev from vd to rvd.
4175	 */
4176	for (int c = 0; c < vd->vdev_children; c++) {
4177
4178		/*
4179		 * Set the vdev id to the first hole, if one exists.
4180		 */
4181		for (id = 0; id < rvd->vdev_children; id++) {
4182			if (rvd->vdev_child[id]->vdev_ishole) {
4183				vdev_free(rvd->vdev_child[id]);
4184				break;
4185			}
4186		}
4187		tvd = vd->vdev_child[c];
4188		vdev_remove_child(vd, tvd);
4189		tvd->vdev_id = id;
4190		vdev_add_child(rvd, tvd);
4191		vdev_config_dirty(tvd);
4192	}
4193
4194	if (nspares != 0) {
4195		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4196		    ZPOOL_CONFIG_SPARES);
4197		spa_load_spares(spa);
4198		spa->spa_spares.sav_sync = B_TRUE;
4199	}
4200
4201	if (nl2cache != 0) {
4202		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4203		    ZPOOL_CONFIG_L2CACHE);
4204		spa_load_l2cache(spa);
4205		spa->spa_l2cache.sav_sync = B_TRUE;
4206	}
4207
4208	/*
4209	 * We have to be careful when adding new vdevs to an existing pool.
4210	 * If other threads start allocating from these vdevs before we
4211	 * sync the config cache, and we lose power, then upon reboot we may
4212	 * fail to open the pool because there are DVAs that the config cache
4213	 * can't translate.  Therefore, we first add the vdevs without
4214	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4215	 * and then let spa_config_update() initialize the new metaslabs.
4216	 *
4217	 * spa_load() checks for added-but-not-initialized vdevs, so that
4218	 * if we lose power at any point in this sequence, the remaining
4219	 * steps will be completed the next time we load the pool.
4220	 */
4221	(void) spa_vdev_exit(spa, vd, txg, 0);
4222
4223	mutex_enter(&spa_namespace_lock);
4224	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4225	mutex_exit(&spa_namespace_lock);
4226
4227	return (0);
4228}
4229
4230/*
4231 * Attach a device to a mirror.  The arguments are the path to any device
4232 * in the mirror, and the nvroot for the new device.  If the path specifies
4233 * a device that is not mirrored, we automatically insert the mirror vdev.
4234 *
4235 * If 'replacing' is specified, the new device is intended to replace the
4236 * existing device; in this case the two devices are made into their own
4237 * mirror using the 'replacing' vdev, which is functionally identical to
4238 * the mirror vdev (it actually reuses all the same ops) but has a few
4239 * extra rules: you can't attach to it after it's been created, and upon
4240 * completion of resilvering, the first disk (the one being replaced)
4241 * is automatically detached.
4242 */
4243int
4244spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4245{
4246	uint64_t txg, dtl_max_txg;
4247	vdev_t *rvd = spa->spa_root_vdev;
4248	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4249	vdev_ops_t *pvops;
4250	char *oldvdpath, *newvdpath;
4251	int newvd_isspare;
4252	int error;
4253
4254	ASSERT(spa_writeable(spa));
4255
4256	txg = spa_vdev_enter(spa);
4257
4258	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4259
4260	if (oldvd == NULL)
4261		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4262
4263	if (!oldvd->vdev_ops->vdev_op_leaf)
4264		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4265
4266	pvd = oldvd->vdev_parent;
4267
4268	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4269	    VDEV_ALLOC_ATTACH)) != 0)
4270		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4271
4272	if (newrootvd->vdev_children != 1)
4273		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4274
4275	newvd = newrootvd->vdev_child[0];
4276
4277	if (!newvd->vdev_ops->vdev_op_leaf)
4278		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4279
4280	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4281		return (spa_vdev_exit(spa, newrootvd, txg, error));
4282
4283	/*
4284	 * Spares can't replace logs
4285	 */
4286	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4287		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4288
4289	if (!replacing) {
4290		/*
4291		 * For attach, the only allowable parent is a mirror or the root
4292		 * vdev.
4293		 */
4294		if (pvd->vdev_ops != &vdev_mirror_ops &&
4295		    pvd->vdev_ops != &vdev_root_ops)
4296			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4297
4298		pvops = &vdev_mirror_ops;
4299	} else {
4300		/*
4301		 * Active hot spares can only be replaced by inactive hot
4302		 * spares.
4303		 */
4304		if (pvd->vdev_ops == &vdev_spare_ops &&
4305		    oldvd->vdev_isspare &&
4306		    !spa_has_spare(spa, newvd->vdev_guid))
4307			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4308
4309		/*
4310		 * If the source is a hot spare, and the parent isn't already a
4311		 * spare, then we want to create a new hot spare.  Otherwise, we
4312		 * want to create a replacing vdev.  The user is not allowed to
4313		 * attach to a spared vdev child unless the 'isspare' state is
4314		 * the same (spare replaces spare, non-spare replaces
4315		 * non-spare).
4316		 */
4317		if (pvd->vdev_ops == &vdev_replacing_ops &&
4318		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4319			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4320		} else if (pvd->vdev_ops == &vdev_spare_ops &&
4321		    newvd->vdev_isspare != oldvd->vdev_isspare) {
4322			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4323		}
4324
4325		if (newvd->vdev_isspare)
4326			pvops = &vdev_spare_ops;
4327		else
4328			pvops = &vdev_replacing_ops;
4329	}
4330
4331	/*
4332	 * Make sure the new device is big enough.
4333	 */
4334	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4335		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4336
4337	/*
4338	 * The new device cannot have a higher alignment requirement
4339	 * than the top-level vdev.
4340	 */
4341	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4342		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4343
4344	/*
4345	 * If this is an in-place replacement, update oldvd's path and devid
4346	 * to make it distinguishable from newvd, and unopenable from now on.
4347	 */
4348	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4349		spa_strfree(oldvd->vdev_path);
4350		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4351		    KM_SLEEP);
4352		(void) sprintf(oldvd->vdev_path, "%s/%s",
4353		    newvd->vdev_path, "old");
4354		if (oldvd->vdev_devid != NULL) {
4355			spa_strfree(oldvd->vdev_devid);
4356			oldvd->vdev_devid = NULL;
4357		}
4358	}
4359
4360	/* mark the device being resilvered */
4361	newvd->vdev_resilvering = B_TRUE;
4362
4363	/*
4364	 * If the parent is not a mirror, or if we're replacing, insert the new
4365	 * mirror/replacing/spare vdev above oldvd.
4366	 */
4367	if (pvd->vdev_ops != pvops)
4368		pvd = vdev_add_parent(oldvd, pvops);
4369
4370	ASSERT(pvd->vdev_top->vdev_parent == rvd);
4371	ASSERT(pvd->vdev_ops == pvops);
4372	ASSERT(oldvd->vdev_parent == pvd);
4373
4374	/*
4375	 * Extract the new device from its root and add it to pvd.
4376	 */
4377	vdev_remove_child(newrootvd, newvd);
4378	newvd->vdev_id = pvd->vdev_children;
4379	newvd->vdev_crtxg = oldvd->vdev_crtxg;
4380	vdev_add_child(pvd, newvd);
4381
4382	tvd = newvd->vdev_top;
4383	ASSERT(pvd->vdev_top == tvd);
4384	ASSERT(tvd->vdev_parent == rvd);
4385
4386	vdev_config_dirty(tvd);
4387
4388	/*
4389	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4390	 * for any dmu_sync-ed blocks.  It will propagate upward when
4391	 * spa_vdev_exit() calls vdev_dtl_reassess().
4392	 */
4393	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4394
4395	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4396	    dtl_max_txg - TXG_INITIAL);
4397
4398	if (newvd->vdev_isspare) {
4399		spa_spare_activate(newvd);
4400		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4401	}
4402
4403	oldvdpath = spa_strdup(oldvd->vdev_path);
4404	newvdpath = spa_strdup(newvd->vdev_path);
4405	newvd_isspare = newvd->vdev_isspare;
4406
4407	/*
4408	 * Mark newvd's DTL dirty in this txg.
4409	 */
4410	vdev_dirty(tvd, VDD_DTL, newvd, txg);
4411
4412	/*
4413	 * Restart the resilver
4414	 */
4415	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4416
4417	/*
4418	 * Commit the config
4419	 */
4420	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4421
4422	spa_history_log_internal(spa, "vdev attach", NULL,
4423	    "%s vdev=%s %s vdev=%s",
4424	    replacing && newvd_isspare ? "spare in" :
4425	    replacing ? "replace" : "attach", newvdpath,
4426	    replacing ? "for" : "to", oldvdpath);
4427
4428	spa_strfree(oldvdpath);
4429	spa_strfree(newvdpath);
4430
4431	if (spa->spa_bootfs)
4432		spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4433
4434	return (0);
4435}
4436
4437/*
4438 * Detach a device from a mirror or replacing vdev.
4439 * If 'replace_done' is specified, only detach if the parent
4440 * is a replacing vdev.
4441 */
4442int
4443spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4444{
4445	uint64_t txg;
4446	int error;
4447	vdev_t *rvd = spa->spa_root_vdev;
4448	vdev_t *vd, *pvd, *cvd, *tvd;
4449	boolean_t unspare = B_FALSE;
4450	uint64_t unspare_guid = 0;
4451	char *vdpath;
4452
4453	ASSERT(spa_writeable(spa));
4454
4455	txg = spa_vdev_enter(spa);
4456
4457	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4458
4459	if (vd == NULL)
4460		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4461
4462	if (!vd->vdev_ops->vdev_op_leaf)
4463		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4464
4465	pvd = vd->vdev_parent;
4466
4467	/*
4468	 * If the parent/child relationship is not as expected, don't do it.
4469	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4470	 * vdev that's replacing B with C.  The user's intent in replacing
4471	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4472	 * the replace by detaching C, the expected behavior is to end up
4473	 * M(A,B).  But suppose that right after deciding to detach C,
4474	 * the replacement of B completes.  We would have M(A,C), and then
4475	 * ask to detach C, which would leave us with just A -- not what
4476	 * the user wanted.  To prevent this, we make sure that the
4477	 * parent/child relationship hasn't changed -- in this example,
4478	 * that C's parent is still the replacing vdev R.
4479	 */
4480	if (pvd->vdev_guid != pguid && pguid != 0)
4481		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4482
4483	/*
4484	 * Only 'replacing' or 'spare' vdevs can be replaced.
4485	 */
4486	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4487	    pvd->vdev_ops != &vdev_spare_ops)
4488		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4489
4490	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4491	    spa_version(spa) >= SPA_VERSION_SPARES);
4492
4493	/*
4494	 * Only mirror, replacing, and spare vdevs support detach.
4495	 */
4496	if (pvd->vdev_ops != &vdev_replacing_ops &&
4497	    pvd->vdev_ops != &vdev_mirror_ops &&
4498	    pvd->vdev_ops != &vdev_spare_ops)
4499		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4500
4501	/*
4502	 * If this device has the only valid copy of some data,
4503	 * we cannot safely detach it.
4504	 */
4505	if (vdev_dtl_required(vd))
4506		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4507
4508	ASSERT(pvd->vdev_children >= 2);
4509
4510	/*
4511	 * If we are detaching the second disk from a replacing vdev, then
4512	 * check to see if we changed the original vdev's path to have "/old"
4513	 * at the end in spa_vdev_attach().  If so, undo that change now.
4514	 */
4515	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4516	    vd->vdev_path != NULL) {
4517		size_t len = strlen(vd->vdev_path);
4518
4519		for (int c = 0; c < pvd->vdev_children; c++) {
4520			cvd = pvd->vdev_child[c];
4521
4522			if (cvd == vd || cvd->vdev_path == NULL)
4523				continue;
4524
4525			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4526			    strcmp(cvd->vdev_path + len, "/old") == 0) {
4527				spa_strfree(cvd->vdev_path);
4528				cvd->vdev_path = spa_strdup(vd->vdev_path);
4529				break;
4530			}
4531		}
4532	}
4533
4534	/*
4535	 * If we are detaching the original disk from a spare, then it implies
4536	 * that the spare should become a real disk, and be removed from the
4537	 * active spare list for the pool.
4538	 */
4539	if (pvd->vdev_ops == &vdev_spare_ops &&
4540	    vd->vdev_id == 0 &&
4541	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4542		unspare = B_TRUE;
4543
4544	/*
4545	 * Erase the disk labels so the disk can be used for other things.
4546	 * This must be done after all other error cases are handled,
4547	 * but before we disembowel vd (so we can still do I/O to it).
4548	 * But if we can't do it, don't treat the error as fatal --
4549	 * it may be that the unwritability of the disk is the reason
4550	 * it's being detached!
4551	 */
4552	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4553
4554	/*
4555	 * Remove vd from its parent and compact the parent's children.
4556	 */
4557	vdev_remove_child(pvd, vd);
4558	vdev_compact_children(pvd);
4559
4560	/*
4561	 * Remember one of the remaining children so we can get tvd below.
4562	 */
4563	cvd = pvd->vdev_child[pvd->vdev_children - 1];
4564
4565	/*
4566	 * If we need to remove the remaining child from the list of hot spares,
4567	 * do it now, marking the vdev as no longer a spare in the process.
4568	 * We must do this before vdev_remove_parent(), because that can
4569	 * change the GUID if it creates a new toplevel GUID.  For a similar
4570	 * reason, we must remove the spare now, in the same txg as the detach;
4571	 * otherwise someone could attach a new sibling, change the GUID, and
4572	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4573	 */
4574	if (unspare) {
4575		ASSERT(cvd->vdev_isspare);
4576		spa_spare_remove(cvd);
4577		unspare_guid = cvd->vdev_guid;
4578		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4579		cvd->vdev_unspare = B_TRUE;
4580	}
4581
4582	/*
4583	 * If the parent mirror/replacing vdev only has one child,
4584	 * the parent is no longer needed.  Remove it from the tree.
4585	 */
4586	if (pvd->vdev_children == 1) {
4587		if (pvd->vdev_ops == &vdev_spare_ops)
4588			cvd->vdev_unspare = B_FALSE;
4589		vdev_remove_parent(cvd);
4590		cvd->vdev_resilvering = B_FALSE;
4591	}
4592
4593
4594	/*
4595	 * We don't set tvd until now because the parent we just removed
4596	 * may have been the previous top-level vdev.
4597	 */
4598	tvd = cvd->vdev_top;
4599	ASSERT(tvd->vdev_parent == rvd);
4600
4601	/*
4602	 * Reevaluate the parent vdev state.
4603	 */
4604	vdev_propagate_state(cvd);
4605
4606	/*
4607	 * If the 'autoexpand' property is set on the pool then automatically
4608	 * try to expand the size of the pool. For example if the device we
4609	 * just detached was smaller than the others, it may be possible to
4610	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4611	 * first so that we can obtain the updated sizes of the leaf vdevs.
4612	 */
4613	if (spa->spa_autoexpand) {
4614		vdev_reopen(tvd);
4615		vdev_expand(tvd, txg);
4616	}
4617
4618	vdev_config_dirty(tvd);
4619
4620	/*
4621	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4622	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4623	 * But first make sure we're not on any *other* txg's DTL list, to
4624	 * prevent vd from being accessed after it's freed.
4625	 */
4626	vdpath = spa_strdup(vd->vdev_path);
4627	for (int t = 0; t < TXG_SIZE; t++)
4628		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4629	vd->vdev_detached = B_TRUE;
4630	vdev_dirty(tvd, VDD_DTL, vd, txg);
4631
4632	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4633
4634	/* hang on to the spa before we release the lock */
4635	spa_open_ref(spa, FTAG);
4636
4637	error = spa_vdev_exit(spa, vd, txg, 0);
4638
4639	spa_history_log_internal(spa, "detach", NULL,
4640	    "vdev=%s", vdpath);
4641	spa_strfree(vdpath);
4642
4643	/*
4644	 * If this was the removal of the original device in a hot spare vdev,
4645	 * then we want to go through and remove the device from the hot spare
4646	 * list of every other pool.
4647	 */
4648	if (unspare) {
4649		spa_t *altspa = NULL;
4650
4651		mutex_enter(&spa_namespace_lock);
4652		while ((altspa = spa_next(altspa)) != NULL) {
4653			if (altspa->spa_state != POOL_STATE_ACTIVE ||
4654			    altspa == spa)
4655				continue;
4656
4657			spa_open_ref(altspa, FTAG);
4658			mutex_exit(&spa_namespace_lock);
4659			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4660			mutex_enter(&spa_namespace_lock);
4661			spa_close(altspa, FTAG);
4662		}
4663		mutex_exit(&spa_namespace_lock);
4664
4665		/* search the rest of the vdevs for spares to remove */
4666		spa_vdev_resilver_done(spa);
4667	}
4668
4669	/* all done with the spa; OK to release */
4670	mutex_enter(&spa_namespace_lock);
4671	spa_close(spa, FTAG);
4672	mutex_exit(&spa_namespace_lock);
4673
4674	return (error);
4675}
4676
4677/*
4678 * Split a set of devices from their mirrors, and create a new pool from them.
4679 */
4680int
4681spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4682    nvlist_t *props, boolean_t exp)
4683{
4684	int error = 0;
4685	uint64_t txg, *glist;
4686	spa_t *newspa;
4687	uint_t c, children, lastlog;
4688	nvlist_t **child, *nvl, *tmp;
4689	dmu_tx_t *tx;
4690	char *altroot = NULL;
4691	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4692	boolean_t activate_slog;
4693
4694	ASSERT(spa_writeable(spa));
4695
4696	txg = spa_vdev_enter(spa);
4697
4698	/* clear the log and flush everything up to now */
4699	activate_slog = spa_passivate_log(spa);
4700	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4701	error = spa_offline_log(spa);
4702	txg = spa_vdev_config_enter(spa);
4703
4704	if (activate_slog)
4705		spa_activate_log(spa);
4706
4707	if (error != 0)
4708		return (spa_vdev_exit(spa, NULL, txg, error));
4709
4710	/* check new spa name before going any further */
4711	if (spa_lookup(newname) != NULL)
4712		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4713
4714	/*
4715	 * scan through all the children to ensure they're all mirrors
4716	 */
4717	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4718	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4719	    &children) != 0)
4720		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4721
4722	/* first, check to ensure we've got the right child count */
4723	rvd = spa->spa_root_vdev;
4724	lastlog = 0;
4725	for (c = 0; c < rvd->vdev_children; c++) {
4726		vdev_t *vd = rvd->vdev_child[c];
4727
4728		/* don't count the holes & logs as children */
4729		if (vd->vdev_islog || vd->vdev_ishole) {
4730			if (lastlog == 0)
4731				lastlog = c;
4732			continue;
4733		}
4734
4735		lastlog = 0;
4736	}
4737	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4738		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4739
4740	/* next, ensure no spare or cache devices are part of the split */
4741	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4742	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4743		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4744
4745	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4746	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4747
4748	/* then, loop over each vdev and validate it */
4749	for (c = 0; c < children; c++) {
4750		uint64_t is_hole = 0;
4751
4752		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4753		    &is_hole);
4754
4755		if (is_hole != 0) {
4756			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4757			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4758				continue;
4759			} else {
4760				error = EINVAL;
4761				break;
4762			}
4763		}
4764
4765		/* which disk is going to be split? */
4766		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4767		    &glist[c]) != 0) {
4768			error = EINVAL;
4769			break;
4770		}
4771
4772		/* look it up in the spa */
4773		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4774		if (vml[c] == NULL) {
4775			error = ENODEV;
4776			break;
4777		}
4778
4779		/* make sure there's nothing stopping the split */
4780		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4781		    vml[c]->vdev_islog ||
4782		    vml[c]->vdev_ishole ||
4783		    vml[c]->vdev_isspare ||
4784		    vml[c]->vdev_isl2cache ||
4785		    !vdev_writeable(vml[c]) ||
4786		    vml[c]->vdev_children != 0 ||
4787		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4788		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4789			error = EINVAL;
4790			break;
4791		}
4792
4793		if (vdev_dtl_required(vml[c])) {
4794			error = EBUSY;
4795			break;
4796		}
4797
4798		/* we need certain info from the top level */
4799		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4800		    vml[c]->vdev_top->vdev_ms_array) == 0);
4801		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4802		    vml[c]->vdev_top->vdev_ms_shift) == 0);
4803		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4804		    vml[c]->vdev_top->vdev_asize) == 0);
4805		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4806		    vml[c]->vdev_top->vdev_ashift) == 0);
4807	}
4808
4809	if (error != 0) {
4810		kmem_free(vml, children * sizeof (vdev_t *));
4811		kmem_free(glist, children * sizeof (uint64_t));
4812		return (spa_vdev_exit(spa, NULL, txg, error));
4813	}
4814
4815	/* stop writers from using the disks */
4816	for (c = 0; c < children; c++) {
4817		if (vml[c] != NULL)
4818			vml[c]->vdev_offline = B_TRUE;
4819	}
4820	vdev_reopen(spa->spa_root_vdev);
4821
4822	/*
4823	 * Temporarily record the splitting vdevs in the spa config.  This
4824	 * will disappear once the config is regenerated.
4825	 */
4826	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4827	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4828	    glist, children) == 0);
4829	kmem_free(glist, children * sizeof (uint64_t));
4830
4831	mutex_enter(&spa->spa_props_lock);
4832	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4833	    nvl) == 0);
4834	mutex_exit(&spa->spa_props_lock);
4835	spa->spa_config_splitting = nvl;
4836	vdev_config_dirty(spa->spa_root_vdev);
4837
4838	/* configure and create the new pool */
4839	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4840	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4841	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4842	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4843	    spa_version(spa)) == 0);
4844	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4845	    spa->spa_config_txg) == 0);
4846	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4847	    spa_generate_guid(NULL)) == 0);
4848	(void) nvlist_lookup_string(props,
4849	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4850
4851	/* add the new pool to the namespace */
4852	newspa = spa_add(newname, config, altroot);
4853	newspa->spa_config_txg = spa->spa_config_txg;
4854	spa_set_log_state(newspa, SPA_LOG_CLEAR);
4855
4856	/* release the spa config lock, retaining the namespace lock */
4857	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4858
4859	if (zio_injection_enabled)
4860		zio_handle_panic_injection(spa, FTAG, 1);
4861
4862	spa_activate(newspa, spa_mode_global);
4863	spa_async_suspend(newspa);
4864
4865	/* create the new pool from the disks of the original pool */
4866	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4867	if (error)
4868		goto out;
4869
4870	/* if that worked, generate a real config for the new pool */
4871	if (newspa->spa_root_vdev != NULL) {
4872		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4873		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4874		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4875		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4876		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4877		    B_TRUE));
4878	}
4879
4880	/* set the props */
4881	if (props != NULL) {
4882		spa_configfile_set(newspa, props, B_FALSE);
4883		error = spa_prop_set(newspa, props);
4884		if (error)
4885			goto out;
4886	}
4887
4888	/* flush everything */
4889	txg = spa_vdev_config_enter(newspa);
4890	vdev_config_dirty(newspa->spa_root_vdev);
4891	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4892
4893	if (zio_injection_enabled)
4894		zio_handle_panic_injection(spa, FTAG, 2);
4895
4896	spa_async_resume(newspa);
4897
4898	/* finally, update the original pool's config */
4899	txg = spa_vdev_config_enter(spa);
4900	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4901	error = dmu_tx_assign(tx, TXG_WAIT);
4902	if (error != 0)
4903		dmu_tx_abort(tx);
4904	for (c = 0; c < children; c++) {
4905		if (vml[c] != NULL) {
4906			vdev_split(vml[c]);
4907			if (error == 0)
4908				spa_history_log_internal(spa, "detach", tx,
4909				    "vdev=%s", vml[c]->vdev_path);
4910			vdev_free(vml[c]);
4911		}
4912	}
4913	vdev_config_dirty(spa->spa_root_vdev);
4914	spa->spa_config_splitting = NULL;
4915	nvlist_free(nvl);
4916	if (error == 0)
4917		dmu_tx_commit(tx);
4918	(void) spa_vdev_exit(spa, NULL, txg, 0);
4919
4920	if (zio_injection_enabled)
4921		zio_handle_panic_injection(spa, FTAG, 3);
4922
4923	/* split is complete; log a history record */
4924	spa_history_log_internal(newspa, "split", NULL,
4925	    "from pool %s", spa_name(spa));
4926
4927	kmem_free(vml, children * sizeof (vdev_t *));
4928
4929	/* if we're not going to mount the filesystems in userland, export */
4930	if (exp)
4931		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4932		    B_FALSE, B_FALSE);
4933
4934	return (error);
4935
4936out:
4937	spa_unload(newspa);
4938	spa_deactivate(newspa);
4939	spa_remove(newspa);
4940
4941	txg = spa_vdev_config_enter(spa);
4942
4943	/* re-online all offlined disks */
4944	for (c = 0; c < children; c++) {
4945		if (vml[c] != NULL)
4946			vml[c]->vdev_offline = B_FALSE;
4947	}
4948	vdev_reopen(spa->spa_root_vdev);
4949
4950	nvlist_free(spa->spa_config_splitting);
4951	spa->spa_config_splitting = NULL;
4952	(void) spa_vdev_exit(spa, NULL, txg, error);
4953
4954	kmem_free(vml, children * sizeof (vdev_t *));
4955	return (error);
4956}
4957
4958static nvlist_t *
4959spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4960{
4961	for (int i = 0; i < count; i++) {
4962		uint64_t guid;
4963
4964		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4965		    &guid) == 0);
4966
4967		if (guid == target_guid)
4968			return (nvpp[i]);
4969	}
4970
4971	return (NULL);
4972}
4973
4974static void
4975spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4976	nvlist_t *dev_to_remove)
4977{
4978	nvlist_t **newdev = NULL;
4979
4980	if (count > 1)
4981		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4982
4983	for (int i = 0, j = 0; i < count; i++) {
4984		if (dev[i] == dev_to_remove)
4985			continue;
4986		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4987	}
4988
4989	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4990	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4991
4992	for (int i = 0; i < count - 1; i++)
4993		nvlist_free(newdev[i]);
4994
4995	if (count > 1)
4996		kmem_free(newdev, (count - 1) * sizeof (void *));
4997}
4998
4999/*
5000 * Evacuate the device.
5001 */
5002static int
5003spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5004{
5005	uint64_t txg;
5006	int error = 0;
5007
5008	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5009	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5010	ASSERT(vd == vd->vdev_top);
5011
5012	/*
5013	 * Evacuate the device.  We don't hold the config lock as writer
5014	 * since we need to do I/O but we do keep the
5015	 * spa_namespace_lock held.  Once this completes the device
5016	 * should no longer have any blocks allocated on it.
5017	 */
5018	if (vd->vdev_islog) {
5019		if (vd->vdev_stat.vs_alloc != 0)
5020			error = spa_offline_log(spa);
5021	} else {
5022		error = ENOTSUP;
5023	}
5024
5025	if (error)
5026		return (error);
5027
5028	/*
5029	 * The evacuation succeeded.  Remove any remaining MOS metadata
5030	 * associated with this vdev, and wait for these changes to sync.
5031	 */
5032	ASSERT0(vd->vdev_stat.vs_alloc);
5033	txg = spa_vdev_config_enter(spa);
5034	vd->vdev_removing = B_TRUE;
5035	vdev_dirty(vd, 0, NULL, txg);
5036	vdev_config_dirty(vd);
5037	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5038
5039	return (0);
5040}
5041
5042/*
5043 * Complete the removal by cleaning up the namespace.
5044 */
5045static void
5046spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5047{
5048	vdev_t *rvd = spa->spa_root_vdev;
5049	uint64_t id = vd->vdev_id;
5050	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5051
5052	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5053	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5054	ASSERT(vd == vd->vdev_top);
5055
5056	/*
5057	 * Only remove any devices which are empty.
5058	 */
5059	if (vd->vdev_stat.vs_alloc != 0)
5060		return;
5061
5062	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5063
5064	if (list_link_active(&vd->vdev_state_dirty_node))
5065		vdev_state_clean(vd);
5066	if (list_link_active(&vd->vdev_config_dirty_node))
5067		vdev_config_clean(vd);
5068
5069	vdev_free(vd);
5070
5071	if (last_vdev) {
5072		vdev_compact_children(rvd);
5073	} else {
5074		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5075		vdev_add_child(rvd, vd);
5076	}
5077	vdev_config_dirty(rvd);
5078
5079	/*
5080	 * Reassess the health of our root vdev.
5081	 */
5082	vdev_reopen(rvd);
5083}
5084
5085/*
5086 * Remove a device from the pool -
5087 *
5088 * Removing a device from the vdev namespace requires several steps
5089 * and can take a significant amount of time.  As a result we use
5090 * the spa_vdev_config_[enter/exit] functions which allow us to
5091 * grab and release the spa_config_lock while still holding the namespace
5092 * lock.  During each step the configuration is synced out.
5093 */
5094
5095/*
5096 * Remove a device from the pool.  Currently, this supports removing only hot
5097 * spares, slogs, and level 2 ARC devices.
5098 */
5099int
5100spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5101{
5102	vdev_t *vd;
5103	metaslab_group_t *mg;
5104	nvlist_t **spares, **l2cache, *nv;
5105	uint64_t txg = 0;
5106	uint_t nspares, nl2cache;
5107	int error = 0;
5108	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5109
5110	ASSERT(spa_writeable(spa));
5111
5112	if (!locked)
5113		txg = spa_vdev_enter(spa);
5114
5115	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5116
5117	if (spa->spa_spares.sav_vdevs != NULL &&
5118	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5119	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5120	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5121		/*
5122		 * Only remove the hot spare if it's not currently in use
5123		 * in this pool.
5124		 */
5125		if (vd == NULL || unspare) {
5126			spa_vdev_remove_aux(spa->spa_spares.sav_config,
5127			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5128			spa_load_spares(spa);
5129			spa->spa_spares.sav_sync = B_TRUE;
5130		} else {
5131			error = EBUSY;
5132		}
5133	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
5134	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5135	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5136	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5137		/*
5138		 * Cache devices can always be removed.
5139		 */
5140		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5141		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5142		spa_load_l2cache(spa);
5143		spa->spa_l2cache.sav_sync = B_TRUE;
5144	} else if (vd != NULL && vd->vdev_islog) {
5145		ASSERT(!locked);
5146		ASSERT(vd == vd->vdev_top);
5147
5148		/*
5149		 * XXX - Once we have bp-rewrite this should
5150		 * become the common case.
5151		 */
5152
5153		mg = vd->vdev_mg;
5154
5155		/*
5156		 * Stop allocating from this vdev.
5157		 */
5158		metaslab_group_passivate(mg);
5159
5160		/*
5161		 * Wait for the youngest allocations and frees to sync,
5162		 * and then wait for the deferral of those frees to finish.
5163		 */
5164		spa_vdev_config_exit(spa, NULL,
5165		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5166
5167		/*
5168		 * Attempt to evacuate the vdev.
5169		 */
5170		error = spa_vdev_remove_evacuate(spa, vd);
5171
5172		txg = spa_vdev_config_enter(spa);
5173
5174		/*
5175		 * If we couldn't evacuate the vdev, unwind.
5176		 */
5177		if (error) {
5178			metaslab_group_activate(mg);
5179			return (spa_vdev_exit(spa, NULL, txg, error));
5180		}
5181
5182		/*
5183		 * Clean up the vdev namespace.
5184		 */
5185		spa_vdev_remove_from_namespace(spa, vd);
5186
5187	} else if (vd != NULL) {
5188		/*
5189		 * Normal vdevs cannot be removed (yet).
5190		 */
5191		error = ENOTSUP;
5192	} else {
5193		/*
5194		 * There is no vdev of any kind with the specified guid.
5195		 */
5196		error = ENOENT;
5197	}
5198
5199	if (!locked)
5200		return (spa_vdev_exit(spa, NULL, txg, error));
5201
5202	return (error);
5203}
5204
5205/*
5206 * Find any device that's done replacing, or a vdev marked 'unspare' that's
5207 * current spared, so we can detach it.
5208 */
5209static vdev_t *
5210spa_vdev_resilver_done_hunt(vdev_t *vd)
5211{
5212	vdev_t *newvd, *oldvd;
5213
5214	for (int c = 0; c < vd->vdev_children; c++) {
5215		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5216		if (oldvd != NULL)
5217			return (oldvd);
5218	}
5219
5220	/*
5221	 * Check for a completed replacement.  We always consider the first
5222	 * vdev in the list to be the oldest vdev, and the last one to be
5223	 * the newest (see spa_vdev_attach() for how that works).  In
5224	 * the case where the newest vdev is faulted, we will not automatically
5225	 * remove it after a resilver completes.  This is OK as it will require
5226	 * user intervention to determine which disk the admin wishes to keep.
5227	 */
5228	if (vd->vdev_ops == &vdev_replacing_ops) {
5229		ASSERT(vd->vdev_children > 1);
5230
5231		newvd = vd->vdev_child[vd->vdev_children - 1];
5232		oldvd = vd->vdev_child[0];
5233
5234		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5235		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5236		    !vdev_dtl_required(oldvd))
5237			return (oldvd);
5238	}
5239
5240	/*
5241	 * Check for a completed resilver with the 'unspare' flag set.
5242	 */
5243	if (vd->vdev_ops == &vdev_spare_ops) {
5244		vdev_t *first = vd->vdev_child[0];
5245		vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5246
5247		if (last->vdev_unspare) {
5248			oldvd = first;
5249			newvd = last;
5250		} else if (first->vdev_unspare) {
5251			oldvd = last;
5252			newvd = first;
5253		} else {
5254			oldvd = NULL;
5255		}
5256
5257		if (oldvd != NULL &&
5258		    vdev_dtl_empty(newvd, DTL_MISSING) &&
5259		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5260		    !vdev_dtl_required(oldvd))
5261			return (oldvd);
5262
5263		/*
5264		 * If there are more than two spares attached to a disk,
5265		 * and those spares are not required, then we want to
5266		 * attempt to free them up now so that they can be used
5267		 * by other pools.  Once we're back down to a single
5268		 * disk+spare, we stop removing them.
5269		 */
5270		if (vd->vdev_children > 2) {
5271			newvd = vd->vdev_child[1];
5272
5273			if (newvd->vdev_isspare && last->vdev_isspare &&
5274			    vdev_dtl_empty(last, DTL_MISSING) &&
5275			    vdev_dtl_empty(last, DTL_OUTAGE) &&
5276			    !vdev_dtl_required(newvd))
5277				return (newvd);
5278		}
5279	}
5280
5281	return (NULL);
5282}
5283
5284static void
5285spa_vdev_resilver_done(spa_t *spa)
5286{
5287	vdev_t *vd, *pvd, *ppvd;
5288	uint64_t guid, sguid, pguid, ppguid;
5289
5290	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5291
5292	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5293		pvd = vd->vdev_parent;
5294		ppvd = pvd->vdev_parent;
5295		guid = vd->vdev_guid;
5296		pguid = pvd->vdev_guid;
5297		ppguid = ppvd->vdev_guid;
5298		sguid = 0;
5299		/*
5300		 * If we have just finished replacing a hot spared device, then
5301		 * we need to detach the parent's first child (the original hot
5302		 * spare) as well.
5303		 */
5304		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5305		    ppvd->vdev_children == 2) {
5306			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5307			sguid = ppvd->vdev_child[1]->vdev_guid;
5308		}
5309		spa_config_exit(spa, SCL_ALL, FTAG);
5310		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5311			return;
5312		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5313			return;
5314		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5315	}
5316
5317	spa_config_exit(spa, SCL_ALL, FTAG);
5318}
5319
5320/*
5321 * Update the stored path or FRU for this vdev.
5322 */
5323int
5324spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5325    boolean_t ispath)
5326{
5327	vdev_t *vd;
5328	boolean_t sync = B_FALSE;
5329
5330	ASSERT(spa_writeable(spa));
5331
5332	spa_vdev_state_enter(spa, SCL_ALL);
5333
5334	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5335		return (spa_vdev_state_exit(spa, NULL, ENOENT));
5336
5337	if (!vd->vdev_ops->vdev_op_leaf)
5338		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5339
5340	if (ispath) {
5341		if (strcmp(value, vd->vdev_path) != 0) {
5342			spa_strfree(vd->vdev_path);
5343			vd->vdev_path = spa_strdup(value);
5344			sync = B_TRUE;
5345		}
5346	} else {
5347		if (vd->vdev_fru == NULL) {
5348			vd->vdev_fru = spa_strdup(value);
5349			sync = B_TRUE;
5350		} else if (strcmp(value, vd->vdev_fru) != 0) {
5351			spa_strfree(vd->vdev_fru);
5352			vd->vdev_fru = spa_strdup(value);
5353			sync = B_TRUE;
5354		}
5355	}
5356
5357	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5358}
5359
5360int
5361spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5362{
5363	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5364}
5365
5366int
5367spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5368{
5369	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5370}
5371
5372/*
5373 * ==========================================================================
5374 * SPA Scanning
5375 * ==========================================================================
5376 */
5377
5378int
5379spa_scan_stop(spa_t *spa)
5380{
5381	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5382	if (dsl_scan_resilvering(spa->spa_dsl_pool))
5383		return (EBUSY);
5384	return (dsl_scan_cancel(spa->spa_dsl_pool));
5385}
5386
5387int
5388spa_scan(spa_t *spa, pool_scan_func_t func)
5389{
5390	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5391
5392	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5393		return (ENOTSUP);
5394
5395	/*
5396	 * If a resilver was requested, but there is no DTL on a
5397	 * writeable leaf device, we have nothing to do.
5398	 */
5399	if (func == POOL_SCAN_RESILVER &&
5400	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5401		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5402		return (0);
5403	}
5404
5405	return (dsl_scan(spa->spa_dsl_pool, func));
5406}
5407
5408/*
5409 * ==========================================================================
5410 * SPA async task processing
5411 * ==========================================================================
5412 */
5413
5414static void
5415spa_async_remove(spa_t *spa, vdev_t *vd)
5416{
5417	if (vd->vdev_remove_wanted) {
5418		vd->vdev_remove_wanted = B_FALSE;
5419		vd->vdev_delayed_close = B_FALSE;
5420		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5421
5422		/*
5423		 * We want to clear the stats, but we don't want to do a full
5424		 * vdev_clear() as that will cause us to throw away
5425		 * degraded/faulted state as well as attempt to reopen the
5426		 * device, all of which is a waste.
5427		 */
5428		vd->vdev_stat.vs_read_errors = 0;
5429		vd->vdev_stat.vs_write_errors = 0;
5430		vd->vdev_stat.vs_checksum_errors = 0;
5431
5432		vdev_state_dirty(vd->vdev_top);
5433	}
5434
5435	for (int c = 0; c < vd->vdev_children; c++)
5436		spa_async_remove(spa, vd->vdev_child[c]);
5437}
5438
5439static void
5440spa_async_probe(spa_t *spa, vdev_t *vd)
5441{
5442	if (vd->vdev_probe_wanted) {
5443		vd->vdev_probe_wanted = B_FALSE;
5444		vdev_reopen(vd);	/* vdev_open() does the actual probe */
5445	}
5446
5447	for (int c = 0; c < vd->vdev_children; c++)
5448		spa_async_probe(spa, vd->vdev_child[c]);
5449}
5450
5451static void
5452spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5453{
5454	sysevent_id_t eid;
5455	nvlist_t *attr;
5456	char *physpath;
5457
5458	if (!spa->spa_autoexpand)
5459		return;
5460
5461	for (int c = 0; c < vd->vdev_children; c++) {
5462		vdev_t *cvd = vd->vdev_child[c];
5463		spa_async_autoexpand(spa, cvd);
5464	}
5465
5466	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5467		return;
5468
5469	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5470	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5471
5472	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5473	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5474
5475	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5476	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5477
5478	nvlist_free(attr);
5479	kmem_free(physpath, MAXPATHLEN);
5480}
5481
5482static void
5483spa_async_thread(spa_t *spa)
5484{
5485	int tasks;
5486
5487	ASSERT(spa->spa_sync_on);
5488
5489	mutex_enter(&spa->spa_async_lock);
5490	tasks = spa->spa_async_tasks;
5491	spa->spa_async_tasks = 0;
5492	mutex_exit(&spa->spa_async_lock);
5493
5494	/*
5495	 * See if the config needs to be updated.
5496	 */
5497	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5498		uint64_t old_space, new_space;
5499
5500		mutex_enter(&spa_namespace_lock);
5501		old_space = metaslab_class_get_space(spa_normal_class(spa));
5502		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5503		new_space = metaslab_class_get_space(spa_normal_class(spa));
5504		mutex_exit(&spa_namespace_lock);
5505
5506		/*
5507		 * If the pool grew as a result of the config update,
5508		 * then log an internal history event.
5509		 */
5510		if (new_space != old_space) {
5511			spa_history_log_internal(spa, "vdev online", NULL,
5512			    "pool '%s' size: %llu(+%llu)",
5513			    spa_name(spa), new_space, new_space - old_space);
5514		}
5515	}
5516
5517	/*
5518	 * See if any devices need to be marked REMOVED.
5519	 */
5520	if (tasks & SPA_ASYNC_REMOVE) {
5521		spa_vdev_state_enter(spa, SCL_NONE);
5522		spa_async_remove(spa, spa->spa_root_vdev);
5523		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5524			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5525		for (int i = 0; i < spa->spa_spares.sav_count; i++)
5526			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5527		(void) spa_vdev_state_exit(spa, NULL, 0);
5528	}
5529
5530	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5531		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5532		spa_async_autoexpand(spa, spa->spa_root_vdev);
5533		spa_config_exit(spa, SCL_CONFIG, FTAG);
5534	}
5535
5536	/*
5537	 * See if any devices need to be probed.
5538	 */
5539	if (tasks & SPA_ASYNC_PROBE) {
5540		spa_vdev_state_enter(spa, SCL_NONE);
5541		spa_async_probe(spa, spa->spa_root_vdev);
5542		(void) spa_vdev_state_exit(spa, NULL, 0);
5543	}
5544
5545	/*
5546	 * If any devices are done replacing, detach them.
5547	 */
5548	if (tasks & SPA_ASYNC_RESILVER_DONE)
5549		spa_vdev_resilver_done(spa);
5550
5551	/*
5552	 * Kick off a resilver.
5553	 */
5554	if (tasks & SPA_ASYNC_RESILVER)
5555		dsl_resilver_restart(spa->spa_dsl_pool, 0);
5556
5557	/*
5558	 * Let the world know that we're done.
5559	 */
5560	mutex_enter(&spa->spa_async_lock);
5561	spa->spa_async_thread = NULL;
5562	cv_broadcast(&spa->spa_async_cv);
5563	mutex_exit(&spa->spa_async_lock);
5564	thread_exit();
5565}
5566
5567void
5568spa_async_suspend(spa_t *spa)
5569{
5570	mutex_enter(&spa->spa_async_lock);
5571	spa->spa_async_suspended++;
5572	while (spa->spa_async_thread != NULL)
5573		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5574	mutex_exit(&spa->spa_async_lock);
5575}
5576
5577void
5578spa_async_resume(spa_t *spa)
5579{
5580	mutex_enter(&spa->spa_async_lock);
5581	ASSERT(spa->spa_async_suspended != 0);
5582	spa->spa_async_suspended--;
5583	mutex_exit(&spa->spa_async_lock);
5584}
5585
5586static void
5587spa_async_dispatch(spa_t *spa)
5588{
5589	mutex_enter(&spa->spa_async_lock);
5590	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5591	    spa->spa_async_thread == NULL &&
5592	    rootdir != NULL && !vn_is_readonly(rootdir))
5593		spa->spa_async_thread = thread_create(NULL, 0,
5594		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5595	mutex_exit(&spa->spa_async_lock);
5596}
5597
5598void
5599spa_async_request(spa_t *spa, int task)
5600{
5601	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5602	mutex_enter(&spa->spa_async_lock);
5603	spa->spa_async_tasks |= task;
5604	mutex_exit(&spa->spa_async_lock);
5605}
5606
5607/*
5608 * ==========================================================================
5609 * SPA syncing routines
5610 * ==========================================================================
5611 */
5612
5613static int
5614bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5615{
5616	bpobj_t *bpo = arg;
5617	bpobj_enqueue(bpo, bp, tx);
5618	return (0);
5619}
5620
5621static int
5622spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5623{
5624	zio_t *zio = arg;
5625
5626	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5627	    zio->io_flags));
5628	return (0);
5629}
5630
5631static void
5632spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5633{
5634	char *packed = NULL;
5635	size_t bufsize;
5636	size_t nvsize = 0;
5637	dmu_buf_t *db;
5638
5639	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5640
5641	/*
5642	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5643	 * information.  This avoids the dbuf_will_dirty() path and
5644	 * saves us a pre-read to get data we don't actually care about.
5645	 */
5646	bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5647	packed = kmem_alloc(bufsize, KM_SLEEP);
5648
5649	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5650	    KM_SLEEP) == 0);
5651	bzero(packed + nvsize, bufsize - nvsize);
5652
5653	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5654
5655	kmem_free(packed, bufsize);
5656
5657	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5658	dmu_buf_will_dirty(db, tx);
5659	*(uint64_t *)db->db_data = nvsize;
5660	dmu_buf_rele(db, FTAG);
5661}
5662
5663static void
5664spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5665    const char *config, const char *entry)
5666{
5667	nvlist_t *nvroot;
5668	nvlist_t **list;
5669	int i;
5670
5671	if (!sav->sav_sync)
5672		return;
5673
5674	/*
5675	 * Update the MOS nvlist describing the list of available devices.
5676	 * spa_validate_aux() will have already made sure this nvlist is
5677	 * valid and the vdevs are labeled appropriately.
5678	 */
5679	if (sav->sav_object == 0) {
5680		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5681		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5682		    sizeof (uint64_t), tx);
5683		VERIFY(zap_update(spa->spa_meta_objset,
5684		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5685		    &sav->sav_object, tx) == 0);
5686	}
5687
5688	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5689	if (sav->sav_count == 0) {
5690		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5691	} else {
5692		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5693		for (i = 0; i < sav->sav_count; i++)
5694			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5695			    B_FALSE, VDEV_CONFIG_L2CACHE);
5696		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5697		    sav->sav_count) == 0);
5698		for (i = 0; i < sav->sav_count; i++)
5699			nvlist_free(list[i]);
5700		kmem_free(list, sav->sav_count * sizeof (void *));
5701	}
5702
5703	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5704	nvlist_free(nvroot);
5705
5706	sav->sav_sync = B_FALSE;
5707}
5708
5709static void
5710spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5711{
5712	nvlist_t *config;
5713
5714	if (list_is_empty(&spa->spa_config_dirty_list))
5715		return;
5716
5717	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5718
5719	config = spa_config_generate(spa, spa->spa_root_vdev,
5720	    dmu_tx_get_txg(tx), B_FALSE);
5721
5722	/*
5723	 * If we're upgrading the spa version then make sure that
5724	 * the config object gets updated with the correct version.
5725	 */
5726	if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5727		fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5728		    spa->spa_uberblock.ub_version);
5729
5730	spa_config_exit(spa, SCL_STATE, FTAG);
5731
5732	if (spa->spa_config_syncing)
5733		nvlist_free(spa->spa_config_syncing);
5734	spa->spa_config_syncing = config;
5735
5736	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5737}
5738
5739static void
5740spa_sync_version(void *arg1, void *arg2, dmu_tx_t *tx)
5741{
5742	spa_t *spa = arg1;
5743	uint64_t version = *(uint64_t *)arg2;
5744
5745	/*
5746	 * Setting the version is special cased when first creating the pool.
5747	 */
5748	ASSERT(tx->tx_txg != TXG_INITIAL);
5749
5750	ASSERT(version <= SPA_VERSION);
5751	ASSERT(version >= spa_version(spa));
5752
5753	spa->spa_uberblock.ub_version = version;
5754	vdev_config_dirty(spa->spa_root_vdev);
5755	spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5756}
5757
5758/*
5759 * Set zpool properties.
5760 */
5761static void
5762spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5763{
5764	spa_t *spa = arg1;
5765	objset_t *mos = spa->spa_meta_objset;
5766	nvlist_t *nvp = arg2;
5767	nvpair_t *elem = NULL;
5768
5769	mutex_enter(&spa->spa_props_lock);
5770
5771	while ((elem = nvlist_next_nvpair(nvp, elem))) {
5772		uint64_t intval;
5773		char *strval, *fname;
5774		zpool_prop_t prop;
5775		const char *propname;
5776		zprop_type_t proptype;
5777		zfeature_info_t *feature;
5778
5779		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5780		case ZPROP_INVAL:
5781			/*
5782			 * We checked this earlier in spa_prop_validate().
5783			 */
5784			ASSERT(zpool_prop_feature(nvpair_name(elem)));
5785
5786			fname = strchr(nvpair_name(elem), '@') + 1;
5787			VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
5788
5789			spa_feature_enable(spa, feature, tx);
5790			spa_history_log_internal(spa, "set", tx,
5791			    "%s=enabled", nvpair_name(elem));
5792			break;
5793
5794		case ZPOOL_PROP_VERSION:
5795			VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5796			/*
5797			 * The version is synced seperatly before other
5798			 * properties and should be correct by now.
5799			 */
5800			ASSERT3U(spa_version(spa), >=, intval);
5801			break;
5802
5803		case ZPOOL_PROP_ALTROOT:
5804			/*
5805			 * 'altroot' is a non-persistent property. It should
5806			 * have been set temporarily at creation or import time.
5807			 */
5808			ASSERT(spa->spa_root != NULL);
5809			break;
5810
5811		case ZPOOL_PROP_READONLY:
5812		case ZPOOL_PROP_CACHEFILE:
5813			/*
5814			 * 'readonly' and 'cachefile' are also non-persisitent
5815			 * properties.
5816			 */
5817			break;
5818		case ZPOOL_PROP_COMMENT:
5819			VERIFY(nvpair_value_string(elem, &strval) == 0);
5820			if (spa->spa_comment != NULL)
5821				spa_strfree(spa->spa_comment);
5822			spa->spa_comment = spa_strdup(strval);
5823			/*
5824			 * We need to dirty the configuration on all the vdevs
5825			 * so that their labels get updated.  It's unnecessary
5826			 * to do this for pool creation since the vdev's
5827			 * configuratoin has already been dirtied.
5828			 */
5829			if (tx->tx_txg != TXG_INITIAL)
5830				vdev_config_dirty(spa->spa_root_vdev);
5831			spa_history_log_internal(spa, "set", tx,
5832			    "%s=%s", nvpair_name(elem), strval);
5833			break;
5834		default:
5835			/*
5836			 * Set pool property values in the poolprops mos object.
5837			 */
5838			if (spa->spa_pool_props_object == 0) {
5839				spa->spa_pool_props_object =
5840				    zap_create_link(mos, DMU_OT_POOL_PROPS,
5841				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5842				    tx);
5843			}
5844
5845			/* normalize the property name */
5846			propname = zpool_prop_to_name(prop);
5847			proptype = zpool_prop_get_type(prop);
5848
5849			if (nvpair_type(elem) == DATA_TYPE_STRING) {
5850				ASSERT(proptype == PROP_TYPE_STRING);
5851				VERIFY(nvpair_value_string(elem, &strval) == 0);
5852				VERIFY(zap_update(mos,
5853				    spa->spa_pool_props_object, propname,
5854				    1, strlen(strval) + 1, strval, tx) == 0);
5855				spa_history_log_internal(spa, "set", tx,
5856				    "%s=%s", nvpair_name(elem), strval);
5857			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5858				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5859
5860				if (proptype == PROP_TYPE_INDEX) {
5861					const char *unused;
5862					VERIFY(zpool_prop_index_to_string(
5863					    prop, intval, &unused) == 0);
5864				}
5865				VERIFY(zap_update(mos,
5866				    spa->spa_pool_props_object, propname,
5867				    8, 1, &intval, tx) == 0);
5868				spa_history_log_internal(spa, "set", tx,
5869				    "%s=%lld", nvpair_name(elem), intval);
5870			} else {
5871				ASSERT(0); /* not allowed */
5872			}
5873
5874			switch (prop) {
5875			case ZPOOL_PROP_DELEGATION:
5876				spa->spa_delegation = intval;
5877				break;
5878			case ZPOOL_PROP_BOOTFS:
5879				spa->spa_bootfs = intval;
5880				break;
5881			case ZPOOL_PROP_FAILUREMODE:
5882				spa->spa_failmode = intval;
5883				break;
5884			case ZPOOL_PROP_AUTOEXPAND:
5885				spa->spa_autoexpand = intval;
5886				if (tx->tx_txg != TXG_INITIAL)
5887					spa_async_request(spa,
5888					    SPA_ASYNC_AUTOEXPAND);
5889				break;
5890			case ZPOOL_PROP_DEDUPDITTO:
5891				spa->spa_dedup_ditto = intval;
5892				break;
5893			default:
5894				break;
5895			}
5896		}
5897
5898	}
5899
5900	mutex_exit(&spa->spa_props_lock);
5901}
5902
5903/*
5904 * Perform one-time upgrade on-disk changes.  spa_version() does not
5905 * reflect the new version this txg, so there must be no changes this
5906 * txg to anything that the upgrade code depends on after it executes.
5907 * Therefore this must be called after dsl_pool_sync() does the sync
5908 * tasks.
5909 */
5910static void
5911spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5912{
5913	dsl_pool_t *dp = spa->spa_dsl_pool;
5914
5915	ASSERT(spa->spa_sync_pass == 1);
5916
5917	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5918	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5919		dsl_pool_create_origin(dp, tx);
5920
5921		/* Keeping the origin open increases spa_minref */
5922		spa->spa_minref += 3;
5923	}
5924
5925	if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5926	    spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5927		dsl_pool_upgrade_clones(dp, tx);
5928	}
5929
5930	if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5931	    spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5932		dsl_pool_upgrade_dir_clones(dp, tx);
5933
5934		/* Keeping the freedir open increases spa_minref */
5935		spa->spa_minref += 3;
5936	}
5937
5938	if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
5939	    spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
5940		spa_feature_create_zap_objects(spa, tx);
5941	}
5942}
5943
5944/*
5945 * Sync the specified transaction group.  New blocks may be dirtied as
5946 * part of the process, so we iterate until it converges.
5947 */
5948void
5949spa_sync(spa_t *spa, uint64_t txg)
5950{
5951	dsl_pool_t *dp = spa->spa_dsl_pool;
5952	objset_t *mos = spa->spa_meta_objset;
5953	bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5954	bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5955	vdev_t *rvd = spa->spa_root_vdev;
5956	vdev_t *vd;
5957	dmu_tx_t *tx;
5958	int error;
5959
5960	VERIFY(spa_writeable(spa));
5961
5962	/*
5963	 * Lock out configuration changes.
5964	 */
5965	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5966
5967	spa->spa_syncing_txg = txg;
5968	spa->spa_sync_pass = 0;
5969
5970	/*
5971	 * If there are any pending vdev state changes, convert them
5972	 * into config changes that go out with this transaction group.
5973	 */
5974	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5975	while (list_head(&spa->spa_state_dirty_list) != NULL) {
5976		/*
5977		 * We need the write lock here because, for aux vdevs,
5978		 * calling vdev_config_dirty() modifies sav_config.
5979		 * This is ugly and will become unnecessary when we
5980		 * eliminate the aux vdev wart by integrating all vdevs
5981		 * into the root vdev tree.
5982		 */
5983		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5984		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5985		while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5986			vdev_state_clean(vd);
5987			vdev_config_dirty(vd);
5988		}
5989		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5990		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5991	}
5992	spa_config_exit(spa, SCL_STATE, FTAG);
5993
5994	tx = dmu_tx_create_assigned(dp, txg);
5995
5996	spa->spa_sync_starttime = gethrtime();
5997	VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
5998	    spa->spa_sync_starttime + spa->spa_deadman_synctime));
5999
6000	/*
6001	 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6002	 * set spa_deflate if we have no raid-z vdevs.
6003	 */
6004	if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6005	    spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6006		int i;
6007
6008		for (i = 0; i < rvd->vdev_children; i++) {
6009			vd = rvd->vdev_child[i];
6010			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6011				break;
6012		}
6013		if (i == rvd->vdev_children) {
6014			spa->spa_deflate = TRUE;
6015			VERIFY(0 == zap_add(spa->spa_meta_objset,
6016			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6017			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6018		}
6019	}
6020
6021	/*
6022	 * If anything has changed in this txg, or if someone is waiting
6023	 * for this txg to sync (eg, spa_vdev_remove()), push the
6024	 * deferred frees from the previous txg.  If not, leave them
6025	 * alone so that we don't generate work on an otherwise idle
6026	 * system.
6027	 */
6028	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6029	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6030	    !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6031	    ((dsl_scan_active(dp->dp_scan) ||
6032	    txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6033		zio_t *zio = zio_root(spa, NULL, NULL, 0);
6034		VERIFY3U(bpobj_iterate(defer_bpo,
6035		    spa_free_sync_cb, zio, tx), ==, 0);
6036		VERIFY0(zio_wait(zio));
6037	}
6038
6039	/*
6040	 * Iterate to convergence.
6041	 */
6042	do {
6043		int pass = ++spa->spa_sync_pass;
6044
6045		spa_sync_config_object(spa, tx);
6046		spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6047		    ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6048		spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6049		    ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6050		spa_errlog_sync(spa, txg);
6051		dsl_pool_sync(dp, txg);
6052
6053		if (pass < zfs_sync_pass_deferred_free) {
6054			zio_t *zio = zio_root(spa, NULL, NULL, 0);
6055			bplist_iterate(free_bpl, spa_free_sync_cb,
6056			    zio, tx);
6057			VERIFY(zio_wait(zio) == 0);
6058		} else {
6059			bplist_iterate(free_bpl, bpobj_enqueue_cb