spa.c revision 4263d13f00c9691fa14620eff82abef795be0693
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
66#ifdef	_KERNEL
67#include <sys/bootprops.h>
68#include <sys/callb.h>
69#include <sys/cpupart.h>
70#include <sys/pool.h>
71#include <sys/sysdc.h>
72#include <sys/zone.h>
73#endif	/* _KERNEL */
74
75#include "zfs_prop.h"
76#include "zfs_comutil.h"
77
78typedef enum zti_modes {
79	zti_mode_fixed,			/* value is # of threads (min 1) */
80	zti_mode_online_percent,	/* value is % of online CPUs */
81	zti_mode_batch,			/* cpu-intensive; value is ignored */
82	zti_mode_null,			/* don't create a taskq */
83	zti_nmodes
84} zti_modes_t;
85
86#define	ZTI_FIX(n)	{ zti_mode_fixed, (n) }
87#define	ZTI_PCT(n)	{ zti_mode_online_percent, (n) }
88#define	ZTI_BATCH	{ zti_mode_batch, 0 }
89#define	ZTI_NULL	{ zti_mode_null, 0 }
90
91#define	ZTI_ONE		ZTI_FIX(1)
92
93typedef struct zio_taskq_info {
94	enum zti_modes zti_mode;
95	uint_t zti_value;
96} zio_taskq_info_t;
97
98static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
99	"issue", "issue_high", "intr", "intr_high"
100};
101
102/*
103 * Define the taskq threads for the following I/O types:
104 * 	NULL, READ, WRITE, FREE, CLAIM, and IOCTL
105 */
106const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
107	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
108	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
109	{ ZTI_FIX(8),	ZTI_NULL,	ZTI_BATCH,	ZTI_NULL },
110	{ ZTI_BATCH,	ZTI_FIX(5),	ZTI_FIX(8),	ZTI_FIX(5) },
111	{ ZTI_FIX(100),	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
112	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
113	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
114};
115
116static dsl_syncfunc_t spa_sync_props;
117static boolean_t spa_has_active_shared_spare(spa_t *spa);
118static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
119    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
120    char **ereport);
121static void spa_vdev_resilver_done(spa_t *spa);
122
123uint_t		zio_taskq_batch_pct = 100;	/* 1 thread per cpu in pset */
124id_t		zio_taskq_psrset_bind = PS_NONE;
125boolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
126uint_t		zio_taskq_basedc = 80;		/* base duty cycle */
127
128boolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
129
130/*
131 * This (illegal) pool name is used when temporarily importing a spa_t in order
132 * to get the vdev stats associated with the imported devices.
133 */
134#define	TRYIMPORT_NAME	"$import"
135
136/*
137 * ==========================================================================
138 * SPA properties routines
139 * ==========================================================================
140 */
141
142/*
143 * Add a (source=src, propname=propval) list to an nvlist.
144 */
145static void
146spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
147    uint64_t intval, zprop_source_t src)
148{
149	const char *propname = zpool_prop_to_name(prop);
150	nvlist_t *propval;
151
152	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
153	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
154
155	if (strval != NULL)
156		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
157	else
158		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
159
160	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
161	nvlist_free(propval);
162}
163
164/*
165 * Get property values from the spa configuration.
166 */
167static void
168spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
169{
170	vdev_t *rvd = spa->spa_root_vdev;
171	uint64_t size;
172	uint64_t alloc;
173	uint64_t space;
174	uint64_t cap, version;
175	zprop_source_t src = ZPROP_SRC_NONE;
176	spa_config_dirent_t *dp;
177
178	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
179
180	if (rvd != NULL) {
181		alloc = metaslab_class_get_alloc(spa_normal_class(spa));
182		size = metaslab_class_get_space(spa_normal_class(spa));
183		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
184		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
185		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
186		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
187		    size - alloc, src);
188
189		space = 0;
190		for (int c = 0; c < rvd->vdev_children; c++) {
191			vdev_t *tvd = rvd->vdev_child[c];
192			space += tvd->vdev_max_asize - tvd->vdev_asize;
193		}
194		spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
195		    src);
196
197		spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
198		    (spa_mode(spa) == FREAD), src);
199
200		cap = (size == 0) ? 0 : (alloc * 100 / size);
201		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
202
203		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
204		    ddt_get_pool_dedup_ratio(spa), src);
205
206		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
207		    rvd->vdev_state, src);
208
209		version = spa_version(spa);
210		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
211			src = ZPROP_SRC_DEFAULT;
212		else
213			src = ZPROP_SRC_LOCAL;
214		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
215	}
216
217	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
218
219	if (spa->spa_comment != NULL) {
220		spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
221		    0, ZPROP_SRC_LOCAL);
222	}
223
224	if (spa->spa_root != NULL)
225		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
226		    0, ZPROP_SRC_LOCAL);
227
228	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
229		if (dp->scd_path == NULL) {
230			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
231			    "none", 0, ZPROP_SRC_LOCAL);
232		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
233			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
234			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
235		}
236	}
237}
238
239/*
240 * Get zpool property values.
241 */
242int
243spa_prop_get(spa_t *spa, nvlist_t **nvp)
244{
245	objset_t *mos = spa->spa_meta_objset;
246	zap_cursor_t zc;
247	zap_attribute_t za;
248	int err;
249
250	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
251
252	mutex_enter(&spa->spa_props_lock);
253
254	/*
255	 * Get properties from the spa config.
256	 */
257	spa_prop_get_config(spa, nvp);
258
259	/* If no pool property object, no more prop to get. */
260	if (mos == NULL || spa->spa_pool_props_object == 0) {
261		mutex_exit(&spa->spa_props_lock);
262		return (0);
263	}
264
265	/*
266	 * Get properties from the MOS pool property object.
267	 */
268	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
269	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
270	    zap_cursor_advance(&zc)) {
271		uint64_t intval = 0;
272		char *strval = NULL;
273		zprop_source_t src = ZPROP_SRC_DEFAULT;
274		zpool_prop_t prop;
275
276		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
277			continue;
278
279		switch (za.za_integer_length) {
280		case 8:
281			/* integer property */
282			if (za.za_first_integer !=
283			    zpool_prop_default_numeric(prop))
284				src = ZPROP_SRC_LOCAL;
285
286			if (prop == ZPOOL_PROP_BOOTFS) {
287				dsl_pool_t *dp;
288				dsl_dataset_t *ds = NULL;
289
290				dp = spa_get_dsl(spa);
291				rw_enter(&dp->dp_config_rwlock, RW_READER);
292				if (err = dsl_dataset_hold_obj(dp,
293				    za.za_first_integer, FTAG, &ds)) {
294					rw_exit(&dp->dp_config_rwlock);
295					break;
296				}
297
298				strval = kmem_alloc(
299				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
300				    KM_SLEEP);
301				dsl_dataset_name(ds, strval);
302				dsl_dataset_rele(ds, FTAG);
303				rw_exit(&dp->dp_config_rwlock);
304			} else {
305				strval = NULL;
306				intval = za.za_first_integer;
307			}
308
309			spa_prop_add_list(*nvp, prop, strval, intval, src);
310
311			if (strval != NULL)
312				kmem_free(strval,
313				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
314
315			break;
316
317		case 1:
318			/* string property */
319			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
320			err = zap_lookup(mos, spa->spa_pool_props_object,
321			    za.za_name, 1, za.za_num_integers, strval);
322			if (err) {
323				kmem_free(strval, za.za_num_integers);
324				break;
325			}
326			spa_prop_add_list(*nvp, prop, strval, 0, src);
327			kmem_free(strval, za.za_num_integers);
328			break;
329
330		default:
331			break;
332		}
333	}
334	zap_cursor_fini(&zc);
335	mutex_exit(&spa->spa_props_lock);
336out:
337	if (err && err != ENOENT) {
338		nvlist_free(*nvp);
339		*nvp = NULL;
340		return (err);
341	}
342
343	return (0);
344}
345
346/*
347 * Validate the given pool properties nvlist and modify the list
348 * for the property values to be set.
349 */
350static int
351spa_prop_validate(spa_t *spa, nvlist_t *props)
352{
353	nvpair_t *elem;
354	int error = 0, reset_bootfs = 0;
355	uint64_t objnum;
356
357	elem = NULL;
358	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
359		zpool_prop_t prop;
360		char *propname, *strval;
361		uint64_t intval;
362		objset_t *os;
363		char *slash, *check;
364
365		propname = nvpair_name(elem);
366
367		if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
368			return (EINVAL);
369
370		switch (prop) {
371		case ZPOOL_PROP_VERSION:
372			error = nvpair_value_uint64(elem, &intval);
373			if (!error &&
374			    (intval < spa_version(spa) || intval > SPA_VERSION))
375				error = EINVAL;
376			break;
377
378		case ZPOOL_PROP_DELEGATION:
379		case ZPOOL_PROP_AUTOREPLACE:
380		case ZPOOL_PROP_LISTSNAPS:
381		case ZPOOL_PROP_AUTOEXPAND:
382			error = nvpair_value_uint64(elem, &intval);
383			if (!error && intval > 1)
384				error = EINVAL;
385			break;
386
387		case ZPOOL_PROP_BOOTFS:
388			/*
389			 * If the pool version is less than SPA_VERSION_BOOTFS,
390			 * or the pool is still being created (version == 0),
391			 * the bootfs property cannot be set.
392			 */
393			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
394				error = ENOTSUP;
395				break;
396			}
397
398			/*
399			 * Make sure the vdev config is bootable
400			 */
401			if (!vdev_is_bootable(spa->spa_root_vdev)) {
402				error = ENOTSUP;
403				break;
404			}
405
406			reset_bootfs = 1;
407
408			error = nvpair_value_string(elem, &strval);
409
410			if (!error) {
411				uint64_t compress;
412
413				if (strval == NULL || strval[0] == '\0') {
414					objnum = zpool_prop_default_numeric(
415					    ZPOOL_PROP_BOOTFS);
416					break;
417				}
418
419				if (error = dmu_objset_hold(strval, FTAG, &os))
420					break;
421
422				/* Must be ZPL and not gzip compressed. */
423
424				if (dmu_objset_type(os) != DMU_OST_ZFS) {
425					error = ENOTSUP;
426				} else if ((error = dsl_prop_get_integer(strval,
427				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
428				    &compress, NULL)) == 0 &&
429				    !BOOTFS_COMPRESS_VALID(compress)) {
430					error = ENOTSUP;
431				} else {
432					objnum = dmu_objset_id(os);
433				}
434				dmu_objset_rele(os, FTAG);
435			}
436			break;
437
438		case ZPOOL_PROP_FAILUREMODE:
439			error = nvpair_value_uint64(elem, &intval);
440			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
441			    intval > ZIO_FAILURE_MODE_PANIC))
442				error = EINVAL;
443
444			/*
445			 * This is a special case which only occurs when
446			 * the pool has completely failed. This allows
447			 * the user to change the in-core failmode property
448			 * without syncing it out to disk (I/Os might
449			 * currently be blocked). We do this by returning
450			 * EIO to the caller (spa_prop_set) to trick it
451			 * into thinking we encountered a property validation
452			 * error.
453			 */
454			if (!error && spa_suspended(spa)) {
455				spa->spa_failmode = intval;
456				error = EIO;
457			}
458			break;
459
460		case ZPOOL_PROP_CACHEFILE:
461			if ((error = nvpair_value_string(elem, &strval)) != 0)
462				break;
463
464			if (strval[0] == '\0')
465				break;
466
467			if (strcmp(strval, "none") == 0)
468				break;
469
470			if (strval[0] != '/') {
471				error = EINVAL;
472				break;
473			}
474
475			slash = strrchr(strval, '/');
476			ASSERT(slash != NULL);
477
478			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
479			    strcmp(slash, "/..") == 0)
480				error = EINVAL;
481			break;
482
483		case ZPOOL_PROP_COMMENT:
484			if ((error = nvpair_value_string(elem, &strval)) != 0)
485				break;
486			for (check = strval; *check != '\0'; check++) {
487				/*
488				 * The kernel doesn't have an easy isprint()
489				 * check.  For this kernel check, we merely
490				 * check ASCII apart from DEL.  Fix this if
491				 * there is an easy-to-use kernel isprint().
492				 */
493				if (*check >= 0x7f) {
494					error = EINVAL;
495					break;
496				}
497				check++;
498			}
499			if (strlen(strval) > ZPROP_MAX_COMMENT)
500				error = E2BIG;
501			break;
502
503		case ZPOOL_PROP_DEDUPDITTO:
504			if (spa_version(spa) < SPA_VERSION_DEDUP)
505				error = ENOTSUP;
506			else
507				error = nvpair_value_uint64(elem, &intval);
508			if (error == 0 &&
509			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
510				error = EINVAL;
511			break;
512		}
513
514		if (error)
515			break;
516	}
517
518	if (!error && reset_bootfs) {
519		error = nvlist_remove(props,
520		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
521
522		if (!error) {
523			error = nvlist_add_uint64(props,
524			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
525		}
526	}
527
528	return (error);
529}
530
531void
532spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
533{
534	char *cachefile;
535	spa_config_dirent_t *dp;
536
537	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
538	    &cachefile) != 0)
539		return;
540
541	dp = kmem_alloc(sizeof (spa_config_dirent_t),
542	    KM_SLEEP);
543
544	if (cachefile[0] == '\0')
545		dp->scd_path = spa_strdup(spa_config_path);
546	else if (strcmp(cachefile, "none") == 0)
547		dp->scd_path = NULL;
548	else
549		dp->scd_path = spa_strdup(cachefile);
550
551	list_insert_head(&spa->spa_config_list, dp);
552	if (need_sync)
553		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
554}
555
556int
557spa_prop_set(spa_t *spa, nvlist_t *nvp)
558{
559	int error;
560	nvpair_t *elem;
561	boolean_t need_sync = B_FALSE;
562	zpool_prop_t prop;
563
564	if ((error = spa_prop_validate(spa, nvp)) != 0)
565		return (error);
566
567	elem = NULL;
568	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
569		if ((prop = zpool_name_to_prop(
570		    nvpair_name(elem))) == ZPROP_INVAL)
571			return (EINVAL);
572
573		if (prop == ZPOOL_PROP_CACHEFILE ||
574		    prop == ZPOOL_PROP_ALTROOT ||
575		    prop == ZPOOL_PROP_READONLY)
576			continue;
577
578		need_sync = B_TRUE;
579		break;
580	}
581
582	if (need_sync)
583		return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
584		    spa, nvp, 3));
585	else
586		return (0);
587}
588
589/*
590 * If the bootfs property value is dsobj, clear it.
591 */
592void
593spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
594{
595	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
596		VERIFY(zap_remove(spa->spa_meta_objset,
597		    spa->spa_pool_props_object,
598		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
599		spa->spa_bootfs = 0;
600	}
601}
602
603/*
604 * Change the GUID for the pool.  This is done so that we can later
605 * re-import a pool built from a clone of our own vdevs.  We will modify
606 * the root vdev's guid, our own pool guid, and then mark all of our
607 * vdevs dirty.  Note that we must make sure that all our vdevs are
608 * online when we do this, or else any vdevs that weren't present
609 * would be orphaned from our pool.  We are also going to issue a
610 * sysevent to update any watchers.
611 */
612int
613spa_change_guid(spa_t *spa)
614{
615	uint64_t	oldguid, newguid;
616	uint64_t	txg;
617
618	if (!(spa_mode_global & FWRITE))
619		return (EROFS);
620
621	txg = spa_vdev_enter(spa);
622
623	if (spa->spa_root_vdev->vdev_state != VDEV_STATE_HEALTHY)
624		return (spa_vdev_exit(spa, NULL, txg, ENXIO));
625
626	oldguid = spa_guid(spa);
627	newguid = spa_generate_guid(NULL);
628	ASSERT3U(oldguid, !=, newguid);
629
630	spa->spa_root_vdev->vdev_guid = newguid;
631	spa->spa_root_vdev->vdev_guid_sum += (newguid - oldguid);
632
633	vdev_config_dirty(spa->spa_root_vdev);
634
635	spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
636
637	return (spa_vdev_exit(spa, NULL, txg, 0));
638}
639
640/*
641 * ==========================================================================
642 * SPA state manipulation (open/create/destroy/import/export)
643 * ==========================================================================
644 */
645
646static int
647spa_error_entry_compare(const void *a, const void *b)
648{
649	spa_error_entry_t *sa = (spa_error_entry_t *)a;
650	spa_error_entry_t *sb = (spa_error_entry_t *)b;
651	int ret;
652
653	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
654	    sizeof (zbookmark_t));
655
656	if (ret < 0)
657		return (-1);
658	else if (ret > 0)
659		return (1);
660	else
661		return (0);
662}
663
664/*
665 * Utility function which retrieves copies of the current logs and
666 * re-initializes them in the process.
667 */
668void
669spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
670{
671	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
672
673	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
674	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
675
676	avl_create(&spa->spa_errlist_scrub,
677	    spa_error_entry_compare, sizeof (spa_error_entry_t),
678	    offsetof(spa_error_entry_t, se_avl));
679	avl_create(&spa->spa_errlist_last,
680	    spa_error_entry_compare, sizeof (spa_error_entry_t),
681	    offsetof(spa_error_entry_t, se_avl));
682}
683
684static taskq_t *
685spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
686    uint_t value)
687{
688	uint_t flags = 0;
689	boolean_t batch = B_FALSE;
690
691	switch (mode) {
692	case zti_mode_null:
693		return (NULL);		/* no taskq needed */
694
695	case zti_mode_fixed:
696		ASSERT3U(value, >=, 1);
697		value = MAX(value, 1);
698		break;
699
700	case zti_mode_batch:
701		batch = B_TRUE;
702		flags |= TASKQ_THREADS_CPU_PCT;
703		value = zio_taskq_batch_pct;
704		break;
705
706	case zti_mode_online_percent:
707		flags |= TASKQ_THREADS_CPU_PCT;
708		break;
709
710	default:
711		panic("unrecognized mode for %s taskq (%u:%u) in "
712		    "spa_activate()",
713		    name, mode, value);
714		break;
715	}
716
717	if (zio_taskq_sysdc && spa->spa_proc != &p0) {
718		if (batch)
719			flags |= TASKQ_DC_BATCH;
720
721		return (taskq_create_sysdc(name, value, 50, INT_MAX,
722		    spa->spa_proc, zio_taskq_basedc, flags));
723	}
724	return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
725	    spa->spa_proc, flags));
726}
727
728static void
729spa_create_zio_taskqs(spa_t *spa)
730{
731	for (int t = 0; t < ZIO_TYPES; t++) {
732		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
733			const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
734			enum zti_modes mode = ztip->zti_mode;
735			uint_t value = ztip->zti_value;
736			char name[32];
737
738			(void) snprintf(name, sizeof (name),
739			    "%s_%s", zio_type_name[t], zio_taskq_types[q]);
740
741			spa->spa_zio_taskq[t][q] =
742			    spa_taskq_create(spa, name, mode, value);
743		}
744	}
745}
746
747#ifdef _KERNEL
748static void
749spa_thread(void *arg)
750{
751	callb_cpr_t cprinfo;
752
753	spa_t *spa = arg;
754	user_t *pu = PTOU(curproc);
755
756	CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
757	    spa->spa_name);
758
759	ASSERT(curproc != &p0);
760	(void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
761	    "zpool-%s", spa->spa_name);
762	(void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
763
764	/* bind this thread to the requested psrset */
765	if (zio_taskq_psrset_bind != PS_NONE) {
766		pool_lock();
767		mutex_enter(&cpu_lock);
768		mutex_enter(&pidlock);
769		mutex_enter(&curproc->p_lock);
770
771		if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
772		    0, NULL, NULL) == 0)  {
773			curthread->t_bind_pset = zio_taskq_psrset_bind;
774		} else {
775			cmn_err(CE_WARN,
776			    "Couldn't bind process for zfs pool \"%s\" to "
777			    "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
778		}
779
780		mutex_exit(&curproc->p_lock);
781		mutex_exit(&pidlock);
782		mutex_exit(&cpu_lock);
783		pool_unlock();
784	}
785
786	if (zio_taskq_sysdc) {
787		sysdc_thread_enter(curthread, 100, 0);
788	}
789
790	spa->spa_proc = curproc;
791	spa->spa_did = curthread->t_did;
792
793	spa_create_zio_taskqs(spa);
794
795	mutex_enter(&spa->spa_proc_lock);
796	ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
797
798	spa->spa_proc_state = SPA_PROC_ACTIVE;
799	cv_broadcast(&spa->spa_proc_cv);
800
801	CALLB_CPR_SAFE_BEGIN(&cprinfo);
802	while (spa->spa_proc_state == SPA_PROC_ACTIVE)
803		cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
804	CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
805
806	ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
807	spa->spa_proc_state = SPA_PROC_GONE;
808	spa->spa_proc = &p0;
809	cv_broadcast(&spa->spa_proc_cv);
810	CALLB_CPR_EXIT(&cprinfo);	/* drops spa_proc_lock */
811
812	mutex_enter(&curproc->p_lock);
813	lwp_exit();
814}
815#endif
816
817/*
818 * Activate an uninitialized pool.
819 */
820static void
821spa_activate(spa_t *spa, int mode)
822{
823	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
824
825	spa->spa_state = POOL_STATE_ACTIVE;
826	spa->spa_mode = mode;
827
828	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
829	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
830
831	/* Try to create a covering process */
832	mutex_enter(&spa->spa_proc_lock);
833	ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
834	ASSERT(spa->spa_proc == &p0);
835	spa->spa_did = 0;
836
837	/* Only create a process if we're going to be around a while. */
838	if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
839		if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
840		    NULL, 0) == 0) {
841			spa->spa_proc_state = SPA_PROC_CREATED;
842			while (spa->spa_proc_state == SPA_PROC_CREATED) {
843				cv_wait(&spa->spa_proc_cv,
844				    &spa->spa_proc_lock);
845			}
846			ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
847			ASSERT(spa->spa_proc != &p0);
848			ASSERT(spa->spa_did != 0);
849		} else {
850#ifdef _KERNEL
851			cmn_err(CE_WARN,
852			    "Couldn't create process for zfs pool \"%s\"\n",
853			    spa->spa_name);
854#endif
855		}
856	}
857	mutex_exit(&spa->spa_proc_lock);
858
859	/* If we didn't create a process, we need to create our taskqs. */
860	if (spa->spa_proc == &p0) {
861		spa_create_zio_taskqs(spa);
862	}
863
864	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
865	    offsetof(vdev_t, vdev_config_dirty_node));
866	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
867	    offsetof(vdev_t, vdev_state_dirty_node));
868
869	txg_list_create(&spa->spa_vdev_txg_list,
870	    offsetof(struct vdev, vdev_txg_node));
871
872	avl_create(&spa->spa_errlist_scrub,
873	    spa_error_entry_compare, sizeof (spa_error_entry_t),
874	    offsetof(spa_error_entry_t, se_avl));
875	avl_create(&spa->spa_errlist_last,
876	    spa_error_entry_compare, sizeof (spa_error_entry_t),
877	    offsetof(spa_error_entry_t, se_avl));
878}
879
880/*
881 * Opposite of spa_activate().
882 */
883static void
884spa_deactivate(spa_t *spa)
885{
886	ASSERT(spa->spa_sync_on == B_FALSE);
887	ASSERT(spa->spa_dsl_pool == NULL);
888	ASSERT(spa->spa_root_vdev == NULL);
889	ASSERT(spa->spa_async_zio_root == NULL);
890	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
891
892	txg_list_destroy(&spa->spa_vdev_txg_list);
893
894	list_destroy(&spa->spa_config_dirty_list);
895	list_destroy(&spa->spa_state_dirty_list);
896
897	for (int t = 0; t < ZIO_TYPES; t++) {
898		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
899			if (spa->spa_zio_taskq[t][q] != NULL)
900				taskq_destroy(spa->spa_zio_taskq[t][q]);
901			spa->spa_zio_taskq[t][q] = NULL;
902		}
903	}
904
905	metaslab_class_destroy(spa->spa_normal_class);
906	spa->spa_normal_class = NULL;
907
908	metaslab_class_destroy(spa->spa_log_class);
909	spa->spa_log_class = NULL;
910
911	/*
912	 * If this was part of an import or the open otherwise failed, we may
913	 * still have errors left in the queues.  Empty them just in case.
914	 */
915	spa_errlog_drain(spa);
916
917	avl_destroy(&spa->spa_errlist_scrub);
918	avl_destroy(&spa->spa_errlist_last);
919
920	spa->spa_state = POOL_STATE_UNINITIALIZED;
921
922	mutex_enter(&spa->spa_proc_lock);
923	if (spa->spa_proc_state != SPA_PROC_NONE) {
924		ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
925		spa->spa_proc_state = SPA_PROC_DEACTIVATE;
926		cv_broadcast(&spa->spa_proc_cv);
927		while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
928			ASSERT(spa->spa_proc != &p0);
929			cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
930		}
931		ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
932		spa->spa_proc_state = SPA_PROC_NONE;
933	}
934	ASSERT(spa->spa_proc == &p0);
935	mutex_exit(&spa->spa_proc_lock);
936
937	/*
938	 * We want to make sure spa_thread() has actually exited the ZFS
939	 * module, so that the module can't be unloaded out from underneath
940	 * it.
941	 */
942	if (spa->spa_did != 0) {
943		thread_join(spa->spa_did);
944		spa->spa_did = 0;
945	}
946}
947
948/*
949 * Verify a pool configuration, and construct the vdev tree appropriately.  This
950 * will create all the necessary vdevs in the appropriate layout, with each vdev
951 * in the CLOSED state.  This will prep the pool before open/creation/import.
952 * All vdev validation is done by the vdev_alloc() routine.
953 */
954static int
955spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
956    uint_t id, int atype)
957{
958	nvlist_t **child;
959	uint_t children;
960	int error;
961
962	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
963		return (error);
964
965	if ((*vdp)->vdev_ops->vdev_op_leaf)
966		return (0);
967
968	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
969	    &child, &children);
970
971	if (error == ENOENT)
972		return (0);
973
974	if (error) {
975		vdev_free(*vdp);
976		*vdp = NULL;
977		return (EINVAL);
978	}
979
980	for (int c = 0; c < children; c++) {
981		vdev_t *vd;
982		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
983		    atype)) != 0) {
984			vdev_free(*vdp);
985			*vdp = NULL;
986			return (error);
987		}
988	}
989
990	ASSERT(*vdp != NULL);
991
992	return (0);
993}
994
995/*
996 * Opposite of spa_load().
997 */
998static void
999spa_unload(spa_t *spa)
1000{
1001	int i;
1002
1003	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1004
1005	/*
1006	 * Stop async tasks.
1007	 */
1008	spa_async_suspend(spa);
1009
1010	/*
1011	 * Stop syncing.
1012	 */
1013	if (spa->spa_sync_on) {
1014		txg_sync_stop(spa->spa_dsl_pool);
1015		spa->spa_sync_on = B_FALSE;
1016	}
1017
1018	/*
1019	 * Wait for any outstanding async I/O to complete.
1020	 */
1021	if (spa->spa_async_zio_root != NULL) {
1022		(void) zio_wait(spa->spa_async_zio_root);
1023		spa->spa_async_zio_root = NULL;
1024	}
1025
1026	bpobj_close(&spa->spa_deferred_bpobj);
1027
1028	/*
1029	 * Close the dsl pool.
1030	 */
1031	if (spa->spa_dsl_pool) {
1032		dsl_pool_close(spa->spa_dsl_pool);
1033		spa->spa_dsl_pool = NULL;
1034		spa->spa_meta_objset = NULL;
1035	}
1036
1037	ddt_unload(spa);
1038
1039	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1040
1041	/*
1042	 * Drop and purge level 2 cache
1043	 */
1044	spa_l2cache_drop(spa);
1045
1046	/*
1047	 * Close all vdevs.
1048	 */
1049	if (spa->spa_root_vdev)
1050		vdev_free(spa->spa_root_vdev);
1051	ASSERT(spa->spa_root_vdev == NULL);
1052
1053	for (i = 0; i < spa->spa_spares.sav_count; i++)
1054		vdev_free(spa->spa_spares.sav_vdevs[i]);
1055	if (spa->spa_spares.sav_vdevs) {
1056		kmem_free(spa->spa_spares.sav_vdevs,
1057		    spa->spa_spares.sav_count * sizeof (void *));
1058		spa->spa_spares.sav_vdevs = NULL;
1059	}
1060	if (spa->spa_spares.sav_config) {
1061		nvlist_free(spa->spa_spares.sav_config);
1062		spa->spa_spares.sav_config = NULL;
1063	}
1064	spa->spa_spares.sav_count = 0;
1065
1066	for (i = 0; i < spa->spa_l2cache.sav_count; i++)
1067		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1068	if (spa->spa_l2cache.sav_vdevs) {
1069		kmem_free(spa->spa_l2cache.sav_vdevs,
1070		    spa->spa_l2cache.sav_count * sizeof (void *));
1071		spa->spa_l2cache.sav_vdevs = NULL;
1072	}
1073	if (spa->spa_l2cache.sav_config) {
1074		nvlist_free(spa->spa_l2cache.sav_config);
1075		spa->spa_l2cache.sav_config = NULL;
1076	}
1077	spa->spa_l2cache.sav_count = 0;
1078
1079	spa->spa_async_suspended = 0;
1080
1081	if (spa->spa_comment != NULL) {
1082		spa_strfree(spa->spa_comment);
1083		spa->spa_comment = NULL;
1084	}
1085
1086	spa_config_exit(spa, SCL_ALL, FTAG);
1087}
1088
1089/*
1090 * Load (or re-load) the current list of vdevs describing the active spares for
1091 * this pool.  When this is called, we have some form of basic information in
1092 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1093 * then re-generate a more complete list including status information.
1094 */
1095static void
1096spa_load_spares(spa_t *spa)
1097{
1098	nvlist_t **spares;
1099	uint_t nspares;
1100	int i;
1101	vdev_t *vd, *tvd;
1102
1103	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1104
1105	/*
1106	 * First, close and free any existing spare vdevs.
1107	 */
1108	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1109		vd = spa->spa_spares.sav_vdevs[i];
1110
1111		/* Undo the call to spa_activate() below */
1112		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1113		    B_FALSE)) != NULL && tvd->vdev_isspare)
1114			spa_spare_remove(tvd);
1115		vdev_close(vd);
1116		vdev_free(vd);
1117	}
1118
1119	if (spa->spa_spares.sav_vdevs)
1120		kmem_free(spa->spa_spares.sav_vdevs,
1121		    spa->spa_spares.sav_count * sizeof (void *));
1122
1123	if (spa->spa_spares.sav_config == NULL)
1124		nspares = 0;
1125	else
1126		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1127		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1128
1129	spa->spa_spares.sav_count = (int)nspares;
1130	spa->spa_spares.sav_vdevs = NULL;
1131
1132	if (nspares == 0)
1133		return;
1134
1135	/*
1136	 * Construct the array of vdevs, opening them to get status in the
1137	 * process.   For each spare, there is potentially two different vdev_t
1138	 * structures associated with it: one in the list of spares (used only
1139	 * for basic validation purposes) and one in the active vdev
1140	 * configuration (if it's spared in).  During this phase we open and
1141	 * validate each vdev on the spare list.  If the vdev also exists in the
1142	 * active configuration, then we also mark this vdev as an active spare.
1143	 */
1144	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1145	    KM_SLEEP);
1146	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1147		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1148		    VDEV_ALLOC_SPARE) == 0);
1149		ASSERT(vd != NULL);
1150
1151		spa->spa_spares.sav_vdevs[i] = vd;
1152
1153		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1154		    B_FALSE)) != NULL) {
1155			if (!tvd->vdev_isspare)
1156				spa_spare_add(tvd);
1157
1158			/*
1159			 * We only mark the spare active if we were successfully
1160			 * able to load the vdev.  Otherwise, importing a pool
1161			 * with a bad active spare would result in strange
1162			 * behavior, because multiple pool would think the spare
1163			 * is actively in use.
1164			 *
1165			 * There is a vulnerability here to an equally bizarre
1166			 * circumstance, where a dead active spare is later
1167			 * brought back to life (onlined or otherwise).  Given
1168			 * the rarity of this scenario, and the extra complexity
1169			 * it adds, we ignore the possibility.
1170			 */
1171			if (!vdev_is_dead(tvd))
1172				spa_spare_activate(tvd);
1173		}
1174
1175		vd->vdev_top = vd;
1176		vd->vdev_aux = &spa->spa_spares;
1177
1178		if (vdev_open(vd) != 0)
1179			continue;
1180
1181		if (vdev_validate_aux(vd) == 0)
1182			spa_spare_add(vd);
1183	}
1184
1185	/*
1186	 * Recompute the stashed list of spares, with status information
1187	 * this time.
1188	 */
1189	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1190	    DATA_TYPE_NVLIST_ARRAY) == 0);
1191
1192	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1193	    KM_SLEEP);
1194	for (i = 0; i < spa->spa_spares.sav_count; i++)
1195		spares[i] = vdev_config_generate(spa,
1196		    spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1197	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1198	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1199	for (i = 0; i < spa->spa_spares.sav_count; i++)
1200		nvlist_free(spares[i]);
1201	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1202}
1203
1204/*
1205 * Load (or re-load) the current list of vdevs describing the active l2cache for
1206 * this pool.  When this is called, we have some form of basic information in
1207 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1208 * then re-generate a more complete list including status information.
1209 * Devices which are already active have their details maintained, and are
1210 * not re-opened.
1211 */
1212static void
1213spa_load_l2cache(spa_t *spa)
1214{
1215	nvlist_t **l2cache;
1216	uint_t nl2cache;
1217	int i, j, oldnvdevs;
1218	uint64_t guid;
1219	vdev_t *vd, **oldvdevs, **newvdevs;
1220	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1221
1222	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1223
1224	if (sav->sav_config != NULL) {
1225		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1226		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1227		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1228	} else {
1229		nl2cache = 0;
1230	}
1231
1232	oldvdevs = sav->sav_vdevs;
1233	oldnvdevs = sav->sav_count;
1234	sav->sav_vdevs = NULL;
1235	sav->sav_count = 0;
1236
1237	/*
1238	 * Process new nvlist of vdevs.
1239	 */
1240	for (i = 0; i < nl2cache; i++) {
1241		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1242		    &guid) == 0);
1243
1244		newvdevs[i] = NULL;
1245		for (j = 0; j < oldnvdevs; j++) {
1246			vd = oldvdevs[j];
1247			if (vd != NULL && guid == vd->vdev_guid) {
1248				/*
1249				 * Retain previous vdev for add/remove ops.
1250				 */
1251				newvdevs[i] = vd;
1252				oldvdevs[j] = NULL;
1253				break;
1254			}
1255		}
1256
1257		if (newvdevs[i] == NULL) {
1258			/*
1259			 * Create new vdev
1260			 */
1261			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1262			    VDEV_ALLOC_L2CACHE) == 0);
1263			ASSERT(vd != NULL);
1264			newvdevs[i] = vd;
1265
1266			/*
1267			 * Commit this vdev as an l2cache device,
1268			 * even if it fails to open.
1269			 */
1270			spa_l2cache_add(vd);
1271
1272			vd->vdev_top = vd;
1273			vd->vdev_aux = sav;
1274
1275			spa_l2cache_activate(vd);
1276
1277			if (vdev_open(vd) != 0)
1278				continue;
1279
1280			(void) vdev_validate_aux(vd);
1281
1282			if (!vdev_is_dead(vd))
1283				l2arc_add_vdev(spa, vd);
1284		}
1285	}
1286
1287	/*
1288	 * Purge vdevs that were dropped
1289	 */
1290	for (i = 0; i < oldnvdevs; i++) {
1291		uint64_t pool;
1292
1293		vd = oldvdevs[i];
1294		if (vd != NULL) {
1295			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1296			    pool != 0ULL && l2arc_vdev_present(vd))
1297				l2arc_remove_vdev(vd);
1298			(void) vdev_close(vd);
1299			spa_l2cache_remove(vd);
1300		}
1301	}
1302
1303	if (oldvdevs)
1304		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1305
1306	if (sav->sav_config == NULL)
1307		goto out;
1308
1309	sav->sav_vdevs = newvdevs;
1310	sav->sav_count = (int)nl2cache;
1311
1312	/*
1313	 * Recompute the stashed list of l2cache devices, with status
1314	 * information this time.
1315	 */
1316	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1317	    DATA_TYPE_NVLIST_ARRAY) == 0);
1318
1319	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1320	for (i = 0; i < sav->sav_count; i++)
1321		l2cache[i] = vdev_config_generate(spa,
1322		    sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1323	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1324	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1325out:
1326	for (i = 0; i < sav->sav_count; i++)
1327		nvlist_free(l2cache[i]);
1328	if (sav->sav_count)
1329		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1330}
1331
1332static int
1333load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1334{
1335	dmu_buf_t *db;
1336	char *packed = NULL;
1337	size_t nvsize = 0;
1338	int error;
1339	*value = NULL;
1340
1341	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1342	nvsize = *(uint64_t *)db->db_data;
1343	dmu_buf_rele(db, FTAG);
1344
1345	packed = kmem_alloc(nvsize, KM_SLEEP);
1346	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1347	    DMU_READ_PREFETCH);
1348	if (error == 0)
1349		error = nvlist_unpack(packed, nvsize, value, 0);
1350	kmem_free(packed, nvsize);
1351
1352	return (error);
1353}
1354
1355/*
1356 * Checks to see if the given vdev could not be opened, in which case we post a
1357 * sysevent to notify the autoreplace code that the device has been removed.
1358 */
1359static void
1360spa_check_removed(vdev_t *vd)
1361{
1362	for (int c = 0; c < vd->vdev_children; c++)
1363		spa_check_removed(vd->vdev_child[c]);
1364
1365	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1366		zfs_post_autoreplace(vd->vdev_spa, vd);
1367		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1368	}
1369}
1370
1371/*
1372 * Validate the current config against the MOS config
1373 */
1374static boolean_t
1375spa_config_valid(spa_t *spa, nvlist_t *config)
1376{
1377	vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1378	nvlist_t *nv;
1379
1380	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1381
1382	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1383	VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1384
1385	ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1386
1387	/*
1388	 * If we're doing a normal import, then build up any additional
1389	 * diagnostic information about missing devices in this config.
1390	 * We'll pass this up to the user for further processing.
1391	 */
1392	if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1393		nvlist_t **child, *nv;
1394		uint64_t idx = 0;
1395
1396		child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1397		    KM_SLEEP);
1398		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1399
1400		for (int c = 0; c < rvd->vdev_children; c++) {
1401			vdev_t *tvd = rvd->vdev_child[c];
1402			vdev_t *mtvd  = mrvd->vdev_child[c];
1403
1404			if (tvd->vdev_ops == &vdev_missing_ops &&
1405			    mtvd->vdev_ops != &vdev_missing_ops &&
1406			    mtvd->vdev_islog)
1407				child[idx++] = vdev_config_generate(spa, mtvd,
1408				    B_FALSE, 0);
1409		}
1410
1411		if (idx) {
1412			VERIFY(nvlist_add_nvlist_array(nv,
1413			    ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1414			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1415			    ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1416
1417			for (int i = 0; i < idx; i++)
1418				nvlist_free(child[i]);
1419		}
1420		nvlist_free(nv);
1421		kmem_free(child, rvd->vdev_children * sizeof (char **));
1422	}
1423
1424	/*
1425	 * Compare the root vdev tree with the information we have
1426	 * from the MOS config (mrvd). Check each top-level vdev
1427	 * with the corresponding MOS config top-level (mtvd).
1428	 */
1429	for (int c = 0; c < rvd->vdev_children; c++) {
1430		vdev_t *tvd = rvd->vdev_child[c];
1431		vdev_t *mtvd  = mrvd->vdev_child[c];
1432
1433		/*
1434		 * Resolve any "missing" vdevs in the current configuration.
1435		 * If we find that the MOS config has more accurate information
1436		 * about the top-level vdev then use that vdev instead.
1437		 */
1438		if (tvd->vdev_ops == &vdev_missing_ops &&
1439		    mtvd->vdev_ops != &vdev_missing_ops) {
1440
1441			if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1442				continue;
1443
1444			/*
1445			 * Device specific actions.
1446			 */
1447			if (mtvd->vdev_islog) {
1448				spa_set_log_state(spa, SPA_LOG_CLEAR);
1449			} else {
1450				/*
1451				 * XXX - once we have 'readonly' pool
1452				 * support we should be able to handle
1453				 * missing data devices by transitioning
1454				 * the pool to readonly.
1455				 */
1456				continue;
1457			}
1458
1459			/*
1460			 * Swap the missing vdev with the data we were
1461			 * able to obtain from the MOS config.
1462			 */
1463			vdev_remove_child(rvd, tvd);
1464			vdev_remove_child(mrvd, mtvd);
1465
1466			vdev_add_child(rvd, mtvd);
1467			vdev_add_child(mrvd, tvd);
1468
1469			spa_config_exit(spa, SCL_ALL, FTAG);
1470			vdev_load(mtvd);
1471			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1472
1473			vdev_reopen(rvd);
1474		} else if (mtvd->vdev_islog) {
1475			/*
1476			 * Load the slog device's state from the MOS config
1477			 * since it's possible that the label does not
1478			 * contain the most up-to-date information.
1479			 */
1480			vdev_load_log_state(tvd, mtvd);
1481			vdev_reopen(tvd);
1482		}
1483	}
1484	vdev_free(mrvd);
1485	spa_config_exit(spa, SCL_ALL, FTAG);
1486
1487	/*
1488	 * Ensure we were able to validate the config.
1489	 */
1490	return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1491}
1492
1493/*
1494 * Check for missing log devices
1495 */
1496static int
1497spa_check_logs(spa_t *spa)
1498{
1499	switch (spa->spa_log_state) {
1500	case SPA_LOG_MISSING:
1501		/* need to recheck in case slog has been restored */
1502	case SPA_LOG_UNKNOWN:
1503		if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1504		    DS_FIND_CHILDREN)) {
1505			spa_set_log_state(spa, SPA_LOG_MISSING);
1506			return (1);
1507		}
1508		break;
1509	}
1510	return (0);
1511}
1512
1513static boolean_t
1514spa_passivate_log(spa_t *spa)
1515{
1516	vdev_t *rvd = spa->spa_root_vdev;
1517	boolean_t slog_found = B_FALSE;
1518
1519	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1520
1521	if (!spa_has_slogs(spa))
1522		return (B_FALSE);
1523
1524	for (int c = 0; c < rvd->vdev_children; c++) {
1525		vdev_t *tvd = rvd->vdev_child[c];
1526		metaslab_group_t *mg = tvd->vdev_mg;
1527
1528		if (tvd->vdev_islog) {
1529			metaslab_group_passivate(mg);
1530			slog_found = B_TRUE;
1531		}
1532	}
1533
1534	return (slog_found);
1535}
1536
1537static void
1538spa_activate_log(spa_t *spa)
1539{
1540	vdev_t *rvd = spa->spa_root_vdev;
1541
1542	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1543
1544	for (int c = 0; c < rvd->vdev_children; c++) {
1545		vdev_t *tvd = rvd->vdev_child[c];
1546		metaslab_group_t *mg = tvd->vdev_mg;
1547
1548		if (tvd->vdev_islog)
1549			metaslab_group_activate(mg);
1550	}
1551}
1552
1553int
1554spa_offline_log(spa_t *spa)
1555{
1556	int error = 0;
1557
1558	if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1559	    NULL, DS_FIND_CHILDREN)) == 0) {
1560
1561		/*
1562		 * We successfully offlined the log device, sync out the
1563		 * current txg so that the "stubby" block can be removed
1564		 * by zil_sync().
1565		 */
1566		txg_wait_synced(spa->spa_dsl_pool, 0);
1567	}
1568	return (error);
1569}
1570
1571static void
1572spa_aux_check_removed(spa_aux_vdev_t *sav)
1573{
1574	for (int i = 0; i < sav->sav_count; i++)
1575		spa_check_removed(sav->sav_vdevs[i]);
1576}
1577
1578void
1579spa_claim_notify(zio_t *zio)
1580{
1581	spa_t *spa = zio->io_spa;
1582
1583	if (zio->io_error)
1584		return;
1585
1586	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1587	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1588		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1589	mutex_exit(&spa->spa_props_lock);
1590}
1591
1592typedef struct spa_load_error {
1593	uint64_t	sle_meta_count;
1594	uint64_t	sle_data_count;
1595} spa_load_error_t;
1596
1597static void
1598spa_load_verify_done(zio_t *zio)
1599{
1600	blkptr_t *bp = zio->io_bp;
1601	spa_load_error_t *sle = zio->io_private;
1602	dmu_object_type_t type = BP_GET_TYPE(bp);
1603	int error = zio->io_error;
1604
1605	if (error) {
1606		if ((BP_GET_LEVEL(bp) != 0 || dmu_ot[type].ot_metadata) &&
1607		    type != DMU_OT_INTENT_LOG)
1608			atomic_add_64(&sle->sle_meta_count, 1);
1609		else
1610			atomic_add_64(&sle->sle_data_count, 1);
1611	}
1612	zio_data_buf_free(zio->io_data, zio->io_size);
1613}
1614
1615/*ARGSUSED*/
1616static int
1617spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1618    arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1619{
1620	if (bp != NULL) {
1621		zio_t *rio = arg;
1622		size_t size = BP_GET_PSIZE(bp);
1623		void *data = zio_data_buf_alloc(size);
1624
1625		zio_nowait(zio_read(rio, spa, bp, data, size,
1626		    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1627		    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1628		    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1629	}
1630	return (0);
1631}
1632
1633static int
1634spa_load_verify(spa_t *spa)
1635{
1636	zio_t *rio;
1637	spa_load_error_t sle = { 0 };
1638	zpool_rewind_policy_t policy;
1639	boolean_t verify_ok = B_FALSE;
1640	int error;
1641
1642	zpool_get_rewind_policy(spa->spa_config, &policy);
1643
1644	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1645		return (0);
1646
1647	rio = zio_root(spa, NULL, &sle,
1648	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1649
1650	error = traverse_pool(spa, spa->spa_verify_min_txg,
1651	    TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1652
1653	(void) zio_wait(rio);
1654
1655	spa->spa_load_meta_errors = sle.sle_meta_count;
1656	spa->spa_load_data_errors = sle.sle_data_count;
1657
1658	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1659	    sle.sle_data_count <= policy.zrp_maxdata) {
1660		int64_t loss = 0;
1661
1662		verify_ok = B_TRUE;
1663		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1664		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1665
1666		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1667		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1668		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1669		VERIFY(nvlist_add_int64(spa->spa_load_info,
1670		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1671		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1672		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1673	} else {
1674		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1675	}
1676
1677	if (error) {
1678		if (error != ENXIO && error != EIO)
1679			error = EIO;
1680		return (error);
1681	}
1682
1683	return (verify_ok ? 0 : EIO);
1684}
1685
1686/*
1687 * Find a value in the pool props object.
1688 */
1689static void
1690spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1691{
1692	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1693	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1694}
1695
1696/*
1697 * Find a value in the pool directory object.
1698 */
1699static int
1700spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1701{
1702	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1703	    name, sizeof (uint64_t), 1, val));
1704}
1705
1706static int
1707spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1708{
1709	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1710	return (err);
1711}
1712
1713/*
1714 * Fix up config after a partly-completed split.  This is done with the
1715 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1716 * pool have that entry in their config, but only the splitting one contains
1717 * a list of all the guids of the vdevs that are being split off.
1718 *
1719 * This function determines what to do with that list: either rejoin
1720 * all the disks to the pool, or complete the splitting process.  To attempt
1721 * the rejoin, each disk that is offlined is marked online again, and
1722 * we do a reopen() call.  If the vdev label for every disk that was
1723 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1724 * then we call vdev_split() on each disk, and complete the split.
1725 *
1726 * Otherwise we leave the config alone, with all the vdevs in place in
1727 * the original pool.
1728 */
1729static void
1730spa_try_repair(spa_t *spa, nvlist_t *config)
1731{
1732	uint_t extracted;
1733	uint64_t *glist;
1734	uint_t i, gcount;
1735	nvlist_t *nvl;
1736	vdev_t **vd;
1737	boolean_t attempt_reopen;
1738
1739	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1740		return;
1741
1742	/* check that the config is complete */
1743	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1744	    &glist, &gcount) != 0)
1745		return;
1746
1747	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1748
1749	/* attempt to online all the vdevs & validate */
1750	attempt_reopen = B_TRUE;
1751	for (i = 0; i < gcount; i++) {
1752		if (glist[i] == 0)	/* vdev is hole */
1753			continue;
1754
1755		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1756		if (vd[i] == NULL) {
1757			/*
1758			 * Don't bother attempting to reopen the disks;
1759			 * just do the split.
1760			 */
1761			attempt_reopen = B_FALSE;
1762		} else {
1763			/* attempt to re-online it */
1764			vd[i]->vdev_offline = B_FALSE;
1765		}
1766	}
1767
1768	if (attempt_reopen) {
1769		vdev_reopen(spa->spa_root_vdev);
1770
1771		/* check each device to see what state it's in */
1772		for (extracted = 0, i = 0; i < gcount; i++) {
1773			if (vd[i] != NULL &&
1774			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1775				break;
1776			++extracted;
1777		}
1778	}
1779
1780	/*
1781	 * If every disk has been moved to the new pool, or if we never
1782	 * even attempted to look at them, then we split them off for
1783	 * good.
1784	 */
1785	if (!attempt_reopen || gcount == extracted) {
1786		for (i = 0; i < gcount; i++)
1787			if (vd[i] != NULL)
1788				vdev_split(vd[i]);
1789		vdev_reopen(spa->spa_root_vdev);
1790	}
1791
1792	kmem_free(vd, gcount * sizeof (vdev_t *));
1793}
1794
1795static int
1796spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1797    boolean_t mosconfig)
1798{
1799	nvlist_t *config = spa->spa_config;
1800	char *ereport = FM_EREPORT_ZFS_POOL;
1801	char *comment;
1802	int error;
1803	uint64_t pool_guid;
1804	nvlist_t *nvl;
1805
1806	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1807		return (EINVAL);
1808
1809	ASSERT(spa->spa_comment == NULL);
1810	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
1811		spa->spa_comment = spa_strdup(comment);
1812
1813	/*
1814	 * Versioning wasn't explicitly added to the label until later, so if
1815	 * it's not present treat it as the initial version.
1816	 */
1817	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1818	    &spa->spa_ubsync.ub_version) != 0)
1819		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1820
1821	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1822	    &spa->spa_config_txg);
1823
1824	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1825	    spa_guid_exists(pool_guid, 0)) {
1826		error = EEXIST;
1827	} else {
1828		spa->spa_config_guid = pool_guid;
1829
1830		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1831		    &nvl) == 0) {
1832			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1833			    KM_SLEEP) == 0);
1834		}
1835
1836		gethrestime(&spa->spa_loaded_ts);
1837		error = spa_load_impl(spa, pool_guid, config, state, type,
1838		    mosconfig, &ereport);
1839	}
1840
1841	spa->spa_minref = refcount_count(&spa->spa_refcount);
1842	if (error) {
1843		if (error != EEXIST) {
1844			spa->spa_loaded_ts.tv_sec = 0;
1845			spa->spa_loaded_ts.tv_nsec = 0;
1846		}
1847		if (error != EBADF) {
1848			zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1849		}
1850	}
1851	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1852	spa->spa_ena = 0;
1853
1854	return (error);
1855}
1856
1857/*
1858 * Load an existing storage pool, using the pool's builtin spa_config as a
1859 * source of configuration information.
1860 */
1861static int
1862spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1863    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1864    char **ereport)
1865{
1866	int error = 0;
1867	nvlist_t *nvroot = NULL;
1868	vdev_t *rvd;
1869	uberblock_t *ub = &spa->spa_uberblock;
1870	uint64_t children, config_cache_txg = spa->spa_config_txg;
1871	int orig_mode = spa->spa_mode;
1872	int parse;
1873	uint64_t obj;
1874
1875	/*
1876	 * If this is an untrusted config, access the pool in read-only mode.
1877	 * This prevents things like resilvering recently removed devices.
1878	 */
1879	if (!mosconfig)
1880		spa->spa_mode = FREAD;
1881
1882	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1883
1884	spa->spa_load_state = state;
1885
1886	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1887		return (EINVAL);
1888
1889	parse = (type == SPA_IMPORT_EXISTING ?
1890	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1891
1892	/*
1893	 * Create "The Godfather" zio to hold all async IOs
1894	 */
1895	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1896	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1897
1898	/*
1899	 * Parse the configuration into a vdev tree.  We explicitly set the
1900	 * value that will be returned by spa_version() since parsing the
1901	 * configuration requires knowing the version number.
1902	 */
1903	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1904	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1905	spa_config_exit(spa, SCL_ALL, FTAG);
1906
1907	if (error != 0)
1908		return (error);
1909
1910	ASSERT(spa->spa_root_vdev == rvd);
1911
1912	if (type != SPA_IMPORT_ASSEMBLE) {
1913		ASSERT(spa_guid(spa) == pool_guid);
1914	}
1915
1916	/*
1917	 * Try to open all vdevs, loading each label in the process.
1918	 */
1919	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1920	error = vdev_open(rvd);
1921	spa_config_exit(spa, SCL_ALL, FTAG);
1922	if (error != 0)
1923		return (error);
1924
1925	/*
1926	 * We need to validate the vdev labels against the configuration that
1927	 * we have in hand, which is dependent on the setting of mosconfig. If
1928	 * mosconfig is true then we're validating the vdev labels based on
1929	 * that config.  Otherwise, we're validating against the cached config
1930	 * (zpool.cache) that was read when we loaded the zfs module, and then
1931	 * later we will recursively call spa_load() and validate against
1932	 * the vdev config.
1933	 *
1934	 * If we're assembling a new pool that's been split off from an
1935	 * existing pool, the labels haven't yet been updated so we skip
1936	 * validation for now.
1937	 */
1938	if (type != SPA_IMPORT_ASSEMBLE) {
1939		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1940		error = vdev_validate(rvd);
1941		spa_config_exit(spa, SCL_ALL, FTAG);
1942
1943		if (error != 0)
1944			return (error);
1945
1946		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1947			return (ENXIO);
1948	}
1949
1950	/*
1951	 * Find the best uberblock.
1952	 */
1953	vdev_uberblock_load(NULL, rvd, ub);
1954
1955	/*
1956	 * If we weren't able to find a single valid uberblock, return failure.
1957	 */
1958	if (ub->ub_txg == 0)
1959		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
1960
1961	/*
1962	 * If the pool is newer than the code, we can't open it.
1963	 */
1964	if (ub->ub_version > SPA_VERSION)
1965		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
1966
1967	/*
1968	 * If the vdev guid sum doesn't match the uberblock, we have an
1969	 * incomplete configuration.  We first check to see if the pool
1970	 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
1971	 * If it is, defer the vdev_guid_sum check till later so we
1972	 * can handle missing vdevs.
1973	 */
1974	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
1975	    &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
1976	    rvd->vdev_guid_sum != ub->ub_guid_sum)
1977		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
1978
1979	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
1980		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1981		spa_try_repair(spa, config);
1982		spa_config_exit(spa, SCL_ALL, FTAG);
1983		nvlist_free(spa->spa_config_splitting);
1984		spa->spa_config_splitting = NULL;
1985	}
1986
1987	/*
1988	 * Initialize internal SPA structures.
1989	 */
1990	spa->spa_state = POOL_STATE_ACTIVE;
1991	spa->spa_ubsync = spa->spa_uberblock;
1992	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
1993	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
1994	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
1995	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
1996	spa->spa_claim_max_txg = spa->spa_first_txg;
1997	spa->spa_prev_software_version = ub->ub_software_version;
1998
1999	error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2000	if (error)
2001		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2002	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2003
2004	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2005		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2006
2007	if (!mosconfig) {
2008		uint64_t hostid;
2009		nvlist_t *policy = NULL, *nvconfig;
2010
2011		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2012			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2013
2014		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2015		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2016			char *hostname;
2017			unsigned long myhostid = 0;
2018
2019			VERIFY(nvlist_lookup_string(nvconfig,
2020			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2021
2022#ifdef	_KERNEL
2023			myhostid = zone_get_hostid(NULL);
2024#else	/* _KERNEL */
2025			/*
2026			 * We're emulating the system's hostid in userland, so
2027			 * we can't use zone_get_hostid().
2028			 */
2029			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2030#endif	/* _KERNEL */
2031			if (hostid != 0 && myhostid != 0 &&
2032			    hostid != myhostid) {
2033				nvlist_free(nvconfig);
2034				cmn_err(CE_WARN, "pool '%s' could not be "
2035				    "loaded as it was last accessed by "
2036				    "another system (host: %s hostid: 0x%lx). "
2037				    "See: http://www.sun.com/msg/ZFS-8000-EY",
2038				    spa_name(spa), hostname,
2039				    (unsigned long)hostid);
2040				return (EBADF);
2041			}
2042		}
2043		if (nvlist_lookup_nvlist(spa->spa_config,
2044		    ZPOOL_REWIND_POLICY, &policy) == 0)
2045			VERIFY(nvlist_add_nvlist(nvconfig,
2046			    ZPOOL_REWIND_POLICY, policy) == 0);
2047
2048		spa_config_set(spa, nvconfig);
2049		spa_unload(spa);
2050		spa_deactivate(spa);
2051		spa_activate(spa, orig_mode);
2052
2053		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2054	}
2055
2056	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2057		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2058	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2059	if (error != 0)
2060		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2061
2062	/*
2063	 * Load the bit that tells us to use the new accounting function
2064	 * (raid-z deflation).  If we have an older pool, this will not
2065	 * be present.
2066	 */
2067	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2068	if (error != 0 && error != ENOENT)
2069		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2070
2071	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2072	    &spa->spa_creation_version);
2073	if (error != 0 && error != ENOENT)
2074		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2075
2076	/*
2077	 * Load the persistent error log.  If we have an older pool, this will
2078	 * not be present.
2079	 */
2080	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2081	if (error != 0 && error != ENOENT)
2082		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2083
2084	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2085	    &spa->spa_errlog_scrub);
2086	if (error != 0 && error != ENOENT)
2087		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2088
2089	/*
2090	 * Load the history object.  If we have an older pool, this
2091	 * will not be present.
2092	 */
2093	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2094	if (error != 0 && error != ENOENT)
2095		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2096
2097	/*
2098	 * If we're assembling the pool from the split-off vdevs of
2099	 * an existing pool, we don't want to attach the spares & cache
2100	 * devices.
2101	 */
2102
2103	/*
2104	 * Load any hot spares for this pool.
2105	 */
2106	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2107	if (error != 0 && error != ENOENT)
2108		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2109	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2110		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2111		if (load_nvlist(spa, spa->spa_spares.sav_object,
2112		    &spa->spa_spares.sav_config) != 0)
2113			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2114
2115		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2116		spa_load_spares(spa);
2117		spa_config_exit(spa, SCL_ALL, FTAG);
2118	} else if (error == 0) {
2119		spa->spa_spares.sav_sync = B_TRUE;
2120	}
2121
2122	/*
2123	 * Load any level 2 ARC devices for this pool.
2124	 */
2125	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2126	    &spa->spa_l2cache.sav_object);
2127	if (error != 0 && error != ENOENT)
2128		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2129	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2130		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2131		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2132		    &spa->spa_l2cache.sav_config) != 0)
2133			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2134
2135		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2136		spa_load_l2cache(spa);
2137		spa_config_exit(spa, SCL_ALL, FTAG);
2138	} else if (error == 0) {
2139		spa->spa_l2cache.sav_sync = B_TRUE;
2140	}
2141
2142	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2143
2144	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2145	if (error && error != ENOENT)
2146		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2147
2148	if (error == 0) {
2149		uint64_t autoreplace;
2150
2151		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2152		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2153		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2154		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2155		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2156		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2157		    &spa->spa_dedup_ditto);
2158
2159		spa->spa_autoreplace = (autoreplace != 0);
2160	}
2161
2162	/*
2163	 * If the 'autoreplace' property is set, then post a resource notifying
2164	 * the ZFS DE that it should not issue any faults for unopenable
2165	 * devices.  We also iterate over the vdevs, and post a sysevent for any
2166	 * unopenable vdevs so that the normal autoreplace handler can take
2167	 * over.
2168	 */
2169	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2170		spa_check_removed(spa->spa_root_vdev);
2171		/*
2172		 * For the import case, this is done in spa_import(), because
2173		 * at this point we're using the spare definitions from
2174		 * the MOS config, not necessarily from the userland config.
2175		 */
2176		if (state != SPA_LOAD_IMPORT) {
2177			spa_aux_check_removed(&spa->spa_spares);
2178			spa_aux_check_removed(&spa->spa_l2cache);
2179		}
2180	}
2181
2182	/*
2183	 * Load the vdev state for all toplevel vdevs.
2184	 */
2185	vdev_load(rvd);
2186
2187	/*
2188	 * Propagate the leaf DTLs we just loaded all the way up the tree.
2189	 */
2190	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2191	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2192	spa_config_exit(spa, SCL_ALL, FTAG);
2193
2194	/*
2195	 * Load the DDTs (dedup tables).
2196	 */
2197	error = ddt_load(spa);
2198	if (error != 0)
2199		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2200
2201	spa_update_dspace(spa);
2202
2203	/*
2204	 * Validate the config, using the MOS config to fill in any
2205	 * information which might be missing.  If we fail to validate
2206	 * the config then declare the pool unfit for use. If we're
2207	 * assembling a pool from a split, the log is not transferred
2208	 * over.
2209	 */
2210	if (type != SPA_IMPORT_ASSEMBLE) {
2211		nvlist_t *nvconfig;
2212
2213		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2214			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2215
2216		if (!spa_config_valid(spa, nvconfig)) {
2217			nvlist_free(nvconfig);
2218			return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2219			    ENXIO));
2220		}
2221		nvlist_free(nvconfig);
2222
2223		/*
2224		 * Now that we've validate the config, check the state of the
2225		 * root vdev.  If it can't be opened, it indicates one or
2226		 * more toplevel vdevs are faulted.
2227		 */
2228		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2229			return (ENXIO);
2230
2231		if (spa_check_logs(spa)) {
2232			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2233			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2234		}
2235	}
2236
2237	/*
2238	 * We've successfully opened the pool, verify that we're ready
2239	 * to start pushing transactions.
2240	 */
2241	if (state != SPA_LOAD_TRYIMPORT) {
2242		if (error = spa_load_verify(spa))
2243			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2244			    error));
2245	}
2246
2247	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2248	    spa->spa_load_max_txg == UINT64_MAX)) {
2249		dmu_tx_t *tx;
2250		int need_update = B_FALSE;
2251
2252		ASSERT(state != SPA_LOAD_TRYIMPORT);
2253
2254		/*
2255		 * Claim log blocks that haven't been committed yet.
2256		 * This must all happen in a single txg.
2257		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2258		 * invoked from zil_claim_log_block()'s i/o done callback.
2259		 * Price of rollback is that we abandon the log.
2260		 */
2261		spa->spa_claiming = B_TRUE;
2262
2263		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2264		    spa_first_txg(spa));
2265		(void) dmu_objset_find(spa_name(spa),
2266		    zil_claim, tx, DS_FIND_CHILDREN);
2267		dmu_tx_commit(tx);
2268
2269		spa->spa_claiming = B_FALSE;
2270
2271		spa_set_log_state(spa, SPA_LOG_GOOD);
2272		spa->spa_sync_on = B_TRUE;
2273		txg_sync_start(spa->spa_dsl_pool);
2274
2275		/*
2276		 * Wait for all claims to sync.  We sync up to the highest
2277		 * claimed log block birth time so that claimed log blocks
2278		 * don't appear to be from the future.  spa_claim_max_txg
2279		 * will have been set for us by either zil_check_log_chain()
2280		 * (invoked from spa_check_logs()) or zil_claim() above.
2281		 */
2282		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2283
2284		/*
2285		 * If the config cache is stale, or we have uninitialized
2286		 * metaslabs (see spa_vdev_add()), then update the config.
2287		 *
2288		 * If this is a verbatim import, trust the current
2289		 * in-core spa_config and update the disk labels.
2290		 */
2291		if (config_cache_txg != spa->spa_config_txg ||
2292		    state == SPA_LOAD_IMPORT ||
2293		    state == SPA_LOAD_RECOVER ||
2294		    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2295			need_update = B_TRUE;
2296
2297		for (int c = 0; c < rvd->vdev_children; c++)
2298			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2299				need_update = B_TRUE;
2300
2301		/*
2302		 * Update the config cache asychronously in case we're the
2303		 * root pool, in which case the config cache isn't writable yet.
2304		 */
2305		if (need_update)
2306			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2307
2308		/*
2309		 * Check all DTLs to see if anything needs resilvering.
2310		 */
2311		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2312		    vdev_resilver_needed(rvd, NULL, NULL))
2313			spa_async_request(spa, SPA_ASYNC_RESILVER);
2314
2315		/*
2316		 * Delete any inconsistent datasets.
2317		 */
2318		(void) dmu_objset_find(spa_name(spa),
2319		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2320
2321		/*
2322		 * Clean up any stale temporary dataset userrefs.
2323		 */
2324		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2325	}
2326
2327	return (0);
2328}
2329
2330static int
2331spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2332{
2333	int mode = spa->spa_mode;
2334
2335	spa_unload(spa);
2336	spa_deactivate(spa);
2337
2338	spa->spa_load_max_txg--;
2339
2340	spa_activate(spa, mode);
2341	spa_async_suspend(spa);
2342
2343	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2344}
2345
2346static int
2347spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2348    uint64_t max_request, int rewind_flags)
2349{
2350	nvlist_t *config = NULL;
2351	int load_error, rewind_error;
2352	uint64_t safe_rewind_txg;
2353	uint64_t min_txg;
2354
2355	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2356		spa->spa_load_max_txg = spa->spa_load_txg;
2357		spa_set_log_state(spa, SPA_LOG_CLEAR);
2358	} else {
2359		spa->spa_load_max_txg = max_request;
2360	}
2361
2362	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2363	    mosconfig);
2364	if (load_error == 0)
2365		return (0);
2366
2367	if (spa->spa_root_vdev != NULL)
2368		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2369
2370	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2371	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2372
2373	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2374		nvlist_free(config);
2375		return (load_error);
2376	}
2377
2378	/* Price of rolling back is discarding txgs, including log */
2379	if (state == SPA_LOAD_RECOVER)
2380		spa_set_log_state(spa, SPA_LOG_CLEAR);
2381
2382	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2383	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2384	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2385	    TXG_INITIAL : safe_rewind_txg;
2386
2387	/*
2388	 * Continue as long as we're finding errors, we're still within
2389	 * the acceptable rewind range, and we're still finding uberblocks
2390	 */
2391	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2392	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2393		if (spa->spa_load_max_txg < safe_rewind_txg)
2394			spa->spa_extreme_rewind = B_TRUE;
2395		rewind_error = spa_load_retry(spa, state, mosconfig);
2396	}
2397
2398	spa->spa_extreme_rewind = B_FALSE;
2399	spa->spa_load_max_txg = UINT64_MAX;
2400
2401	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2402		spa_config_set(spa, config);
2403
2404	return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
2405}
2406
2407/*
2408 * Pool Open/Import
2409 *
2410 * The import case is identical to an open except that the configuration is sent
2411 * down from userland, instead of grabbed from the configuration cache.  For the
2412 * case of an open, the pool configuration will exist in the
2413 * POOL_STATE_UNINITIALIZED state.
2414 *
2415 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2416 * the same time open the pool, without having to keep around the spa_t in some
2417 * ambiguous state.
2418 */
2419static int
2420spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2421    nvlist_t **config)
2422{
2423	spa_t *spa;
2424	spa_load_state_t state = SPA_LOAD_OPEN;
2425	int error;
2426	int locked = B_FALSE;
2427
2428	*spapp = NULL;
2429
2430	/*
2431	 * As disgusting as this is, we need to support recursive calls to this
2432	 * function because dsl_dir_open() is called during spa_load(), and ends
2433	 * up calling spa_open() again.  The real fix is to figure out how to
2434	 * avoid dsl_dir_open() calling this in the first place.
2435	 */
2436	if (mutex_owner(&spa_namespace_lock) != curthread) {
2437		mutex_enter(&spa_namespace_lock);
2438		locked = B_TRUE;
2439	}
2440
2441	if ((spa = spa_lookup(pool)) == NULL) {
2442		if (locked)
2443			mutex_exit(&spa_namespace_lock);
2444		return (ENOENT);
2445	}
2446
2447	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2448		zpool_rewind_policy_t policy;
2449
2450		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2451		    &policy);
2452		if (policy.zrp_request & ZPOOL_DO_REWIND)
2453			state = SPA_LOAD_RECOVER;
2454
2455		spa_activate(spa, spa_mode_global);
2456
2457		if (state != SPA_LOAD_RECOVER)
2458			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2459
2460		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2461		    policy.zrp_request);
2462
2463		if (error == EBADF) {
2464			/*
2465			 * If vdev_validate() returns failure (indicated by
2466			 * EBADF), it indicates that one of the vdevs indicates
2467			 * that the pool has been exported or destroyed.  If
2468			 * this is the case, the config cache is out of sync and
2469			 * we should remove the pool from the namespace.
2470			 */
2471			spa_unload(spa);
2472			spa_deactivate(spa);
2473			spa_config_sync(spa, B_TRUE, B_TRUE);
2474			spa_remove(spa);
2475			if (locked)
2476				mutex_exit(&spa_namespace_lock);
2477			return (ENOENT);
2478		}
2479
2480		if (error) {
2481			/*
2482			 * We can't open the pool, but we still have useful
2483			 * information: the state of each vdev after the
2484			 * attempted vdev_open().  Return this to the user.
2485			 */
2486			if (config != NULL && spa->spa_config) {
2487				VERIFY(nvlist_dup(spa->spa_config, config,
2488				    KM_SLEEP) == 0);
2489				VERIFY(nvlist_add_nvlist(*config,
2490				    ZPOOL_CONFIG_LOAD_INFO,
2491				    spa->spa_load_info) == 0);
2492			}
2493			spa_unload(spa);
2494			spa_deactivate(spa);
2495			spa->spa_last_open_failed = error;
2496			if (locked)
2497				mutex_exit(&spa_namespace_lock);
2498			*spapp = NULL;
2499			return (error);
2500		}
2501	}
2502
2503	spa_open_ref(spa, tag);
2504
2505	if (config != NULL)
2506		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2507
2508	/*
2509	 * If we've recovered the pool, pass back any information we
2510	 * gathered while doing the load.
2511	 */
2512	if (state == SPA_LOAD_RECOVER) {
2513		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2514		    spa->spa_load_info) == 0);
2515	}
2516
2517	if (locked) {
2518		spa->spa_last_open_failed = 0;
2519		spa->spa_last_ubsync_txg = 0;
2520		spa->spa_load_txg = 0;
2521		mutex_exit(&spa_namespace_lock);
2522	}
2523
2524	*spapp = spa;
2525
2526	return (0);
2527}
2528
2529int
2530spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2531    nvlist_t **config)
2532{
2533	return (spa_open_common(name, spapp, tag, policy, config));
2534}
2535
2536int
2537spa_open(const char *name, spa_t **spapp, void *tag)
2538{
2539	return (spa_open_common(name, spapp, tag, NULL, NULL));
2540}
2541
2542/*
2543 * Lookup the given spa_t, incrementing the inject count in the process,
2544 * preventing it from being exported or destroyed.
2545 */
2546spa_t *
2547spa_inject_addref(char *name)
2548{
2549	spa_t *spa;
2550
2551	mutex_enter(&spa_namespace_lock);
2552	if ((spa = spa_lookup(name)) == NULL) {
2553		mutex_exit(&spa_namespace_lock);
2554		return (NULL);
2555	}
2556	spa->spa_inject_ref++;
2557	mutex_exit(&spa_namespace_lock);
2558
2559	return (spa);
2560}
2561
2562void
2563spa_inject_delref(spa_t *spa)
2564{
2565	mutex_enter(&spa_namespace_lock);
2566	spa->spa_inject_ref--;
2567	mutex_exit(&spa_namespace_lock);
2568}
2569
2570/*
2571 * Add spares device information to the nvlist.
2572 */
2573static void
2574spa_add_spares(spa_t *spa, nvlist_t *config)
2575{
2576	nvlist_t **spares;
2577	uint_t i, nspares;
2578	nvlist_t *nvroot;
2579	uint64_t guid;
2580	vdev_stat_t *vs;
2581	uint_t vsc;
2582	uint64_t pool;
2583
2584	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2585
2586	if (spa->spa_spares.sav_count == 0)
2587		return;
2588
2589	VERIFY(nvlist_lookup_nvlist(config,
2590	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2591	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2592	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2593	if (nspares != 0) {
2594		VERIFY(nvlist_add_nvlist_array(nvroot,
2595		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2596		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2597		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2598
2599		/*
2600		 * Go through and find any spares which have since been
2601		 * repurposed as an active spare.  If this is the case, update
2602		 * their status appropriately.
2603		 */
2604		for (i = 0; i < nspares; i++) {
2605			VERIFY(nvlist_lookup_uint64(spares[i],
2606			    ZPOOL_CONFIG_GUID, &guid) == 0);
2607			if (spa_spare_exists(guid, &pool, NULL) &&
2608			    pool != 0ULL) {
2609				VERIFY(nvlist_lookup_uint64_array(
2610				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
2611				    (uint64_t **)&vs, &vsc) == 0);
2612				vs->vs_state = VDEV_STATE_CANT_OPEN;
2613				vs->vs_aux = VDEV_AUX_SPARED;
2614			}
2615		}
2616	}
2617}
2618
2619/*
2620 * Add l2cache device information to the nvlist, including vdev stats.
2621 */
2622static void
2623spa_add_l2cache(spa_t *spa, nvlist_t *config)
2624{
2625	nvlist_t **l2cache;
2626	uint_t i, j, nl2cache;
2627	nvlist_t *nvroot;
2628	uint64_t guid;
2629	vdev_t *vd;
2630	vdev_stat_t *vs;
2631	uint_t vsc;
2632
2633	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2634
2635	if (spa->spa_l2cache.sav_count == 0)
2636		return;
2637
2638	VERIFY(nvlist_lookup_nvlist(config,
2639	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2640	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2641	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2642	if (nl2cache != 0) {
2643		VERIFY(nvlist_add_nvlist_array(nvroot,
2644		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2645		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2646		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2647
2648		/*
2649		 * Update level 2 cache device stats.
2650		 */
2651
2652		for (i = 0; i < nl2cache; i++) {
2653			VERIFY(nvlist_lookup_uint64(l2cache[i],
2654			    ZPOOL_CONFIG_GUID, &guid) == 0);
2655
2656			vd = NULL;
2657			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2658				if (guid ==
2659				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2660					vd = spa->spa_l2cache.sav_vdevs[j];
2661					break;
2662				}
2663			}
2664			ASSERT(vd != NULL);
2665
2666			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2667			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2668			    == 0);
2669			vdev_get_stats(vd, vs);
2670		}
2671	}
2672}
2673
2674int
2675spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2676{
2677	int error;
2678	spa_t *spa;
2679
2680	*config = NULL;
2681	error = spa_open_common(name, &spa, FTAG, NULL, config);
2682
2683	if (spa != NULL) {
2684		/*
2685		 * This still leaves a window of inconsistency where the spares
2686		 * or l2cache devices could change and the config would be
2687		 * self-inconsistent.
2688		 */
2689		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2690
2691		if (*config != NULL) {
2692			uint64_t loadtimes[2];
2693
2694			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
2695			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
2696			VERIFY(nvlist_add_uint64_array(*config,
2697			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
2698
2699			VERIFY(nvlist_add_uint64(*config,
2700			    ZPOOL_CONFIG_ERRCOUNT,
2701			    spa_get_errlog_size(spa)) == 0);
2702
2703			if (spa_suspended(spa))
2704				VERIFY(nvlist_add_uint64(*config,
2705				    ZPOOL_CONFIG_SUSPENDED,
2706				    spa->spa_failmode) == 0);
2707
2708			spa_add_spares(spa, *config);
2709			spa_add_l2cache(spa, *config);
2710		}
2711	}
2712
2713	/*
2714	 * We want to get the alternate root even for faulted pools, so we cheat
2715	 * and call spa_lookup() directly.
2716	 */
2717	if (altroot) {
2718		if (spa == NULL) {
2719			mutex_enter(&spa_namespace_lock);
2720			spa = spa_lookup(name);
2721			if (spa)
2722				spa_altroot(spa, altroot, buflen);
2723			else
2724				altroot[0] = '\0';
2725			spa = NULL;
2726			mutex_exit(&spa_namespace_lock);
2727		} else {
2728			spa_altroot(spa, altroot, buflen);
2729		}
2730	}
2731
2732	if (spa != NULL) {
2733		spa_config_exit(spa, SCL_CONFIG, FTAG);
2734		spa_close(spa, FTAG);
2735	}
2736
2737	return (error);
2738}
2739
2740/*
2741 * Validate that the auxiliary device array is well formed.  We must have an
2742 * array of nvlists, each which describes a valid leaf vdev.  If this is an
2743 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2744 * specified, as long as they are well-formed.
2745 */
2746static int
2747spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2748    spa_aux_vdev_t *sav, const char *config, uint64_t version,
2749    vdev_labeltype_t label)
2750{
2751	nvlist_t **dev;
2752	uint_t i, ndev;
2753	vdev_t *vd;
2754	int error;
2755
2756	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2757
2758	/*
2759	 * It's acceptable to have no devs specified.
2760	 */
2761	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2762		return (0);
2763
2764	if (ndev == 0)
2765		return (EINVAL);
2766
2767	/*
2768	 * Make sure the pool is formatted with a version that supports this
2769	 * device type.
2770	 */
2771	if (spa_version(spa) < version)
2772		return (ENOTSUP);
2773
2774	/*
2775	 * Set the pending device list so we correctly handle device in-use
2776	 * checking.
2777	 */
2778	sav->sav_pending = dev;
2779	sav->sav_npending = ndev;
2780
2781	for (i = 0; i < ndev; i++) {
2782		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2783		    mode)) != 0)
2784			goto out;
2785
2786		if (!vd->vdev_ops->vdev_op_leaf) {
2787			vdev_free(vd);
2788			error = EINVAL;
2789			goto out;
2790		}
2791
2792		/*
2793		 * The L2ARC currently only supports disk devices in
2794		 * kernel context.  For user-level testing, we allow it.
2795		 */
2796#ifdef _KERNEL
2797		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2798		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2799			error = ENOTBLK;
2800			goto out;
2801		}
2802#endif
2803		vd->vdev_top = vd;
2804
2805		if ((error = vdev_open(vd)) == 0 &&
2806		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
2807			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2808			    vd->vdev_guid) == 0);
2809		}
2810
2811		vdev_free(vd);
2812
2813		if (error &&
2814		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2815			goto out;
2816		else
2817			error = 0;
2818	}
2819
2820out:
2821	sav->sav_pending = NULL;
2822	sav->sav_npending = 0;
2823	return (error);
2824}
2825
2826static int
2827spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2828{
2829	int error;
2830
2831	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2832
2833	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2834	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2835	    VDEV_LABEL_SPARE)) != 0) {
2836		return (error);
2837	}
2838
2839	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2840	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2841	    VDEV_LABEL_L2CACHE));
2842}
2843
2844static void
2845spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2846    const char *config)
2847{
2848	int i;
2849
2850	if (sav->sav_config != NULL) {
2851		nvlist_t **olddevs;
2852		uint_t oldndevs;
2853		nvlist_t **newdevs;
2854
2855		/*
2856		 * Generate new dev list by concatentating with the
2857		 * current dev list.
2858		 */
2859		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2860		    &olddevs, &oldndevs) == 0);
2861
2862		newdevs = kmem_alloc(sizeof (void *) *
2863		    (ndevs + oldndevs), KM_SLEEP);
2864		for (i = 0; i < oldndevs; i++)
2865			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2866			    KM_SLEEP) == 0);
2867		for (i = 0; i < ndevs; i++)
2868			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2869			    KM_SLEEP) == 0);
2870
2871		VERIFY(nvlist_remove(sav->sav_config, config,
2872		    DATA_TYPE_NVLIST_ARRAY) == 0);
2873
2874		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2875		    config, newdevs, ndevs + oldndevs) == 0);
2876		for (i = 0; i < oldndevs + ndevs; i++)
2877			nvlist_free(newdevs[i]);
2878		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2879	} else {
2880		/*
2881		 * Generate a new dev list.
2882		 */
2883		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2884		    KM_SLEEP) == 0);
2885		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2886		    devs, ndevs) == 0);
2887	}
2888}
2889
2890/*
2891 * Stop and drop level 2 ARC devices
2892 */
2893void
2894spa_l2cache_drop(spa_t *spa)
2895{
2896	vdev_t *vd;
2897	int i;
2898	spa_aux_vdev_t *sav = &spa->spa_l2cache;
2899
2900	for (i = 0; i < sav->sav_count; i++) {
2901		uint64_t pool;
2902
2903		vd = sav->sav_vdevs[i];
2904		ASSERT(vd != NULL);
2905
2906		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2907		    pool != 0ULL && l2arc_vdev_present(vd))
2908			l2arc_remove_vdev(vd);
2909		if (vd->vdev_isl2cache)
2910			spa_l2cache_remove(vd);
2911		vdev_clear_stats(vd);
2912		(void) vdev_close(vd);
2913	}
2914}
2915
2916/*
2917 * Pool Creation
2918 */
2919int
2920spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2921    const char *history_str, nvlist_t *zplprops)
2922{
2923	spa_t *spa;
2924	char *altroot = NULL;
2925	vdev_t *rvd;
2926	dsl_pool_t *dp;
2927	dmu_tx_t *tx;
2928	int error = 0;
2929	uint64_t txg = TXG_INITIAL;
2930	nvlist_t **spares, **l2cache;
2931	uint_t nspares, nl2cache;
2932	uint64_t version, obj;
2933
2934	/*
2935	 * If this pool already exists, return failure.
2936	 */
2937	mutex_enter(&spa_namespace_lock);
2938	if (spa_lookup(pool) != NULL) {
2939		mutex_exit(&spa_namespace_lock);
2940		return (EEXIST);
2941	}
2942
2943	/*
2944	 * Allocate a new spa_t structure.
2945	 */
2946	(void) nvlist_lookup_string(props,
2947	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2948	spa = spa_add(pool, NULL, altroot);
2949	spa_activate(spa, spa_mode_global);
2950
2951	if (props && (error = spa_prop_validate(spa, props))) {
2952		spa_deactivate(spa);
2953		spa_remove(spa);
2954		mutex_exit(&spa_namespace_lock);
2955		return (error);
2956	}
2957
2958	if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2959	    &version) != 0)
2960		version = SPA_VERSION;
2961	ASSERT(version <= SPA_VERSION);
2962
2963	spa->spa_first_txg = txg;
2964	spa->spa_uberblock.ub_txg = txg - 1;
2965	spa->spa_uberblock.ub_version = version;
2966	spa->spa_ubsync = spa->spa_uberblock;
2967
2968	/*
2969	 * Create "The Godfather" zio to hold all async IOs
2970	 */
2971	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2972	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2973
2974	/*
2975	 * Create the root vdev.
2976	 */
2977	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2978
2979	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2980
2981	ASSERT(error != 0 || rvd != NULL);
2982	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2983
2984	if (error == 0 && !zfs_allocatable_devs(nvroot))
2985		error = EINVAL;
2986
2987	if (error == 0 &&
2988	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2989	    (error = spa_validate_aux(spa, nvroot, txg,
2990	    VDEV_ALLOC_ADD)) == 0) {
2991		for (int c = 0; c < rvd->vdev_children; c++) {
2992			vdev_metaslab_set_size(rvd->vdev_child[c]);
2993			vdev_expand(rvd->vdev_child[c], txg);
2994		}
2995	}
2996
2997	spa_config_exit(spa, SCL_ALL, FTAG);
2998
2999	if (error != 0) {
3000		spa_unload(spa);
3001		spa_deactivate(spa);
3002		spa_remove(spa);
3003		mutex_exit(&spa_namespace_lock);
3004		return (error);
3005	}
3006
3007	/*
3008	 * Get the list of spares, if specified.
3009	 */
3010	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3011	    &spares, &nspares) == 0) {
3012		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3013		    KM_SLEEP) == 0);
3014		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3015		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3016		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3017		spa_load_spares(spa);
3018		spa_config_exit(spa, SCL_ALL, FTAG);
3019		spa->spa_spares.sav_sync = B_TRUE;
3020	}
3021
3022	/*
3023	 * Get the list of level 2 cache devices, if specified.
3024	 */
3025	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3026	    &l2cache, &nl2cache) == 0) {
3027		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3028		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3029		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3030		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3031		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3032		spa_load_l2cache(spa);
3033		spa_config_exit(spa, SCL_ALL, FTAG);
3034		spa->spa_l2cache.sav_sync = B_TRUE;
3035	}
3036
3037	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3038	spa->spa_meta_objset = dp->dp_meta_objset;
3039
3040	/*
3041	 * Create DDTs (dedup tables).
3042	 */
3043	ddt_create(spa);
3044
3045	spa_update_dspace(spa);
3046
3047	tx = dmu_tx_create_assigned(dp, txg);
3048
3049	/*
3050	 * Create the pool config object.
3051	 */
3052	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3053	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3054	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3055
3056	if (zap_add(spa->spa_meta_objset,
3057	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3058	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3059		cmn_err(CE_PANIC, "failed to add pool config");
3060	}
3061
3062	if (zap_add(spa->spa_meta_objset,
3063	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3064	    sizeof (uint64_t), 1, &version, tx) != 0) {
3065		cmn_err(CE_PANIC, "failed to add pool version");
3066	}
3067
3068	/* Newly created pools with the right version are always deflated. */
3069	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3070		spa->spa_deflate = TRUE;
3071		if (zap_add(spa->spa_meta_objset,
3072		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3073		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3074			cmn_err(CE_PANIC, "failed to add deflate");
3075		}
3076	}
3077
3078	/*
3079	 * Create the deferred-free bpobj.  Turn off compression
3080	 * because sync-to-convergence takes longer if the blocksize
3081	 * keeps changing.
3082	 */
3083	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3084	dmu_object_set_compress(spa->spa_meta_objset, obj,
3085	    ZIO_COMPRESS_OFF, tx);
3086	if (zap_add(spa->spa_meta_objset,
3087	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3088	    sizeof (uint64_t), 1, &obj, tx) != 0) {
3089		cmn_err(CE_PANIC, "failed to add bpobj");
3090	}
3091	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3092	    spa->spa_meta_objset, obj));
3093
3094	/*
3095	 * Create the pool's history object.
3096	 */
3097	if (version >= SPA_VERSION_ZPOOL_HISTORY)
3098		spa_history_create_obj(spa, tx);
3099
3100	/*
3101	 * Set pool properties.
3102	 */
3103	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3104	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3105	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3106	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3107
3108	if (props != NULL) {
3109		spa_configfile_set(spa, props, B_FALSE);
3110		spa_sync_props(spa, props, tx);
3111	}
3112
3113	dmu_tx_commit(tx);
3114
3115	spa->spa_sync_on = B_TRUE;
3116	txg_sync_start(spa->spa_dsl_pool);
3117
3118	/*
3119	 * We explicitly wait for the first transaction to complete so that our
3120	 * bean counters are appropriately updated.
3121	 */
3122	txg_wait_synced(spa->spa_dsl_pool, txg);
3123
3124	spa_config_sync(spa, B_FALSE, B_TRUE);
3125
3126	if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3127		(void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3128	spa_history_log_version(spa, LOG_POOL_CREATE);
3129
3130	spa->spa_minref = refcount_count(&spa->spa_refcount);
3131
3132	mutex_exit(&spa_namespace_lock);
3133
3134	return (0);
3135}
3136
3137#ifdef _KERNEL
3138/*
3139 * Get the root pool information from the root disk, then import the root pool
3140 * during the system boot up time.
3141 */
3142extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3143
3144static nvlist_t *
3145spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3146{
3147	nvlist_t *config;
3148	nvlist_t *nvtop, *nvroot;
3149	uint64_t pgid;
3150
3151	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3152		return (NULL);
3153
3154	/*
3155	 * Add this top-level vdev to the child array.
3156	 */
3157	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3158	    &nvtop) == 0);
3159	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3160	    &pgid) == 0);
3161	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3162
3163	/*
3164	 * Put this pool's top-level vdevs into a root vdev.
3165	 */
3166	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3167	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3168	    VDEV_TYPE_ROOT) == 0);
3169	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3170	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3171	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3172	    &nvtop, 1) == 0);
3173
3174	/*
3175	 * Replace the existing vdev_tree with the new root vdev in
3176	 * this pool's configuration (remove the old, add the new).
3177	 */
3178	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3179	nvlist_free(nvroot);
3180	return (config);
3181}
3182
3183/*
3184 * Walk the vdev tree and see if we can find a device with "better"
3185 * configuration. A configuration is "better" if the label on that
3186 * device has a more recent txg.
3187 */
3188static void
3189spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3190{
3191	for (int c = 0; c < vd->vdev_children; c++)
3192		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3193
3194	if (vd->vdev_ops->vdev_op_leaf) {
3195		nvlist_t *label;
3196		uint64_t label_txg;
3197
3198		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3199		    &label) != 0)
3200			return;
3201
3202		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3203		    &label_txg) == 0);
3204
3205		/*
3206		 * Do we have a better boot device?
3207		 */
3208		if (label_txg > *txg) {
3209			*txg = label_txg;
3210			*avd = vd;
3211		}
3212		nvlist_free(label);
3213	}
3214}
3215
3216/*
3217 * Import a root pool.
3218 *
3219 * For x86. devpath_list will consist of devid and/or physpath name of
3220 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3221 * The GRUB "findroot" command will return the vdev we should boot.
3222 *
3223 * For Sparc, devpath_list consists the physpath name of the booting device
3224 * no matter the rootpool is a single device pool or a mirrored pool.
3225 * e.g.
3226 *	"/pci@1f,0/ide@d/disk@0,0:a"
3227 */
3228int
3229spa_import_rootpool(char *devpath, char *devid)
3230{
3231	spa_t *spa;
3232	vdev_t *rvd, *bvd, *avd = NULL;
3233	nvlist_t *config, *nvtop;
3234	uint64_t guid, txg;
3235	char *pname;
3236	int error;
3237
3238	/*
3239	 * Read the label from the boot device and generate a configuration.
3240	 */
3241	config = spa_generate_rootconf(devpath, devid, &guid);
3242#if defined(_OBP) && defined(_KERNEL)
3243	if (config == NULL) {
3244		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3245			/* iscsi boot */
3246			get_iscsi_bootpath_phy(devpath);
3247			config = spa_generate_rootconf(devpath, devid, &guid);
3248		}
3249	}
3250#endif
3251	if (config == NULL) {
3252		cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
3253		    devpath);
3254		return (EIO);
3255	}
3256
3257	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3258	    &pname) == 0);
3259	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3260
3261	mutex_enter(&spa_namespace_lock);
3262	if ((spa = spa_lookup(pname)) != NULL) {
3263		/*
3264		 * Remove the existing root pool from the namespace so that we
3265		 * can replace it with the correct config we just read in.
3266		 */
3267		spa_remove(spa);
3268	}
3269
3270	spa = spa_add(pname, config, NULL);
3271	spa->spa_is_root = B_TRUE;
3272	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3273
3274	/*
3275	 * Build up a vdev tree based on the boot device's label config.
3276	 */
3277	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3278	    &nvtop) == 0);
3279	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3280	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3281	    VDEV_ALLOC_ROOTPOOL);
3282	spa_config_exit(spa, SCL_ALL, FTAG);
3283	if (error) {
3284		mutex_exit(&spa_namespace_lock);
3285		nvlist_free(config);
3286		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3287		    pname);
3288		return (error);
3289	}
3290
3291	/*
3292	 * Get the boot vdev.
3293	 */
3294	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3295		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3296		    (u_longlong_t)guid);
3297		error = ENOENT;
3298		goto out;
3299	}
3300
3301	/*
3302	 * Determine if there is a better boot device.
3303	 */
3304	avd = bvd;
3305	spa_alt_rootvdev(rvd, &avd, &txg);
3306	if (avd != bvd) {
3307		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3308		    "try booting from '%s'", avd->vdev_path);
3309		error = EINVAL;
3310		goto out;
3311	}
3312
3313	/*
3314	 * If the boot device is part of a spare vdev then ensure that
3315	 * we're booting off the active spare.
3316	 */
3317	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3318	    !bvd->vdev_isspare) {
3319		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3320		    "try booting from '%s'",
3321		    bvd->vdev_parent->
3322		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3323		error = EINVAL;
3324		goto out;
3325	}
3326
3327	error = 0;
3328	spa_history_log_version(spa, LOG_POOL_IMPORT);
3329out:
3330	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3331	vdev_free(rvd);
3332	spa_config_exit(spa, SCL_ALL, FTAG);
3333	mutex_exit(&spa_namespace_lock);
3334
3335	nvlist_free(config);
3336	return (error);
3337}
3338
3339#endif
3340
3341/*
3342 * Import a non-root pool into the system.
3343 */
3344int
3345spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3346{
3347	spa_t *spa;
3348	char *altroot = NULL;
3349	spa_load_state_t state = SPA_LOAD_IMPORT;
3350	zpool_rewind_policy_t policy;
3351	uint64_t mode = spa_mode_global;
3352	uint64_t readonly = B_FALSE;
3353	int error;
3354	nvlist_t *nvroot;
3355	nvlist_t **spares, **l2cache;
3356	uint_t nspares, nl2cache;
3357
3358	/*
3359	 * If a pool with this name exists, return failure.
3360	 */
3361	mutex_enter(&spa_namespace_lock);
3362	if (spa_lookup(pool) != NULL) {
3363		mutex_exit(&spa_namespace_lock);
3364		return (EEXIST);
3365	}
3366
3367	/*
3368	 * Create and initialize the spa structure.
3369	 */
3370	(void) nvlist_lookup_string(props,
3371	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3372	(void) nvlist_lookup_uint64(props,
3373	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3374	if (readonly)
3375		mode = FREAD;
3376	spa = spa_add(pool, config, altroot);
3377	spa->spa_import_flags = flags;
3378
3379	/*
3380	 * Verbatim import - Take a pool and insert it into the namespace
3381	 * as if it had been loaded at boot.
3382	 */
3383	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3384		if (props != NULL)
3385			spa_configfile_set(spa, props, B_FALSE);
3386
3387		spa_config_sync(spa, B_FALSE, B_TRUE);
3388
3389		mutex_exit(&spa_namespace_lock);
3390		spa_history_log_version(spa, LOG_POOL_IMPORT);
3391
3392		return (0);
3393	}
3394
3395	spa_activate(spa, mode);
3396
3397	/*
3398	 * Don't start async tasks until we know everything is healthy.
3399	 */
3400	spa_async_suspend(spa);
3401
3402	zpool_get_rewind_policy(config, &policy);
3403	if (policy.zrp_request & ZPOOL_DO_REWIND)
3404		state = SPA_LOAD_RECOVER;
3405
3406	/*
3407	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3408	 * because the user-supplied config is actually the one to trust when
3409	 * doing an import.
3410	 */
3411	if (state != SPA_LOAD_RECOVER)
3412		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3413
3414	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3415	    policy.zrp_request);
3416
3417	/*
3418	 * Propagate anything learned while loading the pool and pass it
3419	 * back to caller (i.e. rewind info, missing devices, etc).
3420	 */
3421	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3422	    spa->spa_load_info) == 0);
3423
3424	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3425	/*
3426	 * Toss any existing sparelist, as it doesn't have any validity
3427	 * anymore, and conflicts with spa_has_spare().
3428	 */
3429	if (spa->spa_spares.sav_config) {
3430		nvlist_free(spa->spa_spares.sav_config);
3431		spa->spa_spares.sav_config = NULL;
3432		spa_load_spares(spa);
3433	}
3434	if (spa->spa_l2cache.sav_config) {
3435		nvlist_free(spa->spa_l2cache.sav_config);
3436		spa->spa_l2cache.sav_config = NULL;
3437		spa_load_l2cache(spa);
3438	}
3439
3440	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3441	    &nvroot) == 0);
3442	if (error == 0)
3443		error = spa_validate_aux(spa, nvroot, -1ULL,
3444		    VDEV_ALLOC_SPARE);
3445	if (error == 0)
3446		error = spa_validate_aux(spa, nvroot, -1ULL,
3447		    VDEV_ALLOC_L2CACHE);
3448	spa_config_exit(spa, SCL_ALL, FTAG);
3449
3450	if (props != NULL)
3451		spa_configfile_set(spa, props, B_FALSE);
3452
3453	if (error != 0 || (props && spa_writeable(spa) &&
3454	    (error = spa_prop_set(spa, props)))) {
3455		spa_unload(spa);
3456		spa_deactivate(spa);
3457		spa_remove(spa);
3458		mutex_exit(&spa_namespace_lock);
3459		return (error);
3460	}
3461
3462	spa_async_resume(spa);
3463
3464	/*
3465	 * Override any spares and level 2 cache devices as specified by
3466	 * the user, as these may have correct device names/devids, etc.
3467	 */
3468	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3469	    &spares, &nspares) == 0) {
3470		if (spa->spa_spares.sav_config)
3471			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3472			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3473		else
3474			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3475			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3476		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3477		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3478		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3479		spa_load_spares(spa);
3480		spa_config_exit(spa, SCL_ALL, FTAG);
3481		spa->spa_spares.sav_sync = B_TRUE;
3482	}
3483	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3484	    &l2cache, &nl2cache) == 0) {
3485		if (spa->spa_l2cache.sav_config)
3486			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3487			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3488		else
3489			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3490			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3491		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3492		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3493		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3494		spa_load_l2cache(spa);
3495		spa_config_exit(spa, SCL_ALL, FTAG);
3496		spa->spa_l2cache.sav_sync = B_TRUE;
3497	}
3498
3499	/*
3500	 * Check for any removed devices.
3501	 */
3502	if (spa->spa_autoreplace) {
3503		spa_aux_check_removed(&spa->spa_spares);
3504		spa_aux_check_removed(&spa->spa_l2cache);
3505	}
3506
3507	if (spa_writeable(spa)) {
3508		/*
3509		 * Update the config cache to include the newly-imported pool.
3510		 */
3511		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3512	}
3513
3514	/*
3515	 * It's possible that the pool was expanded while it was exported.
3516	 * We kick off an async task to handle this for us.
3517	 */
3518	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3519
3520	mutex_exit(&spa_namespace_lock);
3521	spa_history_log_version(spa, LOG_POOL_IMPORT);
3522
3523	return (0);
3524}
3525
3526nvlist_t *
3527spa_tryimport(nvlist_t *tryconfig)
3528{
3529	nvlist_t *config = NULL;
3530	char *poolname;
3531	spa_t *spa;
3532	uint64_t state;
3533	int error;
3534
3535	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3536		return (NULL);
3537
3538	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3539		return (NULL);
3540
3541	/*
3542	 * Create and initialize the spa structure.
3543	 */
3544	mutex_enter(&spa_namespace_lock);
3545	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3546	spa_activate(spa, FREAD);
3547
3548	/*
3549	 * Pass off the heavy lifting to spa_load().
3550	 * Pass TRUE for mosconfig because the user-supplied config
3551	 * is actually the one to trust when doing an import.
3552	 */
3553	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3554
3555	/*
3556	 * If 'tryconfig' was at least parsable, return the current config.
3557	 */
3558	if (spa->spa_root_vdev != NULL) {
3559		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3560		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3561		    poolname) == 0);
3562		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3563		    state) == 0);
3564		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3565		    spa->spa_uberblock.ub_timestamp) == 0);
3566
3567		/*
3568		 * If the bootfs property exists on this pool then we
3569		 * copy it out so that external consumers can tell which
3570		 * pools are bootable.
3571		 */
3572		if ((!error || error == EEXIST) && spa->spa_bootfs) {
3573			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3574
3575			/*
3576			 * We have to play games with the name since the
3577			 * pool was opened as TRYIMPORT_NAME.
3578			 */
3579			if (dsl_dsobj_to_dsname(spa_name(spa),
3580			    spa->spa_bootfs, tmpname) == 0) {
3581				char *cp;
3582				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3583
3584				cp = strchr(tmpname, '/');
3585				if (cp == NULL) {
3586					(void) strlcpy(dsname, tmpname,
3587					    MAXPATHLEN);
3588				} else {
3589					(void) snprintf(dsname, MAXPATHLEN,
3590					    "%s/%s", poolname, ++cp);
3591				}
3592				VERIFY(nvlist_add_string(config,
3593				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3594				kmem_free(dsname, MAXPATHLEN);
3595			}
3596			kmem_free(tmpname, MAXPATHLEN);
3597		}
3598
3599		/*
3600		 * Add the list of hot spares and level 2 cache devices.
3601		 */
3602		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3603		spa_add_spares(spa, config);
3604		spa_add_l2cache(spa, config);
3605		spa_config_exit(spa, SCL_CONFIG, FTAG);
3606	}
3607
3608	spa_unload(spa);
3609	spa_deactivate(spa);
3610	spa_remove(spa);
3611	mutex_exit(&spa_namespace_lock);
3612
3613	return (config);
3614}
3615
3616/*
3617 * Pool export/destroy
3618 *
3619 * The act of destroying or exporting a pool is very simple.  We make sure there
3620 * is no more pending I/O and any references to the pool are gone.  Then, we
3621 * update the pool state and sync all the labels to disk, removing the
3622 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3623 * we don't sync the labels or remove the configuration cache.
3624 */
3625static int
3626spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3627    boolean_t force, boolean_t hardforce)
3628{
3629	spa_t *spa;
3630
3631	if (oldconfig)
3632		*oldconfig = NULL;
3633
3634	if (!(spa_mode_global & FWRITE))
3635		return (EROFS);
3636
3637	mutex_enter(&spa_namespace_lock);
3638	if ((spa = spa_lookup(pool)) == NULL) {
3639		mutex_exit(&spa_namespace_lock);
3640		return (ENOENT);
3641	}
3642
3643	/*
3644	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3645	 * reacquire the namespace lock, and see if we can export.
3646	 */
3647	spa_open_ref(spa, FTAG);
3648	mutex_exit(&spa_namespace_lock);
3649	spa_async_suspend(spa);
3650	mutex_enter(&spa_namespace_lock);
3651	spa_close(spa, FTAG);
3652
3653	/*
3654	 * The pool will be in core if it's openable,
3655	 * in which case we can modify its state.
3656	 */
3657	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3658		/*
3659		 * Objsets may be open only because they're dirty, so we
3660		 * have to force it to sync before checking spa_refcnt.
3661		 */
3662		txg_wait_synced(spa->spa_dsl_pool, 0);
3663
3664		/*
3665		 * A pool cannot be exported or destroyed if there are active
3666		 * references.  If we are resetting a pool, allow references by
3667		 * fault injection handlers.
3668		 */
3669		if (!spa_refcount_zero(spa) ||
3670		    (spa->spa_inject_ref != 0 &&
3671		    new_state != POOL_STATE_UNINITIALIZED)) {
3672			spa_async_resume(spa);
3673			mutex_exit(&spa_namespace_lock);
3674			return (EBUSY);
3675		}
3676
3677		/*
3678		 * A pool cannot be exported if it has an active shared spare.
3679		 * This is to prevent other pools stealing the active spare
3680		 * from an exported pool. At user's own will, such pool can
3681		 * be forcedly exported.
3682		 */
3683		if (!force && new_state == POOL_STATE_EXPORTED &&
3684		    spa_has_active_shared_spare(spa)) {
3685			spa_async_resume(spa);
3686			mutex_exit(&spa_namespace_lock);
3687			return (EXDEV);
3688		}
3689
3690		/*
3691		 * We want this to be reflected on every label,
3692		 * so mark them all dirty.  spa_unload() will do the
3693		 * final sync that pushes these changes out.
3694		 */
3695		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3696			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3697			spa->spa_state = new_state;
3698			spa->spa_final_txg = spa_last_synced_txg(spa) +
3699			    TXG_DEFER_SIZE + 1;
3700			vdev_config_dirty(spa->spa_root_vdev);
3701			spa_config_exit(spa, SCL_ALL, FTAG);
3702		}
3703	}
3704
3705	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
3706
3707	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3708		spa_unload(spa);
3709		spa_deactivate(spa);
3710	}
3711
3712	if (oldconfig && spa->spa_config)
3713		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3714
3715	if (new_state != POOL_STATE_UNINITIALIZED) {
3716		if (!hardforce)
3717			spa_config_sync(spa, B_TRUE, B_TRUE);
3718		spa_remove(spa);
3719	}
3720	mutex_exit(&spa_namespace_lock);
3721
3722	return (0);
3723}
3724
3725/*
3726 * Destroy a storage pool.
3727 */
3728int
3729spa_destroy(char *pool)
3730{
3731	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3732	    B_FALSE, B_FALSE));
3733}
3734
3735/*
3736 * Export a storage pool.
3737 */
3738int
3739spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3740    boolean_t hardforce)
3741{
3742	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3743	    force, hardforce));
3744}
3745
3746/*
3747 * Similar to spa_export(), this unloads the spa_t without actually removing it
3748 * from the namespace in any way.
3749 */
3750int
3751spa_reset(char *pool)
3752{
3753	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3754	    B_FALSE, B_FALSE));
3755}
3756
3757/*
3758 * ==========================================================================
3759 * Device manipulation
3760 * ==========================================================================
3761 */
3762
3763/*
3764 * Add a device to a storage pool.
3765 */
3766int
3767spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3768{
3769	uint64_t txg, id;
3770	int error;
3771	vdev_t *rvd = spa->spa_root_vdev;
3772	vdev_t *vd, *tvd;
3773	nvlist_t **spares, **l2cache;
3774	uint_t nspares, nl2cache;
3775
3776	ASSERT(spa_writeable(spa));
3777
3778	txg = spa_vdev_enter(spa);
3779
3780	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3781	    VDEV_ALLOC_ADD)) != 0)
3782		return (spa_vdev_exit(spa, NULL, txg, error));
3783
3784	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
3785
3786	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3787	    &nspares) != 0)
3788		nspares = 0;
3789
3790	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3791	    &nl2cache) != 0)
3792		nl2cache = 0;
3793
3794	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3795		return (spa_vdev_exit(spa, vd, txg, EINVAL));
3796
3797	if (vd->vdev_children != 0 &&
3798	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
3799		return (spa_vdev_exit(spa, vd, txg, error));
3800
3801	/*
3802	 * We must validate the spares and l2cache devices after checking the
3803	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
3804	 */
3805	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3806		return (spa_vdev_exit(spa, vd, txg, error));
3807
3808	/*
3809	 * Transfer each new top-level vdev from vd to rvd.
3810	 */
3811	for (int c = 0; c < vd->vdev_children; c++) {
3812
3813		/*
3814		 * Set the vdev id to the first hole, if one exists.
3815		 */
3816		for (id = 0; id < rvd->vdev_children; id++) {
3817			if (rvd->vdev_child[id]->vdev_ishole) {
3818				vdev_free(rvd->vdev_child[id]);
3819				break;
3820			}
3821		}
3822		tvd = vd->vdev_child[c];
3823		vdev_remove_child(vd, tvd);
3824		tvd->vdev_id = id;
3825		vdev_add_child(rvd, tvd);
3826		vdev_config_dirty(tvd);
3827	}
3828
3829	if (nspares != 0) {
3830		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3831		    ZPOOL_CONFIG_SPARES);
3832		spa_load_spares(spa);
3833		spa->spa_spares.sav_sync = B_TRUE;
3834	}
3835
3836	if (nl2cache != 0) {
3837		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3838		    ZPOOL_CONFIG_L2CACHE);
3839		spa_load_l2cache(spa);
3840		spa->spa_l2cache.sav_sync = B_TRUE;
3841	}
3842
3843	/*
3844	 * We have to be careful when adding new vdevs to an existing pool.
3845	 * If other threads start allocating from these vdevs before we
3846	 * sync the config cache, and we lose power, then upon reboot we may
3847	 * fail to open the pool because there are DVAs that the config cache
3848	 * can't translate.  Therefore, we first add the vdevs without
3849	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3850	 * and then let spa_config_update() initialize the new metaslabs.
3851	 *
3852	 * spa_load() checks for added-but-not-initialized vdevs, so that
3853	 * if we lose power at any point in this sequence, the remaining
3854	 * steps will be completed the next time we load the pool.
3855	 */
3856	(void) spa_vdev_exit(spa, vd, txg, 0);
3857
3858	mutex_enter(&spa_namespace_lock);
3859	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3860	mutex_exit(&spa_namespace_lock);
3861
3862	return (0);
3863}
3864
3865/*
3866 * Attach a device to a mirror.  The arguments are the path to any device
3867 * in the mirror, and the nvroot for the new device.  If the path specifies
3868 * a device that is not mirrored, we automatically insert the mirror vdev.
3869 *
3870 * If 'replacing' is specified, the new device is intended to replace the
3871 * existing device; in this case the two devices are made into their own
3872 * mirror using the 'replacing' vdev, which is functionally identical to
3873 * the mirror vdev (it actually reuses all the same ops) but has a few
3874 * extra rules: you can't attach to it after it's been created, and upon
3875 * completion of resilvering, the first disk (the one being replaced)
3876 * is automatically detached.
3877 */
3878int
3879spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3880{
3881	uint64_t txg, dtl_max_txg;
3882	vdev_t *rvd = spa->spa_root_vdev;
3883	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3884	vdev_ops_t *pvops;
3885	char *oldvdpath, *newvdpath;
3886	int newvd_isspare;
3887	int error;
3888
3889	ASSERT(spa_writeable(spa));
3890
3891	txg = spa_vdev_enter(spa);
3892
3893	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3894
3895	if (oldvd == NULL)
3896		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3897
3898	if (!oldvd->vdev_ops->vdev_op_leaf)
3899		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3900
3901	pvd = oldvd->vdev_parent;
3902
3903	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3904	    VDEV_ALLOC_ADD)) != 0)
3905		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3906
3907	if (newrootvd->vdev_children != 1)
3908		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3909
3910	newvd = newrootvd->vdev_child[0];
3911
3912	if (!newvd->vdev_ops->vdev_op_leaf)
3913		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3914
3915	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3916		return (spa_vdev_exit(spa, newrootvd, txg, error));
3917
3918	/*
3919	 * Spares can't replace logs
3920	 */
3921	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3922		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3923
3924	if (!replacing) {
3925		/*
3926		 * For attach, the only allowable parent is a mirror or the root
3927		 * vdev.
3928		 */
3929		if (pvd->vdev_ops != &vdev_mirror_ops &&
3930		    pvd->vdev_ops != &vdev_root_ops)
3931			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3932
3933		pvops = &vdev_mirror_ops;
3934	} else {
3935		/*
3936		 * Active hot spares can only be replaced by inactive hot
3937		 * spares.
3938		 */
3939		if (pvd->vdev_ops == &vdev_spare_ops &&
3940		    oldvd->vdev_isspare &&
3941		    !spa_has_spare(spa, newvd->vdev_guid))
3942			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3943
3944		/*
3945		 * If the source is a hot spare, and the parent isn't already a
3946		 * spare, then we want to create a new hot spare.  Otherwise, we
3947		 * want to create a replacing vdev.  The user is not allowed to
3948		 * attach to a spared vdev child unless the 'isspare' state is
3949		 * the same (spare replaces spare, non-spare replaces
3950		 * non-spare).
3951		 */
3952		if (pvd->vdev_ops == &vdev_replacing_ops &&
3953		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
3954			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3955		} else if (pvd->vdev_ops == &vdev_spare_ops &&
3956		    newvd->vdev_isspare != oldvd->vdev_isspare) {
3957			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3958		}
3959
3960		if (newvd->vdev_isspare)
3961			pvops = &vdev_spare_ops;
3962		else
3963			pvops = &vdev_replacing_ops;
3964	}
3965
3966	/*
3967	 * Make sure the new device is big enough.
3968	 */
3969	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3970		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3971
3972	/*
3973	 * The new device cannot have a higher alignment requirement
3974	 * than the top-level vdev.
3975	 */
3976	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3977		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3978
3979	/*
3980	 * If this is an in-place replacement, update oldvd's path and devid
3981	 * to make it distinguishable from newvd, and unopenable from now on.
3982	 */
3983	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3984		spa_strfree(oldvd->vdev_path);
3985		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3986		    KM_SLEEP);
3987		(void) sprintf(oldvd->vdev_path, "%s/%s",
3988		    newvd->vdev_path, "old");
3989		if (oldvd->vdev_devid != NULL) {
3990			spa_strfree(oldvd->vdev_devid);
3991			oldvd->vdev_devid = NULL;
3992		}
3993	}
3994
3995	/* mark the device being resilvered */
3996	newvd->vdev_resilvering = B_TRUE;
3997
3998	/*
3999	 * If the parent is not a mirror, or if we're replacing, insert the new
4000	 * mirror/replacing/spare vdev above oldvd.
4001	 */
4002	if (pvd->vdev_ops != pvops)
4003		pvd = vdev_add_parent(oldvd, pvops);
4004
4005	ASSERT(pvd->vdev_top->vdev_parent == rvd);
4006	ASSERT(pvd->vdev_ops == pvops);
4007	ASSERT(oldvd->vdev_parent == pvd);
4008
4009	/*
4010	 * Extract the new device from its root and add it to pvd.
4011	 */
4012	vdev_remove_child(newrootvd, newvd);
4013	newvd->vdev_id = pvd->vdev_children;
4014	newvd->vdev_crtxg = oldvd->vdev_crtxg;
4015	vdev_add_child(pvd, newvd);
4016
4017	tvd = newvd->vdev_top;
4018	ASSERT(pvd->vdev_top == tvd);
4019	ASSERT(tvd->vdev_parent == rvd);
4020
4021	vdev_config_dirty(tvd);
4022
4023	/*
4024	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4025	 * for any dmu_sync-ed blocks.  It will propagate upward when
4026	 * spa_vdev_exit() calls vdev_dtl_reassess().
4027	 */
4028	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4029
4030	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4031	    dtl_max_txg - TXG_INITIAL);
4032
4033	if (newvd->vdev_isspare) {
4034		spa_spare_activate(newvd);
4035		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4036	}
4037
4038	oldvdpath = spa_strdup(oldvd->vdev_path);
4039	newvdpath = spa_strdup(newvd->vdev_path);
4040	newvd_isspare = newvd->vdev_isspare;
4041
4042	/*
4043	 * Mark newvd's DTL dirty in this txg.
4044	 */
4045	vdev_dirty(tvd, VDD_DTL, newvd, txg);
4046
4047	/*
4048	 * Restart the resilver
4049	 */
4050	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4051
4052	/*
4053	 * Commit the config
4054	 */
4055	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4056
4057	spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
4058	    "%s vdev=%s %s vdev=%s",
4059	    replacing && newvd_isspare ? "spare in" :
4060	    replacing ? "replace" : "attach", newvdpath,
4061	    replacing ? "for" : "to", oldvdpath);
4062
4063	spa_strfree(oldvdpath);
4064	spa_strfree(newvdpath);
4065
4066	if (spa->spa_bootfs)
4067		spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4068
4069	return (0);
4070}
4071
4072/*
4073 * Detach a device from a mirror or replacing vdev.
4074 * If 'replace_done' is specified, only detach if the parent
4075 * is a replacing vdev.
4076 */
4077int
4078spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4079{
4080	uint64_t txg;
4081	int error;
4082	vdev_t *rvd = spa->spa_root_vdev;
4083	vdev_t *vd, *pvd, *cvd, *tvd;
4084	boolean_t unspare = B_FALSE;
4085	uint64_t unspare_guid;
4086	char *vdpath;
4087
4088	ASSERT(spa_writeable(spa));
4089
4090	txg = spa_vdev_enter(spa);
4091
4092	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4093
4094	if (vd == NULL)
4095		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4096
4097	if (!vd->vdev_ops->vdev_op_leaf)
4098		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4099
4100	pvd = vd->vdev_parent;
4101
4102	/*
4103	 * If the parent/child relationship is not as expected, don't do it.
4104	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4105	 * vdev that's replacing B with C.  The user's intent in replacing
4106	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4107	 * the replace by detaching C, the expected behavior is to end up
4108	 * M(A,B).  But suppose that right after deciding to detach C,
4109	 * the replacement of B completes.  We would have M(A,C), and then
4110	 * ask to detach C, which would leave us with just A -- not what
4111	 * the user wanted.  To prevent this, we make sure that the
4112	 * parent/child relationship hasn't changed -- in this example,
4113	 * that C's parent is still the replacing vdev R.
4114	 */
4115	if (pvd->vdev_guid != pguid && pguid != 0)
4116		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4117
4118	/*
4119	 * Only 'replacing' or 'spare' vdevs can be replaced.
4120	 */
4121	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4122	    pvd->vdev_ops != &vdev_spare_ops)
4123		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4124
4125	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4126	    spa_version(spa) >= SPA_VERSION_SPARES);
4127
4128	/*
4129	 * Only mirror, replacing, and spare vdevs support detach.
4130	 */
4131	if (pvd->vdev_ops != &vdev_replacing_ops &&
4132	    pvd->vdev_ops != &vdev_mirror_ops &&
4133	    pvd->vdev_ops != &vdev_spare_ops)
4134		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4135
4136	/*
4137	 * If this device has the only valid copy of some data,
4138	 * we cannot safely detach it.
4139	 */
4140	if (vdev_dtl_required(vd))
4141		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4142
4143	ASSERT(pvd->vdev_children >= 2);
4144
4145	/*
4146	 * If we are detaching the second disk from a replacing vdev, then
4147	 * check to see if we changed the original vdev's path to have "/old"
4148	 * at the end in spa_vdev_attach().  If so, undo that change now.
4149	 */
4150	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4151	    vd->vdev_path != NULL) {
4152		size_t len = strlen(vd->vdev_path);
4153
4154		for (int c = 0; c < pvd->vdev_children; c++) {
4155			cvd = pvd->vdev_child[c];
4156
4157			if (cvd == vd || cvd->vdev_path == NULL)
4158				continue;
4159
4160			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4161			    strcmp(cvd->vdev_path + len, "/old") == 0) {
4162				spa_strfree(cvd->vdev_path);
4163				cvd->vdev_path = spa_strdup(vd->vdev_path);
4164				break;
4165			}
4166		}
4167	}
4168
4169	/*
4170	 * If we are detaching the original disk from a spare, then it implies
4171	 * that the spare should become a real disk, and be removed from the
4172	 * active spare list for the pool.
4173	 */
4174	if (pvd->vdev_ops == &vdev_spare_ops &&
4175	    vd->vdev_id == 0 &&
4176	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4177		unspare = B_TRUE;
4178
4179	/*
4180	 * Erase the disk labels so the disk can be used for other things.
4181	 * This must be done after all other error cases are handled,
4182	 * but before we disembowel vd (so we can still do I/O to it).
4183	 * But if we can't do it, don't treat the error as fatal --
4184	 * it may be that the unwritability of the disk is the reason
4185	 * it's being detached!
4186	 */
4187	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4188
4189	/*
4190	 * Remove vd from its parent and compact the parent's children.
4191	 */
4192	vdev_remove_child(pvd, vd);
4193	vdev_compact_children(pvd);
4194
4195	/*
4196	 * Remember one of the remaining children so we can get tvd below.
4197	 */
4198	cvd = pvd->vdev_child[pvd->vdev_children - 1];
4199
4200	/*
4201	 * If we need to remove the remaining child from the list of hot spares,
4202	 * do it now, marking the vdev as no longer a spare in the process.
4203	 * We must do this before vdev_remove_parent(), because that can
4204	 * change the GUID if it creates a new toplevel GUID.  For a similar
4205	 * reason, we must remove the spare now, in the same txg as the detach;
4206	 * otherwise someone could attach a new sibling, change the GUID, and
4207	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4208	 */
4209	if (unspare) {
4210		ASSERT(cvd->vdev_isspare);
4211		spa_spare_remove(cvd);
4212		unspare_guid = cvd->vdev_guid;
4213		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4214		cvd->vdev_unspare = B_TRUE;
4215	}
4216
4217	/*
4218	 * If the parent mirror/replacing vdev only has one child,
4219	 * the parent is no longer needed.  Remove it from the tree.
4220	 */
4221	if (pvd->vdev_children == 1) {
4222		if (pvd->vdev_ops == &vdev_spare_ops)
4223			cvd->vdev_unspare = B_FALSE;
4224		vdev_remove_parent(cvd);
4225		cvd->vdev_resilvering = B_FALSE;
4226	}
4227
4228
4229	/*
4230	 * We don't set tvd until now because the parent we just removed
4231	 * may have been the previous top-level vdev.
4232	 */
4233	tvd = cvd->vdev_top;
4234	ASSERT(tvd->vdev_parent == rvd);
4235
4236	/*
4237	 * Reevaluate the parent vdev state.
4238	 */
4239	vdev_propagate_state(cvd);
4240
4241	/*
4242	 * If the 'autoexpand' property is set on the pool then automatically
4243	 * try to expand the size of the pool. For example if the device we
4244	 * just detached was smaller than the others, it may be possible to
4245	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4246	 * first so that we can obtain the updated sizes of the leaf vdevs.
4247	 */
4248	if (spa->spa_autoexpand) {
4249		vdev_reopen(tvd);
4250		vdev_expand(tvd, txg);
4251	}
4252
4253	vdev_config_dirty(tvd);
4254
4255	/*
4256	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4257	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4258	 * But first make sure we're not on any *other* txg's DTL list, to
4259	 * prevent vd from being accessed after it's freed.
4260	 */
4261	vdpath = spa_strdup(vd->vdev_path);
4262	for (int t = 0; t < TXG_SIZE; t++)
4263		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4264	vd->vdev_detached = B_TRUE;
4265	vdev_dirty(tvd, VDD_DTL, vd, txg);
4266
4267	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4268
4269	/* hang on to the spa before we release the lock */
4270	spa_open_ref(spa, FTAG);
4271
4272	error = spa_vdev_exit(spa, vd, txg, 0);
4273
4274	spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4275	    "vdev=%s", vdpath);
4276	spa_strfree(vdpath);
4277
4278	/*
4279	 * If this was the removal of the original device in a hot spare vdev,
4280	 * then we want to go through and remove the device from the hot spare
4281	 * list of every other pool.
4282	 */
4283	if (unspare) {
4284		spa_t *altspa = NULL;
4285
4286		mutex_enter(&spa_namespace_lock);
4287		while ((altspa = spa_next(altspa)) != NULL) {
4288			if (altspa->spa_state != POOL_STATE_ACTIVE ||
4289			    altspa == spa)
4290				continue;
4291
4292			spa_open_ref(altspa, FTAG);
4293			mutex_exit(&spa_namespace_lock);
4294			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4295			mutex_enter(&spa_namespace_lock);
4296			spa_close(altspa, FTAG);
4297		}
4298		mutex_exit(&spa_namespace_lock);
4299
4300		/* search the rest of the vdevs for spares to remove */
4301		spa_vdev_resilver_done(spa);
4302	}
4303
4304	/* all done with the spa; OK to release */
4305	mutex_enter(&spa_namespace_lock);
4306	spa_close(spa, FTAG);
4307	mutex_exit(&spa_namespace_lock);
4308
4309	return (error);
4310}
4311
4312/*
4313 * Split a set of devices from their mirrors, and create a new pool from them.
4314 */
4315int
4316spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4317    nvlist_t *props, boolean_t exp)
4318{
4319	int error = 0;
4320	uint64_t txg, *glist;
4321	spa_t *newspa;
4322	uint_t c, children, lastlog;
4323	nvlist_t **child, *nvl, *tmp;
4324	dmu_tx_t *tx;
4325	char *altroot = NULL;
4326	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4327	boolean_t activate_slog;
4328
4329	ASSERT(spa_writeable(spa));
4330
4331	txg = spa_vdev_enter(spa);
4332
4333	/* clear the log and flush everything up to now */
4334	activate_slog = spa_passivate_log(spa);
4335	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4336	error = spa_offline_log(spa);
4337	txg = spa_vdev_config_enter(spa);
4338
4339	if (activate_slog)
4340		spa_activate_log(spa);
4341
4342	if (error != 0)
4343		return (spa_vdev_exit(spa, NULL, txg, error));
4344
4345	/* check new spa name before going any further */
4346	if (spa_lookup(newname) != NULL)
4347		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4348
4349	/*
4350	 * scan through all the children to ensure they're all mirrors
4351	 */
4352	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4353	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4354	    &children) != 0)
4355		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4356
4357	/* first, check to ensure we've got the right child count */
4358	rvd = spa->spa_root_vdev;
4359	lastlog = 0;
4360	for (c = 0; c < rvd->vdev_children; c++) {
4361		vdev_t *vd = rvd->vdev_child[c];
4362
4363		/* don't count the holes & logs as children */
4364		if (vd->vdev_islog || vd->vdev_ishole) {
4365			if (lastlog == 0)
4366				lastlog = c;
4367			continue;
4368		}
4369
4370		lastlog = 0;
4371	}
4372	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4373		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4374
4375	/* next, ensure no spare or cache devices are part of the split */
4376	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4377	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4378		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4379
4380	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4381	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4382
4383	/* then, loop over each vdev and validate it */
4384	for (c = 0; c < children; c++) {
4385		uint64_t is_hole = 0;
4386
4387		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4388		    &is_hole);
4389
4390		if (is_hole != 0) {
4391			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4392			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4393				continue;
4394			} else {
4395				error = EINVAL;
4396				break;
4397			}
4398		}
4399
4400		/* which disk is going to be split? */
4401		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4402		    &glist[c]) != 0) {
4403			error = EINVAL;
4404			break;
4405		}
4406
4407		/* look it up in the spa */
4408		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4409		if (vml[c] == NULL) {
4410			error = ENODEV;
4411			break;
4412		}
4413
4414		/* make sure there's nothing stopping the split */
4415		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4416		    vml[c]->vdev_islog ||
4417		    vml[c]->vdev_ishole ||
4418		    vml[c]->vdev_isspare ||
4419		    vml[c]->vdev_isl2cache ||
4420		    !vdev_writeable(vml[c]) ||
4421		    vml[c]->vdev_children != 0 ||
4422		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4423		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4424			error = EINVAL;
4425			break;
4426		}
4427
4428		if (vdev_dtl_required(vml[c])) {
4429			error = EBUSY;
4430			break;
4431		}
4432
4433		/* we need certain info from the top level */
4434		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4435		    vml[c]->vdev_top->vdev_ms_array) == 0);
4436		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4437		    vml[c]->vdev_top->vdev_ms_shift) == 0);
4438		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4439		    vml[c]->vdev_top->vdev_asize) == 0);
4440		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4441		    vml[c]->vdev_top->vdev_ashift) == 0);
4442	}
4443
4444	if (error != 0) {
4445		kmem_free(vml, children * sizeof (vdev_t *));
4446		kmem_free(glist, children * sizeof (uint64_t));
4447		return (spa_vdev_exit(spa, NULL, txg, error));
4448	}
4449
4450	/* stop writers from using the disks */
4451	for (c = 0; c < children; c++) {
4452		if (vml[c] != NULL)
4453			vml[c]->vdev_offline = B_TRUE;
4454	}
4455	vdev_reopen(spa->spa_root_vdev);
4456
4457	/*
4458	 * Temporarily record the splitting vdevs in the spa config.  This
4459	 * will disappear once the config is regenerated.
4460	 */
4461	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4462	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4463	    glist, children) == 0);
4464	kmem_free(glist, children * sizeof (uint64_t));
4465
4466	mutex_enter(&spa->spa_props_lock);
4467	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4468	    nvl) == 0);
4469	mutex_exit(&spa->spa_props_lock);
4470	spa->spa_config_splitting = nvl;
4471	vdev_config_dirty(spa->spa_root_vdev);
4472
4473	/* configure and create the new pool */
4474	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4475	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4476	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4477	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4478	    spa_version(spa)) == 0);
4479	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4480	    spa->spa_config_txg) == 0);
4481	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4482	    spa_generate_guid(NULL)) == 0);
4483	(void) nvlist_lookup_string(props,
4484	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4485
4486	/* add the new pool to the namespace */
4487	newspa = spa_add(newname, config, altroot);
4488	newspa->spa_config_txg = spa->spa_config_txg;
4489	spa_set_log_state(newspa, SPA_LOG_CLEAR);
4490
4491	/* release the spa config lock, retaining the namespace lock */
4492	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4493
4494	if (zio_injection_enabled)
4495		zio_handle_panic_injection(spa, FTAG, 1);
4496
4497	spa_activate(newspa, spa_mode_global);
4498	spa_async_suspend(newspa);
4499
4500	/* create the new pool from the disks of the original pool */
4501	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4502	if (error)
4503		goto out;
4504
4505	/* if that worked, generate a real config for the new pool */
4506	if (newspa->spa_root_vdev != NULL) {
4507		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4508		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4509		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4510		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4511		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4512		    B_TRUE));
4513	}
4514
4515	/* set the props */
4516	if (props != NULL) {
4517		spa_configfile_set(newspa, props, B_FALSE);
4518		error = spa_prop_set(newspa, props);
4519		if (error)
4520			goto out;
4521	}
4522
4523	/* flush everything */
4524	txg = spa_vdev_config_enter(newspa);
4525	vdev_config_dirty(newspa->spa_root_vdev);
4526	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4527
4528	if (zio_injection_enabled)
4529		zio_handle_panic_injection(spa, FTAG, 2);
4530
4531	spa_async_resume(newspa);
4532
4533	/* finally, update the original pool's config */
4534	txg = spa_vdev_config_enter(spa);
4535	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4536	error = dmu_tx_assign(tx, TXG_WAIT);
4537	if (error != 0)
4538		dmu_tx_abort(tx);
4539	for (c = 0; c < children; c++) {
4540		if (vml[c] != NULL) {
4541			vdev_split(vml[c]);
4542			if (error == 0)
4543				spa_history_log_internal(LOG_POOL_VDEV_DETACH,
4544				    spa, tx, "vdev=%s",
4545				    vml[c]->vdev_path);
4546			vdev_free(vml[c]);
4547		}
4548	}
4549	vdev_config_dirty(spa->spa_root_vdev);
4550	spa->spa_config_splitting = NULL;
4551	nvlist_free(nvl);
4552	if (error == 0)
4553		dmu_tx_commit(tx);
4554	(void) spa_vdev_exit(spa, NULL, txg, 0);
4555
4556	if (zio_injection_enabled)
4557		zio_handle_panic_injection(spa, FTAG, 3);
4558
4559	/* split is complete; log a history record */
4560	spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
4561	    "split new pool %s from pool %s", newname, spa_name(spa));
4562
4563	kmem_free(vml, children * sizeof (vdev_t *));
4564
4565	/* if we're not going to mount the filesystems in userland, export */
4566	if (exp)
4567		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4568		    B_FALSE, B_FALSE);
4569
4570	return (error);
4571
4572out:
4573	spa_unload(newspa);
4574	spa_deactivate(newspa);
4575	spa_remove(newspa);
4576
4577	txg = spa_vdev_config_enter(spa);
4578
4579	/* re-online all offlined disks */
4580	for (c = 0; c < children; c++) {
4581		if (vml[c] != NULL)
4582			vml[c]->vdev_offline = B_FALSE;
4583	}
4584	vdev_reopen(spa->spa_root_vdev);
4585
4586	nvlist_free(spa->spa_config_splitting);
4587	spa->spa_config_splitting = NULL;
4588	(void) spa_vdev_exit(spa, NULL, txg, error);
4589
4590	kmem_free(vml, children * sizeof (vdev_t *));
4591	return (error);
4592}
4593
4594static nvlist_t *
4595spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4596{
4597	for (int i = 0; i < count; i++) {
4598		uint64_t guid;
4599
4600		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4601		    &guid) == 0);
4602
4603		if (guid == target_guid)
4604			return (nvpp[i]);
4605	}
4606
4607	return (NULL);
4608}
4609
4610static void
4611spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4612	nvlist_t *dev_to_remove)
4613{
4614	nvlist_t **newdev = NULL;
4615
4616	if (count > 1)
4617		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4618
4619	for (int i = 0, j = 0; i < count; i++) {
4620		if (dev[i] == dev_to_remove)
4621			continue;
4622		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4623	}
4624
4625	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4626	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4627
4628	for (int i = 0; i < count - 1; i++)
4629		nvlist_free(newdev[i]);
4630
4631	if (count > 1)
4632		kmem_free(newdev, (count - 1) * sizeof (void *));
4633}
4634
4635/*
4636 * Evacuate the device.
4637 */
4638static int
4639spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4640{
4641	uint64_t txg;
4642	int error = 0;
4643
4644	ASSERT(MUTEX_HELD(&spa_namespace_lock));
4645	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4646	ASSERT(vd == vd->vdev_top);
4647
4648	/*
4649	 * Evacuate the device.  We don't hold the config lock as writer
4650	 * since we need to do I/O but we do keep the
4651	 * spa_namespace_lock held.  Once this completes the device
4652	 * should no longer have any blocks allocated on it.
4653	 */
4654	if (vd->vdev_islog) {
4655		if (vd->vdev_stat.vs_alloc != 0)
4656			error = spa_offline_log(spa);
4657	} else {
4658		error = ENOTSUP;
4659	}
4660
4661	if (error)
4662		return (error);
4663
4664	/*
4665	 * The evacuation succeeded.  Remove any remaining MOS metadata
4666	 * associated with this vdev, and wait for these changes to sync.
4667	 */
4668	ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
4669	txg = spa_vdev_config_enter(spa);
4670	vd->vdev_removing = B_TRUE;
4671	vdev_dirty(vd, 0, NULL, txg);
4672	vdev_config_dirty(vd);
4673	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4674
4675	return (0);
4676}
4677
4678/*
4679 * Complete the removal by cleaning up the namespace.
4680 */
4681static void
4682spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
4683{
4684	vdev_t *rvd = spa->spa_root_vdev;
4685	uint64_t id = vd->vdev_id;
4686	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
4687
4688	ASSERT(MUTEX_HELD(&spa_namespace_lock));
4689	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4690	ASSERT(vd == vd->vdev_top);
4691
4692	/*
4693	 * Only remove any devices which are empty.
4694	 */
4695	if (vd->vdev_stat.vs_alloc != 0)
4696		return;
4697
4698	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4699
4700	if (list_link_active(&vd->vdev_state_dirty_node))
4701		vdev_state_clean(vd);
4702	if (list_link_active(&vd->vdev_config_dirty_node))
4703		vdev_config_clean(vd);
4704
4705	vdev_free(vd);
4706
4707	if (last_vdev) {
4708		vdev_compact_children(rvd);
4709	} else {
4710		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
4711		vdev_add_child(rvd, vd);
4712	}
4713	vdev_config_dirty(rvd);
4714
4715	/*
4716	 * Reassess the health of our root vdev.
4717	 */
4718	vdev_reopen(rvd);
4719}
4720
4721/*
4722 * Remove a device from the pool -
4723 *
4724 * Removing a device from the vdev namespace requires several steps
4725 * and can take a significant amount of time.  As a result we use
4726 * the spa_vdev_config_[enter/exit] functions which allow us to
4727 * grab and release the spa_config_lock while still holding the namespace
4728 * lock.  During each step the configuration is synced out.
4729 */
4730
4731/*
4732 * Remove a device from the pool.  Currently, this supports removing only hot
4733 * spares, slogs, and level 2 ARC devices.
4734 */
4735int
4736spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
4737{
4738	vdev_t *vd;
4739	metaslab_group_t *mg;
4740	nvlist_t **spares, **l2cache, *nv;
4741	uint64_t txg = 0;
4742	uint_t nspares, nl2cache;
4743	int error = 0;
4744	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
4745
4746	ASSERT(spa_writeable(spa));
4747
4748	if (!locked)
4749		txg = spa_vdev_enter(spa);
4750
4751	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4752
4753	if (spa->spa_spares.sav_vdevs != NULL &&
4754	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4755	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
4756	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
4757		/*
4758		 * Only remove the hot spare if it's not currently in use
4759		 * in this pool.
4760		 */
4761		if (vd == NULL || unspare) {
4762			spa_vdev_remove_aux(spa->spa_spares.sav_config,
4763			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
4764			spa_load_spares(spa);
4765			spa->spa_spares.sav_sync = B_TRUE;
4766		} else {
4767			error = EBUSY;
4768		}
4769	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
4770	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4771	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
4772	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
4773		/*
4774		 * Cache devices can always be removed.
4775		 */
4776		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
4777		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
4778		spa_load_l2cache(spa);
4779		spa->spa_l2cache.sav_sync = B_TRUE;
4780	} else if (vd != NULL && vd->vdev_islog) {
4781		ASSERT(!locked);
4782		ASSERT(vd == vd->vdev_top);
4783
4784		/*
4785		 * XXX - Once we have bp-rewrite this should
4786		 * become the common case.
4787		 */
4788
4789		mg = vd->vdev_mg;
4790
4791		/*
4792		 * Stop allocating from this vdev.
4793		 */
4794		metaslab_group_passivate(mg);
4795
4796		/*
4797		 * Wait for the youngest allocations and frees to sync,
4798		 * and then wait for the deferral of those frees to finish.
4799		 */
4800		spa_vdev_config_exit(spa, NULL,
4801		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
4802
4803		/*
4804		 * Attempt to evacuate the vdev.
4805		 */
4806		error = spa_vdev_remove_evacuate(spa, vd);
4807
4808		txg = spa_vdev_config_enter(spa);
4809
4810		/*
4811		 * If we couldn't evacuate the vdev, unwind.
4812		 */
4813		if (error) {
4814			metaslab_group_activate(mg);
4815			return (spa_vdev_exit(spa, NULL, txg, error));
4816		}
4817
4818		/*
4819		 * Clean up the vdev namespace.
4820		 */
4821		spa_vdev_remove_from_namespace(spa, vd);
4822
4823	} else if (vd != NULL) {
4824		/*
4825		 * Normal vdevs cannot be removed (yet).
4826		 */
4827		error = ENOTSUP;
4828	} else {
4829		/*
4830		 * There is no vdev of any kind with the specified guid.
4831		 */
4832		error = ENOENT;
4833	}
4834
4835	if (!locked)
4836		return (spa_vdev_exit(spa, NULL, txg, error));
4837
4838	return (error);
4839}
4840
4841/*
4842 * Find any device that's done replacing, or a vdev marked 'unspare' that's
4843 * current spared, so we can detach it.
4844 */
4845static vdev_t *
4846spa_vdev_resilver_done_hunt(vdev_t *vd)
4847{
4848	vdev_t *newvd, *oldvd;
4849
4850	for (int c = 0; c < vd->vdev_children; c++) {
4851		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
4852		if (oldvd != NULL)
4853			return (oldvd);
4854	}
4855
4856	/*
4857	 * Check for a completed replacement.  We always consider the first
4858	 * vdev in the list to be the oldest vdev, and the last one to be
4859	 * the newest (see spa_vdev_attach() for how that works).  In
4860	 * the case where the newest vdev is faulted, we will not automatically
4861	 * remove it after a resilver completes.  This is OK as it will require
4862	 * user intervention to determine which disk the admin wishes to keep.
4863	 */
4864	if (vd->vdev_ops == &vdev_replacing_ops) {
4865		ASSERT(vd->vdev_children > 1);
4866
4867		newvd = vd->vdev_child[vd->vdev_children - 1];
4868		oldvd = vd->vdev_child[0];
4869
4870		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4871		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4872		    !vdev_dtl_required(oldvd))
4873			return (oldvd);
4874	}
4875
4876	/*
4877	 * Check for a completed resilver with the 'unspare' flag set.
4878	 */
4879	if (vd->vdev_ops == &vdev_spare_ops) {
4880		vdev_t *first = vd->vdev_child[0];
4881		vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
4882
4883		if (last->vdev_unspare) {
4884			oldvd = first;
4885			newvd = last;
4886		} else if (first->vdev_unspare) {
4887			oldvd = last;
4888			newvd = first;
4889		} else {
4890			oldvd = NULL;
4891		}
4892
4893		if (oldvd != NULL &&
4894		    vdev_dtl_empty(newvd, DTL_MISSING) &&
4895		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4896		    !vdev_dtl_required(oldvd))
4897			return (oldvd);
4898
4899		/*
4900		 * If there are more than two spares attached to a disk,
4901		 * and those spares are not required, then we want to
4902		 * attempt to free them up now so that they can be used
4903		 * by other pools.  Once we're back down to a single
4904		 * disk+spare, we stop removing them.
4905		 */
4906		if (vd->vdev_children > 2) {
4907			newvd = vd->vdev_child[1];
4908
4909			if (newvd->vdev_isspare && last->vdev_isspare &&
4910			    vdev_dtl_empty(last, DTL_MISSING) &&
4911			    vdev_dtl_empty(last, DTL_OUTAGE) &&
4912			    !vdev_dtl_required(newvd))
4913				return (newvd);
4914		}
4915	}
4916
4917	return (NULL);
4918}
4919
4920static void
4921spa_vdev_resilver_done(spa_t *spa)
4922{
4923	vdev_t *vd, *pvd, *ppvd;
4924	uint64_t guid, sguid, pguid, ppguid;
4925
4926	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4927
4928	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4929		pvd = vd->vdev_parent;
4930		ppvd = pvd->vdev_parent;
4931		guid = vd->vdev_guid;
4932		pguid = pvd->vdev_guid;
4933		ppguid = ppvd->vdev_guid;
4934		sguid = 0;
4935		/*
4936		 * If we have just finished replacing a hot spared device, then
4937		 * we need to detach the parent's first child (the original hot
4938		 * spare) as well.
4939		 */
4940		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
4941		    ppvd->vdev_children == 2) {
4942			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4943			sguid = ppvd->vdev_child[1]->vdev_guid;
4944		}
4945		spa_config_exit(spa, SCL_ALL, FTAG);
4946		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4947			return;
4948		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4949			return;
4950		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4951	}
4952
4953	spa_config_exit(spa, SCL_ALL, FTAG);
4954}
4955
4956/*
4957 * Update the stored path or FRU for this vdev.
4958 */
4959int
4960spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4961    boolean_t ispath)
4962{
4963	vdev_t *vd;
4964	boolean_t sync = B_FALSE;
4965
4966	ASSERT(spa_writeable(spa));
4967
4968	spa_vdev_state_enter(spa, SCL_ALL);
4969
4970	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4971		return (spa_vdev_state_exit(spa, NULL, ENOENT));
4972
4973	if (!vd->vdev_ops->vdev_op_leaf)
4974		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
4975
4976	if (ispath) {
4977		if (strcmp(value, vd->vdev_path) != 0) {
4978			spa_strfree(vd->vdev_path);
4979			vd->vdev_path = spa_strdup(value);
4980			sync = B_TRUE;
4981		}
4982	} else {
4983		if (vd->vdev_fru == NULL) {
4984			vd->vdev_fru = spa_strdup(value);
4985			sync = B_TRUE;
4986		} else if (strcmp(value, vd->vdev_fru) != 0) {
4987			spa_strfree(vd->vdev_fru);
4988			vd->vdev_fru = spa_strdup(value);
4989			sync = B_TRUE;
4990		}
4991	}
4992
4993	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
4994}
4995
4996int
4997spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
4998{
4999	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5000}
5001
5002int
5003spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5004{
5005	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5006}
5007
5008/*
5009 * ==========================================================================
5010 * SPA Scanning
5011 * ==========================================================================
5012 */
5013
5014int
5015spa_scan_stop(spa_t *spa)
5016{
5017	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5018	if (dsl_scan_resilvering(spa->spa_dsl_pool))
5019		return (EBUSY);
5020	return (dsl_scan_cancel(spa->spa_dsl_pool));
5021}
5022
5023int
5024spa_scan(spa_t *spa, pool_scan_func_t func)
5025{
5026	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5027
5028	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5029		return (ENOTSUP);
5030
5031	/*
5032	 * If a resilver was requested, but there is no DTL on a
5033	 * writeable leaf device, we have nothing to do.
5034	 */
5035	if (func == POOL_SCAN_RESILVER &&
5036	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5037		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5038		return (0);
5039	}
5040
5041	return (dsl_scan(spa->spa_dsl_pool, func));
5042}
5043
5044/*
5045 * ==========================================================================
5046 * SPA async task processing
5047 * ==========================================================================
5048 */
5049
5050static void
5051spa_async_remove(spa_t *spa, vdev_t *vd)
5052{
5053	if (vd->vdev_remove_wanted) {
5054		vd->vdev_remove_wanted = B_FALSE;
5055		vd->vdev_delayed_close = B_FALSE;
5056		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5057
5058		/*
5059		 * We want to clear the stats, but we don't want to do a full
5060		 * vdev_clear() as that will cause us to throw away
5061		 * degraded/faulted state as well as attempt to reopen the
5062		 * device, all of which is a waste.
5063		 */
5064		vd->vdev_stat.vs_read_errors = 0;
5065		vd->vdev_stat.vs_write_errors = 0;
5066		vd->vdev_stat.vs_checksum_errors = 0;
5067
5068		vdev_state_dirty(vd->vdev_top);
5069	}
5070
5071	for (int c = 0; c < vd->vdev_children; c++)
5072		spa_async_remove(spa, vd->vdev_child[c]);
5073}
5074
5075static void
5076spa_async_probe(spa_t *spa, vdev_t *vd)
5077{
5078	if (vd->vdev_probe_wanted) {
5079		vd->vdev_probe_wanted = B_FALSE;
5080		vdev_reopen(vd);	/* vdev_open() does the actual probe */
5081	}
5082
5083	for (int c = 0; c < vd->vdev_children; c++)
5084		spa_async_probe(spa, vd->vdev_child[c]);
5085}
5086
5087static void
5088spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5089{
5090	sysevent_id_t eid;
5091	nvlist_t *attr;
5092	char *physpath;
5093
5094	if (!spa->spa_autoexpand)
5095		return;
5096
5097	for (int c = 0; c < vd->vdev_children; c++) {
5098		vdev_t *cvd = vd->vdev_child[c];
5099		spa_async_autoexpand(spa, cvd);
5100	}
5101
5102	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5103		return;
5104
5105	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5106	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5107
5108	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5109	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5110
5111	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5112	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5113
5114	nvlist_free(attr);
5115	kmem_free(physpath, MAXPATHLEN);
5116}
5117
5118static void
5119spa_async_thread(spa_t *spa)
5120{
5121	int tasks;
5122
5123	ASSERT(spa->spa_sync_on);
5124
5125	mutex_enter(&spa->spa_async_lock);
5126	tasks = spa->spa_async_tasks;
5127	spa->spa_async_tasks = 0;
5128	mutex_exit(&spa->spa_async_lock);
5129
5130	/*
5131	 * See if the config needs to be updated.
5132	 */
5133	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5134		uint64_t old_space, new_space;
5135
5136		mutex_enter(&spa_namespace_lock);
5137		old_space = metaslab_class_get_space(spa_normal_class(spa));
5138		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5139		new_space = metaslab_class_get_space(spa_normal_class(spa));
5140		mutex_exit(&spa_namespace_lock);
5141
5142		/*
5143		 * If the pool grew as a result of the config update,
5144		 * then log an internal history event.
5145		 */
5146		if (new_space != old_space) {
5147			spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5148			    spa, NULL,
5149			    "pool '%s' size: %llu(+%llu)",
5150			    spa_name(spa), new_space, new_space - old_space);
5151		}
5152	}
5153
5154	/*
5155	 * See if any devices need to be marked REMOVED.
5156	 */
5157	if (tasks & SPA_ASYNC_REMOVE) {
5158		spa_vdev_state_enter(spa, SCL_NONE);
5159		spa_async_remove(spa, spa->spa_root_vdev);
5160		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5161			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5162		for (int i = 0; i < spa->spa_spares.sav_count; i++)
5163			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5164		(void) spa_vdev_state_exit(spa, NULL, 0);
5165	}
5166
5167	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5168		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5169		spa_async_autoexpand(spa, spa->spa_root_vdev);
5170		spa_config_exit(spa, SCL_CONFIG, FTAG);
5171	}
5172
5173	/*
5174	 * See if any devices need to be probed.
5175	 */
5176	if (tasks & SPA_ASYNC_PROBE) {
5177		spa_vdev_state_enter(spa, SCL_NONE);
5178		spa_async_probe(spa, spa->spa_root_vdev);
5179		(void) spa_vdev_state_exit(spa, NULL, 0);
5180	}
5181
5182	/*
5183	 * If any devices are done replacing, detach them.
5184	 */
5185	if (tasks & SPA_ASYNC_RESILVER_DONE)
5186		spa_vdev_resilver_done(spa);
5187
5188	/*
5189	 * Kick off a resilver.
5190	 */
5191	if (tasks & SPA_ASYNC_RESILVER)
5192		dsl_resilver_restart(spa->spa_dsl_pool, 0);
5193
5194	/*
5195	 * Let the world know that we're done.
5196	 */
5197	mutex_enter(&spa->spa_async_lock);
5198	spa->spa_async_thread = NULL;
5199	cv_broadcast(&spa->spa_async_cv);
5200	mutex_exit(&spa->spa_async_lock);
5201	thread_exit();
5202}
5203
5204void
5205spa_async_suspend(spa_t *spa)
5206{
5207	mutex_enter(&spa->spa_async_lock);
5208	spa->spa_async_suspended++;
5209	while (spa->spa_async_thread != NULL)
5210		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5211	mutex_exit(&spa->spa_async_lock);
5212}
5213
5214void
5215spa_async_resume(spa_t *spa)
5216{
5217	mutex_enter(&spa->spa_async_lock);
5218	ASSERT(spa->spa_async_suspended != 0);
5219	spa->spa_async_suspended--;
5220	mutex_exit(&spa->spa_async_lock);
5221}
5222
5223static void
5224spa_async_dispatch(spa_t *spa)
5225{
5226	mutex_enter(&spa->spa_async_lock);
5227	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5228	    spa->spa_async_thread == NULL &&
5229	    rootdir != NULL && !vn_is_readonly(rootdir))
5230		spa->spa_async_thread = thread_create(NULL, 0,
5231		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5232	mutex_exit(&spa->spa_async_lock);
5233}
5234
5235void
5236spa_async_request(spa_t *spa, int task)
5237{
5238	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5239	mutex_enter(&spa->spa_async_lock);
5240	spa->spa_async_tasks |= task;
5241	mutex_exit(&spa->spa_async_lock);
5242}
5243
5244/*
5245 * ==========================================================================
5246 * SPA syncing routines
5247 * ==========================================================================
5248 */
5249
5250static int
5251bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5252{
5253	bpobj_t *bpo = arg;
5254	bpobj_enqueue(bpo, bp, tx);
5255	return (0);
5256}
5257
5258static int
5259spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5260{
5261	zio_t *zio = arg;
5262
5263	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5264	    zio->io_flags));
5265	return (0);
5266}
5267
5268static void
5269spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5270{
5271	char *packed = NULL;
5272	size_t bufsize;
5273	size_t nvsize = 0;
5274	dmu_buf_t *db;
5275
5276	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5277
5278	/*
5279	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5280	 * information.  This avoids the dbuf_will_dirty() path and
5281	 * saves us a pre-read to get data we don't actually care about.
5282	 */
5283	bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
5284	packed = kmem_alloc(bufsize, KM_SLEEP);
5285
5286	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5287	    KM_SLEEP) == 0);
5288	bzero(packed + nvsize, bufsize - nvsize);
5289
5290	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5291
5292	kmem_free(packed, bufsize);
5293
5294	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5295	dmu_buf_will_dirty(db, tx);
5296	*(uint64_t *)db->db_data = nvsize;
5297	dmu_buf_rele(db, FTAG);
5298}
5299
5300static void
5301spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5302    const char *config, const char *entry)
5303{
5304	nvlist_t *nvroot;
5305	nvlist_t **list;
5306	int i;
5307
5308	if (!sav->sav_sync)
5309		return;
5310
5311	/*
5312	 * Update the MOS nvlist describing the list of available devices.
5313	 * spa_validate_aux() will have already made sure this nvlist is
5314	 * valid and the vdevs are labeled appropriately.
5315	 */
5316	if (sav->sav_object == 0) {
5317		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5318		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5319		    sizeof (uint64_t), tx);
5320		VERIFY(zap_update(spa->spa_meta_objset,
5321		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5322		    &sav->sav_object, tx) == 0);
5323	}
5324
5325	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5326	if (sav->sav_count == 0) {
5327		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5328	} else {
5329		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5330		for (i = 0; i < sav->sav_count; i++)
5331			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5332			    B_FALSE, VDEV_CONFIG_L2CACHE);
5333		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5334		    sav->sav_count) == 0);
5335		for (i = 0; i < sav->sav_count; i++)
5336			nvlist_free(list[i]);
5337		kmem_free(list, sav->sav_count * sizeof (void *));
5338	}
5339
5340	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5341	nvlist_free(nvroot);
5342
5343	sav->sav_sync = B_FALSE;
5344}
5345
5346static void
5347spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5348{
5349	nvlist_t *config;
5350
5351	if (list_is_empty(&spa->spa_config_dirty_list))
5352		return;
5353
5354	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5355
5356	config = spa_config_generate(spa, spa->spa_root_vdev,
5357	    dmu_tx_get_txg(tx), B_FALSE);
5358
5359	spa_config_exit(spa, SCL_STATE, FTAG);
5360
5361	if (spa->spa_config_syncing)
5362		nvlist_free(spa->spa_config_syncing);
5363	spa->spa_config_syncing = config;
5364
5365	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5366}
5367
5368/*
5369 * Set zpool properties.
5370 */
5371static void
5372spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5373{
5374	spa_t *spa = arg1;
5375	objset_t *mos = spa->spa_meta_objset;
5376	nvlist_t *nvp = arg2;
5377	nvpair_t *elem;
5378	uint64_t intval;
5379	char *strval;
5380	zpool_prop_t prop;
5381	const char *propname;
5382	zprop_type_t proptype;
5383
5384	mutex_enter(&spa->spa_props_lock);
5385
5386	elem = NULL;
5387	while ((elem = nvlist_next_nvpair(nvp, elem))) {
5388		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5389		case ZPOOL_PROP_VERSION:
5390			/*
5391			 * Only set version for non-zpool-creation cases
5392			 * (set/import). spa_create() needs special care
5393			 * for version setting.
5394			 */
5395			if (tx->tx_txg != TXG_INITIAL) {
5396				VERIFY(nvpair_value_uint64(elem,
5397				    &intval) == 0);
5398				ASSERT(intval <= SPA_VERSION);
5399				ASSERT(intval >= spa_version(spa));
5400				spa->spa_uberblock.ub_version = intval;
5401				vdev_config_dirty(spa->spa_root_vdev);
5402			}
5403			break;
5404
5405		case ZPOOL_PROP_ALTROOT:
5406			/*
5407			 * 'altroot' is a non-persistent property. It should
5408			 * have been set temporarily at creation or import time.
5409			 */
5410			ASSERT(spa->spa_root != NULL);
5411			break;
5412
5413		case ZPOOL_PROP_READONLY:
5414		case ZPOOL_PROP_CACHEFILE:
5415			/*
5416			 * 'readonly' and 'cachefile' are also non-persisitent
5417			 * properties.
5418			 */
5419			break;
5420		case ZPOOL_PROP_COMMENT:
5421			VERIFY(nvpair_value_string(elem, &strval) == 0);
5422			if (spa->spa_comment != NULL)
5423				spa_strfree(spa->spa_comment);
5424			spa->spa_comment = spa_strdup(strval);
5425			/*
5426			 * We need to dirty the configuration on all the vdevs
5427			 * so that their labels get updated.  It's unnecessary
5428			 * to do this for pool creation since the vdev's
5429			 * configuratoin has already been dirtied.
5430			 */
5431			if (tx->tx_txg != TXG_INITIAL)
5432				vdev_config_dirty(spa->spa_root_vdev);
5433			break;
5434		default:
5435			/*
5436			 * Set pool property values in the poolprops mos object.
5437			 */
5438			if (spa->spa_pool_props_object == 0) {
5439				VERIFY((spa->spa_pool_props_object =
5440				    zap_create(mos, DMU_OT_POOL_PROPS,
5441				    DMU_OT_NONE, 0, tx)) > 0);
5442
5443				VERIFY(zap_update(mos,
5444				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5445				    8, 1, &spa->spa_pool_props_object, tx)
5446				    == 0);
5447			}
5448
5449			/* normalize the property name */
5450			propname = zpool_prop_to_name(prop);
5451			proptype = zpool_prop_get_type(prop);
5452
5453			if (nvpair_type(elem) == DATA_TYPE_STRING) {
5454				ASSERT(proptype == PROP_TYPE_STRING);
5455				VERIFY(nvpair_value_string(elem, &strval) == 0);
5456				VERIFY(zap_update(mos,
5457				    spa->spa_pool_props_object, propname,
5458				    1, strlen(strval) + 1, strval, tx) == 0);
5459
5460			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5461				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5462
5463				if (proptype == PROP_TYPE_INDEX) {
5464					const char *unused;
5465					VERIFY(zpool_prop_index_to_string(
5466					    prop, intval, &unused) == 0);
5467				}
5468				VERIFY(zap_update(mos,
5469				    spa->spa_pool_props_object, propname,
5470				    8, 1, &intval, tx) == 0);
5471			} else {
5472				ASSERT(0); /* not allowed */
5473			}
5474
5475			switch (prop) {
5476			case ZPOOL_PROP_DELEGATION:
5477				spa->spa_delegation = intval;
5478				break;
5479			case ZPOOL_PROP_BOOTFS:
5480				spa->spa_bootfs = intval;
5481				break;
5482			case ZPOOL_PROP_FAILUREMODE:
5483				spa->spa_failmode = intval;
5484				break;
5485			case ZPOOL_PROP_AUTOEXPAND:
5486				spa->spa_autoexpand = intval;
5487				if (tx->tx_txg != TXG_INITIAL)
5488					spa_async_request(spa,
5489					    SPA_ASYNC_AUTOEXPAND);
5490				break;
5491			case ZPOOL_PROP_DEDUPDITTO:
5492				spa->spa_dedup_ditto = intval;
5493				break;
5494			default:
5495				break;
5496			}
5497		}
5498
5499		/* log internal history if this is not a zpool create */
5500		if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5501		    tx->tx_txg != TXG_INITIAL) {
5502			spa_history_log_internal(LOG_POOL_PROPSET,
5503			    spa, tx, "%s %lld %s",
5504			    nvpair_name(elem), intval, spa_name(spa));
5505		}
5506	}
5507
5508	mutex_exit(&spa->spa_props_lock);
5509}
5510
5511/*
5512 * Perform one-time upgrade on-disk changes.  spa_version() does not
5513 * reflect the new version this txg, so there must be no changes this
5514 * txg to anything that the upgrade code depends on after it executes.
5515 * Therefore this must be called after dsl_pool_sync() does the sync
5516 * tasks.
5517 */
5518static void
5519spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5520{
5521	dsl_pool_t *dp = spa->spa_dsl_pool;
5522
5523	ASSERT(spa->spa_sync_pass == 1);
5524
5525	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5526	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5527		dsl_pool_create_origin(dp, tx);
5528
5529		/* Keeping the origin open increases spa_minref */
5530		spa->spa_minref += 3;
5531	}
5532
5533	if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5534	    spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5535		dsl_pool_upgrade_clones(dp, tx);
5536	}
5537
5538	if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5539	    spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5540		dsl_pool_upgrade_dir_clones(dp, tx);
5541
5542		/* Keeping the freedir open increases spa_minref */
5543		spa->spa_minref += 3;
5544	}
5545}
5546
5547/*
5548 * Sync the specified transaction group.  New blocks may be dirtied as
5549 * part of the process, so we iterate until it converges.
5550 */
5551void
5552spa_sync(spa_t *spa, uint64_t txg)
5553{
5554	dsl_pool_t *dp = spa->spa_dsl_pool;
5555	objset_t *mos = spa->spa_meta_objset;
5556	bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5557	bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5558	vdev_t *rvd = spa->spa_root_vdev;
5559	vdev_t *vd;
5560	dmu_tx_t *tx;
5561	int error;
5562
5563	VERIFY(spa_writeable(spa));
5564
5565	/*
5566	 * Lock out configuration changes.
5567	 */
5568	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5569
5570	spa->spa_syncing_txg = txg;
5571	spa->spa_sync_pass = 0;
5572
5573	/*
5574	 * If there are any pending vdev state changes, convert them
5575	 * into config changes that go out with this transaction group.
5576	 */
5577	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5578	while (list_head(&spa->spa_state_dirty_list) != NULL) {
5579		/*
5580		 * We need the write lock here because, for aux vdevs,
5581		 * calling vdev_config_dirty() modifies sav_config.
5582		 * This is ugly and will become unnecessary when we
5583		 * eliminate the aux vdev wart by integrating all vdevs
5584		 * into the root vdev tree.
5585		 */
5586		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5587		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5588		while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5589			vdev_state_clean(vd);
5590			vdev_config_dirty(vd);
5591		}
5592		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5593		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5594	}
5595	spa_config_exit(spa, SCL_STATE, FTAG);
5596
5597	tx = dmu_tx_create_assigned(dp, txg);
5598
5599	/*
5600	 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5601	 * set spa_deflate if we have no raid-z vdevs.
5602	 */
5603	if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5604	    spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5605		int i;
5606
5607		for (i = 0; i < rvd->vdev_children; i++) {
5608			vd = rvd->vdev_child[i];
5609			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5610				break;
5611		}
5612		if (i == rvd->vdev_children) {
5613			spa->spa_deflate = TRUE;
5614			VERIFY(0 == zap_add(spa->spa_meta_objset,
5615			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5616			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5617		}
5618	}
5619
5620	/*
5621	 * If anything has changed in this txg, or if someone is waiting
5622	 * for this txg to sync (eg, spa_vdev_remove()), push the
5623	 * deferred frees from the previous txg.  If not, leave them
5624	 * alone so that we don't generate work on an otherwise idle
5625	 * system.
5626	 */
5627	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5628	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5629	    !txg_list_empty(&dp->dp_sync_tasks, txg) ||
5630	    ((dsl_scan_active(dp->dp_scan) ||
5631	    txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
5632		zio_t *zio = zio_root(spa, NULL, NULL, 0);
5633		VERIFY3U(bpobj_iterate(defer_bpo,
5634		    spa_free_sync_cb, zio, tx), ==, 0);
5635		VERIFY3U(zio_wait(zio), ==, 0);
5636	}
5637
5638	/*
5639	 * Iterate to convergence.
5640	 */
5641	do {
5642		int pass = ++spa->spa_sync_pass;
5643
5644		spa_sync_config_object(spa, tx);
5645		spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5646		    ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5647		spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5648		    ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5649		spa_errlog_sync(spa, txg);
5650		dsl_pool_sync(dp, txg);
5651
5652		if (pass <= SYNC_PASS_DEFERRED_FREE) {
5653			zio_t *zio = zio_root(spa, NULL, NULL, 0);
5654			bplist_iterate(free_bpl, spa_free_sync_cb,
5655			    zio, tx);
5656			VERIFY(zio_wait(zio) == 0);
5657		} else {
5658			bplist_iterate(free_bpl, bpobj_enqueue_cb,
5659			    defer_bpo, tx);
5660		}
5661
5662		ddt_sync(spa, txg);
5663		dsl_scan_sync(dp, tx);
5664
5665		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
5666			vdev_sync(vd, txg);
5667
5668		if (pass == 1)
5669			spa_sync_upgrades(spa, tx);
5670
5671	} while (dmu_objset_is_dirty(mos, txg));
5672
5673	/*
5674	 * Rewrite the vdev configuration (which includes the uberblock)
5675	 * to commit the transaction group.
5676	 *
5677	 * If there are no dirty vdevs, we sync the uberblock to a few
5678	 * random top-level vdevs that are known to be visible in the
5679	 * config cache (see spa_vdev_add() for a complete description).
5680	 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
5681	 */
5682	for (;;) {
5683		/*
5684		 * We hold SCL_STATE to prevent vdev open/close/etc.
5685		 * while we're attempting to write the vdev labels.
5686		 */
5687		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5688
5689		if (list_is_empty(&spa->spa_config_dirty_list)) {
5690			vdev_t *svd[SPA_DVAS_PER_BP];
5691			int svdcount = 0;
5692			int children = rvd->vdev_children;
5693			int c0 = spa_get_random(children);
5694
5695			for (int c = 0; c < children; c++) {
5696				vd = rvd->vdev_child[(c0 + c) % children];
5697				if (vd->vdev_ms_array == 0 || vd->vdev_islog)
5698					continue;
5699				svd[svdcount++] = vd;
5700				if (svdcount == SPA_DVAS_PER_BP)
5701					break;
5702			}
5703			error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
5704			if (error != 0)
5705				error = vdev_config_sync(svd, svdcount, txg,
5706				    B_TRUE);
5707		} else {
5708			error = vdev_config_sync(rvd->vdev_child,
5709			    rvd->vdev_children, txg, B_FALSE);
5710			if (error != 0)
5711				error = vdev_config_sync(rvd->vdev_child,
5712				    rvd->vdev_children, txg, B_TRUE);
5713		}
5714
5715		spa_config_exit(spa, SCL_STATE, FTAG);
5716
5717		if (error == 0)
5718			break;
5719		zio_suspend(spa, NULL);
5720		zio_resume_wait(spa);
5721	}
5722	dmu_tx_commit(tx);
5723
5724	/*
5725	 * Clear the dirty config list.
5726	 */
5727	while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
5728		vdev_config_clean(vd);
5729
5730	/*
5731	 * Now that the new config has synced transactionally,
5732	 * let it become visible to the config cache.
5733	 */
5734	if (spa->spa_config_syncing != NULL) {
5735		spa_config_set(spa, spa->spa_config_syncing);
5736		spa->spa_config_txg = txg;
5737		spa->spa_config_syncing = NULL;
5738	}
5739
5740	spa->spa_ubsync = spa->spa_uberblock;
5741
5742	dsl_pool_sync_done(dp, txg);
5743
5744	/*
5745	 * Update usable space statistics.
5746	 */
5747	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
5748		vdev_sync_done(vd, txg);
5749
5750	spa_update_dspace(spa);
5751
5752	/*
5753	 * It had better be the case that we didn't dirty anything
5754	 * since vdev_config_sync().
5755	 */
5756	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
5757	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
5758	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
5759
5760	spa->spa_sync_pass = 0;
5761
5762	spa_config_exit(spa, SCL_CONFIG, FTAG);
5763
5764	spa_handle_ignored_writes(spa);
5765
5766	/*
5767	 * If any async tasks have been requested, kick them off.
5768	 */
5769	spa_async_dispatch(spa);
5770}
5771
5772/*
5773 * Sync all pools.  We don't want to hold the namespace lock across these
5774 * operations, so we take a reference on the spa_t and drop the lock during the
5775 * sync.
5776 */
5777void
5778spa_sync_allpools(void)
5779{
5780	spa_t *spa = NULL;
5781	mutex_enter(&spa_namespace_lock);
5782	while ((spa = spa_next(spa)) != NULL) {
5783		if (spa_state(spa) != POOL_STATE_ACTIVE ||
5784		    !spa_writeable(spa) || spa_suspended(spa))
5785			continue;
5786		spa_open_ref(spa, FTAG);
5787		mutex_exit(&spa_namespace_lock);
5788		txg_wait_synced(spa_get_dsl(spa), 0);
5789		mutex_enter(&spa_namespace_lock);
5790		spa_close(spa, FTAG);
5791	}
5792	mutex_exit(&spa_namespace_lock);
5793}
5794
5795/*
5796 * ==========================================================================
5797 * Miscellaneous routines
5798 * ==========================================================================
5799 */
5800
5801/*
5802 * Remove all pools in the system.
5803 */
5804void
5805spa_evict_all(void)
5806{
5807	spa_t *spa;
5808
5809	/*
5810	 * Remove all cached state.  All pools should be closed now,
5811	 * so every spa in the AVL tree should be unreferenced.
5812	 */
5813	mutex_enter(&spa_namespace_lock);
5814	while ((spa = spa_next(NULL)) != NULL) {
5815		/*
5816		 * Stop async tasks.  The async thread may need to detach
5817		 * a device that's been replaced, which requires grabbing
5818		 * spa_namespace_lock, so we must drop it here.
5819		 */
5820		spa_open_ref(spa, FTAG);
5821		mutex_exit(&spa_namespace_lock);
5822		spa_async_suspend(spa);
5823		mutex_enter(&spa_namespace_lock);
5824		spa_close(spa, FTAG);
5825
5826		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5827			spa_unload(spa);
5828			spa_deactivate(spa);
5829		}
5830		spa_remove(spa);
5831	}
5832	mutex_exit(&spa_namespace_lock);
5833}
5834
5835vdev_t *
5836spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
5837{
5838	vdev_t *vd;
5839	int i;
5840
5841	if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
5842		return (vd);
5843
5844	if (aux) {
5845		for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
5846			vd = spa->spa_l2cache.sav_vdevs[i];
5847			if (vd->vdev_guid == guid)
5848				return (vd);
5849		}
5850
5851		for (i = 0; i < spa->spa_spares.sav_count; i++) {
5852			vd = spa->spa_spares.sav_vdevs[i];
5853			if (vd->vdev_guid == guid)
5854				return (vd);
5855		}
5856	}
5857
5858	return (NULL);
5859}
5860
5861void
5862spa_upgrade(spa_t *spa, uint64_t version)
5863{
5864	ASSERT(spa_writeable(spa));
5865
5866	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5867
5868	/*
5869	 * This should only be called for a non-faulted pool, and since a
5870	 * future version would result in an unopenable pool, this shouldn't be
5871	 * possible.
5872	 */
5873	ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
5874	ASSERT(version >= spa->spa_uberblock.ub_version);
5875
5876	spa->spa_uberblock.ub_version = version;
5877	vdev_config_dirty(spa->spa_root_vdev);
5878
5879	spa_config_exit(spa, SCL_ALL, FTAG);
5880
5881	txg_wait_synced(spa_get_dsl(spa), 0);
5882}
5883
5884boolean_t
5885spa_has_spare(spa_t *spa, uint64_t guid)
5886{
5887	int i;
5888	uint64_t spareguid;
5889	spa_aux_vdev_t *sav = &spa->spa_spares;
5890
5891	for (i = 0; i < sav->sav_count; i++)
5892		if (sav->sav_vdevs[i]->vdev_guid == guid)
5893			return (B_TRUE);
5894
5895	for (i = 0; i < sav->sav_npending; i++) {
5896		if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
5897		    &spareguid) == 0 && spareguid == guid)
5898			return (B_TRUE);
5899	}
5900
5901	return (B_FALSE);
5902}
5903
5904/*
5905 * Check if a pool has an active shared spare device.
5906 * Note: reference count of an active spare is 2, as a spare and as a replace
5907 */
5908static boolean_t
5909spa_has_active_shared_spare(spa_t *spa)
5910{
5911	int i, refcnt;
5912	uint64_t pool;
5913	spa_aux_vdev_t *sav = &spa->spa_spares;
5914
5915	for (i = 0; i < sav->sav_count; i++) {
5916		if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
5917		    &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
5918		    refcnt > 2)
5919			return (B_TRUE);
5920	}
5921
5922	return (B_FALSE);
5923}
5924
5925/*
5926 * Post a sysevent corresponding to the given event.  The 'name' must be one of
5927 * the event definitions in sys/sysevent/eventdefs.h.  The payload will be
5928 * filled in from the spa and (optionally) the vdev.  This doesn't do anything
5929 * in the userland libzpool, as we don't want consumers to misinterpret ztest
5930 * or zdb as real changes.
5931 */
5932void
5933spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
5934{
5935#ifdef _KERNEL
5936	sysevent_t		*ev;
5937	sysevent_attr_list_t	*attr = NULL;
5938	sysevent_value_t	value;
5939	sysevent_id_t		eid;
5940
5941	ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
5942	    SE_SLEEP);
5943
5944	value.value_type = SE_DATA_TYPE_STRING;
5945	value.value.sv_string = spa_name(spa);
5946	if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
5947		goto done;
5948
5949	value.value_type = SE_DATA_TYPE_UINT64;
5950	value.value.sv_uint64 = spa_guid(spa);
5951	if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
5952		goto done;
5953
5954	if (vd) {
5955		value.value_type = SE_DATA_TYPE_UINT64;
5956		value.value.sv_uint64 = vd->vdev_guid;
5957		if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
5958		    SE_SLEEP) != 0)
5959			goto done;
5960
5961		if (vd->vdev_path) {
5962			value.value_type = SE_DATA_TYPE_STRING;
5963			value.value.sv_string = vd->vdev_path;
5964			if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
5965			    &value, SE_SLEEP) != 0)
5966				goto done;
5967		}
5968	}
5969
5970	if (sysevent_attach_attributes(ev, attr) != 0)
5971		goto done;
5972	attr = NULL;
5973
5974	(void) log_sysevent(ev, SE_SLEEP, &eid);
5975
5976done:
5977	if (attr)
5978		sysevent_free_attr(attr);
5979	sysevent_free(ev);
5980#endif
5981}
5982