spa.c revision 654b400c387942fc00d4f6869539adbd7b25fbce
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_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1068		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1069	}
1070	if (spa->spa_l2cache.sav_vdevs) {
1071		kmem_free(spa->spa_l2cache.sav_vdevs,
1072		    spa->spa_l2cache.sav_count * sizeof (void *));
1073		spa->spa_l2cache.sav_vdevs = NULL;
1074	}
1075	if (spa->spa_l2cache.sav_config) {
1076		nvlist_free(spa->spa_l2cache.sav_config);
1077		spa->spa_l2cache.sav_config = NULL;
1078	}
1079	spa->spa_l2cache.sav_count = 0;
1080
1081	spa->spa_async_suspended = 0;
1082
1083	if (spa->spa_comment != NULL) {
1084		spa_strfree(spa->spa_comment);
1085		spa->spa_comment = NULL;
1086	}
1087
1088	spa_config_exit(spa, SCL_ALL, FTAG);
1089}
1090
1091/*
1092 * Load (or re-load) the current list of vdevs describing the active spares for
1093 * this pool.  When this is called, we have some form of basic information in
1094 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1095 * then re-generate a more complete list including status information.
1096 */
1097static void
1098spa_load_spares(spa_t *spa)
1099{
1100	nvlist_t **spares;
1101	uint_t nspares;
1102	int i;
1103	vdev_t *vd, *tvd;
1104
1105	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1106
1107	/*
1108	 * First, close and free any existing spare vdevs.
1109	 */
1110	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1111		vd = spa->spa_spares.sav_vdevs[i];
1112
1113		/* Undo the call to spa_activate() below */
1114		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1115		    B_FALSE)) != NULL && tvd->vdev_isspare)
1116			spa_spare_remove(tvd);
1117		vdev_close(vd);
1118		vdev_free(vd);
1119	}
1120
1121	if (spa->spa_spares.sav_vdevs)
1122		kmem_free(spa->spa_spares.sav_vdevs,
1123		    spa->spa_spares.sav_count * sizeof (void *));
1124
1125	if (spa->spa_spares.sav_config == NULL)
1126		nspares = 0;
1127	else
1128		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1129		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1130
1131	spa->spa_spares.sav_count = (int)nspares;
1132	spa->spa_spares.sav_vdevs = NULL;
1133
1134	if (nspares == 0)
1135		return;
1136
1137	/*
1138	 * Construct the array of vdevs, opening them to get status in the
1139	 * process.   For each spare, there is potentially two different vdev_t
1140	 * structures associated with it: one in the list of spares (used only
1141	 * for basic validation purposes) and one in the active vdev
1142	 * configuration (if it's spared in).  During this phase we open and
1143	 * validate each vdev on the spare list.  If the vdev also exists in the
1144	 * active configuration, then we also mark this vdev as an active spare.
1145	 */
1146	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1147	    KM_SLEEP);
1148	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1149		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1150		    VDEV_ALLOC_SPARE) == 0);
1151		ASSERT(vd != NULL);
1152
1153		spa->spa_spares.sav_vdevs[i] = vd;
1154
1155		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1156		    B_FALSE)) != NULL) {
1157			if (!tvd->vdev_isspare)
1158				spa_spare_add(tvd);
1159
1160			/*
1161			 * We only mark the spare active if we were successfully
1162			 * able to load the vdev.  Otherwise, importing a pool
1163			 * with a bad active spare would result in strange
1164			 * behavior, because multiple pool would think the spare
1165			 * is actively in use.
1166			 *
1167			 * There is a vulnerability here to an equally bizarre
1168			 * circumstance, where a dead active spare is later
1169			 * brought back to life (onlined or otherwise).  Given
1170			 * the rarity of this scenario, and the extra complexity
1171			 * it adds, we ignore the possibility.
1172			 */
1173			if (!vdev_is_dead(tvd))
1174				spa_spare_activate(tvd);
1175		}
1176
1177		vd->vdev_top = vd;
1178		vd->vdev_aux = &spa->spa_spares;
1179
1180		if (vdev_open(vd) != 0)
1181			continue;
1182
1183		if (vdev_validate_aux(vd) == 0)
1184			spa_spare_add(vd);
1185	}
1186
1187	/*
1188	 * Recompute the stashed list of spares, with status information
1189	 * this time.
1190	 */
1191	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1192	    DATA_TYPE_NVLIST_ARRAY) == 0);
1193
1194	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1195	    KM_SLEEP);
1196	for (i = 0; i < spa->spa_spares.sav_count; i++)
1197		spares[i] = vdev_config_generate(spa,
1198		    spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1199	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1200	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1201	for (i = 0; i < spa->spa_spares.sav_count; i++)
1202		nvlist_free(spares[i]);
1203	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1204}
1205
1206/*
1207 * Load (or re-load) the current list of vdevs describing the active l2cache for
1208 * this pool.  When this is called, we have some form of basic information in
1209 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1210 * then re-generate a more complete list including status information.
1211 * Devices which are already active have their details maintained, and are
1212 * not re-opened.
1213 */
1214static void
1215spa_load_l2cache(spa_t *spa)
1216{
1217	nvlist_t **l2cache;
1218	uint_t nl2cache;
1219	int i, j, oldnvdevs;
1220	uint64_t guid;
1221	vdev_t *vd, **oldvdevs, **newvdevs;
1222	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1223
1224	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1225
1226	if (sav->sav_config != NULL) {
1227		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1228		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1229		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1230	} else {
1231		nl2cache = 0;
1232	}
1233
1234	oldvdevs = sav->sav_vdevs;
1235	oldnvdevs = sav->sav_count;
1236	sav->sav_vdevs = NULL;
1237	sav->sav_count = 0;
1238
1239	/*
1240	 * Process new nvlist of vdevs.
1241	 */
1242	for (i = 0; i < nl2cache; i++) {
1243		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1244		    &guid) == 0);
1245
1246		newvdevs[i] = NULL;
1247		for (j = 0; j < oldnvdevs; j++) {
1248			vd = oldvdevs[j];
1249			if (vd != NULL && guid == vd->vdev_guid) {
1250				/*
1251				 * Retain previous vdev for add/remove ops.
1252				 */
1253				newvdevs[i] = vd;
1254				oldvdevs[j] = NULL;
1255				break;
1256			}
1257		}
1258
1259		if (newvdevs[i] == NULL) {
1260			/*
1261			 * Create new vdev
1262			 */
1263			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1264			    VDEV_ALLOC_L2CACHE) == 0);
1265			ASSERT(vd != NULL);
1266			newvdevs[i] = vd;
1267
1268			/*
1269			 * Commit this vdev as an l2cache device,
1270			 * even if it fails to open.
1271			 */
1272			spa_l2cache_add(vd);
1273
1274			vd->vdev_top = vd;
1275			vd->vdev_aux = sav;
1276
1277			spa_l2cache_activate(vd);
1278
1279			if (vdev_open(vd) != 0)
1280				continue;
1281
1282			(void) vdev_validate_aux(vd);
1283
1284			if (!vdev_is_dead(vd))
1285				l2arc_add_vdev(spa, vd);
1286		}
1287	}
1288
1289	/*
1290	 * Purge vdevs that were dropped
1291	 */
1292	for (i = 0; i < oldnvdevs; i++) {
1293		uint64_t pool;
1294
1295		vd = oldvdevs[i];
1296		if (vd != NULL) {
1297			ASSERT(vd->vdev_isl2cache);
1298
1299			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1300			    pool != 0ULL && l2arc_vdev_present(vd))
1301				l2arc_remove_vdev(vd);
1302			vdev_clear_stats(vd);
1303			vdev_free(vd);
1304		}
1305	}
1306
1307	if (oldvdevs)
1308		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1309
1310	if (sav->sav_config == NULL)
1311		goto out;
1312
1313	sav->sav_vdevs = newvdevs;
1314	sav->sav_count = (int)nl2cache;
1315
1316	/*
1317	 * Recompute the stashed list of l2cache devices, with status
1318	 * information this time.
1319	 */
1320	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1321	    DATA_TYPE_NVLIST_ARRAY) == 0);
1322
1323	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1324	for (i = 0; i < sav->sav_count; i++)
1325		l2cache[i] = vdev_config_generate(spa,
1326		    sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1327	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1328	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1329out:
1330	for (i = 0; i < sav->sav_count; i++)
1331		nvlist_free(l2cache[i]);
1332	if (sav->sav_count)
1333		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1334}
1335
1336static int
1337load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1338{
1339	dmu_buf_t *db;
1340	char *packed = NULL;
1341	size_t nvsize = 0;
1342	int error;
1343	*value = NULL;
1344
1345	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1346	nvsize = *(uint64_t *)db->db_data;
1347	dmu_buf_rele(db, FTAG);
1348
1349	packed = kmem_alloc(nvsize, KM_SLEEP);
1350	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1351	    DMU_READ_PREFETCH);
1352	if (error == 0)
1353		error = nvlist_unpack(packed, nvsize, value, 0);
1354	kmem_free(packed, nvsize);
1355
1356	return (error);
1357}
1358
1359/*
1360 * Checks to see if the given vdev could not be opened, in which case we post a
1361 * sysevent to notify the autoreplace code that the device has been removed.
1362 */
1363static void
1364spa_check_removed(vdev_t *vd)
1365{
1366	for (int c = 0; c < vd->vdev_children; c++)
1367		spa_check_removed(vd->vdev_child[c]);
1368
1369	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1370		zfs_post_autoreplace(vd->vdev_spa, vd);
1371		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1372	}
1373}
1374
1375/*
1376 * Validate the current config against the MOS config
1377 */
1378static boolean_t
1379spa_config_valid(spa_t *spa, nvlist_t *config)
1380{
1381	vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1382	nvlist_t *nv;
1383
1384	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1385
1386	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1387	VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1388
1389	ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1390
1391	/*
1392	 * If we're doing a normal import, then build up any additional
1393	 * diagnostic information about missing devices in this config.
1394	 * We'll pass this up to the user for further processing.
1395	 */
1396	if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1397		nvlist_t **child, *nv;
1398		uint64_t idx = 0;
1399
1400		child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1401		    KM_SLEEP);
1402		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1403
1404		for (int c = 0; c < rvd->vdev_children; c++) {
1405			vdev_t *tvd = rvd->vdev_child[c];
1406			vdev_t *mtvd  = mrvd->vdev_child[c];
1407
1408			if (tvd->vdev_ops == &vdev_missing_ops &&
1409			    mtvd->vdev_ops != &vdev_missing_ops &&
1410			    mtvd->vdev_islog)
1411				child[idx++] = vdev_config_generate(spa, mtvd,
1412				    B_FALSE, 0);
1413		}
1414
1415		if (idx) {
1416			VERIFY(nvlist_add_nvlist_array(nv,
1417			    ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1418			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1419			    ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1420
1421			for (int i = 0; i < idx; i++)
1422				nvlist_free(child[i]);
1423		}
1424		nvlist_free(nv);
1425		kmem_free(child, rvd->vdev_children * sizeof (char **));
1426	}
1427
1428	/*
1429	 * Compare the root vdev tree with the information we have
1430	 * from the MOS config (mrvd). Check each top-level vdev
1431	 * with the corresponding MOS config top-level (mtvd).
1432	 */
1433	for (int c = 0; c < rvd->vdev_children; c++) {
1434		vdev_t *tvd = rvd->vdev_child[c];
1435		vdev_t *mtvd  = mrvd->vdev_child[c];
1436
1437		/*
1438		 * Resolve any "missing" vdevs in the current configuration.
1439		 * If we find that the MOS config has more accurate information
1440		 * about the top-level vdev then use that vdev instead.
1441		 */
1442		if (tvd->vdev_ops == &vdev_missing_ops &&
1443		    mtvd->vdev_ops != &vdev_missing_ops) {
1444
1445			if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1446				continue;
1447
1448			/*
1449			 * Device specific actions.
1450			 */
1451			if (mtvd->vdev_islog) {
1452				spa_set_log_state(spa, SPA_LOG_CLEAR);
1453			} else {
1454				/*
1455				 * XXX - once we have 'readonly' pool
1456				 * support we should be able to handle
1457				 * missing data devices by transitioning
1458				 * the pool to readonly.
1459				 */
1460				continue;
1461			}
1462
1463			/*
1464			 * Swap the missing vdev with the data we were
1465			 * able to obtain from the MOS config.
1466			 */
1467			vdev_remove_child(rvd, tvd);
1468			vdev_remove_child(mrvd, mtvd);
1469
1470			vdev_add_child(rvd, mtvd);
1471			vdev_add_child(mrvd, tvd);
1472
1473			spa_config_exit(spa, SCL_ALL, FTAG);
1474			vdev_load(mtvd);
1475			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1476
1477			vdev_reopen(rvd);
1478		} else if (mtvd->vdev_islog) {
1479			/*
1480			 * Load the slog device's state from the MOS config
1481			 * since it's possible that the label does not
1482			 * contain the most up-to-date information.
1483			 */
1484			vdev_load_log_state(tvd, mtvd);
1485			vdev_reopen(tvd);
1486		}
1487	}
1488	vdev_free(mrvd);
1489	spa_config_exit(spa, SCL_ALL, FTAG);
1490
1491	/*
1492	 * Ensure we were able to validate the config.
1493	 */
1494	return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1495}
1496
1497/*
1498 * Check for missing log devices
1499 */
1500static int
1501spa_check_logs(spa_t *spa)
1502{
1503	switch (spa->spa_log_state) {
1504	case SPA_LOG_MISSING:
1505		/* need to recheck in case slog has been restored */
1506	case SPA_LOG_UNKNOWN:
1507		if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1508		    DS_FIND_CHILDREN)) {
1509			spa_set_log_state(spa, SPA_LOG_MISSING);
1510			return (1);
1511		}
1512		break;
1513	}
1514	return (0);
1515}
1516
1517static boolean_t
1518spa_passivate_log(spa_t *spa)
1519{
1520	vdev_t *rvd = spa->spa_root_vdev;
1521	boolean_t slog_found = B_FALSE;
1522
1523	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1524
1525	if (!spa_has_slogs(spa))
1526		return (B_FALSE);
1527
1528	for (int c = 0; c < rvd->vdev_children; c++) {
1529		vdev_t *tvd = rvd->vdev_child[c];
1530		metaslab_group_t *mg = tvd->vdev_mg;
1531
1532		if (tvd->vdev_islog) {
1533			metaslab_group_passivate(mg);
1534			slog_found = B_TRUE;
1535		}
1536	}
1537
1538	return (slog_found);
1539}
1540
1541static void
1542spa_activate_log(spa_t *spa)
1543{
1544	vdev_t *rvd = spa->spa_root_vdev;
1545
1546	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1547
1548	for (int c = 0; c < rvd->vdev_children; c++) {
1549		vdev_t *tvd = rvd->vdev_child[c];
1550		metaslab_group_t *mg = tvd->vdev_mg;
1551
1552		if (tvd->vdev_islog)
1553			metaslab_group_activate(mg);
1554	}
1555}
1556
1557int
1558spa_offline_log(spa_t *spa)
1559{
1560	int error = 0;
1561
1562	if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1563	    NULL, DS_FIND_CHILDREN)) == 0) {
1564
1565		/*
1566		 * We successfully offlined the log device, sync out the
1567		 * current txg so that the "stubby" block can be removed
1568		 * by zil_sync().
1569		 */
1570		txg_wait_synced(spa->spa_dsl_pool, 0);
1571	}
1572	return (error);
1573}
1574
1575static void
1576spa_aux_check_removed(spa_aux_vdev_t *sav)
1577{
1578	for (int i = 0; i < sav->sav_count; i++)
1579		spa_check_removed(sav->sav_vdevs[i]);
1580}
1581
1582void
1583spa_claim_notify(zio_t *zio)
1584{
1585	spa_t *spa = zio->io_spa;
1586
1587	if (zio->io_error)
1588		return;
1589
1590	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1591	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1592		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1593	mutex_exit(&spa->spa_props_lock);
1594}
1595
1596typedef struct spa_load_error {
1597	uint64_t	sle_meta_count;
1598	uint64_t	sle_data_count;
1599} spa_load_error_t;
1600
1601static void
1602spa_load_verify_done(zio_t *zio)
1603{
1604	blkptr_t *bp = zio->io_bp;
1605	spa_load_error_t *sle = zio->io_private;
1606	dmu_object_type_t type = BP_GET_TYPE(bp);
1607	int error = zio->io_error;
1608
1609	if (error) {
1610		if ((BP_GET_LEVEL(bp) != 0 || dmu_ot[type].ot_metadata) &&
1611		    type != DMU_OT_INTENT_LOG)
1612			atomic_add_64(&sle->sle_meta_count, 1);
1613		else
1614			atomic_add_64(&sle->sle_data_count, 1);
1615	}
1616	zio_data_buf_free(zio->io_data, zio->io_size);
1617}
1618
1619/*ARGSUSED*/
1620static int
1621spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1622    arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1623{
1624	if (bp != NULL) {
1625		zio_t *rio = arg;
1626		size_t size = BP_GET_PSIZE(bp);
1627		void *data = zio_data_buf_alloc(size);
1628
1629		zio_nowait(zio_read(rio, spa, bp, data, size,
1630		    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1631		    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1632		    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1633	}
1634	return (0);
1635}
1636
1637static int
1638spa_load_verify(spa_t *spa)
1639{
1640	zio_t *rio;
1641	spa_load_error_t sle = { 0 };
1642	zpool_rewind_policy_t policy;
1643	boolean_t verify_ok = B_FALSE;
1644	int error;
1645
1646	zpool_get_rewind_policy(spa->spa_config, &policy);
1647
1648	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1649		return (0);
1650
1651	rio = zio_root(spa, NULL, &sle,
1652	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1653
1654	error = traverse_pool(spa, spa->spa_verify_min_txg,
1655	    TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1656
1657	(void) zio_wait(rio);
1658
1659	spa->spa_load_meta_errors = sle.sle_meta_count;
1660	spa->spa_load_data_errors = sle.sle_data_count;
1661
1662	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1663	    sle.sle_data_count <= policy.zrp_maxdata) {
1664		int64_t loss = 0;
1665
1666		verify_ok = B_TRUE;
1667		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1668		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1669
1670		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1671		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1672		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1673		VERIFY(nvlist_add_int64(spa->spa_load_info,
1674		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1675		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1676		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1677	} else {
1678		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1679	}
1680
1681	if (error) {
1682		if (error != ENXIO && error != EIO)
1683			error = EIO;
1684		return (error);
1685	}
1686
1687	return (verify_ok ? 0 : EIO);
1688}
1689
1690/*
1691 * Find a value in the pool props object.
1692 */
1693static void
1694spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1695{
1696	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1697	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1698}
1699
1700/*
1701 * Find a value in the pool directory object.
1702 */
1703static int
1704spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1705{
1706	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1707	    name, sizeof (uint64_t), 1, val));
1708}
1709
1710static int
1711spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1712{
1713	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1714	return (err);
1715}
1716
1717/*
1718 * Fix up config after a partly-completed split.  This is done with the
1719 * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1720 * pool have that entry in their config, but only the splitting one contains
1721 * a list of all the guids of the vdevs that are being split off.
1722 *
1723 * This function determines what to do with that list: either rejoin
1724 * all the disks to the pool, or complete the splitting process.  To attempt
1725 * the rejoin, each disk that is offlined is marked online again, and
1726 * we do a reopen() call.  If the vdev label for every disk that was
1727 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1728 * then we call vdev_split() on each disk, and complete the split.
1729 *
1730 * Otherwise we leave the config alone, with all the vdevs in place in
1731 * the original pool.
1732 */
1733static void
1734spa_try_repair(spa_t *spa, nvlist_t *config)
1735{
1736	uint_t extracted;
1737	uint64_t *glist;
1738	uint_t i, gcount;
1739	nvlist_t *nvl;
1740	vdev_t **vd;
1741	boolean_t attempt_reopen;
1742
1743	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1744		return;
1745
1746	/* check that the config is complete */
1747	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1748	    &glist, &gcount) != 0)
1749		return;
1750
1751	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1752
1753	/* attempt to online all the vdevs & validate */
1754	attempt_reopen = B_TRUE;
1755	for (i = 0; i < gcount; i++) {
1756		if (glist[i] == 0)	/* vdev is hole */
1757			continue;
1758
1759		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1760		if (vd[i] == NULL) {
1761			/*
1762			 * Don't bother attempting to reopen the disks;
1763			 * just do the split.
1764			 */
1765			attempt_reopen = B_FALSE;
1766		} else {
1767			/* attempt to re-online it */
1768			vd[i]->vdev_offline = B_FALSE;
1769		}
1770	}
1771
1772	if (attempt_reopen) {
1773		vdev_reopen(spa->spa_root_vdev);
1774
1775		/* check each device to see what state it's in */
1776		for (extracted = 0, i = 0; i < gcount; i++) {
1777			if (vd[i] != NULL &&
1778			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1779				break;
1780			++extracted;
1781		}
1782	}
1783
1784	/*
1785	 * If every disk has been moved to the new pool, or if we never
1786	 * even attempted to look at them, then we split them off for
1787	 * good.
1788	 */
1789	if (!attempt_reopen || gcount == extracted) {
1790		for (i = 0; i < gcount; i++)
1791			if (vd[i] != NULL)
1792				vdev_split(vd[i]);
1793		vdev_reopen(spa->spa_root_vdev);
1794	}
1795
1796	kmem_free(vd, gcount * sizeof (vdev_t *));
1797}
1798
1799static int
1800spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1801    boolean_t mosconfig)
1802{
1803	nvlist_t *config = spa->spa_config;
1804	char *ereport = FM_EREPORT_ZFS_POOL;
1805	char *comment;
1806	int error;
1807	uint64_t pool_guid;
1808	nvlist_t *nvl;
1809
1810	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1811		return (EINVAL);
1812
1813	ASSERT(spa->spa_comment == NULL);
1814	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
1815		spa->spa_comment = spa_strdup(comment);
1816
1817	/*
1818	 * Versioning wasn't explicitly added to the label until later, so if
1819	 * it's not present treat it as the initial version.
1820	 */
1821	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1822	    &spa->spa_ubsync.ub_version) != 0)
1823		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1824
1825	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1826	    &spa->spa_config_txg);
1827
1828	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1829	    spa_guid_exists(pool_guid, 0)) {
1830		error = EEXIST;
1831	} else {
1832		spa->spa_config_guid = pool_guid;
1833
1834		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1835		    &nvl) == 0) {
1836			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1837			    KM_SLEEP) == 0);
1838		}
1839
1840		gethrestime(&spa->spa_loaded_ts);
1841		error = spa_load_impl(spa, pool_guid, config, state, type,
1842		    mosconfig, &ereport);
1843	}
1844
1845	spa->spa_minref = refcount_count(&spa->spa_refcount);
1846	if (error) {
1847		if (error != EEXIST) {
1848			spa->spa_loaded_ts.tv_sec = 0;
1849			spa->spa_loaded_ts.tv_nsec = 0;
1850		}
1851		if (error != EBADF) {
1852			zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1853		}
1854	}
1855	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1856	spa->spa_ena = 0;
1857
1858	return (error);
1859}
1860
1861/*
1862 * Load an existing storage pool, using the pool's builtin spa_config as a
1863 * source of configuration information.
1864 */
1865static int
1866spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1867    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1868    char **ereport)
1869{
1870	int error = 0;
1871	nvlist_t *nvroot = NULL;
1872	vdev_t *rvd;
1873	uberblock_t *ub = &spa->spa_uberblock;
1874	uint64_t children, config_cache_txg = spa->spa_config_txg;
1875	int orig_mode = spa->spa_mode;
1876	int parse;
1877	uint64_t obj;
1878
1879	/*
1880	 * If this is an untrusted config, access the pool in read-only mode.
1881	 * This prevents things like resilvering recently removed devices.
1882	 */
1883	if (!mosconfig)
1884		spa->spa_mode = FREAD;
1885
1886	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1887
1888	spa->spa_load_state = state;
1889
1890	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1891		return (EINVAL);
1892
1893	parse = (type == SPA_IMPORT_EXISTING ?
1894	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1895
1896	/*
1897	 * Create "The Godfather" zio to hold all async IOs
1898	 */
1899	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1900	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1901
1902	/*
1903	 * Parse the configuration into a vdev tree.  We explicitly set the
1904	 * value that will be returned by spa_version() since parsing the
1905	 * configuration requires knowing the version number.
1906	 */
1907	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1908	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1909	spa_config_exit(spa, SCL_ALL, FTAG);
1910
1911	if (error != 0)
1912		return (error);
1913
1914	ASSERT(spa->spa_root_vdev == rvd);
1915
1916	if (type != SPA_IMPORT_ASSEMBLE) {
1917		ASSERT(spa_guid(spa) == pool_guid);
1918	}
1919
1920	/*
1921	 * Try to open all vdevs, loading each label in the process.
1922	 */
1923	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1924	error = vdev_open(rvd);
1925	spa_config_exit(spa, SCL_ALL, FTAG);
1926	if (error != 0)
1927		return (error);
1928
1929	/*
1930	 * We need to validate the vdev labels against the configuration that
1931	 * we have in hand, which is dependent on the setting of mosconfig. If
1932	 * mosconfig is true then we're validating the vdev labels based on
1933	 * that config.  Otherwise, we're validating against the cached config
1934	 * (zpool.cache) that was read when we loaded the zfs module, and then
1935	 * later we will recursively call spa_load() and validate against
1936	 * the vdev config.
1937	 *
1938	 * If we're assembling a new pool that's been split off from an
1939	 * existing pool, the labels haven't yet been updated so we skip
1940	 * validation for now.
1941	 */
1942	if (type != SPA_IMPORT_ASSEMBLE) {
1943		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1944		error = vdev_validate(rvd, mosconfig);
1945		spa_config_exit(spa, SCL_ALL, FTAG);
1946
1947		if (error != 0)
1948			return (error);
1949
1950		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1951			return (ENXIO);
1952	}
1953
1954	/*
1955	 * Find the best uberblock.
1956	 */
1957	vdev_uberblock_load(NULL, rvd, ub);
1958
1959	/*
1960	 * If we weren't able to find a single valid uberblock, return failure.
1961	 */
1962	if (ub->ub_txg == 0)
1963		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
1964
1965	/*
1966	 * If the pool is newer than the code, we can't open it.
1967	 */
1968	if (ub->ub_version > SPA_VERSION)
1969		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
1970
1971	/*
1972	 * If the vdev guid sum doesn't match the uberblock, we have an
1973	 * incomplete configuration.  We first check to see if the pool
1974	 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
1975	 * If it is, defer the vdev_guid_sum check till later so we
1976	 * can handle missing vdevs.
1977	 */
1978	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
1979	    &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
1980	    rvd->vdev_guid_sum != ub->ub_guid_sum)
1981		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
1982
1983	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
1984		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1985		spa_try_repair(spa, config);
1986		spa_config_exit(spa, SCL_ALL, FTAG);
1987		nvlist_free(spa->spa_config_splitting);
1988		spa->spa_config_splitting = NULL;
1989	}
1990
1991	/*
1992	 * Initialize internal SPA structures.
1993	 */
1994	spa->spa_state = POOL_STATE_ACTIVE;
1995	spa->spa_ubsync = spa->spa_uberblock;
1996	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
1997	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
1998	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
1999	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2000	spa->spa_claim_max_txg = spa->spa_first_txg;
2001	spa->spa_prev_software_version = ub->ub_software_version;
2002
2003	error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2004	if (error)
2005		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2006	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2007
2008	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2009		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2010
2011	if (!mosconfig) {
2012		uint64_t hostid;
2013		nvlist_t *policy = NULL, *nvconfig;
2014
2015		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2016			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2017
2018		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2019		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2020			char *hostname;
2021			unsigned long myhostid = 0;
2022
2023			VERIFY(nvlist_lookup_string(nvconfig,
2024			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2025
2026#ifdef	_KERNEL
2027			myhostid = zone_get_hostid(NULL);
2028#else	/* _KERNEL */
2029			/*
2030			 * We're emulating the system's hostid in userland, so
2031			 * we can't use zone_get_hostid().
2032			 */
2033			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2034#endif	/* _KERNEL */
2035			if (hostid != 0 && myhostid != 0 &&
2036			    hostid != myhostid) {
2037				nvlist_free(nvconfig);
2038				cmn_err(CE_WARN, "pool '%s' could not be "
2039				    "loaded as it was last accessed by "
2040				    "another system (host: %s hostid: 0x%lx). "
2041				    "See: http://illumos.org/msg/ZFS-8000-EY",
2042				    spa_name(spa), hostname,
2043				    (unsigned long)hostid);
2044				return (EBADF);
2045			}
2046		}
2047		if (nvlist_lookup_nvlist(spa->spa_config,
2048		    ZPOOL_REWIND_POLICY, &policy) == 0)
2049			VERIFY(nvlist_add_nvlist(nvconfig,
2050			    ZPOOL_REWIND_POLICY, policy) == 0);
2051
2052		spa_config_set(spa, nvconfig);
2053		spa_unload(spa);
2054		spa_deactivate(spa);
2055		spa_activate(spa, orig_mode);
2056
2057		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2058	}
2059
2060	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2061		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2062	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2063	if (error != 0)
2064		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2065
2066	/*
2067	 * Load the bit that tells us to use the new accounting function
2068	 * (raid-z deflation).  If we have an older pool, this will not
2069	 * be present.
2070	 */
2071	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2072	if (error != 0 && error != ENOENT)
2073		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2074
2075	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2076	    &spa->spa_creation_version);
2077	if (error != 0 && error != ENOENT)
2078		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2079
2080	/*
2081	 * Load the persistent error log.  If we have an older pool, this will
2082	 * not be present.
2083	 */
2084	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2085	if (error != 0 && error != ENOENT)
2086		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2087
2088	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2089	    &spa->spa_errlog_scrub);
2090	if (error != 0 && error != ENOENT)
2091		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2092
2093	/*
2094	 * Load the history object.  If we have an older pool, this
2095	 * will not be present.
2096	 */
2097	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2098	if (error != 0 && error != ENOENT)
2099		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2100
2101	/*
2102	 * If we're assembling the pool from the split-off vdevs of
2103	 * an existing pool, we don't want to attach the spares & cache
2104	 * devices.
2105	 */
2106
2107	/*
2108	 * Load any hot spares for this pool.
2109	 */
2110	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2111	if (error != 0 && error != ENOENT)
2112		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2113	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2114		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2115		if (load_nvlist(spa, spa->spa_spares.sav_object,
2116		    &spa->spa_spares.sav_config) != 0)
2117			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2118
2119		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2120		spa_load_spares(spa);
2121		spa_config_exit(spa, SCL_ALL, FTAG);
2122	} else if (error == 0) {
2123		spa->spa_spares.sav_sync = B_TRUE;
2124	}
2125
2126	/*
2127	 * Load any level 2 ARC devices for this pool.
2128	 */
2129	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2130	    &spa->spa_l2cache.sav_object);
2131	if (error != 0 && error != ENOENT)
2132		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2133	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2134		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2135		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2136		    &spa->spa_l2cache.sav_config) != 0)
2137			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2138
2139		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2140		spa_load_l2cache(spa);
2141		spa_config_exit(spa, SCL_ALL, FTAG);
2142	} else if (error == 0) {
2143		spa->spa_l2cache.sav_sync = B_TRUE;
2144	}
2145
2146	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2147
2148	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2149	if (error && error != ENOENT)
2150		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2151
2152	if (error == 0) {
2153		uint64_t autoreplace;
2154
2155		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2156		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2157		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2158		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2159		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2160		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2161		    &spa->spa_dedup_ditto);
2162
2163		spa->spa_autoreplace = (autoreplace != 0);
2164	}
2165
2166	/*
2167	 * If the 'autoreplace' property is set, then post a resource notifying
2168	 * the ZFS DE that it should not issue any faults for unopenable
2169	 * devices.  We also iterate over the vdevs, and post a sysevent for any
2170	 * unopenable vdevs so that the normal autoreplace handler can take
2171	 * over.
2172	 */
2173	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2174		spa_check_removed(spa->spa_root_vdev);
2175		/*
2176		 * For the import case, this is done in spa_import(), because
2177		 * at this point we're using the spare definitions from
2178		 * the MOS config, not necessarily from the userland config.
2179		 */
2180		if (state != SPA_LOAD_IMPORT) {
2181			spa_aux_check_removed(&spa->spa_spares);
2182			spa_aux_check_removed(&spa->spa_l2cache);
2183		}
2184	}
2185
2186	/*
2187	 * Load the vdev state for all toplevel vdevs.
2188	 */
2189	vdev_load(rvd);
2190
2191	/*
2192	 * Propagate the leaf DTLs we just loaded all the way up the tree.
2193	 */
2194	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2195	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2196	spa_config_exit(spa, SCL_ALL, FTAG);
2197
2198	/*
2199	 * Load the DDTs (dedup tables).
2200	 */
2201	error = ddt_load(spa);
2202	if (error != 0)
2203		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2204
2205	spa_update_dspace(spa);
2206
2207	/*
2208	 * Validate the config, using the MOS config to fill in any
2209	 * information which might be missing.  If we fail to validate
2210	 * the config then declare the pool unfit for use. If we're
2211	 * assembling a pool from a split, the log is not transferred
2212	 * over.
2213	 */
2214	if (type != SPA_IMPORT_ASSEMBLE) {
2215		nvlist_t *nvconfig;
2216
2217		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2218			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2219
2220		if (!spa_config_valid(spa, nvconfig)) {
2221			nvlist_free(nvconfig);
2222			return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2223			    ENXIO));
2224		}
2225		nvlist_free(nvconfig);
2226
2227		/*
2228		 * Now that we've validate the config, check the state of the
2229		 * root vdev.  If it can't be opened, it indicates one or
2230		 * more toplevel vdevs are faulted.
2231		 */
2232		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2233			return (ENXIO);
2234
2235		if (spa_check_logs(spa)) {
2236			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2237			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2238		}
2239	}
2240
2241	/*
2242	 * We've successfully opened the pool, verify that we're ready
2243	 * to start pushing transactions.
2244	 */
2245	if (state != SPA_LOAD_TRYIMPORT) {
2246		if (error = spa_load_verify(spa))
2247			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2248			    error));
2249	}
2250
2251	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2252	    spa->spa_load_max_txg == UINT64_MAX)) {
2253		dmu_tx_t *tx;
2254		int need_update = B_FALSE;
2255
2256		ASSERT(state != SPA_LOAD_TRYIMPORT);
2257
2258		/*
2259		 * Claim log blocks that haven't been committed yet.
2260		 * This must all happen in a single txg.
2261		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2262		 * invoked from zil_claim_log_block()'s i/o done callback.
2263		 * Price of rollback is that we abandon the log.
2264		 */
2265		spa->spa_claiming = B_TRUE;
2266
2267		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2268		    spa_first_txg(spa));
2269		(void) dmu_objset_find(spa_name(spa),
2270		    zil_claim, tx, DS_FIND_CHILDREN);
2271		dmu_tx_commit(tx);
2272
2273		spa->spa_claiming = B_FALSE;
2274
2275		spa_set_log_state(spa, SPA_LOG_GOOD);
2276		spa->spa_sync_on = B_TRUE;
2277		txg_sync_start(spa->spa_dsl_pool);
2278
2279		/*
2280		 * Wait for all claims to sync.  We sync up to the highest
2281		 * claimed log block birth time so that claimed log blocks
2282		 * don't appear to be from the future.  spa_claim_max_txg
2283		 * will have been set for us by either zil_check_log_chain()
2284		 * (invoked from spa_check_logs()) or zil_claim() above.
2285		 */
2286		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2287
2288		/*
2289		 * If the config cache is stale, or we have uninitialized
2290		 * metaslabs (see spa_vdev_add()), then update the config.
2291		 *
2292		 * If this is a verbatim import, trust the current
2293		 * in-core spa_config and update the disk labels.
2294		 */
2295		if (config_cache_txg != spa->spa_config_txg ||
2296		    state == SPA_LOAD_IMPORT ||
2297		    state == SPA_LOAD_RECOVER ||
2298		    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2299			need_update = B_TRUE;
2300
2301		for (int c = 0; c < rvd->vdev_children; c++)
2302			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2303				need_update = B_TRUE;
2304
2305		/*
2306		 * Update the config cache asychronously in case we're the
2307		 * root pool, in which case the config cache isn't writable yet.
2308		 */
2309		if (need_update)
2310			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2311
2312		/*
2313		 * Check all DTLs to see if anything needs resilvering.
2314		 */
2315		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2316		    vdev_resilver_needed(rvd, NULL, NULL))
2317			spa_async_request(spa, SPA_ASYNC_RESILVER);
2318
2319		/*
2320		 * Delete any inconsistent datasets.
2321		 */
2322		(void) dmu_objset_find(spa_name(spa),
2323		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2324
2325		/*
2326		 * Clean up any stale temporary dataset userrefs.
2327		 */
2328		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2329	}
2330
2331	return (0);
2332}
2333
2334static int
2335spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2336{
2337	int mode = spa->spa_mode;
2338
2339	spa_unload(spa);
2340	spa_deactivate(spa);
2341
2342	spa->spa_load_max_txg--;
2343
2344	spa_activate(spa, mode);
2345	spa_async_suspend(spa);
2346
2347	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2348}
2349
2350static int
2351spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2352    uint64_t max_request, int rewind_flags)
2353{
2354	nvlist_t *config = NULL;
2355	int load_error, rewind_error;
2356	uint64_t safe_rewind_txg;
2357	uint64_t min_txg;
2358
2359	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2360		spa->spa_load_max_txg = spa->spa_load_txg;
2361		spa_set_log_state(spa, SPA_LOG_CLEAR);
2362	} else {
2363		spa->spa_load_max_txg = max_request;
2364	}
2365
2366	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2367	    mosconfig);
2368	if (load_error == 0)
2369		return (0);
2370
2371	if (spa->spa_root_vdev != NULL)
2372		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2373
2374	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2375	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2376
2377	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2378		nvlist_free(config);
2379		return (load_error);
2380	}
2381
2382	/* Price of rolling back is discarding txgs, including log */
2383	if (state == SPA_LOAD_RECOVER)
2384		spa_set_log_state(spa, SPA_LOG_CLEAR);
2385
2386	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2387	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2388	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2389	    TXG_INITIAL : safe_rewind_txg;
2390
2391	/*
2392	 * Continue as long as we're finding errors, we're still within
2393	 * the acceptable rewind range, and we're still finding uberblocks
2394	 */
2395	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2396	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2397		if (spa->spa_load_max_txg < safe_rewind_txg)
2398			spa->spa_extreme_rewind = B_TRUE;
2399		rewind_error = spa_load_retry(spa, state, mosconfig);
2400	}
2401
2402	spa->spa_extreme_rewind = B_FALSE;
2403	spa->spa_load_max_txg = UINT64_MAX;
2404
2405	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2406		spa_config_set(spa, config);
2407
2408	return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
2409}
2410
2411/*
2412 * Pool Open/Import
2413 *
2414 * The import case is identical to an open except that the configuration is sent
2415 * down from userland, instead of grabbed from the configuration cache.  For the
2416 * case of an open, the pool configuration will exist in the
2417 * POOL_STATE_UNINITIALIZED state.
2418 *
2419 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2420 * the same time open the pool, without having to keep around the spa_t in some
2421 * ambiguous state.
2422 */
2423static int
2424spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2425    nvlist_t **config)
2426{
2427	spa_t *spa;
2428	spa_load_state_t state = SPA_LOAD_OPEN;
2429	int error;
2430	int locked = B_FALSE;
2431
2432	*spapp = NULL;
2433
2434	/*
2435	 * As disgusting as this is, we need to support recursive calls to this
2436	 * function because dsl_dir_open() is called during spa_load(), and ends
2437	 * up calling spa_open() again.  The real fix is to figure out how to
2438	 * avoid dsl_dir_open() calling this in the first place.
2439	 */
2440	if (mutex_owner(&spa_namespace_lock) != curthread) {
2441		mutex_enter(&spa_namespace_lock);
2442		locked = B_TRUE;
2443	}
2444
2445	if ((spa = spa_lookup(pool)) == NULL) {
2446		if (locked)
2447			mutex_exit(&spa_namespace_lock);
2448		return (ENOENT);
2449	}
2450
2451	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2452		zpool_rewind_policy_t policy;
2453
2454		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2455		    &policy);
2456		if (policy.zrp_request & ZPOOL_DO_REWIND)
2457			state = SPA_LOAD_RECOVER;
2458
2459		spa_activate(spa, spa_mode_global);
2460
2461		if (state != SPA_LOAD_RECOVER)
2462			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2463
2464		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2465		    policy.zrp_request);
2466
2467		if (error == EBADF) {
2468			/*
2469			 * If vdev_validate() returns failure (indicated by
2470			 * EBADF), it indicates that one of the vdevs indicates
2471			 * that the pool has been exported or destroyed.  If
2472			 * this is the case, the config cache is out of sync and
2473			 * we should remove the pool from the namespace.
2474			 */
2475			spa_unload(spa);
2476			spa_deactivate(spa);
2477			spa_config_sync(spa, B_TRUE, B_TRUE);
2478			spa_remove(spa);
2479			if (locked)
2480				mutex_exit(&spa_namespace_lock);
2481			return (ENOENT);
2482		}
2483
2484		if (error) {
2485			/*
2486			 * We can't open the pool, but we still have useful
2487			 * information: the state of each vdev after the
2488			 * attempted vdev_open().  Return this to the user.
2489			 */
2490			if (config != NULL && spa->spa_config) {
2491				VERIFY(nvlist_dup(spa->spa_config, config,
2492				    KM_SLEEP) == 0);
2493				VERIFY(nvlist_add_nvlist(*config,
2494				    ZPOOL_CONFIG_LOAD_INFO,
2495				    spa->spa_load_info) == 0);
2496			}
2497			spa_unload(spa);
2498			spa_deactivate(spa);
2499			spa->spa_last_open_failed = error;
2500			if (locked)
2501				mutex_exit(&spa_namespace_lock);
2502			*spapp = NULL;
2503			return (error);
2504		}
2505	}
2506
2507	spa_open_ref(spa, tag);
2508
2509	if (config != NULL)
2510		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2511
2512	/*
2513	 * If we've recovered the pool, pass back any information we
2514	 * gathered while doing the load.
2515	 */
2516	if (state == SPA_LOAD_RECOVER) {
2517		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2518		    spa->spa_load_info) == 0);
2519	}
2520
2521	if (locked) {
2522		spa->spa_last_open_failed = 0;
2523		spa->spa_last_ubsync_txg = 0;
2524		spa->spa_load_txg = 0;
2525		mutex_exit(&spa_namespace_lock);
2526	}
2527
2528	*spapp = spa;
2529
2530	return (0);
2531}
2532
2533int
2534spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2535    nvlist_t **config)
2536{
2537	return (spa_open_common(name, spapp, tag, policy, config));
2538}
2539
2540int
2541spa_open(const char *name, spa_t **spapp, void *tag)
2542{
2543	return (spa_open_common(name, spapp, tag, NULL, NULL));
2544}
2545
2546/*
2547 * Lookup the given spa_t, incrementing the inject count in the process,
2548 * preventing it from being exported or destroyed.
2549 */
2550spa_t *
2551spa_inject_addref(char *name)
2552{
2553	spa_t *spa;
2554
2555	mutex_enter(&spa_namespace_lock);
2556	if ((spa = spa_lookup(name)) == NULL) {
2557		mutex_exit(&spa_namespace_lock);
2558		return (NULL);
2559	}
2560	spa->spa_inject_ref++;
2561	mutex_exit(&spa_namespace_lock);
2562
2563	return (spa);
2564}
2565
2566void
2567spa_inject_delref(spa_t *spa)
2568{
2569	mutex_enter(&spa_namespace_lock);
2570	spa->spa_inject_ref--;
2571	mutex_exit(&spa_namespace_lock);
2572}
2573
2574/*
2575 * Add spares device information to the nvlist.
2576 */
2577static void
2578spa_add_spares(spa_t *spa, nvlist_t *config)
2579{
2580	nvlist_t **spares;
2581	uint_t i, nspares;
2582	nvlist_t *nvroot;
2583	uint64_t guid;
2584	vdev_stat_t *vs;
2585	uint_t vsc;
2586	uint64_t pool;
2587
2588	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2589
2590	if (spa->spa_spares.sav_count == 0)
2591		return;
2592
2593	VERIFY(nvlist_lookup_nvlist(config,
2594	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2595	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2596	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2597	if (nspares != 0) {
2598		VERIFY(nvlist_add_nvlist_array(nvroot,
2599		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2600		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2601		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2602
2603		/*
2604		 * Go through and find any spares which have since been
2605		 * repurposed as an active spare.  If this is the case, update
2606		 * their status appropriately.
2607		 */
2608		for (i = 0; i < nspares; i++) {
2609			VERIFY(nvlist_lookup_uint64(spares[i],
2610			    ZPOOL_CONFIG_GUID, &guid) == 0);
2611			if (spa_spare_exists(guid, &pool, NULL) &&
2612			    pool != 0ULL) {
2613				VERIFY(nvlist_lookup_uint64_array(
2614				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
2615				    (uint64_t **)&vs, &vsc) == 0);
2616				vs->vs_state = VDEV_STATE_CANT_OPEN;
2617				vs->vs_aux = VDEV_AUX_SPARED;
2618			}
2619		}
2620	}
2621}
2622
2623/*
2624 * Add l2cache device information to the nvlist, including vdev stats.
2625 */
2626static void
2627spa_add_l2cache(spa_t *spa, nvlist_t *config)
2628{
2629	nvlist_t **l2cache;
2630	uint_t i, j, nl2cache;
2631	nvlist_t *nvroot;
2632	uint64_t guid;
2633	vdev_t *vd;
2634	vdev_stat_t *vs;
2635	uint_t vsc;
2636
2637	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2638
2639	if (spa->spa_l2cache.sav_count == 0)
2640		return;
2641
2642	VERIFY(nvlist_lookup_nvlist(config,
2643	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2644	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2645	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2646	if (nl2cache != 0) {
2647		VERIFY(nvlist_add_nvlist_array(nvroot,
2648		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2649		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2650		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2651
2652		/*
2653		 * Update level 2 cache device stats.
2654		 */
2655
2656		for (i = 0; i < nl2cache; i++) {
2657			VERIFY(nvlist_lookup_uint64(l2cache[i],
2658			    ZPOOL_CONFIG_GUID, &guid) == 0);
2659
2660			vd = NULL;
2661			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2662				if (guid ==
2663				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2664					vd = spa->spa_l2cache.sav_vdevs[j];
2665					break;
2666				}
2667			}
2668			ASSERT(vd != NULL);
2669
2670			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2671			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2672			    == 0);
2673			vdev_get_stats(vd, vs);
2674		}
2675	}
2676}
2677
2678int
2679spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2680{
2681	int error;
2682	spa_t *spa;
2683
2684	*config = NULL;
2685	error = spa_open_common(name, &spa, FTAG, NULL, config);
2686
2687	if (spa != NULL) {
2688		/*
2689		 * This still leaves a window of inconsistency where the spares
2690		 * or l2cache devices could change and the config would be
2691		 * self-inconsistent.
2692		 */
2693		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2694
2695		if (*config != NULL) {
2696			uint64_t loadtimes[2];
2697
2698			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
2699			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
2700			VERIFY(nvlist_add_uint64_array(*config,
2701			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
2702
2703			VERIFY(nvlist_add_uint64(*config,
2704			    ZPOOL_CONFIG_ERRCOUNT,
2705			    spa_get_errlog_size(spa)) == 0);
2706
2707			if (spa_suspended(spa))
2708				VERIFY(nvlist_add_uint64(*config,
2709				    ZPOOL_CONFIG_SUSPENDED,
2710				    spa->spa_failmode) == 0);
2711
2712			spa_add_spares(spa, *config);
2713			spa_add_l2cache(spa, *config);
2714		}
2715	}
2716
2717	/*
2718	 * We want to get the alternate root even for faulted pools, so we cheat
2719	 * and call spa_lookup() directly.
2720	 */
2721	if (altroot) {
2722		if (spa == NULL) {
2723			mutex_enter(&spa_namespace_lock);
2724			spa = spa_lookup(name);
2725			if (spa)
2726				spa_altroot(spa, altroot, buflen);
2727			else
2728				altroot[0] = '\0';
2729			spa = NULL;
2730			mutex_exit(&spa_namespace_lock);
2731		} else {
2732			spa_altroot(spa, altroot, buflen);
2733		}
2734	}
2735
2736	if (spa != NULL) {
2737		spa_config_exit(spa, SCL_CONFIG, FTAG);
2738		spa_close(spa, FTAG);
2739	}
2740
2741	return (error);
2742}
2743
2744/*
2745 * Validate that the auxiliary device array is well formed.  We must have an
2746 * array of nvlists, each which describes a valid leaf vdev.  If this is an
2747 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2748 * specified, as long as they are well-formed.
2749 */
2750static int
2751spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2752    spa_aux_vdev_t *sav, const char *config, uint64_t version,
2753    vdev_labeltype_t label)
2754{
2755	nvlist_t **dev;
2756	uint_t i, ndev;
2757	vdev_t *vd;
2758	int error;
2759
2760	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2761
2762	/*
2763	 * It's acceptable to have no devs specified.
2764	 */
2765	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2766		return (0);
2767
2768	if (ndev == 0)
2769		return (EINVAL);
2770
2771	/*
2772	 * Make sure the pool is formatted with a version that supports this
2773	 * device type.
2774	 */
2775	if (spa_version(spa) < version)
2776		return (ENOTSUP);
2777
2778	/*
2779	 * Set the pending device list so we correctly handle device in-use
2780	 * checking.
2781	 */
2782	sav->sav_pending = dev;
2783	sav->sav_npending = ndev;
2784
2785	for (i = 0; i < ndev; i++) {
2786		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2787		    mode)) != 0)
2788			goto out;
2789
2790		if (!vd->vdev_ops->vdev_op_leaf) {
2791			vdev_free(vd);
2792			error = EINVAL;
2793			goto out;
2794		}
2795
2796		/*
2797		 * The L2ARC currently only supports disk devices in
2798		 * kernel context.  For user-level testing, we allow it.
2799		 */
2800#ifdef _KERNEL
2801		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2802		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2803			error = ENOTBLK;
2804			vdev_free(vd);
2805			goto out;
2806		}
2807#endif
2808		vd->vdev_top = vd;
2809
2810		if ((error = vdev_open(vd)) == 0 &&
2811		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
2812			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2813			    vd->vdev_guid) == 0);
2814		}
2815
2816		vdev_free(vd);
2817
2818		if (error &&
2819		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2820			goto out;
2821		else
2822			error = 0;
2823	}
2824
2825out:
2826	sav->sav_pending = NULL;
2827	sav->sav_npending = 0;
2828	return (error);
2829}
2830
2831static int
2832spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2833{
2834	int error;
2835
2836	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2837
2838	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2839	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2840	    VDEV_LABEL_SPARE)) != 0) {
2841		return (error);
2842	}
2843
2844	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2845	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2846	    VDEV_LABEL_L2CACHE));
2847}
2848
2849static void
2850spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2851    const char *config)
2852{
2853	int i;
2854
2855	if (sav->sav_config != NULL) {
2856		nvlist_t **olddevs;
2857		uint_t oldndevs;
2858		nvlist_t **newdevs;
2859
2860		/*
2861		 * Generate new dev list by concatentating with the
2862		 * current dev list.
2863		 */
2864		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2865		    &olddevs, &oldndevs) == 0);
2866
2867		newdevs = kmem_alloc(sizeof (void *) *
2868		    (ndevs + oldndevs), KM_SLEEP);
2869		for (i = 0; i < oldndevs; i++)
2870			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2871			    KM_SLEEP) == 0);
2872		for (i = 0; i < ndevs; i++)
2873			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2874			    KM_SLEEP) == 0);
2875
2876		VERIFY(nvlist_remove(sav->sav_config, config,
2877		    DATA_TYPE_NVLIST_ARRAY) == 0);
2878
2879		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2880		    config, newdevs, ndevs + oldndevs) == 0);
2881		for (i = 0; i < oldndevs + ndevs; i++)
2882			nvlist_free(newdevs[i]);
2883		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2884	} else {
2885		/*
2886		 * Generate a new dev list.
2887		 */
2888		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2889		    KM_SLEEP) == 0);
2890		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2891		    devs, ndevs) == 0);
2892	}
2893}
2894
2895/*
2896 * Stop and drop level 2 ARC devices
2897 */
2898void
2899spa_l2cache_drop(spa_t *spa)
2900{
2901	vdev_t *vd;
2902	int i;
2903	spa_aux_vdev_t *sav = &spa->spa_l2cache;
2904
2905	for (i = 0; i < sav->sav_count; i++) {
2906		uint64_t pool;
2907
2908		vd = sav->sav_vdevs[i];
2909		ASSERT(vd != NULL);
2910
2911		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2912		    pool != 0ULL && l2arc_vdev_present(vd))
2913			l2arc_remove_vdev(vd);
2914	}
2915}
2916
2917/*
2918 * Pool Creation
2919 */
2920int
2921spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2922    const char *history_str, nvlist_t *zplprops)
2923{
2924	spa_t *spa;
2925	char *altroot = NULL;
2926	vdev_t *rvd;
2927	dsl_pool_t *dp;
2928	dmu_tx_t *tx;
2929	int error = 0;
2930	uint64_t txg = TXG_INITIAL;
2931	nvlist_t **spares, **l2cache;
2932	uint_t nspares, nl2cache;
2933	uint64_t version, obj;
2934
2935	/*
2936	 * If this pool already exists, return failure.
2937	 */
2938	mutex_enter(&spa_namespace_lock);
2939	if (spa_lookup(pool) != NULL) {
2940		mutex_exit(&spa_namespace_lock);
2941		return (EEXIST);
2942	}
2943
2944	/*
2945	 * Allocate a new spa_t structure.
2946	 */
2947	(void) nvlist_lookup_string(props,
2948	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2949	spa = spa_add(pool, NULL, altroot);
2950	spa_activate(spa, spa_mode_global);
2951
2952	if (props && (error = spa_prop_validate(spa, props))) {
2953		spa_deactivate(spa);
2954		spa_remove(spa);
2955		mutex_exit(&spa_namespace_lock);
2956		return (error);
2957	}
2958
2959	if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2960	    &version) != 0)
2961		version = SPA_VERSION;
2962	ASSERT(version <= SPA_VERSION);
2963
2964	spa->spa_first_txg = txg;
2965	spa->spa_uberblock.ub_txg = txg - 1;
2966	spa->spa_uberblock.ub_version = version;
2967	spa->spa_ubsync = spa->spa_uberblock;
2968
2969	/*
2970	 * Create "The Godfather" zio to hold all async IOs
2971	 */
2972	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2973	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2974
2975	/*
2976	 * Create the root vdev.
2977	 */
2978	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2979
2980	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2981
2982	ASSERT(error != 0 || rvd != NULL);
2983	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2984
2985	if (error == 0 && !zfs_allocatable_devs(nvroot))
2986		error = EINVAL;
2987
2988	if (error == 0 &&
2989	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2990	    (error = spa_validate_aux(spa, nvroot, txg,
2991	    VDEV_ALLOC_ADD)) == 0) {
2992		for (int c = 0; c < rvd->vdev_children; c++) {
2993			vdev_metaslab_set_size(rvd->vdev_child[c]);
2994			vdev_expand(rvd->vdev_child[c], txg);
2995		}
2996	}
2997
2998	spa_config_exit(spa, SCL_ALL, FTAG);
2999
3000	if (error != 0) {
3001		spa_unload(spa);
3002		spa_deactivate(spa);
3003		spa_remove(spa);
3004		mutex_exit(&spa_namespace_lock);
3005		return (error);
3006	}
3007
3008	/*
3009	 * Get the list of spares, if specified.
3010	 */
3011	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3012	    &spares, &nspares) == 0) {
3013		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3014		    KM_SLEEP) == 0);
3015		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3016		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3017		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3018		spa_load_spares(spa);
3019		spa_config_exit(spa, SCL_ALL, FTAG);
3020		spa->spa_spares.sav_sync = B_TRUE;
3021	}
3022
3023	/*
3024	 * Get the list of level 2 cache devices, if specified.
3025	 */
3026	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3027	    &l2cache, &nl2cache) == 0) {
3028		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3029		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3030		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3031		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3032		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3033		spa_load_l2cache(spa);
3034		spa_config_exit(spa, SCL_ALL, FTAG);
3035		spa->spa_l2cache.sav_sync = B_TRUE;
3036	}
3037
3038	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3039	spa->spa_meta_objset = dp->dp_meta_objset;
3040
3041	/*
3042	 * Create DDTs (dedup tables).
3043	 */
3044	ddt_create(spa);
3045
3046	spa_update_dspace(spa);
3047
3048	tx = dmu_tx_create_assigned(dp, txg);
3049
3050	/*
3051	 * Create the pool config object.
3052	 */
3053	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3054	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3055	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3056
3057	if (zap_add(spa->spa_meta_objset,
3058	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3059	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3060		cmn_err(CE_PANIC, "failed to add pool config");
3061	}
3062
3063	if (zap_add(spa->spa_meta_objset,
3064	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3065	    sizeof (uint64_t), 1, &version, tx) != 0) {
3066		cmn_err(CE_PANIC, "failed to add pool version");
3067	}
3068
3069	/* Newly created pools with the right version are always deflated. */
3070	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3071		spa->spa_deflate = TRUE;
3072		if (zap_add(spa->spa_meta_objset,
3073		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3074		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3075			cmn_err(CE_PANIC, "failed to add deflate");
3076		}
3077	}
3078
3079	/*
3080	 * Create the deferred-free bpobj.  Turn off compression
3081	 * because sync-to-convergence takes longer if the blocksize
3082	 * keeps changing.
3083	 */
3084	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3085	dmu_object_set_compress(spa->spa_meta_objset, obj,
3086	    ZIO_COMPRESS_OFF, tx);
3087	if (zap_add(spa->spa_meta_objset,
3088	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3089	    sizeof (uint64_t), 1, &obj, tx) != 0) {
3090		cmn_err(CE_PANIC, "failed to add bpobj");
3091	}
3092	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3093	    spa->spa_meta_objset, obj));
3094
3095	/*
3096	 * Create the pool's history object.
3097	 */
3098	if (version >= SPA_VERSION_ZPOOL_HISTORY)
3099		spa_history_create_obj(spa, tx);
3100
3101	/*
3102	 * Set pool properties.
3103	 */
3104	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3105	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3106	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3107	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3108
3109	if (props != NULL) {
3110		spa_configfile_set(spa, props, B_FALSE);
3111		spa_sync_props(spa, props, tx);
3112	}
3113
3114	dmu_tx_commit(tx);
3115
3116	spa->spa_sync_on = B_TRUE;
3117	txg_sync_start(spa->spa_dsl_pool);
3118
3119	/*
3120	 * We explicitly wait for the first transaction to complete so that our
3121	 * bean counters are appropriately updated.
3122	 */
3123	txg_wait_synced(spa->spa_dsl_pool, txg);
3124
3125	spa_config_sync(spa, B_FALSE, B_TRUE);
3126
3127	if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3128		(void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3129	spa_history_log_version(spa, LOG_POOL_CREATE);
3130
3131	spa->spa_minref = refcount_count(&spa->spa_refcount);
3132
3133	mutex_exit(&spa_namespace_lock);
3134
3135	return (0);
3136}
3137
3138#ifdef _KERNEL
3139/*
3140 * Get the root pool information from the root disk, then import the root pool
3141 * during the system boot up time.
3142 */
3143extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3144
3145static nvlist_t *
3146spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3147{
3148	nvlist_t *config;
3149	nvlist_t *nvtop, *nvroot;
3150	uint64_t pgid;
3151
3152	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3153		return (NULL);
3154
3155	/*
3156	 * Add this top-level vdev to the child array.
3157	 */
3158	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3159	    &nvtop) == 0);
3160	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3161	    &pgid) == 0);
3162	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3163
3164	/*
3165	 * Put this pool's top-level vdevs into a root vdev.
3166	 */
3167	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3168	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3169	    VDEV_TYPE_ROOT) == 0);
3170	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3171	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3172	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3173	    &nvtop, 1) == 0);
3174
3175	/*
3176	 * Replace the existing vdev_tree with the new root vdev in
3177	 * this pool's configuration (remove the old, add the new).
3178	 */
3179	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3180	nvlist_free(nvroot);
3181	return (config);
3182}
3183
3184/*
3185 * Walk the vdev tree and see if we can find a device with "better"
3186 * configuration. A configuration is "better" if the label on that
3187 * device has a more recent txg.
3188 */
3189static void
3190spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3191{
3192	for (int c = 0; c < vd->vdev_children; c++)
3193		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3194
3195	if (vd->vdev_ops->vdev_op_leaf) {
3196		nvlist_t *label;
3197		uint64_t label_txg;
3198
3199		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3200		    &label) != 0)
3201			return;
3202
3203		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3204		    &label_txg) == 0);
3205
3206		/*
3207		 * Do we have a better boot device?
3208		 */
3209		if (label_txg > *txg) {
3210			*txg = label_txg;
3211			*avd = vd;
3212		}
3213		nvlist_free(label);
3214	}
3215}
3216
3217/*
3218 * Import a root pool.
3219 *
3220 * For x86. devpath_list will consist of devid and/or physpath name of
3221 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3222 * The GRUB "findroot" command will return the vdev we should boot.
3223 *
3224 * For Sparc, devpath_list consists the physpath name of the booting device
3225 * no matter the rootpool is a single device pool or a mirrored pool.
3226 * e.g.
3227 *	"/pci@1f,0/ide@d/disk@0,0:a"
3228 */
3229int
3230spa_import_rootpool(char *devpath, char *devid)
3231{
3232	spa_t *spa;
3233	vdev_t *rvd, *bvd, *avd = NULL;
3234	nvlist_t *config, *nvtop;
3235	uint64_t guid, txg;
3236	char *pname;
3237	int error;
3238
3239	/*
3240	 * Read the label from the boot device and generate a configuration.
3241	 */
3242	config = spa_generate_rootconf(devpath, devid, &guid);
3243#if defined(_OBP) && defined(_KERNEL)
3244	if (config == NULL) {
3245		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3246			/* iscsi boot */
3247			get_iscsi_bootpath_phy(devpath);
3248			config = spa_generate_rootconf(devpath, devid, &guid);
3249		}
3250	}
3251#endif
3252	if (config == NULL) {
3253		cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
3254		    devpath);
3255		return (EIO);
3256	}
3257
3258	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3259	    &pname) == 0);
3260	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3261
3262	mutex_enter(&spa_namespace_lock);
3263	if ((spa = spa_lookup(pname)) != NULL) {
3264		/*
3265		 * Remove the existing root pool from the namespace so that we
3266		 * can replace it with the correct config we just read in.
3267		 */
3268		spa_remove(spa);
3269	}
3270
3271	spa = spa_add(pname, config, NULL);
3272	spa->spa_is_root = B_TRUE;
3273	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3274
3275	/*
3276	 * Build up a vdev tree based on the boot device's label config.
3277	 */
3278	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3279	    &nvtop) == 0);
3280	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3281	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3282	    VDEV_ALLOC_ROOTPOOL);
3283	spa_config_exit(spa, SCL_ALL, FTAG);
3284	if (error) {
3285		mutex_exit(&spa_namespace_lock);
3286		nvlist_free(config);
3287		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3288		    pname);
3289		return (error);
3290	}
3291
3292	/*
3293	 * Get the boot vdev.
3294	 */
3295	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3296		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3297		    (u_longlong_t)guid);
3298		error = ENOENT;
3299		goto out;
3300	}
3301
3302	/*
3303	 * Determine if there is a better boot device.
3304	 */
3305	avd = bvd;
3306	spa_alt_rootvdev(rvd, &avd, &txg);
3307	if (avd != bvd) {
3308		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3309		    "try booting from '%s'", avd->vdev_path);
3310		error = EINVAL;
3311		goto out;
3312	}
3313
3314	/*
3315	 * If the boot device is part of a spare vdev then ensure that
3316	 * we're booting off the active spare.
3317	 */
3318	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3319	    !bvd->vdev_isspare) {
3320		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3321		    "try booting from '%s'",
3322		    bvd->vdev_parent->
3323		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3324		error = EINVAL;
3325		goto out;
3326	}
3327
3328	error = 0;
3329	spa_history_log_version(spa, LOG_POOL_IMPORT);
3330out:
3331	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3332	vdev_free(rvd);
3333	spa_config_exit(spa, SCL_ALL, FTAG);
3334	mutex_exit(&spa_namespace_lock);
3335
3336	nvlist_free(config);
3337	return (error);
3338}
3339
3340#endif
3341
3342/*
3343 * Import a non-root pool into the system.
3344 */
3345int
3346spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3347{
3348	spa_t *spa;
3349	char *altroot = NULL;
3350	spa_load_state_t state = SPA_LOAD_IMPORT;
3351	zpool_rewind_policy_t policy;
3352	uint64_t mode = spa_mode_global;
3353	uint64_t readonly = B_FALSE;
3354	int error;
3355	nvlist_t *nvroot;
3356	nvlist_t **spares, **l2cache;
3357	uint_t nspares, nl2cache;
3358
3359	/*
3360	 * If a pool with this name exists, return failure.
3361	 */
3362	mutex_enter(&spa_namespace_lock);
3363	if (spa_lookup(pool) != NULL) {
3364		mutex_exit(&spa_namespace_lock);
3365		return (EEXIST);
3366	}
3367
3368	/*
3369	 * Create and initialize the spa structure.
3370	 */
3371	(void) nvlist_lookup_string(props,
3372	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3373	(void) nvlist_lookup_uint64(props,
3374	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3375	if (readonly)
3376		mode = FREAD;
3377	spa = spa_add(pool, config, altroot);
3378	spa->spa_import_flags = flags;
3379
3380	/*
3381	 * Verbatim import - Take a pool and insert it into the namespace
3382	 * as if it had been loaded at boot.
3383	 */
3384	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3385		if (props != NULL)
3386			spa_configfile_set(spa, props, B_FALSE);
3387
3388		spa_config_sync(spa, B_FALSE, B_TRUE);
3389
3390		mutex_exit(&spa_namespace_lock);
3391		spa_history_log_version(spa, LOG_POOL_IMPORT);
3392
3393		return (0);
3394	}
3395
3396	spa_activate(spa, mode);
3397
3398	/*
3399	 * Don't start async tasks until we know everything is healthy.
3400	 */
3401	spa_async_suspend(spa);
3402
3403	zpool_get_rewind_policy(config, &policy);
3404	if (policy.zrp_request & ZPOOL_DO_REWIND)
3405		state = SPA_LOAD_RECOVER;
3406
3407	/*
3408	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3409	 * because the user-supplied config is actually the one to trust when
3410	 * doing an import.
3411	 */
3412	if (state != SPA_LOAD_RECOVER)
3413		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3414
3415	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3416	    policy.zrp_request);
3417
3418	/*
3419	 * Propagate anything learned while loading the pool and pass it
3420	 * back to caller (i.e. rewind info, missing devices, etc).
3421	 */
3422	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3423	    spa->spa_load_info) == 0);
3424
3425	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3426	/*
3427	 * Toss any existing sparelist, as it doesn't have any validity
3428	 * anymore, and conflicts with spa_has_spare().
3429	 */
3430	if (spa->spa_spares.sav_config) {
3431		nvlist_free(spa->spa_spares.sav_config);
3432		spa->spa_spares.sav_config = NULL;
3433		spa_load_spares(spa);
3434	}
3435	if (spa->spa_l2cache.sav_config) {
3436		nvlist_free(spa->spa_l2cache.sav_config);
3437		spa->spa_l2cache.sav_config = NULL;
3438		spa_load_l2cache(spa);
3439	}
3440
3441	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3442	    &nvroot) == 0);
3443	if (error == 0)
3444		error = spa_validate_aux(spa, nvroot, -1ULL,
3445		    VDEV_ALLOC_SPARE);
3446	if (error == 0)
3447		error = spa_validate_aux(spa, nvroot, -1ULL,
3448		    VDEV_ALLOC_L2CACHE);
3449	spa_config_exit(spa, SCL_ALL, FTAG);
3450
3451	if (props != NULL)
3452		spa_configfile_set(spa, props, B_FALSE);
3453
3454	if (error != 0 || (props && spa_writeable(spa) &&
3455	    (error = spa_prop_set(spa, props)))) {
3456		spa_unload(spa);
3457		spa_deactivate(spa);
3458		spa_remove(spa);
3459		mutex_exit(&spa_namespace_lock);
3460		return (error);
3461	}
3462
3463	spa_async_resume(spa);
3464
3465	/*
3466	 * Override any spares and level 2 cache devices as specified by
3467	 * the user, as these may have correct device names/devids, etc.
3468	 */
3469	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3470	    &spares, &nspares) == 0) {
3471		if (spa->spa_spares.sav_config)
3472			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3473			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3474		else
3475			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3476			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3477		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3478		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3479		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3480		spa_load_spares(spa);
3481		spa_config_exit(spa, SCL_ALL, FTAG);
3482		spa->spa_spares.sav_sync = B_TRUE;
3483	}
3484	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3485	    &l2cache, &nl2cache) == 0) {
3486		if (spa->spa_l2cache.sav_config)
3487			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3488			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3489		else
3490			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3491			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3492		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3493		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3494		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3495		spa_load_l2cache(spa);
3496		spa_config_exit(spa, SCL_ALL, FTAG);
3497		spa->spa_l2cache.sav_sync = B_TRUE;
3498	}
3499
3500	/*
3501	 * Check for any removed devices.
3502	 */
3503	if (spa->spa_autoreplace) {
3504		spa_aux_check_removed(&spa->spa_spares);
3505		spa_aux_check_removed(&spa->spa_l2cache);
3506	}
3507
3508	if (spa_writeable(spa)) {
3509		/*
3510		 * Update the config cache to include the newly-imported pool.
3511		 */
3512		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3513	}
3514
3515	/*
3516	 * It's possible that the pool was expanded while it was exported.
3517	 * We kick off an async task to handle this for us.
3518	 */
3519	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3520
3521	mutex_exit(&spa_namespace_lock);
3522	spa_history_log_version(spa, LOG_POOL_IMPORT);
3523
3524	return (0);
3525}
3526
3527nvlist_t *
3528spa_tryimport(nvlist_t *tryconfig)
3529{
3530	nvlist_t *config = NULL;
3531	char *poolname;
3532	spa_t *spa;
3533	uint64_t state;
3534	int error;
3535
3536	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3537		return (NULL);
3538
3539	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3540		return (NULL);
3541
3542	/*
3543	 * Create and initialize the spa structure.
3544	 */
3545	mutex_enter(&spa_namespace_lock);
3546	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3547	spa_activate(spa, FREAD);
3548
3549	/*
3550	 * Pass off the heavy lifting to spa_load().
3551	 * Pass TRUE for mosconfig because the user-supplied config
3552	 * is actually the one to trust when doing an import.
3553	 */
3554	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3555
3556	/*
3557	 * If 'tryconfig' was at least parsable, return the current config.
3558	 */
3559	if (spa->spa_root_vdev != NULL) {
3560		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3561		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3562		    poolname) == 0);
3563		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3564		    state) == 0);
3565		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3566		    spa->spa_uberblock.ub_timestamp) == 0);
3567
3568		/*
3569		 * If the bootfs property exists on this pool then we
3570		 * copy it out so that external consumers can tell which
3571		 * pools are bootable.
3572		 */
3573		if ((!error || error == EEXIST) && spa->spa_bootfs) {
3574			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3575
3576			/*
3577			 * We have to play games with the name since the
3578			 * pool was opened as TRYIMPORT_NAME.
3579			 */
3580			if (dsl_dsobj_to_dsname(spa_name(spa),
3581			    spa->spa_bootfs, tmpname) == 0) {
3582				char *cp;
3583				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3584
3585				cp = strchr(tmpname, '/');
3586				if (cp == NULL) {
3587					(void) strlcpy(dsname, tmpname,
3588					    MAXPATHLEN);
3589				} else {
3590					(void) snprintf(dsname, MAXPATHLEN,
3591					    "%s/%s", poolname, ++cp);
3592				}
3593				VERIFY(nvlist_add_string(config,
3594				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3595				kmem_free(dsname, MAXPATHLEN);
3596			}
3597			kmem_free(tmpname, MAXPATHLEN);
3598		}
3599
3600		/*
3601		 * Add the list of hot spares and level 2 cache devices.
3602		 */
3603		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3604		spa_add_spares(spa, config);
3605		spa_add_l2cache(spa, config);
3606		spa_config_exit(spa, SCL_CONFIG, FTAG);
3607	}
3608
3609	spa_unload(spa);
3610	spa_deactivate(spa);
3611	spa_remove(spa);
3612	mutex_exit(&spa_namespace_lock);
3613
3614	return (config);
3615}
3616
3617/*
3618 * Pool export/destroy
3619 *
3620 * The act of destroying or exporting a pool is very simple.  We make sure there
3621 * is no more pending I/O and any references to the pool are gone.  Then, we
3622 * update the pool state and sync all the labels to disk, removing the
3623 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3624 * we don't sync the labels or remove the configuration cache.
3625 */
3626static int
3627spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3628    boolean_t force, boolean_t hardforce)
3629{
3630	spa_t *spa;
3631
3632	if (oldconfig)
3633		*oldconfig = NULL;
3634
3635	if (!(spa_mode_global & FWRITE))
3636		return (EROFS);
3637
3638	mutex_enter(&spa_namespace_lock);
3639	if ((spa = spa_lookup(pool)) == NULL) {
3640		mutex_exit(&spa_namespace_lock);
3641		return (ENOENT);
3642	}
3643
3644	/*
3645	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3646	 * reacquire the namespace lock, and see if we can export.
3647	 */
3648	spa_open_ref(spa, FTAG);
3649	mutex_exit(&spa_namespace_lock);
3650	spa_async_suspend(spa);
3651	mutex_enter(&spa_namespace_lock);
3652	spa_close(spa, FTAG);
3653
3654	/*
3655	 * The pool will be in core if it's openable,
3656	 * in which case we can modify its state.
3657	 */
3658	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3659		/*
3660		 * Objsets may be open only because they're dirty, so we
3661		 * have to force it to sync before checking spa_refcnt.
3662		 */
3663		txg_wait_synced(spa->spa_dsl_pool, 0);
3664
3665		/*
3666		 * A pool cannot be exported or destroyed if there are active
3667		 * references.  If we are resetting a pool, allow references by
3668		 * fault injection handlers.
3669		 */
3670		if (!spa_refcount_zero(spa) ||
3671		    (spa->spa_inject_ref != 0 &&
3672		    new_state != POOL_STATE_UNINITIALIZED)) {
3673			spa_async_resume(spa);
3674			mutex_exit(&spa_namespace_lock);
3675			return (EBUSY);
3676		}
3677
3678		/*
3679		 * A pool cannot be exported if it has an active shared spare.
3680		 * This is to prevent other pools stealing the active spare
3681		 * from an exported pool. At user's own will, such pool can
3682		 * be forcedly exported.
3683		 */
3684		if (!force && new_state == POOL_STATE_EXPORTED &&
3685		    spa_has_active_shared_spare(spa)) {
3686			spa_async_resume(spa);
3687			mutex_exit(&spa_namespace_lock);
3688			return (EXDEV);
3689		}
3690
3691		/*
3692		 * We want this to be reflected on every label,
3693		 * so mark them all dirty.  spa_unload() will do the
3694		 * final sync that pushes these changes out.
3695		 */
3696		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3697			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3698			spa->spa_state = new_state;
3699			spa->spa_final_txg = spa_last_synced_txg(spa) +
3700			    TXG_DEFER_SIZE + 1;
3701			vdev_config_dirty(spa->spa_root_vdev);
3702			spa_config_exit(spa, SCL_ALL, FTAG);
3703		}
3704	}
3705
3706	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
3707
3708	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3709		spa_unload(spa);
3710		spa_deactivate(spa);
3711	}
3712
3713	if (oldconfig && spa->spa_config)
3714		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3715
3716	if (new_state != POOL_STATE_UNINITIALIZED) {
3717		if (!hardforce)
3718			spa_config_sync(spa, B_TRUE, B_TRUE);
3719		spa_remove(spa);
3720	}
3721	mutex_exit(&spa_namespace_lock);
3722
3723	return (0);
3724}
3725
3726/*
3727 * Destroy a storage pool.
3728 */
3729int
3730spa_destroy(char *pool)
3731{
3732	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3733	    B_FALSE, B_FALSE));
3734}
3735
3736/*
3737 * Export a storage pool.
3738 */
3739int
3740spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3741    boolean_t hardforce)
3742{
3743	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3744	    force, hardforce));
3745}
3746
3747/*
3748 * Similar to spa_export(), this unloads the spa_t without actually removing it
3749 * from the namespace in any way.
3750 */
3751int
3752spa_reset(char *pool)
3753{
3754	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3755	    B_FALSE, B_FALSE));
3756}
3757
3758/*
3759 * ==========================================================================
3760 * Device manipulation
3761 * ==========================================================================
3762 */
3763
3764/*
3765 * Add a device to a storage pool.
3766 */
3767int
3768spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3769{
3770	uint64_t txg, id;
3771	int error;
3772	vdev_t *rvd = spa->spa_root_vdev;
3773	vdev_t *vd, *tvd;
3774	nvlist_t **spares, **l2cache;
3775	uint_t nspares, nl2cache;
3776
3777	ASSERT(spa_writeable(spa));
3778
3779	txg = spa_vdev_enter(spa);
3780
3781	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3782	    VDEV_ALLOC_ADD)) != 0)
3783		return (spa_vdev_exit(spa, NULL, txg, error));
3784
3785	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
3786
3787	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3788	    &nspares) != 0)
3789		nspares = 0;
3790
3791	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3792	    &nl2cache) != 0)
3793		nl2cache = 0;
3794
3795	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3796		return (spa_vdev_exit(spa, vd, txg, EINVAL));
3797
3798	if (vd->vdev_children != 0 &&
3799	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
3800		return (spa_vdev_exit(spa, vd, txg, error));
3801
3802	/*
3803	 * We must validate the spares and l2cache devices after checking the
3804	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
3805	 */
3806	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3807		return (spa_vdev_exit(spa, vd, txg, error));
3808
3809	/*
3810	 * Transfer each new top-level vdev from vd to rvd.
3811	 */
3812	for (int c = 0; c < vd->vdev_children; c++) {
3813
3814		/*
3815		 * Set the vdev id to the first hole, if one exists.
3816		 */
3817		for (id = 0; id < rvd->vdev_children; id++) {
3818			if (rvd->vdev_child[id]->vdev_ishole) {
3819				vdev_free(rvd->vdev_child[id]);
3820				break;
3821			}
3822		}
3823		tvd = vd->vdev_child[c];
3824		vdev_remove_child(vd, tvd);
3825		tvd->vdev_id = id;
3826		vdev_add_child(rvd, tvd);
3827		vdev_config_dirty(tvd);
3828	}
3829
3830	if (nspares != 0) {
3831		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3832		    ZPOOL_CONFIG_SPARES);
3833		spa_load_spares(spa);
3834		spa->spa_spares.sav_sync = B_TRUE;
3835	}
3836
3837	if (nl2cache != 0) {
3838		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3839		    ZPOOL_CONFIG_L2CACHE);
3840		spa_load_l2cache(spa);
3841		spa->spa_l2cache.sav_sync = B_TRUE;
3842	}
3843
3844	/*
3845	 * We have to be careful when adding new vdevs to an existing pool.
3846	 * If other threads start allocating from these vdevs before we
3847	 * sync the config cache, and we lose power, then upon reboot we may
3848	 * fail to open the pool because there are DVAs that the config cache
3849	 * can't translate.  Therefore, we first add the vdevs without
3850	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3851	 * and then let spa_config_update() initialize the new metaslabs.
3852	 *
3853	 * spa_load() checks for added-but-not-initialized vdevs, so that
3854	 * if we lose power at any point in this sequence, the remaining
3855	 * steps will be completed the next time we load the pool.
3856	 */
3857	(void) spa_vdev_exit(spa, vd, txg, 0);
3858
3859	mutex_enter(&spa_namespace_lock);
3860	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3861	mutex_exit(&spa_namespace_lock);
3862
3863	return (0);
3864}
3865
3866/*
3867 * Attach a device to a mirror.  The arguments are the path to any device
3868 * in the mirror, and the nvroot for the new device.  If the path specifies
3869 * a device that is not mirrored, we automatically insert the mirror vdev.
3870 *
3871 * If 'replacing' is specified, the new device is intended to replace the
3872 * existing device; in this case the two devices are made into their own
3873 * mirror using the 'replacing' vdev, which is functionally identical to
3874 * the mirror vdev (it actually reuses all the same ops) but has a few
3875 * extra rules: you can't attach to it after it's been created, and upon
3876 * completion of resilvering, the first disk (the one being replaced)
3877 * is automatically detached.
3878 */
3879int
3880spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3881{
3882	uint64_t txg, dtl_max_txg;
3883	vdev_t *rvd = spa->spa_root_vdev;
3884	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3885	vdev_ops_t *pvops;
3886	char *oldvdpath, *newvdpath;
3887	int newvd_isspare;
3888	int error;
3889
3890	ASSERT(spa_writeable(spa));
3891
3892	txg = spa_vdev_enter(spa);
3893
3894	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3895
3896	if (oldvd == NULL)
3897		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3898
3899	if (!oldvd->vdev_ops->vdev_op_leaf)
3900		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3901
3902	pvd = oldvd->vdev_parent;
3903
3904	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3905	    VDEV_ALLOC_ATTACH)) != 0)
3906		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3907
3908	if (newrootvd->vdev_children != 1)
3909		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3910
3911	newvd = newrootvd->vdev_child[0];
3912
3913	if (!newvd->vdev_ops->vdev_op_leaf)
3914		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3915
3916	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3917		return (spa_vdev_exit(spa, newrootvd, txg, error));
3918
3919	/*
3920	 * Spares can't replace logs
3921	 */
3922	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3923		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3924
3925	if (!replacing) {
3926		/*
3927		 * For attach, the only allowable parent is a mirror or the root
3928		 * vdev.
3929		 */
3930		if (pvd->vdev_ops != &vdev_mirror_ops &&
3931		    pvd->vdev_ops != &vdev_root_ops)
3932			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3933
3934		pvops = &vdev_mirror_ops;
3935	} else {
3936		/*
3937		 * Active hot spares can only be replaced by inactive hot
3938		 * spares.
3939		 */
3940		if (pvd->vdev_ops == &vdev_spare_ops &&
3941		    oldvd->vdev_isspare &&
3942		    !spa_has_spare(spa, newvd->vdev_guid))
3943			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3944
3945		/*
3946		 * If the source is a hot spare, and the parent isn't already a
3947		 * spare, then we want to create a new hot spare.  Otherwise, we
3948		 * want to create a replacing vdev.  The user is not allowed to
3949		 * attach to a spared vdev child unless the 'isspare' state is
3950		 * the same (spare replaces spare, non-spare replaces
3951		 * non-spare).
3952		 */
3953		if (pvd->vdev_ops == &vdev_replacing_ops &&
3954		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
3955			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3956		} else if (pvd->vdev_ops == &vdev_spare_ops &&
3957		    newvd->vdev_isspare != oldvd->vdev_isspare) {
3958			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3959		}
3960
3961		if (newvd->vdev_isspare)
3962			pvops = &vdev_spare_ops;
3963		else
3964			pvops = &vdev_replacing_ops;
3965	}
3966
3967	/*
3968	 * Make sure the new device is big enough.
3969	 */
3970	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3971		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3972
3973	/*
3974	 * The new device cannot have a higher alignment requirement
3975	 * than the top-level vdev.
3976	 */
3977	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3978		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3979
3980	/*
3981	 * If this is an in-place replacement, update oldvd's path and devid
3982	 * to make it distinguishable from newvd, and unopenable from now on.
3983	 */
3984	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3985		spa_strfree(oldvd->vdev_path);
3986		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3987		    KM_SLEEP);
3988		(void) sprintf(oldvd->vdev_path, "%s/%s",
3989		    newvd->vdev_path, "old");
3990		if (oldvd->vdev_devid != NULL) {
3991			spa_strfree(oldvd->vdev_devid);
3992			oldvd->vdev_devid = NULL;
3993		}
3994	}
3995
3996	/* mark the device being resilvered */
3997	newvd->vdev_resilvering = B_TRUE;
3998
3999	/*
4000	 * If the parent is not a mirror, or if we're replacing, insert the new
4001	 * mirror/replacing/spare vdev above oldvd.
4002	 */
4003	if (pvd->vdev_ops != pvops)
4004		pvd = vdev_add_parent(oldvd, pvops);
4005
4006	ASSERT(pvd->vdev_top->vdev_parent == rvd);
4007	ASSERT(pvd->vdev_ops == pvops);
4008	ASSERT(oldvd->vdev_parent == pvd);
4009
4010	/*
4011	 * Extract the new device from its root and add it to pvd.
4012	 */
4013	vdev_remove_child(newrootvd, newvd);
4014	newvd->vdev_id = pvd->vdev_children;
4015	newvd->vdev_crtxg = oldvd->vdev_crtxg;
4016	vdev_add_child(pvd, newvd);
4017
4018	tvd = newvd->vdev_top;
4019	ASSERT(pvd->vdev_top == tvd);
4020	ASSERT(tvd->vdev_parent == rvd);
4021
4022	vdev_config_dirty(tvd);
4023
4024	/*
4025	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4026	 * for any dmu_sync-ed blocks.  It will propagate upward when
4027	 * spa_vdev_exit() calls vdev_dtl_reassess().
4028	 */
4029	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4030
4031	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4032	    dtl_max_txg - TXG_INITIAL);
4033
4034	if (newvd->vdev_isspare) {
4035		spa_spare_activate(newvd);
4036		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4037	}
4038
4039	oldvdpath = spa_strdup(oldvd->vdev_path);
4040	newvdpath = spa_strdup(newvd->vdev_path);
4041	newvd_isspare = newvd->vdev_isspare;
4042
4043	/*
4044	 * Mark newvd's DTL dirty in this txg.
4045	 */
4046	vdev_dirty(tvd, VDD_DTL, newvd, txg);
4047
4048	/*
4049	 * Restart the resilver
4050	 */
4051	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4052
4053	/*
4054	 * Commit the config
4055	 */
4056	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4057
4058	spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
4059	    "%s vdev=%s %s vdev=%s",
4060	    replacing && newvd_isspare ? "spare in" :
4061	    replacing ? "replace" : "attach", newvdpath,
4062	    replacing ? "for" : "to", oldvdpath);
4063
4064	spa_strfree(oldvdpath);
4065	spa_strfree(newvdpath);
4066
4067	if (spa->spa_bootfs)
4068		spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4069
4070	return (0);
4071}
4072
4073/*
4074 * Detach a device from a mirror or replacing vdev.
4075 * If 'replace_done' is specified, only detach if the parent
4076 * is a replacing vdev.
4077 */
4078int
4079spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4080{
4081	uint64_t txg;
4082	int error;
4083	vdev_t *rvd = spa->spa_root_vdev;
4084	vdev_t *vd, *pvd, *cvd, *tvd;
4085	boolean_t unspare = B_FALSE;
4086	uint64_t unspare_guid;
4087	char *vdpath;
4088
4089	ASSERT(spa_writeable(spa));
4090
4091	txg = spa_vdev_enter(spa);
4092
4093	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4094
4095	if (vd == NULL)
4096		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4097
4098	if (!vd->vdev_ops->vdev_op_leaf)
4099		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4100
4101	pvd = vd->vdev_parent;
4102
4103	/*
4104	 * If the parent/child relationship is not as expected, don't do it.
4105	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4106	 * vdev that's replacing B with C.  The user's intent in replacing
4107	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4108	 * the replace by detaching C, the expected behavior is to end up
4109	 * M(A,B).  But suppose that right after deciding to detach C,
4110	 * the replacement of B completes.  We would have M(A,C), and then
4111	 * ask to detach C, which would leave us with just A -- not what
4112	 * the user wanted.  To prevent this, we make sure that the
4113	 * parent/child relationship hasn't changed -- in this example,
4114	 * that C's parent is still the replacing vdev R.
4115	 */
4116	if (pvd->vdev_guid != pguid && pguid != 0)
4117		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4118
4119	/*
4120	 * Only 'replacing' or 'spare' vdevs can be replaced.
4121	 */
4122	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4123	    pvd->vdev_ops != &vdev_spare_ops)
4124		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4125
4126	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4127	    spa_version(spa) >= SPA_VERSION_SPARES);
4128
4129	/*
4130	 * Only mirror, replacing, and spare vdevs support detach.
4131	 */
4132	if (pvd->vdev_ops != &vdev_replacing_ops &&
4133	    pvd->vdev_ops != &vdev_mirror_ops &&
4134	    pvd->vdev_ops != &vdev_spare_ops)
4135		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4136
4137	/*
4138	 * If this device has the only valid copy of some data,
4139	 * we cannot safely detach it.
4140	 */
4141	if (vdev_dtl_required(vd))
4142		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4143
4144	ASSERT(pvd->vdev_children >= 2);
4145
4146	/*
4147	 * If we are detaching the second disk from a replacing vdev, then
4148	 * check to see if we changed the original vdev's path to have "/old"
4149	 * at the end in spa_vdev_attach().  If so, undo that change now.
4150	 */
4151	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4152	    vd->vdev_path != NULL) {
4153		size_t len = strlen(vd->vdev_path);
4154
4155		for (int c = 0; c < pvd->vdev_children; c++) {
4156			cvd = pvd->vdev_child[c];
4157
4158			if (cvd == vd || cvd->vdev_path == NULL)
4159				continue;
4160
4161			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4162			    strcmp(cvd->vdev_path + len, "/old") == 0) {
4163				spa_strfree(cvd->vdev_path);
4164				cvd->vdev_path = spa_strdup(vd->vdev_path);
4165				break;
4166			}
4167		}
4168	}
4169
4170	/*
4171	 * If we are detaching the original disk from a spare, then it implies
4172	 * that the spare should become a real disk, and be removed from the
4173	 * active spare list for the pool.
4174	 */
4175	if (pvd->vdev_ops == &vdev_spare_ops &&
4176	    vd->vdev_id == 0 &&
4177	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4178		unspare = B_TRUE;
4179
4180	/*
4181	 * Erase the disk labels so the disk can be used for other things.
4182	 * This must be done after all other error cases are handled,
4183	 * but before we disembowel vd (so we can still do I/O to it).
4184	 * But if we can't do it, don't treat the error as fatal --
4185	 * it may be that the unwritability of the disk is the reason
4186	 * it's being detached!
4187	 */
4188	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4189
4190	/*
4191	 * Remove vd from its parent and compact the parent's children.
4192	 */
4193	vdev_remove_child(pvd, vd);
4194	vdev_compact_children(pvd);
4195
4196	/*
4197	 * Remember one of the remaining children so we can get tvd below.
4198	 */
4199	cvd = pvd->vdev_child[pvd->vdev_children - 1];
4200
4201	/*
4202	 * If we need to remove the remaining child from the list of hot spares,
4203	 * do it now, marking the vdev as no longer a spare in the process.
4204	 * We must do this before vdev_remove_parent(), because that can
4205	 * change the GUID if it creates a new toplevel GUID.  For a similar
4206	 * reason, we must remove the spare now, in the same txg as the detach;
4207	 * otherwise someone could attach a new sibling, change the GUID, and
4208	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4209	 */
4210	if (unspare) {
4211		ASSERT(cvd->vdev_isspare);
4212		spa_spare_remove(cvd);
4213		unspare_guid = cvd->vdev_guid;
4214		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4215		cvd->vdev_unspare = B_TRUE;
4216	}
4217
4218	/*
4219	 * If the parent mirror/replacing vdev only has one child,
4220	 * the parent is no longer needed.  Remove it from the tree.
4221	 */
4222	if (pvd->vdev_children == 1) {
4223		if (pvd->vdev_ops == &vdev_spare_ops)
4224			cvd->vdev_unspare = B_FALSE;
4225		vdev_remove_parent(cvd);
4226		cvd->vdev_resilvering = B_FALSE;
4227	}
4228
4229
4230	/*
4231	 * We don't set tvd until now because the parent we just removed
4232	 * may have been the previous top-level vdev.
4233	 */
4234	tvd = cvd->vdev_top;
4235	ASSERT(tvd->vdev_parent == rvd);
4236
4237	/*
4238	 * Reevaluate the parent vdev state.
4239	 */
4240	vdev_propagate_state(cvd);
4241
4242	/*
4243	 * If the 'autoexpand' property is set on the pool then automatically
4244	 * try to expand the size of the pool. For example if the device we
4245	 * just detached was smaller than the others, it may be possible to
4246	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4247	 * first so that we can obtain the updated sizes of the leaf vdevs.
4248	 */
4249	if (spa->spa_autoexpand) {
4250		vdev_reopen(tvd);
4251		vdev_expand(tvd, txg);
4252	}
4253
4254	vdev_config_dirty(tvd);
4255
4256	/*
4257	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4258	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4259	 * But first make sure we're not on any *other* txg's DTL list, to
4260	 * prevent vd from being accessed after it's freed.
4261	 */
4262	vdpath = spa_strdup(vd->vdev_path);
4263	for (int t = 0; t < TXG_SIZE; t++)
4264		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4265	vd->vdev_detached = B_TRUE;
4266	vdev_dirty(tvd, VDD_DTL, vd, txg);
4267
4268	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4269
4270	/* hang on to the spa before we release the lock */
4271	spa_open_ref(spa, FTAG);
4272
4273	error = spa_vdev_exit(spa, vd, txg, 0);
4274
4275	spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4276	    "vdev=%s", vdpath);
4277	spa_strfree(vdpath);
4278
4279	/*
4280	 * If this was the removal of the original device in a hot spare vdev,
4281	 * then we want to go through and remove the device from the hot spare
4282	 * list of every other pool.
4283	 */
4284	if (unspare) {
4285		spa_t *altspa = NULL;
4286
4287		mutex_enter(&spa_namespace_lock);
4288		while ((altspa = spa_next(altspa)) != NULL) {
4289			if (altspa->spa_state != POOL_STATE_ACTIVE ||
4290			    altspa == spa)
4291				continue;
4292
4293			spa_open_ref(altspa, FTAG);
4294			mutex_exit(&spa_namespace_lock);
4295			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4296			mutex_enter(&spa_namespace_lock);
4297			spa_close(altspa, FTAG);
4298		}
4299		mutex_exit(&spa_namespace_lock);
4300
4301		/* search the rest of the vdevs for spares to remove */
4302		spa_vdev_resilver_done(spa);
4303	}
4304
4305	/* all done with the spa; OK to release */
4306	mutex_enter(&spa_namespace_lock);
4307	spa_close(spa, FTAG);
4308	mutex_exit(&spa_namespace_lock);
4309
4310	return (error);
4311}
4312
4313/*
4314 * Split a set of devices from their mirrors, and create a new pool from them.
4315 */
4316int
4317spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4318    nvlist_t *props, boolean_t exp)
4319{
4320	int error = 0;
4321	uint64_t txg, *glist;
4322	spa_t *newspa;
4323	uint_t c, children, lastlog;
4324	nvlist_t **child, *nvl, *tmp;
4325	dmu_tx_t *tx;
4326	char *altroot = NULL;
4327	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4328	boolean_t activate_slog;
4329
4330	ASSERT(spa_writeable(spa));
4331
4332	txg = spa_vdev_enter(spa);
4333
4334	/* clear the log and flush everything up to now */
4335	activate_slog = spa_passivate_log(spa);
4336	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4337	error = spa_offline_log(spa);
4338	txg = spa_vdev_config_enter(spa);
4339
4340	if (activate_slog)
4341		spa_activate_log(spa);
4342
4343	if (error != 0)
4344		return (spa_vdev_exit(spa, NULL, txg, error));
4345
4346	/* check new spa name before going any further */
4347	if (spa_lookup(newname) != NULL)
4348		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4349
4350	/*
4351	 * scan through all the children to ensure they're all mirrors
4352	 */
4353	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4354	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4355	    &children) != 0)
4356		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4357
4358	/* first, check to ensure we've got the right child count */
4359	rvd = spa->spa_root_vdev;
4360	lastlog = 0;
4361	for (c = 0; c < rvd->vdev_children; c++) {
4362		vdev_t *vd = rvd->vdev_child[c];
4363
4364		/* don't count the holes & logs as children */
4365		if (vd->vdev_islog || vd->vdev_ishole) {
4366			if (lastlog == 0)
4367				lastlog = c;
4368			continue;
4369		}
4370
4371		lastlog = 0;
4372	}
4373	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4374		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4375
4376	/* next, ensure no spare or cache devices are part of the split */
4377	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4378	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4379		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4380
4381	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4382	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4383
4384	/* then, loop over each vdev and validate it */
4385	for (c = 0; c < children; c++) {
4386		uint64_t is_hole = 0;
4387
4388		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4389		    &is_hole);
4390
4391		if (is_hole != 0) {
4392			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4393			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4394				continue;
4395			} else {
4396				error = EINVAL;
4397				break;
4398			}
4399		}
4400
4401		/* which disk is going to be split? */
4402		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4403		    &glist[c]) != 0) {
4404			error = EINVAL;
4405			break;
4406		}
4407
4408		/* look it up in the spa */
4409		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4410		if (vml[c] == NULL) {
4411			error = ENODEV;
4412			break;
4413		}
4414
4415		/* make sure there's nothing stopping the split */
4416		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4417		    vml[c]->vdev_islog ||
4418		    vml[c]->vdev_ishole ||
4419		    vml[c]->vdev_isspare ||
4420		    vml[c]->vdev_isl2cache ||
4421		    !vdev_writeable(vml[c]) ||
4422		    vml[c]->vdev_children != 0 ||
4423		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4424		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4425			error = EINVAL;
4426			break;
4427		}
4428
4429		if (vdev_dtl_required(vml[c])) {
4430			error = EBUSY;
4431			break;
4432		}
4433
4434		/* we need certain info from the top level */
4435		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4436		    vml[c]->vdev_top->vdev_ms_array) == 0);
4437		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4438		    vml[c]->vdev_top->vdev_ms_shift) == 0);
4439		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4440		    vml[c]->vdev_top->vdev_asize) == 0);
4441		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4442		    vml[c]->vdev_top->vdev_ashift) == 0);
4443	}
4444
4445	if (error != 0) {
4446		kmem_free(vml, children * sizeof (vdev_t *));
4447		kmem_free(glist, children * sizeof (uint64_t));
4448		return (spa_vdev_exit(spa, NULL, txg, error));
4449	}
4450
4451	/* stop writers from using the disks */
4452	for (c = 0; c < children; c++) {
4453		if (vml[c] != NULL)
4454			vml[c]->vdev_offline = B_TRUE;
4455	}
4456	vdev_reopen(spa->spa_root_vdev);
4457
4458	/*
4459	 * Temporarily record the splitting vdevs in the spa config.  This
4460	 * will disappear once the config is regenerated.
4461	 */
4462	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4463	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4464	    glist, children) == 0);
4465	kmem_free(glist, children * sizeof (uint64_t));
4466
4467	mutex_enter(&spa->spa_props_lock);
4468	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4469	    nvl) == 0);
4470	mutex_exit(&spa->spa_props_lock);
4471	spa->spa_config_splitting = nvl;
4472	vdev_config_dirty(spa->spa_root_vdev);
4473
4474	/* configure and create the new pool */
4475	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4476	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4477	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4478	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4479	    spa_version(spa)) == 0);
4480	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4481	    spa->spa_config_txg) == 0);
4482	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4483	    spa_generate_guid(NULL)) == 0);
4484	(void) nvlist_lookup_string(props,
4485	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4486
4487	/* add the new pool to the namespace */
4488	newspa = spa_add(newname, config, altroot);
4489	newspa->spa_config_txg = spa->spa_config_txg;
4490	spa_set_log_state(newspa, SPA_LOG_CLEAR);
4491
4492	/* release the spa config lock, retaining the namespace lock */
4493	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4494
4495	if (zio_injection_enabled)
4496		zio_handle_panic_injection(spa, FTAG, 1);
4497
4498	spa_activate(newspa, spa_mode_global);
4499	spa_async_suspend(newspa);
4500
4501	/* create the new pool from the disks of the original pool */
4502	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4503	if (error)
4504		goto out;
4505
4506	/* if that worked, generate a real config for the new pool */
4507	if (newspa->spa_root_vdev != NULL) {
4508		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4509		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4510		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4511		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4512		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4513		    B_TRUE));
4514	}
4515
4516	/* set the props */
4517	if (props != NULL) {
4518		spa_configfile_set(newspa, props, B_FALSE);
4519		error = spa_prop_set(newspa, props);
4520		if (error)
4521			goto out;
4522	}
4523
4524	/* flush everything */
4525	txg = spa_vdev_config_enter(newspa);
4526	vdev_config_dirty(newspa->spa_root_vdev);
4527	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4528
4529	if (zio_injection_enabled)
4530		zio_handle_panic_injection(spa, FTAG, 2);
4531
4532	spa_async_resume(newspa);
4533
4534	/* finally, update the original pool's config */
4535	txg = spa_vdev_config_enter(spa);
4536	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4537	error = dmu_tx_assign(tx, TXG_WAIT);
4538	if (error != 0)
4539		dmu_tx_abort(tx);
4540	for (c = 0; c < children; c++) {
4541		if (vml[c] != NULL) {
4542			vdev_split(vml[c]);
4543			if (error == 0)
4544				spa_history_log_internal(LOG_POOL_VDEV_DETACH,
4545				    spa, tx, "vdev=%s",
4546				    vml[c]->vdev_path);
4547			vdev_free(vml[c]);
4548		}
4549	}
4550	vdev_config_dirty(spa->spa_root_vdev);
4551	spa->spa_config_splitting = NULL;
4552	nvlist_free(nvl);
4553	if (error == 0)
4554		dmu_tx_commit(tx);
4555	(void) spa_vdev_exit(spa, NULL, txg, 0);
4556
4557	if (zio_injection_enabled)
4558		zio_handle_panic_injection(spa, FTAG, 3);
4559
4560	/* split is complete; log a history record */
4561	spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
4562	    "split new pool %s from pool %s", newname, spa_name(spa));
4563
4564	kmem_free(vml, children * sizeof (vdev_t *));
4565
4566	/* if we're not going to mount the filesystems in userland, export */
4567	if (exp)
4568		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4569		    B_FALSE, B_FALSE);
4570
4571	return (error);
4572
4573out:
4574	spa_unload(newspa);
4575	spa_deactivate(newspa);
4576	spa_remove(newspa);
4577
4578	txg = spa_vdev_config_enter(spa);
4579
4580	/* re-online all offlined disks */
4581	for (c = 0; c < children; c++) {
4582		if (vml[c] != NULL)
4583			vml[c]->vdev_offline = B_FALSE;
4584	}
4585	vdev_reopen(spa->spa_root_vdev);
4586
4587	nvlist_free(spa->spa_config_splitting);
4588	spa->spa_config_splitting = NULL;
4589	(void) spa_vdev_exit(spa, NULL, txg, error);
4590
4591	kmem_free(vml, children * sizeof (vdev_t *));
4592	return (error);
4593}
4594
4595static nvlist_t *
4596spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4597{
4598	for (int i = 0; i < count; i++) {
4599		uint64_t guid;
4600
4601		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4602		    &guid) == 0);
4603
4604		if (guid == target_guid)
4605			return (nvpp[i]);
4606	}
4607
4608	return (NULL);
4609}
4610
4611static void
4612spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4613	nvlist_t *dev_to_remove)
4614{
4615	nvlist_t **newdev = NULL;
4616
4617	if (count > 1)
4618		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4619
4620	for (int i = 0, j = 0; i < count; i++) {
4621		if (dev[i] == dev_to_remove)
4622			continue;
4623		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4624	}
4625
4626	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4627	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4628
4629	for (int i = 0; i < count - 1; i++)
4630		nvlist_free(newdev[i]);
4631
4632	if (count > 1)
4633		kmem_free(newdev, (count - 1) * sizeof (void *));
4634}
4635
4636/*
4637 * Evacuate the device.
4638 */
4639static int
4640spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4641{
4642	uint64_t txg;
4643	int error = 0;
4644
4645	ASSERT(MUTEX_HELD(&spa_namespace_lock));
4646	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4647	ASSERT(vd == vd->vdev_top);
4648
4649	/*
4650	 * Evacuate the device.  We don't hold the config lock as writer
4651	 * since we need to do I/O but we do keep the
4652	 * spa_namespace_lock held.  Once this completes the device
4653	 * should no longer have any blocks allocated on it.
4654	 */
4655	if (vd->vdev_islog) {
4656		if (vd->vdev_stat.vs_alloc != 0)
4657			error = spa_offline_log(spa);
4658	} else {
4659		error = ENOTSUP;
4660	}
4661
4662	if (error)
4663		return (error);
4664
4665	/*
4666	 * The evacuation succeeded.  Remove any remaining MOS metadata
4667	 * associated with this vdev, and wait for these changes to sync.
4668	 */
4669	ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
4670	txg = spa_vdev_config_enter(spa);
4671	vd->vdev_removing = B_TRUE;
4672	vdev_dirty(vd, 0, NULL, txg);
4673	vdev_config_dirty(vd);
4674	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4675
4676	return (0);
4677}
4678
4679/*
4680 * Complete the removal by cleaning up the namespace.
4681 */
4682static void
4683spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
4684{
4685	vdev_t *rvd = spa->spa_root_vdev;
4686	uint64_t id = vd->vdev_id;
4687	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
4688
4689	ASSERT(MUTEX_HELD(&spa_namespace_lock));
4690	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4691	ASSERT(vd == vd->vdev_top);
4692
4693	/*
4694	 * Only remove any devices which are empty.
4695	 */
4696	if (vd->vdev_stat.vs_alloc != 0)
4697		return;
4698
4699	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4700
4701	if (list_link_active(&vd->vdev_state_dirty_node))
4702		vdev_state_clean(vd);
4703	if (list_link_active(&vd->vdev_config_dirty_node))
4704		vdev_config_clean(vd);
4705
4706	vdev_free(vd);
4707
4708	if (last_vdev) {
4709		vdev_compact_children(rvd);
4710	} else {
4711		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
4712		vdev_add_child(rvd, vd);
4713	}
4714	vdev_config_dirty(rvd);
4715
4716	/*
4717	 * Reassess the health of our root vdev.
4718	 */
4719	vdev_reopen(rvd);
4720}
4721
4722/*
4723 * Remove a device from the pool -
4724 *
4725 * Removing a device from the vdev namespace requires several steps
4726 * and can take a significant amount of time.  As a result we use
4727 * the spa_vdev_config_[enter/exit] functions which allow us to
4728 * grab and release the spa_config_lock while still holding the namespace
4729 * lock.  During each step the configuration is synced out.
4730 */
4731
4732/*
4733 * Remove a device from the pool.  Currently, this supports removing only hot
4734 * spares, slogs, and level 2 ARC devices.
4735 */
4736int
4737spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
4738{
4739	vdev_t *vd;
4740	metaslab_group_t *mg;
4741	nvlist_t **spares, **l2cache, *nv;
4742	uint64_t txg = 0;
4743	uint_t nspares, nl2cache;
4744	int error = 0;
4745	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
4746
4747	ASSERT(spa_writeable(spa));
4748
4749	if (!locked)
4750		txg = spa_vdev_enter(spa);
4751
4752	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4753
4754	if (spa->spa_spares.sav_vdevs != NULL &&
4755	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4756	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
4757	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
4758		/*
4759		 * Only remove the hot spare if it's not currently in use
4760		 * in this pool.
4761		 */
4762		if (vd == NULL || unspare) {
4763			spa_vdev_remove_aux(spa->spa_spares.sav_config,
4764			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
4765			spa_load_spares(spa);
4766			spa->spa_spares.sav_sync = B_TRUE;
4767		} else {
4768			error = EBUSY;
4769		}
4770	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
4771	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4772	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
4773	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
4774		/*
4775		 * Cache devices can always be removed.
4776		 */
4777		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
4778		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
4779		spa_load_l2cache(spa);
4780		spa->spa_l2cache.sav_sync = B_TRUE;
4781	} else if (vd != NULL && vd->vdev_islog) {
4782		ASSERT(!locked);
4783		ASSERT(vd == vd->vdev_top);
4784
4785		/*
4786		 * XXX - Once we have bp-rewrite this should
4787		 * become the common case.
4788		 */
4789
4790		mg = vd->vdev_mg;
4791
4792		/*
4793		 * Stop allocating from this vdev.
4794		 */
4795		metaslab_group_passivate(mg);
4796
4797		/*
4798		 * Wait for the youngest allocations and frees to sync,
4799		 * and then wait for the deferral of those frees to finish.
4800		 */
4801		spa_vdev_config_exit(spa, NULL,
4802		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
4803
4804		/*
4805		 * Attempt to evacuate the vdev.
4806		 */
4807		error = spa_vdev_remove_evacuate(spa, vd);
4808
4809		txg = spa_vdev_config_enter(spa);
4810
4811		/*
4812		 * If we couldn't evacuate the vdev, unwind.
4813		 */
4814		if (error) {
4815			metaslab_group_activate(mg);
4816			return (spa_vdev_exit(spa, NULL, txg, error));
4817		}
4818
4819		/*
4820		 * Clean up the vdev namespace.
4821		 */
4822		spa_vdev_remove_from_namespace(spa, vd);
4823
4824	} else if (vd != NULL) {
4825		/*
4826		 * Normal vdevs cannot be removed (yet).
4827		 */
4828		error = ENOTSUP;
4829	} else {
4830		/*
4831		 * There is no vdev of any kind with the specified guid.
4832		 */
4833		error = ENOENT;
4834	}
4835
4836	if (!locked)
4837		return (spa_vdev_exit(spa, NULL, txg, error));
4838
4839	return (error);
4840}
4841
4842/*
4843 * Find any device that's done replacing, or a vdev marked 'unspare' that's
4844 * current spared, so we can detach it.
4845 */
4846static vdev_t *
4847spa_vdev_resilver_done_hunt(vdev_t *vd)
4848{
4849	vdev_t *newvd, *oldvd;
4850
4851	for (int c = 0; c < vd->vdev_children; c++) {
4852		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
4853		if (oldvd != NULL)
4854			return (oldvd);
4855	}
4856
4857	/*
4858	 * Check for a completed replacement.  We always consider the first
4859	 * vdev in the list to be the oldest vdev, and the last one to be
4860	 * the newest (see spa_vdev_attach() for how that works).  In
4861	 * the case where the newest vdev is faulted, we will not automatically
4862	 * remove it after a resilver completes.  This is OK as it will require
4863	 * user intervention to determine which disk the admin wishes to keep.
4864	 */
4865	if (vd->vdev_ops == &vdev_replacing_ops) {
4866		ASSERT(vd->vdev_children > 1);
4867
4868		newvd = vd->vdev_child[vd->vdev_children - 1];
4869		oldvd = vd->vdev_child[0];
4870
4871		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4872		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4873		    !vdev_dtl_required(oldvd))
4874			return (oldvd);
4875	}
4876
4877	/*
4878	 * Check for a completed resilver with the 'unspare' flag set.
4879	 */
4880	if (vd->vdev_ops == &vdev_spare_ops) {
4881		vdev_t *first = vd->vdev_child[0];
4882		vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
4883
4884		if (last->vdev_unspare) {
4885			oldvd = first;
4886			newvd = last;
4887		} else if (first->vdev_unspare) {
4888			oldvd = last;
4889			newvd = first;
4890		} else {
4891			oldvd = NULL;
4892		}
4893
4894		if (oldvd != NULL &&
4895		    vdev_dtl_empty(newvd, DTL_MISSING) &&
4896		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4897		    !vdev_dtl_required(oldvd))
4898			return (oldvd);
4899
4900		/*
4901		 * If there are more than two spares attached to a disk,
4902		 * and those spares are not required, then we want to
4903		 * attempt to free them up now so that they can be used
4904		 * by other pools.  Once we're back down to a single
4905		 * disk+spare, we stop removing them.
4906		 */
4907		if (vd->vdev_children > 2) {
4908			newvd = vd->vdev_child[1];
4909
4910			if (newvd->vdev_isspare && last->vdev_isspare &&
4911			    vdev_dtl_empty(last, DTL_MISSING) &&
4912			    vdev_dtl_empty(last, DTL_OUTAGE) &&
4913			    !vdev_dtl_required(newvd))
4914				return (newvd);
4915		}
4916	}
4917
4918	return (NULL);
4919}
4920
4921static void
4922spa_vdev_resilver_done(spa_t *spa)
4923{
4924	vdev_t *vd, *pvd, *ppvd;
4925	uint64_t guid, sguid, pguid, ppguid;
4926
4927	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4928
4929	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4930		pvd = vd->vdev_parent;
4931		ppvd = pvd->vdev_parent;
4932		guid = vd->vdev_guid;
4933		pguid = pvd->vdev_guid;
4934		ppguid = ppvd->vdev_guid;
4935		sguid = 0;
4936		/*
4937		 * If we have just finished replacing a hot spared device, then
4938		 * we need to detach the parent's first child (the original hot
4939		 * spare) as well.
4940		 */
4941		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
4942		    ppvd->vdev_children == 2) {
4943			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4944			sguid = ppvd->vdev_child[1]->vdev_guid;
4945		}
4946		spa_config_exit(spa, SCL_ALL, FTAG);
4947		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4948			return;
4949		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4950			return;
4951		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4952	}
4953
4954	spa_config_exit(spa, SCL_ALL, FTAG);
4955}
4956
4957/*
4958 * Update the stored path or FRU for this vdev.
4959 */
4960int
4961spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4962    boolean_t ispath)
4963{
4964	vdev_t *vd;
4965	boolean_t sync = B_FALSE;
4966
4967	ASSERT(spa_writeable(spa));
4968
4969	spa_vdev_state_enter(spa, SCL_ALL);
4970
4971	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4972		return (spa_vdev_state_exit(spa, NULL, ENOENT));
4973
4974	if (!vd->vdev_ops->vdev_op_leaf)
4975		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
4976
4977	if (ispath) {
4978		if (strcmp(value, vd->vdev_path) != 0) {
4979			spa_strfree(vd->vdev_path);
4980			vd->vdev_path = spa_strdup(value);
4981			sync = B_TRUE;
4982		}
4983	} else {
4984		if (vd->vdev_fru == NULL) {
4985			vd->vdev_fru = spa_strdup(value);
4986			sync = B_TRUE;
4987		} else if (strcmp(value, vd->vdev_fru) != 0) {
4988			spa_strfree(vd->vdev_fru);
4989			vd->vdev_fru = spa_strdup(value);
4990			sync = B_TRUE;
4991		}
4992	}
4993
4994	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
4995}
4996
4997int
4998spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
4999{
5000	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5001}
5002
5003int
5004spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5005{
5006	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5007}
5008
5009/*
5010 * ==========================================================================
5011 * SPA Scanning
5012 * ==========================================================================
5013 */
5014
5015int
5016spa_scan_stop(spa_t *spa)
5017{
5018	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5019	if (dsl_scan_resilvering(spa->spa_dsl_pool))
5020		return (EBUSY);
5021	return (dsl_scan_cancel(spa->spa_dsl_pool));
5022}
5023
5024int
5025spa_scan(spa_t *spa, pool_scan_func_t func)
5026{
5027	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5028
5029	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5030		return (ENOTSUP);
5031
5032	/*
5033	 * If a resilver was requested, but there is no DTL on a
5034	 * writeable leaf device, we have nothing to do.
5035	 */
5036	if (func == POOL_SCAN_RESILVER &&
5037	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5038		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5039		return (0);
5040	}
5041
5042	return (dsl_scan(spa->spa_dsl_pool, func));
5043}
5044
5045/*
5046 * ==========================================================================
5047 * SPA async task processing
5048 * ==========================================================================
5049 */
5050
5051static void
5052spa_async_remove(spa_t *spa, vdev_t *vd)
5053{
5054	if (vd->vdev_remove_wanted) {
5055		vd->vdev_remove_wanted = B_FALSE;
5056		vd->vdev_delayed_close = B_FALSE;
5057		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5058
5059		/*
5060		 * We want to clear the stats, but we don't want to do a full
5061		 * vdev_clear() as that will cause us to throw away
5062		 * degraded/faulted state as well as attempt to reopen the
5063		 * device, all of which is a waste.
5064		 */
5065		vd->vdev_stat.vs_read_errors = 0;
5066		vd->vdev_stat.vs_write_errors = 0;
5067		vd->vdev_stat.vs_checksum_errors = 0;
5068
5069		vdev_state_dirty(vd->vdev_top);
5070	}
5071
5072	for (int c = 0; c < vd->vdev_children; c++)
5073		spa_async_remove(spa, vd->vdev_child[c]);
5074}
5075
5076static void
5077spa_async_probe(spa_t *spa, vdev_t *vd)
5078{
5079	if (vd->vdev_probe_wanted) {
5080		vd->vdev_probe_wanted = B_FALSE;
5081		vdev_reopen(vd);	/* vdev_open() does the actual probe */
5082	}
5083
5084	for (int c = 0; c < vd->vdev_children; c++)
5085		spa_async_probe(spa, vd->vdev_child[c]);
5086}
5087
5088static void
5089spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5090{
5091	sysevent_id_t eid;
5092	nvlist_t *attr;
5093	char *physpath;
5094
5095	if (!spa->spa_autoexpand)
5096		return;
5097
5098	for (int c = 0; c < vd->vdev_children; c++) {
5099		vdev_t *cvd = vd->vdev_child[c];
5100		spa_async_autoexpand(spa, cvd);
5101	}
5102
5103	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5104		return;
5105
5106	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5107	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5108
5109	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5110	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5111
5112	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5113	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5114
5115	nvlist_free(attr);
5116	kmem_free(physpath, MAXPATHLEN);
5117}
5118
5119static void
5120spa_async_thread(spa_t *spa)
5121{
5122	int tasks;
5123
5124	ASSERT(spa->spa_sync_on);
5125
5126	mutex_enter(&spa->spa_async_lock);
5127	tasks = spa->spa_async_tasks;
5128	spa->spa_async_tasks = 0;
5129	mutex_exit(&spa->spa_async_lock);
5130
5131	/*
5132	 * See if the config needs to be updated.
5133	 */
5134	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5135		uint64_t old_space, new_space;
5136
5137		mutex_enter(&spa_namespace_lock);
5138		old_space = metaslab_class_get_space(spa_normal_class(spa));
5139		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5140		new_space = metaslab_class_get_space(spa_normal_class(spa));
5141		mutex_exit(&spa_namespace_lock);
5142
5143		/*
5144		 * If the pool grew as a result of the config update,
5145		 * then log an internal history event.
5146		 */
5147		if (new_space != old_space) {
5148			spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5149			    spa, NULL,
5150			    "pool '%s' size: %llu(+%llu)",
5151			    spa_name(spa), new_space, new_space - old_space);
5152		}
5153	}
5154
5155	/*
5156	 * See if any devices need to be marked REMOVED.
5157	 */
5158	if (tasks & SPA_ASYNC_REMOVE) {
5159		spa_vdev_state_enter(spa, SCL_NONE);
5160		spa_async_remove(spa, spa->spa_root_vdev);
5161		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5162			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5163		for (int i = 0; i < spa->spa_spares.sav_count; i++)
5164			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5165		(void) spa_vdev_state_exit(spa, NULL, 0);
5166	}
5167
5168	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5169		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5170		spa_async_autoexpand(spa, spa->spa_root_vdev);
5171		spa_config_exit(spa, SCL_CONFIG, FTAG);
5172	}
5173
5174	/*
5175	 * See if any devices need to be probed.
5176	 */
5177	if (tasks & SPA_ASYNC_PROBE) {
5178		spa_vdev_state_enter(spa, SCL_NONE);
5179		spa_async_probe(spa, spa->spa_root_vdev);
5180		(void) spa_vdev_state_exit(spa, NULL, 0);
5181	}
5182
5183	/*
5184	 * If any devices are done replacing, detach them.
5185	 */
5186	if (tasks & SPA_ASYNC_RESILVER_DONE)
5187		spa_vdev_resilver_done(spa);
5188
5189	/*
5190	 * Kick off a resilver.
5191	 */
5192	if (tasks & SPA_ASYNC_RESILVER)
5193		dsl_resilver_restart(spa->spa_dsl_pool, 0);
5194
5195	/*
5196	 * Let the world know that we're done.
5197	 */
5198	mutex_enter(&spa->spa_async_lock);
5199	spa->spa_async_thread = NULL;
5200	cv_broadcast(&spa->spa_async_cv);
5201	mutex_exit(&spa->spa_async_lock);
5202	thread_exit();
5203}
5204
5205void
5206spa_async_suspend(spa_t *spa)
5207{
5208	mutex_enter(&spa->spa_async_lock);
5209	spa->spa_async_suspended++;
5210	while (spa->spa_async_thread != NULL)
5211		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5212	mutex_exit(&spa->spa_async_lock);
5213}
5214
5215void
5216spa_async_resume(spa_t *spa)
5217{
5218	mutex_enter(&spa->spa_async_lock);
5219	ASSERT(spa->spa_async_suspended != 0);
5220	spa->spa_async_suspended--;
5221	mutex_exit(&spa->spa_async_lock);
5222}
5223
5224static void
5225spa_async_dispatch(spa_t *spa)
5226{
5227	mutex_enter(&spa->spa_async_lock);
5228	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5229	    spa->spa_async_thread == NULL &&
5230	    rootdir != NULL && !vn_is_readonly(rootdir))
5231		spa->spa_async_thread = thread_create(NULL, 0,
5232		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5233	mutex_exit(&spa->spa_async_lock);
5234}
5235
5236void
5237spa_async_request(spa_t *spa, int task)
5238{
5239	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5240	mutex_enter(&spa->spa_async_lock);
5241	spa->spa_async_tasks |= task;
5242	mutex_exit(&spa->spa_async_lock);
5243}
5244
5245/*
5246 * ==========================================================================
5247 * SPA syncing routines
5248 * ==========================================================================
5249 */
5250
5251static int
5252bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5253{
5254	bpobj_t *bpo = arg;
5255	bpobj_enqueue(bpo, bp, tx);
5256	return (0);
5257}
5258
5259static int
5260spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5261{
5262	zio_t *zio = arg;
5263
5264	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5265	    zio->io_flags));
5266	return (0);
5267}
5268
5269static void
5270spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5271{
5272	char *packed = NULL;
5273	size_t bufsize;
5274	size_t nvsize = 0;
5275	dmu_buf_t *db;
5276
5277	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5278
5279	/*
5280	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5281	 * information.  This avoids the dbuf_will_dirty() path and
5282	 * saves us a pre-read to get data we don't actually care about.
5283	 */
5284	bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
5285	packed = kmem_alloc(bufsize, KM_SLEEP);
5286
5287	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5288	    KM_SLEEP) == 0);
5289	bzero(packed + nvsize, bufsize - nvsize);
5290
5291	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5292
5293	kmem_free(packed, bufsize);
5294
5295	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5296	dmu_buf_will_dirty(db, tx);
5297	*(uint64_t *)db->db_data = nvsize;
5298	dmu_buf_rele(db, FTAG);
5299}
5300
5301static void
5302spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5303    const char *config, const char *entry)
5304{
5305	nvlist_t *nvroot;
5306	nvlist_t **list;
5307	int i;
5308
5309	if (!sav->sav_sync)
5310		return;
5311
5312	/*
5313	 * Update the MOS nvlist describing the list of available devices.
5314	 * spa_validate_aux() will have already made sure this nvlist is
5315	 * valid and the vdevs are labeled appropriately.
5316	 */
5317	if (sav->sav_object == 0) {
5318		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5319		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5320		    sizeof (uint64_t), tx);
5321		VERIFY(zap_update(spa->spa_meta_objset,
5322		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5323		    &sav->sav_object, tx) == 0);
5324	}
5325
5326	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5327	if (sav->sav_count == 0) {
5328		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5329	} else {
5330		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5331		for (i = 0; i < sav->sav_count; i++)
5332			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5333			    B_FALSE, VDEV_CONFIG_L2CACHE);
5334		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5335		    sav->sav_count) == 0);
5336		for (i = 0; i < sav->sav_count; i++)
5337			nvlist_free(list[i]);
5338		kmem_free(list, sav->sav_count * sizeof (void *));
5339	}
5340
5341	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5342	nvlist_free(nvroot);
5343
5344	sav->sav_sync = B_FALSE;
5345}
5346
5347static void
5348spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5349{
5350	nvlist_t *config;
5351
5352	if (list_is_empty(&spa->spa_config_dirty_list))
5353		return;
5354
5355	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5356
5357	config = spa_config_generate(spa, spa->spa_root_vdev,
5358	    dmu_tx_get_txg(tx), B_FALSE);
5359
5360	spa_config_exit(spa, SCL_STATE, FTAG);
5361
5362	if (spa->spa_config_syncing)
5363		nvlist_free(spa->spa_config_syncing);
5364	spa->spa_config_syncing = config;
5365
5366	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5367}
5368
5369/*
5370 * Set zpool properties.
5371 */
5372static void
5373spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5374{
5375	spa_t *spa = arg1;
5376	objset_t *mos = spa->spa_meta_objset;
5377	nvlist_t *nvp = arg2;
5378	nvpair_t *elem;
5379	uint64_t intval;
5380	char *strval;
5381	zpool_prop_t prop;
5382	const char *propname;
5383	zprop_type_t proptype;
5384
5385	mutex_enter(&spa->spa_props_lock);
5386
5387	elem = NULL;
5388	while ((elem = nvlist_next_nvpair(nvp, elem))) {
5389		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5390		case ZPOOL_PROP_VERSION:
5391			/*
5392			 * Only set version for non-zpool-creation cases
5393			 * (set/import). spa_create() needs special care
5394			 * for version setting.
5395			 */
5396			if (tx->tx_txg != TXG_INITIAL) {
5397				VERIFY(nvpair_value_uint64(elem,
5398				    &intval) == 0);
5399				ASSERT(intval <= SPA_VERSION);
5400				ASSERT(intval >= spa_version(spa));
5401				spa->spa_uberblock.ub_version = intval;
5402				vdev_config_dirty(spa->spa_root_vdev);
5403			}
5404			break;
5405
5406		case ZPOOL_PROP_ALTROOT:
5407			/*
5408			 * 'altroot' is a non-persistent property. It should
5409			 * have been set temporarily at creation or import time.
5410			 */
5411			ASSERT(spa->spa_root != NULL);
5412			break;
5413
5414		case ZPOOL_PROP_READONLY:
5415		case ZPOOL_PROP_CACHEFILE:
5416			/*
5417			 * 'readonly' and 'cachefile' are also non-persisitent
5418			 * properties.
5419			 */
5420			break;
5421		case ZPOOL_PROP_COMMENT:
5422			VERIFY(nvpair_value_string(elem, &strval) == 0);
5423			if (spa->spa_comment != NULL)
5424				spa_strfree(spa->spa_comment);
5425			spa->spa_comment = spa_strdup(strval);
5426			/*
5427			 * We need to dirty the configuration on all the vdevs
5428			 * so that their labels get updated.  It's unnecessary
5429			 * to do this for pool creation since the vdev's
5430			 * configuratoin has already been dirtied.
5431			 */
5432			if (tx->tx_txg != TXG_INITIAL)
5433				vdev_config_dirty(spa->spa_root_vdev);
5434			break;
5435		default:
5436			/*
5437			 * Set pool property values in the poolprops mos object.
5438			 */
5439			if (spa->spa_pool_props_object == 0) {
5440				VERIFY((spa->spa_pool_props_object =
5441				    zap_create(mos, DMU_OT_POOL_PROPS,
5442				    DMU_OT_NONE, 0, tx)) > 0);
5443
5444				VERIFY(zap_update(mos,
5445				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5446				    8, 1, &spa->spa_pool_props_object, tx)
5447				    == 0);
5448			}
5449
5450			/* normalize the property name */
5451			propname = zpool_prop_to_name(prop);
5452			proptype = zpool_prop_get_type(prop);
5453
5454			if (nvpair_type(elem) == DATA_TYPE_STRING) {
5455				ASSERT(proptype == PROP_TYPE_STRING);
5456				VERIFY(nvpair_value_string(elem, &strval) == 0);
5457				VERIFY(zap_update(mos,
5458				    spa->spa_pool_props_object, propname,
5459				    1, strlen(strval) + 1, strval, tx) == 0);
5460
5461			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5462				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5463
5464				if (proptype == PROP_TYPE_INDEX) {
5465					const char *unused;
5466					VERIFY(zpool_prop_index_to_string(
5467					    prop, intval, &unused) == 0);
5468				}
5469				VERIFY(zap_update(mos,
5470				    spa->spa_pool_props_object, propname,
5471				    8, 1, &intval, tx) == 0);
5472			} else {
5473				ASSERT(0); /* not allowed */
5474			}
5475
5476			switch (prop) {
5477			case ZPOOL_PROP_DELEGATION:
5478				spa->spa_delegation = intval;
5479				break;
5480			case ZPOOL_PROP_BOOTFS:
5481				spa->spa_bootfs = intval;
5482				break;
5483			case ZPOOL_PROP_FAILUREMODE:
5484				spa->spa_failmode = intval;
5485				break;
5486			case ZPOOL_PROP_AUTOEXPAND:
5487				spa->spa_autoexpand = intval;
5488				if (tx->tx_txg != TXG_INITIAL)
5489					spa_async_request(spa,
5490					    SPA_ASYNC_AUTOEXPAND);
5491				break;
5492			case ZPOOL_PROP_DEDUPDITTO:
5493				spa->spa_dedup_ditto = intval;
5494				break;
5495			default:
5496				break;
5497			}
5498		}
5499
5500		/* log internal history if this is not a zpool create */
5501		if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5502		    tx->tx_txg != TXG_INITIAL) {
5503			spa_history_log_internal(LOG_POOL_PROPSET,
5504			    spa, tx, "%s %lld %s",
5505			    nvpair_name(elem), intval, spa_name(spa));
5506		}
5507	}
5508
5509	mutex_exit(&spa->spa_props_lock);
5510}
5511
5512/*
5513 * Perform one-time upgrade on-disk changes.  spa_version() does not
5514 * reflect the new version this txg, so there must be no changes this
5515 * txg to anything that the upgrade code depends on after it executes.
5516 * Therefore this must be called after dsl_pool_sync() does the sync
5517 * tasks.
5518 */
5519static void
5520spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5521{
5522	dsl_pool_t *dp = spa->spa_dsl_pool;
5523
5524	ASSERT(spa->spa_sync_pass == 1);
5525
5526	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5527	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5528		dsl_pool_create_origin(dp, tx);
5529
5530		/* Keeping the origin open increases spa_minref */
5531		spa->spa_minref += 3;
5532	}
5533
5534	if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5535	    spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5536		dsl_pool_upgrade_clones(dp, tx);
5537	}
5538
5539	if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5540	    spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5541		dsl_pool_upgrade_dir_clones(dp, tx);
5542
5543		/* Keeping the freedir open increases spa_minref */
5544		spa->spa_minref += 3;
5545	}
5546}
5547
5548/*
5549 * Sync the specified transaction group.  New blocks may be dirtied as
5550 * part of the process, so we iterate until it converges.
5551 */
5552void
5553spa_sync(spa_t *spa, uint64_t txg)
5554{
5555	dsl_pool_t *dp = spa->spa_dsl_pool;
5556	objset_t *mos = spa->spa_meta_objset;
5557	bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5558	bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5559	vdev_t *rvd = spa->spa_root_vdev;
5560	vdev_t *vd;
5561	dmu_tx_t *tx;
5562	int error;
5563
5564	VERIFY(spa_writeable(spa));
5565
5566	/*
5567	 * Lock out configuration changes.
5568	 */
5569	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5570
5571	spa->spa_syncing_txg = txg;
5572	spa->spa_sync_pass = 0;
5573
5574	/*
5575	 * If there are any pending vdev state changes, convert them
5576	 * into config changes that go out with this transaction group.
5577	 */
5578	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5579	while (list_head(&spa->spa_state_dirty_list) != NULL) {
5580		/*
5581		 * We need the write lock here because, for aux vdevs,
5582		 * calling vdev_config_dirty() modifies sav_config.
5583		 * This is ugly and will become unnecessary when we
5584		 * eliminate the aux vdev wart by integrating all vdevs
5585		 * into the root vdev tree.
5586		 */
5587		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5588		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5589		while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5590			vdev_state_clean(vd);
5591			vdev_config_dirty(vd);
5592		}
5593		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5594		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5595	}
5596	spa_config_exit(spa, SCL_STATE, FTAG);
5597
5598	tx = dmu_tx_create_assigned(dp, txg);
5599
5600	/*
5601	 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5602	 * set spa_deflate if we have no raid-z vdevs.
5603	 */
5604	if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5605	    spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5606		int i;
5607
5608		for (i = 0; i < rvd->vdev_children; i++) {
5609			vd = rvd->vdev_child[i];
5610			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5611				break;
5612		}
5613		if (i == rvd->vdev_children) {
5614			spa->spa_deflate = TRUE;
5615			VERIFY(0 == zap_add(spa->spa_meta_objset,
5616			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5617			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5618		}
5619	}
5620
5621	/*
5622	 * If anything has changed in this txg, or if someone is waiting
5623	 * for this txg to sync (eg, spa_vdev_remove()), push the
5624	 * deferred frees from the previous txg.  If not, leave them
5625	 * alone so that we don't generate work on an otherwise idle
5626	 * system.
5627	 */
5628	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5629	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5630	    !txg_list_empty(&dp->dp_sync_tasks, txg) ||
5631	    ((dsl_scan_active(dp->dp_scan) ||
5632	    txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
5633		zio_t *zio = zio_root(spa, NULL, NULL, 0);
5634		VERIFY3U(bpobj_iterate(defer_bpo,
5635		    spa_free_sync_cb, zio, tx), ==, 0);
5636		VERIFY3U(zio_wait(zio), ==, 0);
5637	}
5638
5639	/*
5640	 * Iterate to convergence.
5641	 */
5642	do {
5643		int pass = ++spa->spa_sync_pass;
5644
5645		spa_sync_config_object(spa, tx);
5646		spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5647		    ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5648		spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5649		    ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5650		spa_errlog_sync(spa, txg);
5651		dsl_pool_sync(dp, txg);
5652
5653		if (pass <= SYNC_PASS_DEFERRED_FREE) {
5654			zio_t *zio = zio_root(spa, NULL, NULL, 0);
5655			bplist_iterate(free_bpl, spa_free_sync_cb,
5656			    zio, tx);
5657			VERIFY(zio_wait(zio) == 0);
5658		} else {
5659			bplist_iterate(free_bpl, bpobj_enqueue_cb,
5660			    defer_bpo, tx);
5661		}
5662
5663		ddt_sync(spa, txg);
5664		dsl_scan_sync(dp, tx);
5665
5666		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
5667			vdev_sync(vd, txg);
5668
5669		if (pass == 1)
5670			spa_sync_upgrades(spa, tx);
5671
5672	} while (dmu_objset_is_dirty(mos, txg));
5673
5674	/*
5675	 * Rewrite the vdev configuration (which includes the uberblock)
5676	 * to commit the transaction group.
5677	 *
5678	 * If there are no dirty vdevs, we sync the uberblock to a few
5679	 * random top-level vdevs that are known to be visible in the
5680	 * config cache (see spa_vdev_add() for a complete description).
5681	 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
5682	 */
5683	for (;;) {
5684		/*
5685		 * We hold SCL_STATE to prevent vdev open/close/etc.
5686		 * while we're attempting to write the vdev labels.
5687		 */
5688		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5689
5690		if (list_is_empty(&spa->spa_config_dirty_list)) {
5691			vdev_t *svd[SPA_DVAS_PER_BP];
5692			int svdcount = 0;
5693			int children = rvd->vdev_children;
5694			int c0 = spa_get_random(children);
5695
5696			for (int c = 0; c < children; c++) {
5697				vd = rvd->vdev_child[(c0 + c) % children];
5698				if (vd->vdev_ms_array == 0 || vd->vdev_islog)
5699					continue;
5700				svd[svdcount++] = vd;
5701				if (svdcount == SPA_DVAS_PER_BP)
5702					break;
5703			}
5704			error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
5705			if (error != 0)
5706				error = vdev_config_sync(svd, svdcount, txg,
5707				    B_TRUE);
5708		} else {
5709			error = vdev_config_sync(rvd->vdev_child,
5710			    rvd->vdev_children, txg, B_FALSE);
5711			if (error != 0)
5712				error = vdev_config_sync(rvd->vdev_child,
5713				    rvd->vdev_children, txg, B_TRUE);
5714		}
5715
5716		spa_config_exit(spa, SCL_STATE, FTAG);
5717
5718		if (error == 0)
5719			break;
5720		zio_suspend(spa, NULL);
5721		zio_resume_wait(spa);
5722	}
5723	dmu_tx_commit(tx);
5724
5725	/*
5726	 * Clear the dirty config list.
5727	 */
5728	while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
5729		vdev_config_clean(vd);
5730
5731	/*
5732	 * Now that the new config has synced transactionally,
5733	 * let it become visible to the config cache.
5734	 */
5735	if (spa->spa_config_syncing != NULL) {
5736		spa_config_set(spa, spa->spa_config_syncing);
5737		spa->spa_config_txg = txg;
5738		spa->spa_config_syncing = NULL;
5739	}
5740
5741	spa->spa_ubsync = spa->spa_uberblock;
5742
5743	dsl_pool_sync_done(dp, txg);
5744
5745	/*
5746	 * Update usable space statistics.
5747	 */
5748	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
5749		vdev_sync_done(vd, txg);
5750
5751	spa_update_dspace(spa);
5752
5753	/*
5754	 * It had better be the case that we didn't dirty anything
5755	 * since vdev_config_sync().
5756	 */
5757	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
5758	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
5759	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
5760
5761	spa->spa_sync_pass = 0;
5762
5763	spa_config_exit(spa, SCL_CONFIG, FTAG);
5764
5765	spa_handle_ignored_writes(spa);
5766
5767	/*
5768	 * If any async tasks have been requested, kick them off.
5769	 */
5770	spa_async_dispatch(spa);
5771}
5772
5773/*
5774 * Sync all pools.  We don't want to hold the namespace lock across these
5775 * operations, so we take a reference on the spa_t and drop the lock during the
5776 * sync.
5777 */
5778void
5779spa_sync_allpools(void)
5780{
5781	spa_t *spa = NULL;
5782	mutex_enter(&spa_namespace_lock);
5783	while ((spa = spa_next(spa)) != NULL) {
5784		if (spa_state(spa) != POOL_STATE_ACTIVE ||
5785		    !spa_writeable(spa) || spa_suspended(spa))
5786			continue;
5787		spa_open_ref(spa, FTAG);
5788		mutex_exit(&spa_namespace_lock);
5789		txg_wait_synced(spa_get_dsl(spa), 0);
5790		mutex_enter(&spa_namespace_lock);
5791		spa_close(spa, FTAG);
5792	}
5793	mutex_exit(&spa_namespace_lock);
5794}
5795
5796/*
5797 * ==========================================================================
5798 * Miscellaneous routines
5799 * ==========================================================================
5800 */
5801
5802/*
5803 * Remove all pools in the system.
5804 */
5805void
5806spa_evict_all(void)
5807{
5808	spa_t *spa;
5809
5810	/*
5811	 * Remove all cached state.  All pools should be closed now,
5812	 * so every spa in the AVL tree should be unreferenced.
5813	 */
5814	mutex_enter(&spa_namespace_lock);
5815	while ((spa = spa_next(NULL)) != NULL) {
5816		/*
5817		 * Stop async tasks.  The async thread may need to detach
5818		 * a device that's been replaced, which requires grabbing
5819		 * spa_namespace_lock, so we must drop it here.
5820		 */
5821		spa_open_ref(spa, FTAG);
5822		mutex_exit(&spa_namespace_lock);
5823		spa_async_suspend(spa);
5824		mutex_enter(&spa_namespace_lock);
5825		spa_close(spa, FTAG);
5826
5827		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5828			spa_unload(spa);
5829			spa_deactivate(spa);
5830		}
5831		spa_remove(spa);
5832	}
5833	mutex_exit(&spa_namespace_lock);
5834}
5835
5836vdev_t *
5837spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
5838{
5839	vdev_t *vd;
5840	int i;
5841
5842	if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
5843		return (vd);
5844
5845	if (aux) {
5846		for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
5847			vd = spa->spa_l2cache.sav_vdevs[i];
5848			if (vd->vdev_guid == guid)
5849				return (vd);
5850		}
5851
5852		for (i = 0; i < spa->spa_spares.sav_count; i++) {
5853			vd = spa->spa_spares.sav_vdevs[i];
5854			if (vd->vdev_guid == guid)
5855				return (vd);
5856		}
5857	}
5858
5859	return (NULL);
5860}
5861
5862void
5863spa_upgrade(spa_t *spa, uint64_t version)
5864{
5865	ASSERT(spa_writeable(spa));
5866
5867	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5868
5869	/*
5870	 * This should only be called for a non-faulted pool, and since a
5871	 * future version would result in an unopenable pool, this shouldn't be
5872	 * possible.
5873	 */
5874	ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
5875	ASSERT(version >= spa->spa_uberblock.ub_version);
5876
5877	spa->spa_uberblock.ub_version = version;
5878	vdev_config_dirty(spa->spa_root_vdev);
5879
5880	spa_config_exit(spa, SCL_ALL, FTAG);
5881
5882	txg_wait_synced(spa_get_dsl(spa), 0);
5883}
5884
5885boolean_t
5886spa_has_spare(spa_t *spa, uint64_t guid)
5887{
5888	int i;
5889	uint64_t spareguid;
5890	spa_aux_vdev_t *sav = &spa->spa_spares;
5891
5892	for (i = 0; i < sav->sav_count; i++)
5893		if (sav->sav_vdevs[i]->vdev_guid == guid)
5894			return (B_TRUE);
5895
5896	for (i = 0; i < sav->sav_npending; i++) {
5897		if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
5898		    &spareguid) == 0 && spareguid == guid)
5899			return (B_TRUE);
5900	}
5901
5902	return (B_FALSE);
5903}
5904
5905/*
5906 * Check if a pool has an active shared spare device.
5907 * Note: reference count of an active spare is 2, as a spare and as a replace
5908 */
5909static boolean_t
5910spa_has_active_shared_spare(spa_t *spa)
5911{
5912	int i, refcnt;
5913	uint64_t pool;
5914	spa_aux_vdev_t *sav = &spa->spa_spares;
5915
5916	for (i = 0; i < sav->sav_count; i++) {
5917		if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
5918		    &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
5919		    refcnt > 2)
5920			return (B_TRUE);
5921	}
5922
5923	return (B_FALSE);
5924}
5925
5926/*
5927 * Post a sysevent corresponding to the given event.  The 'name' must be one of
5928 * the event definitions in sys/sysevent/eventdefs.h.  The payload will be
5929 * filled in from the spa and (optionally) the vdev.  This doesn't do anything
5930 * in the userland libzpool, as we don't want consumers to misinterpret ztest
5931 * or zdb as real changes.
5932 */
5933void
5934spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
5935{
5936#ifdef _KERNEL
5937	sysevent_t		*ev;
5938	sysevent_attr_list_t	*attr = NULL;
5939	sysevent_value_t	value;
5940	sysevent_id_t		eid;
5941
5942	ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
5943	    SE_SLEEP);
5944
5945	value.value_type = SE_DATA_TYPE_STRING;
5946	value.value.sv_string = spa_name(spa);
5947	if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
5948		goto done;
5949
5950	value.value_type = SE_DATA_TYPE_UINT64;
5951	value.value.sv_uint64 = spa_guid(spa);
5952	if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
5953		goto done;
5954
5955	if (vd) {
5956		value.value_type = SE_DATA_TYPE_UINT64;
5957		value.value.sv_uint64 = vd->vdev_guid;
5958		if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
5959		    SE_SLEEP) != 0)
5960			goto done;
5961
5962		if (vd->vdev_path) {
5963			value.value_type = SE_DATA_TYPE_STRING;
5964			value.value.sv_string = vd->vdev_path;
5965			if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
5966			    &value, SE_SLEEP) != 0)
5967				goto done;
5968		}
5969	}
5970
5971	if (sysevent_attach_attributes(ev, attr) != 0)
5972		goto done;
5973	attr = NULL;
5974
5975	(void) log_sysevent(ev, SE_SLEEP, &eid);
5976
5977done:
5978	if (attr)
5979		sysevent_free_attr(attr);
5980	sysevent_free(ev);
5981#endif
5982}
5983