spa.c revision 7b7154bef660a4eb85192fc7052c4009497b862c
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 2009 Sun Microsystems, Inc.  All rights reserved.
24 * Use is subject to license terms.
25 */
26
27/*
28 * This file contains all the routines used when modifying on-disk SPA state.
29 * This includes opening, importing, destroying, exporting a pool, and syncing a
30 * pool.
31 */
32
33#include <sys/zfs_context.h>
34#include <sys/fm/fs/zfs.h>
35#include <sys/spa_impl.h>
36#include <sys/zio.h>
37#include <sys/zio_checksum.h>
38#include <sys/zio_compress.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/vdev_impl.h>
44#include <sys/metaslab.h>
45#include <sys/uberblock_impl.h>
46#include <sys/txg.h>
47#include <sys/avl.h>
48#include <sys/dmu_traverse.h>
49#include <sys/dmu_objset.h>
50#include <sys/unique.h>
51#include <sys/dsl_pool.h>
52#include <sys/dsl_dataset.h>
53#include <sys/dsl_dir.h>
54#include <sys/dsl_prop.h>
55#include <sys/dsl_synctask.h>
56#include <sys/fs/zfs.h>
57#include <sys/arc.h>
58#include <sys/callb.h>
59#include <sys/systeminfo.h>
60#include <sys/sunddi.h>
61#include <sys/spa_boot.h>
62
63#ifdef	_KERNEL
64#include <sys/zone.h>
65#endif	/* _KERNEL */
66
67#include "zfs_prop.h"
68#include "zfs_comutil.h"
69
70int zio_taskq_threads[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
71	/*	ISSUE	INTR					*/
72	{	1,	1	},	/* ZIO_TYPE_NULL	*/
73	{	1,	8	},	/* ZIO_TYPE_READ	*/
74	{	8,	1	},	/* ZIO_TYPE_WRITE	*/
75	{	1,	1	},	/* ZIO_TYPE_FREE	*/
76	{	1,	1	},	/* ZIO_TYPE_CLAIM	*/
77	{	1,	1	},	/* ZIO_TYPE_IOCTL	*/
78};
79
80static void spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx);
81static boolean_t spa_has_active_shared_spare(spa_t *spa);
82
83/*
84 * ==========================================================================
85 * SPA properties routines
86 * ==========================================================================
87 */
88
89/*
90 * Add a (source=src, propname=propval) list to an nvlist.
91 */
92static void
93spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
94    uint64_t intval, zprop_source_t src)
95{
96	const char *propname = zpool_prop_to_name(prop);
97	nvlist_t *propval;
98
99	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
100	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
101
102	if (strval != NULL)
103		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
104	else
105		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
106
107	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
108	nvlist_free(propval);
109}
110
111/*
112 * Get property values from the spa configuration.
113 */
114static void
115spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
116{
117	uint64_t size;
118	uint64_t used;
119	uint64_t cap, version;
120	zprop_source_t src = ZPROP_SRC_NONE;
121	spa_config_dirent_t *dp;
122
123	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
124
125	if (spa->spa_root_vdev != NULL) {
126		size = spa_get_space(spa);
127		used = spa_get_alloc(spa);
128		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
129		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
130		spa_prop_add_list(*nvp, ZPOOL_PROP_USED, NULL, used, src);
131		spa_prop_add_list(*nvp, ZPOOL_PROP_AVAILABLE, NULL,
132		    size - used, src);
133
134		cap = (size == 0) ? 0 : (used * 100 / size);
135		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
136
137		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
138		    spa->spa_root_vdev->vdev_state, src);
139
140		version = spa_version(spa);
141		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
142			src = ZPROP_SRC_DEFAULT;
143		else
144			src = ZPROP_SRC_LOCAL;
145		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
146	}
147
148	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
149
150	if (spa->spa_root != NULL)
151		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
152		    0, ZPROP_SRC_LOCAL);
153
154	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
155		if (dp->scd_path == NULL) {
156			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
157			    "none", 0, ZPROP_SRC_LOCAL);
158		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
159			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
160			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
161		}
162	}
163}
164
165/*
166 * Get zpool property values.
167 */
168int
169spa_prop_get(spa_t *spa, nvlist_t **nvp)
170{
171	zap_cursor_t zc;
172	zap_attribute_t za;
173	objset_t *mos = spa->spa_meta_objset;
174	int err;
175
176	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
177
178	mutex_enter(&spa->spa_props_lock);
179
180	/*
181	 * Get properties from the spa config.
182	 */
183	spa_prop_get_config(spa, nvp);
184
185	/* If no pool property object, no more prop to get. */
186	if (spa->spa_pool_props_object == 0) {
187		mutex_exit(&spa->spa_props_lock);
188		return (0);
189	}
190
191	/*
192	 * Get properties from the MOS pool property object.
193	 */
194	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
195	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
196	    zap_cursor_advance(&zc)) {
197		uint64_t intval = 0;
198		char *strval = NULL;
199		zprop_source_t src = ZPROP_SRC_DEFAULT;
200		zpool_prop_t prop;
201
202		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
203			continue;
204
205		switch (za.za_integer_length) {
206		case 8:
207			/* integer property */
208			if (za.za_first_integer !=
209			    zpool_prop_default_numeric(prop))
210				src = ZPROP_SRC_LOCAL;
211
212			if (prop == ZPOOL_PROP_BOOTFS) {
213				dsl_pool_t *dp;
214				dsl_dataset_t *ds = NULL;
215
216				dp = spa_get_dsl(spa);
217				rw_enter(&dp->dp_config_rwlock, RW_READER);
218				if (err = dsl_dataset_hold_obj(dp,
219				    za.za_first_integer, FTAG, &ds)) {
220					rw_exit(&dp->dp_config_rwlock);
221					break;
222				}
223
224				strval = kmem_alloc(
225				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
226				    KM_SLEEP);
227				dsl_dataset_name(ds, strval);
228				dsl_dataset_rele(ds, FTAG);
229				rw_exit(&dp->dp_config_rwlock);
230			} else {
231				strval = NULL;
232				intval = za.za_first_integer;
233			}
234
235			spa_prop_add_list(*nvp, prop, strval, intval, src);
236
237			if (strval != NULL)
238				kmem_free(strval,
239				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
240
241			break;
242
243		case 1:
244			/* string property */
245			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
246			err = zap_lookup(mos, spa->spa_pool_props_object,
247			    za.za_name, 1, za.za_num_integers, strval);
248			if (err) {
249				kmem_free(strval, za.za_num_integers);
250				break;
251			}
252			spa_prop_add_list(*nvp, prop, strval, 0, src);
253			kmem_free(strval, za.za_num_integers);
254			break;
255
256		default:
257			break;
258		}
259	}
260	zap_cursor_fini(&zc);
261	mutex_exit(&spa->spa_props_lock);
262out:
263	if (err && err != ENOENT) {
264		nvlist_free(*nvp);
265		*nvp = NULL;
266		return (err);
267	}
268
269	return (0);
270}
271
272/*
273 * Validate the given pool properties nvlist and modify the list
274 * for the property values to be set.
275 */
276static int
277spa_prop_validate(spa_t *spa, nvlist_t *props)
278{
279	nvpair_t *elem;
280	int error = 0, reset_bootfs = 0;
281	uint64_t objnum;
282
283	elem = NULL;
284	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
285		zpool_prop_t prop;
286		char *propname, *strval;
287		uint64_t intval;
288		objset_t *os;
289		char *slash;
290
291		propname = nvpair_name(elem);
292
293		if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
294			return (EINVAL);
295
296		switch (prop) {
297		case ZPOOL_PROP_VERSION:
298			error = nvpair_value_uint64(elem, &intval);
299			if (!error &&
300			    (intval < spa_version(spa) || intval > SPA_VERSION))
301				error = EINVAL;
302			break;
303
304		case ZPOOL_PROP_DELEGATION:
305		case ZPOOL_PROP_AUTOREPLACE:
306		case ZPOOL_PROP_LISTSNAPS:
307			error = nvpair_value_uint64(elem, &intval);
308			if (!error && intval > 1)
309				error = EINVAL;
310			break;
311
312		case ZPOOL_PROP_BOOTFS:
313			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
314				error = ENOTSUP;
315				break;
316			}
317
318			/*
319			 * Make sure the vdev config is bootable
320			 */
321			if (!vdev_is_bootable(spa->spa_root_vdev)) {
322				error = ENOTSUP;
323				break;
324			}
325
326			reset_bootfs = 1;
327
328			error = nvpair_value_string(elem, &strval);
329
330			if (!error) {
331				uint64_t compress;
332
333				if (strval == NULL || strval[0] == '\0') {
334					objnum = zpool_prop_default_numeric(
335					    ZPOOL_PROP_BOOTFS);
336					break;
337				}
338
339				if (error = dmu_objset_open(strval, DMU_OST_ZFS,
340				    DS_MODE_USER | DS_MODE_READONLY, &os))
341					break;
342
343				/* We don't support gzip bootable datasets */
344				if ((error = dsl_prop_get_integer(strval,
345				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
346				    &compress, NULL)) == 0 &&
347				    !BOOTFS_COMPRESS_VALID(compress)) {
348					error = ENOTSUP;
349				} else {
350					objnum = dmu_objset_id(os);
351				}
352				dmu_objset_close(os);
353			}
354			break;
355
356		case ZPOOL_PROP_FAILUREMODE:
357			error = nvpair_value_uint64(elem, &intval);
358			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
359			    intval > ZIO_FAILURE_MODE_PANIC))
360				error = EINVAL;
361
362			/*
363			 * This is a special case which only occurs when
364			 * the pool has completely failed. This allows
365			 * the user to change the in-core failmode property
366			 * without syncing it out to disk (I/Os might
367			 * currently be blocked). We do this by returning
368			 * EIO to the caller (spa_prop_set) to trick it
369			 * into thinking we encountered a property validation
370			 * error.
371			 */
372			if (!error && spa_suspended(spa)) {
373				spa->spa_failmode = intval;
374				error = EIO;
375			}
376			break;
377
378		case ZPOOL_PROP_CACHEFILE:
379			if ((error = nvpair_value_string(elem, &strval)) != 0)
380				break;
381
382			if (strval[0] == '\0')
383				break;
384
385			if (strcmp(strval, "none") == 0)
386				break;
387
388			if (strval[0] != '/') {
389				error = EINVAL;
390				break;
391			}
392
393			slash = strrchr(strval, '/');
394			ASSERT(slash != NULL);
395
396			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
397			    strcmp(slash, "/..") == 0)
398				error = EINVAL;
399			break;
400		}
401
402		if (error)
403			break;
404	}
405
406	if (!error && reset_bootfs) {
407		error = nvlist_remove(props,
408		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
409
410		if (!error) {
411			error = nvlist_add_uint64(props,
412			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
413		}
414	}
415
416	return (error);
417}
418
419void
420spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
421{
422	char *cachefile;
423	spa_config_dirent_t *dp;
424
425	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
426	    &cachefile) != 0)
427		return;
428
429	dp = kmem_alloc(sizeof (spa_config_dirent_t),
430	    KM_SLEEP);
431
432	if (cachefile[0] == '\0')
433		dp->scd_path = spa_strdup(spa_config_path);
434	else if (strcmp(cachefile, "none") == 0)
435		dp->scd_path = NULL;
436	else
437		dp->scd_path = spa_strdup(cachefile);
438
439	list_insert_head(&spa->spa_config_list, dp);
440	if (need_sync)
441		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
442}
443
444int
445spa_prop_set(spa_t *spa, nvlist_t *nvp)
446{
447	int error;
448	nvpair_t *elem;
449	boolean_t need_sync = B_FALSE;
450	zpool_prop_t prop;
451
452	if ((error = spa_prop_validate(spa, nvp)) != 0)
453		return (error);
454
455	elem = NULL;
456	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
457		if ((prop = zpool_name_to_prop(
458		    nvpair_name(elem))) == ZPROP_INVAL)
459			return (EINVAL);
460
461		if (prop == ZPOOL_PROP_CACHEFILE || prop == ZPOOL_PROP_ALTROOT)
462			continue;
463
464		need_sync = B_TRUE;
465		break;
466	}
467
468	if (need_sync)
469		return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
470		    spa, nvp, 3));
471	else
472		return (0);
473}
474
475/*
476 * If the bootfs property value is dsobj, clear it.
477 */
478void
479spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
480{
481	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
482		VERIFY(zap_remove(spa->spa_meta_objset,
483		    spa->spa_pool_props_object,
484		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
485		spa->spa_bootfs = 0;
486	}
487}
488
489/*
490 * ==========================================================================
491 * SPA state manipulation (open/create/destroy/import/export)
492 * ==========================================================================
493 */
494
495static int
496spa_error_entry_compare(const void *a, const void *b)
497{
498	spa_error_entry_t *sa = (spa_error_entry_t *)a;
499	spa_error_entry_t *sb = (spa_error_entry_t *)b;
500	int ret;
501
502	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
503	    sizeof (zbookmark_t));
504
505	if (ret < 0)
506		return (-1);
507	else if (ret > 0)
508		return (1);
509	else
510		return (0);
511}
512
513/*
514 * Utility function which retrieves copies of the current logs and
515 * re-initializes them in the process.
516 */
517void
518spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
519{
520	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
521
522	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
523	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
524
525	avl_create(&spa->spa_errlist_scrub,
526	    spa_error_entry_compare, sizeof (spa_error_entry_t),
527	    offsetof(spa_error_entry_t, se_avl));
528	avl_create(&spa->spa_errlist_last,
529	    spa_error_entry_compare, sizeof (spa_error_entry_t),
530	    offsetof(spa_error_entry_t, se_avl));
531}
532
533/*
534 * Activate an uninitialized pool.
535 */
536static void
537spa_activate(spa_t *spa, int mode)
538{
539	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
540
541	spa->spa_state = POOL_STATE_ACTIVE;
542	spa->spa_mode = mode;
543
544	spa->spa_normal_class = metaslab_class_create();
545	spa->spa_log_class = metaslab_class_create();
546
547	for (int t = 0; t < ZIO_TYPES; t++) {
548		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
549			spa->spa_zio_taskq[t][q] = taskq_create("spa_zio",
550			    zio_taskq_threads[t][q], maxclsyspri, 50,
551			    INT_MAX, TASKQ_PREPOPULATE);
552		}
553	}
554
555	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
556	    offsetof(vdev_t, vdev_config_dirty_node));
557	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
558	    offsetof(vdev_t, vdev_state_dirty_node));
559
560	txg_list_create(&spa->spa_vdev_txg_list,
561	    offsetof(struct vdev, vdev_txg_node));
562
563	avl_create(&spa->spa_errlist_scrub,
564	    spa_error_entry_compare, sizeof (spa_error_entry_t),
565	    offsetof(spa_error_entry_t, se_avl));
566	avl_create(&spa->spa_errlist_last,
567	    spa_error_entry_compare, sizeof (spa_error_entry_t),
568	    offsetof(spa_error_entry_t, se_avl));
569}
570
571/*
572 * Opposite of spa_activate().
573 */
574static void
575spa_deactivate(spa_t *spa)
576{
577	ASSERT(spa->spa_sync_on == B_FALSE);
578	ASSERT(spa->spa_dsl_pool == NULL);
579	ASSERT(spa->spa_root_vdev == NULL);
580
581	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
582
583	txg_list_destroy(&spa->spa_vdev_txg_list);
584
585	list_destroy(&spa->spa_config_dirty_list);
586	list_destroy(&spa->spa_state_dirty_list);
587
588	for (int t = 0; t < ZIO_TYPES; t++) {
589		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
590			taskq_destroy(spa->spa_zio_taskq[t][q]);
591			spa->spa_zio_taskq[t][q] = NULL;
592		}
593	}
594
595	metaslab_class_destroy(spa->spa_normal_class);
596	spa->spa_normal_class = NULL;
597
598	metaslab_class_destroy(spa->spa_log_class);
599	spa->spa_log_class = NULL;
600
601	/*
602	 * If this was part of an import or the open otherwise failed, we may
603	 * still have errors left in the queues.  Empty them just in case.
604	 */
605	spa_errlog_drain(spa);
606
607	avl_destroy(&spa->spa_errlist_scrub);
608	avl_destroy(&spa->spa_errlist_last);
609
610	spa->spa_state = POOL_STATE_UNINITIALIZED;
611}
612
613/*
614 * Verify a pool configuration, and construct the vdev tree appropriately.  This
615 * will create all the necessary vdevs in the appropriate layout, with each vdev
616 * in the CLOSED state.  This will prep the pool before open/creation/import.
617 * All vdev validation is done by the vdev_alloc() routine.
618 */
619static int
620spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
621    uint_t id, int atype)
622{
623	nvlist_t **child;
624	uint_t c, children;
625	int error;
626
627	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
628		return (error);
629
630	if ((*vdp)->vdev_ops->vdev_op_leaf)
631		return (0);
632
633	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
634	    &child, &children);
635
636	if (error == ENOENT)
637		return (0);
638
639	if (error) {
640		vdev_free(*vdp);
641		*vdp = NULL;
642		return (EINVAL);
643	}
644
645	for (c = 0; c < children; c++) {
646		vdev_t *vd;
647		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
648		    atype)) != 0) {
649			vdev_free(*vdp);
650			*vdp = NULL;
651			return (error);
652		}
653	}
654
655	ASSERT(*vdp != NULL);
656
657	return (0);
658}
659
660/*
661 * Opposite of spa_load().
662 */
663static void
664spa_unload(spa_t *spa)
665{
666	int i;
667
668	ASSERT(MUTEX_HELD(&spa_namespace_lock));
669
670	/*
671	 * Stop async tasks.
672	 */
673	spa_async_suspend(spa);
674
675	/*
676	 * Stop syncing.
677	 */
678	if (spa->spa_sync_on) {
679		txg_sync_stop(spa->spa_dsl_pool);
680		spa->spa_sync_on = B_FALSE;
681	}
682
683	/*
684	 * Wait for any outstanding async I/O to complete.
685	 */
686	mutex_enter(&spa->spa_async_root_lock);
687	while (spa->spa_async_root_count != 0)
688		cv_wait(&spa->spa_async_root_cv, &spa->spa_async_root_lock);
689	mutex_exit(&spa->spa_async_root_lock);
690
691	/*
692	 * Close the dsl pool.
693	 */
694	if (spa->spa_dsl_pool) {
695		dsl_pool_close(spa->spa_dsl_pool);
696		spa->spa_dsl_pool = NULL;
697	}
698
699	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
700
701	/*
702	 * Drop and purge level 2 cache
703	 */
704	spa_l2cache_drop(spa);
705
706	/*
707	 * Close all vdevs.
708	 */
709	if (spa->spa_root_vdev)
710		vdev_free(spa->spa_root_vdev);
711	ASSERT(spa->spa_root_vdev == NULL);
712
713	for (i = 0; i < spa->spa_spares.sav_count; i++)
714		vdev_free(spa->spa_spares.sav_vdevs[i]);
715	if (spa->spa_spares.sav_vdevs) {
716		kmem_free(spa->spa_spares.sav_vdevs,
717		    spa->spa_spares.sav_count * sizeof (void *));
718		spa->spa_spares.sav_vdevs = NULL;
719	}
720	if (spa->spa_spares.sav_config) {
721		nvlist_free(spa->spa_spares.sav_config);
722		spa->spa_spares.sav_config = NULL;
723	}
724	spa->spa_spares.sav_count = 0;
725
726	for (i = 0; i < spa->spa_l2cache.sav_count; i++)
727		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
728	if (spa->spa_l2cache.sav_vdevs) {
729		kmem_free(spa->spa_l2cache.sav_vdevs,
730		    spa->spa_l2cache.sav_count * sizeof (void *));
731		spa->spa_l2cache.sav_vdevs = NULL;
732	}
733	if (spa->spa_l2cache.sav_config) {
734		nvlist_free(spa->spa_l2cache.sav_config);
735		spa->spa_l2cache.sav_config = NULL;
736	}
737	spa->spa_l2cache.sav_count = 0;
738
739	spa->spa_async_suspended = 0;
740
741	spa_config_exit(spa, SCL_ALL, FTAG);
742}
743
744/*
745 * Load (or re-load) the current list of vdevs describing the active spares for
746 * this pool.  When this is called, we have some form of basic information in
747 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
748 * then re-generate a more complete list including status information.
749 */
750static void
751spa_load_spares(spa_t *spa)
752{
753	nvlist_t **spares;
754	uint_t nspares;
755	int i;
756	vdev_t *vd, *tvd;
757
758	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
759
760	/*
761	 * First, close and free any existing spare vdevs.
762	 */
763	for (i = 0; i < spa->spa_spares.sav_count; i++) {
764		vd = spa->spa_spares.sav_vdevs[i];
765
766		/* Undo the call to spa_activate() below */
767		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
768		    B_FALSE)) != NULL && tvd->vdev_isspare)
769			spa_spare_remove(tvd);
770		vdev_close(vd);
771		vdev_free(vd);
772	}
773
774	if (spa->spa_spares.sav_vdevs)
775		kmem_free(spa->spa_spares.sav_vdevs,
776		    spa->spa_spares.sav_count * sizeof (void *));
777
778	if (spa->spa_spares.sav_config == NULL)
779		nspares = 0;
780	else
781		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
782		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
783
784	spa->spa_spares.sav_count = (int)nspares;
785	spa->spa_spares.sav_vdevs = NULL;
786
787	if (nspares == 0)
788		return;
789
790	/*
791	 * Construct the array of vdevs, opening them to get status in the
792	 * process.   For each spare, there is potentially two different vdev_t
793	 * structures associated with it: one in the list of spares (used only
794	 * for basic validation purposes) and one in the active vdev
795	 * configuration (if it's spared in).  During this phase we open and
796	 * validate each vdev on the spare list.  If the vdev also exists in the
797	 * active configuration, then we also mark this vdev as an active spare.
798	 */
799	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
800	    KM_SLEEP);
801	for (i = 0; i < spa->spa_spares.sav_count; i++) {
802		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
803		    VDEV_ALLOC_SPARE) == 0);
804		ASSERT(vd != NULL);
805
806		spa->spa_spares.sav_vdevs[i] = vd;
807
808		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
809		    B_FALSE)) != NULL) {
810			if (!tvd->vdev_isspare)
811				spa_spare_add(tvd);
812
813			/*
814			 * We only mark the spare active if we were successfully
815			 * able to load the vdev.  Otherwise, importing a pool
816			 * with a bad active spare would result in strange
817			 * behavior, because multiple pool would think the spare
818			 * is actively in use.
819			 *
820			 * There is a vulnerability here to an equally bizarre
821			 * circumstance, where a dead active spare is later
822			 * brought back to life (onlined or otherwise).  Given
823			 * the rarity of this scenario, and the extra complexity
824			 * it adds, we ignore the possibility.
825			 */
826			if (!vdev_is_dead(tvd))
827				spa_spare_activate(tvd);
828		}
829
830		vd->vdev_top = vd;
831
832		if (vdev_open(vd) != 0)
833			continue;
834
835		if (vdev_validate_aux(vd) == 0)
836			spa_spare_add(vd);
837	}
838
839	/*
840	 * Recompute the stashed list of spares, with status information
841	 * this time.
842	 */
843	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
844	    DATA_TYPE_NVLIST_ARRAY) == 0);
845
846	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
847	    KM_SLEEP);
848	for (i = 0; i < spa->spa_spares.sav_count; i++)
849		spares[i] = vdev_config_generate(spa,
850		    spa->spa_spares.sav_vdevs[i], B_TRUE, B_TRUE, B_FALSE);
851	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
852	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
853	for (i = 0; i < spa->spa_spares.sav_count; i++)
854		nvlist_free(spares[i]);
855	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
856}
857
858/*
859 * Load (or re-load) the current list of vdevs describing the active l2cache for
860 * this pool.  When this is called, we have some form of basic information in
861 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
862 * then re-generate a more complete list including status information.
863 * Devices which are already active have their details maintained, and are
864 * not re-opened.
865 */
866static void
867spa_load_l2cache(spa_t *spa)
868{
869	nvlist_t **l2cache;
870	uint_t nl2cache;
871	int i, j, oldnvdevs;
872	uint64_t guid, size;
873	vdev_t *vd, **oldvdevs, **newvdevs;
874	spa_aux_vdev_t *sav = &spa->spa_l2cache;
875
876	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
877
878	if (sav->sav_config != NULL) {
879		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
880		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
881		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
882	} else {
883		nl2cache = 0;
884	}
885
886	oldvdevs = sav->sav_vdevs;
887	oldnvdevs = sav->sav_count;
888	sav->sav_vdevs = NULL;
889	sav->sav_count = 0;
890
891	/*
892	 * Process new nvlist of vdevs.
893	 */
894	for (i = 0; i < nl2cache; i++) {
895		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
896		    &guid) == 0);
897
898		newvdevs[i] = NULL;
899		for (j = 0; j < oldnvdevs; j++) {
900			vd = oldvdevs[j];
901			if (vd != NULL && guid == vd->vdev_guid) {
902				/*
903				 * Retain previous vdev for add/remove ops.
904				 */
905				newvdevs[i] = vd;
906				oldvdevs[j] = NULL;
907				break;
908			}
909		}
910
911		if (newvdevs[i] == NULL) {
912			/*
913			 * Create new vdev
914			 */
915			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
916			    VDEV_ALLOC_L2CACHE) == 0);
917			ASSERT(vd != NULL);
918			newvdevs[i] = vd;
919
920			/*
921			 * Commit this vdev as an l2cache device,
922			 * even if it fails to open.
923			 */
924			spa_l2cache_add(vd);
925
926			vd->vdev_top = vd;
927			vd->vdev_aux = sav;
928
929			spa_l2cache_activate(vd);
930
931			if (vdev_open(vd) != 0)
932				continue;
933
934			(void) vdev_validate_aux(vd);
935
936			if (!vdev_is_dead(vd)) {
937				size = vdev_get_rsize(vd);
938				l2arc_add_vdev(spa, vd,
939				    VDEV_LABEL_START_SIZE,
940				    size - VDEV_LABEL_START_SIZE);
941			}
942		}
943	}
944
945	/*
946	 * Purge vdevs that were dropped
947	 */
948	for (i = 0; i < oldnvdevs; i++) {
949		uint64_t pool;
950
951		vd = oldvdevs[i];
952		if (vd != NULL) {
953			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
954			    pool != 0ULL && l2arc_vdev_present(vd))
955				l2arc_remove_vdev(vd);
956			(void) vdev_close(vd);
957			spa_l2cache_remove(vd);
958		}
959	}
960
961	if (oldvdevs)
962		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
963
964	if (sav->sav_config == NULL)
965		goto out;
966
967	sav->sav_vdevs = newvdevs;
968	sav->sav_count = (int)nl2cache;
969
970	/*
971	 * Recompute the stashed list of l2cache devices, with status
972	 * information this time.
973	 */
974	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
975	    DATA_TYPE_NVLIST_ARRAY) == 0);
976
977	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
978	for (i = 0; i < sav->sav_count; i++)
979		l2cache[i] = vdev_config_generate(spa,
980		    sav->sav_vdevs[i], B_TRUE, B_FALSE, B_TRUE);
981	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
982	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
983out:
984	for (i = 0; i < sav->sav_count; i++)
985		nvlist_free(l2cache[i]);
986	if (sav->sav_count)
987		kmem_free(l2cache, sav->sav_count * sizeof (void *));
988}
989
990static int
991load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
992{
993	dmu_buf_t *db;
994	char *packed = NULL;
995	size_t nvsize = 0;
996	int error;
997	*value = NULL;
998
999	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1000	nvsize = *(uint64_t *)db->db_data;
1001	dmu_buf_rele(db, FTAG);
1002
1003	packed = kmem_alloc(nvsize, KM_SLEEP);
1004	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed);
1005	if (error == 0)
1006		error = nvlist_unpack(packed, nvsize, value, 0);
1007	kmem_free(packed, nvsize);
1008
1009	return (error);
1010}
1011
1012/*
1013 * Checks to see if the given vdev could not be opened, in which case we post a
1014 * sysevent to notify the autoreplace code that the device has been removed.
1015 */
1016static void
1017spa_check_removed(vdev_t *vd)
1018{
1019	int c;
1020
1021	for (c = 0; c < vd->vdev_children; c++)
1022		spa_check_removed(vd->vdev_child[c]);
1023
1024	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1025		zfs_post_autoreplace(vd->vdev_spa, vd);
1026		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1027	}
1028}
1029
1030/*
1031 * Check for missing log devices
1032 */
1033int
1034spa_check_logs(spa_t *spa)
1035{
1036	switch (spa->spa_log_state) {
1037	case SPA_LOG_MISSING:
1038		/* need to recheck in case slog has been restored */
1039	case SPA_LOG_UNKNOWN:
1040		if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1041		    DS_FIND_CHILDREN)) {
1042			spa->spa_log_state = SPA_LOG_MISSING;
1043			return (1);
1044		}
1045		break;
1046
1047	case SPA_LOG_CLEAR:
1048		(void) dmu_objset_find(spa->spa_name, zil_clear_log_chain, NULL,
1049		    DS_FIND_CHILDREN);
1050		break;
1051	}
1052	spa->spa_log_state = SPA_LOG_GOOD;
1053	return (0);
1054}
1055
1056/*
1057 * Load an existing storage pool, using the pool's builtin spa_config as a
1058 * source of configuration information.
1059 */
1060static int
1061spa_load(spa_t *spa, nvlist_t *config, spa_load_state_t state, int mosconfig)
1062{
1063	int error = 0;
1064	nvlist_t *nvroot = NULL;
1065	vdev_t *rvd;
1066	uberblock_t *ub = &spa->spa_uberblock;
1067	uint64_t config_cache_txg = spa->spa_config_txg;
1068	uint64_t pool_guid;
1069	uint64_t version;
1070	uint64_t autoreplace = 0;
1071	int orig_mode = spa->spa_mode;
1072	char *ereport = FM_EREPORT_ZFS_POOL;
1073
1074	/*
1075	 * If this is an untrusted config, access the pool in read-only mode.
1076	 * This prevents things like resilvering recently removed devices.
1077	 */
1078	if (!mosconfig)
1079		spa->spa_mode = FREAD;
1080
1081	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1082
1083	spa->spa_load_state = state;
1084
1085	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) ||
1086	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
1087		error = EINVAL;
1088		goto out;
1089	}
1090
1091	/*
1092	 * Versioning wasn't explicitly added to the label until later, so if
1093	 * it's not present treat it as the initial version.
1094	 */
1095	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &version) != 0)
1096		version = SPA_VERSION_INITIAL;
1097
1098	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1099	    &spa->spa_config_txg);
1100
1101	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1102	    spa_guid_exists(pool_guid, 0)) {
1103		error = EEXIST;
1104		goto out;
1105	}
1106
1107	spa->spa_load_guid = pool_guid;
1108
1109	/*
1110	 * Parse the configuration into a vdev tree.  We explicitly set the
1111	 * value that will be returned by spa_version() since parsing the
1112	 * configuration requires knowing the version number.
1113	 */
1114	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1115	spa->spa_ubsync.ub_version = version;
1116	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_LOAD);
1117	spa_config_exit(spa, SCL_ALL, FTAG);
1118
1119	if (error != 0)
1120		goto out;
1121
1122	ASSERT(spa->spa_root_vdev == rvd);
1123	ASSERT(spa_guid(spa) == pool_guid);
1124
1125	/*
1126	 * Try to open all vdevs, loading each label in the process.
1127	 */
1128	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1129	error = vdev_open(rvd);
1130	spa_config_exit(spa, SCL_ALL, FTAG);
1131	if (error != 0)
1132		goto out;
1133
1134	/*
1135	 * Validate the labels for all leaf vdevs.  We need to grab the config
1136	 * lock because all label I/O is done with ZIO_FLAG_CONFIG_WRITER.
1137	 */
1138	if (mosconfig) {
1139		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1140		error = vdev_validate(rvd);
1141		spa_config_exit(spa, SCL_ALL, FTAG);
1142		if (error != 0)
1143			goto out;
1144	}
1145
1146	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
1147		error = ENXIO;
1148		goto out;
1149	}
1150
1151	/*
1152	 * Find the best uberblock.
1153	 */
1154	vdev_uberblock_load(NULL, rvd, ub);
1155
1156	/*
1157	 * If we weren't able to find a single valid uberblock, return failure.
1158	 */
1159	if (ub->ub_txg == 0) {
1160		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1161		    VDEV_AUX_CORRUPT_DATA);
1162		error = ENXIO;
1163		goto out;
1164	}
1165
1166	/*
1167	 * If the pool is newer than the code, we can't open it.
1168	 */
1169	if (ub->ub_version > SPA_VERSION) {
1170		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1171		    VDEV_AUX_VERSION_NEWER);
1172		error = ENOTSUP;
1173		goto out;
1174	}
1175
1176	/*
1177	 * If the vdev guid sum doesn't match the uberblock, we have an
1178	 * incomplete configuration.
1179	 */
1180	if (rvd->vdev_guid_sum != ub->ub_guid_sum && mosconfig) {
1181		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1182		    VDEV_AUX_BAD_GUID_SUM);
1183		error = ENXIO;
1184		goto out;
1185	}
1186
1187	/*
1188	 * Initialize internal SPA structures.
1189	 */
1190	spa->spa_state = POOL_STATE_ACTIVE;
1191	spa->spa_ubsync = spa->spa_uberblock;
1192	spa->spa_first_txg = spa_last_synced_txg(spa) + 1;
1193	error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1194	if (error) {
1195		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1196		    VDEV_AUX_CORRUPT_DATA);
1197		goto out;
1198	}
1199	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1200
1201	if (zap_lookup(spa->spa_meta_objset,
1202	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
1203	    sizeof (uint64_t), 1, &spa->spa_config_object) != 0) {
1204		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1205		    VDEV_AUX_CORRUPT_DATA);
1206		error = EIO;
1207		goto out;
1208	}
1209
1210	if (!mosconfig) {
1211		nvlist_t *newconfig;
1212		uint64_t hostid;
1213
1214		if (load_nvlist(spa, spa->spa_config_object, &newconfig) != 0) {
1215			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1216			    VDEV_AUX_CORRUPT_DATA);
1217			error = EIO;
1218			goto out;
1219		}
1220
1221		if (!spa_is_root(spa) && nvlist_lookup_uint64(newconfig,
1222		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
1223			char *hostname;
1224			unsigned long myhostid = 0;
1225
1226			VERIFY(nvlist_lookup_string(newconfig,
1227			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
1228
1229#ifdef	_KERNEL
1230			myhostid = zone_get_hostid(NULL);
1231#else	/* _KERNEL */
1232			/*
1233			 * We're emulating the system's hostid in userland, so
1234			 * we can't use zone_get_hostid().
1235			 */
1236			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
1237#endif	/* _KERNEL */
1238			if (hostid != 0 && myhostid != 0 &&
1239			    hostid != myhostid) {
1240				cmn_err(CE_WARN, "pool '%s' could not be "
1241				    "loaded as it was last accessed by "
1242				    "another system (host: %s hostid: 0x%lx). "
1243				    "See: http://www.sun.com/msg/ZFS-8000-EY",
1244				    spa_name(spa), hostname,
1245				    (unsigned long)hostid);
1246				error = EBADF;
1247				goto out;
1248			}
1249		}
1250
1251		spa_config_set(spa, newconfig);
1252		spa_unload(spa);
1253		spa_deactivate(spa);
1254		spa_activate(spa, orig_mode);
1255
1256		return (spa_load(spa, newconfig, state, B_TRUE));
1257	}
1258
1259	if (zap_lookup(spa->spa_meta_objset,
1260	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
1261	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj) != 0) {
1262		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1263		    VDEV_AUX_CORRUPT_DATA);
1264		error = EIO;
1265		goto out;
1266	}
1267
1268	/*
1269	 * Load the bit that tells us to use the new accounting function
1270	 * (raid-z deflation).  If we have an older pool, this will not
1271	 * be present.
1272	 */
1273	error = zap_lookup(spa->spa_meta_objset,
1274	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
1275	    sizeof (uint64_t), 1, &spa->spa_deflate);
1276	if (error != 0 && error != ENOENT) {
1277		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1278		    VDEV_AUX_CORRUPT_DATA);
1279		error = EIO;
1280		goto out;
1281	}
1282
1283	/*
1284	 * Load the persistent error log.  If we have an older pool, this will
1285	 * not be present.
1286	 */
1287	error = zap_lookup(spa->spa_meta_objset,
1288	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_LAST,
1289	    sizeof (uint64_t), 1, &spa->spa_errlog_last);
1290	if (error != 0 && error != ENOENT) {
1291		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1292		    VDEV_AUX_CORRUPT_DATA);
1293		error = EIO;
1294		goto out;
1295	}
1296
1297	error = zap_lookup(spa->spa_meta_objset,
1298	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_SCRUB,
1299	    sizeof (uint64_t), 1, &spa->spa_errlog_scrub);
1300	if (error != 0 && error != ENOENT) {
1301		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1302		    VDEV_AUX_CORRUPT_DATA);
1303		error = EIO;
1304		goto out;
1305	}
1306
1307	/*
1308	 * Load the history object.  If we have an older pool, this
1309	 * will not be present.
1310	 */
1311	error = zap_lookup(spa->spa_meta_objset,
1312	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_HISTORY,
1313	    sizeof (uint64_t), 1, &spa->spa_history);
1314	if (error != 0 && error != ENOENT) {
1315		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1316		    VDEV_AUX_CORRUPT_DATA);
1317		error = EIO;
1318		goto out;
1319	}
1320
1321	/*
1322	 * Load any hot spares for this pool.
1323	 */
1324	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1325	    DMU_POOL_SPARES, sizeof (uint64_t), 1, &spa->spa_spares.sav_object);
1326	if (error != 0 && error != ENOENT) {
1327		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1328		    VDEV_AUX_CORRUPT_DATA);
1329		error = EIO;
1330		goto out;
1331	}
1332	if (error == 0) {
1333		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
1334		if (load_nvlist(spa, spa->spa_spares.sav_object,
1335		    &spa->spa_spares.sav_config) != 0) {
1336			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1337			    VDEV_AUX_CORRUPT_DATA);
1338			error = EIO;
1339			goto out;
1340		}
1341
1342		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1343		spa_load_spares(spa);
1344		spa_config_exit(spa, SCL_ALL, FTAG);
1345	}
1346
1347	/*
1348	 * Load any level 2 ARC devices for this pool.
1349	 */
1350	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1351	    DMU_POOL_L2CACHE, sizeof (uint64_t), 1,
1352	    &spa->spa_l2cache.sav_object);
1353	if (error != 0 && error != ENOENT) {
1354		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1355		    VDEV_AUX_CORRUPT_DATA);
1356		error = EIO;
1357		goto out;
1358	}
1359	if (error == 0) {
1360		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
1361		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
1362		    &spa->spa_l2cache.sav_config) != 0) {
1363			vdev_set_state(rvd, B_TRUE,
1364			    VDEV_STATE_CANT_OPEN,
1365			    VDEV_AUX_CORRUPT_DATA);
1366			error = EIO;
1367			goto out;
1368		}
1369
1370		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1371		spa_load_l2cache(spa);
1372		spa_config_exit(spa, SCL_ALL, FTAG);
1373	}
1374
1375	if (spa_check_logs(spa)) {
1376		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1377		    VDEV_AUX_BAD_LOG);
1378		error = ENXIO;
1379		ereport = FM_EREPORT_ZFS_LOG_REPLAY;
1380		goto out;
1381	}
1382
1383
1384	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
1385
1386	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1387	    DMU_POOL_PROPS, sizeof (uint64_t), 1, &spa->spa_pool_props_object);
1388
1389	if (error && error != ENOENT) {
1390		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1391		    VDEV_AUX_CORRUPT_DATA);
1392		error = EIO;
1393		goto out;
1394	}
1395
1396	if (error == 0) {
1397		(void) zap_lookup(spa->spa_meta_objset,
1398		    spa->spa_pool_props_object,
1399		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS),
1400		    sizeof (uint64_t), 1, &spa->spa_bootfs);
1401		(void) zap_lookup(spa->spa_meta_objset,
1402		    spa->spa_pool_props_object,
1403		    zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE),
1404		    sizeof (uint64_t), 1, &autoreplace);
1405		(void) zap_lookup(spa->spa_meta_objset,
1406		    spa->spa_pool_props_object,
1407		    zpool_prop_to_name(ZPOOL_PROP_DELEGATION),
1408		    sizeof (uint64_t), 1, &spa->spa_delegation);
1409		(void) zap_lookup(spa->spa_meta_objset,
1410		    spa->spa_pool_props_object,
1411		    zpool_prop_to_name(ZPOOL_PROP_FAILUREMODE),
1412		    sizeof (uint64_t), 1, &spa->spa_failmode);
1413	}
1414
1415	/*
1416	 * If the 'autoreplace' property is set, then post a resource notifying
1417	 * the ZFS DE that it should not issue any faults for unopenable
1418	 * devices.  We also iterate over the vdevs, and post a sysevent for any
1419	 * unopenable vdevs so that the normal autoreplace handler can take
1420	 * over.
1421	 */
1422	if (autoreplace && state != SPA_LOAD_TRYIMPORT)
1423		spa_check_removed(spa->spa_root_vdev);
1424
1425	/*
1426	 * Load the vdev state for all toplevel vdevs.
1427	 */
1428	vdev_load(rvd);
1429
1430	/*
1431	 * Propagate the leaf DTLs we just loaded all the way up the tree.
1432	 */
1433	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1434	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
1435	spa_config_exit(spa, SCL_ALL, FTAG);
1436
1437	/*
1438	 * Check the state of the root vdev.  If it can't be opened, it
1439	 * indicates one or more toplevel vdevs are faulted.
1440	 */
1441	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
1442		error = ENXIO;
1443		goto out;
1444	}
1445
1446	if (spa_writeable(spa)) {
1447		dmu_tx_t *tx;
1448		int need_update = B_FALSE;
1449
1450		ASSERT(state != SPA_LOAD_TRYIMPORT);
1451
1452		/*
1453		 * Claim log blocks that haven't been committed yet.
1454		 * This must all happen in a single txg.
1455		 */
1456		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
1457		    spa_first_txg(spa));
1458		(void) dmu_objset_find(spa_name(spa),
1459		    zil_claim, tx, DS_FIND_CHILDREN);
1460		dmu_tx_commit(tx);
1461
1462		spa->spa_sync_on = B_TRUE;
1463		txg_sync_start(spa->spa_dsl_pool);
1464
1465		/*
1466		 * Wait for all claims to sync.
1467		 */
1468		txg_wait_synced(spa->spa_dsl_pool, 0);
1469
1470		/*
1471		 * If the config cache is stale, or we have uninitialized
1472		 * metaslabs (see spa_vdev_add()), then update the config.
1473		 */
1474		if (config_cache_txg != spa->spa_config_txg ||
1475		    state == SPA_LOAD_IMPORT)
1476			need_update = B_TRUE;
1477
1478		for (int c = 0; c < rvd->vdev_children; c++)
1479			if (rvd->vdev_child[c]->vdev_ms_array == 0)
1480				need_update = B_TRUE;
1481
1482		/*
1483		 * Update the config cache asychronously in case we're the
1484		 * root pool, in which case the config cache isn't writable yet.
1485		 */
1486		if (need_update)
1487			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
1488
1489		/*
1490		 * Check all DTLs to see if anything needs resilvering.
1491		 */
1492		if (vdev_resilver_needed(rvd, NULL, NULL))
1493			spa_async_request(spa, SPA_ASYNC_RESILVER);
1494	}
1495
1496	error = 0;
1497out:
1498	spa->spa_minref = refcount_count(&spa->spa_refcount);
1499	if (error && error != EBADF)
1500		zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1501	spa->spa_load_state = SPA_LOAD_NONE;
1502	spa->spa_ena = 0;
1503
1504	return (error);
1505}
1506
1507/*
1508 * Pool Open/Import
1509 *
1510 * The import case is identical to an open except that the configuration is sent
1511 * down from userland, instead of grabbed from the configuration cache.  For the
1512 * case of an open, the pool configuration will exist in the
1513 * POOL_STATE_UNINITIALIZED state.
1514 *
1515 * The stats information (gen/count/ustats) is used to gather vdev statistics at
1516 * the same time open the pool, without having to keep around the spa_t in some
1517 * ambiguous state.
1518 */
1519static int
1520spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t **config)
1521{
1522	spa_t *spa;
1523	int error;
1524	int locked = B_FALSE;
1525
1526	*spapp = NULL;
1527
1528	/*
1529	 * As disgusting as this is, we need to support recursive calls to this
1530	 * function because dsl_dir_open() is called during spa_load(), and ends
1531	 * up calling spa_open() again.  The real fix is to figure out how to
1532	 * avoid dsl_dir_open() calling this in the first place.
1533	 */
1534	if (mutex_owner(&spa_namespace_lock) != curthread) {
1535		mutex_enter(&spa_namespace_lock);
1536		locked = B_TRUE;
1537	}
1538
1539	if ((spa = spa_lookup(pool)) == NULL) {
1540		if (locked)
1541			mutex_exit(&spa_namespace_lock);
1542		return (ENOENT);
1543	}
1544	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
1545
1546		spa_activate(spa, spa_mode_global);
1547
1548		error = spa_load(spa, spa->spa_config, SPA_LOAD_OPEN, B_FALSE);
1549
1550		if (error == EBADF) {
1551			/*
1552			 * If vdev_validate() returns failure (indicated by
1553			 * EBADF), it indicates that one of the vdevs indicates
1554			 * that the pool has been exported or destroyed.  If
1555			 * this is the case, the config cache is out of sync and
1556			 * we should remove the pool from the namespace.
1557			 */
1558			spa_unload(spa);
1559			spa_deactivate(spa);
1560			spa_config_sync(spa, B_TRUE, B_TRUE);
1561			spa_remove(spa);
1562			if (locked)
1563				mutex_exit(&spa_namespace_lock);
1564			return (ENOENT);
1565		}
1566
1567		if (error) {
1568			/*
1569			 * We can't open the pool, but we still have useful
1570			 * information: the state of each vdev after the
1571			 * attempted vdev_open().  Return this to the user.
1572			 */
1573			if (config != NULL && spa->spa_root_vdev != NULL)
1574				*config = spa_config_generate(spa, NULL, -1ULL,
1575				    B_TRUE);
1576			spa_unload(spa);
1577			spa_deactivate(spa);
1578			spa->spa_last_open_failed = B_TRUE;
1579			if (locked)
1580				mutex_exit(&spa_namespace_lock);
1581			*spapp = NULL;
1582			return (error);
1583		} else {
1584			spa->spa_last_open_failed = B_FALSE;
1585		}
1586	}
1587
1588	spa_open_ref(spa, tag);
1589
1590	if (locked)
1591		mutex_exit(&spa_namespace_lock);
1592
1593	*spapp = spa;
1594
1595	if (config != NULL)
1596		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
1597
1598	return (0);
1599}
1600
1601int
1602spa_open(const char *name, spa_t **spapp, void *tag)
1603{
1604	return (spa_open_common(name, spapp, tag, NULL));
1605}
1606
1607/*
1608 * Lookup the given spa_t, incrementing the inject count in the process,
1609 * preventing it from being exported or destroyed.
1610 */
1611spa_t *
1612spa_inject_addref(char *name)
1613{
1614	spa_t *spa;
1615
1616	mutex_enter(&spa_namespace_lock);
1617	if ((spa = spa_lookup(name)) == NULL) {
1618		mutex_exit(&spa_namespace_lock);
1619		return (NULL);
1620	}
1621	spa->spa_inject_ref++;
1622	mutex_exit(&spa_namespace_lock);
1623
1624	return (spa);
1625}
1626
1627void
1628spa_inject_delref(spa_t *spa)
1629{
1630	mutex_enter(&spa_namespace_lock);
1631	spa->spa_inject_ref--;
1632	mutex_exit(&spa_namespace_lock);
1633}
1634
1635/*
1636 * Add spares device information to the nvlist.
1637 */
1638static void
1639spa_add_spares(spa_t *spa, nvlist_t *config)
1640{
1641	nvlist_t **spares;
1642	uint_t i, nspares;
1643	nvlist_t *nvroot;
1644	uint64_t guid;
1645	vdev_stat_t *vs;
1646	uint_t vsc;
1647	uint64_t pool;
1648
1649	if (spa->spa_spares.sav_count == 0)
1650		return;
1651
1652	VERIFY(nvlist_lookup_nvlist(config,
1653	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1654	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1655	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1656	if (nspares != 0) {
1657		VERIFY(nvlist_add_nvlist_array(nvroot,
1658		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1659		VERIFY(nvlist_lookup_nvlist_array(nvroot,
1660		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1661
1662		/*
1663		 * Go through and find any spares which have since been
1664		 * repurposed as an active spare.  If this is the case, update
1665		 * their status appropriately.
1666		 */
1667		for (i = 0; i < nspares; i++) {
1668			VERIFY(nvlist_lookup_uint64(spares[i],
1669			    ZPOOL_CONFIG_GUID, &guid) == 0);
1670			if (spa_spare_exists(guid, &pool, NULL) &&
1671			    pool != 0ULL) {
1672				VERIFY(nvlist_lookup_uint64_array(
1673				    spares[i], ZPOOL_CONFIG_STATS,
1674				    (uint64_t **)&vs, &vsc) == 0);
1675				vs->vs_state = VDEV_STATE_CANT_OPEN;
1676				vs->vs_aux = VDEV_AUX_SPARED;
1677			}
1678		}
1679	}
1680}
1681
1682/*
1683 * Add l2cache device information to the nvlist, including vdev stats.
1684 */
1685static void
1686spa_add_l2cache(spa_t *spa, nvlist_t *config)
1687{
1688	nvlist_t **l2cache;
1689	uint_t i, j, nl2cache;
1690	nvlist_t *nvroot;
1691	uint64_t guid;
1692	vdev_t *vd;
1693	vdev_stat_t *vs;
1694	uint_t vsc;
1695
1696	if (spa->spa_l2cache.sav_count == 0)
1697		return;
1698
1699	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
1700
1701	VERIFY(nvlist_lookup_nvlist(config,
1702	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1703	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
1704	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1705	if (nl2cache != 0) {
1706		VERIFY(nvlist_add_nvlist_array(nvroot,
1707		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
1708		VERIFY(nvlist_lookup_nvlist_array(nvroot,
1709		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1710
1711		/*
1712		 * Update level 2 cache device stats.
1713		 */
1714
1715		for (i = 0; i < nl2cache; i++) {
1716			VERIFY(nvlist_lookup_uint64(l2cache[i],
1717			    ZPOOL_CONFIG_GUID, &guid) == 0);
1718
1719			vd = NULL;
1720			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
1721				if (guid ==
1722				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
1723					vd = spa->spa_l2cache.sav_vdevs[j];
1724					break;
1725				}
1726			}
1727			ASSERT(vd != NULL);
1728
1729			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
1730			    ZPOOL_CONFIG_STATS, (uint64_t **)&vs, &vsc) == 0);
1731			vdev_get_stats(vd, vs);
1732		}
1733	}
1734
1735	spa_config_exit(spa, SCL_CONFIG, FTAG);
1736}
1737
1738int
1739spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
1740{
1741	int error;
1742	spa_t *spa;
1743
1744	*config = NULL;
1745	error = spa_open_common(name, &spa, FTAG, config);
1746
1747	if (spa && *config != NULL) {
1748		VERIFY(nvlist_add_uint64(*config, ZPOOL_CONFIG_ERRCOUNT,
1749		    spa_get_errlog_size(spa)) == 0);
1750
1751		if (spa_suspended(spa))
1752			VERIFY(nvlist_add_uint64(*config,
1753			    ZPOOL_CONFIG_SUSPENDED, spa->spa_failmode) == 0);
1754
1755		spa_add_spares(spa, *config);
1756		spa_add_l2cache(spa, *config);
1757	}
1758
1759	/*
1760	 * We want to get the alternate root even for faulted pools, so we cheat
1761	 * and call spa_lookup() directly.
1762	 */
1763	if (altroot) {
1764		if (spa == NULL) {
1765			mutex_enter(&spa_namespace_lock);
1766			spa = spa_lookup(name);
1767			if (spa)
1768				spa_altroot(spa, altroot, buflen);
1769			else
1770				altroot[0] = '\0';
1771			spa = NULL;
1772			mutex_exit(&spa_namespace_lock);
1773		} else {
1774			spa_altroot(spa, altroot, buflen);
1775		}
1776	}
1777
1778	if (spa != NULL)
1779		spa_close(spa, FTAG);
1780
1781	return (error);
1782}
1783
1784/*
1785 * Validate that the auxiliary device array is well formed.  We must have an
1786 * array of nvlists, each which describes a valid leaf vdev.  If this is an
1787 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
1788 * specified, as long as they are well-formed.
1789 */
1790static int
1791spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
1792    spa_aux_vdev_t *sav, const char *config, uint64_t version,
1793    vdev_labeltype_t label)
1794{
1795	nvlist_t **dev;
1796	uint_t i, ndev;
1797	vdev_t *vd;
1798	int error;
1799
1800	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1801
1802	/*
1803	 * It's acceptable to have no devs specified.
1804	 */
1805	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
1806		return (0);
1807
1808	if (ndev == 0)
1809		return (EINVAL);
1810
1811	/*
1812	 * Make sure the pool is formatted with a version that supports this
1813	 * device type.
1814	 */
1815	if (spa_version(spa) < version)
1816		return (ENOTSUP);
1817
1818	/*
1819	 * Set the pending device list so we correctly handle device in-use
1820	 * checking.
1821	 */
1822	sav->sav_pending = dev;
1823	sav->sav_npending = ndev;
1824
1825	for (i = 0; i < ndev; i++) {
1826		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
1827		    mode)) != 0)
1828			goto out;
1829
1830		if (!vd->vdev_ops->vdev_op_leaf) {
1831			vdev_free(vd);
1832			error = EINVAL;
1833			goto out;
1834		}
1835
1836		/*
1837		 * The L2ARC currently only supports disk devices in
1838		 * kernel context.  For user-level testing, we allow it.
1839		 */
1840#ifdef _KERNEL
1841		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
1842		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
1843			error = ENOTBLK;
1844			goto out;
1845		}
1846#endif
1847		vd->vdev_top = vd;
1848
1849		if ((error = vdev_open(vd)) == 0 &&
1850		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
1851			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
1852			    vd->vdev_guid) == 0);
1853		}
1854
1855		vdev_free(vd);
1856
1857		if (error &&
1858		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
1859			goto out;
1860		else
1861			error = 0;
1862	}
1863
1864out:
1865	sav->sav_pending = NULL;
1866	sav->sav_npending = 0;
1867	return (error);
1868}
1869
1870static int
1871spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
1872{
1873	int error;
1874
1875	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1876
1877	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
1878	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
1879	    VDEV_LABEL_SPARE)) != 0) {
1880		return (error);
1881	}
1882
1883	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
1884	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
1885	    VDEV_LABEL_L2CACHE));
1886}
1887
1888static void
1889spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
1890    const char *config)
1891{
1892	int i;
1893
1894	if (sav->sav_config != NULL) {
1895		nvlist_t **olddevs;
1896		uint_t oldndevs;
1897		nvlist_t **newdevs;
1898
1899		/*
1900		 * Generate new dev list by concatentating with the
1901		 * current dev list.
1902		 */
1903		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
1904		    &olddevs, &oldndevs) == 0);
1905
1906		newdevs = kmem_alloc(sizeof (void *) *
1907		    (ndevs + oldndevs), KM_SLEEP);
1908		for (i = 0; i < oldndevs; i++)
1909			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
1910			    KM_SLEEP) == 0);
1911		for (i = 0; i < ndevs; i++)
1912			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
1913			    KM_SLEEP) == 0);
1914
1915		VERIFY(nvlist_remove(sav->sav_config, config,
1916		    DATA_TYPE_NVLIST_ARRAY) == 0);
1917
1918		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1919		    config, newdevs, ndevs + oldndevs) == 0);
1920		for (i = 0; i < oldndevs + ndevs; i++)
1921			nvlist_free(newdevs[i]);
1922		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
1923	} else {
1924		/*
1925		 * Generate a new dev list.
1926		 */
1927		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
1928		    KM_SLEEP) == 0);
1929		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
1930		    devs, ndevs) == 0);
1931	}
1932}
1933
1934/*
1935 * Stop and drop level 2 ARC devices
1936 */
1937void
1938spa_l2cache_drop(spa_t *spa)
1939{
1940	vdev_t *vd;
1941	int i;
1942	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1943
1944	for (i = 0; i < sav->sav_count; i++) {
1945		uint64_t pool;
1946
1947		vd = sav->sav_vdevs[i];
1948		ASSERT(vd != NULL);
1949
1950		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1951		    pool != 0ULL && l2arc_vdev_present(vd))
1952			l2arc_remove_vdev(vd);
1953		if (vd->vdev_isl2cache)
1954			spa_l2cache_remove(vd);
1955		vdev_clear_stats(vd);
1956		(void) vdev_close(vd);
1957	}
1958}
1959
1960/*
1961 * Pool Creation
1962 */
1963int
1964spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
1965    const char *history_str, nvlist_t *zplprops)
1966{
1967	spa_t *spa;
1968	char *altroot = NULL;
1969	vdev_t *rvd;
1970	dsl_pool_t *dp;
1971	dmu_tx_t *tx;
1972	int c, error = 0;
1973	uint64_t txg = TXG_INITIAL;
1974	nvlist_t **spares, **l2cache;
1975	uint_t nspares, nl2cache;
1976	uint64_t version;
1977
1978	/*
1979	 * If this pool already exists, return failure.
1980	 */
1981	mutex_enter(&spa_namespace_lock);
1982	if (spa_lookup(pool) != NULL) {
1983		mutex_exit(&spa_namespace_lock);
1984		return (EEXIST);
1985	}
1986
1987	/*
1988	 * Allocate a new spa_t structure.
1989	 */
1990	(void) nvlist_lookup_string(props,
1991	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
1992	spa = spa_add(pool, altroot);
1993	spa_activate(spa, spa_mode_global);
1994
1995	spa->spa_uberblock.ub_txg = txg - 1;
1996
1997	if (props && (error = spa_prop_validate(spa, props))) {
1998		spa_unload(spa);
1999		spa_deactivate(spa);
2000		spa_remove(spa);
2001		mutex_exit(&spa_namespace_lock);
2002		return (error);
2003	}
2004
2005	if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2006	    &version) != 0)
2007		version = SPA_VERSION;
2008	ASSERT(version <= SPA_VERSION);
2009	spa->spa_uberblock.ub_version = version;
2010	spa->spa_ubsync = spa->spa_uberblock;
2011
2012	/*
2013	 * Create the root vdev.
2014	 */
2015	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2016
2017	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2018
2019	ASSERT(error != 0 || rvd != NULL);
2020	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2021
2022	if (error == 0 && !zfs_allocatable_devs(nvroot))
2023		error = EINVAL;
2024
2025	if (error == 0 &&
2026	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2027	    (error = spa_validate_aux(spa, nvroot, txg,
2028	    VDEV_ALLOC_ADD)) == 0) {
2029		for (c = 0; c < rvd->vdev_children; c++)
2030			vdev_init(rvd->vdev_child[c], txg);
2031		vdev_config_dirty(rvd);
2032	}
2033
2034	spa_config_exit(spa, SCL_ALL, FTAG);
2035
2036	if (error != 0) {
2037		spa_unload(spa);
2038		spa_deactivate(spa);
2039		spa_remove(spa);
2040		mutex_exit(&spa_namespace_lock);
2041		return (error);
2042	}
2043
2044	/*
2045	 * Get the list of spares, if specified.
2046	 */
2047	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2048	    &spares, &nspares) == 0) {
2049		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
2050		    KM_SLEEP) == 0);
2051		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2052		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2053		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2054		spa_load_spares(spa);
2055		spa_config_exit(spa, SCL_ALL, FTAG);
2056		spa->spa_spares.sav_sync = B_TRUE;
2057	}
2058
2059	/*
2060	 * Get the list of level 2 cache devices, if specified.
2061	 */
2062	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2063	    &l2cache, &nl2cache) == 0) {
2064		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2065		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2066		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2067		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2068		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2069		spa_load_l2cache(spa);
2070		spa_config_exit(spa, SCL_ALL, FTAG);
2071		spa->spa_l2cache.sav_sync = B_TRUE;
2072	}
2073
2074	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
2075	spa->spa_meta_objset = dp->dp_meta_objset;
2076
2077	tx = dmu_tx_create_assigned(dp, txg);
2078
2079	/*
2080	 * Create the pool config object.
2081	 */
2082	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
2083	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
2084	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2085
2086	if (zap_add(spa->spa_meta_objset,
2087	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
2088	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
2089		cmn_err(CE_PANIC, "failed to add pool config");
2090	}
2091
2092	/* Newly created pools with the right version are always deflated. */
2093	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
2094		spa->spa_deflate = TRUE;
2095		if (zap_add(spa->spa_meta_objset,
2096		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
2097		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
2098			cmn_err(CE_PANIC, "failed to add deflate");
2099		}
2100	}
2101
2102	/*
2103	 * Create the deferred-free bplist object.  Turn off compression
2104	 * because sync-to-convergence takes longer if the blocksize
2105	 * keeps changing.
2106	 */
2107	spa->spa_sync_bplist_obj = bplist_create(spa->spa_meta_objset,
2108	    1 << 14, tx);
2109	dmu_object_set_compress(spa->spa_meta_objset, spa->spa_sync_bplist_obj,
2110	    ZIO_COMPRESS_OFF, tx);
2111
2112	if (zap_add(spa->spa_meta_objset,
2113	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
2114	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj, tx) != 0) {
2115		cmn_err(CE_PANIC, "failed to add bplist");
2116	}
2117
2118	/*
2119	 * Create the pool's history object.
2120	 */
2121	if (version >= SPA_VERSION_ZPOOL_HISTORY)
2122		spa_history_create_obj(spa, tx);
2123
2124	/*
2125	 * Set pool properties.
2126	 */
2127	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
2128	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2129	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
2130	if (props != NULL) {
2131		spa_configfile_set(spa, props, B_FALSE);
2132		spa_sync_props(spa, props, CRED(), tx);
2133	}
2134
2135	dmu_tx_commit(tx);
2136
2137	spa->spa_sync_on = B_TRUE;
2138	txg_sync_start(spa->spa_dsl_pool);
2139
2140	/*
2141	 * We explicitly wait for the first transaction to complete so that our
2142	 * bean counters are appropriately updated.
2143	 */
2144	txg_wait_synced(spa->spa_dsl_pool, txg);
2145
2146	spa_config_sync(spa, B_FALSE, B_TRUE);
2147
2148	if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
2149		(void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
2150
2151	spa->spa_minref = refcount_count(&spa->spa_refcount);
2152
2153	mutex_exit(&spa_namespace_lock);
2154
2155	return (0);
2156}
2157
2158/*
2159 * Import the given pool into the system.  We set up the necessary spa_t and
2160 * then call spa_load() to do the dirty work.
2161 */
2162static int
2163spa_import_common(const char *pool, nvlist_t *config, nvlist_t *props,
2164    boolean_t isroot, boolean_t allowfaulted)
2165{
2166	spa_t *spa;
2167	char *altroot = NULL;
2168	int error, loaderr;
2169	nvlist_t *nvroot;
2170	nvlist_t **spares, **l2cache;
2171	uint_t nspares, nl2cache;
2172
2173	/*
2174	 * If a pool with this name exists, return failure.
2175	 */
2176	mutex_enter(&spa_namespace_lock);
2177	if ((spa = spa_lookup(pool)) != NULL) {
2178		if (isroot) {
2179			/*
2180			 * Remove the existing root pool from the
2181			 * namespace so that we can replace it with
2182			 * the correct config we just read in.
2183			 */
2184			ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
2185			spa_remove(spa);
2186		} else {
2187			mutex_exit(&spa_namespace_lock);
2188			return (EEXIST);
2189		}
2190	}
2191
2192	/*
2193	 * Create and initialize the spa structure.
2194	 */
2195	(void) nvlist_lookup_string(props,
2196	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2197	spa = spa_add(pool, altroot);
2198	spa_activate(spa, spa_mode_global);
2199
2200	if (allowfaulted)
2201		spa->spa_import_faulted = B_TRUE;
2202	spa->spa_is_root = isroot;
2203
2204	/*
2205	 * Pass off the heavy lifting to spa_load().
2206	 * Pass TRUE for mosconfig (unless this is a root pool) because
2207	 * the user-supplied config is actually the one to trust when
2208	 * doing an import.
2209	 */
2210	loaderr = error = spa_load(spa, config, SPA_LOAD_IMPORT, !isroot);
2211
2212	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2213	/*
2214	 * Toss any existing sparelist, as it doesn't have any validity anymore,
2215	 * and conflicts with spa_has_spare().
2216	 */
2217	if (!isroot && spa->spa_spares.sav_config) {
2218		nvlist_free(spa->spa_spares.sav_config);
2219		spa->spa_spares.sav_config = NULL;
2220		spa_load_spares(spa);
2221	}
2222	if (!isroot && spa->spa_l2cache.sav_config) {
2223		nvlist_free(spa->spa_l2cache.sav_config);
2224		spa->spa_l2cache.sav_config = NULL;
2225		spa_load_l2cache(spa);
2226	}
2227
2228	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2229	    &nvroot) == 0);
2230	if (error == 0)
2231		error = spa_validate_aux(spa, nvroot, -1ULL, VDEV_ALLOC_SPARE);
2232	if (error == 0)
2233		error = spa_validate_aux(spa, nvroot, -1ULL,
2234		    VDEV_ALLOC_L2CACHE);
2235	spa_config_exit(spa, SCL_ALL, FTAG);
2236
2237	if (props != NULL)
2238		spa_configfile_set(spa, props, B_FALSE);
2239
2240	if (error != 0 || (props && spa_writeable(spa) &&
2241	    (error = spa_prop_set(spa, props)))) {
2242		if (loaderr != 0 && loaderr != EINVAL && allowfaulted) {
2243			/*
2244			 * If we failed to load the pool, but 'allowfaulted' is
2245			 * set, then manually set the config as if the config
2246			 * passed in was specified in the cache file.
2247			 */
2248			error = 0;
2249			spa->spa_import_faulted = B_FALSE;
2250			if (spa->spa_config == NULL)
2251				spa->spa_config = spa_config_generate(spa,
2252				    NULL, -1ULL, B_TRUE);
2253			spa_unload(spa);
2254			spa_deactivate(spa);
2255			spa_config_sync(spa, B_FALSE, B_TRUE);
2256		} else {
2257			spa_unload(spa);
2258			spa_deactivate(spa);
2259			spa_remove(spa);
2260		}
2261		mutex_exit(&spa_namespace_lock);
2262		return (error);
2263	}
2264
2265	/*
2266	 * Override any spares and level 2 cache devices as specified by
2267	 * the user, as these may have correct device names/devids, etc.
2268	 */
2269	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2270	    &spares, &nspares) == 0) {
2271		if (spa->spa_spares.sav_config)
2272			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
2273			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
2274		else
2275			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
2276			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2277		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2278		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2279		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2280		spa_load_spares(spa);
2281		spa_config_exit(spa, SCL_ALL, FTAG);
2282		spa->spa_spares.sav_sync = B_TRUE;
2283	}
2284	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2285	    &l2cache, &nl2cache) == 0) {
2286		if (spa->spa_l2cache.sav_config)
2287			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
2288			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
2289		else
2290			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2291			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2292		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2293		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2294		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2295		spa_load_l2cache(spa);
2296		spa_config_exit(spa, SCL_ALL, FTAG);
2297		spa->spa_l2cache.sav_sync = B_TRUE;
2298	}
2299
2300	if (spa_writeable(spa)) {
2301		/*
2302		 * Update the config cache to include the newly-imported pool.
2303		 */
2304		spa_config_update_common(spa, SPA_CONFIG_UPDATE_POOL, isroot);
2305	}
2306
2307	spa->spa_import_faulted = B_FALSE;
2308	mutex_exit(&spa_namespace_lock);
2309
2310	return (0);
2311}
2312
2313#ifdef _KERNEL
2314/*
2315 * Build a "root" vdev for a top level vdev read in from a rootpool
2316 * device label.
2317 */
2318static void
2319spa_build_rootpool_config(nvlist_t *config)
2320{
2321	nvlist_t *nvtop, *nvroot;
2322	uint64_t pgid;
2323
2324	/*
2325	 * Add this top-level vdev to the child array.
2326	 */
2327	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtop)
2328	    == 0);
2329	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pgid)
2330	    == 0);
2331
2332	/*
2333	 * Put this pool's top-level vdevs into a root vdev.
2334	 */
2335	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2336	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT)
2337	    == 0);
2338	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
2339	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
2340	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2341	    &nvtop, 1) == 0);
2342
2343	/*
2344	 * Replace the existing vdev_tree with the new root vdev in
2345	 * this pool's configuration (remove the old, add the new).
2346	 */
2347	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
2348	nvlist_free(nvroot);
2349}
2350
2351/*
2352 * Get the root pool information from the root disk, then import the root pool
2353 * during the system boot up time.
2354 */
2355extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
2356
2357int
2358spa_check_rootconf(char *devpath, char *devid, nvlist_t **bestconf,
2359    uint64_t *besttxg)
2360{
2361	nvlist_t *config;
2362	uint64_t txg;
2363	int error;
2364
2365	if (error = vdev_disk_read_rootlabel(devpath, devid, &config))
2366		return (error);
2367
2368	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
2369
2370	if (bestconf != NULL)
2371		*bestconf = config;
2372	else
2373		nvlist_free(config);
2374	*besttxg = txg;
2375	return (0);
2376}
2377
2378boolean_t
2379spa_rootdev_validate(nvlist_t *nv)
2380{
2381	uint64_t ival;
2382
2383	if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
2384	    nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
2385	    nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
2386		return (B_FALSE);
2387
2388	return (B_TRUE);
2389}
2390
2391
2392/*
2393 * Given the boot device's physical path or devid, check if the device
2394 * is in a valid state.  If so, return the configuration from the vdev
2395 * label.
2396 */
2397int
2398spa_get_rootconf(char *devpath, char *devid, nvlist_t **bestconf)
2399{
2400	nvlist_t *conf = NULL;
2401	uint64_t txg = 0;
2402	nvlist_t *nvtop, **child;
2403	char *type;
2404	char *bootpath = NULL;
2405	uint_t children, c;
2406	char *tmp;
2407	int error;
2408
2409	if (devpath && ((tmp = strchr(devpath, ' ')) != NULL))
2410		*tmp = '\0';
2411	if (error = spa_check_rootconf(devpath, devid, &conf, &txg)) {
2412		cmn_err(CE_NOTE, "error reading device label");
2413		return (error);
2414	}
2415	if (txg == 0) {
2416		cmn_err(CE_NOTE, "this device is detached");
2417		nvlist_free(conf);
2418		return (EINVAL);
2419	}
2420
2421	VERIFY(nvlist_lookup_nvlist(conf, ZPOOL_CONFIG_VDEV_TREE,
2422	    &nvtop) == 0);
2423	VERIFY(nvlist_lookup_string(nvtop, ZPOOL_CONFIG_TYPE, &type) == 0);
2424
2425	if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2426		if (spa_rootdev_validate(nvtop)) {
2427			goto out;
2428		} else {
2429			nvlist_free(conf);
2430			return (EINVAL);
2431		}
2432	}
2433
2434	ASSERT(strcmp(type, VDEV_TYPE_MIRROR) == 0);
2435
2436	VERIFY(nvlist_lookup_nvlist_array(nvtop, ZPOOL_CONFIG_CHILDREN,
2437	    &child, &children) == 0);
2438
2439	/*
2440	 * Go thru vdevs in the mirror to see if the given device
2441	 * has the most recent txg. Only the device with the most
2442	 * recent txg has valid information and should be booted.
2443	 */
2444	for (c = 0; c < children; c++) {
2445		char *cdevid, *cpath;
2446		uint64_t tmptxg;
2447
2448		cpath = NULL;
2449		cdevid = NULL;
2450		if (nvlist_lookup_string(child[c], ZPOOL_CONFIG_PHYS_PATH,
2451		    &cpath) != 0 && nvlist_lookup_string(child[c],
2452		    ZPOOL_CONFIG_DEVID, &cdevid) != 0)
2453			return (EINVAL);
2454		if ((spa_check_rootconf(cpath, cdevid, NULL,
2455		    &tmptxg) == 0) && (tmptxg > txg)) {
2456			txg = tmptxg;
2457			VERIFY(nvlist_lookup_string(child[c],
2458			    ZPOOL_CONFIG_PATH, &bootpath) == 0);
2459		}
2460	}
2461
2462	/* Does the best device match the one we've booted from? */
2463	if (bootpath) {
2464		cmn_err(CE_NOTE, "try booting from '%s'", bootpath);
2465		return (EINVAL);
2466	}
2467out:
2468	*bestconf = conf;
2469	return (0);
2470}
2471
2472/*
2473 * Import a root pool.
2474 *
2475 * For x86. devpath_list will consist of devid and/or physpath name of
2476 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
2477 * The GRUB "findroot" command will return the vdev we should boot.
2478 *
2479 * For Sparc, devpath_list consists the physpath name of the booting device
2480 * no matter the rootpool is a single device pool or a mirrored pool.
2481 * e.g.
2482 *	"/pci@1f,0/ide@d/disk@0,0:a"
2483 */
2484int
2485spa_import_rootpool(char *devpath, char *devid)
2486{
2487	nvlist_t *conf = NULL;
2488	char *pname;
2489	int error;
2490
2491	/*
2492	 * Get the vdev pathname and configuation from the most
2493	 * recently updated vdev (highest txg).
2494	 */
2495	if (error = spa_get_rootconf(devpath, devid, &conf))
2496		goto msg_out;
2497
2498	/*
2499	 * Add type "root" vdev to the config.
2500	 */
2501	spa_build_rootpool_config(conf);
2502
2503	VERIFY(nvlist_lookup_string(conf, ZPOOL_CONFIG_POOL_NAME, &pname) == 0);
2504
2505	/*
2506	 * We specify 'allowfaulted' for this to be treated like spa_open()
2507	 * instead of spa_import().  This prevents us from marking vdevs as
2508	 * persistently unavailable, and generates FMA ereports as if it were a
2509	 * pool open, not import.
2510	 */
2511	error = spa_import_common(pname, conf, NULL, B_TRUE, B_TRUE);
2512	ASSERT(error != EEXIST);
2513
2514	nvlist_free(conf);
2515	return (error);
2516
2517msg_out:
2518	cmn_err(CE_NOTE, "\n"
2519	    "  ***************************************************  \n"
2520	    "  *  This device is not bootable!                   *  \n"
2521	    "  *  It is either offlined or detached or faulted.  *  \n"
2522	    "  *  Please try to boot from a different device.    *  \n"
2523	    "  ***************************************************  ");
2524
2525	return (error);
2526}
2527#endif
2528
2529/*
2530 * Import a non-root pool into the system.
2531 */
2532int
2533spa_import(const char *pool, nvlist_t *config, nvlist_t *props)
2534{
2535	return (spa_import_common(pool, config, props, B_FALSE, B_FALSE));
2536}
2537
2538int
2539spa_import_faulted(const char *pool, nvlist_t *config, nvlist_t *props)
2540{
2541	return (spa_import_common(pool, config, props, B_FALSE, B_TRUE));
2542}
2543
2544
2545/*
2546 * This (illegal) pool name is used when temporarily importing a spa_t in order
2547 * to get the vdev stats associated with the imported devices.
2548 */
2549#define	TRYIMPORT_NAME	"$import"
2550
2551nvlist_t *
2552spa_tryimport(nvlist_t *tryconfig)
2553{
2554	nvlist_t *config = NULL;
2555	char *poolname;
2556	spa_t *spa;
2557	uint64_t state;
2558	int error;
2559
2560	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
2561		return (NULL);
2562
2563	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
2564		return (NULL);
2565
2566	/*
2567	 * Create and initialize the spa structure.
2568	 */
2569	mutex_enter(&spa_namespace_lock);
2570	spa = spa_add(TRYIMPORT_NAME, NULL);
2571	spa_activate(spa, FREAD);
2572
2573	/*
2574	 * Pass off the heavy lifting to spa_load().
2575	 * Pass TRUE for mosconfig because the user-supplied config
2576	 * is actually the one to trust when doing an import.
2577	 */
2578	error = spa_load(spa, tryconfig, SPA_LOAD_TRYIMPORT, B_TRUE);
2579
2580	/*
2581	 * If 'tryconfig' was at least parsable, return the current config.
2582	 */
2583	if (spa->spa_root_vdev != NULL) {
2584		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2585		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
2586		    poolname) == 0);
2587		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
2588		    state) == 0);
2589		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
2590		    spa->spa_uberblock.ub_timestamp) == 0);
2591
2592		/*
2593		 * If the bootfs property exists on this pool then we
2594		 * copy it out so that external consumers can tell which
2595		 * pools are bootable.
2596		 */
2597		if ((!error || error == EEXIST) && spa->spa_bootfs) {
2598			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2599
2600			/*
2601			 * We have to play games with the name since the
2602			 * pool was opened as TRYIMPORT_NAME.
2603			 */
2604			if (dsl_dsobj_to_dsname(spa_name(spa),
2605			    spa->spa_bootfs, tmpname) == 0) {
2606				char *cp;
2607				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2608
2609				cp = strchr(tmpname, '/');
2610				if (cp == NULL) {
2611					(void) strlcpy(dsname, tmpname,
2612					    MAXPATHLEN);
2613				} else {
2614					(void) snprintf(dsname, MAXPATHLEN,
2615					    "%s/%s", poolname, ++cp);
2616				}
2617				VERIFY(nvlist_add_string(config,
2618				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
2619				kmem_free(dsname, MAXPATHLEN);
2620			}
2621			kmem_free(tmpname, MAXPATHLEN);
2622		}
2623
2624		/*
2625		 * Add the list of hot spares and level 2 cache devices.
2626		 */
2627		spa_add_spares(spa, config);
2628		spa_add_l2cache(spa, config);
2629	}
2630
2631	spa_unload(spa);
2632	spa_deactivate(spa);
2633	spa_remove(spa);
2634	mutex_exit(&spa_namespace_lock);
2635
2636	return (config);
2637}
2638
2639/*
2640 * Pool export/destroy
2641 *
2642 * The act of destroying or exporting a pool is very simple.  We make sure there
2643 * is no more pending I/O and any references to the pool are gone.  Then, we
2644 * update the pool state and sync all the labels to disk, removing the
2645 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
2646 * we don't sync the labels or remove the configuration cache.
2647 */
2648static int
2649spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
2650    boolean_t force, boolean_t hardforce)
2651{
2652	spa_t *spa;
2653
2654	if (oldconfig)
2655		*oldconfig = NULL;
2656
2657	if (!(spa_mode_global & FWRITE))
2658		return (EROFS);
2659
2660	mutex_enter(&spa_namespace_lock);
2661	if ((spa = spa_lookup(pool)) == NULL) {
2662		mutex_exit(&spa_namespace_lock);
2663		return (ENOENT);
2664	}
2665
2666	/*
2667	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
2668	 * reacquire the namespace lock, and see if we can export.
2669	 */
2670	spa_open_ref(spa, FTAG);
2671	mutex_exit(&spa_namespace_lock);
2672	spa_async_suspend(spa);
2673	mutex_enter(&spa_namespace_lock);
2674	spa_close(spa, FTAG);
2675
2676	/*
2677	 * The pool will be in core if it's openable,
2678	 * in which case we can modify its state.
2679	 */
2680	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
2681		/*
2682		 * Objsets may be open only because they're dirty, so we
2683		 * have to force it to sync before checking spa_refcnt.
2684		 */
2685		txg_wait_synced(spa->spa_dsl_pool, 0);
2686
2687		/*
2688		 * A pool cannot be exported or destroyed if there are active
2689		 * references.  If we are resetting a pool, allow references by
2690		 * fault injection handlers.
2691		 */
2692		if (!spa_refcount_zero(spa) ||
2693		    (spa->spa_inject_ref != 0 &&
2694		    new_state != POOL_STATE_UNINITIALIZED)) {
2695			spa_async_resume(spa);
2696			mutex_exit(&spa_namespace_lock);
2697			return (EBUSY);
2698		}
2699
2700		/*
2701		 * A pool cannot be exported if it has an active shared spare.
2702		 * This is to prevent other pools stealing the active spare
2703		 * from an exported pool. At user's own will, such pool can
2704		 * be forcedly exported.
2705		 */
2706		if (!force && new_state == POOL_STATE_EXPORTED &&
2707		    spa_has_active_shared_spare(spa)) {
2708			spa_async_resume(spa);
2709			mutex_exit(&spa_namespace_lock);
2710			return (EXDEV);
2711		}
2712
2713		/*
2714		 * We want this to be reflected on every label,
2715		 * so mark them all dirty.  spa_unload() will do the
2716		 * final sync that pushes these changes out.
2717		 */
2718		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
2719			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2720			spa->spa_state = new_state;
2721			spa->spa_final_txg = spa_last_synced_txg(spa) + 1;
2722			vdev_config_dirty(spa->spa_root_vdev);
2723			spa_config_exit(spa, SCL_ALL, FTAG);
2724		}
2725	}
2726
2727	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
2728
2729	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
2730		spa_unload(spa);
2731		spa_deactivate(spa);
2732	}
2733
2734	if (oldconfig && spa->spa_config)
2735		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
2736
2737	if (new_state != POOL_STATE_UNINITIALIZED) {
2738		if (!hardforce)
2739			spa_config_sync(spa, B_TRUE, B_TRUE);
2740		spa_remove(spa);
2741	}
2742	mutex_exit(&spa_namespace_lock);
2743
2744	return (0);
2745}
2746
2747/*
2748 * Destroy a storage pool.
2749 */
2750int
2751spa_destroy(char *pool)
2752{
2753	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
2754	    B_FALSE, B_FALSE));
2755}
2756
2757/*
2758 * Export a storage pool.
2759 */
2760int
2761spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
2762    boolean_t hardforce)
2763{
2764	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
2765	    force, hardforce));
2766}
2767
2768/*
2769 * Similar to spa_export(), this unloads the spa_t without actually removing it
2770 * from the namespace in any way.
2771 */
2772int
2773spa_reset(char *pool)
2774{
2775	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
2776	    B_FALSE, B_FALSE));
2777}
2778
2779/*
2780 * ==========================================================================
2781 * Device manipulation
2782 * ==========================================================================
2783 */
2784
2785/*
2786 * Add a device to a storage pool.
2787 */
2788int
2789spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
2790{
2791	uint64_t txg;
2792	int error;
2793	vdev_t *rvd = spa->spa_root_vdev;
2794	vdev_t *vd, *tvd;
2795	nvlist_t **spares, **l2cache;
2796	uint_t nspares, nl2cache;
2797
2798	txg = spa_vdev_enter(spa);
2799
2800	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
2801	    VDEV_ALLOC_ADD)) != 0)
2802		return (spa_vdev_exit(spa, NULL, txg, error));
2803
2804	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
2805
2806	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
2807	    &nspares) != 0)
2808		nspares = 0;
2809
2810	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
2811	    &nl2cache) != 0)
2812		nl2cache = 0;
2813
2814	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
2815		return (spa_vdev_exit(spa, vd, txg, EINVAL));
2816
2817	if (vd->vdev_children != 0 &&
2818	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
2819		return (spa_vdev_exit(spa, vd, txg, error));
2820
2821	/*
2822	 * We must validate the spares and l2cache devices after checking the
2823	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
2824	 */
2825	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
2826		return (spa_vdev_exit(spa, vd, txg, error));
2827
2828	/*
2829	 * Transfer each new top-level vdev from vd to rvd.
2830	 */
2831	for (int c = 0; c < vd->vdev_children; c++) {
2832		tvd = vd->vdev_child[c];
2833		vdev_remove_child(vd, tvd);
2834		tvd->vdev_id = rvd->vdev_children;
2835		vdev_add_child(rvd, tvd);
2836		vdev_config_dirty(tvd);
2837	}
2838
2839	if (nspares != 0) {
2840		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
2841		    ZPOOL_CONFIG_SPARES);
2842		spa_load_spares(spa);
2843		spa->spa_spares.sav_sync = B_TRUE;
2844	}
2845
2846	if (nl2cache != 0) {
2847		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
2848		    ZPOOL_CONFIG_L2CACHE);
2849		spa_load_l2cache(spa);
2850		spa->spa_l2cache.sav_sync = B_TRUE;
2851	}
2852
2853	/*
2854	 * We have to be careful when adding new vdevs to an existing pool.
2855	 * If other threads start allocating from these vdevs before we
2856	 * sync the config cache, and we lose power, then upon reboot we may
2857	 * fail to open the pool because there are DVAs that the config cache
2858	 * can't translate.  Therefore, we first add the vdevs without
2859	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
2860	 * and then let spa_config_update() initialize the new metaslabs.
2861	 *
2862	 * spa_load() checks for added-but-not-initialized vdevs, so that
2863	 * if we lose power at any point in this sequence, the remaining
2864	 * steps will be completed the next time we load the pool.
2865	 */
2866	(void) spa_vdev_exit(spa, vd, txg, 0);
2867
2868	mutex_enter(&spa_namespace_lock);
2869	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
2870	mutex_exit(&spa_namespace_lock);
2871
2872	return (0);
2873}
2874
2875/*
2876 * Attach a device to a mirror.  The arguments are the path to any device
2877 * in the mirror, and the nvroot for the new device.  If the path specifies
2878 * a device that is not mirrored, we automatically insert the mirror vdev.
2879 *
2880 * If 'replacing' is specified, the new device is intended to replace the
2881 * existing device; in this case the two devices are made into their own
2882 * mirror using the 'replacing' vdev, which is functionally identical to
2883 * the mirror vdev (it actually reuses all the same ops) but has a few
2884 * extra rules: you can't attach to it after it's been created, and upon
2885 * completion of resilvering, the first disk (the one being replaced)
2886 * is automatically detached.
2887 */
2888int
2889spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
2890{
2891	uint64_t txg, open_txg;
2892	vdev_t *rvd = spa->spa_root_vdev;
2893	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
2894	vdev_ops_t *pvops;
2895	dmu_tx_t *tx;
2896	char *oldvdpath, *newvdpath;
2897	int newvd_isspare;
2898	int error;
2899
2900	txg = spa_vdev_enter(spa);
2901
2902	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
2903
2904	if (oldvd == NULL)
2905		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
2906
2907	if (!oldvd->vdev_ops->vdev_op_leaf)
2908		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
2909
2910	pvd = oldvd->vdev_parent;
2911
2912	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
2913	    VDEV_ALLOC_ADD)) != 0)
2914		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
2915
2916	if (newrootvd->vdev_children != 1)
2917		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
2918
2919	newvd = newrootvd->vdev_child[0];
2920
2921	if (!newvd->vdev_ops->vdev_op_leaf)
2922		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
2923
2924	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
2925		return (spa_vdev_exit(spa, newrootvd, txg, error));
2926
2927	/*
2928	 * Spares can't replace logs
2929	 */
2930	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
2931		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2932
2933	if (!replacing) {
2934		/*
2935		 * For attach, the only allowable parent is a mirror or the root
2936		 * vdev.
2937		 */
2938		if (pvd->vdev_ops != &vdev_mirror_ops &&
2939		    pvd->vdev_ops != &vdev_root_ops)
2940			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2941
2942		pvops = &vdev_mirror_ops;
2943	} else {
2944		/*
2945		 * Active hot spares can only be replaced by inactive hot
2946		 * spares.
2947		 */
2948		if (pvd->vdev_ops == &vdev_spare_ops &&
2949		    pvd->vdev_child[1] == oldvd &&
2950		    !spa_has_spare(spa, newvd->vdev_guid))
2951			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2952
2953		/*
2954		 * If the source is a hot spare, and the parent isn't already a
2955		 * spare, then we want to create a new hot spare.  Otherwise, we
2956		 * want to create a replacing vdev.  The user is not allowed to
2957		 * attach to a spared vdev child unless the 'isspare' state is
2958		 * the same (spare replaces spare, non-spare replaces
2959		 * non-spare).
2960		 */
2961		if (pvd->vdev_ops == &vdev_replacing_ops)
2962			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2963		else if (pvd->vdev_ops == &vdev_spare_ops &&
2964		    newvd->vdev_isspare != oldvd->vdev_isspare)
2965			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2966		else if (pvd->vdev_ops != &vdev_spare_ops &&
2967		    newvd->vdev_isspare)
2968			pvops = &vdev_spare_ops;
2969		else
2970			pvops = &vdev_replacing_ops;
2971	}
2972
2973	/*
2974	 * Compare the new device size with the replaceable/attachable
2975	 * device size.
2976	 */
2977	if (newvd->vdev_psize < vdev_get_rsize(oldvd))
2978		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
2979
2980	/*
2981	 * The new device cannot have a higher alignment requirement
2982	 * than the top-level vdev.
2983	 */
2984	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
2985		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
2986
2987	/*
2988	 * If this is an in-place replacement, update oldvd's path and devid
2989	 * to make it distinguishable from newvd, and unopenable from now on.
2990	 */
2991	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
2992		spa_strfree(oldvd->vdev_path);
2993		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
2994		    KM_SLEEP);
2995		(void) sprintf(oldvd->vdev_path, "%s/%s",
2996		    newvd->vdev_path, "old");
2997		if (oldvd->vdev_devid != NULL) {
2998			spa_strfree(oldvd->vdev_devid);
2999			oldvd->vdev_devid = NULL;
3000		}
3001	}
3002
3003	/*
3004	 * If the parent is not a mirror, or if we're replacing, insert the new
3005	 * mirror/replacing/spare vdev above oldvd.
3006	 */
3007	if (pvd->vdev_ops != pvops)
3008		pvd = vdev_add_parent(oldvd, pvops);
3009
3010	ASSERT(pvd->vdev_top->vdev_parent == rvd);
3011	ASSERT(pvd->vdev_ops == pvops);
3012	ASSERT(oldvd->vdev_parent == pvd);
3013
3014	/*
3015	 * Extract the new device from its root and add it to pvd.
3016	 */
3017	vdev_remove_child(newrootvd, newvd);
3018	newvd->vdev_id = pvd->vdev_children;
3019	vdev_add_child(pvd, newvd);
3020
3021	/*
3022	 * If newvd is smaller than oldvd, but larger than its rsize,
3023	 * the addition of newvd may have decreased our parent's asize.
3024	 */
3025	pvd->vdev_asize = MIN(pvd->vdev_asize, newvd->vdev_asize);
3026
3027	tvd = newvd->vdev_top;
3028	ASSERT(pvd->vdev_top == tvd);
3029	ASSERT(tvd->vdev_parent == rvd);
3030
3031	vdev_config_dirty(tvd);
3032
3033	/*
3034	 * Set newvd's DTL to [TXG_INITIAL, open_txg].  It will propagate
3035	 * upward when spa_vdev_exit() calls vdev_dtl_reassess().
3036	 */
3037	open_txg = txg + TXG_CONCURRENT_STATES - 1;
3038
3039	vdev_dtl_dirty(newvd, DTL_MISSING,
3040	    TXG_INITIAL, open_txg - TXG_INITIAL + 1);
3041
3042	if (newvd->vdev_isspare)
3043		spa_spare_activate(newvd);
3044	oldvdpath = spa_strdup(oldvd->vdev_path);
3045	newvdpath = spa_strdup(newvd->vdev_path);
3046	newvd_isspare = newvd->vdev_isspare;
3047
3048	/*
3049	 * Mark newvd's DTL dirty in this txg.
3050	 */
3051	vdev_dirty(tvd, VDD_DTL, newvd, txg);
3052
3053	(void) spa_vdev_exit(spa, newrootvd, open_txg, 0);
3054
3055	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
3056	if (dmu_tx_assign(tx, TXG_WAIT) == 0) {
3057		spa_history_internal_log(LOG_POOL_VDEV_ATTACH, spa, tx,
3058		    CRED(),  "%s vdev=%s %s vdev=%s",
3059		    replacing && newvd_isspare ? "spare in" :
3060		    replacing ? "replace" : "attach", newvdpath,
3061		    replacing ? "for" : "to", oldvdpath);
3062		dmu_tx_commit(tx);
3063	} else {
3064		dmu_tx_abort(tx);
3065	}
3066
3067	spa_strfree(oldvdpath);
3068	spa_strfree(newvdpath);
3069
3070	/*
3071	 * Kick off a resilver to update newvd.
3072	 */
3073	VERIFY3U(spa_scrub(spa, POOL_SCRUB_RESILVER), ==, 0);
3074
3075	return (0);
3076}
3077
3078/*
3079 * Detach a device from a mirror or replacing vdev.
3080 * If 'replace_done' is specified, only detach if the parent
3081 * is a replacing vdev.
3082 */
3083int
3084spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
3085{
3086	uint64_t txg;
3087	int error;
3088	vdev_t *rvd = spa->spa_root_vdev;
3089	vdev_t *vd, *pvd, *cvd, *tvd;
3090	boolean_t unspare = B_FALSE;
3091	uint64_t unspare_guid;
3092	size_t len;
3093
3094	txg = spa_vdev_enter(spa);
3095
3096	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3097
3098	if (vd == NULL)
3099		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3100
3101	if (!vd->vdev_ops->vdev_op_leaf)
3102		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3103
3104	pvd = vd->vdev_parent;
3105
3106	/*
3107	 * If the parent/child relationship is not as expected, don't do it.
3108	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
3109	 * vdev that's replacing B with C.  The user's intent in replacing
3110	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
3111	 * the replace by detaching C, the expected behavior is to end up
3112	 * M(A,B).  But suppose that right after deciding to detach C,
3113	 * the replacement of B completes.  We would have M(A,C), and then
3114	 * ask to detach C, which would leave us with just A -- not what
3115	 * the user wanted.  To prevent this, we make sure that the
3116	 * parent/child relationship hasn't changed -- in this example,
3117	 * that C's parent is still the replacing vdev R.
3118	 */
3119	if (pvd->vdev_guid != pguid && pguid != 0)
3120		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3121
3122	/*
3123	 * If replace_done is specified, only remove this device if it's
3124	 * the first child of a replacing vdev.  For the 'spare' vdev, either
3125	 * disk can be removed.
3126	 */
3127	if (replace_done) {
3128		if (pvd->vdev_ops == &vdev_replacing_ops) {
3129			if (vd->vdev_id != 0)
3130				return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3131		} else if (pvd->vdev_ops != &vdev_spare_ops) {
3132			return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3133		}
3134	}
3135
3136	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
3137	    spa_version(spa) >= SPA_VERSION_SPARES);
3138
3139	/*
3140	 * Only mirror, replacing, and spare vdevs support detach.
3141	 */
3142	if (pvd->vdev_ops != &vdev_replacing_ops &&
3143	    pvd->vdev_ops != &vdev_mirror_ops &&
3144	    pvd->vdev_ops != &vdev_spare_ops)
3145		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3146
3147	/*
3148	 * If this device has the only valid copy of some data,
3149	 * we cannot safely detach it.
3150	 */
3151	if (vdev_dtl_required(vd))
3152		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3153
3154	ASSERT(pvd->vdev_children >= 2);
3155
3156	/*
3157	 * If we are detaching the second disk from a replacing vdev, then
3158	 * check to see if we changed the original vdev's path to have "/old"
3159	 * at the end in spa_vdev_attach().  If so, undo that change now.
3160	 */
3161	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id == 1 &&
3162	    pvd->vdev_child[0]->vdev_path != NULL &&
3163	    pvd->vdev_child[1]->vdev_path != NULL) {
3164		ASSERT(pvd->vdev_child[1] == vd);
3165		cvd = pvd->vdev_child[0];
3166		len = strlen(vd->vdev_path);
3167		if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
3168		    strcmp(cvd->vdev_path + len, "/old") == 0) {
3169			spa_strfree(cvd->vdev_path);
3170			cvd->vdev_path = spa_strdup(vd->vdev_path);
3171		}
3172	}
3173
3174	/*
3175	 * If we are detaching the original disk from a spare, then it implies
3176	 * that the spare should become a real disk, and be removed from the
3177	 * active spare list for the pool.
3178	 */
3179	if (pvd->vdev_ops == &vdev_spare_ops &&
3180	    vd->vdev_id == 0 && pvd->vdev_child[1]->vdev_isspare)
3181		unspare = B_TRUE;
3182
3183	/*
3184	 * Erase the disk labels so the disk can be used for other things.
3185	 * This must be done after all other error cases are handled,
3186	 * but before we disembowel vd (so we can still do I/O to it).
3187	 * But if we can't do it, don't treat the error as fatal --
3188	 * it may be that the unwritability of the disk is the reason
3189	 * it's being detached!
3190	 */
3191	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
3192
3193	/*
3194	 * Remove vd from its parent and compact the parent's children.
3195	 */
3196	vdev_remove_child(pvd, vd);
3197	vdev_compact_children(pvd);
3198
3199	/*
3200	 * Remember one of the remaining children so we can get tvd below.
3201	 */
3202	cvd = pvd->vdev_child[0];
3203
3204	/*
3205	 * If we need to remove the remaining child from the list of hot spares,
3206	 * do it now, marking the vdev as no longer a spare in the process.
3207	 * We must do this before vdev_remove_parent(), because that can
3208	 * change the GUID if it creates a new toplevel GUID.  For a similar
3209	 * reason, we must remove the spare now, in the same txg as the detach;
3210	 * otherwise someone could attach a new sibling, change the GUID, and
3211	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
3212	 */
3213	if (unspare) {
3214		ASSERT(cvd->vdev_isspare);
3215		spa_spare_remove(cvd);
3216		unspare_guid = cvd->vdev_guid;
3217		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
3218	}
3219
3220	/*
3221	 * If the parent mirror/replacing vdev only has one child,
3222	 * the parent is no longer needed.  Remove it from the tree.
3223	 */
3224	if (pvd->vdev_children == 1)
3225		vdev_remove_parent(cvd);
3226
3227	/*
3228	 * We don't set tvd until now because the parent we just removed
3229	 * may have been the previous top-level vdev.
3230	 */
3231	tvd = cvd->vdev_top;
3232	ASSERT(tvd->vdev_parent == rvd);
3233
3234	/*
3235	 * Reevaluate the parent vdev state.
3236	 */
3237	vdev_propagate_state(cvd);
3238
3239	/*
3240	 * If the device we just detached was smaller than the others, it may be
3241	 * possible to add metaslabs (i.e. grow the pool).  vdev_metaslab_init()
3242	 * can't fail because the existing metaslabs are already in core, so
3243	 * there's nothing to read from disk.
3244	 */
3245	VERIFY(vdev_metaslab_init(tvd, txg) == 0);
3246
3247	vdev_config_dirty(tvd);
3248
3249	/*
3250	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
3251	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
3252	 * But first make sure we're not on any *other* txg's DTL list, to
3253	 * prevent vd from being accessed after it's freed.
3254	 */
3255	for (int t = 0; t < TXG_SIZE; t++)
3256		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
3257	vd->vdev_detached = B_TRUE;
3258	vdev_dirty(tvd, VDD_DTL, vd, txg);
3259
3260	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
3261
3262	error = spa_vdev_exit(spa, vd, txg, 0);
3263
3264	/*
3265	 * If this was the removal of the original device in a hot spare vdev,
3266	 * then we want to go through and remove the device from the hot spare
3267	 * list of every other pool.
3268	 */
3269	if (unspare) {
3270		spa_t *myspa = spa;
3271		spa = NULL;
3272		mutex_enter(&spa_namespace_lock);
3273		while ((spa = spa_next(spa)) != NULL) {
3274			if (spa->spa_state != POOL_STATE_ACTIVE)
3275				continue;
3276			if (spa == myspa)
3277				continue;
3278			spa_open_ref(spa, FTAG);
3279			mutex_exit(&spa_namespace_lock);
3280			(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
3281			mutex_enter(&spa_namespace_lock);
3282			spa_close(spa, FTAG);
3283		}
3284		mutex_exit(&spa_namespace_lock);
3285	}
3286
3287	return (error);
3288}
3289
3290static nvlist_t *
3291spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
3292{
3293	for (int i = 0; i < count; i++) {
3294		uint64_t guid;
3295
3296		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
3297		    &guid) == 0);
3298
3299		if (guid == target_guid)
3300			return (nvpp[i]);
3301	}
3302
3303	return (NULL);
3304}
3305
3306static void
3307spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
3308	nvlist_t *dev_to_remove)
3309{
3310	nvlist_t **newdev = NULL;
3311
3312	if (count > 1)
3313		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
3314
3315	for (int i = 0, j = 0; i < count; i++) {
3316		if (dev[i] == dev_to_remove)
3317			continue;
3318		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
3319	}
3320
3321	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
3322	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
3323
3324	for (int i = 0; i < count - 1; i++)
3325		nvlist_free(newdev[i]);
3326
3327	if (count > 1)
3328		kmem_free(newdev, (count - 1) * sizeof (void *));
3329}
3330
3331/*
3332 * Remove a device from the pool.  Currently, this supports removing only hot
3333 * spares and level 2 ARC devices.
3334 */
3335int
3336spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
3337{
3338	vdev_t *vd;
3339	nvlist_t **spares, **l2cache, *nv;
3340	uint_t nspares, nl2cache;
3341	uint64_t txg = 0;
3342	int error = 0;
3343	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
3344
3345	if (!locked)
3346		txg = spa_vdev_enter(spa);
3347
3348	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3349
3350	if (spa->spa_spares.sav_vdevs != NULL &&
3351	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3352	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
3353	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
3354		/*
3355		 * Only remove the hot spare if it's not currently in use
3356		 * in this pool.
3357		 */
3358		if (vd == NULL || unspare) {
3359			spa_vdev_remove_aux(spa->spa_spares.sav_config,
3360			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
3361			spa_load_spares(spa);
3362			spa->spa_spares.sav_sync = B_TRUE;
3363		} else {
3364			error = EBUSY;
3365		}
3366	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
3367	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3368	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
3369	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
3370		/*
3371		 * Cache devices can always be removed.
3372		 */
3373		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
3374		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
3375		spa_load_l2cache(spa);
3376		spa->spa_l2cache.sav_sync = B_TRUE;
3377	} else if (vd != NULL) {
3378		/*
3379		 * Normal vdevs cannot be removed (yet).
3380		 */
3381		error = ENOTSUP;
3382	} else {
3383		/*
3384		 * There is no vdev of any kind with the specified guid.
3385		 */
3386		error = ENOENT;
3387	}
3388
3389	if (!locked)
3390		return (spa_vdev_exit(spa, NULL, txg, error));
3391
3392	return (error);
3393}
3394
3395/*
3396 * Find any device that's done replacing, or a vdev marked 'unspare' that's
3397 * current spared, so we can detach it.
3398 */
3399static vdev_t *
3400spa_vdev_resilver_done_hunt(vdev_t *vd)
3401{
3402	vdev_t *newvd, *oldvd;
3403	int c;
3404
3405	for (c = 0; c < vd->vdev_children; c++) {
3406		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
3407		if (oldvd != NULL)
3408			return (oldvd);
3409	}
3410
3411	/*
3412	 * Check for a completed replacement.
3413	 */
3414	if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
3415		oldvd = vd->vdev_child[0];
3416		newvd = vd->vdev_child[1];
3417
3418		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
3419		    !vdev_dtl_required(oldvd))
3420			return (oldvd);
3421	}
3422
3423	/*
3424	 * Check for a completed resilver with the 'unspare' flag set.
3425	 */
3426	if (vd->vdev_ops == &vdev_spare_ops && vd->vdev_children == 2) {
3427		newvd = vd->vdev_child[0];
3428		oldvd = vd->vdev_child[1];
3429
3430		if (newvd->vdev_unspare &&
3431		    vdev_dtl_empty(newvd, DTL_MISSING) &&
3432		    !vdev_dtl_required(oldvd)) {
3433			newvd->vdev_unspare = 0;
3434			return (oldvd);
3435		}
3436	}
3437
3438	return (NULL);
3439}
3440
3441static void
3442spa_vdev_resilver_done(spa_t *spa)
3443{
3444	vdev_t *vd, *pvd, *ppvd;
3445	uint64_t guid, sguid, pguid, ppguid;
3446
3447	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3448
3449	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
3450		pvd = vd->vdev_parent;
3451		ppvd = pvd->vdev_parent;
3452		guid = vd->vdev_guid;
3453		pguid = pvd->vdev_guid;
3454		ppguid = ppvd->vdev_guid;
3455		sguid = 0;
3456		/*
3457		 * If we have just finished replacing a hot spared device, then
3458		 * we need to detach the parent's first child (the original hot
3459		 * spare) as well.
3460		 */
3461		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0) {
3462			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
3463			ASSERT(ppvd->vdev_children == 2);
3464			sguid = ppvd->vdev_child[1]->vdev_guid;
3465		}
3466		spa_config_exit(spa, SCL_ALL, FTAG);
3467		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
3468			return;
3469		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
3470			return;
3471		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3472	}
3473
3474	spa_config_exit(spa, SCL_ALL, FTAG);
3475}
3476
3477/*
3478 * Update the stored path for this vdev.  Dirty the vdev configuration, relying
3479 * on spa_vdev_enter/exit() to synchronize the labels and cache.
3480 */
3481int
3482spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
3483{
3484	vdev_t *vd;
3485	uint64_t txg;
3486
3487	txg = spa_vdev_enter(spa);
3488
3489	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) {
3490		/*
3491		 * Determine if this is a reference to a hot spare device.  If
3492		 * it is, update the path manually as there is no associated
3493		 * vdev_t that can be synced to disk.
3494		 */
3495		nvlist_t **spares;
3496		uint_t i, nspares;
3497
3498		if (spa->spa_spares.sav_config != NULL) {
3499			VERIFY(nvlist_lookup_nvlist_array(
3500			    spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
3501			    &spares, &nspares) == 0);
3502			for (i = 0; i < nspares; i++) {
3503				uint64_t theguid;
3504				VERIFY(nvlist_lookup_uint64(spares[i],
3505				    ZPOOL_CONFIG_GUID, &theguid) == 0);
3506				if (theguid == guid) {
3507					VERIFY(nvlist_add_string(spares[i],
3508					    ZPOOL_CONFIG_PATH, newpath) == 0);
3509					spa_load_spares(spa);
3510					spa->spa_spares.sav_sync = B_TRUE;
3511					return (spa_vdev_exit(spa, NULL, txg,
3512					    0));
3513				}
3514			}
3515		}
3516
3517		return (spa_vdev_exit(spa, NULL, txg, ENOENT));
3518	}
3519
3520	if (!vd->vdev_ops->vdev_op_leaf)
3521		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3522
3523	spa_strfree(vd->vdev_path);
3524	vd->vdev_path = spa_strdup(newpath);
3525
3526	vdev_config_dirty(vd->vdev_top);
3527
3528	return (spa_vdev_exit(spa, NULL, txg, 0));
3529}
3530
3531/*
3532 * ==========================================================================
3533 * SPA Scrubbing
3534 * ==========================================================================
3535 */
3536
3537int
3538spa_scrub(spa_t *spa, pool_scrub_type_t type)
3539{
3540	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
3541
3542	if ((uint_t)type >= POOL_SCRUB_TYPES)
3543		return (ENOTSUP);
3544
3545	/*
3546	 * If a resilver was requested, but there is no DTL on a
3547	 * writeable leaf device, we have nothing to do.
3548	 */
3549	if (type == POOL_SCRUB_RESILVER &&
3550	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
3551		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
3552		return (0);
3553	}
3554
3555	if (type == POOL_SCRUB_EVERYTHING &&
3556	    spa->spa_dsl_pool->dp_scrub_func != SCRUB_FUNC_NONE &&
3557	    spa->spa_dsl_pool->dp_scrub_isresilver)
3558		return (EBUSY);
3559
3560	if (type == POOL_SCRUB_EVERYTHING || type == POOL_SCRUB_RESILVER) {
3561		return (dsl_pool_scrub_clean(spa->spa_dsl_pool));
3562	} else if (type == POOL_SCRUB_NONE) {
3563		return (dsl_pool_scrub_cancel(spa->spa_dsl_pool));
3564	} else {
3565		return (EINVAL);
3566	}
3567}
3568
3569/*
3570 * ==========================================================================
3571 * SPA async task processing
3572 * ==========================================================================
3573 */
3574
3575static void
3576spa_async_remove(spa_t *spa, vdev_t *vd)
3577{
3578	if (vd->vdev_remove_wanted) {
3579		vd->vdev_remove_wanted = 0;
3580		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
3581		vdev_clear(spa, vd);
3582		vdev_state_dirty(vd->vdev_top);
3583	}
3584
3585	for (int c = 0; c < vd->vdev_children; c++)
3586		spa_async_remove(spa, vd->vdev_child[c]);
3587}
3588
3589static void
3590spa_async_probe(spa_t *spa, vdev_t *vd)
3591{
3592	if (vd->vdev_probe_wanted) {
3593		vd->vdev_probe_wanted = 0;
3594		vdev_reopen(vd);	/* vdev_open() does the actual probe */
3595	}
3596
3597	for (int c = 0; c < vd->vdev_children; c++)
3598		spa_async_probe(spa, vd->vdev_child[c]);
3599}
3600
3601static void
3602spa_async_thread(spa_t *spa)
3603{
3604	int tasks;
3605
3606	ASSERT(spa->spa_sync_on);
3607
3608	mutex_enter(&spa->spa_async_lock);
3609	tasks = spa->spa_async_tasks;
3610	spa->spa_async_tasks = 0;
3611	mutex_exit(&spa->spa_async_lock);
3612
3613	/*
3614	 * See if the config needs to be updated.
3615	 */
3616	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
3617		mutex_enter(&spa_namespace_lock);
3618		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3619		mutex_exit(&spa_namespace_lock);
3620	}
3621
3622	/*
3623	 * See if any devices need to be marked REMOVED.
3624	 */
3625	if (tasks & SPA_ASYNC_REMOVE) {
3626		spa_vdev_state_enter(spa);
3627		spa_async_remove(spa, spa->spa_root_vdev);
3628		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
3629			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
3630		for (int i = 0; i < spa->spa_spares.sav_count; i++)
3631			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
3632		(void) spa_vdev_state_exit(spa, NULL, 0);
3633	}
3634
3635	/*
3636	 * See if any devices need to be probed.
3637	 */
3638	if (tasks & SPA_ASYNC_PROBE) {
3639		spa_vdev_state_enter(spa);
3640		spa_async_probe(spa, spa->spa_root_vdev);
3641		(void) spa_vdev_state_exit(spa, NULL, 0);
3642	}
3643
3644	/*
3645	 * If any devices are done replacing, detach them.
3646	 */
3647	if (tasks & SPA_ASYNC_RESILVER_DONE)
3648		spa_vdev_resilver_done(spa);
3649
3650	/*
3651	 * Kick off a resilver.
3652	 */
3653	if (tasks & SPA_ASYNC_RESILVER)
3654		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER) == 0);
3655
3656	/*
3657	 * Let the world know that we're done.
3658	 */
3659	mutex_enter(&spa->spa_async_lock);
3660	spa->spa_async_thread = NULL;
3661	cv_broadcast(&spa->spa_async_cv);
3662	mutex_exit(&spa->spa_async_lock);
3663	thread_exit();
3664}
3665
3666void
3667spa_async_suspend(spa_t *spa)
3668{
3669	mutex_enter(&spa->spa_async_lock);
3670	spa->spa_async_suspended++;
3671	while (spa->spa_async_thread != NULL)
3672		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
3673	mutex_exit(&spa->spa_async_lock);
3674}
3675
3676void
3677spa_async_resume(spa_t *spa)
3678{
3679	mutex_enter(&spa->spa_async_lock);
3680	ASSERT(spa->spa_async_suspended != 0);
3681	spa->spa_async_suspended--;
3682	mutex_exit(&spa->spa_async_lock);
3683}
3684
3685static void
3686spa_async_dispatch(spa_t *spa)
3687{
3688	mutex_enter(&spa->spa_async_lock);
3689	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
3690	    spa->spa_async_thread == NULL &&
3691	    rootdir != NULL && !vn_is_readonly(rootdir))
3692		spa->spa_async_thread = thread_create(NULL, 0,
3693		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
3694	mutex_exit(&spa->spa_async_lock);
3695}
3696
3697void
3698spa_async_request(spa_t *spa, int task)
3699{
3700	mutex_enter(&spa->spa_async_lock);
3701	spa->spa_async_tasks |= task;
3702	mutex_exit(&spa->spa_async_lock);
3703}
3704
3705/*
3706 * ==========================================================================
3707 * SPA syncing routines
3708 * ==========================================================================
3709 */
3710
3711static void
3712spa_sync_deferred_frees(spa_t *spa, uint64_t txg)
3713{
3714	bplist_t *bpl = &spa->spa_sync_bplist;
3715	dmu_tx_t *tx;
3716	blkptr_t blk;
3717	uint64_t itor = 0;
3718	zio_t *zio;
3719	int error;
3720	uint8_t c = 1;
3721
3722	zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
3723
3724	while (bplist_iterate(bpl, &itor, &blk) == 0) {
3725		ASSERT(blk.blk_birth < txg);
3726		zio_nowait(zio_free(zio, spa, txg, &blk, NULL, NULL,
3727		    ZIO_FLAG_MUSTSUCCEED));
3728	}
3729
3730	error = zio_wait(zio);
3731	ASSERT3U(error, ==, 0);
3732
3733	tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3734	bplist_vacate(bpl, tx);
3735
3736	/*
3737	 * Pre-dirty the first block so we sync to convergence faster.
3738	 * (Usually only the first block is needed.)
3739	 */
3740	dmu_write(spa->spa_meta_objset, spa->spa_sync_bplist_obj, 0, 1, &c, tx);
3741	dmu_tx_commit(tx);
3742}
3743
3744static void
3745spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
3746{
3747	char *packed = NULL;
3748	size_t bufsize;
3749	size_t nvsize = 0;
3750	dmu_buf_t *db;
3751
3752	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
3753
3754	/*
3755	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
3756	 * information.  This avoids the dbuf_will_dirty() path and
3757	 * saves us a pre-read to get data we don't actually care about.
3758	 */
3759	bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
3760	packed = kmem_alloc(bufsize, KM_SLEEP);
3761
3762	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
3763	    KM_SLEEP) == 0);
3764	bzero(packed + nvsize, bufsize - nvsize);
3765
3766	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
3767
3768	kmem_free(packed, bufsize);
3769
3770	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
3771	dmu_buf_will_dirty(db, tx);
3772	*(uint64_t *)db->db_data = nvsize;
3773	dmu_buf_rele(db, FTAG);
3774}
3775
3776static void
3777spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
3778    const char *config, const char *entry)
3779{
3780	nvlist_t *nvroot;
3781	nvlist_t **list;
3782	int i;
3783
3784	if (!sav->sav_sync)
3785		return;
3786
3787	/*
3788	 * Update the MOS nvlist describing the list of available devices.
3789	 * spa_validate_aux() will have already made sure this nvlist is
3790	 * valid and the vdevs are labeled appropriately.
3791	 */
3792	if (sav->sav_object == 0) {
3793		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
3794		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
3795		    sizeof (uint64_t), tx);
3796		VERIFY(zap_update(spa->spa_meta_objset,
3797		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
3798		    &sav->sav_object, tx) == 0);
3799	}
3800
3801	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3802	if (sav->sav_count == 0) {
3803		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
3804	} else {
3805		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
3806		for (i = 0; i < sav->sav_count; i++)
3807			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
3808			    B_FALSE, B_FALSE, B_TRUE);
3809		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
3810		    sav->sav_count) == 0);
3811		for (i = 0; i < sav->sav_count; i++)
3812			nvlist_free(list[i]);
3813		kmem_free(list, sav->sav_count * sizeof (void *));
3814	}
3815
3816	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
3817	nvlist_free(nvroot);
3818
3819	sav->sav_sync = B_FALSE;
3820}
3821
3822static void
3823spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
3824{
3825	nvlist_t *config;
3826
3827	if (list_is_empty(&spa->spa_config_dirty_list))
3828		return;
3829
3830	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
3831
3832	config = spa_config_generate(spa, spa->spa_root_vdev,
3833	    dmu_tx_get_txg(tx), B_FALSE);
3834
3835	spa_config_exit(spa, SCL_STATE, FTAG);
3836
3837	if (spa->spa_config_syncing)
3838		nvlist_free(spa->spa_config_syncing);
3839	spa->spa_config_syncing = config;
3840
3841	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
3842}
3843
3844/*
3845 * Set zpool properties.
3846 */
3847static void
3848spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
3849{
3850	spa_t *spa = arg1;
3851	objset_t *mos = spa->spa_meta_objset;
3852	nvlist_t *nvp = arg2;
3853	nvpair_t *elem;
3854	uint64_t intval;
3855	char *strval;
3856	zpool_prop_t prop;
3857	const char *propname;
3858	zprop_type_t proptype;
3859
3860	mutex_enter(&spa->spa_props_lock);
3861
3862	elem = NULL;
3863	while ((elem = nvlist_next_nvpair(nvp, elem))) {
3864		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
3865		case ZPOOL_PROP_VERSION:
3866			/*
3867			 * Only set version for non-zpool-creation cases
3868			 * (set/import). spa_create() needs special care
3869			 * for version setting.
3870			 */
3871			if (tx->tx_txg != TXG_INITIAL) {
3872				VERIFY(nvpair_value_uint64(elem,
3873				    &intval) == 0);
3874				ASSERT(intval <= SPA_VERSION);
3875				ASSERT(intval >= spa_version(spa));
3876				spa->spa_uberblock.ub_version = intval;
3877				vdev_config_dirty(spa->spa_root_vdev);
3878			}
3879			break;
3880
3881		case ZPOOL_PROP_ALTROOT:
3882			/*
3883			 * 'altroot' is a non-persistent property. It should
3884			 * have been set temporarily at creation or import time.
3885			 */
3886			ASSERT(spa->spa_root != NULL);
3887			break;
3888
3889		case ZPOOL_PROP_CACHEFILE:
3890			/*
3891			 * 'cachefile' is also a non-persisitent property.
3892			 */
3893			break;
3894		default:
3895			/*
3896			 * Set pool property values in the poolprops mos object.
3897			 */
3898			if (spa->spa_pool_props_object == 0) {
3899				objset_t *mos = spa->spa_meta_objset;
3900
3901				VERIFY((spa->spa_pool_props_object =
3902				    zap_create(mos, DMU_OT_POOL_PROPS,
3903				    DMU_OT_NONE, 0, tx)) > 0);
3904
3905				VERIFY(zap_update(mos,
3906				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
3907				    8, 1, &spa->spa_pool_props_object, tx)
3908				    == 0);
3909			}
3910
3911			/* normalize the property name */
3912			propname = zpool_prop_to_name(prop);
3913			proptype = zpool_prop_get_type(prop);
3914
3915			if (nvpair_type(elem) == DATA_TYPE_STRING) {
3916				ASSERT(proptype == PROP_TYPE_STRING);
3917				VERIFY(nvpair_value_string(elem, &strval) == 0);
3918				VERIFY(zap_update(mos,
3919				    spa->spa_pool_props_object, propname,
3920				    1, strlen(strval) + 1, strval, tx) == 0);
3921
3922			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
3923				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
3924
3925				if (proptype == PROP_TYPE_INDEX) {
3926					const char *unused;
3927					VERIFY(zpool_prop_index_to_string(
3928					    prop, intval, &unused) == 0);
3929				}
3930				VERIFY(zap_update(mos,
3931				    spa->spa_pool_props_object, propname,
3932				    8, 1, &intval, tx) == 0);
3933			} else {
3934				ASSERT(0); /* not allowed */
3935			}
3936
3937			switch (prop) {
3938			case ZPOOL_PROP_DELEGATION:
3939				spa->spa_delegation = intval;
3940				break;
3941			case ZPOOL_PROP_BOOTFS:
3942				spa->spa_bootfs = intval;
3943				break;
3944			case ZPOOL_PROP_FAILUREMODE:
3945				spa->spa_failmode = intval;
3946				break;
3947			default:
3948				break;
3949			}
3950		}
3951
3952		/* log internal history if this is not a zpool create */
3953		if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
3954		    tx->tx_txg != TXG_INITIAL) {
3955			spa_history_internal_log(LOG_POOL_PROPSET,
3956			    spa, tx, cr, "%s %lld %s",
3957			    nvpair_name(elem), intval, spa_name(spa));
3958		}
3959	}
3960
3961	mutex_exit(&spa->spa_props_lock);
3962}
3963
3964/*
3965 * Sync the specified transaction group.  New blocks may be dirtied as
3966 * part of the process, so we iterate until it converges.
3967 */
3968void
3969spa_sync(spa_t *spa, uint64_t txg)
3970{
3971	dsl_pool_t *dp = spa->spa_dsl_pool;
3972	objset_t *mos = spa->spa_meta_objset;
3973	bplist_t *bpl = &spa->spa_sync_bplist;
3974	vdev_t *rvd = spa->spa_root_vdev;
3975	vdev_t *vd;
3976	dmu_tx_t *tx;
3977	int dirty_vdevs;
3978	int error;
3979
3980	/*
3981	 * Lock out configuration changes.
3982	 */
3983	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3984
3985	spa->spa_syncing_txg = txg;
3986	spa->spa_sync_pass = 0;
3987
3988	/*
3989	 * If there are any pending vdev state changes, convert them
3990	 * into config changes that go out with this transaction group.
3991	 */
3992	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
3993	while (list_head(&spa->spa_state_dirty_list) != NULL) {
3994		/*
3995		 * We need the write lock here because, for aux vdevs,
3996		 * calling vdev_config_dirty() modifies sav_config.
3997		 * This is ugly and will become unnecessary when we
3998		 * eliminate the aux vdev wart by integrating all vdevs
3999		 * into the root vdev tree.
4000		 */
4001		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
4002		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
4003		while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
4004			vdev_state_clean(vd);
4005			vdev_config_dirty(vd);
4006		}
4007		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
4008		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
4009	}
4010	spa_config_exit(spa, SCL_STATE, FTAG);
4011
4012	VERIFY(0 == bplist_open(bpl, mos, spa->spa_sync_bplist_obj));
4013
4014	tx = dmu_tx_create_assigned(dp, txg);
4015
4016	/*
4017	 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
4018	 * set spa_deflate if we have no raid-z vdevs.
4019	 */
4020	if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
4021	    spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
4022		int i;
4023
4024		for (i = 0; i < rvd->vdev_children; i++) {
4025			vd = rvd->vdev_child[i];
4026			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
4027				break;
4028		}
4029		if (i == rvd->vdev_children) {
4030			spa->spa_deflate = TRUE;
4031			VERIFY(0 == zap_add(spa->spa_meta_objset,
4032			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
4033			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
4034		}
4035	}
4036
4037	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
4038	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
4039		dsl_pool_create_origin(dp, tx);
4040
4041		/* Keeping the origin open increases spa_minref */
4042		spa->spa_minref += 3;
4043	}
4044
4045	if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
4046	    spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
4047		dsl_pool_upgrade_clones(dp, tx);
4048	}
4049
4050	/*
4051	 * If anything has changed in this txg, push the deferred frees
4052	 * from the previous txg.  If not, leave them alone so that we
4053	 * don't generate work on an otherwise idle system.
4054	 */
4055	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
4056	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
4057	    !txg_list_empty(&dp->dp_sync_tasks, txg))
4058		spa_sync_deferred_frees(spa, txg);
4059
4060	/*
4061	 * Iterate to convergence.
4062	 */
4063	do {
4064		spa->spa_sync_pass++;
4065
4066		spa_sync_config_object(spa, tx);
4067		spa_sync_aux_dev(spa, &spa->spa_spares, tx,
4068		    ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
4069		spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
4070		    ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
4071		spa_errlog_sync(spa, txg);
4072		dsl_pool_sync(dp, txg);
4073
4074		dirty_vdevs = 0;
4075		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) {
4076			vdev_sync(vd, txg);
4077			dirty_vdevs++;
4078		}
4079
4080		bplist_sync(bpl, tx);
4081	} while (dirty_vdevs);
4082
4083	bplist_close(bpl);
4084
4085	dprintf("txg %llu passes %d\n", txg, spa->spa_sync_pass);
4086
4087	/*
4088	 * Rewrite the vdev configuration (which includes the uberblock)
4089	 * to commit the transaction group.
4090	 *
4091	 * If there are no dirty vdevs, we sync the uberblock to a few
4092	 * random top-level vdevs that are known to be visible in the
4093	 * config cache (see spa_vdev_add() for a complete description).
4094	 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
4095	 */
4096	for (;;) {
4097		/*
4098		 * We hold SCL_STATE to prevent vdev open/close/etc.
4099		 * while we're attempting to write the vdev labels.
4100		 */
4101		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4102
4103		if (list_is_empty(&spa->spa_config_dirty_list)) {
4104			vdev_t *svd[SPA_DVAS_PER_BP];
4105			int svdcount = 0;
4106			int children = rvd->vdev_children;
4107			int c0 = spa_get_random(children);
4108			int c;
4109
4110			for (c = 0; c < children; c++) {
4111				vd = rvd->vdev_child[(c0 + c) % children];
4112				if (vd->vdev_ms_array == 0 || vd->vdev_islog)
4113					continue;
4114				svd[svdcount++] = vd;
4115				if (svdcount == SPA_DVAS_PER_BP)
4116					break;
4117			}
4118			error = vdev_config_sync(svd, svdcount, txg);
4119		} else {
4120			error = vdev_config_sync(rvd->vdev_child,
4121			    rvd->vdev_children, txg);
4122		}
4123
4124		spa_config_exit(spa, SCL_STATE, FTAG);
4125
4126		if (error == 0)
4127			break;
4128		zio_suspend(spa, NULL);
4129		zio_resume_wait(spa);
4130	}
4131	dmu_tx_commit(tx);
4132
4133	/*
4134	 * Clear the dirty config list.
4135	 */
4136	while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
4137		vdev_config_clean(vd);
4138
4139	/*
4140	 * Now that the new config has synced transactionally,
4141	 * let it become visible to the config cache.
4142	 */
4143	if (spa->spa_config_syncing != NULL) {
4144		spa_config_set(spa, spa->spa_config_syncing);
4145		spa->spa_config_txg = txg;
4146		spa->spa_config_syncing = NULL;
4147	}
4148
4149	spa->spa_ubsync = spa->spa_uberblock;
4150
4151	/*
4152	 * Clean up the ZIL records for the synced txg.
4153	 */
4154	dsl_pool_zil_clean(dp);
4155
4156	/*
4157	 * Update usable space statistics.
4158	 */
4159	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
4160		vdev_sync_done(vd, txg);
4161
4162	/*
4163	 * It had better be the case that we didn't dirty anything
4164	 * since vdev_config_sync().
4165	 */
4166	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
4167	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
4168	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
4169	ASSERT(bpl->bpl_queue == NULL);
4170
4171	spa_config_exit(spa, SCL_CONFIG, FTAG);
4172
4173	/*
4174	 * If any async tasks have been requested, kick them off.
4175	 */
4176	spa_async_dispatch(spa);
4177}
4178
4179/*
4180 * Sync all pools.  We don't want to hold the namespace lock across these
4181 * operations, so we take a reference on the spa_t and drop the lock during the
4182 * sync.
4183 */
4184void
4185spa_sync_allpools(void)
4186{
4187	spa_t *spa = NULL;
4188	mutex_enter(&spa_namespace_lock);
4189	while ((spa = spa_next(spa)) != NULL) {
4190		if (spa_state(spa) != POOL_STATE_ACTIVE || spa_suspended(spa))
4191			continue;
4192		spa_open_ref(spa, FTAG);
4193		mutex_exit(&spa_namespace_lock);
4194		txg_wait_synced(spa_get_dsl(spa), 0);
4195		mutex_enter(&spa_namespace_lock);
4196		spa_close(spa, FTAG);
4197	}
4198	mutex_exit(&spa_namespace_lock);
4199}
4200
4201/*
4202 * ==========================================================================
4203 * Miscellaneous routines
4204 * ==========================================================================
4205 */
4206
4207/*
4208 * Remove all pools in the system.
4209 */
4210void
4211spa_evict_all(void)
4212{
4213	spa_t *spa;
4214
4215	/*
4216	 * Remove all cached state.  All pools should be closed now,
4217	 * so every spa in the AVL tree should be unreferenced.
4218	 */
4219	mutex_enter(&spa_namespace_lock);
4220	while ((spa = spa_next(NULL)) != NULL) {
4221		/*
4222		 * Stop async tasks.  The async thread may need to detach
4223		 * a device that's been replaced, which requires grabbing
4224		 * spa_namespace_lock, so we must drop it here.
4225		 */
4226		spa_open_ref(spa, FTAG);
4227		mutex_exit(&spa_namespace_lock);
4228		spa_async_suspend(spa);
4229		mutex_enter(&spa_namespace_lock);
4230		spa_close(spa, FTAG);
4231
4232		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4233			spa_unload(spa);
4234			spa_deactivate(spa);
4235		}
4236		spa_remove(spa);
4237	}
4238	mutex_exit(&spa_namespace_lock);
4239}
4240
4241vdev_t *
4242spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t l2cache)
4243{
4244	vdev_t *vd;
4245	int i;
4246
4247	if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
4248		return (vd);
4249
4250	if (l2cache) {
4251		for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
4252			vd = spa->spa_l2cache.sav_vdevs[i];
4253			if (vd->vdev_guid == guid)
4254				return (vd);
4255		}
4256	}
4257
4258	return (NULL);
4259}
4260
4261void
4262spa_upgrade(spa_t *spa, uint64_t version)
4263{
4264	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4265
4266	/*
4267	 * This should only be called for a non-faulted pool, and since a
4268	 * future version would result in an unopenable pool, this shouldn't be
4269	 * possible.
4270	 */
4271	ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
4272	ASSERT(version >= spa->spa_uberblock.ub_version);
4273
4274	spa->spa_uberblock.ub_version = version;
4275	vdev_config_dirty(spa->spa_root_vdev);
4276
4277	spa_config_exit(spa, SCL_ALL, FTAG);
4278
4279	txg_wait_synced(spa_get_dsl(spa), 0);
4280}
4281
4282boolean_t
4283spa_has_spare(spa_t *spa, uint64_t guid)
4284{
4285	int i;
4286	uint64_t spareguid;
4287	spa_aux_vdev_t *sav = &spa->spa_spares;
4288
4289	for (i = 0; i < sav->sav_count; i++)
4290		if (sav->sav_vdevs[i]->vdev_guid == guid)
4291			return (B_TRUE);
4292
4293	for (i = 0; i < sav->sav_npending; i++) {
4294		if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
4295		    &spareguid) == 0 && spareguid == guid)
4296			return (B_TRUE);
4297	}
4298
4299	return (B_FALSE);
4300}
4301
4302/*
4303 * Check if a pool has an active shared spare device.
4304 * Note: reference count of an active spare is 2, as a spare and as a replace
4305 */
4306static boolean_t
4307spa_has_active_shared_spare(spa_t *spa)
4308{
4309	int i, refcnt;
4310	uint64_t pool;
4311	spa_aux_vdev_t *sav = &spa->spa_spares;
4312
4313	for (i = 0; i < sav->sav_count; i++) {
4314		if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
4315		    &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
4316		    refcnt > 2)
4317			return (B_TRUE);
4318	}
4319
4320	return (B_FALSE);
4321}
4322
4323/*
4324 * Post a sysevent corresponding to the given event.  The 'name' must be one of
4325 * the event definitions in sys/sysevent/eventdefs.h.  The payload will be
4326 * filled in from the spa and (optionally) the vdev.  This doesn't do anything
4327 * in the userland libzpool, as we don't want consumers to misinterpret ztest
4328 * or zdb as real changes.
4329 */
4330void
4331spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
4332{
4333#ifdef _KERNEL
4334	sysevent_t		*ev;
4335	sysevent_attr_list_t	*attr = NULL;
4336	sysevent_value_t	value;
4337	sysevent_id_t		eid;
4338
4339	ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
4340	    SE_SLEEP);
4341
4342	value.value_type = SE_DATA_TYPE_STRING;
4343	value.value.sv_string = spa_name(spa);
4344	if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
4345		goto done;
4346
4347	value.value_type = SE_DATA_TYPE_UINT64;
4348	value.value.sv_uint64 = spa_guid(spa);
4349	if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
4350		goto done;
4351
4352	if (vd) {
4353		value.value_type = SE_DATA_TYPE_UINT64;
4354		value.value.sv_uint64 = vd->vdev_guid;
4355		if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
4356		    SE_SLEEP) != 0)
4357			goto done;
4358
4359		if (vd->vdev_path) {
4360			value.value_type = SE_DATA_TYPE_STRING;
4361			value.value.sv_string = vd->vdev_path;
4362			if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
4363			    &value, SE_SLEEP) != 0)
4364				goto done;
4365		}
4366	}
4367
4368	if (sysevent_attach_attributes(ev, attr) != 0)
4369		goto done;
4370	attr = NULL;
4371
4372	(void) log_sysevent(ev, SE_SLEEP, &eid);
4373
4374done:
4375	if (attr)
4376		sysevent_free_attr(attr);
4377	sysevent_free(ev);
4378#endif
4379}
4380