spa.c revision 6809eb4e2a4a94bcc76229c7b5108b96134537b6
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	{	8,	8	},	/* ZIO_TYPE_READ	*/
74	{	8,	8	},	/* 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	if (spa->spa_async_zio_root != NULL) {
687		(void) zio_wait(spa->spa_async_zio_root);
688		spa->spa_async_zio_root = NULL;
689	}
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		vd->vdev_aux = &spa->spa_spares;
832
833		if (vdev_open(vd) != 0)
834			continue;
835
836		if (vdev_validate_aux(vd) == 0)
837			spa_spare_add(vd);
838	}
839
840	/*
841	 * Recompute the stashed list of spares, with status information
842	 * this time.
843	 */
844	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
845	    DATA_TYPE_NVLIST_ARRAY) == 0);
846
847	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
848	    KM_SLEEP);
849	for (i = 0; i < spa->spa_spares.sav_count; i++)
850		spares[i] = vdev_config_generate(spa,
851		    spa->spa_spares.sav_vdevs[i], B_TRUE, B_TRUE, B_FALSE);
852	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
853	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
854	for (i = 0; i < spa->spa_spares.sav_count; i++)
855		nvlist_free(spares[i]);
856	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
857}
858
859/*
860 * Load (or re-load) the current list of vdevs describing the active l2cache for
861 * this pool.  When this is called, we have some form of basic information in
862 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
863 * then re-generate a more complete list including status information.
864 * Devices which are already active have their details maintained, and are
865 * not re-opened.
866 */
867static void
868spa_load_l2cache(spa_t *spa)
869{
870	nvlist_t **l2cache;
871	uint_t nl2cache;
872	int i, j, oldnvdevs;
873	uint64_t guid, size;
874	vdev_t *vd, **oldvdevs, **newvdevs;
875	spa_aux_vdev_t *sav = &spa->spa_l2cache;
876
877	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
878
879	if (sav->sav_config != NULL) {
880		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
881		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
882		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
883	} else {
884		nl2cache = 0;
885	}
886
887	oldvdevs = sav->sav_vdevs;
888	oldnvdevs = sav->sav_count;
889	sav->sav_vdevs = NULL;
890	sav->sav_count = 0;
891
892	/*
893	 * Process new nvlist of vdevs.
894	 */
895	for (i = 0; i < nl2cache; i++) {
896		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
897		    &guid) == 0);
898
899		newvdevs[i] = NULL;
900		for (j = 0; j < oldnvdevs; j++) {
901			vd = oldvdevs[j];
902			if (vd != NULL && guid == vd->vdev_guid) {
903				/*
904				 * Retain previous vdev for add/remove ops.
905				 */
906				newvdevs[i] = vd;
907				oldvdevs[j] = NULL;
908				break;
909			}
910		}
911
912		if (newvdevs[i] == NULL) {
913			/*
914			 * Create new vdev
915			 */
916			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
917			    VDEV_ALLOC_L2CACHE) == 0);
918			ASSERT(vd != NULL);
919			newvdevs[i] = vd;
920
921			/*
922			 * Commit this vdev as an l2cache device,
923			 * even if it fails to open.
924			 */
925			spa_l2cache_add(vd);
926
927			vd->vdev_top = vd;
928			vd->vdev_aux = sav;
929
930			spa_l2cache_activate(vd);
931
932			if (vdev_open(vd) != 0)
933				continue;
934
935			(void) vdev_validate_aux(vd);
936
937			if (!vdev_is_dead(vd)) {
938				size = vdev_get_rsize(vd);
939				l2arc_add_vdev(spa, vd,
940				    VDEV_LABEL_START_SIZE,
941				    size - VDEV_LABEL_START_SIZE);
942			}
943		}
944	}
945
946	/*
947	 * Purge vdevs that were dropped
948	 */
949	for (i = 0; i < oldnvdevs; i++) {
950		uint64_t pool;
951
952		vd = oldvdevs[i];
953		if (vd != NULL) {
954			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
955			    pool != 0ULL && l2arc_vdev_present(vd))
956				l2arc_remove_vdev(vd);
957			(void) vdev_close(vd);
958			spa_l2cache_remove(vd);
959		}
960	}
961
962	if (oldvdevs)
963		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
964
965	if (sav->sav_config == NULL)
966		goto out;
967
968	sav->sav_vdevs = newvdevs;
969	sav->sav_count = (int)nl2cache;
970
971	/*
972	 * Recompute the stashed list of l2cache devices, with status
973	 * information this time.
974	 */
975	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
976	    DATA_TYPE_NVLIST_ARRAY) == 0);
977
978	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
979	for (i = 0; i < sav->sav_count; i++)
980		l2cache[i] = vdev_config_generate(spa,
981		    sav->sav_vdevs[i], B_TRUE, B_FALSE, B_TRUE);
982	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
983	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
984out:
985	for (i = 0; i < sav->sav_count; i++)
986		nvlist_free(l2cache[i]);
987	if (sav->sav_count)
988		kmem_free(l2cache, sav->sav_count * sizeof (void *));
989}
990
991static int
992load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
993{
994	dmu_buf_t *db;
995	char *packed = NULL;
996	size_t nvsize = 0;
997	int error;
998	*value = NULL;
999
1000	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1001	nvsize = *(uint64_t *)db->db_data;
1002	dmu_buf_rele(db, FTAG);
1003
1004	packed = kmem_alloc(nvsize, KM_SLEEP);
1005	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed);
1006	if (error == 0)
1007		error = nvlist_unpack(packed, nvsize, value, 0);
1008	kmem_free(packed, nvsize);
1009
1010	return (error);
1011}
1012
1013/*
1014 * Checks to see if the given vdev could not be opened, in which case we post a
1015 * sysevent to notify the autoreplace code that the device has been removed.
1016 */
1017static void
1018spa_check_removed(vdev_t *vd)
1019{
1020	int c;
1021
1022	for (c = 0; c < vd->vdev_children; c++)
1023		spa_check_removed(vd->vdev_child[c]);
1024
1025	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1026		zfs_post_autoreplace(vd->vdev_spa, vd);
1027		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1028	}
1029}
1030
1031/*
1032 * Check for missing log devices
1033 */
1034int
1035spa_check_logs(spa_t *spa)
1036{
1037	switch (spa->spa_log_state) {
1038	case SPA_LOG_MISSING:
1039		/* need to recheck in case slog has been restored */
1040	case SPA_LOG_UNKNOWN:
1041		if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1042		    DS_FIND_CHILDREN)) {
1043			spa->spa_log_state = SPA_LOG_MISSING;
1044			return (1);
1045		}
1046		break;
1047
1048	case SPA_LOG_CLEAR:
1049		(void) dmu_objset_find(spa->spa_name, zil_clear_log_chain, NULL,
1050		    DS_FIND_CHILDREN);
1051		break;
1052	}
1053	spa->spa_log_state = SPA_LOG_GOOD;
1054	return (0);
1055}
1056
1057/*
1058 * Load an existing storage pool, using the pool's builtin spa_config as a
1059 * source of configuration information.
1060 */
1061static int
1062spa_load(spa_t *spa, nvlist_t *config, spa_load_state_t state, int mosconfig)
1063{
1064	int error = 0;
1065	nvlist_t *nvroot = NULL;
1066	vdev_t *rvd;
1067	uberblock_t *ub = &spa->spa_uberblock;
1068	uint64_t config_cache_txg = spa->spa_config_txg;
1069	uint64_t pool_guid;
1070	uint64_t version;
1071	uint64_t autoreplace = 0;
1072	int orig_mode = spa->spa_mode;
1073	char *ereport = FM_EREPORT_ZFS_POOL;
1074
1075	/*
1076	 * If this is an untrusted config, access the pool in read-only mode.
1077	 * This prevents things like resilvering recently removed devices.
1078	 */
1079	if (!mosconfig)
1080		spa->spa_mode = FREAD;
1081
1082	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1083
1084	spa->spa_load_state = state;
1085
1086	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) ||
1087	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
1088		error = EINVAL;
1089		goto out;
1090	}
1091
1092	/*
1093	 * Versioning wasn't explicitly added to the label until later, so if
1094	 * it's not present treat it as the initial version.
1095	 */
1096	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &version) != 0)
1097		version = SPA_VERSION_INITIAL;
1098
1099	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1100	    &spa->spa_config_txg);
1101
1102	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1103	    spa_guid_exists(pool_guid, 0)) {
1104		error = EEXIST;
1105		goto out;
1106	}
1107
1108	spa->spa_load_guid = pool_guid;
1109
1110	/*
1111	 * Create "The Godfather" zio to hold all async IOs
1112	 */
1113	if (spa->spa_async_zio_root == NULL)
1114		spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1115		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
1116		    ZIO_FLAG_GODFATHER);
1117
1118	/*
1119	 * Parse the configuration into a vdev tree.  We explicitly set the
1120	 * value that will be returned by spa_version() since parsing the
1121	 * configuration requires knowing the version number.
1122	 */
1123	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1124	spa->spa_ubsync.ub_version = version;
1125	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_LOAD);
1126	spa_config_exit(spa, SCL_ALL, FTAG);
1127
1128	if (error != 0)
1129		goto out;
1130
1131	ASSERT(spa->spa_root_vdev == rvd);
1132	ASSERT(spa_guid(spa) == pool_guid);
1133
1134	/*
1135	 * Try to open all vdevs, loading each label in the process.
1136	 */
1137	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1138	error = vdev_open(rvd);
1139	spa_config_exit(spa, SCL_ALL, FTAG);
1140	if (error != 0)
1141		goto out;
1142
1143	/*
1144	 * We need to validate the vdev labels against the configuration that
1145	 * we have in hand, which is dependent on the setting of mosconfig. If
1146	 * mosconfig is true then we're validating the vdev labels based on
1147	 * that config. Otherwise, we're validating against the cached config
1148	 * (zpool.cache) that was read when we loaded the zfs module, and then
1149	 * later we will recursively call spa_load() and validate against
1150	 * the vdev config.
1151	 */
1152	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1153	error = vdev_validate(rvd);
1154	spa_config_exit(spa, SCL_ALL, FTAG);
1155	if (error != 0)
1156		goto out;
1157
1158	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
1159		error = ENXIO;
1160		goto out;
1161	}
1162
1163	/*
1164	 * Find the best uberblock.
1165	 */
1166	vdev_uberblock_load(NULL, rvd, ub);
1167
1168	/*
1169	 * If we weren't able to find a single valid uberblock, return failure.
1170	 */
1171	if (ub->ub_txg == 0) {
1172		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1173		    VDEV_AUX_CORRUPT_DATA);
1174		error = ENXIO;
1175		goto out;
1176	}
1177
1178	/*
1179	 * If the pool is newer than the code, we can't open it.
1180	 */
1181	if (ub->ub_version > SPA_VERSION) {
1182		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1183		    VDEV_AUX_VERSION_NEWER);
1184		error = ENOTSUP;
1185		goto out;
1186	}
1187
1188	/*
1189	 * If the vdev guid sum doesn't match the uberblock, we have an
1190	 * incomplete configuration.
1191	 */
1192	if (rvd->vdev_guid_sum != ub->ub_guid_sum && mosconfig) {
1193		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1194		    VDEV_AUX_BAD_GUID_SUM);
1195		error = ENXIO;
1196		goto out;
1197	}
1198
1199	/*
1200	 * Initialize internal SPA structures.
1201	 */
1202	spa->spa_state = POOL_STATE_ACTIVE;
1203	spa->spa_ubsync = spa->spa_uberblock;
1204	spa->spa_first_txg = spa_last_synced_txg(spa) + 1;
1205	error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1206	if (error) {
1207		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1208		    VDEV_AUX_CORRUPT_DATA);
1209		goto out;
1210	}
1211	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1212
1213	if (zap_lookup(spa->spa_meta_objset,
1214	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
1215	    sizeof (uint64_t), 1, &spa->spa_config_object) != 0) {
1216		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1217		    VDEV_AUX_CORRUPT_DATA);
1218		error = EIO;
1219		goto out;
1220	}
1221
1222	if (!mosconfig) {
1223		nvlist_t *newconfig;
1224		uint64_t hostid;
1225
1226		if (load_nvlist(spa, spa->spa_config_object, &newconfig) != 0) {
1227			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1228			    VDEV_AUX_CORRUPT_DATA);
1229			error = EIO;
1230			goto out;
1231		}
1232
1233		if (!spa_is_root(spa) && nvlist_lookup_uint64(newconfig,
1234		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
1235			char *hostname;
1236			unsigned long myhostid = 0;
1237
1238			VERIFY(nvlist_lookup_string(newconfig,
1239			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
1240
1241#ifdef	_KERNEL
1242			myhostid = zone_get_hostid(NULL);
1243#else	/* _KERNEL */
1244			/*
1245			 * We're emulating the system's hostid in userland, so
1246			 * we can't use zone_get_hostid().
1247			 */
1248			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
1249#endif	/* _KERNEL */
1250			if (hostid != 0 && myhostid != 0 &&
1251			    hostid != myhostid) {
1252				cmn_err(CE_WARN, "pool '%s' could not be "
1253				    "loaded as it was last accessed by "
1254				    "another system (host: %s hostid: 0x%lx). "
1255				    "See: http://www.sun.com/msg/ZFS-8000-EY",
1256				    spa_name(spa), hostname,
1257				    (unsigned long)hostid);
1258				error = EBADF;
1259				goto out;
1260			}
1261		}
1262
1263		spa_config_set(spa, newconfig);
1264		spa_unload(spa);
1265		spa_deactivate(spa);
1266		spa_activate(spa, orig_mode);
1267
1268		return (spa_load(spa, newconfig, state, B_TRUE));
1269	}
1270
1271	if (zap_lookup(spa->spa_meta_objset,
1272	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
1273	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj) != 0) {
1274		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1275		    VDEV_AUX_CORRUPT_DATA);
1276		error = EIO;
1277		goto out;
1278	}
1279
1280	/*
1281	 * Load the bit that tells us to use the new accounting function
1282	 * (raid-z deflation).  If we have an older pool, this will not
1283	 * be present.
1284	 */
1285	error = zap_lookup(spa->spa_meta_objset,
1286	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
1287	    sizeof (uint64_t), 1, &spa->spa_deflate);
1288	if (error != 0 && error != ENOENT) {
1289		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1290		    VDEV_AUX_CORRUPT_DATA);
1291		error = EIO;
1292		goto out;
1293	}
1294
1295	/*
1296	 * Load the persistent error log.  If we have an older pool, this will
1297	 * not be present.
1298	 */
1299	error = zap_lookup(spa->spa_meta_objset,
1300	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_LAST,
1301	    sizeof (uint64_t), 1, &spa->spa_errlog_last);
1302	if (error != 0 && error != ENOENT) {
1303		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1304		    VDEV_AUX_CORRUPT_DATA);
1305		error = EIO;
1306		goto out;
1307	}
1308
1309	error = zap_lookup(spa->spa_meta_objset,
1310	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_SCRUB,
1311	    sizeof (uint64_t), 1, &spa->spa_errlog_scrub);
1312	if (error != 0 && error != ENOENT) {
1313		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1314		    VDEV_AUX_CORRUPT_DATA);
1315		error = EIO;
1316		goto out;
1317	}
1318
1319	/*
1320	 * Load the history object.  If we have an older pool, this
1321	 * will not be present.
1322	 */
1323	error = zap_lookup(spa->spa_meta_objset,
1324	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_HISTORY,
1325	    sizeof (uint64_t), 1, &spa->spa_history);
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
1333	/*
1334	 * Load any hot spares for this pool.
1335	 */
1336	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1337	    DMU_POOL_SPARES, sizeof (uint64_t), 1, &spa->spa_spares.sav_object);
1338	if (error != 0 && error != ENOENT) {
1339		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1340		    VDEV_AUX_CORRUPT_DATA);
1341		error = EIO;
1342		goto out;
1343	}
1344	if (error == 0) {
1345		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
1346		if (load_nvlist(spa, spa->spa_spares.sav_object,
1347		    &spa->spa_spares.sav_config) != 0) {
1348			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1349			    VDEV_AUX_CORRUPT_DATA);
1350			error = EIO;
1351			goto out;
1352		}
1353
1354		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1355		spa_load_spares(spa);
1356		spa_config_exit(spa, SCL_ALL, FTAG);
1357	}
1358
1359	/*
1360	 * Load any level 2 ARC devices for this pool.
1361	 */
1362	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1363	    DMU_POOL_L2CACHE, sizeof (uint64_t), 1,
1364	    &spa->spa_l2cache.sav_object);
1365	if (error != 0 && error != ENOENT) {
1366		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1367		    VDEV_AUX_CORRUPT_DATA);
1368		error = EIO;
1369		goto out;
1370	}
1371	if (error == 0) {
1372		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
1373		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
1374		    &spa->spa_l2cache.sav_config) != 0) {
1375			vdev_set_state(rvd, B_TRUE,
1376			    VDEV_STATE_CANT_OPEN,
1377			    VDEV_AUX_CORRUPT_DATA);
1378			error = EIO;
1379			goto out;
1380		}
1381
1382		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1383		spa_load_l2cache(spa);
1384		spa_config_exit(spa, SCL_ALL, FTAG);
1385	}
1386
1387	if (spa_check_logs(spa)) {
1388		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1389		    VDEV_AUX_BAD_LOG);
1390		error = ENXIO;
1391		ereport = FM_EREPORT_ZFS_LOG_REPLAY;
1392		goto out;
1393	}
1394
1395
1396	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
1397
1398	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1399	    DMU_POOL_PROPS, sizeof (uint64_t), 1, &spa->spa_pool_props_object);
1400
1401	if (error && error != ENOENT) {
1402		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1403		    VDEV_AUX_CORRUPT_DATA);
1404		error = EIO;
1405		goto out;
1406	}
1407
1408	if (error == 0) {
1409		(void) zap_lookup(spa->spa_meta_objset,
1410		    spa->spa_pool_props_object,
1411		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS),
1412		    sizeof (uint64_t), 1, &spa->spa_bootfs);
1413		(void) zap_lookup(spa->spa_meta_objset,
1414		    spa->spa_pool_props_object,
1415		    zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE),
1416		    sizeof (uint64_t), 1, &autoreplace);
1417		(void) zap_lookup(spa->spa_meta_objset,
1418		    spa->spa_pool_props_object,
1419		    zpool_prop_to_name(ZPOOL_PROP_DELEGATION),
1420		    sizeof (uint64_t), 1, &spa->spa_delegation);
1421		(void) zap_lookup(spa->spa_meta_objset,
1422		    spa->spa_pool_props_object,
1423		    zpool_prop_to_name(ZPOOL_PROP_FAILUREMODE),
1424		    sizeof (uint64_t), 1, &spa->spa_failmode);
1425	}
1426
1427	/*
1428	 * If the 'autoreplace' property is set, then post a resource notifying
1429	 * the ZFS DE that it should not issue any faults for unopenable
1430	 * devices.  We also iterate over the vdevs, and post a sysevent for any
1431	 * unopenable vdevs so that the normal autoreplace handler can take
1432	 * over.
1433	 */
1434	if (autoreplace && state != SPA_LOAD_TRYIMPORT)
1435		spa_check_removed(spa->spa_root_vdev);
1436
1437	/*
1438	 * Load the vdev state for all toplevel vdevs.
1439	 */
1440	vdev_load(rvd);
1441
1442	/*
1443	 * Propagate the leaf DTLs we just loaded all the way up the tree.
1444	 */
1445	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1446	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
1447	spa_config_exit(spa, SCL_ALL, FTAG);
1448
1449	/*
1450	 * Check the state of the root vdev.  If it can't be opened, it
1451	 * indicates one or more toplevel vdevs are faulted.
1452	 */
1453	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
1454		error = ENXIO;
1455		goto out;
1456	}
1457
1458	if (spa_writeable(spa)) {
1459		dmu_tx_t *tx;
1460		int need_update = B_FALSE;
1461
1462		ASSERT(state != SPA_LOAD_TRYIMPORT);
1463
1464		/*
1465		 * Claim log blocks that haven't been committed yet.
1466		 * This must all happen in a single txg.
1467		 */
1468		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
1469		    spa_first_txg(spa));
1470		(void) dmu_objset_find(spa_name(spa),
1471		    zil_claim, tx, DS_FIND_CHILDREN);
1472		dmu_tx_commit(tx);
1473
1474		spa->spa_sync_on = B_TRUE;
1475		txg_sync_start(spa->spa_dsl_pool);
1476
1477		/*
1478		 * Wait for all claims to sync.
1479		 */
1480		txg_wait_synced(spa->spa_dsl_pool, 0);
1481
1482		/*
1483		 * If the config cache is stale, or we have uninitialized
1484		 * metaslabs (see spa_vdev_add()), then update the config.
1485		 */
1486		if (config_cache_txg != spa->spa_config_txg ||
1487		    state == SPA_LOAD_IMPORT)
1488			need_update = B_TRUE;
1489
1490		for (int c = 0; c < rvd->vdev_children; c++)
1491			if (rvd->vdev_child[c]->vdev_ms_array == 0)
1492				need_update = B_TRUE;
1493
1494		/*
1495		 * Update the config cache asychronously in case we're the
1496		 * root pool, in which case the config cache isn't writable yet.
1497		 */
1498		if (need_update)
1499			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
1500
1501		/*
1502		 * Check all DTLs to see if anything needs resilvering.
1503		 */
1504		if (vdev_resilver_needed(rvd, NULL, NULL))
1505			spa_async_request(spa, SPA_ASYNC_RESILVER);
1506	}
1507
1508	error = 0;
1509out:
1510	spa->spa_minref = refcount_count(&spa->spa_refcount);
1511	if (error && error != EBADF)
1512		zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1513	spa->spa_load_state = SPA_LOAD_NONE;
1514	spa->spa_ena = 0;
1515
1516	return (error);
1517}
1518
1519/*
1520 * Pool Open/Import
1521 *
1522 * The import case is identical to an open except that the configuration is sent
1523 * down from userland, instead of grabbed from the configuration cache.  For the
1524 * case of an open, the pool configuration will exist in the
1525 * POOL_STATE_UNINITIALIZED state.
1526 *
1527 * The stats information (gen/count/ustats) is used to gather vdev statistics at
1528 * the same time open the pool, without having to keep around the spa_t in some
1529 * ambiguous state.
1530 */
1531static int
1532spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t **config)
1533{
1534	spa_t *spa;
1535	int error;
1536	int locked = B_FALSE;
1537
1538	*spapp = NULL;
1539
1540	/*
1541	 * As disgusting as this is, we need to support recursive calls to this
1542	 * function because dsl_dir_open() is called during spa_load(), and ends
1543	 * up calling spa_open() again.  The real fix is to figure out how to
1544	 * avoid dsl_dir_open() calling this in the first place.
1545	 */
1546	if (mutex_owner(&spa_namespace_lock) != curthread) {
1547		mutex_enter(&spa_namespace_lock);
1548		locked = B_TRUE;
1549	}
1550
1551	if ((spa = spa_lookup(pool)) == NULL) {
1552		if (locked)
1553			mutex_exit(&spa_namespace_lock);
1554		return (ENOENT);
1555	}
1556	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
1557
1558		spa_activate(spa, spa_mode_global);
1559
1560		error = spa_load(spa, spa->spa_config, SPA_LOAD_OPEN, B_FALSE);
1561
1562		if (error == EBADF) {
1563			/*
1564			 * If vdev_validate() returns failure (indicated by
1565			 * EBADF), it indicates that one of the vdevs indicates
1566			 * that the pool has been exported or destroyed.  If
1567			 * this is the case, the config cache is out of sync and
1568			 * we should remove the pool from the namespace.
1569			 */
1570			spa_unload(spa);
1571			spa_deactivate(spa);
1572			spa_config_sync(spa, B_TRUE, B_TRUE);
1573			spa_remove(spa);
1574			if (locked)
1575				mutex_exit(&spa_namespace_lock);
1576			return (ENOENT);
1577		}
1578
1579		if (error) {
1580			/*
1581			 * We can't open the pool, but we still have useful
1582			 * information: the state of each vdev after the
1583			 * attempted vdev_open().  Return this to the user.
1584			 */
1585			if (config != NULL && spa->spa_root_vdev != NULL)
1586				*config = spa_config_generate(spa, NULL, -1ULL,
1587				    B_TRUE);
1588			spa_unload(spa);
1589			spa_deactivate(spa);
1590			spa->spa_last_open_failed = B_TRUE;
1591			if (locked)
1592				mutex_exit(&spa_namespace_lock);
1593			*spapp = NULL;
1594			return (error);
1595		} else {
1596			spa->spa_last_open_failed = B_FALSE;
1597		}
1598	}
1599
1600	spa_open_ref(spa, tag);
1601
1602	if (locked)
1603		mutex_exit(&spa_namespace_lock);
1604
1605	*spapp = spa;
1606
1607	if (config != NULL)
1608		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
1609
1610	return (0);
1611}
1612
1613int
1614spa_open(const char *name, spa_t **spapp, void *tag)
1615{
1616	return (spa_open_common(name, spapp, tag, NULL));
1617}
1618
1619/*
1620 * Lookup the given spa_t, incrementing the inject count in the process,
1621 * preventing it from being exported or destroyed.
1622 */
1623spa_t *
1624spa_inject_addref(char *name)
1625{
1626	spa_t *spa;
1627
1628	mutex_enter(&spa_namespace_lock);
1629	if ((spa = spa_lookup(name)) == NULL) {
1630		mutex_exit(&spa_namespace_lock);
1631		return (NULL);
1632	}
1633	spa->spa_inject_ref++;
1634	mutex_exit(&spa_namespace_lock);
1635
1636	return (spa);
1637}
1638
1639void
1640spa_inject_delref(spa_t *spa)
1641{
1642	mutex_enter(&spa_namespace_lock);
1643	spa->spa_inject_ref--;
1644	mutex_exit(&spa_namespace_lock);
1645}
1646
1647/*
1648 * Add spares device information to the nvlist.
1649 */
1650static void
1651spa_add_spares(spa_t *spa, nvlist_t *config)
1652{
1653	nvlist_t **spares;
1654	uint_t i, nspares;
1655	nvlist_t *nvroot;
1656	uint64_t guid;
1657	vdev_stat_t *vs;
1658	uint_t vsc;
1659	uint64_t pool;
1660
1661	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
1662
1663	if (spa->spa_spares.sav_count == 0)
1664		return;
1665
1666	VERIFY(nvlist_lookup_nvlist(config,
1667	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1668	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1669	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1670	if (nspares != 0) {
1671		VERIFY(nvlist_add_nvlist_array(nvroot,
1672		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1673		VERIFY(nvlist_lookup_nvlist_array(nvroot,
1674		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1675
1676		/*
1677		 * Go through and find any spares which have since been
1678		 * repurposed as an active spare.  If this is the case, update
1679		 * their status appropriately.
1680		 */
1681		for (i = 0; i < nspares; i++) {
1682			VERIFY(nvlist_lookup_uint64(spares[i],
1683			    ZPOOL_CONFIG_GUID, &guid) == 0);
1684			if (spa_spare_exists(guid, &pool, NULL) &&
1685			    pool != 0ULL) {
1686				VERIFY(nvlist_lookup_uint64_array(
1687				    spares[i], ZPOOL_CONFIG_STATS,
1688				    (uint64_t **)&vs, &vsc) == 0);
1689				vs->vs_state = VDEV_STATE_CANT_OPEN;
1690				vs->vs_aux = VDEV_AUX_SPARED;
1691			}
1692		}
1693	}
1694}
1695
1696/*
1697 * Add l2cache device information to the nvlist, including vdev stats.
1698 */
1699static void
1700spa_add_l2cache(spa_t *spa, nvlist_t *config)
1701{
1702	nvlist_t **l2cache;
1703	uint_t i, j, nl2cache;
1704	nvlist_t *nvroot;
1705	uint64_t guid;
1706	vdev_t *vd;
1707	vdev_stat_t *vs;
1708	uint_t vsc;
1709
1710	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
1711
1712	if (spa->spa_l2cache.sav_count == 0)
1713		return;
1714
1715	VERIFY(nvlist_lookup_nvlist(config,
1716	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1717	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
1718	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1719	if (nl2cache != 0) {
1720		VERIFY(nvlist_add_nvlist_array(nvroot,
1721		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
1722		VERIFY(nvlist_lookup_nvlist_array(nvroot,
1723		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1724
1725		/*
1726		 * Update level 2 cache device stats.
1727		 */
1728
1729		for (i = 0; i < nl2cache; i++) {
1730			VERIFY(nvlist_lookup_uint64(l2cache[i],
1731			    ZPOOL_CONFIG_GUID, &guid) == 0);
1732
1733			vd = NULL;
1734			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
1735				if (guid ==
1736				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
1737					vd = spa->spa_l2cache.sav_vdevs[j];
1738					break;
1739				}
1740			}
1741			ASSERT(vd != NULL);
1742
1743			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
1744			    ZPOOL_CONFIG_STATS, (uint64_t **)&vs, &vsc) == 0);
1745			vdev_get_stats(vd, vs);
1746		}
1747	}
1748}
1749
1750int
1751spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
1752{
1753	int error;
1754	spa_t *spa;
1755
1756	*config = NULL;
1757	error = spa_open_common(name, &spa, FTAG, config);
1758
1759	if (spa != NULL) {
1760		/*
1761		 * This still leaves a window of inconsistency where the spares
1762		 * or l2cache devices could change and the config would be
1763		 * self-inconsistent.
1764		 */
1765		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
1766
1767		if (*config != NULL) {
1768			VERIFY(nvlist_add_uint64(*config,
1769			    ZPOOL_CONFIG_ERRCOUNT,
1770			    spa_get_errlog_size(spa)) == 0);
1771
1772			if (spa_suspended(spa))
1773				VERIFY(nvlist_add_uint64(*config,
1774				    ZPOOL_CONFIG_SUSPENDED,
1775				    spa->spa_failmode) == 0);
1776
1777			spa_add_spares(spa, *config);
1778			spa_add_l2cache(spa, *config);
1779		}
1780	}
1781
1782	/*
1783	 * We want to get the alternate root even for faulted pools, so we cheat
1784	 * and call spa_lookup() directly.
1785	 */
1786	if (altroot) {
1787		if (spa == NULL) {
1788			mutex_enter(&spa_namespace_lock);
1789			spa = spa_lookup(name);
1790			if (spa)
1791				spa_altroot(spa, altroot, buflen);
1792			else
1793				altroot[0] = '\0';
1794			spa = NULL;
1795			mutex_exit(&spa_namespace_lock);
1796		} else {
1797			spa_altroot(spa, altroot, buflen);
1798		}
1799	}
1800
1801	if (spa != NULL) {
1802		spa_config_exit(spa, SCL_CONFIG, FTAG);
1803		spa_close(spa, FTAG);
1804	}
1805
1806	return (error);
1807}
1808
1809/*
1810 * Validate that the auxiliary device array is well formed.  We must have an
1811 * array of nvlists, each which describes a valid leaf vdev.  If this is an
1812 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
1813 * specified, as long as they are well-formed.
1814 */
1815static int
1816spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
1817    spa_aux_vdev_t *sav, const char *config, uint64_t version,
1818    vdev_labeltype_t label)
1819{
1820	nvlist_t **dev;
1821	uint_t i, ndev;
1822	vdev_t *vd;
1823	int error;
1824
1825	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1826
1827	/*
1828	 * It's acceptable to have no devs specified.
1829	 */
1830	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
1831		return (0);
1832
1833	if (ndev == 0)
1834		return (EINVAL);
1835
1836	/*
1837	 * Make sure the pool is formatted with a version that supports this
1838	 * device type.
1839	 */
1840	if (spa_version(spa) < version)
1841		return (ENOTSUP);
1842
1843	/*
1844	 * Set the pending device list so we correctly handle device in-use
1845	 * checking.
1846	 */
1847	sav->sav_pending = dev;
1848	sav->sav_npending = ndev;
1849
1850	for (i = 0; i < ndev; i++) {
1851		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
1852		    mode)) != 0)
1853			goto out;
1854
1855		if (!vd->vdev_ops->vdev_op_leaf) {
1856			vdev_free(vd);
1857			error = EINVAL;
1858			goto out;
1859		}
1860
1861		/*
1862		 * The L2ARC currently only supports disk devices in
1863		 * kernel context.  For user-level testing, we allow it.
1864		 */
1865#ifdef _KERNEL
1866		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
1867		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
1868			error = ENOTBLK;
1869			goto out;
1870		}
1871#endif
1872		vd->vdev_top = vd;
1873
1874		if ((error = vdev_open(vd)) == 0 &&
1875		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
1876			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
1877			    vd->vdev_guid) == 0);
1878		}
1879
1880		vdev_free(vd);
1881
1882		if (error &&
1883		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
1884			goto out;
1885		else
1886			error = 0;
1887	}
1888
1889out:
1890	sav->sav_pending = NULL;
1891	sav->sav_npending = 0;
1892	return (error);
1893}
1894
1895static int
1896spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
1897{
1898	int error;
1899
1900	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1901
1902	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
1903	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
1904	    VDEV_LABEL_SPARE)) != 0) {
1905		return (error);
1906	}
1907
1908	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
1909	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
1910	    VDEV_LABEL_L2CACHE));
1911}
1912
1913static void
1914spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
1915    const char *config)
1916{
1917	int i;
1918
1919	if (sav->sav_config != NULL) {
1920		nvlist_t **olddevs;
1921		uint_t oldndevs;
1922		nvlist_t **newdevs;
1923
1924		/*
1925		 * Generate new dev list by concatentating with the
1926		 * current dev list.
1927		 */
1928		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
1929		    &olddevs, &oldndevs) == 0);
1930
1931		newdevs = kmem_alloc(sizeof (void *) *
1932		    (ndevs + oldndevs), KM_SLEEP);
1933		for (i = 0; i < oldndevs; i++)
1934			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
1935			    KM_SLEEP) == 0);
1936		for (i = 0; i < ndevs; i++)
1937			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
1938			    KM_SLEEP) == 0);
1939
1940		VERIFY(nvlist_remove(sav->sav_config, config,
1941		    DATA_TYPE_NVLIST_ARRAY) == 0);
1942
1943		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1944		    config, newdevs, ndevs + oldndevs) == 0);
1945		for (i = 0; i < oldndevs + ndevs; i++)
1946			nvlist_free(newdevs[i]);
1947		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
1948	} else {
1949		/*
1950		 * Generate a new dev list.
1951		 */
1952		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
1953		    KM_SLEEP) == 0);
1954		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
1955		    devs, ndevs) == 0);
1956	}
1957}
1958
1959/*
1960 * Stop and drop level 2 ARC devices
1961 */
1962void
1963spa_l2cache_drop(spa_t *spa)
1964{
1965	vdev_t *vd;
1966	int i;
1967	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1968
1969	for (i = 0; i < sav->sav_count; i++) {
1970		uint64_t pool;
1971
1972		vd = sav->sav_vdevs[i];
1973		ASSERT(vd != NULL);
1974
1975		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1976		    pool != 0ULL && l2arc_vdev_present(vd))
1977			l2arc_remove_vdev(vd);
1978		if (vd->vdev_isl2cache)
1979			spa_l2cache_remove(vd);
1980		vdev_clear_stats(vd);
1981		(void) vdev_close(vd);
1982	}
1983}
1984
1985/*
1986 * Pool Creation
1987 */
1988int
1989spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
1990    const char *history_str, nvlist_t *zplprops)
1991{
1992	spa_t *spa;
1993	char *altroot = NULL;
1994	vdev_t *rvd;
1995	dsl_pool_t *dp;
1996	dmu_tx_t *tx;
1997	int c, error = 0;
1998	uint64_t txg = TXG_INITIAL;
1999	nvlist_t **spares, **l2cache;
2000	uint_t nspares, nl2cache;
2001	uint64_t version;
2002
2003	/*
2004	 * If this pool already exists, return failure.
2005	 */
2006	mutex_enter(&spa_namespace_lock);
2007	if (spa_lookup(pool) != NULL) {
2008		mutex_exit(&spa_namespace_lock);
2009		return (EEXIST);
2010	}
2011
2012	/*
2013	 * Allocate a new spa_t structure.
2014	 */
2015	(void) nvlist_lookup_string(props,
2016	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2017	spa = spa_add(pool, altroot);
2018	spa_activate(spa, spa_mode_global);
2019
2020	spa->spa_uberblock.ub_txg = txg - 1;
2021
2022	if (props && (error = spa_prop_validate(spa, props))) {
2023		spa_unload(spa);
2024		spa_deactivate(spa);
2025		spa_remove(spa);
2026		mutex_exit(&spa_namespace_lock);
2027		return (error);
2028	}
2029
2030	if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2031	    &version) != 0)
2032		version = SPA_VERSION;
2033	ASSERT(version <= SPA_VERSION);
2034	spa->spa_uberblock.ub_version = version;
2035	spa->spa_ubsync = spa->spa_uberblock;
2036
2037	/*
2038	 * Create "The Godfather" zio to hold all async IOs
2039	 */
2040	if (spa->spa_async_zio_root == NULL)
2041		spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2042		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2043		    ZIO_FLAG_GODFATHER);
2044
2045	/*
2046	 * Create the root vdev.
2047	 */
2048	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2049
2050	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2051
2052	ASSERT(error != 0 || rvd != NULL);
2053	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2054
2055	if (error == 0 && !zfs_allocatable_devs(nvroot))
2056		error = EINVAL;
2057
2058	if (error == 0 &&
2059	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2060	    (error = spa_validate_aux(spa, nvroot, txg,
2061	    VDEV_ALLOC_ADD)) == 0) {
2062		for (c = 0; c < rvd->vdev_children; c++)
2063			vdev_init(rvd->vdev_child[c], txg);
2064		vdev_config_dirty(rvd);
2065	}
2066
2067	spa_config_exit(spa, SCL_ALL, FTAG);
2068
2069	if (error != 0) {
2070		spa_unload(spa);
2071		spa_deactivate(spa);
2072		spa_remove(spa);
2073		mutex_exit(&spa_namespace_lock);
2074		return (error);
2075	}
2076
2077	/*
2078	 * Get the list of spares, if specified.
2079	 */
2080	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2081	    &spares, &nspares) == 0) {
2082		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
2083		    KM_SLEEP) == 0);
2084		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2085		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2086		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2087		spa_load_spares(spa);
2088		spa_config_exit(spa, SCL_ALL, FTAG);
2089		spa->spa_spares.sav_sync = B_TRUE;
2090	}
2091
2092	/*
2093	 * Get the list of level 2 cache devices, if specified.
2094	 */
2095	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2096	    &l2cache, &nl2cache) == 0) {
2097		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2098		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2099		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2100		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2101		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2102		spa_load_l2cache(spa);
2103		spa_config_exit(spa, SCL_ALL, FTAG);
2104		spa->spa_l2cache.sav_sync = B_TRUE;
2105	}
2106
2107	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
2108	spa->spa_meta_objset = dp->dp_meta_objset;
2109
2110	tx = dmu_tx_create_assigned(dp, txg);
2111
2112	/*
2113	 * Create the pool config object.
2114	 */
2115	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
2116	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
2117	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2118
2119	if (zap_add(spa->spa_meta_objset,
2120	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
2121	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
2122		cmn_err(CE_PANIC, "failed to add pool config");
2123	}
2124
2125	/* Newly created pools with the right version are always deflated. */
2126	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
2127		spa->spa_deflate = TRUE;
2128		if (zap_add(spa->spa_meta_objset,
2129		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
2130		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
2131			cmn_err(CE_PANIC, "failed to add deflate");
2132		}
2133	}
2134
2135	/*
2136	 * Create the deferred-free bplist object.  Turn off compression
2137	 * because sync-to-convergence takes longer if the blocksize
2138	 * keeps changing.
2139	 */
2140	spa->spa_sync_bplist_obj = bplist_create(spa->spa_meta_objset,
2141	    1 << 14, tx);
2142	dmu_object_set_compress(spa->spa_meta_objset, spa->spa_sync_bplist_obj,
2143	    ZIO_COMPRESS_OFF, tx);
2144
2145	if (zap_add(spa->spa_meta_objset,
2146	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
2147	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj, tx) != 0) {
2148		cmn_err(CE_PANIC, "failed to add bplist");
2149	}
2150
2151	/*
2152	 * Create the pool's history object.
2153	 */
2154	if (version >= SPA_VERSION_ZPOOL_HISTORY)
2155		spa_history_create_obj(spa, tx);
2156
2157	/*
2158	 * Set pool properties.
2159	 */
2160	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
2161	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2162	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
2163	if (props != NULL) {
2164		spa_configfile_set(spa, props, B_FALSE);
2165		spa_sync_props(spa, props, CRED(), tx);
2166	}
2167
2168	dmu_tx_commit(tx);
2169
2170	spa->spa_sync_on = B_TRUE;
2171	txg_sync_start(spa->spa_dsl_pool);
2172
2173	/*
2174	 * We explicitly wait for the first transaction to complete so that our
2175	 * bean counters are appropriately updated.
2176	 */
2177	txg_wait_synced(spa->spa_dsl_pool, txg);
2178
2179	spa_config_sync(spa, B_FALSE, B_TRUE);
2180
2181	if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
2182		(void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
2183
2184	spa->spa_minref = refcount_count(&spa->spa_refcount);
2185
2186	mutex_exit(&spa_namespace_lock);
2187
2188	return (0);
2189}
2190
2191#ifdef _KERNEL
2192/*
2193 * Build a "root" vdev for a top level vdev read in from a rootpool
2194 * device label.
2195 */
2196static void
2197spa_build_rootpool_config(nvlist_t *config)
2198{
2199	nvlist_t *nvtop, *nvroot;
2200	uint64_t pgid;
2201
2202	/*
2203	 * Add this top-level vdev to the child array.
2204	 */
2205	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtop)
2206	    == 0);
2207	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pgid)
2208	    == 0);
2209
2210	/*
2211	 * Put this pool's top-level vdevs into a root vdev.
2212	 */
2213	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2214	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT)
2215	    == 0);
2216	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
2217	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
2218	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2219	    &nvtop, 1) == 0);
2220
2221	/*
2222	 * Replace the existing vdev_tree with the new root vdev in
2223	 * this pool's configuration (remove the old, add the new).
2224	 */
2225	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
2226	nvlist_free(nvroot);
2227}
2228
2229/*
2230 * Get the root pool information from the root disk, then import the root pool
2231 * during the system boot up time.
2232 */
2233extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
2234
2235int
2236spa_check_rootconf(char *devpath, char *devid, nvlist_t **bestconf,
2237    uint64_t *besttxg)
2238{
2239	nvlist_t *config;
2240	uint64_t txg;
2241	int error;
2242
2243	if (error = vdev_disk_read_rootlabel(devpath, devid, &config))
2244		return (error);
2245
2246	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
2247
2248	if (bestconf != NULL)
2249		*bestconf = config;
2250	else
2251		nvlist_free(config);
2252	*besttxg = txg;
2253	return (0);
2254}
2255
2256boolean_t
2257spa_rootdev_validate(nvlist_t *nv)
2258{
2259	uint64_t ival;
2260
2261	if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
2262	    nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
2263	    nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
2264		return (B_FALSE);
2265
2266	return (B_TRUE);
2267}
2268
2269
2270/*
2271 * Given the boot device's physical path or devid, check if the device
2272 * is in a valid state.  If so, return the configuration from the vdev
2273 * label.
2274 */
2275int
2276spa_get_rootconf(char *devpath, char *devid, nvlist_t **bestconf)
2277{
2278	nvlist_t *conf = NULL;
2279	uint64_t txg = 0;
2280	nvlist_t *nvtop, **child;
2281	char *type;
2282	char *bootpath = NULL;
2283	uint_t children, c;
2284	char *tmp;
2285	int error;
2286
2287	if (devpath && ((tmp = strchr(devpath, ' ')) != NULL))
2288		*tmp = '\0';
2289	if (error = spa_check_rootconf(devpath, devid, &conf, &txg)) {
2290		cmn_err(CE_NOTE, "error reading device label");
2291		return (error);
2292	}
2293	if (txg == 0) {
2294		cmn_err(CE_NOTE, "this device is detached");
2295		nvlist_free(conf);
2296		return (EINVAL);
2297	}
2298
2299	VERIFY(nvlist_lookup_nvlist(conf, ZPOOL_CONFIG_VDEV_TREE,
2300	    &nvtop) == 0);
2301	VERIFY(nvlist_lookup_string(nvtop, ZPOOL_CONFIG_TYPE, &type) == 0);
2302
2303	if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2304		if (spa_rootdev_validate(nvtop)) {
2305			goto out;
2306		} else {
2307			nvlist_free(conf);
2308			return (EINVAL);
2309		}
2310	}
2311
2312	ASSERT(strcmp(type, VDEV_TYPE_MIRROR) == 0);
2313
2314	VERIFY(nvlist_lookup_nvlist_array(nvtop, ZPOOL_CONFIG_CHILDREN,
2315	    &child, &children) == 0);
2316
2317	/*
2318	 * Go thru vdevs in the mirror to see if the given device
2319	 * has the most recent txg. Only the device with the most
2320	 * recent txg has valid information and should be booted.
2321	 */
2322	for (c = 0; c < children; c++) {
2323		char *cdevid, *cpath;
2324		uint64_t tmptxg;
2325
2326		cpath = NULL;
2327		cdevid = NULL;
2328		if (nvlist_lookup_string(child[c], ZPOOL_CONFIG_PHYS_PATH,
2329		    &cpath) != 0 && nvlist_lookup_string(child[c],
2330		    ZPOOL_CONFIG_DEVID, &cdevid) != 0)
2331			return (EINVAL);
2332		if ((spa_check_rootconf(cpath, cdevid, NULL,
2333		    &tmptxg) == 0) && (tmptxg > txg)) {
2334			txg = tmptxg;
2335			VERIFY(nvlist_lookup_string(child[c],
2336			    ZPOOL_CONFIG_PATH, &bootpath) == 0);
2337		}
2338	}
2339
2340	/* Does the best device match the one we've booted from? */
2341	if (bootpath) {
2342		cmn_err(CE_NOTE, "try booting from '%s'", bootpath);
2343		return (EINVAL);
2344	}
2345out:
2346	*bestconf = conf;
2347	return (0);
2348}
2349
2350/*
2351 * Import a root pool.
2352 *
2353 * For x86. devpath_list will consist of devid and/or physpath name of
2354 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
2355 * The GRUB "findroot" command will return the vdev we should boot.
2356 *
2357 * For Sparc, devpath_list consists the physpath name of the booting device
2358 * no matter the rootpool is a single device pool or a mirrored pool.
2359 * e.g.
2360 *	"/pci@1f,0/ide@d/disk@0,0:a"
2361 */
2362int
2363spa_import_rootpool(char *devpath, char *devid)
2364{
2365	nvlist_t *conf = NULL;
2366	char *pname;
2367	int error;
2368	spa_t *spa;
2369
2370	/*
2371	 * Get the vdev pathname and configuation from the most
2372	 * recently updated vdev (highest txg).
2373	 */
2374	if (error = spa_get_rootconf(devpath, devid, &conf))
2375		goto msg_out;
2376
2377	/*
2378	 * Add type "root" vdev to the config.
2379	 */
2380	spa_build_rootpool_config(conf);
2381
2382	VERIFY(nvlist_lookup_string(conf, ZPOOL_CONFIG_POOL_NAME, &pname) == 0);
2383
2384	mutex_enter(&spa_namespace_lock);
2385	if ((spa = spa_lookup(pname)) != NULL) {
2386		/*
2387		 * Remove the existing root pool from the namespace so that we
2388		 * can replace it with the correct config we just read in.
2389		 */
2390		spa_remove(spa);
2391	}
2392
2393	spa = spa_add(pname, NULL);
2394
2395	spa->spa_is_root = B_TRUE;
2396	VERIFY(nvlist_dup(conf, &spa->spa_config, 0) == 0);
2397	mutex_exit(&spa_namespace_lock);
2398
2399	nvlist_free(conf);
2400	return (0);
2401
2402msg_out:
2403	cmn_err(CE_NOTE, "\n"
2404	    "  ***************************************************  \n"
2405	    "  *  This device is not bootable!                   *  \n"
2406	    "  *  It is either offlined or detached or faulted.  *  \n"
2407	    "  *  Please try to boot from a different device.    *  \n"
2408	    "  ***************************************************  ");
2409
2410	return (error);
2411}
2412#endif
2413
2414/*
2415 * Take a pool and insert it into the namespace as if it had been loaded at
2416 * boot.
2417 */
2418int
2419spa_import_verbatim(const char *pool, nvlist_t *config, nvlist_t *props)
2420{
2421	spa_t *spa;
2422	char *altroot = NULL;
2423
2424	mutex_enter(&spa_namespace_lock);
2425	if (spa_lookup(pool) != NULL) {
2426		mutex_exit(&spa_namespace_lock);
2427		return (EEXIST);
2428	}
2429
2430	(void) nvlist_lookup_string(props,
2431	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2432	spa = spa_add(pool, altroot);
2433
2434	VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
2435
2436	if (props != NULL)
2437		spa_configfile_set(spa, props, B_FALSE);
2438
2439	spa_config_sync(spa, B_FALSE, B_TRUE);
2440
2441	mutex_exit(&spa_namespace_lock);
2442
2443	return (0);
2444}
2445
2446/*
2447 * Import a non-root pool into the system.
2448 */
2449int
2450spa_import(const char *pool, nvlist_t *config, nvlist_t *props)
2451{
2452	spa_t *spa;
2453	char *altroot = NULL;
2454	int error;
2455	nvlist_t *nvroot;
2456	nvlist_t **spares, **l2cache;
2457	uint_t nspares, nl2cache;
2458
2459	/*
2460	 * If a pool with this name exists, return failure.
2461	 */
2462	mutex_enter(&spa_namespace_lock);
2463	if ((spa = spa_lookup(pool)) != NULL) {
2464		mutex_exit(&spa_namespace_lock);
2465		return (EEXIST);
2466	}
2467
2468	/*
2469	 * Create and initialize the spa structure.
2470	 */
2471	(void) nvlist_lookup_string(props,
2472	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2473	spa = spa_add(pool, altroot);
2474	spa_activate(spa, spa_mode_global);
2475
2476	/*
2477	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
2478	 * because the user-supplied config is actually the one to trust when
2479	 * doing an import.
2480	 */
2481	error = spa_load(spa, config, SPA_LOAD_IMPORT, B_TRUE);
2482
2483	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2484	/*
2485	 * Toss any existing sparelist, as it doesn't have any validity
2486	 * anymore, and conflicts with spa_has_spare().
2487	 */
2488	if (spa->spa_spares.sav_config) {
2489		nvlist_free(spa->spa_spares.sav_config);
2490		spa->spa_spares.sav_config = NULL;
2491		spa_load_spares(spa);
2492	}
2493	if (spa->spa_l2cache.sav_config) {
2494		nvlist_free(spa->spa_l2cache.sav_config);
2495		spa->spa_l2cache.sav_config = NULL;
2496		spa_load_l2cache(spa);
2497	}
2498
2499	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2500	    &nvroot) == 0);
2501	if (error == 0)
2502		error = spa_validate_aux(spa, nvroot, -1ULL,
2503		    VDEV_ALLOC_SPARE);
2504	if (error == 0)
2505		error = spa_validate_aux(spa, nvroot, -1ULL,
2506		    VDEV_ALLOC_L2CACHE);
2507	spa_config_exit(spa, SCL_ALL, FTAG);
2508
2509	if (props != NULL)
2510		spa_configfile_set(spa, props, B_FALSE);
2511
2512	if (error != 0 || (props && spa_writeable(spa) &&
2513	    (error = spa_prop_set(spa, props)))) {
2514		spa_unload(spa);
2515		spa_deactivate(spa);
2516		spa_remove(spa);
2517		mutex_exit(&spa_namespace_lock);
2518		return (error);
2519	}
2520
2521	/*
2522	 * Override any spares and level 2 cache devices as specified by
2523	 * the user, as these may have correct device names/devids, etc.
2524	 */
2525	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2526	    &spares, &nspares) == 0) {
2527		if (spa->spa_spares.sav_config)
2528			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
2529			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
2530		else
2531			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
2532			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2533		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2534		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2535		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2536		spa_load_spares(spa);
2537		spa_config_exit(spa, SCL_ALL, FTAG);
2538		spa->spa_spares.sav_sync = B_TRUE;
2539	}
2540	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2541	    &l2cache, &nl2cache) == 0) {
2542		if (spa->spa_l2cache.sav_config)
2543			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
2544			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
2545		else
2546			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2547			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2548		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2549		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2550		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2551		spa_load_l2cache(spa);
2552		spa_config_exit(spa, SCL_ALL, FTAG);
2553		spa->spa_l2cache.sav_sync = B_TRUE;
2554	}
2555
2556	if (spa_writeable(spa)) {
2557		/*
2558		 * Update the config cache to include the newly-imported pool.
2559		 */
2560		spa_config_update_common(spa, SPA_CONFIG_UPDATE_POOL, B_FALSE);
2561	}
2562
2563	mutex_exit(&spa_namespace_lock);
2564
2565	return (0);
2566}
2567
2568
2569/*
2570 * This (illegal) pool name is used when temporarily importing a spa_t in order
2571 * to get the vdev stats associated with the imported devices.
2572 */
2573#define	TRYIMPORT_NAME	"$import"
2574
2575nvlist_t *
2576spa_tryimport(nvlist_t *tryconfig)
2577{
2578	nvlist_t *config = NULL;
2579	char *poolname;
2580	spa_t *spa;
2581	uint64_t state;
2582	int error;
2583
2584	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
2585		return (NULL);
2586
2587	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
2588		return (NULL);
2589
2590	/*
2591	 * Create and initialize the spa structure.
2592	 */
2593	mutex_enter(&spa_namespace_lock);
2594	spa = spa_add(TRYIMPORT_NAME, NULL);
2595	spa_activate(spa, FREAD);
2596
2597	/*
2598	 * Pass off the heavy lifting to spa_load().
2599	 * Pass TRUE for mosconfig because the user-supplied config
2600	 * is actually the one to trust when doing an import.
2601	 */
2602	error = spa_load(spa, tryconfig, SPA_LOAD_TRYIMPORT, B_TRUE);
2603
2604	/*
2605	 * If 'tryconfig' was at least parsable, return the current config.
2606	 */
2607	if (spa->spa_root_vdev != NULL) {
2608		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2609		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
2610		    poolname) == 0);
2611		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
2612		    state) == 0);
2613		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
2614		    spa->spa_uberblock.ub_timestamp) == 0);
2615
2616		/*
2617		 * If the bootfs property exists on this pool then we
2618		 * copy it out so that external consumers can tell which
2619		 * pools are bootable.
2620		 */
2621		if ((!error || error == EEXIST) && spa->spa_bootfs) {
2622			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2623
2624			/*
2625			 * We have to play games with the name since the
2626			 * pool was opened as TRYIMPORT_NAME.
2627			 */
2628			if (dsl_dsobj_to_dsname(spa_name(spa),
2629			    spa->spa_bootfs, tmpname) == 0) {
2630				char *cp;
2631				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2632
2633				cp = strchr(tmpname, '/');
2634				if (cp == NULL) {
2635					(void) strlcpy(dsname, tmpname,
2636					    MAXPATHLEN);
2637				} else {
2638					(void) snprintf(dsname, MAXPATHLEN,
2639					    "%s/%s", poolname, ++cp);
2640				}
2641				VERIFY(nvlist_add_string(config,
2642				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
2643				kmem_free(dsname, MAXPATHLEN);
2644			}
2645			kmem_free(tmpname, MAXPATHLEN);
2646		}
2647
2648		/*
2649		 * Add the list of hot spares and level 2 cache devices.
2650		 */
2651		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2652		spa_add_spares(spa, config);
2653		spa_add_l2cache(spa, config);
2654		spa_config_exit(spa, SCL_CONFIG, FTAG);
2655	}
2656
2657	spa_unload(spa);
2658	spa_deactivate(spa);
2659	spa_remove(spa);
2660	mutex_exit(&spa_namespace_lock);
2661
2662	return (config);
2663}
2664
2665/*
2666 * Pool export/destroy
2667 *
2668 * The act of destroying or exporting a pool is very simple.  We make sure there
2669 * is no more pending I/O and any references to the pool are gone.  Then, we
2670 * update the pool state and sync all the labels to disk, removing the
2671 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
2672 * we don't sync the labels or remove the configuration cache.
2673 */
2674static int
2675spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
2676    boolean_t force, boolean_t hardforce)
2677{
2678	spa_t *spa;
2679
2680	if (oldconfig)
2681		*oldconfig = NULL;
2682
2683	if (!(spa_mode_global & FWRITE))
2684		return (EROFS);
2685
2686	mutex_enter(&spa_namespace_lock);
2687	if ((spa = spa_lookup(pool)) == NULL) {
2688		mutex_exit(&spa_namespace_lock);
2689		return (ENOENT);
2690	}
2691
2692	/*
2693	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
2694	 * reacquire the namespace lock, and see if we can export.
2695	 */
2696	spa_open_ref(spa, FTAG);
2697	mutex_exit(&spa_namespace_lock);
2698	spa_async_suspend(spa);
2699	mutex_enter(&spa_namespace_lock);
2700	spa_close(spa, FTAG);
2701
2702	/*
2703	 * The pool will be in core if it's openable,
2704	 * in which case we can modify its state.
2705	 */
2706	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
2707		/*
2708		 * Objsets may be open only because they're dirty, so we
2709		 * have to force it to sync before checking spa_refcnt.
2710		 */
2711		txg_wait_synced(spa->spa_dsl_pool, 0);
2712
2713		/*
2714		 * A pool cannot be exported or destroyed if there are active
2715		 * references.  If we are resetting a pool, allow references by
2716		 * fault injection handlers.
2717		 */
2718		if (!spa_refcount_zero(spa) ||
2719		    (spa->spa_inject_ref != 0 &&
2720		    new_state != POOL_STATE_UNINITIALIZED)) {
2721			spa_async_resume(spa);
2722			mutex_exit(&spa_namespace_lock);
2723			return (EBUSY);
2724		}
2725
2726		/*
2727		 * A pool cannot be exported if it has an active shared spare.
2728		 * This is to prevent other pools stealing the active spare
2729		 * from an exported pool. At user's own will, such pool can
2730		 * be forcedly exported.
2731		 */
2732		if (!force && new_state == POOL_STATE_EXPORTED &&
2733		    spa_has_active_shared_spare(spa)) {
2734			spa_async_resume(spa);
2735			mutex_exit(&spa_namespace_lock);
2736			return (EXDEV);
2737		}
2738
2739		/*
2740		 * We want this to be reflected on every label,
2741		 * so mark them all dirty.  spa_unload() will do the
2742		 * final sync that pushes these changes out.
2743		 */
2744		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
2745			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2746			spa->spa_state = new_state;
2747			spa->spa_final_txg = spa_last_synced_txg(spa) + 1;
2748			vdev_config_dirty(spa->spa_root_vdev);
2749			spa_config_exit(spa, SCL_ALL, FTAG);
2750		}
2751	}
2752
2753	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
2754
2755	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
2756		spa_unload(spa);
2757		spa_deactivate(spa);
2758	}
2759
2760	if (oldconfig && spa->spa_config)
2761		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
2762
2763	if (new_state != POOL_STATE_UNINITIALIZED) {
2764		if (!hardforce)
2765			spa_config_sync(spa, B_TRUE, B_TRUE);
2766		spa_remove(spa);
2767	}
2768	mutex_exit(&spa_namespace_lock);
2769
2770	return (0);
2771}
2772
2773/*
2774 * Destroy a storage pool.
2775 */
2776int
2777spa_destroy(char *pool)
2778{
2779	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
2780	    B_FALSE, B_FALSE));
2781}
2782
2783/*
2784 * Export a storage pool.
2785 */
2786int
2787spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
2788    boolean_t hardforce)
2789{
2790	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
2791	    force, hardforce));
2792}
2793
2794/*
2795 * Similar to spa_export(), this unloads the spa_t without actually removing it
2796 * from the namespace in any way.
2797 */
2798int
2799spa_reset(char *pool)
2800{
2801	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
2802	    B_FALSE, B_FALSE));
2803}
2804
2805/*
2806 * ==========================================================================
2807 * Device manipulation
2808 * ==========================================================================
2809 */
2810
2811/*
2812 * Add a device to a storage pool.
2813 */
2814int
2815spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
2816{
2817	uint64_t txg;
2818	int error;
2819	vdev_t *rvd = spa->spa_root_vdev;
2820	vdev_t *vd, *tvd;
2821	nvlist_t **spares, **l2cache;
2822	uint_t nspares, nl2cache;
2823
2824	txg = spa_vdev_enter(spa);
2825
2826	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
2827	    VDEV_ALLOC_ADD)) != 0)
2828		return (spa_vdev_exit(spa, NULL, txg, error));
2829
2830	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
2831
2832	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
2833	    &nspares) != 0)
2834		nspares = 0;
2835
2836	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
2837	    &nl2cache) != 0)
2838		nl2cache = 0;
2839
2840	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
2841		return (spa_vdev_exit(spa, vd, txg, EINVAL));
2842
2843	if (vd->vdev_children != 0 &&
2844	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
2845		return (spa_vdev_exit(spa, vd, txg, error));
2846
2847	/*
2848	 * We must validate the spares and l2cache devices after checking the
2849	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
2850	 */
2851	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
2852		return (spa_vdev_exit(spa, vd, txg, error));
2853
2854	/*
2855	 * Transfer each new top-level vdev from vd to rvd.
2856	 */
2857	for (int c = 0; c < vd->vdev_children; c++) {
2858		tvd = vd->vdev_child[c];
2859		vdev_remove_child(vd, tvd);
2860		tvd->vdev_id = rvd->vdev_children;
2861		vdev_add_child(rvd, tvd);
2862		vdev_config_dirty(tvd);
2863	}
2864
2865	if (nspares != 0) {
2866		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
2867		    ZPOOL_CONFIG_SPARES);
2868		spa_load_spares(spa);
2869		spa->spa_spares.sav_sync = B_TRUE;
2870	}
2871
2872	if (nl2cache != 0) {
2873		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
2874		    ZPOOL_CONFIG_L2CACHE);
2875		spa_load_l2cache(spa);
2876		spa->spa_l2cache.sav_sync = B_TRUE;
2877	}
2878
2879	/*
2880	 * We have to be careful when adding new vdevs to an existing pool.
2881	 * If other threads start allocating from these vdevs before we
2882	 * sync the config cache, and we lose power, then upon reboot we may
2883	 * fail to open the pool because there are DVAs that the config cache
2884	 * can't translate.  Therefore, we first add the vdevs without
2885	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
2886	 * and then let spa_config_update() initialize the new metaslabs.
2887	 *
2888	 * spa_load() checks for added-but-not-initialized vdevs, so that
2889	 * if we lose power at any point in this sequence, the remaining
2890	 * steps will be completed the next time we load the pool.
2891	 */
2892	(void) spa_vdev_exit(spa, vd, txg, 0);
2893
2894	mutex_enter(&spa_namespace_lock);
2895	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
2896	mutex_exit(&spa_namespace_lock);
2897
2898	return (0);
2899}
2900
2901/*
2902 * Attach a device to a mirror.  The arguments are the path to any device
2903 * in the mirror, and the nvroot for the new device.  If the path specifies
2904 * a device that is not mirrored, we automatically insert the mirror vdev.
2905 *
2906 * If 'replacing' is specified, the new device is intended to replace the
2907 * existing device; in this case the two devices are made into their own
2908 * mirror using the 'replacing' vdev, which is functionally identical to
2909 * the mirror vdev (it actually reuses all the same ops) but has a few
2910 * extra rules: you can't attach to it after it's been created, and upon
2911 * completion of resilvering, the first disk (the one being replaced)
2912 * is automatically detached.
2913 */
2914int
2915spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
2916{
2917	uint64_t txg, open_txg;
2918	vdev_t *rvd = spa->spa_root_vdev;
2919	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
2920	vdev_ops_t *pvops;
2921	dmu_tx_t *tx;
2922	char *oldvdpath, *newvdpath;
2923	int newvd_isspare;
2924	int error;
2925
2926	txg = spa_vdev_enter(spa);
2927
2928	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
2929
2930	if (oldvd == NULL)
2931		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
2932
2933	if (!oldvd->vdev_ops->vdev_op_leaf)
2934		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
2935
2936	pvd = oldvd->vdev_parent;
2937
2938	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
2939	    VDEV_ALLOC_ADD)) != 0)
2940		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
2941
2942	if (newrootvd->vdev_children != 1)
2943		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
2944
2945	newvd = newrootvd->vdev_child[0];
2946
2947	if (!newvd->vdev_ops->vdev_op_leaf)
2948		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
2949
2950	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
2951		return (spa_vdev_exit(spa, newrootvd, txg, error));
2952
2953	/*
2954	 * Spares can't replace logs
2955	 */
2956	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
2957		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2958
2959	if (!replacing) {
2960		/*
2961		 * For attach, the only allowable parent is a mirror or the root
2962		 * vdev.
2963		 */
2964		if (pvd->vdev_ops != &vdev_mirror_ops &&
2965		    pvd->vdev_ops != &vdev_root_ops)
2966			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2967
2968		pvops = &vdev_mirror_ops;
2969	} else {
2970		/*
2971		 * Active hot spares can only be replaced by inactive hot
2972		 * spares.
2973		 */
2974		if (pvd->vdev_ops == &vdev_spare_ops &&
2975		    pvd->vdev_child[1] == oldvd &&
2976		    !spa_has_spare(spa, newvd->vdev_guid))
2977			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2978
2979		/*
2980		 * If the source is a hot spare, and the parent isn't already a
2981		 * spare, then we want to create a new hot spare.  Otherwise, we
2982		 * want to create a replacing vdev.  The user is not allowed to
2983		 * attach to a spared vdev child unless the 'isspare' state is
2984		 * the same (spare replaces spare, non-spare replaces
2985		 * non-spare).
2986		 */
2987		if (pvd->vdev_ops == &vdev_replacing_ops)
2988			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2989		else if (pvd->vdev_ops == &vdev_spare_ops &&
2990		    newvd->vdev_isspare != oldvd->vdev_isspare)
2991			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2992		else if (pvd->vdev_ops != &vdev_spare_ops &&
2993		    newvd->vdev_isspare)
2994			pvops = &vdev_spare_ops;
2995		else
2996			pvops = &vdev_replacing_ops;
2997	}
2998
2999	/*
3000	 * Compare the new device size with the replaceable/attachable
3001	 * device size.
3002	 */
3003	if (newvd->vdev_psize < vdev_get_rsize(oldvd))
3004		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3005
3006	/*
3007	 * The new device cannot have a higher alignment requirement
3008	 * than the top-level vdev.
3009	 */
3010	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3011		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3012
3013	/*
3014	 * If this is an in-place replacement, update oldvd's path and devid
3015	 * to make it distinguishable from newvd, and unopenable from now on.
3016	 */
3017	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3018		spa_strfree(oldvd->vdev_path);
3019		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3020		    KM_SLEEP);
3021		(void) sprintf(oldvd->vdev_path, "%s/%s",
3022		    newvd->vdev_path, "old");
3023		if (oldvd->vdev_devid != NULL) {
3024			spa_strfree(oldvd->vdev_devid);
3025			oldvd->vdev_devid = NULL;
3026		}
3027	}
3028
3029	/*
3030	 * If the parent is not a mirror, or if we're replacing, insert the new
3031	 * mirror/replacing/spare vdev above oldvd.
3032	 */
3033	if (pvd->vdev_ops != pvops)
3034		pvd = vdev_add_parent(oldvd, pvops);
3035
3036	ASSERT(pvd->vdev_top->vdev_parent == rvd);
3037	ASSERT(pvd->vdev_ops == pvops);
3038	ASSERT(oldvd->vdev_parent == pvd);
3039
3040	/*
3041	 * Extract the new device from its root and add it to pvd.
3042	 */
3043	vdev_remove_child(newrootvd, newvd);
3044	newvd->vdev_id = pvd->vdev_children;
3045	vdev_add_child(pvd, newvd);
3046
3047	/*
3048	 * If newvd is smaller than oldvd, but larger than its rsize,
3049	 * the addition of newvd may have decreased our parent's asize.
3050	 */
3051	pvd->vdev_asize = MIN(pvd->vdev_asize, newvd->vdev_asize);
3052
3053	tvd = newvd->vdev_top;
3054	ASSERT(pvd->vdev_top == tvd);
3055	ASSERT(tvd->vdev_parent == rvd);
3056
3057	vdev_config_dirty(tvd);
3058
3059	/*
3060	 * Set newvd's DTL to [TXG_INITIAL, open_txg].  It will propagate
3061	 * upward when spa_vdev_exit() calls vdev_dtl_reassess().
3062	 */
3063	open_txg = txg + TXG_CONCURRENT_STATES - 1;
3064
3065	vdev_dtl_dirty(newvd, DTL_MISSING,
3066	    TXG_INITIAL, open_txg - TXG_INITIAL + 1);
3067
3068	if (newvd->vdev_isspare) {
3069		spa_spare_activate(newvd);
3070		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
3071	}
3072
3073	oldvdpath = spa_strdup(oldvd->vdev_path);
3074	newvdpath = spa_strdup(newvd->vdev_path);
3075	newvd_isspare = newvd->vdev_isspare;
3076
3077	/*
3078	 * Mark newvd's DTL dirty in this txg.
3079	 */
3080	vdev_dirty(tvd, VDD_DTL, newvd, txg);
3081
3082	(void) spa_vdev_exit(spa, newrootvd, open_txg, 0);
3083
3084	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
3085	if (dmu_tx_assign(tx, TXG_WAIT) == 0) {
3086		spa_history_internal_log(LOG_POOL_VDEV_ATTACH, spa, tx,
3087		    CRED(),  "%s vdev=%s %s vdev=%s",
3088		    replacing && newvd_isspare ? "spare in" :
3089		    replacing ? "replace" : "attach", newvdpath,
3090		    replacing ? "for" : "to", oldvdpath);
3091		dmu_tx_commit(tx);
3092	} else {
3093		dmu_tx_abort(tx);
3094	}
3095
3096	spa_strfree(oldvdpath);
3097	spa_strfree(newvdpath);
3098
3099	/*
3100	 * Kick off a resilver to update newvd.
3101	 */
3102	VERIFY3U(spa_scrub(spa, POOL_SCRUB_RESILVER), ==, 0);
3103
3104	return (0);
3105}
3106
3107/*
3108 * Detach a device from a mirror or replacing vdev.
3109 * If 'replace_done' is specified, only detach if the parent
3110 * is a replacing vdev.
3111 */
3112int
3113spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
3114{
3115	uint64_t txg;
3116	int error;
3117	vdev_t *rvd = spa->spa_root_vdev;
3118	vdev_t *vd, *pvd, *cvd, *tvd;
3119	boolean_t unspare = B_FALSE;
3120	uint64_t unspare_guid;
3121	size_t len;
3122
3123	txg = spa_vdev_enter(spa);
3124
3125	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3126
3127	if (vd == NULL)
3128		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3129
3130	if (!vd->vdev_ops->vdev_op_leaf)
3131		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3132
3133	pvd = vd->vdev_parent;
3134
3135	/*
3136	 * If the parent/child relationship is not as expected, don't do it.
3137	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
3138	 * vdev that's replacing B with C.  The user's intent in replacing
3139	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
3140	 * the replace by detaching C, the expected behavior is to end up
3141	 * M(A,B).  But suppose that right after deciding to detach C,
3142	 * the replacement of B completes.  We would have M(A,C), and then
3143	 * ask to detach C, which would leave us with just A -- not what
3144	 * the user wanted.  To prevent this, we make sure that the
3145	 * parent/child relationship hasn't changed -- in this example,
3146	 * that C's parent is still the replacing vdev R.
3147	 */
3148	if (pvd->vdev_guid != pguid && pguid != 0)
3149		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3150
3151	/*
3152	 * If replace_done is specified, only remove this device if it's
3153	 * the first child of a replacing vdev.  For the 'spare' vdev, either
3154	 * disk can be removed.
3155	 */
3156	if (replace_done) {
3157		if (pvd->vdev_ops == &vdev_replacing_ops) {
3158			if (vd->vdev_id != 0)
3159				return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3160		} else if (pvd->vdev_ops != &vdev_spare_ops) {
3161			return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3162		}
3163	}
3164
3165	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
3166	    spa_version(spa) >= SPA_VERSION_SPARES);
3167
3168	/*
3169	 * Only mirror, replacing, and spare vdevs support detach.
3170	 */
3171	if (pvd->vdev_ops != &vdev_replacing_ops &&
3172	    pvd->vdev_ops != &vdev_mirror_ops &&
3173	    pvd->vdev_ops != &vdev_spare_ops)
3174		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3175
3176	/*
3177	 * If this device has the only valid copy of some data,
3178	 * we cannot safely detach it.
3179	 */
3180	if (vdev_dtl_required(vd))
3181		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3182
3183	ASSERT(pvd->vdev_children >= 2);
3184
3185	/*
3186	 * If we are detaching the second disk from a replacing vdev, then
3187	 * check to see if we changed the original vdev's path to have "/old"
3188	 * at the end in spa_vdev_attach().  If so, undo that change now.
3189	 */
3190	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id == 1 &&
3191	    pvd->vdev_child[0]->vdev_path != NULL &&
3192	    pvd->vdev_child[1]->vdev_path != NULL) {
3193		ASSERT(pvd->vdev_child[1] == vd);
3194		cvd = pvd->vdev_child[0];
3195		len = strlen(vd->vdev_path);
3196		if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
3197		    strcmp(cvd->vdev_path + len, "/old") == 0) {
3198			spa_strfree(cvd->vdev_path);
3199			cvd->vdev_path = spa_strdup(vd->vdev_path);
3200		}
3201	}
3202
3203	/*
3204	 * If we are detaching the original disk from a spare, then it implies
3205	 * that the spare should become a real disk, and be removed from the
3206	 * active spare list for the pool.
3207	 */
3208	if (pvd->vdev_ops == &vdev_spare_ops &&
3209	    vd->vdev_id == 0 && pvd->vdev_child[1]->vdev_isspare)
3210		unspare = B_TRUE;
3211
3212	/*
3213	 * Erase the disk labels so the disk can be used for other things.
3214	 * This must be done after all other error cases are handled,
3215	 * but before we disembowel vd (so we can still do I/O to it).
3216	 * But if we can't do it, don't treat the error as fatal --
3217	 * it may be that the unwritability of the disk is the reason
3218	 * it's being detached!
3219	 */
3220	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
3221
3222	/*
3223	 * Remove vd from its parent and compact the parent's children.
3224	 */
3225	vdev_remove_child(pvd, vd);
3226	vdev_compact_children(pvd);
3227
3228	/*
3229	 * Remember one of the remaining children so we can get tvd below.
3230	 */
3231	cvd = pvd->vdev_child[0];
3232
3233	/*
3234	 * If we need to remove the remaining child from the list of hot spares,
3235	 * do it now, marking the vdev as no longer a spare in the process.
3236	 * We must do this before vdev_remove_parent(), because that can
3237	 * change the GUID if it creates a new toplevel GUID.  For a similar
3238	 * reason, we must remove the spare now, in the same txg as the detach;
3239	 * otherwise someone could attach a new sibling, change the GUID, and
3240	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
3241	 */
3242	if (unspare) {
3243		ASSERT(cvd->vdev_isspare);
3244		spa_spare_remove(cvd);
3245		unspare_guid = cvd->vdev_guid;
3246		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
3247	}
3248
3249	/*
3250	 * If the parent mirror/replacing vdev only has one child,
3251	 * the parent is no longer needed.  Remove it from the tree.
3252	 */
3253	if (pvd->vdev_children == 1)
3254		vdev_remove_parent(cvd);
3255
3256	/*
3257	 * We don't set tvd until now because the parent we just removed
3258	 * may have been the previous top-level vdev.
3259	 */
3260	tvd = cvd->vdev_top;
3261	ASSERT(tvd->vdev_parent == rvd);
3262
3263	/*
3264	 * Reevaluate the parent vdev state.
3265	 */
3266	vdev_propagate_state(cvd);
3267
3268	/*
3269	 * If the device we just detached was smaller than the others, it may be
3270	 * possible to add metaslabs (i.e. grow the pool).  vdev_metaslab_init()
3271	 * can't fail because the existing metaslabs are already in core, so
3272	 * there's nothing to read from disk.
3273	 */
3274	VERIFY(vdev_metaslab_init(tvd, txg) == 0);
3275
3276	vdev_config_dirty(tvd);
3277
3278	/*
3279	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
3280	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
3281	 * But first make sure we're not on any *other* txg's DTL list, to
3282	 * prevent vd from being accessed after it's freed.
3283	 */
3284	for (int t = 0; t < TXG_SIZE; t++)
3285		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
3286	vd->vdev_detached = B_TRUE;
3287	vdev_dirty(tvd, VDD_DTL, vd, txg);
3288
3289	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
3290
3291	error = spa_vdev_exit(spa, vd, txg, 0);
3292
3293	/*
3294	 * If this was the removal of the original device in a hot spare vdev,
3295	 * then we want to go through and remove the device from the hot spare
3296	 * list of every other pool.
3297	 */
3298	if (unspare) {
3299		spa_t *myspa = spa;
3300		spa = NULL;
3301		mutex_enter(&spa_namespace_lock);
3302		while ((spa = spa_next(spa)) != NULL) {
3303			if (spa->spa_state != POOL_STATE_ACTIVE)
3304				continue;
3305			if (spa == myspa)
3306				continue;
3307			spa_open_ref(spa, FTAG);
3308			mutex_exit(&spa_namespace_lock);
3309			(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
3310			mutex_enter(&spa_namespace_lock);
3311			spa_close(spa, FTAG);
3312		}
3313		mutex_exit(&spa_namespace_lock);
3314	}
3315
3316	return (error);
3317}
3318
3319static nvlist_t *
3320spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
3321{
3322	for (int i = 0; i < count; i++) {
3323		uint64_t guid;
3324
3325		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
3326		    &guid) == 0);
3327
3328		if (guid == target_guid)
3329			return (nvpp[i]);
3330	}
3331
3332	return (NULL);
3333}
3334
3335static void
3336spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
3337	nvlist_t *dev_to_remove)
3338{
3339	nvlist_t **newdev = NULL;
3340
3341	if (count > 1)
3342		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
3343
3344	for (int i = 0, j = 0; i < count; i++) {
3345		if (dev[i] == dev_to_remove)
3346			continue;
3347		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
3348	}
3349
3350	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
3351	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
3352
3353	for (int i = 0; i < count - 1; i++)
3354		nvlist_free(newdev[i]);
3355
3356	if (count > 1)
3357		kmem_free(newdev, (count - 1) * sizeof (void *));
3358}
3359
3360/*
3361 * Remove a device from the pool.  Currently, this supports removing only hot
3362 * spares and level 2 ARC devices.
3363 */
3364int
3365spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
3366{
3367	vdev_t *vd;
3368	nvlist_t **spares, **l2cache, *nv;
3369	uint_t nspares, nl2cache;
3370	uint64_t txg = 0;
3371	int error = 0;
3372	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
3373
3374	if (!locked)
3375		txg = spa_vdev_enter(spa);
3376
3377	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3378
3379	if (spa->spa_spares.sav_vdevs != NULL &&
3380	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3381	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
3382	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
3383		/*
3384		 * Only remove the hot spare if it's not currently in use
3385		 * in this pool.
3386		 */
3387		if (vd == NULL || unspare) {
3388			spa_vdev_remove_aux(spa->spa_spares.sav_config,
3389			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
3390			spa_load_spares(spa);
3391			spa->spa_spares.sav_sync = B_TRUE;
3392		} else {
3393			error = EBUSY;
3394		}
3395	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
3396	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3397	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
3398	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
3399		/*
3400		 * Cache devices can always be removed.
3401		 */
3402		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
3403		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
3404		spa_load_l2cache(spa);
3405		spa->spa_l2cache.sav_sync = B_TRUE;
3406	} else if (vd != NULL) {
3407		/*
3408		 * Normal vdevs cannot be removed (yet).
3409		 */
3410		error = ENOTSUP;
3411	} else {
3412		/*
3413		 * There is no vdev of any kind with the specified guid.
3414		 */
3415		error = ENOENT;
3416	}
3417
3418	if (!locked)
3419		return (spa_vdev_exit(spa, NULL, txg, error));
3420
3421	return (error);
3422}
3423
3424/*
3425 * Find any device that's done replacing, or a vdev marked 'unspare' that's
3426 * current spared, so we can detach it.
3427 */
3428static vdev_t *
3429spa_vdev_resilver_done_hunt(vdev_t *vd)
3430{
3431	vdev_t *newvd, *oldvd;
3432	int c;
3433
3434	for (c = 0; c < vd->vdev_children; c++) {
3435		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
3436		if (oldvd != NULL)
3437			return (oldvd);
3438	}
3439
3440	/*
3441	 * Check for a completed replacement.
3442	 */
3443	if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
3444		oldvd = vd->vdev_child[0];
3445		newvd = vd->vdev_child[1];
3446
3447		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
3448		    !vdev_dtl_required(oldvd))
3449			return (oldvd);
3450	}
3451
3452	/*
3453	 * Check for a completed resilver with the 'unspare' flag set.
3454	 */
3455	if (vd->vdev_ops == &vdev_spare_ops && vd->vdev_children == 2) {
3456		newvd = vd->vdev_child[0];
3457		oldvd = vd->vdev_child[1];
3458
3459		if (newvd->vdev_unspare &&
3460		    vdev_dtl_empty(newvd, DTL_MISSING) &&
3461		    !vdev_dtl_required(oldvd)) {
3462			newvd->vdev_unspare = 0;
3463			return (oldvd);
3464		}
3465	}
3466
3467	return (NULL);
3468}
3469
3470static void
3471spa_vdev_resilver_done(spa_t *spa)
3472{
3473	vdev_t *vd, *pvd, *ppvd;
3474	uint64_t guid, sguid, pguid, ppguid;
3475
3476	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3477
3478	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
3479		pvd = vd->vdev_parent;
3480		ppvd = pvd->vdev_parent;
3481		guid = vd->vdev_guid;
3482		pguid = pvd->vdev_guid;
3483		ppguid = ppvd->vdev_guid;
3484		sguid = 0;
3485		/*
3486		 * If we have just finished replacing a hot spared device, then
3487		 * we need to detach the parent's first child (the original hot
3488		 * spare) as well.
3489		 */
3490		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0) {
3491			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
3492			ASSERT(ppvd->vdev_children == 2);
3493			sguid = ppvd->vdev_child[1]->vdev_guid;
3494		}
3495		spa_config_exit(spa, SCL_ALL, FTAG);
3496		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
3497			return;
3498		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
3499			return;
3500		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3501	}
3502
3503	spa_config_exit(spa, SCL_ALL, FTAG);
3504}
3505
3506/*
3507 * Update the stored path or FRU for this vdev.  Dirty the vdev configuration,
3508 * relying on spa_vdev_enter/exit() to synchronize the labels and cache.
3509 */
3510int
3511spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
3512    boolean_t ispath)
3513{
3514	vdev_t *vd;
3515	uint64_t txg;
3516
3517	txg = spa_vdev_enter(spa);
3518
3519	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3520		return (spa_vdev_exit(spa, NULL, txg, ENOENT));
3521
3522	if (!vd->vdev_ops->vdev_op_leaf)
3523		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3524
3525	if (ispath) {
3526		spa_strfree(vd->vdev_path);
3527		vd->vdev_path = spa_strdup(value);
3528	} else {
3529		if (vd->vdev_fru != NULL)
3530			spa_strfree(vd->vdev_fru);
3531		vd->vdev_fru = spa_strdup(value);
3532	}
3533
3534	vdev_config_dirty(vd->vdev_top);
3535
3536	return (spa_vdev_exit(spa, NULL, txg, 0));
3537}
3538
3539int
3540spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
3541{
3542	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
3543}
3544
3545int
3546spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
3547{
3548	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
3549}
3550
3551/*
3552 * ==========================================================================
3553 * SPA Scrubbing
3554 * ==========================================================================
3555 */
3556
3557int
3558spa_scrub(spa_t *spa, pool_scrub_type_t type)
3559{
3560	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
3561
3562	if ((uint_t)type >= POOL_SCRUB_TYPES)
3563		return (ENOTSUP);
3564
3565	/*
3566	 * If a resilver was requested, but there is no DTL on a
3567	 * writeable leaf device, we have nothing to do.
3568	 */
3569	if (type == POOL_SCRUB_RESILVER &&
3570	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
3571		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
3572		return (0);
3573	}
3574
3575	if (type == POOL_SCRUB_EVERYTHING &&
3576	    spa->spa_dsl_pool->dp_scrub_func != SCRUB_FUNC_NONE &&
3577	    spa->spa_dsl_pool->dp_scrub_isresilver)
3578		return (EBUSY);
3579
3580	if (type == POOL_SCRUB_EVERYTHING || type == POOL_SCRUB_RESILVER) {
3581		return (dsl_pool_scrub_clean(spa->spa_dsl_pool));
3582	} else if (type == POOL_SCRUB_NONE) {
3583		return (dsl_pool_scrub_cancel(spa->spa_dsl_pool));
3584	} else {
3585		return (EINVAL);
3586	}
3587}
3588
3589/*
3590 * ==========================================================================
3591 * SPA async task processing
3592 * ==========================================================================
3593 */
3594
3595static void
3596spa_async_remove(spa_t *spa, vdev_t *vd)
3597{
3598	if (vd->vdev_remove_wanted) {
3599		vd->vdev_remove_wanted = 0;
3600		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
3601		vdev_clear(spa, vd);
3602		vdev_state_dirty(vd->vdev_top);
3603	}
3604
3605	for (int c = 0; c < vd->vdev_children; c++)
3606		spa_async_remove(spa, vd->vdev_child[c]);
3607}
3608
3609static void
3610spa_async_probe(spa_t *spa, vdev_t *vd)
3611{
3612	if (vd->vdev_probe_wanted) {
3613		vd->vdev_probe_wanted = 0;
3614		vdev_reopen(vd);	/* vdev_open() does the actual probe */
3615	}
3616
3617	for (int c = 0; c < vd->vdev_children; c++)
3618		spa_async_probe(spa, vd->vdev_child[c]);
3619}
3620
3621static void
3622spa_async_thread(spa_t *spa)
3623{
3624	int tasks;
3625
3626	ASSERT(spa->spa_sync_on);
3627
3628	mutex_enter(&spa->spa_async_lock);
3629	tasks = spa->spa_async_tasks;
3630	spa->spa_async_tasks = 0;
3631	mutex_exit(&spa->spa_async_lock);
3632
3633	/*
3634	 * See if the config needs to be updated.
3635	 */
3636	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
3637		mutex_enter(&spa_namespace_lock);
3638		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3639		mutex_exit(&spa_namespace_lock);
3640	}
3641
3642	/*
3643	 * See if any devices need to be marked REMOVED.
3644	 */
3645	if (tasks & SPA_ASYNC_REMOVE) {
3646		spa_vdev_state_enter(spa);
3647		spa_async_remove(spa, spa->spa_root_vdev);
3648		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
3649			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
3650		for (int i = 0; i < spa->spa_spares.sav_count; i++)
3651			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
3652		(void) spa_vdev_state_exit(spa, NULL, 0);
3653	}
3654
3655	/*
3656	 * See if any devices need to be probed.
3657	 */
3658	if (tasks & SPA_ASYNC_PROBE) {
3659		spa_vdev_state_enter(spa);
3660		spa_async_probe(spa, spa->spa_root_vdev);
3661		(void) spa_vdev_state_exit(spa, NULL, 0);
3662	}
3663
3664	/*
3665	 * If any devices are done replacing, detach them.
3666	 */
3667	if (tasks & SPA_ASYNC_RESILVER_DONE)
3668		spa_vdev_resilver_done(spa);
3669
3670	/*
3671	 * Kick off a resilver.
3672	 */
3673	if (tasks & SPA_ASYNC_RESILVER)
3674		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER) == 0);
3675
3676	/*
3677	 * Let the world know that we're done.
3678	 */
3679	mutex_enter(&spa->spa_async_lock);
3680	spa->spa_async_thread = NULL;
3681	cv_broadcast(&spa->spa_async_cv);
3682	mutex_exit(&spa->spa_async_lock);
3683	thread_exit();
3684}
3685
3686void
3687spa_async_suspend(spa_t *spa)
3688{
3689	mutex_enter(&spa->spa_async_lock);
3690	spa->spa_async_suspended++;
3691	while (spa->spa_async_thread != NULL)
3692		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
3693	mutex_exit(&spa->spa_async_lock);
3694}
3695
3696void
3697spa_async_resume(spa_t *spa)
3698{
3699	mutex_enter(&spa->spa_async_lock);
3700	ASSERT(spa->spa_async_suspended != 0);
3701	spa->spa_async_suspended--;
3702	mutex_exit(&spa->spa_async_lock);
3703}
3704
3705static void
3706spa_async_dispatch(spa_t *spa)
3707{
3708	mutex_enter(&spa->spa_async_lock);
3709	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
3710	    spa->spa_async_thread == NULL &&
3711	    rootdir != NULL && !vn_is_readonly(rootdir))
3712		spa->spa_async_thread = thread_create(NULL, 0,
3713		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
3714	mutex_exit(&spa->spa_async_lock);
3715}
3716
3717void
3718spa_async_request(spa_t *spa, int task)
3719{
3720	mutex_enter(&spa->spa_async_lock);
3721	spa->spa_async_tasks |= task;
3722	mutex_exit(&spa->spa_async_lock);
3723}
3724
3725/*
3726 * ==========================================================================
3727 * SPA syncing routines
3728 * ==========================================================================
3729 */
3730
3731static void
3732spa_sync_deferred_frees(spa_t *spa, uint64_t txg)
3733{
3734	bplist_t *bpl = &spa->spa_sync_bplist;
3735	dmu_tx_t *tx;
3736	blkptr_t blk;
3737	uint64_t itor = 0;
3738	zio_t *zio;
3739	int error;
3740	uint8_t c = 1;
3741
3742	zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
3743
3744	while (bplist_iterate(bpl, &itor, &blk) == 0) {
3745		ASSERT(blk.blk_birth < txg);
3746		zio_nowait(zio_free(zio, spa, txg, &blk, NULL, NULL,
3747		    ZIO_FLAG_MUSTSUCCEED));
3748	}
3749
3750	error = zio_wait(zio);
3751	ASSERT3U(error, ==, 0);
3752
3753	tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3754	bplist_vacate(bpl, tx);
3755
3756	/*
3757	 * Pre-dirty the first block so we sync to convergence faster.
3758	 * (Usually only the first block is needed.)
3759	 */
3760	dmu_write(spa->spa_meta_objset, spa->spa_sync_bplist_obj, 0, 1, &c, tx);
3761	dmu_tx_commit(tx);
3762}
3763
3764static void
3765spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
3766{
3767	char *packed = NULL;
3768	size_t bufsize;
3769	size_t nvsize = 0;
3770	dmu_buf_t *db;
3771
3772	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
3773
3774	/*
3775	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
3776	 * information.  This avoids the dbuf_will_dirty() path and
3777	 * saves us a pre-read to get data we don't actually care about.
3778	 */
3779	bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
3780	packed = kmem_alloc(bufsize, KM_SLEEP);
3781
3782	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
3783	    KM_SLEEP) == 0);
3784	bzero(packed + nvsize, bufsize - nvsize);
3785
3786	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
3787
3788	kmem_free(packed, bufsize);
3789
3790	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
3791	dmu_buf_will_dirty(db, tx);
3792	*(uint64_t *)db->db_data = nvsize;
3793	dmu_buf_rele(db, FTAG);
3794}
3795
3796static void
3797spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
3798    const char *config, const char *entry)
3799{
3800	nvlist_t *nvroot;
3801	nvlist_t **list;
3802	int i;
3803
3804	if (!sav->sav_sync)
3805		return;
3806
3807	/*
3808	 * Update the MOS nvlist describing the list of available devices.
3809	 * spa_validate_aux() will have already made sure this nvlist is
3810	 * valid and the vdevs are labeled appropriately.
3811	 */
3812	if (sav->sav_object == 0) {
3813		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
3814		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
3815		    sizeof (uint64_t), tx);
3816		VERIFY(zap_update(spa->spa_meta_objset,
3817		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
3818		    &sav->sav_object, tx) == 0);
3819	}
3820
3821	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3822	if (sav->sav_count == 0) {
3823		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
3824	} else {
3825		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
3826		for (i = 0; i < sav->sav_count; i++)
3827			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
3828			    B_FALSE, B_FALSE, B_TRUE);
3829		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
3830		    sav->sav_count) == 0);
3831		for (i = 0; i < sav->sav_count; i++)
3832			nvlist_free(list[i]);
3833		kmem_free(list, sav->sav_count * sizeof (void *));
3834	}
3835
3836	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
3837	nvlist_free(nvroot);
3838
3839	sav->sav_sync = B_FALSE;
3840}
3841
3842static void
3843spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
3844{
3845	nvlist_t *config;
3846
3847	if (list_is_empty(&spa->spa_config_dirty_list))
3848		return;
3849
3850	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
3851
3852	config = spa_config_generate(spa, spa->spa_root_vdev,
3853	    dmu_tx_get_txg(tx), B_FALSE);
3854
3855	spa_config_exit(spa, SCL_STATE, FTAG);
3856
3857	if (spa->spa_config_syncing)
3858		nvlist_free(spa->spa_config_syncing);
3859	spa->spa_config_syncing = config;
3860
3861	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
3862}
3863
3864/*
3865 * Set zpool properties.
3866 */
3867static void
3868spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
3869{
3870	spa_t *spa = arg1;
3871	objset_t *mos = spa->spa_meta_objset;
3872	nvlist_t *nvp = arg2;
3873	nvpair_t *elem;
3874	uint64_t intval;
3875	char *strval;
3876	zpool_prop_t prop;
3877	const char *propname;
3878	zprop_type_t proptype;
3879
3880	mutex_enter(&spa->spa_props_lock);
3881
3882	elem = NULL;
3883	while ((elem = nvlist_next_nvpair(nvp, elem))) {
3884		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
3885		case ZPOOL_PROP_VERSION:
3886			/*
3887			 * Only set version for non-zpool-creation cases
3888			 * (set/import). spa_create() needs special care
3889			 * for version setting.
3890			 */
3891			if (tx->tx_txg != TXG_INITIAL) {
3892				VERIFY(nvpair_value_uint64(elem,
3893				    &intval) == 0);
3894				ASSERT(intval <= SPA_VERSION);
3895				ASSERT(intval >= spa_version(spa));
3896				spa->spa_uberblock.ub_version = intval;
3897				vdev_config_dirty(spa->spa_root_vdev);
3898			}
3899			break;
3900
3901		case ZPOOL_PROP_ALTROOT:
3902			/*
3903			 * 'altroot' is a non-persistent property. It should
3904			 * have been set temporarily at creation or import time.
3905			 */
3906			ASSERT(spa->spa_root != NULL);
3907			break;
3908
3909		case ZPOOL_PROP_CACHEFILE:
3910			/*
3911			 * 'cachefile' is also a non-persisitent property.
3912			 */
3913			break;
3914		default:
3915			/*
3916			 * Set pool property values in the poolprops mos object.
3917			 */
3918			if (spa->spa_pool_props_object == 0) {
3919				objset_t *mos = spa->spa_meta_objset;
3920
3921				VERIFY((spa->spa_pool_props_object =
3922				    zap_create(mos, DMU_OT_POOL_PROPS,
3923				    DMU_OT_NONE, 0, tx)) > 0);
3924
3925				VERIFY(zap_update(mos,
3926				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
3927				    8, 1, &spa->spa_pool_props_object, tx)
3928				    == 0);
3929			}
3930
3931			/* normalize the property name */
3932			propname = zpool_prop_to_name(prop);
3933			proptype = zpool_prop_get_type(prop);
3934
3935			if (nvpair_type(elem) == DATA_TYPE_STRING) {
3936				ASSERT(proptype == PROP_TYPE_STRING);
3937				VERIFY(nvpair_value_string(elem, &strval) == 0);
3938				VERIFY(zap_update(mos,
3939				    spa->spa_pool_props_object, propname,
3940				    1, strlen(strval) + 1, strval, tx) == 0);
3941
3942			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
3943				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
3944
3945				if (proptype == PROP_TYPE_INDEX) {
3946					const char *unused;
3947					VERIFY(zpool_prop_index_to_string(
3948					    prop, intval, &unused) == 0);
3949				}
3950				VERIFY(zap_update(mos,
3951				    spa->spa_pool_props_object, propname,
3952				    8, 1, &intval, tx) == 0);
3953			} else {
3954				ASSERT(0); /* not allowed */
3955			}
3956
3957			switch (prop) {
3958			case ZPOOL_PROP_DELEGATION:
3959				spa->spa_delegation = intval;
3960				break;
3961			case ZPOOL_PROP_BOOTFS:
3962				spa->spa_bootfs = intval;
3963				break;
3964			case ZPOOL_PROP_FAILUREMODE:
3965				spa->spa_failmode = intval;
3966				break;
3967			default:
3968				break;
3969			}
3970		}
3971
3972		/* log internal history if this is not a zpool create */
3973		if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
3974		    tx->tx_txg != TXG_INITIAL) {
3975			spa_history_internal_log(LOG_POOL_PROPSET,
3976			    spa, tx, cr, "%s %lld %s",
3977			    nvpair_name(elem), intval, spa_name(spa));
3978		}
3979	}
3980
3981	mutex_exit(&spa->spa_props_lock);
3982}
3983
3984/*
3985 * Sync the specified transaction group.  New blocks may be dirtied as
3986 * part of the process, so we iterate until it converges.
3987 */
3988void
3989spa_sync(spa_t *spa, uint64_t txg)
3990{
3991	dsl_pool_t *dp = spa->spa_dsl_pool;
3992	objset_t *mos = spa->spa_meta_objset;
3993	bplist_t *bpl = &spa->spa_sync_bplist;
3994	vdev_t *rvd = spa->spa_root_vdev;
3995	vdev_t *vd;
3996	dmu_tx_t *tx;
3997	int dirty_vdevs;
3998	int error;
3999
4000	/*
4001	 * Lock out configuration changes.
4002	 */
4003	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4004
4005	spa->spa_syncing_txg = txg;
4006	spa->spa_sync_pass = 0;
4007
4008	/*
4009	 * If there are any pending vdev state changes, convert them
4010	 * into config changes that go out with this transaction group.
4011	 */
4012	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4013	while (list_head(&spa->spa_state_dirty_list) != NULL) {
4014		/*
4015		 * We need the write lock here because, for aux vdevs,
4016		 * calling vdev_config_dirty() modifies sav_config.
4017		 * This is ugly and will become unnecessary when we
4018		 * eliminate the aux vdev wart by integrating all vdevs
4019		 * into the root vdev tree.
4020		 */
4021		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
4022		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
4023		while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
4024			vdev_state_clean(vd);
4025			vdev_config_dirty(vd);
4026		}
4027		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
4028		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
4029	}
4030	spa_config_exit(spa, SCL_STATE, FTAG);
4031
4032	VERIFY(0 == bplist_open(bpl, mos, spa->spa_sync_bplist_obj));
4033
4034	tx = dmu_tx_create_assigned(dp, txg);
4035
4036	/*
4037	 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
4038	 * set spa_deflate if we have no raid-z vdevs.
4039	 */
4040	if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
4041	    spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
4042		int i;
4043
4044		for (i = 0; i < rvd->vdev_children; i++) {
4045			vd = rvd->vdev_child[i];
4046			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
4047				break;
4048		}
4049		if (i == rvd->vdev_children) {
4050			spa->spa_deflate = TRUE;
4051			VERIFY(0 == zap_add(spa->spa_meta_objset,
4052			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
4053			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
4054		}
4055	}
4056
4057	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
4058	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
4059		dsl_pool_create_origin(dp, tx);
4060
4061		/* Keeping the origin open increases spa_minref */
4062		spa->spa_minref += 3;
4063	}
4064
4065	if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
4066	    spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
4067		dsl_pool_upgrade_clones(dp, tx);
4068	}
4069
4070	/*
4071	 * If anything has changed in this txg, push the deferred frees
4072	 * from the previous txg.  If not, leave them alone so that we
4073	 * don't generate work on an otherwise idle system.
4074	 */
4075	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
4076	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
4077	    !txg_list_empty(&dp->dp_sync_tasks, txg))
4078		spa_sync_deferred_frees(spa, txg);
4079
4080	/*
4081	 * Iterate to convergence.
4082	 */
4083	do {
4084		spa->spa_sync_pass++;
4085
4086		spa_sync_config_object(spa, tx);
4087		spa_sync_aux_dev(spa, &spa->spa_spares, tx,
4088		    ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
4089		spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
4090		    ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
4091		spa_errlog_sync(spa, txg);
4092		dsl_pool_sync(dp, txg);
4093
4094		dirty_vdevs = 0;
4095		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) {
4096			vdev_sync(vd, txg);
4097			dirty_vdevs++;
4098		}
4099
4100		bplist_sync(bpl, tx);
4101	} while (dirty_vdevs);
4102
4103	bplist_close(bpl);
4104
4105	dprintf("txg %llu passes %d\n", txg, spa->spa_sync_pass);
4106
4107	/*
4108	 * Rewrite the vdev configuration (which includes the uberblock)
4109	 * to commit the transaction group.
4110	 *
4111	 * If there are no dirty vdevs, we sync the uberblock to a few
4112	 * random top-level vdevs that are known to be visible in the
4113	 * config cache (see spa_vdev_add() for a complete description).
4114	 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
4115	 */
4116	for (;;) {
4117		/*
4118		 * We hold SCL_STATE to prevent vdev open/close/etc.
4119		 * while we're attempting to write the vdev labels.
4120		 */
4121		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4122
4123		if (list_is_empty(&spa->spa_config_dirty_list)) {
4124			vdev_t *svd[SPA_DVAS_PER_BP];
4125			int svdcount = 0;
4126			int children = rvd->vdev_children;
4127			int c0 = spa_get_random(children);
4128			int c;
4129
4130			for (c = 0; c < children; c++) {
4131				vd = rvd->vdev_child[(c0 + c) % children];
4132				if (vd->vdev_ms_array == 0 || vd->vdev_islog)
4133					continue;
4134				svd[svdcount++] = vd;
4135				if (svdcount == SPA_DVAS_PER_BP)
4136					break;
4137			}
4138			error = vdev_config_sync(svd, svdcount, txg);
4139		} else {
4140			error = vdev_config_sync(rvd->vdev_child,
4141			    rvd->vdev_children, txg);
4142		}
4143
4144		spa_config_exit(spa, SCL_STATE, FTAG);
4145
4146		if (error == 0)
4147			break;
4148		zio_suspend(spa, NULL);
4149		zio_resume_wait(spa);
4150	}
4151	dmu_tx_commit(tx);
4152
4153	/*
4154	 * Clear the dirty config list.
4155	 */
4156	while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
4157		vdev_config_clean(vd);
4158
4159	/*
4160	 * Now that the new config has synced transactionally,
4161	 * let it become visible to the config cache.
4162	 */
4163	if (spa->spa_config_syncing != NULL) {
4164		spa_config_set(spa, spa->spa_config_syncing);
4165		spa->spa_config_txg = txg;
4166		spa->spa_config_syncing = NULL;
4167	}
4168
4169	spa->spa_ubsync = spa->spa_uberblock;
4170
4171	/*
4172	 * Clean up the ZIL records for the synced txg.
4173	 */
4174	dsl_pool_zil_clean(dp);
4175
4176	/*
4177	 * Update usable space statistics.
4178	 */
4179	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
4180		vdev_sync_done(vd, txg);
4181
4182	/*
4183	 * It had better be the case that we didn't dirty anything
4184	 * since vdev_config_sync().
4185	 */
4186	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
4187	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
4188	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
4189	ASSERT(bpl->bpl_queue == NULL);
4190
4191	spa_config_exit(spa, SCL_CONFIG, FTAG);
4192
4193	/*
4194	 * If any async tasks have been requested, kick them off.
4195	 */
4196	spa_async_dispatch(spa);
4197}
4198
4199/*
4200 * Sync all pools.  We don't want to hold the namespace lock across these
4201 * operations, so we take a reference on the spa_t and drop the lock during the
4202 * sync.
4203 */
4204void
4205spa_sync_allpools(void)
4206{
4207	spa_t *spa = NULL;
4208	mutex_enter(&spa_namespace_lock);
4209	while ((spa = spa_next(spa)) != NULL) {
4210		if (spa_state(spa) != POOL_STATE_ACTIVE || spa_suspended(spa))
4211			continue;
4212		spa_open_ref(spa, FTAG);
4213		mutex_exit(&spa_namespace_lock);
4214		txg_wait_synced(spa_get_dsl(spa), 0);
4215		mutex_enter(&spa_namespace_lock);
4216		spa_close(spa, FTAG);
4217	}
4218	mutex_exit(&spa_namespace_lock);
4219}
4220
4221/*
4222 * ==========================================================================
4223 * Miscellaneous routines
4224 * ==========================================================================
4225 */
4226
4227/*
4228 * Remove all pools in the system.
4229 */
4230void
4231spa_evict_all(void)
4232{
4233	spa_t *spa;
4234
4235	/*
4236	 * Remove all cached state.  All pools should be closed now,
4237	 * so every spa in the AVL tree should be unreferenced.
4238	 */
4239	mutex_enter(&spa_namespace_lock);
4240	while ((spa = spa_next(NULL)) != NULL) {
4241		/*
4242		 * Stop async tasks.  The async thread may need to detach
4243		 * a device that's been replaced, which requires grabbing
4244		 * spa_namespace_lock, so we must drop it here.
4245		 */
4246		spa_open_ref(spa, FTAG);
4247		mutex_exit(&spa_namespace_lock);
4248		spa_async_suspend(spa);
4249		mutex_enter(&spa_namespace_lock);
4250		spa_close(spa, FTAG);
4251
4252		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4253			spa_unload(spa);
4254			spa_deactivate(spa);
4255		}
4256		spa_remove(spa);
4257	}
4258	mutex_exit(&spa_namespace_lock);
4259}
4260
4261vdev_t *
4262spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
4263{
4264	vdev_t *vd;
4265	int i;
4266
4267	if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
4268		return (vd);
4269
4270	if (aux) {
4271		for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
4272			vd = spa->spa_l2cache.sav_vdevs[i];
4273			if (vd->vdev_guid == guid)
4274				return (vd);
4275		}
4276
4277		for (i = 0; i < spa->spa_spares.sav_count; i++) {
4278			vd = spa->spa_spares.sav_vdevs[i];
4279			if (vd->vdev_guid == guid)
4280				return (vd);
4281		}
4282	}
4283
4284	return (NULL);
4285}
4286
4287void
4288spa_upgrade(spa_t *spa, uint64_t version)
4289{
4290	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4291
4292	/*
4293	 * This should only be called for a non-faulted pool, and since a
4294	 * future version would result in an unopenable pool, this shouldn't be
4295	 * possible.
4296	 */
4297	ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
4298	ASSERT(version >= spa->spa_uberblock.ub_version);
4299
4300	spa->spa_uberblock.ub_version = version;
4301	vdev_config_dirty(spa->spa_root_vdev);
4302
4303	spa_config_exit(spa, SCL_ALL, FTAG);
4304
4305	txg_wait_synced(spa_get_dsl(spa), 0);
4306}
4307
4308boolean_t
4309spa_has_spare(spa_t *spa, uint64_t guid)
4310{
4311	int i;
4312	uint64_t spareguid;
4313	spa_aux_vdev_t *sav = &spa->spa_spares;
4314
4315	for (i = 0; i < sav->sav_count; i++)
4316		if (sav->sav_vdevs[i]->vdev_guid == guid)
4317			return (B_TRUE);
4318
4319	for (i = 0; i < sav->sav_npending; i++) {
4320		if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
4321		    &spareguid) == 0 && spareguid == guid)
4322			return (B_TRUE);
4323	}
4324
4325	return (B_FALSE);
4326}
4327
4328/*
4329 * Check if a pool has an active shared spare device.
4330 * Note: reference count of an active spare is 2, as a spare and as a replace
4331 */
4332static boolean_t
4333spa_has_active_shared_spare(spa_t *spa)
4334{
4335	int i, refcnt;
4336	uint64_t pool;
4337	spa_aux_vdev_t *sav = &spa->spa_spares;
4338
4339	for (i = 0; i < sav->sav_count; i++) {
4340		if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
4341		    &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
4342		    refcnt > 2)
4343			return (B_TRUE);
4344	}
4345
4346	return (B_FALSE);
4347}
4348
4349/*
4350 * Post a sysevent corresponding to the given event.  The 'name' must be one of
4351 * the event definitions in sys/sysevent/eventdefs.h.  The payload will be
4352 * filled in from the spa and (optionally) the vdev.  This doesn't do anything
4353 * in the userland libzpool, as we don't want consumers to misinterpret ztest
4354 * or zdb as real changes.
4355 */
4356void
4357spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
4358{
4359#ifdef _KERNEL
4360	sysevent_t		*ev;
4361	sysevent_attr_list_t	*attr = NULL;
4362	sysevent_value_t	value;
4363	sysevent_id_t		eid;
4364
4365	ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
4366	    SE_SLEEP);
4367
4368	value.value_type = SE_DATA_TYPE_STRING;
4369	value.value.sv_string = spa_name(spa);
4370	if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
4371		goto done;
4372
4373	value.value_type = SE_DATA_TYPE_UINT64;
4374	value.value.sv_uint64 = spa_guid(spa);
4375	if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
4376		goto done;
4377
4378	if (vd) {
4379		value.value_type = SE_DATA_TYPE_UINT64;
4380		value.value.sv_uint64 = vd->vdev_guid;
4381		if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
4382		    SE_SLEEP) != 0)
4383			goto done;
4384
4385		if (vd->vdev_path) {
4386			value.value_type = SE_DATA_TYPE_STRING;
4387			value.value.sv_string = vd->vdev_path;
4388			if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
4389			    &value, SE_SLEEP) != 0)
4390				goto done;
4391		}
4392	}
4393
4394	if (sysevent_attach_attributes(ev, attr) != 0)
4395		goto done;
4396	attr = NULL;
4397
4398	(void) log_sysevent(ev, SE_SLEEP, &eid);
4399
4400done:
4401	if (attr)
4402		sysevent_free_attr(attr);
4403	sysevent_free(ev);
4404#endif
4405}
4406