spa.c revision bf82a41b568b2bd31bf9814587eb25ee2e7b05ff
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 2008 Sun Microsystems, Inc.  All rights reserved.
24 * Use is subject to license terms.
25 */
26
27#pragma ident	"%Z%%M%	%I%	%E% SMI"
28
29/*
30 * This file contains all the routines used when modifying on-disk SPA state.
31 * This includes opening, importing, destroying, exporting a pool, and syncing a
32 * pool.
33 */
34
35#include <sys/zfs_context.h>
36#include <sys/fm/fs/zfs.h>
37#include <sys/spa_impl.h>
38#include <sys/zio.h>
39#include <sys/zio_checksum.h>
40#include <sys/zio_compress.h>
41#include <sys/dmu.h>
42#include <sys/dmu_tx.h>
43#include <sys/zap.h>
44#include <sys/zil.h>
45#include <sys/vdev_impl.h>
46#include <sys/metaslab.h>
47#include <sys/uberblock_impl.h>
48#include <sys/txg.h>
49#include <sys/avl.h>
50#include <sys/dmu_traverse.h>
51#include <sys/dmu_objset.h>
52#include <sys/unique.h>
53#include <sys/dsl_pool.h>
54#include <sys/dsl_dataset.h>
55#include <sys/dsl_dir.h>
56#include <sys/dsl_prop.h>
57#include <sys/dsl_synctask.h>
58#include <sys/fs/zfs.h>
59#include <sys/arc.h>
60#include <sys/callb.h>
61#include <sys/systeminfo.h>
62#include <sys/sunddi.h>
63#include <sys/spa_boot.h>
64
65#include "zfs_prop.h"
66#include "zfs_comutil.h"
67
68int zio_taskq_threads = 8;
69
70static void spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx);
71
72/*
73 * ==========================================================================
74 * SPA properties routines
75 * ==========================================================================
76 */
77
78/*
79 * Add a (source=src, propname=propval) list to an nvlist.
80 */
81static void
82spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
83    uint64_t intval, zprop_source_t src)
84{
85	const char *propname = zpool_prop_to_name(prop);
86	nvlist_t *propval;
87
88	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
89	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
90
91	if (strval != NULL)
92		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
93	else
94		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
95
96	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
97	nvlist_free(propval);
98}
99
100/*
101 * Get property values from the spa configuration.
102 */
103static void
104spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
105{
106	uint64_t size = spa_get_space(spa);
107	uint64_t used = spa_get_alloc(spa);
108	uint64_t cap, version;
109	zprop_source_t src = ZPROP_SRC_NONE;
110	spa_config_dirent_t *dp;
111
112	/*
113	 * readonly properties
114	 */
115	spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa->spa_name, 0, src);
116	spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
117	spa_prop_add_list(*nvp, ZPOOL_PROP_USED, NULL, used, src);
118	spa_prop_add_list(*nvp, ZPOOL_PROP_AVAILABLE, NULL, size - used, src);
119
120	cap = (size == 0) ? 0 : (used * 100 / size);
121	spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
122
123	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
124	spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
125	    spa->spa_root_vdev->vdev_state, src);
126
127	/*
128	 * settable properties that are not stored in the pool property object.
129	 */
130	version = spa_version(spa);
131	if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
132		src = ZPROP_SRC_DEFAULT;
133	else
134		src = ZPROP_SRC_LOCAL;
135	spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
136
137	if (spa->spa_root != NULL)
138		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
139		    0, ZPROP_SRC_LOCAL);
140
141	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
142		if (dp->scd_path == NULL) {
143			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
144			    "none", 0, ZPROP_SRC_LOCAL);
145		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
146			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
147			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
148		}
149	}
150}
151
152/*
153 * Get zpool property values.
154 */
155int
156spa_prop_get(spa_t *spa, nvlist_t **nvp)
157{
158	zap_cursor_t zc;
159	zap_attribute_t za;
160	objset_t *mos = spa->spa_meta_objset;
161	int err;
162
163	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
164
165	/*
166	 * Get properties from the spa config.
167	 */
168	spa_prop_get_config(spa, nvp);
169
170	mutex_enter(&spa->spa_props_lock);
171	/* If no pool property object, no more prop to get. */
172	if (spa->spa_pool_props_object == 0) {
173		mutex_exit(&spa->spa_props_lock);
174		return (0);
175	}
176
177	/*
178	 * Get properties from the MOS pool property object.
179	 */
180	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
181	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
182	    zap_cursor_advance(&zc)) {
183		uint64_t intval = 0;
184		char *strval = NULL;
185		zprop_source_t src = ZPROP_SRC_DEFAULT;
186		zpool_prop_t prop;
187
188		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
189			continue;
190
191		switch (za.za_integer_length) {
192		case 8:
193			/* integer property */
194			if (za.za_first_integer !=
195			    zpool_prop_default_numeric(prop))
196				src = ZPROP_SRC_LOCAL;
197
198			if (prop == ZPOOL_PROP_BOOTFS) {
199				dsl_pool_t *dp;
200				dsl_dataset_t *ds = NULL;
201
202				dp = spa_get_dsl(spa);
203				rw_enter(&dp->dp_config_rwlock, RW_READER);
204				if (err = dsl_dataset_open_obj(dp,
205				    za.za_first_integer, NULL, DS_MODE_NONE,
206				    FTAG, &ds)) {
207					rw_exit(&dp->dp_config_rwlock);
208					break;
209				}
210
211				strval = kmem_alloc(
212				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
213				    KM_SLEEP);
214				dsl_dataset_name(ds, strval);
215				dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
216				rw_exit(&dp->dp_config_rwlock);
217			} else {
218				strval = NULL;
219				intval = za.za_first_integer;
220			}
221
222			spa_prop_add_list(*nvp, prop, strval, intval, src);
223
224			if (strval != NULL)
225				kmem_free(strval,
226				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
227
228			break;
229
230		case 1:
231			/* string property */
232			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
233			err = zap_lookup(mos, spa->spa_pool_props_object,
234			    za.za_name, 1, za.za_num_integers, strval);
235			if (err) {
236				kmem_free(strval, za.za_num_integers);
237				break;
238			}
239			spa_prop_add_list(*nvp, prop, strval, 0, src);
240			kmem_free(strval, za.za_num_integers);
241			break;
242
243		default:
244			break;
245		}
246	}
247	zap_cursor_fini(&zc);
248	mutex_exit(&spa->spa_props_lock);
249out:
250	if (err && err != ENOENT) {
251		nvlist_free(*nvp);
252		*nvp = NULL;
253		return (err);
254	}
255
256	return (0);
257}
258
259/*
260 * Validate the given pool properties nvlist and modify the list
261 * for the property values to be set.
262 */
263static int
264spa_prop_validate(spa_t *spa, nvlist_t *props)
265{
266	nvpair_t *elem;
267	int error = 0, reset_bootfs = 0;
268	uint64_t objnum;
269
270	elem = NULL;
271	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
272		zpool_prop_t prop;
273		char *propname, *strval;
274		uint64_t intval;
275		vdev_t *rvdev;
276		char *vdev_type;
277		objset_t *os;
278		char *slash;
279
280		propname = nvpair_name(elem);
281
282		if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
283			return (EINVAL);
284
285		switch (prop) {
286		case ZPOOL_PROP_VERSION:
287			error = nvpair_value_uint64(elem, &intval);
288			if (!error &&
289			    (intval < spa_version(spa) || intval > SPA_VERSION))
290				error = EINVAL;
291			break;
292
293		case ZPOOL_PROP_DELEGATION:
294		case ZPOOL_PROP_AUTOREPLACE:
295			error = nvpair_value_uint64(elem, &intval);
296			if (!error && intval > 1)
297				error = EINVAL;
298			break;
299
300		case ZPOOL_PROP_BOOTFS:
301			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
302				error = ENOTSUP;
303				break;
304			}
305
306			/*
307			 * A bootable filesystem can not be on a RAIDZ pool
308			 * nor a striped pool with more than 1 device.
309			 */
310			rvdev = spa->spa_root_vdev;
311			vdev_type =
312			    rvdev->vdev_child[0]->vdev_ops->vdev_op_type;
313			if (rvdev->vdev_children > 1 ||
314			    strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
315			    strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
316				error = ENOTSUP;
317				break;
318			}
319
320			reset_bootfs = 1;
321
322			error = nvpair_value_string(elem, &strval);
323
324			if (!error) {
325				if (strval == NULL || strval[0] == '\0') {
326					objnum = zpool_prop_default_numeric(
327					    ZPOOL_PROP_BOOTFS);
328					break;
329				}
330
331				if (error = dmu_objset_open(strval, DMU_OST_ZFS,
332				    DS_MODE_STANDARD | DS_MODE_READONLY, &os))
333					break;
334				objnum = dmu_objset_id(os);
335				dmu_objset_close(os);
336			}
337			break;
338		case ZPOOL_PROP_FAILUREMODE:
339			error = nvpair_value_uint64(elem, &intval);
340			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
341			    intval > ZIO_FAILURE_MODE_PANIC))
342				error = EINVAL;
343
344			/*
345			 * This is a special case which only occurs when
346			 * the pool has completely failed. This allows
347			 * the user to change the in-core failmode property
348			 * without syncing it out to disk (I/Os might
349			 * currently be blocked). We do this by returning
350			 * EIO to the caller (spa_prop_set) to trick it
351			 * into thinking we encountered a property validation
352			 * error.
353			 */
354			if (!error && spa_state(spa) == POOL_STATE_IO_FAILURE) {
355				spa->spa_failmode = intval;
356				error = EIO;
357			}
358			break;
359
360		case ZPOOL_PROP_CACHEFILE:
361			if ((error = nvpair_value_string(elem, &strval)) != 0)
362				break;
363
364			if (strval[0] == '\0')
365				break;
366
367			if (strcmp(strval, "none") == 0)
368				break;
369
370			if (strval[0] != '/') {
371				error = EINVAL;
372				break;
373			}
374
375			slash = strrchr(strval, '/');
376			ASSERT(slash != NULL);
377
378			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
379			    strcmp(slash, "/..") == 0)
380				error = EINVAL;
381			break;
382		}
383
384		if (error)
385			break;
386	}
387
388	if (!error && reset_bootfs) {
389		error = nvlist_remove(props,
390		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
391
392		if (!error) {
393			error = nvlist_add_uint64(props,
394			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
395		}
396	}
397
398	return (error);
399}
400
401int
402spa_prop_set(spa_t *spa, nvlist_t *nvp)
403{
404	int error;
405
406	if ((error = spa_prop_validate(spa, nvp)) != 0)
407		return (error);
408
409	return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
410	    spa, nvp, 3));
411}
412
413/*
414 * If the bootfs property value is dsobj, clear it.
415 */
416void
417spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
418{
419	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
420		VERIFY(zap_remove(spa->spa_meta_objset,
421		    spa->spa_pool_props_object,
422		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
423		spa->spa_bootfs = 0;
424	}
425}
426
427/*
428 * ==========================================================================
429 * SPA state manipulation (open/create/destroy/import/export)
430 * ==========================================================================
431 */
432
433static int
434spa_error_entry_compare(const void *a, const void *b)
435{
436	spa_error_entry_t *sa = (spa_error_entry_t *)a;
437	spa_error_entry_t *sb = (spa_error_entry_t *)b;
438	int ret;
439
440	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
441	    sizeof (zbookmark_t));
442
443	if (ret < 0)
444		return (-1);
445	else if (ret > 0)
446		return (1);
447	else
448		return (0);
449}
450
451/*
452 * Utility function which retrieves copies of the current logs and
453 * re-initializes them in the process.
454 */
455void
456spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
457{
458	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
459
460	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
461	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
462
463	avl_create(&spa->spa_errlist_scrub,
464	    spa_error_entry_compare, sizeof (spa_error_entry_t),
465	    offsetof(spa_error_entry_t, se_avl));
466	avl_create(&spa->spa_errlist_last,
467	    spa_error_entry_compare, sizeof (spa_error_entry_t),
468	    offsetof(spa_error_entry_t, se_avl));
469}
470
471/*
472 * Activate an uninitialized pool.
473 */
474static void
475spa_activate(spa_t *spa)
476{
477	int t;
478
479	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
480
481	spa->spa_state = POOL_STATE_ACTIVE;
482
483	spa->spa_normal_class = metaslab_class_create();
484	spa->spa_log_class = metaslab_class_create();
485
486	for (t = 0; t < ZIO_TYPES; t++) {
487		spa->spa_zio_issue_taskq[t] = taskq_create("spa_zio_issue",
488		    zio_taskq_threads, maxclsyspri, 50, INT_MAX,
489		    TASKQ_PREPOPULATE);
490		spa->spa_zio_intr_taskq[t] = taskq_create("spa_zio_intr",
491		    zio_taskq_threads, maxclsyspri, 50, INT_MAX,
492		    TASKQ_PREPOPULATE);
493	}
494
495	list_create(&spa->spa_dirty_list, sizeof (vdev_t),
496	    offsetof(vdev_t, vdev_dirty_node));
497	list_create(&spa->spa_zio_list, sizeof (zio_t),
498	    offsetof(zio_t, zio_link_node));
499
500	txg_list_create(&spa->spa_vdev_txg_list,
501	    offsetof(struct vdev, vdev_txg_node));
502
503	avl_create(&spa->spa_errlist_scrub,
504	    spa_error_entry_compare, sizeof (spa_error_entry_t),
505	    offsetof(spa_error_entry_t, se_avl));
506	avl_create(&spa->spa_errlist_last,
507	    spa_error_entry_compare, sizeof (spa_error_entry_t),
508	    offsetof(spa_error_entry_t, se_avl));
509}
510
511/*
512 * Opposite of spa_activate().
513 */
514static void
515spa_deactivate(spa_t *spa)
516{
517	int t;
518
519	ASSERT(spa->spa_sync_on == B_FALSE);
520	ASSERT(spa->spa_dsl_pool == NULL);
521	ASSERT(spa->spa_root_vdev == NULL);
522
523	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
524
525	txg_list_destroy(&spa->spa_vdev_txg_list);
526
527	list_destroy(&spa->spa_dirty_list);
528	list_destroy(&spa->spa_zio_list);
529
530	for (t = 0; t < ZIO_TYPES; t++) {
531		taskq_destroy(spa->spa_zio_issue_taskq[t]);
532		taskq_destroy(spa->spa_zio_intr_taskq[t]);
533		spa->spa_zio_issue_taskq[t] = NULL;
534		spa->spa_zio_intr_taskq[t] = NULL;
535	}
536
537	metaslab_class_destroy(spa->spa_normal_class);
538	spa->spa_normal_class = NULL;
539
540	metaslab_class_destroy(spa->spa_log_class);
541	spa->spa_log_class = NULL;
542
543	/*
544	 * If this was part of an import or the open otherwise failed, we may
545	 * still have errors left in the queues.  Empty them just in case.
546	 */
547	spa_errlog_drain(spa);
548
549	avl_destroy(&spa->spa_errlist_scrub);
550	avl_destroy(&spa->spa_errlist_last);
551
552	spa->spa_state = POOL_STATE_UNINITIALIZED;
553}
554
555/*
556 * Verify a pool configuration, and construct the vdev tree appropriately.  This
557 * will create all the necessary vdevs in the appropriate layout, with each vdev
558 * in the CLOSED state.  This will prep the pool before open/creation/import.
559 * All vdev validation is done by the vdev_alloc() routine.
560 */
561static int
562spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
563    uint_t id, int atype)
564{
565	nvlist_t **child;
566	uint_t c, children;
567	int error;
568
569	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
570		return (error);
571
572	if ((*vdp)->vdev_ops->vdev_op_leaf)
573		return (0);
574
575	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
576	    &child, &children) != 0) {
577		vdev_free(*vdp);
578		*vdp = NULL;
579		return (EINVAL);
580	}
581
582	for (c = 0; c < children; c++) {
583		vdev_t *vd;
584		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
585		    atype)) != 0) {
586			vdev_free(*vdp);
587			*vdp = NULL;
588			return (error);
589		}
590	}
591
592	ASSERT(*vdp != NULL);
593
594	return (0);
595}
596
597/*
598 * Opposite of spa_load().
599 */
600static void
601spa_unload(spa_t *spa)
602{
603	int i;
604
605	/*
606	 * Stop async tasks.
607	 */
608	spa_async_suspend(spa);
609
610	/*
611	 * Stop syncing.
612	 */
613	if (spa->spa_sync_on) {
614		txg_sync_stop(spa->spa_dsl_pool);
615		spa->spa_sync_on = B_FALSE;
616	}
617
618	/*
619	 * Wait for any outstanding prefetch I/O to complete.
620	 */
621	spa_config_enter(spa, RW_WRITER, FTAG);
622	spa_config_exit(spa, FTAG);
623
624	/*
625	 * Drop and purge level 2 cache
626	 */
627	spa_l2cache_drop(spa);
628
629	/*
630	 * Close the dsl pool.
631	 */
632	if (spa->spa_dsl_pool) {
633		dsl_pool_close(spa->spa_dsl_pool);
634		spa->spa_dsl_pool = NULL;
635	}
636
637	/*
638	 * Close all vdevs.
639	 */
640	if (spa->spa_root_vdev)
641		vdev_free(spa->spa_root_vdev);
642	ASSERT(spa->spa_root_vdev == NULL);
643
644	for (i = 0; i < spa->spa_spares.sav_count; i++)
645		vdev_free(spa->spa_spares.sav_vdevs[i]);
646	if (spa->spa_spares.sav_vdevs) {
647		kmem_free(spa->spa_spares.sav_vdevs,
648		    spa->spa_spares.sav_count * sizeof (void *));
649		spa->spa_spares.sav_vdevs = NULL;
650	}
651	if (spa->spa_spares.sav_config) {
652		nvlist_free(spa->spa_spares.sav_config);
653		spa->spa_spares.sav_config = NULL;
654	}
655
656	for (i = 0; i < spa->spa_l2cache.sav_count; i++)
657		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
658	if (spa->spa_l2cache.sav_vdevs) {
659		kmem_free(spa->spa_l2cache.sav_vdevs,
660		    spa->spa_l2cache.sav_count * sizeof (void *));
661		spa->spa_l2cache.sav_vdevs = NULL;
662	}
663	if (spa->spa_l2cache.sav_config) {
664		nvlist_free(spa->spa_l2cache.sav_config);
665		spa->spa_l2cache.sav_config = NULL;
666	}
667
668	spa->spa_async_suspended = 0;
669}
670
671/*
672 * Load (or re-load) the current list of vdevs describing the active spares for
673 * this pool.  When this is called, we have some form of basic information in
674 * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
675 * then re-generate a more complete list including status information.
676 */
677static void
678spa_load_spares(spa_t *spa)
679{
680	nvlist_t **spares;
681	uint_t nspares;
682	int i;
683	vdev_t *vd, *tvd;
684
685	/*
686	 * First, close and free any existing spare vdevs.
687	 */
688	for (i = 0; i < spa->spa_spares.sav_count; i++) {
689		vd = spa->spa_spares.sav_vdevs[i];
690
691		/* Undo the call to spa_activate() below */
692		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
693		    B_FALSE)) != NULL && tvd->vdev_isspare)
694			spa_spare_remove(tvd);
695		vdev_close(vd);
696		vdev_free(vd);
697	}
698
699	if (spa->spa_spares.sav_vdevs)
700		kmem_free(spa->spa_spares.sav_vdevs,
701		    spa->spa_spares.sav_count * sizeof (void *));
702
703	if (spa->spa_spares.sav_config == NULL)
704		nspares = 0;
705	else
706		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
707		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
708
709	spa->spa_spares.sav_count = (int)nspares;
710	spa->spa_spares.sav_vdevs = NULL;
711
712	if (nspares == 0)
713		return;
714
715	/*
716	 * Construct the array of vdevs, opening them to get status in the
717	 * process.   For each spare, there is potentially two different vdev_t
718	 * structures associated with it: one in the list of spares (used only
719	 * for basic validation purposes) and one in the active vdev
720	 * configuration (if it's spared in).  During this phase we open and
721	 * validate each vdev on the spare list.  If the vdev also exists in the
722	 * active configuration, then we also mark this vdev as an active spare.
723	 */
724	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
725	    KM_SLEEP);
726	for (i = 0; i < spa->spa_spares.sav_count; i++) {
727		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
728		    VDEV_ALLOC_SPARE) == 0);
729		ASSERT(vd != NULL);
730
731		spa->spa_spares.sav_vdevs[i] = vd;
732
733		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
734		    B_FALSE)) != NULL) {
735			if (!tvd->vdev_isspare)
736				spa_spare_add(tvd);
737
738			/*
739			 * We only mark the spare active if we were successfully
740			 * able to load the vdev.  Otherwise, importing a pool
741			 * with a bad active spare would result in strange
742			 * behavior, because multiple pool would think the spare
743			 * is actively in use.
744			 *
745			 * There is a vulnerability here to an equally bizarre
746			 * circumstance, where a dead active spare is later
747			 * brought back to life (onlined or otherwise).  Given
748			 * the rarity of this scenario, and the extra complexity
749			 * it adds, we ignore the possibility.
750			 */
751			if (!vdev_is_dead(tvd))
752				spa_spare_activate(tvd);
753		}
754
755		if (vdev_open(vd) != 0)
756			continue;
757
758		vd->vdev_top = vd;
759		if (vdev_validate_aux(vd) == 0)
760			spa_spare_add(vd);
761	}
762
763	/*
764	 * Recompute the stashed list of spares, with status information
765	 * this time.
766	 */
767	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
768	    DATA_TYPE_NVLIST_ARRAY) == 0);
769
770	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
771	    KM_SLEEP);
772	for (i = 0; i < spa->spa_spares.sav_count; i++)
773		spares[i] = vdev_config_generate(spa,
774		    spa->spa_spares.sav_vdevs[i], B_TRUE, B_TRUE, B_FALSE);
775	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
776	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
777	for (i = 0; i < spa->spa_spares.sav_count; i++)
778		nvlist_free(spares[i]);
779	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
780}
781
782/*
783 * Load (or re-load) the current list of vdevs describing the active l2cache for
784 * this pool.  When this is called, we have some form of basic information in
785 * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
786 * then re-generate a more complete list including status information.
787 * Devices which are already active have their details maintained, and are
788 * not re-opened.
789 */
790static void
791spa_load_l2cache(spa_t *spa)
792{
793	nvlist_t **l2cache;
794	uint_t nl2cache;
795	int i, j, oldnvdevs;
796	uint64_t guid, size;
797	vdev_t *vd, **oldvdevs, **newvdevs;
798	spa_aux_vdev_t *sav = &spa->spa_l2cache;
799
800	if (sav->sav_config != NULL) {
801		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
802		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
803		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
804	} else {
805		nl2cache = 0;
806	}
807
808	oldvdevs = sav->sav_vdevs;
809	oldnvdevs = sav->sav_count;
810	sav->sav_vdevs = NULL;
811	sav->sav_count = 0;
812
813	/*
814	 * Process new nvlist of vdevs.
815	 */
816	for (i = 0; i < nl2cache; i++) {
817		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
818		    &guid) == 0);
819
820		newvdevs[i] = NULL;
821		for (j = 0; j < oldnvdevs; j++) {
822			vd = oldvdevs[j];
823			if (vd != NULL && guid == vd->vdev_guid) {
824				/*
825				 * Retain previous vdev for add/remove ops.
826				 */
827				newvdevs[i] = vd;
828				oldvdevs[j] = NULL;
829				break;
830			}
831		}
832
833		if (newvdevs[i] == NULL) {
834			/*
835			 * Create new vdev
836			 */
837			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
838			    VDEV_ALLOC_L2CACHE) == 0);
839			ASSERT(vd != NULL);
840			newvdevs[i] = vd;
841
842			/*
843			 * Commit this vdev as an l2cache device,
844			 * even if it fails to open.
845			 */
846			spa_l2cache_add(vd);
847
848			vd->vdev_top = vd;
849			vd->vdev_aux = sav;
850
851			spa_l2cache_activate(vd);
852
853			if (vdev_open(vd) != 0)
854				continue;
855
856			(void) vdev_validate_aux(vd);
857
858			if (!vdev_is_dead(vd)) {
859				size = vdev_get_rsize(vd);
860				l2arc_add_vdev(spa, vd,
861				    VDEV_LABEL_START_SIZE,
862				    size - VDEV_LABEL_START_SIZE);
863			}
864		}
865	}
866
867	/*
868	 * Purge vdevs that were dropped
869	 */
870	for (i = 0; i < oldnvdevs; i++) {
871		uint64_t pool;
872
873		vd = oldvdevs[i];
874		if (vd != NULL) {
875			if (spa_mode & FWRITE &&
876			    spa_l2cache_exists(vd->vdev_guid, &pool) &&
877			    pool != 0ULL &&
878			    l2arc_vdev_present(vd)) {
879				l2arc_remove_vdev(vd);
880			}
881			(void) vdev_close(vd);
882			spa_l2cache_remove(vd);
883		}
884	}
885
886	if (oldvdevs)
887		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
888
889	if (sav->sav_config == NULL)
890		goto out;
891
892	sav->sav_vdevs = newvdevs;
893	sav->sav_count = (int)nl2cache;
894
895	/*
896	 * Recompute the stashed list of l2cache devices, with status
897	 * information this time.
898	 */
899	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
900	    DATA_TYPE_NVLIST_ARRAY) == 0);
901
902	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
903	for (i = 0; i < sav->sav_count; i++)
904		l2cache[i] = vdev_config_generate(spa,
905		    sav->sav_vdevs[i], B_TRUE, B_FALSE, B_TRUE);
906	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
907	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
908out:
909	for (i = 0; i < sav->sav_count; i++)
910		nvlist_free(l2cache[i]);
911	if (sav->sav_count)
912		kmem_free(l2cache, sav->sav_count * sizeof (void *));
913}
914
915static int
916load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
917{
918	dmu_buf_t *db;
919	char *packed = NULL;
920	size_t nvsize = 0;
921	int error;
922	*value = NULL;
923
924	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
925	nvsize = *(uint64_t *)db->db_data;
926	dmu_buf_rele(db, FTAG);
927
928	packed = kmem_alloc(nvsize, KM_SLEEP);
929	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed);
930	if (error == 0)
931		error = nvlist_unpack(packed, nvsize, value, 0);
932	kmem_free(packed, nvsize);
933
934	return (error);
935}
936
937/*
938 * Checks to see if the given vdev could not be opened, in which case we post a
939 * sysevent to notify the autoreplace code that the device has been removed.
940 */
941static void
942spa_check_removed(vdev_t *vd)
943{
944	int c;
945
946	for (c = 0; c < vd->vdev_children; c++)
947		spa_check_removed(vd->vdev_child[c]);
948
949	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
950		zfs_post_autoreplace(vd->vdev_spa, vd);
951		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
952	}
953}
954
955/*
956 * Load an existing storage pool, using the pool's builtin spa_config as a
957 * source of configuration information.
958 */
959static int
960spa_load(spa_t *spa, nvlist_t *config, spa_load_state_t state, int mosconfig)
961{
962	int error = 0;
963	nvlist_t *nvroot = NULL;
964	vdev_t *rvd;
965	uberblock_t *ub = &spa->spa_uberblock;
966	uint64_t config_cache_txg = spa->spa_config_txg;
967	uint64_t pool_guid;
968	uint64_t version;
969	zio_t *zio;
970	uint64_t autoreplace = 0;
971
972	spa->spa_load_state = state;
973
974	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) ||
975	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
976		error = EINVAL;
977		goto out;
978	}
979
980	/*
981	 * Versioning wasn't explicitly added to the label until later, so if
982	 * it's not present treat it as the initial version.
983	 */
984	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &version) != 0)
985		version = SPA_VERSION_INITIAL;
986
987	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
988	    &spa->spa_config_txg);
989
990	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
991	    spa_guid_exists(pool_guid, 0)) {
992		error = EEXIST;
993		goto out;
994	}
995
996	spa->spa_load_guid = pool_guid;
997
998	/*
999	 * Parse the configuration into a vdev tree.  We explicitly set the
1000	 * value that will be returned by spa_version() since parsing the
1001	 * configuration requires knowing the version number.
1002	 */
1003	spa_config_enter(spa, RW_WRITER, FTAG);
1004	spa->spa_ubsync.ub_version = version;
1005	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_LOAD);
1006	spa_config_exit(spa, FTAG);
1007
1008	if (error != 0)
1009		goto out;
1010
1011	ASSERT(spa->spa_root_vdev == rvd);
1012	ASSERT(spa_guid(spa) == pool_guid);
1013
1014	/*
1015	 * Try to open all vdevs, loading each label in the process.
1016	 */
1017	error = vdev_open(rvd);
1018	if (error != 0)
1019		goto out;
1020
1021	/*
1022	 * Validate the labels for all leaf vdevs.  We need to grab the config
1023	 * lock because all label I/O is done with the ZIO_FLAG_CONFIG_HELD
1024	 * flag.
1025	 */
1026	spa_config_enter(spa, RW_READER, FTAG);
1027	error = vdev_validate(rvd);
1028	spa_config_exit(spa, FTAG);
1029
1030	if (error != 0)
1031		goto out;
1032
1033	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
1034		error = ENXIO;
1035		goto out;
1036	}
1037
1038	/*
1039	 * Find the best uberblock.
1040	 */
1041	bzero(ub, sizeof (uberblock_t));
1042
1043	zio = zio_root(spa, NULL, NULL,
1044	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1045	vdev_uberblock_load(zio, rvd, ub);
1046	error = zio_wait(zio);
1047
1048	/*
1049	 * If we weren't able to find a single valid uberblock, return failure.
1050	 */
1051	if (ub->ub_txg == 0) {
1052		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1053		    VDEV_AUX_CORRUPT_DATA);
1054		error = ENXIO;
1055		goto out;
1056	}
1057
1058	/*
1059	 * If the pool is newer than the code, we can't open it.
1060	 */
1061	if (ub->ub_version > SPA_VERSION) {
1062		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1063		    VDEV_AUX_VERSION_NEWER);
1064		error = ENOTSUP;
1065		goto out;
1066	}
1067
1068	/*
1069	 * If the vdev guid sum doesn't match the uberblock, we have an
1070	 * incomplete configuration.
1071	 */
1072	if (rvd->vdev_guid_sum != ub->ub_guid_sum && mosconfig) {
1073		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1074		    VDEV_AUX_BAD_GUID_SUM);
1075		error = ENXIO;
1076		goto out;
1077	}
1078
1079	/*
1080	 * Initialize internal SPA structures.
1081	 */
1082	spa->spa_state = POOL_STATE_ACTIVE;
1083	spa->spa_ubsync = spa->spa_uberblock;
1084	spa->spa_first_txg = spa_last_synced_txg(spa) + 1;
1085	error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1086	if (error) {
1087		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1088		    VDEV_AUX_CORRUPT_DATA);
1089		goto out;
1090	}
1091	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1092
1093	if (zap_lookup(spa->spa_meta_objset,
1094	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
1095	    sizeof (uint64_t), 1, &spa->spa_config_object) != 0) {
1096		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1097		    VDEV_AUX_CORRUPT_DATA);
1098		error = EIO;
1099		goto out;
1100	}
1101
1102	if (!mosconfig) {
1103		nvlist_t *newconfig;
1104		uint64_t hostid;
1105
1106		if (load_nvlist(spa, spa->spa_config_object, &newconfig) != 0) {
1107			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1108			    VDEV_AUX_CORRUPT_DATA);
1109			error = EIO;
1110			goto out;
1111		}
1112
1113		if (nvlist_lookup_uint64(newconfig, ZPOOL_CONFIG_HOSTID,
1114		    &hostid) == 0) {
1115			char *hostname;
1116			unsigned long myhostid = 0;
1117
1118			VERIFY(nvlist_lookup_string(newconfig,
1119			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
1120
1121			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
1122			if (hostid != 0 && myhostid != 0 &&
1123			    (unsigned long)hostid != myhostid) {
1124				cmn_err(CE_WARN, "pool '%s' could not be "
1125				    "loaded as it was last accessed by "
1126				    "another system (host: %s hostid: 0x%lx).  "
1127				    "See: http://www.sun.com/msg/ZFS-8000-EY",
1128				    spa->spa_name, hostname,
1129				    (unsigned long)hostid);
1130				error = EBADF;
1131				goto out;
1132			}
1133		}
1134
1135		spa_config_set(spa, newconfig);
1136		spa_unload(spa);
1137		spa_deactivate(spa);
1138		spa_activate(spa);
1139
1140		return (spa_load(spa, newconfig, state, B_TRUE));
1141	}
1142
1143	if (zap_lookup(spa->spa_meta_objset,
1144	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
1145	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj) != 0) {
1146		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1147		    VDEV_AUX_CORRUPT_DATA);
1148		error = EIO;
1149		goto out;
1150	}
1151
1152	/*
1153	 * Load the bit that tells us to use the new accounting function
1154	 * (raid-z deflation).  If we have an older pool, this will not
1155	 * be present.
1156	 */
1157	error = zap_lookup(spa->spa_meta_objset,
1158	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
1159	    sizeof (uint64_t), 1, &spa->spa_deflate);
1160	if (error != 0 && error != ENOENT) {
1161		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1162		    VDEV_AUX_CORRUPT_DATA);
1163		error = EIO;
1164		goto out;
1165	}
1166
1167	/*
1168	 * Load the persistent error log.  If we have an older pool, this will
1169	 * not be present.
1170	 */
1171	error = zap_lookup(spa->spa_meta_objset,
1172	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_LAST,
1173	    sizeof (uint64_t), 1, &spa->spa_errlog_last);
1174	if (error != 0 && error != ENOENT) {
1175		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1176		    VDEV_AUX_CORRUPT_DATA);
1177		error = EIO;
1178		goto out;
1179	}
1180
1181	error = zap_lookup(spa->spa_meta_objset,
1182	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_SCRUB,
1183	    sizeof (uint64_t), 1, &spa->spa_errlog_scrub);
1184	if (error != 0 && error != ENOENT) {
1185		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1186		    VDEV_AUX_CORRUPT_DATA);
1187		error = EIO;
1188		goto out;
1189	}
1190
1191	/*
1192	 * Load the history object.  If we have an older pool, this
1193	 * will not be present.
1194	 */
1195	error = zap_lookup(spa->spa_meta_objset,
1196	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_HISTORY,
1197	    sizeof (uint64_t), 1, &spa->spa_history);
1198	if (error != 0 && error != ENOENT) {
1199		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1200		    VDEV_AUX_CORRUPT_DATA);
1201		error = EIO;
1202		goto out;
1203	}
1204
1205	/*
1206	 * Load any hot spares for this pool.
1207	 */
1208	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1209	    DMU_POOL_SPARES, sizeof (uint64_t), 1, &spa->spa_spares.sav_object);
1210	if (error != 0 && error != ENOENT) {
1211		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1212		    VDEV_AUX_CORRUPT_DATA);
1213		error = EIO;
1214		goto out;
1215	}
1216	if (error == 0) {
1217		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
1218		if (load_nvlist(spa, spa->spa_spares.sav_object,
1219		    &spa->spa_spares.sav_config) != 0) {
1220			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1221			    VDEV_AUX_CORRUPT_DATA);
1222			error = EIO;
1223			goto out;
1224		}
1225
1226		spa_config_enter(spa, RW_WRITER, FTAG);
1227		spa_load_spares(spa);
1228		spa_config_exit(spa, FTAG);
1229	}
1230
1231	/*
1232	 * Load any level 2 ARC devices for this pool.
1233	 */
1234	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1235	    DMU_POOL_L2CACHE, sizeof (uint64_t), 1,
1236	    &spa->spa_l2cache.sav_object);
1237	if (error != 0 && error != ENOENT) {
1238		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1239		    VDEV_AUX_CORRUPT_DATA);
1240		error = EIO;
1241		goto out;
1242	}
1243	if (error == 0) {
1244		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
1245		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
1246		    &spa->spa_l2cache.sav_config) != 0) {
1247			vdev_set_state(rvd, B_TRUE,
1248			    VDEV_STATE_CANT_OPEN,
1249			    VDEV_AUX_CORRUPT_DATA);
1250			error = EIO;
1251			goto out;
1252		}
1253
1254		spa_config_enter(spa, RW_WRITER, FTAG);
1255		spa_load_l2cache(spa);
1256		spa_config_exit(spa, FTAG);
1257	}
1258
1259	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
1260
1261	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1262	    DMU_POOL_PROPS, sizeof (uint64_t), 1, &spa->spa_pool_props_object);
1263
1264	if (error && error != ENOENT) {
1265		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1266		    VDEV_AUX_CORRUPT_DATA);
1267		error = EIO;
1268		goto out;
1269	}
1270
1271	if (error == 0) {
1272		(void) zap_lookup(spa->spa_meta_objset,
1273		    spa->spa_pool_props_object,
1274		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS),
1275		    sizeof (uint64_t), 1, &spa->spa_bootfs);
1276		(void) zap_lookup(spa->spa_meta_objset,
1277		    spa->spa_pool_props_object,
1278		    zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE),
1279		    sizeof (uint64_t), 1, &autoreplace);
1280		(void) zap_lookup(spa->spa_meta_objset,
1281		    spa->spa_pool_props_object,
1282		    zpool_prop_to_name(ZPOOL_PROP_DELEGATION),
1283		    sizeof (uint64_t), 1, &spa->spa_delegation);
1284		(void) zap_lookup(spa->spa_meta_objset,
1285		    spa->spa_pool_props_object,
1286		    zpool_prop_to_name(ZPOOL_PROP_FAILUREMODE),
1287		    sizeof (uint64_t), 1, &spa->spa_failmode);
1288	}
1289
1290	/*
1291	 * If the 'autoreplace' property is set, then post a resource notifying
1292	 * the ZFS DE that it should not issue any faults for unopenable
1293	 * devices.  We also iterate over the vdevs, and post a sysevent for any
1294	 * unopenable vdevs so that the normal autoreplace handler can take
1295	 * over.
1296	 */
1297	if (autoreplace && state != SPA_LOAD_TRYIMPORT)
1298		spa_check_removed(spa->spa_root_vdev);
1299
1300	/*
1301	 * Load the vdev state for all toplevel vdevs.
1302	 */
1303	vdev_load(rvd);
1304
1305	/*
1306	 * Propagate the leaf DTLs we just loaded all the way up the tree.
1307	 */
1308	spa_config_enter(spa, RW_WRITER, FTAG);
1309	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
1310	spa_config_exit(spa, FTAG);
1311
1312	/*
1313	 * Check the state of the root vdev.  If it can't be opened, it
1314	 * indicates one or more toplevel vdevs are faulted.
1315	 */
1316	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
1317		error = ENXIO;
1318		goto out;
1319	}
1320
1321	if ((spa_mode & FWRITE) && state != SPA_LOAD_TRYIMPORT) {
1322		dmu_tx_t *tx;
1323		int need_update = B_FALSE;
1324		int c;
1325
1326		/*
1327		 * Claim log blocks that haven't been committed yet.
1328		 * This must all happen in a single txg.
1329		 */
1330		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
1331		    spa_first_txg(spa));
1332		(void) dmu_objset_find(spa->spa_name,
1333		    zil_claim, tx, DS_FIND_CHILDREN);
1334		dmu_tx_commit(tx);
1335
1336		spa->spa_sync_on = B_TRUE;
1337		txg_sync_start(spa->spa_dsl_pool);
1338
1339		/*
1340		 * Wait for all claims to sync.
1341		 */
1342		txg_wait_synced(spa->spa_dsl_pool, 0);
1343
1344		/*
1345		 * If the config cache is stale, or we have uninitialized
1346		 * metaslabs (see spa_vdev_add()), then update the config.
1347		 */
1348		if (config_cache_txg != spa->spa_config_txg ||
1349		    state == SPA_LOAD_IMPORT)
1350			need_update = B_TRUE;
1351
1352		for (c = 0; c < rvd->vdev_children; c++)
1353			if (rvd->vdev_child[c]->vdev_ms_array == 0)
1354				need_update = B_TRUE;
1355
1356		/*
1357		 * Update the config cache asychronously in case we're the
1358		 * root pool, in which case the config cache isn't writable yet.
1359		 */
1360		if (need_update)
1361			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
1362	}
1363
1364	error = 0;
1365out:
1366	if (error && error != EBADF)
1367		zfs_ereport_post(FM_EREPORT_ZFS_POOL, spa, NULL, NULL, 0, 0);
1368	spa->spa_load_state = SPA_LOAD_NONE;
1369	spa->spa_ena = 0;
1370
1371	return (error);
1372}
1373
1374/*
1375 * Pool Open/Import
1376 *
1377 * The import case is identical to an open except that the configuration is sent
1378 * down from userland, instead of grabbed from the configuration cache.  For the
1379 * case of an open, the pool configuration will exist in the
1380 * POOL_STATE_UNINITIALIZED state.
1381 *
1382 * The stats information (gen/count/ustats) is used to gather vdev statistics at
1383 * the same time open the pool, without having to keep around the spa_t in some
1384 * ambiguous state.
1385 */
1386static int
1387spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t **config)
1388{
1389	spa_t *spa;
1390	int error;
1391	int loaded = B_FALSE;
1392	int locked = B_FALSE;
1393
1394	*spapp = NULL;
1395
1396	/*
1397	 * As disgusting as this is, we need to support recursive calls to this
1398	 * function because dsl_dir_open() is called during spa_load(), and ends
1399	 * up calling spa_open() again.  The real fix is to figure out how to
1400	 * avoid dsl_dir_open() calling this in the first place.
1401	 */
1402	if (mutex_owner(&spa_namespace_lock) != curthread) {
1403		mutex_enter(&spa_namespace_lock);
1404		locked = B_TRUE;
1405	}
1406
1407	if ((spa = spa_lookup(pool)) == NULL) {
1408		if (locked)
1409			mutex_exit(&spa_namespace_lock);
1410		return (ENOENT);
1411	}
1412	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
1413
1414		spa_activate(spa);
1415
1416		error = spa_load(spa, spa->spa_config, SPA_LOAD_OPEN, B_FALSE);
1417
1418		if (error == EBADF) {
1419			/*
1420			 * If vdev_validate() returns failure (indicated by
1421			 * EBADF), it indicates that one of the vdevs indicates
1422			 * that the pool has been exported or destroyed.  If
1423			 * this is the case, the config cache is out of sync and
1424			 * we should remove the pool from the namespace.
1425			 */
1426			spa_unload(spa);
1427			spa_deactivate(spa);
1428			spa_config_sync(spa, B_TRUE, B_TRUE);
1429			spa_remove(spa);
1430			if (locked)
1431				mutex_exit(&spa_namespace_lock);
1432			return (ENOENT);
1433		}
1434
1435		if (error) {
1436			/*
1437			 * We can't open the pool, but we still have useful
1438			 * information: the state of each vdev after the
1439			 * attempted vdev_open().  Return this to the user.
1440			 */
1441			if (config != NULL && spa->spa_root_vdev != NULL) {
1442				spa_config_enter(spa, RW_READER, FTAG);
1443				*config = spa_config_generate(spa, NULL, -1ULL,
1444				    B_TRUE);
1445				spa_config_exit(spa, FTAG);
1446			}
1447			spa_unload(spa);
1448			spa_deactivate(spa);
1449			spa->spa_last_open_failed = B_TRUE;
1450			if (locked)
1451				mutex_exit(&spa_namespace_lock);
1452			*spapp = NULL;
1453			return (error);
1454		} else {
1455			spa->spa_last_open_failed = B_FALSE;
1456		}
1457
1458		loaded = B_TRUE;
1459	}
1460
1461	spa_open_ref(spa, tag);
1462
1463	/*
1464	 * If we just loaded the pool, resilver anything that's out of date.
1465	 */
1466	if (loaded && (spa_mode & FWRITE))
1467		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
1468
1469	if (locked)
1470		mutex_exit(&spa_namespace_lock);
1471
1472	*spapp = spa;
1473
1474	if (config != NULL) {
1475		spa_config_enter(spa, RW_READER, FTAG);
1476		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
1477		spa_config_exit(spa, FTAG);
1478	}
1479
1480	return (0);
1481}
1482
1483int
1484spa_open(const char *name, spa_t **spapp, void *tag)
1485{
1486	return (spa_open_common(name, spapp, tag, NULL));
1487}
1488
1489/*
1490 * Lookup the given spa_t, incrementing the inject count in the process,
1491 * preventing it from being exported or destroyed.
1492 */
1493spa_t *
1494spa_inject_addref(char *name)
1495{
1496	spa_t *spa;
1497
1498	mutex_enter(&spa_namespace_lock);
1499	if ((spa = spa_lookup(name)) == NULL) {
1500		mutex_exit(&spa_namespace_lock);
1501		return (NULL);
1502	}
1503	spa->spa_inject_ref++;
1504	mutex_exit(&spa_namespace_lock);
1505
1506	return (spa);
1507}
1508
1509void
1510spa_inject_delref(spa_t *spa)
1511{
1512	mutex_enter(&spa_namespace_lock);
1513	spa->spa_inject_ref--;
1514	mutex_exit(&spa_namespace_lock);
1515}
1516
1517/*
1518 * Add spares device information to the nvlist.
1519 */
1520static void
1521spa_add_spares(spa_t *spa, nvlist_t *config)
1522{
1523	nvlist_t **spares;
1524	uint_t i, nspares;
1525	nvlist_t *nvroot;
1526	uint64_t guid;
1527	vdev_stat_t *vs;
1528	uint_t vsc;
1529	uint64_t pool;
1530
1531	if (spa->spa_spares.sav_count == 0)
1532		return;
1533
1534	VERIFY(nvlist_lookup_nvlist(config,
1535	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1536	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1537	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1538	if (nspares != 0) {
1539		VERIFY(nvlist_add_nvlist_array(nvroot,
1540		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1541		VERIFY(nvlist_lookup_nvlist_array(nvroot,
1542		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1543
1544		/*
1545		 * Go through and find any spares which have since been
1546		 * repurposed as an active spare.  If this is the case, update
1547		 * their status appropriately.
1548		 */
1549		for (i = 0; i < nspares; i++) {
1550			VERIFY(nvlist_lookup_uint64(spares[i],
1551			    ZPOOL_CONFIG_GUID, &guid) == 0);
1552			if (spa_spare_exists(guid, &pool) && pool != 0ULL) {
1553				VERIFY(nvlist_lookup_uint64_array(
1554				    spares[i], ZPOOL_CONFIG_STATS,
1555				    (uint64_t **)&vs, &vsc) == 0);
1556				vs->vs_state = VDEV_STATE_CANT_OPEN;
1557				vs->vs_aux = VDEV_AUX_SPARED;
1558			}
1559		}
1560	}
1561}
1562
1563/*
1564 * Add l2cache device information to the nvlist, including vdev stats.
1565 */
1566static void
1567spa_add_l2cache(spa_t *spa, nvlist_t *config)
1568{
1569	nvlist_t **l2cache;
1570	uint_t i, j, nl2cache;
1571	nvlist_t *nvroot;
1572	uint64_t guid;
1573	vdev_t *vd;
1574	vdev_stat_t *vs;
1575	uint_t vsc;
1576
1577	if (spa->spa_l2cache.sav_count == 0)
1578		return;
1579
1580	spa_config_enter(spa, RW_READER, FTAG);
1581
1582	VERIFY(nvlist_lookup_nvlist(config,
1583	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1584	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
1585	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1586	if (nl2cache != 0) {
1587		VERIFY(nvlist_add_nvlist_array(nvroot,
1588		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
1589		VERIFY(nvlist_lookup_nvlist_array(nvroot,
1590		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1591
1592		/*
1593		 * Update level 2 cache device stats.
1594		 */
1595
1596		for (i = 0; i < nl2cache; i++) {
1597			VERIFY(nvlist_lookup_uint64(l2cache[i],
1598			    ZPOOL_CONFIG_GUID, &guid) == 0);
1599
1600			vd = NULL;
1601			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
1602				if (guid ==
1603				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
1604					vd = spa->spa_l2cache.sav_vdevs[j];
1605					break;
1606				}
1607			}
1608			ASSERT(vd != NULL);
1609
1610			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
1611			    ZPOOL_CONFIG_STATS, (uint64_t **)&vs, &vsc) == 0);
1612			vdev_get_stats(vd, vs);
1613		}
1614	}
1615
1616	spa_config_exit(spa, FTAG);
1617}
1618
1619int
1620spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
1621{
1622	int error;
1623	spa_t *spa;
1624
1625	*config = NULL;
1626	error = spa_open_common(name, &spa, FTAG, config);
1627
1628	if (spa && *config != NULL) {
1629		VERIFY(nvlist_add_uint64(*config, ZPOOL_CONFIG_ERRCOUNT,
1630		    spa_get_errlog_size(spa)) == 0);
1631
1632		spa_add_spares(spa, *config);
1633		spa_add_l2cache(spa, *config);
1634	}
1635
1636	/*
1637	 * We want to get the alternate root even for faulted pools, so we cheat
1638	 * and call spa_lookup() directly.
1639	 */
1640	if (altroot) {
1641		if (spa == NULL) {
1642			mutex_enter(&spa_namespace_lock);
1643			spa = spa_lookup(name);
1644			if (spa)
1645				spa_altroot(spa, altroot, buflen);
1646			else
1647				altroot[0] = '\0';
1648			spa = NULL;
1649			mutex_exit(&spa_namespace_lock);
1650		} else {
1651			spa_altroot(spa, altroot, buflen);
1652		}
1653	}
1654
1655	if (spa != NULL)
1656		spa_close(spa, FTAG);
1657
1658	return (error);
1659}
1660
1661/*
1662 * Validate that the auxiliary device array is well formed.  We must have an
1663 * array of nvlists, each which describes a valid leaf vdev.  If this is an
1664 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
1665 * specified, as long as they are well-formed.
1666 */
1667static int
1668spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
1669    spa_aux_vdev_t *sav, const char *config, uint64_t version,
1670    vdev_labeltype_t label)
1671{
1672	nvlist_t **dev;
1673	uint_t i, ndev;
1674	vdev_t *vd;
1675	int error;
1676
1677	/*
1678	 * It's acceptable to have no devs specified.
1679	 */
1680	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
1681		return (0);
1682
1683	if (ndev == 0)
1684		return (EINVAL);
1685
1686	/*
1687	 * Make sure the pool is formatted with a version that supports this
1688	 * device type.
1689	 */
1690	if (spa_version(spa) < version)
1691		return (ENOTSUP);
1692
1693	/*
1694	 * Set the pending device list so we correctly handle device in-use
1695	 * checking.
1696	 */
1697	sav->sav_pending = dev;
1698	sav->sav_npending = ndev;
1699
1700	for (i = 0; i < ndev; i++) {
1701		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
1702		    mode)) != 0)
1703			goto out;
1704
1705		if (!vd->vdev_ops->vdev_op_leaf) {
1706			vdev_free(vd);
1707			error = EINVAL;
1708			goto out;
1709		}
1710
1711		/*
1712		 * The L2ARC currently only supports disk devices.
1713		 */
1714		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
1715		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
1716			error = ENOTBLK;
1717			goto out;
1718		}
1719
1720		vd->vdev_top = vd;
1721
1722		if ((error = vdev_open(vd)) == 0 &&
1723		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
1724			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
1725			    vd->vdev_guid) == 0);
1726		}
1727
1728		vdev_free(vd);
1729
1730		if (error &&
1731		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
1732			goto out;
1733		else
1734			error = 0;
1735	}
1736
1737out:
1738	sav->sav_pending = NULL;
1739	sav->sav_npending = 0;
1740	return (error);
1741}
1742
1743static int
1744spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
1745{
1746	int error;
1747
1748	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
1749	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
1750	    VDEV_LABEL_SPARE)) != 0) {
1751		return (error);
1752	}
1753
1754	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
1755	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
1756	    VDEV_LABEL_L2CACHE));
1757}
1758
1759static void
1760spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
1761    const char *config)
1762{
1763	int i;
1764
1765	if (sav->sav_config != NULL) {
1766		nvlist_t **olddevs;
1767		uint_t oldndevs;
1768		nvlist_t **newdevs;
1769
1770		/*
1771		 * Generate new dev list by concatentating with the
1772		 * current dev list.
1773		 */
1774		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
1775		    &olddevs, &oldndevs) == 0);
1776
1777		newdevs = kmem_alloc(sizeof (void *) *
1778		    (ndevs + oldndevs), KM_SLEEP);
1779		for (i = 0; i < oldndevs; i++)
1780			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
1781			    KM_SLEEP) == 0);
1782		for (i = 0; i < ndevs; i++)
1783			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
1784			    KM_SLEEP) == 0);
1785
1786		VERIFY(nvlist_remove(sav->sav_config, config,
1787		    DATA_TYPE_NVLIST_ARRAY) == 0);
1788
1789		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1790		    config, newdevs, ndevs + oldndevs) == 0);
1791		for (i = 0; i < oldndevs + ndevs; i++)
1792			nvlist_free(newdevs[i]);
1793		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
1794	} else {
1795		/*
1796		 * Generate a new dev list.
1797		 */
1798		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
1799		    KM_SLEEP) == 0);
1800		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
1801		    devs, ndevs) == 0);
1802	}
1803}
1804
1805/*
1806 * Stop and drop level 2 ARC devices
1807 */
1808void
1809spa_l2cache_drop(spa_t *spa)
1810{
1811	vdev_t *vd;
1812	int i;
1813	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1814
1815	for (i = 0; i < sav->sav_count; i++) {
1816		uint64_t pool;
1817
1818		vd = sav->sav_vdevs[i];
1819		ASSERT(vd != NULL);
1820
1821		if (spa_mode & FWRITE &&
1822		    spa_l2cache_exists(vd->vdev_guid, &pool) && pool != 0ULL &&
1823		    l2arc_vdev_present(vd)) {
1824			l2arc_remove_vdev(vd);
1825		}
1826		if (vd->vdev_isl2cache)
1827			spa_l2cache_remove(vd);
1828		vdev_clear_stats(vd);
1829		(void) vdev_close(vd);
1830	}
1831}
1832
1833/*
1834 * Pool Creation
1835 */
1836int
1837spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
1838    const char *history_str)
1839{
1840	spa_t *spa;
1841	char *altroot = NULL;
1842	vdev_t *rvd;
1843	dsl_pool_t *dp;
1844	dmu_tx_t *tx;
1845	int c, error = 0;
1846	uint64_t txg = TXG_INITIAL;
1847	nvlist_t **spares, **l2cache;
1848	uint_t nspares, nl2cache;
1849	uint64_t version;
1850
1851	/*
1852	 * If this pool already exists, return failure.
1853	 */
1854	mutex_enter(&spa_namespace_lock);
1855	if (spa_lookup(pool) != NULL) {
1856		mutex_exit(&spa_namespace_lock);
1857		return (EEXIST);
1858	}
1859
1860	/*
1861	 * Allocate a new spa_t structure.
1862	 */
1863	(void) nvlist_lookup_string(props,
1864	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
1865	spa = spa_add(pool, altroot);
1866	spa_activate(spa);
1867
1868	spa->spa_uberblock.ub_txg = txg - 1;
1869
1870	if (props && (error = spa_prop_validate(spa, props))) {
1871		spa_unload(spa);
1872		spa_deactivate(spa);
1873		spa_remove(spa);
1874		mutex_exit(&spa_namespace_lock);
1875		return (error);
1876	}
1877
1878	if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
1879	    &version) != 0)
1880		version = SPA_VERSION;
1881	ASSERT(version <= SPA_VERSION);
1882	spa->spa_uberblock.ub_version = version;
1883	spa->spa_ubsync = spa->spa_uberblock;
1884
1885	/*
1886	 * Create the root vdev.
1887	 */
1888	spa_config_enter(spa, RW_WRITER, FTAG);
1889
1890	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
1891
1892	ASSERT(error != 0 || rvd != NULL);
1893	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
1894
1895	if (error == 0 && !zfs_allocatable_devs(nvroot))
1896		error = EINVAL;
1897
1898	if (error == 0 &&
1899	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
1900	    (error = spa_validate_aux(spa, nvroot, txg,
1901	    VDEV_ALLOC_ADD)) == 0) {
1902		for (c = 0; c < rvd->vdev_children; c++)
1903			vdev_init(rvd->vdev_child[c], txg);
1904		vdev_config_dirty(rvd);
1905	}
1906
1907	spa_config_exit(spa, FTAG);
1908
1909	if (error != 0) {
1910		spa_unload(spa);
1911		spa_deactivate(spa);
1912		spa_remove(spa);
1913		mutex_exit(&spa_namespace_lock);
1914		return (error);
1915	}
1916
1917	/*
1918	 * Get the list of spares, if specified.
1919	 */
1920	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1921	    &spares, &nspares) == 0) {
1922		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
1923		    KM_SLEEP) == 0);
1924		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1925		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1926		spa_config_enter(spa, RW_WRITER, FTAG);
1927		spa_load_spares(spa);
1928		spa_config_exit(spa, FTAG);
1929		spa->spa_spares.sav_sync = B_TRUE;
1930	}
1931
1932	/*
1933	 * Get the list of level 2 cache devices, if specified.
1934	 */
1935	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1936	    &l2cache, &nl2cache) == 0) {
1937		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
1938		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
1939		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
1940		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
1941		spa_config_enter(spa, RW_WRITER, FTAG);
1942		spa_load_l2cache(spa);
1943		spa_config_exit(spa, FTAG);
1944		spa->spa_l2cache.sav_sync = B_TRUE;
1945	}
1946
1947	spa->spa_dsl_pool = dp = dsl_pool_create(spa, txg);
1948	spa->spa_meta_objset = dp->dp_meta_objset;
1949
1950	tx = dmu_tx_create_assigned(dp, txg);
1951
1952	/*
1953	 * Create the pool config object.
1954	 */
1955	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
1956	    DMU_OT_PACKED_NVLIST, 1 << 14,
1957	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
1958
1959	if (zap_add(spa->spa_meta_objset,
1960	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
1961	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
1962		cmn_err(CE_PANIC, "failed to add pool config");
1963	}
1964
1965	/* Newly created pools with the right version are always deflated. */
1966	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
1967		spa->spa_deflate = TRUE;
1968		if (zap_add(spa->spa_meta_objset,
1969		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
1970		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
1971			cmn_err(CE_PANIC, "failed to add deflate");
1972		}
1973	}
1974
1975	/*
1976	 * Create the deferred-free bplist object.  Turn off compression
1977	 * because sync-to-convergence takes longer if the blocksize
1978	 * keeps changing.
1979	 */
1980	spa->spa_sync_bplist_obj = bplist_create(spa->spa_meta_objset,
1981	    1 << 14, tx);
1982	dmu_object_set_compress(spa->spa_meta_objset, spa->spa_sync_bplist_obj,
1983	    ZIO_COMPRESS_OFF, tx);
1984
1985	if (zap_add(spa->spa_meta_objset,
1986	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
1987	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj, tx) != 0) {
1988		cmn_err(CE_PANIC, "failed to add bplist");
1989	}
1990
1991	/*
1992	 * Create the pool's history object.
1993	 */
1994	if (version >= SPA_VERSION_ZPOOL_HISTORY)
1995		spa_history_create_obj(spa, tx);
1996
1997	/*
1998	 * Set pool properties.
1999	 */
2000	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
2001	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2002	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
2003	if (props)
2004		spa_sync_props(spa, props, CRED(), tx);
2005
2006	dmu_tx_commit(tx);
2007
2008	spa->spa_sync_on = B_TRUE;
2009	txg_sync_start(spa->spa_dsl_pool);
2010
2011	/*
2012	 * We explicitly wait for the first transaction to complete so that our
2013	 * bean counters are appropriately updated.
2014	 */
2015	txg_wait_synced(spa->spa_dsl_pool, txg);
2016
2017	spa_config_sync(spa, B_FALSE, B_TRUE);
2018
2019	if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
2020		(void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
2021
2022	mutex_exit(&spa_namespace_lock);
2023
2024	return (0);
2025}
2026
2027/*
2028 * Import the given pool into the system.  We set up the necessary spa_t and
2029 * then call spa_load() to do the dirty work.
2030 */
2031static int
2032spa_import_common(const char *pool, nvlist_t *config, nvlist_t *props,
2033    boolean_t isroot, boolean_t allowfaulted)
2034{
2035	spa_t *spa;
2036	char *altroot = NULL;
2037	int error, loaderr;
2038	nvlist_t *nvroot;
2039	nvlist_t **spares, **l2cache;
2040	uint_t nspares, nl2cache;
2041	int mosconfig = isroot? B_FALSE : B_TRUE;
2042
2043	/*
2044	 * If a pool with this name exists, return failure.
2045	 */
2046	mutex_enter(&spa_namespace_lock);
2047	if (spa_lookup(pool) != NULL) {
2048		mutex_exit(&spa_namespace_lock);
2049		return (EEXIST);
2050	}
2051
2052	/*
2053	 * Create and initialize the spa structure.
2054	 */
2055	(void) nvlist_lookup_string(props,
2056	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2057	spa = spa_add(pool, altroot);
2058	spa_activate(spa);
2059
2060	if (allowfaulted)
2061		spa->spa_import_faulted = B_TRUE;
2062	spa->spa_is_root = isroot;
2063
2064	/*
2065	 * Pass off the heavy lifting to spa_load().
2066	 * Pass TRUE for mosconfig because the user-supplied config
2067	 * is actually the one to trust when doing an import.
2068	 */
2069	loaderr = error = spa_load(spa, config, SPA_LOAD_IMPORT, mosconfig);
2070
2071	spa_config_enter(spa, RW_WRITER, FTAG);
2072	/*
2073	 * Toss any existing sparelist, as it doesn't have any validity anymore,
2074	 * and conflicts with spa_has_spare().
2075	 */
2076	if (!isroot && spa->spa_spares.sav_config) {
2077		nvlist_free(spa->spa_spares.sav_config);
2078		spa->spa_spares.sav_config = NULL;
2079		spa_load_spares(spa);
2080	}
2081	if (!isroot && spa->spa_l2cache.sav_config) {
2082		nvlist_free(spa->spa_l2cache.sav_config);
2083		spa->spa_l2cache.sav_config = NULL;
2084		spa_load_l2cache(spa);
2085	}
2086
2087	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2088	    &nvroot) == 0);
2089	if (error == 0)
2090		error = spa_validate_aux(spa, nvroot, -1ULL, VDEV_ALLOC_SPARE);
2091	if (error == 0)
2092		error = spa_validate_aux(spa, nvroot, -1ULL,
2093		    VDEV_ALLOC_L2CACHE);
2094	spa_config_exit(spa, FTAG);
2095
2096	if (error != 0 || (props && (error = spa_prop_set(spa, props)))) {
2097		if (loaderr != 0 && loaderr != EINVAL && allowfaulted) {
2098			/*
2099			 * If we failed to load the pool, but 'allowfaulted' is
2100			 * set, then manually set the config as if the config
2101			 * passed in was specified in the cache file.
2102			 */
2103			error = 0;
2104			spa->spa_import_faulted = B_FALSE;
2105			if (spa->spa_config == NULL) {
2106				spa_config_enter(spa, RW_READER, FTAG);
2107				spa->spa_config = spa_config_generate(spa,
2108				    NULL, -1ULL, B_TRUE);
2109				spa_config_exit(spa, FTAG);
2110			}
2111			spa_unload(spa);
2112			spa_deactivate(spa);
2113			spa_config_sync(spa, B_FALSE, B_TRUE);
2114		} else {
2115			spa_unload(spa);
2116			spa_deactivate(spa);
2117			spa_remove(spa);
2118		}
2119		mutex_exit(&spa_namespace_lock);
2120		return (error);
2121	}
2122
2123	/*
2124	 * Override any spares and level 2 cache devices as specified by
2125	 * the user, as these may have correct device names/devids, etc.
2126	 */
2127	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2128	    &spares, &nspares) == 0) {
2129		if (spa->spa_spares.sav_config)
2130			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
2131			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
2132		else
2133			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
2134			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2135		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2136		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2137		spa_config_enter(spa, RW_WRITER, FTAG);
2138		spa_load_spares(spa);
2139		spa_config_exit(spa, FTAG);
2140		spa->spa_spares.sav_sync = B_TRUE;
2141	}
2142	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2143	    &l2cache, &nl2cache) == 0) {
2144		if (spa->spa_l2cache.sav_config)
2145			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
2146			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
2147		else
2148			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2149			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2150		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2151		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2152		spa_config_enter(spa, RW_WRITER, FTAG);
2153		spa_load_l2cache(spa);
2154		spa_config_exit(spa, FTAG);
2155		spa->spa_l2cache.sav_sync = B_TRUE;
2156	}
2157
2158	if (spa_mode & FWRITE) {
2159		/*
2160		 * Update the config cache to include the newly-imported pool.
2161		 */
2162		spa_config_update_common(spa, SPA_CONFIG_UPDATE_POOL, isroot);
2163
2164		/*
2165		 * Resilver anything that's out of date.
2166		 */
2167		if (!isroot)
2168			VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER,
2169			    B_TRUE) == 0);
2170	}
2171
2172	spa->spa_import_faulted = B_FALSE;
2173	mutex_exit(&spa_namespace_lock);
2174
2175	return (0);
2176}
2177
2178#ifdef _KERNEL
2179/*
2180 * Build a "root" vdev for a top level vdev read in from a rootpool
2181 * device label.
2182 */
2183static void
2184spa_build_rootpool_config(nvlist_t *config)
2185{
2186	nvlist_t *nvtop, *nvroot;
2187	uint64_t pgid;
2188
2189	/*
2190	 * Add this top-level vdev to the child array.
2191	 */
2192	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtop)
2193	    == 0);
2194	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pgid)
2195	    == 0);
2196
2197	/*
2198	 * Put this pool's top-level vdevs into a root vdev.
2199	 */
2200	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2201	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT)
2202	    == 0);
2203	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
2204	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
2205	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2206	    &nvtop, 1) == 0);
2207
2208	/*
2209	 * Replace the existing vdev_tree with the new root vdev in
2210	 * this pool's configuration (remove the old, add the new).
2211	 */
2212	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
2213	nvlist_free(nvroot);
2214}
2215
2216/*
2217 * Get the root pool information from the root disk, then import the root pool
2218 * during the system boot up time.
2219 */
2220extern nvlist_t *vdev_disk_read_rootlabel(char *);
2221
2222void
2223spa_check_rootconf(char *devpath, char **bestdev, nvlist_t **bestconf,
2224    uint64_t *besttxg)
2225{
2226	nvlist_t *config;
2227	uint64_t txg;
2228
2229	if ((config = vdev_disk_read_rootlabel(devpath)) == NULL)
2230		return;
2231
2232	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
2233
2234	if (txg > *besttxg) {
2235		*besttxg = txg;
2236		if (*bestconf != NULL)
2237			nvlist_free(*bestconf);
2238		*bestconf = config;
2239		*bestdev = devpath;
2240	}
2241}
2242
2243boolean_t
2244spa_rootdev_validate(nvlist_t *nv)
2245{
2246	uint64_t ival;
2247
2248	if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
2249	    nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
2250	    nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, &ival) == 0 ||
2251	    nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
2252		return (B_FALSE);
2253
2254	return (B_TRUE);
2255}
2256
2257/*
2258 * Import a root pool.
2259 *
2260 * For x86. devpath_list will consist the physpath name of the vdev in a single
2261 * disk root pool or a list of physnames for the vdevs in a mirrored rootpool.
2262 * e.g.
2263 *	"/pci@1f,0/ide@d/disk@0,0:a /pci@1f,o/ide@d/disk@2,0:a"
2264 *
2265 * For Sparc, devpath_list consists the physpath name of the booting device
2266 * no matter the rootpool is a single device pool or a mirrored pool.
2267 * e.g.
2268 *	"/pci@1f,0/ide@d/disk@0,0:a"
2269 */
2270int
2271spa_import_rootpool(char *devpath_list)
2272{
2273	nvlist_t *conf = NULL;
2274	char *dev = NULL;
2275	char *pname;
2276	int error;
2277
2278	/*
2279	 * Get the vdev pathname and configuation from the most
2280	 * recently updated vdev (highest txg).
2281	 */
2282	if (error = spa_get_rootconf(devpath_list, &dev, &conf))
2283		goto msg_out;
2284
2285	/*
2286	 * Add type "root" vdev to the config.
2287	 */
2288	spa_build_rootpool_config(conf);
2289
2290	VERIFY(nvlist_lookup_string(conf, ZPOOL_CONFIG_POOL_NAME, &pname) == 0);
2291
2292	/*
2293	 * We specify 'allowfaulted' for this to be treated like spa_open()
2294	 * instead of spa_import().  This prevents us from marking vdevs as
2295	 * persistently unavailable, and generates FMA ereports as if it were a
2296	 * pool open, not import.
2297	 */
2298	error = spa_import_common(pname, conf, NULL, B_TRUE, B_TRUE);
2299	if (error == EEXIST)
2300		error = 0;
2301
2302	nvlist_free(conf);
2303	return (error);
2304
2305msg_out:
2306	cmn_err(CE_NOTE, "\n\n"
2307	    "  ***************************************************  \n"
2308	    "  *  This device is not bootable!                   *  \n"
2309	    "  *  It is either offlined or detached or faulted.  *  \n"
2310	    "  *  Please try to boot from a different device.    *  \n"
2311	    "  ***************************************************  \n\n");
2312
2313	return (error);
2314}
2315#endif
2316
2317/*
2318 * Import a non-root pool into the system.
2319 */
2320int
2321spa_import(const char *pool, nvlist_t *config, nvlist_t *props)
2322{
2323	return (spa_import_common(pool, config, props, B_FALSE, B_FALSE));
2324}
2325
2326int
2327spa_import_faulted(const char *pool, nvlist_t *config, nvlist_t *props)
2328{
2329	return (spa_import_common(pool, config, props, B_FALSE, B_TRUE));
2330}
2331
2332
2333/*
2334 * This (illegal) pool name is used when temporarily importing a spa_t in order
2335 * to get the vdev stats associated with the imported devices.
2336 */
2337#define	TRYIMPORT_NAME	"$import"
2338
2339nvlist_t *
2340spa_tryimport(nvlist_t *tryconfig)
2341{
2342	nvlist_t *config = NULL;
2343	char *poolname;
2344	spa_t *spa;
2345	uint64_t state;
2346
2347	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
2348		return (NULL);
2349
2350	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
2351		return (NULL);
2352
2353	/*
2354	 * Create and initialize the spa structure.
2355	 */
2356	mutex_enter(&spa_namespace_lock);
2357	spa = spa_add(TRYIMPORT_NAME, NULL);
2358	spa_activate(spa);
2359
2360	/*
2361	 * Pass off the heavy lifting to spa_load().
2362	 * Pass TRUE for mosconfig because the user-supplied config
2363	 * is actually the one to trust when doing an import.
2364	 */
2365	(void) spa_load(spa, tryconfig, SPA_LOAD_TRYIMPORT, B_TRUE);
2366
2367	/*
2368	 * If 'tryconfig' was at least parsable, return the current config.
2369	 */
2370	if (spa->spa_root_vdev != NULL) {
2371		spa_config_enter(spa, RW_READER, FTAG);
2372		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2373		spa_config_exit(spa, FTAG);
2374		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
2375		    poolname) == 0);
2376		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
2377		    state) == 0);
2378		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
2379		    spa->spa_uberblock.ub_timestamp) == 0);
2380
2381		/*
2382		 * If the bootfs property exists on this pool then we
2383		 * copy it out so that external consumers can tell which
2384		 * pools are bootable.
2385		 */
2386		if (spa->spa_bootfs) {
2387			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2388
2389			/*
2390			 * We have to play games with the name since the
2391			 * pool was opened as TRYIMPORT_NAME.
2392			 */
2393			if (dsl_dsobj_to_dsname(spa->spa_name,
2394			    spa->spa_bootfs, tmpname) == 0) {
2395				char *cp;
2396				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2397
2398				cp = strchr(tmpname, '/');
2399				if (cp == NULL) {
2400					(void) strlcpy(dsname, tmpname,
2401					    MAXPATHLEN);
2402				} else {
2403					(void) snprintf(dsname, MAXPATHLEN,
2404					    "%s/%s", poolname, ++cp);
2405				}
2406				VERIFY(nvlist_add_string(config,
2407				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
2408				kmem_free(dsname, MAXPATHLEN);
2409			}
2410			kmem_free(tmpname, MAXPATHLEN);
2411		}
2412
2413		/*
2414		 * Add the list of hot spares and level 2 cache devices.
2415		 */
2416		spa_add_spares(spa, config);
2417		spa_add_l2cache(spa, config);
2418	}
2419
2420	spa_unload(spa);
2421	spa_deactivate(spa);
2422	spa_remove(spa);
2423	mutex_exit(&spa_namespace_lock);
2424
2425	return (config);
2426}
2427
2428/*
2429 * Pool export/destroy
2430 *
2431 * The act of destroying or exporting a pool is very simple.  We make sure there
2432 * is no more pending I/O and any references to the pool are gone.  Then, we
2433 * update the pool state and sync all the labels to disk, removing the
2434 * configuration from the cache afterwards.
2435 */
2436static int
2437spa_export_common(char *pool, int new_state, nvlist_t **oldconfig)
2438{
2439	spa_t *spa;
2440
2441	if (oldconfig)
2442		*oldconfig = NULL;
2443
2444	if (!(spa_mode & FWRITE))
2445		return (EROFS);
2446
2447	mutex_enter(&spa_namespace_lock);
2448	if ((spa = spa_lookup(pool)) == NULL) {
2449		mutex_exit(&spa_namespace_lock);
2450		return (ENOENT);
2451	}
2452
2453	/*
2454	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
2455	 * reacquire the namespace lock, and see if we can export.
2456	 */
2457	spa_open_ref(spa, FTAG);
2458	mutex_exit(&spa_namespace_lock);
2459	spa_async_suspend(spa);
2460	mutex_enter(&spa_namespace_lock);
2461	spa_close(spa, FTAG);
2462
2463	/*
2464	 * The pool will be in core if it's openable,
2465	 * in which case we can modify its state.
2466	 */
2467	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
2468		/*
2469		 * Objsets may be open only because they're dirty, so we
2470		 * have to force it to sync before checking spa_refcnt.
2471		 */
2472		spa_scrub_suspend(spa);
2473		txg_wait_synced(spa->spa_dsl_pool, 0);
2474
2475		/*
2476		 * A pool cannot be exported or destroyed if there are active
2477		 * references.  If we are resetting a pool, allow references by
2478		 * fault injection handlers.
2479		 */
2480		if (!spa_refcount_zero(spa) ||
2481		    (spa->spa_inject_ref != 0 &&
2482		    new_state != POOL_STATE_UNINITIALIZED)) {
2483			spa_scrub_resume(spa);
2484			spa_async_resume(spa);
2485			mutex_exit(&spa_namespace_lock);
2486			return (EBUSY);
2487		}
2488
2489		spa_scrub_resume(spa);
2490		VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0);
2491
2492		/*
2493		 * We want this to be reflected on every label,
2494		 * so mark them all dirty.  spa_unload() will do the
2495		 * final sync that pushes these changes out.
2496		 */
2497		if (new_state != POOL_STATE_UNINITIALIZED) {
2498			spa_config_enter(spa, RW_WRITER, FTAG);
2499			spa->spa_state = new_state;
2500			spa->spa_final_txg = spa_last_synced_txg(spa) + 1;
2501			vdev_config_dirty(spa->spa_root_vdev);
2502			spa_config_exit(spa, FTAG);
2503		}
2504	}
2505
2506	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
2507
2508	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
2509		spa_unload(spa);
2510		spa_deactivate(spa);
2511	}
2512
2513	if (oldconfig && spa->spa_config)
2514		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
2515
2516	if (new_state != POOL_STATE_UNINITIALIZED) {
2517		spa_config_sync(spa, B_TRUE, B_TRUE);
2518		spa_remove(spa);
2519	}
2520	mutex_exit(&spa_namespace_lock);
2521
2522	return (0);
2523}
2524
2525/*
2526 * Destroy a storage pool.
2527 */
2528int
2529spa_destroy(char *pool)
2530{
2531	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL));
2532}
2533
2534/*
2535 * Export a storage pool.
2536 */
2537int
2538spa_export(char *pool, nvlist_t **oldconfig)
2539{
2540	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig));
2541}
2542
2543/*
2544 * Similar to spa_export(), this unloads the spa_t without actually removing it
2545 * from the namespace in any way.
2546 */
2547int
2548spa_reset(char *pool)
2549{
2550	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL));
2551}
2552
2553
2554/*
2555 * ==========================================================================
2556 * Device manipulation
2557 * ==========================================================================
2558 */
2559
2560/*
2561 * Add a device to a storage pool.
2562 */
2563int
2564spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
2565{
2566	uint64_t txg;
2567	int c, error;
2568	vdev_t *rvd = spa->spa_root_vdev;
2569	vdev_t *vd, *tvd;
2570	nvlist_t **spares, **l2cache;
2571	uint_t nspares, nl2cache;
2572
2573	txg = spa_vdev_enter(spa);
2574
2575	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
2576	    VDEV_ALLOC_ADD)) != 0)
2577		return (spa_vdev_exit(spa, NULL, txg, error));
2578
2579	spa->spa_pending_vdev = vd;
2580
2581	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
2582	    &nspares) != 0)
2583		nspares = 0;
2584
2585	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
2586	    &nl2cache) != 0)
2587		nl2cache = 0;
2588
2589	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) {
2590		spa->spa_pending_vdev = NULL;
2591		return (spa_vdev_exit(spa, vd, txg, EINVAL));
2592	}
2593
2594	if (vd->vdev_children != 0) {
2595		if ((error = vdev_create(vd, txg, B_FALSE)) != 0) {
2596			spa->spa_pending_vdev = NULL;
2597			return (spa_vdev_exit(spa, vd, txg, error));
2598		}
2599	}
2600
2601	/*
2602	 * We must validate the spares and l2cache devices after checking the
2603	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
2604	 */
2605	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) {
2606		spa->spa_pending_vdev = NULL;
2607		return (spa_vdev_exit(spa, vd, txg, error));
2608	}
2609
2610	spa->spa_pending_vdev = NULL;
2611
2612	/*
2613	 * Transfer each new top-level vdev from vd to rvd.
2614	 */
2615	for (c = 0; c < vd->vdev_children; c++) {
2616		tvd = vd->vdev_child[c];
2617		vdev_remove_child(vd, tvd);
2618		tvd->vdev_id = rvd->vdev_children;
2619		vdev_add_child(rvd, tvd);
2620		vdev_config_dirty(tvd);
2621	}
2622
2623	if (nspares != 0) {
2624		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
2625		    ZPOOL_CONFIG_SPARES);
2626		spa_load_spares(spa);
2627		spa->spa_spares.sav_sync = B_TRUE;
2628	}
2629
2630	if (nl2cache != 0) {
2631		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
2632		    ZPOOL_CONFIG_L2CACHE);
2633		spa_load_l2cache(spa);
2634		spa->spa_l2cache.sav_sync = B_TRUE;
2635	}
2636
2637	/*
2638	 * We have to be careful when adding new vdevs to an existing pool.
2639	 * If other threads start allocating from these vdevs before we
2640	 * sync the config cache, and we lose power, then upon reboot we may
2641	 * fail to open the pool because there are DVAs that the config cache
2642	 * can't translate.  Therefore, we first add the vdevs without
2643	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
2644	 * and then let spa_config_update() initialize the new metaslabs.
2645	 *
2646	 * spa_load() checks for added-but-not-initialized vdevs, so that
2647	 * if we lose power at any point in this sequence, the remaining
2648	 * steps will be completed the next time we load the pool.
2649	 */
2650	(void) spa_vdev_exit(spa, vd, txg, 0);
2651
2652	mutex_enter(&spa_namespace_lock);
2653	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
2654	mutex_exit(&spa_namespace_lock);
2655
2656	return (0);
2657}
2658
2659/*
2660 * Attach a device to a mirror.  The arguments are the path to any device
2661 * in the mirror, and the nvroot for the new device.  If the path specifies
2662 * a device that is not mirrored, we automatically insert the mirror vdev.
2663 *
2664 * If 'replacing' is specified, the new device is intended to replace the
2665 * existing device; in this case the two devices are made into their own
2666 * mirror using the 'replacing' vdev, which is functionally identical to
2667 * the mirror vdev (it actually reuses all the same ops) but has a few
2668 * extra rules: you can't attach to it after it's been created, and upon
2669 * completion of resilvering, the first disk (the one being replaced)
2670 * is automatically detached.
2671 */
2672int
2673spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
2674{
2675	uint64_t txg, open_txg;
2676	int error;
2677	vdev_t *rvd = spa->spa_root_vdev;
2678	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
2679	vdev_ops_t *pvops;
2680	int is_log;
2681
2682	txg = spa_vdev_enter(spa);
2683
2684	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
2685
2686	if (oldvd == NULL)
2687		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
2688
2689	if (!oldvd->vdev_ops->vdev_op_leaf)
2690		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
2691
2692	pvd = oldvd->vdev_parent;
2693
2694	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
2695	    VDEV_ALLOC_ADD)) != 0)
2696		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
2697
2698	if (newrootvd->vdev_children != 1)
2699		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
2700
2701	newvd = newrootvd->vdev_child[0];
2702
2703	if (!newvd->vdev_ops->vdev_op_leaf)
2704		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
2705
2706	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
2707		return (spa_vdev_exit(spa, newrootvd, txg, error));
2708
2709	/*
2710	 * Spares can't replace logs
2711	 */
2712	is_log = oldvd->vdev_islog;
2713	if (is_log && newvd->vdev_isspare)
2714		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2715
2716	if (!replacing) {
2717		/*
2718		 * For attach, the only allowable parent is a mirror or the root
2719		 * vdev.
2720		 */
2721		if (pvd->vdev_ops != &vdev_mirror_ops &&
2722		    pvd->vdev_ops != &vdev_root_ops)
2723			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2724
2725		pvops = &vdev_mirror_ops;
2726	} else {
2727		/*
2728		 * Active hot spares can only be replaced by inactive hot
2729		 * spares.
2730		 */
2731		if (pvd->vdev_ops == &vdev_spare_ops &&
2732		    pvd->vdev_child[1] == oldvd &&
2733		    !spa_has_spare(spa, newvd->vdev_guid))
2734			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2735
2736		/*
2737		 * If the source is a hot spare, and the parent isn't already a
2738		 * spare, then we want to create a new hot spare.  Otherwise, we
2739		 * want to create a replacing vdev.  The user is not allowed to
2740		 * attach to a spared vdev child unless the 'isspare' state is
2741		 * the same (spare replaces spare, non-spare replaces
2742		 * non-spare).
2743		 */
2744		if (pvd->vdev_ops == &vdev_replacing_ops)
2745			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2746		else if (pvd->vdev_ops == &vdev_spare_ops &&
2747		    newvd->vdev_isspare != oldvd->vdev_isspare)
2748			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
2749		else if (pvd->vdev_ops != &vdev_spare_ops &&
2750		    newvd->vdev_isspare)
2751			pvops = &vdev_spare_ops;
2752		else
2753			pvops = &vdev_replacing_ops;
2754	}
2755
2756	/*
2757	 * Compare the new device size with the replaceable/attachable
2758	 * device size.
2759	 */
2760	if (newvd->vdev_psize < vdev_get_rsize(oldvd))
2761		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
2762
2763	/*
2764	 * The new device cannot have a higher alignment requirement
2765	 * than the top-level vdev.
2766	 */
2767	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
2768		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
2769
2770	/*
2771	 * If this is an in-place replacement, update oldvd's path and devid
2772	 * to make it distinguishable from newvd, and unopenable from now on.
2773	 */
2774	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
2775		spa_strfree(oldvd->vdev_path);
2776		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
2777		    KM_SLEEP);
2778		(void) sprintf(oldvd->vdev_path, "%s/%s",
2779		    newvd->vdev_path, "old");
2780		if (oldvd->vdev_devid != NULL) {
2781			spa_strfree(oldvd->vdev_devid);
2782			oldvd->vdev_devid = NULL;
2783		}
2784	}
2785
2786	/*
2787	 * If the parent is not a mirror, or if we're replacing, insert the new
2788	 * mirror/replacing/spare vdev above oldvd.
2789	 */
2790	if (pvd->vdev_ops != pvops)
2791		pvd = vdev_add_parent(oldvd, pvops);
2792
2793	ASSERT(pvd->vdev_top->vdev_parent == rvd);
2794	ASSERT(pvd->vdev_ops == pvops);
2795	ASSERT(oldvd->vdev_parent == pvd);
2796
2797	/*
2798	 * Extract the new device from its root and add it to pvd.
2799	 */
2800	vdev_remove_child(newrootvd, newvd);
2801	newvd->vdev_id = pvd->vdev_children;
2802	vdev_add_child(pvd, newvd);
2803
2804	/*
2805	 * If newvd is smaller than oldvd, but larger than its rsize,
2806	 * the addition of newvd may have decreased our parent's asize.
2807	 */
2808	pvd->vdev_asize = MIN(pvd->vdev_asize, newvd->vdev_asize);
2809
2810	tvd = newvd->vdev_top;
2811	ASSERT(pvd->vdev_top == tvd);
2812	ASSERT(tvd->vdev_parent == rvd);
2813
2814	vdev_config_dirty(tvd);
2815
2816	/*
2817	 * Set newvd's DTL to [TXG_INITIAL, open_txg].  It will propagate
2818	 * upward when spa_vdev_exit() calls vdev_dtl_reassess().
2819	 */
2820	open_txg = txg + TXG_CONCURRENT_STATES - 1;
2821
2822	mutex_enter(&newvd->vdev_dtl_lock);
2823	space_map_add(&newvd->vdev_dtl_map, TXG_INITIAL,
2824	    open_txg - TXG_INITIAL + 1);
2825	mutex_exit(&newvd->vdev_dtl_lock);
2826
2827	if (newvd->vdev_isspare)
2828		spa_spare_activate(newvd);
2829
2830	/*
2831	 * Mark newvd's DTL dirty in this txg.
2832	 */
2833	vdev_dirty(tvd, VDD_DTL, newvd, txg);
2834
2835	(void) spa_vdev_exit(spa, newrootvd, open_txg, 0);
2836
2837	/*
2838	 * Kick off a resilver to update newvd.  We need to grab the namespace
2839	 * lock because spa_scrub() needs to post a sysevent with the pool name.
2840	 */
2841	mutex_enter(&spa_namespace_lock);
2842	VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
2843	mutex_exit(&spa_namespace_lock);
2844
2845	return (0);
2846}
2847
2848/*
2849 * Detach a device from a mirror or replacing vdev.
2850 * If 'replace_done' is specified, only detach if the parent
2851 * is a replacing vdev.
2852 */
2853int
2854spa_vdev_detach(spa_t *spa, uint64_t guid, int replace_done)
2855{
2856	uint64_t txg;
2857	int c, t, error;
2858	vdev_t *rvd = spa->spa_root_vdev;
2859	vdev_t *vd, *pvd, *cvd, *tvd;
2860	boolean_t unspare = B_FALSE;
2861	uint64_t unspare_guid;
2862	size_t len;
2863
2864	txg = spa_vdev_enter(spa);
2865
2866	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
2867
2868	if (vd == NULL)
2869		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
2870
2871	if (!vd->vdev_ops->vdev_op_leaf)
2872		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
2873
2874	pvd = vd->vdev_parent;
2875
2876	/*
2877	 * If replace_done is specified, only remove this device if it's
2878	 * the first child of a replacing vdev.  For the 'spare' vdev, either
2879	 * disk can be removed.
2880	 */
2881	if (replace_done) {
2882		if (pvd->vdev_ops == &vdev_replacing_ops) {
2883			if (vd->vdev_id != 0)
2884				return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
2885		} else if (pvd->vdev_ops != &vdev_spare_ops) {
2886			return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
2887		}
2888	}
2889
2890	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
2891	    spa_version(spa) >= SPA_VERSION_SPARES);
2892
2893	/*
2894	 * Only mirror, replacing, and spare vdevs support detach.
2895	 */
2896	if (pvd->vdev_ops != &vdev_replacing_ops &&
2897	    pvd->vdev_ops != &vdev_mirror_ops &&
2898	    pvd->vdev_ops != &vdev_spare_ops)
2899		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
2900
2901	/*
2902	 * If there's only one replica, you can't detach it.
2903	 */
2904	if (pvd->vdev_children <= 1)
2905		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
2906
2907	/*
2908	 * If all siblings have non-empty DTLs, this device may have the only
2909	 * valid copy of the data, which means we cannot safely detach it.
2910	 *
2911	 * XXX -- as in the vdev_offline() case, we really want a more
2912	 * precise DTL check.
2913	 */
2914	for (c = 0; c < pvd->vdev_children; c++) {
2915		uint64_t dirty;
2916
2917		cvd = pvd->vdev_child[c];
2918		if (cvd == vd)
2919			continue;
2920		if (vdev_is_dead(cvd))
2921			continue;
2922		mutex_enter(&cvd->vdev_dtl_lock);
2923		dirty = cvd->vdev_dtl_map.sm_space |
2924		    cvd->vdev_dtl_scrub.sm_space;
2925		mutex_exit(&cvd->vdev_dtl_lock);
2926		if (!dirty)
2927			break;
2928	}
2929
2930	/*
2931	 * If we are a replacing or spare vdev, then we can always detach the
2932	 * latter child, as that is how one cancels the operation.
2933	 */
2934	if ((pvd->vdev_ops == &vdev_mirror_ops || vd->vdev_id != 1) &&
2935	    c == pvd->vdev_children)
2936		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
2937
2938	/*
2939	 * If we are detaching the second disk from a replacing vdev, then
2940	 * check to see if we changed the original vdev's path to have "/old"
2941	 * at the end in spa_vdev_attach().  If so, undo that change now.
2942	 */
2943	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id == 1 &&
2944	    pvd->vdev_child[0]->vdev_path != NULL &&
2945	    pvd->vdev_child[1]->vdev_path != NULL) {
2946		ASSERT(pvd->vdev_child[1] == vd);
2947		cvd = pvd->vdev_child[0];
2948		len = strlen(vd->vdev_path);
2949		if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
2950		    strcmp(cvd->vdev_path + len, "/old") == 0) {
2951			spa_strfree(cvd->vdev_path);
2952			cvd->vdev_path = spa_strdup(vd->vdev_path);
2953		}
2954	}
2955
2956	/*
2957	 * If we are detaching the original disk from a spare, then it implies
2958	 * that the spare should become a real disk, and be removed from the
2959	 * active spare list for the pool.
2960	 */
2961	if (pvd->vdev_ops == &vdev_spare_ops &&
2962	    vd->vdev_id == 0)
2963		unspare = B_TRUE;
2964
2965	/*
2966	 * Erase the disk labels so the disk can be used for other things.
2967	 * This must be done after all other error cases are handled,
2968	 * but before we disembowel vd (so we can still do I/O to it).
2969	 * But if we can't do it, don't treat the error as fatal --
2970	 * it may be that the unwritability of the disk is the reason
2971	 * it's being detached!
2972	 */
2973	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
2974
2975	/*
2976	 * Remove vd from its parent and compact the parent's children.
2977	 */
2978	vdev_remove_child(pvd, vd);
2979	vdev_compact_children(pvd);
2980
2981	/*
2982	 * Remember one of the remaining children so we can get tvd below.
2983	 */
2984	cvd = pvd->vdev_child[0];
2985
2986	/*
2987	 * If we need to remove the remaining child from the list of hot spares,
2988	 * do it now, marking the vdev as no longer a spare in the process.  We
2989	 * must do this before vdev_remove_parent(), because that can change the
2990	 * GUID if it creates a new toplevel GUID.
2991	 */
2992	if (unspare) {
2993		ASSERT(cvd->vdev_isspare);
2994		spa_spare_remove(cvd);
2995		unspare_guid = cvd->vdev_guid;
2996	}
2997
2998	/*
2999	 * If the parent mirror/replacing vdev only has one child,
3000	 * the parent is no longer needed.  Remove it from the tree.
3001	 */
3002	if (pvd->vdev_children == 1)
3003		vdev_remove_parent(cvd);
3004
3005	/*
3006	 * We don't set tvd until now because the parent we just removed
3007	 * may have been the previous top-level vdev.
3008	 */
3009	tvd = cvd->vdev_top;
3010	ASSERT(tvd->vdev_parent == rvd);
3011
3012	/*
3013	 * Reevaluate the parent vdev state.
3014	 */
3015	vdev_propagate_state(cvd);
3016
3017	/*
3018	 * If the device we just detached was smaller than the others, it may be
3019	 * possible to add metaslabs (i.e. grow the pool).  vdev_metaslab_init()
3020	 * can't fail because the existing metaslabs are already in core, so
3021	 * there's nothing to read from disk.
3022	 */
3023	VERIFY(vdev_metaslab_init(tvd, txg) == 0);
3024
3025	vdev_config_dirty(tvd);
3026
3027	/*
3028	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
3029	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
3030	 * But first make sure we're not on any *other* txg's DTL list, to
3031	 * prevent vd from being accessed after it's freed.
3032	 */
3033	for (t = 0; t < TXG_SIZE; t++)
3034		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
3035	vd->vdev_detached = B_TRUE;
3036	vdev_dirty(tvd, VDD_DTL, vd, txg);
3037
3038	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
3039
3040	error = spa_vdev_exit(spa, vd, txg, 0);
3041
3042	/*
3043	 * If this was the removal of the original device in a hot spare vdev,
3044	 * then we want to go through and remove the device from the hot spare
3045	 * list of every other pool.
3046	 */
3047	if (unspare) {
3048		spa = NULL;
3049		mutex_enter(&spa_namespace_lock);
3050		while ((spa = spa_next(spa)) != NULL) {
3051			if (spa->spa_state != POOL_STATE_ACTIVE)
3052				continue;
3053
3054			(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
3055		}
3056		mutex_exit(&spa_namespace_lock);
3057	}
3058
3059	return (error);
3060}
3061
3062/*
3063 * Remove a spares vdev from the nvlist config.
3064 */
3065static int
3066spa_remove_spares(spa_aux_vdev_t *sav, uint64_t guid, boolean_t unspare,
3067    nvlist_t **spares, int nspares, vdev_t *vd)
3068{
3069	nvlist_t *nv, **newspares;
3070	int i, j;
3071
3072	nv = NULL;
3073	for (i = 0; i < nspares; i++) {
3074		uint64_t theguid;
3075
3076		VERIFY(nvlist_lookup_uint64(spares[i],
3077		    ZPOOL_CONFIG_GUID, &theguid) == 0);
3078		if (theguid == guid) {
3079			nv = spares[i];
3080			break;
3081		}
3082	}
3083
3084	/*
3085	 * Only remove the hot spare if it's not currently in use in this pool.
3086	 */
3087	if (nv == NULL && vd == NULL)
3088		return (ENOENT);
3089
3090	if (nv == NULL && vd != NULL)
3091		return (ENOTSUP);
3092
3093	if (!unspare && nv != NULL && vd != NULL)
3094		return (EBUSY);
3095
3096	if (nspares == 1) {
3097		newspares = NULL;
3098	} else {
3099		newspares = kmem_alloc((nspares - 1) * sizeof (void *),
3100		    KM_SLEEP);
3101		for (i = 0, j = 0; i < nspares; i++) {
3102			if (spares[i] != nv)
3103				VERIFY(nvlist_dup(spares[i],
3104				    &newspares[j++], KM_SLEEP) == 0);
3105		}
3106	}
3107
3108	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_SPARES,
3109	    DATA_TYPE_NVLIST_ARRAY) == 0);
3110	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3111	    ZPOOL_CONFIG_SPARES, newspares, nspares - 1) == 0);
3112	for (i = 0; i < nspares - 1; i++)
3113		nvlist_free(newspares[i]);
3114	kmem_free(newspares, (nspares - 1) * sizeof (void *));
3115
3116	return (0);
3117}
3118
3119/*
3120 * Remove an l2cache vdev from the nvlist config.
3121 */
3122static int
3123spa_remove_l2cache(spa_aux_vdev_t *sav, uint64_t guid, nvlist_t **l2cache,
3124    int nl2cache, vdev_t *vd)
3125{
3126	nvlist_t *nv, **newl2cache;
3127	int i, j;
3128
3129	nv = NULL;
3130	for (i = 0; i < nl2cache; i++) {
3131		uint64_t theguid;
3132
3133		VERIFY(nvlist_lookup_uint64(l2cache[i],
3134		    ZPOOL_CONFIG_GUID, &theguid) == 0);
3135		if (theguid == guid) {
3136			nv = l2cache[i];
3137			break;
3138		}
3139	}
3140
3141	if (vd == NULL) {
3142		for (i = 0; i < nl2cache; i++) {
3143			if (sav->sav_vdevs[i]->vdev_guid == guid) {
3144				vd = sav->sav_vdevs[i];
3145				break;
3146			}
3147		}
3148	}
3149
3150	if (nv == NULL && vd == NULL)
3151		return (ENOENT);
3152
3153	if (nv == NULL && vd != NULL)
3154		return (ENOTSUP);
3155
3156	if (nl2cache == 1) {
3157		newl2cache = NULL;
3158	} else {
3159		newl2cache = kmem_alloc((nl2cache - 1) * sizeof (void *),
3160		    KM_SLEEP);
3161		for (i = 0, j = 0; i < nl2cache; i++) {
3162			if (l2cache[i] != nv)
3163				VERIFY(nvlist_dup(l2cache[i],
3164				    &newl2cache[j++], KM_SLEEP) == 0);
3165		}
3166	}
3167
3168	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
3169	    DATA_TYPE_NVLIST_ARRAY) == 0);
3170	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3171	    ZPOOL_CONFIG_L2CACHE, newl2cache, nl2cache - 1) == 0);
3172	for (i = 0; i < nl2cache - 1; i++)
3173		nvlist_free(newl2cache[i]);
3174	kmem_free(newl2cache, (nl2cache - 1) * sizeof (void *));
3175
3176	return (0);
3177}
3178
3179/*
3180 * Remove a device from the pool.  Currently, this supports removing only hot
3181 * spares and level 2 ARC devices.
3182 */
3183int
3184spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
3185{
3186	vdev_t *vd;
3187	nvlist_t **spares, **l2cache;
3188	uint_t nspares, nl2cache;
3189	int error = 0;
3190
3191	spa_config_enter(spa, RW_WRITER, FTAG);
3192
3193	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3194
3195	if (spa->spa_spares.sav_vdevs != NULL &&
3196	    spa_spare_exists(guid, NULL) &&
3197	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3198	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0) {
3199		if ((error = spa_remove_spares(&spa->spa_spares, guid, unspare,
3200		    spares, nspares, vd)) != 0)
3201			goto out;
3202		spa_load_spares(spa);
3203		spa->spa_spares.sav_sync = B_TRUE;
3204		goto out;
3205	}
3206
3207	if (spa->spa_l2cache.sav_vdevs != NULL &&
3208	    spa_l2cache_exists(guid, NULL) &&
3209	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3210	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0) {
3211		if ((error = spa_remove_l2cache(&spa->spa_l2cache, guid,
3212		    l2cache, nl2cache, vd)) != 0)
3213			goto out;
3214		spa_load_l2cache(spa);
3215		spa->spa_l2cache.sav_sync = B_TRUE;
3216	}
3217
3218out:
3219	spa_config_exit(spa, FTAG);
3220	return (error);
3221}
3222
3223/*
3224 * Find any device that's done replacing, or a vdev marked 'unspare' that's
3225 * current spared, so we can detach it.
3226 */
3227static vdev_t *
3228spa_vdev_resilver_done_hunt(vdev_t *vd)
3229{
3230	vdev_t *newvd, *oldvd;
3231	int c;
3232
3233	for (c = 0; c < vd->vdev_children; c++) {
3234		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
3235		if (oldvd != NULL)
3236			return (oldvd);
3237	}
3238
3239	/*
3240	 * Check for a completed replacement.
3241	 */
3242	if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
3243		oldvd = vd->vdev_child[0];
3244		newvd = vd->vdev_child[1];
3245
3246		mutex_enter(&newvd->vdev_dtl_lock);
3247		if (newvd->vdev_dtl_map.sm_space == 0 &&
3248		    newvd->vdev_dtl_scrub.sm_space == 0) {
3249			mutex_exit(&newvd->vdev_dtl_lock);
3250			return (oldvd);
3251		}
3252		mutex_exit(&newvd->vdev_dtl_lock);
3253	}
3254
3255	/*
3256	 * Check for a completed resilver with the 'unspare' flag set.
3257	 */
3258	if (vd->vdev_ops == &vdev_spare_ops && vd->vdev_children == 2) {
3259		newvd = vd->vdev_child[0];
3260		oldvd = vd->vdev_child[1];
3261
3262		mutex_enter(&newvd->vdev_dtl_lock);
3263		if (newvd->vdev_unspare &&
3264		    newvd->vdev_dtl_map.sm_space == 0 &&
3265		    newvd->vdev_dtl_scrub.sm_space == 0) {
3266			newvd->vdev_unspare = 0;
3267			mutex_exit(&newvd->vdev_dtl_lock);
3268			return (oldvd);
3269		}
3270		mutex_exit(&newvd->vdev_dtl_lock);
3271	}
3272
3273	return (NULL);
3274}
3275
3276static void
3277spa_vdev_resilver_done(spa_t *spa)
3278{
3279	vdev_t *vd;
3280	vdev_t *pvd;
3281	uint64_t guid;
3282	uint64_t pguid = 0;
3283
3284	spa_config_enter(spa, RW_READER, FTAG);
3285
3286	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
3287		guid = vd->vdev_guid;
3288		/*
3289		 * If we have just finished replacing a hot spared device, then
3290		 * we need to detach the parent's first child (the original hot
3291		 * spare) as well.
3292		 */
3293		pvd = vd->vdev_parent;
3294		if (pvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3295		    pvd->vdev_id == 0) {
3296			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
3297			ASSERT(pvd->vdev_parent->vdev_children == 2);
3298			pguid = pvd->vdev_parent->vdev_child[1]->vdev_guid;
3299		}
3300		spa_config_exit(spa, FTAG);
3301		if (spa_vdev_detach(spa, guid, B_TRUE) != 0)
3302			return;
3303		if (pguid != 0 && spa_vdev_detach(spa, pguid, B_TRUE) != 0)
3304			return;
3305		spa_config_enter(spa, RW_READER, FTAG);
3306	}
3307
3308	spa_config_exit(spa, FTAG);
3309}
3310
3311/*
3312 * Update the stored path for this vdev.  Dirty the vdev configuration, relying
3313 * on spa_vdev_enter/exit() to synchronize the labels and cache.
3314 */
3315int
3316spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
3317{
3318	vdev_t *vd;
3319	uint64_t txg;
3320
3321	txg = spa_vdev_enter(spa);
3322
3323	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) {
3324		/*
3325		 * Determine if this is a reference to a hot spare device.  If
3326		 * it is, update the path manually as there is no associated
3327		 * vdev_t that can be synced to disk.
3328		 */
3329		nvlist_t **spares;
3330		uint_t i, nspares;
3331
3332		if (spa->spa_spares.sav_config != NULL) {
3333			VERIFY(nvlist_lookup_nvlist_array(
3334			    spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
3335			    &spares, &nspares) == 0);
3336			for (i = 0; i < nspares; i++) {
3337				uint64_t theguid;
3338				VERIFY(nvlist_lookup_uint64(spares[i],
3339				    ZPOOL_CONFIG_GUID, &theguid) == 0);
3340				if (theguid == guid) {
3341					VERIFY(nvlist_add_string(spares[i],
3342					    ZPOOL_CONFIG_PATH, newpath) == 0);
3343					spa_load_spares(spa);
3344					spa->spa_spares.sav_sync = B_TRUE;
3345					return (spa_vdev_exit(spa, NULL, txg,
3346					    0));
3347				}
3348			}
3349		}
3350
3351		return (spa_vdev_exit(spa, NULL, txg, ENOENT));
3352	}
3353
3354	if (!vd->vdev_ops->vdev_op_leaf)
3355		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3356
3357	spa_strfree(vd->vdev_path);
3358	vd->vdev_path = spa_strdup(newpath);
3359
3360	vdev_config_dirty(vd->vdev_top);
3361
3362	return (spa_vdev_exit(spa, NULL, txg, 0));
3363}
3364
3365/*
3366 * ==========================================================================
3367 * SPA Scrubbing
3368 * ==========================================================================
3369 */
3370
3371static void
3372spa_scrub_io_done(zio_t *zio)
3373{
3374	spa_t *spa = zio->io_spa;
3375
3376	arc_data_buf_free(zio->io_data, zio->io_size);
3377
3378	mutex_enter(&spa->spa_scrub_lock);
3379	if (zio->io_error && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
3380		vdev_t *vd = zio->io_vd ? zio->io_vd : spa->spa_root_vdev;
3381		spa->spa_scrub_errors++;
3382		mutex_enter(&vd->vdev_stat_lock);
3383		vd->vdev_stat.vs_scrub_errors++;
3384		mutex_exit(&vd->vdev_stat_lock);
3385	}
3386
3387	if (--spa->spa_scrub_inflight < spa->spa_scrub_maxinflight)
3388		cv_broadcast(&spa->spa_scrub_io_cv);
3389
3390	ASSERT(spa->spa_scrub_inflight >= 0);
3391
3392	mutex_exit(&spa->spa_scrub_lock);
3393}
3394
3395static void
3396spa_scrub_io_start(spa_t *spa, blkptr_t *bp, int priority, int flags,
3397    zbookmark_t *zb)
3398{
3399	size_t size = BP_GET_LSIZE(bp);
3400	void *data;
3401
3402	mutex_enter(&spa->spa_scrub_lock);
3403	/*
3404	 * Do not give too much work to vdev(s).
3405	 */
3406	while (spa->spa_scrub_inflight >= spa->spa_scrub_maxinflight) {
3407		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
3408	}
3409	spa->spa_scrub_inflight++;
3410	mutex_exit(&spa->spa_scrub_lock);
3411
3412	data = arc_data_buf_alloc(size);
3413
3414	if (zb->zb_level == -1 && BP_GET_TYPE(bp) != DMU_OT_OBJSET)
3415		flags |= ZIO_FLAG_SPECULATIVE;	/* intent log block */
3416
3417	flags |= ZIO_FLAG_SCRUB_THREAD | ZIO_FLAG_CANFAIL;
3418
3419	zio_nowait(zio_read(NULL, spa, bp, data, size,
3420	    spa_scrub_io_done, NULL, priority, flags, zb));
3421}
3422
3423/* ARGSUSED */
3424static int
3425spa_scrub_cb(traverse_blk_cache_t *bc, spa_t *spa, void *a)
3426{
3427	blkptr_t *bp = &bc->bc_blkptr;
3428	vdev_t *vd = spa->spa_root_vdev;
3429	dva_t *dva = bp->blk_dva;
3430	int needs_resilver = B_FALSE;
3431	int d;
3432
3433	if (bc->bc_errno) {
3434		/*
3435		 * We can't scrub this block, but we can continue to scrub
3436		 * the rest of the pool.  Note the error and move along.
3437		 */
3438		mutex_enter(&spa->spa_scrub_lock);
3439		spa->spa_scrub_errors++;
3440		mutex_exit(&spa->spa_scrub_lock);
3441
3442		mutex_enter(&vd->vdev_stat_lock);
3443		vd->vdev_stat.vs_scrub_errors++;
3444		mutex_exit(&vd->vdev_stat_lock);
3445
3446		return (ERESTART);
3447	}
3448
3449	ASSERT(bp->blk_birth < spa->spa_scrub_maxtxg);
3450
3451	for (d = 0; d < BP_GET_NDVAS(bp); d++) {
3452		vd = vdev_lookup_top(spa, DVA_GET_VDEV(&dva[d]));
3453
3454		ASSERT(vd != NULL);
3455
3456		/*
3457		 * Keep track of how much data we've examined so that
3458		 * zpool(1M) status can make useful progress reports.
3459		 */
3460		mutex_enter(&vd->vdev_stat_lock);
3461		vd->vdev_stat.vs_scrub_examined += DVA_GET_ASIZE(&dva[d]);
3462		mutex_exit(&vd->vdev_stat_lock);
3463
3464		if (spa->spa_scrub_type == POOL_SCRUB_RESILVER) {
3465			if (DVA_GET_GANG(&dva[d])) {
3466				/*
3467				 * Gang members may be spread across multiple
3468				 * vdevs, so the best we can do is look at the
3469				 * pool-wide DTL.
3470				 * XXX -- it would be better to change our
3471				 * allocation policy to ensure that this can't
3472				 * happen.
3473				 */
3474				vd = spa->spa_root_vdev;
3475			}
3476			if (vdev_dtl_contains(&vd->vdev_dtl_map,
3477			    bp->blk_birth, 1))
3478				needs_resilver = B_TRUE;
3479		}
3480	}
3481
3482	if (spa->spa_scrub_type == POOL_SCRUB_EVERYTHING)
3483		spa_scrub_io_start(spa, bp, ZIO_PRIORITY_SCRUB,
3484		    ZIO_FLAG_SCRUB, &bc->bc_bookmark);
3485	else if (needs_resilver)
3486		spa_scrub_io_start(spa, bp, ZIO_PRIORITY_RESILVER,
3487		    ZIO_FLAG_RESILVER, &bc->bc_bookmark);
3488
3489	return (0);
3490}
3491
3492static void
3493spa_scrub_thread(spa_t *spa)
3494{
3495	callb_cpr_t cprinfo;
3496	traverse_handle_t *th = spa->spa_scrub_th;
3497	vdev_t *rvd = spa->spa_root_vdev;
3498	pool_scrub_type_t scrub_type = spa->spa_scrub_type;
3499	int error = 0;
3500	boolean_t complete;
3501
3502	CALLB_CPR_INIT(&cprinfo, &spa->spa_scrub_lock, callb_generic_cpr, FTAG);
3503
3504	/*
3505	 * If we're restarting due to a snapshot create/delete,
3506	 * wait for that to complete.
3507	 */
3508	txg_wait_synced(spa_get_dsl(spa), 0);
3509
3510	dprintf("start %s mintxg=%llu maxtxg=%llu\n",
3511	    scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub",
3512	    spa->spa_scrub_mintxg, spa->spa_scrub_maxtxg);
3513
3514	spa_config_enter(spa, RW_WRITER, FTAG);
3515	vdev_reopen(rvd);		/* purge all vdev caches */
3516	vdev_config_dirty(rvd);		/* rewrite all disk labels */
3517	vdev_scrub_stat_update(rvd, scrub_type, B_FALSE);
3518	spa_config_exit(spa, FTAG);
3519
3520	mutex_enter(&spa->spa_scrub_lock);
3521	spa->spa_scrub_errors = 0;
3522	spa->spa_scrub_active = 1;
3523	ASSERT(spa->spa_scrub_inflight == 0);
3524
3525	while (!spa->spa_scrub_stop) {
3526		CALLB_CPR_SAFE_BEGIN(&cprinfo);
3527		while (spa->spa_scrub_suspended) {
3528			spa->spa_scrub_active = 0;
3529			cv_broadcast(&spa->spa_scrub_cv);
3530			cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
3531			spa->spa_scrub_active = 1;
3532		}
3533		CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_scrub_lock);
3534
3535		if (spa->spa_scrub_restart_txg != 0)
3536			break;
3537
3538		mutex_exit(&spa->spa_scrub_lock);
3539		error = traverse_more(th);
3540		mutex_enter(&spa->spa_scrub_lock);
3541		if (error != EAGAIN)
3542			break;
3543	}
3544
3545	while (spa->spa_scrub_inflight)
3546		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
3547
3548	spa->spa_scrub_active = 0;
3549	cv_broadcast(&spa->spa_scrub_cv);
3550
3551	mutex_exit(&spa->spa_scrub_lock);
3552
3553	spa_config_enter(spa, RW_WRITER, FTAG);
3554
3555	mutex_enter(&spa->spa_scrub_lock);
3556
3557	/*
3558	 * Note: we check spa_scrub_restart_txg under both spa_scrub_lock
3559	 * AND the spa config lock to synchronize with any config changes
3560	 * that revise the DTLs under spa_vdev_enter() / spa_vdev_exit().
3561	 */
3562	if (spa->spa_scrub_restart_txg != 0)
3563		error = ERESTART;
3564
3565	if (spa->spa_scrub_stop)
3566		error = EINTR;
3567
3568	/*
3569	 * Even if there were uncorrectable errors, we consider the scrub
3570	 * completed.  The downside is that if there is a transient error during
3571	 * a resilver, we won't resilver the data properly to the target.  But
3572	 * if the damage is permanent (more likely) we will resilver forever,
3573	 * which isn't really acceptable.  Since there is enough information for
3574	 * the user to know what has failed and why, this seems like a more
3575	 * tractable approach.
3576	 */
3577	complete = (error == 0);
3578
3579	dprintf("end %s to maxtxg=%llu %s, traverse=%d, %llu errors, stop=%u\n",
3580	    scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub",
3581	    spa->spa_scrub_maxtxg, complete ? "done" : "FAILED",
3582	    error, spa->spa_scrub_errors, spa->spa_scrub_stop);
3583
3584	mutex_exit(&spa->spa_scrub_lock);
3585
3586	/*
3587	 * If the scrub/resilver completed, update all DTLs to reflect this.
3588	 * Whether it succeeded or not, vacate all temporary scrub DTLs.
3589	 */
3590	vdev_dtl_reassess(rvd, spa_last_synced_txg(spa) + 1,
3591	    complete ? spa->spa_scrub_maxtxg : 0, B_TRUE);
3592	vdev_scrub_stat_update(rvd, POOL_SCRUB_NONE, complete);
3593	spa_errlog_rotate(spa);
3594
3595	if (scrub_type == POOL_SCRUB_RESILVER && complete)
3596		spa_event_notify(spa, NULL, ESC_ZFS_RESILVER_FINISH);
3597
3598	spa_config_exit(spa, FTAG);
3599
3600	mutex_enter(&spa->spa_scrub_lock);
3601
3602	/*
3603	 * We may have finished replacing a device.
3604	 * Let the async thread assess this and handle the detach.
3605	 */
3606	spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
3607
3608	/*
3609	 * If we were told to restart, our final act is to start a new scrub.
3610	 */
3611	if (error == ERESTART)
3612		spa_async_request(spa, scrub_type == POOL_SCRUB_RESILVER ?
3613		    SPA_ASYNC_RESILVER : SPA_ASYNC_SCRUB);
3614
3615	spa->spa_scrub_type = POOL_SCRUB_NONE;
3616	spa->spa_scrub_active = 0;
3617	spa->spa_scrub_thread = NULL;
3618	cv_broadcast(&spa->spa_scrub_cv);
3619	CALLB_CPR_EXIT(&cprinfo);	/* drops &spa->spa_scrub_lock */
3620	thread_exit();
3621}
3622
3623void
3624spa_scrub_suspend(spa_t *spa)
3625{
3626	mutex_enter(&spa->spa_scrub_lock);
3627	spa->spa_scrub_suspended++;
3628	while (spa->spa_scrub_active) {
3629		cv_broadcast(&spa->spa_scrub_cv);
3630		cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
3631	}
3632	while (spa->spa_scrub_inflight)
3633		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
3634	mutex_exit(&spa->spa_scrub_lock);
3635}
3636
3637void
3638spa_scrub_resume(spa_t *spa)
3639{
3640	mutex_enter(&spa->spa_scrub_lock);
3641	ASSERT(spa->spa_scrub_suspended != 0);
3642	if (--spa->spa_scrub_suspended == 0)
3643		cv_broadcast(&spa->spa_scrub_cv);
3644	mutex_exit(&spa->spa_scrub_lock);
3645}
3646
3647void
3648spa_scrub_restart(spa_t *spa, uint64_t txg)
3649{
3650	/*
3651	 * Something happened (e.g. snapshot create/delete) that means
3652	 * we must restart any in-progress scrubs.  The itinerary will
3653	 * fix this properly.
3654	 */
3655	mutex_enter(&spa->spa_scrub_lock);
3656	spa->spa_scrub_restart_txg = txg;
3657	mutex_exit(&spa->spa_scrub_lock);
3658}
3659
3660int
3661spa_scrub(spa_t *spa, pool_scrub_type_t type, boolean_t force)
3662{
3663	space_seg_t *ss;
3664	uint64_t mintxg, maxtxg;
3665	vdev_t *rvd = spa->spa_root_vdev;
3666
3667	ASSERT(MUTEX_HELD(&spa_namespace_lock));
3668	ASSERT(!spa_config_held(spa, RW_WRITER));
3669
3670	if ((uint_t)type >= POOL_SCRUB_TYPES)
3671		return (ENOTSUP);
3672
3673	mutex_enter(&spa->spa_scrub_lock);
3674
3675	/*
3676	 * If there's a scrub or resilver already in progress, stop it.
3677	 */
3678	while (spa->spa_scrub_thread != NULL) {
3679		/*
3680		 * Don't stop a resilver unless forced.
3681		 */
3682		if (spa->spa_scrub_type == POOL_SCRUB_RESILVER && !force) {
3683			mutex_exit(&spa->spa_scrub_lock);
3684			return (EBUSY);
3685		}
3686		spa->spa_scrub_stop = 1;
3687		cv_broadcast(&spa->spa_scrub_cv);
3688		cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock);
3689	}
3690
3691	/*
3692	 * Terminate the previous traverse.
3693	 */
3694	if (spa->spa_scrub_th != NULL) {
3695		traverse_fini(spa->spa_scrub_th);
3696		spa->spa_scrub_th = NULL;
3697	}
3698
3699	if (rvd == NULL) {
3700		ASSERT(spa->spa_scrub_stop == 0);
3701		ASSERT(spa->spa_scrub_type == type);
3702		ASSERT(spa->spa_scrub_restart_txg == 0);
3703		mutex_exit(&spa->spa_scrub_lock);
3704		return (0);
3705	}
3706
3707	mintxg = TXG_INITIAL - 1;
3708	maxtxg = spa_last_synced_txg(spa) + 1;
3709
3710	mutex_enter(&rvd->vdev_dtl_lock);
3711
3712	if (rvd->vdev_dtl_map.sm_space == 0) {
3713		/*
3714		 * The pool-wide DTL is empty.
3715		 * If this is a resilver, there's nothing to do except
3716		 * check whether any in-progress replacements have completed.
3717		 */
3718		if (type == POOL_SCRUB_RESILVER) {
3719			type = POOL_SCRUB_NONE;
3720			spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
3721		}
3722	} else {
3723		/*
3724		 * The pool-wide DTL is non-empty.
3725		 * If this is a normal scrub, upgrade to a resilver instead.
3726		 */
3727		if (type == POOL_SCRUB_EVERYTHING)
3728			type = POOL_SCRUB_RESILVER;
3729	}
3730
3731	if (type == POOL_SCRUB_RESILVER) {
3732		/*
3733		 * Determine the resilvering boundaries.
3734		 *
3735		 * Note: (mintxg, maxtxg) is an open interval,
3736		 * i.e. mintxg and maxtxg themselves are not included.
3737		 *
3738		 * Note: for maxtxg, we MIN with spa_last_synced_txg(spa) + 1
3739		 * so we don't claim to resilver a txg that's still changing.
3740		 */
3741		ss = avl_first(&rvd->vdev_dtl_map.sm_root);
3742		mintxg = ss->ss_start - 1;
3743		ss = avl_last(&rvd->vdev_dtl_map.sm_root);
3744		maxtxg = MIN(ss->ss_end, maxtxg);
3745
3746		spa_event_notify(spa, NULL, ESC_ZFS_RESILVER_START);
3747	}
3748
3749	mutex_exit(&rvd->vdev_dtl_lock);
3750
3751	spa->spa_scrub_stop = 0;
3752	spa->spa_scrub_type = type;
3753	spa->spa_scrub_restart_txg = 0;
3754
3755	if (type != POOL_SCRUB_NONE) {
3756		spa->spa_scrub_mintxg = mintxg;
3757		spa->spa_scrub_maxtxg = maxtxg;
3758		spa->spa_scrub_th = traverse_init(spa, spa_scrub_cb, NULL,
3759		    ADVANCE_PRE | ADVANCE_PRUNE | ADVANCE_ZIL,
3760		    ZIO_FLAG_CANFAIL);
3761		traverse_add_pool(spa->spa_scrub_th, mintxg, maxtxg);
3762		spa->spa_scrub_thread = thread_create(NULL, 0,
3763		    spa_scrub_thread, spa, 0, &p0, TS_RUN, minclsyspri);
3764	}
3765
3766	mutex_exit(&spa->spa_scrub_lock);
3767
3768	return (0);
3769}
3770
3771/*
3772 * ==========================================================================
3773 * SPA async task processing
3774 * ==========================================================================
3775 */
3776
3777static void
3778spa_async_remove(spa_t *spa, vdev_t *vd)
3779{
3780	vdev_t *tvd;
3781	int c;
3782
3783	for (c = 0; c < vd->vdev_children; c++) {
3784		tvd = vd->vdev_child[c];
3785		if (tvd->vdev_remove_wanted) {
3786			tvd->vdev_remove_wanted = 0;
3787			vdev_set_state(tvd, B_FALSE, VDEV_STATE_REMOVED,
3788			    VDEV_AUX_NONE);
3789			vdev_clear(spa, tvd, B_TRUE);
3790			vdev_config_dirty(tvd->vdev_top);
3791		}
3792		spa_async_remove(spa, tvd);
3793	}
3794}
3795
3796static void
3797spa_async_thread(spa_t *spa)
3798{
3799	int tasks;
3800	uint64_t txg;
3801
3802	ASSERT(spa->spa_sync_on);
3803
3804	mutex_enter(&spa->spa_async_lock);
3805	tasks = spa->spa_async_tasks;
3806	spa->spa_async_tasks = 0;
3807	mutex_exit(&spa->spa_async_lock);
3808
3809	/*
3810	 * See if the config needs to be updated.
3811	 */
3812	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
3813		mutex_enter(&spa_namespace_lock);
3814		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3815		mutex_exit(&spa_namespace_lock);
3816	}
3817
3818	/*
3819	 * See if any devices need to be marked REMOVED.
3820	 *
3821	 * XXX - We avoid doing this when we are in
3822	 * I/O failure state since spa_vdev_enter() grabs
3823	 * the namespace lock and would not be able to obtain
3824	 * the writer config lock.
3825	 */
3826	if (tasks & SPA_ASYNC_REMOVE &&
3827	    spa_state(spa) != POOL_STATE_IO_FAILURE) {
3828		txg = spa_vdev_enter(spa);
3829		spa_async_remove(spa, spa->spa_root_vdev);
3830		(void) spa_vdev_exit(spa, NULL, txg, 0);
3831	}
3832
3833	/*
3834	 * If any devices are done replacing, detach them.
3835	 */
3836	if (tasks & SPA_ASYNC_RESILVER_DONE)
3837		spa_vdev_resilver_done(spa);
3838
3839	/*
3840	 * Kick off a scrub.  When starting a RESILVER scrub (or an EVERYTHING
3841	 * scrub which can become a resilver), we need to hold
3842	 * spa_namespace_lock() because the sysevent we post via
3843	 * spa_event_notify() needs to get the name of the pool.
3844	 */
3845	if (tasks & SPA_ASYNC_SCRUB) {
3846		mutex_enter(&spa_namespace_lock);
3847		VERIFY(spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_TRUE) == 0);
3848		mutex_exit(&spa_namespace_lock);
3849	}
3850
3851	/*
3852	 * Kick off a resilver.
3853	 */
3854	if (tasks & SPA_ASYNC_RESILVER) {
3855		mutex_enter(&spa_namespace_lock);
3856		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0);
3857		mutex_exit(&spa_namespace_lock);
3858	}
3859
3860	/*
3861	 * Let the world know that we're done.
3862	 */
3863	mutex_enter(&spa->spa_async_lock);
3864	spa->spa_async_thread = NULL;
3865	cv_broadcast(&spa->spa_async_cv);
3866	mutex_exit(&spa->spa_async_lock);
3867	thread_exit();
3868}
3869
3870void
3871spa_async_suspend(spa_t *spa)
3872{
3873	mutex_enter(&spa->spa_async_lock);
3874	spa->spa_async_suspended++;
3875	while (spa->spa_async_thread != NULL)
3876		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
3877	mutex_exit(&spa->spa_async_lock);
3878}
3879
3880void
3881spa_async_resume(spa_t *spa)
3882{
3883	mutex_enter(&spa->spa_async_lock);
3884	ASSERT(spa->spa_async_suspended != 0);
3885	spa->spa_async_suspended--;
3886	mutex_exit(&spa->spa_async_lock);
3887}
3888
3889static void
3890spa_async_dispatch(spa_t *spa)
3891{
3892	mutex_enter(&spa->spa_async_lock);
3893	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
3894	    spa->spa_async_thread == NULL &&
3895	    rootdir != NULL && !vn_is_readonly(rootdir))
3896		spa->spa_async_thread = thread_create(NULL, 0,
3897		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
3898	mutex_exit(&spa->spa_async_lock);
3899}
3900
3901void
3902spa_async_request(spa_t *spa, int task)
3903{
3904	mutex_enter(&spa->spa_async_lock);
3905	spa->spa_async_tasks |= task;
3906	mutex_exit(&spa->spa_async_lock);
3907}
3908
3909/*
3910 * ==========================================================================
3911 * SPA syncing routines
3912 * ==========================================================================
3913 */
3914
3915static void
3916spa_sync_deferred_frees(spa_t *spa, uint64_t txg)
3917{
3918	bplist_t *bpl = &spa->spa_sync_bplist;
3919	dmu_tx_t *tx;
3920	blkptr_t blk;
3921	uint64_t itor = 0;
3922	zio_t *zio;
3923	int error;
3924	uint8_t c = 1;
3925
3926	zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CONFIG_HELD);
3927
3928	while (bplist_iterate(bpl, &itor, &blk) == 0)
3929		zio_nowait(zio_free(zio, spa, txg, &blk, NULL, NULL));
3930
3931	error = zio_wait(zio);
3932	ASSERT3U(error, ==, 0);
3933
3934	tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3935	bplist_vacate(bpl, tx);
3936
3937	/*
3938	 * Pre-dirty the first block so we sync to convergence faster.
3939	 * (Usually only the first block is needed.)
3940	 */
3941	dmu_write(spa->spa_meta_objset, spa->spa_sync_bplist_obj, 0, 1, &c, tx);
3942	dmu_tx_commit(tx);
3943}
3944
3945static void
3946spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
3947{
3948	char *packed = NULL;
3949	size_t nvsize = 0;
3950	dmu_buf_t *db;
3951
3952	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
3953
3954	packed = kmem_alloc(nvsize, KM_SLEEP);
3955
3956	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
3957	    KM_SLEEP) == 0);
3958
3959	dmu_write(spa->spa_meta_objset, obj, 0, nvsize, packed, tx);
3960
3961	kmem_free(packed, nvsize);
3962
3963	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
3964	dmu_buf_will_dirty(db, tx);
3965	*(uint64_t *)db->db_data = nvsize;
3966	dmu_buf_rele(db, FTAG);
3967}
3968
3969static void
3970spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
3971    const char *config, const char *entry)
3972{
3973	nvlist_t *nvroot;
3974	nvlist_t **list;
3975	int i;
3976
3977	if (!sav->sav_sync)
3978		return;
3979
3980	/*
3981	 * Update the MOS nvlist describing the list of available devices.
3982	 * spa_validate_aux() will have already made sure this nvlist is
3983	 * valid and the vdevs are labeled appropriately.
3984	 */
3985	if (sav->sav_object == 0) {
3986		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
3987		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
3988		    sizeof (uint64_t), tx);
3989		VERIFY(zap_update(spa->spa_meta_objset,
3990		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
3991		    &sav->sav_object, tx) == 0);
3992	}
3993
3994	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3995	if (sav->sav_count == 0) {
3996		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
3997	} else {
3998		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
3999		for (i = 0; i < sav->sav_count; i++)
4000			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
4001			    B_FALSE, B_FALSE, B_TRUE);
4002		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
4003		    sav->sav_count) == 0);
4004		for (i = 0; i < sav->sav_count; i++)
4005			nvlist_free(list[i]);
4006		kmem_free(list, sav->sav_count * sizeof (void *));
4007	}
4008
4009	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
4010	nvlist_free(nvroot);
4011
4012	sav->sav_sync = B_FALSE;
4013}
4014
4015static void
4016spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
4017{
4018	nvlist_t *config;
4019
4020	if (list_is_empty(&spa->spa_dirty_list))
4021		return;
4022
4023	config = spa_config_generate(spa, NULL, dmu_tx_get_txg(tx), B_FALSE);
4024
4025	if (spa->spa_config_syncing)
4026		nvlist_free(spa->spa_config_syncing);
4027	spa->spa_config_syncing = config;
4028
4029	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
4030}
4031
4032/*
4033 * Set zpool properties.
4034 */
4035static void
4036spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
4037{
4038	spa_t *spa = arg1;
4039	objset_t *mos = spa->spa_meta_objset;
4040	nvlist_t *nvp = arg2;
4041	nvpair_t *elem;
4042	uint64_t intval;
4043	char *strval;
4044	zpool_prop_t prop;
4045	const char *propname;
4046	zprop_type_t proptype;
4047	spa_config_dirent_t *dp;
4048
4049	elem = NULL;
4050	while ((elem = nvlist_next_nvpair(nvp, elem))) {
4051		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
4052		case ZPOOL_PROP_VERSION:
4053			/*
4054			 * Only set version for non-zpool-creation cases
4055			 * (set/import). spa_create() needs special care
4056			 * for version setting.
4057			 */
4058			if (tx->tx_txg != TXG_INITIAL) {
4059				VERIFY(nvpair_value_uint64(elem,
4060				    &intval) == 0);
4061				ASSERT(intval <= SPA_VERSION);
4062				ASSERT(intval >= spa_version(spa));
4063				spa->spa_uberblock.ub_version = intval;
4064				vdev_config_dirty(spa->spa_root_vdev);
4065			}
4066			break;
4067
4068		case ZPOOL_PROP_ALTROOT:
4069			/*
4070			 * 'altroot' is a non-persistent property. It should
4071			 * have been set temporarily at creation or import time.
4072			 */
4073			ASSERT(spa->spa_root != NULL);
4074			break;
4075
4076		case ZPOOL_PROP_CACHEFILE:
4077			/*
4078			 * 'cachefile' is a non-persistent property, but note
4079			 * an async request that the config cache needs to be
4080			 * udpated.
4081			 */
4082			VERIFY(nvpair_value_string(elem, &strval) == 0);
4083
4084			dp = kmem_alloc(sizeof (spa_config_dirent_t),
4085			    KM_SLEEP);
4086
4087			if (strval[0] == '\0')
4088				dp->scd_path = spa_strdup(spa_config_path);
4089			else if (strcmp(strval, "none") == 0)
4090				dp->scd_path = NULL;
4091			else
4092				dp->scd_path = spa_strdup(strval);
4093
4094			list_insert_head(&spa->spa_config_list, dp);
4095			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
4096			break;
4097		default:
4098			/*
4099			 * Set pool property values in the poolprops mos object.
4100			 */
4101			mutex_enter(&spa->spa_props_lock);
4102			if (spa->spa_pool_props_object == 0) {
4103				objset_t *mos = spa->spa_meta_objset;
4104
4105				VERIFY((spa->spa_pool_props_object =
4106				    zap_create(mos, DMU_OT_POOL_PROPS,
4107				    DMU_OT_NONE, 0, tx)) > 0);
4108
4109				VERIFY(zap_update(mos,
4110				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
4111				    8, 1, &spa->spa_pool_props_object, tx)
4112				    == 0);
4113			}
4114			mutex_exit(&spa->spa_props_lock);
4115
4116			/* normalize the property name */
4117			propname = zpool_prop_to_name(prop);
4118			proptype = zpool_prop_get_type(prop);
4119
4120			if (nvpair_type(elem) == DATA_TYPE_STRING) {
4121				ASSERT(proptype == PROP_TYPE_STRING);
4122				VERIFY(nvpair_value_string(elem, &strval) == 0);
4123				VERIFY(zap_update(mos,
4124				    spa->spa_pool_props_object, propname,
4125				    1, strlen(strval) + 1, strval, tx) == 0);
4126
4127			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
4128				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
4129
4130				if (proptype == PROP_TYPE_INDEX) {
4131					const char *unused;
4132					VERIFY(zpool_prop_index_to_string(
4133					    prop, intval, &unused) == 0);
4134				}
4135				VERIFY(zap_update(mos,
4136				    spa->spa_pool_props_object, propname,
4137				    8, 1, &intval, tx) == 0);
4138			} else {
4139				ASSERT(0); /* not allowed */
4140			}
4141
4142			switch (prop) {
4143			case ZPOOL_PROP_DELEGATION:
4144				spa->spa_delegation = intval;
4145				break;
4146			case ZPOOL_PROP_BOOTFS:
4147				spa->spa_bootfs = intval;
4148				break;
4149			case ZPOOL_PROP_FAILUREMODE:
4150				spa->spa_failmode = intval;
4151				break;
4152			default:
4153				break;
4154			}
4155		}
4156
4157		/* log internal history if this is not a zpool create */
4158		if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
4159		    tx->tx_txg != TXG_INITIAL) {
4160			spa_history_internal_log(LOG_POOL_PROPSET,
4161			    spa, tx, cr, "%s %lld %s",
4162			    nvpair_name(elem), intval, spa->spa_name);
4163		}
4164	}
4165}
4166
4167/*
4168 * Sync the specified transaction group.  New blocks may be dirtied as
4169 * part of the process, so we iterate until it converges.
4170 */
4171void
4172spa_sync(spa_t *spa, uint64_t txg)
4173{
4174	dsl_pool_t *dp = spa->spa_dsl_pool;
4175	objset_t *mos = spa->spa_meta_objset;
4176	bplist_t *bpl = &spa->spa_sync_bplist;
4177	vdev_t *rvd = spa->spa_root_vdev;
4178	vdev_t *vd;
4179	dmu_tx_t *tx;
4180	int dirty_vdevs;
4181
4182	/*
4183	 * Lock out configuration changes.
4184	 */
4185	spa_config_enter(spa, RW_READER, FTAG);
4186
4187	spa->spa_syncing_txg = txg;
4188	spa->spa_sync_pass = 0;
4189
4190	VERIFY(0 == bplist_open(bpl, mos, spa->spa_sync_bplist_obj));
4191
4192	tx = dmu_tx_create_assigned(dp, txg);
4193
4194	/*
4195	 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
4196	 * set spa_deflate if we have no raid-z vdevs.
4197	 */
4198	if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
4199	    spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
4200		int i;
4201
4202		for (i = 0; i < rvd->vdev_children; i++) {
4203			vd = rvd->vdev_child[i];
4204			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
4205				break;
4206		}
4207		if (i == rvd->vdev_children) {
4208			spa->spa_deflate = TRUE;
4209			VERIFY(0 == zap_add(spa->spa_meta_objset,
4210			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
4211			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
4212		}
4213	}
4214
4215	/*
4216	 * If anything has changed in this txg, push the deferred frees
4217	 * from the previous txg.  If not, leave them alone so that we
4218	 * don't generate work on an otherwise idle system.
4219	 */
4220	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
4221	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
4222	    !txg_list_empty(&dp->dp_sync_tasks, txg))
4223		spa_sync_deferred_frees(spa, txg);
4224
4225	/*
4226	 * Iterate to convergence.
4227	 */
4228	do {
4229		spa->spa_sync_pass++;
4230
4231		spa_sync_config_object(spa, tx);
4232		spa_sync_aux_dev(spa, &spa->spa_spares, tx,
4233		    ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
4234		spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
4235		    ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
4236		spa_errlog_sync(spa, txg);
4237		dsl_pool_sync(dp, txg);
4238
4239		dirty_vdevs = 0;
4240		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) {
4241			vdev_sync(vd, txg);
4242			dirty_vdevs++;
4243		}
4244
4245		bplist_sync(bpl, tx);
4246	} while (dirty_vdevs);
4247
4248	bplist_close(bpl);
4249
4250	dprintf("txg %llu passes %d\n", txg, spa->spa_sync_pass);
4251
4252	/*
4253	 * Rewrite the vdev configuration (which includes the uberblock)
4254	 * to commit the transaction group.
4255	 *
4256	 * If there are no dirty vdevs, we sync the uberblock to a few
4257	 * random top-level vdevs that are known to be visible in the
4258	 * config cache (see spa_vdev_add() for details).  If there *are*
4259	 * dirty vdevs -- or if the sync to our random subset fails --
4260	 * then sync the uberblock to all vdevs.
4261	 */
4262	if (list_is_empty(&spa->spa_dirty_list)) {
4263		vdev_t *svd[SPA_DVAS_PER_BP];
4264		int svdcount = 0;
4265		int children = rvd->vdev_children;
4266		int c0 = spa_get_random(children);
4267		int c;
4268
4269		for (c = 0; c < children; c++) {
4270			vd = rvd->vdev_child[(c0 + c) % children];
4271			if (vd->vdev_ms_array == 0 || vd->vdev_islog)
4272				continue;
4273			svd[svdcount++] = vd;
4274			if (svdcount == SPA_DVAS_PER_BP)
4275				break;
4276		}
4277		vdev_config_sync(svd, svdcount, txg);
4278	} else {
4279		vdev_config_sync(rvd->vdev_child, rvd->vdev_children, txg);
4280	}
4281	dmu_tx_commit(tx);
4282
4283	/*
4284	 * Clear the dirty config list.
4285	 */
4286	while ((vd = list_head(&spa->spa_dirty_list)) != NULL)
4287		vdev_config_clean(vd);
4288
4289	/*
4290	 * Now that the new config has synced transactionally,
4291	 * let it become visible to the config cache.
4292	 */
4293	if (spa->spa_config_syncing != NULL) {
4294		spa_config_set(spa, spa->spa_config_syncing);
4295		spa->spa_config_txg = txg;
4296		spa->spa_config_syncing = NULL;
4297	}
4298
4299	/*
4300	 * Make a stable copy of the fully synced uberblock.
4301	 * We use this as the root for pool traversals.
4302	 */
4303	spa->spa_traverse_wanted = 1;	/* tells traverse_more() to stop */
4304
4305	spa_scrub_suspend(spa);		/* stop scrubbing and finish I/Os */
4306
4307	rw_enter(&spa->spa_traverse_lock, RW_WRITER);
4308	spa->spa_traverse_wanted = 0;
4309	spa->spa_ubsync = spa->spa_uberblock;
4310	rw_exit(&spa->spa_traverse_lock);
4311
4312	spa_scrub_resume(spa);		/* resume scrub with new ubsync */
4313
4314	/*
4315	 * Clean up the ZIL records for the synced txg.
4316	 */
4317	dsl_pool_zil_clean(dp);
4318
4319	/*
4320	 * Update usable space statistics.
4321	 */
4322	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
4323		vdev_sync_done(vd, txg);
4324
4325	/*
4326	 * It had better be the case that we didn't dirty anything
4327	 * since vdev_config_sync().
4328	 */
4329	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
4330	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
4331	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
4332	ASSERT(bpl->bpl_queue == NULL);
4333
4334	spa_config_exit(spa, FTAG);
4335
4336	/*
4337	 * If any async tasks have been requested, kick them off.
4338	 */
4339	spa_async_dispatch(spa);
4340}
4341
4342/*
4343 * Sync all pools.  We don't want to hold the namespace lock across these
4344 * operations, so we take a reference on the spa_t and drop the lock during the
4345 * sync.
4346 */
4347void
4348spa_sync_allpools(void)
4349{
4350	spa_t *spa = NULL;
4351	mutex_enter(&spa_namespace_lock);
4352	while ((spa = spa_next(spa)) != NULL) {
4353		if (spa_state(spa) != POOL_STATE_ACTIVE)
4354			continue;
4355		spa_open_ref(spa, FTAG);
4356		mutex_exit(&spa_namespace_lock);
4357		txg_wait_synced(spa_get_dsl(spa), 0);
4358		mutex_enter(&spa_namespace_lock);
4359		spa_close(spa, FTAG);
4360	}
4361	mutex_exit(&spa_namespace_lock);
4362}
4363
4364/*
4365 * ==========================================================================
4366 * Miscellaneous routines
4367 * ==========================================================================
4368 */
4369
4370/*
4371 * Remove all pools in the system.
4372 */
4373void
4374spa_evict_all(void)
4375{
4376	spa_t *spa;
4377
4378	/*
4379	 * Remove all cached state.  All pools should be closed now,
4380	 * so every spa in the AVL tree should be unreferenced.
4381	 */
4382	mutex_enter(&spa_namespace_lock);
4383	while ((spa = spa_next(NULL)) != NULL) {
4384		/*
4385		 * Stop async tasks.  The async thread may need to detach
4386		 * a device that's been replaced, which requires grabbing
4387		 * spa_namespace_lock, so we must drop it here.
4388		 */
4389		spa_open_ref(spa, FTAG);
4390		mutex_exit(&spa_namespace_lock);
4391		spa_async_suspend(spa);
4392		mutex_enter(&spa_namespace_lock);
4393		VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0);
4394		spa_close(spa, FTAG);
4395
4396		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4397			spa_unload(spa);
4398			spa_deactivate(spa);
4399		}
4400		spa_remove(spa);
4401	}
4402	mutex_exit(&spa_namespace_lock);
4403}
4404
4405vdev_t *
4406spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t l2cache)
4407{
4408	vdev_t *vd;
4409	int i;
4410
4411	if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
4412		return (vd);
4413
4414	if (l2cache) {
4415		for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
4416			vd = spa->spa_l2cache.sav_vdevs[i];
4417			if (vd->vdev_guid == guid)
4418				return (vd);
4419		}
4420	}
4421
4422	return (NULL);
4423}
4424
4425void
4426spa_upgrade(spa_t *spa, uint64_t version)
4427{
4428	spa_config_enter(spa, RW_WRITER, FTAG);
4429
4430	/*
4431	 * This should only be called for a non-faulted pool, and since a
4432	 * future version would result in an unopenable pool, this shouldn't be
4433	 * possible.
4434	 */
4435	ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
4436	ASSERT(version >= spa->spa_uberblock.ub_version);
4437
4438	spa->spa_uberblock.ub_version = version;
4439	vdev_config_dirty(spa->spa_root_vdev);
4440
4441	spa_config_exit(spa, FTAG);
4442
4443	txg_wait_synced(spa_get_dsl(spa), 0);
4444}
4445
4446boolean_t
4447spa_has_spare(spa_t *spa, uint64_t guid)
4448{
4449	int i;
4450	uint64_t spareguid;
4451	spa_aux_vdev_t *sav = &spa->spa_spares;
4452
4453	for (i = 0; i < sav->sav_count; i++)
4454		if (sav->sav_vdevs[i]->vdev_guid == guid)
4455			return (B_TRUE);
4456
4457	for (i = 0; i < sav->sav_npending; i++) {
4458		if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
4459		    &spareguid) == 0 && spareguid == guid)
4460			return (B_TRUE);
4461	}
4462
4463	return (B_FALSE);
4464}
4465
4466/*
4467 * Post a sysevent corresponding to the given event.  The 'name' must be one of
4468 * the event definitions in sys/sysevent/eventdefs.h.  The payload will be
4469 * filled in from the spa and (optionally) the vdev.  This doesn't do anything
4470 * in the userland libzpool, as we don't want consumers to misinterpret ztest
4471 * or zdb as real changes.
4472 */
4473void
4474spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
4475{
4476#ifdef _KERNEL
4477	sysevent_t		*ev;
4478	sysevent_attr_list_t	*attr = NULL;
4479	sysevent_value_t	value;
4480	sysevent_id_t		eid;
4481
4482	ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
4483	    SE_SLEEP);
4484
4485	value.value_type = SE_DATA_TYPE_STRING;
4486	value.value.sv_string = spa_name(spa);
4487	if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
4488		goto done;
4489
4490	value.value_type = SE_DATA_TYPE_UINT64;
4491	value.value.sv_uint64 = spa_guid(spa);
4492	if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
4493		goto done;
4494
4495	if (vd) {
4496		value.value_type = SE_DATA_TYPE_UINT64;
4497		value.value.sv_uint64 = vd->vdev_guid;
4498		if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
4499		    SE_SLEEP) != 0)
4500			goto done;
4501
4502		if (vd->vdev_path) {
4503			value.value_type = SE_DATA_TYPE_STRING;
4504			value.value.sv_string = vd->vdev_path;
4505			if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
4506			    &value, SE_SLEEP) != 0)
4507				goto done;
4508		}
4509	}
4510
4511	if (sysevent_attach_attributes(ev, attr) != 0)
4512		goto done;
4513	attr = NULL;
4514
4515	(void) log_sysevent(ev, SE_SLEEP, &eid);
4516
4517done:
4518	if (attr)
4519		sysevent_free_attr(attr);
4520	sysevent_free(ev);
4521#endif
4522}
4523