xref: /titanic_41/usr/src/uts/common/fs/zfs/spa.c (revision 56e2cc86321ec889bf83a888d902c60d6fb2ef8d)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*
28  * This file contains all the routines used when modifying on-disk SPA state.
29  * This includes opening, importing, destroying, exporting a pool, and syncing a
30  * pool.
31  */
32 
33 #include <sys/zfs_context.h>
34 #include <sys/fm/fs/zfs.h>
35 #include <sys/spa_impl.h>
36 #include <sys/zio.h>
37 #include <sys/zio_checksum.h>
38 #include <sys/zio_compress.h>
39 #include <sys/dmu.h>
40 #include <sys/dmu_tx.h>
41 #include <sys/zap.h>
42 #include <sys/zil.h>
43 #include <sys/vdev_impl.h>
44 #include <sys/metaslab.h>
45 #include <sys/uberblock_impl.h>
46 #include <sys/txg.h>
47 #include <sys/avl.h>
48 #include <sys/dmu_traverse.h>
49 #include <sys/dmu_objset.h>
50 #include <sys/unique.h>
51 #include <sys/dsl_pool.h>
52 #include <sys/dsl_dataset.h>
53 #include <sys/dsl_dir.h>
54 #include <sys/dsl_prop.h>
55 #include <sys/dsl_synctask.h>
56 #include <sys/fs/zfs.h>
57 #include <sys/arc.h>
58 #include <sys/callb.h>
59 #include <sys/systeminfo.h>
60 #include <sys/sunddi.h>
61 #include <sys/spa_boot.h>
62 
63 #ifdef	_KERNEL
64 #include <sys/zone.h>
65 #endif	/* _KERNEL */
66 
67 #include "zfs_prop.h"
68 #include "zfs_comutil.h"
69 
70 enum zti_modes {
71 	zti_mode_fixed,			/* value is # of threads (min 1) */
72 	zti_mode_online_percent,	/* value is % of online CPUs */
73 	zti_mode_tune,			/* fill from zio_taskq_tune_* */
74 	zti_nmodes
75 };
76 
77 #define	ZTI_THREAD_FIX(n)	{ zti_mode_fixed, (n) }
78 #define	ZTI_THREAD_PCT(n)	{ zti_mode_online_percent, (n) }
79 #define	ZTI_THREAD_TUNE		{ zti_mode_tune, 0 }
80 
81 #define	ZTI_THREAD_ONE		ZTI_THREAD_FIX(1)
82 
83 typedef struct zio_taskq_info {
84 	const char *zti_name;
85 	struct {
86 		enum zti_modes zti_mode;
87 		uint_t zti_value;
88 	} zti_nthreads[ZIO_TASKQ_TYPES];
89 } zio_taskq_info_t;
90 
91 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
92 				"issue",		"intr"
93 };
94 
95 const zio_taskq_info_t zio_taskqs[ZIO_TYPES] = {
96 	/*			ISSUE			INTR		*/
97 	{ "spa_zio_null",	{ ZTI_THREAD_ONE,	ZTI_THREAD_ONE } },
98 	{ "spa_zio_read",	{ ZTI_THREAD_FIX(8),	ZTI_THREAD_TUNE } },
99 	{ "spa_zio_write",	{ ZTI_THREAD_TUNE,	ZTI_THREAD_FIX(8) } },
100 	{ "spa_zio_free",	{ ZTI_THREAD_ONE,	ZTI_THREAD_ONE } },
101 	{ "spa_zio_claim",	{ ZTI_THREAD_ONE,	ZTI_THREAD_ONE } },
102 	{ "spa_zio_ioctl",	{ ZTI_THREAD_ONE,	ZTI_THREAD_ONE } },
103 };
104 
105 enum zti_modes zio_taskq_tune_mode = zti_mode_online_percent;
106 uint_t zio_taskq_tune_value = 80;	/* #threads = 80% of # online CPUs */
107 
108 static void spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx);
109 static boolean_t spa_has_active_shared_spare(spa_t *spa);
110 
111 /*
112  * ==========================================================================
113  * SPA properties routines
114  * ==========================================================================
115  */
116 
117 /*
118  * Add a (source=src, propname=propval) list to an nvlist.
119  */
120 static void
121 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
122     uint64_t intval, zprop_source_t src)
123 {
124 	const char *propname = zpool_prop_to_name(prop);
125 	nvlist_t *propval;
126 
127 	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
128 	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
129 
130 	if (strval != NULL)
131 		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
132 	else
133 		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
134 
135 	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
136 	nvlist_free(propval);
137 }
138 
139 /*
140  * Get property values from the spa configuration.
141  */
142 static void
143 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
144 {
145 	uint64_t size;
146 	uint64_t used;
147 	uint64_t cap, version;
148 	zprop_source_t src = ZPROP_SRC_NONE;
149 	spa_config_dirent_t *dp;
150 
151 	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
152 
153 	if (spa->spa_root_vdev != NULL) {
154 		size = spa_get_space(spa);
155 		used = spa_get_alloc(spa);
156 		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
157 		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
158 		spa_prop_add_list(*nvp, ZPOOL_PROP_USED, NULL, used, src);
159 		spa_prop_add_list(*nvp, ZPOOL_PROP_AVAILABLE, NULL,
160 		    size - used, src);
161 
162 		cap = (size == 0) ? 0 : (used * 100 / size);
163 		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
164 
165 		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
166 		    spa->spa_root_vdev->vdev_state, src);
167 
168 		version = spa_version(spa);
169 		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
170 			src = ZPROP_SRC_DEFAULT;
171 		else
172 			src = ZPROP_SRC_LOCAL;
173 		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
174 	}
175 
176 	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
177 
178 	if (spa->spa_root != NULL)
179 		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
180 		    0, ZPROP_SRC_LOCAL);
181 
182 	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
183 		if (dp->scd_path == NULL) {
184 			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
185 			    "none", 0, ZPROP_SRC_LOCAL);
186 		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
187 			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
188 			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
189 		}
190 	}
191 }
192 
193 /*
194  * Get zpool property values.
195  */
196 int
197 spa_prop_get(spa_t *spa, nvlist_t **nvp)
198 {
199 	zap_cursor_t zc;
200 	zap_attribute_t za;
201 	objset_t *mos = spa->spa_meta_objset;
202 	int err;
203 
204 	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
205 
206 	mutex_enter(&spa->spa_props_lock);
207 
208 	/*
209 	 * Get properties from the spa config.
210 	 */
211 	spa_prop_get_config(spa, nvp);
212 
213 	/* If no pool property object, no more prop to get. */
214 	if (spa->spa_pool_props_object == 0) {
215 		mutex_exit(&spa->spa_props_lock);
216 		return (0);
217 	}
218 
219 	/*
220 	 * Get properties from the MOS pool property object.
221 	 */
222 	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
223 	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
224 	    zap_cursor_advance(&zc)) {
225 		uint64_t intval = 0;
226 		char *strval = NULL;
227 		zprop_source_t src = ZPROP_SRC_DEFAULT;
228 		zpool_prop_t prop;
229 
230 		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
231 			continue;
232 
233 		switch (za.za_integer_length) {
234 		case 8:
235 			/* integer property */
236 			if (za.za_first_integer !=
237 			    zpool_prop_default_numeric(prop))
238 				src = ZPROP_SRC_LOCAL;
239 
240 			if (prop == ZPOOL_PROP_BOOTFS) {
241 				dsl_pool_t *dp;
242 				dsl_dataset_t *ds = NULL;
243 
244 				dp = spa_get_dsl(spa);
245 				rw_enter(&dp->dp_config_rwlock, RW_READER);
246 				if (err = dsl_dataset_hold_obj(dp,
247 				    za.za_first_integer, FTAG, &ds)) {
248 					rw_exit(&dp->dp_config_rwlock);
249 					break;
250 				}
251 
252 				strval = kmem_alloc(
253 				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
254 				    KM_SLEEP);
255 				dsl_dataset_name(ds, strval);
256 				dsl_dataset_rele(ds, FTAG);
257 				rw_exit(&dp->dp_config_rwlock);
258 			} else {
259 				strval = NULL;
260 				intval = za.za_first_integer;
261 			}
262 
263 			spa_prop_add_list(*nvp, prop, strval, intval, src);
264 
265 			if (strval != NULL)
266 				kmem_free(strval,
267 				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
268 
269 			break;
270 
271 		case 1:
272 			/* string property */
273 			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
274 			err = zap_lookup(mos, spa->spa_pool_props_object,
275 			    za.za_name, 1, za.za_num_integers, strval);
276 			if (err) {
277 				kmem_free(strval, za.za_num_integers);
278 				break;
279 			}
280 			spa_prop_add_list(*nvp, prop, strval, 0, src);
281 			kmem_free(strval, za.za_num_integers);
282 			break;
283 
284 		default:
285 			break;
286 		}
287 	}
288 	zap_cursor_fini(&zc);
289 	mutex_exit(&spa->spa_props_lock);
290 out:
291 	if (err && err != ENOENT) {
292 		nvlist_free(*nvp);
293 		*nvp = NULL;
294 		return (err);
295 	}
296 
297 	return (0);
298 }
299 
300 /*
301  * Validate the given pool properties nvlist and modify the list
302  * for the property values to be set.
303  */
304 static int
305 spa_prop_validate(spa_t *spa, nvlist_t *props)
306 {
307 	nvpair_t *elem;
308 	int error = 0, reset_bootfs = 0;
309 	uint64_t objnum;
310 
311 	elem = NULL;
312 	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
313 		zpool_prop_t prop;
314 		char *propname, *strval;
315 		uint64_t intval;
316 		objset_t *os;
317 		char *slash;
318 
319 		propname = nvpair_name(elem);
320 
321 		if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
322 			return (EINVAL);
323 
324 		switch (prop) {
325 		case ZPOOL_PROP_VERSION:
326 			error = nvpair_value_uint64(elem, &intval);
327 			if (!error &&
328 			    (intval < spa_version(spa) || intval > SPA_VERSION))
329 				error = EINVAL;
330 			break;
331 
332 		case ZPOOL_PROP_DELEGATION:
333 		case ZPOOL_PROP_AUTOREPLACE:
334 		case ZPOOL_PROP_LISTSNAPS:
335 			error = nvpair_value_uint64(elem, &intval);
336 			if (!error && intval > 1)
337 				error = EINVAL;
338 			break;
339 
340 		case ZPOOL_PROP_BOOTFS:
341 			/*
342 			 * If the pool version is less than SPA_VERSION_BOOTFS,
343 			 * or the pool is still being created (version == 0),
344 			 * the bootfs property cannot be set.
345 			 */
346 			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
347 				error = ENOTSUP;
348 				break;
349 			}
350 
351 			/*
352 			 * Make sure the vdev config is bootable
353 			 */
354 			if (!vdev_is_bootable(spa->spa_root_vdev)) {
355 				error = ENOTSUP;
356 				break;
357 			}
358 
359 			reset_bootfs = 1;
360 
361 			error = nvpair_value_string(elem, &strval);
362 
363 			if (!error) {
364 				uint64_t compress;
365 
366 				if (strval == NULL || strval[0] == '\0') {
367 					objnum = zpool_prop_default_numeric(
368 					    ZPOOL_PROP_BOOTFS);
369 					break;
370 				}
371 
372 				if (error = dmu_objset_open(strval, DMU_OST_ZFS,
373 				    DS_MODE_USER | DS_MODE_READONLY, &os))
374 					break;
375 
376 				/* We don't support gzip bootable datasets */
377 				if ((error = dsl_prop_get_integer(strval,
378 				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
379 				    &compress, NULL)) == 0 &&
380 				    !BOOTFS_COMPRESS_VALID(compress)) {
381 					error = ENOTSUP;
382 				} else {
383 					objnum = dmu_objset_id(os);
384 				}
385 				dmu_objset_close(os);
386 			}
387 			break;
388 
389 		case ZPOOL_PROP_FAILUREMODE:
390 			error = nvpair_value_uint64(elem, &intval);
391 			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
392 			    intval > ZIO_FAILURE_MODE_PANIC))
393 				error = EINVAL;
394 
395 			/*
396 			 * This is a special case which only occurs when
397 			 * the pool has completely failed. This allows
398 			 * the user to change the in-core failmode property
399 			 * without syncing it out to disk (I/Os might
400 			 * currently be blocked). We do this by returning
401 			 * EIO to the caller (spa_prop_set) to trick it
402 			 * into thinking we encountered a property validation
403 			 * error.
404 			 */
405 			if (!error && spa_suspended(spa)) {
406 				spa->spa_failmode = intval;
407 				error = EIO;
408 			}
409 			break;
410 
411 		case ZPOOL_PROP_CACHEFILE:
412 			if ((error = nvpair_value_string(elem, &strval)) != 0)
413 				break;
414 
415 			if (strval[0] == '\0')
416 				break;
417 
418 			if (strcmp(strval, "none") == 0)
419 				break;
420 
421 			if (strval[0] != '/') {
422 				error = EINVAL;
423 				break;
424 			}
425 
426 			slash = strrchr(strval, '/');
427 			ASSERT(slash != NULL);
428 
429 			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
430 			    strcmp(slash, "/..") == 0)
431 				error = EINVAL;
432 			break;
433 		}
434 
435 		if (error)
436 			break;
437 	}
438 
439 	if (!error && reset_bootfs) {
440 		error = nvlist_remove(props,
441 		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
442 
443 		if (!error) {
444 			error = nvlist_add_uint64(props,
445 			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
446 		}
447 	}
448 
449 	return (error);
450 }
451 
452 void
453 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
454 {
455 	char *cachefile;
456 	spa_config_dirent_t *dp;
457 
458 	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
459 	    &cachefile) != 0)
460 		return;
461 
462 	dp = kmem_alloc(sizeof (spa_config_dirent_t),
463 	    KM_SLEEP);
464 
465 	if (cachefile[0] == '\0')
466 		dp->scd_path = spa_strdup(spa_config_path);
467 	else if (strcmp(cachefile, "none") == 0)
468 		dp->scd_path = NULL;
469 	else
470 		dp->scd_path = spa_strdup(cachefile);
471 
472 	list_insert_head(&spa->spa_config_list, dp);
473 	if (need_sync)
474 		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
475 }
476 
477 int
478 spa_prop_set(spa_t *spa, nvlist_t *nvp)
479 {
480 	int error;
481 	nvpair_t *elem;
482 	boolean_t need_sync = B_FALSE;
483 	zpool_prop_t prop;
484 
485 	if ((error = spa_prop_validate(spa, nvp)) != 0)
486 		return (error);
487 
488 	elem = NULL;
489 	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
490 		if ((prop = zpool_name_to_prop(
491 		    nvpair_name(elem))) == ZPROP_INVAL)
492 			return (EINVAL);
493 
494 		if (prop == ZPOOL_PROP_CACHEFILE || prop == ZPOOL_PROP_ALTROOT)
495 			continue;
496 
497 		need_sync = B_TRUE;
498 		break;
499 	}
500 
501 	if (need_sync)
502 		return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
503 		    spa, nvp, 3));
504 	else
505 		return (0);
506 }
507 
508 /*
509  * If the bootfs property value is dsobj, clear it.
510  */
511 void
512 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
513 {
514 	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
515 		VERIFY(zap_remove(spa->spa_meta_objset,
516 		    spa->spa_pool_props_object,
517 		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
518 		spa->spa_bootfs = 0;
519 	}
520 }
521 
522 /*
523  * ==========================================================================
524  * SPA state manipulation (open/create/destroy/import/export)
525  * ==========================================================================
526  */
527 
528 static int
529 spa_error_entry_compare(const void *a, const void *b)
530 {
531 	spa_error_entry_t *sa = (spa_error_entry_t *)a;
532 	spa_error_entry_t *sb = (spa_error_entry_t *)b;
533 	int ret;
534 
535 	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
536 	    sizeof (zbookmark_t));
537 
538 	if (ret < 0)
539 		return (-1);
540 	else if (ret > 0)
541 		return (1);
542 	else
543 		return (0);
544 }
545 
546 /*
547  * Utility function which retrieves copies of the current logs and
548  * re-initializes them in the process.
549  */
550 void
551 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
552 {
553 	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
554 
555 	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
556 	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
557 
558 	avl_create(&spa->spa_errlist_scrub,
559 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
560 	    offsetof(spa_error_entry_t, se_avl));
561 	avl_create(&spa->spa_errlist_last,
562 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
563 	    offsetof(spa_error_entry_t, se_avl));
564 }
565 
566 /*
567  * Activate an uninitialized pool.
568  */
569 static void
570 spa_activate(spa_t *spa, int mode)
571 {
572 	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
573 
574 	spa->spa_state = POOL_STATE_ACTIVE;
575 	spa->spa_mode = mode;
576 
577 	spa->spa_normal_class = metaslab_class_create(zfs_metaslab_ops);
578 	spa->spa_log_class = metaslab_class_create(zfs_metaslab_ops);
579 
580 	for (int t = 0; t < ZIO_TYPES; t++) {
581 		const zio_taskq_info_t *ztip = &zio_taskqs[t];
582 		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
583 			enum zti_modes mode = ztip->zti_nthreads[q].zti_mode;
584 			uint_t value = ztip->zti_nthreads[q].zti_value;
585 			char name[32];
586 
587 			(void) snprintf(name, sizeof (name),
588 			    "%s_%s", ztip->zti_name, zio_taskq_types[q]);
589 
590 			if (mode == zti_mode_tune) {
591 				mode = zio_taskq_tune_mode;
592 				value = zio_taskq_tune_value;
593 				if (mode == zti_mode_tune)
594 					mode = zti_mode_online_percent;
595 			}
596 
597 			switch (mode) {
598 			case zti_mode_fixed:
599 				ASSERT3U(value, >=, 1);
600 				value = MAX(value, 1);
601 
602 				spa->spa_zio_taskq[t][q] = taskq_create(name,
603 				    value, maxclsyspri, 50, INT_MAX,
604 				    TASKQ_PREPOPULATE);
605 				break;
606 
607 			case zti_mode_online_percent:
608 				spa->spa_zio_taskq[t][q] = taskq_create(name,
609 				    value, maxclsyspri, 50, INT_MAX,
610 				    TASKQ_PREPOPULATE | TASKQ_THREADS_CPU_PCT);
611 				break;
612 
613 			case zti_mode_tune:
614 			default:
615 				panic("unrecognized mode for "
616 				    "zio_taskqs[%u]->zti_nthreads[%u] (%u:%u) "
617 				    "in spa_activate()",
618 				    t, q, mode, value);
619 				break;
620 			}
621 		}
622 	}
623 
624 	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
625 	    offsetof(vdev_t, vdev_config_dirty_node));
626 	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
627 	    offsetof(vdev_t, vdev_state_dirty_node));
628 
629 	txg_list_create(&spa->spa_vdev_txg_list,
630 	    offsetof(struct vdev, vdev_txg_node));
631 
632 	avl_create(&spa->spa_errlist_scrub,
633 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
634 	    offsetof(spa_error_entry_t, se_avl));
635 	avl_create(&spa->spa_errlist_last,
636 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
637 	    offsetof(spa_error_entry_t, se_avl));
638 }
639 
640 /*
641  * Opposite of spa_activate().
642  */
643 static void
644 spa_deactivate(spa_t *spa)
645 {
646 	ASSERT(spa->spa_sync_on == B_FALSE);
647 	ASSERT(spa->spa_dsl_pool == NULL);
648 	ASSERT(spa->spa_root_vdev == NULL);
649 	ASSERT(spa->spa_async_zio_root == NULL);
650 	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
651 
652 	txg_list_destroy(&spa->spa_vdev_txg_list);
653 
654 	list_destroy(&spa->spa_config_dirty_list);
655 	list_destroy(&spa->spa_state_dirty_list);
656 
657 	for (int t = 0; t < ZIO_TYPES; t++) {
658 		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
659 			taskq_destroy(spa->spa_zio_taskq[t][q]);
660 			spa->spa_zio_taskq[t][q] = NULL;
661 		}
662 	}
663 
664 	metaslab_class_destroy(spa->spa_normal_class);
665 	spa->spa_normal_class = NULL;
666 
667 	metaslab_class_destroy(spa->spa_log_class);
668 	spa->spa_log_class = NULL;
669 
670 	/*
671 	 * If this was part of an import or the open otherwise failed, we may
672 	 * still have errors left in the queues.  Empty them just in case.
673 	 */
674 	spa_errlog_drain(spa);
675 
676 	avl_destroy(&spa->spa_errlist_scrub);
677 	avl_destroy(&spa->spa_errlist_last);
678 
679 	spa->spa_state = POOL_STATE_UNINITIALIZED;
680 }
681 
682 /*
683  * Verify a pool configuration, and construct the vdev tree appropriately.  This
684  * will create all the necessary vdevs in the appropriate layout, with each vdev
685  * in the CLOSED state.  This will prep the pool before open/creation/import.
686  * All vdev validation is done by the vdev_alloc() routine.
687  */
688 static int
689 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
690     uint_t id, int atype)
691 {
692 	nvlist_t **child;
693 	uint_t c, children;
694 	int error;
695 
696 	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
697 		return (error);
698 
699 	if ((*vdp)->vdev_ops->vdev_op_leaf)
700 		return (0);
701 
702 	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
703 	    &child, &children);
704 
705 	if (error == ENOENT)
706 		return (0);
707 
708 	if (error) {
709 		vdev_free(*vdp);
710 		*vdp = NULL;
711 		return (EINVAL);
712 	}
713 
714 	for (c = 0; c < children; c++) {
715 		vdev_t *vd;
716 		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
717 		    atype)) != 0) {
718 			vdev_free(*vdp);
719 			*vdp = NULL;
720 			return (error);
721 		}
722 	}
723 
724 	ASSERT(*vdp != NULL);
725 
726 	return (0);
727 }
728 
729 /*
730  * Opposite of spa_load().
731  */
732 static void
733 spa_unload(spa_t *spa)
734 {
735 	int i;
736 
737 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
738 
739 	/*
740 	 * Stop async tasks.
741 	 */
742 	spa_async_suspend(spa);
743 
744 	/*
745 	 * Stop syncing.
746 	 */
747 	if (spa->spa_sync_on) {
748 		txg_sync_stop(spa->spa_dsl_pool);
749 		spa->spa_sync_on = B_FALSE;
750 	}
751 
752 	/*
753 	 * Wait for any outstanding async I/O to complete.
754 	 */
755 	if (spa->spa_async_zio_root != NULL) {
756 		(void) zio_wait(spa->spa_async_zio_root);
757 		spa->spa_async_zio_root = NULL;
758 	}
759 
760 	/*
761 	 * Close the dsl pool.
762 	 */
763 	if (spa->spa_dsl_pool) {
764 		dsl_pool_close(spa->spa_dsl_pool);
765 		spa->spa_dsl_pool = NULL;
766 	}
767 
768 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
769 
770 	/*
771 	 * Drop and purge level 2 cache
772 	 */
773 	spa_l2cache_drop(spa);
774 
775 	/*
776 	 * Close all vdevs.
777 	 */
778 	if (spa->spa_root_vdev)
779 		vdev_free(spa->spa_root_vdev);
780 	ASSERT(spa->spa_root_vdev == NULL);
781 
782 	for (i = 0; i < spa->spa_spares.sav_count; i++)
783 		vdev_free(spa->spa_spares.sav_vdevs[i]);
784 	if (spa->spa_spares.sav_vdevs) {
785 		kmem_free(spa->spa_spares.sav_vdevs,
786 		    spa->spa_spares.sav_count * sizeof (void *));
787 		spa->spa_spares.sav_vdevs = NULL;
788 	}
789 	if (spa->spa_spares.sav_config) {
790 		nvlist_free(spa->spa_spares.sav_config);
791 		spa->spa_spares.sav_config = NULL;
792 	}
793 	spa->spa_spares.sav_count = 0;
794 
795 	for (i = 0; i < spa->spa_l2cache.sav_count; i++)
796 		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
797 	if (spa->spa_l2cache.sav_vdevs) {
798 		kmem_free(spa->spa_l2cache.sav_vdevs,
799 		    spa->spa_l2cache.sav_count * sizeof (void *));
800 		spa->spa_l2cache.sav_vdevs = NULL;
801 	}
802 	if (spa->spa_l2cache.sav_config) {
803 		nvlist_free(spa->spa_l2cache.sav_config);
804 		spa->spa_l2cache.sav_config = NULL;
805 	}
806 	spa->spa_l2cache.sav_count = 0;
807 
808 	spa->spa_async_suspended = 0;
809 
810 	spa_config_exit(spa, SCL_ALL, FTAG);
811 }
812 
813 /*
814  * Load (or re-load) the current list of vdevs describing the active spares for
815  * this pool.  When this is called, we have some form of basic information in
816  * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
817  * then re-generate a more complete list including status information.
818  */
819 static void
820 spa_load_spares(spa_t *spa)
821 {
822 	nvlist_t **spares;
823 	uint_t nspares;
824 	int i;
825 	vdev_t *vd, *tvd;
826 
827 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
828 
829 	/*
830 	 * First, close and free any existing spare vdevs.
831 	 */
832 	for (i = 0; i < spa->spa_spares.sav_count; i++) {
833 		vd = spa->spa_spares.sav_vdevs[i];
834 
835 		/* Undo the call to spa_activate() below */
836 		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
837 		    B_FALSE)) != NULL && tvd->vdev_isspare)
838 			spa_spare_remove(tvd);
839 		vdev_close(vd);
840 		vdev_free(vd);
841 	}
842 
843 	if (spa->spa_spares.sav_vdevs)
844 		kmem_free(spa->spa_spares.sav_vdevs,
845 		    spa->spa_spares.sav_count * sizeof (void *));
846 
847 	if (spa->spa_spares.sav_config == NULL)
848 		nspares = 0;
849 	else
850 		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
851 		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
852 
853 	spa->spa_spares.sav_count = (int)nspares;
854 	spa->spa_spares.sav_vdevs = NULL;
855 
856 	if (nspares == 0)
857 		return;
858 
859 	/*
860 	 * Construct the array of vdevs, opening them to get status in the
861 	 * process.   For each spare, there is potentially two different vdev_t
862 	 * structures associated with it: one in the list of spares (used only
863 	 * for basic validation purposes) and one in the active vdev
864 	 * configuration (if it's spared in).  During this phase we open and
865 	 * validate each vdev on the spare list.  If the vdev also exists in the
866 	 * active configuration, then we also mark this vdev as an active spare.
867 	 */
868 	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
869 	    KM_SLEEP);
870 	for (i = 0; i < spa->spa_spares.sav_count; i++) {
871 		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
872 		    VDEV_ALLOC_SPARE) == 0);
873 		ASSERT(vd != NULL);
874 
875 		spa->spa_spares.sav_vdevs[i] = vd;
876 
877 		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
878 		    B_FALSE)) != NULL) {
879 			if (!tvd->vdev_isspare)
880 				spa_spare_add(tvd);
881 
882 			/*
883 			 * We only mark the spare active if we were successfully
884 			 * able to load the vdev.  Otherwise, importing a pool
885 			 * with a bad active spare would result in strange
886 			 * behavior, because multiple pool would think the spare
887 			 * is actively in use.
888 			 *
889 			 * There is a vulnerability here to an equally bizarre
890 			 * circumstance, where a dead active spare is later
891 			 * brought back to life (onlined or otherwise).  Given
892 			 * the rarity of this scenario, and the extra complexity
893 			 * it adds, we ignore the possibility.
894 			 */
895 			if (!vdev_is_dead(tvd))
896 				spa_spare_activate(tvd);
897 		}
898 
899 		vd->vdev_top = vd;
900 		vd->vdev_aux = &spa->spa_spares;
901 
902 		if (vdev_open(vd) != 0)
903 			continue;
904 
905 		if (vdev_validate_aux(vd) == 0)
906 			spa_spare_add(vd);
907 	}
908 
909 	/*
910 	 * Recompute the stashed list of spares, with status information
911 	 * this time.
912 	 */
913 	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
914 	    DATA_TYPE_NVLIST_ARRAY) == 0);
915 
916 	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
917 	    KM_SLEEP);
918 	for (i = 0; i < spa->spa_spares.sav_count; i++)
919 		spares[i] = vdev_config_generate(spa,
920 		    spa->spa_spares.sav_vdevs[i], B_TRUE, B_TRUE, B_FALSE);
921 	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
922 	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
923 	for (i = 0; i < spa->spa_spares.sav_count; i++)
924 		nvlist_free(spares[i]);
925 	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
926 }
927 
928 /*
929  * Load (or re-load) the current list of vdevs describing the active l2cache for
930  * this pool.  When this is called, we have some form of basic information in
931  * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
932  * then re-generate a more complete list including status information.
933  * Devices which are already active have their details maintained, and are
934  * not re-opened.
935  */
936 static void
937 spa_load_l2cache(spa_t *spa)
938 {
939 	nvlist_t **l2cache;
940 	uint_t nl2cache;
941 	int i, j, oldnvdevs;
942 	uint64_t guid, size;
943 	vdev_t *vd, **oldvdevs, **newvdevs;
944 	spa_aux_vdev_t *sav = &spa->spa_l2cache;
945 
946 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
947 
948 	if (sav->sav_config != NULL) {
949 		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
950 		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
951 		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
952 	} else {
953 		nl2cache = 0;
954 	}
955 
956 	oldvdevs = sav->sav_vdevs;
957 	oldnvdevs = sav->sav_count;
958 	sav->sav_vdevs = NULL;
959 	sav->sav_count = 0;
960 
961 	/*
962 	 * Process new nvlist of vdevs.
963 	 */
964 	for (i = 0; i < nl2cache; i++) {
965 		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
966 		    &guid) == 0);
967 
968 		newvdevs[i] = NULL;
969 		for (j = 0; j < oldnvdevs; j++) {
970 			vd = oldvdevs[j];
971 			if (vd != NULL && guid == vd->vdev_guid) {
972 				/*
973 				 * Retain previous vdev for add/remove ops.
974 				 */
975 				newvdevs[i] = vd;
976 				oldvdevs[j] = NULL;
977 				break;
978 			}
979 		}
980 
981 		if (newvdevs[i] == NULL) {
982 			/*
983 			 * Create new vdev
984 			 */
985 			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
986 			    VDEV_ALLOC_L2CACHE) == 0);
987 			ASSERT(vd != NULL);
988 			newvdevs[i] = vd;
989 
990 			/*
991 			 * Commit this vdev as an l2cache device,
992 			 * even if it fails to open.
993 			 */
994 			spa_l2cache_add(vd);
995 
996 			vd->vdev_top = vd;
997 			vd->vdev_aux = sav;
998 
999 			spa_l2cache_activate(vd);
1000 
1001 			if (vdev_open(vd) != 0)
1002 				continue;
1003 
1004 			(void) vdev_validate_aux(vd);
1005 
1006 			if (!vdev_is_dead(vd)) {
1007 				size = vdev_get_rsize(vd);
1008 				l2arc_add_vdev(spa, vd,
1009 				    VDEV_LABEL_START_SIZE,
1010 				    size - VDEV_LABEL_START_SIZE);
1011 			}
1012 		}
1013 	}
1014 
1015 	/*
1016 	 * Purge vdevs that were dropped
1017 	 */
1018 	for (i = 0; i < oldnvdevs; i++) {
1019 		uint64_t pool;
1020 
1021 		vd = oldvdevs[i];
1022 		if (vd != NULL) {
1023 			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1024 			    pool != 0ULL && l2arc_vdev_present(vd))
1025 				l2arc_remove_vdev(vd);
1026 			(void) vdev_close(vd);
1027 			spa_l2cache_remove(vd);
1028 		}
1029 	}
1030 
1031 	if (oldvdevs)
1032 		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1033 
1034 	if (sav->sav_config == NULL)
1035 		goto out;
1036 
1037 	sav->sav_vdevs = newvdevs;
1038 	sav->sav_count = (int)nl2cache;
1039 
1040 	/*
1041 	 * Recompute the stashed list of l2cache devices, with status
1042 	 * information this time.
1043 	 */
1044 	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1045 	    DATA_TYPE_NVLIST_ARRAY) == 0);
1046 
1047 	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1048 	for (i = 0; i < sav->sav_count; i++)
1049 		l2cache[i] = vdev_config_generate(spa,
1050 		    sav->sav_vdevs[i], B_TRUE, B_FALSE, B_TRUE);
1051 	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1052 	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1053 out:
1054 	for (i = 0; i < sav->sav_count; i++)
1055 		nvlist_free(l2cache[i]);
1056 	if (sav->sav_count)
1057 		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1058 }
1059 
1060 static int
1061 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1062 {
1063 	dmu_buf_t *db;
1064 	char *packed = NULL;
1065 	size_t nvsize = 0;
1066 	int error;
1067 	*value = NULL;
1068 
1069 	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1070 	nvsize = *(uint64_t *)db->db_data;
1071 	dmu_buf_rele(db, FTAG);
1072 
1073 	packed = kmem_alloc(nvsize, KM_SLEEP);
1074 	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1075 	    DMU_READ_PREFETCH);
1076 	if (error == 0)
1077 		error = nvlist_unpack(packed, nvsize, value, 0);
1078 	kmem_free(packed, nvsize);
1079 
1080 	return (error);
1081 }
1082 
1083 /*
1084  * Checks to see if the given vdev could not be opened, in which case we post a
1085  * sysevent to notify the autoreplace code that the device has been removed.
1086  */
1087 static void
1088 spa_check_removed(vdev_t *vd)
1089 {
1090 	int c;
1091 
1092 	for (c = 0; c < vd->vdev_children; c++)
1093 		spa_check_removed(vd->vdev_child[c]);
1094 
1095 	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1096 		zfs_post_autoreplace(vd->vdev_spa, vd);
1097 		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1098 	}
1099 }
1100 
1101 /*
1102  * Check for missing log devices
1103  */
1104 int
1105 spa_check_logs(spa_t *spa)
1106 {
1107 	switch (spa->spa_log_state) {
1108 	case SPA_LOG_MISSING:
1109 		/* need to recheck in case slog has been restored */
1110 	case SPA_LOG_UNKNOWN:
1111 		if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1112 		    DS_FIND_CHILDREN)) {
1113 			spa->spa_log_state = SPA_LOG_MISSING;
1114 			return (1);
1115 		}
1116 		break;
1117 
1118 	case SPA_LOG_CLEAR:
1119 		(void) dmu_objset_find(spa->spa_name, zil_clear_log_chain, NULL,
1120 		    DS_FIND_CHILDREN);
1121 		break;
1122 	}
1123 	spa->spa_log_state = SPA_LOG_GOOD;
1124 	return (0);
1125 }
1126 
1127 /*
1128  * Load an existing storage pool, using the pool's builtin spa_config as a
1129  * source of configuration information.
1130  */
1131 static int
1132 spa_load(spa_t *spa, nvlist_t *config, spa_load_state_t state, int mosconfig)
1133 {
1134 	int error = 0;
1135 	nvlist_t *nvroot = NULL;
1136 	vdev_t *rvd;
1137 	uberblock_t *ub = &spa->spa_uberblock;
1138 	uint64_t config_cache_txg = spa->spa_config_txg;
1139 	uint64_t pool_guid;
1140 	uint64_t version;
1141 	uint64_t autoreplace = 0;
1142 	int orig_mode = spa->spa_mode;
1143 	char *ereport = FM_EREPORT_ZFS_POOL;
1144 
1145 	/*
1146 	 * If this is an untrusted config, access the pool in read-only mode.
1147 	 * This prevents things like resilvering recently removed devices.
1148 	 */
1149 	if (!mosconfig)
1150 		spa->spa_mode = FREAD;
1151 
1152 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1153 
1154 	spa->spa_load_state = state;
1155 
1156 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) ||
1157 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
1158 		error = EINVAL;
1159 		goto out;
1160 	}
1161 
1162 	/*
1163 	 * Versioning wasn't explicitly added to the label until later, so if
1164 	 * it's not present treat it as the initial version.
1165 	 */
1166 	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &version) != 0)
1167 		version = SPA_VERSION_INITIAL;
1168 
1169 	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1170 	    &spa->spa_config_txg);
1171 
1172 	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1173 	    spa_guid_exists(pool_guid, 0)) {
1174 		error = EEXIST;
1175 		goto out;
1176 	}
1177 
1178 	spa->spa_load_guid = pool_guid;
1179 
1180 	/*
1181 	 * Create "The Godfather" zio to hold all async IOs
1182 	 */
1183 	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1184 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1185 
1186 	/*
1187 	 * Parse the configuration into a vdev tree.  We explicitly set the
1188 	 * value that will be returned by spa_version() since parsing the
1189 	 * configuration requires knowing the version number.
1190 	 */
1191 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1192 	spa->spa_ubsync.ub_version = version;
1193 	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_LOAD);
1194 	spa_config_exit(spa, SCL_ALL, FTAG);
1195 
1196 	if (error != 0)
1197 		goto out;
1198 
1199 	ASSERT(spa->spa_root_vdev == rvd);
1200 	ASSERT(spa_guid(spa) == pool_guid);
1201 
1202 	/*
1203 	 * Try to open all vdevs, loading each label in the process.
1204 	 */
1205 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1206 	error = vdev_open(rvd);
1207 	spa_config_exit(spa, SCL_ALL, FTAG);
1208 	if (error != 0)
1209 		goto out;
1210 
1211 	/*
1212 	 * We need to validate the vdev labels against the configuration that
1213 	 * we have in hand, which is dependent on the setting of mosconfig. If
1214 	 * mosconfig is true then we're validating the vdev labels based on
1215 	 * that config. Otherwise, we're validating against the cached config
1216 	 * (zpool.cache) that was read when we loaded the zfs module, and then
1217 	 * later we will recursively call spa_load() and validate against
1218 	 * the vdev config.
1219 	 */
1220 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1221 	error = vdev_validate(rvd);
1222 	spa_config_exit(spa, SCL_ALL, FTAG);
1223 	if (error != 0)
1224 		goto out;
1225 
1226 	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
1227 		error = ENXIO;
1228 		goto out;
1229 	}
1230 
1231 	/*
1232 	 * Find the best uberblock.
1233 	 */
1234 	vdev_uberblock_load(NULL, rvd, ub);
1235 
1236 	/*
1237 	 * If we weren't able to find a single valid uberblock, return failure.
1238 	 */
1239 	if (ub->ub_txg == 0) {
1240 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1241 		    VDEV_AUX_CORRUPT_DATA);
1242 		error = ENXIO;
1243 		goto out;
1244 	}
1245 
1246 	/*
1247 	 * If the pool is newer than the code, we can't open it.
1248 	 */
1249 	if (ub->ub_version > SPA_VERSION) {
1250 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1251 		    VDEV_AUX_VERSION_NEWER);
1252 		error = ENOTSUP;
1253 		goto out;
1254 	}
1255 
1256 	/*
1257 	 * If the vdev guid sum doesn't match the uberblock, we have an
1258 	 * incomplete configuration.
1259 	 */
1260 	if (rvd->vdev_guid_sum != ub->ub_guid_sum && mosconfig) {
1261 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1262 		    VDEV_AUX_BAD_GUID_SUM);
1263 		error = ENXIO;
1264 		goto out;
1265 	}
1266 
1267 	/*
1268 	 * Initialize internal SPA structures.
1269 	 */
1270 	spa->spa_state = POOL_STATE_ACTIVE;
1271 	spa->spa_ubsync = spa->spa_uberblock;
1272 	spa->spa_first_txg = spa_last_synced_txg(spa) + 1;
1273 	error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1274 	if (error) {
1275 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1276 		    VDEV_AUX_CORRUPT_DATA);
1277 		goto out;
1278 	}
1279 	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1280 
1281 	if (zap_lookup(spa->spa_meta_objset,
1282 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
1283 	    sizeof (uint64_t), 1, &spa->spa_config_object) != 0) {
1284 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1285 		    VDEV_AUX_CORRUPT_DATA);
1286 		error = EIO;
1287 		goto out;
1288 	}
1289 
1290 	if (!mosconfig) {
1291 		nvlist_t *newconfig;
1292 		uint64_t hostid;
1293 
1294 		if (load_nvlist(spa, spa->spa_config_object, &newconfig) != 0) {
1295 			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1296 			    VDEV_AUX_CORRUPT_DATA);
1297 			error = EIO;
1298 			goto out;
1299 		}
1300 
1301 		if (!spa_is_root(spa) && nvlist_lookup_uint64(newconfig,
1302 		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
1303 			char *hostname;
1304 			unsigned long myhostid = 0;
1305 
1306 			VERIFY(nvlist_lookup_string(newconfig,
1307 			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
1308 
1309 #ifdef	_KERNEL
1310 			myhostid = zone_get_hostid(NULL);
1311 #else	/* _KERNEL */
1312 			/*
1313 			 * We're emulating the system's hostid in userland, so
1314 			 * we can't use zone_get_hostid().
1315 			 */
1316 			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
1317 #endif	/* _KERNEL */
1318 			if (hostid != 0 && myhostid != 0 &&
1319 			    hostid != myhostid) {
1320 				cmn_err(CE_WARN, "pool '%s' could not be "
1321 				    "loaded as it was last accessed by "
1322 				    "another system (host: %s hostid: 0x%lx). "
1323 				    "See: http://www.sun.com/msg/ZFS-8000-EY",
1324 				    spa_name(spa), hostname,
1325 				    (unsigned long)hostid);
1326 				error = EBADF;
1327 				goto out;
1328 			}
1329 		}
1330 
1331 		spa_config_set(spa, newconfig);
1332 		spa_unload(spa);
1333 		spa_deactivate(spa);
1334 		spa_activate(spa, orig_mode);
1335 
1336 		return (spa_load(spa, newconfig, state, B_TRUE));
1337 	}
1338 
1339 	if (zap_lookup(spa->spa_meta_objset,
1340 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
1341 	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj) != 0) {
1342 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1343 		    VDEV_AUX_CORRUPT_DATA);
1344 		error = EIO;
1345 		goto out;
1346 	}
1347 
1348 	/*
1349 	 * Load the bit that tells us to use the new accounting function
1350 	 * (raid-z deflation).  If we have an older pool, this will not
1351 	 * be present.
1352 	 */
1353 	error = zap_lookup(spa->spa_meta_objset,
1354 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
1355 	    sizeof (uint64_t), 1, &spa->spa_deflate);
1356 	if (error != 0 && error != ENOENT) {
1357 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1358 		    VDEV_AUX_CORRUPT_DATA);
1359 		error = EIO;
1360 		goto out;
1361 	}
1362 
1363 	/*
1364 	 * Load the persistent error log.  If we have an older pool, this will
1365 	 * not be present.
1366 	 */
1367 	error = zap_lookup(spa->spa_meta_objset,
1368 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_LAST,
1369 	    sizeof (uint64_t), 1, &spa->spa_errlog_last);
1370 	if (error != 0 && error != ENOENT) {
1371 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1372 		    VDEV_AUX_CORRUPT_DATA);
1373 		error = EIO;
1374 		goto out;
1375 	}
1376 
1377 	error = zap_lookup(spa->spa_meta_objset,
1378 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_SCRUB,
1379 	    sizeof (uint64_t), 1, &spa->spa_errlog_scrub);
1380 	if (error != 0 && error != ENOENT) {
1381 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1382 		    VDEV_AUX_CORRUPT_DATA);
1383 		error = EIO;
1384 		goto out;
1385 	}
1386 
1387 	/*
1388 	 * Load the history object.  If we have an older pool, this
1389 	 * will not be present.
1390 	 */
1391 	error = zap_lookup(spa->spa_meta_objset,
1392 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_HISTORY,
1393 	    sizeof (uint64_t), 1, &spa->spa_history);
1394 	if (error != 0 && error != ENOENT) {
1395 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1396 		    VDEV_AUX_CORRUPT_DATA);
1397 		error = EIO;
1398 		goto out;
1399 	}
1400 
1401 	/*
1402 	 * Load any hot spares for this pool.
1403 	 */
1404 	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1405 	    DMU_POOL_SPARES, sizeof (uint64_t), 1, &spa->spa_spares.sav_object);
1406 	if (error != 0 && error != ENOENT) {
1407 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1408 		    VDEV_AUX_CORRUPT_DATA);
1409 		error = EIO;
1410 		goto out;
1411 	}
1412 	if (error == 0) {
1413 		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
1414 		if (load_nvlist(spa, spa->spa_spares.sav_object,
1415 		    &spa->spa_spares.sav_config) != 0) {
1416 			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1417 			    VDEV_AUX_CORRUPT_DATA);
1418 			error = EIO;
1419 			goto out;
1420 		}
1421 
1422 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1423 		spa_load_spares(spa);
1424 		spa_config_exit(spa, SCL_ALL, FTAG);
1425 	}
1426 
1427 	/*
1428 	 * Load any level 2 ARC devices for this pool.
1429 	 */
1430 	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1431 	    DMU_POOL_L2CACHE, sizeof (uint64_t), 1,
1432 	    &spa->spa_l2cache.sav_object);
1433 	if (error != 0 && error != ENOENT) {
1434 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1435 		    VDEV_AUX_CORRUPT_DATA);
1436 		error = EIO;
1437 		goto out;
1438 	}
1439 	if (error == 0) {
1440 		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
1441 		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
1442 		    &spa->spa_l2cache.sav_config) != 0) {
1443 			vdev_set_state(rvd, B_TRUE,
1444 			    VDEV_STATE_CANT_OPEN,
1445 			    VDEV_AUX_CORRUPT_DATA);
1446 			error = EIO;
1447 			goto out;
1448 		}
1449 
1450 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1451 		spa_load_l2cache(spa);
1452 		spa_config_exit(spa, SCL_ALL, FTAG);
1453 	}
1454 
1455 	if (spa_check_logs(spa)) {
1456 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1457 		    VDEV_AUX_BAD_LOG);
1458 		error = ENXIO;
1459 		ereport = FM_EREPORT_ZFS_LOG_REPLAY;
1460 		goto out;
1461 	}
1462 
1463 
1464 	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
1465 
1466 	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1467 	    DMU_POOL_PROPS, sizeof (uint64_t), 1, &spa->spa_pool_props_object);
1468 
1469 	if (error && error != ENOENT) {
1470 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1471 		    VDEV_AUX_CORRUPT_DATA);
1472 		error = EIO;
1473 		goto out;
1474 	}
1475 
1476 	if (error == 0) {
1477 		(void) zap_lookup(spa->spa_meta_objset,
1478 		    spa->spa_pool_props_object,
1479 		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS),
1480 		    sizeof (uint64_t), 1, &spa->spa_bootfs);
1481 		(void) zap_lookup(spa->spa_meta_objset,
1482 		    spa->spa_pool_props_object,
1483 		    zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE),
1484 		    sizeof (uint64_t), 1, &autoreplace);
1485 		(void) zap_lookup(spa->spa_meta_objset,
1486 		    spa->spa_pool_props_object,
1487 		    zpool_prop_to_name(ZPOOL_PROP_DELEGATION),
1488 		    sizeof (uint64_t), 1, &spa->spa_delegation);
1489 		(void) zap_lookup(spa->spa_meta_objset,
1490 		    spa->spa_pool_props_object,
1491 		    zpool_prop_to_name(ZPOOL_PROP_FAILUREMODE),
1492 		    sizeof (uint64_t), 1, &spa->spa_failmode);
1493 	}
1494 
1495 	/*
1496 	 * If the 'autoreplace' property is set, then post a resource notifying
1497 	 * the ZFS DE that it should not issue any faults for unopenable
1498 	 * devices.  We also iterate over the vdevs, and post a sysevent for any
1499 	 * unopenable vdevs so that the normal autoreplace handler can take
1500 	 * over.
1501 	 */
1502 	if (autoreplace && state != SPA_LOAD_TRYIMPORT)
1503 		spa_check_removed(spa->spa_root_vdev);
1504 
1505 	/*
1506 	 * Load the vdev state for all toplevel vdevs.
1507 	 */
1508 	vdev_load(rvd);
1509 
1510 	/*
1511 	 * Propagate the leaf DTLs we just loaded all the way up the tree.
1512 	 */
1513 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1514 	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
1515 	spa_config_exit(spa, SCL_ALL, FTAG);
1516 
1517 	/*
1518 	 * Check the state of the root vdev.  If it can't be opened, it
1519 	 * indicates one or more toplevel vdevs are faulted.
1520 	 */
1521 	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
1522 		error = ENXIO;
1523 		goto out;
1524 	}
1525 
1526 	if (spa_writeable(spa)) {
1527 		dmu_tx_t *tx;
1528 		int need_update = B_FALSE;
1529 
1530 		ASSERT(state != SPA_LOAD_TRYIMPORT);
1531 
1532 		/*
1533 		 * Claim log blocks that haven't been committed yet.
1534 		 * This must all happen in a single txg.
1535 		 */
1536 		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
1537 		    spa_first_txg(spa));
1538 		(void) dmu_objset_find(spa_name(spa),
1539 		    zil_claim, tx, DS_FIND_CHILDREN);
1540 		dmu_tx_commit(tx);
1541 
1542 		spa->spa_sync_on = B_TRUE;
1543 		txg_sync_start(spa->spa_dsl_pool);
1544 
1545 		/*
1546 		 * Wait for all claims to sync.
1547 		 */
1548 		txg_wait_synced(spa->spa_dsl_pool, 0);
1549 
1550 		/*
1551 		 * If the config cache is stale, or we have uninitialized
1552 		 * metaslabs (see spa_vdev_add()), then update the config.
1553 		 */
1554 		if (config_cache_txg != spa->spa_config_txg ||
1555 		    state == SPA_LOAD_IMPORT)
1556 			need_update = B_TRUE;
1557 
1558 		for (int c = 0; c < rvd->vdev_children; c++)
1559 			if (rvd->vdev_child[c]->vdev_ms_array == 0)
1560 				need_update = B_TRUE;
1561 
1562 		/*
1563 		 * Update the config cache asychronously in case we're the
1564 		 * root pool, in which case the config cache isn't writable yet.
1565 		 */
1566 		if (need_update)
1567 			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
1568 
1569 		/*
1570 		 * Check all DTLs to see if anything needs resilvering.
1571 		 */
1572 		if (vdev_resilver_needed(rvd, NULL, NULL))
1573 			spa_async_request(spa, SPA_ASYNC_RESILVER);
1574 	}
1575 
1576 	error = 0;
1577 out:
1578 	spa->spa_minref = refcount_count(&spa->spa_refcount);
1579 	if (error && error != EBADF)
1580 		zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1581 	spa->spa_load_state = SPA_LOAD_NONE;
1582 	spa->spa_ena = 0;
1583 
1584 	return (error);
1585 }
1586 
1587 /*
1588  * Pool Open/Import
1589  *
1590  * The import case is identical to an open except that the configuration is sent
1591  * down from userland, instead of grabbed from the configuration cache.  For the
1592  * case of an open, the pool configuration will exist in the
1593  * POOL_STATE_UNINITIALIZED state.
1594  *
1595  * The stats information (gen/count/ustats) is used to gather vdev statistics at
1596  * the same time open the pool, without having to keep around the spa_t in some
1597  * ambiguous state.
1598  */
1599 static int
1600 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t **config)
1601 {
1602 	spa_t *spa;
1603 	int error;
1604 	int locked = B_FALSE;
1605 
1606 	*spapp = NULL;
1607 
1608 	/*
1609 	 * As disgusting as this is, we need to support recursive calls to this
1610 	 * function because dsl_dir_open() is called during spa_load(), and ends
1611 	 * up calling spa_open() again.  The real fix is to figure out how to
1612 	 * avoid dsl_dir_open() calling this in the first place.
1613 	 */
1614 	if (mutex_owner(&spa_namespace_lock) != curthread) {
1615 		mutex_enter(&spa_namespace_lock);
1616 		locked = B_TRUE;
1617 	}
1618 
1619 	if ((spa = spa_lookup(pool)) == NULL) {
1620 		if (locked)
1621 			mutex_exit(&spa_namespace_lock);
1622 		return (ENOENT);
1623 	}
1624 	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
1625 
1626 		spa_activate(spa, spa_mode_global);
1627 
1628 		error = spa_load(spa, spa->spa_config, SPA_LOAD_OPEN, B_FALSE);
1629 
1630 		if (error == EBADF) {
1631 			/*
1632 			 * If vdev_validate() returns failure (indicated by
1633 			 * EBADF), it indicates that one of the vdevs indicates
1634 			 * that the pool has been exported or destroyed.  If
1635 			 * this is the case, the config cache is out of sync and
1636 			 * we should remove the pool from the namespace.
1637 			 */
1638 			spa_unload(spa);
1639 			spa_deactivate(spa);
1640 			spa_config_sync(spa, B_TRUE, B_TRUE);
1641 			spa_remove(spa);
1642 			if (locked)
1643 				mutex_exit(&spa_namespace_lock);
1644 			return (ENOENT);
1645 		}
1646 
1647 		if (error) {
1648 			/*
1649 			 * We can't open the pool, but we still have useful
1650 			 * information: the state of each vdev after the
1651 			 * attempted vdev_open().  Return this to the user.
1652 			 */
1653 			if (config != NULL && spa->spa_root_vdev != NULL)
1654 				*config = spa_config_generate(spa, NULL, -1ULL,
1655 				    B_TRUE);
1656 			spa_unload(spa);
1657 			spa_deactivate(spa);
1658 			spa->spa_last_open_failed = B_TRUE;
1659 			if (locked)
1660 				mutex_exit(&spa_namespace_lock);
1661 			*spapp = NULL;
1662 			return (error);
1663 		} else {
1664 			spa->spa_last_open_failed = B_FALSE;
1665 		}
1666 	}
1667 
1668 	spa_open_ref(spa, tag);
1669 
1670 	if (locked)
1671 		mutex_exit(&spa_namespace_lock);
1672 
1673 	*spapp = spa;
1674 
1675 	if (config != NULL)
1676 		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
1677 
1678 	return (0);
1679 }
1680 
1681 int
1682 spa_open(const char *name, spa_t **spapp, void *tag)
1683 {
1684 	return (spa_open_common(name, spapp, tag, NULL));
1685 }
1686 
1687 /*
1688  * Lookup the given spa_t, incrementing the inject count in the process,
1689  * preventing it from being exported or destroyed.
1690  */
1691 spa_t *
1692 spa_inject_addref(char *name)
1693 {
1694 	spa_t *spa;
1695 
1696 	mutex_enter(&spa_namespace_lock);
1697 	if ((spa = spa_lookup(name)) == NULL) {
1698 		mutex_exit(&spa_namespace_lock);
1699 		return (NULL);
1700 	}
1701 	spa->spa_inject_ref++;
1702 	mutex_exit(&spa_namespace_lock);
1703 
1704 	return (spa);
1705 }
1706 
1707 void
1708 spa_inject_delref(spa_t *spa)
1709 {
1710 	mutex_enter(&spa_namespace_lock);
1711 	spa->spa_inject_ref--;
1712 	mutex_exit(&spa_namespace_lock);
1713 }
1714 
1715 /*
1716  * Add spares device information to the nvlist.
1717  */
1718 static void
1719 spa_add_spares(spa_t *spa, nvlist_t *config)
1720 {
1721 	nvlist_t **spares;
1722 	uint_t i, nspares;
1723 	nvlist_t *nvroot;
1724 	uint64_t guid;
1725 	vdev_stat_t *vs;
1726 	uint_t vsc;
1727 	uint64_t pool;
1728 
1729 	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
1730 
1731 	if (spa->spa_spares.sav_count == 0)
1732 		return;
1733 
1734 	VERIFY(nvlist_lookup_nvlist(config,
1735 	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1736 	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1737 	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1738 	if (nspares != 0) {
1739 		VERIFY(nvlist_add_nvlist_array(nvroot,
1740 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1741 		VERIFY(nvlist_lookup_nvlist_array(nvroot,
1742 		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1743 
1744 		/*
1745 		 * Go through and find any spares which have since been
1746 		 * repurposed as an active spare.  If this is the case, update
1747 		 * their status appropriately.
1748 		 */
1749 		for (i = 0; i < nspares; i++) {
1750 			VERIFY(nvlist_lookup_uint64(spares[i],
1751 			    ZPOOL_CONFIG_GUID, &guid) == 0);
1752 			if (spa_spare_exists(guid, &pool, NULL) &&
1753 			    pool != 0ULL) {
1754 				VERIFY(nvlist_lookup_uint64_array(
1755 				    spares[i], ZPOOL_CONFIG_STATS,
1756 				    (uint64_t **)&vs, &vsc) == 0);
1757 				vs->vs_state = VDEV_STATE_CANT_OPEN;
1758 				vs->vs_aux = VDEV_AUX_SPARED;
1759 			}
1760 		}
1761 	}
1762 }
1763 
1764 /*
1765  * Add l2cache device information to the nvlist, including vdev stats.
1766  */
1767 static void
1768 spa_add_l2cache(spa_t *spa, nvlist_t *config)
1769 {
1770 	nvlist_t **l2cache;
1771 	uint_t i, j, nl2cache;
1772 	nvlist_t *nvroot;
1773 	uint64_t guid;
1774 	vdev_t *vd;
1775 	vdev_stat_t *vs;
1776 	uint_t vsc;
1777 
1778 	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
1779 
1780 	if (spa->spa_l2cache.sav_count == 0)
1781 		return;
1782 
1783 	VERIFY(nvlist_lookup_nvlist(config,
1784 	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1785 	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
1786 	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1787 	if (nl2cache != 0) {
1788 		VERIFY(nvlist_add_nvlist_array(nvroot,
1789 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
1790 		VERIFY(nvlist_lookup_nvlist_array(nvroot,
1791 		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1792 
1793 		/*
1794 		 * Update level 2 cache device stats.
1795 		 */
1796 
1797 		for (i = 0; i < nl2cache; i++) {
1798 			VERIFY(nvlist_lookup_uint64(l2cache[i],
1799 			    ZPOOL_CONFIG_GUID, &guid) == 0);
1800 
1801 			vd = NULL;
1802 			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
1803 				if (guid ==
1804 				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
1805 					vd = spa->spa_l2cache.sav_vdevs[j];
1806 					break;
1807 				}
1808 			}
1809 			ASSERT(vd != NULL);
1810 
1811 			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
1812 			    ZPOOL_CONFIG_STATS, (uint64_t **)&vs, &vsc) == 0);
1813 			vdev_get_stats(vd, vs);
1814 		}
1815 	}
1816 }
1817 
1818 int
1819 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
1820 {
1821 	int error;
1822 	spa_t *spa;
1823 
1824 	*config = NULL;
1825 	error = spa_open_common(name, &spa, FTAG, config);
1826 
1827 	if (spa != NULL) {
1828 		/*
1829 		 * This still leaves a window of inconsistency where the spares
1830 		 * or l2cache devices could change and the config would be
1831 		 * self-inconsistent.
1832 		 */
1833 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
1834 
1835 		if (*config != NULL) {
1836 			VERIFY(nvlist_add_uint64(*config,
1837 			    ZPOOL_CONFIG_ERRCOUNT,
1838 			    spa_get_errlog_size(spa)) == 0);
1839 
1840 			if (spa_suspended(spa))
1841 				VERIFY(nvlist_add_uint64(*config,
1842 				    ZPOOL_CONFIG_SUSPENDED,
1843 				    spa->spa_failmode) == 0);
1844 
1845 			spa_add_spares(spa, *config);
1846 			spa_add_l2cache(spa, *config);
1847 		}
1848 	}
1849 
1850 	/*
1851 	 * We want to get the alternate root even for faulted pools, so we cheat
1852 	 * and call spa_lookup() directly.
1853 	 */
1854 	if (altroot) {
1855 		if (spa == NULL) {
1856 			mutex_enter(&spa_namespace_lock);
1857 			spa = spa_lookup(name);
1858 			if (spa)
1859 				spa_altroot(spa, altroot, buflen);
1860 			else
1861 				altroot[0] = '\0';
1862 			spa = NULL;
1863 			mutex_exit(&spa_namespace_lock);
1864 		} else {
1865 			spa_altroot(spa, altroot, buflen);
1866 		}
1867 	}
1868 
1869 	if (spa != NULL) {
1870 		spa_config_exit(spa, SCL_CONFIG, FTAG);
1871 		spa_close(spa, FTAG);
1872 	}
1873 
1874 	return (error);
1875 }
1876 
1877 /*
1878  * Validate that the auxiliary device array is well formed.  We must have an
1879  * array of nvlists, each which describes a valid leaf vdev.  If this is an
1880  * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
1881  * specified, as long as they are well-formed.
1882  */
1883 static int
1884 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
1885     spa_aux_vdev_t *sav, const char *config, uint64_t version,
1886     vdev_labeltype_t label)
1887 {
1888 	nvlist_t **dev;
1889 	uint_t i, ndev;
1890 	vdev_t *vd;
1891 	int error;
1892 
1893 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1894 
1895 	/*
1896 	 * It's acceptable to have no devs specified.
1897 	 */
1898 	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
1899 		return (0);
1900 
1901 	if (ndev == 0)
1902 		return (EINVAL);
1903 
1904 	/*
1905 	 * Make sure the pool is formatted with a version that supports this
1906 	 * device type.
1907 	 */
1908 	if (spa_version(spa) < version)
1909 		return (ENOTSUP);
1910 
1911 	/*
1912 	 * Set the pending device list so we correctly handle device in-use
1913 	 * checking.
1914 	 */
1915 	sav->sav_pending = dev;
1916 	sav->sav_npending = ndev;
1917 
1918 	for (i = 0; i < ndev; i++) {
1919 		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
1920 		    mode)) != 0)
1921 			goto out;
1922 
1923 		if (!vd->vdev_ops->vdev_op_leaf) {
1924 			vdev_free(vd);
1925 			error = EINVAL;
1926 			goto out;
1927 		}
1928 
1929 		/*
1930 		 * The L2ARC currently only supports disk devices in
1931 		 * kernel context.  For user-level testing, we allow it.
1932 		 */
1933 #ifdef _KERNEL
1934 		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
1935 		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
1936 			error = ENOTBLK;
1937 			goto out;
1938 		}
1939 #endif
1940 		vd->vdev_top = vd;
1941 
1942 		if ((error = vdev_open(vd)) == 0 &&
1943 		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
1944 			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
1945 			    vd->vdev_guid) == 0);
1946 		}
1947 
1948 		vdev_free(vd);
1949 
1950 		if (error &&
1951 		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
1952 			goto out;
1953 		else
1954 			error = 0;
1955 	}
1956 
1957 out:
1958 	sav->sav_pending = NULL;
1959 	sav->sav_npending = 0;
1960 	return (error);
1961 }
1962 
1963 static int
1964 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
1965 {
1966 	int error;
1967 
1968 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1969 
1970 	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
1971 	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
1972 	    VDEV_LABEL_SPARE)) != 0) {
1973 		return (error);
1974 	}
1975 
1976 	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
1977 	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
1978 	    VDEV_LABEL_L2CACHE));
1979 }
1980 
1981 static void
1982 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
1983     const char *config)
1984 {
1985 	int i;
1986 
1987 	if (sav->sav_config != NULL) {
1988 		nvlist_t **olddevs;
1989 		uint_t oldndevs;
1990 		nvlist_t **newdevs;
1991 
1992 		/*
1993 		 * Generate new dev list by concatentating with the
1994 		 * current dev list.
1995 		 */
1996 		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
1997 		    &olddevs, &oldndevs) == 0);
1998 
1999 		newdevs = kmem_alloc(sizeof (void *) *
2000 		    (ndevs + oldndevs), KM_SLEEP);
2001 		for (i = 0; i < oldndevs; i++)
2002 			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2003 			    KM_SLEEP) == 0);
2004 		for (i = 0; i < ndevs; i++)
2005 			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2006 			    KM_SLEEP) == 0);
2007 
2008 		VERIFY(nvlist_remove(sav->sav_config, config,
2009 		    DATA_TYPE_NVLIST_ARRAY) == 0);
2010 
2011 		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2012 		    config, newdevs, ndevs + oldndevs) == 0);
2013 		for (i = 0; i < oldndevs + ndevs; i++)
2014 			nvlist_free(newdevs[i]);
2015 		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2016 	} else {
2017 		/*
2018 		 * Generate a new dev list.
2019 		 */
2020 		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2021 		    KM_SLEEP) == 0);
2022 		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2023 		    devs, ndevs) == 0);
2024 	}
2025 }
2026 
2027 /*
2028  * Stop and drop level 2 ARC devices
2029  */
2030 void
2031 spa_l2cache_drop(spa_t *spa)
2032 {
2033 	vdev_t *vd;
2034 	int i;
2035 	spa_aux_vdev_t *sav = &spa->spa_l2cache;
2036 
2037 	for (i = 0; i < sav->sav_count; i++) {
2038 		uint64_t pool;
2039 
2040 		vd = sav->sav_vdevs[i];
2041 		ASSERT(vd != NULL);
2042 
2043 		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2044 		    pool != 0ULL && l2arc_vdev_present(vd))
2045 			l2arc_remove_vdev(vd);
2046 		if (vd->vdev_isl2cache)
2047 			spa_l2cache_remove(vd);
2048 		vdev_clear_stats(vd);
2049 		(void) vdev_close(vd);
2050 	}
2051 }
2052 
2053 /*
2054  * Pool Creation
2055  */
2056 int
2057 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2058     const char *history_str, nvlist_t *zplprops)
2059 {
2060 	spa_t *spa;
2061 	char *altroot = NULL;
2062 	vdev_t *rvd;
2063 	dsl_pool_t *dp;
2064 	dmu_tx_t *tx;
2065 	int c, error = 0;
2066 	uint64_t txg = TXG_INITIAL;
2067 	nvlist_t **spares, **l2cache;
2068 	uint_t nspares, nl2cache;
2069 	uint64_t version;
2070 
2071 	/*
2072 	 * If this pool already exists, return failure.
2073 	 */
2074 	mutex_enter(&spa_namespace_lock);
2075 	if (spa_lookup(pool) != NULL) {
2076 		mutex_exit(&spa_namespace_lock);
2077 		return (EEXIST);
2078 	}
2079 
2080 	/*
2081 	 * Allocate a new spa_t structure.
2082 	 */
2083 	(void) nvlist_lookup_string(props,
2084 	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2085 	spa = spa_add(pool, altroot);
2086 	spa_activate(spa, spa_mode_global);
2087 
2088 	spa->spa_uberblock.ub_txg = txg - 1;
2089 
2090 	if (props && (error = spa_prop_validate(spa, props))) {
2091 		spa_deactivate(spa);
2092 		spa_remove(spa);
2093 		mutex_exit(&spa_namespace_lock);
2094 		return (error);
2095 	}
2096 
2097 	if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2098 	    &version) != 0)
2099 		version = SPA_VERSION;
2100 	ASSERT(version <= SPA_VERSION);
2101 	spa->spa_uberblock.ub_version = version;
2102 	spa->spa_ubsync = spa->spa_uberblock;
2103 
2104 	/*
2105 	 * Create "The Godfather" zio to hold all async IOs
2106 	 */
2107 	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2108 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2109 
2110 	/*
2111 	 * Create the root vdev.
2112 	 */
2113 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2114 
2115 	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2116 
2117 	ASSERT(error != 0 || rvd != NULL);
2118 	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2119 
2120 	if (error == 0 && !zfs_allocatable_devs(nvroot))
2121 		error = EINVAL;
2122 
2123 	if (error == 0 &&
2124 	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2125 	    (error = spa_validate_aux(spa, nvroot, txg,
2126 	    VDEV_ALLOC_ADD)) == 0) {
2127 		for (c = 0; c < rvd->vdev_children; c++)
2128 			vdev_init(rvd->vdev_child[c], txg);
2129 		vdev_config_dirty(rvd);
2130 	}
2131 
2132 	spa_config_exit(spa, SCL_ALL, FTAG);
2133 
2134 	if (error != 0) {
2135 		spa_unload(spa);
2136 		spa_deactivate(spa);
2137 		spa_remove(spa);
2138 		mutex_exit(&spa_namespace_lock);
2139 		return (error);
2140 	}
2141 
2142 	/*
2143 	 * Get the list of spares, if specified.
2144 	 */
2145 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2146 	    &spares, &nspares) == 0) {
2147 		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
2148 		    KM_SLEEP) == 0);
2149 		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2150 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2151 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2152 		spa_load_spares(spa);
2153 		spa_config_exit(spa, SCL_ALL, FTAG);
2154 		spa->spa_spares.sav_sync = B_TRUE;
2155 	}
2156 
2157 	/*
2158 	 * Get the list of level 2 cache devices, if specified.
2159 	 */
2160 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2161 	    &l2cache, &nl2cache) == 0) {
2162 		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2163 		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2164 		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2165 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2166 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2167 		spa_load_l2cache(spa);
2168 		spa_config_exit(spa, SCL_ALL, FTAG);
2169 		spa->spa_l2cache.sav_sync = B_TRUE;
2170 	}
2171 
2172 	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
2173 	spa->spa_meta_objset = dp->dp_meta_objset;
2174 
2175 	tx = dmu_tx_create_assigned(dp, txg);
2176 
2177 	/*
2178 	 * Create the pool config object.
2179 	 */
2180 	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
2181 	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
2182 	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2183 
2184 	if (zap_add(spa->spa_meta_objset,
2185 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
2186 	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
2187 		cmn_err(CE_PANIC, "failed to add pool config");
2188 	}
2189 
2190 	/* Newly created pools with the right version are always deflated. */
2191 	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
2192 		spa->spa_deflate = TRUE;
2193 		if (zap_add(spa->spa_meta_objset,
2194 		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
2195 		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
2196 			cmn_err(CE_PANIC, "failed to add deflate");
2197 		}
2198 	}
2199 
2200 	/*
2201 	 * Create the deferred-free bplist object.  Turn off compression
2202 	 * because sync-to-convergence takes longer if the blocksize
2203 	 * keeps changing.
2204 	 */
2205 	spa->spa_sync_bplist_obj = bplist_create(spa->spa_meta_objset,
2206 	    1 << 14, tx);
2207 	dmu_object_set_compress(spa->spa_meta_objset, spa->spa_sync_bplist_obj,
2208 	    ZIO_COMPRESS_OFF, tx);
2209 
2210 	if (zap_add(spa->spa_meta_objset,
2211 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
2212 	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj, tx) != 0) {
2213 		cmn_err(CE_PANIC, "failed to add bplist");
2214 	}
2215 
2216 	/*
2217 	 * Create the pool's history object.
2218 	 */
2219 	if (version >= SPA_VERSION_ZPOOL_HISTORY)
2220 		spa_history_create_obj(spa, tx);
2221 
2222 	/*
2223 	 * Set pool properties.
2224 	 */
2225 	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
2226 	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2227 	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
2228 	if (props != NULL) {
2229 		spa_configfile_set(spa, props, B_FALSE);
2230 		spa_sync_props(spa, props, CRED(), tx);
2231 	}
2232 
2233 	dmu_tx_commit(tx);
2234 
2235 	spa->spa_sync_on = B_TRUE;
2236 	txg_sync_start(spa->spa_dsl_pool);
2237 
2238 	/*
2239 	 * We explicitly wait for the first transaction to complete so that our
2240 	 * bean counters are appropriately updated.
2241 	 */
2242 	txg_wait_synced(spa->spa_dsl_pool, txg);
2243 
2244 	spa_config_sync(spa, B_FALSE, B_TRUE);
2245 
2246 	if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
2247 		(void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
2248 
2249 	spa->spa_minref = refcount_count(&spa->spa_refcount);
2250 
2251 	mutex_exit(&spa_namespace_lock);
2252 
2253 	return (0);
2254 }
2255 
2256 #ifdef _KERNEL
2257 /*
2258  * Build a "root" vdev for a top level vdev read in from a rootpool
2259  * device label.
2260  */
2261 static void
2262 spa_build_rootpool_config(nvlist_t *config)
2263 {
2264 	nvlist_t *nvtop, *nvroot;
2265 	uint64_t pgid;
2266 
2267 	/*
2268 	 * Add this top-level vdev to the child array.
2269 	 */
2270 	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtop)
2271 	    == 0);
2272 	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pgid)
2273 	    == 0);
2274 
2275 	/*
2276 	 * Put this pool's top-level vdevs into a root vdev.
2277 	 */
2278 	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2279 	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT)
2280 	    == 0);
2281 	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
2282 	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
2283 	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2284 	    &nvtop, 1) == 0);
2285 
2286 	/*
2287 	 * Replace the existing vdev_tree with the new root vdev in
2288 	 * this pool's configuration (remove the old, add the new).
2289 	 */
2290 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
2291 	nvlist_free(nvroot);
2292 }
2293 
2294 /*
2295  * Get the root pool information from the root disk, then import the root pool
2296  * during the system boot up time.
2297  */
2298 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
2299 
2300 int
2301 spa_check_rootconf(char *devpath, char *devid, nvlist_t **bestconf,
2302     uint64_t *besttxg)
2303 {
2304 	nvlist_t *config;
2305 	uint64_t txg;
2306 	int error;
2307 
2308 	if (error = vdev_disk_read_rootlabel(devpath, devid, &config))
2309 		return (error);
2310 
2311 	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
2312 
2313 	if (bestconf != NULL)
2314 		*bestconf = config;
2315 	else
2316 		nvlist_free(config);
2317 	*besttxg = txg;
2318 	return (0);
2319 }
2320 
2321 boolean_t
2322 spa_rootdev_validate(nvlist_t *nv)
2323 {
2324 	uint64_t ival;
2325 
2326 	if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
2327 	    nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
2328 	    nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
2329 		return (B_FALSE);
2330 
2331 	return (B_TRUE);
2332 }
2333 
2334 
2335 /*
2336  * Given the boot device's physical path or devid, check if the device
2337  * is in a valid state.  If so, return the configuration from the vdev
2338  * label.
2339  */
2340 int
2341 spa_get_rootconf(char *devpath, char *devid, nvlist_t **bestconf)
2342 {
2343 	nvlist_t *conf = NULL;
2344 	uint64_t txg = 0;
2345 	nvlist_t *nvtop, **child;
2346 	char *type;
2347 	char *bootpath = NULL;
2348 	uint_t children, c;
2349 	char *tmp;
2350 	int error;
2351 
2352 	if (devpath && ((tmp = strchr(devpath, ' ')) != NULL))
2353 		*tmp = '\0';
2354 	if (error = spa_check_rootconf(devpath, devid, &conf, &txg)) {
2355 		cmn_err(CE_NOTE, "error reading device label");
2356 		return (error);
2357 	}
2358 	if (txg == 0) {
2359 		cmn_err(CE_NOTE, "this device is detached");
2360 		nvlist_free(conf);
2361 		return (EINVAL);
2362 	}
2363 
2364 	VERIFY(nvlist_lookup_nvlist(conf, ZPOOL_CONFIG_VDEV_TREE,
2365 	    &nvtop) == 0);
2366 	VERIFY(nvlist_lookup_string(nvtop, ZPOOL_CONFIG_TYPE, &type) == 0);
2367 
2368 	if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2369 		if (spa_rootdev_validate(nvtop)) {
2370 			goto out;
2371 		} else {
2372 			nvlist_free(conf);
2373 			return (EINVAL);
2374 		}
2375 	}
2376 
2377 	ASSERT(strcmp(type, VDEV_TYPE_MIRROR) == 0);
2378 
2379 	VERIFY(nvlist_lookup_nvlist_array(nvtop, ZPOOL_CONFIG_CHILDREN,
2380 	    &child, &children) == 0);
2381 
2382 	/*
2383 	 * Go thru vdevs in the mirror to see if the given device
2384 	 * has the most recent txg. Only the device with the most
2385 	 * recent txg has valid information and should be booted.
2386 	 */
2387 	for (c = 0; c < children; c++) {
2388 		char *cdevid, *cpath;
2389 		uint64_t tmptxg;
2390 
2391 		cpath = NULL;
2392 		cdevid = NULL;
2393 		(void) nvlist_lookup_string(child[c], ZPOOL_CONFIG_PHYS_PATH,
2394 		    &cpath);
2395 		(void) nvlist_lookup_string(child[c], ZPOOL_CONFIG_DEVID,
2396 		    &cdevid);
2397 		if (cpath == NULL && cdevid == NULL)
2398 			return (EINVAL);
2399 		if ((spa_check_rootconf(cpath, cdevid, NULL,
2400 		    &tmptxg) == 0) && (tmptxg > txg)) {
2401 			txg = tmptxg;
2402 			VERIFY(nvlist_lookup_string(child[c],
2403 			    ZPOOL_CONFIG_PATH, &bootpath) == 0);
2404 		}
2405 	}
2406 
2407 	/* Does the best device match the one we've booted from? */
2408 	if (bootpath) {
2409 		cmn_err(CE_NOTE, "try booting from '%s'", bootpath);
2410 		return (EINVAL);
2411 	}
2412 out:
2413 	*bestconf = conf;
2414 	return (0);
2415 }
2416 
2417 /*
2418  * Import a root pool.
2419  *
2420  * For x86. devpath_list will consist of devid and/or physpath name of
2421  * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
2422  * The GRUB "findroot" command will return the vdev we should boot.
2423  *
2424  * For Sparc, devpath_list consists the physpath name of the booting device
2425  * no matter the rootpool is a single device pool or a mirrored pool.
2426  * e.g.
2427  *	"/pci@1f,0/ide@d/disk@0,0:a"
2428  */
2429 int
2430 spa_import_rootpool(char *devpath, char *devid)
2431 {
2432 	nvlist_t *conf = NULL;
2433 	char *pname;
2434 	int error;
2435 	spa_t *spa;
2436 
2437 	/*
2438 	 * Get the vdev pathname and configuation from the most
2439 	 * recently updated vdev (highest txg).
2440 	 */
2441 	if (error = spa_get_rootconf(devpath, devid, &conf))
2442 		goto msg_out;
2443 
2444 	/*
2445 	 * Add type "root" vdev to the config.
2446 	 */
2447 	spa_build_rootpool_config(conf);
2448 
2449 	VERIFY(nvlist_lookup_string(conf, ZPOOL_CONFIG_POOL_NAME, &pname) == 0);
2450 
2451 	mutex_enter(&spa_namespace_lock);
2452 	if ((spa = spa_lookup(pname)) != NULL) {
2453 		/*
2454 		 * Remove the existing root pool from the namespace so that we
2455 		 * can replace it with the correct config we just read in.
2456 		 */
2457 		spa_remove(spa);
2458 	}
2459 
2460 	spa = spa_add(pname, NULL);
2461 
2462 	spa->spa_is_root = B_TRUE;
2463 	VERIFY(nvlist_dup(conf, &spa->spa_config, 0) == 0);
2464 	mutex_exit(&spa_namespace_lock);
2465 
2466 	nvlist_free(conf);
2467 	return (0);
2468 
2469 msg_out:
2470 	cmn_err(CE_NOTE, "\n"
2471 	    "  ***************************************************  \n"
2472 	    "  *  This device is not bootable!                   *  \n"
2473 	    "  *  It is either offlined or detached or faulted.  *  \n"
2474 	    "  *  Please try to boot from a different device.    *  \n"
2475 	    "  ***************************************************  ");
2476 
2477 	return (error);
2478 }
2479 #endif
2480 
2481 /*
2482  * Take a pool and insert it into the namespace as if it had been loaded at
2483  * boot.
2484  */
2485 int
2486 spa_import_verbatim(const char *pool, nvlist_t *config, nvlist_t *props)
2487 {
2488 	spa_t *spa;
2489 	char *altroot = NULL;
2490 
2491 	mutex_enter(&spa_namespace_lock);
2492 	if (spa_lookup(pool) != NULL) {
2493 		mutex_exit(&spa_namespace_lock);
2494 		return (EEXIST);
2495 	}
2496 
2497 	(void) nvlist_lookup_string(props,
2498 	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2499 	spa = spa_add(pool, altroot);
2500 
2501 	VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
2502 
2503 	if (props != NULL)
2504 		spa_configfile_set(spa, props, B_FALSE);
2505 
2506 	spa_config_sync(spa, B_FALSE, B_TRUE);
2507 
2508 	mutex_exit(&spa_namespace_lock);
2509 
2510 	return (0);
2511 }
2512 
2513 /*
2514  * Import a non-root pool into the system.
2515  */
2516 int
2517 spa_import(const char *pool, nvlist_t *config, nvlist_t *props)
2518 {
2519 	spa_t *spa;
2520 	char *altroot = NULL;
2521 	int error;
2522 	nvlist_t *nvroot;
2523 	nvlist_t **spares, **l2cache;
2524 	uint_t nspares, nl2cache;
2525 
2526 	/*
2527 	 * If a pool with this name exists, return failure.
2528 	 */
2529 	mutex_enter(&spa_namespace_lock);
2530 	if ((spa = spa_lookup(pool)) != NULL) {
2531 		mutex_exit(&spa_namespace_lock);
2532 		return (EEXIST);
2533 	}
2534 
2535 	/*
2536 	 * Create and initialize the spa structure.
2537 	 */
2538 	(void) nvlist_lookup_string(props,
2539 	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2540 	spa = spa_add(pool, altroot);
2541 	spa_activate(spa, spa_mode_global);
2542 
2543 	/*
2544 	 * Don't start async tasks until we know everything is healthy.
2545 	 */
2546 	spa_async_suspend(spa);
2547 
2548 	/*
2549 	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
2550 	 * because the user-supplied config is actually the one to trust when
2551 	 * doing an import.
2552 	 */
2553 	error = spa_load(spa, config, SPA_LOAD_IMPORT, B_TRUE);
2554 
2555 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2556 	/*
2557 	 * Toss any existing sparelist, as it doesn't have any validity
2558 	 * anymore, and conflicts with spa_has_spare().
2559 	 */
2560 	if (spa->spa_spares.sav_config) {
2561 		nvlist_free(spa->spa_spares.sav_config);
2562 		spa->spa_spares.sav_config = NULL;
2563 		spa_load_spares(spa);
2564 	}
2565 	if (spa->spa_l2cache.sav_config) {
2566 		nvlist_free(spa->spa_l2cache.sav_config);
2567 		spa->spa_l2cache.sav_config = NULL;
2568 		spa_load_l2cache(spa);
2569 	}
2570 
2571 	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2572 	    &nvroot) == 0);
2573 	if (error == 0)
2574 		error = spa_validate_aux(spa, nvroot, -1ULL,
2575 		    VDEV_ALLOC_SPARE);
2576 	if (error == 0)
2577 		error = spa_validate_aux(spa, nvroot, -1ULL,
2578 		    VDEV_ALLOC_L2CACHE);
2579 	spa_config_exit(spa, SCL_ALL, FTAG);
2580 
2581 	if (props != NULL)
2582 		spa_configfile_set(spa, props, B_FALSE);
2583 
2584 	if (error != 0 || (props && spa_writeable(spa) &&
2585 	    (error = spa_prop_set(spa, props)))) {
2586 		spa_unload(spa);
2587 		spa_deactivate(spa);
2588 		spa_remove(spa);
2589 		mutex_exit(&spa_namespace_lock);
2590 		return (error);
2591 	}
2592 
2593 	spa_async_resume(spa);
2594 
2595 	/*
2596 	 * Override any spares and level 2 cache devices as specified by
2597 	 * the user, as these may have correct device names/devids, etc.
2598 	 */
2599 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2600 	    &spares, &nspares) == 0) {
2601 		if (spa->spa_spares.sav_config)
2602 			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
2603 			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
2604 		else
2605 			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
2606 			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2607 		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2608 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2609 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2610 		spa_load_spares(spa);
2611 		spa_config_exit(spa, SCL_ALL, FTAG);
2612 		spa->spa_spares.sav_sync = B_TRUE;
2613 	}
2614 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2615 	    &l2cache, &nl2cache) == 0) {
2616 		if (spa->spa_l2cache.sav_config)
2617 			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
2618 			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
2619 		else
2620 			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2621 			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2622 		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2623 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2624 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2625 		spa_load_l2cache(spa);
2626 		spa_config_exit(spa, SCL_ALL, FTAG);
2627 		spa->spa_l2cache.sav_sync = B_TRUE;
2628 	}
2629 
2630 	if (spa_writeable(spa)) {
2631 		/*
2632 		 * Update the config cache to include the newly-imported pool.
2633 		 */
2634 		spa_config_update_common(spa, SPA_CONFIG_UPDATE_POOL, B_FALSE);
2635 	}
2636 
2637 	mutex_exit(&spa_namespace_lock);
2638 
2639 	return (0);
2640 }
2641 
2642 
2643 /*
2644  * This (illegal) pool name is used when temporarily importing a spa_t in order
2645  * to get the vdev stats associated with the imported devices.
2646  */
2647 #define	TRYIMPORT_NAME	"$import"
2648 
2649 nvlist_t *
2650 spa_tryimport(nvlist_t *tryconfig)
2651 {
2652 	nvlist_t *config = NULL;
2653 	char *poolname;
2654 	spa_t *spa;
2655 	uint64_t state;
2656 	int error;
2657 
2658 	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
2659 		return (NULL);
2660 
2661 	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
2662 		return (NULL);
2663 
2664 	/*
2665 	 * Create and initialize the spa structure.
2666 	 */
2667 	mutex_enter(&spa_namespace_lock);
2668 	spa = spa_add(TRYIMPORT_NAME, NULL);
2669 	spa_activate(spa, FREAD);
2670 
2671 	/*
2672 	 * Pass off the heavy lifting to spa_load().
2673 	 * Pass TRUE for mosconfig because the user-supplied config
2674 	 * is actually the one to trust when doing an import.
2675 	 */
2676 	error = spa_load(spa, tryconfig, SPA_LOAD_TRYIMPORT, B_TRUE);
2677 
2678 	/*
2679 	 * If 'tryconfig' was at least parsable, return the current config.
2680 	 */
2681 	if (spa->spa_root_vdev != NULL) {
2682 		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2683 		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
2684 		    poolname) == 0);
2685 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
2686 		    state) == 0);
2687 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
2688 		    spa->spa_uberblock.ub_timestamp) == 0);
2689 
2690 		/*
2691 		 * If the bootfs property exists on this pool then we
2692 		 * copy it out so that external consumers can tell which
2693 		 * pools are bootable.
2694 		 */
2695 		if ((!error || error == EEXIST) && spa->spa_bootfs) {
2696 			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2697 
2698 			/*
2699 			 * We have to play games with the name since the
2700 			 * pool was opened as TRYIMPORT_NAME.
2701 			 */
2702 			if (dsl_dsobj_to_dsname(spa_name(spa),
2703 			    spa->spa_bootfs, tmpname) == 0) {
2704 				char *cp;
2705 				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2706 
2707 				cp = strchr(tmpname, '/');
2708 				if (cp == NULL) {
2709 					(void) strlcpy(dsname, tmpname,
2710 					    MAXPATHLEN);
2711 				} else {
2712 					(void) snprintf(dsname, MAXPATHLEN,
2713 					    "%s/%s", poolname, ++cp);
2714 				}
2715 				VERIFY(nvlist_add_string(config,
2716 				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
2717 				kmem_free(dsname, MAXPATHLEN);
2718 			}
2719 			kmem_free(tmpname, MAXPATHLEN);
2720 		}
2721 
2722 		/*
2723 		 * Add the list of hot spares and level 2 cache devices.
2724 		 */
2725 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2726 		spa_add_spares(spa, config);
2727 		spa_add_l2cache(spa, config);
2728 		spa_config_exit(spa, SCL_CONFIG, FTAG);
2729 	}
2730 
2731 	spa_unload(spa);
2732 	spa_deactivate(spa);
2733 	spa_remove(spa);
2734 	mutex_exit(&spa_namespace_lock);
2735 
2736 	return (config);
2737 }
2738 
2739 /*
2740  * Pool export/destroy
2741  *
2742  * The act of destroying or exporting a pool is very simple.  We make sure there
2743  * is no more pending I/O and any references to the pool are gone.  Then, we
2744  * update the pool state and sync all the labels to disk, removing the
2745  * configuration from the cache afterwards. If the 'hardforce' flag is set, then
2746  * we don't sync the labels or remove the configuration cache.
2747  */
2748 static int
2749 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
2750     boolean_t force, boolean_t hardforce)
2751 {
2752 	spa_t *spa;
2753 
2754 	if (oldconfig)
2755 		*oldconfig = NULL;
2756 
2757 	if (!(spa_mode_global & FWRITE))
2758 		return (EROFS);
2759 
2760 	mutex_enter(&spa_namespace_lock);
2761 	if ((spa = spa_lookup(pool)) == NULL) {
2762 		mutex_exit(&spa_namespace_lock);
2763 		return (ENOENT);
2764 	}
2765 
2766 	/*
2767 	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
2768 	 * reacquire the namespace lock, and see if we can export.
2769 	 */
2770 	spa_open_ref(spa, FTAG);
2771 	mutex_exit(&spa_namespace_lock);
2772 	spa_async_suspend(spa);
2773 	mutex_enter(&spa_namespace_lock);
2774 	spa_close(spa, FTAG);
2775 
2776 	/*
2777 	 * The pool will be in core if it's openable,
2778 	 * in which case we can modify its state.
2779 	 */
2780 	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
2781 		/*
2782 		 * Objsets may be open only because they're dirty, so we
2783 		 * have to force it to sync before checking spa_refcnt.
2784 		 */
2785 		txg_wait_synced(spa->spa_dsl_pool, 0);
2786 
2787 		/*
2788 		 * A pool cannot be exported or destroyed if there are active
2789 		 * references.  If we are resetting a pool, allow references by
2790 		 * fault injection handlers.
2791 		 */
2792 		if (!spa_refcount_zero(spa) ||
2793 		    (spa->spa_inject_ref != 0 &&
2794 		    new_state != POOL_STATE_UNINITIALIZED)) {
2795 			spa_async_resume(spa);
2796 			mutex_exit(&spa_namespace_lock);
2797 			return (EBUSY);
2798 		}
2799 
2800 		/*
2801 		 * A pool cannot be exported if it has an active shared spare.
2802 		 * This is to prevent other pools stealing the active spare
2803 		 * from an exported pool. At user's own will, such pool can
2804 		 * be forcedly exported.
2805 		 */
2806 		if (!force && new_state == POOL_STATE_EXPORTED &&
2807 		    spa_has_active_shared_spare(spa)) {
2808 			spa_async_resume(spa);
2809 			mutex_exit(&spa_namespace_lock);
2810 			return (EXDEV);
2811 		}
2812 
2813 		/*
2814 		 * We want this to be reflected on every label,
2815 		 * so mark them all dirty.  spa_unload() will do the
2816 		 * final sync that pushes these changes out.
2817 		 */
2818 		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
2819 			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2820 			spa->spa_state = new_state;
2821 			spa->spa_final_txg = spa_last_synced_txg(spa) + 1;
2822 			vdev_config_dirty(spa->spa_root_vdev);
2823 			spa_config_exit(spa, SCL_ALL, FTAG);
2824 		}
2825 	}
2826 
2827 	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
2828 
2829 	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
2830 		spa_unload(spa);
2831 		spa_deactivate(spa);
2832 	}
2833 
2834 	if (oldconfig && spa->spa_config)
2835 		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
2836 
2837 	if (new_state != POOL_STATE_UNINITIALIZED) {
2838 		if (!hardforce)
2839 			spa_config_sync(spa, B_TRUE, B_TRUE);
2840 		spa_remove(spa);
2841 	}
2842 	mutex_exit(&spa_namespace_lock);
2843 
2844 	return (0);
2845 }
2846 
2847 /*
2848  * Destroy a storage pool.
2849  */
2850 int
2851 spa_destroy(char *pool)
2852 {
2853 	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
2854 	    B_FALSE, B_FALSE));
2855 }
2856 
2857 /*
2858  * Export a storage pool.
2859  */
2860 int
2861 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
2862     boolean_t hardforce)
2863 {
2864 	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
2865 	    force, hardforce));
2866 }
2867 
2868 /*
2869  * Similar to spa_export(), this unloads the spa_t without actually removing it
2870  * from the namespace in any way.
2871  */
2872 int
2873 spa_reset(char *pool)
2874 {
2875 	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
2876 	    B_FALSE, B_FALSE));
2877 }
2878 
2879 /*
2880  * ==========================================================================
2881  * Device manipulation
2882  * ==========================================================================
2883  */
2884 
2885 /*
2886  * Add a device to a storage pool.
2887  */
2888 int
2889 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
2890 {
2891 	uint64_t txg;
2892 	int error;
2893 	vdev_t *rvd = spa->spa_root_vdev;
2894 	vdev_t *vd, *tvd;
2895 	nvlist_t **spares, **l2cache;
2896 	uint_t nspares, nl2cache;
2897 
2898 	txg = spa_vdev_enter(spa);
2899 
2900 	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
2901 	    VDEV_ALLOC_ADD)) != 0)
2902 		return (spa_vdev_exit(spa, NULL, txg, error));
2903 
2904 	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
2905 
2906 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
2907 	    &nspares) != 0)
2908 		nspares = 0;
2909 
2910 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
2911 	    &nl2cache) != 0)
2912 		nl2cache = 0;
2913 
2914 	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
2915 		return (spa_vdev_exit(spa, vd, txg, EINVAL));
2916 
2917 	if (vd->vdev_children != 0 &&
2918 	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
2919 		return (spa_vdev_exit(spa, vd, txg, error));
2920 
2921 	/*
2922 	 * We must validate the spares and l2cache devices after checking the
2923 	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
2924 	 */
2925 	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
2926 		return (spa_vdev_exit(spa, vd, txg, error));
2927 
2928 	/*
2929 	 * Transfer each new top-level vdev from vd to rvd.
2930 	 */
2931 	for (int c = 0; c < vd->vdev_children; c++) {
2932 		tvd = vd->vdev_child[c];
2933 		vdev_remove_child(vd, tvd);
2934 		tvd->vdev_id = rvd->vdev_children;
2935 		vdev_add_child(rvd, tvd);
2936 		vdev_config_dirty(tvd);
2937 	}
2938 
2939 	if (nspares != 0) {
2940 		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
2941 		    ZPOOL_CONFIG_SPARES);
2942 		spa_load_spares(spa);
2943 		spa->spa_spares.sav_sync = B_TRUE;
2944 	}
2945 
2946 	if (nl2cache != 0) {
2947 		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
2948 		    ZPOOL_CONFIG_L2CACHE);
2949 		spa_load_l2cache(spa);
2950 		spa->spa_l2cache.sav_sync = B_TRUE;
2951 	}
2952 
2953 	/*
2954 	 * We have to be careful when adding new vdevs to an existing pool.
2955 	 * If other threads start allocating from these vdevs before we
2956 	 * sync the config cache, and we lose power, then upon reboot we may
2957 	 * fail to open the pool because there are DVAs that the config cache
2958 	 * can't translate.  Therefore, we first add the vdevs without
2959 	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
2960 	 * and then let spa_config_update() initialize the new metaslabs.
2961 	 *
2962 	 * spa_load() checks for added-but-not-initialized vdevs, so that
2963 	 * if we lose power at any point in this sequence, the remaining
2964 	 * steps will be completed the next time we load the pool.
2965 	 */
2966 	(void) spa_vdev_exit(spa, vd, txg, 0);
2967 
2968 	mutex_enter(&spa_namespace_lock);
2969 	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
2970 	mutex_exit(&spa_namespace_lock);
2971 
2972 	return (0);
2973 }
2974 
2975 /*
2976  * Attach a device to a mirror.  The arguments are the path to any device
2977  * in the mirror, and the nvroot for the new device.  If the path specifies
2978  * a device that is not mirrored, we automatically insert the mirror vdev.
2979  *
2980  * If 'replacing' is specified, the new device is intended to replace the
2981  * existing device; in this case the two devices are made into their own
2982  * mirror using the 'replacing' vdev, which is functionally identical to
2983  * the mirror vdev (it actually reuses all the same ops) but has a few
2984  * extra rules: you can't attach to it after it's been created, and upon
2985  * completion of resilvering, the first disk (the one being replaced)
2986  * is automatically detached.
2987  */
2988 int
2989 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
2990 {
2991 	uint64_t txg, open_txg;
2992 	vdev_t *rvd = spa->spa_root_vdev;
2993 	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
2994 	vdev_ops_t *pvops;
2995 	dmu_tx_t *tx;
2996 	char *oldvdpath, *newvdpath;
2997 	int newvd_isspare;
2998 	int error;
2999 
3000 	txg = spa_vdev_enter(spa);
3001 
3002 	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3003 
3004 	if (oldvd == NULL)
3005 		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3006 
3007 	if (!oldvd->vdev_ops->vdev_op_leaf)
3008 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3009 
3010 	pvd = oldvd->vdev_parent;
3011 
3012 	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3013 	    VDEV_ALLOC_ADD)) != 0)
3014 		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3015 
3016 	if (newrootvd->vdev_children != 1)
3017 		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3018 
3019 	newvd = newrootvd->vdev_child[0];
3020 
3021 	if (!newvd->vdev_ops->vdev_op_leaf)
3022 		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3023 
3024 	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3025 		return (spa_vdev_exit(spa, newrootvd, txg, error));
3026 
3027 	/*
3028 	 * Spares can't replace logs
3029 	 */
3030 	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3031 		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3032 
3033 	if (!replacing) {
3034 		/*
3035 		 * For attach, the only allowable parent is a mirror or the root
3036 		 * vdev.
3037 		 */
3038 		if (pvd->vdev_ops != &vdev_mirror_ops &&
3039 		    pvd->vdev_ops != &vdev_root_ops)
3040 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3041 
3042 		pvops = &vdev_mirror_ops;
3043 	} else {
3044 		/*
3045 		 * Active hot spares can only be replaced by inactive hot
3046 		 * spares.
3047 		 */
3048 		if (pvd->vdev_ops == &vdev_spare_ops &&
3049 		    pvd->vdev_child[1] == oldvd &&
3050 		    !spa_has_spare(spa, newvd->vdev_guid))
3051 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3052 
3053 		/*
3054 		 * If the source is a hot spare, and the parent isn't already a
3055 		 * spare, then we want to create a new hot spare.  Otherwise, we
3056 		 * want to create a replacing vdev.  The user is not allowed to
3057 		 * attach to a spared vdev child unless the 'isspare' state is
3058 		 * the same (spare replaces spare, non-spare replaces
3059 		 * non-spare).
3060 		 */
3061 		if (pvd->vdev_ops == &vdev_replacing_ops)
3062 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3063 		else if (pvd->vdev_ops == &vdev_spare_ops &&
3064 		    newvd->vdev_isspare != oldvd->vdev_isspare)
3065 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3066 		else if (pvd->vdev_ops != &vdev_spare_ops &&
3067 		    newvd->vdev_isspare)
3068 			pvops = &vdev_spare_ops;
3069 		else
3070 			pvops = &vdev_replacing_ops;
3071 	}
3072 
3073 	/*
3074 	 * Compare the new device size with the replaceable/attachable
3075 	 * device size.
3076 	 */
3077 	if (newvd->vdev_psize < vdev_get_rsize(oldvd))
3078 		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3079 
3080 	/*
3081 	 * The new device cannot have a higher alignment requirement
3082 	 * than the top-level vdev.
3083 	 */
3084 	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3085 		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3086 
3087 	/*
3088 	 * If this is an in-place replacement, update oldvd's path and devid
3089 	 * to make it distinguishable from newvd, and unopenable from now on.
3090 	 */
3091 	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3092 		spa_strfree(oldvd->vdev_path);
3093 		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3094 		    KM_SLEEP);
3095 		(void) sprintf(oldvd->vdev_path, "%s/%s",
3096 		    newvd->vdev_path, "old");
3097 		if (oldvd->vdev_devid != NULL) {
3098 			spa_strfree(oldvd->vdev_devid);
3099 			oldvd->vdev_devid = NULL;
3100 		}
3101 	}
3102 
3103 	/*
3104 	 * If the parent is not a mirror, or if we're replacing, insert the new
3105 	 * mirror/replacing/spare vdev above oldvd.
3106 	 */
3107 	if (pvd->vdev_ops != pvops)
3108 		pvd = vdev_add_parent(oldvd, pvops);
3109 
3110 	ASSERT(pvd->vdev_top->vdev_parent == rvd);
3111 	ASSERT(pvd->vdev_ops == pvops);
3112 	ASSERT(oldvd->vdev_parent == pvd);
3113 
3114 	/*
3115 	 * Extract the new device from its root and add it to pvd.
3116 	 */
3117 	vdev_remove_child(newrootvd, newvd);
3118 	newvd->vdev_id = pvd->vdev_children;
3119 	vdev_add_child(pvd, newvd);
3120 
3121 	/*
3122 	 * If newvd is smaller than oldvd, but larger than its rsize,
3123 	 * the addition of newvd may have decreased our parent's asize.
3124 	 */
3125 	pvd->vdev_asize = MIN(pvd->vdev_asize, newvd->vdev_asize);
3126 
3127 	tvd = newvd->vdev_top;
3128 	ASSERT(pvd->vdev_top == tvd);
3129 	ASSERT(tvd->vdev_parent == rvd);
3130 
3131 	vdev_config_dirty(tvd);
3132 
3133 	/*
3134 	 * Set newvd's DTL to [TXG_INITIAL, open_txg].  It will propagate
3135 	 * upward when spa_vdev_exit() calls vdev_dtl_reassess().
3136 	 */
3137 	open_txg = txg + TXG_CONCURRENT_STATES - 1;
3138 
3139 	vdev_dtl_dirty(newvd, DTL_MISSING,
3140 	    TXG_INITIAL, open_txg - TXG_INITIAL + 1);
3141 
3142 	if (newvd->vdev_isspare) {
3143 		spa_spare_activate(newvd);
3144 		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
3145 	}
3146 
3147 	oldvdpath = spa_strdup(oldvd->vdev_path);
3148 	newvdpath = spa_strdup(newvd->vdev_path);
3149 	newvd_isspare = newvd->vdev_isspare;
3150 
3151 	/*
3152 	 * Mark newvd's DTL dirty in this txg.
3153 	 */
3154 	vdev_dirty(tvd, VDD_DTL, newvd, txg);
3155 
3156 	(void) spa_vdev_exit(spa, newrootvd, open_txg, 0);
3157 
3158 	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
3159 	if (dmu_tx_assign(tx, TXG_WAIT) == 0) {
3160 		spa_history_internal_log(LOG_POOL_VDEV_ATTACH, spa, tx,
3161 		    CRED(),  "%s vdev=%s %s vdev=%s",
3162 		    replacing && newvd_isspare ? "spare in" :
3163 		    replacing ? "replace" : "attach", newvdpath,
3164 		    replacing ? "for" : "to", oldvdpath);
3165 		dmu_tx_commit(tx);
3166 	} else {
3167 		dmu_tx_abort(tx);
3168 	}
3169 
3170 	spa_strfree(oldvdpath);
3171 	spa_strfree(newvdpath);
3172 
3173 	/*
3174 	 * Kick off a resilver to update newvd.
3175 	 */
3176 	VERIFY3U(spa_scrub(spa, POOL_SCRUB_RESILVER), ==, 0);
3177 
3178 	return (0);
3179 }
3180 
3181 /*
3182  * Detach a device from a mirror or replacing vdev.
3183  * If 'replace_done' is specified, only detach if the parent
3184  * is a replacing vdev.
3185  */
3186 int
3187 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
3188 {
3189 	uint64_t txg;
3190 	int error;
3191 	vdev_t *rvd = spa->spa_root_vdev;
3192 	vdev_t *vd, *pvd, *cvd, *tvd;
3193 	boolean_t unspare = B_FALSE;
3194 	uint64_t unspare_guid;
3195 	size_t len;
3196 
3197 	txg = spa_vdev_enter(spa);
3198 
3199 	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3200 
3201 	if (vd == NULL)
3202 		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3203 
3204 	if (!vd->vdev_ops->vdev_op_leaf)
3205 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3206 
3207 	pvd = vd->vdev_parent;
3208 
3209 	/*
3210 	 * If the parent/child relationship is not as expected, don't do it.
3211 	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
3212 	 * vdev that's replacing B with C.  The user's intent in replacing
3213 	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
3214 	 * the replace by detaching C, the expected behavior is to end up
3215 	 * M(A,B).  But suppose that right after deciding to detach C,
3216 	 * the replacement of B completes.  We would have M(A,C), and then
3217 	 * ask to detach C, which would leave us with just A -- not what
3218 	 * the user wanted.  To prevent this, we make sure that the
3219 	 * parent/child relationship hasn't changed -- in this example,
3220 	 * that C's parent is still the replacing vdev R.
3221 	 */
3222 	if (pvd->vdev_guid != pguid && pguid != 0)
3223 		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3224 
3225 	/*
3226 	 * If replace_done is specified, only remove this device if it's
3227 	 * the first child of a replacing vdev.  For the 'spare' vdev, either
3228 	 * disk can be removed.
3229 	 */
3230 	if (replace_done) {
3231 		if (pvd->vdev_ops == &vdev_replacing_ops) {
3232 			if (vd->vdev_id != 0)
3233 				return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3234 		} else if (pvd->vdev_ops != &vdev_spare_ops) {
3235 			return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3236 		}
3237 	}
3238 
3239 	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
3240 	    spa_version(spa) >= SPA_VERSION_SPARES);
3241 
3242 	/*
3243 	 * Only mirror, replacing, and spare vdevs support detach.
3244 	 */
3245 	if (pvd->vdev_ops != &vdev_replacing_ops &&
3246 	    pvd->vdev_ops != &vdev_mirror_ops &&
3247 	    pvd->vdev_ops != &vdev_spare_ops)
3248 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3249 
3250 	/*
3251 	 * If this device has the only valid copy of some data,
3252 	 * we cannot safely detach it.
3253 	 */
3254 	if (vdev_dtl_required(vd))
3255 		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3256 
3257 	ASSERT(pvd->vdev_children >= 2);
3258 
3259 	/*
3260 	 * If we are detaching the second disk from a replacing vdev, then
3261 	 * check to see if we changed the original vdev's path to have "/old"
3262 	 * at the end in spa_vdev_attach().  If so, undo that change now.
3263 	 */
3264 	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id == 1 &&
3265 	    pvd->vdev_child[0]->vdev_path != NULL &&
3266 	    pvd->vdev_child[1]->vdev_path != NULL) {
3267 		ASSERT(pvd->vdev_child[1] == vd);
3268 		cvd = pvd->vdev_child[0];
3269 		len = strlen(vd->vdev_path);
3270 		if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
3271 		    strcmp(cvd->vdev_path + len, "/old") == 0) {
3272 			spa_strfree(cvd->vdev_path);
3273 			cvd->vdev_path = spa_strdup(vd->vdev_path);
3274 		}
3275 	}
3276 
3277 	/*
3278 	 * If we are detaching the original disk from a spare, then it implies
3279 	 * that the spare should become a real disk, and be removed from the
3280 	 * active spare list for the pool.
3281 	 */
3282 	if (pvd->vdev_ops == &vdev_spare_ops &&
3283 	    vd->vdev_id == 0 && pvd->vdev_child[1]->vdev_isspare)
3284 		unspare = B_TRUE;
3285 
3286 	/*
3287 	 * Erase the disk labels so the disk can be used for other things.
3288 	 * This must be done after all other error cases are handled,
3289 	 * but before we disembowel vd (so we can still do I/O to it).
3290 	 * But if we can't do it, don't treat the error as fatal --
3291 	 * it may be that the unwritability of the disk is the reason
3292 	 * it's being detached!
3293 	 */
3294 	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
3295 
3296 	/*
3297 	 * Remove vd from its parent and compact the parent's children.
3298 	 */
3299 	vdev_remove_child(pvd, vd);
3300 	vdev_compact_children(pvd);
3301 
3302 	/*
3303 	 * Remember one of the remaining children so we can get tvd below.
3304 	 */
3305 	cvd = pvd->vdev_child[0];
3306 
3307 	/*
3308 	 * If we need to remove the remaining child from the list of hot spares,
3309 	 * do it now, marking the vdev as no longer a spare in the process.
3310 	 * We must do this before vdev_remove_parent(), because that can
3311 	 * change the GUID if it creates a new toplevel GUID.  For a similar
3312 	 * reason, we must remove the spare now, in the same txg as the detach;
3313 	 * otherwise someone could attach a new sibling, change the GUID, and
3314 	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
3315 	 */
3316 	if (unspare) {
3317 		ASSERT(cvd->vdev_isspare);
3318 		spa_spare_remove(cvd);
3319 		unspare_guid = cvd->vdev_guid;
3320 		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
3321 	}
3322 
3323 	/*
3324 	 * If the parent mirror/replacing vdev only has one child,
3325 	 * the parent is no longer needed.  Remove it from the tree.
3326 	 */
3327 	if (pvd->vdev_children == 1)
3328 		vdev_remove_parent(cvd);
3329 
3330 	/*
3331 	 * We don't set tvd until now because the parent we just removed
3332 	 * may have been the previous top-level vdev.
3333 	 */
3334 	tvd = cvd->vdev_top;
3335 	ASSERT(tvd->vdev_parent == rvd);
3336 
3337 	/*
3338 	 * Reevaluate the parent vdev state.
3339 	 */
3340 	vdev_propagate_state(cvd);
3341 
3342 	/*
3343 	 * If the device we just detached was smaller than the others, it may be
3344 	 * possible to add metaslabs (i.e. grow the pool).  vdev_metaslab_init()
3345 	 * can't fail because the existing metaslabs are already in core, so
3346 	 * there's nothing to read from disk.
3347 	 */
3348 	VERIFY(vdev_metaslab_init(tvd, txg) == 0);
3349 
3350 	vdev_config_dirty(tvd);
3351 
3352 	/*
3353 	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
3354 	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
3355 	 * But first make sure we're not on any *other* txg's DTL list, to
3356 	 * prevent vd from being accessed after it's freed.
3357 	 */
3358 	for (int t = 0; t < TXG_SIZE; t++)
3359 		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
3360 	vd->vdev_detached = B_TRUE;
3361 	vdev_dirty(tvd, VDD_DTL, vd, txg);
3362 
3363 	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
3364 
3365 	error = spa_vdev_exit(spa, vd, txg, 0);
3366 
3367 	/*
3368 	 * If this was the removal of the original device in a hot spare vdev,
3369 	 * then we want to go through and remove the device from the hot spare
3370 	 * list of every other pool.
3371 	 */
3372 	if (unspare) {
3373 		spa_t *myspa = spa;
3374 		spa = NULL;
3375 		mutex_enter(&spa_namespace_lock);
3376 		while ((spa = spa_next(spa)) != NULL) {
3377 			if (spa->spa_state != POOL_STATE_ACTIVE)
3378 				continue;
3379 			if (spa == myspa)
3380 				continue;
3381 			spa_open_ref(spa, FTAG);
3382 			mutex_exit(&spa_namespace_lock);
3383 			(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
3384 			mutex_enter(&spa_namespace_lock);
3385 			spa_close(spa, FTAG);
3386 		}
3387 		mutex_exit(&spa_namespace_lock);
3388 	}
3389 
3390 	return (error);
3391 }
3392 
3393 static nvlist_t *
3394 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
3395 {
3396 	for (int i = 0; i < count; i++) {
3397 		uint64_t guid;
3398 
3399 		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
3400 		    &guid) == 0);
3401 
3402 		if (guid == target_guid)
3403 			return (nvpp[i]);
3404 	}
3405 
3406 	return (NULL);
3407 }
3408 
3409 static void
3410 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
3411 	nvlist_t *dev_to_remove)
3412 {
3413 	nvlist_t **newdev = NULL;
3414 
3415 	if (count > 1)
3416 		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
3417 
3418 	for (int i = 0, j = 0; i < count; i++) {
3419 		if (dev[i] == dev_to_remove)
3420 			continue;
3421 		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
3422 	}
3423 
3424 	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
3425 	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
3426 
3427 	for (int i = 0; i < count - 1; i++)
3428 		nvlist_free(newdev[i]);
3429 
3430 	if (count > 1)
3431 		kmem_free(newdev, (count - 1) * sizeof (void *));
3432 }
3433 
3434 /*
3435  * Remove a device from the pool.  Currently, this supports removing only hot
3436  * spares and level 2 ARC devices.
3437  */
3438 int
3439 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
3440 {
3441 	vdev_t *vd;
3442 	nvlist_t **spares, **l2cache, *nv;
3443 	uint_t nspares, nl2cache;
3444 	uint64_t txg = 0;
3445 	int error = 0;
3446 	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
3447 
3448 	if (!locked)
3449 		txg = spa_vdev_enter(spa);
3450 
3451 	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3452 
3453 	if (spa->spa_spares.sav_vdevs != NULL &&
3454 	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3455 	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
3456 	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
3457 		/*
3458 		 * Only remove the hot spare if it's not currently in use
3459 		 * in this pool.
3460 		 */
3461 		if (vd == NULL || unspare) {
3462 			spa_vdev_remove_aux(spa->spa_spares.sav_config,
3463 			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
3464 			spa_load_spares(spa);
3465 			spa->spa_spares.sav_sync = B_TRUE;
3466 		} else {
3467 			error = EBUSY;
3468 		}
3469 	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
3470 	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3471 	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
3472 	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
3473 		/*
3474 		 * Cache devices can always be removed.
3475 		 */
3476 		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
3477 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
3478 		spa_load_l2cache(spa);
3479 		spa->spa_l2cache.sav_sync = B_TRUE;
3480 	} else if (vd != NULL) {
3481 		/*
3482 		 * Normal vdevs cannot be removed (yet).
3483 		 */
3484 		error = ENOTSUP;
3485 	} else {
3486 		/*
3487 		 * There is no vdev of any kind with the specified guid.
3488 		 */
3489 		error = ENOENT;
3490 	}
3491 
3492 	if (!locked)
3493 		return (spa_vdev_exit(spa, NULL, txg, error));
3494 
3495 	return (error);
3496 }
3497 
3498 /*
3499  * Find any device that's done replacing, or a vdev marked 'unspare' that's
3500  * current spared, so we can detach it.
3501  */
3502 static vdev_t *
3503 spa_vdev_resilver_done_hunt(vdev_t *vd)
3504 {
3505 	vdev_t *newvd, *oldvd;
3506 	int c;
3507 
3508 	for (c = 0; c < vd->vdev_children; c++) {
3509 		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
3510 		if (oldvd != NULL)
3511 			return (oldvd);
3512 	}
3513 
3514 	/*
3515 	 * Check for a completed replacement.
3516 	 */
3517 	if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
3518 		oldvd = vd->vdev_child[0];
3519 		newvd = vd->vdev_child[1];
3520 
3521 		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
3522 		    !vdev_dtl_required(oldvd))
3523 			return (oldvd);
3524 	}
3525 
3526 	/*
3527 	 * Check for a completed resilver with the 'unspare' flag set.
3528 	 */
3529 	if (vd->vdev_ops == &vdev_spare_ops && vd->vdev_children == 2) {
3530 		newvd = vd->vdev_child[0];
3531 		oldvd = vd->vdev_child[1];
3532 
3533 		if (newvd->vdev_unspare &&
3534 		    vdev_dtl_empty(newvd, DTL_MISSING) &&
3535 		    !vdev_dtl_required(oldvd)) {
3536 			newvd->vdev_unspare = 0;
3537 			return (oldvd);
3538 		}
3539 	}
3540 
3541 	return (NULL);
3542 }
3543 
3544 static void
3545 spa_vdev_resilver_done(spa_t *spa)
3546 {
3547 	vdev_t *vd, *pvd, *ppvd;
3548 	uint64_t guid, sguid, pguid, ppguid;
3549 
3550 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3551 
3552 	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
3553 		pvd = vd->vdev_parent;
3554 		ppvd = pvd->vdev_parent;
3555 		guid = vd->vdev_guid;
3556 		pguid = pvd->vdev_guid;
3557 		ppguid = ppvd->vdev_guid;
3558 		sguid = 0;
3559 		/*
3560 		 * If we have just finished replacing a hot spared device, then
3561 		 * we need to detach the parent's first child (the original hot
3562 		 * spare) as well.
3563 		 */
3564 		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0) {
3565 			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
3566 			ASSERT(ppvd->vdev_children == 2);
3567 			sguid = ppvd->vdev_child[1]->vdev_guid;
3568 		}
3569 		spa_config_exit(spa, SCL_ALL, FTAG);
3570 		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
3571 			return;
3572 		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
3573 			return;
3574 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3575 	}
3576 
3577 	spa_config_exit(spa, SCL_ALL, FTAG);
3578 }
3579 
3580 /*
3581  * Update the stored path or FRU for this vdev.  Dirty the vdev configuration,
3582  * relying on spa_vdev_enter/exit() to synchronize the labels and cache.
3583  */
3584 int
3585 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
3586     boolean_t ispath)
3587 {
3588 	vdev_t *vd;
3589 	uint64_t txg;
3590 
3591 	txg = spa_vdev_enter(spa);
3592 
3593 	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3594 		return (spa_vdev_exit(spa, NULL, txg, ENOENT));
3595 
3596 	if (!vd->vdev_ops->vdev_op_leaf)
3597 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3598 
3599 	if (ispath) {
3600 		spa_strfree(vd->vdev_path);
3601 		vd->vdev_path = spa_strdup(value);
3602 	} else {
3603 		if (vd->vdev_fru != NULL)
3604 			spa_strfree(vd->vdev_fru);
3605 		vd->vdev_fru = spa_strdup(value);
3606 	}
3607 
3608 	vdev_config_dirty(vd->vdev_top);
3609 
3610 	return (spa_vdev_exit(spa, NULL, txg, 0));
3611 }
3612 
3613 int
3614 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
3615 {
3616 	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
3617 }
3618 
3619 int
3620 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
3621 {
3622 	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
3623 }
3624 
3625 /*
3626  * ==========================================================================
3627  * SPA Scrubbing
3628  * ==========================================================================
3629  */
3630 
3631 int
3632 spa_scrub(spa_t *spa, pool_scrub_type_t type)
3633 {
3634 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
3635 
3636 	if ((uint_t)type >= POOL_SCRUB_TYPES)
3637 		return (ENOTSUP);
3638 
3639 	/*
3640 	 * If a resilver was requested, but there is no DTL on a
3641 	 * writeable leaf device, we have nothing to do.
3642 	 */
3643 	if (type == POOL_SCRUB_RESILVER &&
3644 	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
3645 		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
3646 		return (0);
3647 	}
3648 
3649 	if (type == POOL_SCRUB_EVERYTHING &&
3650 	    spa->spa_dsl_pool->dp_scrub_func != SCRUB_FUNC_NONE &&
3651 	    spa->spa_dsl_pool->dp_scrub_isresilver)
3652 		return (EBUSY);
3653 
3654 	if (type == POOL_SCRUB_EVERYTHING || type == POOL_SCRUB_RESILVER) {
3655 		return (dsl_pool_scrub_clean(spa->spa_dsl_pool));
3656 	} else if (type == POOL_SCRUB_NONE) {
3657 		return (dsl_pool_scrub_cancel(spa->spa_dsl_pool));
3658 	} else {
3659 		return (EINVAL);
3660 	}
3661 }
3662 
3663 /*
3664  * ==========================================================================
3665  * SPA async task processing
3666  * ==========================================================================
3667  */
3668 
3669 static void
3670 spa_async_remove(spa_t *spa, vdev_t *vd)
3671 {
3672 	if (vd->vdev_remove_wanted) {
3673 		vd->vdev_remove_wanted = 0;
3674 		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
3675 		vdev_clear(spa, vd);
3676 		vdev_state_dirty(vd->vdev_top);
3677 	}
3678 
3679 	for (int c = 0; c < vd->vdev_children; c++)
3680 		spa_async_remove(spa, vd->vdev_child[c]);
3681 }
3682 
3683 static void
3684 spa_async_probe(spa_t *spa, vdev_t *vd)
3685 {
3686 	if (vd->vdev_probe_wanted) {
3687 		vd->vdev_probe_wanted = 0;
3688 		vdev_reopen(vd);	/* vdev_open() does the actual probe */
3689 	}
3690 
3691 	for (int c = 0; c < vd->vdev_children; c++)
3692 		spa_async_probe(spa, vd->vdev_child[c]);
3693 }
3694 
3695 static void
3696 spa_async_thread(spa_t *spa)
3697 {
3698 	int tasks;
3699 
3700 	ASSERT(spa->spa_sync_on);
3701 
3702 	mutex_enter(&spa->spa_async_lock);
3703 	tasks = spa->spa_async_tasks;
3704 	spa->spa_async_tasks = 0;
3705 	mutex_exit(&spa->spa_async_lock);
3706 
3707 	/*
3708 	 * See if the config needs to be updated.
3709 	 */
3710 	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
3711 		mutex_enter(&spa_namespace_lock);
3712 		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3713 		mutex_exit(&spa_namespace_lock);
3714 	}
3715 
3716 	/*
3717 	 * See if any devices need to be marked REMOVED.
3718 	 */
3719 	if (tasks & SPA_ASYNC_REMOVE) {
3720 		spa_vdev_state_enter(spa);
3721 		spa_async_remove(spa, spa->spa_root_vdev);
3722 		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
3723 			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
3724 		for (int i = 0; i < spa->spa_spares.sav_count; i++)
3725 			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
3726 		(void) spa_vdev_state_exit(spa, NULL, 0);
3727 	}
3728 
3729 	/*
3730 	 * See if any devices need to be probed.
3731 	 */
3732 	if (tasks & SPA_ASYNC_PROBE) {
3733 		spa_vdev_state_enter(spa);
3734 		spa_async_probe(spa, spa->spa_root_vdev);
3735 		(void) spa_vdev_state_exit(spa, NULL, 0);
3736 	}
3737 
3738 	/*
3739 	 * If any devices are done replacing, detach them.
3740 	 */
3741 	if (tasks & SPA_ASYNC_RESILVER_DONE)
3742 		spa_vdev_resilver_done(spa);
3743 
3744 	/*
3745 	 * Kick off a resilver.
3746 	 */
3747 	if (tasks & SPA_ASYNC_RESILVER)
3748 		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER) == 0);
3749 
3750 	/*
3751 	 * Let the world know that we're done.
3752 	 */
3753 	mutex_enter(&spa->spa_async_lock);
3754 	spa->spa_async_thread = NULL;
3755 	cv_broadcast(&spa->spa_async_cv);
3756 	mutex_exit(&spa->spa_async_lock);
3757 	thread_exit();
3758 }
3759 
3760 void
3761 spa_async_suspend(spa_t *spa)
3762 {
3763 	mutex_enter(&spa->spa_async_lock);
3764 	spa->spa_async_suspended++;
3765 	while (spa->spa_async_thread != NULL)
3766 		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
3767 	mutex_exit(&spa->spa_async_lock);
3768 }
3769 
3770 void
3771 spa_async_resume(spa_t *spa)
3772 {
3773 	mutex_enter(&spa->spa_async_lock);
3774 	ASSERT(spa->spa_async_suspended != 0);
3775 	spa->spa_async_suspended--;
3776 	mutex_exit(&spa->spa_async_lock);
3777 }
3778 
3779 static void
3780 spa_async_dispatch(spa_t *spa)
3781 {
3782 	mutex_enter(&spa->spa_async_lock);
3783 	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
3784 	    spa->spa_async_thread == NULL &&
3785 	    rootdir != NULL && !vn_is_readonly(rootdir))
3786 		spa->spa_async_thread = thread_create(NULL, 0,
3787 		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
3788 	mutex_exit(&spa->spa_async_lock);
3789 }
3790 
3791 void
3792 spa_async_request(spa_t *spa, int task)
3793 {
3794 	mutex_enter(&spa->spa_async_lock);
3795 	spa->spa_async_tasks |= task;
3796 	mutex_exit(&spa->spa_async_lock);
3797 }
3798 
3799 /*
3800  * ==========================================================================
3801  * SPA syncing routines
3802  * ==========================================================================
3803  */
3804 
3805 static void
3806 spa_sync_deferred_frees(spa_t *spa, uint64_t txg)
3807 {
3808 	bplist_t *bpl = &spa->spa_sync_bplist;
3809 	dmu_tx_t *tx;
3810 	blkptr_t blk;
3811 	uint64_t itor = 0;
3812 	zio_t *zio;
3813 	int error;
3814 	uint8_t c = 1;
3815 
3816 	zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
3817 
3818 	while (bplist_iterate(bpl, &itor, &blk) == 0) {
3819 		ASSERT(blk.blk_birth < txg);
3820 		zio_nowait(zio_free(zio, spa, txg, &blk, NULL, NULL,
3821 		    ZIO_FLAG_MUSTSUCCEED));
3822 	}
3823 
3824 	error = zio_wait(zio);
3825 	ASSERT3U(error, ==, 0);
3826 
3827 	tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3828 	bplist_vacate(bpl, tx);
3829 
3830 	/*
3831 	 * Pre-dirty the first block so we sync to convergence faster.
3832 	 * (Usually only the first block is needed.)
3833 	 */
3834 	dmu_write(spa->spa_meta_objset, spa->spa_sync_bplist_obj, 0, 1, &c, tx);
3835 	dmu_tx_commit(tx);
3836 }
3837 
3838 static void
3839 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
3840 {
3841 	char *packed = NULL;
3842 	size_t bufsize;
3843 	size_t nvsize = 0;
3844 	dmu_buf_t *db;
3845 
3846 	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
3847 
3848 	/*
3849 	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
3850 	 * information.  This avoids the dbuf_will_dirty() path and
3851 	 * saves us a pre-read to get data we don't actually care about.
3852 	 */
3853 	bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
3854 	packed = kmem_alloc(bufsize, KM_SLEEP);
3855 
3856 	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
3857 	    KM_SLEEP) == 0);
3858 	bzero(packed + nvsize, bufsize - nvsize);
3859 
3860 	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
3861 
3862 	kmem_free(packed, bufsize);
3863 
3864 	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
3865 	dmu_buf_will_dirty(db, tx);
3866 	*(uint64_t *)db->db_data = nvsize;
3867 	dmu_buf_rele(db, FTAG);
3868 }
3869 
3870 static void
3871 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
3872     const char *config, const char *entry)
3873 {
3874 	nvlist_t *nvroot;
3875 	nvlist_t **list;
3876 	int i;
3877 
3878 	if (!sav->sav_sync)
3879 		return;
3880 
3881 	/*
3882 	 * Update the MOS nvlist describing the list of available devices.
3883 	 * spa_validate_aux() will have already made sure this nvlist is
3884 	 * valid and the vdevs are labeled appropriately.
3885 	 */
3886 	if (sav->sav_object == 0) {
3887 		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
3888 		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
3889 		    sizeof (uint64_t), tx);
3890 		VERIFY(zap_update(spa->spa_meta_objset,
3891 		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
3892 		    &sav->sav_object, tx) == 0);
3893 	}
3894 
3895 	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3896 	if (sav->sav_count == 0) {
3897 		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
3898 	} else {
3899 		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
3900 		for (i = 0; i < sav->sav_count; i++)
3901 			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
3902 			    B_FALSE, B_FALSE, B_TRUE);
3903 		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
3904 		    sav->sav_count) == 0);
3905 		for (i = 0; i < sav->sav_count; i++)
3906 			nvlist_free(list[i]);
3907 		kmem_free(list, sav->sav_count * sizeof (void *));
3908 	}
3909 
3910 	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
3911 	nvlist_free(nvroot);
3912 
3913 	sav->sav_sync = B_FALSE;
3914 }
3915 
3916 static void
3917 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
3918 {
3919 	nvlist_t *config;
3920 
3921 	if (list_is_empty(&spa->spa_config_dirty_list))
3922 		return;
3923 
3924 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
3925 
3926 	config = spa_config_generate(spa, spa->spa_root_vdev,
3927 	    dmu_tx_get_txg(tx), B_FALSE);
3928 
3929 	spa_config_exit(spa, SCL_STATE, FTAG);
3930 
3931 	if (spa->spa_config_syncing)
3932 		nvlist_free(spa->spa_config_syncing);
3933 	spa->spa_config_syncing = config;
3934 
3935 	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
3936 }
3937 
3938 /*
3939  * Set zpool properties.
3940  */
3941 static void
3942 spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
3943 {
3944 	spa_t *spa = arg1;
3945 	objset_t *mos = spa->spa_meta_objset;
3946 	nvlist_t *nvp = arg2;
3947 	nvpair_t *elem;
3948 	uint64_t intval;
3949 	char *strval;
3950 	zpool_prop_t prop;
3951 	const char *propname;
3952 	zprop_type_t proptype;
3953 
3954 	mutex_enter(&spa->spa_props_lock);
3955 
3956 	elem = NULL;
3957 	while ((elem = nvlist_next_nvpair(nvp, elem))) {
3958 		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
3959 		case ZPOOL_PROP_VERSION:
3960 			/*
3961 			 * Only set version for non-zpool-creation cases
3962 			 * (set/import). spa_create() needs special care
3963 			 * for version setting.
3964 			 */
3965 			if (tx->tx_txg != TXG_INITIAL) {
3966 				VERIFY(nvpair_value_uint64(elem,
3967 				    &intval) == 0);
3968 				ASSERT(intval <= SPA_VERSION);
3969 				ASSERT(intval >= spa_version(spa));
3970 				spa->spa_uberblock.ub_version = intval;
3971 				vdev_config_dirty(spa->spa_root_vdev);
3972 			}
3973 			break;
3974 
3975 		case ZPOOL_PROP_ALTROOT:
3976 			/*
3977 			 * 'altroot' is a non-persistent property. It should
3978 			 * have been set temporarily at creation or import time.
3979 			 */
3980 			ASSERT(spa->spa_root != NULL);
3981 			break;
3982 
3983 		case ZPOOL_PROP_CACHEFILE:
3984 			/*
3985 			 * 'cachefile' is also a non-persisitent property.
3986 			 */
3987 			break;
3988 		default:
3989 			/*
3990 			 * Set pool property values in the poolprops mos object.
3991 			 */
3992 			if (spa->spa_pool_props_object == 0) {
3993 				objset_t *mos = spa->spa_meta_objset;
3994 
3995 				VERIFY((spa->spa_pool_props_object =
3996 				    zap_create(mos, DMU_OT_POOL_PROPS,
3997 				    DMU_OT_NONE, 0, tx)) > 0);
3998 
3999 				VERIFY(zap_update(mos,
4000 				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
4001 				    8, 1, &spa->spa_pool_props_object, tx)
4002 				    == 0);
4003 			}
4004 
4005 			/* normalize the property name */
4006 			propname = zpool_prop_to_name(prop);
4007 			proptype = zpool_prop_get_type(prop);
4008 
4009 			if (nvpair_type(elem) == DATA_TYPE_STRING) {
4010 				ASSERT(proptype == PROP_TYPE_STRING);
4011 				VERIFY(nvpair_value_string(elem, &strval) == 0);
4012 				VERIFY(zap_update(mos,
4013 				    spa->spa_pool_props_object, propname,
4014 				    1, strlen(strval) + 1, strval, tx) == 0);
4015 
4016 			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
4017 				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
4018 
4019 				if (proptype == PROP_TYPE_INDEX) {
4020 					const char *unused;
4021 					VERIFY(zpool_prop_index_to_string(
4022 					    prop, intval, &unused) == 0);
4023 				}
4024 				VERIFY(zap_update(mos,
4025 				    spa->spa_pool_props_object, propname,
4026 				    8, 1, &intval, tx) == 0);
4027 			} else {
4028 				ASSERT(0); /* not allowed */
4029 			}
4030 
4031 			switch (prop) {
4032 			case ZPOOL_PROP_DELEGATION:
4033 				spa->spa_delegation = intval;
4034 				break;
4035 			case ZPOOL_PROP_BOOTFS:
4036 				spa->spa_bootfs = intval;
4037 				break;
4038 			case ZPOOL_PROP_FAILUREMODE:
4039 				spa->spa_failmode = intval;
4040 				break;
4041 			default:
4042 				break;
4043 			}
4044 		}
4045 
4046 		/* log internal history if this is not a zpool create */
4047 		if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
4048 		    tx->tx_txg != TXG_INITIAL) {
4049 			spa_history_internal_log(LOG_POOL_PROPSET,
4050 			    spa, tx, cr, "%s %lld %s",
4051 			    nvpair_name(elem), intval, spa_name(spa));
4052 		}
4053 	}
4054 
4055 	mutex_exit(&spa->spa_props_lock);
4056 }
4057 
4058 /*
4059  * Sync the specified transaction group.  New blocks may be dirtied as
4060  * part of the process, so we iterate until it converges.
4061  */
4062 void
4063 spa_sync(spa_t *spa, uint64_t txg)
4064 {
4065 	dsl_pool_t *dp = spa->spa_dsl_pool;
4066 	objset_t *mos = spa->spa_meta_objset;
4067 	bplist_t *bpl = &spa->spa_sync_bplist;
4068 	vdev_t *rvd = spa->spa_root_vdev;
4069 	vdev_t *vd;
4070 	dmu_tx_t *tx;
4071 	int dirty_vdevs;
4072 	int error;
4073 
4074 	/*
4075 	 * Lock out configuration changes.
4076 	 */
4077 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4078 
4079 	spa->spa_syncing_txg = txg;
4080 	spa->spa_sync_pass = 0;
4081 
4082 	/*
4083 	 * If there are any pending vdev state changes, convert them
4084 	 * into config changes that go out with this transaction group.
4085 	 */
4086 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4087 	while (list_head(&spa->spa_state_dirty_list) != NULL) {
4088 		/*
4089 		 * We need the write lock here because, for aux vdevs,
4090 		 * calling vdev_config_dirty() modifies sav_config.
4091 		 * This is ugly and will become unnecessary when we
4092 		 * eliminate the aux vdev wart by integrating all vdevs
4093 		 * into the root vdev tree.
4094 		 */
4095 		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
4096 		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
4097 		while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
4098 			vdev_state_clean(vd);
4099 			vdev_config_dirty(vd);
4100 		}
4101 		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
4102 		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
4103 	}
4104 	spa_config_exit(spa, SCL_STATE, FTAG);
4105 
4106 	VERIFY(0 == bplist_open(bpl, mos, spa->spa_sync_bplist_obj));
4107 
4108 	tx = dmu_tx_create_assigned(dp, txg);
4109 
4110 	/*
4111 	 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
4112 	 * set spa_deflate if we have no raid-z vdevs.
4113 	 */
4114 	if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
4115 	    spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
4116 		int i;
4117 
4118 		for (i = 0; i < rvd->vdev_children; i++) {
4119 			vd = rvd->vdev_child[i];
4120 			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
4121 				break;
4122 		}
4123 		if (i == rvd->vdev_children) {
4124 			spa->spa_deflate = TRUE;
4125 			VERIFY(0 == zap_add(spa->spa_meta_objset,
4126 			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
4127 			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
4128 		}
4129 	}
4130 
4131 	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
4132 	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
4133 		dsl_pool_create_origin(dp, tx);
4134 
4135 		/* Keeping the origin open increases spa_minref */
4136 		spa->spa_minref += 3;
4137 	}
4138 
4139 	if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
4140 	    spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
4141 		dsl_pool_upgrade_clones(dp, tx);
4142 	}
4143 
4144 	/*
4145 	 * If anything has changed in this txg, push the deferred frees
4146 	 * from the previous txg.  If not, leave them alone so that we
4147 	 * don't generate work on an otherwise idle system.
4148 	 */
4149 	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
4150 	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
4151 	    !txg_list_empty(&dp->dp_sync_tasks, txg))
4152 		spa_sync_deferred_frees(spa, txg);
4153 
4154 	/*
4155 	 * Iterate to convergence.
4156 	 */
4157 	do {
4158 		spa->spa_sync_pass++;
4159 
4160 		spa_sync_config_object(spa, tx);
4161 		spa_sync_aux_dev(spa, &spa->spa_spares, tx,
4162 		    ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
4163 		spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
4164 		    ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
4165 		spa_errlog_sync(spa, txg);
4166 		dsl_pool_sync(dp, txg);
4167 
4168 		dirty_vdevs = 0;
4169 		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) {
4170 			vdev_sync(vd, txg);
4171 			dirty_vdevs++;
4172 		}
4173 
4174 		bplist_sync(bpl, tx);
4175 	} while (dirty_vdevs);
4176 
4177 	bplist_close(bpl);
4178 
4179 	dprintf("txg %llu passes %d\n", txg, spa->spa_sync_pass);
4180 
4181 	/*
4182 	 * Rewrite the vdev configuration (which includes the uberblock)
4183 	 * to commit the transaction group.
4184 	 *
4185 	 * If there are no dirty vdevs, we sync the uberblock to a few
4186 	 * random top-level vdevs that are known to be visible in the
4187 	 * config cache (see spa_vdev_add() for a complete description).
4188 	 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
4189 	 */
4190 	for (;;) {
4191 		/*
4192 		 * We hold SCL_STATE to prevent vdev open/close/etc.
4193 		 * while we're attempting to write the vdev labels.
4194 		 */
4195 		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4196 
4197 		if (list_is_empty(&spa->spa_config_dirty_list)) {
4198 			vdev_t *svd[SPA_DVAS_PER_BP];
4199 			int svdcount = 0;
4200 			int children = rvd->vdev_children;
4201 			int c0 = spa_get_random(children);
4202 			int c;
4203 
4204 			for (c = 0; c < children; c++) {
4205 				vd = rvd->vdev_child[(c0 + c) % children];
4206 				if (vd->vdev_ms_array == 0 || vd->vdev_islog)
4207 					continue;
4208 				svd[svdcount++] = vd;
4209 				if (svdcount == SPA_DVAS_PER_BP)
4210 					break;
4211 			}
4212 			error = vdev_config_sync(svd, svdcount, txg);
4213 		} else {
4214 			error = vdev_config_sync(rvd->vdev_child,
4215 			    rvd->vdev_children, txg);
4216 		}
4217 
4218 		spa_config_exit(spa, SCL_STATE, FTAG);
4219 
4220 		if (error == 0)
4221 			break;
4222 		zio_suspend(spa, NULL);
4223 		zio_resume_wait(spa);
4224 	}
4225 	dmu_tx_commit(tx);
4226 
4227 	/*
4228 	 * Clear the dirty config list.
4229 	 */
4230 	while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
4231 		vdev_config_clean(vd);
4232 
4233 	/*
4234 	 * Now that the new config has synced transactionally,
4235 	 * let it become visible to the config cache.
4236 	 */
4237 	if (spa->spa_config_syncing != NULL) {
4238 		spa_config_set(spa, spa->spa_config_syncing);
4239 		spa->spa_config_txg = txg;
4240 		spa->spa_config_syncing = NULL;
4241 	}
4242 
4243 	spa->spa_ubsync = spa->spa_uberblock;
4244 
4245 	/*
4246 	 * Clean up the ZIL records for the synced txg.
4247 	 */
4248 	dsl_pool_zil_clean(dp);
4249 
4250 	/*
4251 	 * Update usable space statistics.
4252 	 */
4253 	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
4254 		vdev_sync_done(vd, txg);
4255 
4256 	/*
4257 	 * It had better be the case that we didn't dirty anything
4258 	 * since vdev_config_sync().
4259 	 */
4260 	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
4261 	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
4262 	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
4263 	ASSERT(bpl->bpl_queue == NULL);
4264 
4265 	spa_config_exit(spa, SCL_CONFIG, FTAG);
4266 
4267 	/*
4268 	 * If any async tasks have been requested, kick them off.
4269 	 */
4270 	spa_async_dispatch(spa);
4271 }
4272 
4273 /*
4274  * Sync all pools.  We don't want to hold the namespace lock across these
4275  * operations, so we take a reference on the spa_t and drop the lock during the
4276  * sync.
4277  */
4278 void
4279 spa_sync_allpools(void)
4280 {
4281 	spa_t *spa = NULL;
4282 	mutex_enter(&spa_namespace_lock);
4283 	while ((spa = spa_next(spa)) != NULL) {
4284 		if (spa_state(spa) != POOL_STATE_ACTIVE || spa_suspended(spa))
4285 			continue;
4286 		spa_open_ref(spa, FTAG);
4287 		mutex_exit(&spa_namespace_lock);
4288 		txg_wait_synced(spa_get_dsl(spa), 0);
4289 		mutex_enter(&spa_namespace_lock);
4290 		spa_close(spa, FTAG);
4291 	}
4292 	mutex_exit(&spa_namespace_lock);
4293 }
4294 
4295 /*
4296  * ==========================================================================
4297  * Miscellaneous routines
4298  * ==========================================================================
4299  */
4300 
4301 /*
4302  * Remove all pools in the system.
4303  */
4304 void
4305 spa_evict_all(void)
4306 {
4307 	spa_t *spa;
4308 
4309 	/*
4310 	 * Remove all cached state.  All pools should be closed now,
4311 	 * so every spa in the AVL tree should be unreferenced.
4312 	 */
4313 	mutex_enter(&spa_namespace_lock);
4314 	while ((spa = spa_next(NULL)) != NULL) {
4315 		/*
4316 		 * Stop async tasks.  The async thread may need to detach
4317 		 * a device that's been replaced, which requires grabbing
4318 		 * spa_namespace_lock, so we must drop it here.
4319 		 */
4320 		spa_open_ref(spa, FTAG);
4321 		mutex_exit(&spa_namespace_lock);
4322 		spa_async_suspend(spa);
4323 		mutex_enter(&spa_namespace_lock);
4324 		spa_close(spa, FTAG);
4325 
4326 		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4327 			spa_unload(spa);
4328 			spa_deactivate(spa);
4329 		}
4330 		spa_remove(spa);
4331 	}
4332 	mutex_exit(&spa_namespace_lock);
4333 }
4334 
4335 vdev_t *
4336 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
4337 {
4338 	vdev_t *vd;
4339 	int i;
4340 
4341 	if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
4342 		return (vd);
4343 
4344 	if (aux) {
4345 		for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
4346 			vd = spa->spa_l2cache.sav_vdevs[i];
4347 			if (vd->vdev_guid == guid)
4348 				return (vd);
4349 		}
4350 
4351 		for (i = 0; i < spa->spa_spares.sav_count; i++) {
4352 			vd = spa->spa_spares.sav_vdevs[i];
4353 			if (vd->vdev_guid == guid)
4354 				return (vd);
4355 		}
4356 	}
4357 
4358 	return (NULL);
4359 }
4360 
4361 void
4362 spa_upgrade(spa_t *spa, uint64_t version)
4363 {
4364 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4365 
4366 	/*
4367 	 * This should only be called for a non-faulted pool, and since a
4368 	 * future version would result in an unopenable pool, this shouldn't be
4369 	 * possible.
4370 	 */
4371 	ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
4372 	ASSERT(version >= spa->spa_uberblock.ub_version);
4373 
4374 	spa->spa_uberblock.ub_version = version;
4375 	vdev_config_dirty(spa->spa_root_vdev);
4376 
4377 	spa_config_exit(spa, SCL_ALL, FTAG);
4378 
4379 	txg_wait_synced(spa_get_dsl(spa), 0);
4380 }
4381 
4382 boolean_t
4383 spa_has_spare(spa_t *spa, uint64_t guid)
4384 {
4385 	int i;
4386 	uint64_t spareguid;
4387 	spa_aux_vdev_t *sav = &spa->spa_spares;
4388 
4389 	for (i = 0; i < sav->sav_count; i++)
4390 		if (sav->sav_vdevs[i]->vdev_guid == guid)
4391 			return (B_TRUE);
4392 
4393 	for (i = 0; i < sav->sav_npending; i++) {
4394 		if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
4395 		    &spareguid) == 0 && spareguid == guid)
4396 			return (B_TRUE);
4397 	}
4398 
4399 	return (B_FALSE);
4400 }
4401 
4402 /*
4403  * Check if a pool has an active shared spare device.
4404  * Note: reference count of an active spare is 2, as a spare and as a replace
4405  */
4406 static boolean_t
4407 spa_has_active_shared_spare(spa_t *spa)
4408 {
4409 	int i, refcnt;
4410 	uint64_t pool;
4411 	spa_aux_vdev_t *sav = &spa->spa_spares;
4412 
4413 	for (i = 0; i < sav->sav_count; i++) {
4414 		if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
4415 		    &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
4416 		    refcnt > 2)
4417 			return (B_TRUE);
4418 	}
4419 
4420 	return (B_FALSE);
4421 }
4422 
4423 /*
4424  * Post a sysevent corresponding to the given event.  The 'name' must be one of
4425  * the event definitions in sys/sysevent/eventdefs.h.  The payload will be
4426  * filled in from the spa and (optionally) the vdev.  This doesn't do anything
4427  * in the userland libzpool, as we don't want consumers to misinterpret ztest
4428  * or zdb as real changes.
4429  */
4430 void
4431 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
4432 {
4433 #ifdef _KERNEL
4434 	sysevent_t		*ev;
4435 	sysevent_attr_list_t	*attr = NULL;
4436 	sysevent_value_t	value;
4437 	sysevent_id_t		eid;
4438 
4439 	ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
4440 	    SE_SLEEP);
4441 
4442 	value.value_type = SE_DATA_TYPE_STRING;
4443 	value.value.sv_string = spa_name(spa);
4444 	if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
4445 		goto done;
4446 
4447 	value.value_type = SE_DATA_TYPE_UINT64;
4448 	value.value.sv_uint64 = spa_guid(spa);
4449 	if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
4450 		goto done;
4451 
4452 	if (vd) {
4453 		value.value_type = SE_DATA_TYPE_UINT64;
4454 		value.value.sv_uint64 = vd->vdev_guid;
4455 		if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
4456 		    SE_SLEEP) != 0)
4457 			goto done;
4458 
4459 		if (vd->vdev_path) {
4460 			value.value_type = SE_DATA_TYPE_STRING;
4461 			value.value.sv_string = vd->vdev_path;
4462 			if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
4463 			    &value, SE_SLEEP) != 0)
4464 				goto done;
4465 		}
4466 	}
4467 
4468 	if (sysevent_attach_attributes(ev, attr) != 0)
4469 		goto done;
4470 	attr = NULL;
4471 
4472 	(void) log_sysevent(ev, SE_SLEEP, &eid);
4473 
4474 done:
4475 	if (attr)
4476 		sysevent_free_attr(attr);
4477 	sysevent_free(ev);
4478 #endif
4479 }
4480