xref: /titanic_51/usr/src/uts/common/fs/zfs/spa.c (revision f4a94ada79e5d2be49a574fa7fba9364c57b05d9)
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  * Load the slog device state from the config object since it's possible
1103  * that the label does not contain the most up-to-date information.
1104  */
1105 void
1106 spa_load_log_state(spa_t *spa)
1107 {
1108 	nvlist_t *nv, *nvroot, **child;
1109 	uint64_t is_log;
1110 	uint_t children, c;
1111 	vdev_t *rvd = spa->spa_root_vdev;
1112 
1113 	VERIFY(load_nvlist(spa, spa->spa_config_object, &nv) == 0);
1114 	VERIFY(nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1115 	VERIFY(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1116 	    &child, &children) == 0);
1117 
1118 	for (c = 0; c < children; c++) {
1119 		vdev_t *tvd = rvd->vdev_child[c];
1120 
1121 		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
1122 		    &is_log) == 0 && is_log)
1123 			vdev_load_log_state(tvd, child[c]);
1124 	}
1125 	nvlist_free(nv);
1126 }
1127 
1128 /*
1129  * Check for missing log devices
1130  */
1131 int
1132 spa_check_logs(spa_t *spa)
1133 {
1134 	switch (spa->spa_log_state) {
1135 	case SPA_LOG_MISSING:
1136 		/* need to recheck in case slog has been restored */
1137 	case SPA_LOG_UNKNOWN:
1138 		if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1139 		    DS_FIND_CHILDREN)) {
1140 			spa->spa_log_state = SPA_LOG_MISSING;
1141 			return (1);
1142 		}
1143 		break;
1144 	}
1145 	return (0);
1146 }
1147 
1148 /*
1149  * Load an existing storage pool, using the pool's builtin spa_config as a
1150  * source of configuration information.
1151  */
1152 static int
1153 spa_load(spa_t *spa, nvlist_t *config, spa_load_state_t state, int mosconfig)
1154 {
1155 	int error = 0;
1156 	nvlist_t *nvroot = NULL;
1157 	vdev_t *rvd;
1158 	uberblock_t *ub = &spa->spa_uberblock;
1159 	uint64_t config_cache_txg = spa->spa_config_txg;
1160 	uint64_t pool_guid;
1161 	uint64_t version;
1162 	uint64_t autoreplace = 0;
1163 	int orig_mode = spa->spa_mode;
1164 	char *ereport = FM_EREPORT_ZFS_POOL;
1165 
1166 	/*
1167 	 * If this is an untrusted config, access the pool in read-only mode.
1168 	 * This prevents things like resilvering recently removed devices.
1169 	 */
1170 	if (!mosconfig)
1171 		spa->spa_mode = FREAD;
1172 
1173 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1174 
1175 	spa->spa_load_state = state;
1176 
1177 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) ||
1178 	    nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
1179 		error = EINVAL;
1180 		goto out;
1181 	}
1182 
1183 	/*
1184 	 * Versioning wasn't explicitly added to the label until later, so if
1185 	 * it's not present treat it as the initial version.
1186 	 */
1187 	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &version) != 0)
1188 		version = SPA_VERSION_INITIAL;
1189 
1190 	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1191 	    &spa->spa_config_txg);
1192 
1193 	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1194 	    spa_guid_exists(pool_guid, 0)) {
1195 		error = EEXIST;
1196 		goto out;
1197 	}
1198 
1199 	spa->spa_load_guid = pool_guid;
1200 
1201 	/*
1202 	 * Create "The Godfather" zio to hold all async IOs
1203 	 */
1204 	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1205 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1206 
1207 	/*
1208 	 * Parse the configuration into a vdev tree.  We explicitly set the
1209 	 * value that will be returned by spa_version() since parsing the
1210 	 * configuration requires knowing the version number.
1211 	 */
1212 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1213 	spa->spa_ubsync.ub_version = version;
1214 	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_LOAD);
1215 	spa_config_exit(spa, SCL_ALL, FTAG);
1216 
1217 	if (error != 0)
1218 		goto out;
1219 
1220 	ASSERT(spa->spa_root_vdev == rvd);
1221 	ASSERT(spa_guid(spa) == pool_guid);
1222 
1223 	/*
1224 	 * Try to open all vdevs, loading each label in the process.
1225 	 */
1226 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1227 	error = vdev_open(rvd);
1228 	spa_config_exit(spa, SCL_ALL, FTAG);
1229 	if (error != 0)
1230 		goto out;
1231 
1232 	/*
1233 	 * We need to validate the vdev labels against the configuration that
1234 	 * we have in hand, which is dependent on the setting of mosconfig. If
1235 	 * mosconfig is true then we're validating the vdev labels based on
1236 	 * that config. Otherwise, we're validating against the cached config
1237 	 * (zpool.cache) that was read when we loaded the zfs module, and then
1238 	 * later we will recursively call spa_load() and validate against
1239 	 * the vdev config.
1240 	 */
1241 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1242 	error = vdev_validate(rvd);
1243 	spa_config_exit(spa, SCL_ALL, FTAG);
1244 	if (error != 0)
1245 		goto out;
1246 
1247 	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
1248 		error = ENXIO;
1249 		goto out;
1250 	}
1251 
1252 	/*
1253 	 * Find the best uberblock.
1254 	 */
1255 	vdev_uberblock_load(NULL, rvd, ub);
1256 
1257 	/*
1258 	 * If we weren't able to find a single valid uberblock, return failure.
1259 	 */
1260 	if (ub->ub_txg == 0) {
1261 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1262 		    VDEV_AUX_CORRUPT_DATA);
1263 		error = ENXIO;
1264 		goto out;
1265 	}
1266 
1267 	/*
1268 	 * If the pool is newer than the code, we can't open it.
1269 	 */
1270 	if (ub->ub_version > SPA_VERSION) {
1271 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1272 		    VDEV_AUX_VERSION_NEWER);
1273 		error = ENOTSUP;
1274 		goto out;
1275 	}
1276 
1277 	/*
1278 	 * If the vdev guid sum doesn't match the uberblock, we have an
1279 	 * incomplete configuration.
1280 	 */
1281 	if (rvd->vdev_guid_sum != ub->ub_guid_sum && mosconfig) {
1282 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1283 		    VDEV_AUX_BAD_GUID_SUM);
1284 		error = ENXIO;
1285 		goto out;
1286 	}
1287 
1288 	/*
1289 	 * Initialize internal SPA structures.
1290 	 */
1291 	spa->spa_state = POOL_STATE_ACTIVE;
1292 	spa->spa_ubsync = spa->spa_uberblock;
1293 	spa->spa_first_txg = spa_last_synced_txg(spa) + 1;
1294 	error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1295 	if (error) {
1296 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1297 		    VDEV_AUX_CORRUPT_DATA);
1298 		goto out;
1299 	}
1300 	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1301 
1302 	if (zap_lookup(spa->spa_meta_objset,
1303 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
1304 	    sizeof (uint64_t), 1, &spa->spa_config_object) != 0) {
1305 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1306 		    VDEV_AUX_CORRUPT_DATA);
1307 		error = EIO;
1308 		goto out;
1309 	}
1310 
1311 	if (!mosconfig) {
1312 		nvlist_t *newconfig;
1313 		uint64_t hostid;
1314 
1315 		if (load_nvlist(spa, spa->spa_config_object, &newconfig) != 0) {
1316 			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1317 			    VDEV_AUX_CORRUPT_DATA);
1318 			error = EIO;
1319 			goto out;
1320 		}
1321 
1322 		if (!spa_is_root(spa) && nvlist_lookup_uint64(newconfig,
1323 		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
1324 			char *hostname;
1325 			unsigned long myhostid = 0;
1326 
1327 			VERIFY(nvlist_lookup_string(newconfig,
1328 			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
1329 
1330 #ifdef	_KERNEL
1331 			myhostid = zone_get_hostid(NULL);
1332 #else	/* _KERNEL */
1333 			/*
1334 			 * We're emulating the system's hostid in userland, so
1335 			 * we can't use zone_get_hostid().
1336 			 */
1337 			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
1338 #endif	/* _KERNEL */
1339 			if (hostid != 0 && myhostid != 0 &&
1340 			    hostid != myhostid) {
1341 				cmn_err(CE_WARN, "pool '%s' could not be "
1342 				    "loaded as it was last accessed by "
1343 				    "another system (host: %s hostid: 0x%lx). "
1344 				    "See: http://www.sun.com/msg/ZFS-8000-EY",
1345 				    spa_name(spa), hostname,
1346 				    (unsigned long)hostid);
1347 				error = EBADF;
1348 				goto out;
1349 			}
1350 		}
1351 
1352 		spa_config_set(spa, newconfig);
1353 		spa_unload(spa);
1354 		spa_deactivate(spa);
1355 		spa_activate(spa, orig_mode);
1356 
1357 		return (spa_load(spa, newconfig, state, B_TRUE));
1358 	}
1359 
1360 	if (zap_lookup(spa->spa_meta_objset,
1361 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
1362 	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj) != 0) {
1363 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1364 		    VDEV_AUX_CORRUPT_DATA);
1365 		error = EIO;
1366 		goto out;
1367 	}
1368 
1369 	/*
1370 	 * Load the bit that tells us to use the new accounting function
1371 	 * (raid-z deflation).  If we have an older pool, this will not
1372 	 * be present.
1373 	 */
1374 	error = zap_lookup(spa->spa_meta_objset,
1375 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
1376 	    sizeof (uint64_t), 1, &spa->spa_deflate);
1377 	if (error != 0 && error != ENOENT) {
1378 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1379 		    VDEV_AUX_CORRUPT_DATA);
1380 		error = EIO;
1381 		goto out;
1382 	}
1383 
1384 	/*
1385 	 * Load the persistent error log.  If we have an older pool, this will
1386 	 * not be present.
1387 	 */
1388 	error = zap_lookup(spa->spa_meta_objset,
1389 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_LAST,
1390 	    sizeof (uint64_t), 1, &spa->spa_errlog_last);
1391 	if (error != 0 && error != ENOENT) {
1392 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1393 		    VDEV_AUX_CORRUPT_DATA);
1394 		error = EIO;
1395 		goto out;
1396 	}
1397 
1398 	error = zap_lookup(spa->spa_meta_objset,
1399 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_SCRUB,
1400 	    sizeof (uint64_t), 1, &spa->spa_errlog_scrub);
1401 	if (error != 0 && error != ENOENT) {
1402 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1403 		    VDEV_AUX_CORRUPT_DATA);
1404 		error = EIO;
1405 		goto out;
1406 	}
1407 
1408 	/*
1409 	 * Load the history object.  If we have an older pool, this
1410 	 * will not be present.
1411 	 */
1412 	error = zap_lookup(spa->spa_meta_objset,
1413 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_HISTORY,
1414 	    sizeof (uint64_t), 1, &spa->spa_history);
1415 	if (error != 0 && error != ENOENT) {
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 	/*
1423 	 * Load any hot spares for this pool.
1424 	 */
1425 	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1426 	    DMU_POOL_SPARES, sizeof (uint64_t), 1, &spa->spa_spares.sav_object);
1427 	if (error != 0 && error != ENOENT) {
1428 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1429 		    VDEV_AUX_CORRUPT_DATA);
1430 		error = EIO;
1431 		goto out;
1432 	}
1433 	if (error == 0) {
1434 		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
1435 		if (load_nvlist(spa, spa->spa_spares.sav_object,
1436 		    &spa->spa_spares.sav_config) != 0) {
1437 			vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1438 			    VDEV_AUX_CORRUPT_DATA);
1439 			error = EIO;
1440 			goto out;
1441 		}
1442 
1443 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1444 		spa_load_spares(spa);
1445 		spa_config_exit(spa, SCL_ALL, FTAG);
1446 	}
1447 
1448 	/*
1449 	 * Load any level 2 ARC devices for this pool.
1450 	 */
1451 	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1452 	    DMU_POOL_L2CACHE, sizeof (uint64_t), 1,
1453 	    &spa->spa_l2cache.sav_object);
1454 	if (error != 0 && error != ENOENT) {
1455 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1456 		    VDEV_AUX_CORRUPT_DATA);
1457 		error = EIO;
1458 		goto out;
1459 	}
1460 	if (error == 0) {
1461 		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
1462 		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
1463 		    &spa->spa_l2cache.sav_config) != 0) {
1464 			vdev_set_state(rvd, B_TRUE,
1465 			    VDEV_STATE_CANT_OPEN,
1466 			    VDEV_AUX_CORRUPT_DATA);
1467 			error = EIO;
1468 			goto out;
1469 		}
1470 
1471 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1472 		spa_load_l2cache(spa);
1473 		spa_config_exit(spa, SCL_ALL, FTAG);
1474 	}
1475 
1476 	spa_load_log_state(spa);
1477 
1478 	if (spa_check_logs(spa)) {
1479 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1480 		    VDEV_AUX_BAD_LOG);
1481 		error = ENXIO;
1482 		ereport = FM_EREPORT_ZFS_LOG_REPLAY;
1483 		goto out;
1484 	}
1485 
1486 
1487 	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
1488 
1489 	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1490 	    DMU_POOL_PROPS, sizeof (uint64_t), 1, &spa->spa_pool_props_object);
1491 
1492 	if (error && error != ENOENT) {
1493 		vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN,
1494 		    VDEV_AUX_CORRUPT_DATA);
1495 		error = EIO;
1496 		goto out;
1497 	}
1498 
1499 	if (error == 0) {
1500 		(void) zap_lookup(spa->spa_meta_objset,
1501 		    spa->spa_pool_props_object,
1502 		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS),
1503 		    sizeof (uint64_t), 1, &spa->spa_bootfs);
1504 		(void) zap_lookup(spa->spa_meta_objset,
1505 		    spa->spa_pool_props_object,
1506 		    zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE),
1507 		    sizeof (uint64_t), 1, &autoreplace);
1508 		(void) zap_lookup(spa->spa_meta_objset,
1509 		    spa->spa_pool_props_object,
1510 		    zpool_prop_to_name(ZPOOL_PROP_DELEGATION),
1511 		    sizeof (uint64_t), 1, &spa->spa_delegation);
1512 		(void) zap_lookup(spa->spa_meta_objset,
1513 		    spa->spa_pool_props_object,
1514 		    zpool_prop_to_name(ZPOOL_PROP_FAILUREMODE),
1515 		    sizeof (uint64_t), 1, &spa->spa_failmode);
1516 	}
1517 
1518 	/*
1519 	 * If the 'autoreplace' property is set, then post a resource notifying
1520 	 * the ZFS DE that it should not issue any faults for unopenable
1521 	 * devices.  We also iterate over the vdevs, and post a sysevent for any
1522 	 * unopenable vdevs so that the normal autoreplace handler can take
1523 	 * over.
1524 	 */
1525 	if (autoreplace && state != SPA_LOAD_TRYIMPORT)
1526 		spa_check_removed(spa->spa_root_vdev);
1527 
1528 	/*
1529 	 * Load the vdev state for all toplevel vdevs.
1530 	 */
1531 	vdev_load(rvd);
1532 
1533 	/*
1534 	 * Propagate the leaf DTLs we just loaded all the way up the tree.
1535 	 */
1536 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1537 	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
1538 	spa_config_exit(spa, SCL_ALL, FTAG);
1539 
1540 	/*
1541 	 * Check the state of the root vdev.  If it can't be opened, it
1542 	 * indicates one or more toplevel vdevs are faulted.
1543 	 */
1544 	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
1545 		error = ENXIO;
1546 		goto out;
1547 	}
1548 
1549 	if (spa_writeable(spa)) {
1550 		dmu_tx_t *tx;
1551 		int need_update = B_FALSE;
1552 
1553 		ASSERT(state != SPA_LOAD_TRYIMPORT);
1554 
1555 		/*
1556 		 * Claim log blocks that haven't been committed yet.
1557 		 * This must all happen in a single txg.
1558 		 */
1559 		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
1560 		    spa_first_txg(spa));
1561 		(void) dmu_objset_find(spa_name(spa),
1562 		    zil_claim, tx, DS_FIND_CHILDREN);
1563 		dmu_tx_commit(tx);
1564 
1565 		spa->spa_log_state = SPA_LOG_GOOD;
1566 		spa->spa_sync_on = B_TRUE;
1567 		txg_sync_start(spa->spa_dsl_pool);
1568 
1569 		/*
1570 		 * Wait for all claims to sync.
1571 		 */
1572 		txg_wait_synced(spa->spa_dsl_pool, 0);
1573 
1574 		/*
1575 		 * If the config cache is stale, or we have uninitialized
1576 		 * metaslabs (see spa_vdev_add()), then update the config.
1577 		 */
1578 		if (config_cache_txg != spa->spa_config_txg ||
1579 		    state == SPA_LOAD_IMPORT)
1580 			need_update = B_TRUE;
1581 
1582 		for (int c = 0; c < rvd->vdev_children; c++)
1583 			if (rvd->vdev_child[c]->vdev_ms_array == 0)
1584 				need_update = B_TRUE;
1585 
1586 		/*
1587 		 * Update the config cache asychronously in case we're the
1588 		 * root pool, in which case the config cache isn't writable yet.
1589 		 */
1590 		if (need_update)
1591 			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
1592 
1593 		/*
1594 		 * Check all DTLs to see if anything needs resilvering.
1595 		 */
1596 		if (vdev_resilver_needed(rvd, NULL, NULL))
1597 			spa_async_request(spa, SPA_ASYNC_RESILVER);
1598 	}
1599 
1600 	error = 0;
1601 out:
1602 	spa->spa_minref = refcount_count(&spa->spa_refcount);
1603 	if (error && error != EBADF)
1604 		zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1605 	spa->spa_load_state = SPA_LOAD_NONE;
1606 	spa->spa_ena = 0;
1607 
1608 	return (error);
1609 }
1610 
1611 /*
1612  * Pool Open/Import
1613  *
1614  * The import case is identical to an open except that the configuration is sent
1615  * down from userland, instead of grabbed from the configuration cache.  For the
1616  * case of an open, the pool configuration will exist in the
1617  * POOL_STATE_UNINITIALIZED state.
1618  *
1619  * The stats information (gen/count/ustats) is used to gather vdev statistics at
1620  * the same time open the pool, without having to keep around the spa_t in some
1621  * ambiguous state.
1622  */
1623 static int
1624 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t **config)
1625 {
1626 	spa_t *spa;
1627 	int error;
1628 	int locked = B_FALSE;
1629 
1630 	*spapp = NULL;
1631 
1632 	/*
1633 	 * As disgusting as this is, we need to support recursive calls to this
1634 	 * function because dsl_dir_open() is called during spa_load(), and ends
1635 	 * up calling spa_open() again.  The real fix is to figure out how to
1636 	 * avoid dsl_dir_open() calling this in the first place.
1637 	 */
1638 	if (mutex_owner(&spa_namespace_lock) != curthread) {
1639 		mutex_enter(&spa_namespace_lock);
1640 		locked = B_TRUE;
1641 	}
1642 
1643 	if ((spa = spa_lookup(pool)) == NULL) {
1644 		if (locked)
1645 			mutex_exit(&spa_namespace_lock);
1646 		return (ENOENT);
1647 	}
1648 	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
1649 
1650 		spa_activate(spa, spa_mode_global);
1651 
1652 		error = spa_load(spa, spa->spa_config, SPA_LOAD_OPEN, B_FALSE);
1653 
1654 		if (error == EBADF) {
1655 			/*
1656 			 * If vdev_validate() returns failure (indicated by
1657 			 * EBADF), it indicates that one of the vdevs indicates
1658 			 * that the pool has been exported or destroyed.  If
1659 			 * this is the case, the config cache is out of sync and
1660 			 * we should remove the pool from the namespace.
1661 			 */
1662 			spa_unload(spa);
1663 			spa_deactivate(spa);
1664 			spa_config_sync(spa, B_TRUE, B_TRUE);
1665 			spa_remove(spa);
1666 			if (locked)
1667 				mutex_exit(&spa_namespace_lock);
1668 			return (ENOENT);
1669 		}
1670 
1671 		if (error) {
1672 			/*
1673 			 * We can't open the pool, but we still have useful
1674 			 * information: the state of each vdev after the
1675 			 * attempted vdev_open().  Return this to the user.
1676 			 */
1677 			if (config != NULL && spa->spa_root_vdev != NULL)
1678 				*config = spa_config_generate(spa, NULL, -1ULL,
1679 				    B_TRUE);
1680 			spa_unload(spa);
1681 			spa_deactivate(spa);
1682 			spa->spa_last_open_failed = B_TRUE;
1683 			if (locked)
1684 				mutex_exit(&spa_namespace_lock);
1685 			*spapp = NULL;
1686 			return (error);
1687 		} else {
1688 			spa->spa_last_open_failed = B_FALSE;
1689 		}
1690 	}
1691 
1692 	spa_open_ref(spa, tag);
1693 
1694 	if (locked)
1695 		mutex_exit(&spa_namespace_lock);
1696 
1697 	*spapp = spa;
1698 
1699 	if (config != NULL)
1700 		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
1701 
1702 	return (0);
1703 }
1704 
1705 int
1706 spa_open(const char *name, spa_t **spapp, void *tag)
1707 {
1708 	return (spa_open_common(name, spapp, tag, NULL));
1709 }
1710 
1711 /*
1712  * Lookup the given spa_t, incrementing the inject count in the process,
1713  * preventing it from being exported or destroyed.
1714  */
1715 spa_t *
1716 spa_inject_addref(char *name)
1717 {
1718 	spa_t *spa;
1719 
1720 	mutex_enter(&spa_namespace_lock);
1721 	if ((spa = spa_lookup(name)) == NULL) {
1722 		mutex_exit(&spa_namespace_lock);
1723 		return (NULL);
1724 	}
1725 	spa->spa_inject_ref++;
1726 	mutex_exit(&spa_namespace_lock);
1727 
1728 	return (spa);
1729 }
1730 
1731 void
1732 spa_inject_delref(spa_t *spa)
1733 {
1734 	mutex_enter(&spa_namespace_lock);
1735 	spa->spa_inject_ref--;
1736 	mutex_exit(&spa_namespace_lock);
1737 }
1738 
1739 /*
1740  * Add spares device information to the nvlist.
1741  */
1742 static void
1743 spa_add_spares(spa_t *spa, nvlist_t *config)
1744 {
1745 	nvlist_t **spares;
1746 	uint_t i, nspares;
1747 	nvlist_t *nvroot;
1748 	uint64_t guid;
1749 	vdev_stat_t *vs;
1750 	uint_t vsc;
1751 	uint64_t pool;
1752 
1753 	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
1754 
1755 	if (spa->spa_spares.sav_count == 0)
1756 		return;
1757 
1758 	VERIFY(nvlist_lookup_nvlist(config,
1759 	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1760 	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1761 	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1762 	if (nspares != 0) {
1763 		VERIFY(nvlist_add_nvlist_array(nvroot,
1764 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
1765 		VERIFY(nvlist_lookup_nvlist_array(nvroot,
1766 		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1767 
1768 		/*
1769 		 * Go through and find any spares which have since been
1770 		 * repurposed as an active spare.  If this is the case, update
1771 		 * their status appropriately.
1772 		 */
1773 		for (i = 0; i < nspares; i++) {
1774 			VERIFY(nvlist_lookup_uint64(spares[i],
1775 			    ZPOOL_CONFIG_GUID, &guid) == 0);
1776 			if (spa_spare_exists(guid, &pool, NULL) &&
1777 			    pool != 0ULL) {
1778 				VERIFY(nvlist_lookup_uint64_array(
1779 				    spares[i], ZPOOL_CONFIG_STATS,
1780 				    (uint64_t **)&vs, &vsc) == 0);
1781 				vs->vs_state = VDEV_STATE_CANT_OPEN;
1782 				vs->vs_aux = VDEV_AUX_SPARED;
1783 			}
1784 		}
1785 	}
1786 }
1787 
1788 /*
1789  * Add l2cache device information to the nvlist, including vdev stats.
1790  */
1791 static void
1792 spa_add_l2cache(spa_t *spa, nvlist_t *config)
1793 {
1794 	nvlist_t **l2cache;
1795 	uint_t i, j, nl2cache;
1796 	nvlist_t *nvroot;
1797 	uint64_t guid;
1798 	vdev_t *vd;
1799 	vdev_stat_t *vs;
1800 	uint_t vsc;
1801 
1802 	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
1803 
1804 	if (spa->spa_l2cache.sav_count == 0)
1805 		return;
1806 
1807 	VERIFY(nvlist_lookup_nvlist(config,
1808 	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1809 	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
1810 	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1811 	if (nl2cache != 0) {
1812 		VERIFY(nvlist_add_nvlist_array(nvroot,
1813 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
1814 		VERIFY(nvlist_lookup_nvlist_array(nvroot,
1815 		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1816 
1817 		/*
1818 		 * Update level 2 cache device stats.
1819 		 */
1820 
1821 		for (i = 0; i < nl2cache; i++) {
1822 			VERIFY(nvlist_lookup_uint64(l2cache[i],
1823 			    ZPOOL_CONFIG_GUID, &guid) == 0);
1824 
1825 			vd = NULL;
1826 			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
1827 				if (guid ==
1828 				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
1829 					vd = spa->spa_l2cache.sav_vdevs[j];
1830 					break;
1831 				}
1832 			}
1833 			ASSERT(vd != NULL);
1834 
1835 			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
1836 			    ZPOOL_CONFIG_STATS, (uint64_t **)&vs, &vsc) == 0);
1837 			vdev_get_stats(vd, vs);
1838 		}
1839 	}
1840 }
1841 
1842 int
1843 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
1844 {
1845 	int error;
1846 	spa_t *spa;
1847 
1848 	*config = NULL;
1849 	error = spa_open_common(name, &spa, FTAG, config);
1850 
1851 	if (spa != NULL) {
1852 		/*
1853 		 * This still leaves a window of inconsistency where the spares
1854 		 * or l2cache devices could change and the config would be
1855 		 * self-inconsistent.
1856 		 */
1857 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
1858 
1859 		if (*config != NULL) {
1860 			VERIFY(nvlist_add_uint64(*config,
1861 			    ZPOOL_CONFIG_ERRCOUNT,
1862 			    spa_get_errlog_size(spa)) == 0);
1863 
1864 			if (spa_suspended(spa))
1865 				VERIFY(nvlist_add_uint64(*config,
1866 				    ZPOOL_CONFIG_SUSPENDED,
1867 				    spa->spa_failmode) == 0);
1868 
1869 			spa_add_spares(spa, *config);
1870 			spa_add_l2cache(spa, *config);
1871 		}
1872 	}
1873 
1874 	/*
1875 	 * We want to get the alternate root even for faulted pools, so we cheat
1876 	 * and call spa_lookup() directly.
1877 	 */
1878 	if (altroot) {
1879 		if (spa == NULL) {
1880 			mutex_enter(&spa_namespace_lock);
1881 			spa = spa_lookup(name);
1882 			if (spa)
1883 				spa_altroot(spa, altroot, buflen);
1884 			else
1885 				altroot[0] = '\0';
1886 			spa = NULL;
1887 			mutex_exit(&spa_namespace_lock);
1888 		} else {
1889 			spa_altroot(spa, altroot, buflen);
1890 		}
1891 	}
1892 
1893 	if (spa != NULL) {
1894 		spa_config_exit(spa, SCL_CONFIG, FTAG);
1895 		spa_close(spa, FTAG);
1896 	}
1897 
1898 	return (error);
1899 }
1900 
1901 /*
1902  * Validate that the auxiliary device array is well formed.  We must have an
1903  * array of nvlists, each which describes a valid leaf vdev.  If this is an
1904  * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
1905  * specified, as long as they are well-formed.
1906  */
1907 static int
1908 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
1909     spa_aux_vdev_t *sav, const char *config, uint64_t version,
1910     vdev_labeltype_t label)
1911 {
1912 	nvlist_t **dev;
1913 	uint_t i, ndev;
1914 	vdev_t *vd;
1915 	int error;
1916 
1917 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1918 
1919 	/*
1920 	 * It's acceptable to have no devs specified.
1921 	 */
1922 	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
1923 		return (0);
1924 
1925 	if (ndev == 0)
1926 		return (EINVAL);
1927 
1928 	/*
1929 	 * Make sure the pool is formatted with a version that supports this
1930 	 * device type.
1931 	 */
1932 	if (spa_version(spa) < version)
1933 		return (ENOTSUP);
1934 
1935 	/*
1936 	 * Set the pending device list so we correctly handle device in-use
1937 	 * checking.
1938 	 */
1939 	sav->sav_pending = dev;
1940 	sav->sav_npending = ndev;
1941 
1942 	for (i = 0; i < ndev; i++) {
1943 		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
1944 		    mode)) != 0)
1945 			goto out;
1946 
1947 		if (!vd->vdev_ops->vdev_op_leaf) {
1948 			vdev_free(vd);
1949 			error = EINVAL;
1950 			goto out;
1951 		}
1952 
1953 		/*
1954 		 * The L2ARC currently only supports disk devices in
1955 		 * kernel context.  For user-level testing, we allow it.
1956 		 */
1957 #ifdef _KERNEL
1958 		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
1959 		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
1960 			error = ENOTBLK;
1961 			goto out;
1962 		}
1963 #endif
1964 		vd->vdev_top = vd;
1965 
1966 		if ((error = vdev_open(vd)) == 0 &&
1967 		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
1968 			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
1969 			    vd->vdev_guid) == 0);
1970 		}
1971 
1972 		vdev_free(vd);
1973 
1974 		if (error &&
1975 		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
1976 			goto out;
1977 		else
1978 			error = 0;
1979 	}
1980 
1981 out:
1982 	sav->sav_pending = NULL;
1983 	sav->sav_npending = 0;
1984 	return (error);
1985 }
1986 
1987 static int
1988 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
1989 {
1990 	int error;
1991 
1992 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1993 
1994 	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
1995 	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
1996 	    VDEV_LABEL_SPARE)) != 0) {
1997 		return (error);
1998 	}
1999 
2000 	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2001 	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2002 	    VDEV_LABEL_L2CACHE));
2003 }
2004 
2005 static void
2006 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2007     const char *config)
2008 {
2009 	int i;
2010 
2011 	if (sav->sav_config != NULL) {
2012 		nvlist_t **olddevs;
2013 		uint_t oldndevs;
2014 		nvlist_t **newdevs;
2015 
2016 		/*
2017 		 * Generate new dev list by concatentating with the
2018 		 * current dev list.
2019 		 */
2020 		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2021 		    &olddevs, &oldndevs) == 0);
2022 
2023 		newdevs = kmem_alloc(sizeof (void *) *
2024 		    (ndevs + oldndevs), KM_SLEEP);
2025 		for (i = 0; i < oldndevs; i++)
2026 			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2027 			    KM_SLEEP) == 0);
2028 		for (i = 0; i < ndevs; i++)
2029 			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2030 			    KM_SLEEP) == 0);
2031 
2032 		VERIFY(nvlist_remove(sav->sav_config, config,
2033 		    DATA_TYPE_NVLIST_ARRAY) == 0);
2034 
2035 		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2036 		    config, newdevs, ndevs + oldndevs) == 0);
2037 		for (i = 0; i < oldndevs + ndevs; i++)
2038 			nvlist_free(newdevs[i]);
2039 		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2040 	} else {
2041 		/*
2042 		 * Generate a new dev list.
2043 		 */
2044 		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2045 		    KM_SLEEP) == 0);
2046 		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2047 		    devs, ndevs) == 0);
2048 	}
2049 }
2050 
2051 /*
2052  * Stop and drop level 2 ARC devices
2053  */
2054 void
2055 spa_l2cache_drop(spa_t *spa)
2056 {
2057 	vdev_t *vd;
2058 	int i;
2059 	spa_aux_vdev_t *sav = &spa->spa_l2cache;
2060 
2061 	for (i = 0; i < sav->sav_count; i++) {
2062 		uint64_t pool;
2063 
2064 		vd = sav->sav_vdevs[i];
2065 		ASSERT(vd != NULL);
2066 
2067 		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2068 		    pool != 0ULL && l2arc_vdev_present(vd))
2069 			l2arc_remove_vdev(vd);
2070 		if (vd->vdev_isl2cache)
2071 			spa_l2cache_remove(vd);
2072 		vdev_clear_stats(vd);
2073 		(void) vdev_close(vd);
2074 	}
2075 }
2076 
2077 /*
2078  * Pool Creation
2079  */
2080 int
2081 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2082     const char *history_str, nvlist_t *zplprops)
2083 {
2084 	spa_t *spa;
2085 	char *altroot = NULL;
2086 	vdev_t *rvd;
2087 	dsl_pool_t *dp;
2088 	dmu_tx_t *tx;
2089 	int c, error = 0;
2090 	uint64_t txg = TXG_INITIAL;
2091 	nvlist_t **spares, **l2cache;
2092 	uint_t nspares, nl2cache;
2093 	uint64_t version;
2094 
2095 	/*
2096 	 * If this pool already exists, return failure.
2097 	 */
2098 	mutex_enter(&spa_namespace_lock);
2099 	if (spa_lookup(pool) != NULL) {
2100 		mutex_exit(&spa_namespace_lock);
2101 		return (EEXIST);
2102 	}
2103 
2104 	/*
2105 	 * Allocate a new spa_t structure.
2106 	 */
2107 	(void) nvlist_lookup_string(props,
2108 	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2109 	spa = spa_add(pool, altroot);
2110 	spa_activate(spa, spa_mode_global);
2111 
2112 	spa->spa_uberblock.ub_txg = txg - 1;
2113 
2114 	if (props && (error = spa_prop_validate(spa, props))) {
2115 		spa_deactivate(spa);
2116 		spa_remove(spa);
2117 		mutex_exit(&spa_namespace_lock);
2118 		return (error);
2119 	}
2120 
2121 	if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2122 	    &version) != 0)
2123 		version = SPA_VERSION;
2124 	ASSERT(version <= SPA_VERSION);
2125 	spa->spa_uberblock.ub_version = version;
2126 	spa->spa_ubsync = spa->spa_uberblock;
2127 
2128 	/*
2129 	 * Create "The Godfather" zio to hold all async IOs
2130 	 */
2131 	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2132 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2133 
2134 	/*
2135 	 * Create the root vdev.
2136 	 */
2137 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2138 
2139 	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2140 
2141 	ASSERT(error != 0 || rvd != NULL);
2142 	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2143 
2144 	if (error == 0 && !zfs_allocatable_devs(nvroot))
2145 		error = EINVAL;
2146 
2147 	if (error == 0 &&
2148 	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2149 	    (error = spa_validate_aux(spa, nvroot, txg,
2150 	    VDEV_ALLOC_ADD)) == 0) {
2151 		for (c = 0; c < rvd->vdev_children; c++)
2152 			vdev_init(rvd->vdev_child[c], txg);
2153 		vdev_config_dirty(rvd);
2154 	}
2155 
2156 	spa_config_exit(spa, SCL_ALL, FTAG);
2157 
2158 	if (error != 0) {
2159 		spa_unload(spa);
2160 		spa_deactivate(spa);
2161 		spa_remove(spa);
2162 		mutex_exit(&spa_namespace_lock);
2163 		return (error);
2164 	}
2165 
2166 	/*
2167 	 * Get the list of spares, if specified.
2168 	 */
2169 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2170 	    &spares, &nspares) == 0) {
2171 		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
2172 		    KM_SLEEP) == 0);
2173 		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2174 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2175 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2176 		spa_load_spares(spa);
2177 		spa_config_exit(spa, SCL_ALL, FTAG);
2178 		spa->spa_spares.sav_sync = B_TRUE;
2179 	}
2180 
2181 	/*
2182 	 * Get the list of level 2 cache devices, if specified.
2183 	 */
2184 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2185 	    &l2cache, &nl2cache) == 0) {
2186 		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2187 		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2188 		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2189 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2190 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2191 		spa_load_l2cache(spa);
2192 		spa_config_exit(spa, SCL_ALL, FTAG);
2193 		spa->spa_l2cache.sav_sync = B_TRUE;
2194 	}
2195 
2196 	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
2197 	spa->spa_meta_objset = dp->dp_meta_objset;
2198 
2199 	tx = dmu_tx_create_assigned(dp, txg);
2200 
2201 	/*
2202 	 * Create the pool config object.
2203 	 */
2204 	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
2205 	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
2206 	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2207 
2208 	if (zap_add(spa->spa_meta_objset,
2209 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
2210 	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
2211 		cmn_err(CE_PANIC, "failed to add pool config");
2212 	}
2213 
2214 	/* Newly created pools with the right version are always deflated. */
2215 	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
2216 		spa->spa_deflate = TRUE;
2217 		if (zap_add(spa->spa_meta_objset,
2218 		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
2219 		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
2220 			cmn_err(CE_PANIC, "failed to add deflate");
2221 		}
2222 	}
2223 
2224 	/*
2225 	 * Create the deferred-free bplist object.  Turn off compression
2226 	 * because sync-to-convergence takes longer if the blocksize
2227 	 * keeps changing.
2228 	 */
2229 	spa->spa_sync_bplist_obj = bplist_create(spa->spa_meta_objset,
2230 	    1 << 14, tx);
2231 	dmu_object_set_compress(spa->spa_meta_objset, spa->spa_sync_bplist_obj,
2232 	    ZIO_COMPRESS_OFF, tx);
2233 
2234 	if (zap_add(spa->spa_meta_objset,
2235 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
2236 	    sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj, tx) != 0) {
2237 		cmn_err(CE_PANIC, "failed to add bplist");
2238 	}
2239 
2240 	/*
2241 	 * Create the pool's history object.
2242 	 */
2243 	if (version >= SPA_VERSION_ZPOOL_HISTORY)
2244 		spa_history_create_obj(spa, tx);
2245 
2246 	/*
2247 	 * Set pool properties.
2248 	 */
2249 	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
2250 	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2251 	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
2252 	if (props != NULL) {
2253 		spa_configfile_set(spa, props, B_FALSE);
2254 		spa_sync_props(spa, props, CRED(), tx);
2255 	}
2256 
2257 	dmu_tx_commit(tx);
2258 
2259 	spa->spa_sync_on = B_TRUE;
2260 	txg_sync_start(spa->spa_dsl_pool);
2261 
2262 	/*
2263 	 * We explicitly wait for the first transaction to complete so that our
2264 	 * bean counters are appropriately updated.
2265 	 */
2266 	txg_wait_synced(spa->spa_dsl_pool, txg);
2267 
2268 	spa_config_sync(spa, B_FALSE, B_TRUE);
2269 
2270 	if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
2271 		(void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
2272 
2273 	spa->spa_minref = refcount_count(&spa->spa_refcount);
2274 
2275 	mutex_exit(&spa_namespace_lock);
2276 
2277 	return (0);
2278 }
2279 
2280 #ifdef _KERNEL
2281 /*
2282  * Build a "root" vdev for a top level vdev read in from a rootpool
2283  * device label.
2284  */
2285 static void
2286 spa_build_rootpool_config(nvlist_t *config)
2287 {
2288 	nvlist_t *nvtop, *nvroot;
2289 	uint64_t pgid;
2290 
2291 	/*
2292 	 * Add this top-level vdev to the child array.
2293 	 */
2294 	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtop)
2295 	    == 0);
2296 	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pgid)
2297 	    == 0);
2298 
2299 	/*
2300 	 * Put this pool's top-level vdevs into a root vdev.
2301 	 */
2302 	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2303 	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT)
2304 	    == 0);
2305 	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
2306 	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
2307 	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2308 	    &nvtop, 1) == 0);
2309 
2310 	/*
2311 	 * Replace the existing vdev_tree with the new root vdev in
2312 	 * this pool's configuration (remove the old, add the new).
2313 	 */
2314 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
2315 	nvlist_free(nvroot);
2316 }
2317 
2318 /*
2319  * Get the root pool information from the root disk, then import the root pool
2320  * during the system boot up time.
2321  */
2322 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
2323 
2324 int
2325 spa_check_rootconf(char *devpath, char *devid, nvlist_t **bestconf,
2326     uint64_t *besttxg)
2327 {
2328 	nvlist_t *config;
2329 	uint64_t txg;
2330 	int error;
2331 
2332 	if (error = vdev_disk_read_rootlabel(devpath, devid, &config))
2333 		return (error);
2334 
2335 	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
2336 
2337 	if (bestconf != NULL)
2338 		*bestconf = config;
2339 	else
2340 		nvlist_free(config);
2341 	*besttxg = txg;
2342 	return (0);
2343 }
2344 
2345 boolean_t
2346 spa_rootdev_validate(nvlist_t *nv)
2347 {
2348 	uint64_t ival;
2349 
2350 	if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
2351 	    nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
2352 	    nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
2353 		return (B_FALSE);
2354 
2355 	return (B_TRUE);
2356 }
2357 
2358 
2359 /*
2360  * Given the boot device's physical path or devid, check if the device
2361  * is in a valid state.  If so, return the configuration from the vdev
2362  * label.
2363  */
2364 int
2365 spa_get_rootconf(char *devpath, char *devid, nvlist_t **bestconf)
2366 {
2367 	nvlist_t *conf = NULL;
2368 	uint64_t txg = 0;
2369 	nvlist_t *nvtop, **child;
2370 	char *type;
2371 	char *bootpath = NULL;
2372 	uint_t children, c;
2373 	char *tmp;
2374 	int error;
2375 
2376 	if (devpath && ((tmp = strchr(devpath, ' ')) != NULL))
2377 		*tmp = '\0';
2378 	if (error = spa_check_rootconf(devpath, devid, &conf, &txg)) {
2379 		cmn_err(CE_NOTE, "error reading device label");
2380 		return (error);
2381 	}
2382 	if (txg == 0) {
2383 		cmn_err(CE_NOTE, "this device is detached");
2384 		nvlist_free(conf);
2385 		return (EINVAL);
2386 	}
2387 
2388 	VERIFY(nvlist_lookup_nvlist(conf, ZPOOL_CONFIG_VDEV_TREE,
2389 	    &nvtop) == 0);
2390 	VERIFY(nvlist_lookup_string(nvtop, ZPOOL_CONFIG_TYPE, &type) == 0);
2391 
2392 	if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2393 		if (spa_rootdev_validate(nvtop)) {
2394 			goto out;
2395 		} else {
2396 			nvlist_free(conf);
2397 			return (EINVAL);
2398 		}
2399 	}
2400 
2401 	ASSERT(strcmp(type, VDEV_TYPE_MIRROR) == 0);
2402 
2403 	VERIFY(nvlist_lookup_nvlist_array(nvtop, ZPOOL_CONFIG_CHILDREN,
2404 	    &child, &children) == 0);
2405 
2406 	/*
2407 	 * Go thru vdevs in the mirror to see if the given device
2408 	 * has the most recent txg. Only the device with the most
2409 	 * recent txg has valid information and should be booted.
2410 	 */
2411 	for (c = 0; c < children; c++) {
2412 		char *cdevid, *cpath;
2413 		uint64_t tmptxg;
2414 
2415 		cpath = NULL;
2416 		cdevid = NULL;
2417 		(void) nvlist_lookup_string(child[c], ZPOOL_CONFIG_PHYS_PATH,
2418 		    &cpath);
2419 		(void) nvlist_lookup_string(child[c], ZPOOL_CONFIG_DEVID,
2420 		    &cdevid);
2421 		if (cpath == NULL && cdevid == NULL)
2422 			return (EINVAL);
2423 		if ((spa_check_rootconf(cpath, cdevid, NULL,
2424 		    &tmptxg) == 0) && (tmptxg > txg)) {
2425 			txg = tmptxg;
2426 			VERIFY(nvlist_lookup_string(child[c],
2427 			    ZPOOL_CONFIG_PATH, &bootpath) == 0);
2428 		}
2429 	}
2430 
2431 	/* Does the best device match the one we've booted from? */
2432 	if (bootpath) {
2433 		cmn_err(CE_NOTE, "try booting from '%s'", bootpath);
2434 		return (EINVAL);
2435 	}
2436 out:
2437 	*bestconf = conf;
2438 	return (0);
2439 }
2440 
2441 /*
2442  * Import a root pool.
2443  *
2444  * For x86. devpath_list will consist of devid and/or physpath name of
2445  * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
2446  * The GRUB "findroot" command will return the vdev we should boot.
2447  *
2448  * For Sparc, devpath_list consists the physpath name of the booting device
2449  * no matter the rootpool is a single device pool or a mirrored pool.
2450  * e.g.
2451  *	"/pci@1f,0/ide@d/disk@0,0:a"
2452  */
2453 int
2454 spa_import_rootpool(char *devpath, char *devid)
2455 {
2456 	nvlist_t *conf = NULL;
2457 	char *pname;
2458 	int error;
2459 	spa_t *spa;
2460 
2461 	/*
2462 	 * Get the vdev pathname and configuation from the most
2463 	 * recently updated vdev (highest txg).
2464 	 */
2465 	if (error = spa_get_rootconf(devpath, devid, &conf))
2466 		goto msg_out;
2467 
2468 	/*
2469 	 * Add type "root" vdev to the config.
2470 	 */
2471 	spa_build_rootpool_config(conf);
2472 
2473 	VERIFY(nvlist_lookup_string(conf, ZPOOL_CONFIG_POOL_NAME, &pname) == 0);
2474 
2475 	mutex_enter(&spa_namespace_lock);
2476 	if ((spa = spa_lookup(pname)) != NULL) {
2477 		/*
2478 		 * Remove the existing root pool from the namespace so that we
2479 		 * can replace it with the correct config we just read in.
2480 		 */
2481 		spa_remove(spa);
2482 	}
2483 
2484 	spa = spa_add(pname, NULL);
2485 
2486 	spa->spa_is_root = B_TRUE;
2487 	VERIFY(nvlist_dup(conf, &spa->spa_config, 0) == 0);
2488 	mutex_exit(&spa_namespace_lock);
2489 
2490 	nvlist_free(conf);
2491 	return (0);
2492 
2493 msg_out:
2494 	cmn_err(CE_NOTE, "\n"
2495 	    "  ***************************************************  \n"
2496 	    "  *  This device is not bootable!                   *  \n"
2497 	    "  *  It is either offlined or detached or faulted.  *  \n"
2498 	    "  *  Please try to boot from a different device.    *  \n"
2499 	    "  ***************************************************  ");
2500 
2501 	return (error);
2502 }
2503 #endif
2504 
2505 /*
2506  * Take a pool and insert it into the namespace as if it had been loaded at
2507  * boot.
2508  */
2509 int
2510 spa_import_verbatim(const char *pool, nvlist_t *config, nvlist_t *props)
2511 {
2512 	spa_t *spa;
2513 	char *altroot = NULL;
2514 
2515 	mutex_enter(&spa_namespace_lock);
2516 	if (spa_lookup(pool) != NULL) {
2517 		mutex_exit(&spa_namespace_lock);
2518 		return (EEXIST);
2519 	}
2520 
2521 	(void) nvlist_lookup_string(props,
2522 	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2523 	spa = spa_add(pool, altroot);
2524 
2525 	VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
2526 
2527 	if (props != NULL)
2528 		spa_configfile_set(spa, props, B_FALSE);
2529 
2530 	spa_config_sync(spa, B_FALSE, B_TRUE);
2531 
2532 	mutex_exit(&spa_namespace_lock);
2533 
2534 	return (0);
2535 }
2536 
2537 /*
2538  * Import a non-root pool into the system.
2539  */
2540 int
2541 spa_import(const char *pool, nvlist_t *config, nvlist_t *props)
2542 {
2543 	spa_t *spa;
2544 	char *altroot = NULL;
2545 	int error;
2546 	nvlist_t *nvroot;
2547 	nvlist_t **spares, **l2cache;
2548 	uint_t nspares, nl2cache;
2549 
2550 	/*
2551 	 * If a pool with this name exists, return failure.
2552 	 */
2553 	mutex_enter(&spa_namespace_lock);
2554 	if ((spa = spa_lookup(pool)) != NULL) {
2555 		mutex_exit(&spa_namespace_lock);
2556 		return (EEXIST);
2557 	}
2558 
2559 	/*
2560 	 * Create and initialize the spa structure.
2561 	 */
2562 	(void) nvlist_lookup_string(props,
2563 	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2564 	spa = spa_add(pool, altroot);
2565 	spa_activate(spa, spa_mode_global);
2566 
2567 	/*
2568 	 * Don't start async tasks until we know everything is healthy.
2569 	 */
2570 	spa_async_suspend(spa);
2571 
2572 	/*
2573 	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
2574 	 * because the user-supplied config is actually the one to trust when
2575 	 * doing an import.
2576 	 */
2577 	error = spa_load(spa, config, SPA_LOAD_IMPORT, B_TRUE);
2578 
2579 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2580 	/*
2581 	 * Toss any existing sparelist, as it doesn't have any validity
2582 	 * anymore, and conflicts with spa_has_spare().
2583 	 */
2584 	if (spa->spa_spares.sav_config) {
2585 		nvlist_free(spa->spa_spares.sav_config);
2586 		spa->spa_spares.sav_config = NULL;
2587 		spa_load_spares(spa);
2588 	}
2589 	if (spa->spa_l2cache.sav_config) {
2590 		nvlist_free(spa->spa_l2cache.sav_config);
2591 		spa->spa_l2cache.sav_config = NULL;
2592 		spa_load_l2cache(spa);
2593 	}
2594 
2595 	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2596 	    &nvroot) == 0);
2597 	if (error == 0)
2598 		error = spa_validate_aux(spa, nvroot, -1ULL,
2599 		    VDEV_ALLOC_SPARE);
2600 	if (error == 0)
2601 		error = spa_validate_aux(spa, nvroot, -1ULL,
2602 		    VDEV_ALLOC_L2CACHE);
2603 	spa_config_exit(spa, SCL_ALL, FTAG);
2604 
2605 	if (props != NULL)
2606 		spa_configfile_set(spa, props, B_FALSE);
2607 
2608 	if (error != 0 || (props && spa_writeable(spa) &&
2609 	    (error = spa_prop_set(spa, props)))) {
2610 		spa_unload(spa);
2611 		spa_deactivate(spa);
2612 		spa_remove(spa);
2613 		mutex_exit(&spa_namespace_lock);
2614 		return (error);
2615 	}
2616 
2617 	spa_async_resume(spa);
2618 
2619 	/*
2620 	 * Override any spares and level 2 cache devices as specified by
2621 	 * the user, as these may have correct device names/devids, etc.
2622 	 */
2623 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2624 	    &spares, &nspares) == 0) {
2625 		if (spa->spa_spares.sav_config)
2626 			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
2627 			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
2628 		else
2629 			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
2630 			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2631 		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2632 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2633 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2634 		spa_load_spares(spa);
2635 		spa_config_exit(spa, SCL_ALL, FTAG);
2636 		spa->spa_spares.sav_sync = B_TRUE;
2637 	}
2638 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2639 	    &l2cache, &nl2cache) == 0) {
2640 		if (spa->spa_l2cache.sav_config)
2641 			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
2642 			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
2643 		else
2644 			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2645 			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2646 		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2647 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2648 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2649 		spa_load_l2cache(spa);
2650 		spa_config_exit(spa, SCL_ALL, FTAG);
2651 		spa->spa_l2cache.sav_sync = B_TRUE;
2652 	}
2653 
2654 	if (spa_writeable(spa)) {
2655 		/*
2656 		 * Update the config cache to include the newly-imported pool.
2657 		 */
2658 		spa_config_update_common(spa, SPA_CONFIG_UPDATE_POOL, B_FALSE);
2659 	}
2660 
2661 	mutex_exit(&spa_namespace_lock);
2662 
2663 	return (0);
2664 }
2665 
2666 
2667 /*
2668  * This (illegal) pool name is used when temporarily importing a spa_t in order
2669  * to get the vdev stats associated with the imported devices.
2670  */
2671 #define	TRYIMPORT_NAME	"$import"
2672 
2673 nvlist_t *
2674 spa_tryimport(nvlist_t *tryconfig)
2675 {
2676 	nvlist_t *config = NULL;
2677 	char *poolname;
2678 	spa_t *spa;
2679 	uint64_t state;
2680 	int error;
2681 
2682 	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
2683 		return (NULL);
2684 
2685 	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
2686 		return (NULL);
2687 
2688 	/*
2689 	 * Create and initialize the spa structure.
2690 	 */
2691 	mutex_enter(&spa_namespace_lock);
2692 	spa = spa_add(TRYIMPORT_NAME, NULL);
2693 	spa_activate(spa, FREAD);
2694 
2695 	/*
2696 	 * Pass off the heavy lifting to spa_load().
2697 	 * Pass TRUE for mosconfig because the user-supplied config
2698 	 * is actually the one to trust when doing an import.
2699 	 */
2700 	error = spa_load(spa, tryconfig, SPA_LOAD_TRYIMPORT, B_TRUE);
2701 
2702 	/*
2703 	 * If 'tryconfig' was at least parsable, return the current config.
2704 	 */
2705 	if (spa->spa_root_vdev != NULL) {
2706 		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2707 		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
2708 		    poolname) == 0);
2709 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
2710 		    state) == 0);
2711 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
2712 		    spa->spa_uberblock.ub_timestamp) == 0);
2713 
2714 		/*
2715 		 * If the bootfs property exists on this pool then we
2716 		 * copy it out so that external consumers can tell which
2717 		 * pools are bootable.
2718 		 */
2719 		if ((!error || error == EEXIST) && spa->spa_bootfs) {
2720 			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2721 
2722 			/*
2723 			 * We have to play games with the name since the
2724 			 * pool was opened as TRYIMPORT_NAME.
2725 			 */
2726 			if (dsl_dsobj_to_dsname(spa_name(spa),
2727 			    spa->spa_bootfs, tmpname) == 0) {
2728 				char *cp;
2729 				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2730 
2731 				cp = strchr(tmpname, '/');
2732 				if (cp == NULL) {
2733 					(void) strlcpy(dsname, tmpname,
2734 					    MAXPATHLEN);
2735 				} else {
2736 					(void) snprintf(dsname, MAXPATHLEN,
2737 					    "%s/%s", poolname, ++cp);
2738 				}
2739 				VERIFY(nvlist_add_string(config,
2740 				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
2741 				kmem_free(dsname, MAXPATHLEN);
2742 			}
2743 			kmem_free(tmpname, MAXPATHLEN);
2744 		}
2745 
2746 		/*
2747 		 * Add the list of hot spares and level 2 cache devices.
2748 		 */
2749 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2750 		spa_add_spares(spa, config);
2751 		spa_add_l2cache(spa, config);
2752 		spa_config_exit(spa, SCL_CONFIG, FTAG);
2753 	}
2754 
2755 	spa_unload(spa);
2756 	spa_deactivate(spa);
2757 	spa_remove(spa);
2758 	mutex_exit(&spa_namespace_lock);
2759 
2760 	return (config);
2761 }
2762 
2763 /*
2764  * Pool export/destroy
2765  *
2766  * The act of destroying or exporting a pool is very simple.  We make sure there
2767  * is no more pending I/O and any references to the pool are gone.  Then, we
2768  * update the pool state and sync all the labels to disk, removing the
2769  * configuration from the cache afterwards. If the 'hardforce' flag is set, then
2770  * we don't sync the labels or remove the configuration cache.
2771  */
2772 static int
2773 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
2774     boolean_t force, boolean_t hardforce)
2775 {
2776 	spa_t *spa;
2777 
2778 	if (oldconfig)
2779 		*oldconfig = NULL;
2780 
2781 	if (!(spa_mode_global & FWRITE))
2782 		return (EROFS);
2783 
2784 	mutex_enter(&spa_namespace_lock);
2785 	if ((spa = spa_lookup(pool)) == NULL) {
2786 		mutex_exit(&spa_namespace_lock);
2787 		return (ENOENT);
2788 	}
2789 
2790 	/*
2791 	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
2792 	 * reacquire the namespace lock, and see if we can export.
2793 	 */
2794 	spa_open_ref(spa, FTAG);
2795 	mutex_exit(&spa_namespace_lock);
2796 	spa_async_suspend(spa);
2797 	mutex_enter(&spa_namespace_lock);
2798 	spa_close(spa, FTAG);
2799 
2800 	/*
2801 	 * The pool will be in core if it's openable,
2802 	 * in which case we can modify its state.
2803 	 */
2804 	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
2805 		/*
2806 		 * Objsets may be open only because they're dirty, so we
2807 		 * have to force it to sync before checking spa_refcnt.
2808 		 */
2809 		txg_wait_synced(spa->spa_dsl_pool, 0);
2810 
2811 		/*
2812 		 * A pool cannot be exported or destroyed if there are active
2813 		 * references.  If we are resetting a pool, allow references by
2814 		 * fault injection handlers.
2815 		 */
2816 		if (!spa_refcount_zero(spa) ||
2817 		    (spa->spa_inject_ref != 0 &&
2818 		    new_state != POOL_STATE_UNINITIALIZED)) {
2819 			spa_async_resume(spa);
2820 			mutex_exit(&spa_namespace_lock);
2821 			return (EBUSY);
2822 		}
2823 
2824 		/*
2825 		 * A pool cannot be exported if it has an active shared spare.
2826 		 * This is to prevent other pools stealing the active spare
2827 		 * from an exported pool. At user's own will, such pool can
2828 		 * be forcedly exported.
2829 		 */
2830 		if (!force && new_state == POOL_STATE_EXPORTED &&
2831 		    spa_has_active_shared_spare(spa)) {
2832 			spa_async_resume(spa);
2833 			mutex_exit(&spa_namespace_lock);
2834 			return (EXDEV);
2835 		}
2836 
2837 		/*
2838 		 * We want this to be reflected on every label,
2839 		 * so mark them all dirty.  spa_unload() will do the
2840 		 * final sync that pushes these changes out.
2841 		 */
2842 		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
2843 			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2844 			spa->spa_state = new_state;
2845 			spa->spa_final_txg = spa_last_synced_txg(spa) + 1;
2846 			vdev_config_dirty(spa->spa_root_vdev);
2847 			spa_config_exit(spa, SCL_ALL, FTAG);
2848 		}
2849 	}
2850 
2851 	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
2852 
2853 	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
2854 		spa_unload(spa);
2855 		spa_deactivate(spa);
2856 	}
2857 
2858 	if (oldconfig && spa->spa_config)
2859 		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
2860 
2861 	if (new_state != POOL_STATE_UNINITIALIZED) {
2862 		if (!hardforce)
2863 			spa_config_sync(spa, B_TRUE, B_TRUE);
2864 		spa_remove(spa);
2865 	}
2866 	mutex_exit(&spa_namespace_lock);
2867 
2868 	return (0);
2869 }
2870 
2871 /*
2872  * Destroy a storage pool.
2873  */
2874 int
2875 spa_destroy(char *pool)
2876 {
2877 	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
2878 	    B_FALSE, B_FALSE));
2879 }
2880 
2881 /*
2882  * Export a storage pool.
2883  */
2884 int
2885 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
2886     boolean_t hardforce)
2887 {
2888 	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
2889 	    force, hardforce));
2890 }
2891 
2892 /*
2893  * Similar to spa_export(), this unloads the spa_t without actually removing it
2894  * from the namespace in any way.
2895  */
2896 int
2897 spa_reset(char *pool)
2898 {
2899 	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
2900 	    B_FALSE, B_FALSE));
2901 }
2902 
2903 /*
2904  * ==========================================================================
2905  * Device manipulation
2906  * ==========================================================================
2907  */
2908 
2909 /*
2910  * Add a device to a storage pool.
2911  */
2912 int
2913 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
2914 {
2915 	uint64_t txg;
2916 	int error;
2917 	vdev_t *rvd = spa->spa_root_vdev;
2918 	vdev_t *vd, *tvd;
2919 	nvlist_t **spares, **l2cache;
2920 	uint_t nspares, nl2cache;
2921 
2922 	txg = spa_vdev_enter(spa);
2923 
2924 	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
2925 	    VDEV_ALLOC_ADD)) != 0)
2926 		return (spa_vdev_exit(spa, NULL, txg, error));
2927 
2928 	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
2929 
2930 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
2931 	    &nspares) != 0)
2932 		nspares = 0;
2933 
2934 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
2935 	    &nl2cache) != 0)
2936 		nl2cache = 0;
2937 
2938 	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
2939 		return (spa_vdev_exit(spa, vd, txg, EINVAL));
2940 
2941 	if (vd->vdev_children != 0 &&
2942 	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
2943 		return (spa_vdev_exit(spa, vd, txg, error));
2944 
2945 	/*
2946 	 * We must validate the spares and l2cache devices after checking the
2947 	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
2948 	 */
2949 	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
2950 		return (spa_vdev_exit(spa, vd, txg, error));
2951 
2952 	/*
2953 	 * Transfer each new top-level vdev from vd to rvd.
2954 	 */
2955 	for (int c = 0; c < vd->vdev_children; c++) {
2956 		tvd = vd->vdev_child[c];
2957 		vdev_remove_child(vd, tvd);
2958 		tvd->vdev_id = rvd->vdev_children;
2959 		vdev_add_child(rvd, tvd);
2960 		vdev_config_dirty(tvd);
2961 	}
2962 
2963 	if (nspares != 0) {
2964 		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
2965 		    ZPOOL_CONFIG_SPARES);
2966 		spa_load_spares(spa);
2967 		spa->spa_spares.sav_sync = B_TRUE;
2968 	}
2969 
2970 	if (nl2cache != 0) {
2971 		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
2972 		    ZPOOL_CONFIG_L2CACHE);
2973 		spa_load_l2cache(spa);
2974 		spa->spa_l2cache.sav_sync = B_TRUE;
2975 	}
2976 
2977 	/*
2978 	 * We have to be careful when adding new vdevs to an existing pool.
2979 	 * If other threads start allocating from these vdevs before we
2980 	 * sync the config cache, and we lose power, then upon reboot we may
2981 	 * fail to open the pool because there are DVAs that the config cache
2982 	 * can't translate.  Therefore, we first add the vdevs without
2983 	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
2984 	 * and then let spa_config_update() initialize the new metaslabs.
2985 	 *
2986 	 * spa_load() checks for added-but-not-initialized vdevs, so that
2987 	 * if we lose power at any point in this sequence, the remaining
2988 	 * steps will be completed the next time we load the pool.
2989 	 */
2990 	(void) spa_vdev_exit(spa, vd, txg, 0);
2991 
2992 	mutex_enter(&spa_namespace_lock);
2993 	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
2994 	mutex_exit(&spa_namespace_lock);
2995 
2996 	return (0);
2997 }
2998 
2999 /*
3000  * Attach a device to a mirror.  The arguments are the path to any device
3001  * in the mirror, and the nvroot for the new device.  If the path specifies
3002  * a device that is not mirrored, we automatically insert the mirror vdev.
3003  *
3004  * If 'replacing' is specified, the new device is intended to replace the
3005  * existing device; in this case the two devices are made into their own
3006  * mirror using the 'replacing' vdev, which is functionally identical to
3007  * the mirror vdev (it actually reuses all the same ops) but has a few
3008  * extra rules: you can't attach to it after it's been created, and upon
3009  * completion of resilvering, the first disk (the one being replaced)
3010  * is automatically detached.
3011  */
3012 int
3013 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3014 {
3015 	uint64_t txg, open_txg;
3016 	vdev_t *rvd = spa->spa_root_vdev;
3017 	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3018 	vdev_ops_t *pvops;
3019 	dmu_tx_t *tx;
3020 	char *oldvdpath, *newvdpath;
3021 	int newvd_isspare;
3022 	int error;
3023 
3024 	txg = spa_vdev_enter(spa);
3025 
3026 	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3027 
3028 	if (oldvd == NULL)
3029 		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3030 
3031 	if (!oldvd->vdev_ops->vdev_op_leaf)
3032 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3033 
3034 	pvd = oldvd->vdev_parent;
3035 
3036 	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3037 	    VDEV_ALLOC_ADD)) != 0)
3038 		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3039 
3040 	if (newrootvd->vdev_children != 1)
3041 		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3042 
3043 	newvd = newrootvd->vdev_child[0];
3044 
3045 	if (!newvd->vdev_ops->vdev_op_leaf)
3046 		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3047 
3048 	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3049 		return (spa_vdev_exit(spa, newrootvd, txg, error));
3050 
3051 	/*
3052 	 * Spares can't replace logs
3053 	 */
3054 	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3055 		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3056 
3057 	if (!replacing) {
3058 		/*
3059 		 * For attach, the only allowable parent is a mirror or the root
3060 		 * vdev.
3061 		 */
3062 		if (pvd->vdev_ops != &vdev_mirror_ops &&
3063 		    pvd->vdev_ops != &vdev_root_ops)
3064 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3065 
3066 		pvops = &vdev_mirror_ops;
3067 	} else {
3068 		/*
3069 		 * Active hot spares can only be replaced by inactive hot
3070 		 * spares.
3071 		 */
3072 		if (pvd->vdev_ops == &vdev_spare_ops &&
3073 		    pvd->vdev_child[1] == oldvd &&
3074 		    !spa_has_spare(spa, newvd->vdev_guid))
3075 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3076 
3077 		/*
3078 		 * If the source is a hot spare, and the parent isn't already a
3079 		 * spare, then we want to create a new hot spare.  Otherwise, we
3080 		 * want to create a replacing vdev.  The user is not allowed to
3081 		 * attach to a spared vdev child unless the 'isspare' state is
3082 		 * the same (spare replaces spare, non-spare replaces
3083 		 * non-spare).
3084 		 */
3085 		if (pvd->vdev_ops == &vdev_replacing_ops)
3086 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3087 		else if (pvd->vdev_ops == &vdev_spare_ops &&
3088 		    newvd->vdev_isspare != oldvd->vdev_isspare)
3089 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3090 		else if (pvd->vdev_ops != &vdev_spare_ops &&
3091 		    newvd->vdev_isspare)
3092 			pvops = &vdev_spare_ops;
3093 		else
3094 			pvops = &vdev_replacing_ops;
3095 	}
3096 
3097 	/*
3098 	 * Compare the new device size with the replaceable/attachable
3099 	 * device size.
3100 	 */
3101 	if (newvd->vdev_psize < vdev_get_rsize(oldvd))
3102 		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3103 
3104 	/*
3105 	 * The new device cannot have a higher alignment requirement
3106 	 * than the top-level vdev.
3107 	 */
3108 	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3109 		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3110 
3111 	/*
3112 	 * If this is an in-place replacement, update oldvd's path and devid
3113 	 * to make it distinguishable from newvd, and unopenable from now on.
3114 	 */
3115 	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3116 		spa_strfree(oldvd->vdev_path);
3117 		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3118 		    KM_SLEEP);
3119 		(void) sprintf(oldvd->vdev_path, "%s/%s",
3120 		    newvd->vdev_path, "old");
3121 		if (oldvd->vdev_devid != NULL) {
3122 			spa_strfree(oldvd->vdev_devid);
3123 			oldvd->vdev_devid = NULL;
3124 		}
3125 	}
3126 
3127 	/*
3128 	 * If the parent is not a mirror, or if we're replacing, insert the new
3129 	 * mirror/replacing/spare vdev above oldvd.
3130 	 */
3131 	if (pvd->vdev_ops != pvops)
3132 		pvd = vdev_add_parent(oldvd, pvops);
3133 
3134 	ASSERT(pvd->vdev_top->vdev_parent == rvd);
3135 	ASSERT(pvd->vdev_ops == pvops);
3136 	ASSERT(oldvd->vdev_parent == pvd);
3137 
3138 	/*
3139 	 * Extract the new device from its root and add it to pvd.
3140 	 */
3141 	vdev_remove_child(newrootvd, newvd);
3142 	newvd->vdev_id = pvd->vdev_children;
3143 	vdev_add_child(pvd, newvd);
3144 
3145 	/*
3146 	 * If newvd is smaller than oldvd, but larger than its rsize,
3147 	 * the addition of newvd may have decreased our parent's asize.
3148 	 */
3149 	pvd->vdev_asize = MIN(pvd->vdev_asize, newvd->vdev_asize);
3150 
3151 	tvd = newvd->vdev_top;
3152 	ASSERT(pvd->vdev_top == tvd);
3153 	ASSERT(tvd->vdev_parent == rvd);
3154 
3155 	vdev_config_dirty(tvd);
3156 
3157 	/*
3158 	 * Set newvd's DTL to [TXG_INITIAL, open_txg].  It will propagate
3159 	 * upward when spa_vdev_exit() calls vdev_dtl_reassess().
3160 	 */
3161 	open_txg = txg + TXG_CONCURRENT_STATES - 1;
3162 
3163 	vdev_dtl_dirty(newvd, DTL_MISSING,
3164 	    TXG_INITIAL, open_txg - TXG_INITIAL + 1);
3165 
3166 	if (newvd->vdev_isspare) {
3167 		spa_spare_activate(newvd);
3168 		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
3169 	}
3170 
3171 	oldvdpath = spa_strdup(oldvd->vdev_path);
3172 	newvdpath = spa_strdup(newvd->vdev_path);
3173 	newvd_isspare = newvd->vdev_isspare;
3174 
3175 	/*
3176 	 * Mark newvd's DTL dirty in this txg.
3177 	 */
3178 	vdev_dirty(tvd, VDD_DTL, newvd, txg);
3179 
3180 	(void) spa_vdev_exit(spa, newrootvd, open_txg, 0);
3181 
3182 	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
3183 	if (dmu_tx_assign(tx, TXG_WAIT) == 0) {
3184 		spa_history_internal_log(LOG_POOL_VDEV_ATTACH, spa, tx,
3185 		    CRED(),  "%s vdev=%s %s vdev=%s",
3186 		    replacing && newvd_isspare ? "spare in" :
3187 		    replacing ? "replace" : "attach", newvdpath,
3188 		    replacing ? "for" : "to", oldvdpath);
3189 		dmu_tx_commit(tx);
3190 	} else {
3191 		dmu_tx_abort(tx);
3192 	}
3193 
3194 	spa_strfree(oldvdpath);
3195 	spa_strfree(newvdpath);
3196 
3197 	/*
3198 	 * Kick off a resilver to update newvd.
3199 	 */
3200 	VERIFY3U(spa_scrub(spa, POOL_SCRUB_RESILVER), ==, 0);
3201 
3202 	return (0);
3203 }
3204 
3205 /*
3206  * Detach a device from a mirror or replacing vdev.
3207  * If 'replace_done' is specified, only detach if the parent
3208  * is a replacing vdev.
3209  */
3210 int
3211 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
3212 {
3213 	uint64_t txg;
3214 	int error;
3215 	vdev_t *rvd = spa->spa_root_vdev;
3216 	vdev_t *vd, *pvd, *cvd, *tvd;
3217 	boolean_t unspare = B_FALSE;
3218 	uint64_t unspare_guid;
3219 	size_t len;
3220 
3221 	txg = spa_vdev_enter(spa);
3222 
3223 	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3224 
3225 	if (vd == NULL)
3226 		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3227 
3228 	if (!vd->vdev_ops->vdev_op_leaf)
3229 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3230 
3231 	pvd = vd->vdev_parent;
3232 
3233 	/*
3234 	 * If the parent/child relationship is not as expected, don't do it.
3235 	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
3236 	 * vdev that's replacing B with C.  The user's intent in replacing
3237 	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
3238 	 * the replace by detaching C, the expected behavior is to end up
3239 	 * M(A,B).  But suppose that right after deciding to detach C,
3240 	 * the replacement of B completes.  We would have M(A,C), and then
3241 	 * ask to detach C, which would leave us with just A -- not what
3242 	 * the user wanted.  To prevent this, we make sure that the
3243 	 * parent/child relationship hasn't changed -- in this example,
3244 	 * that C's parent is still the replacing vdev R.
3245 	 */
3246 	if (pvd->vdev_guid != pguid && pguid != 0)
3247 		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3248 
3249 	/*
3250 	 * If replace_done is specified, only remove this device if it's
3251 	 * the first child of a replacing vdev.  For the 'spare' vdev, either
3252 	 * disk can be removed.
3253 	 */
3254 	if (replace_done) {
3255 		if (pvd->vdev_ops == &vdev_replacing_ops) {
3256 			if (vd->vdev_id != 0)
3257 				return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3258 		} else if (pvd->vdev_ops != &vdev_spare_ops) {
3259 			return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3260 		}
3261 	}
3262 
3263 	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
3264 	    spa_version(spa) >= SPA_VERSION_SPARES);
3265 
3266 	/*
3267 	 * Only mirror, replacing, and spare vdevs support detach.
3268 	 */
3269 	if (pvd->vdev_ops != &vdev_replacing_ops &&
3270 	    pvd->vdev_ops != &vdev_mirror_ops &&
3271 	    pvd->vdev_ops != &vdev_spare_ops)
3272 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3273 
3274 	/*
3275 	 * If this device has the only valid copy of some data,
3276 	 * we cannot safely detach it.
3277 	 */
3278 	if (vdev_dtl_required(vd))
3279 		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3280 
3281 	ASSERT(pvd->vdev_children >= 2);
3282 
3283 	/*
3284 	 * If we are detaching the second disk from a replacing vdev, then
3285 	 * check to see if we changed the original vdev's path to have "/old"
3286 	 * at the end in spa_vdev_attach().  If so, undo that change now.
3287 	 */
3288 	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id == 1 &&
3289 	    pvd->vdev_child[0]->vdev_path != NULL &&
3290 	    pvd->vdev_child[1]->vdev_path != NULL) {
3291 		ASSERT(pvd->vdev_child[1] == vd);
3292 		cvd = pvd->vdev_child[0];
3293 		len = strlen(vd->vdev_path);
3294 		if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
3295 		    strcmp(cvd->vdev_path + len, "/old") == 0) {
3296 			spa_strfree(cvd->vdev_path);
3297 			cvd->vdev_path = spa_strdup(vd->vdev_path);
3298 		}
3299 	}
3300 
3301 	/*
3302 	 * If we are detaching the original disk from a spare, then it implies
3303 	 * that the spare should become a real disk, and be removed from the
3304 	 * active spare list for the pool.
3305 	 */
3306 	if (pvd->vdev_ops == &vdev_spare_ops &&
3307 	    vd->vdev_id == 0 && pvd->vdev_child[1]->vdev_isspare)
3308 		unspare = B_TRUE;
3309 
3310 	/*
3311 	 * Erase the disk labels so the disk can be used for other things.
3312 	 * This must be done after all other error cases are handled,
3313 	 * but before we disembowel vd (so we can still do I/O to it).
3314 	 * But if we can't do it, don't treat the error as fatal --
3315 	 * it may be that the unwritability of the disk is the reason
3316 	 * it's being detached!
3317 	 */
3318 	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
3319 
3320 	/*
3321 	 * Remove vd from its parent and compact the parent's children.
3322 	 */
3323 	vdev_remove_child(pvd, vd);
3324 	vdev_compact_children(pvd);
3325 
3326 	/*
3327 	 * Remember one of the remaining children so we can get tvd below.
3328 	 */
3329 	cvd = pvd->vdev_child[0];
3330 
3331 	/*
3332 	 * If we need to remove the remaining child from the list of hot spares,
3333 	 * do it now, marking the vdev as no longer a spare in the process.
3334 	 * We must do this before vdev_remove_parent(), because that can
3335 	 * change the GUID if it creates a new toplevel GUID.  For a similar
3336 	 * reason, we must remove the spare now, in the same txg as the detach;
3337 	 * otherwise someone could attach a new sibling, change the GUID, and
3338 	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
3339 	 */
3340 	if (unspare) {
3341 		ASSERT(cvd->vdev_isspare);
3342 		spa_spare_remove(cvd);
3343 		unspare_guid = cvd->vdev_guid;
3344 		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
3345 	}
3346 
3347 	/*
3348 	 * If the parent mirror/replacing vdev only has one child,
3349 	 * the parent is no longer needed.  Remove it from the tree.
3350 	 */
3351 	if (pvd->vdev_children == 1)
3352 		vdev_remove_parent(cvd);
3353 
3354 	/*
3355 	 * We don't set tvd until now because the parent we just removed
3356 	 * may have been the previous top-level vdev.
3357 	 */
3358 	tvd = cvd->vdev_top;
3359 	ASSERT(tvd->vdev_parent == rvd);
3360 
3361 	/*
3362 	 * Reevaluate the parent vdev state.
3363 	 */
3364 	vdev_propagate_state(cvd);
3365 
3366 	/*
3367 	 * If the device we just detached was smaller than the others, it may be
3368 	 * possible to add metaslabs (i.e. grow the pool).  vdev_metaslab_init()
3369 	 * can't fail because the existing metaslabs are already in core, so
3370 	 * there's nothing to read from disk.
3371 	 */
3372 	VERIFY(vdev_metaslab_init(tvd, txg) == 0);
3373 
3374 	vdev_config_dirty(tvd);
3375 
3376 	/*
3377 	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
3378 	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
3379 	 * But first make sure we're not on any *other* txg's DTL list, to
3380 	 * prevent vd from being accessed after it's freed.
3381 	 */
3382 	for (int t = 0; t < TXG_SIZE; t++)
3383 		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
3384 	vd->vdev_detached = B_TRUE;
3385 	vdev_dirty(tvd, VDD_DTL, vd, txg);
3386 
3387 	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
3388 
3389 	error = spa_vdev_exit(spa, vd, txg, 0);
3390 
3391 	/*
3392 	 * If this was the removal of the original device in a hot spare vdev,
3393 	 * then we want to go through and remove the device from the hot spare
3394 	 * list of every other pool.
3395 	 */
3396 	if (unspare) {
3397 		spa_t *myspa = spa;
3398 		spa = NULL;
3399 		mutex_enter(&spa_namespace_lock);
3400 		while ((spa = spa_next(spa)) != NULL) {
3401 			if (spa->spa_state != POOL_STATE_ACTIVE)
3402 				continue;
3403 			if (spa == myspa)
3404 				continue;
3405 			spa_open_ref(spa, FTAG);
3406 			mutex_exit(&spa_namespace_lock);
3407 			(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
3408 			mutex_enter(&spa_namespace_lock);
3409 			spa_close(spa, FTAG);
3410 		}
3411 		mutex_exit(&spa_namespace_lock);
3412 	}
3413 
3414 	return (error);
3415 }
3416 
3417 static nvlist_t *
3418 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
3419 {
3420 	for (int i = 0; i < count; i++) {
3421 		uint64_t guid;
3422 
3423 		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
3424 		    &guid) == 0);
3425 
3426 		if (guid == target_guid)
3427 			return (nvpp[i]);
3428 	}
3429 
3430 	return (NULL);
3431 }
3432 
3433 static void
3434 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
3435 	nvlist_t *dev_to_remove)
3436 {
3437 	nvlist_t **newdev = NULL;
3438 
3439 	if (count > 1)
3440 		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
3441 
3442 	for (int i = 0, j = 0; i < count; i++) {
3443 		if (dev[i] == dev_to_remove)
3444 			continue;
3445 		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
3446 	}
3447 
3448 	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
3449 	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
3450 
3451 	for (int i = 0; i < count - 1; i++)
3452 		nvlist_free(newdev[i]);
3453 
3454 	if (count > 1)
3455 		kmem_free(newdev, (count - 1) * sizeof (void *));
3456 }
3457 
3458 /*
3459  * Remove a device from the pool.  Currently, this supports removing only hot
3460  * spares and level 2 ARC devices.
3461  */
3462 int
3463 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
3464 {
3465 	vdev_t *vd;
3466 	nvlist_t **spares, **l2cache, *nv;
3467 	uint_t nspares, nl2cache;
3468 	uint64_t txg = 0;
3469 	int error = 0;
3470 	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
3471 
3472 	if (!locked)
3473 		txg = spa_vdev_enter(spa);
3474 
3475 	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3476 
3477 	if (spa->spa_spares.sav_vdevs != NULL &&
3478 	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3479 	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
3480 	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
3481 		/*
3482 		 * Only remove the hot spare if it's not currently in use
3483 		 * in this pool.
3484 		 */
3485 		if (vd == NULL || unspare) {
3486 			spa_vdev_remove_aux(spa->spa_spares.sav_config,
3487 			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
3488 			spa_load_spares(spa);
3489 			spa->spa_spares.sav_sync = B_TRUE;
3490 		} else {
3491 			error = EBUSY;
3492 		}
3493 	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
3494 	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3495 	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
3496 	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
3497 		/*
3498 		 * Cache devices can always be removed.
3499 		 */
3500 		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
3501 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
3502 		spa_load_l2cache(spa);
3503 		spa->spa_l2cache.sav_sync = B_TRUE;
3504 	} else if (vd != NULL) {
3505 		/*
3506 		 * Normal vdevs cannot be removed (yet).
3507 		 */
3508 		error = ENOTSUP;
3509 	} else {
3510 		/*
3511 		 * There is no vdev of any kind with the specified guid.
3512 		 */
3513 		error = ENOENT;
3514 	}
3515 
3516 	if (!locked)
3517 		return (spa_vdev_exit(spa, NULL, txg, error));
3518 
3519 	return (error);
3520 }
3521 
3522 /*
3523  * Find any device that's done replacing, or a vdev marked 'unspare' that's
3524  * current spared, so we can detach it.
3525  */
3526 static vdev_t *
3527 spa_vdev_resilver_done_hunt(vdev_t *vd)
3528 {
3529 	vdev_t *newvd, *oldvd;
3530 	int c;
3531 
3532 	for (c = 0; c < vd->vdev_children; c++) {
3533 		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
3534 		if (oldvd != NULL)
3535 			return (oldvd);
3536 	}
3537 
3538 	/*
3539 	 * Check for a completed replacement.
3540 	 */
3541 	if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
3542 		oldvd = vd->vdev_child[0];
3543 		newvd = vd->vdev_child[1];
3544 
3545 		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
3546 		    !vdev_dtl_required(oldvd))
3547 			return (oldvd);
3548 	}
3549 
3550 	/*
3551 	 * Check for a completed resilver with the 'unspare' flag set.
3552 	 */
3553 	if (vd->vdev_ops == &vdev_spare_ops && vd->vdev_children == 2) {
3554 		newvd = vd->vdev_child[0];
3555 		oldvd = vd->vdev_child[1];
3556 
3557 		if (newvd->vdev_unspare &&
3558 		    vdev_dtl_empty(newvd, DTL_MISSING) &&
3559 		    !vdev_dtl_required(oldvd)) {
3560 			newvd->vdev_unspare = 0;
3561 			return (oldvd);
3562 		}
3563 	}
3564 
3565 	return (NULL);
3566 }
3567 
3568 static void
3569 spa_vdev_resilver_done(spa_t *spa)
3570 {
3571 	vdev_t *vd, *pvd, *ppvd;
3572 	uint64_t guid, sguid, pguid, ppguid;
3573 
3574 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3575 
3576 	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
3577 		pvd = vd->vdev_parent;
3578 		ppvd = pvd->vdev_parent;
3579 		guid = vd->vdev_guid;
3580 		pguid = pvd->vdev_guid;
3581 		ppguid = ppvd->vdev_guid;
3582 		sguid = 0;
3583 		/*
3584 		 * If we have just finished replacing a hot spared device, then
3585 		 * we need to detach the parent's first child (the original hot
3586 		 * spare) as well.
3587 		 */
3588 		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0) {
3589 			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
3590 			ASSERT(ppvd->vdev_children == 2);
3591 			sguid = ppvd->vdev_child[1]->vdev_guid;
3592 		}
3593 		spa_config_exit(spa, SCL_ALL, FTAG);
3594 		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
3595 			return;
3596 		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
3597 			return;
3598 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3599 	}
3600 
3601 	spa_config_exit(spa, SCL_ALL, FTAG);
3602 }
3603 
3604 /*
3605  * Update the stored path or FRU for this vdev.  Dirty the vdev configuration,
3606  * relying on spa_vdev_enter/exit() to synchronize the labels and cache.
3607  */
3608 int
3609 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
3610     boolean_t ispath)
3611 {
3612 	vdev_t *vd;
3613 	uint64_t txg;
3614 
3615 	txg = spa_vdev_enter(spa);
3616 
3617 	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
3618 		return (spa_vdev_exit(spa, NULL, txg, ENOENT));
3619 
3620 	if (!vd->vdev_ops->vdev_op_leaf)
3621 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3622 
3623 	if (ispath) {
3624 		spa_strfree(vd->vdev_path);
3625 		vd->vdev_path = spa_strdup(value);
3626 	} else {
3627 		if (vd->vdev_fru != NULL)
3628 			spa_strfree(vd->vdev_fru);
3629 		vd->vdev_fru = spa_strdup(value);
3630 	}
3631 
3632 	vdev_config_dirty(vd->vdev_top);
3633 
3634 	return (spa_vdev_exit(spa, NULL, txg, 0));
3635 }
3636 
3637 int
3638 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
3639 {
3640 	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
3641 }
3642 
3643 int
3644 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
3645 {
3646 	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
3647 }
3648 
3649 /*
3650  * ==========================================================================
3651  * SPA Scrubbing
3652  * ==========================================================================
3653  */
3654 
3655 int
3656 spa_scrub(spa_t *spa, pool_scrub_type_t type)
3657 {
3658 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
3659 
3660 	if ((uint_t)type >= POOL_SCRUB_TYPES)
3661 		return (ENOTSUP);
3662 
3663 	/*
3664 	 * If a resilver was requested, but there is no DTL on a
3665 	 * writeable leaf device, we have nothing to do.
3666 	 */
3667 	if (type == POOL_SCRUB_RESILVER &&
3668 	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
3669 		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
3670 		return (0);
3671 	}
3672 
3673 	if (type == POOL_SCRUB_EVERYTHING &&
3674 	    spa->spa_dsl_pool->dp_scrub_func != SCRUB_FUNC_NONE &&
3675 	    spa->spa_dsl_pool->dp_scrub_isresilver)
3676 		return (EBUSY);
3677 
3678 	if (type == POOL_SCRUB_EVERYTHING || type == POOL_SCRUB_RESILVER) {
3679 		return (dsl_pool_scrub_clean(spa->spa_dsl_pool));
3680 	} else if (type == POOL_SCRUB_NONE) {
3681 		return (dsl_pool_scrub_cancel(spa->spa_dsl_pool));
3682 	} else {
3683 		return (EINVAL);
3684 	}
3685 }
3686 
3687 /*
3688  * ==========================================================================
3689  * SPA async task processing
3690  * ==========================================================================
3691  */
3692 
3693 static void
3694 spa_async_remove(spa_t *spa, vdev_t *vd)
3695 {
3696 	if (vd->vdev_remove_wanted) {
3697 		vd->vdev_remove_wanted = 0;
3698 		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
3699 		vdev_clear(spa, vd);
3700 		vdev_state_dirty(vd->vdev_top);
3701 	}
3702 
3703 	for (int c = 0; c < vd->vdev_children; c++)
3704 		spa_async_remove(spa, vd->vdev_child[c]);
3705 }
3706 
3707 static void
3708 spa_async_probe(spa_t *spa, vdev_t *vd)
3709 {
3710 	if (vd->vdev_probe_wanted) {
3711 		vd->vdev_probe_wanted = 0;
3712 		vdev_reopen(vd);	/* vdev_open() does the actual probe */
3713 	}
3714 
3715 	for (int c = 0; c < vd->vdev_children; c++)
3716 		spa_async_probe(spa, vd->vdev_child[c]);
3717 }
3718 
3719 static void
3720 spa_async_thread(spa_t *spa)
3721 {
3722 	int tasks;
3723 
3724 	ASSERT(spa->spa_sync_on);
3725 
3726 	mutex_enter(&spa->spa_async_lock);
3727 	tasks = spa->spa_async_tasks;
3728 	spa->spa_async_tasks = 0;
3729 	mutex_exit(&spa->spa_async_lock);
3730 
3731 	/*
3732 	 * See if the config needs to be updated.
3733 	 */
3734 	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
3735 		mutex_enter(&spa_namespace_lock);
3736 		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3737 		mutex_exit(&spa_namespace_lock);
3738 	}
3739 
3740 	/*
3741 	 * See if any devices need to be marked REMOVED.
3742 	 */
3743 	if (tasks & SPA_ASYNC_REMOVE) {
3744 		spa_vdev_state_enter(spa);
3745 		spa_async_remove(spa, spa->spa_root_vdev);
3746 		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
3747 			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
3748 		for (int i = 0; i < spa->spa_spares.sav_count; i++)
3749 			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
3750 		(void) spa_vdev_state_exit(spa, NULL, 0);
3751 	}
3752 
3753 	/*
3754 	 * See if any devices need to be probed.
3755 	 */
3756 	if (tasks & SPA_ASYNC_PROBE) {
3757 		spa_vdev_state_enter(spa);
3758 		spa_async_probe(spa, spa->spa_root_vdev);
3759 		(void) spa_vdev_state_exit(spa, NULL, 0);
3760 	}
3761 
3762 	/*
3763 	 * If any devices are done replacing, detach them.
3764 	 */
3765 	if (tasks & SPA_ASYNC_RESILVER_DONE)
3766 		spa_vdev_resilver_done(spa);
3767 
3768 	/*
3769 	 * Kick off a resilver.
3770 	 */
3771 	if (tasks & SPA_ASYNC_RESILVER)
3772 		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER) == 0);
3773 
3774 	/*
3775 	 * Let the world know that we're done.
3776 	 */
3777 	mutex_enter(&spa->spa_async_lock);
3778 	spa->spa_async_thread = NULL;
3779 	cv_broadcast(&spa->spa_async_cv);
3780 	mutex_exit(&spa->spa_async_lock);
3781 	thread_exit();
3782 }
3783 
3784 void
3785 spa_async_suspend(spa_t *spa)
3786 {
3787 	mutex_enter(&spa->spa_async_lock);
3788 	spa->spa_async_suspended++;
3789 	while (spa->spa_async_thread != NULL)
3790 		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
3791 	mutex_exit(&spa->spa_async_lock);
3792 }
3793 
3794 void
3795 spa_async_resume(spa_t *spa)
3796 {
3797 	mutex_enter(&spa->spa_async_lock);
3798 	ASSERT(spa->spa_async_suspended != 0);
3799 	spa->spa_async_suspended--;
3800 	mutex_exit(&spa->spa_async_lock);
3801 }
3802 
3803 static void
3804 spa_async_dispatch(spa_t *spa)
3805 {
3806 	mutex_enter(&spa->spa_async_lock);
3807 	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
3808 	    spa->spa_async_thread == NULL &&
3809 	    rootdir != NULL && !vn_is_readonly(rootdir))
3810 		spa->spa_async_thread = thread_create(NULL, 0,
3811 		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
3812 	mutex_exit(&spa->spa_async_lock);
3813 }
3814 
3815 void
3816 spa_async_request(spa_t *spa, int task)
3817 {
3818 	mutex_enter(&spa->spa_async_lock);
3819 	spa->spa_async_tasks |= task;
3820 	mutex_exit(&spa->spa_async_lock);
3821 }
3822 
3823 /*
3824  * ==========================================================================
3825  * SPA syncing routines
3826  * ==========================================================================
3827  */
3828 
3829 static void
3830 spa_sync_deferred_frees(spa_t *spa, uint64_t txg)
3831 {
3832 	bplist_t *bpl = &spa->spa_sync_bplist;
3833 	dmu_tx_t *tx;
3834 	blkptr_t blk;
3835 	uint64_t itor = 0;
3836 	zio_t *zio;
3837 	int error;
3838 	uint8_t c = 1;
3839 
3840 	zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL);
3841 
3842 	while (bplist_iterate(bpl, &itor, &blk) == 0) {
3843 		ASSERT(blk.blk_birth < txg);
3844 		zio_nowait(zio_free(zio, spa, txg, &blk, NULL, NULL,
3845 		    ZIO_FLAG_MUSTSUCCEED));
3846 	}
3847 
3848 	error = zio_wait(zio);
3849 	ASSERT3U(error, ==, 0);
3850 
3851 	tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
3852 	bplist_vacate(bpl, tx);
3853 
3854 	/*
3855 	 * Pre-dirty the first block so we sync to convergence faster.
3856 	 * (Usually only the first block is needed.)
3857 	 */
3858 	dmu_write(spa->spa_meta_objset, spa->spa_sync_bplist_obj, 0, 1, &c, tx);
3859 	dmu_tx_commit(tx);
3860 }
3861 
3862 static void
3863 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
3864 {
3865 	char *packed = NULL;
3866 	size_t bufsize;
3867 	size_t nvsize = 0;
3868 	dmu_buf_t *db;
3869 
3870 	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
3871 
3872 	/*
3873 	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
3874 	 * information.  This avoids the dbuf_will_dirty() path and
3875 	 * saves us a pre-read to get data we don't actually care about.
3876 	 */
3877 	bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
3878 	packed = kmem_alloc(bufsize, KM_SLEEP);
3879 
3880 	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
3881 	    KM_SLEEP) == 0);
3882 	bzero(packed + nvsize, bufsize - nvsize);
3883 
3884 	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
3885 
3886 	kmem_free(packed, bufsize);
3887 
3888 	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
3889 	dmu_buf_will_dirty(db, tx);
3890 	*(uint64_t *)db->db_data = nvsize;
3891 	dmu_buf_rele(db, FTAG);
3892 }
3893 
3894 static void
3895 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
3896     const char *config, const char *entry)
3897 {
3898 	nvlist_t *nvroot;
3899 	nvlist_t **list;
3900 	int i;
3901 
3902 	if (!sav->sav_sync)
3903 		return;
3904 
3905 	/*
3906 	 * Update the MOS nvlist describing the list of available devices.
3907 	 * spa_validate_aux() will have already made sure this nvlist is
3908 	 * valid and the vdevs are labeled appropriately.
3909 	 */
3910 	if (sav->sav_object == 0) {
3911 		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
3912 		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
3913 		    sizeof (uint64_t), tx);
3914 		VERIFY(zap_update(spa->spa_meta_objset,
3915 		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
3916 		    &sav->sav_object, tx) == 0);
3917 	}
3918 
3919 	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3920 	if (sav->sav_count == 0) {
3921 		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
3922 	} else {
3923 		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
3924 		for (i = 0; i < sav->sav_count; i++)
3925 			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
3926 			    B_FALSE, B_FALSE, B_TRUE);
3927 		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
3928 		    sav->sav_count) == 0);
3929 		for (i = 0; i < sav->sav_count; i++)
3930 			nvlist_free(list[i]);
3931 		kmem_free(list, sav->sav_count * sizeof (void *));
3932 	}
3933 
3934 	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
3935 	nvlist_free(nvroot);
3936 
3937 	sav->sav_sync = B_FALSE;
3938 }
3939 
3940 static void
3941 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
3942 {
3943 	nvlist_t *config;
3944 
3945 	if (list_is_empty(&spa->spa_config_dirty_list))
3946 		return;
3947 
3948 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
3949 
3950 	config = spa_config_generate(spa, spa->spa_root_vdev,
3951 	    dmu_tx_get_txg(tx), B_FALSE);
3952 
3953 	spa_config_exit(spa, SCL_STATE, FTAG);
3954 
3955 	if (spa->spa_config_syncing)
3956 		nvlist_free(spa->spa_config_syncing);
3957 	spa->spa_config_syncing = config;
3958 
3959 	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
3960 }
3961 
3962 /*
3963  * Set zpool properties.
3964  */
3965 static void
3966 spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
3967 {
3968 	spa_t *spa = arg1;
3969 	objset_t *mos = spa->spa_meta_objset;
3970 	nvlist_t *nvp = arg2;
3971 	nvpair_t *elem;
3972 	uint64_t intval;
3973 	char *strval;
3974 	zpool_prop_t prop;
3975 	const char *propname;
3976 	zprop_type_t proptype;
3977 
3978 	mutex_enter(&spa->spa_props_lock);
3979 
3980 	elem = NULL;
3981 	while ((elem = nvlist_next_nvpair(nvp, elem))) {
3982 		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
3983 		case ZPOOL_PROP_VERSION:
3984 			/*
3985 			 * Only set version for non-zpool-creation cases
3986 			 * (set/import). spa_create() needs special care
3987 			 * for version setting.
3988 			 */
3989 			if (tx->tx_txg != TXG_INITIAL) {
3990 				VERIFY(nvpair_value_uint64(elem,
3991 				    &intval) == 0);
3992 				ASSERT(intval <= SPA_VERSION);
3993 				ASSERT(intval >= spa_version(spa));
3994 				spa->spa_uberblock.ub_version = intval;
3995 				vdev_config_dirty(spa->spa_root_vdev);
3996 			}
3997 			break;
3998 
3999 		case ZPOOL_PROP_ALTROOT:
4000 			/*
4001 			 * 'altroot' is a non-persistent property. It should
4002 			 * have been set temporarily at creation or import time.
4003 			 */
4004 			ASSERT(spa->spa_root != NULL);
4005 			break;
4006 
4007 		case ZPOOL_PROP_CACHEFILE:
4008 			/*
4009 			 * 'cachefile' is also a non-persisitent property.
4010 			 */
4011 			break;
4012 		default:
4013 			/*
4014 			 * Set pool property values in the poolprops mos object.
4015 			 */
4016 			if (spa->spa_pool_props_object == 0) {
4017 				objset_t *mos = spa->spa_meta_objset;
4018 
4019 				VERIFY((spa->spa_pool_props_object =
4020 				    zap_create(mos, DMU_OT_POOL_PROPS,
4021 				    DMU_OT_NONE, 0, tx)) > 0);
4022 
4023 				VERIFY(zap_update(mos,
4024 				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
4025 				    8, 1, &spa->spa_pool_props_object, tx)
4026 				    == 0);
4027 			}
4028 
4029 			/* normalize the property name */
4030 			propname = zpool_prop_to_name(prop);
4031 			proptype = zpool_prop_get_type(prop);
4032 
4033 			if (nvpair_type(elem) == DATA_TYPE_STRING) {
4034 				ASSERT(proptype == PROP_TYPE_STRING);
4035 				VERIFY(nvpair_value_string(elem, &strval) == 0);
4036 				VERIFY(zap_update(mos,
4037 				    spa->spa_pool_props_object, propname,
4038 				    1, strlen(strval) + 1, strval, tx) == 0);
4039 
4040 			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
4041 				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
4042 
4043 				if (proptype == PROP_TYPE_INDEX) {
4044 					const char *unused;
4045 					VERIFY(zpool_prop_index_to_string(
4046 					    prop, intval, &unused) == 0);
4047 				}
4048 				VERIFY(zap_update(mos,
4049 				    spa->spa_pool_props_object, propname,
4050 				    8, 1, &intval, tx) == 0);
4051 			} else {
4052 				ASSERT(0); /* not allowed */
4053 			}
4054 
4055 			switch (prop) {
4056 			case ZPOOL_PROP_DELEGATION:
4057 				spa->spa_delegation = intval;
4058 				break;
4059 			case ZPOOL_PROP_BOOTFS:
4060 				spa->spa_bootfs = intval;
4061 				break;
4062 			case ZPOOL_PROP_FAILUREMODE:
4063 				spa->spa_failmode = intval;
4064 				break;
4065 			default:
4066 				break;
4067 			}
4068 		}
4069 
4070 		/* log internal history if this is not a zpool create */
4071 		if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
4072 		    tx->tx_txg != TXG_INITIAL) {
4073 			spa_history_internal_log(LOG_POOL_PROPSET,
4074 			    spa, tx, cr, "%s %lld %s",
4075 			    nvpair_name(elem), intval, spa_name(spa));
4076 		}
4077 	}
4078 
4079 	mutex_exit(&spa->spa_props_lock);
4080 }
4081 
4082 /*
4083  * Sync the specified transaction group.  New blocks may be dirtied as
4084  * part of the process, so we iterate until it converges.
4085  */
4086 void
4087 spa_sync(spa_t *spa, uint64_t txg)
4088 {
4089 	dsl_pool_t *dp = spa->spa_dsl_pool;
4090 	objset_t *mos = spa->spa_meta_objset;
4091 	bplist_t *bpl = &spa->spa_sync_bplist;
4092 	vdev_t *rvd = spa->spa_root_vdev;
4093 	vdev_t *vd;
4094 	dmu_tx_t *tx;
4095 	int dirty_vdevs;
4096 	int error;
4097 
4098 	/*
4099 	 * Lock out configuration changes.
4100 	 */
4101 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4102 
4103 	spa->spa_syncing_txg = txg;
4104 	spa->spa_sync_pass = 0;
4105 
4106 	/*
4107 	 * If there are any pending vdev state changes, convert them
4108 	 * into config changes that go out with this transaction group.
4109 	 */
4110 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4111 	while (list_head(&spa->spa_state_dirty_list) != NULL) {
4112 		/*
4113 		 * We need the write lock here because, for aux vdevs,
4114 		 * calling vdev_config_dirty() modifies sav_config.
4115 		 * This is ugly and will become unnecessary when we
4116 		 * eliminate the aux vdev wart by integrating all vdevs
4117 		 * into the root vdev tree.
4118 		 */
4119 		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
4120 		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
4121 		while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
4122 			vdev_state_clean(vd);
4123 			vdev_config_dirty(vd);
4124 		}
4125 		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
4126 		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
4127 	}
4128 	spa_config_exit(spa, SCL_STATE, FTAG);
4129 
4130 	VERIFY(0 == bplist_open(bpl, mos, spa->spa_sync_bplist_obj));
4131 
4132 	tx = dmu_tx_create_assigned(dp, txg);
4133 
4134 	/*
4135 	 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
4136 	 * set spa_deflate if we have no raid-z vdevs.
4137 	 */
4138 	if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
4139 	    spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
4140 		int i;
4141 
4142 		for (i = 0; i < rvd->vdev_children; i++) {
4143 			vd = rvd->vdev_child[i];
4144 			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
4145 				break;
4146 		}
4147 		if (i == rvd->vdev_children) {
4148 			spa->spa_deflate = TRUE;
4149 			VERIFY(0 == zap_add(spa->spa_meta_objset,
4150 			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
4151 			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
4152 		}
4153 	}
4154 
4155 	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
4156 	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
4157 		dsl_pool_create_origin(dp, tx);
4158 
4159 		/* Keeping the origin open increases spa_minref */
4160 		spa->spa_minref += 3;
4161 	}
4162 
4163 	if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
4164 	    spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
4165 		dsl_pool_upgrade_clones(dp, tx);
4166 	}
4167 
4168 	/*
4169 	 * If anything has changed in this txg, push the deferred frees
4170 	 * from the previous txg.  If not, leave them alone so that we
4171 	 * don't generate work on an otherwise idle system.
4172 	 */
4173 	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
4174 	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
4175 	    !txg_list_empty(&dp->dp_sync_tasks, txg))
4176 		spa_sync_deferred_frees(spa, txg);
4177 
4178 	/*
4179 	 * Iterate to convergence.
4180 	 */
4181 	do {
4182 		spa->spa_sync_pass++;
4183 
4184 		spa_sync_config_object(spa, tx);
4185 		spa_sync_aux_dev(spa, &spa->spa_spares, tx,
4186 		    ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
4187 		spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
4188 		    ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
4189 		spa_errlog_sync(spa, txg);
4190 		dsl_pool_sync(dp, txg);
4191 
4192 		dirty_vdevs = 0;
4193 		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) {
4194 			vdev_sync(vd, txg);
4195 			dirty_vdevs++;
4196 		}
4197 
4198 		bplist_sync(bpl, tx);
4199 	} while (dirty_vdevs);
4200 
4201 	bplist_close(bpl);
4202 
4203 	dprintf("txg %llu passes %d\n", txg, spa->spa_sync_pass);
4204 
4205 	/*
4206 	 * Rewrite the vdev configuration (which includes the uberblock)
4207 	 * to commit the transaction group.
4208 	 *
4209 	 * If there are no dirty vdevs, we sync the uberblock to a few
4210 	 * random top-level vdevs that are known to be visible in the
4211 	 * config cache (see spa_vdev_add() for a complete description).
4212 	 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
4213 	 */
4214 	for (;;) {
4215 		/*
4216 		 * We hold SCL_STATE to prevent vdev open/close/etc.
4217 		 * while we're attempting to write the vdev labels.
4218 		 */
4219 		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4220 
4221 		if (list_is_empty(&spa->spa_config_dirty_list)) {
4222 			vdev_t *svd[SPA_DVAS_PER_BP];
4223 			int svdcount = 0;
4224 			int children = rvd->vdev_children;
4225 			int c0 = spa_get_random(children);
4226 			int c;
4227 
4228 			for (c = 0; c < children; c++) {
4229 				vd = rvd->vdev_child[(c0 + c) % children];
4230 				if (vd->vdev_ms_array == 0 || vd->vdev_islog)
4231 					continue;
4232 				svd[svdcount++] = vd;
4233 				if (svdcount == SPA_DVAS_PER_BP)
4234 					break;
4235 			}
4236 			error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
4237 			if (error != 0)
4238 				error = vdev_config_sync(svd, svdcount, txg,
4239 				    B_TRUE);
4240 		} else {
4241 			error = vdev_config_sync(rvd->vdev_child,
4242 			    rvd->vdev_children, txg, B_FALSE);
4243 			if (error != 0)
4244 				error = vdev_config_sync(rvd->vdev_child,
4245 				    rvd->vdev_children, txg, B_TRUE);
4246 		}
4247 
4248 		spa_config_exit(spa, SCL_STATE, FTAG);
4249 
4250 		if (error == 0)
4251 			break;
4252 		zio_suspend(spa, NULL);
4253 		zio_resume_wait(spa);
4254 	}
4255 	dmu_tx_commit(tx);
4256 
4257 	/*
4258 	 * Clear the dirty config list.
4259 	 */
4260 	while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
4261 		vdev_config_clean(vd);
4262 
4263 	/*
4264 	 * Now that the new config has synced transactionally,
4265 	 * let it become visible to the config cache.
4266 	 */
4267 	if (spa->spa_config_syncing != NULL) {
4268 		spa_config_set(spa, spa->spa_config_syncing);
4269 		spa->spa_config_txg = txg;
4270 		spa->spa_config_syncing = NULL;
4271 	}
4272 
4273 	spa->spa_ubsync = spa->spa_uberblock;
4274 
4275 	/*
4276 	 * Clean up the ZIL records for the synced txg.
4277 	 */
4278 	dsl_pool_zil_clean(dp);
4279 
4280 	/*
4281 	 * Update usable space statistics.
4282 	 */
4283 	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
4284 		vdev_sync_done(vd, txg);
4285 
4286 	/*
4287 	 * It had better be the case that we didn't dirty anything
4288 	 * since vdev_config_sync().
4289 	 */
4290 	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
4291 	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
4292 	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
4293 	ASSERT(bpl->bpl_queue == NULL);
4294 
4295 	spa_config_exit(spa, SCL_CONFIG, FTAG);
4296 
4297 	/*
4298 	 * If any async tasks have been requested, kick them off.
4299 	 */
4300 	spa_async_dispatch(spa);
4301 }
4302 
4303 /*
4304  * Sync all pools.  We don't want to hold the namespace lock across these
4305  * operations, so we take a reference on the spa_t and drop the lock during the
4306  * sync.
4307  */
4308 void
4309 spa_sync_allpools(void)
4310 {
4311 	spa_t *spa = NULL;
4312 	mutex_enter(&spa_namespace_lock);
4313 	while ((spa = spa_next(spa)) != NULL) {
4314 		if (spa_state(spa) != POOL_STATE_ACTIVE || spa_suspended(spa))
4315 			continue;
4316 		spa_open_ref(spa, FTAG);
4317 		mutex_exit(&spa_namespace_lock);
4318 		txg_wait_synced(spa_get_dsl(spa), 0);
4319 		mutex_enter(&spa_namespace_lock);
4320 		spa_close(spa, FTAG);
4321 	}
4322 	mutex_exit(&spa_namespace_lock);
4323 }
4324 
4325 /*
4326  * ==========================================================================
4327  * Miscellaneous routines
4328  * ==========================================================================
4329  */
4330 
4331 /*
4332  * Remove all pools in the system.
4333  */
4334 void
4335 spa_evict_all(void)
4336 {
4337 	spa_t *spa;
4338 
4339 	/*
4340 	 * Remove all cached state.  All pools should be closed now,
4341 	 * so every spa in the AVL tree should be unreferenced.
4342 	 */
4343 	mutex_enter(&spa_namespace_lock);
4344 	while ((spa = spa_next(NULL)) != NULL) {
4345 		/*
4346 		 * Stop async tasks.  The async thread may need to detach
4347 		 * a device that's been replaced, which requires grabbing
4348 		 * spa_namespace_lock, so we must drop it here.
4349 		 */
4350 		spa_open_ref(spa, FTAG);
4351 		mutex_exit(&spa_namespace_lock);
4352 		spa_async_suspend(spa);
4353 		mutex_enter(&spa_namespace_lock);
4354 		spa_close(spa, FTAG);
4355 
4356 		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4357 			spa_unload(spa);
4358 			spa_deactivate(spa);
4359 		}
4360 		spa_remove(spa);
4361 	}
4362 	mutex_exit(&spa_namespace_lock);
4363 }
4364 
4365 vdev_t *
4366 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
4367 {
4368 	vdev_t *vd;
4369 	int i;
4370 
4371 	if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
4372 		return (vd);
4373 
4374 	if (aux) {
4375 		for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
4376 			vd = spa->spa_l2cache.sav_vdevs[i];
4377 			if (vd->vdev_guid == guid)
4378 				return (vd);
4379 		}
4380 
4381 		for (i = 0; i < spa->spa_spares.sav_count; i++) {
4382 			vd = spa->spa_spares.sav_vdevs[i];
4383 			if (vd->vdev_guid == guid)
4384 				return (vd);
4385 		}
4386 	}
4387 
4388 	return (NULL);
4389 }
4390 
4391 void
4392 spa_upgrade(spa_t *spa, uint64_t version)
4393 {
4394 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4395 
4396 	/*
4397 	 * This should only be called for a non-faulted pool, and since a
4398 	 * future version would result in an unopenable pool, this shouldn't be
4399 	 * possible.
4400 	 */
4401 	ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
4402 	ASSERT(version >= spa->spa_uberblock.ub_version);
4403 
4404 	spa->spa_uberblock.ub_version = version;
4405 	vdev_config_dirty(spa->spa_root_vdev);
4406 
4407 	spa_config_exit(spa, SCL_ALL, FTAG);
4408 
4409 	txg_wait_synced(spa_get_dsl(spa), 0);
4410 }
4411 
4412 boolean_t
4413 spa_has_spare(spa_t *spa, uint64_t guid)
4414 {
4415 	int i;
4416 	uint64_t spareguid;
4417 	spa_aux_vdev_t *sav = &spa->spa_spares;
4418 
4419 	for (i = 0; i < sav->sav_count; i++)
4420 		if (sav->sav_vdevs[i]->vdev_guid == guid)
4421 			return (B_TRUE);
4422 
4423 	for (i = 0; i < sav->sav_npending; i++) {
4424 		if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
4425 		    &spareguid) == 0 && spareguid == guid)
4426 			return (B_TRUE);
4427 	}
4428 
4429 	return (B_FALSE);
4430 }
4431 
4432 /*
4433  * Check if a pool has an active shared spare device.
4434  * Note: reference count of an active spare is 2, as a spare and as a replace
4435  */
4436 static boolean_t
4437 spa_has_active_shared_spare(spa_t *spa)
4438 {
4439 	int i, refcnt;
4440 	uint64_t pool;
4441 	spa_aux_vdev_t *sav = &spa->spa_spares;
4442 
4443 	for (i = 0; i < sav->sav_count; i++) {
4444 		if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
4445 		    &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
4446 		    refcnt > 2)
4447 			return (B_TRUE);
4448 	}
4449 
4450 	return (B_FALSE);
4451 }
4452 
4453 /*
4454  * Post a sysevent corresponding to the given event.  The 'name' must be one of
4455  * the event definitions in sys/sysevent/eventdefs.h.  The payload will be
4456  * filled in from the spa and (optionally) the vdev.  This doesn't do anything
4457  * in the userland libzpool, as we don't want consumers to misinterpret ztest
4458  * or zdb as real changes.
4459  */
4460 void
4461 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
4462 {
4463 #ifdef _KERNEL
4464 	sysevent_t		*ev;
4465 	sysevent_attr_list_t	*attr = NULL;
4466 	sysevent_value_t	value;
4467 	sysevent_id_t		eid;
4468 
4469 	ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
4470 	    SE_SLEEP);
4471 
4472 	value.value_type = SE_DATA_TYPE_STRING;
4473 	value.value.sv_string = spa_name(spa);
4474 	if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
4475 		goto done;
4476 
4477 	value.value_type = SE_DATA_TYPE_UINT64;
4478 	value.value.sv_uint64 = spa_guid(spa);
4479 	if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
4480 		goto done;
4481 
4482 	if (vd) {
4483 		value.value_type = SE_DATA_TYPE_UINT64;
4484 		value.value.sv_uint64 = vd->vdev_guid;
4485 		if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
4486 		    SE_SLEEP) != 0)
4487 			goto done;
4488 
4489 		if (vd->vdev_path) {
4490 			value.value_type = SE_DATA_TYPE_STRING;
4491 			value.value.sv_string = vd->vdev_path;
4492 			if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
4493 			    &value, SE_SLEEP) != 0)
4494 				goto done;
4495 		}
4496 	}
4497 
4498 	if (sysevent_attach_attributes(ev, attr) != 0)
4499 		goto done;
4500 	attr = NULL;
4501 
4502 	(void) log_sysevent(ev, SE_SLEEP, &eid);
4503 
4504 done:
4505 	if (attr)
4506 		sysevent_free_attr(attr);
4507 	sysevent_free(ev);
4508 #endif
4509 }
4510