xref: /titanic_41/usr/src/uts/common/fs/zfs/spa.c (revision 3fb517f786391b507780c78aabb8d98bfea9efe9)
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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24  */
25 
26 /*
27  * This file contains all the routines used when modifying on-disk SPA state.
28  * This includes opening, importing, destroying, exporting a pool, and syncing a
29  * pool.
30  */
31 
32 #include <sys/zfs_context.h>
33 #include <sys/fm/fs/zfs.h>
34 #include <sys/spa_impl.h>
35 #include <sys/zio.h>
36 #include <sys/zio_checksum.h>
37 #include <sys/dmu.h>
38 #include <sys/dmu_tx.h>
39 #include <sys/zap.h>
40 #include <sys/zil.h>
41 #include <sys/ddt.h>
42 #include <sys/vdev_impl.h>
43 #include <sys/metaslab.h>
44 #include <sys/metaslab_impl.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/spa_boot.h>
61 #include <sys/zfs_ioctl.h>
62 #include <sys/dsl_scan.h>
63 
64 #ifdef	_KERNEL
65 #include <sys/bootprops.h>
66 #include <sys/callb.h>
67 #include <sys/cpupart.h>
68 #include <sys/pool.h>
69 #include <sys/sysdc.h>
70 #include <sys/zone.h>
71 #endif	/* _KERNEL */
72 
73 #include "zfs_prop.h"
74 #include "zfs_comutil.h"
75 
76 typedef enum zti_modes {
77 	zti_mode_fixed,			/* value is # of threads (min 1) */
78 	zti_mode_online_percent,	/* value is % of online CPUs */
79 	zti_mode_batch,			/* cpu-intensive; value is ignored */
80 	zti_mode_null,			/* don't create a taskq */
81 	zti_nmodes
82 } zti_modes_t;
83 
84 #define	ZTI_FIX(n)	{ zti_mode_fixed, (n) }
85 #define	ZTI_PCT(n)	{ zti_mode_online_percent, (n) }
86 #define	ZTI_BATCH	{ zti_mode_batch, 0 }
87 #define	ZTI_NULL	{ zti_mode_null, 0 }
88 
89 #define	ZTI_ONE		ZTI_FIX(1)
90 
91 typedef struct zio_taskq_info {
92 	enum zti_modes zti_mode;
93 	uint_t zti_value;
94 } zio_taskq_info_t;
95 
96 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
97 	"issue", "issue_high", "intr", "intr_high"
98 };
99 
100 /*
101  * Define the taskq threads for the following I/O types:
102  * 	NULL, READ, WRITE, FREE, CLAIM, and IOCTL
103  */
104 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
105 	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
106 	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
107 	{ ZTI_FIX(8),	ZTI_NULL,	ZTI_BATCH,	ZTI_NULL },
108 	{ ZTI_BATCH,	ZTI_FIX(5),	ZTI_FIX(8),	ZTI_FIX(5) },
109 	{ ZTI_FIX(10),	ZTI_NULL,	ZTI_FIX(10),	ZTI_NULL },
110 	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
111 	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
112 };
113 
114 static dsl_syncfunc_t spa_sync_props;
115 static boolean_t spa_has_active_shared_spare(spa_t *spa);
116 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
117     spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
118     char **ereport);
119 
120 uint_t		zio_taskq_batch_pct = 100;	/* 1 thread per cpu in pset */
121 id_t		zio_taskq_psrset_bind = PS_NONE;
122 boolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
123 uint_t		zio_taskq_basedc = 80;		/* base duty cycle */
124 
125 boolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
126 
127 /*
128  * This (illegal) pool name is used when temporarily importing a spa_t in order
129  * to get the vdev stats associated with the imported devices.
130  */
131 #define	TRYIMPORT_NAME	"$import"
132 
133 /*
134  * ==========================================================================
135  * SPA properties routines
136  * ==========================================================================
137  */
138 
139 /*
140  * Add a (source=src, propname=propval) list to an nvlist.
141  */
142 static void
143 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
144     uint64_t intval, zprop_source_t src)
145 {
146 	const char *propname = zpool_prop_to_name(prop);
147 	nvlist_t *propval;
148 
149 	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
150 	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
151 
152 	if (strval != NULL)
153 		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
154 	else
155 		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
156 
157 	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
158 	nvlist_free(propval);
159 }
160 
161 /*
162  * Get property values from the spa configuration.
163  */
164 static void
165 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
166 {
167 	uint64_t size;
168 	uint64_t alloc;
169 	uint64_t cap, version;
170 	zprop_source_t src = ZPROP_SRC_NONE;
171 	spa_config_dirent_t *dp;
172 
173 	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
174 
175 	if (spa->spa_root_vdev != NULL) {
176 		alloc = metaslab_class_get_alloc(spa_normal_class(spa));
177 		size = metaslab_class_get_space(spa_normal_class(spa));
178 		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
179 		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
180 		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
181 		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
182 		    size - alloc, src);
183 
184 		cap = (size == 0) ? 0 : (alloc * 100 / size);
185 		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
186 
187 		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
188 		    ddt_get_pool_dedup_ratio(spa), src);
189 
190 		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
191 		    spa->spa_root_vdev->vdev_state, src);
192 
193 		version = spa_version(spa);
194 		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
195 			src = ZPROP_SRC_DEFAULT;
196 		else
197 			src = ZPROP_SRC_LOCAL;
198 		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
199 	}
200 
201 	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
202 
203 	if (spa->spa_root != NULL)
204 		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
205 		    0, ZPROP_SRC_LOCAL);
206 
207 	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
208 		if (dp->scd_path == NULL) {
209 			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
210 			    "none", 0, ZPROP_SRC_LOCAL);
211 		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
212 			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
213 			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
214 		}
215 	}
216 }
217 
218 /*
219  * Get zpool property values.
220  */
221 int
222 spa_prop_get(spa_t *spa, nvlist_t **nvp)
223 {
224 	objset_t *mos = spa->spa_meta_objset;
225 	zap_cursor_t zc;
226 	zap_attribute_t za;
227 	int err;
228 
229 	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
230 
231 	mutex_enter(&spa->spa_props_lock);
232 
233 	/*
234 	 * Get properties from the spa config.
235 	 */
236 	spa_prop_get_config(spa, nvp);
237 
238 	/* If no pool property object, no more prop to get. */
239 	if (mos == NULL || spa->spa_pool_props_object == 0) {
240 		mutex_exit(&spa->spa_props_lock);
241 		return (0);
242 	}
243 
244 	/*
245 	 * Get properties from the MOS pool property object.
246 	 */
247 	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
248 	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
249 	    zap_cursor_advance(&zc)) {
250 		uint64_t intval = 0;
251 		char *strval = NULL;
252 		zprop_source_t src = ZPROP_SRC_DEFAULT;
253 		zpool_prop_t prop;
254 
255 		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
256 			continue;
257 
258 		switch (za.za_integer_length) {
259 		case 8:
260 			/* integer property */
261 			if (za.za_first_integer !=
262 			    zpool_prop_default_numeric(prop))
263 				src = ZPROP_SRC_LOCAL;
264 
265 			if (prop == ZPOOL_PROP_BOOTFS) {
266 				dsl_pool_t *dp;
267 				dsl_dataset_t *ds = NULL;
268 
269 				dp = spa_get_dsl(spa);
270 				rw_enter(&dp->dp_config_rwlock, RW_READER);
271 				if (err = dsl_dataset_hold_obj(dp,
272 				    za.za_first_integer, FTAG, &ds)) {
273 					rw_exit(&dp->dp_config_rwlock);
274 					break;
275 				}
276 
277 				strval = kmem_alloc(
278 				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
279 				    KM_SLEEP);
280 				dsl_dataset_name(ds, strval);
281 				dsl_dataset_rele(ds, FTAG);
282 				rw_exit(&dp->dp_config_rwlock);
283 			} else {
284 				strval = NULL;
285 				intval = za.za_first_integer;
286 			}
287 
288 			spa_prop_add_list(*nvp, prop, strval, intval, src);
289 
290 			if (strval != NULL)
291 				kmem_free(strval,
292 				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
293 
294 			break;
295 
296 		case 1:
297 			/* string property */
298 			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
299 			err = zap_lookup(mos, spa->spa_pool_props_object,
300 			    za.za_name, 1, za.za_num_integers, strval);
301 			if (err) {
302 				kmem_free(strval, za.za_num_integers);
303 				break;
304 			}
305 			spa_prop_add_list(*nvp, prop, strval, 0, src);
306 			kmem_free(strval, za.za_num_integers);
307 			break;
308 
309 		default:
310 			break;
311 		}
312 	}
313 	zap_cursor_fini(&zc);
314 	mutex_exit(&spa->spa_props_lock);
315 out:
316 	if (err && err != ENOENT) {
317 		nvlist_free(*nvp);
318 		*nvp = NULL;
319 		return (err);
320 	}
321 
322 	return (0);
323 }
324 
325 /*
326  * Validate the given pool properties nvlist and modify the list
327  * for the property values to be set.
328  */
329 static int
330 spa_prop_validate(spa_t *spa, nvlist_t *props)
331 {
332 	nvpair_t *elem;
333 	int error = 0, reset_bootfs = 0;
334 	uint64_t objnum;
335 
336 	elem = NULL;
337 	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
338 		zpool_prop_t prop;
339 		char *propname, *strval;
340 		uint64_t intval;
341 		objset_t *os;
342 		char *slash;
343 
344 		propname = nvpair_name(elem);
345 
346 		if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
347 			return (EINVAL);
348 
349 		switch (prop) {
350 		case ZPOOL_PROP_VERSION:
351 			error = nvpair_value_uint64(elem, &intval);
352 			if (!error &&
353 			    (intval < spa_version(spa) || intval > SPA_VERSION))
354 				error = EINVAL;
355 			break;
356 
357 		case ZPOOL_PROP_DELEGATION:
358 		case ZPOOL_PROP_AUTOREPLACE:
359 		case ZPOOL_PROP_LISTSNAPS:
360 		case ZPOOL_PROP_AUTOEXPAND:
361 			error = nvpair_value_uint64(elem, &intval);
362 			if (!error && intval > 1)
363 				error = EINVAL;
364 			break;
365 
366 		case ZPOOL_PROP_BOOTFS:
367 			/*
368 			 * If the pool version is less than SPA_VERSION_BOOTFS,
369 			 * or the pool is still being created (version == 0),
370 			 * the bootfs property cannot be set.
371 			 */
372 			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
373 				error = ENOTSUP;
374 				break;
375 			}
376 
377 			/*
378 			 * Make sure the vdev config is bootable
379 			 */
380 			if (!vdev_is_bootable(spa->spa_root_vdev)) {
381 				error = ENOTSUP;
382 				break;
383 			}
384 
385 			reset_bootfs = 1;
386 
387 			error = nvpair_value_string(elem, &strval);
388 
389 			if (!error) {
390 				uint64_t compress;
391 
392 				if (strval == NULL || strval[0] == '\0') {
393 					objnum = zpool_prop_default_numeric(
394 					    ZPOOL_PROP_BOOTFS);
395 					break;
396 				}
397 
398 				if (error = dmu_objset_hold(strval, FTAG, &os))
399 					break;
400 
401 				/* Must be ZPL and not gzip compressed. */
402 
403 				if (dmu_objset_type(os) != DMU_OST_ZFS) {
404 					error = ENOTSUP;
405 				} else if ((error = dsl_prop_get_integer(strval,
406 				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
407 				    &compress, NULL)) == 0 &&
408 				    !BOOTFS_COMPRESS_VALID(compress)) {
409 					error = ENOTSUP;
410 				} else {
411 					objnum = dmu_objset_id(os);
412 				}
413 				dmu_objset_rele(os, FTAG);
414 			}
415 			break;
416 
417 		case ZPOOL_PROP_FAILUREMODE:
418 			error = nvpair_value_uint64(elem, &intval);
419 			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
420 			    intval > ZIO_FAILURE_MODE_PANIC))
421 				error = EINVAL;
422 
423 			/*
424 			 * This is a special case which only occurs when
425 			 * the pool has completely failed. This allows
426 			 * the user to change the in-core failmode property
427 			 * without syncing it out to disk (I/Os might
428 			 * currently be blocked). We do this by returning
429 			 * EIO to the caller (spa_prop_set) to trick it
430 			 * into thinking we encountered a property validation
431 			 * error.
432 			 */
433 			if (!error && spa_suspended(spa)) {
434 				spa->spa_failmode = intval;
435 				error = EIO;
436 			}
437 			break;
438 
439 		case ZPOOL_PROP_CACHEFILE:
440 			if ((error = nvpair_value_string(elem, &strval)) != 0)
441 				break;
442 
443 			if (strval[0] == '\0')
444 				break;
445 
446 			if (strcmp(strval, "none") == 0)
447 				break;
448 
449 			if (strval[0] != '/') {
450 				error = EINVAL;
451 				break;
452 			}
453 
454 			slash = strrchr(strval, '/');
455 			ASSERT(slash != NULL);
456 
457 			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
458 			    strcmp(slash, "/..") == 0)
459 				error = EINVAL;
460 			break;
461 
462 		case ZPOOL_PROP_DEDUPDITTO:
463 			if (spa_version(spa) < SPA_VERSION_DEDUP)
464 				error = ENOTSUP;
465 			else
466 				error = nvpair_value_uint64(elem, &intval);
467 			if (error == 0 &&
468 			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
469 				error = EINVAL;
470 			break;
471 		}
472 
473 		if (error)
474 			break;
475 	}
476 
477 	if (!error && reset_bootfs) {
478 		error = nvlist_remove(props,
479 		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
480 
481 		if (!error) {
482 			error = nvlist_add_uint64(props,
483 			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
484 		}
485 	}
486 
487 	return (error);
488 }
489 
490 void
491 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
492 {
493 	char *cachefile;
494 	spa_config_dirent_t *dp;
495 
496 	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
497 	    &cachefile) != 0)
498 		return;
499 
500 	dp = kmem_alloc(sizeof (spa_config_dirent_t),
501 	    KM_SLEEP);
502 
503 	if (cachefile[0] == '\0')
504 		dp->scd_path = spa_strdup(spa_config_path);
505 	else if (strcmp(cachefile, "none") == 0)
506 		dp->scd_path = NULL;
507 	else
508 		dp->scd_path = spa_strdup(cachefile);
509 
510 	list_insert_head(&spa->spa_config_list, dp);
511 	if (need_sync)
512 		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
513 }
514 
515 int
516 spa_prop_set(spa_t *spa, nvlist_t *nvp)
517 {
518 	int error;
519 	nvpair_t *elem;
520 	boolean_t need_sync = B_FALSE;
521 	zpool_prop_t prop;
522 
523 	if ((error = spa_prop_validate(spa, nvp)) != 0)
524 		return (error);
525 
526 	elem = NULL;
527 	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
528 		if ((prop = zpool_name_to_prop(
529 		    nvpair_name(elem))) == ZPROP_INVAL)
530 			return (EINVAL);
531 
532 		if (prop == ZPOOL_PROP_CACHEFILE || prop == ZPOOL_PROP_ALTROOT)
533 			continue;
534 
535 		need_sync = B_TRUE;
536 		break;
537 	}
538 
539 	if (need_sync)
540 		return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
541 		    spa, nvp, 3));
542 	else
543 		return (0);
544 }
545 
546 /*
547  * If the bootfs property value is dsobj, clear it.
548  */
549 void
550 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
551 {
552 	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
553 		VERIFY(zap_remove(spa->spa_meta_objset,
554 		    spa->spa_pool_props_object,
555 		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
556 		spa->spa_bootfs = 0;
557 	}
558 }
559 
560 /*
561  * ==========================================================================
562  * SPA state manipulation (open/create/destroy/import/export)
563  * ==========================================================================
564  */
565 
566 static int
567 spa_error_entry_compare(const void *a, const void *b)
568 {
569 	spa_error_entry_t *sa = (spa_error_entry_t *)a;
570 	spa_error_entry_t *sb = (spa_error_entry_t *)b;
571 	int ret;
572 
573 	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
574 	    sizeof (zbookmark_t));
575 
576 	if (ret < 0)
577 		return (-1);
578 	else if (ret > 0)
579 		return (1);
580 	else
581 		return (0);
582 }
583 
584 /*
585  * Utility function which retrieves copies of the current logs and
586  * re-initializes them in the process.
587  */
588 void
589 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
590 {
591 	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
592 
593 	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
594 	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
595 
596 	avl_create(&spa->spa_errlist_scrub,
597 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
598 	    offsetof(spa_error_entry_t, se_avl));
599 	avl_create(&spa->spa_errlist_last,
600 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
601 	    offsetof(spa_error_entry_t, se_avl));
602 }
603 
604 static taskq_t *
605 spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
606     uint_t value)
607 {
608 	uint_t flags = TASKQ_PREPOPULATE;
609 	boolean_t batch = B_FALSE;
610 
611 	switch (mode) {
612 	case zti_mode_null:
613 		return (NULL);		/* no taskq needed */
614 
615 	case zti_mode_fixed:
616 		ASSERT3U(value, >=, 1);
617 		value = MAX(value, 1);
618 		break;
619 
620 	case zti_mode_batch:
621 		batch = B_TRUE;
622 		flags |= TASKQ_THREADS_CPU_PCT;
623 		value = zio_taskq_batch_pct;
624 		break;
625 
626 	case zti_mode_online_percent:
627 		flags |= TASKQ_THREADS_CPU_PCT;
628 		break;
629 
630 	default:
631 		panic("unrecognized mode for %s taskq (%u:%u) in "
632 		    "spa_activate()",
633 		    name, mode, value);
634 		break;
635 	}
636 
637 	if (zio_taskq_sysdc && spa->spa_proc != &p0) {
638 		if (batch)
639 			flags |= TASKQ_DC_BATCH;
640 
641 		return (taskq_create_sysdc(name, value, 50, INT_MAX,
642 		    spa->spa_proc, zio_taskq_basedc, flags));
643 	}
644 	return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
645 	    spa->spa_proc, flags));
646 }
647 
648 static void
649 spa_create_zio_taskqs(spa_t *spa)
650 {
651 	for (int t = 0; t < ZIO_TYPES; t++) {
652 		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
653 			const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
654 			enum zti_modes mode = ztip->zti_mode;
655 			uint_t value = ztip->zti_value;
656 			char name[32];
657 
658 			(void) snprintf(name, sizeof (name),
659 			    "%s_%s", zio_type_name[t], zio_taskq_types[q]);
660 
661 			spa->spa_zio_taskq[t][q] =
662 			    spa_taskq_create(spa, name, mode, value);
663 		}
664 	}
665 }
666 
667 #ifdef _KERNEL
668 static void
669 spa_thread(void *arg)
670 {
671 	callb_cpr_t cprinfo;
672 
673 	spa_t *spa = arg;
674 	user_t *pu = PTOU(curproc);
675 
676 	CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
677 	    spa->spa_name);
678 
679 	ASSERT(curproc != &p0);
680 	(void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
681 	    "zpool-%s", spa->spa_name);
682 	(void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
683 
684 	/* bind this thread to the requested psrset */
685 	if (zio_taskq_psrset_bind != PS_NONE) {
686 		pool_lock();
687 		mutex_enter(&cpu_lock);
688 		mutex_enter(&pidlock);
689 		mutex_enter(&curproc->p_lock);
690 
691 		if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
692 		    0, NULL, NULL) == 0)  {
693 			curthread->t_bind_pset = zio_taskq_psrset_bind;
694 		} else {
695 			cmn_err(CE_WARN,
696 			    "Couldn't bind process for zfs pool \"%s\" to "
697 			    "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
698 		}
699 
700 		mutex_exit(&curproc->p_lock);
701 		mutex_exit(&pidlock);
702 		mutex_exit(&cpu_lock);
703 		pool_unlock();
704 	}
705 
706 	if (zio_taskq_sysdc) {
707 		sysdc_thread_enter(curthread, 100, 0);
708 	}
709 
710 	spa->spa_proc = curproc;
711 	spa->spa_did = curthread->t_did;
712 
713 	spa_create_zio_taskqs(spa);
714 
715 	mutex_enter(&spa->spa_proc_lock);
716 	ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
717 
718 	spa->spa_proc_state = SPA_PROC_ACTIVE;
719 	cv_broadcast(&spa->spa_proc_cv);
720 
721 	CALLB_CPR_SAFE_BEGIN(&cprinfo);
722 	while (spa->spa_proc_state == SPA_PROC_ACTIVE)
723 		cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
724 	CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
725 
726 	ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
727 	spa->spa_proc_state = SPA_PROC_GONE;
728 	spa->spa_proc = &p0;
729 	cv_broadcast(&spa->spa_proc_cv);
730 	CALLB_CPR_EXIT(&cprinfo);	/* drops spa_proc_lock */
731 
732 	mutex_enter(&curproc->p_lock);
733 	lwp_exit();
734 }
735 #endif
736 
737 /*
738  * Activate an uninitialized pool.
739  */
740 static void
741 spa_activate(spa_t *spa, int mode)
742 {
743 	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
744 
745 	spa->spa_state = POOL_STATE_ACTIVE;
746 	spa->spa_mode = mode;
747 
748 	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
749 	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
750 
751 	/* Try to create a covering process */
752 	mutex_enter(&spa->spa_proc_lock);
753 	ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
754 	ASSERT(spa->spa_proc == &p0);
755 	spa->spa_did = 0;
756 
757 	/* Only create a process if we're going to be around a while. */
758 	if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
759 		if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
760 		    NULL, 0) == 0) {
761 			spa->spa_proc_state = SPA_PROC_CREATED;
762 			while (spa->spa_proc_state == SPA_PROC_CREATED) {
763 				cv_wait(&spa->spa_proc_cv,
764 				    &spa->spa_proc_lock);
765 			}
766 			ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
767 			ASSERT(spa->spa_proc != &p0);
768 			ASSERT(spa->spa_did != 0);
769 		} else {
770 #ifdef _KERNEL
771 			cmn_err(CE_WARN,
772 			    "Couldn't create process for zfs pool \"%s\"\n",
773 			    spa->spa_name);
774 #endif
775 		}
776 	}
777 	mutex_exit(&spa->spa_proc_lock);
778 
779 	/* If we didn't create a process, we need to create our taskqs. */
780 	if (spa->spa_proc == &p0) {
781 		spa_create_zio_taskqs(spa);
782 	}
783 
784 	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
785 	    offsetof(vdev_t, vdev_config_dirty_node));
786 	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
787 	    offsetof(vdev_t, vdev_state_dirty_node));
788 
789 	txg_list_create(&spa->spa_vdev_txg_list,
790 	    offsetof(struct vdev, vdev_txg_node));
791 
792 	avl_create(&spa->spa_errlist_scrub,
793 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
794 	    offsetof(spa_error_entry_t, se_avl));
795 	avl_create(&spa->spa_errlist_last,
796 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
797 	    offsetof(spa_error_entry_t, se_avl));
798 }
799 
800 /*
801  * Opposite of spa_activate().
802  */
803 static void
804 spa_deactivate(spa_t *spa)
805 {
806 	ASSERT(spa->spa_sync_on == B_FALSE);
807 	ASSERT(spa->spa_dsl_pool == NULL);
808 	ASSERT(spa->spa_root_vdev == NULL);
809 	ASSERT(spa->spa_async_zio_root == NULL);
810 	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
811 
812 	txg_list_destroy(&spa->spa_vdev_txg_list);
813 
814 	list_destroy(&spa->spa_config_dirty_list);
815 	list_destroy(&spa->spa_state_dirty_list);
816 
817 	for (int t = 0; t < ZIO_TYPES; t++) {
818 		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
819 			if (spa->spa_zio_taskq[t][q] != NULL)
820 				taskq_destroy(spa->spa_zio_taskq[t][q]);
821 			spa->spa_zio_taskq[t][q] = NULL;
822 		}
823 	}
824 
825 	metaslab_class_destroy(spa->spa_normal_class);
826 	spa->spa_normal_class = NULL;
827 
828 	metaslab_class_destroy(spa->spa_log_class);
829 	spa->spa_log_class = NULL;
830 
831 	/*
832 	 * If this was part of an import or the open otherwise failed, we may
833 	 * still have errors left in the queues.  Empty them just in case.
834 	 */
835 	spa_errlog_drain(spa);
836 
837 	avl_destroy(&spa->spa_errlist_scrub);
838 	avl_destroy(&spa->spa_errlist_last);
839 
840 	spa->spa_state = POOL_STATE_UNINITIALIZED;
841 
842 	mutex_enter(&spa->spa_proc_lock);
843 	if (spa->spa_proc_state != SPA_PROC_NONE) {
844 		ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
845 		spa->spa_proc_state = SPA_PROC_DEACTIVATE;
846 		cv_broadcast(&spa->spa_proc_cv);
847 		while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
848 			ASSERT(spa->spa_proc != &p0);
849 			cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
850 		}
851 		ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
852 		spa->spa_proc_state = SPA_PROC_NONE;
853 	}
854 	ASSERT(spa->spa_proc == &p0);
855 	mutex_exit(&spa->spa_proc_lock);
856 
857 	/*
858 	 * We want to make sure spa_thread() has actually exited the ZFS
859 	 * module, so that the module can't be unloaded out from underneath
860 	 * it.
861 	 */
862 	if (spa->spa_did != 0) {
863 		thread_join(spa->spa_did);
864 		spa->spa_did = 0;
865 	}
866 }
867 
868 /*
869  * Verify a pool configuration, and construct the vdev tree appropriately.  This
870  * will create all the necessary vdevs in the appropriate layout, with each vdev
871  * in the CLOSED state.  This will prep the pool before open/creation/import.
872  * All vdev validation is done by the vdev_alloc() routine.
873  */
874 static int
875 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
876     uint_t id, int atype)
877 {
878 	nvlist_t **child;
879 	uint_t children;
880 	int error;
881 
882 	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
883 		return (error);
884 
885 	if ((*vdp)->vdev_ops->vdev_op_leaf)
886 		return (0);
887 
888 	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
889 	    &child, &children);
890 
891 	if (error == ENOENT)
892 		return (0);
893 
894 	if (error) {
895 		vdev_free(*vdp);
896 		*vdp = NULL;
897 		return (EINVAL);
898 	}
899 
900 	for (int c = 0; c < children; c++) {
901 		vdev_t *vd;
902 		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
903 		    atype)) != 0) {
904 			vdev_free(*vdp);
905 			*vdp = NULL;
906 			return (error);
907 		}
908 	}
909 
910 	ASSERT(*vdp != NULL);
911 
912 	return (0);
913 }
914 
915 /*
916  * Opposite of spa_load().
917  */
918 static void
919 spa_unload(spa_t *spa)
920 {
921 	int i;
922 
923 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
924 
925 	/*
926 	 * Stop async tasks.
927 	 */
928 	spa_async_suspend(spa);
929 
930 	/*
931 	 * Stop syncing.
932 	 */
933 	if (spa->spa_sync_on) {
934 		txg_sync_stop(spa->spa_dsl_pool);
935 		spa->spa_sync_on = B_FALSE;
936 	}
937 
938 	/*
939 	 * Wait for any outstanding async I/O to complete.
940 	 */
941 	if (spa->spa_async_zio_root != NULL) {
942 		(void) zio_wait(spa->spa_async_zio_root);
943 		spa->spa_async_zio_root = NULL;
944 	}
945 
946 	/*
947 	 * Close the dsl pool.
948 	 */
949 	if (spa->spa_dsl_pool) {
950 		dsl_pool_close(spa->spa_dsl_pool);
951 		spa->spa_dsl_pool = NULL;
952 		spa->spa_meta_objset = NULL;
953 	}
954 
955 	ddt_unload(spa);
956 
957 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
958 
959 	/*
960 	 * Drop and purge level 2 cache
961 	 */
962 	spa_l2cache_drop(spa);
963 
964 	/*
965 	 * Close all vdevs.
966 	 */
967 	if (spa->spa_root_vdev)
968 		vdev_free(spa->spa_root_vdev);
969 	ASSERT(spa->spa_root_vdev == NULL);
970 
971 	for (i = 0; i < spa->spa_spares.sav_count; i++)
972 		vdev_free(spa->spa_spares.sav_vdevs[i]);
973 	if (spa->spa_spares.sav_vdevs) {
974 		kmem_free(spa->spa_spares.sav_vdevs,
975 		    spa->spa_spares.sav_count * sizeof (void *));
976 		spa->spa_spares.sav_vdevs = NULL;
977 	}
978 	if (spa->spa_spares.sav_config) {
979 		nvlist_free(spa->spa_spares.sav_config);
980 		spa->spa_spares.sav_config = NULL;
981 	}
982 	spa->spa_spares.sav_count = 0;
983 
984 	for (i = 0; i < spa->spa_l2cache.sav_count; i++)
985 		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
986 	if (spa->spa_l2cache.sav_vdevs) {
987 		kmem_free(spa->spa_l2cache.sav_vdevs,
988 		    spa->spa_l2cache.sav_count * sizeof (void *));
989 		spa->spa_l2cache.sav_vdevs = NULL;
990 	}
991 	if (spa->spa_l2cache.sav_config) {
992 		nvlist_free(spa->spa_l2cache.sav_config);
993 		spa->spa_l2cache.sav_config = NULL;
994 	}
995 	spa->spa_l2cache.sav_count = 0;
996 
997 	spa->spa_async_suspended = 0;
998 
999 	spa_config_exit(spa, SCL_ALL, FTAG);
1000 }
1001 
1002 /*
1003  * Load (or re-load) the current list of vdevs describing the active spares for
1004  * this pool.  When this is called, we have some form of basic information in
1005  * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1006  * then re-generate a more complete list including status information.
1007  */
1008 static void
1009 spa_load_spares(spa_t *spa)
1010 {
1011 	nvlist_t **spares;
1012 	uint_t nspares;
1013 	int i;
1014 	vdev_t *vd, *tvd;
1015 
1016 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1017 
1018 	/*
1019 	 * First, close and free any existing spare vdevs.
1020 	 */
1021 	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1022 		vd = spa->spa_spares.sav_vdevs[i];
1023 
1024 		/* Undo the call to spa_activate() below */
1025 		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1026 		    B_FALSE)) != NULL && tvd->vdev_isspare)
1027 			spa_spare_remove(tvd);
1028 		vdev_close(vd);
1029 		vdev_free(vd);
1030 	}
1031 
1032 	if (spa->spa_spares.sav_vdevs)
1033 		kmem_free(spa->spa_spares.sav_vdevs,
1034 		    spa->spa_spares.sav_count * sizeof (void *));
1035 
1036 	if (spa->spa_spares.sav_config == NULL)
1037 		nspares = 0;
1038 	else
1039 		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1040 		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1041 
1042 	spa->spa_spares.sav_count = (int)nspares;
1043 	spa->spa_spares.sav_vdevs = NULL;
1044 
1045 	if (nspares == 0)
1046 		return;
1047 
1048 	/*
1049 	 * Construct the array of vdevs, opening them to get status in the
1050 	 * process.   For each spare, there is potentially two different vdev_t
1051 	 * structures associated with it: one in the list of spares (used only
1052 	 * for basic validation purposes) and one in the active vdev
1053 	 * configuration (if it's spared in).  During this phase we open and
1054 	 * validate each vdev on the spare list.  If the vdev also exists in the
1055 	 * active configuration, then we also mark this vdev as an active spare.
1056 	 */
1057 	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1058 	    KM_SLEEP);
1059 	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1060 		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1061 		    VDEV_ALLOC_SPARE) == 0);
1062 		ASSERT(vd != NULL);
1063 
1064 		spa->spa_spares.sav_vdevs[i] = vd;
1065 
1066 		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1067 		    B_FALSE)) != NULL) {
1068 			if (!tvd->vdev_isspare)
1069 				spa_spare_add(tvd);
1070 
1071 			/*
1072 			 * We only mark the spare active if we were successfully
1073 			 * able to load the vdev.  Otherwise, importing a pool
1074 			 * with a bad active spare would result in strange
1075 			 * behavior, because multiple pool would think the spare
1076 			 * is actively in use.
1077 			 *
1078 			 * There is a vulnerability here to an equally bizarre
1079 			 * circumstance, where a dead active spare is later
1080 			 * brought back to life (onlined or otherwise).  Given
1081 			 * the rarity of this scenario, and the extra complexity
1082 			 * it adds, we ignore the possibility.
1083 			 */
1084 			if (!vdev_is_dead(tvd))
1085 				spa_spare_activate(tvd);
1086 		}
1087 
1088 		vd->vdev_top = vd;
1089 		vd->vdev_aux = &spa->spa_spares;
1090 
1091 		if (vdev_open(vd) != 0)
1092 			continue;
1093 
1094 		if (vdev_validate_aux(vd) == 0)
1095 			spa_spare_add(vd);
1096 	}
1097 
1098 	/*
1099 	 * Recompute the stashed list of spares, with status information
1100 	 * this time.
1101 	 */
1102 	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1103 	    DATA_TYPE_NVLIST_ARRAY) == 0);
1104 
1105 	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1106 	    KM_SLEEP);
1107 	for (i = 0; i < spa->spa_spares.sav_count; i++)
1108 		spares[i] = vdev_config_generate(spa,
1109 		    spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1110 	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1111 	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1112 	for (i = 0; i < spa->spa_spares.sav_count; i++)
1113 		nvlist_free(spares[i]);
1114 	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1115 }
1116 
1117 /*
1118  * Load (or re-load) the current list of vdevs describing the active l2cache for
1119  * this pool.  When this is called, we have some form of basic information in
1120  * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1121  * then re-generate a more complete list including status information.
1122  * Devices which are already active have their details maintained, and are
1123  * not re-opened.
1124  */
1125 static void
1126 spa_load_l2cache(spa_t *spa)
1127 {
1128 	nvlist_t **l2cache;
1129 	uint_t nl2cache;
1130 	int i, j, oldnvdevs;
1131 	uint64_t guid;
1132 	vdev_t *vd, **oldvdevs, **newvdevs;
1133 	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1134 
1135 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1136 
1137 	if (sav->sav_config != NULL) {
1138 		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1139 		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1140 		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1141 	} else {
1142 		nl2cache = 0;
1143 	}
1144 
1145 	oldvdevs = sav->sav_vdevs;
1146 	oldnvdevs = sav->sav_count;
1147 	sav->sav_vdevs = NULL;
1148 	sav->sav_count = 0;
1149 
1150 	/*
1151 	 * Process new nvlist of vdevs.
1152 	 */
1153 	for (i = 0; i < nl2cache; i++) {
1154 		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1155 		    &guid) == 0);
1156 
1157 		newvdevs[i] = NULL;
1158 		for (j = 0; j < oldnvdevs; j++) {
1159 			vd = oldvdevs[j];
1160 			if (vd != NULL && guid == vd->vdev_guid) {
1161 				/*
1162 				 * Retain previous vdev for add/remove ops.
1163 				 */
1164 				newvdevs[i] = vd;
1165 				oldvdevs[j] = NULL;
1166 				break;
1167 			}
1168 		}
1169 
1170 		if (newvdevs[i] == NULL) {
1171 			/*
1172 			 * Create new vdev
1173 			 */
1174 			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1175 			    VDEV_ALLOC_L2CACHE) == 0);
1176 			ASSERT(vd != NULL);
1177 			newvdevs[i] = vd;
1178 
1179 			/*
1180 			 * Commit this vdev as an l2cache device,
1181 			 * even if it fails to open.
1182 			 */
1183 			spa_l2cache_add(vd);
1184 
1185 			vd->vdev_top = vd;
1186 			vd->vdev_aux = sav;
1187 
1188 			spa_l2cache_activate(vd);
1189 
1190 			if (vdev_open(vd) != 0)
1191 				continue;
1192 
1193 			(void) vdev_validate_aux(vd);
1194 
1195 			if (!vdev_is_dead(vd))
1196 				l2arc_add_vdev(spa, vd);
1197 		}
1198 	}
1199 
1200 	/*
1201 	 * Purge vdevs that were dropped
1202 	 */
1203 	for (i = 0; i < oldnvdevs; i++) {
1204 		uint64_t pool;
1205 
1206 		vd = oldvdevs[i];
1207 		if (vd != NULL) {
1208 			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1209 			    pool != 0ULL && l2arc_vdev_present(vd))
1210 				l2arc_remove_vdev(vd);
1211 			(void) vdev_close(vd);
1212 			spa_l2cache_remove(vd);
1213 		}
1214 	}
1215 
1216 	if (oldvdevs)
1217 		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1218 
1219 	if (sav->sav_config == NULL)
1220 		goto out;
1221 
1222 	sav->sav_vdevs = newvdevs;
1223 	sav->sav_count = (int)nl2cache;
1224 
1225 	/*
1226 	 * Recompute the stashed list of l2cache devices, with status
1227 	 * information this time.
1228 	 */
1229 	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1230 	    DATA_TYPE_NVLIST_ARRAY) == 0);
1231 
1232 	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1233 	for (i = 0; i < sav->sav_count; i++)
1234 		l2cache[i] = vdev_config_generate(spa,
1235 		    sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1236 	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1237 	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1238 out:
1239 	for (i = 0; i < sav->sav_count; i++)
1240 		nvlist_free(l2cache[i]);
1241 	if (sav->sav_count)
1242 		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1243 }
1244 
1245 static int
1246 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1247 {
1248 	dmu_buf_t *db;
1249 	char *packed = NULL;
1250 	size_t nvsize = 0;
1251 	int error;
1252 	*value = NULL;
1253 
1254 	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1255 	nvsize = *(uint64_t *)db->db_data;
1256 	dmu_buf_rele(db, FTAG);
1257 
1258 	packed = kmem_alloc(nvsize, KM_SLEEP);
1259 	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1260 	    DMU_READ_PREFETCH);
1261 	if (error == 0)
1262 		error = nvlist_unpack(packed, nvsize, value, 0);
1263 	kmem_free(packed, nvsize);
1264 
1265 	return (error);
1266 }
1267 
1268 /*
1269  * Checks to see if the given vdev could not be opened, in which case we post a
1270  * sysevent to notify the autoreplace code that the device has been removed.
1271  */
1272 static void
1273 spa_check_removed(vdev_t *vd)
1274 {
1275 	for (int c = 0; c < vd->vdev_children; c++)
1276 		spa_check_removed(vd->vdev_child[c]);
1277 
1278 	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1279 		zfs_post_autoreplace(vd->vdev_spa, vd);
1280 		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1281 	}
1282 }
1283 
1284 /*
1285  * Load the slog device state from the config object since it's possible
1286  * that the label does not contain the most up-to-date information.
1287  */
1288 void
1289 spa_load_log_state(spa_t *spa, nvlist_t *nv)
1290 {
1291 	vdev_t *ovd, *rvd = spa->spa_root_vdev;
1292 
1293 	/*
1294 	 * Load the original root vdev tree from the passed config.
1295 	 */
1296 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1297 	VERIFY(spa_config_parse(spa, &ovd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1298 
1299 	for (int c = 0; c < rvd->vdev_children; c++) {
1300 		vdev_t *cvd = rvd->vdev_child[c];
1301 		if (cvd->vdev_islog)
1302 			vdev_load_log_state(cvd, ovd->vdev_child[c]);
1303 	}
1304 	vdev_free(ovd);
1305 	spa_config_exit(spa, SCL_ALL, FTAG);
1306 }
1307 
1308 /*
1309  * Check for missing log devices
1310  */
1311 int
1312 spa_check_logs(spa_t *spa)
1313 {
1314 	switch (spa->spa_log_state) {
1315 	case SPA_LOG_MISSING:
1316 		/* need to recheck in case slog has been restored */
1317 	case SPA_LOG_UNKNOWN:
1318 		if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1319 		    DS_FIND_CHILDREN)) {
1320 			spa_set_log_state(spa, SPA_LOG_MISSING);
1321 			return (1);
1322 		}
1323 		break;
1324 	}
1325 	return (0);
1326 }
1327 
1328 static boolean_t
1329 spa_passivate_log(spa_t *spa)
1330 {
1331 	vdev_t *rvd = spa->spa_root_vdev;
1332 	boolean_t slog_found = B_FALSE;
1333 
1334 	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1335 
1336 	if (!spa_has_slogs(spa))
1337 		return (B_FALSE);
1338 
1339 	for (int c = 0; c < rvd->vdev_children; c++) {
1340 		vdev_t *tvd = rvd->vdev_child[c];
1341 		metaslab_group_t *mg = tvd->vdev_mg;
1342 
1343 		if (tvd->vdev_islog) {
1344 			metaslab_group_passivate(mg);
1345 			slog_found = B_TRUE;
1346 		}
1347 	}
1348 
1349 	return (slog_found);
1350 }
1351 
1352 static void
1353 spa_activate_log(spa_t *spa)
1354 {
1355 	vdev_t *rvd = spa->spa_root_vdev;
1356 
1357 	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1358 
1359 	for (int c = 0; c < rvd->vdev_children; c++) {
1360 		vdev_t *tvd = rvd->vdev_child[c];
1361 		metaslab_group_t *mg = tvd->vdev_mg;
1362 
1363 		if (tvd->vdev_islog)
1364 			metaslab_group_activate(mg);
1365 	}
1366 }
1367 
1368 int
1369 spa_offline_log(spa_t *spa)
1370 {
1371 	int error = 0;
1372 
1373 	if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1374 	    NULL, DS_FIND_CHILDREN)) == 0) {
1375 
1376 		/*
1377 		 * We successfully offlined the log device, sync out the
1378 		 * current txg so that the "stubby" block can be removed
1379 		 * by zil_sync().
1380 		 */
1381 		txg_wait_synced(spa->spa_dsl_pool, 0);
1382 	}
1383 	return (error);
1384 }
1385 
1386 static void
1387 spa_aux_check_removed(spa_aux_vdev_t *sav)
1388 {
1389 	for (int i = 0; i < sav->sav_count; i++)
1390 		spa_check_removed(sav->sav_vdevs[i]);
1391 }
1392 
1393 void
1394 spa_claim_notify(zio_t *zio)
1395 {
1396 	spa_t *spa = zio->io_spa;
1397 
1398 	if (zio->io_error)
1399 		return;
1400 
1401 	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1402 	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1403 		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1404 	mutex_exit(&spa->spa_props_lock);
1405 }
1406 
1407 typedef struct spa_load_error {
1408 	uint64_t	sle_meta_count;
1409 	uint64_t	sle_data_count;
1410 } spa_load_error_t;
1411 
1412 static void
1413 spa_load_verify_done(zio_t *zio)
1414 {
1415 	blkptr_t *bp = zio->io_bp;
1416 	spa_load_error_t *sle = zio->io_private;
1417 	dmu_object_type_t type = BP_GET_TYPE(bp);
1418 	int error = zio->io_error;
1419 
1420 	if (error) {
1421 		if ((BP_GET_LEVEL(bp) != 0 || dmu_ot[type].ot_metadata) &&
1422 		    type != DMU_OT_INTENT_LOG)
1423 			atomic_add_64(&sle->sle_meta_count, 1);
1424 		else
1425 			atomic_add_64(&sle->sle_data_count, 1);
1426 	}
1427 	zio_data_buf_free(zio->io_data, zio->io_size);
1428 }
1429 
1430 /*ARGSUSED*/
1431 static int
1432 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1433     arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1434 {
1435 	if (bp != NULL) {
1436 		zio_t *rio = arg;
1437 		size_t size = BP_GET_PSIZE(bp);
1438 		void *data = zio_data_buf_alloc(size);
1439 
1440 		zio_nowait(zio_read(rio, spa, bp, data, size,
1441 		    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1442 		    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1443 		    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1444 	}
1445 	return (0);
1446 }
1447 
1448 static int
1449 spa_load_verify(spa_t *spa)
1450 {
1451 	zio_t *rio;
1452 	spa_load_error_t sle = { 0 };
1453 	zpool_rewind_policy_t policy;
1454 	boolean_t verify_ok = B_FALSE;
1455 	int error;
1456 
1457 	zpool_get_rewind_policy(spa->spa_config, &policy);
1458 
1459 	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1460 		return (0);
1461 
1462 	rio = zio_root(spa, NULL, &sle,
1463 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1464 
1465 	error = traverse_pool(spa, spa->spa_verify_min_txg,
1466 	    TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1467 
1468 	(void) zio_wait(rio);
1469 
1470 	spa->spa_load_meta_errors = sle.sle_meta_count;
1471 	spa->spa_load_data_errors = sle.sle_data_count;
1472 
1473 	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1474 	    sle.sle_data_count <= policy.zrp_maxdata) {
1475 		verify_ok = B_TRUE;
1476 		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1477 		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1478 	} else {
1479 		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1480 	}
1481 
1482 	if (error) {
1483 		if (error != ENXIO && error != EIO)
1484 			error = EIO;
1485 		return (error);
1486 	}
1487 
1488 	return (verify_ok ? 0 : EIO);
1489 }
1490 
1491 /*
1492  * Find a value in the pool props object.
1493  */
1494 static void
1495 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1496 {
1497 	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1498 	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1499 }
1500 
1501 /*
1502  * Find a value in the pool directory object.
1503  */
1504 static int
1505 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1506 {
1507 	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1508 	    name, sizeof (uint64_t), 1, val));
1509 }
1510 
1511 static int
1512 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1513 {
1514 	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1515 	return (err);
1516 }
1517 
1518 /*
1519  * Fix up config after a partly-completed split.  This is done with the
1520  * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1521  * pool have that entry in their config, but only the splitting one contains
1522  * a list of all the guids of the vdevs that are being split off.
1523  *
1524  * This function determines what to do with that list: either rejoin
1525  * all the disks to the pool, or complete the splitting process.  To attempt
1526  * the rejoin, each disk that is offlined is marked online again, and
1527  * we do a reopen() call.  If the vdev label for every disk that was
1528  * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1529  * then we call vdev_split() on each disk, and complete the split.
1530  *
1531  * Otherwise we leave the config alone, with all the vdevs in place in
1532  * the original pool.
1533  */
1534 static void
1535 spa_try_repair(spa_t *spa, nvlist_t *config)
1536 {
1537 	uint_t extracted;
1538 	uint64_t *glist;
1539 	uint_t i, gcount;
1540 	nvlist_t *nvl;
1541 	vdev_t **vd;
1542 	boolean_t attempt_reopen;
1543 
1544 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1545 		return;
1546 
1547 	/* check that the config is complete */
1548 	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1549 	    &glist, &gcount) != 0)
1550 		return;
1551 
1552 	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1553 
1554 	/* attempt to online all the vdevs & validate */
1555 	attempt_reopen = B_TRUE;
1556 	for (i = 0; i < gcount; i++) {
1557 		if (glist[i] == 0)	/* vdev is hole */
1558 			continue;
1559 
1560 		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1561 		if (vd[i] == NULL) {
1562 			/*
1563 			 * Don't bother attempting to reopen the disks;
1564 			 * just do the split.
1565 			 */
1566 			attempt_reopen = B_FALSE;
1567 		} else {
1568 			/* attempt to re-online it */
1569 			vd[i]->vdev_offline = B_FALSE;
1570 		}
1571 	}
1572 
1573 	if (attempt_reopen) {
1574 		vdev_reopen(spa->spa_root_vdev);
1575 
1576 		/* check each device to see what state it's in */
1577 		for (extracted = 0, i = 0; i < gcount; i++) {
1578 			if (vd[i] != NULL &&
1579 			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1580 				break;
1581 			++extracted;
1582 		}
1583 	}
1584 
1585 	/*
1586 	 * If every disk has been moved to the new pool, or if we never
1587 	 * even attempted to look at them, then we split them off for
1588 	 * good.
1589 	 */
1590 	if (!attempt_reopen || gcount == extracted) {
1591 		for (i = 0; i < gcount; i++)
1592 			if (vd[i] != NULL)
1593 				vdev_split(vd[i]);
1594 		vdev_reopen(spa->spa_root_vdev);
1595 	}
1596 
1597 	kmem_free(vd, gcount * sizeof (vdev_t *));
1598 }
1599 
1600 static int
1601 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1602     boolean_t mosconfig)
1603 {
1604 	nvlist_t *config = spa->spa_config;
1605 	char *ereport = FM_EREPORT_ZFS_POOL;
1606 	int error;
1607 	uint64_t pool_guid;
1608 	nvlist_t *nvl;
1609 
1610 	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1611 		return (EINVAL);
1612 
1613 	/*
1614 	 * Versioning wasn't explicitly added to the label until later, so if
1615 	 * it's not present treat it as the initial version.
1616 	 */
1617 	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1618 	    &spa->spa_ubsync.ub_version) != 0)
1619 		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1620 
1621 	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1622 	    &spa->spa_config_txg);
1623 
1624 	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1625 	    spa_guid_exists(pool_guid, 0)) {
1626 		error = EEXIST;
1627 	} else {
1628 		spa->spa_load_guid = pool_guid;
1629 
1630 		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1631 		    &nvl) == 0) {
1632 			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1633 			    KM_SLEEP) == 0);
1634 		}
1635 
1636 		error = spa_load_impl(spa, pool_guid, config, state, type,
1637 		    mosconfig, &ereport);
1638 	}
1639 
1640 	spa->spa_minref = refcount_count(&spa->spa_refcount);
1641 	if (error && error != EBADF)
1642 		zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1643 	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1644 	spa->spa_ena = 0;
1645 
1646 	return (error);
1647 }
1648 
1649 /*
1650  * Load an existing storage pool, using the pool's builtin spa_config as a
1651  * source of configuration information.
1652  */
1653 static int
1654 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1655     spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1656     char **ereport)
1657 {
1658 	int error = 0;
1659 	nvlist_t *nvroot = NULL;
1660 	vdev_t *rvd;
1661 	uberblock_t *ub = &spa->spa_uberblock;
1662 	uint64_t config_cache_txg = spa->spa_config_txg;
1663 	int orig_mode = spa->spa_mode;
1664 	int parse;
1665 
1666 	/*
1667 	 * If this is an untrusted config, access the pool in read-only mode.
1668 	 * This prevents things like resilvering recently removed devices.
1669 	 */
1670 	if (!mosconfig)
1671 		spa->spa_mode = FREAD;
1672 
1673 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1674 
1675 	spa->spa_load_state = state;
1676 
1677 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1678 		return (EINVAL);
1679 
1680 	parse = (type == SPA_IMPORT_EXISTING ?
1681 	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1682 
1683 	/*
1684 	 * Create "The Godfather" zio to hold all async IOs
1685 	 */
1686 	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1687 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1688 
1689 	/*
1690 	 * Parse the configuration into a vdev tree.  We explicitly set the
1691 	 * value that will be returned by spa_version() since parsing the
1692 	 * configuration requires knowing the version number.
1693 	 */
1694 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1695 	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1696 	spa_config_exit(spa, SCL_ALL, FTAG);
1697 
1698 	if (error != 0)
1699 		return (error);
1700 
1701 	ASSERT(spa->spa_root_vdev == rvd);
1702 
1703 	if (type != SPA_IMPORT_ASSEMBLE) {
1704 		ASSERT(spa_guid(spa) == pool_guid);
1705 	}
1706 
1707 	/*
1708 	 * Try to open all vdevs, loading each label in the process.
1709 	 */
1710 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1711 	error = vdev_open(rvd);
1712 	spa_config_exit(spa, SCL_ALL, FTAG);
1713 	if (error != 0)
1714 		return (error);
1715 
1716 	/*
1717 	 * We need to validate the vdev labels against the configuration that
1718 	 * we have in hand, which is dependent on the setting of mosconfig. If
1719 	 * mosconfig is true then we're validating the vdev labels based on
1720 	 * that config.  Otherwise, we're validating against the cached config
1721 	 * (zpool.cache) that was read when we loaded the zfs module, and then
1722 	 * later we will recursively call spa_load() and validate against
1723 	 * the vdev config.
1724 	 *
1725 	 * If we're assembling a new pool that's been split off from an
1726 	 * existing pool, the labels haven't yet been updated so we skip
1727 	 * validation for now.
1728 	 */
1729 	if (type != SPA_IMPORT_ASSEMBLE) {
1730 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1731 		error = vdev_validate(rvd);
1732 		spa_config_exit(spa, SCL_ALL, FTAG);
1733 
1734 		if (error != 0)
1735 			return (error);
1736 
1737 		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1738 			return (ENXIO);
1739 	}
1740 
1741 	/*
1742 	 * Find the best uberblock.
1743 	 */
1744 	vdev_uberblock_load(NULL, rvd, ub);
1745 
1746 	/*
1747 	 * If we weren't able to find a single valid uberblock, return failure.
1748 	 */
1749 	if (ub->ub_txg == 0)
1750 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
1751 
1752 	/*
1753 	 * If the pool is newer than the code, we can't open it.
1754 	 */
1755 	if (ub->ub_version > SPA_VERSION)
1756 		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
1757 
1758 	/*
1759 	 * If the vdev guid sum doesn't match the uberblock, we have an
1760 	 * incomplete configuration.
1761 	 */
1762 	if (mosconfig && type != SPA_IMPORT_ASSEMBLE &&
1763 	    rvd->vdev_guid_sum != ub->ub_guid_sum)
1764 		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
1765 
1766 	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
1767 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1768 		spa_try_repair(spa, config);
1769 		spa_config_exit(spa, SCL_ALL, FTAG);
1770 		nvlist_free(spa->spa_config_splitting);
1771 		spa->spa_config_splitting = NULL;
1772 	}
1773 
1774 	/*
1775 	 * Initialize internal SPA structures.
1776 	 */
1777 	spa->spa_state = POOL_STATE_ACTIVE;
1778 	spa->spa_ubsync = spa->spa_uberblock;
1779 	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
1780 	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
1781 	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
1782 	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
1783 	spa->spa_claim_max_txg = spa->spa_first_txg;
1784 	spa->spa_prev_software_version = ub->ub_software_version;
1785 
1786 	error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1787 	if (error)
1788 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1789 	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1790 
1791 	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
1792 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1793 
1794 	if (!mosconfig) {
1795 		uint64_t hostid;
1796 		nvlist_t *policy = NULL, *nvconfig;
1797 
1798 		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
1799 			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1800 
1801 		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
1802 		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
1803 			char *hostname;
1804 			unsigned long myhostid = 0;
1805 
1806 			VERIFY(nvlist_lookup_string(nvconfig,
1807 			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
1808 
1809 #ifdef	_KERNEL
1810 			myhostid = zone_get_hostid(NULL);
1811 #else	/* _KERNEL */
1812 			/*
1813 			 * We're emulating the system's hostid in userland, so
1814 			 * we can't use zone_get_hostid().
1815 			 */
1816 			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
1817 #endif	/* _KERNEL */
1818 			if (hostid != 0 && myhostid != 0 &&
1819 			    hostid != myhostid) {
1820 				nvlist_free(nvconfig);
1821 				cmn_err(CE_WARN, "pool '%s' could not be "
1822 				    "loaded as it was last accessed by "
1823 				    "another system (host: %s hostid: 0x%lx). "
1824 				    "See: http://www.sun.com/msg/ZFS-8000-EY",
1825 				    spa_name(spa), hostname,
1826 				    (unsigned long)hostid);
1827 				return (EBADF);
1828 			}
1829 		}
1830 		if (nvlist_lookup_nvlist(spa->spa_config,
1831 		    ZPOOL_REWIND_POLICY, &policy) == 0)
1832 			VERIFY(nvlist_add_nvlist(nvconfig,
1833 			    ZPOOL_REWIND_POLICY, policy) == 0);
1834 
1835 		spa_config_set(spa, nvconfig);
1836 		spa_unload(spa);
1837 		spa_deactivate(spa);
1838 		spa_activate(spa, orig_mode);
1839 
1840 		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
1841 	}
1842 
1843 	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPLIST,
1844 	    &spa->spa_deferred_bplist_obj) != 0)
1845 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1846 
1847 	/*
1848 	 * Load the bit that tells us to use the new accounting function
1849 	 * (raid-z deflation).  If we have an older pool, this will not
1850 	 * be present.
1851 	 */
1852 	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
1853 	if (error != 0 && error != ENOENT)
1854 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1855 
1856 	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
1857 	    &spa->spa_creation_version);
1858 	if (error != 0 && error != ENOENT)
1859 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1860 
1861 	/*
1862 	 * Load the persistent error log.  If we have an older pool, this will
1863 	 * not be present.
1864 	 */
1865 	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
1866 	if (error != 0 && error != ENOENT)
1867 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1868 
1869 	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
1870 	    &spa->spa_errlog_scrub);
1871 	if (error != 0 && error != ENOENT)
1872 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1873 
1874 	/*
1875 	 * Load the history object.  If we have an older pool, this
1876 	 * will not be present.
1877 	 */
1878 	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
1879 	if (error != 0 && error != ENOENT)
1880 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1881 
1882 	/*
1883 	 * If we're assembling the pool from the split-off vdevs of
1884 	 * an existing pool, we don't want to attach the spares & cache
1885 	 * devices.
1886 	 */
1887 
1888 	/*
1889 	 * Load any hot spares for this pool.
1890 	 */
1891 	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
1892 	if (error != 0 && error != ENOENT)
1893 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1894 	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
1895 		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
1896 		if (load_nvlist(spa, spa->spa_spares.sav_object,
1897 		    &spa->spa_spares.sav_config) != 0)
1898 			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1899 
1900 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1901 		spa_load_spares(spa);
1902 		spa_config_exit(spa, SCL_ALL, FTAG);
1903 	} else if (error == 0) {
1904 		spa->spa_spares.sav_sync = B_TRUE;
1905 	}
1906 
1907 	/*
1908 	 * Load any level 2 ARC devices for this pool.
1909 	 */
1910 	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
1911 	    &spa->spa_l2cache.sav_object);
1912 	if (error != 0 && error != ENOENT)
1913 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1914 	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
1915 		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
1916 		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
1917 		    &spa->spa_l2cache.sav_config) != 0)
1918 			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1919 
1920 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1921 		spa_load_l2cache(spa);
1922 		spa_config_exit(spa, SCL_ALL, FTAG);
1923 	} else if (error == 0) {
1924 		spa->spa_l2cache.sav_sync = B_TRUE;
1925 	}
1926 
1927 	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
1928 
1929 	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
1930 	if (error && error != ENOENT)
1931 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1932 
1933 	if (error == 0) {
1934 		uint64_t autoreplace;
1935 
1936 		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
1937 		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
1938 		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
1939 		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
1940 		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
1941 		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
1942 		    &spa->spa_dedup_ditto);
1943 
1944 		spa->spa_autoreplace = (autoreplace != 0);
1945 	}
1946 
1947 	/*
1948 	 * If the 'autoreplace' property is set, then post a resource notifying
1949 	 * the ZFS DE that it should not issue any faults for unopenable
1950 	 * devices.  We also iterate over the vdevs, and post a sysevent for any
1951 	 * unopenable vdevs so that the normal autoreplace handler can take
1952 	 * over.
1953 	 */
1954 	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
1955 		spa_check_removed(spa->spa_root_vdev);
1956 		/*
1957 		 * For the import case, this is done in spa_import(), because
1958 		 * at this point we're using the spare definitions from
1959 		 * the MOS config, not necessarily from the userland config.
1960 		 */
1961 		if (state != SPA_LOAD_IMPORT) {
1962 			spa_aux_check_removed(&spa->spa_spares);
1963 			spa_aux_check_removed(&spa->spa_l2cache);
1964 		}
1965 	}
1966 
1967 	/*
1968 	 * Load the vdev state for all toplevel vdevs.
1969 	 */
1970 	vdev_load(rvd);
1971 
1972 	/*
1973 	 * Propagate the leaf DTLs we just loaded all the way up the tree.
1974 	 */
1975 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1976 	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
1977 	spa_config_exit(spa, SCL_ALL, FTAG);
1978 
1979 	/*
1980 	 * Check the state of the root vdev.  If it can't be opened, it
1981 	 * indicates one or more toplevel vdevs are faulted.
1982 	 */
1983 	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1984 		return (ENXIO);
1985 
1986 	/*
1987 	 * Load the DDTs (dedup tables).
1988 	 */
1989 	error = ddt_load(spa);
1990 	if (error != 0)
1991 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1992 
1993 	spa_update_dspace(spa);
1994 
1995 	if (state != SPA_LOAD_TRYIMPORT) {
1996 		error = spa_load_verify(spa);
1997 		if (error)
1998 			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
1999 			    error));
2000 	}
2001 
2002 	/*
2003 	 * Load the intent log state and check log integrity.  If we're
2004 	 * assembling a pool from a split, the log is not transferred over.
2005 	 */
2006 	if (type != SPA_IMPORT_ASSEMBLE) {
2007 		nvlist_t *nvconfig;
2008 
2009 		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2010 			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2011 
2012 		VERIFY(nvlist_lookup_nvlist(nvconfig, ZPOOL_CONFIG_VDEV_TREE,
2013 		    &nvroot) == 0);
2014 		spa_load_log_state(spa, nvroot);
2015 		nvlist_free(nvconfig);
2016 
2017 		if (spa_check_logs(spa)) {
2018 			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2019 			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2020 		}
2021 	}
2022 
2023 	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2024 	    spa->spa_load_max_txg == UINT64_MAX)) {
2025 		dmu_tx_t *tx;
2026 		int need_update = B_FALSE;
2027 
2028 		ASSERT(state != SPA_LOAD_TRYIMPORT);
2029 
2030 		/*
2031 		 * Claim log blocks that haven't been committed yet.
2032 		 * This must all happen in a single txg.
2033 		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2034 		 * invoked from zil_claim_log_block()'s i/o done callback.
2035 		 * Price of rollback is that we abandon the log.
2036 		 */
2037 		spa->spa_claiming = B_TRUE;
2038 
2039 		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2040 		    spa_first_txg(spa));
2041 		(void) dmu_objset_find(spa_name(spa),
2042 		    zil_claim, tx, DS_FIND_CHILDREN);
2043 		dmu_tx_commit(tx);
2044 
2045 		spa->spa_claiming = B_FALSE;
2046 
2047 		spa_set_log_state(spa, SPA_LOG_GOOD);
2048 		spa->spa_sync_on = B_TRUE;
2049 		txg_sync_start(spa->spa_dsl_pool);
2050 
2051 		/*
2052 		 * Wait for all claims to sync.  We sync up to the highest
2053 		 * claimed log block birth time so that claimed log blocks
2054 		 * don't appear to be from the future.  spa_claim_max_txg
2055 		 * will have been set for us by either zil_check_log_chain()
2056 		 * (invoked from spa_check_logs()) or zil_claim() above.
2057 		 */
2058 		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2059 
2060 		/*
2061 		 * If the config cache is stale, or we have uninitialized
2062 		 * metaslabs (see spa_vdev_add()), then update the config.
2063 		 *
2064 		 * If spa_load_verbatim is true, trust the current
2065 		 * in-core spa_config and update the disk labels.
2066 		 */
2067 		if (config_cache_txg != spa->spa_config_txg ||
2068 		    state == SPA_LOAD_IMPORT || spa->spa_load_verbatim ||
2069 		    state == SPA_LOAD_RECOVER)
2070 			need_update = B_TRUE;
2071 
2072 		for (int c = 0; c < rvd->vdev_children; c++)
2073 			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2074 				need_update = B_TRUE;
2075 
2076 		/*
2077 		 * Update the config cache asychronously in case we're the
2078 		 * root pool, in which case the config cache isn't writable yet.
2079 		 */
2080 		if (need_update)
2081 			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2082 
2083 		/*
2084 		 * Check all DTLs to see if anything needs resilvering.
2085 		 */
2086 		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2087 		    vdev_resilver_needed(rvd, NULL, NULL))
2088 			spa_async_request(spa, SPA_ASYNC_RESILVER);
2089 
2090 		/*
2091 		 * Delete any inconsistent datasets.
2092 		 */
2093 		(void) dmu_objset_find(spa_name(spa),
2094 		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2095 
2096 		/*
2097 		 * Clean up any stale temporary dataset userrefs.
2098 		 */
2099 		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2100 	}
2101 
2102 	return (0);
2103 }
2104 
2105 static int
2106 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2107 {
2108 	spa_unload(spa);
2109 	spa_deactivate(spa);
2110 
2111 	spa->spa_load_max_txg--;
2112 
2113 	spa_activate(spa, spa_mode_global);
2114 	spa_async_suspend(spa);
2115 
2116 	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2117 }
2118 
2119 static int
2120 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2121     uint64_t max_request, int rewind_flags)
2122 {
2123 	nvlist_t *config = NULL;
2124 	int load_error, rewind_error;
2125 	uint64_t safe_rewind_txg;
2126 	uint64_t min_txg;
2127 
2128 	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2129 		spa->spa_load_max_txg = spa->spa_load_txg;
2130 		spa_set_log_state(spa, SPA_LOG_CLEAR);
2131 	} else {
2132 		spa->spa_load_max_txg = max_request;
2133 	}
2134 
2135 	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2136 	    mosconfig);
2137 	if (load_error == 0)
2138 		return (0);
2139 
2140 	if (spa->spa_root_vdev != NULL)
2141 		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2142 
2143 	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2144 	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2145 
2146 	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2147 		nvlist_free(config);
2148 		return (load_error);
2149 	}
2150 
2151 	/* Price of rolling back is discarding txgs, including log */
2152 	if (state == SPA_LOAD_RECOVER)
2153 		spa_set_log_state(spa, SPA_LOG_CLEAR);
2154 
2155 	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2156 	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2157 	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2158 	    TXG_INITIAL : safe_rewind_txg;
2159 
2160 	/*
2161 	 * Continue as long as we're finding errors, we're still within
2162 	 * the acceptable rewind range, and we're still finding uberblocks
2163 	 */
2164 	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2165 	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2166 		if (spa->spa_load_max_txg < safe_rewind_txg)
2167 			spa->spa_extreme_rewind = B_TRUE;
2168 		rewind_error = spa_load_retry(spa, state, mosconfig);
2169 	}
2170 
2171 	if (config)
2172 		spa_rewind_data_to_nvlist(spa, config);
2173 
2174 	spa->spa_extreme_rewind = B_FALSE;
2175 	spa->spa_load_max_txg = UINT64_MAX;
2176 
2177 	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2178 		spa_config_set(spa, config);
2179 
2180 	return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
2181 }
2182 
2183 /*
2184  * Pool Open/Import
2185  *
2186  * The import case is identical to an open except that the configuration is sent
2187  * down from userland, instead of grabbed from the configuration cache.  For the
2188  * case of an open, the pool configuration will exist in the
2189  * POOL_STATE_UNINITIALIZED state.
2190  *
2191  * The stats information (gen/count/ustats) is used to gather vdev statistics at
2192  * the same time open the pool, without having to keep around the spa_t in some
2193  * ambiguous state.
2194  */
2195 static int
2196 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2197     nvlist_t **config)
2198 {
2199 	spa_t *spa;
2200 	int error;
2201 	int locked = B_FALSE;
2202 
2203 	*spapp = NULL;
2204 
2205 	/*
2206 	 * As disgusting as this is, we need to support recursive calls to this
2207 	 * function because dsl_dir_open() is called during spa_load(), and ends
2208 	 * up calling spa_open() again.  The real fix is to figure out how to
2209 	 * avoid dsl_dir_open() calling this in the first place.
2210 	 */
2211 	if (mutex_owner(&spa_namespace_lock) != curthread) {
2212 		mutex_enter(&spa_namespace_lock);
2213 		locked = B_TRUE;
2214 	}
2215 
2216 	if ((spa = spa_lookup(pool)) == NULL) {
2217 		if (locked)
2218 			mutex_exit(&spa_namespace_lock);
2219 		return (ENOENT);
2220 	}
2221 
2222 	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2223 		spa_load_state_t state = SPA_LOAD_OPEN;
2224 		zpool_rewind_policy_t policy;
2225 
2226 		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2227 		    &policy);
2228 		if (policy.zrp_request & ZPOOL_DO_REWIND)
2229 			state = SPA_LOAD_RECOVER;
2230 
2231 		spa_activate(spa, spa_mode_global);
2232 
2233 		if (spa->spa_last_open_failed && (policy.zrp_request &
2234 		    (ZPOOL_NO_REWIND | ZPOOL_NEVER_REWIND))) {
2235 			if (config != NULL && spa->spa_config)
2236 				VERIFY(nvlist_dup(spa->spa_config,
2237 				    config, KM_SLEEP) == 0);
2238 			spa_deactivate(spa);
2239 			if (locked)
2240 				mutex_exit(&spa_namespace_lock);
2241 			return (spa->spa_last_open_failed);
2242 		}
2243 
2244 		if (state != SPA_LOAD_RECOVER)
2245 			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2246 
2247 		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2248 		    policy.zrp_request);
2249 
2250 		if (error == EBADF) {
2251 			/*
2252 			 * If vdev_validate() returns failure (indicated by
2253 			 * EBADF), it indicates that one of the vdevs indicates
2254 			 * that the pool has been exported or destroyed.  If
2255 			 * this is the case, the config cache is out of sync and
2256 			 * we should remove the pool from the namespace.
2257 			 */
2258 			spa_unload(spa);
2259 			spa_deactivate(spa);
2260 			spa_config_sync(spa, B_TRUE, B_TRUE);
2261 			spa_remove(spa);
2262 			if (locked)
2263 				mutex_exit(&spa_namespace_lock);
2264 			return (ENOENT);
2265 		}
2266 
2267 		if (error) {
2268 			/*
2269 			 * We can't open the pool, but we still have useful
2270 			 * information: the state of each vdev after the
2271 			 * attempted vdev_open().  Return this to the user.
2272 			 */
2273 			if (config != NULL && spa->spa_config)
2274 				VERIFY(nvlist_dup(spa->spa_config, config,
2275 				    KM_SLEEP) == 0);
2276 			spa_unload(spa);
2277 			spa_deactivate(spa);
2278 			spa->spa_last_open_failed = error;
2279 			if (locked)
2280 				mutex_exit(&spa_namespace_lock);
2281 			*spapp = NULL;
2282 			return (error);
2283 		}
2284 
2285 	}
2286 
2287 	spa_open_ref(spa, tag);
2288 
2289 
2290 	if (config != NULL)
2291 		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2292 
2293 	if (locked) {
2294 		spa->spa_last_open_failed = 0;
2295 		spa->spa_last_ubsync_txg = 0;
2296 		spa->spa_load_txg = 0;
2297 		mutex_exit(&spa_namespace_lock);
2298 	}
2299 
2300 	*spapp = spa;
2301 
2302 	return (0);
2303 }
2304 
2305 int
2306 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2307     nvlist_t **config)
2308 {
2309 	return (spa_open_common(name, spapp, tag, policy, config));
2310 }
2311 
2312 int
2313 spa_open(const char *name, spa_t **spapp, void *tag)
2314 {
2315 	return (spa_open_common(name, spapp, tag, NULL, NULL));
2316 }
2317 
2318 /*
2319  * Lookup the given spa_t, incrementing the inject count in the process,
2320  * preventing it from being exported or destroyed.
2321  */
2322 spa_t *
2323 spa_inject_addref(char *name)
2324 {
2325 	spa_t *spa;
2326 
2327 	mutex_enter(&spa_namespace_lock);
2328 	if ((spa = spa_lookup(name)) == NULL) {
2329 		mutex_exit(&spa_namespace_lock);
2330 		return (NULL);
2331 	}
2332 	spa->spa_inject_ref++;
2333 	mutex_exit(&spa_namespace_lock);
2334 
2335 	return (spa);
2336 }
2337 
2338 void
2339 spa_inject_delref(spa_t *spa)
2340 {
2341 	mutex_enter(&spa_namespace_lock);
2342 	spa->spa_inject_ref--;
2343 	mutex_exit(&spa_namespace_lock);
2344 }
2345 
2346 /*
2347  * Add spares device information to the nvlist.
2348  */
2349 static void
2350 spa_add_spares(spa_t *spa, nvlist_t *config)
2351 {
2352 	nvlist_t **spares;
2353 	uint_t i, nspares;
2354 	nvlist_t *nvroot;
2355 	uint64_t guid;
2356 	vdev_stat_t *vs;
2357 	uint_t vsc;
2358 	uint64_t pool;
2359 
2360 	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2361 
2362 	if (spa->spa_spares.sav_count == 0)
2363 		return;
2364 
2365 	VERIFY(nvlist_lookup_nvlist(config,
2366 	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2367 	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2368 	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2369 	if (nspares != 0) {
2370 		VERIFY(nvlist_add_nvlist_array(nvroot,
2371 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2372 		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2373 		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2374 
2375 		/*
2376 		 * Go through and find any spares which have since been
2377 		 * repurposed as an active spare.  If this is the case, update
2378 		 * their status appropriately.
2379 		 */
2380 		for (i = 0; i < nspares; i++) {
2381 			VERIFY(nvlist_lookup_uint64(spares[i],
2382 			    ZPOOL_CONFIG_GUID, &guid) == 0);
2383 			if (spa_spare_exists(guid, &pool, NULL) &&
2384 			    pool != 0ULL) {
2385 				VERIFY(nvlist_lookup_uint64_array(
2386 				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
2387 				    (uint64_t **)&vs, &vsc) == 0);
2388 				vs->vs_state = VDEV_STATE_CANT_OPEN;
2389 				vs->vs_aux = VDEV_AUX_SPARED;
2390 			}
2391 		}
2392 	}
2393 }
2394 
2395 /*
2396  * Add l2cache device information to the nvlist, including vdev stats.
2397  */
2398 static void
2399 spa_add_l2cache(spa_t *spa, nvlist_t *config)
2400 {
2401 	nvlist_t **l2cache;
2402 	uint_t i, j, nl2cache;
2403 	nvlist_t *nvroot;
2404 	uint64_t guid;
2405 	vdev_t *vd;
2406 	vdev_stat_t *vs;
2407 	uint_t vsc;
2408 
2409 	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2410 
2411 	if (spa->spa_l2cache.sav_count == 0)
2412 		return;
2413 
2414 	VERIFY(nvlist_lookup_nvlist(config,
2415 	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2416 	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2417 	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2418 	if (nl2cache != 0) {
2419 		VERIFY(nvlist_add_nvlist_array(nvroot,
2420 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2421 		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2422 		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2423 
2424 		/*
2425 		 * Update level 2 cache device stats.
2426 		 */
2427 
2428 		for (i = 0; i < nl2cache; i++) {
2429 			VERIFY(nvlist_lookup_uint64(l2cache[i],
2430 			    ZPOOL_CONFIG_GUID, &guid) == 0);
2431 
2432 			vd = NULL;
2433 			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2434 				if (guid ==
2435 				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2436 					vd = spa->spa_l2cache.sav_vdevs[j];
2437 					break;
2438 				}
2439 			}
2440 			ASSERT(vd != NULL);
2441 
2442 			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2443 			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2444 			    == 0);
2445 			vdev_get_stats(vd, vs);
2446 		}
2447 	}
2448 }
2449 
2450 int
2451 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2452 {
2453 	int error;
2454 	spa_t *spa;
2455 
2456 	*config = NULL;
2457 	error = spa_open_common(name, &spa, FTAG, NULL, config);
2458 
2459 	if (spa != NULL) {
2460 		/*
2461 		 * This still leaves a window of inconsistency where the spares
2462 		 * or l2cache devices could change and the config would be
2463 		 * self-inconsistent.
2464 		 */
2465 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2466 
2467 		if (*config != NULL) {
2468 			VERIFY(nvlist_add_uint64(*config,
2469 			    ZPOOL_CONFIG_ERRCOUNT,
2470 			    spa_get_errlog_size(spa)) == 0);
2471 
2472 			if (spa_suspended(spa))
2473 				VERIFY(nvlist_add_uint64(*config,
2474 				    ZPOOL_CONFIG_SUSPENDED,
2475 				    spa->spa_failmode) == 0);
2476 
2477 			spa_add_spares(spa, *config);
2478 			spa_add_l2cache(spa, *config);
2479 		}
2480 	}
2481 
2482 	/*
2483 	 * We want to get the alternate root even for faulted pools, so we cheat
2484 	 * and call spa_lookup() directly.
2485 	 */
2486 	if (altroot) {
2487 		if (spa == NULL) {
2488 			mutex_enter(&spa_namespace_lock);
2489 			spa = spa_lookup(name);
2490 			if (spa)
2491 				spa_altroot(spa, altroot, buflen);
2492 			else
2493 				altroot[0] = '\0';
2494 			spa = NULL;
2495 			mutex_exit(&spa_namespace_lock);
2496 		} else {
2497 			spa_altroot(spa, altroot, buflen);
2498 		}
2499 	}
2500 
2501 	if (spa != NULL) {
2502 		spa_config_exit(spa, SCL_CONFIG, FTAG);
2503 		spa_close(spa, FTAG);
2504 	}
2505 
2506 	return (error);
2507 }
2508 
2509 /*
2510  * Validate that the auxiliary device array is well formed.  We must have an
2511  * array of nvlists, each which describes a valid leaf vdev.  If this is an
2512  * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2513  * specified, as long as they are well-formed.
2514  */
2515 static int
2516 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2517     spa_aux_vdev_t *sav, const char *config, uint64_t version,
2518     vdev_labeltype_t label)
2519 {
2520 	nvlist_t **dev;
2521 	uint_t i, ndev;
2522 	vdev_t *vd;
2523 	int error;
2524 
2525 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2526 
2527 	/*
2528 	 * It's acceptable to have no devs specified.
2529 	 */
2530 	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2531 		return (0);
2532 
2533 	if (ndev == 0)
2534 		return (EINVAL);
2535 
2536 	/*
2537 	 * Make sure the pool is formatted with a version that supports this
2538 	 * device type.
2539 	 */
2540 	if (spa_version(spa) < version)
2541 		return (ENOTSUP);
2542 
2543 	/*
2544 	 * Set the pending device list so we correctly handle device in-use
2545 	 * checking.
2546 	 */
2547 	sav->sav_pending = dev;
2548 	sav->sav_npending = ndev;
2549 
2550 	for (i = 0; i < ndev; i++) {
2551 		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2552 		    mode)) != 0)
2553 			goto out;
2554 
2555 		if (!vd->vdev_ops->vdev_op_leaf) {
2556 			vdev_free(vd);
2557 			error = EINVAL;
2558 			goto out;
2559 		}
2560 
2561 		/*
2562 		 * The L2ARC currently only supports disk devices in
2563 		 * kernel context.  For user-level testing, we allow it.
2564 		 */
2565 #ifdef _KERNEL
2566 		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2567 		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2568 			error = ENOTBLK;
2569 			goto out;
2570 		}
2571 #endif
2572 		vd->vdev_top = vd;
2573 
2574 		if ((error = vdev_open(vd)) == 0 &&
2575 		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
2576 			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2577 			    vd->vdev_guid) == 0);
2578 		}
2579 
2580 		vdev_free(vd);
2581 
2582 		if (error &&
2583 		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2584 			goto out;
2585 		else
2586 			error = 0;
2587 	}
2588 
2589 out:
2590 	sav->sav_pending = NULL;
2591 	sav->sav_npending = 0;
2592 	return (error);
2593 }
2594 
2595 static int
2596 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2597 {
2598 	int error;
2599 
2600 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2601 
2602 	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2603 	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2604 	    VDEV_LABEL_SPARE)) != 0) {
2605 		return (error);
2606 	}
2607 
2608 	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2609 	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2610 	    VDEV_LABEL_L2CACHE));
2611 }
2612 
2613 static void
2614 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2615     const char *config)
2616 {
2617 	int i;
2618 
2619 	if (sav->sav_config != NULL) {
2620 		nvlist_t **olddevs;
2621 		uint_t oldndevs;
2622 		nvlist_t **newdevs;
2623 
2624 		/*
2625 		 * Generate new dev list by concatentating with the
2626 		 * current dev list.
2627 		 */
2628 		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2629 		    &olddevs, &oldndevs) == 0);
2630 
2631 		newdevs = kmem_alloc(sizeof (void *) *
2632 		    (ndevs + oldndevs), KM_SLEEP);
2633 		for (i = 0; i < oldndevs; i++)
2634 			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2635 			    KM_SLEEP) == 0);
2636 		for (i = 0; i < ndevs; i++)
2637 			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2638 			    KM_SLEEP) == 0);
2639 
2640 		VERIFY(nvlist_remove(sav->sav_config, config,
2641 		    DATA_TYPE_NVLIST_ARRAY) == 0);
2642 
2643 		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2644 		    config, newdevs, ndevs + oldndevs) == 0);
2645 		for (i = 0; i < oldndevs + ndevs; i++)
2646 			nvlist_free(newdevs[i]);
2647 		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2648 	} else {
2649 		/*
2650 		 * Generate a new dev list.
2651 		 */
2652 		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2653 		    KM_SLEEP) == 0);
2654 		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2655 		    devs, ndevs) == 0);
2656 	}
2657 }
2658 
2659 /*
2660  * Stop and drop level 2 ARC devices
2661  */
2662 void
2663 spa_l2cache_drop(spa_t *spa)
2664 {
2665 	vdev_t *vd;
2666 	int i;
2667 	spa_aux_vdev_t *sav = &spa->spa_l2cache;
2668 
2669 	for (i = 0; i < sav->sav_count; i++) {
2670 		uint64_t pool;
2671 
2672 		vd = sav->sav_vdevs[i];
2673 		ASSERT(vd != NULL);
2674 
2675 		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2676 		    pool != 0ULL && l2arc_vdev_present(vd))
2677 			l2arc_remove_vdev(vd);
2678 		if (vd->vdev_isl2cache)
2679 			spa_l2cache_remove(vd);
2680 		vdev_clear_stats(vd);
2681 		(void) vdev_close(vd);
2682 	}
2683 }
2684 
2685 /*
2686  * Pool Creation
2687  */
2688 int
2689 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2690     const char *history_str, nvlist_t *zplprops)
2691 {
2692 	spa_t *spa;
2693 	char *altroot = NULL;
2694 	vdev_t *rvd;
2695 	dsl_pool_t *dp;
2696 	dmu_tx_t *tx;
2697 	int error = 0;
2698 	uint64_t txg = TXG_INITIAL;
2699 	nvlist_t **spares, **l2cache;
2700 	uint_t nspares, nl2cache;
2701 	uint64_t version;
2702 
2703 	/*
2704 	 * If this pool already exists, return failure.
2705 	 */
2706 	mutex_enter(&spa_namespace_lock);
2707 	if (spa_lookup(pool) != NULL) {
2708 		mutex_exit(&spa_namespace_lock);
2709 		return (EEXIST);
2710 	}
2711 
2712 	/*
2713 	 * Allocate a new spa_t structure.
2714 	 */
2715 	(void) nvlist_lookup_string(props,
2716 	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2717 	spa = spa_add(pool, NULL, altroot);
2718 	spa_activate(spa, spa_mode_global);
2719 
2720 	if (props && (error = spa_prop_validate(spa, props))) {
2721 		spa_deactivate(spa);
2722 		spa_remove(spa);
2723 		mutex_exit(&spa_namespace_lock);
2724 		return (error);
2725 	}
2726 
2727 	if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2728 	    &version) != 0)
2729 		version = SPA_VERSION;
2730 	ASSERT(version <= SPA_VERSION);
2731 
2732 	spa->spa_first_txg = txg;
2733 	spa->spa_uberblock.ub_txg = txg - 1;
2734 	spa->spa_uberblock.ub_version = version;
2735 	spa->spa_ubsync = spa->spa_uberblock;
2736 
2737 	/*
2738 	 * Create "The Godfather" zio to hold all async IOs
2739 	 */
2740 	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2741 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2742 
2743 	/*
2744 	 * Create the root vdev.
2745 	 */
2746 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2747 
2748 	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2749 
2750 	ASSERT(error != 0 || rvd != NULL);
2751 	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2752 
2753 	if (error == 0 && !zfs_allocatable_devs(nvroot))
2754 		error = EINVAL;
2755 
2756 	if (error == 0 &&
2757 	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2758 	    (error = spa_validate_aux(spa, nvroot, txg,
2759 	    VDEV_ALLOC_ADD)) == 0) {
2760 		for (int c = 0; c < rvd->vdev_children; c++) {
2761 			vdev_metaslab_set_size(rvd->vdev_child[c]);
2762 			vdev_expand(rvd->vdev_child[c], txg);
2763 		}
2764 	}
2765 
2766 	spa_config_exit(spa, SCL_ALL, FTAG);
2767 
2768 	if (error != 0) {
2769 		spa_unload(spa);
2770 		spa_deactivate(spa);
2771 		spa_remove(spa);
2772 		mutex_exit(&spa_namespace_lock);
2773 		return (error);
2774 	}
2775 
2776 	/*
2777 	 * Get the list of spares, if specified.
2778 	 */
2779 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2780 	    &spares, &nspares) == 0) {
2781 		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
2782 		    KM_SLEEP) == 0);
2783 		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2784 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2785 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2786 		spa_load_spares(spa);
2787 		spa_config_exit(spa, SCL_ALL, FTAG);
2788 		spa->spa_spares.sav_sync = B_TRUE;
2789 	}
2790 
2791 	/*
2792 	 * Get the list of level 2 cache devices, if specified.
2793 	 */
2794 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2795 	    &l2cache, &nl2cache) == 0) {
2796 		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2797 		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2798 		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2799 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2800 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2801 		spa_load_l2cache(spa);
2802 		spa_config_exit(spa, SCL_ALL, FTAG);
2803 		spa->spa_l2cache.sav_sync = B_TRUE;
2804 	}
2805 
2806 	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
2807 	spa->spa_meta_objset = dp->dp_meta_objset;
2808 
2809 	/*
2810 	 * Create DDTs (dedup tables).
2811 	 */
2812 	ddt_create(spa);
2813 
2814 	spa_update_dspace(spa);
2815 
2816 	tx = dmu_tx_create_assigned(dp, txg);
2817 
2818 	/*
2819 	 * Create the pool config object.
2820 	 */
2821 	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
2822 	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
2823 	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2824 
2825 	if (zap_add(spa->spa_meta_objset,
2826 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
2827 	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
2828 		cmn_err(CE_PANIC, "failed to add pool config");
2829 	}
2830 
2831 	if (zap_add(spa->spa_meta_objset,
2832 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
2833 	    sizeof (uint64_t), 1, &version, tx) != 0) {
2834 		cmn_err(CE_PANIC, "failed to add pool version");
2835 	}
2836 
2837 	/* Newly created pools with the right version are always deflated. */
2838 	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
2839 		spa->spa_deflate = TRUE;
2840 		if (zap_add(spa->spa_meta_objset,
2841 		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
2842 		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
2843 			cmn_err(CE_PANIC, "failed to add deflate");
2844 		}
2845 	}
2846 
2847 	/*
2848 	 * Create the deferred-free bplist object.  Turn off compression
2849 	 * because sync-to-convergence takes longer if the blocksize
2850 	 * keeps changing.
2851 	 */
2852 	spa->spa_deferred_bplist_obj = bplist_create(spa->spa_meta_objset,
2853 	    1 << 14, tx);
2854 	dmu_object_set_compress(spa->spa_meta_objset,
2855 	    spa->spa_deferred_bplist_obj, ZIO_COMPRESS_OFF, tx);
2856 
2857 	if (zap_add(spa->spa_meta_objset,
2858 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
2859 	    sizeof (uint64_t), 1, &spa->spa_deferred_bplist_obj, tx) != 0) {
2860 		cmn_err(CE_PANIC, "failed to add bplist");
2861 	}
2862 
2863 	/*
2864 	 * Create the pool's history object.
2865 	 */
2866 	if (version >= SPA_VERSION_ZPOOL_HISTORY)
2867 		spa_history_create_obj(spa, tx);
2868 
2869 	/*
2870 	 * Set pool properties.
2871 	 */
2872 	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
2873 	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2874 	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
2875 	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
2876 
2877 	if (props != NULL) {
2878 		spa_configfile_set(spa, props, B_FALSE);
2879 		spa_sync_props(spa, props, tx);
2880 	}
2881 
2882 	dmu_tx_commit(tx);
2883 
2884 	spa->spa_sync_on = B_TRUE;
2885 	txg_sync_start(spa->spa_dsl_pool);
2886 
2887 	/*
2888 	 * We explicitly wait for the first transaction to complete so that our
2889 	 * bean counters are appropriately updated.
2890 	 */
2891 	txg_wait_synced(spa->spa_dsl_pool, txg);
2892 
2893 	spa_config_sync(spa, B_FALSE, B_TRUE);
2894 
2895 	if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
2896 		(void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
2897 	spa_history_log_version(spa, LOG_POOL_CREATE);
2898 
2899 	spa->spa_minref = refcount_count(&spa->spa_refcount);
2900 
2901 	mutex_exit(&spa_namespace_lock);
2902 
2903 	return (0);
2904 }
2905 
2906 #ifdef _KERNEL
2907 /*
2908  * Get the root pool information from the root disk, then import the root pool
2909  * during the system boot up time.
2910  */
2911 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
2912 
2913 static nvlist_t *
2914 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
2915 {
2916 	nvlist_t *config;
2917 	nvlist_t *nvtop, *nvroot;
2918 	uint64_t pgid;
2919 
2920 	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
2921 		return (NULL);
2922 
2923 	/*
2924 	 * Add this top-level vdev to the child array.
2925 	 */
2926 	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2927 	    &nvtop) == 0);
2928 	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
2929 	    &pgid) == 0);
2930 	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
2931 
2932 	/*
2933 	 * Put this pool's top-level vdevs into a root vdev.
2934 	 */
2935 	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2936 	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
2937 	    VDEV_TYPE_ROOT) == 0);
2938 	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
2939 	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
2940 	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2941 	    &nvtop, 1) == 0);
2942 
2943 	/*
2944 	 * Replace the existing vdev_tree with the new root vdev in
2945 	 * this pool's configuration (remove the old, add the new).
2946 	 */
2947 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
2948 	nvlist_free(nvroot);
2949 	return (config);
2950 }
2951 
2952 /*
2953  * Walk the vdev tree and see if we can find a device with "better"
2954  * configuration. A configuration is "better" if the label on that
2955  * device has a more recent txg.
2956  */
2957 static void
2958 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
2959 {
2960 	for (int c = 0; c < vd->vdev_children; c++)
2961 		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
2962 
2963 	if (vd->vdev_ops->vdev_op_leaf) {
2964 		nvlist_t *label;
2965 		uint64_t label_txg;
2966 
2967 		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
2968 		    &label) != 0)
2969 			return;
2970 
2971 		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
2972 		    &label_txg) == 0);
2973 
2974 		/*
2975 		 * Do we have a better boot device?
2976 		 */
2977 		if (label_txg > *txg) {
2978 			*txg = label_txg;
2979 			*avd = vd;
2980 		}
2981 		nvlist_free(label);
2982 	}
2983 }
2984 
2985 /*
2986  * Import a root pool.
2987  *
2988  * For x86. devpath_list will consist of devid and/or physpath name of
2989  * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
2990  * The GRUB "findroot" command will return the vdev we should boot.
2991  *
2992  * For Sparc, devpath_list consists the physpath name of the booting device
2993  * no matter the rootpool is a single device pool or a mirrored pool.
2994  * e.g.
2995  *	"/pci@1f,0/ide@d/disk@0,0:a"
2996  */
2997 int
2998 spa_import_rootpool(char *devpath, char *devid)
2999 {
3000 	spa_t *spa;
3001 	vdev_t *rvd, *bvd, *avd = NULL;
3002 	nvlist_t *config, *nvtop;
3003 	uint64_t guid, txg;
3004 	char *pname;
3005 	int error;
3006 
3007 	/*
3008 	 * Read the label from the boot device and generate a configuration.
3009 	 */
3010 	config = spa_generate_rootconf(devpath, devid, &guid);
3011 #if defined(_OBP) && defined(_KERNEL)
3012 	if (config == NULL) {
3013 		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3014 			/* iscsi boot */
3015 			get_iscsi_bootpath_phy(devpath);
3016 			config = spa_generate_rootconf(devpath, devid, &guid);
3017 		}
3018 	}
3019 #endif
3020 	if (config == NULL) {
3021 		cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
3022 		    devpath);
3023 		return (EIO);
3024 	}
3025 
3026 	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3027 	    &pname) == 0);
3028 	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3029 
3030 	mutex_enter(&spa_namespace_lock);
3031 	if ((spa = spa_lookup(pname)) != NULL) {
3032 		/*
3033 		 * Remove the existing root pool from the namespace so that we
3034 		 * can replace it with the correct config we just read in.
3035 		 */
3036 		spa_remove(spa);
3037 	}
3038 
3039 	spa = spa_add(pname, config, NULL);
3040 	spa->spa_is_root = B_TRUE;
3041 	spa->spa_load_verbatim = B_TRUE;
3042 
3043 	/*
3044 	 * Build up a vdev tree based on the boot device's label config.
3045 	 */
3046 	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3047 	    &nvtop) == 0);
3048 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3049 	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3050 	    VDEV_ALLOC_ROOTPOOL);
3051 	spa_config_exit(spa, SCL_ALL, FTAG);
3052 	if (error) {
3053 		mutex_exit(&spa_namespace_lock);
3054 		nvlist_free(config);
3055 		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3056 		    pname);
3057 		return (error);
3058 	}
3059 
3060 	/*
3061 	 * Get the boot vdev.
3062 	 */
3063 	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3064 		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3065 		    (u_longlong_t)guid);
3066 		error = ENOENT;
3067 		goto out;
3068 	}
3069 
3070 	/*
3071 	 * Determine if there is a better boot device.
3072 	 */
3073 	avd = bvd;
3074 	spa_alt_rootvdev(rvd, &avd, &txg);
3075 	if (avd != bvd) {
3076 		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3077 		    "try booting from '%s'", avd->vdev_path);
3078 		error = EINVAL;
3079 		goto out;
3080 	}
3081 
3082 	/*
3083 	 * If the boot device is part of a spare vdev then ensure that
3084 	 * we're booting off the active spare.
3085 	 */
3086 	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3087 	    !bvd->vdev_isspare) {
3088 		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3089 		    "try booting from '%s'",
3090 		    bvd->vdev_parent->vdev_child[1]->vdev_path);
3091 		error = EINVAL;
3092 		goto out;
3093 	}
3094 
3095 	error = 0;
3096 	spa_history_log_version(spa, LOG_POOL_IMPORT);
3097 out:
3098 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3099 	vdev_free(rvd);
3100 	spa_config_exit(spa, SCL_ALL, FTAG);
3101 	mutex_exit(&spa_namespace_lock);
3102 
3103 	nvlist_free(config);
3104 	return (error);
3105 }
3106 
3107 #endif
3108 
3109 /*
3110  * Take a pool and insert it into the namespace as if it had been loaded at
3111  * boot.
3112  */
3113 int
3114 spa_import_verbatim(const char *pool, nvlist_t *config, nvlist_t *props)
3115 {
3116 	spa_t *spa;
3117 	char *altroot = NULL;
3118 
3119 	mutex_enter(&spa_namespace_lock);
3120 	if (spa_lookup(pool) != NULL) {
3121 		mutex_exit(&spa_namespace_lock);
3122 		return (EEXIST);
3123 	}
3124 
3125 	(void) nvlist_lookup_string(props,
3126 	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3127 	spa = spa_add(pool, config, altroot);
3128 
3129 	spa->spa_load_verbatim = B_TRUE;
3130 
3131 	if (props != NULL)
3132 		spa_configfile_set(spa, props, B_FALSE);
3133 
3134 	spa_config_sync(spa, B_FALSE, B_TRUE);
3135 
3136 	mutex_exit(&spa_namespace_lock);
3137 	spa_history_log_version(spa, LOG_POOL_IMPORT);
3138 
3139 	return (0);
3140 }
3141 
3142 /*
3143  * Import a non-root pool into the system.
3144  */
3145 int
3146 spa_import(const char *pool, nvlist_t *config, nvlist_t *props)
3147 {
3148 	spa_t *spa;
3149 	char *altroot = NULL;
3150 	spa_load_state_t state = SPA_LOAD_IMPORT;
3151 	zpool_rewind_policy_t policy;
3152 	int error;
3153 	nvlist_t *nvroot;
3154 	nvlist_t **spares, **l2cache;
3155 	uint_t nspares, nl2cache;
3156 
3157 	/*
3158 	 * If a pool with this name exists, return failure.
3159 	 */
3160 	mutex_enter(&spa_namespace_lock);
3161 	if (spa_lookup(pool) != NULL) {
3162 		mutex_exit(&spa_namespace_lock);
3163 		return (EEXIST);
3164 	}
3165 
3166 	zpool_get_rewind_policy(config, &policy);
3167 	if (policy.zrp_request & ZPOOL_DO_REWIND)
3168 		state = SPA_LOAD_RECOVER;
3169 
3170 	/*
3171 	 * Create and initialize the spa structure.
3172 	 */
3173 	(void) nvlist_lookup_string(props,
3174 	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3175 	spa = spa_add(pool, config, altroot);
3176 	spa_activate(spa, spa_mode_global);
3177 
3178 	/*
3179 	 * Don't start async tasks until we know everything is healthy.
3180 	 */
3181 	spa_async_suspend(spa);
3182 
3183 	/*
3184 	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3185 	 * because the user-supplied config is actually the one to trust when
3186 	 * doing an import.
3187 	 */
3188 	if (state != SPA_LOAD_RECOVER)
3189 		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3190 	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3191 	    policy.zrp_request);
3192 
3193 	/*
3194 	 * Propagate anything learned about failing or best txgs
3195 	 * back to caller
3196 	 */
3197 	spa_rewind_data_to_nvlist(spa, config);
3198 
3199 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3200 	/*
3201 	 * Toss any existing sparelist, as it doesn't have any validity
3202 	 * anymore, and conflicts with spa_has_spare().
3203 	 */
3204 	if (spa->spa_spares.sav_config) {
3205 		nvlist_free(spa->spa_spares.sav_config);
3206 		spa->spa_spares.sav_config = NULL;
3207 		spa_load_spares(spa);
3208 	}
3209 	if (spa->spa_l2cache.sav_config) {
3210 		nvlist_free(spa->spa_l2cache.sav_config);
3211 		spa->spa_l2cache.sav_config = NULL;
3212 		spa_load_l2cache(spa);
3213 	}
3214 
3215 	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3216 	    &nvroot) == 0);
3217 	if (error == 0)
3218 		error = spa_validate_aux(spa, nvroot, -1ULL,
3219 		    VDEV_ALLOC_SPARE);
3220 	if (error == 0)
3221 		error = spa_validate_aux(spa, nvroot, -1ULL,
3222 		    VDEV_ALLOC_L2CACHE);
3223 	spa_config_exit(spa, SCL_ALL, FTAG);
3224 
3225 	if (props != NULL)
3226 		spa_configfile_set(spa, props, B_FALSE);
3227 
3228 	if (error != 0 || (props && spa_writeable(spa) &&
3229 	    (error = spa_prop_set(spa, props)))) {
3230 		spa_unload(spa);
3231 		spa_deactivate(spa);
3232 		spa_remove(spa);
3233 		mutex_exit(&spa_namespace_lock);
3234 		return (error);
3235 	}
3236 
3237 	/*
3238 	 * Override any spares and level 2 cache devices as specified by
3239 	 * the user, as these may have correct device names/devids, etc.
3240 	 */
3241 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3242 	    &spares, &nspares) == 0) {
3243 		if (spa->spa_spares.sav_config)
3244 			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3245 			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3246 		else
3247 			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3248 			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3249 		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3250 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3251 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3252 		spa_load_spares(spa);
3253 		spa_config_exit(spa, SCL_ALL, FTAG);
3254 		spa->spa_spares.sav_sync = B_TRUE;
3255 	}
3256 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3257 	    &l2cache, &nl2cache) == 0) {
3258 		if (spa->spa_l2cache.sav_config)
3259 			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3260 			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3261 		else
3262 			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3263 			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3264 		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3265 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3266 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3267 		spa_load_l2cache(spa);
3268 		spa_config_exit(spa, SCL_ALL, FTAG);
3269 		spa->spa_l2cache.sav_sync = B_TRUE;
3270 	}
3271 
3272 	/*
3273 	 * Check for any removed devices.
3274 	 */
3275 	if (spa->spa_autoreplace) {
3276 		spa_aux_check_removed(&spa->spa_spares);
3277 		spa_aux_check_removed(&spa->spa_l2cache);
3278 	}
3279 
3280 	if (spa_writeable(spa)) {
3281 		/*
3282 		 * Update the config cache to include the newly-imported pool.
3283 		 */
3284 		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3285 	}
3286 
3287 	spa_async_resume(spa);
3288 
3289 	/*
3290 	 * It's possible that the pool was expanded while it was exported.
3291 	 * We kick off an async task to handle this for us.
3292 	 */
3293 	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3294 
3295 	mutex_exit(&spa_namespace_lock);
3296 	spa_history_log_version(spa, LOG_POOL_IMPORT);
3297 
3298 	return (0);
3299 }
3300 
3301 nvlist_t *
3302 spa_tryimport(nvlist_t *tryconfig)
3303 {
3304 	nvlist_t *config = NULL;
3305 	char *poolname;
3306 	spa_t *spa;
3307 	uint64_t state;
3308 	int error;
3309 
3310 	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3311 		return (NULL);
3312 
3313 	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3314 		return (NULL);
3315 
3316 	/*
3317 	 * Create and initialize the spa structure.
3318 	 */
3319 	mutex_enter(&spa_namespace_lock);
3320 	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3321 	spa_activate(spa, FREAD);
3322 
3323 	/*
3324 	 * Pass off the heavy lifting to spa_load().
3325 	 * Pass TRUE for mosconfig because the user-supplied config
3326 	 * is actually the one to trust when doing an import.
3327 	 */
3328 	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3329 
3330 	/*
3331 	 * If 'tryconfig' was at least parsable, return the current config.
3332 	 */
3333 	if (spa->spa_root_vdev != NULL) {
3334 		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3335 		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3336 		    poolname) == 0);
3337 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3338 		    state) == 0);
3339 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3340 		    spa->spa_uberblock.ub_timestamp) == 0);
3341 
3342 		/*
3343 		 * If the bootfs property exists on this pool then we
3344 		 * copy it out so that external consumers can tell which
3345 		 * pools are bootable.
3346 		 */
3347 		if ((!error || error == EEXIST) && spa->spa_bootfs) {
3348 			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3349 
3350 			/*
3351 			 * We have to play games with the name since the
3352 			 * pool was opened as TRYIMPORT_NAME.
3353 			 */
3354 			if (dsl_dsobj_to_dsname(spa_name(spa),
3355 			    spa->spa_bootfs, tmpname) == 0) {
3356 				char *cp;
3357 				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3358 
3359 				cp = strchr(tmpname, '/');
3360 				if (cp == NULL) {
3361 					(void) strlcpy(dsname, tmpname,
3362 					    MAXPATHLEN);
3363 				} else {
3364 					(void) snprintf(dsname, MAXPATHLEN,
3365 					    "%s/%s", poolname, ++cp);
3366 				}
3367 				VERIFY(nvlist_add_string(config,
3368 				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3369 				kmem_free(dsname, MAXPATHLEN);
3370 			}
3371 			kmem_free(tmpname, MAXPATHLEN);
3372 		}
3373 
3374 		/*
3375 		 * Add the list of hot spares and level 2 cache devices.
3376 		 */
3377 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3378 		spa_add_spares(spa, config);
3379 		spa_add_l2cache(spa, config);
3380 		spa_config_exit(spa, SCL_CONFIG, FTAG);
3381 	}
3382 
3383 	spa_unload(spa);
3384 	spa_deactivate(spa);
3385 	spa_remove(spa);
3386 	mutex_exit(&spa_namespace_lock);
3387 
3388 	return (config);
3389 }
3390 
3391 /*
3392  * Pool export/destroy
3393  *
3394  * The act of destroying or exporting a pool is very simple.  We make sure there
3395  * is no more pending I/O and any references to the pool are gone.  Then, we
3396  * update the pool state and sync all the labels to disk, removing the
3397  * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3398  * we don't sync the labels or remove the configuration cache.
3399  */
3400 static int
3401 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3402     boolean_t force, boolean_t hardforce)
3403 {
3404 	spa_t *spa;
3405 
3406 	if (oldconfig)
3407 		*oldconfig = NULL;
3408 
3409 	if (!(spa_mode_global & FWRITE))
3410 		return (EROFS);
3411 
3412 	mutex_enter(&spa_namespace_lock);
3413 	if ((spa = spa_lookup(pool)) == NULL) {
3414 		mutex_exit(&spa_namespace_lock);
3415 		return (ENOENT);
3416 	}
3417 
3418 	/*
3419 	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3420 	 * reacquire the namespace lock, and see if we can export.
3421 	 */
3422 	spa_open_ref(spa, FTAG);
3423 	mutex_exit(&spa_namespace_lock);
3424 	spa_async_suspend(spa);
3425 	mutex_enter(&spa_namespace_lock);
3426 	spa_close(spa, FTAG);
3427 
3428 	/*
3429 	 * The pool will be in core if it's openable,
3430 	 * in which case we can modify its state.
3431 	 */
3432 	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3433 		/*
3434 		 * Objsets may be open only because they're dirty, so we
3435 		 * have to force it to sync before checking spa_refcnt.
3436 		 */
3437 		txg_wait_synced(spa->spa_dsl_pool, 0);
3438 
3439 		/*
3440 		 * A pool cannot be exported or destroyed if there are active
3441 		 * references.  If we are resetting a pool, allow references by
3442 		 * fault injection handlers.
3443 		 */
3444 		if (!spa_refcount_zero(spa) ||
3445 		    (spa->spa_inject_ref != 0 &&
3446 		    new_state != POOL_STATE_UNINITIALIZED)) {
3447 			spa_async_resume(spa);
3448 			mutex_exit(&spa_namespace_lock);
3449 			return (EBUSY);
3450 		}
3451 
3452 		/*
3453 		 * A pool cannot be exported if it has an active shared spare.
3454 		 * This is to prevent other pools stealing the active spare
3455 		 * from an exported pool. At user's own will, such pool can
3456 		 * be forcedly exported.
3457 		 */
3458 		if (!force && new_state == POOL_STATE_EXPORTED &&
3459 		    spa_has_active_shared_spare(spa)) {
3460 			spa_async_resume(spa);
3461 			mutex_exit(&spa_namespace_lock);
3462 			return (EXDEV);
3463 		}
3464 
3465 		/*
3466 		 * We want this to be reflected on every label,
3467 		 * so mark them all dirty.  spa_unload() will do the
3468 		 * final sync that pushes these changes out.
3469 		 */
3470 		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3471 			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3472 			spa->spa_state = new_state;
3473 			spa->spa_final_txg = spa_last_synced_txg(spa) +
3474 			    TXG_DEFER_SIZE + 1;
3475 			vdev_config_dirty(spa->spa_root_vdev);
3476 			spa_config_exit(spa, SCL_ALL, FTAG);
3477 		}
3478 	}
3479 
3480 	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
3481 
3482 	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3483 		spa_unload(spa);
3484 		spa_deactivate(spa);
3485 	}
3486 
3487 	if (oldconfig && spa->spa_config)
3488 		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3489 
3490 	if (new_state != POOL_STATE_UNINITIALIZED) {
3491 		if (!hardforce)
3492 			spa_config_sync(spa, B_TRUE, B_TRUE);
3493 		spa_remove(spa);
3494 	}
3495 	mutex_exit(&spa_namespace_lock);
3496 
3497 	return (0);
3498 }
3499 
3500 /*
3501  * Destroy a storage pool.
3502  */
3503 int
3504 spa_destroy(char *pool)
3505 {
3506 	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3507 	    B_FALSE, B_FALSE));
3508 }
3509 
3510 /*
3511  * Export a storage pool.
3512  */
3513 int
3514 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3515     boolean_t hardforce)
3516 {
3517 	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3518 	    force, hardforce));
3519 }
3520 
3521 /*
3522  * Similar to spa_export(), this unloads the spa_t without actually removing it
3523  * from the namespace in any way.
3524  */
3525 int
3526 spa_reset(char *pool)
3527 {
3528 	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3529 	    B_FALSE, B_FALSE));
3530 }
3531 
3532 /*
3533  * ==========================================================================
3534  * Device manipulation
3535  * ==========================================================================
3536  */
3537 
3538 /*
3539  * Add a device to a storage pool.
3540  */
3541 int
3542 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3543 {
3544 	uint64_t txg, id;
3545 	int error;
3546 	vdev_t *rvd = spa->spa_root_vdev;
3547 	vdev_t *vd, *tvd;
3548 	nvlist_t **spares, **l2cache;
3549 	uint_t nspares, nl2cache;
3550 
3551 	txg = spa_vdev_enter(spa);
3552 
3553 	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3554 	    VDEV_ALLOC_ADD)) != 0)
3555 		return (spa_vdev_exit(spa, NULL, txg, error));
3556 
3557 	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
3558 
3559 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3560 	    &nspares) != 0)
3561 		nspares = 0;
3562 
3563 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3564 	    &nl2cache) != 0)
3565 		nl2cache = 0;
3566 
3567 	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3568 		return (spa_vdev_exit(spa, vd, txg, EINVAL));
3569 
3570 	if (vd->vdev_children != 0 &&
3571 	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
3572 		return (spa_vdev_exit(spa, vd, txg, error));
3573 
3574 	/*
3575 	 * We must validate the spares and l2cache devices after checking the
3576 	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
3577 	 */
3578 	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3579 		return (spa_vdev_exit(spa, vd, txg, error));
3580 
3581 	/*
3582 	 * Transfer each new top-level vdev from vd to rvd.
3583 	 */
3584 	for (int c = 0; c < vd->vdev_children; c++) {
3585 
3586 		/*
3587 		 * Set the vdev id to the first hole, if one exists.
3588 		 */
3589 		for (id = 0; id < rvd->vdev_children; id++) {
3590 			if (rvd->vdev_child[id]->vdev_ishole) {
3591 				vdev_free(rvd->vdev_child[id]);
3592 				break;
3593 			}
3594 		}
3595 		tvd = vd->vdev_child[c];
3596 		vdev_remove_child(vd, tvd);
3597 		tvd->vdev_id = id;
3598 		vdev_add_child(rvd, tvd);
3599 		vdev_config_dirty(tvd);
3600 	}
3601 
3602 	if (nspares != 0) {
3603 		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3604 		    ZPOOL_CONFIG_SPARES);
3605 		spa_load_spares(spa);
3606 		spa->spa_spares.sav_sync = B_TRUE;
3607 	}
3608 
3609 	if (nl2cache != 0) {
3610 		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3611 		    ZPOOL_CONFIG_L2CACHE);
3612 		spa_load_l2cache(spa);
3613 		spa->spa_l2cache.sav_sync = B_TRUE;
3614 	}
3615 
3616 	/*
3617 	 * We have to be careful when adding new vdevs to an existing pool.
3618 	 * If other threads start allocating from these vdevs before we
3619 	 * sync the config cache, and we lose power, then upon reboot we may
3620 	 * fail to open the pool because there are DVAs that the config cache
3621 	 * can't translate.  Therefore, we first add the vdevs without
3622 	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3623 	 * and then let spa_config_update() initialize the new metaslabs.
3624 	 *
3625 	 * spa_load() checks for added-but-not-initialized vdevs, so that
3626 	 * if we lose power at any point in this sequence, the remaining
3627 	 * steps will be completed the next time we load the pool.
3628 	 */
3629 	(void) spa_vdev_exit(spa, vd, txg, 0);
3630 
3631 	mutex_enter(&spa_namespace_lock);
3632 	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3633 	mutex_exit(&spa_namespace_lock);
3634 
3635 	return (0);
3636 }
3637 
3638 /*
3639  * Attach a device to a mirror.  The arguments are the path to any device
3640  * in the mirror, and the nvroot for the new device.  If the path specifies
3641  * a device that is not mirrored, we automatically insert the mirror vdev.
3642  *
3643  * If 'replacing' is specified, the new device is intended to replace the
3644  * existing device; in this case the two devices are made into their own
3645  * mirror using the 'replacing' vdev, which is functionally identical to
3646  * the mirror vdev (it actually reuses all the same ops) but has a few
3647  * extra rules: you can't attach to it after it's been created, and upon
3648  * completion of resilvering, the first disk (the one being replaced)
3649  * is automatically detached.
3650  */
3651 int
3652 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3653 {
3654 	uint64_t txg, dtl_max_txg;
3655 	vdev_t *rvd = spa->spa_root_vdev;
3656 	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3657 	vdev_ops_t *pvops;
3658 	char *oldvdpath, *newvdpath;
3659 	int newvd_isspare;
3660 	int error;
3661 
3662 	txg = spa_vdev_enter(spa);
3663 
3664 	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3665 
3666 	if (oldvd == NULL)
3667 		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3668 
3669 	if (!oldvd->vdev_ops->vdev_op_leaf)
3670 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3671 
3672 	pvd = oldvd->vdev_parent;
3673 
3674 	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3675 	    VDEV_ALLOC_ADD)) != 0)
3676 		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3677 
3678 	if (newrootvd->vdev_children != 1)
3679 		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3680 
3681 	newvd = newrootvd->vdev_child[0];
3682 
3683 	if (!newvd->vdev_ops->vdev_op_leaf)
3684 		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3685 
3686 	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3687 		return (spa_vdev_exit(spa, newrootvd, txg, error));
3688 
3689 	/*
3690 	 * Spares can't replace logs
3691 	 */
3692 	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3693 		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3694 
3695 	if (!replacing) {
3696 		/*
3697 		 * For attach, the only allowable parent is a mirror or the root
3698 		 * vdev.
3699 		 */
3700 		if (pvd->vdev_ops != &vdev_mirror_ops &&
3701 		    pvd->vdev_ops != &vdev_root_ops)
3702 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3703 
3704 		pvops = &vdev_mirror_ops;
3705 	} else {
3706 		/*
3707 		 * Active hot spares can only be replaced by inactive hot
3708 		 * spares.
3709 		 */
3710 		if (pvd->vdev_ops == &vdev_spare_ops &&
3711 		    pvd->vdev_child[1] == oldvd &&
3712 		    !spa_has_spare(spa, newvd->vdev_guid))
3713 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3714 
3715 		/*
3716 		 * If the source is a hot spare, and the parent isn't already a
3717 		 * spare, then we want to create a new hot spare.  Otherwise, we
3718 		 * want to create a replacing vdev.  The user is not allowed to
3719 		 * attach to a spared vdev child unless the 'isspare' state is
3720 		 * the same (spare replaces spare, non-spare replaces
3721 		 * non-spare).
3722 		 */
3723 		if (pvd->vdev_ops == &vdev_replacing_ops)
3724 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3725 		else if (pvd->vdev_ops == &vdev_spare_ops &&
3726 		    newvd->vdev_isspare != oldvd->vdev_isspare)
3727 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3728 		else if (pvd->vdev_ops != &vdev_spare_ops &&
3729 		    newvd->vdev_isspare)
3730 			pvops = &vdev_spare_ops;
3731 		else
3732 			pvops = &vdev_replacing_ops;
3733 	}
3734 
3735 	/*
3736 	 * Make sure the new device is big enough.
3737 	 */
3738 	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3739 		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3740 
3741 	/*
3742 	 * The new device cannot have a higher alignment requirement
3743 	 * than the top-level vdev.
3744 	 */
3745 	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3746 		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3747 
3748 	/*
3749 	 * If this is an in-place replacement, update oldvd's path and devid
3750 	 * to make it distinguishable from newvd, and unopenable from now on.
3751 	 */
3752 	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3753 		spa_strfree(oldvd->vdev_path);
3754 		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3755 		    KM_SLEEP);
3756 		(void) sprintf(oldvd->vdev_path, "%s/%s",
3757 		    newvd->vdev_path, "old");
3758 		if (oldvd->vdev_devid != NULL) {
3759 			spa_strfree(oldvd->vdev_devid);
3760 			oldvd->vdev_devid = NULL;
3761 		}
3762 	}
3763 
3764 	/*
3765 	 * If the parent is not a mirror, or if we're replacing, insert the new
3766 	 * mirror/replacing/spare vdev above oldvd.
3767 	 */
3768 	if (pvd->vdev_ops != pvops)
3769 		pvd = vdev_add_parent(oldvd, pvops);
3770 
3771 	ASSERT(pvd->vdev_top->vdev_parent == rvd);
3772 	ASSERT(pvd->vdev_ops == pvops);
3773 	ASSERT(oldvd->vdev_parent == pvd);
3774 
3775 	/*
3776 	 * Extract the new device from its root and add it to pvd.
3777 	 */
3778 	vdev_remove_child(newrootvd, newvd);
3779 	newvd->vdev_id = pvd->vdev_children;
3780 	newvd->vdev_crtxg = oldvd->vdev_crtxg;
3781 	vdev_add_child(pvd, newvd);
3782 
3783 	tvd = newvd->vdev_top;
3784 	ASSERT(pvd->vdev_top == tvd);
3785 	ASSERT(tvd->vdev_parent == rvd);
3786 
3787 	vdev_config_dirty(tvd);
3788 
3789 	/*
3790 	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
3791 	 * for any dmu_sync-ed blocks.  It will propagate upward when
3792 	 * spa_vdev_exit() calls vdev_dtl_reassess().
3793 	 */
3794 	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
3795 
3796 	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
3797 	    dtl_max_txg - TXG_INITIAL);
3798 
3799 	if (newvd->vdev_isspare) {
3800 		spa_spare_activate(newvd);
3801 		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
3802 	}
3803 
3804 	oldvdpath = spa_strdup(oldvd->vdev_path);
3805 	newvdpath = spa_strdup(newvd->vdev_path);
3806 	newvd_isspare = newvd->vdev_isspare;
3807 
3808 	/*
3809 	 * Mark newvd's DTL dirty in this txg.
3810 	 */
3811 	vdev_dirty(tvd, VDD_DTL, newvd, txg);
3812 
3813 	/*
3814 	 * Restart the resilver
3815 	 */
3816 	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
3817 
3818 	/*
3819 	 * Commit the config
3820 	 */
3821 	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
3822 
3823 	spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
3824 	    "%s vdev=%s %s vdev=%s",
3825 	    replacing && newvd_isspare ? "spare in" :
3826 	    replacing ? "replace" : "attach", newvdpath,
3827 	    replacing ? "for" : "to", oldvdpath);
3828 
3829 	spa_strfree(oldvdpath);
3830 	spa_strfree(newvdpath);
3831 
3832 	return (0);
3833 }
3834 
3835 /*
3836  * Detach a device from a mirror or replacing vdev.
3837  * If 'replace_done' is specified, only detach if the parent
3838  * is a replacing vdev.
3839  */
3840 int
3841 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
3842 {
3843 	uint64_t txg;
3844 	int error;
3845 	vdev_t *rvd = spa->spa_root_vdev;
3846 	vdev_t *vd, *pvd, *cvd, *tvd;
3847 	boolean_t unspare = B_FALSE;
3848 	uint64_t unspare_guid;
3849 	size_t len;
3850 	char *vdpath;
3851 
3852 	txg = spa_vdev_enter(spa);
3853 
3854 	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3855 
3856 	if (vd == NULL)
3857 		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3858 
3859 	if (!vd->vdev_ops->vdev_op_leaf)
3860 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3861 
3862 	pvd = vd->vdev_parent;
3863 
3864 	/*
3865 	 * If the parent/child relationship is not as expected, don't do it.
3866 	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
3867 	 * vdev that's replacing B with C.  The user's intent in replacing
3868 	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
3869 	 * the replace by detaching C, the expected behavior is to end up
3870 	 * M(A,B).  But suppose that right after deciding to detach C,
3871 	 * the replacement of B completes.  We would have M(A,C), and then
3872 	 * ask to detach C, which would leave us with just A -- not what
3873 	 * the user wanted.  To prevent this, we make sure that the
3874 	 * parent/child relationship hasn't changed -- in this example,
3875 	 * that C's parent is still the replacing vdev R.
3876 	 */
3877 	if (pvd->vdev_guid != pguid && pguid != 0)
3878 		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3879 
3880 	/*
3881 	 * If replace_done is specified, only remove this device if it's
3882 	 * the first child of a replacing vdev.  For the 'spare' vdev, either
3883 	 * disk can be removed.
3884 	 */
3885 	if (replace_done) {
3886 		if (pvd->vdev_ops == &vdev_replacing_ops) {
3887 			if (vd->vdev_id != 0)
3888 				return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3889 		} else if (pvd->vdev_ops != &vdev_spare_ops) {
3890 			return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3891 		}
3892 	}
3893 
3894 	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
3895 	    spa_version(spa) >= SPA_VERSION_SPARES);
3896 
3897 	/*
3898 	 * Only mirror, replacing, and spare vdevs support detach.
3899 	 */
3900 	if (pvd->vdev_ops != &vdev_replacing_ops &&
3901 	    pvd->vdev_ops != &vdev_mirror_ops &&
3902 	    pvd->vdev_ops != &vdev_spare_ops)
3903 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3904 
3905 	/*
3906 	 * If this device has the only valid copy of some data,
3907 	 * we cannot safely detach it.
3908 	 */
3909 	if (vdev_dtl_required(vd))
3910 		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3911 
3912 	ASSERT(pvd->vdev_children >= 2);
3913 
3914 	/*
3915 	 * If we are detaching the second disk from a replacing vdev, then
3916 	 * check to see if we changed the original vdev's path to have "/old"
3917 	 * at the end in spa_vdev_attach().  If so, undo that change now.
3918 	 */
3919 	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id == 1 &&
3920 	    pvd->vdev_child[0]->vdev_path != NULL &&
3921 	    pvd->vdev_child[1]->vdev_path != NULL) {
3922 		ASSERT(pvd->vdev_child[1] == vd);
3923 		cvd = pvd->vdev_child[0];
3924 		len = strlen(vd->vdev_path);
3925 		if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
3926 		    strcmp(cvd->vdev_path + len, "/old") == 0) {
3927 			spa_strfree(cvd->vdev_path);
3928 			cvd->vdev_path = spa_strdup(vd->vdev_path);
3929 		}
3930 	}
3931 
3932 	/*
3933 	 * If we are detaching the original disk from a spare, then it implies
3934 	 * that the spare should become a real disk, and be removed from the
3935 	 * active spare list for the pool.
3936 	 */
3937 	if (pvd->vdev_ops == &vdev_spare_ops &&
3938 	    vd->vdev_id == 0 && pvd->vdev_child[1]->vdev_isspare)
3939 		unspare = B_TRUE;
3940 
3941 	/*
3942 	 * Erase the disk labels so the disk can be used for other things.
3943 	 * This must be done after all other error cases are handled,
3944 	 * but before we disembowel vd (so we can still do I/O to it).
3945 	 * But if we can't do it, don't treat the error as fatal --
3946 	 * it may be that the unwritability of the disk is the reason
3947 	 * it's being detached!
3948 	 */
3949 	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
3950 
3951 	/*
3952 	 * Remove vd from its parent and compact the parent's children.
3953 	 */
3954 	vdev_remove_child(pvd, vd);
3955 	vdev_compact_children(pvd);
3956 
3957 	/*
3958 	 * Remember one of the remaining children so we can get tvd below.
3959 	 */
3960 	cvd = pvd->vdev_child[0];
3961 
3962 	/*
3963 	 * If we need to remove the remaining child from the list of hot spares,
3964 	 * do it now, marking the vdev as no longer a spare in the process.
3965 	 * We must do this before vdev_remove_parent(), because that can
3966 	 * change the GUID if it creates a new toplevel GUID.  For a similar
3967 	 * reason, we must remove the spare now, in the same txg as the detach;
3968 	 * otherwise someone could attach a new sibling, change the GUID, and
3969 	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
3970 	 */
3971 	if (unspare) {
3972 		ASSERT(cvd->vdev_isspare);
3973 		spa_spare_remove(cvd);
3974 		unspare_guid = cvd->vdev_guid;
3975 		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
3976 	}
3977 
3978 	/*
3979 	 * If the parent mirror/replacing vdev only has one child,
3980 	 * the parent is no longer needed.  Remove it from the tree.
3981 	 */
3982 	if (pvd->vdev_children == 1)
3983 		vdev_remove_parent(cvd);
3984 
3985 	/*
3986 	 * We don't set tvd until now because the parent we just removed
3987 	 * may have been the previous top-level vdev.
3988 	 */
3989 	tvd = cvd->vdev_top;
3990 	ASSERT(tvd->vdev_parent == rvd);
3991 
3992 	/*
3993 	 * Reevaluate the parent vdev state.
3994 	 */
3995 	vdev_propagate_state(cvd);
3996 
3997 	/*
3998 	 * If the 'autoexpand' property is set on the pool then automatically
3999 	 * try to expand the size of the pool. For example if the device we
4000 	 * just detached was smaller than the others, it may be possible to
4001 	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4002 	 * first so that we can obtain the updated sizes of the leaf vdevs.
4003 	 */
4004 	if (spa->spa_autoexpand) {
4005 		vdev_reopen(tvd);
4006 		vdev_expand(tvd, txg);
4007 	}
4008 
4009 	vdev_config_dirty(tvd);
4010 
4011 	/*
4012 	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4013 	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4014 	 * But first make sure we're not on any *other* txg's DTL list, to
4015 	 * prevent vd from being accessed after it's freed.
4016 	 */
4017 	vdpath = spa_strdup(vd->vdev_path);
4018 	for (int t = 0; t < TXG_SIZE; t++)
4019 		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4020 	vd->vdev_detached = B_TRUE;
4021 	vdev_dirty(tvd, VDD_DTL, vd, txg);
4022 
4023 	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4024 
4025 	error = spa_vdev_exit(spa, vd, txg, 0);
4026 
4027 	spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4028 	    "vdev=%s", vdpath);
4029 	spa_strfree(vdpath);
4030 
4031 	/*
4032 	 * If this was the removal of the original device in a hot spare vdev,
4033 	 * then we want to go through and remove the device from the hot spare
4034 	 * list of every other pool.
4035 	 */
4036 	if (unspare) {
4037 		spa_t *myspa = spa;
4038 		spa = NULL;
4039 		mutex_enter(&spa_namespace_lock);
4040 		while ((spa = spa_next(spa)) != NULL) {
4041 			if (spa->spa_state != POOL_STATE_ACTIVE)
4042 				continue;
4043 			if (spa == myspa)
4044 				continue;
4045 			spa_open_ref(spa, FTAG);
4046 			mutex_exit(&spa_namespace_lock);
4047 			(void) spa_vdev_remove(spa, unspare_guid,
4048 			    B_TRUE);
4049 			mutex_enter(&spa_namespace_lock);
4050 			spa_close(spa, FTAG);
4051 		}
4052 		mutex_exit(&spa_namespace_lock);
4053 	}
4054 
4055 	return (error);
4056 }
4057 
4058 /*
4059  * Split a set of devices from their mirrors, and create a new pool from them.
4060  */
4061 int
4062 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4063     nvlist_t *props, boolean_t exp)
4064 {
4065 	int error = 0;
4066 	uint64_t txg, *glist;
4067 	spa_t *newspa;
4068 	uint_t c, children, lastlog;
4069 	nvlist_t **child, *nvl, *tmp;
4070 	dmu_tx_t *tx;
4071 	char *altroot = NULL;
4072 	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4073 	boolean_t activate_slog;
4074 
4075 	if (!spa_writeable(spa))
4076 		return (EROFS);
4077 
4078 	txg = spa_vdev_enter(spa);
4079 
4080 	/* clear the log and flush everything up to now */
4081 	activate_slog = spa_passivate_log(spa);
4082 	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4083 	error = spa_offline_log(spa);
4084 	txg = spa_vdev_config_enter(spa);
4085 
4086 	if (activate_slog)
4087 		spa_activate_log(spa);
4088 
4089 	if (error != 0)
4090 		return (spa_vdev_exit(spa, NULL, txg, error));
4091 
4092 	/* check new spa name before going any further */
4093 	if (spa_lookup(newname) != NULL)
4094 		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4095 
4096 	/*
4097 	 * scan through all the children to ensure they're all mirrors
4098 	 */
4099 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4100 	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4101 	    &children) != 0)
4102 		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4103 
4104 	/* first, check to ensure we've got the right child count */
4105 	rvd = spa->spa_root_vdev;
4106 	lastlog = 0;
4107 	for (c = 0; c < rvd->vdev_children; c++) {
4108 		vdev_t *vd = rvd->vdev_child[c];
4109 
4110 		/* don't count the holes & logs as children */
4111 		if (vd->vdev_islog || vd->vdev_ishole) {
4112 			if (lastlog == 0)
4113 				lastlog = c;
4114 			continue;
4115 		}
4116 
4117 		lastlog = 0;
4118 	}
4119 	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4120 		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4121 
4122 	/* next, ensure no spare or cache devices are part of the split */
4123 	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4124 	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4125 		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4126 
4127 	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4128 	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4129 
4130 	/* then, loop over each vdev and validate it */
4131 	for (c = 0; c < children; c++) {
4132 		uint64_t is_hole = 0;
4133 
4134 		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4135 		    &is_hole);
4136 
4137 		if (is_hole != 0) {
4138 			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4139 			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4140 				continue;
4141 			} else {
4142 				error = EINVAL;
4143 				break;
4144 			}
4145 		}
4146 
4147 		/* which disk is going to be split? */
4148 		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4149 		    &glist[c]) != 0) {
4150 			error = EINVAL;
4151 			break;
4152 		}
4153 
4154 		/* look it up in the spa */
4155 		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4156 		if (vml[c] == NULL) {
4157 			error = ENODEV;
4158 			break;
4159 		}
4160 
4161 		/* make sure there's nothing stopping the split */
4162 		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4163 		    vml[c]->vdev_islog ||
4164 		    vml[c]->vdev_ishole ||
4165 		    vml[c]->vdev_isspare ||
4166 		    vml[c]->vdev_isl2cache ||
4167 		    !vdev_writeable(vml[c]) ||
4168 		    vml[c]->vdev_children != 0 ||
4169 		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4170 		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4171 			error = EINVAL;
4172 			break;
4173 		}
4174 
4175 		if (vdev_dtl_required(vml[c])) {
4176 			error = EBUSY;
4177 			break;
4178 		}
4179 
4180 		/* we need certain info from the top level */
4181 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4182 		    vml[c]->vdev_top->vdev_ms_array) == 0);
4183 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4184 		    vml[c]->vdev_top->vdev_ms_shift) == 0);
4185 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4186 		    vml[c]->vdev_top->vdev_asize) == 0);
4187 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4188 		    vml[c]->vdev_top->vdev_ashift) == 0);
4189 	}
4190 
4191 	if (error != 0) {
4192 		kmem_free(vml, children * sizeof (vdev_t *));
4193 		kmem_free(glist, children * sizeof (uint64_t));
4194 		return (spa_vdev_exit(spa, NULL, txg, error));
4195 	}
4196 
4197 	/* stop writers from using the disks */
4198 	for (c = 0; c < children; c++) {
4199 		if (vml[c] != NULL)
4200 			vml[c]->vdev_offline = B_TRUE;
4201 	}
4202 	vdev_reopen(spa->spa_root_vdev);
4203 
4204 	/*
4205 	 * Temporarily record the splitting vdevs in the spa config.  This
4206 	 * will disappear once the config is regenerated.
4207 	 */
4208 	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4209 	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4210 	    glist, children) == 0);
4211 	kmem_free(glist, children * sizeof (uint64_t));
4212 
4213 	mutex_enter(&spa->spa_props_lock);
4214 	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4215 	    nvl) == 0);
4216 	mutex_exit(&spa->spa_props_lock);
4217 	spa->spa_config_splitting = nvl;
4218 	vdev_config_dirty(spa->spa_root_vdev);
4219 
4220 	/* configure and create the new pool */
4221 	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4222 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4223 	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4224 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4225 	    spa_version(spa)) == 0);
4226 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4227 	    spa->spa_config_txg) == 0);
4228 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4229 	    spa_generate_guid(NULL)) == 0);
4230 	(void) nvlist_lookup_string(props,
4231 	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4232 
4233 	/* add the new pool to the namespace */
4234 	newspa = spa_add(newname, config, altroot);
4235 	newspa->spa_config_txg = spa->spa_config_txg;
4236 	spa_set_log_state(newspa, SPA_LOG_CLEAR);
4237 
4238 	/* release the spa config lock, retaining the namespace lock */
4239 	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4240 
4241 	if (zio_injection_enabled)
4242 		zio_handle_panic_injection(spa, FTAG, 1);
4243 
4244 	spa_activate(newspa, spa_mode_global);
4245 	spa_async_suspend(newspa);
4246 
4247 	/* create the new pool from the disks of the original pool */
4248 	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4249 	if (error)
4250 		goto out;
4251 
4252 	/* if that worked, generate a real config for the new pool */
4253 	if (newspa->spa_root_vdev != NULL) {
4254 		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4255 		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4256 		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4257 		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4258 		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4259 		    B_TRUE));
4260 	}
4261 
4262 	/* set the props */
4263 	if (props != NULL) {
4264 		spa_configfile_set(newspa, props, B_FALSE);
4265 		error = spa_prop_set(newspa, props);
4266 		if (error)
4267 			goto out;
4268 	}
4269 
4270 	/* flush everything */
4271 	txg = spa_vdev_config_enter(newspa);
4272 	vdev_config_dirty(newspa->spa_root_vdev);
4273 	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4274 
4275 	if (zio_injection_enabled)
4276 		zio_handle_panic_injection(spa, FTAG, 2);
4277 
4278 	spa_async_resume(newspa);
4279 
4280 	/* finally, update the original pool's config */
4281 	txg = spa_vdev_config_enter(spa);
4282 	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4283 	error = dmu_tx_assign(tx, TXG_WAIT);
4284 	if (error != 0)
4285 		dmu_tx_abort(tx);
4286 	for (c = 0; c < children; c++) {
4287 		if (vml[c] != NULL) {
4288 			vdev_split(vml[c]);
4289 			if (error == 0)
4290 				spa_history_log_internal(LOG_POOL_VDEV_DETACH,
4291 				    spa, tx, "vdev=%s",
4292 				    vml[c]->vdev_path);
4293 			vdev_free(vml[c]);
4294 		}
4295 	}
4296 	vdev_config_dirty(spa->spa_root_vdev);
4297 	spa->spa_config_splitting = NULL;
4298 	nvlist_free(nvl);
4299 	if (error == 0)
4300 		dmu_tx_commit(tx);
4301 	(void) spa_vdev_exit(spa, NULL, txg, 0);
4302 
4303 	if (zio_injection_enabled)
4304 		zio_handle_panic_injection(spa, FTAG, 3);
4305 
4306 	/* split is complete; log a history record */
4307 	spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
4308 	    "split new pool %s from pool %s", newname, spa_name(spa));
4309 
4310 	kmem_free(vml, children * sizeof (vdev_t *));
4311 
4312 	/* if we're not going to mount the filesystems in userland, export */
4313 	if (exp)
4314 		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4315 		    B_FALSE, B_FALSE);
4316 
4317 	return (error);
4318 
4319 out:
4320 	spa_unload(newspa);
4321 	spa_deactivate(newspa);
4322 	spa_remove(newspa);
4323 
4324 	txg = spa_vdev_config_enter(spa);
4325 
4326 	/* re-online all offlined disks */
4327 	for (c = 0; c < children; c++) {
4328 		if (vml[c] != NULL)
4329 			vml[c]->vdev_offline = B_FALSE;
4330 	}
4331 	vdev_reopen(spa->spa_root_vdev);
4332 
4333 	nvlist_free(spa->spa_config_splitting);
4334 	spa->spa_config_splitting = NULL;
4335 	(void) spa_vdev_exit(spa, NULL, txg, error);
4336 
4337 	kmem_free(vml, children * sizeof (vdev_t *));
4338 	return (error);
4339 }
4340 
4341 static nvlist_t *
4342 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4343 {
4344 	for (int i = 0; i < count; i++) {
4345 		uint64_t guid;
4346 
4347 		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4348 		    &guid) == 0);
4349 
4350 		if (guid == target_guid)
4351 			return (nvpp[i]);
4352 	}
4353 
4354 	return (NULL);
4355 }
4356 
4357 static void
4358 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4359 	nvlist_t *dev_to_remove)
4360 {
4361 	nvlist_t **newdev = NULL;
4362 
4363 	if (count > 1)
4364 		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4365 
4366 	for (int i = 0, j = 0; i < count; i++) {
4367 		if (dev[i] == dev_to_remove)
4368 			continue;
4369 		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4370 	}
4371 
4372 	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4373 	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4374 
4375 	for (int i = 0; i < count - 1; i++)
4376 		nvlist_free(newdev[i]);
4377 
4378 	if (count > 1)
4379 		kmem_free(newdev, (count - 1) * sizeof (void *));
4380 }
4381 
4382 /*
4383  * Evacuate the device.
4384  */
4385 static int
4386 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4387 {
4388 	uint64_t txg;
4389 	int error = 0;
4390 
4391 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
4392 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4393 	ASSERT(vd == vd->vdev_top);
4394 
4395 	/*
4396 	 * Evacuate the device.  We don't hold the config lock as writer
4397 	 * since we need to do I/O but we do keep the
4398 	 * spa_namespace_lock held.  Once this completes the device
4399 	 * should no longer have any blocks allocated on it.
4400 	 */
4401 	if (vd->vdev_islog) {
4402 		if (vd->vdev_stat.vs_alloc != 0)
4403 			error = spa_offline_log(spa);
4404 	} else {
4405 		error = ENOTSUP;
4406 	}
4407 
4408 	if (error)
4409 		return (error);
4410 
4411 	/*
4412 	 * The evacuation succeeded.  Remove any remaining MOS metadata
4413 	 * associated with this vdev, and wait for these changes to sync.
4414 	 */
4415 	ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
4416 	txg = spa_vdev_config_enter(spa);
4417 	vd->vdev_removing = B_TRUE;
4418 	vdev_dirty(vd, 0, NULL, txg);
4419 	vdev_config_dirty(vd);
4420 	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4421 
4422 	return (0);
4423 }
4424 
4425 /*
4426  * Complete the removal by cleaning up the namespace.
4427  */
4428 static void
4429 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
4430 {
4431 	vdev_t *rvd = spa->spa_root_vdev;
4432 	uint64_t id = vd->vdev_id;
4433 	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
4434 
4435 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
4436 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4437 	ASSERT(vd == vd->vdev_top);
4438 
4439 	/*
4440 	 * Only remove any devices which are empty.
4441 	 */
4442 	if (vd->vdev_stat.vs_alloc != 0)
4443 		return;
4444 
4445 	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4446 
4447 	if (list_link_active(&vd->vdev_state_dirty_node))
4448 		vdev_state_clean(vd);
4449 	if (list_link_active(&vd->vdev_config_dirty_node))
4450 		vdev_config_clean(vd);
4451 
4452 	vdev_free(vd);
4453 
4454 	if (last_vdev) {
4455 		vdev_compact_children(rvd);
4456 	} else {
4457 		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
4458 		vdev_add_child(rvd, vd);
4459 	}
4460 }
4461 
4462 /*
4463  * Remove a device from the pool -
4464  *
4465  * Removing a device from the vdev namespace requires several steps
4466  * and can take a significant amount of time.  As a result we use
4467  * the spa_vdev_config_[enter/exit] functions which allow us to
4468  * grab and release the spa_config_lock while still holding the namespace
4469  * lock.  During each step the configuration is synced out.
4470  */
4471 
4472 /*
4473  * Remove a device from the pool.  Currently, this supports removing only hot
4474  * spares, slogs, and level 2 ARC devices.
4475  */
4476 int
4477 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
4478 {
4479 	vdev_t *vd;
4480 	metaslab_group_t *mg;
4481 	nvlist_t **spares, **l2cache, *nv;
4482 	uint64_t txg = 0;
4483 	uint_t nspares, nl2cache;
4484 	int error = 0;
4485 	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
4486 
4487 	if (!locked)
4488 		txg = spa_vdev_enter(spa);
4489 
4490 	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4491 
4492 	if (spa->spa_spares.sav_vdevs != NULL &&
4493 	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4494 	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
4495 	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
4496 		/*
4497 		 * Only remove the hot spare if it's not currently in use
4498 		 * in this pool.
4499 		 */
4500 		if (vd == NULL || unspare) {
4501 			spa_vdev_remove_aux(spa->spa_spares.sav_config,
4502 			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
4503 			spa_load_spares(spa);
4504 			spa->spa_spares.sav_sync = B_TRUE;
4505 		} else {
4506 			error = EBUSY;
4507 		}
4508 	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
4509 	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4510 	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
4511 	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
4512 		/*
4513 		 * Cache devices can always be removed.
4514 		 */
4515 		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
4516 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
4517 		spa_load_l2cache(spa);
4518 		spa->spa_l2cache.sav_sync = B_TRUE;
4519 	} else if (vd != NULL && vd->vdev_islog) {
4520 		ASSERT(!locked);
4521 		ASSERT(vd == vd->vdev_top);
4522 
4523 		/*
4524 		 * XXX - Once we have bp-rewrite this should
4525 		 * become the common case.
4526 		 */
4527 
4528 		mg = vd->vdev_mg;
4529 
4530 		/*
4531 		 * Stop allocating from this vdev.
4532 		 */
4533 		metaslab_group_passivate(mg);
4534 
4535 		/*
4536 		 * Wait for the youngest allocations and frees to sync,
4537 		 * and then wait for the deferral of those frees to finish.
4538 		 */
4539 		spa_vdev_config_exit(spa, NULL,
4540 		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
4541 
4542 		/*
4543 		 * Attempt to evacuate the vdev.
4544 		 */
4545 		error = spa_vdev_remove_evacuate(spa, vd);
4546 
4547 		txg = spa_vdev_config_enter(spa);
4548 
4549 		/*
4550 		 * If we couldn't evacuate the vdev, unwind.
4551 		 */
4552 		if (error) {
4553 			metaslab_group_activate(mg);
4554 			return (spa_vdev_exit(spa, NULL, txg, error));
4555 		}
4556 
4557 		/*
4558 		 * Clean up the vdev namespace.
4559 		 */
4560 		spa_vdev_remove_from_namespace(spa, vd);
4561 
4562 	} else if (vd != NULL) {
4563 		/*
4564 		 * Normal vdevs cannot be removed (yet).
4565 		 */
4566 		error = ENOTSUP;
4567 	} else {
4568 		/*
4569 		 * There is no vdev of any kind with the specified guid.
4570 		 */
4571 		error = ENOENT;
4572 	}
4573 
4574 	if (!locked)
4575 		return (spa_vdev_exit(spa, NULL, txg, error));
4576 
4577 	return (error);
4578 }
4579 
4580 /*
4581  * Find any device that's done replacing, or a vdev marked 'unspare' that's
4582  * current spared, so we can detach it.
4583  */
4584 static vdev_t *
4585 spa_vdev_resilver_done_hunt(vdev_t *vd)
4586 {
4587 	vdev_t *newvd, *oldvd;
4588 
4589 	for (int c = 0; c < vd->vdev_children; c++) {
4590 		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
4591 		if (oldvd != NULL)
4592 			return (oldvd);
4593 	}
4594 
4595 	/*
4596 	 * Check for a completed replacement.
4597 	 */
4598 	if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
4599 		oldvd = vd->vdev_child[0];
4600 		newvd = vd->vdev_child[1];
4601 
4602 		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4603 		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4604 		    !vdev_dtl_required(oldvd))
4605 			return (oldvd);
4606 	}
4607 
4608 	/*
4609 	 * Check for a completed resilver with the 'unspare' flag set.
4610 	 */
4611 	if (vd->vdev_ops == &vdev_spare_ops && vd->vdev_children == 2) {
4612 		newvd = vd->vdev_child[0];
4613 		oldvd = vd->vdev_child[1];
4614 
4615 		if (newvd->vdev_unspare &&
4616 		    vdev_dtl_empty(newvd, DTL_MISSING) &&
4617 		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4618 		    !vdev_dtl_required(oldvd)) {
4619 			newvd->vdev_unspare = 0;
4620 			return (oldvd);
4621 		}
4622 	}
4623 
4624 	return (NULL);
4625 }
4626 
4627 static void
4628 spa_vdev_resilver_done(spa_t *spa)
4629 {
4630 	vdev_t *vd, *pvd, *ppvd;
4631 	uint64_t guid, sguid, pguid, ppguid;
4632 
4633 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4634 
4635 	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4636 		pvd = vd->vdev_parent;
4637 		ppvd = pvd->vdev_parent;
4638 		guid = vd->vdev_guid;
4639 		pguid = pvd->vdev_guid;
4640 		ppguid = ppvd->vdev_guid;
4641 		sguid = 0;
4642 		/*
4643 		 * If we have just finished replacing a hot spared device, then
4644 		 * we need to detach the parent's first child (the original hot
4645 		 * spare) as well.
4646 		 */
4647 		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0) {
4648 			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4649 			ASSERT(ppvd->vdev_children == 2);
4650 			sguid = ppvd->vdev_child[1]->vdev_guid;
4651 		}
4652 		spa_config_exit(spa, SCL_ALL, FTAG);
4653 		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4654 			return;
4655 		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4656 			return;
4657 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4658 	}
4659 
4660 	spa_config_exit(spa, SCL_ALL, FTAG);
4661 }
4662 
4663 /*
4664  * Update the stored path or FRU for this vdev.
4665  */
4666 int
4667 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4668     boolean_t ispath)
4669 {
4670 	vdev_t *vd;
4671 	boolean_t sync = B_FALSE;
4672 
4673 	spa_vdev_state_enter(spa, SCL_ALL);
4674 
4675 	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4676 		return (spa_vdev_state_exit(spa, NULL, ENOENT));
4677 
4678 	if (!vd->vdev_ops->vdev_op_leaf)
4679 		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
4680 
4681 	if (ispath) {
4682 		if (strcmp(value, vd->vdev_path) != 0) {
4683 			spa_strfree(vd->vdev_path);
4684 			vd->vdev_path = spa_strdup(value);
4685 			sync = B_TRUE;
4686 		}
4687 	} else {
4688 		if (vd->vdev_fru == NULL) {
4689 			vd->vdev_fru = spa_strdup(value);
4690 			sync = B_TRUE;
4691 		} else if (strcmp(value, vd->vdev_fru) != 0) {
4692 			spa_strfree(vd->vdev_fru);
4693 			vd->vdev_fru = spa_strdup(value);
4694 			sync = B_TRUE;
4695 		}
4696 	}
4697 
4698 	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
4699 }
4700 
4701 int
4702 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
4703 {
4704 	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
4705 }
4706 
4707 int
4708 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
4709 {
4710 	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
4711 }
4712 
4713 /*
4714  * ==========================================================================
4715  * SPA Scanning
4716  * ==========================================================================
4717  */
4718 
4719 int
4720 spa_scan_stop(spa_t *spa)
4721 {
4722 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4723 	if (dsl_scan_resilvering(spa->spa_dsl_pool))
4724 		return (EBUSY);
4725 	return (dsl_scan_cancel(spa->spa_dsl_pool));
4726 }
4727 
4728 int
4729 spa_scan(spa_t *spa, pool_scan_func_t func)
4730 {
4731 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4732 
4733 	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
4734 		return (ENOTSUP);
4735 
4736 	/*
4737 	 * If a resilver was requested, but there is no DTL on a
4738 	 * writeable leaf device, we have nothing to do.
4739 	 */
4740 	if (func == POOL_SCAN_RESILVER &&
4741 	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
4742 		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
4743 		return (0);
4744 	}
4745 
4746 	return (dsl_scan(spa->spa_dsl_pool, func));
4747 }
4748 
4749 /*
4750  * ==========================================================================
4751  * SPA async task processing
4752  * ==========================================================================
4753  */
4754 
4755 static void
4756 spa_async_remove(spa_t *spa, vdev_t *vd)
4757 {
4758 	if (vd->vdev_remove_wanted) {
4759 		vd->vdev_remove_wanted = B_FALSE;
4760 		vd->vdev_delayed_close = B_FALSE;
4761 		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
4762 
4763 		/*
4764 		 * We want to clear the stats, but we don't want to do a full
4765 		 * vdev_clear() as that will cause us to throw away
4766 		 * degraded/faulted state as well as attempt to reopen the
4767 		 * device, all of which is a waste.
4768 		 */
4769 		vd->vdev_stat.vs_read_errors = 0;
4770 		vd->vdev_stat.vs_write_errors = 0;
4771 		vd->vdev_stat.vs_checksum_errors = 0;
4772 
4773 		vdev_state_dirty(vd->vdev_top);
4774 	}
4775 
4776 	for (int c = 0; c < vd->vdev_children; c++)
4777 		spa_async_remove(spa, vd->vdev_child[c]);
4778 }
4779 
4780 static void
4781 spa_async_probe(spa_t *spa, vdev_t *vd)
4782 {
4783 	if (vd->vdev_probe_wanted) {
4784 		vd->vdev_probe_wanted = B_FALSE;
4785 		vdev_reopen(vd);	/* vdev_open() does the actual probe */
4786 	}
4787 
4788 	for (int c = 0; c < vd->vdev_children; c++)
4789 		spa_async_probe(spa, vd->vdev_child[c]);
4790 }
4791 
4792 static void
4793 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
4794 {
4795 	sysevent_id_t eid;
4796 	nvlist_t *attr;
4797 	char *physpath;
4798 
4799 	if (!spa->spa_autoexpand)
4800 		return;
4801 
4802 	for (int c = 0; c < vd->vdev_children; c++) {
4803 		vdev_t *cvd = vd->vdev_child[c];
4804 		spa_async_autoexpand(spa, cvd);
4805 	}
4806 
4807 	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
4808 		return;
4809 
4810 	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
4811 	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
4812 
4813 	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4814 	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
4815 
4816 	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
4817 	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
4818 
4819 	nvlist_free(attr);
4820 	kmem_free(physpath, MAXPATHLEN);
4821 }
4822 
4823 static void
4824 spa_async_thread(spa_t *spa)
4825 {
4826 	int tasks;
4827 
4828 	ASSERT(spa->spa_sync_on);
4829 
4830 	mutex_enter(&spa->spa_async_lock);
4831 	tasks = spa->spa_async_tasks;
4832 	spa->spa_async_tasks = 0;
4833 	mutex_exit(&spa->spa_async_lock);
4834 
4835 	/*
4836 	 * See if the config needs to be updated.
4837 	 */
4838 	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
4839 		uint64_t old_space, new_space;
4840 
4841 		mutex_enter(&spa_namespace_lock);
4842 		old_space = metaslab_class_get_space(spa_normal_class(spa));
4843 		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4844 		new_space = metaslab_class_get_space(spa_normal_class(spa));
4845 		mutex_exit(&spa_namespace_lock);
4846 
4847 		/*
4848 		 * If the pool grew as a result of the config update,
4849 		 * then log an internal history event.
4850 		 */
4851 		if (new_space != old_space) {
4852 			spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
4853 			    spa, NULL,
4854 			    "pool '%s' size: %llu(+%llu)",
4855 			    spa_name(spa), new_space, new_space - old_space);
4856 		}
4857 	}
4858 
4859 	/*
4860 	 * See if any devices need to be marked REMOVED.
4861 	 */
4862 	if (tasks & SPA_ASYNC_REMOVE) {
4863 		spa_vdev_state_enter(spa, SCL_NONE);
4864 		spa_async_remove(spa, spa->spa_root_vdev);
4865 		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
4866 			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
4867 		for (int i = 0; i < spa->spa_spares.sav_count; i++)
4868 			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
4869 		(void) spa_vdev_state_exit(spa, NULL, 0);
4870 	}
4871 
4872 	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
4873 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4874 		spa_async_autoexpand(spa, spa->spa_root_vdev);
4875 		spa_config_exit(spa, SCL_CONFIG, FTAG);
4876 	}
4877 
4878 	/*
4879 	 * See if any devices need to be probed.
4880 	 */
4881 	if (tasks & SPA_ASYNC_PROBE) {
4882 		spa_vdev_state_enter(spa, SCL_NONE);
4883 		spa_async_probe(spa, spa->spa_root_vdev);
4884 		(void) spa_vdev_state_exit(spa, NULL, 0);
4885 	}
4886 
4887 	/*
4888 	 * If any devices are done replacing, detach them.
4889 	 */
4890 	if (tasks & SPA_ASYNC_RESILVER_DONE)
4891 		spa_vdev_resilver_done(spa);
4892 
4893 	/*
4894 	 * Kick off a resilver.
4895 	 */
4896 	if (tasks & SPA_ASYNC_RESILVER)
4897 		dsl_resilver_restart(spa->spa_dsl_pool, 0);
4898 
4899 	/*
4900 	 * Let the world know that we're done.
4901 	 */
4902 	mutex_enter(&spa->spa_async_lock);
4903 	spa->spa_async_thread = NULL;
4904 	cv_broadcast(&spa->spa_async_cv);
4905 	mutex_exit(&spa->spa_async_lock);
4906 	thread_exit();
4907 }
4908 
4909 void
4910 spa_async_suspend(spa_t *spa)
4911 {
4912 	mutex_enter(&spa->spa_async_lock);
4913 	spa->spa_async_suspended++;
4914 	while (spa->spa_async_thread != NULL)
4915 		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
4916 	mutex_exit(&spa->spa_async_lock);
4917 }
4918 
4919 void
4920 spa_async_resume(spa_t *spa)
4921 {
4922 	mutex_enter(&spa->spa_async_lock);
4923 	ASSERT(spa->spa_async_suspended != 0);
4924 	spa->spa_async_suspended--;
4925 	mutex_exit(&spa->spa_async_lock);
4926 }
4927 
4928 static void
4929 spa_async_dispatch(spa_t *spa)
4930 {
4931 	mutex_enter(&spa->spa_async_lock);
4932 	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
4933 	    spa->spa_async_thread == NULL &&
4934 	    rootdir != NULL && !vn_is_readonly(rootdir))
4935 		spa->spa_async_thread = thread_create(NULL, 0,
4936 		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
4937 	mutex_exit(&spa->spa_async_lock);
4938 }
4939 
4940 void
4941 spa_async_request(spa_t *spa, int task)
4942 {
4943 	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
4944 	mutex_enter(&spa->spa_async_lock);
4945 	spa->spa_async_tasks |= task;
4946 	mutex_exit(&spa->spa_async_lock);
4947 }
4948 
4949 /*
4950  * ==========================================================================
4951  * SPA syncing routines
4952  * ==========================================================================
4953  */
4954 static void
4955 spa_sync_deferred_bplist(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx, uint64_t txg)
4956 {
4957 	blkptr_t blk;
4958 	uint64_t itor = 0;
4959 	uint8_t c = 1;
4960 
4961 	while (bplist_iterate(bpl, &itor, &blk) == 0) {
4962 		ASSERT(blk.blk_birth < txg);
4963 		zio_free(spa, txg, &blk);
4964 	}
4965 
4966 	bplist_vacate(bpl, tx);
4967 
4968 	/*
4969 	 * Pre-dirty the first block so we sync to convergence faster.
4970 	 * (Usually only the first block is needed.)
4971 	 */
4972 	dmu_write(bpl->bpl_mos, spa->spa_deferred_bplist_obj, 0, 1, &c, tx);
4973 }
4974 
4975 static void
4976 spa_sync_free(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
4977 {
4978 	zio_t *zio = arg;
4979 
4980 	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
4981 	    zio->io_flags));
4982 }
4983 
4984 static void
4985 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
4986 {
4987 	char *packed = NULL;
4988 	size_t bufsize;
4989 	size_t nvsize = 0;
4990 	dmu_buf_t *db;
4991 
4992 	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
4993 
4994 	/*
4995 	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
4996 	 * information.  This avoids the dbuf_will_dirty() path and
4997 	 * saves us a pre-read to get data we don't actually care about.
4998 	 */
4999 	bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
5000 	packed = kmem_alloc(bufsize, KM_SLEEP);
5001 
5002 	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5003 	    KM_SLEEP) == 0);
5004 	bzero(packed + nvsize, bufsize - nvsize);
5005 
5006 	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5007 
5008 	kmem_free(packed, bufsize);
5009 
5010 	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5011 	dmu_buf_will_dirty(db, tx);
5012 	*(uint64_t *)db->db_data = nvsize;
5013 	dmu_buf_rele(db, FTAG);
5014 }
5015 
5016 static void
5017 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5018     const char *config, const char *entry)
5019 {
5020 	nvlist_t *nvroot;
5021 	nvlist_t **list;
5022 	int i;
5023 
5024 	if (!sav->sav_sync)
5025 		return;
5026 
5027 	/*
5028 	 * Update the MOS nvlist describing the list of available devices.
5029 	 * spa_validate_aux() will have already made sure this nvlist is
5030 	 * valid and the vdevs are labeled appropriately.
5031 	 */
5032 	if (sav->sav_object == 0) {
5033 		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5034 		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5035 		    sizeof (uint64_t), tx);
5036 		VERIFY(zap_update(spa->spa_meta_objset,
5037 		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5038 		    &sav->sav_object, tx) == 0);
5039 	}
5040 
5041 	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5042 	if (sav->sav_count == 0) {
5043 		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5044 	} else {
5045 		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5046 		for (i = 0; i < sav->sav_count; i++)
5047 			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5048 			    B_FALSE, VDEV_CONFIG_L2CACHE);
5049 		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5050 		    sav->sav_count) == 0);
5051 		for (i = 0; i < sav->sav_count; i++)
5052 			nvlist_free(list[i]);
5053 		kmem_free(list, sav->sav_count * sizeof (void *));
5054 	}
5055 
5056 	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5057 	nvlist_free(nvroot);
5058 
5059 	sav->sav_sync = B_FALSE;
5060 }
5061 
5062 static void
5063 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5064 {
5065 	nvlist_t *config;
5066 
5067 	if (list_is_empty(&spa->spa_config_dirty_list))
5068 		return;
5069 
5070 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5071 
5072 	config = spa_config_generate(spa, spa->spa_root_vdev,
5073 	    dmu_tx_get_txg(tx), B_FALSE);
5074 
5075 	spa_config_exit(spa, SCL_STATE, FTAG);
5076 
5077 	if (spa->spa_config_syncing)
5078 		nvlist_free(spa->spa_config_syncing);
5079 	spa->spa_config_syncing = config;
5080 
5081 	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5082 }
5083 
5084 /*
5085  * Set zpool properties.
5086  */
5087 static void
5088 spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5089 {
5090 	spa_t *spa = arg1;
5091 	objset_t *mos = spa->spa_meta_objset;
5092 	nvlist_t *nvp = arg2;
5093 	nvpair_t *elem;
5094 	uint64_t intval;
5095 	char *strval;
5096 	zpool_prop_t prop;
5097 	const char *propname;
5098 	zprop_type_t proptype;
5099 
5100 	mutex_enter(&spa->spa_props_lock);
5101 
5102 	elem = NULL;
5103 	while ((elem = nvlist_next_nvpair(nvp, elem))) {
5104 		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5105 		case ZPOOL_PROP_VERSION:
5106 			/*
5107 			 * Only set version for non-zpool-creation cases
5108 			 * (set/import). spa_create() needs special care
5109 			 * for version setting.
5110 			 */
5111 			if (tx->tx_txg != TXG_INITIAL) {
5112 				VERIFY(nvpair_value_uint64(elem,
5113 				    &intval) == 0);
5114 				ASSERT(intval <= SPA_VERSION);
5115 				ASSERT(intval >= spa_version(spa));
5116 				spa->spa_uberblock.ub_version = intval;
5117 				vdev_config_dirty(spa->spa_root_vdev);
5118 			}
5119 			break;
5120 
5121 		case ZPOOL_PROP_ALTROOT:
5122 			/*
5123 			 * 'altroot' is a non-persistent property. It should
5124 			 * have been set temporarily at creation or import time.
5125 			 */
5126 			ASSERT(spa->spa_root != NULL);
5127 			break;
5128 
5129 		case ZPOOL_PROP_CACHEFILE:
5130 			/*
5131 			 * 'cachefile' is also a non-persisitent property.
5132 			 */
5133 			break;
5134 		default:
5135 			/*
5136 			 * Set pool property values in the poolprops mos object.
5137 			 */
5138 			if (spa->spa_pool_props_object == 0) {
5139 				VERIFY((spa->spa_pool_props_object =
5140 				    zap_create(mos, DMU_OT_POOL_PROPS,
5141 				    DMU_OT_NONE, 0, tx)) > 0);
5142 
5143 				VERIFY(zap_update(mos,
5144 				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5145 				    8, 1, &spa->spa_pool_props_object, tx)
5146 				    == 0);
5147 			}
5148 
5149 			/* normalize the property name */
5150 			propname = zpool_prop_to_name(prop);
5151 			proptype = zpool_prop_get_type(prop);
5152 
5153 			if (nvpair_type(elem) == DATA_TYPE_STRING) {
5154 				ASSERT(proptype == PROP_TYPE_STRING);
5155 				VERIFY(nvpair_value_string(elem, &strval) == 0);
5156 				VERIFY(zap_update(mos,
5157 				    spa->spa_pool_props_object, propname,
5158 				    1, strlen(strval) + 1, strval, tx) == 0);
5159 
5160 			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5161 				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5162 
5163 				if (proptype == PROP_TYPE_INDEX) {
5164 					const char *unused;
5165 					VERIFY(zpool_prop_index_to_string(
5166 					    prop, intval, &unused) == 0);
5167 				}
5168 				VERIFY(zap_update(mos,
5169 				    spa->spa_pool_props_object, propname,
5170 				    8, 1, &intval, tx) == 0);
5171 			} else {
5172 				ASSERT(0); /* not allowed */
5173 			}
5174 
5175 			switch (prop) {
5176 			case ZPOOL_PROP_DELEGATION:
5177 				spa->spa_delegation = intval;
5178 				break;
5179 			case ZPOOL_PROP_BOOTFS:
5180 				spa->spa_bootfs = intval;
5181 				break;
5182 			case ZPOOL_PROP_FAILUREMODE:
5183 				spa->spa_failmode = intval;
5184 				break;
5185 			case ZPOOL_PROP_AUTOEXPAND:
5186 				spa->spa_autoexpand = intval;
5187 				spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
5188 				break;
5189 			case ZPOOL_PROP_DEDUPDITTO:
5190 				spa->spa_dedup_ditto = intval;
5191 				break;
5192 			default:
5193 				break;
5194 			}
5195 		}
5196 
5197 		/* log internal history if this is not a zpool create */
5198 		if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5199 		    tx->tx_txg != TXG_INITIAL) {
5200 			spa_history_log_internal(LOG_POOL_PROPSET,
5201 			    spa, tx, "%s %lld %s",
5202 			    nvpair_name(elem), intval, spa_name(spa));
5203 		}
5204 	}
5205 
5206 	mutex_exit(&spa->spa_props_lock);
5207 }
5208 
5209 /*
5210  * Sync the specified transaction group.  New blocks may be dirtied as
5211  * part of the process, so we iterate until it converges.
5212  */
5213 void
5214 spa_sync(spa_t *spa, uint64_t txg)
5215 {
5216 	dsl_pool_t *dp = spa->spa_dsl_pool;
5217 	objset_t *mos = spa->spa_meta_objset;
5218 	bplist_t *defer_bpl = &spa->spa_deferred_bplist;
5219 	bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5220 	vdev_t *rvd = spa->spa_root_vdev;
5221 	vdev_t *vd;
5222 	dmu_tx_t *tx;
5223 	int error;
5224 
5225 	/*
5226 	 * Lock out configuration changes.
5227 	 */
5228 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5229 
5230 	spa->spa_syncing_txg = txg;
5231 	spa->spa_sync_pass = 0;
5232 
5233 	/*
5234 	 * If there are any pending vdev state changes, convert them
5235 	 * into config changes that go out with this transaction group.
5236 	 */
5237 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5238 	while (list_head(&spa->spa_state_dirty_list) != NULL) {
5239 		/*
5240 		 * We need the write lock here because, for aux vdevs,
5241 		 * calling vdev_config_dirty() modifies sav_config.
5242 		 * This is ugly and will become unnecessary when we
5243 		 * eliminate the aux vdev wart by integrating all vdevs
5244 		 * into the root vdev tree.
5245 		 */
5246 		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5247 		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5248 		while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5249 			vdev_state_clean(vd);
5250 			vdev_config_dirty(vd);
5251 		}
5252 		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5253 		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5254 	}
5255 	spa_config_exit(spa, SCL_STATE, FTAG);
5256 
5257 	VERIFY(0 == bplist_open(defer_bpl, mos, spa->spa_deferred_bplist_obj));
5258 
5259 	tx = dmu_tx_create_assigned(dp, txg);
5260 
5261 	/*
5262 	 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5263 	 * set spa_deflate if we have no raid-z vdevs.
5264 	 */
5265 	if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5266 	    spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5267 		int i;
5268 
5269 		for (i = 0; i < rvd->vdev_children; i++) {
5270 			vd = rvd->vdev_child[i];
5271 			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5272 				break;
5273 		}
5274 		if (i == rvd->vdev_children) {
5275 			spa->spa_deflate = TRUE;
5276 			VERIFY(0 == zap_add(spa->spa_meta_objset,
5277 			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5278 			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5279 		}
5280 	}
5281 
5282 	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5283 	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5284 		dsl_pool_create_origin(dp, tx);
5285 
5286 		/* Keeping the origin open increases spa_minref */
5287 		spa->spa_minref += 3;
5288 	}
5289 
5290 	if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5291 	    spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5292 		dsl_pool_upgrade_clones(dp, tx);
5293 	}
5294 
5295 	/*
5296 	 * If anything has changed in this txg, or if someone is waiting
5297 	 * for this txg to sync (eg, spa_vdev_remove()), push the
5298 	 * deferred frees from the previous txg.  If not, leave them
5299 	 * alone so that we don't generate work on an otherwise idle
5300 	 * system.
5301 	 */
5302 	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5303 	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5304 	    !txg_list_empty(&dp->dp_sync_tasks, txg) ||
5305 	    ((dp->dp_scan->scn_phys.scn_state == DSS_SCANNING ||
5306 	    txg_sync_waiting(dp)) && !spa_shutting_down(spa)))
5307 		spa_sync_deferred_bplist(spa, defer_bpl, tx, txg);
5308 
5309 	/*
5310 	 * Iterate to convergence.
5311 	 */
5312 	do {
5313 		int pass = ++spa->spa_sync_pass;
5314 
5315 		spa_sync_config_object(spa, tx);
5316 		spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5317 		    ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5318 		spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5319 		    ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5320 		spa_errlog_sync(spa, txg);
5321 		dsl_pool_sync(dp, txg);
5322 
5323 		if (pass <= SYNC_PASS_DEFERRED_FREE) {
5324 			zio_t *zio = zio_root(spa, NULL, NULL, 0);
5325 			bplist_sync(free_bpl, spa_sync_free, zio, tx);
5326 			VERIFY(zio_wait(zio) == 0);
5327 		} else {
5328 			bplist_sync(free_bpl, bplist_enqueue_cb, defer_bpl, tx);
5329 		}
5330 
5331 		ddt_sync(spa, txg);
5332 		dsl_scan_sync(dp, tx);
5333 
5334 		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
5335 			vdev_sync(vd, txg);
5336 
5337 	} while (dmu_objset_is_dirty(mos, txg));
5338 
5339 	ASSERT(list_is_empty(&free_bpl->bpl_queue));
5340 
5341 	bplist_close(defer_bpl);
5342 
5343 	/*
5344 	 * Rewrite the vdev configuration (which includes the uberblock)
5345 	 * to commit the transaction group.
5346 	 *
5347 	 * If there are no dirty vdevs, we sync the uberblock to a few
5348 	 * random top-level vdevs that are known to be visible in the
5349 	 * config cache (see spa_vdev_add() for a complete description).
5350 	 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
5351 	 */
5352 	for (;;) {
5353 		/*
5354 		 * We hold SCL_STATE to prevent vdev open/close/etc.
5355 		 * while we're attempting to write the vdev labels.
5356 		 */
5357 		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5358 
5359 		if (list_is_empty(&spa->spa_config_dirty_list)) {
5360 			vdev_t *svd[SPA_DVAS_PER_BP];
5361 			int svdcount = 0;
5362 			int children = rvd->vdev_children;
5363 			int c0 = spa_get_random(children);
5364 
5365 			for (int c = 0; c < children; c++) {
5366 				vd = rvd->vdev_child[(c0 + c) % children];
5367 				if (vd->vdev_ms_array == 0 || vd->vdev_islog)
5368 					continue;
5369 				svd[svdcount++] = vd;
5370 				if (svdcount == SPA_DVAS_PER_BP)
5371 					break;
5372 			}
5373 			error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
5374 			if (error != 0)
5375 				error = vdev_config_sync(svd, svdcount, txg,
5376 				    B_TRUE);
5377 		} else {
5378 			error = vdev_config_sync(rvd->vdev_child,
5379 			    rvd->vdev_children, txg, B_FALSE);
5380 			if (error != 0)
5381 				error = vdev_config_sync(rvd->vdev_child,
5382 				    rvd->vdev_children, txg, B_TRUE);
5383 		}
5384 
5385 		spa_config_exit(spa, SCL_STATE, FTAG);
5386 
5387 		if (error == 0)
5388 			break;
5389 		zio_suspend(spa, NULL);
5390 		zio_resume_wait(spa);
5391 	}
5392 	dmu_tx_commit(tx);
5393 
5394 	/*
5395 	 * Clear the dirty config list.
5396 	 */
5397 	while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
5398 		vdev_config_clean(vd);
5399 
5400 	/*
5401 	 * Now that the new config has synced transactionally,
5402 	 * let it become visible to the config cache.
5403 	 */
5404 	if (spa->spa_config_syncing != NULL) {
5405 		spa_config_set(spa, spa->spa_config_syncing);
5406 		spa->spa_config_txg = txg;
5407 		spa->spa_config_syncing = NULL;
5408 	}
5409 
5410 	spa->spa_ubsync = spa->spa_uberblock;
5411 
5412 	dsl_pool_sync_done(dp, txg);
5413 
5414 	/*
5415 	 * Update usable space statistics.
5416 	 */
5417 	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
5418 		vdev_sync_done(vd, txg);
5419 
5420 	spa_update_dspace(spa);
5421 
5422 	/*
5423 	 * It had better be the case that we didn't dirty anything
5424 	 * since vdev_config_sync().
5425 	 */
5426 	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
5427 	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
5428 	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
5429 	ASSERT(list_is_empty(&defer_bpl->bpl_queue));
5430 	ASSERT(list_is_empty(&free_bpl->bpl_queue));
5431 
5432 	spa->spa_sync_pass = 0;
5433 
5434 	spa_config_exit(spa, SCL_CONFIG, FTAG);
5435 
5436 	spa_handle_ignored_writes(spa);
5437 
5438 	/*
5439 	 * If any async tasks have been requested, kick them off.
5440 	 */
5441 	spa_async_dispatch(spa);
5442 }
5443 
5444 /*
5445  * Sync all pools.  We don't want to hold the namespace lock across these
5446  * operations, so we take a reference on the spa_t and drop the lock during the
5447  * sync.
5448  */
5449 void
5450 spa_sync_allpools(void)
5451 {
5452 	spa_t *spa = NULL;
5453 	mutex_enter(&spa_namespace_lock);
5454 	while ((spa = spa_next(spa)) != NULL) {
5455 		if (spa_state(spa) != POOL_STATE_ACTIVE || spa_suspended(spa))
5456 			continue;
5457 		spa_open_ref(spa, FTAG);
5458 		mutex_exit(&spa_namespace_lock);
5459 		txg_wait_synced(spa_get_dsl(spa), 0);
5460 		mutex_enter(&spa_namespace_lock);
5461 		spa_close(spa, FTAG);
5462 	}
5463 	mutex_exit(&spa_namespace_lock);
5464 }
5465 
5466 /*
5467  * ==========================================================================
5468  * Miscellaneous routines
5469  * ==========================================================================
5470  */
5471 
5472 /*
5473  * Remove all pools in the system.
5474  */
5475 void
5476 spa_evict_all(void)
5477 {
5478 	spa_t *spa;
5479 
5480 	/*
5481 	 * Remove all cached state.  All pools should be closed now,
5482 	 * so every spa in the AVL tree should be unreferenced.
5483 	 */
5484 	mutex_enter(&spa_namespace_lock);
5485 	while ((spa = spa_next(NULL)) != NULL) {
5486 		/*
5487 		 * Stop async tasks.  The async thread may need to detach
5488 		 * a device that's been replaced, which requires grabbing
5489 		 * spa_namespace_lock, so we must drop it here.
5490 		 */
5491 		spa_open_ref(spa, FTAG);
5492 		mutex_exit(&spa_namespace_lock);
5493 		spa_async_suspend(spa);
5494 		mutex_enter(&spa_namespace_lock);
5495 		spa_close(spa, FTAG);
5496 
5497 		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5498 			spa_unload(spa);
5499 			spa_deactivate(spa);
5500 		}
5501 		spa_remove(spa);
5502 	}
5503 	mutex_exit(&spa_namespace_lock);
5504 }
5505 
5506 vdev_t *
5507 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
5508 {
5509 	vdev_t *vd;
5510 	int i;
5511 
5512 	if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
5513 		return (vd);
5514 
5515 	if (aux) {
5516 		for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
5517 			vd = spa->spa_l2cache.sav_vdevs[i];
5518 			if (vd->vdev_guid == guid)
5519 				return (vd);
5520 		}
5521 
5522 		for (i = 0; i < spa->spa_spares.sav_count; i++) {
5523 			vd = spa->spa_spares.sav_vdevs[i];
5524 			if (vd->vdev_guid == guid)
5525 				return (vd);
5526 		}
5527 	}
5528 
5529 	return (NULL);
5530 }
5531 
5532 void
5533 spa_upgrade(spa_t *spa, uint64_t version)
5534 {
5535 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5536 
5537 	/*
5538 	 * This should only be called for a non-faulted pool, and since a
5539 	 * future version would result in an unopenable pool, this shouldn't be
5540 	 * possible.
5541 	 */
5542 	ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
5543 	ASSERT(version >= spa->spa_uberblock.ub_version);
5544 
5545 	spa->spa_uberblock.ub_version = version;
5546 	vdev_config_dirty(spa->spa_root_vdev);
5547 
5548 	spa_config_exit(spa, SCL_ALL, FTAG);
5549 
5550 	txg_wait_synced(spa_get_dsl(spa), 0);
5551 }
5552 
5553 boolean_t
5554 spa_has_spare(spa_t *spa, uint64_t guid)
5555 {
5556 	int i;
5557 	uint64_t spareguid;
5558 	spa_aux_vdev_t *sav = &spa->spa_spares;
5559 
5560 	for (i = 0; i < sav->sav_count; i++)
5561 		if (sav->sav_vdevs[i]->vdev_guid == guid)
5562 			return (B_TRUE);
5563 
5564 	for (i = 0; i < sav->sav_npending; i++) {
5565 		if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
5566 		    &spareguid) == 0 && spareguid == guid)
5567 			return (B_TRUE);
5568 	}
5569 
5570 	return (B_FALSE);
5571 }
5572 
5573 /*
5574  * Check if a pool has an active shared spare device.
5575  * Note: reference count of an active spare is 2, as a spare and as a replace
5576  */
5577 static boolean_t
5578 spa_has_active_shared_spare(spa_t *spa)
5579 {
5580 	int i, refcnt;
5581 	uint64_t pool;
5582 	spa_aux_vdev_t *sav = &spa->spa_spares;
5583 
5584 	for (i = 0; i < sav->sav_count; i++) {
5585 		if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
5586 		    &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
5587 		    refcnt > 2)
5588 			return (B_TRUE);
5589 	}
5590 
5591 	return (B_FALSE);
5592 }
5593 
5594 /*
5595  * Post a sysevent corresponding to the given event.  The 'name' must be one of
5596  * the event definitions in sys/sysevent/eventdefs.h.  The payload will be
5597  * filled in from the spa and (optionally) the vdev.  This doesn't do anything
5598  * in the userland libzpool, as we don't want consumers to misinterpret ztest
5599  * or zdb as real changes.
5600  */
5601 void
5602 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
5603 {
5604 #ifdef _KERNEL
5605 	sysevent_t		*ev;
5606 	sysevent_attr_list_t	*attr = NULL;
5607 	sysevent_value_t	value;
5608 	sysevent_id_t		eid;
5609 
5610 	ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
5611 	    SE_SLEEP);
5612 
5613 	value.value_type = SE_DATA_TYPE_STRING;
5614 	value.value.sv_string = spa_name(spa);
5615 	if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
5616 		goto done;
5617 
5618 	value.value_type = SE_DATA_TYPE_UINT64;
5619 	value.value.sv_uint64 = spa_guid(spa);
5620 	if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
5621 		goto done;
5622 
5623 	if (vd) {
5624 		value.value_type = SE_DATA_TYPE_UINT64;
5625 		value.value.sv_uint64 = vd->vdev_guid;
5626 		if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
5627 		    SE_SLEEP) != 0)
5628 			goto done;
5629 
5630 		if (vd->vdev_path) {
5631 			value.value_type = SE_DATA_TYPE_STRING;
5632 			value.value.sv_string = vd->vdev_path;
5633 			if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
5634 			    &value, SE_SLEEP) != 0)
5635 				goto done;
5636 		}
5637 	}
5638 
5639 	if (sysevent_attach_attributes(ev, attr) != 0)
5640 		goto done;
5641 	attr = NULL;
5642 
5643 	(void) log_sysevent(ev, SE_SLEEP, &eid);
5644 
5645 done:
5646 	if (attr)
5647 		sysevent_free_attr(attr);
5648 	sysevent_free(ev);
5649 #endif
5650 }
5651