xref: /illumos-gate/usr/src/uts/common/fs/zfs/spa.c (revision e69acc927b15c46de46ae637e53e1b808990a5bc)
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 2010 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*
28  * This file contains all the routines used when modifying on-disk SPA state.
29  * This includes opening, importing, destroying, exporting a pool, and syncing a
30  * pool.
31  */
32 
33 #include <sys/zfs_context.h>
34 #include <sys/fm/fs/zfs.h>
35 #include <sys/spa_impl.h>
36 #include <sys/zio.h>
37 #include <sys/zio_checksum.h>
38 #include <sys/dmu.h>
39 #include <sys/dmu_tx.h>
40 #include <sys/zap.h>
41 #include <sys/zil.h>
42 #include <sys/ddt.h>
43 #include <sys/vdev_impl.h>
44 #include <sys/metaslab.h>
45 #include <sys/metaslab_impl.h>
46 #include <sys/uberblock_impl.h>
47 #include <sys/txg.h>
48 #include <sys/avl.h>
49 #include <sys/dmu_traverse.h>
50 #include <sys/dmu_objset.h>
51 #include <sys/unique.h>
52 #include <sys/dsl_pool.h>
53 #include <sys/dsl_dataset.h>
54 #include <sys/dsl_dir.h>
55 #include <sys/dsl_prop.h>
56 #include <sys/dsl_synctask.h>
57 #include <sys/fs/zfs.h>
58 #include <sys/arc.h>
59 #include <sys/callb.h>
60 #include <sys/systeminfo.h>
61 #include <sys/spa_boot.h>
62 #include <sys/zfs_ioctl.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_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
110 	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
111 	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
112 };
113 
114 static void spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx);
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, B_TRUE, B_FALSE);
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, B_FALSE, B_TRUE);
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     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 
1785 	error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1786 	if (error)
1787 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1788 	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1789 
1790 	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
1791 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1792 
1793 	if (!mosconfig) {
1794 		uint64_t hostid;
1795 		nvlist_t *policy = NULL, *nvconfig;
1796 
1797 		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
1798 			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1799 
1800 		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
1801 		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
1802 			char *hostname;
1803 			unsigned long myhostid = 0;
1804 
1805 			VERIFY(nvlist_lookup_string(nvconfig,
1806 			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
1807 
1808 #ifdef	_KERNEL
1809 			myhostid = zone_get_hostid(NULL);
1810 #else	/* _KERNEL */
1811 			/*
1812 			 * We're emulating the system's hostid in userland, so
1813 			 * we can't use zone_get_hostid().
1814 			 */
1815 			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
1816 #endif	/* _KERNEL */
1817 			if (hostid != 0 && myhostid != 0 &&
1818 			    hostid != myhostid) {
1819 				nvlist_free(nvconfig);
1820 				cmn_err(CE_WARN, "pool '%s' could not be "
1821 				    "loaded as it was last accessed by "
1822 				    "another system (host: %s hostid: 0x%lx). "
1823 				    "See: http://www.sun.com/msg/ZFS-8000-EY",
1824 				    spa_name(spa), hostname,
1825 				    (unsigned long)hostid);
1826 				return (EBADF);
1827 			}
1828 		}
1829 		if (nvlist_lookup_nvlist(spa->spa_config,
1830 		    ZPOOL_REWIND_POLICY, &policy) == 0)
1831 			VERIFY(nvlist_add_nvlist(nvconfig,
1832 			    ZPOOL_REWIND_POLICY, policy) == 0);
1833 
1834 		spa_config_set(spa, nvconfig);
1835 		spa_unload(spa);
1836 		spa_deactivate(spa);
1837 		spa_activate(spa, orig_mode);
1838 
1839 		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
1840 	}
1841 
1842 	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPLIST,
1843 	    &spa->spa_deferred_bplist_obj) != 0)
1844 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1845 
1846 	/*
1847 	 * Load the bit that tells us to use the new accounting function
1848 	 * (raid-z deflation).  If we have an older pool, this will not
1849 	 * be present.
1850 	 */
1851 	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
1852 	if (error != 0 && error != ENOENT)
1853 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1854 
1855 	/*
1856 	 * Load the persistent error log.  If we have an older pool, this will
1857 	 * not be present.
1858 	 */
1859 	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
1860 	if (error != 0 && error != ENOENT)
1861 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1862 
1863 	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
1864 	    &spa->spa_errlog_scrub);
1865 	if (error != 0 && error != ENOENT)
1866 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1867 
1868 	/*
1869 	 * Load the history object.  If we have an older pool, this
1870 	 * will not be present.
1871 	 */
1872 	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
1873 	if (error != 0 && error != ENOENT)
1874 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1875 
1876 	/*
1877 	 * If we're assembling the pool from the split-off vdevs of
1878 	 * an existing pool, we don't want to attach the spares & cache
1879 	 * devices.
1880 	 */
1881 
1882 	/*
1883 	 * Load any hot spares for this pool.
1884 	 */
1885 	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
1886 	if (error != 0 && error != ENOENT)
1887 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1888 	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
1889 		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
1890 		if (load_nvlist(spa, spa->spa_spares.sav_object,
1891 		    &spa->spa_spares.sav_config) != 0)
1892 			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1893 
1894 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1895 		spa_load_spares(spa);
1896 		spa_config_exit(spa, SCL_ALL, FTAG);
1897 	} else if (error == 0) {
1898 		spa->spa_spares.sav_sync = B_TRUE;
1899 	}
1900 
1901 	/*
1902 	 * Load any level 2 ARC devices for this pool.
1903 	 */
1904 	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
1905 	    &spa->spa_l2cache.sav_object);
1906 	if (error != 0 && error != ENOENT)
1907 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1908 	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
1909 		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
1910 		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
1911 		    &spa->spa_l2cache.sav_config) != 0)
1912 			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1913 
1914 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1915 		spa_load_l2cache(spa);
1916 		spa_config_exit(spa, SCL_ALL, FTAG);
1917 	} else if (error == 0) {
1918 		spa->spa_l2cache.sav_sync = B_TRUE;
1919 	}
1920 
1921 	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
1922 
1923 	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
1924 	if (error && error != ENOENT)
1925 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1926 
1927 	if (error == 0) {
1928 		uint64_t autoreplace;
1929 
1930 		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
1931 		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
1932 		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
1933 		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
1934 		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
1935 		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
1936 		    &spa->spa_dedup_ditto);
1937 
1938 		spa->spa_autoreplace = (autoreplace != 0);
1939 	}
1940 
1941 	/*
1942 	 * If the 'autoreplace' property is set, then post a resource notifying
1943 	 * the ZFS DE that it should not issue any faults for unopenable
1944 	 * devices.  We also iterate over the vdevs, and post a sysevent for any
1945 	 * unopenable vdevs so that the normal autoreplace handler can take
1946 	 * over.
1947 	 */
1948 	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
1949 		spa_check_removed(spa->spa_root_vdev);
1950 		/*
1951 		 * For the import case, this is done in spa_import(), because
1952 		 * at this point we're using the spare definitions from
1953 		 * the MOS config, not necessarily from the userland config.
1954 		 */
1955 		if (state != SPA_LOAD_IMPORT) {
1956 			spa_aux_check_removed(&spa->spa_spares);
1957 			spa_aux_check_removed(&spa->spa_l2cache);
1958 		}
1959 	}
1960 
1961 	/*
1962 	 * Load the vdev state for all toplevel vdevs.
1963 	 */
1964 	vdev_load(rvd);
1965 
1966 	/*
1967 	 * Propagate the leaf DTLs we just loaded all the way up the tree.
1968 	 */
1969 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1970 	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
1971 	spa_config_exit(spa, SCL_ALL, FTAG);
1972 
1973 	/*
1974 	 * Check the state of the root vdev.  If it can't be opened, it
1975 	 * indicates one or more toplevel vdevs are faulted.
1976 	 */
1977 	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1978 		return (ENXIO);
1979 
1980 	/*
1981 	 * Load the DDTs (dedup tables).
1982 	 */
1983 	error = ddt_load(spa);
1984 	if (error != 0)
1985 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1986 
1987 	spa_update_dspace(spa);
1988 
1989 	if (state != SPA_LOAD_TRYIMPORT) {
1990 		error = spa_load_verify(spa);
1991 		if (error)
1992 			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
1993 			    error));
1994 	}
1995 
1996 	/*
1997 	 * Load the intent log state and check log integrity.  If we're
1998 	 * assembling a pool from a split, the log is not transferred over.
1999 	 */
2000 	if (type != SPA_IMPORT_ASSEMBLE) {
2001 		nvlist_t *nvconfig;
2002 
2003 		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2004 			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2005 
2006 		VERIFY(nvlist_lookup_nvlist(nvconfig, ZPOOL_CONFIG_VDEV_TREE,
2007 		    &nvroot) == 0);
2008 		spa_load_log_state(spa, nvroot);
2009 		nvlist_free(nvconfig);
2010 
2011 		if (spa_check_logs(spa)) {
2012 			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2013 			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2014 		}
2015 	}
2016 
2017 	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2018 	    spa->spa_load_max_txg == UINT64_MAX)) {
2019 		dmu_tx_t *tx;
2020 		int need_update = B_FALSE;
2021 
2022 		ASSERT(state != SPA_LOAD_TRYIMPORT);
2023 
2024 		/*
2025 		 * Claim log blocks that haven't been committed yet.
2026 		 * This must all happen in a single txg.
2027 		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2028 		 * invoked from zil_claim_log_block()'s i/o done callback.
2029 		 * Price of rollback is that we abandon the log.
2030 		 */
2031 		spa->spa_claiming = B_TRUE;
2032 
2033 		tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2034 		    spa_first_txg(spa));
2035 		(void) dmu_objset_find(spa_name(spa),
2036 		    zil_claim, tx, DS_FIND_CHILDREN);
2037 		dmu_tx_commit(tx);
2038 
2039 		spa->spa_claiming = B_FALSE;
2040 
2041 		spa_set_log_state(spa, SPA_LOG_GOOD);
2042 		spa->spa_sync_on = B_TRUE;
2043 		txg_sync_start(spa->spa_dsl_pool);
2044 
2045 		/*
2046 		 * Wait for all claims to sync.  We sync up to the highest
2047 		 * claimed log block birth time so that claimed log blocks
2048 		 * don't appear to be from the future.  spa_claim_max_txg
2049 		 * will have been set for us by either zil_check_log_chain()
2050 		 * (invoked from spa_check_logs()) or zil_claim() above.
2051 		 */
2052 		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2053 
2054 		/*
2055 		 * If the config cache is stale, or we have uninitialized
2056 		 * metaslabs (see spa_vdev_add()), then update the config.
2057 		 *
2058 		 * If spa_load_verbatim is true, trust the current
2059 		 * in-core spa_config and update the disk labels.
2060 		 */
2061 		if (config_cache_txg != spa->spa_config_txg ||
2062 		    state == SPA_LOAD_IMPORT || spa->spa_load_verbatim ||
2063 		    state == SPA_LOAD_RECOVER)
2064 			need_update = B_TRUE;
2065 
2066 		for (int c = 0; c < rvd->vdev_children; c++)
2067 			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2068 				need_update = B_TRUE;
2069 
2070 		/*
2071 		 * Update the config cache asychronously in case we're the
2072 		 * root pool, in which case the config cache isn't writable yet.
2073 		 */
2074 		if (need_update)
2075 			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2076 
2077 		/*
2078 		 * Check all DTLs to see if anything needs resilvering.
2079 		 */
2080 		if (vdev_resilver_needed(rvd, NULL, NULL))
2081 			spa_async_request(spa, SPA_ASYNC_RESILVER);
2082 
2083 		/*
2084 		 * Delete any inconsistent datasets.
2085 		 */
2086 		(void) dmu_objset_find(spa_name(spa),
2087 		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2088 
2089 		/*
2090 		 * Clean up any stale temporary dataset userrefs.
2091 		 */
2092 		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2093 	}
2094 
2095 	return (0);
2096 }
2097 
2098 static int
2099 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2100 {
2101 	spa_unload(spa);
2102 	spa_deactivate(spa);
2103 
2104 	spa->spa_load_max_txg--;
2105 
2106 	spa_activate(spa, spa_mode_global);
2107 	spa_async_suspend(spa);
2108 
2109 	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2110 }
2111 
2112 static int
2113 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2114     uint64_t max_request, int rewind_flags)
2115 {
2116 	nvlist_t *config = NULL;
2117 	int load_error, rewind_error;
2118 	uint64_t safe_rewind_txg;
2119 	uint64_t min_txg;
2120 
2121 	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2122 		spa->spa_load_max_txg = spa->spa_load_txg;
2123 		spa_set_log_state(spa, SPA_LOG_CLEAR);
2124 	} else {
2125 		spa->spa_load_max_txg = max_request;
2126 	}
2127 
2128 	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2129 	    mosconfig);
2130 	if (load_error == 0)
2131 		return (0);
2132 
2133 	if (spa->spa_root_vdev != NULL)
2134 		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2135 
2136 	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2137 	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2138 
2139 	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2140 		nvlist_free(config);
2141 		return (load_error);
2142 	}
2143 
2144 	/* Price of rolling back is discarding txgs, including log */
2145 	if (state == SPA_LOAD_RECOVER)
2146 		spa_set_log_state(spa, SPA_LOG_CLEAR);
2147 
2148 	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2149 	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2150 	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2151 	    TXG_INITIAL : safe_rewind_txg;
2152 
2153 	/*
2154 	 * Continue as long as we're finding errors, we're still within
2155 	 * the acceptable rewind range, and we're still finding uberblocks
2156 	 */
2157 	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2158 	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2159 		if (spa->spa_load_max_txg < safe_rewind_txg)
2160 			spa->spa_extreme_rewind = B_TRUE;
2161 		rewind_error = spa_load_retry(spa, state, mosconfig);
2162 	}
2163 
2164 	if (config)
2165 		spa_rewind_data_to_nvlist(spa, config);
2166 
2167 	spa->spa_extreme_rewind = B_FALSE;
2168 	spa->spa_load_max_txg = UINT64_MAX;
2169 
2170 	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2171 		spa_config_set(spa, config);
2172 
2173 	return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
2174 }
2175 
2176 /*
2177  * Pool Open/Import
2178  *
2179  * The import case is identical to an open except that the configuration is sent
2180  * down from userland, instead of grabbed from the configuration cache.  For the
2181  * case of an open, the pool configuration will exist in the
2182  * POOL_STATE_UNINITIALIZED state.
2183  *
2184  * The stats information (gen/count/ustats) is used to gather vdev statistics at
2185  * the same time open the pool, without having to keep around the spa_t in some
2186  * ambiguous state.
2187  */
2188 static int
2189 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2190     nvlist_t **config)
2191 {
2192 	spa_t *spa;
2193 	int error;
2194 	int locked = B_FALSE;
2195 
2196 	*spapp = NULL;
2197 
2198 	/*
2199 	 * As disgusting as this is, we need to support recursive calls to this
2200 	 * function because dsl_dir_open() is called during spa_load(), and ends
2201 	 * up calling spa_open() again.  The real fix is to figure out how to
2202 	 * avoid dsl_dir_open() calling this in the first place.
2203 	 */
2204 	if (mutex_owner(&spa_namespace_lock) != curthread) {
2205 		mutex_enter(&spa_namespace_lock);
2206 		locked = B_TRUE;
2207 	}
2208 
2209 	if ((spa = spa_lookup(pool)) == NULL) {
2210 		if (locked)
2211 			mutex_exit(&spa_namespace_lock);
2212 		return (ENOENT);
2213 	}
2214 
2215 	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2216 		spa_load_state_t state = SPA_LOAD_OPEN;
2217 		zpool_rewind_policy_t policy;
2218 
2219 		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2220 		    &policy);
2221 		if (policy.zrp_request & ZPOOL_DO_REWIND)
2222 			state = SPA_LOAD_RECOVER;
2223 
2224 		spa_activate(spa, spa_mode_global);
2225 
2226 		if (spa->spa_last_open_failed && (policy.zrp_request &
2227 		    (ZPOOL_NO_REWIND | ZPOOL_NEVER_REWIND))) {
2228 			if (config != NULL && spa->spa_config)
2229 				VERIFY(nvlist_dup(spa->spa_config,
2230 				    config, KM_SLEEP) == 0);
2231 			spa_deactivate(spa);
2232 			if (locked)
2233 				mutex_exit(&spa_namespace_lock);
2234 			return (spa->spa_last_open_failed);
2235 		}
2236 
2237 		if (state != SPA_LOAD_RECOVER)
2238 			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2239 
2240 		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2241 		    policy.zrp_request);
2242 
2243 		if (error == EBADF) {
2244 			/*
2245 			 * If vdev_validate() returns failure (indicated by
2246 			 * EBADF), it indicates that one of the vdevs indicates
2247 			 * that the pool has been exported or destroyed.  If
2248 			 * this is the case, the config cache is out of sync and
2249 			 * we should remove the pool from the namespace.
2250 			 */
2251 			spa_unload(spa);
2252 			spa_deactivate(spa);
2253 			spa_config_sync(spa, B_TRUE, B_TRUE);
2254 			spa_remove(spa);
2255 			if (locked)
2256 				mutex_exit(&spa_namespace_lock);
2257 			return (ENOENT);
2258 		}
2259 
2260 		if (error) {
2261 			/*
2262 			 * We can't open the pool, but we still have useful
2263 			 * information: the state of each vdev after the
2264 			 * attempted vdev_open().  Return this to the user.
2265 			 */
2266 			if (config != NULL && spa->spa_config)
2267 				VERIFY(nvlist_dup(spa->spa_config, config,
2268 				    KM_SLEEP) == 0);
2269 			spa_unload(spa);
2270 			spa_deactivate(spa);
2271 			spa->spa_last_open_failed = error;
2272 			if (locked)
2273 				mutex_exit(&spa_namespace_lock);
2274 			*spapp = NULL;
2275 			return (error);
2276 		}
2277 
2278 	}
2279 
2280 	spa_open_ref(spa, tag);
2281 
2282 
2283 	if (config != NULL)
2284 		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2285 
2286 	if (locked) {
2287 		spa->spa_last_open_failed = 0;
2288 		spa->spa_last_ubsync_txg = 0;
2289 		spa->spa_load_txg = 0;
2290 		mutex_exit(&spa_namespace_lock);
2291 	}
2292 
2293 	*spapp = spa;
2294 
2295 	return (0);
2296 }
2297 
2298 int
2299 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2300     nvlist_t **config)
2301 {
2302 	return (spa_open_common(name, spapp, tag, policy, config));
2303 }
2304 
2305 int
2306 spa_open(const char *name, spa_t **spapp, void *tag)
2307 {
2308 	return (spa_open_common(name, spapp, tag, NULL, NULL));
2309 }
2310 
2311 /*
2312  * Lookup the given spa_t, incrementing the inject count in the process,
2313  * preventing it from being exported or destroyed.
2314  */
2315 spa_t *
2316 spa_inject_addref(char *name)
2317 {
2318 	spa_t *spa;
2319 
2320 	mutex_enter(&spa_namespace_lock);
2321 	if ((spa = spa_lookup(name)) == NULL) {
2322 		mutex_exit(&spa_namespace_lock);
2323 		return (NULL);
2324 	}
2325 	spa->spa_inject_ref++;
2326 	mutex_exit(&spa_namespace_lock);
2327 
2328 	return (spa);
2329 }
2330 
2331 void
2332 spa_inject_delref(spa_t *spa)
2333 {
2334 	mutex_enter(&spa_namespace_lock);
2335 	spa->spa_inject_ref--;
2336 	mutex_exit(&spa_namespace_lock);
2337 }
2338 
2339 /*
2340  * Add spares device information to the nvlist.
2341  */
2342 static void
2343 spa_add_spares(spa_t *spa, nvlist_t *config)
2344 {
2345 	nvlist_t **spares;
2346 	uint_t i, nspares;
2347 	nvlist_t *nvroot;
2348 	uint64_t guid;
2349 	vdev_stat_t *vs;
2350 	uint_t vsc;
2351 	uint64_t pool;
2352 
2353 	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2354 
2355 	if (spa->spa_spares.sav_count == 0)
2356 		return;
2357 
2358 	VERIFY(nvlist_lookup_nvlist(config,
2359 	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2360 	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2361 	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2362 	if (nspares != 0) {
2363 		VERIFY(nvlist_add_nvlist_array(nvroot,
2364 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2365 		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2366 		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2367 
2368 		/*
2369 		 * Go through and find any spares which have since been
2370 		 * repurposed as an active spare.  If this is the case, update
2371 		 * their status appropriately.
2372 		 */
2373 		for (i = 0; i < nspares; i++) {
2374 			VERIFY(nvlist_lookup_uint64(spares[i],
2375 			    ZPOOL_CONFIG_GUID, &guid) == 0);
2376 			if (spa_spare_exists(guid, &pool, NULL) &&
2377 			    pool != 0ULL) {
2378 				VERIFY(nvlist_lookup_uint64_array(
2379 				    spares[i], ZPOOL_CONFIG_STATS,
2380 				    (uint64_t **)&vs, &vsc) == 0);
2381 				vs->vs_state = VDEV_STATE_CANT_OPEN;
2382 				vs->vs_aux = VDEV_AUX_SPARED;
2383 			}
2384 		}
2385 	}
2386 }
2387 
2388 /*
2389  * Add l2cache device information to the nvlist, including vdev stats.
2390  */
2391 static void
2392 spa_add_l2cache(spa_t *spa, nvlist_t *config)
2393 {
2394 	nvlist_t **l2cache;
2395 	uint_t i, j, nl2cache;
2396 	nvlist_t *nvroot;
2397 	uint64_t guid;
2398 	vdev_t *vd;
2399 	vdev_stat_t *vs;
2400 	uint_t vsc;
2401 
2402 	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2403 
2404 	if (spa->spa_l2cache.sav_count == 0)
2405 		return;
2406 
2407 	VERIFY(nvlist_lookup_nvlist(config,
2408 	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2409 	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2410 	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2411 	if (nl2cache != 0) {
2412 		VERIFY(nvlist_add_nvlist_array(nvroot,
2413 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2414 		VERIFY(nvlist_lookup_nvlist_array(nvroot,
2415 		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2416 
2417 		/*
2418 		 * Update level 2 cache device stats.
2419 		 */
2420 
2421 		for (i = 0; i < nl2cache; i++) {
2422 			VERIFY(nvlist_lookup_uint64(l2cache[i],
2423 			    ZPOOL_CONFIG_GUID, &guid) == 0);
2424 
2425 			vd = NULL;
2426 			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2427 				if (guid ==
2428 				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2429 					vd = spa->spa_l2cache.sav_vdevs[j];
2430 					break;
2431 				}
2432 			}
2433 			ASSERT(vd != NULL);
2434 
2435 			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2436 			    ZPOOL_CONFIG_STATS, (uint64_t **)&vs, &vsc) == 0);
2437 			vdev_get_stats(vd, vs);
2438 		}
2439 	}
2440 }
2441 
2442 int
2443 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2444 {
2445 	int error;
2446 	spa_t *spa;
2447 
2448 	*config = NULL;
2449 	error = spa_open_common(name, &spa, FTAG, NULL, config);
2450 
2451 	if (spa != NULL) {
2452 		/*
2453 		 * This still leaves a window of inconsistency where the spares
2454 		 * or l2cache devices could change and the config would be
2455 		 * self-inconsistent.
2456 		 */
2457 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2458 
2459 		if (*config != NULL) {
2460 			VERIFY(nvlist_add_uint64(*config,
2461 			    ZPOOL_CONFIG_ERRCOUNT,
2462 			    spa_get_errlog_size(spa)) == 0);
2463 
2464 			if (spa_suspended(spa))
2465 				VERIFY(nvlist_add_uint64(*config,
2466 				    ZPOOL_CONFIG_SUSPENDED,
2467 				    spa->spa_failmode) == 0);
2468 
2469 			spa_add_spares(spa, *config);
2470 			spa_add_l2cache(spa, *config);
2471 		}
2472 	}
2473 
2474 	/*
2475 	 * We want to get the alternate root even for faulted pools, so we cheat
2476 	 * and call spa_lookup() directly.
2477 	 */
2478 	if (altroot) {
2479 		if (spa == NULL) {
2480 			mutex_enter(&spa_namespace_lock);
2481 			spa = spa_lookup(name);
2482 			if (spa)
2483 				spa_altroot(spa, altroot, buflen);
2484 			else
2485 				altroot[0] = '\0';
2486 			spa = NULL;
2487 			mutex_exit(&spa_namespace_lock);
2488 		} else {
2489 			spa_altroot(spa, altroot, buflen);
2490 		}
2491 	}
2492 
2493 	if (spa != NULL) {
2494 		spa_config_exit(spa, SCL_CONFIG, FTAG);
2495 		spa_close(spa, FTAG);
2496 	}
2497 
2498 	return (error);
2499 }
2500 
2501 /*
2502  * Validate that the auxiliary device array is well formed.  We must have an
2503  * array of nvlists, each which describes a valid leaf vdev.  If this is an
2504  * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2505  * specified, as long as they are well-formed.
2506  */
2507 static int
2508 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2509     spa_aux_vdev_t *sav, const char *config, uint64_t version,
2510     vdev_labeltype_t label)
2511 {
2512 	nvlist_t **dev;
2513 	uint_t i, ndev;
2514 	vdev_t *vd;
2515 	int error;
2516 
2517 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2518 
2519 	/*
2520 	 * It's acceptable to have no devs specified.
2521 	 */
2522 	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2523 		return (0);
2524 
2525 	if (ndev == 0)
2526 		return (EINVAL);
2527 
2528 	/*
2529 	 * Make sure the pool is formatted with a version that supports this
2530 	 * device type.
2531 	 */
2532 	if (spa_version(spa) < version)
2533 		return (ENOTSUP);
2534 
2535 	/*
2536 	 * Set the pending device list so we correctly handle device in-use
2537 	 * checking.
2538 	 */
2539 	sav->sav_pending = dev;
2540 	sav->sav_npending = ndev;
2541 
2542 	for (i = 0; i < ndev; i++) {
2543 		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2544 		    mode)) != 0)
2545 			goto out;
2546 
2547 		if (!vd->vdev_ops->vdev_op_leaf) {
2548 			vdev_free(vd);
2549 			error = EINVAL;
2550 			goto out;
2551 		}
2552 
2553 		/*
2554 		 * The L2ARC currently only supports disk devices in
2555 		 * kernel context.  For user-level testing, we allow it.
2556 		 */
2557 #ifdef _KERNEL
2558 		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2559 		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2560 			error = ENOTBLK;
2561 			goto out;
2562 		}
2563 #endif
2564 		vd->vdev_top = vd;
2565 
2566 		if ((error = vdev_open(vd)) == 0 &&
2567 		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
2568 			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2569 			    vd->vdev_guid) == 0);
2570 		}
2571 
2572 		vdev_free(vd);
2573 
2574 		if (error &&
2575 		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2576 			goto out;
2577 		else
2578 			error = 0;
2579 	}
2580 
2581 out:
2582 	sav->sav_pending = NULL;
2583 	sav->sav_npending = 0;
2584 	return (error);
2585 }
2586 
2587 static int
2588 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2589 {
2590 	int error;
2591 
2592 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2593 
2594 	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2595 	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2596 	    VDEV_LABEL_SPARE)) != 0) {
2597 		return (error);
2598 	}
2599 
2600 	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2601 	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2602 	    VDEV_LABEL_L2CACHE));
2603 }
2604 
2605 static void
2606 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2607     const char *config)
2608 {
2609 	int i;
2610 
2611 	if (sav->sav_config != NULL) {
2612 		nvlist_t **olddevs;
2613 		uint_t oldndevs;
2614 		nvlist_t **newdevs;
2615 
2616 		/*
2617 		 * Generate new dev list by concatentating with the
2618 		 * current dev list.
2619 		 */
2620 		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2621 		    &olddevs, &oldndevs) == 0);
2622 
2623 		newdevs = kmem_alloc(sizeof (void *) *
2624 		    (ndevs + oldndevs), KM_SLEEP);
2625 		for (i = 0; i < oldndevs; i++)
2626 			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2627 			    KM_SLEEP) == 0);
2628 		for (i = 0; i < ndevs; i++)
2629 			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2630 			    KM_SLEEP) == 0);
2631 
2632 		VERIFY(nvlist_remove(sav->sav_config, config,
2633 		    DATA_TYPE_NVLIST_ARRAY) == 0);
2634 
2635 		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2636 		    config, newdevs, ndevs + oldndevs) == 0);
2637 		for (i = 0; i < oldndevs + ndevs; i++)
2638 			nvlist_free(newdevs[i]);
2639 		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2640 	} else {
2641 		/*
2642 		 * Generate a new dev list.
2643 		 */
2644 		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2645 		    KM_SLEEP) == 0);
2646 		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2647 		    devs, ndevs) == 0);
2648 	}
2649 }
2650 
2651 /*
2652  * Stop and drop level 2 ARC devices
2653  */
2654 void
2655 spa_l2cache_drop(spa_t *spa)
2656 {
2657 	vdev_t *vd;
2658 	int i;
2659 	spa_aux_vdev_t *sav = &spa->spa_l2cache;
2660 
2661 	for (i = 0; i < sav->sav_count; i++) {
2662 		uint64_t pool;
2663 
2664 		vd = sav->sav_vdevs[i];
2665 		ASSERT(vd != NULL);
2666 
2667 		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2668 		    pool != 0ULL && l2arc_vdev_present(vd))
2669 			l2arc_remove_vdev(vd);
2670 		if (vd->vdev_isl2cache)
2671 			spa_l2cache_remove(vd);
2672 		vdev_clear_stats(vd);
2673 		(void) vdev_close(vd);
2674 	}
2675 }
2676 
2677 /*
2678  * Pool Creation
2679  */
2680 int
2681 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2682     const char *history_str, nvlist_t *zplprops)
2683 {
2684 	spa_t *spa;
2685 	char *altroot = NULL;
2686 	vdev_t *rvd;
2687 	dsl_pool_t *dp;
2688 	dmu_tx_t *tx;
2689 	int error = 0;
2690 	uint64_t txg = TXG_INITIAL;
2691 	nvlist_t **spares, **l2cache;
2692 	uint_t nspares, nl2cache;
2693 	uint64_t version;
2694 
2695 	/*
2696 	 * If this pool already exists, return failure.
2697 	 */
2698 	mutex_enter(&spa_namespace_lock);
2699 	if (spa_lookup(pool) != NULL) {
2700 		mutex_exit(&spa_namespace_lock);
2701 		return (EEXIST);
2702 	}
2703 
2704 	/*
2705 	 * Allocate a new spa_t structure.
2706 	 */
2707 	(void) nvlist_lookup_string(props,
2708 	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2709 	spa = spa_add(pool, NULL, altroot);
2710 	spa_activate(spa, spa_mode_global);
2711 
2712 	if (props && (error = spa_prop_validate(spa, props))) {
2713 		spa_deactivate(spa);
2714 		spa_remove(spa);
2715 		mutex_exit(&spa_namespace_lock);
2716 		return (error);
2717 	}
2718 
2719 	if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2720 	    &version) != 0)
2721 		version = SPA_VERSION;
2722 	ASSERT(version <= SPA_VERSION);
2723 
2724 	spa->spa_first_txg = txg;
2725 	spa->spa_uberblock.ub_txg = txg - 1;
2726 	spa->spa_uberblock.ub_version = version;
2727 	spa->spa_ubsync = spa->spa_uberblock;
2728 
2729 	/*
2730 	 * Create "The Godfather" zio to hold all async IOs
2731 	 */
2732 	spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2733 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2734 
2735 	/*
2736 	 * Create the root vdev.
2737 	 */
2738 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2739 
2740 	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2741 
2742 	ASSERT(error != 0 || rvd != NULL);
2743 	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2744 
2745 	if (error == 0 && !zfs_allocatable_devs(nvroot))
2746 		error = EINVAL;
2747 
2748 	if (error == 0 &&
2749 	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2750 	    (error = spa_validate_aux(spa, nvroot, txg,
2751 	    VDEV_ALLOC_ADD)) == 0) {
2752 		for (int c = 0; c < rvd->vdev_children; c++) {
2753 			vdev_metaslab_set_size(rvd->vdev_child[c]);
2754 			vdev_expand(rvd->vdev_child[c], txg);
2755 		}
2756 	}
2757 
2758 	spa_config_exit(spa, SCL_ALL, FTAG);
2759 
2760 	if (error != 0) {
2761 		spa_unload(spa);
2762 		spa_deactivate(spa);
2763 		spa_remove(spa);
2764 		mutex_exit(&spa_namespace_lock);
2765 		return (error);
2766 	}
2767 
2768 	/*
2769 	 * Get the list of spares, if specified.
2770 	 */
2771 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2772 	    &spares, &nspares) == 0) {
2773 		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
2774 		    KM_SLEEP) == 0);
2775 		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2776 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2777 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2778 		spa_load_spares(spa);
2779 		spa_config_exit(spa, SCL_ALL, FTAG);
2780 		spa->spa_spares.sav_sync = B_TRUE;
2781 	}
2782 
2783 	/*
2784 	 * Get the list of level 2 cache devices, if specified.
2785 	 */
2786 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2787 	    &l2cache, &nl2cache) == 0) {
2788 		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2789 		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
2790 		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2791 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2792 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2793 		spa_load_l2cache(spa);
2794 		spa_config_exit(spa, SCL_ALL, FTAG);
2795 		spa->spa_l2cache.sav_sync = B_TRUE;
2796 	}
2797 
2798 	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
2799 	spa->spa_meta_objset = dp->dp_meta_objset;
2800 
2801 	/*
2802 	 * Create DDTs (dedup tables).
2803 	 */
2804 	ddt_create(spa);
2805 
2806 	spa_update_dspace(spa);
2807 
2808 	tx = dmu_tx_create_assigned(dp, txg);
2809 
2810 	/*
2811 	 * Create the pool config object.
2812 	 */
2813 	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
2814 	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
2815 	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2816 
2817 	if (zap_add(spa->spa_meta_objset,
2818 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
2819 	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
2820 		cmn_err(CE_PANIC, "failed to add pool config");
2821 	}
2822 
2823 	/* Newly created pools with the right version are always deflated. */
2824 	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
2825 		spa->spa_deflate = TRUE;
2826 		if (zap_add(spa->spa_meta_objset,
2827 		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
2828 		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
2829 			cmn_err(CE_PANIC, "failed to add deflate");
2830 		}
2831 	}
2832 
2833 	/*
2834 	 * Create the deferred-free bplist object.  Turn off compression
2835 	 * because sync-to-convergence takes longer if the blocksize
2836 	 * keeps changing.
2837 	 */
2838 	spa->spa_deferred_bplist_obj = bplist_create(spa->spa_meta_objset,
2839 	    1 << 14, tx);
2840 	dmu_object_set_compress(spa->spa_meta_objset,
2841 	    spa->spa_deferred_bplist_obj, ZIO_COMPRESS_OFF, tx);
2842 
2843 	if (zap_add(spa->spa_meta_objset,
2844 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST,
2845 	    sizeof (uint64_t), 1, &spa->spa_deferred_bplist_obj, tx) != 0) {
2846 		cmn_err(CE_PANIC, "failed to add bplist");
2847 	}
2848 
2849 	/*
2850 	 * Create the pool's history object.
2851 	 */
2852 	if (version >= SPA_VERSION_ZPOOL_HISTORY)
2853 		spa_history_create_obj(spa, tx);
2854 
2855 	/*
2856 	 * Set pool properties.
2857 	 */
2858 	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
2859 	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2860 	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
2861 	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
2862 
2863 	if (props != NULL) {
2864 		spa_configfile_set(spa, props, B_FALSE);
2865 		spa_sync_props(spa, props, CRED(), tx);
2866 	}
2867 
2868 	dmu_tx_commit(tx);
2869 
2870 	spa->spa_sync_on = B_TRUE;
2871 	txg_sync_start(spa->spa_dsl_pool);
2872 
2873 	/*
2874 	 * We explicitly wait for the first transaction to complete so that our
2875 	 * bean counters are appropriately updated.
2876 	 */
2877 	txg_wait_synced(spa->spa_dsl_pool, txg);
2878 
2879 	spa_config_sync(spa, B_FALSE, B_TRUE);
2880 
2881 	if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
2882 		(void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
2883 	spa_history_log_version(spa, LOG_POOL_CREATE);
2884 
2885 	spa->spa_minref = refcount_count(&spa->spa_refcount);
2886 
2887 	mutex_exit(&spa_namespace_lock);
2888 
2889 	return (0);
2890 }
2891 
2892 #ifdef _KERNEL
2893 /*
2894  * Get the root pool information from the root disk, then import the root pool
2895  * during the system boot up time.
2896  */
2897 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
2898 
2899 static nvlist_t *
2900 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
2901 {
2902 	nvlist_t *config;
2903 	nvlist_t *nvtop, *nvroot;
2904 	uint64_t pgid;
2905 
2906 	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
2907 		return (NULL);
2908 
2909 	/*
2910 	 * Add this top-level vdev to the child array.
2911 	 */
2912 	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2913 	    &nvtop) == 0);
2914 	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
2915 	    &pgid) == 0);
2916 	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
2917 
2918 	/*
2919 	 * Put this pool's top-level vdevs into a root vdev.
2920 	 */
2921 	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2922 	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
2923 	    VDEV_TYPE_ROOT) == 0);
2924 	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
2925 	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
2926 	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2927 	    &nvtop, 1) == 0);
2928 
2929 	/*
2930 	 * Replace the existing vdev_tree with the new root vdev in
2931 	 * this pool's configuration (remove the old, add the new).
2932 	 */
2933 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
2934 	nvlist_free(nvroot);
2935 	return (config);
2936 }
2937 
2938 /*
2939  * Walk the vdev tree and see if we can find a device with "better"
2940  * configuration. A configuration is "better" if the label on that
2941  * device has a more recent txg.
2942  */
2943 static void
2944 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
2945 {
2946 	for (int c = 0; c < vd->vdev_children; c++)
2947 		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
2948 
2949 	if (vd->vdev_ops->vdev_op_leaf) {
2950 		nvlist_t *label;
2951 		uint64_t label_txg;
2952 
2953 		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
2954 		    &label) != 0)
2955 			return;
2956 
2957 		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
2958 		    &label_txg) == 0);
2959 
2960 		/*
2961 		 * Do we have a better boot device?
2962 		 */
2963 		if (label_txg > *txg) {
2964 			*txg = label_txg;
2965 			*avd = vd;
2966 		}
2967 		nvlist_free(label);
2968 	}
2969 }
2970 
2971 /*
2972  * Import a root pool.
2973  *
2974  * For x86. devpath_list will consist of devid and/or physpath name of
2975  * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
2976  * The GRUB "findroot" command will return the vdev we should boot.
2977  *
2978  * For Sparc, devpath_list consists the physpath name of the booting device
2979  * no matter the rootpool is a single device pool or a mirrored pool.
2980  * e.g.
2981  *	"/pci@1f,0/ide@d/disk@0,0:a"
2982  */
2983 int
2984 spa_import_rootpool(char *devpath, char *devid)
2985 {
2986 	spa_t *spa;
2987 	vdev_t *rvd, *bvd, *avd = NULL;
2988 	nvlist_t *config, *nvtop;
2989 	uint64_t guid, txg;
2990 	char *pname;
2991 	int error;
2992 
2993 	/*
2994 	 * Read the label from the boot device and generate a configuration.
2995 	 */
2996 	config = spa_generate_rootconf(devpath, devid, &guid);
2997 #if defined(_OBP) && defined(_KERNEL)
2998 	if (config == NULL) {
2999 		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3000 			/* iscsi boot */
3001 			get_iscsi_bootpath_phy(devpath);
3002 			config = spa_generate_rootconf(devpath, devid, &guid);
3003 		}
3004 	}
3005 #endif
3006 	if (config == NULL) {
3007 		cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
3008 		    devpath);
3009 		return (EIO);
3010 	}
3011 
3012 	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3013 	    &pname) == 0);
3014 	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3015 
3016 	mutex_enter(&spa_namespace_lock);
3017 	if ((spa = spa_lookup(pname)) != NULL) {
3018 		/*
3019 		 * Remove the existing root pool from the namespace so that we
3020 		 * can replace it with the correct config we just read in.
3021 		 */
3022 		spa_remove(spa);
3023 	}
3024 
3025 	spa = spa_add(pname, config, NULL);
3026 	spa->spa_is_root = B_TRUE;
3027 	spa->spa_load_verbatim = B_TRUE;
3028 
3029 	/*
3030 	 * Build up a vdev tree based on the boot device's label config.
3031 	 */
3032 	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3033 	    &nvtop) == 0);
3034 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3035 	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3036 	    VDEV_ALLOC_ROOTPOOL);
3037 	spa_config_exit(spa, SCL_ALL, FTAG);
3038 	if (error) {
3039 		mutex_exit(&spa_namespace_lock);
3040 		nvlist_free(config);
3041 		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3042 		    pname);
3043 		return (error);
3044 	}
3045 
3046 	/*
3047 	 * Get the boot vdev.
3048 	 */
3049 	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3050 		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3051 		    (u_longlong_t)guid);
3052 		error = ENOENT;
3053 		goto out;
3054 	}
3055 
3056 	/*
3057 	 * Determine if there is a better boot device.
3058 	 */
3059 	avd = bvd;
3060 	spa_alt_rootvdev(rvd, &avd, &txg);
3061 	if (avd != bvd) {
3062 		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3063 		    "try booting from '%s'", avd->vdev_path);
3064 		error = EINVAL;
3065 		goto out;
3066 	}
3067 
3068 	/*
3069 	 * If the boot device is part of a spare vdev then ensure that
3070 	 * we're booting off the active spare.
3071 	 */
3072 	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3073 	    !bvd->vdev_isspare) {
3074 		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3075 		    "try booting from '%s'",
3076 		    bvd->vdev_parent->vdev_child[1]->vdev_path);
3077 		error = EINVAL;
3078 		goto out;
3079 	}
3080 
3081 	error = 0;
3082 	spa_history_log_version(spa, LOG_POOL_IMPORT);
3083 out:
3084 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3085 	vdev_free(rvd);
3086 	spa_config_exit(spa, SCL_ALL, FTAG);
3087 	mutex_exit(&spa_namespace_lock);
3088 
3089 	nvlist_free(config);
3090 	return (error);
3091 }
3092 
3093 #endif
3094 
3095 /*
3096  * Take a pool and insert it into the namespace as if it had been loaded at
3097  * boot.
3098  */
3099 int
3100 spa_import_verbatim(const char *pool, nvlist_t *config, nvlist_t *props)
3101 {
3102 	spa_t *spa;
3103 	char *altroot = NULL;
3104 
3105 	mutex_enter(&spa_namespace_lock);
3106 	if (spa_lookup(pool) != NULL) {
3107 		mutex_exit(&spa_namespace_lock);
3108 		return (EEXIST);
3109 	}
3110 
3111 	(void) nvlist_lookup_string(props,
3112 	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3113 	spa = spa_add(pool, config, altroot);
3114 
3115 	spa->spa_load_verbatim = B_TRUE;
3116 
3117 	if (props != NULL)
3118 		spa_configfile_set(spa, props, B_FALSE);
3119 
3120 	spa_config_sync(spa, B_FALSE, B_TRUE);
3121 
3122 	mutex_exit(&spa_namespace_lock);
3123 	spa_history_log_version(spa, LOG_POOL_IMPORT);
3124 
3125 	return (0);
3126 }
3127 
3128 /*
3129  * Import a non-root pool into the system.
3130  */
3131 int
3132 spa_import(const char *pool, nvlist_t *config, nvlist_t *props)
3133 {
3134 	spa_t *spa;
3135 	char *altroot = NULL;
3136 	spa_load_state_t state = SPA_LOAD_IMPORT;
3137 	zpool_rewind_policy_t policy;
3138 	int error;
3139 	nvlist_t *nvroot;
3140 	nvlist_t **spares, **l2cache;
3141 	uint_t nspares, nl2cache;
3142 
3143 	/*
3144 	 * If a pool with this name exists, return failure.
3145 	 */
3146 	mutex_enter(&spa_namespace_lock);
3147 	if (spa_lookup(pool) != NULL) {
3148 		mutex_exit(&spa_namespace_lock);
3149 		return (EEXIST);
3150 	}
3151 
3152 	zpool_get_rewind_policy(config, &policy);
3153 	if (policy.zrp_request & ZPOOL_DO_REWIND)
3154 		state = SPA_LOAD_RECOVER;
3155 
3156 	/*
3157 	 * Create and initialize the spa structure.
3158 	 */
3159 	(void) nvlist_lookup_string(props,
3160 	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3161 	spa = spa_add(pool, config, altroot);
3162 	spa_activate(spa, spa_mode_global);
3163 
3164 	/*
3165 	 * Don't start async tasks until we know everything is healthy.
3166 	 */
3167 	spa_async_suspend(spa);
3168 
3169 	/*
3170 	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3171 	 * because the user-supplied config is actually the one to trust when
3172 	 * doing an import.
3173 	 */
3174 	if (state != SPA_LOAD_RECOVER)
3175 		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3176 	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3177 	    policy.zrp_request);
3178 
3179 	/*
3180 	 * Propagate anything learned about failing or best txgs
3181 	 * back to caller
3182 	 */
3183 	spa_rewind_data_to_nvlist(spa, config);
3184 
3185 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3186 	/*
3187 	 * Toss any existing sparelist, as it doesn't have any validity
3188 	 * anymore, and conflicts with spa_has_spare().
3189 	 */
3190 	if (spa->spa_spares.sav_config) {
3191 		nvlist_free(spa->spa_spares.sav_config);
3192 		spa->spa_spares.sav_config = NULL;
3193 		spa_load_spares(spa);
3194 	}
3195 	if (spa->spa_l2cache.sav_config) {
3196 		nvlist_free(spa->spa_l2cache.sav_config);
3197 		spa->spa_l2cache.sav_config = NULL;
3198 		spa_load_l2cache(spa);
3199 	}
3200 
3201 	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3202 	    &nvroot) == 0);
3203 	if (error == 0)
3204 		error = spa_validate_aux(spa, nvroot, -1ULL,
3205 		    VDEV_ALLOC_SPARE);
3206 	if (error == 0)
3207 		error = spa_validate_aux(spa, nvroot, -1ULL,
3208 		    VDEV_ALLOC_L2CACHE);
3209 	spa_config_exit(spa, SCL_ALL, FTAG);
3210 
3211 	if (props != NULL)
3212 		spa_configfile_set(spa, props, B_FALSE);
3213 
3214 	if (error != 0 || (props && spa_writeable(spa) &&
3215 	    (error = spa_prop_set(spa, props)))) {
3216 		spa_unload(spa);
3217 		spa_deactivate(spa);
3218 		spa_remove(spa);
3219 		mutex_exit(&spa_namespace_lock);
3220 		return (error);
3221 	}
3222 
3223 	spa_async_resume(spa);
3224 
3225 	/*
3226 	 * Override any spares and level 2 cache devices as specified by
3227 	 * the user, as these may have correct device names/devids, etc.
3228 	 */
3229 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3230 	    &spares, &nspares) == 0) {
3231 		if (spa->spa_spares.sav_config)
3232 			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3233 			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3234 		else
3235 			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3236 			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3237 		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3238 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3239 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3240 		spa_load_spares(spa);
3241 		spa_config_exit(spa, SCL_ALL, FTAG);
3242 		spa->spa_spares.sav_sync = B_TRUE;
3243 	}
3244 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3245 	    &l2cache, &nl2cache) == 0) {
3246 		if (spa->spa_l2cache.sav_config)
3247 			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3248 			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3249 		else
3250 			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3251 			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3252 		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3253 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3254 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3255 		spa_load_l2cache(spa);
3256 		spa_config_exit(spa, SCL_ALL, FTAG);
3257 		spa->spa_l2cache.sav_sync = B_TRUE;
3258 	}
3259 
3260 	/*
3261 	 * Check for any removed devices.
3262 	 */
3263 	if (spa->spa_autoreplace) {
3264 		spa_aux_check_removed(&spa->spa_spares);
3265 		spa_aux_check_removed(&spa->spa_l2cache);
3266 	}
3267 
3268 	if (spa_writeable(spa)) {
3269 		/*
3270 		 * Update the config cache to include the newly-imported pool.
3271 		 */
3272 		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3273 	}
3274 
3275 	/*
3276 	 * It's possible that the pool was expanded while it was exported.
3277 	 * We kick off an async task to handle this for us.
3278 	 */
3279 	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3280 
3281 	mutex_exit(&spa_namespace_lock);
3282 	spa_history_log_version(spa, LOG_POOL_IMPORT);
3283 
3284 	return (0);
3285 }
3286 
3287 nvlist_t *
3288 spa_tryimport(nvlist_t *tryconfig)
3289 {
3290 	nvlist_t *config = NULL;
3291 	char *poolname;
3292 	spa_t *spa;
3293 	uint64_t state;
3294 	int error;
3295 
3296 	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3297 		return (NULL);
3298 
3299 	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3300 		return (NULL);
3301 
3302 	/*
3303 	 * Create and initialize the spa structure.
3304 	 */
3305 	mutex_enter(&spa_namespace_lock);
3306 	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3307 	spa_activate(spa, FREAD);
3308 
3309 	/*
3310 	 * Pass off the heavy lifting to spa_load().
3311 	 * Pass TRUE for mosconfig because the user-supplied config
3312 	 * is actually the one to trust when doing an import.
3313 	 */
3314 	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3315 
3316 	/*
3317 	 * If 'tryconfig' was at least parsable, return the current config.
3318 	 */
3319 	if (spa->spa_root_vdev != NULL) {
3320 		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3321 		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3322 		    poolname) == 0);
3323 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3324 		    state) == 0);
3325 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3326 		    spa->spa_uberblock.ub_timestamp) == 0);
3327 
3328 		/*
3329 		 * If the bootfs property exists on this pool then we
3330 		 * copy it out so that external consumers can tell which
3331 		 * pools are bootable.
3332 		 */
3333 		if ((!error || error == EEXIST) && spa->spa_bootfs) {
3334 			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3335 
3336 			/*
3337 			 * We have to play games with the name since the
3338 			 * pool was opened as TRYIMPORT_NAME.
3339 			 */
3340 			if (dsl_dsobj_to_dsname(spa_name(spa),
3341 			    spa->spa_bootfs, tmpname) == 0) {
3342 				char *cp;
3343 				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3344 
3345 				cp = strchr(tmpname, '/');
3346 				if (cp == NULL) {
3347 					(void) strlcpy(dsname, tmpname,
3348 					    MAXPATHLEN);
3349 				} else {
3350 					(void) snprintf(dsname, MAXPATHLEN,
3351 					    "%s/%s", poolname, ++cp);
3352 				}
3353 				VERIFY(nvlist_add_string(config,
3354 				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3355 				kmem_free(dsname, MAXPATHLEN);
3356 			}
3357 			kmem_free(tmpname, MAXPATHLEN);
3358 		}
3359 
3360 		/*
3361 		 * Add the list of hot spares and level 2 cache devices.
3362 		 */
3363 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3364 		spa_add_spares(spa, config);
3365 		spa_add_l2cache(spa, config);
3366 		spa_config_exit(spa, SCL_CONFIG, FTAG);
3367 	}
3368 
3369 	spa_unload(spa);
3370 	spa_deactivate(spa);
3371 	spa_remove(spa);
3372 	mutex_exit(&spa_namespace_lock);
3373 
3374 	return (config);
3375 }
3376 
3377 /*
3378  * Pool export/destroy
3379  *
3380  * The act of destroying or exporting a pool is very simple.  We make sure there
3381  * is no more pending I/O and any references to the pool are gone.  Then, we
3382  * update the pool state and sync all the labels to disk, removing the
3383  * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3384  * we don't sync the labels or remove the configuration cache.
3385  */
3386 static int
3387 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3388     boolean_t force, boolean_t hardforce)
3389 {
3390 	spa_t *spa;
3391 
3392 	if (oldconfig)
3393 		*oldconfig = NULL;
3394 
3395 	if (!(spa_mode_global & FWRITE))
3396 		return (EROFS);
3397 
3398 	mutex_enter(&spa_namespace_lock);
3399 	if ((spa = spa_lookup(pool)) == NULL) {
3400 		mutex_exit(&spa_namespace_lock);
3401 		return (ENOENT);
3402 	}
3403 
3404 	/*
3405 	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3406 	 * reacquire the namespace lock, and see if we can export.
3407 	 */
3408 	spa_open_ref(spa, FTAG);
3409 	mutex_exit(&spa_namespace_lock);
3410 	spa_async_suspend(spa);
3411 	mutex_enter(&spa_namespace_lock);
3412 	spa_close(spa, FTAG);
3413 
3414 	/*
3415 	 * The pool will be in core if it's openable,
3416 	 * in which case we can modify its state.
3417 	 */
3418 	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3419 		/*
3420 		 * Objsets may be open only because they're dirty, so we
3421 		 * have to force it to sync before checking spa_refcnt.
3422 		 */
3423 		txg_wait_synced(spa->spa_dsl_pool, 0);
3424 
3425 		/*
3426 		 * A pool cannot be exported or destroyed if there are active
3427 		 * references.  If we are resetting a pool, allow references by
3428 		 * fault injection handlers.
3429 		 */
3430 		if (!spa_refcount_zero(spa) ||
3431 		    (spa->spa_inject_ref != 0 &&
3432 		    new_state != POOL_STATE_UNINITIALIZED)) {
3433 			spa_async_resume(spa);
3434 			mutex_exit(&spa_namespace_lock);
3435 			return (EBUSY);
3436 		}
3437 
3438 		/*
3439 		 * A pool cannot be exported if it has an active shared spare.
3440 		 * This is to prevent other pools stealing the active spare
3441 		 * from an exported pool. At user's own will, such pool can
3442 		 * be forcedly exported.
3443 		 */
3444 		if (!force && new_state == POOL_STATE_EXPORTED &&
3445 		    spa_has_active_shared_spare(spa)) {
3446 			spa_async_resume(spa);
3447 			mutex_exit(&spa_namespace_lock);
3448 			return (EXDEV);
3449 		}
3450 
3451 		/*
3452 		 * We want this to be reflected on every label,
3453 		 * so mark them all dirty.  spa_unload() will do the
3454 		 * final sync that pushes these changes out.
3455 		 */
3456 		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3457 			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3458 			spa->spa_state = new_state;
3459 			spa->spa_final_txg = spa_last_synced_txg(spa) + 1;
3460 			vdev_config_dirty(spa->spa_root_vdev);
3461 			spa_config_exit(spa, SCL_ALL, FTAG);
3462 		}
3463 	}
3464 
3465 	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
3466 
3467 	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3468 		spa_unload(spa);
3469 		spa_deactivate(spa);
3470 	}
3471 
3472 	if (oldconfig && spa->spa_config)
3473 		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3474 
3475 	if (new_state != POOL_STATE_UNINITIALIZED) {
3476 		if (!hardforce)
3477 			spa_config_sync(spa, B_TRUE, B_TRUE);
3478 		spa_remove(spa);
3479 	}
3480 	mutex_exit(&spa_namespace_lock);
3481 
3482 	return (0);
3483 }
3484 
3485 /*
3486  * Destroy a storage pool.
3487  */
3488 int
3489 spa_destroy(char *pool)
3490 {
3491 	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3492 	    B_FALSE, B_FALSE));
3493 }
3494 
3495 /*
3496  * Export a storage pool.
3497  */
3498 int
3499 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3500     boolean_t hardforce)
3501 {
3502 	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3503 	    force, hardforce));
3504 }
3505 
3506 /*
3507  * Similar to spa_export(), this unloads the spa_t without actually removing it
3508  * from the namespace in any way.
3509  */
3510 int
3511 spa_reset(char *pool)
3512 {
3513 	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3514 	    B_FALSE, B_FALSE));
3515 }
3516 
3517 /*
3518  * ==========================================================================
3519  * Device manipulation
3520  * ==========================================================================
3521  */
3522 
3523 /*
3524  * Add a device to a storage pool.
3525  */
3526 int
3527 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3528 {
3529 	uint64_t txg, id;
3530 	int error;
3531 	vdev_t *rvd = spa->spa_root_vdev;
3532 	vdev_t *vd, *tvd;
3533 	nvlist_t **spares, **l2cache;
3534 	uint_t nspares, nl2cache;
3535 
3536 	txg = spa_vdev_enter(spa);
3537 
3538 	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3539 	    VDEV_ALLOC_ADD)) != 0)
3540 		return (spa_vdev_exit(spa, NULL, txg, error));
3541 
3542 	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
3543 
3544 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3545 	    &nspares) != 0)
3546 		nspares = 0;
3547 
3548 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3549 	    &nl2cache) != 0)
3550 		nl2cache = 0;
3551 
3552 	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3553 		return (spa_vdev_exit(spa, vd, txg, EINVAL));
3554 
3555 	if (vd->vdev_children != 0 &&
3556 	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
3557 		return (spa_vdev_exit(spa, vd, txg, error));
3558 
3559 	/*
3560 	 * We must validate the spares and l2cache devices after checking the
3561 	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
3562 	 */
3563 	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3564 		return (spa_vdev_exit(spa, vd, txg, error));
3565 
3566 	/*
3567 	 * Transfer each new top-level vdev from vd to rvd.
3568 	 */
3569 	for (int c = 0; c < vd->vdev_children; c++) {
3570 
3571 		/*
3572 		 * Set the vdev id to the first hole, if one exists.
3573 		 */
3574 		for (id = 0; id < rvd->vdev_children; id++) {
3575 			if (rvd->vdev_child[id]->vdev_ishole) {
3576 				vdev_free(rvd->vdev_child[id]);
3577 				break;
3578 			}
3579 		}
3580 		tvd = vd->vdev_child[c];
3581 		vdev_remove_child(vd, tvd);
3582 		tvd->vdev_id = id;
3583 		vdev_add_child(rvd, tvd);
3584 		vdev_config_dirty(tvd);
3585 	}
3586 
3587 	if (nspares != 0) {
3588 		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3589 		    ZPOOL_CONFIG_SPARES);
3590 		spa_load_spares(spa);
3591 		spa->spa_spares.sav_sync = B_TRUE;
3592 	}
3593 
3594 	if (nl2cache != 0) {
3595 		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3596 		    ZPOOL_CONFIG_L2CACHE);
3597 		spa_load_l2cache(spa);
3598 		spa->spa_l2cache.sav_sync = B_TRUE;
3599 	}
3600 
3601 	/*
3602 	 * We have to be careful when adding new vdevs to an existing pool.
3603 	 * If other threads start allocating from these vdevs before we
3604 	 * sync the config cache, and we lose power, then upon reboot we may
3605 	 * fail to open the pool because there are DVAs that the config cache
3606 	 * can't translate.  Therefore, we first add the vdevs without
3607 	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3608 	 * and then let spa_config_update() initialize the new metaslabs.
3609 	 *
3610 	 * spa_load() checks for added-but-not-initialized vdevs, so that
3611 	 * if we lose power at any point in this sequence, the remaining
3612 	 * steps will be completed the next time we load the pool.
3613 	 */
3614 	(void) spa_vdev_exit(spa, vd, txg, 0);
3615 
3616 	mutex_enter(&spa_namespace_lock);
3617 	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3618 	mutex_exit(&spa_namespace_lock);
3619 
3620 	return (0);
3621 }
3622 
3623 /*
3624  * Attach a device to a mirror.  The arguments are the path to any device
3625  * in the mirror, and the nvroot for the new device.  If the path specifies
3626  * a device that is not mirrored, we automatically insert the mirror vdev.
3627  *
3628  * If 'replacing' is specified, the new device is intended to replace the
3629  * existing device; in this case the two devices are made into their own
3630  * mirror using the 'replacing' vdev, which is functionally identical to
3631  * the mirror vdev (it actually reuses all the same ops) but has a few
3632  * extra rules: you can't attach to it after it's been created, and upon
3633  * completion of resilvering, the first disk (the one being replaced)
3634  * is automatically detached.
3635  */
3636 int
3637 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3638 {
3639 	uint64_t txg, open_txg;
3640 	vdev_t *rvd = spa->spa_root_vdev;
3641 	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3642 	vdev_ops_t *pvops;
3643 	char *oldvdpath, *newvdpath;
3644 	int newvd_isspare;
3645 	int error;
3646 
3647 	txg = spa_vdev_enter(spa);
3648 
3649 	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3650 
3651 	if (oldvd == NULL)
3652 		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3653 
3654 	if (!oldvd->vdev_ops->vdev_op_leaf)
3655 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3656 
3657 	pvd = oldvd->vdev_parent;
3658 
3659 	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3660 	    VDEV_ALLOC_ADD)) != 0)
3661 		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3662 
3663 	if (newrootvd->vdev_children != 1)
3664 		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3665 
3666 	newvd = newrootvd->vdev_child[0];
3667 
3668 	if (!newvd->vdev_ops->vdev_op_leaf)
3669 		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3670 
3671 	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3672 		return (spa_vdev_exit(spa, newrootvd, txg, error));
3673 
3674 	/*
3675 	 * Spares can't replace logs
3676 	 */
3677 	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3678 		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3679 
3680 	if (!replacing) {
3681 		/*
3682 		 * For attach, the only allowable parent is a mirror or the root
3683 		 * vdev.
3684 		 */
3685 		if (pvd->vdev_ops != &vdev_mirror_ops &&
3686 		    pvd->vdev_ops != &vdev_root_ops)
3687 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3688 
3689 		pvops = &vdev_mirror_ops;
3690 	} else {
3691 		/*
3692 		 * Active hot spares can only be replaced by inactive hot
3693 		 * spares.
3694 		 */
3695 		if (pvd->vdev_ops == &vdev_spare_ops &&
3696 		    pvd->vdev_child[1] == oldvd &&
3697 		    !spa_has_spare(spa, newvd->vdev_guid))
3698 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3699 
3700 		/*
3701 		 * If the source is a hot spare, and the parent isn't already a
3702 		 * spare, then we want to create a new hot spare.  Otherwise, we
3703 		 * want to create a replacing vdev.  The user is not allowed to
3704 		 * attach to a spared vdev child unless the 'isspare' state is
3705 		 * the same (spare replaces spare, non-spare replaces
3706 		 * non-spare).
3707 		 */
3708 		if (pvd->vdev_ops == &vdev_replacing_ops)
3709 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3710 		else if (pvd->vdev_ops == &vdev_spare_ops &&
3711 		    newvd->vdev_isspare != oldvd->vdev_isspare)
3712 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3713 		else if (pvd->vdev_ops != &vdev_spare_ops &&
3714 		    newvd->vdev_isspare)
3715 			pvops = &vdev_spare_ops;
3716 		else
3717 			pvops = &vdev_replacing_ops;
3718 	}
3719 
3720 	/*
3721 	 * Make sure the new device is big enough.
3722 	 */
3723 	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3724 		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3725 
3726 	/*
3727 	 * The new device cannot have a higher alignment requirement
3728 	 * than the top-level vdev.
3729 	 */
3730 	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3731 		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3732 
3733 	/*
3734 	 * If this is an in-place replacement, update oldvd's path and devid
3735 	 * to make it distinguishable from newvd, and unopenable from now on.
3736 	 */
3737 	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3738 		spa_strfree(oldvd->vdev_path);
3739 		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3740 		    KM_SLEEP);
3741 		(void) sprintf(oldvd->vdev_path, "%s/%s",
3742 		    newvd->vdev_path, "old");
3743 		if (oldvd->vdev_devid != NULL) {
3744 			spa_strfree(oldvd->vdev_devid);
3745 			oldvd->vdev_devid = NULL;
3746 		}
3747 	}
3748 
3749 	/*
3750 	 * If the parent is not a mirror, or if we're replacing, insert the new
3751 	 * mirror/replacing/spare vdev above oldvd.
3752 	 */
3753 	if (pvd->vdev_ops != pvops)
3754 		pvd = vdev_add_parent(oldvd, pvops);
3755 
3756 	ASSERT(pvd->vdev_top->vdev_parent == rvd);
3757 	ASSERT(pvd->vdev_ops == pvops);
3758 	ASSERT(oldvd->vdev_parent == pvd);
3759 
3760 	/*
3761 	 * Extract the new device from its root and add it to pvd.
3762 	 */
3763 	vdev_remove_child(newrootvd, newvd);
3764 	newvd->vdev_id = pvd->vdev_children;
3765 	newvd->vdev_crtxg = oldvd->vdev_crtxg;
3766 	vdev_add_child(pvd, newvd);
3767 
3768 	tvd = newvd->vdev_top;
3769 	ASSERT(pvd->vdev_top == tvd);
3770 	ASSERT(tvd->vdev_parent == rvd);
3771 
3772 	vdev_config_dirty(tvd);
3773 
3774 	/*
3775 	 * Set newvd's DTL to [TXG_INITIAL, open_txg].  It will propagate
3776 	 * upward when spa_vdev_exit() calls vdev_dtl_reassess().
3777 	 */
3778 	open_txg = txg + TXG_CONCURRENT_STATES - 1;
3779 
3780 	vdev_dtl_dirty(newvd, DTL_MISSING,
3781 	    TXG_INITIAL, open_txg - TXG_INITIAL + 1);
3782 
3783 	if (newvd->vdev_isspare) {
3784 		spa_spare_activate(newvd);
3785 		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
3786 	}
3787 
3788 	oldvdpath = spa_strdup(oldvd->vdev_path);
3789 	newvdpath = spa_strdup(newvd->vdev_path);
3790 	newvd_isspare = newvd->vdev_isspare;
3791 
3792 	/*
3793 	 * Mark newvd's DTL dirty in this txg.
3794 	 */
3795 	vdev_dirty(tvd, VDD_DTL, newvd, txg);
3796 
3797 	(void) spa_vdev_exit(spa, newrootvd, open_txg, 0);
3798 
3799 	spa_history_internal_log(LOG_POOL_VDEV_ATTACH, spa, NULL,
3800 	    CRED(),  "%s vdev=%s %s vdev=%s",
3801 	    replacing && newvd_isspare ? "spare in" :
3802 	    replacing ? "replace" : "attach", newvdpath,
3803 	    replacing ? "for" : "to", oldvdpath);
3804 
3805 	spa_strfree(oldvdpath);
3806 	spa_strfree(newvdpath);
3807 
3808 	/*
3809 	 * Kick off a resilver to update newvd.
3810 	 */
3811 	VERIFY3U(spa_scrub(spa, POOL_SCRUB_RESILVER), ==, 0);
3812 
3813 	return (0);
3814 }
3815 
3816 /*
3817  * Detach a device from a mirror or replacing vdev.
3818  * If 'replace_done' is specified, only detach if the parent
3819  * is a replacing vdev.
3820  */
3821 int
3822 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
3823 {
3824 	uint64_t txg;
3825 	int error;
3826 	vdev_t *rvd = spa->spa_root_vdev;
3827 	vdev_t *vd, *pvd, *cvd, *tvd;
3828 	boolean_t unspare = B_FALSE;
3829 	uint64_t unspare_guid;
3830 	size_t len;
3831 	char *vdpath;
3832 
3833 	txg = spa_vdev_enter(spa);
3834 
3835 	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
3836 
3837 	if (vd == NULL)
3838 		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3839 
3840 	if (!vd->vdev_ops->vdev_op_leaf)
3841 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3842 
3843 	pvd = vd->vdev_parent;
3844 
3845 	/*
3846 	 * If the parent/child relationship is not as expected, don't do it.
3847 	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
3848 	 * vdev that's replacing B with C.  The user's intent in replacing
3849 	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
3850 	 * the replace by detaching C, the expected behavior is to end up
3851 	 * M(A,B).  But suppose that right after deciding to detach C,
3852 	 * the replacement of B completes.  We would have M(A,C), and then
3853 	 * ask to detach C, which would leave us with just A -- not what
3854 	 * the user wanted.  To prevent this, we make sure that the
3855 	 * parent/child relationship hasn't changed -- in this example,
3856 	 * that C's parent is still the replacing vdev R.
3857 	 */
3858 	if (pvd->vdev_guid != pguid && pguid != 0)
3859 		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3860 
3861 	/*
3862 	 * If replace_done is specified, only remove this device if it's
3863 	 * the first child of a replacing vdev.  For the 'spare' vdev, either
3864 	 * disk can be removed.
3865 	 */
3866 	if (replace_done) {
3867 		if (pvd->vdev_ops == &vdev_replacing_ops) {
3868 			if (vd->vdev_id != 0)
3869 				return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3870 		} else if (pvd->vdev_ops != &vdev_spare_ops) {
3871 			return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3872 		}
3873 	}
3874 
3875 	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
3876 	    spa_version(spa) >= SPA_VERSION_SPARES);
3877 
3878 	/*
3879 	 * Only mirror, replacing, and spare vdevs support detach.
3880 	 */
3881 	if (pvd->vdev_ops != &vdev_replacing_ops &&
3882 	    pvd->vdev_ops != &vdev_mirror_ops &&
3883 	    pvd->vdev_ops != &vdev_spare_ops)
3884 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3885 
3886 	/*
3887 	 * If this device has the only valid copy of some data,
3888 	 * we cannot safely detach it.
3889 	 */
3890 	if (vdev_dtl_required(vd))
3891 		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
3892 
3893 	ASSERT(pvd->vdev_children >= 2);
3894 
3895 	/*
3896 	 * If we are detaching the second disk from a replacing vdev, then
3897 	 * check to see if we changed the original vdev's path to have "/old"
3898 	 * at the end in spa_vdev_attach().  If so, undo that change now.
3899 	 */
3900 	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id == 1 &&
3901 	    pvd->vdev_child[0]->vdev_path != NULL &&
3902 	    pvd->vdev_child[1]->vdev_path != NULL) {
3903 		ASSERT(pvd->vdev_child[1] == vd);
3904 		cvd = pvd->vdev_child[0];
3905 		len = strlen(vd->vdev_path);
3906 		if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
3907 		    strcmp(cvd->vdev_path + len, "/old") == 0) {
3908 			spa_strfree(cvd->vdev_path);
3909 			cvd->vdev_path = spa_strdup(vd->vdev_path);
3910 		}
3911 	}
3912 
3913 	/*
3914 	 * If we are detaching the original disk from a spare, then it implies
3915 	 * that the spare should become a real disk, and be removed from the
3916 	 * active spare list for the pool.
3917 	 */
3918 	if (pvd->vdev_ops == &vdev_spare_ops &&
3919 	    vd->vdev_id == 0 && pvd->vdev_child[1]->vdev_isspare)
3920 		unspare = B_TRUE;
3921 
3922 	/*
3923 	 * Erase the disk labels so the disk can be used for other things.
3924 	 * This must be done after all other error cases are handled,
3925 	 * but before we disembowel vd (so we can still do I/O to it).
3926 	 * But if we can't do it, don't treat the error as fatal --
3927 	 * it may be that the unwritability of the disk is the reason
3928 	 * it's being detached!
3929 	 */
3930 	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
3931 
3932 	/*
3933 	 * Remove vd from its parent and compact the parent's children.
3934 	 */
3935 	vdev_remove_child(pvd, vd);
3936 	vdev_compact_children(pvd);
3937 
3938 	/*
3939 	 * Remember one of the remaining children so we can get tvd below.
3940 	 */
3941 	cvd = pvd->vdev_child[0];
3942 
3943 	/*
3944 	 * If we need to remove the remaining child from the list of hot spares,
3945 	 * do it now, marking the vdev as no longer a spare in the process.
3946 	 * We must do this before vdev_remove_parent(), because that can
3947 	 * change the GUID if it creates a new toplevel GUID.  For a similar
3948 	 * reason, we must remove the spare now, in the same txg as the detach;
3949 	 * otherwise someone could attach a new sibling, change the GUID, and
3950 	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
3951 	 */
3952 	if (unspare) {
3953 		ASSERT(cvd->vdev_isspare);
3954 		spa_spare_remove(cvd);
3955 		unspare_guid = cvd->vdev_guid;
3956 		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
3957 	}
3958 
3959 	/*
3960 	 * If the parent mirror/replacing vdev only has one child,
3961 	 * the parent is no longer needed.  Remove it from the tree.
3962 	 */
3963 	if (pvd->vdev_children == 1)
3964 		vdev_remove_parent(cvd);
3965 
3966 	/*
3967 	 * We don't set tvd until now because the parent we just removed
3968 	 * may have been the previous top-level vdev.
3969 	 */
3970 	tvd = cvd->vdev_top;
3971 	ASSERT(tvd->vdev_parent == rvd);
3972 
3973 	/*
3974 	 * Reevaluate the parent vdev state.
3975 	 */
3976 	vdev_propagate_state(cvd);
3977 
3978 	/*
3979 	 * If the 'autoexpand' property is set on the pool then automatically
3980 	 * try to expand the size of the pool. For example if the device we
3981 	 * just detached was smaller than the others, it may be possible to
3982 	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
3983 	 * first so that we can obtain the updated sizes of the leaf vdevs.
3984 	 */
3985 	if (spa->spa_autoexpand) {
3986 		vdev_reopen(tvd);
3987 		vdev_expand(tvd, txg);
3988 	}
3989 
3990 	vdev_config_dirty(tvd);
3991 
3992 	/*
3993 	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
3994 	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
3995 	 * But first make sure we're not on any *other* txg's DTL list, to
3996 	 * prevent vd from being accessed after it's freed.
3997 	 */
3998 	vdpath = spa_strdup(vd->vdev_path);
3999 	for (int t = 0; t < TXG_SIZE; t++)
4000 		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4001 	vd->vdev_detached = B_TRUE;
4002 	vdev_dirty(tvd, VDD_DTL, vd, txg);
4003 
4004 	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4005 
4006 	error = spa_vdev_exit(spa, vd, txg, 0);
4007 
4008 	spa_history_internal_log(LOG_POOL_VDEV_DETACH, spa, NULL, CRED(),
4009 	    "vdev=%s", vdpath);
4010 	spa_strfree(vdpath);
4011 
4012 	/*
4013 	 * If this was the removal of the original device in a hot spare vdev,
4014 	 * then we want to go through and remove the device from the hot spare
4015 	 * list of every other pool.
4016 	 */
4017 	if (unspare) {
4018 		spa_t *myspa = spa;
4019 		spa = NULL;
4020 		mutex_enter(&spa_namespace_lock);
4021 		while ((spa = spa_next(spa)) != NULL) {
4022 			if (spa->spa_state != POOL_STATE_ACTIVE)
4023 				continue;
4024 			if (spa == myspa)
4025 				continue;
4026 			spa_open_ref(spa, FTAG);
4027 			mutex_exit(&spa_namespace_lock);
4028 			(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4029 			mutex_enter(&spa_namespace_lock);
4030 			spa_close(spa, FTAG);
4031 		}
4032 		mutex_exit(&spa_namespace_lock);
4033 	}
4034 
4035 	return (error);
4036 }
4037 
4038 /*
4039  * Split a set of devices from their mirrors, and create a new pool from them.
4040  */
4041 int
4042 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4043     nvlist_t *props, boolean_t exp)
4044 {
4045 	int error = 0;
4046 	uint64_t txg, *glist;
4047 	spa_t *newspa;
4048 	uint_t c, children, lastlog;
4049 	nvlist_t **child, *nvl, *tmp;
4050 	dmu_tx_t *tx;
4051 	char *altroot = NULL;
4052 	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
4053 	boolean_t activate_slog;
4054 
4055 	if (!spa_writeable(spa))
4056 		return (EROFS);
4057 
4058 	txg = spa_vdev_enter(spa);
4059 
4060 	/* clear the log and flush everything up to now */
4061 	activate_slog = spa_passivate_log(spa);
4062 	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4063 	error = spa_offline_log(spa);
4064 	txg = spa_vdev_config_enter(spa);
4065 
4066 	if (activate_slog)
4067 		spa_activate_log(spa);
4068 
4069 	if (error != 0)
4070 		return (spa_vdev_exit(spa, NULL, txg, error));
4071 
4072 	/* check new spa name before going any further */
4073 	if (spa_lookup(newname) != NULL)
4074 		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4075 
4076 	/*
4077 	 * scan through all the children to ensure they're all mirrors
4078 	 */
4079 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4080 	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4081 	    &children) != 0)
4082 		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4083 
4084 	/* first, check to ensure we've got the right child count */
4085 	rvd = spa->spa_root_vdev;
4086 	lastlog = 0;
4087 	for (c = 0; c < rvd->vdev_children; c++) {
4088 		vdev_t *vd = rvd->vdev_child[c];
4089 
4090 		/* don't count the holes & logs as children */
4091 		if (vd->vdev_islog || vd->vdev_ishole) {
4092 			if (lastlog == 0)
4093 				lastlog = c;
4094 			continue;
4095 		}
4096 
4097 		lastlog = 0;
4098 	}
4099 	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4100 		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4101 
4102 	/* next, ensure no spare or cache devices are part of the split */
4103 	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4104 	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4105 		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4106 
4107 	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4108 	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4109 
4110 	/* then, loop over each vdev and validate it */
4111 	for (c = 0; c < children; c++) {
4112 		uint64_t is_hole = 0;
4113 
4114 		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4115 		    &is_hole);
4116 
4117 		if (is_hole != 0) {
4118 			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4119 			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4120 				continue;
4121 			} else {
4122 				error = EINVAL;
4123 				break;
4124 			}
4125 		}
4126 
4127 		/* which disk is going to be split? */
4128 		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4129 		    &glist[c]) != 0) {
4130 			error = EINVAL;
4131 			break;
4132 		}
4133 
4134 		/* look it up in the spa */
4135 		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4136 		if (vml[c] == NULL) {
4137 			error = ENODEV;
4138 			break;
4139 		}
4140 
4141 		/* make sure there's nothing stopping the split */
4142 		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4143 		    vml[c]->vdev_islog ||
4144 		    vml[c]->vdev_ishole ||
4145 		    vml[c]->vdev_isspare ||
4146 		    vml[c]->vdev_isl2cache ||
4147 		    !vdev_writeable(vml[c]) ||
4148 		    vml[c]->vdev_children != 0 ||
4149 		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4150 		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4151 			error = EINVAL;
4152 			break;
4153 		}
4154 
4155 		if (vdev_dtl_required(vml[c])) {
4156 			error = EBUSY;
4157 			break;
4158 		}
4159 
4160 		/* we need certain info from the top level */
4161 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4162 		    vml[c]->vdev_top->vdev_ms_array) == 0);
4163 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4164 		    vml[c]->vdev_top->vdev_ms_shift) == 0);
4165 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4166 		    vml[c]->vdev_top->vdev_asize) == 0);
4167 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4168 		    vml[c]->vdev_top->vdev_ashift) == 0);
4169 	}
4170 
4171 	if (error != 0) {
4172 		kmem_free(vml, children * sizeof (vdev_t *));
4173 		kmem_free(glist, children * sizeof (uint64_t));
4174 		return (spa_vdev_exit(spa, NULL, txg, error));
4175 	}
4176 
4177 	/* stop writers from using the disks */
4178 	for (c = 0; c < children; c++) {
4179 		if (vml[c] != NULL)
4180 			vml[c]->vdev_offline = B_TRUE;
4181 	}
4182 	vdev_reopen(spa->spa_root_vdev);
4183 
4184 	/*
4185 	 * Temporarily record the splitting vdevs in the spa config.  This
4186 	 * will disappear once the config is regenerated.
4187 	 */
4188 	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4189 	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4190 	    glist, children) == 0);
4191 	kmem_free(glist, children * sizeof (uint64_t));
4192 
4193 	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4194 	    nvl) == 0);
4195 	spa->spa_config_splitting = nvl;
4196 	vdev_config_dirty(spa->spa_root_vdev);
4197 
4198 	/* configure and create the new pool */
4199 	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4200 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4201 	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4202 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4203 	    spa_version(spa)) == 0);
4204 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4205 	    spa->spa_config_txg) == 0);
4206 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4207 	    spa_generate_guid(NULL)) == 0);
4208 	(void) nvlist_lookup_string(props,
4209 	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4210 
4211 	/* add the new pool to the namespace */
4212 	newspa = spa_add(newname, config, altroot);
4213 	newspa->spa_config_txg = spa->spa_config_txg;
4214 	spa_set_log_state(newspa, SPA_LOG_CLEAR);
4215 
4216 	/* release the spa config lock, retaining the namespace lock */
4217 	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4218 
4219 	if (zio_injection_enabled)
4220 		zio_handle_panic_injection(spa, FTAG, 1);
4221 
4222 	spa_activate(newspa, spa_mode_global);
4223 	spa_async_suspend(newspa);
4224 
4225 	/* create the new pool from the disks of the original pool */
4226 	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4227 	if (error)
4228 		goto out;
4229 
4230 	/* if that worked, generate a real config for the new pool */
4231 	if (newspa->spa_root_vdev != NULL) {
4232 		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4233 		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4234 		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4235 		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4236 		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4237 		    B_TRUE));
4238 	}
4239 
4240 	/* set the props */
4241 	if (props != NULL) {
4242 		spa_configfile_set(newspa, props, B_FALSE);
4243 		error = spa_prop_set(newspa, props);
4244 		if (error)
4245 			goto out;
4246 	}
4247 
4248 	/* flush everything */
4249 	txg = spa_vdev_config_enter(newspa);
4250 	vdev_config_dirty(newspa->spa_root_vdev);
4251 	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4252 
4253 	if (zio_injection_enabled)
4254 		zio_handle_panic_injection(spa, FTAG, 2);
4255 
4256 	spa_async_resume(newspa);
4257 
4258 	/* finally, update the original pool's config */
4259 	txg = spa_vdev_config_enter(spa);
4260 	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4261 	error = dmu_tx_assign(tx, TXG_WAIT);
4262 	if (error != 0)
4263 		dmu_tx_abort(tx);
4264 	for (c = 0; c < children; c++) {
4265 		if (vml[c] != NULL) {
4266 			vdev_split(vml[c]);
4267 			if (error == 0)
4268 				spa_history_internal_log(LOG_POOL_VDEV_DETACH,
4269 				    spa, tx, CRED(), "vdev=%s",
4270 				    vml[c]->vdev_path);
4271 			vdev_free(vml[c]);
4272 		}
4273 	}
4274 	vdev_config_dirty(spa->spa_root_vdev);
4275 	spa->spa_config_splitting = NULL;
4276 	nvlist_free(nvl);
4277 	if (error == 0)
4278 		dmu_tx_commit(tx);
4279 	(void) spa_vdev_exit(spa, NULL, txg, 0);
4280 
4281 	if (zio_injection_enabled)
4282 		zio_handle_panic_injection(spa, FTAG, 3);
4283 
4284 	/* split is complete; log a history record */
4285 	spa_history_internal_log(LOG_POOL_SPLIT, newspa, NULL, CRED(),
4286 	    "split new pool %s from pool %s", newname, spa_name(spa));
4287 
4288 	kmem_free(vml, children * sizeof (vdev_t *));
4289 
4290 	/* if we're not going to mount the filesystems in userland, export */
4291 	if (exp)
4292 		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4293 		    B_FALSE, B_FALSE);
4294 
4295 	return (error);
4296 
4297 out:
4298 	spa_unload(newspa);
4299 	spa_deactivate(newspa);
4300 	spa_remove(newspa);
4301 
4302 	txg = spa_vdev_config_enter(spa);
4303 	nvlist_free(spa->spa_config_splitting);
4304 	spa->spa_config_splitting = NULL;
4305 	(void) spa_vdev_exit(spa, NULL, txg, error);
4306 
4307 	kmem_free(vml, children * sizeof (vdev_t *));
4308 	return (error);
4309 }
4310 
4311 static nvlist_t *
4312 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4313 {
4314 	for (int i = 0; i < count; i++) {
4315 		uint64_t guid;
4316 
4317 		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4318 		    &guid) == 0);
4319 
4320 		if (guid == target_guid)
4321 			return (nvpp[i]);
4322 	}
4323 
4324 	return (NULL);
4325 }
4326 
4327 static void
4328 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4329 	nvlist_t *dev_to_remove)
4330 {
4331 	nvlist_t **newdev = NULL;
4332 
4333 	if (count > 1)
4334 		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4335 
4336 	for (int i = 0, j = 0; i < count; i++) {
4337 		if (dev[i] == dev_to_remove)
4338 			continue;
4339 		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4340 	}
4341 
4342 	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4343 	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4344 
4345 	for (int i = 0; i < count - 1; i++)
4346 		nvlist_free(newdev[i]);
4347 
4348 	if (count > 1)
4349 		kmem_free(newdev, (count - 1) * sizeof (void *));
4350 }
4351 
4352 /*
4353  * Removing a device from the vdev namespace requires several steps
4354  * and can take a significant amount of time.  As a result we use
4355  * the spa_vdev_config_[enter/exit] functions which allow us to
4356  * grab and release the spa_config_lock while still holding the namespace
4357  * lock.  During each step the configuration is synced out.
4358  */
4359 
4360 /*
4361  * Evacuate the device.
4362  */
4363 int
4364 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4365 {
4366 	int error = 0;
4367 	uint64_t txg;
4368 
4369 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
4370 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4371 	ASSERT(vd == vd->vdev_top);
4372 
4373 	/*
4374 	 * Evacuate the device.  We don't hold the config lock as writer
4375 	 * since we need to do I/O but we do keep the
4376 	 * spa_namespace_lock held.  Once this completes the device
4377 	 * should no longer have any blocks allocated on it.
4378 	 */
4379 	if (vd->vdev_islog) {
4380 		error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
4381 		    NULL, DS_FIND_CHILDREN);
4382 	} else {
4383 		error = ENOTSUP;	/* until we have bp rewrite */
4384 	}
4385 
4386 	txg_wait_synced(spa_get_dsl(spa), 0);
4387 
4388 	if (error)
4389 		return (error);
4390 
4391 	/*
4392 	 * The evacuation succeeded.  Remove any remaining MOS metadata
4393 	 * associated with this vdev, and wait for these changes to sync.
4394 	 */
4395 	txg = spa_vdev_config_enter(spa);
4396 	vd->vdev_removing = B_TRUE;
4397 	vdev_dirty(vd, 0, NULL, txg);
4398 	vdev_config_dirty(vd);
4399 	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4400 
4401 	return (0);
4402 }
4403 
4404 /*
4405  * Complete the removal by cleaning up the namespace.
4406  */
4407 void
4408 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
4409 {
4410 	vdev_t *rvd = spa->spa_root_vdev;
4411 	uint64_t id = vd->vdev_id;
4412 	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
4413 
4414 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
4415 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4416 	ASSERT(vd == vd->vdev_top);
4417 
4418 	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4419 
4420 	if (list_link_active(&vd->vdev_state_dirty_node))
4421 		vdev_state_clean(vd);
4422 	if (list_link_active(&vd->vdev_config_dirty_node))
4423 		vdev_config_clean(vd);
4424 
4425 	vdev_free(vd);
4426 
4427 	if (last_vdev) {
4428 		vdev_compact_children(rvd);
4429 	} else {
4430 		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
4431 		vdev_add_child(rvd, vd);
4432 	}
4433 	vdev_config_dirty(rvd);
4434 
4435 	/*
4436 	 * Reassess the health of our root vdev.
4437 	 */
4438 	vdev_reopen(rvd);
4439 }
4440 
4441 /*
4442  * Remove a device from the pool.  Currently, this supports removing only hot
4443  * spares, slogs, and level 2 ARC devices.
4444  */
4445 int
4446 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
4447 {
4448 	vdev_t *vd;
4449 	metaslab_group_t *mg;
4450 	nvlist_t **spares, **l2cache, *nv;
4451 	uint64_t txg = 0;
4452 	uint_t nspares, nl2cache;
4453 	int error = 0;
4454 	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
4455 
4456 	if (!locked)
4457 		txg = spa_vdev_enter(spa);
4458 
4459 	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4460 
4461 	if (spa->spa_spares.sav_vdevs != NULL &&
4462 	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4463 	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
4464 	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
4465 		/*
4466 		 * Only remove the hot spare if it's not currently in use
4467 		 * in this pool.
4468 		 */
4469 		if (vd == NULL || unspare) {
4470 			spa_vdev_remove_aux(spa->spa_spares.sav_config,
4471 			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
4472 			spa_load_spares(spa);
4473 			spa->spa_spares.sav_sync = B_TRUE;
4474 		} else {
4475 			error = EBUSY;
4476 		}
4477 	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
4478 	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4479 	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
4480 	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
4481 		/*
4482 		 * Cache devices can always be removed.
4483 		 */
4484 		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
4485 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
4486 		spa_load_l2cache(spa);
4487 		spa->spa_l2cache.sav_sync = B_TRUE;
4488 	} else if (vd != NULL && vd->vdev_islog) {
4489 		ASSERT(!locked);
4490 		ASSERT(vd == vd->vdev_top);
4491 
4492 		/*
4493 		 * XXX - Once we have bp-rewrite this should
4494 		 * become the common case.
4495 		 */
4496 
4497 		mg = vd->vdev_mg;
4498 
4499 		/*
4500 		 * Stop allocating from this vdev.
4501 		 */
4502 		metaslab_group_passivate(mg);
4503 
4504 		/*
4505 		 * Wait for the youngest allocations and frees to sync,
4506 		 * and then wait for the deferral of those frees to finish.
4507 		 */
4508 		spa_vdev_config_exit(spa, NULL,
4509 		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
4510 
4511 		/*
4512 		 * Attempt to evacuate the vdev.
4513 		 */
4514 		error = spa_vdev_remove_evacuate(spa, vd);
4515 
4516 		txg = spa_vdev_config_enter(spa);
4517 
4518 		/*
4519 		 * If we couldn't evacuate the vdev, unwind.
4520 		 */
4521 		if (error) {
4522 			metaslab_group_activate(mg);
4523 			return (spa_vdev_exit(spa, NULL, txg, error));
4524 		}
4525 
4526 		/*
4527 		 * Clean up the vdev namespace.
4528 		 */
4529 		spa_vdev_remove_from_namespace(spa, vd);
4530 
4531 	} else if (vd != NULL) {
4532 		/*
4533 		 * Normal vdevs cannot be removed (yet).
4534 		 */
4535 		error = ENOTSUP;
4536 	} else {
4537 		/*
4538 		 * There is no vdev of any kind with the specified guid.
4539 		 */
4540 		error = ENOENT;
4541 	}
4542 
4543 	if (!locked)
4544 		return (spa_vdev_exit(spa, NULL, txg, error));
4545 
4546 	return (error);
4547 }
4548 
4549 /*
4550  * Find any device that's done replacing, or a vdev marked 'unspare' that's
4551  * current spared, so we can detach it.
4552  */
4553 static vdev_t *
4554 spa_vdev_resilver_done_hunt(vdev_t *vd)
4555 {
4556 	vdev_t *newvd, *oldvd;
4557 
4558 	for (int c = 0; c < vd->vdev_children; c++) {
4559 		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
4560 		if (oldvd != NULL)
4561 			return (oldvd);
4562 	}
4563 
4564 	/*
4565 	 * Check for a completed replacement.
4566 	 */
4567 	if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) {
4568 		oldvd = vd->vdev_child[0];
4569 		newvd = vd->vdev_child[1];
4570 
4571 		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4572 		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4573 		    !vdev_dtl_required(oldvd))
4574 			return (oldvd);
4575 	}
4576 
4577 	/*
4578 	 * Check for a completed resilver with the 'unspare' flag set.
4579 	 */
4580 	if (vd->vdev_ops == &vdev_spare_ops && vd->vdev_children == 2) {
4581 		newvd = vd->vdev_child[0];
4582 		oldvd = vd->vdev_child[1];
4583 
4584 		if (newvd->vdev_unspare &&
4585 		    vdev_dtl_empty(newvd, DTL_MISSING) &&
4586 		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4587 		    !vdev_dtl_required(oldvd)) {
4588 			newvd->vdev_unspare = 0;
4589 			return (oldvd);
4590 		}
4591 	}
4592 
4593 	return (NULL);
4594 }
4595 
4596 static void
4597 spa_vdev_resilver_done(spa_t *spa)
4598 {
4599 	vdev_t *vd, *pvd, *ppvd;
4600 	uint64_t guid, sguid, pguid, ppguid;
4601 
4602 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4603 
4604 	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4605 		pvd = vd->vdev_parent;
4606 		ppvd = pvd->vdev_parent;
4607 		guid = vd->vdev_guid;
4608 		pguid = pvd->vdev_guid;
4609 		ppguid = ppvd->vdev_guid;
4610 		sguid = 0;
4611 		/*
4612 		 * If we have just finished replacing a hot spared device, then
4613 		 * we need to detach the parent's first child (the original hot
4614 		 * spare) as well.
4615 		 */
4616 		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0) {
4617 			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4618 			ASSERT(ppvd->vdev_children == 2);
4619 			sguid = ppvd->vdev_child[1]->vdev_guid;
4620 		}
4621 		spa_config_exit(spa, SCL_ALL, FTAG);
4622 		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4623 			return;
4624 		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4625 			return;
4626 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4627 	}
4628 
4629 	spa_config_exit(spa, SCL_ALL, FTAG);
4630 }
4631 
4632 /*
4633  * Update the stored path or FRU for this vdev.
4634  */
4635 int
4636 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4637     boolean_t ispath)
4638 {
4639 	vdev_t *vd;
4640 	boolean_t sync = B_FALSE;
4641 
4642 	spa_vdev_state_enter(spa, SCL_ALL);
4643 
4644 	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4645 		return (spa_vdev_state_exit(spa, NULL, ENOENT));
4646 
4647 	if (!vd->vdev_ops->vdev_op_leaf)
4648 		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
4649 
4650 	if (ispath) {
4651 		if (strcmp(value, vd->vdev_path) != 0) {
4652 			spa_strfree(vd->vdev_path);
4653 			vd->vdev_path = spa_strdup(value);
4654 			sync = B_TRUE;
4655 		}
4656 	} else {
4657 		if (vd->vdev_fru == NULL) {
4658 			vd->vdev_fru = spa_strdup(value);
4659 			sync = B_TRUE;
4660 		} else if (strcmp(value, vd->vdev_fru) != 0) {
4661 			spa_strfree(vd->vdev_fru);
4662 			vd->vdev_fru = spa_strdup(value);
4663 			sync = B_TRUE;
4664 		}
4665 	}
4666 
4667 	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
4668 }
4669 
4670 int
4671 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
4672 {
4673 	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
4674 }
4675 
4676 int
4677 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
4678 {
4679 	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
4680 }
4681 
4682 /*
4683  * ==========================================================================
4684  * SPA Scrubbing
4685  * ==========================================================================
4686  */
4687 
4688 int
4689 spa_scrub(spa_t *spa, pool_scrub_type_t type)
4690 {
4691 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4692 
4693 	if ((uint_t)type >= POOL_SCRUB_TYPES)
4694 		return (ENOTSUP);
4695 
4696 	/*
4697 	 * If a resilver was requested, but there is no DTL on a
4698 	 * writeable leaf device, we have nothing to do.
4699 	 */
4700 	if (type == POOL_SCRUB_RESILVER &&
4701 	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
4702 		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
4703 		return (0);
4704 	}
4705 
4706 	if (type == POOL_SCRUB_EVERYTHING &&
4707 	    spa->spa_dsl_pool->dp_scrub_func != SCRUB_FUNC_NONE &&
4708 	    spa->spa_dsl_pool->dp_scrub_isresilver)
4709 		return (EBUSY);
4710 
4711 	if (type == POOL_SCRUB_EVERYTHING || type == POOL_SCRUB_RESILVER) {
4712 		return (dsl_pool_scrub_clean(spa->spa_dsl_pool));
4713 	} else if (type == POOL_SCRUB_NONE) {
4714 		return (dsl_pool_scrub_cancel(spa->spa_dsl_pool));
4715 	} else {
4716 		return (EINVAL);
4717 	}
4718 }
4719 
4720 /*
4721  * ==========================================================================
4722  * SPA async task processing
4723  * ==========================================================================
4724  */
4725 
4726 static void
4727 spa_async_remove(spa_t *spa, vdev_t *vd)
4728 {
4729 	if (vd->vdev_remove_wanted) {
4730 		vd->vdev_remove_wanted = 0;
4731 		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
4732 
4733 		/*
4734 		 * We want to clear the stats, but we don't want to do a full
4735 		 * vdev_clear() as that will cause us to throw away
4736 		 * degraded/faulted state as well as attempt to reopen the
4737 		 * device, all of which is a waste.
4738 		 */
4739 		vd->vdev_stat.vs_read_errors = 0;
4740 		vd->vdev_stat.vs_write_errors = 0;
4741 		vd->vdev_stat.vs_checksum_errors = 0;
4742 
4743 		vdev_state_dirty(vd->vdev_top);
4744 	}
4745 
4746 	for (int c = 0; c < vd->vdev_children; c++)
4747 		spa_async_remove(spa, vd->vdev_child[c]);
4748 }
4749 
4750 static void
4751 spa_async_probe(spa_t *spa, vdev_t *vd)
4752 {
4753 	if (vd->vdev_probe_wanted) {
4754 		vd->vdev_probe_wanted = 0;
4755 		vdev_reopen(vd);	/* vdev_open() does the actual probe */
4756 	}
4757 
4758 	for (int c = 0; c < vd->vdev_children; c++)
4759 		spa_async_probe(spa, vd->vdev_child[c]);
4760 }
4761 
4762 static void
4763 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
4764 {
4765 	sysevent_id_t eid;
4766 	nvlist_t *attr;
4767 	char *physpath;
4768 
4769 	if (!spa->spa_autoexpand)
4770 		return;
4771 
4772 	for (int c = 0; c < vd->vdev_children; c++) {
4773 		vdev_t *cvd = vd->vdev_child[c];
4774 		spa_async_autoexpand(spa, cvd);
4775 	}
4776 
4777 	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
4778 		return;
4779 
4780 	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
4781 	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
4782 
4783 	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4784 	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
4785 
4786 	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
4787 	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
4788 
4789 	nvlist_free(attr);
4790 	kmem_free(physpath, MAXPATHLEN);
4791 }
4792 
4793 static void
4794 spa_async_thread(spa_t *spa)
4795 {
4796 	int tasks;
4797 
4798 	ASSERT(spa->spa_sync_on);
4799 
4800 	mutex_enter(&spa->spa_async_lock);
4801 	tasks = spa->spa_async_tasks;
4802 	spa->spa_async_tasks = 0;
4803 	mutex_exit(&spa->spa_async_lock);
4804 
4805 	/*
4806 	 * See if the config needs to be updated.
4807 	 */
4808 	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
4809 		uint64_t old_space, new_space;
4810 
4811 		mutex_enter(&spa_namespace_lock);
4812 		old_space = metaslab_class_get_space(spa_normal_class(spa));
4813 		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4814 		new_space = metaslab_class_get_space(spa_normal_class(spa));
4815 		mutex_exit(&spa_namespace_lock);
4816 
4817 		/*
4818 		 * If the pool grew as a result of the config update,
4819 		 * then log an internal history event.
4820 		 */
4821 		if (new_space != old_space) {
4822 			spa_history_internal_log(LOG_POOL_VDEV_ONLINE,
4823 			    spa, NULL, CRED(),
4824 			    "pool '%s' size: %llu(+%llu)",
4825 			    spa_name(spa), new_space, new_space - old_space);
4826 		}
4827 	}
4828 
4829 	/*
4830 	 * See if any devices need to be marked REMOVED.
4831 	 */
4832 	if (tasks & SPA_ASYNC_REMOVE) {
4833 		spa_vdev_state_enter(spa, SCL_NONE);
4834 		spa_async_remove(spa, spa->spa_root_vdev);
4835 		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
4836 			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
4837 		for (int i = 0; i < spa->spa_spares.sav_count; i++)
4838 			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
4839 		(void) spa_vdev_state_exit(spa, NULL, 0);
4840 	}
4841 
4842 	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
4843 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4844 		spa_async_autoexpand(spa, spa->spa_root_vdev);
4845 		spa_config_exit(spa, SCL_CONFIG, FTAG);
4846 	}
4847 
4848 	/*
4849 	 * See if any devices need to be probed.
4850 	 */
4851 	if (tasks & SPA_ASYNC_PROBE) {
4852 		spa_vdev_state_enter(spa, SCL_NONE);
4853 		spa_async_probe(spa, spa->spa_root_vdev);
4854 		(void) spa_vdev_state_exit(spa, NULL, 0);
4855 	}
4856 
4857 	/*
4858 	 * If any devices are done replacing, detach them.
4859 	 */
4860 	if (tasks & SPA_ASYNC_RESILVER_DONE)
4861 		spa_vdev_resilver_done(spa);
4862 
4863 	/*
4864 	 * Kick off a resilver.
4865 	 */
4866 	if (tasks & SPA_ASYNC_RESILVER)
4867 		VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER) == 0);
4868 
4869 	/*
4870 	 * Let the world know that we're done.
4871 	 */
4872 	mutex_enter(&spa->spa_async_lock);
4873 	spa->spa_async_thread = NULL;
4874 	cv_broadcast(&spa->spa_async_cv);
4875 	mutex_exit(&spa->spa_async_lock);
4876 	thread_exit();
4877 }
4878 
4879 void
4880 spa_async_suspend(spa_t *spa)
4881 {
4882 	mutex_enter(&spa->spa_async_lock);
4883 	spa->spa_async_suspended++;
4884 	while (spa->spa_async_thread != NULL)
4885 		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
4886 	mutex_exit(&spa->spa_async_lock);
4887 }
4888 
4889 void
4890 spa_async_resume(spa_t *spa)
4891 {
4892 	mutex_enter(&spa->spa_async_lock);
4893 	ASSERT(spa->spa_async_suspended != 0);
4894 	spa->spa_async_suspended--;
4895 	mutex_exit(&spa->spa_async_lock);
4896 }
4897 
4898 static void
4899 spa_async_dispatch(spa_t *spa)
4900 {
4901 	mutex_enter(&spa->spa_async_lock);
4902 	if (spa->spa_async_tasks && !spa->spa_async_suspended &&
4903 	    spa->spa_async_thread == NULL &&
4904 	    rootdir != NULL && !vn_is_readonly(rootdir))
4905 		spa->spa_async_thread = thread_create(NULL, 0,
4906 		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
4907 	mutex_exit(&spa->spa_async_lock);
4908 }
4909 
4910 void
4911 spa_async_request(spa_t *spa, int task)
4912 {
4913 	mutex_enter(&spa->spa_async_lock);
4914 	spa->spa_async_tasks |= task;
4915 	mutex_exit(&spa->spa_async_lock);
4916 }
4917 
4918 /*
4919  * ==========================================================================
4920  * SPA syncing routines
4921  * ==========================================================================
4922  */
4923 static void
4924 spa_sync_deferred_bplist(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx, uint64_t txg)
4925 {
4926 	blkptr_t blk;
4927 	uint64_t itor = 0;
4928 	uint8_t c = 1;
4929 
4930 	while (bplist_iterate(bpl, &itor, &blk) == 0) {
4931 		ASSERT(blk.blk_birth < txg);
4932 		zio_free(spa, txg, &blk);
4933 	}
4934 
4935 	bplist_vacate(bpl, tx);
4936 
4937 	/*
4938 	 * Pre-dirty the first block so we sync to convergence faster.
4939 	 * (Usually only the first block is needed.)
4940 	 */
4941 	dmu_write(bpl->bpl_mos, spa->spa_deferred_bplist_obj, 0, 1, &c, tx);
4942 }
4943 
4944 static void
4945 spa_sync_free(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
4946 {
4947 	zio_t *zio = arg;
4948 
4949 	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
4950 	    zio->io_flags));
4951 }
4952 
4953 static void
4954 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
4955 {
4956 	char *packed = NULL;
4957 	size_t bufsize;
4958 	size_t nvsize = 0;
4959 	dmu_buf_t *db;
4960 
4961 	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
4962 
4963 	/*
4964 	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
4965 	 * information.  This avoids the dbuf_will_dirty() path and
4966 	 * saves us a pre-read to get data we don't actually care about.
4967 	 */
4968 	bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
4969 	packed = kmem_alloc(bufsize, KM_SLEEP);
4970 
4971 	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
4972 	    KM_SLEEP) == 0);
4973 	bzero(packed + nvsize, bufsize - nvsize);
4974 
4975 	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
4976 
4977 	kmem_free(packed, bufsize);
4978 
4979 	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
4980 	dmu_buf_will_dirty(db, tx);
4981 	*(uint64_t *)db->db_data = nvsize;
4982 	dmu_buf_rele(db, FTAG);
4983 }
4984 
4985 static void
4986 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
4987     const char *config, const char *entry)
4988 {
4989 	nvlist_t *nvroot;
4990 	nvlist_t **list;
4991 	int i;
4992 
4993 	if (!sav->sav_sync)
4994 		return;
4995 
4996 	/*
4997 	 * Update the MOS nvlist describing the list of available devices.
4998 	 * spa_validate_aux() will have already made sure this nvlist is
4999 	 * valid and the vdevs are labeled appropriately.
5000 	 */
5001 	if (sav->sav_object == 0) {
5002 		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5003 		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5004 		    sizeof (uint64_t), tx);
5005 		VERIFY(zap_update(spa->spa_meta_objset,
5006 		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5007 		    &sav->sav_object, tx) == 0);
5008 	}
5009 
5010 	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5011 	if (sav->sav_count == 0) {
5012 		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5013 	} else {
5014 		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5015 		for (i = 0; i < sav->sav_count; i++)
5016 			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5017 			    B_FALSE, B_FALSE, B_TRUE);
5018 		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5019 		    sav->sav_count) == 0);
5020 		for (i = 0; i < sav->sav_count; i++)
5021 			nvlist_free(list[i]);
5022 		kmem_free(list, sav->sav_count * sizeof (void *));
5023 	}
5024 
5025 	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5026 	nvlist_free(nvroot);
5027 
5028 	sav->sav_sync = B_FALSE;
5029 }
5030 
5031 static void
5032 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5033 {
5034 	nvlist_t *config;
5035 
5036 	if (list_is_empty(&spa->spa_config_dirty_list))
5037 		return;
5038 
5039 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5040 
5041 	config = spa_config_generate(spa, spa->spa_root_vdev,
5042 	    dmu_tx_get_txg(tx), B_FALSE);
5043 
5044 	spa_config_exit(spa, SCL_STATE, FTAG);
5045 
5046 	if (spa->spa_config_syncing)
5047 		nvlist_free(spa->spa_config_syncing);
5048 	spa->spa_config_syncing = config;
5049 
5050 	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5051 }
5052 
5053 /*
5054  * Set zpool properties.
5055  */
5056 static void
5057 spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
5058 {
5059 	spa_t *spa = arg1;
5060 	objset_t *mos = spa->spa_meta_objset;
5061 	nvlist_t *nvp = arg2;
5062 	nvpair_t *elem;
5063 	uint64_t intval;
5064 	char *strval;
5065 	zpool_prop_t prop;
5066 	const char *propname;
5067 	zprop_type_t proptype;
5068 
5069 	mutex_enter(&spa->spa_props_lock);
5070 
5071 	elem = NULL;
5072 	while ((elem = nvlist_next_nvpair(nvp, elem))) {
5073 		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5074 		case ZPOOL_PROP_VERSION:
5075 			/*
5076 			 * Only set version for non-zpool-creation cases
5077 			 * (set/import). spa_create() needs special care
5078 			 * for version setting.
5079 			 */
5080 			if (tx->tx_txg != TXG_INITIAL) {
5081 				VERIFY(nvpair_value_uint64(elem,
5082 				    &intval) == 0);
5083 				ASSERT(intval <= SPA_VERSION);
5084 				ASSERT(intval >= spa_version(spa));
5085 				spa->spa_uberblock.ub_version = intval;
5086 				vdev_config_dirty(spa->spa_root_vdev);
5087 			}
5088 			break;
5089 
5090 		case ZPOOL_PROP_ALTROOT:
5091 			/*
5092 			 * 'altroot' is a non-persistent property. It should
5093 			 * have been set temporarily at creation or import time.
5094 			 */
5095 			ASSERT(spa->spa_root != NULL);
5096 			break;
5097 
5098 		case ZPOOL_PROP_CACHEFILE:
5099 			/*
5100 			 * 'cachefile' is also a non-persisitent property.
5101 			 */
5102 			break;
5103 		default:
5104 			/*
5105 			 * Set pool property values in the poolprops mos object.
5106 			 */
5107 			if (spa->spa_pool_props_object == 0) {
5108 				VERIFY((spa->spa_pool_props_object =
5109 				    zap_create(mos, DMU_OT_POOL_PROPS,
5110 				    DMU_OT_NONE, 0, tx)) > 0);
5111 
5112 				VERIFY(zap_update(mos,
5113 				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5114 				    8, 1, &spa->spa_pool_props_object, tx)
5115 				    == 0);
5116 			}
5117 
5118 			/* normalize the property name */
5119 			propname = zpool_prop_to_name(prop);
5120 			proptype = zpool_prop_get_type(prop);
5121 
5122 			if (nvpair_type(elem) == DATA_TYPE_STRING) {
5123 				ASSERT(proptype == PROP_TYPE_STRING);
5124 				VERIFY(nvpair_value_string(elem, &strval) == 0);
5125 				VERIFY(zap_update(mos,
5126 				    spa->spa_pool_props_object, propname,
5127 				    1, strlen(strval) + 1, strval, tx) == 0);
5128 
5129 			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5130 				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5131 
5132 				if (proptype == PROP_TYPE_INDEX) {
5133 					const char *unused;
5134 					VERIFY(zpool_prop_index_to_string(
5135 					    prop, intval, &unused) == 0);
5136 				}
5137 				VERIFY(zap_update(mos,
5138 				    spa->spa_pool_props_object, propname,
5139 				    8, 1, &intval, tx) == 0);
5140 			} else {
5141 				ASSERT(0); /* not allowed */
5142 			}
5143 
5144 			switch (prop) {
5145 			case ZPOOL_PROP_DELEGATION:
5146 				spa->spa_delegation = intval;
5147 				break;
5148 			case ZPOOL_PROP_BOOTFS:
5149 				spa->spa_bootfs = intval;
5150 				break;
5151 			case ZPOOL_PROP_FAILUREMODE:
5152 				spa->spa_failmode = intval;
5153 				break;
5154 			case ZPOOL_PROP_AUTOEXPAND:
5155 				spa->spa_autoexpand = intval;
5156 				spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
5157 				break;
5158 			case ZPOOL_PROP_DEDUPDITTO:
5159 				spa->spa_dedup_ditto = intval;
5160 				break;
5161 			default:
5162 				break;
5163 			}
5164 		}
5165 
5166 		/* log internal history if this is not a zpool create */
5167 		if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5168 		    tx->tx_txg != TXG_INITIAL) {
5169 			spa_history_internal_log(LOG_POOL_PROPSET,
5170 			    spa, tx, cr, "%s %lld %s",
5171 			    nvpair_name(elem), intval, spa_name(spa));
5172 		}
5173 	}
5174 
5175 	mutex_exit(&spa->spa_props_lock);
5176 }
5177 
5178 /*
5179  * Sync the specified transaction group.  New blocks may be dirtied as
5180  * part of the process, so we iterate until it converges.
5181  */
5182 void
5183 spa_sync(spa_t *spa, uint64_t txg)
5184 {
5185 	dsl_pool_t *dp = spa->spa_dsl_pool;
5186 	objset_t *mos = spa->spa_meta_objset;
5187 	bplist_t *defer_bpl = &spa->spa_deferred_bplist;
5188 	bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5189 	vdev_t *rvd = spa->spa_root_vdev;
5190 	vdev_t *vd;
5191 	dmu_tx_t *tx;
5192 	int error;
5193 
5194 	/*
5195 	 * Lock out configuration changes.
5196 	 */
5197 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5198 
5199 	spa->spa_syncing_txg = txg;
5200 	spa->spa_sync_pass = 0;
5201 
5202 	/*
5203 	 * If there are any pending vdev state changes, convert them
5204 	 * into config changes that go out with this transaction group.
5205 	 */
5206 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5207 	while (list_head(&spa->spa_state_dirty_list) != NULL) {
5208 		/*
5209 		 * We need the write lock here because, for aux vdevs,
5210 		 * calling vdev_config_dirty() modifies sav_config.
5211 		 * This is ugly and will become unnecessary when we
5212 		 * eliminate the aux vdev wart by integrating all vdevs
5213 		 * into the root vdev tree.
5214 		 */
5215 		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5216 		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5217 		while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5218 			vdev_state_clean(vd);
5219 			vdev_config_dirty(vd);
5220 		}
5221 		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5222 		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5223 	}
5224 	spa_config_exit(spa, SCL_STATE, FTAG);
5225 
5226 	VERIFY(0 == bplist_open(defer_bpl, mos, spa->spa_deferred_bplist_obj));
5227 
5228 	tx = dmu_tx_create_assigned(dp, txg);
5229 
5230 	/*
5231 	 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5232 	 * set spa_deflate if we have no raid-z vdevs.
5233 	 */
5234 	if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5235 	    spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5236 		int i;
5237 
5238 		for (i = 0; i < rvd->vdev_children; i++) {
5239 			vd = rvd->vdev_child[i];
5240 			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5241 				break;
5242 		}
5243 		if (i == rvd->vdev_children) {
5244 			spa->spa_deflate = TRUE;
5245 			VERIFY(0 == zap_add(spa->spa_meta_objset,
5246 			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5247 			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5248 		}
5249 	}
5250 
5251 	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5252 	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5253 		dsl_pool_create_origin(dp, tx);
5254 
5255 		/* Keeping the origin open increases spa_minref */
5256 		spa->spa_minref += 3;
5257 	}
5258 
5259 	if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5260 	    spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5261 		dsl_pool_upgrade_clones(dp, tx);
5262 	}
5263 
5264 	/*
5265 	 * If anything has changed in this txg, push the deferred frees
5266 	 * from the previous txg.  If not, leave them alone so that we
5267 	 * don't generate work on an otherwise idle system.
5268 	 */
5269 	if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5270 	    !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5271 	    !txg_list_empty(&dp->dp_sync_tasks, txg))
5272 		spa_sync_deferred_bplist(spa, defer_bpl, tx, txg);
5273 
5274 	/*
5275 	 * Iterate to convergence.
5276 	 */
5277 	do {
5278 		int pass = ++spa->spa_sync_pass;
5279 
5280 		spa_sync_config_object(spa, tx);
5281 		spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5282 		    ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5283 		spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5284 		    ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5285 		spa_errlog_sync(spa, txg);
5286 		dsl_pool_sync(dp, txg);
5287 
5288 		if (pass <= SYNC_PASS_DEFERRED_FREE) {
5289 			zio_t *zio = zio_root(spa, NULL, NULL, 0);
5290 			bplist_sync(free_bpl, spa_sync_free, zio, tx);
5291 			VERIFY(zio_wait(zio) == 0);
5292 		} else {
5293 			bplist_sync(free_bpl, bplist_enqueue_cb, defer_bpl, tx);
5294 		}
5295 
5296 		ddt_sync(spa, txg);
5297 
5298 		mutex_enter(&spa->spa_scrub_lock);
5299 		while (spa->spa_scrub_inflight > 0)
5300 			cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
5301 		mutex_exit(&spa->spa_scrub_lock);
5302 
5303 		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
5304 			vdev_sync(vd, txg);
5305 
5306 	} while (dmu_objset_is_dirty(mos, txg));
5307 
5308 	ASSERT(free_bpl->bpl_queue == NULL);
5309 
5310 	bplist_close(defer_bpl);
5311 
5312 	/*
5313 	 * Rewrite the vdev configuration (which includes the uberblock)
5314 	 * to commit the transaction group.
5315 	 *
5316 	 * If there are no dirty vdevs, we sync the uberblock to a few
5317 	 * random top-level vdevs that are known to be visible in the
5318 	 * config cache (see spa_vdev_add() for a complete description).
5319 	 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
5320 	 */
5321 	for (;;) {
5322 		/*
5323 		 * We hold SCL_STATE to prevent vdev open/close/etc.
5324 		 * while we're attempting to write the vdev labels.
5325 		 */
5326 		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5327 
5328 		if (list_is_empty(&spa->spa_config_dirty_list)) {
5329 			vdev_t *svd[SPA_DVAS_PER_BP];
5330 			int svdcount = 0;
5331 			int children = rvd->vdev_children;
5332 			int c0 = spa_get_random(children);
5333 
5334 			for (int c = 0; c < children; c++) {
5335 				vd = rvd->vdev_child[(c0 + c) % children];
5336 				if (vd->vdev_ms_array == 0 || vd->vdev_islog)
5337 					continue;
5338 				svd[svdcount++] = vd;
5339 				if (svdcount == SPA_DVAS_PER_BP)
5340 					break;
5341 			}
5342 			error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
5343 			if (error != 0)
5344 				error = vdev_config_sync(svd, svdcount, txg,
5345 				    B_TRUE);
5346 		} else {
5347 			error = vdev_config_sync(rvd->vdev_child,
5348 			    rvd->vdev_children, txg, B_FALSE);
5349 			if (error != 0)
5350 				error = vdev_config_sync(rvd->vdev_child,
5351 				    rvd->vdev_children, txg, B_TRUE);
5352 		}
5353 
5354 		spa_config_exit(spa, SCL_STATE, FTAG);
5355 
5356 		if (error == 0)
5357 			break;
5358 		zio_suspend(spa, NULL);
5359 		zio_resume_wait(spa);
5360 	}
5361 	dmu_tx_commit(tx);
5362 
5363 	/*
5364 	 * Clear the dirty config list.
5365 	 */
5366 	while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
5367 		vdev_config_clean(vd);
5368 
5369 	/*
5370 	 * Now that the new config has synced transactionally,
5371 	 * let it become visible to the config cache.
5372 	 */
5373 	if (spa->spa_config_syncing != NULL) {
5374 		spa_config_set(spa, spa->spa_config_syncing);
5375 		spa->spa_config_txg = txg;
5376 		spa->spa_config_syncing = NULL;
5377 	}
5378 
5379 	spa->spa_ubsync = spa->spa_uberblock;
5380 
5381 	dsl_pool_sync_done(dp, txg);
5382 
5383 	/*
5384 	 * Update usable space statistics.
5385 	 */
5386 	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
5387 		vdev_sync_done(vd, txg);
5388 
5389 	spa_update_dspace(spa);
5390 
5391 	/*
5392 	 * It had better be the case that we didn't dirty anything
5393 	 * since vdev_config_sync().
5394 	 */
5395 	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
5396 	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
5397 	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
5398 	ASSERT(defer_bpl->bpl_queue == NULL);
5399 	ASSERT(free_bpl->bpl_queue == NULL);
5400 
5401 	spa->spa_sync_pass = 0;
5402 
5403 	spa_config_exit(spa, SCL_CONFIG, FTAG);
5404 
5405 	spa_handle_ignored_writes(spa);
5406 
5407 	/*
5408 	 * If any async tasks have been requested, kick them off.
5409 	 */
5410 	spa_async_dispatch(spa);
5411 }
5412 
5413 /*
5414  * Sync all pools.  We don't want to hold the namespace lock across these
5415  * operations, so we take a reference on the spa_t and drop the lock during the
5416  * sync.
5417  */
5418 void
5419 spa_sync_allpools(void)
5420 {
5421 	spa_t *spa = NULL;
5422 	mutex_enter(&spa_namespace_lock);
5423 	while ((spa = spa_next(spa)) != NULL) {
5424 		if (spa_state(spa) != POOL_STATE_ACTIVE || spa_suspended(spa))
5425 			continue;
5426 		spa_open_ref(spa, FTAG);
5427 		mutex_exit(&spa_namespace_lock);
5428 		txg_wait_synced(spa_get_dsl(spa), 0);
5429 		mutex_enter(&spa_namespace_lock);
5430 		spa_close(spa, FTAG);
5431 	}
5432 	mutex_exit(&spa_namespace_lock);
5433 }
5434 
5435 /*
5436  * ==========================================================================
5437  * Miscellaneous routines
5438  * ==========================================================================
5439  */
5440 
5441 /*
5442  * Remove all pools in the system.
5443  */
5444 void
5445 spa_evict_all(void)
5446 {
5447 	spa_t *spa;
5448 
5449 	/*
5450 	 * Remove all cached state.  All pools should be closed now,
5451 	 * so every spa in the AVL tree should be unreferenced.
5452 	 */
5453 	mutex_enter(&spa_namespace_lock);
5454 	while ((spa = spa_next(NULL)) != NULL) {
5455 		/*
5456 		 * Stop async tasks.  The async thread may need to detach
5457 		 * a device that's been replaced, which requires grabbing
5458 		 * spa_namespace_lock, so we must drop it here.
5459 		 */
5460 		spa_open_ref(spa, FTAG);
5461 		mutex_exit(&spa_namespace_lock);
5462 		spa_async_suspend(spa);
5463 		mutex_enter(&spa_namespace_lock);
5464 		spa_close(spa, FTAG);
5465 
5466 		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5467 			spa_unload(spa);
5468 			spa_deactivate(spa);
5469 		}
5470 		spa_remove(spa);
5471 	}
5472 	mutex_exit(&spa_namespace_lock);
5473 }
5474 
5475 vdev_t *
5476 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
5477 {
5478 	vdev_t *vd;
5479 	int i;
5480 
5481 	if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
5482 		return (vd);
5483 
5484 	if (aux) {
5485 		for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
5486 			vd = spa->spa_l2cache.sav_vdevs[i];
5487 			if (vd->vdev_guid == guid)
5488 				return (vd);
5489 		}
5490 
5491 		for (i = 0; i < spa->spa_spares.sav_count; i++) {
5492 			vd = spa->spa_spares.sav_vdevs[i];
5493 			if (vd->vdev_guid == guid)
5494 				return (vd);
5495 		}
5496 	}
5497 
5498 	return (NULL);
5499 }
5500 
5501 void
5502 spa_upgrade(spa_t *spa, uint64_t version)
5503 {
5504 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5505 
5506 	/*
5507 	 * This should only be called for a non-faulted pool, and since a
5508 	 * future version would result in an unopenable pool, this shouldn't be
5509 	 * possible.
5510 	 */
5511 	ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
5512 	ASSERT(version >= spa->spa_uberblock.ub_version);
5513 
5514 	spa->spa_uberblock.ub_version = version;
5515 	vdev_config_dirty(spa->spa_root_vdev);
5516 
5517 	spa_config_exit(spa, SCL_ALL, FTAG);
5518 
5519 	txg_wait_synced(spa_get_dsl(spa), 0);
5520 }
5521 
5522 boolean_t
5523 spa_has_spare(spa_t *spa, uint64_t guid)
5524 {
5525 	int i;
5526 	uint64_t spareguid;
5527 	spa_aux_vdev_t *sav = &spa->spa_spares;
5528 
5529 	for (i = 0; i < sav->sav_count; i++)
5530 		if (sav->sav_vdevs[i]->vdev_guid == guid)
5531 			return (B_TRUE);
5532 
5533 	for (i = 0; i < sav->sav_npending; i++) {
5534 		if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
5535 		    &spareguid) == 0 && spareguid == guid)
5536 			return (B_TRUE);
5537 	}
5538 
5539 	return (B_FALSE);
5540 }
5541 
5542 /*
5543  * Check if a pool has an active shared spare device.
5544  * Note: reference count of an active spare is 2, as a spare and as a replace
5545  */
5546 static boolean_t
5547 spa_has_active_shared_spare(spa_t *spa)
5548 {
5549 	int i, refcnt;
5550 	uint64_t pool;
5551 	spa_aux_vdev_t *sav = &spa->spa_spares;
5552 
5553 	for (i = 0; i < sav->sav_count; i++) {
5554 		if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
5555 		    &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
5556 		    refcnt > 2)
5557 			return (B_TRUE);
5558 	}
5559 
5560 	return (B_FALSE);
5561 }
5562 
5563 /*
5564  * Post a sysevent corresponding to the given event.  The 'name' must be one of
5565  * the event definitions in sys/sysevent/eventdefs.h.  The payload will be
5566  * filled in from the spa and (optionally) the vdev.  This doesn't do anything
5567  * in the userland libzpool, as we don't want consumers to misinterpret ztest
5568  * or zdb as real changes.
5569  */
5570 void
5571 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
5572 {
5573 #ifdef _KERNEL
5574 	sysevent_t		*ev;
5575 	sysevent_attr_list_t	*attr = NULL;
5576 	sysevent_value_t	value;
5577 	sysevent_id_t		eid;
5578 
5579 	ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
5580 	    SE_SLEEP);
5581 
5582 	value.value_type = SE_DATA_TYPE_STRING;
5583 	value.value.sv_string = spa_name(spa);
5584 	if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
5585 		goto done;
5586 
5587 	value.value_type = SE_DATA_TYPE_UINT64;
5588 	value.value.sv_uint64 = spa_guid(spa);
5589 	if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
5590 		goto done;
5591 
5592 	if (vd) {
5593 		value.value_type = SE_DATA_TYPE_UINT64;
5594 		value.value.sv_uint64 = vd->vdev_guid;
5595 		if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
5596 		    SE_SLEEP) != 0)
5597 			goto done;
5598 
5599 		if (vd->vdev_path) {
5600 			value.value_type = SE_DATA_TYPE_STRING;
5601 			value.value.sv_string = vd->vdev_path;
5602 			if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
5603 			    &value, SE_SLEEP) != 0)
5604 				goto done;
5605 		}
5606 	}
5607 
5608 	if (sysevent_attach_attributes(ev, attr) != 0)
5609 		goto done;
5610 	attr = NULL;
5611 
5612 	(void) log_sysevent(ev, SE_SLEEP, &eid);
5613 
5614 done:
5615 	if (attr)
5616 		sysevent_free_attr(attr);
5617 	sysevent_free(ev);
5618 #endif
5619 }
5620