xref: /illumos-gate/usr/src/uts/common/fs/zfs/spa.c (revision b7b2590dd9f11b12a0b4878db3886068cce176af)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
25  * Copyright (c) 2015, Nexenta Systems, Inc.  All rights reserved.
26  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27  * Copyright 2013 Saso Kiselkov. All rights reserved.
28  * Copyright (c) 2014 Integros [integros.com]
29  * Copyright 2016 Toomas Soome <tsoome@me.com>
30  */
31 
32 /*
33  * SPA: Storage Pool Allocator
34  *
35  * This file contains all the routines used when modifying on-disk SPA state.
36  * This includes opening, importing, destroying, exporting a pool, and syncing a
37  * pool.
38  */
39 
40 #include <sys/zfs_context.h>
41 #include <sys/fm/fs/zfs.h>
42 #include <sys/spa_impl.h>
43 #include <sys/zio.h>
44 #include <sys/zio_checksum.h>
45 #include <sys/dmu.h>
46 #include <sys/dmu_tx.h>
47 #include <sys/zap.h>
48 #include <sys/zil.h>
49 #include <sys/ddt.h>
50 #include <sys/vdev_impl.h>
51 #include <sys/metaslab.h>
52 #include <sys/metaslab_impl.h>
53 #include <sys/uberblock_impl.h>
54 #include <sys/txg.h>
55 #include <sys/avl.h>
56 #include <sys/dmu_traverse.h>
57 #include <sys/dmu_objset.h>
58 #include <sys/unique.h>
59 #include <sys/dsl_pool.h>
60 #include <sys/dsl_dataset.h>
61 #include <sys/dsl_dir.h>
62 #include <sys/dsl_prop.h>
63 #include <sys/dsl_synctask.h>
64 #include <sys/fs/zfs.h>
65 #include <sys/arc.h>
66 #include <sys/callb.h>
67 #include <sys/systeminfo.h>
68 #include <sys/spa_boot.h>
69 #include <sys/zfs_ioctl.h>
70 #include <sys/dsl_scan.h>
71 #include <sys/zfeature.h>
72 #include <sys/dsl_destroy.h>
73 
74 #ifdef	_KERNEL
75 #include <sys/bootprops.h>
76 #include <sys/callb.h>
77 #include <sys/cpupart.h>
78 #include <sys/pool.h>
79 #include <sys/sysdc.h>
80 #include <sys/zone.h>
81 #endif	/* _KERNEL */
82 
83 #include "zfs_prop.h"
84 #include "zfs_comutil.h"
85 
86 /*
87  * The interval, in seconds, at which failed configuration cache file writes
88  * should be retried.
89  */
90 static int zfs_ccw_retry_interval = 300;
91 
92 typedef enum zti_modes {
93 	ZTI_MODE_FIXED,			/* value is # of threads (min 1) */
94 	ZTI_MODE_BATCH,			/* cpu-intensive; value is ignored */
95 	ZTI_MODE_NULL,			/* don't create a taskq */
96 	ZTI_NMODES
97 } zti_modes_t;
98 
99 #define	ZTI_P(n, q)	{ ZTI_MODE_FIXED, (n), (q) }
100 #define	ZTI_BATCH	{ ZTI_MODE_BATCH, 0, 1 }
101 #define	ZTI_NULL	{ ZTI_MODE_NULL, 0, 0 }
102 
103 #define	ZTI_N(n)	ZTI_P(n, 1)
104 #define	ZTI_ONE		ZTI_N(1)
105 
106 typedef struct zio_taskq_info {
107 	zti_modes_t zti_mode;
108 	uint_t zti_value;
109 	uint_t zti_count;
110 } zio_taskq_info_t;
111 
112 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
113 	"issue", "issue_high", "intr", "intr_high"
114 };
115 
116 /*
117  * This table defines the taskq settings for each ZFS I/O type. When
118  * initializing a pool, we use this table to create an appropriately sized
119  * taskq. Some operations are low volume and therefore have a small, static
120  * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
121  * macros. Other operations process a large amount of data; the ZTI_BATCH
122  * macro causes us to create a taskq oriented for throughput. Some operations
123  * are so high frequency and short-lived that the taskq itself can become a a
124  * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
125  * additional degree of parallelism specified by the number of threads per-
126  * taskq and the number of taskqs; when dispatching an event in this case, the
127  * particular taskq is chosen at random.
128  *
129  * The different taskq priorities are to handle the different contexts (issue
130  * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
131  * need to be handled with minimum delay.
132  */
133 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
134 	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
135 	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* NULL */
136 	{ ZTI_N(8),	ZTI_NULL,	ZTI_P(12, 8),	ZTI_NULL }, /* READ */
137 	{ ZTI_BATCH,	ZTI_N(5),	ZTI_N(8),	ZTI_N(5) }, /* WRITE */
138 	{ ZTI_P(12, 8),	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* FREE */
139 	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* CLAIM */
140 	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* IOCTL */
141 };
142 
143 static sysevent_t *spa_event_create(spa_t *spa, vdev_t *vd, const char *name);
144 static void spa_event_post(sysevent_t *ev);
145 static void spa_sync_version(void *arg, dmu_tx_t *tx);
146 static void spa_sync_props(void *arg, dmu_tx_t *tx);
147 static boolean_t spa_has_active_shared_spare(spa_t *spa);
148 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
149     spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
150     char **ereport);
151 static void spa_vdev_resilver_done(spa_t *spa);
152 
153 uint_t		zio_taskq_batch_pct = 75;	/* 1 thread per cpu in pset */
154 id_t		zio_taskq_psrset_bind = PS_NONE;
155 boolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
156 uint_t		zio_taskq_basedc = 80;		/* base duty cycle */
157 
158 boolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
159 extern int	zfs_sync_pass_deferred_free;
160 
161 /*
162  * This (illegal) pool name is used when temporarily importing a spa_t in order
163  * to get the vdev stats associated with the imported devices.
164  */
165 #define	TRYIMPORT_NAME	"$import"
166 
167 /*
168  * ==========================================================================
169  * SPA properties routines
170  * ==========================================================================
171  */
172 
173 /*
174  * Add a (source=src, propname=propval) list to an nvlist.
175  */
176 static void
177 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
178     uint64_t intval, zprop_source_t src)
179 {
180 	const char *propname = zpool_prop_to_name(prop);
181 	nvlist_t *propval;
182 
183 	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
184 	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
185 
186 	if (strval != NULL)
187 		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
188 	else
189 		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
190 
191 	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
192 	nvlist_free(propval);
193 }
194 
195 /*
196  * Get property values from the spa configuration.
197  */
198 static void
199 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
200 {
201 	vdev_t *rvd = spa->spa_root_vdev;
202 	dsl_pool_t *pool = spa->spa_dsl_pool;
203 	uint64_t size, alloc, cap, version;
204 	zprop_source_t src = ZPROP_SRC_NONE;
205 	spa_config_dirent_t *dp;
206 	metaslab_class_t *mc = spa_normal_class(spa);
207 
208 	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
209 
210 	if (rvd != NULL) {
211 		alloc = metaslab_class_get_alloc(spa_normal_class(spa));
212 		size = metaslab_class_get_space(spa_normal_class(spa));
213 		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
214 		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
215 		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
216 		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
217 		    size - alloc, src);
218 
219 		spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
220 		    metaslab_class_fragmentation(mc), src);
221 		spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
222 		    metaslab_class_expandable_space(mc), src);
223 		spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
224 		    (spa_mode(spa) == FREAD), src);
225 
226 		cap = (size == 0) ? 0 : (alloc * 100 / size);
227 		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
228 
229 		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
230 		    ddt_get_pool_dedup_ratio(spa), src);
231 
232 		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
233 		    rvd->vdev_state, src);
234 
235 		version = spa_version(spa);
236 		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
237 			src = ZPROP_SRC_DEFAULT;
238 		else
239 			src = ZPROP_SRC_LOCAL;
240 		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
241 	}
242 
243 	if (pool != NULL) {
244 		/*
245 		 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
246 		 * when opening pools before this version freedir will be NULL.
247 		 */
248 		if (pool->dp_free_dir != NULL) {
249 			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
250 			    dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
251 			    src);
252 		} else {
253 			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
254 			    NULL, 0, src);
255 		}
256 
257 		if (pool->dp_leak_dir != NULL) {
258 			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
259 			    dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
260 			    src);
261 		} else {
262 			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
263 			    NULL, 0, src);
264 		}
265 	}
266 
267 	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
268 
269 	if (spa->spa_comment != NULL) {
270 		spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
271 		    0, ZPROP_SRC_LOCAL);
272 	}
273 
274 	if (spa->spa_root != NULL)
275 		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
276 		    0, ZPROP_SRC_LOCAL);
277 
278 	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
279 		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
280 		    MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
281 	} else {
282 		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
283 		    SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
284 	}
285 
286 	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
287 		if (dp->scd_path == NULL) {
288 			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
289 			    "none", 0, ZPROP_SRC_LOCAL);
290 		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
291 			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
292 			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
293 		}
294 	}
295 }
296 
297 /*
298  * Get zpool property values.
299  */
300 int
301 spa_prop_get(spa_t *spa, nvlist_t **nvp)
302 {
303 	objset_t *mos = spa->spa_meta_objset;
304 	zap_cursor_t zc;
305 	zap_attribute_t za;
306 	int err;
307 
308 	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
309 
310 	mutex_enter(&spa->spa_props_lock);
311 
312 	/*
313 	 * Get properties from the spa config.
314 	 */
315 	spa_prop_get_config(spa, nvp);
316 
317 	/* If no pool property object, no more prop to get. */
318 	if (mos == NULL || spa->spa_pool_props_object == 0) {
319 		mutex_exit(&spa->spa_props_lock);
320 		return (0);
321 	}
322 
323 	/*
324 	 * Get properties from the MOS pool property object.
325 	 */
326 	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
327 	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
328 	    zap_cursor_advance(&zc)) {
329 		uint64_t intval = 0;
330 		char *strval = NULL;
331 		zprop_source_t src = ZPROP_SRC_DEFAULT;
332 		zpool_prop_t prop;
333 
334 		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
335 			continue;
336 
337 		switch (za.za_integer_length) {
338 		case 8:
339 			/* integer property */
340 			if (za.za_first_integer !=
341 			    zpool_prop_default_numeric(prop))
342 				src = ZPROP_SRC_LOCAL;
343 
344 			if (prop == ZPOOL_PROP_BOOTFS) {
345 				dsl_pool_t *dp;
346 				dsl_dataset_t *ds = NULL;
347 
348 				dp = spa_get_dsl(spa);
349 				dsl_pool_config_enter(dp, FTAG);
350 				if (err = dsl_dataset_hold_obj(dp,
351 				    za.za_first_integer, FTAG, &ds)) {
352 					dsl_pool_config_exit(dp, FTAG);
353 					break;
354 				}
355 
356 				strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN,
357 				    KM_SLEEP);
358 				dsl_dataset_name(ds, strval);
359 				dsl_dataset_rele(ds, FTAG);
360 				dsl_pool_config_exit(dp, FTAG);
361 			} else {
362 				strval = NULL;
363 				intval = za.za_first_integer;
364 			}
365 
366 			spa_prop_add_list(*nvp, prop, strval, intval, src);
367 
368 			if (strval != NULL)
369 				kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN);
370 
371 			break;
372 
373 		case 1:
374 			/* string property */
375 			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
376 			err = zap_lookup(mos, spa->spa_pool_props_object,
377 			    za.za_name, 1, za.za_num_integers, strval);
378 			if (err) {
379 				kmem_free(strval, za.za_num_integers);
380 				break;
381 			}
382 			spa_prop_add_list(*nvp, prop, strval, 0, src);
383 			kmem_free(strval, za.za_num_integers);
384 			break;
385 
386 		default:
387 			break;
388 		}
389 	}
390 	zap_cursor_fini(&zc);
391 	mutex_exit(&spa->spa_props_lock);
392 out:
393 	if (err && err != ENOENT) {
394 		nvlist_free(*nvp);
395 		*nvp = NULL;
396 		return (err);
397 	}
398 
399 	return (0);
400 }
401 
402 /*
403  * Validate the given pool properties nvlist and modify the list
404  * for the property values to be set.
405  */
406 static int
407 spa_prop_validate(spa_t *spa, nvlist_t *props)
408 {
409 	nvpair_t *elem;
410 	int error = 0, reset_bootfs = 0;
411 	uint64_t objnum = 0;
412 	boolean_t has_feature = B_FALSE;
413 
414 	elem = NULL;
415 	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
416 		uint64_t intval;
417 		char *strval, *slash, *check, *fname;
418 		const char *propname = nvpair_name(elem);
419 		zpool_prop_t prop = zpool_name_to_prop(propname);
420 
421 		switch (prop) {
422 		case ZPROP_INVAL:
423 			if (!zpool_prop_feature(propname)) {
424 				error = SET_ERROR(EINVAL);
425 				break;
426 			}
427 
428 			/*
429 			 * Sanitize the input.
430 			 */
431 			if (nvpair_type(elem) != DATA_TYPE_UINT64) {
432 				error = SET_ERROR(EINVAL);
433 				break;
434 			}
435 
436 			if (nvpair_value_uint64(elem, &intval) != 0) {
437 				error = SET_ERROR(EINVAL);
438 				break;
439 			}
440 
441 			if (intval != 0) {
442 				error = SET_ERROR(EINVAL);
443 				break;
444 			}
445 
446 			fname = strchr(propname, '@') + 1;
447 			if (zfeature_lookup_name(fname, NULL) != 0) {
448 				error = SET_ERROR(EINVAL);
449 				break;
450 			}
451 
452 			has_feature = B_TRUE;
453 			break;
454 
455 		case ZPOOL_PROP_VERSION:
456 			error = nvpair_value_uint64(elem, &intval);
457 			if (!error &&
458 			    (intval < spa_version(spa) ||
459 			    intval > SPA_VERSION_BEFORE_FEATURES ||
460 			    has_feature))
461 				error = SET_ERROR(EINVAL);
462 			break;
463 
464 		case ZPOOL_PROP_DELEGATION:
465 		case ZPOOL_PROP_AUTOREPLACE:
466 		case ZPOOL_PROP_LISTSNAPS:
467 		case ZPOOL_PROP_AUTOEXPAND:
468 			error = nvpair_value_uint64(elem, &intval);
469 			if (!error && intval > 1)
470 				error = SET_ERROR(EINVAL);
471 			break;
472 
473 		case ZPOOL_PROP_BOOTFS:
474 			/*
475 			 * If the pool version is less than SPA_VERSION_BOOTFS,
476 			 * or the pool is still being created (version == 0),
477 			 * the bootfs property cannot be set.
478 			 */
479 			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
480 				error = SET_ERROR(ENOTSUP);
481 				break;
482 			}
483 
484 			/*
485 			 * Make sure the vdev config is bootable
486 			 */
487 			if (!vdev_is_bootable(spa->spa_root_vdev)) {
488 				error = SET_ERROR(ENOTSUP);
489 				break;
490 			}
491 
492 			reset_bootfs = 1;
493 
494 			error = nvpair_value_string(elem, &strval);
495 
496 			if (!error) {
497 				objset_t *os;
498 				uint64_t propval;
499 
500 				if (strval == NULL || strval[0] == '\0') {
501 					objnum = zpool_prop_default_numeric(
502 					    ZPOOL_PROP_BOOTFS);
503 					break;
504 				}
505 
506 				if (error = dmu_objset_hold(strval, FTAG, &os))
507 					break;
508 
509 				/*
510 				 * Must be ZPL, and its property settings
511 				 * must be supported by GRUB (compression
512 				 * is not gzip, and large blocks are not used).
513 				 */
514 
515 				if (dmu_objset_type(os) != DMU_OST_ZFS) {
516 					error = SET_ERROR(ENOTSUP);
517 				} else if ((error =
518 				    dsl_prop_get_int_ds(dmu_objset_ds(os),
519 				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
520 				    &propval)) == 0 &&
521 				    !BOOTFS_COMPRESS_VALID(propval)) {
522 					error = SET_ERROR(ENOTSUP);
523 				} else {
524 					objnum = dmu_objset_id(os);
525 				}
526 				dmu_objset_rele(os, FTAG);
527 			}
528 			break;
529 
530 		case ZPOOL_PROP_FAILUREMODE:
531 			error = nvpair_value_uint64(elem, &intval);
532 			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
533 			    intval > ZIO_FAILURE_MODE_PANIC))
534 				error = SET_ERROR(EINVAL);
535 
536 			/*
537 			 * This is a special case which only occurs when
538 			 * the pool has completely failed. This allows
539 			 * the user to change the in-core failmode property
540 			 * without syncing it out to disk (I/Os might
541 			 * currently be blocked). We do this by returning
542 			 * EIO to the caller (spa_prop_set) to trick it
543 			 * into thinking we encountered a property validation
544 			 * error.
545 			 */
546 			if (!error && spa_suspended(spa)) {
547 				spa->spa_failmode = intval;
548 				error = SET_ERROR(EIO);
549 			}
550 			break;
551 
552 		case ZPOOL_PROP_CACHEFILE:
553 			if ((error = nvpair_value_string(elem, &strval)) != 0)
554 				break;
555 
556 			if (strval[0] == '\0')
557 				break;
558 
559 			if (strcmp(strval, "none") == 0)
560 				break;
561 
562 			if (strval[0] != '/') {
563 				error = SET_ERROR(EINVAL);
564 				break;
565 			}
566 
567 			slash = strrchr(strval, '/');
568 			ASSERT(slash != NULL);
569 
570 			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
571 			    strcmp(slash, "/..") == 0)
572 				error = SET_ERROR(EINVAL);
573 			break;
574 
575 		case ZPOOL_PROP_COMMENT:
576 			if ((error = nvpair_value_string(elem, &strval)) != 0)
577 				break;
578 			for (check = strval; *check != '\0'; check++) {
579 				/*
580 				 * The kernel doesn't have an easy isprint()
581 				 * check.  For this kernel check, we merely
582 				 * check ASCII apart from DEL.  Fix this if
583 				 * there is an easy-to-use kernel isprint().
584 				 */
585 				if (*check >= 0x7f) {
586 					error = SET_ERROR(EINVAL);
587 					break;
588 				}
589 			}
590 			if (strlen(strval) > ZPROP_MAX_COMMENT)
591 				error = E2BIG;
592 			break;
593 
594 		case ZPOOL_PROP_DEDUPDITTO:
595 			if (spa_version(spa) < SPA_VERSION_DEDUP)
596 				error = SET_ERROR(ENOTSUP);
597 			else
598 				error = nvpair_value_uint64(elem, &intval);
599 			if (error == 0 &&
600 			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
601 				error = SET_ERROR(EINVAL);
602 			break;
603 		}
604 
605 		if (error)
606 			break;
607 	}
608 
609 	if (!error && reset_bootfs) {
610 		error = nvlist_remove(props,
611 		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
612 
613 		if (!error) {
614 			error = nvlist_add_uint64(props,
615 			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
616 		}
617 	}
618 
619 	return (error);
620 }
621 
622 void
623 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
624 {
625 	char *cachefile;
626 	spa_config_dirent_t *dp;
627 
628 	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
629 	    &cachefile) != 0)
630 		return;
631 
632 	dp = kmem_alloc(sizeof (spa_config_dirent_t),
633 	    KM_SLEEP);
634 
635 	if (cachefile[0] == '\0')
636 		dp->scd_path = spa_strdup(spa_config_path);
637 	else if (strcmp(cachefile, "none") == 0)
638 		dp->scd_path = NULL;
639 	else
640 		dp->scd_path = spa_strdup(cachefile);
641 
642 	list_insert_head(&spa->spa_config_list, dp);
643 	if (need_sync)
644 		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
645 }
646 
647 int
648 spa_prop_set(spa_t *spa, nvlist_t *nvp)
649 {
650 	int error;
651 	nvpair_t *elem = NULL;
652 	boolean_t need_sync = B_FALSE;
653 
654 	if ((error = spa_prop_validate(spa, nvp)) != 0)
655 		return (error);
656 
657 	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
658 		zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
659 
660 		if (prop == ZPOOL_PROP_CACHEFILE ||
661 		    prop == ZPOOL_PROP_ALTROOT ||
662 		    prop == ZPOOL_PROP_READONLY)
663 			continue;
664 
665 		if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
666 			uint64_t ver;
667 
668 			if (prop == ZPOOL_PROP_VERSION) {
669 				VERIFY(nvpair_value_uint64(elem, &ver) == 0);
670 			} else {
671 				ASSERT(zpool_prop_feature(nvpair_name(elem)));
672 				ver = SPA_VERSION_FEATURES;
673 				need_sync = B_TRUE;
674 			}
675 
676 			/* Save time if the version is already set. */
677 			if (ver == spa_version(spa))
678 				continue;
679 
680 			/*
681 			 * In addition to the pool directory object, we might
682 			 * create the pool properties object, the features for
683 			 * read object, the features for write object, or the
684 			 * feature descriptions object.
685 			 */
686 			error = dsl_sync_task(spa->spa_name, NULL,
687 			    spa_sync_version, &ver,
688 			    6, ZFS_SPACE_CHECK_RESERVED);
689 			if (error)
690 				return (error);
691 			continue;
692 		}
693 
694 		need_sync = B_TRUE;
695 		break;
696 	}
697 
698 	if (need_sync) {
699 		return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
700 		    nvp, 6, ZFS_SPACE_CHECK_RESERVED));
701 	}
702 
703 	return (0);
704 }
705 
706 /*
707  * If the bootfs property value is dsobj, clear it.
708  */
709 void
710 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
711 {
712 	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
713 		VERIFY(zap_remove(spa->spa_meta_objset,
714 		    spa->spa_pool_props_object,
715 		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
716 		spa->spa_bootfs = 0;
717 	}
718 }
719 
720 /*ARGSUSED*/
721 static int
722 spa_change_guid_check(void *arg, dmu_tx_t *tx)
723 {
724 	uint64_t *newguid = arg;
725 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
726 	vdev_t *rvd = spa->spa_root_vdev;
727 	uint64_t vdev_state;
728 
729 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
730 	vdev_state = rvd->vdev_state;
731 	spa_config_exit(spa, SCL_STATE, FTAG);
732 
733 	if (vdev_state != VDEV_STATE_HEALTHY)
734 		return (SET_ERROR(ENXIO));
735 
736 	ASSERT3U(spa_guid(spa), !=, *newguid);
737 
738 	return (0);
739 }
740 
741 static void
742 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
743 {
744 	uint64_t *newguid = arg;
745 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
746 	uint64_t oldguid;
747 	vdev_t *rvd = spa->spa_root_vdev;
748 
749 	oldguid = spa_guid(spa);
750 
751 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
752 	rvd->vdev_guid = *newguid;
753 	rvd->vdev_guid_sum += (*newguid - oldguid);
754 	vdev_config_dirty(rvd);
755 	spa_config_exit(spa, SCL_STATE, FTAG);
756 
757 	spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
758 	    oldguid, *newguid);
759 }
760 
761 /*
762  * Change the GUID for the pool.  This is done so that we can later
763  * re-import a pool built from a clone of our own vdevs.  We will modify
764  * the root vdev's guid, our own pool guid, and then mark all of our
765  * vdevs dirty.  Note that we must make sure that all our vdevs are
766  * online when we do this, or else any vdevs that weren't present
767  * would be orphaned from our pool.  We are also going to issue a
768  * sysevent to update any watchers.
769  */
770 int
771 spa_change_guid(spa_t *spa)
772 {
773 	int error;
774 	uint64_t guid;
775 
776 	mutex_enter(&spa->spa_vdev_top_lock);
777 	mutex_enter(&spa_namespace_lock);
778 	guid = spa_generate_guid(NULL);
779 
780 	error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
781 	    spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
782 
783 	if (error == 0) {
784 		spa_config_sync(spa, B_FALSE, B_TRUE);
785 		spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
786 	}
787 
788 	mutex_exit(&spa_namespace_lock);
789 	mutex_exit(&spa->spa_vdev_top_lock);
790 
791 	return (error);
792 }
793 
794 /*
795  * ==========================================================================
796  * SPA state manipulation (open/create/destroy/import/export)
797  * ==========================================================================
798  */
799 
800 static int
801 spa_error_entry_compare(const void *a, const void *b)
802 {
803 	spa_error_entry_t *sa = (spa_error_entry_t *)a;
804 	spa_error_entry_t *sb = (spa_error_entry_t *)b;
805 	int ret;
806 
807 	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
808 	    sizeof (zbookmark_phys_t));
809 
810 	if (ret < 0)
811 		return (-1);
812 	else if (ret > 0)
813 		return (1);
814 	else
815 		return (0);
816 }
817 
818 /*
819  * Utility function which retrieves copies of the current logs and
820  * re-initializes them in the process.
821  */
822 void
823 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
824 {
825 	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
826 
827 	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
828 	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
829 
830 	avl_create(&spa->spa_errlist_scrub,
831 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
832 	    offsetof(spa_error_entry_t, se_avl));
833 	avl_create(&spa->spa_errlist_last,
834 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
835 	    offsetof(spa_error_entry_t, se_avl));
836 }
837 
838 static void
839 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
840 {
841 	const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
842 	enum zti_modes mode = ztip->zti_mode;
843 	uint_t value = ztip->zti_value;
844 	uint_t count = ztip->zti_count;
845 	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
846 	char name[32];
847 	uint_t flags = 0;
848 	boolean_t batch = B_FALSE;
849 
850 	if (mode == ZTI_MODE_NULL) {
851 		tqs->stqs_count = 0;
852 		tqs->stqs_taskq = NULL;
853 		return;
854 	}
855 
856 	ASSERT3U(count, >, 0);
857 
858 	tqs->stqs_count = count;
859 	tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
860 
861 	switch (mode) {
862 	case ZTI_MODE_FIXED:
863 		ASSERT3U(value, >=, 1);
864 		value = MAX(value, 1);
865 		break;
866 
867 	case ZTI_MODE_BATCH:
868 		batch = B_TRUE;
869 		flags |= TASKQ_THREADS_CPU_PCT;
870 		value = zio_taskq_batch_pct;
871 		break;
872 
873 	default:
874 		panic("unrecognized mode for %s_%s taskq (%u:%u) in "
875 		    "spa_activate()",
876 		    zio_type_name[t], zio_taskq_types[q], mode, value);
877 		break;
878 	}
879 
880 	for (uint_t i = 0; i < count; i++) {
881 		taskq_t *tq;
882 
883 		if (count > 1) {
884 			(void) snprintf(name, sizeof (name), "%s_%s_%u",
885 			    zio_type_name[t], zio_taskq_types[q], i);
886 		} else {
887 			(void) snprintf(name, sizeof (name), "%s_%s",
888 			    zio_type_name[t], zio_taskq_types[q]);
889 		}
890 
891 		if (zio_taskq_sysdc && spa->spa_proc != &p0) {
892 			if (batch)
893 				flags |= TASKQ_DC_BATCH;
894 
895 			tq = taskq_create_sysdc(name, value, 50, INT_MAX,
896 			    spa->spa_proc, zio_taskq_basedc, flags);
897 		} else {
898 			pri_t pri = maxclsyspri;
899 			/*
900 			 * The write issue taskq can be extremely CPU
901 			 * intensive.  Run it at slightly lower priority
902 			 * than the other taskqs.
903 			 */
904 			if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
905 				pri--;
906 
907 			tq = taskq_create_proc(name, value, pri, 50,
908 			    INT_MAX, spa->spa_proc, flags);
909 		}
910 
911 		tqs->stqs_taskq[i] = tq;
912 	}
913 }
914 
915 static void
916 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
917 {
918 	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
919 
920 	if (tqs->stqs_taskq == NULL) {
921 		ASSERT0(tqs->stqs_count);
922 		return;
923 	}
924 
925 	for (uint_t i = 0; i < tqs->stqs_count; i++) {
926 		ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
927 		taskq_destroy(tqs->stqs_taskq[i]);
928 	}
929 
930 	kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
931 	tqs->stqs_taskq = NULL;
932 }
933 
934 /*
935  * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
936  * Note that a type may have multiple discrete taskqs to avoid lock contention
937  * on the taskq itself. In that case we choose which taskq at random by using
938  * the low bits of gethrtime().
939  */
940 void
941 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
942     task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
943 {
944 	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
945 	taskq_t *tq;
946 
947 	ASSERT3P(tqs->stqs_taskq, !=, NULL);
948 	ASSERT3U(tqs->stqs_count, !=, 0);
949 
950 	if (tqs->stqs_count == 1) {
951 		tq = tqs->stqs_taskq[0];
952 	} else {
953 		tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
954 	}
955 
956 	taskq_dispatch_ent(tq, func, arg, flags, ent);
957 }
958 
959 static void
960 spa_create_zio_taskqs(spa_t *spa)
961 {
962 	for (int t = 0; t < ZIO_TYPES; t++) {
963 		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
964 			spa_taskqs_init(spa, t, q);
965 		}
966 	}
967 }
968 
969 #ifdef _KERNEL
970 static void
971 spa_thread(void *arg)
972 {
973 	callb_cpr_t cprinfo;
974 
975 	spa_t *spa = arg;
976 	user_t *pu = PTOU(curproc);
977 
978 	CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
979 	    spa->spa_name);
980 
981 	ASSERT(curproc != &p0);
982 	(void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
983 	    "zpool-%s", spa->spa_name);
984 	(void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
985 
986 	/* bind this thread to the requested psrset */
987 	if (zio_taskq_psrset_bind != PS_NONE) {
988 		pool_lock();
989 		mutex_enter(&cpu_lock);
990 		mutex_enter(&pidlock);
991 		mutex_enter(&curproc->p_lock);
992 
993 		if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
994 		    0, NULL, NULL) == 0)  {
995 			curthread->t_bind_pset = zio_taskq_psrset_bind;
996 		} else {
997 			cmn_err(CE_WARN,
998 			    "Couldn't bind process for zfs pool \"%s\" to "
999 			    "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1000 		}
1001 
1002 		mutex_exit(&curproc->p_lock);
1003 		mutex_exit(&pidlock);
1004 		mutex_exit(&cpu_lock);
1005 		pool_unlock();
1006 	}
1007 
1008 	if (zio_taskq_sysdc) {
1009 		sysdc_thread_enter(curthread, 100, 0);
1010 	}
1011 
1012 	spa->spa_proc = curproc;
1013 	spa->spa_did = curthread->t_did;
1014 
1015 	spa_create_zio_taskqs(spa);
1016 
1017 	mutex_enter(&spa->spa_proc_lock);
1018 	ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1019 
1020 	spa->spa_proc_state = SPA_PROC_ACTIVE;
1021 	cv_broadcast(&spa->spa_proc_cv);
1022 
1023 	CALLB_CPR_SAFE_BEGIN(&cprinfo);
1024 	while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1025 		cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1026 	CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1027 
1028 	ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1029 	spa->spa_proc_state = SPA_PROC_GONE;
1030 	spa->spa_proc = &p0;
1031 	cv_broadcast(&spa->spa_proc_cv);
1032 	CALLB_CPR_EXIT(&cprinfo);	/* drops spa_proc_lock */
1033 
1034 	mutex_enter(&curproc->p_lock);
1035 	lwp_exit();
1036 }
1037 #endif
1038 
1039 /*
1040  * Activate an uninitialized pool.
1041  */
1042 static void
1043 spa_activate(spa_t *spa, int mode)
1044 {
1045 	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1046 
1047 	spa->spa_state = POOL_STATE_ACTIVE;
1048 	spa->spa_mode = mode;
1049 
1050 	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1051 	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1052 
1053 	/* Try to create a covering process */
1054 	mutex_enter(&spa->spa_proc_lock);
1055 	ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1056 	ASSERT(spa->spa_proc == &p0);
1057 	spa->spa_did = 0;
1058 
1059 	/* Only create a process if we're going to be around a while. */
1060 	if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1061 		if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1062 		    NULL, 0) == 0) {
1063 			spa->spa_proc_state = SPA_PROC_CREATED;
1064 			while (spa->spa_proc_state == SPA_PROC_CREATED) {
1065 				cv_wait(&spa->spa_proc_cv,
1066 				    &spa->spa_proc_lock);
1067 			}
1068 			ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1069 			ASSERT(spa->spa_proc != &p0);
1070 			ASSERT(spa->spa_did != 0);
1071 		} else {
1072 #ifdef _KERNEL
1073 			cmn_err(CE_WARN,
1074 			    "Couldn't create process for zfs pool \"%s\"\n",
1075 			    spa->spa_name);
1076 #endif
1077 		}
1078 	}
1079 	mutex_exit(&spa->spa_proc_lock);
1080 
1081 	/* If we didn't create a process, we need to create our taskqs. */
1082 	if (spa->spa_proc == &p0) {
1083 		spa_create_zio_taskqs(spa);
1084 	}
1085 
1086 	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1087 	    offsetof(vdev_t, vdev_config_dirty_node));
1088 	list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1089 	    offsetof(objset_t, os_evicting_node));
1090 	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1091 	    offsetof(vdev_t, vdev_state_dirty_node));
1092 
1093 	txg_list_create(&spa->spa_vdev_txg_list, spa,
1094 	    offsetof(struct vdev, vdev_txg_node));
1095 
1096 	avl_create(&spa->spa_errlist_scrub,
1097 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1098 	    offsetof(spa_error_entry_t, se_avl));
1099 	avl_create(&spa->spa_errlist_last,
1100 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1101 	    offsetof(spa_error_entry_t, se_avl));
1102 }
1103 
1104 /*
1105  * Opposite of spa_activate().
1106  */
1107 static void
1108 spa_deactivate(spa_t *spa)
1109 {
1110 	ASSERT(spa->spa_sync_on == B_FALSE);
1111 	ASSERT(spa->spa_dsl_pool == NULL);
1112 	ASSERT(spa->spa_root_vdev == NULL);
1113 	ASSERT(spa->spa_async_zio_root == NULL);
1114 	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1115 
1116 	spa_evicting_os_wait(spa);
1117 
1118 	txg_list_destroy(&spa->spa_vdev_txg_list);
1119 
1120 	list_destroy(&spa->spa_config_dirty_list);
1121 	list_destroy(&spa->spa_evicting_os_list);
1122 	list_destroy(&spa->spa_state_dirty_list);
1123 
1124 	for (int t = 0; t < ZIO_TYPES; t++) {
1125 		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1126 			spa_taskqs_fini(spa, t, q);
1127 		}
1128 	}
1129 
1130 	metaslab_class_destroy(spa->spa_normal_class);
1131 	spa->spa_normal_class = NULL;
1132 
1133 	metaslab_class_destroy(spa->spa_log_class);
1134 	spa->spa_log_class = NULL;
1135 
1136 	/*
1137 	 * If this was part of an import or the open otherwise failed, we may
1138 	 * still have errors left in the queues.  Empty them just in case.
1139 	 */
1140 	spa_errlog_drain(spa);
1141 
1142 	avl_destroy(&spa->spa_errlist_scrub);
1143 	avl_destroy(&spa->spa_errlist_last);
1144 
1145 	spa->spa_state = POOL_STATE_UNINITIALIZED;
1146 
1147 	mutex_enter(&spa->spa_proc_lock);
1148 	if (spa->spa_proc_state != SPA_PROC_NONE) {
1149 		ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1150 		spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1151 		cv_broadcast(&spa->spa_proc_cv);
1152 		while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1153 			ASSERT(spa->spa_proc != &p0);
1154 			cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1155 		}
1156 		ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1157 		spa->spa_proc_state = SPA_PROC_NONE;
1158 	}
1159 	ASSERT(spa->spa_proc == &p0);
1160 	mutex_exit(&spa->spa_proc_lock);
1161 
1162 	/*
1163 	 * We want to make sure spa_thread() has actually exited the ZFS
1164 	 * module, so that the module can't be unloaded out from underneath
1165 	 * it.
1166 	 */
1167 	if (spa->spa_did != 0) {
1168 		thread_join(spa->spa_did);
1169 		spa->spa_did = 0;
1170 	}
1171 }
1172 
1173 /*
1174  * Verify a pool configuration, and construct the vdev tree appropriately.  This
1175  * will create all the necessary vdevs in the appropriate layout, with each vdev
1176  * in the CLOSED state.  This will prep the pool before open/creation/import.
1177  * All vdev validation is done by the vdev_alloc() routine.
1178  */
1179 static int
1180 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1181     uint_t id, int atype)
1182 {
1183 	nvlist_t **child;
1184 	uint_t children;
1185 	int error;
1186 
1187 	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1188 		return (error);
1189 
1190 	if ((*vdp)->vdev_ops->vdev_op_leaf)
1191 		return (0);
1192 
1193 	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1194 	    &child, &children);
1195 
1196 	if (error == ENOENT)
1197 		return (0);
1198 
1199 	if (error) {
1200 		vdev_free(*vdp);
1201 		*vdp = NULL;
1202 		return (SET_ERROR(EINVAL));
1203 	}
1204 
1205 	for (int c = 0; c < children; c++) {
1206 		vdev_t *vd;
1207 		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1208 		    atype)) != 0) {
1209 			vdev_free(*vdp);
1210 			*vdp = NULL;
1211 			return (error);
1212 		}
1213 	}
1214 
1215 	ASSERT(*vdp != NULL);
1216 
1217 	return (0);
1218 }
1219 
1220 /*
1221  * Opposite of spa_load().
1222  */
1223 static void
1224 spa_unload(spa_t *spa)
1225 {
1226 	int i;
1227 
1228 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1229 
1230 	/*
1231 	 * Stop async tasks.
1232 	 */
1233 	spa_async_suspend(spa);
1234 
1235 	/*
1236 	 * Stop syncing.
1237 	 */
1238 	if (spa->spa_sync_on) {
1239 		txg_sync_stop(spa->spa_dsl_pool);
1240 		spa->spa_sync_on = B_FALSE;
1241 	}
1242 
1243 	/*
1244 	 * Even though vdev_free() also calls vdev_metaslab_fini, we need
1245 	 * to call it earlier, before we wait for async i/o to complete.
1246 	 * This ensures that there is no async metaslab prefetching, by
1247 	 * calling taskq_wait(mg_taskq).
1248 	 */
1249 	if (spa->spa_root_vdev != NULL) {
1250 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1251 		for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++)
1252 			vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]);
1253 		spa_config_exit(spa, SCL_ALL, FTAG);
1254 	}
1255 
1256 	/*
1257 	 * Wait for any outstanding async I/O to complete.
1258 	 */
1259 	if (spa->spa_async_zio_root != NULL) {
1260 		for (int i = 0; i < max_ncpus; i++)
1261 			(void) zio_wait(spa->spa_async_zio_root[i]);
1262 		kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1263 		spa->spa_async_zio_root = NULL;
1264 	}
1265 
1266 	bpobj_close(&spa->spa_deferred_bpobj);
1267 
1268 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1269 
1270 	/*
1271 	 * Close all vdevs.
1272 	 */
1273 	if (spa->spa_root_vdev)
1274 		vdev_free(spa->spa_root_vdev);
1275 	ASSERT(spa->spa_root_vdev == NULL);
1276 
1277 	/*
1278 	 * Close the dsl pool.
1279 	 */
1280 	if (spa->spa_dsl_pool) {
1281 		dsl_pool_close(spa->spa_dsl_pool);
1282 		spa->spa_dsl_pool = NULL;
1283 		spa->spa_meta_objset = NULL;
1284 	}
1285 
1286 	ddt_unload(spa);
1287 
1288 	/*
1289 	 * Drop and purge level 2 cache
1290 	 */
1291 	spa_l2cache_drop(spa);
1292 
1293 	for (i = 0; i < spa->spa_spares.sav_count; i++)
1294 		vdev_free(spa->spa_spares.sav_vdevs[i]);
1295 	if (spa->spa_spares.sav_vdevs) {
1296 		kmem_free(spa->spa_spares.sav_vdevs,
1297 		    spa->spa_spares.sav_count * sizeof (void *));
1298 		spa->spa_spares.sav_vdevs = NULL;
1299 	}
1300 	if (spa->spa_spares.sav_config) {
1301 		nvlist_free(spa->spa_spares.sav_config);
1302 		spa->spa_spares.sav_config = NULL;
1303 	}
1304 	spa->spa_spares.sav_count = 0;
1305 
1306 	for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1307 		vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1308 		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1309 	}
1310 	if (spa->spa_l2cache.sav_vdevs) {
1311 		kmem_free(spa->spa_l2cache.sav_vdevs,
1312 		    spa->spa_l2cache.sav_count * sizeof (void *));
1313 		spa->spa_l2cache.sav_vdevs = NULL;
1314 	}
1315 	if (spa->spa_l2cache.sav_config) {
1316 		nvlist_free(spa->spa_l2cache.sav_config);
1317 		spa->spa_l2cache.sav_config = NULL;
1318 	}
1319 	spa->spa_l2cache.sav_count = 0;
1320 
1321 	spa->spa_async_suspended = 0;
1322 
1323 	if (spa->spa_comment != NULL) {
1324 		spa_strfree(spa->spa_comment);
1325 		spa->spa_comment = NULL;
1326 	}
1327 
1328 	spa_config_exit(spa, SCL_ALL, FTAG);
1329 }
1330 
1331 /*
1332  * Load (or re-load) the current list of vdevs describing the active spares for
1333  * this pool.  When this is called, we have some form of basic information in
1334  * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1335  * then re-generate a more complete list including status information.
1336  */
1337 static void
1338 spa_load_spares(spa_t *spa)
1339 {
1340 	nvlist_t **spares;
1341 	uint_t nspares;
1342 	int i;
1343 	vdev_t *vd, *tvd;
1344 
1345 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1346 
1347 	/*
1348 	 * First, close and free any existing spare vdevs.
1349 	 */
1350 	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1351 		vd = spa->spa_spares.sav_vdevs[i];
1352 
1353 		/* Undo the call to spa_activate() below */
1354 		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1355 		    B_FALSE)) != NULL && tvd->vdev_isspare)
1356 			spa_spare_remove(tvd);
1357 		vdev_close(vd);
1358 		vdev_free(vd);
1359 	}
1360 
1361 	if (spa->spa_spares.sav_vdevs)
1362 		kmem_free(spa->spa_spares.sav_vdevs,
1363 		    spa->spa_spares.sav_count * sizeof (void *));
1364 
1365 	if (spa->spa_spares.sav_config == NULL)
1366 		nspares = 0;
1367 	else
1368 		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1369 		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1370 
1371 	spa->spa_spares.sav_count = (int)nspares;
1372 	spa->spa_spares.sav_vdevs = NULL;
1373 
1374 	if (nspares == 0)
1375 		return;
1376 
1377 	/*
1378 	 * Construct the array of vdevs, opening them to get status in the
1379 	 * process.   For each spare, there is potentially two different vdev_t
1380 	 * structures associated with it: one in the list of spares (used only
1381 	 * for basic validation purposes) and one in the active vdev
1382 	 * configuration (if it's spared in).  During this phase we open and
1383 	 * validate each vdev on the spare list.  If the vdev also exists in the
1384 	 * active configuration, then we also mark this vdev as an active spare.
1385 	 */
1386 	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1387 	    KM_SLEEP);
1388 	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1389 		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1390 		    VDEV_ALLOC_SPARE) == 0);
1391 		ASSERT(vd != NULL);
1392 
1393 		spa->spa_spares.sav_vdevs[i] = vd;
1394 
1395 		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1396 		    B_FALSE)) != NULL) {
1397 			if (!tvd->vdev_isspare)
1398 				spa_spare_add(tvd);
1399 
1400 			/*
1401 			 * We only mark the spare active if we were successfully
1402 			 * able to load the vdev.  Otherwise, importing a pool
1403 			 * with a bad active spare would result in strange
1404 			 * behavior, because multiple pool would think the spare
1405 			 * is actively in use.
1406 			 *
1407 			 * There is a vulnerability here to an equally bizarre
1408 			 * circumstance, where a dead active spare is later
1409 			 * brought back to life (onlined or otherwise).  Given
1410 			 * the rarity of this scenario, and the extra complexity
1411 			 * it adds, we ignore the possibility.
1412 			 */
1413 			if (!vdev_is_dead(tvd))
1414 				spa_spare_activate(tvd);
1415 		}
1416 
1417 		vd->vdev_top = vd;
1418 		vd->vdev_aux = &spa->spa_spares;
1419 
1420 		if (vdev_open(vd) != 0)
1421 			continue;
1422 
1423 		if (vdev_validate_aux(vd) == 0)
1424 			spa_spare_add(vd);
1425 	}
1426 
1427 	/*
1428 	 * Recompute the stashed list of spares, with status information
1429 	 * this time.
1430 	 */
1431 	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1432 	    DATA_TYPE_NVLIST_ARRAY) == 0);
1433 
1434 	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1435 	    KM_SLEEP);
1436 	for (i = 0; i < spa->spa_spares.sav_count; i++)
1437 		spares[i] = vdev_config_generate(spa,
1438 		    spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1439 	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1440 	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1441 	for (i = 0; i < spa->spa_spares.sav_count; i++)
1442 		nvlist_free(spares[i]);
1443 	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1444 }
1445 
1446 /*
1447  * Load (or re-load) the current list of vdevs describing the active l2cache for
1448  * this pool.  When this is called, we have some form of basic information in
1449  * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1450  * then re-generate a more complete list including status information.
1451  * Devices which are already active have their details maintained, and are
1452  * not re-opened.
1453  */
1454 static void
1455 spa_load_l2cache(spa_t *spa)
1456 {
1457 	nvlist_t **l2cache;
1458 	uint_t nl2cache;
1459 	int i, j, oldnvdevs;
1460 	uint64_t guid;
1461 	vdev_t *vd, **oldvdevs, **newvdevs;
1462 	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1463 
1464 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1465 
1466 	if (sav->sav_config != NULL) {
1467 		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1468 		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1469 		newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1470 	} else {
1471 		nl2cache = 0;
1472 		newvdevs = NULL;
1473 	}
1474 
1475 	oldvdevs = sav->sav_vdevs;
1476 	oldnvdevs = sav->sav_count;
1477 	sav->sav_vdevs = NULL;
1478 	sav->sav_count = 0;
1479 
1480 	/*
1481 	 * Process new nvlist of vdevs.
1482 	 */
1483 	for (i = 0; i < nl2cache; i++) {
1484 		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1485 		    &guid) == 0);
1486 
1487 		newvdevs[i] = NULL;
1488 		for (j = 0; j < oldnvdevs; j++) {
1489 			vd = oldvdevs[j];
1490 			if (vd != NULL && guid == vd->vdev_guid) {
1491 				/*
1492 				 * Retain previous vdev for add/remove ops.
1493 				 */
1494 				newvdevs[i] = vd;
1495 				oldvdevs[j] = NULL;
1496 				break;
1497 			}
1498 		}
1499 
1500 		if (newvdevs[i] == NULL) {
1501 			/*
1502 			 * Create new vdev
1503 			 */
1504 			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1505 			    VDEV_ALLOC_L2CACHE) == 0);
1506 			ASSERT(vd != NULL);
1507 			newvdevs[i] = vd;
1508 
1509 			/*
1510 			 * Commit this vdev as an l2cache device,
1511 			 * even if it fails to open.
1512 			 */
1513 			spa_l2cache_add(vd);
1514 
1515 			vd->vdev_top = vd;
1516 			vd->vdev_aux = sav;
1517 
1518 			spa_l2cache_activate(vd);
1519 
1520 			if (vdev_open(vd) != 0)
1521 				continue;
1522 
1523 			(void) vdev_validate_aux(vd);
1524 
1525 			if (!vdev_is_dead(vd))
1526 				l2arc_add_vdev(spa, vd);
1527 		}
1528 	}
1529 
1530 	/*
1531 	 * Purge vdevs that were dropped
1532 	 */
1533 	for (i = 0; i < oldnvdevs; i++) {
1534 		uint64_t pool;
1535 
1536 		vd = oldvdevs[i];
1537 		if (vd != NULL) {
1538 			ASSERT(vd->vdev_isl2cache);
1539 
1540 			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1541 			    pool != 0ULL && l2arc_vdev_present(vd))
1542 				l2arc_remove_vdev(vd);
1543 			vdev_clear_stats(vd);
1544 			vdev_free(vd);
1545 		}
1546 	}
1547 
1548 	if (oldvdevs)
1549 		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1550 
1551 	if (sav->sav_config == NULL)
1552 		goto out;
1553 
1554 	sav->sav_vdevs = newvdevs;
1555 	sav->sav_count = (int)nl2cache;
1556 
1557 	/*
1558 	 * Recompute the stashed list of l2cache devices, with status
1559 	 * information this time.
1560 	 */
1561 	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1562 	    DATA_TYPE_NVLIST_ARRAY) == 0);
1563 
1564 	l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1565 	for (i = 0; i < sav->sav_count; i++)
1566 		l2cache[i] = vdev_config_generate(spa,
1567 		    sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1568 	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1569 	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1570 out:
1571 	for (i = 0; i < sav->sav_count; i++)
1572 		nvlist_free(l2cache[i]);
1573 	if (sav->sav_count)
1574 		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1575 }
1576 
1577 static int
1578 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1579 {
1580 	dmu_buf_t *db;
1581 	char *packed = NULL;
1582 	size_t nvsize = 0;
1583 	int error;
1584 	*value = NULL;
1585 
1586 	error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1587 	if (error != 0)
1588 		return (error);
1589 
1590 	nvsize = *(uint64_t *)db->db_data;
1591 	dmu_buf_rele(db, FTAG);
1592 
1593 	packed = kmem_alloc(nvsize, KM_SLEEP);
1594 	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1595 	    DMU_READ_PREFETCH);
1596 	if (error == 0)
1597 		error = nvlist_unpack(packed, nvsize, value, 0);
1598 	kmem_free(packed, nvsize);
1599 
1600 	return (error);
1601 }
1602 
1603 /*
1604  * Checks to see if the given vdev could not be opened, in which case we post a
1605  * sysevent to notify the autoreplace code that the device has been removed.
1606  */
1607 static void
1608 spa_check_removed(vdev_t *vd)
1609 {
1610 	for (int c = 0; c < vd->vdev_children; c++)
1611 		spa_check_removed(vd->vdev_child[c]);
1612 
1613 	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1614 	    !vd->vdev_ishole) {
1615 		zfs_post_autoreplace(vd->vdev_spa, vd);
1616 		spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1617 	}
1618 }
1619 
1620 static void
1621 spa_config_valid_zaps(vdev_t *vd, vdev_t *mvd)
1622 {
1623 	ASSERT3U(vd->vdev_children, ==, mvd->vdev_children);
1624 
1625 	vd->vdev_top_zap = mvd->vdev_top_zap;
1626 	vd->vdev_leaf_zap = mvd->vdev_leaf_zap;
1627 
1628 	for (uint64_t i = 0; i < vd->vdev_children; i++) {
1629 		spa_config_valid_zaps(vd->vdev_child[i], mvd->vdev_child[i]);
1630 	}
1631 }
1632 
1633 /*
1634  * Validate the current config against the MOS config
1635  */
1636 static boolean_t
1637 spa_config_valid(spa_t *spa, nvlist_t *config)
1638 {
1639 	vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1640 	nvlist_t *nv;
1641 
1642 	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1643 
1644 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1645 	VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1646 
1647 	ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1648 
1649 	/*
1650 	 * If we're doing a normal import, then build up any additional
1651 	 * diagnostic information about missing devices in this config.
1652 	 * We'll pass this up to the user for further processing.
1653 	 */
1654 	if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1655 		nvlist_t **child, *nv;
1656 		uint64_t idx = 0;
1657 
1658 		child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1659 		    KM_SLEEP);
1660 		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1661 
1662 		for (int c = 0; c < rvd->vdev_children; c++) {
1663 			vdev_t *tvd = rvd->vdev_child[c];
1664 			vdev_t *mtvd  = mrvd->vdev_child[c];
1665 
1666 			if (tvd->vdev_ops == &vdev_missing_ops &&
1667 			    mtvd->vdev_ops != &vdev_missing_ops &&
1668 			    mtvd->vdev_islog)
1669 				child[idx++] = vdev_config_generate(spa, mtvd,
1670 				    B_FALSE, 0);
1671 		}
1672 
1673 		if (idx) {
1674 			VERIFY(nvlist_add_nvlist_array(nv,
1675 			    ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1676 			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1677 			    ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1678 
1679 			for (int i = 0; i < idx; i++)
1680 				nvlist_free(child[i]);
1681 		}
1682 		nvlist_free(nv);
1683 		kmem_free(child, rvd->vdev_children * sizeof (char **));
1684 	}
1685 
1686 	/*
1687 	 * Compare the root vdev tree with the information we have
1688 	 * from the MOS config (mrvd). Check each top-level vdev
1689 	 * with the corresponding MOS config top-level (mtvd).
1690 	 */
1691 	for (int c = 0; c < rvd->vdev_children; c++) {
1692 		vdev_t *tvd = rvd->vdev_child[c];
1693 		vdev_t *mtvd  = mrvd->vdev_child[c];
1694 
1695 		/*
1696 		 * Resolve any "missing" vdevs in the current configuration.
1697 		 * If we find that the MOS config has more accurate information
1698 		 * about the top-level vdev then use that vdev instead.
1699 		 */
1700 		if (tvd->vdev_ops == &vdev_missing_ops &&
1701 		    mtvd->vdev_ops != &vdev_missing_ops) {
1702 
1703 			if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1704 				continue;
1705 
1706 			/*
1707 			 * Device specific actions.
1708 			 */
1709 			if (mtvd->vdev_islog) {
1710 				spa_set_log_state(spa, SPA_LOG_CLEAR);
1711 			} else {
1712 				/*
1713 				 * XXX - once we have 'readonly' pool
1714 				 * support we should be able to handle
1715 				 * missing data devices by transitioning
1716 				 * the pool to readonly.
1717 				 */
1718 				continue;
1719 			}
1720 
1721 			/*
1722 			 * Swap the missing vdev with the data we were
1723 			 * able to obtain from the MOS config.
1724 			 */
1725 			vdev_remove_child(rvd, tvd);
1726 			vdev_remove_child(mrvd, mtvd);
1727 
1728 			vdev_add_child(rvd, mtvd);
1729 			vdev_add_child(mrvd, tvd);
1730 
1731 			spa_config_exit(spa, SCL_ALL, FTAG);
1732 			vdev_load(mtvd);
1733 			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1734 
1735 			vdev_reopen(rvd);
1736 		} else {
1737 			if (mtvd->vdev_islog) {
1738 				/*
1739 				 * Load the slog device's state from the MOS
1740 				 * config since it's possible that the label
1741 				 * does not contain the most up-to-date
1742 				 * information.
1743 				 */
1744 				vdev_load_log_state(tvd, mtvd);
1745 				vdev_reopen(tvd);
1746 			}
1747 
1748 			/*
1749 			 * Per-vdev ZAP info is stored exclusively in the MOS.
1750 			 */
1751 			spa_config_valid_zaps(tvd, mtvd);
1752 		}
1753 	}
1754 
1755 	vdev_free(mrvd);
1756 	spa_config_exit(spa, SCL_ALL, FTAG);
1757 
1758 	/*
1759 	 * Ensure we were able to validate the config.
1760 	 */
1761 	return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1762 }
1763 
1764 /*
1765  * Check for missing log devices
1766  */
1767 static boolean_t
1768 spa_check_logs(spa_t *spa)
1769 {
1770 	boolean_t rv = B_FALSE;
1771 	dsl_pool_t *dp = spa_get_dsl(spa);
1772 
1773 	switch (spa->spa_log_state) {
1774 	case SPA_LOG_MISSING:
1775 		/* need to recheck in case slog has been restored */
1776 	case SPA_LOG_UNKNOWN:
1777 		rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1778 		    zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
1779 		if (rv)
1780 			spa_set_log_state(spa, SPA_LOG_MISSING);
1781 		break;
1782 	}
1783 	return (rv);
1784 }
1785 
1786 static boolean_t
1787 spa_passivate_log(spa_t *spa)
1788 {
1789 	vdev_t *rvd = spa->spa_root_vdev;
1790 	boolean_t slog_found = B_FALSE;
1791 
1792 	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1793 
1794 	if (!spa_has_slogs(spa))
1795 		return (B_FALSE);
1796 
1797 	for (int c = 0; c < rvd->vdev_children; c++) {
1798 		vdev_t *tvd = rvd->vdev_child[c];
1799 		metaslab_group_t *mg = tvd->vdev_mg;
1800 
1801 		if (tvd->vdev_islog) {
1802 			metaslab_group_passivate(mg);
1803 			slog_found = B_TRUE;
1804 		}
1805 	}
1806 
1807 	return (slog_found);
1808 }
1809 
1810 static void
1811 spa_activate_log(spa_t *spa)
1812 {
1813 	vdev_t *rvd = spa->spa_root_vdev;
1814 
1815 	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1816 
1817 	for (int c = 0; c < rvd->vdev_children; c++) {
1818 		vdev_t *tvd = rvd->vdev_child[c];
1819 		metaslab_group_t *mg = tvd->vdev_mg;
1820 
1821 		if (tvd->vdev_islog)
1822 			metaslab_group_activate(mg);
1823 	}
1824 }
1825 
1826 int
1827 spa_offline_log(spa_t *spa)
1828 {
1829 	int error;
1830 
1831 	error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1832 	    NULL, DS_FIND_CHILDREN);
1833 	if (error == 0) {
1834 		/*
1835 		 * We successfully offlined the log device, sync out the
1836 		 * current txg so that the "stubby" block can be removed
1837 		 * by zil_sync().
1838 		 */
1839 		txg_wait_synced(spa->spa_dsl_pool, 0);
1840 	}
1841 	return (error);
1842 }
1843 
1844 static void
1845 spa_aux_check_removed(spa_aux_vdev_t *sav)
1846 {
1847 	for (int i = 0; i < sav->sav_count; i++)
1848 		spa_check_removed(sav->sav_vdevs[i]);
1849 }
1850 
1851 void
1852 spa_claim_notify(zio_t *zio)
1853 {
1854 	spa_t *spa = zio->io_spa;
1855 
1856 	if (zio->io_error)
1857 		return;
1858 
1859 	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
1860 	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1861 		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1862 	mutex_exit(&spa->spa_props_lock);
1863 }
1864 
1865 typedef struct spa_load_error {
1866 	uint64_t	sle_meta_count;
1867 	uint64_t	sle_data_count;
1868 } spa_load_error_t;
1869 
1870 static void
1871 spa_load_verify_done(zio_t *zio)
1872 {
1873 	blkptr_t *bp = zio->io_bp;
1874 	spa_load_error_t *sle = zio->io_private;
1875 	dmu_object_type_t type = BP_GET_TYPE(bp);
1876 	int error = zio->io_error;
1877 	spa_t *spa = zio->io_spa;
1878 
1879 	if (error) {
1880 		if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1881 		    type != DMU_OT_INTENT_LOG)
1882 			atomic_inc_64(&sle->sle_meta_count);
1883 		else
1884 			atomic_inc_64(&sle->sle_data_count);
1885 	}
1886 	zio_data_buf_free(zio->io_data, zio->io_size);
1887 
1888 	mutex_enter(&spa->spa_scrub_lock);
1889 	spa->spa_scrub_inflight--;
1890 	cv_broadcast(&spa->spa_scrub_io_cv);
1891 	mutex_exit(&spa->spa_scrub_lock);
1892 }
1893 
1894 /*
1895  * Maximum number of concurrent scrub i/os to create while verifying
1896  * a pool while importing it.
1897  */
1898 int spa_load_verify_maxinflight = 10000;
1899 boolean_t spa_load_verify_metadata = B_TRUE;
1900 boolean_t spa_load_verify_data = B_TRUE;
1901 
1902 /*ARGSUSED*/
1903 static int
1904 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1905     const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
1906 {
1907 	if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
1908 		return (0);
1909 	/*
1910 	 * Note: normally this routine will not be called if
1911 	 * spa_load_verify_metadata is not set.  However, it may be useful
1912 	 * to manually set the flag after the traversal has begun.
1913 	 */
1914 	if (!spa_load_verify_metadata)
1915 		return (0);
1916 	if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data)
1917 		return (0);
1918 
1919 	zio_t *rio = arg;
1920 	size_t size = BP_GET_PSIZE(bp);
1921 	void *data = zio_data_buf_alloc(size);
1922 
1923 	mutex_enter(&spa->spa_scrub_lock);
1924 	while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
1925 		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1926 	spa->spa_scrub_inflight++;
1927 	mutex_exit(&spa->spa_scrub_lock);
1928 
1929 	zio_nowait(zio_read(rio, spa, bp, data, size,
1930 	    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1931 	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1932 	    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1933 	return (0);
1934 }
1935 
1936 /* ARGSUSED */
1937 int
1938 verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1939 {
1940 	if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN)
1941 		return (SET_ERROR(ENAMETOOLONG));
1942 
1943 	return (0);
1944 }
1945 
1946 static int
1947 spa_load_verify(spa_t *spa)
1948 {
1949 	zio_t *rio;
1950 	spa_load_error_t sle = { 0 };
1951 	zpool_rewind_policy_t policy;
1952 	boolean_t verify_ok = B_FALSE;
1953 	int error = 0;
1954 
1955 	zpool_get_rewind_policy(spa->spa_config, &policy);
1956 
1957 	if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1958 		return (0);
1959 
1960 	dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
1961 	error = dmu_objset_find_dp(spa->spa_dsl_pool,
1962 	    spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL,
1963 	    DS_FIND_CHILDREN);
1964 	dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
1965 	if (error != 0)
1966 		return (error);
1967 
1968 	rio = zio_root(spa, NULL, &sle,
1969 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1970 
1971 	if (spa_load_verify_metadata) {
1972 		error = traverse_pool(spa, spa->spa_verify_min_txg,
1973 		    TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
1974 		    spa_load_verify_cb, rio);
1975 	}
1976 
1977 	(void) zio_wait(rio);
1978 
1979 	spa->spa_load_meta_errors = sle.sle_meta_count;
1980 	spa->spa_load_data_errors = sle.sle_data_count;
1981 
1982 	if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1983 	    sle.sle_data_count <= policy.zrp_maxdata) {
1984 		int64_t loss = 0;
1985 
1986 		verify_ok = B_TRUE;
1987 		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1988 		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1989 
1990 		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1991 		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1992 		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1993 		VERIFY(nvlist_add_int64(spa->spa_load_info,
1994 		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1995 		VERIFY(nvlist_add_uint64(spa->spa_load_info,
1996 		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1997 	} else {
1998 		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1999 	}
2000 
2001 	if (error) {
2002 		if (error != ENXIO && error != EIO)
2003 			error = SET_ERROR(EIO);
2004 		return (error);
2005 	}
2006 
2007 	return (verify_ok ? 0 : EIO);
2008 }
2009 
2010 /*
2011  * Find a value in the pool props object.
2012  */
2013 static void
2014 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2015 {
2016 	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2017 	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2018 }
2019 
2020 /*
2021  * Find a value in the pool directory object.
2022  */
2023 static int
2024 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
2025 {
2026 	return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2027 	    name, sizeof (uint64_t), 1, val));
2028 }
2029 
2030 static int
2031 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2032 {
2033 	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2034 	return (err);
2035 }
2036 
2037 /*
2038  * Fix up config after a partly-completed split.  This is done with the
2039  * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
2040  * pool have that entry in their config, but only the splitting one contains
2041  * a list of all the guids of the vdevs that are being split off.
2042  *
2043  * This function determines what to do with that list: either rejoin
2044  * all the disks to the pool, or complete the splitting process.  To attempt
2045  * the rejoin, each disk that is offlined is marked online again, and
2046  * we do a reopen() call.  If the vdev label for every disk that was
2047  * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2048  * then we call vdev_split() on each disk, and complete the split.
2049  *
2050  * Otherwise we leave the config alone, with all the vdevs in place in
2051  * the original pool.
2052  */
2053 static void
2054 spa_try_repair(spa_t *spa, nvlist_t *config)
2055 {
2056 	uint_t extracted;
2057 	uint64_t *glist;
2058 	uint_t i, gcount;
2059 	nvlist_t *nvl;
2060 	vdev_t **vd;
2061 	boolean_t attempt_reopen;
2062 
2063 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2064 		return;
2065 
2066 	/* check that the config is complete */
2067 	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2068 	    &glist, &gcount) != 0)
2069 		return;
2070 
2071 	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2072 
2073 	/* attempt to online all the vdevs & validate */
2074 	attempt_reopen = B_TRUE;
2075 	for (i = 0; i < gcount; i++) {
2076 		if (glist[i] == 0)	/* vdev is hole */
2077 			continue;
2078 
2079 		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2080 		if (vd[i] == NULL) {
2081 			/*
2082 			 * Don't bother attempting to reopen the disks;
2083 			 * just do the split.
2084 			 */
2085 			attempt_reopen = B_FALSE;
2086 		} else {
2087 			/* attempt to re-online it */
2088 			vd[i]->vdev_offline = B_FALSE;
2089 		}
2090 	}
2091 
2092 	if (attempt_reopen) {
2093 		vdev_reopen(spa->spa_root_vdev);
2094 
2095 		/* check each device to see what state it's in */
2096 		for (extracted = 0, i = 0; i < gcount; i++) {
2097 			if (vd[i] != NULL &&
2098 			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2099 				break;
2100 			++extracted;
2101 		}
2102 	}
2103 
2104 	/*
2105 	 * If every disk has been moved to the new pool, or if we never
2106 	 * even attempted to look at them, then we split them off for
2107 	 * good.
2108 	 */
2109 	if (!attempt_reopen || gcount == extracted) {
2110 		for (i = 0; i < gcount; i++)
2111 			if (vd[i] != NULL)
2112 				vdev_split(vd[i]);
2113 		vdev_reopen(spa->spa_root_vdev);
2114 	}
2115 
2116 	kmem_free(vd, gcount * sizeof (vdev_t *));
2117 }
2118 
2119 static int
2120 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2121     boolean_t mosconfig)
2122 {
2123 	nvlist_t *config = spa->spa_config;
2124 	char *ereport = FM_EREPORT_ZFS_POOL;
2125 	char *comment;
2126 	int error;
2127 	uint64_t pool_guid;
2128 	nvlist_t *nvl;
2129 
2130 	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2131 		return (SET_ERROR(EINVAL));
2132 
2133 	ASSERT(spa->spa_comment == NULL);
2134 	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2135 		spa->spa_comment = spa_strdup(comment);
2136 
2137 	/*
2138 	 * Versioning wasn't explicitly added to the label until later, so if
2139 	 * it's not present treat it as the initial version.
2140 	 */
2141 	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2142 	    &spa->spa_ubsync.ub_version) != 0)
2143 		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2144 
2145 	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2146 	    &spa->spa_config_txg);
2147 
2148 	if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2149 	    spa_guid_exists(pool_guid, 0)) {
2150 		error = SET_ERROR(EEXIST);
2151 	} else {
2152 		spa->spa_config_guid = pool_guid;
2153 
2154 		if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2155 		    &nvl) == 0) {
2156 			VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2157 			    KM_SLEEP) == 0);
2158 		}
2159 
2160 		nvlist_free(spa->spa_load_info);
2161 		spa->spa_load_info = fnvlist_alloc();
2162 
2163 		gethrestime(&spa->spa_loaded_ts);
2164 		error = spa_load_impl(spa, pool_guid, config, state, type,
2165 		    mosconfig, &ereport);
2166 	}
2167 
2168 	/*
2169 	 * Don't count references from objsets that are already closed
2170 	 * and are making their way through the eviction process.
2171 	 */
2172 	spa_evicting_os_wait(spa);
2173 	spa->spa_minref = refcount_count(&spa->spa_refcount);
2174 	if (error) {
2175 		if (error != EEXIST) {
2176 			spa->spa_loaded_ts.tv_sec = 0;
2177 			spa->spa_loaded_ts.tv_nsec = 0;
2178 		}
2179 		if (error != EBADF) {
2180 			zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2181 		}
2182 	}
2183 	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2184 	spa->spa_ena = 0;
2185 
2186 	return (error);
2187 }
2188 
2189 /*
2190  * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2191  * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2192  * spa's per-vdev ZAP list.
2193  */
2194 static uint64_t
2195 vdev_count_verify_zaps(vdev_t *vd)
2196 {
2197 	spa_t *spa = vd->vdev_spa;
2198 	uint64_t total = 0;
2199 	if (vd->vdev_top_zap != 0) {
2200 		total++;
2201 		ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2202 		    spa->spa_all_vdev_zaps, vd->vdev_top_zap));
2203 	}
2204 	if (vd->vdev_leaf_zap != 0) {
2205 		total++;
2206 		ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2207 		    spa->spa_all_vdev_zaps, vd->vdev_leaf_zap));
2208 	}
2209 
2210 	for (uint64_t i = 0; i < vd->vdev_children; i++) {
2211 		total += vdev_count_verify_zaps(vd->vdev_child[i]);
2212 	}
2213 
2214 	return (total);
2215 }
2216 
2217 /*
2218  * Load an existing storage pool, using the pool's builtin spa_config as a
2219  * source of configuration information.
2220  */
2221 static int
2222 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2223     spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2224     char **ereport)
2225 {
2226 	int error = 0;
2227 	nvlist_t *nvroot = NULL;
2228 	nvlist_t *label;
2229 	vdev_t *rvd;
2230 	uberblock_t *ub = &spa->spa_uberblock;
2231 	uint64_t children, config_cache_txg = spa->spa_config_txg;
2232 	int orig_mode = spa->spa_mode;
2233 	int parse;
2234 	uint64_t obj;
2235 	boolean_t missing_feat_write = B_FALSE;
2236 
2237 	/*
2238 	 * If this is an untrusted config, access the pool in read-only mode.
2239 	 * This prevents things like resilvering recently removed devices.
2240 	 */
2241 	if (!mosconfig)
2242 		spa->spa_mode = FREAD;
2243 
2244 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
2245 
2246 	spa->spa_load_state = state;
2247 
2248 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2249 		return (SET_ERROR(EINVAL));
2250 
2251 	parse = (type == SPA_IMPORT_EXISTING ?
2252 	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2253 
2254 	/*
2255 	 * Create "The Godfather" zio to hold all async IOs
2256 	 */
2257 	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2258 	    KM_SLEEP);
2259 	for (int i = 0; i < max_ncpus; i++) {
2260 		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2261 		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2262 		    ZIO_FLAG_GODFATHER);
2263 	}
2264 
2265 	/*
2266 	 * Parse the configuration into a vdev tree.  We explicitly set the
2267 	 * value that will be returned by spa_version() since parsing the
2268 	 * configuration requires knowing the version number.
2269 	 */
2270 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2271 	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2272 	spa_config_exit(spa, SCL_ALL, FTAG);
2273 
2274 	if (error != 0)
2275 		return (error);
2276 
2277 	ASSERT(spa->spa_root_vdev == rvd);
2278 	ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
2279 	ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
2280 
2281 	if (type != SPA_IMPORT_ASSEMBLE) {
2282 		ASSERT(spa_guid(spa) == pool_guid);
2283 	}
2284 
2285 	/*
2286 	 * Try to open all vdevs, loading each label in the process.
2287 	 */
2288 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2289 	error = vdev_open(rvd);
2290 	spa_config_exit(spa, SCL_ALL, FTAG);
2291 	if (error != 0)
2292 		return (error);
2293 
2294 	/*
2295 	 * We need to validate the vdev labels against the configuration that
2296 	 * we have in hand, which is dependent on the setting of mosconfig. If
2297 	 * mosconfig is true then we're validating the vdev labels based on
2298 	 * that config.  Otherwise, we're validating against the cached config
2299 	 * (zpool.cache) that was read when we loaded the zfs module, and then
2300 	 * later we will recursively call spa_load() and validate against
2301 	 * the vdev config.
2302 	 *
2303 	 * If we're assembling a new pool that's been split off from an
2304 	 * existing pool, the labels haven't yet been updated so we skip
2305 	 * validation for now.
2306 	 */
2307 	if (type != SPA_IMPORT_ASSEMBLE) {
2308 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2309 		error = vdev_validate(rvd, mosconfig);
2310 		spa_config_exit(spa, SCL_ALL, FTAG);
2311 
2312 		if (error != 0)
2313 			return (error);
2314 
2315 		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2316 			return (SET_ERROR(ENXIO));
2317 	}
2318 
2319 	/*
2320 	 * Find the best uberblock.
2321 	 */
2322 	vdev_uberblock_load(rvd, ub, &label);
2323 
2324 	/*
2325 	 * If we weren't able to find a single valid uberblock, return failure.
2326 	 */
2327 	if (ub->ub_txg == 0) {
2328 		nvlist_free(label);
2329 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2330 	}
2331 
2332 	/*
2333 	 * If the pool has an unsupported version we can't open it.
2334 	 */
2335 	if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2336 		nvlist_free(label);
2337 		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2338 	}
2339 
2340 	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2341 		nvlist_t *features;
2342 
2343 		/*
2344 		 * If we weren't able to find what's necessary for reading the
2345 		 * MOS in the label, return failure.
2346 		 */
2347 		if (label == NULL || nvlist_lookup_nvlist(label,
2348 		    ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2349 			nvlist_free(label);
2350 			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2351 			    ENXIO));
2352 		}
2353 
2354 		/*
2355 		 * Update our in-core representation with the definitive values
2356 		 * from the label.
2357 		 */
2358 		nvlist_free(spa->spa_label_features);
2359 		VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2360 	}
2361 
2362 	nvlist_free(label);
2363 
2364 	/*
2365 	 * Look through entries in the label nvlist's features_for_read. If
2366 	 * there is a feature listed there which we don't understand then we
2367 	 * cannot open a pool.
2368 	 */
2369 	if (ub->ub_version >= SPA_VERSION_FEATURES) {
2370 		nvlist_t *unsup_feat;
2371 
2372 		VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2373 		    0);
2374 
2375 		for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2376 		    NULL); nvp != NULL;
2377 		    nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2378 			if (!zfeature_is_supported(nvpair_name(nvp))) {
2379 				VERIFY(nvlist_add_string(unsup_feat,
2380 				    nvpair_name(nvp), "") == 0);
2381 			}
2382 		}
2383 
2384 		if (!nvlist_empty(unsup_feat)) {
2385 			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2386 			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2387 			nvlist_free(unsup_feat);
2388 			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2389 			    ENOTSUP));
2390 		}
2391 
2392 		nvlist_free(unsup_feat);
2393 	}
2394 
2395 	/*
2396 	 * If the vdev guid sum doesn't match the uberblock, we have an
2397 	 * incomplete configuration.  We first check to see if the pool
2398 	 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2399 	 * If it is, defer the vdev_guid_sum check till later so we
2400 	 * can handle missing vdevs.
2401 	 */
2402 	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2403 	    &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2404 	    rvd->vdev_guid_sum != ub->ub_guid_sum)
2405 		return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2406 
2407 	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2408 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2409 		spa_try_repair(spa, config);
2410 		spa_config_exit(spa, SCL_ALL, FTAG);
2411 		nvlist_free(spa->spa_config_splitting);
2412 		spa->spa_config_splitting = NULL;
2413 	}
2414 
2415 	/*
2416 	 * Initialize internal SPA structures.
2417 	 */
2418 	spa->spa_state = POOL_STATE_ACTIVE;
2419 	spa->spa_ubsync = spa->spa_uberblock;
2420 	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2421 	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2422 	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2423 	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2424 	spa->spa_claim_max_txg = spa->spa_first_txg;
2425 	spa->spa_prev_software_version = ub->ub_software_version;
2426 
2427 	error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2428 	if (error)
2429 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2430 	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2431 
2432 	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2433 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2434 
2435 	if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2436 		boolean_t missing_feat_read = B_FALSE;
2437 		nvlist_t *unsup_feat, *enabled_feat;
2438 
2439 		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2440 		    &spa->spa_feat_for_read_obj) != 0) {
2441 			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2442 		}
2443 
2444 		if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2445 		    &spa->spa_feat_for_write_obj) != 0) {
2446 			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2447 		}
2448 
2449 		if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2450 		    &spa->spa_feat_desc_obj) != 0) {
2451 			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2452 		}
2453 
2454 		enabled_feat = fnvlist_alloc();
2455 		unsup_feat = fnvlist_alloc();
2456 
2457 		if (!spa_features_check(spa, B_FALSE,
2458 		    unsup_feat, enabled_feat))
2459 			missing_feat_read = B_TRUE;
2460 
2461 		if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2462 			if (!spa_features_check(spa, B_TRUE,
2463 			    unsup_feat, enabled_feat)) {
2464 				missing_feat_write = B_TRUE;
2465 			}
2466 		}
2467 
2468 		fnvlist_add_nvlist(spa->spa_load_info,
2469 		    ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2470 
2471 		if (!nvlist_empty(unsup_feat)) {
2472 			fnvlist_add_nvlist(spa->spa_load_info,
2473 			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2474 		}
2475 
2476 		fnvlist_free(enabled_feat);
2477 		fnvlist_free(unsup_feat);
2478 
2479 		if (!missing_feat_read) {
2480 			fnvlist_add_boolean(spa->spa_load_info,
2481 			    ZPOOL_CONFIG_CAN_RDONLY);
2482 		}
2483 
2484 		/*
2485 		 * If the state is SPA_LOAD_TRYIMPORT, our objective is
2486 		 * twofold: to determine whether the pool is available for
2487 		 * import in read-write mode and (if it is not) whether the
2488 		 * pool is available for import in read-only mode. If the pool
2489 		 * is available for import in read-write mode, it is displayed
2490 		 * as available in userland; if it is not available for import
2491 		 * in read-only mode, it is displayed as unavailable in
2492 		 * userland. If the pool is available for import in read-only
2493 		 * mode but not read-write mode, it is displayed as unavailable
2494 		 * in userland with a special note that the pool is actually
2495 		 * available for open in read-only mode.
2496 		 *
2497 		 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2498 		 * missing a feature for write, we must first determine whether
2499 		 * the pool can be opened read-only before returning to
2500 		 * userland in order to know whether to display the
2501 		 * abovementioned note.
2502 		 */
2503 		if (missing_feat_read || (missing_feat_write &&
2504 		    spa_writeable(spa))) {
2505 			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2506 			    ENOTSUP));
2507 		}
2508 
2509 		/*
2510 		 * Load refcounts for ZFS features from disk into an in-memory
2511 		 * cache during SPA initialization.
2512 		 */
2513 		for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
2514 			uint64_t refcount;
2515 
2516 			error = feature_get_refcount_from_disk(spa,
2517 			    &spa_feature_table[i], &refcount);
2518 			if (error == 0) {
2519 				spa->spa_feat_refcount_cache[i] = refcount;
2520 			} else if (error == ENOTSUP) {
2521 				spa->spa_feat_refcount_cache[i] =
2522 				    SPA_FEATURE_DISABLED;
2523 			} else {
2524 				return (spa_vdev_err(rvd,
2525 				    VDEV_AUX_CORRUPT_DATA, EIO));
2526 			}
2527 		}
2528 	}
2529 
2530 	if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
2531 		if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
2532 		    &spa->spa_feat_enabled_txg_obj) != 0)
2533 			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2534 	}
2535 
2536 	spa->spa_is_initializing = B_TRUE;
2537 	error = dsl_pool_open(spa->spa_dsl_pool);
2538 	spa->spa_is_initializing = B_FALSE;
2539 	if (error != 0)
2540 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2541 
2542 	if (!mosconfig) {
2543 		uint64_t hostid;
2544 		nvlist_t *policy = NULL, *nvconfig;
2545 
2546 		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2547 			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2548 
2549 		if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2550 		    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2551 			char *hostname;
2552 			unsigned long myhostid = 0;
2553 
2554 			VERIFY(nvlist_lookup_string(nvconfig,
2555 			    ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2556 
2557 #ifdef	_KERNEL
2558 			myhostid = zone_get_hostid(NULL);
2559 #else	/* _KERNEL */
2560 			/*
2561 			 * We're emulating the system's hostid in userland, so
2562 			 * we can't use zone_get_hostid().
2563 			 */
2564 			(void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2565 #endif	/* _KERNEL */
2566 			if (hostid != 0 && myhostid != 0 &&
2567 			    hostid != myhostid) {
2568 				nvlist_free(nvconfig);
2569 				cmn_err(CE_WARN, "pool '%s' could not be "
2570 				    "loaded as it was last accessed by "
2571 				    "another system (host: %s hostid: 0x%lx). "
2572 				    "See: http://illumos.org/msg/ZFS-8000-EY",
2573 				    spa_name(spa), hostname,
2574 				    (unsigned long)hostid);
2575 				return (SET_ERROR(EBADF));
2576 			}
2577 		}
2578 		if (nvlist_lookup_nvlist(spa->spa_config,
2579 		    ZPOOL_REWIND_POLICY, &policy) == 0)
2580 			VERIFY(nvlist_add_nvlist(nvconfig,
2581 			    ZPOOL_REWIND_POLICY, policy) == 0);
2582 
2583 		spa_config_set(spa, nvconfig);
2584 		spa_unload(spa);
2585 		spa_deactivate(spa);
2586 		spa_activate(spa, orig_mode);
2587 
2588 		return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2589 	}
2590 
2591 	/* Grab the secret checksum salt from the MOS. */
2592 	error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2593 	    DMU_POOL_CHECKSUM_SALT, 1,
2594 	    sizeof (spa->spa_cksum_salt.zcs_bytes),
2595 	    spa->spa_cksum_salt.zcs_bytes);
2596 	if (error == ENOENT) {
2597 		/* Generate a new salt for subsequent use */
2598 		(void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
2599 		    sizeof (spa->spa_cksum_salt.zcs_bytes));
2600 	} else if (error != 0) {
2601 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2602 	}
2603 
2604 	if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2605 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2606 	error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2607 	if (error != 0)
2608 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2609 
2610 	/*
2611 	 * Load the bit that tells us to use the new accounting function
2612 	 * (raid-z deflation).  If we have an older pool, this will not
2613 	 * be present.
2614 	 */
2615 	error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2616 	if (error != 0 && error != ENOENT)
2617 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2618 
2619 	error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2620 	    &spa->spa_creation_version);
2621 	if (error != 0 && error != ENOENT)
2622 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2623 
2624 	/*
2625 	 * Load the persistent error log.  If we have an older pool, this will
2626 	 * not be present.
2627 	 */
2628 	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2629 	if (error != 0 && error != ENOENT)
2630 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2631 
2632 	error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2633 	    &spa->spa_errlog_scrub);
2634 	if (error != 0 && error != ENOENT)
2635 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2636 
2637 	/*
2638 	 * Load the history object.  If we have an older pool, this
2639 	 * will not be present.
2640 	 */
2641 	error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2642 	if (error != 0 && error != ENOENT)
2643 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2644 
2645 	/*
2646 	 * Load the per-vdev ZAP map. If we have an older pool, this will not
2647 	 * be present; in this case, defer its creation to a later time to
2648 	 * avoid dirtying the MOS this early / out of sync context. See
2649 	 * spa_sync_config_object.
2650 	 */
2651 
2652 	/* The sentinel is only available in the MOS config. */
2653 	nvlist_t *mos_config;
2654 	if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0)
2655 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2656 
2657 	error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP,
2658 	    &spa->spa_all_vdev_zaps);
2659 
2660 	if (error == ENOENT) {
2661 		VERIFY(!nvlist_exists(mos_config,
2662 		    ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
2663 		spa->spa_avz_action = AVZ_ACTION_INITIALIZE;
2664 		ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
2665 	} else if (error != 0) {
2666 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2667 	} else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) {
2668 		/*
2669 		 * An older version of ZFS overwrote the sentinel value, so
2670 		 * we have orphaned per-vdev ZAPs in the MOS. Defer their
2671 		 * destruction to later; see spa_sync_config_object.
2672 		 */
2673 		spa->spa_avz_action = AVZ_ACTION_DESTROY;
2674 		/*
2675 		 * We're assuming that no vdevs have had their ZAPs created
2676 		 * before this. Better be sure of it.
2677 		 */
2678 		ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
2679 	}
2680 	nvlist_free(mos_config);
2681 
2682 	/*
2683 	 * If we're assembling the pool from the split-off vdevs of
2684 	 * an existing pool, we don't want to attach the spares & cache
2685 	 * devices.
2686 	 */
2687 
2688 	/*
2689 	 * Load any hot spares for this pool.
2690 	 */
2691 	error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2692 	if (error != 0 && error != ENOENT)
2693 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2694 	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2695 		ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2696 		if (load_nvlist(spa, spa->spa_spares.sav_object,
2697 		    &spa->spa_spares.sav_config) != 0)
2698 			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2699 
2700 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2701 		spa_load_spares(spa);
2702 		spa_config_exit(spa, SCL_ALL, FTAG);
2703 	} else if (error == 0) {
2704 		spa->spa_spares.sav_sync = B_TRUE;
2705 	}
2706 
2707 	/*
2708 	 * Load any level 2 ARC devices for this pool.
2709 	 */
2710 	error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2711 	    &spa->spa_l2cache.sav_object);
2712 	if (error != 0 && error != ENOENT)
2713 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2714 	if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2715 		ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2716 		if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2717 		    &spa->spa_l2cache.sav_config) != 0)
2718 			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2719 
2720 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2721 		spa_load_l2cache(spa);
2722 		spa_config_exit(spa, SCL_ALL, FTAG);
2723 	} else if (error == 0) {
2724 		spa->spa_l2cache.sav_sync = B_TRUE;
2725 	}
2726 
2727 	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2728 
2729 	error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2730 	if (error && error != ENOENT)
2731 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2732 
2733 	if (error == 0) {
2734 		uint64_t autoreplace;
2735 
2736 		spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2737 		spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2738 		spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2739 		spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2740 		spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2741 		spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2742 		    &spa->spa_dedup_ditto);
2743 
2744 		spa->spa_autoreplace = (autoreplace != 0);
2745 	}
2746 
2747 	/*
2748 	 * If the 'autoreplace' property is set, then post a resource notifying
2749 	 * the ZFS DE that it should not issue any faults for unopenable
2750 	 * devices.  We also iterate over the vdevs, and post a sysevent for any
2751 	 * unopenable vdevs so that the normal autoreplace handler can take
2752 	 * over.
2753 	 */
2754 	if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2755 		spa_check_removed(spa->spa_root_vdev);
2756 		/*
2757 		 * For the import case, this is done in spa_import(), because
2758 		 * at this point we're using the spare definitions from
2759 		 * the MOS config, not necessarily from the userland config.
2760 		 */
2761 		if (state != SPA_LOAD_IMPORT) {
2762 			spa_aux_check_removed(&spa->spa_spares);
2763 			spa_aux_check_removed(&spa->spa_l2cache);
2764 		}
2765 	}
2766 
2767 	/*
2768 	 * Load the vdev state for all toplevel vdevs.
2769 	 */
2770 	vdev_load(rvd);
2771 
2772 	/*
2773 	 * Propagate the leaf DTLs we just loaded all the way up the tree.
2774 	 */
2775 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2776 	vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2777 	spa_config_exit(spa, SCL_ALL, FTAG);
2778 
2779 	/*
2780 	 * Load the DDTs (dedup tables).
2781 	 */
2782 	error = ddt_load(spa);
2783 	if (error != 0)
2784 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2785 
2786 	spa_update_dspace(spa);
2787 
2788 	/*
2789 	 * Validate the config, using the MOS config to fill in any
2790 	 * information which might be missing.  If we fail to validate
2791 	 * the config then declare the pool unfit for use. If we're
2792 	 * assembling a pool from a split, the log is not transferred
2793 	 * over.
2794 	 */
2795 	if (type != SPA_IMPORT_ASSEMBLE) {
2796 		nvlist_t *nvconfig;
2797 
2798 		if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2799 			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2800 
2801 		if (!spa_config_valid(spa, nvconfig)) {
2802 			nvlist_free(nvconfig);
2803 			return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2804 			    ENXIO));
2805 		}
2806 		nvlist_free(nvconfig);
2807 
2808 		/*
2809 		 * Now that we've validated the config, check the state of the
2810 		 * root vdev.  If it can't be opened, it indicates one or
2811 		 * more toplevel vdevs are faulted.
2812 		 */
2813 		if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2814 			return (SET_ERROR(ENXIO));
2815 
2816 		if (spa_writeable(spa) && spa_check_logs(spa)) {
2817 			*ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2818 			return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2819 		}
2820 	}
2821 
2822 	if (missing_feat_write) {
2823 		ASSERT(state == SPA_LOAD_TRYIMPORT);
2824 
2825 		/*
2826 		 * At this point, we know that we can open the pool in
2827 		 * read-only mode but not read-write mode. We now have enough
2828 		 * information and can return to userland.
2829 		 */
2830 		return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2831 	}
2832 
2833 	/*
2834 	 * We've successfully opened the pool, verify that we're ready
2835 	 * to start pushing transactions.
2836 	 */
2837 	if (state != SPA_LOAD_TRYIMPORT) {
2838 		if (error = spa_load_verify(spa))
2839 			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2840 			    error));
2841 	}
2842 
2843 	if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2844 	    spa->spa_load_max_txg == UINT64_MAX)) {
2845 		dmu_tx_t *tx;
2846 		int need_update = B_FALSE;
2847 		dsl_pool_t *dp = spa_get_dsl(spa);
2848 
2849 		ASSERT(state != SPA_LOAD_TRYIMPORT);
2850 
2851 		/*
2852 		 * Claim log blocks that haven't been committed yet.
2853 		 * This must all happen in a single txg.
2854 		 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2855 		 * invoked from zil_claim_log_block()'s i/o done callback.
2856 		 * Price of rollback is that we abandon the log.
2857 		 */
2858 		spa->spa_claiming = B_TRUE;
2859 
2860 		tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
2861 		(void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
2862 		    zil_claim, tx, DS_FIND_CHILDREN);
2863 		dmu_tx_commit(tx);
2864 
2865 		spa->spa_claiming = B_FALSE;
2866 
2867 		spa_set_log_state(spa, SPA_LOG_GOOD);
2868 		spa->spa_sync_on = B_TRUE;
2869 		txg_sync_start(spa->spa_dsl_pool);
2870 
2871 		/*
2872 		 * Wait for all claims to sync.  We sync up to the highest
2873 		 * claimed log block birth time so that claimed log blocks
2874 		 * don't appear to be from the future.  spa_claim_max_txg
2875 		 * will have been set for us by either zil_check_log_chain()
2876 		 * (invoked from spa_check_logs()) or zil_claim() above.
2877 		 */
2878 		txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2879 
2880 		/*
2881 		 * If the config cache is stale, or we have uninitialized
2882 		 * metaslabs (see spa_vdev_add()), then update the config.
2883 		 *
2884 		 * If this is a verbatim import, trust the current
2885 		 * in-core spa_config and update the disk labels.
2886 		 */
2887 		if (config_cache_txg != spa->spa_config_txg ||
2888 		    state == SPA_LOAD_IMPORT ||
2889 		    state == SPA_LOAD_RECOVER ||
2890 		    (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2891 			need_update = B_TRUE;
2892 
2893 		for (int c = 0; c < rvd->vdev_children; c++)
2894 			if (rvd->vdev_child[c]->vdev_ms_array == 0)
2895 				need_update = B_TRUE;
2896 
2897 		/*
2898 		 * Update the config cache asychronously in case we're the
2899 		 * root pool, in which case the config cache isn't writable yet.
2900 		 */
2901 		if (need_update)
2902 			spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2903 
2904 		/*
2905 		 * Check all DTLs to see if anything needs resilvering.
2906 		 */
2907 		if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2908 		    vdev_resilver_needed(rvd, NULL, NULL))
2909 			spa_async_request(spa, SPA_ASYNC_RESILVER);
2910 
2911 		/*
2912 		 * Log the fact that we booted up (so that we can detect if
2913 		 * we rebooted in the middle of an operation).
2914 		 */
2915 		spa_history_log_version(spa, "open");
2916 
2917 		/*
2918 		 * Delete any inconsistent datasets.
2919 		 */
2920 		(void) dmu_objset_find(spa_name(spa),
2921 		    dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2922 
2923 		/*
2924 		 * Clean up any stale temporary dataset userrefs.
2925 		 */
2926 		dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2927 	}
2928 
2929 	return (0);
2930 }
2931 
2932 static int
2933 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2934 {
2935 	int mode = spa->spa_mode;
2936 
2937 	spa_unload(spa);
2938 	spa_deactivate(spa);
2939 
2940 	spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
2941 
2942 	spa_activate(spa, mode);
2943 	spa_async_suspend(spa);
2944 
2945 	return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2946 }
2947 
2948 /*
2949  * If spa_load() fails this function will try loading prior txg's. If
2950  * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2951  * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2952  * function will not rewind the pool and will return the same error as
2953  * spa_load().
2954  */
2955 static int
2956 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2957     uint64_t max_request, int rewind_flags)
2958 {
2959 	nvlist_t *loadinfo = NULL;
2960 	nvlist_t *config = NULL;
2961 	int load_error, rewind_error;
2962 	uint64_t safe_rewind_txg;
2963 	uint64_t min_txg;
2964 
2965 	if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2966 		spa->spa_load_max_txg = spa->spa_load_txg;
2967 		spa_set_log_state(spa, SPA_LOG_CLEAR);
2968 	} else {
2969 		spa->spa_load_max_txg = max_request;
2970 		if (max_request != UINT64_MAX)
2971 			spa->spa_extreme_rewind = B_TRUE;
2972 	}
2973 
2974 	load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2975 	    mosconfig);
2976 	if (load_error == 0)
2977 		return (0);
2978 
2979 	if (spa->spa_root_vdev != NULL)
2980 		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2981 
2982 	spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2983 	spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2984 
2985 	if (rewind_flags & ZPOOL_NEVER_REWIND) {
2986 		nvlist_free(config);
2987 		return (load_error);
2988 	}
2989 
2990 	if (state == SPA_LOAD_RECOVER) {
2991 		/* Price of rolling back is discarding txgs, including log */
2992 		spa_set_log_state(spa, SPA_LOG_CLEAR);
2993 	} else {
2994 		/*
2995 		 * If we aren't rolling back save the load info from our first
2996 		 * import attempt so that we can restore it after attempting
2997 		 * to rewind.
2998 		 */
2999 		loadinfo = spa->spa_load_info;
3000 		spa->spa_load_info = fnvlist_alloc();
3001 	}
3002 
3003 	spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
3004 	safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
3005 	min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
3006 	    TXG_INITIAL : safe_rewind_txg;
3007 
3008 	/*
3009 	 * Continue as long as we're finding errors, we're still within
3010 	 * the acceptable rewind range, and we're still finding uberblocks
3011 	 */
3012 	while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
3013 	    spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
3014 		if (spa->spa_load_max_txg < safe_rewind_txg)
3015 			spa->spa_extreme_rewind = B_TRUE;
3016 		rewind_error = spa_load_retry(spa, state, mosconfig);
3017 	}
3018 
3019 	spa->spa_extreme_rewind = B_FALSE;
3020 	spa->spa_load_max_txg = UINT64_MAX;
3021 
3022 	if (config && (rewind_error || state != SPA_LOAD_RECOVER))
3023 		spa_config_set(spa, config);
3024 
3025 	if (state == SPA_LOAD_RECOVER) {
3026 		ASSERT3P(loadinfo, ==, NULL);
3027 		return (rewind_error);
3028 	} else {
3029 		/* Store the rewind info as part of the initial load info */
3030 		fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
3031 		    spa->spa_load_info);
3032 
3033 		/* Restore the initial load info */
3034 		fnvlist_free(spa->spa_load_info);
3035 		spa->spa_load_info = loadinfo;
3036 
3037 		return (load_error);
3038 	}
3039 }
3040 
3041 /*
3042  * Pool Open/Import
3043  *
3044  * The import case is identical to an open except that the configuration is sent
3045  * down from userland, instead of grabbed from the configuration cache.  For the
3046  * case of an open, the pool configuration will exist in the
3047  * POOL_STATE_UNINITIALIZED state.
3048  *
3049  * The stats information (gen/count/ustats) is used to gather vdev statistics at
3050  * the same time open the pool, without having to keep around the spa_t in some
3051  * ambiguous state.
3052  */
3053 static int
3054 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
3055     nvlist_t **config)
3056 {
3057 	spa_t *spa;
3058 	spa_load_state_t state = SPA_LOAD_OPEN;
3059 	int error;
3060 	int locked = B_FALSE;
3061 
3062 	*spapp = NULL;
3063 
3064 	/*
3065 	 * As disgusting as this is, we need to support recursive calls to this
3066 	 * function because dsl_dir_open() is called during spa_load(), and ends
3067 	 * up calling spa_open() again.  The real fix is to figure out how to
3068 	 * avoid dsl_dir_open() calling this in the first place.
3069 	 */
3070 	if (mutex_owner(&spa_namespace_lock) != curthread) {
3071 		mutex_enter(&spa_namespace_lock);
3072 		locked = B_TRUE;
3073 	}
3074 
3075 	if ((spa = spa_lookup(pool)) == NULL) {
3076 		if (locked)
3077 			mutex_exit(&spa_namespace_lock);
3078 		return (SET_ERROR(ENOENT));
3079 	}
3080 
3081 	if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
3082 		zpool_rewind_policy_t policy;
3083 
3084 		zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
3085 		    &policy);
3086 		if (policy.zrp_request & ZPOOL_DO_REWIND)
3087 			state = SPA_LOAD_RECOVER;
3088 
3089 		spa_activate(spa, spa_mode_global);
3090 
3091 		if (state != SPA_LOAD_RECOVER)
3092 			spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3093 
3094 		error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
3095 		    policy.zrp_request);
3096 
3097 		if (error == EBADF) {
3098 			/*
3099 			 * If vdev_validate() returns failure (indicated by
3100 			 * EBADF), it indicates that one of the vdevs indicates
3101 			 * that the pool has been exported or destroyed.  If
3102 			 * this is the case, the config cache is out of sync and
3103 			 * we should remove the pool from the namespace.
3104 			 */
3105 			spa_unload(spa);
3106 			spa_deactivate(spa);
3107 			spa_config_sync(spa, B_TRUE, B_TRUE);
3108 			spa_remove(spa);
3109 			if (locked)
3110 				mutex_exit(&spa_namespace_lock);
3111 			return (SET_ERROR(ENOENT));
3112 		}
3113 
3114 		if (error) {
3115 			/*
3116 			 * We can't open the pool, but we still have useful
3117 			 * information: the state of each vdev after the
3118 			 * attempted vdev_open().  Return this to the user.
3119 			 */
3120 			if (config != NULL && spa->spa_config) {
3121 				VERIFY(nvlist_dup(spa->spa_config, config,
3122 				    KM_SLEEP) == 0);
3123 				VERIFY(nvlist_add_nvlist(*config,
3124 				    ZPOOL_CONFIG_LOAD_INFO,
3125 				    spa->spa_load_info) == 0);
3126 			}
3127 			spa_unload(spa);
3128 			spa_deactivate(spa);
3129 			spa->spa_last_open_failed = error;
3130 			if (locked)
3131 				mutex_exit(&spa_namespace_lock);
3132 			*spapp = NULL;
3133 			return (error);
3134 		}
3135 	}
3136 
3137 	spa_open_ref(spa, tag);
3138 
3139 	if (config != NULL)
3140 		*config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3141 
3142 	/*
3143 	 * If we've recovered the pool, pass back any information we
3144 	 * gathered while doing the load.
3145 	 */
3146 	if (state == SPA_LOAD_RECOVER) {
3147 		VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
3148 		    spa->spa_load_info) == 0);
3149 	}
3150 
3151 	if (locked) {
3152 		spa->spa_last_open_failed = 0;
3153 		spa->spa_last_ubsync_txg = 0;
3154 		spa->spa_load_txg = 0;
3155 		mutex_exit(&spa_namespace_lock);
3156 	}
3157 
3158 	*spapp = spa;
3159 
3160 	return (0);
3161 }
3162 
3163 int
3164 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
3165     nvlist_t **config)
3166 {
3167 	return (spa_open_common(name, spapp, tag, policy, config));
3168 }
3169 
3170 int
3171 spa_open(const char *name, spa_t **spapp, void *tag)
3172 {
3173 	return (spa_open_common(name, spapp, tag, NULL, NULL));
3174 }
3175 
3176 /*
3177  * Lookup the given spa_t, incrementing the inject count in the process,
3178  * preventing it from being exported or destroyed.
3179  */
3180 spa_t *
3181 spa_inject_addref(char *name)
3182 {
3183 	spa_t *spa;
3184 
3185 	mutex_enter(&spa_namespace_lock);
3186 	if ((spa = spa_lookup(name)) == NULL) {
3187 		mutex_exit(&spa_namespace_lock);
3188 		return (NULL);
3189 	}
3190 	spa->spa_inject_ref++;
3191 	mutex_exit(&spa_namespace_lock);
3192 
3193 	return (spa);
3194 }
3195 
3196 void
3197 spa_inject_delref(spa_t *spa)
3198 {
3199 	mutex_enter(&spa_namespace_lock);
3200 	spa->spa_inject_ref--;
3201 	mutex_exit(&spa_namespace_lock);
3202 }
3203 
3204 /*
3205  * Add spares device information to the nvlist.
3206  */
3207 static void
3208 spa_add_spares(spa_t *spa, nvlist_t *config)
3209 {
3210 	nvlist_t **spares;
3211 	uint_t i, nspares;
3212 	nvlist_t *nvroot;
3213 	uint64_t guid;
3214 	vdev_stat_t *vs;
3215 	uint_t vsc;
3216 	uint64_t pool;
3217 
3218 	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3219 
3220 	if (spa->spa_spares.sav_count == 0)
3221 		return;
3222 
3223 	VERIFY(nvlist_lookup_nvlist(config,
3224 	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3225 	VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
3226 	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3227 	if (nspares != 0) {
3228 		VERIFY(nvlist_add_nvlist_array(nvroot,
3229 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3230 		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3231 		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
3232 
3233 		/*
3234 		 * Go through and find any spares which have since been
3235 		 * repurposed as an active spare.  If this is the case, update
3236 		 * their status appropriately.
3237 		 */
3238 		for (i = 0; i < nspares; i++) {
3239 			VERIFY(nvlist_lookup_uint64(spares[i],
3240 			    ZPOOL_CONFIG_GUID, &guid) == 0);
3241 			if (spa_spare_exists(guid, &pool, NULL) &&
3242 			    pool != 0ULL) {
3243 				VERIFY(nvlist_lookup_uint64_array(
3244 				    spares[i], ZPOOL_CONFIG_VDEV_STATS,
3245 				    (uint64_t **)&vs, &vsc) == 0);
3246 				vs->vs_state = VDEV_STATE_CANT_OPEN;
3247 				vs->vs_aux = VDEV_AUX_SPARED;
3248 			}
3249 		}
3250 	}
3251 }
3252 
3253 /*
3254  * Add l2cache device information to the nvlist, including vdev stats.
3255  */
3256 static void
3257 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3258 {
3259 	nvlist_t **l2cache;
3260 	uint_t i, j, nl2cache;
3261 	nvlist_t *nvroot;
3262 	uint64_t guid;
3263 	vdev_t *vd;
3264 	vdev_stat_t *vs;
3265 	uint_t vsc;
3266 
3267 	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3268 
3269 	if (spa->spa_l2cache.sav_count == 0)
3270 		return;
3271 
3272 	VERIFY(nvlist_lookup_nvlist(config,
3273 	    ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3274 	VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3275 	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3276 	if (nl2cache != 0) {
3277 		VERIFY(nvlist_add_nvlist_array(nvroot,
3278 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3279 		VERIFY(nvlist_lookup_nvlist_array(nvroot,
3280 		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3281 
3282 		/*
3283 		 * Update level 2 cache device stats.
3284 		 */
3285 
3286 		for (i = 0; i < nl2cache; i++) {
3287 			VERIFY(nvlist_lookup_uint64(l2cache[i],
3288 			    ZPOOL_CONFIG_GUID, &guid) == 0);
3289 
3290 			vd = NULL;
3291 			for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3292 				if (guid ==
3293 				    spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3294 					vd = spa->spa_l2cache.sav_vdevs[j];
3295 					break;
3296 				}
3297 			}
3298 			ASSERT(vd != NULL);
3299 
3300 			VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3301 			    ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3302 			    == 0);
3303 			vdev_get_stats(vd, vs);
3304 		}
3305 	}
3306 }
3307 
3308 static void
3309 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3310 {
3311 	nvlist_t *features;
3312 	zap_cursor_t zc;
3313 	zap_attribute_t za;
3314 
3315 	ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3316 	VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3317 
3318 	if (spa->spa_feat_for_read_obj != 0) {
3319 		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3320 		    spa->spa_feat_for_read_obj);
3321 		    zap_cursor_retrieve(&zc, &za) == 0;
3322 		    zap_cursor_advance(&zc)) {
3323 			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3324 			    za.za_num_integers == 1);
3325 			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3326 			    za.za_first_integer));
3327 		}
3328 		zap_cursor_fini(&zc);
3329 	}
3330 
3331 	if (spa->spa_feat_for_write_obj != 0) {
3332 		for (zap_cursor_init(&zc, spa->spa_meta_objset,
3333 		    spa->spa_feat_for_write_obj);
3334 		    zap_cursor_retrieve(&zc, &za) == 0;
3335 		    zap_cursor_advance(&zc)) {
3336 			ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3337 			    za.za_num_integers == 1);
3338 			VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3339 			    za.za_first_integer));
3340 		}
3341 		zap_cursor_fini(&zc);
3342 	}
3343 
3344 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3345 	    features) == 0);
3346 	nvlist_free(features);
3347 }
3348 
3349 int
3350 spa_get_stats(const char *name, nvlist_t **config,
3351     char *altroot, size_t buflen)
3352 {
3353 	int error;
3354 	spa_t *spa;
3355 
3356 	*config = NULL;
3357 	error = spa_open_common(name, &spa, FTAG, NULL, config);
3358 
3359 	if (spa != NULL) {
3360 		/*
3361 		 * This still leaves a window of inconsistency where the spares
3362 		 * or l2cache devices could change and the config would be
3363 		 * self-inconsistent.
3364 		 */
3365 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3366 
3367 		if (*config != NULL) {
3368 			uint64_t loadtimes[2];
3369 
3370 			loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3371 			loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3372 			VERIFY(nvlist_add_uint64_array(*config,
3373 			    ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3374 
3375 			VERIFY(nvlist_add_uint64(*config,
3376 			    ZPOOL_CONFIG_ERRCOUNT,
3377 			    spa_get_errlog_size(spa)) == 0);
3378 
3379 			if (spa_suspended(spa))
3380 				VERIFY(nvlist_add_uint64(*config,
3381 				    ZPOOL_CONFIG_SUSPENDED,
3382 				    spa->spa_failmode) == 0);
3383 
3384 			spa_add_spares(spa, *config);
3385 			spa_add_l2cache(spa, *config);
3386 			spa_add_feature_stats(spa, *config);
3387 		}
3388 	}
3389 
3390 	/*
3391 	 * We want to get the alternate root even for faulted pools, so we cheat
3392 	 * and call spa_lookup() directly.
3393 	 */
3394 	if (altroot) {
3395 		if (spa == NULL) {
3396 			mutex_enter(&spa_namespace_lock);
3397 			spa = spa_lookup(name);
3398 			if (spa)
3399 				spa_altroot(spa, altroot, buflen);
3400 			else
3401 				altroot[0] = '\0';
3402 			spa = NULL;
3403 			mutex_exit(&spa_namespace_lock);
3404 		} else {
3405 			spa_altroot(spa, altroot, buflen);
3406 		}
3407 	}
3408 
3409 	if (spa != NULL) {
3410 		spa_config_exit(spa, SCL_CONFIG, FTAG);
3411 		spa_close(spa, FTAG);
3412 	}
3413 
3414 	return (error);
3415 }
3416 
3417 /*
3418  * Validate that the auxiliary device array is well formed.  We must have an
3419  * array of nvlists, each which describes a valid leaf vdev.  If this is an
3420  * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3421  * specified, as long as they are well-formed.
3422  */
3423 static int
3424 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3425     spa_aux_vdev_t *sav, const char *config, uint64_t version,
3426     vdev_labeltype_t label)
3427 {
3428 	nvlist_t **dev;
3429 	uint_t i, ndev;
3430 	vdev_t *vd;
3431 	int error;
3432 
3433 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3434 
3435 	/*
3436 	 * It's acceptable to have no devs specified.
3437 	 */
3438 	if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3439 		return (0);
3440 
3441 	if (ndev == 0)
3442 		return (SET_ERROR(EINVAL));
3443 
3444 	/*
3445 	 * Make sure the pool is formatted with a version that supports this
3446 	 * device type.
3447 	 */
3448 	if (spa_version(spa) < version)
3449 		return (SET_ERROR(ENOTSUP));
3450 
3451 	/*
3452 	 * Set the pending device list so we correctly handle device in-use
3453 	 * checking.
3454 	 */
3455 	sav->sav_pending = dev;
3456 	sav->sav_npending = ndev;
3457 
3458 	for (i = 0; i < ndev; i++) {
3459 		if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3460 		    mode)) != 0)
3461 			goto out;
3462 
3463 		if (!vd->vdev_ops->vdev_op_leaf) {
3464 			vdev_free(vd);
3465 			error = SET_ERROR(EINVAL);
3466 			goto out;
3467 		}
3468 
3469 		/*
3470 		 * The L2ARC currently only supports disk devices in
3471 		 * kernel context.  For user-level testing, we allow it.
3472 		 */
3473 #ifdef _KERNEL
3474 		if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3475 		    strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3476 			error = SET_ERROR(ENOTBLK);
3477 			vdev_free(vd);
3478 			goto out;
3479 		}
3480 #endif
3481 		vd->vdev_top = vd;
3482 
3483 		if ((error = vdev_open(vd)) == 0 &&
3484 		    (error = vdev_label_init(vd, crtxg, label)) == 0) {
3485 			VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3486 			    vd->vdev_guid) == 0);
3487 		}
3488 
3489 		vdev_free(vd);
3490 
3491 		if (error &&
3492 		    (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3493 			goto out;
3494 		else
3495 			error = 0;
3496 	}
3497 
3498 out:
3499 	sav->sav_pending = NULL;
3500 	sav->sav_npending = 0;
3501 	return (error);
3502 }
3503 
3504 static int
3505 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3506 {
3507 	int error;
3508 
3509 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3510 
3511 	if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3512 	    &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3513 	    VDEV_LABEL_SPARE)) != 0) {
3514 		return (error);
3515 	}
3516 
3517 	return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3518 	    &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3519 	    VDEV_LABEL_L2CACHE));
3520 }
3521 
3522 static void
3523 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3524     const char *config)
3525 {
3526 	int i;
3527 
3528 	if (sav->sav_config != NULL) {
3529 		nvlist_t **olddevs;
3530 		uint_t oldndevs;
3531 		nvlist_t **newdevs;
3532 
3533 		/*
3534 		 * Generate new dev list by concatentating with the
3535 		 * current dev list.
3536 		 */
3537 		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3538 		    &olddevs, &oldndevs) == 0);
3539 
3540 		newdevs = kmem_alloc(sizeof (void *) *
3541 		    (ndevs + oldndevs), KM_SLEEP);
3542 		for (i = 0; i < oldndevs; i++)
3543 			VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3544 			    KM_SLEEP) == 0);
3545 		for (i = 0; i < ndevs; i++)
3546 			VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3547 			    KM_SLEEP) == 0);
3548 
3549 		VERIFY(nvlist_remove(sav->sav_config, config,
3550 		    DATA_TYPE_NVLIST_ARRAY) == 0);
3551 
3552 		VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3553 		    config, newdevs, ndevs + oldndevs) == 0);
3554 		for (i = 0; i < oldndevs + ndevs; i++)
3555 			nvlist_free(newdevs[i]);
3556 		kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3557 	} else {
3558 		/*
3559 		 * Generate a new dev list.
3560 		 */
3561 		VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3562 		    KM_SLEEP) == 0);
3563 		VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3564 		    devs, ndevs) == 0);
3565 	}
3566 }
3567 
3568 /*
3569  * Stop and drop level 2 ARC devices
3570  */
3571 void
3572 spa_l2cache_drop(spa_t *spa)
3573 {
3574 	vdev_t *vd;
3575 	int i;
3576 	spa_aux_vdev_t *sav = &spa->spa_l2cache;
3577 
3578 	for (i = 0; i < sav->sav_count; i++) {
3579 		uint64_t pool;
3580 
3581 		vd = sav->sav_vdevs[i];
3582 		ASSERT(vd != NULL);
3583 
3584 		if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3585 		    pool != 0ULL && l2arc_vdev_present(vd))
3586 			l2arc_remove_vdev(vd);
3587 	}
3588 }
3589 
3590 /*
3591  * Pool Creation
3592  */
3593 int
3594 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3595     nvlist_t *zplprops)
3596 {
3597 	spa_t *spa;
3598 	char *altroot = NULL;
3599 	vdev_t *rvd;
3600 	dsl_pool_t *dp;
3601 	dmu_tx_t *tx;
3602 	int error = 0;
3603 	uint64_t txg = TXG_INITIAL;
3604 	nvlist_t **spares, **l2cache;
3605 	uint_t nspares, nl2cache;
3606 	uint64_t version, obj;
3607 	boolean_t has_features;
3608 
3609 	/*
3610 	 * If this pool already exists, return failure.
3611 	 */
3612 	mutex_enter(&spa_namespace_lock);
3613 	if (spa_lookup(pool) != NULL) {
3614 		mutex_exit(&spa_namespace_lock);
3615 		return (SET_ERROR(EEXIST));
3616 	}
3617 
3618 	/*
3619 	 * Allocate a new spa_t structure.
3620 	 */
3621 	(void) nvlist_lookup_string(props,
3622 	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3623 	spa = spa_add(pool, NULL, altroot);
3624 	spa_activate(spa, spa_mode_global);
3625 
3626 	if (props && (error = spa_prop_validate(spa, props))) {
3627 		spa_deactivate(spa);
3628 		spa_remove(spa);
3629 		mutex_exit(&spa_namespace_lock);
3630 		return (error);
3631 	}
3632 
3633 	has_features = B_FALSE;
3634 	for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3635 	    elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3636 		if (zpool_prop_feature(nvpair_name(elem)))
3637 			has_features = B_TRUE;
3638 	}
3639 
3640 	if (has_features || nvlist_lookup_uint64(props,
3641 	    zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3642 		version = SPA_VERSION;
3643 	}
3644 	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3645 
3646 	spa->spa_first_txg = txg;
3647 	spa->spa_uberblock.ub_txg = txg - 1;
3648 	spa->spa_uberblock.ub_version = version;
3649 	spa->spa_ubsync = spa->spa_uberblock;
3650 	spa->spa_load_state = SPA_LOAD_CREATE;
3651 
3652 	/*
3653 	 * Create "The Godfather" zio to hold all async IOs
3654 	 */
3655 	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
3656 	    KM_SLEEP);
3657 	for (int i = 0; i < max_ncpus; i++) {
3658 		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
3659 		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
3660 		    ZIO_FLAG_GODFATHER);
3661 	}
3662 
3663 	/*
3664 	 * Create the root vdev.
3665 	 */
3666 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3667 
3668 	error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3669 
3670 	ASSERT(error != 0 || rvd != NULL);
3671 	ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3672 
3673 	if (error == 0 && !zfs_allocatable_devs(nvroot))
3674 		error = SET_ERROR(EINVAL);
3675 
3676 	if (error == 0 &&
3677 	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3678 	    (error = spa_validate_aux(spa, nvroot, txg,
3679 	    VDEV_ALLOC_ADD)) == 0) {
3680 		for (int c = 0; c < rvd->vdev_children; c++) {
3681 			vdev_metaslab_set_size(rvd->vdev_child[c]);
3682 			vdev_expand(rvd->vdev_child[c], txg);
3683 		}
3684 	}
3685 
3686 	spa_config_exit(spa, SCL_ALL, FTAG);
3687 
3688 	if (error != 0) {
3689 		spa_unload(spa);
3690 		spa_deactivate(spa);
3691 		spa_remove(spa);
3692 		mutex_exit(&spa_namespace_lock);
3693 		return (error);
3694 	}
3695 
3696 	/*
3697 	 * Get the list of spares, if specified.
3698 	 */
3699 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3700 	    &spares, &nspares) == 0) {
3701 		VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3702 		    KM_SLEEP) == 0);
3703 		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3704 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3705 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3706 		spa_load_spares(spa);
3707 		spa_config_exit(spa, SCL_ALL, FTAG);
3708 		spa->spa_spares.sav_sync = B_TRUE;
3709 	}
3710 
3711 	/*
3712 	 * Get the list of level 2 cache devices, if specified.
3713 	 */
3714 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3715 	    &l2cache, &nl2cache) == 0) {
3716 		VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3717 		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
3718 		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3719 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3720 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3721 		spa_load_l2cache(spa);
3722 		spa_config_exit(spa, SCL_ALL, FTAG);
3723 		spa->spa_l2cache.sav_sync = B_TRUE;
3724 	}
3725 
3726 	spa->spa_is_initializing = B_TRUE;
3727 	spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3728 	spa->spa_meta_objset = dp->dp_meta_objset;
3729 	spa->spa_is_initializing = B_FALSE;
3730 
3731 	/*
3732 	 * Create DDTs (dedup tables).
3733 	 */
3734 	ddt_create(spa);
3735 
3736 	spa_update_dspace(spa);
3737 
3738 	tx = dmu_tx_create_assigned(dp, txg);
3739 
3740 	/*
3741 	 * Create the pool config object.
3742 	 */
3743 	spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3744 	    DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3745 	    DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3746 
3747 	if (zap_add(spa->spa_meta_objset,
3748 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3749 	    sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3750 		cmn_err(CE_PANIC, "failed to add pool config");
3751 	}
3752 
3753 	if (spa_version(spa) >= SPA_VERSION_FEATURES)
3754 		spa_feature_create_zap_objects(spa, tx);
3755 
3756 	if (zap_add(spa->spa_meta_objset,
3757 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3758 	    sizeof (uint64_t), 1, &version, tx) != 0) {
3759 		cmn_err(CE_PANIC, "failed to add pool version");
3760 	}
3761 
3762 	/* Newly created pools with the right version are always deflated. */
3763 	if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3764 		spa->spa_deflate = TRUE;
3765 		if (zap_add(spa->spa_meta_objset,
3766 		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3767 		    sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3768 			cmn_err(CE_PANIC, "failed to add deflate");
3769 		}
3770 	}
3771 
3772 	/*
3773 	 * Create the deferred-free bpobj.  Turn off compression
3774 	 * because sync-to-convergence takes longer if the blocksize
3775 	 * keeps changing.
3776 	 */
3777 	obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3778 	dmu_object_set_compress(spa->spa_meta_objset, obj,
3779 	    ZIO_COMPRESS_OFF, tx);
3780 	if (zap_add(spa->spa_meta_objset,
3781 	    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3782 	    sizeof (uint64_t), 1, &obj, tx) != 0) {
3783 		cmn_err(CE_PANIC, "failed to add bpobj");
3784 	}
3785 	VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3786 	    spa->spa_meta_objset, obj));
3787 
3788 	/*
3789 	 * Create the pool's history object.
3790 	 */
3791 	if (version >= SPA_VERSION_ZPOOL_HISTORY)
3792 		spa_history_create_obj(spa, tx);
3793 
3794 	/*
3795 	 * Generate some random noise for salted checksums to operate on.
3796 	 */
3797 	(void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
3798 	    sizeof (spa->spa_cksum_salt.zcs_bytes));
3799 
3800 	/*
3801 	 * Set pool properties.
3802 	 */
3803 	spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3804 	spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3805 	spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3806 	spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3807 
3808 	if (props != NULL) {
3809 		spa_configfile_set(spa, props, B_FALSE);
3810 		spa_sync_props(props, tx);
3811 	}
3812 
3813 	dmu_tx_commit(tx);
3814 
3815 	spa->spa_sync_on = B_TRUE;
3816 	txg_sync_start(spa->spa_dsl_pool);
3817 
3818 	/*
3819 	 * We explicitly wait for the first transaction to complete so that our
3820 	 * bean counters are appropriately updated.
3821 	 */
3822 	txg_wait_synced(spa->spa_dsl_pool, txg);
3823 
3824 	spa_config_sync(spa, B_FALSE, B_TRUE);
3825 	spa_event_notify(spa, NULL, ESC_ZFS_POOL_CREATE);
3826 
3827 	spa_history_log_version(spa, "create");
3828 
3829 	/*
3830 	 * Don't count references from objsets that are already closed
3831 	 * and are making their way through the eviction process.
3832 	 */
3833 	spa_evicting_os_wait(spa);
3834 	spa->spa_minref = refcount_count(&spa->spa_refcount);
3835 	spa->spa_load_state = SPA_LOAD_NONE;
3836 
3837 	mutex_exit(&spa_namespace_lock);
3838 
3839 	return (0);
3840 }
3841 
3842 #ifdef _KERNEL
3843 /*
3844  * Get the root pool information from the root disk, then import the root pool
3845  * during the system boot up time.
3846  */
3847 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3848 
3849 static nvlist_t *
3850 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3851 {
3852 	nvlist_t *config;
3853 	nvlist_t *nvtop, *nvroot;
3854 	uint64_t pgid;
3855 
3856 	if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3857 		return (NULL);
3858 
3859 	/*
3860 	 * Add this top-level vdev to the child array.
3861 	 */
3862 	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3863 	    &nvtop) == 0);
3864 	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3865 	    &pgid) == 0);
3866 	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3867 
3868 	/*
3869 	 * Put this pool's top-level vdevs into a root vdev.
3870 	 */
3871 	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3872 	VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3873 	    VDEV_TYPE_ROOT) == 0);
3874 	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3875 	VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3876 	VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3877 	    &nvtop, 1) == 0);
3878 
3879 	/*
3880 	 * Replace the existing vdev_tree with the new root vdev in
3881 	 * this pool's configuration (remove the old, add the new).
3882 	 */
3883 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3884 	nvlist_free(nvroot);
3885 	return (config);
3886 }
3887 
3888 /*
3889  * Walk the vdev tree and see if we can find a device with "better"
3890  * configuration. A configuration is "better" if the label on that
3891  * device has a more recent txg.
3892  */
3893 static void
3894 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3895 {
3896 	for (int c = 0; c < vd->vdev_children; c++)
3897 		spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3898 
3899 	if (vd->vdev_ops->vdev_op_leaf) {
3900 		nvlist_t *label;
3901 		uint64_t label_txg;
3902 
3903 		if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3904 		    &label) != 0)
3905 			return;
3906 
3907 		VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3908 		    &label_txg) == 0);
3909 
3910 		/*
3911 		 * Do we have a better boot device?
3912 		 */
3913 		if (label_txg > *txg) {
3914 			*txg = label_txg;
3915 			*avd = vd;
3916 		}
3917 		nvlist_free(label);
3918 	}
3919 }
3920 
3921 /*
3922  * Import a root pool.
3923  *
3924  * For x86. devpath_list will consist of devid and/or physpath name of
3925  * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3926  * The GRUB "findroot" command will return the vdev we should boot.
3927  *
3928  * For Sparc, devpath_list consists the physpath name of the booting device
3929  * no matter the rootpool is a single device pool or a mirrored pool.
3930  * e.g.
3931  *	"/pci@1f,0/ide@d/disk@0,0:a"
3932  */
3933 int
3934 spa_import_rootpool(char *devpath, char *devid)
3935 {
3936 	spa_t *spa;
3937 	vdev_t *rvd, *bvd, *avd = NULL;
3938 	nvlist_t *config, *nvtop;
3939 	uint64_t guid, txg;
3940 	char *pname;
3941 	int error;
3942 
3943 	/*
3944 	 * Read the label from the boot device and generate a configuration.
3945 	 */
3946 	config = spa_generate_rootconf(devpath, devid, &guid);
3947 #if defined(_OBP) && defined(_KERNEL)
3948 	if (config == NULL) {
3949 		if (strstr(devpath, "/iscsi/ssd") != NULL) {
3950 			/* iscsi boot */
3951 			get_iscsi_bootpath_phy(devpath);
3952 			config = spa_generate_rootconf(devpath, devid, &guid);
3953 		}
3954 	}
3955 #endif
3956 	if (config == NULL) {
3957 		cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3958 		    devpath);
3959 		return (SET_ERROR(EIO));
3960 	}
3961 
3962 	VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3963 	    &pname) == 0);
3964 	VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3965 
3966 	mutex_enter(&spa_namespace_lock);
3967 	if ((spa = spa_lookup(pname)) != NULL) {
3968 		/*
3969 		 * Remove the existing root pool from the namespace so that we
3970 		 * can replace it with the correct config we just read in.
3971 		 */
3972 		spa_remove(spa);
3973 	}
3974 
3975 	spa = spa_add(pname, config, NULL);
3976 	spa->spa_is_root = B_TRUE;
3977 	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3978 
3979 	/*
3980 	 * Build up a vdev tree based on the boot device's label config.
3981 	 */
3982 	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3983 	    &nvtop) == 0);
3984 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3985 	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3986 	    VDEV_ALLOC_ROOTPOOL);
3987 	spa_config_exit(spa, SCL_ALL, FTAG);
3988 	if (error) {
3989 		mutex_exit(&spa_namespace_lock);
3990 		nvlist_free(config);
3991 		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3992 		    pname);
3993 		return (error);
3994 	}
3995 
3996 	/*
3997 	 * Get the boot vdev.
3998 	 */
3999 	if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
4000 		cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
4001 		    (u_longlong_t)guid);
4002 		error = SET_ERROR(ENOENT);
4003 		goto out;
4004 	}
4005 
4006 	/*
4007 	 * Determine if there is a better boot device.
4008 	 */
4009 	avd = bvd;
4010 	spa_alt_rootvdev(rvd, &avd, &txg);
4011 	if (avd != bvd) {
4012 		cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
4013 		    "try booting from '%s'", avd->vdev_path);
4014 		error = SET_ERROR(EINVAL);
4015 		goto out;
4016 	}
4017 
4018 	/*
4019 	 * If the boot device is part of a spare vdev then ensure that
4020 	 * we're booting off the active spare.
4021 	 */
4022 	if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
4023 	    !bvd->vdev_isspare) {
4024 		cmn_err(CE_NOTE, "The boot device is currently spared. Please "
4025 		    "try booting from '%s'",
4026 		    bvd->vdev_parent->
4027 		    vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
4028 		error = SET_ERROR(EINVAL);
4029 		goto out;
4030 	}
4031 
4032 	error = 0;
4033 out:
4034 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4035 	vdev_free(rvd);
4036 	spa_config_exit(spa, SCL_ALL, FTAG);
4037 	mutex_exit(&spa_namespace_lock);
4038 
4039 	nvlist_free(config);
4040 	return (error);
4041 }
4042 
4043 #endif
4044 
4045 /*
4046  * Import a non-root pool into the system.
4047  */
4048 int
4049 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
4050 {
4051 	spa_t *spa;
4052 	char *altroot = NULL;
4053 	spa_load_state_t state = SPA_LOAD_IMPORT;
4054 	zpool_rewind_policy_t policy;
4055 	uint64_t mode = spa_mode_global;
4056 	uint64_t readonly = B_FALSE;
4057 	int error;
4058 	nvlist_t *nvroot;
4059 	nvlist_t **spares, **l2cache;
4060 	uint_t nspares, nl2cache;
4061 
4062 	/*
4063 	 * If a pool with this name exists, return failure.
4064 	 */
4065 	mutex_enter(&spa_namespace_lock);
4066 	if (spa_lookup(pool) != NULL) {
4067 		mutex_exit(&spa_namespace_lock);
4068 		return (SET_ERROR(EEXIST));
4069 	}
4070 
4071 	/*
4072 	 * Create and initialize the spa structure.
4073 	 */
4074 	(void) nvlist_lookup_string(props,
4075 	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4076 	(void) nvlist_lookup_uint64(props,
4077 	    zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
4078 	if (readonly)
4079 		mode = FREAD;
4080 	spa = spa_add(pool, config, altroot);
4081 	spa->spa_import_flags = flags;
4082 
4083 	/*
4084 	 * Verbatim import - Take a pool and insert it into the namespace
4085 	 * as if it had been loaded at boot.
4086 	 */
4087 	if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
4088 		if (props != NULL)
4089 			spa_configfile_set(spa, props, B_FALSE);
4090 
4091 		spa_config_sync(spa, B_FALSE, B_TRUE);
4092 		spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4093 
4094 		mutex_exit(&spa_namespace_lock);
4095 		return (0);
4096 	}
4097 
4098 	spa_activate(spa, mode);
4099 
4100 	/*
4101 	 * Don't start async tasks until we know everything is healthy.
4102 	 */
4103 	spa_async_suspend(spa);
4104 
4105 	zpool_get_rewind_policy(config, &policy);
4106 	if (policy.zrp_request & ZPOOL_DO_REWIND)
4107 		state = SPA_LOAD_RECOVER;
4108 
4109 	/*
4110 	 * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
4111 	 * because the user-supplied config is actually the one to trust when
4112 	 * doing an import.
4113 	 */
4114 	if (state != SPA_LOAD_RECOVER)
4115 		spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4116 
4117 	error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
4118 	    policy.zrp_request);
4119 
4120 	/*
4121 	 * Propagate anything learned while loading the pool and pass it
4122 	 * back to caller (i.e. rewind info, missing devices, etc).
4123 	 */
4124 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4125 	    spa->spa_load_info) == 0);
4126 
4127 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4128 	/*
4129 	 * Toss any existing sparelist, as it doesn't have any validity
4130 	 * anymore, and conflicts with spa_has_spare().
4131 	 */
4132 	if (spa->spa_spares.sav_config) {
4133 		nvlist_free(spa->spa_spares.sav_config);
4134 		spa->spa_spares.sav_config = NULL;
4135 		spa_load_spares(spa);
4136 	}
4137 	if (spa->spa_l2cache.sav_config) {
4138 		nvlist_free(spa->spa_l2cache.sav_config);
4139 		spa->spa_l2cache.sav_config = NULL;
4140 		spa_load_l2cache(spa);
4141 	}
4142 
4143 	VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4144 	    &nvroot) == 0);
4145 	if (error == 0)
4146 		error = spa_validate_aux(spa, nvroot, -1ULL,
4147 		    VDEV_ALLOC_SPARE);
4148 	if (error == 0)
4149 		error = spa_validate_aux(spa, nvroot, -1ULL,
4150 		    VDEV_ALLOC_L2CACHE);
4151 	spa_config_exit(spa, SCL_ALL, FTAG);
4152 
4153 	if (props != NULL)
4154 		spa_configfile_set(spa, props, B_FALSE);
4155 
4156 	if (error != 0 || (props && spa_writeable(spa) &&
4157 	    (error = spa_prop_set(spa, props)))) {
4158 		spa_unload(spa);
4159 		spa_deactivate(spa);
4160 		spa_remove(spa);
4161 		mutex_exit(&spa_namespace_lock);
4162 		return (error);
4163 	}
4164 
4165 	spa_async_resume(spa);
4166 
4167 	/*
4168 	 * Override any spares and level 2 cache devices as specified by
4169 	 * the user, as these may have correct device names/devids, etc.
4170 	 */
4171 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4172 	    &spares, &nspares) == 0) {
4173 		if (spa->spa_spares.sav_config)
4174 			VERIFY(nvlist_remove(spa->spa_spares.sav_config,
4175 			    ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
4176 		else
4177 			VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
4178 			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4179 		VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4180 		    ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4181 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4182 		spa_load_spares(spa);
4183 		spa_config_exit(spa, SCL_ALL, FTAG);
4184 		spa->spa_spares.sav_sync = B_TRUE;
4185 	}
4186 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4187 	    &l2cache, &nl2cache) == 0) {
4188 		if (spa->spa_l2cache.sav_config)
4189 			VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
4190 			    ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
4191 		else
4192 			VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4193 			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
4194 		VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4195 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4196 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4197 		spa_load_l2cache(spa);
4198 		spa_config_exit(spa, SCL_ALL, FTAG);
4199 		spa->spa_l2cache.sav_sync = B_TRUE;
4200 	}
4201 
4202 	/*
4203 	 * Check for any removed devices.
4204 	 */
4205 	if (spa->spa_autoreplace) {
4206 		spa_aux_check_removed(&spa->spa_spares);
4207 		spa_aux_check_removed(&spa->spa_l2cache);
4208 	}
4209 
4210 	if (spa_writeable(spa)) {
4211 		/*
4212 		 * Update the config cache to include the newly-imported pool.
4213 		 */
4214 		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4215 	}
4216 
4217 	/*
4218 	 * It's possible that the pool was expanded while it was exported.
4219 	 * We kick off an async task to handle this for us.
4220 	 */
4221 	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
4222 
4223 	spa_history_log_version(spa, "import");
4224 
4225 	spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
4226 
4227 	mutex_exit(&spa_namespace_lock);
4228 
4229 	return (0);
4230 }
4231 
4232 nvlist_t *
4233 spa_tryimport(nvlist_t *tryconfig)
4234 {
4235 	nvlist_t *config = NULL;
4236 	char *poolname;
4237 	spa_t *spa;
4238 	uint64_t state;
4239 	int error;
4240 
4241 	if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
4242 		return (NULL);
4243 
4244 	if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
4245 		return (NULL);
4246 
4247 	/*
4248 	 * Create and initialize the spa structure.
4249 	 */
4250 	mutex_enter(&spa_namespace_lock);
4251 	spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
4252 	spa_activate(spa, FREAD);
4253 
4254 	/*
4255 	 * Pass off the heavy lifting to spa_load().
4256 	 * Pass TRUE for mosconfig because the user-supplied config
4257 	 * is actually the one to trust when doing an import.
4258 	 */
4259 	error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4260 
4261 	/*
4262 	 * If 'tryconfig' was at least parsable, return the current config.
4263 	 */
4264 	if (spa->spa_root_vdev != NULL) {
4265 		config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4266 		VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4267 		    poolname) == 0);
4268 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4269 		    state) == 0);
4270 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4271 		    spa->spa_uberblock.ub_timestamp) == 0);
4272 		VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4273 		    spa->spa_load_info) == 0);
4274 
4275 		/*
4276 		 * If the bootfs property exists on this pool then we
4277 		 * copy it out so that external consumers can tell which
4278 		 * pools are bootable.
4279 		 */
4280 		if ((!error || error == EEXIST) && spa->spa_bootfs) {
4281 			char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4282 
4283 			/*
4284 			 * We have to play games with the name since the
4285 			 * pool was opened as TRYIMPORT_NAME.
4286 			 */
4287 			if (dsl_dsobj_to_dsname(spa_name(spa),
4288 			    spa->spa_bootfs, tmpname) == 0) {
4289 				char *cp;
4290 				char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4291 
4292 				cp = strchr(tmpname, '/');
4293 				if (cp == NULL) {
4294 					(void) strlcpy(dsname, tmpname,
4295 					    MAXPATHLEN);
4296 				} else {
4297 					(void) snprintf(dsname, MAXPATHLEN,
4298 					    "%s/%s", poolname, ++cp);
4299 				}
4300 				VERIFY(nvlist_add_string(config,
4301 				    ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4302 				kmem_free(dsname, MAXPATHLEN);
4303 			}
4304 			kmem_free(tmpname, MAXPATHLEN);
4305 		}
4306 
4307 		/*
4308 		 * Add the list of hot spares and level 2 cache devices.
4309 		 */
4310 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4311 		spa_add_spares(spa, config);
4312 		spa_add_l2cache(spa, config);
4313 		spa_config_exit(spa, SCL_CONFIG, FTAG);
4314 	}
4315 
4316 	spa_unload(spa);
4317 	spa_deactivate(spa);
4318 	spa_remove(spa);
4319 	mutex_exit(&spa_namespace_lock);
4320 
4321 	return (config);
4322 }
4323 
4324 /*
4325  * Pool export/destroy
4326  *
4327  * The act of destroying or exporting a pool is very simple.  We make sure there
4328  * is no more pending I/O and any references to the pool are gone.  Then, we
4329  * update the pool state and sync all the labels to disk, removing the
4330  * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4331  * we don't sync the labels or remove the configuration cache.
4332  */
4333 static int
4334 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4335     boolean_t force, boolean_t hardforce)
4336 {
4337 	spa_t *spa;
4338 
4339 	if (oldconfig)
4340 		*oldconfig = NULL;
4341 
4342 	if (!(spa_mode_global & FWRITE))
4343 		return (SET_ERROR(EROFS));
4344 
4345 	mutex_enter(&spa_namespace_lock);
4346 	if ((spa = spa_lookup(pool)) == NULL) {
4347 		mutex_exit(&spa_namespace_lock);
4348 		return (SET_ERROR(ENOENT));
4349 	}
4350 
4351 	/*
4352 	 * Put a hold on the pool, drop the namespace lock, stop async tasks,
4353 	 * reacquire the namespace lock, and see if we can export.
4354 	 */
4355 	spa_open_ref(spa, FTAG);
4356 	mutex_exit(&spa_namespace_lock);
4357 	spa_async_suspend(spa);
4358 	mutex_enter(&spa_namespace_lock);
4359 	spa_close(spa, FTAG);
4360 
4361 	/*
4362 	 * The pool will be in core if it's openable,
4363 	 * in which case we can modify its state.
4364 	 */
4365 	if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4366 		/*
4367 		 * Objsets may be open only because they're dirty, so we
4368 		 * have to force it to sync before checking spa_refcnt.
4369 		 */
4370 		txg_wait_synced(spa->spa_dsl_pool, 0);
4371 		spa_evicting_os_wait(spa);
4372 
4373 		/*
4374 		 * A pool cannot be exported or destroyed if there are active
4375 		 * references.  If we are resetting a pool, allow references by
4376 		 * fault injection handlers.
4377 		 */
4378 		if (!spa_refcount_zero(spa) ||
4379 		    (spa->spa_inject_ref != 0 &&
4380 		    new_state != POOL_STATE_UNINITIALIZED)) {
4381 			spa_async_resume(spa);
4382 			mutex_exit(&spa_namespace_lock);
4383 			return (SET_ERROR(EBUSY));
4384 		}
4385 
4386 		/*
4387 		 * A pool cannot be exported if it has an active shared spare.
4388 		 * This is to prevent other pools stealing the active spare
4389 		 * from an exported pool. At user's own will, such pool can
4390 		 * be forcedly exported.
4391 		 */
4392 		if (!force && new_state == POOL_STATE_EXPORTED &&
4393 		    spa_has_active_shared_spare(spa)) {
4394 			spa_async_resume(spa);
4395 			mutex_exit(&spa_namespace_lock);
4396 			return (SET_ERROR(EXDEV));
4397 		}
4398 
4399 		/*
4400 		 * We want this to be reflected on every label,
4401 		 * so mark them all dirty.  spa_unload() will do the
4402 		 * final sync that pushes these changes out.
4403 		 */
4404 		if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4405 			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4406 			spa->spa_state = new_state;
4407 			spa->spa_final_txg = spa_last_synced_txg(spa) +
4408 			    TXG_DEFER_SIZE + 1;
4409 			vdev_config_dirty(spa->spa_root_vdev);
4410 			spa_config_exit(spa, SCL_ALL, FTAG);
4411 		}
4412 	}
4413 
4414 	spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4415 
4416 	if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4417 		spa_unload(spa);
4418 		spa_deactivate(spa);
4419 	}
4420 
4421 	if (oldconfig && spa->spa_config)
4422 		VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4423 
4424 	if (new_state != POOL_STATE_UNINITIALIZED) {
4425 		if (!hardforce)
4426 			spa_config_sync(spa, B_TRUE, B_TRUE);
4427 		spa_remove(spa);
4428 	}
4429 	mutex_exit(&spa_namespace_lock);
4430 
4431 	return (0);
4432 }
4433 
4434 /*
4435  * Destroy a storage pool.
4436  */
4437 int
4438 spa_destroy(char *pool)
4439 {
4440 	return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4441 	    B_FALSE, B_FALSE));
4442 }
4443 
4444 /*
4445  * Export a storage pool.
4446  */
4447 int
4448 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4449     boolean_t hardforce)
4450 {
4451 	return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4452 	    force, hardforce));
4453 }
4454 
4455 /*
4456  * Similar to spa_export(), this unloads the spa_t without actually removing it
4457  * from the namespace in any way.
4458  */
4459 int
4460 spa_reset(char *pool)
4461 {
4462 	return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4463 	    B_FALSE, B_FALSE));
4464 }
4465 
4466 /*
4467  * ==========================================================================
4468  * Device manipulation
4469  * ==========================================================================
4470  */
4471 
4472 /*
4473  * Add a device to a storage pool.
4474  */
4475 int
4476 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4477 {
4478 	uint64_t txg, id;
4479 	int error;
4480 	vdev_t *rvd = spa->spa_root_vdev;
4481 	vdev_t *vd, *tvd;
4482 	nvlist_t **spares, **l2cache;
4483 	uint_t nspares, nl2cache;
4484 
4485 	ASSERT(spa_writeable(spa));
4486 
4487 	txg = spa_vdev_enter(spa);
4488 
4489 	if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4490 	    VDEV_ALLOC_ADD)) != 0)
4491 		return (spa_vdev_exit(spa, NULL, txg, error));
4492 
4493 	spa->spa_pending_vdev = vd;	/* spa_vdev_exit() will clear this */
4494 
4495 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4496 	    &nspares) != 0)
4497 		nspares = 0;
4498 
4499 	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4500 	    &nl2cache) != 0)
4501 		nl2cache = 0;
4502 
4503 	if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4504 		return (spa_vdev_exit(spa, vd, txg, EINVAL));
4505 
4506 	if (vd->vdev_children != 0 &&
4507 	    (error = vdev_create(vd, txg, B_FALSE)) != 0)
4508 		return (spa_vdev_exit(spa, vd, txg, error));
4509 
4510 	/*
4511 	 * We must validate the spares and l2cache devices after checking the
4512 	 * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
4513 	 */
4514 	if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4515 		return (spa_vdev_exit(spa, vd, txg, error));
4516 
4517 	/*
4518 	 * Transfer each new top-level vdev from vd to rvd.
4519 	 */
4520 	for (int c = 0; c < vd->vdev_children; c++) {
4521 
4522 		/*
4523 		 * Set the vdev id to the first hole, if one exists.
4524 		 */
4525 		for (id = 0; id < rvd->vdev_children; id++) {
4526 			if (rvd->vdev_child[id]->vdev_ishole) {
4527 				vdev_free(rvd->vdev_child[id]);
4528 				break;
4529 			}
4530 		}
4531 		tvd = vd->vdev_child[c];
4532 		vdev_remove_child(vd, tvd);
4533 		tvd->vdev_id = id;
4534 		vdev_add_child(rvd, tvd);
4535 		vdev_config_dirty(tvd);
4536 	}
4537 
4538 	if (nspares != 0) {
4539 		spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4540 		    ZPOOL_CONFIG_SPARES);
4541 		spa_load_spares(spa);
4542 		spa->spa_spares.sav_sync = B_TRUE;
4543 	}
4544 
4545 	if (nl2cache != 0) {
4546 		spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4547 		    ZPOOL_CONFIG_L2CACHE);
4548 		spa_load_l2cache(spa);
4549 		spa->spa_l2cache.sav_sync = B_TRUE;
4550 	}
4551 
4552 	/*
4553 	 * We have to be careful when adding new vdevs to an existing pool.
4554 	 * If other threads start allocating from these vdevs before we
4555 	 * sync the config cache, and we lose power, then upon reboot we may
4556 	 * fail to open the pool because there are DVAs that the config cache
4557 	 * can't translate.  Therefore, we first add the vdevs without
4558 	 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4559 	 * and then let spa_config_update() initialize the new metaslabs.
4560 	 *
4561 	 * spa_load() checks for added-but-not-initialized vdevs, so that
4562 	 * if we lose power at any point in this sequence, the remaining
4563 	 * steps will be completed the next time we load the pool.
4564 	 */
4565 	(void) spa_vdev_exit(spa, vd, txg, 0);
4566 
4567 	mutex_enter(&spa_namespace_lock);
4568 	spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4569 	spa_event_notify(spa, NULL, ESC_ZFS_VDEV_ADD);
4570 	mutex_exit(&spa_namespace_lock);
4571 
4572 	return (0);
4573 }
4574 
4575 /*
4576  * Attach a device to a mirror.  The arguments are the path to any device
4577  * in the mirror, and the nvroot for the new device.  If the path specifies
4578  * a device that is not mirrored, we automatically insert the mirror vdev.
4579  *
4580  * If 'replacing' is specified, the new device is intended to replace the
4581  * existing device; in this case the two devices are made into their own
4582  * mirror using the 'replacing' vdev, which is functionally identical to
4583  * the mirror vdev (it actually reuses all the same ops) but has a few
4584  * extra rules: you can't attach to it after it's been created, and upon
4585  * completion of resilvering, the first disk (the one being replaced)
4586  * is automatically detached.
4587  */
4588 int
4589 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4590 {
4591 	uint64_t txg, dtl_max_txg;
4592 	vdev_t *rvd = spa->spa_root_vdev;
4593 	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4594 	vdev_ops_t *pvops;
4595 	char *oldvdpath, *newvdpath;
4596 	int newvd_isspare;
4597 	int error;
4598 
4599 	ASSERT(spa_writeable(spa));
4600 
4601 	txg = spa_vdev_enter(spa);
4602 
4603 	oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4604 
4605 	if (oldvd == NULL)
4606 		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4607 
4608 	if (!oldvd->vdev_ops->vdev_op_leaf)
4609 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4610 
4611 	pvd = oldvd->vdev_parent;
4612 
4613 	if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4614 	    VDEV_ALLOC_ATTACH)) != 0)
4615 		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4616 
4617 	if (newrootvd->vdev_children != 1)
4618 		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4619 
4620 	newvd = newrootvd->vdev_child[0];
4621 
4622 	if (!newvd->vdev_ops->vdev_op_leaf)
4623 		return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4624 
4625 	if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4626 		return (spa_vdev_exit(spa, newrootvd, txg, error));
4627 
4628 	/*
4629 	 * Spares can't replace logs
4630 	 */
4631 	if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4632 		return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4633 
4634 	if (!replacing) {
4635 		/*
4636 		 * For attach, the only allowable parent is a mirror or the root
4637 		 * vdev.
4638 		 */
4639 		if (pvd->vdev_ops != &vdev_mirror_ops &&
4640 		    pvd->vdev_ops != &vdev_root_ops)
4641 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4642 
4643 		pvops = &vdev_mirror_ops;
4644 	} else {
4645 		/*
4646 		 * Active hot spares can only be replaced by inactive hot
4647 		 * spares.
4648 		 */
4649 		if (pvd->vdev_ops == &vdev_spare_ops &&
4650 		    oldvd->vdev_isspare &&
4651 		    !spa_has_spare(spa, newvd->vdev_guid))
4652 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4653 
4654 		/*
4655 		 * If the source is a hot spare, and the parent isn't already a
4656 		 * spare, then we want to create a new hot spare.  Otherwise, we
4657 		 * want to create a replacing vdev.  The user is not allowed to
4658 		 * attach to a spared vdev child unless the 'isspare' state is
4659 		 * the same (spare replaces spare, non-spare replaces
4660 		 * non-spare).
4661 		 */
4662 		if (pvd->vdev_ops == &vdev_replacing_ops &&
4663 		    spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4664 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4665 		} else if (pvd->vdev_ops == &vdev_spare_ops &&
4666 		    newvd->vdev_isspare != oldvd->vdev_isspare) {
4667 			return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4668 		}
4669 
4670 		if (newvd->vdev_isspare)
4671 			pvops = &vdev_spare_ops;
4672 		else
4673 			pvops = &vdev_replacing_ops;
4674 	}
4675 
4676 	/*
4677 	 * Make sure the new device is big enough.
4678 	 */
4679 	if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4680 		return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4681 
4682 	/*
4683 	 * The new device cannot have a higher alignment requirement
4684 	 * than the top-level vdev.
4685 	 */
4686 	if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4687 		return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4688 
4689 	/*
4690 	 * If this is an in-place replacement, update oldvd's path and devid
4691 	 * to make it distinguishable from newvd, and unopenable from now on.
4692 	 */
4693 	if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4694 		spa_strfree(oldvd->vdev_path);
4695 		oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4696 		    KM_SLEEP);
4697 		(void) sprintf(oldvd->vdev_path, "%s/%s",
4698 		    newvd->vdev_path, "old");
4699 		if (oldvd->vdev_devid != NULL) {
4700 			spa_strfree(oldvd->vdev_devid);
4701 			oldvd->vdev_devid = NULL;
4702 		}
4703 	}
4704 
4705 	/* mark the device being resilvered */
4706 	newvd->vdev_resilver_txg = txg;
4707 
4708 	/*
4709 	 * If the parent is not a mirror, or if we're replacing, insert the new
4710 	 * mirror/replacing/spare vdev above oldvd.
4711 	 */
4712 	if (pvd->vdev_ops != pvops)
4713 		pvd = vdev_add_parent(oldvd, pvops);
4714 
4715 	ASSERT(pvd->vdev_top->vdev_parent == rvd);
4716 	ASSERT(pvd->vdev_ops == pvops);
4717 	ASSERT(oldvd->vdev_parent == pvd);
4718 
4719 	/*
4720 	 * Extract the new device from its root and add it to pvd.
4721 	 */
4722 	vdev_remove_child(newrootvd, newvd);
4723 	newvd->vdev_id = pvd->vdev_children;
4724 	newvd->vdev_crtxg = oldvd->vdev_crtxg;
4725 	vdev_add_child(pvd, newvd);
4726 
4727 	tvd = newvd->vdev_top;
4728 	ASSERT(pvd->vdev_top == tvd);
4729 	ASSERT(tvd->vdev_parent == rvd);
4730 
4731 	vdev_config_dirty(tvd);
4732 
4733 	/*
4734 	 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4735 	 * for any dmu_sync-ed blocks.  It will propagate upward when
4736 	 * spa_vdev_exit() calls vdev_dtl_reassess().
4737 	 */
4738 	dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4739 
4740 	vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4741 	    dtl_max_txg - TXG_INITIAL);
4742 
4743 	if (newvd->vdev_isspare) {
4744 		spa_spare_activate(newvd);
4745 		spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4746 	}
4747 
4748 	oldvdpath = spa_strdup(oldvd->vdev_path);
4749 	newvdpath = spa_strdup(newvd->vdev_path);
4750 	newvd_isspare = newvd->vdev_isspare;
4751 
4752 	/*
4753 	 * Mark newvd's DTL dirty in this txg.
4754 	 */
4755 	vdev_dirty(tvd, VDD_DTL, newvd, txg);
4756 
4757 	/*
4758 	 * Schedule the resilver to restart in the future. We do this to
4759 	 * ensure that dmu_sync-ed blocks have been stitched into the
4760 	 * respective datasets.
4761 	 */
4762 	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4763 
4764 	if (spa->spa_bootfs)
4765 		spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4766 
4767 	spa_event_notify(spa, newvd, ESC_ZFS_VDEV_ATTACH);
4768 
4769 	/*
4770 	 * Commit the config
4771 	 */
4772 	(void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4773 
4774 	spa_history_log_internal(spa, "vdev attach", NULL,
4775 	    "%s vdev=%s %s vdev=%s",
4776 	    replacing && newvd_isspare ? "spare in" :
4777 	    replacing ? "replace" : "attach", newvdpath,
4778 	    replacing ? "for" : "to", oldvdpath);
4779 
4780 	spa_strfree(oldvdpath);
4781 	spa_strfree(newvdpath);
4782 
4783 	return (0);
4784 }
4785 
4786 /*
4787  * Detach a device from a mirror or replacing vdev.
4788  *
4789  * If 'replace_done' is specified, only detach if the parent
4790  * is a replacing vdev.
4791  */
4792 int
4793 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4794 {
4795 	uint64_t txg;
4796 	int error;
4797 	vdev_t *rvd = spa->spa_root_vdev;
4798 	vdev_t *vd, *pvd, *cvd, *tvd;
4799 	boolean_t unspare = B_FALSE;
4800 	uint64_t unspare_guid = 0;
4801 	char *vdpath;
4802 
4803 	ASSERT(spa_writeable(spa));
4804 
4805 	txg = spa_vdev_enter(spa);
4806 
4807 	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4808 
4809 	if (vd == NULL)
4810 		return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4811 
4812 	if (!vd->vdev_ops->vdev_op_leaf)
4813 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4814 
4815 	pvd = vd->vdev_parent;
4816 
4817 	/*
4818 	 * If the parent/child relationship is not as expected, don't do it.
4819 	 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4820 	 * vdev that's replacing B with C.  The user's intent in replacing
4821 	 * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4822 	 * the replace by detaching C, the expected behavior is to end up
4823 	 * M(A,B).  But suppose that right after deciding to detach C,
4824 	 * the replacement of B completes.  We would have M(A,C), and then
4825 	 * ask to detach C, which would leave us with just A -- not what
4826 	 * the user wanted.  To prevent this, we make sure that the
4827 	 * parent/child relationship hasn't changed -- in this example,
4828 	 * that C's parent is still the replacing vdev R.
4829 	 */
4830 	if (pvd->vdev_guid != pguid && pguid != 0)
4831 		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4832 
4833 	/*
4834 	 * Only 'replacing' or 'spare' vdevs can be replaced.
4835 	 */
4836 	if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4837 	    pvd->vdev_ops != &vdev_spare_ops)
4838 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4839 
4840 	ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4841 	    spa_version(spa) >= SPA_VERSION_SPARES);
4842 
4843 	/*
4844 	 * Only mirror, replacing, and spare vdevs support detach.
4845 	 */
4846 	if (pvd->vdev_ops != &vdev_replacing_ops &&
4847 	    pvd->vdev_ops != &vdev_mirror_ops &&
4848 	    pvd->vdev_ops != &vdev_spare_ops)
4849 		return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4850 
4851 	/*
4852 	 * If this device has the only valid copy of some data,
4853 	 * we cannot safely detach it.
4854 	 */
4855 	if (vdev_dtl_required(vd))
4856 		return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4857 
4858 	ASSERT(pvd->vdev_children >= 2);
4859 
4860 	/*
4861 	 * If we are detaching the second disk from a replacing vdev, then
4862 	 * check to see if we changed the original vdev's path to have "/old"
4863 	 * at the end in spa_vdev_attach().  If so, undo that change now.
4864 	 */
4865 	if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4866 	    vd->vdev_path != NULL) {
4867 		size_t len = strlen(vd->vdev_path);
4868 
4869 		for (int c = 0; c < pvd->vdev_children; c++) {
4870 			cvd = pvd->vdev_child[c];
4871 
4872 			if (cvd == vd || cvd->vdev_path == NULL)
4873 				continue;
4874 
4875 			if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4876 			    strcmp(cvd->vdev_path + len, "/old") == 0) {
4877 				spa_strfree(cvd->vdev_path);
4878 				cvd->vdev_path = spa_strdup(vd->vdev_path);
4879 				break;
4880 			}
4881 		}
4882 	}
4883 
4884 	/*
4885 	 * If we are detaching the original disk from a spare, then it implies
4886 	 * that the spare should become a real disk, and be removed from the
4887 	 * active spare list for the pool.
4888 	 */
4889 	if (pvd->vdev_ops == &vdev_spare_ops &&
4890 	    vd->vdev_id == 0 &&
4891 	    pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4892 		unspare = B_TRUE;
4893 
4894 	/*
4895 	 * Erase the disk labels so the disk can be used for other things.
4896 	 * This must be done after all other error cases are handled,
4897 	 * but before we disembowel vd (so we can still do I/O to it).
4898 	 * But if we can't do it, don't treat the error as fatal --
4899 	 * it may be that the unwritability of the disk is the reason
4900 	 * it's being detached!
4901 	 */
4902 	error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4903 
4904 	/*
4905 	 * Remove vd from its parent and compact the parent's children.
4906 	 */
4907 	vdev_remove_child(pvd, vd);
4908 	vdev_compact_children(pvd);
4909 
4910 	/*
4911 	 * Remember one of the remaining children so we can get tvd below.
4912 	 */
4913 	cvd = pvd->vdev_child[pvd->vdev_children - 1];
4914 
4915 	/*
4916 	 * If we need to remove the remaining child from the list of hot spares,
4917 	 * do it now, marking the vdev as no longer a spare in the process.
4918 	 * We must do this before vdev_remove_parent(), because that can
4919 	 * change the GUID if it creates a new toplevel GUID.  For a similar
4920 	 * reason, we must remove the spare now, in the same txg as the detach;
4921 	 * otherwise someone could attach a new sibling, change the GUID, and
4922 	 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4923 	 */
4924 	if (unspare) {
4925 		ASSERT(cvd->vdev_isspare);
4926 		spa_spare_remove(cvd);
4927 		unspare_guid = cvd->vdev_guid;
4928 		(void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4929 		cvd->vdev_unspare = B_TRUE;
4930 	}
4931 
4932 	/*
4933 	 * If the parent mirror/replacing vdev only has one child,
4934 	 * the parent is no longer needed.  Remove it from the tree.
4935 	 */
4936 	if (pvd->vdev_children == 1) {
4937 		if (pvd->vdev_ops == &vdev_spare_ops)
4938 			cvd->vdev_unspare = B_FALSE;
4939 		vdev_remove_parent(cvd);
4940 	}
4941 
4942 
4943 	/*
4944 	 * We don't set tvd until now because the parent we just removed
4945 	 * may have been the previous top-level vdev.
4946 	 */
4947 	tvd = cvd->vdev_top;
4948 	ASSERT(tvd->vdev_parent == rvd);
4949 
4950 	/*
4951 	 * Reevaluate the parent vdev state.
4952 	 */
4953 	vdev_propagate_state(cvd);
4954 
4955 	/*
4956 	 * If the 'autoexpand' property is set on the pool then automatically
4957 	 * try to expand the size of the pool. For example if the device we
4958 	 * just detached was smaller than the others, it may be possible to
4959 	 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4960 	 * first so that we can obtain the updated sizes of the leaf vdevs.
4961 	 */
4962 	if (spa->spa_autoexpand) {
4963 		vdev_reopen(tvd);
4964 		vdev_expand(tvd, txg);
4965 	}
4966 
4967 	vdev_config_dirty(tvd);
4968 
4969 	/*
4970 	 * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4971 	 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4972 	 * But first make sure we're not on any *other* txg's DTL list, to
4973 	 * prevent vd from being accessed after it's freed.
4974 	 */
4975 	vdpath = spa_strdup(vd->vdev_path);
4976 	for (int t = 0; t < TXG_SIZE; t++)
4977 		(void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4978 	vd->vdev_detached = B_TRUE;
4979 	vdev_dirty(tvd, VDD_DTL, vd, txg);
4980 
4981 	spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4982 
4983 	/* hang on to the spa before we release the lock */
4984 	spa_open_ref(spa, FTAG);
4985 
4986 	error = spa_vdev_exit(spa, vd, txg, 0);
4987 
4988 	spa_history_log_internal(spa, "detach", NULL,
4989 	    "vdev=%s", vdpath);
4990 	spa_strfree(vdpath);
4991 
4992 	/*
4993 	 * If this was the removal of the original device in a hot spare vdev,
4994 	 * then we want to go through and remove the device from the hot spare
4995 	 * list of every other pool.
4996 	 */
4997 	if (unspare) {
4998 		spa_t *altspa = NULL;
4999 
5000 		mutex_enter(&spa_namespace_lock);
5001 		while ((altspa = spa_next(altspa)) != NULL) {
5002 			if (altspa->spa_state != POOL_STATE_ACTIVE ||
5003 			    altspa == spa)
5004 				continue;
5005 
5006 			spa_open_ref(altspa, FTAG);
5007 			mutex_exit(&spa_namespace_lock);
5008 			(void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
5009 			mutex_enter(&spa_namespace_lock);
5010 			spa_close(altspa, FTAG);
5011 		}
5012 		mutex_exit(&spa_namespace_lock);
5013 
5014 		/* search the rest of the vdevs for spares to remove */
5015 		spa_vdev_resilver_done(spa);
5016 	}
5017 
5018 	/* all done with the spa; OK to release */
5019 	mutex_enter(&spa_namespace_lock);
5020 	spa_close(spa, FTAG);
5021 	mutex_exit(&spa_namespace_lock);
5022 
5023 	return (error);
5024 }
5025 
5026 /*
5027  * Split a set of devices from their mirrors, and create a new pool from them.
5028  */
5029 int
5030 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
5031     nvlist_t *props, boolean_t exp)
5032 {
5033 	int error = 0;
5034 	uint64_t txg, *glist;
5035 	spa_t *newspa;
5036 	uint_t c, children, lastlog;
5037 	nvlist_t **child, *nvl, *tmp;
5038 	dmu_tx_t *tx;
5039 	char *altroot = NULL;
5040 	vdev_t *rvd, **vml = NULL;			/* vdev modify list */
5041 	boolean_t activate_slog;
5042 
5043 	ASSERT(spa_writeable(spa));
5044 
5045 	txg = spa_vdev_enter(spa);
5046 
5047 	/* clear the log and flush everything up to now */
5048 	activate_slog = spa_passivate_log(spa);
5049 	(void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5050 	error = spa_offline_log(spa);
5051 	txg = spa_vdev_config_enter(spa);
5052 
5053 	if (activate_slog)
5054 		spa_activate_log(spa);
5055 
5056 	if (error != 0)
5057 		return (spa_vdev_exit(spa, NULL, txg, error));
5058 
5059 	/* check new spa name before going any further */
5060 	if (spa_lookup(newname) != NULL)
5061 		return (spa_vdev_exit(spa, NULL, txg, EEXIST));
5062 
5063 	/*
5064 	 * scan through all the children to ensure they're all mirrors
5065 	 */
5066 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
5067 	    nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
5068 	    &children) != 0)
5069 		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5070 
5071 	/* first, check to ensure we've got the right child count */
5072 	rvd = spa->spa_root_vdev;
5073 	lastlog = 0;
5074 	for (c = 0; c < rvd->vdev_children; c++) {
5075 		vdev_t *vd = rvd->vdev_child[c];
5076 
5077 		/* don't count the holes & logs as children */
5078 		if (vd->vdev_islog || vd->vdev_ishole) {
5079 			if (lastlog == 0)
5080 				lastlog = c;
5081 			continue;
5082 		}
5083 
5084 		lastlog = 0;
5085 	}
5086 	if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
5087 		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5088 
5089 	/* next, ensure no spare or cache devices are part of the split */
5090 	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
5091 	    nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
5092 		return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5093 
5094 	vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
5095 	glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
5096 
5097 	/* then, loop over each vdev and validate it */
5098 	for (c = 0; c < children; c++) {
5099 		uint64_t is_hole = 0;
5100 
5101 		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
5102 		    &is_hole);
5103 
5104 		if (is_hole != 0) {
5105 			if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
5106 			    spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
5107 				continue;
5108 			} else {
5109 				error = SET_ERROR(EINVAL);
5110 				break;
5111 			}
5112 		}
5113 
5114 		/* which disk is going to be split? */
5115 		if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
5116 		    &glist[c]) != 0) {
5117 			error = SET_ERROR(EINVAL);
5118 			break;
5119 		}
5120 
5121 		/* look it up in the spa */
5122 		vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
5123 		if (vml[c] == NULL) {
5124 			error = SET_ERROR(ENODEV);
5125 			break;
5126 		}
5127 
5128 		/* make sure there's nothing stopping the split */
5129 		if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
5130 		    vml[c]->vdev_islog ||
5131 		    vml[c]->vdev_ishole ||
5132 		    vml[c]->vdev_isspare ||
5133 		    vml[c]->vdev_isl2cache ||
5134 		    !vdev_writeable(vml[c]) ||
5135 		    vml[c]->vdev_children != 0 ||
5136 		    vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
5137 		    c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
5138 			error = SET_ERROR(EINVAL);
5139 			break;
5140 		}
5141 
5142 		if (vdev_dtl_required(vml[c])) {
5143 			error = SET_ERROR(EBUSY);
5144 			break;
5145 		}
5146 
5147 		/* we need certain info from the top level */
5148 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
5149 		    vml[c]->vdev_top->vdev_ms_array) == 0);
5150 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
5151 		    vml[c]->vdev_top->vdev_ms_shift) == 0);
5152 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
5153 		    vml[c]->vdev_top->vdev_asize) == 0);
5154 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
5155 		    vml[c]->vdev_top->vdev_ashift) == 0);
5156 
5157 		/* transfer per-vdev ZAPs */
5158 		ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0);
5159 		VERIFY0(nvlist_add_uint64(child[c],
5160 		    ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap));
5161 
5162 		ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0);
5163 		VERIFY0(nvlist_add_uint64(child[c],
5164 		    ZPOOL_CONFIG_VDEV_TOP_ZAP,
5165 		    vml[c]->vdev_parent->vdev_top_zap));
5166 	}
5167 
5168 	if (error != 0) {
5169 		kmem_free(vml, children * sizeof (vdev_t *));
5170 		kmem_free(glist, children * sizeof (uint64_t));
5171 		return (spa_vdev_exit(spa, NULL, txg, error));
5172 	}
5173 
5174 	/* stop writers from using the disks */
5175 	for (c = 0; c < children; c++) {
5176 		if (vml[c] != NULL)
5177 			vml[c]->vdev_offline = B_TRUE;
5178 	}
5179 	vdev_reopen(spa->spa_root_vdev);
5180 
5181 	/*
5182 	 * Temporarily record the splitting vdevs in the spa config.  This
5183 	 * will disappear once the config is regenerated.
5184 	 */
5185 	VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5186 	VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
5187 	    glist, children) == 0);
5188 	kmem_free(glist, children * sizeof (uint64_t));
5189 
5190 	mutex_enter(&spa->spa_props_lock);
5191 	VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
5192 	    nvl) == 0);
5193 	mutex_exit(&spa->spa_props_lock);
5194 	spa->spa_config_splitting = nvl;
5195 	vdev_config_dirty(spa->spa_root_vdev);
5196 
5197 	/* configure and create the new pool */
5198 	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
5199 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5200 	    exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
5201 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5202 	    spa_version(spa)) == 0);
5203 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
5204 	    spa->spa_config_txg) == 0);
5205 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5206 	    spa_generate_guid(NULL)) == 0);
5207 	VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
5208 	(void) nvlist_lookup_string(props,
5209 	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5210 
5211 	/* add the new pool to the namespace */
5212 	newspa = spa_add(newname, config, altroot);
5213 	newspa->spa_avz_action = AVZ_ACTION_REBUILD;
5214 	newspa->spa_config_txg = spa->spa_config_txg;
5215 	spa_set_log_state(newspa, SPA_LOG_CLEAR);
5216 
5217 	/* release the spa config lock, retaining the namespace lock */
5218 	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5219 
5220 	if (zio_injection_enabled)
5221 		zio_handle_panic_injection(spa, FTAG, 1);
5222 
5223 	spa_activate(newspa, spa_mode_global);
5224 	spa_async_suspend(newspa);
5225 
5226 	/* create the new pool from the disks of the original pool */
5227 	error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
5228 	if (error)
5229 		goto out;
5230 
5231 	/* if that worked, generate a real config for the new pool */
5232 	if (newspa->spa_root_vdev != NULL) {
5233 		VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
5234 		    NV_UNIQUE_NAME, KM_SLEEP) == 0);
5235 		VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
5236 		    ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
5237 		spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
5238 		    B_TRUE));
5239 	}
5240 
5241 	/* set the props */
5242 	if (props != NULL) {
5243 		spa_configfile_set(newspa, props, B_FALSE);
5244 		error = spa_prop_set(newspa, props);
5245 		if (error)
5246 			goto out;
5247 	}
5248 
5249 	/* flush everything */
5250 	txg = spa_vdev_config_enter(newspa);
5251 	vdev_config_dirty(newspa->spa_root_vdev);
5252 	(void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
5253 
5254 	if (zio_injection_enabled)
5255 		zio_handle_panic_injection(spa, FTAG, 2);
5256 
5257 	spa_async_resume(newspa);
5258 
5259 	/* finally, update the original pool's config */
5260 	txg = spa_vdev_config_enter(spa);
5261 	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
5262 	error = dmu_tx_assign(tx, TXG_WAIT);
5263 	if (error != 0)
5264 		dmu_tx_abort(tx);
5265 	for (c = 0; c < children; c++) {
5266 		if (vml[c] != NULL) {
5267 			vdev_split(vml[c]);
5268 			if (error == 0)
5269 				spa_history_log_internal(spa, "detach", tx,
5270 				    "vdev=%s", vml[c]->vdev_path);
5271 
5272 			vdev_free(vml[c]);
5273 		}
5274 	}
5275 	spa->spa_avz_action = AVZ_ACTION_REBUILD;
5276 	vdev_config_dirty(spa->spa_root_vdev);
5277 	spa->spa_config_splitting = NULL;
5278 	nvlist_free(nvl);
5279 	if (error == 0)
5280 		dmu_tx_commit(tx);
5281 	(void) spa_vdev_exit(spa, NULL, txg, 0);
5282 
5283 	if (zio_injection_enabled)
5284 		zio_handle_panic_injection(spa, FTAG, 3);
5285 
5286 	/* split is complete; log a history record */
5287 	spa_history_log_internal(newspa, "split", NULL,
5288 	    "from pool %s", spa_name(spa));
5289 
5290 	kmem_free(vml, children * sizeof (vdev_t *));
5291 
5292 	/* if we're not going to mount the filesystems in userland, export */
5293 	if (exp)
5294 		error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5295 		    B_FALSE, B_FALSE);
5296 
5297 	return (error);
5298 
5299 out:
5300 	spa_unload(newspa);
5301 	spa_deactivate(newspa);
5302 	spa_remove(newspa);
5303 
5304 	txg = spa_vdev_config_enter(spa);
5305 
5306 	/* re-online all offlined disks */
5307 	for (c = 0; c < children; c++) {
5308 		if (vml[c] != NULL)
5309 			vml[c]->vdev_offline = B_FALSE;
5310 	}
5311 	vdev_reopen(spa->spa_root_vdev);
5312 
5313 	nvlist_free(spa->spa_config_splitting);
5314 	spa->spa_config_splitting = NULL;
5315 	(void) spa_vdev_exit(spa, NULL, txg, error);
5316 
5317 	kmem_free(vml, children * sizeof (vdev_t *));
5318 	return (error);
5319 }
5320 
5321 static nvlist_t *
5322 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5323 {
5324 	for (int i = 0; i < count; i++) {
5325 		uint64_t guid;
5326 
5327 		VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5328 		    &guid) == 0);
5329 
5330 		if (guid == target_guid)
5331 			return (nvpp[i]);
5332 	}
5333 
5334 	return (NULL);
5335 }
5336 
5337 static void
5338 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5339     nvlist_t *dev_to_remove)
5340 {
5341 	nvlist_t **newdev = NULL;
5342 
5343 	if (count > 1)
5344 		newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5345 
5346 	for (int i = 0, j = 0; i < count; i++) {
5347 		if (dev[i] == dev_to_remove)
5348 			continue;
5349 		VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5350 	}
5351 
5352 	VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5353 	VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5354 
5355 	for (int i = 0; i < count - 1; i++)
5356 		nvlist_free(newdev[i]);
5357 
5358 	if (count > 1)
5359 		kmem_free(newdev, (count - 1) * sizeof (void *));
5360 }
5361 
5362 /*
5363  * Evacuate the device.
5364  */
5365 static int
5366 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5367 {
5368 	uint64_t txg;
5369 	int error = 0;
5370 
5371 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5372 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5373 	ASSERT(vd == vd->vdev_top);
5374 
5375 	/*
5376 	 * Evacuate the device.  We don't hold the config lock as writer
5377 	 * since we need to do I/O but we do keep the
5378 	 * spa_namespace_lock held.  Once this completes the device
5379 	 * should no longer have any blocks allocated on it.
5380 	 */
5381 	if (vd->vdev_islog) {
5382 		if (vd->vdev_stat.vs_alloc != 0)
5383 			error = spa_offline_log(spa);
5384 	} else {
5385 		error = SET_ERROR(ENOTSUP);
5386 	}
5387 
5388 	if (error)
5389 		return (error);
5390 
5391 	/*
5392 	 * The evacuation succeeded.  Remove any remaining MOS metadata
5393 	 * associated with this vdev, and wait for these changes to sync.
5394 	 */
5395 	ASSERT0(vd->vdev_stat.vs_alloc);
5396 	txg = spa_vdev_config_enter(spa);
5397 	vd->vdev_removing = B_TRUE;
5398 	vdev_dirty_leaves(vd, VDD_DTL, txg);
5399 	vdev_config_dirty(vd);
5400 	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5401 
5402 	return (0);
5403 }
5404 
5405 /*
5406  * Complete the removal by cleaning up the namespace.
5407  */
5408 static void
5409 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5410 {
5411 	vdev_t *rvd = spa->spa_root_vdev;
5412 	uint64_t id = vd->vdev_id;
5413 	boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5414 
5415 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
5416 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5417 	ASSERT(vd == vd->vdev_top);
5418 
5419 	/*
5420 	 * Only remove any devices which are empty.
5421 	 */
5422 	if (vd->vdev_stat.vs_alloc != 0)
5423 		return;
5424 
5425 	(void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5426 
5427 	if (list_link_active(&vd->vdev_state_dirty_node))
5428 		vdev_state_clean(vd);
5429 	if (list_link_active(&vd->vdev_config_dirty_node))
5430 		vdev_config_clean(vd);
5431 
5432 	vdev_free(vd);
5433 
5434 	if (last_vdev) {
5435 		vdev_compact_children(rvd);
5436 	} else {
5437 		vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5438 		vdev_add_child(rvd, vd);
5439 	}
5440 	vdev_config_dirty(rvd);
5441 
5442 	/*
5443 	 * Reassess the health of our root vdev.
5444 	 */
5445 	vdev_reopen(rvd);
5446 }
5447 
5448 /*
5449  * Remove a device from the pool -
5450  *
5451  * Removing a device from the vdev namespace requires several steps
5452  * and can take a significant amount of time.  As a result we use
5453  * the spa_vdev_config_[enter/exit] functions which allow us to
5454  * grab and release the spa_config_lock while still holding the namespace
5455  * lock.  During each step the configuration is synced out.
5456  *
5457  * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5458  * devices.
5459  */
5460 int
5461 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5462 {
5463 	vdev_t *vd;
5464 	sysevent_t *ev = NULL;
5465 	metaslab_group_t *mg;
5466 	nvlist_t **spares, **l2cache, *nv;
5467 	uint64_t txg = 0;
5468 	uint_t nspares, nl2cache;
5469 	int error = 0;
5470 	boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5471 
5472 	ASSERT(spa_writeable(spa));
5473 
5474 	if (!locked)
5475 		txg = spa_vdev_enter(spa);
5476 
5477 	vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5478 
5479 	if (spa->spa_spares.sav_vdevs != NULL &&
5480 	    nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5481 	    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5482 	    (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5483 		/*
5484 		 * Only remove the hot spare if it's not currently in use
5485 		 * in this pool.
5486 		 */
5487 		if (vd == NULL || unspare) {
5488 			if (vd == NULL)
5489 				vd = spa_lookup_by_guid(spa, guid, B_TRUE);
5490 			ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX);
5491 			spa_vdev_remove_aux(spa->spa_spares.sav_config,
5492 			    ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5493 			spa_load_spares(spa);
5494 			spa->spa_spares.sav_sync = B_TRUE;
5495 		} else {
5496 			error = SET_ERROR(EBUSY);
5497 		}
5498 	} else if (spa->spa_l2cache.sav_vdevs != NULL &&
5499 	    nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5500 	    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5501 	    (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5502 		/*
5503 		 * Cache devices can always be removed.
5504 		 */
5505 		vd = spa_lookup_by_guid(spa, guid, B_TRUE);
5506 		ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX);
5507 		spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5508 		    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5509 		spa_load_l2cache(spa);
5510 		spa->spa_l2cache.sav_sync = B_TRUE;
5511 	} else if (vd != NULL && vd->vdev_islog) {
5512 		ASSERT(!locked);
5513 		ASSERT(vd == vd->vdev_top);
5514 
5515 		mg = vd->vdev_mg;
5516 
5517 		/*
5518 		 * Stop allocating from this vdev.
5519 		 */
5520 		metaslab_group_passivate(mg);
5521 
5522 		/*
5523 		 * Wait for the youngest allocations and frees to sync,
5524 		 * and then wait for the deferral of those frees to finish.
5525 		 */
5526 		spa_vdev_config_exit(spa, NULL,
5527 		    txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5528 
5529 		/*
5530 		 * Attempt to evacuate the vdev.
5531 		 */
5532 		error = spa_vdev_remove_evacuate(spa, vd);
5533 
5534 		txg = spa_vdev_config_enter(spa);
5535 
5536 		/*
5537 		 * If we couldn't evacuate the vdev, unwind.
5538 		 */
5539 		if (error) {
5540 			metaslab_group_activate(mg);
5541 			return (spa_vdev_exit(spa, NULL, txg, error));
5542 		}
5543 
5544 		/*
5545 		 * Clean up the vdev namespace.
5546 		 */
5547 		ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_DEV);
5548 		spa_vdev_remove_from_namespace(spa, vd);
5549 
5550 	} else if (vd != NULL) {
5551 		/*
5552 		 * Normal vdevs cannot be removed (yet).
5553 		 */
5554 		error = SET_ERROR(ENOTSUP);
5555 	} else {
5556 		/*
5557 		 * There is no vdev of any kind with the specified guid.
5558 		 */
5559 		error = SET_ERROR(ENOENT);
5560 	}
5561 
5562 	if (!locked)
5563 		error = spa_vdev_exit(spa, NULL, txg, error);
5564 
5565 	if (ev)
5566 		spa_event_post(ev);
5567 
5568 	return (error);
5569 }
5570 
5571 /*
5572  * Find any device that's done replacing, or a vdev marked 'unspare' that's
5573  * currently spared, so we can detach it.
5574  */
5575 static vdev_t *
5576 spa_vdev_resilver_done_hunt(vdev_t *vd)
5577 {
5578 	vdev_t *newvd, *oldvd;
5579 
5580 	for (int c = 0; c < vd->vdev_children; c++) {
5581 		oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5582 		if (oldvd != NULL)
5583 			return (oldvd);
5584 	}
5585 
5586 	/*
5587 	 * Check for a completed replacement.  We always consider the first
5588 	 * vdev in the list to be the oldest vdev, and the last one to be
5589 	 * the newest (see spa_vdev_attach() for how that works).  In
5590 	 * the case where the newest vdev is faulted, we will not automatically
5591 	 * remove it after a resilver completes.  This is OK as it will require
5592 	 * user intervention to determine which disk the admin wishes to keep.
5593 	 */
5594 	if (vd->vdev_ops == &vdev_replacing_ops) {
5595 		ASSERT(vd->vdev_children > 1);
5596 
5597 		newvd = vd->vdev_child[vd->vdev_children - 1];
5598 		oldvd = vd->vdev_child[0];
5599 
5600 		if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5601 		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5602 		    !vdev_dtl_required(oldvd))
5603 			return (oldvd);
5604 	}
5605 
5606 	/*
5607 	 * Check for a completed resilver with the 'unspare' flag set.
5608 	 */
5609 	if (vd->vdev_ops == &vdev_spare_ops) {
5610 		vdev_t *first = vd->vdev_child[0];
5611 		vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5612 
5613 		if (last->vdev_unspare) {
5614 			oldvd = first;
5615 			newvd = last;
5616 		} else if (first->vdev_unspare) {
5617 			oldvd = last;
5618 			newvd = first;
5619 		} else {
5620 			oldvd = NULL;
5621 		}
5622 
5623 		if (oldvd != NULL &&
5624 		    vdev_dtl_empty(newvd, DTL_MISSING) &&
5625 		    vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5626 		    !vdev_dtl_required(oldvd))
5627 			return (oldvd);
5628 
5629 		/*
5630 		 * If there are more than two spares attached to a disk,
5631 		 * and those spares are not required, then we want to
5632 		 * attempt to free them up now so that they can be used
5633 		 * by other pools.  Once we're back down to a single
5634 		 * disk+spare, we stop removing them.
5635 		 */
5636 		if (vd->vdev_children > 2) {
5637 			newvd = vd->vdev_child[1];
5638 
5639 			if (newvd->vdev_isspare && last->vdev_isspare &&
5640 			    vdev_dtl_empty(last, DTL_MISSING) &&
5641 			    vdev_dtl_empty(last, DTL_OUTAGE) &&
5642 			    !vdev_dtl_required(newvd))
5643 				return (newvd);
5644 		}
5645 	}
5646 
5647 	return (NULL);
5648 }
5649 
5650 static void
5651 spa_vdev_resilver_done(spa_t *spa)
5652 {
5653 	vdev_t *vd, *pvd, *ppvd;
5654 	uint64_t guid, sguid, pguid, ppguid;
5655 
5656 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5657 
5658 	while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5659 		pvd = vd->vdev_parent;
5660 		ppvd = pvd->vdev_parent;
5661 		guid = vd->vdev_guid;
5662 		pguid = pvd->vdev_guid;
5663 		ppguid = ppvd->vdev_guid;
5664 		sguid = 0;
5665 		/*
5666 		 * If we have just finished replacing a hot spared device, then
5667 		 * we need to detach the parent's first child (the original hot
5668 		 * spare) as well.
5669 		 */
5670 		if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5671 		    ppvd->vdev_children == 2) {
5672 			ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5673 			sguid = ppvd->vdev_child[1]->vdev_guid;
5674 		}
5675 		ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5676 
5677 		spa_config_exit(spa, SCL_ALL, FTAG);
5678 		if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5679 			return;
5680 		if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5681 			return;
5682 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5683 	}
5684 
5685 	spa_config_exit(spa, SCL_ALL, FTAG);
5686 }
5687 
5688 /*
5689  * Update the stored path or FRU for this vdev.
5690  */
5691 int
5692 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5693     boolean_t ispath)
5694 {
5695 	vdev_t *vd;
5696 	boolean_t sync = B_FALSE;
5697 
5698 	ASSERT(spa_writeable(spa));
5699 
5700 	spa_vdev_state_enter(spa, SCL_ALL);
5701 
5702 	if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5703 		return (spa_vdev_state_exit(spa, NULL, ENOENT));
5704 
5705 	if (!vd->vdev_ops->vdev_op_leaf)
5706 		return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5707 
5708 	if (ispath) {
5709 		if (strcmp(value, vd->vdev_path) != 0) {
5710 			spa_strfree(vd->vdev_path);
5711 			vd->vdev_path = spa_strdup(value);
5712 			sync = B_TRUE;
5713 		}
5714 	} else {
5715 		if (vd->vdev_fru == NULL) {
5716 			vd->vdev_fru = spa_strdup(value);
5717 			sync = B_TRUE;
5718 		} else if (strcmp(value, vd->vdev_fru) != 0) {
5719 			spa_strfree(vd->vdev_fru);
5720 			vd->vdev_fru = spa_strdup(value);
5721 			sync = B_TRUE;
5722 		}
5723 	}
5724 
5725 	return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5726 }
5727 
5728 int
5729 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5730 {
5731 	return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5732 }
5733 
5734 int
5735 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5736 {
5737 	return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5738 }
5739 
5740 /*
5741  * ==========================================================================
5742  * SPA Scanning
5743  * ==========================================================================
5744  */
5745 
5746 int
5747 spa_scan_stop(spa_t *spa)
5748 {
5749 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5750 	if (dsl_scan_resilvering(spa->spa_dsl_pool))
5751 		return (SET_ERROR(EBUSY));
5752 	return (dsl_scan_cancel(spa->spa_dsl_pool));
5753 }
5754 
5755 int
5756 spa_scan(spa_t *spa, pool_scan_func_t func)
5757 {
5758 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5759 
5760 	if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5761 		return (SET_ERROR(ENOTSUP));
5762 
5763 	/*
5764 	 * If a resilver was requested, but there is no DTL on a
5765 	 * writeable leaf device, we have nothing to do.
5766 	 */
5767 	if (func == POOL_SCAN_RESILVER &&
5768 	    !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5769 		spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5770 		return (0);
5771 	}
5772 
5773 	return (dsl_scan(spa->spa_dsl_pool, func));
5774 }
5775 
5776 /*
5777  * ==========================================================================
5778  * SPA async task processing
5779  * ==========================================================================
5780  */
5781 
5782 static void
5783 spa_async_remove(spa_t *spa, vdev_t *vd)
5784 {
5785 	if (vd->vdev_remove_wanted) {
5786 		vd->vdev_remove_wanted = B_FALSE;
5787 		vd->vdev_delayed_close = B_FALSE;
5788 		vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5789 
5790 		/*
5791 		 * We want to clear the stats, but we don't want to do a full
5792 		 * vdev_clear() as that will cause us to throw away
5793 		 * degraded/faulted state as well as attempt to reopen the
5794 		 * device, all of which is a waste.
5795 		 */
5796 		vd->vdev_stat.vs_read_errors = 0;
5797 		vd->vdev_stat.vs_write_errors = 0;
5798 		vd->vdev_stat.vs_checksum_errors = 0;
5799 
5800 		vdev_state_dirty(vd->vdev_top);
5801 	}
5802 
5803 	for (int c = 0; c < vd->vdev_children; c++)
5804 		spa_async_remove(spa, vd->vdev_child[c]);
5805 }
5806 
5807 static void
5808 spa_async_probe(spa_t *spa, vdev_t *vd)
5809 {
5810 	if (vd->vdev_probe_wanted) {
5811 		vd->vdev_probe_wanted = B_FALSE;
5812 		vdev_reopen(vd);	/* vdev_open() does the actual probe */
5813 	}
5814 
5815 	for (int c = 0; c < vd->vdev_children; c++)
5816 		spa_async_probe(spa, vd->vdev_child[c]);
5817 }
5818 
5819 static void
5820 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5821 {
5822 	sysevent_id_t eid;
5823 	nvlist_t *attr;
5824 	char *physpath;
5825 
5826 	if (!spa->spa_autoexpand)
5827 		return;
5828 
5829 	for (int c = 0; c < vd->vdev_children; c++) {
5830 		vdev_t *cvd = vd->vdev_child[c];
5831 		spa_async_autoexpand(spa, cvd);
5832 	}
5833 
5834 	if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5835 		return;
5836 
5837 	physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5838 	(void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5839 
5840 	VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5841 	VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5842 
5843 	(void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5844 	    ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5845 
5846 	nvlist_free(attr);
5847 	kmem_free(physpath, MAXPATHLEN);
5848 }
5849 
5850 static void
5851 spa_async_thread(spa_t *spa)
5852 {
5853 	int tasks;
5854 
5855 	ASSERT(spa->spa_sync_on);
5856 
5857 	mutex_enter(&spa->spa_async_lock);
5858 	tasks = spa->spa_async_tasks;
5859 	spa->spa_async_tasks = 0;
5860 	mutex_exit(&spa->spa_async_lock);
5861 
5862 	/*
5863 	 * See if the config needs to be updated.
5864 	 */
5865 	if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5866 		uint64_t old_space, new_space;
5867 
5868 		mutex_enter(&spa_namespace_lock);
5869 		old_space = metaslab_class_get_space(spa_normal_class(spa));
5870 		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5871 		new_space = metaslab_class_get_space(spa_normal_class(spa));
5872 		mutex_exit(&spa_namespace_lock);
5873 
5874 		/*
5875 		 * If the pool grew as a result of the config update,
5876 		 * then log an internal history event.
5877 		 */
5878 		if (new_space != old_space) {
5879 			spa_history_log_internal(spa, "vdev online", NULL,
5880 			    "pool '%s' size: %llu(+%llu)",
5881 			    spa_name(spa), new_space, new_space - old_space);
5882 		}
5883 	}
5884 
5885 	/*
5886 	 * See if any devices need to be marked REMOVED.
5887 	 */
5888 	if (tasks & SPA_ASYNC_REMOVE) {
5889 		spa_vdev_state_enter(spa, SCL_NONE);
5890 		spa_async_remove(spa, spa->spa_root_vdev);
5891 		for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5892 			spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5893 		for (int i = 0; i < spa->spa_spares.sav_count; i++)
5894 			spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5895 		(void) spa_vdev_state_exit(spa, NULL, 0);
5896 	}
5897 
5898 	if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5899 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5900 		spa_async_autoexpand(spa, spa->spa_root_vdev);
5901 		spa_config_exit(spa, SCL_CONFIG, FTAG);
5902 	}
5903 
5904 	/*
5905 	 * See if any devices need to be probed.
5906 	 */
5907 	if (tasks & SPA_ASYNC_PROBE) {
5908 		spa_vdev_state_enter(spa, SCL_NONE);
5909 		spa_async_probe(spa, spa->spa_root_vdev);
5910 		(void) spa_vdev_state_exit(spa, NULL, 0);
5911 	}
5912 
5913 	/*
5914 	 * If any devices are done replacing, detach them.
5915 	 */
5916 	if (tasks & SPA_ASYNC_RESILVER_DONE)
5917 		spa_vdev_resilver_done(spa);
5918 
5919 	/*
5920 	 * Kick off a resilver.
5921 	 */
5922 	if (tasks & SPA_ASYNC_RESILVER)
5923 		dsl_resilver_restart(spa->spa_dsl_pool, 0);
5924 
5925 	/*
5926 	 * Let the world know that we're done.
5927 	 */
5928 	mutex_enter(&spa->spa_async_lock);
5929 	spa->spa_async_thread = NULL;
5930 	cv_broadcast(&spa->spa_async_cv);
5931 	mutex_exit(&spa->spa_async_lock);
5932 	thread_exit();
5933 }
5934 
5935 void
5936 spa_async_suspend(spa_t *spa)
5937 {
5938 	mutex_enter(&spa->spa_async_lock);
5939 	spa->spa_async_suspended++;
5940 	while (spa->spa_async_thread != NULL)
5941 		cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5942 	mutex_exit(&spa->spa_async_lock);
5943 }
5944 
5945 void
5946 spa_async_resume(spa_t *spa)
5947 {
5948 	mutex_enter(&spa->spa_async_lock);
5949 	ASSERT(spa->spa_async_suspended != 0);
5950 	spa->spa_async_suspended--;
5951 	mutex_exit(&spa->spa_async_lock);
5952 }
5953 
5954 static boolean_t
5955 spa_async_tasks_pending(spa_t *spa)
5956 {
5957 	uint_t non_config_tasks;
5958 	uint_t config_task;
5959 	boolean_t config_task_suspended;
5960 
5961 	non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5962 	config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5963 	if (spa->spa_ccw_fail_time == 0) {
5964 		config_task_suspended = B_FALSE;
5965 	} else {
5966 		config_task_suspended =
5967 		    (gethrtime() - spa->spa_ccw_fail_time) <
5968 		    (zfs_ccw_retry_interval * NANOSEC);
5969 	}
5970 
5971 	return (non_config_tasks || (config_task && !config_task_suspended));
5972 }
5973 
5974 static void
5975 spa_async_dispatch(spa_t *spa)
5976 {
5977 	mutex_enter(&spa->spa_async_lock);
5978 	if (spa_async_tasks_pending(spa) &&
5979 	    !spa->spa_async_suspended &&
5980 	    spa->spa_async_thread == NULL &&
5981 	    rootdir != NULL)
5982 		spa->spa_async_thread = thread_create(NULL, 0,
5983 		    spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5984 	mutex_exit(&spa->spa_async_lock);
5985 }
5986 
5987 void
5988 spa_async_request(spa_t *spa, int task)
5989 {
5990 	zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5991 	mutex_enter(&spa->spa_async_lock);
5992 	spa->spa_async_tasks |= task;
5993 	mutex_exit(&spa->spa_async_lock);
5994 }
5995 
5996 /*
5997  * ==========================================================================
5998  * SPA syncing routines
5999  * ==========================================================================
6000  */
6001 
6002 static int
6003 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6004 {
6005 	bpobj_t *bpo = arg;
6006 	bpobj_enqueue(bpo, bp, tx);
6007 	return (0);
6008 }
6009 
6010 static int
6011 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
6012 {
6013 	zio_t *zio = arg;
6014 
6015 	zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
6016 	    zio->io_flags));
6017 	return (0);
6018 }
6019 
6020 /*
6021  * Note: this simple function is not inlined to make it easier to dtrace the
6022  * amount of time spent syncing frees.
6023  */
6024 static void
6025 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
6026 {
6027 	zio_t *zio = zio_root(spa, NULL, NULL, 0);
6028 	bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
6029 	VERIFY(zio_wait(zio) == 0);
6030 }
6031 
6032 /*
6033  * Note: this simple function is not inlined to make it easier to dtrace the
6034  * amount of time spent syncing deferred frees.
6035  */
6036 static void
6037 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
6038 {
6039 	zio_t *zio = zio_root(spa, NULL, NULL, 0);
6040 	VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
6041 	    spa_free_sync_cb, zio, tx), ==, 0);
6042 	VERIFY0(zio_wait(zio));
6043 }
6044 
6045 
6046 static void
6047 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
6048 {
6049 	char *packed = NULL;
6050 	size_t bufsize;
6051 	size_t nvsize = 0;
6052 	dmu_buf_t *db;
6053 
6054 	VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
6055 
6056 	/*
6057 	 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
6058 	 * information.  This avoids the dmu_buf_will_dirty() path and
6059 	 * saves us a pre-read to get data we don't actually care about.
6060 	 */
6061 	bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
6062 	packed = kmem_alloc(bufsize, KM_SLEEP);
6063 
6064 	VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
6065 	    KM_SLEEP) == 0);
6066 	bzero(packed + nvsize, bufsize - nvsize);
6067 
6068 	dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
6069 
6070 	kmem_free(packed, bufsize);
6071 
6072 	VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
6073 	dmu_buf_will_dirty(db, tx);
6074 	*(uint64_t *)db->db_data = nvsize;
6075 	dmu_buf_rele(db, FTAG);
6076 }
6077 
6078 static void
6079 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
6080     const char *config, const char *entry)
6081 {
6082 	nvlist_t *nvroot;
6083 	nvlist_t **list;
6084 	int i;
6085 
6086 	if (!sav->sav_sync)
6087 		return;
6088 
6089 	/*
6090 	 * Update the MOS nvlist describing the list of available devices.
6091 	 * spa_validate_aux() will have already made sure this nvlist is
6092 	 * valid and the vdevs are labeled appropriately.
6093 	 */
6094 	if (sav->sav_object == 0) {
6095 		sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
6096 		    DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
6097 		    sizeof (uint64_t), tx);
6098 		VERIFY(zap_update(spa->spa_meta_objset,
6099 		    DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
6100 		    &sav->sav_object, tx) == 0);
6101 	}
6102 
6103 	VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6104 	if (sav->sav_count == 0) {
6105 		VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
6106 	} else {
6107 		list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
6108 		for (i = 0; i < sav->sav_count; i++)
6109 			list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
6110 			    B_FALSE, VDEV_CONFIG_L2CACHE);
6111 		VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
6112 		    sav->sav_count) == 0);
6113 		for (i = 0; i < sav->sav_count; i++)
6114 			nvlist_free(list[i]);
6115 		kmem_free(list, sav->sav_count * sizeof (void *));
6116 	}
6117 
6118 	spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
6119 	nvlist_free(nvroot);
6120 
6121 	sav->sav_sync = B_FALSE;
6122 }
6123 
6124 /*
6125  * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
6126  * The all-vdev ZAP must be empty.
6127  */
6128 static void
6129 spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx)
6130 {
6131 	spa_t *spa = vd->vdev_spa;
6132 	if (vd->vdev_top_zap != 0) {
6133 		VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
6134 		    vd->vdev_top_zap, tx));
6135 	}
6136 	if (vd->vdev_leaf_zap != 0) {
6137 		VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
6138 		    vd->vdev_leaf_zap, tx));
6139 	}
6140 	for (uint64_t i = 0; i < vd->vdev_children; i++) {
6141 		spa_avz_build(vd->vdev_child[i], avz, tx);
6142 	}
6143 }
6144 
6145 static void
6146 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
6147 {
6148 	nvlist_t *config;
6149 
6150 	/*
6151 	 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
6152 	 * its config may not be dirty but we still need to build per-vdev ZAPs.
6153 	 * Similarly, if the pool is being assembled (e.g. after a split), we
6154 	 * need to rebuild the AVZ although the config may not be dirty.
6155 	 */
6156 	if (list_is_empty(&spa->spa_config_dirty_list) &&
6157 	    spa->spa_avz_action == AVZ_ACTION_NONE)
6158 		return;
6159 
6160 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6161 
6162 	ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE ||
6163 	    spa->spa_avz_action == AVZ_ACTION_INITIALIZE ||
6164 	    spa->spa_all_vdev_zaps != 0);
6165 
6166 	if (spa->spa_avz_action == AVZ_ACTION_REBUILD) {
6167 		/* Make and build the new AVZ */
6168 		uint64_t new_avz = zap_create(spa->spa_meta_objset,
6169 		    DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx);
6170 		spa_avz_build(spa->spa_root_vdev, new_avz, tx);
6171 
6172 		/* Diff old AVZ with new one */
6173 		zap_cursor_t zc;
6174 		zap_attribute_t za;
6175 
6176 		for (zap_cursor_init(&zc, spa->spa_meta_objset,
6177 		    spa->spa_all_vdev_zaps);
6178 		    zap_cursor_retrieve(&zc, &za) == 0;
6179 		    zap_cursor_advance(&zc)) {
6180 			uint64_t vdzap = za.za_first_integer;
6181 			if (zap_lookup_int(spa->spa_meta_objset, new_avz,
6182 			    vdzap) == ENOENT) {
6183 				/*
6184 				 * ZAP is listed in old AVZ but not in new one;
6185 				 * destroy it
6186 				 */
6187 				VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap,
6188 				    tx));
6189 			}
6190 		}
6191 
6192 		zap_cursor_fini(&zc);
6193 
6194 		/* Destroy the old AVZ */
6195 		VERIFY0(zap_destroy(spa->spa_meta_objset,
6196 		    spa->spa_all_vdev_zaps, tx));
6197 
6198 		/* Replace the old AVZ in the dir obj with the new one */
6199 		VERIFY0(zap_update(spa->spa_meta_objset,
6200 		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP,
6201 		    sizeof (new_avz), 1, &new_avz, tx));
6202 
6203 		spa->spa_all_vdev_zaps = new_avz;
6204 	} else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) {
6205 		zap_cursor_t zc;
6206 		zap_attribute_t za;
6207 
6208 		/* Walk through the AVZ and destroy all listed ZAPs */
6209 		for (zap_cursor_init(&zc, spa->spa_meta_objset,
6210 		    spa->spa_all_vdev_zaps);
6211 		    zap_cursor_retrieve(&zc, &za) == 0;
6212 		    zap_cursor_advance(&zc)) {
6213 			uint64_t zap = za.za_first_integer;
6214 			VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx));
6215 		}
6216 
6217 		zap_cursor_fini(&zc);
6218 
6219 		/* Destroy and unlink the AVZ itself */
6220 		VERIFY0(zap_destroy(spa->spa_meta_objset,
6221 		    spa->spa_all_vdev_zaps, tx));
6222 		VERIFY0(zap_remove(spa->spa_meta_objset,
6223 		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx));
6224 		spa->spa_all_vdev_zaps = 0;
6225 	}
6226 
6227 	if (spa->spa_all_vdev_zaps == 0) {
6228 		spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset,
6229 		    DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT,
6230 		    DMU_POOL_VDEV_ZAP_MAP, tx);
6231 	}
6232 	spa->spa_avz_action = AVZ_ACTION_NONE;
6233 
6234 	/* Create ZAPs for vdevs that don't have them. */
6235 	vdev_construct_zaps(spa->spa_root_vdev, tx);
6236 
6237 	config = spa_config_generate(spa, spa->spa_root_vdev,
6238 	    dmu_tx_get_txg(tx), B_FALSE);
6239 
6240 	/*
6241 	 * If we're upgrading the spa version then make sure that
6242 	 * the config object gets updated with the correct version.
6243 	 */
6244 	if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
6245 		fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6246 		    spa->spa_uberblock.ub_version);
6247 
6248 	spa_config_exit(spa, SCL_STATE, FTAG);
6249 
6250 	nvlist_free(spa->spa_config_syncing);
6251 	spa->spa_config_syncing = config;
6252 
6253 	spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
6254 }
6255 
6256 static void
6257 spa_sync_version(void *arg, dmu_tx_t *tx)
6258 {
6259 	uint64_t *versionp = arg;
6260 	uint64_t version = *versionp;
6261 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6262 
6263 	/*
6264 	 * Setting the version is special cased when first creating the pool.
6265 	 */
6266 	ASSERT(tx->tx_txg != TXG_INITIAL);
6267 
6268 	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
6269 	ASSERT(version >= spa_version(spa));
6270 
6271 	spa->spa_uberblock.ub_version = version;
6272 	vdev_config_dirty(spa->spa_root_vdev);
6273 	spa_history_log_internal(spa, "set", tx, "version=%lld", version);
6274 }
6275 
6276 /*
6277  * Set zpool properties.
6278  */
6279 static void
6280 spa_sync_props(void *arg, dmu_tx_t *tx)
6281 {
6282 	nvlist_t *nvp = arg;
6283 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
6284 	objset_t *mos = spa->spa_meta_objset;
6285 	nvpair_t *elem = NULL;
6286 
6287 	mutex_enter(&spa->spa_props_lock);
6288 
6289 	while ((elem = nvlist_next_nvpair(nvp, elem))) {
6290 		uint64_t intval;
6291 		char *strval, *fname;
6292 		zpool_prop_t prop;
6293 		const char *propname;
6294 		zprop_type_t proptype;
6295 		spa_feature_t fid;
6296 
6297 		switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
6298 		case ZPROP_INVAL:
6299 			/*
6300 			 * We checked this earlier in spa_prop_validate().
6301 			 */
6302 			ASSERT(zpool_prop_feature(nvpair_name(elem)));
6303 
6304 			fname = strchr(nvpair_name(elem), '@') + 1;
6305 			VERIFY0(zfeature_lookup_name(fname, &fid));
6306 
6307 			spa_feature_enable(spa, fid, tx);
6308 			spa_history_log_internal(spa, "set", tx,
6309 			    "%s=enabled", nvpair_name(elem));
6310 			break;
6311 
6312 		case ZPOOL_PROP_VERSION:
6313 			intval = fnvpair_value_uint64(elem);
6314 			/*
6315 			 * The version is synced seperatly before other
6316 			 * properties and should be correct by now.
6317 			 */
6318 			ASSERT3U(spa_version(spa), >=, intval);
6319 			break;
6320 
6321 		case ZPOOL_PROP_ALTROOT:
6322 			/*
6323 			 * 'altroot' is a non-persistent property. It should
6324 			 * have been set temporarily at creation or import time.
6325 			 */
6326 			ASSERT(spa->spa_root != NULL);
6327 			break;
6328 
6329 		case ZPOOL_PROP_READONLY:
6330 		case ZPOOL_PROP_CACHEFILE:
6331 			/*
6332 			 * 'readonly' and 'cachefile' are also non-persisitent
6333 			 * properties.
6334 			 */
6335 			break;
6336 		case ZPOOL_PROP_COMMENT:
6337 			strval = fnvpair_value_string(elem);
6338 			if (spa->spa_comment != NULL)
6339 				spa_strfree(spa->spa_comment);
6340 			spa->spa_comment = spa_strdup(strval);
6341 			/*
6342 			 * We need to dirty the configuration on all the vdevs
6343 			 * so that their labels get updated.  It's unnecessary
6344 			 * to do this for pool creation since the vdev's
6345 			 * configuratoin has already been dirtied.
6346 			 */
6347 			if (tx->tx_txg != TXG_INITIAL)
6348 				vdev_config_dirty(spa->spa_root_vdev);
6349 			spa_history_log_internal(spa, "set", tx,
6350 			    "%s=%s", nvpair_name(elem), strval);
6351 			break;
6352 		default:
6353 			/*
6354 			 * Set pool property values in the poolprops mos object.
6355 			 */
6356 			if (spa->spa_pool_props_object == 0) {
6357 				spa->spa_pool_props_object =
6358 				    zap_create_link(mos, DMU_OT_POOL_PROPS,
6359 				    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
6360 				    tx);
6361 			}
6362 
6363 			/* normalize the property name */
6364 			propname = zpool_prop_to_name(prop);
6365 			proptype = zpool_prop_get_type(prop);
6366 
6367 			if (nvpair_type(elem) == DATA_TYPE_STRING) {
6368 				ASSERT(proptype == PROP_TYPE_STRING);
6369 				strval = fnvpair_value_string(elem);
6370 				VERIFY0(zap_update(mos,
6371 				    spa->spa_pool_props_object, propname,
6372 				    1, strlen(strval) + 1, strval, tx));
6373 				spa_history_log_internal(spa, "set", tx,
6374 				    "%s=%s", nvpair_name(elem), strval);
6375 			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
6376 				intval = fnvpair_value_uint64(elem);
6377 
6378 				if (proptype == PROP_TYPE_INDEX) {
6379 					const char *unused;
6380 					VERIFY0(zpool_prop_index_to_string(
6381 					    prop, intval, &unused));
6382 				}
6383 				VERIFY0(zap_update(mos,
6384 				    spa->spa_pool_props_object, propname,
6385 				    8, 1, &intval, tx));
6386 				spa_history_log_internal(spa, "set", tx,
6387 				    "%s=%lld", nvpair_name(elem), intval);
6388 			} else {
6389 				ASSERT(0); /* not allowed */
6390 			}
6391 
6392 			switch (prop) {
6393 			case ZPOOL_PROP_DELEGATION:
6394 				spa->spa_delegation = intval;
6395 				break;
6396 			case ZPOOL_PROP_BOOTFS:
6397 				spa->spa_bootfs = intval;
6398 				break;
6399 			case ZPOOL_PROP_FAILUREMODE:
6400 				spa->spa_failmode = intval;
6401 				break;
6402 			case ZPOOL_PROP_AUTOEXPAND:
6403 				spa->spa_autoexpand = intval;
6404 				if (tx->tx_txg != TXG_INITIAL)
6405 					spa_async_request(spa,
6406 					    SPA_ASYNC_AUTOEXPAND);
6407 				break;
6408 			case ZPOOL_PROP_DEDUPDITTO:
6409 				spa->spa_dedup_ditto = intval;
6410 				break;
6411 			default:
6412 				break;
6413 			}
6414 		}
6415 
6416 	}
6417 
6418 	mutex_exit(&spa->spa_props_lock);
6419 }
6420 
6421 /*
6422  * Perform one-time upgrade on-disk changes.  spa_version() does not
6423  * reflect the new version this txg, so there must be no changes this
6424  * txg to anything that the upgrade code depends on after it executes.
6425  * Therefore this must be called after dsl_pool_sync() does the sync
6426  * tasks.
6427  */
6428 static void
6429 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6430 {
6431 	dsl_pool_t *dp = spa->spa_dsl_pool;
6432 
6433 	ASSERT(spa->spa_sync_pass == 1);
6434 
6435 	rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6436 
6437 	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6438 	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6439 		dsl_pool_create_origin(dp, tx);
6440 
6441 		/* Keeping the origin open increases spa_minref */
6442 		spa->spa_minref += 3;
6443 	}
6444 
6445 	if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6446 	    spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6447 		dsl_pool_upgrade_clones(dp, tx);
6448 	}
6449 
6450 	if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6451 	    spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6452 		dsl_pool_upgrade_dir_clones(dp, tx);
6453 
6454 		/* Keeping the freedir open increases spa_minref */
6455 		spa->spa_minref += 3;
6456 	}
6457 
6458 	if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6459 	    spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6460 		spa_feature_create_zap_objects(spa, tx);
6461 	}
6462 
6463 	/*
6464 	 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
6465 	 * when possibility to use lz4 compression for metadata was added
6466 	 * Old pools that have this feature enabled must be upgraded to have
6467 	 * this feature active
6468 	 */
6469 	if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6470 		boolean_t lz4_en = spa_feature_is_enabled(spa,
6471 		    SPA_FEATURE_LZ4_COMPRESS);
6472 		boolean_t lz4_ac = spa_feature_is_active(spa,
6473 		    SPA_FEATURE_LZ4_COMPRESS);
6474 
6475 		if (lz4_en && !lz4_ac)
6476 			spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
6477 	}
6478 
6479 	/*
6480 	 * If we haven't written the salt, do so now.  Note that the
6481 	 * feature may not be activated yet, but that's fine since
6482 	 * the presence of this ZAP entry is backwards compatible.
6483 	 */
6484 	if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
6485 	    DMU_POOL_CHECKSUM_SALT) == ENOENT) {
6486 		VERIFY0(zap_add(spa->spa_meta_objset,
6487 		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1,
6488 		    sizeof (spa->spa_cksum_salt.zcs_bytes),
6489 		    spa->spa_cksum_salt.zcs_bytes, tx));
6490 	}
6491 
6492 	rrw_exit(&dp->dp_config_rwlock, FTAG);
6493 }
6494 
6495 /*
6496  * Sync the specified transaction group.  New blocks may be dirtied as
6497  * part of the process, so we iterate until it converges.
6498  */
6499 void
6500 spa_sync(spa_t *spa, uint64_t txg)
6501 {
6502 	dsl_pool_t *dp = spa->spa_dsl_pool;
6503 	objset_t *mos = spa->spa_meta_objset;
6504 	bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6505 	vdev_t *rvd = spa->spa_root_vdev;
6506 	vdev_t *vd;
6507 	dmu_tx_t *tx;
6508 	int error;
6509 	uint32_t max_queue_depth = zfs_vdev_async_write_max_active *
6510 	    zfs_vdev_queue_depth_pct / 100;
6511 
6512 	VERIFY(spa_writeable(spa));
6513 
6514 	/*
6515 	 * Lock out configuration changes.
6516 	 */
6517 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6518 
6519 	spa->spa_syncing_txg = txg;
6520 	spa->spa_sync_pass = 0;
6521 
6522 	mutex_enter(&spa->spa_alloc_lock);
6523 	VERIFY0(avl_numnodes(&spa->spa_alloc_tree));
6524 	mutex_exit(&spa->spa_alloc_lock);
6525 
6526 	/*
6527 	 * If there are any pending vdev state changes, convert them
6528 	 * into config changes that go out with this transaction group.
6529 	 */
6530 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6531 	while (list_head(&spa->spa_state_dirty_list) != NULL) {
6532 		/*
6533 		 * We need the write lock here because, for aux vdevs,
6534 		 * calling vdev_config_dirty() modifies sav_config.
6535 		 * This is ugly and will become unnecessary when we
6536 		 * eliminate the aux vdev wart by integrating all vdevs
6537 		 * into the root vdev tree.
6538 		 */
6539 		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6540 		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6541 		while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6542 			vdev_state_clean(vd);
6543 			vdev_config_dirty(vd);
6544 		}
6545 		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6546 		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6547 	}
6548 	spa_config_exit(spa, SCL_STATE, FTAG);
6549 
6550 	tx = dmu_tx_create_assigned(dp, txg);
6551 
6552 	spa->spa_sync_starttime = gethrtime();
6553 	VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6554 	    spa->spa_sync_starttime + spa->spa_deadman_synctime));
6555 
6556 	/*
6557 	 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6558 	 * set spa_deflate if we have no raid-z vdevs.
6559 	 */
6560 	if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6561 	    spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6562 		int i;
6563 
6564 		for (i = 0; i < rvd->vdev_children; i++) {
6565 			vd = rvd->vdev_child[i];
6566 			if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6567 				break;
6568 		}
6569 		if (i == rvd->vdev_children) {
6570 			spa->spa_deflate = TRUE;
6571 			VERIFY(0 == zap_add(spa->spa_meta_objset,
6572 			    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6573 			    sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6574 		}
6575 	}
6576 
6577 	/*
6578 	 * Set the top-level vdev's max queue depth. Evaluate each
6579 	 * top-level's async write queue depth in case it changed.
6580 	 * The max queue depth will not change in the middle of syncing
6581 	 * out this txg.
6582 	 */
6583 	uint64_t queue_depth_total = 0;
6584 	for (int c = 0; c < rvd->vdev_children; c++) {
6585 		vdev_t *tvd = rvd->vdev_child[c];
6586 		metaslab_group_t *mg = tvd->vdev_mg;
6587 
6588 		if (mg == NULL || mg->mg_class != spa_normal_class(spa) ||
6589 		    !metaslab_group_initialized(mg))
6590 			continue;
6591 
6592 		/*
6593 		 * It is safe to do a lock-free check here because only async
6594 		 * allocations look at mg_max_alloc_queue_depth, and async
6595 		 * allocations all happen from spa_sync().
6596 		 */
6597 		ASSERT0(refcount_count(&mg->mg_alloc_queue_depth));
6598 		mg->mg_max_alloc_queue_depth = max_queue_depth;
6599 		queue_depth_total += mg->mg_max_alloc_queue_depth;
6600 	}
6601 	metaslab_class_t *mc = spa_normal_class(spa);
6602 	ASSERT0(refcount_count(&mc->mc_alloc_slots));
6603 	mc->mc_alloc_max_slots = queue_depth_total;
6604 	mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
6605 
6606 	ASSERT3U(mc->mc_alloc_max_slots, <=,
6607 	    max_queue_depth * rvd->vdev_children);
6608 
6609 	/*
6610 	 * Iterate to convergence.
6611 	 */
6612 	do {
6613 		int pass = ++spa->spa_sync_pass;
6614 
6615 		spa_sync_config_object(spa, tx);
6616 		spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6617 		    ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6618 		spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6619 		    ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6620 		spa_errlog_sync(spa, txg);
6621 		dsl_pool_sync(dp, txg);
6622 
6623 		if (pass < zfs_sync_pass_deferred_free) {
6624 			spa_sync_frees(spa, free_bpl, tx);
6625 		} else {
6626 			/*
6627 			 * We can not defer frees in pass 1, because
6628 			 * we sync the deferred frees later in pass 1.
6629 			 */
6630 			ASSERT3U(pass, >, 1);
6631 			bplist_iterate(free_bpl, bpobj_enqueue_cb,
6632 			    &spa->spa_deferred_bpobj, tx);
6633 		}
6634 
6635 		ddt_sync(spa, txg);
6636 		dsl_scan_sync(dp, tx);
6637 
6638 		while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6639 			vdev_sync(vd, txg);
6640 
6641 		if (pass == 1) {
6642 			spa_sync_upgrades(spa, tx);
6643 			ASSERT3U(txg, >=,
6644 			    spa->spa_uberblock.ub_rootbp.blk_birth);
6645 			/*
6646 			 * Note: We need to check if the MOS is dirty
6647 			 * because we could have marked the MOS dirty
6648 			 * without updating the uberblock (e.g. if we
6649 			 * have sync tasks but no dirty user data).  We
6650 			 * need to check the uberblock's rootbp because
6651 			 * it is updated if we have synced out dirty
6652 			 * data (though in this case the MOS will most
6653 			 * likely also be dirty due to second order
6654 			 * effects, we don't want to rely on that here).
6655 			 */
6656 			if (spa->spa_uberblock.ub_rootbp.blk_birth < txg &&
6657 			    !dmu_objset_is_dirty(mos, txg)) {
6658 				/*
6659 				 * Nothing changed on the first pass,
6660 				 * therefore this TXG is a no-op.  Avoid
6661 				 * syncing deferred frees, so that we
6662 				 * can keep this TXG as a no-op.
6663 				 */
6664 				ASSERT(txg_list_empty(&dp->dp_dirty_datasets,
6665 				    txg));
6666 				ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6667 				ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg));
6668 				break;
6669 			}
6670 			spa_sync_deferred_frees(spa, tx);
6671 		}
6672 
6673 	} while (dmu_objset_is_dirty(mos, txg));
6674 
6675 	if (!list_is_empty(&spa->spa_config_dirty_list)) {
6676 		/*
6677 		 * Make sure that the number of ZAPs for all the vdevs matches
6678 		 * the number of ZAPs in the per-vdev ZAP list. This only gets
6679 		 * called if the config is dirty; otherwise there may be
6680 		 * outstanding AVZ operations that weren't completed in
6681 		 * spa_sync_config_object.
6682 		 */
6683 		uint64_t all_vdev_zap_entry_count;
6684 		ASSERT0(zap_count(spa->spa_meta_objset,
6685 		    spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count));
6686 		ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==,
6687 		    all_vdev_zap_entry_count);
6688 	}
6689 
6690 	/*
6691 	 * Rewrite the vdev configuration (which includes the uberblock)
6692 	 * to commit the transaction group.
6693 	 *
6694 	 * If there are no dirty vdevs, we sync the uberblock to a few
6695 	 * random top-level vdevs that are known to be visible in the
6696 	 * config cache (see spa_vdev_add() for a complete description).
6697 	 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6698 	 */
6699 	for (;;) {
6700 		/*
6701 		 * We hold SCL_STATE to prevent vdev open/close/etc.
6702 		 * while we're attempting to write the vdev labels.
6703 		 */
6704 		spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6705 
6706 		if (list_is_empty(&spa->spa_config_dirty_list)) {
6707 			vdev_t *svd[SPA_DVAS_PER_BP];
6708 			int svdcount = 0;
6709 			int children = rvd->vdev_children;
6710 			int c0 = spa_get_random(children);
6711 
6712 			for (int c = 0; c < children; c++) {
6713 				vd = rvd->vdev_child[(c0 + c) % children];
6714 				if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6715 					continue;
6716 				svd[svdcount++] = vd;
6717 				if (svdcount == SPA_DVAS_PER_BP)
6718 					break;
6719 			}
6720 			error = vdev_config_sync(svd, svdcount, txg);
6721 		} else {
6722 			error = vdev_config_sync(rvd->vdev_child,
6723 			    rvd->vdev_children, txg);
6724 		}
6725 
6726 		if (error == 0)
6727 			spa->spa_last_synced_guid = rvd->vdev_guid;
6728 
6729 		spa_config_exit(spa, SCL_STATE, FTAG);
6730 
6731 		if (error == 0)
6732 			break;
6733 		zio_suspend(spa, NULL);
6734 		zio_resume_wait(spa);
6735 	}
6736 	dmu_tx_commit(tx);
6737 
6738 	VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6739 
6740 	/*
6741 	 * Clear the dirty config list.
6742 	 */
6743 	while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6744 		vdev_config_clean(vd);
6745 
6746 	/*
6747 	 * Now that the new config has synced transactionally,
6748 	 * let it become visible to the config cache.
6749 	 */
6750 	if (spa->spa_config_syncing != NULL) {
6751 		spa_config_set(spa, spa->spa_config_syncing);
6752 		spa->spa_config_txg = txg;
6753 		spa->spa_config_syncing = NULL;
6754 	}
6755 
6756 	dsl_pool_sync_done(dp, txg);
6757 
6758 	mutex_enter(&spa->spa_alloc_lock);
6759 	VERIFY0(avl_numnodes(&spa->spa_alloc_tree));
6760 	mutex_exit(&spa->spa_alloc_lock);
6761 
6762 	/*
6763 	 * Update usable space statistics.
6764 	 */
6765 	while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6766 		vdev_sync_done(vd, txg);
6767 
6768 	spa_update_dspace(spa);
6769 
6770 	/*
6771 	 * It had better be the case that we didn't dirty anything
6772 	 * since vdev_config_sync().
6773 	 */
6774 	ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6775 	ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6776 	ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6777 
6778 	spa->spa_sync_pass = 0;
6779 
6780 	/*
6781 	 * Update the last synced uberblock here. We want to do this at
6782 	 * the end of spa_sync() so that consumers of spa_last_synced_txg()
6783 	 * will be guaranteed that all the processing associated with
6784 	 * that txg has been completed.
6785 	 */
6786 	spa->spa_ubsync = spa->spa_uberblock;
6787 	spa_config_exit(spa, SCL_CONFIG, FTAG);
6788 
6789 	spa_handle_ignored_writes(spa);
6790 
6791 	/*
6792 	 * If any async tasks have been requested, kick them off.
6793 	 */
6794 	spa_async_dispatch(spa);
6795 }
6796 
6797 /*
6798  * Sync all pools.  We don't want to hold the namespace lock across these
6799  * operations, so we take a reference on the spa_t and drop the lock during the
6800  * sync.
6801  */
6802 void
6803 spa_sync_allpools(void)
6804 {
6805 	spa_t *spa = NULL;
6806 	mutex_enter(&spa_namespace_lock);
6807 	while ((spa = spa_next(spa)) != NULL) {
6808 		if (spa_state(spa) != POOL_STATE_ACTIVE ||
6809 		    !spa_writeable(spa) || spa_suspended(spa))
6810 			continue;
6811 		spa_open_ref(spa, FTAG);
6812 		mutex_exit(&spa_namespace_lock);
6813 		txg_wait_synced(spa_get_dsl(spa), 0);
6814 		mutex_enter(&spa_namespace_lock);
6815 		spa_close(spa, FTAG);
6816 	}
6817 	mutex_exit(&spa_namespace_lock);
6818 }
6819 
6820 /*
6821  * ==========================================================================
6822  * Miscellaneous routines
6823  * ==========================================================================
6824  */
6825 
6826 /*
6827  * Remove all pools in the system.
6828  */
6829 void
6830 spa_evict_all(void)
6831 {
6832 	spa_t *spa;
6833 
6834 	/*
6835 	 * Remove all cached state.  All pools should be closed now,
6836 	 * so every spa in the AVL tree should be unreferenced.
6837 	 */
6838 	mutex_enter(&spa_namespace_lock);
6839 	while ((spa = spa_next(NULL)) != NULL) {
6840 		/*
6841 		 * Stop async tasks.  The async thread may need to detach
6842 		 * a device that's been replaced, which requires grabbing
6843 		 * spa_namespace_lock, so we must drop it here.
6844 		 */
6845 		spa_open_ref(spa, FTAG);
6846 		mutex_exit(&spa_namespace_lock);
6847 		spa_async_suspend(spa);
6848 		mutex_enter(&spa_namespace_lock);
6849 		spa_close(spa, FTAG);
6850 
6851 		if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6852 			spa_unload(spa);
6853 			spa_deactivate(spa);
6854 		}
6855 		spa_remove(spa);
6856 	}
6857 	mutex_exit(&spa_namespace_lock);
6858 }
6859 
6860 vdev_t *
6861 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6862 {
6863 	vdev_t *vd;
6864 	int i;
6865 
6866 	if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6867 		return (vd);
6868 
6869 	if (aux) {
6870 		for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6871 			vd = spa->spa_l2cache.sav_vdevs[i];
6872 			if (vd->vdev_guid == guid)
6873 				return (vd);
6874 		}
6875 
6876 		for (i = 0; i < spa->spa_spares.sav_count; i++) {
6877 			vd = spa->spa_spares.sav_vdevs[i];
6878 			if (vd->vdev_guid == guid)
6879 				return (vd);
6880 		}
6881 	}
6882 
6883 	return (NULL);
6884 }
6885 
6886 void
6887 spa_upgrade(spa_t *spa, uint64_t version)
6888 {
6889 	ASSERT(spa_writeable(spa));
6890 
6891 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6892 
6893 	/*
6894 	 * This should only be called for a non-faulted pool, and since a
6895 	 * future version would result in an unopenable pool, this shouldn't be
6896 	 * possible.
6897 	 */
6898 	ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6899 	ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
6900 
6901 	spa->spa_uberblock.ub_version = version;
6902 	vdev_config_dirty(spa->spa_root_vdev);
6903 
6904 	spa_config_exit(spa, SCL_ALL, FTAG);
6905 
6906 	txg_wait_synced(spa_get_dsl(spa), 0);
6907 }
6908 
6909 boolean_t
6910 spa_has_spare(spa_t *spa, uint64_t guid)
6911 {
6912 	int i;
6913 	uint64_t spareguid;
6914 	spa_aux_vdev_t *sav = &spa->spa_spares;
6915 
6916 	for (i = 0; i < sav->sav_count; i++)
6917 		if (sav->sav_vdevs[i]->vdev_guid == guid)
6918 			return (B_TRUE);
6919 
6920 	for (i = 0; i < sav->sav_npending; i++) {
6921 		if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6922 		    &spareguid) == 0 && spareguid == guid)
6923 			return (B_TRUE);
6924 	}
6925 
6926 	return (B_FALSE);
6927 }
6928 
6929 /*
6930  * Check if a pool has an active shared spare device.
6931  * Note: reference count of an active spare is 2, as a spare and as a replace
6932  */
6933 static boolean_t
6934 spa_has_active_shared_spare(spa_t *spa)
6935 {
6936 	int i, refcnt;
6937 	uint64_t pool;
6938 	spa_aux_vdev_t *sav = &spa->spa_spares;
6939 
6940 	for (i = 0; i < sav->sav_count; i++) {
6941 		if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6942 		    &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6943 		    refcnt > 2)
6944 			return (B_TRUE);
6945 	}
6946 
6947 	return (B_FALSE);
6948 }
6949 
6950 static sysevent_t *
6951 spa_event_create(spa_t *spa, vdev_t *vd, const char *name)
6952 {
6953 	sysevent_t		*ev = NULL;
6954 #ifdef _KERNEL
6955 	sysevent_attr_list_t	*attr = NULL;
6956 	sysevent_value_t	value;
6957 
6958 	ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6959 	    SE_SLEEP);
6960 	ASSERT(ev != NULL);
6961 
6962 	value.value_type = SE_DATA_TYPE_STRING;
6963 	value.value.sv_string = spa_name(spa);
6964 	if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6965 		goto done;
6966 
6967 	value.value_type = SE_DATA_TYPE_UINT64;
6968 	value.value.sv_uint64 = spa_guid(spa);
6969 	if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6970 		goto done;
6971 
6972 	if (vd) {
6973 		value.value_type = SE_DATA_TYPE_UINT64;
6974 		value.value.sv_uint64 = vd->vdev_guid;
6975 		if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6976 		    SE_SLEEP) != 0)
6977 			goto done;
6978 
6979 		if (vd->vdev_path) {
6980 			value.value_type = SE_DATA_TYPE_STRING;
6981 			value.value.sv_string = vd->vdev_path;
6982 			if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6983 			    &value, SE_SLEEP) != 0)
6984 				goto done;
6985 		}
6986 	}
6987 
6988 	if (sysevent_attach_attributes(ev, attr) != 0)
6989 		goto done;
6990 	attr = NULL;
6991 
6992 done:
6993 	if (attr)
6994 		sysevent_free_attr(attr);
6995 
6996 #endif
6997 	return (ev);
6998 }
6999 
7000 static void
7001 spa_event_post(sysevent_t *ev)
7002 {
7003 #ifdef _KERNEL
7004 	sysevent_id_t		eid;
7005 
7006 	(void) log_sysevent(ev, SE_SLEEP, &eid);
7007 	sysevent_free(ev);
7008 #endif
7009 }
7010 
7011 /*
7012  * Post a sysevent corresponding to the given event.  The 'name' must be one of
7013  * the event definitions in sys/sysevent/eventdefs.h.  The payload will be
7014  * filled in from the spa and (optionally) the vdev.  This doesn't do anything
7015  * in the userland libzpool, as we don't want consumers to misinterpret ztest
7016  * or zdb as real changes.
7017  */
7018 void
7019 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
7020 {
7021 	spa_event_post(spa_event_create(spa, vd, name));
7022 }
7023